blob: d1edf26025dcb018ec434309aacc839f9dbf2cfb [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199struct vfsmount *alloc_vfsmnt(const char *name)
200{
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
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800469void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700470{
471 mnt->mnt_sb = sb;
472 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700473}
474
475EXPORT_SYMBOL(simple_set_mnt);
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477void free_vfsmnt(struct vfsmount *mnt)
478{
479 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100480 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000481#ifdef CONFIG_SMP
Nick Pigginb3e19d92011-01-07 17:50:11 +1100482 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000483#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 kmem_cache_free(mnt_cache, mnt);
485}
486
487/*
Ram Paia05964f2005-11-07 17:20:17 -0500488 * find the first or last mount at @dentry on vfsmount @mnt depending on
489 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000490 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 */
Ram Paia05964f2005-11-07 17:20:17 -0500492struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
493 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Ram Paib58fed82005-11-07 17:16:09 -0500495 struct list_head *head = mount_hashtable + hash(mnt, dentry);
496 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 struct vfsmount *p, *found = NULL;
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500500 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 p = NULL;
502 if (tmp == head)
503 break;
504 p = list_entry(tmp, struct vfsmount, mnt_hash);
505 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500506 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 break;
508 }
509 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return found;
511}
512
Ram Paia05964f2005-11-07 17:20:17 -0500513/*
514 * lookup_mnt increments the ref count before returning
515 * the vfsmount struct.
516 */
Al Viro1c755af2009-04-18 14:06:57 -0400517struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500518{
519 struct vfsmount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000520
521 br_read_lock(vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400522 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500523 mntget(child_mnt);
Nick Piggin99b7db72010-08-18 04:37:39 +1000524 br_read_unlock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500525 return child_mnt;
526}
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528static inline int check_mnt(struct vfsmount *mnt)
529{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800530 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
Nick Piggin99b7db72010-08-18 04:37:39 +1000533/*
534 * vfsmount lock must be held for write
535 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800536static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500537{
538 if (ns) {
539 ns->event = ++event;
540 wake_up_interruptible(&ns->poll);
541 }
542}
543
Nick Piggin99b7db72010-08-18 04:37:39 +1000544/*
545 * vfsmount lock must be held for write
546 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800547static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500548{
549 if (ns && ns->event != event) {
550 ns->event = event;
551 wake_up_interruptible(&ns->poll);
552 }
553}
554
Nick Piggin99b7db72010-08-18 04:37:39 +1000555/*
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100556 * Clear dentry's mounted state if it has no remaining mounts.
557 * vfsmount_lock must be held for write.
558 */
559static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
560{
561 unsigned u;
562
563 for (u = 0; u < HASH_SIZE; u++) {
564 struct vfsmount *p;
565
566 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
567 if (p->mnt_mountpoint == dentry)
568 return;
569 }
570 }
571 spin_lock(&dentry->d_lock);
572 dentry->d_flags &= ~DCACHE_MOUNTED;
573 spin_unlock(&dentry->d_lock);
574}
575
576/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000577 * vfsmount lock must be held for write
578 */
Al Viro1a390682008-03-21 20:48:19 -0400579static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Al Viro1a390682008-03-21 20:48:19 -0400581 old_path->dentry = mnt->mnt_mountpoint;
582 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 mnt->mnt_parent = mnt;
584 mnt->mnt_mountpoint = mnt->mnt_root;
585 list_del_init(&mnt->mnt_child);
586 list_del_init(&mnt->mnt_hash);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100587 dentry_reset_mounted(old_path->mnt, old_path->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
589
Nick Piggin99b7db72010-08-18 04:37:39 +1000590/*
591 * vfsmount lock must be held for write
592 */
Ram Paib90fa9a2005-11-07 17:19:50 -0500593void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
594 struct vfsmount *child_mnt)
595{
596 child_mnt->mnt_parent = mntget(mnt);
597 child_mnt->mnt_mountpoint = dget(dentry);
Nick Piggin5f57cbc2011-01-07 17:49:54 +1100598 spin_lock(&dentry->d_lock);
599 dentry->d_flags |= DCACHE_MOUNTED;
600 spin_unlock(&dentry->d_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500601}
602
Nick Piggin99b7db72010-08-18 04:37:39 +1000603/*
604 * vfsmount lock must be held for write
605 */
Al Viro1a390682008-03-21 20:48:19 -0400606static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Al Viro1a390682008-03-21 20:48:19 -0400608 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500609 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400610 hash(path->mnt, path->dentry));
611 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500612}
613
Al Viro7e3d0eb2011-01-16 16:32:11 -0500614static inline void __mnt_make_longterm(struct vfsmount *mnt)
615{
616#ifdef CONFIG_SMP
617 atomic_inc(&mnt->mnt_longterm);
618#endif
619}
620
621/* needs vfsmount lock for write */
622static inline void __mnt_make_shortterm(struct vfsmount *mnt)
623{
624#ifdef CONFIG_SMP
625 atomic_dec(&mnt->mnt_longterm);
626#endif
627}
628
Ram Paib90fa9a2005-11-07 17:19:50 -0500629/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000630 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500631 */
632static void commit_tree(struct vfsmount *mnt)
633{
634 struct vfsmount *parent = mnt->mnt_parent;
635 struct vfsmount *m;
636 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800637 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500638
639 BUG_ON(parent == mnt);
640
641 list_add_tail(&head, &mnt->mnt_list);
Al Virof03c6592011-01-14 22:30:21 -0500642 list_for_each_entry(m, &head, mnt_list) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800643 m->mnt_ns = n;
Al Viro7e3d0eb2011-01-16 16:32:11 -0500644 __mnt_make_longterm(m);
Al Virof03c6592011-01-14 22:30:21 -0500645 }
646
Ram Paib90fa9a2005-11-07 17:19:50 -0500647 list_splice(&head, n->list.prev);
648
649 list_add_tail(&mnt->mnt_hash, mount_hashtable +
650 hash(parent, mnt->mnt_mountpoint));
651 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800652 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
655static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
656{
657 struct list_head *next = p->mnt_mounts.next;
658 if (next == &p->mnt_mounts) {
659 while (1) {
660 if (p == root)
661 return NULL;
662 next = p->mnt_child.next;
663 if (next != &p->mnt_parent->mnt_mounts)
664 break;
665 p = p->mnt_parent;
666 }
667 }
668 return list_entry(next, struct vfsmount, mnt_child);
669}
670
Ram Pai9676f0c2005-11-07 17:21:20 -0500671static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
672{
673 struct list_head *prev = p->mnt_mounts.prev;
674 while (prev != &p->mnt_mounts) {
675 p = list_entry(prev, struct vfsmount, mnt_child);
676 prev = p->mnt_mounts.prev;
677 }
678 return p;
679}
680
Ram Pai36341f62005-11-07 17:17:22 -0500681static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
682 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 struct super_block *sb = old->mnt_sb;
685 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
686
687 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100688 if (flag & (CL_SLAVE | CL_PRIVATE))
689 mnt->mnt_group_id = 0; /* not a peer of original */
690 else
691 mnt->mnt_group_id = old->mnt_group_id;
692
693 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
694 int err = mnt_alloc_group_id(mnt);
695 if (err)
696 goto out_free;
697 }
698
Miklos Szeredibe1a16a2010-10-05 12:31:09 +0200699 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 atomic_inc(&sb->s_active);
701 mnt->mnt_sb = sb;
702 mnt->mnt_root = dget(root);
703 mnt->mnt_mountpoint = mnt->mnt_root;
704 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500705
Ram Pai5afe0022005-11-07 17:21:01 -0500706 if (flag & CL_SLAVE) {
707 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
708 mnt->mnt_master = old;
709 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400710 } else if (!(flag & CL_PRIVATE)) {
Al Viro796a6b52010-01-16 13:28:47 -0500711 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500712 list_add(&mnt->mnt_share, &old->mnt_share);
713 if (IS_MNT_SLAVE(old))
714 list_add(&mnt->mnt_slave, &old->mnt_slave);
715 mnt->mnt_master = old->mnt_master;
716 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500717 if (flag & CL_MAKE_SHARED)
718 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720 /* stick the duplicate mount on the same expiry list
721 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500722 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500723 if (!list_empty(&old->mnt_expire))
724 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
727 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100728
729 out_free:
730 free_vfsmnt(mnt);
731 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
733
Nick Pigginb3e19d92011-01-07 17:50:11 +1100734static inline void mntfree(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
736 struct super_block *sb = mnt->mnt_sb;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100737
Dave Hansen3d733632008-02-15 14:37:59 -0800738 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800739 * This probably indicates that somebody messed
740 * up a mnt_want/drop_write() pair. If this
741 * happens, the filesystem was probably unable
742 * to make r/w->r/o transitions.
743 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000744 /*
Nick Pigginb3e19d92011-01-07 17:50:11 +1100745 * The locking used to deal with mnt_count decrement provides barriers,
746 * so mnt_get_writers() below is safe.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000747 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100748 WARN_ON(mnt_get_writers(mnt));
Andreas Gruenbacherca9c7262009-12-17 21:24:27 -0500749 fsnotify_vfsmount_delete(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 dput(mnt->mnt_root);
751 free_vfsmnt(mnt);
752 deactivate_super(sb);
753}
754
Al Virof03c6592011-01-14 22:30:21 -0500755static void mntput_no_expire(struct vfsmount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -0500756{
Nick Pigginb3e19d92011-01-07 17:50:11 +1100757put_again:
Al Virof03c6592011-01-14 22:30:21 -0500758#ifdef CONFIG_SMP
759 br_read_lock(vfsmount_lock);
760 if (likely(atomic_read(&mnt->mnt_longterm))) {
761 mnt_dec_count(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100762 br_read_unlock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500763 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100764 }
Al Virof03c6592011-01-14 22:30:21 -0500765 br_read_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100766
767 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500768 mnt_dec_count(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100769 if (mnt_get_count(mnt)) {
770 br_write_unlock(vfsmount_lock);
771 return;
772 }
Nick Pigginb3e19d92011-01-07 17:50:11 +1100773#else
Nick Pigginb3e19d92011-01-07 17:50:11 +1100774 mnt_dec_count(mnt);
775 if (likely(mnt_get_count(mnt)))
Nick Piggin99b7db72010-08-18 04:37:39 +1000776 return;
777 br_write_lock(vfsmount_lock);
Al Virof03c6592011-01-14 22:30:21 -0500778#endif
Nick Pigginb3e19d92011-01-07 17:50:11 +1100779 if (unlikely(mnt->mnt_pinned)) {
780 mnt_add_count(mnt, mnt->mnt_pinned + 1);
781 mnt->mnt_pinned = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +1000782 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100783 acct_auto_close_mnt(mnt);
784 goto put_again;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500785 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000786 br_write_unlock(vfsmount_lock);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100787 mntfree(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500788}
Nick Pigginb3e19d92011-01-07 17:50:11 +1100789
790void mntput(struct vfsmount *mnt)
791{
792 if (mnt) {
793 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
794 if (unlikely(mnt->mnt_expiry_mark))
795 mnt->mnt_expiry_mark = 0;
Al Virof03c6592011-01-14 22:30:21 -0500796 mntput_no_expire(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100797 }
798}
799EXPORT_SYMBOL(mntput);
800
801struct vfsmount *mntget(struct vfsmount *mnt)
802{
803 if (mnt)
804 mnt_inc_count(mnt);
805 return mnt;
806}
807EXPORT_SYMBOL(mntget);
808
Al Viro7b7b1ac2005-11-07 17:13:39 -0500809void mnt_pin(struct vfsmount *mnt)
810{
Nick Piggin99b7db72010-08-18 04:37:39 +1000811 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500812 mnt->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000813 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500814}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500815EXPORT_SYMBOL(mnt_pin);
816
817void mnt_unpin(struct vfsmount *mnt)
818{
Nick Piggin99b7db72010-08-18 04:37:39 +1000819 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500820 if (mnt->mnt_pinned) {
Nick Pigginb3e19d92011-01-07 17:50:11 +1100821 mnt_inc_count(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500822 mnt->mnt_pinned--;
823 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000824 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500825}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500826EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800828static inline void mangle(struct seq_file *m, const char *s)
829{
830 seq_escape(m, s, " \t\n\\");
831}
832
833/*
834 * Simple .show_options callback for filesystems which don't want to
835 * implement more complex mount option showing.
836 *
837 * See also save_mount_options().
838 */
839int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
840{
Al Viro2a32ceb2009-05-08 16:05:57 -0400841 const char *options;
842
843 rcu_read_lock();
844 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800845
846 if (options != NULL && options[0]) {
847 seq_putc(m, ',');
848 mangle(m, options);
849 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400850 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800851
852 return 0;
853}
854EXPORT_SYMBOL(generic_show_options);
855
856/*
857 * If filesystem uses generic_show_options(), this function should be
858 * called from the fill_super() callback.
859 *
860 * The .remount_fs callback usually needs to be handled in a special
861 * way, to make sure, that previous options are not overwritten if the
862 * remount fails.
863 *
864 * Also note, that if the filesystem's .remount_fs function doesn't
865 * reset all options to their default value, but changes only newly
866 * given options, then the displayed options will not reflect reality
867 * any more.
868 */
869void save_mount_options(struct super_block *sb, char *options)
870{
Al Viro2a32ceb2009-05-08 16:05:57 -0400871 BUG_ON(sb->s_options);
872 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800873}
874EXPORT_SYMBOL(save_mount_options);
875
Al Viro2a32ceb2009-05-08 16:05:57 -0400876void replace_mount_options(struct super_block *sb, char *options)
877{
878 char *old = sb->s_options;
879 rcu_assign_pointer(sb->s_options, options);
880 if (old) {
881 synchronize_rcu();
882 kfree(old);
883 }
884}
885EXPORT_SYMBOL(replace_mount_options);
886
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100887#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888/* iterator */
889static void *m_start(struct seq_file *m, loff_t *pos)
890{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100891 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
Ram Pai390c6842005-11-07 17:17:51 -0500893 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100894 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
897static void *m_next(struct seq_file *m, void *v, loff_t *pos)
898{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100899 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700900
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100901 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902}
903
904static void m_stop(struct seq_file *m, void *v)
905{
Ram Pai390c6842005-11-07 17:17:51 -0500906 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Al Viro9f5596a2010-02-05 00:40:25 -0500909int mnt_had_events(struct proc_mounts *p)
910{
911 struct mnt_namespace *ns = p->ns;
912 int res = 0;
913
Nick Piggin99b7db72010-08-18 04:37:39 +1000914 br_read_lock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500915 if (p->event != ns->event) {
916 p->event = ns->event;
917 res = 1;
918 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000919 br_read_unlock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500920
921 return res;
922}
923
Ram Pai2d4d4862008-03-27 13:06:25 +0100924struct proc_fs_info {
925 int flag;
926 const char *str;
927};
928
Eric Paris2069f452008-07-04 09:47:13 +1000929static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Ram Pai2d4d4862008-03-27 13:06:25 +0100931 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 { MS_SYNCHRONOUS, ",sync" },
933 { MS_DIRSYNC, ",dirsync" },
934 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 { 0, NULL }
936 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100937 const struct proc_fs_info *fs_infop;
938
939 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
940 if (sb->s_flags & fs_infop->flag)
941 seq_puts(m, fs_infop->str);
942 }
Eric Paris2069f452008-07-04 09:47:13 +1000943
944 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100945}
946
947static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
948{
949 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 { MNT_NOSUID, ",nosuid" },
951 { MNT_NODEV, ",nodev" },
952 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800953 { MNT_NOATIME, ",noatime" },
954 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800955 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 { 0, NULL }
957 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100958 const struct proc_fs_info *fs_infop;
959
960 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
961 if (mnt->mnt_flags & fs_infop->flag)
962 seq_puts(m, fs_infop->str);
963 }
964}
965
966static void show_type(struct seq_file *m, struct super_block *sb)
967{
968 mangle(m, sb->s_type->name);
969 if (sb->s_subtype && sb->s_subtype[0]) {
970 seq_putc(m, '.');
971 mangle(m, sb->s_subtype);
972 }
973}
974
975static int show_vfsmnt(struct seq_file *m, void *v)
976{
977 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
978 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800979 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
982 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800983 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100985 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800986 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000987 err = show_sb_opts(m, mnt->mnt_sb);
988 if (err)
989 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100990 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (mnt->mnt_sb->s_op->show_options)
992 err = mnt->mnt_sb->s_op->show_options(m, mnt);
993 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000994out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return err;
996}
997
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100998const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 .start = m_start,
1000 .next = m_next,
1001 .stop = m_stop,
1002 .show = show_vfsmnt
1003};
1004
Ram Pai2d4d4862008-03-27 13:06:25 +01001005static int show_mountinfo(struct seq_file *m, void *v)
1006{
1007 struct proc_mounts *p = m->private;
1008 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1009 struct super_block *sb = mnt->mnt_sb;
1010 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1011 struct path root = p->root;
1012 int err = 0;
1013
1014 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1015 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1016 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1017 seq_putc(m, ' ');
1018 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1019 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1020 /*
1021 * Mountpoint is outside root, discard that one. Ugly,
1022 * but less so than trying to do that in iterator in a
1023 * race-free way (due to renames).
1024 */
1025 return SEQ_SKIP;
1026 }
1027 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1028 show_mnt_opts(m, mnt);
1029
1030 /* Tagged fields ("foo:X" or "bar") */
1031 if (IS_MNT_SHARED(mnt))
1032 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +01001033 if (IS_MNT_SLAVE(mnt)) {
1034 int master = mnt->mnt_master->mnt_group_id;
1035 int dom = get_dominating_id(mnt, &p->root);
1036 seq_printf(m, " master:%i", master);
1037 if (dom && dom != master)
1038 seq_printf(m, " propagate_from:%i", dom);
1039 }
Ram Pai2d4d4862008-03-27 13:06:25 +01001040 if (IS_MNT_UNBINDABLE(mnt))
1041 seq_puts(m, " unbindable");
1042
1043 /* Filesystem specific data */
1044 seq_puts(m, " - ");
1045 show_type(m, sb);
1046 seq_putc(m, ' ');
1047 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1048 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +10001049 err = show_sb_opts(m, sb);
1050 if (err)
1051 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +01001052 if (sb->s_op->show_options)
1053 err = sb->s_op->show_options(m, mnt);
1054 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +10001055out:
Ram Pai2d4d4862008-03-27 13:06:25 +01001056 return err;
1057}
1058
1059const struct seq_operations mountinfo_op = {
1060 .start = m_start,
1061 .next = m_next,
1062 .stop = m_stop,
1063 .show = show_mountinfo,
1064};
1065
Chuck Leverb4629fe2006-03-20 13:44:12 -05001066static int show_vfsstat(struct seq_file *m, void *v)
1067{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001068 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -08001069 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -05001070 int err = 0;
1071
1072 /* device */
1073 if (mnt->mnt_devname) {
1074 seq_puts(m, "device ");
1075 mangle(m, mnt->mnt_devname);
1076 } else
1077 seq_puts(m, "no device");
1078
1079 /* mount point */
1080 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -08001081 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -05001082 seq_putc(m, ' ');
1083
1084 /* file system type */
1085 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +01001086 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -05001087
1088 /* optional statistics */
1089 if (mnt->mnt_sb->s_op->show_stats) {
1090 seq_putc(m, ' ');
1091 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1092 }
1093
1094 seq_putc(m, '\n');
1095 return err;
1096}
1097
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001098const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -05001099 .start = m_start,
1100 .next = m_next,
1101 .stop = m_stop,
1102 .show = show_vfsstat,
1103};
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001104#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106/**
1107 * may_umount_tree - check if a mount tree is busy
1108 * @mnt: root of mount tree
1109 *
1110 * This is called to check if a tree of mounts has any
1111 * open files, pwds, chroots or sub mounts that are
1112 * busy.
1113 */
1114int may_umount_tree(struct vfsmount *mnt)
1115{
Ram Pai36341f62005-11-07 17:17:22 -05001116 int actual_refs = 0;
1117 int minimum_refs = 0;
1118 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Nick Pigginb3e19d92011-01-07 17:50:11 +11001120 /* write lock needed for mnt_get_count */
1121 br_write_lock(vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -05001122 for (p = mnt; p; p = next_mnt(p, mnt)) {
Nick Pigginb3e19d92011-01-07 17:50:11 +11001123 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Nick Pigginb3e19d92011-01-07 17:50:11 +11001126 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001129 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Ian Kente3474a82006-03-27 01:14:51 -08001131 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132}
1133
1134EXPORT_SYMBOL(may_umount_tree);
1135
1136/**
1137 * may_umount - check if a mount point is busy
1138 * @mnt: root of mount
1139 *
1140 * This is called to check if a mount point has any
1141 * open files, pwds, chroots or sub mounts. If the
1142 * mount has sub mounts this will return busy
1143 * regardless of whether the sub mounts are busy.
1144 *
1145 * Doesn't take quota and stuff into account. IOW, in some cases it will
1146 * give false negatives. The main reason why it's here is that we need
1147 * a non-destructive way to look for easily umountable filesystems.
1148 */
1149int may_umount(struct vfsmount *mnt)
1150{
Ian Kente3474a82006-03-27 01:14:51 -08001151 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001152 down_read(&namespace_sem);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001153 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001154 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -08001155 ret = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001156 br_write_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001157 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001158 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159}
1160
1161EXPORT_SYMBOL(may_umount);
1162
Ram Paib90fa9a2005-11-07 17:19:50 -05001163void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164{
Ram Pai70fbcdf2005-11-07 17:17:04 -05001165 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001166 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001167 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001168 list_del_init(&mnt->mnt_hash);
1169 if (mnt->mnt_parent != mnt) {
1170 struct dentry *dentry;
1171 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001172
1173 br_write_lock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001174 dentry = mnt->mnt_mountpoint;
1175 m = mnt->mnt_parent;
1176 mnt->mnt_mountpoint = mnt->mnt_root;
1177 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001178 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001179 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001180 dput(dentry);
1181 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
Al Virof03c6592011-01-14 22:30:21 -05001183 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001184 }
1185}
1186
Nick Piggin99b7db72010-08-18 04:37:39 +10001187/*
1188 * vfsmount lock must be held for write
1189 * namespace_sem must be held for write
1190 */
Ram Paia05964f2005-11-07 17:20:17 -05001191void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001192{
Al Viro7b8a53f2011-01-15 20:08:44 -05001193 LIST_HEAD(tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001194 struct vfsmount *p;
1195
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001196 for (p = mnt; p; p = next_mnt(p, mnt))
Al Viro7b8a53f2011-01-15 20:08:44 -05001197 list_move(&p->mnt_hash, &tmp_list);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001198
Ram Paia05964f2005-11-07 17:20:17 -05001199 if (propagate)
Al Viro7b8a53f2011-01-15 20:08:44 -05001200 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001201
Al Viro7b8a53f2011-01-15 20:08:44 -05001202 list_for_each_entry(p, &tmp_list, mnt_hash) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001203 list_del_init(&p->mnt_expire);
1204 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001205 __touch_mnt_namespace(p->mnt_ns);
1206 p->mnt_ns = NULL;
Al Viro7e3d0eb2011-01-16 16:32:11 -05001207 __mnt_make_shortterm(p);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001208 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001209 if (p->mnt_parent != p) {
1210 p->mnt_parent->mnt_ghosts++;
Nick Piggin5f57cbc2011-01-07 17:49:54 +11001211 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
Al Viro7c4b93d2008-03-21 23:59:49 -04001212 }
Ram Paia05964f2005-11-07 17:20:17 -05001213 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
Al Viro7b8a53f2011-01-15 20:08:44 -05001215 list_splice(&tmp_list, kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
Al Viroc35038b2008-03-22 00:46:23 -04001218static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1219
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220static int do_umount(struct vfsmount *mnt, int flags)
1221{
Ram Paib58fed82005-11-07 17:16:09 -05001222 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001224 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
1226 retval = security_sb_umount(mnt, flags);
1227 if (retval)
1228 return retval;
1229
1230 /*
1231 * Allow userspace to request a mountpoint be expired rather than
1232 * unmounting unconditionally. Unmount only happens if:
1233 * (1) the mark is already set (the mark is cleared by mntput())
1234 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1235 */
1236 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001237 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 flags & (MNT_FORCE | MNT_DETACH))
1239 return -EINVAL;
1240
Nick Pigginb3e19d92011-01-07 17:50:11 +11001241 /*
1242 * probably don't strictly need the lock here if we examined
1243 * all race cases, but it's a slowpath.
1244 */
1245 br_write_lock(vfsmount_lock);
1246 if (mnt_get_count(mnt) != 2) {
J. R. Okajimabf9faa22011-02-23 16:59:49 +09001247 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001249 }
1250 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 if (!xchg(&mnt->mnt_expiry_mark, 1))
1253 return -EAGAIN;
1254 }
1255
1256 /*
1257 * If we may have to abort operations to get out of this
1258 * mount, and they will themselves hold resources we must
1259 * allow the fs to do things. In the Unix tradition of
1260 * 'Gee thats tricky lets do it in userspace' the umount_begin
1261 * might fail to complete on the first run through as other tasks
1262 * must return, and the like. Thats for the mount program to worry
1263 * about for the moment.
1264 */
1265
Al Viro42faad92008-04-24 07:21:56 -04001266 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001267 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 /*
1271 * No sense to grab the lock for this test, but test itself looks
1272 * somewhat bogus. Suggestions for better replacement?
1273 * Ho-hum... In principle, we might treat that as umount + switch
1274 * to rootfs. GC would eventually take care of the old vfsmount.
1275 * Actually it makes sense, especially if rootfs would contain a
1276 * /reboot - static binary that would close all descriptors and
1277 * call reboot(9). Then init(8) could umount root and exec /reboot.
1278 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001279 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 /*
1281 * Special case for "unmounting" root ...
1282 * we just try to remount it readonly.
1283 */
1284 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001285 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 up_write(&sb->s_umount);
1288 return retval;
1289 }
1290
Ram Pai390c6842005-11-07 17:17:51 -05001291 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001292 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001293 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Al Viroc35038b2008-03-22 00:46:23 -04001295 if (!(flags & MNT_DETACH))
1296 shrink_submounts(mnt, &umount_list);
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001299 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001301 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 retval = 0;
1303 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001304 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001305 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001306 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 return retval;
1308}
1309
1310/*
1311 * Now umount can handle mount points as well as block devices.
1312 * This is important for filesystems which use unnamed block devices.
1313 *
1314 * We now support a flag for forced unmount like the other 'big iron'
1315 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1316 */
1317
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001318SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319{
Al Viro2d8f3032008-07-22 09:59:21 -04001320 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001322 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001324 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1325 return -EINVAL;
1326
1327 if (!(flags & UMOUNT_NOFOLLOW))
1328 lookup_flags |= LOOKUP_FOLLOW;
1329
1330 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 if (retval)
1332 goto out;
1333 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001334 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001336 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 goto dput_and_out;
1338
1339 retval = -EPERM;
1340 if (!capable(CAP_SYS_ADMIN))
1341 goto dput_and_out;
1342
Al Viro2d8f3032008-07-22 09:59:21 -04001343 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001345 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001346 dput(path.dentry);
1347 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348out:
1349 return retval;
1350}
1351
1352#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1353
1354/*
Ram Paib58fed82005-11-07 17:16:09 -05001355 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001357SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
Ram Paib58fed82005-11-07 17:16:09 -05001359 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360}
1361
1362#endif
1363
Al Viro2d92ab32008-08-02 00:51:11 -04001364static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365{
1366 if (capable(CAP_SYS_ADMIN))
1367 return 0;
1368 return -EPERM;
1369#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001370 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001372 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001373 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 return -EPERM;
1375 }
Al Viro2d92ab32008-08-02 00:51:11 -04001376 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return -EPERM;
1378 return 0;
1379#endif
1380}
1381
Ram Paib90fa9a2005-11-07 17:19:50 -05001382struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001383 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384{
1385 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001386 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
Ram Pai9676f0c2005-11-07 17:21:20 -05001388 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1389 return NULL;
1390
Ram Pai36341f62005-11-07 17:17:22 -05001391 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 if (!q)
1393 goto Enomem;
1394 q->mnt_mountpoint = mnt->mnt_mountpoint;
1395
1396 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001397 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001398 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 continue;
1400
1401 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001402 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1403 s = skip_mnt_tree(s);
1404 continue;
1405 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 while (p != s->mnt_parent) {
1407 p = p->mnt_parent;
1408 q = q->mnt_parent;
1409 }
1410 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001411 path.mnt = q;
1412 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001413 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (!q)
1415 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001416 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001418 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001419 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 }
1421 }
1422 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001423Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001425 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001426 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001427 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001428 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001429 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 }
1431 return NULL;
1432}
1433
Al Viro589ff872009-04-18 03:28:19 -04001434struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001435{
1436 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001437 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001438 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001439 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001440 return tree;
1441}
1442
1443void drop_collected_mounts(struct vfsmount *mnt)
1444{
1445 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001446 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001447 br_write_lock(vfsmount_lock);
Al Viro8aec0802007-06-07 12:20:32 -04001448 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001449 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001450 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001451 release_mounts(&umount_list);
1452}
1453
Al Viro1f707132010-01-30 22:51:25 -05001454int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1455 struct vfsmount *root)
1456{
1457 struct vfsmount *mnt;
1458 int res = f(root, arg);
1459 if (res)
1460 return res;
1461 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1462 res = f(mnt, arg);
1463 if (res)
1464 return res;
1465 }
1466 return 0;
1467}
1468
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001469static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1470{
1471 struct vfsmount *p;
1472
1473 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1474 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1475 mnt_release_group_id(p);
1476 }
1477}
1478
1479static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1480{
1481 struct vfsmount *p;
1482
1483 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1484 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1485 int err = mnt_alloc_group_id(p);
1486 if (err) {
1487 cleanup_group_ids(mnt, p);
1488 return err;
1489 }
1490 }
1491 }
1492
1493 return 0;
1494}
1495
Ram Paib90fa9a2005-11-07 17:19:50 -05001496/*
1497 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001498 * @nd : place the mount tree @source_mnt is attached
1499 * @parent_nd : if non-null, detach the source_mnt from its parent and
1500 * store the parent mount and mountpoint dentry.
1501 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001502 *
1503 * NOTE: in the table below explains the semantics when a source mount
1504 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001505 * ---------------------------------------------------------------------------
1506 * | BIND MOUNT OPERATION |
1507 * |**************************************************************************
1508 * | source-->| shared | private | slave | unbindable |
1509 * | dest | | | | |
1510 * | | | | | | |
1511 * | v | | | | |
1512 * |**************************************************************************
1513 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1514 * | | | | | |
1515 * |non-shared| shared (+) | private | slave (*) | invalid |
1516 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001517 * A bind operation clones the source mount and mounts the clone on the
1518 * destination mount.
1519 *
1520 * (++) the cloned mount is propagated to all the mounts in the propagation
1521 * tree of the destination mount and the cloned mount is added to
1522 * the peer group of the source mount.
1523 * (+) the cloned mount is created under the destination mount and is marked
1524 * as shared. The cloned mount is added to the peer group of the source
1525 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001526 * (+++) the mount is propagated to all the mounts in the propagation tree
1527 * of the destination mount and the cloned mount is made slave
1528 * of the same master as that of the source mount. The cloned mount
1529 * is marked as 'shared and slave'.
1530 * (*) the cloned mount is made a slave of the same master as that of the
1531 * source mount.
1532 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001533 * ---------------------------------------------------------------------------
1534 * | MOVE MOUNT OPERATION |
1535 * |**************************************************************************
1536 * | source-->| shared | private | slave | unbindable |
1537 * | dest | | | | |
1538 * | | | | | | |
1539 * | v | | | | |
1540 * |**************************************************************************
1541 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1542 * | | | | | |
1543 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1544 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001545 *
1546 * (+) the mount is moved to the destination. And is then propagated to
1547 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001548 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001549 * (+++) the mount is moved to the destination and is then propagated to
1550 * all the mounts belonging to the destination mount's propagation tree.
1551 * the mount is marked as 'shared and slave'.
1552 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001553 *
1554 * if the source mount is a tree, the operations explained above is
1555 * applied to each mount in the tree.
1556 * Must be called without spinlocks held, since this function can sleep
1557 * in allocations.
1558 */
1559static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001560 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001561{
1562 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001563 struct vfsmount *dest_mnt = path->mnt;
1564 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001565 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001566 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001567
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001568 if (IS_MNT_SHARED(dest_mnt)) {
1569 err = invent_group_ids(source_mnt, true);
1570 if (err)
1571 goto out;
1572 }
1573 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1574 if (err)
1575 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001576
Nick Piggin99b7db72010-08-18 04:37:39 +10001577 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001578
Ram Paib90fa9a2005-11-07 17:19:50 -05001579 if (IS_MNT_SHARED(dest_mnt)) {
1580 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1581 set_mnt_shared(p);
1582 }
Al Viro1a390682008-03-21 20:48:19 -04001583 if (parent_path) {
1584 detach_mnt(source_mnt, parent_path);
1585 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001586 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001587 } else {
1588 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1589 commit_tree(source_mnt);
1590 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001591
1592 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1593 list_del_init(&child->mnt_hash);
1594 commit_tree(child);
1595 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001596 br_write_unlock(vfsmount_lock);
1597
Ram Paib90fa9a2005-11-07 17:19:50 -05001598 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001599
1600 out_cleanup_ids:
1601 if (IS_MNT_SHARED(dest_mnt))
1602 cleanup_group_ids(source_mnt, NULL);
1603 out:
1604 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001605}
1606
Al Viro8c3ee422008-03-22 18:00:39 -04001607static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608{
1609 int err;
1610 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1611 return -EINVAL;
1612
Al Viro8c3ee422008-03-22 18:00:39 -04001613 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1615 return -ENOTDIR;
1616
1617 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001618 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001619 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 goto out_unlock;
1621
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001622 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001623 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001625 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return err;
1627}
1628
1629/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001630 * Sanity check the flags to change_mnt_propagation.
1631 */
1632
1633static int flags_to_propagation_type(int flags)
1634{
1635 int type = flags & ~MS_REC;
1636
1637 /* Fail if any non-propagation flags are set */
1638 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1639 return 0;
1640 /* Only one propagation flag should be set */
1641 if (!is_power_of_2(type))
1642 return 0;
1643 return type;
1644}
1645
1646/*
Ram Pai07b20882005-11-07 17:19:07 -05001647 * recursively change the type of the mountpoint.
1648 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001649static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001650{
Al Viro2d92ab32008-08-02 00:51:11 -04001651 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001652 int recurse = flag & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001653 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001654 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001655
Miklos Szerediee6f9582007-05-08 00:30:40 -07001656 if (!capable(CAP_SYS_ADMIN))
1657 return -EPERM;
1658
Al Viro2d92ab32008-08-02 00:51:11 -04001659 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001660 return -EINVAL;
1661
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07001662 type = flags_to_propagation_type(flag);
1663 if (!type)
1664 return -EINVAL;
1665
Ram Pai07b20882005-11-07 17:19:07 -05001666 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001667 if (type == MS_SHARED) {
1668 err = invent_group_ids(mnt, recurse);
1669 if (err)
1670 goto out_unlock;
1671 }
1672
Nick Piggin99b7db72010-08-18 04:37:39 +10001673 br_write_lock(vfsmount_lock);
Ram Pai07b20882005-11-07 17:19:07 -05001674 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1675 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001676 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001677
1678 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001679 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001680 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001681}
1682
1683/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 * do loopback mount.
1685 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001686static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001687 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688{
Al Viro2d92ab32008-08-02 00:51:11 -04001689 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001691 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 if (err)
1693 return err;
1694 if (!old_name || !*old_name)
1695 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001696 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (err)
1698 return err;
1699
Ram Pai390c6842005-11-07 17:17:51 -05001700 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001702 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001703 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001704
Al Viro2d92ab32008-08-02 00:51:11 -04001705 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001706 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
Al Viroccd48bc2005-11-07 17:15:04 -05001708 err = -ENOMEM;
1709 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001710 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001711 else
Al Viro2d92ab32008-08-02 00:51:11 -04001712 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001713
1714 if (!mnt)
1715 goto out;
1716
Al Viro2d92ab32008-08-02 00:51:11 -04001717 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001718 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001719 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001720
1721 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001722 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001723 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001724 release_mounts(&umount_list);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726
Al Viroccd48bc2005-11-07 17:15:04 -05001727out:
Ram Pai390c6842005-11-07 17:17:51 -05001728 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001729 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 return err;
1731}
1732
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001733static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1734{
1735 int error = 0;
1736 int readonly_request = 0;
1737
1738 if (ms_flags & MS_RDONLY)
1739 readonly_request = 1;
1740 if (readonly_request == __mnt_is_readonly(mnt))
1741 return 0;
1742
1743 if (readonly_request)
1744 error = mnt_make_readonly(mnt);
1745 else
1746 __mnt_unmake_readonly(mnt);
1747 return error;
1748}
1749
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750/*
1751 * change filesystem flags. dir should be a physical root of filesystem.
1752 * If you've mounted a non-root directory somewhere and want to do remount
1753 * on it - tough luck.
1754 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001755static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 void *data)
1757{
1758 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001759 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
1761 if (!capable(CAP_SYS_ADMIN))
1762 return -EPERM;
1763
Al Viro2d92ab32008-08-02 00:51:11 -04001764 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 return -EINVAL;
1766
Al Viro2d92ab32008-08-02 00:51:11 -04001767 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 return -EINVAL;
1769
1770 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001771 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001772 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001773 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001774 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001775 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001776 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001777 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001778 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001779 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001782 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001783 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001784 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001785 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 return err;
1788}
1789
Ram Pai9676f0c2005-11-07 17:21:20 -05001790static inline int tree_contains_unbindable(struct vfsmount *mnt)
1791{
1792 struct vfsmount *p;
1793 for (p = mnt; p; p = next_mnt(p, mnt)) {
1794 if (IS_MNT_UNBINDABLE(p))
1795 return 1;
1796 }
1797 return 0;
1798}
1799
Al Viro0a0d8a42008-08-02 00:55:27 -04001800static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
Al Viro2d92ab32008-08-02 00:51:11 -04001802 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 struct vfsmount *p;
1804 int err = 0;
1805 if (!capable(CAP_SYS_ADMIN))
1806 return -EPERM;
1807 if (!old_name || !*old_name)
1808 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001809 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 if (err)
1811 return err;
1812
Ram Pai390c6842005-11-07 17:17:51 -05001813 down_write(&namespace_sem);
David Howellscc53ce52011-01-14 18:45:26 +00001814 err = follow_down(path, true);
1815 if (err < 0)
1816 goto out;
1817
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001819 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 goto out;
1821
1822 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001823 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001824 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 goto out1;
1826
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001827 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001828 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001831 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001832 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Al Viro2d92ab32008-08-02 00:51:11 -04001834 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001835 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Al Viro2d92ab32008-08-02 00:51:11 -04001837 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1838 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001839 goto out1;
1840 /*
1841 * Don't move a mount residing in a shared parent.
1842 */
Al Viro2d92ab32008-08-02 00:51:11 -04001843 if (old_path.mnt->mnt_parent &&
1844 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001845 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001846 /*
1847 * Don't move a mount tree containing unbindable mounts to a destination
1848 * mount which is shared.
1849 */
Al Viro2d92ab32008-08-02 00:51:11 -04001850 if (IS_MNT_SHARED(path->mnt) &&
1851 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001852 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001854 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1855 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001856 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
Al Viro2d92ab32008-08-02 00:51:11 -04001858 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001859 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001860 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862 /* if the mount is moved, it should no longer be expire
1863 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001864 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001866 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867out:
Ram Pai390c6842005-11-07 17:17:51 -05001868 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001870 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001871 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 return err;
1873}
1874
Al Virob1e75df2011-01-17 01:47:59 -05001875static int do_add_mount(struct vfsmount *, struct path *, int);
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877/*
1878 * create a new mount for userspace and request it to be added into the
1879 * namespace's tree
1880 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001881static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 int mnt_flags, char *name, void *data)
1883{
1884 struct vfsmount *mnt;
Al Viro15f9a3f2011-01-17 01:41:58 -05001885 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Vegard Nossumeca6f532009-09-18 13:05:45 -07001887 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 return -EINVAL;
1889
1890 /* we need capabilities... */
1891 if (!capable(CAP_SYS_ADMIN))
1892 return -EPERM;
1893
1894 mnt = do_kern_mount(type, flags, name, data);
1895 if (IS_ERR(mnt))
1896 return PTR_ERR(mnt);
1897
Al Viro15f9a3f2011-01-17 01:41:58 -05001898 err = do_add_mount(mnt, path, mnt_flags);
1899 if (err)
1900 mntput(mnt);
1901 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
Al Viro19a167a2011-01-17 01:35:23 -05001904int finish_automount(struct vfsmount *m, struct path *path)
1905{
1906 int err;
1907 /* The new mount record should have at least 2 refs to prevent it being
1908 * expired before we get a chance to add it
1909 */
1910 BUG_ON(mnt_get_count(m) < 2);
1911
1912 if (m->mnt_sb == path->mnt->mnt_sb &&
1913 m->mnt_root == path->dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05001914 err = -ELOOP;
1915 goto fail;
Al Viro19a167a2011-01-17 01:35:23 -05001916 }
1917
Al Viro19a167a2011-01-17 01:35:23 -05001918 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
Al Virob1e75df2011-01-17 01:47:59 -05001919 if (!err)
1920 return 0;
1921fail:
1922 /* remove m from any expiration list it may be on */
1923 if (!list_empty(&m->mnt_expire)) {
1924 down_write(&namespace_sem);
1925 br_write_lock(vfsmount_lock);
1926 list_del_init(&m->mnt_expire);
1927 br_write_unlock(vfsmount_lock);
1928 up_write(&namespace_sem);
Al Viro19a167a2011-01-17 01:35:23 -05001929 }
Al Virob1e75df2011-01-17 01:47:59 -05001930 mntput(m);
1931 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05001932 return err;
1933}
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935/*
1936 * add a mount into a namespace's mount tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 */
Al Virob1e75df2011-01-17 01:47:59 -05001938static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
1940 int err;
1941
Al Viro80893522010-02-05 09:30:46 -05001942 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
Al Viro27d55f12010-01-16 13:07:36 -05001943
Ram Pai390c6842005-11-07 17:17:51 -05001944 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 /* Something was mounted here while we slept */
David Howellscc53ce52011-01-14 18:45:26 +00001946 err = follow_down(path, true);
1947 if (err < 0)
1948 goto unlock;
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001951 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 goto unlock;
1953
1954 /* Refuse the same filesystem on the same mount point */
1955 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001956 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1957 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 goto unlock;
1959
1960 err = -EINVAL;
1961 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1962 goto unlock;
1963
1964 newmnt->mnt_flags = mnt_flags;
Al Virob1e75df2011-01-17 01:47:59 -05001965 err = graft_tree(newmnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001968 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return err;
1970}
1971
David Howellsea5b7782011-01-14 19:10:03 +00001972/**
1973 * mnt_set_expiry - Put a mount on an expiration list
1974 * @mnt: The mount to list.
1975 * @expiry_list: The list to add the mount to.
1976 */
1977void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1978{
1979 down_write(&namespace_sem);
1980 br_write_lock(vfsmount_lock);
1981
1982 list_add_tail(&mnt->mnt_expire, expiry_list);
1983
1984 br_write_unlock(vfsmount_lock);
1985 up_write(&namespace_sem);
1986}
1987EXPORT_SYMBOL(mnt_set_expiry);
1988
1989/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 * process a list of expirable mountpoints with the intent of discarding any
1991 * mountpoints that aren't in use and haven't been touched since last we came
1992 * here
1993 */
1994void mark_mounts_for_expiry(struct list_head *mounts)
1995{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 struct vfsmount *mnt, *next;
1997 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001998 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
2000 if (list_empty(mounts))
2001 return;
2002
Al Virobcc5c7d2008-03-22 00:21:53 -04002003 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002004 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006 /* extract from the expiration list every vfsmount that matches the
2007 * following criteria:
2008 * - only referenced by its parent vfsmount
2009 * - still marked for expiry (marked on the last call here; marks are
2010 * cleared by mntput())
2011 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002012 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04002014 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07002016 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 }
Al Virobcc5c7d2008-03-22 00:21:53 -04002018 while (!list_empty(&graveyard)) {
2019 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2020 touch_mnt_namespace(mnt->mnt_ns);
2021 umount_tree(mnt, 1, &umounts);
2022 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002023 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04002024 up_write(&namespace_sem);
2025
2026 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027}
2028
2029EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2030
2031/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04002032 * Ripoff of 'select_parent()'
2033 *
2034 * search the list of submounts for a given mountpoint, and move any
2035 * shrinkable submounts to the 'graveyard' list.
2036 */
2037static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2038{
2039 struct vfsmount *this_parent = parent;
2040 struct list_head *next;
2041 int found = 0;
2042
2043repeat:
2044 next = this_parent->mnt_mounts.next;
2045resume:
2046 while (next != &this_parent->mnt_mounts) {
2047 struct list_head *tmp = next;
2048 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2049
2050 next = tmp->next;
2051 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2052 continue;
2053 /*
2054 * Descend a level if the d_mounts list is non-empty.
2055 */
2056 if (!list_empty(&mnt->mnt_mounts)) {
2057 this_parent = mnt;
2058 goto repeat;
2059 }
2060
2061 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04002062 list_move_tail(&mnt->mnt_expire, graveyard);
2063 found++;
2064 }
2065 }
2066 /*
2067 * All done at this level ... ascend and resume the search
2068 */
2069 if (this_parent != parent) {
2070 next = this_parent->mnt_child.next;
2071 this_parent = this_parent->mnt_parent;
2072 goto resume;
2073 }
2074 return found;
2075}
2076
2077/*
2078 * process a list of expirable mountpoints with the intent of discarding any
2079 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10002080 *
2081 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04002082 */
Al Viroc35038b2008-03-22 00:46:23 -04002083static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04002084{
2085 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04002086 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04002087
Trond Myklebust5528f9112006-06-09 09:34:17 -04002088 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04002089 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04002090 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04002091 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04002092 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08002093 touch_mnt_namespace(m->mnt_ns);
2094 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04002095 }
2096 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04002097}
2098
Trond Myklebust5528f9112006-06-09 09:34:17 -04002099/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * Some copy_from_user() implementations do not return the exact number of
2101 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2102 * Note that this function differs from copy_from_user() in that it will oops
2103 * on bad values of `to', rather than returning a short copy.
2104 */
Ram Paib58fed82005-11-07 17:16:09 -05002105static long exact_copy_from_user(void *to, const void __user * from,
2106 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107{
2108 char *t = to;
2109 const char __user *f = from;
2110 char c;
2111
2112 if (!access_ok(VERIFY_READ, from, n))
2113 return n;
2114
2115 while (n) {
2116 if (__get_user(c, f)) {
2117 memset(t, 0, n);
2118 break;
2119 }
2120 *t++ = c;
2121 f++;
2122 n--;
2123 }
2124 return n;
2125}
2126
Ram Paib58fed82005-11-07 17:16:09 -05002127int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128{
2129 int i;
2130 unsigned long page;
2131 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05002132
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 *where = 0;
2134 if (!data)
2135 return 0;
2136
2137 if (!(page = __get_free_page(GFP_KERNEL)))
2138 return -ENOMEM;
2139
2140 /* We only care that *some* data at the address the user
2141 * gave us is valid. Just in case, we'll zero
2142 * the remainder of the page.
2143 */
2144 /* copy_from_user cannot cross TASK_SIZE ! */
2145 size = TASK_SIZE - (unsigned long)data;
2146 if (size > PAGE_SIZE)
2147 size = PAGE_SIZE;
2148
2149 i = size - exact_copy_from_user((void *)page, data, size);
2150 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05002151 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 return -EFAULT;
2153 }
2154 if (i != PAGE_SIZE)
2155 memset((char *)page + i, 0, PAGE_SIZE - i);
2156 *where = page;
2157 return 0;
2158}
2159
Vegard Nossumeca6f532009-09-18 13:05:45 -07002160int copy_mount_string(const void __user *data, char **where)
2161{
2162 char *tmp;
2163
2164 if (!data) {
2165 *where = NULL;
2166 return 0;
2167 }
2168
2169 tmp = strndup_user(data, PAGE_SIZE);
2170 if (IS_ERR(tmp))
2171 return PTR_ERR(tmp);
2172
2173 *where = tmp;
2174 return 0;
2175}
2176
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177/*
2178 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2179 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2180 *
2181 * data is a (void *) that can point to any structure up to
2182 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2183 * information (or be NULL).
2184 *
2185 * Pre-0.97 versions of mount() didn't have a flags word.
2186 * When the flags word was introduced its top half was required
2187 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2188 * Therefore, if this magic number is present, it carries no information
2189 * and must be discarded.
2190 */
Ram Paib58fed82005-11-07 17:16:09 -05002191long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 unsigned long flags, void *data_page)
2193{
Al Viro2d92ab32008-08-02 00:51:11 -04002194 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 int retval = 0;
2196 int mnt_flags = 0;
2197
2198 /* Discard magic */
2199 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2200 flags &= ~MS_MGC_MSK;
2201
2202 /* Basic sanity checks */
2203
2204 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2205 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
2207 if (data_page)
2208 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2209
Tetsuo Handaa27ab9f2009-10-04 21:49:49 +09002210 /* ... and get the mountpoint */
2211 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2212 if (retval)
2213 return retval;
2214
2215 retval = security_sb_mount(dev_name, &path,
2216 type_page, flags, data_page);
2217 if (retval)
2218 goto dput_out;
2219
Andi Kleen613cbe32009-04-19 18:40:43 +02002220 /* Default to relatime unless overriden */
2221 if (!(flags & MS_NOATIME))
2222 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002223
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 /* Separate the per-mountpoint flags */
2225 if (flags & MS_NOSUID)
2226 mnt_flags |= MNT_NOSUID;
2227 if (flags & MS_NODEV)
2228 mnt_flags |= MNT_NODEV;
2229 if (flags & MS_NOEXEC)
2230 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002231 if (flags & MS_NOATIME)
2232 mnt_flags |= MNT_NOATIME;
2233 if (flags & MS_NODIRATIME)
2234 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002235 if (flags & MS_STRICTATIME)
2236 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002237 if (flags & MS_RDONLY)
2238 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002239
Al Viro7a4dec52010-08-09 12:05:43 -04002240 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002241 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2242 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002245 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 data_page);
2247 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002248 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002249 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002250 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002252 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 else
Al Viro2d92ab32008-08-02 00:51:11 -04002254 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 dev_name, data_page);
2256dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002257 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 return retval;
2259}
2260
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002261static struct mnt_namespace *alloc_mnt_ns(void)
2262{
2263 struct mnt_namespace *new_ns;
2264
2265 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2266 if (!new_ns)
2267 return ERR_PTR(-ENOMEM);
2268 atomic_set(&new_ns->count, 1);
2269 new_ns->root = NULL;
2270 INIT_LIST_HEAD(&new_ns->list);
2271 init_waitqueue_head(&new_ns->poll);
2272 new_ns->event = 0;
2273 return new_ns;
2274}
2275
Al Virof03c6592011-01-14 22:30:21 -05002276void mnt_make_longterm(struct vfsmount *mnt)
2277{
Al Viro7e3d0eb2011-01-16 16:32:11 -05002278 __mnt_make_longterm(mnt);
Al Virof03c6592011-01-14 22:30:21 -05002279}
2280
2281void mnt_make_shortterm(struct vfsmount *mnt)
2282{
Al Viro7e3d0eb2011-01-16 16:32:11 -05002283#ifdef CONFIG_SMP
Al Virof03c6592011-01-14 22:30:21 -05002284 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2285 return;
2286 br_write_lock(vfsmount_lock);
2287 atomic_dec(&mnt->mnt_longterm);
2288 br_write_unlock(vfsmount_lock);
Al Viro7e3d0eb2011-01-16 16:32:11 -05002289#endif
Al Virof03c6592011-01-14 22:30:21 -05002290}
2291
JANAK DESAI741a2952006-02-07 12:59:00 -08002292/*
2293 * Allocate a new namespace structure and populate it with contents
2294 * copied from the namespace of the passed in task structure.
2295 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002296static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002297 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002299 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002300 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 struct vfsmount *p, *q;
2302
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002303 new_ns = alloc_mnt_ns();
2304 if (IS_ERR(new_ns))
2305 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Ram Pai390c6842005-11-07 17:17:51 -05002307 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002309 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002310 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002312 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002314 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002316 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002318 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 /*
2321 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2322 * as belonging to new namespace. We have already acquired a private
2323 * fs_struct, so tsk->fs->lock is not needed.
2324 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002325 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 q = new_ns->root;
2327 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002328 q->mnt_ns = new_ns;
Al Viro7e3d0eb2011-01-16 16:32:11 -05002329 __mnt_make_longterm(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002331 if (p == fs->root.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002332 fs->root.mnt = mntget(q);
Al Viro7e3d0eb2011-01-16 16:32:11 -05002333 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002334 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 rootmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002337 if (p == fs->pwd.mnt) {
Al Virof03c6592011-01-14 22:30:21 -05002338 fs->pwd.mnt = mntget(q);
Al Viro7e3d0eb2011-01-16 16:32:11 -05002339 __mnt_make_longterm(q);
Al Virof03c6592011-01-14 22:30:21 -05002340 mnt_make_shortterm(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 pwdmnt = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002344 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 q = next_mnt(q, new_ns->root);
2346 }
Ram Pai390c6842005-11-07 17:17:51 -05002347 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05002350 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05002352 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
JANAK DESAI741a2952006-02-07 12:59:00 -08002354 return new_ns;
2355}
2356
Eric W. Biederman213dd262007-07-15 23:41:15 -07002357struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002358 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002359{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002360 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002361
Badari Pulavartye3222c42007-05-08 00:25:21 -07002362 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002363 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002364
2365 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002366 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002367
Badari Pulavartye3222c42007-05-08 00:25:21 -07002368 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002369
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002370 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002371 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372}
2373
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002374/**
2375 * create_mnt_ns - creates a private namespace and adds a root filesystem
2376 * @mnt: pointer to the new root filesystem mountpoint
2377 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002378struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002379{
2380 struct mnt_namespace *new_ns;
2381
2382 new_ns = alloc_mnt_ns();
2383 if (!IS_ERR(new_ns)) {
2384 mnt->mnt_ns = new_ns;
Al Viro7e3d0eb2011-01-16 16:32:11 -05002385 __mnt_make_longterm(mnt);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002386 new_ns->root = mnt;
2387 list_add(&new_ns->list, &new_ns->root->mnt_list);
2388 }
2389 return new_ns;
2390}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002391EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002392
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002393SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2394 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002396 int ret;
2397 char *kernel_type;
2398 char *kernel_dir;
2399 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
Vegard Nossumeca6f532009-09-18 13:05:45 -07002402 ret = copy_mount_string(type, &kernel_type);
2403 if (ret < 0)
2404 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Vegard Nossumeca6f532009-09-18 13:05:45 -07002406 kernel_dir = getname(dir_name);
2407 if (IS_ERR(kernel_dir)) {
2408 ret = PTR_ERR(kernel_dir);
2409 goto out_dir;
2410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Vegard Nossumeca6f532009-09-18 13:05:45 -07002412 ret = copy_mount_string(dev_name, &kernel_dev);
2413 if (ret < 0)
2414 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Vegard Nossumeca6f532009-09-18 13:05:45 -07002416 ret = copy_mount_options(data, &data_page);
2417 if (ret < 0)
2418 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Vegard Nossumeca6f532009-09-18 13:05:45 -07002420 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2421 (void *) data_page);
2422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002424out_data:
2425 kfree(kernel_dev);
2426out_dev:
2427 putname(kernel_dir);
2428out_dir:
2429 kfree(kernel_type);
2430out_type:
2431 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432}
2433
2434/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 * pivot_root Semantics:
2436 * Moves the root file system of the current process to the directory put_old,
2437 * makes new_root as the new root file system of the current process, and sets
2438 * root/cwd of all processes which had them on the current root to new_root.
2439 *
2440 * Restrictions:
2441 * The new_root and put_old must be directories, and must not be on the
2442 * same file system as the current process root. The put_old must be
2443 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2444 * pointed to by put_old must yield the same directory as new_root. No other
2445 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2446 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002447 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2448 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2449 * in this situation.
2450 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 * Notes:
2452 * - we don't move root/cwd if they are not at the root (reason: if something
2453 * cared enough to change them, it's probably wrong to force them elsewhere)
2454 * - it's okay to pick a root that isn't the root of a file system, e.g.
2455 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2456 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2457 * first.
2458 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002459SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2460 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461{
2462 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002463 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 int error;
2465
2466 if (!capable(CAP_SYS_ADMIN))
2467 return -EPERM;
2468
Al Viro2d8f3032008-07-22 09:59:21 -04002469 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 if (error)
2471 goto out0;
2472 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002473 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 goto out1;
2475
Al Viro2d8f3032008-07-22 09:59:21 -04002476 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 if (error)
2478 goto out1;
2479
Al Viro2d8f3032008-07-22 09:59:21 -04002480 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002482 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 goto out1;
2484 }
2485
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002486 get_fs_root(current->fs, &root);
Ram Pai390c6842005-11-07 17:17:51 -05002487 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002488 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002490 if (IS_MNT_SHARED(old.mnt) ||
2491 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002492 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002493 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002494 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 goto out2;
2496 error = -ENOENT;
Al Virod83c49f2010-04-30 17:17:09 -04002497 if (cant_mount(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002499 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002501 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 goto out2;
2503 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002504 if (new.mnt == root.mnt ||
2505 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 goto out2; /* loop, on the same file system */
2507 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002508 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002510 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002511 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002512 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002514 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002515 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002516 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002517 tmp = old.mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +10002518 br_write_lock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002519 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 for (;;) {
2521 if (tmp->mnt_parent == tmp)
2522 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002523 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 break;
2525 tmp = tmp->mnt_parent;
2526 }
Al Viro2d8f3032008-07-22 09:59:21 -04002527 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002529 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002531 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002532 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002533 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002534 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002535 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002536 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002537 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002538 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002539 chroot_fs_refs(&root, &new);
Nick Pigginb3e19d92011-01-07 17:50:11 +11002540
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002542 path_put(&root_parent);
2543 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002545 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002546 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002547 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002548 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002550 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 return error;
2553out3:
Nick Piggin99b7db72010-08-18 04:37:39 +10002554 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 goto out2;
2556}
2557
2558static void __init init_mount_tree(void)
2559{
2560 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002561 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002562 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2565 if (IS_ERR(mnt))
2566 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11002567
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002568 ns = create_mnt_ns(mnt);
2569 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002572 init_task.nsproxy->mnt_ns = ns;
2573 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574
Jan Blunckac748a02008-02-14 19:34:39 -08002575 root.mnt = ns->root;
2576 root.dentry = ns->root->mnt_root;
2577
2578 set_fs_pwd(current->fs, &root);
2579 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580}
2581
Denis Cheng74bf17c2007-10-16 23:26:30 -07002582void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002584 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002585 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Ram Pai390c6842005-11-07 17:17:51 -05002587 init_rwsem(&namespace_sem);
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002590 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
Ram Paib58fed82005-11-07 17:16:09 -05002592 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
2594 if (!mount_hashtable)
2595 panic("Failed to allocate mount hash table\n");
2596
Eric Dumazet13f14b42008-02-06 01:37:57 -08002597 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Eric Dumazet13f14b42008-02-06 01:37:57 -08002599 for (u = 0; u < HASH_SIZE; u++)
2600 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
Nick Piggin99b7db72010-08-18 04:37:39 +10002602 br_lock_init(vfsmount_lock);
2603
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002604 err = sysfs_init();
2605 if (err)
2606 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002607 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002608 fs_kobj = kobject_create_and_add("fs", NULL);
2609 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002610 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 init_rootfs();
2612 init_mount_tree();
2613}
2614
Trond Myklebust616511d2009-06-22 15:09:13 -04002615void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002617 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002618
Al Virod498b252010-02-05 02:21:06 -05002619 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002620 return;
Ram Pai390c6842005-11-07 17:17:51 -05002621 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002622 br_write_lock(vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002623 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002624 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002625 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002626 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002627 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002629EXPORT_SYMBOL(put_mnt_ns);