vfs: mnt_ns moved to struct mount
[linux-2.6-block.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 */
b105e270 81static int mnt_alloc_id(struct mount *mnt)
73cd49ec
MS
82{
83 int res;
84
85retry:
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
99b7db7b 87 spin_lock(&mnt_id_lock);
b105e270 88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt.mnt_id);
f21f6220 89 if (!res)
b105e270 90 mnt_id_start = mnt->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
b105e270 98static void mnt_free_id(struct mount *mnt)
73cd49ec 99{
b105e270 100 int id = mnt->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 */
4b8b21f4 113static int mnt_alloc_group_id(struct mount *mnt)
719f5d7f 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,
4b8b21f4 122 &mnt->mnt.mnt_group_id);
f21f6220 123 if (!res)
4b8b21f4 124 mnt_group_start = mnt->mnt.mnt_group_id + 1;
f21f6220
AV
125
126 return res;
719f5d7f
MS
127}
128
129/*
130 * Release a peer group ID
131 */
4b8b21f4 132void mnt_release_group_id(struct mount *mnt)
719f5d7f 133{
4b8b21f4 134 int id = mnt->mnt.mnt_group_id;
f21f6220
AV
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
4b8b21f4 138 mnt->mnt.mnt_group_id = 0;
719f5d7f
MS
139}
140
b3e19d92
NP
141/*
142 * vfsmount lock must be held for read
143 */
83adc753 144static inline void mnt_add_count(struct mount *mnt, int n)
b3e19d92
NP
145{
146#ifdef CONFIG_SMP
68e8a9fe 147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
b3e19d92
NP
148#else
149 preempt_disable();
68e8a9fe 150 mnt->mnt_count += n;
b3e19d92
NP
151 preempt_enable();
152#endif
153}
154
b3e19d92
NP
155/*
156 * vfsmount lock must be held for write
157 */
83adc753 158unsigned int mnt_get_count(struct mount *mnt)
b3e19d92
NP
159{
160#ifdef CONFIG_SMP
f03c6599 161 unsigned int count = 0;
b3e19d92
NP
162 int cpu;
163
164 for_each_possible_cpu(cpu) {
68e8a9fe 165 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
b3e19d92
NP
166 }
167
168 return count;
169#else
68e8a9fe 170 return mnt->mnt_count;
b3e19d92
NP
171#endif
172}
173
b105e270 174static struct mount *alloc_vfsmnt(const char *name)
1da177e4 175{
7d6fec45
AV
176 struct mount *p = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
177 if (p) {
178 struct vfsmount *mnt = &p->mnt;
73cd49ec
MS
179 int err;
180
b105e270 181 err = mnt_alloc_id(p);
88b38782
LZ
182 if (err)
183 goto out_free_cache;
184
185 if (name) {
186 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
188 goto out_free_id;
73cd49ec
MS
189 }
190
b3e19d92 191#ifdef CONFIG_SMP
68e8a9fe
AV
192 p->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 if (!p->mnt_pcp)
b3e19d92
NP
194 goto out_free_devname;
195
68e8a9fe 196 this_cpu_add(p->mnt_pcp->mnt_count, 1);
b3e19d92 197#else
68e8a9fe
AV
198 p->mnt_count = 1;
199 p->mnt_writers = 0;
b3e19d92
NP
200#endif
201
1b8e5564 202 INIT_LIST_HEAD(&p->mnt_hash);
6b41d536
AV
203 INIT_LIST_HEAD(&p->mnt_child);
204 INIT_LIST_HEAD(&p->mnt_mounts);
1da177e4 205 INIT_LIST_HEAD(&mnt->mnt_list);
6776db3d
AV
206 INIT_LIST_HEAD(&p->mnt_expire);
207 INIT_LIST_HEAD(&p->mnt_share);
208 INIT_LIST_HEAD(&p->mnt_slave_list);
209 INIT_LIST_HEAD(&p->mnt_slave);
2504c5d6
AG
210#ifdef CONFIG_FSNOTIFY
211 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
d3ef3d73 212#endif
1da177e4 213 }
b105e270 214 return p;
88b38782 215
d3ef3d73 216#ifdef CONFIG_SMP
217out_free_devname:
7d6fec45 218 kfree(p->mnt.mnt_devname);
d3ef3d73 219#endif
88b38782 220out_free_id:
b105e270 221 mnt_free_id(p);
88b38782 222out_free_cache:
7d6fec45 223 kmem_cache_free(mnt_cache, p);
88b38782 224 return NULL;
1da177e4
LT
225}
226
3d733633
DH
227/*
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
233 * a filesystem.
234 */
235/*
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
238 *
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
244 * r/w.
245 */
246int __mnt_is_readonly(struct vfsmount *mnt)
247{
2e4b7fcd
DH
248 if (mnt->mnt_flags & MNT_READONLY)
249 return 1;
250 if (mnt->mnt_sb->s_flags & MS_RDONLY)
251 return 1;
252 return 0;
3d733633
DH
253}
254EXPORT_SYMBOL_GPL(__mnt_is_readonly);
255
83adc753 256static inline void mnt_inc_writers(struct mount *mnt)
d3ef3d73 257{
258#ifdef CONFIG_SMP
68e8a9fe 259 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
d3ef3d73 260#else
68e8a9fe 261 mnt->mnt_writers++;
d3ef3d73 262#endif
263}
3d733633 264
83adc753 265static inline void mnt_dec_writers(struct mount *mnt)
3d733633 266{
d3ef3d73 267#ifdef CONFIG_SMP
68e8a9fe 268 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
d3ef3d73 269#else
68e8a9fe 270 mnt->mnt_writers--;
d3ef3d73 271#endif
3d733633 272}
3d733633 273
83adc753 274static unsigned int mnt_get_writers(struct mount *mnt)
3d733633 275{
d3ef3d73 276#ifdef CONFIG_SMP
277 unsigned int count = 0;
3d733633 278 int cpu;
3d733633
DH
279
280 for_each_possible_cpu(cpu) {
68e8a9fe 281 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
3d733633 282 }
3d733633 283
d3ef3d73 284 return count;
285#else
286 return mnt->mnt_writers;
287#endif
3d733633
DH
288}
289
8366025e
DH
290/*
291 * Most r/o checks on a fs are for operations that take
292 * discrete amounts of time, like a write() or unlink().
293 * We must keep track of when those operations start
294 * (for permission checks) and when they end, so that
295 * we can determine when writes are able to occur to
296 * a filesystem.
297 */
298/**
299 * mnt_want_write - get write access to a mount
83adc753 300 * @m: the mount on which to take a write
8366025e
DH
301 *
302 * This tells the low-level filesystem that a write is
303 * about to be performed to it, and makes sure that
304 * writes are allowed before returning success. When
305 * the write operation is finished, mnt_drop_write()
306 * must be called. This is effectively a refcount.
307 */
83adc753 308int mnt_want_write(struct vfsmount *m)
8366025e 309{
83adc753 310 struct mount *mnt = real_mount(m);
3d733633 311 int ret = 0;
3d733633 312
d3ef3d73 313 preempt_disable();
c6653a83 314 mnt_inc_writers(mnt);
d3ef3d73 315 /*
c6653a83 316 * The store to mnt_inc_writers must be visible before we pass
d3ef3d73 317 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
318 * incremented count after it has set MNT_WRITE_HOLD.
319 */
320 smp_mb();
83adc753 321 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
d3ef3d73 322 cpu_relax();
323 /*
324 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
325 * be set to match its requirements. So we must not load that until
326 * MNT_WRITE_HOLD is cleared.
327 */
328 smp_rmb();
83adc753 329 if (__mnt_is_readonly(m)) {
c6653a83 330 mnt_dec_writers(mnt);
3d733633
DH
331 ret = -EROFS;
332 goto out;
333 }
3d733633 334out:
d3ef3d73 335 preempt_enable();
3d733633 336 return ret;
8366025e
DH
337}
338EXPORT_SYMBOL_GPL(mnt_want_write);
339
96029c4e 340/**
341 * mnt_clone_write - get write access to a mount
342 * @mnt: the mount on which to take a write
343 *
344 * This is effectively like mnt_want_write, except
345 * it must only be used to take an extra write reference
346 * on a mountpoint that we already know has a write reference
347 * on it. This allows some optimisation.
348 *
349 * After finished, mnt_drop_write must be called as usual to
350 * drop the reference.
351 */
352int mnt_clone_write(struct vfsmount *mnt)
353{
354 /* superblock may be r/o */
355 if (__mnt_is_readonly(mnt))
356 return -EROFS;
357 preempt_disable();
83adc753 358 mnt_inc_writers(real_mount(mnt));
96029c4e 359 preempt_enable();
360 return 0;
361}
362EXPORT_SYMBOL_GPL(mnt_clone_write);
363
364/**
365 * mnt_want_write_file - get write access to a file's mount
366 * @file: the file who's mount on which to take a write
367 *
368 * This is like mnt_want_write, but it takes a file and can
369 * do some optimisations if the file is open for write already
370 */
371int mnt_want_write_file(struct file *file)
372{
2d8dd38a
OH
373 struct inode *inode = file->f_dentry->d_inode;
374 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
96029c4e 375 return mnt_want_write(file->f_path.mnt);
376 else
377 return mnt_clone_write(file->f_path.mnt);
378}
379EXPORT_SYMBOL_GPL(mnt_want_write_file);
380
8366025e
DH
381/**
382 * mnt_drop_write - give up write access to a mount
383 * @mnt: the mount on which to give up write access
384 *
385 * Tells the low-level filesystem that we are done
386 * performing writes to it. Must be matched with
387 * mnt_want_write() call above.
388 */
389void mnt_drop_write(struct vfsmount *mnt)
390{
d3ef3d73 391 preempt_disable();
83adc753 392 mnt_dec_writers(real_mount(mnt));
d3ef3d73 393 preempt_enable();
8366025e
DH
394}
395EXPORT_SYMBOL_GPL(mnt_drop_write);
396
2a79f17e
AV
397void mnt_drop_write_file(struct file *file)
398{
399 mnt_drop_write(file->f_path.mnt);
400}
401EXPORT_SYMBOL(mnt_drop_write_file);
402
83adc753 403static int mnt_make_readonly(struct mount *mnt)
8366025e 404{
3d733633
DH
405 int ret = 0;
406
99b7db7b 407 br_write_lock(vfsmount_lock);
83adc753 408 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
3d733633 409 /*
d3ef3d73 410 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
411 * should be visible before we do.
3d733633 412 */
d3ef3d73 413 smp_mb();
414
3d733633 415 /*
d3ef3d73 416 * With writers on hold, if this value is zero, then there are
417 * definitely no active writers (although held writers may subsequently
418 * increment the count, they'll have to wait, and decrement it after
419 * seeing MNT_READONLY).
420 *
421 * It is OK to have counter incremented on one CPU and decremented on
422 * another: the sum will add up correctly. The danger would be when we
423 * sum up each counter, if we read a counter before it is incremented,
424 * but then read another CPU's count which it has been subsequently
425 * decremented from -- we would see more decrements than we should.
426 * MNT_WRITE_HOLD protects against this scenario, because
427 * mnt_want_write first increments count, then smp_mb, then spins on
428 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
429 * we're counting up here.
3d733633 430 */
c6653a83 431 if (mnt_get_writers(mnt) > 0)
d3ef3d73 432 ret = -EBUSY;
433 else
83adc753 434 mnt->mnt.mnt_flags |= MNT_READONLY;
d3ef3d73 435 /*
436 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
437 * that become unheld will see MNT_READONLY.
438 */
439 smp_wmb();
83adc753 440 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
99b7db7b 441 br_write_unlock(vfsmount_lock);
3d733633 442 return ret;
8366025e 443}
8366025e 444
83adc753 445static void __mnt_unmake_readonly(struct mount *mnt)
2e4b7fcd 446{
99b7db7b 447 br_write_lock(vfsmount_lock);
83adc753 448 mnt->mnt.mnt_flags &= ~MNT_READONLY;
99b7db7b 449 br_write_unlock(vfsmount_lock);
2e4b7fcd
DH
450}
451
b105e270 452static void free_vfsmnt(struct mount *mnt)
1da177e4 453{
b105e270 454 kfree(mnt->mnt.mnt_devname);
73cd49ec 455 mnt_free_id(mnt);
d3ef3d73 456#ifdef CONFIG_SMP
68e8a9fe 457 free_percpu(mnt->mnt_pcp);
d3ef3d73 458#endif
b105e270 459 kmem_cache_free(mnt_cache, mnt);
1da177e4
LT
460}
461
462/*
a05964f3
RP
463 * find the first or last mount at @dentry on vfsmount @mnt depending on
464 * @dir. If @dir is set return the first mount else return the last mount.
99b7db7b 465 * vfsmount_lock must be held for read or write.
1da177e4 466 */
c7105365 467struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
a05964f3 468 int dir)
1da177e4 469{
b58fed8b
RP
470 struct list_head *head = mount_hashtable + hash(mnt, dentry);
471 struct list_head *tmp = head;
c7105365 472 struct mount *p, *found = NULL;
1da177e4 473
1da177e4 474 for (;;) {
a05964f3 475 tmp = dir ? tmp->next : tmp->prev;
1da177e4
LT
476 p = NULL;
477 if (tmp == head)
478 break;
1b8e5564 479 p = list_entry(tmp, struct mount, mnt_hash);
a73324da 480 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
a05964f3 481 found = p;
1da177e4
LT
482 break;
483 }
484 }
1da177e4
LT
485 return found;
486}
487
a05964f3
RP
488/*
489 * lookup_mnt increments the ref count before returning
490 * the vfsmount struct.
491 */
1c755af4 492struct vfsmount *lookup_mnt(struct path *path)
a05964f3 493{
c7105365 494 struct mount *child_mnt;
99b7db7b
NP
495
496 br_read_lock(vfsmount_lock);
c7105365
AV
497 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
498 if (child_mnt) {
499 mnt_add_count(child_mnt, 1);
500 br_read_unlock(vfsmount_lock);
501 return &child_mnt->mnt;
502 } else {
503 br_read_unlock(vfsmount_lock);
504 return NULL;
505 }
a05964f3
RP
506}
507
143c8c91 508static inline int check_mnt(struct mount *mnt)
1da177e4 509{
6b3286ed 510 return mnt->mnt_ns == current->nsproxy->mnt_ns;
1da177e4
LT
511}
512
99b7db7b
NP
513/*
514 * vfsmount lock must be held for write
515 */
6b3286ed 516static void touch_mnt_namespace(struct mnt_namespace *ns)
5addc5dd
AV
517{
518 if (ns) {
519 ns->event = ++event;
520 wake_up_interruptible(&ns->poll);
521 }
522}
523
99b7db7b
NP
524/*
525 * vfsmount lock must be held for write
526 */
6b3286ed 527static void __touch_mnt_namespace(struct mnt_namespace *ns)
5addc5dd
AV
528{
529 if (ns && ns->event != event) {
530 ns->event = event;
531 wake_up_interruptible(&ns->poll);
532 }
533}
534
5f57cbcc
NP
535/*
536 * Clear dentry's mounted state if it has no remaining mounts.
537 * vfsmount_lock must be held for write.
538 */
aa0a4cf0 539static void dentry_reset_mounted(struct dentry *dentry)
5f57cbcc
NP
540{
541 unsigned u;
542
543 for (u = 0; u < HASH_SIZE; u++) {
d5e50f74 544 struct mount *p;
5f57cbcc 545
1b8e5564 546 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
a73324da 547 if (p->mnt_mountpoint == dentry)
5f57cbcc
NP
548 return;
549 }
550 }
551 spin_lock(&dentry->d_lock);
552 dentry->d_flags &= ~DCACHE_MOUNTED;
553 spin_unlock(&dentry->d_lock);
554}
555
99b7db7b
NP
556/*
557 * vfsmount lock must be held for write
558 */
419148da
AV
559static void detach_mnt(struct mount *mnt, struct path *old_path)
560{
a73324da 561 old_path->dentry = mnt->mnt_mountpoint;
0714a533
AV
562 old_path->mnt = &mnt->mnt_parent->mnt;
563 mnt->mnt_parent = mnt;
a73324da 564 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
6b41d536 565 list_del_init(&mnt->mnt_child);
1b8e5564 566 list_del_init(&mnt->mnt_hash);
aa0a4cf0 567 dentry_reset_mounted(old_path->dentry);
1da177e4
LT
568}
569
99b7db7b
NP
570/*
571 * vfsmount lock must be held for write
572 */
14cf1fa8 573void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
44d964d6 574 struct mount *child_mnt)
b90fa9ae 575{
14cf1fa8 576 child_mnt->mnt_parent = real_mount(mntget(&mnt->mnt));
a73324da 577 child_mnt->mnt_mountpoint = dget(dentry);
5f57cbcc
NP
578 spin_lock(&dentry->d_lock);
579 dentry->d_flags |= DCACHE_MOUNTED;
580 spin_unlock(&dentry->d_lock);
b90fa9ae
RP
581}
582
99b7db7b
NP
583/*
584 * vfsmount lock must be held for write
585 */
419148da 586static void attach_mnt(struct mount *mnt, struct path *path)
1da177e4 587{
14cf1fa8 588 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
1b8e5564 589 list_add_tail(&mnt->mnt_hash, mount_hashtable +
1a390689 590 hash(path->mnt, path->dentry));
6b41d536 591 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
b90fa9ae
RP
592}
593
83adc753 594static inline void __mnt_make_longterm(struct mount *mnt)
7e3d0eb0
AV
595{
596#ifdef CONFIG_SMP
68e8a9fe 597 atomic_inc(&mnt->mnt_longterm);
7e3d0eb0
AV
598#endif
599}
600
601/* needs vfsmount lock for write */
83adc753 602static inline void __mnt_make_shortterm(struct mount *mnt)
7e3d0eb0
AV
603{
604#ifdef CONFIG_SMP
68e8a9fe 605 atomic_dec(&mnt->mnt_longterm);
7e3d0eb0
AV
606#endif
607}
608
b90fa9ae 609/*
99b7db7b 610 * vfsmount lock must be held for write
b90fa9ae 611 */
4b2619a5 612static void commit_tree(struct mount *mnt)
b90fa9ae 613{
0714a533 614 struct mount *parent = mnt->mnt_parent;
83adc753 615 struct mount *m;
b90fa9ae 616 LIST_HEAD(head);
143c8c91 617 struct mnt_namespace *n = parent->mnt_ns;
b90fa9ae 618
0714a533 619 BUG_ON(parent == mnt);
b90fa9ae 620
4b2619a5 621 list_add_tail(&head, &mnt->mnt.mnt_list);
83adc753 622 list_for_each_entry(m, &head, mnt.mnt_list) {
143c8c91 623 m->mnt_ns = n;
7e3d0eb0 624 __mnt_make_longterm(m);
f03c6599
AV
625 }
626
b90fa9ae
RP
627 list_splice(&head, n->list.prev);
628
1b8e5564 629 list_add_tail(&mnt->mnt_hash, mount_hashtable +
a73324da 630 hash(&parent->mnt, mnt->mnt_mountpoint));
6b41d536 631 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
6b3286ed 632 touch_mnt_namespace(n);
1da177e4
LT
633}
634
315fc83e 635static struct mount *next_mnt(struct mount *p, struct vfsmount *root)
1da177e4 636{
6b41d536
AV
637 struct list_head *next = p->mnt_mounts.next;
638 if (next == &p->mnt_mounts) {
1da177e4 639 while (1) {
315fc83e 640 if (&p->mnt == root)
1da177e4 641 return NULL;
6b41d536
AV
642 next = p->mnt_child.next;
643 if (next != &p->mnt_parent->mnt_mounts)
1da177e4 644 break;
0714a533 645 p = p->mnt_parent;
1da177e4
LT
646 }
647 }
6b41d536 648 return list_entry(next, struct mount, mnt_child);
1da177e4
LT
649}
650
315fc83e 651static struct mount *skip_mnt_tree(struct mount *p)
9676f0c6 652{
6b41d536
AV
653 struct list_head *prev = p->mnt_mounts.prev;
654 while (prev != &p->mnt_mounts) {
655 p = list_entry(prev, struct mount, mnt_child);
656 prev = p->mnt_mounts.prev;
9676f0c6
RP
657 }
658 return p;
659}
660
9d412a43
AV
661struct vfsmount *
662vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
663{
b105e270 664 struct mount *mnt;
9d412a43
AV
665 struct dentry *root;
666
667 if (!type)
668 return ERR_PTR(-ENODEV);
669
670 mnt = alloc_vfsmnt(name);
671 if (!mnt)
672 return ERR_PTR(-ENOMEM);
673
674 if (flags & MS_KERNMOUNT)
b105e270 675 mnt->mnt.mnt_flags = MNT_INTERNAL;
9d412a43
AV
676
677 root = mount_fs(type, flags, name, data);
678 if (IS_ERR(root)) {
679 free_vfsmnt(mnt);
680 return ERR_CAST(root);
681 }
682
b105e270
AV
683 mnt->mnt.mnt_root = root;
684 mnt->mnt.mnt_sb = root->d_sb;
a73324da 685 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
0714a533 686 mnt->mnt_parent = mnt;
b105e270 687 return &mnt->mnt;
9d412a43
AV
688}
689EXPORT_SYMBOL_GPL(vfs_kern_mount);
690
87129cc0 691static struct mount *clone_mnt(struct mount *old, struct dentry *root,
36341f64 692 int flag)
1da177e4 693{
87129cc0
AV
694 struct super_block *sb = old->mnt.mnt_sb;
695 struct mount *mnt = alloc_vfsmnt(old->mnt.mnt_devname);
1da177e4
LT
696
697 if (mnt) {
719f5d7f 698 if (flag & (CL_SLAVE | CL_PRIVATE))
b105e270 699 mnt->mnt.mnt_group_id = 0; /* not a peer of original */
719f5d7f 700 else
87129cc0 701 mnt->mnt.mnt_group_id = old->mnt.mnt_group_id;
719f5d7f 702
b105e270
AV
703 if ((flag & CL_MAKE_SHARED) && !mnt->mnt.mnt_group_id) {
704 int err = mnt_alloc_group_id(mnt);
719f5d7f
MS
705 if (err)
706 goto out_free;
707 }
708
87129cc0 709 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
1da177e4 710 atomic_inc(&sb->s_active);
b105e270
AV
711 mnt->mnt.mnt_sb = sb;
712 mnt->mnt.mnt_root = dget(root);
a73324da 713 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
0714a533 714 mnt->mnt_parent = mnt;
b90fa9ae 715
5afe0022 716 if (flag & CL_SLAVE) {
6776db3d 717 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
32301920 718 mnt->mnt_master = old;
b105e270 719 CLEAR_MNT_SHARED(&mnt->mnt);
8aec0809 720 } else if (!(flag & CL_PRIVATE)) {
87129cc0 721 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(&old->mnt))
6776db3d 722 list_add(&mnt->mnt_share, &old->mnt_share);
d10e8def 723 if (IS_MNT_SLAVE(old))
6776db3d 724 list_add(&mnt->mnt_slave, &old->mnt_slave);
d10e8def 725 mnt->mnt_master = old->mnt_master;
5afe0022 726 }
b90fa9ae 727 if (flag & CL_MAKE_SHARED)
0f0afb1d 728 set_mnt_shared(mnt);
1da177e4
LT
729
730 /* stick the duplicate mount on the same expiry list
731 * as the original if that was on one */
36341f64 732 if (flag & CL_EXPIRE) {
6776db3d
AV
733 if (!list_empty(&old->mnt_expire))
734 list_add(&mnt->mnt_expire, &old->mnt_expire);
36341f64 735 }
1da177e4 736 }
cb338d06 737 return mnt;
719f5d7f
MS
738
739 out_free:
740 free_vfsmnt(mnt);
741 return NULL;
1da177e4
LT
742}
743
83adc753 744static inline void mntfree(struct mount *mnt)
1da177e4 745{
83adc753
AV
746 struct vfsmount *m = &mnt->mnt;
747 struct super_block *sb = m->mnt_sb;
b3e19d92 748
3d733633
DH
749 /*
750 * This probably indicates that somebody messed
751 * up a mnt_want/drop_write() pair. If this
752 * happens, the filesystem was probably unable
753 * to make r/w->r/o transitions.
754 */
d3ef3d73 755 /*
b3e19d92
NP
756 * The locking used to deal with mnt_count decrement provides barriers,
757 * so mnt_get_writers() below is safe.
d3ef3d73 758 */
c6653a83 759 WARN_ON(mnt_get_writers(mnt));
83adc753
AV
760 fsnotify_vfsmount_delete(m);
761 dput(m->mnt_root);
762 free_vfsmnt(mnt);
1da177e4
LT
763 deactivate_super(sb);
764}
765
900148dc 766static void mntput_no_expire(struct mount *mnt)
b3e19d92 767{
b3e19d92 768put_again:
f03c6599
AV
769#ifdef CONFIG_SMP
770 br_read_lock(vfsmount_lock);
68e8a9fe 771 if (likely(atomic_read(&mnt->mnt_longterm))) {
aa9c0e07 772 mnt_add_count(mnt, -1);
b3e19d92 773 br_read_unlock(vfsmount_lock);
f03c6599 774 return;
b3e19d92 775 }
f03c6599 776 br_read_unlock(vfsmount_lock);
b3e19d92 777
99b7db7b 778 br_write_lock(vfsmount_lock);
aa9c0e07 779 mnt_add_count(mnt, -1);
b3e19d92 780 if (mnt_get_count(mnt)) {
99b7db7b
NP
781 br_write_unlock(vfsmount_lock);
782 return;
783 }
b3e19d92 784#else
aa9c0e07 785 mnt_add_count(mnt, -1);
b3e19d92 786 if (likely(mnt_get_count(mnt)))
99b7db7b 787 return;
b3e19d92 788 br_write_lock(vfsmount_lock);
f03c6599 789#endif
83adc753
AV
790 if (unlikely(mnt->mnt.mnt_pinned)) {
791 mnt_add_count(mnt, mnt->mnt.mnt_pinned + 1);
792 mnt->mnt.mnt_pinned = 0;
b3e19d92 793 br_write_unlock(vfsmount_lock);
900148dc 794 acct_auto_close_mnt(&mnt->mnt);
b3e19d92 795 goto put_again;
7b7b1ace 796 }
99b7db7b 797 br_write_unlock(vfsmount_lock);
b3e19d92
NP
798 mntfree(mnt);
799}
b3e19d92
NP
800
801void mntput(struct vfsmount *mnt)
802{
803 if (mnt) {
804 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
805 if (unlikely(mnt->mnt_expiry_mark))
806 mnt->mnt_expiry_mark = 0;
900148dc 807 mntput_no_expire(real_mount(mnt));
b3e19d92
NP
808 }
809}
810EXPORT_SYMBOL(mntput);
811
812struct vfsmount *mntget(struct vfsmount *mnt)
813{
814 if (mnt)
83adc753 815 mnt_add_count(real_mount(mnt), 1);
b3e19d92
NP
816 return mnt;
817}
818EXPORT_SYMBOL(mntget);
819
7b7b1ace
AV
820void mnt_pin(struct vfsmount *mnt)
821{
99b7db7b 822 br_write_lock(vfsmount_lock);
7b7b1ace 823 mnt->mnt_pinned++;
99b7db7b 824 br_write_unlock(vfsmount_lock);
7b7b1ace 825}
7b7b1ace
AV
826EXPORT_SYMBOL(mnt_pin);
827
828void mnt_unpin(struct vfsmount *mnt)
829{
99b7db7b 830 br_write_lock(vfsmount_lock);
7b7b1ace 831 if (mnt->mnt_pinned) {
83adc753 832 mnt_add_count(real_mount(mnt), 1);
7b7b1ace
AV
833 mnt->mnt_pinned--;
834 }
99b7db7b 835 br_write_unlock(vfsmount_lock);
7b7b1ace 836}
7b7b1ace 837EXPORT_SYMBOL(mnt_unpin);
1da177e4 838
b3b304a2
MS
839static inline void mangle(struct seq_file *m, const char *s)
840{
841 seq_escape(m, s, " \t\n\\");
842}
843
844/*
845 * Simple .show_options callback for filesystems which don't want to
846 * implement more complex mount option showing.
847 *
848 * See also save_mount_options().
849 */
850int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
851{
2a32cebd
AV
852 const char *options;
853
854 rcu_read_lock();
855 options = rcu_dereference(mnt->mnt_sb->s_options);
b3b304a2
MS
856
857 if (options != NULL && options[0]) {
858 seq_putc(m, ',');
859 mangle(m, options);
860 }
2a32cebd 861 rcu_read_unlock();
b3b304a2
MS
862
863 return 0;
864}
865EXPORT_SYMBOL(generic_show_options);
866
867/*
868 * If filesystem uses generic_show_options(), this function should be
869 * called from the fill_super() callback.
870 *
871 * The .remount_fs callback usually needs to be handled in a special
872 * way, to make sure, that previous options are not overwritten if the
873 * remount fails.
874 *
875 * Also note, that if the filesystem's .remount_fs function doesn't
876 * reset all options to their default value, but changes only newly
877 * given options, then the displayed options will not reflect reality
878 * any more.
879 */
880void save_mount_options(struct super_block *sb, char *options)
881{
2a32cebd
AV
882 BUG_ON(sb->s_options);
883 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
b3b304a2
MS
884}
885EXPORT_SYMBOL(save_mount_options);
886
2a32cebd
AV
887void replace_mount_options(struct super_block *sb, char *options)
888{
889 char *old = sb->s_options;
890 rcu_assign_pointer(sb->s_options, options);
891 if (old) {
892 synchronize_rcu();
893 kfree(old);
894 }
895}
896EXPORT_SYMBOL(replace_mount_options);
897
a1a2c409 898#ifdef CONFIG_PROC_FS
1da177e4
LT
899/* iterator */
900static void *m_start(struct seq_file *m, loff_t *pos)
901{
a1a2c409 902 struct proc_mounts *p = m->private;
1da177e4 903
390c6843 904 down_read(&namespace_sem);
a1a2c409 905 return seq_list_start(&p->ns->list, *pos);
1da177e4
LT
906}
907
908static void *m_next(struct seq_file *m, void *v, loff_t *pos)
909{
a1a2c409 910 struct proc_mounts *p = m->private;
b0765fb8 911
a1a2c409 912 return seq_list_next(v, &p->ns->list, pos);
1da177e4
LT
913}
914
915static void m_stop(struct seq_file *m, void *v)
916{
390c6843 917 up_read(&namespace_sem);
1da177e4
LT
918}
919
9f5596af
AV
920int mnt_had_events(struct proc_mounts *p)
921{
922 struct mnt_namespace *ns = p->ns;
923 int res = 0;
924
99b7db7b 925 br_read_lock(vfsmount_lock);
f1514638
KS
926 if (p->m.poll_event != ns->event) {
927 p->m.poll_event = ns->event;
9f5596af
AV
928 res = 1;
929 }
99b7db7b 930 br_read_unlock(vfsmount_lock);
9f5596af
AV
931
932 return res;
933}
934
2d4d4864
RP
935struct proc_fs_info {
936 int flag;
937 const char *str;
938};
939
2069f457 940static int show_sb_opts(struct seq_file *m, struct super_block *sb)
1da177e4 941{
2d4d4864 942 static const struct proc_fs_info fs_info[] = {
1da177e4
LT
943 { MS_SYNCHRONOUS, ",sync" },
944 { MS_DIRSYNC, ",dirsync" },
945 { MS_MANDLOCK, ",mand" },
1da177e4
LT
946 { 0, NULL }
947 };
2d4d4864
RP
948 const struct proc_fs_info *fs_infop;
949
950 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
951 if (sb->s_flags & fs_infop->flag)
952 seq_puts(m, fs_infop->str);
953 }
2069f457
EP
954
955 return security_sb_show_options(m, sb);
2d4d4864
RP
956}
957
958static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
959{
960 static const struct proc_fs_info mnt_info[] = {
1da177e4
LT
961 { MNT_NOSUID, ",nosuid" },
962 { MNT_NODEV, ",nodev" },
963 { MNT_NOEXEC, ",noexec" },
fc33a7bb
CH
964 { MNT_NOATIME, ",noatime" },
965 { MNT_NODIRATIME, ",nodiratime" },
47ae32d6 966 { MNT_RELATIME, ",relatime" },
1da177e4
LT
967 { 0, NULL }
968 };
2d4d4864
RP
969 const struct proc_fs_info *fs_infop;
970
971 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
972 if (mnt->mnt_flags & fs_infop->flag)
973 seq_puts(m, fs_infop->str);
974 }
975}
976
977static void show_type(struct seq_file *m, struct super_block *sb)
978{
979 mangle(m, sb->s_type->name);
980 if (sb->s_subtype && sb->s_subtype[0]) {
981 seq_putc(m, '.');
982 mangle(m, sb->s_subtype);
983 }
984}
985
986static int show_vfsmnt(struct seq_file *m, void *v)
987{
988 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
989 int err = 0;
c32c2f63 990 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1da177e4 991
c7f404b4
AV
992 if (mnt->mnt_sb->s_op->show_devname) {
993 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
994 if (err)
995 goto out;
996 } else {
997 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
998 }
1da177e4 999 seq_putc(m, ' ');
c32c2f63 1000 seq_path(m, &mnt_path, " \t\n\\");
1da177e4 1001 seq_putc(m, ' ');
2d4d4864 1002 show_type(m, mnt->mnt_sb);
2e4b7fcd 1003 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
2069f457
EP
1004 err = show_sb_opts(m, mnt->mnt_sb);
1005 if (err)
1006 goto out;
2d4d4864 1007 show_mnt_opts(m, mnt);
1da177e4
LT
1008 if (mnt->mnt_sb->s_op->show_options)
1009 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1010 seq_puts(m, " 0 0\n");
2069f457 1011out:
1da177e4
LT
1012 return err;
1013}
1014
a1a2c409 1015const struct seq_operations mounts_op = {
1da177e4
LT
1016 .start = m_start,
1017 .next = m_next,
1018 .stop = m_stop,
1019 .show = show_vfsmnt
1020};
1021
2d4d4864
RP
1022static int show_mountinfo(struct seq_file *m, void *v)
1023{
1024 struct proc_mounts *p = m->private;
1025 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
3376f34f 1026 struct mount *r = real_mount(mnt);
2d4d4864
RP
1027 struct super_block *sb = mnt->mnt_sb;
1028 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1029 struct path root = p->root;
1030 int err = 0;
1031
0714a533 1032 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, r->mnt_parent->mnt.mnt_id,
2d4d4864 1033 MAJOR(sb->s_dev), MINOR(sb->s_dev));
c7f404b4
AV
1034 if (sb->s_op->show_path)
1035 err = sb->s_op->show_path(m, mnt);
1036 else
1037 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1038 if (err)
1039 goto out;
2d4d4864 1040 seq_putc(m, ' ');
02125a82
AV
1041
1042 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1043 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1044 if (err)
1045 goto out;
1046
2d4d4864
RP
1047 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1048 show_mnt_opts(m, mnt);
1049
1050 /* Tagged fields ("foo:X" or "bar") */
1051 if (IS_MNT_SHARED(mnt))
1052 seq_printf(m, " shared:%i", mnt->mnt_group_id);
d10e8def 1053 if (IS_MNT_SLAVE(r)) {
32301920 1054 int master = r->mnt_master->mnt.mnt_group_id;
6fc7871f 1055 int dom = get_dominating_id(r, &p->root);
97e7e0f7
MS
1056 seq_printf(m, " master:%i", master);
1057 if (dom && dom != master)
1058 seq_printf(m, " propagate_from:%i", dom);
1059 }
2d4d4864
RP
1060 if (IS_MNT_UNBINDABLE(mnt))
1061 seq_puts(m, " unbindable");
1062
1063 /* Filesystem specific data */
1064 seq_puts(m, " - ");
1065 show_type(m, sb);
1066 seq_putc(m, ' ');
c7f404b4
AV
1067 if (sb->s_op->show_devname)
1068 err = sb->s_op->show_devname(m, mnt);
1069 else
1070 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1071 if (err)
1072 goto out;
2d4d4864 1073 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
2069f457
EP
1074 err = show_sb_opts(m, sb);
1075 if (err)
1076 goto out;
2d4d4864
RP
1077 if (sb->s_op->show_options)
1078 err = sb->s_op->show_options(m, mnt);
1079 seq_putc(m, '\n');
2069f457 1080out:
2d4d4864
RP
1081 return err;
1082}
1083
1084const struct seq_operations mountinfo_op = {
1085 .start = m_start,
1086 .next = m_next,
1087 .stop = m_stop,
1088 .show = show_mountinfo,
1089};
1090
b4629fe2
CL
1091static int show_vfsstat(struct seq_file *m, void *v)
1092{
b0765fb8 1093 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
c32c2f63 1094 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
b4629fe2
CL
1095 int err = 0;
1096
1097 /* device */
c7f404b4 1098 if (mnt->mnt_sb->s_op->show_devname) {
a877ee03 1099 seq_puts(m, "device ");
c7f404b4
AV
1100 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1101 } else {
1102 if (mnt->mnt_devname) {
1103 seq_puts(m, "device ");
1104 mangle(m, mnt->mnt_devname);
1105 } else
1106 seq_puts(m, "no device");
1107 }
b4629fe2
CL
1108
1109 /* mount point */
1110 seq_puts(m, " mounted on ");
c32c2f63 1111 seq_path(m, &mnt_path, " \t\n\\");
b4629fe2
CL
1112 seq_putc(m, ' ');
1113
1114 /* file system type */
1115 seq_puts(m, "with fstype ");
2d4d4864 1116 show_type(m, mnt->mnt_sb);
b4629fe2
CL
1117
1118 /* optional statistics */
1119 if (mnt->mnt_sb->s_op->show_stats) {
1120 seq_putc(m, ' ');
c7f404b4
AV
1121 if (!err)
1122 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
b4629fe2
CL
1123 }
1124
1125 seq_putc(m, '\n');
1126 return err;
1127}
1128
a1a2c409 1129const struct seq_operations mountstats_op = {
b4629fe2
CL
1130 .start = m_start,
1131 .next = m_next,
1132 .stop = m_stop,
1133 .show = show_vfsstat,
1134};
a1a2c409 1135#endif /* CONFIG_PROC_FS */
b4629fe2 1136
1da177e4
LT
1137/**
1138 * may_umount_tree - check if a mount tree is busy
1139 * @mnt: root of mount tree
1140 *
1141 * This is called to check if a tree of mounts has any
1142 * open files, pwds, chroots or sub mounts that are
1143 * busy.
1144 */
1145int may_umount_tree(struct vfsmount *mnt)
1146{
36341f64
RP
1147 int actual_refs = 0;
1148 int minimum_refs = 0;
315fc83e
AV
1149 struct mount *p;
1150 BUG_ON(!mnt);
1da177e4 1151
b3e19d92
NP
1152 /* write lock needed for mnt_get_count */
1153 br_write_lock(vfsmount_lock);
315fc83e 1154 for (p = real_mount(mnt); p; p = next_mnt(p, mnt)) {
83adc753 1155 actual_refs += mnt_get_count(p);
1da177e4 1156 minimum_refs += 2;
1da177e4 1157 }
b3e19d92 1158 br_write_unlock(vfsmount_lock);
1da177e4
LT
1159
1160 if (actual_refs > minimum_refs)
e3474a8e 1161 return 0;
1da177e4 1162
e3474a8e 1163 return 1;
1da177e4
LT
1164}
1165
1166EXPORT_SYMBOL(may_umount_tree);
1167
1168/**
1169 * may_umount - check if a mount point is busy
1170 * @mnt: root of mount
1171 *
1172 * This is called to check if a mount point has any
1173 * open files, pwds, chroots or sub mounts. If the
1174 * mount has sub mounts this will return busy
1175 * regardless of whether the sub mounts are busy.
1176 *
1177 * Doesn't take quota and stuff into account. IOW, in some cases it will
1178 * give false negatives. The main reason why it's here is that we need
1179 * a non-destructive way to look for easily umountable filesystems.
1180 */
1181int may_umount(struct vfsmount *mnt)
1182{
e3474a8e 1183 int ret = 1;
8ad08d8a 1184 down_read(&namespace_sem);
b3e19d92 1185 br_write_lock(vfsmount_lock);
1ab59738 1186 if (propagate_mount_busy(real_mount(mnt), 2))
e3474a8e 1187 ret = 0;
b3e19d92 1188 br_write_unlock(vfsmount_lock);
8ad08d8a 1189 up_read(&namespace_sem);
a05964f3 1190 return ret;
1da177e4
LT
1191}
1192
1193EXPORT_SYMBOL(may_umount);
1194
b90fa9ae 1195void release_mounts(struct list_head *head)
70fbcdf4 1196{
d5e50f74 1197 struct mount *mnt;
bf066c7d 1198 while (!list_empty(head)) {
1b8e5564
AV
1199 mnt = list_first_entry(head, struct mount, mnt_hash);
1200 list_del_init(&mnt->mnt_hash);
676da58d 1201 if (mnt_has_parent(mnt)) {
70fbcdf4
RP
1202 struct dentry *dentry;
1203 struct vfsmount *m;
99b7db7b
NP
1204
1205 br_write_lock(vfsmount_lock);
a73324da 1206 dentry = mnt->mnt_mountpoint;
0714a533 1207 m = &mnt->mnt_parent->mnt;
a73324da 1208 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
0714a533 1209 mnt->mnt_parent = mnt;
7c4b93d8 1210 m->mnt_ghosts--;
99b7db7b 1211 br_write_unlock(vfsmount_lock);
70fbcdf4
RP
1212 dput(dentry);
1213 mntput(m);
1214 }
d5e50f74 1215 mntput(&mnt->mnt);
70fbcdf4
RP
1216 }
1217}
1218
99b7db7b
NP
1219/*
1220 * vfsmount lock must be held for write
1221 * namespace_sem must be held for write
1222 */
761d5c38 1223void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1da177e4 1224{
7b8a53fd 1225 LIST_HEAD(tmp_list);
315fc83e 1226 struct mount *p;
1da177e4 1227
761d5c38 1228 for (p = mnt; p; p = next_mnt(p, &mnt->mnt))
1b8e5564 1229 list_move(&p->mnt_hash, &tmp_list);
1da177e4 1230
a05964f3 1231 if (propagate)
7b8a53fd 1232 propagate_umount(&tmp_list);
a05964f3 1233
1b8e5564 1234 list_for_each_entry(p, &tmp_list, mnt_hash) {
6776db3d 1235 list_del_init(&p->mnt_expire);
315fc83e 1236 list_del_init(&p->mnt.mnt_list);
143c8c91
AV
1237 __touch_mnt_namespace(p->mnt_ns);
1238 p->mnt_ns = NULL;
83adc753 1239 __mnt_make_shortterm(p);
6b41d536 1240 list_del_init(&p->mnt_child);
676da58d 1241 if (mnt_has_parent(p)) {
0714a533 1242 p->mnt_parent->mnt.mnt_ghosts++;
a73324da 1243 dentry_reset_mounted(p->mnt_mountpoint);
7c4b93d8 1244 }
0f0afb1d 1245 change_mnt_propagation(p, MS_PRIVATE);
1da177e4 1246 }
7b8a53fd 1247 list_splice(&tmp_list, kill);
1da177e4
LT
1248}
1249
692afc31 1250static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
c35038be 1251
1ab59738 1252static int do_umount(struct mount *mnt, int flags)
1da177e4 1253{
1ab59738 1254 struct super_block *sb = mnt->mnt.mnt_sb;
1da177e4 1255 int retval;
70fbcdf4 1256 LIST_HEAD(umount_list);
1da177e4 1257
1ab59738 1258 retval = security_sb_umount(&mnt->mnt, flags);
1da177e4
LT
1259 if (retval)
1260 return retval;
1261
1262 /*
1263 * Allow userspace to request a mountpoint be expired rather than
1264 * unmounting unconditionally. Unmount only happens if:
1265 * (1) the mark is already set (the mark is cleared by mntput())
1266 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1267 */
1268 if (flags & MNT_EXPIRE) {
1ab59738 1269 if (&mnt->mnt == current->fs->root.mnt ||
1da177e4
LT
1270 flags & (MNT_FORCE | MNT_DETACH))
1271 return -EINVAL;
1272
b3e19d92
NP
1273 /*
1274 * probably don't strictly need the lock here if we examined
1275 * all race cases, but it's a slowpath.
1276 */
1277 br_write_lock(vfsmount_lock);
83adc753 1278 if (mnt_get_count(mnt) != 2) {
bf9faa2a 1279 br_write_unlock(vfsmount_lock);
1da177e4 1280 return -EBUSY;
b3e19d92
NP
1281 }
1282 br_write_unlock(vfsmount_lock);
1da177e4 1283
1ab59738 1284 if (!xchg(&mnt->mnt.mnt_expiry_mark, 1))
1da177e4
LT
1285 return -EAGAIN;
1286 }
1287
1288 /*
1289 * If we may have to abort operations to get out of this
1290 * mount, and they will themselves hold resources we must
1291 * allow the fs to do things. In the Unix tradition of
1292 * 'Gee thats tricky lets do it in userspace' the umount_begin
1293 * might fail to complete on the first run through as other tasks
1294 * must return, and the like. Thats for the mount program to worry
1295 * about for the moment.
1296 */
1297
42faad99 1298 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
42faad99 1299 sb->s_op->umount_begin(sb);
42faad99 1300 }
1da177e4
LT
1301
1302 /*
1303 * No sense to grab the lock for this test, but test itself looks
1304 * somewhat bogus. Suggestions for better replacement?
1305 * Ho-hum... In principle, we might treat that as umount + switch
1306 * to rootfs. GC would eventually take care of the old vfsmount.
1307 * Actually it makes sense, especially if rootfs would contain a
1308 * /reboot - static binary that would close all descriptors and
1309 * call reboot(9). Then init(8) could umount root and exec /reboot.
1310 */
1ab59738 1311 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1da177e4
LT
1312 /*
1313 * Special case for "unmounting" root ...
1314 * we just try to remount it readonly.
1315 */
1316 down_write(&sb->s_umount);
4aa98cf7 1317 if (!(sb->s_flags & MS_RDONLY))
1da177e4 1318 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1da177e4
LT
1319 up_write(&sb->s_umount);
1320 return retval;
1321 }
1322
390c6843 1323 down_write(&namespace_sem);
99b7db7b 1324 br_write_lock(vfsmount_lock);
5addc5dd 1325 event++;
1da177e4 1326
c35038be 1327 if (!(flags & MNT_DETACH))
1ab59738 1328 shrink_submounts(mnt, &umount_list);
c35038be 1329
1da177e4 1330 retval = -EBUSY;
a05964f3 1331 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1ab59738
AV
1332 if (!list_empty(&mnt->mnt.mnt_list))
1333 umount_tree(mnt, 1, &umount_list);
1da177e4
LT
1334 retval = 0;
1335 }
99b7db7b 1336 br_write_unlock(vfsmount_lock);
390c6843 1337 up_write(&namespace_sem);
70fbcdf4 1338 release_mounts(&umount_list);
1da177e4
LT
1339 return retval;
1340}
1341
1342/*
1343 * Now umount can handle mount points as well as block devices.
1344 * This is important for filesystems which use unnamed block devices.
1345 *
1346 * We now support a flag for forced unmount like the other 'big iron'
1347 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1348 */
1349
bdc480e3 1350SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1da177e4 1351{
2d8f3038 1352 struct path path;
900148dc 1353 struct mount *mnt;
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;
900148dc 1366 mnt = real_mount(path.mnt);
1da177e4 1367 retval = -EINVAL;
2d8f3038 1368 if (path.dentry != path.mnt->mnt_root)
1da177e4 1369 goto dput_and_out;
143c8c91 1370 if (!check_mnt(mnt))
1da177e4
LT
1371 goto dput_and_out;
1372
1373 retval = -EPERM;
1374 if (!capable(CAP_SYS_ADMIN))
1375 goto dput_and_out;
1376
900148dc 1377 retval = do_umount(mnt, flags);
1da177e4 1378dput_and_out:
429731b1 1379 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
2d8f3038 1380 dput(path.dentry);
900148dc 1381 mntput_no_expire(mnt);
1da177e4
LT
1382out:
1383 return retval;
1384}
1385
1386#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1387
1388/*
b58fed8b 1389 * The 2.0 compatible umount. No flags.
1da177e4 1390 */
bdc480e3 1391SYSCALL_DEFINE1(oldumount, char __user *, name)
1da177e4 1392{
b58fed8b 1393 return sys_umount(name, 0);
1da177e4
LT
1394}
1395
1396#endif
1397
2d92ab3c 1398static int mount_is_safe(struct path *path)
1da177e4
LT
1399{
1400 if (capable(CAP_SYS_ADMIN))
1401 return 0;
1402 return -EPERM;
1403#ifdef notyet
2d92ab3c 1404 if (S_ISLNK(path->dentry->d_inode->i_mode))
1da177e4 1405 return -EPERM;
2d92ab3c 1406 if (path->dentry->d_inode->i_mode & S_ISVTX) {
da9592ed 1407 if (current_uid() != path->dentry->d_inode->i_uid)
1da177e4
LT
1408 return -EPERM;
1409 }
2d92ab3c 1410 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1da177e4
LT
1411 return -EPERM;
1412 return 0;
1413#endif
1414}
1415
87129cc0 1416struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
36341f64 1417 int flag)
1da177e4 1418{
a73324da 1419 struct mount *res, *p, *q, *r;
1a390689 1420 struct path path;
1da177e4 1421
87129cc0 1422 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(&mnt->mnt))
9676f0c6
RP
1423 return NULL;
1424
36341f64 1425 res = q = clone_mnt(mnt, dentry, flag);
1da177e4
LT
1426 if (!q)
1427 goto Enomem;
a73324da 1428 q->mnt_mountpoint = mnt->mnt_mountpoint;
1da177e4
LT
1429
1430 p = mnt;
6b41d536 1431 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
315fc83e 1432 struct mount *s;
7ec02ef1 1433 if (!is_subdir(r->mnt_mountpoint, dentry))
1da177e4
LT
1434 continue;
1435
a73324da 1436 for (s = r; s; s = next_mnt(s, &r->mnt)) {
315fc83e 1437 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(&s->mnt)) {
9676f0c6
RP
1438 s = skip_mnt_tree(s);
1439 continue;
1440 }
0714a533
AV
1441 while (p != s->mnt_parent) {
1442 p = p->mnt_parent;
1443 q = q->mnt_parent;
1da177e4 1444 }
87129cc0 1445 p = s;
cb338d06 1446 path.mnt = &q->mnt;
a73324da 1447 path.dentry = p->mnt_mountpoint;
87129cc0 1448 q = clone_mnt(p, p->mnt.mnt_root, flag);
1da177e4
LT
1449 if (!q)
1450 goto Enomem;
99b7db7b 1451 br_write_lock(vfsmount_lock);
cb338d06
AV
1452 list_add_tail(&q->mnt.mnt_list, &res->mnt.mnt_list);
1453 attach_mnt(q, &path);
99b7db7b 1454 br_write_unlock(vfsmount_lock);
1da177e4
LT
1455 }
1456 }
1457 return res;
b58fed8b 1458Enomem:
1da177e4 1459 if (res) {
70fbcdf4 1460 LIST_HEAD(umount_list);
99b7db7b 1461 br_write_lock(vfsmount_lock);
761d5c38 1462 umount_tree(res, 0, &umount_list);
99b7db7b 1463 br_write_unlock(vfsmount_lock);
70fbcdf4 1464 release_mounts(&umount_list);
1da177e4
LT
1465 }
1466 return NULL;
1467}
1468
589ff870 1469struct vfsmount *collect_mounts(struct path *path)
8aec0809 1470{
cb338d06 1471 struct mount *tree;
1a60a280 1472 down_write(&namespace_sem);
87129cc0
AV
1473 tree = copy_tree(real_mount(path->mnt), path->dentry,
1474 CL_COPY_ALL | CL_PRIVATE);
1a60a280 1475 up_write(&namespace_sem);
cb338d06 1476 return tree ? &tree->mnt : NULL;
8aec0809
AV
1477}
1478
1479void drop_collected_mounts(struct vfsmount *mnt)
1480{
1481 LIST_HEAD(umount_list);
1a60a280 1482 down_write(&namespace_sem);
99b7db7b 1483 br_write_lock(vfsmount_lock);
761d5c38 1484 umount_tree(real_mount(mnt), 0, &umount_list);
99b7db7b 1485 br_write_unlock(vfsmount_lock);
1a60a280 1486 up_write(&namespace_sem);
8aec0809
AV
1487 release_mounts(&umount_list);
1488}
1489
1f707137
AV
1490int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1491 struct vfsmount *root)
1492{
1493 struct vfsmount *mnt;
1494 int res = f(root, arg);
1495 if (res)
1496 return res;
1497 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1498 res = f(mnt, arg);
1499 if (res)
1500 return res;
1501 }
1502 return 0;
1503}
1504
4b8b21f4 1505static void cleanup_group_ids(struct mount *mnt, struct mount *end)
719f5d7f 1506{
315fc83e 1507 struct mount *p;
719f5d7f 1508
4b8b21f4 1509 for (p = mnt; p != end; p = next_mnt(p, &mnt->mnt)) {
315fc83e 1510 if (p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt))
4b8b21f4 1511 mnt_release_group_id(p);
719f5d7f
MS
1512 }
1513}
1514
4b8b21f4 1515static int invent_group_ids(struct mount *mnt, bool recurse)
719f5d7f 1516{
315fc83e 1517 struct mount *p;
719f5d7f 1518
4b8b21f4 1519 for (p = mnt; p; p = recurse ? next_mnt(p, &mnt->mnt) : NULL) {
315fc83e 1520 if (!p->mnt.mnt_group_id && !IS_MNT_SHARED(&p->mnt)) {
4b8b21f4 1521 int err = mnt_alloc_group_id(p);
719f5d7f 1522 if (err) {
4b8b21f4 1523 cleanup_group_ids(mnt, p);
719f5d7f
MS
1524 return err;
1525 }
1526 }
1527 }
1528
1529 return 0;
1530}
1531
b90fa9ae
RP
1532/*
1533 * @source_mnt : mount tree to be attached
21444403
RP
1534 * @nd : place the mount tree @source_mnt is attached
1535 * @parent_nd : if non-null, detach the source_mnt from its parent and
1536 * store the parent mount and mountpoint dentry.
1537 * (done when source_mnt is moved)
b90fa9ae
RP
1538 *
1539 * NOTE: in the table below explains the semantics when a source mount
1540 * of a given type is attached to a destination mount of a given type.
9676f0c6
RP
1541 * ---------------------------------------------------------------------------
1542 * | BIND MOUNT OPERATION |
1543 * |**************************************************************************
1544 * | source-->| shared | private | slave | unbindable |
1545 * | dest | | | | |
1546 * | | | | | | |
1547 * | v | | | | |
1548 * |**************************************************************************
1549 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1550 * | | | | | |
1551 * |non-shared| shared (+) | private | slave (*) | invalid |
1552 * ***************************************************************************
b90fa9ae
RP
1553 * A bind operation clones the source mount and mounts the clone on the
1554 * destination mount.
1555 *
1556 * (++) the cloned mount is propagated to all the mounts in the propagation
1557 * tree of the destination mount and the cloned mount is added to
1558 * the peer group of the source mount.
1559 * (+) the cloned mount is created under the destination mount and is marked
1560 * as shared. The cloned mount is added to the peer group of the source
1561 * mount.
5afe0022
RP
1562 * (+++) the mount is propagated to all the mounts in the propagation tree
1563 * of the destination mount and the cloned mount is made slave
1564 * of the same master as that of the source mount. The cloned mount
1565 * is marked as 'shared and slave'.
1566 * (*) the cloned mount is made a slave of the same master as that of the
1567 * source mount.
1568 *
9676f0c6
RP
1569 * ---------------------------------------------------------------------------
1570 * | MOVE MOUNT OPERATION |
1571 * |**************************************************************************
1572 * | source-->| shared | private | slave | unbindable |
1573 * | dest | | | | |
1574 * | | | | | | |
1575 * | v | | | | |
1576 * |**************************************************************************
1577 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1578 * | | | | | |
1579 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1580 * ***************************************************************************
5afe0022
RP
1581 *
1582 * (+) the mount is moved to the destination. And is then propagated to
1583 * all the mounts in the propagation tree of the destination mount.
21444403 1584 * (+*) the mount is moved to the destination.
5afe0022
RP
1585 * (+++) the mount is moved to the destination and is then propagated to
1586 * all the mounts belonging to the destination mount's propagation tree.
1587 * the mount is marked as 'shared and slave'.
1588 * (*) the mount continues to be a slave at the new location.
b90fa9ae
RP
1589 *
1590 * if the source mount is a tree, the operations explained above is
1591 * applied to each mount in the tree.
1592 * Must be called without spinlocks held, since this function can sleep
1593 * in allocations.
1594 */
0fb54e50 1595static int attach_recursive_mnt(struct mount *source_mnt,
1a390689 1596 struct path *path, struct path *parent_path)
b90fa9ae
RP
1597{
1598 LIST_HEAD(tree_list);
a8d56d8e 1599 struct mount *dest_mnt = real_mount(path->mnt);
1a390689 1600 struct dentry *dest_dentry = path->dentry;
315fc83e 1601 struct mount *child, *p;
719f5d7f 1602 int err;
b90fa9ae 1603
a8d56d8e 1604 if (IS_MNT_SHARED(&dest_mnt->mnt)) {
0fb54e50 1605 err = invent_group_ids(source_mnt, true);
719f5d7f
MS
1606 if (err)
1607 goto out;
1608 }
a8d56d8e 1609 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
719f5d7f
MS
1610 if (err)
1611 goto out_cleanup_ids;
b90fa9ae 1612
99b7db7b 1613 br_write_lock(vfsmount_lock);
df1a1ad2 1614
a8d56d8e 1615 if (IS_MNT_SHARED(&dest_mnt->mnt)) {
0fb54e50 1616 for (p = source_mnt; p; p = next_mnt(p, &source_mnt->mnt))
0f0afb1d 1617 set_mnt_shared(p);
b90fa9ae 1618 }
1a390689 1619 if (parent_path) {
0fb54e50
AV
1620 detach_mnt(source_mnt, parent_path);
1621 attach_mnt(source_mnt, path);
143c8c91 1622 touch_mnt_namespace(source_mnt->mnt_ns);
21444403 1623 } else {
14cf1fa8 1624 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
0fb54e50 1625 commit_tree(source_mnt);
21444403 1626 }
b90fa9ae 1627
1b8e5564
AV
1628 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1629 list_del_init(&child->mnt_hash);
4b2619a5 1630 commit_tree(child);
b90fa9ae 1631 }
99b7db7b
NP
1632 br_write_unlock(vfsmount_lock);
1633
b90fa9ae 1634 return 0;
719f5d7f
MS
1635
1636 out_cleanup_ids:
a8d56d8e 1637 if (IS_MNT_SHARED(&dest_mnt->mnt))
0fb54e50 1638 cleanup_group_ids(source_mnt, NULL);
719f5d7f
MS
1639 out:
1640 return err;
b90fa9ae
RP
1641}
1642
b12cea91
AV
1643static int lock_mount(struct path *path)
1644{
1645 struct vfsmount *mnt;
1646retry:
1647 mutex_lock(&path->dentry->d_inode->i_mutex);
1648 if (unlikely(cant_mount(path->dentry))) {
1649 mutex_unlock(&path->dentry->d_inode->i_mutex);
1650 return -ENOENT;
1651 }
1652 down_write(&namespace_sem);
1653 mnt = lookup_mnt(path);
1654 if (likely(!mnt))
1655 return 0;
1656 up_write(&namespace_sem);
1657 mutex_unlock(&path->dentry->d_inode->i_mutex);
1658 path_put(path);
1659 path->mnt = mnt;
1660 path->dentry = dget(mnt->mnt_root);
1661 goto retry;
1662}
1663
1664static void unlock_mount(struct path *path)
1665{
1666 up_write(&namespace_sem);
1667 mutex_unlock(&path->dentry->d_inode->i_mutex);
1668}
1669
95bc5f25 1670static int graft_tree(struct mount *mnt, struct path *path)
1da177e4 1671{
95bc5f25 1672 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1da177e4
LT
1673 return -EINVAL;
1674
8c3ee42e 1675 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
95bc5f25 1676 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1da177e4
LT
1677 return -ENOTDIR;
1678
b12cea91
AV
1679 if (d_unlinked(path->dentry))
1680 return -ENOENT;
1da177e4 1681
95bc5f25 1682 return attach_recursive_mnt(mnt, path, NULL);
1da177e4
LT
1683}
1684
7a2e8a8f
VA
1685/*
1686 * Sanity check the flags to change_mnt_propagation.
1687 */
1688
1689static int flags_to_propagation_type(int flags)
1690{
7c6e984d 1691 int type = flags & ~(MS_REC | MS_SILENT);
7a2e8a8f
VA
1692
1693 /* Fail if any non-propagation flags are set */
1694 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1695 return 0;
1696 /* Only one propagation flag should be set */
1697 if (!is_power_of_2(type))
1698 return 0;
1699 return type;
1700}
1701
07b20889
RP
1702/*
1703 * recursively change the type of the mountpoint.
1704 */
0a0d8a46 1705static int do_change_type(struct path *path, int flag)
07b20889 1706{
315fc83e 1707 struct mount *m;
4b8b21f4 1708 struct mount *mnt = real_mount(path->mnt);
07b20889 1709 int recurse = flag & MS_REC;
7a2e8a8f 1710 int type;
719f5d7f 1711 int err = 0;
07b20889 1712
ee6f9582
MS
1713 if (!capable(CAP_SYS_ADMIN))
1714 return -EPERM;
1715
2d92ab3c 1716 if (path->dentry != path->mnt->mnt_root)
07b20889
RP
1717 return -EINVAL;
1718
7a2e8a8f
VA
1719 type = flags_to_propagation_type(flag);
1720 if (!type)
1721 return -EINVAL;
1722
07b20889 1723 down_write(&namespace_sem);
719f5d7f
MS
1724 if (type == MS_SHARED) {
1725 err = invent_group_ids(mnt, recurse);
1726 if (err)
1727 goto out_unlock;
1728 }
1729
99b7db7b 1730 br_write_lock(vfsmount_lock);
4b8b21f4 1731 for (m = mnt; m; m = (recurse ? next_mnt(m, &mnt->mnt) : NULL))
0f0afb1d 1732 change_mnt_propagation(m, type);
99b7db7b 1733 br_write_unlock(vfsmount_lock);
719f5d7f
MS
1734
1735 out_unlock:
07b20889 1736 up_write(&namespace_sem);
719f5d7f 1737 return err;
07b20889
RP
1738}
1739
1da177e4
LT
1740/*
1741 * do loopback mount.
1742 */
0a0d8a46 1743static int do_loopback(struct path *path, char *old_name,
2dafe1c4 1744 int recurse)
1da177e4 1745{
b12cea91 1746 LIST_HEAD(umount_list);
2d92ab3c 1747 struct path old_path;
87129cc0 1748 struct mount *mnt = NULL, *old;
2d92ab3c 1749 int err = mount_is_safe(path);
1da177e4
LT
1750 if (err)
1751 return err;
1752 if (!old_name || !*old_name)
1753 return -EINVAL;
815d405c 1754 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1da177e4
LT
1755 if (err)
1756 return err;
1757
b12cea91
AV
1758 err = lock_mount(path);
1759 if (err)
1760 goto out;
1761
87129cc0
AV
1762 old = real_mount(old_path.mnt);
1763
1da177e4 1764 err = -EINVAL;
2d92ab3c 1765 if (IS_MNT_UNBINDABLE(old_path.mnt))
b12cea91 1766 goto out2;
9676f0c6 1767
143c8c91 1768 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
b12cea91 1769 goto out2;
1da177e4 1770
ccd48bc7
AV
1771 err = -ENOMEM;
1772 if (recurse)
87129cc0 1773 mnt = copy_tree(old, old_path.dentry, 0);
ccd48bc7 1774 else
87129cc0 1775 mnt = clone_mnt(old, old_path.dentry, 0);
ccd48bc7
AV
1776
1777 if (!mnt)
b12cea91 1778 goto out2;
ccd48bc7 1779
95bc5f25 1780 err = graft_tree(mnt, path);
ccd48bc7 1781 if (err) {
99b7db7b 1782 br_write_lock(vfsmount_lock);
761d5c38 1783 umount_tree(mnt, 0, &umount_list);
99b7db7b 1784 br_write_unlock(vfsmount_lock);
5b83d2c5 1785 }
b12cea91
AV
1786out2:
1787 unlock_mount(path);
1788 release_mounts(&umount_list);
ccd48bc7 1789out:
2d92ab3c 1790 path_put(&old_path);
1da177e4
LT
1791 return err;
1792}
1793
2e4b7fcd
DH
1794static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1795{
1796 int error = 0;
1797 int readonly_request = 0;
1798
1799 if (ms_flags & MS_RDONLY)
1800 readonly_request = 1;
1801 if (readonly_request == __mnt_is_readonly(mnt))
1802 return 0;
1803
1804 if (readonly_request)
83adc753 1805 error = mnt_make_readonly(real_mount(mnt));
2e4b7fcd 1806 else
83adc753 1807 __mnt_unmake_readonly(real_mount(mnt));
2e4b7fcd
DH
1808 return error;
1809}
1810
1da177e4
LT
1811/*
1812 * change filesystem flags. dir should be a physical root of filesystem.
1813 * If you've mounted a non-root directory somewhere and want to do remount
1814 * on it - tough luck.
1815 */
0a0d8a46 1816static int do_remount(struct path *path, int flags, int mnt_flags,
1da177e4
LT
1817 void *data)
1818{
1819 int err;
2d92ab3c 1820 struct super_block *sb = path->mnt->mnt_sb;
143c8c91 1821 struct mount *mnt = real_mount(path->mnt);
1da177e4
LT
1822
1823 if (!capable(CAP_SYS_ADMIN))
1824 return -EPERM;
1825
143c8c91 1826 if (!check_mnt(mnt))
1da177e4
LT
1827 return -EINVAL;
1828
2d92ab3c 1829 if (path->dentry != path->mnt->mnt_root)
1da177e4
LT
1830 return -EINVAL;
1831
ff36fe2c
EP
1832 err = security_sb_remount(sb, data);
1833 if (err)
1834 return err;
1835
1da177e4 1836 down_write(&sb->s_umount);
2e4b7fcd 1837 if (flags & MS_BIND)
2d92ab3c 1838 err = change_mount_flags(path->mnt, flags);
4aa98cf7 1839 else
2e4b7fcd 1840 err = do_remount_sb(sb, flags, data, 0);
7b43a79f 1841 if (!err) {
99b7db7b 1842 br_write_lock(vfsmount_lock);
143c8c91
AV
1843 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1844 mnt->mnt.mnt_flags = mnt_flags;
99b7db7b 1845 br_write_unlock(vfsmount_lock);
7b43a79f 1846 }
1da177e4 1847 up_write(&sb->s_umount);
0e55a7cc 1848 if (!err) {
99b7db7b 1849 br_write_lock(vfsmount_lock);
143c8c91 1850 touch_mnt_namespace(mnt->mnt_ns);
99b7db7b 1851 br_write_unlock(vfsmount_lock);
0e55a7cc 1852 }
1da177e4
LT
1853 return err;
1854}
1855
cbbe362c 1856static inline int tree_contains_unbindable(struct mount *mnt)
9676f0c6 1857{
315fc83e 1858 struct mount *p;
cbbe362c 1859 for (p = mnt; p; p = next_mnt(p, &mnt->mnt)) {
315fc83e 1860 if (IS_MNT_UNBINDABLE(&p->mnt))
9676f0c6
RP
1861 return 1;
1862 }
1863 return 0;
1864}
1865
0a0d8a46 1866static int do_move_mount(struct path *path, char *old_name)
1da177e4 1867{
2d92ab3c 1868 struct path old_path, parent_path;
676da58d 1869 struct mount *p;
0fb54e50 1870 struct mount *old;
1da177e4
LT
1871 int err = 0;
1872 if (!capable(CAP_SYS_ADMIN))
1873 return -EPERM;
1874 if (!old_name || !*old_name)
1875 return -EINVAL;
2d92ab3c 1876 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1da177e4
LT
1877 if (err)
1878 return err;
1879
b12cea91 1880 err = lock_mount(path);
cc53ce53
DH
1881 if (err < 0)
1882 goto out;
1883
143c8c91
AV
1884 old = real_mount(old_path.mnt);
1885
1da177e4 1886 err = -EINVAL;
143c8c91 1887 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1da177e4
LT
1888 goto out1;
1889
f3da392e 1890 if (d_unlinked(path->dentry))
21444403 1891 goto out1;
1da177e4
LT
1892
1893 err = -EINVAL;
2d92ab3c 1894 if (old_path.dentry != old_path.mnt->mnt_root)
21444403 1895 goto out1;
1da177e4 1896
676da58d 1897 if (!mnt_has_parent(old))
21444403 1898 goto out1;
1da177e4 1899
2d92ab3c
AV
1900 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1901 S_ISDIR(old_path.dentry->d_inode->i_mode))
21444403
RP
1902 goto out1;
1903 /*
1904 * Don't move a mount residing in a shared parent.
1905 */
0714a533 1906 if (IS_MNT_SHARED(&old->mnt_parent->mnt))
21444403 1907 goto out1;
9676f0c6
RP
1908 /*
1909 * Don't move a mount tree containing unbindable mounts to a destination
1910 * mount which is shared.
1911 */
2d92ab3c 1912 if (IS_MNT_SHARED(path->mnt) &&
cbbe362c 1913 tree_contains_unbindable(old))
9676f0c6 1914 goto out1;
1da177e4 1915 err = -ELOOP;
0714a533 1916 for (p = real_mount(path->mnt); mnt_has_parent(p); p = p->mnt_parent)
676da58d 1917 if (p == old)
21444403 1918 goto out1;
1da177e4 1919
0fb54e50 1920 err = attach_recursive_mnt(old, path, &parent_path);
4ac91378 1921 if (err)
21444403 1922 goto out1;
1da177e4
LT
1923
1924 /* if the mount is moved, it should no longer be expire
1925 * automatically */
6776db3d 1926 list_del_init(&old->mnt_expire);
1da177e4 1927out1:
b12cea91 1928 unlock_mount(path);
1da177e4 1929out:
1da177e4 1930 if (!err)
1a390689 1931 path_put(&parent_path);
2d92ab3c 1932 path_put(&old_path);
1da177e4
LT
1933 return err;
1934}
1935
9d412a43
AV
1936static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1937{
1938 int err;
1939 const char *subtype = strchr(fstype, '.');
1940 if (subtype) {
1941 subtype++;
1942 err = -EINVAL;
1943 if (!subtype[0])
1944 goto err;
1945 } else
1946 subtype = "";
1947
1948 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1949 err = -ENOMEM;
1950 if (!mnt->mnt_sb->s_subtype)
1951 goto err;
1952 return mnt;
1953
1954 err:
1955 mntput(mnt);
1956 return ERR_PTR(err);
1957}
1958
79e801a9 1959static struct vfsmount *
9d412a43
AV
1960do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1961{
1962 struct file_system_type *type = get_fs_type(fstype);
1963 struct vfsmount *mnt;
1964 if (!type)
1965 return ERR_PTR(-ENODEV);
1966 mnt = vfs_kern_mount(type, flags, name, data);
1967 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1968 !mnt->mnt_sb->s_subtype)
1969 mnt = fs_set_subtype(mnt, fstype);
1970 put_filesystem(type);
1971 return mnt;
1972}
9d412a43
AV
1973
1974/*
1975 * add a mount into a namespace's mount tree
1976 */
95bc5f25 1977static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
9d412a43
AV
1978{
1979 int err;
1980
1981 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1982
b12cea91
AV
1983 err = lock_mount(path);
1984 if (err)
1985 return err;
9d412a43
AV
1986
1987 err = -EINVAL;
143c8c91 1988 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
9d412a43
AV
1989 goto unlock;
1990
1991 /* Refuse the same filesystem on the same mount point */
1992 err = -EBUSY;
95bc5f25 1993 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
9d412a43
AV
1994 path->mnt->mnt_root == path->dentry)
1995 goto unlock;
1996
1997 err = -EINVAL;
95bc5f25 1998 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
9d412a43
AV
1999 goto unlock;
2000
95bc5f25 2001 newmnt->mnt.mnt_flags = mnt_flags;
9d412a43
AV
2002 err = graft_tree(newmnt, path);
2003
2004unlock:
b12cea91 2005 unlock_mount(path);
9d412a43
AV
2006 return err;
2007}
b1e75df4 2008
1da177e4
LT
2009/*
2010 * create a new mount for userspace and request it to be added into the
2011 * namespace's tree
2012 */
0a0d8a46 2013static int do_new_mount(struct path *path, char *type, int flags,
1da177e4
LT
2014 int mnt_flags, char *name, void *data)
2015{
2016 struct vfsmount *mnt;
15f9a3f3 2017 int err;
1da177e4 2018
eca6f534 2019 if (!type)
1da177e4
LT
2020 return -EINVAL;
2021
2022 /* we need capabilities... */
2023 if (!capable(CAP_SYS_ADMIN))
2024 return -EPERM;
2025
2026 mnt = do_kern_mount(type, flags, name, data);
2027 if (IS_ERR(mnt))
2028 return PTR_ERR(mnt);
2029
95bc5f25 2030 err = do_add_mount(real_mount(mnt), path, mnt_flags);
15f9a3f3
AV
2031 if (err)
2032 mntput(mnt);
2033 return err;
1da177e4
LT
2034}
2035
19a167af
AV
2036int finish_automount(struct vfsmount *m, struct path *path)
2037{
6776db3d 2038 struct mount *mnt = real_mount(m);
19a167af
AV
2039 int err;
2040 /* The new mount record should have at least 2 refs to prevent it being
2041 * expired before we get a chance to add it
2042 */
6776db3d 2043 BUG_ON(mnt_get_count(mnt) < 2);
19a167af
AV
2044
2045 if (m->mnt_sb == path->mnt->mnt_sb &&
2046 m->mnt_root == path->dentry) {
b1e75df4
AV
2047 err = -ELOOP;
2048 goto fail;
19a167af
AV
2049 }
2050
95bc5f25 2051 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
b1e75df4
AV
2052 if (!err)
2053 return 0;
2054fail:
2055 /* remove m from any expiration list it may be on */
6776db3d 2056 if (!list_empty(&mnt->mnt_expire)) {
b1e75df4
AV
2057 down_write(&namespace_sem);
2058 br_write_lock(vfsmount_lock);
6776db3d 2059 list_del_init(&mnt->mnt_expire);
b1e75df4
AV
2060 br_write_unlock(vfsmount_lock);
2061 up_write(&namespace_sem);
19a167af 2062 }
b1e75df4
AV
2063 mntput(m);
2064 mntput(m);
19a167af
AV
2065 return err;
2066}
2067
ea5b778a
DH
2068/**
2069 * mnt_set_expiry - Put a mount on an expiration list
2070 * @mnt: The mount to list.
2071 * @expiry_list: The list to add the mount to.
2072 */
2073void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2074{
2075 down_write(&namespace_sem);
2076 br_write_lock(vfsmount_lock);
2077
6776db3d 2078 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
ea5b778a
DH
2079
2080 br_write_unlock(vfsmount_lock);
2081 up_write(&namespace_sem);
2082}
2083EXPORT_SYMBOL(mnt_set_expiry);
2084
1da177e4
LT
2085/*
2086 * process a list of expirable mountpoints with the intent of discarding any
2087 * mountpoints that aren't in use and haven't been touched since last we came
2088 * here
2089 */
2090void mark_mounts_for_expiry(struct list_head *mounts)
2091{
761d5c38 2092 struct mount *mnt, *next;
1da177e4 2093 LIST_HEAD(graveyard);
bcc5c7d2 2094 LIST_HEAD(umounts);
1da177e4
LT
2095
2096 if (list_empty(mounts))
2097 return;
2098
bcc5c7d2 2099 down_write(&namespace_sem);
99b7db7b 2100 br_write_lock(vfsmount_lock);
1da177e4
LT
2101
2102 /* extract from the expiration list every vfsmount that matches the
2103 * following criteria:
2104 * - only referenced by its parent vfsmount
2105 * - still marked for expiry (marked on the last call here; marks are
2106 * cleared by mntput())
2107 */
6776db3d 2108 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
761d5c38 2109 if (!xchg(&mnt->mnt.mnt_expiry_mark, 1) ||
1ab59738 2110 propagate_mount_busy(mnt, 1))
1da177e4 2111 continue;
6776db3d 2112 list_move(&mnt->mnt_expire, &graveyard);
1da177e4 2113 }
bcc5c7d2 2114 while (!list_empty(&graveyard)) {
6776db3d 2115 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
143c8c91 2116 touch_mnt_namespace(mnt->mnt_ns);
bcc5c7d2
AV
2117 umount_tree(mnt, 1, &umounts);
2118 }
99b7db7b 2119 br_write_unlock(vfsmount_lock);
bcc5c7d2
AV
2120 up_write(&namespace_sem);
2121
2122 release_mounts(&umounts);
5528f911
TM
2123}
2124
2125EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2126
2127/*
2128 * Ripoff of 'select_parent()'
2129 *
2130 * search the list of submounts for a given mountpoint, and move any
2131 * shrinkable submounts to the 'graveyard' list.
2132 */
692afc31 2133static int select_submounts(struct mount *parent, struct list_head *graveyard)
5528f911 2134{
692afc31 2135 struct mount *this_parent = parent;
5528f911
TM
2136 struct list_head *next;
2137 int found = 0;
2138
2139repeat:
6b41d536 2140 next = this_parent->mnt_mounts.next;
5528f911 2141resume:
6b41d536 2142 while (next != &this_parent->mnt_mounts) {
5528f911 2143 struct list_head *tmp = next;
6b41d536 2144 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
5528f911
TM
2145
2146 next = tmp->next;
692afc31 2147 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
1da177e4 2148 continue;
5528f911
TM
2149 /*
2150 * Descend a level if the d_mounts list is non-empty.
2151 */
6b41d536 2152 if (!list_empty(&mnt->mnt_mounts)) {
5528f911
TM
2153 this_parent = mnt;
2154 goto repeat;
2155 }
1da177e4 2156
1ab59738 2157 if (!propagate_mount_busy(mnt, 1)) {
6776db3d 2158 list_move_tail(&mnt->mnt_expire, graveyard);
5528f911
TM
2159 found++;
2160 }
1da177e4 2161 }
5528f911
TM
2162 /*
2163 * All done at this level ... ascend and resume the search
2164 */
2165 if (this_parent != parent) {
6b41d536 2166 next = this_parent->mnt_child.next;
0714a533 2167 this_parent = this_parent->mnt_parent;
5528f911
TM
2168 goto resume;
2169 }
2170 return found;
2171}
2172
2173/*
2174 * process a list of expirable mountpoints with the intent of discarding any
2175 * submounts of a specific parent mountpoint
99b7db7b
NP
2176 *
2177 * vfsmount_lock must be held for write
5528f911 2178 */
692afc31 2179static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
5528f911
TM
2180{
2181 LIST_HEAD(graveyard);
761d5c38 2182 struct mount *m;
5528f911 2183
5528f911 2184 /* extract submounts of 'mountpoint' from the expiration list */
c35038be 2185 while (select_submounts(mnt, &graveyard)) {
bcc5c7d2 2186 while (!list_empty(&graveyard)) {
761d5c38 2187 m = list_first_entry(&graveyard, struct mount,
6776db3d 2188 mnt_expire);
143c8c91 2189 touch_mnt_namespace(m->mnt_ns);
afef80b3 2190 umount_tree(m, 1, umounts);
bcc5c7d2
AV
2191 }
2192 }
1da177e4
LT
2193}
2194
1da177e4
LT
2195/*
2196 * Some copy_from_user() implementations do not return the exact number of
2197 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2198 * Note that this function differs from copy_from_user() in that it will oops
2199 * on bad values of `to', rather than returning a short copy.
2200 */
b58fed8b
RP
2201static long exact_copy_from_user(void *to, const void __user * from,
2202 unsigned long n)
1da177e4
LT
2203{
2204 char *t = to;
2205 const char __user *f = from;
2206 char c;
2207
2208 if (!access_ok(VERIFY_READ, from, n))
2209 return n;
2210
2211 while (n) {
2212 if (__get_user(c, f)) {
2213 memset(t, 0, n);
2214 break;
2215 }
2216 *t++ = c;
2217 f++;
2218 n--;
2219 }
2220 return n;
2221}
2222
b58fed8b 2223int copy_mount_options(const void __user * data, unsigned long *where)
1da177e4
LT
2224{
2225 int i;
2226 unsigned long page;
2227 unsigned long size;
b58fed8b 2228
1da177e4
LT
2229 *where = 0;
2230 if (!data)
2231 return 0;
2232
2233 if (!(page = __get_free_page(GFP_KERNEL)))
2234 return -ENOMEM;
2235
2236 /* We only care that *some* data at the address the user
2237 * gave us is valid. Just in case, we'll zero
2238 * the remainder of the page.
2239 */
2240 /* copy_from_user cannot cross TASK_SIZE ! */
2241 size = TASK_SIZE - (unsigned long)data;
2242 if (size > PAGE_SIZE)
2243 size = PAGE_SIZE;
2244
2245 i = size - exact_copy_from_user((void *)page, data, size);
2246 if (!i) {
b58fed8b 2247 free_page(page);
1da177e4
LT
2248 return -EFAULT;
2249 }
2250 if (i != PAGE_SIZE)
2251 memset((char *)page + i, 0, PAGE_SIZE - i);
2252 *where = page;
2253 return 0;
2254}
2255
eca6f534
VN
2256int copy_mount_string(const void __user *data, char **where)
2257{
2258 char *tmp;
2259
2260 if (!data) {
2261 *where = NULL;
2262 return 0;
2263 }
2264
2265 tmp = strndup_user(data, PAGE_SIZE);
2266 if (IS_ERR(tmp))
2267 return PTR_ERR(tmp);
2268
2269 *where = tmp;
2270 return 0;
2271}
2272
1da177e4
LT
2273/*
2274 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2275 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2276 *
2277 * data is a (void *) that can point to any structure up to
2278 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2279 * information (or be NULL).
2280 *
2281 * Pre-0.97 versions of mount() didn't have a flags word.
2282 * When the flags word was introduced its top half was required
2283 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2284 * Therefore, if this magic number is present, it carries no information
2285 * and must be discarded.
2286 */
b58fed8b 2287long do_mount(char *dev_name, char *dir_name, char *type_page,
1da177e4
LT
2288 unsigned long flags, void *data_page)
2289{
2d92ab3c 2290 struct path path;
1da177e4
LT
2291 int retval = 0;
2292 int mnt_flags = 0;
2293
2294 /* Discard magic */
2295 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2296 flags &= ~MS_MGC_MSK;
2297
2298 /* Basic sanity checks */
2299
2300 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2301 return -EINVAL;
1da177e4
LT
2302
2303 if (data_page)
2304 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2305
a27ab9f2
TH
2306 /* ... and get the mountpoint */
2307 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2308 if (retval)
2309 return retval;
2310
2311 retval = security_sb_mount(dev_name, &path,
2312 type_page, flags, data_page);
2313 if (retval)
2314 goto dput_out;
2315
613cbe3d
AK
2316 /* Default to relatime unless overriden */
2317 if (!(flags & MS_NOATIME))
2318 mnt_flags |= MNT_RELATIME;
0a1c01c9 2319
1da177e4
LT
2320 /* Separate the per-mountpoint flags */
2321 if (flags & MS_NOSUID)
2322 mnt_flags |= MNT_NOSUID;
2323 if (flags & MS_NODEV)
2324 mnt_flags |= MNT_NODEV;
2325 if (flags & MS_NOEXEC)
2326 mnt_flags |= MNT_NOEXEC;
fc33a7bb
CH
2327 if (flags & MS_NOATIME)
2328 mnt_flags |= MNT_NOATIME;
2329 if (flags & MS_NODIRATIME)
2330 mnt_flags |= MNT_NODIRATIME;
d0adde57
MG
2331 if (flags & MS_STRICTATIME)
2332 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2e4b7fcd
DH
2333 if (flags & MS_RDONLY)
2334 mnt_flags |= MNT_READONLY;
fc33a7bb 2335
7a4dec53 2336 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
d0adde57
MG
2337 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2338 MS_STRICTATIME);
1da177e4 2339
1da177e4 2340 if (flags & MS_REMOUNT)
2d92ab3c 2341 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1da177e4
LT
2342 data_page);
2343 else if (flags & MS_BIND)
2d92ab3c 2344 retval = do_loopback(&path, dev_name, flags & MS_REC);
9676f0c6 2345 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2d92ab3c 2346 retval = do_change_type(&path, flags);
1da177e4 2347 else if (flags & MS_MOVE)
2d92ab3c 2348 retval = do_move_mount(&path, dev_name);
1da177e4 2349 else
2d92ab3c 2350 retval = do_new_mount(&path, type_page, flags, mnt_flags,
1da177e4
LT
2351 dev_name, data_page);
2352dput_out:
2d92ab3c 2353 path_put(&path);
1da177e4
LT
2354 return retval;
2355}
2356
cf8d2c11
TM
2357static struct mnt_namespace *alloc_mnt_ns(void)
2358{
2359 struct mnt_namespace *new_ns;
2360
2361 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2362 if (!new_ns)
2363 return ERR_PTR(-ENOMEM);
2364 atomic_set(&new_ns->count, 1);
2365 new_ns->root = NULL;
2366 INIT_LIST_HEAD(&new_ns->list);
2367 init_waitqueue_head(&new_ns->poll);
2368 new_ns->event = 0;
2369 return new_ns;
2370}
2371
f03c6599
AV
2372void mnt_make_longterm(struct vfsmount *mnt)
2373{
83adc753 2374 __mnt_make_longterm(real_mount(mnt));
f03c6599
AV
2375}
2376
83adc753 2377void mnt_make_shortterm(struct vfsmount *m)
f03c6599 2378{
7e3d0eb0 2379#ifdef CONFIG_SMP
83adc753 2380 struct mount *mnt = real_mount(m);
68e8a9fe 2381 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
f03c6599
AV
2382 return;
2383 br_write_lock(vfsmount_lock);
68e8a9fe 2384 atomic_dec(&mnt->mnt_longterm);
f03c6599 2385 br_write_unlock(vfsmount_lock);
7e3d0eb0 2386#endif
f03c6599
AV
2387}
2388
741a2951
JD
2389/*
2390 * Allocate a new namespace structure and populate it with contents
2391 * copied from the namespace of the passed in task structure.
2392 */
e3222c4e 2393static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
6b3286ed 2394 struct fs_struct *fs)
1da177e4 2395{
6b3286ed 2396 struct mnt_namespace *new_ns;
7f2da1e7 2397 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
315fc83e 2398 struct mount *p, *q;
cb338d06 2399 struct mount *new;
1da177e4 2400
cf8d2c11
TM
2401 new_ns = alloc_mnt_ns();
2402 if (IS_ERR(new_ns))
2403 return new_ns;
1da177e4 2404
390c6843 2405 down_write(&namespace_sem);
1da177e4 2406 /* First pass: copy the tree topology */
87129cc0 2407 new = copy_tree(real_mount(mnt_ns->root), mnt_ns->root->mnt_root,
9676f0c6 2408 CL_COPY_ALL | CL_EXPIRE);
cb338d06 2409 if (!new) {
390c6843 2410 up_write(&namespace_sem);
1da177e4 2411 kfree(new_ns);
5cc4a034 2412 return ERR_PTR(-ENOMEM);
1da177e4 2413 }
cb338d06 2414 new_ns->root = &new->mnt;
99b7db7b 2415 br_write_lock(vfsmount_lock);
1da177e4 2416 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
99b7db7b 2417 br_write_unlock(vfsmount_lock);
1da177e4
LT
2418
2419 /*
2420 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2421 * as belonging to new namespace. We have already acquired a private
2422 * fs_struct, so tsk->fs->lock is not needed.
2423 */
315fc83e 2424 p = real_mount(mnt_ns->root);
cb338d06 2425 q = new;
1da177e4 2426 while (p) {
143c8c91 2427 q->mnt_ns = new_ns;
83adc753 2428 __mnt_make_longterm(q);
1da177e4 2429 if (fs) {
315fc83e
AV
2430 if (&p->mnt == fs->root.mnt) {
2431 fs->root.mnt = mntget(&q->mnt);
83adc753 2432 __mnt_make_longterm(q);
315fc83e
AV
2433 mnt_make_shortterm(&p->mnt);
2434 rootmnt = &p->mnt;
1da177e4 2435 }
315fc83e
AV
2436 if (&p->mnt == fs->pwd.mnt) {
2437 fs->pwd.mnt = mntget(&q->mnt);
83adc753 2438 __mnt_make_longterm(q);
315fc83e
AV
2439 mnt_make_shortterm(&p->mnt);
2440 pwdmnt = &p->mnt;
1da177e4 2441 }
1da177e4 2442 }
6b3286ed 2443 p = next_mnt(p, mnt_ns->root);
1da177e4
LT
2444 q = next_mnt(q, new_ns->root);
2445 }
390c6843 2446 up_write(&namespace_sem);
1da177e4 2447
1da177e4 2448 if (rootmnt)
f03c6599 2449 mntput(rootmnt);
1da177e4 2450 if (pwdmnt)
f03c6599 2451 mntput(pwdmnt);
1da177e4 2452
741a2951
JD
2453 return new_ns;
2454}
2455
213dd266 2456struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
e3222c4e 2457 struct fs_struct *new_fs)
741a2951 2458{
6b3286ed 2459 struct mnt_namespace *new_ns;
741a2951 2460
e3222c4e 2461 BUG_ON(!ns);
6b3286ed 2462 get_mnt_ns(ns);
741a2951
JD
2463
2464 if (!(flags & CLONE_NEWNS))
e3222c4e 2465 return ns;
741a2951 2466
e3222c4e 2467 new_ns = dup_mnt_ns(ns, new_fs);
741a2951 2468
6b3286ed 2469 put_mnt_ns(ns);
e3222c4e 2470 return new_ns;
1da177e4
LT
2471}
2472
cf8d2c11
TM
2473/**
2474 * create_mnt_ns - creates a private namespace and adds a root filesystem
2475 * @mnt: pointer to the new root filesystem mountpoint
2476 */
6c449c8d 2477static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
cf8d2c11
TM
2478{
2479 struct mnt_namespace *new_ns;
2480
2481 new_ns = alloc_mnt_ns();
2482 if (!IS_ERR(new_ns)) {
143c8c91 2483 real_mount(mnt)->mnt_ns = new_ns;
83adc753 2484 __mnt_make_longterm(real_mount(mnt));
cf8d2c11
TM
2485 new_ns->root = mnt;
2486 list_add(&new_ns->list, &new_ns->root->mnt_list);
c1334495
AV
2487 } else {
2488 mntput(mnt);
cf8d2c11
TM
2489 }
2490 return new_ns;
2491}
cf8d2c11 2492
ea441d11
AV
2493struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2494{
2495 struct mnt_namespace *ns;
d31da0f0 2496 struct super_block *s;
ea441d11
AV
2497 struct path path;
2498 int err;
2499
2500 ns = create_mnt_ns(mnt);
2501 if (IS_ERR(ns))
2502 return ERR_CAST(ns);
2503
2504 err = vfs_path_lookup(mnt->mnt_root, mnt,
2505 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2506
2507 put_mnt_ns(ns);
2508
2509 if (err)
2510 return ERR_PTR(err);
2511
2512 /* trade a vfsmount reference for active sb one */
d31da0f0
AV
2513 s = path.mnt->mnt_sb;
2514 atomic_inc(&s->s_active);
ea441d11
AV
2515 mntput(path.mnt);
2516 /* lock the sucker */
d31da0f0 2517 down_write(&s->s_umount);
ea441d11
AV
2518 /* ... and return the root of (sub)tree on it */
2519 return path.dentry;
2520}
2521EXPORT_SYMBOL(mount_subtree);
2522
bdc480e3
HC
2523SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2524 char __user *, type, unsigned long, flags, void __user *, data)
1da177e4 2525{
eca6f534
VN
2526 int ret;
2527 char *kernel_type;
2528 char *kernel_dir;
2529 char *kernel_dev;
1da177e4 2530 unsigned long data_page;
1da177e4 2531
eca6f534
VN
2532 ret = copy_mount_string(type, &kernel_type);
2533 if (ret < 0)
2534 goto out_type;
1da177e4 2535
eca6f534
VN
2536 kernel_dir = getname(dir_name);
2537 if (IS_ERR(kernel_dir)) {
2538 ret = PTR_ERR(kernel_dir);
2539 goto out_dir;
2540 }
1da177e4 2541
eca6f534
VN
2542 ret = copy_mount_string(dev_name, &kernel_dev);
2543 if (ret < 0)
2544 goto out_dev;
1da177e4 2545
eca6f534
VN
2546 ret = copy_mount_options(data, &data_page);
2547 if (ret < 0)
2548 goto out_data;
1da177e4 2549
eca6f534
VN
2550 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2551 (void *) data_page);
1da177e4 2552
eca6f534
VN
2553 free_page(data_page);
2554out_data:
2555 kfree(kernel_dev);
2556out_dev:
2557 putname(kernel_dir);
2558out_dir:
2559 kfree(kernel_type);
2560out_type:
2561 return ret;
1da177e4
LT
2562}
2563
afac7cba
AV
2564/*
2565 * Return true if path is reachable from root
2566 *
2567 * namespace_sem or vfsmount_lock is held
2568 */
643822b4 2569bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
afac7cba
AV
2570 const struct path *root)
2571{
643822b4 2572 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
a73324da 2573 dentry = mnt->mnt_mountpoint;
0714a533 2574 mnt = mnt->mnt_parent;
afac7cba 2575 }
643822b4 2576 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
afac7cba
AV
2577}
2578
2579int path_is_under(struct path *path1, struct path *path2)
2580{
2581 int res;
2582 br_read_lock(vfsmount_lock);
643822b4 2583 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
afac7cba
AV
2584 br_read_unlock(vfsmount_lock);
2585 return res;
2586}
2587EXPORT_SYMBOL(path_is_under);
2588
1da177e4
LT
2589/*
2590 * pivot_root Semantics:
2591 * Moves the root file system of the current process to the directory put_old,
2592 * makes new_root as the new root file system of the current process, and sets
2593 * root/cwd of all processes which had them on the current root to new_root.
2594 *
2595 * Restrictions:
2596 * The new_root and put_old must be directories, and must not be on the
2597 * same file system as the current process root. The put_old must be
2598 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2599 * pointed to by put_old must yield the same directory as new_root. No other
2600 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2601 *
4a0d11fa
NB
2602 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2603 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2604 * in this situation.
2605 *
1da177e4
LT
2606 * Notes:
2607 * - we don't move root/cwd if they are not at the root (reason: if something
2608 * cared enough to change them, it's probably wrong to force them elsewhere)
2609 * - it's okay to pick a root that isn't the root of a file system, e.g.
2610 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2611 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2612 * first.
2613 */
3480b257
HC
2614SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2615 const char __user *, put_old)
1da177e4 2616{
2d8f3038 2617 struct path new, old, parent_path, root_parent, root;
419148da 2618 struct mount *new_mnt, *root_mnt;
1da177e4
LT
2619 int error;
2620
2621 if (!capable(CAP_SYS_ADMIN))
2622 return -EPERM;
2623
2d8f3038 2624 error = user_path_dir(new_root, &new);
1da177e4
LT
2625 if (error)
2626 goto out0;
1da177e4 2627
2d8f3038 2628 error = user_path_dir(put_old, &old);
1da177e4
LT
2629 if (error)
2630 goto out1;
2631
2d8f3038 2632 error = security_sb_pivotroot(&old, &new);
b12cea91
AV
2633 if (error)
2634 goto out2;
1da177e4 2635
f7ad3c6b 2636 get_fs_root(current->fs, &root);
b12cea91
AV
2637 error = lock_mount(&old);
2638 if (error)
2639 goto out3;
2640
1da177e4 2641 error = -EINVAL;
419148da
AV
2642 new_mnt = real_mount(new.mnt);
2643 root_mnt = real_mount(root.mnt);
2d8f3038 2644 if (IS_MNT_SHARED(old.mnt) ||
0714a533
AV
2645 IS_MNT_SHARED(&new_mnt->mnt_parent->mnt) ||
2646 IS_MNT_SHARED(&root_mnt->mnt_parent->mnt))
b12cea91 2647 goto out4;
143c8c91 2648 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
b12cea91 2649 goto out4;
1da177e4 2650 error = -ENOENT;
f3da392e 2651 if (d_unlinked(new.dentry))
b12cea91 2652 goto out4;
f3da392e 2653 if (d_unlinked(old.dentry))
b12cea91 2654 goto out4;
1da177e4 2655 error = -EBUSY;
2d8f3038
AV
2656 if (new.mnt == root.mnt ||
2657 old.mnt == root.mnt)
b12cea91 2658 goto out4; /* loop, on the same file system */
1da177e4 2659 error = -EINVAL;
8c3ee42e 2660 if (root.mnt->mnt_root != root.dentry)
b12cea91 2661 goto out4; /* not a mountpoint */
676da58d 2662 if (!mnt_has_parent(root_mnt))
b12cea91 2663 goto out4; /* not attached */
2d8f3038 2664 if (new.mnt->mnt_root != new.dentry)
b12cea91 2665 goto out4; /* not a mountpoint */
676da58d 2666 if (!mnt_has_parent(new_mnt))
b12cea91 2667 goto out4; /* not attached */
4ac91378 2668 /* make sure we can reach put_old from new_root */
643822b4 2669 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
b12cea91 2670 goto out4;
27cb1572 2671 br_write_lock(vfsmount_lock);
419148da
AV
2672 detach_mnt(new_mnt, &parent_path);
2673 detach_mnt(root_mnt, &root_parent);
4ac91378 2674 /* mount old root on put_old */
419148da 2675 attach_mnt(root_mnt, &old);
4ac91378 2676 /* mount new_root on / */
419148da 2677 attach_mnt(new_mnt, &root_parent);
6b3286ed 2678 touch_mnt_namespace(current->nsproxy->mnt_ns);
99b7db7b 2679 br_write_unlock(vfsmount_lock);
2d8f3038 2680 chroot_fs_refs(&root, &new);
1da177e4 2681 error = 0;
b12cea91
AV
2682out4:
2683 unlock_mount(&old);
2684 if (!error) {
2685 path_put(&root_parent);
2686 path_put(&parent_path);
2687 }
2688out3:
8c3ee42e 2689 path_put(&root);
b12cea91 2690out2:
2d8f3038 2691 path_put(&old);
1da177e4 2692out1:
2d8f3038 2693 path_put(&new);
1da177e4 2694out0:
1da177e4 2695 return error;
1da177e4
LT
2696}
2697
2698static void __init init_mount_tree(void)
2699{
2700 struct vfsmount *mnt;
6b3286ed 2701 struct mnt_namespace *ns;
ac748a09 2702 struct path root;
1da177e4
LT
2703
2704 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2705 if (IS_ERR(mnt))
2706 panic("Can't create rootfs");
b3e19d92 2707
3b22edc5
TM
2708 ns = create_mnt_ns(mnt);
2709 if (IS_ERR(ns))
1da177e4 2710 panic("Can't allocate initial namespace");
6b3286ed
KK
2711
2712 init_task.nsproxy->mnt_ns = ns;
2713 get_mnt_ns(ns);
2714
ac748a09
JB
2715 root.mnt = ns->root;
2716 root.dentry = ns->root->mnt_root;
2717
2718 set_fs_pwd(current->fs, &root);
2719 set_fs_root(current->fs, &root);
1da177e4
LT
2720}
2721
74bf17cf 2722void __init mnt_init(void)
1da177e4 2723{
13f14b4d 2724 unsigned u;
15a67dd8 2725 int err;
1da177e4 2726
390c6843
RP
2727 init_rwsem(&namespace_sem);
2728
7d6fec45 2729 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
20c2df83 2730 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1da177e4 2731
b58fed8b 2732 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1da177e4
LT
2733
2734 if (!mount_hashtable)
2735 panic("Failed to allocate mount hash table\n");
2736
80cdc6da 2737 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
13f14b4d
ED
2738
2739 for (u = 0; u < HASH_SIZE; u++)
2740 INIT_LIST_HEAD(&mount_hashtable[u]);
1da177e4 2741
99b7db7b
NP
2742 br_lock_init(vfsmount_lock);
2743
15a67dd8
RD
2744 err = sysfs_init();
2745 if (err)
2746 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
8e24eea7 2747 __func__, err);
00d26666
GKH
2748 fs_kobj = kobject_create_and_add("fs", NULL);
2749 if (!fs_kobj)
8e24eea7 2750 printk(KERN_WARNING "%s: kobj create error\n", __func__);
1da177e4
LT
2751 init_rootfs();
2752 init_mount_tree();
2753}
2754
616511d0 2755void put_mnt_ns(struct mnt_namespace *ns)
1da177e4 2756{
70fbcdf4 2757 LIST_HEAD(umount_list);
616511d0 2758
d498b25a 2759 if (!atomic_dec_and_test(&ns->count))
616511d0 2760 return;
390c6843 2761 down_write(&namespace_sem);
99b7db7b 2762 br_write_lock(vfsmount_lock);
761d5c38 2763 umount_tree(real_mount(ns->root), 0, &umount_list);
99b7db7b 2764 br_write_unlock(vfsmount_lock);
390c6843 2765 up_write(&namespace_sem);
70fbcdf4 2766 release_mounts(&umount_list);
6b3286ed 2767 kfree(ns);
1da177e4 2768}
9d412a43
AV
2769
2770struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2771{
423e0ab0
TC
2772 struct vfsmount *mnt;
2773 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2774 if (!IS_ERR(mnt)) {
2775 /*
2776 * it is a longterm mount, don't release mnt until
2777 * we unmount before file sys is unregistered
2778 */
2779 mnt_make_longterm(mnt);
2780 }
2781 return mnt;
9d412a43
AV
2782}
2783EXPORT_SYMBOL_GPL(kern_mount_data);
423e0ab0
TC
2784
2785void kern_unmount(struct vfsmount *mnt)
2786{
2787 /* release long term mount so mount point can be released */
2788 if (!IS_ERR_OR_NULL(mnt)) {
2789 mnt_make_shortterm(mnt);
2790 mntput(mnt);
2791 }
2792}
2793EXPORT_SYMBOL(kern_unmount);
02125a82
AV
2794
2795bool our_mnt(struct vfsmount *mnt)
2796{
143c8c91 2797 return check_mnt(real_mount(mnt));
02125a82 2798}