blob: 6b5e0436acfdc9529f93a981efa8b371d8f237ac [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>
12#include <linux/slab.h>
13#include <linux/sched.h>
Nick Piggin99b7db72010-08-18 04:37:39 +100014#include <linux/spinlock.h>
15#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070017#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080020#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070022#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080024#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/namei.h>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040026#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/security.h>
28#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080030#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010031#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040032#include <linux/fs_struct.h>
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -050033#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
35#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050036#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazet13f14b42008-02-06 01:37:57 -080039#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
Al Viro5addc5d2005-11-07 17:15:49 -050042static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010043static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010044static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100045static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040046static int mnt_id_start = 0;
47static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Eric Dumazetfa3536c2006-03-26 01:37:24 -080049static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050051static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060054struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080056
Nick Piggin99b7db72010-08-18 04:37:39 +100057/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
Ram Paib58fed82005-11-07 17:16:09 -050069 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080071 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073}
74
Dave Hansen3d733632008-02-15 14:37:59 -080075#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
Nick Piggin99b7db72010-08-18 04:37:39 +100077/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
Al Virob105e272011-11-24 20:38:33 -050081static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010082{
83 int res;
84
85retry:
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100087 spin_lock(&mnt_id_lock);
Al Virob105e272011-11-24 20:38:33 -050088 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt.mnt_id);
Al Virof21f6222009-06-24 03:12:00 -040089 if (!res)
Al Virob105e272011-11-24 20:38:33 -050090 mnt_id_start = mnt->mnt.mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100091 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010092 if (res == -EAGAIN)
93 goto retry;
94
95 return res;
96}
97
Al Virob105e272011-11-24 20:38:33 -050098static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010099{
Al Virob105e272011-11-24 20:38:33 -0500100 int id = mnt->mnt.mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000101 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
104 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000105 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100106}
107
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100108/*
109 * Allocate a new peer group ID
110 *
111 * mnt_group_ida is protected by namespace_sem
112 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500113static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100114{
Al Virof21f6222009-06-24 03:12:00 -0400115 int res;
116
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118 return -ENOMEM;
119
Al Virof21f6222009-06-24 03:12:00 -0400120 res = ida_get_new_above(&mnt_group_ida,
121 mnt_group_start,
Al Viro4b8b21f2011-11-24 19:54:23 -0500122 &mnt->mnt.mnt_group_id);
Al Virof21f6222009-06-24 03:12:00 -0400123 if (!res)
Al Viro4b8b21f2011-11-24 19:54:23 -0500124 mnt_group_start = mnt->mnt.mnt_group_id + 1;
Al Virof21f6222009-06-24 03:12:00 -0400125
126 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127}
128
129/*
130 * Release a peer group ID
131 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500132void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100133{
Al Viro4b8b21f2011-11-24 19:54:23 -0500134 int id = mnt->mnt.mnt_group_id;
Al Virof21f6222009-06-24 03:12:00 -0400135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
Al Viro4b8b21f2011-11-24 19:54:23 -0500138 mnt->mnt.mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100139}
140
Nick Pigginb3e19d92011-01-07 17:50:11 +1100141/*
142 * vfsmount lock must be held for read
143 */
Al Viro83adc752011-11-24 22:37:54 -0500144static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100145{
146#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100148#else
149 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500150 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100151 preempt_enable();
152#endif
153}
154
Nick Pigginb3e19d92011-01-07 17:50:11 +1100155/*
156 * vfsmount lock must be held for write
157 */
Al Viro83adc752011-11-24 22:37:54 -0500158unsigned int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100159{
160#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500161 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100162 int cpu;
163
164 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500165 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100166 }
167
168 return count;
169#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500170 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100171#endif
172}
173
Al Virob105e272011-11-24 20:38:33 -0500174static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
Al Viro7d6fec42011-11-23 12:14:10 -0500176 struct mount *p = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
177 if (p) {
178 struct vfsmount *mnt = &p->mnt;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100179 int err;
180
Al Virob105e272011-11-24 20:38:33 -0500181 err = mnt_alloc_id(p);
Li Zefan88b38782008-07-21 18:06:36 +0800182 if (err)
183 goto out_free_cache;
184
185 if (name) {
186 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
188 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100189 }
190
Nick Pigginb3e19d92011-01-07 17:50:11 +1100191#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500192 p->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 if (!p->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100194 goto out_free_devname;
195
Al Viro68e8a9f2011-11-24 22:53:09 -0500196 this_cpu_add(p->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100197#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500198 p->mnt_count = 1;
199 p->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100200#endif
201
Al Viro1b8e5562011-11-24 21:01:32 -0500202 INIT_LIST_HEAD(&p->mnt_hash);
Al Viro6b41d532011-11-24 23:24:33 -0500203 INIT_LIST_HEAD(&p->mnt_child);
204 INIT_LIST_HEAD(&p->mnt_mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 INIT_LIST_HEAD(&mnt->mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -0500206 INIT_LIST_HEAD(&p->mnt_expire);
207 INIT_LIST_HEAD(&p->mnt_share);
208 INIT_LIST_HEAD(&p->mnt_slave_list);
209 INIT_LIST_HEAD(&p->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500210#ifdef CONFIG_FSNOTIFY
211 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
212#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 }
Al Virob105e272011-11-24 20:38:33 -0500214 return p;
Li Zefan88b38782008-07-21 18:06:36 +0800215
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000216#ifdef CONFIG_SMP
217out_free_devname:
Al Viro7d6fec42011-11-23 12:14:10 -0500218 kfree(p->mnt.mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000219#endif
Li Zefan88b38782008-07-21 18:06:36 +0800220out_free_id:
Al Virob105e272011-11-24 20:38:33 -0500221 mnt_free_id(p);
Li Zefan88b38782008-07-21 18:06:36 +0800222out_free_cache:
Al Viro7d6fec42011-11-23 12:14:10 -0500223 kmem_cache_free(mnt_cache, p);
Li Zefan88b38782008-07-21 18:06:36 +0800224 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Dave Hansen83660252008-02-15 14:37:30 -0800227/*
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
233 * a filesystem.
234 */
Dave Hansen3d733632008-02-15 14:37:59 -0800235/*
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
238 *
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
244 * r/w.
245 */
246int __mnt_is_readonly(struct vfsmount *mnt)
247{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800248 if (mnt->mnt_flags & MNT_READONLY)
249 return 1;
250 if (mnt->mnt_sb->s_flags & MS_RDONLY)
251 return 1;
252 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800253}
254EXPORT_SYMBOL_GPL(__mnt_is_readonly);
255
Al Viro83adc752011-11-24 22:37:54 -0500256static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800257{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000258#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500259 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000260#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500261 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000262#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800263}
Dave Hansen3d733632008-02-15 14:37:59 -0800264
Al Viro83adc752011-11-24 22:37:54 -0500265static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800266{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000267#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500268 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000269#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500270 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000271#endif
272}
273
Al Viro83adc752011-11-24 22:37:54 -0500274static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000275{
276#ifdef CONFIG_SMP
277 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800278 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800279
280 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500281 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800282 }
Dave Hansen3d733632008-02-15 14:37:59 -0800283
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000284 return count;
285#else
286 return mnt->mnt_writers;
287#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800288}
289
290/*
291 * Most r/o checks on a fs are for operations that take
292 * discrete amounts of time, like a write() or unlink().
293 * We must keep track of when those operations start
294 * (for permission checks) and when they end, so that
295 * we can determine when writes are able to occur to
296 * a filesystem.
297 */
Dave Hansen83660252008-02-15 14:37:30 -0800298/**
299 * mnt_want_write - get write access to a mount
Al Viro83adc752011-11-24 22:37:54 -0500300 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800301 *
302 * This tells the low-level filesystem that a write is
303 * about to be performed to it, and makes sure that
304 * writes are allowed before returning success. When
305 * the write operation is finished, mnt_drop_write()
306 * must be called. This is effectively a refcount.
307 */
Al Viro83adc752011-11-24 22:37:54 -0500308int mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800309{
Al Viro83adc752011-11-24 22:37:54 -0500310 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800311 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800312
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000313 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100314 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000315 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100316 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000317 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
318 * incremented count after it has set MNT_WRITE_HOLD.
319 */
320 smp_mb();
Al Viro83adc752011-11-24 22:37:54 -0500321 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000322 cpu_relax();
323 /*
324 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
325 * be set to match its requirements. So we must not load that until
326 * MNT_WRITE_HOLD is cleared.
327 */
328 smp_rmb();
Al Viro83adc752011-11-24 22:37:54 -0500329 if (__mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100330 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800331 ret = -EROFS;
332 goto out;
333 }
Dave Hansen3d733632008-02-15 14:37:59 -0800334out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000335 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800336 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800337}
338EXPORT_SYMBOL_GPL(mnt_want_write);
339
340/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000341 * mnt_clone_write - get write access to a mount
342 * @mnt: the mount on which to take a write
343 *
344 * This is effectively like mnt_want_write, except
345 * it must only be used to take an extra write reference
346 * on a mountpoint that we already know has a write reference
347 * on it. This allows some optimisation.
348 *
349 * After finished, mnt_drop_write must be called as usual to
350 * drop the reference.
351 */
352int mnt_clone_write(struct vfsmount *mnt)
353{
354 /* superblock may be r/o */
355 if (__mnt_is_readonly(mnt))
356 return -EROFS;
357 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500358 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000359 preempt_enable();
360 return 0;
361}
362EXPORT_SYMBOL_GPL(mnt_clone_write);
363
364/**
365 * mnt_want_write_file - get write access to a file's mount
366 * @file: the file who's mount on which to take a write
367 *
368 * This is like mnt_want_write, but it takes a file and can
369 * do some optimisations if the file is open for write already
370 */
371int mnt_want_write_file(struct file *file)
372{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700373 struct inode *inode = file->f_dentry->d_inode;
374 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000375 return mnt_want_write(file->f_path.mnt);
376 else
377 return mnt_clone_write(file->f_path.mnt);
378}
379EXPORT_SYMBOL_GPL(mnt_want_write_file);
380
381/**
Dave Hansen83660252008-02-15 14:37:30 -0800382 * mnt_drop_write - give up write access to a mount
383 * @mnt: the mount on which to give up write access
384 *
385 * Tells the low-level filesystem that we are done
386 * performing writes to it. Must be matched with
387 * mnt_want_write() call above.
388 */
389void mnt_drop_write(struct vfsmount *mnt)
390{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000391 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500392 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000393 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800394}
395EXPORT_SYMBOL_GPL(mnt_drop_write);
396
Al Viro2a79f172011-12-09 08:06:57 -0500397void mnt_drop_write_file(struct file *file)
398{
399 mnt_drop_write(file->f_path.mnt);
400}
401EXPORT_SYMBOL(mnt_drop_write_file);
402
Al Viro83adc752011-11-24 22:37:54 -0500403static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800404{
Dave Hansen3d733632008-02-15 14:37:59 -0800405 int ret = 0;
406
Nick Piggin99b7db72010-08-18 04:37:39 +1000407 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500408 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000409 /*
410 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
411 * should be visible before we do.
412 */
413 smp_mb();
414
415 /*
416 * With writers on hold, if this value is zero, then there are
417 * definitely no active writers (although held writers may subsequently
418 * increment the count, they'll have to wait, and decrement it after
419 * seeing MNT_READONLY).
420 *
421 * It is OK to have counter incremented on one CPU and decremented on
422 * another: the sum will add up correctly. The danger would be when we
423 * sum up each counter, if we read a counter before it is incremented,
424 * but then read another CPU's count which it has been subsequently
425 * decremented from -- we would see more decrements than we should.
426 * MNT_WRITE_HOLD protects against this scenario, because
427 * mnt_want_write first increments count, then smp_mb, then spins on
428 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
429 * we're counting up here.
430 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100431 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000432 ret = -EBUSY;
433 else
Al Viro83adc752011-11-24 22:37:54 -0500434 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000435 /*
436 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
437 * that become unheld will see MNT_READONLY.
438 */
439 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500440 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Nick Piggin99b7db72010-08-18 04:37:39 +1000441 br_write_unlock(vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800442 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800443}
Dave Hansen83660252008-02-15 14:37:30 -0800444
Al Viro83adc752011-11-24 22:37:54 -0500445static void __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800446{
Nick Piggin99b7db72010-08-18 04:37:39 +1000447 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500448 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Nick Piggin99b7db72010-08-18 04:37:39 +1000449 br_write_unlock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800450}
451
Al Virob105e272011-11-24 20:38:33 -0500452static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Al Virob105e272011-11-24 20:38:33 -0500454 kfree(mnt->mnt.mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100455 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000456#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500457 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000458#endif
Al Virob105e272011-11-24 20:38:33 -0500459 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
462/*
Ram Paia05964f2005-11-07 17:20:17 -0500463 * find the first or last mount at @dentry on vfsmount @mnt depending on
464 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000465 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
Al Viroc7105362011-11-24 18:22:03 -0500467struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
Ram Paia05964f2005-11-07 17:20:17 -0500468 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Ram Paib58fed82005-11-07 17:16:09 -0500470 struct list_head *head = mount_hashtable + hash(mnt, dentry);
471 struct list_head *tmp = head;
Al Viroc7105362011-11-24 18:22:03 -0500472 struct mount *p, *found = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500475 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 p = NULL;
477 if (tmp == head)
478 break;
Al Viro1b8e5562011-11-24 21:01:32 -0500479 p = list_entry(tmp, struct mount, mnt_hash);
Al Viroa73324d2011-11-24 22:25:07 -0500480 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500481 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 break;
483 }
484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return found;
486}
487
Ram Paia05964f2005-11-07 17:20:17 -0500488/*
489 * lookup_mnt increments the ref count before returning
490 * the vfsmount struct.
491 */
Al Viro1c755af2009-04-18 14:06:57 -0400492struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500493{
Al Viroc7105362011-11-24 18:22:03 -0500494 struct mount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000495
496 br_read_lock(vfsmount_lock);
Al Viroc7105362011-11-24 18:22:03 -0500497 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
498 if (child_mnt) {
499 mnt_add_count(child_mnt, 1);
500 br_read_unlock(vfsmount_lock);
501 return &child_mnt->mnt;
502 } else {
503 br_read_unlock(vfsmount_lock);
504 return NULL;
505 }
Ram Paia05964f2005-11-07 17:20:17 -0500506}
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508static inline int check_mnt(struct vfsmount *mnt)
509{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800510 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
Nick Piggin99b7db72010-08-18 04:37:39 +1000513/*
514 * vfsmount lock must be held for write
515 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800516static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500517{
518 if (ns) {
519 ns->event = ++event;
520 wake_up_interruptible(&ns->poll);
521 }
522}
523
Nick Piggin99b7db72010-08-18 04:37:39 +1000524/*
525 * vfsmount lock must be held for write
526 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800527static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500528{
529 if (ns && ns->event != event) {
530 ns->event = event;
531 wake_up_interruptible(&ns->poll);
532 }
533}
534
Nick Piggin99b7db72010-08-18 04:37:39 +1000535/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100536 * Clear dentry's mounted state if it has no remaining mounts.
537 * vfsmount_lock must be held for write.
538 */
Al Viroaa0a4cf2011-11-24 19:31:36 -0500539static void dentry_reset_mounted(struct dentry *dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100540{
541 unsigned u;
542
543 for (u = 0; u < HASH_SIZE; u++) {
Al Virod5e50f72011-11-24 20:58:57 -0500544 struct mount *p;
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100545
Al Viro1b8e5562011-11-24 21:01:32 -0500546 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
Al Viroa73324d2011-11-24 22:25:07 -0500547 if (p->mnt_mountpoint == dentry)
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100548 return;
549 }
550 }
551 spin_lock(&dentry->d_lock);
552 dentry->d_flags &= ~DCACHE_MOUNTED;
553 spin_unlock(&dentry->d_lock);
554}
555
556/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000557 * vfsmount lock must be held for write
558 */
Al Viro419148d2011-11-24 19:41:16 -0500559static void detach_mnt(struct mount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Al Viroa73324d2011-11-24 22:25:07 -0500561 old_path->dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500562 old_path->mnt = &mnt->mnt_parent->mnt;
563 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500564 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500565 list_del_init(&mnt->mnt_child);
Al Viro1b8e5562011-11-24 21:01:32 -0500566 list_del_init(&mnt->mnt_hash);
Al Viroaa0a4cf2011-11-24 19:31:36 -0500567 dentry_reset_mounted(old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
Nick Piggin99b7db72010-08-18 04:37:39 +1000570/*
571 * vfsmount lock must be held for write
572 */
Al Viro14cf1fa2011-11-25 00:01:17 -0500573void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
Al Viro44d964d62011-11-24 21:28:22 -0500574 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500575{
Al Viro14cf1fa2011-11-25 00:01:17 -0500576 child_mnt->mnt_parent = real_mount(mntget(&mnt->mnt));
Al Viroa73324d2011-11-24 22:25:07 -0500577 child_mnt->mnt_mountpoint = dget(dentry);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100578 spin_lock(&dentry->d_lock);
579 dentry->d_flags |= DCACHE_MOUNTED;
580 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500581}
582
Nick Piggin99b7db72010-08-18 04:37:39 +1000583/*
584 * vfsmount lock must be held for write
585 */
Al Viro419148d2011-11-24 19:41:16 -0500586static void attach_mnt(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
Al Viro14cf1fa2011-11-25 00:01:17 -0500588 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
Al Viro1b8e5562011-11-24 21:01:32 -0500589 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400590 hash(path->mnt, path->dentry));
Al Viro6b41d532011-11-24 23:24:33 -0500591 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500592}
593
Al Viro83adc752011-11-24 22:37:54 -0500594static inline void __mnt_make_longterm(struct mount *mnt)
Al Viro7e3d0eb02011-01-16 16:32:11 -0500595{
596#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500597 atomic_inc(&mnt->mnt_longterm);
Al Viro7e3d0eb02011-01-16 16:32:11 -0500598#endif
599}
600
601/* needs vfsmount lock for write */
Al Viro83adc752011-11-24 22:37:54 -0500602static inline void __mnt_make_shortterm(struct mount *mnt)
Al Viro7e3d0eb02011-01-16 16:32:11 -0500603{
604#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500605 atomic_dec(&mnt->mnt_longterm);
Al Viro7e3d0eb02011-01-16 16:32:11 -0500606#endif
607}
608
Ram Paib90fa9a2005-11-07 17:19:50 -0500609/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000610 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500611 */
Al Viro4b2619a2011-11-24 19:47:15 -0500612static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500613{
Al Viro0714a532011-11-24 22:19:58 -0500614 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500615 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500616 LIST_HEAD(head);
Al Viro0714a532011-11-24 22:19:58 -0500617 struct mnt_namespace *n = parent->mnt.mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500618
Al Viro0714a532011-11-24 22:19:58 -0500619 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500620
Al Viro4b2619a2011-11-24 19:47:15 -0500621 list_add_tail(&head, &mnt->mnt.mnt_list);
Al Viro83adc752011-11-24 22:37:54 -0500622 list_for_each_entry(m, &head, mnt.mnt_list) {
623 m->mnt.mnt_ns = n;
Al Viro7e3d0eb02011-01-16 16:32:11 -0500624 __mnt_make_longterm(m);
Al Virof03c6592011-01-14 22:30:21 -0500625 }
626
Ram Paib90fa9a2005-11-07 17:19:50 -0500627 list_splice(&head, n->list.prev);
628
Al Viro1b8e5562011-11-24 21:01:32 -0500629 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viroa73324d2011-11-24 22:25:07 -0500630 hash(&parent->mnt, mnt->mnt_mountpoint));
Al Viro6b41d532011-11-24 23:24:33 -0500631 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800632 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
Al Viro315fc832011-11-24 18:57:30 -0500635static struct mount *next_mnt(struct mount *p, struct vfsmount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Al Viro6b41d532011-11-24 23:24:33 -0500637 struct list_head *next = p->mnt_mounts.next;
638 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 while (1) {
Al Viro315fc832011-11-24 18:57:30 -0500640 if (&p->mnt == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500642 next = p->mnt_child.next;
643 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 break;
Al Viro0714a532011-11-24 22:19:58 -0500645 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647 }
Al Viro6b41d532011-11-24 23:24:33 -0500648 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Al Viro315fc832011-11-24 18:57:30 -0500651static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500652{
Al Viro6b41d532011-11-24 23:24:33 -0500653 struct list_head *prev = p->mnt_mounts.prev;
654 while (prev != &p->mnt_mounts) {
655 p = list_entry(prev, struct mount, mnt_child);
656 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500657 }
658 return p;
659}
660
Al Viro9d412a42011-03-17 22:08:28 -0400661struct vfsmount *
662vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
663{
Al Virob105e272011-11-24 20:38:33 -0500664 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400665 struct dentry *root;
666
667 if (!type)
668 return ERR_PTR(-ENODEV);
669
670 mnt = alloc_vfsmnt(name);
671 if (!mnt)
672 return ERR_PTR(-ENOMEM);
673
674 if (flags & MS_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500675 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400676
677 root = mount_fs(type, flags, name, data);
678 if (IS_ERR(root)) {
679 free_vfsmnt(mnt);
680 return ERR_CAST(root);
681 }
682
Al Virob105e272011-11-24 20:38:33 -0500683 mnt->mnt.mnt_root = root;
684 mnt->mnt.mnt_sb = root->d_sb;
Al Viroa73324d2011-11-24 22:25:07 -0500685 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500686 mnt->mnt_parent = mnt;
Al Virob105e272011-11-24 20:38:33 -0500687 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400688}
689EXPORT_SYMBOL_GPL(vfs_kern_mount);
690
Al Viro87129cc2011-11-24 21:24:27 -0500691static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -0500692 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
Al Viro87129cc2011-11-24 21:24:27 -0500694 struct super_block *sb = old->mnt.mnt_sb;
695 struct mount *mnt = alloc_vfsmnt(old->mnt.mnt_devname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100698 if (flag & (CL_SLAVE | CL_PRIVATE))
Al Virob105e272011-11-24 20:38:33 -0500699 mnt->mnt.mnt_group_id = 0; /* not a peer of original */
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100700 else
Al Viro87129cc2011-11-24 21:24:27 -0500701 mnt->mnt.mnt_group_id = old->mnt.mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100702
Al Virob105e272011-11-24 20:38:33 -0500703 if ((flag & CL_MAKE_SHARED) && !mnt->mnt.mnt_group_id) {
704 int err = mnt_alloc_group_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100705 if (err)
706 goto out_free;
707 }
708
Al Viro87129cc2011-11-24 21:24:27 -0500709 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 atomic_inc(&sb->s_active);
Al Virob105e272011-11-24 20:38:33 -0500711 mnt->mnt.mnt_sb = sb;
712 mnt->mnt.mnt_root = dget(root);
Al Viroa73324d2011-11-24 22:25:07 -0500713 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -0500714 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500715
Ram Pai5afe0022005-11-07 17:21:01 -0500716 if (flag & CL_SLAVE) {
Al Viro6776db3d2011-11-25 00:22:05 -0500717 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
Al Viro32301922011-11-25 00:10:28 -0500718 mnt->mnt_master = old;
Al Virob105e272011-11-24 20:38:33 -0500719 CLEAR_MNT_SHARED(&mnt->mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400720 } else if (!(flag & CL_PRIVATE)) {
Al Viro87129cc2011-11-24 21:24:27 -0500721 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(&old->mnt))
Al Viro6776db3d2011-11-25 00:22:05 -0500722 list_add(&mnt->mnt_share, &old->mnt_share);
Al Virod10e8de2011-11-25 00:07:16 -0500723 if (IS_MNT_SLAVE(old))
Al Viro6776db3d2011-11-25 00:22:05 -0500724 list_add(&mnt->mnt_slave, &old->mnt_slave);
Al Virod10e8de2011-11-25 00:07:16 -0500725 mnt->mnt_master = old->mnt_master;
Ram Pai5afe0022005-11-07 17:21:01 -0500726 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500727 if (flag & CL_MAKE_SHARED)
Al Viro0f0afb12011-11-24 20:43:10 -0500728 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 /* stick the duplicate mount on the same expiry list
731 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500732 if (flag & CL_EXPIRE) {
Al Viro6776db3d2011-11-25 00:22:05 -0500733 if (!list_empty(&old->mnt_expire))
734 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 }
Al Virocb338d02011-11-24 20:55:08 -0500737 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100738
739 out_free:
740 free_vfsmnt(mnt);
741 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
Al Viro83adc752011-11-24 22:37:54 -0500744static inline void mntfree(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Al Viro83adc752011-11-24 22:37:54 -0500746 struct vfsmount *m = &mnt->mnt;
747 struct super_block *sb = m->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100748
Dave Hansen3d733632008-02-15 14:37:59 -0800749 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800750 * This probably indicates that somebody messed
751 * up a mnt_want/drop_write() pair. If this
752 * happens, the filesystem was probably unable
753 * to make r/w->r/o transitions.
754 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000755 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100756 * The locking used to deal with mnt_count decrement provides barriers,
757 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000758 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100759 WARN_ON(mnt_get_writers(mnt));
Al Viro83adc752011-11-24 22:37:54 -0500760 fsnotify_vfsmount_delete(m);
761 dput(m->mnt_root);
762 free_vfsmnt(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 deactivate_super(sb);
764}
765
Al Viro83adc752011-11-24 22:37:54 -0500766static void mntput_no_expire(struct vfsmount *m)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500767{
Al Viro83adc752011-11-24 22:37:54 -0500768 struct mount *mnt = real_mount(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100769put_again:
Al Virof03c6592011-01-14 22:30:21 -0500770#ifdef CONFIG_SMP
771 br_read_lock(vfsmount_lock);
Al Viro68e8a9f2011-11-24 22:53:09 -0500772 if (likely(atomic_read(&mnt->mnt_longterm))) {
Al Viroaa9c0e02011-11-23 19:04:52 -0500773 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100774 br_read_unlock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500775 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100776 }
Al Virof03c6592011-01-14 22:30:21 -0500777 br_read_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100778
779 br_write_lock(vfsmount_lock);
Al Viroaa9c0e02011-11-23 19:04:52 -0500780 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100781 if (mnt_get_count(mnt)) {
782 br_write_unlock(vfsmount_lock);
783 return;
784 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100785#else
Al Viroaa9c0e02011-11-23 19:04:52 -0500786 mnt_add_count(mnt, -1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100787 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000788 return;
789 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500790#endif
Al Viro83adc752011-11-24 22:37:54 -0500791 if (unlikely(mnt->mnt.mnt_pinned)) {
792 mnt_add_count(mnt, mnt->mnt.mnt_pinned + 1);
793 mnt->mnt.mnt_pinned = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +1000794 br_write_unlock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -0500795 acct_auto_close_mnt(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100796 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500797 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000798 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100799 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500800}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100801
802void mntput(struct vfsmount *mnt)
803{
804 if (mnt) {
805 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
806 if (unlikely(mnt->mnt_expiry_mark))
807 mnt->mnt_expiry_mark = 0;
Al Virof03c6592011-01-14 22:30:21 -0500808 mntput_no_expire(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100809 }
810}
811EXPORT_SYMBOL(mntput);
812
813struct vfsmount *mntget(struct vfsmount *mnt)
814{
815 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -0500816 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100817 return mnt;
818}
819EXPORT_SYMBOL(mntget);
820
Al Viro7b7b1ac2005-11-07 17:13:39 -0500821void mnt_pin(struct vfsmount *mnt)
822{
Nick Piggin99b7db72010-08-18 04:37:39 +1000823 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500824 mnt->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000825 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500826}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500827EXPORT_SYMBOL(mnt_pin);
828
829void mnt_unpin(struct vfsmount *mnt)
830{
Nick Piggin99b7db72010-08-18 04:37:39 +1000831 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500832 if (mnt->mnt_pinned) {
Al Viro83adc752011-11-24 22:37:54 -0500833 mnt_add_count(real_mount(mnt), 1);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500834 mnt->mnt_pinned--;
835 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000836 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500837}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500838EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800840static inline void mangle(struct seq_file *m, const char *s)
841{
842 seq_escape(m, s, " \t\n\\");
843}
844
845/*
846 * Simple .show_options callback for filesystems which don't want to
847 * implement more complex mount option showing.
848 *
849 * See also save_mount_options().
850 */
851int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
852{
Al Viro2a32ceb2009-05-08 16:05:57 -0400853 const char *options;
854
855 rcu_read_lock();
856 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800857
858 if (options != NULL && options[0]) {
859 seq_putc(m, ',');
860 mangle(m, options);
861 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400862 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800863
864 return 0;
865}
866EXPORT_SYMBOL(generic_show_options);
867
868/*
869 * If filesystem uses generic_show_options(), this function should be
870 * called from the fill_super() callback.
871 *
872 * The .remount_fs callback usually needs to be handled in a special
873 * way, to make sure, that previous options are not overwritten if the
874 * remount fails.
875 *
876 * Also note, that if the filesystem's .remount_fs function doesn't
877 * reset all options to their default value, but changes only newly
878 * given options, then the displayed options will not reflect reality
879 * any more.
880 */
881void save_mount_options(struct super_block *sb, char *options)
882{
Al Viro2a32ceb2009-05-08 16:05:57 -0400883 BUG_ON(sb->s_options);
884 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800885}
886EXPORT_SYMBOL(save_mount_options);
887
Al Viro2a32ceb2009-05-08 16:05:57 -0400888void replace_mount_options(struct super_block *sb, char *options)
889{
890 char *old = sb->s_options;
891 rcu_assign_pointer(sb->s_options, options);
892 if (old) {
893 synchronize_rcu();
894 kfree(old);
895 }
896}
897EXPORT_SYMBOL(replace_mount_options);
898
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100899#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900/* iterator */
901static void *m_start(struct seq_file *m, loff_t *pos)
902{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100903 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Ram Pai390c6842005-11-07 17:17:51 -0500905 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100906 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
909static void *m_next(struct seq_file *m, void *v, loff_t *pos)
910{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100911 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700912
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100913 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914}
915
916static void m_stop(struct seq_file *m, void *v)
917{
Ram Pai390c6842005-11-07 17:17:51 -0500918 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919}
920
Al Viro9f5596a2010-02-05 00:40:25 -0500921int mnt_had_events(struct proc_mounts *p)
922{
923 struct mnt_namespace *ns = p->ns;
924 int res = 0;
925
Nick Piggin99b7db72010-08-18 04:37:39 +1000926 br_read_lock(vfsmount_lock);
Kay Sieversf1514632011-07-12 20:48:39 +0200927 if (p->m.poll_event != ns->event) {
928 p->m.poll_event = ns->event;
Al Viro9f5596a2010-02-05 00:40:25 -0500929 res = 1;
930 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000931 br_read_unlock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500932
933 return res;
934}
935
Ram Pai2d4d4862008-03-27 13:06:25 +0100936struct proc_fs_info {
937 int flag;
938 const char *str;
939};
940
Eric Paris2069f452008-07-04 09:47:13 +1000941static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Ram Pai2d4d4862008-03-27 13:06:25 +0100943 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 { MS_SYNCHRONOUS, ",sync" },
945 { MS_DIRSYNC, ",dirsync" },
946 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 { 0, NULL }
948 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100949 const struct proc_fs_info *fs_infop;
950
951 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
952 if (sb->s_flags & fs_infop->flag)
953 seq_puts(m, fs_infop->str);
954 }
Eric Paris2069f452008-07-04 09:47:13 +1000955
956 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100957}
958
959static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
960{
961 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 { MNT_NOSUID, ",nosuid" },
963 { MNT_NODEV, ",nodev" },
964 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800965 { MNT_NOATIME, ",noatime" },
966 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800967 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 { 0, NULL }
969 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100970 const struct proc_fs_info *fs_infop;
971
972 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
973 if (mnt->mnt_flags & fs_infop->flag)
974 seq_puts(m, fs_infop->str);
975 }
976}
977
978static void show_type(struct seq_file *m, struct super_block *sb)
979{
980 mangle(m, sb->s_type->name);
981 if (sb->s_subtype && sb->s_subtype[0]) {
982 seq_putc(m, '.');
983 mangle(m, sb->s_subtype);
984 }
985}
986
987static int show_vfsmnt(struct seq_file *m, void *v)
988{
989 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
990 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800991 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Al Viroc7f404b2011-03-16 06:59:40 -0400993 if (mnt->mnt_sb->s_op->show_devname) {
994 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
995 if (err)
996 goto out;
997 } else {
998 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -08001001 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +01001003 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001004 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001005 err = show_sb_opts(m, mnt->mnt_sb);
1006 if (err)
1007 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001008 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (mnt->mnt_sb->s_op->show_options)
1010 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1011 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +10001012out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return err;
1014}
1015
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001016const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 .start = m_start,
1018 .next = m_next,
1019 .stop = m_stop,
1020 .show = show_vfsmnt
1021};
1022
Ram Pai2d4d4862008-03-27 13:06:25 +01001023static int show_mountinfo(struct seq_file *m, void *v)
1024{
1025 struct proc_mounts *p = m->private;
1026 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Al Viro3376f342011-11-24 22:05:19 -05001027 struct mount *r = real_mount(mnt);
Ram Pai2d4d4862008-03-27 13:06:25 +01001028 struct super_block *sb = mnt->mnt_sb;
1029 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1030 struct path root = p->root;
1031 int err = 0;
1032
Al Viro0714a532011-11-24 22:19:58 -05001033 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, r->mnt_parent->mnt.mnt_id,
Ram Pai2d4d4862008-03-27 13:06:25 +01001034 MAJOR(sb->s_dev), MINOR(sb->s_dev));
Al Viroc7f404b2011-03-16 06:59:40 -04001035 if (sb->s_op->show_path)
1036 err = sb->s_op->show_path(m, mnt);
1037 else
1038 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1039 if (err)
1040 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001041 seq_putc(m, ' ');
Al Viro02125a82011-12-05 08:43:34 -05001042
1043 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1044 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1045 if (err)
1046 goto out;
1047
Ram Pai2d4d4862008-03-27 13:06:25 +01001048 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1049 show_mnt_opts(m, mnt);
1050
1051 /* Tagged fields ("foo:X" or "bar") */
1052 if (IS_MNT_SHARED(mnt))
1053 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Al Virod10e8de2011-11-25 00:07:16 -05001054 if (IS_MNT_SLAVE(r)) {
Al Viro32301922011-11-25 00:10:28 -05001055 int master = r->mnt_master->mnt.mnt_group_id;
Al Viro6fc78712011-11-24 23:35:54 -05001056 int dom = get_dominating_id(r, &p->root);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +01001057 seq_printf(m, " master:%i", master);
1058 if (dom && dom != master)
1059 seq_printf(m, " propagate_from:%i", dom);
1060 }
Ram Pai2d4d4862008-03-27 13:06:25 +01001061 if (IS_MNT_UNBINDABLE(mnt))
1062 seq_puts(m, " unbindable");
1063
1064 /* Filesystem specific data */
1065 seq_puts(m, " - ");
1066 show_type(m, sb);
1067 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001068 if (sb->s_op->show_devname)
1069 err = sb->s_op->show_devname(m, mnt);
1070 else
1071 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1072 if (err)
1073 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001074 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001075 err = show_sb_opts(m, sb);
1076 if (err)
1077 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001078 if (sb->s_op->show_options)
1079 err = sb->s_op->show_options(m, mnt);
1080 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +10001081out:
Ram Pai2d4d4862008-03-27 13:06:25 +01001082 return err;
1083}
1084
1085const struct seq_operations mountinfo_op = {
1086 .start = m_start,
1087 .next = m_next,
1088 .stop = m_stop,
1089 .show = show_mountinfo,
1090};
1091
Chuck Leverb4629fe2006-03-20 13:44:12 -05001092static int show_vfsstat(struct seq_file *m, void *v)
1093{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001094 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -08001095 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -05001096 int err = 0;
1097
1098 /* device */
Al Viroc7f404b2011-03-16 06:59:40 -04001099 if (mnt->mnt_sb->s_op->show_devname) {
Bryan Schumakera877ee02011-10-07 13:41:15 -04001100 seq_puts(m, "device ");
Al Viroc7f404b2011-03-16 06:59:40 -04001101 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1102 } else {
1103 if (mnt->mnt_devname) {
1104 seq_puts(m, "device ");
1105 mangle(m, mnt->mnt_devname);
1106 } else
1107 seq_puts(m, "no device");
1108 }
Chuck Leverb4629fe2006-03-20 13:44:12 -05001109
1110 /* mount point */
1111 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08001112 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -05001113 seq_putc(m, ' ');
1114
1115 /* file system type */
1116 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +01001117 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001118
1119 /* optional statistics */
1120 if (mnt->mnt_sb->s_op->show_stats) {
1121 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001122 if (!err)
1123 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001124 }
1125
1126 seq_putc(m, '\n');
1127 return err;
1128}
1129
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001130const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -05001131 .start = m_start,
1132 .next = m_next,
1133 .stop = m_stop,
1134 .show = show_vfsstat,
1135};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001136#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138/**
1139 * may_umount_tree - check if a mount tree is busy
1140 * @mnt: root of mount tree
1141 *
1142 * This is called to check if a tree of mounts has any
1143 * open files, pwds, chroots or sub mounts that are
1144 * busy.
1145 */
1146int may_umount_tree(struct vfsmount *mnt)
1147{
Ram Pai36341f62005-11-07 17:17:22 -05001148 int actual_refs = 0;
1149 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001150 struct mount *p;
1151 BUG_ON(!mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Nick Pigginb3e19d92011-01-07 17:50:11 +11001153 /* write lock needed for mnt_get_count */
1154 br_write_lock(vfsmount_lock);
Al Viro315fc832011-11-24 18:57:30 -05001155 for (p = real_mount(mnt); p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001156 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
Nick Pigginb3e19d92011-01-07 17:50:11 +11001159 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001162 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Ian Kente3474a82006-03-27 01:14:51 -08001164 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
1167EXPORT_SYMBOL(may_umount_tree);
1168
1169/**
1170 * may_umount - check if a mount point is busy
1171 * @mnt: root of mount
1172 *
1173 * This is called to check if a mount point has any
1174 * open files, pwds, chroots or sub mounts. If the
1175 * mount has sub mounts this will return busy
1176 * regardless of whether the sub mounts are busy.
1177 *
1178 * Doesn't take quota and stuff into account. IOW, in some cases it will
1179 * give false negatives. The main reason why it's here is that we need
1180 * a non-destructive way to look for easily umountable filesystems.
1181 */
1182int may_umount(struct vfsmount *mnt)
1183{
Ian Kente3474a82006-03-27 01:14:51 -08001184 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001185 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001186 br_write_lock(vfsmount_lock);
Al Viro1ab59732011-11-24 21:35:16 -05001187 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001188 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001189 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001190 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001191 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
1193
1194EXPORT_SYMBOL(may_umount);
1195
Ram Paib90fa9a2005-11-07 17:19:50 -05001196void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
Al Virod5e50f72011-11-24 20:58:57 -05001198 struct mount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001199 while (!list_empty(head)) {
Al Viro1b8e5562011-11-24 21:01:32 -05001200 mnt = list_first_entry(head, struct mount, mnt_hash);
1201 list_del_init(&mnt->mnt_hash);
Al Viro676da582011-11-24 21:47:05 -05001202 if (mnt_has_parent(mnt)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001203 struct dentry *dentry;
1204 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001205
1206 br_write_lock(vfsmount_lock);
Al Viroa73324d2011-11-24 22:25:07 -05001207 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05001208 m = &mnt->mnt_parent->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001209 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro0714a532011-11-24 22:19:58 -05001210 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001211 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001212 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001213 dput(dentry);
1214 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
Al Virod5e50f72011-11-24 20:58:57 -05001216 mntput(&mnt->mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001217 }
1218}
1219
Nick Piggin99b7db72010-08-18 04:37:39 +10001220/*
1221 * vfsmount lock must be held for write
1222 * namespace_sem must be held for write
1223 */
Al Viro761d5c32011-11-24 21:07:43 -05001224void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001225{
Al Viro7b8a53f2011-01-15 20:08:44 -05001226 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001227 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001228
Al Viro761d5c32011-11-24 21:07:43 -05001229 for (p = mnt; p; p = next_mnt(p, &mnt->mnt))
Al Viro1b8e5562011-11-24 21:01:32 -05001230 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001231
Ram Paia05964f2005-11-07 17:20:17 -05001232 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001233 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001234
Al Viro1b8e5562011-11-24 21:01:32 -05001235 list_for_each_entry(p, &tmp_list, mnt_hash) {
Al Viro6776db3d2011-11-25 00:22:05 -05001236 list_del_init(&p->mnt_expire);
Al Viro315fc832011-11-24 18:57:30 -05001237 list_del_init(&p->mnt.mnt_list);
1238 __touch_mnt_namespace(p->mnt.mnt_ns);
1239 p->mnt.mnt_ns = NULL;
Al Viro83adc752011-11-24 22:37:54 -05001240 __mnt_make_shortterm(p);
Al Viro6b41d532011-11-24 23:24:33 -05001241 list_del_init(&p->mnt_child);
Al Viro676da582011-11-24 21:47:05 -05001242 if (mnt_has_parent(p)) {
Al Viro0714a532011-11-24 22:19:58 -05001243 p->mnt_parent->mnt.mnt_ghosts++;
Al Viroa73324d2011-11-24 22:25:07 -05001244 dentry_reset_mounted(p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001245 }
Al Viro0f0afb12011-11-24 20:43:10 -05001246 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001248 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249}
1250
Al Viro692afc32011-11-24 21:15:14 -05001251static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
Al Viroc35038b2008-03-22 00:46:23 -04001252
Al Viro1ab59732011-11-24 21:35:16 -05001253static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Al Viro1ab59732011-11-24 21:35:16 -05001255 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001257 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Al Viro1ab59732011-11-24 21:35:16 -05001259 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 if (retval)
1261 return retval;
1262
1263 /*
1264 * Allow userspace to request a mountpoint be expired rather than
1265 * unmounting unconditionally. Unmount only happens if:
1266 * (1) the mark is already set (the mark is cleared by mntput())
1267 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1268 */
1269 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001270 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 flags & (MNT_FORCE | MNT_DETACH))
1272 return -EINVAL;
1273
Nick Pigginb3e19d92011-01-07 17:50:11 +11001274 /*
1275 * probably don't strictly need the lock here if we examined
1276 * all race cases, but it's a slowpath.
1277 */
1278 br_write_lock(vfsmount_lock);
Al Viro83adc752011-11-24 22:37:54 -05001279 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001280 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001282 }
1283 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Al Viro1ab59732011-11-24 21:35:16 -05001285 if (!xchg(&mnt->mnt.mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return -EAGAIN;
1287 }
1288
1289 /*
1290 * If we may have to abort operations to get out of this
1291 * mount, and they will themselves hold resources we must
1292 * allow the fs to do things. In the Unix tradition of
1293 * 'Gee thats tricky lets do it in userspace' the umount_begin
1294 * might fail to complete on the first run through as other tasks
1295 * must return, and the like. Thats for the mount program to worry
1296 * about for the moment.
1297 */
1298
Al Viro42faad92008-04-24 07:21:56 -04001299 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001300 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 /*
1304 * No sense to grab the lock for this test, but test itself looks
1305 * somewhat bogus. Suggestions for better replacement?
1306 * Ho-hum... In principle, we might treat that as umount + switch
1307 * to rootfs. GC would eventually take care of the old vfsmount.
1308 * Actually it makes sense, especially if rootfs would contain a
1309 * /reboot - static binary that would close all descriptors and
1310 * call reboot(9). Then init(8) could umount root and exec /reboot.
1311 */
Al Viro1ab59732011-11-24 21:35:16 -05001312 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 /*
1314 * Special case for "unmounting" root ...
1315 * we just try to remount it readonly.
1316 */
1317 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001318 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 up_write(&sb->s_umount);
1321 return retval;
1322 }
1323
Ram Pai390c6842005-11-07 17:17:51 -05001324 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001325 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001326 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Al Viroc35038b2008-03-22 00:46:23 -04001328 if (!(flags & MNT_DETACH))
Al Viro1ab59732011-11-24 21:35:16 -05001329 shrink_submounts(mnt, &umount_list);
Al Viroc35038b2008-03-22 00:46:23 -04001330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001332 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Al Viro1ab59732011-11-24 21:35:16 -05001333 if (!list_empty(&mnt->mnt.mnt_list))
1334 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 retval = 0;
1336 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001337 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001338 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001339 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 return retval;
1341}
1342
1343/*
1344 * Now umount can handle mount points as well as block devices.
1345 * This is important for filesystems which use unnamed block devices.
1346 *
1347 * We now support a flag for forced unmount like the other 'big iron'
1348 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1349 */
1350
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001351SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
Al Viro2d8f3032008-07-22 09:59:21 -04001353 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001355 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001357 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1358 return -EINVAL;
1359
1360 if (!(flags & UMOUNT_NOFOLLOW))
1361 lookup_flags |= LOOKUP_FOLLOW;
1362
1363 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (retval)
1365 goto out;
1366 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001367 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001369 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 goto dput_and_out;
1371
1372 retval = -EPERM;
1373 if (!capable(CAP_SYS_ADMIN))
1374 goto dput_and_out;
1375
Al Viro1ab59732011-11-24 21:35:16 -05001376 retval = do_umount(real_mount(path.mnt), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001378 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001379 dput(path.dentry);
1380 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381out:
1382 return retval;
1383}
1384
1385#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1386
1387/*
Ram Paib58fed82005-11-07 17:16:09 -05001388 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001390SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391{
Ram Paib58fed82005-11-07 17:16:09 -05001392 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393}
1394
1395#endif
1396
Al Viro2d92ab32008-08-02 00:51:11 -04001397static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398{
1399 if (capable(CAP_SYS_ADMIN))
1400 return 0;
1401 return -EPERM;
1402#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001403 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001405 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001406 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 return -EPERM;
1408 }
Al Viro2d92ab32008-08-02 00:51:11 -04001409 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 return -EPERM;
1411 return 0;
1412#endif
1413}
1414
Al Viro87129cc2011-11-24 21:24:27 -05001415struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001416 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Al Viroa73324d2011-11-24 22:25:07 -05001418 struct mount *res, *p, *q, *r;
Al Viro1a390682008-03-21 20:48:19 -04001419 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Al Viro87129cc2011-11-24 21:24:27 -05001421 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(&mnt->mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001422 return NULL;
1423
Ram Pai36341f62005-11-07 17:17:22 -05001424 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (!q)
1426 goto Enomem;
Al Viroa73324d2011-11-24 22:25:07 -05001427 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001430 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001431 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001432 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 continue;
1434
Al Viroa73324d2011-11-24 22:25:07 -05001435 for (s = r; s; s = next_mnt(s, &r->mnt)) {
Al Viro315fc832011-11-24 18:57:30 -05001436 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(&s->mnt)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001437 s = skip_mnt_tree(s);
1438 continue;
1439 }
Al Viro0714a532011-11-24 22:19:58 -05001440 while (p != s->mnt_parent) {
1441 p = p->mnt_parent;
1442 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
Al Viro87129cc2011-11-24 21:24:27 -05001444 p = s;
Al Virocb338d02011-11-24 20:55:08 -05001445 path.mnt = &q->mnt;
Al Viroa73324d2011-11-24 22:25:07 -05001446 path.dentry = p->mnt_mountpoint;
Al Viro87129cc2011-11-24 21:24:27 -05001447 q = clone_mnt(p, p->mnt.mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 if (!q)
1449 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001450 br_write_lock(vfsmount_lock);
Al Virocb338d02011-11-24 20:55:08 -05001451 list_add_tail(&q->mnt.mnt_list, &res->mnt.mnt_list);
1452 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001453 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
1455 }
1456 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001457Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001459 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001460 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001461 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001462 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001463 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 }
1465 return NULL;
1466}
1467
Al Viro589ff872009-04-18 03:28:19 -04001468struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001469{
Al Virocb338d02011-11-24 20:55:08 -05001470 struct mount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001471 down_write(&namespace_sem);
Al Viro87129cc2011-11-24 21:24:27 -05001472 tree = copy_tree(real_mount(path->mnt), path->dentry,
1473 CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001474 up_write(&namespace_sem);
Al Virocb338d02011-11-24 20:55:08 -05001475 return tree ? &tree->mnt : NULL;
Al Viro8aec0802007-06-07 12:20:32 -04001476}
1477
1478void drop_collected_mounts(struct vfsmount *mnt)
1479{
1480 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001481 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001482 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001483 umount_tree(real_mount(mnt), 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001484 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001485 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001486 release_mounts(&umount_list);
1487}
1488
Al Viro1f707132010-01-30 22:51:25 -05001489int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1490 struct vfsmount *root)
1491{
1492 struct vfsmount *mnt;
1493 int res = f(root, arg);
1494 if (res)
1495 return res;
1496 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1497 res = f(mnt, arg);
1498 if (res)
1499 return res;
1500 }
1501 return 0;
1502}
1503
Al Viro4b8b21f2011-11-24 19:54:23 -05001504static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001505{
Al Viro315fc832011-11-24 18:57:30 -05001506 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001507
Al Viro4b8b21f2011-11-24 19:54:23 -05001508 for (p = mnt; p != end; p = next_mnt(p, &mnt->mnt)) {
Al Viro315fc832011-11-24 18:57:30 -05001509 if (p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt))
Al Viro4b8b21f2011-11-24 19:54:23 -05001510 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001511 }
1512}
1513
Al Viro4b8b21f2011-11-24 19:54:23 -05001514static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001515{
Al Viro315fc832011-11-24 18:57:30 -05001516 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001517
Al Viro4b8b21f2011-11-24 19:54:23 -05001518 for (p = mnt; p; p = recurse ? next_mnt(p, &mnt->mnt) : NULL) {
Al Viro315fc832011-11-24 18:57:30 -05001519 if (!p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001520 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001521 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05001522 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001523 return err;
1524 }
1525 }
1526 }
1527
1528 return 0;
1529}
1530
Ram Paib90fa9a2005-11-07 17:19:50 -05001531/*
1532 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001533 * @nd : place the mount tree @source_mnt is attached
1534 * @parent_nd : if non-null, detach the source_mnt from its parent and
1535 * store the parent mount and mountpoint dentry.
1536 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001537 *
1538 * NOTE: in the table below explains the semantics when a source mount
1539 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001540 * ---------------------------------------------------------------------------
1541 * | BIND MOUNT OPERATION |
1542 * |**************************************************************************
1543 * | source-->| shared | private | slave | unbindable |
1544 * | dest | | | | |
1545 * | | | | | | |
1546 * | v | | | | |
1547 * |**************************************************************************
1548 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1549 * | | | | | |
1550 * |non-shared| shared (+) | private | slave (*) | invalid |
1551 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001552 * A bind operation clones the source mount and mounts the clone on the
1553 * destination mount.
1554 *
1555 * (++) the cloned mount is propagated to all the mounts in the propagation
1556 * tree of the destination mount and the cloned mount is added to
1557 * the peer group of the source mount.
1558 * (+) the cloned mount is created under the destination mount and is marked
1559 * as shared. The cloned mount is added to the peer group of the source
1560 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001561 * (+++) the mount is propagated to all the mounts in the propagation tree
1562 * of the destination mount and the cloned mount is made slave
1563 * of the same master as that of the source mount. The cloned mount
1564 * is marked as 'shared and slave'.
1565 * (*) the cloned mount is made a slave of the same master as that of the
1566 * source mount.
1567 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001568 * ---------------------------------------------------------------------------
1569 * | MOVE MOUNT OPERATION |
1570 * |**************************************************************************
1571 * | source-->| shared | private | slave | unbindable |
1572 * | dest | | | | |
1573 * | | | | | | |
1574 * | v | | | | |
1575 * |**************************************************************************
1576 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1577 * | | | | | |
1578 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1579 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001580 *
1581 * (+) the mount is moved to the destination. And is then propagated to
1582 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001583 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001584 * (+++) the mount is moved to the destination and is then propagated to
1585 * all the mounts belonging to the destination mount's propagation tree.
1586 * the mount is marked as 'shared and slave'.
1587 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001588 *
1589 * if the source mount is a tree, the operations explained above is
1590 * applied to each mount in the tree.
1591 * Must be called without spinlocks held, since this function can sleep
1592 * in allocations.
1593 */
Al Viro0fb54e52011-11-24 19:59:16 -05001594static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001595 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001596{
1597 LIST_HEAD(tree_list);
Al Viroa8d56d82011-11-24 23:59:29 -05001598 struct mount *dest_mnt = real_mount(path->mnt);
Al Viro1a390682008-03-21 20:48:19 -04001599 struct dentry *dest_dentry = path->dentry;
Al Viro315fc832011-11-24 18:57:30 -05001600 struct mount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001601 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001602
Al Viroa8d56d82011-11-24 23:59:29 -05001603 if (IS_MNT_SHARED(&dest_mnt->mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001604 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001605 if (err)
1606 goto out;
1607 }
Al Viroa8d56d82011-11-24 23:59:29 -05001608 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001609 if (err)
1610 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001611
Nick Piggin99b7db72010-08-18 04:37:39 +10001612 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001613
Al Viroa8d56d82011-11-24 23:59:29 -05001614 if (IS_MNT_SHARED(&dest_mnt->mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05001615 for (p = source_mnt; p; p = next_mnt(p, &source_mnt->mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05001616 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05001617 }
Al Viro1a390682008-03-21 20:48:19 -04001618 if (parent_path) {
Al Viro0fb54e52011-11-24 19:59:16 -05001619 detach_mnt(source_mnt, parent_path);
1620 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001621 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001622 } else {
Al Viro14cf1fa2011-11-25 00:01:17 -05001623 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
Al Viro0fb54e52011-11-24 19:59:16 -05001624 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05001625 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001626
Al Viro1b8e5562011-11-24 21:01:32 -05001627 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1628 list_del_init(&child->mnt_hash);
Al Viro4b2619a2011-11-24 19:47:15 -05001629 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05001630 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001631 br_write_unlock(vfsmount_lock);
1632
Ram Paib90fa9a2005-11-07 17:19:50 -05001633 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001634
1635 out_cleanup_ids:
Al Viroa8d56d82011-11-24 23:59:29 -05001636 if (IS_MNT_SHARED(&dest_mnt->mnt))
Al Viro0fb54e52011-11-24 19:59:16 -05001637 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001638 out:
1639 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001640}
1641
Al Virob12cea92011-03-18 08:55:38 -04001642static int lock_mount(struct path *path)
1643{
1644 struct vfsmount *mnt;
1645retry:
1646 mutex_lock(&path->dentry->d_inode->i_mutex);
1647 if (unlikely(cant_mount(path->dentry))) {
1648 mutex_unlock(&path->dentry->d_inode->i_mutex);
1649 return -ENOENT;
1650 }
1651 down_write(&namespace_sem);
1652 mnt = lookup_mnt(path);
1653 if (likely(!mnt))
1654 return 0;
1655 up_write(&namespace_sem);
1656 mutex_unlock(&path->dentry->d_inode->i_mutex);
1657 path_put(path);
1658 path->mnt = mnt;
1659 path->dentry = dget(mnt->mnt_root);
1660 goto retry;
1661}
1662
1663static void unlock_mount(struct path *path)
1664{
1665 up_write(&namespace_sem);
1666 mutex_unlock(&path->dentry->d_inode->i_mutex);
1667}
1668
Al Viro95bc5f22011-11-25 00:30:56 -05001669static int graft_tree(struct mount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
Al Viro95bc5f22011-11-25 00:30:56 -05001671 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 return -EINVAL;
1673
Al Viro8c3ee422008-03-22 18:00:39 -04001674 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Al Viro95bc5f22011-11-25 00:30:56 -05001675 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 return -ENOTDIR;
1677
Al Virob12cea92011-03-18 08:55:38 -04001678 if (d_unlinked(path->dentry))
1679 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
Al Viro95bc5f22011-11-25 00:30:56 -05001681 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682}
1683
1684/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001685 * Sanity check the flags to change_mnt_propagation.
1686 */
1687
1688static int flags_to_propagation_type(int flags)
1689{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001690 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001691
1692 /* Fail if any non-propagation flags are set */
1693 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1694 return 0;
1695 /* Only one propagation flag should be set */
1696 if (!is_power_of_2(type))
1697 return 0;
1698 return type;
1699}
1700
1701/*
Ram Pai07b20882005-11-07 17:19:07 -05001702 * recursively change the type of the mountpoint.
1703 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001704static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001705{
Al Viro315fc832011-11-24 18:57:30 -05001706 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05001707 struct mount *mnt = real_mount(path->mnt);
Ram Pai07b20882005-11-07 17:19:07 -05001708 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001709 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001710 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001711
Miklos Szerediee6f9582007-05-08 00:30:40 -07001712 if (!capable(CAP_SYS_ADMIN))
1713 return -EPERM;
1714
Al Viro2d92ab32008-08-02 00:51:11 -04001715 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001716 return -EINVAL;
1717
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001718 type = flags_to_propagation_type(flag);
1719 if (!type)
1720 return -EINVAL;
1721
Ram Pai07b20882005-11-07 17:19:07 -05001722 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001723 if (type == MS_SHARED) {
1724 err = invent_group_ids(mnt, recurse);
1725 if (err)
1726 goto out_unlock;
1727 }
1728
Nick Piggin99b7db72010-08-18 04:37:39 +10001729 br_write_lock(vfsmount_lock);
Al Viro4b8b21f2011-11-24 19:54:23 -05001730 for (m = mnt; m; m = (recurse ? next_mnt(m, &mnt->mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05001731 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001732 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001733
1734 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001735 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001736 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001737}
1738
1739/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 * do loopback mount.
1741 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001742static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001743 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744{
Al Virob12cea92011-03-18 08:55:38 -04001745 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001746 struct path old_path;
Al Viro87129cc2011-11-24 21:24:27 -05001747 struct mount *mnt = NULL, *old;
Al Viro2d92ab32008-08-02 00:51:11 -04001748 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 if (err)
1750 return err;
1751 if (!old_name || !*old_name)
1752 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001753 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 if (err)
1755 return err;
1756
Al Virob12cea92011-03-18 08:55:38 -04001757 err = lock_mount(path);
1758 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001759 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001760
Al Viro87129cc2011-11-24 21:24:27 -05001761 old = real_mount(old_path.mnt);
1762
Al Virob12cea92011-03-18 08:55:38 -04001763 err = -EINVAL;
1764 if (IS_MNT_UNBINDABLE(old_path.mnt))
1765 goto out2;
1766
Al Viro2d92ab32008-08-02 00:51:11 -04001767 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Virob12cea92011-03-18 08:55:38 -04001768 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Al Viroccd48bc2005-11-07 17:15:04 -05001770 err = -ENOMEM;
1771 if (recurse)
Al Viro87129cc2011-11-24 21:24:27 -05001772 mnt = copy_tree(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001773 else
Al Viro87129cc2011-11-24 21:24:27 -05001774 mnt = clone_mnt(old, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001775
1776 if (!mnt)
Al Virob12cea92011-03-18 08:55:38 -04001777 goto out2;
Al Viroccd48bc2005-11-07 17:15:04 -05001778
Al Viro95bc5f22011-11-25 00:30:56 -05001779 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001780 if (err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001781 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05001782 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001783 br_write_unlock(vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001784 }
Al Virob12cea92011-03-18 08:55:38 -04001785out2:
1786 unlock_mount(path);
1787 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001788out:
Al Viro2d92ab32008-08-02 00:51:11 -04001789 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 return err;
1791}
1792
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001793static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1794{
1795 int error = 0;
1796 int readonly_request = 0;
1797
1798 if (ms_flags & MS_RDONLY)
1799 readonly_request = 1;
1800 if (readonly_request == __mnt_is_readonly(mnt))
1801 return 0;
1802
1803 if (readonly_request)
Al Viro83adc752011-11-24 22:37:54 -05001804 error = mnt_make_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001805 else
Al Viro83adc752011-11-24 22:37:54 -05001806 __mnt_unmake_readonly(real_mount(mnt));
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001807 return error;
1808}
1809
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810/*
1811 * change filesystem flags. dir should be a physical root of filesystem.
1812 * If you've mounted a non-root directory somewhere and want to do remount
1813 * on it - tough luck.
1814 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001815static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 void *data)
1817{
1818 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001819 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 if (!capable(CAP_SYS_ADMIN))
1822 return -EPERM;
1823
Al Viro2d92ab32008-08-02 00:51:11 -04001824 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return -EINVAL;
1826
Al Viro2d92ab32008-08-02 00:51:11 -04001827 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return -EINVAL;
1829
Eric Parisff36fe22011-03-03 16:09:14 -05001830 err = security_sb_remount(sb, data);
1831 if (err)
1832 return err;
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001835 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001836 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001837 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001838 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001839 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001840 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001841 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001842 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001843 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001846 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001847 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001848 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001849 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 return err;
1852}
1853
Al Virocbbe3622011-11-24 20:01:19 -05001854static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05001855{
Al Viro315fc832011-11-24 18:57:30 -05001856 struct mount *p;
Al Virocbbe3622011-11-24 20:01:19 -05001857 for (p = mnt; p; p = next_mnt(p, &mnt->mnt)) {
Al Viro315fc832011-11-24 18:57:30 -05001858 if (IS_MNT_UNBINDABLE(&p->mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001859 return 1;
1860 }
1861 return 0;
1862}
1863
Al Viro0a0d8a42008-08-02 00:55:27 -04001864static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865{
Al Viro2d92ab32008-08-02 00:51:11 -04001866 struct path old_path, parent_path;
Al Viro676da582011-11-24 21:47:05 -05001867 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05001868 struct mount *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 int err = 0;
1870 if (!capable(CAP_SYS_ADMIN))
1871 return -EPERM;
1872 if (!old_name || !*old_name)
1873 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001874 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (err)
1876 return err;
1877
Al Virob12cea92011-03-18 08:55:38 -04001878 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001879 if (err < 0)
1880 goto out;
1881
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001883 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 goto out1;
1885
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001886 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001887 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
1889 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001890 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001891 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Al Viro0fb54e52011-11-24 19:59:16 -05001893 old = real_mount(old_path.mnt);
1894
Al Viro676da582011-11-24 21:47:05 -05001895 if (!mnt_has_parent(old))
Ram Pai21444402005-11-07 17:20:03 -05001896 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Al Viro2d92ab32008-08-02 00:51:11 -04001898 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1899 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001900 goto out1;
1901 /*
1902 * Don't move a mount residing in a shared parent.
1903 */
Al Viro0714a532011-11-24 22:19:58 -05001904 if (IS_MNT_SHARED(&old->mnt_parent->mnt))
Ram Pai21444402005-11-07 17:20:03 -05001905 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001906 /*
1907 * Don't move a mount tree containing unbindable mounts to a destination
1908 * mount which is shared.
1909 */
Al Viro2d92ab32008-08-02 00:51:11 -04001910 if (IS_MNT_SHARED(path->mnt) &&
Al Virocbbe3622011-11-24 20:01:19 -05001911 tree_contains_unbindable(old))
Ram Pai9676f0c2005-11-07 17:21:20 -05001912 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 err = -ELOOP;
Al Viro0714a532011-11-24 22:19:58 -05001914 for (p = real_mount(path->mnt); mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05001915 if (p == old)
Ram Pai21444402005-11-07 17:20:03 -05001916 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917
Al Viro0fb54e52011-11-24 19:59:16 -05001918 err = attach_recursive_mnt(old, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001919 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001920 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922 /* if the mount is moved, it should no longer be expire
1923 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05001924 list_del_init(&old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925out1:
Al Virob12cea92011-03-18 08:55:38 -04001926 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001929 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001930 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 return err;
1932}
1933
Al Viro9d412a42011-03-17 22:08:28 -04001934static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1935{
1936 int err;
1937 const char *subtype = strchr(fstype, '.');
1938 if (subtype) {
1939 subtype++;
1940 err = -EINVAL;
1941 if (!subtype[0])
1942 goto err;
1943 } else
1944 subtype = "";
1945
1946 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1947 err = -ENOMEM;
1948 if (!mnt->mnt_sb->s_subtype)
1949 goto err;
1950 return mnt;
1951
1952 err:
1953 mntput(mnt);
1954 return ERR_PTR(err);
1955}
1956
Al Viro79e801a2011-12-12 23:07:05 -05001957static struct vfsmount *
Al Viro9d412a42011-03-17 22:08:28 -04001958do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1959{
1960 struct file_system_type *type = get_fs_type(fstype);
1961 struct vfsmount *mnt;
1962 if (!type)
1963 return ERR_PTR(-ENODEV);
1964 mnt = vfs_kern_mount(type, flags, name, data);
1965 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1966 !mnt->mnt_sb->s_subtype)
1967 mnt = fs_set_subtype(mnt, fstype);
1968 put_filesystem(type);
1969 return mnt;
1970}
Al Viro9d412a42011-03-17 22:08:28 -04001971
1972/*
1973 * add a mount into a namespace's mount tree
1974 */
Al Viro95bc5f22011-11-25 00:30:56 -05001975static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04001976{
1977 int err;
1978
1979 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1980
Al Virob12cea92011-03-18 08:55:38 -04001981 err = lock_mount(path);
1982 if (err)
1983 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001984
1985 err = -EINVAL;
1986 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1987 goto unlock;
1988
1989 /* Refuse the same filesystem on the same mount point */
1990 err = -EBUSY;
Al Viro95bc5f22011-11-25 00:30:56 -05001991 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04001992 path->mnt->mnt_root == path->dentry)
1993 goto unlock;
1994
1995 err = -EINVAL;
Al Viro95bc5f22011-11-25 00:30:56 -05001996 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
Al Viro9d412a42011-03-17 22:08:28 -04001997 goto unlock;
1998
Al Viro95bc5f22011-11-25 00:30:56 -05001999 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro9d412a42011-03-17 22:08:28 -04002000 err = graft_tree(newmnt, path);
2001
2002unlock:
Al Virob12cea92011-03-18 08:55:38 -04002003 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04002004 return err;
2005}
Al Virob1e75df2011-01-17 01:47:59 -05002006
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007/*
2008 * create a new mount for userspace and request it to be added into the
2009 * namespace's tree
2010 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002011static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 int mnt_flags, char *name, void *data)
2013{
2014 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002015 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
Vegard Nossumeca6f532009-09-18 13:05:45 -07002017 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 return -EINVAL;
2019
2020 /* we need capabilities... */
2021 if (!capable(CAP_SYS_ADMIN))
2022 return -EPERM;
2023
2024 mnt = do_kern_mount(type, flags, name, data);
2025 if (IS_ERR(mnt))
2026 return PTR_ERR(mnt);
2027
Al Viro95bc5f22011-11-25 00:30:56 -05002028 err = do_add_mount(real_mount(mnt), path, mnt_flags);
Al Viro15f9a3f2011-01-17 01:41:58 -05002029 if (err)
2030 mntput(mnt);
2031 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
Al Viro19a167a2011-01-17 01:35:23 -05002034int finish_automount(struct vfsmount *m, struct path *path)
2035{
Al Viro6776db3d2011-11-25 00:22:05 -05002036 struct mount *mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002037 int err;
2038 /* The new mount record should have at least 2 refs to prevent it being
2039 * expired before we get a chance to add it
2040 */
Al Viro6776db3d2011-11-25 00:22:05 -05002041 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002042
2043 if (m->mnt_sb == path->mnt->mnt_sb &&
2044 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002045 err = -ELOOP;
2046 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002047 }
2048
Al Viro95bc5f22011-11-25 00:30:56 -05002049 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002050 if (!err)
2051 return 0;
2052fail:
2053 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002054 if (!list_empty(&mnt->mnt_expire)) {
Al Virob1e75df2011-01-17 01:47:59 -05002055 down_write(&namespace_sem);
2056 br_write_lock(vfsmount_lock);
Al Viro6776db3d2011-11-25 00:22:05 -05002057 list_del_init(&mnt->mnt_expire);
Al Virob1e75df2011-01-17 01:47:59 -05002058 br_write_unlock(vfsmount_lock);
2059 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05002060 }
Al Virob1e75df2011-01-17 01:47:59 -05002061 mntput(m);
2062 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002063 return err;
2064}
2065
David Howellsea5b7782011-01-14 19:10:03 +00002066/**
2067 * mnt_set_expiry - Put a mount on an expiration list
2068 * @mnt: The mount to list.
2069 * @expiry_list: The list to add the mount to.
2070 */
2071void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2072{
2073 down_write(&namespace_sem);
2074 br_write_lock(vfsmount_lock);
2075
Al Viro6776db3d2011-11-25 00:22:05 -05002076 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002077
2078 br_write_unlock(vfsmount_lock);
2079 up_write(&namespace_sem);
2080}
2081EXPORT_SYMBOL(mnt_set_expiry);
2082
2083/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 * process a list of expirable mountpoints with the intent of discarding any
2085 * mountpoints that aren't in use and haven't been touched since last we came
2086 * here
2087 */
2088void mark_mounts_for_expiry(struct list_head *mounts)
2089{
Al Viro761d5c32011-11-24 21:07:43 -05002090 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002092 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
2094 if (list_empty(mounts))
2095 return;
2096
Al Virobcc5c7d2008-03-22 00:21:53 -04002097 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002098 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
2100 /* extract from the expiration list every vfsmount that matches the
2101 * following criteria:
2102 * - only referenced by its parent vfsmount
2103 * - still marked for expiry (marked on the last call here; marks are
2104 * cleared by mntput())
2105 */
Al Viro6776db3d2011-11-25 00:22:05 -05002106 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro761d5c32011-11-24 21:07:43 -05002107 if (!xchg(&mnt->mnt.mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002108 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05002110 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002112 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002113 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro761d5c32011-11-24 21:07:43 -05002114 touch_mnt_namespace(mnt->mnt.mnt_ns);
Al Virobcc5c7d2008-03-22 00:21:53 -04002115 umount_tree(mnt, 1, &umounts);
2116 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002117 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002118 up_write(&namespace_sem);
2119
2120 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121}
2122
2123EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2124
2125/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002126 * Ripoff of 'select_parent()'
2127 *
2128 * search the list of submounts for a given mountpoint, and move any
2129 * shrinkable submounts to the 'graveyard' list.
2130 */
Al Viro692afc32011-11-24 21:15:14 -05002131static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002132{
Al Viro692afc32011-11-24 21:15:14 -05002133 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002134 struct list_head *next;
2135 int found = 0;
2136
2137repeat:
Al Viro6b41d532011-11-24 23:24:33 -05002138 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002139resume:
Al Viro6b41d532011-11-24 23:24:33 -05002140 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002141 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05002142 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002143
2144 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05002145 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04002146 continue;
2147 /*
2148 * Descend a level if the d_mounts list is non-empty.
2149 */
Al Viro6b41d532011-11-24 23:24:33 -05002150 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002151 this_parent = mnt;
2152 goto repeat;
2153 }
2154
Al Viro1ab59732011-11-24 21:35:16 -05002155 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05002156 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04002157 found++;
2158 }
2159 }
2160 /*
2161 * All done at this level ... ascend and resume the search
2162 */
2163 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05002164 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05002165 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002166 goto resume;
2167 }
2168 return found;
2169}
2170
2171/*
2172 * process a list of expirable mountpoints with the intent of discarding any
2173 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002174 *
2175 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002176 */
Al Viro692afc32011-11-24 21:15:14 -05002177static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002178{
2179 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05002180 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002181
Trond Myklebust5528f9112006-06-09 09:34:17 -04002182 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002183 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002184 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05002185 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05002186 mnt_expire);
Al Viro761d5c32011-11-24 21:07:43 -05002187 touch_mnt_namespace(m->mnt.mnt_ns);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002188 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002189 }
2190 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002191}
2192
Trond Myklebust5528f9112006-06-09 09:34:17 -04002193/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 * Some copy_from_user() implementations do not return the exact number of
2195 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2196 * Note that this function differs from copy_from_user() in that it will oops
2197 * on bad values of `to', rather than returning a short copy.
2198 */
Ram Paib58fed82005-11-07 17:16:09 -05002199static long exact_copy_from_user(void *to, const void __user * from,
2200 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
2202 char *t = to;
2203 const char __user *f = from;
2204 char c;
2205
2206 if (!access_ok(VERIFY_READ, from, n))
2207 return n;
2208
2209 while (n) {
2210 if (__get_user(c, f)) {
2211 memset(t, 0, n);
2212 break;
2213 }
2214 *t++ = c;
2215 f++;
2216 n--;
2217 }
2218 return n;
2219}
2220
Ram Paib58fed82005-11-07 17:16:09 -05002221int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222{
2223 int i;
2224 unsigned long page;
2225 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002226
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 *where = 0;
2228 if (!data)
2229 return 0;
2230
2231 if (!(page = __get_free_page(GFP_KERNEL)))
2232 return -ENOMEM;
2233
2234 /* We only care that *some* data at the address the user
2235 * gave us is valid. Just in case, we'll zero
2236 * the remainder of the page.
2237 */
2238 /* copy_from_user cannot cross TASK_SIZE ! */
2239 size = TASK_SIZE - (unsigned long)data;
2240 if (size > PAGE_SIZE)
2241 size = PAGE_SIZE;
2242
2243 i = size - exact_copy_from_user((void *)page, data, size);
2244 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002245 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 return -EFAULT;
2247 }
2248 if (i != PAGE_SIZE)
2249 memset((char *)page + i, 0, PAGE_SIZE - i);
2250 *where = page;
2251 return 0;
2252}
2253
Vegard Nossumeca6f532009-09-18 13:05:45 -07002254int copy_mount_string(const void __user *data, char **where)
2255{
2256 char *tmp;
2257
2258 if (!data) {
2259 *where = NULL;
2260 return 0;
2261 }
2262
2263 tmp = strndup_user(data, PAGE_SIZE);
2264 if (IS_ERR(tmp))
2265 return PTR_ERR(tmp);
2266
2267 *where = tmp;
2268 return 0;
2269}
2270
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271/*
2272 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2273 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2274 *
2275 * data is a (void *) that can point to any structure up to
2276 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2277 * information (or be NULL).
2278 *
2279 * Pre-0.97 versions of mount() didn't have a flags word.
2280 * When the flags word was introduced its top half was required
2281 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2282 * Therefore, if this magic number is present, it carries no information
2283 * and must be discarded.
2284 */
Ram Paib58fed82005-11-07 17:16:09 -05002285long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 unsigned long flags, void *data_page)
2287{
Al Viro2d92ab32008-08-02 00:51:11 -04002288 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 int retval = 0;
2290 int mnt_flags = 0;
2291
2292 /* Discard magic */
2293 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2294 flags &= ~MS_MGC_MSK;
2295
2296 /* Basic sanity checks */
2297
2298 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2299 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
2301 if (data_page)
2302 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2303
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002304 /* ... and get the mountpoint */
2305 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2306 if (retval)
2307 return retval;
2308
2309 retval = security_sb_mount(dev_name, &path,
2310 type_page, flags, data_page);
2311 if (retval)
2312 goto dput_out;
2313
Andi Kleen613cbe32009-04-19 18:40:43 +02002314 /* Default to relatime unless overriden */
2315 if (!(flags & MS_NOATIME))
2316 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002317
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 /* Separate the per-mountpoint flags */
2319 if (flags & MS_NOSUID)
2320 mnt_flags |= MNT_NOSUID;
2321 if (flags & MS_NODEV)
2322 mnt_flags |= MNT_NODEV;
2323 if (flags & MS_NOEXEC)
2324 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002325 if (flags & MS_NOATIME)
2326 mnt_flags |= MNT_NOATIME;
2327 if (flags & MS_NODIRATIME)
2328 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002329 if (flags & MS_STRICTATIME)
2330 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002331 if (flags & MS_RDONLY)
2332 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002333
Al Viro7a4dec52010-08-09 12:05:43 -04002334 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002335 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2336 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002339 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 data_page);
2341 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002342 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002343 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002344 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002346 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 else
Al Viro2d92ab32008-08-02 00:51:11 -04002348 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 dev_name, data_page);
2350dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002351 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 return retval;
2353}
2354
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002355static struct mnt_namespace *alloc_mnt_ns(void)
2356{
2357 struct mnt_namespace *new_ns;
2358
2359 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2360 if (!new_ns)
2361 return ERR_PTR(-ENOMEM);
2362 atomic_set(&new_ns->count, 1);
2363 new_ns->root = NULL;
2364 INIT_LIST_HEAD(&new_ns->list);
2365 init_waitqueue_head(&new_ns->poll);
2366 new_ns->event = 0;
2367 return new_ns;
2368}
2369
Al Virof03c6592011-01-14 22:30:21 -05002370void mnt_make_longterm(struct vfsmount *mnt)
2371{
Al Viro83adc752011-11-24 22:37:54 -05002372 __mnt_make_longterm(real_mount(mnt));
Al Virof03c6592011-01-14 22:30:21 -05002373}
2374
Al Viro83adc752011-11-24 22:37:54 -05002375void mnt_make_shortterm(struct vfsmount *m)
Al Virof03c6592011-01-14 22:30:21 -05002376{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002377#ifdef CONFIG_SMP
Al Viro83adc752011-11-24 22:37:54 -05002378 struct mount *mnt = real_mount(m);
Al Viro68e8a9f2011-11-24 22:53:09 -05002379 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
Al Virof03c6592011-01-14 22:30:21 -05002380 return;
2381 br_write_lock(vfsmount_lock);
Al Viro68e8a9f2011-11-24 22:53:09 -05002382 atomic_dec(&mnt->mnt_longterm);
Al Virof03c6592011-01-14 22:30:21 -05002383 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002384#endif
Al Virof03c6592011-01-14 22:30:21 -05002385}
2386
JANAK DESAI741a2952006-02-07 12:59:00 -08002387/*
2388 * Allocate a new namespace structure and populate it with contents
2389 * copied from the namespace of the passed in task structure.
2390 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002391static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002392 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002394 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002395 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05002396 struct mount *p, *q;
Al Virocb338d02011-11-24 20:55:08 -05002397 struct mount *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002399 new_ns = alloc_mnt_ns();
2400 if (IS_ERR(new_ns))
2401 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Ram Pai390c6842005-11-07 17:17:51 -05002403 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 /* First pass: copy the tree topology */
Al Viro87129cc2011-11-24 21:24:27 -05002405 new = copy_tree(real_mount(mnt_ns->root), mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002406 CL_COPY_ALL | CL_EXPIRE);
Al Virocb338d02011-11-24 20:55:08 -05002407 if (!new) {
Ram Pai390c6842005-11-07 17:17:51 -05002408 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002410 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 }
Al Virocb338d02011-11-24 20:55:08 -05002412 new_ns->root = &new->mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +10002413 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002415 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
2417 /*
2418 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2419 * as belonging to new namespace. We have already acquired a private
2420 * fs_struct, so tsk->fs->lock is not needed.
2421 */
Al Viro315fc832011-11-24 18:57:30 -05002422 p = real_mount(mnt_ns->root);
Al Virocb338d02011-11-24 20:55:08 -05002423 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 while (p) {
Al Viro315fc832011-11-24 18:57:30 -05002425 q->mnt.mnt_ns = new_ns;
Al Viro83adc752011-11-24 22:37:54 -05002426 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (fs) {
Al Viro315fc832011-11-24 18:57:30 -05002428 if (&p->mnt == fs->root.mnt) {
2429 fs->root.mnt = mntget(&q->mnt);
Al Viro83adc752011-11-24 22:37:54 -05002430 __mnt_make_longterm(q);
Al Viro315fc832011-11-24 18:57:30 -05002431 mnt_make_shortterm(&p->mnt);
2432 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 }
Al Viro315fc832011-11-24 18:57:30 -05002434 if (&p->mnt == fs->pwd.mnt) {
2435 fs->pwd.mnt = mntget(&q->mnt);
Al Viro83adc752011-11-24 22:37:54 -05002436 __mnt_make_longterm(q);
Al Viro315fc832011-11-24 18:57:30 -05002437 mnt_make_shortterm(&p->mnt);
2438 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002441 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 q = next_mnt(q, new_ns->root);
2443 }
Ram Pai390c6842005-11-07 17:17:51 -05002444 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002447 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002449 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
JANAK DESAI741a2952006-02-07 12:59:00 -08002451 return new_ns;
2452}
2453
Eric W. Biederman213dd262007-07-15 23:41:15 -07002454struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002455 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002456{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002457 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002458
Badari Pulavartye3222c42007-05-08 00:25:21 -07002459 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002460 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002461
2462 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002463 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002464
Badari Pulavartye3222c42007-05-08 00:25:21 -07002465 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002466
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002467 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002468 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469}
2470
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002471/**
2472 * create_mnt_ns - creates a private namespace and adds a root filesystem
2473 * @mnt: pointer to the new root filesystem mountpoint
2474 */
Al Viro6c449c82011-11-16 22:01:36 -05002475static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002476{
2477 struct mnt_namespace *new_ns;
2478
2479 new_ns = alloc_mnt_ns();
2480 if (!IS_ERR(new_ns)) {
2481 mnt->mnt_ns = new_ns;
Al Viro83adc752011-11-24 22:37:54 -05002482 __mnt_make_longterm(real_mount(mnt));
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002483 new_ns->root = mnt;
2484 list_add(&new_ns->list, &new_ns->root->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002485 } else {
2486 mntput(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002487 }
2488 return new_ns;
2489}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002490
Al Viroea441d12011-11-16 21:43:59 -05002491struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2492{
2493 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002494 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002495 struct path path;
2496 int err;
2497
2498 ns = create_mnt_ns(mnt);
2499 if (IS_ERR(ns))
2500 return ERR_CAST(ns);
2501
2502 err = vfs_path_lookup(mnt->mnt_root, mnt,
2503 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2504
2505 put_mnt_ns(ns);
2506
2507 if (err)
2508 return ERR_PTR(err);
2509
2510 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002511 s = path.mnt->mnt_sb;
2512 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002513 mntput(path.mnt);
2514 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002515 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002516 /* ... and return the root of (sub)tree on it */
2517 return path.dentry;
2518}
2519EXPORT_SYMBOL(mount_subtree);
2520
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002521SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2522 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002524 int ret;
2525 char *kernel_type;
2526 char *kernel_dir;
2527 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529
Vegard Nossumeca6f532009-09-18 13:05:45 -07002530 ret = copy_mount_string(type, &kernel_type);
2531 if (ret < 0)
2532 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Vegard Nossumeca6f532009-09-18 13:05:45 -07002534 kernel_dir = getname(dir_name);
2535 if (IS_ERR(kernel_dir)) {
2536 ret = PTR_ERR(kernel_dir);
2537 goto out_dir;
2538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Vegard Nossumeca6f532009-09-18 13:05:45 -07002540 ret = copy_mount_string(dev_name, &kernel_dev);
2541 if (ret < 0)
2542 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
Vegard Nossumeca6f532009-09-18 13:05:45 -07002544 ret = copy_mount_options(data, &data_page);
2545 if (ret < 0)
2546 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
Vegard Nossumeca6f532009-09-18 13:05:45 -07002548 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2549 (void *) data_page);
2550
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002552out_data:
2553 kfree(kernel_dev);
2554out_dev:
2555 putname(kernel_dir);
2556out_dir:
2557 kfree(kernel_type);
2558out_type:
2559 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
2562/*
Al Viroafac7cb2011-11-23 19:34:49 -05002563 * Return true if path is reachable from root
2564 *
2565 * namespace_sem or vfsmount_lock is held
2566 */
Al Viro643822b2011-11-24 22:00:28 -05002567bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05002568 const struct path *root)
2569{
Al Viro643822b2011-11-24 22:00:28 -05002570 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05002571 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002572 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05002573 }
Al Viro643822b2011-11-24 22:00:28 -05002574 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05002575}
2576
2577int path_is_under(struct path *path1, struct path *path2)
2578{
2579 int res;
2580 br_read_lock(vfsmount_lock);
Al Viro643822b2011-11-24 22:00:28 -05002581 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viroafac7cb2011-11-23 19:34:49 -05002582 br_read_unlock(vfsmount_lock);
2583 return res;
2584}
2585EXPORT_SYMBOL(path_is_under);
2586
2587/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 * pivot_root Semantics:
2589 * Moves the root file system of the current process to the directory put_old,
2590 * makes new_root as the new root file system of the current process, and sets
2591 * root/cwd of all processes which had them on the current root to new_root.
2592 *
2593 * Restrictions:
2594 * The new_root and put_old must be directories, and must not be on the
2595 * same file system as the current process root. The put_old must be
2596 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2597 * pointed to by put_old must yield the same directory as new_root. No other
2598 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2599 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002600 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2601 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2602 * in this situation.
2603 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 * Notes:
2605 * - we don't move root/cwd if they are not at the root (reason: if something
2606 * cared enough to change them, it's probably wrong to force them elsewhere)
2607 * - it's okay to pick a root that isn't the root of a file system, e.g.
2608 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2609 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2610 * first.
2611 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002612SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2613 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614{
Al Viro2d8f3032008-07-22 09:59:21 -04002615 struct path new, old, parent_path, root_parent, root;
Al Viro419148d2011-11-24 19:41:16 -05002616 struct mount *new_mnt, *root_mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 int error;
2618
2619 if (!capable(CAP_SYS_ADMIN))
2620 return -EPERM;
2621
Al Viro2d8f3032008-07-22 09:59:21 -04002622 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 if (error)
2624 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Al Viro2d8f3032008-07-22 09:59:21 -04002626 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 if (error)
2628 goto out1;
2629
Al Viro2d8f3032008-07-22 09:59:21 -04002630 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002631 if (error)
2632 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002634 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002635 error = lock_mount(&old);
2636 if (error)
2637 goto out3;
2638
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05002640 new_mnt = real_mount(new.mnt);
2641 root_mnt = real_mount(root.mnt);
Al Viro2d8f3032008-07-22 09:59:21 -04002642 if (IS_MNT_SHARED(old.mnt) ||
Al Viro0714a532011-11-24 22:19:58 -05002643 IS_MNT_SHARED(&new_mnt->mnt_parent->mnt) ||
2644 IS_MNT_SHARED(&root_mnt->mnt_parent->mnt))
Al Virob12cea92011-03-18 08:55:38 -04002645 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002646 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002647 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002649 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002650 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002651 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002652 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002654 if (new.mnt == root.mnt ||
2655 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002656 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002658 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002659 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002660 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002661 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002662 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002663 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05002664 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04002665 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002666 /* make sure we can reach put_old from new_root */
Al Viro643822b2011-11-24 22:00:28 -05002667 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04002668 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002669 br_write_lock(vfsmount_lock);
Al Viro419148d2011-11-24 19:41:16 -05002670 detach_mnt(new_mnt, &parent_path);
2671 detach_mnt(root_mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002672 /* mount old root on put_old */
Al Viro419148d2011-11-24 19:41:16 -05002673 attach_mnt(root_mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002674 /* mount new_root on / */
Al Viro419148d2011-11-24 19:41:16 -05002675 attach_mnt(new_mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002676 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002677 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002678 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002680out4:
2681 unlock_mount(&old);
2682 if (!error) {
2683 path_put(&root_parent);
2684 path_put(&parent_path);
2685 }
2686out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002687 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002688out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002689 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002691 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694}
2695
2696static void __init init_mount_tree(void)
2697{
2698 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002699 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002700 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
2702 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2703 if (IS_ERR(mnt))
2704 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002705
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002706 ns = create_mnt_ns(mnt);
2707 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002710 init_task.nsproxy->mnt_ns = ns;
2711 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
Jan Blunckac748a02008-02-14 19:34:39 -08002713 root.mnt = ns->root;
2714 root.dentry = ns->root->mnt_root;
2715
2716 set_fs_pwd(current->fs, &root);
2717 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718}
2719
Denis Cheng74bf17c2007-10-16 23:26:30 -07002720void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002722 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002723 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
Ram Pai390c6842005-11-07 17:17:51 -05002725 init_rwsem(&namespace_sem);
2726
Al Viro7d6fec42011-11-23 12:14:10 -05002727 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002728 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
Ram Paib58fed82005-11-07 17:16:09 -05002730 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731
2732 if (!mount_hashtable)
2733 panic("Failed to allocate mount hash table\n");
2734
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002735 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
Eric Dumazet13f14b42008-02-06 01:37:57 -08002737 for (u = 0; u < HASH_SIZE; u++)
2738 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739
Nick Piggin99b7db72010-08-18 04:37:39 +10002740 br_lock_init(vfsmount_lock);
2741
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002742 err = sysfs_init();
2743 if (err)
2744 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002745 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002746 fs_kobj = kobject_create_and_add("fs", NULL);
2747 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002748 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 init_rootfs();
2750 init_mount_tree();
2751}
2752
Trond Myklebust616511d2009-06-22 15:09:13 -04002753void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002755 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002756
Al Virod498b252010-02-05 02:21:06 -05002757 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002758 return;
Ram Pai390c6842005-11-07 17:17:51 -05002759 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002760 br_write_lock(vfsmount_lock);
Al Viro761d5c32011-11-24 21:07:43 -05002761 umount_tree(real_mount(ns->root), 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002762 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002763 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002764 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002765 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766}
Al Viro9d412a42011-03-17 22:08:28 -04002767
2768struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2769{
Tim Chen423e0ab2011-07-19 09:32:38 -07002770 struct vfsmount *mnt;
2771 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2772 if (!IS_ERR(mnt)) {
2773 /*
2774 * it is a longterm mount, don't release mnt until
2775 * we unmount before file sys is unregistered
2776 */
2777 mnt_make_longterm(mnt);
2778 }
2779 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002780}
2781EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002782
2783void kern_unmount(struct vfsmount *mnt)
2784{
2785 /* release long term mount so mount point can be released */
2786 if (!IS_ERR_OR_NULL(mnt)) {
2787 mnt_make_shortterm(mnt);
2788 mntput(mnt);
2789 }
2790}
2791EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002792
2793bool our_mnt(struct vfsmount *mnt)
2794{
2795 return check_mnt(mnt);
2796}