blob: e069a4c5e389c48eb88e3a6847422750ae693b2b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11#include <linux/config.h>
12#include <linux/syscalls.h>
13#include <linux/slab.h>
14#include <linux/sched.h>
15#include <linux/smp_lock.h>
16#include <linux/init.h>
17#include <linux/quotaops.h>
18#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
21#include <linux/seq_file.h>
22#include <linux/namespace.h>
23#include <linux/namei.h>
24#include <linux/security.h>
25#include <linux/mount.h>
26#include <asm/uaccess.h>
27#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050028#include "pnode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30extern int __init init_rootfs(void);
31
32#ifdef CONFIG_SYSFS
33extern int __init sysfs_init(void);
34#else
35static inline int sysfs_init(void)
36{
37 return 0;
38}
39#endif
40
41/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050042__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43
44static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Eric Dumazetfa3536c2006-03-26 01:37:24 -080046static struct list_head *mount_hashtable __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070047static int hash_mask __read_mostly, hash_bits __read_mostly;
Eric Dumazetfa3536c2006-03-26 01:37:24 -080048static kmem_cache_t *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050049static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080051/* /sys/fs */
52decl_subsys(fs, NULL, NULL);
53EXPORT_SYMBOL_GPL(fs_subsys);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56{
Ram Paib58fed82005-11-07 17:16:09 -050057 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
58 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 tmp = tmp + (tmp >> hash_bits);
60 return tmp & hash_mask;
61}
62
63struct vfsmount *alloc_vfsmnt(const char *name)
64{
Ram Paib58fed82005-11-07 17:16:09 -050065 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 if (mnt) {
67 memset(mnt, 0, sizeof(struct vfsmount));
Ram Paib58fed82005-11-07 17:16:09 -050068 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 INIT_LIST_HEAD(&mnt->mnt_hash);
70 INIT_LIST_HEAD(&mnt->mnt_child);
71 INIT_LIST_HEAD(&mnt->mnt_mounts);
72 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070073 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050074 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050075 INIT_LIST_HEAD(&mnt->mnt_slave_list);
76 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050078 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 char *newname = kmalloc(size, GFP_KERNEL);
80 if (newname) {
81 memcpy(newname, name, size);
82 mnt->mnt_devname = newname;
83 }
84 }
85 }
86 return mnt;
87}
88
89void free_vfsmnt(struct vfsmount *mnt)
90{
91 kfree(mnt->mnt_devname);
92 kmem_cache_free(mnt_cache, mnt);
93}
94
95/*
Ram Paia05964f2005-11-07 17:20:17 -050096 * find the first or last mount at @dentry on vfsmount @mnt depending on
97 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 */
Ram Paia05964f2005-11-07 17:20:17 -050099struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
100 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Ram Paib58fed82005-11-07 17:16:09 -0500102 struct list_head *head = mount_hashtable + hash(mnt, dentry);
103 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 struct vfsmount *p, *found = NULL;
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500107 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 p = NULL;
109 if (tmp == head)
110 break;
111 p = list_entry(tmp, struct vfsmount, mnt_hash);
112 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500113 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 break;
115 }
116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return found;
118}
119
Ram Paia05964f2005-11-07 17:20:17 -0500120/*
121 * lookup_mnt increments the ref count before returning
122 * the vfsmount struct.
123 */
124struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
125{
126 struct vfsmount *child_mnt;
127 spin_lock(&vfsmount_lock);
128 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
129 mntget(child_mnt);
130 spin_unlock(&vfsmount_lock);
131 return child_mnt;
132}
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134static inline int check_mnt(struct vfsmount *mnt)
135{
136 return mnt->mnt_namespace == current->namespace;
137}
138
Al Viro5addc5d2005-11-07 17:15:49 -0500139static void touch_namespace(struct namespace *ns)
140{
141 if (ns) {
142 ns->event = ++event;
143 wake_up_interruptible(&ns->poll);
144 }
145}
146
147static void __touch_namespace(struct namespace *ns)
148{
149 if (ns && ns->event != event) {
150 ns->event = event;
151 wake_up_interruptible(&ns->poll);
152 }
153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
156{
157 old_nd->dentry = mnt->mnt_mountpoint;
158 old_nd->mnt = mnt->mnt_parent;
159 mnt->mnt_parent = mnt;
160 mnt->mnt_mountpoint = mnt->mnt_root;
161 list_del_init(&mnt->mnt_child);
162 list_del_init(&mnt->mnt_hash);
163 old_nd->dentry->d_mounted--;
164}
165
Ram Paib90fa9a2005-11-07 17:19:50 -0500166void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
167 struct vfsmount *child_mnt)
168{
169 child_mnt->mnt_parent = mntget(mnt);
170 child_mnt->mnt_mountpoint = dget(dentry);
171 dentry->d_mounted++;
172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
175{
Ram Paib90fa9a2005-11-07 17:19:50 -0500176 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
177 list_add_tail(&mnt->mnt_hash, mount_hashtable +
178 hash(nd->mnt, nd->dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500180}
181
182/*
183 * the caller must hold vfsmount_lock
184 */
185static void commit_tree(struct vfsmount *mnt)
186{
187 struct vfsmount *parent = mnt->mnt_parent;
188 struct vfsmount *m;
189 LIST_HEAD(head);
190 struct namespace *n = parent->mnt_namespace;
191
192 BUG_ON(parent == mnt);
193
194 list_add_tail(&head, &mnt->mnt_list);
195 list_for_each_entry(m, &head, mnt_list)
196 m->mnt_namespace = n;
197 list_splice(&head, n->list.prev);
198
199 list_add_tail(&mnt->mnt_hash, mount_hashtable +
200 hash(parent, mnt->mnt_mountpoint));
201 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
202 touch_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
206{
207 struct list_head *next = p->mnt_mounts.next;
208 if (next == &p->mnt_mounts) {
209 while (1) {
210 if (p == root)
211 return NULL;
212 next = p->mnt_child.next;
213 if (next != &p->mnt_parent->mnt_mounts)
214 break;
215 p = p->mnt_parent;
216 }
217 }
218 return list_entry(next, struct vfsmount, mnt_child);
219}
220
Ram Pai9676f0c2005-11-07 17:21:20 -0500221static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
222{
223 struct list_head *prev = p->mnt_mounts.prev;
224 while (prev != &p->mnt_mounts) {
225 p = list_entry(prev, struct vfsmount, mnt_child);
226 prev = p->mnt_mounts.prev;
227 }
228 return p;
229}
230
Ram Pai36341f62005-11-07 17:17:22 -0500231static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
232 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
234 struct super_block *sb = old->mnt_sb;
235 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
236
237 if (mnt) {
238 mnt->mnt_flags = old->mnt_flags;
239 atomic_inc(&sb->s_active);
240 mnt->mnt_sb = sb;
241 mnt->mnt_root = dget(root);
242 mnt->mnt_mountpoint = mnt->mnt_root;
243 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500244
Ram Pai5afe0022005-11-07 17:21:01 -0500245 if (flag & CL_SLAVE) {
246 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
247 mnt->mnt_master = old;
248 CLEAR_MNT_SHARED(mnt);
249 } else {
250 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
251 list_add(&mnt->mnt_share, &old->mnt_share);
252 if (IS_MNT_SLAVE(old))
253 list_add(&mnt->mnt_slave, &old->mnt_slave);
254 mnt->mnt_master = old->mnt_master;
255 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500256 if (flag & CL_MAKE_SHARED)
257 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 /* stick the duplicate mount on the same expiry list
260 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500261 if (flag & CL_EXPIRE) {
262 spin_lock(&vfsmount_lock);
263 if (!list_empty(&old->mnt_expire))
264 list_add(&mnt->mnt_expire, &old->mnt_expire);
265 spin_unlock(&vfsmount_lock);
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 }
268 return mnt;
269}
270
Al Viro7b7b1ac2005-11-07 17:13:39 -0500271static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct super_block *sb = mnt->mnt_sb;
274 dput(mnt->mnt_root);
275 free_vfsmnt(mnt);
276 deactivate_super(sb);
277}
278
Al Viro7b7b1ac2005-11-07 17:13:39 -0500279void mntput_no_expire(struct vfsmount *mnt)
280{
281repeat:
282 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
283 if (likely(!mnt->mnt_pinned)) {
284 spin_unlock(&vfsmount_lock);
285 __mntput(mnt);
286 return;
287 }
288 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
289 mnt->mnt_pinned = 0;
290 spin_unlock(&vfsmount_lock);
291 acct_auto_close_mnt(mnt);
292 security_sb_umount_close(mnt);
293 goto repeat;
294 }
295}
296
297EXPORT_SYMBOL(mntput_no_expire);
298
299void mnt_pin(struct vfsmount *mnt)
300{
301 spin_lock(&vfsmount_lock);
302 mnt->mnt_pinned++;
303 spin_unlock(&vfsmount_lock);
304}
305
306EXPORT_SYMBOL(mnt_pin);
307
308void mnt_unpin(struct vfsmount *mnt)
309{
310 spin_lock(&vfsmount_lock);
311 if (mnt->mnt_pinned) {
312 atomic_inc(&mnt->mnt_count);
313 mnt->mnt_pinned--;
314 }
315 spin_unlock(&vfsmount_lock);
316}
317
318EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320/* iterator */
321static void *m_start(struct seq_file *m, loff_t *pos)
322{
323 struct namespace *n = m->private;
324 struct list_head *p;
325 loff_t l = *pos;
326
Ram Pai390c6842005-11-07 17:17:51 -0500327 down_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 list_for_each(p, &n->list)
329 if (!l--)
330 return list_entry(p, struct vfsmount, mnt_list);
331 return NULL;
332}
333
334static void *m_next(struct seq_file *m, void *v, loff_t *pos)
335{
336 struct namespace *n = m->private;
337 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
338 (*pos)++;
Ram Paib58fed82005-11-07 17:16:09 -0500339 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342static void m_stop(struct seq_file *m, void *v)
343{
Ram Pai390c6842005-11-07 17:17:51 -0500344 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
347static inline void mangle(struct seq_file *m, const char *s)
348{
349 seq_escape(m, s, " \t\n\\");
350}
351
352static int show_vfsmnt(struct seq_file *m, void *v)
353{
354 struct vfsmount *mnt = v;
355 int err = 0;
356 static struct proc_fs_info {
357 int flag;
358 char *str;
359 } fs_info[] = {
360 { MS_SYNCHRONOUS, ",sync" },
361 { MS_DIRSYNC, ",dirsync" },
362 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 { 0, NULL }
364 };
365 static struct proc_fs_info mnt_info[] = {
366 { MNT_NOSUID, ",nosuid" },
367 { MNT_NODEV, ",nodev" },
368 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800369 { MNT_NOATIME, ",noatime" },
370 { MNT_NODIRATIME, ",nodiratime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 { 0, NULL }
372 };
373 struct proc_fs_info *fs_infop;
374
375 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
376 seq_putc(m, ' ');
377 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
378 seq_putc(m, ' ');
379 mangle(m, mnt->mnt_sb->s_type->name);
380 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
381 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
382 if (mnt->mnt_sb->s_flags & fs_infop->flag)
383 seq_puts(m, fs_infop->str);
384 }
385 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
386 if (mnt->mnt_flags & fs_infop->flag)
387 seq_puts(m, fs_infop->str);
388 }
389 if (mnt->mnt_sb->s_op->show_options)
390 err = mnt->mnt_sb->s_op->show_options(m, mnt);
391 seq_puts(m, " 0 0\n");
392 return err;
393}
394
395struct seq_operations mounts_op = {
396 .start = m_start,
397 .next = m_next,
398 .stop = m_stop,
399 .show = show_vfsmnt
400};
401
Chuck Leverb4629fe2006-03-20 13:44:12 -0500402static int show_vfsstat(struct seq_file *m, void *v)
403{
404 struct vfsmount *mnt = v;
405 int err = 0;
406
407 /* device */
408 if (mnt->mnt_devname) {
409 seq_puts(m, "device ");
410 mangle(m, mnt->mnt_devname);
411 } else
412 seq_puts(m, "no device");
413
414 /* mount point */
415 seq_puts(m, " mounted on ");
416 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
417 seq_putc(m, ' ');
418
419 /* file system type */
420 seq_puts(m, "with fstype ");
421 mangle(m, mnt->mnt_sb->s_type->name);
422
423 /* optional statistics */
424 if (mnt->mnt_sb->s_op->show_stats) {
425 seq_putc(m, ' ');
426 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
427 }
428
429 seq_putc(m, '\n');
430 return err;
431}
432
433struct seq_operations mountstats_op = {
434 .start = m_start,
435 .next = m_next,
436 .stop = m_stop,
437 .show = show_vfsstat,
438};
439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440/**
441 * may_umount_tree - check if a mount tree is busy
442 * @mnt: root of mount tree
443 *
444 * This is called to check if a tree of mounts has any
445 * open files, pwds, chroots or sub mounts that are
446 * busy.
447 */
448int may_umount_tree(struct vfsmount *mnt)
449{
Ram Pai36341f62005-11-07 17:17:22 -0500450 int actual_refs = 0;
451 int minimum_refs = 0;
452 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500455 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 actual_refs += atomic_read(&p->mnt_count);
457 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 }
459 spin_unlock(&vfsmount_lock);
460
461 if (actual_refs > minimum_refs)
462 return -EBUSY;
463
464 return 0;
465}
466
467EXPORT_SYMBOL(may_umount_tree);
468
469/**
470 * may_umount - check if a mount point is busy
471 * @mnt: root of mount
472 *
473 * This is called to check if a mount point has any
474 * open files, pwds, chroots or sub mounts. If the
475 * mount has sub mounts this will return busy
476 * regardless of whether the sub mounts are busy.
477 *
478 * Doesn't take quota and stuff into account. IOW, in some cases it will
479 * give false negatives. The main reason why it's here is that we need
480 * a non-destructive way to look for easily umountable filesystems.
481 */
482int may_umount(struct vfsmount *mnt)
483{
Ram Paia05964f2005-11-07 17:20:17 -0500484 int ret = 0;
485 spin_lock(&vfsmount_lock);
486 if (propagate_mount_busy(mnt, 2))
487 ret = -EBUSY;
488 spin_unlock(&vfsmount_lock);
489 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
492EXPORT_SYMBOL(may_umount);
493
Ram Paib90fa9a2005-11-07 17:19:50 -0500494void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500496 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800497 while (!list_empty(head)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500498 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
499 list_del_init(&mnt->mnt_hash);
500 if (mnt->mnt_parent != mnt) {
501 struct dentry *dentry;
502 struct vfsmount *m;
503 spin_lock(&vfsmount_lock);
504 dentry = mnt->mnt_mountpoint;
505 m = mnt->mnt_parent;
506 mnt->mnt_mountpoint = mnt->mnt_root;
507 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500509 dput(dentry);
510 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500513 }
514}
515
Ram Paia05964f2005-11-07 17:20:17 -0500516void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500517{
518 struct vfsmount *p;
519
520 for (p = mnt; p; p = next_mnt(p, mnt)) {
521 list_del(&p->mnt_hash);
522 list_add(&p->mnt_hash, kill);
523 }
524
Ram Paia05964f2005-11-07 17:20:17 -0500525 if (propagate)
526 propagate_umount(kill);
527
Ram Pai70fbcdf2005-11-07 17:17:04 -0500528 list_for_each_entry(p, kill, mnt_hash) {
529 list_del_init(&p->mnt_expire);
530 list_del_init(&p->mnt_list);
531 __touch_namespace(p->mnt_namespace);
532 p->mnt_namespace = NULL;
533 list_del_init(&p->mnt_child);
534 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500535 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500536 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538}
539
540static int do_umount(struct vfsmount *mnt, int flags)
541{
Ram Paib58fed82005-11-07 17:16:09 -0500542 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500544 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
546 retval = security_sb_umount(mnt, flags);
547 if (retval)
548 return retval;
549
550 /*
551 * Allow userspace to request a mountpoint be expired rather than
552 * unmounting unconditionally. Unmount only happens if:
553 * (1) the mark is already set (the mark is cleared by mntput())
554 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
555 */
556 if (flags & MNT_EXPIRE) {
557 if (mnt == current->fs->rootmnt ||
558 flags & (MNT_FORCE | MNT_DETACH))
559 return -EINVAL;
560
561 if (atomic_read(&mnt->mnt_count) != 2)
562 return -EBUSY;
563
564 if (!xchg(&mnt->mnt_expiry_mark, 1))
565 return -EAGAIN;
566 }
567
568 /*
569 * If we may have to abort operations to get out of this
570 * mount, and they will themselves hold resources we must
571 * allow the fs to do things. In the Unix tradition of
572 * 'Gee thats tricky lets do it in userspace' the umount_begin
573 * might fail to complete on the first run through as other tasks
574 * must return, and the like. Thats for the mount program to worry
575 * about for the moment.
576 */
577
578 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -0500579 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 sb->s_op->umount_begin(sb);
581 unlock_kernel();
582
583 /*
584 * No sense to grab the lock for this test, but test itself looks
585 * somewhat bogus. Suggestions for better replacement?
586 * Ho-hum... In principle, we might treat that as umount + switch
587 * to rootfs. GC would eventually take care of the old vfsmount.
588 * Actually it makes sense, especially if rootfs would contain a
589 * /reboot - static binary that would close all descriptors and
590 * call reboot(9). Then init(8) could umount root and exec /reboot.
591 */
592 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
593 /*
594 * Special case for "unmounting" root ...
595 * we just try to remount it readonly.
596 */
597 down_write(&sb->s_umount);
598 if (!(sb->s_flags & MS_RDONLY)) {
599 lock_kernel();
600 DQUOT_OFF(sb);
601 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
602 unlock_kernel();
603 }
604 up_write(&sb->s_umount);
605 return retval;
606 }
607
Ram Pai390c6842005-11-07 17:17:51 -0500608 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500610 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500613 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500615 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 retval = 0;
617 }
618 spin_unlock(&vfsmount_lock);
619 if (retval)
620 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500621 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500622 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 return retval;
624}
625
626/*
627 * Now umount can handle mount points as well as block devices.
628 * This is important for filesystems which use unnamed block devices.
629 *
630 * We now support a flag for forced unmount like the other 'big iron'
631 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
632 */
633
634asmlinkage long sys_umount(char __user * name, int flags)
635{
636 struct nameidata nd;
637 int retval;
638
639 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
640 if (retval)
641 goto out;
642 retval = -EINVAL;
643 if (nd.dentry != nd.mnt->mnt_root)
644 goto dput_and_out;
645 if (!check_mnt(nd.mnt))
646 goto dput_and_out;
647
648 retval = -EPERM;
649 if (!capable(CAP_SYS_ADMIN))
650 goto dput_and_out;
651
652 retval = do_umount(nd.mnt, flags);
653dput_and_out:
654 path_release_on_umount(&nd);
655out:
656 return retval;
657}
658
659#ifdef __ARCH_WANT_SYS_OLDUMOUNT
660
661/*
Ram Paib58fed82005-11-07 17:16:09 -0500662 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664asmlinkage long sys_oldumount(char __user * name)
665{
Ram Paib58fed82005-11-07 17:16:09 -0500666 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
669#endif
670
671static int mount_is_safe(struct nameidata *nd)
672{
673 if (capable(CAP_SYS_ADMIN))
674 return 0;
675 return -EPERM;
676#ifdef notyet
677 if (S_ISLNK(nd->dentry->d_inode->i_mode))
678 return -EPERM;
679 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
680 if (current->uid != nd->dentry->d_inode->i_uid)
681 return -EPERM;
682 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800683 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 return -EPERM;
685 return 0;
686#endif
687}
688
Ram Paib58fed82005-11-07 17:16:09 -0500689static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 while (1) {
692 if (d == dentry)
693 return 1;
694 if (d == NULL || d == d->d_parent)
695 return 0;
696 d = d->d_parent;
697 }
698}
699
Ram Paib90fa9a2005-11-07 17:19:50 -0500700struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500701 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct nameidata nd;
705
Ram Pai9676f0c2005-11-07 17:21:20 -0500706 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
707 return NULL;
708
Ram Pai36341f62005-11-07 17:17:22 -0500709 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if (!q)
711 goto Enomem;
712 q->mnt_mountpoint = mnt->mnt_mountpoint;
713
714 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700715 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
717 continue;
718
719 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500720 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
721 s = skip_mnt_tree(s);
722 continue;
723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 while (p != s->mnt_parent) {
725 p = p->mnt_parent;
726 q = q->mnt_parent;
727 }
728 p = s;
729 nd.mnt = q;
730 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500731 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 if (!q)
733 goto Enomem;
734 spin_lock(&vfsmount_lock);
735 list_add_tail(&q->mnt_list, &res->mnt_list);
736 attach_mnt(q, &nd);
737 spin_unlock(&vfsmount_lock);
738 }
739 }
740 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500741Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500743 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500745 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500747 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749 return NULL;
750}
751
Ram Paib90fa9a2005-11-07 17:19:50 -0500752/*
753 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500754 * @nd : place the mount tree @source_mnt is attached
755 * @parent_nd : if non-null, detach the source_mnt from its parent and
756 * store the parent mount and mountpoint dentry.
757 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500758 *
759 * NOTE: in the table below explains the semantics when a source mount
760 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500761 * ---------------------------------------------------------------------------
762 * | BIND MOUNT OPERATION |
763 * |**************************************************************************
764 * | source-->| shared | private | slave | unbindable |
765 * | dest | | | | |
766 * | | | | | | |
767 * | v | | | | |
768 * |**************************************************************************
769 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
770 * | | | | | |
771 * |non-shared| shared (+) | private | slave (*) | invalid |
772 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500773 * A bind operation clones the source mount and mounts the clone on the
774 * destination mount.
775 *
776 * (++) the cloned mount is propagated to all the mounts in the propagation
777 * tree of the destination mount and the cloned mount is added to
778 * the peer group of the source mount.
779 * (+) the cloned mount is created under the destination mount and is marked
780 * as shared. The cloned mount is added to the peer group of the source
781 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500782 * (+++) the mount is propagated to all the mounts in the propagation tree
783 * of the destination mount and the cloned mount is made slave
784 * of the same master as that of the source mount. The cloned mount
785 * is marked as 'shared and slave'.
786 * (*) the cloned mount is made a slave of the same master as that of the
787 * source mount.
788 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500789 * ---------------------------------------------------------------------------
790 * | MOVE MOUNT OPERATION |
791 * |**************************************************************************
792 * | source-->| shared | private | slave | unbindable |
793 * | dest | | | | |
794 * | | | | | | |
795 * | v | | | | |
796 * |**************************************************************************
797 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
798 * | | | | | |
799 * |non-shared| shared (+*) | private | slave (*) | unbindable |
800 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500801 *
802 * (+) the mount is moved to the destination. And is then propagated to
803 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500804 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500805 * (+++) the mount is moved to the destination and is then propagated to
806 * all the mounts belonging to the destination mount's propagation tree.
807 * the mount is marked as 'shared and slave'.
808 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500809 *
810 * if the source mount is a tree, the operations explained above is
811 * applied to each mount in the tree.
812 * Must be called without spinlocks held, since this function can sleep
813 * in allocations.
814 */
815static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500816 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500817{
818 LIST_HEAD(tree_list);
819 struct vfsmount *dest_mnt = nd->mnt;
820 struct dentry *dest_dentry = nd->dentry;
821 struct vfsmount *child, *p;
822
823 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
824 return -EINVAL;
825
826 if (IS_MNT_SHARED(dest_mnt)) {
827 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
828 set_mnt_shared(p);
829 }
830
831 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500832 if (parent_nd) {
833 detach_mnt(source_mnt, parent_nd);
834 attach_mnt(source_mnt, nd);
835 touch_namespace(current->namespace);
836 } else {
837 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
838 commit_tree(source_mnt);
839 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500840
841 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
842 list_del_init(&child->mnt_hash);
843 commit_tree(child);
844 }
845 spin_unlock(&vfsmount_lock);
846 return 0;
847}
848
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
850{
851 int err;
852 if (mnt->mnt_sb->s_flags & MS_NOUSER)
853 return -EINVAL;
854
855 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
856 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
857 return -ENOTDIR;
858
859 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800860 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 if (IS_DEADDIR(nd->dentry->d_inode))
862 goto out_unlock;
863
864 err = security_sb_check_sb(mnt, nd);
865 if (err)
866 goto out_unlock;
867
868 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500869 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500870 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800872 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (!err)
874 security_sb_post_addmount(mnt, nd);
875 return err;
876}
877
878/*
Ram Pai07b20882005-11-07 17:19:07 -0500879 * recursively change the type of the mountpoint.
880 */
881static int do_change_type(struct nameidata *nd, int flag)
882{
883 struct vfsmount *m, *mnt = nd->mnt;
884 int recurse = flag & MS_REC;
885 int type = flag & ~MS_REC;
886
887 if (nd->dentry != nd->mnt->mnt_root)
888 return -EINVAL;
889
890 down_write(&namespace_sem);
891 spin_lock(&vfsmount_lock);
892 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
893 change_mnt_propagation(m, type);
894 spin_unlock(&vfsmount_lock);
895 up_write(&namespace_sem);
896 return 0;
897}
898
899/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 * do loopback mount.
901 */
902static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
903{
904 struct nameidata old_nd;
905 struct vfsmount *mnt = NULL;
906 int err = mount_is_safe(nd);
907 if (err)
908 return err;
909 if (!old_name || !*old_name)
910 return -EINVAL;
911 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
912 if (err)
913 return err;
914
Ram Pai390c6842005-11-07 17:17:51 -0500915 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500917 if (IS_MNT_UNBINDABLE(old_nd.mnt))
918 goto out;
919
Al Viroccd48bc2005-11-07 17:15:04 -0500920 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
921 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Al Viroccd48bc2005-11-07 17:15:04 -0500923 err = -ENOMEM;
924 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500925 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500926 else
Ram Pai36341f62005-11-07 17:17:22 -0500927 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500928
929 if (!mnt)
930 goto out;
931
Al Viroccd48bc2005-11-07 17:15:04 -0500932 err = graft_tree(mnt, nd);
933 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500934 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500936 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500938 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -0500939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Al Viroccd48bc2005-11-07 17:15:04 -0500941out:
Ram Pai390c6842005-11-07 17:17:51 -0500942 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 path_release(&old_nd);
944 return err;
945}
946
947/*
948 * change filesystem flags. dir should be a physical root of filesystem.
949 * If you've mounted a non-root directory somewhere and want to do remount
950 * on it - tough luck.
951 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
953 void *data)
954{
955 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500956 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 if (!capable(CAP_SYS_ADMIN))
959 return -EPERM;
960
961 if (!check_mnt(nd->mnt))
962 return -EINVAL;
963
964 if (nd->dentry != nd->mnt->mnt_root)
965 return -EINVAL;
966
967 down_write(&sb->s_umount);
968 err = do_remount_sb(sb, flags, data, 0);
969 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500970 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 up_write(&sb->s_umount);
972 if (!err)
973 security_sb_post_remount(nd->mnt, flags, data);
974 return err;
975}
976
Ram Pai9676f0c2005-11-07 17:21:20 -0500977static inline int tree_contains_unbindable(struct vfsmount *mnt)
978{
979 struct vfsmount *p;
980 for (p = mnt; p; p = next_mnt(p, mnt)) {
981 if (IS_MNT_UNBINDABLE(p))
982 return 1;
983 }
984 return 0;
985}
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987static int do_move_mount(struct nameidata *nd, char *old_name)
988{
989 struct nameidata old_nd, parent_nd;
990 struct vfsmount *p;
991 int err = 0;
992 if (!capable(CAP_SYS_ADMIN))
993 return -EPERM;
994 if (!old_name || !*old_name)
995 return -EINVAL;
996 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
997 if (err)
998 return err;
999
Ram Pai390c6842005-11-07 17:17:51 -05001000 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -05001001 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 ;
1003 err = -EINVAL;
1004 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1005 goto out;
1006
1007 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001008 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (IS_DEADDIR(nd->dentry->d_inode))
1010 goto out1;
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001013 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
1015 err = -EINVAL;
1016 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001017 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001020 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1023 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001024 goto out1;
1025 /*
1026 * Don't move a mount residing in a shared parent.
1027 */
1028 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1029 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001030 /*
1031 * Don't move a mount tree containing unbindable mounts to a destination
1032 * mount which is shared.
1033 */
1034 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1035 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -05001037 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001039 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Ram Pai21444402005-11-07 17:20:03 -05001041 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1042 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Ram Pai21444402005-11-07 17:20:03 -05001044 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 /* if the mount is moved, it should no longer be expire
1046 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001047 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 spin_unlock(&vfsmount_lock);
1049out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001050 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051out:
Ram Pai390c6842005-11-07 17:17:51 -05001052 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 if (!err)
1054 path_release(&parent_nd);
1055 path_release(&old_nd);
1056 return err;
1057}
1058
1059/*
1060 * create a new mount for userspace and request it to be added into the
1061 * namespace's tree
1062 */
1063static int do_new_mount(struct nameidata *nd, char *type, int flags,
1064 int mnt_flags, char *name, void *data)
1065{
1066 struct vfsmount *mnt;
1067
1068 if (!type || !memchr(type, 0, PAGE_SIZE))
1069 return -EINVAL;
1070
1071 /* we need capabilities... */
1072 if (!capable(CAP_SYS_ADMIN))
1073 return -EPERM;
1074
1075 mnt = do_kern_mount(type, flags, name, data);
1076 if (IS_ERR(mnt))
1077 return PTR_ERR(mnt);
1078
1079 return do_add_mount(mnt, nd, mnt_flags, NULL);
1080}
1081
1082/*
1083 * add a mount into a namespace's mount tree
1084 * - provide the option of adding the new mount to an expiration list
1085 */
1086int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1087 int mnt_flags, struct list_head *fslist)
1088{
1089 int err;
1090
Ram Pai390c6842005-11-07 17:17:51 -05001091 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001093 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 ;
1095 err = -EINVAL;
1096 if (!check_mnt(nd->mnt))
1097 goto unlock;
1098
1099 /* Refuse the same filesystem on the same mount point */
1100 err = -EBUSY;
1101 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1102 nd->mnt->mnt_root == nd->dentry)
1103 goto unlock;
1104
1105 err = -EINVAL;
1106 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1107 goto unlock;
1108
1109 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001110 if ((err = graft_tree(newmnt, nd)))
1111 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001113 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 /* add to the specified expiration list */
1115 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001116 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 spin_unlock(&vfsmount_lock);
1118 }
Ram Pai390c6842005-11-07 17:17:51 -05001119 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001120 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121
1122unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001123 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 mntput(newmnt);
1125 return err;
1126}
1127
1128EXPORT_SYMBOL_GPL(do_add_mount);
1129
Ram Pai70fbcdf2005-11-07 17:17:04 -05001130static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1131 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001132{
1133 spin_lock(&vfsmount_lock);
1134
1135 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001136 * Check if mount is still attached, if not, let whoever holds it deal
1137 * with the sucker
1138 */
1139 if (mnt->mnt_parent == mnt) {
1140 spin_unlock(&vfsmount_lock);
1141 return;
1142 }
1143
1144 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001145 * Check that it is still dead: the count should now be 2 - as
1146 * contributed by the vfsmount parent and the mntget above
1147 */
Ram Paia05964f2005-11-07 17:20:17 -05001148 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001149 /* delete from the namespace */
Al Viro5addc5d2005-11-07 17:15:49 -05001150 touch_namespace(mnt->mnt_namespace);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001151 list_del_init(&mnt->mnt_list);
Miklos Szerediac081152005-07-07 17:57:27 -07001152 mnt->mnt_namespace = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001153 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001154 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001155 } else {
1156 /*
1157 * Someone brought it back to life whilst we didn't have any
1158 * locks held so return it to the expiration list
1159 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001160 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001161 spin_unlock(&vfsmount_lock);
1162 }
1163}
1164
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165/*
1166 * process a list of expirable mountpoints with the intent of discarding any
1167 * mountpoints that aren't in use and haven't been touched since last we came
1168 * here
1169 */
1170void mark_mounts_for_expiry(struct list_head *mounts)
1171{
1172 struct namespace *namespace;
1173 struct vfsmount *mnt, *next;
1174 LIST_HEAD(graveyard);
1175
1176 if (list_empty(mounts))
1177 return;
1178
1179 spin_lock(&vfsmount_lock);
1180
1181 /* extract from the expiration list every vfsmount that matches the
1182 * following criteria:
1183 * - only referenced by its parent vfsmount
1184 * - still marked for expiry (marked on the last call here; marks are
1185 * cleared by mntput())
1186 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001187 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1189 atomic_read(&mnt->mnt_count) != 1)
1190 continue;
1191
1192 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001193 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 }
1195
1196 /*
1197 * go through the vfsmounts we've just consigned to the graveyard to
1198 * - check that they're still dead
1199 * - delete the vfsmount from the appropriate namespace under lock
1200 * - dispose of the corpse
1201 */
1202 while (!list_empty(&graveyard)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001203 LIST_HEAD(umounts);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001204 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1205 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 /* don't do anything if the namespace is dead - all the
1208 * vfsmounts from it are going away anyway */
1209 namespace = mnt->mnt_namespace;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001210 if (!namespace || !namespace->root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 continue;
1212 get_namespace(namespace);
1213
1214 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001215 down_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001216 expire_mount(mnt, mounts, &umounts);
Ram Pai390c6842005-11-07 17:17:51 -05001217 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001218 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 mntput(mnt);
1220 put_namespace(namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 spin_lock(&vfsmount_lock);
1222 }
1223
1224 spin_unlock(&vfsmount_lock);
1225}
1226
1227EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1228
1229/*
1230 * Some copy_from_user() implementations do not return the exact number of
1231 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1232 * Note that this function differs from copy_from_user() in that it will oops
1233 * on bad values of `to', rather than returning a short copy.
1234 */
Ram Paib58fed82005-11-07 17:16:09 -05001235static long exact_copy_from_user(void *to, const void __user * from,
1236 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237{
1238 char *t = to;
1239 const char __user *f = from;
1240 char c;
1241
1242 if (!access_ok(VERIFY_READ, from, n))
1243 return n;
1244
1245 while (n) {
1246 if (__get_user(c, f)) {
1247 memset(t, 0, n);
1248 break;
1249 }
1250 *t++ = c;
1251 f++;
1252 n--;
1253 }
1254 return n;
1255}
1256
Ram Paib58fed82005-11-07 17:16:09 -05001257int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258{
1259 int i;
1260 unsigned long page;
1261 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 *where = 0;
1264 if (!data)
1265 return 0;
1266
1267 if (!(page = __get_free_page(GFP_KERNEL)))
1268 return -ENOMEM;
1269
1270 /* We only care that *some* data at the address the user
1271 * gave us is valid. Just in case, we'll zero
1272 * the remainder of the page.
1273 */
1274 /* copy_from_user cannot cross TASK_SIZE ! */
1275 size = TASK_SIZE - (unsigned long)data;
1276 if (size > PAGE_SIZE)
1277 size = PAGE_SIZE;
1278
1279 i = size - exact_copy_from_user((void *)page, data, size);
1280 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001281 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 return -EFAULT;
1283 }
1284 if (i != PAGE_SIZE)
1285 memset((char *)page + i, 0, PAGE_SIZE - i);
1286 *where = page;
1287 return 0;
1288}
1289
1290/*
1291 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1292 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1293 *
1294 * data is a (void *) that can point to any structure up to
1295 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1296 * information (or be NULL).
1297 *
1298 * Pre-0.97 versions of mount() didn't have a flags word.
1299 * When the flags word was introduced its top half was required
1300 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1301 * Therefore, if this magic number is present, it carries no information
1302 * and must be discarded.
1303 */
Ram Paib58fed82005-11-07 17:16:09 -05001304long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 unsigned long flags, void *data_page)
1306{
1307 struct nameidata nd;
1308 int retval = 0;
1309 int mnt_flags = 0;
1310
1311 /* Discard magic */
1312 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1313 flags &= ~MS_MGC_MSK;
1314
1315 /* Basic sanity checks */
1316
1317 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1318 return -EINVAL;
1319 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1320 return -EINVAL;
1321
1322 if (data_page)
1323 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1324
1325 /* Separate the per-mountpoint flags */
1326 if (flags & MS_NOSUID)
1327 mnt_flags |= MNT_NOSUID;
1328 if (flags & MS_NODEV)
1329 mnt_flags |= MNT_NODEV;
1330 if (flags & MS_NOEXEC)
1331 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001332 if (flags & MS_NOATIME)
1333 mnt_flags |= MNT_NOATIME;
1334 if (flags & MS_NODIRATIME)
1335 mnt_flags |= MNT_NODIRATIME;
1336
1337 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1338 MS_NOATIME | MS_NODIRATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
1340 /* ... and get the mountpoint */
1341 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1342 if (retval)
1343 return retval;
1344
1345 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1346 if (retval)
1347 goto dput_out;
1348
1349 if (flags & MS_REMOUNT)
1350 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1351 data_page);
1352 else if (flags & MS_BIND)
1353 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001354 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001355 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 else if (flags & MS_MOVE)
1357 retval = do_move_mount(&nd, dev_name);
1358 else
1359 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1360 dev_name, data_page);
1361dput_out:
1362 path_release(&nd);
1363 return retval;
1364}
1365
JANAK DESAI741a2952006-02-07 12:59:00 -08001366/*
1367 * Allocate a new namespace structure and populate it with contents
1368 * copied from the namespace of the passed in task structure.
1369 */
1370struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{
1372 struct namespace *namespace = tsk->namespace;
1373 struct namespace *new_ns;
1374 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 struct vfsmount *p, *q;
1376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1378 if (!new_ns)
Adrian Bunkf13b8352006-03-15 17:37:32 +01001379 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001383 init_waitqueue_head(&new_ns->poll);
1384 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Ram Pai390c6842005-11-07 17:17:51 -05001386 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 /* First pass: copy the tree topology */
Ram Pai36341f62005-11-07 17:17:22 -05001388 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001389 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001391 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 kfree(new_ns);
Adrian Bunkf13b8352006-03-15 17:37:32 +01001393 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 }
1395 spin_lock(&vfsmount_lock);
1396 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1397 spin_unlock(&vfsmount_lock);
1398
1399 /*
1400 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1401 * as belonging to new namespace. We have already acquired a private
1402 * fs_struct, so tsk->fs->lock is not needed.
1403 */
1404 p = namespace->root;
1405 q = new_ns->root;
1406 while (p) {
1407 q->mnt_namespace = new_ns;
1408 if (fs) {
1409 if (p == fs->rootmnt) {
1410 rootmnt = p;
1411 fs->rootmnt = mntget(q);
1412 }
1413 if (p == fs->pwdmnt) {
1414 pwdmnt = p;
1415 fs->pwdmnt = mntget(q);
1416 }
1417 if (p == fs->altrootmnt) {
1418 altrootmnt = p;
1419 fs->altrootmnt = mntget(q);
1420 }
1421 }
1422 p = next_mnt(p, namespace->root);
1423 q = next_mnt(q, new_ns->root);
1424 }
Ram Pai390c6842005-11-07 17:17:51 -05001425 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (rootmnt)
1428 mntput(rootmnt);
1429 if (pwdmnt)
1430 mntput(pwdmnt);
1431 if (altrootmnt)
1432 mntput(altrootmnt);
1433
JANAK DESAI741a2952006-02-07 12:59:00 -08001434 return new_ns;
1435}
1436
1437int copy_namespace(int flags, struct task_struct *tsk)
1438{
1439 struct namespace *namespace = tsk->namespace;
1440 struct namespace *new_ns;
1441 int err = 0;
1442
1443 if (!namespace)
1444 return 0;
1445
1446 get_namespace(namespace);
1447
1448 if (!(flags & CLONE_NEWNS))
1449 return 0;
1450
1451 if (!capable(CAP_SYS_ADMIN)) {
1452 err = -EPERM;
1453 goto out;
1454 }
1455
1456 new_ns = dup_namespace(tsk, tsk->fs);
1457 if (!new_ns) {
1458 err = -ENOMEM;
1459 goto out;
1460 }
1461
1462 tsk->namespace = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
1464out:
1465 put_namespace(namespace);
JANAK DESAI741a2952006-02-07 12:59:00 -08001466 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
1468
1469asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1470 char __user * type, unsigned long flags,
1471 void __user * data)
1472{
1473 int retval;
1474 unsigned long data_page;
1475 unsigned long type_page;
1476 unsigned long dev_page;
1477 char *dir_page;
1478
Ram Paib58fed82005-11-07 17:16:09 -05001479 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (retval < 0)
1481 return retval;
1482
1483 dir_page = getname(dir_name);
1484 retval = PTR_ERR(dir_page);
1485 if (IS_ERR(dir_page))
1486 goto out1;
1487
Ram Paib58fed82005-11-07 17:16:09 -05001488 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 if (retval < 0)
1490 goto out2;
1491
Ram Paib58fed82005-11-07 17:16:09 -05001492 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (retval < 0)
1494 goto out3;
1495
1496 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001497 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1498 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 unlock_kernel();
1500 free_page(data_page);
1501
1502out3:
1503 free_page(dev_page);
1504out2:
1505 putname(dir_page);
1506out1:
1507 free_page(type_page);
1508 return retval;
1509}
1510
1511/*
1512 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1513 * It can block. Requires the big lock held.
1514 */
1515void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1516 struct dentry *dentry)
1517{
1518 struct dentry *old_root;
1519 struct vfsmount *old_rootmnt;
1520 write_lock(&fs->lock);
1521 old_root = fs->root;
1522 old_rootmnt = fs->rootmnt;
1523 fs->rootmnt = mntget(mnt);
1524 fs->root = dget(dentry);
1525 write_unlock(&fs->lock);
1526 if (old_root) {
1527 dput(old_root);
1528 mntput(old_rootmnt);
1529 }
1530}
1531
1532/*
1533 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1534 * It can block. Requires the big lock held.
1535 */
1536void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1537 struct dentry *dentry)
1538{
1539 struct dentry *old_pwd;
1540 struct vfsmount *old_pwdmnt;
1541
1542 write_lock(&fs->lock);
1543 old_pwd = fs->pwd;
1544 old_pwdmnt = fs->pwdmnt;
1545 fs->pwdmnt = mntget(mnt);
1546 fs->pwd = dget(dentry);
1547 write_unlock(&fs->lock);
1548
1549 if (old_pwd) {
1550 dput(old_pwd);
1551 mntput(old_pwdmnt);
1552 }
1553}
1554
1555static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1556{
1557 struct task_struct *g, *p;
1558 struct fs_struct *fs;
1559
1560 read_lock(&tasklist_lock);
1561 do_each_thread(g, p) {
1562 task_lock(p);
1563 fs = p->fs;
1564 if (fs) {
1565 atomic_inc(&fs->count);
1566 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001567 if (fs->root == old_nd->dentry
1568 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001570 if (fs->pwd == old_nd->dentry
1571 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1573 put_fs_struct(fs);
1574 } else
1575 task_unlock(p);
1576 } while_each_thread(g, p);
1577 read_unlock(&tasklist_lock);
1578}
1579
1580/*
1581 * pivot_root Semantics:
1582 * Moves the root file system of the current process to the directory put_old,
1583 * makes new_root as the new root file system of the current process, and sets
1584 * root/cwd of all processes which had them on the current root to new_root.
1585 *
1586 * Restrictions:
1587 * The new_root and put_old must be directories, and must not be on the
1588 * same file system as the current process root. The put_old must be
1589 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1590 * pointed to by put_old must yield the same directory as new_root. No other
1591 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1592 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001593 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1594 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1595 * in this situation.
1596 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 * Notes:
1598 * - we don't move root/cwd if they are not at the root (reason: if something
1599 * cared enough to change them, it's probably wrong to force them elsewhere)
1600 * - it's okay to pick a root that isn't the root of a file system, e.g.
1601 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1602 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1603 * first.
1604 */
Ram Paib58fed82005-11-07 17:16:09 -05001605asmlinkage long sys_pivot_root(const char __user * new_root,
1606 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607{
1608 struct vfsmount *tmp;
1609 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1610 int error;
1611
1612 if (!capable(CAP_SYS_ADMIN))
1613 return -EPERM;
1614
1615 lock_kernel();
1616
Ram Paib58fed82005-11-07 17:16:09 -05001617 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1618 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 if (error)
1620 goto out0;
1621 error = -EINVAL;
1622 if (!check_mnt(new_nd.mnt))
1623 goto out1;
1624
Ram Paib58fed82005-11-07 17:16:09 -05001625 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 if (error)
1627 goto out1;
1628
1629 error = security_sb_pivotroot(&old_nd, &new_nd);
1630 if (error) {
1631 path_release(&old_nd);
1632 goto out1;
1633 }
1634
1635 read_lock(&current->fs->lock);
1636 user_nd.mnt = mntget(current->fs->rootmnt);
1637 user_nd.dentry = dget(current->fs->root);
1638 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001639 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001640 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001642 if (IS_MNT_SHARED(old_nd.mnt) ||
1643 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1644 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1645 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 if (!check_mnt(user_nd.mnt))
1647 goto out2;
1648 error = -ENOENT;
1649 if (IS_DEADDIR(new_nd.dentry->d_inode))
1650 goto out2;
1651 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1652 goto out2;
1653 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1654 goto out2;
1655 error = -EBUSY;
1656 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1657 goto out2; /* loop, on the same file system */
1658 error = -EINVAL;
1659 if (user_nd.mnt->mnt_root != user_nd.dentry)
1660 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001661 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1662 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (new_nd.mnt->mnt_root != new_nd.dentry)
1664 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001665 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1666 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1668 spin_lock(&vfsmount_lock);
1669 if (tmp != new_nd.mnt) {
1670 for (;;) {
1671 if (tmp->mnt_parent == tmp)
1672 goto out3; /* already mounted on put_old */
1673 if (tmp->mnt_parent == new_nd.mnt)
1674 break;
1675 tmp = tmp->mnt_parent;
1676 }
1677 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1678 goto out3;
1679 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1680 goto out3;
1681 detach_mnt(new_nd.mnt, &parent_nd);
1682 detach_mnt(user_nd.mnt, &root_parent);
1683 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1684 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Al Viro5addc5d2005-11-07 17:15:49 -05001685 touch_namespace(current->namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 spin_unlock(&vfsmount_lock);
1687 chroot_fs_refs(&user_nd, &new_nd);
1688 security_sb_post_pivotroot(&user_nd, &new_nd);
1689 error = 0;
1690 path_release(&root_parent);
1691 path_release(&parent_nd);
1692out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001693 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001694 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 path_release(&user_nd);
1696 path_release(&old_nd);
1697out1:
1698 path_release(&new_nd);
1699out0:
1700 unlock_kernel();
1701 return error;
1702out3:
1703 spin_unlock(&vfsmount_lock);
1704 goto out2;
1705}
1706
1707static void __init init_mount_tree(void)
1708{
1709 struct vfsmount *mnt;
1710 struct namespace *namespace;
1711 struct task_struct *g, *p;
1712
1713 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1714 if (IS_ERR(mnt))
1715 panic("Can't create rootfs");
1716 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1717 if (!namespace)
1718 panic("Can't allocate initial namespace");
1719 atomic_set(&namespace->count, 1);
1720 INIT_LIST_HEAD(&namespace->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001721 init_waitqueue_head(&namespace->poll);
1722 namespace->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 list_add(&mnt->mnt_list, &namespace->list);
1724 namespace->root = mnt;
1725 mnt->mnt_namespace = namespace;
1726
1727 init_task.namespace = namespace;
1728 read_lock(&tasklist_lock);
1729 do_each_thread(g, p) {
1730 get_namespace(namespace);
1731 p->namespace = namespace;
1732 } while_each_thread(g, p);
1733 read_unlock(&tasklist_lock);
1734
1735 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1736 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1737}
1738
1739void __init mnt_init(unsigned long mempages)
1740{
1741 struct list_head *d;
1742 unsigned int nr_hash;
1743 int i;
1744
Ram Pai390c6842005-11-07 17:17:51 -05001745 init_rwsem(&namespace_sem);
1746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Ram Paib58fed82005-11-07 17:16:09 -05001748 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
Ram Paib58fed82005-11-07 17:16:09 -05001750 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752 if (!mount_hashtable)
1753 panic("Failed to allocate mount hash table\n");
1754
1755 /*
1756 * Find the power-of-two list-heads that can fit into the allocation..
1757 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1758 * a power-of-two.
1759 */
1760 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1761 hash_bits = 0;
1762 do {
1763 hash_bits++;
1764 } while ((nr_hash >> hash_bits) != 0);
1765 hash_bits--;
1766
1767 /*
1768 * Re-calculate the actual number of entries and the mask
1769 * from the number of bits we can fit.
1770 */
1771 nr_hash = 1UL << hash_bits;
Ram Paib58fed82005-11-07 17:16:09 -05001772 hash_mask = nr_hash - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 printk("Mount-cache hash table entries: %d\n", nr_hash);
1775
1776 /* And initialize the newly allocated array */
1777 d = mount_hashtable;
1778 i = nr_hash;
1779 do {
1780 INIT_LIST_HEAD(d);
1781 d++;
1782 i--;
1783 } while (i);
1784 sysfs_init();
Miklos Szeredif87fd4c2006-01-16 22:14:23 -08001785 subsystem_register(&fs_subsys);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 init_rootfs();
1787 init_mount_tree();
1788}
1789
1790void __put_namespace(struct namespace *namespace)
1791{
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001792 struct vfsmount *root = namespace->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001793 LIST_HEAD(umount_list);
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001794 namespace->root = NULL;
1795 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001796 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001798 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001800 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001801 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 kfree(namespace);
1803}