]>
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 | ||
454e2398 DH |
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 | ||
1da177e4 LT |
92 | void free_vfsmnt(struct vfsmount *mnt) |
93 | { | |
94 | kfree(mnt->mnt_devname); | |
95 | kmem_cache_free(mnt_cache, mnt); | |
96 | } | |
97 | ||
98 | /* | |
a05964f3 RP |
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. | |
1da177e4 | 101 | */ |
a05964f3 RP |
102 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
103 | int dir) | |
1da177e4 | 104 | { |
b58fed8b RP |
105 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
106 | struct list_head *tmp = head; | |
1da177e4 LT |
107 | struct vfsmount *p, *found = NULL; |
108 | ||
1da177e4 | 109 | for (;;) { |
a05964f3 | 110 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
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) { | |
a05964f3 | 116 | found = p; |
1da177e4 LT |
117 | break; |
118 | } | |
119 | } | |
1da177e4 LT |
120 | return found; |
121 | } | |
122 | ||
a05964f3 RP |
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 | ||
1da177e4 LT |
137 | static inline int check_mnt(struct vfsmount *mnt) |
138 | { | |
6b3286ed | 139 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
140 | } |
141 | ||
6b3286ed | 142 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
143 | { |
144 | if (ns) { | |
145 | ns->event = ++event; | |
146 | wake_up_interruptible(&ns->poll); | |
147 | } | |
148 | } | |
149 | ||
6b3286ed | 150 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
151 | { |
152 | if (ns && ns->event != event) { | |
153 | ns->event = event; | |
154 | wake_up_interruptible(&ns->poll); | |
155 | } | |
156 | } | |
157 | ||
1a390689 | 158 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 159 | { |
1a390689 AV |
160 | old_path->dentry = mnt->mnt_mountpoint; |
161 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
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); | |
1a390689 | 166 | old_path->dentry->d_mounted--; |
1da177e4 LT |
167 | } |
168 | ||
b90fa9ae RP |
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 | ||
1a390689 | 177 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 178 | { |
1a390689 | 179 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 180 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
181 | hash(path->mnt, path->dentry)); |
182 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
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); | |
6b3286ed | 193 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
194 | |
195 | BUG_ON(parent == mnt); | |
196 | ||
197 | list_add_tail(&head, &mnt->mnt_list); | |
198 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 199 | m->mnt_ns = n; |
b90fa9ae RP |
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); | |
6b3286ed | 205 | touch_mnt_namespace(n); |
1da177e4 LT |
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 | ||
9676f0c6 RP |
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 | ||
36341f64 RP |
234 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
235 | int flag) | |
1da177e4 LT |
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; | |
b90fa9ae | 247 | |
5afe0022 RP |
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); | |
8aec0809 | 252 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
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 | } | |
b90fa9ae RP |
259 | if (flag & CL_MAKE_SHARED) |
260 | set_mnt_shared(mnt); | |
1da177e4 LT |
261 | |
262 | /* stick the duplicate mount on the same expiry list | |
263 | * as the original if that was on one */ | |
36341f64 RP |
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 | } | |
1da177e4 LT |
270 | } |
271 | return mnt; | |
272 | } | |
273 | ||
7b7b1ace | 274 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
275 | { |
276 | struct super_block *sb = mnt->mnt_sb; | |
277 | dput(mnt->mnt_root); | |
278 | free_vfsmnt(mnt); | |
279 | deactivate_super(sb); | |
280 | } | |
281 | ||
7b7b1ace AV |
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); | |
1da177e4 | 322 | |
b3b304a2 MS |
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 | ||
1da177e4 LT |
367 | /* iterator */ |
368 | static void *m_start(struct seq_file *m, loff_t *pos) | |
369 | { | |
6b3286ed | 370 | struct mnt_namespace *n = m->private; |
1da177e4 | 371 | |
390c6843 | 372 | down_read(&namespace_sem); |
b0765fb8 | 373 | return seq_list_start(&n->list, *pos); |
1da177e4 LT |
374 | } |
375 | ||
376 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
377 | { | |
6b3286ed | 378 | struct mnt_namespace *n = m->private; |
b0765fb8 PE |
379 | |
380 | return seq_list_next(v, &n->list, pos); | |
1da177e4 LT |
381 | } |
382 | ||
383 | static void m_stop(struct seq_file *m, void *v) | |
384 | { | |
390c6843 | 385 | up_read(&namespace_sem); |
1da177e4 LT |
386 | } |
387 | ||
1da177e4 LT |
388 | static int show_vfsmnt(struct seq_file *m, void *v) |
389 | { | |
b0765fb8 | 390 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
1da177e4 LT |
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" }, | |
1da177e4 LT |
399 | { 0, NULL } |
400 | }; | |
401 | static struct proc_fs_info mnt_info[] = { | |
402 | { MNT_NOSUID, ",nosuid" }, | |
403 | { MNT_NODEV, ",nodev" }, | |
404 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
405 | { MNT_NOATIME, ",noatime" }, |
406 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 407 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
408 | { 0, NULL } |
409 | }; | |
410 | struct proc_fs_info *fs_infop; | |
c32c2f63 | 411 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
412 | |
413 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
414 | seq_putc(m, ' '); | |
c32c2f63 | 415 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 LT |
416 | seq_putc(m, ' '); |
417 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
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 | } | |
1da177e4 LT |
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 | ||
b4629fe2 CL |
444 | static int show_vfsstat(struct seq_file *m, void *v) |
445 | { | |
b0765fb8 | 446 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 447 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
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 "); | |
c32c2f63 | 459 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
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 | ||
1da177e4 LT |
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 | { | |
36341f64 RP |
493 | int actual_refs = 0; |
494 | int minimum_refs = 0; | |
495 | struct vfsmount *p; | |
1da177e4 LT |
496 | |
497 | spin_lock(&vfsmount_lock); | |
36341f64 | 498 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
499 | actual_refs += atomic_read(&p->mnt_count); |
500 | minimum_refs += 2; | |
1da177e4 LT |
501 | } |
502 | spin_unlock(&vfsmount_lock); | |
503 | ||
504 | if (actual_refs > minimum_refs) | |
e3474a8e | 505 | return 0; |
1da177e4 | 506 | |
e3474a8e | 507 | return 1; |
1da177e4 LT |
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 | { | |
e3474a8e | 527 | int ret = 1; |
a05964f3 RP |
528 | spin_lock(&vfsmount_lock); |
529 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 530 | ret = 0; |
a05964f3 RP |
531 | spin_unlock(&vfsmount_lock); |
532 | return ret; | |
1da177e4 LT |
533 | } |
534 | ||
535 | EXPORT_SYMBOL(may_umount); | |
536 | ||
b90fa9ae | 537 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
538 | { |
539 | struct vfsmount *mnt; | |
bf066c7d | 540 | while (!list_empty(head)) { |
b5e61818 | 541 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
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; | |
7c4b93d8 | 551 | m->mnt_ghosts--; |
70fbcdf4 RP |
552 | spin_unlock(&vfsmount_lock); |
553 | dput(dentry); | |
554 | mntput(m); | |
555 | } | |
556 | mntput(mnt); | |
557 | } | |
558 | } | |
559 | ||
a05964f3 | 560 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
561 | { |
562 | struct vfsmount *p; | |
1da177e4 | 563 | |
1bfba4e8 AM |
564 | for (p = mnt; p; p = next_mnt(p, mnt)) |
565 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 566 | |
a05964f3 RP |
567 | if (propagate) |
568 | propagate_umount(kill); | |
569 | ||
70fbcdf4 RP |
570 | list_for_each_entry(p, kill, mnt_hash) { |
571 | list_del_init(&p->mnt_expire); | |
572 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
573 | __touch_mnt_namespace(p->mnt_ns); |
574 | p->mnt_ns = NULL; | |
70fbcdf4 | 575 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
576 | if (p->mnt_parent != p) { |
577 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 578 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 579 | } |
a05964f3 | 580 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
581 | } |
582 | } | |
583 | ||
584 | static int do_umount(struct vfsmount *mnt, int flags) | |
585 | { | |
b58fed8b | 586 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 587 | int retval; |
70fbcdf4 | 588 | LIST_HEAD(umount_list); |
1da177e4 LT |
589 | |
590 | retval = security_sb_umount(mnt, flags); | |
591 | if (retval) | |
592 | return retval; | |
593 | ||
594 | /* | |
595 | * Allow userspace to request a mountpoint be expired rather than | |
596 | * unmounting unconditionally. Unmount only happens if: | |
597 | * (1) the mark is already set (the mark is cleared by mntput()) | |
598 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
599 | */ | |
600 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 601 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
602 | flags & (MNT_FORCE | MNT_DETACH)) |
603 | return -EINVAL; | |
604 | ||
605 | if (atomic_read(&mnt->mnt_count) != 2) | |
606 | return -EBUSY; | |
607 | ||
608 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
609 | return -EAGAIN; | |
610 | } | |
611 | ||
612 | /* | |
613 | * If we may have to abort operations to get out of this | |
614 | * mount, and they will themselves hold resources we must | |
615 | * allow the fs to do things. In the Unix tradition of | |
616 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
617 | * might fail to complete on the first run through as other tasks | |
618 | * must return, and the like. Thats for the mount program to worry | |
619 | * about for the moment. | |
620 | */ | |
621 | ||
622 | lock_kernel(); | |
8b512d9a TM |
623 | if (sb->s_op->umount_begin) |
624 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
625 | unlock_kernel(); |
626 | ||
627 | /* | |
628 | * No sense to grab the lock for this test, but test itself looks | |
629 | * somewhat bogus. Suggestions for better replacement? | |
630 | * Ho-hum... In principle, we might treat that as umount + switch | |
631 | * to rootfs. GC would eventually take care of the old vfsmount. | |
632 | * Actually it makes sense, especially if rootfs would contain a | |
633 | * /reboot - static binary that would close all descriptors and | |
634 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
635 | */ | |
6ac08c39 | 636 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
637 | /* |
638 | * Special case for "unmounting" root ... | |
639 | * we just try to remount it readonly. | |
640 | */ | |
641 | down_write(&sb->s_umount); | |
642 | if (!(sb->s_flags & MS_RDONLY)) { | |
643 | lock_kernel(); | |
644 | DQUOT_OFF(sb); | |
645 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
646 | unlock_kernel(); | |
647 | } | |
648 | up_write(&sb->s_umount); | |
649 | return retval; | |
650 | } | |
651 | ||
390c6843 | 652 | down_write(&namespace_sem); |
1da177e4 | 653 | spin_lock(&vfsmount_lock); |
5addc5dd | 654 | event++; |
1da177e4 | 655 | |
1da177e4 | 656 | retval = -EBUSY; |
a05964f3 | 657 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 658 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 659 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
660 | retval = 0; |
661 | } | |
662 | spin_unlock(&vfsmount_lock); | |
663 | if (retval) | |
664 | security_sb_umount_busy(mnt); | |
390c6843 | 665 | up_write(&namespace_sem); |
70fbcdf4 | 666 | release_mounts(&umount_list); |
1da177e4 LT |
667 | return retval; |
668 | } | |
669 | ||
670 | /* | |
671 | * Now umount can handle mount points as well as block devices. | |
672 | * This is important for filesystems which use unnamed block devices. | |
673 | * | |
674 | * We now support a flag for forced unmount like the other 'big iron' | |
675 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
676 | */ | |
677 | ||
678 | asmlinkage long sys_umount(char __user * name, int flags) | |
679 | { | |
680 | struct nameidata nd; | |
681 | int retval; | |
682 | ||
683 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
684 | if (retval) | |
685 | goto out; | |
686 | retval = -EINVAL; | |
4ac91378 | 687 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 688 | goto dput_and_out; |
4ac91378 | 689 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
690 | goto dput_and_out; |
691 | ||
692 | retval = -EPERM; | |
693 | if (!capable(CAP_SYS_ADMIN)) | |
694 | goto dput_and_out; | |
695 | ||
4ac91378 | 696 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 697 | dput_and_out: |
429731b1 | 698 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
699 | dput(nd.path.dentry); |
700 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
701 | out: |
702 | return retval; | |
703 | } | |
704 | ||
705 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
706 | ||
707 | /* | |
b58fed8b | 708 | * The 2.0 compatible umount. No flags. |
1da177e4 | 709 | */ |
1da177e4 LT |
710 | asmlinkage long sys_oldumount(char __user * name) |
711 | { | |
b58fed8b | 712 | return sys_umount(name, 0); |
1da177e4 LT |
713 | } |
714 | ||
715 | #endif | |
716 | ||
717 | static int mount_is_safe(struct nameidata *nd) | |
718 | { | |
719 | if (capable(CAP_SYS_ADMIN)) | |
720 | return 0; | |
721 | return -EPERM; | |
722 | #ifdef notyet | |
4ac91378 | 723 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 724 | return -EPERM; |
4ac91378 JB |
725 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
726 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
727 | return -EPERM; |
728 | } | |
e4543edd | 729 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
730 | return -EPERM; |
731 | return 0; | |
732 | #endif | |
733 | } | |
734 | ||
b58fed8b | 735 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
736 | { |
737 | while (1) { | |
738 | if (d == dentry) | |
739 | return 1; | |
740 | if (d == NULL || d == d->d_parent) | |
741 | return 0; | |
742 | d = d->d_parent; | |
743 | } | |
744 | } | |
745 | ||
b90fa9ae | 746 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 747 | int flag) |
1da177e4 LT |
748 | { |
749 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 750 | struct path path; |
1da177e4 | 751 | |
9676f0c6 RP |
752 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
753 | return NULL; | |
754 | ||
36341f64 | 755 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
756 | if (!q) |
757 | goto Enomem; | |
758 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
759 | ||
760 | p = mnt; | |
fdadd65f | 761 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
762 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
763 | continue; | |
764 | ||
765 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
766 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
767 | s = skip_mnt_tree(s); | |
768 | continue; | |
769 | } | |
1da177e4 LT |
770 | while (p != s->mnt_parent) { |
771 | p = p->mnt_parent; | |
772 | q = q->mnt_parent; | |
773 | } | |
774 | p = s; | |
1a390689 AV |
775 | path.mnt = q; |
776 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 777 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
778 | if (!q) |
779 | goto Enomem; | |
780 | spin_lock(&vfsmount_lock); | |
781 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 782 | attach_mnt(q, &path); |
1da177e4 LT |
783 | spin_unlock(&vfsmount_lock); |
784 | } | |
785 | } | |
786 | return res; | |
b58fed8b | 787 | Enomem: |
1da177e4 | 788 | if (res) { |
70fbcdf4 | 789 | LIST_HEAD(umount_list); |
1da177e4 | 790 | spin_lock(&vfsmount_lock); |
a05964f3 | 791 | umount_tree(res, 0, &umount_list); |
1da177e4 | 792 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 793 | release_mounts(&umount_list); |
1da177e4 LT |
794 | } |
795 | return NULL; | |
796 | } | |
797 | ||
8aec0809 AV |
798 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
799 | { | |
800 | struct vfsmount *tree; | |
801 | down_read(&namespace_sem); | |
802 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); | |
803 | up_read(&namespace_sem); | |
804 | return tree; | |
805 | } | |
806 | ||
807 | void drop_collected_mounts(struct vfsmount *mnt) | |
808 | { | |
809 | LIST_HEAD(umount_list); | |
810 | down_read(&namespace_sem); | |
811 | spin_lock(&vfsmount_lock); | |
812 | umount_tree(mnt, 0, &umount_list); | |
813 | spin_unlock(&vfsmount_lock); | |
814 | up_read(&namespace_sem); | |
815 | release_mounts(&umount_list); | |
816 | } | |
817 | ||
b90fa9ae RP |
818 | /* |
819 | * @source_mnt : mount tree to be attached | |
21444403 RP |
820 | * @nd : place the mount tree @source_mnt is attached |
821 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
822 | * store the parent mount and mountpoint dentry. | |
823 | * (done when source_mnt is moved) | |
b90fa9ae RP |
824 | * |
825 | * NOTE: in the table below explains the semantics when a source mount | |
826 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
827 | * --------------------------------------------------------------------------- |
828 | * | BIND MOUNT OPERATION | | |
829 | * |************************************************************************** | |
830 | * | source-->| shared | private | slave | unbindable | | |
831 | * | dest | | | | | | |
832 | * | | | | | | | | |
833 | * | v | | | | | | |
834 | * |************************************************************************** | |
835 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
836 | * | | | | | | | |
837 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
838 | * *************************************************************************** | |
b90fa9ae RP |
839 | * A bind operation clones the source mount and mounts the clone on the |
840 | * destination mount. | |
841 | * | |
842 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
843 | * tree of the destination mount and the cloned mount is added to | |
844 | * the peer group of the source mount. | |
845 | * (+) the cloned mount is created under the destination mount and is marked | |
846 | * as shared. The cloned mount is added to the peer group of the source | |
847 | * mount. | |
5afe0022 RP |
848 | * (+++) the mount is propagated to all the mounts in the propagation tree |
849 | * of the destination mount and the cloned mount is made slave | |
850 | * of the same master as that of the source mount. The cloned mount | |
851 | * is marked as 'shared and slave'. | |
852 | * (*) the cloned mount is made a slave of the same master as that of the | |
853 | * source mount. | |
854 | * | |
9676f0c6 RP |
855 | * --------------------------------------------------------------------------- |
856 | * | MOVE MOUNT OPERATION | | |
857 | * |************************************************************************** | |
858 | * | source-->| shared | private | slave | unbindable | | |
859 | * | dest | | | | | | |
860 | * | | | | | | | | |
861 | * | v | | | | | | |
862 | * |************************************************************************** | |
863 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
864 | * | | | | | | | |
865 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
866 | * *************************************************************************** | |
5afe0022 RP |
867 | * |
868 | * (+) the mount is moved to the destination. And is then propagated to | |
869 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 870 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
871 | * (+++) the mount is moved to the destination and is then propagated to |
872 | * all the mounts belonging to the destination mount's propagation tree. | |
873 | * the mount is marked as 'shared and slave'. | |
874 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
875 | * |
876 | * if the source mount is a tree, the operations explained above is | |
877 | * applied to each mount in the tree. | |
878 | * Must be called without spinlocks held, since this function can sleep | |
879 | * in allocations. | |
880 | */ | |
881 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 882 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
883 | { |
884 | LIST_HEAD(tree_list); | |
1a390689 AV |
885 | struct vfsmount *dest_mnt = path->mnt; |
886 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae RP |
887 | struct vfsmount *child, *p; |
888 | ||
889 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
890 | return -EINVAL; | |
891 | ||
892 | if (IS_MNT_SHARED(dest_mnt)) { | |
893 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
894 | set_mnt_shared(p); | |
895 | } | |
896 | ||
897 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
898 | if (parent_path) { |
899 | detach_mnt(source_mnt, parent_path); | |
900 | attach_mnt(source_mnt, path); | |
6b3286ed | 901 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
902 | } else { |
903 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
904 | commit_tree(source_mnt); | |
905 | } | |
b90fa9ae RP |
906 | |
907 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
908 | list_del_init(&child->mnt_hash); | |
909 | commit_tree(child); | |
910 | } | |
911 | spin_unlock(&vfsmount_lock); | |
912 | return 0; | |
913 | } | |
914 | ||
1da177e4 LT |
915 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
916 | { | |
917 | int err; | |
918 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
919 | return -EINVAL; | |
920 | ||
4ac91378 | 921 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1da177e4 LT |
922 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
923 | return -ENOTDIR; | |
924 | ||
925 | err = -ENOENT; | |
4ac91378 JB |
926 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
927 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
928 | goto out_unlock; |
929 | ||
930 | err = security_sb_check_sb(mnt, nd); | |
931 | if (err) | |
932 | goto out_unlock; | |
933 | ||
934 | err = -ENOENT; | |
4ac91378 | 935 | if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) |
1a390689 | 936 | err = attach_recursive_mnt(mnt, &nd->path, NULL); |
1da177e4 | 937 | out_unlock: |
4ac91378 | 938 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 LT |
939 | if (!err) |
940 | security_sb_post_addmount(mnt, nd); | |
941 | return err; | |
942 | } | |
943 | ||
07b20889 RP |
944 | /* |
945 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 946 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 947 | */ |
2dafe1c4 | 948 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 949 | { |
4ac91378 | 950 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
951 | int recurse = flag & MS_REC; |
952 | int type = flag & ~MS_REC; | |
953 | ||
ee6f9582 MS |
954 | if (!capable(CAP_SYS_ADMIN)) |
955 | return -EPERM; | |
956 | ||
4ac91378 | 957 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
958 | return -EINVAL; |
959 | ||
960 | down_write(&namespace_sem); | |
961 | spin_lock(&vfsmount_lock); | |
962 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
963 | change_mnt_propagation(m, type); | |
964 | spin_unlock(&vfsmount_lock); | |
965 | up_write(&namespace_sem); | |
966 | return 0; | |
967 | } | |
968 | ||
1da177e4 LT |
969 | /* |
970 | * do loopback mount. | |
2dafe1c4 | 971 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 972 | */ |
2dafe1c4 ES |
973 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
974 | int recurse) | |
1da177e4 LT |
975 | { |
976 | struct nameidata old_nd; | |
977 | struct vfsmount *mnt = NULL; | |
978 | int err = mount_is_safe(nd); | |
979 | if (err) | |
980 | return err; | |
981 | if (!old_name || !*old_name) | |
982 | return -EINVAL; | |
983 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
984 | if (err) | |
985 | return err; | |
986 | ||
390c6843 | 987 | down_write(&namespace_sem); |
1da177e4 | 988 | err = -EINVAL; |
4ac91378 JB |
989 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
990 | goto out; | |
9676f0c6 | 991 | |
4ac91378 | 992 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 993 | goto out; |
1da177e4 | 994 | |
ccd48bc7 AV |
995 | err = -ENOMEM; |
996 | if (recurse) | |
4ac91378 | 997 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 998 | else |
4ac91378 | 999 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1000 | |
1001 | if (!mnt) | |
1002 | goto out; | |
1003 | ||
ccd48bc7 AV |
1004 | err = graft_tree(mnt, nd); |
1005 | if (err) { | |
70fbcdf4 | 1006 | LIST_HEAD(umount_list); |
1da177e4 | 1007 | spin_lock(&vfsmount_lock); |
a05964f3 | 1008 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1009 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1010 | release_mounts(&umount_list); |
5b83d2c5 | 1011 | } |
1da177e4 | 1012 | |
ccd48bc7 | 1013 | out: |
390c6843 | 1014 | up_write(&namespace_sem); |
1d957f9b | 1015 | path_put(&old_nd.path); |
1da177e4 LT |
1016 | return err; |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * change filesystem flags. dir should be a physical root of filesystem. | |
1021 | * If you've mounted a non-root directory somewhere and want to do remount | |
1022 | * on it - tough luck. | |
2dafe1c4 | 1023 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1024 | */ |
2dafe1c4 | 1025 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1026 | void *data) |
1027 | { | |
1028 | int err; | |
4ac91378 | 1029 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1030 | |
1031 | if (!capable(CAP_SYS_ADMIN)) | |
1032 | return -EPERM; | |
1033 | ||
4ac91378 | 1034 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1035 | return -EINVAL; |
1036 | ||
4ac91378 | 1037 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1038 | return -EINVAL; |
1039 | ||
1040 | down_write(&sb->s_umount); | |
1041 | err = do_remount_sb(sb, flags, data, 0); | |
1042 | if (!err) | |
4ac91378 | 1043 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1044 | up_write(&sb->s_umount); |
1045 | if (!err) | |
4ac91378 | 1046 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1047 | return err; |
1048 | } | |
1049 | ||
9676f0c6 RP |
1050 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1051 | { | |
1052 | struct vfsmount *p; | |
1053 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1054 | if (IS_MNT_UNBINDABLE(p)) | |
1055 | return 1; | |
1056 | } | |
1057 | return 0; | |
1058 | } | |
1059 | ||
2dafe1c4 ES |
1060 | /* |
1061 | * noinline this do_mount helper to save do_mount stack space. | |
1062 | */ | |
1063 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1064 | { |
1a390689 AV |
1065 | struct nameidata old_nd; |
1066 | struct path parent_path; | |
1da177e4 LT |
1067 | struct vfsmount *p; |
1068 | int err = 0; | |
1069 | if (!capable(CAP_SYS_ADMIN)) | |
1070 | return -EPERM; | |
1071 | if (!old_name || !*old_name) | |
1072 | return -EINVAL; | |
1073 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1074 | if (err) | |
1075 | return err; | |
1076 | ||
390c6843 | 1077 | down_write(&namespace_sem); |
4ac91378 JB |
1078 | while (d_mountpoint(nd->path.dentry) && |
1079 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1080 | ; |
1081 | err = -EINVAL; | |
4ac91378 | 1082 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1083 | goto out; |
1084 | ||
1085 | err = -ENOENT; | |
4ac91378 JB |
1086 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1087 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1088 | goto out1; |
1089 | ||
4ac91378 | 1090 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1091 | goto out1; |
1da177e4 LT |
1092 | |
1093 | err = -EINVAL; | |
4ac91378 | 1094 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1095 | goto out1; |
1da177e4 | 1096 | |
4ac91378 | 1097 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1098 | goto out1; |
1da177e4 | 1099 | |
4ac91378 JB |
1100 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1101 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1102 | goto out1; |
1103 | /* | |
1104 | * Don't move a mount residing in a shared parent. | |
1105 | */ | |
4ac91378 JB |
1106 | if (old_nd.path.mnt->mnt_parent && |
1107 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1108 | goto out1; |
9676f0c6 RP |
1109 | /* |
1110 | * Don't move a mount tree containing unbindable mounts to a destination | |
1111 | * mount which is shared. | |
1112 | */ | |
4ac91378 JB |
1113 | if (IS_MNT_SHARED(nd->path.mnt) && |
1114 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1115 | goto out1; |
1da177e4 | 1116 | err = -ELOOP; |
4ac91378 JB |
1117 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1118 | if (p == old_nd.path.mnt) | |
21444403 | 1119 | goto out1; |
1da177e4 | 1120 | |
1a390689 | 1121 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1122 | if (err) |
21444403 | 1123 | goto out1; |
1da177e4 | 1124 | |
21444403 | 1125 | spin_lock(&vfsmount_lock); |
1da177e4 LT |
1126 | /* if the mount is moved, it should no longer be expire |
1127 | * automatically */ | |
4ac91378 | 1128 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 LT |
1129 | spin_unlock(&vfsmount_lock); |
1130 | out1: | |
4ac91378 | 1131 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1132 | out: |
390c6843 | 1133 | up_write(&namespace_sem); |
1da177e4 | 1134 | if (!err) |
1a390689 | 1135 | path_put(&parent_path); |
1d957f9b | 1136 | path_put(&old_nd.path); |
1da177e4 LT |
1137 | return err; |
1138 | } | |
1139 | ||
1140 | /* | |
1141 | * create a new mount for userspace and request it to be added into the | |
1142 | * namespace's tree | |
2dafe1c4 | 1143 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1144 | */ |
2dafe1c4 | 1145 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1146 | int mnt_flags, char *name, void *data) |
1147 | { | |
1148 | struct vfsmount *mnt; | |
1149 | ||
1150 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1151 | return -EINVAL; | |
1152 | ||
1153 | /* we need capabilities... */ | |
1154 | if (!capable(CAP_SYS_ADMIN)) | |
1155 | return -EPERM; | |
1156 | ||
1157 | mnt = do_kern_mount(type, flags, name, data); | |
1158 | if (IS_ERR(mnt)) | |
1159 | return PTR_ERR(mnt); | |
1160 | ||
1161 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * add a mount into a namespace's mount tree | |
1166 | * - provide the option of adding the new mount to an expiration list | |
1167 | */ | |
1168 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1169 | int mnt_flags, struct list_head *fslist) | |
1170 | { | |
1171 | int err; | |
1172 | ||
390c6843 | 1173 | down_write(&namespace_sem); |
1da177e4 | 1174 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1175 | while (d_mountpoint(nd->path.dentry) && |
1176 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1177 | ; |
1178 | err = -EINVAL; | |
4ac91378 | 1179 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1180 | goto unlock; |
1181 | ||
1182 | /* Refuse the same filesystem on the same mount point */ | |
1183 | err = -EBUSY; | |
4ac91378 JB |
1184 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1185 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1186 | goto unlock; |
1187 | ||
1188 | err = -EINVAL; | |
1189 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1190 | goto unlock; | |
1191 | ||
1192 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1193 | if ((err = graft_tree(newmnt, nd))) |
1194 | goto unlock; | |
1da177e4 | 1195 | |
5b83d2c5 | 1196 | if (fslist) { |
1da177e4 LT |
1197 | /* add to the specified expiration list */ |
1198 | spin_lock(&vfsmount_lock); | |
55e700b9 | 1199 | list_add_tail(&newmnt->mnt_expire, fslist); |
1da177e4 LT |
1200 | spin_unlock(&vfsmount_lock); |
1201 | } | |
390c6843 | 1202 | up_write(&namespace_sem); |
5b83d2c5 | 1203 | return 0; |
1da177e4 LT |
1204 | |
1205 | unlock: | |
390c6843 | 1206 | up_write(&namespace_sem); |
1da177e4 LT |
1207 | mntput(newmnt); |
1208 | return err; | |
1209 | } | |
1210 | ||
1211 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1212 | ||
70fbcdf4 RP |
1213 | static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, |
1214 | struct list_head *umounts) | |
24ca2af1 MS |
1215 | { |
1216 | spin_lock(&vfsmount_lock); | |
1217 | ||
ed42c879 MS |
1218 | /* |
1219 | * Check if mount is still attached, if not, let whoever holds it deal | |
1220 | * with the sucker | |
1221 | */ | |
1222 | if (mnt->mnt_parent == mnt) { | |
1223 | spin_unlock(&vfsmount_lock); | |
1224 | return; | |
1225 | } | |
1226 | ||
24ca2af1 MS |
1227 | /* |
1228 | * Check that it is still dead: the count should now be 2 - as | |
1229 | * contributed by the vfsmount parent and the mntget above | |
1230 | */ | |
a05964f3 | 1231 | if (!propagate_mount_busy(mnt, 2)) { |
24ca2af1 | 1232 | /* delete from the namespace */ |
6b3286ed | 1233 | touch_mnt_namespace(mnt->mnt_ns); |
24ca2af1 | 1234 | list_del_init(&mnt->mnt_list); |
6b3286ed | 1235 | mnt->mnt_ns = NULL; |
a05964f3 | 1236 | umount_tree(mnt, 1, umounts); |
24ca2af1 | 1237 | spin_unlock(&vfsmount_lock); |
24ca2af1 MS |
1238 | } else { |
1239 | /* | |
1240 | * Someone brought it back to life whilst we didn't have any | |
1241 | * locks held so return it to the expiration list | |
1242 | */ | |
55e700b9 | 1243 | list_add_tail(&mnt->mnt_expire, mounts); |
24ca2af1 MS |
1244 | spin_unlock(&vfsmount_lock); |
1245 | } | |
1246 | } | |
1247 | ||
5528f911 TM |
1248 | /* |
1249 | * go through the vfsmounts we've just consigned to the graveyard to | |
1250 | * - check that they're still dead | |
1251 | * - delete the vfsmount from the appropriate namespace under lock | |
1252 | * - dispose of the corpse | |
1253 | */ | |
1254 | static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts) | |
1255 | { | |
6b3286ed | 1256 | struct mnt_namespace *ns; |
5528f911 TM |
1257 | struct vfsmount *mnt; |
1258 | ||
1259 | while (!list_empty(graveyard)) { | |
1260 | LIST_HEAD(umounts); | |
b5e61818 | 1261 | mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire); |
5528f911 TM |
1262 | list_del_init(&mnt->mnt_expire); |
1263 | ||
1264 | /* don't do anything if the namespace is dead - all the | |
1265 | * vfsmounts from it are going away anyway */ | |
6b3286ed KK |
1266 | ns = mnt->mnt_ns; |
1267 | if (!ns || !ns->root) | |
5528f911 | 1268 | continue; |
6b3286ed | 1269 | get_mnt_ns(ns); |
5528f911 TM |
1270 | |
1271 | spin_unlock(&vfsmount_lock); | |
1272 | down_write(&namespace_sem); | |
1273 | expire_mount(mnt, mounts, &umounts); | |
1274 | up_write(&namespace_sem); | |
1275 | release_mounts(&umounts); | |
1276 | mntput(mnt); | |
6b3286ed | 1277 | put_mnt_ns(ns); |
5528f911 TM |
1278 | spin_lock(&vfsmount_lock); |
1279 | } | |
1280 | } | |
1281 | ||
1da177e4 LT |
1282 | /* |
1283 | * process a list of expirable mountpoints with the intent of discarding any | |
1284 | * mountpoints that aren't in use and haven't been touched since last we came | |
1285 | * here | |
1286 | */ | |
1287 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1288 | { | |
1da177e4 LT |
1289 | struct vfsmount *mnt, *next; |
1290 | LIST_HEAD(graveyard); | |
1291 | ||
1292 | if (list_empty(mounts)) | |
1293 | return; | |
1294 | ||
1295 | spin_lock(&vfsmount_lock); | |
1296 | ||
1297 | /* extract from the expiration list every vfsmount that matches the | |
1298 | * following criteria: | |
1299 | * - only referenced by its parent vfsmount | |
1300 | * - still marked for expiry (marked on the last call here; marks are | |
1301 | * cleared by mntput()) | |
1302 | */ | |
55e700b9 | 1303 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 LT |
1304 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1305 | atomic_read(&mnt->mnt_count) != 1) | |
1306 | continue; | |
1307 | ||
1308 | mntget(mnt); | |
55e700b9 | 1309 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 LT |
1310 | } |
1311 | ||
5528f911 | 1312 | expire_mount_list(&graveyard, mounts); |
1da177e4 | 1313 | |
5528f911 TM |
1314 | spin_unlock(&vfsmount_lock); |
1315 | } | |
1316 | ||
1317 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1318 | ||
1319 | /* | |
1320 | * Ripoff of 'select_parent()' | |
1321 | * | |
1322 | * search the list of submounts for a given mountpoint, and move any | |
1323 | * shrinkable submounts to the 'graveyard' list. | |
1324 | */ | |
1325 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1326 | { | |
1327 | struct vfsmount *this_parent = parent; | |
1328 | struct list_head *next; | |
1329 | int found = 0; | |
1330 | ||
1331 | repeat: | |
1332 | next = this_parent->mnt_mounts.next; | |
1333 | resume: | |
1334 | while (next != &this_parent->mnt_mounts) { | |
1335 | struct list_head *tmp = next; | |
1336 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1337 | ||
1338 | next = tmp->next; | |
1339 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1340 | continue; |
5528f911 TM |
1341 | /* |
1342 | * Descend a level if the d_mounts list is non-empty. | |
1343 | */ | |
1344 | if (!list_empty(&mnt->mnt_mounts)) { | |
1345 | this_parent = mnt; | |
1346 | goto repeat; | |
1347 | } | |
1da177e4 | 1348 | |
5528f911 TM |
1349 | if (!propagate_mount_busy(mnt, 1)) { |
1350 | mntget(mnt); | |
1351 | list_move_tail(&mnt->mnt_expire, graveyard); | |
1352 | found++; | |
1353 | } | |
1da177e4 | 1354 | } |
5528f911 TM |
1355 | /* |
1356 | * All done at this level ... ascend and resume the search | |
1357 | */ | |
1358 | if (this_parent != parent) { | |
1359 | next = this_parent->mnt_child.next; | |
1360 | this_parent = this_parent->mnt_parent; | |
1361 | goto resume; | |
1362 | } | |
1363 | return found; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * process a list of expirable mountpoints with the intent of discarding any | |
1368 | * submounts of a specific parent mountpoint | |
1369 | */ | |
1370 | void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts) | |
1371 | { | |
1372 | LIST_HEAD(graveyard); | |
1373 | int found; | |
1374 | ||
1375 | spin_lock(&vfsmount_lock); | |
1376 | ||
1377 | /* extract submounts of 'mountpoint' from the expiration list */ | |
1378 | while ((found = select_submounts(mountpoint, &graveyard)) != 0) | |
1379 | expire_mount_list(&graveyard, mounts); | |
1da177e4 LT |
1380 | |
1381 | spin_unlock(&vfsmount_lock); | |
1382 | } | |
1383 | ||
5528f911 | 1384 | EXPORT_SYMBOL_GPL(shrink_submounts); |
1da177e4 LT |
1385 | |
1386 | /* | |
1387 | * Some copy_from_user() implementations do not return the exact number of | |
1388 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1389 | * Note that this function differs from copy_from_user() in that it will oops | |
1390 | * on bad values of `to', rather than returning a short copy. | |
1391 | */ | |
b58fed8b RP |
1392 | static long exact_copy_from_user(void *to, const void __user * from, |
1393 | unsigned long n) | |
1da177e4 LT |
1394 | { |
1395 | char *t = to; | |
1396 | const char __user *f = from; | |
1397 | char c; | |
1398 | ||
1399 | if (!access_ok(VERIFY_READ, from, n)) | |
1400 | return n; | |
1401 | ||
1402 | while (n) { | |
1403 | if (__get_user(c, f)) { | |
1404 | memset(t, 0, n); | |
1405 | break; | |
1406 | } | |
1407 | *t++ = c; | |
1408 | f++; | |
1409 | n--; | |
1410 | } | |
1411 | return n; | |
1412 | } | |
1413 | ||
b58fed8b | 1414 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1415 | { |
1416 | int i; | |
1417 | unsigned long page; | |
1418 | unsigned long size; | |
b58fed8b | 1419 | |
1da177e4 LT |
1420 | *where = 0; |
1421 | if (!data) | |
1422 | return 0; | |
1423 | ||
1424 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1425 | return -ENOMEM; | |
1426 | ||
1427 | /* We only care that *some* data at the address the user | |
1428 | * gave us is valid. Just in case, we'll zero | |
1429 | * the remainder of the page. | |
1430 | */ | |
1431 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1432 | size = TASK_SIZE - (unsigned long)data; | |
1433 | if (size > PAGE_SIZE) | |
1434 | size = PAGE_SIZE; | |
1435 | ||
1436 | i = size - exact_copy_from_user((void *)page, data, size); | |
1437 | if (!i) { | |
b58fed8b | 1438 | free_page(page); |
1da177e4 LT |
1439 | return -EFAULT; |
1440 | } | |
1441 | if (i != PAGE_SIZE) | |
1442 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1443 | *where = page; | |
1444 | return 0; | |
1445 | } | |
1446 | ||
1447 | /* | |
1448 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1449 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1450 | * | |
1451 | * data is a (void *) that can point to any structure up to | |
1452 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1453 | * information (or be NULL). | |
1454 | * | |
1455 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1456 | * When the flags word was introduced its top half was required | |
1457 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1458 | * Therefore, if this magic number is present, it carries no information | |
1459 | * and must be discarded. | |
1460 | */ | |
b58fed8b | 1461 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1462 | unsigned long flags, void *data_page) |
1463 | { | |
1464 | struct nameidata nd; | |
1465 | int retval = 0; | |
1466 | int mnt_flags = 0; | |
1467 | ||
1468 | /* Discard magic */ | |
1469 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1470 | flags &= ~MS_MGC_MSK; | |
1471 | ||
1472 | /* Basic sanity checks */ | |
1473 | ||
1474 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1475 | return -EINVAL; | |
1476 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1477 | return -EINVAL; | |
1478 | ||
1479 | if (data_page) | |
1480 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1481 | ||
1482 | /* Separate the per-mountpoint flags */ | |
1483 | if (flags & MS_NOSUID) | |
1484 | mnt_flags |= MNT_NOSUID; | |
1485 | if (flags & MS_NODEV) | |
1486 | mnt_flags |= MNT_NODEV; | |
1487 | if (flags & MS_NOEXEC) | |
1488 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1489 | if (flags & MS_NOATIME) |
1490 | mnt_flags |= MNT_NOATIME; | |
1491 | if (flags & MS_NODIRATIME) | |
1492 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1493 | if (flags & MS_RELATIME) |
1494 | mnt_flags |= MNT_RELATIME; | |
fc33a7bb CH |
1495 | |
1496 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1497 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1498 | |
1499 | /* ... and get the mountpoint */ | |
1500 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1501 | if (retval) | |
1502 | return retval; | |
1503 | ||
1504 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1505 | if (retval) | |
1506 | goto dput_out; | |
1507 | ||
1508 | if (flags & MS_REMOUNT) | |
1509 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1510 | data_page); | |
1511 | else if (flags & MS_BIND) | |
eee391a6 | 1512 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1513 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1514 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1515 | else if (flags & MS_MOVE) |
1516 | retval = do_move_mount(&nd, dev_name); | |
1517 | else | |
1518 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1519 | dev_name, data_page); | |
1520 | dput_out: | |
1d957f9b | 1521 | path_put(&nd.path); |
1da177e4 LT |
1522 | return retval; |
1523 | } | |
1524 | ||
741a2951 JD |
1525 | /* |
1526 | * Allocate a new namespace structure and populate it with contents | |
1527 | * copied from the namespace of the passed in task structure. | |
1528 | */ | |
e3222c4e | 1529 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1530 | struct fs_struct *fs) |
1da177e4 | 1531 | { |
6b3286ed | 1532 | struct mnt_namespace *new_ns; |
1da177e4 | 1533 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1534 | struct vfsmount *p, *q; |
1535 | ||
6b3286ed | 1536 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1537 | if (!new_ns) |
467e9f4b | 1538 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1539 | |
1540 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1541 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1542 | init_waitqueue_head(&new_ns->poll); |
1543 | new_ns->event = 0; | |
1da177e4 | 1544 | |
390c6843 | 1545 | down_write(&namespace_sem); |
1da177e4 | 1546 | /* First pass: copy the tree topology */ |
6b3286ed | 1547 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1548 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1549 | if (!new_ns->root) { |
390c6843 | 1550 | up_write(&namespace_sem); |
1da177e4 | 1551 | kfree(new_ns); |
467e9f4b | 1552 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1553 | } |
1554 | spin_lock(&vfsmount_lock); | |
1555 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1556 | spin_unlock(&vfsmount_lock); | |
1557 | ||
1558 | /* | |
1559 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1560 | * as belonging to new namespace. We have already acquired a private | |
1561 | * fs_struct, so tsk->fs->lock is not needed. | |
1562 | */ | |
6b3286ed | 1563 | p = mnt_ns->root; |
1da177e4 LT |
1564 | q = new_ns->root; |
1565 | while (p) { | |
6b3286ed | 1566 | q->mnt_ns = new_ns; |
1da177e4 | 1567 | if (fs) { |
6ac08c39 | 1568 | if (p == fs->root.mnt) { |
1da177e4 | 1569 | rootmnt = p; |
6ac08c39 | 1570 | fs->root.mnt = mntget(q); |
1da177e4 | 1571 | } |
6ac08c39 | 1572 | if (p == fs->pwd.mnt) { |
1da177e4 | 1573 | pwdmnt = p; |
6ac08c39 | 1574 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1575 | } |
6ac08c39 | 1576 | if (p == fs->altroot.mnt) { |
1da177e4 | 1577 | altrootmnt = p; |
6ac08c39 | 1578 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1579 | } |
1580 | } | |
6b3286ed | 1581 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1582 | q = next_mnt(q, new_ns->root); |
1583 | } | |
390c6843 | 1584 | up_write(&namespace_sem); |
1da177e4 | 1585 | |
1da177e4 LT |
1586 | if (rootmnt) |
1587 | mntput(rootmnt); | |
1588 | if (pwdmnt) | |
1589 | mntput(pwdmnt); | |
1590 | if (altrootmnt) | |
1591 | mntput(altrootmnt); | |
1592 | ||
741a2951 JD |
1593 | return new_ns; |
1594 | } | |
1595 | ||
213dd266 | 1596 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1597 | struct fs_struct *new_fs) |
741a2951 | 1598 | { |
6b3286ed | 1599 | struct mnt_namespace *new_ns; |
741a2951 | 1600 | |
e3222c4e | 1601 | BUG_ON(!ns); |
6b3286ed | 1602 | get_mnt_ns(ns); |
741a2951 JD |
1603 | |
1604 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1605 | return ns; |
741a2951 | 1606 | |
e3222c4e | 1607 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1608 | |
6b3286ed | 1609 | put_mnt_ns(ns); |
e3222c4e | 1610 | return new_ns; |
1da177e4 LT |
1611 | } |
1612 | ||
1613 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1614 | char __user * type, unsigned long flags, | |
1615 | void __user * data) | |
1616 | { | |
1617 | int retval; | |
1618 | unsigned long data_page; | |
1619 | unsigned long type_page; | |
1620 | unsigned long dev_page; | |
1621 | char *dir_page; | |
1622 | ||
b58fed8b | 1623 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1624 | if (retval < 0) |
1625 | return retval; | |
1626 | ||
1627 | dir_page = getname(dir_name); | |
1628 | retval = PTR_ERR(dir_page); | |
1629 | if (IS_ERR(dir_page)) | |
1630 | goto out1; | |
1631 | ||
b58fed8b | 1632 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1633 | if (retval < 0) |
1634 | goto out2; | |
1635 | ||
b58fed8b | 1636 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1637 | if (retval < 0) |
1638 | goto out3; | |
1639 | ||
1640 | lock_kernel(); | |
b58fed8b RP |
1641 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1642 | flags, (void *)data_page); | |
1da177e4 LT |
1643 | unlock_kernel(); |
1644 | free_page(data_page); | |
1645 | ||
1646 | out3: | |
1647 | free_page(dev_page); | |
1648 | out2: | |
1649 | putname(dir_page); | |
1650 | out1: | |
1651 | free_page(type_page); | |
1652 | return retval; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1657 | * It can block. Requires the big lock held. | |
1658 | */ | |
ac748a09 | 1659 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 1660 | { |
6ac08c39 JB |
1661 | struct path old_root; |
1662 | ||
1da177e4 LT |
1663 | write_lock(&fs->lock); |
1664 | old_root = fs->root; | |
ac748a09 JB |
1665 | fs->root = *path; |
1666 | path_get(path); | |
1da177e4 | 1667 | write_unlock(&fs->lock); |
6ac08c39 JB |
1668 | if (old_root.dentry) |
1669 | path_put(&old_root); | |
1da177e4 LT |
1670 | } |
1671 | ||
1672 | /* | |
1673 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1674 | * It can block. Requires the big lock held. | |
1675 | */ | |
ac748a09 | 1676 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 1677 | { |
6ac08c39 | 1678 | struct path old_pwd; |
1da177e4 LT |
1679 | |
1680 | write_lock(&fs->lock); | |
1681 | old_pwd = fs->pwd; | |
ac748a09 JB |
1682 | fs->pwd = *path; |
1683 | path_get(path); | |
1da177e4 LT |
1684 | write_unlock(&fs->lock); |
1685 | ||
6ac08c39 JB |
1686 | if (old_pwd.dentry) |
1687 | path_put(&old_pwd); | |
1da177e4 LT |
1688 | } |
1689 | ||
1a390689 | 1690 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
1691 | { |
1692 | struct task_struct *g, *p; | |
1693 | struct fs_struct *fs; | |
1694 | ||
1695 | read_lock(&tasklist_lock); | |
1696 | do_each_thread(g, p) { | |
1697 | task_lock(p); | |
1698 | fs = p->fs; | |
1699 | if (fs) { | |
1700 | atomic_inc(&fs->count); | |
1701 | task_unlock(p); | |
1a390689 AV |
1702 | if (fs->root.dentry == old_root->dentry |
1703 | && fs->root.mnt == old_root->mnt) | |
1704 | set_fs_root(fs, new_root); | |
1705 | if (fs->pwd.dentry == old_root->dentry | |
1706 | && fs->pwd.mnt == old_root->mnt) | |
1707 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
1708 | put_fs_struct(fs); |
1709 | } else | |
1710 | task_unlock(p); | |
1711 | } while_each_thread(g, p); | |
1712 | read_unlock(&tasklist_lock); | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * pivot_root Semantics: | |
1717 | * Moves the root file system of the current process to the directory put_old, | |
1718 | * makes new_root as the new root file system of the current process, and sets | |
1719 | * root/cwd of all processes which had them on the current root to new_root. | |
1720 | * | |
1721 | * Restrictions: | |
1722 | * The new_root and put_old must be directories, and must not be on the | |
1723 | * same file system as the current process root. The put_old must be | |
1724 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1725 | * pointed to by put_old must yield the same directory as new_root. No other | |
1726 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1727 | * | |
4a0d11fa NB |
1728 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
1729 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1730 | * in this situation. | |
1731 | * | |
1da177e4 LT |
1732 | * Notes: |
1733 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1734 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1735 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1736 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1737 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1738 | * first. | |
1739 | */ | |
b58fed8b RP |
1740 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1741 | const char __user * put_old) | |
1da177e4 LT |
1742 | { |
1743 | struct vfsmount *tmp; | |
1a390689 AV |
1744 | struct nameidata new_nd, old_nd, user_nd; |
1745 | struct path parent_path, root_parent; | |
1da177e4 LT |
1746 | int error; |
1747 | ||
1748 | if (!capable(CAP_SYS_ADMIN)) | |
1749 | return -EPERM; | |
1750 | ||
1751 | lock_kernel(); | |
1752 | ||
b58fed8b RP |
1753 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1754 | &new_nd); | |
1da177e4 LT |
1755 | if (error) |
1756 | goto out0; | |
1757 | error = -EINVAL; | |
4ac91378 | 1758 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
1759 | goto out1; |
1760 | ||
b58fed8b | 1761 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1762 | if (error) |
1763 | goto out1; | |
1764 | ||
1765 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1766 | if (error) { | |
1d957f9b | 1767 | path_put(&old_nd.path); |
1da177e4 LT |
1768 | goto out1; |
1769 | } | |
1770 | ||
1771 | read_lock(¤t->fs->lock); | |
6ac08c39 JB |
1772 | user_nd.path = current->fs->root; |
1773 | path_get(¤t->fs->root); | |
1da177e4 | 1774 | read_unlock(¤t->fs->lock); |
390c6843 | 1775 | down_write(&namespace_sem); |
4ac91378 | 1776 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 1777 | error = -EINVAL; |
4ac91378 JB |
1778 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
1779 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
1780 | IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) | |
21444403 | 1781 | goto out2; |
4ac91378 | 1782 | if (!check_mnt(user_nd.path.mnt)) |
1da177e4 LT |
1783 | goto out2; |
1784 | error = -ENOENT; | |
4ac91378 | 1785 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 1786 | goto out2; |
4ac91378 | 1787 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 1788 | goto out2; |
4ac91378 | 1789 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
1790 | goto out2; |
1791 | error = -EBUSY; | |
4ac91378 JB |
1792 | if (new_nd.path.mnt == user_nd.path.mnt || |
1793 | old_nd.path.mnt == user_nd.path.mnt) | |
1da177e4 LT |
1794 | goto out2; /* loop, on the same file system */ |
1795 | error = -EINVAL; | |
4ac91378 | 1796 | if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) |
1da177e4 | 1797 | goto out2; /* not a mountpoint */ |
4ac91378 | 1798 | if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) |
0bb6fcc1 | 1799 | goto out2; /* not attached */ |
4ac91378 | 1800 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 1801 | goto out2; /* not a mountpoint */ |
4ac91378 | 1802 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 1803 | goto out2; /* not attached */ |
4ac91378 JB |
1804 | /* make sure we can reach put_old from new_root */ |
1805 | tmp = old_nd.path.mnt; | |
1da177e4 | 1806 | spin_lock(&vfsmount_lock); |
4ac91378 | 1807 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
1808 | for (;;) { |
1809 | if (tmp->mnt_parent == tmp) | |
1810 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 1811 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
1812 | break; |
1813 | tmp = tmp->mnt_parent; | |
1814 | } | |
4ac91378 | 1815 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 1816 | goto out3; |
4ac91378 | 1817 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 1818 | goto out3; |
1a390689 | 1819 | detach_mnt(new_nd.path.mnt, &parent_path); |
4ac91378 JB |
1820 | detach_mnt(user_nd.path.mnt, &root_parent); |
1821 | /* mount old root on put_old */ | |
1a390689 | 1822 | attach_mnt(user_nd.path.mnt, &old_nd.path); |
4ac91378 JB |
1823 | /* mount new_root on / */ |
1824 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 1825 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 1826 | spin_unlock(&vfsmount_lock); |
1a390689 | 1827 | chroot_fs_refs(&user_nd.path, &new_nd.path); |
1da177e4 LT |
1828 | security_sb_post_pivotroot(&user_nd, &new_nd); |
1829 | error = 0; | |
1a390689 AV |
1830 | path_put(&root_parent); |
1831 | path_put(&parent_path); | |
1da177e4 | 1832 | out2: |
4ac91378 | 1833 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 1834 | up_write(&namespace_sem); |
1d957f9b JB |
1835 | path_put(&user_nd.path); |
1836 | path_put(&old_nd.path); | |
1da177e4 | 1837 | out1: |
1d957f9b | 1838 | path_put(&new_nd.path); |
1da177e4 LT |
1839 | out0: |
1840 | unlock_kernel(); | |
1841 | return error; | |
1842 | out3: | |
1843 | spin_unlock(&vfsmount_lock); | |
1844 | goto out2; | |
1845 | } | |
1846 | ||
1847 | static void __init init_mount_tree(void) | |
1848 | { | |
1849 | struct vfsmount *mnt; | |
6b3286ed | 1850 | struct mnt_namespace *ns; |
ac748a09 | 1851 | struct path root; |
1da177e4 LT |
1852 | |
1853 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1854 | if (IS_ERR(mnt)) | |
1855 | panic("Can't create rootfs"); | |
6b3286ed KK |
1856 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
1857 | if (!ns) | |
1da177e4 | 1858 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
1859 | atomic_set(&ns->count, 1); |
1860 | INIT_LIST_HEAD(&ns->list); | |
1861 | init_waitqueue_head(&ns->poll); | |
1862 | ns->event = 0; | |
1863 | list_add(&mnt->mnt_list, &ns->list); | |
1864 | ns->root = mnt; | |
1865 | mnt->mnt_ns = ns; | |
1866 | ||
1867 | init_task.nsproxy->mnt_ns = ns; | |
1868 | get_mnt_ns(ns); | |
1869 | ||
ac748a09 JB |
1870 | root.mnt = ns->root; |
1871 | root.dentry = ns->root->mnt_root; | |
1872 | ||
1873 | set_fs_pwd(current->fs, &root); | |
1874 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
1875 | } |
1876 | ||
74bf17cf | 1877 | void __init mnt_init(void) |
1da177e4 | 1878 | { |
13f14b4d | 1879 | unsigned u; |
15a67dd8 | 1880 | int err; |
1da177e4 | 1881 | |
390c6843 RP |
1882 | init_rwsem(&namespace_sem); |
1883 | ||
1da177e4 | 1884 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 1885 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 1886 | |
b58fed8b | 1887 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1888 | |
1889 | if (!mount_hashtable) | |
1890 | panic("Failed to allocate mount hash table\n"); | |
1891 | ||
13f14b4d ED |
1892 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
1893 | ||
1894 | for (u = 0; u < HASH_SIZE; u++) | |
1895 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 1896 | |
15a67dd8 RD |
1897 | err = sysfs_init(); |
1898 | if (err) | |
1899 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
1900 | __FUNCTION__, err); | |
00d26666 GKH |
1901 | fs_kobj = kobject_create_and_add("fs", NULL); |
1902 | if (!fs_kobj) | |
1903 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
1904 | init_rootfs(); |
1905 | init_mount_tree(); | |
1906 | } | |
1907 | ||
6b3286ed | 1908 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 1909 | { |
6b3286ed | 1910 | struct vfsmount *root = ns->root; |
70fbcdf4 | 1911 | LIST_HEAD(umount_list); |
6b3286ed | 1912 | ns->root = NULL; |
1ce88cf4 | 1913 | spin_unlock(&vfsmount_lock); |
390c6843 | 1914 | down_write(&namespace_sem); |
1da177e4 | 1915 | spin_lock(&vfsmount_lock); |
a05964f3 | 1916 | umount_tree(root, 0, &umount_list); |
1da177e4 | 1917 | spin_unlock(&vfsmount_lock); |
390c6843 | 1918 | up_write(&namespace_sem); |
70fbcdf4 | 1919 | release_mounts(&umount_list); |
6b3286ed | 1920 | kfree(ns); |
1da177e4 | 1921 | } |