blob: 3dc283fd4716beacfe037839d2fc5fd5974c925d [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>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080019#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070021#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080023#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/namei.h>
25#include <linux/security.h>
26#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010029#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040030#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/uaccess.h>
32#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050033#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070034#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Eric Dumazet13f14b42008-02-06 01:37:57 -080036#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37#define HASH_SIZE (1UL << HASH_SHIFT)
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050040__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
41
42static 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);
Al Virof21f6222009-06-24 03:12:00 -040045static int mnt_id_start = 0;
46static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Eric Dumazetfa3536c2006-03-26 01:37:24 -080048static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080049static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050050static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080052/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060053struct kobject *fs_kobj;
54EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
57{
Ram Paib58fed82005-11-07 17:16:09 -050058 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
59 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080060 tmp = tmp + (tmp >> HASH_SHIFT);
61 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062}
63
Dave Hansen3d733632008-02-15 14:37:59 -080064#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
65
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010066/* allocation is serialized by namespace_sem */
67static int mnt_alloc_id(struct vfsmount *mnt)
68{
69 int res;
70
71retry:
72 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
73 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040074 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
75 if (!res)
76 mnt_id_start = mnt->mnt_id + 1;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010077 spin_unlock(&vfsmount_lock);
78 if (res == -EAGAIN)
79 goto retry;
80
81 return res;
82}
83
84static void mnt_free_id(struct vfsmount *mnt)
85{
Al Virof21f6222009-06-24 03:12:00 -040086 int id = mnt->mnt_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010087 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040088 ida_remove(&mnt_id_ida, id);
89 if (mnt_id_start > id)
90 mnt_id_start = id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010091 spin_unlock(&vfsmount_lock);
92}
93
Miklos Szeredi719f5d72008-03-27 13:06:23 +010094/*
95 * Allocate a new peer group ID
96 *
97 * mnt_group_ida is protected by namespace_sem
98 */
99static int mnt_alloc_group_id(struct vfsmount *mnt)
100{
Al Virof21f6222009-06-24 03:12:00 -0400101 int res;
102
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100103 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
104 return -ENOMEM;
105
Al Virof21f6222009-06-24 03:12:00 -0400106 res = ida_get_new_above(&mnt_group_ida,
107 mnt_group_start,
108 &mnt->mnt_group_id);
109 if (!res)
110 mnt_group_start = mnt->mnt_group_id + 1;
111
112 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100113}
114
115/*
116 * Release a peer group ID
117 */
118void mnt_release_group_id(struct vfsmount *mnt)
119{
Al Virof21f6222009-06-24 03:12:00 -0400120 int id = mnt->mnt_group_id;
121 ida_remove(&mnt_group_ida, id);
122 if (mnt_group_start > id)
123 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100124 mnt->mnt_group_id = 0;
125}
126
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127struct vfsmount *alloc_vfsmnt(const char *name)
128{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800129 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100131 int err;
132
133 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800134 if (err)
135 goto out_free_cache;
136
137 if (name) {
138 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
139 if (!mnt->mnt_devname)
140 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100141 }
142
Ram Paib58fed82005-11-07 17:16:09 -0500143 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 INIT_LIST_HEAD(&mnt->mnt_hash);
145 INIT_LIST_HEAD(&mnt->mnt_child);
146 INIT_LIST_HEAD(&mnt->mnt_mounts);
147 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700148 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500149 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500150 INIT_LIST_HEAD(&mnt->mnt_slave_list);
151 INIT_LIST_HEAD(&mnt->mnt_slave);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000152#ifdef CONFIG_SMP
153 mnt->mnt_writers = alloc_percpu(int);
154 if (!mnt->mnt_writers)
155 goto out_free_devname;
156#else
157 mnt->mnt_writers = 0;
158#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 }
160 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800161
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000162#ifdef CONFIG_SMP
163out_free_devname:
164 kfree(mnt->mnt_devname);
165#endif
Li Zefan88b38782008-07-21 18:06:36 +0800166out_free_id:
167 mnt_free_id(mnt);
168out_free_cache:
169 kmem_cache_free(mnt_cache, mnt);
170 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
Dave Hansen83660252008-02-15 14:37:30 -0800173/*
174 * Most r/o checks on a fs are for operations that take
175 * discrete amounts of time, like a write() or unlink().
176 * We must keep track of when those operations start
177 * (for permission checks) and when they end, so that
178 * we can determine when writes are able to occur to
179 * a filesystem.
180 */
Dave Hansen3d733632008-02-15 14:37:59 -0800181/*
182 * __mnt_is_readonly: check whether a mount is read-only
183 * @mnt: the mount to check for its write status
184 *
185 * This shouldn't be used directly ouside of the VFS.
186 * It does not guarantee that the filesystem will stay
187 * r/w, just that it is right *now*. This can not and
188 * should not be used in place of IS_RDONLY(inode).
189 * mnt_want/drop_write() will _keep_ the filesystem
190 * r/w.
191 */
192int __mnt_is_readonly(struct vfsmount *mnt)
193{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800194 if (mnt->mnt_flags & MNT_READONLY)
195 return 1;
196 if (mnt->mnt_sb->s_flags & MS_RDONLY)
197 return 1;
198 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800199}
200EXPORT_SYMBOL_GPL(__mnt_is_readonly);
201
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000202static inline void inc_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800203{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000204#ifdef CONFIG_SMP
205 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
206#else
207 mnt->mnt_writers++;
208#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800209}
Dave Hansen3d733632008-02-15 14:37:59 -0800210
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000211static inline void dec_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800212{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000213#ifdef CONFIG_SMP
214 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
215#else
216 mnt->mnt_writers--;
217#endif
218}
219
220static unsigned int count_mnt_writers(struct vfsmount *mnt)
221{
222#ifdef CONFIG_SMP
223 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800224 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800225
226 for_each_possible_cpu(cpu) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000227 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800228 }
Dave Hansen3d733632008-02-15 14:37:59 -0800229
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000230 return count;
231#else
232 return mnt->mnt_writers;
233#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800234}
235
236/*
237 * Most r/o checks on a fs are for operations that take
238 * discrete amounts of time, like a write() or unlink().
239 * We must keep track of when those operations start
240 * (for permission checks) and when they end, so that
241 * we can determine when writes are able to occur to
242 * a filesystem.
243 */
Dave Hansen83660252008-02-15 14:37:30 -0800244/**
245 * mnt_want_write - get write access to a mount
246 * @mnt: the mount on which to take a write
247 *
248 * This tells the low-level filesystem that a write is
249 * about to be performed to it, and makes sure that
250 * writes are allowed before returning success. When
251 * the write operation is finished, mnt_drop_write()
252 * must be called. This is effectively a refcount.
253 */
254int mnt_want_write(struct vfsmount *mnt)
255{
Dave Hansen3d733632008-02-15 14:37:59 -0800256 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800257
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000258 preempt_disable();
259 inc_mnt_writers(mnt);
260 /*
261 * The store to inc_mnt_writers must be visible before we pass
262 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
263 * incremented count after it has set MNT_WRITE_HOLD.
264 */
265 smp_mb();
266 while (mnt->mnt_flags & MNT_WRITE_HOLD)
267 cpu_relax();
268 /*
269 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
270 * be set to match its requirements. So we must not load that until
271 * MNT_WRITE_HOLD is cleared.
272 */
273 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800274 if (__mnt_is_readonly(mnt)) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000275 dec_mnt_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800276 ret = -EROFS;
277 goto out;
278 }
Dave Hansen3d733632008-02-15 14:37:59 -0800279out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000280 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800281 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800282}
283EXPORT_SYMBOL_GPL(mnt_want_write);
284
285/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000286 * mnt_clone_write - get write access to a mount
287 * @mnt: the mount on which to take a write
288 *
289 * This is effectively like mnt_want_write, except
290 * it must only be used to take an extra write reference
291 * on a mountpoint that we already know has a write reference
292 * on it. This allows some optimisation.
293 *
294 * After finished, mnt_drop_write must be called as usual to
295 * drop the reference.
296 */
297int mnt_clone_write(struct vfsmount *mnt)
298{
299 /* superblock may be r/o */
300 if (__mnt_is_readonly(mnt))
301 return -EROFS;
302 preempt_disable();
303 inc_mnt_writers(mnt);
304 preempt_enable();
305 return 0;
306}
307EXPORT_SYMBOL_GPL(mnt_clone_write);
308
309/**
310 * mnt_want_write_file - get write access to a file's mount
311 * @file: the file who's mount on which to take a write
312 *
313 * This is like mnt_want_write, but it takes a file and can
314 * do some optimisations if the file is open for write already
315 */
316int mnt_want_write_file(struct file *file)
317{
318 if (!(file->f_mode & FMODE_WRITE))
319 return mnt_want_write(file->f_path.mnt);
320 else
321 return mnt_clone_write(file->f_path.mnt);
322}
323EXPORT_SYMBOL_GPL(mnt_want_write_file);
324
325/**
Dave Hansen83660252008-02-15 14:37:30 -0800326 * mnt_drop_write - give up write access to a mount
327 * @mnt: the mount on which to give up write access
328 *
329 * Tells the low-level filesystem that we are done
330 * performing writes to it. Must be matched with
331 * mnt_want_write() call above.
332 */
333void mnt_drop_write(struct vfsmount *mnt)
334{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000335 preempt_disable();
336 dec_mnt_writers(mnt);
337 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800338}
339EXPORT_SYMBOL_GPL(mnt_drop_write);
340
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800341static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800342{
Dave Hansen3d733632008-02-15 14:37:59 -0800343 int ret = 0;
344
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800345 spin_lock(&vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000346 mnt->mnt_flags |= MNT_WRITE_HOLD;
347 /*
348 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
349 * should be visible before we do.
350 */
351 smp_mb();
352
353 /*
354 * With writers on hold, if this value is zero, then there are
355 * definitely no active writers (although held writers may subsequently
356 * increment the count, they'll have to wait, and decrement it after
357 * seeing MNT_READONLY).
358 *
359 * It is OK to have counter incremented on one CPU and decremented on
360 * another: the sum will add up correctly. The danger would be when we
361 * sum up each counter, if we read a counter before it is incremented,
362 * but then read another CPU's count which it has been subsequently
363 * decremented from -- we would see more decrements than we should.
364 * MNT_WRITE_HOLD protects against this scenario, because
365 * mnt_want_write first increments count, then smp_mb, then spins on
366 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
367 * we're counting up here.
368 */
369 if (count_mnt_writers(mnt) > 0)
370 ret = -EBUSY;
371 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800372 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000373 /*
374 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
375 * that become unheld will see MNT_READONLY.
376 */
377 smp_wmb();
378 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800379 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800380 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800381}
Dave Hansen83660252008-02-15 14:37:30 -0800382
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800383static void __mnt_unmake_readonly(struct vfsmount *mnt)
384{
385 spin_lock(&vfsmount_lock);
386 mnt->mnt_flags &= ~MNT_READONLY;
387 spin_unlock(&vfsmount_lock);
388}
389
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800390void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700391{
392 mnt->mnt_sb = sb;
393 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700394}
395
396EXPORT_SYMBOL(simple_set_mnt);
397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398void free_vfsmnt(struct vfsmount *mnt)
399{
400 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100401 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000402#ifdef CONFIG_SMP
403 free_percpu(mnt->mnt_writers);
404#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 kmem_cache_free(mnt_cache, mnt);
406}
407
408/*
Ram Paia05964f2005-11-07 17:20:17 -0500409 * find the first or last mount at @dentry on vfsmount @mnt depending on
410 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
Ram Paia05964f2005-11-07 17:20:17 -0500412struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
413 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414{
Ram Paib58fed82005-11-07 17:16:09 -0500415 struct list_head *head = mount_hashtable + hash(mnt, dentry);
416 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 struct vfsmount *p, *found = NULL;
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500420 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 p = NULL;
422 if (tmp == head)
423 break;
424 p = list_entry(tmp, struct vfsmount, mnt_hash);
425 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500426 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 break;
428 }
429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 return found;
431}
432
Ram Paia05964f2005-11-07 17:20:17 -0500433/*
434 * lookup_mnt increments the ref count before returning
435 * the vfsmount struct.
436 */
Al Viro1c755af2009-04-18 14:06:57 -0400437struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500438{
439 struct vfsmount *child_mnt;
440 spin_lock(&vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400441 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500442 mntget(child_mnt);
443 spin_unlock(&vfsmount_lock);
444 return child_mnt;
445}
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447static inline int check_mnt(struct vfsmount *mnt)
448{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800449 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
451
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800452static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500453{
454 if (ns) {
455 ns->event = ++event;
456 wake_up_interruptible(&ns->poll);
457 }
458}
459
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800460static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500461{
462 if (ns && ns->event != event) {
463 ns->event = event;
464 wake_up_interruptible(&ns->poll);
465 }
466}
467
Al Viro1a390682008-03-21 20:48:19 -0400468static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Al Viro1a390682008-03-21 20:48:19 -0400470 old_path->dentry = mnt->mnt_mountpoint;
471 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 mnt->mnt_parent = mnt;
473 mnt->mnt_mountpoint = mnt->mnt_root;
474 list_del_init(&mnt->mnt_child);
475 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400476 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
Ram Paib90fa9a2005-11-07 17:19:50 -0500479void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
480 struct vfsmount *child_mnt)
481{
482 child_mnt->mnt_parent = mntget(mnt);
483 child_mnt->mnt_mountpoint = dget(dentry);
484 dentry->d_mounted++;
485}
486
Al Viro1a390682008-03-21 20:48:19 -0400487static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
Al Viro1a390682008-03-21 20:48:19 -0400489 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500490 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400491 hash(path->mnt, path->dentry));
492 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500493}
494
495/*
496 * the caller must hold vfsmount_lock
497 */
498static void commit_tree(struct vfsmount *mnt)
499{
500 struct vfsmount *parent = mnt->mnt_parent;
501 struct vfsmount *m;
502 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800503 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500504
505 BUG_ON(parent == mnt);
506
507 list_add_tail(&head, &mnt->mnt_list);
508 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800509 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500510 list_splice(&head, n->list.prev);
511
512 list_add_tail(&mnt->mnt_hash, mount_hashtable +
513 hash(parent, mnt->mnt_mountpoint));
514 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800515 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516}
517
518static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
519{
520 struct list_head *next = p->mnt_mounts.next;
521 if (next == &p->mnt_mounts) {
522 while (1) {
523 if (p == root)
524 return NULL;
525 next = p->mnt_child.next;
526 if (next != &p->mnt_parent->mnt_mounts)
527 break;
528 p = p->mnt_parent;
529 }
530 }
531 return list_entry(next, struct vfsmount, mnt_child);
532}
533
Ram Pai9676f0c2005-11-07 17:21:20 -0500534static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
535{
536 struct list_head *prev = p->mnt_mounts.prev;
537 while (prev != &p->mnt_mounts) {
538 p = list_entry(prev, struct vfsmount, mnt_child);
539 prev = p->mnt_mounts.prev;
540 }
541 return p;
542}
543
Ram Pai36341f62005-11-07 17:17:22 -0500544static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
545 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546{
547 struct super_block *sb = old->mnt_sb;
548 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
549
550 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100551 if (flag & (CL_SLAVE | CL_PRIVATE))
552 mnt->mnt_group_id = 0; /* not a peer of original */
553 else
554 mnt->mnt_group_id = old->mnt_group_id;
555
556 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
557 int err = mnt_alloc_group_id(mnt);
558 if (err)
559 goto out_free;
560 }
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 mnt->mnt_flags = old->mnt_flags;
563 atomic_inc(&sb->s_active);
564 mnt->mnt_sb = sb;
565 mnt->mnt_root = dget(root);
566 mnt->mnt_mountpoint = mnt->mnt_root;
567 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500568
Ram Pai5afe0022005-11-07 17:21:01 -0500569 if (flag & CL_SLAVE) {
570 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
571 mnt->mnt_master = old;
572 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400573 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500574 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
575 list_add(&mnt->mnt_share, &old->mnt_share);
576 if (IS_MNT_SLAVE(old))
577 list_add(&mnt->mnt_slave, &old->mnt_slave);
578 mnt->mnt_master = old->mnt_master;
579 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500580 if (flag & CL_MAKE_SHARED)
581 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583 /* stick the duplicate mount on the same expiry list
584 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500585 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500586 if (!list_empty(&old->mnt_expire))
587 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500588 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 }
590 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100591
592 out_free:
593 free_vfsmnt(mnt);
594 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
596
Al Viro7b7b1ac2005-11-07 17:13:39 -0500597static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
599 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800600 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800601 * This probably indicates that somebody messed
602 * up a mnt_want/drop_write() pair. If this
603 * happens, the filesystem was probably unable
604 * to make r/w->r/o transitions.
605 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000606 /*
607 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
608 * provides barriers, so count_mnt_writers() below is safe. AV
609 */
610 WARN_ON(count_mnt_writers(mnt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 dput(mnt->mnt_root);
612 free_vfsmnt(mnt);
613 deactivate_super(sb);
614}
615
Al Viro7b7b1ac2005-11-07 17:13:39 -0500616void mntput_no_expire(struct vfsmount *mnt)
617{
618repeat:
619 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
620 if (likely(!mnt->mnt_pinned)) {
621 spin_unlock(&vfsmount_lock);
622 __mntput(mnt);
623 return;
624 }
625 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
626 mnt->mnt_pinned = 0;
627 spin_unlock(&vfsmount_lock);
628 acct_auto_close_mnt(mnt);
629 security_sb_umount_close(mnt);
630 goto repeat;
631 }
632}
633
634EXPORT_SYMBOL(mntput_no_expire);
635
636void mnt_pin(struct vfsmount *mnt)
637{
638 spin_lock(&vfsmount_lock);
639 mnt->mnt_pinned++;
640 spin_unlock(&vfsmount_lock);
641}
642
643EXPORT_SYMBOL(mnt_pin);
644
645void mnt_unpin(struct vfsmount *mnt)
646{
647 spin_lock(&vfsmount_lock);
648 if (mnt->mnt_pinned) {
649 atomic_inc(&mnt->mnt_count);
650 mnt->mnt_pinned--;
651 }
652 spin_unlock(&vfsmount_lock);
653}
654
655EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800657static inline void mangle(struct seq_file *m, const char *s)
658{
659 seq_escape(m, s, " \t\n\\");
660}
661
662/*
663 * Simple .show_options callback for filesystems which don't want to
664 * implement more complex mount option showing.
665 *
666 * See also save_mount_options().
667 */
668int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
669{
Al Viro2a32ceb2009-05-08 16:05:57 -0400670 const char *options;
671
672 rcu_read_lock();
673 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800674
675 if (options != NULL && options[0]) {
676 seq_putc(m, ',');
677 mangle(m, options);
678 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400679 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800680
681 return 0;
682}
683EXPORT_SYMBOL(generic_show_options);
684
685/*
686 * If filesystem uses generic_show_options(), this function should be
687 * called from the fill_super() callback.
688 *
689 * The .remount_fs callback usually needs to be handled in a special
690 * way, to make sure, that previous options are not overwritten if the
691 * remount fails.
692 *
693 * Also note, that if the filesystem's .remount_fs function doesn't
694 * reset all options to their default value, but changes only newly
695 * given options, then the displayed options will not reflect reality
696 * any more.
697 */
698void save_mount_options(struct super_block *sb, char *options)
699{
Al Viro2a32ceb2009-05-08 16:05:57 -0400700 BUG_ON(sb->s_options);
701 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800702}
703EXPORT_SYMBOL(save_mount_options);
704
Al Viro2a32ceb2009-05-08 16:05:57 -0400705void replace_mount_options(struct super_block *sb, char *options)
706{
707 char *old = sb->s_options;
708 rcu_assign_pointer(sb->s_options, options);
709 if (old) {
710 synchronize_rcu();
711 kfree(old);
712 }
713}
714EXPORT_SYMBOL(replace_mount_options);
715
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100716#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717/* iterator */
718static void *m_start(struct seq_file *m, loff_t *pos)
719{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100720 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Ram Pai390c6842005-11-07 17:17:51 -0500722 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100723 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724}
725
726static void *m_next(struct seq_file *m, void *v, loff_t *pos)
727{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100728 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700729
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100730 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731}
732
733static void m_stop(struct seq_file *m, void *v)
734{
Ram Pai390c6842005-11-07 17:17:51 -0500735 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736}
737
Ram Pai2d4d4862008-03-27 13:06:25 +0100738struct proc_fs_info {
739 int flag;
740 const char *str;
741};
742
Eric Paris2069f452008-07-04 09:47:13 +1000743static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Ram Pai2d4d4862008-03-27 13:06:25 +0100745 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 { MS_SYNCHRONOUS, ",sync" },
747 { MS_DIRSYNC, ",dirsync" },
748 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 { 0, NULL }
750 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100751 const struct proc_fs_info *fs_infop;
752
753 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
754 if (sb->s_flags & fs_infop->flag)
755 seq_puts(m, fs_infop->str);
756 }
Eric Paris2069f452008-07-04 09:47:13 +1000757
758 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100759}
760
761static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
762{
763 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 { MNT_NOSUID, ",nosuid" },
765 { MNT_NODEV, ",nodev" },
766 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800767 { MNT_NOATIME, ",noatime" },
768 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800769 { MNT_RELATIME, ",relatime" },
Matthew Garrettd0adde52009-03-26 17:49:56 +0000770 { MNT_STRICTATIME, ",strictatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 { 0, NULL }
772 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100773 const struct proc_fs_info *fs_infop;
774
775 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
776 if (mnt->mnt_flags & fs_infop->flag)
777 seq_puts(m, fs_infop->str);
778 }
779}
780
781static void show_type(struct seq_file *m, struct super_block *sb)
782{
783 mangle(m, sb->s_type->name);
784 if (sb->s_subtype && sb->s_subtype[0]) {
785 seq_putc(m, '.');
786 mangle(m, sb->s_subtype);
787 }
788}
789
790static int show_vfsmnt(struct seq_file *m, void *v)
791{
792 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
793 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800794 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
797 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800798 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100800 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800801 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000802 err = show_sb_opts(m, mnt->mnt_sb);
803 if (err)
804 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100805 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 if (mnt->mnt_sb->s_op->show_options)
807 err = mnt->mnt_sb->s_op->show_options(m, mnt);
808 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000809out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return err;
811}
812
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100813const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 .start = m_start,
815 .next = m_next,
816 .stop = m_stop,
817 .show = show_vfsmnt
818};
819
Ram Pai2d4d4862008-03-27 13:06:25 +0100820static int show_mountinfo(struct seq_file *m, void *v)
821{
822 struct proc_mounts *p = m->private;
823 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
824 struct super_block *sb = mnt->mnt_sb;
825 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
826 struct path root = p->root;
827 int err = 0;
828
829 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
830 MAJOR(sb->s_dev), MINOR(sb->s_dev));
831 seq_dentry(m, mnt->mnt_root, " \t\n\\");
832 seq_putc(m, ' ');
833 seq_path_root(m, &mnt_path, &root, " \t\n\\");
834 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
835 /*
836 * Mountpoint is outside root, discard that one. Ugly,
837 * but less so than trying to do that in iterator in a
838 * race-free way (due to renames).
839 */
840 return SEQ_SKIP;
841 }
842 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
843 show_mnt_opts(m, mnt);
844
845 /* Tagged fields ("foo:X" or "bar") */
846 if (IS_MNT_SHARED(mnt))
847 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100848 if (IS_MNT_SLAVE(mnt)) {
849 int master = mnt->mnt_master->mnt_group_id;
850 int dom = get_dominating_id(mnt, &p->root);
851 seq_printf(m, " master:%i", master);
852 if (dom && dom != master)
853 seq_printf(m, " propagate_from:%i", dom);
854 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100855 if (IS_MNT_UNBINDABLE(mnt))
856 seq_puts(m, " unbindable");
857
858 /* Filesystem specific data */
859 seq_puts(m, " - ");
860 show_type(m, sb);
861 seq_putc(m, ' ');
862 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
863 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000864 err = show_sb_opts(m, sb);
865 if (err)
866 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100867 if (sb->s_op->show_options)
868 err = sb->s_op->show_options(m, mnt);
869 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000870out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100871 return err;
872}
873
874const struct seq_operations mountinfo_op = {
875 .start = m_start,
876 .next = m_next,
877 .stop = m_stop,
878 .show = show_mountinfo,
879};
880
Chuck Leverb4629fe2006-03-20 13:44:12 -0500881static int show_vfsstat(struct seq_file *m, void *v)
882{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700883 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800884 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500885 int err = 0;
886
887 /* device */
888 if (mnt->mnt_devname) {
889 seq_puts(m, "device ");
890 mangle(m, mnt->mnt_devname);
891 } else
892 seq_puts(m, "no device");
893
894 /* mount point */
895 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800896 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500897 seq_putc(m, ' ');
898
899 /* file system type */
900 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100901 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500902
903 /* optional statistics */
904 if (mnt->mnt_sb->s_op->show_stats) {
905 seq_putc(m, ' ');
906 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
907 }
908
909 seq_putc(m, '\n');
910 return err;
911}
912
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100913const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500914 .start = m_start,
915 .next = m_next,
916 .stop = m_stop,
917 .show = show_vfsstat,
918};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100919#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921/**
922 * may_umount_tree - check if a mount tree is busy
923 * @mnt: root of mount tree
924 *
925 * This is called to check if a tree of mounts has any
926 * open files, pwds, chroots or sub mounts that are
927 * busy.
928 */
929int may_umount_tree(struct vfsmount *mnt)
930{
Ram Pai36341f62005-11-07 17:17:22 -0500931 int actual_refs = 0;
932 int minimum_refs = 0;
933 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500936 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 actual_refs += atomic_read(&p->mnt_count);
938 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 }
940 spin_unlock(&vfsmount_lock);
941
942 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800943 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Ian Kente3474a82006-03-27 01:14:51 -0800945 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
948EXPORT_SYMBOL(may_umount_tree);
949
950/**
951 * may_umount - check if a mount point is busy
952 * @mnt: root of mount
953 *
954 * This is called to check if a mount point has any
955 * open files, pwds, chroots or sub mounts. If the
956 * mount has sub mounts this will return busy
957 * regardless of whether the sub mounts are busy.
958 *
959 * Doesn't take quota and stuff into account. IOW, in some cases it will
960 * give false negatives. The main reason why it's here is that we need
961 * a non-destructive way to look for easily umountable filesystems.
962 */
963int may_umount(struct vfsmount *mnt)
964{
Ian Kente3474a82006-03-27 01:14:51 -0800965 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500966 spin_lock(&vfsmount_lock);
967 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800968 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500969 spin_unlock(&vfsmount_lock);
970 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973EXPORT_SYMBOL(may_umount);
974
Ram Paib90fa9a2005-11-07 17:19:50 -0500975void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500977 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800978 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700979 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500980 list_del_init(&mnt->mnt_hash);
981 if (mnt->mnt_parent != mnt) {
982 struct dentry *dentry;
983 struct vfsmount *m;
984 spin_lock(&vfsmount_lock);
985 dentry = mnt->mnt_mountpoint;
986 m = mnt->mnt_parent;
987 mnt->mnt_mountpoint = mnt->mnt_root;
988 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -0400989 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500991 dput(dentry);
992 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 }
994 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500995 }
996}
997
Ram Paia05964f2005-11-07 17:20:17 -0500998void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500999{
1000 struct vfsmount *p;
1001
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001002 for (p = mnt; p; p = next_mnt(p, mnt))
1003 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001004
Ram Paia05964f2005-11-07 17:20:17 -05001005 if (propagate)
1006 propagate_umount(kill);
1007
Ram Pai70fbcdf2005-11-07 17:17:04 -05001008 list_for_each_entry(p, kill, mnt_hash) {
1009 list_del_init(&p->mnt_expire);
1010 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001011 __touch_mnt_namespace(p->mnt_ns);
1012 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001013 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001014 if (p->mnt_parent != p) {
1015 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001016 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001017 }
Ram Paia05964f2005-11-07 17:20:17 -05001018 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
1020}
1021
Al Viroc35038b2008-03-22 00:46:23 -04001022static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024static int do_umount(struct vfsmount *mnt, int flags)
1025{
Ram Paib58fed82005-11-07 17:16:09 -05001026 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001028 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 retval = security_sb_umount(mnt, flags);
1031 if (retval)
1032 return retval;
1033
1034 /*
1035 * Allow userspace to request a mountpoint be expired rather than
1036 * unmounting unconditionally. Unmount only happens if:
1037 * (1) the mark is already set (the mark is cleared by mntput())
1038 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1039 */
1040 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001041 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 flags & (MNT_FORCE | MNT_DETACH))
1043 return -EINVAL;
1044
1045 if (atomic_read(&mnt->mnt_count) != 2)
1046 return -EBUSY;
1047
1048 if (!xchg(&mnt->mnt_expiry_mark, 1))
1049 return -EAGAIN;
1050 }
1051
1052 /*
1053 * If we may have to abort operations to get out of this
1054 * mount, and they will themselves hold resources we must
1055 * allow the fs to do things. In the Unix tradition of
1056 * 'Gee thats tricky lets do it in userspace' the umount_begin
1057 * might fail to complete on the first run through as other tasks
1058 * must return, and the like. Thats for the mount program to worry
1059 * about for the moment.
1060 */
1061
Al Viro42faad92008-04-24 07:21:56 -04001062 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001063 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
1066 /*
1067 * No sense to grab the lock for this test, but test itself looks
1068 * somewhat bogus. Suggestions for better replacement?
1069 * Ho-hum... In principle, we might treat that as umount + switch
1070 * to rootfs. GC would eventually take care of the old vfsmount.
1071 * Actually it makes sense, especially if rootfs would contain a
1072 * /reboot - static binary that would close all descriptors and
1073 * call reboot(9). Then init(8) could umount root and exec /reboot.
1074 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001075 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 /*
1077 * Special case for "unmounting" root ...
1078 * we just try to remount it readonly.
1079 */
1080 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001081 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 up_write(&sb->s_umount);
1084 return retval;
1085 }
1086
Ram Pai390c6842005-11-07 17:17:51 -05001087 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001089 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Al Viroc35038b2008-03-22 00:46:23 -04001091 if (!(flags & MNT_DETACH))
1092 shrink_submounts(mnt, &umount_list);
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001095 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001097 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 retval = 0;
1099 }
1100 spin_unlock(&vfsmount_lock);
1101 if (retval)
1102 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001103 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001104 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 return retval;
1106}
1107
1108/*
1109 * Now umount can handle mount points as well as block devices.
1110 * This is important for filesystems which use unnamed block devices.
1111 *
1112 * We now support a flag for forced unmount like the other 'big iron'
1113 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1114 */
1115
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001116SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
Al Viro2d8f3032008-07-22 09:59:21 -04001118 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 int retval;
1120
Al Viro2d8f3032008-07-22 09:59:21 -04001121 retval = user_path(name, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (retval)
1123 goto out;
1124 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001125 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001127 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 goto dput_and_out;
1129
1130 retval = -EPERM;
1131 if (!capable(CAP_SYS_ADMIN))
1132 goto dput_and_out;
1133
Al Viro2d8f3032008-07-22 09:59:21 -04001134 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001136 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001137 dput(path.dentry);
1138 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139out:
1140 return retval;
1141}
1142
1143#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1144
1145/*
Ram Paib58fed82005-11-07 17:16:09 -05001146 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001148SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Ram Paib58fed82005-11-07 17:16:09 -05001150 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151}
1152
1153#endif
1154
Al Viro2d92ab32008-08-02 00:51:11 -04001155static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
1157 if (capable(CAP_SYS_ADMIN))
1158 return 0;
1159 return -EPERM;
1160#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001161 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001163 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001164 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 return -EPERM;
1166 }
Al Viro2d92ab32008-08-02 00:51:11 -04001167 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 return -EPERM;
1169 return 0;
1170#endif
1171}
1172
Ram Paib90fa9a2005-11-07 17:19:50 -05001173struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001174 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
1176 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001177 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
Ram Pai9676f0c2005-11-07 17:21:20 -05001179 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1180 return NULL;
1181
Ram Pai36341f62005-11-07 17:17:22 -05001182 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if (!q)
1184 goto Enomem;
1185 q->mnt_mountpoint = mnt->mnt_mountpoint;
1186
1187 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001188 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001189 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 continue;
1191
1192 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001193 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1194 s = skip_mnt_tree(s);
1195 continue;
1196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 while (p != s->mnt_parent) {
1198 p = p->mnt_parent;
1199 q = q->mnt_parent;
1200 }
1201 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001202 path.mnt = q;
1203 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001204 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (!q)
1206 goto Enomem;
1207 spin_lock(&vfsmount_lock);
1208 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001209 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 spin_unlock(&vfsmount_lock);
1211 }
1212 }
1213 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001214Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001216 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001218 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001220 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
1222 return NULL;
1223}
1224
Al Viro589ff872009-04-18 03:28:19 -04001225struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001226{
1227 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001228 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001229 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001230 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001231 return tree;
1232}
1233
1234void drop_collected_mounts(struct vfsmount *mnt)
1235{
1236 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001237 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001238 spin_lock(&vfsmount_lock);
1239 umount_tree(mnt, 0, &umount_list);
1240 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001241 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001242 release_mounts(&umount_list);
1243}
1244
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001245static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1246{
1247 struct vfsmount *p;
1248
1249 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1250 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1251 mnt_release_group_id(p);
1252 }
1253}
1254
1255static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1256{
1257 struct vfsmount *p;
1258
1259 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1260 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1261 int err = mnt_alloc_group_id(p);
1262 if (err) {
1263 cleanup_group_ids(mnt, p);
1264 return err;
1265 }
1266 }
1267 }
1268
1269 return 0;
1270}
1271
Ram Paib90fa9a2005-11-07 17:19:50 -05001272/*
1273 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001274 * @nd : place the mount tree @source_mnt is attached
1275 * @parent_nd : if non-null, detach the source_mnt from its parent and
1276 * store the parent mount and mountpoint dentry.
1277 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001278 *
1279 * NOTE: in the table below explains the semantics when a source mount
1280 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001281 * ---------------------------------------------------------------------------
1282 * | BIND MOUNT OPERATION |
1283 * |**************************************************************************
1284 * | source-->| shared | private | slave | unbindable |
1285 * | dest | | | | |
1286 * | | | | | | |
1287 * | v | | | | |
1288 * |**************************************************************************
1289 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1290 * | | | | | |
1291 * |non-shared| shared (+) | private | slave (*) | invalid |
1292 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001293 * A bind operation clones the source mount and mounts the clone on the
1294 * destination mount.
1295 *
1296 * (++) the cloned mount is propagated to all the mounts in the propagation
1297 * tree of the destination mount and the cloned mount is added to
1298 * the peer group of the source mount.
1299 * (+) the cloned mount is created under the destination mount and is marked
1300 * as shared. The cloned mount is added to the peer group of the source
1301 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001302 * (+++) the mount is propagated to all the mounts in the propagation tree
1303 * of the destination mount and the cloned mount is made slave
1304 * of the same master as that of the source mount. The cloned mount
1305 * is marked as 'shared and slave'.
1306 * (*) the cloned mount is made a slave of the same master as that of the
1307 * source mount.
1308 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001309 * ---------------------------------------------------------------------------
1310 * | MOVE MOUNT OPERATION |
1311 * |**************************************************************************
1312 * | source-->| shared | private | slave | unbindable |
1313 * | dest | | | | |
1314 * | | | | | | |
1315 * | v | | | | |
1316 * |**************************************************************************
1317 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1318 * | | | | | |
1319 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1320 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001321 *
1322 * (+) the mount is moved to the destination. And is then propagated to
1323 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001324 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001325 * (+++) the mount is moved to the destination and is then propagated to
1326 * all the mounts belonging to the destination mount's propagation tree.
1327 * the mount is marked as 'shared and slave'.
1328 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001329 *
1330 * if the source mount is a tree, the operations explained above is
1331 * applied to each mount in the tree.
1332 * Must be called without spinlocks held, since this function can sleep
1333 * in allocations.
1334 */
1335static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001336 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001337{
1338 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001339 struct vfsmount *dest_mnt = path->mnt;
1340 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001341 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001342 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001343
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001344 if (IS_MNT_SHARED(dest_mnt)) {
1345 err = invent_group_ids(source_mnt, true);
1346 if (err)
1347 goto out;
1348 }
1349 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1350 if (err)
1351 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001352
1353 if (IS_MNT_SHARED(dest_mnt)) {
1354 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1355 set_mnt_shared(p);
1356 }
1357
1358 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001359 if (parent_path) {
1360 detach_mnt(source_mnt, parent_path);
1361 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001362 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001363 } else {
1364 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1365 commit_tree(source_mnt);
1366 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001367
1368 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1369 list_del_init(&child->mnt_hash);
1370 commit_tree(child);
1371 }
1372 spin_unlock(&vfsmount_lock);
1373 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001374
1375 out_cleanup_ids:
1376 if (IS_MNT_SHARED(dest_mnt))
1377 cleanup_group_ids(source_mnt, NULL);
1378 out:
1379 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001380}
1381
Al Viro8c3ee422008-03-22 18:00:39 -04001382static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
1384 int err;
1385 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1386 return -EINVAL;
1387
Al Viro8c3ee422008-03-22 18:00:39 -04001388 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1390 return -ENOTDIR;
1391
1392 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001393 mutex_lock(&path->dentry->d_inode->i_mutex);
1394 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 goto out_unlock;
1396
Al Viro8c3ee422008-03-22 18:00:39 -04001397 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 if (err)
1399 goto out_unlock;
1400
1401 err = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001402 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001403 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001405 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001407 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 return err;
1409}
1410
1411/*
Ram Pai07b20882005-11-07 17:19:07 -05001412 * recursively change the type of the mountpoint.
1413 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001414static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001415{
Al Viro2d92ab32008-08-02 00:51:11 -04001416 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001417 int recurse = flag & MS_REC;
1418 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001419 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001420
Miklos Szerediee6f9582007-05-08 00:30:40 -07001421 if (!capable(CAP_SYS_ADMIN))
1422 return -EPERM;
1423
Al Viro2d92ab32008-08-02 00:51:11 -04001424 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001425 return -EINVAL;
1426
1427 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001428 if (type == MS_SHARED) {
1429 err = invent_group_ids(mnt, recurse);
1430 if (err)
1431 goto out_unlock;
1432 }
1433
Ram Pai07b20882005-11-07 17:19:07 -05001434 spin_lock(&vfsmount_lock);
1435 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1436 change_mnt_propagation(m, type);
1437 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001438
1439 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001440 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001441 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001442}
1443
1444/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 * do loopback mount.
1446 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001447static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001448 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Al Viro2d92ab32008-08-02 00:51:11 -04001450 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001452 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 if (err)
1454 return err;
1455 if (!old_name || !*old_name)
1456 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001457 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 if (err)
1459 return err;
1460
Ram Pai390c6842005-11-07 17:17:51 -05001461 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001463 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001464 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001465
Al Viro2d92ab32008-08-02 00:51:11 -04001466 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001467 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Al Viroccd48bc2005-11-07 17:15:04 -05001469 err = -ENOMEM;
1470 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001471 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001472 else
Al Viro2d92ab32008-08-02 00:51:11 -04001473 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001474
1475 if (!mnt)
1476 goto out;
1477
Al Viro2d92ab32008-08-02 00:51:11 -04001478 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001479 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001480 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001482 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001484 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Al Viroccd48bc2005-11-07 17:15:04 -05001487out:
Ram Pai390c6842005-11-07 17:17:51 -05001488 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001489 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 return err;
1491}
1492
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001493static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1494{
1495 int error = 0;
1496 int readonly_request = 0;
1497
1498 if (ms_flags & MS_RDONLY)
1499 readonly_request = 1;
1500 if (readonly_request == __mnt_is_readonly(mnt))
1501 return 0;
1502
1503 if (readonly_request)
1504 error = mnt_make_readonly(mnt);
1505 else
1506 __mnt_unmake_readonly(mnt);
1507 return error;
1508}
1509
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510/*
1511 * change filesystem flags. dir should be a physical root of filesystem.
1512 * If you've mounted a non-root directory somewhere and want to do remount
1513 * on it - tough luck.
1514 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001515static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 void *data)
1517{
1518 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001519 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521 if (!capable(CAP_SYS_ADMIN))
1522 return -EPERM;
1523
Al Viro2d92ab32008-08-02 00:51:11 -04001524 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 return -EINVAL;
1526
Al Viro2d92ab32008-08-02 00:51:11 -04001527 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 return -EINVAL;
1529
1530 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001531 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001532 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001533 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001534 err = do_remount_sb(sb, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (!err)
Al Viro2d92ab32008-08-02 00:51:11 -04001536 path->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001538 if (!err) {
Al Viro2d92ab32008-08-02 00:51:11 -04001539 security_sb_post_remount(path->mnt, flags, data);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001540
1541 spin_lock(&vfsmount_lock);
1542 touch_mnt_namespace(path->mnt->mnt_ns);
1543 spin_unlock(&vfsmount_lock);
1544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return err;
1546}
1547
Ram Pai9676f0c2005-11-07 17:21:20 -05001548static inline int tree_contains_unbindable(struct vfsmount *mnt)
1549{
1550 struct vfsmount *p;
1551 for (p = mnt; p; p = next_mnt(p, mnt)) {
1552 if (IS_MNT_UNBINDABLE(p))
1553 return 1;
1554 }
1555 return 0;
1556}
1557
Al Viro0a0d8a42008-08-02 00:55:27 -04001558static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
Al Viro2d92ab32008-08-02 00:51:11 -04001560 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 struct vfsmount *p;
1562 int err = 0;
1563 if (!capable(CAP_SYS_ADMIN))
1564 return -EPERM;
1565 if (!old_name || !*old_name)
1566 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001567 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 if (err)
1569 return err;
1570
Ram Pai390c6842005-11-07 17:17:51 -05001571 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001572 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001573 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 ;
1575 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001576 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 goto out;
1578
1579 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001580 mutex_lock(&path->dentry->d_inode->i_mutex);
1581 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 goto out1;
1583
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001584 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001585 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
1587 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001588 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001589 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
Al Viro2d92ab32008-08-02 00:51:11 -04001591 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001592 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
Al Viro2d92ab32008-08-02 00:51:11 -04001594 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1595 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001596 goto out1;
1597 /*
1598 * Don't move a mount residing in a shared parent.
1599 */
Al Viro2d92ab32008-08-02 00:51:11 -04001600 if (old_path.mnt->mnt_parent &&
1601 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001602 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001603 /*
1604 * Don't move a mount tree containing unbindable mounts to a destination
1605 * mount which is shared.
1606 */
Al Viro2d92ab32008-08-02 00:51:11 -04001607 if (IS_MNT_SHARED(path->mnt) &&
1608 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001609 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001611 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1612 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001613 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
Al Viro2d92ab32008-08-02 00:51:11 -04001615 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001616 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001617 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619 /* if the mount is moved, it should no longer be expire
1620 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001621 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001623 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624out:
Ram Pai390c6842005-11-07 17:17:51 -05001625 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001627 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001628 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 return err;
1630}
1631
1632/*
1633 * create a new mount for userspace and request it to be added into the
1634 * namespace's tree
1635 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001636static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 int mnt_flags, char *name, void *data)
1638{
1639 struct vfsmount *mnt;
1640
1641 if (!type || !memchr(type, 0, PAGE_SIZE))
1642 return -EINVAL;
1643
1644 /* we need capabilities... */
1645 if (!capable(CAP_SYS_ADMIN))
1646 return -EPERM;
1647
Al Viro7f78d4c2009-05-08 13:34:06 -04001648 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001650 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (IS_ERR(mnt))
1652 return PTR_ERR(mnt);
1653
Al Viro2d92ab32008-08-02 00:51:11 -04001654 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655}
1656
1657/*
1658 * add a mount into a namespace's mount tree
1659 * - provide the option of adding the new mount to an expiration list
1660 */
Al Viro8d66bf52008-08-01 09:05:54 -04001661int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 int mnt_flags, struct list_head *fslist)
1663{
1664 int err;
1665
Ram Pai390c6842005-11-07 17:17:51 -05001666 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001668 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001669 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 ;
1671 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001672 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 goto unlock;
1674
1675 /* Refuse the same filesystem on the same mount point */
1676 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001677 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1678 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 goto unlock;
1680
1681 err = -EINVAL;
1682 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1683 goto unlock;
1684
1685 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001686 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001687 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
Al Viro6758f952008-03-22 16:14:30 -04001689 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001690 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001691
Ram Pai390c6842005-11-07 17:17:51 -05001692 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001693 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001696 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 mntput(newmnt);
1698 return err;
1699}
1700
1701EXPORT_SYMBOL_GPL(do_add_mount);
1702
Trond Myklebust5528f9112006-06-09 09:34:17 -04001703/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 * process a list of expirable mountpoints with the intent of discarding any
1705 * mountpoints that aren't in use and haven't been touched since last we came
1706 * here
1707 */
1708void mark_mounts_for_expiry(struct list_head *mounts)
1709{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 struct vfsmount *mnt, *next;
1711 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001712 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
1714 if (list_empty(mounts))
1715 return;
1716
Al Virobcc5c7d2008-03-22 00:21:53 -04001717 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 spin_lock(&vfsmount_lock);
1719
1720 /* extract from the expiration list every vfsmount that matches the
1721 * following criteria:
1722 * - only referenced by its parent vfsmount
1723 * - still marked for expiry (marked on the last call here; marks are
1724 * cleared by mntput())
1725 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001726 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001728 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001730 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001732 while (!list_empty(&graveyard)) {
1733 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1734 touch_mnt_namespace(mnt->mnt_ns);
1735 umount_tree(mnt, 1, &umounts);
1736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001738 up_write(&namespace_sem);
1739
1740 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741}
1742
1743EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1744
1745/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001746 * Ripoff of 'select_parent()'
1747 *
1748 * search the list of submounts for a given mountpoint, and move any
1749 * shrinkable submounts to the 'graveyard' list.
1750 */
1751static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1752{
1753 struct vfsmount *this_parent = parent;
1754 struct list_head *next;
1755 int found = 0;
1756
1757repeat:
1758 next = this_parent->mnt_mounts.next;
1759resume:
1760 while (next != &this_parent->mnt_mounts) {
1761 struct list_head *tmp = next;
1762 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1763
1764 next = tmp->next;
1765 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1766 continue;
1767 /*
1768 * Descend a level if the d_mounts list is non-empty.
1769 */
1770 if (!list_empty(&mnt->mnt_mounts)) {
1771 this_parent = mnt;
1772 goto repeat;
1773 }
1774
1775 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001776 list_move_tail(&mnt->mnt_expire, graveyard);
1777 found++;
1778 }
1779 }
1780 /*
1781 * All done at this level ... ascend and resume the search
1782 */
1783 if (this_parent != parent) {
1784 next = this_parent->mnt_child.next;
1785 this_parent = this_parent->mnt_parent;
1786 goto resume;
1787 }
1788 return found;
1789}
1790
1791/*
1792 * process a list of expirable mountpoints with the intent of discarding any
1793 * submounts of a specific parent mountpoint
1794 */
Al Viroc35038b2008-03-22 00:46:23 -04001795static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001796{
1797 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001798 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001799
Trond Myklebust5528f9112006-06-09 09:34:17 -04001800 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001801 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001802 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001803 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001804 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001805 touch_mnt_namespace(m->mnt_ns);
1806 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001807 }
1808 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001809}
1810
Trond Myklebust5528f9112006-06-09 09:34:17 -04001811/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 * Some copy_from_user() implementations do not return the exact number of
1813 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1814 * Note that this function differs from copy_from_user() in that it will oops
1815 * on bad values of `to', rather than returning a short copy.
1816 */
Ram Paib58fed82005-11-07 17:16:09 -05001817static long exact_copy_from_user(void *to, const void __user * from,
1818 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
1820 char *t = to;
1821 const char __user *f = from;
1822 char c;
1823
1824 if (!access_ok(VERIFY_READ, from, n))
1825 return n;
1826
1827 while (n) {
1828 if (__get_user(c, f)) {
1829 memset(t, 0, n);
1830 break;
1831 }
1832 *t++ = c;
1833 f++;
1834 n--;
1835 }
1836 return n;
1837}
1838
Ram Paib58fed82005-11-07 17:16:09 -05001839int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840{
1841 int i;
1842 unsigned long page;
1843 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 *where = 0;
1846 if (!data)
1847 return 0;
1848
1849 if (!(page = __get_free_page(GFP_KERNEL)))
1850 return -ENOMEM;
1851
1852 /* We only care that *some* data at the address the user
1853 * gave us is valid. Just in case, we'll zero
1854 * the remainder of the page.
1855 */
1856 /* copy_from_user cannot cross TASK_SIZE ! */
1857 size = TASK_SIZE - (unsigned long)data;
1858 if (size > PAGE_SIZE)
1859 size = PAGE_SIZE;
1860
1861 i = size - exact_copy_from_user((void *)page, data, size);
1862 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001863 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 return -EFAULT;
1865 }
1866 if (i != PAGE_SIZE)
1867 memset((char *)page + i, 0, PAGE_SIZE - i);
1868 *where = page;
1869 return 0;
1870}
1871
1872/*
1873 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1874 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1875 *
1876 * data is a (void *) that can point to any structure up to
1877 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1878 * information (or be NULL).
1879 *
1880 * Pre-0.97 versions of mount() didn't have a flags word.
1881 * When the flags word was introduced its top half was required
1882 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1883 * Therefore, if this magic number is present, it carries no information
1884 * and must be discarded.
1885 */
Ram Paib58fed82005-11-07 17:16:09 -05001886long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 unsigned long flags, void *data_page)
1888{
Al Viro2d92ab32008-08-02 00:51:11 -04001889 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 int retval = 0;
1891 int mnt_flags = 0;
1892
1893 /* Discard magic */
1894 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1895 flags &= ~MS_MGC_MSK;
1896
1897 /* Basic sanity checks */
1898
1899 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1900 return -EINVAL;
1901 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1902 return -EINVAL;
1903
1904 if (data_page)
1905 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1906
Andi Kleen613cbe32009-04-19 18:40:43 +02001907 /* Default to relatime unless overriden */
1908 if (!(flags & MS_NOATIME))
1909 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 /* Separate the per-mountpoint flags */
1912 if (flags & MS_NOSUID)
1913 mnt_flags |= MNT_NOSUID;
1914 if (flags & MS_NODEV)
1915 mnt_flags |= MNT_NODEV;
1916 if (flags & MS_NOEXEC)
1917 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001918 if (flags & MS_NOATIME)
1919 mnt_flags |= MNT_NOATIME;
1920 if (flags & MS_NODIRATIME)
1921 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001922 if (flags & MS_STRICTATIME)
1923 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001924 if (flags & MS_RDONLY)
1925 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001926
1927 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Matthew Garrettd0adde52009-03-26 17:49:56 +00001928 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1929 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 /* ... and get the mountpoint */
Al Viro2d92ab32008-08-02 00:51:11 -04001932 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 if (retval)
1934 return retval;
1935
Al Viro2d92ab32008-08-02 00:51:11 -04001936 retval = security_sb_mount(dev_name, &path,
Al Virob5266eb2008-03-22 17:48:24 -04001937 type_page, flags, data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 if (retval)
1939 goto dput_out;
1940
1941 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001942 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 data_page);
1944 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001945 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001946 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04001947 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04001949 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 else
Al Viro2d92ab32008-08-02 00:51:11 -04001951 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 dev_name, data_page);
1953dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04001954 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return retval;
1956}
1957
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04001958static struct mnt_namespace *alloc_mnt_ns(void)
1959{
1960 struct mnt_namespace *new_ns;
1961
1962 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1963 if (!new_ns)
1964 return ERR_PTR(-ENOMEM);
1965 atomic_set(&new_ns->count, 1);
1966 new_ns->root = NULL;
1967 INIT_LIST_HEAD(&new_ns->list);
1968 init_waitqueue_head(&new_ns->poll);
1969 new_ns->event = 0;
1970 return new_ns;
1971}
1972
JANAK DESAI741a2952006-02-07 12:59:00 -08001973/*
1974 * Allocate a new namespace structure and populate it with contents
1975 * copied from the namespace of the passed in task structure.
1976 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001977static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001978 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001980 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04001981 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 struct vfsmount *p, *q;
1983
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04001984 new_ns = alloc_mnt_ns();
1985 if (IS_ERR(new_ns))
1986 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Ram Pai390c6842005-11-07 17:17:51 -05001988 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001990 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001991 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001993 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08001995 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 }
1997 spin_lock(&vfsmount_lock);
1998 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1999 spin_unlock(&vfsmount_lock);
2000
2001 /*
2002 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2003 * as belonging to new namespace. We have already acquired a private
2004 * fs_struct, so tsk->fs->lock is not needed.
2005 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002006 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 q = new_ns->root;
2008 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002009 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002011 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002013 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002015 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002017 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002020 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 q = next_mnt(q, new_ns->root);
2022 }
Ram Pai390c6842005-11-07 17:17:51 -05002023 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 if (rootmnt)
2026 mntput(rootmnt);
2027 if (pwdmnt)
2028 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
JANAK DESAI741a2952006-02-07 12:59:00 -08002030 return new_ns;
2031}
2032
Eric W. Biederman213dd262007-07-15 23:41:15 -07002033struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002034 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002035{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002036 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002037
Badari Pulavartye3222c42007-05-08 00:25:21 -07002038 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002039 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002040
2041 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002042 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002043
Badari Pulavartye3222c42007-05-08 00:25:21 -07002044 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002045
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002046 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002047 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048}
2049
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002050/**
2051 * create_mnt_ns - creates a private namespace and adds a root filesystem
2052 * @mnt: pointer to the new root filesystem mountpoint
2053 */
2054struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2055{
2056 struct mnt_namespace *new_ns;
2057
2058 new_ns = alloc_mnt_ns();
2059 if (!IS_ERR(new_ns)) {
2060 mnt->mnt_ns = new_ns;
2061 new_ns->root = mnt;
2062 list_add(&new_ns->list, &new_ns->root->mnt_list);
2063 }
2064 return new_ns;
2065}
2066EXPORT_SYMBOL(create_mnt_ns);
2067
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002068SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2069 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
2071 int retval;
2072 unsigned long data_page;
2073 unsigned long type_page;
2074 unsigned long dev_page;
2075 char *dir_page;
2076
Ram Paib58fed82005-11-07 17:16:09 -05002077 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 if (retval < 0)
2079 return retval;
2080
2081 dir_page = getname(dir_name);
2082 retval = PTR_ERR(dir_page);
2083 if (IS_ERR(dir_page))
2084 goto out1;
2085
Ram Paib58fed82005-11-07 17:16:09 -05002086 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 if (retval < 0)
2088 goto out2;
2089
Ram Paib58fed82005-11-07 17:16:09 -05002090 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 if (retval < 0)
2092 goto out3;
2093
Ram Paib58fed82005-11-07 17:16:09 -05002094 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2095 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 free_page(data_page);
2097
2098out3:
2099 free_page(dev_page);
2100out2:
2101 putname(dir_page);
2102out1:
2103 free_page(type_page);
2104 return retval;
2105}
2106
2107/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 * pivot_root Semantics:
2109 * Moves the root file system of the current process to the directory put_old,
2110 * makes new_root as the new root file system of the current process, and sets
2111 * root/cwd of all processes which had them on the current root to new_root.
2112 *
2113 * Restrictions:
2114 * The new_root and put_old must be directories, and must not be on the
2115 * same file system as the current process root. The put_old must be
2116 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2117 * pointed to by put_old must yield the same directory as new_root. No other
2118 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2119 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002120 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2121 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2122 * in this situation.
2123 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 * Notes:
2125 * - we don't move root/cwd if they are not at the root (reason: if something
2126 * cared enough to change them, it's probably wrong to force them elsewhere)
2127 * - it's okay to pick a root that isn't the root of a file system, e.g.
2128 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2129 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2130 * first.
2131 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002132SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2133 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134{
2135 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002136 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 int error;
2138
2139 if (!capable(CAP_SYS_ADMIN))
2140 return -EPERM;
2141
Al Viro2d8f3032008-07-22 09:59:21 -04002142 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 if (error)
2144 goto out0;
2145 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002146 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 goto out1;
2148
Al Viro2d8f3032008-07-22 09:59:21 -04002149 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 if (error)
2151 goto out1;
2152
Al Viro2d8f3032008-07-22 09:59:21 -04002153 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002155 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 goto out1;
2157 }
2158
2159 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002160 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002161 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002163 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002164 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002166 if (IS_MNT_SHARED(old.mnt) ||
2167 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002168 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002169 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002170 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 goto out2;
2172 error = -ENOENT;
Al Viro2d8f3032008-07-22 09:59:21 -04002173 if (IS_DEADDIR(new.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002175 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002177 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 goto out2;
2179 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002180 if (new.mnt == root.mnt ||
2181 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 goto out2; /* loop, on the same file system */
2183 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002184 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002186 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002187 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002188 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002190 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002191 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002192 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002193 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002195 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 for (;;) {
2197 if (tmp->mnt_parent == tmp)
2198 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002199 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 break;
2201 tmp = tmp->mnt_parent;
2202 }
Al Viro2d8f3032008-07-22 09:59:21 -04002203 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002205 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002207 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002208 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002209 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002210 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002211 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002212 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002213 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002215 chroot_fs_refs(&root, &new);
2216 security_sb_post_pivotroot(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002218 path_put(&root_parent);
2219 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002221 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002222 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002223 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002224 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002226 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 return error;
2229out3:
2230 spin_unlock(&vfsmount_lock);
2231 goto out2;
2232}
2233
2234static void __init init_mount_tree(void)
2235{
2236 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002237 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002238 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2241 if (IS_ERR(mnt))
2242 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002243 ns = create_mnt_ns(mnt);
2244 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002247 init_task.nsproxy->mnt_ns = ns;
2248 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
Jan Blunckac748a02008-02-14 19:34:39 -08002250 root.mnt = ns->root;
2251 root.dentry = ns->root->mnt_root;
2252
2253 set_fs_pwd(current->fs, &root);
2254 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255}
2256
Denis Cheng74bf17c2007-10-16 23:26:30 -07002257void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002259 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002260 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261
Ram Pai390c6842005-11-07 17:17:51 -05002262 init_rwsem(&namespace_sem);
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002265 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Ram Paib58fed82005-11-07 17:16:09 -05002267 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
2269 if (!mount_hashtable)
2270 panic("Failed to allocate mount hash table\n");
2271
Eric Dumazet13f14b42008-02-06 01:37:57 -08002272 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Eric Dumazet13f14b42008-02-06 01:37:57 -08002274 for (u = 0; u < HASH_SIZE; u++)
2275 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002277 err = sysfs_init();
2278 if (err)
2279 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002280 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002281 fs_kobj = kobject_create_and_add("fs", NULL);
2282 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002283 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 init_rootfs();
2285 init_mount_tree();
2286}
2287
Trond Myklebust616511d2009-06-22 15:09:13 -04002288void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Trond Myklebust616511d2009-06-22 15:09:13 -04002290 struct vfsmount *root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05002291 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002292
2293 if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock))
2294 return;
2295 root = ns->root;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002296 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07002297 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002298 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05002300 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002302 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002303 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002304 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002306EXPORT_SYMBOL(put_mnt_ns);