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