]>
Commit | Line | Data |
---|---|---|
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 | ||
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> | |
19 | #include <linux/module.h> | |
20 | #include <linux/seq_file.h> | |
21 | #include <linux/namespace.h> | |
22 | #include <linux/namei.h> | |
23 | #include <linux/security.h> | |
24 | #include <linux/mount.h> | |
25 | #include <asm/uaccess.h> | |
26 | #include <asm/unistd.h> | |
07b20889 | 27 | #include "pnode.h" |
1da177e4 LT |
28 | |
29 | extern int __init init_rootfs(void); | |
30 | ||
31 | #ifdef CONFIG_SYSFS | |
32 | extern int __init sysfs_init(void); | |
33 | #else | |
34 | static inline int sysfs_init(void) | |
35 | { | |
36 | return 0; | |
37 | } | |
38 | #endif | |
39 | ||
40 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ | |
5addc5dd AV |
41 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
42 | ||
43 | static int event; | |
1da177e4 LT |
44 | |
45 | static struct list_head *mount_hashtable; | |
6c231b7b | 46 | static int hash_mask __read_mostly, hash_bits __read_mostly; |
b58fed8b | 47 | static kmem_cache_t *mnt_cache; |
390c6843 | 48 | static struct rw_semaphore namespace_sem; |
1da177e4 LT |
49 | |
50 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) | |
51 | { | |
b58fed8b RP |
52 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
53 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
1da177e4 LT |
54 | tmp = tmp + (tmp >> hash_bits); |
55 | return tmp & hash_mask; | |
56 | } | |
57 | ||
58 | struct vfsmount *alloc_vfsmnt(const char *name) | |
59 | { | |
b58fed8b | 60 | struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); |
1da177e4 LT |
61 | if (mnt) { |
62 | memset(mnt, 0, sizeof(struct vfsmount)); | |
b58fed8b | 63 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
64 | INIT_LIST_HEAD(&mnt->mnt_hash); |
65 | INIT_LIST_HEAD(&mnt->mnt_child); | |
66 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
67 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 68 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 69 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
70 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
71 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
1da177e4 | 72 | if (name) { |
b58fed8b | 73 | int size = strlen(name) + 1; |
1da177e4 LT |
74 | char *newname = kmalloc(size, GFP_KERNEL); |
75 | if (newname) { | |
76 | memcpy(newname, name, size); | |
77 | mnt->mnt_devname = newname; | |
78 | } | |
79 | } | |
80 | } | |
81 | return mnt; | |
82 | } | |
83 | ||
84 | void free_vfsmnt(struct vfsmount *mnt) | |
85 | { | |
86 | kfree(mnt->mnt_devname); | |
87 | kmem_cache_free(mnt_cache, mnt); | |
88 | } | |
89 | ||
90 | /* | |
a05964f3 RP |
91 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
92 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 93 | */ |
a05964f3 RP |
94 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
95 | int dir) | |
1da177e4 | 96 | { |
b58fed8b RP |
97 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
98 | struct list_head *tmp = head; | |
1da177e4 LT |
99 | struct vfsmount *p, *found = NULL; |
100 | ||
1da177e4 | 101 | for (;;) { |
a05964f3 | 102 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
103 | p = NULL; |
104 | if (tmp == head) | |
105 | break; | |
106 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
107 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 108 | found = p; |
1da177e4 LT |
109 | break; |
110 | } | |
111 | } | |
1da177e4 LT |
112 | return found; |
113 | } | |
114 | ||
a05964f3 RP |
115 | /* |
116 | * lookup_mnt increments the ref count before returning | |
117 | * the vfsmount struct. | |
118 | */ | |
119 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
120 | { | |
121 | struct vfsmount *child_mnt; | |
122 | spin_lock(&vfsmount_lock); | |
123 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
124 | mntget(child_mnt); | |
125 | spin_unlock(&vfsmount_lock); | |
126 | return child_mnt; | |
127 | } | |
128 | ||
1da177e4 LT |
129 | static inline int check_mnt(struct vfsmount *mnt) |
130 | { | |
131 | return mnt->mnt_namespace == current->namespace; | |
132 | } | |
133 | ||
5addc5dd AV |
134 | static void touch_namespace(struct namespace *ns) |
135 | { | |
136 | if (ns) { | |
137 | ns->event = ++event; | |
138 | wake_up_interruptible(&ns->poll); | |
139 | } | |
140 | } | |
141 | ||
142 | static void __touch_namespace(struct namespace *ns) | |
143 | { | |
144 | if (ns && ns->event != event) { | |
145 | ns->event = event; | |
146 | wake_up_interruptible(&ns->poll); | |
147 | } | |
148 | } | |
149 | ||
1da177e4 LT |
150 | static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) |
151 | { | |
152 | old_nd->dentry = mnt->mnt_mountpoint; | |
153 | old_nd->mnt = mnt->mnt_parent; | |
154 | mnt->mnt_parent = mnt; | |
155 | mnt->mnt_mountpoint = mnt->mnt_root; | |
156 | list_del_init(&mnt->mnt_child); | |
157 | list_del_init(&mnt->mnt_hash); | |
158 | old_nd->dentry->d_mounted--; | |
159 | } | |
160 | ||
b90fa9ae RP |
161 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
162 | struct vfsmount *child_mnt) | |
163 | { | |
164 | child_mnt->mnt_parent = mntget(mnt); | |
165 | child_mnt->mnt_mountpoint = dget(dentry); | |
166 | dentry->d_mounted++; | |
167 | } | |
168 | ||
1da177e4 LT |
169 | static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) |
170 | { | |
b90fa9ae RP |
171 | mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); |
172 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
173 | hash(nd->mnt, nd->dentry)); | |
1da177e4 | 174 | list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); |
b90fa9ae RP |
175 | } |
176 | ||
177 | /* | |
178 | * the caller must hold vfsmount_lock | |
179 | */ | |
180 | static void commit_tree(struct vfsmount *mnt) | |
181 | { | |
182 | struct vfsmount *parent = mnt->mnt_parent; | |
183 | struct vfsmount *m; | |
184 | LIST_HEAD(head); | |
185 | struct namespace *n = parent->mnt_namespace; | |
186 | ||
187 | BUG_ON(parent == mnt); | |
188 | ||
189 | list_add_tail(&head, &mnt->mnt_list); | |
190 | list_for_each_entry(m, &head, mnt_list) | |
191 | m->mnt_namespace = n; | |
192 | list_splice(&head, n->list.prev); | |
193 | ||
194 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
195 | hash(parent, mnt->mnt_mountpoint)); | |
196 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
197 | touch_namespace(n); | |
1da177e4 LT |
198 | } |
199 | ||
200 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
201 | { | |
202 | struct list_head *next = p->mnt_mounts.next; | |
203 | if (next == &p->mnt_mounts) { | |
204 | while (1) { | |
205 | if (p == root) | |
206 | return NULL; | |
207 | next = p->mnt_child.next; | |
208 | if (next != &p->mnt_parent->mnt_mounts) | |
209 | break; | |
210 | p = p->mnt_parent; | |
211 | } | |
212 | } | |
213 | return list_entry(next, struct vfsmount, mnt_child); | |
214 | } | |
215 | ||
36341f64 RP |
216 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
217 | int flag) | |
1da177e4 LT |
218 | { |
219 | struct super_block *sb = old->mnt_sb; | |
220 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
221 | ||
222 | if (mnt) { | |
223 | mnt->mnt_flags = old->mnt_flags; | |
224 | atomic_inc(&sb->s_active); | |
225 | mnt->mnt_sb = sb; | |
226 | mnt->mnt_root = dget(root); | |
227 | mnt->mnt_mountpoint = mnt->mnt_root; | |
228 | mnt->mnt_parent = mnt; | |
b90fa9ae RP |
229 | |
230 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) | |
231 | list_add(&mnt->mnt_share, &old->mnt_share); | |
232 | if (flag & CL_MAKE_SHARED) | |
233 | set_mnt_shared(mnt); | |
1da177e4 LT |
234 | |
235 | /* stick the duplicate mount on the same expiry list | |
236 | * as the original if that was on one */ | |
36341f64 RP |
237 | if (flag & CL_EXPIRE) { |
238 | spin_lock(&vfsmount_lock); | |
239 | if (!list_empty(&old->mnt_expire)) | |
240 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
241 | spin_unlock(&vfsmount_lock); | |
242 | } | |
1da177e4 LT |
243 | } |
244 | return mnt; | |
245 | } | |
246 | ||
7b7b1ace | 247 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
248 | { |
249 | struct super_block *sb = mnt->mnt_sb; | |
250 | dput(mnt->mnt_root); | |
251 | free_vfsmnt(mnt); | |
252 | deactivate_super(sb); | |
253 | } | |
254 | ||
7b7b1ace AV |
255 | void mntput_no_expire(struct vfsmount *mnt) |
256 | { | |
257 | repeat: | |
258 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
259 | if (likely(!mnt->mnt_pinned)) { | |
260 | spin_unlock(&vfsmount_lock); | |
261 | __mntput(mnt); | |
262 | return; | |
263 | } | |
264 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
265 | mnt->mnt_pinned = 0; | |
266 | spin_unlock(&vfsmount_lock); | |
267 | acct_auto_close_mnt(mnt); | |
268 | security_sb_umount_close(mnt); | |
269 | goto repeat; | |
270 | } | |
271 | } | |
272 | ||
273 | EXPORT_SYMBOL(mntput_no_expire); | |
274 | ||
275 | void mnt_pin(struct vfsmount *mnt) | |
276 | { | |
277 | spin_lock(&vfsmount_lock); | |
278 | mnt->mnt_pinned++; | |
279 | spin_unlock(&vfsmount_lock); | |
280 | } | |
281 | ||
282 | EXPORT_SYMBOL(mnt_pin); | |
283 | ||
284 | void mnt_unpin(struct vfsmount *mnt) | |
285 | { | |
286 | spin_lock(&vfsmount_lock); | |
287 | if (mnt->mnt_pinned) { | |
288 | atomic_inc(&mnt->mnt_count); | |
289 | mnt->mnt_pinned--; | |
290 | } | |
291 | spin_unlock(&vfsmount_lock); | |
292 | } | |
293 | ||
294 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 LT |
295 | |
296 | /* iterator */ | |
297 | static void *m_start(struct seq_file *m, loff_t *pos) | |
298 | { | |
299 | struct namespace *n = m->private; | |
300 | struct list_head *p; | |
301 | loff_t l = *pos; | |
302 | ||
390c6843 | 303 | down_read(&namespace_sem); |
1da177e4 LT |
304 | list_for_each(p, &n->list) |
305 | if (!l--) | |
306 | return list_entry(p, struct vfsmount, mnt_list); | |
307 | return NULL; | |
308 | } | |
309 | ||
310 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
311 | { | |
312 | struct namespace *n = m->private; | |
313 | struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; | |
314 | (*pos)++; | |
b58fed8b | 315 | return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); |
1da177e4 LT |
316 | } |
317 | ||
318 | static void m_stop(struct seq_file *m, void *v) | |
319 | { | |
390c6843 | 320 | up_read(&namespace_sem); |
1da177e4 LT |
321 | } |
322 | ||
323 | static inline void mangle(struct seq_file *m, const char *s) | |
324 | { | |
325 | seq_escape(m, s, " \t\n\\"); | |
326 | } | |
327 | ||
328 | static int show_vfsmnt(struct seq_file *m, void *v) | |
329 | { | |
330 | struct vfsmount *mnt = v; | |
331 | int err = 0; | |
332 | static struct proc_fs_info { | |
333 | int flag; | |
334 | char *str; | |
335 | } fs_info[] = { | |
336 | { MS_SYNCHRONOUS, ",sync" }, | |
337 | { MS_DIRSYNC, ",dirsync" }, | |
338 | { MS_MANDLOCK, ",mand" }, | |
339 | { MS_NOATIME, ",noatime" }, | |
340 | { MS_NODIRATIME, ",nodiratime" }, | |
341 | { 0, NULL } | |
342 | }; | |
343 | static struct proc_fs_info mnt_info[] = { | |
344 | { MNT_NOSUID, ",nosuid" }, | |
345 | { MNT_NODEV, ",nodev" }, | |
346 | { MNT_NOEXEC, ",noexec" }, | |
347 | { 0, NULL } | |
348 | }; | |
349 | struct proc_fs_info *fs_infop; | |
350 | ||
351 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
352 | seq_putc(m, ' '); | |
353 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
354 | seq_putc(m, ' '); | |
355 | mangle(m, mnt->mnt_sb->s_type->name); | |
356 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
357 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
358 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
359 | seq_puts(m, fs_infop->str); | |
360 | } | |
361 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
362 | if (mnt->mnt_flags & fs_infop->flag) | |
363 | seq_puts(m, fs_infop->str); | |
364 | } | |
365 | if (mnt->mnt_sb->s_op->show_options) | |
366 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
367 | seq_puts(m, " 0 0\n"); | |
368 | return err; | |
369 | } | |
370 | ||
371 | struct seq_operations mounts_op = { | |
372 | .start = m_start, | |
373 | .next = m_next, | |
374 | .stop = m_stop, | |
375 | .show = show_vfsmnt | |
376 | }; | |
377 | ||
378 | /** | |
379 | * may_umount_tree - check if a mount tree is busy | |
380 | * @mnt: root of mount tree | |
381 | * | |
382 | * This is called to check if a tree of mounts has any | |
383 | * open files, pwds, chroots or sub mounts that are | |
384 | * busy. | |
385 | */ | |
386 | int may_umount_tree(struct vfsmount *mnt) | |
387 | { | |
36341f64 RP |
388 | int actual_refs = 0; |
389 | int minimum_refs = 0; | |
390 | struct vfsmount *p; | |
1da177e4 LT |
391 | |
392 | spin_lock(&vfsmount_lock); | |
36341f64 | 393 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
394 | actual_refs += atomic_read(&p->mnt_count); |
395 | minimum_refs += 2; | |
1da177e4 LT |
396 | } |
397 | spin_unlock(&vfsmount_lock); | |
398 | ||
399 | if (actual_refs > minimum_refs) | |
400 | return -EBUSY; | |
401 | ||
402 | return 0; | |
403 | } | |
404 | ||
405 | EXPORT_SYMBOL(may_umount_tree); | |
406 | ||
407 | /** | |
408 | * may_umount - check if a mount point is busy | |
409 | * @mnt: root of mount | |
410 | * | |
411 | * This is called to check if a mount point has any | |
412 | * open files, pwds, chroots or sub mounts. If the | |
413 | * mount has sub mounts this will return busy | |
414 | * regardless of whether the sub mounts are busy. | |
415 | * | |
416 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
417 | * give false negatives. The main reason why it's here is that we need | |
418 | * a non-destructive way to look for easily umountable filesystems. | |
419 | */ | |
420 | int may_umount(struct vfsmount *mnt) | |
421 | { | |
a05964f3 RP |
422 | int ret = 0; |
423 | spin_lock(&vfsmount_lock); | |
424 | if (propagate_mount_busy(mnt, 2)) | |
425 | ret = -EBUSY; | |
426 | spin_unlock(&vfsmount_lock); | |
427 | return ret; | |
1da177e4 LT |
428 | } |
429 | ||
430 | EXPORT_SYMBOL(may_umount); | |
431 | ||
b90fa9ae | 432 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
433 | { |
434 | struct vfsmount *mnt; | |
435 | while(!list_empty(head)) { | |
436 | mnt = list_entry(head->next, struct vfsmount, mnt_hash); | |
437 | list_del_init(&mnt->mnt_hash); | |
438 | if (mnt->mnt_parent != mnt) { | |
439 | struct dentry *dentry; | |
440 | struct vfsmount *m; | |
441 | spin_lock(&vfsmount_lock); | |
442 | dentry = mnt->mnt_mountpoint; | |
443 | m = mnt->mnt_parent; | |
444 | mnt->mnt_mountpoint = mnt->mnt_root; | |
445 | mnt->mnt_parent = mnt; | |
446 | spin_unlock(&vfsmount_lock); | |
447 | dput(dentry); | |
448 | mntput(m); | |
449 | } | |
450 | mntput(mnt); | |
451 | } | |
452 | } | |
453 | ||
a05964f3 | 454 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
455 | { |
456 | struct vfsmount *p; | |
1da177e4 LT |
457 | |
458 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
70fbcdf4 RP |
459 | list_del(&p->mnt_hash); |
460 | list_add(&p->mnt_hash, kill); | |
1da177e4 LT |
461 | } |
462 | ||
a05964f3 RP |
463 | if (propagate) |
464 | propagate_umount(kill); | |
465 | ||
70fbcdf4 RP |
466 | list_for_each_entry(p, kill, mnt_hash) { |
467 | list_del_init(&p->mnt_expire); | |
468 | list_del_init(&p->mnt_list); | |
469 | __touch_namespace(p->mnt_namespace); | |
470 | p->mnt_namespace = NULL; | |
471 | list_del_init(&p->mnt_child); | |
472 | if (p->mnt_parent != p) | |
473 | mnt->mnt_mountpoint->d_mounted--; | |
a05964f3 | 474 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
475 | } |
476 | } | |
477 | ||
478 | static int do_umount(struct vfsmount *mnt, int flags) | |
479 | { | |
b58fed8b | 480 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 481 | int retval; |
70fbcdf4 | 482 | LIST_HEAD(umount_list); |
1da177e4 LT |
483 | |
484 | retval = security_sb_umount(mnt, flags); | |
485 | if (retval) | |
486 | return retval; | |
487 | ||
488 | /* | |
489 | * Allow userspace to request a mountpoint be expired rather than | |
490 | * unmounting unconditionally. Unmount only happens if: | |
491 | * (1) the mark is already set (the mark is cleared by mntput()) | |
492 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
493 | */ | |
494 | if (flags & MNT_EXPIRE) { | |
495 | if (mnt == current->fs->rootmnt || | |
496 | flags & (MNT_FORCE | MNT_DETACH)) | |
497 | return -EINVAL; | |
498 | ||
499 | if (atomic_read(&mnt->mnt_count) != 2) | |
500 | return -EBUSY; | |
501 | ||
502 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
503 | return -EAGAIN; | |
504 | } | |
505 | ||
506 | /* | |
507 | * If we may have to abort operations to get out of this | |
508 | * mount, and they will themselves hold resources we must | |
509 | * allow the fs to do things. In the Unix tradition of | |
510 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
511 | * might fail to complete on the first run through as other tasks | |
512 | * must return, and the like. Thats for the mount program to worry | |
513 | * about for the moment. | |
514 | */ | |
515 | ||
516 | lock_kernel(); | |
b58fed8b | 517 | if ((flags & MNT_FORCE) && sb->s_op->umount_begin) |
1da177e4 LT |
518 | sb->s_op->umount_begin(sb); |
519 | unlock_kernel(); | |
520 | ||
521 | /* | |
522 | * No sense to grab the lock for this test, but test itself looks | |
523 | * somewhat bogus. Suggestions for better replacement? | |
524 | * Ho-hum... In principle, we might treat that as umount + switch | |
525 | * to rootfs. GC would eventually take care of the old vfsmount. | |
526 | * Actually it makes sense, especially if rootfs would contain a | |
527 | * /reboot - static binary that would close all descriptors and | |
528 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
529 | */ | |
530 | if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { | |
531 | /* | |
532 | * Special case for "unmounting" root ... | |
533 | * we just try to remount it readonly. | |
534 | */ | |
535 | down_write(&sb->s_umount); | |
536 | if (!(sb->s_flags & MS_RDONLY)) { | |
537 | lock_kernel(); | |
538 | DQUOT_OFF(sb); | |
539 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
540 | unlock_kernel(); | |
541 | } | |
542 | up_write(&sb->s_umount); | |
543 | return retval; | |
544 | } | |
545 | ||
390c6843 | 546 | down_write(&namespace_sem); |
1da177e4 | 547 | spin_lock(&vfsmount_lock); |
5addc5dd | 548 | event++; |
1da177e4 | 549 | |
1da177e4 | 550 | retval = -EBUSY; |
a05964f3 | 551 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 552 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 553 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
554 | retval = 0; |
555 | } | |
556 | spin_unlock(&vfsmount_lock); | |
557 | if (retval) | |
558 | security_sb_umount_busy(mnt); | |
390c6843 | 559 | up_write(&namespace_sem); |
70fbcdf4 | 560 | release_mounts(&umount_list); |
1da177e4 LT |
561 | return retval; |
562 | } | |
563 | ||
564 | /* | |
565 | * Now umount can handle mount points as well as block devices. | |
566 | * This is important for filesystems which use unnamed block devices. | |
567 | * | |
568 | * We now support a flag for forced unmount like the other 'big iron' | |
569 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
570 | */ | |
571 | ||
572 | asmlinkage long sys_umount(char __user * name, int flags) | |
573 | { | |
574 | struct nameidata nd; | |
575 | int retval; | |
576 | ||
577 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
578 | if (retval) | |
579 | goto out; | |
580 | retval = -EINVAL; | |
581 | if (nd.dentry != nd.mnt->mnt_root) | |
582 | goto dput_and_out; | |
583 | if (!check_mnt(nd.mnt)) | |
584 | goto dput_and_out; | |
585 | ||
586 | retval = -EPERM; | |
587 | if (!capable(CAP_SYS_ADMIN)) | |
588 | goto dput_and_out; | |
589 | ||
590 | retval = do_umount(nd.mnt, flags); | |
591 | dput_and_out: | |
592 | path_release_on_umount(&nd); | |
593 | out: | |
594 | return retval; | |
595 | } | |
596 | ||
597 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
598 | ||
599 | /* | |
b58fed8b | 600 | * The 2.0 compatible umount. No flags. |
1da177e4 | 601 | */ |
1da177e4 LT |
602 | asmlinkage long sys_oldumount(char __user * name) |
603 | { | |
b58fed8b | 604 | return sys_umount(name, 0); |
1da177e4 LT |
605 | } |
606 | ||
607 | #endif | |
608 | ||
609 | static int mount_is_safe(struct nameidata *nd) | |
610 | { | |
611 | if (capable(CAP_SYS_ADMIN)) | |
612 | return 0; | |
613 | return -EPERM; | |
614 | #ifdef notyet | |
615 | if (S_ISLNK(nd->dentry->d_inode->i_mode)) | |
616 | return -EPERM; | |
617 | if (nd->dentry->d_inode->i_mode & S_ISVTX) { | |
618 | if (current->uid != nd->dentry->d_inode->i_uid) | |
619 | return -EPERM; | |
620 | } | |
621 | if (permission(nd->dentry->d_inode, MAY_WRITE, nd)) | |
622 | return -EPERM; | |
623 | return 0; | |
624 | #endif | |
625 | } | |
626 | ||
b58fed8b | 627 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
628 | { |
629 | while (1) { | |
630 | if (d == dentry) | |
631 | return 1; | |
632 | if (d == NULL || d == d->d_parent) | |
633 | return 0; | |
634 | d = d->d_parent; | |
635 | } | |
636 | } | |
637 | ||
b90fa9ae | 638 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 639 | int flag) |
1da177e4 LT |
640 | { |
641 | struct vfsmount *res, *p, *q, *r, *s; | |
1da177e4 LT |
642 | struct nameidata nd; |
643 | ||
36341f64 | 644 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
645 | if (!q) |
646 | goto Enomem; | |
647 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
648 | ||
649 | p = mnt; | |
fdadd65f | 650 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
651 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
652 | continue; | |
653 | ||
654 | for (s = r; s; s = next_mnt(s, r)) { | |
655 | while (p != s->mnt_parent) { | |
656 | p = p->mnt_parent; | |
657 | q = q->mnt_parent; | |
658 | } | |
659 | p = s; | |
660 | nd.mnt = q; | |
661 | nd.dentry = p->mnt_mountpoint; | |
36341f64 | 662 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
663 | if (!q) |
664 | goto Enomem; | |
665 | spin_lock(&vfsmount_lock); | |
666 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
667 | attach_mnt(q, &nd); | |
668 | spin_unlock(&vfsmount_lock); | |
669 | } | |
670 | } | |
671 | return res; | |
b58fed8b | 672 | Enomem: |
1da177e4 | 673 | if (res) { |
70fbcdf4 | 674 | LIST_HEAD(umount_list); |
1da177e4 | 675 | spin_lock(&vfsmount_lock); |
a05964f3 | 676 | umount_tree(res, 0, &umount_list); |
1da177e4 | 677 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 678 | release_mounts(&umount_list); |
1da177e4 LT |
679 | } |
680 | return NULL; | |
681 | } | |
682 | ||
b90fa9ae RP |
683 | /* |
684 | * @source_mnt : mount tree to be attached | |
21444403 RP |
685 | * @nd : place the mount tree @source_mnt is attached |
686 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
687 | * store the parent mount and mountpoint dentry. | |
688 | * (done when source_mnt is moved) | |
b90fa9ae RP |
689 | * |
690 | * NOTE: in the table below explains the semantics when a source mount | |
691 | * of a given type is attached to a destination mount of a given type. | |
692 | * --------------------------------------------- | |
693 | * | BIND MOUNT OPERATION | | |
694 | * |******************************************** | |
695 | * | source-->| shared | private | | |
696 | * | dest | | | | |
697 | * | | | | | | |
698 | * | v | | | | |
699 | * |******************************************** | |
700 | * | shared | shared (++) | shared (+) | | |
701 | * | | | | | |
702 | * |non-shared| shared (+) | private | | |
703 | * ********************************************* | |
704 | * A bind operation clones the source mount and mounts the clone on the | |
705 | * destination mount. | |
706 | * | |
707 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
708 | * tree of the destination mount and the cloned mount is added to | |
709 | * the peer group of the source mount. | |
710 | * (+) the cloned mount is created under the destination mount and is marked | |
711 | * as shared. The cloned mount is added to the peer group of the source | |
712 | * mount. | |
21444403 RP |
713 | * --------------------------------------------- |
714 | * | MOVE MOUNT OPERATION | | |
715 | * |******************************************** | |
716 | * | source-->| shared | private | | |
717 | * | dest | | | | |
718 | * | | | | | | |
719 | * | v | | | | |
720 | * |******************************************** | |
721 | * | shared | shared (+) | shared (+) | | |
722 | * | | | | | |
723 | * |non-shared| shared (+*) | private | | |
724 | * ********************************************* | |
725 | * (+) the mount is moved to the destination. And is then propagated to all | |
726 | * the mounts in the propagation tree of the destination mount. | |
727 | * (+*) the mount is moved to the destination. | |
b90fa9ae RP |
728 | * |
729 | * if the source mount is a tree, the operations explained above is | |
730 | * applied to each mount in the tree. | |
731 | * Must be called without spinlocks held, since this function can sleep | |
732 | * in allocations. | |
733 | */ | |
734 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
21444403 | 735 | struct nameidata *nd, struct nameidata *parent_nd) |
b90fa9ae RP |
736 | { |
737 | LIST_HEAD(tree_list); | |
738 | struct vfsmount *dest_mnt = nd->mnt; | |
739 | struct dentry *dest_dentry = nd->dentry; | |
740 | struct vfsmount *child, *p; | |
741 | ||
742 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
743 | return -EINVAL; | |
744 | ||
745 | if (IS_MNT_SHARED(dest_mnt)) { | |
746 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
747 | set_mnt_shared(p); | |
748 | } | |
749 | ||
750 | spin_lock(&vfsmount_lock); | |
21444403 RP |
751 | if (parent_nd) { |
752 | detach_mnt(source_mnt, parent_nd); | |
753 | attach_mnt(source_mnt, nd); | |
754 | touch_namespace(current->namespace); | |
755 | } else { | |
756 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
757 | commit_tree(source_mnt); | |
758 | } | |
b90fa9ae RP |
759 | |
760 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
761 | list_del_init(&child->mnt_hash); | |
762 | commit_tree(child); | |
763 | } | |
764 | spin_unlock(&vfsmount_lock); | |
765 | return 0; | |
766 | } | |
767 | ||
1da177e4 LT |
768 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
769 | { | |
770 | int err; | |
771 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
772 | return -EINVAL; | |
773 | ||
774 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
775 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) | |
776 | return -ENOTDIR; | |
777 | ||
778 | err = -ENOENT; | |
779 | down(&nd->dentry->d_inode->i_sem); | |
780 | if (IS_DEADDIR(nd->dentry->d_inode)) | |
781 | goto out_unlock; | |
782 | ||
783 | err = security_sb_check_sb(mnt, nd); | |
784 | if (err) | |
785 | goto out_unlock; | |
786 | ||
787 | err = -ENOENT; | |
b90fa9ae | 788 | if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) |
21444403 | 789 | err = attach_recursive_mnt(mnt, nd, NULL); |
1da177e4 LT |
790 | out_unlock: |
791 | up(&nd->dentry->d_inode->i_sem); | |
792 | if (!err) | |
793 | security_sb_post_addmount(mnt, nd); | |
794 | return err; | |
795 | } | |
796 | ||
07b20889 RP |
797 | /* |
798 | * recursively change the type of the mountpoint. | |
799 | */ | |
800 | static int do_change_type(struct nameidata *nd, int flag) | |
801 | { | |
802 | struct vfsmount *m, *mnt = nd->mnt; | |
803 | int recurse = flag & MS_REC; | |
804 | int type = flag & ~MS_REC; | |
805 | ||
806 | if (nd->dentry != nd->mnt->mnt_root) | |
807 | return -EINVAL; | |
808 | ||
809 | down_write(&namespace_sem); | |
810 | spin_lock(&vfsmount_lock); | |
811 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
812 | change_mnt_propagation(m, type); | |
813 | spin_unlock(&vfsmount_lock); | |
814 | up_write(&namespace_sem); | |
815 | return 0; | |
816 | } | |
817 | ||
1da177e4 LT |
818 | /* |
819 | * do loopback mount. | |
820 | */ | |
821 | static int do_loopback(struct nameidata *nd, char *old_name, int recurse) | |
822 | { | |
823 | struct nameidata old_nd; | |
824 | struct vfsmount *mnt = NULL; | |
825 | int err = mount_is_safe(nd); | |
826 | if (err) | |
827 | return err; | |
828 | if (!old_name || !*old_name) | |
829 | return -EINVAL; | |
830 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
831 | if (err) | |
832 | return err; | |
833 | ||
390c6843 | 834 | down_write(&namespace_sem); |
1da177e4 | 835 | err = -EINVAL; |
ccd48bc7 AV |
836 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) |
837 | goto out; | |
1da177e4 | 838 | |
ccd48bc7 AV |
839 | err = -ENOMEM; |
840 | if (recurse) | |
36341f64 | 841 | mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 | 842 | else |
36341f64 | 843 | mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 AV |
844 | |
845 | if (!mnt) | |
846 | goto out; | |
847 | ||
ccd48bc7 AV |
848 | err = graft_tree(mnt, nd); |
849 | if (err) { | |
70fbcdf4 | 850 | LIST_HEAD(umount_list); |
1da177e4 | 851 | spin_lock(&vfsmount_lock); |
a05964f3 | 852 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 853 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 854 | release_mounts(&umount_list); |
5b83d2c5 | 855 | } |
1da177e4 | 856 | |
ccd48bc7 | 857 | out: |
390c6843 | 858 | up_write(&namespace_sem); |
1da177e4 LT |
859 | path_release(&old_nd); |
860 | return err; | |
861 | } | |
862 | ||
863 | /* | |
864 | * change filesystem flags. dir should be a physical root of filesystem. | |
865 | * If you've mounted a non-root directory somewhere and want to do remount | |
866 | * on it - tough luck. | |
867 | */ | |
1da177e4 LT |
868 | static int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
869 | void *data) | |
870 | { | |
871 | int err; | |
b58fed8b | 872 | struct super_block *sb = nd->mnt->mnt_sb; |
1da177e4 LT |
873 | |
874 | if (!capable(CAP_SYS_ADMIN)) | |
875 | return -EPERM; | |
876 | ||
877 | if (!check_mnt(nd->mnt)) | |
878 | return -EINVAL; | |
879 | ||
880 | if (nd->dentry != nd->mnt->mnt_root) | |
881 | return -EINVAL; | |
882 | ||
883 | down_write(&sb->s_umount); | |
884 | err = do_remount_sb(sb, flags, data, 0); | |
885 | if (!err) | |
b58fed8b | 886 | nd->mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
887 | up_write(&sb->s_umount); |
888 | if (!err) | |
889 | security_sb_post_remount(nd->mnt, flags, data); | |
890 | return err; | |
891 | } | |
892 | ||
893 | static int do_move_mount(struct nameidata *nd, char *old_name) | |
894 | { | |
895 | struct nameidata old_nd, parent_nd; | |
896 | struct vfsmount *p; | |
897 | int err = 0; | |
898 | if (!capable(CAP_SYS_ADMIN)) | |
899 | return -EPERM; | |
900 | if (!old_name || !*old_name) | |
901 | return -EINVAL; | |
902 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
903 | if (err) | |
904 | return err; | |
905 | ||
390c6843 | 906 | down_write(&namespace_sem); |
b58fed8b | 907 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
908 | ; |
909 | err = -EINVAL; | |
910 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) | |
911 | goto out; | |
912 | ||
913 | err = -ENOENT; | |
914 | down(&nd->dentry->d_inode->i_sem); | |
915 | if (IS_DEADDIR(nd->dentry->d_inode)) | |
916 | goto out1; | |
917 | ||
1da177e4 | 918 | if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) |
21444403 | 919 | goto out1; |
1da177e4 LT |
920 | |
921 | err = -EINVAL; | |
922 | if (old_nd.dentry != old_nd.mnt->mnt_root) | |
21444403 | 923 | goto out1; |
1da177e4 LT |
924 | |
925 | if (old_nd.mnt == old_nd.mnt->mnt_parent) | |
21444403 | 926 | goto out1; |
1da177e4 LT |
927 | |
928 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
929 | S_ISDIR(old_nd.dentry->d_inode->i_mode)) | |
21444403 RP |
930 | goto out1; |
931 | /* | |
932 | * Don't move a mount residing in a shared parent. | |
933 | */ | |
934 | if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) | |
935 | goto out1; | |
1da177e4 | 936 | err = -ELOOP; |
b58fed8b | 937 | for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1da177e4 | 938 | if (p == old_nd.mnt) |
21444403 | 939 | goto out1; |
1da177e4 | 940 | |
21444403 RP |
941 | if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) |
942 | goto out1; | |
1da177e4 | 943 | |
21444403 | 944 | spin_lock(&vfsmount_lock); |
1da177e4 LT |
945 | /* if the mount is moved, it should no longer be expire |
946 | * automatically */ | |
55e700b9 | 947 | list_del_init(&old_nd.mnt->mnt_expire); |
1da177e4 LT |
948 | spin_unlock(&vfsmount_lock); |
949 | out1: | |
950 | up(&nd->dentry->d_inode->i_sem); | |
951 | out: | |
390c6843 | 952 | up_write(&namespace_sem); |
1da177e4 LT |
953 | if (!err) |
954 | path_release(&parent_nd); | |
955 | path_release(&old_nd); | |
956 | return err; | |
957 | } | |
958 | ||
959 | /* | |
960 | * create a new mount for userspace and request it to be added into the | |
961 | * namespace's tree | |
962 | */ | |
963 | static int do_new_mount(struct nameidata *nd, char *type, int flags, | |
964 | int mnt_flags, char *name, void *data) | |
965 | { | |
966 | struct vfsmount *mnt; | |
967 | ||
968 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
969 | return -EINVAL; | |
970 | ||
971 | /* we need capabilities... */ | |
972 | if (!capable(CAP_SYS_ADMIN)) | |
973 | return -EPERM; | |
974 | ||
975 | mnt = do_kern_mount(type, flags, name, data); | |
976 | if (IS_ERR(mnt)) | |
977 | return PTR_ERR(mnt); | |
978 | ||
979 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
980 | } | |
981 | ||
982 | /* | |
983 | * add a mount into a namespace's mount tree | |
984 | * - provide the option of adding the new mount to an expiration list | |
985 | */ | |
986 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
987 | int mnt_flags, struct list_head *fslist) | |
988 | { | |
989 | int err; | |
990 | ||
390c6843 | 991 | down_write(&namespace_sem); |
1da177e4 | 992 | /* Something was mounted here while we slept */ |
b58fed8b | 993 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
994 | ; |
995 | err = -EINVAL; | |
996 | if (!check_mnt(nd->mnt)) | |
997 | goto unlock; | |
998 | ||
999 | /* Refuse the same filesystem on the same mount point */ | |
1000 | err = -EBUSY; | |
1001 | if (nd->mnt->mnt_sb == newmnt->mnt_sb && | |
1002 | nd->mnt->mnt_root == nd->dentry) | |
1003 | goto unlock; | |
1004 | ||
1005 | err = -EINVAL; | |
1006 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1007 | goto unlock; | |
1008 | ||
1009 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1010 | if ((err = graft_tree(newmnt, nd))) |
1011 | goto unlock; | |
1da177e4 | 1012 | |
5b83d2c5 | 1013 | if (fslist) { |
1da177e4 LT |
1014 | /* add to the specified expiration list */ |
1015 | spin_lock(&vfsmount_lock); | |
55e700b9 | 1016 | list_add_tail(&newmnt->mnt_expire, fslist); |
1da177e4 LT |
1017 | spin_unlock(&vfsmount_lock); |
1018 | } | |
390c6843 | 1019 | up_write(&namespace_sem); |
5b83d2c5 | 1020 | return 0; |
1da177e4 LT |
1021 | |
1022 | unlock: | |
390c6843 | 1023 | up_write(&namespace_sem); |
1da177e4 LT |
1024 | mntput(newmnt); |
1025 | return err; | |
1026 | } | |
1027 | ||
1028 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1029 | ||
70fbcdf4 RP |
1030 | static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, |
1031 | struct list_head *umounts) | |
24ca2af1 MS |
1032 | { |
1033 | spin_lock(&vfsmount_lock); | |
1034 | ||
ed42c879 MS |
1035 | /* |
1036 | * Check if mount is still attached, if not, let whoever holds it deal | |
1037 | * with the sucker | |
1038 | */ | |
1039 | if (mnt->mnt_parent == mnt) { | |
1040 | spin_unlock(&vfsmount_lock); | |
1041 | return; | |
1042 | } | |
1043 | ||
24ca2af1 MS |
1044 | /* |
1045 | * Check that it is still dead: the count should now be 2 - as | |
1046 | * contributed by the vfsmount parent and the mntget above | |
1047 | */ | |
a05964f3 | 1048 | if (!propagate_mount_busy(mnt, 2)) { |
24ca2af1 | 1049 | /* delete from the namespace */ |
5addc5dd | 1050 | touch_namespace(mnt->mnt_namespace); |
24ca2af1 | 1051 | list_del_init(&mnt->mnt_list); |
ac081153 | 1052 | mnt->mnt_namespace = NULL; |
a05964f3 | 1053 | umount_tree(mnt, 1, umounts); |
24ca2af1 | 1054 | spin_unlock(&vfsmount_lock); |
24ca2af1 MS |
1055 | } else { |
1056 | /* | |
1057 | * Someone brought it back to life whilst we didn't have any | |
1058 | * locks held so return it to the expiration list | |
1059 | */ | |
55e700b9 | 1060 | list_add_tail(&mnt->mnt_expire, mounts); |
24ca2af1 MS |
1061 | spin_unlock(&vfsmount_lock); |
1062 | } | |
1063 | } | |
1064 | ||
1da177e4 LT |
1065 | /* |
1066 | * process a list of expirable mountpoints with the intent of discarding any | |
1067 | * mountpoints that aren't in use and haven't been touched since last we came | |
1068 | * here | |
1069 | */ | |
1070 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1071 | { | |
1072 | struct namespace *namespace; | |
1073 | struct vfsmount *mnt, *next; | |
1074 | LIST_HEAD(graveyard); | |
1075 | ||
1076 | if (list_empty(mounts)) | |
1077 | return; | |
1078 | ||
1079 | spin_lock(&vfsmount_lock); | |
1080 | ||
1081 | /* extract from the expiration list every vfsmount that matches the | |
1082 | * following criteria: | |
1083 | * - only referenced by its parent vfsmount | |
1084 | * - still marked for expiry (marked on the last call here; marks are | |
1085 | * cleared by mntput()) | |
1086 | */ | |
55e700b9 | 1087 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 LT |
1088 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1089 | atomic_read(&mnt->mnt_count) != 1) | |
1090 | continue; | |
1091 | ||
1092 | mntget(mnt); | |
55e700b9 | 1093 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 LT |
1094 | } |
1095 | ||
1096 | /* | |
1097 | * go through the vfsmounts we've just consigned to the graveyard to | |
1098 | * - check that they're still dead | |
1099 | * - delete the vfsmount from the appropriate namespace under lock | |
1100 | * - dispose of the corpse | |
1101 | */ | |
1102 | while (!list_empty(&graveyard)) { | |
70fbcdf4 | 1103 | LIST_HEAD(umounts); |
55e700b9 MS |
1104 | mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire); |
1105 | list_del_init(&mnt->mnt_expire); | |
1da177e4 LT |
1106 | |
1107 | /* don't do anything if the namespace is dead - all the | |
1108 | * vfsmounts from it are going away anyway */ | |
1109 | namespace = mnt->mnt_namespace; | |
1ce88cf4 | 1110 | if (!namespace || !namespace->root) |
1da177e4 LT |
1111 | continue; |
1112 | get_namespace(namespace); | |
1113 | ||
1114 | spin_unlock(&vfsmount_lock); | |
390c6843 | 1115 | down_write(&namespace_sem); |
70fbcdf4 | 1116 | expire_mount(mnt, mounts, &umounts); |
390c6843 | 1117 | up_write(&namespace_sem); |
70fbcdf4 | 1118 | release_mounts(&umounts); |
1da177e4 LT |
1119 | mntput(mnt); |
1120 | put_namespace(namespace); | |
1da177e4 LT |
1121 | spin_lock(&vfsmount_lock); |
1122 | } | |
1123 | ||
1124 | spin_unlock(&vfsmount_lock); | |
1125 | } | |
1126 | ||
1127 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1128 | ||
1129 | /* | |
1130 | * Some copy_from_user() implementations do not return the exact number of | |
1131 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1132 | * Note that this function differs from copy_from_user() in that it will oops | |
1133 | * on bad values of `to', rather than returning a short copy. | |
1134 | */ | |
b58fed8b RP |
1135 | static long exact_copy_from_user(void *to, const void __user * from, |
1136 | unsigned long n) | |
1da177e4 LT |
1137 | { |
1138 | char *t = to; | |
1139 | const char __user *f = from; | |
1140 | char c; | |
1141 | ||
1142 | if (!access_ok(VERIFY_READ, from, n)) | |
1143 | return n; | |
1144 | ||
1145 | while (n) { | |
1146 | if (__get_user(c, f)) { | |
1147 | memset(t, 0, n); | |
1148 | break; | |
1149 | } | |
1150 | *t++ = c; | |
1151 | f++; | |
1152 | n--; | |
1153 | } | |
1154 | return n; | |
1155 | } | |
1156 | ||
b58fed8b | 1157 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1158 | { |
1159 | int i; | |
1160 | unsigned long page; | |
1161 | unsigned long size; | |
b58fed8b | 1162 | |
1da177e4 LT |
1163 | *where = 0; |
1164 | if (!data) | |
1165 | return 0; | |
1166 | ||
1167 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1168 | return -ENOMEM; | |
1169 | ||
1170 | /* We only care that *some* data at the address the user | |
1171 | * gave us is valid. Just in case, we'll zero | |
1172 | * the remainder of the page. | |
1173 | */ | |
1174 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1175 | size = TASK_SIZE - (unsigned long)data; | |
1176 | if (size > PAGE_SIZE) | |
1177 | size = PAGE_SIZE; | |
1178 | ||
1179 | i = size - exact_copy_from_user((void *)page, data, size); | |
1180 | if (!i) { | |
b58fed8b | 1181 | free_page(page); |
1da177e4 LT |
1182 | return -EFAULT; |
1183 | } | |
1184 | if (i != PAGE_SIZE) | |
1185 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1186 | *where = page; | |
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1192 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1193 | * | |
1194 | * data is a (void *) that can point to any structure up to | |
1195 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1196 | * information (or be NULL). | |
1197 | * | |
1198 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1199 | * When the flags word was introduced its top half was required | |
1200 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1201 | * Therefore, if this magic number is present, it carries no information | |
1202 | * and must be discarded. | |
1203 | */ | |
b58fed8b | 1204 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1205 | unsigned long flags, void *data_page) |
1206 | { | |
1207 | struct nameidata nd; | |
1208 | int retval = 0; | |
1209 | int mnt_flags = 0; | |
1210 | ||
1211 | /* Discard magic */ | |
1212 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1213 | flags &= ~MS_MGC_MSK; | |
1214 | ||
1215 | /* Basic sanity checks */ | |
1216 | ||
1217 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1218 | return -EINVAL; | |
1219 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1220 | return -EINVAL; | |
1221 | ||
1222 | if (data_page) | |
1223 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1224 | ||
1225 | /* Separate the per-mountpoint flags */ | |
1226 | if (flags & MS_NOSUID) | |
1227 | mnt_flags |= MNT_NOSUID; | |
1228 | if (flags & MS_NODEV) | |
1229 | mnt_flags |= MNT_NODEV; | |
1230 | if (flags & MS_NOEXEC) | |
1231 | mnt_flags |= MNT_NOEXEC; | |
b58fed8b | 1232 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE); |
1da177e4 LT |
1233 | |
1234 | /* ... and get the mountpoint */ | |
1235 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1236 | if (retval) | |
1237 | return retval; | |
1238 | ||
1239 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1240 | if (retval) | |
1241 | goto dput_out; | |
1242 | ||
1243 | if (flags & MS_REMOUNT) | |
1244 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1245 | data_page); | |
1246 | else if (flags & MS_BIND) | |
1247 | retval = do_loopback(&nd, dev_name, flags & MS_REC); | |
a58b0eb8 | 1248 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE)) |
07b20889 | 1249 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1250 | else if (flags & MS_MOVE) |
1251 | retval = do_move_mount(&nd, dev_name); | |
1252 | else | |
1253 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1254 | dev_name, data_page); | |
1255 | dput_out: | |
1256 | path_release(&nd); | |
1257 | return retval; | |
1258 | } | |
1259 | ||
1260 | int copy_namespace(int flags, struct task_struct *tsk) | |
1261 | { | |
1262 | struct namespace *namespace = tsk->namespace; | |
1263 | struct namespace *new_ns; | |
1264 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; | |
1265 | struct fs_struct *fs = tsk->fs; | |
1266 | struct vfsmount *p, *q; | |
1267 | ||
1268 | if (!namespace) | |
1269 | return 0; | |
1270 | ||
1271 | get_namespace(namespace); | |
1272 | ||
1273 | if (!(flags & CLONE_NEWNS)) | |
1274 | return 0; | |
1275 | ||
1276 | if (!capable(CAP_SYS_ADMIN)) { | |
1277 | put_namespace(namespace); | |
1278 | return -EPERM; | |
1279 | } | |
1280 | ||
1281 | new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); | |
1282 | if (!new_ns) | |
1283 | goto out; | |
1284 | ||
1285 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1286 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1287 | init_waitqueue_head(&new_ns->poll); |
1288 | new_ns->event = 0; | |
1da177e4 | 1289 | |
390c6843 | 1290 | down_write(&namespace_sem); |
1da177e4 | 1291 | /* First pass: copy the tree topology */ |
36341f64 RP |
1292 | new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, |
1293 | CL_EXPIRE); | |
1da177e4 | 1294 | if (!new_ns->root) { |
390c6843 | 1295 | up_write(&namespace_sem); |
1da177e4 LT |
1296 | kfree(new_ns); |
1297 | goto out; | |
1298 | } | |
1299 | spin_lock(&vfsmount_lock); | |
1300 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1301 | spin_unlock(&vfsmount_lock); | |
1302 | ||
1303 | /* | |
1304 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1305 | * as belonging to new namespace. We have already acquired a private | |
1306 | * fs_struct, so tsk->fs->lock is not needed. | |
1307 | */ | |
1308 | p = namespace->root; | |
1309 | q = new_ns->root; | |
1310 | while (p) { | |
1311 | q->mnt_namespace = new_ns; | |
1312 | if (fs) { | |
1313 | if (p == fs->rootmnt) { | |
1314 | rootmnt = p; | |
1315 | fs->rootmnt = mntget(q); | |
1316 | } | |
1317 | if (p == fs->pwdmnt) { | |
1318 | pwdmnt = p; | |
1319 | fs->pwdmnt = mntget(q); | |
1320 | } | |
1321 | if (p == fs->altrootmnt) { | |
1322 | altrootmnt = p; | |
1323 | fs->altrootmnt = mntget(q); | |
1324 | } | |
1325 | } | |
1326 | p = next_mnt(p, namespace->root); | |
1327 | q = next_mnt(q, new_ns->root); | |
1328 | } | |
390c6843 | 1329 | up_write(&namespace_sem); |
1da177e4 LT |
1330 | |
1331 | tsk->namespace = new_ns; | |
1332 | ||
1333 | if (rootmnt) | |
1334 | mntput(rootmnt); | |
1335 | if (pwdmnt) | |
1336 | mntput(pwdmnt); | |
1337 | if (altrootmnt) | |
1338 | mntput(altrootmnt); | |
1339 | ||
1340 | put_namespace(namespace); | |
1341 | return 0; | |
1342 | ||
1343 | out: | |
1344 | put_namespace(namespace); | |
1345 | return -ENOMEM; | |
1346 | } | |
1347 | ||
1348 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1349 | char __user * type, unsigned long flags, | |
1350 | void __user * data) | |
1351 | { | |
1352 | int retval; | |
1353 | unsigned long data_page; | |
1354 | unsigned long type_page; | |
1355 | unsigned long dev_page; | |
1356 | char *dir_page; | |
1357 | ||
b58fed8b | 1358 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1359 | if (retval < 0) |
1360 | return retval; | |
1361 | ||
1362 | dir_page = getname(dir_name); | |
1363 | retval = PTR_ERR(dir_page); | |
1364 | if (IS_ERR(dir_page)) | |
1365 | goto out1; | |
1366 | ||
b58fed8b | 1367 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1368 | if (retval < 0) |
1369 | goto out2; | |
1370 | ||
b58fed8b | 1371 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1372 | if (retval < 0) |
1373 | goto out3; | |
1374 | ||
1375 | lock_kernel(); | |
b58fed8b RP |
1376 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1377 | flags, (void *)data_page); | |
1da177e4 LT |
1378 | unlock_kernel(); |
1379 | free_page(data_page); | |
1380 | ||
1381 | out3: | |
1382 | free_page(dev_page); | |
1383 | out2: | |
1384 | putname(dir_page); | |
1385 | out1: | |
1386 | free_page(type_page); | |
1387 | return retval; | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1392 | * It can block. Requires the big lock held. | |
1393 | */ | |
1394 | void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, | |
1395 | struct dentry *dentry) | |
1396 | { | |
1397 | struct dentry *old_root; | |
1398 | struct vfsmount *old_rootmnt; | |
1399 | write_lock(&fs->lock); | |
1400 | old_root = fs->root; | |
1401 | old_rootmnt = fs->rootmnt; | |
1402 | fs->rootmnt = mntget(mnt); | |
1403 | fs->root = dget(dentry); | |
1404 | write_unlock(&fs->lock); | |
1405 | if (old_root) { | |
1406 | dput(old_root); | |
1407 | mntput(old_rootmnt); | |
1408 | } | |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1413 | * It can block. Requires the big lock held. | |
1414 | */ | |
1415 | void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, | |
1416 | struct dentry *dentry) | |
1417 | { | |
1418 | struct dentry *old_pwd; | |
1419 | struct vfsmount *old_pwdmnt; | |
1420 | ||
1421 | write_lock(&fs->lock); | |
1422 | old_pwd = fs->pwd; | |
1423 | old_pwdmnt = fs->pwdmnt; | |
1424 | fs->pwdmnt = mntget(mnt); | |
1425 | fs->pwd = dget(dentry); | |
1426 | write_unlock(&fs->lock); | |
1427 | ||
1428 | if (old_pwd) { | |
1429 | dput(old_pwd); | |
1430 | mntput(old_pwdmnt); | |
1431 | } | |
1432 | } | |
1433 | ||
1434 | static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) | |
1435 | { | |
1436 | struct task_struct *g, *p; | |
1437 | struct fs_struct *fs; | |
1438 | ||
1439 | read_lock(&tasklist_lock); | |
1440 | do_each_thread(g, p) { | |
1441 | task_lock(p); | |
1442 | fs = p->fs; | |
1443 | if (fs) { | |
1444 | atomic_inc(&fs->count); | |
1445 | task_unlock(p); | |
b58fed8b RP |
1446 | if (fs->root == old_nd->dentry |
1447 | && fs->rootmnt == old_nd->mnt) | |
1da177e4 | 1448 | set_fs_root(fs, new_nd->mnt, new_nd->dentry); |
b58fed8b RP |
1449 | if (fs->pwd == old_nd->dentry |
1450 | && fs->pwdmnt == old_nd->mnt) | |
1da177e4 LT |
1451 | set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); |
1452 | put_fs_struct(fs); | |
1453 | } else | |
1454 | task_unlock(p); | |
1455 | } while_each_thread(g, p); | |
1456 | read_unlock(&tasklist_lock); | |
1457 | } | |
1458 | ||
1459 | /* | |
1460 | * pivot_root Semantics: | |
1461 | * Moves the root file system of the current process to the directory put_old, | |
1462 | * makes new_root as the new root file system of the current process, and sets | |
1463 | * root/cwd of all processes which had them on the current root to new_root. | |
1464 | * | |
1465 | * Restrictions: | |
1466 | * The new_root and put_old must be directories, and must not be on the | |
1467 | * same file system as the current process root. The put_old must be | |
1468 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1469 | * pointed to by put_old must yield the same directory as new_root. No other | |
1470 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1471 | * | |
1472 | * Notes: | |
1473 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1474 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1475 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1476 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1477 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1478 | * first. | |
1479 | */ | |
b58fed8b RP |
1480 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1481 | const char __user * put_old) | |
1da177e4 LT |
1482 | { |
1483 | struct vfsmount *tmp; | |
1484 | struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; | |
1485 | int error; | |
1486 | ||
1487 | if (!capable(CAP_SYS_ADMIN)) | |
1488 | return -EPERM; | |
1489 | ||
1490 | lock_kernel(); | |
1491 | ||
b58fed8b RP |
1492 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1493 | &new_nd); | |
1da177e4 LT |
1494 | if (error) |
1495 | goto out0; | |
1496 | error = -EINVAL; | |
1497 | if (!check_mnt(new_nd.mnt)) | |
1498 | goto out1; | |
1499 | ||
b58fed8b | 1500 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1501 | if (error) |
1502 | goto out1; | |
1503 | ||
1504 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1505 | if (error) { | |
1506 | path_release(&old_nd); | |
1507 | goto out1; | |
1508 | } | |
1509 | ||
1510 | read_lock(¤t->fs->lock); | |
1511 | user_nd.mnt = mntget(current->fs->rootmnt); | |
1512 | user_nd.dentry = dget(current->fs->root); | |
1513 | read_unlock(¤t->fs->lock); | |
390c6843 | 1514 | down_write(&namespace_sem); |
1da177e4 LT |
1515 | down(&old_nd.dentry->d_inode->i_sem); |
1516 | error = -EINVAL; | |
21444403 RP |
1517 | if (IS_MNT_SHARED(old_nd.mnt) || |
1518 | IS_MNT_SHARED(new_nd.mnt->mnt_parent) || | |
1519 | IS_MNT_SHARED(user_nd.mnt->mnt_parent)) | |
1520 | goto out2; | |
1da177e4 LT |
1521 | if (!check_mnt(user_nd.mnt)) |
1522 | goto out2; | |
1523 | error = -ENOENT; | |
1524 | if (IS_DEADDIR(new_nd.dentry->d_inode)) | |
1525 | goto out2; | |
1526 | if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) | |
1527 | goto out2; | |
1528 | if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) | |
1529 | goto out2; | |
1530 | error = -EBUSY; | |
1531 | if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) | |
1532 | goto out2; /* loop, on the same file system */ | |
1533 | error = -EINVAL; | |
1534 | if (user_nd.mnt->mnt_root != user_nd.dentry) | |
1535 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1536 | if (user_nd.mnt->mnt_parent == user_nd.mnt) |
1537 | goto out2; /* not attached */ | |
1da177e4 LT |
1538 | if (new_nd.mnt->mnt_root != new_nd.dentry) |
1539 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1540 | if (new_nd.mnt->mnt_parent == new_nd.mnt) |
1541 | goto out2; /* not attached */ | |
1da177e4 LT |
1542 | tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ |
1543 | spin_lock(&vfsmount_lock); | |
1544 | if (tmp != new_nd.mnt) { | |
1545 | for (;;) { | |
1546 | if (tmp->mnt_parent == tmp) | |
1547 | goto out3; /* already mounted on put_old */ | |
1548 | if (tmp->mnt_parent == new_nd.mnt) | |
1549 | break; | |
1550 | tmp = tmp->mnt_parent; | |
1551 | } | |
1552 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) | |
1553 | goto out3; | |
1554 | } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) | |
1555 | goto out3; | |
1556 | detach_mnt(new_nd.mnt, &parent_nd); | |
1557 | detach_mnt(user_nd.mnt, &root_parent); | |
1558 | attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ | |
1559 | attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ | |
5addc5dd | 1560 | touch_namespace(current->namespace); |
1da177e4 LT |
1561 | spin_unlock(&vfsmount_lock); |
1562 | chroot_fs_refs(&user_nd, &new_nd); | |
1563 | security_sb_post_pivotroot(&user_nd, &new_nd); | |
1564 | error = 0; | |
1565 | path_release(&root_parent); | |
1566 | path_release(&parent_nd); | |
1567 | out2: | |
1568 | up(&old_nd.dentry->d_inode->i_sem); | |
390c6843 | 1569 | up_write(&namespace_sem); |
1da177e4 LT |
1570 | path_release(&user_nd); |
1571 | path_release(&old_nd); | |
1572 | out1: | |
1573 | path_release(&new_nd); | |
1574 | out0: | |
1575 | unlock_kernel(); | |
1576 | return error; | |
1577 | out3: | |
1578 | spin_unlock(&vfsmount_lock); | |
1579 | goto out2; | |
1580 | } | |
1581 | ||
1582 | static void __init init_mount_tree(void) | |
1583 | { | |
1584 | struct vfsmount *mnt; | |
1585 | struct namespace *namespace; | |
1586 | struct task_struct *g, *p; | |
1587 | ||
1588 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1589 | if (IS_ERR(mnt)) | |
1590 | panic("Can't create rootfs"); | |
1591 | namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); | |
1592 | if (!namespace) | |
1593 | panic("Can't allocate initial namespace"); | |
1594 | atomic_set(&namespace->count, 1); | |
1595 | INIT_LIST_HEAD(&namespace->list); | |
5addc5dd AV |
1596 | init_waitqueue_head(&namespace->poll); |
1597 | namespace->event = 0; | |
1da177e4 LT |
1598 | list_add(&mnt->mnt_list, &namespace->list); |
1599 | namespace->root = mnt; | |
1600 | mnt->mnt_namespace = namespace; | |
1601 | ||
1602 | init_task.namespace = namespace; | |
1603 | read_lock(&tasklist_lock); | |
1604 | do_each_thread(g, p) { | |
1605 | get_namespace(namespace); | |
1606 | p->namespace = namespace; | |
1607 | } while_each_thread(g, p); | |
1608 | read_unlock(&tasklist_lock); | |
1609 | ||
1610 | set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); | |
1611 | set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); | |
1612 | } | |
1613 | ||
1614 | void __init mnt_init(unsigned long mempages) | |
1615 | { | |
1616 | struct list_head *d; | |
1617 | unsigned int nr_hash; | |
1618 | int i; | |
1619 | ||
390c6843 RP |
1620 | init_rwsem(&namespace_sem); |
1621 | ||
1da177e4 | 1622 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
b58fed8b | 1623 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); |
1da177e4 | 1624 | |
b58fed8b | 1625 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1626 | |
1627 | if (!mount_hashtable) | |
1628 | panic("Failed to allocate mount hash table\n"); | |
1629 | ||
1630 | /* | |
1631 | * Find the power-of-two list-heads that can fit into the allocation.. | |
1632 | * We don't guarantee that "sizeof(struct list_head)" is necessarily | |
1633 | * a power-of-two. | |
1634 | */ | |
1635 | nr_hash = PAGE_SIZE / sizeof(struct list_head); | |
1636 | hash_bits = 0; | |
1637 | do { | |
1638 | hash_bits++; | |
1639 | } while ((nr_hash >> hash_bits) != 0); | |
1640 | hash_bits--; | |
1641 | ||
1642 | /* | |
1643 | * Re-calculate the actual number of entries and the mask | |
1644 | * from the number of bits we can fit. | |
1645 | */ | |
1646 | nr_hash = 1UL << hash_bits; | |
b58fed8b | 1647 | hash_mask = nr_hash - 1; |
1da177e4 LT |
1648 | |
1649 | printk("Mount-cache hash table entries: %d\n", nr_hash); | |
1650 | ||
1651 | /* And initialize the newly allocated array */ | |
1652 | d = mount_hashtable; | |
1653 | i = nr_hash; | |
1654 | do { | |
1655 | INIT_LIST_HEAD(d); | |
1656 | d++; | |
1657 | i--; | |
1658 | } while (i); | |
1659 | sysfs_init(); | |
1660 | init_rootfs(); | |
1661 | init_mount_tree(); | |
1662 | } | |
1663 | ||
1664 | void __put_namespace(struct namespace *namespace) | |
1665 | { | |
1ce88cf4 | 1666 | struct vfsmount *root = namespace->root; |
70fbcdf4 | 1667 | LIST_HEAD(umount_list); |
1ce88cf4 MS |
1668 | namespace->root = NULL; |
1669 | spin_unlock(&vfsmount_lock); | |
390c6843 | 1670 | down_write(&namespace_sem); |
1da177e4 | 1671 | spin_lock(&vfsmount_lock); |
a05964f3 | 1672 | umount_tree(root, 0, &umount_list); |
1da177e4 | 1673 | spin_unlock(&vfsmount_lock); |
390c6843 | 1674 | up_write(&namespace_sem); |
70fbcdf4 | 1675 | release_mounts(&umount_list); |
1da177e4 LT |
1676 | kfree(namespace); |
1677 | } |