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 | ||
1da177e4 LT |
158 | static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) |
159 | { | |
4ac91378 JB |
160 | old_nd->path.dentry = mnt->mnt_mountpoint; |
161 | old_nd->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); | |
4ac91378 | 166 | old_nd->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 | ||
1da177e4 LT |
177 | static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) |
178 | { | |
4ac91378 | 179 | mnt_set_mountpoint(nd->path.mnt, nd->path.dentry, mnt); |
b90fa9ae | 180 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
4ac91378 JB |
181 | hash(nd->path.mnt, nd->path.dentry)); |
182 | list_add_tail(&mnt->mnt_child, &nd->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; | |
411 | ||
412 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
413 | seq_putc(m, ' '); | |
414 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
415 | seq_putc(m, ' '); | |
416 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
417 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { |
418 | seq_putc(m, '.'); | |
419 | mangle(m, mnt->mnt_sb->s_subtype); | |
420 | } | |
1da177e4 LT |
421 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); |
422 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
423 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
424 | seq_puts(m, fs_infop->str); | |
425 | } | |
426 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
427 | if (mnt->mnt_flags & fs_infop->flag) | |
428 | seq_puts(m, fs_infop->str); | |
429 | } | |
430 | if (mnt->mnt_sb->s_op->show_options) | |
431 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
432 | seq_puts(m, " 0 0\n"); | |
433 | return err; | |
434 | } | |
435 | ||
436 | struct seq_operations mounts_op = { | |
437 | .start = m_start, | |
438 | .next = m_next, | |
439 | .stop = m_stop, | |
440 | .show = show_vfsmnt | |
441 | }; | |
442 | ||
b4629fe2 CL |
443 | static int show_vfsstat(struct seq_file *m, void *v) |
444 | { | |
b0765fb8 | 445 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
b4629fe2 CL |
446 | int err = 0; |
447 | ||
448 | /* device */ | |
449 | if (mnt->mnt_devname) { | |
450 | seq_puts(m, "device "); | |
451 | mangle(m, mnt->mnt_devname); | |
452 | } else | |
453 | seq_puts(m, "no device"); | |
454 | ||
455 | /* mount point */ | |
456 | seq_puts(m, " mounted on "); | |
457 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
458 | seq_putc(m, ' '); | |
459 | ||
460 | /* file system type */ | |
461 | seq_puts(m, "with fstype "); | |
462 | mangle(m, mnt->mnt_sb->s_type->name); | |
463 | ||
464 | /* optional statistics */ | |
465 | if (mnt->mnt_sb->s_op->show_stats) { | |
466 | seq_putc(m, ' '); | |
467 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
468 | } | |
469 | ||
470 | seq_putc(m, '\n'); | |
471 | return err; | |
472 | } | |
473 | ||
474 | struct seq_operations mountstats_op = { | |
475 | .start = m_start, | |
476 | .next = m_next, | |
477 | .stop = m_stop, | |
478 | .show = show_vfsstat, | |
479 | }; | |
480 | ||
1da177e4 LT |
481 | /** |
482 | * may_umount_tree - check if a mount tree is busy | |
483 | * @mnt: root of mount tree | |
484 | * | |
485 | * This is called to check if a tree of mounts has any | |
486 | * open files, pwds, chroots or sub mounts that are | |
487 | * busy. | |
488 | */ | |
489 | int may_umount_tree(struct vfsmount *mnt) | |
490 | { | |
36341f64 RP |
491 | int actual_refs = 0; |
492 | int minimum_refs = 0; | |
493 | struct vfsmount *p; | |
1da177e4 LT |
494 | |
495 | spin_lock(&vfsmount_lock); | |
36341f64 | 496 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
497 | actual_refs += atomic_read(&p->mnt_count); |
498 | minimum_refs += 2; | |
1da177e4 LT |
499 | } |
500 | spin_unlock(&vfsmount_lock); | |
501 | ||
502 | if (actual_refs > minimum_refs) | |
e3474a8e | 503 | return 0; |
1da177e4 | 504 | |
e3474a8e | 505 | return 1; |
1da177e4 LT |
506 | } |
507 | ||
508 | EXPORT_SYMBOL(may_umount_tree); | |
509 | ||
510 | /** | |
511 | * may_umount - check if a mount point is busy | |
512 | * @mnt: root of mount | |
513 | * | |
514 | * This is called to check if a mount point has any | |
515 | * open files, pwds, chroots or sub mounts. If the | |
516 | * mount has sub mounts this will return busy | |
517 | * regardless of whether the sub mounts are busy. | |
518 | * | |
519 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
520 | * give false negatives. The main reason why it's here is that we need | |
521 | * a non-destructive way to look for easily umountable filesystems. | |
522 | */ | |
523 | int may_umount(struct vfsmount *mnt) | |
524 | { | |
e3474a8e | 525 | int ret = 1; |
a05964f3 RP |
526 | spin_lock(&vfsmount_lock); |
527 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 528 | ret = 0; |
a05964f3 RP |
529 | spin_unlock(&vfsmount_lock); |
530 | return ret; | |
1da177e4 LT |
531 | } |
532 | ||
533 | EXPORT_SYMBOL(may_umount); | |
534 | ||
b90fa9ae | 535 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
536 | { |
537 | struct vfsmount *mnt; | |
bf066c7d | 538 | while (!list_empty(head)) { |
b5e61818 | 539 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
540 | list_del_init(&mnt->mnt_hash); |
541 | if (mnt->mnt_parent != mnt) { | |
542 | struct dentry *dentry; | |
543 | struct vfsmount *m; | |
544 | spin_lock(&vfsmount_lock); | |
545 | dentry = mnt->mnt_mountpoint; | |
546 | m = mnt->mnt_parent; | |
547 | mnt->mnt_mountpoint = mnt->mnt_root; | |
548 | mnt->mnt_parent = mnt; | |
549 | spin_unlock(&vfsmount_lock); | |
550 | dput(dentry); | |
551 | mntput(m); | |
552 | } | |
553 | mntput(mnt); | |
554 | } | |
555 | } | |
556 | ||
a05964f3 | 557 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
558 | { |
559 | struct vfsmount *p; | |
1da177e4 | 560 | |
1bfba4e8 AM |
561 | for (p = mnt; p; p = next_mnt(p, mnt)) |
562 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 563 | |
a05964f3 RP |
564 | if (propagate) |
565 | propagate_umount(kill); | |
566 | ||
70fbcdf4 RP |
567 | list_for_each_entry(p, kill, mnt_hash) { |
568 | list_del_init(&p->mnt_expire); | |
569 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
570 | __touch_mnt_namespace(p->mnt_ns); |
571 | p->mnt_ns = NULL; | |
70fbcdf4 RP |
572 | list_del_init(&p->mnt_child); |
573 | if (p->mnt_parent != p) | |
f30ac319 | 574 | p->mnt_mountpoint->d_mounted--; |
a05964f3 | 575 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
576 | } |
577 | } | |
578 | ||
579 | static int do_umount(struct vfsmount *mnt, int flags) | |
580 | { | |
b58fed8b | 581 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 582 | int retval; |
70fbcdf4 | 583 | LIST_HEAD(umount_list); |
1da177e4 LT |
584 | |
585 | retval = security_sb_umount(mnt, flags); | |
586 | if (retval) | |
587 | return retval; | |
588 | ||
589 | /* | |
590 | * Allow userspace to request a mountpoint be expired rather than | |
591 | * unmounting unconditionally. Unmount only happens if: | |
592 | * (1) the mark is already set (the mark is cleared by mntput()) | |
593 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
594 | */ | |
595 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 596 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
597 | flags & (MNT_FORCE | MNT_DETACH)) |
598 | return -EINVAL; | |
599 | ||
600 | if (atomic_read(&mnt->mnt_count) != 2) | |
601 | return -EBUSY; | |
602 | ||
603 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
604 | return -EAGAIN; | |
605 | } | |
606 | ||
607 | /* | |
608 | * If we may have to abort operations to get out of this | |
609 | * mount, and they will themselves hold resources we must | |
610 | * allow the fs to do things. In the Unix tradition of | |
611 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
612 | * might fail to complete on the first run through as other tasks | |
613 | * must return, and the like. Thats for the mount program to worry | |
614 | * about for the moment. | |
615 | */ | |
616 | ||
617 | lock_kernel(); | |
8b512d9a TM |
618 | if (sb->s_op->umount_begin) |
619 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
620 | unlock_kernel(); |
621 | ||
622 | /* | |
623 | * No sense to grab the lock for this test, but test itself looks | |
624 | * somewhat bogus. Suggestions for better replacement? | |
625 | * Ho-hum... In principle, we might treat that as umount + switch | |
626 | * to rootfs. GC would eventually take care of the old vfsmount. | |
627 | * Actually it makes sense, especially if rootfs would contain a | |
628 | * /reboot - static binary that would close all descriptors and | |
629 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
630 | */ | |
6ac08c39 | 631 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
632 | /* |
633 | * Special case for "unmounting" root ... | |
634 | * we just try to remount it readonly. | |
635 | */ | |
636 | down_write(&sb->s_umount); | |
637 | if (!(sb->s_flags & MS_RDONLY)) { | |
638 | lock_kernel(); | |
639 | DQUOT_OFF(sb); | |
640 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
641 | unlock_kernel(); | |
642 | } | |
643 | up_write(&sb->s_umount); | |
644 | return retval; | |
645 | } | |
646 | ||
390c6843 | 647 | down_write(&namespace_sem); |
1da177e4 | 648 | spin_lock(&vfsmount_lock); |
5addc5dd | 649 | event++; |
1da177e4 | 650 | |
1da177e4 | 651 | retval = -EBUSY; |
a05964f3 | 652 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 653 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 654 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
655 | retval = 0; |
656 | } | |
657 | spin_unlock(&vfsmount_lock); | |
658 | if (retval) | |
659 | security_sb_umount_busy(mnt); | |
390c6843 | 660 | up_write(&namespace_sem); |
70fbcdf4 | 661 | release_mounts(&umount_list); |
1da177e4 LT |
662 | return retval; |
663 | } | |
664 | ||
665 | /* | |
666 | * Now umount can handle mount points as well as block devices. | |
667 | * This is important for filesystems which use unnamed block devices. | |
668 | * | |
669 | * We now support a flag for forced unmount like the other 'big iron' | |
670 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
671 | */ | |
672 | ||
673 | asmlinkage long sys_umount(char __user * name, int flags) | |
674 | { | |
675 | struct nameidata nd; | |
676 | int retval; | |
677 | ||
678 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
679 | if (retval) | |
680 | goto out; | |
681 | retval = -EINVAL; | |
4ac91378 | 682 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 683 | goto dput_and_out; |
4ac91378 | 684 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
685 | goto dput_and_out; |
686 | ||
687 | retval = -EPERM; | |
688 | if (!capable(CAP_SYS_ADMIN)) | |
689 | goto dput_and_out; | |
690 | ||
4ac91378 | 691 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 692 | dput_and_out: |
429731b1 | 693 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
694 | dput(nd.path.dentry); |
695 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
696 | out: |
697 | return retval; | |
698 | } | |
699 | ||
700 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
701 | ||
702 | /* | |
b58fed8b | 703 | * The 2.0 compatible umount. No flags. |
1da177e4 | 704 | */ |
1da177e4 LT |
705 | asmlinkage long sys_oldumount(char __user * name) |
706 | { | |
b58fed8b | 707 | return sys_umount(name, 0); |
1da177e4 LT |
708 | } |
709 | ||
710 | #endif | |
711 | ||
712 | static int mount_is_safe(struct nameidata *nd) | |
713 | { | |
714 | if (capable(CAP_SYS_ADMIN)) | |
715 | return 0; | |
716 | return -EPERM; | |
717 | #ifdef notyet | |
4ac91378 | 718 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 719 | return -EPERM; |
4ac91378 JB |
720 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
721 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
722 | return -EPERM; |
723 | } | |
e4543edd | 724 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
725 | return -EPERM; |
726 | return 0; | |
727 | #endif | |
728 | } | |
729 | ||
b58fed8b | 730 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
731 | { |
732 | while (1) { | |
733 | if (d == dentry) | |
734 | return 1; | |
735 | if (d == NULL || d == d->d_parent) | |
736 | return 0; | |
737 | d = d->d_parent; | |
738 | } | |
739 | } | |
740 | ||
b90fa9ae | 741 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 742 | int flag) |
1da177e4 LT |
743 | { |
744 | struct vfsmount *res, *p, *q, *r, *s; | |
1da177e4 LT |
745 | struct nameidata nd; |
746 | ||
9676f0c6 RP |
747 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
748 | return NULL; | |
749 | ||
36341f64 | 750 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
751 | if (!q) |
752 | goto Enomem; | |
753 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
754 | ||
755 | p = mnt; | |
fdadd65f | 756 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
757 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
758 | continue; | |
759 | ||
760 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
761 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
762 | s = skip_mnt_tree(s); | |
763 | continue; | |
764 | } | |
1da177e4 LT |
765 | while (p != s->mnt_parent) { |
766 | p = p->mnt_parent; | |
767 | q = q->mnt_parent; | |
768 | } | |
769 | p = s; | |
4ac91378 JB |
770 | nd.path.mnt = q; |
771 | nd.path.dentry = p->mnt_mountpoint; | |
36341f64 | 772 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
773 | if (!q) |
774 | goto Enomem; | |
775 | spin_lock(&vfsmount_lock); | |
776 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
777 | attach_mnt(q, &nd); | |
778 | spin_unlock(&vfsmount_lock); | |
779 | } | |
780 | } | |
781 | return res; | |
b58fed8b | 782 | Enomem: |
1da177e4 | 783 | if (res) { |
70fbcdf4 | 784 | LIST_HEAD(umount_list); |
1da177e4 | 785 | spin_lock(&vfsmount_lock); |
a05964f3 | 786 | umount_tree(res, 0, &umount_list); |
1da177e4 | 787 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 788 | release_mounts(&umount_list); |
1da177e4 LT |
789 | } |
790 | return NULL; | |
791 | } | |
792 | ||
8aec0809 AV |
793 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
794 | { | |
795 | struct vfsmount *tree; | |
796 | down_read(&namespace_sem); | |
797 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); | |
798 | up_read(&namespace_sem); | |
799 | return tree; | |
800 | } | |
801 | ||
802 | void drop_collected_mounts(struct vfsmount *mnt) | |
803 | { | |
804 | LIST_HEAD(umount_list); | |
805 | down_read(&namespace_sem); | |
806 | spin_lock(&vfsmount_lock); | |
807 | umount_tree(mnt, 0, &umount_list); | |
808 | spin_unlock(&vfsmount_lock); | |
809 | up_read(&namespace_sem); | |
810 | release_mounts(&umount_list); | |
811 | } | |
812 | ||
b90fa9ae RP |
813 | /* |
814 | * @source_mnt : mount tree to be attached | |
21444403 RP |
815 | * @nd : place the mount tree @source_mnt is attached |
816 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
817 | * store the parent mount and mountpoint dentry. | |
818 | * (done when source_mnt is moved) | |
b90fa9ae RP |
819 | * |
820 | * NOTE: in the table below explains the semantics when a source mount | |
821 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
822 | * --------------------------------------------------------------------------- |
823 | * | BIND MOUNT OPERATION | | |
824 | * |************************************************************************** | |
825 | * | source-->| shared | private | slave | unbindable | | |
826 | * | dest | | | | | | |
827 | * | | | | | | | | |
828 | * | v | | | | | | |
829 | * |************************************************************************** | |
830 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
831 | * | | | | | | | |
832 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
833 | * *************************************************************************** | |
b90fa9ae RP |
834 | * A bind operation clones the source mount and mounts the clone on the |
835 | * destination mount. | |
836 | * | |
837 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
838 | * tree of the destination mount and the cloned mount is added to | |
839 | * the peer group of the source mount. | |
840 | * (+) the cloned mount is created under the destination mount and is marked | |
841 | * as shared. The cloned mount is added to the peer group of the source | |
842 | * mount. | |
5afe0022 RP |
843 | * (+++) the mount is propagated to all the mounts in the propagation tree |
844 | * of the destination mount and the cloned mount is made slave | |
845 | * of the same master as that of the source mount. The cloned mount | |
846 | * is marked as 'shared and slave'. | |
847 | * (*) the cloned mount is made a slave of the same master as that of the | |
848 | * source mount. | |
849 | * | |
9676f0c6 RP |
850 | * --------------------------------------------------------------------------- |
851 | * | MOVE MOUNT OPERATION | | |
852 | * |************************************************************************** | |
853 | * | source-->| shared | private | slave | unbindable | | |
854 | * | dest | | | | | | |
855 | * | | | | | | | | |
856 | * | v | | | | | | |
857 | * |************************************************************************** | |
858 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
859 | * | | | | | | | |
860 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
861 | * *************************************************************************** | |
5afe0022 RP |
862 | * |
863 | * (+) the mount is moved to the destination. And is then propagated to | |
864 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 865 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
866 | * (+++) the mount is moved to the destination and is then propagated to |
867 | * all the mounts belonging to the destination mount's propagation tree. | |
868 | * the mount is marked as 'shared and slave'. | |
869 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
870 | * |
871 | * if the source mount is a tree, the operations explained above is | |
872 | * applied to each mount in the tree. | |
873 | * Must be called without spinlocks held, since this function can sleep | |
874 | * in allocations. | |
875 | */ | |
876 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
21444403 | 877 | struct nameidata *nd, struct nameidata *parent_nd) |
b90fa9ae RP |
878 | { |
879 | LIST_HEAD(tree_list); | |
4ac91378 JB |
880 | struct vfsmount *dest_mnt = nd->path.mnt; |
881 | struct dentry *dest_dentry = nd->path.dentry; | |
b90fa9ae RP |
882 | struct vfsmount *child, *p; |
883 | ||
884 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
885 | return -EINVAL; | |
886 | ||
887 | if (IS_MNT_SHARED(dest_mnt)) { | |
888 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
889 | set_mnt_shared(p); | |
890 | } | |
891 | ||
892 | spin_lock(&vfsmount_lock); | |
21444403 RP |
893 | if (parent_nd) { |
894 | detach_mnt(source_mnt, parent_nd); | |
895 | attach_mnt(source_mnt, nd); | |
6b3286ed | 896 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
897 | } else { |
898 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
899 | commit_tree(source_mnt); | |
900 | } | |
b90fa9ae RP |
901 | |
902 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
903 | list_del_init(&child->mnt_hash); | |
904 | commit_tree(child); | |
905 | } | |
906 | spin_unlock(&vfsmount_lock); | |
907 | return 0; | |
908 | } | |
909 | ||
1da177e4 LT |
910 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
911 | { | |
912 | int err; | |
913 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
914 | return -EINVAL; | |
915 | ||
4ac91378 | 916 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1da177e4 LT |
917 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
918 | return -ENOTDIR; | |
919 | ||
920 | err = -ENOENT; | |
4ac91378 JB |
921 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
922 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
923 | goto out_unlock; |
924 | ||
925 | err = security_sb_check_sb(mnt, nd); | |
926 | if (err) | |
927 | goto out_unlock; | |
928 | ||
929 | err = -ENOENT; | |
4ac91378 | 930 | if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) |
21444403 | 931 | err = attach_recursive_mnt(mnt, nd, NULL); |
1da177e4 | 932 | out_unlock: |
4ac91378 | 933 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 LT |
934 | if (!err) |
935 | security_sb_post_addmount(mnt, nd); | |
936 | return err; | |
937 | } | |
938 | ||
07b20889 RP |
939 | /* |
940 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 941 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 942 | */ |
2dafe1c4 | 943 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 944 | { |
4ac91378 | 945 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
946 | int recurse = flag & MS_REC; |
947 | int type = flag & ~MS_REC; | |
948 | ||
ee6f9582 MS |
949 | if (!capable(CAP_SYS_ADMIN)) |
950 | return -EPERM; | |
951 | ||
4ac91378 | 952 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
953 | return -EINVAL; |
954 | ||
955 | down_write(&namespace_sem); | |
956 | spin_lock(&vfsmount_lock); | |
957 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
958 | change_mnt_propagation(m, type); | |
959 | spin_unlock(&vfsmount_lock); | |
960 | up_write(&namespace_sem); | |
961 | return 0; | |
962 | } | |
963 | ||
1da177e4 LT |
964 | /* |
965 | * do loopback mount. | |
2dafe1c4 | 966 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 967 | */ |
2dafe1c4 ES |
968 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
969 | int recurse) | |
1da177e4 LT |
970 | { |
971 | struct nameidata old_nd; | |
972 | struct vfsmount *mnt = NULL; | |
973 | int err = mount_is_safe(nd); | |
974 | if (err) | |
975 | return err; | |
976 | if (!old_name || !*old_name) | |
977 | return -EINVAL; | |
978 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
979 | if (err) | |
980 | return err; | |
981 | ||
390c6843 | 982 | down_write(&namespace_sem); |
1da177e4 | 983 | err = -EINVAL; |
4ac91378 JB |
984 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
985 | goto out; | |
9676f0c6 | 986 | |
4ac91378 | 987 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 988 | goto out; |
1da177e4 | 989 | |
ccd48bc7 AV |
990 | err = -ENOMEM; |
991 | if (recurse) | |
4ac91378 | 992 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 993 | else |
4ac91378 | 994 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
995 | |
996 | if (!mnt) | |
997 | goto out; | |
998 | ||
ccd48bc7 AV |
999 | err = graft_tree(mnt, nd); |
1000 | if (err) { | |
70fbcdf4 | 1001 | LIST_HEAD(umount_list); |
1da177e4 | 1002 | spin_lock(&vfsmount_lock); |
a05964f3 | 1003 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1004 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1005 | release_mounts(&umount_list); |
5b83d2c5 | 1006 | } |
1da177e4 | 1007 | |
ccd48bc7 | 1008 | out: |
390c6843 | 1009 | up_write(&namespace_sem); |
1d957f9b | 1010 | path_put(&old_nd.path); |
1da177e4 LT |
1011 | return err; |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * change filesystem flags. dir should be a physical root of filesystem. | |
1016 | * If you've mounted a non-root directory somewhere and want to do remount | |
1017 | * on it - tough luck. | |
2dafe1c4 | 1018 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1019 | */ |
2dafe1c4 | 1020 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1021 | void *data) |
1022 | { | |
1023 | int err; | |
4ac91378 | 1024 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1025 | |
1026 | if (!capable(CAP_SYS_ADMIN)) | |
1027 | return -EPERM; | |
1028 | ||
4ac91378 | 1029 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1030 | return -EINVAL; |
1031 | ||
4ac91378 | 1032 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1033 | return -EINVAL; |
1034 | ||
1035 | down_write(&sb->s_umount); | |
1036 | err = do_remount_sb(sb, flags, data, 0); | |
1037 | if (!err) | |
4ac91378 | 1038 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1039 | up_write(&sb->s_umount); |
1040 | if (!err) | |
4ac91378 | 1041 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1042 | return err; |
1043 | } | |
1044 | ||
9676f0c6 RP |
1045 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1046 | { | |
1047 | struct vfsmount *p; | |
1048 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1049 | if (IS_MNT_UNBINDABLE(p)) | |
1050 | return 1; | |
1051 | } | |
1052 | return 0; | |
1053 | } | |
1054 | ||
2dafe1c4 ES |
1055 | /* |
1056 | * noinline this do_mount helper to save do_mount stack space. | |
1057 | */ | |
1058 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 LT |
1059 | { |
1060 | struct nameidata old_nd, parent_nd; | |
1061 | struct vfsmount *p; | |
1062 | int err = 0; | |
1063 | if (!capable(CAP_SYS_ADMIN)) | |
1064 | return -EPERM; | |
1065 | if (!old_name || !*old_name) | |
1066 | return -EINVAL; | |
1067 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1068 | if (err) | |
1069 | return err; | |
1070 | ||
390c6843 | 1071 | down_write(&namespace_sem); |
4ac91378 JB |
1072 | while (d_mountpoint(nd->path.dentry) && |
1073 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1074 | ; |
1075 | err = -EINVAL; | |
4ac91378 | 1076 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1077 | goto out; |
1078 | ||
1079 | err = -ENOENT; | |
4ac91378 JB |
1080 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1081 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1082 | goto out1; |
1083 | ||
4ac91378 | 1084 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1085 | goto out1; |
1da177e4 LT |
1086 | |
1087 | err = -EINVAL; | |
4ac91378 | 1088 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1089 | goto out1; |
1da177e4 | 1090 | |
4ac91378 | 1091 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1092 | goto out1; |
1da177e4 | 1093 | |
4ac91378 JB |
1094 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1095 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1096 | goto out1; |
1097 | /* | |
1098 | * Don't move a mount residing in a shared parent. | |
1099 | */ | |
4ac91378 JB |
1100 | if (old_nd.path.mnt->mnt_parent && |
1101 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1102 | goto out1; |
9676f0c6 RP |
1103 | /* |
1104 | * Don't move a mount tree containing unbindable mounts to a destination | |
1105 | * mount which is shared. | |
1106 | */ | |
4ac91378 JB |
1107 | if (IS_MNT_SHARED(nd->path.mnt) && |
1108 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1109 | goto out1; |
1da177e4 | 1110 | err = -ELOOP; |
4ac91378 JB |
1111 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1112 | if (p == old_nd.path.mnt) | |
21444403 | 1113 | goto out1; |
1da177e4 | 1114 | |
4ac91378 JB |
1115 | err = attach_recursive_mnt(old_nd.path.mnt, nd, &parent_nd); |
1116 | if (err) | |
21444403 | 1117 | goto out1; |
1da177e4 | 1118 | |
21444403 | 1119 | spin_lock(&vfsmount_lock); |
1da177e4 LT |
1120 | /* if the mount is moved, it should no longer be expire |
1121 | * automatically */ | |
4ac91378 | 1122 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 LT |
1123 | spin_unlock(&vfsmount_lock); |
1124 | out1: | |
4ac91378 | 1125 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1126 | out: |
390c6843 | 1127 | up_write(&namespace_sem); |
1da177e4 | 1128 | if (!err) |
1d957f9b JB |
1129 | path_put(&parent_nd.path); |
1130 | path_put(&old_nd.path); | |
1da177e4 LT |
1131 | return err; |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * create a new mount for userspace and request it to be added into the | |
1136 | * namespace's tree | |
2dafe1c4 | 1137 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1138 | */ |
2dafe1c4 | 1139 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1140 | int mnt_flags, char *name, void *data) |
1141 | { | |
1142 | struct vfsmount *mnt; | |
1143 | ||
1144 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1145 | return -EINVAL; | |
1146 | ||
1147 | /* we need capabilities... */ | |
1148 | if (!capable(CAP_SYS_ADMIN)) | |
1149 | return -EPERM; | |
1150 | ||
1151 | mnt = do_kern_mount(type, flags, name, data); | |
1152 | if (IS_ERR(mnt)) | |
1153 | return PTR_ERR(mnt); | |
1154 | ||
1155 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * add a mount into a namespace's mount tree | |
1160 | * - provide the option of adding the new mount to an expiration list | |
1161 | */ | |
1162 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1163 | int mnt_flags, struct list_head *fslist) | |
1164 | { | |
1165 | int err; | |
1166 | ||
390c6843 | 1167 | down_write(&namespace_sem); |
1da177e4 | 1168 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1169 | while (d_mountpoint(nd->path.dentry) && |
1170 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1171 | ; |
1172 | err = -EINVAL; | |
4ac91378 | 1173 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1174 | goto unlock; |
1175 | ||
1176 | /* Refuse the same filesystem on the same mount point */ | |
1177 | err = -EBUSY; | |
4ac91378 JB |
1178 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1179 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1180 | goto unlock; |
1181 | ||
1182 | err = -EINVAL; | |
1183 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1184 | goto unlock; | |
1185 | ||
1186 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1187 | if ((err = graft_tree(newmnt, nd))) |
1188 | goto unlock; | |
1da177e4 | 1189 | |
5b83d2c5 | 1190 | if (fslist) { |
1da177e4 LT |
1191 | /* add to the specified expiration list */ |
1192 | spin_lock(&vfsmount_lock); | |
55e700b9 | 1193 | list_add_tail(&newmnt->mnt_expire, fslist); |
1da177e4 LT |
1194 | spin_unlock(&vfsmount_lock); |
1195 | } | |
390c6843 | 1196 | up_write(&namespace_sem); |
5b83d2c5 | 1197 | return 0; |
1da177e4 LT |
1198 | |
1199 | unlock: | |
390c6843 | 1200 | up_write(&namespace_sem); |
1da177e4 LT |
1201 | mntput(newmnt); |
1202 | return err; | |
1203 | } | |
1204 | ||
1205 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1206 | ||
70fbcdf4 RP |
1207 | static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, |
1208 | struct list_head *umounts) | |
24ca2af1 MS |
1209 | { |
1210 | spin_lock(&vfsmount_lock); | |
1211 | ||
ed42c879 MS |
1212 | /* |
1213 | * Check if mount is still attached, if not, let whoever holds it deal | |
1214 | * with the sucker | |
1215 | */ | |
1216 | if (mnt->mnt_parent == mnt) { | |
1217 | spin_unlock(&vfsmount_lock); | |
1218 | return; | |
1219 | } | |
1220 | ||
24ca2af1 MS |
1221 | /* |
1222 | * Check that it is still dead: the count should now be 2 - as | |
1223 | * contributed by the vfsmount parent and the mntget above | |
1224 | */ | |
a05964f3 | 1225 | if (!propagate_mount_busy(mnt, 2)) { |
24ca2af1 | 1226 | /* delete from the namespace */ |
6b3286ed | 1227 | touch_mnt_namespace(mnt->mnt_ns); |
24ca2af1 | 1228 | list_del_init(&mnt->mnt_list); |
6b3286ed | 1229 | mnt->mnt_ns = NULL; |
a05964f3 | 1230 | umount_tree(mnt, 1, umounts); |
24ca2af1 | 1231 | spin_unlock(&vfsmount_lock); |
24ca2af1 MS |
1232 | } else { |
1233 | /* | |
1234 | * Someone brought it back to life whilst we didn't have any | |
1235 | * locks held so return it to the expiration list | |
1236 | */ | |
55e700b9 | 1237 | list_add_tail(&mnt->mnt_expire, mounts); |
24ca2af1 MS |
1238 | spin_unlock(&vfsmount_lock); |
1239 | } | |
1240 | } | |
1241 | ||
5528f911 TM |
1242 | /* |
1243 | * go through the vfsmounts we've just consigned to the graveyard to | |
1244 | * - check that they're still dead | |
1245 | * - delete the vfsmount from the appropriate namespace under lock | |
1246 | * - dispose of the corpse | |
1247 | */ | |
1248 | static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts) | |
1249 | { | |
6b3286ed | 1250 | struct mnt_namespace *ns; |
5528f911 TM |
1251 | struct vfsmount *mnt; |
1252 | ||
1253 | while (!list_empty(graveyard)) { | |
1254 | LIST_HEAD(umounts); | |
b5e61818 | 1255 | mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire); |
5528f911 TM |
1256 | list_del_init(&mnt->mnt_expire); |
1257 | ||
1258 | /* don't do anything if the namespace is dead - all the | |
1259 | * vfsmounts from it are going away anyway */ | |
6b3286ed KK |
1260 | ns = mnt->mnt_ns; |
1261 | if (!ns || !ns->root) | |
5528f911 | 1262 | continue; |
6b3286ed | 1263 | get_mnt_ns(ns); |
5528f911 TM |
1264 | |
1265 | spin_unlock(&vfsmount_lock); | |
1266 | down_write(&namespace_sem); | |
1267 | expire_mount(mnt, mounts, &umounts); | |
1268 | up_write(&namespace_sem); | |
1269 | release_mounts(&umounts); | |
1270 | mntput(mnt); | |
6b3286ed | 1271 | put_mnt_ns(ns); |
5528f911 TM |
1272 | spin_lock(&vfsmount_lock); |
1273 | } | |
1274 | } | |
1275 | ||
1da177e4 LT |
1276 | /* |
1277 | * process a list of expirable mountpoints with the intent of discarding any | |
1278 | * mountpoints that aren't in use and haven't been touched since last we came | |
1279 | * here | |
1280 | */ | |
1281 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1282 | { | |
1da177e4 LT |
1283 | struct vfsmount *mnt, *next; |
1284 | LIST_HEAD(graveyard); | |
1285 | ||
1286 | if (list_empty(mounts)) | |
1287 | return; | |
1288 | ||
1289 | spin_lock(&vfsmount_lock); | |
1290 | ||
1291 | /* extract from the expiration list every vfsmount that matches the | |
1292 | * following criteria: | |
1293 | * - only referenced by its parent vfsmount | |
1294 | * - still marked for expiry (marked on the last call here; marks are | |
1295 | * cleared by mntput()) | |
1296 | */ | |
55e700b9 | 1297 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 LT |
1298 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1299 | atomic_read(&mnt->mnt_count) != 1) | |
1300 | continue; | |
1301 | ||
1302 | mntget(mnt); | |
55e700b9 | 1303 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 LT |
1304 | } |
1305 | ||
5528f911 | 1306 | expire_mount_list(&graveyard, mounts); |
1da177e4 | 1307 | |
5528f911 TM |
1308 | spin_unlock(&vfsmount_lock); |
1309 | } | |
1310 | ||
1311 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1312 | ||
1313 | /* | |
1314 | * Ripoff of 'select_parent()' | |
1315 | * | |
1316 | * search the list of submounts for a given mountpoint, and move any | |
1317 | * shrinkable submounts to the 'graveyard' list. | |
1318 | */ | |
1319 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1320 | { | |
1321 | struct vfsmount *this_parent = parent; | |
1322 | struct list_head *next; | |
1323 | int found = 0; | |
1324 | ||
1325 | repeat: | |
1326 | next = this_parent->mnt_mounts.next; | |
1327 | resume: | |
1328 | while (next != &this_parent->mnt_mounts) { | |
1329 | struct list_head *tmp = next; | |
1330 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1331 | ||
1332 | next = tmp->next; | |
1333 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1334 | continue; |
5528f911 TM |
1335 | /* |
1336 | * Descend a level if the d_mounts list is non-empty. | |
1337 | */ | |
1338 | if (!list_empty(&mnt->mnt_mounts)) { | |
1339 | this_parent = mnt; | |
1340 | goto repeat; | |
1341 | } | |
1da177e4 | 1342 | |
5528f911 TM |
1343 | if (!propagate_mount_busy(mnt, 1)) { |
1344 | mntget(mnt); | |
1345 | list_move_tail(&mnt->mnt_expire, graveyard); | |
1346 | found++; | |
1347 | } | |
1da177e4 | 1348 | } |
5528f911 TM |
1349 | /* |
1350 | * All done at this level ... ascend and resume the search | |
1351 | */ | |
1352 | if (this_parent != parent) { | |
1353 | next = this_parent->mnt_child.next; | |
1354 | this_parent = this_parent->mnt_parent; | |
1355 | goto resume; | |
1356 | } | |
1357 | return found; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * process a list of expirable mountpoints with the intent of discarding any | |
1362 | * submounts of a specific parent mountpoint | |
1363 | */ | |
1364 | void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts) | |
1365 | { | |
1366 | LIST_HEAD(graveyard); | |
1367 | int found; | |
1368 | ||
1369 | spin_lock(&vfsmount_lock); | |
1370 | ||
1371 | /* extract submounts of 'mountpoint' from the expiration list */ | |
1372 | while ((found = select_submounts(mountpoint, &graveyard)) != 0) | |
1373 | expire_mount_list(&graveyard, mounts); | |
1da177e4 LT |
1374 | |
1375 | spin_unlock(&vfsmount_lock); | |
1376 | } | |
1377 | ||
5528f911 | 1378 | EXPORT_SYMBOL_GPL(shrink_submounts); |
1da177e4 LT |
1379 | |
1380 | /* | |
1381 | * Some copy_from_user() implementations do not return the exact number of | |
1382 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1383 | * Note that this function differs from copy_from_user() in that it will oops | |
1384 | * on bad values of `to', rather than returning a short copy. | |
1385 | */ | |
b58fed8b RP |
1386 | static long exact_copy_from_user(void *to, const void __user * from, |
1387 | unsigned long n) | |
1da177e4 LT |
1388 | { |
1389 | char *t = to; | |
1390 | const char __user *f = from; | |
1391 | char c; | |
1392 | ||
1393 | if (!access_ok(VERIFY_READ, from, n)) | |
1394 | return n; | |
1395 | ||
1396 | while (n) { | |
1397 | if (__get_user(c, f)) { | |
1398 | memset(t, 0, n); | |
1399 | break; | |
1400 | } | |
1401 | *t++ = c; | |
1402 | f++; | |
1403 | n--; | |
1404 | } | |
1405 | return n; | |
1406 | } | |
1407 | ||
b58fed8b | 1408 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1409 | { |
1410 | int i; | |
1411 | unsigned long page; | |
1412 | unsigned long size; | |
b58fed8b | 1413 | |
1da177e4 LT |
1414 | *where = 0; |
1415 | if (!data) | |
1416 | return 0; | |
1417 | ||
1418 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1419 | return -ENOMEM; | |
1420 | ||
1421 | /* We only care that *some* data at the address the user | |
1422 | * gave us is valid. Just in case, we'll zero | |
1423 | * the remainder of the page. | |
1424 | */ | |
1425 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1426 | size = TASK_SIZE - (unsigned long)data; | |
1427 | if (size > PAGE_SIZE) | |
1428 | size = PAGE_SIZE; | |
1429 | ||
1430 | i = size - exact_copy_from_user((void *)page, data, size); | |
1431 | if (!i) { | |
b58fed8b | 1432 | free_page(page); |
1da177e4 LT |
1433 | return -EFAULT; |
1434 | } | |
1435 | if (i != PAGE_SIZE) | |
1436 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1437 | *where = page; | |
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | /* | |
1442 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1443 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1444 | * | |
1445 | * data is a (void *) that can point to any structure up to | |
1446 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1447 | * information (or be NULL). | |
1448 | * | |
1449 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1450 | * When the flags word was introduced its top half was required | |
1451 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1452 | * Therefore, if this magic number is present, it carries no information | |
1453 | * and must be discarded. | |
1454 | */ | |
b58fed8b | 1455 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1456 | unsigned long flags, void *data_page) |
1457 | { | |
1458 | struct nameidata nd; | |
1459 | int retval = 0; | |
1460 | int mnt_flags = 0; | |
1461 | ||
1462 | /* Discard magic */ | |
1463 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1464 | flags &= ~MS_MGC_MSK; | |
1465 | ||
1466 | /* Basic sanity checks */ | |
1467 | ||
1468 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1469 | return -EINVAL; | |
1470 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1471 | return -EINVAL; | |
1472 | ||
1473 | if (data_page) | |
1474 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1475 | ||
1476 | /* Separate the per-mountpoint flags */ | |
1477 | if (flags & MS_NOSUID) | |
1478 | mnt_flags |= MNT_NOSUID; | |
1479 | if (flags & MS_NODEV) | |
1480 | mnt_flags |= MNT_NODEV; | |
1481 | if (flags & MS_NOEXEC) | |
1482 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1483 | if (flags & MS_NOATIME) |
1484 | mnt_flags |= MNT_NOATIME; | |
1485 | if (flags & MS_NODIRATIME) | |
1486 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1487 | if (flags & MS_RELATIME) |
1488 | mnt_flags |= MNT_RELATIME; | |
fc33a7bb CH |
1489 | |
1490 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1491 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1492 | |
1493 | /* ... and get the mountpoint */ | |
1494 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1495 | if (retval) | |
1496 | return retval; | |
1497 | ||
1498 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1499 | if (retval) | |
1500 | goto dput_out; | |
1501 | ||
1502 | if (flags & MS_REMOUNT) | |
1503 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1504 | data_page); | |
1505 | else if (flags & MS_BIND) | |
eee391a6 | 1506 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1507 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1508 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1509 | else if (flags & MS_MOVE) |
1510 | retval = do_move_mount(&nd, dev_name); | |
1511 | else | |
1512 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1513 | dev_name, data_page); | |
1514 | dput_out: | |
1d957f9b | 1515 | path_put(&nd.path); |
1da177e4 LT |
1516 | return retval; |
1517 | } | |
1518 | ||
741a2951 JD |
1519 | /* |
1520 | * Allocate a new namespace structure and populate it with contents | |
1521 | * copied from the namespace of the passed in task structure. | |
1522 | */ | |
e3222c4e | 1523 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1524 | struct fs_struct *fs) |
1da177e4 | 1525 | { |
6b3286ed | 1526 | struct mnt_namespace *new_ns; |
1da177e4 | 1527 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1528 | struct vfsmount *p, *q; |
1529 | ||
6b3286ed | 1530 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1531 | if (!new_ns) |
467e9f4b | 1532 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1533 | |
1534 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1535 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1536 | init_waitqueue_head(&new_ns->poll); |
1537 | new_ns->event = 0; | |
1da177e4 | 1538 | |
390c6843 | 1539 | down_write(&namespace_sem); |
1da177e4 | 1540 | /* First pass: copy the tree topology */ |
6b3286ed | 1541 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1542 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1543 | if (!new_ns->root) { |
390c6843 | 1544 | up_write(&namespace_sem); |
1da177e4 | 1545 | kfree(new_ns); |
467e9f4b | 1546 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1547 | } |
1548 | spin_lock(&vfsmount_lock); | |
1549 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1550 | spin_unlock(&vfsmount_lock); | |
1551 | ||
1552 | /* | |
1553 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1554 | * as belonging to new namespace. We have already acquired a private | |
1555 | * fs_struct, so tsk->fs->lock is not needed. | |
1556 | */ | |
6b3286ed | 1557 | p = mnt_ns->root; |
1da177e4 LT |
1558 | q = new_ns->root; |
1559 | while (p) { | |
6b3286ed | 1560 | q->mnt_ns = new_ns; |
1da177e4 | 1561 | if (fs) { |
6ac08c39 | 1562 | if (p == fs->root.mnt) { |
1da177e4 | 1563 | rootmnt = p; |
6ac08c39 | 1564 | fs->root.mnt = mntget(q); |
1da177e4 | 1565 | } |
6ac08c39 | 1566 | if (p == fs->pwd.mnt) { |
1da177e4 | 1567 | pwdmnt = p; |
6ac08c39 | 1568 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1569 | } |
6ac08c39 | 1570 | if (p == fs->altroot.mnt) { |
1da177e4 | 1571 | altrootmnt = p; |
6ac08c39 | 1572 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1573 | } |
1574 | } | |
6b3286ed | 1575 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1576 | q = next_mnt(q, new_ns->root); |
1577 | } | |
390c6843 | 1578 | up_write(&namespace_sem); |
1da177e4 | 1579 | |
1da177e4 LT |
1580 | if (rootmnt) |
1581 | mntput(rootmnt); | |
1582 | if (pwdmnt) | |
1583 | mntput(pwdmnt); | |
1584 | if (altrootmnt) | |
1585 | mntput(altrootmnt); | |
1586 | ||
741a2951 JD |
1587 | return new_ns; |
1588 | } | |
1589 | ||
213dd266 | 1590 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1591 | struct fs_struct *new_fs) |
741a2951 | 1592 | { |
6b3286ed | 1593 | struct mnt_namespace *new_ns; |
741a2951 | 1594 | |
e3222c4e | 1595 | BUG_ON(!ns); |
6b3286ed | 1596 | get_mnt_ns(ns); |
741a2951 JD |
1597 | |
1598 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1599 | return ns; |
741a2951 | 1600 | |
e3222c4e | 1601 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1602 | |
6b3286ed | 1603 | put_mnt_ns(ns); |
e3222c4e | 1604 | return new_ns; |
1da177e4 LT |
1605 | } |
1606 | ||
1607 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1608 | char __user * type, unsigned long flags, | |
1609 | void __user * data) | |
1610 | { | |
1611 | int retval; | |
1612 | unsigned long data_page; | |
1613 | unsigned long type_page; | |
1614 | unsigned long dev_page; | |
1615 | char *dir_page; | |
1616 | ||
b58fed8b | 1617 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1618 | if (retval < 0) |
1619 | return retval; | |
1620 | ||
1621 | dir_page = getname(dir_name); | |
1622 | retval = PTR_ERR(dir_page); | |
1623 | if (IS_ERR(dir_page)) | |
1624 | goto out1; | |
1625 | ||
b58fed8b | 1626 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1627 | if (retval < 0) |
1628 | goto out2; | |
1629 | ||
b58fed8b | 1630 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1631 | if (retval < 0) |
1632 | goto out3; | |
1633 | ||
1634 | lock_kernel(); | |
b58fed8b RP |
1635 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1636 | flags, (void *)data_page); | |
1da177e4 LT |
1637 | unlock_kernel(); |
1638 | free_page(data_page); | |
1639 | ||
1640 | out3: | |
1641 | free_page(dev_page); | |
1642 | out2: | |
1643 | putname(dir_page); | |
1644 | out1: | |
1645 | free_page(type_page); | |
1646 | return retval; | |
1647 | } | |
1648 | ||
1649 | /* | |
1650 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1651 | * It can block. Requires the big lock held. | |
1652 | */ | |
ac748a09 | 1653 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 1654 | { |
6ac08c39 JB |
1655 | struct path old_root; |
1656 | ||
1da177e4 LT |
1657 | write_lock(&fs->lock); |
1658 | old_root = fs->root; | |
ac748a09 JB |
1659 | fs->root = *path; |
1660 | path_get(path); | |
1da177e4 | 1661 | write_unlock(&fs->lock); |
6ac08c39 JB |
1662 | if (old_root.dentry) |
1663 | path_put(&old_root); | |
1da177e4 LT |
1664 | } |
1665 | ||
1666 | /* | |
1667 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1668 | * It can block. Requires the big lock held. | |
1669 | */ | |
ac748a09 | 1670 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 1671 | { |
6ac08c39 | 1672 | struct path old_pwd; |
1da177e4 LT |
1673 | |
1674 | write_lock(&fs->lock); | |
1675 | old_pwd = fs->pwd; | |
ac748a09 JB |
1676 | fs->pwd = *path; |
1677 | path_get(path); | |
1da177e4 LT |
1678 | write_unlock(&fs->lock); |
1679 | ||
6ac08c39 JB |
1680 | if (old_pwd.dentry) |
1681 | path_put(&old_pwd); | |
1da177e4 LT |
1682 | } |
1683 | ||
1684 | static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) | |
1685 | { | |
1686 | struct task_struct *g, *p; | |
1687 | struct fs_struct *fs; | |
1688 | ||
1689 | read_lock(&tasklist_lock); | |
1690 | do_each_thread(g, p) { | |
1691 | task_lock(p); | |
1692 | fs = p->fs; | |
1693 | if (fs) { | |
1694 | atomic_inc(&fs->count); | |
1695 | task_unlock(p); | |
6ac08c39 JB |
1696 | if (fs->root.dentry == old_nd->path.dentry |
1697 | && fs->root.mnt == old_nd->path.mnt) | |
ac748a09 | 1698 | set_fs_root(fs, &new_nd->path); |
6ac08c39 JB |
1699 | if (fs->pwd.dentry == old_nd->path.dentry |
1700 | && fs->pwd.mnt == old_nd->path.mnt) | |
ac748a09 | 1701 | set_fs_pwd(fs, &new_nd->path); |
1da177e4 LT |
1702 | put_fs_struct(fs); |
1703 | } else | |
1704 | task_unlock(p); | |
1705 | } while_each_thread(g, p); | |
1706 | read_unlock(&tasklist_lock); | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * pivot_root Semantics: | |
1711 | * Moves the root file system of the current process to the directory put_old, | |
1712 | * makes new_root as the new root file system of the current process, and sets | |
1713 | * root/cwd of all processes which had them on the current root to new_root. | |
1714 | * | |
1715 | * Restrictions: | |
1716 | * The new_root and put_old must be directories, and must not be on the | |
1717 | * same file system as the current process root. The put_old must be | |
1718 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1719 | * pointed to by put_old must yield the same directory as new_root. No other | |
1720 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1721 | * | |
4a0d11fa NB |
1722 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
1723 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1724 | * in this situation. | |
1725 | * | |
1da177e4 LT |
1726 | * Notes: |
1727 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1728 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1729 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1730 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1731 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1732 | * first. | |
1733 | */ | |
b58fed8b RP |
1734 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1735 | const char __user * put_old) | |
1da177e4 LT |
1736 | { |
1737 | struct vfsmount *tmp; | |
1738 | struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; | |
1739 | int error; | |
1740 | ||
1741 | if (!capable(CAP_SYS_ADMIN)) | |
1742 | return -EPERM; | |
1743 | ||
1744 | lock_kernel(); | |
1745 | ||
b58fed8b RP |
1746 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1747 | &new_nd); | |
1da177e4 LT |
1748 | if (error) |
1749 | goto out0; | |
1750 | error = -EINVAL; | |
4ac91378 | 1751 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
1752 | goto out1; |
1753 | ||
b58fed8b | 1754 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1755 | if (error) |
1756 | goto out1; | |
1757 | ||
1758 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1759 | if (error) { | |
1d957f9b | 1760 | path_put(&old_nd.path); |
1da177e4 LT |
1761 | goto out1; |
1762 | } | |
1763 | ||
1764 | read_lock(¤t->fs->lock); | |
6ac08c39 JB |
1765 | user_nd.path = current->fs->root; |
1766 | path_get(¤t->fs->root); | |
1da177e4 | 1767 | read_unlock(¤t->fs->lock); |
390c6843 | 1768 | down_write(&namespace_sem); |
4ac91378 | 1769 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 1770 | error = -EINVAL; |
4ac91378 JB |
1771 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
1772 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
1773 | IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) | |
21444403 | 1774 | goto out2; |
4ac91378 | 1775 | if (!check_mnt(user_nd.path.mnt)) |
1da177e4 LT |
1776 | goto out2; |
1777 | error = -ENOENT; | |
4ac91378 | 1778 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 1779 | goto out2; |
4ac91378 | 1780 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 1781 | goto out2; |
4ac91378 | 1782 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
1783 | goto out2; |
1784 | error = -EBUSY; | |
4ac91378 JB |
1785 | if (new_nd.path.mnt == user_nd.path.mnt || |
1786 | old_nd.path.mnt == user_nd.path.mnt) | |
1da177e4 LT |
1787 | goto out2; /* loop, on the same file system */ |
1788 | error = -EINVAL; | |
4ac91378 | 1789 | if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) |
1da177e4 | 1790 | goto out2; /* not a mountpoint */ |
4ac91378 | 1791 | if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) |
0bb6fcc1 | 1792 | goto out2; /* not attached */ |
4ac91378 | 1793 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 1794 | goto out2; /* not a mountpoint */ |
4ac91378 | 1795 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 1796 | goto out2; /* not attached */ |
4ac91378 JB |
1797 | /* make sure we can reach put_old from new_root */ |
1798 | tmp = old_nd.path.mnt; | |
1da177e4 | 1799 | spin_lock(&vfsmount_lock); |
4ac91378 | 1800 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
1801 | for (;;) { |
1802 | if (tmp->mnt_parent == tmp) | |
1803 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 1804 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
1805 | break; |
1806 | tmp = tmp->mnt_parent; | |
1807 | } | |
4ac91378 | 1808 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 1809 | goto out3; |
4ac91378 | 1810 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 1811 | goto out3; |
4ac91378 JB |
1812 | detach_mnt(new_nd.path.mnt, &parent_nd); |
1813 | detach_mnt(user_nd.path.mnt, &root_parent); | |
1814 | /* mount old root on put_old */ | |
1815 | attach_mnt(user_nd.path.mnt, &old_nd); | |
1816 | /* mount new_root on / */ | |
1817 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 1818 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 LT |
1819 | spin_unlock(&vfsmount_lock); |
1820 | chroot_fs_refs(&user_nd, &new_nd); | |
1821 | security_sb_post_pivotroot(&user_nd, &new_nd); | |
1822 | error = 0; | |
1d957f9b JB |
1823 | path_put(&root_parent.path); |
1824 | path_put(&parent_nd.path); | |
1da177e4 | 1825 | out2: |
4ac91378 | 1826 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 1827 | up_write(&namespace_sem); |
1d957f9b JB |
1828 | path_put(&user_nd.path); |
1829 | path_put(&old_nd.path); | |
1da177e4 | 1830 | out1: |
1d957f9b | 1831 | path_put(&new_nd.path); |
1da177e4 LT |
1832 | out0: |
1833 | unlock_kernel(); | |
1834 | return error; | |
1835 | out3: | |
1836 | spin_unlock(&vfsmount_lock); | |
1837 | goto out2; | |
1838 | } | |
1839 | ||
1840 | static void __init init_mount_tree(void) | |
1841 | { | |
1842 | struct vfsmount *mnt; | |
6b3286ed | 1843 | struct mnt_namespace *ns; |
ac748a09 | 1844 | struct path root; |
1da177e4 LT |
1845 | |
1846 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1847 | if (IS_ERR(mnt)) | |
1848 | panic("Can't create rootfs"); | |
6b3286ed KK |
1849 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
1850 | if (!ns) | |
1da177e4 | 1851 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
1852 | atomic_set(&ns->count, 1); |
1853 | INIT_LIST_HEAD(&ns->list); | |
1854 | init_waitqueue_head(&ns->poll); | |
1855 | ns->event = 0; | |
1856 | list_add(&mnt->mnt_list, &ns->list); | |
1857 | ns->root = mnt; | |
1858 | mnt->mnt_ns = ns; | |
1859 | ||
1860 | init_task.nsproxy->mnt_ns = ns; | |
1861 | get_mnt_ns(ns); | |
1862 | ||
ac748a09 JB |
1863 | root.mnt = ns->root; |
1864 | root.dentry = ns->root->mnt_root; | |
1865 | ||
1866 | set_fs_pwd(current->fs, &root); | |
1867 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
1868 | } |
1869 | ||
74bf17cf | 1870 | void __init mnt_init(void) |
1da177e4 | 1871 | { |
13f14b4d | 1872 | unsigned u; |
15a67dd8 | 1873 | int err; |
1da177e4 | 1874 | |
390c6843 RP |
1875 | init_rwsem(&namespace_sem); |
1876 | ||
1da177e4 | 1877 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 1878 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 1879 | |
b58fed8b | 1880 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1881 | |
1882 | if (!mount_hashtable) | |
1883 | panic("Failed to allocate mount hash table\n"); | |
1884 | ||
13f14b4d ED |
1885 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
1886 | ||
1887 | for (u = 0; u < HASH_SIZE; u++) | |
1888 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 1889 | |
15a67dd8 RD |
1890 | err = sysfs_init(); |
1891 | if (err) | |
1892 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
1893 | __FUNCTION__, err); | |
00d26666 GKH |
1894 | fs_kobj = kobject_create_and_add("fs", NULL); |
1895 | if (!fs_kobj) | |
1896 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
1897 | init_rootfs(); |
1898 | init_mount_tree(); | |
1899 | } | |
1900 | ||
6b3286ed | 1901 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 1902 | { |
6b3286ed | 1903 | struct vfsmount *root = ns->root; |
70fbcdf4 | 1904 | LIST_HEAD(umount_list); |
6b3286ed | 1905 | ns->root = NULL; |
1ce88cf4 | 1906 | spin_unlock(&vfsmount_lock); |
390c6843 | 1907 | down_write(&namespace_sem); |
1da177e4 | 1908 | spin_lock(&vfsmount_lock); |
a05964f3 | 1909 | umount_tree(root, 0, &umount_list); |
1da177e4 | 1910 | spin_unlock(&vfsmount_lock); |
390c6843 | 1911 | up_write(&namespace_sem); |
70fbcdf4 | 1912 | release_mounts(&umount_list); |
6b3286ed | 1913 | kfree(ns); |
1da177e4 | 1914 | } |