blob: 8174c8ab5c70e71316403e2f879a51c77566e422 [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)) {
Al Viro796a6b52010-01-16 13:28:47 -0500576 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500577 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
Al Viro9f5596a2010-02-05 00:40:25 -0500740int mnt_had_events(struct proc_mounts *p)
741{
742 struct mnt_namespace *ns = p->ns;
743 int res = 0;
744
745 spin_lock(&vfsmount_lock);
746 if (p->event != ns->event) {
747 p->event = ns->event;
748 res = 1;
749 }
750 spin_unlock(&vfsmount_lock);
751
752 return res;
753}
754
Ram Pai2d4d4862008-03-27 13:06:25 +0100755struct proc_fs_info {
756 int flag;
757 const char *str;
758};
759
Eric Paris2069f452008-07-04 09:47:13 +1000760static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
Ram Pai2d4d4862008-03-27 13:06:25 +0100762 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 { MS_SYNCHRONOUS, ",sync" },
764 { MS_DIRSYNC, ",dirsync" },
765 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 { 0, NULL }
767 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100768 const struct proc_fs_info *fs_infop;
769
770 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
771 if (sb->s_flags & fs_infop->flag)
772 seq_puts(m, fs_infop->str);
773 }
Eric Paris2069f452008-07-04 09:47:13 +1000774
775 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100776}
777
778static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
779{
780 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 { MNT_NOSUID, ",nosuid" },
782 { MNT_NODEV, ",nodev" },
783 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800784 { MNT_NOATIME, ",noatime" },
785 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800786 { MNT_RELATIME, ",relatime" },
Matthew Garrettd0adde52009-03-26 17:49:56 +0000787 { MNT_STRICTATIME, ",strictatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 { 0, NULL }
789 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100790 const struct proc_fs_info *fs_infop;
791
792 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
793 if (mnt->mnt_flags & fs_infop->flag)
794 seq_puts(m, fs_infop->str);
795 }
796}
797
798static void show_type(struct seq_file *m, struct super_block *sb)
799{
800 mangle(m, sb->s_type->name);
801 if (sb->s_subtype && sb->s_subtype[0]) {
802 seq_putc(m, '.');
803 mangle(m, sb->s_subtype);
804 }
805}
806
807static int show_vfsmnt(struct seq_file *m, void *v)
808{
809 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
810 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800811 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
814 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800815 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100817 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800818 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000819 err = show_sb_opts(m, mnt->mnt_sb);
820 if (err)
821 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100822 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 if (mnt->mnt_sb->s_op->show_options)
824 err = mnt->mnt_sb->s_op->show_options(m, mnt);
825 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000826out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return err;
828}
829
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100830const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 .start = m_start,
832 .next = m_next,
833 .stop = m_stop,
834 .show = show_vfsmnt
835};
836
Ram Pai2d4d4862008-03-27 13:06:25 +0100837static int show_mountinfo(struct seq_file *m, void *v)
838{
839 struct proc_mounts *p = m->private;
840 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
841 struct super_block *sb = mnt->mnt_sb;
842 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
843 struct path root = p->root;
844 int err = 0;
845
846 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
847 MAJOR(sb->s_dev), MINOR(sb->s_dev));
848 seq_dentry(m, mnt->mnt_root, " \t\n\\");
849 seq_putc(m, ' ');
850 seq_path_root(m, &mnt_path, &root, " \t\n\\");
851 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
852 /*
853 * Mountpoint is outside root, discard that one. Ugly,
854 * but less so than trying to do that in iterator in a
855 * race-free way (due to renames).
856 */
857 return SEQ_SKIP;
858 }
859 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
860 show_mnt_opts(m, mnt);
861
862 /* Tagged fields ("foo:X" or "bar") */
863 if (IS_MNT_SHARED(mnt))
864 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100865 if (IS_MNT_SLAVE(mnt)) {
866 int master = mnt->mnt_master->mnt_group_id;
867 int dom = get_dominating_id(mnt, &p->root);
868 seq_printf(m, " master:%i", master);
869 if (dom && dom != master)
870 seq_printf(m, " propagate_from:%i", dom);
871 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100872 if (IS_MNT_UNBINDABLE(mnt))
873 seq_puts(m, " unbindable");
874
875 /* Filesystem specific data */
876 seq_puts(m, " - ");
877 show_type(m, sb);
878 seq_putc(m, ' ');
879 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
880 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000881 err = show_sb_opts(m, sb);
882 if (err)
883 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100884 if (sb->s_op->show_options)
885 err = sb->s_op->show_options(m, mnt);
886 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000887out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100888 return err;
889}
890
891const struct seq_operations mountinfo_op = {
892 .start = m_start,
893 .next = m_next,
894 .stop = m_stop,
895 .show = show_mountinfo,
896};
897
Chuck Leverb4629fe2006-03-20 13:44:12 -0500898static int show_vfsstat(struct seq_file *m, void *v)
899{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700900 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800901 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500902 int err = 0;
903
904 /* device */
905 if (mnt->mnt_devname) {
906 seq_puts(m, "device ");
907 mangle(m, mnt->mnt_devname);
908 } else
909 seq_puts(m, "no device");
910
911 /* mount point */
912 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800913 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500914 seq_putc(m, ' ');
915
916 /* file system type */
917 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100918 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500919
920 /* optional statistics */
921 if (mnt->mnt_sb->s_op->show_stats) {
922 seq_putc(m, ' ');
923 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
924 }
925
926 seq_putc(m, '\n');
927 return err;
928}
929
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100930const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500931 .start = m_start,
932 .next = m_next,
933 .stop = m_stop,
934 .show = show_vfsstat,
935};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100936#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500937
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938/**
939 * may_umount_tree - check if a mount tree is busy
940 * @mnt: root of mount tree
941 *
942 * This is called to check if a tree of mounts has any
943 * open files, pwds, chroots or sub mounts that are
944 * busy.
945 */
946int may_umount_tree(struct vfsmount *mnt)
947{
Ram Pai36341f62005-11-07 17:17:22 -0500948 int actual_refs = 0;
949 int minimum_refs = 0;
950 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951
952 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500953 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 actual_refs += atomic_read(&p->mnt_count);
955 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 }
957 spin_unlock(&vfsmount_lock);
958
959 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800960 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Ian Kente3474a82006-03-27 01:14:51 -0800962 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
965EXPORT_SYMBOL(may_umount_tree);
966
967/**
968 * may_umount - check if a mount point is busy
969 * @mnt: root of mount
970 *
971 * This is called to check if a mount point has any
972 * open files, pwds, chroots or sub mounts. If the
973 * mount has sub mounts this will return busy
974 * regardless of whether the sub mounts are busy.
975 *
976 * Doesn't take quota and stuff into account. IOW, in some cases it will
977 * give false negatives. The main reason why it's here is that we need
978 * a non-destructive way to look for easily umountable filesystems.
979 */
980int may_umount(struct vfsmount *mnt)
981{
Ian Kente3474a82006-03-27 01:14:51 -0800982 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -0500983 down_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500984 spin_lock(&vfsmount_lock);
985 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800986 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500987 spin_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -0500988 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500989 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990}
991
992EXPORT_SYMBOL(may_umount);
993
Ram Paib90fa9a2005-11-07 17:19:50 -0500994void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500996 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800997 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700998 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500999 list_del_init(&mnt->mnt_hash);
1000 if (mnt->mnt_parent != mnt) {
1001 struct dentry *dentry;
1002 struct vfsmount *m;
1003 spin_lock(&vfsmount_lock);
1004 dentry = mnt->mnt_mountpoint;
1005 m = mnt->mnt_parent;
1006 mnt->mnt_mountpoint = mnt->mnt_root;
1007 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001008 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001010 dput(dentry);
1011 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 }
1013 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001014 }
1015}
1016
Ram Paia05964f2005-11-07 17:20:17 -05001017void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001018{
1019 struct vfsmount *p;
1020
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001021 for (p = mnt; p; p = next_mnt(p, mnt))
1022 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001023
Ram Paia05964f2005-11-07 17:20:17 -05001024 if (propagate)
1025 propagate_umount(kill);
1026
Ram Pai70fbcdf2005-11-07 17:17:04 -05001027 list_for_each_entry(p, kill, mnt_hash) {
1028 list_del_init(&p->mnt_expire);
1029 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001030 __touch_mnt_namespace(p->mnt_ns);
1031 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001032 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001033 if (p->mnt_parent != p) {
1034 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001035 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001036 }
Ram Paia05964f2005-11-07 17:20:17 -05001037 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
1039}
1040
Al Viroc35038b2008-03-22 00:46:23 -04001041static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043static int do_umount(struct vfsmount *mnt, int flags)
1044{
Ram Paib58fed82005-11-07 17:16:09 -05001045 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001047 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
1049 retval = security_sb_umount(mnt, flags);
1050 if (retval)
1051 return retval;
1052
1053 /*
1054 * Allow userspace to request a mountpoint be expired rather than
1055 * unmounting unconditionally. Unmount only happens if:
1056 * (1) the mark is already set (the mark is cleared by mntput())
1057 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1058 */
1059 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001060 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 flags & (MNT_FORCE | MNT_DETACH))
1062 return -EINVAL;
1063
1064 if (atomic_read(&mnt->mnt_count) != 2)
1065 return -EBUSY;
1066
1067 if (!xchg(&mnt->mnt_expiry_mark, 1))
1068 return -EAGAIN;
1069 }
1070
1071 /*
1072 * If we may have to abort operations to get out of this
1073 * mount, and they will themselves hold resources we must
1074 * allow the fs to do things. In the Unix tradition of
1075 * 'Gee thats tricky lets do it in userspace' the umount_begin
1076 * might fail to complete on the first run through as other tasks
1077 * must return, and the like. Thats for the mount program to worry
1078 * about for the moment.
1079 */
1080
Al Viro42faad92008-04-24 07:21:56 -04001081 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001082 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001083 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085 /*
1086 * No sense to grab the lock for this test, but test itself looks
1087 * somewhat bogus. Suggestions for better replacement?
1088 * Ho-hum... In principle, we might treat that as umount + switch
1089 * to rootfs. GC would eventually take care of the old vfsmount.
1090 * Actually it makes sense, especially if rootfs would contain a
1091 * /reboot - static binary that would close all descriptors and
1092 * call reboot(9). Then init(8) could umount root and exec /reboot.
1093 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001094 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 /*
1096 * Special case for "unmounting" root ...
1097 * we just try to remount it readonly.
1098 */
1099 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001100 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 up_write(&sb->s_umount);
1103 return retval;
1104 }
1105
Ram Pai390c6842005-11-07 17:17:51 -05001106 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001108 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Al Viroc35038b2008-03-22 00:46:23 -04001110 if (!(flags & MNT_DETACH))
1111 shrink_submounts(mnt, &umount_list);
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001114 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001116 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 retval = 0;
1118 }
1119 spin_unlock(&vfsmount_lock);
1120 if (retval)
1121 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001122 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001123 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return retval;
1125}
1126
1127/*
1128 * Now umount can handle mount points as well as block devices.
1129 * This is important for filesystems which use unnamed block devices.
1130 *
1131 * We now support a flag for forced unmount like the other 'big iron'
1132 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1133 */
1134
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001135SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Al Viro2d8f3032008-07-22 09:59:21 -04001137 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001139 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001141 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1142 return -EINVAL;
1143
1144 if (!(flags & UMOUNT_NOFOLLOW))
1145 lookup_flags |= LOOKUP_FOLLOW;
1146
1147 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 if (retval)
1149 goto out;
1150 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001151 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001153 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 goto dput_and_out;
1155
1156 retval = -EPERM;
1157 if (!capable(CAP_SYS_ADMIN))
1158 goto dput_and_out;
1159
Al Viro2d8f3032008-07-22 09:59:21 -04001160 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001162 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001163 dput(path.dentry);
1164 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165out:
1166 return retval;
1167}
1168
1169#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1170
1171/*
Ram Paib58fed82005-11-07 17:16:09 -05001172 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001174SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
Ram Paib58fed82005-11-07 17:16:09 -05001176 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
1178
1179#endif
1180
Al Viro2d92ab32008-08-02 00:51:11 -04001181static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
1183 if (capable(CAP_SYS_ADMIN))
1184 return 0;
1185 return -EPERM;
1186#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001187 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001189 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001190 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 return -EPERM;
1192 }
Al Viro2d92ab32008-08-02 00:51:11 -04001193 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 return -EPERM;
1195 return 0;
1196#endif
1197}
1198
Ram Paib90fa9a2005-11-07 17:19:50 -05001199struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001200 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
1202 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001203 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Ram Pai9676f0c2005-11-07 17:21:20 -05001205 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1206 return NULL;
1207
Ram Pai36341f62005-11-07 17:17:22 -05001208 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 if (!q)
1210 goto Enomem;
1211 q->mnt_mountpoint = mnt->mnt_mountpoint;
1212
1213 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001214 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001215 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 continue;
1217
1218 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001219 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1220 s = skip_mnt_tree(s);
1221 continue;
1222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 while (p != s->mnt_parent) {
1224 p = p->mnt_parent;
1225 q = q->mnt_parent;
1226 }
1227 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001228 path.mnt = q;
1229 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001230 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 if (!q)
1232 goto Enomem;
1233 spin_lock(&vfsmount_lock);
1234 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001235 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 spin_unlock(&vfsmount_lock);
1237 }
1238 }
1239 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001240Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001242 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001244 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001246 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248 return NULL;
1249}
1250
Al Viro589ff872009-04-18 03:28:19 -04001251struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001252{
1253 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001254 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001255 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001256 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001257 return tree;
1258}
1259
1260void drop_collected_mounts(struct vfsmount *mnt)
1261{
1262 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001263 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001264 spin_lock(&vfsmount_lock);
1265 umount_tree(mnt, 0, &umount_list);
1266 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001267 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001268 release_mounts(&umount_list);
1269}
1270
Al Viro1f707132010-01-30 22:51:25 -05001271int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1272 struct vfsmount *root)
1273{
1274 struct vfsmount *mnt;
1275 int res = f(root, arg);
1276 if (res)
1277 return res;
1278 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1279 res = f(mnt, arg);
1280 if (res)
1281 return res;
1282 }
1283 return 0;
1284}
1285
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001286static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1287{
1288 struct vfsmount *p;
1289
1290 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1291 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1292 mnt_release_group_id(p);
1293 }
1294}
1295
1296static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1297{
1298 struct vfsmount *p;
1299
1300 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1301 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1302 int err = mnt_alloc_group_id(p);
1303 if (err) {
1304 cleanup_group_ids(mnt, p);
1305 return err;
1306 }
1307 }
1308 }
1309
1310 return 0;
1311}
1312
Ram Paib90fa9a2005-11-07 17:19:50 -05001313/*
1314 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001315 * @nd : place the mount tree @source_mnt is attached
1316 * @parent_nd : if non-null, detach the source_mnt from its parent and
1317 * store the parent mount and mountpoint dentry.
1318 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001319 *
1320 * NOTE: in the table below explains the semantics when a source mount
1321 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001322 * ---------------------------------------------------------------------------
1323 * | BIND MOUNT OPERATION |
1324 * |**************************************************************************
1325 * | source-->| shared | private | slave | unbindable |
1326 * | dest | | | | |
1327 * | | | | | | |
1328 * | v | | | | |
1329 * |**************************************************************************
1330 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1331 * | | | | | |
1332 * |non-shared| shared (+) | private | slave (*) | invalid |
1333 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001334 * A bind operation clones the source mount and mounts the clone on the
1335 * destination mount.
1336 *
1337 * (++) the cloned mount is propagated to all the mounts in the propagation
1338 * tree of the destination mount and the cloned mount is added to
1339 * the peer group of the source mount.
1340 * (+) the cloned mount is created under the destination mount and is marked
1341 * as shared. The cloned mount is added to the peer group of the source
1342 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001343 * (+++) the mount is propagated to all the mounts in the propagation tree
1344 * of the destination mount and the cloned mount is made slave
1345 * of the same master as that of the source mount. The cloned mount
1346 * is marked as 'shared and slave'.
1347 * (*) the cloned mount is made a slave of the same master as that of the
1348 * source mount.
1349 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001350 * ---------------------------------------------------------------------------
1351 * | MOVE MOUNT OPERATION |
1352 * |**************************************************************************
1353 * | source-->| shared | private | slave | unbindable |
1354 * | dest | | | | |
1355 * | | | | | | |
1356 * | v | | | | |
1357 * |**************************************************************************
1358 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1359 * | | | | | |
1360 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1361 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001362 *
1363 * (+) the mount is moved to the destination. And is then propagated to
1364 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001365 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001366 * (+++) the mount is moved to the destination and is then propagated to
1367 * all the mounts belonging to the destination mount's propagation tree.
1368 * the mount is marked as 'shared and slave'.
1369 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001370 *
1371 * if the source mount is a tree, the operations explained above is
1372 * applied to each mount in the tree.
1373 * Must be called without spinlocks held, since this function can sleep
1374 * in allocations.
1375 */
1376static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001377 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001378{
1379 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001380 struct vfsmount *dest_mnt = path->mnt;
1381 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001382 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001383 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001384
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001385 if (IS_MNT_SHARED(dest_mnt)) {
1386 err = invent_group_ids(source_mnt, true);
1387 if (err)
1388 goto out;
1389 }
1390 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1391 if (err)
1392 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001393
Al Virodf1a1ad2010-01-16 12:57:40 -05001394 spin_lock(&vfsmount_lock);
1395
Ram Paib90fa9a2005-11-07 17:19:50 -05001396 if (IS_MNT_SHARED(dest_mnt)) {
1397 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1398 set_mnt_shared(p);
1399 }
Al Viro1a390682008-03-21 20:48:19 -04001400 if (parent_path) {
1401 detach_mnt(source_mnt, parent_path);
1402 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001403 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001404 } else {
1405 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1406 commit_tree(source_mnt);
1407 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001408
1409 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1410 list_del_init(&child->mnt_hash);
1411 commit_tree(child);
1412 }
1413 spin_unlock(&vfsmount_lock);
1414 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001415
1416 out_cleanup_ids:
1417 if (IS_MNT_SHARED(dest_mnt))
1418 cleanup_group_ids(source_mnt, NULL);
1419 out:
1420 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001421}
1422
Al Viro8c3ee422008-03-22 18:00:39 -04001423static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 int err;
1426 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1427 return -EINVAL;
1428
Al Viro8c3ee422008-03-22 18:00:39 -04001429 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1431 return -ENOTDIR;
1432
1433 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001434 mutex_lock(&path->dentry->d_inode->i_mutex);
1435 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 goto out_unlock;
1437
Al Viro8c3ee422008-03-22 18:00:39 -04001438 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (err)
1440 goto out_unlock;
1441
1442 err = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001443 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001444 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001446 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001448 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return err;
1450}
1451
1452/*
Ram Pai07b20882005-11-07 17:19:07 -05001453 * recursively change the type of the mountpoint.
1454 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001455static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001456{
Al Viro2d92ab32008-08-02 00:51:11 -04001457 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001458 int recurse = flag & MS_REC;
1459 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001460 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001461
Miklos Szerediee6f9582007-05-08 00:30:40 -07001462 if (!capable(CAP_SYS_ADMIN))
1463 return -EPERM;
1464
Al Viro2d92ab32008-08-02 00:51:11 -04001465 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001466 return -EINVAL;
1467
1468 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001469 if (type == MS_SHARED) {
1470 err = invent_group_ids(mnt, recurse);
1471 if (err)
1472 goto out_unlock;
1473 }
1474
Ram Pai07b20882005-11-07 17:19:07 -05001475 spin_lock(&vfsmount_lock);
1476 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1477 change_mnt_propagation(m, type);
1478 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001479
1480 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001481 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001482 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001483}
1484
1485/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 * do loopback mount.
1487 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001488static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001489 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
Al Viro2d92ab32008-08-02 00:51:11 -04001491 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001493 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (err)
1495 return err;
1496 if (!old_name || !*old_name)
1497 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001498 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 if (err)
1500 return err;
1501
Ram Pai390c6842005-11-07 17:17:51 -05001502 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001504 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001505 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001506
Al Viro2d92ab32008-08-02 00:51:11 -04001507 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001508 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Al Viroccd48bc2005-11-07 17:15:04 -05001510 err = -ENOMEM;
1511 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001512 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001513 else
Al Viro2d92ab32008-08-02 00:51:11 -04001514 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001515
1516 if (!mnt)
1517 goto out;
1518
Al Viro2d92ab32008-08-02 00:51:11 -04001519 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001520 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001521 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001523 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001525 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001526 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Al Viroccd48bc2005-11-07 17:15:04 -05001528out:
Ram Pai390c6842005-11-07 17:17:51 -05001529 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001530 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 return err;
1532}
1533
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001534static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1535{
1536 int error = 0;
1537 int readonly_request = 0;
1538
1539 if (ms_flags & MS_RDONLY)
1540 readonly_request = 1;
1541 if (readonly_request == __mnt_is_readonly(mnt))
1542 return 0;
1543
1544 if (readonly_request)
1545 error = mnt_make_readonly(mnt);
1546 else
1547 __mnt_unmake_readonly(mnt);
1548 return error;
1549}
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551/*
1552 * change filesystem flags. dir should be a physical root of filesystem.
1553 * If you've mounted a non-root directory somewhere and want to do remount
1554 * on it - tough luck.
1555 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001556static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 void *data)
1558{
1559 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001560 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562 if (!capable(CAP_SYS_ADMIN))
1563 return -EPERM;
1564
Al Viro2d92ab32008-08-02 00:51:11 -04001565 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return -EINVAL;
1567
Al Viro2d92ab32008-08-02 00:51:11 -04001568 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 return -EINVAL;
1570
1571 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001572 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001573 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001574 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001575 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001576 if (!err) {
1577 spin_lock(&vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001578 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001579 path->mnt->mnt_flags = mnt_flags;
Al Viro7b43a792010-01-16 13:01:26 -05001580 spin_unlock(&vfsmount_lock);
1581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001583 if (!err) {
Al Viro2d92ab32008-08-02 00:51:11 -04001584 security_sb_post_remount(path->mnt, flags, data);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001585
1586 spin_lock(&vfsmount_lock);
1587 touch_mnt_namespace(path->mnt->mnt_ns);
1588 spin_unlock(&vfsmount_lock);
1589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 return err;
1591}
1592
Ram Pai9676f0c2005-11-07 17:21:20 -05001593static inline int tree_contains_unbindable(struct vfsmount *mnt)
1594{
1595 struct vfsmount *p;
1596 for (p = mnt; p; p = next_mnt(p, mnt)) {
1597 if (IS_MNT_UNBINDABLE(p))
1598 return 1;
1599 }
1600 return 0;
1601}
1602
Al Viro0a0d8a42008-08-02 00:55:27 -04001603static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Al Viro2d92ab32008-08-02 00:51:11 -04001605 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 struct vfsmount *p;
1607 int err = 0;
1608 if (!capable(CAP_SYS_ADMIN))
1609 return -EPERM;
1610 if (!old_name || !*old_name)
1611 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001612 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (err)
1614 return err;
1615
Ram Pai390c6842005-11-07 17:17:51 -05001616 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001617 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001618 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 ;
1620 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001621 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 goto out;
1623
1624 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001625 mutex_lock(&path->dentry->d_inode->i_mutex);
1626 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 goto out1;
1628
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001629 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001630 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001633 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001634 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Al Viro2d92ab32008-08-02 00:51:11 -04001636 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001637 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
Al Viro2d92ab32008-08-02 00:51:11 -04001639 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1640 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001641 goto out1;
1642 /*
1643 * Don't move a mount residing in a shared parent.
1644 */
Al Viro2d92ab32008-08-02 00:51:11 -04001645 if (old_path.mnt->mnt_parent &&
1646 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001647 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001648 /*
1649 * Don't move a mount tree containing unbindable mounts to a destination
1650 * mount which is shared.
1651 */
Al Viro2d92ab32008-08-02 00:51:11 -04001652 if (IS_MNT_SHARED(path->mnt) &&
1653 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001654 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001656 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1657 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001658 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Al Viro2d92ab32008-08-02 00:51:11 -04001660 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001661 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001662 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
1664 /* if the mount is moved, it should no longer be expire
1665 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001666 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001668 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669out:
Ram Pai390c6842005-11-07 17:17:51 -05001670 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001672 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001673 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return err;
1675}
1676
1677/*
1678 * create a new mount for userspace and request it to be added into the
1679 * namespace's tree
1680 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001681static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 int mnt_flags, char *name, void *data)
1683{
1684 struct vfsmount *mnt;
1685
Vegard Nossumeca6f532009-09-18 13:05:45 -07001686 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 return -EINVAL;
1688
1689 /* we need capabilities... */
1690 if (!capable(CAP_SYS_ADMIN))
1691 return -EPERM;
1692
Al Viro7f78d4c2009-05-08 13:34:06 -04001693 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001695 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (IS_ERR(mnt))
1697 return PTR_ERR(mnt);
1698
Al Viro2d92ab32008-08-02 00:51:11 -04001699 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
1702/*
1703 * add a mount into a namespace's mount tree
1704 * - provide the option of adding the new mount to an expiration list
1705 */
Al Viro8d66bf52008-08-01 09:05:54 -04001706int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 int mnt_flags, struct list_head *fslist)
1708{
1709 int err;
1710
Al Viro80893522010-02-05 09:30:46 -05001711 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
Al Viro27d55f12010-01-16 13:07:36 -05001712
Ram Pai390c6842005-11-07 17:17:51 -05001713 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001715 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001716 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 ;
1718 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001719 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 goto unlock;
1721
1722 /* Refuse the same filesystem on the same mount point */
1723 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001724 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1725 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 goto unlock;
1727
1728 err = -EINVAL;
1729 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1730 goto unlock;
1731
1732 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001733 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001734 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Al Viro6758f952008-03-22 16:14:30 -04001736 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001737 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001738
Ram Pai390c6842005-11-07 17:17:51 -05001739 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001740 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001743 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 mntput(newmnt);
1745 return err;
1746}
1747
1748EXPORT_SYMBOL_GPL(do_add_mount);
1749
Trond Myklebust5528f9112006-06-09 09:34:17 -04001750/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 * process a list of expirable mountpoints with the intent of discarding any
1752 * mountpoints that aren't in use and haven't been touched since last we came
1753 * here
1754 */
1755void mark_mounts_for_expiry(struct list_head *mounts)
1756{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 struct vfsmount *mnt, *next;
1758 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001759 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
1761 if (list_empty(mounts))
1762 return;
1763
Al Virobcc5c7d2008-03-22 00:21:53 -04001764 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 spin_lock(&vfsmount_lock);
1766
1767 /* extract from the expiration list every vfsmount that matches the
1768 * following criteria:
1769 * - only referenced by its parent vfsmount
1770 * - still marked for expiry (marked on the last call here; marks are
1771 * cleared by mntput())
1772 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001773 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001775 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001777 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001779 while (!list_empty(&graveyard)) {
1780 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1781 touch_mnt_namespace(mnt->mnt_ns);
1782 umount_tree(mnt, 1, &umounts);
1783 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001785 up_write(&namespace_sem);
1786
1787 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788}
1789
1790EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1791
1792/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001793 * Ripoff of 'select_parent()'
1794 *
1795 * search the list of submounts for a given mountpoint, and move any
1796 * shrinkable submounts to the 'graveyard' list.
1797 */
1798static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1799{
1800 struct vfsmount *this_parent = parent;
1801 struct list_head *next;
1802 int found = 0;
1803
1804repeat:
1805 next = this_parent->mnt_mounts.next;
1806resume:
1807 while (next != &this_parent->mnt_mounts) {
1808 struct list_head *tmp = next;
1809 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1810
1811 next = tmp->next;
1812 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1813 continue;
1814 /*
1815 * Descend a level if the d_mounts list is non-empty.
1816 */
1817 if (!list_empty(&mnt->mnt_mounts)) {
1818 this_parent = mnt;
1819 goto repeat;
1820 }
1821
1822 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001823 list_move_tail(&mnt->mnt_expire, graveyard);
1824 found++;
1825 }
1826 }
1827 /*
1828 * All done at this level ... ascend and resume the search
1829 */
1830 if (this_parent != parent) {
1831 next = this_parent->mnt_child.next;
1832 this_parent = this_parent->mnt_parent;
1833 goto resume;
1834 }
1835 return found;
1836}
1837
1838/*
1839 * process a list of expirable mountpoints with the intent of discarding any
1840 * submounts of a specific parent mountpoint
1841 */
Al Viroc35038b2008-03-22 00:46:23 -04001842static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001843{
1844 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001845 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001846
Trond Myklebust5528f9112006-06-09 09:34:17 -04001847 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001848 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001849 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001850 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001851 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001852 touch_mnt_namespace(m->mnt_ns);
1853 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001854 }
1855 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001856}
1857
Trond Myklebust5528f9112006-06-09 09:34:17 -04001858/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 * Some copy_from_user() implementations do not return the exact number of
1860 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1861 * Note that this function differs from copy_from_user() in that it will oops
1862 * on bad values of `to', rather than returning a short copy.
1863 */
Ram Paib58fed82005-11-07 17:16:09 -05001864static long exact_copy_from_user(void *to, const void __user * from,
1865 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
1867 char *t = to;
1868 const char __user *f = from;
1869 char c;
1870
1871 if (!access_ok(VERIFY_READ, from, n))
1872 return n;
1873
1874 while (n) {
1875 if (__get_user(c, f)) {
1876 memset(t, 0, n);
1877 break;
1878 }
1879 *t++ = c;
1880 f++;
1881 n--;
1882 }
1883 return n;
1884}
1885
Ram Paib58fed82005-11-07 17:16:09 -05001886int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
1888 int i;
1889 unsigned long page;
1890 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 *where = 0;
1893 if (!data)
1894 return 0;
1895
1896 if (!(page = __get_free_page(GFP_KERNEL)))
1897 return -ENOMEM;
1898
1899 /* We only care that *some* data at the address the user
1900 * gave us is valid. Just in case, we'll zero
1901 * the remainder of the page.
1902 */
1903 /* copy_from_user cannot cross TASK_SIZE ! */
1904 size = TASK_SIZE - (unsigned long)data;
1905 if (size > PAGE_SIZE)
1906 size = PAGE_SIZE;
1907
1908 i = size - exact_copy_from_user((void *)page, data, size);
1909 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001910 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 return -EFAULT;
1912 }
1913 if (i != PAGE_SIZE)
1914 memset((char *)page + i, 0, PAGE_SIZE - i);
1915 *where = page;
1916 return 0;
1917}
1918
Vegard Nossumeca6f532009-09-18 13:05:45 -07001919int copy_mount_string(const void __user *data, char **where)
1920{
1921 char *tmp;
1922
1923 if (!data) {
1924 *where = NULL;
1925 return 0;
1926 }
1927
1928 tmp = strndup_user(data, PAGE_SIZE);
1929 if (IS_ERR(tmp))
1930 return PTR_ERR(tmp);
1931
1932 *where = tmp;
1933 return 0;
1934}
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936/*
1937 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1938 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1939 *
1940 * data is a (void *) that can point to any structure up to
1941 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1942 * information (or be NULL).
1943 *
1944 * Pre-0.97 versions of mount() didn't have a flags word.
1945 * When the flags word was introduced its top half was required
1946 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1947 * Therefore, if this magic number is present, it carries no information
1948 * and must be discarded.
1949 */
Ram Paib58fed82005-11-07 17:16:09 -05001950long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 unsigned long flags, void *data_page)
1952{
Al Viro2d92ab32008-08-02 00:51:11 -04001953 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 int retval = 0;
1955 int mnt_flags = 0;
1956
1957 /* Discard magic */
1958 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1959 flags &= ~MS_MGC_MSK;
1960
1961 /* Basic sanity checks */
1962
1963 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1964 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966 if (data_page)
1967 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1968
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09001969 /* ... and get the mountpoint */
1970 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1971 if (retval)
1972 return retval;
1973
1974 retval = security_sb_mount(dev_name, &path,
1975 type_page, flags, data_page);
1976 if (retval)
1977 goto dput_out;
1978
Andi Kleen613cbe32009-04-19 18:40:43 +02001979 /* Default to relatime unless overriden */
1980 if (!(flags & MS_NOATIME))
1981 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 /* Separate the per-mountpoint flags */
1984 if (flags & MS_NOSUID)
1985 mnt_flags |= MNT_NOSUID;
1986 if (flags & MS_NODEV)
1987 mnt_flags |= MNT_NODEV;
1988 if (flags & MS_NOEXEC)
1989 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001990 if (flags & MS_NOATIME)
1991 mnt_flags |= MNT_NOATIME;
1992 if (flags & MS_NODIRATIME)
1993 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001994 if (flags & MS_STRICTATIME)
1995 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001996 if (flags & MS_RDONLY)
1997 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001998
1999 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002000 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2001 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002004 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 data_page);
2006 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002007 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002008 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002009 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002011 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 else
Al Viro2d92ab32008-08-02 00:51:11 -04002013 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 dev_name, data_page);
2015dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002016 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 return retval;
2018}
2019
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002020static struct mnt_namespace *alloc_mnt_ns(void)
2021{
2022 struct mnt_namespace *new_ns;
2023
2024 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2025 if (!new_ns)
2026 return ERR_PTR(-ENOMEM);
2027 atomic_set(&new_ns->count, 1);
2028 new_ns->root = NULL;
2029 INIT_LIST_HEAD(&new_ns->list);
2030 init_waitqueue_head(&new_ns->poll);
2031 new_ns->event = 0;
2032 return new_ns;
2033}
2034
JANAK DESAI741a2952006-02-07 12:59:00 -08002035/*
2036 * Allocate a new namespace structure and populate it with contents
2037 * copied from the namespace of the passed in task structure.
2038 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002039static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002040 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002042 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002043 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 struct vfsmount *p, *q;
2045
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002046 new_ns = alloc_mnt_ns();
2047 if (IS_ERR(new_ns))
2048 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Ram Pai390c6842005-11-07 17:17:51 -05002050 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002052 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002053 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002055 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002057 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
2059 spin_lock(&vfsmount_lock);
2060 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2061 spin_unlock(&vfsmount_lock);
2062
2063 /*
2064 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2065 * as belonging to new namespace. We have already acquired a private
2066 * fs_struct, so tsk->fs->lock is not needed.
2067 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002068 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 q = new_ns->root;
2070 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002071 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002073 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002075 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002077 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002079 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002082 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 q = next_mnt(q, new_ns->root);
2084 }
Ram Pai390c6842005-11-07 17:17:51 -05002085 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 if (rootmnt)
2088 mntput(rootmnt);
2089 if (pwdmnt)
2090 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
JANAK DESAI741a2952006-02-07 12:59:00 -08002092 return new_ns;
2093}
2094
Eric W. Biederman213dd262007-07-15 23:41:15 -07002095struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002096 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002097{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002098 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002099
Badari Pulavartye3222c42007-05-08 00:25:21 -07002100 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002101 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002102
2103 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002104 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002105
Badari Pulavartye3222c42007-05-08 00:25:21 -07002106 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002107
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002108 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002109 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110}
2111
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002112/**
2113 * create_mnt_ns - creates a private namespace and adds a root filesystem
2114 * @mnt: pointer to the new root filesystem mountpoint
2115 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002116struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002117{
2118 struct mnt_namespace *new_ns;
2119
2120 new_ns = alloc_mnt_ns();
2121 if (!IS_ERR(new_ns)) {
2122 mnt->mnt_ns = new_ns;
2123 new_ns->root = mnt;
2124 list_add(&new_ns->list, &new_ns->root->mnt_list);
2125 }
2126 return new_ns;
2127}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002128EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002129
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002130SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2131 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002133 int ret;
2134 char *kernel_type;
2135 char *kernel_dir;
2136 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
Vegard Nossumeca6f532009-09-18 13:05:45 -07002139 ret = copy_mount_string(type, &kernel_type);
2140 if (ret < 0)
2141 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Vegard Nossumeca6f532009-09-18 13:05:45 -07002143 kernel_dir = getname(dir_name);
2144 if (IS_ERR(kernel_dir)) {
2145 ret = PTR_ERR(kernel_dir);
2146 goto out_dir;
2147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
Vegard Nossumeca6f532009-09-18 13:05:45 -07002149 ret = copy_mount_string(dev_name, &kernel_dev);
2150 if (ret < 0)
2151 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
Vegard Nossumeca6f532009-09-18 13:05:45 -07002153 ret = copy_mount_options(data, &data_page);
2154 if (ret < 0)
2155 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
Vegard Nossumeca6f532009-09-18 13:05:45 -07002157 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2158 (void *) data_page);
2159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002161out_data:
2162 kfree(kernel_dev);
2163out_dev:
2164 putname(kernel_dir);
2165out_dir:
2166 kfree(kernel_type);
2167out_type:
2168 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169}
2170
2171/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 * pivot_root Semantics:
2173 * Moves the root file system of the current process to the directory put_old,
2174 * makes new_root as the new root file system of the current process, and sets
2175 * root/cwd of all processes which had them on the current root to new_root.
2176 *
2177 * Restrictions:
2178 * The new_root and put_old must be directories, and must not be on the
2179 * same file system as the current process root. The put_old must be
2180 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2181 * pointed to by put_old must yield the same directory as new_root. No other
2182 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2183 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002184 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2185 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2186 * in this situation.
2187 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 * Notes:
2189 * - we don't move root/cwd if they are not at the root (reason: if something
2190 * cared enough to change them, it's probably wrong to force them elsewhere)
2191 * - it's okay to pick a root that isn't the root of a file system, e.g.
2192 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2193 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2194 * first.
2195 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002196SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2197 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198{
2199 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002200 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 int error;
2202
2203 if (!capable(CAP_SYS_ADMIN))
2204 return -EPERM;
2205
Al Viro2d8f3032008-07-22 09:59:21 -04002206 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 if (error)
2208 goto out0;
2209 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002210 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 goto out1;
2212
Al Viro2d8f3032008-07-22 09:59:21 -04002213 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 if (error)
2215 goto out1;
2216
Al Viro2d8f3032008-07-22 09:59:21 -04002217 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002219 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 goto out1;
2221 }
2222
2223 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002224 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002225 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002227 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002228 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002230 if (IS_MNT_SHARED(old.mnt) ||
2231 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002232 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002233 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002234 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 goto out2;
2236 error = -ENOENT;
Al Viro2d8f3032008-07-22 09:59:21 -04002237 if (IS_DEADDIR(new.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002239 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002241 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 goto out2;
2243 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002244 if (new.mnt == root.mnt ||
2245 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 goto out2; /* loop, on the same file system */
2247 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002248 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002250 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002251 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002252 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002254 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002255 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002256 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002257 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002259 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 for (;;) {
2261 if (tmp->mnt_parent == tmp)
2262 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002263 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 break;
2265 tmp = tmp->mnt_parent;
2266 }
Al Viro2d8f3032008-07-22 09:59:21 -04002267 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002269 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002271 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002272 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002273 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002274 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002275 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002276 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002277 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002279 chroot_fs_refs(&root, &new);
2280 security_sb_post_pivotroot(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002282 path_put(&root_parent);
2283 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002285 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002286 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002287 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002288 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002290 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 return error;
2293out3:
2294 spin_unlock(&vfsmount_lock);
2295 goto out2;
2296}
2297
2298static void __init init_mount_tree(void)
2299{
2300 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002301 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002302 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303
2304 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2305 if (IS_ERR(mnt))
2306 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002307 ns = create_mnt_ns(mnt);
2308 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002311 init_task.nsproxy->mnt_ns = ns;
2312 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
Jan Blunckac748a02008-02-14 19:34:39 -08002314 root.mnt = ns->root;
2315 root.dentry = ns->root->mnt_root;
2316
2317 set_fs_pwd(current->fs, &root);
2318 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319}
2320
Denis Cheng74bf17c2007-10-16 23:26:30 -07002321void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002323 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002324 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Ram Pai390c6842005-11-07 17:17:51 -05002326 init_rwsem(&namespace_sem);
2327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002329 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
Ram Paib58fed82005-11-07 17:16:09 -05002331 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
2333 if (!mount_hashtable)
2334 panic("Failed to allocate mount hash table\n");
2335
Eric Dumazet13f14b42008-02-06 01:37:57 -08002336 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
Eric Dumazet13f14b42008-02-06 01:37:57 -08002338 for (u = 0; u < HASH_SIZE; u++)
2339 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002341 err = sysfs_init();
2342 if (err)
2343 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002344 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002345 fs_kobj = kobject_create_and_add("fs", NULL);
2346 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002347 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 init_rootfs();
2349 init_mount_tree();
2350}
2351
Trond Myklebust616511d2009-06-22 15:09:13 -04002352void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002354 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002355
Al Virod498b252010-02-05 02:21:06 -05002356 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002357 return;
Ram Pai390c6842005-11-07 17:17:51 -05002358 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 spin_lock(&vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002360 umount_tree(ns->root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002362 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002363 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002364 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002366EXPORT_SYMBOL(put_mnt_ns);