]>
Commit | Line | Data |
---|---|---|
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/syscalls.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/smp_lock.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/quotaops.h> | |
18 | #include <linux/acct.h> | |
19 | #include <linux/capability.h> | |
20 | #include <linux/module.h> | |
21 | #include <linux/sysfs.h> | |
22 | #include <linux/seq_file.h> | |
23 | #include <linux/mnt_namespace.h> | |
24 | #include <linux/namei.h> | |
25 | #include <linux/security.h> | |
26 | #include <linux/mount.h> | |
27 | #include <linux/ramfs.h> | |
28 | #include <linux/log2.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <asm/unistd.h> | |
31 | #include "pnode.h" | |
32 | #include "internal.h" | |
33 | ||
34 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) | |
35 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
36 | ||
37 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ | |
38 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); | |
39 | ||
40 | static int event; | |
41 | ||
42 | static struct list_head *mount_hashtable __read_mostly; | |
43 | static struct kmem_cache *mnt_cache __read_mostly; | |
44 | static struct rw_semaphore namespace_sem; | |
45 | ||
46 | /* /sys/fs */ | |
47 | struct kobject *fs_kobj; | |
48 | EXPORT_SYMBOL_GPL(fs_kobj); | |
49 | ||
50 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) | |
51 | { | |
52 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); | |
53 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
54 | tmp = tmp + (tmp >> HASH_SHIFT); | |
55 | return tmp & (HASH_SIZE - 1); | |
56 | } | |
57 | ||
58 | struct vfsmount *alloc_vfsmnt(const char *name) | |
59 | { | |
60 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); | |
61 | if (mnt) { | |
62 | atomic_set(&mnt->mnt_count, 1); | |
63 | INIT_LIST_HEAD(&mnt->mnt_hash); | |
64 | INIT_LIST_HEAD(&mnt->mnt_child); | |
65 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
66 | INIT_LIST_HEAD(&mnt->mnt_list); | |
67 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
68 | INIT_LIST_HEAD(&mnt->mnt_share); | |
69 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
70 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
71 | if (name) { | |
72 | int size = strlen(name) + 1; | |
73 | char *newname = kmalloc(size, GFP_KERNEL); | |
74 | if (newname) { | |
75 | memcpy(newname, name, size); | |
76 | mnt->mnt_devname = newname; | |
77 | } | |
78 | } | |
79 | } | |
80 | return mnt; | |
81 | } | |
82 | ||
83 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) | |
84 | { | |
85 | mnt->mnt_sb = sb; | |
86 | mnt->mnt_root = dget(sb->s_root); | |
87 | return 0; | |
88 | } | |
89 | ||
90 | EXPORT_SYMBOL(simple_set_mnt); | |
91 | ||
92 | void free_vfsmnt(struct vfsmount *mnt) | |
93 | { | |
94 | kfree(mnt->mnt_devname); | |
95 | kmem_cache_free(mnt_cache, mnt); | |
96 | } | |
97 | ||
98 | /* | |
99 | * find the first or last mount at @dentry on vfsmount @mnt depending on | |
100 | * @dir. If @dir is set return the first mount else return the last mount. | |
101 | */ | |
102 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, | |
103 | int dir) | |
104 | { | |
105 | struct list_head *head = mount_hashtable + hash(mnt, dentry); | |
106 | struct list_head *tmp = head; | |
107 | struct vfsmount *p, *found = NULL; | |
108 | ||
109 | for (;;) { | |
110 | tmp = dir ? tmp->next : tmp->prev; | |
111 | p = NULL; | |
112 | if (tmp == head) | |
113 | break; | |
114 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
115 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
116 | found = p; | |
117 | break; | |
118 | } | |
119 | } | |
120 | return found; | |
121 | } | |
122 | ||
123 | /* | |
124 | * lookup_mnt increments the ref count before returning | |
125 | * the vfsmount struct. | |
126 | */ | |
127 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
128 | { | |
129 | struct vfsmount *child_mnt; | |
130 | spin_lock(&vfsmount_lock); | |
131 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
132 | mntget(child_mnt); | |
133 | spin_unlock(&vfsmount_lock); | |
134 | return child_mnt; | |
135 | } | |
136 | ||
137 | static inline int check_mnt(struct vfsmount *mnt) | |
138 | { | |
139 | return mnt->mnt_ns == current->nsproxy->mnt_ns; | |
140 | } | |
141 | ||
142 | static void touch_mnt_namespace(struct mnt_namespace *ns) | |
143 | { | |
144 | if (ns) { | |
145 | ns->event = ++event; | |
146 | wake_up_interruptible(&ns->poll); | |
147 | } | |
148 | } | |
149 | ||
150 | static void __touch_mnt_namespace(struct mnt_namespace *ns) | |
151 | { | |
152 | if (ns && ns->event != event) { | |
153 | ns->event = event; | |
154 | wake_up_interruptible(&ns->poll); | |
155 | } | |
156 | } | |
157 | ||
158 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) | |
159 | { | |
160 | old_path->dentry = mnt->mnt_mountpoint; | |
161 | old_path->mnt = mnt->mnt_parent; | |
162 | mnt->mnt_parent = mnt; | |
163 | mnt->mnt_mountpoint = mnt->mnt_root; | |
164 | list_del_init(&mnt->mnt_child); | |
165 | list_del_init(&mnt->mnt_hash); | |
166 | old_path->dentry->d_mounted--; | |
167 | } | |
168 | ||
169 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, | |
170 | struct vfsmount *child_mnt) | |
171 | { | |
172 | child_mnt->mnt_parent = mntget(mnt); | |
173 | child_mnt->mnt_mountpoint = dget(dentry); | |
174 | dentry->d_mounted++; | |
175 | } | |
176 | ||
177 | static void attach_mnt(struct vfsmount *mnt, struct path *path) | |
178 | { | |
179 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); | |
180 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
181 | hash(path->mnt, path->dentry)); | |
182 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
183 | } | |
184 | ||
185 | /* | |
186 | * the caller must hold vfsmount_lock | |
187 | */ | |
188 | static void commit_tree(struct vfsmount *mnt) | |
189 | { | |
190 | struct vfsmount *parent = mnt->mnt_parent; | |
191 | struct vfsmount *m; | |
192 | LIST_HEAD(head); | |
193 | struct mnt_namespace *n = parent->mnt_ns; | |
194 | ||
195 | BUG_ON(parent == mnt); | |
196 | ||
197 | list_add_tail(&head, &mnt->mnt_list); | |
198 | list_for_each_entry(m, &head, mnt_list) | |
199 | m->mnt_ns = n; | |
200 | list_splice(&head, n->list.prev); | |
201 | ||
202 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
203 | hash(parent, mnt->mnt_mountpoint)); | |
204 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
205 | touch_mnt_namespace(n); | |
206 | } | |
207 | ||
208 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
209 | { | |
210 | struct list_head *next = p->mnt_mounts.next; | |
211 | if (next == &p->mnt_mounts) { | |
212 | while (1) { | |
213 | if (p == root) | |
214 | return NULL; | |
215 | next = p->mnt_child.next; | |
216 | if (next != &p->mnt_parent->mnt_mounts) | |
217 | break; | |
218 | p = p->mnt_parent; | |
219 | } | |
220 | } | |
221 | return list_entry(next, struct vfsmount, mnt_child); | |
222 | } | |
223 | ||
224 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) | |
225 | { | |
226 | struct list_head *prev = p->mnt_mounts.prev; | |
227 | while (prev != &p->mnt_mounts) { | |
228 | p = list_entry(prev, struct vfsmount, mnt_child); | |
229 | prev = p->mnt_mounts.prev; | |
230 | } | |
231 | return p; | |
232 | } | |
233 | ||
234 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, | |
235 | int flag) | |
236 | { | |
237 | struct super_block *sb = old->mnt_sb; | |
238 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
239 | ||
240 | if (mnt) { | |
241 | mnt->mnt_flags = old->mnt_flags; | |
242 | atomic_inc(&sb->s_active); | |
243 | mnt->mnt_sb = sb; | |
244 | mnt->mnt_root = dget(root); | |
245 | mnt->mnt_mountpoint = mnt->mnt_root; | |
246 | mnt->mnt_parent = mnt; | |
247 | ||
248 | if (flag & CL_SLAVE) { | |
249 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
250 | mnt->mnt_master = old; | |
251 | CLEAR_MNT_SHARED(mnt); | |
252 | } else if (!(flag & CL_PRIVATE)) { | |
253 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) | |
254 | list_add(&mnt->mnt_share, &old->mnt_share); | |
255 | if (IS_MNT_SLAVE(old)) | |
256 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
257 | mnt->mnt_master = old->mnt_master; | |
258 | } | |
259 | if (flag & CL_MAKE_SHARED) | |
260 | set_mnt_shared(mnt); | |
261 | ||
262 | /* stick the duplicate mount on the same expiry list | |
263 | * as the original if that was on one */ | |
264 | if (flag & CL_EXPIRE) { | |
265 | spin_lock(&vfsmount_lock); | |
266 | if (!list_empty(&old->mnt_expire)) | |
267 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
268 | spin_unlock(&vfsmount_lock); | |
269 | } | |
270 | } | |
271 | return mnt; | |
272 | } | |
273 | ||
274 | static inline void __mntput(struct vfsmount *mnt) | |
275 | { | |
276 | struct super_block *sb = mnt->mnt_sb; | |
277 | dput(mnt->mnt_root); | |
278 | free_vfsmnt(mnt); | |
279 | deactivate_super(sb); | |
280 | } | |
281 | ||
282 | void mntput_no_expire(struct vfsmount *mnt) | |
283 | { | |
284 | repeat: | |
285 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
286 | if (likely(!mnt->mnt_pinned)) { | |
287 | spin_unlock(&vfsmount_lock); | |
288 | __mntput(mnt); | |
289 | return; | |
290 | } | |
291 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
292 | mnt->mnt_pinned = 0; | |
293 | spin_unlock(&vfsmount_lock); | |
294 | acct_auto_close_mnt(mnt); | |
295 | security_sb_umount_close(mnt); | |
296 | goto repeat; | |
297 | } | |
298 | } | |
299 | ||
300 | EXPORT_SYMBOL(mntput_no_expire); | |
301 | ||
302 | void mnt_pin(struct vfsmount *mnt) | |
303 | { | |
304 | spin_lock(&vfsmount_lock); | |
305 | mnt->mnt_pinned++; | |
306 | spin_unlock(&vfsmount_lock); | |
307 | } | |
308 | ||
309 | EXPORT_SYMBOL(mnt_pin); | |
310 | ||
311 | void mnt_unpin(struct vfsmount *mnt) | |
312 | { | |
313 | spin_lock(&vfsmount_lock); | |
314 | if (mnt->mnt_pinned) { | |
315 | atomic_inc(&mnt->mnt_count); | |
316 | mnt->mnt_pinned--; | |
317 | } | |
318 | spin_unlock(&vfsmount_lock); | |
319 | } | |
320 | ||
321 | EXPORT_SYMBOL(mnt_unpin); | |
322 | ||
323 | static inline void mangle(struct seq_file *m, const char *s) | |
324 | { | |
325 | seq_escape(m, s, " \t\n\\"); | |
326 | } | |
327 | ||
328 | /* | |
329 | * Simple .show_options callback for filesystems which don't want to | |
330 | * implement more complex mount option showing. | |
331 | * | |
332 | * See also save_mount_options(). | |
333 | */ | |
334 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
335 | { | |
336 | const char *options = mnt->mnt_sb->s_options; | |
337 | ||
338 | if (options != NULL && options[0]) { | |
339 | seq_putc(m, ','); | |
340 | mangle(m, options); | |
341 | } | |
342 | ||
343 | return 0; | |
344 | } | |
345 | EXPORT_SYMBOL(generic_show_options); | |
346 | ||
347 | /* | |
348 | * If filesystem uses generic_show_options(), this function should be | |
349 | * called from the fill_super() callback. | |
350 | * | |
351 | * The .remount_fs callback usually needs to be handled in a special | |
352 | * way, to make sure, that previous options are not overwritten if the | |
353 | * remount fails. | |
354 | * | |
355 | * Also note, that if the filesystem's .remount_fs function doesn't | |
356 | * reset all options to their default value, but changes only newly | |
357 | * given options, then the displayed options will not reflect reality | |
358 | * any more. | |
359 | */ | |
360 | void save_mount_options(struct super_block *sb, char *options) | |
361 | { | |
362 | kfree(sb->s_options); | |
363 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
364 | } | |
365 | EXPORT_SYMBOL(save_mount_options); | |
366 | ||
367 | /* iterator */ | |
368 | static void *m_start(struct seq_file *m, loff_t *pos) | |
369 | { | |
370 | struct mnt_namespace *n = m->private; | |
371 | ||
372 | down_read(&namespace_sem); | |
373 | return seq_list_start(&n->list, *pos); | |
374 | } | |
375 | ||
376 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
377 | { | |
378 | struct mnt_namespace *n = m->private; | |
379 | ||
380 | return seq_list_next(v, &n->list, pos); | |
381 | } | |
382 | ||
383 | static void m_stop(struct seq_file *m, void *v) | |
384 | { | |
385 | up_read(&namespace_sem); | |
386 | } | |
387 | ||
388 | static int show_vfsmnt(struct seq_file *m, void *v) | |
389 | { | |
390 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
391 | int err = 0; | |
392 | static struct proc_fs_info { | |
393 | int flag; | |
394 | char *str; | |
395 | } fs_info[] = { | |
396 | { MS_SYNCHRONOUS, ",sync" }, | |
397 | { MS_DIRSYNC, ",dirsync" }, | |
398 | { MS_MANDLOCK, ",mand" }, | |
399 | { 0, NULL } | |
400 | }; | |
401 | static struct proc_fs_info mnt_info[] = { | |
402 | { MNT_NOSUID, ",nosuid" }, | |
403 | { MNT_NODEV, ",nodev" }, | |
404 | { MNT_NOEXEC, ",noexec" }, | |
405 | { MNT_NOATIME, ",noatime" }, | |
406 | { MNT_NODIRATIME, ",nodiratime" }, | |
407 | { MNT_RELATIME, ",relatime" }, | |
408 | { 0, NULL } | |
409 | }; | |
410 | struct proc_fs_info *fs_infop; | |
411 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
412 | ||
413 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
414 | seq_putc(m, ' '); | |
415 | seq_path(m, &mnt_path, " \t\n\\"); | |
416 | seq_putc(m, ' '); | |
417 | mangle(m, mnt->mnt_sb->s_type->name); | |
418 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { | |
419 | seq_putc(m, '.'); | |
420 | mangle(m, mnt->mnt_sb->s_subtype); | |
421 | } | |
422 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
423 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
424 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
425 | seq_puts(m, fs_infop->str); | |
426 | } | |
427 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
428 | if (mnt->mnt_flags & fs_infop->flag) | |
429 | seq_puts(m, fs_infop->str); | |
430 | } | |
431 | if (mnt->mnt_sb->s_op->show_options) | |
432 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
433 | seq_puts(m, " 0 0\n"); | |
434 | return err; | |
435 | } | |
436 | ||
437 | struct seq_operations mounts_op = { | |
438 | .start = m_start, | |
439 | .next = m_next, | |
440 | .stop = m_stop, | |
441 | .show = show_vfsmnt | |
442 | }; | |
443 | ||
444 | static int show_vfsstat(struct seq_file *m, void *v) | |
445 | { | |
446 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
447 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
448 | int err = 0; | |
449 | ||
450 | /* device */ | |
451 | if (mnt->mnt_devname) { | |
452 | seq_puts(m, "device "); | |
453 | mangle(m, mnt->mnt_devname); | |
454 | } else | |
455 | seq_puts(m, "no device"); | |
456 | ||
457 | /* mount point */ | |
458 | seq_puts(m, " mounted on "); | |
459 | seq_path(m, &mnt_path, " \t\n\\"); | |
460 | seq_putc(m, ' '); | |
461 | ||
462 | /* file system type */ | |
463 | seq_puts(m, "with fstype "); | |
464 | mangle(m, mnt->mnt_sb->s_type->name); | |
465 | ||
466 | /* optional statistics */ | |
467 | if (mnt->mnt_sb->s_op->show_stats) { | |
468 | seq_putc(m, ' '); | |
469 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
470 | } | |
471 | ||
472 | seq_putc(m, '\n'); | |
473 | return err; | |
474 | } | |
475 | ||
476 | struct seq_operations mountstats_op = { | |
477 | .start = m_start, | |
478 | .next = m_next, | |
479 | .stop = m_stop, | |
480 | .show = show_vfsstat, | |
481 | }; | |
482 | ||
483 | /** | |
484 | * may_umount_tree - check if a mount tree is busy | |
485 | * @mnt: root of mount tree | |
486 | * | |
487 | * This is called to check if a tree of mounts has any | |
488 | * open files, pwds, chroots or sub mounts that are | |
489 | * busy. | |
490 | */ | |
491 | int may_umount_tree(struct vfsmount *mnt) | |
492 | { | |
493 | int actual_refs = 0; | |
494 | int minimum_refs = 0; | |
495 | struct vfsmount *p; | |
496 | ||
497 | spin_lock(&vfsmount_lock); | |
498 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
499 | actual_refs += atomic_read(&p->mnt_count); | |
500 | minimum_refs += 2; | |
501 | } | |
502 | spin_unlock(&vfsmount_lock); | |
503 | ||
504 | if (actual_refs > minimum_refs) | |
505 | return 0; | |
506 | ||
507 | return 1; | |
508 | } | |
509 | ||
510 | EXPORT_SYMBOL(may_umount_tree); | |
511 | ||
512 | /** | |
513 | * may_umount - check if a mount point is busy | |
514 | * @mnt: root of mount | |
515 | * | |
516 | * This is called to check if a mount point has any | |
517 | * open files, pwds, chroots or sub mounts. If the | |
518 | * mount has sub mounts this will return busy | |
519 | * regardless of whether the sub mounts are busy. | |
520 | * | |
521 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
522 | * give false negatives. The main reason why it's here is that we need | |
523 | * a non-destructive way to look for easily umountable filesystems. | |
524 | */ | |
525 | int may_umount(struct vfsmount *mnt) | |
526 | { | |
527 | int ret = 1; | |
528 | spin_lock(&vfsmount_lock); | |
529 | if (propagate_mount_busy(mnt, 2)) | |
530 | ret = 0; | |
531 | spin_unlock(&vfsmount_lock); | |
532 | return ret; | |
533 | } | |
534 | ||
535 | EXPORT_SYMBOL(may_umount); | |
536 | ||
537 | void release_mounts(struct list_head *head) | |
538 | { | |
539 | struct vfsmount *mnt; | |
540 | while (!list_empty(head)) { | |
541 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); | |
542 | list_del_init(&mnt->mnt_hash); | |
543 | if (mnt->mnt_parent != mnt) { | |
544 | struct dentry *dentry; | |
545 | struct vfsmount *m; | |
546 | spin_lock(&vfsmount_lock); | |
547 | dentry = mnt->mnt_mountpoint; | |
548 | m = mnt->mnt_parent; | |
549 | mnt->mnt_mountpoint = mnt->mnt_root; | |
550 | mnt->mnt_parent = mnt; | |
551 | m->mnt_ghosts--; | |
552 | spin_unlock(&vfsmount_lock); | |
553 | dput(dentry); | |
554 | mntput(m); | |
555 | } | |
556 | mntput(mnt); | |
557 | } | |
558 | } | |
559 | ||
560 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) | |
561 | { | |
562 | struct vfsmount *p; | |
563 | ||
564 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
565 | list_move(&p->mnt_hash, kill); | |
566 | ||
567 | if (propagate) | |
568 | propagate_umount(kill); | |
569 | ||
570 | list_for_each_entry(p, kill, mnt_hash) { | |
571 | list_del_init(&p->mnt_expire); | |
572 | list_del_init(&p->mnt_list); | |
573 | __touch_mnt_namespace(p->mnt_ns); | |
574 | p->mnt_ns = NULL; | |
575 | list_del_init(&p->mnt_child); | |
576 | if (p->mnt_parent != p) { | |
577 | p->mnt_parent->mnt_ghosts++; | |
578 | p->mnt_mountpoint->d_mounted--; | |
579 | } | |
580 | change_mnt_propagation(p, MS_PRIVATE); | |
581 | } | |
582 | } | |
583 | ||
584 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); | |
585 | ||
586 | static int do_umount(struct vfsmount *mnt, int flags) | |
587 | { | |
588 | struct super_block *sb = mnt->mnt_sb; | |
589 | int retval; | |
590 | LIST_HEAD(umount_list); | |
591 | ||
592 | retval = security_sb_umount(mnt, flags); | |
593 | if (retval) | |
594 | return retval; | |
595 | ||
596 | /* | |
597 | * Allow userspace to request a mountpoint be expired rather than | |
598 | * unmounting unconditionally. Unmount only happens if: | |
599 | * (1) the mark is already set (the mark is cleared by mntput()) | |
600 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
601 | */ | |
602 | if (flags & MNT_EXPIRE) { | |
603 | if (mnt == current->fs->root.mnt || | |
604 | flags & (MNT_FORCE | MNT_DETACH)) | |
605 | return -EINVAL; | |
606 | ||
607 | if (atomic_read(&mnt->mnt_count) != 2) | |
608 | return -EBUSY; | |
609 | ||
610 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
611 | return -EAGAIN; | |
612 | } | |
613 | ||
614 | /* | |
615 | * If we may have to abort operations to get out of this | |
616 | * mount, and they will themselves hold resources we must | |
617 | * allow the fs to do things. In the Unix tradition of | |
618 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
619 | * might fail to complete on the first run through as other tasks | |
620 | * must return, and the like. Thats for the mount program to worry | |
621 | * about for the moment. | |
622 | */ | |
623 | ||
624 | lock_kernel(); | |
625 | if (sb->s_op->umount_begin) | |
626 | sb->s_op->umount_begin(mnt, flags); | |
627 | unlock_kernel(); | |
628 | ||
629 | /* | |
630 | * No sense to grab the lock for this test, but test itself looks | |
631 | * somewhat bogus. Suggestions for better replacement? | |
632 | * Ho-hum... In principle, we might treat that as umount + switch | |
633 | * to rootfs. GC would eventually take care of the old vfsmount. | |
634 | * Actually it makes sense, especially if rootfs would contain a | |
635 | * /reboot - static binary that would close all descriptors and | |
636 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
637 | */ | |
638 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { | |
639 | /* | |
640 | * Special case for "unmounting" root ... | |
641 | * we just try to remount it readonly. | |
642 | */ | |
643 | down_write(&sb->s_umount); | |
644 | if (!(sb->s_flags & MS_RDONLY)) { | |
645 | lock_kernel(); | |
646 | DQUOT_OFF(sb); | |
647 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
648 | unlock_kernel(); | |
649 | } | |
650 | up_write(&sb->s_umount); | |
651 | return retval; | |
652 | } | |
653 | ||
654 | down_write(&namespace_sem); | |
655 | spin_lock(&vfsmount_lock); | |
656 | event++; | |
657 | ||
658 | if (!(flags & MNT_DETACH)) | |
659 | shrink_submounts(mnt, &umount_list); | |
660 | ||
661 | retval = -EBUSY; | |
662 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { | |
663 | if (!list_empty(&mnt->mnt_list)) | |
664 | umount_tree(mnt, 1, &umount_list); | |
665 | retval = 0; | |
666 | } | |
667 | spin_unlock(&vfsmount_lock); | |
668 | if (retval) | |
669 | security_sb_umount_busy(mnt); | |
670 | up_write(&namespace_sem); | |
671 | release_mounts(&umount_list); | |
672 | return retval; | |
673 | } | |
674 | ||
675 | /* | |
676 | * Now umount can handle mount points as well as block devices. | |
677 | * This is important for filesystems which use unnamed block devices. | |
678 | * | |
679 | * We now support a flag for forced unmount like the other 'big iron' | |
680 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
681 | */ | |
682 | ||
683 | asmlinkage long sys_umount(char __user * name, int flags) | |
684 | { | |
685 | struct nameidata nd; | |
686 | int retval; | |
687 | ||
688 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
689 | if (retval) | |
690 | goto out; | |
691 | retval = -EINVAL; | |
692 | if (nd.path.dentry != nd.path.mnt->mnt_root) | |
693 | goto dput_and_out; | |
694 | if (!check_mnt(nd.path.mnt)) | |
695 | goto dput_and_out; | |
696 | ||
697 | retval = -EPERM; | |
698 | if (!capable(CAP_SYS_ADMIN)) | |
699 | goto dput_and_out; | |
700 | ||
701 | retval = do_umount(nd.path.mnt, flags); | |
702 | dput_and_out: | |
703 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ | |
704 | dput(nd.path.dentry); | |
705 | mntput_no_expire(nd.path.mnt); | |
706 | out: | |
707 | return retval; | |
708 | } | |
709 | ||
710 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
711 | ||
712 | /* | |
713 | * The 2.0 compatible umount. No flags. | |
714 | */ | |
715 | asmlinkage long sys_oldumount(char __user * name) | |
716 | { | |
717 | return sys_umount(name, 0); | |
718 | } | |
719 | ||
720 | #endif | |
721 | ||
722 | static int mount_is_safe(struct nameidata *nd) | |
723 | { | |
724 | if (capable(CAP_SYS_ADMIN)) | |
725 | return 0; | |
726 | return -EPERM; | |
727 | #ifdef notyet | |
728 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) | |
729 | return -EPERM; | |
730 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { | |
731 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
732 | return -EPERM; | |
733 | } | |
734 | if (vfs_permission(nd, MAY_WRITE)) | |
735 | return -EPERM; | |
736 | return 0; | |
737 | #endif | |
738 | } | |
739 | ||
740 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) | |
741 | { | |
742 | while (1) { | |
743 | if (d == dentry) | |
744 | return 1; | |
745 | if (d == NULL || d == d->d_parent) | |
746 | return 0; | |
747 | d = d->d_parent; | |
748 | } | |
749 | } | |
750 | ||
751 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, | |
752 | int flag) | |
753 | { | |
754 | struct vfsmount *res, *p, *q, *r, *s; | |
755 | struct path path; | |
756 | ||
757 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) | |
758 | return NULL; | |
759 | ||
760 | res = q = clone_mnt(mnt, dentry, flag); | |
761 | if (!q) | |
762 | goto Enomem; | |
763 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
764 | ||
765 | p = mnt; | |
766 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { | |
767 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) | |
768 | continue; | |
769 | ||
770 | for (s = r; s; s = next_mnt(s, r)) { | |
771 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { | |
772 | s = skip_mnt_tree(s); | |
773 | continue; | |
774 | } | |
775 | while (p != s->mnt_parent) { | |
776 | p = p->mnt_parent; | |
777 | q = q->mnt_parent; | |
778 | } | |
779 | p = s; | |
780 | path.mnt = q; | |
781 | path.dentry = p->mnt_mountpoint; | |
782 | q = clone_mnt(p, p->mnt_root, flag); | |
783 | if (!q) | |
784 | goto Enomem; | |
785 | spin_lock(&vfsmount_lock); | |
786 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
787 | attach_mnt(q, &path); | |
788 | spin_unlock(&vfsmount_lock); | |
789 | } | |
790 | } | |
791 | return res; | |
792 | Enomem: | |
793 | if (res) { | |
794 | LIST_HEAD(umount_list); | |
795 | spin_lock(&vfsmount_lock); | |
796 | umount_tree(res, 0, &umount_list); | |
797 | spin_unlock(&vfsmount_lock); | |
798 | release_mounts(&umount_list); | |
799 | } | |
800 | return NULL; | |
801 | } | |
802 | ||
803 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) | |
804 | { | |
805 | struct vfsmount *tree; | |
806 | down_read(&namespace_sem); | |
807 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); | |
808 | up_read(&namespace_sem); | |
809 | return tree; | |
810 | } | |
811 | ||
812 | void drop_collected_mounts(struct vfsmount *mnt) | |
813 | { | |
814 | LIST_HEAD(umount_list); | |
815 | down_read(&namespace_sem); | |
816 | spin_lock(&vfsmount_lock); | |
817 | umount_tree(mnt, 0, &umount_list); | |
818 | spin_unlock(&vfsmount_lock); | |
819 | up_read(&namespace_sem); | |
820 | release_mounts(&umount_list); | |
821 | } | |
822 | ||
823 | /* | |
824 | * @source_mnt : mount tree to be attached | |
825 | * @nd : place the mount tree @source_mnt is attached | |
826 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
827 | * store the parent mount and mountpoint dentry. | |
828 | * (done when source_mnt is moved) | |
829 | * | |
830 | * NOTE: in the table below explains the semantics when a source mount | |
831 | * of a given type is attached to a destination mount of a given type. | |
832 | * --------------------------------------------------------------------------- | |
833 | * | BIND MOUNT OPERATION | | |
834 | * |************************************************************************** | |
835 | * | source-->| shared | private | slave | unbindable | | |
836 | * | dest | | | | | | |
837 | * | | | | | | | | |
838 | * | v | | | | | | |
839 | * |************************************************************************** | |
840 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
841 | * | | | | | | | |
842 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
843 | * *************************************************************************** | |
844 | * A bind operation clones the source mount and mounts the clone on the | |
845 | * destination mount. | |
846 | * | |
847 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
848 | * tree of the destination mount and the cloned mount is added to | |
849 | * the peer group of the source mount. | |
850 | * (+) the cloned mount is created under the destination mount and is marked | |
851 | * as shared. The cloned mount is added to the peer group of the source | |
852 | * mount. | |
853 | * (+++) the mount is propagated to all the mounts in the propagation tree | |
854 | * of the destination mount and the cloned mount is made slave | |
855 | * of the same master as that of the source mount. The cloned mount | |
856 | * is marked as 'shared and slave'. | |
857 | * (*) the cloned mount is made a slave of the same master as that of the | |
858 | * source mount. | |
859 | * | |
860 | * --------------------------------------------------------------------------- | |
861 | * | MOVE MOUNT OPERATION | | |
862 | * |************************************************************************** | |
863 | * | source-->| shared | private | slave | unbindable | | |
864 | * | dest | | | | | | |
865 | * | | | | | | | | |
866 | * | v | | | | | | |
867 | * |************************************************************************** | |
868 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
869 | * | | | | | | | |
870 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
871 | * *************************************************************************** | |
872 | * | |
873 | * (+) the mount is moved to the destination. And is then propagated to | |
874 | * all the mounts in the propagation tree of the destination mount. | |
875 | * (+*) the mount is moved to the destination. | |
876 | * (+++) the mount is moved to the destination and is then propagated to | |
877 | * all the mounts belonging to the destination mount's propagation tree. | |
878 | * the mount is marked as 'shared and slave'. | |
879 | * (*) the mount continues to be a slave at the new location. | |
880 | * | |
881 | * if the source mount is a tree, the operations explained above is | |
882 | * applied to each mount in the tree. | |
883 | * Must be called without spinlocks held, since this function can sleep | |
884 | * in allocations. | |
885 | */ | |
886 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
887 | struct path *path, struct path *parent_path) | |
888 | { | |
889 | LIST_HEAD(tree_list); | |
890 | struct vfsmount *dest_mnt = path->mnt; | |
891 | struct dentry *dest_dentry = path->dentry; | |
892 | struct vfsmount *child, *p; | |
893 | ||
894 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
895 | return -EINVAL; | |
896 | ||
897 | if (IS_MNT_SHARED(dest_mnt)) { | |
898 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
899 | set_mnt_shared(p); | |
900 | } | |
901 | ||
902 | spin_lock(&vfsmount_lock); | |
903 | if (parent_path) { | |
904 | detach_mnt(source_mnt, parent_path); | |
905 | attach_mnt(source_mnt, path); | |
906 | touch_mnt_namespace(current->nsproxy->mnt_ns); | |
907 | } else { | |
908 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
909 | commit_tree(source_mnt); | |
910 | } | |
911 | ||
912 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
913 | list_del_init(&child->mnt_hash); | |
914 | commit_tree(child); | |
915 | } | |
916 | spin_unlock(&vfsmount_lock); | |
917 | return 0; | |
918 | } | |
919 | ||
920 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) | |
921 | { | |
922 | int err; | |
923 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
924 | return -EINVAL; | |
925 | ||
926 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != | |
927 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) | |
928 | return -ENOTDIR; | |
929 | ||
930 | err = -ENOENT; | |
931 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); | |
932 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
933 | goto out_unlock; | |
934 | ||
935 | err = security_sb_check_sb(mnt, nd); | |
936 | if (err) | |
937 | goto out_unlock; | |
938 | ||
939 | err = -ENOENT; | |
940 | if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) | |
941 | err = attach_recursive_mnt(mnt, &nd->path, NULL); | |
942 | out_unlock: | |
943 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); | |
944 | if (!err) | |
945 | security_sb_post_addmount(mnt, nd); | |
946 | return err; | |
947 | } | |
948 | ||
949 | /* | |
950 | * recursively change the type of the mountpoint. | |
951 | * noinline this do_mount helper to save do_mount stack space. | |
952 | */ | |
953 | static noinline int do_change_type(struct nameidata *nd, int flag) | |
954 | { | |
955 | struct vfsmount *m, *mnt = nd->path.mnt; | |
956 | int recurse = flag & MS_REC; | |
957 | int type = flag & ~MS_REC; | |
958 | ||
959 | if (!capable(CAP_SYS_ADMIN)) | |
960 | return -EPERM; | |
961 | ||
962 | if (nd->path.dentry != nd->path.mnt->mnt_root) | |
963 | return -EINVAL; | |
964 | ||
965 | down_write(&namespace_sem); | |
966 | spin_lock(&vfsmount_lock); | |
967 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
968 | change_mnt_propagation(m, type); | |
969 | spin_unlock(&vfsmount_lock); | |
970 | up_write(&namespace_sem); | |
971 | return 0; | |
972 | } | |
973 | ||
974 | /* | |
975 | * do loopback mount. | |
976 | * noinline this do_mount helper to save do_mount stack space. | |
977 | */ | |
978 | static noinline int do_loopback(struct nameidata *nd, char *old_name, | |
979 | int recurse) | |
980 | { | |
981 | struct nameidata old_nd; | |
982 | struct vfsmount *mnt = NULL; | |
983 | int err = mount_is_safe(nd); | |
984 | if (err) | |
985 | return err; | |
986 | if (!old_name || !*old_name) | |
987 | return -EINVAL; | |
988 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
989 | if (err) | |
990 | return err; | |
991 | ||
992 | down_write(&namespace_sem); | |
993 | err = -EINVAL; | |
994 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) | |
995 | goto out; | |
996 | ||
997 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) | |
998 | goto out; | |
999 | ||
1000 | err = -ENOMEM; | |
1001 | if (recurse) | |
1002 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); | |
1003 | else | |
1004 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); | |
1005 | ||
1006 | if (!mnt) | |
1007 | goto out; | |
1008 | ||
1009 | err = graft_tree(mnt, nd); | |
1010 | if (err) { | |
1011 | LIST_HEAD(umount_list); | |
1012 | spin_lock(&vfsmount_lock); | |
1013 | umount_tree(mnt, 0, &umount_list); | |
1014 | spin_unlock(&vfsmount_lock); | |
1015 | release_mounts(&umount_list); | |
1016 | } | |
1017 | ||
1018 | out: | |
1019 | up_write(&namespace_sem); | |
1020 | path_put(&old_nd.path); | |
1021 | return err; | |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * change filesystem flags. dir should be a physical root of filesystem. | |
1026 | * If you've mounted a non-root directory somewhere and want to do remount | |
1027 | * on it - tough luck. | |
1028 | * noinline this do_mount helper to save do_mount stack space. | |
1029 | */ | |
1030 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, | |
1031 | void *data) | |
1032 | { | |
1033 | int err; | |
1034 | struct super_block *sb = nd->path.mnt->mnt_sb; | |
1035 | ||
1036 | if (!capable(CAP_SYS_ADMIN)) | |
1037 | return -EPERM; | |
1038 | ||
1039 | if (!check_mnt(nd->path.mnt)) | |
1040 | return -EINVAL; | |
1041 | ||
1042 | if (nd->path.dentry != nd->path.mnt->mnt_root) | |
1043 | return -EINVAL; | |
1044 | ||
1045 | down_write(&sb->s_umount); | |
1046 | err = do_remount_sb(sb, flags, data, 0); | |
1047 | if (!err) | |
1048 | nd->path.mnt->mnt_flags = mnt_flags; | |
1049 | up_write(&sb->s_umount); | |
1050 | if (!err) | |
1051 | security_sb_post_remount(nd->path.mnt, flags, data); | |
1052 | return err; | |
1053 | } | |
1054 | ||
1055 | static inline int tree_contains_unbindable(struct vfsmount *mnt) | |
1056 | { | |
1057 | struct vfsmount *p; | |
1058 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1059 | if (IS_MNT_UNBINDABLE(p)) | |
1060 | return 1; | |
1061 | } | |
1062 | return 0; | |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * noinline this do_mount helper to save do_mount stack space. | |
1067 | */ | |
1068 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1069 | { | |
1070 | struct nameidata old_nd; | |
1071 | struct path parent_path; | |
1072 | struct vfsmount *p; | |
1073 | int err = 0; | |
1074 | if (!capable(CAP_SYS_ADMIN)) | |
1075 | return -EPERM; | |
1076 | if (!old_name || !*old_name) | |
1077 | return -EINVAL; | |
1078 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1079 | if (err) | |
1080 | return err; | |
1081 | ||
1082 | down_write(&namespace_sem); | |
1083 | while (d_mountpoint(nd->path.dentry) && | |
1084 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1085 | ; | |
1086 | err = -EINVAL; | |
1087 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) | |
1088 | goto out; | |
1089 | ||
1090 | err = -ENOENT; | |
1091 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); | |
1092 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1093 | goto out1; | |
1094 | ||
1095 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) | |
1096 | goto out1; | |
1097 | ||
1098 | err = -EINVAL; | |
1099 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) | |
1100 | goto out1; | |
1101 | ||
1102 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) | |
1103 | goto out1; | |
1104 | ||
1105 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != | |
1106 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
1107 | goto out1; | |
1108 | /* | |
1109 | * Don't move a mount residing in a shared parent. | |
1110 | */ | |
1111 | if (old_nd.path.mnt->mnt_parent && | |
1112 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
1113 | goto out1; | |
1114 | /* | |
1115 | * Don't move a mount tree containing unbindable mounts to a destination | |
1116 | * mount which is shared. | |
1117 | */ | |
1118 | if (IS_MNT_SHARED(nd->path.mnt) && | |
1119 | tree_contains_unbindable(old_nd.path.mnt)) | |
1120 | goto out1; | |
1121 | err = -ELOOP; | |
1122 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) | |
1123 | if (p == old_nd.path.mnt) | |
1124 | goto out1; | |
1125 | ||
1126 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); | |
1127 | if (err) | |
1128 | goto out1; | |
1129 | ||
1130 | spin_lock(&vfsmount_lock); | |
1131 | /* if the mount is moved, it should no longer be expire | |
1132 | * automatically */ | |
1133 | list_del_init(&old_nd.path.mnt->mnt_expire); | |
1134 | spin_unlock(&vfsmount_lock); | |
1135 | out1: | |
1136 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); | |
1137 | out: | |
1138 | up_write(&namespace_sem); | |
1139 | if (!err) | |
1140 | path_put(&parent_path); | |
1141 | path_put(&old_nd.path); | |
1142 | return err; | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * create a new mount for userspace and request it to be added into the | |
1147 | * namespace's tree | |
1148 | * noinline this do_mount helper to save do_mount stack space. | |
1149 | */ | |
1150 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, | |
1151 | int mnt_flags, char *name, void *data) | |
1152 | { | |
1153 | struct vfsmount *mnt; | |
1154 | ||
1155 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1156 | return -EINVAL; | |
1157 | ||
1158 | /* we need capabilities... */ | |
1159 | if (!capable(CAP_SYS_ADMIN)) | |
1160 | return -EPERM; | |
1161 | ||
1162 | mnt = do_kern_mount(type, flags, name, data); | |
1163 | if (IS_ERR(mnt)) | |
1164 | return PTR_ERR(mnt); | |
1165 | ||
1166 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * add a mount into a namespace's mount tree | |
1171 | * - provide the option of adding the new mount to an expiration list | |
1172 | */ | |
1173 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1174 | int mnt_flags, struct list_head *fslist) | |
1175 | { | |
1176 | int err; | |
1177 | ||
1178 | down_write(&namespace_sem); | |
1179 | /* Something was mounted here while we slept */ | |
1180 | while (d_mountpoint(nd->path.dentry) && | |
1181 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1182 | ; | |
1183 | err = -EINVAL; | |
1184 | if (!check_mnt(nd->path.mnt)) | |
1185 | goto unlock; | |
1186 | ||
1187 | /* Refuse the same filesystem on the same mount point */ | |
1188 | err = -EBUSY; | |
1189 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && | |
1190 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1191 | goto unlock; | |
1192 | ||
1193 | err = -EINVAL; | |
1194 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1195 | goto unlock; | |
1196 | ||
1197 | newmnt->mnt_flags = mnt_flags; | |
1198 | if ((err = graft_tree(newmnt, nd))) | |
1199 | goto unlock; | |
1200 | ||
1201 | if (fslist) { | |
1202 | /* add to the specified expiration list */ | |
1203 | spin_lock(&vfsmount_lock); | |
1204 | list_add_tail(&newmnt->mnt_expire, fslist); | |
1205 | spin_unlock(&vfsmount_lock); | |
1206 | } | |
1207 | up_write(&namespace_sem); | |
1208 | return 0; | |
1209 | ||
1210 | unlock: | |
1211 | up_write(&namespace_sem); | |
1212 | mntput(newmnt); | |
1213 | return err; | |
1214 | } | |
1215 | ||
1216 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1217 | ||
1218 | /* | |
1219 | * process a list of expirable mountpoints with the intent of discarding any | |
1220 | * mountpoints that aren't in use and haven't been touched since last we came | |
1221 | * here | |
1222 | */ | |
1223 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1224 | { | |
1225 | struct vfsmount *mnt, *next; | |
1226 | LIST_HEAD(graveyard); | |
1227 | LIST_HEAD(umounts); | |
1228 | ||
1229 | if (list_empty(mounts)) | |
1230 | return; | |
1231 | ||
1232 | down_write(&namespace_sem); | |
1233 | spin_lock(&vfsmount_lock); | |
1234 | ||
1235 | /* extract from the expiration list every vfsmount that matches the | |
1236 | * following criteria: | |
1237 | * - only referenced by its parent vfsmount | |
1238 | * - still marked for expiry (marked on the last call here; marks are | |
1239 | * cleared by mntput()) | |
1240 | */ | |
1241 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { | |
1242 | if (!xchg(&mnt->mnt_expiry_mark, 1) || | |
1243 | propagate_mount_busy(mnt, 1)) | |
1244 | continue; | |
1245 | list_move(&mnt->mnt_expire, &graveyard); | |
1246 | } | |
1247 | while (!list_empty(&graveyard)) { | |
1248 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1249 | touch_mnt_namespace(mnt->mnt_ns); | |
1250 | umount_tree(mnt, 1, &umounts); | |
1251 | } | |
1252 | spin_unlock(&vfsmount_lock); | |
1253 | up_write(&namespace_sem); | |
1254 | ||
1255 | release_mounts(&umounts); | |
1256 | } | |
1257 | ||
1258 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1259 | ||
1260 | /* | |
1261 | * Ripoff of 'select_parent()' | |
1262 | * | |
1263 | * search the list of submounts for a given mountpoint, and move any | |
1264 | * shrinkable submounts to the 'graveyard' list. | |
1265 | */ | |
1266 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1267 | { | |
1268 | struct vfsmount *this_parent = parent; | |
1269 | struct list_head *next; | |
1270 | int found = 0; | |
1271 | ||
1272 | repeat: | |
1273 | next = this_parent->mnt_mounts.next; | |
1274 | resume: | |
1275 | while (next != &this_parent->mnt_mounts) { | |
1276 | struct list_head *tmp = next; | |
1277 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1278 | ||
1279 | next = tmp->next; | |
1280 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1281 | continue; | |
1282 | /* | |
1283 | * Descend a level if the d_mounts list is non-empty. | |
1284 | */ | |
1285 | if (!list_empty(&mnt->mnt_mounts)) { | |
1286 | this_parent = mnt; | |
1287 | goto repeat; | |
1288 | } | |
1289 | ||
1290 | if (!propagate_mount_busy(mnt, 1)) { | |
1291 | list_move_tail(&mnt->mnt_expire, graveyard); | |
1292 | found++; | |
1293 | } | |
1294 | } | |
1295 | /* | |
1296 | * All done at this level ... ascend and resume the search | |
1297 | */ | |
1298 | if (this_parent != parent) { | |
1299 | next = this_parent->mnt_child.next; | |
1300 | this_parent = this_parent->mnt_parent; | |
1301 | goto resume; | |
1302 | } | |
1303 | return found; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * process a list of expirable mountpoints with the intent of discarding any | |
1308 | * submounts of a specific parent mountpoint | |
1309 | */ | |
1310 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) | |
1311 | { | |
1312 | LIST_HEAD(graveyard); | |
1313 | struct vfsmount *m; | |
1314 | ||
1315 | /* extract submounts of 'mountpoint' from the expiration list */ | |
1316 | while (select_submounts(mnt, &graveyard)) { | |
1317 | while (!list_empty(&graveyard)) { | |
1318 | m = list_first_entry(&graveyard, struct vfsmount, | |
1319 | mnt_expire); | |
1320 | touch_mnt_namespace(mnt->mnt_ns); | |
1321 | umount_tree(mnt, 1, umounts); | |
1322 | } | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Some copy_from_user() implementations do not return the exact number of | |
1328 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1329 | * Note that this function differs from copy_from_user() in that it will oops | |
1330 | * on bad values of `to', rather than returning a short copy. | |
1331 | */ | |
1332 | static long exact_copy_from_user(void *to, const void __user * from, | |
1333 | unsigned long n) | |
1334 | { | |
1335 | char *t = to; | |
1336 | const char __user *f = from; | |
1337 | char c; | |
1338 | ||
1339 | if (!access_ok(VERIFY_READ, from, n)) | |
1340 | return n; | |
1341 | ||
1342 | while (n) { | |
1343 | if (__get_user(c, f)) { | |
1344 | memset(t, 0, n); | |
1345 | break; | |
1346 | } | |
1347 | *t++ = c; | |
1348 | f++; | |
1349 | n--; | |
1350 | } | |
1351 | return n; | |
1352 | } | |
1353 | ||
1354 | int copy_mount_options(const void __user * data, unsigned long *where) | |
1355 | { | |
1356 | int i; | |
1357 | unsigned long page; | |
1358 | unsigned long size; | |
1359 | ||
1360 | *where = 0; | |
1361 | if (!data) | |
1362 | return 0; | |
1363 | ||
1364 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1365 | return -ENOMEM; | |
1366 | ||
1367 | /* We only care that *some* data at the address the user | |
1368 | * gave us is valid. Just in case, we'll zero | |
1369 | * the remainder of the page. | |
1370 | */ | |
1371 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1372 | size = TASK_SIZE - (unsigned long)data; | |
1373 | if (size > PAGE_SIZE) | |
1374 | size = PAGE_SIZE; | |
1375 | ||
1376 | i = size - exact_copy_from_user((void *)page, data, size); | |
1377 | if (!i) { | |
1378 | free_page(page); | |
1379 | return -EFAULT; | |
1380 | } | |
1381 | if (i != PAGE_SIZE) | |
1382 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1383 | *where = page; | |
1384 | return 0; | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1389 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1390 | * | |
1391 | * data is a (void *) that can point to any structure up to | |
1392 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1393 | * information (or be NULL). | |
1394 | * | |
1395 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1396 | * When the flags word was introduced its top half was required | |
1397 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1398 | * Therefore, if this magic number is present, it carries no information | |
1399 | * and must be discarded. | |
1400 | */ | |
1401 | long do_mount(char *dev_name, char *dir_name, char *type_page, | |
1402 | unsigned long flags, void *data_page) | |
1403 | { | |
1404 | struct nameidata nd; | |
1405 | int retval = 0; | |
1406 | int mnt_flags = 0; | |
1407 | ||
1408 | /* Discard magic */ | |
1409 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1410 | flags &= ~MS_MGC_MSK; | |
1411 | ||
1412 | /* Basic sanity checks */ | |
1413 | ||
1414 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1415 | return -EINVAL; | |
1416 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1417 | return -EINVAL; | |
1418 | ||
1419 | if (data_page) | |
1420 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1421 | ||
1422 | /* Separate the per-mountpoint flags */ | |
1423 | if (flags & MS_NOSUID) | |
1424 | mnt_flags |= MNT_NOSUID; | |
1425 | if (flags & MS_NODEV) | |
1426 | mnt_flags |= MNT_NODEV; | |
1427 | if (flags & MS_NOEXEC) | |
1428 | mnt_flags |= MNT_NOEXEC; | |
1429 | if (flags & MS_NOATIME) | |
1430 | mnt_flags |= MNT_NOATIME; | |
1431 | if (flags & MS_NODIRATIME) | |
1432 | mnt_flags |= MNT_NODIRATIME; | |
1433 | if (flags & MS_RELATIME) | |
1434 | mnt_flags |= MNT_RELATIME; | |
1435 | ||
1436 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
1437 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); | |
1438 | ||
1439 | /* ... and get the mountpoint */ | |
1440 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1441 | if (retval) | |
1442 | return retval; | |
1443 | ||
1444 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1445 | if (retval) | |
1446 | goto dput_out; | |
1447 | ||
1448 | if (flags & MS_REMOUNT) | |
1449 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1450 | data_page); | |
1451 | else if (flags & MS_BIND) | |
1452 | retval = do_loopback(&nd, dev_name, flags & MS_REC); | |
1453 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1454 | retval = do_change_type(&nd, flags); | |
1455 | else if (flags & MS_MOVE) | |
1456 | retval = do_move_mount(&nd, dev_name); | |
1457 | else | |
1458 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1459 | dev_name, data_page); | |
1460 | dput_out: | |
1461 | path_put(&nd.path); | |
1462 | return retval; | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Allocate a new namespace structure and populate it with contents | |
1467 | * copied from the namespace of the passed in task structure. | |
1468 | */ | |
1469 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, | |
1470 | struct fs_struct *fs) | |
1471 | { | |
1472 | struct mnt_namespace *new_ns; | |
1473 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; | |
1474 | struct vfsmount *p, *q; | |
1475 | ||
1476 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
1477 | if (!new_ns) | |
1478 | return ERR_PTR(-ENOMEM); | |
1479 | ||
1480 | atomic_set(&new_ns->count, 1); | |
1481 | INIT_LIST_HEAD(&new_ns->list); | |
1482 | init_waitqueue_head(&new_ns->poll); | |
1483 | new_ns->event = 0; | |
1484 | ||
1485 | down_write(&namespace_sem); | |
1486 | /* First pass: copy the tree topology */ | |
1487 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, | |
1488 | CL_COPY_ALL | CL_EXPIRE); | |
1489 | if (!new_ns->root) { | |
1490 | up_write(&namespace_sem); | |
1491 | kfree(new_ns); | |
1492 | return ERR_PTR(-ENOMEM);; | |
1493 | } | |
1494 | spin_lock(&vfsmount_lock); | |
1495 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1496 | spin_unlock(&vfsmount_lock); | |
1497 | ||
1498 | /* | |
1499 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1500 | * as belonging to new namespace. We have already acquired a private | |
1501 | * fs_struct, so tsk->fs->lock is not needed. | |
1502 | */ | |
1503 | p = mnt_ns->root; | |
1504 | q = new_ns->root; | |
1505 | while (p) { | |
1506 | q->mnt_ns = new_ns; | |
1507 | if (fs) { | |
1508 | if (p == fs->root.mnt) { | |
1509 | rootmnt = p; | |
1510 | fs->root.mnt = mntget(q); | |
1511 | } | |
1512 | if (p == fs->pwd.mnt) { | |
1513 | pwdmnt = p; | |
1514 | fs->pwd.mnt = mntget(q); | |
1515 | } | |
1516 | if (p == fs->altroot.mnt) { | |
1517 | altrootmnt = p; | |
1518 | fs->altroot.mnt = mntget(q); | |
1519 | } | |
1520 | } | |
1521 | p = next_mnt(p, mnt_ns->root); | |
1522 | q = next_mnt(q, new_ns->root); | |
1523 | } | |
1524 | up_write(&namespace_sem); | |
1525 | ||
1526 | if (rootmnt) | |
1527 | mntput(rootmnt); | |
1528 | if (pwdmnt) | |
1529 | mntput(pwdmnt); | |
1530 | if (altrootmnt) | |
1531 | mntput(altrootmnt); | |
1532 | ||
1533 | return new_ns; | |
1534 | } | |
1535 | ||
1536 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, | |
1537 | struct fs_struct *new_fs) | |
1538 | { | |
1539 | struct mnt_namespace *new_ns; | |
1540 | ||
1541 | BUG_ON(!ns); | |
1542 | get_mnt_ns(ns); | |
1543 | ||
1544 | if (!(flags & CLONE_NEWNS)) | |
1545 | return ns; | |
1546 | ||
1547 | new_ns = dup_mnt_ns(ns, new_fs); | |
1548 | ||
1549 | put_mnt_ns(ns); | |
1550 | return new_ns; | |
1551 | } | |
1552 | ||
1553 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1554 | char __user * type, unsigned long flags, | |
1555 | void __user * data) | |
1556 | { | |
1557 | int retval; | |
1558 | unsigned long data_page; | |
1559 | unsigned long type_page; | |
1560 | unsigned long dev_page; | |
1561 | char *dir_page; | |
1562 | ||
1563 | retval = copy_mount_options(type, &type_page); | |
1564 | if (retval < 0) | |
1565 | return retval; | |
1566 | ||
1567 | dir_page = getname(dir_name); | |
1568 | retval = PTR_ERR(dir_page); | |
1569 | if (IS_ERR(dir_page)) | |
1570 | goto out1; | |
1571 | ||
1572 | retval = copy_mount_options(dev_name, &dev_page); | |
1573 | if (retval < 0) | |
1574 | goto out2; | |
1575 | ||
1576 | retval = copy_mount_options(data, &data_page); | |
1577 | if (retval < 0) | |
1578 | goto out3; | |
1579 | ||
1580 | lock_kernel(); | |
1581 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, | |
1582 | flags, (void *)data_page); | |
1583 | unlock_kernel(); | |
1584 | free_page(data_page); | |
1585 | ||
1586 | out3: | |
1587 | free_page(dev_page); | |
1588 | out2: | |
1589 | putname(dir_page); | |
1590 | out1: | |
1591 | free_page(type_page); | |
1592 | return retval; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1597 | * It can block. Requires the big lock held. | |
1598 | */ | |
1599 | void set_fs_root(struct fs_struct *fs, struct path *path) | |
1600 | { | |
1601 | struct path old_root; | |
1602 | ||
1603 | write_lock(&fs->lock); | |
1604 | old_root = fs->root; | |
1605 | fs->root = *path; | |
1606 | path_get(path); | |
1607 | write_unlock(&fs->lock); | |
1608 | if (old_root.dentry) | |
1609 | path_put(&old_root); | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1614 | * It can block. Requires the big lock held. | |
1615 | */ | |
1616 | void set_fs_pwd(struct fs_struct *fs, struct path *path) | |
1617 | { | |
1618 | struct path old_pwd; | |
1619 | ||
1620 | write_lock(&fs->lock); | |
1621 | old_pwd = fs->pwd; | |
1622 | fs->pwd = *path; | |
1623 | path_get(path); | |
1624 | write_unlock(&fs->lock); | |
1625 | ||
1626 | if (old_pwd.dentry) | |
1627 | path_put(&old_pwd); | |
1628 | } | |
1629 | ||
1630 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) | |
1631 | { | |
1632 | struct task_struct *g, *p; | |
1633 | struct fs_struct *fs; | |
1634 | ||
1635 | read_lock(&tasklist_lock); | |
1636 | do_each_thread(g, p) { | |
1637 | task_lock(p); | |
1638 | fs = p->fs; | |
1639 | if (fs) { | |
1640 | atomic_inc(&fs->count); | |
1641 | task_unlock(p); | |
1642 | if (fs->root.dentry == old_root->dentry | |
1643 | && fs->root.mnt == old_root->mnt) | |
1644 | set_fs_root(fs, new_root); | |
1645 | if (fs->pwd.dentry == old_root->dentry | |
1646 | && fs->pwd.mnt == old_root->mnt) | |
1647 | set_fs_pwd(fs, new_root); | |
1648 | put_fs_struct(fs); | |
1649 | } else | |
1650 | task_unlock(p); | |
1651 | } while_each_thread(g, p); | |
1652 | read_unlock(&tasklist_lock); | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * pivot_root Semantics: | |
1657 | * Moves the root file system of the current process to the directory put_old, | |
1658 | * makes new_root as the new root file system of the current process, and sets | |
1659 | * root/cwd of all processes which had them on the current root to new_root. | |
1660 | * | |
1661 | * Restrictions: | |
1662 | * The new_root and put_old must be directories, and must not be on the | |
1663 | * same file system as the current process root. The put_old must be | |
1664 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1665 | * pointed to by put_old must yield the same directory as new_root. No other | |
1666 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1667 | * | |
1668 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. | |
1669 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1670 | * in this situation. | |
1671 | * | |
1672 | * Notes: | |
1673 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1674 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1675 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1676 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1677 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1678 | * first. | |
1679 | */ | |
1680 | asmlinkage long sys_pivot_root(const char __user * new_root, | |
1681 | const char __user * put_old) | |
1682 | { | |
1683 | struct vfsmount *tmp; | |
1684 | struct nameidata new_nd, old_nd, user_nd; | |
1685 | struct path parent_path, root_parent; | |
1686 | int error; | |
1687 | ||
1688 | if (!capable(CAP_SYS_ADMIN)) | |
1689 | return -EPERM; | |
1690 | ||
1691 | lock_kernel(); | |
1692 | ||
1693 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, | |
1694 | &new_nd); | |
1695 | if (error) | |
1696 | goto out0; | |
1697 | error = -EINVAL; | |
1698 | if (!check_mnt(new_nd.path.mnt)) | |
1699 | goto out1; | |
1700 | ||
1701 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); | |
1702 | if (error) | |
1703 | goto out1; | |
1704 | ||
1705 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1706 | if (error) { | |
1707 | path_put(&old_nd.path); | |
1708 | goto out1; | |
1709 | } | |
1710 | ||
1711 | read_lock(¤t->fs->lock); | |
1712 | user_nd.path = current->fs->root; | |
1713 | path_get(¤t->fs->root); | |
1714 | read_unlock(¤t->fs->lock); | |
1715 | down_write(&namespace_sem); | |
1716 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); | |
1717 | error = -EINVAL; | |
1718 | if (IS_MNT_SHARED(old_nd.path.mnt) || | |
1719 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
1720 | IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) | |
1721 | goto out2; | |
1722 | if (!check_mnt(user_nd.path.mnt)) | |
1723 | goto out2; | |
1724 | error = -ENOENT; | |
1725 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) | |
1726 | goto out2; | |
1727 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) | |
1728 | goto out2; | |
1729 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) | |
1730 | goto out2; | |
1731 | error = -EBUSY; | |
1732 | if (new_nd.path.mnt == user_nd.path.mnt || | |
1733 | old_nd.path.mnt == user_nd.path.mnt) | |
1734 | goto out2; /* loop, on the same file system */ | |
1735 | error = -EINVAL; | |
1736 | if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) | |
1737 | goto out2; /* not a mountpoint */ | |
1738 | if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) | |
1739 | goto out2; /* not attached */ | |
1740 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) | |
1741 | goto out2; /* not a mountpoint */ | |
1742 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) | |
1743 | goto out2; /* not attached */ | |
1744 | /* make sure we can reach put_old from new_root */ | |
1745 | tmp = old_nd.path.mnt; | |
1746 | spin_lock(&vfsmount_lock); | |
1747 | if (tmp != new_nd.path.mnt) { | |
1748 | for (;;) { | |
1749 | if (tmp->mnt_parent == tmp) | |
1750 | goto out3; /* already mounted on put_old */ | |
1751 | if (tmp->mnt_parent == new_nd.path.mnt) | |
1752 | break; | |
1753 | tmp = tmp->mnt_parent; | |
1754 | } | |
1755 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) | |
1756 | goto out3; | |
1757 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) | |
1758 | goto out3; | |
1759 | detach_mnt(new_nd.path.mnt, &parent_path); | |
1760 | detach_mnt(user_nd.path.mnt, &root_parent); | |
1761 | /* mount old root on put_old */ | |
1762 | attach_mnt(user_nd.path.mnt, &old_nd.path); | |
1763 | /* mount new_root on / */ | |
1764 | attach_mnt(new_nd.path.mnt, &root_parent); | |
1765 | touch_mnt_namespace(current->nsproxy->mnt_ns); | |
1766 | spin_unlock(&vfsmount_lock); | |
1767 | chroot_fs_refs(&user_nd.path, &new_nd.path); | |
1768 | security_sb_post_pivotroot(&user_nd, &new_nd); | |
1769 | error = 0; | |
1770 | path_put(&root_parent); | |
1771 | path_put(&parent_path); | |
1772 | out2: | |
1773 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); | |
1774 | up_write(&namespace_sem); | |
1775 | path_put(&user_nd.path); | |
1776 | path_put(&old_nd.path); | |
1777 | out1: | |
1778 | path_put(&new_nd.path); | |
1779 | out0: | |
1780 | unlock_kernel(); | |
1781 | return error; | |
1782 | out3: | |
1783 | spin_unlock(&vfsmount_lock); | |
1784 | goto out2; | |
1785 | } | |
1786 | ||
1787 | static void __init init_mount_tree(void) | |
1788 | { | |
1789 | struct vfsmount *mnt; | |
1790 | struct mnt_namespace *ns; | |
1791 | struct path root; | |
1792 | ||
1793 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1794 | if (IS_ERR(mnt)) | |
1795 | panic("Can't create rootfs"); | |
1796 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); | |
1797 | if (!ns) | |
1798 | panic("Can't allocate initial namespace"); | |
1799 | atomic_set(&ns->count, 1); | |
1800 | INIT_LIST_HEAD(&ns->list); | |
1801 | init_waitqueue_head(&ns->poll); | |
1802 | ns->event = 0; | |
1803 | list_add(&mnt->mnt_list, &ns->list); | |
1804 | ns->root = mnt; | |
1805 | mnt->mnt_ns = ns; | |
1806 | ||
1807 | init_task.nsproxy->mnt_ns = ns; | |
1808 | get_mnt_ns(ns); | |
1809 | ||
1810 | root.mnt = ns->root; | |
1811 | root.dentry = ns->root->mnt_root; | |
1812 | ||
1813 | set_fs_pwd(current->fs, &root); | |
1814 | set_fs_root(current->fs, &root); | |
1815 | } | |
1816 | ||
1817 | void __init mnt_init(void) | |
1818 | { | |
1819 | unsigned u; | |
1820 | int err; | |
1821 | ||
1822 | init_rwsem(&namespace_sem); | |
1823 | ||
1824 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), | |
1825 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
1826 | ||
1827 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); | |
1828 | ||
1829 | if (!mount_hashtable) | |
1830 | panic("Failed to allocate mount hash table\n"); | |
1831 | ||
1832 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); | |
1833 | ||
1834 | for (u = 0; u < HASH_SIZE; u++) | |
1835 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1836 | ||
1837 | err = sysfs_init(); | |
1838 | if (err) | |
1839 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
1840 | __FUNCTION__, err); | |
1841 | fs_kobj = kobject_create_and_add("fs", NULL); | |
1842 | if (!fs_kobj) | |
1843 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1844 | init_rootfs(); | |
1845 | init_mount_tree(); | |
1846 | } | |
1847 | ||
1848 | void __put_mnt_ns(struct mnt_namespace *ns) | |
1849 | { | |
1850 | struct vfsmount *root = ns->root; | |
1851 | LIST_HEAD(umount_list); | |
1852 | ns->root = NULL; | |
1853 | spin_unlock(&vfsmount_lock); | |
1854 | down_write(&namespace_sem); | |
1855 | spin_lock(&vfsmount_lock); | |
1856 | umount_tree(root, 0, &umount_list); | |
1857 | spin_unlock(&vfsmount_lock); | |
1858 | up_write(&namespace_sem); | |
1859 | release_mounts(&umount_list); | |
1860 | kfree(ns); | |
1861 | } |