blob: 6d3a1963879b0f13fd4929195a19c958d9deb4bb [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 Howells07f3f05c2006-09-30 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 */
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010081static int mnt_alloc_id(struct vfsmount *mnt)
82{
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 Virof21f6222009-06-24 03:12:00 -040088 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
89 if (!res)
90 mnt_id_start = 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
98static void mnt_free_id(struct vfsmount *mnt)
99{
Al Virof21f6222009-06-24 03:12:00 -0400100 int id = 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 */
113static int mnt_alloc_group_id(struct vfsmount *mnt)
114{
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,
122 &mnt->mnt_group_id);
123 if (!res)
124 mnt_group_start = mnt->mnt_group_id + 1;
125
126 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100127}
128
129/*
130 * Release a peer group ID
131 */
132void mnt_release_group_id(struct vfsmount *mnt)
133{
Al Virof21f6222009-06-24 03:12:00 -0400134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100138 mnt->mnt_group_id = 0;
139}
140
Nick Pigginb3e19d92011-01-07 17:50:11 +1100141/*
142 * vfsmount lock must be held for read
143 */
144static inline void mnt_add_count(struct vfsmount *mnt, int n)
145{
146#ifdef CONFIG_SMP
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
148#else
149 preempt_disable();
150 mnt->mnt_count += n;
151 preempt_enable();
152#endif
153}
154
155static inline void mnt_set_count(struct vfsmount *mnt, int n)
156{
157#ifdef CONFIG_SMP
158 this_cpu_write(mnt->mnt_pcp->mnt_count, n);
159#else
160 mnt->mnt_count = n;
161#endif
162}
163
164/*
165 * vfsmount lock must be held for read
166 */
167static inline void mnt_inc_count(struct vfsmount *mnt)
168{
169 mnt_add_count(mnt, 1);
170}
171
172/*
173 * vfsmount lock must be held for read
174 */
175static inline void mnt_dec_count(struct vfsmount *mnt)
176{
177 mnt_add_count(mnt, -1);
178}
179
180/*
181 * vfsmount lock must be held for write
182 */
183unsigned int mnt_get_count(struct vfsmount *mnt)
184{
185#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -0500186 unsigned int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100187 int cpu;
188
189 for_each_possible_cpu(cpu) {
190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191 }
192
193 return count;
194#else
195 return mnt->mnt_count;
196#endif
197}
198
Al Viro9d412a42011-03-17 22:08:28 -0400199static struct vfsmount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100203 int err;
204
205 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800206 if (err)
207 goto out_free_cache;
208
209 if (name) {
210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211 if (!mnt->mnt_devname)
212 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100213 }
214
Nick Pigginb3e19d92011-01-07 17:50:11 +1100215#ifdef CONFIG_SMP
216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 if (!mnt->mnt_pcp)
218 goto out_free_devname;
219
Al Virof03c6592011-01-14 22:30:21 -0500220 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100221#else
222 mnt->mnt_count = 1;
223 mnt->mnt_writers = 0;
224#endif
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 INIT_LIST_HEAD(&mnt->mnt_hash);
227 INIT_LIST_HEAD(&mnt->mnt_child);
228 INIT_LIST_HEAD(&mnt->mnt_mounts);
229 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700230 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500231 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500232 INIT_LIST_HEAD(&mnt->mnt_slave_list);
233 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500234#ifdef CONFIG_FSNOTIFY
235 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
236#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 }
238 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800239
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000240#ifdef CONFIG_SMP
241out_free_devname:
242 kfree(mnt->mnt_devname);
243#endif
Li Zefan88b38782008-07-21 18:06:36 +0800244out_free_id:
245 mnt_free_id(mnt);
246out_free_cache:
247 kmem_cache_free(mnt_cache, mnt);
248 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
Dave Hansen83660252008-02-15 14:37:30 -0800251/*
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
257 * a filesystem.
258 */
Dave Hansen3d733632008-02-15 14:37:59 -0800259/*
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
262 *
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
268 * r/w.
269 */
270int __mnt_is_readonly(struct vfsmount *mnt)
271{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800272 if (mnt->mnt_flags & MNT_READONLY)
273 return 1;
274 if (mnt->mnt_sb->s_flags & MS_RDONLY)
275 return 1;
276 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800277}
278EXPORT_SYMBOL_GPL(__mnt_is_readonly);
279
Nick Pigginc6653a82011-01-07 17:50:10 +1100280static inline void mnt_inc_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800281{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000282#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000284#else
285 mnt->mnt_writers++;
286#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800287}
Dave Hansen3d733632008-02-15 14:37:59 -0800288
Nick Pigginc6653a82011-01-07 17:50:10 +1100289static inline void mnt_dec_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800290{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000291#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293#else
294 mnt->mnt_writers--;
295#endif
296}
297
Nick Pigginc6653a82011-01-07 17:50:10 +1100298static unsigned int mnt_get_writers(struct vfsmount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000299{
300#ifdef CONFIG_SMP
301 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800302 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800303
304 for_each_possible_cpu(cpu) {
Nick Pigginb3e19d92011-01-07 17:50:11 +1100305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800306 }
Dave Hansen3d733632008-02-15 14:37:59 -0800307
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000308 return count;
309#else
310 return mnt->mnt_writers;
311#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800312}
313
314/*
315 * Most r/o checks on a fs are for operations that take
316 * discrete amounts of time, like a write() or unlink().
317 * We must keep track of when those operations start
318 * (for permission checks) and when they end, so that
319 * we can determine when writes are able to occur to
320 * a filesystem.
321 */
Dave Hansen83660252008-02-15 14:37:30 -0800322/**
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
325 *
326 * This tells the low-level filesystem that a write is
327 * about to be performed to it, and makes sure that
328 * writes are allowed before returning success. When
329 * the write operation is finished, mnt_drop_write()
330 * must be called. This is effectively a refcount.
331 */
332int mnt_want_write(struct vfsmount *mnt)
333{
Dave Hansen3d733632008-02-15 14:37:59 -0800334 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800335
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000336 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100337 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000338 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100339 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000340 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
341 * incremented count after it has set MNT_WRITE_HOLD.
342 */
343 smp_mb();
344 while (mnt->mnt_flags & MNT_WRITE_HOLD)
345 cpu_relax();
346 /*
347 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
348 * be set to match its requirements. So we must not load that until
349 * MNT_WRITE_HOLD is cleared.
350 */
351 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800352 if (__mnt_is_readonly(mnt)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100353 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800354 ret = -EROFS;
355 goto out;
356 }
Dave Hansen3d733632008-02-15 14:37:59 -0800357out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000358 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800359 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800360}
361EXPORT_SYMBOL_GPL(mnt_want_write);
362
363/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
366 *
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
371 *
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
374 */
375int mnt_clone_write(struct vfsmount *mnt)
376{
377 /* superblock may be r/o */
378 if (__mnt_is_readonly(mnt))
379 return -EROFS;
380 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100381 mnt_inc_writers(mnt);
npiggin@suse.de96029c42009-04-26 20:25:55 +1000382 preempt_enable();
383 return 0;
384}
385EXPORT_SYMBOL_GPL(mnt_clone_write);
386
387/**
388 * mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
390 *
391 * This is like mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
393 */
394int mnt_want_write_file(struct file *file)
395{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700396 struct inode *inode = file->f_dentry->d_inode;
397 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000398 return mnt_want_write(file->f_path.mnt);
399 else
400 return mnt_clone_write(file->f_path.mnt);
401}
402EXPORT_SYMBOL_GPL(mnt_want_write_file);
403
404/**
Dave Hansen83660252008-02-15 14:37:30 -0800405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
407 *
408 * Tells the low-level filesystem that we are done
409 * performing writes to it. Must be matched with
410 * mnt_want_write() call above.
411 */
412void mnt_drop_write(struct vfsmount *mnt)
413{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000414 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100415 mnt_dec_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000416 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800417}
418EXPORT_SYMBOL_GPL(mnt_drop_write);
419
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800420static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800421{
Dave Hansen3d733632008-02-15 14:37:59 -0800422 int ret = 0;
423
Nick Piggin99b7db72010-08-18 04:37:39 +1000424 br_write_lock(vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000425 mnt->mnt_flags |= MNT_WRITE_HOLD;
426 /*
427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428 * should be visible before we do.
429 */
430 smp_mb();
431
432 /*
433 * With writers on hold, if this value is zero, then there are
434 * definitely no active writers (although held writers may subsequently
435 * increment the count, they'll have to wait, and decrement it after
436 * seeing MNT_READONLY).
437 *
438 * It is OK to have counter incremented on one CPU and decremented on
439 * another: the sum will add up correctly. The danger would be when we
440 * sum up each counter, if we read a counter before it is incremented,
441 * but then read another CPU's count which it has been subsequently
442 * decremented from -- we would see more decrements than we should.
443 * MNT_WRITE_HOLD protects against this scenario, because
444 * mnt_want_write first increments count, then smp_mb, then spins on
445 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
446 * we're counting up here.
447 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100448 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000449 ret = -EBUSY;
450 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800451 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000452 /*
453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454 * that become unheld will see MNT_READONLY.
455 */
456 smp_wmb();
457 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Nick Piggin99b7db72010-08-18 04:37:39 +1000458 br_write_unlock(vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800459 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800460}
Dave Hansen83660252008-02-15 14:37:30 -0800461
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800462static void __mnt_unmake_readonly(struct vfsmount *mnt)
463{
Nick Piggin99b7db72010-08-18 04:37:39 +1000464 br_write_lock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800465 mnt->mnt_flags &= ~MNT_READONLY;
Nick Piggin99b7db72010-08-18 04:37:39 +1000466 br_write_unlock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800467}
468
Al Viro9d412a42011-03-17 22:08:28 -0400469static void free_vfsmnt(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470{
471 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100472 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000473#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100474 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000475#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 kmem_cache_free(mnt_cache, mnt);
477}
478
479/*
Ram Paia05964f2005-11-07 17:20:17 -0500480 * find the first or last mount at @dentry on vfsmount @mnt depending on
481 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000482 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
Ram Paia05964f2005-11-07 17:20:17 -0500484struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
485 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486{
Ram Paib58fed82005-11-07 17:16:09 -0500487 struct list_head *head = mount_hashtable + hash(mnt, dentry);
488 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 struct vfsmount *p, *found = NULL;
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500492 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 p = NULL;
494 if (tmp == head)
495 break;
496 p = list_entry(tmp, struct vfsmount, mnt_hash);
497 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500498 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 break;
500 }
501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return found;
503}
504
Ram Paia05964f2005-11-07 17:20:17 -0500505/*
506 * lookup_mnt increments the ref count before returning
507 * the vfsmount struct.
508 */
Al Viro1c755af2009-04-18 14:06:57 -0400509struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500510{
511 struct vfsmount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000512
513 br_read_lock(vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400514 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500515 mntget(child_mnt);
Nick Piggin99b7db72010-08-18 04:37:39 +1000516 br_read_unlock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500517 return child_mnt;
518}
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520static inline int check_mnt(struct vfsmount *mnt)
521{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800522 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523}
524
Nick Piggin99b7db72010-08-18 04:37:39 +1000525/*
526 * vfsmount lock must be held for write
527 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800528static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500529{
530 if (ns) {
531 ns->event = ++event;
532 wake_up_interruptible(&ns->poll);
533 }
534}
535
Nick Piggin99b7db72010-08-18 04:37:39 +1000536/*
537 * vfsmount lock must be held for write
538 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800539static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500540{
541 if (ns && ns->event != event) {
542 ns->event = event;
543 wake_up_interruptible(&ns->poll);
544 }
545}
546
Nick Piggin99b7db72010-08-18 04:37:39 +1000547/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100548 * Clear dentry's mounted state if it has no remaining mounts.
549 * vfsmount_lock must be held for write.
550 */
551static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
552{
553 unsigned u;
554
555 for (u = 0; u < HASH_SIZE; u++) {
556 struct vfsmount *p;
557
558 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
559 if (p->mnt_mountpoint == dentry)
560 return;
561 }
562 }
563 spin_lock(&dentry->d_lock);
564 dentry->d_flags &= ~DCACHE_MOUNTED;
565 spin_unlock(&dentry->d_lock);
566}
567
568/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000569 * vfsmount lock must be held for write
570 */
Al Viro1a390682008-03-21 20:48:19 -0400571static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Al Viro1a390682008-03-21 20:48:19 -0400573 old_path->dentry = mnt->mnt_mountpoint;
574 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 mnt->mnt_parent = mnt;
576 mnt->mnt_mountpoint = mnt->mnt_root;
577 list_del_init(&mnt->mnt_child);
578 list_del_init(&mnt->mnt_hash);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100579 dentry_reset_mounted(old_path->mnt, old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
Nick Piggin99b7db72010-08-18 04:37:39 +1000582/*
583 * vfsmount lock must be held for write
584 */
Ram Paib90fa9a2005-11-07 17:19:50 -0500585void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
586 struct vfsmount *child_mnt)
587{
588 child_mnt->mnt_parent = mntget(mnt);
589 child_mnt->mnt_mountpoint = dget(dentry);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100590 spin_lock(&dentry->d_lock);
591 dentry->d_flags |= DCACHE_MOUNTED;
592 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500593}
594
Nick Piggin99b7db72010-08-18 04:37:39 +1000595/*
596 * vfsmount lock must be held for write
597 */
Al Viro1a390682008-03-21 20:48:19 -0400598static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599{
Al Viro1a390682008-03-21 20:48:19 -0400600 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500601 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400602 hash(path->mnt, path->dentry));
603 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500604}
605
Al Viro7e3d0eb02011-01-16 16:32:11 -0500606static inline void __mnt_make_longterm(struct vfsmount *mnt)
607{
608#ifdef CONFIG_SMP
609 atomic_inc(&mnt->mnt_longterm);
610#endif
611}
612
613/* needs vfsmount lock for write */
614static inline void __mnt_make_shortterm(struct vfsmount *mnt)
615{
616#ifdef CONFIG_SMP
617 atomic_dec(&mnt->mnt_longterm);
618#endif
619}
620
Ram Paib90fa9a2005-11-07 17:19:50 -0500621/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000622 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500623 */
624static void commit_tree(struct vfsmount *mnt)
625{
626 struct vfsmount *parent = mnt->mnt_parent;
627 struct vfsmount *m;
628 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800629 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500630
631 BUG_ON(parent == mnt);
632
633 list_add_tail(&head, &mnt->mnt_list);
Al Virof03c6592011-01-14 22:30:21 -0500634 list_for_each_entry(m, &head, mnt_list) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800635 m->mnt_ns = n;
Al Viro7e3d0eb02011-01-16 16:32:11 -0500636 __mnt_make_longterm(m);
Al Virof03c6592011-01-14 22:30:21 -0500637 }
638
Ram Paib90fa9a2005-11-07 17:19:50 -0500639 list_splice(&head, n->list.prev);
640
641 list_add_tail(&mnt->mnt_hash, mount_hashtable +
642 hash(parent, mnt->mnt_mountpoint));
643 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800644 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
648{
649 struct list_head *next = p->mnt_mounts.next;
650 if (next == &p->mnt_mounts) {
651 while (1) {
652 if (p == root)
653 return NULL;
654 next = p->mnt_child.next;
655 if (next != &p->mnt_parent->mnt_mounts)
656 break;
657 p = p->mnt_parent;
658 }
659 }
660 return list_entry(next, struct vfsmount, mnt_child);
661}
662
Ram Pai9676f0c2005-11-07 17:21:20 -0500663static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
664{
665 struct list_head *prev = p->mnt_mounts.prev;
666 while (prev != &p->mnt_mounts) {
667 p = list_entry(prev, struct vfsmount, mnt_child);
668 prev = p->mnt_mounts.prev;
669 }
670 return p;
671}
672
Al Viro9d412a42011-03-17 22:08:28 -0400673struct vfsmount *
674vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
675{
676 struct vfsmount *mnt;
677 struct dentry *root;
678
679 if (!type)
680 return ERR_PTR(-ENODEV);
681
682 mnt = alloc_vfsmnt(name);
683 if (!mnt)
684 return ERR_PTR(-ENOMEM);
685
686 if (flags & MS_KERNMOUNT)
687 mnt->mnt_flags = MNT_INTERNAL;
688
689 root = mount_fs(type, flags, name, data);
690 if (IS_ERR(root)) {
691 free_vfsmnt(mnt);
692 return ERR_CAST(root);
693 }
694
695 mnt->mnt_root = root;
696 mnt->mnt_sb = root->d_sb;
697 mnt->mnt_mountpoint = mnt->mnt_root;
698 mnt->mnt_parent = mnt;
699 return mnt;
700}
701EXPORT_SYMBOL_GPL(vfs_kern_mount);
702
Ram Pai36341f62005-11-07 17:17:22 -0500703static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
704 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
706 struct super_block *sb = old->mnt_sb;
707 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
708
709 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100710 if (flag & (CL_SLAVE | CL_PRIVATE))
711 mnt->mnt_group_id = 0; /* not a peer of original */
712 else
713 mnt->mnt_group_id = old->mnt_group_id;
714
715 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
716 int err = mnt_alloc_group_id(mnt);
717 if (err)
718 goto out_free;
719 }
720
Miklos Szeredibe1a16a2010-10-05 12:31:09 +0200721 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 atomic_inc(&sb->s_active);
723 mnt->mnt_sb = sb;
724 mnt->mnt_root = dget(root);
725 mnt->mnt_mountpoint = mnt->mnt_root;
726 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500727
Ram Pai5afe0022005-11-07 17:21:01 -0500728 if (flag & CL_SLAVE) {
729 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
730 mnt->mnt_master = old;
731 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400732 } else if (!(flag & CL_PRIVATE)) {
Al Viro796a6b52010-01-16 13:28:47 -0500733 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500734 list_add(&mnt->mnt_share, &old->mnt_share);
735 if (IS_MNT_SLAVE(old))
736 list_add(&mnt->mnt_slave, &old->mnt_slave);
737 mnt->mnt_master = old->mnt_master;
738 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500739 if (flag & CL_MAKE_SHARED)
740 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
742 /* stick the duplicate mount on the same expiry list
743 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500744 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500745 if (!list_empty(&old->mnt_expire))
746 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500747 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100750
751 out_free:
752 free_vfsmnt(mnt);
753 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754}
755
Nick Pigginb3e19d92011-01-07 17:50:11 +1100756static inline void mntfree(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
758 struct super_block *sb = mnt->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100759
Dave Hansen3d733632008-02-15 14:37:59 -0800760 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800761 * This probably indicates that somebody messed
762 * up a mnt_want/drop_write() pair. If this
763 * happens, the filesystem was probably unable
764 * to make r/w->r/o transitions.
765 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000766 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100767 * The locking used to deal with mnt_count decrement provides barriers,
768 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000769 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100770 WARN_ON(mnt_get_writers(mnt));
Andreas Gruenbacherca9c7262009-12-17 21:24:27 -0500771 fsnotify_vfsmount_delete(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 dput(mnt->mnt_root);
773 free_vfsmnt(mnt);
774 deactivate_super(sb);
775}
776
Al Virof03c6592011-01-14 22:30:21 -0500777static void mntput_no_expire(struct vfsmount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500778{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100779put_again:
Al Virof03c6592011-01-14 22:30:21 -0500780#ifdef CONFIG_SMP
781 br_read_lock(vfsmount_lock);
782 if (likely(atomic_read(&mnt->mnt_longterm))) {
783 mnt_dec_count(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100784 br_read_unlock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500785 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100786 }
Al Virof03c6592011-01-14 22:30:21 -0500787 br_read_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100788
789 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500790 mnt_dec_count(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100791 if (mnt_get_count(mnt)) {
792 br_write_unlock(vfsmount_lock);
793 return;
794 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100795#else
Nick Pigginb3e19d92011-01-07 17:50:11 +1100796 mnt_dec_count(mnt);
797 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000798 return;
799 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500800#endif
Nick Pigginb3e19d92011-01-07 17:50:11 +1100801 if (unlikely(mnt->mnt_pinned)) {
802 mnt_add_count(mnt, mnt->mnt_pinned + 1);
803 mnt->mnt_pinned = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +1000804 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100805 acct_auto_close_mnt(mnt);
806 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500807 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000808 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100809 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500810}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100811
812void mntput(struct vfsmount *mnt)
813{
814 if (mnt) {
815 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
816 if (unlikely(mnt->mnt_expiry_mark))
817 mnt->mnt_expiry_mark = 0;
Al Virof03c6592011-01-14 22:30:21 -0500818 mntput_no_expire(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100819 }
820}
821EXPORT_SYMBOL(mntput);
822
823struct vfsmount *mntget(struct vfsmount *mnt)
824{
825 if (mnt)
826 mnt_inc_count(mnt);
827 return mnt;
828}
829EXPORT_SYMBOL(mntget);
830
Al Viro7b7b1ac2005-11-07 17:13:39 -0500831void mnt_pin(struct vfsmount *mnt)
832{
Nick Piggin99b7db72010-08-18 04:37:39 +1000833 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500834 mnt->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000835 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500836}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500837EXPORT_SYMBOL(mnt_pin);
838
839void mnt_unpin(struct vfsmount *mnt)
840{
Nick Piggin99b7db72010-08-18 04:37:39 +1000841 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500842 if (mnt->mnt_pinned) {
Nick Pigginb3e19d92011-01-07 17:50:11 +1100843 mnt_inc_count(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500844 mnt->mnt_pinned--;
845 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000846 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500847}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500848EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800850static inline void mangle(struct seq_file *m, const char *s)
851{
852 seq_escape(m, s, " \t\n\\");
853}
854
855/*
856 * Simple .show_options callback for filesystems which don't want to
857 * implement more complex mount option showing.
858 *
859 * See also save_mount_options().
860 */
861int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
862{
Al Viro2a32ceb2009-05-08 16:05:57 -0400863 const char *options;
864
865 rcu_read_lock();
866 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800867
868 if (options != NULL && options[0]) {
869 seq_putc(m, ',');
870 mangle(m, options);
871 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400872 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800873
874 return 0;
875}
876EXPORT_SYMBOL(generic_show_options);
877
878/*
879 * If filesystem uses generic_show_options(), this function should be
880 * called from the fill_super() callback.
881 *
882 * The .remount_fs callback usually needs to be handled in a special
883 * way, to make sure, that previous options are not overwritten if the
884 * remount fails.
885 *
886 * Also note, that if the filesystem's .remount_fs function doesn't
887 * reset all options to their default value, but changes only newly
888 * given options, then the displayed options will not reflect reality
889 * any more.
890 */
891void save_mount_options(struct super_block *sb, char *options)
892{
Al Viro2a32ceb2009-05-08 16:05:57 -0400893 BUG_ON(sb->s_options);
894 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800895}
896EXPORT_SYMBOL(save_mount_options);
897
Al Viro2a32ceb2009-05-08 16:05:57 -0400898void replace_mount_options(struct super_block *sb, char *options)
899{
900 char *old = sb->s_options;
901 rcu_assign_pointer(sb->s_options, options);
902 if (old) {
903 synchronize_rcu();
904 kfree(old);
905 }
906}
907EXPORT_SYMBOL(replace_mount_options);
908
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100909#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910/* iterator */
911static void *m_start(struct seq_file *m, loff_t *pos)
912{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100913 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
Ram Pai390c6842005-11-07 17:17:51 -0500915 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100916 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917}
918
919static void *m_next(struct seq_file *m, void *v, loff_t *pos)
920{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100921 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700922
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100923 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
926static void m_stop(struct seq_file *m, void *v)
927{
Ram Pai390c6842005-11-07 17:17:51 -0500928 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Al Viro9f5596a2010-02-05 00:40:25 -0500931int mnt_had_events(struct proc_mounts *p)
932{
933 struct mnt_namespace *ns = p->ns;
934 int res = 0;
935
Nick Piggin99b7db72010-08-18 04:37:39 +1000936 br_read_lock(vfsmount_lock);
Kay Sieversf1514632011-07-12 20:48:39 +0200937 if (p->m.poll_event != ns->event) {
938 p->m.poll_event = ns->event;
Al Viro9f5596a2010-02-05 00:40:25 -0500939 res = 1;
940 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000941 br_read_unlock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500942
943 return res;
944}
945
Ram Pai2d4d4862008-03-27 13:06:25 +0100946struct proc_fs_info {
947 int flag;
948 const char *str;
949};
950
Eric Paris2069f452008-07-04 09:47:13 +1000951static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952{
Ram Pai2d4d4862008-03-27 13:06:25 +0100953 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 { MS_SYNCHRONOUS, ",sync" },
955 { MS_DIRSYNC, ",dirsync" },
956 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 { 0, NULL }
958 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100959 const struct proc_fs_info *fs_infop;
960
961 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
962 if (sb->s_flags & fs_infop->flag)
963 seq_puts(m, fs_infop->str);
964 }
Eric Paris2069f452008-07-04 09:47:13 +1000965
966 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100967}
968
969static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
970{
971 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 { MNT_NOSUID, ",nosuid" },
973 { MNT_NODEV, ",nodev" },
974 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800975 { MNT_NOATIME, ",noatime" },
976 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800977 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 { 0, NULL }
979 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100980 const struct proc_fs_info *fs_infop;
981
982 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
983 if (mnt->mnt_flags & fs_infop->flag)
984 seq_puts(m, fs_infop->str);
985 }
986}
987
988static void show_type(struct seq_file *m, struct super_block *sb)
989{
990 mangle(m, sb->s_type->name);
991 if (sb->s_subtype && sb->s_subtype[0]) {
992 seq_putc(m, '.');
993 mangle(m, sb->s_subtype);
994 }
995}
996
997static int show_vfsmnt(struct seq_file *m, void *v)
998{
999 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1000 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001001 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Al Viroc7f404b2011-03-16 06:59:40 -04001003 if (mnt->mnt_sb->s_op->show_devname) {
1004 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1005 if (err)
1006 goto out;
1007 } else {
1008 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -08001011 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +01001013 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001014 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001015 err = show_sb_opts(m, mnt->mnt_sb);
1016 if (err)
1017 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001018 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (mnt->mnt_sb->s_op->show_options)
1020 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1021 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +10001022out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 return err;
1024}
1025
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001026const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 .start = m_start,
1028 .next = m_next,
1029 .stop = m_stop,
1030 .show = show_vfsmnt
1031};
1032
Ram Pai2d4d4862008-03-27 13:06:25 +01001033static int show_mountinfo(struct seq_file *m, void *v)
1034{
1035 struct proc_mounts *p = m->private;
1036 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1037 struct super_block *sb = mnt->mnt_sb;
1038 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1039 struct path root = p->root;
1040 int err = 0;
1041
1042 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1043 MAJOR(sb->s_dev), MINOR(sb->s_dev));
Al Viroc7f404b2011-03-16 06:59:40 -04001044 if (sb->s_op->show_path)
1045 err = sb->s_op->show_path(m, mnt);
1046 else
1047 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1048 if (err)
1049 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001050 seq_putc(m, ' ');
1051 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1052 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1053 /*
1054 * Mountpoint is outside root, discard that one. Ugly,
1055 * but less so than trying to do that in iterator in a
1056 * race-free way (due to renames).
1057 */
1058 return SEQ_SKIP;
1059 }
1060 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1061 show_mnt_opts(m, mnt);
1062
1063 /* Tagged fields ("foo:X" or "bar") */
1064 if (IS_MNT_SHARED(mnt))
1065 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +01001066 if (IS_MNT_SLAVE(mnt)) {
1067 int master = mnt->mnt_master->mnt_group_id;
1068 int dom = get_dominating_id(mnt, &p->root);
1069 seq_printf(m, " master:%i", master);
1070 if (dom && dom != master)
1071 seq_printf(m, " propagate_from:%i", dom);
1072 }
Ram Pai2d4d4862008-03-27 13:06:25 +01001073 if (IS_MNT_UNBINDABLE(mnt))
1074 seq_puts(m, " unbindable");
1075
1076 /* Filesystem specific data */
1077 seq_puts(m, " - ");
1078 show_type(m, sb);
1079 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001080 if (sb->s_op->show_devname)
1081 err = sb->s_op->show_devname(m, mnt);
1082 else
1083 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084 if (err)
1085 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001086 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001087 err = show_sb_opts(m, sb);
1088 if (err)
1089 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001090 if (sb->s_op->show_options)
1091 err = sb->s_op->show_options(m, mnt);
1092 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +10001093out:
Ram Pai2d4d4862008-03-27 13:06:25 +01001094 return err;
1095}
1096
1097const struct seq_operations mountinfo_op = {
1098 .start = m_start,
1099 .next = m_next,
1100 .stop = m_stop,
1101 .show = show_mountinfo,
1102};
1103
Chuck Leverb4629fe2006-03-20 13:44:12 -05001104static int show_vfsstat(struct seq_file *m, void *v)
1105{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001106 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -08001107 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -05001108 int err = 0;
1109
1110 /* device */
Al Viroc7f404b2011-03-16 06:59:40 -04001111 if (mnt->mnt_sb->s_op->show_devname) {
Bryan Schumakera877ee02011-10-07 13:41:15 -04001112 seq_puts(m, "device ");
Al Viroc7f404b2011-03-16 06:59:40 -04001113 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1114 } else {
1115 if (mnt->mnt_devname) {
1116 seq_puts(m, "device ");
1117 mangle(m, mnt->mnt_devname);
1118 } else
1119 seq_puts(m, "no device");
1120 }
Chuck Leverb4629fe2006-03-20 13:44:12 -05001121
1122 /* mount point */
1123 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08001124 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -05001125 seq_putc(m, ' ');
1126
1127 /* file system type */
1128 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +01001129 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001130
1131 /* optional statistics */
1132 if (mnt->mnt_sb->s_op->show_stats) {
1133 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001134 if (!err)
1135 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001136 }
1137
1138 seq_putc(m, '\n');
1139 return err;
1140}
1141
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001142const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -05001143 .start = m_start,
1144 .next = m_next,
1145 .stop = m_stop,
1146 .show = show_vfsstat,
1147};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001148#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001149
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150/**
1151 * may_umount_tree - check if a mount tree is busy
1152 * @mnt: root of mount tree
1153 *
1154 * This is called to check if a tree of mounts has any
1155 * open files, pwds, chroots or sub mounts that are
1156 * busy.
1157 */
1158int may_umount_tree(struct vfsmount *mnt)
1159{
Ram Pai36341f62005-11-07 17:17:22 -05001160 int actual_refs = 0;
1161 int minimum_refs = 0;
1162 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Nick Pigginb3e19d92011-01-07 17:50:11 +11001164 /* write lock needed for mnt_get_count */
1165 br_write_lock(vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -05001166 for (p = mnt; p; p = next_mnt(p, mnt)) {
Nick Pigginb3e19d92011-01-07 17:50:11 +11001167 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
Nick Pigginb3e19d92011-01-07 17:50:11 +11001170 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001173 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Ian Kente3474a82006-03-27 01:14:51 -08001175 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
1177
1178EXPORT_SYMBOL(may_umount_tree);
1179
1180/**
1181 * may_umount - check if a mount point is busy
1182 * @mnt: root of mount
1183 *
1184 * This is called to check if a mount point has any
1185 * open files, pwds, chroots or sub mounts. If the
1186 * mount has sub mounts this will return busy
1187 * regardless of whether the sub mounts are busy.
1188 *
1189 * Doesn't take quota and stuff into account. IOW, in some cases it will
1190 * give false negatives. The main reason why it's here is that we need
1191 * a non-destructive way to look for easily umountable filesystems.
1192 */
1193int may_umount(struct vfsmount *mnt)
1194{
Ian Kente3474a82006-03-27 01:14:51 -08001195 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001196 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001197 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001198 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -08001199 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001200 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001201 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001202 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
1205EXPORT_SYMBOL(may_umount);
1206
Ram Paib90fa9a2005-11-07 17:19:50 -05001207void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Ram Pai70fbcdf2005-11-07 17:17:04 -05001209 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001210 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001211 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001212 list_del_init(&mnt->mnt_hash);
1213 if (mnt->mnt_parent != mnt) {
1214 struct dentry *dentry;
1215 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001216
1217 br_write_lock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001218 dentry = mnt->mnt_mountpoint;
1219 m = mnt->mnt_parent;
1220 mnt->mnt_mountpoint = mnt->mnt_root;
1221 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001222 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001223 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001224 dput(dentry);
1225 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
Al Virof03c6592011-01-14 22:30:21 -05001227 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001228 }
1229}
1230
Nick Piggin99b7db72010-08-18 04:37:39 +10001231/*
1232 * vfsmount lock must be held for write
1233 * namespace_sem must be held for write
1234 */
Ram Paia05964f2005-11-07 17:20:17 -05001235void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001236{
Al Viro7b8a53f2011-01-15 20:08:44 -05001237 LIST_HEAD(tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001238 struct vfsmount *p;
1239
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001240 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro7b8a53f2011-01-15 20:08:44 -05001241 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001242
Ram Paia05964f2005-11-07 17:20:17 -05001243 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001244 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001245
Al Viro7b8a53f2011-01-15 20:08:44 -05001246 list_for_each_entry(p, &tmp_list, mnt_hash) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001247 list_del_init(&p->mnt_expire);
1248 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001249 __touch_mnt_namespace(p->mnt_ns);
1250 p->mnt_ns = NULL;
Al Viro7e3d0eb02011-01-16 16:32:11 -05001251 __mnt_make_shortterm(p);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001252 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001253 if (p->mnt_parent != p) {
1254 p->mnt_parent->mnt_ghosts++;
Nick Piggin5f57cbc2011-01-07 17:49:54 +11001255 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001256 }
Ram Paia05964f2005-11-07 17:20:17 -05001257 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001259 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
Al Viroc35038b2008-03-22 00:46:23 -04001262static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264static int do_umount(struct vfsmount *mnt, int flags)
1265{
Ram Paib58fed82005-11-07 17:16:09 -05001266 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001268 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 retval = security_sb_umount(mnt, flags);
1271 if (retval)
1272 return retval;
1273
1274 /*
1275 * Allow userspace to request a mountpoint be expired rather than
1276 * unmounting unconditionally. Unmount only happens if:
1277 * (1) the mark is already set (the mark is cleared by mntput())
1278 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1279 */
1280 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001281 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 flags & (MNT_FORCE | MNT_DETACH))
1283 return -EINVAL;
1284
Nick Pigginb3e19d92011-01-07 17:50:11 +11001285 /*
1286 * probably don't strictly need the lock here if we examined
1287 * all race cases, but it's a slowpath.
1288 */
1289 br_write_lock(vfsmount_lock);
1290 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001291 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001293 }
1294 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 if (!xchg(&mnt->mnt_expiry_mark, 1))
1297 return -EAGAIN;
1298 }
1299
1300 /*
1301 * If we may have to abort operations to get out of this
1302 * mount, and they will themselves hold resources we must
1303 * allow the fs to do things. In the Unix tradition of
1304 * 'Gee thats tricky lets do it in userspace' the umount_begin
1305 * might fail to complete on the first run through as other tasks
1306 * must return, and the like. Thats for the mount program to worry
1307 * about for the moment.
1308 */
1309
Al Viro42faad92008-04-24 07:21:56 -04001310 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001311 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001312 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 /*
1315 * No sense to grab the lock for this test, but test itself looks
1316 * somewhat bogus. Suggestions for better replacement?
1317 * Ho-hum... In principle, we might treat that as umount + switch
1318 * to rootfs. GC would eventually take care of the old vfsmount.
1319 * Actually it makes sense, especially if rootfs would contain a
1320 * /reboot - static binary that would close all descriptors and
1321 * call reboot(9). Then init(8) could umount root and exec /reboot.
1322 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001323 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 /*
1325 * Special case for "unmounting" root ...
1326 * we just try to remount it readonly.
1327 */
1328 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001329 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 up_write(&sb->s_umount);
1332 return retval;
1333 }
1334
Ram Pai390c6842005-11-07 17:17:51 -05001335 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001336 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001337 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
Al Viroc35038b2008-03-22 00:46:23 -04001339 if (!(flags & MNT_DETACH))
1340 shrink_submounts(mnt, &umount_list);
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001343 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001345 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 retval = 0;
1347 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001348 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001349 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001350 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return retval;
1352}
1353
1354/*
1355 * Now umount can handle mount points as well as block devices.
1356 * This is important for filesystems which use unnamed block devices.
1357 *
1358 * We now support a flag for forced unmount like the other 'big iron'
1359 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1360 */
1361
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001362SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
Al Viro2d8f3032008-07-22 09:59:21 -04001364 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001366 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001368 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1369 return -EINVAL;
1370
1371 if (!(flags & UMOUNT_NOFOLLOW))
1372 lookup_flags |= LOOKUP_FOLLOW;
1373
1374 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 if (retval)
1376 goto out;
1377 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001378 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001380 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 goto dput_and_out;
1382
1383 retval = -EPERM;
1384 if (!capable(CAP_SYS_ADMIN))
1385 goto dput_and_out;
1386
Al Viro2d8f3032008-07-22 09:59:21 -04001387 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001389 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001390 dput(path.dentry);
1391 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392out:
1393 return retval;
1394}
1395
1396#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1397
1398/*
Ram Paib58fed82005-11-07 17:16:09 -05001399 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001401SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402{
Ram Paib58fed82005-11-07 17:16:09 -05001403 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404}
1405
1406#endif
1407
Al Viro2d92ab32008-08-02 00:51:11 -04001408static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
1410 if (capable(CAP_SYS_ADMIN))
1411 return 0;
1412 return -EPERM;
1413#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001414 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001416 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001417 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 return -EPERM;
1419 }
Al Viro2d92ab32008-08-02 00:51:11 -04001420 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return -EPERM;
1422 return 0;
1423#endif
1424}
1425
Ram Paib90fa9a2005-11-07 17:19:50 -05001426struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001427 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
1429 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001430 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Ram Pai9676f0c2005-11-07 17:21:20 -05001432 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1433 return NULL;
1434
Ram Pai36341f62005-11-07 17:17:22 -05001435 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (!q)
1437 goto Enomem;
1438 q->mnt_mountpoint = mnt->mnt_mountpoint;
1439
1440 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001441 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001442 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 continue;
1444
1445 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001446 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1447 s = skip_mnt_tree(s);
1448 continue;
1449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 while (p != s->mnt_parent) {
1451 p = p->mnt_parent;
1452 q = q->mnt_parent;
1453 }
1454 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001455 path.mnt = q;
1456 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001457 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 if (!q)
1459 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001460 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001462 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001463 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 }
1465 }
1466 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001467Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001469 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001470 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001471 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001472 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001473 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
1475 return NULL;
1476}
1477
Al Viro589ff872009-04-18 03:28:19 -04001478struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001479{
1480 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001481 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001482 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001483 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001484 return tree;
1485}
1486
1487void drop_collected_mounts(struct vfsmount *mnt)
1488{
1489 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001490 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001491 br_write_lock(vfsmount_lock);
Al Viro8aec0802007-06-07 12:20:32 -04001492 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001493 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001494 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001495 release_mounts(&umount_list);
1496}
1497
Al Viro1f707132010-01-30 22:51:25 -05001498int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1499 struct vfsmount *root)
1500{
1501 struct vfsmount *mnt;
1502 int res = f(root, arg);
1503 if (res)
1504 return res;
1505 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1506 res = f(mnt, arg);
1507 if (res)
1508 return res;
1509 }
1510 return 0;
1511}
1512
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001513static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1514{
1515 struct vfsmount *p;
1516
1517 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1518 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1519 mnt_release_group_id(p);
1520 }
1521}
1522
1523static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1524{
1525 struct vfsmount *p;
1526
1527 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1528 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1529 int err = mnt_alloc_group_id(p);
1530 if (err) {
1531 cleanup_group_ids(mnt, p);
1532 return err;
1533 }
1534 }
1535 }
1536
1537 return 0;
1538}
1539
Ram Paib90fa9a2005-11-07 17:19:50 -05001540/*
1541 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001542 * @nd : place the mount tree @source_mnt is attached
1543 * @parent_nd : if non-null, detach the source_mnt from its parent and
1544 * store the parent mount and mountpoint dentry.
1545 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001546 *
1547 * NOTE: in the table below explains the semantics when a source mount
1548 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001549 * ---------------------------------------------------------------------------
1550 * | BIND MOUNT OPERATION |
1551 * |**************************************************************************
1552 * | source-->| shared | private | slave | unbindable |
1553 * | dest | | | | |
1554 * | | | | | | |
1555 * | v | | | | |
1556 * |**************************************************************************
1557 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1558 * | | | | | |
1559 * |non-shared| shared (+) | private | slave (*) | invalid |
1560 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001561 * A bind operation clones the source mount and mounts the clone on the
1562 * destination mount.
1563 *
1564 * (++) the cloned mount is propagated to all the mounts in the propagation
1565 * tree of the destination mount and the cloned mount is added to
1566 * the peer group of the source mount.
1567 * (+) the cloned mount is created under the destination mount and is marked
1568 * as shared. The cloned mount is added to the peer group of the source
1569 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001570 * (+++) the mount is propagated to all the mounts in the propagation tree
1571 * of the destination mount and the cloned mount is made slave
1572 * of the same master as that of the source mount. The cloned mount
1573 * is marked as 'shared and slave'.
1574 * (*) the cloned mount is made a slave of the same master as that of the
1575 * source mount.
1576 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001577 * ---------------------------------------------------------------------------
1578 * | MOVE MOUNT OPERATION |
1579 * |**************************************************************************
1580 * | source-->| shared | private | slave | unbindable |
1581 * | dest | | | | |
1582 * | | | | | | |
1583 * | v | | | | |
1584 * |**************************************************************************
1585 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1586 * | | | | | |
1587 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1588 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001589 *
1590 * (+) the mount is moved to the destination. And is then propagated to
1591 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001592 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001593 * (+++) the mount is moved to the destination and is then propagated to
1594 * all the mounts belonging to the destination mount's propagation tree.
1595 * the mount is marked as 'shared and slave'.
1596 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001597 *
1598 * if the source mount is a tree, the operations explained above is
1599 * applied to each mount in the tree.
1600 * Must be called without spinlocks held, since this function can sleep
1601 * in allocations.
1602 */
1603static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001604 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001605{
1606 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001607 struct vfsmount *dest_mnt = path->mnt;
1608 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001609 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001610 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001611
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001612 if (IS_MNT_SHARED(dest_mnt)) {
1613 err = invent_group_ids(source_mnt, true);
1614 if (err)
1615 goto out;
1616 }
1617 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1618 if (err)
1619 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001620
Nick Piggin99b7db72010-08-18 04:37:39 +10001621 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001622
Ram Paib90fa9a2005-11-07 17:19:50 -05001623 if (IS_MNT_SHARED(dest_mnt)) {
1624 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1625 set_mnt_shared(p);
1626 }
Al Viro1a390682008-03-21 20:48:19 -04001627 if (parent_path) {
1628 detach_mnt(source_mnt, parent_path);
1629 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001630 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001631 } else {
1632 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1633 commit_tree(source_mnt);
1634 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001635
1636 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1637 list_del_init(&child->mnt_hash);
1638 commit_tree(child);
1639 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001640 br_write_unlock(vfsmount_lock);
1641
Ram Paib90fa9a2005-11-07 17:19:50 -05001642 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001643
1644 out_cleanup_ids:
1645 if (IS_MNT_SHARED(dest_mnt))
1646 cleanup_group_ids(source_mnt, NULL);
1647 out:
1648 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001649}
1650
Al Virob12cea92011-03-18 08:55:38 -04001651static int lock_mount(struct path *path)
1652{
1653 struct vfsmount *mnt;
1654retry:
1655 mutex_lock(&path->dentry->d_inode->i_mutex);
1656 if (unlikely(cant_mount(path->dentry))) {
1657 mutex_unlock(&path->dentry->d_inode->i_mutex);
1658 return -ENOENT;
1659 }
1660 down_write(&namespace_sem);
1661 mnt = lookup_mnt(path);
1662 if (likely(!mnt))
1663 return 0;
1664 up_write(&namespace_sem);
1665 mutex_unlock(&path->dentry->d_inode->i_mutex);
1666 path_put(path);
1667 path->mnt = mnt;
1668 path->dentry = dget(mnt->mnt_root);
1669 goto retry;
1670}
1671
1672static void unlock_mount(struct path *path)
1673{
1674 up_write(&namespace_sem);
1675 mutex_unlock(&path->dentry->d_inode->i_mutex);
1676}
1677
Al Viro8c3ee422008-03-22 18:00:39 -04001678static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1681 return -EINVAL;
1682
Al Viro8c3ee422008-03-22 18:00:39 -04001683 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1685 return -ENOTDIR;
1686
Al Virob12cea92011-03-18 08:55:38 -04001687 if (d_unlinked(path->dentry))
1688 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
Al Virob12cea92011-03-18 08:55:38 -04001690 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691}
1692
1693/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001694 * Sanity check the flags to change_mnt_propagation.
1695 */
1696
1697static int flags_to_propagation_type(int flags)
1698{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001699 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001700
1701 /* Fail if any non-propagation flags are set */
1702 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1703 return 0;
1704 /* Only one propagation flag should be set */
1705 if (!is_power_of_2(type))
1706 return 0;
1707 return type;
1708}
1709
1710/*
Ram Pai07b20882005-11-07 17:19:07 -05001711 * recursively change the type of the mountpoint.
1712 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001713static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001714{
Al Viro2d92ab32008-08-02 00:51:11 -04001715 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001716 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001717 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001718 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001719
Miklos Szerediee6f9582007-05-08 00:30:40 -07001720 if (!capable(CAP_SYS_ADMIN))
1721 return -EPERM;
1722
Al Viro2d92ab32008-08-02 00:51:11 -04001723 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001724 return -EINVAL;
1725
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001726 type = flags_to_propagation_type(flag);
1727 if (!type)
1728 return -EINVAL;
1729
Ram Pai07b20882005-11-07 17:19:07 -05001730 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001731 if (type == MS_SHARED) {
1732 err = invent_group_ids(mnt, recurse);
1733 if (err)
1734 goto out_unlock;
1735 }
1736
Nick Piggin99b7db72010-08-18 04:37:39 +10001737 br_write_lock(vfsmount_lock);
Ram Pai07b20882005-11-07 17:19:07 -05001738 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1739 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001740 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001741
1742 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001743 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001744 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001745}
1746
1747/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 * do loopback mount.
1749 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001750static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001751 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Al Virob12cea92011-03-18 08:55:38 -04001753 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001754 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001756 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (err)
1758 return err;
1759 if (!old_name || !*old_name)
1760 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001761 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 if (err)
1763 return err;
1764
Al Virob12cea92011-03-18 08:55:38 -04001765 err = lock_mount(path);
1766 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001767 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001768
Al Virob12cea92011-03-18 08:55:38 -04001769 err = -EINVAL;
1770 if (IS_MNT_UNBINDABLE(old_path.mnt))
1771 goto out2;
1772
Al Viro2d92ab32008-08-02 00:51:11 -04001773 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Virob12cea92011-03-18 08:55:38 -04001774 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
Al Viroccd48bc2005-11-07 17:15:04 -05001776 err = -ENOMEM;
1777 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001778 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001779 else
Al Viro2d92ab32008-08-02 00:51:11 -04001780 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001781
1782 if (!mnt)
Al Virob12cea92011-03-18 08:55:38 -04001783 goto out2;
Al Viroccd48bc2005-11-07 17:15:04 -05001784
Al Viro2d92ab32008-08-02 00:51:11 -04001785 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001786 if (err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001787 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001788 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001789 br_write_unlock(vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001790 }
Al Virob12cea92011-03-18 08:55:38 -04001791out2:
1792 unlock_mount(path);
1793 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001794out:
Al Viro2d92ab32008-08-02 00:51:11 -04001795 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 return err;
1797}
1798
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001799static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1800{
1801 int error = 0;
1802 int readonly_request = 0;
1803
1804 if (ms_flags & MS_RDONLY)
1805 readonly_request = 1;
1806 if (readonly_request == __mnt_is_readonly(mnt))
1807 return 0;
1808
1809 if (readonly_request)
1810 error = mnt_make_readonly(mnt);
1811 else
1812 __mnt_unmake_readonly(mnt);
1813 return error;
1814}
1815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816/*
1817 * change filesystem flags. dir should be a physical root of filesystem.
1818 * If you've mounted a non-root directory somewhere and want to do remount
1819 * on it - tough luck.
1820 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001821static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 void *data)
1823{
1824 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001825 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 if (!capable(CAP_SYS_ADMIN))
1828 return -EPERM;
1829
Al Viro2d92ab32008-08-02 00:51:11 -04001830 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 return -EINVAL;
1832
Al Viro2d92ab32008-08-02 00:51:11 -04001833 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return -EINVAL;
1835
Eric Parisff36fe22011-03-03 16:09:14 -05001836 err = security_sb_remount(sb, data);
1837 if (err)
1838 return err;
1839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001841 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001842 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001843 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001844 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001845 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001846 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001847 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001848 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001849 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001850 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001852 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001853 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001854 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001855 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 return err;
1858}
1859
Ram Pai9676f0c2005-11-07 17:21:20 -05001860static inline int tree_contains_unbindable(struct vfsmount *mnt)
1861{
1862 struct vfsmount *p;
1863 for (p = mnt; p; p = next_mnt(p, mnt)) {
1864 if (IS_MNT_UNBINDABLE(p))
1865 return 1;
1866 }
1867 return 0;
1868}
1869
Al Viro0a0d8a42008-08-02 00:55:27 -04001870static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
Al Viro2d92ab32008-08-02 00:51:11 -04001872 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 struct vfsmount *p;
1874 int err = 0;
1875 if (!capable(CAP_SYS_ADMIN))
1876 return -EPERM;
1877 if (!old_name || !*old_name)
1878 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001879 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 if (err)
1881 return err;
1882
Al Virob12cea92011-03-18 08:55:38 -04001883 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001884 if (err < 0)
1885 goto out;
1886
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001888 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 goto out1;
1890
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001891 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001892 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001895 if (old_path.dentry != old_path.mnt->mnt_root)
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 (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001899 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
Al Viro2d92ab32008-08-02 00:51:11 -04001901 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1902 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001903 goto out1;
1904 /*
1905 * Don't move a mount residing in a shared parent.
1906 */
Al Viro2d92ab32008-08-02 00:51:11 -04001907 if (old_path.mnt->mnt_parent &&
1908 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001909 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001910 /*
1911 * Don't move a mount tree containing unbindable mounts to a destination
1912 * mount which is shared.
1913 */
Al Viro2d92ab32008-08-02 00:51:11 -04001914 if (IS_MNT_SHARED(path->mnt) &&
1915 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001916 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001918 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1919 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001920 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
Al Viro2d92ab32008-08-02 00:51:11 -04001922 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001923 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001924 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
1926 /* if the mount is moved, it should no longer be expire
1927 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001928 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929out1:
Al Virob12cea92011-03-18 08:55:38 -04001930 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001933 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001934 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return err;
1936}
1937
Al Viro9d412a42011-03-17 22:08:28 -04001938static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1939{
1940 int err;
1941 const char *subtype = strchr(fstype, '.');
1942 if (subtype) {
1943 subtype++;
1944 err = -EINVAL;
1945 if (!subtype[0])
1946 goto err;
1947 } else
1948 subtype = "";
1949
1950 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1951 err = -ENOMEM;
1952 if (!mnt->mnt_sb->s_subtype)
1953 goto err;
1954 return mnt;
1955
1956 err:
1957 mntput(mnt);
1958 return ERR_PTR(err);
1959}
1960
1961struct vfsmount *
1962do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1963{
1964 struct file_system_type *type = get_fs_type(fstype);
1965 struct vfsmount *mnt;
1966 if (!type)
1967 return ERR_PTR(-ENODEV);
1968 mnt = vfs_kern_mount(type, flags, name, data);
1969 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1970 !mnt->mnt_sb->s_subtype)
1971 mnt = fs_set_subtype(mnt, fstype);
1972 put_filesystem(type);
1973 return mnt;
1974}
1975EXPORT_SYMBOL_GPL(do_kern_mount);
1976
1977/*
1978 * add a mount into a namespace's mount tree
1979 */
1980static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1981{
1982 int err;
1983
1984 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1985
Al Virob12cea92011-03-18 08:55:38 -04001986 err = lock_mount(path);
1987 if (err)
1988 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001989
1990 err = -EINVAL;
1991 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1992 goto unlock;
1993
1994 /* Refuse the same filesystem on the same mount point */
1995 err = -EBUSY;
1996 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1997 path->mnt->mnt_root == path->dentry)
1998 goto unlock;
1999
2000 err = -EINVAL;
2001 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2002 goto unlock;
2003
2004 newmnt->mnt_flags = mnt_flags;
2005 err = graft_tree(newmnt, path);
2006
2007unlock:
Al Virob12cea92011-03-18 08:55:38 -04002008 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04002009 return err;
2010}
Al Virob1e75df2011-01-17 01:47:59 -05002011
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012/*
2013 * create a new mount for userspace and request it to be added into the
2014 * namespace's tree
2015 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002016static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 int mnt_flags, char *name, void *data)
2018{
2019 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002020 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Vegard Nossumeca6f532009-09-18 13:05:45 -07002022 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 return -EINVAL;
2024
2025 /* we need capabilities... */
2026 if (!capable(CAP_SYS_ADMIN))
2027 return -EPERM;
2028
2029 mnt = do_kern_mount(type, flags, name, data);
2030 if (IS_ERR(mnt))
2031 return PTR_ERR(mnt);
2032
Al Viro15f9a3f2011-01-17 01:41:58 -05002033 err = do_add_mount(mnt, path, mnt_flags);
2034 if (err)
2035 mntput(mnt);
2036 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037}
2038
Al Viro19a167a2011-01-17 01:35:23 -05002039int finish_automount(struct vfsmount *m, struct path *path)
2040{
2041 int err;
2042 /* The new mount record should have at least 2 refs to prevent it being
2043 * expired before we get a chance to add it
2044 */
2045 BUG_ON(mnt_get_count(m) < 2);
2046
2047 if (m->mnt_sb == path->mnt->mnt_sb &&
2048 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002049 err = -ELOOP;
2050 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002051 }
2052
Al Viro19a167a2011-01-17 01:35:23 -05002053 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002054 if (!err)
2055 return 0;
2056fail:
2057 /* remove m from any expiration list it may be on */
2058 if (!list_empty(&m->mnt_expire)) {
2059 down_write(&namespace_sem);
2060 br_write_lock(vfsmount_lock);
2061 list_del_init(&m->mnt_expire);
2062 br_write_unlock(vfsmount_lock);
2063 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05002064 }
Al Virob1e75df2011-01-17 01:47:59 -05002065 mntput(m);
2066 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002067 return err;
2068}
2069
David Howellsea5b7782011-01-14 19:10:03 +00002070/**
2071 * mnt_set_expiry - Put a mount on an expiration list
2072 * @mnt: The mount to list.
2073 * @expiry_list: The list to add the mount to.
2074 */
2075void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2076{
2077 down_write(&namespace_sem);
2078 br_write_lock(vfsmount_lock);
2079
2080 list_add_tail(&mnt->mnt_expire, expiry_list);
2081
2082 br_write_unlock(vfsmount_lock);
2083 up_write(&namespace_sem);
2084}
2085EXPORT_SYMBOL(mnt_set_expiry);
2086
2087/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 * process a list of expirable mountpoints with the intent of discarding any
2089 * mountpoints that aren't in use and haven't been touched since last we came
2090 * here
2091 */
2092void mark_mounts_for_expiry(struct list_head *mounts)
2093{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 struct vfsmount *mnt, *next;
2095 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002096 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 if (list_empty(mounts))
2099 return;
2100
Al Virobcc5c7d2008-03-22 00:21:53 -04002101 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002102 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
2104 /* extract from the expiration list every vfsmount that matches the
2105 * following criteria:
2106 * - only referenced by its parent vfsmount
2107 * - still marked for expiry (marked on the last call here; marks are
2108 * cleared by mntput())
2109 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002110 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04002112 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002114 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002116 while (!list_empty(&graveyard)) {
2117 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2118 touch_mnt_namespace(mnt->mnt_ns);
2119 umount_tree(mnt, 1, &umounts);
2120 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002121 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002122 up_write(&namespace_sem);
2123
2124 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125}
2126
2127EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2128
2129/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002130 * Ripoff of 'select_parent()'
2131 *
2132 * search the list of submounts for a given mountpoint, and move any
2133 * shrinkable submounts to the 'graveyard' list.
2134 */
2135static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2136{
2137 struct vfsmount *this_parent = parent;
2138 struct list_head *next;
2139 int found = 0;
2140
2141repeat:
2142 next = this_parent->mnt_mounts.next;
2143resume:
2144 while (next != &this_parent->mnt_mounts) {
2145 struct list_head *tmp = next;
2146 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2147
2148 next = tmp->next;
2149 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2150 continue;
2151 /*
2152 * Descend a level if the d_mounts list is non-empty.
2153 */
2154 if (!list_empty(&mnt->mnt_mounts)) {
2155 this_parent = mnt;
2156 goto repeat;
2157 }
2158
2159 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002160 list_move_tail(&mnt->mnt_expire, graveyard);
2161 found++;
2162 }
2163 }
2164 /*
2165 * All done at this level ... ascend and resume the search
2166 */
2167 if (this_parent != parent) {
2168 next = this_parent->mnt_child.next;
2169 this_parent = this_parent->mnt_parent;
2170 goto resume;
2171 }
2172 return found;
2173}
2174
2175/*
2176 * process a list of expirable mountpoints with the intent of discarding any
2177 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002178 *
2179 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002180 */
Al Viroc35038b2008-03-22 00:46:23 -04002181static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002182{
2183 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04002184 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002185
Trond Myklebust5528f9112006-06-09 09:34:17 -04002186 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002187 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002188 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04002189 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04002190 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002191 touch_mnt_namespace(m->mnt_ns);
2192 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002193 }
2194 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002195}
2196
Trond Myklebust5528f9112006-06-09 09:34:17 -04002197/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 * Some copy_from_user() implementations do not return the exact number of
2199 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2200 * Note that this function differs from copy_from_user() in that it will oops
2201 * on bad values of `to', rather than returning a short copy.
2202 */
Ram Paib58fed82005-11-07 17:16:09 -05002203static long exact_copy_from_user(void *to, const void __user * from,
2204 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
2206 char *t = to;
2207 const char __user *f = from;
2208 char c;
2209
2210 if (!access_ok(VERIFY_READ, from, n))
2211 return n;
2212
2213 while (n) {
2214 if (__get_user(c, f)) {
2215 memset(t, 0, n);
2216 break;
2217 }
2218 *t++ = c;
2219 f++;
2220 n--;
2221 }
2222 return n;
2223}
2224
Ram Paib58fed82005-11-07 17:16:09 -05002225int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
2227 int i;
2228 unsigned long page;
2229 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 *where = 0;
2232 if (!data)
2233 return 0;
2234
2235 if (!(page = __get_free_page(GFP_KERNEL)))
2236 return -ENOMEM;
2237
2238 /* We only care that *some* data at the address the user
2239 * gave us is valid. Just in case, we'll zero
2240 * the remainder of the page.
2241 */
2242 /* copy_from_user cannot cross TASK_SIZE ! */
2243 size = TASK_SIZE - (unsigned long)data;
2244 if (size > PAGE_SIZE)
2245 size = PAGE_SIZE;
2246
2247 i = size - exact_copy_from_user((void *)page, data, size);
2248 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002249 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 return -EFAULT;
2251 }
2252 if (i != PAGE_SIZE)
2253 memset((char *)page + i, 0, PAGE_SIZE - i);
2254 *where = page;
2255 return 0;
2256}
2257
Vegard Nossumeca6f532009-09-18 13:05:45 -07002258int copy_mount_string(const void __user *data, char **where)
2259{
2260 char *tmp;
2261
2262 if (!data) {
2263 *where = NULL;
2264 return 0;
2265 }
2266
2267 tmp = strndup_user(data, PAGE_SIZE);
2268 if (IS_ERR(tmp))
2269 return PTR_ERR(tmp);
2270
2271 *where = tmp;
2272 return 0;
2273}
2274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275/*
2276 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2277 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2278 *
2279 * data is a (void *) that can point to any structure up to
2280 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2281 * information (or be NULL).
2282 *
2283 * Pre-0.97 versions of mount() didn't have a flags word.
2284 * When the flags word was introduced its top half was required
2285 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2286 * Therefore, if this magic number is present, it carries no information
2287 * and must be discarded.
2288 */
Ram Paib58fed82005-11-07 17:16:09 -05002289long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 unsigned long flags, void *data_page)
2291{
Al Viro2d92ab32008-08-02 00:51:11 -04002292 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 int retval = 0;
2294 int mnt_flags = 0;
2295
2296 /* Discard magic */
2297 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2298 flags &= ~MS_MGC_MSK;
2299
2300 /* Basic sanity checks */
2301
2302 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2303 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 if (data_page)
2306 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2307
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002308 /* ... and get the mountpoint */
2309 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2310 if (retval)
2311 return retval;
2312
2313 retval = security_sb_mount(dev_name, &path,
2314 type_page, flags, data_page);
2315 if (retval)
2316 goto dput_out;
2317
Andi Kleen613cbe32009-04-19 18:40:43 +02002318 /* Default to relatime unless overriden */
2319 if (!(flags & MS_NOATIME))
2320 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 /* Separate the per-mountpoint flags */
2323 if (flags & MS_NOSUID)
2324 mnt_flags |= MNT_NOSUID;
2325 if (flags & MS_NODEV)
2326 mnt_flags |= MNT_NODEV;
2327 if (flags & MS_NOEXEC)
2328 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002329 if (flags & MS_NOATIME)
2330 mnt_flags |= MNT_NOATIME;
2331 if (flags & MS_NODIRATIME)
2332 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002333 if (flags & MS_STRICTATIME)
2334 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002335 if (flags & MS_RDONLY)
2336 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002337
Al Viro7a4dec52010-08-09 12:05:43 -04002338 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002339 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2340 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002343 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 data_page);
2345 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002346 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002347 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002348 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002350 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 else
Al Viro2d92ab32008-08-02 00:51:11 -04002352 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 dev_name, data_page);
2354dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002355 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 return retval;
2357}
2358
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002359static struct mnt_namespace *alloc_mnt_ns(void)
2360{
2361 struct mnt_namespace *new_ns;
2362
2363 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2364 if (!new_ns)
2365 return ERR_PTR(-ENOMEM);
2366 atomic_set(&new_ns->count, 1);
2367 new_ns->root = NULL;
2368 INIT_LIST_HEAD(&new_ns->list);
2369 init_waitqueue_head(&new_ns->poll);
2370 new_ns->event = 0;
2371 return new_ns;
2372}
2373
Al Virof03c6592011-01-14 22:30:21 -05002374void mnt_make_longterm(struct vfsmount *mnt)
2375{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002376 __mnt_make_longterm(mnt);
Al Virof03c6592011-01-14 22:30:21 -05002377}
2378
2379void mnt_make_shortterm(struct vfsmount *mnt)
2380{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002381#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -05002382 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2383 return;
2384 br_write_lock(vfsmount_lock);
2385 atomic_dec(&mnt->mnt_longterm);
2386 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002387#endif
Al Virof03c6592011-01-14 22:30:21 -05002388}
2389
JANAK DESAI741a2952006-02-07 12:59:00 -08002390/*
2391 * Allocate a new namespace structure and populate it with contents
2392 * copied from the namespace of the passed in task structure.
2393 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002394static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002395 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002397 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002398 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 struct vfsmount *p, *q;
2400
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002401 new_ns = alloc_mnt_ns();
2402 if (IS_ERR(new_ns))
2403 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
Ram Pai390c6842005-11-07 17:17:51 -05002405 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002407 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002408 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002410 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002412 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002414 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002416 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417
2418 /*
2419 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2420 * as belonging to new namespace. We have already acquired a private
2421 * fs_struct, so tsk->fs->lock is not needed.
2422 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002423 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 q = new_ns->root;
2425 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002426 q->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002427 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002429 if (p == fs->root.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002430 fs->root.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002431 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002432 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 rootmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002435 if (p == fs->pwd.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002436 fs->pwd.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002437 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002438 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 pwdmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002442 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 q = next_mnt(q, new_ns->root);
2444 }
Ram Pai390c6842005-11-07 17:17:51 -05002445 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002448 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002450 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
JANAK DESAI741a2952006-02-07 12:59:00 -08002452 return new_ns;
2453}
2454
Eric W. Biederman213dd262007-07-15 23:41:15 -07002455struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002456 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002457{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002458 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002459
Badari Pulavartye3222c42007-05-08 00:25:21 -07002460 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002461 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002462
2463 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002464 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002465
Badari Pulavartye3222c42007-05-08 00:25:21 -07002466 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002467
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002468 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002469 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470}
2471
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002472/**
2473 * create_mnt_ns - creates a private namespace and adds a root filesystem
2474 * @mnt: pointer to the new root filesystem mountpoint
2475 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002476struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002477{
2478 struct mnt_namespace *new_ns;
2479
2480 new_ns = alloc_mnt_ns();
2481 if (!IS_ERR(new_ns)) {
2482 mnt->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002483 __mnt_make_longterm(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002484 new_ns->root = mnt;
2485 list_add(&new_ns->list, &new_ns->root->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002486 } else {
2487 mntput(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002488 }
2489 return new_ns;
2490}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002491EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002492
Al Viroea441d12011-11-16 21:43:59 -05002493struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2494{
2495 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002496 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002497 struct path path;
2498 int err;
2499
2500 ns = create_mnt_ns(mnt);
2501 if (IS_ERR(ns))
2502 return ERR_CAST(ns);
2503
2504 err = vfs_path_lookup(mnt->mnt_root, mnt,
2505 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2506
2507 put_mnt_ns(ns);
2508
2509 if (err)
2510 return ERR_PTR(err);
2511
2512 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002513 s = path.mnt->mnt_sb;
2514 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002515 mntput(path.mnt);
2516 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002517 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002518 /* ... and return the root of (sub)tree on it */
2519 return path.dentry;
2520}
2521EXPORT_SYMBOL(mount_subtree);
2522
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002523SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2524 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002526 int ret;
2527 char *kernel_type;
2528 char *kernel_dir;
2529 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Vegard Nossumeca6f532009-09-18 13:05:45 -07002532 ret = copy_mount_string(type, &kernel_type);
2533 if (ret < 0)
2534 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535
Vegard Nossumeca6f532009-09-18 13:05:45 -07002536 kernel_dir = getname(dir_name);
2537 if (IS_ERR(kernel_dir)) {
2538 ret = PTR_ERR(kernel_dir);
2539 goto out_dir;
2540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
Vegard Nossumeca6f532009-09-18 13:05:45 -07002542 ret = copy_mount_string(dev_name, &kernel_dev);
2543 if (ret < 0)
2544 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Vegard Nossumeca6f532009-09-18 13:05:45 -07002546 ret = copy_mount_options(data, &data_page);
2547 if (ret < 0)
2548 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549
Vegard Nossumeca6f532009-09-18 13:05:45 -07002550 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2551 (void *) data_page);
2552
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002554out_data:
2555 kfree(kernel_dev);
2556out_dev:
2557 putname(kernel_dir);
2558out_dir:
2559 kfree(kernel_type);
2560out_type:
2561 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562}
2563
2564/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 * pivot_root Semantics:
2566 * Moves the root file system of the current process to the directory put_old,
2567 * makes new_root as the new root file system of the current process, and sets
2568 * root/cwd of all processes which had them on the current root to new_root.
2569 *
2570 * Restrictions:
2571 * The new_root and put_old must be directories, and must not be on the
2572 * same file system as the current process root. The put_old must be
2573 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2574 * pointed to by put_old must yield the same directory as new_root. No other
2575 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2576 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002577 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2578 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2579 * in this situation.
2580 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 * Notes:
2582 * - we don't move root/cwd if they are not at the root (reason: if something
2583 * cared enough to change them, it's probably wrong to force them elsewhere)
2584 * - it's okay to pick a root that isn't the root of a file system, e.g.
2585 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2586 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2587 * first.
2588 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002589SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2590 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591{
2592 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002593 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 int error;
2595
2596 if (!capable(CAP_SYS_ADMIN))
2597 return -EPERM;
2598
Al Viro2d8f3032008-07-22 09:59:21 -04002599 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 if (error)
2601 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
Al Viro2d8f3032008-07-22 09:59:21 -04002603 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 if (error)
2605 goto out1;
2606
Al Viro2d8f3032008-07-22 09:59:21 -04002607 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002608 if (error)
2609 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002611 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002612 error = lock_mount(&old);
2613 if (error)
2614 goto out3;
2615
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002617 if (IS_MNT_SHARED(old.mnt) ||
2618 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002619 IS_MNT_SHARED(root.mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002620 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002621 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002622 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002624 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002625 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002626 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002627 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002629 if (new.mnt == root.mnt ||
2630 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002631 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002633 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002634 goto out4; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002635 if (root.mnt->mnt_parent == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002636 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002637 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002638 goto out4; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002639 if (new.mnt->mnt_parent == new.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002640 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002641 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002642 tmp = old.mnt;
Al Viro2d8f3032008-07-22 09:59:21 -04002643 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 for (;;) {
2645 if (tmp->mnt_parent == tmp)
Al Virob12cea92011-03-18 08:55:38 -04002646 goto out4; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002647 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 break;
2649 tmp = tmp->mnt_parent;
2650 }
Al Viro2d8f3032008-07-22 09:59:21 -04002651 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002652 goto out4;
Al Viro2d8f3032008-07-22 09:59:21 -04002653 } else if (!is_subdir(old.dentry, new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002654 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002655 br_write_lock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002656 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002657 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002658 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002659 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002660 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002661 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002662 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002663 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002664 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002666out4:
2667 unlock_mount(&old);
2668 if (!error) {
2669 path_put(&root_parent);
2670 path_put(&parent_path);
2671 }
2672out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002673 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002674out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002675 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002677 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680}
2681
2682static void __init init_mount_tree(void)
2683{
2684 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002685 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002686 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
2688 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2689 if (IS_ERR(mnt))
2690 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002691
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002692 ns = create_mnt_ns(mnt);
2693 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002696 init_task.nsproxy->mnt_ns = ns;
2697 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
Jan Blunckac748a02008-02-14 19:34:39 -08002699 root.mnt = ns->root;
2700 root.dentry = ns->root->mnt_root;
2701
2702 set_fs_pwd(current->fs, &root);
2703 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704}
2705
Denis Cheng74bf17c2007-10-16 23:26:30 -07002706void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002708 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002709 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Ram Pai390c6842005-11-07 17:17:51 -05002711 init_rwsem(&namespace_sem);
2712
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002714 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Ram Paib58fed82005-11-07 17:16:09 -05002716 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
2718 if (!mount_hashtable)
2719 panic("Failed to allocate mount hash table\n");
2720
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002721 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Eric Dumazet13f14b42008-02-06 01:37:57 -08002723 for (u = 0; u < HASH_SIZE; u++)
2724 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Nick Piggin99b7db72010-08-18 04:37:39 +10002726 br_lock_init(vfsmount_lock);
2727
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002728 err = sysfs_init();
2729 if (err)
2730 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002731 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002732 fs_kobj = kobject_create_and_add("fs", NULL);
2733 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002734 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 init_rootfs();
2736 init_mount_tree();
2737}
2738
Trond Myklebust616511d2009-06-22 15:09:13 -04002739void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002741 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002742
Al Virod498b252010-02-05 02:21:06 -05002743 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002744 return;
Ram Pai390c6842005-11-07 17:17:51 -05002745 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002746 br_write_lock(vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002747 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002748 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002749 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002750 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002751 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002753EXPORT_SYMBOL(put_mnt_ns);
Al Viro9d412a42011-03-17 22:08:28 -04002754
2755struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2756{
Tim Chen423e0ab2011-07-19 09:32:38 -07002757 struct vfsmount *mnt;
2758 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2759 if (!IS_ERR(mnt)) {
2760 /*
2761 * it is a longterm mount, don't release mnt until
2762 * we unmount before file sys is unregistered
2763 */
2764 mnt_make_longterm(mnt);
2765 }
2766 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002767}
2768EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002769
2770void kern_unmount(struct vfsmount *mnt)
2771{
2772 /* release long term mount so mount point can be released */
2773 if (!IS_ERR_OR_NULL(mnt)) {
2774 mnt_make_shortterm(mnt);
2775 mntput(mnt);
2776 }
2777}
2778EXPORT_SYMBOL(kern_unmount);