blob: 461090a57b334c89654734a59a3782aac8f29caa [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>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040025#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/security.h>
27#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020028#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080029#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010030#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040031#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/uaccess.h>
33#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050034#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Eric Dumazet13f14b42008-02-06 01:37:57 -080037#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
38#define HASH_SIZE (1UL << HASH_SHIFT)
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050041__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42
43static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010044static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010045static DEFINE_IDA(mnt_group_ida);
Al Virof21f6222009-06-24 03:12:00 -040046static int mnt_id_start = 0;
47static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Eric Dumazetfa3536c2006-03-26 01:37:24 -080049static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050051static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060054struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
58{
Ram Paib58fed82005-11-07 17:16:09 -050059 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
60 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080061 tmp = tmp + (tmp >> HASH_SHIFT);
62 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063}
64
Dave Hansen3d733632008-02-15 14:37:59 -080065#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
66
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010067/* allocation is serialized by namespace_sem */
68static int mnt_alloc_id(struct vfsmount *mnt)
69{
70 int res;
71
72retry:
73 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
74 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040075 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
76 if (!res)
77 mnt_id_start = mnt->mnt_id + 1;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010078 spin_unlock(&vfsmount_lock);
79 if (res == -EAGAIN)
80 goto retry;
81
82 return res;
83}
84
85static void mnt_free_id(struct vfsmount *mnt)
86{
Al Virof21f6222009-06-24 03:12:00 -040087 int id = mnt->mnt_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010088 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040089 ida_remove(&mnt_id_ida, id);
90 if (mnt_id_start > id)
91 mnt_id_start = id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010092 spin_unlock(&vfsmount_lock);
93}
94
Miklos Szeredi719f5d72008-03-27 13:06:23 +010095/*
96 * Allocate a new peer group ID
97 *
98 * mnt_group_ida is protected by namespace_sem
99 */
100static int mnt_alloc_group_id(struct vfsmount *mnt)
101{
Al Virof21f6222009-06-24 03:12:00 -0400102 int res;
103
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100104 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
105 return -ENOMEM;
106
Al Virof21f6222009-06-24 03:12:00 -0400107 res = ida_get_new_above(&mnt_group_ida,
108 mnt_group_start,
109 &mnt->mnt_group_id);
110 if (!res)
111 mnt_group_start = mnt->mnt_group_id + 1;
112
113 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100114}
115
116/*
117 * Release a peer group ID
118 */
119void mnt_release_group_id(struct vfsmount *mnt)
120{
Al Virof21f6222009-06-24 03:12:00 -0400121 int id = mnt->mnt_group_id;
122 ida_remove(&mnt_group_ida, id);
123 if (mnt_group_start > id)
124 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100125 mnt->mnt_group_id = 0;
126}
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128struct vfsmount *alloc_vfsmnt(const char *name)
129{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800130 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100132 int err;
133
134 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800135 if (err)
136 goto out_free_cache;
137
138 if (name) {
139 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
140 if (!mnt->mnt_devname)
141 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100142 }
143
Ram Paib58fed82005-11-07 17:16:09 -0500144 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 INIT_LIST_HEAD(&mnt->mnt_hash);
146 INIT_LIST_HEAD(&mnt->mnt_child);
147 INIT_LIST_HEAD(&mnt->mnt_mounts);
148 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700149 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500150 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500151 INIT_LIST_HEAD(&mnt->mnt_slave_list);
152 INIT_LIST_HEAD(&mnt->mnt_slave);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000153#ifdef CONFIG_SMP
154 mnt->mnt_writers = alloc_percpu(int);
155 if (!mnt->mnt_writers)
156 goto out_free_devname;
157#else
158 mnt->mnt_writers = 0;
159#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 }
161 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800162
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000163#ifdef CONFIG_SMP
164out_free_devname:
165 kfree(mnt->mnt_devname);
166#endif
Li Zefan88b38782008-07-21 18:06:36 +0800167out_free_id:
168 mnt_free_id(mnt);
169out_free_cache:
170 kmem_cache_free(mnt_cache, mnt);
171 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
173
Dave Hansen83660252008-02-15 14:37:30 -0800174/*
175 * Most r/o checks on a fs are for operations that take
176 * discrete amounts of time, like a write() or unlink().
177 * We must keep track of when those operations start
178 * (for permission checks) and when they end, so that
179 * we can determine when writes are able to occur to
180 * a filesystem.
181 */
Dave Hansen3d733632008-02-15 14:37:59 -0800182/*
183 * __mnt_is_readonly: check whether a mount is read-only
184 * @mnt: the mount to check for its write status
185 *
186 * This shouldn't be used directly ouside of the VFS.
187 * It does not guarantee that the filesystem will stay
188 * r/w, just that it is right *now*. This can not and
189 * should not be used in place of IS_RDONLY(inode).
190 * mnt_want/drop_write() will _keep_ the filesystem
191 * r/w.
192 */
193int __mnt_is_readonly(struct vfsmount *mnt)
194{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800195 if (mnt->mnt_flags & MNT_READONLY)
196 return 1;
197 if (mnt->mnt_sb->s_flags & MS_RDONLY)
198 return 1;
199 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800200}
201EXPORT_SYMBOL_GPL(__mnt_is_readonly);
202
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000203static inline void inc_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800204{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000205#ifdef CONFIG_SMP
206 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
207#else
208 mnt->mnt_writers++;
209#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800210}
Dave Hansen3d733632008-02-15 14:37:59 -0800211
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000212static inline void dec_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800213{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000214#ifdef CONFIG_SMP
215 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
216#else
217 mnt->mnt_writers--;
218#endif
219}
220
221static unsigned int count_mnt_writers(struct vfsmount *mnt)
222{
223#ifdef CONFIG_SMP
224 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800225 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800226
227 for_each_possible_cpu(cpu) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000228 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800229 }
Dave Hansen3d733632008-02-15 14:37:59 -0800230
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000231 return count;
232#else
233 return mnt->mnt_writers;
234#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800235}
236
237/*
238 * Most r/o checks on a fs are for operations that take
239 * discrete amounts of time, like a write() or unlink().
240 * We must keep track of when those operations start
241 * (for permission checks) and when they end, so that
242 * we can determine when writes are able to occur to
243 * a filesystem.
244 */
Dave Hansen83660252008-02-15 14:37:30 -0800245/**
246 * mnt_want_write - get write access to a mount
247 * @mnt: the mount on which to take a write
248 *
249 * This tells the low-level filesystem that a write is
250 * about to be performed to it, and makes sure that
251 * writes are allowed before returning success. When
252 * the write operation is finished, mnt_drop_write()
253 * must be called. This is effectively a refcount.
254 */
255int mnt_want_write(struct vfsmount *mnt)
256{
Dave Hansen3d733632008-02-15 14:37:59 -0800257 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800258
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000259 preempt_disable();
260 inc_mnt_writers(mnt);
261 /*
262 * The store to inc_mnt_writers must be visible before we pass
263 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
264 * incremented count after it has set MNT_WRITE_HOLD.
265 */
266 smp_mb();
267 while (mnt->mnt_flags & MNT_WRITE_HOLD)
268 cpu_relax();
269 /*
270 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
271 * be set to match its requirements. So we must not load that until
272 * MNT_WRITE_HOLD is cleared.
273 */
274 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800275 if (__mnt_is_readonly(mnt)) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000276 dec_mnt_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800277 ret = -EROFS;
278 goto out;
279 }
Dave Hansen3d733632008-02-15 14:37:59 -0800280out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000281 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800282 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800283}
284EXPORT_SYMBOL_GPL(mnt_want_write);
285
286/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000287 * mnt_clone_write - get write access to a mount
288 * @mnt: the mount on which to take a write
289 *
290 * This is effectively like mnt_want_write, except
291 * it must only be used to take an extra write reference
292 * on a mountpoint that we already know has a write reference
293 * on it. This allows some optimisation.
294 *
295 * After finished, mnt_drop_write must be called as usual to
296 * drop the reference.
297 */
298int mnt_clone_write(struct vfsmount *mnt)
299{
300 /* superblock may be r/o */
301 if (__mnt_is_readonly(mnt))
302 return -EROFS;
303 preempt_disable();
304 inc_mnt_writers(mnt);
305 preempt_enable();
306 return 0;
307}
308EXPORT_SYMBOL_GPL(mnt_clone_write);
309
310/**
311 * mnt_want_write_file - get write access to a file's mount
312 * @file: the file who's mount on which to take a write
313 *
314 * This is like mnt_want_write, but it takes a file and can
315 * do some optimisations if the file is open for write already
316 */
317int mnt_want_write_file(struct file *file)
318{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700319 struct inode *inode = file->f_dentry->d_inode;
320 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000321 return mnt_want_write(file->f_path.mnt);
322 else
323 return mnt_clone_write(file->f_path.mnt);
324}
325EXPORT_SYMBOL_GPL(mnt_want_write_file);
326
327/**
Dave Hansen83660252008-02-15 14:37:30 -0800328 * mnt_drop_write - give up write access to a mount
329 * @mnt: the mount on which to give up write access
330 *
331 * Tells the low-level filesystem that we are done
332 * performing writes to it. Must be matched with
333 * mnt_want_write() call above.
334 */
335void mnt_drop_write(struct vfsmount *mnt)
336{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000337 preempt_disable();
338 dec_mnt_writers(mnt);
339 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800340}
341EXPORT_SYMBOL_GPL(mnt_drop_write);
342
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800343static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800344{
Dave Hansen3d733632008-02-15 14:37:59 -0800345 int ret = 0;
346
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800347 spin_lock(&vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000348 mnt->mnt_flags |= MNT_WRITE_HOLD;
349 /*
350 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
351 * should be visible before we do.
352 */
353 smp_mb();
354
355 /*
356 * With writers on hold, if this value is zero, then there are
357 * definitely no active writers (although held writers may subsequently
358 * increment the count, they'll have to wait, and decrement it after
359 * seeing MNT_READONLY).
360 *
361 * It is OK to have counter incremented on one CPU and decremented on
362 * another: the sum will add up correctly. The danger would be when we
363 * sum up each counter, if we read a counter before it is incremented,
364 * but then read another CPU's count which it has been subsequently
365 * decremented from -- we would see more decrements than we should.
366 * MNT_WRITE_HOLD protects against this scenario, because
367 * mnt_want_write first increments count, then smp_mb, then spins on
368 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
369 * we're counting up here.
370 */
371 if (count_mnt_writers(mnt) > 0)
372 ret = -EBUSY;
373 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800374 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000375 /*
376 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
377 * that become unheld will see MNT_READONLY.
378 */
379 smp_wmb();
380 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800381 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800382 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800383}
Dave Hansen83660252008-02-15 14:37:30 -0800384
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800385static void __mnt_unmake_readonly(struct vfsmount *mnt)
386{
387 spin_lock(&vfsmount_lock);
388 mnt->mnt_flags &= ~MNT_READONLY;
389 spin_unlock(&vfsmount_lock);
390}
391
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800392void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700393{
394 mnt->mnt_sb = sb;
395 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700396}
397
398EXPORT_SYMBOL(simple_set_mnt);
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400void free_vfsmnt(struct vfsmount *mnt)
401{
402 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100403 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000404#ifdef CONFIG_SMP
405 free_percpu(mnt->mnt_writers);
406#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 kmem_cache_free(mnt_cache, mnt);
408}
409
410/*
Ram Paia05964f2005-11-07 17:20:17 -0500411 * find the first or last mount at @dentry on vfsmount @mnt depending on
412 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 */
Ram Paia05964f2005-11-07 17:20:17 -0500414struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
415 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416{
Ram Paib58fed82005-11-07 17:16:09 -0500417 struct list_head *head = mount_hashtable + hash(mnt, dentry);
418 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 struct vfsmount *p, *found = NULL;
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500422 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 p = NULL;
424 if (tmp == head)
425 break;
426 p = list_entry(tmp, struct vfsmount, mnt_hash);
427 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500428 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 break;
430 }
431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 return found;
433}
434
Ram Paia05964f2005-11-07 17:20:17 -0500435/*
436 * lookup_mnt increments the ref count before returning
437 * the vfsmount struct.
438 */
Al Viro1c755af2009-04-18 14:06:57 -0400439struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500440{
441 struct vfsmount *child_mnt;
442 spin_lock(&vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400443 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500444 mntget(child_mnt);
445 spin_unlock(&vfsmount_lock);
446 return child_mnt;
447}
448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449static inline int check_mnt(struct vfsmount *mnt)
450{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800451 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800454static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500455{
456 if (ns) {
457 ns->event = ++event;
458 wake_up_interruptible(&ns->poll);
459 }
460}
461
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800462static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500463{
464 if (ns && ns->event != event) {
465 ns->event = event;
466 wake_up_interruptible(&ns->poll);
467 }
468}
469
Al Viro1a390682008-03-21 20:48:19 -0400470static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Al Viro1a390682008-03-21 20:48:19 -0400472 old_path->dentry = mnt->mnt_mountpoint;
473 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 mnt->mnt_parent = mnt;
475 mnt->mnt_mountpoint = mnt->mnt_root;
476 list_del_init(&mnt->mnt_child);
477 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400478 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
Ram Paib90fa9a2005-11-07 17:19:50 -0500481void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
482 struct vfsmount *child_mnt)
483{
484 child_mnt->mnt_parent = mntget(mnt);
485 child_mnt->mnt_mountpoint = dget(dentry);
486 dentry->d_mounted++;
487}
488
Al Viro1a390682008-03-21 20:48:19 -0400489static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490{
Al Viro1a390682008-03-21 20:48:19 -0400491 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500492 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400493 hash(path->mnt, path->dentry));
494 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500495}
496
497/*
498 * the caller must hold vfsmount_lock
499 */
500static void commit_tree(struct vfsmount *mnt)
501{
502 struct vfsmount *parent = mnt->mnt_parent;
503 struct vfsmount *m;
504 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800505 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500506
507 BUG_ON(parent == mnt);
508
509 list_add_tail(&head, &mnt->mnt_list);
510 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800511 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500512 list_splice(&head, n->list.prev);
513
514 list_add_tail(&mnt->mnt_hash, mount_hashtable +
515 hash(parent, mnt->mnt_mountpoint));
516 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800517 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
519
520static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
521{
522 struct list_head *next = p->mnt_mounts.next;
523 if (next == &p->mnt_mounts) {
524 while (1) {
525 if (p == root)
526 return NULL;
527 next = p->mnt_child.next;
528 if (next != &p->mnt_parent->mnt_mounts)
529 break;
530 p = p->mnt_parent;
531 }
532 }
533 return list_entry(next, struct vfsmount, mnt_child);
534}
535
Ram Pai9676f0c2005-11-07 17:21:20 -0500536static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
537{
538 struct list_head *prev = p->mnt_mounts.prev;
539 while (prev != &p->mnt_mounts) {
540 p = list_entry(prev, struct vfsmount, mnt_child);
541 prev = p->mnt_mounts.prev;
542 }
543 return p;
544}
545
Ram Pai36341f62005-11-07 17:17:22 -0500546static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
547 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549 struct super_block *sb = old->mnt_sb;
550 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
551
552 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100553 if (flag & (CL_SLAVE | CL_PRIVATE))
554 mnt->mnt_group_id = 0; /* not a peer of original */
555 else
556 mnt->mnt_group_id = old->mnt_group_id;
557
558 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
559 int err = mnt_alloc_group_id(mnt);
560 if (err)
561 goto out_free;
562 }
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 mnt->mnt_flags = old->mnt_flags;
565 atomic_inc(&sb->s_active);
566 mnt->mnt_sb = sb;
567 mnt->mnt_root = dget(root);
568 mnt->mnt_mountpoint = mnt->mnt_root;
569 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500570
Ram Pai5afe0022005-11-07 17:21:01 -0500571 if (flag & CL_SLAVE) {
572 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
573 mnt->mnt_master = old;
574 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400575 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500576 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
577 list_add(&mnt->mnt_share, &old->mnt_share);
578 if (IS_MNT_SLAVE(old))
579 list_add(&mnt->mnt_slave, &old->mnt_slave);
580 mnt->mnt_master = old->mnt_master;
581 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500582 if (flag & CL_MAKE_SHARED)
583 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 /* stick the duplicate mount on the same expiry list
586 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500587 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500588 if (!list_empty(&old->mnt_expire))
589 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
592 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100593
594 out_free:
595 free_vfsmnt(mnt);
596 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
Al Viro7b7b1ac2005-11-07 17:13:39 -0500599static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800602 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800603 * This probably indicates that somebody messed
604 * up a mnt_want/drop_write() pair. If this
605 * happens, the filesystem was probably unable
606 * to make r/w->r/o transitions.
607 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000608 /*
609 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
610 * provides barriers, so count_mnt_writers() below is safe. AV
611 */
612 WARN_ON(count_mnt_writers(mnt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 dput(mnt->mnt_root);
614 free_vfsmnt(mnt);
615 deactivate_super(sb);
616}
617
Al Viro7b7b1ac2005-11-07 17:13:39 -0500618void mntput_no_expire(struct vfsmount *mnt)
619{
620repeat:
621 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
622 if (likely(!mnt->mnt_pinned)) {
623 spin_unlock(&vfsmount_lock);
624 __mntput(mnt);
625 return;
626 }
627 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
628 mnt->mnt_pinned = 0;
629 spin_unlock(&vfsmount_lock);
630 acct_auto_close_mnt(mnt);
631 security_sb_umount_close(mnt);
632 goto repeat;
633 }
634}
635
636EXPORT_SYMBOL(mntput_no_expire);
637
638void mnt_pin(struct vfsmount *mnt)
639{
640 spin_lock(&vfsmount_lock);
641 mnt->mnt_pinned++;
642 spin_unlock(&vfsmount_lock);
643}
644
645EXPORT_SYMBOL(mnt_pin);
646
647void mnt_unpin(struct vfsmount *mnt)
648{
649 spin_lock(&vfsmount_lock);
650 if (mnt->mnt_pinned) {
651 atomic_inc(&mnt->mnt_count);
652 mnt->mnt_pinned--;
653 }
654 spin_unlock(&vfsmount_lock);
655}
656
657EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800659static inline void mangle(struct seq_file *m, const char *s)
660{
661 seq_escape(m, s, " \t\n\\");
662}
663
664/*
665 * Simple .show_options callback for filesystems which don't want to
666 * implement more complex mount option showing.
667 *
668 * See also save_mount_options().
669 */
670int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
671{
Al Viro2a32ceb2009-05-08 16:05:57 -0400672 const char *options;
673
674 rcu_read_lock();
675 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800676
677 if (options != NULL && options[0]) {
678 seq_putc(m, ',');
679 mangle(m, options);
680 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400681 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800682
683 return 0;
684}
685EXPORT_SYMBOL(generic_show_options);
686
687/*
688 * If filesystem uses generic_show_options(), this function should be
689 * called from the fill_super() callback.
690 *
691 * The .remount_fs callback usually needs to be handled in a special
692 * way, to make sure, that previous options are not overwritten if the
693 * remount fails.
694 *
695 * Also note, that if the filesystem's .remount_fs function doesn't
696 * reset all options to their default value, but changes only newly
697 * given options, then the displayed options will not reflect reality
698 * any more.
699 */
700void save_mount_options(struct super_block *sb, char *options)
701{
Al Viro2a32ceb2009-05-08 16:05:57 -0400702 BUG_ON(sb->s_options);
703 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800704}
705EXPORT_SYMBOL(save_mount_options);
706
Al Viro2a32ceb2009-05-08 16:05:57 -0400707void replace_mount_options(struct super_block *sb, char *options)
708{
709 char *old = sb->s_options;
710 rcu_assign_pointer(sb->s_options, options);
711 if (old) {
712 synchronize_rcu();
713 kfree(old);
714 }
715}
716EXPORT_SYMBOL(replace_mount_options);
717
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100718#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/* iterator */
720static void *m_start(struct seq_file *m, loff_t *pos)
721{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100722 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Ram Pai390c6842005-11-07 17:17:51 -0500724 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100725 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
728static void *m_next(struct seq_file *m, void *v, loff_t *pos)
729{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100730 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700731
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100732 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
735static void m_stop(struct seq_file *m, void *v)
736{
Ram Pai390c6842005-11-07 17:17:51 -0500737 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
739
Ram Pai2d4d4862008-03-27 13:06:25 +0100740struct proc_fs_info {
741 int flag;
742 const char *str;
743};
744
Eric Paris2069f452008-07-04 09:47:13 +1000745static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Ram Pai2d4d4862008-03-27 13:06:25 +0100747 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 { MS_SYNCHRONOUS, ",sync" },
749 { MS_DIRSYNC, ",dirsync" },
750 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 { 0, NULL }
752 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100753 const struct proc_fs_info *fs_infop;
754
755 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
756 if (sb->s_flags & fs_infop->flag)
757 seq_puts(m, fs_infop->str);
758 }
Eric Paris2069f452008-07-04 09:47:13 +1000759
760 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100761}
762
763static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
764{
765 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 { MNT_NOSUID, ",nosuid" },
767 { MNT_NODEV, ",nodev" },
768 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800769 { MNT_NOATIME, ",noatime" },
770 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800771 { MNT_RELATIME, ",relatime" },
Matthew Garrettd0adde52009-03-26 17:49:56 +0000772 { MNT_STRICTATIME, ",strictatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 { 0, NULL }
774 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100775 const struct proc_fs_info *fs_infop;
776
777 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
778 if (mnt->mnt_flags & fs_infop->flag)
779 seq_puts(m, fs_infop->str);
780 }
781}
782
783static void show_type(struct seq_file *m, struct super_block *sb)
784{
785 mangle(m, sb->s_type->name);
786 if (sb->s_subtype && sb->s_subtype[0]) {
787 seq_putc(m, '.');
788 mangle(m, sb->s_subtype);
789 }
790}
791
792static int show_vfsmnt(struct seq_file *m, void *v)
793{
794 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
795 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800796 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
799 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800800 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100802 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800803 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000804 err = show_sb_opts(m, mnt->mnt_sb);
805 if (err)
806 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100807 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (mnt->mnt_sb->s_op->show_options)
809 err = mnt->mnt_sb->s_op->show_options(m, mnt);
810 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000811out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return err;
813}
814
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100815const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 .start = m_start,
817 .next = m_next,
818 .stop = m_stop,
819 .show = show_vfsmnt
820};
821
Ram Pai2d4d4862008-03-27 13:06:25 +0100822static int show_mountinfo(struct seq_file *m, void *v)
823{
824 struct proc_mounts *p = m->private;
825 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
826 struct super_block *sb = mnt->mnt_sb;
827 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
828 struct path root = p->root;
829 int err = 0;
830
831 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
832 MAJOR(sb->s_dev), MINOR(sb->s_dev));
833 seq_dentry(m, mnt->mnt_root, " \t\n\\");
834 seq_putc(m, ' ');
835 seq_path_root(m, &mnt_path, &root, " \t\n\\");
836 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
837 /*
838 * Mountpoint is outside root, discard that one. Ugly,
839 * but less so than trying to do that in iterator in a
840 * race-free way (due to renames).
841 */
842 return SEQ_SKIP;
843 }
844 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
845 show_mnt_opts(m, mnt);
846
847 /* Tagged fields ("foo:X" or "bar") */
848 if (IS_MNT_SHARED(mnt))
849 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100850 if (IS_MNT_SLAVE(mnt)) {
851 int master = mnt->mnt_master->mnt_group_id;
852 int dom = get_dominating_id(mnt, &p->root);
853 seq_printf(m, " master:%i", master);
854 if (dom && dom != master)
855 seq_printf(m, " propagate_from:%i", dom);
856 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100857 if (IS_MNT_UNBINDABLE(mnt))
858 seq_puts(m, " unbindable");
859
860 /* Filesystem specific data */
861 seq_puts(m, " - ");
862 show_type(m, sb);
863 seq_putc(m, ' ');
864 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
865 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000866 err = show_sb_opts(m, sb);
867 if (err)
868 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100869 if (sb->s_op->show_options)
870 err = sb->s_op->show_options(m, mnt);
871 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000872out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100873 return err;
874}
875
876const struct seq_operations mountinfo_op = {
877 .start = m_start,
878 .next = m_next,
879 .stop = m_stop,
880 .show = show_mountinfo,
881};
882
Chuck Leverb4629fe2006-03-20 13:44:12 -0500883static int show_vfsstat(struct seq_file *m, void *v)
884{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700885 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800886 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500887 int err = 0;
888
889 /* device */
890 if (mnt->mnt_devname) {
891 seq_puts(m, "device ");
892 mangle(m, mnt->mnt_devname);
893 } else
894 seq_puts(m, "no device");
895
896 /* mount point */
897 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800898 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500899 seq_putc(m, ' ');
900
901 /* file system type */
902 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100903 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500904
905 /* optional statistics */
906 if (mnt->mnt_sb->s_op->show_stats) {
907 seq_putc(m, ' ');
908 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
909 }
910
911 seq_putc(m, '\n');
912 return err;
913}
914
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100915const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500916 .start = m_start,
917 .next = m_next,
918 .stop = m_stop,
919 .show = show_vfsstat,
920};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100921#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923/**
924 * may_umount_tree - check if a mount tree is busy
925 * @mnt: root of mount tree
926 *
927 * This is called to check if a tree of mounts has any
928 * open files, pwds, chroots or sub mounts that are
929 * busy.
930 */
931int may_umount_tree(struct vfsmount *mnt)
932{
Ram Pai36341f62005-11-07 17:17:22 -0500933 int actual_refs = 0;
934 int minimum_refs = 0;
935 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500938 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 actual_refs += atomic_read(&p->mnt_count);
940 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 }
942 spin_unlock(&vfsmount_lock);
943
944 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800945 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Ian Kente3474a82006-03-27 01:14:51 -0800947 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948}
949
950EXPORT_SYMBOL(may_umount_tree);
951
952/**
953 * may_umount - check if a mount point is busy
954 * @mnt: root of mount
955 *
956 * This is called to check if a mount point has any
957 * open files, pwds, chroots or sub mounts. If the
958 * mount has sub mounts this will return busy
959 * regardless of whether the sub mounts are busy.
960 *
961 * Doesn't take quota and stuff into account. IOW, in some cases it will
962 * give false negatives. The main reason why it's here is that we need
963 * a non-destructive way to look for easily umountable filesystems.
964 */
965int may_umount(struct vfsmount *mnt)
966{
Ian Kente3474a82006-03-27 01:14:51 -0800967 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -0500968 down_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500969 spin_lock(&vfsmount_lock);
970 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800971 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500972 spin_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -0500973 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500974 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
977EXPORT_SYMBOL(may_umount);
978
Ram Paib90fa9a2005-11-07 17:19:50 -0500979void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500981 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800982 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700983 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500984 list_del_init(&mnt->mnt_hash);
985 if (mnt->mnt_parent != mnt) {
986 struct dentry *dentry;
987 struct vfsmount *m;
988 spin_lock(&vfsmount_lock);
989 dentry = mnt->mnt_mountpoint;
990 m = mnt->mnt_parent;
991 mnt->mnt_mountpoint = mnt->mnt_root;
992 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -0400993 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500995 dput(dentry);
996 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 }
998 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500999 }
1000}
1001
Ram Paia05964f2005-11-07 17:20:17 -05001002void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001003{
1004 struct vfsmount *p;
1005
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001006 for (p = mnt; p; p = next_mnt(p, mnt))
1007 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001008
Ram Paia05964f2005-11-07 17:20:17 -05001009 if (propagate)
1010 propagate_umount(kill);
1011
Ram Pai70fbcdf2005-11-07 17:17:04 -05001012 list_for_each_entry(p, kill, mnt_hash) {
1013 list_del_init(&p->mnt_expire);
1014 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001015 __touch_mnt_namespace(p->mnt_ns);
1016 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001017 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001018 if (p->mnt_parent != p) {
1019 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001020 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001021 }
Ram Paia05964f2005-11-07 17:20:17 -05001022 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
1024}
1025
Al Viroc35038b2008-03-22 00:46:23 -04001026static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028static int do_umount(struct vfsmount *mnt, int flags)
1029{
Ram Paib58fed82005-11-07 17:16:09 -05001030 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001032 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 retval = security_sb_umount(mnt, flags);
1035 if (retval)
1036 return retval;
1037
1038 /*
1039 * Allow userspace to request a mountpoint be expired rather than
1040 * unmounting unconditionally. Unmount only happens if:
1041 * (1) the mark is already set (the mark is cleared by mntput())
1042 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1043 */
1044 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001045 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 flags & (MNT_FORCE | MNT_DETACH))
1047 return -EINVAL;
1048
1049 if (atomic_read(&mnt->mnt_count) != 2)
1050 return -EBUSY;
1051
1052 if (!xchg(&mnt->mnt_expiry_mark, 1))
1053 return -EAGAIN;
1054 }
1055
1056 /*
1057 * If we may have to abort operations to get out of this
1058 * mount, and they will themselves hold resources we must
1059 * allow the fs to do things. In the Unix tradition of
1060 * 'Gee thats tricky lets do it in userspace' the umount_begin
1061 * might fail to complete on the first run through as other tasks
1062 * must return, and the like. Thats for the mount program to worry
1063 * about for the moment.
1064 */
1065
Al Viro42faad92008-04-24 07:21:56 -04001066 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001067 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
1070 /*
1071 * No sense to grab the lock for this test, but test itself looks
1072 * somewhat bogus. Suggestions for better replacement?
1073 * Ho-hum... In principle, we might treat that as umount + switch
1074 * to rootfs. GC would eventually take care of the old vfsmount.
1075 * Actually it makes sense, especially if rootfs would contain a
1076 * /reboot - static binary that would close all descriptors and
1077 * call reboot(9). Then init(8) could umount root and exec /reboot.
1078 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001079 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 /*
1081 * Special case for "unmounting" root ...
1082 * we just try to remount it readonly.
1083 */
1084 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001085 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 up_write(&sb->s_umount);
1088 return retval;
1089 }
1090
Ram Pai390c6842005-11-07 17:17:51 -05001091 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001093 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Al Viroc35038b2008-03-22 00:46:23 -04001095 if (!(flags & MNT_DETACH))
1096 shrink_submounts(mnt, &umount_list);
1097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001099 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001101 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 retval = 0;
1103 }
1104 spin_unlock(&vfsmount_lock);
1105 if (retval)
1106 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001107 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001108 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 return retval;
1110}
1111
1112/*
1113 * Now umount can handle mount points as well as block devices.
1114 * This is important for filesystems which use unnamed block devices.
1115 *
1116 * We now support a flag for forced unmount like the other 'big iron'
1117 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1118 */
1119
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001120SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Al Viro2d8f3032008-07-22 09:59:21 -04001122 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 int retval;
1124
Al Viro2d8f3032008-07-22 09:59:21 -04001125 retval = user_path(name, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (retval)
1127 goto out;
1128 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001129 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001131 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 goto dput_and_out;
1133
1134 retval = -EPERM;
1135 if (!capable(CAP_SYS_ADMIN))
1136 goto dput_and_out;
1137
Al Viro2d8f3032008-07-22 09:59:21 -04001138 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001140 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001141 dput(path.dentry);
1142 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143out:
1144 return retval;
1145}
1146
1147#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1148
1149/*
Ram Paib58fed82005-11-07 17:16:09 -05001150 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001152SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153{
Ram Paib58fed82005-11-07 17:16:09 -05001154 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
1157#endif
1158
Al Viro2d92ab32008-08-02 00:51:11 -04001159static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
1161 if (capable(CAP_SYS_ADMIN))
1162 return 0;
1163 return -EPERM;
1164#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001165 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001167 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001168 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 return -EPERM;
1170 }
Al Viro2d92ab32008-08-02 00:51:11 -04001171 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 return -EPERM;
1173 return 0;
1174#endif
1175}
1176
Ram Paib90fa9a2005-11-07 17:19:50 -05001177struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001178 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179{
1180 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001181 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
Ram Pai9676f0c2005-11-07 17:21:20 -05001183 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1184 return NULL;
1185
Ram Pai36341f62005-11-07 17:17:22 -05001186 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (!q)
1188 goto Enomem;
1189 q->mnt_mountpoint = mnt->mnt_mountpoint;
1190
1191 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001192 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001193 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 continue;
1195
1196 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001197 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1198 s = skip_mnt_tree(s);
1199 continue;
1200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 while (p != s->mnt_parent) {
1202 p = p->mnt_parent;
1203 q = q->mnt_parent;
1204 }
1205 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001206 path.mnt = q;
1207 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001208 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (!q)
1210 goto Enomem;
1211 spin_lock(&vfsmount_lock);
1212 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001213 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 spin_unlock(&vfsmount_lock);
1215 }
1216 }
1217 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001218Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001220 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001222 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001224 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 }
1226 return NULL;
1227}
1228
Al Viro589ff872009-04-18 03:28:19 -04001229struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001230{
1231 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001232 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001233 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001234 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001235 return tree;
1236}
1237
1238void drop_collected_mounts(struct vfsmount *mnt)
1239{
1240 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001241 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001242 spin_lock(&vfsmount_lock);
1243 umount_tree(mnt, 0, &umount_list);
1244 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001245 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001246 release_mounts(&umount_list);
1247}
1248
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001249static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1250{
1251 struct vfsmount *p;
1252
1253 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1254 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1255 mnt_release_group_id(p);
1256 }
1257}
1258
1259static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1260{
1261 struct vfsmount *p;
1262
1263 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1264 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1265 int err = mnt_alloc_group_id(p);
1266 if (err) {
1267 cleanup_group_ids(mnt, p);
1268 return err;
1269 }
1270 }
1271 }
1272
1273 return 0;
1274}
1275
Ram Paib90fa9a2005-11-07 17:19:50 -05001276/*
1277 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001278 * @nd : place the mount tree @source_mnt is attached
1279 * @parent_nd : if non-null, detach the source_mnt from its parent and
1280 * store the parent mount and mountpoint dentry.
1281 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001282 *
1283 * NOTE: in the table below explains the semantics when a source mount
1284 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001285 * ---------------------------------------------------------------------------
1286 * | BIND MOUNT OPERATION |
1287 * |**************************************************************************
1288 * | source-->| shared | private | slave | unbindable |
1289 * | dest | | | | |
1290 * | | | | | | |
1291 * | v | | | | |
1292 * |**************************************************************************
1293 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1294 * | | | | | |
1295 * |non-shared| shared (+) | private | slave (*) | invalid |
1296 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001297 * A bind operation clones the source mount and mounts the clone on the
1298 * destination mount.
1299 *
1300 * (++) the cloned mount is propagated to all the mounts in the propagation
1301 * tree of the destination mount and the cloned mount is added to
1302 * the peer group of the source mount.
1303 * (+) the cloned mount is created under the destination mount and is marked
1304 * as shared. The cloned mount is added to the peer group of the source
1305 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001306 * (+++) the mount is propagated to all the mounts in the propagation tree
1307 * of the destination mount and the cloned mount is made slave
1308 * of the same master as that of the source mount. The cloned mount
1309 * is marked as 'shared and slave'.
1310 * (*) the cloned mount is made a slave of the same master as that of the
1311 * source mount.
1312 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001313 * ---------------------------------------------------------------------------
1314 * | MOVE MOUNT OPERATION |
1315 * |**************************************************************************
1316 * | source-->| shared | private | slave | unbindable |
1317 * | dest | | | | |
1318 * | | | | | | |
1319 * | v | | | | |
1320 * |**************************************************************************
1321 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1322 * | | | | | |
1323 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1324 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001325 *
1326 * (+) the mount is moved to the destination. And is then propagated to
1327 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001328 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001329 * (+++) the mount is moved to the destination and is then propagated to
1330 * all the mounts belonging to the destination mount's propagation tree.
1331 * the mount is marked as 'shared and slave'.
1332 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001333 *
1334 * if the source mount is a tree, the operations explained above is
1335 * applied to each mount in the tree.
1336 * Must be called without spinlocks held, since this function can sleep
1337 * in allocations.
1338 */
1339static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001340 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001341{
1342 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001343 struct vfsmount *dest_mnt = path->mnt;
1344 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001345 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001346 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001347
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001348 if (IS_MNT_SHARED(dest_mnt)) {
1349 err = invent_group_ids(source_mnt, true);
1350 if (err)
1351 goto out;
1352 }
1353 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1354 if (err)
1355 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001356
1357 if (IS_MNT_SHARED(dest_mnt)) {
1358 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1359 set_mnt_shared(p);
1360 }
1361
1362 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001363 if (parent_path) {
1364 detach_mnt(source_mnt, parent_path);
1365 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001366 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001367 } else {
1368 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1369 commit_tree(source_mnt);
1370 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001371
1372 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1373 list_del_init(&child->mnt_hash);
1374 commit_tree(child);
1375 }
1376 spin_unlock(&vfsmount_lock);
1377 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001378
1379 out_cleanup_ids:
1380 if (IS_MNT_SHARED(dest_mnt))
1381 cleanup_group_ids(source_mnt, NULL);
1382 out:
1383 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001384}
1385
Al Viro8c3ee422008-03-22 18:00:39 -04001386static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
1388 int err;
1389 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1390 return -EINVAL;
1391
Al Viro8c3ee422008-03-22 18:00:39 -04001392 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1394 return -ENOTDIR;
1395
1396 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001397 mutex_lock(&path->dentry->d_inode->i_mutex);
1398 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 goto out_unlock;
1400
Al Viro8c3ee422008-03-22 18:00:39 -04001401 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (err)
1403 goto out_unlock;
1404
1405 err = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001406 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001407 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001409 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001411 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return err;
1413}
1414
1415/*
Ram Pai07b20882005-11-07 17:19:07 -05001416 * recursively change the type of the mountpoint.
1417 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001418static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001419{
Al Viro2d92ab32008-08-02 00:51:11 -04001420 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001421 int recurse = flag & MS_REC;
1422 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001423 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001424
Miklos Szerediee6f9582007-05-08 00:30:40 -07001425 if (!capable(CAP_SYS_ADMIN))
1426 return -EPERM;
1427
Al Viro2d92ab32008-08-02 00:51:11 -04001428 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001429 return -EINVAL;
1430
1431 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001432 if (type == MS_SHARED) {
1433 err = invent_group_ids(mnt, recurse);
1434 if (err)
1435 goto out_unlock;
1436 }
1437
Ram Pai07b20882005-11-07 17:19:07 -05001438 spin_lock(&vfsmount_lock);
1439 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1440 change_mnt_propagation(m, type);
1441 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001442
1443 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001444 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001445 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001446}
1447
1448/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 * do loopback mount.
1450 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001451static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001452 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
Al Viro2d92ab32008-08-02 00:51:11 -04001454 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001456 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 if (err)
1458 return err;
1459 if (!old_name || !*old_name)
1460 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001461 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 if (err)
1463 return err;
1464
Ram Pai390c6842005-11-07 17:17:51 -05001465 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001467 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001468 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001469
Al Viro2d92ab32008-08-02 00:51:11 -04001470 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001471 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Al Viroccd48bc2005-11-07 17:15:04 -05001473 err = -ENOMEM;
1474 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001475 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001476 else
Al Viro2d92ab32008-08-02 00:51:11 -04001477 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001478
1479 if (!mnt)
1480 goto out;
1481
Al Viro2d92ab32008-08-02 00:51:11 -04001482 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001483 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001484 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001486 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001488 release_mounts(&umount_list);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Al Viroccd48bc2005-11-07 17:15:04 -05001491out:
Ram Pai390c6842005-11-07 17:17:51 -05001492 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001493 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 return err;
1495}
1496
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001497static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1498{
1499 int error = 0;
1500 int readonly_request = 0;
1501
1502 if (ms_flags & MS_RDONLY)
1503 readonly_request = 1;
1504 if (readonly_request == __mnt_is_readonly(mnt))
1505 return 0;
1506
1507 if (readonly_request)
1508 error = mnt_make_readonly(mnt);
1509 else
1510 __mnt_unmake_readonly(mnt);
1511 return error;
1512}
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514/*
1515 * change filesystem flags. dir should be a physical root of filesystem.
1516 * If you've mounted a non-root directory somewhere and want to do remount
1517 * on it - tough luck.
1518 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001519static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 void *data)
1521{
1522 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001523 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
1525 if (!capable(CAP_SYS_ADMIN))
1526 return -EPERM;
1527
Al Viro2d92ab32008-08-02 00:51:11 -04001528 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return -EINVAL;
1530
Al Viro2d92ab32008-08-02 00:51:11 -04001531 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return -EINVAL;
1533
1534 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001535 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001536 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001537 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001538 err = do_remount_sb(sb, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (!err)
Al Viro2d92ab32008-08-02 00:51:11 -04001540 path->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001542 if (!err) {
Al Viro2d92ab32008-08-02 00:51:11 -04001543 security_sb_post_remount(path->mnt, flags, data);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001544
1545 spin_lock(&vfsmount_lock);
1546 touch_mnt_namespace(path->mnt->mnt_ns);
1547 spin_unlock(&vfsmount_lock);
1548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 return err;
1550}
1551
Ram Pai9676f0c2005-11-07 17:21:20 -05001552static inline int tree_contains_unbindable(struct vfsmount *mnt)
1553{
1554 struct vfsmount *p;
1555 for (p = mnt; p; p = next_mnt(p, mnt)) {
1556 if (IS_MNT_UNBINDABLE(p))
1557 return 1;
1558 }
1559 return 0;
1560}
1561
Al Viro0a0d8a42008-08-02 00:55:27 -04001562static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Al Viro2d92ab32008-08-02 00:51:11 -04001564 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct vfsmount *p;
1566 int err = 0;
1567 if (!capable(CAP_SYS_ADMIN))
1568 return -EPERM;
1569 if (!old_name || !*old_name)
1570 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001571 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (err)
1573 return err;
1574
Ram Pai390c6842005-11-07 17:17:51 -05001575 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001576 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001577 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 ;
1579 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001580 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 goto out;
1582
1583 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001584 mutex_lock(&path->dentry->d_inode->i_mutex);
1585 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 goto out1;
1587
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001588 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001589 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001592 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001593 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
Al Viro2d92ab32008-08-02 00:51:11 -04001595 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001596 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Al Viro2d92ab32008-08-02 00:51:11 -04001598 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1599 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001600 goto out1;
1601 /*
1602 * Don't move a mount residing in a shared parent.
1603 */
Al Viro2d92ab32008-08-02 00:51:11 -04001604 if (old_path.mnt->mnt_parent &&
1605 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001606 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001607 /*
1608 * Don't move a mount tree containing unbindable mounts to a destination
1609 * mount which is shared.
1610 */
Al Viro2d92ab32008-08-02 00:51:11 -04001611 if (IS_MNT_SHARED(path->mnt) &&
1612 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001613 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001615 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1616 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001617 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Al Viro2d92ab32008-08-02 00:51:11 -04001619 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001620 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001621 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
1623 /* if the mount is moved, it should no longer be expire
1624 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001625 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001627 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628out:
Ram Pai390c6842005-11-07 17:17:51 -05001629 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001631 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001632 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 return err;
1634}
1635
1636/*
1637 * create a new mount for userspace and request it to be added into the
1638 * namespace's tree
1639 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001640static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 int mnt_flags, char *name, void *data)
1642{
1643 struct vfsmount *mnt;
1644
Vegard Nossumeca6f532009-09-18 13:05:45 -07001645 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return -EINVAL;
1647
1648 /* we need capabilities... */
1649 if (!capable(CAP_SYS_ADMIN))
1650 return -EPERM;
1651
Al Viro7f78d4c2009-05-08 13:34:06 -04001652 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001654 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 if (IS_ERR(mnt))
1656 return PTR_ERR(mnt);
1657
Al Viro2d92ab32008-08-02 00:51:11 -04001658 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
1661/*
1662 * add a mount into a namespace's mount tree
1663 * - provide the option of adding the new mount to an expiration list
1664 */
Al Viro8d66bf52008-08-01 09:05:54 -04001665int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 int mnt_flags, struct list_head *fslist)
1667{
1668 int err;
1669
Ram Pai390c6842005-11-07 17:17:51 -05001670 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001672 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001673 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 ;
1675 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001676 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 goto unlock;
1678
1679 /* Refuse the same filesystem on the same mount point */
1680 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001681 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1682 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 goto unlock;
1684
1685 err = -EINVAL;
1686 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1687 goto unlock;
1688
1689 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001690 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c2005-11-07 17:16:29 -05001691 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Al Viro6758f952008-03-22 16:14:30 -04001693 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001694 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001695
Ram Pai390c6842005-11-07 17:17:51 -05001696 up_write(&namespace_sem);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001697 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
1699unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001700 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 mntput(newmnt);
1702 return err;
1703}
1704
1705EXPORT_SYMBOL_GPL(do_add_mount);
1706
Trond Myklebust5528f9112006-06-09 09:34:17 -04001707/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 * process a list of expirable mountpoints with the intent of discarding any
1709 * mountpoints that aren't in use and haven't been touched since last we came
1710 * here
1711 */
1712void mark_mounts_for_expiry(struct list_head *mounts)
1713{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 struct vfsmount *mnt, *next;
1715 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001716 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 if (list_empty(mounts))
1719 return;
1720
Al Virobcc5c7d2008-03-22 00:21:53 -04001721 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 spin_lock(&vfsmount_lock);
1723
1724 /* extract from the expiration list every vfsmount that matches the
1725 * following criteria:
1726 * - only referenced by its parent vfsmount
1727 * - still marked for expiry (marked on the last call here; marks are
1728 * cleared by mntput())
1729 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001730 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001732 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001734 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001736 while (!list_empty(&graveyard)) {
1737 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1738 touch_mnt_namespace(mnt->mnt_ns);
1739 umount_tree(mnt, 1, &umounts);
1740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001742 up_write(&namespace_sem);
1743
1744 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745}
1746
1747EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1748
1749/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001750 * Ripoff of 'select_parent()'
1751 *
1752 * search the list of submounts for a given mountpoint, and move any
1753 * shrinkable submounts to the 'graveyard' list.
1754 */
1755static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1756{
1757 struct vfsmount *this_parent = parent;
1758 struct list_head *next;
1759 int found = 0;
1760
1761repeat:
1762 next = this_parent->mnt_mounts.next;
1763resume:
1764 while (next != &this_parent->mnt_mounts) {
1765 struct list_head *tmp = next;
1766 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1767
1768 next = tmp->next;
1769 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1770 continue;
1771 /*
1772 * Descend a level if the d_mounts list is non-empty.
1773 */
1774 if (!list_empty(&mnt->mnt_mounts)) {
1775 this_parent = mnt;
1776 goto repeat;
1777 }
1778
1779 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001780 list_move_tail(&mnt->mnt_expire, graveyard);
1781 found++;
1782 }
1783 }
1784 /*
1785 * All done at this level ... ascend and resume the search
1786 */
1787 if (this_parent != parent) {
1788 next = this_parent->mnt_child.next;
1789 this_parent = this_parent->mnt_parent;
1790 goto resume;
1791 }
1792 return found;
1793}
1794
1795/*
1796 * process a list of expirable mountpoints with the intent of discarding any
1797 * submounts of a specific parent mountpoint
1798 */
Al Viroc35038b2008-03-22 00:46:23 -04001799static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001800{
1801 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001802 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001803
Trond Myklebust5528f9112006-06-09 09:34:17 -04001804 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001805 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001806 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001807 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001808 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001809 touch_mnt_namespace(m->mnt_ns);
1810 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001811 }
1812 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001813}
1814
Trond Myklebust5528f9112006-06-09 09:34:17 -04001815/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 * Some copy_from_user() implementations do not return the exact number of
1817 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1818 * Note that this function differs from copy_from_user() in that it will oops
1819 * on bad values of `to', rather than returning a short copy.
1820 */
Ram Paib58fed82005-11-07 17:16:09 -05001821static long exact_copy_from_user(void *to, const void __user * from,
1822 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823{
1824 char *t = to;
1825 const char __user *f = from;
1826 char c;
1827
1828 if (!access_ok(VERIFY_READ, from, n))
1829 return n;
1830
1831 while (n) {
1832 if (__get_user(c, f)) {
1833 memset(t, 0, n);
1834 break;
1835 }
1836 *t++ = c;
1837 f++;
1838 n--;
1839 }
1840 return n;
1841}
1842
Ram Paib58fed82005-11-07 17:16:09 -05001843int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844{
1845 int i;
1846 unsigned long page;
1847 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 *where = 0;
1850 if (!data)
1851 return 0;
1852
1853 if (!(page = __get_free_page(GFP_KERNEL)))
1854 return -ENOMEM;
1855
1856 /* We only care that *some* data at the address the user
1857 * gave us is valid. Just in case, we'll zero
1858 * the remainder of the page.
1859 */
1860 /* copy_from_user cannot cross TASK_SIZE ! */
1861 size = TASK_SIZE - (unsigned long)data;
1862 if (size > PAGE_SIZE)
1863 size = PAGE_SIZE;
1864
1865 i = size - exact_copy_from_user((void *)page, data, size);
1866 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001867 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 return -EFAULT;
1869 }
1870 if (i != PAGE_SIZE)
1871 memset((char *)page + i, 0, PAGE_SIZE - i);
1872 *where = page;
1873 return 0;
1874}
1875
Vegard Nossumeca6f532009-09-18 13:05:45 -07001876int copy_mount_string(const void __user *data, char **where)
1877{
1878 char *tmp;
1879
1880 if (!data) {
1881 *where = NULL;
1882 return 0;
1883 }
1884
1885 tmp = strndup_user(data, PAGE_SIZE);
1886 if (IS_ERR(tmp))
1887 return PTR_ERR(tmp);
1888
1889 *where = tmp;
1890 return 0;
1891}
1892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893/*
1894 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1895 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1896 *
1897 * data is a (void *) that can point to any structure up to
1898 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1899 * information (or be NULL).
1900 *
1901 * Pre-0.97 versions of mount() didn't have a flags word.
1902 * When the flags word was introduced its top half was required
1903 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1904 * Therefore, if this magic number is present, it carries no information
1905 * and must be discarded.
1906 */
Ram Paib58fed82005-11-07 17:16:09 -05001907long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 unsigned long flags, void *data_page)
1909{
Al Viro2d92ab32008-08-02 00:51:11 -04001910 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 int retval = 0;
1912 int mnt_flags = 0;
1913
1914 /* Discard magic */
1915 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1916 flags &= ~MS_MGC_MSK;
1917
1918 /* Basic sanity checks */
1919
1920 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1921 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
1923 if (data_page)
1924 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1925
Tetsuo Handaa27ab9f2009-10-04 21:49:49 +09001926 /* ... and get the mountpoint */
1927 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1928 if (retval)
1929 return retval;
1930
1931 retval = security_sb_mount(dev_name, &path,
1932 type_page, flags, data_page);
1933 if (retval)
1934 goto dput_out;
1935
Andi Kleen613cbe32009-04-19 18:40:43 +02001936 /* Default to relatime unless overriden */
1937 if (!(flags & MS_NOATIME))
1938 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 /* Separate the per-mountpoint flags */
1941 if (flags & MS_NOSUID)
1942 mnt_flags |= MNT_NOSUID;
1943 if (flags & MS_NODEV)
1944 mnt_flags |= MNT_NODEV;
1945 if (flags & MS_NOEXEC)
1946 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001947 if (flags & MS_NOATIME)
1948 mnt_flags |= MNT_NOATIME;
1949 if (flags & MS_NODIRATIME)
1950 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001951 if (flags & MS_STRICTATIME)
1952 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001953 if (flags & MS_RDONLY)
1954 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001955
1956 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Matthew Garrettd0adde52009-03-26 17:49:56 +00001957 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1958 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001961 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 data_page);
1963 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001964 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001965 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04001966 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04001968 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 else
Al Viro2d92ab32008-08-02 00:51:11 -04001970 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 dev_name, data_page);
1972dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04001973 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return retval;
1975}
1976
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04001977static struct mnt_namespace *alloc_mnt_ns(void)
1978{
1979 struct mnt_namespace *new_ns;
1980
1981 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1982 if (!new_ns)
1983 return ERR_PTR(-ENOMEM);
1984 atomic_set(&new_ns->count, 1);
1985 new_ns->root = NULL;
1986 INIT_LIST_HEAD(&new_ns->list);
1987 init_waitqueue_head(&new_ns->poll);
1988 new_ns->event = 0;
1989 return new_ns;
1990}
1991
JANAK DESAI741a2952006-02-07 12:59:00 -08001992/*
1993 * Allocate a new namespace structure and populate it with contents
1994 * copied from the namespace of the passed in task structure.
1995 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001996static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001997 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001999 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002000 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 struct vfsmount *p, *q;
2002
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002003 new_ns = alloc_mnt_ns();
2004 if (IS_ERR(new_ns))
2005 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Ram Pai390c6842005-11-07 17:17:51 -05002007 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002009 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002010 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002012 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002014 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 }
2016 spin_lock(&vfsmount_lock);
2017 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2018 spin_unlock(&vfsmount_lock);
2019
2020 /*
2021 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2022 * as belonging to new namespace. We have already acquired a private
2023 * fs_struct, so tsk->fs->lock is not needed.
2024 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002025 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 q = new_ns->root;
2027 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002028 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002030 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002032 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002034 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002036 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002039 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 q = next_mnt(q, new_ns->root);
2041 }
Ram Pai390c6842005-11-07 17:17:51 -05002042 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 if (rootmnt)
2045 mntput(rootmnt);
2046 if (pwdmnt)
2047 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
JANAK DESAI741a2952006-02-07 12:59:00 -08002049 return new_ns;
2050}
2051
Eric W. Biederman213dd262007-07-15 23:41:15 -07002052struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002053 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002054{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002055 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002056
Badari Pulavartye3222c42007-05-08 00:25:21 -07002057 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002058 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002059
2060 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002061 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002062
Badari Pulavartye3222c42007-05-08 00:25:21 -07002063 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002064
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002065 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002066 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067}
2068
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002069/**
2070 * create_mnt_ns - creates a private namespace and adds a root filesystem
2071 * @mnt: pointer to the new root filesystem mountpoint
2072 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002073struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002074{
2075 struct mnt_namespace *new_ns;
2076
2077 new_ns = alloc_mnt_ns();
2078 if (!IS_ERR(new_ns)) {
2079 mnt->mnt_ns = new_ns;
2080 new_ns->root = mnt;
2081 list_add(&new_ns->list, &new_ns->root->mnt_list);
2082 }
2083 return new_ns;
2084}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002085EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002086
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002087SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2088 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002090 int ret;
2091 char *kernel_type;
2092 char *kernel_dir;
2093 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095
Vegard Nossumeca6f532009-09-18 13:05:45 -07002096 ret = copy_mount_string(type, &kernel_type);
2097 if (ret < 0)
2098 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Vegard Nossumeca6f532009-09-18 13:05:45 -07002100 kernel_dir = getname(dir_name);
2101 if (IS_ERR(kernel_dir)) {
2102 ret = PTR_ERR(kernel_dir);
2103 goto out_dir;
2104 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Vegard Nossumeca6f532009-09-18 13:05:45 -07002106 ret = copy_mount_string(dev_name, &kernel_dev);
2107 if (ret < 0)
2108 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
Vegard Nossumeca6f532009-09-18 13:05:45 -07002110 ret = copy_mount_options(data, &data_page);
2111 if (ret < 0)
2112 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113
Vegard Nossumeca6f532009-09-18 13:05:45 -07002114 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2115 (void *) data_page);
2116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002118out_data:
2119 kfree(kernel_dev);
2120out_dev:
2121 putname(kernel_dir);
2122out_dir:
2123 kfree(kernel_type);
2124out_type:
2125 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
2127
2128/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 * pivot_root Semantics:
2130 * Moves the root file system of the current process to the directory put_old,
2131 * makes new_root as the new root file system of the current process, and sets
2132 * root/cwd of all processes which had them on the current root to new_root.
2133 *
2134 * Restrictions:
2135 * The new_root and put_old must be directories, and must not be on the
2136 * same file system as the current process root. The put_old must be
2137 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2138 * pointed to by put_old must yield the same directory as new_root. No other
2139 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2140 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002141 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2142 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2143 * in this situation.
2144 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * Notes:
2146 * - we don't move root/cwd if they are not at the root (reason: if something
2147 * cared enough to change them, it's probably wrong to force them elsewhere)
2148 * - it's okay to pick a root that isn't the root of a file system, e.g.
2149 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2150 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2151 * first.
2152 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002153SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2154 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002157 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 int error;
2159
2160 if (!capable(CAP_SYS_ADMIN))
2161 return -EPERM;
2162
Al Viro2d8f3032008-07-22 09:59:21 -04002163 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (error)
2165 goto out0;
2166 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002167 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 goto out1;
2169
Al Viro2d8f3032008-07-22 09:59:21 -04002170 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 if (error)
2172 goto out1;
2173
Al Viro2d8f3032008-07-22 09:59:21 -04002174 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002176 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 goto out1;
2178 }
2179
2180 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002181 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002182 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002184 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002185 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002187 if (IS_MNT_SHARED(old.mnt) ||
2188 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002189 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002190 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002191 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 goto out2;
2193 error = -ENOENT;
Al Viro2d8f3032008-07-22 09:59:21 -04002194 if (IS_DEADDIR(new.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002196 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002198 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 goto out2;
2200 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002201 if (new.mnt == root.mnt ||
2202 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 goto out2; /* loop, on the same file system */
2204 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002205 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002207 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002208 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002209 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002211 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002212 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002213 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002214 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002216 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 for (;;) {
2218 if (tmp->mnt_parent == tmp)
2219 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002220 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 break;
2222 tmp = tmp->mnt_parent;
2223 }
Al Viro2d8f3032008-07-22 09:59:21 -04002224 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002226 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002228 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002229 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002230 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002231 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002232 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002233 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002234 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002236 chroot_fs_refs(&root, &new);
2237 security_sb_post_pivotroot(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002239 path_put(&root_parent);
2240 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002242 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002243 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002244 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002245 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002247 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 return error;
2250out3:
2251 spin_unlock(&vfsmount_lock);
2252 goto out2;
2253}
2254
2255static void __init init_mount_tree(void)
2256{
2257 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002258 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002259 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
2261 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2262 if (IS_ERR(mnt))
2263 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002264 ns = create_mnt_ns(mnt);
2265 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002268 init_task.nsproxy->mnt_ns = ns;
2269 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Jan Blunckac748a02008-02-14 19:34:39 -08002271 root.mnt = ns->root;
2272 root.dentry = ns->root->mnt_root;
2273
2274 set_fs_pwd(current->fs, &root);
2275 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276}
2277
Denis Cheng74bf17c2007-10-16 23:26:30 -07002278void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002280 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002281 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
Ram Pai390c6842005-11-07 17:17:51 -05002283 init_rwsem(&namespace_sem);
2284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002286 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Ram Paib58fed82005-11-07 17:16:09 -05002288 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
2290 if (!mount_hashtable)
2291 panic("Failed to allocate mount hash table\n");
2292
Eric Dumazet13f14b42008-02-06 01:37:57 -08002293 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294
Eric Dumazet13f14b42008-02-06 01:37:57 -08002295 for (u = 0; u < HASH_SIZE; u++)
2296 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002298 err = sysfs_init();
2299 if (err)
2300 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002301 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002302 fs_kobj = kobject_create_and_add("fs", NULL);
2303 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002304 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 init_rootfs();
2306 init_mount_tree();
2307}
2308
Trond Myklebust616511d2009-06-22 15:09:13 -04002309void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Trond Myklebust616511d2009-06-22 15:09:13 -04002311 struct vfsmount *root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05002312 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002313
2314 if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock))
2315 return;
2316 root = ns->root;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002317 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07002318 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002319 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05002321 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002323 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002324 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002325 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002327EXPORT_SYMBOL(put_mnt_ns);