]>
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 | ||
1da177e4 LT |
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> | |
15a67dd8 | 16 | #include <linux/kernel.h> |
1da177e4 LT |
17 | #include <linux/quotaops.h> |
18 | #include <linux/acct.h> | |
16f7e0fe | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/module.h> |
f20a9ead | 21 | #include <linux/sysfs.h> |
1da177e4 | 22 | #include <linux/seq_file.h> |
6b3286ed | 23 | #include <linux/mnt_namespace.h> |
1da177e4 LT |
24 | #include <linux/namei.h> |
25 | #include <linux/security.h> | |
26 | #include <linux/mount.h> | |
07f3f05c | 27 | #include <linux/ramfs.h> |
13f14b4d | 28 | #include <linux/log2.h> |
1da177e4 LT |
29 | #include <asm/uaccess.h> |
30 | #include <asm/unistd.h> | |
07b20889 | 31 | #include "pnode.h" |
948730b0 | 32 | #include "internal.h" |
1da177e4 | 33 | |
13f14b4d ED |
34 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
35 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
36 | ||
1da177e4 | 37 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
38 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
39 | ||
40 | static int event; | |
1da177e4 | 41 | |
fa3536cc | 42 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 43 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 44 | static struct rw_semaphore namespace_sem; |
1da177e4 | 45 | |
f87fd4c2 | 46 | /* /sys/fs */ |
00d26666 GKH |
47 | struct kobject *fs_kobj; |
48 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 49 | |
1da177e4 LT |
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); | |
13f14b4d ED |
54 | tmp = tmp + (tmp >> HASH_SHIFT); |
55 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
56 | } |
57 | ||
58 | struct vfsmount *alloc_vfsmnt(const char *name) | |
59 | { | |
c3762229 | 60 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 61 | if (mnt) { |
b58fed8b | 62 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
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); | |
55e700b9 | 67 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 68 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
69 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
70 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
1da177e4 | 71 | if (name) { |
b58fed8b | 72 | int size = strlen(name) + 1; |
1da177e4 LT |
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 | ||
8366025e DH |
83 | /* |
84 | * Most r/o checks on a fs are for operations that take | |
85 | * discrete amounts of time, like a write() or unlink(). | |
86 | * We must keep track of when those operations start | |
87 | * (for permission checks) and when they end, so that | |
88 | * we can determine when writes are able to occur to | |
89 | * a filesystem. | |
90 | */ | |
91 | /** | |
92 | * mnt_want_write - get write access to a mount | |
93 | * @mnt: the mount on which to take a write | |
94 | * | |
95 | * This tells the low-level filesystem that a write is | |
96 | * about to be performed to it, and makes sure that | |
97 | * writes are allowed before returning success. When | |
98 | * the write operation is finished, mnt_drop_write() | |
99 | * must be called. This is effectively a refcount. | |
100 | */ | |
101 | int mnt_want_write(struct vfsmount *mnt) | |
102 | { | |
103 | if (__mnt_is_readonly(mnt)) | |
104 | return -EROFS; | |
105 | return 0; | |
106 | } | |
107 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
108 | ||
109 | /** | |
110 | * mnt_drop_write - give up write access to a mount | |
111 | * @mnt: the mount on which to give up write access | |
112 | * | |
113 | * Tells the low-level filesystem that we are done | |
114 | * performing writes to it. Must be matched with | |
115 | * mnt_want_write() call above. | |
116 | */ | |
117 | void mnt_drop_write(struct vfsmount *mnt) | |
118 | { | |
119 | } | |
120 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
121 | ||
122 | /* | |
123 | * __mnt_is_readonly: check whether a mount is read-only | |
124 | * @mnt: the mount to check for its write status | |
125 | * | |
126 | * This shouldn't be used directly ouside of the VFS. | |
127 | * It does not guarantee that the filesystem will stay | |
128 | * r/w, just that it is right *now*. This can not and | |
129 | * should not be used in place of IS_RDONLY(inode). | |
130 | */ | |
131 | int __mnt_is_readonly(struct vfsmount *mnt) | |
132 | { | |
133 | return (mnt->mnt_sb->s_flags & MS_RDONLY); | |
134 | } | |
135 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
136 | ||
454e2398 DH |
137 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
138 | { | |
139 | mnt->mnt_sb = sb; | |
140 | mnt->mnt_root = dget(sb->s_root); | |
141 | return 0; | |
142 | } | |
143 | ||
144 | EXPORT_SYMBOL(simple_set_mnt); | |
145 | ||
1da177e4 LT |
146 | void free_vfsmnt(struct vfsmount *mnt) |
147 | { | |
148 | kfree(mnt->mnt_devname); | |
149 | kmem_cache_free(mnt_cache, mnt); | |
150 | } | |
151 | ||
152 | /* | |
a05964f3 RP |
153 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
154 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 155 | */ |
a05964f3 RP |
156 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
157 | int dir) | |
1da177e4 | 158 | { |
b58fed8b RP |
159 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
160 | struct list_head *tmp = head; | |
1da177e4 LT |
161 | struct vfsmount *p, *found = NULL; |
162 | ||
1da177e4 | 163 | for (;;) { |
a05964f3 | 164 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
165 | p = NULL; |
166 | if (tmp == head) | |
167 | break; | |
168 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
169 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 170 | found = p; |
1da177e4 LT |
171 | break; |
172 | } | |
173 | } | |
1da177e4 LT |
174 | return found; |
175 | } | |
176 | ||
a05964f3 RP |
177 | /* |
178 | * lookup_mnt increments the ref count before returning | |
179 | * the vfsmount struct. | |
180 | */ | |
181 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
182 | { | |
183 | struct vfsmount *child_mnt; | |
184 | spin_lock(&vfsmount_lock); | |
185 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
186 | mntget(child_mnt); | |
187 | spin_unlock(&vfsmount_lock); | |
188 | return child_mnt; | |
189 | } | |
190 | ||
1da177e4 LT |
191 | static inline int check_mnt(struct vfsmount *mnt) |
192 | { | |
6b3286ed | 193 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
194 | } |
195 | ||
6b3286ed | 196 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
197 | { |
198 | if (ns) { | |
199 | ns->event = ++event; | |
200 | wake_up_interruptible(&ns->poll); | |
201 | } | |
202 | } | |
203 | ||
6b3286ed | 204 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
205 | { |
206 | if (ns && ns->event != event) { | |
207 | ns->event = event; | |
208 | wake_up_interruptible(&ns->poll); | |
209 | } | |
210 | } | |
211 | ||
1a390689 | 212 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 213 | { |
1a390689 AV |
214 | old_path->dentry = mnt->mnt_mountpoint; |
215 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
216 | mnt->mnt_parent = mnt; |
217 | mnt->mnt_mountpoint = mnt->mnt_root; | |
218 | list_del_init(&mnt->mnt_child); | |
219 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 220 | old_path->dentry->d_mounted--; |
1da177e4 LT |
221 | } |
222 | ||
b90fa9ae RP |
223 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
224 | struct vfsmount *child_mnt) | |
225 | { | |
226 | child_mnt->mnt_parent = mntget(mnt); | |
227 | child_mnt->mnt_mountpoint = dget(dentry); | |
228 | dentry->d_mounted++; | |
229 | } | |
230 | ||
1a390689 | 231 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 232 | { |
1a390689 | 233 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 234 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
235 | hash(path->mnt, path->dentry)); |
236 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
237 | } |
238 | ||
239 | /* | |
240 | * the caller must hold vfsmount_lock | |
241 | */ | |
242 | static void commit_tree(struct vfsmount *mnt) | |
243 | { | |
244 | struct vfsmount *parent = mnt->mnt_parent; | |
245 | struct vfsmount *m; | |
246 | LIST_HEAD(head); | |
6b3286ed | 247 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
248 | |
249 | BUG_ON(parent == mnt); | |
250 | ||
251 | list_add_tail(&head, &mnt->mnt_list); | |
252 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 253 | m->mnt_ns = n; |
b90fa9ae RP |
254 | list_splice(&head, n->list.prev); |
255 | ||
256 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
257 | hash(parent, mnt->mnt_mountpoint)); | |
258 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 259 | touch_mnt_namespace(n); |
1da177e4 LT |
260 | } |
261 | ||
262 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
263 | { | |
264 | struct list_head *next = p->mnt_mounts.next; | |
265 | if (next == &p->mnt_mounts) { | |
266 | while (1) { | |
267 | if (p == root) | |
268 | return NULL; | |
269 | next = p->mnt_child.next; | |
270 | if (next != &p->mnt_parent->mnt_mounts) | |
271 | break; | |
272 | p = p->mnt_parent; | |
273 | } | |
274 | } | |
275 | return list_entry(next, struct vfsmount, mnt_child); | |
276 | } | |
277 | ||
9676f0c6 RP |
278 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
279 | { | |
280 | struct list_head *prev = p->mnt_mounts.prev; | |
281 | while (prev != &p->mnt_mounts) { | |
282 | p = list_entry(prev, struct vfsmount, mnt_child); | |
283 | prev = p->mnt_mounts.prev; | |
284 | } | |
285 | return p; | |
286 | } | |
287 | ||
36341f64 RP |
288 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
289 | int flag) | |
1da177e4 LT |
290 | { |
291 | struct super_block *sb = old->mnt_sb; | |
292 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
293 | ||
294 | if (mnt) { | |
295 | mnt->mnt_flags = old->mnt_flags; | |
296 | atomic_inc(&sb->s_active); | |
297 | mnt->mnt_sb = sb; | |
298 | mnt->mnt_root = dget(root); | |
299 | mnt->mnt_mountpoint = mnt->mnt_root; | |
300 | mnt->mnt_parent = mnt; | |
b90fa9ae | 301 | |
5afe0022 RP |
302 | if (flag & CL_SLAVE) { |
303 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
304 | mnt->mnt_master = old; | |
305 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 306 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
307 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
308 | list_add(&mnt->mnt_share, &old->mnt_share); | |
309 | if (IS_MNT_SLAVE(old)) | |
310 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
311 | mnt->mnt_master = old->mnt_master; | |
312 | } | |
b90fa9ae RP |
313 | if (flag & CL_MAKE_SHARED) |
314 | set_mnt_shared(mnt); | |
1da177e4 LT |
315 | |
316 | /* stick the duplicate mount on the same expiry list | |
317 | * as the original if that was on one */ | |
36341f64 | 318 | if (flag & CL_EXPIRE) { |
36341f64 RP |
319 | if (!list_empty(&old->mnt_expire)) |
320 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 321 | } |
1da177e4 LT |
322 | } |
323 | return mnt; | |
324 | } | |
325 | ||
7b7b1ace | 326 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
327 | { |
328 | struct super_block *sb = mnt->mnt_sb; | |
329 | dput(mnt->mnt_root); | |
330 | free_vfsmnt(mnt); | |
331 | deactivate_super(sb); | |
332 | } | |
333 | ||
7b7b1ace AV |
334 | void mntput_no_expire(struct vfsmount *mnt) |
335 | { | |
336 | repeat: | |
337 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
338 | if (likely(!mnt->mnt_pinned)) { | |
339 | spin_unlock(&vfsmount_lock); | |
340 | __mntput(mnt); | |
341 | return; | |
342 | } | |
343 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
344 | mnt->mnt_pinned = 0; | |
345 | spin_unlock(&vfsmount_lock); | |
346 | acct_auto_close_mnt(mnt); | |
347 | security_sb_umount_close(mnt); | |
348 | goto repeat; | |
349 | } | |
350 | } | |
351 | ||
352 | EXPORT_SYMBOL(mntput_no_expire); | |
353 | ||
354 | void mnt_pin(struct vfsmount *mnt) | |
355 | { | |
356 | spin_lock(&vfsmount_lock); | |
357 | mnt->mnt_pinned++; | |
358 | spin_unlock(&vfsmount_lock); | |
359 | } | |
360 | ||
361 | EXPORT_SYMBOL(mnt_pin); | |
362 | ||
363 | void mnt_unpin(struct vfsmount *mnt) | |
364 | { | |
365 | spin_lock(&vfsmount_lock); | |
366 | if (mnt->mnt_pinned) { | |
367 | atomic_inc(&mnt->mnt_count); | |
368 | mnt->mnt_pinned--; | |
369 | } | |
370 | spin_unlock(&vfsmount_lock); | |
371 | } | |
372 | ||
373 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 374 | |
b3b304a2 MS |
375 | static inline void mangle(struct seq_file *m, const char *s) |
376 | { | |
377 | seq_escape(m, s, " \t\n\\"); | |
378 | } | |
379 | ||
380 | /* | |
381 | * Simple .show_options callback for filesystems which don't want to | |
382 | * implement more complex mount option showing. | |
383 | * | |
384 | * See also save_mount_options(). | |
385 | */ | |
386 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
387 | { | |
388 | const char *options = mnt->mnt_sb->s_options; | |
389 | ||
390 | if (options != NULL && options[0]) { | |
391 | seq_putc(m, ','); | |
392 | mangle(m, options); | |
393 | } | |
394 | ||
395 | return 0; | |
396 | } | |
397 | EXPORT_SYMBOL(generic_show_options); | |
398 | ||
399 | /* | |
400 | * If filesystem uses generic_show_options(), this function should be | |
401 | * called from the fill_super() callback. | |
402 | * | |
403 | * The .remount_fs callback usually needs to be handled in a special | |
404 | * way, to make sure, that previous options are not overwritten if the | |
405 | * remount fails. | |
406 | * | |
407 | * Also note, that if the filesystem's .remount_fs function doesn't | |
408 | * reset all options to their default value, but changes only newly | |
409 | * given options, then the displayed options will not reflect reality | |
410 | * any more. | |
411 | */ | |
412 | void save_mount_options(struct super_block *sb, char *options) | |
413 | { | |
414 | kfree(sb->s_options); | |
415 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
416 | } | |
417 | EXPORT_SYMBOL(save_mount_options); | |
418 | ||
1da177e4 LT |
419 | /* iterator */ |
420 | static void *m_start(struct seq_file *m, loff_t *pos) | |
421 | { | |
6b3286ed | 422 | struct mnt_namespace *n = m->private; |
1da177e4 | 423 | |
390c6843 | 424 | down_read(&namespace_sem); |
b0765fb8 | 425 | return seq_list_start(&n->list, *pos); |
1da177e4 LT |
426 | } |
427 | ||
428 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
429 | { | |
6b3286ed | 430 | struct mnt_namespace *n = m->private; |
b0765fb8 PE |
431 | |
432 | return seq_list_next(v, &n->list, pos); | |
1da177e4 LT |
433 | } |
434 | ||
435 | static void m_stop(struct seq_file *m, void *v) | |
436 | { | |
390c6843 | 437 | up_read(&namespace_sem); |
1da177e4 LT |
438 | } |
439 | ||
1da177e4 LT |
440 | static int show_vfsmnt(struct seq_file *m, void *v) |
441 | { | |
b0765fb8 | 442 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
1da177e4 LT |
443 | int err = 0; |
444 | static struct proc_fs_info { | |
445 | int flag; | |
446 | char *str; | |
447 | } fs_info[] = { | |
448 | { MS_SYNCHRONOUS, ",sync" }, | |
449 | { MS_DIRSYNC, ",dirsync" }, | |
450 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
451 | { 0, NULL } |
452 | }; | |
453 | static struct proc_fs_info mnt_info[] = { | |
454 | { MNT_NOSUID, ",nosuid" }, | |
455 | { MNT_NODEV, ",nodev" }, | |
456 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
457 | { MNT_NOATIME, ",noatime" }, |
458 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 459 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
460 | { 0, NULL } |
461 | }; | |
462 | struct proc_fs_info *fs_infop; | |
c32c2f63 | 463 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
464 | |
465 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
466 | seq_putc(m, ' '); | |
c32c2f63 | 467 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 LT |
468 | seq_putc(m, ' '); |
469 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
470 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { |
471 | seq_putc(m, '.'); | |
472 | mangle(m, mnt->mnt_sb->s_subtype); | |
473 | } | |
1da177e4 LT |
474 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); |
475 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
476 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
477 | seq_puts(m, fs_infop->str); | |
478 | } | |
479 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
480 | if (mnt->mnt_flags & fs_infop->flag) | |
481 | seq_puts(m, fs_infop->str); | |
482 | } | |
483 | if (mnt->mnt_sb->s_op->show_options) | |
484 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
485 | seq_puts(m, " 0 0\n"); | |
486 | return err; | |
487 | } | |
488 | ||
489 | struct seq_operations mounts_op = { | |
490 | .start = m_start, | |
491 | .next = m_next, | |
492 | .stop = m_stop, | |
493 | .show = show_vfsmnt | |
494 | }; | |
495 | ||
b4629fe2 CL |
496 | static int show_vfsstat(struct seq_file *m, void *v) |
497 | { | |
b0765fb8 | 498 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 499 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
500 | int err = 0; |
501 | ||
502 | /* device */ | |
503 | if (mnt->mnt_devname) { | |
504 | seq_puts(m, "device "); | |
505 | mangle(m, mnt->mnt_devname); | |
506 | } else | |
507 | seq_puts(m, "no device"); | |
508 | ||
509 | /* mount point */ | |
510 | seq_puts(m, " mounted on "); | |
c32c2f63 | 511 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
512 | seq_putc(m, ' '); |
513 | ||
514 | /* file system type */ | |
515 | seq_puts(m, "with fstype "); | |
516 | mangle(m, mnt->mnt_sb->s_type->name); | |
517 | ||
518 | /* optional statistics */ | |
519 | if (mnt->mnt_sb->s_op->show_stats) { | |
520 | seq_putc(m, ' '); | |
521 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
522 | } | |
523 | ||
524 | seq_putc(m, '\n'); | |
525 | return err; | |
526 | } | |
527 | ||
528 | struct seq_operations mountstats_op = { | |
529 | .start = m_start, | |
530 | .next = m_next, | |
531 | .stop = m_stop, | |
532 | .show = show_vfsstat, | |
533 | }; | |
534 | ||
1da177e4 LT |
535 | /** |
536 | * may_umount_tree - check if a mount tree is busy | |
537 | * @mnt: root of mount tree | |
538 | * | |
539 | * This is called to check if a tree of mounts has any | |
540 | * open files, pwds, chroots or sub mounts that are | |
541 | * busy. | |
542 | */ | |
543 | int may_umount_tree(struct vfsmount *mnt) | |
544 | { | |
36341f64 RP |
545 | int actual_refs = 0; |
546 | int minimum_refs = 0; | |
547 | struct vfsmount *p; | |
1da177e4 LT |
548 | |
549 | spin_lock(&vfsmount_lock); | |
36341f64 | 550 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
551 | actual_refs += atomic_read(&p->mnt_count); |
552 | minimum_refs += 2; | |
1da177e4 LT |
553 | } |
554 | spin_unlock(&vfsmount_lock); | |
555 | ||
556 | if (actual_refs > minimum_refs) | |
e3474a8e | 557 | return 0; |
1da177e4 | 558 | |
e3474a8e | 559 | return 1; |
1da177e4 LT |
560 | } |
561 | ||
562 | EXPORT_SYMBOL(may_umount_tree); | |
563 | ||
564 | /** | |
565 | * may_umount - check if a mount point is busy | |
566 | * @mnt: root of mount | |
567 | * | |
568 | * This is called to check if a mount point has any | |
569 | * open files, pwds, chroots or sub mounts. If the | |
570 | * mount has sub mounts this will return busy | |
571 | * regardless of whether the sub mounts are busy. | |
572 | * | |
573 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
574 | * give false negatives. The main reason why it's here is that we need | |
575 | * a non-destructive way to look for easily umountable filesystems. | |
576 | */ | |
577 | int may_umount(struct vfsmount *mnt) | |
578 | { | |
e3474a8e | 579 | int ret = 1; |
a05964f3 RP |
580 | spin_lock(&vfsmount_lock); |
581 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 582 | ret = 0; |
a05964f3 RP |
583 | spin_unlock(&vfsmount_lock); |
584 | return ret; | |
1da177e4 LT |
585 | } |
586 | ||
587 | EXPORT_SYMBOL(may_umount); | |
588 | ||
b90fa9ae | 589 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
590 | { |
591 | struct vfsmount *mnt; | |
bf066c7d | 592 | while (!list_empty(head)) { |
b5e61818 | 593 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
594 | list_del_init(&mnt->mnt_hash); |
595 | if (mnt->mnt_parent != mnt) { | |
596 | struct dentry *dentry; | |
597 | struct vfsmount *m; | |
598 | spin_lock(&vfsmount_lock); | |
599 | dentry = mnt->mnt_mountpoint; | |
600 | m = mnt->mnt_parent; | |
601 | mnt->mnt_mountpoint = mnt->mnt_root; | |
602 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 603 | m->mnt_ghosts--; |
70fbcdf4 RP |
604 | spin_unlock(&vfsmount_lock); |
605 | dput(dentry); | |
606 | mntput(m); | |
607 | } | |
608 | mntput(mnt); | |
609 | } | |
610 | } | |
611 | ||
a05964f3 | 612 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
613 | { |
614 | struct vfsmount *p; | |
1da177e4 | 615 | |
1bfba4e8 AM |
616 | for (p = mnt; p; p = next_mnt(p, mnt)) |
617 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 618 | |
a05964f3 RP |
619 | if (propagate) |
620 | propagate_umount(kill); | |
621 | ||
70fbcdf4 RP |
622 | list_for_each_entry(p, kill, mnt_hash) { |
623 | list_del_init(&p->mnt_expire); | |
624 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
625 | __touch_mnt_namespace(p->mnt_ns); |
626 | p->mnt_ns = NULL; | |
70fbcdf4 | 627 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
628 | if (p->mnt_parent != p) { |
629 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 630 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 631 | } |
a05964f3 | 632 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
633 | } |
634 | } | |
635 | ||
c35038be AV |
636 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
637 | ||
1da177e4 LT |
638 | static int do_umount(struct vfsmount *mnt, int flags) |
639 | { | |
b58fed8b | 640 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 641 | int retval; |
70fbcdf4 | 642 | LIST_HEAD(umount_list); |
1da177e4 LT |
643 | |
644 | retval = security_sb_umount(mnt, flags); | |
645 | if (retval) | |
646 | return retval; | |
647 | ||
648 | /* | |
649 | * Allow userspace to request a mountpoint be expired rather than | |
650 | * unmounting unconditionally. Unmount only happens if: | |
651 | * (1) the mark is already set (the mark is cleared by mntput()) | |
652 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
653 | */ | |
654 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 655 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
656 | flags & (MNT_FORCE | MNT_DETACH)) |
657 | return -EINVAL; | |
658 | ||
659 | if (atomic_read(&mnt->mnt_count) != 2) | |
660 | return -EBUSY; | |
661 | ||
662 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
663 | return -EAGAIN; | |
664 | } | |
665 | ||
666 | /* | |
667 | * If we may have to abort operations to get out of this | |
668 | * mount, and they will themselves hold resources we must | |
669 | * allow the fs to do things. In the Unix tradition of | |
670 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
671 | * might fail to complete on the first run through as other tasks | |
672 | * must return, and the like. Thats for the mount program to worry | |
673 | * about for the moment. | |
674 | */ | |
675 | ||
676 | lock_kernel(); | |
8b512d9a TM |
677 | if (sb->s_op->umount_begin) |
678 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
679 | unlock_kernel(); |
680 | ||
681 | /* | |
682 | * No sense to grab the lock for this test, but test itself looks | |
683 | * somewhat bogus. Suggestions for better replacement? | |
684 | * Ho-hum... In principle, we might treat that as umount + switch | |
685 | * to rootfs. GC would eventually take care of the old vfsmount. | |
686 | * Actually it makes sense, especially if rootfs would contain a | |
687 | * /reboot - static binary that would close all descriptors and | |
688 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
689 | */ | |
6ac08c39 | 690 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
691 | /* |
692 | * Special case for "unmounting" root ... | |
693 | * we just try to remount it readonly. | |
694 | */ | |
695 | down_write(&sb->s_umount); | |
696 | if (!(sb->s_flags & MS_RDONLY)) { | |
697 | lock_kernel(); | |
698 | DQUOT_OFF(sb); | |
699 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
700 | unlock_kernel(); | |
701 | } | |
702 | up_write(&sb->s_umount); | |
703 | return retval; | |
704 | } | |
705 | ||
390c6843 | 706 | down_write(&namespace_sem); |
1da177e4 | 707 | spin_lock(&vfsmount_lock); |
5addc5dd | 708 | event++; |
1da177e4 | 709 | |
c35038be AV |
710 | if (!(flags & MNT_DETACH)) |
711 | shrink_submounts(mnt, &umount_list); | |
712 | ||
1da177e4 | 713 | retval = -EBUSY; |
a05964f3 | 714 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 715 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 716 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
717 | retval = 0; |
718 | } | |
719 | spin_unlock(&vfsmount_lock); | |
720 | if (retval) | |
721 | security_sb_umount_busy(mnt); | |
390c6843 | 722 | up_write(&namespace_sem); |
70fbcdf4 | 723 | release_mounts(&umount_list); |
1da177e4 LT |
724 | return retval; |
725 | } | |
726 | ||
727 | /* | |
728 | * Now umount can handle mount points as well as block devices. | |
729 | * This is important for filesystems which use unnamed block devices. | |
730 | * | |
731 | * We now support a flag for forced unmount like the other 'big iron' | |
732 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
733 | */ | |
734 | ||
735 | asmlinkage long sys_umount(char __user * name, int flags) | |
736 | { | |
737 | struct nameidata nd; | |
738 | int retval; | |
739 | ||
740 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
741 | if (retval) | |
742 | goto out; | |
743 | retval = -EINVAL; | |
4ac91378 | 744 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 745 | goto dput_and_out; |
4ac91378 | 746 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
747 | goto dput_and_out; |
748 | ||
749 | retval = -EPERM; | |
750 | if (!capable(CAP_SYS_ADMIN)) | |
751 | goto dput_and_out; | |
752 | ||
4ac91378 | 753 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 754 | dput_and_out: |
429731b1 | 755 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
756 | dput(nd.path.dentry); |
757 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
758 | out: |
759 | return retval; | |
760 | } | |
761 | ||
762 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
763 | ||
764 | /* | |
b58fed8b | 765 | * The 2.0 compatible umount. No flags. |
1da177e4 | 766 | */ |
1da177e4 LT |
767 | asmlinkage long sys_oldumount(char __user * name) |
768 | { | |
b58fed8b | 769 | return sys_umount(name, 0); |
1da177e4 LT |
770 | } |
771 | ||
772 | #endif | |
773 | ||
774 | static int mount_is_safe(struct nameidata *nd) | |
775 | { | |
776 | if (capable(CAP_SYS_ADMIN)) | |
777 | return 0; | |
778 | return -EPERM; | |
779 | #ifdef notyet | |
4ac91378 | 780 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 781 | return -EPERM; |
4ac91378 JB |
782 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
783 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
784 | return -EPERM; |
785 | } | |
e4543edd | 786 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
787 | return -EPERM; |
788 | return 0; | |
789 | #endif | |
790 | } | |
791 | ||
b58fed8b | 792 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
793 | { |
794 | while (1) { | |
795 | if (d == dentry) | |
796 | return 1; | |
797 | if (d == NULL || d == d->d_parent) | |
798 | return 0; | |
799 | d = d->d_parent; | |
800 | } | |
801 | } | |
802 | ||
b90fa9ae | 803 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 804 | int flag) |
1da177e4 LT |
805 | { |
806 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 807 | struct path path; |
1da177e4 | 808 | |
9676f0c6 RP |
809 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
810 | return NULL; | |
811 | ||
36341f64 | 812 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
813 | if (!q) |
814 | goto Enomem; | |
815 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
816 | ||
817 | p = mnt; | |
fdadd65f | 818 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
819 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
820 | continue; | |
821 | ||
822 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
823 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
824 | s = skip_mnt_tree(s); | |
825 | continue; | |
826 | } | |
1da177e4 LT |
827 | while (p != s->mnt_parent) { |
828 | p = p->mnt_parent; | |
829 | q = q->mnt_parent; | |
830 | } | |
831 | p = s; | |
1a390689 AV |
832 | path.mnt = q; |
833 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 834 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
835 | if (!q) |
836 | goto Enomem; | |
837 | spin_lock(&vfsmount_lock); | |
838 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 839 | attach_mnt(q, &path); |
1da177e4 LT |
840 | spin_unlock(&vfsmount_lock); |
841 | } | |
842 | } | |
843 | return res; | |
b58fed8b | 844 | Enomem: |
1da177e4 | 845 | if (res) { |
70fbcdf4 | 846 | LIST_HEAD(umount_list); |
1da177e4 | 847 | spin_lock(&vfsmount_lock); |
a05964f3 | 848 | umount_tree(res, 0, &umount_list); |
1da177e4 | 849 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 850 | release_mounts(&umount_list); |
1da177e4 LT |
851 | } |
852 | return NULL; | |
853 | } | |
854 | ||
8aec0809 AV |
855 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
856 | { | |
857 | struct vfsmount *tree; | |
858 | down_read(&namespace_sem); | |
859 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); | |
860 | up_read(&namespace_sem); | |
861 | return tree; | |
862 | } | |
863 | ||
864 | void drop_collected_mounts(struct vfsmount *mnt) | |
865 | { | |
866 | LIST_HEAD(umount_list); | |
867 | down_read(&namespace_sem); | |
868 | spin_lock(&vfsmount_lock); | |
869 | umount_tree(mnt, 0, &umount_list); | |
870 | spin_unlock(&vfsmount_lock); | |
871 | up_read(&namespace_sem); | |
872 | release_mounts(&umount_list); | |
873 | } | |
874 | ||
b90fa9ae RP |
875 | /* |
876 | * @source_mnt : mount tree to be attached | |
21444403 RP |
877 | * @nd : place the mount tree @source_mnt is attached |
878 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
879 | * store the parent mount and mountpoint dentry. | |
880 | * (done when source_mnt is moved) | |
b90fa9ae RP |
881 | * |
882 | * NOTE: in the table below explains the semantics when a source mount | |
883 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
884 | * --------------------------------------------------------------------------- |
885 | * | BIND MOUNT OPERATION | | |
886 | * |************************************************************************** | |
887 | * | source-->| shared | private | slave | unbindable | | |
888 | * | dest | | | | | | |
889 | * | | | | | | | | |
890 | * | v | | | | | | |
891 | * |************************************************************************** | |
892 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
893 | * | | | | | | | |
894 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
895 | * *************************************************************************** | |
b90fa9ae RP |
896 | * A bind operation clones the source mount and mounts the clone on the |
897 | * destination mount. | |
898 | * | |
899 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
900 | * tree of the destination mount and the cloned mount is added to | |
901 | * the peer group of the source mount. | |
902 | * (+) the cloned mount is created under the destination mount and is marked | |
903 | * as shared. The cloned mount is added to the peer group of the source | |
904 | * mount. | |
5afe0022 RP |
905 | * (+++) the mount is propagated to all the mounts in the propagation tree |
906 | * of the destination mount and the cloned mount is made slave | |
907 | * of the same master as that of the source mount. The cloned mount | |
908 | * is marked as 'shared and slave'. | |
909 | * (*) the cloned mount is made a slave of the same master as that of the | |
910 | * source mount. | |
911 | * | |
9676f0c6 RP |
912 | * --------------------------------------------------------------------------- |
913 | * | MOVE MOUNT OPERATION | | |
914 | * |************************************************************************** | |
915 | * | source-->| shared | private | slave | unbindable | | |
916 | * | dest | | | | | | |
917 | * | | | | | | | | |
918 | * | v | | | | | | |
919 | * |************************************************************************** | |
920 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
921 | * | | | | | | | |
922 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
923 | * *************************************************************************** | |
5afe0022 RP |
924 | * |
925 | * (+) the mount is moved to the destination. And is then propagated to | |
926 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 927 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
928 | * (+++) the mount is moved to the destination and is then propagated to |
929 | * all the mounts belonging to the destination mount's propagation tree. | |
930 | * the mount is marked as 'shared and slave'. | |
931 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
932 | * |
933 | * if the source mount is a tree, the operations explained above is | |
934 | * applied to each mount in the tree. | |
935 | * Must be called without spinlocks held, since this function can sleep | |
936 | * in allocations. | |
937 | */ | |
938 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 939 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
940 | { |
941 | LIST_HEAD(tree_list); | |
1a390689 AV |
942 | struct vfsmount *dest_mnt = path->mnt; |
943 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae RP |
944 | struct vfsmount *child, *p; |
945 | ||
946 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
947 | return -EINVAL; | |
948 | ||
949 | if (IS_MNT_SHARED(dest_mnt)) { | |
950 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
951 | set_mnt_shared(p); | |
952 | } | |
953 | ||
954 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
955 | if (parent_path) { |
956 | detach_mnt(source_mnt, parent_path); | |
957 | attach_mnt(source_mnt, path); | |
6b3286ed | 958 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
959 | } else { |
960 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
961 | commit_tree(source_mnt); | |
962 | } | |
b90fa9ae RP |
963 | |
964 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
965 | list_del_init(&child->mnt_hash); | |
966 | commit_tree(child); | |
967 | } | |
968 | spin_unlock(&vfsmount_lock); | |
969 | return 0; | |
970 | } | |
971 | ||
1da177e4 LT |
972 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
973 | { | |
974 | int err; | |
975 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
976 | return -EINVAL; | |
977 | ||
4ac91378 | 978 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1da177e4 LT |
979 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
980 | return -ENOTDIR; | |
981 | ||
982 | err = -ENOENT; | |
4ac91378 JB |
983 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
984 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
985 | goto out_unlock; |
986 | ||
987 | err = security_sb_check_sb(mnt, nd); | |
988 | if (err) | |
989 | goto out_unlock; | |
990 | ||
991 | err = -ENOENT; | |
4ac91378 | 992 | if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) |
1a390689 | 993 | err = attach_recursive_mnt(mnt, &nd->path, NULL); |
1da177e4 | 994 | out_unlock: |
4ac91378 | 995 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 LT |
996 | if (!err) |
997 | security_sb_post_addmount(mnt, nd); | |
998 | return err; | |
999 | } | |
1000 | ||
07b20889 RP |
1001 | /* |
1002 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 1003 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 1004 | */ |
2dafe1c4 | 1005 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 1006 | { |
4ac91378 | 1007 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
1008 | int recurse = flag & MS_REC; |
1009 | int type = flag & ~MS_REC; | |
1010 | ||
ee6f9582 MS |
1011 | if (!capable(CAP_SYS_ADMIN)) |
1012 | return -EPERM; | |
1013 | ||
4ac91378 | 1014 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
1015 | return -EINVAL; |
1016 | ||
1017 | down_write(&namespace_sem); | |
1018 | spin_lock(&vfsmount_lock); | |
1019 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1020 | change_mnt_propagation(m, type); | |
1021 | spin_unlock(&vfsmount_lock); | |
1022 | up_write(&namespace_sem); | |
1023 | return 0; | |
1024 | } | |
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * do loopback mount. | |
2dafe1c4 | 1028 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1029 | */ |
2dafe1c4 ES |
1030 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
1031 | int recurse) | |
1da177e4 LT |
1032 | { |
1033 | struct nameidata old_nd; | |
1034 | struct vfsmount *mnt = NULL; | |
1035 | int err = mount_is_safe(nd); | |
1036 | if (err) | |
1037 | return err; | |
1038 | if (!old_name || !*old_name) | |
1039 | return -EINVAL; | |
1040 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1041 | if (err) | |
1042 | return err; | |
1043 | ||
390c6843 | 1044 | down_write(&namespace_sem); |
1da177e4 | 1045 | err = -EINVAL; |
4ac91378 JB |
1046 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
1047 | goto out; | |
9676f0c6 | 1048 | |
4ac91378 | 1049 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 1050 | goto out; |
1da177e4 | 1051 | |
ccd48bc7 AV |
1052 | err = -ENOMEM; |
1053 | if (recurse) | |
4ac91378 | 1054 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 1055 | else |
4ac91378 | 1056 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1057 | |
1058 | if (!mnt) | |
1059 | goto out; | |
1060 | ||
ccd48bc7 AV |
1061 | err = graft_tree(mnt, nd); |
1062 | if (err) { | |
70fbcdf4 | 1063 | LIST_HEAD(umount_list); |
1da177e4 | 1064 | spin_lock(&vfsmount_lock); |
a05964f3 | 1065 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1066 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1067 | release_mounts(&umount_list); |
5b83d2c5 | 1068 | } |
1da177e4 | 1069 | |
ccd48bc7 | 1070 | out: |
390c6843 | 1071 | up_write(&namespace_sem); |
1d957f9b | 1072 | path_put(&old_nd.path); |
1da177e4 LT |
1073 | return err; |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * change filesystem flags. dir should be a physical root of filesystem. | |
1078 | * If you've mounted a non-root directory somewhere and want to do remount | |
1079 | * on it - tough luck. | |
2dafe1c4 | 1080 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1081 | */ |
2dafe1c4 | 1082 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1083 | void *data) |
1084 | { | |
1085 | int err; | |
4ac91378 | 1086 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1087 | |
1088 | if (!capable(CAP_SYS_ADMIN)) | |
1089 | return -EPERM; | |
1090 | ||
4ac91378 | 1091 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1092 | return -EINVAL; |
1093 | ||
4ac91378 | 1094 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1095 | return -EINVAL; |
1096 | ||
1097 | down_write(&sb->s_umount); | |
1098 | err = do_remount_sb(sb, flags, data, 0); | |
1099 | if (!err) | |
4ac91378 | 1100 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1101 | up_write(&sb->s_umount); |
1102 | if (!err) | |
4ac91378 | 1103 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1104 | return err; |
1105 | } | |
1106 | ||
9676f0c6 RP |
1107 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1108 | { | |
1109 | struct vfsmount *p; | |
1110 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1111 | if (IS_MNT_UNBINDABLE(p)) | |
1112 | return 1; | |
1113 | } | |
1114 | return 0; | |
1115 | } | |
1116 | ||
2dafe1c4 ES |
1117 | /* |
1118 | * noinline this do_mount helper to save do_mount stack space. | |
1119 | */ | |
1120 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1121 | { |
1a390689 AV |
1122 | struct nameidata old_nd; |
1123 | struct path parent_path; | |
1da177e4 LT |
1124 | struct vfsmount *p; |
1125 | int err = 0; | |
1126 | if (!capable(CAP_SYS_ADMIN)) | |
1127 | return -EPERM; | |
1128 | if (!old_name || !*old_name) | |
1129 | return -EINVAL; | |
1130 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1131 | if (err) | |
1132 | return err; | |
1133 | ||
390c6843 | 1134 | down_write(&namespace_sem); |
4ac91378 JB |
1135 | while (d_mountpoint(nd->path.dentry) && |
1136 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1137 | ; |
1138 | err = -EINVAL; | |
4ac91378 | 1139 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1140 | goto out; |
1141 | ||
1142 | err = -ENOENT; | |
4ac91378 JB |
1143 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1144 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1145 | goto out1; |
1146 | ||
4ac91378 | 1147 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1148 | goto out1; |
1da177e4 LT |
1149 | |
1150 | err = -EINVAL; | |
4ac91378 | 1151 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1152 | goto out1; |
1da177e4 | 1153 | |
4ac91378 | 1154 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1155 | goto out1; |
1da177e4 | 1156 | |
4ac91378 JB |
1157 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1158 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1159 | goto out1; |
1160 | /* | |
1161 | * Don't move a mount residing in a shared parent. | |
1162 | */ | |
4ac91378 JB |
1163 | if (old_nd.path.mnt->mnt_parent && |
1164 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1165 | goto out1; |
9676f0c6 RP |
1166 | /* |
1167 | * Don't move a mount tree containing unbindable mounts to a destination | |
1168 | * mount which is shared. | |
1169 | */ | |
4ac91378 JB |
1170 | if (IS_MNT_SHARED(nd->path.mnt) && |
1171 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1172 | goto out1; |
1da177e4 | 1173 | err = -ELOOP; |
4ac91378 JB |
1174 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1175 | if (p == old_nd.path.mnt) | |
21444403 | 1176 | goto out1; |
1da177e4 | 1177 | |
1a390689 | 1178 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1179 | if (err) |
21444403 | 1180 | goto out1; |
1da177e4 LT |
1181 | |
1182 | /* if the mount is moved, it should no longer be expire | |
1183 | * automatically */ | |
4ac91378 | 1184 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 | 1185 | out1: |
4ac91378 | 1186 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1187 | out: |
390c6843 | 1188 | up_write(&namespace_sem); |
1da177e4 | 1189 | if (!err) |
1a390689 | 1190 | path_put(&parent_path); |
1d957f9b | 1191 | path_put(&old_nd.path); |
1da177e4 LT |
1192 | return err; |
1193 | } | |
1194 | ||
1195 | /* | |
1196 | * create a new mount for userspace and request it to be added into the | |
1197 | * namespace's tree | |
2dafe1c4 | 1198 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1199 | */ |
2dafe1c4 | 1200 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1201 | int mnt_flags, char *name, void *data) |
1202 | { | |
1203 | struct vfsmount *mnt; | |
1204 | ||
1205 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1206 | return -EINVAL; | |
1207 | ||
1208 | /* we need capabilities... */ | |
1209 | if (!capable(CAP_SYS_ADMIN)) | |
1210 | return -EPERM; | |
1211 | ||
1212 | mnt = do_kern_mount(type, flags, name, data); | |
1213 | if (IS_ERR(mnt)) | |
1214 | return PTR_ERR(mnt); | |
1215 | ||
1216 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * add a mount into a namespace's mount tree | |
1221 | * - provide the option of adding the new mount to an expiration list | |
1222 | */ | |
1223 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1224 | int mnt_flags, struct list_head *fslist) | |
1225 | { | |
1226 | int err; | |
1227 | ||
390c6843 | 1228 | down_write(&namespace_sem); |
1da177e4 | 1229 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1230 | while (d_mountpoint(nd->path.dentry) && |
1231 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1232 | ; |
1233 | err = -EINVAL; | |
4ac91378 | 1234 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1235 | goto unlock; |
1236 | ||
1237 | /* Refuse the same filesystem on the same mount point */ | |
1238 | err = -EBUSY; | |
4ac91378 JB |
1239 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1240 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1241 | goto unlock; |
1242 | ||
1243 | err = -EINVAL; | |
1244 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1245 | goto unlock; | |
1246 | ||
1247 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1248 | if ((err = graft_tree(newmnt, nd))) |
1249 | goto unlock; | |
1da177e4 | 1250 | |
6758f953 | 1251 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1252 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1253 | |
390c6843 | 1254 | up_write(&namespace_sem); |
5b83d2c5 | 1255 | return 0; |
1da177e4 LT |
1256 | |
1257 | unlock: | |
390c6843 | 1258 | up_write(&namespace_sem); |
1da177e4 LT |
1259 | mntput(newmnt); |
1260 | return err; | |
1261 | } | |
1262 | ||
1263 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1264 | ||
1265 | /* | |
1266 | * process a list of expirable mountpoints with the intent of discarding any | |
1267 | * mountpoints that aren't in use and haven't been touched since last we came | |
1268 | * here | |
1269 | */ | |
1270 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1271 | { | |
1da177e4 LT |
1272 | struct vfsmount *mnt, *next; |
1273 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1274 | LIST_HEAD(umounts); |
1da177e4 LT |
1275 | |
1276 | if (list_empty(mounts)) | |
1277 | return; | |
1278 | ||
bcc5c7d2 | 1279 | down_write(&namespace_sem); |
1da177e4 LT |
1280 | spin_lock(&vfsmount_lock); |
1281 | ||
1282 | /* extract from the expiration list every vfsmount that matches the | |
1283 | * following criteria: | |
1284 | * - only referenced by its parent vfsmount | |
1285 | * - still marked for expiry (marked on the last call here; marks are | |
1286 | * cleared by mntput()) | |
1287 | */ | |
55e700b9 | 1288 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1289 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1290 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1291 | continue; |
55e700b9 | 1292 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1293 | } |
bcc5c7d2 AV |
1294 | while (!list_empty(&graveyard)) { |
1295 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1296 | touch_mnt_namespace(mnt->mnt_ns); | |
1297 | umount_tree(mnt, 1, &umounts); | |
1298 | } | |
5528f911 | 1299 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1300 | up_write(&namespace_sem); |
1301 | ||
1302 | release_mounts(&umounts); | |
5528f911 TM |
1303 | } |
1304 | ||
1305 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1306 | ||
1307 | /* | |
1308 | * Ripoff of 'select_parent()' | |
1309 | * | |
1310 | * search the list of submounts for a given mountpoint, and move any | |
1311 | * shrinkable submounts to the 'graveyard' list. | |
1312 | */ | |
1313 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1314 | { | |
1315 | struct vfsmount *this_parent = parent; | |
1316 | struct list_head *next; | |
1317 | int found = 0; | |
1318 | ||
1319 | repeat: | |
1320 | next = this_parent->mnt_mounts.next; | |
1321 | resume: | |
1322 | while (next != &this_parent->mnt_mounts) { | |
1323 | struct list_head *tmp = next; | |
1324 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1325 | ||
1326 | next = tmp->next; | |
1327 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1328 | continue; |
5528f911 TM |
1329 | /* |
1330 | * Descend a level if the d_mounts list is non-empty. | |
1331 | */ | |
1332 | if (!list_empty(&mnt->mnt_mounts)) { | |
1333 | this_parent = mnt; | |
1334 | goto repeat; | |
1335 | } | |
1da177e4 | 1336 | |
5528f911 | 1337 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1338 | list_move_tail(&mnt->mnt_expire, graveyard); |
1339 | found++; | |
1340 | } | |
1da177e4 | 1341 | } |
5528f911 TM |
1342 | /* |
1343 | * All done at this level ... ascend and resume the search | |
1344 | */ | |
1345 | if (this_parent != parent) { | |
1346 | next = this_parent->mnt_child.next; | |
1347 | this_parent = this_parent->mnt_parent; | |
1348 | goto resume; | |
1349 | } | |
1350 | return found; | |
1351 | } | |
1352 | ||
1353 | /* | |
1354 | * process a list of expirable mountpoints with the intent of discarding any | |
1355 | * submounts of a specific parent mountpoint | |
1356 | */ | |
c35038be | 1357 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1358 | { |
1359 | LIST_HEAD(graveyard); | |
c35038be | 1360 | struct vfsmount *m; |
5528f911 | 1361 | |
5528f911 | 1362 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1363 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1364 | while (!list_empty(&graveyard)) { |
c35038be | 1365 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 AV |
1366 | mnt_expire); |
1367 | touch_mnt_namespace(mnt->mnt_ns); | |
c35038be | 1368 | umount_tree(mnt, 1, umounts); |
bcc5c7d2 AV |
1369 | } |
1370 | } | |
1da177e4 LT |
1371 | } |
1372 | ||
1da177e4 LT |
1373 | /* |
1374 | * Some copy_from_user() implementations do not return the exact number of | |
1375 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1376 | * Note that this function differs from copy_from_user() in that it will oops | |
1377 | * on bad values of `to', rather than returning a short copy. | |
1378 | */ | |
b58fed8b RP |
1379 | static long exact_copy_from_user(void *to, const void __user * from, |
1380 | unsigned long n) | |
1da177e4 LT |
1381 | { |
1382 | char *t = to; | |
1383 | const char __user *f = from; | |
1384 | char c; | |
1385 | ||
1386 | if (!access_ok(VERIFY_READ, from, n)) | |
1387 | return n; | |
1388 | ||
1389 | while (n) { | |
1390 | if (__get_user(c, f)) { | |
1391 | memset(t, 0, n); | |
1392 | break; | |
1393 | } | |
1394 | *t++ = c; | |
1395 | f++; | |
1396 | n--; | |
1397 | } | |
1398 | return n; | |
1399 | } | |
1400 | ||
b58fed8b | 1401 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1402 | { |
1403 | int i; | |
1404 | unsigned long page; | |
1405 | unsigned long size; | |
b58fed8b | 1406 | |
1da177e4 LT |
1407 | *where = 0; |
1408 | if (!data) | |
1409 | return 0; | |
1410 | ||
1411 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1412 | return -ENOMEM; | |
1413 | ||
1414 | /* We only care that *some* data at the address the user | |
1415 | * gave us is valid. Just in case, we'll zero | |
1416 | * the remainder of the page. | |
1417 | */ | |
1418 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1419 | size = TASK_SIZE - (unsigned long)data; | |
1420 | if (size > PAGE_SIZE) | |
1421 | size = PAGE_SIZE; | |
1422 | ||
1423 | i = size - exact_copy_from_user((void *)page, data, size); | |
1424 | if (!i) { | |
b58fed8b | 1425 | free_page(page); |
1da177e4 LT |
1426 | return -EFAULT; |
1427 | } | |
1428 | if (i != PAGE_SIZE) | |
1429 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1430 | *where = page; | |
1431 | return 0; | |
1432 | } | |
1433 | ||
1434 | /* | |
1435 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1436 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1437 | * | |
1438 | * data is a (void *) that can point to any structure up to | |
1439 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1440 | * information (or be NULL). | |
1441 | * | |
1442 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1443 | * When the flags word was introduced its top half was required | |
1444 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1445 | * Therefore, if this magic number is present, it carries no information | |
1446 | * and must be discarded. | |
1447 | */ | |
b58fed8b | 1448 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1449 | unsigned long flags, void *data_page) |
1450 | { | |
1451 | struct nameidata nd; | |
1452 | int retval = 0; | |
1453 | int mnt_flags = 0; | |
1454 | ||
1455 | /* Discard magic */ | |
1456 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1457 | flags &= ~MS_MGC_MSK; | |
1458 | ||
1459 | /* Basic sanity checks */ | |
1460 | ||
1461 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1462 | return -EINVAL; | |
1463 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1464 | return -EINVAL; | |
1465 | ||
1466 | if (data_page) | |
1467 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1468 | ||
1469 | /* Separate the per-mountpoint flags */ | |
1470 | if (flags & MS_NOSUID) | |
1471 | mnt_flags |= MNT_NOSUID; | |
1472 | if (flags & MS_NODEV) | |
1473 | mnt_flags |= MNT_NODEV; | |
1474 | if (flags & MS_NOEXEC) | |
1475 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1476 | if (flags & MS_NOATIME) |
1477 | mnt_flags |= MNT_NOATIME; | |
1478 | if (flags & MS_NODIRATIME) | |
1479 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1480 | if (flags & MS_RELATIME) |
1481 | mnt_flags |= MNT_RELATIME; | |
fc33a7bb CH |
1482 | |
1483 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1484 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1485 | |
1486 | /* ... and get the mountpoint */ | |
1487 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1488 | if (retval) | |
1489 | return retval; | |
1490 | ||
1491 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1492 | if (retval) | |
1493 | goto dput_out; | |
1494 | ||
1495 | if (flags & MS_REMOUNT) | |
1496 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1497 | data_page); | |
1498 | else if (flags & MS_BIND) | |
eee391a6 | 1499 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1500 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1501 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1502 | else if (flags & MS_MOVE) |
1503 | retval = do_move_mount(&nd, dev_name); | |
1504 | else | |
1505 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1506 | dev_name, data_page); | |
1507 | dput_out: | |
1d957f9b | 1508 | path_put(&nd.path); |
1da177e4 LT |
1509 | return retval; |
1510 | } | |
1511 | ||
741a2951 JD |
1512 | /* |
1513 | * Allocate a new namespace structure and populate it with contents | |
1514 | * copied from the namespace of the passed in task structure. | |
1515 | */ | |
e3222c4e | 1516 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1517 | struct fs_struct *fs) |
1da177e4 | 1518 | { |
6b3286ed | 1519 | struct mnt_namespace *new_ns; |
1da177e4 | 1520 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1521 | struct vfsmount *p, *q; |
1522 | ||
6b3286ed | 1523 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1524 | if (!new_ns) |
467e9f4b | 1525 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1526 | |
1527 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1528 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1529 | init_waitqueue_head(&new_ns->poll); |
1530 | new_ns->event = 0; | |
1da177e4 | 1531 | |
390c6843 | 1532 | down_write(&namespace_sem); |
1da177e4 | 1533 | /* First pass: copy the tree topology */ |
6b3286ed | 1534 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1535 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1536 | if (!new_ns->root) { |
390c6843 | 1537 | up_write(&namespace_sem); |
1da177e4 | 1538 | kfree(new_ns); |
467e9f4b | 1539 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1540 | } |
1541 | spin_lock(&vfsmount_lock); | |
1542 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1543 | spin_unlock(&vfsmount_lock); | |
1544 | ||
1545 | /* | |
1546 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1547 | * as belonging to new namespace. We have already acquired a private | |
1548 | * fs_struct, so tsk->fs->lock is not needed. | |
1549 | */ | |
6b3286ed | 1550 | p = mnt_ns->root; |
1da177e4 LT |
1551 | q = new_ns->root; |
1552 | while (p) { | |
6b3286ed | 1553 | q->mnt_ns = new_ns; |
1da177e4 | 1554 | if (fs) { |
6ac08c39 | 1555 | if (p == fs->root.mnt) { |
1da177e4 | 1556 | rootmnt = p; |
6ac08c39 | 1557 | fs->root.mnt = mntget(q); |
1da177e4 | 1558 | } |
6ac08c39 | 1559 | if (p == fs->pwd.mnt) { |
1da177e4 | 1560 | pwdmnt = p; |
6ac08c39 | 1561 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1562 | } |
6ac08c39 | 1563 | if (p == fs->altroot.mnt) { |
1da177e4 | 1564 | altrootmnt = p; |
6ac08c39 | 1565 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1566 | } |
1567 | } | |
6b3286ed | 1568 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1569 | q = next_mnt(q, new_ns->root); |
1570 | } | |
390c6843 | 1571 | up_write(&namespace_sem); |
1da177e4 | 1572 | |
1da177e4 LT |
1573 | if (rootmnt) |
1574 | mntput(rootmnt); | |
1575 | if (pwdmnt) | |
1576 | mntput(pwdmnt); | |
1577 | if (altrootmnt) | |
1578 | mntput(altrootmnt); | |
1579 | ||
741a2951 JD |
1580 | return new_ns; |
1581 | } | |
1582 | ||
213dd266 | 1583 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1584 | struct fs_struct *new_fs) |
741a2951 | 1585 | { |
6b3286ed | 1586 | struct mnt_namespace *new_ns; |
741a2951 | 1587 | |
e3222c4e | 1588 | BUG_ON(!ns); |
6b3286ed | 1589 | get_mnt_ns(ns); |
741a2951 JD |
1590 | |
1591 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1592 | return ns; |
741a2951 | 1593 | |
e3222c4e | 1594 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1595 | |
6b3286ed | 1596 | put_mnt_ns(ns); |
e3222c4e | 1597 | return new_ns; |
1da177e4 LT |
1598 | } |
1599 | ||
1600 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1601 | char __user * type, unsigned long flags, | |
1602 | void __user * data) | |
1603 | { | |
1604 | int retval; | |
1605 | unsigned long data_page; | |
1606 | unsigned long type_page; | |
1607 | unsigned long dev_page; | |
1608 | char *dir_page; | |
1609 | ||
b58fed8b | 1610 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1611 | if (retval < 0) |
1612 | return retval; | |
1613 | ||
1614 | dir_page = getname(dir_name); | |
1615 | retval = PTR_ERR(dir_page); | |
1616 | if (IS_ERR(dir_page)) | |
1617 | goto out1; | |
1618 | ||
b58fed8b | 1619 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1620 | if (retval < 0) |
1621 | goto out2; | |
1622 | ||
b58fed8b | 1623 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1624 | if (retval < 0) |
1625 | goto out3; | |
1626 | ||
1627 | lock_kernel(); | |
b58fed8b RP |
1628 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1629 | flags, (void *)data_page); | |
1da177e4 LT |
1630 | unlock_kernel(); |
1631 | free_page(data_page); | |
1632 | ||
1633 | out3: | |
1634 | free_page(dev_page); | |
1635 | out2: | |
1636 | putname(dir_page); | |
1637 | out1: | |
1638 | free_page(type_page); | |
1639 | return retval; | |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1644 | * It can block. Requires the big lock held. | |
1645 | */ | |
ac748a09 | 1646 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 1647 | { |
6ac08c39 JB |
1648 | struct path old_root; |
1649 | ||
1da177e4 LT |
1650 | write_lock(&fs->lock); |
1651 | old_root = fs->root; | |
ac748a09 JB |
1652 | fs->root = *path; |
1653 | path_get(path); | |
1da177e4 | 1654 | write_unlock(&fs->lock); |
6ac08c39 JB |
1655 | if (old_root.dentry) |
1656 | path_put(&old_root); | |
1da177e4 LT |
1657 | } |
1658 | ||
1659 | /* | |
1660 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1661 | * It can block. Requires the big lock held. | |
1662 | */ | |
ac748a09 | 1663 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 1664 | { |
6ac08c39 | 1665 | struct path old_pwd; |
1da177e4 LT |
1666 | |
1667 | write_lock(&fs->lock); | |
1668 | old_pwd = fs->pwd; | |
ac748a09 JB |
1669 | fs->pwd = *path; |
1670 | path_get(path); | |
1da177e4 LT |
1671 | write_unlock(&fs->lock); |
1672 | ||
6ac08c39 JB |
1673 | if (old_pwd.dentry) |
1674 | path_put(&old_pwd); | |
1da177e4 LT |
1675 | } |
1676 | ||
1a390689 | 1677 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
1678 | { |
1679 | struct task_struct *g, *p; | |
1680 | struct fs_struct *fs; | |
1681 | ||
1682 | read_lock(&tasklist_lock); | |
1683 | do_each_thread(g, p) { | |
1684 | task_lock(p); | |
1685 | fs = p->fs; | |
1686 | if (fs) { | |
1687 | atomic_inc(&fs->count); | |
1688 | task_unlock(p); | |
1a390689 AV |
1689 | if (fs->root.dentry == old_root->dentry |
1690 | && fs->root.mnt == old_root->mnt) | |
1691 | set_fs_root(fs, new_root); | |
1692 | if (fs->pwd.dentry == old_root->dentry | |
1693 | && fs->pwd.mnt == old_root->mnt) | |
1694 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
1695 | put_fs_struct(fs); |
1696 | } else | |
1697 | task_unlock(p); | |
1698 | } while_each_thread(g, p); | |
1699 | read_unlock(&tasklist_lock); | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * pivot_root Semantics: | |
1704 | * Moves the root file system of the current process to the directory put_old, | |
1705 | * makes new_root as the new root file system of the current process, and sets | |
1706 | * root/cwd of all processes which had them on the current root to new_root. | |
1707 | * | |
1708 | * Restrictions: | |
1709 | * The new_root and put_old must be directories, and must not be on the | |
1710 | * same file system as the current process root. The put_old must be | |
1711 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1712 | * pointed to by put_old must yield the same directory as new_root. No other | |
1713 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1714 | * | |
4a0d11fa NB |
1715 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
1716 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1717 | * in this situation. | |
1718 | * | |
1da177e4 LT |
1719 | * Notes: |
1720 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1721 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1722 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1723 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1724 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1725 | * first. | |
1726 | */ | |
b58fed8b RP |
1727 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1728 | const char __user * put_old) | |
1da177e4 LT |
1729 | { |
1730 | struct vfsmount *tmp; | |
1a390689 AV |
1731 | struct nameidata new_nd, old_nd, user_nd; |
1732 | struct path parent_path, root_parent; | |
1da177e4 LT |
1733 | int error; |
1734 | ||
1735 | if (!capable(CAP_SYS_ADMIN)) | |
1736 | return -EPERM; | |
1737 | ||
1738 | lock_kernel(); | |
1739 | ||
b58fed8b RP |
1740 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1741 | &new_nd); | |
1da177e4 LT |
1742 | if (error) |
1743 | goto out0; | |
1744 | error = -EINVAL; | |
4ac91378 | 1745 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
1746 | goto out1; |
1747 | ||
b58fed8b | 1748 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1749 | if (error) |
1750 | goto out1; | |
1751 | ||
1752 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1753 | if (error) { | |
1d957f9b | 1754 | path_put(&old_nd.path); |
1da177e4 LT |
1755 | goto out1; |
1756 | } | |
1757 | ||
1758 | read_lock(¤t->fs->lock); | |
6ac08c39 JB |
1759 | user_nd.path = current->fs->root; |
1760 | path_get(¤t->fs->root); | |
1da177e4 | 1761 | read_unlock(¤t->fs->lock); |
390c6843 | 1762 | down_write(&namespace_sem); |
4ac91378 | 1763 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 1764 | error = -EINVAL; |
4ac91378 JB |
1765 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
1766 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
1767 | IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) | |
21444403 | 1768 | goto out2; |
4ac91378 | 1769 | if (!check_mnt(user_nd.path.mnt)) |
1da177e4 LT |
1770 | goto out2; |
1771 | error = -ENOENT; | |
4ac91378 | 1772 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 1773 | goto out2; |
4ac91378 | 1774 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 1775 | goto out2; |
4ac91378 | 1776 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
1777 | goto out2; |
1778 | error = -EBUSY; | |
4ac91378 JB |
1779 | if (new_nd.path.mnt == user_nd.path.mnt || |
1780 | old_nd.path.mnt == user_nd.path.mnt) | |
1da177e4 LT |
1781 | goto out2; /* loop, on the same file system */ |
1782 | error = -EINVAL; | |
4ac91378 | 1783 | if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) |
1da177e4 | 1784 | goto out2; /* not a mountpoint */ |
4ac91378 | 1785 | if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) |
0bb6fcc1 | 1786 | goto out2; /* not attached */ |
4ac91378 | 1787 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 1788 | goto out2; /* not a mountpoint */ |
4ac91378 | 1789 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 1790 | goto out2; /* not attached */ |
4ac91378 JB |
1791 | /* make sure we can reach put_old from new_root */ |
1792 | tmp = old_nd.path.mnt; | |
1da177e4 | 1793 | spin_lock(&vfsmount_lock); |
4ac91378 | 1794 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
1795 | for (;;) { |
1796 | if (tmp->mnt_parent == tmp) | |
1797 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 1798 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
1799 | break; |
1800 | tmp = tmp->mnt_parent; | |
1801 | } | |
4ac91378 | 1802 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 1803 | goto out3; |
4ac91378 | 1804 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 1805 | goto out3; |
1a390689 | 1806 | detach_mnt(new_nd.path.mnt, &parent_path); |
4ac91378 JB |
1807 | detach_mnt(user_nd.path.mnt, &root_parent); |
1808 | /* mount old root on put_old */ | |
1a390689 | 1809 | attach_mnt(user_nd.path.mnt, &old_nd.path); |
4ac91378 JB |
1810 | /* mount new_root on / */ |
1811 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 1812 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 1813 | spin_unlock(&vfsmount_lock); |
1a390689 | 1814 | chroot_fs_refs(&user_nd.path, &new_nd.path); |
1da177e4 LT |
1815 | security_sb_post_pivotroot(&user_nd, &new_nd); |
1816 | error = 0; | |
1a390689 AV |
1817 | path_put(&root_parent); |
1818 | path_put(&parent_path); | |
1da177e4 | 1819 | out2: |
4ac91378 | 1820 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 1821 | up_write(&namespace_sem); |
1d957f9b JB |
1822 | path_put(&user_nd.path); |
1823 | path_put(&old_nd.path); | |
1da177e4 | 1824 | out1: |
1d957f9b | 1825 | path_put(&new_nd.path); |
1da177e4 LT |
1826 | out0: |
1827 | unlock_kernel(); | |
1828 | return error; | |
1829 | out3: | |
1830 | spin_unlock(&vfsmount_lock); | |
1831 | goto out2; | |
1832 | } | |
1833 | ||
1834 | static void __init init_mount_tree(void) | |
1835 | { | |
1836 | struct vfsmount *mnt; | |
6b3286ed | 1837 | struct mnt_namespace *ns; |
ac748a09 | 1838 | struct path root; |
1da177e4 LT |
1839 | |
1840 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1841 | if (IS_ERR(mnt)) | |
1842 | panic("Can't create rootfs"); | |
6b3286ed KK |
1843 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
1844 | if (!ns) | |
1da177e4 | 1845 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
1846 | atomic_set(&ns->count, 1); |
1847 | INIT_LIST_HEAD(&ns->list); | |
1848 | init_waitqueue_head(&ns->poll); | |
1849 | ns->event = 0; | |
1850 | list_add(&mnt->mnt_list, &ns->list); | |
1851 | ns->root = mnt; | |
1852 | mnt->mnt_ns = ns; | |
1853 | ||
1854 | init_task.nsproxy->mnt_ns = ns; | |
1855 | get_mnt_ns(ns); | |
1856 | ||
ac748a09 JB |
1857 | root.mnt = ns->root; |
1858 | root.dentry = ns->root->mnt_root; | |
1859 | ||
1860 | set_fs_pwd(current->fs, &root); | |
1861 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
1862 | } |
1863 | ||
74bf17cf | 1864 | void __init mnt_init(void) |
1da177e4 | 1865 | { |
13f14b4d | 1866 | unsigned u; |
15a67dd8 | 1867 | int err; |
1da177e4 | 1868 | |
390c6843 RP |
1869 | init_rwsem(&namespace_sem); |
1870 | ||
1da177e4 | 1871 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 1872 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 1873 | |
b58fed8b | 1874 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1875 | |
1876 | if (!mount_hashtable) | |
1877 | panic("Failed to allocate mount hash table\n"); | |
1878 | ||
13f14b4d ED |
1879 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
1880 | ||
1881 | for (u = 0; u < HASH_SIZE; u++) | |
1882 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 1883 | |
15a67dd8 RD |
1884 | err = sysfs_init(); |
1885 | if (err) | |
1886 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
1887 | __FUNCTION__, err); | |
00d26666 GKH |
1888 | fs_kobj = kobject_create_and_add("fs", NULL); |
1889 | if (!fs_kobj) | |
1890 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
1891 | init_rootfs(); |
1892 | init_mount_tree(); | |
1893 | } | |
1894 | ||
6b3286ed | 1895 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 1896 | { |
6b3286ed | 1897 | struct vfsmount *root = ns->root; |
70fbcdf4 | 1898 | LIST_HEAD(umount_list); |
6b3286ed | 1899 | ns->root = NULL; |
1ce88cf4 | 1900 | spin_unlock(&vfsmount_lock); |
390c6843 | 1901 | down_write(&namespace_sem); |
1da177e4 | 1902 | spin_lock(&vfsmount_lock); |
a05964f3 | 1903 | umount_tree(root, 0, &umount_list); |
1da177e4 | 1904 | spin_unlock(&vfsmount_lock); |
390c6843 | 1905 | up_write(&namespace_sem); |
70fbcdf4 | 1906 | release_mounts(&umount_list); |
6b3286ed | 1907 | kfree(ns); |
1da177e4 | 1908 | } |