blob: cfc6d4448aa54bdc538131b1e53285a92bb20073 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
Nick Piggin99b7db72010-08-18 04:37:39 +100014#include <linux/spinlock.h>
15#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070017#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080020#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070022#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080024#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/namei.h>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040026#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/security.h>
28#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080030#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010031#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040032#include <linux/fs_struct.h>
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -050033#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/uaccess.h>
35#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050036#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric Dumazet13f14b42008-02-06 01:37:57 -080039#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
Al Viro5addc5d2005-11-07 17:15:49 -050042static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010043static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010044static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100045static DEFINE_SPINLOCK(mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040046static int mnt_id_start = 0;
47static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Eric Dumazetfa3536c2006-03-26 01:37:24 -080049static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050051static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060054struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080056
Nick Piggin99b7db72010-08-18 04:37:39 +100057/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
Ram Paib58fed82005-11-07 17:16:09 -050069 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080071 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073}
74
Dave Hansen3d733632008-02-15 14:37:59 -080075#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
Nick Piggin99b7db72010-08-18 04:37:39 +100077/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
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, ' ');
Al Viro02125a82011-12-05 08:43:34 -05001051
1052 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1053 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1054 if (err)
1055 goto out;
1056
Ram Pai2d4d4862008-03-27 13:06:25 +01001057 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1058 show_mnt_opts(m, mnt);
1059
1060 /* Tagged fields ("foo:X" or "bar") */
1061 if (IS_MNT_SHARED(mnt))
1062 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +01001063 if (IS_MNT_SLAVE(mnt)) {
1064 int master = mnt->mnt_master->mnt_group_id;
1065 int dom = get_dominating_id(mnt, &p->root);
1066 seq_printf(m, " master:%i", master);
1067 if (dom && dom != master)
1068 seq_printf(m, " propagate_from:%i", dom);
1069 }
Ram Pai2d4d4862008-03-27 13:06:25 +01001070 if (IS_MNT_UNBINDABLE(mnt))
1071 seq_puts(m, " unbindable");
1072
1073 /* Filesystem specific data */
1074 seq_puts(m, " - ");
1075 show_type(m, sb);
1076 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001077 if (sb->s_op->show_devname)
1078 err = sb->s_op->show_devname(m, mnt);
1079 else
1080 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1081 if (err)
1082 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001083 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001084 err = show_sb_opts(m, sb);
1085 if (err)
1086 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001087 if (sb->s_op->show_options)
1088 err = sb->s_op->show_options(m, mnt);
1089 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +10001090out:
Ram Pai2d4d4862008-03-27 13:06:25 +01001091 return err;
1092}
1093
1094const struct seq_operations mountinfo_op = {
1095 .start = m_start,
1096 .next = m_next,
1097 .stop = m_stop,
1098 .show = show_mountinfo,
1099};
1100
Chuck Leverb4629fe2006-03-20 13:44:12 -05001101static int show_vfsstat(struct seq_file *m, void *v)
1102{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001103 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -08001104 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -05001105 int err = 0;
1106
1107 /* device */
Al Viroc7f404b2011-03-16 06:59:40 -04001108 if (mnt->mnt_sb->s_op->show_devname) {
Bryan Schumakera877ee02011-10-07 13:41:15 -04001109 seq_puts(m, "device ");
Al Viroc7f404b2011-03-16 06:59:40 -04001110 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1111 } else {
1112 if (mnt->mnt_devname) {
1113 seq_puts(m, "device ");
1114 mangle(m, mnt->mnt_devname);
1115 } else
1116 seq_puts(m, "no device");
1117 }
Chuck Leverb4629fe2006-03-20 13:44:12 -05001118
1119 /* mount point */
1120 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08001121 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -05001122 seq_putc(m, ' ');
1123
1124 /* file system type */
1125 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +01001126 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001127
1128 /* optional statistics */
1129 if (mnt->mnt_sb->s_op->show_stats) {
1130 seq_putc(m, ' ');
Al Viroc7f404b2011-03-16 06:59:40 -04001131 if (!err)
1132 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001133 }
1134
1135 seq_putc(m, '\n');
1136 return err;
1137}
1138
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001139const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -05001140 .start = m_start,
1141 .next = m_next,
1142 .stop = m_stop,
1143 .show = show_vfsstat,
1144};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001145#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147/**
1148 * may_umount_tree - check if a mount tree is busy
1149 * @mnt: root of mount tree
1150 *
1151 * This is called to check if a tree of mounts has any
1152 * open files, pwds, chroots or sub mounts that are
1153 * busy.
1154 */
1155int may_umount_tree(struct vfsmount *mnt)
1156{
Ram Pai36341f62005-11-07 17:17:22 -05001157 int actual_refs = 0;
1158 int minimum_refs = 0;
1159 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Nick Pigginb3e19d92011-01-07 17:50:11 +11001161 /* write lock needed for mnt_get_count */
1162 br_write_lock(vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -05001163 for (p = mnt; p; p = next_mnt(p, mnt)) {
Nick Pigginb3e19d92011-01-07 17:50:11 +11001164 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Nick Pigginb3e19d92011-01-07 17:50:11 +11001167 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001170 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Ian Kente3474a82006-03-27 01:14:51 -08001172 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175EXPORT_SYMBOL(may_umount_tree);
1176
1177/**
1178 * may_umount - check if a mount point is busy
1179 * @mnt: root of mount
1180 *
1181 * This is called to check if a mount point has any
1182 * open files, pwds, chroots or sub mounts. If the
1183 * mount has sub mounts this will return busy
1184 * regardless of whether the sub mounts are busy.
1185 *
1186 * Doesn't take quota and stuff into account. IOW, in some cases it will
1187 * give false negatives. The main reason why it's here is that we need
1188 * a non-destructive way to look for easily umountable filesystems.
1189 */
1190int may_umount(struct vfsmount *mnt)
1191{
Ian Kente3474a82006-03-27 01:14:51 -08001192 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001193 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001194 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001195 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -08001196 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001197 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001198 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001199 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
1201
1202EXPORT_SYMBOL(may_umount);
1203
Ram Paib90fa9a2005-11-07 17:19:50 -05001204void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Ram Pai70fbcdf2005-11-07 17:17:04 -05001206 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001207 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001208 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001209 list_del_init(&mnt->mnt_hash);
1210 if (mnt->mnt_parent != mnt) {
1211 struct dentry *dentry;
1212 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001213
1214 br_write_lock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001215 dentry = mnt->mnt_mountpoint;
1216 m = mnt->mnt_parent;
1217 mnt->mnt_mountpoint = mnt->mnt_root;
1218 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001219 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001220 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001221 dput(dentry);
1222 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
Al Virof03c6592011-01-14 22:30:21 -05001224 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001225 }
1226}
1227
Nick Piggin99b7db72010-08-18 04:37:39 +10001228/*
1229 * vfsmount lock must be held for write
1230 * namespace_sem must be held for write
1231 */
Ram Paia05964f2005-11-07 17:20:17 -05001232void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001233{
Al Viro7b8a53f2011-01-15 20:08:44 -05001234 LIST_HEAD(tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001235 struct vfsmount *p;
1236
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001237 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro7b8a53f2011-01-15 20:08:44 -05001238 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001239
Ram Paia05964f2005-11-07 17:20:17 -05001240 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001241 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001242
Al Viro7b8a53f2011-01-15 20:08:44 -05001243 list_for_each_entry(p, &tmp_list, mnt_hash) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001244 list_del_init(&p->mnt_expire);
1245 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001246 __touch_mnt_namespace(p->mnt_ns);
1247 p->mnt_ns = NULL;
Al Viro7e3d0eb02011-01-16 16:32:11 -05001248 __mnt_make_shortterm(p);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001249 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001250 if (p->mnt_parent != p) {
1251 p->mnt_parent->mnt_ghosts++;
Nick Piggin5f57cbc2011-01-07 17:49:54 +11001252 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001253 }
Ram Paia05964f2005-11-07 17:20:17 -05001254 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001256 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
Al Viroc35038b2008-03-22 00:46:23 -04001259static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261static int do_umount(struct vfsmount *mnt, int flags)
1262{
Ram Paib58fed82005-11-07 17:16:09 -05001263 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001265 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267 retval = security_sb_umount(mnt, flags);
1268 if (retval)
1269 return retval;
1270
1271 /*
1272 * Allow userspace to request a mountpoint be expired rather than
1273 * unmounting unconditionally. Unmount only happens if:
1274 * (1) the mark is already set (the mark is cleared by mntput())
1275 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1276 */
1277 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001278 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 flags & (MNT_FORCE | MNT_DETACH))
1280 return -EINVAL;
1281
Nick Pigginb3e19d92011-01-07 17:50:11 +11001282 /*
1283 * probably don't strictly need the lock here if we examined
1284 * all race cases, but it's a slowpath.
1285 */
1286 br_write_lock(vfsmount_lock);
1287 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001288 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001290 }
1291 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
1293 if (!xchg(&mnt->mnt_expiry_mark, 1))
1294 return -EAGAIN;
1295 }
1296
1297 /*
1298 * If we may have to abort operations to get out of this
1299 * mount, and they will themselves hold resources we must
1300 * allow the fs to do things. In the Unix tradition of
1301 * 'Gee thats tricky lets do it in userspace' the umount_begin
1302 * might fail to complete on the first run through as other tasks
1303 * must return, and the like. Thats for the mount program to worry
1304 * about for the moment.
1305 */
1306
Al Viro42faad92008-04-24 07:21:56 -04001307 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001308 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 /*
1312 * No sense to grab the lock for this test, but test itself looks
1313 * somewhat bogus. Suggestions for better replacement?
1314 * Ho-hum... In principle, we might treat that as umount + switch
1315 * to rootfs. GC would eventually take care of the old vfsmount.
1316 * Actually it makes sense, especially if rootfs would contain a
1317 * /reboot - static binary that would close all descriptors and
1318 * call reboot(9). Then init(8) could umount root and exec /reboot.
1319 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001320 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 /*
1322 * Special case for "unmounting" root ...
1323 * we just try to remount it readonly.
1324 */
1325 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001326 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 up_write(&sb->s_umount);
1329 return retval;
1330 }
1331
Ram Pai390c6842005-11-07 17:17:51 -05001332 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001333 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001334 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Al Viroc35038b2008-03-22 00:46:23 -04001336 if (!(flags & MNT_DETACH))
1337 shrink_submounts(mnt, &umount_list);
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001340 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001342 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 retval = 0;
1344 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001345 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001346 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001347 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return retval;
1349}
1350
1351/*
1352 * Now umount can handle mount points as well as block devices.
1353 * This is important for filesystems which use unnamed block devices.
1354 *
1355 * We now support a flag for forced unmount like the other 'big iron'
1356 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1357 */
1358
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001359SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
Al Viro2d8f3032008-07-22 09:59:21 -04001361 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001363 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001365 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1366 return -EINVAL;
1367
1368 if (!(flags & UMOUNT_NOFOLLOW))
1369 lookup_flags |= LOOKUP_FOLLOW;
1370
1371 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (retval)
1373 goto out;
1374 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001375 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001377 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 goto dput_and_out;
1379
1380 retval = -EPERM;
1381 if (!capable(CAP_SYS_ADMIN))
1382 goto dput_and_out;
1383
Al Viro2d8f3032008-07-22 09:59:21 -04001384 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001386 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001387 dput(path.dentry);
1388 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389out:
1390 return retval;
1391}
1392
1393#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1394
1395/*
Ram Paib58fed82005-11-07 17:16:09 -05001396 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001398SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
Ram Paib58fed82005-11-07 17:16:09 -05001400 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402
1403#endif
1404
Al Viro2d92ab32008-08-02 00:51:11 -04001405static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
1407 if (capable(CAP_SYS_ADMIN))
1408 return 0;
1409 return -EPERM;
1410#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001411 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001413 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001414 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return -EPERM;
1416 }
Al Viro2d92ab32008-08-02 00:51:11 -04001417 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 return -EPERM;
1419 return 0;
1420#endif
1421}
1422
Ram Paib90fa9a2005-11-07 17:19:50 -05001423struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001424 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001427 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Ram Pai9676f0c2005-11-07 17:21:20 -05001429 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1430 return NULL;
1431
Ram Pai36341f62005-11-07 17:17:22 -05001432 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 if (!q)
1434 goto Enomem;
1435 q->mnt_mountpoint = mnt->mnt_mountpoint;
1436
1437 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001438 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001439 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 continue;
1441
1442 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001443 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1444 s = skip_mnt_tree(s);
1445 continue;
1446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 while (p != s->mnt_parent) {
1448 p = p->mnt_parent;
1449 q = q->mnt_parent;
1450 }
1451 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001452 path.mnt = q;
1453 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001454 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (!q)
1456 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001457 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001459 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001460 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 }
1462 }
1463 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001464Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001466 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001467 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001468 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001469 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001470 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 }
1472 return NULL;
1473}
1474
Al Viro589ff872009-04-18 03:28:19 -04001475struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001476{
1477 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001478 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001479 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001480 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001481 return tree;
1482}
1483
1484void drop_collected_mounts(struct vfsmount *mnt)
1485{
1486 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001487 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001488 br_write_lock(vfsmount_lock);
Al Viro8aec0802007-06-07 12:20:32 -04001489 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001490 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001491 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001492 release_mounts(&umount_list);
1493}
1494
Al Viro1f707132010-01-30 22:51:25 -05001495int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1496 struct vfsmount *root)
1497{
1498 struct vfsmount *mnt;
1499 int res = f(root, arg);
1500 if (res)
1501 return res;
1502 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1503 res = f(mnt, arg);
1504 if (res)
1505 return res;
1506 }
1507 return 0;
1508}
1509
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001510static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1511{
1512 struct vfsmount *p;
1513
1514 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1515 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1516 mnt_release_group_id(p);
1517 }
1518}
1519
1520static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1521{
1522 struct vfsmount *p;
1523
1524 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1525 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1526 int err = mnt_alloc_group_id(p);
1527 if (err) {
1528 cleanup_group_ids(mnt, p);
1529 return err;
1530 }
1531 }
1532 }
1533
1534 return 0;
1535}
1536
Ram Paib90fa9a2005-11-07 17:19:50 -05001537/*
1538 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001539 * @nd : place the mount tree @source_mnt is attached
1540 * @parent_nd : if non-null, detach the source_mnt from its parent and
1541 * store the parent mount and mountpoint dentry.
1542 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001543 *
1544 * NOTE: in the table below explains the semantics when a source mount
1545 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001546 * ---------------------------------------------------------------------------
1547 * | BIND MOUNT OPERATION |
1548 * |**************************************************************************
1549 * | source-->| shared | private | slave | unbindable |
1550 * | dest | | | | |
1551 * | | | | | | |
1552 * | v | | | | |
1553 * |**************************************************************************
1554 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1555 * | | | | | |
1556 * |non-shared| shared (+) | private | slave (*) | invalid |
1557 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001558 * A bind operation clones the source mount and mounts the clone on the
1559 * destination mount.
1560 *
1561 * (++) the cloned mount is propagated to all the mounts in the propagation
1562 * tree of the destination mount and the cloned mount is added to
1563 * the peer group of the source mount.
1564 * (+) the cloned mount is created under the destination mount and is marked
1565 * as shared. The cloned mount is added to the peer group of the source
1566 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001567 * (+++) the mount is propagated to all the mounts in the propagation tree
1568 * of the destination mount and the cloned mount is made slave
1569 * of the same master as that of the source mount. The cloned mount
1570 * is marked as 'shared and slave'.
1571 * (*) the cloned mount is made a slave of the same master as that of the
1572 * source mount.
1573 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001574 * ---------------------------------------------------------------------------
1575 * | MOVE MOUNT OPERATION |
1576 * |**************************************************************************
1577 * | source-->| shared | private | slave | unbindable |
1578 * | dest | | | | |
1579 * | | | | | | |
1580 * | v | | | | |
1581 * |**************************************************************************
1582 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1583 * | | | | | |
1584 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1585 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001586 *
1587 * (+) the mount is moved to the destination. And is then propagated to
1588 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001589 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001590 * (+++) the mount is moved to the destination and is then propagated to
1591 * all the mounts belonging to the destination mount's propagation tree.
1592 * the mount is marked as 'shared and slave'.
1593 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001594 *
1595 * if the source mount is a tree, the operations explained above is
1596 * applied to each mount in the tree.
1597 * Must be called without spinlocks held, since this function can sleep
1598 * in allocations.
1599 */
1600static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001601 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001602{
1603 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001604 struct vfsmount *dest_mnt = path->mnt;
1605 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001606 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001607 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001608
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001609 if (IS_MNT_SHARED(dest_mnt)) {
1610 err = invent_group_ids(source_mnt, true);
1611 if (err)
1612 goto out;
1613 }
1614 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1615 if (err)
1616 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001617
Nick Piggin99b7db72010-08-18 04:37:39 +10001618 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001619
Ram Paib90fa9a2005-11-07 17:19:50 -05001620 if (IS_MNT_SHARED(dest_mnt)) {
1621 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1622 set_mnt_shared(p);
1623 }
Al Viro1a390682008-03-21 20:48:19 -04001624 if (parent_path) {
1625 detach_mnt(source_mnt, parent_path);
1626 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001627 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001628 } else {
1629 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1630 commit_tree(source_mnt);
1631 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001632
1633 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1634 list_del_init(&child->mnt_hash);
1635 commit_tree(child);
1636 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001637 br_write_unlock(vfsmount_lock);
1638
Ram Paib90fa9a2005-11-07 17:19:50 -05001639 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001640
1641 out_cleanup_ids:
1642 if (IS_MNT_SHARED(dest_mnt))
1643 cleanup_group_ids(source_mnt, NULL);
1644 out:
1645 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001646}
1647
Al Virob12cea92011-03-18 08:55:38 -04001648static int lock_mount(struct path *path)
1649{
1650 struct vfsmount *mnt;
1651retry:
1652 mutex_lock(&path->dentry->d_inode->i_mutex);
1653 if (unlikely(cant_mount(path->dentry))) {
1654 mutex_unlock(&path->dentry->d_inode->i_mutex);
1655 return -ENOENT;
1656 }
1657 down_write(&namespace_sem);
1658 mnt = lookup_mnt(path);
1659 if (likely(!mnt))
1660 return 0;
1661 up_write(&namespace_sem);
1662 mutex_unlock(&path->dentry->d_inode->i_mutex);
1663 path_put(path);
1664 path->mnt = mnt;
1665 path->dentry = dget(mnt->mnt_root);
1666 goto retry;
1667}
1668
1669static void unlock_mount(struct path *path)
1670{
1671 up_write(&namespace_sem);
1672 mutex_unlock(&path->dentry->d_inode->i_mutex);
1673}
1674
Al Viro8c3ee422008-03-22 18:00:39 -04001675static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1678 return -EINVAL;
1679
Al Viro8c3ee422008-03-22 18:00:39 -04001680 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1682 return -ENOTDIR;
1683
Al Virob12cea92011-03-18 08:55:38 -04001684 if (d_unlinked(path->dentry))
1685 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
Al Virob12cea92011-03-18 08:55:38 -04001687 return attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
1690/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001691 * Sanity check the flags to change_mnt_propagation.
1692 */
1693
1694static int flags_to_propagation_type(int flags)
1695{
Roman Borisov7c6e9842011-05-25 16:26:48 -07001696 int type = flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001697
1698 /* Fail if any non-propagation flags are set */
1699 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1700 return 0;
1701 /* Only one propagation flag should be set */
1702 if (!is_power_of_2(type))
1703 return 0;
1704 return type;
1705}
1706
1707/*
Ram Pai07b20882005-11-07 17:19:07 -05001708 * recursively change the type of the mountpoint.
1709 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001710static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001711{
Al Viro2d92ab32008-08-02 00:51:11 -04001712 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001713 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001714 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001715 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001716
Miklos Szerediee6f9582007-05-08 00:30:40 -07001717 if (!capable(CAP_SYS_ADMIN))
1718 return -EPERM;
1719
Al Viro2d92ab32008-08-02 00:51:11 -04001720 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001721 return -EINVAL;
1722
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001723 type = flags_to_propagation_type(flag);
1724 if (!type)
1725 return -EINVAL;
1726
Ram Pai07b20882005-11-07 17:19:07 -05001727 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001728 if (type == MS_SHARED) {
1729 err = invent_group_ids(mnt, recurse);
1730 if (err)
1731 goto out_unlock;
1732 }
1733
Nick Piggin99b7db72010-08-18 04:37:39 +10001734 br_write_lock(vfsmount_lock);
Ram Pai07b20882005-11-07 17:19:07 -05001735 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1736 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001737 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001738
1739 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001740 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001741 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001742}
1743
1744/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 * do loopback mount.
1746 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001747static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001748 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749{
Al Virob12cea92011-03-18 08:55:38 -04001750 LIST_HEAD(umount_list);
Al Viro2d92ab32008-08-02 00:51:11 -04001751 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001753 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 if (err)
1755 return err;
1756 if (!old_name || !*old_name)
1757 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04001758 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (err)
1760 return err;
1761
Al Virob12cea92011-03-18 08:55:38 -04001762 err = lock_mount(path);
1763 if (err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001764 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001765
Al Virob12cea92011-03-18 08:55:38 -04001766 err = -EINVAL;
1767 if (IS_MNT_UNBINDABLE(old_path.mnt))
1768 goto out2;
1769
Al Viro2d92ab32008-08-02 00:51:11 -04001770 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Virob12cea92011-03-18 08:55:38 -04001771 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
Al Viroccd48bc2005-11-07 17:15:04 -05001773 err = -ENOMEM;
1774 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001775 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001776 else
Al Viro2d92ab32008-08-02 00:51:11 -04001777 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001778
1779 if (!mnt)
Al Virob12cea92011-03-18 08:55:38 -04001780 goto out2;
Al Viroccd48bc2005-11-07 17:15:04 -05001781
Al Viro2d92ab32008-08-02 00:51:11 -04001782 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001783 if (err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001784 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001785 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001786 br_write_unlock(vfsmount_lock);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001787 }
Al Virob12cea92011-03-18 08:55:38 -04001788out2:
1789 unlock_mount(path);
1790 release_mounts(&umount_list);
Al Viroccd48bc2005-11-07 17:15:04 -05001791out:
Al Viro2d92ab32008-08-02 00:51:11 -04001792 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 return err;
1794}
1795
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001796static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1797{
1798 int error = 0;
1799 int readonly_request = 0;
1800
1801 if (ms_flags & MS_RDONLY)
1802 readonly_request = 1;
1803 if (readonly_request == __mnt_is_readonly(mnt))
1804 return 0;
1805
1806 if (readonly_request)
1807 error = mnt_make_readonly(mnt);
1808 else
1809 __mnt_unmake_readonly(mnt);
1810 return error;
1811}
1812
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813/*
1814 * change filesystem flags. dir should be a physical root of filesystem.
1815 * If you've mounted a non-root directory somewhere and want to do remount
1816 * on it - tough luck.
1817 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001818static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 void *data)
1820{
1821 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001822 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
1824 if (!capable(CAP_SYS_ADMIN))
1825 return -EPERM;
1826
Al Viro2d92ab32008-08-02 00:51:11 -04001827 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return -EINVAL;
1829
Al Viro2d92ab32008-08-02 00:51:11 -04001830 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 return -EINVAL;
1832
Eric Parisff36fe22011-03-03 16:09:14 -05001833 err = security_sb_remount(sb, data);
1834 if (err)
1835 return err;
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001838 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001839 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001840 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001841 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001842 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001843 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001844 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001845 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001846 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001849 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001850 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001851 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001852 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 return err;
1855}
1856
Ram Pai9676f0c2005-11-07 17:21:20 -05001857static inline int tree_contains_unbindable(struct vfsmount *mnt)
1858{
1859 struct vfsmount *p;
1860 for (p = mnt; p; p = next_mnt(p, mnt)) {
1861 if (IS_MNT_UNBINDABLE(p))
1862 return 1;
1863 }
1864 return 0;
1865}
1866
Al Viro0a0d8a42008-08-02 00:55:27 -04001867static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
Al Viro2d92ab32008-08-02 00:51:11 -04001869 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 struct vfsmount *p;
1871 int err = 0;
1872 if (!capable(CAP_SYS_ADMIN))
1873 return -EPERM;
1874 if (!old_name || !*old_name)
1875 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001876 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (err)
1878 return err;
1879
Al Virob12cea92011-03-18 08:55:38 -04001880 err = lock_mount(path);
David Howellscc53ce52011-01-14 18:45:26 +00001881 if (err < 0)
1882 goto out;
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001885 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 goto out1;
1887
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001888 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001889 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
1891 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001892 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001893 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
Al Viro2d92ab32008-08-02 00:51:11 -04001895 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001896 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Al Viro2d92ab32008-08-02 00:51:11 -04001898 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1899 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001900 goto out1;
1901 /*
1902 * Don't move a mount residing in a shared parent.
1903 */
Al Viro2d92ab32008-08-02 00:51:11 -04001904 if (old_path.mnt->mnt_parent &&
1905 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001906 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001907 /*
1908 * Don't move a mount tree containing unbindable mounts to a destination
1909 * mount which is shared.
1910 */
Al Viro2d92ab32008-08-02 00:51:11 -04001911 if (IS_MNT_SHARED(path->mnt) &&
1912 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001913 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001915 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1916 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001917 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Al Viro2d92ab32008-08-02 00:51:11 -04001919 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001920 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001921 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
1923 /* if the mount is moved, it should no longer be expire
1924 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001925 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926out1:
Al Virob12cea92011-03-18 08:55:38 -04001927 unlock_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001930 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001931 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 return err;
1933}
1934
Al Viro9d412a42011-03-17 22:08:28 -04001935static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1936{
1937 int err;
1938 const char *subtype = strchr(fstype, '.');
1939 if (subtype) {
1940 subtype++;
1941 err = -EINVAL;
1942 if (!subtype[0])
1943 goto err;
1944 } else
1945 subtype = "";
1946
1947 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1948 err = -ENOMEM;
1949 if (!mnt->mnt_sb->s_subtype)
1950 goto err;
1951 return mnt;
1952
1953 err:
1954 mntput(mnt);
1955 return ERR_PTR(err);
1956}
1957
1958struct vfsmount *
1959do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1960{
1961 struct file_system_type *type = get_fs_type(fstype);
1962 struct vfsmount *mnt;
1963 if (!type)
1964 return ERR_PTR(-ENODEV);
1965 mnt = vfs_kern_mount(type, flags, name, data);
1966 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1967 !mnt->mnt_sb->s_subtype)
1968 mnt = fs_set_subtype(mnt, fstype);
1969 put_filesystem(type);
1970 return mnt;
1971}
1972EXPORT_SYMBOL_GPL(do_kern_mount);
1973
1974/*
1975 * add a mount into a namespace's mount tree
1976 */
1977static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1978{
1979 int err;
1980
1981 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1982
Al Virob12cea92011-03-18 08:55:38 -04001983 err = lock_mount(path);
1984 if (err)
1985 return err;
Al Viro9d412a42011-03-17 22:08:28 -04001986
1987 err = -EINVAL;
1988 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1989 goto unlock;
1990
1991 /* Refuse the same filesystem on the same mount point */
1992 err = -EBUSY;
1993 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1994 path->mnt->mnt_root == path->dentry)
1995 goto unlock;
1996
1997 err = -EINVAL;
1998 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1999 goto unlock;
2000
2001 newmnt->mnt_flags = mnt_flags;
2002 err = graft_tree(newmnt, path);
2003
2004unlock:
Al Virob12cea92011-03-18 08:55:38 -04002005 unlock_mount(path);
Al Viro9d412a42011-03-17 22:08:28 -04002006 return err;
2007}
Al Virob1e75df2011-01-17 01:47:59 -05002008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009/*
2010 * create a new mount for userspace and request it to be added into the
2011 * namespace's tree
2012 */
Al Viro0a0d8a42008-08-02 00:55:27 -04002013static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 int mnt_flags, char *name, void *data)
2015{
2016 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05002017 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018
Vegard Nossumeca6f532009-09-18 13:05:45 -07002019 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return -EINVAL;
2021
2022 /* we need capabilities... */
2023 if (!capable(CAP_SYS_ADMIN))
2024 return -EPERM;
2025
2026 mnt = do_kern_mount(type, flags, name, data);
2027 if (IS_ERR(mnt))
2028 return PTR_ERR(mnt);
2029
Al Viro15f9a3f2011-01-17 01:41:58 -05002030 err = do_add_mount(mnt, path, mnt_flags);
2031 if (err)
2032 mntput(mnt);
2033 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034}
2035
Al Viro19a167a2011-01-17 01:35:23 -05002036int finish_automount(struct vfsmount *m, struct path *path)
2037{
2038 int err;
2039 /* The new mount record should have at least 2 refs to prevent it being
2040 * expired before we get a chance to add it
2041 */
2042 BUG_ON(mnt_get_count(m) < 2);
2043
2044 if (m->mnt_sb == path->mnt->mnt_sb &&
2045 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002046 err = -ELOOP;
2047 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05002048 }
2049
Al Viro19a167a2011-01-17 01:35:23 -05002050 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05002051 if (!err)
2052 return 0;
2053fail:
2054 /* remove m from any expiration list it may be on */
2055 if (!list_empty(&m->mnt_expire)) {
2056 down_write(&namespace_sem);
2057 br_write_lock(vfsmount_lock);
2058 list_del_init(&m->mnt_expire);
2059 br_write_unlock(vfsmount_lock);
2060 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05002061 }
Al Virob1e75df2011-01-17 01:47:59 -05002062 mntput(m);
2063 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002064 return err;
2065}
2066
David Howellsea5b7782011-01-14 19:10:03 +00002067/**
2068 * mnt_set_expiry - Put a mount on an expiration list
2069 * @mnt: The mount to list.
2070 * @expiry_list: The list to add the mount to.
2071 */
2072void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2073{
2074 down_write(&namespace_sem);
2075 br_write_lock(vfsmount_lock);
2076
2077 list_add_tail(&mnt->mnt_expire, expiry_list);
2078
2079 br_write_unlock(vfsmount_lock);
2080 up_write(&namespace_sem);
2081}
2082EXPORT_SYMBOL(mnt_set_expiry);
2083
2084/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 * process a list of expirable mountpoints with the intent of discarding any
2086 * mountpoints that aren't in use and haven't been touched since last we came
2087 * here
2088 */
2089void mark_mounts_for_expiry(struct list_head *mounts)
2090{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 struct vfsmount *mnt, *next;
2092 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04002093 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
2095 if (list_empty(mounts))
2096 return;
2097
Al Virobcc5c7d2008-03-22 00:21:53 -04002098 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002099 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
2101 /* extract from the expiration list every vfsmount that matches the
2102 * following criteria:
2103 * - only referenced by its parent vfsmount
2104 * - still marked for expiry (marked on the last call here; marks are
2105 * cleared by mntput())
2106 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002107 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04002109 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002111 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002113 while (!list_empty(&graveyard)) {
2114 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2115 touch_mnt_namespace(mnt->mnt_ns);
2116 umount_tree(mnt, 1, &umounts);
2117 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002118 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002119 up_write(&namespace_sem);
2120
2121 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122}
2123
2124EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2125
2126/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002127 * Ripoff of 'select_parent()'
2128 *
2129 * search the list of submounts for a given mountpoint, and move any
2130 * shrinkable submounts to the 'graveyard' list.
2131 */
2132static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2133{
2134 struct vfsmount *this_parent = parent;
2135 struct list_head *next;
2136 int found = 0;
2137
2138repeat:
2139 next = this_parent->mnt_mounts.next;
2140resume:
2141 while (next != &this_parent->mnt_mounts) {
2142 struct list_head *tmp = next;
2143 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2144
2145 next = tmp->next;
2146 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2147 continue;
2148 /*
2149 * Descend a level if the d_mounts list is non-empty.
2150 */
2151 if (!list_empty(&mnt->mnt_mounts)) {
2152 this_parent = mnt;
2153 goto repeat;
2154 }
2155
2156 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002157 list_move_tail(&mnt->mnt_expire, graveyard);
2158 found++;
2159 }
2160 }
2161 /*
2162 * All done at this level ... ascend and resume the search
2163 */
2164 if (this_parent != parent) {
2165 next = this_parent->mnt_child.next;
2166 this_parent = this_parent->mnt_parent;
2167 goto resume;
2168 }
2169 return found;
2170}
2171
2172/*
2173 * process a list of expirable mountpoints with the intent of discarding any
2174 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002175 *
2176 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002177 */
Al Viroc35038b2008-03-22 00:46:23 -04002178static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002179{
2180 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04002181 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002182
Trond Myklebust5528f9112006-06-09 09:34:17 -04002183 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002184 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002185 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04002186 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04002187 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002188 touch_mnt_namespace(m->mnt_ns);
2189 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002190 }
2191 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002192}
2193
Trond Myklebust5528f9112006-06-09 09:34:17 -04002194/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 * Some copy_from_user() implementations do not return the exact number of
2196 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2197 * Note that this function differs from copy_from_user() in that it will oops
2198 * on bad values of `to', rather than returning a short copy.
2199 */
Ram Paib58fed82005-11-07 17:16:09 -05002200static long exact_copy_from_user(void *to, const void __user * from,
2201 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
2203 char *t = to;
2204 const char __user *f = from;
2205 char c;
2206
2207 if (!access_ok(VERIFY_READ, from, n))
2208 return n;
2209
2210 while (n) {
2211 if (__get_user(c, f)) {
2212 memset(t, 0, n);
2213 break;
2214 }
2215 *t++ = c;
2216 f++;
2217 n--;
2218 }
2219 return n;
2220}
2221
Ram Paib58fed82005-11-07 17:16:09 -05002222int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
2224 int i;
2225 unsigned long page;
2226 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 *where = 0;
2229 if (!data)
2230 return 0;
2231
2232 if (!(page = __get_free_page(GFP_KERNEL)))
2233 return -ENOMEM;
2234
2235 /* We only care that *some* data at the address the user
2236 * gave us is valid. Just in case, we'll zero
2237 * the remainder of the page.
2238 */
2239 /* copy_from_user cannot cross TASK_SIZE ! */
2240 size = TASK_SIZE - (unsigned long)data;
2241 if (size > PAGE_SIZE)
2242 size = PAGE_SIZE;
2243
2244 i = size - exact_copy_from_user((void *)page, data, size);
2245 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002246 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 return -EFAULT;
2248 }
2249 if (i != PAGE_SIZE)
2250 memset((char *)page + i, 0, PAGE_SIZE - i);
2251 *where = page;
2252 return 0;
2253}
2254
Vegard Nossumeca6f532009-09-18 13:05:45 -07002255int copy_mount_string(const void __user *data, char **where)
2256{
2257 char *tmp;
2258
2259 if (!data) {
2260 *where = NULL;
2261 return 0;
2262 }
2263
2264 tmp = strndup_user(data, PAGE_SIZE);
2265 if (IS_ERR(tmp))
2266 return PTR_ERR(tmp);
2267
2268 *where = tmp;
2269 return 0;
2270}
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272/*
2273 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2274 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2275 *
2276 * data is a (void *) that can point to any structure up to
2277 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2278 * information (or be NULL).
2279 *
2280 * Pre-0.97 versions of mount() didn't have a flags word.
2281 * When the flags word was introduced its top half was required
2282 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2283 * Therefore, if this magic number is present, it carries no information
2284 * and must be discarded.
2285 */
Ram Paib58fed82005-11-07 17:16:09 -05002286long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 unsigned long flags, void *data_page)
2288{
Al Viro2d92ab32008-08-02 00:51:11 -04002289 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 int retval = 0;
2291 int mnt_flags = 0;
2292
2293 /* Discard magic */
2294 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2295 flags &= ~MS_MGC_MSK;
2296
2297 /* Basic sanity checks */
2298
2299 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2300 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302 if (data_page)
2303 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2304
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09002305 /* ... and get the mountpoint */
2306 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2307 if (retval)
2308 return retval;
2309
2310 retval = security_sb_mount(dev_name, &path,
2311 type_page, flags, data_page);
2312 if (retval)
2313 goto dput_out;
2314
Andi Kleen613cbe32009-04-19 18:40:43 +02002315 /* Default to relatime unless overriden */
2316 if (!(flags & MS_NOATIME))
2317 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002318
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 /* Separate the per-mountpoint flags */
2320 if (flags & MS_NOSUID)
2321 mnt_flags |= MNT_NOSUID;
2322 if (flags & MS_NODEV)
2323 mnt_flags |= MNT_NODEV;
2324 if (flags & MS_NOEXEC)
2325 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002326 if (flags & MS_NOATIME)
2327 mnt_flags |= MNT_NOATIME;
2328 if (flags & MS_NODIRATIME)
2329 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002330 if (flags & MS_STRICTATIME)
2331 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002332 if (flags & MS_RDONLY)
2333 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002334
Al Viro7a4dec52010-08-09 12:05:43 -04002335 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002336 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2337 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002340 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 data_page);
2342 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002343 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002344 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002345 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002347 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 else
Al Viro2d92ab32008-08-02 00:51:11 -04002349 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 dev_name, data_page);
2351dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002352 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 return retval;
2354}
2355
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002356static struct mnt_namespace *alloc_mnt_ns(void)
2357{
2358 struct mnt_namespace *new_ns;
2359
2360 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2361 if (!new_ns)
2362 return ERR_PTR(-ENOMEM);
2363 atomic_set(&new_ns->count, 1);
2364 new_ns->root = NULL;
2365 INIT_LIST_HEAD(&new_ns->list);
2366 init_waitqueue_head(&new_ns->poll);
2367 new_ns->event = 0;
2368 return new_ns;
2369}
2370
Al Virof03c6592011-01-14 22:30:21 -05002371void mnt_make_longterm(struct vfsmount *mnt)
2372{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002373 __mnt_make_longterm(mnt);
Al Virof03c6592011-01-14 22:30:21 -05002374}
2375
2376void mnt_make_shortterm(struct vfsmount *mnt)
2377{
Al Viro7e3d0eb02011-01-16 16:32:11 -05002378#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -05002379 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2380 return;
2381 br_write_lock(vfsmount_lock);
2382 atomic_dec(&mnt->mnt_longterm);
2383 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002384#endif
Al Virof03c6592011-01-14 22:30:21 -05002385}
2386
JANAK DESAI741a2952006-02-07 12:59:00 -08002387/*
2388 * Allocate a new namespace structure and populate it with contents
2389 * copied from the namespace of the passed in task structure.
2390 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002391static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002392 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002394 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002395 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 struct vfsmount *p, *q;
2397
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002398 new_ns = alloc_mnt_ns();
2399 if (IS_ERR(new_ns))
2400 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
Ram Pai390c6842005-11-07 17:17:51 -05002402 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002404 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002405 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002407 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002409 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002411 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002413 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
2415 /*
2416 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2417 * as belonging to new namespace. We have already acquired a private
2418 * fs_struct, so tsk->fs->lock is not needed.
2419 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002420 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 q = new_ns->root;
2422 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002423 q->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002424 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002426 if (p == fs->root.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002427 fs->root.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002428 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002429 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 rootmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002432 if (p == fs->pwd.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002433 fs->pwd.mnt = mntget(q);
Al Viro7e3d0eb02011-01-16 16:32:11 -05002434 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002435 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 pwdmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002439 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 q = next_mnt(q, new_ns->root);
2441 }
Ram Pai390c6842005-11-07 17:17:51 -05002442 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002445 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002447 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448
JANAK DESAI741a2952006-02-07 12:59:00 -08002449 return new_ns;
2450}
2451
Eric W. Biederman213dd262007-07-15 23:41:15 -07002452struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002453 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002454{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002455 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002456
Badari Pulavartye3222c42007-05-08 00:25:21 -07002457 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002458 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002459
2460 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002461 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002462
Badari Pulavartye3222c42007-05-08 00:25:21 -07002463 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002464
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002465 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002466 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467}
2468
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002469/**
2470 * create_mnt_ns - creates a private namespace and adds a root filesystem
2471 * @mnt: pointer to the new root filesystem mountpoint
2472 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002473struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002474{
2475 struct mnt_namespace *new_ns;
2476
2477 new_ns = alloc_mnt_ns();
2478 if (!IS_ERR(new_ns)) {
2479 mnt->mnt_ns = new_ns;
Al Viro7e3d0eb02011-01-16 16:32:11 -05002480 __mnt_make_longterm(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002481 new_ns->root = mnt;
2482 list_add(&new_ns->list, &new_ns->root->mnt_list);
Al Viroc1334492011-11-16 16:12:14 -05002483 } else {
2484 mntput(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002485 }
2486 return new_ns;
2487}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002488EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002489
Al Viroea441d12011-11-16 21:43:59 -05002490struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2491{
2492 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05002493 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05002494 struct path path;
2495 int err;
2496
2497 ns = create_mnt_ns(mnt);
2498 if (IS_ERR(ns))
2499 return ERR_CAST(ns);
2500
2501 err = vfs_path_lookup(mnt->mnt_root, mnt,
2502 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2503
2504 put_mnt_ns(ns);
2505
2506 if (err)
2507 return ERR_PTR(err);
2508
2509 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05002510 s = path.mnt->mnt_sb;
2511 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05002512 mntput(path.mnt);
2513 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05002514 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05002515 /* ... and return the root of (sub)tree on it */
2516 return path.dentry;
2517}
2518EXPORT_SYMBOL(mount_subtree);
2519
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002520SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2521 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002523 int ret;
2524 char *kernel_type;
2525 char *kernel_dir;
2526 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Vegard Nossumeca6f532009-09-18 13:05:45 -07002529 ret = copy_mount_string(type, &kernel_type);
2530 if (ret < 0)
2531 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Vegard Nossumeca6f532009-09-18 13:05:45 -07002533 kernel_dir = getname(dir_name);
2534 if (IS_ERR(kernel_dir)) {
2535 ret = PTR_ERR(kernel_dir);
2536 goto out_dir;
2537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Vegard Nossumeca6f532009-09-18 13:05:45 -07002539 ret = copy_mount_string(dev_name, &kernel_dev);
2540 if (ret < 0)
2541 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Vegard Nossumeca6f532009-09-18 13:05:45 -07002543 ret = copy_mount_options(data, &data_page);
2544 if (ret < 0)
2545 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Vegard Nossumeca6f532009-09-18 13:05:45 -07002547 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2548 (void *) data_page);
2549
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002551out_data:
2552 kfree(kernel_dev);
2553out_dev:
2554 putname(kernel_dir);
2555out_dir:
2556 kfree(kernel_type);
2557out_type:
2558 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559}
2560
2561/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 * pivot_root Semantics:
2563 * Moves the root file system of the current process to the directory put_old,
2564 * makes new_root as the new root file system of the current process, and sets
2565 * root/cwd of all processes which had them on the current root to new_root.
2566 *
2567 * Restrictions:
2568 * The new_root and put_old must be directories, and must not be on the
2569 * same file system as the current process root. The put_old must be
2570 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2571 * pointed to by put_old must yield the same directory as new_root. No other
2572 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2573 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002574 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2575 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2576 * in this situation.
2577 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 * Notes:
2579 * - we don't move root/cwd if they are not at the root (reason: if something
2580 * cared enough to change them, it's probably wrong to force them elsewhere)
2581 * - it's okay to pick a root that isn't the root of a file system, e.g.
2582 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2583 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2584 * first.
2585 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002586SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2587 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588{
2589 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002590 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 int error;
2592
2593 if (!capable(CAP_SYS_ADMIN))
2594 return -EPERM;
2595
Al Viro2d8f3032008-07-22 09:59:21 -04002596 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 if (error)
2598 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
Al Viro2d8f3032008-07-22 09:59:21 -04002600 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 if (error)
2602 goto out1;
2603
Al Viro2d8f3032008-07-22 09:59:21 -04002604 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04002605 if (error)
2606 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002608 get_fs_root(current->fs, &root);
Al Virob12cea92011-03-18 08:55:38 -04002609 error = lock_mount(&old);
2610 if (error)
2611 goto out3;
2612
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002614 if (IS_MNT_SHARED(old.mnt) ||
2615 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002616 IS_MNT_SHARED(root.mnt->mnt_parent))
Al Virob12cea92011-03-18 08:55:38 -04002617 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002618 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
Al Virob12cea92011-03-18 08:55:38 -04002619 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002621 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002622 goto out4;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002623 if (d_unlinked(old.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002624 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002626 if (new.mnt == root.mnt ||
2627 old.mnt == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002628 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002630 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002631 goto out4; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002632 if (root.mnt->mnt_parent == root.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002633 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002634 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04002635 goto out4; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002636 if (new.mnt->mnt_parent == new.mnt)
Al Virob12cea92011-03-18 08:55:38 -04002637 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002638 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002639 tmp = old.mnt;
Al Viro2d8f3032008-07-22 09:59:21 -04002640 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 for (;;) {
2642 if (tmp->mnt_parent == tmp)
Al Virob12cea92011-03-18 08:55:38 -04002643 goto out4; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002644 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 break;
2646 tmp = tmp->mnt_parent;
2647 }
Al Viro2d8f3032008-07-22 09:59:21 -04002648 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002649 goto out4;
Al Viro2d8f3032008-07-22 09:59:21 -04002650 } else if (!is_subdir(old.dentry, new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04002651 goto out4;
Al Viro27cb1572011-03-18 08:29:36 -04002652 br_write_lock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002653 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002654 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002655 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002656 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002657 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002658 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002659 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002660 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002661 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04002663out4:
2664 unlock_mount(&old);
2665 if (!error) {
2666 path_put(&root_parent);
2667 path_put(&parent_path);
2668 }
2669out3:
Al Viro8c3ee422008-03-22 18:00:39 -04002670 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04002671out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002672 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002674 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677}
2678
2679static void __init init_mount_tree(void)
2680{
2681 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002682 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002683 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684
2685 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2686 if (IS_ERR(mnt))
2687 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002688
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002689 ns = create_mnt_ns(mnt);
2690 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002693 init_task.nsproxy->mnt_ns = ns;
2694 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Jan Blunckac748a02008-02-14 19:34:39 -08002696 root.mnt = ns->root;
2697 root.dentry = ns->root->mnt_root;
2698
2699 set_fs_pwd(current->fs, &root);
2700 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701}
2702
Denis Cheng74bf17c2007-10-16 23:26:30 -07002703void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002705 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002706 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
Ram Pai390c6842005-11-07 17:17:51 -05002708 init_rwsem(&namespace_sem);
2709
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002711 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
Ram Paib58fed82005-11-07 17:16:09 -05002713 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
2715 if (!mount_hashtable)
2716 panic("Failed to allocate mount hash table\n");
2717
Mandeep Singh Baines80cdc6d2011-03-22 16:33:54 -07002718 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
Eric Dumazet13f14b42008-02-06 01:37:57 -08002720 for (u = 0; u < HASH_SIZE; u++)
2721 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722
Nick Piggin99b7db72010-08-18 04:37:39 +10002723 br_lock_init(vfsmount_lock);
2724
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002725 err = sysfs_init();
2726 if (err)
2727 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002728 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002729 fs_kobj = kobject_create_and_add("fs", NULL);
2730 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002731 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 init_rootfs();
2733 init_mount_tree();
2734}
2735
Trond Myklebust616511d2009-06-22 15:09:13 -04002736void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002738 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002739
Al Virod498b252010-02-05 02:21:06 -05002740 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002741 return;
Ram Pai390c6842005-11-07 17:17:51 -05002742 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002743 br_write_lock(vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002744 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002745 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002746 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002747 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002748 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002750EXPORT_SYMBOL(put_mnt_ns);
Al Viro9d412a42011-03-17 22:08:28 -04002751
2752struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2753{
Tim Chen423e0ab2011-07-19 09:32:38 -07002754 struct vfsmount *mnt;
2755 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2756 if (!IS_ERR(mnt)) {
2757 /*
2758 * it is a longterm mount, don't release mnt until
2759 * we unmount before file sys is unregistered
2760 */
2761 mnt_make_longterm(mnt);
2762 }
2763 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04002764}
2765EXPORT_SYMBOL_GPL(kern_mount_data);
Tim Chen423e0ab2011-07-19 09:32:38 -07002766
2767void kern_unmount(struct vfsmount *mnt)
2768{
2769 /* release long term mount so mount point can be released */
2770 if (!IS_ERR_OR_NULL(mnt)) {
2771 mnt_make_shortterm(mnt);
2772 mntput(mnt);
2773 }
2774}
2775EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05002776
2777bool our_mnt(struct vfsmount *mnt)
2778{
2779 return check_mnt(mnt);
2780}