blob: 88058de59c7c2ca9edf14453f1fbeeaa4017bfa8 [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);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500631 goto repeat;
632 }
633}
634
635EXPORT_SYMBOL(mntput_no_expire);
636
637void mnt_pin(struct vfsmount *mnt)
638{
639 spin_lock(&vfsmount_lock);
640 mnt->mnt_pinned++;
641 spin_unlock(&vfsmount_lock);
642}
643
644EXPORT_SYMBOL(mnt_pin);
645
646void mnt_unpin(struct vfsmount *mnt)
647{
648 spin_lock(&vfsmount_lock);
649 if (mnt->mnt_pinned) {
650 atomic_inc(&mnt->mnt_count);
651 mnt->mnt_pinned--;
652 }
653 spin_unlock(&vfsmount_lock);
654}
655
656EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800658static inline void mangle(struct seq_file *m, const char *s)
659{
660 seq_escape(m, s, " \t\n\\");
661}
662
663/*
664 * Simple .show_options callback for filesystems which don't want to
665 * implement more complex mount option showing.
666 *
667 * See also save_mount_options().
668 */
669int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
670{
Al Viro2a32ceb2009-05-08 16:05:57 -0400671 const char *options;
672
673 rcu_read_lock();
674 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800675
676 if (options != NULL && options[0]) {
677 seq_putc(m, ',');
678 mangle(m, options);
679 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400680 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800681
682 return 0;
683}
684EXPORT_SYMBOL(generic_show_options);
685
686/*
687 * If filesystem uses generic_show_options(), this function should be
688 * called from the fill_super() callback.
689 *
690 * The .remount_fs callback usually needs to be handled in a special
691 * way, to make sure, that previous options are not overwritten if the
692 * remount fails.
693 *
694 * Also note, that if the filesystem's .remount_fs function doesn't
695 * reset all options to their default value, but changes only newly
696 * given options, then the displayed options will not reflect reality
697 * any more.
698 */
699void save_mount_options(struct super_block *sb, char *options)
700{
Al Viro2a32ceb2009-05-08 16:05:57 -0400701 BUG_ON(sb->s_options);
702 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800703}
704EXPORT_SYMBOL(save_mount_options);
705
Al Viro2a32ceb2009-05-08 16:05:57 -0400706void replace_mount_options(struct super_block *sb, char *options)
707{
708 char *old = sb->s_options;
709 rcu_assign_pointer(sb->s_options, options);
710 if (old) {
711 synchronize_rcu();
712 kfree(old);
713 }
714}
715EXPORT_SYMBOL(replace_mount_options);
716
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100717#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718/* iterator */
719static void *m_start(struct seq_file *m, loff_t *pos)
720{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100721 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Ram Pai390c6842005-11-07 17:17:51 -0500723 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100724 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
727static void *m_next(struct seq_file *m, void *v, loff_t *pos)
728{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100729 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700730
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100731 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
733
734static void m_stop(struct seq_file *m, void *v)
735{
Ram Pai390c6842005-11-07 17:17:51 -0500736 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
Al Viro9f5596a2010-02-05 00:40:25 -0500739int mnt_had_events(struct proc_mounts *p)
740{
741 struct mnt_namespace *ns = p->ns;
742 int res = 0;
743
744 spin_lock(&vfsmount_lock);
745 if (p->event != ns->event) {
746 p->event = ns->event;
747 res = 1;
748 }
749 spin_unlock(&vfsmount_lock);
750
751 return res;
752}
753
Ram Pai2d4d4862008-03-27 13:06:25 +0100754struct proc_fs_info {
755 int flag;
756 const char *str;
757};
758
Eric Paris2069f452008-07-04 09:47:13 +1000759static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
Ram Pai2d4d4862008-03-27 13:06:25 +0100761 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 { MS_SYNCHRONOUS, ",sync" },
763 { MS_DIRSYNC, ",dirsync" },
764 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 { 0, NULL }
766 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100767 const struct proc_fs_info *fs_infop;
768
769 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
770 if (sb->s_flags & fs_infop->flag)
771 seq_puts(m, fs_infop->str);
772 }
Eric Paris2069f452008-07-04 09:47:13 +1000773
774 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100775}
776
777static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
778{
779 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 { MNT_NOSUID, ",nosuid" },
781 { MNT_NODEV, ",nodev" },
782 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800783 { MNT_NOATIME, ",noatime" },
784 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800785 { MNT_RELATIME, ",relatime" },
Matthew Garrettd0adde52009-03-26 17:49:56 +0000786 { MNT_STRICTATIME, ",strictatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 { 0, NULL }
788 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100789 const struct proc_fs_info *fs_infop;
790
791 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
792 if (mnt->mnt_flags & fs_infop->flag)
793 seq_puts(m, fs_infop->str);
794 }
795}
796
797static void show_type(struct seq_file *m, struct super_block *sb)
798{
799 mangle(m, sb->s_type->name);
800 if (sb->s_subtype && sb->s_subtype[0]) {
801 seq_putc(m, '.');
802 mangle(m, sb->s_subtype);
803 }
804}
805
806static int show_vfsmnt(struct seq_file *m, void *v)
807{
808 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
809 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800810 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
813 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800814 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100816 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800817 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000818 err = show_sb_opts(m, mnt->mnt_sb);
819 if (err)
820 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100821 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 if (mnt->mnt_sb->s_op->show_options)
823 err = mnt->mnt_sb->s_op->show_options(m, mnt);
824 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000825out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 return err;
827}
828
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100829const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 .start = m_start,
831 .next = m_next,
832 .stop = m_stop,
833 .show = show_vfsmnt
834};
835
Ram Pai2d4d4862008-03-27 13:06:25 +0100836static int show_mountinfo(struct seq_file *m, void *v)
837{
838 struct proc_mounts *p = m->private;
839 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
840 struct super_block *sb = mnt->mnt_sb;
841 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
842 struct path root = p->root;
843 int err = 0;
844
845 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
846 MAJOR(sb->s_dev), MINOR(sb->s_dev));
847 seq_dentry(m, mnt->mnt_root, " \t\n\\");
848 seq_putc(m, ' ');
849 seq_path_root(m, &mnt_path, &root, " \t\n\\");
850 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
851 /*
852 * Mountpoint is outside root, discard that one. Ugly,
853 * but less so than trying to do that in iterator in a
854 * race-free way (due to renames).
855 */
856 return SEQ_SKIP;
857 }
858 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
859 show_mnt_opts(m, mnt);
860
861 /* Tagged fields ("foo:X" or "bar") */
862 if (IS_MNT_SHARED(mnt))
863 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100864 if (IS_MNT_SLAVE(mnt)) {
865 int master = mnt->mnt_master->mnt_group_id;
866 int dom = get_dominating_id(mnt, &p->root);
867 seq_printf(m, " master:%i", master);
868 if (dom && dom != master)
869 seq_printf(m, " propagate_from:%i", dom);
870 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100871 if (IS_MNT_UNBINDABLE(mnt))
872 seq_puts(m, " unbindable");
873
874 /* Filesystem specific data */
875 seq_puts(m, " - ");
876 show_type(m, sb);
877 seq_putc(m, ' ');
878 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
879 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000880 err = show_sb_opts(m, sb);
881 if (err)
882 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100883 if (sb->s_op->show_options)
884 err = sb->s_op->show_options(m, mnt);
885 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000886out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100887 return err;
888}
889
890const struct seq_operations mountinfo_op = {
891 .start = m_start,
892 .next = m_next,
893 .stop = m_stop,
894 .show = show_mountinfo,
895};
896
Chuck Leverb4629fe2006-03-20 13:44:12 -0500897static int show_vfsstat(struct seq_file *m, void *v)
898{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700899 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800900 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500901 int err = 0;
902
903 /* device */
904 if (mnt->mnt_devname) {
905 seq_puts(m, "device ");
906 mangle(m, mnt->mnt_devname);
907 } else
908 seq_puts(m, "no device");
909
910 /* mount point */
911 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800912 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500913 seq_putc(m, ' ');
914
915 /* file system type */
916 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100917 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500918
919 /* optional statistics */
920 if (mnt->mnt_sb->s_op->show_stats) {
921 seq_putc(m, ' ');
922 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
923 }
924
925 seq_putc(m, '\n');
926 return err;
927}
928
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100929const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500930 .start = m_start,
931 .next = m_next,
932 .stop = m_stop,
933 .show = show_vfsstat,
934};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100935#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500936
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937/**
938 * may_umount_tree - check if a mount tree is busy
939 * @mnt: root of mount tree
940 *
941 * This is called to check if a tree of mounts has any
942 * open files, pwds, chroots or sub mounts that are
943 * busy.
944 */
945int may_umount_tree(struct vfsmount *mnt)
946{
Ram Pai36341f62005-11-07 17:17:22 -0500947 int actual_refs = 0;
948 int minimum_refs = 0;
949 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500952 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 actual_refs += atomic_read(&p->mnt_count);
954 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956 spin_unlock(&vfsmount_lock);
957
958 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800959 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Ian Kente3474a82006-03-27 01:14:51 -0800961 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
964EXPORT_SYMBOL(may_umount_tree);
965
966/**
967 * may_umount - check if a mount point is busy
968 * @mnt: root of mount
969 *
970 * This is called to check if a mount point has any
971 * open files, pwds, chroots or sub mounts. If the
972 * mount has sub mounts this will return busy
973 * regardless of whether the sub mounts are busy.
974 *
975 * Doesn't take quota and stuff into account. IOW, in some cases it will
976 * give false negatives. The main reason why it's here is that we need
977 * a non-destructive way to look for easily umountable filesystems.
978 */
979int may_umount(struct vfsmount *mnt)
980{
Ian Kente3474a82006-03-27 01:14:51 -0800981 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -0500982 down_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500983 spin_lock(&vfsmount_lock);
984 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800985 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500986 spin_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -0500987 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500988 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989}
990
991EXPORT_SYMBOL(may_umount);
992
Ram Paib90fa9a2005-11-07 17:19:50 -0500993void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500995 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800996 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700997 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500998 list_del_init(&mnt->mnt_hash);
999 if (mnt->mnt_parent != mnt) {
1000 struct dentry *dentry;
1001 struct vfsmount *m;
1002 spin_lock(&vfsmount_lock);
1003 dentry = mnt->mnt_mountpoint;
1004 m = mnt->mnt_parent;
1005 mnt->mnt_mountpoint = mnt->mnt_root;
1006 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001007 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001009 dput(dentry);
1010 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001013 }
1014}
1015
Ram Paia05964f2005-11-07 17:20:17 -05001016void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001017{
1018 struct vfsmount *p;
1019
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001020 for (p = mnt; p; p = next_mnt(p, mnt))
1021 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001022
Ram Paia05964f2005-11-07 17:20:17 -05001023 if (propagate)
1024 propagate_umount(kill);
1025
Ram Pai70fbcdf2005-11-07 17:17:04 -05001026 list_for_each_entry(p, kill, mnt_hash) {
1027 list_del_init(&p->mnt_expire);
1028 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001029 __touch_mnt_namespace(p->mnt_ns);
1030 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001031 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001032 if (p->mnt_parent != p) {
1033 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001034 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001035 }
Ram Paia05964f2005-11-07 17:20:17 -05001036 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 }
1038}
1039
Al Viroc35038b2008-03-22 00:46:23 -04001040static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042static int do_umount(struct vfsmount *mnt, int flags)
1043{
Ram Paib58fed82005-11-07 17:16:09 -05001044 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001046 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
1048 retval = security_sb_umount(mnt, flags);
1049 if (retval)
1050 return retval;
1051
1052 /*
1053 * Allow userspace to request a mountpoint be expired rather than
1054 * unmounting unconditionally. Unmount only happens if:
1055 * (1) the mark is already set (the mark is cleared by mntput())
1056 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1057 */
1058 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001059 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 flags & (MNT_FORCE | MNT_DETACH))
1061 return -EINVAL;
1062
1063 if (atomic_read(&mnt->mnt_count) != 2)
1064 return -EBUSY;
1065
1066 if (!xchg(&mnt->mnt_expiry_mark, 1))
1067 return -EAGAIN;
1068 }
1069
1070 /*
1071 * If we may have to abort operations to get out of this
1072 * mount, and they will themselves hold resources we must
1073 * allow the fs to do things. In the Unix tradition of
1074 * 'Gee thats tricky lets do it in userspace' the umount_begin
1075 * might fail to complete on the first run through as other tasks
1076 * must return, and the like. Thats for the mount program to worry
1077 * about for the moment.
1078 */
1079
Al Viro42faad92008-04-24 07:21:56 -04001080 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001081 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084 /*
1085 * No sense to grab the lock for this test, but test itself looks
1086 * somewhat bogus. Suggestions for better replacement?
1087 * Ho-hum... In principle, we might treat that as umount + switch
1088 * to rootfs. GC would eventually take care of the old vfsmount.
1089 * Actually it makes sense, especially if rootfs would contain a
1090 * /reboot - static binary that would close all descriptors and
1091 * call reboot(9). Then init(8) could umount root and exec /reboot.
1092 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001093 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 /*
1095 * Special case for "unmounting" root ...
1096 * we just try to remount it readonly.
1097 */
1098 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001099 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 up_write(&sb->s_umount);
1102 return retval;
1103 }
1104
Ram Pai390c6842005-11-07 17:17:51 -05001105 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001107 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Al Viroc35038b2008-03-22 00:46:23 -04001109 if (!(flags & MNT_DETACH))
1110 shrink_submounts(mnt, &umount_list);
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001113 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001115 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 retval = 0;
1117 }
1118 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001119 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001120 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 return retval;
1122}
1123
1124/*
1125 * Now umount can handle mount points as well as block devices.
1126 * This is important for filesystems which use unnamed block devices.
1127 *
1128 * We now support a flag for forced unmount like the other 'big iron'
1129 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1130 */
1131
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001132SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133{
Al Viro2d8f3032008-07-22 09:59:21 -04001134 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001136 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001138 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1139 return -EINVAL;
1140
1141 if (!(flags & UMOUNT_NOFOLLOW))
1142 lookup_flags |= LOOKUP_FOLLOW;
1143
1144 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (retval)
1146 goto out;
1147 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001148 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001150 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 goto dput_and_out;
1152
1153 retval = -EPERM;
1154 if (!capable(CAP_SYS_ADMIN))
1155 goto dput_and_out;
1156
Al Viro2d8f3032008-07-22 09:59:21 -04001157 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001159 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001160 dput(path.dentry);
1161 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162out:
1163 return retval;
1164}
1165
1166#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1167
1168/*
Ram Paib58fed82005-11-07 17:16:09 -05001169 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001171SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172{
Ram Paib58fed82005-11-07 17:16:09 -05001173 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
1176#endif
1177
Al Viro2d92ab32008-08-02 00:51:11 -04001178static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179{
1180 if (capable(CAP_SYS_ADMIN))
1181 return 0;
1182 return -EPERM;
1183#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001184 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001186 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001187 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return -EPERM;
1189 }
Al Viro2d92ab32008-08-02 00:51:11 -04001190 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 return -EPERM;
1192 return 0;
1193#endif
1194}
1195
Ram Paib90fa9a2005-11-07 17:19:50 -05001196struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001197 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
1199 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001200 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Ram Pai9676f0c2005-11-07 17:21:20 -05001202 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1203 return NULL;
1204
Ram Pai36341f62005-11-07 17:17:22 -05001205 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 if (!q)
1207 goto Enomem;
1208 q->mnt_mountpoint = mnt->mnt_mountpoint;
1209
1210 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001211 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001212 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 continue;
1214
1215 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001216 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1217 s = skip_mnt_tree(s);
1218 continue;
1219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 while (p != s->mnt_parent) {
1221 p = p->mnt_parent;
1222 q = q->mnt_parent;
1223 }
1224 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001225 path.mnt = q;
1226 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001227 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (!q)
1229 goto Enomem;
1230 spin_lock(&vfsmount_lock);
1231 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001232 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 spin_unlock(&vfsmount_lock);
1234 }
1235 }
1236 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001237Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001239 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001241 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001243 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245 return NULL;
1246}
1247
Al Viro589ff872009-04-18 03:28:19 -04001248struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001249{
1250 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001251 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001252 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001253 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001254 return tree;
1255}
1256
1257void drop_collected_mounts(struct vfsmount *mnt)
1258{
1259 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001260 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001261 spin_lock(&vfsmount_lock);
1262 umount_tree(mnt, 0, &umount_list);
1263 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001264 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001265 release_mounts(&umount_list);
1266}
1267
Al Viro1f707132010-01-30 22:51:25 -05001268int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1269 struct vfsmount *root)
1270{
1271 struct vfsmount *mnt;
1272 int res = f(root, arg);
1273 if (res)
1274 return res;
1275 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1276 res = f(mnt, arg);
1277 if (res)
1278 return res;
1279 }
1280 return 0;
1281}
1282
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001283static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1284{
1285 struct vfsmount *p;
1286
1287 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1288 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1289 mnt_release_group_id(p);
1290 }
1291}
1292
1293static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1294{
1295 struct vfsmount *p;
1296
1297 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1298 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1299 int err = mnt_alloc_group_id(p);
1300 if (err) {
1301 cleanup_group_ids(mnt, p);
1302 return err;
1303 }
1304 }
1305 }
1306
1307 return 0;
1308}
1309
Ram Paib90fa9a2005-11-07 17:19:50 -05001310/*
1311 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001312 * @nd : place the mount tree @source_mnt is attached
1313 * @parent_nd : if non-null, detach the source_mnt from its parent and
1314 * store the parent mount and mountpoint dentry.
1315 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001316 *
1317 * NOTE: in the table below explains the semantics when a source mount
1318 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001319 * ---------------------------------------------------------------------------
1320 * | BIND MOUNT OPERATION |
1321 * |**************************************************************************
1322 * | source-->| shared | private | slave | unbindable |
1323 * | dest | | | | |
1324 * | | | | | | |
1325 * | v | | | | |
1326 * |**************************************************************************
1327 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1328 * | | | | | |
1329 * |non-shared| shared (+) | private | slave (*) | invalid |
1330 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001331 * A bind operation clones the source mount and mounts the clone on the
1332 * destination mount.
1333 *
1334 * (++) the cloned mount is propagated to all the mounts in the propagation
1335 * tree of the destination mount and the cloned mount is added to
1336 * the peer group of the source mount.
1337 * (+) the cloned mount is created under the destination mount and is marked
1338 * as shared. The cloned mount is added to the peer group of the source
1339 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001340 * (+++) the mount is propagated to all the mounts in the propagation tree
1341 * of the destination mount and the cloned mount is made slave
1342 * of the same master as that of the source mount. The cloned mount
1343 * is marked as 'shared and slave'.
1344 * (*) the cloned mount is made a slave of the same master as that of the
1345 * source mount.
1346 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001347 * ---------------------------------------------------------------------------
1348 * | MOVE MOUNT OPERATION |
1349 * |**************************************************************************
1350 * | source-->| shared | private | slave | unbindable |
1351 * | dest | | | | |
1352 * | | | | | | |
1353 * | v | | | | |
1354 * |**************************************************************************
1355 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1356 * | | | | | |
1357 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1358 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001359 *
1360 * (+) the mount is moved to the destination. And is then propagated to
1361 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001362 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001363 * (+++) the mount is moved to the destination and is then propagated to
1364 * all the mounts belonging to the destination mount's propagation tree.
1365 * the mount is marked as 'shared and slave'.
1366 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001367 *
1368 * if the source mount is a tree, the operations explained above is
1369 * applied to each mount in the tree.
1370 * Must be called without spinlocks held, since this function can sleep
1371 * in allocations.
1372 */
1373static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001374 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001375{
1376 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001377 struct vfsmount *dest_mnt = path->mnt;
1378 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001379 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001380 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001381
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001382 if (IS_MNT_SHARED(dest_mnt)) {
1383 err = invent_group_ids(source_mnt, true);
1384 if (err)
1385 goto out;
1386 }
1387 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1388 if (err)
1389 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001390
Al Virodf1a1ad2010-01-16 12:57:40 -05001391 spin_lock(&vfsmount_lock);
1392
Ram Paib90fa9a2005-11-07 17:19:50 -05001393 if (IS_MNT_SHARED(dest_mnt)) {
1394 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1395 set_mnt_shared(p);
1396 }
Al Viro1a390682008-03-21 20:48:19 -04001397 if (parent_path) {
1398 detach_mnt(source_mnt, parent_path);
1399 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001400 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001401 } else {
1402 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1403 commit_tree(source_mnt);
1404 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001405
1406 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1407 list_del_init(&child->mnt_hash);
1408 commit_tree(child);
1409 }
1410 spin_unlock(&vfsmount_lock);
1411 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001412
1413 out_cleanup_ids:
1414 if (IS_MNT_SHARED(dest_mnt))
1415 cleanup_group_ids(source_mnt, NULL);
1416 out:
1417 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001418}
1419
Al Viro8c3ee422008-03-22 18:00:39 -04001420static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421{
1422 int err;
1423 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1424 return -EINVAL;
1425
Al Viro8c3ee422008-03-22 18:00:39 -04001426 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1428 return -ENOTDIR;
1429
1430 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001431 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001432 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 goto out_unlock;
1434
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001435 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001436 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001438 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return err;
1440}
1441
1442/*
Ram Pai07b20882005-11-07 17:19:07 -05001443 * recursively change the type of the mountpoint.
1444 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001445static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001446{
Al Viro2d92ab32008-08-02 00:51:11 -04001447 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001448 int recurse = flag & MS_REC;
1449 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001450 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001451
Miklos Szerediee6f9582007-05-08 00:30:40 -07001452 if (!capable(CAP_SYS_ADMIN))
1453 return -EPERM;
1454
Al Viro2d92ab32008-08-02 00:51:11 -04001455 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001456 return -EINVAL;
1457
1458 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001459 if (type == MS_SHARED) {
1460 err = invent_group_ids(mnt, recurse);
1461 if (err)
1462 goto out_unlock;
1463 }
1464
Ram Pai07b20882005-11-07 17:19:07 -05001465 spin_lock(&vfsmount_lock);
1466 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1467 change_mnt_propagation(m, type);
1468 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001469
1470 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001471 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001472 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001473}
1474
1475/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 * do loopback mount.
1477 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001478static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001479 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480{
Al Viro2d92ab32008-08-02 00:51:11 -04001481 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001483 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 if (err)
1485 return err;
1486 if (!old_name || !*old_name)
1487 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001488 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 if (err)
1490 return err;
1491
Ram Pai390c6842005-11-07 17:17:51 -05001492 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001494 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001495 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001496
Al Viro2d92ab32008-08-02 00:51:11 -04001497 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001498 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Al Viroccd48bc2005-11-07 17:15:04 -05001500 err = -ENOMEM;
1501 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001502 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001503 else
Al Viro2d92ab32008-08-02 00:51:11 -04001504 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001505
1506 if (!mnt)
1507 goto out;
1508
Al Viro2d92ab32008-08-02 00:51:11 -04001509 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001510 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001511 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001513 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001515 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Al Viroccd48bc2005-11-07 17:15:04 -05001518out:
Ram Pai390c6842005-11-07 17:17:51 -05001519 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001520 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return err;
1522}
1523
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001524static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1525{
1526 int error = 0;
1527 int readonly_request = 0;
1528
1529 if (ms_flags & MS_RDONLY)
1530 readonly_request = 1;
1531 if (readonly_request == __mnt_is_readonly(mnt))
1532 return 0;
1533
1534 if (readonly_request)
1535 error = mnt_make_readonly(mnt);
1536 else
1537 __mnt_unmake_readonly(mnt);
1538 return error;
1539}
1540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541/*
1542 * change filesystem flags. dir should be a physical root of filesystem.
1543 * If you've mounted a non-root directory somewhere and want to do remount
1544 * on it - tough luck.
1545 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001546static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 void *data)
1548{
1549 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001550 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551
1552 if (!capable(CAP_SYS_ADMIN))
1553 return -EPERM;
1554
Al Viro2d92ab32008-08-02 00:51:11 -04001555 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 return -EINVAL;
1557
Al Viro2d92ab32008-08-02 00:51:11 -04001558 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 return -EINVAL;
1560
1561 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001562 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001563 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001564 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001565 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001566 if (!err) {
1567 spin_lock(&vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001568 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001569 path->mnt->mnt_flags = mnt_flags;
Al Viro7b43a792010-01-16 13:01:26 -05001570 spin_unlock(&vfsmount_lock);
1571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001573 if (!err) {
Dan Williams0e55a7c2008-09-26 19:01:20 -07001574 spin_lock(&vfsmount_lock);
1575 touch_mnt_namespace(path->mnt->mnt_ns);
1576 spin_unlock(&vfsmount_lock);
1577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return err;
1579}
1580
Ram Pai9676f0c2005-11-07 17:21:20 -05001581static inline int tree_contains_unbindable(struct vfsmount *mnt)
1582{
1583 struct vfsmount *p;
1584 for (p = mnt; p; p = next_mnt(p, mnt)) {
1585 if (IS_MNT_UNBINDABLE(p))
1586 return 1;
1587 }
1588 return 0;
1589}
1590
Al Viro0a0d8a42008-08-02 00:55:27 -04001591static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
Al Viro2d92ab32008-08-02 00:51:11 -04001593 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 struct vfsmount *p;
1595 int err = 0;
1596 if (!capable(CAP_SYS_ADMIN))
1597 return -EPERM;
1598 if (!old_name || !*old_name)
1599 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001600 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (err)
1602 return err;
1603
Ram Pai390c6842005-11-07 17:17:51 -05001604 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001605 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001606 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 ;
1608 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001609 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 goto out;
1611
1612 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001613 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001614 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 goto out1;
1616
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001617 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001618 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
1620 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001621 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001622 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Al Viro2d92ab32008-08-02 00:51:11 -04001624 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001625 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Al Viro2d92ab32008-08-02 00:51:11 -04001627 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1628 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001629 goto out1;
1630 /*
1631 * Don't move a mount residing in a shared parent.
1632 */
Al Viro2d92ab32008-08-02 00:51:11 -04001633 if (old_path.mnt->mnt_parent &&
1634 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001635 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001636 /*
1637 * Don't move a mount tree containing unbindable mounts to a destination
1638 * mount which is shared.
1639 */
Al Viro2d92ab32008-08-02 00:51:11 -04001640 if (IS_MNT_SHARED(path->mnt) &&
1641 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001642 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001644 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1645 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001646 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
Al Viro2d92ab32008-08-02 00:51:11 -04001648 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001649 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001650 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 /* if the mount is moved, it should no longer be expire
1653 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001654 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001656 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657out:
Ram Pai390c6842005-11-07 17:17:51 -05001658 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001660 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001661 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 return err;
1663}
1664
1665/*
1666 * create a new mount for userspace and request it to be added into the
1667 * namespace's tree
1668 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001669static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 int mnt_flags, char *name, void *data)
1671{
1672 struct vfsmount *mnt;
1673
Vegard Nossumeca6f532009-09-18 13:05:45 -07001674 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 return -EINVAL;
1676
1677 /* we need capabilities... */
1678 if (!capable(CAP_SYS_ADMIN))
1679 return -EPERM;
1680
Al Viro7f78d4c2009-05-08 13:34:06 -04001681 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001683 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 if (IS_ERR(mnt))
1685 return PTR_ERR(mnt);
1686
Al Viro2d92ab32008-08-02 00:51:11 -04001687 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
1690/*
1691 * add a mount into a namespace's mount tree
1692 * - provide the option of adding the new mount to an expiration list
1693 */
Al Viro8d66bf52008-08-01 09:05:54 -04001694int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 int mnt_flags, struct list_head *fslist)
1696{
1697 int err;
1698
Al Viro80893522010-02-05 09:30:46 -05001699 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
Al Viro27d55f12010-01-16 13:07:36 -05001700
Ram Pai390c6842005-11-07 17:17:51 -05001701 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001703 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001704 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 ;
1706 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001707 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 goto unlock;
1709
1710 /* Refuse the same filesystem on the same mount point */
1711 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001712 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1713 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 goto unlock;
1715
1716 err = -EINVAL;
1717 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1718 goto unlock;
1719
1720 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001721 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001722 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Al Viro6758f952008-03-22 16:14:30 -04001724 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001725 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001726
Ram Pai390c6842005-11-07 17:17:51 -05001727 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001728 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001731 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 mntput(newmnt);
1733 return err;
1734}
1735
1736EXPORT_SYMBOL_GPL(do_add_mount);
1737
Trond Myklebust5528f9112006-06-09 09:34:17 -04001738/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 * process a list of expirable mountpoints with the intent of discarding any
1740 * mountpoints that aren't in use and haven't been touched since last we came
1741 * here
1742 */
1743void mark_mounts_for_expiry(struct list_head *mounts)
1744{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 struct vfsmount *mnt, *next;
1746 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001747 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
1749 if (list_empty(mounts))
1750 return;
1751
Al Virobcc5c7d2008-03-22 00:21:53 -04001752 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 spin_lock(&vfsmount_lock);
1754
1755 /* extract from the expiration list every vfsmount that matches the
1756 * following criteria:
1757 * - only referenced by its parent vfsmount
1758 * - still marked for expiry (marked on the last call here; marks are
1759 * cleared by mntput())
1760 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001761 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001763 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001765 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001767 while (!list_empty(&graveyard)) {
1768 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1769 touch_mnt_namespace(mnt->mnt_ns);
1770 umount_tree(mnt, 1, &umounts);
1771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001773 up_write(&namespace_sem);
1774
1775 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
1778EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1779
1780/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001781 * Ripoff of 'select_parent()'
1782 *
1783 * search the list of submounts for a given mountpoint, and move any
1784 * shrinkable submounts to the 'graveyard' list.
1785 */
1786static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1787{
1788 struct vfsmount *this_parent = parent;
1789 struct list_head *next;
1790 int found = 0;
1791
1792repeat:
1793 next = this_parent->mnt_mounts.next;
1794resume:
1795 while (next != &this_parent->mnt_mounts) {
1796 struct list_head *tmp = next;
1797 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1798
1799 next = tmp->next;
1800 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1801 continue;
1802 /*
1803 * Descend a level if the d_mounts list is non-empty.
1804 */
1805 if (!list_empty(&mnt->mnt_mounts)) {
1806 this_parent = mnt;
1807 goto repeat;
1808 }
1809
1810 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001811 list_move_tail(&mnt->mnt_expire, graveyard);
1812 found++;
1813 }
1814 }
1815 /*
1816 * All done at this level ... ascend and resume the search
1817 */
1818 if (this_parent != parent) {
1819 next = this_parent->mnt_child.next;
1820 this_parent = this_parent->mnt_parent;
1821 goto resume;
1822 }
1823 return found;
1824}
1825
1826/*
1827 * process a list of expirable mountpoints with the intent of discarding any
1828 * submounts of a specific parent mountpoint
1829 */
Al Viroc35038b2008-03-22 00:46:23 -04001830static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001831{
1832 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001833 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001834
Trond Myklebust5528f9112006-06-09 09:34:17 -04001835 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001836 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001837 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001838 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001839 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001840 touch_mnt_namespace(m->mnt_ns);
1841 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001842 }
1843 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001844}
1845
Trond Myklebust5528f9112006-06-09 09:34:17 -04001846/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 * Some copy_from_user() implementations do not return the exact number of
1848 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1849 * Note that this function differs from copy_from_user() in that it will oops
1850 * on bad values of `to', rather than returning a short copy.
1851 */
Ram Paib58fed82005-11-07 17:16:09 -05001852static long exact_copy_from_user(void *to, const void __user * from,
1853 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854{
1855 char *t = to;
1856 const char __user *f = from;
1857 char c;
1858
1859 if (!access_ok(VERIFY_READ, from, n))
1860 return n;
1861
1862 while (n) {
1863 if (__get_user(c, f)) {
1864 memset(t, 0, n);
1865 break;
1866 }
1867 *t++ = c;
1868 f++;
1869 n--;
1870 }
1871 return n;
1872}
1873
Ram Paib58fed82005-11-07 17:16:09 -05001874int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875{
1876 int i;
1877 unsigned long page;
1878 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001879
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 *where = 0;
1881 if (!data)
1882 return 0;
1883
1884 if (!(page = __get_free_page(GFP_KERNEL)))
1885 return -ENOMEM;
1886
1887 /* We only care that *some* data at the address the user
1888 * gave us is valid. Just in case, we'll zero
1889 * the remainder of the page.
1890 */
1891 /* copy_from_user cannot cross TASK_SIZE ! */
1892 size = TASK_SIZE - (unsigned long)data;
1893 if (size > PAGE_SIZE)
1894 size = PAGE_SIZE;
1895
1896 i = size - exact_copy_from_user((void *)page, data, size);
1897 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001898 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return -EFAULT;
1900 }
1901 if (i != PAGE_SIZE)
1902 memset((char *)page + i, 0, PAGE_SIZE - i);
1903 *where = page;
1904 return 0;
1905}
1906
Vegard Nossumeca6f532009-09-18 13:05:45 -07001907int copy_mount_string(const void __user *data, char **where)
1908{
1909 char *tmp;
1910
1911 if (!data) {
1912 *where = NULL;
1913 return 0;
1914 }
1915
1916 tmp = strndup_user(data, PAGE_SIZE);
1917 if (IS_ERR(tmp))
1918 return PTR_ERR(tmp);
1919
1920 *where = tmp;
1921 return 0;
1922}
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924/*
1925 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1926 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1927 *
1928 * data is a (void *) that can point to any structure up to
1929 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1930 * information (or be NULL).
1931 *
1932 * Pre-0.97 versions of mount() didn't have a flags word.
1933 * When the flags word was introduced its top half was required
1934 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1935 * Therefore, if this magic number is present, it carries no information
1936 * and must be discarded.
1937 */
Ram Paib58fed82005-11-07 17:16:09 -05001938long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 unsigned long flags, void *data_page)
1940{
Al Viro2d92ab32008-08-02 00:51:11 -04001941 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 int retval = 0;
1943 int mnt_flags = 0;
1944
1945 /* Discard magic */
1946 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1947 flags &= ~MS_MGC_MSK;
1948
1949 /* Basic sanity checks */
1950
1951 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1952 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 if (data_page)
1955 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1956
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09001957 /* ... and get the mountpoint */
1958 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1959 if (retval)
1960 return retval;
1961
1962 retval = security_sb_mount(dev_name, &path,
1963 type_page, flags, data_page);
1964 if (retval)
1965 goto dput_out;
1966
Andi Kleen613cbe32009-04-19 18:40:43 +02001967 /* Default to relatime unless overriden */
1968 if (!(flags & MS_NOATIME))
1969 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001970
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 /* Separate the per-mountpoint flags */
1972 if (flags & MS_NOSUID)
1973 mnt_flags |= MNT_NOSUID;
1974 if (flags & MS_NODEV)
1975 mnt_flags |= MNT_NODEV;
1976 if (flags & MS_NOEXEC)
1977 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001978 if (flags & MS_NOATIME)
1979 mnt_flags |= MNT_NOATIME;
1980 if (flags & MS_NODIRATIME)
1981 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001982 if (flags & MS_STRICTATIME)
1983 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001984 if (flags & MS_RDONLY)
1985 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001986
1987 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Matthew Garrettd0adde52009-03-26 17:49:56 +00001988 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1989 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001992 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 data_page);
1994 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001995 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001996 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04001997 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04001999 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 else
Al Viro2d92ab32008-08-02 00:51:11 -04002001 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 dev_name, data_page);
2003dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002004 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 return retval;
2006}
2007
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002008static struct mnt_namespace *alloc_mnt_ns(void)
2009{
2010 struct mnt_namespace *new_ns;
2011
2012 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2013 if (!new_ns)
2014 return ERR_PTR(-ENOMEM);
2015 atomic_set(&new_ns->count, 1);
2016 new_ns->root = NULL;
2017 INIT_LIST_HEAD(&new_ns->list);
2018 init_waitqueue_head(&new_ns->poll);
2019 new_ns->event = 0;
2020 return new_ns;
2021}
2022
JANAK DESAI741a2952006-02-07 12:59:00 -08002023/*
2024 * Allocate a new namespace structure and populate it with contents
2025 * copied from the namespace of the passed in task structure.
2026 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002027static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002028 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002030 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002031 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 struct vfsmount *p, *q;
2033
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002034 new_ns = alloc_mnt_ns();
2035 if (IS_ERR(new_ns))
2036 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Ram Pai390c6842005-11-07 17:17:51 -05002038 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002040 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002041 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002043 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002045 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 }
2047 spin_lock(&vfsmount_lock);
2048 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2049 spin_unlock(&vfsmount_lock);
2050
2051 /*
2052 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2053 * as belonging to new namespace. We have already acquired a private
2054 * fs_struct, so tsk->fs->lock is not needed.
2055 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002056 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 q = new_ns->root;
2058 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002059 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002061 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002063 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002065 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002067 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002070 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 q = next_mnt(q, new_ns->root);
2072 }
Ram Pai390c6842005-11-07 17:17:51 -05002073 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (rootmnt)
2076 mntput(rootmnt);
2077 if (pwdmnt)
2078 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
JANAK DESAI741a2952006-02-07 12:59:00 -08002080 return new_ns;
2081}
2082
Eric W. Biederman213dd262007-07-15 23:41:15 -07002083struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002084 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002085{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002086 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002087
Badari Pulavartye3222c42007-05-08 00:25:21 -07002088 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002089 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002090
2091 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002092 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002093
Badari Pulavartye3222c42007-05-08 00:25:21 -07002094 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002095
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002096 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002097 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098}
2099
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002100/**
2101 * create_mnt_ns - creates a private namespace and adds a root filesystem
2102 * @mnt: pointer to the new root filesystem mountpoint
2103 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002104struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002105{
2106 struct mnt_namespace *new_ns;
2107
2108 new_ns = alloc_mnt_ns();
2109 if (!IS_ERR(new_ns)) {
2110 mnt->mnt_ns = new_ns;
2111 new_ns->root = mnt;
2112 list_add(&new_ns->list, &new_ns->root->mnt_list);
2113 }
2114 return new_ns;
2115}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002116EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002117
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002118SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2119 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002121 int ret;
2122 char *kernel_type;
2123 char *kernel_dir;
2124 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
Vegard Nossumeca6f532009-09-18 13:05:45 -07002127 ret = copy_mount_string(type, &kernel_type);
2128 if (ret < 0)
2129 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Vegard Nossumeca6f532009-09-18 13:05:45 -07002131 kernel_dir = getname(dir_name);
2132 if (IS_ERR(kernel_dir)) {
2133 ret = PTR_ERR(kernel_dir);
2134 goto out_dir;
2135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
Vegard Nossumeca6f532009-09-18 13:05:45 -07002137 ret = copy_mount_string(dev_name, &kernel_dev);
2138 if (ret < 0)
2139 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Vegard Nossumeca6f532009-09-18 13:05:45 -07002141 ret = copy_mount_options(data, &data_page);
2142 if (ret < 0)
2143 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
Vegard Nossumeca6f532009-09-18 13:05:45 -07002145 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2146 (void *) data_page);
2147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002149out_data:
2150 kfree(kernel_dev);
2151out_dev:
2152 putname(kernel_dir);
2153out_dir:
2154 kfree(kernel_type);
2155out_type:
2156 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157}
2158
2159/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 * pivot_root Semantics:
2161 * Moves the root file system of the current process to the directory put_old,
2162 * makes new_root as the new root file system of the current process, and sets
2163 * root/cwd of all processes which had them on the current root to new_root.
2164 *
2165 * Restrictions:
2166 * The new_root and put_old must be directories, and must not be on the
2167 * same file system as the current process root. The put_old must be
2168 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2169 * pointed to by put_old must yield the same directory as new_root. No other
2170 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2171 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002172 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2173 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2174 * in this situation.
2175 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 * Notes:
2177 * - we don't move root/cwd if they are not at the root (reason: if something
2178 * cared enough to change them, it's probably wrong to force them elsewhere)
2179 * - it's okay to pick a root that isn't the root of a file system, e.g.
2180 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2181 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2182 * first.
2183 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002184SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2185 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186{
2187 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002188 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 int error;
2190
2191 if (!capable(CAP_SYS_ADMIN))
2192 return -EPERM;
2193
Al Viro2d8f3032008-07-22 09:59:21 -04002194 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (error)
2196 goto out0;
2197 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002198 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 goto out1;
2200
Al Viro2d8f3032008-07-22 09:59:21 -04002201 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 if (error)
2203 goto out1;
2204
Al Viro2d8f3032008-07-22 09:59:21 -04002205 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002207 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 goto out1;
2209 }
2210
2211 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002212 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002213 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002215 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002216 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002218 if (IS_MNT_SHARED(old.mnt) ||
2219 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002220 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002221 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002222 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 goto out2;
2224 error = -ENOENT;
Al Virod83c49f2010-04-30 17:17:09 -04002225 if (cant_mount(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002227 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002229 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 goto out2;
2231 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002232 if (new.mnt == root.mnt ||
2233 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 goto out2; /* loop, on the same file system */
2235 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002236 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002238 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002239 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002240 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002242 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002243 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002244 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002245 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002247 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 for (;;) {
2249 if (tmp->mnt_parent == tmp)
2250 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002251 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 break;
2253 tmp = tmp->mnt_parent;
2254 }
Al Viro2d8f3032008-07-22 09:59:21 -04002255 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002257 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002259 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002260 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002261 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002262 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002263 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002264 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002265 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002267 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002269 path_put(&root_parent);
2270 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002272 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002273 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002274 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002275 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002277 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 return error;
2280out3:
2281 spin_unlock(&vfsmount_lock);
2282 goto out2;
2283}
2284
2285static void __init init_mount_tree(void)
2286{
2287 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002288 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002289 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2292 if (IS_ERR(mnt))
2293 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002294 ns = create_mnt_ns(mnt);
2295 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002298 init_task.nsproxy->mnt_ns = ns;
2299 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300
Jan Blunckac748a02008-02-14 19:34:39 -08002301 root.mnt = ns->root;
2302 root.dentry = ns->root->mnt_root;
2303
2304 set_fs_pwd(current->fs, &root);
2305 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306}
2307
Denis Cheng74bf17c2007-10-16 23:26:30 -07002308void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002310 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002311 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
Ram Pai390c6842005-11-07 17:17:51 -05002313 init_rwsem(&namespace_sem);
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002316 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Ram Paib58fed82005-11-07 17:16:09 -05002318 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 if (!mount_hashtable)
2321 panic("Failed to allocate mount hash table\n");
2322
Eric Dumazet13f14b42008-02-06 01:37:57 -08002323 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
Eric Dumazet13f14b42008-02-06 01:37:57 -08002325 for (u = 0; u < HASH_SIZE; u++)
2326 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002328 err = sysfs_init();
2329 if (err)
2330 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002331 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002332 fs_kobj = kobject_create_and_add("fs", NULL);
2333 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002334 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 init_rootfs();
2336 init_mount_tree();
2337}
2338
Trond Myklebust616511d2009-06-22 15:09:13 -04002339void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002341 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002342
Al Virod498b252010-02-05 02:21:06 -05002343 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002344 return;
Ram Pai390c6842005-11-07 17:17:51 -05002345 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 spin_lock(&vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002347 umount_tree(ns->root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002349 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002350 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002351 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002353EXPORT_SYMBOL(put_mnt_ns);