]>
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 | 17 | #include <linux/acct.h> |
16f7e0fe | 18 | #include <linux/capability.h> |
3d733633 | 19 | #include <linux/cpumask.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> |
73cd49ec | 29 | #include <linux/idr.h> |
5ad4e53b | 30 | #include <linux/fs_struct.h> |
1da177e4 LT |
31 | #include <asm/uaccess.h> |
32 | #include <asm/unistd.h> | |
07b20889 | 33 | #include "pnode.h" |
948730b0 | 34 | #include "internal.h" |
1da177e4 | 35 | |
13f14b4d ED |
36 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
37 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
38 | ||
1da177e4 | 39 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
40 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
41 | ||
42 | static int event; | |
73cd49ec | 43 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 44 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 45 | |
fa3536cc | 46 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 47 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 48 | static struct rw_semaphore namespace_sem; |
1da177e4 | 49 | |
f87fd4c2 | 50 | /* /sys/fs */ |
00d26666 GKH |
51 | struct kobject *fs_kobj; |
52 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 53 | |
1da177e4 LT |
54 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
55 | { | |
b58fed8b RP |
56 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
57 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
58 | tmp = tmp + (tmp >> HASH_SHIFT); |
59 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
60 | } |
61 | ||
3d733633 DH |
62 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
63 | ||
73cd49ec MS |
64 | /* allocation is serialized by namespace_sem */ |
65 | static int mnt_alloc_id(struct vfsmount *mnt) | |
66 | { | |
67 | int res; | |
68 | ||
69 | retry: | |
70 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
71 | spin_lock(&vfsmount_lock); | |
72 | res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); | |
73 | spin_unlock(&vfsmount_lock); | |
74 | if (res == -EAGAIN) | |
75 | goto retry; | |
76 | ||
77 | return res; | |
78 | } | |
79 | ||
80 | static void mnt_free_id(struct vfsmount *mnt) | |
81 | { | |
82 | spin_lock(&vfsmount_lock); | |
83 | ida_remove(&mnt_id_ida, mnt->mnt_id); | |
84 | spin_unlock(&vfsmount_lock); | |
85 | } | |
86 | ||
719f5d7f MS |
87 | /* |
88 | * Allocate a new peer group ID | |
89 | * | |
90 | * mnt_group_ida is protected by namespace_sem | |
91 | */ | |
92 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
93 | { | |
94 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) | |
95 | return -ENOMEM; | |
96 | ||
97 | return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id); | |
98 | } | |
99 | ||
100 | /* | |
101 | * Release a peer group ID | |
102 | */ | |
103 | void mnt_release_group_id(struct vfsmount *mnt) | |
104 | { | |
105 | ida_remove(&mnt_group_ida, mnt->mnt_group_id); | |
106 | mnt->mnt_group_id = 0; | |
107 | } | |
108 | ||
1da177e4 LT |
109 | struct vfsmount *alloc_vfsmnt(const char *name) |
110 | { | |
c3762229 | 111 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 112 | if (mnt) { |
73cd49ec MS |
113 | int err; |
114 | ||
115 | err = mnt_alloc_id(mnt); | |
88b38782 LZ |
116 | if (err) |
117 | goto out_free_cache; | |
118 | ||
119 | if (name) { | |
120 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
121 | if (!mnt->mnt_devname) | |
122 | goto out_free_id; | |
73cd49ec MS |
123 | } |
124 | ||
b58fed8b | 125 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
126 | INIT_LIST_HEAD(&mnt->mnt_hash); |
127 | INIT_LIST_HEAD(&mnt->mnt_child); | |
128 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
129 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 130 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 131 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
132 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
133 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
d3ef3d73 | 134 | #ifdef CONFIG_SMP |
135 | mnt->mnt_writers = alloc_percpu(int); | |
136 | if (!mnt->mnt_writers) | |
137 | goto out_free_devname; | |
138 | #else | |
139 | mnt->mnt_writers = 0; | |
140 | #endif | |
1da177e4 LT |
141 | } |
142 | return mnt; | |
88b38782 | 143 | |
d3ef3d73 | 144 | #ifdef CONFIG_SMP |
145 | out_free_devname: | |
146 | kfree(mnt->mnt_devname); | |
147 | #endif | |
88b38782 LZ |
148 | out_free_id: |
149 | mnt_free_id(mnt); | |
150 | out_free_cache: | |
151 | kmem_cache_free(mnt_cache, mnt); | |
152 | return NULL; | |
1da177e4 LT |
153 | } |
154 | ||
3d733633 DH |
155 | /* |
156 | * Most r/o checks on a fs are for operations that take | |
157 | * discrete amounts of time, like a write() or unlink(). | |
158 | * We must keep track of when those operations start | |
159 | * (for permission checks) and when they end, so that | |
160 | * we can determine when writes are able to occur to | |
161 | * a filesystem. | |
162 | */ | |
163 | /* | |
164 | * __mnt_is_readonly: check whether a mount is read-only | |
165 | * @mnt: the mount to check for its write status | |
166 | * | |
167 | * This shouldn't be used directly ouside of the VFS. | |
168 | * It does not guarantee that the filesystem will stay | |
169 | * r/w, just that it is right *now*. This can not and | |
170 | * should not be used in place of IS_RDONLY(inode). | |
171 | * mnt_want/drop_write() will _keep_ the filesystem | |
172 | * r/w. | |
173 | */ | |
174 | int __mnt_is_readonly(struct vfsmount *mnt) | |
175 | { | |
2e4b7fcd DH |
176 | if (mnt->mnt_flags & MNT_READONLY) |
177 | return 1; | |
178 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
179 | return 1; | |
180 | return 0; | |
3d733633 DH |
181 | } |
182 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
183 | ||
d3ef3d73 | 184 | static inline void inc_mnt_writers(struct vfsmount *mnt) |
185 | { | |
186 | #ifdef CONFIG_SMP | |
187 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++; | |
188 | #else | |
189 | mnt->mnt_writers++; | |
190 | #endif | |
191 | } | |
3d733633 | 192 | |
d3ef3d73 | 193 | static inline void dec_mnt_writers(struct vfsmount *mnt) |
3d733633 | 194 | { |
d3ef3d73 | 195 | #ifdef CONFIG_SMP |
196 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--; | |
197 | #else | |
198 | mnt->mnt_writers--; | |
199 | #endif | |
3d733633 | 200 | } |
3d733633 | 201 | |
d3ef3d73 | 202 | static unsigned int count_mnt_writers(struct vfsmount *mnt) |
3d733633 | 203 | { |
d3ef3d73 | 204 | #ifdef CONFIG_SMP |
205 | unsigned int count = 0; | |
3d733633 | 206 | int cpu; |
3d733633 DH |
207 | |
208 | for_each_possible_cpu(cpu) { | |
d3ef3d73 | 209 | count += *per_cpu_ptr(mnt->mnt_writers, cpu); |
3d733633 | 210 | } |
3d733633 | 211 | |
d3ef3d73 | 212 | return count; |
213 | #else | |
214 | return mnt->mnt_writers; | |
215 | #endif | |
3d733633 DH |
216 | } |
217 | ||
8366025e DH |
218 | /* |
219 | * Most r/o checks on a fs are for operations that take | |
220 | * discrete amounts of time, like a write() or unlink(). | |
221 | * We must keep track of when those operations start | |
222 | * (for permission checks) and when they end, so that | |
223 | * we can determine when writes are able to occur to | |
224 | * a filesystem. | |
225 | */ | |
226 | /** | |
227 | * mnt_want_write - get write access to a mount | |
228 | * @mnt: the mount on which to take a write | |
229 | * | |
230 | * This tells the low-level filesystem that a write is | |
231 | * about to be performed to it, and makes sure that | |
232 | * writes are allowed before returning success. When | |
233 | * the write operation is finished, mnt_drop_write() | |
234 | * must be called. This is effectively a refcount. | |
235 | */ | |
236 | int mnt_want_write(struct vfsmount *mnt) | |
237 | { | |
3d733633 | 238 | int ret = 0; |
3d733633 | 239 | |
d3ef3d73 | 240 | preempt_disable(); |
241 | inc_mnt_writers(mnt); | |
242 | /* | |
243 | * The store to inc_mnt_writers must be visible before we pass | |
244 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our | |
245 | * incremented count after it has set MNT_WRITE_HOLD. | |
246 | */ | |
247 | smp_mb(); | |
248 | while (mnt->mnt_flags & MNT_WRITE_HOLD) | |
249 | cpu_relax(); | |
250 | /* | |
251 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
252 | * be set to match its requirements. So we must not load that until | |
253 | * MNT_WRITE_HOLD is cleared. | |
254 | */ | |
255 | smp_rmb(); | |
3d733633 | 256 | if (__mnt_is_readonly(mnt)) { |
d3ef3d73 | 257 | dec_mnt_writers(mnt); |
3d733633 DH |
258 | ret = -EROFS; |
259 | goto out; | |
260 | } | |
3d733633 | 261 | out: |
d3ef3d73 | 262 | preempt_enable(); |
3d733633 | 263 | return ret; |
8366025e DH |
264 | } |
265 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
266 | ||
267 | /** | |
268 | * mnt_drop_write - give up write access to a mount | |
269 | * @mnt: the mount on which to give up write access | |
270 | * | |
271 | * Tells the low-level filesystem that we are done | |
272 | * performing writes to it. Must be matched with | |
273 | * mnt_want_write() call above. | |
274 | */ | |
275 | void mnt_drop_write(struct vfsmount *mnt) | |
276 | { | |
d3ef3d73 | 277 | preempt_disable(); |
278 | dec_mnt_writers(mnt); | |
279 | preempt_enable(); | |
8366025e DH |
280 | } |
281 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
282 | ||
2e4b7fcd | 283 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 284 | { |
3d733633 DH |
285 | int ret = 0; |
286 | ||
d3ef3d73 | 287 | spin_lock(&vfsmount_lock); |
288 | mnt->mnt_flags |= MNT_WRITE_HOLD; | |
3d733633 | 289 | /* |
d3ef3d73 | 290 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
291 | * should be visible before we do. | |
3d733633 | 292 | */ |
d3ef3d73 | 293 | smp_mb(); |
294 | ||
3d733633 | 295 | /* |
d3ef3d73 | 296 | * With writers on hold, if this value is zero, then there are |
297 | * definitely no active writers (although held writers may subsequently | |
298 | * increment the count, they'll have to wait, and decrement it after | |
299 | * seeing MNT_READONLY). | |
300 | * | |
301 | * It is OK to have counter incremented on one CPU and decremented on | |
302 | * another: the sum will add up correctly. The danger would be when we | |
303 | * sum up each counter, if we read a counter before it is incremented, | |
304 | * but then read another CPU's count which it has been subsequently | |
305 | * decremented from -- we would see more decrements than we should. | |
306 | * MNT_WRITE_HOLD protects against this scenario, because | |
307 | * mnt_want_write first increments count, then smp_mb, then spins on | |
308 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
309 | * we're counting up here. | |
3d733633 | 310 | */ |
d3ef3d73 | 311 | if (count_mnt_writers(mnt) > 0) |
312 | ret = -EBUSY; | |
313 | else | |
2e4b7fcd | 314 | mnt->mnt_flags |= MNT_READONLY; |
d3ef3d73 | 315 | /* |
316 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
317 | * that become unheld will see MNT_READONLY. | |
318 | */ | |
319 | smp_wmb(); | |
320 | mnt->mnt_flags &= ~MNT_WRITE_HOLD; | |
2e4b7fcd | 321 | spin_unlock(&vfsmount_lock); |
3d733633 | 322 | return ret; |
8366025e | 323 | } |
8366025e | 324 | |
2e4b7fcd DH |
325 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
326 | { | |
327 | spin_lock(&vfsmount_lock); | |
328 | mnt->mnt_flags &= ~MNT_READONLY; | |
329 | spin_unlock(&vfsmount_lock); | |
330 | } | |
331 | ||
a3ec947c | 332 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
454e2398 DH |
333 | { |
334 | mnt->mnt_sb = sb; | |
335 | mnt->mnt_root = dget(sb->s_root); | |
454e2398 DH |
336 | } |
337 | ||
338 | EXPORT_SYMBOL(simple_set_mnt); | |
339 | ||
1da177e4 LT |
340 | void free_vfsmnt(struct vfsmount *mnt) |
341 | { | |
342 | kfree(mnt->mnt_devname); | |
73cd49ec | 343 | mnt_free_id(mnt); |
d3ef3d73 | 344 | #ifdef CONFIG_SMP |
345 | free_percpu(mnt->mnt_writers); | |
346 | #endif | |
1da177e4 LT |
347 | kmem_cache_free(mnt_cache, mnt); |
348 | } | |
349 | ||
350 | /* | |
a05964f3 RP |
351 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
352 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 353 | */ |
a05964f3 RP |
354 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
355 | int dir) | |
1da177e4 | 356 | { |
b58fed8b RP |
357 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
358 | struct list_head *tmp = head; | |
1da177e4 LT |
359 | struct vfsmount *p, *found = NULL; |
360 | ||
1da177e4 | 361 | for (;;) { |
a05964f3 | 362 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
363 | p = NULL; |
364 | if (tmp == head) | |
365 | break; | |
366 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
367 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 368 | found = p; |
1da177e4 LT |
369 | break; |
370 | } | |
371 | } | |
1da177e4 LT |
372 | return found; |
373 | } | |
374 | ||
a05964f3 RP |
375 | /* |
376 | * lookup_mnt increments the ref count before returning | |
377 | * the vfsmount struct. | |
378 | */ | |
1c755af4 | 379 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 RP |
380 | { |
381 | struct vfsmount *child_mnt; | |
382 | spin_lock(&vfsmount_lock); | |
1c755af4 | 383 | if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) |
a05964f3 RP |
384 | mntget(child_mnt); |
385 | spin_unlock(&vfsmount_lock); | |
386 | return child_mnt; | |
387 | } | |
388 | ||
1da177e4 LT |
389 | static inline int check_mnt(struct vfsmount *mnt) |
390 | { | |
6b3286ed | 391 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
392 | } |
393 | ||
6b3286ed | 394 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
395 | { |
396 | if (ns) { | |
397 | ns->event = ++event; | |
398 | wake_up_interruptible(&ns->poll); | |
399 | } | |
400 | } | |
401 | ||
6b3286ed | 402 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
403 | { |
404 | if (ns && ns->event != event) { | |
405 | ns->event = event; | |
406 | wake_up_interruptible(&ns->poll); | |
407 | } | |
408 | } | |
409 | ||
1a390689 | 410 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 411 | { |
1a390689 AV |
412 | old_path->dentry = mnt->mnt_mountpoint; |
413 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
414 | mnt->mnt_parent = mnt; |
415 | mnt->mnt_mountpoint = mnt->mnt_root; | |
416 | list_del_init(&mnt->mnt_child); | |
417 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 418 | old_path->dentry->d_mounted--; |
1da177e4 LT |
419 | } |
420 | ||
b90fa9ae RP |
421 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
422 | struct vfsmount *child_mnt) | |
423 | { | |
424 | child_mnt->mnt_parent = mntget(mnt); | |
425 | child_mnt->mnt_mountpoint = dget(dentry); | |
426 | dentry->d_mounted++; | |
427 | } | |
428 | ||
1a390689 | 429 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 430 | { |
1a390689 | 431 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 432 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
433 | hash(path->mnt, path->dentry)); |
434 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
435 | } |
436 | ||
437 | /* | |
438 | * the caller must hold vfsmount_lock | |
439 | */ | |
440 | static void commit_tree(struct vfsmount *mnt) | |
441 | { | |
442 | struct vfsmount *parent = mnt->mnt_parent; | |
443 | struct vfsmount *m; | |
444 | LIST_HEAD(head); | |
6b3286ed | 445 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
446 | |
447 | BUG_ON(parent == mnt); | |
448 | ||
449 | list_add_tail(&head, &mnt->mnt_list); | |
450 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 451 | m->mnt_ns = n; |
b90fa9ae RP |
452 | list_splice(&head, n->list.prev); |
453 | ||
454 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
455 | hash(parent, mnt->mnt_mountpoint)); | |
456 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 457 | touch_mnt_namespace(n); |
1da177e4 LT |
458 | } |
459 | ||
460 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
461 | { | |
462 | struct list_head *next = p->mnt_mounts.next; | |
463 | if (next == &p->mnt_mounts) { | |
464 | while (1) { | |
465 | if (p == root) | |
466 | return NULL; | |
467 | next = p->mnt_child.next; | |
468 | if (next != &p->mnt_parent->mnt_mounts) | |
469 | break; | |
470 | p = p->mnt_parent; | |
471 | } | |
472 | } | |
473 | return list_entry(next, struct vfsmount, mnt_child); | |
474 | } | |
475 | ||
9676f0c6 RP |
476 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
477 | { | |
478 | struct list_head *prev = p->mnt_mounts.prev; | |
479 | while (prev != &p->mnt_mounts) { | |
480 | p = list_entry(prev, struct vfsmount, mnt_child); | |
481 | prev = p->mnt_mounts.prev; | |
482 | } | |
483 | return p; | |
484 | } | |
485 | ||
36341f64 RP |
486 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
487 | int flag) | |
1da177e4 LT |
488 | { |
489 | struct super_block *sb = old->mnt_sb; | |
490 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
491 | ||
492 | if (mnt) { | |
719f5d7f MS |
493 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
494 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
495 | else | |
496 | mnt->mnt_group_id = old->mnt_group_id; | |
497 | ||
498 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
499 | int err = mnt_alloc_group_id(mnt); | |
500 | if (err) | |
501 | goto out_free; | |
502 | } | |
503 | ||
1da177e4 LT |
504 | mnt->mnt_flags = old->mnt_flags; |
505 | atomic_inc(&sb->s_active); | |
506 | mnt->mnt_sb = sb; | |
507 | mnt->mnt_root = dget(root); | |
508 | mnt->mnt_mountpoint = mnt->mnt_root; | |
509 | mnt->mnt_parent = mnt; | |
b90fa9ae | 510 | |
5afe0022 RP |
511 | if (flag & CL_SLAVE) { |
512 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
513 | mnt->mnt_master = old; | |
514 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 515 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
516 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
517 | list_add(&mnt->mnt_share, &old->mnt_share); | |
518 | if (IS_MNT_SLAVE(old)) | |
519 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
520 | mnt->mnt_master = old->mnt_master; | |
521 | } | |
b90fa9ae RP |
522 | if (flag & CL_MAKE_SHARED) |
523 | set_mnt_shared(mnt); | |
1da177e4 LT |
524 | |
525 | /* stick the duplicate mount on the same expiry list | |
526 | * as the original if that was on one */ | |
36341f64 | 527 | if (flag & CL_EXPIRE) { |
36341f64 RP |
528 | if (!list_empty(&old->mnt_expire)) |
529 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 530 | } |
1da177e4 LT |
531 | } |
532 | return mnt; | |
719f5d7f MS |
533 | |
534 | out_free: | |
535 | free_vfsmnt(mnt); | |
536 | return NULL; | |
1da177e4 LT |
537 | } |
538 | ||
7b7b1ace | 539 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
540 | { |
541 | struct super_block *sb = mnt->mnt_sb; | |
3d733633 DH |
542 | /* |
543 | * This probably indicates that somebody messed | |
544 | * up a mnt_want/drop_write() pair. If this | |
545 | * happens, the filesystem was probably unable | |
546 | * to make r/w->r/o transitions. | |
547 | */ | |
d3ef3d73 | 548 | /* |
549 | * atomic_dec_and_lock() used to deal with ->mnt_count decrements | |
550 | * provides barriers, so count_mnt_writers() below is safe. AV | |
551 | */ | |
552 | WARN_ON(count_mnt_writers(mnt)); | |
1da177e4 LT |
553 | dput(mnt->mnt_root); |
554 | free_vfsmnt(mnt); | |
555 | deactivate_super(sb); | |
556 | } | |
557 | ||
7b7b1ace AV |
558 | void mntput_no_expire(struct vfsmount *mnt) |
559 | { | |
560 | repeat: | |
561 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
562 | if (likely(!mnt->mnt_pinned)) { | |
563 | spin_unlock(&vfsmount_lock); | |
564 | __mntput(mnt); | |
565 | return; | |
566 | } | |
567 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
568 | mnt->mnt_pinned = 0; | |
569 | spin_unlock(&vfsmount_lock); | |
570 | acct_auto_close_mnt(mnt); | |
571 | security_sb_umount_close(mnt); | |
572 | goto repeat; | |
573 | } | |
574 | } | |
575 | ||
576 | EXPORT_SYMBOL(mntput_no_expire); | |
577 | ||
578 | void mnt_pin(struct vfsmount *mnt) | |
579 | { | |
580 | spin_lock(&vfsmount_lock); | |
581 | mnt->mnt_pinned++; | |
582 | spin_unlock(&vfsmount_lock); | |
583 | } | |
584 | ||
585 | EXPORT_SYMBOL(mnt_pin); | |
586 | ||
587 | void mnt_unpin(struct vfsmount *mnt) | |
588 | { | |
589 | spin_lock(&vfsmount_lock); | |
590 | if (mnt->mnt_pinned) { | |
591 | atomic_inc(&mnt->mnt_count); | |
592 | mnt->mnt_pinned--; | |
593 | } | |
594 | spin_unlock(&vfsmount_lock); | |
595 | } | |
596 | ||
597 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 598 | |
b3b304a2 MS |
599 | static inline void mangle(struct seq_file *m, const char *s) |
600 | { | |
601 | seq_escape(m, s, " \t\n\\"); | |
602 | } | |
603 | ||
604 | /* | |
605 | * Simple .show_options callback for filesystems which don't want to | |
606 | * implement more complex mount option showing. | |
607 | * | |
608 | * See also save_mount_options(). | |
609 | */ | |
610 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
611 | { | |
2a32cebd AV |
612 | const char *options; |
613 | ||
614 | rcu_read_lock(); | |
615 | options = rcu_dereference(mnt->mnt_sb->s_options); | |
b3b304a2 MS |
616 | |
617 | if (options != NULL && options[0]) { | |
618 | seq_putc(m, ','); | |
619 | mangle(m, options); | |
620 | } | |
2a32cebd | 621 | rcu_read_unlock(); |
b3b304a2 MS |
622 | |
623 | return 0; | |
624 | } | |
625 | EXPORT_SYMBOL(generic_show_options); | |
626 | ||
627 | /* | |
628 | * If filesystem uses generic_show_options(), this function should be | |
629 | * called from the fill_super() callback. | |
630 | * | |
631 | * The .remount_fs callback usually needs to be handled in a special | |
632 | * way, to make sure, that previous options are not overwritten if the | |
633 | * remount fails. | |
634 | * | |
635 | * Also note, that if the filesystem's .remount_fs function doesn't | |
636 | * reset all options to their default value, but changes only newly | |
637 | * given options, then the displayed options will not reflect reality | |
638 | * any more. | |
639 | */ | |
640 | void save_mount_options(struct super_block *sb, char *options) | |
641 | { | |
2a32cebd AV |
642 | BUG_ON(sb->s_options); |
643 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
644 | } |
645 | EXPORT_SYMBOL(save_mount_options); | |
646 | ||
2a32cebd AV |
647 | void replace_mount_options(struct super_block *sb, char *options) |
648 | { | |
649 | char *old = sb->s_options; | |
650 | rcu_assign_pointer(sb->s_options, options); | |
651 | if (old) { | |
652 | synchronize_rcu(); | |
653 | kfree(old); | |
654 | } | |
655 | } | |
656 | EXPORT_SYMBOL(replace_mount_options); | |
657 | ||
a1a2c409 | 658 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
659 | /* iterator */ |
660 | static void *m_start(struct seq_file *m, loff_t *pos) | |
661 | { | |
a1a2c409 | 662 | struct proc_mounts *p = m->private; |
1da177e4 | 663 | |
390c6843 | 664 | down_read(&namespace_sem); |
a1a2c409 | 665 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
666 | } |
667 | ||
668 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
669 | { | |
a1a2c409 | 670 | struct proc_mounts *p = m->private; |
b0765fb8 | 671 | |
a1a2c409 | 672 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
673 | } |
674 | ||
675 | static void m_stop(struct seq_file *m, void *v) | |
676 | { | |
390c6843 | 677 | up_read(&namespace_sem); |
1da177e4 LT |
678 | } |
679 | ||
2d4d4864 RP |
680 | struct proc_fs_info { |
681 | int flag; | |
682 | const char *str; | |
683 | }; | |
684 | ||
2069f457 | 685 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) |
1da177e4 | 686 | { |
2d4d4864 | 687 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
688 | { MS_SYNCHRONOUS, ",sync" }, |
689 | { MS_DIRSYNC, ",dirsync" }, | |
690 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
691 | { 0, NULL } |
692 | }; | |
2d4d4864 RP |
693 | const struct proc_fs_info *fs_infop; |
694 | ||
695 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
696 | if (sb->s_flags & fs_infop->flag) | |
697 | seq_puts(m, fs_infop->str); | |
698 | } | |
2069f457 EP |
699 | |
700 | return security_sb_show_options(m, sb); | |
2d4d4864 RP |
701 | } |
702 | ||
703 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
704 | { | |
705 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
706 | { MNT_NOSUID, ",nosuid" }, |
707 | { MNT_NODEV, ",nodev" }, | |
708 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
709 | { MNT_NOATIME, ",noatime" }, |
710 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 711 | { MNT_RELATIME, ",relatime" }, |
d0adde57 | 712 | { MNT_STRICTATIME, ",strictatime" }, |
1da177e4 LT |
713 | { 0, NULL } |
714 | }; | |
2d4d4864 RP |
715 | const struct proc_fs_info *fs_infop; |
716 | ||
717 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
718 | if (mnt->mnt_flags & fs_infop->flag) | |
719 | seq_puts(m, fs_infop->str); | |
720 | } | |
721 | } | |
722 | ||
723 | static void show_type(struct seq_file *m, struct super_block *sb) | |
724 | { | |
725 | mangle(m, sb->s_type->name); | |
726 | if (sb->s_subtype && sb->s_subtype[0]) { | |
727 | seq_putc(m, '.'); | |
728 | mangle(m, sb->s_subtype); | |
729 | } | |
730 | } | |
731 | ||
732 | static int show_vfsmnt(struct seq_file *m, void *v) | |
733 | { | |
734 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
735 | int err = 0; | |
c32c2f63 | 736 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
737 | |
738 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
739 | seq_putc(m, ' '); | |
c32c2f63 | 740 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 741 | seq_putc(m, ' '); |
2d4d4864 | 742 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 743 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2069f457 EP |
744 | err = show_sb_opts(m, mnt->mnt_sb); |
745 | if (err) | |
746 | goto out; | |
2d4d4864 | 747 | show_mnt_opts(m, mnt); |
1da177e4 LT |
748 | if (mnt->mnt_sb->s_op->show_options) |
749 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
750 | seq_puts(m, " 0 0\n"); | |
2069f457 | 751 | out: |
1da177e4 LT |
752 | return err; |
753 | } | |
754 | ||
a1a2c409 | 755 | const struct seq_operations mounts_op = { |
1da177e4 LT |
756 | .start = m_start, |
757 | .next = m_next, | |
758 | .stop = m_stop, | |
759 | .show = show_vfsmnt | |
760 | }; | |
761 | ||
2d4d4864 RP |
762 | static int show_mountinfo(struct seq_file *m, void *v) |
763 | { | |
764 | struct proc_mounts *p = m->private; | |
765 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
766 | struct super_block *sb = mnt->mnt_sb; | |
767 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
768 | struct path root = p->root; | |
769 | int err = 0; | |
770 | ||
771 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
772 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
773 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
774 | seq_putc(m, ' '); | |
775 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
776 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
777 | /* | |
778 | * Mountpoint is outside root, discard that one. Ugly, | |
779 | * but less so than trying to do that in iterator in a | |
780 | * race-free way (due to renames). | |
781 | */ | |
782 | return SEQ_SKIP; | |
783 | } | |
784 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
785 | show_mnt_opts(m, mnt); | |
786 | ||
787 | /* Tagged fields ("foo:X" or "bar") */ | |
788 | if (IS_MNT_SHARED(mnt)) | |
789 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
790 | if (IS_MNT_SLAVE(mnt)) { |
791 | int master = mnt->mnt_master->mnt_group_id; | |
792 | int dom = get_dominating_id(mnt, &p->root); | |
793 | seq_printf(m, " master:%i", master); | |
794 | if (dom && dom != master) | |
795 | seq_printf(m, " propagate_from:%i", dom); | |
796 | } | |
2d4d4864 RP |
797 | if (IS_MNT_UNBINDABLE(mnt)) |
798 | seq_puts(m, " unbindable"); | |
799 | ||
800 | /* Filesystem specific data */ | |
801 | seq_puts(m, " - "); | |
802 | show_type(m, sb); | |
803 | seq_putc(m, ' '); | |
804 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
805 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
2069f457 EP |
806 | err = show_sb_opts(m, sb); |
807 | if (err) | |
808 | goto out; | |
2d4d4864 RP |
809 | if (sb->s_op->show_options) |
810 | err = sb->s_op->show_options(m, mnt); | |
811 | seq_putc(m, '\n'); | |
2069f457 | 812 | out: |
2d4d4864 RP |
813 | return err; |
814 | } | |
815 | ||
816 | const struct seq_operations mountinfo_op = { | |
817 | .start = m_start, | |
818 | .next = m_next, | |
819 | .stop = m_stop, | |
820 | .show = show_mountinfo, | |
821 | }; | |
822 | ||
b4629fe2 CL |
823 | static int show_vfsstat(struct seq_file *m, void *v) |
824 | { | |
b0765fb8 | 825 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 826 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
827 | int err = 0; |
828 | ||
829 | /* device */ | |
830 | if (mnt->mnt_devname) { | |
831 | seq_puts(m, "device "); | |
832 | mangle(m, mnt->mnt_devname); | |
833 | } else | |
834 | seq_puts(m, "no device"); | |
835 | ||
836 | /* mount point */ | |
837 | seq_puts(m, " mounted on "); | |
c32c2f63 | 838 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
839 | seq_putc(m, ' '); |
840 | ||
841 | /* file system type */ | |
842 | seq_puts(m, "with fstype "); | |
2d4d4864 | 843 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
844 | |
845 | /* optional statistics */ | |
846 | if (mnt->mnt_sb->s_op->show_stats) { | |
847 | seq_putc(m, ' '); | |
848 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
849 | } | |
850 | ||
851 | seq_putc(m, '\n'); | |
852 | return err; | |
853 | } | |
854 | ||
a1a2c409 | 855 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
856 | .start = m_start, |
857 | .next = m_next, | |
858 | .stop = m_stop, | |
859 | .show = show_vfsstat, | |
860 | }; | |
a1a2c409 | 861 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 862 | |
1da177e4 LT |
863 | /** |
864 | * may_umount_tree - check if a mount tree is busy | |
865 | * @mnt: root of mount tree | |
866 | * | |
867 | * This is called to check if a tree of mounts has any | |
868 | * open files, pwds, chroots or sub mounts that are | |
869 | * busy. | |
870 | */ | |
871 | int may_umount_tree(struct vfsmount *mnt) | |
872 | { | |
36341f64 RP |
873 | int actual_refs = 0; |
874 | int minimum_refs = 0; | |
875 | struct vfsmount *p; | |
1da177e4 LT |
876 | |
877 | spin_lock(&vfsmount_lock); | |
36341f64 | 878 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
879 | actual_refs += atomic_read(&p->mnt_count); |
880 | minimum_refs += 2; | |
1da177e4 LT |
881 | } |
882 | spin_unlock(&vfsmount_lock); | |
883 | ||
884 | if (actual_refs > minimum_refs) | |
e3474a8e | 885 | return 0; |
1da177e4 | 886 | |
e3474a8e | 887 | return 1; |
1da177e4 LT |
888 | } |
889 | ||
890 | EXPORT_SYMBOL(may_umount_tree); | |
891 | ||
892 | /** | |
893 | * may_umount - check if a mount point is busy | |
894 | * @mnt: root of mount | |
895 | * | |
896 | * This is called to check if a mount point has any | |
897 | * open files, pwds, chroots or sub mounts. If the | |
898 | * mount has sub mounts this will return busy | |
899 | * regardless of whether the sub mounts are busy. | |
900 | * | |
901 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
902 | * give false negatives. The main reason why it's here is that we need | |
903 | * a non-destructive way to look for easily umountable filesystems. | |
904 | */ | |
905 | int may_umount(struct vfsmount *mnt) | |
906 | { | |
e3474a8e | 907 | int ret = 1; |
a05964f3 RP |
908 | spin_lock(&vfsmount_lock); |
909 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 910 | ret = 0; |
a05964f3 RP |
911 | spin_unlock(&vfsmount_lock); |
912 | return ret; | |
1da177e4 LT |
913 | } |
914 | ||
915 | EXPORT_SYMBOL(may_umount); | |
916 | ||
b90fa9ae | 917 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
918 | { |
919 | struct vfsmount *mnt; | |
bf066c7d | 920 | while (!list_empty(head)) { |
b5e61818 | 921 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
922 | list_del_init(&mnt->mnt_hash); |
923 | if (mnt->mnt_parent != mnt) { | |
924 | struct dentry *dentry; | |
925 | struct vfsmount *m; | |
926 | spin_lock(&vfsmount_lock); | |
927 | dentry = mnt->mnt_mountpoint; | |
928 | m = mnt->mnt_parent; | |
929 | mnt->mnt_mountpoint = mnt->mnt_root; | |
930 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 931 | m->mnt_ghosts--; |
70fbcdf4 RP |
932 | spin_unlock(&vfsmount_lock); |
933 | dput(dentry); | |
934 | mntput(m); | |
935 | } | |
936 | mntput(mnt); | |
937 | } | |
938 | } | |
939 | ||
a05964f3 | 940 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
941 | { |
942 | struct vfsmount *p; | |
1da177e4 | 943 | |
1bfba4e8 AM |
944 | for (p = mnt; p; p = next_mnt(p, mnt)) |
945 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 946 | |
a05964f3 RP |
947 | if (propagate) |
948 | propagate_umount(kill); | |
949 | ||
70fbcdf4 RP |
950 | list_for_each_entry(p, kill, mnt_hash) { |
951 | list_del_init(&p->mnt_expire); | |
952 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
953 | __touch_mnt_namespace(p->mnt_ns); |
954 | p->mnt_ns = NULL; | |
70fbcdf4 | 955 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
956 | if (p->mnt_parent != p) { |
957 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 958 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 959 | } |
a05964f3 | 960 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
961 | } |
962 | } | |
963 | ||
c35038be AV |
964 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
965 | ||
1da177e4 LT |
966 | static int do_umount(struct vfsmount *mnt, int flags) |
967 | { | |
b58fed8b | 968 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 969 | int retval; |
70fbcdf4 | 970 | LIST_HEAD(umount_list); |
1da177e4 LT |
971 | |
972 | retval = security_sb_umount(mnt, flags); | |
973 | if (retval) | |
974 | return retval; | |
975 | ||
976 | /* | |
977 | * Allow userspace to request a mountpoint be expired rather than | |
978 | * unmounting unconditionally. Unmount only happens if: | |
979 | * (1) the mark is already set (the mark is cleared by mntput()) | |
980 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
981 | */ | |
982 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 983 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
984 | flags & (MNT_FORCE | MNT_DETACH)) |
985 | return -EINVAL; | |
986 | ||
987 | if (atomic_read(&mnt->mnt_count) != 2) | |
988 | return -EBUSY; | |
989 | ||
990 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
991 | return -EAGAIN; | |
992 | } | |
993 | ||
994 | /* | |
995 | * If we may have to abort operations to get out of this | |
996 | * mount, and they will themselves hold resources we must | |
997 | * allow the fs to do things. In the Unix tradition of | |
998 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
999 | * might fail to complete on the first run through as other tasks | |
1000 | * must return, and the like. Thats for the mount program to worry | |
1001 | * about for the moment. | |
1002 | */ | |
1003 | ||
42faad99 | 1004 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1005 | sb->s_op->umount_begin(sb); |
42faad99 | 1006 | } |
1da177e4 LT |
1007 | |
1008 | /* | |
1009 | * No sense to grab the lock for this test, but test itself looks | |
1010 | * somewhat bogus. Suggestions for better replacement? | |
1011 | * Ho-hum... In principle, we might treat that as umount + switch | |
1012 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1013 | * Actually it makes sense, especially if rootfs would contain a | |
1014 | * /reboot - static binary that would close all descriptors and | |
1015 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1016 | */ | |
6ac08c39 | 1017 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1018 | /* |
1019 | * Special case for "unmounting" root ... | |
1020 | * we just try to remount it readonly. | |
1021 | */ | |
1022 | down_write(&sb->s_umount); | |
1023 | if (!(sb->s_flags & MS_RDONLY)) { | |
1024 | lock_kernel(); | |
1da177e4 LT |
1025 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1026 | unlock_kernel(); | |
1027 | } | |
1028 | up_write(&sb->s_umount); | |
1029 | return retval; | |
1030 | } | |
1031 | ||
390c6843 | 1032 | down_write(&namespace_sem); |
1da177e4 | 1033 | spin_lock(&vfsmount_lock); |
5addc5dd | 1034 | event++; |
1da177e4 | 1035 | |
c35038be AV |
1036 | if (!(flags & MNT_DETACH)) |
1037 | shrink_submounts(mnt, &umount_list); | |
1038 | ||
1da177e4 | 1039 | retval = -EBUSY; |
a05964f3 | 1040 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1041 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1042 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1043 | retval = 0; |
1044 | } | |
1045 | spin_unlock(&vfsmount_lock); | |
1046 | if (retval) | |
1047 | security_sb_umount_busy(mnt); | |
390c6843 | 1048 | up_write(&namespace_sem); |
70fbcdf4 | 1049 | release_mounts(&umount_list); |
1da177e4 LT |
1050 | return retval; |
1051 | } | |
1052 | ||
1053 | /* | |
1054 | * Now umount can handle mount points as well as block devices. | |
1055 | * This is important for filesystems which use unnamed block devices. | |
1056 | * | |
1057 | * We now support a flag for forced unmount like the other 'big iron' | |
1058 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1059 | */ | |
1060 | ||
bdc480e3 | 1061 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1062 | { |
2d8f3038 | 1063 | struct path path; |
1da177e4 LT |
1064 | int retval; |
1065 | ||
2d8f3038 | 1066 | retval = user_path(name, &path); |
1da177e4 LT |
1067 | if (retval) |
1068 | goto out; | |
1069 | retval = -EINVAL; | |
2d8f3038 | 1070 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1071 | goto dput_and_out; |
2d8f3038 | 1072 | if (!check_mnt(path.mnt)) |
1da177e4 LT |
1073 | goto dput_and_out; |
1074 | ||
1075 | retval = -EPERM; | |
1076 | if (!capable(CAP_SYS_ADMIN)) | |
1077 | goto dput_and_out; | |
1078 | ||
2d8f3038 | 1079 | retval = do_umount(path.mnt, flags); |
1da177e4 | 1080 | dput_and_out: |
429731b1 | 1081 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 AV |
1082 | dput(path.dentry); |
1083 | mntput_no_expire(path.mnt); | |
1da177e4 LT |
1084 | out: |
1085 | return retval; | |
1086 | } | |
1087 | ||
1088 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1089 | ||
1090 | /* | |
b58fed8b | 1091 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1092 | */ |
bdc480e3 | 1093 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1094 | { |
b58fed8b | 1095 | return sys_umount(name, 0); |
1da177e4 LT |
1096 | } |
1097 | ||
1098 | #endif | |
1099 | ||
2d92ab3c | 1100 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1101 | { |
1102 | if (capable(CAP_SYS_ADMIN)) | |
1103 | return 0; | |
1104 | return -EPERM; | |
1105 | #ifdef notyet | |
2d92ab3c | 1106 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1107 | return -EPERM; |
2d92ab3c | 1108 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1109 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1110 | return -EPERM; |
1111 | } | |
2d92ab3c | 1112 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1113 | return -EPERM; |
1114 | return 0; | |
1115 | #endif | |
1116 | } | |
1117 | ||
b90fa9ae | 1118 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1119 | int flag) |
1da177e4 LT |
1120 | { |
1121 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1122 | struct path path; |
1da177e4 | 1123 | |
9676f0c6 RP |
1124 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1125 | return NULL; | |
1126 | ||
36341f64 | 1127 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1128 | if (!q) |
1129 | goto Enomem; | |
1130 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1131 | ||
1132 | p = mnt; | |
fdadd65f | 1133 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
7ec02ef1 | 1134 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1135 | continue; |
1136 | ||
1137 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1138 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1139 | s = skip_mnt_tree(s); | |
1140 | continue; | |
1141 | } | |
1da177e4 LT |
1142 | while (p != s->mnt_parent) { |
1143 | p = p->mnt_parent; | |
1144 | q = q->mnt_parent; | |
1145 | } | |
1146 | p = s; | |
1a390689 AV |
1147 | path.mnt = q; |
1148 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1149 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1150 | if (!q) |
1151 | goto Enomem; | |
1152 | spin_lock(&vfsmount_lock); | |
1153 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1154 | attach_mnt(q, &path); |
1da177e4 LT |
1155 | spin_unlock(&vfsmount_lock); |
1156 | } | |
1157 | } | |
1158 | return res; | |
b58fed8b | 1159 | Enomem: |
1da177e4 | 1160 | if (res) { |
70fbcdf4 | 1161 | LIST_HEAD(umount_list); |
1da177e4 | 1162 | spin_lock(&vfsmount_lock); |
a05964f3 | 1163 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1164 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1165 | release_mounts(&umount_list); |
1da177e4 LT |
1166 | } |
1167 | return NULL; | |
1168 | } | |
1169 | ||
589ff870 | 1170 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 AV |
1171 | { |
1172 | struct vfsmount *tree; | |
1a60a280 | 1173 | down_write(&namespace_sem); |
589ff870 | 1174 | tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1175 | up_write(&namespace_sem); |
8aec0809 AV |
1176 | return tree; |
1177 | } | |
1178 | ||
1179 | void drop_collected_mounts(struct vfsmount *mnt) | |
1180 | { | |
1181 | LIST_HEAD(umount_list); | |
1a60a280 | 1182 | down_write(&namespace_sem); |
8aec0809 AV |
1183 | spin_lock(&vfsmount_lock); |
1184 | umount_tree(mnt, 0, &umount_list); | |
1185 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1186 | up_write(&namespace_sem); |
8aec0809 AV |
1187 | release_mounts(&umount_list); |
1188 | } | |
1189 | ||
719f5d7f MS |
1190 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1191 | { | |
1192 | struct vfsmount *p; | |
1193 | ||
1194 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1195 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1196 | mnt_release_group_id(p); | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1201 | { | |
1202 | struct vfsmount *p; | |
1203 | ||
1204 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1205 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1206 | int err = mnt_alloc_group_id(p); | |
1207 | if (err) { | |
1208 | cleanup_group_ids(mnt, p); | |
1209 | return err; | |
1210 | } | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | return 0; | |
1215 | } | |
1216 | ||
b90fa9ae RP |
1217 | /* |
1218 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1219 | * @nd : place the mount tree @source_mnt is attached |
1220 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1221 | * store the parent mount and mountpoint dentry. | |
1222 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1223 | * |
1224 | * NOTE: in the table below explains the semantics when a source mount | |
1225 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1226 | * --------------------------------------------------------------------------- |
1227 | * | BIND MOUNT OPERATION | | |
1228 | * |************************************************************************** | |
1229 | * | source-->| shared | private | slave | unbindable | | |
1230 | * | dest | | | | | | |
1231 | * | | | | | | | | |
1232 | * | v | | | | | | |
1233 | * |************************************************************************** | |
1234 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1235 | * | | | | | | | |
1236 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1237 | * *************************************************************************** | |
b90fa9ae RP |
1238 | * A bind operation clones the source mount and mounts the clone on the |
1239 | * destination mount. | |
1240 | * | |
1241 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1242 | * tree of the destination mount and the cloned mount is added to | |
1243 | * the peer group of the source mount. | |
1244 | * (+) the cloned mount is created under the destination mount and is marked | |
1245 | * as shared. The cloned mount is added to the peer group of the source | |
1246 | * mount. | |
5afe0022 RP |
1247 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1248 | * of the destination mount and the cloned mount is made slave | |
1249 | * of the same master as that of the source mount. The cloned mount | |
1250 | * is marked as 'shared and slave'. | |
1251 | * (*) the cloned mount is made a slave of the same master as that of the | |
1252 | * source mount. | |
1253 | * | |
9676f0c6 RP |
1254 | * --------------------------------------------------------------------------- |
1255 | * | MOVE MOUNT OPERATION | | |
1256 | * |************************************************************************** | |
1257 | * | source-->| shared | private | slave | unbindable | | |
1258 | * | dest | | | | | | |
1259 | * | | | | | | | | |
1260 | * | v | | | | | | |
1261 | * |************************************************************************** | |
1262 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1263 | * | | | | | | | |
1264 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1265 | * *************************************************************************** | |
5afe0022 RP |
1266 | * |
1267 | * (+) the mount is moved to the destination. And is then propagated to | |
1268 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1269 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1270 | * (+++) the mount is moved to the destination and is then propagated to |
1271 | * all the mounts belonging to the destination mount's propagation tree. | |
1272 | * the mount is marked as 'shared and slave'. | |
1273 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1274 | * |
1275 | * if the source mount is a tree, the operations explained above is | |
1276 | * applied to each mount in the tree. | |
1277 | * Must be called without spinlocks held, since this function can sleep | |
1278 | * in allocations. | |
1279 | */ | |
1280 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1281 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1282 | { |
1283 | LIST_HEAD(tree_list); | |
1a390689 AV |
1284 | struct vfsmount *dest_mnt = path->mnt; |
1285 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1286 | struct vfsmount *child, *p; |
719f5d7f | 1287 | int err; |
b90fa9ae | 1288 | |
719f5d7f MS |
1289 | if (IS_MNT_SHARED(dest_mnt)) { |
1290 | err = invent_group_ids(source_mnt, true); | |
1291 | if (err) | |
1292 | goto out; | |
1293 | } | |
1294 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1295 | if (err) | |
1296 | goto out_cleanup_ids; | |
b90fa9ae RP |
1297 | |
1298 | if (IS_MNT_SHARED(dest_mnt)) { | |
1299 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1300 | set_mnt_shared(p); | |
1301 | } | |
1302 | ||
1303 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1304 | if (parent_path) { |
1305 | detach_mnt(source_mnt, parent_path); | |
1306 | attach_mnt(source_mnt, path); | |
e5d67f07 | 1307 | touch_mnt_namespace(parent_path->mnt->mnt_ns); |
21444403 RP |
1308 | } else { |
1309 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1310 | commit_tree(source_mnt); | |
1311 | } | |
b90fa9ae RP |
1312 | |
1313 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1314 | list_del_init(&child->mnt_hash); | |
1315 | commit_tree(child); | |
1316 | } | |
1317 | spin_unlock(&vfsmount_lock); | |
1318 | return 0; | |
719f5d7f MS |
1319 | |
1320 | out_cleanup_ids: | |
1321 | if (IS_MNT_SHARED(dest_mnt)) | |
1322 | cleanup_group_ids(source_mnt, NULL); | |
1323 | out: | |
1324 | return err; | |
b90fa9ae RP |
1325 | } |
1326 | ||
8c3ee42e | 1327 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1328 | { |
1329 | int err; | |
1330 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1331 | return -EINVAL; | |
1332 | ||
8c3ee42e | 1333 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1334 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1335 | return -ENOTDIR; | |
1336 | ||
1337 | err = -ENOENT; | |
8c3ee42e AV |
1338 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1339 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1340 | goto out_unlock; |
1341 | ||
8c3ee42e | 1342 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1343 | if (err) |
1344 | goto out_unlock; | |
1345 | ||
1346 | err = -ENOENT; | |
8c3ee42e AV |
1347 | if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) |
1348 | err = attach_recursive_mnt(mnt, path, NULL); | |
1da177e4 | 1349 | out_unlock: |
8c3ee42e | 1350 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1351 | if (!err) |
8c3ee42e | 1352 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1353 | return err; |
1354 | } | |
1355 | ||
07b20889 RP |
1356 | /* |
1357 | * recursively change the type of the mountpoint. | |
1358 | */ | |
0a0d8a46 | 1359 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1360 | { |
2d92ab3c | 1361 | struct vfsmount *m, *mnt = path->mnt; |
07b20889 RP |
1362 | int recurse = flag & MS_REC; |
1363 | int type = flag & ~MS_REC; | |
719f5d7f | 1364 | int err = 0; |
07b20889 | 1365 | |
ee6f9582 MS |
1366 | if (!capable(CAP_SYS_ADMIN)) |
1367 | return -EPERM; | |
1368 | ||
2d92ab3c | 1369 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1370 | return -EINVAL; |
1371 | ||
1372 | down_write(&namespace_sem); | |
719f5d7f MS |
1373 | if (type == MS_SHARED) { |
1374 | err = invent_group_ids(mnt, recurse); | |
1375 | if (err) | |
1376 | goto out_unlock; | |
1377 | } | |
1378 | ||
07b20889 RP |
1379 | spin_lock(&vfsmount_lock); |
1380 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1381 | change_mnt_propagation(m, type); | |
1382 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1383 | |
1384 | out_unlock: | |
07b20889 | 1385 | up_write(&namespace_sem); |
719f5d7f | 1386 | return err; |
07b20889 RP |
1387 | } |
1388 | ||
1da177e4 LT |
1389 | /* |
1390 | * do loopback mount. | |
1391 | */ | |
0a0d8a46 | 1392 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1393 | int recurse) |
1da177e4 | 1394 | { |
2d92ab3c | 1395 | struct path old_path; |
1da177e4 | 1396 | struct vfsmount *mnt = NULL; |
2d92ab3c | 1397 | int err = mount_is_safe(path); |
1da177e4 LT |
1398 | if (err) |
1399 | return err; | |
1400 | if (!old_name || !*old_name) | |
1401 | return -EINVAL; | |
2d92ab3c | 1402 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1403 | if (err) |
1404 | return err; | |
1405 | ||
390c6843 | 1406 | down_write(&namespace_sem); |
1da177e4 | 1407 | err = -EINVAL; |
2d92ab3c | 1408 | if (IS_MNT_UNBINDABLE(old_path.mnt)) |
4ac91378 | 1409 | goto out; |
9676f0c6 | 1410 | |
2d92ab3c | 1411 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
ccd48bc7 | 1412 | goto out; |
1da177e4 | 1413 | |
ccd48bc7 AV |
1414 | err = -ENOMEM; |
1415 | if (recurse) | |
2d92ab3c | 1416 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 | 1417 | else |
2d92ab3c | 1418 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 AV |
1419 | |
1420 | if (!mnt) | |
1421 | goto out; | |
1422 | ||
2d92ab3c | 1423 | err = graft_tree(mnt, path); |
ccd48bc7 | 1424 | if (err) { |
70fbcdf4 | 1425 | LIST_HEAD(umount_list); |
1da177e4 | 1426 | spin_lock(&vfsmount_lock); |
a05964f3 | 1427 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1428 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1429 | release_mounts(&umount_list); |
5b83d2c5 | 1430 | } |
1da177e4 | 1431 | |
ccd48bc7 | 1432 | out: |
390c6843 | 1433 | up_write(&namespace_sem); |
2d92ab3c | 1434 | path_put(&old_path); |
1da177e4 LT |
1435 | return err; |
1436 | } | |
1437 | ||
2e4b7fcd DH |
1438 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1439 | { | |
1440 | int error = 0; | |
1441 | int readonly_request = 0; | |
1442 | ||
1443 | if (ms_flags & MS_RDONLY) | |
1444 | readonly_request = 1; | |
1445 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1446 | return 0; | |
1447 | ||
1448 | if (readonly_request) | |
1449 | error = mnt_make_readonly(mnt); | |
1450 | else | |
1451 | __mnt_unmake_readonly(mnt); | |
1452 | return error; | |
1453 | } | |
1454 | ||
1da177e4 LT |
1455 | /* |
1456 | * change filesystem flags. dir should be a physical root of filesystem. | |
1457 | * If you've mounted a non-root directory somewhere and want to do remount | |
1458 | * on it - tough luck. | |
1459 | */ | |
0a0d8a46 | 1460 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1461 | void *data) |
1462 | { | |
1463 | int err; | |
2d92ab3c | 1464 | struct super_block *sb = path->mnt->mnt_sb; |
1da177e4 LT |
1465 | |
1466 | if (!capable(CAP_SYS_ADMIN)) | |
1467 | return -EPERM; | |
1468 | ||
2d92ab3c | 1469 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1470 | return -EINVAL; |
1471 | ||
2d92ab3c | 1472 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1473 | return -EINVAL; |
1474 | ||
1475 | down_write(&sb->s_umount); | |
2e4b7fcd | 1476 | if (flags & MS_BIND) |
2d92ab3c | 1477 | err = change_mount_flags(path->mnt, flags); |
2e4b7fcd DH |
1478 | else |
1479 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1480 | if (!err) |
2d92ab3c | 1481 | path->mnt->mnt_flags = mnt_flags; |
1da177e4 | 1482 | up_write(&sb->s_umount); |
0e55a7cc | 1483 | if (!err) { |
2d92ab3c | 1484 | security_sb_post_remount(path->mnt, flags, data); |
0e55a7cc DW |
1485 | |
1486 | spin_lock(&vfsmount_lock); | |
1487 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1488 | spin_unlock(&vfsmount_lock); | |
1489 | } | |
1da177e4 LT |
1490 | return err; |
1491 | } | |
1492 | ||
9676f0c6 RP |
1493 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1494 | { | |
1495 | struct vfsmount *p; | |
1496 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1497 | if (IS_MNT_UNBINDABLE(p)) | |
1498 | return 1; | |
1499 | } | |
1500 | return 0; | |
1501 | } | |
1502 | ||
0a0d8a46 | 1503 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1504 | { |
2d92ab3c | 1505 | struct path old_path, parent_path; |
1da177e4 LT |
1506 | struct vfsmount *p; |
1507 | int err = 0; | |
1508 | if (!capable(CAP_SYS_ADMIN)) | |
1509 | return -EPERM; | |
1510 | if (!old_name || !*old_name) | |
1511 | return -EINVAL; | |
2d92ab3c | 1512 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1513 | if (err) |
1514 | return err; | |
1515 | ||
390c6843 | 1516 | down_write(&namespace_sem); |
2d92ab3c | 1517 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1518 | follow_down(path)) |
1da177e4 LT |
1519 | ; |
1520 | err = -EINVAL; | |
2d92ab3c | 1521 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
1da177e4 LT |
1522 | goto out; |
1523 | ||
1524 | err = -ENOENT; | |
2d92ab3c AV |
1525 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1526 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1527 | goto out1; |
1528 | ||
2d92ab3c | 1529 | if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry)) |
21444403 | 1530 | goto out1; |
1da177e4 LT |
1531 | |
1532 | err = -EINVAL; | |
2d92ab3c | 1533 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1534 | goto out1; |
1da177e4 | 1535 | |
2d92ab3c | 1536 | if (old_path.mnt == old_path.mnt->mnt_parent) |
21444403 | 1537 | goto out1; |
1da177e4 | 1538 | |
2d92ab3c AV |
1539 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1540 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1541 | goto out1; |
1542 | /* | |
1543 | * Don't move a mount residing in a shared parent. | |
1544 | */ | |
2d92ab3c AV |
1545 | if (old_path.mnt->mnt_parent && |
1546 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
21444403 | 1547 | goto out1; |
9676f0c6 RP |
1548 | /* |
1549 | * Don't move a mount tree containing unbindable mounts to a destination | |
1550 | * mount which is shared. | |
1551 | */ | |
2d92ab3c AV |
1552 | if (IS_MNT_SHARED(path->mnt) && |
1553 | tree_contains_unbindable(old_path.mnt)) | |
9676f0c6 | 1554 | goto out1; |
1da177e4 | 1555 | err = -ELOOP; |
2d92ab3c AV |
1556 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1557 | if (p == old_path.mnt) | |
21444403 | 1558 | goto out1; |
1da177e4 | 1559 | |
2d92ab3c | 1560 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); |
4ac91378 | 1561 | if (err) |
21444403 | 1562 | goto out1; |
1da177e4 LT |
1563 | |
1564 | /* if the mount is moved, it should no longer be expire | |
1565 | * automatically */ | |
2d92ab3c | 1566 | list_del_init(&old_path.mnt->mnt_expire); |
1da177e4 | 1567 | out1: |
2d92ab3c | 1568 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1569 | out: |
390c6843 | 1570 | up_write(&namespace_sem); |
1da177e4 | 1571 | if (!err) |
1a390689 | 1572 | path_put(&parent_path); |
2d92ab3c | 1573 | path_put(&old_path); |
1da177e4 LT |
1574 | return err; |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * create a new mount for userspace and request it to be added into the | |
1579 | * namespace's tree | |
1580 | */ | |
0a0d8a46 | 1581 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1582 | int mnt_flags, char *name, void *data) |
1583 | { | |
1584 | struct vfsmount *mnt; | |
1585 | ||
1586 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1587 | return -EINVAL; | |
1588 | ||
1589 | /* we need capabilities... */ | |
1590 | if (!capable(CAP_SYS_ADMIN)) | |
1591 | return -EPERM; | |
1592 | ||
1593 | mnt = do_kern_mount(type, flags, name, data); | |
1594 | if (IS_ERR(mnt)) | |
1595 | return PTR_ERR(mnt); | |
1596 | ||
2d92ab3c | 1597 | return do_add_mount(mnt, path, mnt_flags, NULL); |
1da177e4 LT |
1598 | } |
1599 | ||
1600 | /* | |
1601 | * add a mount into a namespace's mount tree | |
1602 | * - provide the option of adding the new mount to an expiration list | |
1603 | */ | |
8d66bf54 | 1604 | int do_add_mount(struct vfsmount *newmnt, struct path *path, |
1da177e4 LT |
1605 | int mnt_flags, struct list_head *fslist) |
1606 | { | |
1607 | int err; | |
1608 | ||
390c6843 | 1609 | down_write(&namespace_sem); |
1da177e4 | 1610 | /* Something was mounted here while we slept */ |
8d66bf54 | 1611 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1612 | follow_down(path)) |
1da177e4 LT |
1613 | ; |
1614 | err = -EINVAL; | |
dd5cae6e | 1615 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) |
1da177e4 LT |
1616 | goto unlock; |
1617 | ||
1618 | /* Refuse the same filesystem on the same mount point */ | |
1619 | err = -EBUSY; | |
8d66bf54 AV |
1620 | if (path->mnt->mnt_sb == newmnt->mnt_sb && |
1621 | path->mnt->mnt_root == path->dentry) | |
1da177e4 LT |
1622 | goto unlock; |
1623 | ||
1624 | err = -EINVAL; | |
1625 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1626 | goto unlock; | |
1627 | ||
1628 | newmnt->mnt_flags = mnt_flags; | |
8d66bf54 | 1629 | if ((err = graft_tree(newmnt, path))) |
5b83d2c5 | 1630 | goto unlock; |
1da177e4 | 1631 | |
6758f953 | 1632 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1633 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1634 | |
390c6843 | 1635 | up_write(&namespace_sem); |
5b83d2c5 | 1636 | return 0; |
1da177e4 LT |
1637 | |
1638 | unlock: | |
390c6843 | 1639 | up_write(&namespace_sem); |
1da177e4 LT |
1640 | mntput(newmnt); |
1641 | return err; | |
1642 | } | |
1643 | ||
1644 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1645 | ||
1646 | /* | |
1647 | * process a list of expirable mountpoints with the intent of discarding any | |
1648 | * mountpoints that aren't in use and haven't been touched since last we came | |
1649 | * here | |
1650 | */ | |
1651 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1652 | { | |
1da177e4 LT |
1653 | struct vfsmount *mnt, *next; |
1654 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1655 | LIST_HEAD(umounts); |
1da177e4 LT |
1656 | |
1657 | if (list_empty(mounts)) | |
1658 | return; | |
1659 | ||
bcc5c7d2 | 1660 | down_write(&namespace_sem); |
1da177e4 LT |
1661 | spin_lock(&vfsmount_lock); |
1662 | ||
1663 | /* extract from the expiration list every vfsmount that matches the | |
1664 | * following criteria: | |
1665 | * - only referenced by its parent vfsmount | |
1666 | * - still marked for expiry (marked on the last call here; marks are | |
1667 | * cleared by mntput()) | |
1668 | */ | |
55e700b9 | 1669 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1670 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1671 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1672 | continue; |
55e700b9 | 1673 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1674 | } |
bcc5c7d2 AV |
1675 | while (!list_empty(&graveyard)) { |
1676 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1677 | touch_mnt_namespace(mnt->mnt_ns); | |
1678 | umount_tree(mnt, 1, &umounts); | |
1679 | } | |
5528f911 | 1680 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1681 | up_write(&namespace_sem); |
1682 | ||
1683 | release_mounts(&umounts); | |
5528f911 TM |
1684 | } |
1685 | ||
1686 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1687 | ||
1688 | /* | |
1689 | * Ripoff of 'select_parent()' | |
1690 | * | |
1691 | * search the list of submounts for a given mountpoint, and move any | |
1692 | * shrinkable submounts to the 'graveyard' list. | |
1693 | */ | |
1694 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1695 | { | |
1696 | struct vfsmount *this_parent = parent; | |
1697 | struct list_head *next; | |
1698 | int found = 0; | |
1699 | ||
1700 | repeat: | |
1701 | next = this_parent->mnt_mounts.next; | |
1702 | resume: | |
1703 | while (next != &this_parent->mnt_mounts) { | |
1704 | struct list_head *tmp = next; | |
1705 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1706 | ||
1707 | next = tmp->next; | |
1708 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1709 | continue; |
5528f911 TM |
1710 | /* |
1711 | * Descend a level if the d_mounts list is non-empty. | |
1712 | */ | |
1713 | if (!list_empty(&mnt->mnt_mounts)) { | |
1714 | this_parent = mnt; | |
1715 | goto repeat; | |
1716 | } | |
1da177e4 | 1717 | |
5528f911 | 1718 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1719 | list_move_tail(&mnt->mnt_expire, graveyard); |
1720 | found++; | |
1721 | } | |
1da177e4 | 1722 | } |
5528f911 TM |
1723 | /* |
1724 | * All done at this level ... ascend and resume the search | |
1725 | */ | |
1726 | if (this_parent != parent) { | |
1727 | next = this_parent->mnt_child.next; | |
1728 | this_parent = this_parent->mnt_parent; | |
1729 | goto resume; | |
1730 | } | |
1731 | return found; | |
1732 | } | |
1733 | ||
1734 | /* | |
1735 | * process a list of expirable mountpoints with the intent of discarding any | |
1736 | * submounts of a specific parent mountpoint | |
1737 | */ | |
c35038be | 1738 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1739 | { |
1740 | LIST_HEAD(graveyard); | |
c35038be | 1741 | struct vfsmount *m; |
5528f911 | 1742 | |
5528f911 | 1743 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1744 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1745 | while (!list_empty(&graveyard)) { |
c35038be | 1746 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 | 1747 | mnt_expire); |
afef80b3 EB |
1748 | touch_mnt_namespace(m->mnt_ns); |
1749 | umount_tree(m, 1, umounts); | |
bcc5c7d2 AV |
1750 | } |
1751 | } | |
1da177e4 LT |
1752 | } |
1753 | ||
1da177e4 LT |
1754 | /* |
1755 | * Some copy_from_user() implementations do not return the exact number of | |
1756 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1757 | * Note that this function differs from copy_from_user() in that it will oops | |
1758 | * on bad values of `to', rather than returning a short copy. | |
1759 | */ | |
b58fed8b RP |
1760 | static long exact_copy_from_user(void *to, const void __user * from, |
1761 | unsigned long n) | |
1da177e4 LT |
1762 | { |
1763 | char *t = to; | |
1764 | const char __user *f = from; | |
1765 | char c; | |
1766 | ||
1767 | if (!access_ok(VERIFY_READ, from, n)) | |
1768 | return n; | |
1769 | ||
1770 | while (n) { | |
1771 | if (__get_user(c, f)) { | |
1772 | memset(t, 0, n); | |
1773 | break; | |
1774 | } | |
1775 | *t++ = c; | |
1776 | f++; | |
1777 | n--; | |
1778 | } | |
1779 | return n; | |
1780 | } | |
1781 | ||
b58fed8b | 1782 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1783 | { |
1784 | int i; | |
1785 | unsigned long page; | |
1786 | unsigned long size; | |
b58fed8b | 1787 | |
1da177e4 LT |
1788 | *where = 0; |
1789 | if (!data) | |
1790 | return 0; | |
1791 | ||
1792 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1793 | return -ENOMEM; | |
1794 | ||
1795 | /* We only care that *some* data at the address the user | |
1796 | * gave us is valid. Just in case, we'll zero | |
1797 | * the remainder of the page. | |
1798 | */ | |
1799 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1800 | size = TASK_SIZE - (unsigned long)data; | |
1801 | if (size > PAGE_SIZE) | |
1802 | size = PAGE_SIZE; | |
1803 | ||
1804 | i = size - exact_copy_from_user((void *)page, data, size); | |
1805 | if (!i) { | |
b58fed8b | 1806 | free_page(page); |
1da177e4 LT |
1807 | return -EFAULT; |
1808 | } | |
1809 | if (i != PAGE_SIZE) | |
1810 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1811 | *where = page; | |
1812 | return 0; | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1817 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1818 | * | |
1819 | * data is a (void *) that can point to any structure up to | |
1820 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1821 | * information (or be NULL). | |
1822 | * | |
1823 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1824 | * When the flags word was introduced its top half was required | |
1825 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1826 | * Therefore, if this magic number is present, it carries no information | |
1827 | * and must be discarded. | |
1828 | */ | |
b58fed8b | 1829 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1830 | unsigned long flags, void *data_page) |
1831 | { | |
2d92ab3c | 1832 | struct path path; |
1da177e4 LT |
1833 | int retval = 0; |
1834 | int mnt_flags = 0; | |
1835 | ||
1836 | /* Discard magic */ | |
1837 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1838 | flags &= ~MS_MGC_MSK; | |
1839 | ||
1840 | /* Basic sanity checks */ | |
1841 | ||
1842 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1843 | return -EINVAL; | |
1844 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1845 | return -EINVAL; | |
1846 | ||
1847 | if (data_page) | |
1848 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1849 | ||
613cbe3d AK |
1850 | /* Default to relatime unless overriden */ |
1851 | if (!(flags & MS_NOATIME)) | |
1852 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 1853 | |
1da177e4 LT |
1854 | /* Separate the per-mountpoint flags */ |
1855 | if (flags & MS_NOSUID) | |
1856 | mnt_flags |= MNT_NOSUID; | |
1857 | if (flags & MS_NODEV) | |
1858 | mnt_flags |= MNT_NODEV; | |
1859 | if (flags & MS_NOEXEC) | |
1860 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1861 | if (flags & MS_NOATIME) |
1862 | mnt_flags |= MNT_NOATIME; | |
1863 | if (flags & MS_NODIRATIME) | |
1864 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
1865 | if (flags & MS_STRICTATIME) |
1866 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
1867 | if (flags & MS_RDONLY) |
1868 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1869 | |
1870 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
d0adde57 MG |
1871 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
1872 | MS_STRICTATIME); | |
1da177e4 LT |
1873 | |
1874 | /* ... and get the mountpoint */ | |
2d92ab3c | 1875 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1876 | if (retval) |
1877 | return retval; | |
1878 | ||
2d92ab3c | 1879 | retval = security_sb_mount(dev_name, &path, |
b5266eb4 | 1880 | type_page, flags, data_page); |
1da177e4 LT |
1881 | if (retval) |
1882 | goto dput_out; | |
1883 | ||
1884 | if (flags & MS_REMOUNT) | |
2d92ab3c | 1885 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
1886 | data_page); |
1887 | else if (flags & MS_BIND) | |
2d92ab3c | 1888 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 1889 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 1890 | retval = do_change_type(&path, flags); |
1da177e4 | 1891 | else if (flags & MS_MOVE) |
2d92ab3c | 1892 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 1893 | else |
2d92ab3c | 1894 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
1895 | dev_name, data_page); |
1896 | dput_out: | |
2d92ab3c | 1897 | path_put(&path); |
1da177e4 LT |
1898 | return retval; |
1899 | } | |
1900 | ||
741a2951 JD |
1901 | /* |
1902 | * Allocate a new namespace structure and populate it with contents | |
1903 | * copied from the namespace of the passed in task structure. | |
1904 | */ | |
e3222c4e | 1905 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1906 | struct fs_struct *fs) |
1da177e4 | 1907 | { |
6b3286ed | 1908 | struct mnt_namespace *new_ns; |
7f2da1e7 | 1909 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
1da177e4 LT |
1910 | struct vfsmount *p, *q; |
1911 | ||
6b3286ed | 1912 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1913 | if (!new_ns) |
467e9f4b | 1914 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1915 | |
1916 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1917 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1918 | init_waitqueue_head(&new_ns->poll); |
1919 | new_ns->event = 0; | |
1da177e4 | 1920 | |
390c6843 | 1921 | down_write(&namespace_sem); |
1da177e4 | 1922 | /* First pass: copy the tree topology */ |
6b3286ed | 1923 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1924 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1925 | if (!new_ns->root) { |
390c6843 | 1926 | up_write(&namespace_sem); |
1da177e4 | 1927 | kfree(new_ns); |
5cc4a034 | 1928 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1929 | } |
1930 | spin_lock(&vfsmount_lock); | |
1931 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1932 | spin_unlock(&vfsmount_lock); | |
1933 | ||
1934 | /* | |
1935 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1936 | * as belonging to new namespace. We have already acquired a private | |
1937 | * fs_struct, so tsk->fs->lock is not needed. | |
1938 | */ | |
6b3286ed | 1939 | p = mnt_ns->root; |
1da177e4 LT |
1940 | q = new_ns->root; |
1941 | while (p) { | |
6b3286ed | 1942 | q->mnt_ns = new_ns; |
1da177e4 | 1943 | if (fs) { |
6ac08c39 | 1944 | if (p == fs->root.mnt) { |
1da177e4 | 1945 | rootmnt = p; |
6ac08c39 | 1946 | fs->root.mnt = mntget(q); |
1da177e4 | 1947 | } |
6ac08c39 | 1948 | if (p == fs->pwd.mnt) { |
1da177e4 | 1949 | pwdmnt = p; |
6ac08c39 | 1950 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1951 | } |
1da177e4 | 1952 | } |
6b3286ed | 1953 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1954 | q = next_mnt(q, new_ns->root); |
1955 | } | |
390c6843 | 1956 | up_write(&namespace_sem); |
1da177e4 | 1957 | |
1da177e4 LT |
1958 | if (rootmnt) |
1959 | mntput(rootmnt); | |
1960 | if (pwdmnt) | |
1961 | mntput(pwdmnt); | |
1da177e4 | 1962 | |
741a2951 JD |
1963 | return new_ns; |
1964 | } | |
1965 | ||
213dd266 | 1966 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1967 | struct fs_struct *new_fs) |
741a2951 | 1968 | { |
6b3286ed | 1969 | struct mnt_namespace *new_ns; |
741a2951 | 1970 | |
e3222c4e | 1971 | BUG_ON(!ns); |
6b3286ed | 1972 | get_mnt_ns(ns); |
741a2951 JD |
1973 | |
1974 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1975 | return ns; |
741a2951 | 1976 | |
e3222c4e | 1977 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1978 | |
6b3286ed | 1979 | put_mnt_ns(ns); |
e3222c4e | 1980 | return new_ns; |
1da177e4 LT |
1981 | } |
1982 | ||
bdc480e3 HC |
1983 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
1984 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 LT |
1985 | { |
1986 | int retval; | |
1987 | unsigned long data_page; | |
1988 | unsigned long type_page; | |
1989 | unsigned long dev_page; | |
1990 | char *dir_page; | |
1991 | ||
b58fed8b | 1992 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1993 | if (retval < 0) |
1994 | return retval; | |
1995 | ||
1996 | dir_page = getname(dir_name); | |
1997 | retval = PTR_ERR(dir_page); | |
1998 | if (IS_ERR(dir_page)) | |
1999 | goto out1; | |
2000 | ||
b58fed8b | 2001 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2002 | if (retval < 0) |
2003 | goto out2; | |
2004 | ||
b58fed8b | 2005 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2006 | if (retval < 0) |
2007 | goto out3; | |
2008 | ||
2009 | lock_kernel(); | |
b58fed8b RP |
2010 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2011 | flags, (void *)data_page); | |
1da177e4 LT |
2012 | unlock_kernel(); |
2013 | free_page(data_page); | |
2014 | ||
2015 | out3: | |
2016 | free_page(dev_page); | |
2017 | out2: | |
2018 | putname(dir_page); | |
2019 | out1: | |
2020 | free_page(type_page); | |
2021 | return retval; | |
2022 | } | |
2023 | ||
1da177e4 LT |
2024 | /* |
2025 | * pivot_root Semantics: | |
2026 | * Moves the root file system of the current process to the directory put_old, | |
2027 | * makes new_root as the new root file system of the current process, and sets | |
2028 | * root/cwd of all processes which had them on the current root to new_root. | |
2029 | * | |
2030 | * Restrictions: | |
2031 | * The new_root and put_old must be directories, and must not be on the | |
2032 | * same file system as the current process root. The put_old must be | |
2033 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2034 | * pointed to by put_old must yield the same directory as new_root. No other | |
2035 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2036 | * | |
4a0d11fa NB |
2037 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2038 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2039 | * in this situation. | |
2040 | * | |
1da177e4 LT |
2041 | * Notes: |
2042 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2043 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2044 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2045 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2046 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2047 | * first. | |
2048 | */ | |
3480b257 HC |
2049 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2050 | const char __user *, put_old) | |
1da177e4 LT |
2051 | { |
2052 | struct vfsmount *tmp; | |
2d8f3038 | 2053 | struct path new, old, parent_path, root_parent, root; |
1da177e4 LT |
2054 | int error; |
2055 | ||
2056 | if (!capable(CAP_SYS_ADMIN)) | |
2057 | return -EPERM; | |
2058 | ||
2d8f3038 | 2059 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2060 | if (error) |
2061 | goto out0; | |
2062 | error = -EINVAL; | |
2d8f3038 | 2063 | if (!check_mnt(new.mnt)) |
1da177e4 LT |
2064 | goto out1; |
2065 | ||
2d8f3038 | 2066 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2067 | if (error) |
2068 | goto out1; | |
2069 | ||
2d8f3038 | 2070 | error = security_sb_pivotroot(&old, &new); |
1da177e4 | 2071 | if (error) { |
2d8f3038 | 2072 | path_put(&old); |
1da177e4 LT |
2073 | goto out1; |
2074 | } | |
2075 | ||
2076 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2077 | root = current->fs->root; |
6ac08c39 | 2078 | path_get(¤t->fs->root); |
1da177e4 | 2079 | read_unlock(¤t->fs->lock); |
390c6843 | 2080 | down_write(&namespace_sem); |
2d8f3038 | 2081 | mutex_lock(&old.dentry->d_inode->i_mutex); |
1da177e4 | 2082 | error = -EINVAL; |
2d8f3038 AV |
2083 | if (IS_MNT_SHARED(old.mnt) || |
2084 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
8c3ee42e | 2085 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2086 | goto out2; |
8c3ee42e | 2087 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2088 | goto out2; |
2089 | error = -ENOENT; | |
2d8f3038 | 2090 | if (IS_DEADDIR(new.dentry->d_inode)) |
1da177e4 | 2091 | goto out2; |
2d8f3038 | 2092 | if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry)) |
1da177e4 | 2093 | goto out2; |
2d8f3038 | 2094 | if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry)) |
1da177e4 LT |
2095 | goto out2; |
2096 | error = -EBUSY; | |
2d8f3038 AV |
2097 | if (new.mnt == root.mnt || |
2098 | old.mnt == root.mnt) | |
1da177e4 LT |
2099 | goto out2; /* loop, on the same file system */ |
2100 | error = -EINVAL; | |
8c3ee42e | 2101 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2102 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2103 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2104 | goto out2; /* not attached */ |
2d8f3038 | 2105 | if (new.mnt->mnt_root != new.dentry) |
1da177e4 | 2106 | goto out2; /* not a mountpoint */ |
2d8f3038 | 2107 | if (new.mnt->mnt_parent == new.mnt) |
0bb6fcc1 | 2108 | goto out2; /* not attached */ |
4ac91378 | 2109 | /* make sure we can reach put_old from new_root */ |
2d8f3038 | 2110 | tmp = old.mnt; |
1da177e4 | 2111 | spin_lock(&vfsmount_lock); |
2d8f3038 | 2112 | if (tmp != new.mnt) { |
1da177e4 LT |
2113 | for (;;) { |
2114 | if (tmp->mnt_parent == tmp) | |
2115 | goto out3; /* already mounted on put_old */ | |
2d8f3038 | 2116 | if (tmp->mnt_parent == new.mnt) |
1da177e4 LT |
2117 | break; |
2118 | tmp = tmp->mnt_parent; | |
2119 | } | |
2d8f3038 | 2120 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) |
1da177e4 | 2121 | goto out3; |
2d8f3038 | 2122 | } else if (!is_subdir(old.dentry, new.dentry)) |
1da177e4 | 2123 | goto out3; |
2d8f3038 | 2124 | detach_mnt(new.mnt, &parent_path); |
8c3ee42e | 2125 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2126 | /* mount old root on put_old */ |
2d8f3038 | 2127 | attach_mnt(root.mnt, &old); |
4ac91378 | 2128 | /* mount new_root on / */ |
2d8f3038 | 2129 | attach_mnt(new.mnt, &root_parent); |
6b3286ed | 2130 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2131 | spin_unlock(&vfsmount_lock); |
2d8f3038 AV |
2132 | chroot_fs_refs(&root, &new); |
2133 | security_sb_post_pivotroot(&root, &new); | |
1da177e4 | 2134 | error = 0; |
1a390689 AV |
2135 | path_put(&root_parent); |
2136 | path_put(&parent_path); | |
1da177e4 | 2137 | out2: |
2d8f3038 | 2138 | mutex_unlock(&old.dentry->d_inode->i_mutex); |
390c6843 | 2139 | up_write(&namespace_sem); |
8c3ee42e | 2140 | path_put(&root); |
2d8f3038 | 2141 | path_put(&old); |
1da177e4 | 2142 | out1: |
2d8f3038 | 2143 | path_put(&new); |
1da177e4 | 2144 | out0: |
1da177e4 LT |
2145 | return error; |
2146 | out3: | |
2147 | spin_unlock(&vfsmount_lock); | |
2148 | goto out2; | |
2149 | } | |
2150 | ||
2151 | static void __init init_mount_tree(void) | |
2152 | { | |
2153 | struct vfsmount *mnt; | |
6b3286ed | 2154 | struct mnt_namespace *ns; |
ac748a09 | 2155 | struct path root; |
1da177e4 LT |
2156 | |
2157 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2158 | if (IS_ERR(mnt)) | |
2159 | panic("Can't create rootfs"); | |
6b3286ed KK |
2160 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2161 | if (!ns) | |
1da177e4 | 2162 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2163 | atomic_set(&ns->count, 1); |
2164 | INIT_LIST_HEAD(&ns->list); | |
2165 | init_waitqueue_head(&ns->poll); | |
2166 | ns->event = 0; | |
2167 | list_add(&mnt->mnt_list, &ns->list); | |
2168 | ns->root = mnt; | |
2169 | mnt->mnt_ns = ns; | |
2170 | ||
2171 | init_task.nsproxy->mnt_ns = ns; | |
2172 | get_mnt_ns(ns); | |
2173 | ||
ac748a09 JB |
2174 | root.mnt = ns->root; |
2175 | root.dentry = ns->root->mnt_root; | |
2176 | ||
2177 | set_fs_pwd(current->fs, &root); | |
2178 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2179 | } |
2180 | ||
74bf17cf | 2181 | void __init mnt_init(void) |
1da177e4 | 2182 | { |
13f14b4d | 2183 | unsigned u; |
15a67dd8 | 2184 | int err; |
1da177e4 | 2185 | |
390c6843 RP |
2186 | init_rwsem(&namespace_sem); |
2187 | ||
1da177e4 | 2188 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2189 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2190 | |
b58fed8b | 2191 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2192 | |
2193 | if (!mount_hashtable) | |
2194 | panic("Failed to allocate mount hash table\n"); | |
2195 | ||
13f14b4d ED |
2196 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2197 | ||
2198 | for (u = 0; u < HASH_SIZE; u++) | |
2199 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2200 | |
15a67dd8 RD |
2201 | err = sysfs_init(); |
2202 | if (err) | |
2203 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2204 | __func__, err); |
00d26666 GKH |
2205 | fs_kobj = kobject_create_and_add("fs", NULL); |
2206 | if (!fs_kobj) | |
8e24eea7 | 2207 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2208 | init_rootfs(); |
2209 | init_mount_tree(); | |
2210 | } | |
2211 | ||
6b3286ed | 2212 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2213 | { |
6b3286ed | 2214 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2215 | LIST_HEAD(umount_list); |
6b3286ed | 2216 | ns->root = NULL; |
1ce88cf4 | 2217 | spin_unlock(&vfsmount_lock); |
390c6843 | 2218 | down_write(&namespace_sem); |
1da177e4 | 2219 | spin_lock(&vfsmount_lock); |
a05964f3 | 2220 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2221 | spin_unlock(&vfsmount_lock); |
390c6843 | 2222 | up_write(&namespace_sem); |
70fbcdf4 | 2223 | release_mounts(&umount_list); |
6b3286ed | 2224 | kfree(ns); |
1da177e4 | 2225 | } |