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> |
1da177e4 LT |
30 | #include <asm/uaccess.h> |
31 | #include <asm/unistd.h> | |
07b20889 | 32 | #include "pnode.h" |
948730b0 | 33 | #include "internal.h" |
1da177e4 | 34 | |
13f14b4d ED |
35 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
36 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
37 | ||
1da177e4 | 38 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
39 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
40 | ||
41 | static int event; | |
73cd49ec | 42 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 43 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 44 | |
fa3536cc | 45 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 46 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 47 | static struct rw_semaphore namespace_sem; |
1da177e4 | 48 | |
f87fd4c2 | 49 | /* /sys/fs */ |
00d26666 GKH |
50 | struct kobject *fs_kobj; |
51 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 52 | |
1da177e4 LT |
53 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
54 | { | |
b58fed8b RP |
55 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
56 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
57 | tmp = tmp + (tmp >> HASH_SHIFT); |
58 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
59 | } |
60 | ||
3d733633 DH |
61 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
62 | ||
73cd49ec MS |
63 | /* allocation is serialized by namespace_sem */ |
64 | static int mnt_alloc_id(struct vfsmount *mnt) | |
65 | { | |
66 | int res; | |
67 | ||
68 | retry: | |
69 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
70 | spin_lock(&vfsmount_lock); | |
71 | res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); | |
72 | spin_unlock(&vfsmount_lock); | |
73 | if (res == -EAGAIN) | |
74 | goto retry; | |
75 | ||
76 | return res; | |
77 | } | |
78 | ||
79 | static void mnt_free_id(struct vfsmount *mnt) | |
80 | { | |
81 | spin_lock(&vfsmount_lock); | |
82 | ida_remove(&mnt_id_ida, mnt->mnt_id); | |
83 | spin_unlock(&vfsmount_lock); | |
84 | } | |
85 | ||
719f5d7f MS |
86 | /* |
87 | * Allocate a new peer group ID | |
88 | * | |
89 | * mnt_group_ida is protected by namespace_sem | |
90 | */ | |
91 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
92 | { | |
93 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) | |
94 | return -ENOMEM; | |
95 | ||
96 | return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id); | |
97 | } | |
98 | ||
99 | /* | |
100 | * Release a peer group ID | |
101 | */ | |
102 | void mnt_release_group_id(struct vfsmount *mnt) | |
103 | { | |
104 | ida_remove(&mnt_group_ida, mnt->mnt_group_id); | |
105 | mnt->mnt_group_id = 0; | |
106 | } | |
107 | ||
1da177e4 LT |
108 | struct vfsmount *alloc_vfsmnt(const char *name) |
109 | { | |
c3762229 | 110 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 111 | if (mnt) { |
73cd49ec MS |
112 | int err; |
113 | ||
114 | err = mnt_alloc_id(mnt); | |
115 | if (err) { | |
116 | kmem_cache_free(mnt_cache, mnt); | |
117 | return NULL; | |
118 | } | |
119 | ||
b58fed8b | 120 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
121 | INIT_LIST_HEAD(&mnt->mnt_hash); |
122 | INIT_LIST_HEAD(&mnt->mnt_child); | |
123 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
124 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 125 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 126 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
127 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
128 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
3d733633 | 129 | atomic_set(&mnt->__mnt_writers, 0); |
1da177e4 | 130 | if (name) { |
b58fed8b | 131 | int size = strlen(name) + 1; |
1da177e4 LT |
132 | char *newname = kmalloc(size, GFP_KERNEL); |
133 | if (newname) { | |
134 | memcpy(newname, name, size); | |
135 | mnt->mnt_devname = newname; | |
136 | } | |
137 | } | |
138 | } | |
139 | return mnt; | |
140 | } | |
141 | ||
3d733633 DH |
142 | /* |
143 | * Most r/o checks on a fs are for operations that take | |
144 | * discrete amounts of time, like a write() or unlink(). | |
145 | * We must keep track of when those operations start | |
146 | * (for permission checks) and when they end, so that | |
147 | * we can determine when writes are able to occur to | |
148 | * a filesystem. | |
149 | */ | |
150 | /* | |
151 | * __mnt_is_readonly: check whether a mount is read-only | |
152 | * @mnt: the mount to check for its write status | |
153 | * | |
154 | * This shouldn't be used directly ouside of the VFS. | |
155 | * It does not guarantee that the filesystem will stay | |
156 | * r/w, just that it is right *now*. This can not and | |
157 | * should not be used in place of IS_RDONLY(inode). | |
158 | * mnt_want/drop_write() will _keep_ the filesystem | |
159 | * r/w. | |
160 | */ | |
161 | int __mnt_is_readonly(struct vfsmount *mnt) | |
162 | { | |
2e4b7fcd DH |
163 | if (mnt->mnt_flags & MNT_READONLY) |
164 | return 1; | |
165 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
166 | return 1; | |
167 | return 0; | |
3d733633 DH |
168 | } |
169 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
170 | ||
171 | struct mnt_writer { | |
172 | /* | |
173 | * If holding multiple instances of this lock, they | |
174 | * must be ordered by cpu number. | |
175 | */ | |
176 | spinlock_t lock; | |
177 | struct lock_class_key lock_class; /* compiles out with !lockdep */ | |
178 | unsigned long count; | |
179 | struct vfsmount *mnt; | |
180 | } ____cacheline_aligned_in_smp; | |
181 | static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); | |
182 | ||
183 | static int __init init_mnt_writers(void) | |
184 | { | |
185 | int cpu; | |
186 | for_each_possible_cpu(cpu) { | |
187 | struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); | |
188 | spin_lock_init(&writer->lock); | |
189 | lockdep_set_class(&writer->lock, &writer->lock_class); | |
190 | writer->count = 0; | |
191 | } | |
192 | return 0; | |
193 | } | |
194 | fs_initcall(init_mnt_writers); | |
195 | ||
196 | static void unlock_mnt_writers(void) | |
197 | { | |
198 | int cpu; | |
199 | struct mnt_writer *cpu_writer; | |
200 | ||
201 | for_each_possible_cpu(cpu) { | |
202 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
203 | spin_unlock(&cpu_writer->lock); | |
204 | } | |
205 | } | |
206 | ||
207 | static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) | |
208 | { | |
209 | if (!cpu_writer->mnt) | |
210 | return; | |
211 | /* | |
212 | * This is in case anyone ever leaves an invalid, | |
213 | * old ->mnt and a count of 0. | |
214 | */ | |
215 | if (!cpu_writer->count) | |
216 | return; | |
217 | atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); | |
218 | cpu_writer->count = 0; | |
219 | } | |
220 | /* | |
221 | * must hold cpu_writer->lock | |
222 | */ | |
223 | static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, | |
224 | struct vfsmount *mnt) | |
225 | { | |
226 | if (cpu_writer->mnt == mnt) | |
227 | return; | |
228 | __clear_mnt_count(cpu_writer); | |
229 | cpu_writer->mnt = mnt; | |
230 | } | |
231 | ||
8366025e DH |
232 | /* |
233 | * Most r/o checks on a fs are for operations that take | |
234 | * discrete amounts of time, like a write() or unlink(). | |
235 | * We must keep track of when those operations start | |
236 | * (for permission checks) and when they end, so that | |
237 | * we can determine when writes are able to occur to | |
238 | * a filesystem. | |
239 | */ | |
240 | /** | |
241 | * mnt_want_write - get write access to a mount | |
242 | * @mnt: the mount on which to take a write | |
243 | * | |
244 | * This tells the low-level filesystem that a write is | |
245 | * about to be performed to it, and makes sure that | |
246 | * writes are allowed before returning success. When | |
247 | * the write operation is finished, mnt_drop_write() | |
248 | * must be called. This is effectively a refcount. | |
249 | */ | |
250 | int mnt_want_write(struct vfsmount *mnt) | |
251 | { | |
3d733633 DH |
252 | int ret = 0; |
253 | struct mnt_writer *cpu_writer; | |
254 | ||
255 | cpu_writer = &get_cpu_var(mnt_writers); | |
256 | spin_lock(&cpu_writer->lock); | |
257 | if (__mnt_is_readonly(mnt)) { | |
258 | ret = -EROFS; | |
259 | goto out; | |
260 | } | |
261 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
262 | cpu_writer->count++; | |
263 | out: | |
264 | spin_unlock(&cpu_writer->lock); | |
265 | put_cpu_var(mnt_writers); | |
266 | return ret; | |
8366025e DH |
267 | } |
268 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
269 | ||
3d733633 DH |
270 | static void lock_mnt_writers(void) |
271 | { | |
272 | int cpu; | |
273 | struct mnt_writer *cpu_writer; | |
274 | ||
275 | for_each_possible_cpu(cpu) { | |
276 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
277 | spin_lock(&cpu_writer->lock); | |
278 | __clear_mnt_count(cpu_writer); | |
279 | cpu_writer->mnt = NULL; | |
280 | } | |
281 | } | |
282 | ||
283 | /* | |
284 | * These per-cpu write counts are not guaranteed to have | |
285 | * matched increments and decrements on any given cpu. | |
286 | * A file open()ed for write on one cpu and close()d on | |
287 | * another cpu will imbalance this count. Make sure it | |
288 | * does not get too far out of whack. | |
289 | */ | |
290 | static void handle_write_count_underflow(struct vfsmount *mnt) | |
291 | { | |
292 | if (atomic_read(&mnt->__mnt_writers) >= | |
293 | MNT_WRITER_UNDERFLOW_LIMIT) | |
294 | return; | |
295 | /* | |
296 | * It isn't necessary to hold all of the locks | |
297 | * at the same time, but doing it this way makes | |
298 | * us share a lot more code. | |
299 | */ | |
300 | lock_mnt_writers(); | |
301 | /* | |
302 | * vfsmount_lock is for mnt_flags. | |
303 | */ | |
304 | spin_lock(&vfsmount_lock); | |
305 | /* | |
306 | * If coalescing the per-cpu writer counts did not | |
307 | * get us back to a positive writer count, we have | |
308 | * a bug. | |
309 | */ | |
310 | if ((atomic_read(&mnt->__mnt_writers) < 0) && | |
311 | !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { | |
312 | printk(KERN_DEBUG "leak detected on mount(%p) writers " | |
313 | "count: %d\n", | |
314 | mnt, atomic_read(&mnt->__mnt_writers)); | |
315 | WARN_ON(1); | |
316 | /* use the flag to keep the dmesg spam down */ | |
317 | mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; | |
318 | } | |
319 | spin_unlock(&vfsmount_lock); | |
320 | unlock_mnt_writers(); | |
321 | } | |
322 | ||
8366025e DH |
323 | /** |
324 | * mnt_drop_write - give up write access to a mount | |
325 | * @mnt: the mount on which to give up write access | |
326 | * | |
327 | * Tells the low-level filesystem that we are done | |
328 | * performing writes to it. Must be matched with | |
329 | * mnt_want_write() call above. | |
330 | */ | |
331 | void mnt_drop_write(struct vfsmount *mnt) | |
332 | { | |
3d733633 DH |
333 | int must_check_underflow = 0; |
334 | struct mnt_writer *cpu_writer; | |
335 | ||
336 | cpu_writer = &get_cpu_var(mnt_writers); | |
337 | spin_lock(&cpu_writer->lock); | |
338 | ||
339 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
340 | if (cpu_writer->count > 0) { | |
341 | cpu_writer->count--; | |
342 | } else { | |
343 | must_check_underflow = 1; | |
344 | atomic_dec(&mnt->__mnt_writers); | |
345 | } | |
346 | ||
347 | spin_unlock(&cpu_writer->lock); | |
348 | /* | |
349 | * Logically, we could call this each time, | |
350 | * but the __mnt_writers cacheline tends to | |
351 | * be cold, and makes this expensive. | |
352 | */ | |
353 | if (must_check_underflow) | |
354 | handle_write_count_underflow(mnt); | |
355 | /* | |
356 | * This could be done right after the spinlock | |
357 | * is taken because the spinlock keeps us on | |
358 | * the cpu, and disables preemption. However, | |
359 | * putting it here bounds the amount that | |
360 | * __mnt_writers can underflow. Without it, | |
361 | * we could theoretically wrap __mnt_writers. | |
362 | */ | |
363 | put_cpu_var(mnt_writers); | |
8366025e DH |
364 | } |
365 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
366 | ||
2e4b7fcd | 367 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 368 | { |
3d733633 DH |
369 | int ret = 0; |
370 | ||
371 | lock_mnt_writers(); | |
372 | /* | |
373 | * With all the locks held, this value is stable | |
374 | */ | |
375 | if (atomic_read(&mnt->__mnt_writers) > 0) { | |
376 | ret = -EBUSY; | |
377 | goto out; | |
378 | } | |
379 | /* | |
2e4b7fcd DH |
380 | * nobody can do a successful mnt_want_write() with all |
381 | * of the counts in MNT_DENIED_WRITE and the locks held. | |
3d733633 | 382 | */ |
2e4b7fcd DH |
383 | spin_lock(&vfsmount_lock); |
384 | if (!ret) | |
385 | mnt->mnt_flags |= MNT_READONLY; | |
386 | spin_unlock(&vfsmount_lock); | |
3d733633 DH |
387 | out: |
388 | unlock_mnt_writers(); | |
389 | return ret; | |
8366025e | 390 | } |
8366025e | 391 | |
2e4b7fcd DH |
392 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
393 | { | |
394 | spin_lock(&vfsmount_lock); | |
395 | mnt->mnt_flags &= ~MNT_READONLY; | |
396 | spin_unlock(&vfsmount_lock); | |
397 | } | |
398 | ||
454e2398 DH |
399 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
400 | { | |
401 | mnt->mnt_sb = sb; | |
402 | mnt->mnt_root = dget(sb->s_root); | |
403 | return 0; | |
404 | } | |
405 | ||
406 | EXPORT_SYMBOL(simple_set_mnt); | |
407 | ||
1da177e4 LT |
408 | void free_vfsmnt(struct vfsmount *mnt) |
409 | { | |
410 | kfree(mnt->mnt_devname); | |
73cd49ec | 411 | mnt_free_id(mnt); |
1da177e4 LT |
412 | kmem_cache_free(mnt_cache, mnt); |
413 | } | |
414 | ||
415 | /* | |
a05964f3 RP |
416 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
417 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 418 | */ |
a05964f3 RP |
419 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
420 | int dir) | |
1da177e4 | 421 | { |
b58fed8b RP |
422 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
423 | struct list_head *tmp = head; | |
1da177e4 LT |
424 | struct vfsmount *p, *found = NULL; |
425 | ||
1da177e4 | 426 | for (;;) { |
a05964f3 | 427 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
428 | p = NULL; |
429 | if (tmp == head) | |
430 | break; | |
431 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
432 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 433 | found = p; |
1da177e4 LT |
434 | break; |
435 | } | |
436 | } | |
1da177e4 LT |
437 | return found; |
438 | } | |
439 | ||
a05964f3 RP |
440 | /* |
441 | * lookup_mnt increments the ref count before returning | |
442 | * the vfsmount struct. | |
443 | */ | |
444 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
445 | { | |
446 | struct vfsmount *child_mnt; | |
447 | spin_lock(&vfsmount_lock); | |
448 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
449 | mntget(child_mnt); | |
450 | spin_unlock(&vfsmount_lock); | |
451 | return child_mnt; | |
452 | } | |
453 | ||
1da177e4 LT |
454 | static inline int check_mnt(struct vfsmount *mnt) |
455 | { | |
6b3286ed | 456 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
457 | } |
458 | ||
6b3286ed | 459 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
460 | { |
461 | if (ns) { | |
462 | ns->event = ++event; | |
463 | wake_up_interruptible(&ns->poll); | |
464 | } | |
465 | } | |
466 | ||
6b3286ed | 467 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
468 | { |
469 | if (ns && ns->event != event) { | |
470 | ns->event = event; | |
471 | wake_up_interruptible(&ns->poll); | |
472 | } | |
473 | } | |
474 | ||
1a390689 | 475 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 476 | { |
1a390689 AV |
477 | old_path->dentry = mnt->mnt_mountpoint; |
478 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
479 | mnt->mnt_parent = mnt; |
480 | mnt->mnt_mountpoint = mnt->mnt_root; | |
481 | list_del_init(&mnt->mnt_child); | |
482 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 483 | old_path->dentry->d_mounted--; |
1da177e4 LT |
484 | } |
485 | ||
b90fa9ae RP |
486 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
487 | struct vfsmount *child_mnt) | |
488 | { | |
489 | child_mnt->mnt_parent = mntget(mnt); | |
490 | child_mnt->mnt_mountpoint = dget(dentry); | |
491 | dentry->d_mounted++; | |
492 | } | |
493 | ||
1a390689 | 494 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 495 | { |
1a390689 | 496 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 497 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
498 | hash(path->mnt, path->dentry)); |
499 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
500 | } |
501 | ||
502 | /* | |
503 | * the caller must hold vfsmount_lock | |
504 | */ | |
505 | static void commit_tree(struct vfsmount *mnt) | |
506 | { | |
507 | struct vfsmount *parent = mnt->mnt_parent; | |
508 | struct vfsmount *m; | |
509 | LIST_HEAD(head); | |
6b3286ed | 510 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
511 | |
512 | BUG_ON(parent == mnt); | |
513 | ||
514 | list_add_tail(&head, &mnt->mnt_list); | |
515 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 516 | m->mnt_ns = n; |
b90fa9ae RP |
517 | list_splice(&head, n->list.prev); |
518 | ||
519 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
520 | hash(parent, mnt->mnt_mountpoint)); | |
521 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 522 | touch_mnt_namespace(n); |
1da177e4 LT |
523 | } |
524 | ||
525 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
526 | { | |
527 | struct list_head *next = p->mnt_mounts.next; | |
528 | if (next == &p->mnt_mounts) { | |
529 | while (1) { | |
530 | if (p == root) | |
531 | return NULL; | |
532 | next = p->mnt_child.next; | |
533 | if (next != &p->mnt_parent->mnt_mounts) | |
534 | break; | |
535 | p = p->mnt_parent; | |
536 | } | |
537 | } | |
538 | return list_entry(next, struct vfsmount, mnt_child); | |
539 | } | |
540 | ||
9676f0c6 RP |
541 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
542 | { | |
543 | struct list_head *prev = p->mnt_mounts.prev; | |
544 | while (prev != &p->mnt_mounts) { | |
545 | p = list_entry(prev, struct vfsmount, mnt_child); | |
546 | prev = p->mnt_mounts.prev; | |
547 | } | |
548 | return p; | |
549 | } | |
550 | ||
36341f64 RP |
551 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
552 | int flag) | |
1da177e4 LT |
553 | { |
554 | struct super_block *sb = old->mnt_sb; | |
555 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
556 | ||
557 | if (mnt) { | |
719f5d7f MS |
558 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
559 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
560 | else | |
561 | mnt->mnt_group_id = old->mnt_group_id; | |
562 | ||
563 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
564 | int err = mnt_alloc_group_id(mnt); | |
565 | if (err) | |
566 | goto out_free; | |
567 | } | |
568 | ||
1da177e4 LT |
569 | mnt->mnt_flags = old->mnt_flags; |
570 | atomic_inc(&sb->s_active); | |
571 | mnt->mnt_sb = sb; | |
572 | mnt->mnt_root = dget(root); | |
573 | mnt->mnt_mountpoint = mnt->mnt_root; | |
574 | mnt->mnt_parent = mnt; | |
b90fa9ae | 575 | |
5afe0022 RP |
576 | if (flag & CL_SLAVE) { |
577 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
578 | mnt->mnt_master = old; | |
579 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 580 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
581 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
582 | list_add(&mnt->mnt_share, &old->mnt_share); | |
583 | if (IS_MNT_SLAVE(old)) | |
584 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
585 | mnt->mnt_master = old->mnt_master; | |
586 | } | |
b90fa9ae RP |
587 | if (flag & CL_MAKE_SHARED) |
588 | set_mnt_shared(mnt); | |
1da177e4 LT |
589 | |
590 | /* stick the duplicate mount on the same expiry list | |
591 | * as the original if that was on one */ | |
36341f64 | 592 | if (flag & CL_EXPIRE) { |
36341f64 RP |
593 | if (!list_empty(&old->mnt_expire)) |
594 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 595 | } |
1da177e4 LT |
596 | } |
597 | return mnt; | |
719f5d7f MS |
598 | |
599 | out_free: | |
600 | free_vfsmnt(mnt); | |
601 | return NULL; | |
1da177e4 LT |
602 | } |
603 | ||
7b7b1ace | 604 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 | 605 | { |
3d733633 | 606 | int cpu; |
1da177e4 | 607 | struct super_block *sb = mnt->mnt_sb; |
3d733633 DH |
608 | /* |
609 | * We don't have to hold all of the locks at the | |
610 | * same time here because we know that we're the | |
611 | * last reference to mnt and that no new writers | |
612 | * can come in. | |
613 | */ | |
614 | for_each_possible_cpu(cpu) { | |
615 | struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); | |
616 | if (cpu_writer->mnt != mnt) | |
617 | continue; | |
618 | spin_lock(&cpu_writer->lock); | |
619 | atomic_add(cpu_writer->count, &mnt->__mnt_writers); | |
620 | cpu_writer->count = 0; | |
621 | /* | |
622 | * Might as well do this so that no one | |
623 | * ever sees the pointer and expects | |
624 | * it to be valid. | |
625 | */ | |
626 | cpu_writer->mnt = NULL; | |
627 | spin_unlock(&cpu_writer->lock); | |
628 | } | |
629 | /* | |
630 | * This probably indicates that somebody messed | |
631 | * up a mnt_want/drop_write() pair. If this | |
632 | * happens, the filesystem was probably unable | |
633 | * to make r/w->r/o transitions. | |
634 | */ | |
635 | WARN_ON(atomic_read(&mnt->__mnt_writers)); | |
1da177e4 LT |
636 | dput(mnt->mnt_root); |
637 | free_vfsmnt(mnt); | |
638 | deactivate_super(sb); | |
639 | } | |
640 | ||
7b7b1ace AV |
641 | void mntput_no_expire(struct vfsmount *mnt) |
642 | { | |
643 | repeat: | |
644 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
645 | if (likely(!mnt->mnt_pinned)) { | |
646 | spin_unlock(&vfsmount_lock); | |
647 | __mntput(mnt); | |
648 | return; | |
649 | } | |
650 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
651 | mnt->mnt_pinned = 0; | |
652 | spin_unlock(&vfsmount_lock); | |
653 | acct_auto_close_mnt(mnt); | |
654 | security_sb_umount_close(mnt); | |
655 | goto repeat; | |
656 | } | |
657 | } | |
658 | ||
659 | EXPORT_SYMBOL(mntput_no_expire); | |
660 | ||
661 | void mnt_pin(struct vfsmount *mnt) | |
662 | { | |
663 | spin_lock(&vfsmount_lock); | |
664 | mnt->mnt_pinned++; | |
665 | spin_unlock(&vfsmount_lock); | |
666 | } | |
667 | ||
668 | EXPORT_SYMBOL(mnt_pin); | |
669 | ||
670 | void mnt_unpin(struct vfsmount *mnt) | |
671 | { | |
672 | spin_lock(&vfsmount_lock); | |
673 | if (mnt->mnt_pinned) { | |
674 | atomic_inc(&mnt->mnt_count); | |
675 | mnt->mnt_pinned--; | |
676 | } | |
677 | spin_unlock(&vfsmount_lock); | |
678 | } | |
679 | ||
680 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 681 | |
b3b304a2 MS |
682 | static inline void mangle(struct seq_file *m, const char *s) |
683 | { | |
684 | seq_escape(m, s, " \t\n\\"); | |
685 | } | |
686 | ||
687 | /* | |
688 | * Simple .show_options callback for filesystems which don't want to | |
689 | * implement more complex mount option showing. | |
690 | * | |
691 | * See also save_mount_options(). | |
692 | */ | |
693 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
694 | { | |
695 | const char *options = mnt->mnt_sb->s_options; | |
696 | ||
697 | if (options != NULL && options[0]) { | |
698 | seq_putc(m, ','); | |
699 | mangle(m, options); | |
700 | } | |
701 | ||
702 | return 0; | |
703 | } | |
704 | EXPORT_SYMBOL(generic_show_options); | |
705 | ||
706 | /* | |
707 | * If filesystem uses generic_show_options(), this function should be | |
708 | * called from the fill_super() callback. | |
709 | * | |
710 | * The .remount_fs callback usually needs to be handled in a special | |
711 | * way, to make sure, that previous options are not overwritten if the | |
712 | * remount fails. | |
713 | * | |
714 | * Also note, that if the filesystem's .remount_fs function doesn't | |
715 | * reset all options to their default value, but changes only newly | |
716 | * given options, then the displayed options will not reflect reality | |
717 | * any more. | |
718 | */ | |
719 | void save_mount_options(struct super_block *sb, char *options) | |
720 | { | |
721 | kfree(sb->s_options); | |
722 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
723 | } | |
724 | EXPORT_SYMBOL(save_mount_options); | |
725 | ||
a1a2c409 | 726 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
727 | /* iterator */ |
728 | static void *m_start(struct seq_file *m, loff_t *pos) | |
729 | { | |
a1a2c409 | 730 | struct proc_mounts *p = m->private; |
1da177e4 | 731 | |
390c6843 | 732 | down_read(&namespace_sem); |
a1a2c409 | 733 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
734 | } |
735 | ||
736 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
737 | { | |
a1a2c409 | 738 | struct proc_mounts *p = m->private; |
b0765fb8 | 739 | |
a1a2c409 | 740 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
741 | } |
742 | ||
743 | static void m_stop(struct seq_file *m, void *v) | |
744 | { | |
390c6843 | 745 | up_read(&namespace_sem); |
1da177e4 LT |
746 | } |
747 | ||
2d4d4864 RP |
748 | struct proc_fs_info { |
749 | int flag; | |
750 | const char *str; | |
751 | }; | |
752 | ||
753 | static void show_sb_opts(struct seq_file *m, struct super_block *sb) | |
1da177e4 | 754 | { |
2d4d4864 | 755 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
756 | { MS_SYNCHRONOUS, ",sync" }, |
757 | { MS_DIRSYNC, ",dirsync" }, | |
758 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
759 | { 0, NULL } |
760 | }; | |
2d4d4864 RP |
761 | const struct proc_fs_info *fs_infop; |
762 | ||
763 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
764 | if (sb->s_flags & fs_infop->flag) | |
765 | seq_puts(m, fs_infop->str); | |
766 | } | |
767 | } | |
768 | ||
769 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
770 | { | |
771 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
772 | { MNT_NOSUID, ",nosuid" }, |
773 | { MNT_NODEV, ",nodev" }, | |
774 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
775 | { MNT_NOATIME, ",noatime" }, |
776 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 777 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
778 | { 0, NULL } |
779 | }; | |
2d4d4864 RP |
780 | const struct proc_fs_info *fs_infop; |
781 | ||
782 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
783 | if (mnt->mnt_flags & fs_infop->flag) | |
784 | seq_puts(m, fs_infop->str); | |
785 | } | |
786 | } | |
787 | ||
788 | static void show_type(struct seq_file *m, struct super_block *sb) | |
789 | { | |
790 | mangle(m, sb->s_type->name); | |
791 | if (sb->s_subtype && sb->s_subtype[0]) { | |
792 | seq_putc(m, '.'); | |
793 | mangle(m, sb->s_subtype); | |
794 | } | |
795 | } | |
796 | ||
797 | static int show_vfsmnt(struct seq_file *m, void *v) | |
798 | { | |
799 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
800 | int err = 0; | |
c32c2f63 | 801 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
802 | |
803 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
804 | seq_putc(m, ' '); | |
c32c2f63 | 805 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 806 | seq_putc(m, ' '); |
2d4d4864 | 807 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 808 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2d4d4864 RP |
809 | show_sb_opts(m, mnt->mnt_sb); |
810 | show_mnt_opts(m, mnt); | |
1da177e4 LT |
811 | if (mnt->mnt_sb->s_op->show_options) |
812 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
813 | seq_puts(m, " 0 0\n"); | |
814 | return err; | |
815 | } | |
816 | ||
a1a2c409 | 817 | const struct seq_operations mounts_op = { |
1da177e4 LT |
818 | .start = m_start, |
819 | .next = m_next, | |
820 | .stop = m_stop, | |
821 | .show = show_vfsmnt | |
822 | }; | |
823 | ||
2d4d4864 RP |
824 | static int show_mountinfo(struct seq_file *m, void *v) |
825 | { | |
826 | struct proc_mounts *p = m->private; | |
827 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
828 | struct super_block *sb = mnt->mnt_sb; | |
829 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
830 | struct path root = p->root; | |
831 | int err = 0; | |
832 | ||
833 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
834 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
835 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
836 | seq_putc(m, ' '); | |
837 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
838 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
839 | /* | |
840 | * Mountpoint is outside root, discard that one. Ugly, | |
841 | * but less so than trying to do that in iterator in a | |
842 | * race-free way (due to renames). | |
843 | */ | |
844 | return SEQ_SKIP; | |
845 | } | |
846 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
847 | show_mnt_opts(m, mnt); | |
848 | ||
849 | /* Tagged fields ("foo:X" or "bar") */ | |
850 | if (IS_MNT_SHARED(mnt)) | |
851 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
852 | if (IS_MNT_SLAVE(mnt)) { |
853 | int master = mnt->mnt_master->mnt_group_id; | |
854 | int dom = get_dominating_id(mnt, &p->root); | |
855 | seq_printf(m, " master:%i", master); | |
856 | if (dom && dom != master) | |
857 | seq_printf(m, " propagate_from:%i", dom); | |
858 | } | |
2d4d4864 RP |
859 | if (IS_MNT_UNBINDABLE(mnt)) |
860 | seq_puts(m, " unbindable"); | |
861 | ||
862 | /* Filesystem specific data */ | |
863 | seq_puts(m, " - "); | |
864 | show_type(m, sb); | |
865 | seq_putc(m, ' '); | |
866 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
867 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
868 | show_sb_opts(m, sb); | |
869 | if (sb->s_op->show_options) | |
870 | err = sb->s_op->show_options(m, mnt); | |
871 | seq_putc(m, '\n'); | |
872 | return err; | |
873 | } | |
874 | ||
875 | const struct seq_operations mountinfo_op = { | |
876 | .start = m_start, | |
877 | .next = m_next, | |
878 | .stop = m_stop, | |
879 | .show = show_mountinfo, | |
880 | }; | |
881 | ||
b4629fe2 CL |
882 | static int show_vfsstat(struct seq_file *m, void *v) |
883 | { | |
b0765fb8 | 884 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 885 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
886 | int err = 0; |
887 | ||
888 | /* device */ | |
889 | if (mnt->mnt_devname) { | |
890 | seq_puts(m, "device "); | |
891 | mangle(m, mnt->mnt_devname); | |
892 | } else | |
893 | seq_puts(m, "no device"); | |
894 | ||
895 | /* mount point */ | |
896 | seq_puts(m, " mounted on "); | |
c32c2f63 | 897 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
898 | seq_putc(m, ' '); |
899 | ||
900 | /* file system type */ | |
901 | seq_puts(m, "with fstype "); | |
2d4d4864 | 902 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
903 | |
904 | /* optional statistics */ | |
905 | if (mnt->mnt_sb->s_op->show_stats) { | |
906 | seq_putc(m, ' '); | |
907 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
908 | } | |
909 | ||
910 | seq_putc(m, '\n'); | |
911 | return err; | |
912 | } | |
913 | ||
a1a2c409 | 914 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
915 | .start = m_start, |
916 | .next = m_next, | |
917 | .stop = m_stop, | |
918 | .show = show_vfsstat, | |
919 | }; | |
a1a2c409 | 920 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 921 | |
1da177e4 LT |
922 | /** |
923 | * may_umount_tree - check if a mount tree is busy | |
924 | * @mnt: root of mount tree | |
925 | * | |
926 | * This is called to check if a tree of mounts has any | |
927 | * open files, pwds, chroots or sub mounts that are | |
928 | * busy. | |
929 | */ | |
930 | int may_umount_tree(struct vfsmount *mnt) | |
931 | { | |
36341f64 RP |
932 | int actual_refs = 0; |
933 | int minimum_refs = 0; | |
934 | struct vfsmount *p; | |
1da177e4 LT |
935 | |
936 | spin_lock(&vfsmount_lock); | |
36341f64 | 937 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
938 | actual_refs += atomic_read(&p->mnt_count); |
939 | minimum_refs += 2; | |
1da177e4 LT |
940 | } |
941 | spin_unlock(&vfsmount_lock); | |
942 | ||
943 | if (actual_refs > minimum_refs) | |
e3474a8e | 944 | return 0; |
1da177e4 | 945 | |
e3474a8e | 946 | return 1; |
1da177e4 LT |
947 | } |
948 | ||
949 | EXPORT_SYMBOL(may_umount_tree); | |
950 | ||
951 | /** | |
952 | * may_umount - check if a mount point is busy | |
953 | * @mnt: root of mount | |
954 | * | |
955 | * This is called to check if a mount point has any | |
956 | * open files, pwds, chroots or sub mounts. If the | |
957 | * mount has sub mounts this will return busy | |
958 | * regardless of whether the sub mounts are busy. | |
959 | * | |
960 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
961 | * give false negatives. The main reason why it's here is that we need | |
962 | * a non-destructive way to look for easily umountable filesystems. | |
963 | */ | |
964 | int may_umount(struct vfsmount *mnt) | |
965 | { | |
e3474a8e | 966 | int ret = 1; |
a05964f3 RP |
967 | spin_lock(&vfsmount_lock); |
968 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 969 | ret = 0; |
a05964f3 RP |
970 | spin_unlock(&vfsmount_lock); |
971 | return ret; | |
1da177e4 LT |
972 | } |
973 | ||
974 | EXPORT_SYMBOL(may_umount); | |
975 | ||
b90fa9ae | 976 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
977 | { |
978 | struct vfsmount *mnt; | |
bf066c7d | 979 | while (!list_empty(head)) { |
b5e61818 | 980 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
981 | list_del_init(&mnt->mnt_hash); |
982 | if (mnt->mnt_parent != mnt) { | |
983 | struct dentry *dentry; | |
984 | struct vfsmount *m; | |
985 | spin_lock(&vfsmount_lock); | |
986 | dentry = mnt->mnt_mountpoint; | |
987 | m = mnt->mnt_parent; | |
988 | mnt->mnt_mountpoint = mnt->mnt_root; | |
989 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 990 | m->mnt_ghosts--; |
70fbcdf4 RP |
991 | spin_unlock(&vfsmount_lock); |
992 | dput(dentry); | |
993 | mntput(m); | |
994 | } | |
995 | mntput(mnt); | |
996 | } | |
997 | } | |
998 | ||
a05964f3 | 999 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
1000 | { |
1001 | struct vfsmount *p; | |
1da177e4 | 1002 | |
1bfba4e8 AM |
1003 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1004 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 1005 | |
a05964f3 RP |
1006 | if (propagate) |
1007 | propagate_umount(kill); | |
1008 | ||
70fbcdf4 RP |
1009 | list_for_each_entry(p, kill, mnt_hash) { |
1010 | list_del_init(&p->mnt_expire); | |
1011 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
1012 | __touch_mnt_namespace(p->mnt_ns); |
1013 | p->mnt_ns = NULL; | |
70fbcdf4 | 1014 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
1015 | if (p->mnt_parent != p) { |
1016 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 1017 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 1018 | } |
a05964f3 | 1019 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1020 | } |
1021 | } | |
1022 | ||
c35038be AV |
1023 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
1024 | ||
1da177e4 LT |
1025 | static int do_umount(struct vfsmount *mnt, int flags) |
1026 | { | |
b58fed8b | 1027 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 1028 | int retval; |
70fbcdf4 | 1029 | LIST_HEAD(umount_list); |
1da177e4 LT |
1030 | |
1031 | retval = security_sb_umount(mnt, flags); | |
1032 | if (retval) | |
1033 | return retval; | |
1034 | ||
1035 | /* | |
1036 | * Allow userspace to request a mountpoint be expired rather than | |
1037 | * unmounting unconditionally. Unmount only happens if: | |
1038 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1039 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1040 | */ | |
1041 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 1042 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
1043 | flags & (MNT_FORCE | MNT_DETACH)) |
1044 | return -EINVAL; | |
1045 | ||
1046 | if (atomic_read(&mnt->mnt_count) != 2) | |
1047 | return -EBUSY; | |
1048 | ||
1049 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1050 | return -EAGAIN; | |
1051 | } | |
1052 | ||
1053 | /* | |
1054 | * If we may have to abort operations to get out of this | |
1055 | * mount, and they will themselves hold resources we must | |
1056 | * allow the fs to do things. In the Unix tradition of | |
1057 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1058 | * might fail to complete on the first run through as other tasks | |
1059 | * must return, and the like. Thats for the mount program to worry | |
1060 | * about for the moment. | |
1061 | */ | |
1062 | ||
42faad99 AV |
1063 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
1064 | lock_kernel(); | |
1065 | sb->s_op->umount_begin(sb); | |
1066 | unlock_kernel(); | |
1067 | } | |
1da177e4 LT |
1068 | |
1069 | /* | |
1070 | * No sense to grab the lock for this test, but test itself looks | |
1071 | * somewhat bogus. Suggestions for better replacement? | |
1072 | * Ho-hum... In principle, we might treat that as umount + switch | |
1073 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1074 | * Actually it makes sense, especially if rootfs would contain a | |
1075 | * /reboot - static binary that would close all descriptors and | |
1076 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1077 | */ | |
6ac08c39 | 1078 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1079 | /* |
1080 | * Special case for "unmounting" root ... | |
1081 | * we just try to remount it readonly. | |
1082 | */ | |
1083 | down_write(&sb->s_umount); | |
1084 | if (!(sb->s_flags & MS_RDONLY)) { | |
1085 | lock_kernel(); | |
1da177e4 LT |
1086 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1087 | unlock_kernel(); | |
1088 | } | |
1089 | up_write(&sb->s_umount); | |
1090 | return retval; | |
1091 | } | |
1092 | ||
390c6843 | 1093 | down_write(&namespace_sem); |
1da177e4 | 1094 | spin_lock(&vfsmount_lock); |
5addc5dd | 1095 | event++; |
1da177e4 | 1096 | |
c35038be AV |
1097 | if (!(flags & MNT_DETACH)) |
1098 | shrink_submounts(mnt, &umount_list); | |
1099 | ||
1da177e4 | 1100 | retval = -EBUSY; |
a05964f3 | 1101 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1102 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1103 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1104 | retval = 0; |
1105 | } | |
1106 | spin_unlock(&vfsmount_lock); | |
1107 | if (retval) | |
1108 | security_sb_umount_busy(mnt); | |
390c6843 | 1109 | up_write(&namespace_sem); |
70fbcdf4 | 1110 | release_mounts(&umount_list); |
1da177e4 LT |
1111 | return retval; |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Now umount can handle mount points as well as block devices. | |
1116 | * This is important for filesystems which use unnamed block devices. | |
1117 | * | |
1118 | * We now support a flag for forced unmount like the other 'big iron' | |
1119 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1120 | */ | |
1121 | ||
1122 | asmlinkage long sys_umount(char __user * name, int flags) | |
1123 | { | |
1124 | struct nameidata nd; | |
1125 | int retval; | |
1126 | ||
1127 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
1128 | if (retval) | |
1129 | goto out; | |
1130 | retval = -EINVAL; | |
4ac91378 | 1131 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 1132 | goto dput_and_out; |
4ac91378 | 1133 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
1134 | goto dput_and_out; |
1135 | ||
1136 | retval = -EPERM; | |
1137 | if (!capable(CAP_SYS_ADMIN)) | |
1138 | goto dput_and_out; | |
1139 | ||
4ac91378 | 1140 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 1141 | dput_and_out: |
429731b1 | 1142 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
1143 | dput(nd.path.dentry); |
1144 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
1145 | out: |
1146 | return retval; | |
1147 | } | |
1148 | ||
1149 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1150 | ||
1151 | /* | |
b58fed8b | 1152 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1153 | */ |
1da177e4 LT |
1154 | asmlinkage long sys_oldumount(char __user * name) |
1155 | { | |
b58fed8b | 1156 | return sys_umount(name, 0); |
1da177e4 LT |
1157 | } |
1158 | ||
1159 | #endif | |
1160 | ||
1161 | static int mount_is_safe(struct nameidata *nd) | |
1162 | { | |
1163 | if (capable(CAP_SYS_ADMIN)) | |
1164 | return 0; | |
1165 | return -EPERM; | |
1166 | #ifdef notyet | |
4ac91378 | 1167 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 1168 | return -EPERM; |
4ac91378 JB |
1169 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
1170 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
1171 | return -EPERM; |
1172 | } | |
e4543edd | 1173 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
1174 | return -EPERM; |
1175 | return 0; | |
1176 | #endif | |
1177 | } | |
1178 | ||
b58fed8b | 1179 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
1180 | { |
1181 | while (1) { | |
1182 | if (d == dentry) | |
1183 | return 1; | |
1184 | if (d == NULL || d == d->d_parent) | |
1185 | return 0; | |
1186 | d = d->d_parent; | |
1187 | } | |
1188 | } | |
1189 | ||
b90fa9ae | 1190 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1191 | int flag) |
1da177e4 LT |
1192 | { |
1193 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1194 | struct path path; |
1da177e4 | 1195 | |
9676f0c6 RP |
1196 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1197 | return NULL; | |
1198 | ||
36341f64 | 1199 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1200 | if (!q) |
1201 | goto Enomem; | |
1202 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1203 | ||
1204 | p = mnt; | |
fdadd65f | 1205 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
1206 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
1207 | continue; | |
1208 | ||
1209 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1210 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1211 | s = skip_mnt_tree(s); | |
1212 | continue; | |
1213 | } | |
1da177e4 LT |
1214 | while (p != s->mnt_parent) { |
1215 | p = p->mnt_parent; | |
1216 | q = q->mnt_parent; | |
1217 | } | |
1218 | p = s; | |
1a390689 AV |
1219 | path.mnt = q; |
1220 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1221 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1222 | if (!q) |
1223 | goto Enomem; | |
1224 | spin_lock(&vfsmount_lock); | |
1225 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1226 | attach_mnt(q, &path); |
1da177e4 LT |
1227 | spin_unlock(&vfsmount_lock); |
1228 | } | |
1229 | } | |
1230 | return res; | |
b58fed8b | 1231 | Enomem: |
1da177e4 | 1232 | if (res) { |
70fbcdf4 | 1233 | LIST_HEAD(umount_list); |
1da177e4 | 1234 | spin_lock(&vfsmount_lock); |
a05964f3 | 1235 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1236 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1237 | release_mounts(&umount_list); |
1da177e4 LT |
1238 | } |
1239 | return NULL; | |
1240 | } | |
1241 | ||
8aec0809 AV |
1242 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
1243 | { | |
1244 | struct vfsmount *tree; | |
1a60a280 | 1245 | down_write(&namespace_sem); |
8aec0809 | 1246 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1247 | up_write(&namespace_sem); |
8aec0809 AV |
1248 | return tree; |
1249 | } | |
1250 | ||
1251 | void drop_collected_mounts(struct vfsmount *mnt) | |
1252 | { | |
1253 | LIST_HEAD(umount_list); | |
1a60a280 | 1254 | down_write(&namespace_sem); |
8aec0809 AV |
1255 | spin_lock(&vfsmount_lock); |
1256 | umount_tree(mnt, 0, &umount_list); | |
1257 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1258 | up_write(&namespace_sem); |
8aec0809 AV |
1259 | release_mounts(&umount_list); |
1260 | } | |
1261 | ||
719f5d7f MS |
1262 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1263 | { | |
1264 | struct vfsmount *p; | |
1265 | ||
1266 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1267 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1268 | mnt_release_group_id(p); | |
1269 | } | |
1270 | } | |
1271 | ||
1272 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1273 | { | |
1274 | struct vfsmount *p; | |
1275 | ||
1276 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1277 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1278 | int err = mnt_alloc_group_id(p); | |
1279 | if (err) { | |
1280 | cleanup_group_ids(mnt, p); | |
1281 | return err; | |
1282 | } | |
1283 | } | |
1284 | } | |
1285 | ||
1286 | return 0; | |
1287 | } | |
1288 | ||
b90fa9ae RP |
1289 | /* |
1290 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1291 | * @nd : place the mount tree @source_mnt is attached |
1292 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1293 | * store the parent mount and mountpoint dentry. | |
1294 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1295 | * |
1296 | * NOTE: in the table below explains the semantics when a source mount | |
1297 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1298 | * --------------------------------------------------------------------------- |
1299 | * | BIND MOUNT OPERATION | | |
1300 | * |************************************************************************** | |
1301 | * | source-->| shared | private | slave | unbindable | | |
1302 | * | dest | | | | | | |
1303 | * | | | | | | | | |
1304 | * | v | | | | | | |
1305 | * |************************************************************************** | |
1306 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1307 | * | | | | | | | |
1308 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1309 | * *************************************************************************** | |
b90fa9ae RP |
1310 | * A bind operation clones the source mount and mounts the clone on the |
1311 | * destination mount. | |
1312 | * | |
1313 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1314 | * tree of the destination mount and the cloned mount is added to | |
1315 | * the peer group of the source mount. | |
1316 | * (+) the cloned mount is created under the destination mount and is marked | |
1317 | * as shared. The cloned mount is added to the peer group of the source | |
1318 | * mount. | |
5afe0022 RP |
1319 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1320 | * of the destination mount and the cloned mount is made slave | |
1321 | * of the same master as that of the source mount. The cloned mount | |
1322 | * is marked as 'shared and slave'. | |
1323 | * (*) the cloned mount is made a slave of the same master as that of the | |
1324 | * source mount. | |
1325 | * | |
9676f0c6 RP |
1326 | * --------------------------------------------------------------------------- |
1327 | * | MOVE MOUNT OPERATION | | |
1328 | * |************************************************************************** | |
1329 | * | source-->| shared | private | slave | unbindable | | |
1330 | * | dest | | | | | | |
1331 | * | | | | | | | | |
1332 | * | v | | | | | | |
1333 | * |************************************************************************** | |
1334 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1335 | * | | | | | | | |
1336 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1337 | * *************************************************************************** | |
5afe0022 RP |
1338 | * |
1339 | * (+) the mount is moved to the destination. And is then propagated to | |
1340 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1341 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1342 | * (+++) the mount is moved to the destination and is then propagated to |
1343 | * all the mounts belonging to the destination mount's propagation tree. | |
1344 | * the mount is marked as 'shared and slave'. | |
1345 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1346 | * |
1347 | * if the source mount is a tree, the operations explained above is | |
1348 | * applied to each mount in the tree. | |
1349 | * Must be called without spinlocks held, since this function can sleep | |
1350 | * in allocations. | |
1351 | */ | |
1352 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1353 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1354 | { |
1355 | LIST_HEAD(tree_list); | |
1a390689 AV |
1356 | struct vfsmount *dest_mnt = path->mnt; |
1357 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1358 | struct vfsmount *child, *p; |
719f5d7f | 1359 | int err; |
b90fa9ae | 1360 | |
719f5d7f MS |
1361 | if (IS_MNT_SHARED(dest_mnt)) { |
1362 | err = invent_group_ids(source_mnt, true); | |
1363 | if (err) | |
1364 | goto out; | |
1365 | } | |
1366 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1367 | if (err) | |
1368 | goto out_cleanup_ids; | |
b90fa9ae RP |
1369 | |
1370 | if (IS_MNT_SHARED(dest_mnt)) { | |
1371 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1372 | set_mnt_shared(p); | |
1373 | } | |
1374 | ||
1375 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1376 | if (parent_path) { |
1377 | detach_mnt(source_mnt, parent_path); | |
1378 | attach_mnt(source_mnt, path); | |
6b3286ed | 1379 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
1380 | } else { |
1381 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1382 | commit_tree(source_mnt); | |
1383 | } | |
b90fa9ae RP |
1384 | |
1385 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1386 | list_del_init(&child->mnt_hash); | |
1387 | commit_tree(child); | |
1388 | } | |
1389 | spin_unlock(&vfsmount_lock); | |
1390 | return 0; | |
719f5d7f MS |
1391 | |
1392 | out_cleanup_ids: | |
1393 | if (IS_MNT_SHARED(dest_mnt)) | |
1394 | cleanup_group_ids(source_mnt, NULL); | |
1395 | out: | |
1396 | return err; | |
b90fa9ae RP |
1397 | } |
1398 | ||
8c3ee42e | 1399 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1400 | { |
1401 | int err; | |
1402 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1403 | return -EINVAL; | |
1404 | ||
8c3ee42e | 1405 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1406 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1407 | return -ENOTDIR; | |
1408 | ||
1409 | err = -ENOENT; | |
8c3ee42e AV |
1410 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1411 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1412 | goto out_unlock; |
1413 | ||
8c3ee42e | 1414 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1415 | if (err) |
1416 | goto out_unlock; | |
1417 | ||
1418 | err = -ENOENT; | |
8c3ee42e AV |
1419 | if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) |
1420 | err = attach_recursive_mnt(mnt, path, NULL); | |
1da177e4 | 1421 | out_unlock: |
8c3ee42e | 1422 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1423 | if (!err) |
8c3ee42e | 1424 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1425 | return err; |
1426 | } | |
1427 | ||
07b20889 RP |
1428 | /* |
1429 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 1430 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 1431 | */ |
2dafe1c4 | 1432 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 1433 | { |
4ac91378 | 1434 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
1435 | int recurse = flag & MS_REC; |
1436 | int type = flag & ~MS_REC; | |
719f5d7f | 1437 | int err = 0; |
07b20889 | 1438 | |
ee6f9582 MS |
1439 | if (!capable(CAP_SYS_ADMIN)) |
1440 | return -EPERM; | |
1441 | ||
4ac91378 | 1442 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
1443 | return -EINVAL; |
1444 | ||
1445 | down_write(&namespace_sem); | |
719f5d7f MS |
1446 | if (type == MS_SHARED) { |
1447 | err = invent_group_ids(mnt, recurse); | |
1448 | if (err) | |
1449 | goto out_unlock; | |
1450 | } | |
1451 | ||
07b20889 RP |
1452 | spin_lock(&vfsmount_lock); |
1453 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1454 | change_mnt_propagation(m, type); | |
1455 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1456 | |
1457 | out_unlock: | |
07b20889 | 1458 | up_write(&namespace_sem); |
719f5d7f | 1459 | return err; |
07b20889 RP |
1460 | } |
1461 | ||
1da177e4 LT |
1462 | /* |
1463 | * do loopback mount. | |
2dafe1c4 | 1464 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1465 | */ |
2dafe1c4 ES |
1466 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
1467 | int recurse) | |
1da177e4 LT |
1468 | { |
1469 | struct nameidata old_nd; | |
1470 | struct vfsmount *mnt = NULL; | |
1471 | int err = mount_is_safe(nd); | |
1472 | if (err) | |
1473 | return err; | |
1474 | if (!old_name || !*old_name) | |
1475 | return -EINVAL; | |
1476 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1477 | if (err) | |
1478 | return err; | |
1479 | ||
390c6843 | 1480 | down_write(&namespace_sem); |
1da177e4 | 1481 | err = -EINVAL; |
4ac91378 JB |
1482 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
1483 | goto out; | |
9676f0c6 | 1484 | |
4ac91378 | 1485 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 1486 | goto out; |
1da177e4 | 1487 | |
ccd48bc7 AV |
1488 | err = -ENOMEM; |
1489 | if (recurse) | |
4ac91378 | 1490 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 1491 | else |
4ac91378 | 1492 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1493 | |
1494 | if (!mnt) | |
1495 | goto out; | |
1496 | ||
8c3ee42e | 1497 | err = graft_tree(mnt, &nd->path); |
ccd48bc7 | 1498 | if (err) { |
70fbcdf4 | 1499 | LIST_HEAD(umount_list); |
1da177e4 | 1500 | spin_lock(&vfsmount_lock); |
a05964f3 | 1501 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1502 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1503 | release_mounts(&umount_list); |
5b83d2c5 | 1504 | } |
1da177e4 | 1505 | |
ccd48bc7 | 1506 | out: |
390c6843 | 1507 | up_write(&namespace_sem); |
1d957f9b | 1508 | path_put(&old_nd.path); |
1da177e4 LT |
1509 | return err; |
1510 | } | |
1511 | ||
2e4b7fcd DH |
1512 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1513 | { | |
1514 | int error = 0; | |
1515 | int readonly_request = 0; | |
1516 | ||
1517 | if (ms_flags & MS_RDONLY) | |
1518 | readonly_request = 1; | |
1519 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1520 | return 0; | |
1521 | ||
1522 | if (readonly_request) | |
1523 | error = mnt_make_readonly(mnt); | |
1524 | else | |
1525 | __mnt_unmake_readonly(mnt); | |
1526 | return error; | |
1527 | } | |
1528 | ||
1da177e4 LT |
1529 | /* |
1530 | * change filesystem flags. dir should be a physical root of filesystem. | |
1531 | * If you've mounted a non-root directory somewhere and want to do remount | |
1532 | * on it - tough luck. | |
2dafe1c4 | 1533 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1534 | */ |
2dafe1c4 | 1535 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1536 | void *data) |
1537 | { | |
1538 | int err; | |
4ac91378 | 1539 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1540 | |
1541 | if (!capable(CAP_SYS_ADMIN)) | |
1542 | return -EPERM; | |
1543 | ||
4ac91378 | 1544 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1545 | return -EINVAL; |
1546 | ||
4ac91378 | 1547 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1548 | return -EINVAL; |
1549 | ||
1550 | down_write(&sb->s_umount); | |
2e4b7fcd DH |
1551 | if (flags & MS_BIND) |
1552 | err = change_mount_flags(nd->path.mnt, flags); | |
1553 | else | |
1554 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1555 | if (!err) |
4ac91378 | 1556 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1557 | up_write(&sb->s_umount); |
1558 | if (!err) | |
4ac91378 | 1559 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1560 | return err; |
1561 | } | |
1562 | ||
9676f0c6 RP |
1563 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1564 | { | |
1565 | struct vfsmount *p; | |
1566 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1567 | if (IS_MNT_UNBINDABLE(p)) | |
1568 | return 1; | |
1569 | } | |
1570 | return 0; | |
1571 | } | |
1572 | ||
2dafe1c4 ES |
1573 | /* |
1574 | * noinline this do_mount helper to save do_mount stack space. | |
1575 | */ | |
1576 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1577 | { |
1a390689 AV |
1578 | struct nameidata old_nd; |
1579 | struct path parent_path; | |
1da177e4 LT |
1580 | struct vfsmount *p; |
1581 | int err = 0; | |
1582 | if (!capable(CAP_SYS_ADMIN)) | |
1583 | return -EPERM; | |
1584 | if (!old_name || !*old_name) | |
1585 | return -EINVAL; | |
1586 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1587 | if (err) | |
1588 | return err; | |
1589 | ||
390c6843 | 1590 | down_write(&namespace_sem); |
4ac91378 JB |
1591 | while (d_mountpoint(nd->path.dentry) && |
1592 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1593 | ; |
1594 | err = -EINVAL; | |
4ac91378 | 1595 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1596 | goto out; |
1597 | ||
1598 | err = -ENOENT; | |
4ac91378 JB |
1599 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1600 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1601 | goto out1; |
1602 | ||
4ac91378 | 1603 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1604 | goto out1; |
1da177e4 LT |
1605 | |
1606 | err = -EINVAL; | |
4ac91378 | 1607 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1608 | goto out1; |
1da177e4 | 1609 | |
4ac91378 | 1610 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1611 | goto out1; |
1da177e4 | 1612 | |
4ac91378 JB |
1613 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1614 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1615 | goto out1; |
1616 | /* | |
1617 | * Don't move a mount residing in a shared parent. | |
1618 | */ | |
4ac91378 JB |
1619 | if (old_nd.path.mnt->mnt_parent && |
1620 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1621 | goto out1; |
9676f0c6 RP |
1622 | /* |
1623 | * Don't move a mount tree containing unbindable mounts to a destination | |
1624 | * mount which is shared. | |
1625 | */ | |
4ac91378 JB |
1626 | if (IS_MNT_SHARED(nd->path.mnt) && |
1627 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1628 | goto out1; |
1da177e4 | 1629 | err = -ELOOP; |
4ac91378 JB |
1630 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1631 | if (p == old_nd.path.mnt) | |
21444403 | 1632 | goto out1; |
1da177e4 | 1633 | |
1a390689 | 1634 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1635 | if (err) |
21444403 | 1636 | goto out1; |
1da177e4 LT |
1637 | |
1638 | /* if the mount is moved, it should no longer be expire | |
1639 | * automatically */ | |
4ac91378 | 1640 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 | 1641 | out1: |
4ac91378 | 1642 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1643 | out: |
390c6843 | 1644 | up_write(&namespace_sem); |
1da177e4 | 1645 | if (!err) |
1a390689 | 1646 | path_put(&parent_path); |
1d957f9b | 1647 | path_put(&old_nd.path); |
1da177e4 LT |
1648 | return err; |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * create a new mount for userspace and request it to be added into the | |
1653 | * namespace's tree | |
2dafe1c4 | 1654 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1655 | */ |
2dafe1c4 | 1656 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1657 | int mnt_flags, char *name, void *data) |
1658 | { | |
1659 | struct vfsmount *mnt; | |
1660 | ||
1661 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1662 | return -EINVAL; | |
1663 | ||
1664 | /* we need capabilities... */ | |
1665 | if (!capable(CAP_SYS_ADMIN)) | |
1666 | return -EPERM; | |
1667 | ||
1668 | mnt = do_kern_mount(type, flags, name, data); | |
1669 | if (IS_ERR(mnt)) | |
1670 | return PTR_ERR(mnt); | |
1671 | ||
1672 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * add a mount into a namespace's mount tree | |
1677 | * - provide the option of adding the new mount to an expiration list | |
1678 | */ | |
1679 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1680 | int mnt_flags, struct list_head *fslist) | |
1681 | { | |
1682 | int err; | |
1683 | ||
390c6843 | 1684 | down_write(&namespace_sem); |
1da177e4 | 1685 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1686 | while (d_mountpoint(nd->path.dentry) && |
1687 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1688 | ; |
1689 | err = -EINVAL; | |
4ac91378 | 1690 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1691 | goto unlock; |
1692 | ||
1693 | /* Refuse the same filesystem on the same mount point */ | |
1694 | err = -EBUSY; | |
4ac91378 JB |
1695 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1696 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1697 | goto unlock; |
1698 | ||
1699 | err = -EINVAL; | |
1700 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1701 | goto unlock; | |
1702 | ||
1703 | newmnt->mnt_flags = mnt_flags; | |
8c3ee42e | 1704 | if ((err = graft_tree(newmnt, &nd->path))) |
5b83d2c5 | 1705 | goto unlock; |
1da177e4 | 1706 | |
6758f953 | 1707 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1708 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1709 | |
390c6843 | 1710 | up_write(&namespace_sem); |
5b83d2c5 | 1711 | return 0; |
1da177e4 LT |
1712 | |
1713 | unlock: | |
390c6843 | 1714 | up_write(&namespace_sem); |
1da177e4 LT |
1715 | mntput(newmnt); |
1716 | return err; | |
1717 | } | |
1718 | ||
1719 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1720 | ||
1721 | /* | |
1722 | * process a list of expirable mountpoints with the intent of discarding any | |
1723 | * mountpoints that aren't in use and haven't been touched since last we came | |
1724 | * here | |
1725 | */ | |
1726 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1727 | { | |
1da177e4 LT |
1728 | struct vfsmount *mnt, *next; |
1729 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1730 | LIST_HEAD(umounts); |
1da177e4 LT |
1731 | |
1732 | if (list_empty(mounts)) | |
1733 | return; | |
1734 | ||
bcc5c7d2 | 1735 | down_write(&namespace_sem); |
1da177e4 LT |
1736 | spin_lock(&vfsmount_lock); |
1737 | ||
1738 | /* extract from the expiration list every vfsmount that matches the | |
1739 | * following criteria: | |
1740 | * - only referenced by its parent vfsmount | |
1741 | * - still marked for expiry (marked on the last call here; marks are | |
1742 | * cleared by mntput()) | |
1743 | */ | |
55e700b9 | 1744 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1745 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1746 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1747 | continue; |
55e700b9 | 1748 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1749 | } |
bcc5c7d2 AV |
1750 | while (!list_empty(&graveyard)) { |
1751 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1752 | touch_mnt_namespace(mnt->mnt_ns); | |
1753 | umount_tree(mnt, 1, &umounts); | |
1754 | } | |
5528f911 | 1755 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1756 | up_write(&namespace_sem); |
1757 | ||
1758 | release_mounts(&umounts); | |
5528f911 TM |
1759 | } |
1760 | ||
1761 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1762 | ||
1763 | /* | |
1764 | * Ripoff of 'select_parent()' | |
1765 | * | |
1766 | * search the list of submounts for a given mountpoint, and move any | |
1767 | * shrinkable submounts to the 'graveyard' list. | |
1768 | */ | |
1769 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1770 | { | |
1771 | struct vfsmount *this_parent = parent; | |
1772 | struct list_head *next; | |
1773 | int found = 0; | |
1774 | ||
1775 | repeat: | |
1776 | next = this_parent->mnt_mounts.next; | |
1777 | resume: | |
1778 | while (next != &this_parent->mnt_mounts) { | |
1779 | struct list_head *tmp = next; | |
1780 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1781 | ||
1782 | next = tmp->next; | |
1783 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1784 | continue; |
5528f911 TM |
1785 | /* |
1786 | * Descend a level if the d_mounts list is non-empty. | |
1787 | */ | |
1788 | if (!list_empty(&mnt->mnt_mounts)) { | |
1789 | this_parent = mnt; | |
1790 | goto repeat; | |
1791 | } | |
1da177e4 | 1792 | |
5528f911 | 1793 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1794 | list_move_tail(&mnt->mnt_expire, graveyard); |
1795 | found++; | |
1796 | } | |
1da177e4 | 1797 | } |
5528f911 TM |
1798 | /* |
1799 | * All done at this level ... ascend and resume the search | |
1800 | */ | |
1801 | if (this_parent != parent) { | |
1802 | next = this_parent->mnt_child.next; | |
1803 | this_parent = this_parent->mnt_parent; | |
1804 | goto resume; | |
1805 | } | |
1806 | return found; | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * process a list of expirable mountpoints with the intent of discarding any | |
1811 | * submounts of a specific parent mountpoint | |
1812 | */ | |
c35038be | 1813 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1814 | { |
1815 | LIST_HEAD(graveyard); | |
c35038be | 1816 | struct vfsmount *m; |
5528f911 | 1817 | |
5528f911 | 1818 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1819 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1820 | while (!list_empty(&graveyard)) { |
c35038be | 1821 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 AV |
1822 | mnt_expire); |
1823 | touch_mnt_namespace(mnt->mnt_ns); | |
c35038be | 1824 | umount_tree(mnt, 1, umounts); |
bcc5c7d2 AV |
1825 | } |
1826 | } | |
1da177e4 LT |
1827 | } |
1828 | ||
1da177e4 LT |
1829 | /* |
1830 | * Some copy_from_user() implementations do not return the exact number of | |
1831 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1832 | * Note that this function differs from copy_from_user() in that it will oops | |
1833 | * on bad values of `to', rather than returning a short copy. | |
1834 | */ | |
b58fed8b RP |
1835 | static long exact_copy_from_user(void *to, const void __user * from, |
1836 | unsigned long n) | |
1da177e4 LT |
1837 | { |
1838 | char *t = to; | |
1839 | const char __user *f = from; | |
1840 | char c; | |
1841 | ||
1842 | if (!access_ok(VERIFY_READ, from, n)) | |
1843 | return n; | |
1844 | ||
1845 | while (n) { | |
1846 | if (__get_user(c, f)) { | |
1847 | memset(t, 0, n); | |
1848 | break; | |
1849 | } | |
1850 | *t++ = c; | |
1851 | f++; | |
1852 | n--; | |
1853 | } | |
1854 | return n; | |
1855 | } | |
1856 | ||
b58fed8b | 1857 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1858 | { |
1859 | int i; | |
1860 | unsigned long page; | |
1861 | unsigned long size; | |
b58fed8b | 1862 | |
1da177e4 LT |
1863 | *where = 0; |
1864 | if (!data) | |
1865 | return 0; | |
1866 | ||
1867 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1868 | return -ENOMEM; | |
1869 | ||
1870 | /* We only care that *some* data at the address the user | |
1871 | * gave us is valid. Just in case, we'll zero | |
1872 | * the remainder of the page. | |
1873 | */ | |
1874 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1875 | size = TASK_SIZE - (unsigned long)data; | |
1876 | if (size > PAGE_SIZE) | |
1877 | size = PAGE_SIZE; | |
1878 | ||
1879 | i = size - exact_copy_from_user((void *)page, data, size); | |
1880 | if (!i) { | |
b58fed8b | 1881 | free_page(page); |
1da177e4 LT |
1882 | return -EFAULT; |
1883 | } | |
1884 | if (i != PAGE_SIZE) | |
1885 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1886 | *where = page; | |
1887 | return 0; | |
1888 | } | |
1889 | ||
1890 | /* | |
1891 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1892 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1893 | * | |
1894 | * data is a (void *) that can point to any structure up to | |
1895 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1896 | * information (or be NULL). | |
1897 | * | |
1898 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1899 | * When the flags word was introduced its top half was required | |
1900 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1901 | * Therefore, if this magic number is present, it carries no information | |
1902 | * and must be discarded. | |
1903 | */ | |
b58fed8b | 1904 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1905 | unsigned long flags, void *data_page) |
1906 | { | |
1907 | struct nameidata nd; | |
1908 | int retval = 0; | |
1909 | int mnt_flags = 0; | |
1910 | ||
1911 | /* Discard magic */ | |
1912 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1913 | flags &= ~MS_MGC_MSK; | |
1914 | ||
1915 | /* Basic sanity checks */ | |
1916 | ||
1917 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1918 | return -EINVAL; | |
1919 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1920 | return -EINVAL; | |
1921 | ||
1922 | if (data_page) | |
1923 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1924 | ||
1925 | /* Separate the per-mountpoint flags */ | |
1926 | if (flags & MS_NOSUID) | |
1927 | mnt_flags |= MNT_NOSUID; | |
1928 | if (flags & MS_NODEV) | |
1929 | mnt_flags |= MNT_NODEV; | |
1930 | if (flags & MS_NOEXEC) | |
1931 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1932 | if (flags & MS_NOATIME) |
1933 | mnt_flags |= MNT_NOATIME; | |
1934 | if (flags & MS_NODIRATIME) | |
1935 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1936 | if (flags & MS_RELATIME) |
1937 | mnt_flags |= MNT_RELATIME; | |
2e4b7fcd DH |
1938 | if (flags & MS_RDONLY) |
1939 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1940 | |
1941 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1942 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1943 | |
1944 | /* ... and get the mountpoint */ | |
1945 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1946 | if (retval) | |
1947 | return retval; | |
1948 | ||
b5266eb4 AV |
1949 | retval = security_sb_mount(dev_name, &nd.path, |
1950 | type_page, flags, data_page); | |
1da177e4 LT |
1951 | if (retval) |
1952 | goto dput_out; | |
1953 | ||
1954 | if (flags & MS_REMOUNT) | |
1955 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1956 | data_page); | |
1957 | else if (flags & MS_BIND) | |
eee391a6 | 1958 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1959 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1960 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1961 | else if (flags & MS_MOVE) |
1962 | retval = do_move_mount(&nd, dev_name); | |
1963 | else | |
1964 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1965 | dev_name, data_page); | |
1966 | dput_out: | |
1d957f9b | 1967 | path_put(&nd.path); |
1da177e4 LT |
1968 | return retval; |
1969 | } | |
1970 | ||
741a2951 JD |
1971 | /* |
1972 | * Allocate a new namespace structure and populate it with contents | |
1973 | * copied from the namespace of the passed in task structure. | |
1974 | */ | |
e3222c4e | 1975 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1976 | struct fs_struct *fs) |
1da177e4 | 1977 | { |
6b3286ed | 1978 | struct mnt_namespace *new_ns; |
1da177e4 | 1979 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1980 | struct vfsmount *p, *q; |
1981 | ||
6b3286ed | 1982 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1983 | if (!new_ns) |
467e9f4b | 1984 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1985 | |
1986 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1987 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1988 | init_waitqueue_head(&new_ns->poll); |
1989 | new_ns->event = 0; | |
1da177e4 | 1990 | |
390c6843 | 1991 | down_write(&namespace_sem); |
1da177e4 | 1992 | /* First pass: copy the tree topology */ |
6b3286ed | 1993 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1994 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1995 | if (!new_ns->root) { |
390c6843 | 1996 | up_write(&namespace_sem); |
1da177e4 | 1997 | kfree(new_ns); |
467e9f4b | 1998 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1999 | } |
2000 | spin_lock(&vfsmount_lock); | |
2001 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
2002 | spin_unlock(&vfsmount_lock); | |
2003 | ||
2004 | /* | |
2005 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2006 | * as belonging to new namespace. We have already acquired a private | |
2007 | * fs_struct, so tsk->fs->lock is not needed. | |
2008 | */ | |
6b3286ed | 2009 | p = mnt_ns->root; |
1da177e4 LT |
2010 | q = new_ns->root; |
2011 | while (p) { | |
6b3286ed | 2012 | q->mnt_ns = new_ns; |
1da177e4 | 2013 | if (fs) { |
6ac08c39 | 2014 | if (p == fs->root.mnt) { |
1da177e4 | 2015 | rootmnt = p; |
6ac08c39 | 2016 | fs->root.mnt = mntget(q); |
1da177e4 | 2017 | } |
6ac08c39 | 2018 | if (p == fs->pwd.mnt) { |
1da177e4 | 2019 | pwdmnt = p; |
6ac08c39 | 2020 | fs->pwd.mnt = mntget(q); |
1da177e4 | 2021 | } |
6ac08c39 | 2022 | if (p == fs->altroot.mnt) { |
1da177e4 | 2023 | altrootmnt = p; |
6ac08c39 | 2024 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
2025 | } |
2026 | } | |
6b3286ed | 2027 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
2028 | q = next_mnt(q, new_ns->root); |
2029 | } | |
390c6843 | 2030 | up_write(&namespace_sem); |
1da177e4 | 2031 | |
1da177e4 LT |
2032 | if (rootmnt) |
2033 | mntput(rootmnt); | |
2034 | if (pwdmnt) | |
2035 | mntput(pwdmnt); | |
2036 | if (altrootmnt) | |
2037 | mntput(altrootmnt); | |
2038 | ||
741a2951 JD |
2039 | return new_ns; |
2040 | } | |
2041 | ||
213dd266 | 2042 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2043 | struct fs_struct *new_fs) |
741a2951 | 2044 | { |
6b3286ed | 2045 | struct mnt_namespace *new_ns; |
741a2951 | 2046 | |
e3222c4e | 2047 | BUG_ON(!ns); |
6b3286ed | 2048 | get_mnt_ns(ns); |
741a2951 JD |
2049 | |
2050 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2051 | return ns; |
741a2951 | 2052 | |
e3222c4e | 2053 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2054 | |
6b3286ed | 2055 | put_mnt_ns(ns); |
e3222c4e | 2056 | return new_ns; |
1da177e4 LT |
2057 | } |
2058 | ||
2059 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
2060 | char __user * type, unsigned long flags, | |
2061 | void __user * data) | |
2062 | { | |
2063 | int retval; | |
2064 | unsigned long data_page; | |
2065 | unsigned long type_page; | |
2066 | unsigned long dev_page; | |
2067 | char *dir_page; | |
2068 | ||
b58fed8b | 2069 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
2070 | if (retval < 0) |
2071 | return retval; | |
2072 | ||
2073 | dir_page = getname(dir_name); | |
2074 | retval = PTR_ERR(dir_page); | |
2075 | if (IS_ERR(dir_page)) | |
2076 | goto out1; | |
2077 | ||
b58fed8b | 2078 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2079 | if (retval < 0) |
2080 | goto out2; | |
2081 | ||
b58fed8b | 2082 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2083 | if (retval < 0) |
2084 | goto out3; | |
2085 | ||
2086 | lock_kernel(); | |
b58fed8b RP |
2087 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2088 | flags, (void *)data_page); | |
1da177e4 LT |
2089 | unlock_kernel(); |
2090 | free_page(data_page); | |
2091 | ||
2092 | out3: | |
2093 | free_page(dev_page); | |
2094 | out2: | |
2095 | putname(dir_page); | |
2096 | out1: | |
2097 | free_page(type_page); | |
2098 | return retval; | |
2099 | } | |
2100 | ||
2101 | /* | |
2102 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
2103 | * It can block. Requires the big lock held. | |
2104 | */ | |
ac748a09 | 2105 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 2106 | { |
6ac08c39 JB |
2107 | struct path old_root; |
2108 | ||
1da177e4 LT |
2109 | write_lock(&fs->lock); |
2110 | old_root = fs->root; | |
ac748a09 JB |
2111 | fs->root = *path; |
2112 | path_get(path); | |
1da177e4 | 2113 | write_unlock(&fs->lock); |
6ac08c39 JB |
2114 | if (old_root.dentry) |
2115 | path_put(&old_root); | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
2120 | * It can block. Requires the big lock held. | |
2121 | */ | |
ac748a09 | 2122 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 2123 | { |
6ac08c39 | 2124 | struct path old_pwd; |
1da177e4 LT |
2125 | |
2126 | write_lock(&fs->lock); | |
2127 | old_pwd = fs->pwd; | |
ac748a09 JB |
2128 | fs->pwd = *path; |
2129 | path_get(path); | |
1da177e4 LT |
2130 | write_unlock(&fs->lock); |
2131 | ||
6ac08c39 JB |
2132 | if (old_pwd.dentry) |
2133 | path_put(&old_pwd); | |
1da177e4 LT |
2134 | } |
2135 | ||
1a390689 | 2136 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
2137 | { |
2138 | struct task_struct *g, *p; | |
2139 | struct fs_struct *fs; | |
2140 | ||
2141 | read_lock(&tasklist_lock); | |
2142 | do_each_thread(g, p) { | |
2143 | task_lock(p); | |
2144 | fs = p->fs; | |
2145 | if (fs) { | |
2146 | atomic_inc(&fs->count); | |
2147 | task_unlock(p); | |
1a390689 AV |
2148 | if (fs->root.dentry == old_root->dentry |
2149 | && fs->root.mnt == old_root->mnt) | |
2150 | set_fs_root(fs, new_root); | |
2151 | if (fs->pwd.dentry == old_root->dentry | |
2152 | && fs->pwd.mnt == old_root->mnt) | |
2153 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
2154 | put_fs_struct(fs); |
2155 | } else | |
2156 | task_unlock(p); | |
2157 | } while_each_thread(g, p); | |
2158 | read_unlock(&tasklist_lock); | |
2159 | } | |
2160 | ||
2161 | /* | |
2162 | * pivot_root Semantics: | |
2163 | * Moves the root file system of the current process to the directory put_old, | |
2164 | * makes new_root as the new root file system of the current process, and sets | |
2165 | * root/cwd of all processes which had them on the current root to new_root. | |
2166 | * | |
2167 | * Restrictions: | |
2168 | * The new_root and put_old must be directories, and must not be on the | |
2169 | * same file system as the current process root. The put_old must be | |
2170 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2171 | * pointed to by put_old must yield the same directory as new_root. No other | |
2172 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2173 | * | |
4a0d11fa NB |
2174 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2175 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2176 | * in this situation. | |
2177 | * | |
1da177e4 LT |
2178 | * Notes: |
2179 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2180 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2181 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2182 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2183 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2184 | * first. | |
2185 | */ | |
b58fed8b RP |
2186 | asmlinkage long sys_pivot_root(const char __user * new_root, |
2187 | const char __user * put_old) | |
1da177e4 LT |
2188 | { |
2189 | struct vfsmount *tmp; | |
8c3ee42e AV |
2190 | struct nameidata new_nd, old_nd; |
2191 | struct path parent_path, root_parent, root; | |
1da177e4 LT |
2192 | int error; |
2193 | ||
2194 | if (!capable(CAP_SYS_ADMIN)) | |
2195 | return -EPERM; | |
2196 | ||
b58fed8b RP |
2197 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
2198 | &new_nd); | |
1da177e4 LT |
2199 | if (error) |
2200 | goto out0; | |
2201 | error = -EINVAL; | |
4ac91378 | 2202 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
2203 | goto out1; |
2204 | ||
b58fed8b | 2205 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
2206 | if (error) |
2207 | goto out1; | |
2208 | ||
b5266eb4 | 2209 | error = security_sb_pivotroot(&old_nd.path, &new_nd.path); |
1da177e4 | 2210 | if (error) { |
1d957f9b | 2211 | path_put(&old_nd.path); |
1da177e4 LT |
2212 | goto out1; |
2213 | } | |
2214 | ||
2215 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2216 | root = current->fs->root; |
6ac08c39 | 2217 | path_get(¤t->fs->root); |
1da177e4 | 2218 | read_unlock(¤t->fs->lock); |
390c6843 | 2219 | down_write(&namespace_sem); |
4ac91378 | 2220 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2221 | error = -EINVAL; |
4ac91378 JB |
2222 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
2223 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
8c3ee42e | 2224 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2225 | goto out2; |
8c3ee42e | 2226 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2227 | goto out2; |
2228 | error = -ENOENT; | |
4ac91378 | 2229 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 2230 | goto out2; |
4ac91378 | 2231 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 2232 | goto out2; |
4ac91378 | 2233 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
2234 | goto out2; |
2235 | error = -EBUSY; | |
8c3ee42e AV |
2236 | if (new_nd.path.mnt == root.mnt || |
2237 | old_nd.path.mnt == root.mnt) | |
1da177e4 LT |
2238 | goto out2; /* loop, on the same file system */ |
2239 | error = -EINVAL; | |
8c3ee42e | 2240 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2241 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2242 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2243 | goto out2; /* not attached */ |
4ac91378 | 2244 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 2245 | goto out2; /* not a mountpoint */ |
4ac91378 | 2246 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 2247 | goto out2; /* not attached */ |
4ac91378 JB |
2248 | /* make sure we can reach put_old from new_root */ |
2249 | tmp = old_nd.path.mnt; | |
1da177e4 | 2250 | spin_lock(&vfsmount_lock); |
4ac91378 | 2251 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
2252 | for (;;) { |
2253 | if (tmp->mnt_parent == tmp) | |
2254 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 2255 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
2256 | break; |
2257 | tmp = tmp->mnt_parent; | |
2258 | } | |
4ac91378 | 2259 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 2260 | goto out3; |
4ac91378 | 2261 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 2262 | goto out3; |
1a390689 | 2263 | detach_mnt(new_nd.path.mnt, &parent_path); |
8c3ee42e | 2264 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2265 | /* mount old root on put_old */ |
8c3ee42e | 2266 | attach_mnt(root.mnt, &old_nd.path); |
4ac91378 JB |
2267 | /* mount new_root on / */ |
2268 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 2269 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2270 | spin_unlock(&vfsmount_lock); |
8c3ee42e AV |
2271 | chroot_fs_refs(&root, &new_nd.path); |
2272 | security_sb_post_pivotroot(&root, &new_nd.path); | |
1da177e4 | 2273 | error = 0; |
1a390689 AV |
2274 | path_put(&root_parent); |
2275 | path_put(&parent_path); | |
1da177e4 | 2276 | out2: |
4ac91378 | 2277 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 2278 | up_write(&namespace_sem); |
8c3ee42e | 2279 | path_put(&root); |
1d957f9b | 2280 | path_put(&old_nd.path); |
1da177e4 | 2281 | out1: |
1d957f9b | 2282 | path_put(&new_nd.path); |
1da177e4 | 2283 | out0: |
1da177e4 LT |
2284 | return error; |
2285 | out3: | |
2286 | spin_unlock(&vfsmount_lock); | |
2287 | goto out2; | |
2288 | } | |
2289 | ||
2290 | static void __init init_mount_tree(void) | |
2291 | { | |
2292 | struct vfsmount *mnt; | |
6b3286ed | 2293 | struct mnt_namespace *ns; |
ac748a09 | 2294 | struct path root; |
1da177e4 LT |
2295 | |
2296 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2297 | if (IS_ERR(mnt)) | |
2298 | panic("Can't create rootfs"); | |
6b3286ed KK |
2299 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2300 | if (!ns) | |
1da177e4 | 2301 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2302 | atomic_set(&ns->count, 1); |
2303 | INIT_LIST_HEAD(&ns->list); | |
2304 | init_waitqueue_head(&ns->poll); | |
2305 | ns->event = 0; | |
2306 | list_add(&mnt->mnt_list, &ns->list); | |
2307 | ns->root = mnt; | |
2308 | mnt->mnt_ns = ns; | |
2309 | ||
2310 | init_task.nsproxy->mnt_ns = ns; | |
2311 | get_mnt_ns(ns); | |
2312 | ||
ac748a09 JB |
2313 | root.mnt = ns->root; |
2314 | root.dentry = ns->root->mnt_root; | |
2315 | ||
2316 | set_fs_pwd(current->fs, &root); | |
2317 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2318 | } |
2319 | ||
74bf17cf | 2320 | void __init mnt_init(void) |
1da177e4 | 2321 | { |
13f14b4d | 2322 | unsigned u; |
15a67dd8 | 2323 | int err; |
1da177e4 | 2324 | |
390c6843 RP |
2325 | init_rwsem(&namespace_sem); |
2326 | ||
1da177e4 | 2327 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2328 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2329 | |
b58fed8b | 2330 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2331 | |
2332 | if (!mount_hashtable) | |
2333 | panic("Failed to allocate mount hash table\n"); | |
2334 | ||
13f14b4d ED |
2335 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2336 | ||
2337 | for (u = 0; u < HASH_SIZE; u++) | |
2338 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2339 | |
15a67dd8 RD |
2340 | err = sysfs_init(); |
2341 | if (err) | |
2342 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
2343 | __FUNCTION__, err); | |
00d26666 GKH |
2344 | fs_kobj = kobject_create_and_add("fs", NULL); |
2345 | if (!fs_kobj) | |
2346 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
2347 | init_rootfs(); |
2348 | init_mount_tree(); | |
2349 | } | |
2350 | ||
6b3286ed | 2351 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2352 | { |
6b3286ed | 2353 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2354 | LIST_HEAD(umount_list); |
6b3286ed | 2355 | ns->root = NULL; |
1ce88cf4 | 2356 | spin_unlock(&vfsmount_lock); |
390c6843 | 2357 | down_write(&namespace_sem); |
1da177e4 | 2358 | spin_lock(&vfsmount_lock); |
a05964f3 | 2359 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2360 | spin_unlock(&vfsmount_lock); |
390c6843 | 2361 | up_write(&namespace_sem); |
70fbcdf4 | 2362 | release_mounts(&umount_list); |
6b3286ed | 2363 | kfree(ns); |
1da177e4 | 2364 | } |