]> git.ipfire.org Git - people/ms/linux.git/blame - fs/namespace.c
autofs4: Merge the remaining dentry ops tables
[people/ms/linux.git] / fs / namespace.c
CommitLineData
1da177e4
LT
1/*
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
1da177e4
LT
11#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
99b7db7b
NP
14#include <linux/spinlock.h>
15#include <linux/percpu.h>
1da177e4 16#include <linux/init.h>
15a67dd8 17#include <linux/kernel.h>
1da177e4 18#include <linux/acct.h>
16f7e0fe 19#include <linux/capability.h>
3d733633 20#include <linux/cpumask.h>
1da177e4 21#include <linux/module.h>
f20a9ead 22#include <linux/sysfs.h>
1da177e4 23#include <linux/seq_file.h>
6b3286ed 24#include <linux/mnt_namespace.h>
1da177e4 25#include <linux/namei.h>
b43f3cbd 26#include <linux/nsproxy.h>
1da177e4
LT
27#include <linux/security.h>
28#include <linux/mount.h>
07f3f05c 29#include <linux/ramfs.h>
13f14b4d 30#include <linux/log2.h>
73cd49ec 31#include <linux/idr.h>
5ad4e53b 32#include <linux/fs_struct.h>
2504c5d6 33#include <linux/fsnotify.h>
1da177e4
LT
34#include <asm/uaccess.h>
35#include <asm/unistd.h>
07b20889 36#include "pnode.h"
948730b0 37#include "internal.h"
1da177e4 38
13f14b4d
ED
39#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
5addc5dd 42static int event;
73cd49ec 43static DEFINE_IDA(mnt_id_ida);
719f5d7f 44static DEFINE_IDA(mnt_group_ida);
99b7db7b 45static DEFINE_SPINLOCK(mnt_id_lock);
f21f6220
AV
46static int mnt_id_start = 0;
47static int mnt_group_start = 1;
1da177e4 48
fa3536cc 49static struct list_head *mount_hashtable __read_mostly;
e18b890b 50static struct kmem_cache *mnt_cache __read_mostly;
390c6843 51static struct rw_semaphore namespace_sem;
1da177e4 52
f87fd4c2 53/* /sys/fs */
00d26666
GKH
54struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
f87fd4c2 56
99b7db7b
NP
57/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
1da177e4
LT
67static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
b58fed8b
RP
69 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
13f14b4d
ED
71 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
1da177e4
LT
73}
74
3d733633
DH
75#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
99b7db7b
NP
77/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
73cd49ec
MS
81static int mnt_alloc_id(struct vfsmount *mnt)
82{
83 int res;
84
85retry:
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
99b7db7b 87 spin_lock(&mnt_id_lock);
f21f6220
AV
88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
89 if (!res)
90 mnt_id_start = mnt->mnt_id + 1;
99b7db7b 91 spin_unlock(&mnt_id_lock);
73cd49ec
MS
92 if (res == -EAGAIN)
93 goto retry;
94
95 return res;
96}
97
98static void mnt_free_id(struct vfsmount *mnt)
99{
f21f6220 100 int id = mnt->mnt_id;
99b7db7b 101 spin_lock(&mnt_id_lock);
f21f6220
AV
102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
104 mnt_id_start = id;
99b7db7b 105 spin_unlock(&mnt_id_lock);
73cd49ec
MS
106}
107
719f5d7f
MS
108/*
109 * Allocate a new peer group ID
110 *
111 * mnt_group_ida is protected by namespace_sem
112 */
113static int mnt_alloc_group_id(struct vfsmount *mnt)
114{
f21f6220
AV
115 int res;
116
719f5d7f
MS
117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118 return -ENOMEM;
119
f21f6220
AV
120 res = ida_get_new_above(&mnt_group_ida,
121 mnt_group_start,
122 &mnt->mnt_group_id);
123 if (!res)
124 mnt_group_start = mnt->mnt_group_id + 1;
125
126 return res;
719f5d7f
MS
127}
128
129/*
130 * Release a peer group ID
131 */
132void mnt_release_group_id(struct vfsmount *mnt)
133{
f21f6220
AV
134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
719f5d7f
MS
138 mnt->mnt_group_id = 0;
139}
140
b3e19d92
NP
141/*
142 * vfsmount lock must be held for read
143 */
144static inline void mnt_add_count(struct vfsmount *mnt, int n)
145{
146#ifdef CONFIG_SMP
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
148#else
149 preempt_disable();
150 mnt->mnt_count += n;
151 preempt_enable();
152#endif
153}
154
155static inline void mnt_set_count(struct vfsmount *mnt, int n)
156{
157#ifdef CONFIG_SMP
158 this_cpu_write(mnt->mnt_pcp->mnt_count, n);
159#else
160 mnt->mnt_count = n;
161#endif
162}
163
164/*
165 * vfsmount lock must be held for read
166 */
167static inline void mnt_inc_count(struct vfsmount *mnt)
168{
169 mnt_add_count(mnt, 1);
170}
171
172/*
173 * vfsmount lock must be held for read
174 */
175static inline void mnt_dec_count(struct vfsmount *mnt)
176{
177 mnt_add_count(mnt, -1);
178}
179
180/*
181 * vfsmount lock must be held for write
182 */
183unsigned int mnt_get_count(struct vfsmount *mnt)
184{
185#ifdef CONFIG_SMP
186 unsigned int count = atomic_read(&mnt->mnt_longrefs);
187 int cpu;
188
189 for_each_possible_cpu(cpu) {
190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191 }
192
193 return count;
194#else
195 return mnt->mnt_count;
196#endif
197}
198
1da177e4
LT
199struct vfsmount *alloc_vfsmnt(const char *name)
200{
c3762229 201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
1da177e4 202 if (mnt) {
73cd49ec
MS
203 int err;
204
205 err = mnt_alloc_id(mnt);
88b38782
LZ
206 if (err)
207 goto out_free_cache;
208
209 if (name) {
210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211 if (!mnt->mnt_devname)
212 goto out_free_id;
73cd49ec
MS
213 }
214
b3e19d92
NP
215#ifdef CONFIG_SMP
216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 if (!mnt->mnt_pcp)
218 goto out_free_devname;
219
220 atomic_set(&mnt->mnt_longrefs, 1);
221#else
222 mnt->mnt_count = 1;
223 mnt->mnt_writers = 0;
224#endif
225
1da177e4
LT
226 INIT_LIST_HEAD(&mnt->mnt_hash);
227 INIT_LIST_HEAD(&mnt->mnt_child);
228 INIT_LIST_HEAD(&mnt->mnt_mounts);
229 INIT_LIST_HEAD(&mnt->mnt_list);
55e700b9 230 INIT_LIST_HEAD(&mnt->mnt_expire);
03e06e68 231 INIT_LIST_HEAD(&mnt->mnt_share);
a58b0eb8
RP
232 INIT_LIST_HEAD(&mnt->mnt_slave_list);
233 INIT_LIST_HEAD(&mnt->mnt_slave);
2504c5d6
AG
234#ifdef CONFIG_FSNOTIFY
235 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
d3ef3d73 236#endif
1da177e4
LT
237 }
238 return mnt;
88b38782 239
d3ef3d73
NP
240#ifdef CONFIG_SMP
241out_free_devname:
242 kfree(mnt->mnt_devname);
243#endif
88b38782
LZ
244out_free_id:
245 mnt_free_id(mnt);
246out_free_cache:
247 kmem_cache_free(mnt_cache, mnt);
248 return NULL;
1da177e4
LT
249}
250
3d733633
DH
251/*
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
257 * a filesystem.
258 */
259/*
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
262 *
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
268 * r/w.
269 */
270int __mnt_is_readonly(struct vfsmount *mnt)
271{
2e4b7fcd
DH
272 if (mnt->mnt_flags & MNT_READONLY)
273 return 1;
274 if (mnt->mnt_sb->s_flags & MS_RDONLY)
275 return 1;
276 return 0;
3d733633
DH
277}
278EXPORT_SYMBOL_GPL(__mnt_is_readonly);
279
c6653a83 280static inline void mnt_inc_writers(struct vfsmount *mnt)
d3ef3d73
NP
281{
282#ifdef CONFIG_SMP
b3e19d92 283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
d3ef3d73
NP
284#else
285 mnt->mnt_writers++;
286#endif
287}
3d733633 288
c6653a83 289static inline void mnt_dec_writers(struct vfsmount *mnt)
3d733633 290{
d3ef3d73 291#ifdef CONFIG_SMP
b3e19d92 292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
d3ef3d73
NP
293#else
294 mnt->mnt_writers--;
295#endif
3d733633 296}
3d733633 297
c6653a83 298static unsigned int mnt_get_writers(struct vfsmount *mnt)
3d733633 299{
d3ef3d73
NP
300#ifdef CONFIG_SMP
301 unsigned int count = 0;
3d733633 302 int cpu;
3d733633
DH
303
304 for_each_possible_cpu(cpu) {
b3e19d92 305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
3d733633 306 }
3d733633 307
d3ef3d73
NP
308 return count;
309#else
310 return mnt->mnt_writers;
311#endif
3d733633
DH
312}
313
8366025e
DH
314/*
315 * Most r/o checks on a fs are for operations that take
316 * discrete amounts of time, like a write() or unlink().
317 * We must keep track of when those operations start
318 * (for permission checks) and when they end, so that
319 * we can determine when writes are able to occur to
320 * a filesystem.
321 */
322/**
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
325 *
326 * This tells the low-level filesystem that a write is
327 * about to be performed to it, and makes sure that
328 * writes are allowed before returning success. When
329 * the write operation is finished, mnt_drop_write()
330 * must be called. This is effectively a refcount.
331 */
332int mnt_want_write(struct vfsmount *mnt)
333{
3d733633 334 int ret = 0;
3d733633 335
d3ef3d73 336 preempt_disable();
c6653a83 337 mnt_inc_writers(mnt);
d3ef3d73 338 /*
c6653a83 339 * The store to mnt_inc_writers must be visible before we pass
d3ef3d73
NP
340 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
341 * incremented count after it has set MNT_WRITE_HOLD.
342 */
343 smp_mb();
344 while (mnt->mnt_flags & MNT_WRITE_HOLD)
345 cpu_relax();
346 /*
347 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
348 * be set to match its requirements. So we must not load that until
349 * MNT_WRITE_HOLD is cleared.
350 */
351 smp_rmb();
3d733633 352 if (__mnt_is_readonly(mnt)) {
c6653a83 353 mnt_dec_writers(mnt);
3d733633
DH
354 ret = -EROFS;
355 goto out;
356 }
3d733633 357out:
d3ef3d73 358 preempt_enable();
3d733633 359 return ret;
8366025e
DH
360}
361EXPORT_SYMBOL_GPL(mnt_want_write);
362
96029c4e
NP
363/**
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
366 *
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
371 *
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
374 */
375int mnt_clone_write(struct vfsmount *mnt)
376{
377 /* superblock may be r/o */
378 if (__mnt_is_readonly(mnt))
379 return -EROFS;
380 preempt_disable();
c6653a83 381 mnt_inc_writers(mnt);
96029c4e
NP
382 preempt_enable();
383 return 0;
384}
385EXPORT_SYMBOL_GPL(mnt_clone_write);
386
387/**
388 * mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
390 *
391 * This is like mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
393 */
394int mnt_want_write_file(struct file *file)
395{
2d8dd38a
OH
396 struct inode *inode = file->f_dentry->d_inode;
397 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
96029c4e
NP
398 return mnt_want_write(file->f_path.mnt);
399 else
400 return mnt_clone_write(file->f_path.mnt);
401}
402EXPORT_SYMBOL_GPL(mnt_want_write_file);
403
8366025e
DH
404/**
405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
407 *
408 * Tells the low-level filesystem that we are done
409 * performing writes to it. Must be matched with
410 * mnt_want_write() call above.
411 */
412void mnt_drop_write(struct vfsmount *mnt)
413{
d3ef3d73 414 preempt_disable();
c6653a83 415 mnt_dec_writers(mnt);
d3ef3d73 416 preempt_enable();
8366025e
DH
417}
418EXPORT_SYMBOL_GPL(mnt_drop_write);
419
2e4b7fcd 420static int mnt_make_readonly(struct vfsmount *mnt)
8366025e 421{
3d733633
DH
422 int ret = 0;
423
99b7db7b 424 br_write_lock(vfsmount_lock);
d3ef3d73 425 mnt->mnt_flags |= MNT_WRITE_HOLD;
3d733633 426 /*
d3ef3d73
NP
427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428 * should be visible before we do.
3d733633 429 */
d3ef3d73
NP
430 smp_mb();
431
3d733633 432 /*
d3ef3d73
NP
433 * With writers on hold, if this value is zero, then there are
434 * definitely no active writers (although held writers may subsequently
435 * increment the count, they'll have to wait, and decrement it after
436 * seeing MNT_READONLY).
437 *
438 * It is OK to have counter incremented on one CPU and decremented on
439 * another: the sum will add up correctly. The danger would be when we
440 * sum up each counter, if we read a counter before it is incremented,
441 * but then read another CPU's count which it has been subsequently
442 * decremented from -- we would see more decrements than we should.
443 * MNT_WRITE_HOLD protects against this scenario, because
444 * mnt_want_write first increments count, then smp_mb, then spins on
445 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
446 * we're counting up here.
3d733633 447 */
c6653a83 448 if (mnt_get_writers(mnt) > 0)
d3ef3d73
NP
449 ret = -EBUSY;
450 else
2e4b7fcd 451 mnt->mnt_flags |= MNT_READONLY;
d3ef3d73
NP
452 /*
453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454 * that become unheld will see MNT_READONLY.
455 */
456 smp_wmb();
457 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
99b7db7b 458 br_write_unlock(vfsmount_lock);
3d733633 459 return ret;
8366025e 460}
8366025e 461
2e4b7fcd
DH
462static void __mnt_unmake_readonly(struct vfsmount *mnt)
463{
99b7db7b 464 br_write_lock(vfsmount_lock);
2e4b7fcd 465 mnt->mnt_flags &= ~MNT_READONLY;
99b7db7b 466 br_write_unlock(vfsmount_lock);
2e4b7fcd
DH
467}
468
a3ec947c 469void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
454e2398
DH
470{
471 mnt->mnt_sb = sb;
472 mnt->mnt_root = dget(sb->s_root);
454e2398
DH
473}
474
475EXPORT_SYMBOL(simple_set_mnt);
476
1da177e4
LT
477void free_vfsmnt(struct vfsmount *mnt)
478{
479 kfree(mnt->mnt_devname);
73cd49ec 480 mnt_free_id(mnt);
d3ef3d73 481#ifdef CONFIG_SMP
b3e19d92 482 free_percpu(mnt->mnt_pcp);
d3ef3d73 483#endif
1da177e4
LT
484 kmem_cache_free(mnt_cache, mnt);
485}
486
487/*
a05964f3
RP
488 * find the first or last mount at @dentry on vfsmount @mnt depending on
489 * @dir. If @dir is set return the first mount else return the last mount.
99b7db7b 490 * vfsmount_lock must be held for read or write.
1da177e4 491 */
a05964f3
RP
492struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
493 int dir)
1da177e4 494{
b58fed8b
RP
495 struct list_head *head = mount_hashtable + hash(mnt, dentry);
496 struct list_head *tmp = head;
1da177e4
LT
497 struct vfsmount *p, *found = NULL;
498
1da177e4 499 for (;;) {
a05964f3 500 tmp = dir ? tmp->next : tmp->prev;
1da177e4
LT
501 p = NULL;
502 if (tmp == head)
503 break;
504 p = list_entry(tmp, struct vfsmount, mnt_hash);
505 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
a05964f3 506 found = p;
1da177e4
LT
507 break;
508 }
509 }
1da177e4
LT
510 return found;
511}
512
a05964f3
RP
513/*
514 * lookup_mnt increments the ref count before returning
515 * the vfsmount struct.
516 */
1c755af4 517struct vfsmount *lookup_mnt(struct path *path)
a05964f3
RP
518{
519 struct vfsmount *child_mnt;
99b7db7b
NP
520
521 br_read_lock(vfsmount_lock);
1c755af4 522 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
a05964f3 523 mntget(child_mnt);
99b7db7b 524 br_read_unlock(vfsmount_lock);
a05964f3
RP
525 return child_mnt;
526}
527
1da177e4
LT
528static inline int check_mnt(struct vfsmount *mnt)
529{
6b3286ed 530 return mnt->mnt_ns == current->nsproxy->mnt_ns;
1da177e4
LT
531}
532
99b7db7b
NP
533/*
534 * vfsmount lock must be held for write
535 */
6b3286ed 536static void touch_mnt_namespace(struct mnt_namespace *ns)
5addc5dd
AV
537{
538 if (ns) {
539 ns->event = ++event;
540 wake_up_interruptible(&ns->poll);
541 }
542}
543
99b7db7b
NP
544/*
545 * vfsmount lock must be held for write
546 */
6b3286ed 547static void __touch_mnt_namespace(struct mnt_namespace *ns)
5addc5dd
AV
548{
549 if (ns && ns->event != event) {
550 ns->event = event;
551 wake_up_interruptible(&ns->poll);
552 }
553}
554
5f57cbcc
NP
555/*
556 * Clear dentry's mounted state if it has no remaining mounts.
557 * vfsmount_lock must be held for write.
558 */
559static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
560{
561 unsigned u;
562
563 for (u = 0; u < HASH_SIZE; u++) {
564 struct vfsmount *p;
565
566 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
567 if (p->mnt_mountpoint == dentry)
568 return;
569 }
570 }
571 spin_lock(&dentry->d_lock);
572 dentry->d_flags &= ~DCACHE_MOUNTED;
573 spin_unlock(&dentry->d_lock);
574}
575
99b7db7b
NP
576/*
577 * vfsmount lock must be held for write
578 */
1a390689 579static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
1da177e4 580{
1a390689
AV
581 old_path->dentry = mnt->mnt_mountpoint;
582 old_path->mnt = mnt->mnt_parent;
1da177e4
LT
583 mnt->mnt_parent = mnt;
584 mnt->mnt_mountpoint = mnt->mnt_root;
585 list_del_init(&mnt->mnt_child);
586 list_del_init(&mnt->mnt_hash);
5f57cbcc 587 dentry_reset_mounted(old_path->mnt, old_path->dentry);
1da177e4
LT
588}
589
99b7db7b
NP
590/*
591 * vfsmount lock must be held for write
592 */
b90fa9ae
RP
593void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
594 struct vfsmount *child_mnt)
595{
596 child_mnt->mnt_parent = mntget(mnt);
597 child_mnt->mnt_mountpoint = dget(dentry);
5f57cbcc
NP
598 spin_lock(&dentry->d_lock);
599 dentry->d_flags |= DCACHE_MOUNTED;
600 spin_unlock(&dentry->d_lock);
b90fa9ae
RP
601}
602
99b7db7b
NP
603/*
604 * vfsmount lock must be held for write
605 */
1a390689 606static void attach_mnt(struct vfsmount *mnt, struct path *path)
1da177e4 607{
1a390689 608 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
b90fa9ae 609 list_add_tail(&mnt->mnt_hash, mount_hashtable +
1a390689
AV
610 hash(path->mnt, path->dentry));
611 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
b90fa9ae
RP
612}
613
614/*
99b7db7b 615 * vfsmount lock must be held for write
b90fa9ae
RP
616 */
617static void commit_tree(struct vfsmount *mnt)
618{
619 struct vfsmount *parent = mnt->mnt_parent;
620 struct vfsmount *m;
621 LIST_HEAD(head);
6b3286ed 622 struct mnt_namespace *n = parent->mnt_ns;
b90fa9ae
RP
623
624 BUG_ON(parent == mnt);
625
626 list_add_tail(&head, &mnt->mnt_list);
627 list_for_each_entry(m, &head, mnt_list)
6b3286ed 628 m->mnt_ns = n;
b90fa9ae
RP
629 list_splice(&head, n->list.prev);
630
631 list_add_tail(&mnt->mnt_hash, mount_hashtable +
632 hash(parent, mnt->mnt_mountpoint));
633 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
6b3286ed 634 touch_mnt_namespace(n);
1da177e4
LT
635}
636
637static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
638{
639 struct list_head *next = p->mnt_mounts.next;
640 if (next == &p->mnt_mounts) {
641 while (1) {
642 if (p == root)
643 return NULL;
644 next = p->mnt_child.next;
645 if (next != &p->mnt_parent->mnt_mounts)
646 break;
647 p = p->mnt_parent;
648 }
649 }
650 return list_entry(next, struct vfsmount, mnt_child);
651}
652
9676f0c6
RP
653static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
654{
655 struct list_head *prev = p->mnt_mounts.prev;
656 while (prev != &p->mnt_mounts) {
657 p = list_entry(prev, struct vfsmount, mnt_child);
658 prev = p->mnt_mounts.prev;
659 }
660 return p;
661}
662
36341f64
RP
663static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
664 int flag)
1da177e4
LT
665{
666 struct super_block *sb = old->mnt_sb;
667 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
668
669 if (mnt) {
719f5d7f
MS
670 if (flag & (CL_SLAVE | CL_PRIVATE))
671 mnt->mnt_group_id = 0; /* not a peer of original */
672 else
673 mnt->mnt_group_id = old->mnt_group_id;
674
675 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
676 int err = mnt_alloc_group_id(mnt);
677 if (err)
678 goto out_free;
679 }
680
be1a16a0 681 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
1da177e4
LT
682 atomic_inc(&sb->s_active);
683 mnt->mnt_sb = sb;
684 mnt->mnt_root = dget(root);
685 mnt->mnt_mountpoint = mnt->mnt_root;
686 mnt->mnt_parent = mnt;
b90fa9ae 687
5afe0022
RP
688 if (flag & CL_SLAVE) {
689 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
690 mnt->mnt_master = old;
691 CLEAR_MNT_SHARED(mnt);
8aec0809 692 } else if (!(flag & CL_PRIVATE)) {
796a6b52 693 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
5afe0022
RP
694 list_add(&mnt->mnt_share, &old->mnt_share);
695 if (IS_MNT_SLAVE(old))
696 list_add(&mnt->mnt_slave, &old->mnt_slave);
697 mnt->mnt_master = old->mnt_master;
698 }
b90fa9ae
RP
699 if (flag & CL_MAKE_SHARED)
700 set_mnt_shared(mnt);
1da177e4
LT
701
702 /* stick the duplicate mount on the same expiry list
703 * as the original if that was on one */
36341f64 704 if (flag & CL_EXPIRE) {
36341f64
RP
705 if (!list_empty(&old->mnt_expire))
706 list_add(&mnt->mnt_expire, &old->mnt_expire);
36341f64 707 }
1da177e4
LT
708 }
709 return mnt;
719f5d7f
MS
710
711 out_free:
712 free_vfsmnt(mnt);
713 return NULL;
1da177e4
LT
714}
715
b3e19d92 716static inline void mntfree(struct vfsmount *mnt)
1da177e4
LT
717{
718 struct super_block *sb = mnt->mnt_sb;
b3e19d92 719
3d733633
DH
720 /*
721 * This probably indicates that somebody messed
722 * up a mnt_want/drop_write() pair. If this
723 * happens, the filesystem was probably unable
724 * to make r/w->r/o transitions.
725 */
d3ef3d73 726 /*
b3e19d92
NP
727 * The locking used to deal with mnt_count decrement provides barriers,
728 * so mnt_get_writers() below is safe.
d3ef3d73 729 */
c6653a83 730 WARN_ON(mnt_get_writers(mnt));
ca9c726e 731 fsnotify_vfsmount_delete(mnt);
1da177e4
LT
732 dput(mnt->mnt_root);
733 free_vfsmnt(mnt);
734 deactivate_super(sb);
735}
736
b3e19d92
NP
737#ifdef CONFIG_SMP
738static inline void __mntput(struct vfsmount *mnt, int longrefs)
739{
740 if (!longrefs) {
741put_again:
742 br_read_lock(vfsmount_lock);
743 if (likely(atomic_read(&mnt->mnt_longrefs))) {
744 mnt_dec_count(mnt);
745 br_read_unlock(vfsmount_lock);
746 return;
747 }
748 br_read_unlock(vfsmount_lock);
749 } else {
750 BUG_ON(!atomic_read(&mnt->mnt_longrefs));
751 if (atomic_add_unless(&mnt->mnt_longrefs, -1, 1))
752 return;
753 }
754
99b7db7b 755 br_write_lock(vfsmount_lock);
b3e19d92
NP
756 if (!longrefs)
757 mnt_dec_count(mnt);
758 else
759 atomic_dec(&mnt->mnt_longrefs);
760 if (mnt_get_count(mnt)) {
99b7db7b
NP
761 br_write_unlock(vfsmount_lock);
762 return;
763 }
b3e19d92
NP
764 if (unlikely(mnt->mnt_pinned)) {
765 mnt_add_count(mnt, mnt->mnt_pinned + 1);
766 mnt->mnt_pinned = 0;
99b7db7b 767 br_write_unlock(vfsmount_lock);
b3e19d92
NP
768 acct_auto_close_mnt(mnt);
769 goto put_again;
770 }
771 br_write_unlock(vfsmount_lock);
772 mntfree(mnt);
773}
774#else
775static inline void __mntput(struct vfsmount *mnt, int longrefs)
776{
777put_again:
778 mnt_dec_count(mnt);
779 if (likely(mnt_get_count(mnt)))
99b7db7b 780 return;
b3e19d92
NP
781 br_write_lock(vfsmount_lock);
782 if (unlikely(mnt->mnt_pinned)) {
783 mnt_add_count(mnt, mnt->mnt_pinned + 1);
784 mnt->mnt_pinned = 0;
785 br_write_unlock(vfsmount_lock);
786 acct_auto_close_mnt(mnt);
787 goto put_again;
7b7b1ace 788 }
99b7db7b 789 br_write_unlock(vfsmount_lock);
b3e19d92
NP
790 mntfree(mnt);
791}
792#endif
793
794static void mntput_no_expire(struct vfsmount *mnt)
795{
796 __mntput(mnt, 0);
797}
798
799void mntput(struct vfsmount *mnt)
800{
801 if (mnt) {
802 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
803 if (unlikely(mnt->mnt_expiry_mark))
804 mnt->mnt_expiry_mark = 0;
805 __mntput(mnt, 0);
806 }
807}
808EXPORT_SYMBOL(mntput);
809
810struct vfsmount *mntget(struct vfsmount *mnt)
811{
812 if (mnt)
813 mnt_inc_count(mnt);
814 return mnt;
815}
816EXPORT_SYMBOL(mntget);
817
818void mntput_long(struct vfsmount *mnt)
819{
820#ifdef CONFIG_SMP
821 if (mnt) {
822 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
823 if (unlikely(mnt->mnt_expiry_mark))
824 mnt->mnt_expiry_mark = 0;
825 __mntput(mnt, 1);
826 }
827#else
828 mntput(mnt);
829#endif
7b7b1ace 830}
b3e19d92
NP
831EXPORT_SYMBOL(mntput_long);
832
833struct vfsmount *mntget_long(struct vfsmount *mnt)
834{
835#ifdef CONFIG_SMP
836 if (mnt)
837 atomic_inc(&mnt->mnt_longrefs);
838 return mnt;
839#else
840 return mntget(mnt);
841#endif
842}
843EXPORT_SYMBOL(mntget_long);
7b7b1ace
AV
844
845void mnt_pin(struct vfsmount *mnt)
846{
99b7db7b 847 br_write_lock(vfsmount_lock);
7b7b1ace 848 mnt->mnt_pinned++;
99b7db7b 849 br_write_unlock(vfsmount_lock);
7b7b1ace 850}
7b7b1ace
AV
851EXPORT_SYMBOL(mnt_pin);
852
853void mnt_unpin(struct vfsmount *mnt)
854{
99b7db7b 855 br_write_lock(vfsmount_lock);
7b7b1ace 856 if (mnt->mnt_pinned) {
b3e19d92 857 mnt_inc_count(mnt);
7b7b1ace
AV
858 mnt->mnt_pinned--;
859 }
99b7db7b 860 br_write_unlock(vfsmount_lock);
7b7b1ace 861}
7b7b1ace 862EXPORT_SYMBOL(mnt_unpin);
1da177e4 863
b3b304a2
MS
864static inline void mangle(struct seq_file *m, const char *s)
865{
866 seq_escape(m, s, " \t\n\\");
867}
868
869/*
870 * Simple .show_options callback for filesystems which don't want to
871 * implement more complex mount option showing.
872 *
873 * See also save_mount_options().
874 */
875int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
876{
2a32cebd
AV
877 const char *options;
878
879 rcu_read_lock();
880 options = rcu_dereference(mnt->mnt_sb->s_options);
b3b304a2
MS
881
882 if (options != NULL && options[0]) {
883 seq_putc(m, ',');
884 mangle(m, options);
885 }
2a32cebd 886 rcu_read_unlock();
b3b304a2
MS
887
888 return 0;
889}
890EXPORT_SYMBOL(generic_show_options);
891
892/*
893 * If filesystem uses generic_show_options(), this function should be
894 * called from the fill_super() callback.
895 *
896 * The .remount_fs callback usually needs to be handled in a special
897 * way, to make sure, that previous options are not overwritten if the
898 * remount fails.
899 *
900 * Also note, that if the filesystem's .remount_fs function doesn't
901 * reset all options to their default value, but changes only newly
902 * given options, then the displayed options will not reflect reality
903 * any more.
904 */
905void save_mount_options(struct super_block *sb, char *options)
906{
2a32cebd
AV
907 BUG_ON(sb->s_options);
908 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
b3b304a2
MS
909}
910EXPORT_SYMBOL(save_mount_options);
911
2a32cebd
AV
912void replace_mount_options(struct super_block *sb, char *options)
913{
914 char *old = sb->s_options;
915 rcu_assign_pointer(sb->s_options, options);
916 if (old) {
917 synchronize_rcu();
918 kfree(old);
919 }
920}
921EXPORT_SYMBOL(replace_mount_options);
922
a1a2c409 923#ifdef CONFIG_PROC_FS
1da177e4
LT
924/* iterator */
925static void *m_start(struct seq_file *m, loff_t *pos)
926{
a1a2c409 927 struct proc_mounts *p = m->private;
1da177e4 928
390c6843 929 down_read(&namespace_sem);
a1a2c409 930 return seq_list_start(&p->ns->list, *pos);
1da177e4
LT
931}
932
933static void *m_next(struct seq_file *m, void *v, loff_t *pos)
934{
a1a2c409 935 struct proc_mounts *p = m->private;
b0765fb8 936
a1a2c409 937 return seq_list_next(v, &p->ns->list, pos);
1da177e4
LT
938}
939
940static void m_stop(struct seq_file *m, void *v)
941{
390c6843 942 up_read(&namespace_sem);
1da177e4
LT
943}
944
9f5596af
AV
945int mnt_had_events(struct proc_mounts *p)
946{
947 struct mnt_namespace *ns = p->ns;
948 int res = 0;
949
99b7db7b 950 br_read_lock(vfsmount_lock);
9f5596af
AV
951 if (p->event != ns->event) {
952 p->event = ns->event;
953 res = 1;
954 }
99b7db7b 955 br_read_unlock(vfsmount_lock);
9f5596af
AV
956
957 return res;
958}
959
2d4d4864
RP
960struct proc_fs_info {
961 int flag;
962 const char *str;
963};
964
2069f457 965static int show_sb_opts(struct seq_file *m, struct super_block *sb)
1da177e4 966{
2d4d4864 967 static const struct proc_fs_info fs_info[] = {
1da177e4
LT
968 { MS_SYNCHRONOUS, ",sync" },
969 { MS_DIRSYNC, ",dirsync" },
970 { MS_MANDLOCK, ",mand" },
1da177e4
LT
971 { 0, NULL }
972 };
2d4d4864
RP
973 const struct proc_fs_info *fs_infop;
974
975 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
976 if (sb->s_flags & fs_infop->flag)
977 seq_puts(m, fs_infop->str);
978 }
2069f457
EP
979
980 return security_sb_show_options(m, sb);
2d4d4864
RP
981}
982
983static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
984{
985 static const struct proc_fs_info mnt_info[] = {
1da177e4
LT
986 { MNT_NOSUID, ",nosuid" },
987 { MNT_NODEV, ",nodev" },
988 { MNT_NOEXEC, ",noexec" },
fc33a7bb
CH
989 { MNT_NOATIME, ",noatime" },
990 { MNT_NODIRATIME, ",nodiratime" },
47ae32d6 991 { MNT_RELATIME, ",relatime" },
1da177e4
LT
992 { 0, NULL }
993 };
2d4d4864
RP
994 const struct proc_fs_info *fs_infop;
995
996 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
997 if (mnt->mnt_flags & fs_infop->flag)
998 seq_puts(m, fs_infop->str);
999 }
1000}
1001
1002static void show_type(struct seq_file *m, struct super_block *sb)
1003{
1004 mangle(m, sb->s_type->name);
1005 if (sb->s_subtype && sb->s_subtype[0]) {
1006 seq_putc(m, '.');
1007 mangle(m, sb->s_subtype);
1008 }
1009}
1010
1011static int show_vfsmnt(struct seq_file *m, void *v)
1012{
1013 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1014 int err = 0;
c32c2f63 1015 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1da177e4
LT
1016
1017 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1018 seq_putc(m, ' ');
c32c2f63 1019 seq_path(m, &mnt_path, " \t\n\\");
1da177e4 1020 seq_putc(m, ' ');
2d4d4864 1021 show_type(m, mnt->mnt_sb);
2e4b7fcd 1022 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
2069f457
EP
1023 err = show_sb_opts(m, mnt->mnt_sb);
1024 if (err)
1025 goto out;
2d4d4864 1026 show_mnt_opts(m, mnt);
1da177e4
LT
1027 if (mnt->mnt_sb->s_op->show_options)
1028 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1029 seq_puts(m, " 0 0\n");
2069f457 1030out:
1da177e4
LT
1031 return err;
1032}
1033
a1a2c409 1034const struct seq_operations mounts_op = {
1da177e4
LT
1035 .start = m_start,
1036 .next = m_next,
1037 .stop = m_stop,
1038 .show = show_vfsmnt
1039};
1040
2d4d4864
RP
1041static int show_mountinfo(struct seq_file *m, void *v)
1042{
1043 struct proc_mounts *p = m->private;
1044 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1045 struct super_block *sb = mnt->mnt_sb;
1046 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1047 struct path root = p->root;
1048 int err = 0;
1049
1050 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1051 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1052 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1053 seq_putc(m, ' ');
1054 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1055 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1056 /*
1057 * Mountpoint is outside root, discard that one. Ugly,
1058 * but less so than trying to do that in iterator in a
1059 * race-free way (due to renames).
1060 */
1061 return SEQ_SKIP;
1062 }
1063 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1064 show_mnt_opts(m, mnt);
1065
1066 /* Tagged fields ("foo:X" or "bar") */
1067 if (IS_MNT_SHARED(mnt))
1068 seq_printf(m, " shared:%i", mnt->mnt_group_id);
97e7e0f7
MS
1069 if (IS_MNT_SLAVE(mnt)) {
1070 int master = mnt->mnt_master->mnt_group_id;
1071 int dom = get_dominating_id(mnt, &p->root);
1072 seq_printf(m, " master:%i", master);
1073 if (dom && dom != master)
1074 seq_printf(m, " propagate_from:%i", dom);
1075 }
2d4d4864
RP
1076 if (IS_MNT_UNBINDABLE(mnt))
1077 seq_puts(m, " unbindable");
1078
1079 /* Filesystem specific data */
1080 seq_puts(m, " - ");
1081 show_type(m, sb);
1082 seq_putc(m, ' ');
1083 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
2069f457
EP
1085 err = show_sb_opts(m, sb);
1086 if (err)
1087 goto out;
2d4d4864
RP
1088 if (sb->s_op->show_options)
1089 err = sb->s_op->show_options(m, mnt);
1090 seq_putc(m, '\n');
2069f457 1091out:
2d4d4864
RP
1092 return err;
1093}
1094
1095const struct seq_operations mountinfo_op = {
1096 .start = m_start,
1097 .next = m_next,
1098 .stop = m_stop,
1099 .show = show_mountinfo,
1100};
1101
b4629fe2
CL
1102static int show_vfsstat(struct seq_file *m, void *v)
1103{
b0765fb8 1104 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
c32c2f63 1105 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
b4629fe2
CL
1106 int err = 0;
1107
1108 /* device */
1109 if (mnt->mnt_devname) {
1110 seq_puts(m, "device ");
1111 mangle(m, mnt->mnt_devname);
1112 } else
1113 seq_puts(m, "no device");
1114
1115 /* mount point */
1116 seq_puts(m, " mounted on ");
c32c2f63 1117 seq_path(m, &mnt_path, " \t\n\\");
b4629fe2
CL
1118 seq_putc(m, ' ');
1119
1120 /* file system type */
1121 seq_puts(m, "with fstype ");
2d4d4864 1122 show_type(m, mnt->mnt_sb);
b4629fe2
CL
1123
1124 /* optional statistics */
1125 if (mnt->mnt_sb->s_op->show_stats) {
1126 seq_putc(m, ' ');
1127 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1128 }
1129
1130 seq_putc(m, '\n');
1131 return err;
1132}
1133
a1a2c409 1134const struct seq_operations mountstats_op = {
b4629fe2
CL
1135 .start = m_start,
1136 .next = m_next,
1137 .stop = m_stop,
1138 .show = show_vfsstat,
1139};
a1a2c409 1140#endif /* CONFIG_PROC_FS */
b4629fe2 1141
1da177e4
LT
1142/**
1143 * may_umount_tree - check if a mount tree is busy
1144 * @mnt: root of mount tree
1145 *
1146 * This is called to check if a tree of mounts has any
1147 * open files, pwds, chroots or sub mounts that are
1148 * busy.
1149 */
1150int may_umount_tree(struct vfsmount *mnt)
1151{
36341f64
RP
1152 int actual_refs = 0;
1153 int minimum_refs = 0;
1154 struct vfsmount *p;
1da177e4 1155
b3e19d92
NP
1156 /* write lock needed for mnt_get_count */
1157 br_write_lock(vfsmount_lock);
36341f64 1158 for (p = mnt; p; p = next_mnt(p, mnt)) {
b3e19d92 1159 actual_refs += mnt_get_count(p);
1da177e4 1160 minimum_refs += 2;
1da177e4 1161 }
b3e19d92 1162 br_write_unlock(vfsmount_lock);
1da177e4
LT
1163
1164 if (actual_refs > minimum_refs)
e3474a8e 1165 return 0;
1da177e4 1166
e3474a8e 1167 return 1;
1da177e4
LT
1168}
1169
1170EXPORT_SYMBOL(may_umount_tree);
1171
1172/**
1173 * may_umount - check if a mount point is busy
1174 * @mnt: root of mount
1175 *
1176 * This is called to check if a mount point has any
1177 * open files, pwds, chroots or sub mounts. If the
1178 * mount has sub mounts this will return busy
1179 * regardless of whether the sub mounts are busy.
1180 *
1181 * Doesn't take quota and stuff into account. IOW, in some cases it will
1182 * give false negatives. The main reason why it's here is that we need
1183 * a non-destructive way to look for easily umountable filesystems.
1184 */
1185int may_umount(struct vfsmount *mnt)
1186{
e3474a8e 1187 int ret = 1;
8ad08d8a 1188 down_read(&namespace_sem);
b3e19d92 1189 br_write_lock(vfsmount_lock);
a05964f3 1190 if (propagate_mount_busy(mnt, 2))
e3474a8e 1191 ret = 0;
b3e19d92 1192 br_write_unlock(vfsmount_lock);
8ad08d8a 1193 up_read(&namespace_sem);
a05964f3 1194 return ret;
1da177e4
LT
1195}
1196
1197EXPORT_SYMBOL(may_umount);
1198
b90fa9ae 1199void release_mounts(struct list_head *head)
70fbcdf4
RP
1200{
1201 struct vfsmount *mnt;
bf066c7d 1202 while (!list_empty(head)) {
b5e61818 1203 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
70fbcdf4
RP
1204 list_del_init(&mnt->mnt_hash);
1205 if (mnt->mnt_parent != mnt) {
1206 struct dentry *dentry;
1207 struct vfsmount *m;
99b7db7b
NP
1208
1209 br_write_lock(vfsmount_lock);
70fbcdf4
RP
1210 dentry = mnt->mnt_mountpoint;
1211 m = mnt->mnt_parent;
1212 mnt->mnt_mountpoint = mnt->mnt_root;
1213 mnt->mnt_parent = mnt;
7c4b93d8 1214 m->mnt_ghosts--;
99b7db7b 1215 br_write_unlock(vfsmount_lock);
70fbcdf4
RP
1216 dput(dentry);
1217 mntput(m);
1218 }
b3e19d92 1219 mntput_long(mnt);
70fbcdf4
RP
1220 }
1221}
1222
99b7db7b
NP
1223/*
1224 * vfsmount lock must be held for write
1225 * namespace_sem must be held for write
1226 */
a05964f3 1227void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1da177e4
LT
1228{
1229 struct vfsmount *p;
1da177e4 1230
1bfba4e8
AM
1231 for (p = mnt; p; p = next_mnt(p, mnt))
1232 list_move(&p->mnt_hash, kill);
1da177e4 1233
a05964f3
RP
1234 if (propagate)
1235 propagate_umount(kill);
1236
70fbcdf4
RP
1237 list_for_each_entry(p, kill, mnt_hash) {
1238 list_del_init(&p->mnt_expire);
1239 list_del_init(&p->mnt_list);
6b3286ed
KK
1240 __touch_mnt_namespace(p->mnt_ns);
1241 p->mnt_ns = NULL;
70fbcdf4 1242 list_del_init(&p->mnt_child);
7c4b93d8
AV
1243 if (p->mnt_parent != p) {
1244 p->mnt_parent->mnt_ghosts++;
5f57cbcc 1245 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
7c4b93d8 1246 }
a05964f3 1247 change_mnt_propagation(p, MS_PRIVATE);
1da177e4
LT
1248 }
1249}
1250
c35038be
AV
1251static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1252
1da177e4
LT
1253static int do_umount(struct vfsmount *mnt, int flags)
1254{
b58fed8b 1255 struct super_block *sb = mnt->mnt_sb;
1da177e4 1256 int retval;
70fbcdf4 1257 LIST_HEAD(umount_list);
1da177e4
LT
1258
1259 retval = security_sb_umount(mnt, flags);
1260 if (retval)
1261 return retval;
1262
1263 /*
1264 * Allow userspace to request a mountpoint be expired rather than
1265 * unmounting unconditionally. Unmount only happens if:
1266 * (1) the mark is already set (the mark is cleared by mntput())
1267 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1268 */
1269 if (flags & MNT_EXPIRE) {
6ac08c39 1270 if (mnt == current->fs->root.mnt ||
1da177e4
LT
1271 flags & (MNT_FORCE | MNT_DETACH))
1272 return -EINVAL;
1273
b3e19d92
NP
1274 /*
1275 * probably don't strictly need the lock here if we examined
1276 * all race cases, but it's a slowpath.
1277 */
1278 br_write_lock(vfsmount_lock);
1279 if (mnt_get_count(mnt) != 2) {
1280 br_write_lock(vfsmount_lock);
1da177e4 1281 return -EBUSY;
b3e19d92
NP
1282 }
1283 br_write_unlock(vfsmount_lock);
1da177e4
LT
1284
1285 if (!xchg(&mnt->mnt_expiry_mark, 1))
1286 return -EAGAIN;
1287 }
1288
1289 /*
1290 * If we may have to abort operations to get out of this
1291 * mount, and they will themselves hold resources we must
1292 * allow the fs to do things. In the Unix tradition of
1293 * 'Gee thats tricky lets do it in userspace' the umount_begin
1294 * might fail to complete on the first run through as other tasks
1295 * must return, and the like. Thats for the mount program to worry
1296 * about for the moment.
1297 */
1298
42faad99 1299 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
42faad99 1300 sb->s_op->umount_begin(sb);
42faad99 1301 }
1da177e4
LT
1302
1303 /*
1304 * No sense to grab the lock for this test, but test itself looks
1305 * somewhat bogus. Suggestions for better replacement?
1306 * Ho-hum... In principle, we might treat that as umount + switch
1307 * to rootfs. GC would eventually take care of the old vfsmount.
1308 * Actually it makes sense, especially if rootfs would contain a
1309 * /reboot - static binary that would close all descriptors and
1310 * call reboot(9). Then init(8) could umount root and exec /reboot.
1311 */
6ac08c39 1312 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1da177e4
LT
1313 /*
1314 * Special case for "unmounting" root ...
1315 * we just try to remount it readonly.
1316 */
1317 down_write(&sb->s_umount);
4aa98cf7 1318 if (!(sb->s_flags & MS_RDONLY))
1da177e4 1319 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1da177e4
LT
1320 up_write(&sb->s_umount);
1321 return retval;
1322 }
1323
390c6843 1324 down_write(&namespace_sem);
99b7db7b 1325 br_write_lock(vfsmount_lock);
5addc5dd 1326 event++;
1da177e4 1327
c35038be
AV
1328 if (!(flags & MNT_DETACH))
1329 shrink_submounts(mnt, &umount_list);
1330
1da177e4 1331 retval = -EBUSY;
a05964f3 1332 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1da177e4 1333 if (!list_empty(&mnt->mnt_list))
a05964f3 1334 umount_tree(mnt, 1, &umount_list);
1da177e4
LT
1335 retval = 0;
1336 }
99b7db7b 1337 br_write_unlock(vfsmount_lock);
390c6843 1338 up_write(&namespace_sem);
70fbcdf4 1339 release_mounts(&umount_list);
1da177e4
LT
1340 return retval;
1341}
1342
1343/*
1344 * Now umount can handle mount points as well as block devices.
1345 * This is important for filesystems which use unnamed block devices.
1346 *
1347 * We now support a flag for forced unmount like the other 'big iron'
1348 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1349 */
1350
bdc480e3 1351SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1da177e4 1352{
2d8f3038 1353 struct path path;
1da177e4 1354 int retval;
db1f05bb 1355 int lookup_flags = 0;
1da177e4 1356
db1f05bb
MS
1357 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1358 return -EINVAL;
1359
1360 if (!(flags & UMOUNT_NOFOLLOW))
1361 lookup_flags |= LOOKUP_FOLLOW;
1362
1363 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1da177e4
LT
1364 if (retval)
1365 goto out;
1366 retval = -EINVAL;
2d8f3038 1367 if (path.dentry != path.mnt->mnt_root)
1da177e4 1368 goto dput_and_out;
2d8f3038 1369 if (!check_mnt(path.mnt))
1da177e4
LT
1370 goto dput_and_out;
1371
1372 retval = -EPERM;
1373 if (!capable(CAP_SYS_ADMIN))
1374 goto dput_and_out;
1375
2d8f3038 1376 retval = do_umount(path.mnt, flags);
1da177e4 1377dput_and_out:
429731b1 1378 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
2d8f3038
AV
1379 dput(path.dentry);
1380 mntput_no_expire(path.mnt);
1da177e4
LT
1381out:
1382 return retval;
1383}
1384
1385#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1386
1387/*
b58fed8b 1388 * The 2.0 compatible umount. No flags.
1da177e4 1389 */
bdc480e3 1390SYSCALL_DEFINE1(oldumount, char __user *, name)
1da177e4 1391{
b58fed8b 1392 return sys_umount(name, 0);
1da177e4
LT
1393}
1394
1395#endif
1396
2d92ab3c 1397static int mount_is_safe(struct path *path)
1da177e4
LT
1398{
1399 if (capable(CAP_SYS_ADMIN))
1400 return 0;
1401 return -EPERM;
1402#ifdef notyet
2d92ab3c 1403 if (S_ISLNK(path->dentry->d_inode->i_mode))
1da177e4 1404 return -EPERM;
2d92ab3c 1405 if (path->dentry->d_inode->i_mode & S_ISVTX) {
da9592ed 1406 if (current_uid() != path->dentry->d_inode->i_uid)
1da177e4
LT
1407 return -EPERM;
1408 }
2d92ab3c 1409 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1da177e4
LT
1410 return -EPERM;
1411 return 0;
1412#endif
1413}
1414
b90fa9ae 1415struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
36341f64 1416 int flag)
1da177e4
LT
1417{
1418 struct vfsmount *res, *p, *q, *r, *s;
1a390689 1419 struct path path;
1da177e4 1420
9676f0c6
RP
1421 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1422 return NULL;
1423
36341f64 1424 res = q = clone_mnt(mnt, dentry, flag);
1da177e4
LT
1425 if (!q)
1426 goto Enomem;
1427 q->mnt_mountpoint = mnt->mnt_mountpoint;
1428
1429 p = mnt;
fdadd65f 1430 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
7ec02ef1 1431 if (!is_subdir(r->mnt_mountpoint, dentry))
1da177e4
LT
1432 continue;
1433
1434 for (s = r; s; s = next_mnt(s, r)) {
9676f0c6
RP
1435 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1436 s = skip_mnt_tree(s);
1437 continue;
1438 }
1da177e4
LT
1439 while (p != s->mnt_parent) {
1440 p = p->mnt_parent;
1441 q = q->mnt_parent;
1442 }
1443 p = s;
1a390689
AV
1444 path.mnt = q;
1445 path.dentry = p->mnt_mountpoint;
36341f64 1446 q = clone_mnt(p, p->mnt_root, flag);
1da177e4
LT
1447 if (!q)
1448 goto Enomem;
99b7db7b 1449 br_write_lock(vfsmount_lock);
1da177e4 1450 list_add_tail(&q->mnt_list, &res->mnt_list);
1a390689 1451 attach_mnt(q, &path);
99b7db7b 1452 br_write_unlock(vfsmount_lock);
1da177e4
LT
1453 }
1454 }
1455 return res;
b58fed8b 1456Enomem:
1da177e4 1457 if (res) {
70fbcdf4 1458 LIST_HEAD(umount_list);
99b7db7b 1459 br_write_lock(vfsmount_lock);
a05964f3 1460 umount_tree(res, 0, &umount_list);
99b7db7b 1461 br_write_unlock(vfsmount_lock);
70fbcdf4 1462 release_mounts(&umount_list);
1da177e4
LT
1463 }
1464 return NULL;
1465}
1466
589ff870 1467struct vfsmount *collect_mounts(struct path *path)
8aec0809
AV
1468{
1469 struct vfsmount *tree;
1a60a280 1470 down_write(&namespace_sem);
589ff870 1471 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1a60a280 1472 up_write(&namespace_sem);
8aec0809
AV
1473 return tree;
1474}
1475
1476void drop_collected_mounts(struct vfsmount *mnt)
1477{
1478 LIST_HEAD(umount_list);
1a60a280 1479 down_write(&namespace_sem);
99b7db7b 1480 br_write_lock(vfsmount_lock);
8aec0809 1481 umount_tree(mnt, 0, &umount_list);
99b7db7b 1482 br_write_unlock(vfsmount_lock);
1a60a280 1483 up_write(&namespace_sem);
8aec0809
AV
1484 release_mounts(&umount_list);
1485}
1486
1f707137
AV
1487int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1488 struct vfsmount *root)
1489{
1490 struct vfsmount *mnt;
1491 int res = f(root, arg);
1492 if (res)
1493 return res;
1494 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1495 res = f(mnt, arg);
1496 if (res)
1497 return res;
1498 }
1499 return 0;
1500}
1501
719f5d7f
MS
1502static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1503{
1504 struct vfsmount *p;
1505
1506 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1507 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1508 mnt_release_group_id(p);
1509 }
1510}
1511
1512static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1513{
1514 struct vfsmount *p;
1515
1516 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1517 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1518 int err = mnt_alloc_group_id(p);
1519 if (err) {
1520 cleanup_group_ids(mnt, p);
1521 return err;
1522 }
1523 }
1524 }
1525
1526 return 0;
1527}
1528
b90fa9ae
RP
1529/*
1530 * @source_mnt : mount tree to be attached
21444403
RP
1531 * @nd : place the mount tree @source_mnt is attached
1532 * @parent_nd : if non-null, detach the source_mnt from its parent and
1533 * store the parent mount and mountpoint dentry.
1534 * (done when source_mnt is moved)
b90fa9ae
RP
1535 *
1536 * NOTE: in the table below explains the semantics when a source mount
1537 * of a given type is attached to a destination mount of a given type.
9676f0c6
RP
1538 * ---------------------------------------------------------------------------
1539 * | BIND MOUNT OPERATION |
1540 * |**************************************************************************
1541 * | source-->| shared | private | slave | unbindable |
1542 * | dest | | | | |
1543 * | | | | | | |
1544 * | v | | | | |
1545 * |**************************************************************************
1546 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1547 * | | | | | |
1548 * |non-shared| shared (+) | private | slave (*) | invalid |
1549 * ***************************************************************************
b90fa9ae
RP
1550 * A bind operation clones the source mount and mounts the clone on the
1551 * destination mount.
1552 *
1553 * (++) the cloned mount is propagated to all the mounts in the propagation
1554 * tree of the destination mount and the cloned mount is added to
1555 * the peer group of the source mount.
1556 * (+) the cloned mount is created under the destination mount and is marked
1557 * as shared. The cloned mount is added to the peer group of the source
1558 * mount.
5afe0022
RP
1559 * (+++) the mount is propagated to all the mounts in the propagation tree
1560 * of the destination mount and the cloned mount is made slave
1561 * of the same master as that of the source mount. The cloned mount
1562 * is marked as 'shared and slave'.
1563 * (*) the cloned mount is made a slave of the same master as that of the
1564 * source mount.
1565 *
9676f0c6
RP
1566 * ---------------------------------------------------------------------------
1567 * | MOVE MOUNT OPERATION |
1568 * |**************************************************************************
1569 * | source-->| shared | private | slave | unbindable |
1570 * | dest | | | | |
1571 * | | | | | | |
1572 * | v | | | | |
1573 * |**************************************************************************
1574 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1575 * | | | | | |
1576 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1577 * ***************************************************************************
5afe0022
RP
1578 *
1579 * (+) the mount is moved to the destination. And is then propagated to
1580 * all the mounts in the propagation tree of the destination mount.
21444403 1581 * (+*) the mount is moved to the destination.
5afe0022
RP
1582 * (+++) the mount is moved to the destination and is then propagated to
1583 * all the mounts belonging to the destination mount's propagation tree.
1584 * the mount is marked as 'shared and slave'.
1585 * (*) the mount continues to be a slave at the new location.
b90fa9ae
RP
1586 *
1587 * if the source mount is a tree, the operations explained above is
1588 * applied to each mount in the tree.
1589 * Must be called without spinlocks held, since this function can sleep
1590 * in allocations.
1591 */
1592static int attach_recursive_mnt(struct vfsmount *source_mnt,
1a390689 1593 struct path *path, struct path *parent_path)
b90fa9ae
RP
1594{
1595 LIST_HEAD(tree_list);
1a390689
AV
1596 struct vfsmount *dest_mnt = path->mnt;
1597 struct dentry *dest_dentry = path->dentry;
b90fa9ae 1598 struct vfsmount *child, *p;
719f5d7f 1599 int err;
b90fa9ae 1600
719f5d7f
MS
1601 if (IS_MNT_SHARED(dest_mnt)) {
1602 err = invent_group_ids(source_mnt, true);
1603 if (err)
1604 goto out;
1605 }
1606 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1607 if (err)
1608 goto out_cleanup_ids;
b90fa9ae 1609
99b7db7b 1610 br_write_lock(vfsmount_lock);
df1a1ad2 1611
b90fa9ae
RP
1612 if (IS_MNT_SHARED(dest_mnt)) {
1613 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1614 set_mnt_shared(p);
1615 }
1a390689
AV
1616 if (parent_path) {
1617 detach_mnt(source_mnt, parent_path);
1618 attach_mnt(source_mnt, path);
e5d67f07 1619 touch_mnt_namespace(parent_path->mnt->mnt_ns);
21444403
RP
1620 } else {
1621 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1622 commit_tree(source_mnt);
1623 }
b90fa9ae
RP
1624
1625 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1626 list_del_init(&child->mnt_hash);
1627 commit_tree(child);
1628 }
99b7db7b
NP
1629 br_write_unlock(vfsmount_lock);
1630
b90fa9ae 1631 return 0;
719f5d7f
MS
1632
1633 out_cleanup_ids:
1634 if (IS_MNT_SHARED(dest_mnt))
1635 cleanup_group_ids(source_mnt, NULL);
1636 out:
1637 return err;
b90fa9ae
RP
1638}
1639
8c3ee42e 1640static int graft_tree(struct vfsmount *mnt, struct path *path)
1da177e4
LT
1641{
1642 int err;
1643 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1644 return -EINVAL;
1645
8c3ee42e 1646 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1da177e4
LT
1647 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1648 return -ENOTDIR;
1649
1650 err = -ENOENT;
8c3ee42e 1651 mutex_lock(&path->dentry->d_inode->i_mutex);
d83c49f3 1652 if (cant_mount(path->dentry))
1da177e4
LT
1653 goto out_unlock;
1654
f3da392e 1655 if (!d_unlinked(path->dentry))
8c3ee42e 1656 err = attach_recursive_mnt(mnt, path, NULL);
1da177e4 1657out_unlock:
8c3ee42e 1658 mutex_unlock(&path->dentry->d_inode->i_mutex);
1da177e4
LT
1659 return err;
1660}
1661
7a2e8a8f
VA
1662/*
1663 * Sanity check the flags to change_mnt_propagation.
1664 */
1665
1666static int flags_to_propagation_type(int flags)
1667{
1668 int type = flags & ~MS_REC;
1669
1670 /* Fail if any non-propagation flags are set */
1671 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1672 return 0;
1673 /* Only one propagation flag should be set */
1674 if (!is_power_of_2(type))
1675 return 0;
1676 return type;
1677}
1678
07b20889
RP
1679/*
1680 * recursively change the type of the mountpoint.
1681 */
0a0d8a46 1682static int do_change_type(struct path *path, int flag)
07b20889 1683{
2d92ab3c 1684 struct vfsmount *m, *mnt = path->mnt;
07b20889 1685 int recurse = flag & MS_REC;
7a2e8a8f 1686 int type;
719f5d7f 1687 int err = 0;
07b20889 1688
ee6f9582
MS
1689 if (!capable(CAP_SYS_ADMIN))
1690 return -EPERM;
1691
2d92ab3c 1692 if (path->dentry != path->mnt->mnt_root)
07b20889
RP
1693 return -EINVAL;
1694
7a2e8a8f
VA
1695 type = flags_to_propagation_type(flag);
1696 if (!type)
1697 return -EINVAL;
1698
07b20889 1699 down_write(&namespace_sem);
719f5d7f
MS
1700 if (type == MS_SHARED) {
1701 err = invent_group_ids(mnt, recurse);
1702 if (err)
1703 goto out_unlock;
1704 }
1705
99b7db7b 1706 br_write_lock(vfsmount_lock);
07b20889
RP
1707 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1708 change_mnt_propagation(m, type);
99b7db7b 1709 br_write_unlock(vfsmount_lock);
719f5d7f
MS
1710
1711 out_unlock:
07b20889 1712 up_write(&namespace_sem);
719f5d7f 1713 return err;
07b20889
RP
1714}
1715
1da177e4
LT
1716/*
1717 * do loopback mount.
1718 */
0a0d8a46 1719static int do_loopback(struct path *path, char *old_name,
2dafe1c4 1720 int recurse)
1da177e4 1721{
2d92ab3c 1722 struct path old_path;
1da177e4 1723 struct vfsmount *mnt = NULL;
2d92ab3c 1724 int err = mount_is_safe(path);
1da177e4
LT
1725 if (err)
1726 return err;
1727 if (!old_name || !*old_name)
1728 return -EINVAL;
2d92ab3c 1729 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1da177e4
LT
1730 if (err)
1731 return err;
1732
390c6843 1733 down_write(&namespace_sem);
1da177e4 1734 err = -EINVAL;
2d92ab3c 1735 if (IS_MNT_UNBINDABLE(old_path.mnt))
4ac91378 1736 goto out;
9676f0c6 1737
2d92ab3c 1738 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
ccd48bc7 1739 goto out;
1da177e4 1740
ccd48bc7
AV
1741 err = -ENOMEM;
1742 if (recurse)
2d92ab3c 1743 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
ccd48bc7 1744 else
2d92ab3c 1745 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
ccd48bc7
AV
1746
1747 if (!mnt)
1748 goto out;
1749
2d92ab3c 1750 err = graft_tree(mnt, path);
ccd48bc7 1751 if (err) {
70fbcdf4 1752 LIST_HEAD(umount_list);
99b7db7b
NP
1753
1754 br_write_lock(vfsmount_lock);
a05964f3 1755 umount_tree(mnt, 0, &umount_list);
99b7db7b 1756 br_write_unlock(vfsmount_lock);
70fbcdf4 1757 release_mounts(&umount_list);
5b83d2c5 1758 }
1da177e4 1759
ccd48bc7 1760out:
390c6843 1761 up_write(&namespace_sem);
2d92ab3c 1762 path_put(&old_path);
1da177e4
LT
1763 return err;
1764}
1765
2e4b7fcd
DH
1766static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1767{
1768 int error = 0;
1769 int readonly_request = 0;
1770
1771 if (ms_flags & MS_RDONLY)
1772 readonly_request = 1;
1773 if (readonly_request == __mnt_is_readonly(mnt))
1774 return 0;
1775
1776 if (readonly_request)
1777 error = mnt_make_readonly(mnt);
1778 else
1779 __mnt_unmake_readonly(mnt);
1780 return error;
1781}
1782
1da177e4
LT
1783/*
1784 * change filesystem flags. dir should be a physical root of filesystem.
1785 * If you've mounted a non-root directory somewhere and want to do remount
1786 * on it - tough luck.
1787 */
0a0d8a46 1788static int do_remount(struct path *path, int flags, int mnt_flags,
1da177e4
LT
1789 void *data)
1790{
1791 int err;
2d92ab3c 1792 struct super_block *sb = path->mnt->mnt_sb;
1da177e4
LT
1793
1794 if (!capable(CAP_SYS_ADMIN))
1795 return -EPERM;
1796
2d92ab3c 1797 if (!check_mnt(path->mnt))
1da177e4
LT
1798 return -EINVAL;
1799
2d92ab3c 1800 if (path->dentry != path->mnt->mnt_root)
1da177e4
LT
1801 return -EINVAL;
1802
1803 down_write(&sb->s_umount);
2e4b7fcd 1804 if (flags & MS_BIND)
2d92ab3c 1805 err = change_mount_flags(path->mnt, flags);
4aa98cf7 1806 else
2e4b7fcd 1807 err = do_remount_sb(sb, flags, data, 0);
7b43a79f 1808 if (!err) {
99b7db7b 1809 br_write_lock(vfsmount_lock);
495d6c9c 1810 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
2d92ab3c 1811 path->mnt->mnt_flags = mnt_flags;
99b7db7b 1812 br_write_unlock(vfsmount_lock);
7b43a79f 1813 }
1da177e4 1814 up_write(&sb->s_umount);
0e55a7cc 1815 if (!err) {
99b7db7b 1816 br_write_lock(vfsmount_lock);
0e55a7cc 1817 touch_mnt_namespace(path->mnt->mnt_ns);
99b7db7b 1818 br_write_unlock(vfsmount_lock);
0e55a7cc 1819 }
1da177e4
LT
1820 return err;
1821}
1822
9676f0c6
RP
1823static inline int tree_contains_unbindable(struct vfsmount *mnt)
1824{
1825 struct vfsmount *p;
1826 for (p = mnt; p; p = next_mnt(p, mnt)) {
1827 if (IS_MNT_UNBINDABLE(p))
1828 return 1;
1829 }
1830 return 0;
1831}
1832
0a0d8a46 1833static int do_move_mount(struct path *path, char *old_name)
1da177e4 1834{
2d92ab3c 1835 struct path old_path, parent_path;
1da177e4
LT
1836 struct vfsmount *p;
1837 int err = 0;
1838 if (!capable(CAP_SYS_ADMIN))
1839 return -EPERM;
1840 if (!old_name || !*old_name)
1841 return -EINVAL;
2d92ab3c 1842 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1da177e4
LT
1843 if (err)
1844 return err;
1845
390c6843 1846 down_write(&namespace_sem);
cc53ce53
DH
1847 err = follow_down(path, true);
1848 if (err < 0)
1849 goto out;
1850
1da177e4 1851 err = -EINVAL;
2d92ab3c 1852 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1da177e4
LT
1853 goto out;
1854
1855 err = -ENOENT;
2d92ab3c 1856 mutex_lock(&path->dentry->d_inode->i_mutex);
d83c49f3 1857 if (cant_mount(path->dentry))
1da177e4
LT
1858 goto out1;
1859
f3da392e 1860 if (d_unlinked(path->dentry))
21444403 1861 goto out1;
1da177e4
LT
1862
1863 err = -EINVAL;
2d92ab3c 1864 if (old_path.dentry != old_path.mnt->mnt_root)
21444403 1865 goto out1;
1da177e4 1866
2d92ab3c 1867 if (old_path.mnt == old_path.mnt->mnt_parent)
21444403 1868 goto out1;
1da177e4 1869
2d92ab3c
AV
1870 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1871 S_ISDIR(old_path.dentry->d_inode->i_mode))
21444403
RP
1872 goto out1;
1873 /*
1874 * Don't move a mount residing in a shared parent.
1875 */
2d92ab3c
AV
1876 if (old_path.mnt->mnt_parent &&
1877 IS_MNT_SHARED(old_path.mnt->mnt_parent))
21444403 1878 goto out1;
9676f0c6
RP
1879 /*
1880 * Don't move a mount tree containing unbindable mounts to a destination
1881 * mount which is shared.
1882 */
2d92ab3c
AV
1883 if (IS_MNT_SHARED(path->mnt) &&
1884 tree_contains_unbindable(old_path.mnt))
9676f0c6 1885 goto out1;
1da177e4 1886 err = -ELOOP;
2d92ab3c
AV
1887 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1888 if (p == old_path.mnt)
21444403 1889 goto out1;
1da177e4 1890
2d92ab3c 1891 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
4ac91378 1892 if (err)
21444403 1893 goto out1;
1da177e4
LT
1894
1895 /* if the mount is moved, it should no longer be expire
1896 * automatically */
2d92ab3c 1897 list_del_init(&old_path.mnt->mnt_expire);
1da177e4 1898out1:
2d92ab3c 1899 mutex_unlock(&path->dentry->d_inode->i_mutex);
1da177e4 1900out:
390c6843 1901 up_write(&namespace_sem);
1da177e4 1902 if (!err)
1a390689 1903 path_put(&parent_path);
2d92ab3c 1904 path_put(&old_path);
1da177e4
LT
1905 return err;
1906}
1907
1908/*
1909 * create a new mount for userspace and request it to be added into the
1910 * namespace's tree
1911 */
0a0d8a46 1912static int do_new_mount(struct path *path, char *type, int flags,
1da177e4
LT
1913 int mnt_flags, char *name, void *data)
1914{
1915 struct vfsmount *mnt;
1916
eca6f534 1917 if (!type)
1da177e4
LT
1918 return -EINVAL;
1919
1920 /* we need capabilities... */
1921 if (!capable(CAP_SYS_ADMIN))
1922 return -EPERM;
1923
1924 mnt = do_kern_mount(type, flags, name, data);
1925 if (IS_ERR(mnt))
1926 return PTR_ERR(mnt);
1927
ea5b778a 1928 return do_add_mount(mnt, path, mnt_flags);
1da177e4
LT
1929}
1930
1931/*
1932 * add a mount into a namespace's mount tree
ea5b778a 1933 * - this unconditionally eats one of the caller's references to newmnt.
1da177e4 1934 */
ea5b778a 1935int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1da177e4
LT
1936{
1937 int err;
1938
8089352a 1939 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
27d55f1f 1940
390c6843 1941 down_write(&namespace_sem);
1da177e4 1942 /* Something was mounted here while we slept */
cc53ce53
DH
1943 err = follow_down(path, true);
1944 if (err < 0)
1945 goto unlock;
1946
1da177e4 1947 err = -EINVAL;
dd5cae6e 1948 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1da177e4
LT
1949 goto unlock;
1950
1951 /* Refuse the same filesystem on the same mount point */
1952 err = -EBUSY;
8d66bf54
AV
1953 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1954 path->mnt->mnt_root == path->dentry)
1da177e4
LT
1955 goto unlock;
1956
1957 err = -EINVAL;
1958 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1959 goto unlock;
1960
1961 newmnt->mnt_flags = mnt_flags;
8d66bf54 1962 if ((err = graft_tree(newmnt, path)))
5b83d2c5 1963 goto unlock;
1da177e4 1964
390c6843 1965 up_write(&namespace_sem);
5b83d2c5 1966 return 0;
1da177e4
LT
1967
1968unlock:
390c6843 1969 up_write(&namespace_sem);
b3e19d92 1970 mntput_long(newmnt);
1da177e4
LT
1971 return err;
1972}
1973
ea5b778a
DH
1974/**
1975 * mnt_set_expiry - Put a mount on an expiration list
1976 * @mnt: The mount to list.
1977 * @expiry_list: The list to add the mount to.
1978 */
1979void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1980{
1981 down_write(&namespace_sem);
1982 br_write_lock(vfsmount_lock);
1983
1984 list_add_tail(&mnt->mnt_expire, expiry_list);
1985
1986 br_write_unlock(vfsmount_lock);
1987 up_write(&namespace_sem);
1988}
1989EXPORT_SYMBOL(mnt_set_expiry);
1990
1991/*
1992 * Remove a vfsmount from any expiration list it may be on
1993 */
1994void mnt_clear_expiry(struct vfsmount *mnt)
1995{
1996 if (!list_empty(&mnt->mnt_expire)) {
1997 down_write(&namespace_sem);
1998 br_write_lock(vfsmount_lock);
1999 list_del_init(&mnt->mnt_expire);
2000 br_write_unlock(vfsmount_lock);
2001 up_write(&namespace_sem);
2002 }
2003}
1da177e4
LT
2004
2005/*
2006 * process a list of expirable mountpoints with the intent of discarding any
2007 * mountpoints that aren't in use and haven't been touched since last we came
2008 * here
2009 */
2010void mark_mounts_for_expiry(struct list_head *mounts)
2011{
1da177e4
LT
2012 struct vfsmount *mnt, *next;
2013 LIST_HEAD(graveyard);
bcc5c7d2 2014 LIST_HEAD(umounts);
1da177e4
LT
2015
2016 if (list_empty(mounts))
2017 return;
2018
bcc5c7d2 2019 down_write(&namespace_sem);
99b7db7b 2020 br_write_lock(vfsmount_lock);
1da177e4
LT
2021
2022 /* extract from the expiration list every vfsmount that matches the
2023 * following criteria:
2024 * - only referenced by its parent vfsmount
2025 * - still marked for expiry (marked on the last call here; marks are
2026 * cleared by mntput())
2027 */
55e700b9 2028 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1da177e4 2029 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
bcc5c7d2 2030 propagate_mount_busy(mnt, 1))
1da177e4 2031 continue;
55e700b9 2032 list_move(&mnt->mnt_expire, &graveyard);
1da177e4 2033 }
bcc5c7d2
AV
2034 while (!list_empty(&graveyard)) {
2035 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2036 touch_mnt_namespace(mnt->mnt_ns);
2037 umount_tree(mnt, 1, &umounts);
2038 }
99b7db7b 2039 br_write_unlock(vfsmount_lock);
bcc5c7d2
AV
2040 up_write(&namespace_sem);
2041
2042 release_mounts(&umounts);
5528f911
TM
2043}
2044
2045EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2046
2047/*
2048 * Ripoff of 'select_parent()'
2049 *
2050 * search the list of submounts for a given mountpoint, and move any
2051 * shrinkable submounts to the 'graveyard' list.
2052 */
2053static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2054{
2055 struct vfsmount *this_parent = parent;
2056 struct list_head *next;
2057 int found = 0;
2058
2059repeat:
2060 next = this_parent->mnt_mounts.next;
2061resume:
2062 while (next != &this_parent->mnt_mounts) {
2063 struct list_head *tmp = next;
2064 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2065
2066 next = tmp->next;
2067 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1da177e4 2068 continue;
5528f911
TM
2069 /*
2070 * Descend a level if the d_mounts list is non-empty.
2071 */
2072 if (!list_empty(&mnt->mnt_mounts)) {
2073 this_parent = mnt;
2074 goto repeat;
2075 }
1da177e4 2076
5528f911 2077 if (!propagate_mount_busy(mnt, 1)) {
5528f911
TM
2078 list_move_tail(&mnt->mnt_expire, graveyard);
2079 found++;
2080 }
1da177e4 2081 }
5528f911
TM
2082 /*
2083 * All done at this level ... ascend and resume the search
2084 */
2085 if (this_parent != parent) {
2086 next = this_parent->mnt_child.next;
2087 this_parent = this_parent->mnt_parent;
2088 goto resume;
2089 }
2090 return found;
2091}
2092
2093/*
2094 * process a list of expirable mountpoints with the intent of discarding any
2095 * submounts of a specific parent mountpoint
99b7db7b
NP
2096 *
2097 * vfsmount_lock must be held for write
5528f911 2098 */
c35038be 2099static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
5528f911
TM
2100{
2101 LIST_HEAD(graveyard);
c35038be 2102 struct vfsmount *m;
5528f911 2103
5528f911 2104 /* extract submounts of 'mountpoint' from the expiration list */
c35038be 2105 while (select_submounts(mnt, &graveyard)) {
bcc5c7d2 2106 while (!list_empty(&graveyard)) {
c35038be 2107 m = list_first_entry(&graveyard, struct vfsmount,
bcc5c7d2 2108 mnt_expire);
afef80b3
EB
2109 touch_mnt_namespace(m->mnt_ns);
2110 umount_tree(m, 1, umounts);
bcc5c7d2
AV
2111 }
2112 }
1da177e4
LT
2113}
2114
1da177e4
LT
2115/*
2116 * Some copy_from_user() implementations do not return the exact number of
2117 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2118 * Note that this function differs from copy_from_user() in that it will oops
2119 * on bad values of `to', rather than returning a short copy.
2120 */
b58fed8b
RP
2121static long exact_copy_from_user(void *to, const void __user * from,
2122 unsigned long n)
1da177e4
LT
2123{
2124 char *t = to;
2125 const char __user *f = from;
2126 char c;
2127
2128 if (!access_ok(VERIFY_READ, from, n))
2129 return n;
2130
2131 while (n) {
2132 if (__get_user(c, f)) {
2133 memset(t, 0, n);
2134 break;
2135 }
2136 *t++ = c;
2137 f++;
2138 n--;
2139 }
2140 return n;
2141}
2142
b58fed8b 2143int copy_mount_options(const void __user * data, unsigned long *where)
1da177e4
LT
2144{
2145 int i;
2146 unsigned long page;
2147 unsigned long size;
b58fed8b 2148
1da177e4
LT
2149 *where = 0;
2150 if (!data)
2151 return 0;
2152
2153 if (!(page = __get_free_page(GFP_KERNEL)))
2154 return -ENOMEM;
2155
2156 /* We only care that *some* data at the address the user
2157 * gave us is valid. Just in case, we'll zero
2158 * the remainder of the page.
2159 */
2160 /* copy_from_user cannot cross TASK_SIZE ! */
2161 size = TASK_SIZE - (unsigned long)data;
2162 if (size > PAGE_SIZE)
2163 size = PAGE_SIZE;
2164
2165 i = size - exact_copy_from_user((void *)page, data, size);
2166 if (!i) {
b58fed8b 2167 free_page(page);
1da177e4
LT
2168 return -EFAULT;
2169 }
2170 if (i != PAGE_SIZE)
2171 memset((char *)page + i, 0, PAGE_SIZE - i);
2172 *where = page;
2173 return 0;
2174}
2175
eca6f534
VN
2176int copy_mount_string(const void __user *data, char **where)
2177{
2178 char *tmp;
2179
2180 if (!data) {
2181 *where = NULL;
2182 return 0;
2183 }
2184
2185 tmp = strndup_user(data, PAGE_SIZE);
2186 if (IS_ERR(tmp))
2187 return PTR_ERR(tmp);
2188
2189 *where = tmp;
2190 return 0;
2191}
2192
1da177e4
LT
2193/*
2194 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2195 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2196 *
2197 * data is a (void *) that can point to any structure up to
2198 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2199 * information (or be NULL).
2200 *
2201 * Pre-0.97 versions of mount() didn't have a flags word.
2202 * When the flags word was introduced its top half was required
2203 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2204 * Therefore, if this magic number is present, it carries no information
2205 * and must be discarded.
2206 */
b58fed8b 2207long do_mount(char *dev_name, char *dir_name, char *type_page,
1da177e4
LT
2208 unsigned long flags, void *data_page)
2209{
2d92ab3c 2210 struct path path;
1da177e4
LT
2211 int retval = 0;
2212 int mnt_flags = 0;
2213
2214 /* Discard magic */
2215 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2216 flags &= ~MS_MGC_MSK;
2217
2218 /* Basic sanity checks */
2219
2220 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2221 return -EINVAL;
1da177e4
LT
2222
2223 if (data_page)
2224 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2225
a27ab9f2
TH
2226 /* ... and get the mountpoint */
2227 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2228 if (retval)
2229 return retval;
2230
2231 retval = security_sb_mount(dev_name, &path,
2232 type_page, flags, data_page);
2233 if (retval)
2234 goto dput_out;
2235
613cbe3d
AK
2236 /* Default to relatime unless overriden */
2237 if (!(flags & MS_NOATIME))
2238 mnt_flags |= MNT_RELATIME;
0a1c01c9 2239
1da177e4
LT
2240 /* Separate the per-mountpoint flags */
2241 if (flags & MS_NOSUID)
2242 mnt_flags |= MNT_NOSUID;
2243 if (flags & MS_NODEV)
2244 mnt_flags |= MNT_NODEV;
2245 if (flags & MS_NOEXEC)
2246 mnt_flags |= MNT_NOEXEC;
fc33a7bb
CH
2247 if (flags & MS_NOATIME)
2248 mnt_flags |= MNT_NOATIME;
2249 if (flags & MS_NODIRATIME)
2250 mnt_flags |= MNT_NODIRATIME;
d0adde57
MG
2251 if (flags & MS_STRICTATIME)
2252 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2e4b7fcd
DH
2253 if (flags & MS_RDONLY)
2254 mnt_flags |= MNT_READONLY;
fc33a7bb 2255
7a4dec53 2256 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
d0adde57
MG
2257 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2258 MS_STRICTATIME);
1da177e4 2259
1da177e4 2260 if (flags & MS_REMOUNT)
2d92ab3c 2261 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1da177e4
LT
2262 data_page);
2263 else if (flags & MS_BIND)
2d92ab3c 2264 retval = do_loopback(&path, dev_name, flags & MS_REC);
9676f0c6 2265 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2d92ab3c 2266 retval = do_change_type(&path, flags);
1da177e4 2267 else if (flags & MS_MOVE)
2d92ab3c 2268 retval = do_move_mount(&path, dev_name);
1da177e4 2269 else
2d92ab3c 2270 retval = do_new_mount(&path, type_page, flags, mnt_flags,
1da177e4
LT
2271 dev_name, data_page);
2272dput_out:
2d92ab3c 2273 path_put(&path);
1da177e4
LT
2274 return retval;
2275}
2276
cf8d2c11
TM
2277static struct mnt_namespace *alloc_mnt_ns(void)
2278{
2279 struct mnt_namespace *new_ns;
2280
2281 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2282 if (!new_ns)
2283 return ERR_PTR(-ENOMEM);
2284 atomic_set(&new_ns->count, 1);
2285 new_ns->root = NULL;
2286 INIT_LIST_HEAD(&new_ns->list);
2287 init_waitqueue_head(&new_ns->poll);
2288 new_ns->event = 0;
2289 return new_ns;
2290}
2291
741a2951
JD
2292/*
2293 * Allocate a new namespace structure and populate it with contents
2294 * copied from the namespace of the passed in task structure.
2295 */
e3222c4e 2296static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
6b3286ed 2297 struct fs_struct *fs)
1da177e4 2298{
6b3286ed 2299 struct mnt_namespace *new_ns;
7f2da1e7 2300 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1da177e4
LT
2301 struct vfsmount *p, *q;
2302
cf8d2c11
TM
2303 new_ns = alloc_mnt_ns();
2304 if (IS_ERR(new_ns))
2305 return new_ns;
1da177e4 2306
390c6843 2307 down_write(&namespace_sem);
1da177e4 2308 /* First pass: copy the tree topology */
6b3286ed 2309 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
9676f0c6 2310 CL_COPY_ALL | CL_EXPIRE);
1da177e4 2311 if (!new_ns->root) {
390c6843 2312 up_write(&namespace_sem);
1da177e4 2313 kfree(new_ns);
5cc4a034 2314 return ERR_PTR(-ENOMEM);
1da177e4 2315 }
99b7db7b 2316 br_write_lock(vfsmount_lock);
1da177e4 2317 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
99b7db7b 2318 br_write_unlock(vfsmount_lock);
1da177e4
LT
2319
2320 /*
2321 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2322 * as belonging to new namespace. We have already acquired a private
2323 * fs_struct, so tsk->fs->lock is not needed.
2324 */
6b3286ed 2325 p = mnt_ns->root;
1da177e4
LT
2326 q = new_ns->root;
2327 while (p) {
6b3286ed 2328 q->mnt_ns = new_ns;
1da177e4 2329 if (fs) {
6ac08c39 2330 if (p == fs->root.mnt) {
1da177e4 2331 rootmnt = p;
b3e19d92 2332 fs->root.mnt = mntget_long(q);
1da177e4 2333 }
6ac08c39 2334 if (p == fs->pwd.mnt) {
1da177e4 2335 pwdmnt = p;
b3e19d92 2336 fs->pwd.mnt = mntget_long(q);
1da177e4 2337 }
1da177e4 2338 }
6b3286ed 2339 p = next_mnt(p, mnt_ns->root);
1da177e4
LT
2340 q = next_mnt(q, new_ns->root);
2341 }
390c6843 2342 up_write(&namespace_sem);
1da177e4 2343
1da177e4 2344 if (rootmnt)
b3e19d92 2345 mntput_long(rootmnt);
1da177e4 2346 if (pwdmnt)
b3e19d92 2347 mntput_long(pwdmnt);
1da177e4 2348
741a2951
JD
2349 return new_ns;
2350}
2351
213dd266 2352struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
e3222c4e 2353 struct fs_struct *new_fs)
741a2951 2354{
6b3286ed 2355 struct mnt_namespace *new_ns;
741a2951 2356
e3222c4e 2357 BUG_ON(!ns);
6b3286ed 2358 get_mnt_ns(ns);
741a2951
JD
2359
2360 if (!(flags & CLONE_NEWNS))
e3222c4e 2361 return ns;
741a2951 2362
e3222c4e 2363 new_ns = dup_mnt_ns(ns, new_fs);
741a2951 2364
6b3286ed 2365 put_mnt_ns(ns);
e3222c4e 2366 return new_ns;
1da177e4
LT
2367}
2368
cf8d2c11
TM
2369/**
2370 * create_mnt_ns - creates a private namespace and adds a root filesystem
2371 * @mnt: pointer to the new root filesystem mountpoint
2372 */
a2770d86 2373struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
cf8d2c11
TM
2374{
2375 struct mnt_namespace *new_ns;
2376
2377 new_ns = alloc_mnt_ns();
2378 if (!IS_ERR(new_ns)) {
2379 mnt->mnt_ns = new_ns;
2380 new_ns->root = mnt;
2381 list_add(&new_ns->list, &new_ns->root->mnt_list);
2382 }
2383 return new_ns;
2384}
a2770d86 2385EXPORT_SYMBOL(create_mnt_ns);
cf8d2c11 2386
bdc480e3
HC
2387SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2388 char __user *, type, unsigned long, flags, void __user *, data)
1da177e4 2389{
eca6f534
VN
2390 int ret;
2391 char *kernel_type;
2392 char *kernel_dir;
2393 char *kernel_dev;
1da177e4 2394 unsigned long data_page;
1da177e4 2395
eca6f534
VN
2396 ret = copy_mount_string(type, &kernel_type);
2397 if (ret < 0)
2398 goto out_type;
1da177e4 2399
eca6f534
VN
2400 kernel_dir = getname(dir_name);
2401 if (IS_ERR(kernel_dir)) {
2402 ret = PTR_ERR(kernel_dir);
2403 goto out_dir;
2404 }
1da177e4 2405
eca6f534
VN
2406 ret = copy_mount_string(dev_name, &kernel_dev);
2407 if (ret < 0)
2408 goto out_dev;
1da177e4 2409
eca6f534
VN
2410 ret = copy_mount_options(data, &data_page);
2411 if (ret < 0)
2412 goto out_data;
1da177e4 2413
eca6f534
VN
2414 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2415 (void *) data_page);
1da177e4 2416
eca6f534
VN
2417 free_page(data_page);
2418out_data:
2419 kfree(kernel_dev);
2420out_dev:
2421 putname(kernel_dir);
2422out_dir:
2423 kfree(kernel_type);
2424out_type:
2425 return ret;
1da177e4
LT
2426}
2427
1da177e4
LT
2428/*
2429 * pivot_root Semantics:
2430 * Moves the root file system of the current process to the directory put_old,
2431 * makes new_root as the new root file system of the current process, and sets
2432 * root/cwd of all processes which had them on the current root to new_root.
2433 *
2434 * Restrictions:
2435 * The new_root and put_old must be directories, and must not be on the
2436 * same file system as the current process root. The put_old must be
2437 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2438 * pointed to by put_old must yield the same directory as new_root. No other
2439 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2440 *
4a0d11fa
NB
2441 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2442 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2443 * in this situation.
2444 *
1da177e4
LT
2445 * Notes:
2446 * - we don't move root/cwd if they are not at the root (reason: if something
2447 * cared enough to change them, it's probably wrong to force them elsewhere)
2448 * - it's okay to pick a root that isn't the root of a file system, e.g.
2449 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2450 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2451 * first.
2452 */
3480b257
HC
2453SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2454 const char __user *, put_old)
1da177e4
LT
2455{
2456 struct vfsmount *tmp;
2d8f3038 2457 struct path new, old, parent_path, root_parent, root;
1da177e4
LT
2458 int error;
2459
2460 if (!capable(CAP_SYS_ADMIN))
2461 return -EPERM;
2462
2d8f3038 2463 error = user_path_dir(new_root, &new);
1da177e4
LT
2464 if (error)
2465 goto out0;
2466 error = -EINVAL;
2d8f3038 2467 if (!check_mnt(new.mnt))
1da177e4
LT
2468 goto out1;
2469
2d8f3038 2470 error = user_path_dir(put_old, &old);
1da177e4
LT
2471 if (error)
2472 goto out1;
2473
2d8f3038 2474 error = security_sb_pivotroot(&old, &new);
1da177e4 2475 if (error) {
2d8f3038 2476 path_put(&old);
1da177e4
LT
2477 goto out1;
2478 }
2479
f7ad3c6b 2480 get_fs_root(current->fs, &root);
390c6843 2481 down_write(&namespace_sem);
2d8f3038 2482 mutex_lock(&old.dentry->d_inode->i_mutex);
1da177e4 2483 error = -EINVAL;
2d8f3038
AV
2484 if (IS_MNT_SHARED(old.mnt) ||
2485 IS_MNT_SHARED(new.mnt->mnt_parent) ||
8c3ee42e 2486 IS_MNT_SHARED(root.mnt->mnt_parent))
21444403 2487 goto out2;
8c3ee42e 2488 if (!check_mnt(root.mnt))
1da177e4
LT
2489 goto out2;
2490 error = -ENOENT;
d83c49f3 2491 if (cant_mount(old.dentry))
1da177e4 2492 goto out2;
f3da392e 2493 if (d_unlinked(new.dentry))
1da177e4 2494 goto out2;
f3da392e 2495 if (d_unlinked(old.dentry))
1da177e4
LT
2496 goto out2;
2497 error = -EBUSY;
2d8f3038
AV
2498 if (new.mnt == root.mnt ||
2499 old.mnt == root.mnt)
1da177e4
LT
2500 goto out2; /* loop, on the same file system */
2501 error = -EINVAL;
8c3ee42e 2502 if (root.mnt->mnt_root != root.dentry)
1da177e4 2503 goto out2; /* not a mountpoint */
8c3ee42e 2504 if (root.mnt->mnt_parent == root.mnt)
0bb6fcc1 2505 goto out2; /* not attached */
2d8f3038 2506 if (new.mnt->mnt_root != new.dentry)
1da177e4 2507 goto out2; /* not a mountpoint */
2d8f3038 2508 if (new.mnt->mnt_parent == new.mnt)
0bb6fcc1 2509 goto out2; /* not attached */
4ac91378 2510 /* make sure we can reach put_old from new_root */
2d8f3038 2511 tmp = old.mnt;
99b7db7b 2512 br_write_lock(vfsmount_lock);
2d8f3038 2513 if (tmp != new.mnt) {
1da177e4
LT
2514 for (;;) {
2515 if (tmp->mnt_parent == tmp)
2516 goto out3; /* already mounted on put_old */
2d8f3038 2517 if (tmp->mnt_parent == new.mnt)
1da177e4
LT
2518 break;
2519 tmp = tmp->mnt_parent;
2520 }
2d8f3038 2521 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
1da177e4 2522 goto out3;
2d8f3038 2523 } else if (!is_subdir(old.dentry, new.dentry))
1da177e4 2524 goto out3;
2d8f3038 2525 detach_mnt(new.mnt, &parent_path);
8c3ee42e 2526 detach_mnt(root.mnt, &root_parent);
4ac91378 2527 /* mount old root on put_old */
2d8f3038 2528 attach_mnt(root.mnt, &old);
4ac91378 2529 /* mount new_root on / */
2d8f3038 2530 attach_mnt(new.mnt, &root_parent);
6b3286ed 2531 touch_mnt_namespace(current->nsproxy->mnt_ns);
99b7db7b 2532 br_write_unlock(vfsmount_lock);
2d8f3038 2533 chroot_fs_refs(&root, &new);
b3e19d92 2534
1da177e4 2535 error = 0;
1a390689
AV
2536 path_put(&root_parent);
2537 path_put(&parent_path);
1da177e4 2538out2:
2d8f3038 2539 mutex_unlock(&old.dentry->d_inode->i_mutex);
390c6843 2540 up_write(&namespace_sem);
8c3ee42e 2541 path_put(&root);
2d8f3038 2542 path_put(&old);
1da177e4 2543out1:
2d8f3038 2544 path_put(&new);
1da177e4 2545out0:
1da177e4
LT
2546 return error;
2547out3:
99b7db7b 2548 br_write_unlock(vfsmount_lock);
1da177e4
LT
2549 goto out2;
2550}
2551
2552static void __init init_mount_tree(void)
2553{
2554 struct vfsmount *mnt;
6b3286ed 2555 struct mnt_namespace *ns;
ac748a09 2556 struct path root;
1da177e4
LT
2557
2558 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2559 if (IS_ERR(mnt))
2560 panic("Can't create rootfs");
b3e19d92 2561
3b22edc5
TM
2562 ns = create_mnt_ns(mnt);
2563 if (IS_ERR(ns))
1da177e4 2564 panic("Can't allocate initial namespace");
6b3286ed
KK
2565
2566 init_task.nsproxy->mnt_ns = ns;
2567 get_mnt_ns(ns);
2568
ac748a09
JB
2569 root.mnt = ns->root;
2570 root.dentry = ns->root->mnt_root;
2571
2572 set_fs_pwd(current->fs, &root);
2573 set_fs_root(current->fs, &root);
1da177e4
LT
2574}
2575
74bf17cf 2576void __init mnt_init(void)
1da177e4 2577{
13f14b4d 2578 unsigned u;
15a67dd8 2579 int err;
1da177e4 2580
390c6843
RP
2581 init_rwsem(&namespace_sem);
2582
1da177e4 2583 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
20c2df83 2584 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1da177e4 2585
b58fed8b 2586 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1da177e4
LT
2587
2588 if (!mount_hashtable)
2589 panic("Failed to allocate mount hash table\n");
2590
13f14b4d
ED
2591 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2592
2593 for (u = 0; u < HASH_SIZE; u++)
2594 INIT_LIST_HEAD(&mount_hashtable[u]);
1da177e4 2595
99b7db7b
NP
2596 br_lock_init(vfsmount_lock);
2597
15a67dd8
RD
2598 err = sysfs_init();
2599 if (err)
2600 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
8e24eea7 2601 __func__, err);
00d26666
GKH
2602 fs_kobj = kobject_create_and_add("fs", NULL);
2603 if (!fs_kobj)
8e24eea7 2604 printk(KERN_WARNING "%s: kobj create error\n", __func__);
1da177e4
LT
2605 init_rootfs();
2606 init_mount_tree();
2607}
2608
616511d0 2609void put_mnt_ns(struct mnt_namespace *ns)
1da177e4 2610{
70fbcdf4 2611 LIST_HEAD(umount_list);
616511d0 2612
d498b25a 2613 if (!atomic_dec_and_test(&ns->count))
616511d0 2614 return;
390c6843 2615 down_write(&namespace_sem);
99b7db7b 2616 br_write_lock(vfsmount_lock);
d498b25a 2617 umount_tree(ns->root, 0, &umount_list);
99b7db7b 2618 br_write_unlock(vfsmount_lock);
390c6843 2619 up_write(&namespace_sem);
70fbcdf4 2620 release_mounts(&umount_list);
6b3286ed 2621 kfree(ns);
1da177e4 2622}
cf8d2c11 2623EXPORT_SYMBOL(put_mnt_ns);