]>
Commit | Line | Data |
---|---|---|
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 | ||
11 | #include <linux/syscalls.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/spinlock.h> | |
15 | #include <linux/percpu.h> | |
16 | #include <linux/init.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <linux/acct.h> | |
19 | #include <linux/capability.h> | |
20 | #include <linux/cpumask.h> | |
21 | #include <linux/module.h> | |
22 | #include <linux/sysfs.h> | |
23 | #include <linux/seq_file.h> | |
24 | #include <linux/mnt_namespace.h> | |
25 | #include <linux/namei.h> | |
26 | #include <linux/nsproxy.h> | |
27 | #include <linux/security.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/ramfs.h> | |
30 | #include <linux/log2.h> | |
31 | #include <linux/idr.h> | |
32 | #include <linux/fs_struct.h> | |
33 | #include <linux/fsnotify.h> | |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/unistd.h> | |
36 | #include "pnode.h" | |
37 | #include "internal.h" | |
38 | ||
39 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) | |
40 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
41 | ||
42 | static int event; | |
43 | static DEFINE_IDA(mnt_id_ida); | |
44 | static DEFINE_IDA(mnt_group_ida); | |
45 | static DEFINE_SPINLOCK(mnt_id_lock); | |
46 | static int mnt_id_start = 0; | |
47 | static int mnt_group_start = 1; | |
48 | ||
49 | static struct list_head *mount_hashtable __read_mostly; | |
50 | static struct kmem_cache *mnt_cache __read_mostly; | |
51 | static struct rw_semaphore namespace_sem; | |
52 | ||
53 | /* /sys/fs */ | |
54 | struct kobject *fs_kobj; | |
55 | EXPORT_SYMBOL_GPL(fs_kobj); | |
56 | ||
57 | /* | |
58 | * vfsmount lock may be taken for read to prevent changes to the | |
59 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
60 | * up the tree. | |
61 | * | |
62 | * It should be taken for write in all cases where the vfsmount | |
63 | * tree or hash is modified or when a vfsmount structure is modified. | |
64 | */ | |
65 | DEFINE_BRLOCK(vfsmount_lock); | |
66 | ||
67 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) | |
68 | { | |
69 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); | |
70 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
71 | tmp = tmp + (tmp >> HASH_SHIFT); | |
72 | return tmp & (HASH_SIZE - 1); | |
73 | } | |
74 | ||
75 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) | |
76 | ||
77 | /* | |
78 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
79 | * serialize with freeing. | |
80 | */ | |
81 | static int mnt_alloc_id(struct vfsmount *mnt) | |
82 | { | |
83 | int res; | |
84 | ||
85 | retry: | |
86 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
87 | spin_lock(&mnt_id_lock); | |
88 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); | |
89 | if (!res) | |
90 | mnt_id_start = mnt->mnt_id + 1; | |
91 | spin_unlock(&mnt_id_lock); | |
92 | if (res == -EAGAIN) | |
93 | goto retry; | |
94 | ||
95 | return res; | |
96 | } | |
97 | ||
98 | static void mnt_free_id(struct vfsmount *mnt) | |
99 | { | |
100 | int id = mnt->mnt_id; | |
101 | spin_lock(&mnt_id_lock); | |
102 | ida_remove(&mnt_id_ida, id); | |
103 | if (mnt_id_start > id) | |
104 | mnt_id_start = id; | |
105 | spin_unlock(&mnt_id_lock); | |
106 | } | |
107 | ||
108 | /* | |
109 | * Allocate a new peer group ID | |
110 | * | |
111 | * mnt_group_ida is protected by namespace_sem | |
112 | */ | |
113 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
114 | { | |
115 | int res; | |
116 | ||
117 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) | |
118 | return -ENOMEM; | |
119 | ||
120 | res = ida_get_new_above(&mnt_group_ida, | |
121 | mnt_group_start, | |
122 | &mnt->mnt_group_id); | |
123 | if (!res) | |
124 | mnt_group_start = mnt->mnt_group_id + 1; | |
125 | ||
126 | return res; | |
127 | } | |
128 | ||
129 | /* | |
130 | * Release a peer group ID | |
131 | */ | |
132 | void mnt_release_group_id(struct vfsmount *mnt) | |
133 | { | |
134 | int id = mnt->mnt_group_id; | |
135 | ida_remove(&mnt_group_ida, id); | |
136 | if (mnt_group_start > id) | |
137 | mnt_group_start = id; | |
138 | mnt->mnt_group_id = 0; | |
139 | } | |
140 | ||
141 | /* | |
142 | * vfsmount lock must be held for read | |
143 | */ | |
144 | static inline void mnt_add_count(struct vfsmount *mnt, int n) | |
145 | { | |
146 | #ifdef CONFIG_SMP | |
147 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); | |
148 | #else | |
149 | preempt_disable(); | |
150 | mnt->mnt_count += n; | |
151 | preempt_enable(); | |
152 | #endif | |
153 | } | |
154 | ||
155 | static inline void mnt_set_count(struct vfsmount *mnt, int n) | |
156 | { | |
157 | #ifdef CONFIG_SMP | |
158 | this_cpu_write(mnt->mnt_pcp->mnt_count, n); | |
159 | #else | |
160 | mnt->mnt_count = n; | |
161 | #endif | |
162 | } | |
163 | ||
164 | /* | |
165 | * vfsmount lock must be held for read | |
166 | */ | |
167 | static inline void mnt_inc_count(struct vfsmount *mnt) | |
168 | { | |
169 | mnt_add_count(mnt, 1); | |
170 | } | |
171 | ||
172 | /* | |
173 | * vfsmount lock must be held for read | |
174 | */ | |
175 | static inline void mnt_dec_count(struct vfsmount *mnt) | |
176 | { | |
177 | mnt_add_count(mnt, -1); | |
178 | } | |
179 | ||
180 | /* | |
181 | * vfsmount lock must be held for write | |
182 | */ | |
183 | unsigned int mnt_get_count(struct vfsmount *mnt) | |
184 | { | |
185 | #ifdef CONFIG_SMP | |
186 | unsigned int count = 0; | |
187 | int cpu; | |
188 | ||
189 | for_each_possible_cpu(cpu) { | |
190 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; | |
191 | } | |
192 | ||
193 | return count; | |
194 | #else | |
195 | return mnt->mnt_count; | |
196 | #endif | |
197 | } | |
198 | ||
199 | struct vfsmount *alloc_vfsmnt(const char *name) | |
200 | { | |
201 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); | |
202 | if (mnt) { | |
203 | int err; | |
204 | ||
205 | err = mnt_alloc_id(mnt); | |
206 | if (err) | |
207 | goto out_free_cache; | |
208 | ||
209 | if (name) { | |
210 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
211 | if (!mnt->mnt_devname) | |
212 | goto out_free_id; | |
213 | } | |
214 | ||
215 | #ifdef CONFIG_SMP | |
216 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); | |
217 | if (!mnt->mnt_pcp) | |
218 | goto out_free_devname; | |
219 | ||
220 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); | |
221 | #else | |
222 | mnt->mnt_count = 1; | |
223 | mnt->mnt_writers = 0; | |
224 | #endif | |
225 | ||
226 | INIT_LIST_HEAD(&mnt->mnt_hash); | |
227 | INIT_LIST_HEAD(&mnt->mnt_child); | |
228 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
229 | INIT_LIST_HEAD(&mnt->mnt_list); | |
230 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
231 | INIT_LIST_HEAD(&mnt->mnt_share); | |
232 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
233 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
234 | #ifdef CONFIG_FSNOTIFY | |
235 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
236 | #endif | |
237 | } | |
238 | return mnt; | |
239 | ||
240 | #ifdef CONFIG_SMP | |
241 | out_free_devname: | |
242 | kfree(mnt->mnt_devname); | |
243 | #endif | |
244 | out_free_id: | |
245 | mnt_free_id(mnt); | |
246 | out_free_cache: | |
247 | kmem_cache_free(mnt_cache, mnt); | |
248 | return NULL; | |
249 | } | |
250 | ||
251 | /* | |
252 | * Most r/o checks on a fs are for operations that take | |
253 | * discrete amounts of time, like a write() or unlink(). | |
254 | * We must keep track of when those operations start | |
255 | * (for permission checks) and when they end, so that | |
256 | * we can determine when writes are able to occur to | |
257 | * a filesystem. | |
258 | */ | |
259 | /* | |
260 | * __mnt_is_readonly: check whether a mount is read-only | |
261 | * @mnt: the mount to check for its write status | |
262 | * | |
263 | * This shouldn't be used directly ouside of the VFS. | |
264 | * It does not guarantee that the filesystem will stay | |
265 | * r/w, just that it is right *now*. This can not and | |
266 | * should not be used in place of IS_RDONLY(inode). | |
267 | * mnt_want/drop_write() will _keep_ the filesystem | |
268 | * r/w. | |
269 | */ | |
270 | int __mnt_is_readonly(struct vfsmount *mnt) | |
271 | { | |
272 | if (mnt->mnt_flags & MNT_READONLY) | |
273 | return 1; | |
274 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
275 | return 1; | |
276 | return 0; | |
277 | } | |
278 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
279 | ||
280 | static inline void mnt_inc_writers(struct vfsmount *mnt) | |
281 | { | |
282 | #ifdef CONFIG_SMP | |
283 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); | |
284 | #else | |
285 | mnt->mnt_writers++; | |
286 | #endif | |
287 | } | |
288 | ||
289 | static inline void mnt_dec_writers(struct vfsmount *mnt) | |
290 | { | |
291 | #ifdef CONFIG_SMP | |
292 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); | |
293 | #else | |
294 | mnt->mnt_writers--; | |
295 | #endif | |
296 | } | |
297 | ||
298 | static unsigned int mnt_get_writers(struct vfsmount *mnt) | |
299 | { | |
300 | #ifdef CONFIG_SMP | |
301 | unsigned int count = 0; | |
302 | int cpu; | |
303 | ||
304 | for_each_possible_cpu(cpu) { | |
305 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; | |
306 | } | |
307 | ||
308 | return count; | |
309 | #else | |
310 | return mnt->mnt_writers; | |
311 | #endif | |
312 | } | |
313 | ||
314 | /* | |
315 | * Most r/o checks on a fs are for operations that take | |
316 | * discrete amounts of time, like a write() or unlink(). | |
317 | * We must keep track of when those operations start | |
318 | * (for permission checks) and when they end, so that | |
319 | * we can determine when writes are able to occur to | |
320 | * a filesystem. | |
321 | */ | |
322 | /** | |
323 | * mnt_want_write - get write access to a mount | |
324 | * @mnt: the mount on which to take a write | |
325 | * | |
326 | * This tells the low-level filesystem that a write is | |
327 | * about to be performed to it, and makes sure that | |
328 | * writes are allowed before returning success. When | |
329 | * the write operation is finished, mnt_drop_write() | |
330 | * must be called. This is effectively a refcount. | |
331 | */ | |
332 | int mnt_want_write(struct vfsmount *mnt) | |
333 | { | |
334 | int ret = 0; | |
335 | ||
336 | preempt_disable(); | |
337 | mnt_inc_writers(mnt); | |
338 | /* | |
339 | * The store to mnt_inc_writers must be visible before we pass | |
340 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our | |
341 | * incremented count after it has set MNT_WRITE_HOLD. | |
342 | */ | |
343 | smp_mb(); | |
344 | while (mnt->mnt_flags & MNT_WRITE_HOLD) | |
345 | cpu_relax(); | |
346 | /* | |
347 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
348 | * be set to match its requirements. So we must not load that until | |
349 | * MNT_WRITE_HOLD is cleared. | |
350 | */ | |
351 | smp_rmb(); | |
352 | if (__mnt_is_readonly(mnt)) { | |
353 | mnt_dec_writers(mnt); | |
354 | ret = -EROFS; | |
355 | goto out; | |
356 | } | |
357 | out: | |
358 | preempt_enable(); | |
359 | return ret; | |
360 | } | |
361 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
362 | ||
363 | /** | |
364 | * mnt_clone_write - get write access to a mount | |
365 | * @mnt: the mount on which to take a write | |
366 | * | |
367 | * This is effectively like mnt_want_write, except | |
368 | * it must only be used to take an extra write reference | |
369 | * on a mountpoint that we already know has a write reference | |
370 | * on it. This allows some optimisation. | |
371 | * | |
372 | * After finished, mnt_drop_write must be called as usual to | |
373 | * drop the reference. | |
374 | */ | |
375 | int mnt_clone_write(struct vfsmount *mnt) | |
376 | { | |
377 | /* superblock may be r/o */ | |
378 | if (__mnt_is_readonly(mnt)) | |
379 | return -EROFS; | |
380 | preempt_disable(); | |
381 | mnt_inc_writers(mnt); | |
382 | preempt_enable(); | |
383 | return 0; | |
384 | } | |
385 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
386 | ||
387 | /** | |
388 | * mnt_want_write_file - get write access to a file's mount | |
389 | * @file: the file who's mount on which to take a write | |
390 | * | |
391 | * This is like mnt_want_write, but it takes a file and can | |
392 | * do some optimisations if the file is open for write already | |
393 | */ | |
394 | int mnt_want_write_file(struct file *file) | |
395 | { | |
396 | struct inode *inode = file->f_dentry->d_inode; | |
397 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) | |
398 | return mnt_want_write(file->f_path.mnt); | |
399 | else | |
400 | return mnt_clone_write(file->f_path.mnt); | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
403 | ||
404 | /** | |
405 | * mnt_drop_write - give up write access to a mount | |
406 | * @mnt: the mount on which to give up write access | |
407 | * | |
408 | * Tells the low-level filesystem that we are done | |
409 | * performing writes to it. Must be matched with | |
410 | * mnt_want_write() call above. | |
411 | */ | |
412 | void mnt_drop_write(struct vfsmount *mnt) | |
413 | { | |
414 | preempt_disable(); | |
415 | mnt_dec_writers(mnt); | |
416 | preempt_enable(); | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
419 | ||
420 | static int mnt_make_readonly(struct vfsmount *mnt) | |
421 | { | |
422 | int ret = 0; | |
423 | ||
424 | br_write_lock(vfsmount_lock); | |
425 | mnt->mnt_flags |= MNT_WRITE_HOLD; | |
426 | /* | |
427 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store | |
428 | * should be visible before we do. | |
429 | */ | |
430 | smp_mb(); | |
431 | ||
432 | /* | |
433 | * With writers on hold, if this value is zero, then there are | |
434 | * definitely no active writers (although held writers may subsequently | |
435 | * increment the count, they'll have to wait, and decrement it after | |
436 | * seeing MNT_READONLY). | |
437 | * | |
438 | * It is OK to have counter incremented on one CPU and decremented on | |
439 | * another: the sum will add up correctly. The danger would be when we | |
440 | * sum up each counter, if we read a counter before it is incremented, | |
441 | * but then read another CPU's count which it has been subsequently | |
442 | * decremented from -- we would see more decrements than we should. | |
443 | * MNT_WRITE_HOLD protects against this scenario, because | |
444 | * mnt_want_write first increments count, then smp_mb, then spins on | |
445 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
446 | * we're counting up here. | |
447 | */ | |
448 | if (mnt_get_writers(mnt) > 0) | |
449 | ret = -EBUSY; | |
450 | else | |
451 | mnt->mnt_flags |= MNT_READONLY; | |
452 | /* | |
453 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
454 | * that become unheld will see MNT_READONLY. | |
455 | */ | |
456 | smp_wmb(); | |
457 | mnt->mnt_flags &= ~MNT_WRITE_HOLD; | |
458 | br_write_unlock(vfsmount_lock); | |
459 | return ret; | |
460 | } | |
461 | ||
462 | static void __mnt_unmake_readonly(struct vfsmount *mnt) | |
463 | { | |
464 | br_write_lock(vfsmount_lock); | |
465 | mnt->mnt_flags &= ~MNT_READONLY; | |
466 | br_write_unlock(vfsmount_lock); | |
467 | } | |
468 | ||
469 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) | |
470 | { | |
471 | mnt->mnt_sb = sb; | |
472 | mnt->mnt_root = dget(sb->s_root); | |
473 | } | |
474 | ||
475 | EXPORT_SYMBOL(simple_set_mnt); | |
476 | ||
477 | void free_vfsmnt(struct vfsmount *mnt) | |
478 | { | |
479 | kfree(mnt->mnt_devname); | |
480 | mnt_free_id(mnt); | |
481 | #ifdef CONFIG_SMP | |
482 | free_percpu(mnt->mnt_pcp); | |
483 | #endif | |
484 | kmem_cache_free(mnt_cache, mnt); | |
485 | } | |
486 | ||
487 | /* | |
488 | * find the first or last mount at @dentry on vfsmount @mnt depending on | |
489 | * @dir. If @dir is set return the first mount else return the last mount. | |
490 | * vfsmount_lock must be held for read or write. | |
491 | */ | |
492 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, | |
493 | int dir) | |
494 | { | |
495 | struct list_head *head = mount_hashtable + hash(mnt, dentry); | |
496 | struct list_head *tmp = head; | |
497 | struct vfsmount *p, *found = NULL; | |
498 | ||
499 | for (;;) { | |
500 | tmp = dir ? tmp->next : tmp->prev; | |
501 | p = NULL; | |
502 | if (tmp == head) | |
503 | break; | |
504 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
505 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
506 | found = p; | |
507 | break; | |
508 | } | |
509 | } | |
510 | return found; | |
511 | } | |
512 | ||
513 | /* | |
514 | * lookup_mnt increments the ref count before returning | |
515 | * the vfsmount struct. | |
516 | */ | |
517 | struct vfsmount *lookup_mnt(struct path *path) | |
518 | { | |
519 | struct vfsmount *child_mnt; | |
520 | ||
521 | br_read_lock(vfsmount_lock); | |
522 | if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) | |
523 | mntget(child_mnt); | |
524 | br_read_unlock(vfsmount_lock); | |
525 | return child_mnt; | |
526 | } | |
527 | ||
528 | static inline int check_mnt(struct vfsmount *mnt) | |
529 | { | |
530 | return mnt->mnt_ns == current->nsproxy->mnt_ns; | |
531 | } | |
532 | ||
533 | /* | |
534 | * vfsmount lock must be held for write | |
535 | */ | |
536 | static void touch_mnt_namespace(struct mnt_namespace *ns) | |
537 | { | |
538 | if (ns) { | |
539 | ns->event = ++event; | |
540 | wake_up_interruptible(&ns->poll); | |
541 | } | |
542 | } | |
543 | ||
544 | /* | |
545 | * vfsmount lock must be held for write | |
546 | */ | |
547 | static void __touch_mnt_namespace(struct mnt_namespace *ns) | |
548 | { | |
549 | if (ns && ns->event != event) { | |
550 | ns->event = event; | |
551 | wake_up_interruptible(&ns->poll); | |
552 | } | |
553 | } | |
554 | ||
555 | /* | |
556 | * Clear dentry's mounted state if it has no remaining mounts. | |
557 | * vfsmount_lock must be held for write. | |
558 | */ | |
559 | static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry) | |
560 | { | |
561 | unsigned u; | |
562 | ||
563 | for (u = 0; u < HASH_SIZE; u++) { | |
564 | struct vfsmount *p; | |
565 | ||
566 | list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { | |
567 | if (p->mnt_mountpoint == dentry) | |
568 | return; | |
569 | } | |
570 | } | |
571 | spin_lock(&dentry->d_lock); | |
572 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
573 | spin_unlock(&dentry->d_lock); | |
574 | } | |
575 | ||
576 | /* | |
577 | * vfsmount lock must be held for write | |
578 | */ | |
579 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) | |
580 | { | |
581 | old_path->dentry = mnt->mnt_mountpoint; | |
582 | old_path->mnt = mnt->mnt_parent; | |
583 | mnt->mnt_parent = mnt; | |
584 | mnt->mnt_mountpoint = mnt->mnt_root; | |
585 | list_del_init(&mnt->mnt_child); | |
586 | list_del_init(&mnt->mnt_hash); | |
587 | dentry_reset_mounted(old_path->mnt, old_path->dentry); | |
588 | } | |
589 | ||
590 | /* | |
591 | * vfsmount lock must be held for write | |
592 | */ | |
593 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, | |
594 | struct vfsmount *child_mnt) | |
595 | { | |
596 | child_mnt->mnt_parent = mntget(mnt); | |
597 | child_mnt->mnt_mountpoint = dget(dentry); | |
598 | spin_lock(&dentry->d_lock); | |
599 | dentry->d_flags |= DCACHE_MOUNTED; | |
600 | spin_unlock(&dentry->d_lock); | |
601 | } | |
602 | ||
603 | /* | |
604 | * vfsmount lock must be held for write | |
605 | */ | |
606 | static void attach_mnt(struct vfsmount *mnt, struct path *path) | |
607 | { | |
608 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); | |
609 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
610 | hash(path->mnt, path->dentry)); | |
611 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
612 | } | |
613 | ||
614 | static inline void __mnt_make_longterm(struct vfsmount *mnt) | |
615 | { | |
616 | #ifdef CONFIG_SMP | |
617 | atomic_inc(&mnt->mnt_longterm); | |
618 | #endif | |
619 | } | |
620 | ||
621 | /* needs vfsmount lock for write */ | |
622 | static inline void __mnt_make_shortterm(struct vfsmount *mnt) | |
623 | { | |
624 | #ifdef CONFIG_SMP | |
625 | atomic_dec(&mnt->mnt_longterm); | |
626 | #endif | |
627 | } | |
628 | ||
629 | /* | |
630 | * vfsmount lock must be held for write | |
631 | */ | |
632 | static void commit_tree(struct vfsmount *mnt) | |
633 | { | |
634 | struct vfsmount *parent = mnt->mnt_parent; | |
635 | struct vfsmount *m; | |
636 | LIST_HEAD(head); | |
637 | struct mnt_namespace *n = parent->mnt_ns; | |
638 | ||
639 | BUG_ON(parent == mnt); | |
640 | ||
641 | list_add_tail(&head, &mnt->mnt_list); | |
642 | list_for_each_entry(m, &head, mnt_list) { | |
643 | m->mnt_ns = n; | |
644 | __mnt_make_longterm(m); | |
645 | } | |
646 | ||
647 | list_splice(&head, n->list.prev); | |
648 | ||
649 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
650 | hash(parent, mnt->mnt_mountpoint)); | |
651 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
652 | touch_mnt_namespace(n); | |
653 | } | |
654 | ||
655 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
656 | { | |
657 | struct list_head *next = p->mnt_mounts.next; | |
658 | if (next == &p->mnt_mounts) { | |
659 | while (1) { | |
660 | if (p == root) | |
661 | return NULL; | |
662 | next = p->mnt_child.next; | |
663 | if (next != &p->mnt_parent->mnt_mounts) | |
664 | break; | |
665 | p = p->mnt_parent; | |
666 | } | |
667 | } | |
668 | return list_entry(next, struct vfsmount, mnt_child); | |
669 | } | |
670 | ||
671 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) | |
672 | { | |
673 | struct list_head *prev = p->mnt_mounts.prev; | |
674 | while (prev != &p->mnt_mounts) { | |
675 | p = list_entry(prev, struct vfsmount, mnt_child); | |
676 | prev = p->mnt_mounts.prev; | |
677 | } | |
678 | return p; | |
679 | } | |
680 | ||
681 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, | |
682 | int flag) | |
683 | { | |
684 | struct super_block *sb = old->mnt_sb; | |
685 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
686 | ||
687 | if (mnt) { | |
688 | if (flag & (CL_SLAVE | CL_PRIVATE)) | |
689 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
690 | else | |
691 | mnt->mnt_group_id = old->mnt_group_id; | |
692 | ||
693 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
694 | int err = mnt_alloc_group_id(mnt); | |
695 | if (err) | |
696 | goto out_free; | |
697 | } | |
698 | ||
699 | mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD; | |
700 | atomic_inc(&sb->s_active); | |
701 | mnt->mnt_sb = sb; | |
702 | mnt->mnt_root = dget(root); | |
703 | mnt->mnt_mountpoint = mnt->mnt_root; | |
704 | mnt->mnt_parent = mnt; | |
705 | ||
706 | if (flag & CL_SLAVE) { | |
707 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
708 | mnt->mnt_master = old; | |
709 | CLEAR_MNT_SHARED(mnt); | |
710 | } else if (!(flag & CL_PRIVATE)) { | |
711 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
712 | list_add(&mnt->mnt_share, &old->mnt_share); | |
713 | if (IS_MNT_SLAVE(old)) | |
714 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
715 | mnt->mnt_master = old->mnt_master; | |
716 | } | |
717 | if (flag & CL_MAKE_SHARED) | |
718 | set_mnt_shared(mnt); | |
719 | ||
720 | /* stick the duplicate mount on the same expiry list | |
721 | * as the original if that was on one */ | |
722 | if (flag & CL_EXPIRE) { | |
723 | if (!list_empty(&old->mnt_expire)) | |
724 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
725 | } | |
726 | } | |
727 | return mnt; | |
728 | ||
729 | out_free: | |
730 | free_vfsmnt(mnt); | |
731 | return NULL; | |
732 | } | |
733 | ||
734 | static inline void mntfree(struct vfsmount *mnt) | |
735 | { | |
736 | struct super_block *sb = mnt->mnt_sb; | |
737 | ||
738 | /* | |
739 | * This probably indicates that somebody messed | |
740 | * up a mnt_want/drop_write() pair. If this | |
741 | * happens, the filesystem was probably unable | |
742 | * to make r/w->r/o transitions. | |
743 | */ | |
744 | /* | |
745 | * The locking used to deal with mnt_count decrement provides barriers, | |
746 | * so mnt_get_writers() below is safe. | |
747 | */ | |
748 | WARN_ON(mnt_get_writers(mnt)); | |
749 | fsnotify_vfsmount_delete(mnt); | |
750 | dput(mnt->mnt_root); | |
751 | free_vfsmnt(mnt); | |
752 | deactivate_super(sb); | |
753 | } | |
754 | ||
755 | static void mntput_no_expire(struct vfsmount *mnt) | |
756 | { | |
757 | put_again: | |
758 | #ifdef CONFIG_SMP | |
759 | br_read_lock(vfsmount_lock); | |
760 | if (likely(atomic_read(&mnt->mnt_longterm))) { | |
761 | mnt_dec_count(mnt); | |
762 | br_read_unlock(vfsmount_lock); | |
763 | return; | |
764 | } | |
765 | br_read_unlock(vfsmount_lock); | |
766 | ||
767 | br_write_lock(vfsmount_lock); | |
768 | mnt_dec_count(mnt); | |
769 | if (mnt_get_count(mnt)) { | |
770 | br_write_unlock(vfsmount_lock); | |
771 | return; | |
772 | } | |
773 | #else | |
774 | mnt_dec_count(mnt); | |
775 | if (likely(mnt_get_count(mnt))) | |
776 | return; | |
777 | br_write_lock(vfsmount_lock); | |
778 | #endif | |
779 | if (unlikely(mnt->mnt_pinned)) { | |
780 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
781 | mnt->mnt_pinned = 0; | |
782 | br_write_unlock(vfsmount_lock); | |
783 | acct_auto_close_mnt(mnt); | |
784 | goto put_again; | |
785 | } | |
786 | br_write_unlock(vfsmount_lock); | |
787 | mntfree(mnt); | |
788 | } | |
789 | ||
790 | void mntput(struct vfsmount *mnt) | |
791 | { | |
792 | if (mnt) { | |
793 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ | |
794 | if (unlikely(mnt->mnt_expiry_mark)) | |
795 | mnt->mnt_expiry_mark = 0; | |
796 | mntput_no_expire(mnt); | |
797 | } | |
798 | } | |
799 | EXPORT_SYMBOL(mntput); | |
800 | ||
801 | struct vfsmount *mntget(struct vfsmount *mnt) | |
802 | { | |
803 | if (mnt) | |
804 | mnt_inc_count(mnt); | |
805 | return mnt; | |
806 | } | |
807 | EXPORT_SYMBOL(mntget); | |
808 | ||
809 | void mnt_pin(struct vfsmount *mnt) | |
810 | { | |
811 | br_write_lock(vfsmount_lock); | |
812 | mnt->mnt_pinned++; | |
813 | br_write_unlock(vfsmount_lock); | |
814 | } | |
815 | EXPORT_SYMBOL(mnt_pin); | |
816 | ||
817 | void mnt_unpin(struct vfsmount *mnt) | |
818 | { | |
819 | br_write_lock(vfsmount_lock); | |
820 | if (mnt->mnt_pinned) { | |
821 | mnt_inc_count(mnt); | |
822 | mnt->mnt_pinned--; | |
823 | } | |
824 | br_write_unlock(vfsmount_lock); | |
825 | } | |
826 | EXPORT_SYMBOL(mnt_unpin); | |
827 | ||
828 | static inline void mangle(struct seq_file *m, const char *s) | |
829 | { | |
830 | seq_escape(m, s, " \t\n\\"); | |
831 | } | |
832 | ||
833 | /* | |
834 | * Simple .show_options callback for filesystems which don't want to | |
835 | * implement more complex mount option showing. | |
836 | * | |
837 | * See also save_mount_options(). | |
838 | */ | |
839 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
840 | { | |
841 | const char *options; | |
842 | ||
843 | rcu_read_lock(); | |
844 | options = rcu_dereference(mnt->mnt_sb->s_options); | |
845 | ||
846 | if (options != NULL && options[0]) { | |
847 | seq_putc(m, ','); | |
848 | mangle(m, options); | |
849 | } | |
850 | rcu_read_unlock(); | |
851 | ||
852 | return 0; | |
853 | } | |
854 | EXPORT_SYMBOL(generic_show_options); | |
855 | ||
856 | /* | |
857 | * If filesystem uses generic_show_options(), this function should be | |
858 | * called from the fill_super() callback. | |
859 | * | |
860 | * The .remount_fs callback usually needs to be handled in a special | |
861 | * way, to make sure, that previous options are not overwritten if the | |
862 | * remount fails. | |
863 | * | |
864 | * Also note, that if the filesystem's .remount_fs function doesn't | |
865 | * reset all options to their default value, but changes only newly | |
866 | * given options, then the displayed options will not reflect reality | |
867 | * any more. | |
868 | */ | |
869 | void save_mount_options(struct super_block *sb, char *options) | |
870 | { | |
871 | BUG_ON(sb->s_options); | |
872 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
873 | } | |
874 | EXPORT_SYMBOL(save_mount_options); | |
875 | ||
876 | void replace_mount_options(struct super_block *sb, char *options) | |
877 | { | |
878 | char *old = sb->s_options; | |
879 | rcu_assign_pointer(sb->s_options, options); | |
880 | if (old) { | |
881 | synchronize_rcu(); | |
882 | kfree(old); | |
883 | } | |
884 | } | |
885 | EXPORT_SYMBOL(replace_mount_options); | |
886 | ||
887 | #ifdef CONFIG_PROC_FS | |
888 | /* iterator */ | |
889 | static void *m_start(struct seq_file *m, loff_t *pos) | |
890 | { | |
891 | struct proc_mounts *p = m->private; | |
892 | ||
893 | down_read(&namespace_sem); | |
894 | return seq_list_start(&p->ns->list, *pos); | |
895 | } | |
896 | ||
897 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
898 | { | |
899 | struct proc_mounts *p = m->private; | |
900 | ||
901 | return seq_list_next(v, &p->ns->list, pos); | |
902 | } | |
903 | ||
904 | static void m_stop(struct seq_file *m, void *v) | |
905 | { | |
906 | up_read(&namespace_sem); | |
907 | } | |
908 | ||
909 | int mnt_had_events(struct proc_mounts *p) | |
910 | { | |
911 | struct mnt_namespace *ns = p->ns; | |
912 | int res = 0; | |
913 | ||
914 | br_read_lock(vfsmount_lock); | |
915 | if (p->event != ns->event) { | |
916 | p->event = ns->event; | |
917 | res = 1; | |
918 | } | |
919 | br_read_unlock(vfsmount_lock); | |
920 | ||
921 | return res; | |
922 | } | |
923 | ||
924 | struct proc_fs_info { | |
925 | int flag; | |
926 | const char *str; | |
927 | }; | |
928 | ||
929 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) | |
930 | { | |
931 | static const struct proc_fs_info fs_info[] = { | |
932 | { MS_SYNCHRONOUS, ",sync" }, | |
933 | { MS_DIRSYNC, ",dirsync" }, | |
934 | { MS_MANDLOCK, ",mand" }, | |
935 | { 0, NULL } | |
936 | }; | |
937 | const struct proc_fs_info *fs_infop; | |
938 | ||
939 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
940 | if (sb->s_flags & fs_infop->flag) | |
941 | seq_puts(m, fs_infop->str); | |
942 | } | |
943 | ||
944 | return security_sb_show_options(m, sb); | |
945 | } | |
946 | ||
947 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
948 | { | |
949 | static const struct proc_fs_info mnt_info[] = { | |
950 | { MNT_NOSUID, ",nosuid" }, | |
951 | { MNT_NODEV, ",nodev" }, | |
952 | { MNT_NOEXEC, ",noexec" }, | |
953 | { MNT_NOATIME, ",noatime" }, | |
954 | { MNT_NODIRATIME, ",nodiratime" }, | |
955 | { MNT_RELATIME, ",relatime" }, | |
956 | { 0, NULL } | |
957 | }; | |
958 | const struct proc_fs_info *fs_infop; | |
959 | ||
960 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
961 | if (mnt->mnt_flags & fs_infop->flag) | |
962 | seq_puts(m, fs_infop->str); | |
963 | } | |
964 | } | |
965 | ||
966 | static void show_type(struct seq_file *m, struct super_block *sb) | |
967 | { | |
968 | mangle(m, sb->s_type->name); | |
969 | if (sb->s_subtype && sb->s_subtype[0]) { | |
970 | seq_putc(m, '.'); | |
971 | mangle(m, sb->s_subtype); | |
972 | } | |
973 | } | |
974 | ||
975 | static int show_vfsmnt(struct seq_file *m, void *v) | |
976 | { | |
977 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
978 | int err = 0; | |
979 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
980 | ||
981 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
982 | seq_putc(m, ' '); | |
983 | seq_path(m, &mnt_path, " \t\n\\"); | |
984 | seq_putc(m, ' '); | |
985 | show_type(m, mnt->mnt_sb); | |
986 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); | |
987 | err = show_sb_opts(m, mnt->mnt_sb); | |
988 | if (err) | |
989 | goto out; | |
990 | show_mnt_opts(m, mnt); | |
991 | if (mnt->mnt_sb->s_op->show_options) | |
992 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
993 | seq_puts(m, " 0 0\n"); | |
994 | out: | |
995 | return err; | |
996 | } | |
997 | ||
998 | const struct seq_operations mounts_op = { | |
999 | .start = m_start, | |
1000 | .next = m_next, | |
1001 | .stop = m_stop, | |
1002 | .show = show_vfsmnt | |
1003 | }; | |
1004 | ||
1005 | static int show_mountinfo(struct seq_file *m, void *v) | |
1006 | { | |
1007 | struct proc_mounts *p = m->private; | |
1008 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
1009 | struct super_block *sb = mnt->mnt_sb; | |
1010 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
1011 | struct path root = p->root; | |
1012 | int err = 0; | |
1013 | ||
1014 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
1015 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
1016 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
1017 | seq_putc(m, ' '); | |
1018 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
1019 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
1020 | /* | |
1021 | * Mountpoint is outside root, discard that one. Ugly, | |
1022 | * but less so than trying to do that in iterator in a | |
1023 | * race-free way (due to renames). | |
1024 | */ | |
1025 | return SEQ_SKIP; | |
1026 | } | |
1027 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
1028 | show_mnt_opts(m, mnt); | |
1029 | ||
1030 | /* Tagged fields ("foo:X" or "bar") */ | |
1031 | if (IS_MNT_SHARED(mnt)) | |
1032 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
1033 | if (IS_MNT_SLAVE(mnt)) { | |
1034 | int master = mnt->mnt_master->mnt_group_id; | |
1035 | int dom = get_dominating_id(mnt, &p->root); | |
1036 | seq_printf(m, " master:%i", master); | |
1037 | if (dom && dom != master) | |
1038 | seq_printf(m, " propagate_from:%i", dom); | |
1039 | } | |
1040 | if (IS_MNT_UNBINDABLE(mnt)) | |
1041 | seq_puts(m, " unbindable"); | |
1042 | ||
1043 | /* Filesystem specific data */ | |
1044 | seq_puts(m, " - "); | |
1045 | show_type(m, sb); | |
1046 | seq_putc(m, ' '); | |
1047 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
1048 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
1049 | err = show_sb_opts(m, sb); | |
1050 | if (err) | |
1051 | goto out; | |
1052 | if (sb->s_op->show_options) | |
1053 | err = sb->s_op->show_options(m, mnt); | |
1054 | seq_putc(m, '\n'); | |
1055 | out: | |
1056 | return err; | |
1057 | } | |
1058 | ||
1059 | const struct seq_operations mountinfo_op = { | |
1060 | .start = m_start, | |
1061 | .next = m_next, | |
1062 | .stop = m_stop, | |
1063 | .show = show_mountinfo, | |
1064 | }; | |
1065 | ||
1066 | static int show_vfsstat(struct seq_file *m, void *v) | |
1067 | { | |
1068 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
1069 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
1070 | int err = 0; | |
1071 | ||
1072 | /* device */ | |
1073 | if (mnt->mnt_devname) { | |
1074 | seq_puts(m, "device "); | |
1075 | mangle(m, mnt->mnt_devname); | |
1076 | } else | |
1077 | seq_puts(m, "no device"); | |
1078 | ||
1079 | /* mount point */ | |
1080 | seq_puts(m, " mounted on "); | |
1081 | seq_path(m, &mnt_path, " \t\n\\"); | |
1082 | seq_putc(m, ' '); | |
1083 | ||
1084 | /* file system type */ | |
1085 | seq_puts(m, "with fstype "); | |
1086 | show_type(m, mnt->mnt_sb); | |
1087 | ||
1088 | /* optional statistics */ | |
1089 | if (mnt->mnt_sb->s_op->show_stats) { | |
1090 | seq_putc(m, ' '); | |
1091 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
1092 | } | |
1093 | ||
1094 | seq_putc(m, '\n'); | |
1095 | return err; | |
1096 | } | |
1097 | ||
1098 | const struct seq_operations mountstats_op = { | |
1099 | .start = m_start, | |
1100 | .next = m_next, | |
1101 | .stop = m_stop, | |
1102 | .show = show_vfsstat, | |
1103 | }; | |
1104 | #endif /* CONFIG_PROC_FS */ | |
1105 | ||
1106 | /** | |
1107 | * may_umount_tree - check if a mount tree is busy | |
1108 | * @mnt: root of mount tree | |
1109 | * | |
1110 | * This is called to check if a tree of mounts has any | |
1111 | * open files, pwds, chroots or sub mounts that are | |
1112 | * busy. | |
1113 | */ | |
1114 | int may_umount_tree(struct vfsmount *mnt) | |
1115 | { | |
1116 | int actual_refs = 0; | |
1117 | int minimum_refs = 0; | |
1118 | struct vfsmount *p; | |
1119 | ||
1120 | /* write lock needed for mnt_get_count */ | |
1121 | br_write_lock(vfsmount_lock); | |
1122 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1123 | actual_refs += mnt_get_count(p); | |
1124 | minimum_refs += 2; | |
1125 | } | |
1126 | br_write_unlock(vfsmount_lock); | |
1127 | ||
1128 | if (actual_refs > minimum_refs) | |
1129 | return 0; | |
1130 | ||
1131 | return 1; | |
1132 | } | |
1133 | ||
1134 | EXPORT_SYMBOL(may_umount_tree); | |
1135 | ||
1136 | /** | |
1137 | * may_umount - check if a mount point is busy | |
1138 | * @mnt: root of mount | |
1139 | * | |
1140 | * This is called to check if a mount point has any | |
1141 | * open files, pwds, chroots or sub mounts. If the | |
1142 | * mount has sub mounts this will return busy | |
1143 | * regardless of whether the sub mounts are busy. | |
1144 | * | |
1145 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1146 | * give false negatives. The main reason why it's here is that we need | |
1147 | * a non-destructive way to look for easily umountable filesystems. | |
1148 | */ | |
1149 | int may_umount(struct vfsmount *mnt) | |
1150 | { | |
1151 | int ret = 1; | |
1152 | down_read(&namespace_sem); | |
1153 | br_write_lock(vfsmount_lock); | |
1154 | if (propagate_mount_busy(mnt, 2)) | |
1155 | ret = 0; | |
1156 | br_write_unlock(vfsmount_lock); | |
1157 | up_read(&namespace_sem); | |
1158 | return ret; | |
1159 | } | |
1160 | ||
1161 | EXPORT_SYMBOL(may_umount); | |
1162 | ||
1163 | void release_mounts(struct list_head *head) | |
1164 | { | |
1165 | struct vfsmount *mnt; | |
1166 | while (!list_empty(head)) { | |
1167 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); | |
1168 | list_del_init(&mnt->mnt_hash); | |
1169 | if (mnt->mnt_parent != mnt) { | |
1170 | struct dentry *dentry; | |
1171 | struct vfsmount *m; | |
1172 | ||
1173 | br_write_lock(vfsmount_lock); | |
1174 | dentry = mnt->mnt_mountpoint; | |
1175 | m = mnt->mnt_parent; | |
1176 | mnt->mnt_mountpoint = mnt->mnt_root; | |
1177 | mnt->mnt_parent = mnt; | |
1178 | m->mnt_ghosts--; | |
1179 | br_write_unlock(vfsmount_lock); | |
1180 | dput(dentry); | |
1181 | mntput(m); | |
1182 | } | |
1183 | mntput(mnt); | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * vfsmount lock must be held for write | |
1189 | * namespace_sem must be held for write | |
1190 | */ | |
1191 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) | |
1192 | { | |
1193 | LIST_HEAD(tmp_list); | |
1194 | struct vfsmount *p; | |
1195 | ||
1196 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1197 | list_move(&p->mnt_hash, &tmp_list); | |
1198 | ||
1199 | if (propagate) | |
1200 | propagate_umount(&tmp_list); | |
1201 | ||
1202 | list_for_each_entry(p, &tmp_list, mnt_hash) { | |
1203 | list_del_init(&p->mnt_expire); | |
1204 | list_del_init(&p->mnt_list); | |
1205 | __touch_mnt_namespace(p->mnt_ns); | |
1206 | p->mnt_ns = NULL; | |
1207 | __mnt_make_shortterm(p); | |
1208 | list_del_init(&p->mnt_child); | |
1209 | if (p->mnt_parent != p) { | |
1210 | p->mnt_parent->mnt_ghosts++; | |
1211 | dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint); | |
1212 | } | |
1213 | change_mnt_propagation(p, MS_PRIVATE); | |
1214 | } | |
1215 | list_splice(&tmp_list, kill); | |
1216 | } | |
1217 | ||
1218 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); | |
1219 | ||
1220 | static int do_umount(struct vfsmount *mnt, int flags) | |
1221 | { | |
1222 | struct super_block *sb = mnt->mnt_sb; | |
1223 | int retval; | |
1224 | LIST_HEAD(umount_list); | |
1225 | ||
1226 | retval = security_sb_umount(mnt, flags); | |
1227 | if (retval) | |
1228 | return retval; | |
1229 | ||
1230 | /* | |
1231 | * Allow userspace to request a mountpoint be expired rather than | |
1232 | * unmounting unconditionally. Unmount only happens if: | |
1233 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1234 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1235 | */ | |
1236 | if (flags & MNT_EXPIRE) { | |
1237 | if (mnt == current->fs->root.mnt || | |
1238 | flags & (MNT_FORCE | MNT_DETACH)) | |
1239 | return -EINVAL; | |
1240 | ||
1241 | /* | |
1242 | * probably don't strictly need the lock here if we examined | |
1243 | * all race cases, but it's a slowpath. | |
1244 | */ | |
1245 | br_write_lock(vfsmount_lock); | |
1246 | if (mnt_get_count(mnt) != 2) { | |
1247 | br_write_lock(vfsmount_lock); | |
1248 | return -EBUSY; | |
1249 | } | |
1250 | br_write_unlock(vfsmount_lock); | |
1251 | ||
1252 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1253 | return -EAGAIN; | |
1254 | } | |
1255 | ||
1256 | /* | |
1257 | * If we may have to abort operations to get out of this | |
1258 | * mount, and they will themselves hold resources we must | |
1259 | * allow the fs to do things. In the Unix tradition of | |
1260 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1261 | * might fail to complete on the first run through as other tasks | |
1262 | * must return, and the like. Thats for the mount program to worry | |
1263 | * about for the moment. | |
1264 | */ | |
1265 | ||
1266 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { | |
1267 | sb->s_op->umount_begin(sb); | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * No sense to grab the lock for this test, but test itself looks | |
1272 | * somewhat bogus. Suggestions for better replacement? | |
1273 | * Ho-hum... In principle, we might treat that as umount + switch | |
1274 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1275 | * Actually it makes sense, especially if rootfs would contain a | |
1276 | * /reboot - static binary that would close all descriptors and | |
1277 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1278 | */ | |
1279 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { | |
1280 | /* | |
1281 | * Special case for "unmounting" root ... | |
1282 | * we just try to remount it readonly. | |
1283 | */ | |
1284 | down_write(&sb->s_umount); | |
1285 | if (!(sb->s_flags & MS_RDONLY)) | |
1286 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
1287 | up_write(&sb->s_umount); | |
1288 | return retval; | |
1289 | } | |
1290 | ||
1291 | down_write(&namespace_sem); | |
1292 | br_write_lock(vfsmount_lock); | |
1293 | event++; | |
1294 | ||
1295 | if (!(flags & MNT_DETACH)) | |
1296 | shrink_submounts(mnt, &umount_list); | |
1297 | ||
1298 | retval = -EBUSY; | |
1299 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { | |
1300 | if (!list_empty(&mnt->mnt_list)) | |
1301 | umount_tree(mnt, 1, &umount_list); | |
1302 | retval = 0; | |
1303 | } | |
1304 | br_write_unlock(vfsmount_lock); | |
1305 | up_write(&namespace_sem); | |
1306 | release_mounts(&umount_list); | |
1307 | return retval; | |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * Now umount can handle mount points as well as block devices. | |
1312 | * This is important for filesystems which use unnamed block devices. | |
1313 | * | |
1314 | * We now support a flag for forced unmount like the other 'big iron' | |
1315 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1316 | */ | |
1317 | ||
1318 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) | |
1319 | { | |
1320 | struct path path; | |
1321 | int retval; | |
1322 | int lookup_flags = 0; | |
1323 | ||
1324 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1325 | return -EINVAL; | |
1326 | ||
1327 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1328 | lookup_flags |= LOOKUP_FOLLOW; | |
1329 | ||
1330 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1331 | if (retval) | |
1332 | goto out; | |
1333 | retval = -EINVAL; | |
1334 | if (path.dentry != path.mnt->mnt_root) | |
1335 | goto dput_and_out; | |
1336 | if (!check_mnt(path.mnt)) | |
1337 | goto dput_and_out; | |
1338 | ||
1339 | retval = -EPERM; | |
1340 | if (!capable(CAP_SYS_ADMIN)) | |
1341 | goto dput_and_out; | |
1342 | ||
1343 | retval = do_umount(path.mnt, flags); | |
1344 | dput_and_out: | |
1345 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ | |
1346 | dput(path.dentry); | |
1347 | mntput_no_expire(path.mnt); | |
1348 | out: | |
1349 | return retval; | |
1350 | } | |
1351 | ||
1352 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1353 | ||
1354 | /* | |
1355 | * The 2.0 compatible umount. No flags. | |
1356 | */ | |
1357 | SYSCALL_DEFINE1(oldumount, char __user *, name) | |
1358 | { | |
1359 | return sys_umount(name, 0); | |
1360 | } | |
1361 | ||
1362 | #endif | |
1363 | ||
1364 | static int mount_is_safe(struct path *path) | |
1365 | { | |
1366 | if (capable(CAP_SYS_ADMIN)) | |
1367 | return 0; | |
1368 | return -EPERM; | |
1369 | #ifdef notyet | |
1370 | if (S_ISLNK(path->dentry->d_inode->i_mode)) | |
1371 | return -EPERM; | |
1372 | if (path->dentry->d_inode->i_mode & S_ISVTX) { | |
1373 | if (current_uid() != path->dentry->d_inode->i_uid) | |
1374 | return -EPERM; | |
1375 | } | |
1376 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) | |
1377 | return -EPERM; | |
1378 | return 0; | |
1379 | #endif | |
1380 | } | |
1381 | ||
1382 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, | |
1383 | int flag) | |
1384 | { | |
1385 | struct vfsmount *res, *p, *q, *r, *s; | |
1386 | struct path path; | |
1387 | ||
1388 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) | |
1389 | return NULL; | |
1390 | ||
1391 | res = q = clone_mnt(mnt, dentry, flag); | |
1392 | if (!q) | |
1393 | goto Enomem; | |
1394 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1395 | ||
1396 | p = mnt; | |
1397 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { | |
1398 | if (!is_subdir(r->mnt_mountpoint, dentry)) | |
1399 | continue; | |
1400 | ||
1401 | for (s = r; s; s = next_mnt(s, r)) { | |
1402 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { | |
1403 | s = skip_mnt_tree(s); | |
1404 | continue; | |
1405 | } | |
1406 | while (p != s->mnt_parent) { | |
1407 | p = p->mnt_parent; | |
1408 | q = q->mnt_parent; | |
1409 | } | |
1410 | p = s; | |
1411 | path.mnt = q; | |
1412 | path.dentry = p->mnt_mountpoint; | |
1413 | q = clone_mnt(p, p->mnt_root, flag); | |
1414 | if (!q) | |
1415 | goto Enomem; | |
1416 | br_write_lock(vfsmount_lock); | |
1417 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1418 | attach_mnt(q, &path); | |
1419 | br_write_unlock(vfsmount_lock); | |
1420 | } | |
1421 | } | |
1422 | return res; | |
1423 | Enomem: | |
1424 | if (res) { | |
1425 | LIST_HEAD(umount_list); | |
1426 | br_write_lock(vfsmount_lock); | |
1427 | umount_tree(res, 0, &umount_list); | |
1428 | br_write_unlock(vfsmount_lock); | |
1429 | release_mounts(&umount_list); | |
1430 | } | |
1431 | return NULL; | |
1432 | } | |
1433 | ||
1434 | struct vfsmount *collect_mounts(struct path *path) | |
1435 | { | |
1436 | struct vfsmount *tree; | |
1437 | down_write(&namespace_sem); | |
1438 | tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); | |
1439 | up_write(&namespace_sem); | |
1440 | return tree; | |
1441 | } | |
1442 | ||
1443 | void drop_collected_mounts(struct vfsmount *mnt) | |
1444 | { | |
1445 | LIST_HEAD(umount_list); | |
1446 | down_write(&namespace_sem); | |
1447 | br_write_lock(vfsmount_lock); | |
1448 | umount_tree(mnt, 0, &umount_list); | |
1449 | br_write_unlock(vfsmount_lock); | |
1450 | up_write(&namespace_sem); | |
1451 | release_mounts(&umount_list); | |
1452 | } | |
1453 | ||
1454 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, | |
1455 | struct vfsmount *root) | |
1456 | { | |
1457 | struct vfsmount *mnt; | |
1458 | int res = f(root, arg); | |
1459 | if (res) | |
1460 | return res; | |
1461 | list_for_each_entry(mnt, &root->mnt_list, mnt_list) { | |
1462 | res = f(mnt, arg); | |
1463 | if (res) | |
1464 | return res; | |
1465 | } | |
1466 | return 0; | |
1467 | } | |
1468 | ||
1469 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) | |
1470 | { | |
1471 | struct vfsmount *p; | |
1472 | ||
1473 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1474 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1475 | mnt_release_group_id(p); | |
1476 | } | |
1477 | } | |
1478 | ||
1479 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1480 | { | |
1481 | struct vfsmount *p; | |
1482 | ||
1483 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1484 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1485 | int err = mnt_alloc_group_id(p); | |
1486 | if (err) { | |
1487 | cleanup_group_ids(mnt, p); | |
1488 | return err; | |
1489 | } | |
1490 | } | |
1491 | } | |
1492 | ||
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * @source_mnt : mount tree to be attached | |
1498 | * @nd : place the mount tree @source_mnt is attached | |
1499 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1500 | * store the parent mount and mountpoint dentry. | |
1501 | * (done when source_mnt is moved) | |
1502 | * | |
1503 | * NOTE: in the table below explains the semantics when a source mount | |
1504 | * of a given type is attached to a destination mount of a given type. | |
1505 | * --------------------------------------------------------------------------- | |
1506 | * | BIND MOUNT OPERATION | | |
1507 | * |************************************************************************** | |
1508 | * | source-->| shared | private | slave | unbindable | | |
1509 | * | dest | | | | | | |
1510 | * | | | | | | | | |
1511 | * | v | | | | | | |
1512 | * |************************************************************************** | |
1513 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1514 | * | | | | | | | |
1515 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1516 | * *************************************************************************** | |
1517 | * A bind operation clones the source mount and mounts the clone on the | |
1518 | * destination mount. | |
1519 | * | |
1520 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1521 | * tree of the destination mount and the cloned mount is added to | |
1522 | * the peer group of the source mount. | |
1523 | * (+) the cloned mount is created under the destination mount and is marked | |
1524 | * as shared. The cloned mount is added to the peer group of the source | |
1525 | * mount. | |
1526 | * (+++) the mount is propagated to all the mounts in the propagation tree | |
1527 | * of the destination mount and the cloned mount is made slave | |
1528 | * of the same master as that of the source mount. The cloned mount | |
1529 | * is marked as 'shared and slave'. | |
1530 | * (*) the cloned mount is made a slave of the same master as that of the | |
1531 | * source mount. | |
1532 | * | |
1533 | * --------------------------------------------------------------------------- | |
1534 | * | MOVE MOUNT OPERATION | | |
1535 | * |************************************************************************** | |
1536 | * | source-->| shared | private | slave | unbindable | | |
1537 | * | dest | | | | | | |
1538 | * | | | | | | | | |
1539 | * | v | | | | | | |
1540 | * |************************************************************************** | |
1541 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1542 | * | | | | | | | |
1543 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1544 | * *************************************************************************** | |
1545 | * | |
1546 | * (+) the mount is moved to the destination. And is then propagated to | |
1547 | * all the mounts in the propagation tree of the destination mount. | |
1548 | * (+*) the mount is moved to the destination. | |
1549 | * (+++) the mount is moved to the destination and is then propagated to | |
1550 | * all the mounts belonging to the destination mount's propagation tree. | |
1551 | * the mount is marked as 'shared and slave'. | |
1552 | * (*) the mount continues to be a slave at the new location. | |
1553 | * | |
1554 | * if the source mount is a tree, the operations explained above is | |
1555 | * applied to each mount in the tree. | |
1556 | * Must be called without spinlocks held, since this function can sleep | |
1557 | * in allocations. | |
1558 | */ | |
1559 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1560 | struct path *path, struct path *parent_path) | |
1561 | { | |
1562 | LIST_HEAD(tree_list); | |
1563 | struct vfsmount *dest_mnt = path->mnt; | |
1564 | struct dentry *dest_dentry = path->dentry; | |
1565 | struct vfsmount *child, *p; | |
1566 | int err; | |
1567 | ||
1568 | if (IS_MNT_SHARED(dest_mnt)) { | |
1569 | err = invent_group_ids(source_mnt, true); | |
1570 | if (err) | |
1571 | goto out; | |
1572 | } | |
1573 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1574 | if (err) | |
1575 | goto out_cleanup_ids; | |
1576 | ||
1577 | br_write_lock(vfsmount_lock); | |
1578 | ||
1579 | if (IS_MNT_SHARED(dest_mnt)) { | |
1580 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1581 | set_mnt_shared(p); | |
1582 | } | |
1583 | if (parent_path) { | |
1584 | detach_mnt(source_mnt, parent_path); | |
1585 | attach_mnt(source_mnt, path); | |
1586 | touch_mnt_namespace(parent_path->mnt->mnt_ns); | |
1587 | } else { | |
1588 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1589 | commit_tree(source_mnt); | |
1590 | } | |
1591 | ||
1592 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1593 | list_del_init(&child->mnt_hash); | |
1594 | commit_tree(child); | |
1595 | } | |
1596 | br_write_unlock(vfsmount_lock); | |
1597 | ||
1598 | return 0; | |
1599 | ||
1600 | out_cleanup_ids: | |
1601 | if (IS_MNT_SHARED(dest_mnt)) | |
1602 | cleanup_group_ids(source_mnt, NULL); | |
1603 | out: | |
1604 | return err; | |
1605 | } | |
1606 | ||
1607 | static int graft_tree(struct vfsmount *mnt, struct path *path) | |
1608 | { | |
1609 | int err; | |
1610 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1611 | return -EINVAL; | |
1612 | ||
1613 | if (S_ISDIR(path->dentry->d_inode->i_mode) != | |
1614 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) | |
1615 | return -ENOTDIR; | |
1616 | ||
1617 | err = -ENOENT; | |
1618 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1619 | if (cant_mount(path->dentry)) | |
1620 | goto out_unlock; | |
1621 | ||
1622 | if (!d_unlinked(path->dentry)) | |
1623 | err = attach_recursive_mnt(mnt, path, NULL); | |
1624 | out_unlock: | |
1625 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1626 | return err; | |
1627 | } | |
1628 | ||
1629 | /* | |
1630 | * Sanity check the flags to change_mnt_propagation. | |
1631 | */ | |
1632 | ||
1633 | static int flags_to_propagation_type(int flags) | |
1634 | { | |
1635 | int type = flags & ~MS_REC; | |
1636 | ||
1637 | /* Fail if any non-propagation flags are set */ | |
1638 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1639 | return 0; | |
1640 | /* Only one propagation flag should be set */ | |
1641 | if (!is_power_of_2(type)) | |
1642 | return 0; | |
1643 | return type; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * recursively change the type of the mountpoint. | |
1648 | */ | |
1649 | static int do_change_type(struct path *path, int flag) | |
1650 | { | |
1651 | struct vfsmount *m, *mnt = path->mnt; | |
1652 | int recurse = flag & MS_REC; | |
1653 | int type; | |
1654 | int err = 0; | |
1655 | ||
1656 | if (!capable(CAP_SYS_ADMIN)) | |
1657 | return -EPERM; | |
1658 | ||
1659 | if (path->dentry != path->mnt->mnt_root) | |
1660 | return -EINVAL; | |
1661 | ||
1662 | type = flags_to_propagation_type(flag); | |
1663 | if (!type) | |
1664 | return -EINVAL; | |
1665 | ||
1666 | down_write(&namespace_sem); | |
1667 | if (type == MS_SHARED) { | |
1668 | err = invent_group_ids(mnt, recurse); | |
1669 | if (err) | |
1670 | goto out_unlock; | |
1671 | } | |
1672 | ||
1673 | br_write_lock(vfsmount_lock); | |
1674 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1675 | change_mnt_propagation(m, type); | |
1676 | br_write_unlock(vfsmount_lock); | |
1677 | ||
1678 | out_unlock: | |
1679 | up_write(&namespace_sem); | |
1680 | return err; | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * do loopback mount. | |
1685 | */ | |
1686 | static int do_loopback(struct path *path, char *old_name, | |
1687 | int recurse) | |
1688 | { | |
1689 | struct path old_path; | |
1690 | struct vfsmount *mnt = NULL; | |
1691 | int err = mount_is_safe(path); | |
1692 | if (err) | |
1693 | return err; | |
1694 | if (!old_name || !*old_name) | |
1695 | return -EINVAL; | |
1696 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
1697 | if (err) | |
1698 | return err; | |
1699 | ||
1700 | down_write(&namespace_sem); | |
1701 | err = -EINVAL; | |
1702 | if (IS_MNT_UNBINDABLE(old_path.mnt)) | |
1703 | goto out; | |
1704 | ||
1705 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) | |
1706 | goto out; | |
1707 | ||
1708 | err = -ENOMEM; | |
1709 | if (recurse) | |
1710 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); | |
1711 | else | |
1712 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); | |
1713 | ||
1714 | if (!mnt) | |
1715 | goto out; | |
1716 | ||
1717 | err = graft_tree(mnt, path); | |
1718 | if (err) { | |
1719 | LIST_HEAD(umount_list); | |
1720 | ||
1721 | br_write_lock(vfsmount_lock); | |
1722 | umount_tree(mnt, 0, &umount_list); | |
1723 | br_write_unlock(vfsmount_lock); | |
1724 | release_mounts(&umount_list); | |
1725 | } | |
1726 | ||
1727 | out: | |
1728 | up_write(&namespace_sem); | |
1729 | path_put(&old_path); | |
1730 | return err; | |
1731 | } | |
1732 | ||
1733 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) | |
1734 | { | |
1735 | int error = 0; | |
1736 | int readonly_request = 0; | |
1737 | ||
1738 | if (ms_flags & MS_RDONLY) | |
1739 | readonly_request = 1; | |
1740 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1741 | return 0; | |
1742 | ||
1743 | if (readonly_request) | |
1744 | error = mnt_make_readonly(mnt); | |
1745 | else | |
1746 | __mnt_unmake_readonly(mnt); | |
1747 | return error; | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * change filesystem flags. dir should be a physical root of filesystem. | |
1752 | * If you've mounted a non-root directory somewhere and want to do remount | |
1753 | * on it - tough luck. | |
1754 | */ | |
1755 | static int do_remount(struct path *path, int flags, int mnt_flags, | |
1756 | void *data) | |
1757 | { | |
1758 | int err; | |
1759 | struct super_block *sb = path->mnt->mnt_sb; | |
1760 | ||
1761 | if (!capable(CAP_SYS_ADMIN)) | |
1762 | return -EPERM; | |
1763 | ||
1764 | if (!check_mnt(path->mnt)) | |
1765 | return -EINVAL; | |
1766 | ||
1767 | if (path->dentry != path->mnt->mnt_root) | |
1768 | return -EINVAL; | |
1769 | ||
1770 | down_write(&sb->s_umount); | |
1771 | if (flags & MS_BIND) | |
1772 | err = change_mount_flags(path->mnt, flags); | |
1773 | else | |
1774 | err = do_remount_sb(sb, flags, data, 0); | |
1775 | if (!err) { | |
1776 | br_write_lock(vfsmount_lock); | |
1777 | mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; | |
1778 | path->mnt->mnt_flags = mnt_flags; | |
1779 | br_write_unlock(vfsmount_lock); | |
1780 | } | |
1781 | up_write(&sb->s_umount); | |
1782 | if (!err) { | |
1783 | br_write_lock(vfsmount_lock); | |
1784 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1785 | br_write_unlock(vfsmount_lock); | |
1786 | } | |
1787 | return err; | |
1788 | } | |
1789 | ||
1790 | static inline int tree_contains_unbindable(struct vfsmount *mnt) | |
1791 | { | |
1792 | struct vfsmount *p; | |
1793 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1794 | if (IS_MNT_UNBINDABLE(p)) | |
1795 | return 1; | |
1796 | } | |
1797 | return 0; | |
1798 | } | |
1799 | ||
1800 | static int do_move_mount(struct path *path, char *old_name) | |
1801 | { | |
1802 | struct path old_path, parent_path; | |
1803 | struct vfsmount *p; | |
1804 | int err = 0; | |
1805 | if (!capable(CAP_SYS_ADMIN)) | |
1806 | return -EPERM; | |
1807 | if (!old_name || !*old_name) | |
1808 | return -EINVAL; | |
1809 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
1810 | if (err) | |
1811 | return err; | |
1812 | ||
1813 | down_write(&namespace_sem); | |
1814 | err = follow_down(path, true); | |
1815 | if (err < 0) | |
1816 | goto out; | |
1817 | ||
1818 | err = -EINVAL; | |
1819 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) | |
1820 | goto out; | |
1821 | ||
1822 | err = -ENOENT; | |
1823 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1824 | if (cant_mount(path->dentry)) | |
1825 | goto out1; | |
1826 | ||
1827 | if (d_unlinked(path->dentry)) | |
1828 | goto out1; | |
1829 | ||
1830 | err = -EINVAL; | |
1831 | if (old_path.dentry != old_path.mnt->mnt_root) | |
1832 | goto out1; | |
1833 | ||
1834 | if (old_path.mnt == old_path.mnt->mnt_parent) | |
1835 | goto out1; | |
1836 | ||
1837 | if (S_ISDIR(path->dentry->d_inode->i_mode) != | |
1838 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
1839 | goto out1; | |
1840 | /* | |
1841 | * Don't move a mount residing in a shared parent. | |
1842 | */ | |
1843 | if (old_path.mnt->mnt_parent && | |
1844 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
1845 | goto out1; | |
1846 | /* | |
1847 | * Don't move a mount tree containing unbindable mounts to a destination | |
1848 | * mount which is shared. | |
1849 | */ | |
1850 | if (IS_MNT_SHARED(path->mnt) && | |
1851 | tree_contains_unbindable(old_path.mnt)) | |
1852 | goto out1; | |
1853 | err = -ELOOP; | |
1854 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) | |
1855 | if (p == old_path.mnt) | |
1856 | goto out1; | |
1857 | ||
1858 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); | |
1859 | if (err) | |
1860 | goto out1; | |
1861 | ||
1862 | /* if the mount is moved, it should no longer be expire | |
1863 | * automatically */ | |
1864 | list_del_init(&old_path.mnt->mnt_expire); | |
1865 | out1: | |
1866 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1867 | out: | |
1868 | up_write(&namespace_sem); | |
1869 | if (!err) | |
1870 | path_put(&parent_path); | |
1871 | path_put(&old_path); | |
1872 | return err; | |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * create a new mount for userspace and request it to be added into the | |
1877 | * namespace's tree | |
1878 | */ | |
1879 | static int do_new_mount(struct path *path, char *type, int flags, | |
1880 | int mnt_flags, char *name, void *data) | |
1881 | { | |
1882 | struct vfsmount *mnt; | |
1883 | ||
1884 | if (!type) | |
1885 | return -EINVAL; | |
1886 | ||
1887 | /* we need capabilities... */ | |
1888 | if (!capable(CAP_SYS_ADMIN)) | |
1889 | return -EPERM; | |
1890 | ||
1891 | mnt = do_kern_mount(type, flags, name, data); | |
1892 | if (IS_ERR(mnt)) | |
1893 | return PTR_ERR(mnt); | |
1894 | ||
1895 | return do_add_mount(mnt, path, mnt_flags); | |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * add a mount into a namespace's mount tree | |
1900 | * - this unconditionally eats one of the caller's references to newmnt. | |
1901 | */ | |
1902 | int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags) | |
1903 | { | |
1904 | int err; | |
1905 | ||
1906 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1907 | ||
1908 | down_write(&namespace_sem); | |
1909 | /* Something was mounted here while we slept */ | |
1910 | err = follow_down(path, true); | |
1911 | if (err < 0) | |
1912 | goto unlock; | |
1913 | ||
1914 | err = -EINVAL; | |
1915 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) | |
1916 | goto unlock; | |
1917 | ||
1918 | /* Refuse the same filesystem on the same mount point */ | |
1919 | err = -EBUSY; | |
1920 | if (path->mnt->mnt_sb == newmnt->mnt_sb && | |
1921 | path->mnt->mnt_root == path->dentry) | |
1922 | goto unlock; | |
1923 | ||
1924 | err = -EINVAL; | |
1925 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1926 | goto unlock; | |
1927 | ||
1928 | newmnt->mnt_flags = mnt_flags; | |
1929 | if ((err = graft_tree(newmnt, path))) | |
1930 | goto unlock; | |
1931 | ||
1932 | up_write(&namespace_sem); | |
1933 | return 0; | |
1934 | ||
1935 | unlock: | |
1936 | up_write(&namespace_sem); | |
1937 | mntput(newmnt); | |
1938 | return err; | |
1939 | } | |
1940 | ||
1941 | /** | |
1942 | * mnt_set_expiry - Put a mount on an expiration list | |
1943 | * @mnt: The mount to list. | |
1944 | * @expiry_list: The list to add the mount to. | |
1945 | */ | |
1946 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
1947 | { | |
1948 | down_write(&namespace_sem); | |
1949 | br_write_lock(vfsmount_lock); | |
1950 | ||
1951 | list_add_tail(&mnt->mnt_expire, expiry_list); | |
1952 | ||
1953 | br_write_unlock(vfsmount_lock); | |
1954 | up_write(&namespace_sem); | |
1955 | } | |
1956 | EXPORT_SYMBOL(mnt_set_expiry); | |
1957 | ||
1958 | /* | |
1959 | * Remove a vfsmount from any expiration list it may be on | |
1960 | */ | |
1961 | void mnt_clear_expiry(struct vfsmount *mnt) | |
1962 | { | |
1963 | if (!list_empty(&mnt->mnt_expire)) { | |
1964 | down_write(&namespace_sem); | |
1965 | br_write_lock(vfsmount_lock); | |
1966 | list_del_init(&mnt->mnt_expire); | |
1967 | br_write_unlock(vfsmount_lock); | |
1968 | up_write(&namespace_sem); | |
1969 | } | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * process a list of expirable mountpoints with the intent of discarding any | |
1974 | * mountpoints that aren't in use and haven't been touched since last we came | |
1975 | * here | |
1976 | */ | |
1977 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1978 | { | |
1979 | struct vfsmount *mnt, *next; | |
1980 | LIST_HEAD(graveyard); | |
1981 | LIST_HEAD(umounts); | |
1982 | ||
1983 | if (list_empty(mounts)) | |
1984 | return; | |
1985 | ||
1986 | down_write(&namespace_sem); | |
1987 | br_write_lock(vfsmount_lock); | |
1988 | ||
1989 | /* extract from the expiration list every vfsmount that matches the | |
1990 | * following criteria: | |
1991 | * - only referenced by its parent vfsmount | |
1992 | * - still marked for expiry (marked on the last call here; marks are | |
1993 | * cleared by mntput()) | |
1994 | */ | |
1995 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { | |
1996 | if (!xchg(&mnt->mnt_expiry_mark, 1) || | |
1997 | propagate_mount_busy(mnt, 1)) | |
1998 | continue; | |
1999 | list_move(&mnt->mnt_expire, &graveyard); | |
2000 | } | |
2001 | while (!list_empty(&graveyard)) { | |
2002 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
2003 | touch_mnt_namespace(mnt->mnt_ns); | |
2004 | umount_tree(mnt, 1, &umounts); | |
2005 | } | |
2006 | br_write_unlock(vfsmount_lock); | |
2007 | up_write(&namespace_sem); | |
2008 | ||
2009 | release_mounts(&umounts); | |
2010 | } | |
2011 | ||
2012 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2013 | ||
2014 | /* | |
2015 | * Ripoff of 'select_parent()' | |
2016 | * | |
2017 | * search the list of submounts for a given mountpoint, and move any | |
2018 | * shrinkable submounts to the 'graveyard' list. | |
2019 | */ | |
2020 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
2021 | { | |
2022 | struct vfsmount *this_parent = parent; | |
2023 | struct list_head *next; | |
2024 | int found = 0; | |
2025 | ||
2026 | repeat: | |
2027 | next = this_parent->mnt_mounts.next; | |
2028 | resume: | |
2029 | while (next != &this_parent->mnt_mounts) { | |
2030 | struct list_head *tmp = next; | |
2031 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
2032 | ||
2033 | next = tmp->next; | |
2034 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
2035 | continue; | |
2036 | /* | |
2037 | * Descend a level if the d_mounts list is non-empty. | |
2038 | */ | |
2039 | if (!list_empty(&mnt->mnt_mounts)) { | |
2040 | this_parent = mnt; | |
2041 | goto repeat; | |
2042 | } | |
2043 | ||
2044 | if (!propagate_mount_busy(mnt, 1)) { | |
2045 | list_move_tail(&mnt->mnt_expire, graveyard); | |
2046 | found++; | |
2047 | } | |
2048 | } | |
2049 | /* | |
2050 | * All done at this level ... ascend and resume the search | |
2051 | */ | |
2052 | if (this_parent != parent) { | |
2053 | next = this_parent->mnt_child.next; | |
2054 | this_parent = this_parent->mnt_parent; | |
2055 | goto resume; | |
2056 | } | |
2057 | return found; | |
2058 | } | |
2059 | ||
2060 | /* | |
2061 | * process a list of expirable mountpoints with the intent of discarding any | |
2062 | * submounts of a specific parent mountpoint | |
2063 | * | |
2064 | * vfsmount_lock must be held for write | |
2065 | */ | |
2066 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) | |
2067 | { | |
2068 | LIST_HEAD(graveyard); | |
2069 | struct vfsmount *m; | |
2070 | ||
2071 | /* extract submounts of 'mountpoint' from the expiration list */ | |
2072 | while (select_submounts(mnt, &graveyard)) { | |
2073 | while (!list_empty(&graveyard)) { | |
2074 | m = list_first_entry(&graveyard, struct vfsmount, | |
2075 | mnt_expire); | |
2076 | touch_mnt_namespace(m->mnt_ns); | |
2077 | umount_tree(m, 1, umounts); | |
2078 | } | |
2079 | } | |
2080 | } | |
2081 | ||
2082 | /* | |
2083 | * Some copy_from_user() implementations do not return the exact number of | |
2084 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2085 | * Note that this function differs from copy_from_user() in that it will oops | |
2086 | * on bad values of `to', rather than returning a short copy. | |
2087 | */ | |
2088 | static long exact_copy_from_user(void *to, const void __user * from, | |
2089 | unsigned long n) | |
2090 | { | |
2091 | char *t = to; | |
2092 | const char __user *f = from; | |
2093 | char c; | |
2094 | ||
2095 | if (!access_ok(VERIFY_READ, from, n)) | |
2096 | return n; | |
2097 | ||
2098 | while (n) { | |
2099 | if (__get_user(c, f)) { | |
2100 | memset(t, 0, n); | |
2101 | break; | |
2102 | } | |
2103 | *t++ = c; | |
2104 | f++; | |
2105 | n--; | |
2106 | } | |
2107 | return n; | |
2108 | } | |
2109 | ||
2110 | int copy_mount_options(const void __user * data, unsigned long *where) | |
2111 | { | |
2112 | int i; | |
2113 | unsigned long page; | |
2114 | unsigned long size; | |
2115 | ||
2116 | *where = 0; | |
2117 | if (!data) | |
2118 | return 0; | |
2119 | ||
2120 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2121 | return -ENOMEM; | |
2122 | ||
2123 | /* We only care that *some* data at the address the user | |
2124 | * gave us is valid. Just in case, we'll zero | |
2125 | * the remainder of the page. | |
2126 | */ | |
2127 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2128 | size = TASK_SIZE - (unsigned long)data; | |
2129 | if (size > PAGE_SIZE) | |
2130 | size = PAGE_SIZE; | |
2131 | ||
2132 | i = size - exact_copy_from_user((void *)page, data, size); | |
2133 | if (!i) { | |
2134 | free_page(page); | |
2135 | return -EFAULT; | |
2136 | } | |
2137 | if (i != PAGE_SIZE) | |
2138 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2139 | *where = page; | |
2140 | return 0; | |
2141 | } | |
2142 | ||
2143 | int copy_mount_string(const void __user *data, char **where) | |
2144 | { | |
2145 | char *tmp; | |
2146 | ||
2147 | if (!data) { | |
2148 | *where = NULL; | |
2149 | return 0; | |
2150 | } | |
2151 | ||
2152 | tmp = strndup_user(data, PAGE_SIZE); | |
2153 | if (IS_ERR(tmp)) | |
2154 | return PTR_ERR(tmp); | |
2155 | ||
2156 | *where = tmp; | |
2157 | return 0; | |
2158 | } | |
2159 | ||
2160 | /* | |
2161 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2162 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2163 | * | |
2164 | * data is a (void *) that can point to any structure up to | |
2165 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2166 | * information (or be NULL). | |
2167 | * | |
2168 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2169 | * When the flags word was introduced its top half was required | |
2170 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2171 | * Therefore, if this magic number is present, it carries no information | |
2172 | * and must be discarded. | |
2173 | */ | |
2174 | long do_mount(char *dev_name, char *dir_name, char *type_page, | |
2175 | unsigned long flags, void *data_page) | |
2176 | { | |
2177 | struct path path; | |
2178 | int retval = 0; | |
2179 | int mnt_flags = 0; | |
2180 | ||
2181 | /* Discard magic */ | |
2182 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2183 | flags &= ~MS_MGC_MSK; | |
2184 | ||
2185 | /* Basic sanity checks */ | |
2186 | ||
2187 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2188 | return -EINVAL; | |
2189 | ||
2190 | if (data_page) | |
2191 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2192 | ||
2193 | /* ... and get the mountpoint */ | |
2194 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2195 | if (retval) | |
2196 | return retval; | |
2197 | ||
2198 | retval = security_sb_mount(dev_name, &path, | |
2199 | type_page, flags, data_page); | |
2200 | if (retval) | |
2201 | goto dput_out; | |
2202 | ||
2203 | /* Default to relatime unless overriden */ | |
2204 | if (!(flags & MS_NOATIME)) | |
2205 | mnt_flags |= MNT_RELATIME; | |
2206 | ||
2207 | /* Separate the per-mountpoint flags */ | |
2208 | if (flags & MS_NOSUID) | |
2209 | mnt_flags |= MNT_NOSUID; | |
2210 | if (flags & MS_NODEV) | |
2211 | mnt_flags |= MNT_NODEV; | |
2212 | if (flags & MS_NOEXEC) | |
2213 | mnt_flags |= MNT_NOEXEC; | |
2214 | if (flags & MS_NOATIME) | |
2215 | mnt_flags |= MNT_NOATIME; | |
2216 | if (flags & MS_NODIRATIME) | |
2217 | mnt_flags |= MNT_NODIRATIME; | |
2218 | if (flags & MS_STRICTATIME) | |
2219 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2220 | if (flags & MS_RDONLY) | |
2221 | mnt_flags |= MNT_READONLY; | |
2222 | ||
2223 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | | |
2224 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | | |
2225 | MS_STRICTATIME); | |
2226 | ||
2227 | if (flags & MS_REMOUNT) | |
2228 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, | |
2229 | data_page); | |
2230 | else if (flags & MS_BIND) | |
2231 | retval = do_loopback(&path, dev_name, flags & MS_REC); | |
2232 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2233 | retval = do_change_type(&path, flags); | |
2234 | else if (flags & MS_MOVE) | |
2235 | retval = do_move_mount(&path, dev_name); | |
2236 | else | |
2237 | retval = do_new_mount(&path, type_page, flags, mnt_flags, | |
2238 | dev_name, data_page); | |
2239 | dput_out: | |
2240 | path_put(&path); | |
2241 | return retval; | |
2242 | } | |
2243 | ||
2244 | static struct mnt_namespace *alloc_mnt_ns(void) | |
2245 | { | |
2246 | struct mnt_namespace *new_ns; | |
2247 | ||
2248 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2249 | if (!new_ns) | |
2250 | return ERR_PTR(-ENOMEM); | |
2251 | atomic_set(&new_ns->count, 1); | |
2252 | new_ns->root = NULL; | |
2253 | INIT_LIST_HEAD(&new_ns->list); | |
2254 | init_waitqueue_head(&new_ns->poll); | |
2255 | new_ns->event = 0; | |
2256 | return new_ns; | |
2257 | } | |
2258 | ||
2259 | void mnt_make_longterm(struct vfsmount *mnt) | |
2260 | { | |
2261 | __mnt_make_longterm(mnt); | |
2262 | } | |
2263 | ||
2264 | void mnt_make_shortterm(struct vfsmount *mnt) | |
2265 | { | |
2266 | #ifdef CONFIG_SMP | |
2267 | if (atomic_add_unless(&mnt->mnt_longterm, -1, 1)) | |
2268 | return; | |
2269 | br_write_lock(vfsmount_lock); | |
2270 | atomic_dec(&mnt->mnt_longterm); | |
2271 | br_write_unlock(vfsmount_lock); | |
2272 | #endif | |
2273 | } | |
2274 | ||
2275 | /* | |
2276 | * Allocate a new namespace structure and populate it with contents | |
2277 | * copied from the namespace of the passed in task structure. | |
2278 | */ | |
2279 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, | |
2280 | struct fs_struct *fs) | |
2281 | { | |
2282 | struct mnt_namespace *new_ns; | |
2283 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; | |
2284 | struct vfsmount *p, *q; | |
2285 | ||
2286 | new_ns = alloc_mnt_ns(); | |
2287 | if (IS_ERR(new_ns)) | |
2288 | return new_ns; | |
2289 | ||
2290 | down_write(&namespace_sem); | |
2291 | /* First pass: copy the tree topology */ | |
2292 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, | |
2293 | CL_COPY_ALL | CL_EXPIRE); | |
2294 | if (!new_ns->root) { | |
2295 | up_write(&namespace_sem); | |
2296 | kfree(new_ns); | |
2297 | return ERR_PTR(-ENOMEM); | |
2298 | } | |
2299 | br_write_lock(vfsmount_lock); | |
2300 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
2301 | br_write_unlock(vfsmount_lock); | |
2302 | ||
2303 | /* | |
2304 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2305 | * as belonging to new namespace. We have already acquired a private | |
2306 | * fs_struct, so tsk->fs->lock is not needed. | |
2307 | */ | |
2308 | p = mnt_ns->root; | |
2309 | q = new_ns->root; | |
2310 | while (p) { | |
2311 | q->mnt_ns = new_ns; | |
2312 | __mnt_make_longterm(q); | |
2313 | if (fs) { | |
2314 | if (p == fs->root.mnt) { | |
2315 | fs->root.mnt = mntget(q); | |
2316 | __mnt_make_longterm(q); | |
2317 | mnt_make_shortterm(p); | |
2318 | rootmnt = p; | |
2319 | } | |
2320 | if (p == fs->pwd.mnt) { | |
2321 | fs->pwd.mnt = mntget(q); | |
2322 | __mnt_make_longterm(q); | |
2323 | mnt_make_shortterm(p); | |
2324 | pwdmnt = p; | |
2325 | } | |
2326 | } | |
2327 | p = next_mnt(p, mnt_ns->root); | |
2328 | q = next_mnt(q, new_ns->root); | |
2329 | } | |
2330 | up_write(&namespace_sem); | |
2331 | ||
2332 | if (rootmnt) | |
2333 | mntput(rootmnt); | |
2334 | if (pwdmnt) | |
2335 | mntput(pwdmnt); | |
2336 | ||
2337 | return new_ns; | |
2338 | } | |
2339 | ||
2340 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, | |
2341 | struct fs_struct *new_fs) | |
2342 | { | |
2343 | struct mnt_namespace *new_ns; | |
2344 | ||
2345 | BUG_ON(!ns); | |
2346 | get_mnt_ns(ns); | |
2347 | ||
2348 | if (!(flags & CLONE_NEWNS)) | |
2349 | return ns; | |
2350 | ||
2351 | new_ns = dup_mnt_ns(ns, new_fs); | |
2352 | ||
2353 | put_mnt_ns(ns); | |
2354 | return new_ns; | |
2355 | } | |
2356 | ||
2357 | /** | |
2358 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2359 | * @mnt: pointer to the new root filesystem mountpoint | |
2360 | */ | |
2361 | struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) | |
2362 | { | |
2363 | struct mnt_namespace *new_ns; | |
2364 | ||
2365 | new_ns = alloc_mnt_ns(); | |
2366 | if (!IS_ERR(new_ns)) { | |
2367 | mnt->mnt_ns = new_ns; | |
2368 | __mnt_make_longterm(mnt); | |
2369 | new_ns->root = mnt; | |
2370 | list_add(&new_ns->list, &new_ns->root->mnt_list); | |
2371 | } | |
2372 | return new_ns; | |
2373 | } | |
2374 | EXPORT_SYMBOL(create_mnt_ns); | |
2375 | ||
2376 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, | |
2377 | char __user *, type, unsigned long, flags, void __user *, data) | |
2378 | { | |
2379 | int ret; | |
2380 | char *kernel_type; | |
2381 | char *kernel_dir; | |
2382 | char *kernel_dev; | |
2383 | unsigned long data_page; | |
2384 | ||
2385 | ret = copy_mount_string(type, &kernel_type); | |
2386 | if (ret < 0) | |
2387 | goto out_type; | |
2388 | ||
2389 | kernel_dir = getname(dir_name); | |
2390 | if (IS_ERR(kernel_dir)) { | |
2391 | ret = PTR_ERR(kernel_dir); | |
2392 | goto out_dir; | |
2393 | } | |
2394 | ||
2395 | ret = copy_mount_string(dev_name, &kernel_dev); | |
2396 | if (ret < 0) | |
2397 | goto out_dev; | |
2398 | ||
2399 | ret = copy_mount_options(data, &data_page); | |
2400 | if (ret < 0) | |
2401 | goto out_data; | |
2402 | ||
2403 | ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, | |
2404 | (void *) data_page); | |
2405 | ||
2406 | free_page(data_page); | |
2407 | out_data: | |
2408 | kfree(kernel_dev); | |
2409 | out_dev: | |
2410 | putname(kernel_dir); | |
2411 | out_dir: | |
2412 | kfree(kernel_type); | |
2413 | out_type: | |
2414 | return ret; | |
2415 | } | |
2416 | ||
2417 | /* | |
2418 | * pivot_root Semantics: | |
2419 | * Moves the root file system of the current process to the directory put_old, | |
2420 | * makes new_root as the new root file system of the current process, and sets | |
2421 | * root/cwd of all processes which had them on the current root to new_root. | |
2422 | * | |
2423 | * Restrictions: | |
2424 | * The new_root and put_old must be directories, and must not be on the | |
2425 | * same file system as the current process root. The put_old must be | |
2426 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2427 | * pointed to by put_old must yield the same directory as new_root. No other | |
2428 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2429 | * | |
2430 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. | |
2431 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2432 | * in this situation. | |
2433 | * | |
2434 | * Notes: | |
2435 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2436 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2437 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2438 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2439 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2440 | * first. | |
2441 | */ | |
2442 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, | |
2443 | const char __user *, put_old) | |
2444 | { | |
2445 | struct vfsmount *tmp; | |
2446 | struct path new, old, parent_path, root_parent, root; | |
2447 | int error; | |
2448 | ||
2449 | if (!capable(CAP_SYS_ADMIN)) | |
2450 | return -EPERM; | |
2451 | ||
2452 | error = user_path_dir(new_root, &new); | |
2453 | if (error) | |
2454 | goto out0; | |
2455 | error = -EINVAL; | |
2456 | if (!check_mnt(new.mnt)) | |
2457 | goto out1; | |
2458 | ||
2459 | error = user_path_dir(put_old, &old); | |
2460 | if (error) | |
2461 | goto out1; | |
2462 | ||
2463 | error = security_sb_pivotroot(&old, &new); | |
2464 | if (error) { | |
2465 | path_put(&old); | |
2466 | goto out1; | |
2467 | } | |
2468 | ||
2469 | get_fs_root(current->fs, &root); | |
2470 | down_write(&namespace_sem); | |
2471 | mutex_lock(&old.dentry->d_inode->i_mutex); | |
2472 | error = -EINVAL; | |
2473 | if (IS_MNT_SHARED(old.mnt) || | |
2474 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
2475 | IS_MNT_SHARED(root.mnt->mnt_parent)) | |
2476 | goto out2; | |
2477 | if (!check_mnt(root.mnt)) | |
2478 | goto out2; | |
2479 | error = -ENOENT; | |
2480 | if (cant_mount(old.dentry)) | |
2481 | goto out2; | |
2482 | if (d_unlinked(new.dentry)) | |
2483 | goto out2; | |
2484 | if (d_unlinked(old.dentry)) | |
2485 | goto out2; | |
2486 | error = -EBUSY; | |
2487 | if (new.mnt == root.mnt || | |
2488 | old.mnt == root.mnt) | |
2489 | goto out2; /* loop, on the same file system */ | |
2490 | error = -EINVAL; | |
2491 | if (root.mnt->mnt_root != root.dentry) | |
2492 | goto out2; /* not a mountpoint */ | |
2493 | if (root.mnt->mnt_parent == root.mnt) | |
2494 | goto out2; /* not attached */ | |
2495 | if (new.mnt->mnt_root != new.dentry) | |
2496 | goto out2; /* not a mountpoint */ | |
2497 | if (new.mnt->mnt_parent == new.mnt) | |
2498 | goto out2; /* not attached */ | |
2499 | /* make sure we can reach put_old from new_root */ | |
2500 | tmp = old.mnt; | |
2501 | br_write_lock(vfsmount_lock); | |
2502 | if (tmp != new.mnt) { | |
2503 | for (;;) { | |
2504 | if (tmp->mnt_parent == tmp) | |
2505 | goto out3; /* already mounted on put_old */ | |
2506 | if (tmp->mnt_parent == new.mnt) | |
2507 | break; | |
2508 | tmp = tmp->mnt_parent; | |
2509 | } | |
2510 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) | |
2511 | goto out3; | |
2512 | } else if (!is_subdir(old.dentry, new.dentry)) | |
2513 | goto out3; | |
2514 | detach_mnt(new.mnt, &parent_path); | |
2515 | detach_mnt(root.mnt, &root_parent); | |
2516 | /* mount old root on put_old */ | |
2517 | attach_mnt(root.mnt, &old); | |
2518 | /* mount new_root on / */ | |
2519 | attach_mnt(new.mnt, &root_parent); | |
2520 | touch_mnt_namespace(current->nsproxy->mnt_ns); | |
2521 | br_write_unlock(vfsmount_lock); | |
2522 | chroot_fs_refs(&root, &new); | |
2523 | ||
2524 | error = 0; | |
2525 | path_put(&root_parent); | |
2526 | path_put(&parent_path); | |
2527 | out2: | |
2528 | mutex_unlock(&old.dentry->d_inode->i_mutex); | |
2529 | up_write(&namespace_sem); | |
2530 | path_put(&root); | |
2531 | path_put(&old); | |
2532 | out1: | |
2533 | path_put(&new); | |
2534 | out0: | |
2535 | return error; | |
2536 | out3: | |
2537 | br_write_unlock(vfsmount_lock); | |
2538 | goto out2; | |
2539 | } | |
2540 | ||
2541 | static void __init init_mount_tree(void) | |
2542 | { | |
2543 | struct vfsmount *mnt; | |
2544 | struct mnt_namespace *ns; | |
2545 | struct path root; | |
2546 | ||
2547 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2548 | if (IS_ERR(mnt)) | |
2549 | panic("Can't create rootfs"); | |
2550 | ||
2551 | ns = create_mnt_ns(mnt); | |
2552 | if (IS_ERR(ns)) | |
2553 | panic("Can't allocate initial namespace"); | |
2554 | ||
2555 | init_task.nsproxy->mnt_ns = ns; | |
2556 | get_mnt_ns(ns); | |
2557 | ||
2558 | root.mnt = ns->root; | |
2559 | root.dentry = ns->root->mnt_root; | |
2560 | ||
2561 | set_fs_pwd(current->fs, &root); | |
2562 | set_fs_root(current->fs, &root); | |
2563 | } | |
2564 | ||
2565 | void __init mnt_init(void) | |
2566 | { | |
2567 | unsigned u; | |
2568 | int err; | |
2569 | ||
2570 | init_rwsem(&namespace_sem); | |
2571 | ||
2572 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), | |
2573 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
2574 | ||
2575 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); | |
2576 | ||
2577 | if (!mount_hashtable) | |
2578 | panic("Failed to allocate mount hash table\n"); | |
2579 | ||
2580 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); | |
2581 | ||
2582 | for (u = 0; u < HASH_SIZE; u++) | |
2583 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
2584 | ||
2585 | br_lock_init(vfsmount_lock); | |
2586 | ||
2587 | err = sysfs_init(); | |
2588 | if (err) | |
2589 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
2590 | __func__, err); | |
2591 | fs_kobj = kobject_create_and_add("fs", NULL); | |
2592 | if (!fs_kobj) | |
2593 | printk(KERN_WARNING "%s: kobj create error\n", __func__); | |
2594 | init_rootfs(); | |
2595 | init_mount_tree(); | |
2596 | } | |
2597 | ||
2598 | void put_mnt_ns(struct mnt_namespace *ns) | |
2599 | { | |
2600 | LIST_HEAD(umount_list); | |
2601 | ||
2602 | if (!atomic_dec_and_test(&ns->count)) | |
2603 | return; | |
2604 | down_write(&namespace_sem); | |
2605 | br_write_lock(vfsmount_lock); | |
2606 | umount_tree(ns->root, 0, &umount_list); | |
2607 | br_write_unlock(vfsmount_lock); | |
2608 | up_write(&namespace_sem); | |
2609 | release_mounts(&umount_list); | |
2610 | kfree(ns); | |
2611 | } | |
2612 | EXPORT_SYMBOL(put_mnt_ns); |