]>
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 | ||
96029c4e | 267 | /** |
268 | * mnt_clone_write - get write access to a mount | |
269 | * @mnt: the mount on which to take a write | |
270 | * | |
271 | * This is effectively like mnt_want_write, except | |
272 | * it must only be used to take an extra write reference | |
273 | * on a mountpoint that we already know has a write reference | |
274 | * on it. This allows some optimisation. | |
275 | * | |
276 | * After finished, mnt_drop_write must be called as usual to | |
277 | * drop the reference. | |
278 | */ | |
279 | int mnt_clone_write(struct vfsmount *mnt) | |
280 | { | |
281 | /* superblock may be r/o */ | |
282 | if (__mnt_is_readonly(mnt)) | |
283 | return -EROFS; | |
284 | preempt_disable(); | |
285 | inc_mnt_writers(mnt); | |
286 | preempt_enable(); | |
287 | return 0; | |
288 | } | |
289 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
290 | ||
291 | /** | |
292 | * mnt_want_write_file - get write access to a file's mount | |
293 | * @file: the file who's mount on which to take a write | |
294 | * | |
295 | * This is like mnt_want_write, but it takes a file and can | |
296 | * do some optimisations if the file is open for write already | |
297 | */ | |
298 | int mnt_want_write_file(struct file *file) | |
299 | { | |
300 | if (!(file->f_mode & FMODE_WRITE)) | |
301 | return mnt_want_write(file->f_path.mnt); | |
302 | else | |
303 | return mnt_clone_write(file->f_path.mnt); | |
304 | } | |
305 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
306 | ||
8366025e DH |
307 | /** |
308 | * mnt_drop_write - give up write access to a mount | |
309 | * @mnt: the mount on which to give up write access | |
310 | * | |
311 | * Tells the low-level filesystem that we are done | |
312 | * performing writes to it. Must be matched with | |
313 | * mnt_want_write() call above. | |
314 | */ | |
315 | void mnt_drop_write(struct vfsmount *mnt) | |
316 | { | |
d3ef3d73 | 317 | preempt_disable(); |
318 | dec_mnt_writers(mnt); | |
319 | preempt_enable(); | |
8366025e DH |
320 | } |
321 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
322 | ||
2e4b7fcd | 323 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 324 | { |
3d733633 DH |
325 | int ret = 0; |
326 | ||
d3ef3d73 | 327 | spin_lock(&vfsmount_lock); |
328 | mnt->mnt_flags |= MNT_WRITE_HOLD; | |
3d733633 | 329 | /* |
d3ef3d73 | 330 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
331 | * should be visible before we do. | |
3d733633 | 332 | */ |
d3ef3d73 | 333 | smp_mb(); |
334 | ||
3d733633 | 335 | /* |
d3ef3d73 | 336 | * With writers on hold, if this value is zero, then there are |
337 | * definitely no active writers (although held writers may subsequently | |
338 | * increment the count, they'll have to wait, and decrement it after | |
339 | * seeing MNT_READONLY). | |
340 | * | |
341 | * It is OK to have counter incremented on one CPU and decremented on | |
342 | * another: the sum will add up correctly. The danger would be when we | |
343 | * sum up each counter, if we read a counter before it is incremented, | |
344 | * but then read another CPU's count which it has been subsequently | |
345 | * decremented from -- we would see more decrements than we should. | |
346 | * MNT_WRITE_HOLD protects against this scenario, because | |
347 | * mnt_want_write first increments count, then smp_mb, then spins on | |
348 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
349 | * we're counting up here. | |
3d733633 | 350 | */ |
d3ef3d73 | 351 | if (count_mnt_writers(mnt) > 0) |
352 | ret = -EBUSY; | |
353 | else | |
2e4b7fcd | 354 | mnt->mnt_flags |= MNT_READONLY; |
d3ef3d73 | 355 | /* |
356 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
357 | * that become unheld will see MNT_READONLY. | |
358 | */ | |
359 | smp_wmb(); | |
360 | mnt->mnt_flags &= ~MNT_WRITE_HOLD; | |
2e4b7fcd | 361 | spin_unlock(&vfsmount_lock); |
3d733633 | 362 | return ret; |
8366025e | 363 | } |
8366025e | 364 | |
2e4b7fcd DH |
365 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
366 | { | |
367 | spin_lock(&vfsmount_lock); | |
368 | mnt->mnt_flags &= ~MNT_READONLY; | |
369 | spin_unlock(&vfsmount_lock); | |
370 | } | |
371 | ||
a3ec947c | 372 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
454e2398 DH |
373 | { |
374 | mnt->mnt_sb = sb; | |
375 | mnt->mnt_root = dget(sb->s_root); | |
454e2398 DH |
376 | } |
377 | ||
378 | EXPORT_SYMBOL(simple_set_mnt); | |
379 | ||
1da177e4 LT |
380 | void free_vfsmnt(struct vfsmount *mnt) |
381 | { | |
382 | kfree(mnt->mnt_devname); | |
73cd49ec | 383 | mnt_free_id(mnt); |
d3ef3d73 | 384 | #ifdef CONFIG_SMP |
385 | free_percpu(mnt->mnt_writers); | |
386 | #endif | |
1da177e4 LT |
387 | kmem_cache_free(mnt_cache, mnt); |
388 | } | |
389 | ||
390 | /* | |
a05964f3 RP |
391 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
392 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 393 | */ |
a05964f3 RP |
394 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
395 | int dir) | |
1da177e4 | 396 | { |
b58fed8b RP |
397 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
398 | struct list_head *tmp = head; | |
1da177e4 LT |
399 | struct vfsmount *p, *found = NULL; |
400 | ||
1da177e4 | 401 | for (;;) { |
a05964f3 | 402 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
403 | p = NULL; |
404 | if (tmp == head) | |
405 | break; | |
406 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
407 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 408 | found = p; |
1da177e4 LT |
409 | break; |
410 | } | |
411 | } | |
1da177e4 LT |
412 | return found; |
413 | } | |
414 | ||
a05964f3 RP |
415 | /* |
416 | * lookup_mnt increments the ref count before returning | |
417 | * the vfsmount struct. | |
418 | */ | |
1c755af4 | 419 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 RP |
420 | { |
421 | struct vfsmount *child_mnt; | |
422 | spin_lock(&vfsmount_lock); | |
1c755af4 | 423 | if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) |
a05964f3 RP |
424 | mntget(child_mnt); |
425 | spin_unlock(&vfsmount_lock); | |
426 | return child_mnt; | |
427 | } | |
428 | ||
1da177e4 LT |
429 | static inline int check_mnt(struct vfsmount *mnt) |
430 | { | |
6b3286ed | 431 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
432 | } |
433 | ||
6b3286ed | 434 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
435 | { |
436 | if (ns) { | |
437 | ns->event = ++event; | |
438 | wake_up_interruptible(&ns->poll); | |
439 | } | |
440 | } | |
441 | ||
6b3286ed | 442 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
443 | { |
444 | if (ns && ns->event != event) { | |
445 | ns->event = event; | |
446 | wake_up_interruptible(&ns->poll); | |
447 | } | |
448 | } | |
449 | ||
1a390689 | 450 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 451 | { |
1a390689 AV |
452 | old_path->dentry = mnt->mnt_mountpoint; |
453 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
454 | mnt->mnt_parent = mnt; |
455 | mnt->mnt_mountpoint = mnt->mnt_root; | |
456 | list_del_init(&mnt->mnt_child); | |
457 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 458 | old_path->dentry->d_mounted--; |
1da177e4 LT |
459 | } |
460 | ||
b90fa9ae RP |
461 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
462 | struct vfsmount *child_mnt) | |
463 | { | |
464 | child_mnt->mnt_parent = mntget(mnt); | |
465 | child_mnt->mnt_mountpoint = dget(dentry); | |
466 | dentry->d_mounted++; | |
467 | } | |
468 | ||
1a390689 | 469 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 470 | { |
1a390689 | 471 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 472 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
473 | hash(path->mnt, path->dentry)); |
474 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
475 | } |
476 | ||
477 | /* | |
478 | * the caller must hold vfsmount_lock | |
479 | */ | |
480 | static void commit_tree(struct vfsmount *mnt) | |
481 | { | |
482 | struct vfsmount *parent = mnt->mnt_parent; | |
483 | struct vfsmount *m; | |
484 | LIST_HEAD(head); | |
6b3286ed | 485 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
486 | |
487 | BUG_ON(parent == mnt); | |
488 | ||
489 | list_add_tail(&head, &mnt->mnt_list); | |
490 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 491 | m->mnt_ns = n; |
b90fa9ae RP |
492 | list_splice(&head, n->list.prev); |
493 | ||
494 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
495 | hash(parent, mnt->mnt_mountpoint)); | |
496 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 497 | touch_mnt_namespace(n); |
1da177e4 LT |
498 | } |
499 | ||
500 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
501 | { | |
502 | struct list_head *next = p->mnt_mounts.next; | |
503 | if (next == &p->mnt_mounts) { | |
504 | while (1) { | |
505 | if (p == root) | |
506 | return NULL; | |
507 | next = p->mnt_child.next; | |
508 | if (next != &p->mnt_parent->mnt_mounts) | |
509 | break; | |
510 | p = p->mnt_parent; | |
511 | } | |
512 | } | |
513 | return list_entry(next, struct vfsmount, mnt_child); | |
514 | } | |
515 | ||
9676f0c6 RP |
516 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
517 | { | |
518 | struct list_head *prev = p->mnt_mounts.prev; | |
519 | while (prev != &p->mnt_mounts) { | |
520 | p = list_entry(prev, struct vfsmount, mnt_child); | |
521 | prev = p->mnt_mounts.prev; | |
522 | } | |
523 | return p; | |
524 | } | |
525 | ||
36341f64 RP |
526 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
527 | int flag) | |
1da177e4 LT |
528 | { |
529 | struct super_block *sb = old->mnt_sb; | |
530 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
531 | ||
532 | if (mnt) { | |
719f5d7f MS |
533 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
534 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
535 | else | |
536 | mnt->mnt_group_id = old->mnt_group_id; | |
537 | ||
538 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
539 | int err = mnt_alloc_group_id(mnt); | |
540 | if (err) | |
541 | goto out_free; | |
542 | } | |
543 | ||
1da177e4 LT |
544 | mnt->mnt_flags = old->mnt_flags; |
545 | atomic_inc(&sb->s_active); | |
546 | mnt->mnt_sb = sb; | |
547 | mnt->mnt_root = dget(root); | |
548 | mnt->mnt_mountpoint = mnt->mnt_root; | |
549 | mnt->mnt_parent = mnt; | |
b90fa9ae | 550 | |
5afe0022 RP |
551 | if (flag & CL_SLAVE) { |
552 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
553 | mnt->mnt_master = old; | |
554 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 555 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
556 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
557 | list_add(&mnt->mnt_share, &old->mnt_share); | |
558 | if (IS_MNT_SLAVE(old)) | |
559 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
560 | mnt->mnt_master = old->mnt_master; | |
561 | } | |
b90fa9ae RP |
562 | if (flag & CL_MAKE_SHARED) |
563 | set_mnt_shared(mnt); | |
1da177e4 LT |
564 | |
565 | /* stick the duplicate mount on the same expiry list | |
566 | * as the original if that was on one */ | |
36341f64 | 567 | if (flag & CL_EXPIRE) { |
36341f64 RP |
568 | if (!list_empty(&old->mnt_expire)) |
569 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 570 | } |
1da177e4 LT |
571 | } |
572 | return mnt; | |
719f5d7f MS |
573 | |
574 | out_free: | |
575 | free_vfsmnt(mnt); | |
576 | return NULL; | |
1da177e4 LT |
577 | } |
578 | ||
7b7b1ace | 579 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
580 | { |
581 | struct super_block *sb = mnt->mnt_sb; | |
3d733633 DH |
582 | /* |
583 | * This probably indicates that somebody messed | |
584 | * up a mnt_want/drop_write() pair. If this | |
585 | * happens, the filesystem was probably unable | |
586 | * to make r/w->r/o transitions. | |
587 | */ | |
d3ef3d73 | 588 | /* |
589 | * atomic_dec_and_lock() used to deal with ->mnt_count decrements | |
590 | * provides barriers, so count_mnt_writers() below is safe. AV | |
591 | */ | |
592 | WARN_ON(count_mnt_writers(mnt)); | |
1da177e4 LT |
593 | dput(mnt->mnt_root); |
594 | free_vfsmnt(mnt); | |
595 | deactivate_super(sb); | |
596 | } | |
597 | ||
7b7b1ace AV |
598 | void mntput_no_expire(struct vfsmount *mnt) |
599 | { | |
600 | repeat: | |
601 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
602 | if (likely(!mnt->mnt_pinned)) { | |
603 | spin_unlock(&vfsmount_lock); | |
604 | __mntput(mnt); | |
605 | return; | |
606 | } | |
607 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
608 | mnt->mnt_pinned = 0; | |
609 | spin_unlock(&vfsmount_lock); | |
610 | acct_auto_close_mnt(mnt); | |
611 | security_sb_umount_close(mnt); | |
612 | goto repeat; | |
613 | } | |
614 | } | |
615 | ||
616 | EXPORT_SYMBOL(mntput_no_expire); | |
617 | ||
618 | void mnt_pin(struct vfsmount *mnt) | |
619 | { | |
620 | spin_lock(&vfsmount_lock); | |
621 | mnt->mnt_pinned++; | |
622 | spin_unlock(&vfsmount_lock); | |
623 | } | |
624 | ||
625 | EXPORT_SYMBOL(mnt_pin); | |
626 | ||
627 | void mnt_unpin(struct vfsmount *mnt) | |
628 | { | |
629 | spin_lock(&vfsmount_lock); | |
630 | if (mnt->mnt_pinned) { | |
631 | atomic_inc(&mnt->mnt_count); | |
632 | mnt->mnt_pinned--; | |
633 | } | |
634 | spin_unlock(&vfsmount_lock); | |
635 | } | |
636 | ||
637 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 638 | |
b3b304a2 MS |
639 | static inline void mangle(struct seq_file *m, const char *s) |
640 | { | |
641 | seq_escape(m, s, " \t\n\\"); | |
642 | } | |
643 | ||
644 | /* | |
645 | * Simple .show_options callback for filesystems which don't want to | |
646 | * implement more complex mount option showing. | |
647 | * | |
648 | * See also save_mount_options(). | |
649 | */ | |
650 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
651 | { | |
2a32cebd AV |
652 | const char *options; |
653 | ||
654 | rcu_read_lock(); | |
655 | options = rcu_dereference(mnt->mnt_sb->s_options); | |
b3b304a2 MS |
656 | |
657 | if (options != NULL && options[0]) { | |
658 | seq_putc(m, ','); | |
659 | mangle(m, options); | |
660 | } | |
2a32cebd | 661 | rcu_read_unlock(); |
b3b304a2 MS |
662 | |
663 | return 0; | |
664 | } | |
665 | EXPORT_SYMBOL(generic_show_options); | |
666 | ||
667 | /* | |
668 | * If filesystem uses generic_show_options(), this function should be | |
669 | * called from the fill_super() callback. | |
670 | * | |
671 | * The .remount_fs callback usually needs to be handled in a special | |
672 | * way, to make sure, that previous options are not overwritten if the | |
673 | * remount fails. | |
674 | * | |
675 | * Also note, that if the filesystem's .remount_fs function doesn't | |
676 | * reset all options to their default value, but changes only newly | |
677 | * given options, then the displayed options will not reflect reality | |
678 | * any more. | |
679 | */ | |
680 | void save_mount_options(struct super_block *sb, char *options) | |
681 | { | |
2a32cebd AV |
682 | BUG_ON(sb->s_options); |
683 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
684 | } |
685 | EXPORT_SYMBOL(save_mount_options); | |
686 | ||
2a32cebd AV |
687 | void replace_mount_options(struct super_block *sb, char *options) |
688 | { | |
689 | char *old = sb->s_options; | |
690 | rcu_assign_pointer(sb->s_options, options); | |
691 | if (old) { | |
692 | synchronize_rcu(); | |
693 | kfree(old); | |
694 | } | |
695 | } | |
696 | EXPORT_SYMBOL(replace_mount_options); | |
697 | ||
a1a2c409 | 698 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
699 | /* iterator */ |
700 | static void *m_start(struct seq_file *m, loff_t *pos) | |
701 | { | |
a1a2c409 | 702 | struct proc_mounts *p = m->private; |
1da177e4 | 703 | |
390c6843 | 704 | down_read(&namespace_sem); |
a1a2c409 | 705 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
706 | } |
707 | ||
708 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
709 | { | |
a1a2c409 | 710 | struct proc_mounts *p = m->private; |
b0765fb8 | 711 | |
a1a2c409 | 712 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
713 | } |
714 | ||
715 | static void m_stop(struct seq_file *m, void *v) | |
716 | { | |
390c6843 | 717 | up_read(&namespace_sem); |
1da177e4 LT |
718 | } |
719 | ||
2d4d4864 RP |
720 | struct proc_fs_info { |
721 | int flag; | |
722 | const char *str; | |
723 | }; | |
724 | ||
2069f457 | 725 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) |
1da177e4 | 726 | { |
2d4d4864 | 727 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
728 | { MS_SYNCHRONOUS, ",sync" }, |
729 | { MS_DIRSYNC, ",dirsync" }, | |
730 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
731 | { 0, NULL } |
732 | }; | |
2d4d4864 RP |
733 | const struct proc_fs_info *fs_infop; |
734 | ||
735 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
736 | if (sb->s_flags & fs_infop->flag) | |
737 | seq_puts(m, fs_infop->str); | |
738 | } | |
2069f457 EP |
739 | |
740 | return security_sb_show_options(m, sb); | |
2d4d4864 RP |
741 | } |
742 | ||
743 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
744 | { | |
745 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
746 | { MNT_NOSUID, ",nosuid" }, |
747 | { MNT_NODEV, ",nodev" }, | |
748 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
749 | { MNT_NOATIME, ",noatime" }, |
750 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 751 | { MNT_RELATIME, ",relatime" }, |
d0adde57 | 752 | { MNT_STRICTATIME, ",strictatime" }, |
1da177e4 LT |
753 | { 0, NULL } |
754 | }; | |
2d4d4864 RP |
755 | const struct proc_fs_info *fs_infop; |
756 | ||
757 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
758 | if (mnt->mnt_flags & fs_infop->flag) | |
759 | seq_puts(m, fs_infop->str); | |
760 | } | |
761 | } | |
762 | ||
763 | static void show_type(struct seq_file *m, struct super_block *sb) | |
764 | { | |
765 | mangle(m, sb->s_type->name); | |
766 | if (sb->s_subtype && sb->s_subtype[0]) { | |
767 | seq_putc(m, '.'); | |
768 | mangle(m, sb->s_subtype); | |
769 | } | |
770 | } | |
771 | ||
772 | static int show_vfsmnt(struct seq_file *m, void *v) | |
773 | { | |
774 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
775 | int err = 0; | |
c32c2f63 | 776 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
777 | |
778 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
779 | seq_putc(m, ' '); | |
c32c2f63 | 780 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 781 | seq_putc(m, ' '); |
2d4d4864 | 782 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 783 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2069f457 EP |
784 | err = show_sb_opts(m, mnt->mnt_sb); |
785 | if (err) | |
786 | goto out; | |
2d4d4864 | 787 | show_mnt_opts(m, mnt); |
1da177e4 LT |
788 | if (mnt->mnt_sb->s_op->show_options) |
789 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
790 | seq_puts(m, " 0 0\n"); | |
2069f457 | 791 | out: |
1da177e4 LT |
792 | return err; |
793 | } | |
794 | ||
a1a2c409 | 795 | const struct seq_operations mounts_op = { |
1da177e4 LT |
796 | .start = m_start, |
797 | .next = m_next, | |
798 | .stop = m_stop, | |
799 | .show = show_vfsmnt | |
800 | }; | |
801 | ||
2d4d4864 RP |
802 | static int show_mountinfo(struct seq_file *m, void *v) |
803 | { | |
804 | struct proc_mounts *p = m->private; | |
805 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
806 | struct super_block *sb = mnt->mnt_sb; | |
807 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
808 | struct path root = p->root; | |
809 | int err = 0; | |
810 | ||
811 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
812 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
813 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
814 | seq_putc(m, ' '); | |
815 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
816 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
817 | /* | |
818 | * Mountpoint is outside root, discard that one. Ugly, | |
819 | * but less so than trying to do that in iterator in a | |
820 | * race-free way (due to renames). | |
821 | */ | |
822 | return SEQ_SKIP; | |
823 | } | |
824 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
825 | show_mnt_opts(m, mnt); | |
826 | ||
827 | /* Tagged fields ("foo:X" or "bar") */ | |
828 | if (IS_MNT_SHARED(mnt)) | |
829 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
830 | if (IS_MNT_SLAVE(mnt)) { |
831 | int master = mnt->mnt_master->mnt_group_id; | |
832 | int dom = get_dominating_id(mnt, &p->root); | |
833 | seq_printf(m, " master:%i", master); | |
834 | if (dom && dom != master) | |
835 | seq_printf(m, " propagate_from:%i", dom); | |
836 | } | |
2d4d4864 RP |
837 | if (IS_MNT_UNBINDABLE(mnt)) |
838 | seq_puts(m, " unbindable"); | |
839 | ||
840 | /* Filesystem specific data */ | |
841 | seq_puts(m, " - "); | |
842 | show_type(m, sb); | |
843 | seq_putc(m, ' '); | |
844 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
845 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
2069f457 EP |
846 | err = show_sb_opts(m, sb); |
847 | if (err) | |
848 | goto out; | |
2d4d4864 RP |
849 | if (sb->s_op->show_options) |
850 | err = sb->s_op->show_options(m, mnt); | |
851 | seq_putc(m, '\n'); | |
2069f457 | 852 | out: |
2d4d4864 RP |
853 | return err; |
854 | } | |
855 | ||
856 | const struct seq_operations mountinfo_op = { | |
857 | .start = m_start, | |
858 | .next = m_next, | |
859 | .stop = m_stop, | |
860 | .show = show_mountinfo, | |
861 | }; | |
862 | ||
b4629fe2 CL |
863 | static int show_vfsstat(struct seq_file *m, void *v) |
864 | { | |
b0765fb8 | 865 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 866 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
867 | int err = 0; |
868 | ||
869 | /* device */ | |
870 | if (mnt->mnt_devname) { | |
871 | seq_puts(m, "device "); | |
872 | mangle(m, mnt->mnt_devname); | |
873 | } else | |
874 | seq_puts(m, "no device"); | |
875 | ||
876 | /* mount point */ | |
877 | seq_puts(m, " mounted on "); | |
c32c2f63 | 878 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
879 | seq_putc(m, ' '); |
880 | ||
881 | /* file system type */ | |
882 | seq_puts(m, "with fstype "); | |
2d4d4864 | 883 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
884 | |
885 | /* optional statistics */ | |
886 | if (mnt->mnt_sb->s_op->show_stats) { | |
887 | seq_putc(m, ' '); | |
888 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
889 | } | |
890 | ||
891 | seq_putc(m, '\n'); | |
892 | return err; | |
893 | } | |
894 | ||
a1a2c409 | 895 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
896 | .start = m_start, |
897 | .next = m_next, | |
898 | .stop = m_stop, | |
899 | .show = show_vfsstat, | |
900 | }; | |
a1a2c409 | 901 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 902 | |
1da177e4 LT |
903 | /** |
904 | * may_umount_tree - check if a mount tree is busy | |
905 | * @mnt: root of mount tree | |
906 | * | |
907 | * This is called to check if a tree of mounts has any | |
908 | * open files, pwds, chroots or sub mounts that are | |
909 | * busy. | |
910 | */ | |
911 | int may_umount_tree(struct vfsmount *mnt) | |
912 | { | |
36341f64 RP |
913 | int actual_refs = 0; |
914 | int minimum_refs = 0; | |
915 | struct vfsmount *p; | |
1da177e4 LT |
916 | |
917 | spin_lock(&vfsmount_lock); | |
36341f64 | 918 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
919 | actual_refs += atomic_read(&p->mnt_count); |
920 | minimum_refs += 2; | |
1da177e4 LT |
921 | } |
922 | spin_unlock(&vfsmount_lock); | |
923 | ||
924 | if (actual_refs > minimum_refs) | |
e3474a8e | 925 | return 0; |
1da177e4 | 926 | |
e3474a8e | 927 | return 1; |
1da177e4 LT |
928 | } |
929 | ||
930 | EXPORT_SYMBOL(may_umount_tree); | |
931 | ||
932 | /** | |
933 | * may_umount - check if a mount point is busy | |
934 | * @mnt: root of mount | |
935 | * | |
936 | * This is called to check if a mount point has any | |
937 | * open files, pwds, chroots or sub mounts. If the | |
938 | * mount has sub mounts this will return busy | |
939 | * regardless of whether the sub mounts are busy. | |
940 | * | |
941 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
942 | * give false negatives. The main reason why it's here is that we need | |
943 | * a non-destructive way to look for easily umountable filesystems. | |
944 | */ | |
945 | int may_umount(struct vfsmount *mnt) | |
946 | { | |
e3474a8e | 947 | int ret = 1; |
a05964f3 RP |
948 | spin_lock(&vfsmount_lock); |
949 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 950 | ret = 0; |
a05964f3 RP |
951 | spin_unlock(&vfsmount_lock); |
952 | return ret; | |
1da177e4 LT |
953 | } |
954 | ||
955 | EXPORT_SYMBOL(may_umount); | |
956 | ||
b90fa9ae | 957 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
958 | { |
959 | struct vfsmount *mnt; | |
bf066c7d | 960 | while (!list_empty(head)) { |
b5e61818 | 961 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
962 | list_del_init(&mnt->mnt_hash); |
963 | if (mnt->mnt_parent != mnt) { | |
964 | struct dentry *dentry; | |
965 | struct vfsmount *m; | |
966 | spin_lock(&vfsmount_lock); | |
967 | dentry = mnt->mnt_mountpoint; | |
968 | m = mnt->mnt_parent; | |
969 | mnt->mnt_mountpoint = mnt->mnt_root; | |
970 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 971 | m->mnt_ghosts--; |
70fbcdf4 RP |
972 | spin_unlock(&vfsmount_lock); |
973 | dput(dentry); | |
974 | mntput(m); | |
975 | } | |
976 | mntput(mnt); | |
977 | } | |
978 | } | |
979 | ||
a05964f3 | 980 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
981 | { |
982 | struct vfsmount *p; | |
1da177e4 | 983 | |
1bfba4e8 AM |
984 | for (p = mnt; p; p = next_mnt(p, mnt)) |
985 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 986 | |
a05964f3 RP |
987 | if (propagate) |
988 | propagate_umount(kill); | |
989 | ||
70fbcdf4 RP |
990 | list_for_each_entry(p, kill, mnt_hash) { |
991 | list_del_init(&p->mnt_expire); | |
992 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
993 | __touch_mnt_namespace(p->mnt_ns); |
994 | p->mnt_ns = NULL; | |
70fbcdf4 | 995 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
996 | if (p->mnt_parent != p) { |
997 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 998 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 999 | } |
a05964f3 | 1000 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1001 | } |
1002 | } | |
1003 | ||
c35038be AV |
1004 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
1005 | ||
1da177e4 LT |
1006 | static int do_umount(struct vfsmount *mnt, int flags) |
1007 | { | |
b58fed8b | 1008 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 1009 | int retval; |
70fbcdf4 | 1010 | LIST_HEAD(umount_list); |
1da177e4 LT |
1011 | |
1012 | retval = security_sb_umount(mnt, flags); | |
1013 | if (retval) | |
1014 | return retval; | |
1015 | ||
1016 | /* | |
1017 | * Allow userspace to request a mountpoint be expired rather than | |
1018 | * unmounting unconditionally. Unmount only happens if: | |
1019 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1020 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1021 | */ | |
1022 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 1023 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
1024 | flags & (MNT_FORCE | MNT_DETACH)) |
1025 | return -EINVAL; | |
1026 | ||
1027 | if (atomic_read(&mnt->mnt_count) != 2) | |
1028 | return -EBUSY; | |
1029 | ||
1030 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1031 | return -EAGAIN; | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * If we may have to abort operations to get out of this | |
1036 | * mount, and they will themselves hold resources we must | |
1037 | * allow the fs to do things. In the Unix tradition of | |
1038 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1039 | * might fail to complete on the first run through as other tasks | |
1040 | * must return, and the like. Thats for the mount program to worry | |
1041 | * about for the moment. | |
1042 | */ | |
1043 | ||
42faad99 | 1044 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1045 | sb->s_op->umount_begin(sb); |
42faad99 | 1046 | } |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * No sense to grab the lock for this test, but test itself looks | |
1050 | * somewhat bogus. Suggestions for better replacement? | |
1051 | * Ho-hum... In principle, we might treat that as umount + switch | |
1052 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1053 | * Actually it makes sense, especially if rootfs would contain a | |
1054 | * /reboot - static binary that would close all descriptors and | |
1055 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1056 | */ | |
6ac08c39 | 1057 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1058 | /* |
1059 | * Special case for "unmounting" root ... | |
1060 | * we just try to remount it readonly. | |
1061 | */ | |
1062 | down_write(&sb->s_umount); | |
4aa98cf7 | 1063 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1064 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1065 | up_write(&sb->s_umount); |
1066 | return retval; | |
1067 | } | |
1068 | ||
390c6843 | 1069 | down_write(&namespace_sem); |
1da177e4 | 1070 | spin_lock(&vfsmount_lock); |
5addc5dd | 1071 | event++; |
1da177e4 | 1072 | |
c35038be AV |
1073 | if (!(flags & MNT_DETACH)) |
1074 | shrink_submounts(mnt, &umount_list); | |
1075 | ||
1da177e4 | 1076 | retval = -EBUSY; |
a05964f3 | 1077 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1078 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1079 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1080 | retval = 0; |
1081 | } | |
1082 | spin_unlock(&vfsmount_lock); | |
1083 | if (retval) | |
1084 | security_sb_umount_busy(mnt); | |
390c6843 | 1085 | up_write(&namespace_sem); |
70fbcdf4 | 1086 | release_mounts(&umount_list); |
1da177e4 LT |
1087 | return retval; |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * Now umount can handle mount points as well as block devices. | |
1092 | * This is important for filesystems which use unnamed block devices. | |
1093 | * | |
1094 | * We now support a flag for forced unmount like the other 'big iron' | |
1095 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1096 | */ | |
1097 | ||
bdc480e3 | 1098 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1099 | { |
2d8f3038 | 1100 | struct path path; |
1da177e4 LT |
1101 | int retval; |
1102 | ||
2d8f3038 | 1103 | retval = user_path(name, &path); |
1da177e4 LT |
1104 | if (retval) |
1105 | goto out; | |
1106 | retval = -EINVAL; | |
2d8f3038 | 1107 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1108 | goto dput_and_out; |
2d8f3038 | 1109 | if (!check_mnt(path.mnt)) |
1da177e4 LT |
1110 | goto dput_and_out; |
1111 | ||
1112 | retval = -EPERM; | |
1113 | if (!capable(CAP_SYS_ADMIN)) | |
1114 | goto dput_and_out; | |
1115 | ||
2d8f3038 | 1116 | retval = do_umount(path.mnt, flags); |
1da177e4 | 1117 | dput_and_out: |
429731b1 | 1118 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 AV |
1119 | dput(path.dentry); |
1120 | mntput_no_expire(path.mnt); | |
1da177e4 LT |
1121 | out: |
1122 | return retval; | |
1123 | } | |
1124 | ||
1125 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1126 | ||
1127 | /* | |
b58fed8b | 1128 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1129 | */ |
bdc480e3 | 1130 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1131 | { |
b58fed8b | 1132 | return sys_umount(name, 0); |
1da177e4 LT |
1133 | } |
1134 | ||
1135 | #endif | |
1136 | ||
2d92ab3c | 1137 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1138 | { |
1139 | if (capable(CAP_SYS_ADMIN)) | |
1140 | return 0; | |
1141 | return -EPERM; | |
1142 | #ifdef notyet | |
2d92ab3c | 1143 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1144 | return -EPERM; |
2d92ab3c | 1145 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1146 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1147 | return -EPERM; |
1148 | } | |
2d92ab3c | 1149 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1150 | return -EPERM; |
1151 | return 0; | |
1152 | #endif | |
1153 | } | |
1154 | ||
b90fa9ae | 1155 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1156 | int flag) |
1da177e4 LT |
1157 | { |
1158 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1159 | struct path path; |
1da177e4 | 1160 | |
9676f0c6 RP |
1161 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1162 | return NULL; | |
1163 | ||
36341f64 | 1164 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1165 | if (!q) |
1166 | goto Enomem; | |
1167 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1168 | ||
1169 | p = mnt; | |
fdadd65f | 1170 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
7ec02ef1 | 1171 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1172 | continue; |
1173 | ||
1174 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1175 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1176 | s = skip_mnt_tree(s); | |
1177 | continue; | |
1178 | } | |
1da177e4 LT |
1179 | while (p != s->mnt_parent) { |
1180 | p = p->mnt_parent; | |
1181 | q = q->mnt_parent; | |
1182 | } | |
1183 | p = s; | |
1a390689 AV |
1184 | path.mnt = q; |
1185 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1186 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1187 | if (!q) |
1188 | goto Enomem; | |
1189 | spin_lock(&vfsmount_lock); | |
1190 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1191 | attach_mnt(q, &path); |
1da177e4 LT |
1192 | spin_unlock(&vfsmount_lock); |
1193 | } | |
1194 | } | |
1195 | return res; | |
b58fed8b | 1196 | Enomem: |
1da177e4 | 1197 | if (res) { |
70fbcdf4 | 1198 | LIST_HEAD(umount_list); |
1da177e4 | 1199 | spin_lock(&vfsmount_lock); |
a05964f3 | 1200 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1201 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1202 | release_mounts(&umount_list); |
1da177e4 LT |
1203 | } |
1204 | return NULL; | |
1205 | } | |
1206 | ||
589ff870 | 1207 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 AV |
1208 | { |
1209 | struct vfsmount *tree; | |
1a60a280 | 1210 | down_write(&namespace_sem); |
589ff870 | 1211 | tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1212 | up_write(&namespace_sem); |
8aec0809 AV |
1213 | return tree; |
1214 | } | |
1215 | ||
1216 | void drop_collected_mounts(struct vfsmount *mnt) | |
1217 | { | |
1218 | LIST_HEAD(umount_list); | |
1a60a280 | 1219 | down_write(&namespace_sem); |
8aec0809 AV |
1220 | spin_lock(&vfsmount_lock); |
1221 | umount_tree(mnt, 0, &umount_list); | |
1222 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1223 | up_write(&namespace_sem); |
8aec0809 AV |
1224 | release_mounts(&umount_list); |
1225 | } | |
1226 | ||
719f5d7f MS |
1227 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1228 | { | |
1229 | struct vfsmount *p; | |
1230 | ||
1231 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1232 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1233 | mnt_release_group_id(p); | |
1234 | } | |
1235 | } | |
1236 | ||
1237 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1238 | { | |
1239 | struct vfsmount *p; | |
1240 | ||
1241 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1242 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1243 | int err = mnt_alloc_group_id(p); | |
1244 | if (err) { | |
1245 | cleanup_group_ids(mnt, p); | |
1246 | return err; | |
1247 | } | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | return 0; | |
1252 | } | |
1253 | ||
b90fa9ae RP |
1254 | /* |
1255 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1256 | * @nd : place the mount tree @source_mnt is attached |
1257 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1258 | * store the parent mount and mountpoint dentry. | |
1259 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1260 | * |
1261 | * NOTE: in the table below explains the semantics when a source mount | |
1262 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1263 | * --------------------------------------------------------------------------- |
1264 | * | BIND MOUNT OPERATION | | |
1265 | * |************************************************************************** | |
1266 | * | source-->| shared | private | slave | unbindable | | |
1267 | * | dest | | | | | | |
1268 | * | | | | | | | | |
1269 | * | v | | | | | | |
1270 | * |************************************************************************** | |
1271 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1272 | * | | | | | | | |
1273 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1274 | * *************************************************************************** | |
b90fa9ae RP |
1275 | * A bind operation clones the source mount and mounts the clone on the |
1276 | * destination mount. | |
1277 | * | |
1278 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1279 | * tree of the destination mount and the cloned mount is added to | |
1280 | * the peer group of the source mount. | |
1281 | * (+) the cloned mount is created under the destination mount and is marked | |
1282 | * as shared. The cloned mount is added to the peer group of the source | |
1283 | * mount. | |
5afe0022 RP |
1284 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1285 | * of the destination mount and the cloned mount is made slave | |
1286 | * of the same master as that of the source mount. The cloned mount | |
1287 | * is marked as 'shared and slave'. | |
1288 | * (*) the cloned mount is made a slave of the same master as that of the | |
1289 | * source mount. | |
1290 | * | |
9676f0c6 RP |
1291 | * --------------------------------------------------------------------------- |
1292 | * | MOVE MOUNT OPERATION | | |
1293 | * |************************************************************************** | |
1294 | * | source-->| shared | private | slave | unbindable | | |
1295 | * | dest | | | | | | |
1296 | * | | | | | | | | |
1297 | * | v | | | | | | |
1298 | * |************************************************************************** | |
1299 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1300 | * | | | | | | | |
1301 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1302 | * *************************************************************************** | |
5afe0022 RP |
1303 | * |
1304 | * (+) the mount is moved to the destination. And is then propagated to | |
1305 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1306 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1307 | * (+++) the mount is moved to the destination and is then propagated to |
1308 | * all the mounts belonging to the destination mount's propagation tree. | |
1309 | * the mount is marked as 'shared and slave'. | |
1310 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1311 | * |
1312 | * if the source mount is a tree, the operations explained above is | |
1313 | * applied to each mount in the tree. | |
1314 | * Must be called without spinlocks held, since this function can sleep | |
1315 | * in allocations. | |
1316 | */ | |
1317 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1318 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1319 | { |
1320 | LIST_HEAD(tree_list); | |
1a390689 AV |
1321 | struct vfsmount *dest_mnt = path->mnt; |
1322 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1323 | struct vfsmount *child, *p; |
719f5d7f | 1324 | int err; |
b90fa9ae | 1325 | |
719f5d7f MS |
1326 | if (IS_MNT_SHARED(dest_mnt)) { |
1327 | err = invent_group_ids(source_mnt, true); | |
1328 | if (err) | |
1329 | goto out; | |
1330 | } | |
1331 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1332 | if (err) | |
1333 | goto out_cleanup_ids; | |
b90fa9ae RP |
1334 | |
1335 | if (IS_MNT_SHARED(dest_mnt)) { | |
1336 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1337 | set_mnt_shared(p); | |
1338 | } | |
1339 | ||
1340 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1341 | if (parent_path) { |
1342 | detach_mnt(source_mnt, parent_path); | |
1343 | attach_mnt(source_mnt, path); | |
e5d67f07 | 1344 | touch_mnt_namespace(parent_path->mnt->mnt_ns); |
21444403 RP |
1345 | } else { |
1346 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1347 | commit_tree(source_mnt); | |
1348 | } | |
b90fa9ae RP |
1349 | |
1350 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1351 | list_del_init(&child->mnt_hash); | |
1352 | commit_tree(child); | |
1353 | } | |
1354 | spin_unlock(&vfsmount_lock); | |
1355 | return 0; | |
719f5d7f MS |
1356 | |
1357 | out_cleanup_ids: | |
1358 | if (IS_MNT_SHARED(dest_mnt)) | |
1359 | cleanup_group_ids(source_mnt, NULL); | |
1360 | out: | |
1361 | return err; | |
b90fa9ae RP |
1362 | } |
1363 | ||
8c3ee42e | 1364 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1365 | { |
1366 | int err; | |
1367 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1368 | return -EINVAL; | |
1369 | ||
8c3ee42e | 1370 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1371 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1372 | return -ENOTDIR; | |
1373 | ||
1374 | err = -ENOENT; | |
8c3ee42e AV |
1375 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1376 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1377 | goto out_unlock; |
1378 | ||
8c3ee42e | 1379 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1380 | if (err) |
1381 | goto out_unlock; | |
1382 | ||
1383 | err = -ENOENT; | |
f3da392e | 1384 | if (!d_unlinked(path->dentry)) |
8c3ee42e | 1385 | err = attach_recursive_mnt(mnt, path, NULL); |
1da177e4 | 1386 | out_unlock: |
8c3ee42e | 1387 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1388 | if (!err) |
8c3ee42e | 1389 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1390 | return err; |
1391 | } | |
1392 | ||
07b20889 RP |
1393 | /* |
1394 | * recursively change the type of the mountpoint. | |
1395 | */ | |
0a0d8a46 | 1396 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1397 | { |
2d92ab3c | 1398 | struct vfsmount *m, *mnt = path->mnt; |
07b20889 RP |
1399 | int recurse = flag & MS_REC; |
1400 | int type = flag & ~MS_REC; | |
719f5d7f | 1401 | int err = 0; |
07b20889 | 1402 | |
ee6f9582 MS |
1403 | if (!capable(CAP_SYS_ADMIN)) |
1404 | return -EPERM; | |
1405 | ||
2d92ab3c | 1406 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1407 | return -EINVAL; |
1408 | ||
1409 | down_write(&namespace_sem); | |
719f5d7f MS |
1410 | if (type == MS_SHARED) { |
1411 | err = invent_group_ids(mnt, recurse); | |
1412 | if (err) | |
1413 | goto out_unlock; | |
1414 | } | |
1415 | ||
07b20889 RP |
1416 | spin_lock(&vfsmount_lock); |
1417 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1418 | change_mnt_propagation(m, type); | |
1419 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1420 | |
1421 | out_unlock: | |
07b20889 | 1422 | up_write(&namespace_sem); |
719f5d7f | 1423 | return err; |
07b20889 RP |
1424 | } |
1425 | ||
1da177e4 LT |
1426 | /* |
1427 | * do loopback mount. | |
1428 | */ | |
0a0d8a46 | 1429 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1430 | int recurse) |
1da177e4 | 1431 | { |
2d92ab3c | 1432 | struct path old_path; |
1da177e4 | 1433 | struct vfsmount *mnt = NULL; |
2d92ab3c | 1434 | int err = mount_is_safe(path); |
1da177e4 LT |
1435 | if (err) |
1436 | return err; | |
1437 | if (!old_name || !*old_name) | |
1438 | return -EINVAL; | |
2d92ab3c | 1439 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1440 | if (err) |
1441 | return err; | |
1442 | ||
390c6843 | 1443 | down_write(&namespace_sem); |
1da177e4 | 1444 | err = -EINVAL; |
2d92ab3c | 1445 | if (IS_MNT_UNBINDABLE(old_path.mnt)) |
4ac91378 | 1446 | goto out; |
9676f0c6 | 1447 | |
2d92ab3c | 1448 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
ccd48bc7 | 1449 | goto out; |
1da177e4 | 1450 | |
ccd48bc7 AV |
1451 | err = -ENOMEM; |
1452 | if (recurse) | |
2d92ab3c | 1453 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 | 1454 | else |
2d92ab3c | 1455 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 AV |
1456 | |
1457 | if (!mnt) | |
1458 | goto out; | |
1459 | ||
2d92ab3c | 1460 | err = graft_tree(mnt, path); |
ccd48bc7 | 1461 | if (err) { |
70fbcdf4 | 1462 | LIST_HEAD(umount_list); |
1da177e4 | 1463 | spin_lock(&vfsmount_lock); |
a05964f3 | 1464 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1465 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1466 | release_mounts(&umount_list); |
5b83d2c5 | 1467 | } |
1da177e4 | 1468 | |
ccd48bc7 | 1469 | out: |
390c6843 | 1470 | up_write(&namespace_sem); |
2d92ab3c | 1471 | path_put(&old_path); |
1da177e4 LT |
1472 | return err; |
1473 | } | |
1474 | ||
2e4b7fcd DH |
1475 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1476 | { | |
1477 | int error = 0; | |
1478 | int readonly_request = 0; | |
1479 | ||
1480 | if (ms_flags & MS_RDONLY) | |
1481 | readonly_request = 1; | |
1482 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1483 | return 0; | |
1484 | ||
1485 | if (readonly_request) | |
1486 | error = mnt_make_readonly(mnt); | |
1487 | else | |
1488 | __mnt_unmake_readonly(mnt); | |
1489 | return error; | |
1490 | } | |
1491 | ||
1da177e4 LT |
1492 | /* |
1493 | * change filesystem flags. dir should be a physical root of filesystem. | |
1494 | * If you've mounted a non-root directory somewhere and want to do remount | |
1495 | * on it - tough luck. | |
1496 | */ | |
0a0d8a46 | 1497 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1498 | void *data) |
1499 | { | |
1500 | int err; | |
2d92ab3c | 1501 | struct super_block *sb = path->mnt->mnt_sb; |
1da177e4 LT |
1502 | |
1503 | if (!capable(CAP_SYS_ADMIN)) | |
1504 | return -EPERM; | |
1505 | ||
2d92ab3c | 1506 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1507 | return -EINVAL; |
1508 | ||
2d92ab3c | 1509 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1510 | return -EINVAL; |
1511 | ||
1512 | down_write(&sb->s_umount); | |
2e4b7fcd | 1513 | if (flags & MS_BIND) |
2d92ab3c | 1514 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1515 | else |
2e4b7fcd | 1516 | err = do_remount_sb(sb, flags, data, 0); |
1da177e4 | 1517 | if (!err) |
2d92ab3c | 1518 | path->mnt->mnt_flags = mnt_flags; |
1da177e4 | 1519 | up_write(&sb->s_umount); |
0e55a7cc | 1520 | if (!err) { |
2d92ab3c | 1521 | security_sb_post_remount(path->mnt, flags, data); |
0e55a7cc DW |
1522 | |
1523 | spin_lock(&vfsmount_lock); | |
1524 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1525 | spin_unlock(&vfsmount_lock); | |
1526 | } | |
1da177e4 LT |
1527 | return err; |
1528 | } | |
1529 | ||
9676f0c6 RP |
1530 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1531 | { | |
1532 | struct vfsmount *p; | |
1533 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1534 | if (IS_MNT_UNBINDABLE(p)) | |
1535 | return 1; | |
1536 | } | |
1537 | return 0; | |
1538 | } | |
1539 | ||
0a0d8a46 | 1540 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1541 | { |
2d92ab3c | 1542 | struct path old_path, parent_path; |
1da177e4 LT |
1543 | struct vfsmount *p; |
1544 | int err = 0; | |
1545 | if (!capable(CAP_SYS_ADMIN)) | |
1546 | return -EPERM; | |
1547 | if (!old_name || !*old_name) | |
1548 | return -EINVAL; | |
2d92ab3c | 1549 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1550 | if (err) |
1551 | return err; | |
1552 | ||
390c6843 | 1553 | down_write(&namespace_sem); |
2d92ab3c | 1554 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1555 | follow_down(path)) |
1da177e4 LT |
1556 | ; |
1557 | err = -EINVAL; | |
2d92ab3c | 1558 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
1da177e4 LT |
1559 | goto out; |
1560 | ||
1561 | err = -ENOENT; | |
2d92ab3c AV |
1562 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1563 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1564 | goto out1; |
1565 | ||
f3da392e | 1566 | if (d_unlinked(path->dentry)) |
21444403 | 1567 | goto out1; |
1da177e4 LT |
1568 | |
1569 | err = -EINVAL; | |
2d92ab3c | 1570 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1571 | goto out1; |
1da177e4 | 1572 | |
2d92ab3c | 1573 | if (old_path.mnt == old_path.mnt->mnt_parent) |
21444403 | 1574 | goto out1; |
1da177e4 | 1575 | |
2d92ab3c AV |
1576 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1577 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1578 | goto out1; |
1579 | /* | |
1580 | * Don't move a mount residing in a shared parent. | |
1581 | */ | |
2d92ab3c AV |
1582 | if (old_path.mnt->mnt_parent && |
1583 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
21444403 | 1584 | goto out1; |
9676f0c6 RP |
1585 | /* |
1586 | * Don't move a mount tree containing unbindable mounts to a destination | |
1587 | * mount which is shared. | |
1588 | */ | |
2d92ab3c AV |
1589 | if (IS_MNT_SHARED(path->mnt) && |
1590 | tree_contains_unbindable(old_path.mnt)) | |
9676f0c6 | 1591 | goto out1; |
1da177e4 | 1592 | err = -ELOOP; |
2d92ab3c AV |
1593 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1594 | if (p == old_path.mnt) | |
21444403 | 1595 | goto out1; |
1da177e4 | 1596 | |
2d92ab3c | 1597 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); |
4ac91378 | 1598 | if (err) |
21444403 | 1599 | goto out1; |
1da177e4 LT |
1600 | |
1601 | /* if the mount is moved, it should no longer be expire | |
1602 | * automatically */ | |
2d92ab3c | 1603 | list_del_init(&old_path.mnt->mnt_expire); |
1da177e4 | 1604 | out1: |
2d92ab3c | 1605 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1606 | out: |
390c6843 | 1607 | up_write(&namespace_sem); |
1da177e4 | 1608 | if (!err) |
1a390689 | 1609 | path_put(&parent_path); |
2d92ab3c | 1610 | path_put(&old_path); |
1da177e4 LT |
1611 | return err; |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * create a new mount for userspace and request it to be added into the | |
1616 | * namespace's tree | |
1617 | */ | |
0a0d8a46 | 1618 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1619 | int mnt_flags, char *name, void *data) |
1620 | { | |
1621 | struct vfsmount *mnt; | |
1622 | ||
1623 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1624 | return -EINVAL; | |
1625 | ||
1626 | /* we need capabilities... */ | |
1627 | if (!capable(CAP_SYS_ADMIN)) | |
1628 | return -EPERM; | |
1629 | ||
7f78d4cd | 1630 | lock_kernel(); |
1da177e4 | 1631 | mnt = do_kern_mount(type, flags, name, data); |
7f78d4cd | 1632 | unlock_kernel(); |
1da177e4 LT |
1633 | if (IS_ERR(mnt)) |
1634 | return PTR_ERR(mnt); | |
1635 | ||
2d92ab3c | 1636 | return do_add_mount(mnt, path, mnt_flags, NULL); |
1da177e4 LT |
1637 | } |
1638 | ||
1639 | /* | |
1640 | * add a mount into a namespace's mount tree | |
1641 | * - provide the option of adding the new mount to an expiration list | |
1642 | */ | |
8d66bf54 | 1643 | int do_add_mount(struct vfsmount *newmnt, struct path *path, |
1da177e4 LT |
1644 | int mnt_flags, struct list_head *fslist) |
1645 | { | |
1646 | int err; | |
1647 | ||
390c6843 | 1648 | down_write(&namespace_sem); |
1da177e4 | 1649 | /* Something was mounted here while we slept */ |
8d66bf54 | 1650 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1651 | follow_down(path)) |
1da177e4 LT |
1652 | ; |
1653 | err = -EINVAL; | |
dd5cae6e | 1654 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) |
1da177e4 LT |
1655 | goto unlock; |
1656 | ||
1657 | /* Refuse the same filesystem on the same mount point */ | |
1658 | err = -EBUSY; | |
8d66bf54 AV |
1659 | if (path->mnt->mnt_sb == newmnt->mnt_sb && |
1660 | path->mnt->mnt_root == path->dentry) | |
1da177e4 LT |
1661 | goto unlock; |
1662 | ||
1663 | err = -EINVAL; | |
1664 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1665 | goto unlock; | |
1666 | ||
1667 | newmnt->mnt_flags = mnt_flags; | |
8d66bf54 | 1668 | if ((err = graft_tree(newmnt, path))) |
5b83d2c5 | 1669 | goto unlock; |
1da177e4 | 1670 | |
6758f953 | 1671 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1672 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1673 | |
390c6843 | 1674 | up_write(&namespace_sem); |
5b83d2c5 | 1675 | return 0; |
1da177e4 LT |
1676 | |
1677 | unlock: | |
390c6843 | 1678 | up_write(&namespace_sem); |
1da177e4 LT |
1679 | mntput(newmnt); |
1680 | return err; | |
1681 | } | |
1682 | ||
1683 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1684 | ||
1685 | /* | |
1686 | * process a list of expirable mountpoints with the intent of discarding any | |
1687 | * mountpoints that aren't in use and haven't been touched since last we came | |
1688 | * here | |
1689 | */ | |
1690 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1691 | { | |
1da177e4 LT |
1692 | struct vfsmount *mnt, *next; |
1693 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1694 | LIST_HEAD(umounts); |
1da177e4 LT |
1695 | |
1696 | if (list_empty(mounts)) | |
1697 | return; | |
1698 | ||
bcc5c7d2 | 1699 | down_write(&namespace_sem); |
1da177e4 LT |
1700 | spin_lock(&vfsmount_lock); |
1701 | ||
1702 | /* extract from the expiration list every vfsmount that matches the | |
1703 | * following criteria: | |
1704 | * - only referenced by its parent vfsmount | |
1705 | * - still marked for expiry (marked on the last call here; marks are | |
1706 | * cleared by mntput()) | |
1707 | */ | |
55e700b9 | 1708 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1709 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1710 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1711 | continue; |
55e700b9 | 1712 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1713 | } |
bcc5c7d2 AV |
1714 | while (!list_empty(&graveyard)) { |
1715 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1716 | touch_mnt_namespace(mnt->mnt_ns); | |
1717 | umount_tree(mnt, 1, &umounts); | |
1718 | } | |
5528f911 | 1719 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1720 | up_write(&namespace_sem); |
1721 | ||
1722 | release_mounts(&umounts); | |
5528f911 TM |
1723 | } |
1724 | ||
1725 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1726 | ||
1727 | /* | |
1728 | * Ripoff of 'select_parent()' | |
1729 | * | |
1730 | * search the list of submounts for a given mountpoint, and move any | |
1731 | * shrinkable submounts to the 'graveyard' list. | |
1732 | */ | |
1733 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1734 | { | |
1735 | struct vfsmount *this_parent = parent; | |
1736 | struct list_head *next; | |
1737 | int found = 0; | |
1738 | ||
1739 | repeat: | |
1740 | next = this_parent->mnt_mounts.next; | |
1741 | resume: | |
1742 | while (next != &this_parent->mnt_mounts) { | |
1743 | struct list_head *tmp = next; | |
1744 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1745 | ||
1746 | next = tmp->next; | |
1747 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1748 | continue; |
5528f911 TM |
1749 | /* |
1750 | * Descend a level if the d_mounts list is non-empty. | |
1751 | */ | |
1752 | if (!list_empty(&mnt->mnt_mounts)) { | |
1753 | this_parent = mnt; | |
1754 | goto repeat; | |
1755 | } | |
1da177e4 | 1756 | |
5528f911 | 1757 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1758 | list_move_tail(&mnt->mnt_expire, graveyard); |
1759 | found++; | |
1760 | } | |
1da177e4 | 1761 | } |
5528f911 TM |
1762 | /* |
1763 | * All done at this level ... ascend and resume the search | |
1764 | */ | |
1765 | if (this_parent != parent) { | |
1766 | next = this_parent->mnt_child.next; | |
1767 | this_parent = this_parent->mnt_parent; | |
1768 | goto resume; | |
1769 | } | |
1770 | return found; | |
1771 | } | |
1772 | ||
1773 | /* | |
1774 | * process a list of expirable mountpoints with the intent of discarding any | |
1775 | * submounts of a specific parent mountpoint | |
1776 | */ | |
c35038be | 1777 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1778 | { |
1779 | LIST_HEAD(graveyard); | |
c35038be | 1780 | struct vfsmount *m; |
5528f911 | 1781 | |
5528f911 | 1782 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1783 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1784 | while (!list_empty(&graveyard)) { |
c35038be | 1785 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 | 1786 | mnt_expire); |
afef80b3 EB |
1787 | touch_mnt_namespace(m->mnt_ns); |
1788 | umount_tree(m, 1, umounts); | |
bcc5c7d2 AV |
1789 | } |
1790 | } | |
1da177e4 LT |
1791 | } |
1792 | ||
1da177e4 LT |
1793 | /* |
1794 | * Some copy_from_user() implementations do not return the exact number of | |
1795 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1796 | * Note that this function differs from copy_from_user() in that it will oops | |
1797 | * on bad values of `to', rather than returning a short copy. | |
1798 | */ | |
b58fed8b RP |
1799 | static long exact_copy_from_user(void *to, const void __user * from, |
1800 | unsigned long n) | |
1da177e4 LT |
1801 | { |
1802 | char *t = to; | |
1803 | const char __user *f = from; | |
1804 | char c; | |
1805 | ||
1806 | if (!access_ok(VERIFY_READ, from, n)) | |
1807 | return n; | |
1808 | ||
1809 | while (n) { | |
1810 | if (__get_user(c, f)) { | |
1811 | memset(t, 0, n); | |
1812 | break; | |
1813 | } | |
1814 | *t++ = c; | |
1815 | f++; | |
1816 | n--; | |
1817 | } | |
1818 | return n; | |
1819 | } | |
1820 | ||
b58fed8b | 1821 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1822 | { |
1823 | int i; | |
1824 | unsigned long page; | |
1825 | unsigned long size; | |
b58fed8b | 1826 | |
1da177e4 LT |
1827 | *where = 0; |
1828 | if (!data) | |
1829 | return 0; | |
1830 | ||
1831 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1832 | return -ENOMEM; | |
1833 | ||
1834 | /* We only care that *some* data at the address the user | |
1835 | * gave us is valid. Just in case, we'll zero | |
1836 | * the remainder of the page. | |
1837 | */ | |
1838 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1839 | size = TASK_SIZE - (unsigned long)data; | |
1840 | if (size > PAGE_SIZE) | |
1841 | size = PAGE_SIZE; | |
1842 | ||
1843 | i = size - exact_copy_from_user((void *)page, data, size); | |
1844 | if (!i) { | |
b58fed8b | 1845 | free_page(page); |
1da177e4 LT |
1846 | return -EFAULT; |
1847 | } | |
1848 | if (i != PAGE_SIZE) | |
1849 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1850 | *where = page; | |
1851 | return 0; | |
1852 | } | |
1853 | ||
1854 | /* | |
1855 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1856 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1857 | * | |
1858 | * data is a (void *) that can point to any structure up to | |
1859 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1860 | * information (or be NULL). | |
1861 | * | |
1862 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1863 | * When the flags word was introduced its top half was required | |
1864 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1865 | * Therefore, if this magic number is present, it carries no information | |
1866 | * and must be discarded. | |
1867 | */ | |
b58fed8b | 1868 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1869 | unsigned long flags, void *data_page) |
1870 | { | |
2d92ab3c | 1871 | struct path path; |
1da177e4 LT |
1872 | int retval = 0; |
1873 | int mnt_flags = 0; | |
1874 | ||
1875 | /* Discard magic */ | |
1876 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1877 | flags &= ~MS_MGC_MSK; | |
1878 | ||
1879 | /* Basic sanity checks */ | |
1880 | ||
1881 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1882 | return -EINVAL; | |
1883 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1884 | return -EINVAL; | |
1885 | ||
1886 | if (data_page) | |
1887 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1888 | ||
613cbe3d AK |
1889 | /* Default to relatime unless overriden */ |
1890 | if (!(flags & MS_NOATIME)) | |
1891 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 1892 | |
1da177e4 LT |
1893 | /* Separate the per-mountpoint flags */ |
1894 | if (flags & MS_NOSUID) | |
1895 | mnt_flags |= MNT_NOSUID; | |
1896 | if (flags & MS_NODEV) | |
1897 | mnt_flags |= MNT_NODEV; | |
1898 | if (flags & MS_NOEXEC) | |
1899 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1900 | if (flags & MS_NOATIME) |
1901 | mnt_flags |= MNT_NOATIME; | |
1902 | if (flags & MS_NODIRATIME) | |
1903 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
1904 | if (flags & MS_STRICTATIME) |
1905 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
1906 | if (flags & MS_RDONLY) |
1907 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1908 | |
1909 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
d0adde57 MG |
1910 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
1911 | MS_STRICTATIME); | |
1da177e4 LT |
1912 | |
1913 | /* ... and get the mountpoint */ | |
2d92ab3c | 1914 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1915 | if (retval) |
1916 | return retval; | |
1917 | ||
2d92ab3c | 1918 | retval = security_sb_mount(dev_name, &path, |
b5266eb4 | 1919 | type_page, flags, data_page); |
1da177e4 LT |
1920 | if (retval) |
1921 | goto dput_out; | |
1922 | ||
1923 | if (flags & MS_REMOUNT) | |
2d92ab3c | 1924 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
1925 | data_page); |
1926 | else if (flags & MS_BIND) | |
2d92ab3c | 1927 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 1928 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 1929 | retval = do_change_type(&path, flags); |
1da177e4 | 1930 | else if (flags & MS_MOVE) |
2d92ab3c | 1931 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 1932 | else |
2d92ab3c | 1933 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
1934 | dev_name, data_page); |
1935 | dput_out: | |
2d92ab3c | 1936 | path_put(&path); |
1da177e4 LT |
1937 | return retval; |
1938 | } | |
1939 | ||
741a2951 JD |
1940 | /* |
1941 | * Allocate a new namespace structure and populate it with contents | |
1942 | * copied from the namespace of the passed in task structure. | |
1943 | */ | |
e3222c4e | 1944 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1945 | struct fs_struct *fs) |
1da177e4 | 1946 | { |
6b3286ed | 1947 | struct mnt_namespace *new_ns; |
7f2da1e7 | 1948 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
1da177e4 LT |
1949 | struct vfsmount *p, *q; |
1950 | ||
6b3286ed | 1951 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1952 | if (!new_ns) |
467e9f4b | 1953 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1954 | |
1955 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1956 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1957 | init_waitqueue_head(&new_ns->poll); |
1958 | new_ns->event = 0; | |
1da177e4 | 1959 | |
390c6843 | 1960 | down_write(&namespace_sem); |
1da177e4 | 1961 | /* First pass: copy the tree topology */ |
6b3286ed | 1962 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1963 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1964 | if (!new_ns->root) { |
390c6843 | 1965 | up_write(&namespace_sem); |
1da177e4 | 1966 | kfree(new_ns); |
5cc4a034 | 1967 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1968 | } |
1969 | spin_lock(&vfsmount_lock); | |
1970 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1971 | spin_unlock(&vfsmount_lock); | |
1972 | ||
1973 | /* | |
1974 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1975 | * as belonging to new namespace. We have already acquired a private | |
1976 | * fs_struct, so tsk->fs->lock is not needed. | |
1977 | */ | |
6b3286ed | 1978 | p = mnt_ns->root; |
1da177e4 LT |
1979 | q = new_ns->root; |
1980 | while (p) { | |
6b3286ed | 1981 | q->mnt_ns = new_ns; |
1da177e4 | 1982 | if (fs) { |
6ac08c39 | 1983 | if (p == fs->root.mnt) { |
1da177e4 | 1984 | rootmnt = p; |
6ac08c39 | 1985 | fs->root.mnt = mntget(q); |
1da177e4 | 1986 | } |
6ac08c39 | 1987 | if (p == fs->pwd.mnt) { |
1da177e4 | 1988 | pwdmnt = p; |
6ac08c39 | 1989 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1990 | } |
1da177e4 | 1991 | } |
6b3286ed | 1992 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1993 | q = next_mnt(q, new_ns->root); |
1994 | } | |
390c6843 | 1995 | up_write(&namespace_sem); |
1da177e4 | 1996 | |
1da177e4 LT |
1997 | if (rootmnt) |
1998 | mntput(rootmnt); | |
1999 | if (pwdmnt) | |
2000 | mntput(pwdmnt); | |
1da177e4 | 2001 | |
741a2951 JD |
2002 | return new_ns; |
2003 | } | |
2004 | ||
213dd266 | 2005 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2006 | struct fs_struct *new_fs) |
741a2951 | 2007 | { |
6b3286ed | 2008 | struct mnt_namespace *new_ns; |
741a2951 | 2009 | |
e3222c4e | 2010 | BUG_ON(!ns); |
6b3286ed | 2011 | get_mnt_ns(ns); |
741a2951 JD |
2012 | |
2013 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2014 | return ns; |
741a2951 | 2015 | |
e3222c4e | 2016 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2017 | |
6b3286ed | 2018 | put_mnt_ns(ns); |
e3222c4e | 2019 | return new_ns; |
1da177e4 LT |
2020 | } |
2021 | ||
bdc480e3 HC |
2022 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2023 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 LT |
2024 | { |
2025 | int retval; | |
2026 | unsigned long data_page; | |
2027 | unsigned long type_page; | |
2028 | unsigned long dev_page; | |
2029 | char *dir_page; | |
2030 | ||
b58fed8b | 2031 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
2032 | if (retval < 0) |
2033 | return retval; | |
2034 | ||
2035 | dir_page = getname(dir_name); | |
2036 | retval = PTR_ERR(dir_page); | |
2037 | if (IS_ERR(dir_page)) | |
2038 | goto out1; | |
2039 | ||
b58fed8b | 2040 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2041 | if (retval < 0) |
2042 | goto out2; | |
2043 | ||
b58fed8b | 2044 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2045 | if (retval < 0) |
2046 | goto out3; | |
2047 | ||
b58fed8b RP |
2048 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2049 | flags, (void *)data_page); | |
1da177e4 LT |
2050 | free_page(data_page); |
2051 | ||
2052 | out3: | |
2053 | free_page(dev_page); | |
2054 | out2: | |
2055 | putname(dir_page); | |
2056 | out1: | |
2057 | free_page(type_page); | |
2058 | return retval; | |
2059 | } | |
2060 | ||
1da177e4 LT |
2061 | /* |
2062 | * pivot_root Semantics: | |
2063 | * Moves the root file system of the current process to the directory put_old, | |
2064 | * makes new_root as the new root file system of the current process, and sets | |
2065 | * root/cwd of all processes which had them on the current root to new_root. | |
2066 | * | |
2067 | * Restrictions: | |
2068 | * The new_root and put_old must be directories, and must not be on the | |
2069 | * same file system as the current process root. The put_old must be | |
2070 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2071 | * pointed to by put_old must yield the same directory as new_root. No other | |
2072 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2073 | * | |
4a0d11fa NB |
2074 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2075 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2076 | * in this situation. | |
2077 | * | |
1da177e4 LT |
2078 | * Notes: |
2079 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2080 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2081 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2082 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2083 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2084 | * first. | |
2085 | */ | |
3480b257 HC |
2086 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2087 | const char __user *, put_old) | |
1da177e4 LT |
2088 | { |
2089 | struct vfsmount *tmp; | |
2d8f3038 | 2090 | struct path new, old, parent_path, root_parent, root; |
1da177e4 LT |
2091 | int error; |
2092 | ||
2093 | if (!capable(CAP_SYS_ADMIN)) | |
2094 | return -EPERM; | |
2095 | ||
2d8f3038 | 2096 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2097 | if (error) |
2098 | goto out0; | |
2099 | error = -EINVAL; | |
2d8f3038 | 2100 | if (!check_mnt(new.mnt)) |
1da177e4 LT |
2101 | goto out1; |
2102 | ||
2d8f3038 | 2103 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2104 | if (error) |
2105 | goto out1; | |
2106 | ||
2d8f3038 | 2107 | error = security_sb_pivotroot(&old, &new); |
1da177e4 | 2108 | if (error) { |
2d8f3038 | 2109 | path_put(&old); |
1da177e4 LT |
2110 | goto out1; |
2111 | } | |
2112 | ||
2113 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2114 | root = current->fs->root; |
6ac08c39 | 2115 | path_get(¤t->fs->root); |
1da177e4 | 2116 | read_unlock(¤t->fs->lock); |
390c6843 | 2117 | down_write(&namespace_sem); |
2d8f3038 | 2118 | mutex_lock(&old.dentry->d_inode->i_mutex); |
1da177e4 | 2119 | error = -EINVAL; |
2d8f3038 AV |
2120 | if (IS_MNT_SHARED(old.mnt) || |
2121 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
8c3ee42e | 2122 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2123 | goto out2; |
8c3ee42e | 2124 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2125 | goto out2; |
2126 | error = -ENOENT; | |
2d8f3038 | 2127 | if (IS_DEADDIR(new.dentry->d_inode)) |
1da177e4 | 2128 | goto out2; |
f3da392e | 2129 | if (d_unlinked(new.dentry)) |
1da177e4 | 2130 | goto out2; |
f3da392e | 2131 | if (d_unlinked(old.dentry)) |
1da177e4 LT |
2132 | goto out2; |
2133 | error = -EBUSY; | |
2d8f3038 AV |
2134 | if (new.mnt == root.mnt || |
2135 | old.mnt == root.mnt) | |
1da177e4 LT |
2136 | goto out2; /* loop, on the same file system */ |
2137 | error = -EINVAL; | |
8c3ee42e | 2138 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2139 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2140 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2141 | goto out2; /* not attached */ |
2d8f3038 | 2142 | if (new.mnt->mnt_root != new.dentry) |
1da177e4 | 2143 | goto out2; /* not a mountpoint */ |
2d8f3038 | 2144 | if (new.mnt->mnt_parent == new.mnt) |
0bb6fcc1 | 2145 | goto out2; /* not attached */ |
4ac91378 | 2146 | /* make sure we can reach put_old from new_root */ |
2d8f3038 | 2147 | tmp = old.mnt; |
1da177e4 | 2148 | spin_lock(&vfsmount_lock); |
2d8f3038 | 2149 | if (tmp != new.mnt) { |
1da177e4 LT |
2150 | for (;;) { |
2151 | if (tmp->mnt_parent == tmp) | |
2152 | goto out3; /* already mounted on put_old */ | |
2d8f3038 | 2153 | if (tmp->mnt_parent == new.mnt) |
1da177e4 LT |
2154 | break; |
2155 | tmp = tmp->mnt_parent; | |
2156 | } | |
2d8f3038 | 2157 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) |
1da177e4 | 2158 | goto out3; |
2d8f3038 | 2159 | } else if (!is_subdir(old.dentry, new.dentry)) |
1da177e4 | 2160 | goto out3; |
2d8f3038 | 2161 | detach_mnt(new.mnt, &parent_path); |
8c3ee42e | 2162 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2163 | /* mount old root on put_old */ |
2d8f3038 | 2164 | attach_mnt(root.mnt, &old); |
4ac91378 | 2165 | /* mount new_root on / */ |
2d8f3038 | 2166 | attach_mnt(new.mnt, &root_parent); |
6b3286ed | 2167 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2168 | spin_unlock(&vfsmount_lock); |
2d8f3038 AV |
2169 | chroot_fs_refs(&root, &new); |
2170 | security_sb_post_pivotroot(&root, &new); | |
1da177e4 | 2171 | error = 0; |
1a390689 AV |
2172 | path_put(&root_parent); |
2173 | path_put(&parent_path); | |
1da177e4 | 2174 | out2: |
2d8f3038 | 2175 | mutex_unlock(&old.dentry->d_inode->i_mutex); |
390c6843 | 2176 | up_write(&namespace_sem); |
8c3ee42e | 2177 | path_put(&root); |
2d8f3038 | 2178 | path_put(&old); |
1da177e4 | 2179 | out1: |
2d8f3038 | 2180 | path_put(&new); |
1da177e4 | 2181 | out0: |
1da177e4 LT |
2182 | return error; |
2183 | out3: | |
2184 | spin_unlock(&vfsmount_lock); | |
2185 | goto out2; | |
2186 | } | |
2187 | ||
2188 | static void __init init_mount_tree(void) | |
2189 | { | |
2190 | struct vfsmount *mnt; | |
6b3286ed | 2191 | struct mnt_namespace *ns; |
ac748a09 | 2192 | struct path root; |
1da177e4 LT |
2193 | |
2194 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2195 | if (IS_ERR(mnt)) | |
2196 | panic("Can't create rootfs"); | |
6b3286ed KK |
2197 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2198 | if (!ns) | |
1da177e4 | 2199 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2200 | atomic_set(&ns->count, 1); |
2201 | INIT_LIST_HEAD(&ns->list); | |
2202 | init_waitqueue_head(&ns->poll); | |
2203 | ns->event = 0; | |
2204 | list_add(&mnt->mnt_list, &ns->list); | |
2205 | ns->root = mnt; | |
2206 | mnt->mnt_ns = ns; | |
2207 | ||
2208 | init_task.nsproxy->mnt_ns = ns; | |
2209 | get_mnt_ns(ns); | |
2210 | ||
ac748a09 JB |
2211 | root.mnt = ns->root; |
2212 | root.dentry = ns->root->mnt_root; | |
2213 | ||
2214 | set_fs_pwd(current->fs, &root); | |
2215 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2216 | } |
2217 | ||
74bf17cf | 2218 | void __init mnt_init(void) |
1da177e4 | 2219 | { |
13f14b4d | 2220 | unsigned u; |
15a67dd8 | 2221 | int err; |
1da177e4 | 2222 | |
390c6843 RP |
2223 | init_rwsem(&namespace_sem); |
2224 | ||
1da177e4 | 2225 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2226 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2227 | |
b58fed8b | 2228 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2229 | |
2230 | if (!mount_hashtable) | |
2231 | panic("Failed to allocate mount hash table\n"); | |
2232 | ||
13f14b4d ED |
2233 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2234 | ||
2235 | for (u = 0; u < HASH_SIZE; u++) | |
2236 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2237 | |
15a67dd8 RD |
2238 | err = sysfs_init(); |
2239 | if (err) | |
2240 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2241 | __func__, err); |
00d26666 GKH |
2242 | fs_kobj = kobject_create_and_add("fs", NULL); |
2243 | if (!fs_kobj) | |
8e24eea7 | 2244 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2245 | init_rootfs(); |
2246 | init_mount_tree(); | |
2247 | } | |
2248 | ||
6b3286ed | 2249 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2250 | { |
6b3286ed | 2251 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2252 | LIST_HEAD(umount_list); |
6b3286ed | 2253 | ns->root = NULL; |
1ce88cf4 | 2254 | spin_unlock(&vfsmount_lock); |
390c6843 | 2255 | down_write(&namespace_sem); |
1da177e4 | 2256 | spin_lock(&vfsmount_lock); |
a05964f3 | 2257 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2258 | spin_unlock(&vfsmount_lock); |
390c6843 | 2259 | up_write(&namespace_sem); |
70fbcdf4 | 2260 | release_mounts(&umount_list); |
6b3286ed | 2261 | kfree(ns); |
1da177e4 | 2262 | } |