]>
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); | |
88b38782 LZ |
115 | if (err) |
116 | goto out_free_cache; | |
117 | ||
118 | if (name) { | |
119 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
120 | if (!mnt->mnt_devname) | |
121 | goto out_free_id; | |
73cd49ec MS |
122 | } |
123 | ||
b58fed8b | 124 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
125 | INIT_LIST_HEAD(&mnt->mnt_hash); |
126 | INIT_LIST_HEAD(&mnt->mnt_child); | |
127 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
128 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 129 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 130 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
131 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
132 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
3d733633 | 133 | atomic_set(&mnt->__mnt_writers, 0); |
1da177e4 LT |
134 | } |
135 | return mnt; | |
88b38782 LZ |
136 | |
137 | out_free_id: | |
138 | mnt_free_id(mnt); | |
139 | out_free_cache: | |
140 | kmem_cache_free(mnt_cache, mnt); | |
141 | return NULL; | |
1da177e4 LT |
142 | } |
143 | ||
3d733633 DH |
144 | /* |
145 | * Most r/o checks on a fs are for operations that take | |
146 | * discrete amounts of time, like a write() or unlink(). | |
147 | * We must keep track of when those operations start | |
148 | * (for permission checks) and when they end, so that | |
149 | * we can determine when writes are able to occur to | |
150 | * a filesystem. | |
151 | */ | |
152 | /* | |
153 | * __mnt_is_readonly: check whether a mount is read-only | |
154 | * @mnt: the mount to check for its write status | |
155 | * | |
156 | * This shouldn't be used directly ouside of the VFS. | |
157 | * It does not guarantee that the filesystem will stay | |
158 | * r/w, just that it is right *now*. This can not and | |
159 | * should not be used in place of IS_RDONLY(inode). | |
160 | * mnt_want/drop_write() will _keep_ the filesystem | |
161 | * r/w. | |
162 | */ | |
163 | int __mnt_is_readonly(struct vfsmount *mnt) | |
164 | { | |
2e4b7fcd DH |
165 | if (mnt->mnt_flags & MNT_READONLY) |
166 | return 1; | |
167 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
168 | return 1; | |
169 | return 0; | |
3d733633 DH |
170 | } |
171 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
172 | ||
173 | struct mnt_writer { | |
174 | /* | |
175 | * If holding multiple instances of this lock, they | |
176 | * must be ordered by cpu number. | |
177 | */ | |
178 | spinlock_t lock; | |
179 | struct lock_class_key lock_class; /* compiles out with !lockdep */ | |
180 | unsigned long count; | |
181 | struct vfsmount *mnt; | |
182 | } ____cacheline_aligned_in_smp; | |
183 | static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); | |
184 | ||
185 | static int __init init_mnt_writers(void) | |
186 | { | |
187 | int cpu; | |
188 | for_each_possible_cpu(cpu) { | |
189 | struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); | |
190 | spin_lock_init(&writer->lock); | |
191 | lockdep_set_class(&writer->lock, &writer->lock_class); | |
192 | writer->count = 0; | |
193 | } | |
194 | return 0; | |
195 | } | |
196 | fs_initcall(init_mnt_writers); | |
197 | ||
198 | static void unlock_mnt_writers(void) | |
199 | { | |
200 | int cpu; | |
201 | struct mnt_writer *cpu_writer; | |
202 | ||
203 | for_each_possible_cpu(cpu) { | |
204 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
205 | spin_unlock(&cpu_writer->lock); | |
206 | } | |
207 | } | |
208 | ||
209 | static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) | |
210 | { | |
211 | if (!cpu_writer->mnt) | |
212 | return; | |
213 | /* | |
214 | * This is in case anyone ever leaves an invalid, | |
215 | * old ->mnt and a count of 0. | |
216 | */ | |
217 | if (!cpu_writer->count) | |
218 | return; | |
219 | atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); | |
220 | cpu_writer->count = 0; | |
221 | } | |
222 | /* | |
223 | * must hold cpu_writer->lock | |
224 | */ | |
225 | static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, | |
226 | struct vfsmount *mnt) | |
227 | { | |
228 | if (cpu_writer->mnt == mnt) | |
229 | return; | |
230 | __clear_mnt_count(cpu_writer); | |
231 | cpu_writer->mnt = mnt; | |
232 | } | |
233 | ||
8366025e DH |
234 | /* |
235 | * Most r/o checks on a fs are for operations that take | |
236 | * discrete amounts of time, like a write() or unlink(). | |
237 | * We must keep track of when those operations start | |
238 | * (for permission checks) and when they end, so that | |
239 | * we can determine when writes are able to occur to | |
240 | * a filesystem. | |
241 | */ | |
242 | /** | |
243 | * mnt_want_write - get write access to a mount | |
244 | * @mnt: the mount on which to take a write | |
245 | * | |
246 | * This tells the low-level filesystem that a write is | |
247 | * about to be performed to it, and makes sure that | |
248 | * writes are allowed before returning success. When | |
249 | * the write operation is finished, mnt_drop_write() | |
250 | * must be called. This is effectively a refcount. | |
251 | */ | |
252 | int mnt_want_write(struct vfsmount *mnt) | |
253 | { | |
3d733633 DH |
254 | int ret = 0; |
255 | struct mnt_writer *cpu_writer; | |
256 | ||
257 | cpu_writer = &get_cpu_var(mnt_writers); | |
258 | spin_lock(&cpu_writer->lock); | |
259 | if (__mnt_is_readonly(mnt)) { | |
260 | ret = -EROFS; | |
261 | goto out; | |
262 | } | |
263 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
264 | cpu_writer->count++; | |
265 | out: | |
266 | spin_unlock(&cpu_writer->lock); | |
267 | put_cpu_var(mnt_writers); | |
268 | return ret; | |
8366025e DH |
269 | } |
270 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
271 | ||
3d733633 DH |
272 | static void lock_mnt_writers(void) |
273 | { | |
274 | int cpu; | |
275 | struct mnt_writer *cpu_writer; | |
276 | ||
277 | for_each_possible_cpu(cpu) { | |
278 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
279 | spin_lock(&cpu_writer->lock); | |
280 | __clear_mnt_count(cpu_writer); | |
281 | cpu_writer->mnt = NULL; | |
282 | } | |
283 | } | |
284 | ||
285 | /* | |
286 | * These per-cpu write counts are not guaranteed to have | |
287 | * matched increments and decrements on any given cpu. | |
288 | * A file open()ed for write on one cpu and close()d on | |
289 | * another cpu will imbalance this count. Make sure it | |
290 | * does not get too far out of whack. | |
291 | */ | |
292 | static void handle_write_count_underflow(struct vfsmount *mnt) | |
293 | { | |
294 | if (atomic_read(&mnt->__mnt_writers) >= | |
295 | MNT_WRITER_UNDERFLOW_LIMIT) | |
296 | return; | |
297 | /* | |
298 | * It isn't necessary to hold all of the locks | |
299 | * at the same time, but doing it this way makes | |
300 | * us share a lot more code. | |
301 | */ | |
302 | lock_mnt_writers(); | |
303 | /* | |
304 | * vfsmount_lock is for mnt_flags. | |
305 | */ | |
306 | spin_lock(&vfsmount_lock); | |
307 | /* | |
308 | * If coalescing the per-cpu writer counts did not | |
309 | * get us back to a positive writer count, we have | |
310 | * a bug. | |
311 | */ | |
312 | if ((atomic_read(&mnt->__mnt_writers) < 0) && | |
313 | !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { | |
5c752ad9 | 314 | WARN(1, KERN_DEBUG "leak detected on mount(%p) writers " |
3d733633 DH |
315 | "count: %d\n", |
316 | mnt, atomic_read(&mnt->__mnt_writers)); | |
3d733633 DH |
317 | /* use the flag to keep the dmesg spam down */ |
318 | mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; | |
319 | } | |
320 | spin_unlock(&vfsmount_lock); | |
321 | unlock_mnt_writers(); | |
322 | } | |
323 | ||
8366025e DH |
324 | /** |
325 | * mnt_drop_write - give up write access to a mount | |
326 | * @mnt: the mount on which to give up write access | |
327 | * | |
328 | * Tells the low-level filesystem that we are done | |
329 | * performing writes to it. Must be matched with | |
330 | * mnt_want_write() call above. | |
331 | */ | |
332 | void mnt_drop_write(struct vfsmount *mnt) | |
333 | { | |
3d733633 DH |
334 | int must_check_underflow = 0; |
335 | struct mnt_writer *cpu_writer; | |
336 | ||
337 | cpu_writer = &get_cpu_var(mnt_writers); | |
338 | spin_lock(&cpu_writer->lock); | |
339 | ||
340 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
341 | if (cpu_writer->count > 0) { | |
342 | cpu_writer->count--; | |
343 | } else { | |
344 | must_check_underflow = 1; | |
345 | atomic_dec(&mnt->__mnt_writers); | |
346 | } | |
347 | ||
348 | spin_unlock(&cpu_writer->lock); | |
349 | /* | |
350 | * Logically, we could call this each time, | |
351 | * but the __mnt_writers cacheline tends to | |
352 | * be cold, and makes this expensive. | |
353 | */ | |
354 | if (must_check_underflow) | |
355 | handle_write_count_underflow(mnt); | |
356 | /* | |
357 | * This could be done right after the spinlock | |
358 | * is taken because the spinlock keeps us on | |
359 | * the cpu, and disables preemption. However, | |
360 | * putting it here bounds the amount that | |
361 | * __mnt_writers can underflow. Without it, | |
362 | * we could theoretically wrap __mnt_writers. | |
363 | */ | |
364 | put_cpu_var(mnt_writers); | |
8366025e DH |
365 | } |
366 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
367 | ||
2e4b7fcd | 368 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 369 | { |
3d733633 DH |
370 | int ret = 0; |
371 | ||
372 | lock_mnt_writers(); | |
373 | /* | |
374 | * With all the locks held, this value is stable | |
375 | */ | |
376 | if (atomic_read(&mnt->__mnt_writers) > 0) { | |
377 | ret = -EBUSY; | |
378 | goto out; | |
379 | } | |
380 | /* | |
2e4b7fcd DH |
381 | * nobody can do a successful mnt_want_write() with all |
382 | * of the counts in MNT_DENIED_WRITE and the locks held. | |
3d733633 | 383 | */ |
2e4b7fcd DH |
384 | spin_lock(&vfsmount_lock); |
385 | if (!ret) | |
386 | mnt->mnt_flags |= MNT_READONLY; | |
387 | spin_unlock(&vfsmount_lock); | |
3d733633 DH |
388 | out: |
389 | unlock_mnt_writers(); | |
390 | return ret; | |
8366025e | 391 | } |
8366025e | 392 | |
2e4b7fcd DH |
393 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
394 | { | |
395 | spin_lock(&vfsmount_lock); | |
396 | mnt->mnt_flags &= ~MNT_READONLY; | |
397 | spin_unlock(&vfsmount_lock); | |
398 | } | |
399 | ||
a3ec947c | 400 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
454e2398 DH |
401 | { |
402 | mnt->mnt_sb = sb; | |
403 | mnt->mnt_root = dget(sb->s_root); | |
454e2398 DH |
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); | |
3d733633 | 616 | spin_lock(&cpu_writer->lock); |
1a88b536 AV |
617 | if (cpu_writer->mnt != mnt) { |
618 | spin_unlock(&cpu_writer->lock); | |
619 | continue; | |
620 | } | |
3d733633 DH |
621 | atomic_add(cpu_writer->count, &mnt->__mnt_writers); |
622 | cpu_writer->count = 0; | |
623 | /* | |
624 | * Might as well do this so that no one | |
625 | * ever sees the pointer and expects | |
626 | * it to be valid. | |
627 | */ | |
628 | cpu_writer->mnt = NULL; | |
629 | spin_unlock(&cpu_writer->lock); | |
630 | } | |
631 | /* | |
632 | * This probably indicates that somebody messed | |
633 | * up a mnt_want/drop_write() pair. If this | |
634 | * happens, the filesystem was probably unable | |
635 | * to make r/w->r/o transitions. | |
636 | */ | |
637 | WARN_ON(atomic_read(&mnt->__mnt_writers)); | |
1da177e4 LT |
638 | dput(mnt->mnt_root); |
639 | free_vfsmnt(mnt); | |
640 | deactivate_super(sb); | |
641 | } | |
642 | ||
7b7b1ace AV |
643 | void mntput_no_expire(struct vfsmount *mnt) |
644 | { | |
645 | repeat: | |
646 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
647 | if (likely(!mnt->mnt_pinned)) { | |
648 | spin_unlock(&vfsmount_lock); | |
649 | __mntput(mnt); | |
650 | return; | |
651 | } | |
652 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
653 | mnt->mnt_pinned = 0; | |
654 | spin_unlock(&vfsmount_lock); | |
655 | acct_auto_close_mnt(mnt); | |
656 | security_sb_umount_close(mnt); | |
657 | goto repeat; | |
658 | } | |
659 | } | |
660 | ||
661 | EXPORT_SYMBOL(mntput_no_expire); | |
662 | ||
663 | void mnt_pin(struct vfsmount *mnt) | |
664 | { | |
665 | spin_lock(&vfsmount_lock); | |
666 | mnt->mnt_pinned++; | |
667 | spin_unlock(&vfsmount_lock); | |
668 | } | |
669 | ||
670 | EXPORT_SYMBOL(mnt_pin); | |
671 | ||
672 | void mnt_unpin(struct vfsmount *mnt) | |
673 | { | |
674 | spin_lock(&vfsmount_lock); | |
675 | if (mnt->mnt_pinned) { | |
676 | atomic_inc(&mnt->mnt_count); | |
677 | mnt->mnt_pinned--; | |
678 | } | |
679 | spin_unlock(&vfsmount_lock); | |
680 | } | |
681 | ||
682 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 683 | |
b3b304a2 MS |
684 | static inline void mangle(struct seq_file *m, const char *s) |
685 | { | |
686 | seq_escape(m, s, " \t\n\\"); | |
687 | } | |
688 | ||
689 | /* | |
690 | * Simple .show_options callback for filesystems which don't want to | |
691 | * implement more complex mount option showing. | |
692 | * | |
693 | * See also save_mount_options(). | |
694 | */ | |
695 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
696 | { | |
697 | const char *options = mnt->mnt_sb->s_options; | |
698 | ||
699 | if (options != NULL && options[0]) { | |
700 | seq_putc(m, ','); | |
701 | mangle(m, options); | |
702 | } | |
703 | ||
704 | return 0; | |
705 | } | |
706 | EXPORT_SYMBOL(generic_show_options); | |
707 | ||
708 | /* | |
709 | * If filesystem uses generic_show_options(), this function should be | |
710 | * called from the fill_super() callback. | |
711 | * | |
712 | * The .remount_fs callback usually needs to be handled in a special | |
713 | * way, to make sure, that previous options are not overwritten if the | |
714 | * remount fails. | |
715 | * | |
716 | * Also note, that if the filesystem's .remount_fs function doesn't | |
717 | * reset all options to their default value, but changes only newly | |
718 | * given options, then the displayed options will not reflect reality | |
719 | * any more. | |
720 | */ | |
721 | void save_mount_options(struct super_block *sb, char *options) | |
722 | { | |
723 | kfree(sb->s_options); | |
724 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
725 | } | |
726 | EXPORT_SYMBOL(save_mount_options); | |
727 | ||
a1a2c409 | 728 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
729 | /* iterator */ |
730 | static void *m_start(struct seq_file *m, loff_t *pos) | |
731 | { | |
a1a2c409 | 732 | struct proc_mounts *p = m->private; |
1da177e4 | 733 | |
390c6843 | 734 | down_read(&namespace_sem); |
a1a2c409 | 735 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
736 | } |
737 | ||
738 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
739 | { | |
a1a2c409 | 740 | struct proc_mounts *p = m->private; |
b0765fb8 | 741 | |
a1a2c409 | 742 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
743 | } |
744 | ||
745 | static void m_stop(struct seq_file *m, void *v) | |
746 | { | |
390c6843 | 747 | up_read(&namespace_sem); |
1da177e4 LT |
748 | } |
749 | ||
2d4d4864 RP |
750 | struct proc_fs_info { |
751 | int flag; | |
752 | const char *str; | |
753 | }; | |
754 | ||
2069f457 | 755 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) |
1da177e4 | 756 | { |
2d4d4864 | 757 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
758 | { MS_SYNCHRONOUS, ",sync" }, |
759 | { MS_DIRSYNC, ",dirsync" }, | |
760 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
761 | { 0, NULL } |
762 | }; | |
2d4d4864 RP |
763 | const struct proc_fs_info *fs_infop; |
764 | ||
765 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
766 | if (sb->s_flags & fs_infop->flag) | |
767 | seq_puts(m, fs_infop->str); | |
768 | } | |
2069f457 EP |
769 | |
770 | return security_sb_show_options(m, sb); | |
2d4d4864 RP |
771 | } |
772 | ||
773 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
774 | { | |
775 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
776 | { MNT_NOSUID, ",nosuid" }, |
777 | { MNT_NODEV, ",nodev" }, | |
778 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
779 | { MNT_NOATIME, ",noatime" }, |
780 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 781 | { MNT_RELATIME, ",relatime" }, |
d0adde57 | 782 | { MNT_STRICTATIME, ",strictatime" }, |
1da177e4 LT |
783 | { 0, NULL } |
784 | }; | |
2d4d4864 RP |
785 | const struct proc_fs_info *fs_infop; |
786 | ||
787 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
788 | if (mnt->mnt_flags & fs_infop->flag) | |
789 | seq_puts(m, fs_infop->str); | |
790 | } | |
791 | } | |
792 | ||
793 | static void show_type(struct seq_file *m, struct super_block *sb) | |
794 | { | |
795 | mangle(m, sb->s_type->name); | |
796 | if (sb->s_subtype && sb->s_subtype[0]) { | |
797 | seq_putc(m, '.'); | |
798 | mangle(m, sb->s_subtype); | |
799 | } | |
800 | } | |
801 | ||
802 | static int show_vfsmnt(struct seq_file *m, void *v) | |
803 | { | |
804 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
805 | int err = 0; | |
c32c2f63 | 806 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
807 | |
808 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
809 | seq_putc(m, ' '); | |
c32c2f63 | 810 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 811 | seq_putc(m, ' '); |
2d4d4864 | 812 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 813 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2069f457 EP |
814 | err = show_sb_opts(m, mnt->mnt_sb); |
815 | if (err) | |
816 | goto out; | |
2d4d4864 | 817 | show_mnt_opts(m, mnt); |
1da177e4 LT |
818 | if (mnt->mnt_sb->s_op->show_options) |
819 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
820 | seq_puts(m, " 0 0\n"); | |
2069f457 | 821 | out: |
1da177e4 LT |
822 | return err; |
823 | } | |
824 | ||
a1a2c409 | 825 | const struct seq_operations mounts_op = { |
1da177e4 LT |
826 | .start = m_start, |
827 | .next = m_next, | |
828 | .stop = m_stop, | |
829 | .show = show_vfsmnt | |
830 | }; | |
831 | ||
2d4d4864 RP |
832 | static int show_mountinfo(struct seq_file *m, void *v) |
833 | { | |
834 | struct proc_mounts *p = m->private; | |
835 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
836 | struct super_block *sb = mnt->mnt_sb; | |
837 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
838 | struct path root = p->root; | |
839 | int err = 0; | |
840 | ||
841 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
842 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
843 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
844 | seq_putc(m, ' '); | |
845 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
846 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
847 | /* | |
848 | * Mountpoint is outside root, discard that one. Ugly, | |
849 | * but less so than trying to do that in iterator in a | |
850 | * race-free way (due to renames). | |
851 | */ | |
852 | return SEQ_SKIP; | |
853 | } | |
854 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
855 | show_mnt_opts(m, mnt); | |
856 | ||
857 | /* Tagged fields ("foo:X" or "bar") */ | |
858 | if (IS_MNT_SHARED(mnt)) | |
859 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
860 | if (IS_MNT_SLAVE(mnt)) { |
861 | int master = mnt->mnt_master->mnt_group_id; | |
862 | int dom = get_dominating_id(mnt, &p->root); | |
863 | seq_printf(m, " master:%i", master); | |
864 | if (dom && dom != master) | |
865 | seq_printf(m, " propagate_from:%i", dom); | |
866 | } | |
2d4d4864 RP |
867 | if (IS_MNT_UNBINDABLE(mnt)) |
868 | seq_puts(m, " unbindable"); | |
869 | ||
870 | /* Filesystem specific data */ | |
871 | seq_puts(m, " - "); | |
872 | show_type(m, sb); | |
873 | seq_putc(m, ' '); | |
874 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
875 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
2069f457 EP |
876 | err = show_sb_opts(m, sb); |
877 | if (err) | |
878 | goto out; | |
2d4d4864 RP |
879 | if (sb->s_op->show_options) |
880 | err = sb->s_op->show_options(m, mnt); | |
881 | seq_putc(m, '\n'); | |
2069f457 | 882 | out: |
2d4d4864 RP |
883 | return err; |
884 | } | |
885 | ||
886 | const struct seq_operations mountinfo_op = { | |
887 | .start = m_start, | |
888 | .next = m_next, | |
889 | .stop = m_stop, | |
890 | .show = show_mountinfo, | |
891 | }; | |
892 | ||
b4629fe2 CL |
893 | static int show_vfsstat(struct seq_file *m, void *v) |
894 | { | |
b0765fb8 | 895 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 896 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
897 | int err = 0; |
898 | ||
899 | /* device */ | |
900 | if (mnt->mnt_devname) { | |
901 | seq_puts(m, "device "); | |
902 | mangle(m, mnt->mnt_devname); | |
903 | } else | |
904 | seq_puts(m, "no device"); | |
905 | ||
906 | /* mount point */ | |
907 | seq_puts(m, " mounted on "); | |
c32c2f63 | 908 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
909 | seq_putc(m, ' '); |
910 | ||
911 | /* file system type */ | |
912 | seq_puts(m, "with fstype "); | |
2d4d4864 | 913 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
914 | |
915 | /* optional statistics */ | |
916 | if (mnt->mnt_sb->s_op->show_stats) { | |
917 | seq_putc(m, ' '); | |
918 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
919 | } | |
920 | ||
921 | seq_putc(m, '\n'); | |
922 | return err; | |
923 | } | |
924 | ||
a1a2c409 | 925 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
926 | .start = m_start, |
927 | .next = m_next, | |
928 | .stop = m_stop, | |
929 | .show = show_vfsstat, | |
930 | }; | |
a1a2c409 | 931 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 932 | |
1da177e4 LT |
933 | /** |
934 | * may_umount_tree - check if a mount tree is busy | |
935 | * @mnt: root of mount tree | |
936 | * | |
937 | * This is called to check if a tree of mounts has any | |
938 | * open files, pwds, chroots or sub mounts that are | |
939 | * busy. | |
940 | */ | |
941 | int may_umount_tree(struct vfsmount *mnt) | |
942 | { | |
36341f64 RP |
943 | int actual_refs = 0; |
944 | int minimum_refs = 0; | |
945 | struct vfsmount *p; | |
1da177e4 LT |
946 | |
947 | spin_lock(&vfsmount_lock); | |
36341f64 | 948 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
949 | actual_refs += atomic_read(&p->mnt_count); |
950 | minimum_refs += 2; | |
1da177e4 LT |
951 | } |
952 | spin_unlock(&vfsmount_lock); | |
953 | ||
954 | if (actual_refs > minimum_refs) | |
e3474a8e | 955 | return 0; |
1da177e4 | 956 | |
e3474a8e | 957 | return 1; |
1da177e4 LT |
958 | } |
959 | ||
960 | EXPORT_SYMBOL(may_umount_tree); | |
961 | ||
962 | /** | |
963 | * may_umount - check if a mount point is busy | |
964 | * @mnt: root of mount | |
965 | * | |
966 | * This is called to check if a mount point has any | |
967 | * open files, pwds, chroots or sub mounts. If the | |
968 | * mount has sub mounts this will return busy | |
969 | * regardless of whether the sub mounts are busy. | |
970 | * | |
971 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
972 | * give false negatives. The main reason why it's here is that we need | |
973 | * a non-destructive way to look for easily umountable filesystems. | |
974 | */ | |
975 | int may_umount(struct vfsmount *mnt) | |
976 | { | |
e3474a8e | 977 | int ret = 1; |
a05964f3 RP |
978 | spin_lock(&vfsmount_lock); |
979 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 980 | ret = 0; |
a05964f3 RP |
981 | spin_unlock(&vfsmount_lock); |
982 | return ret; | |
1da177e4 LT |
983 | } |
984 | ||
985 | EXPORT_SYMBOL(may_umount); | |
986 | ||
b90fa9ae | 987 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
988 | { |
989 | struct vfsmount *mnt; | |
bf066c7d | 990 | while (!list_empty(head)) { |
b5e61818 | 991 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
992 | list_del_init(&mnt->mnt_hash); |
993 | if (mnt->mnt_parent != mnt) { | |
994 | struct dentry *dentry; | |
995 | struct vfsmount *m; | |
996 | spin_lock(&vfsmount_lock); | |
997 | dentry = mnt->mnt_mountpoint; | |
998 | m = mnt->mnt_parent; | |
999 | mnt->mnt_mountpoint = mnt->mnt_root; | |
1000 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 1001 | m->mnt_ghosts--; |
70fbcdf4 RP |
1002 | spin_unlock(&vfsmount_lock); |
1003 | dput(dentry); | |
1004 | mntput(m); | |
1005 | } | |
1006 | mntput(mnt); | |
1007 | } | |
1008 | } | |
1009 | ||
a05964f3 | 1010 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
1011 | { |
1012 | struct vfsmount *p; | |
1da177e4 | 1013 | |
1bfba4e8 AM |
1014 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1015 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 1016 | |
a05964f3 RP |
1017 | if (propagate) |
1018 | propagate_umount(kill); | |
1019 | ||
70fbcdf4 RP |
1020 | list_for_each_entry(p, kill, mnt_hash) { |
1021 | list_del_init(&p->mnt_expire); | |
1022 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
1023 | __touch_mnt_namespace(p->mnt_ns); |
1024 | p->mnt_ns = NULL; | |
70fbcdf4 | 1025 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
1026 | if (p->mnt_parent != p) { |
1027 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 1028 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 1029 | } |
a05964f3 | 1030 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1031 | } |
1032 | } | |
1033 | ||
c35038be AV |
1034 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
1035 | ||
1da177e4 LT |
1036 | static int do_umount(struct vfsmount *mnt, int flags) |
1037 | { | |
b58fed8b | 1038 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 1039 | int retval; |
70fbcdf4 | 1040 | LIST_HEAD(umount_list); |
1da177e4 LT |
1041 | |
1042 | retval = security_sb_umount(mnt, flags); | |
1043 | if (retval) | |
1044 | return retval; | |
1045 | ||
1046 | /* | |
1047 | * Allow userspace to request a mountpoint be expired rather than | |
1048 | * unmounting unconditionally. Unmount only happens if: | |
1049 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1050 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1051 | */ | |
1052 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 1053 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
1054 | flags & (MNT_FORCE | MNT_DETACH)) |
1055 | return -EINVAL; | |
1056 | ||
1057 | if (atomic_read(&mnt->mnt_count) != 2) | |
1058 | return -EBUSY; | |
1059 | ||
1060 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1061 | return -EAGAIN; | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * If we may have to abort operations to get out of this | |
1066 | * mount, and they will themselves hold resources we must | |
1067 | * allow the fs to do things. In the Unix tradition of | |
1068 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1069 | * might fail to complete on the first run through as other tasks | |
1070 | * must return, and the like. Thats for the mount program to worry | |
1071 | * about for the moment. | |
1072 | */ | |
1073 | ||
42faad99 AV |
1074 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
1075 | lock_kernel(); | |
1076 | sb->s_op->umount_begin(sb); | |
1077 | unlock_kernel(); | |
1078 | } | |
1da177e4 LT |
1079 | |
1080 | /* | |
1081 | * No sense to grab the lock for this test, but test itself looks | |
1082 | * somewhat bogus. Suggestions for better replacement? | |
1083 | * Ho-hum... In principle, we might treat that as umount + switch | |
1084 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1085 | * Actually it makes sense, especially if rootfs would contain a | |
1086 | * /reboot - static binary that would close all descriptors and | |
1087 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1088 | */ | |
6ac08c39 | 1089 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1090 | /* |
1091 | * Special case for "unmounting" root ... | |
1092 | * we just try to remount it readonly. | |
1093 | */ | |
1094 | down_write(&sb->s_umount); | |
1095 | if (!(sb->s_flags & MS_RDONLY)) { | |
1096 | lock_kernel(); | |
1da177e4 LT |
1097 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1098 | unlock_kernel(); | |
1099 | } | |
1100 | up_write(&sb->s_umount); | |
1101 | return retval; | |
1102 | } | |
1103 | ||
390c6843 | 1104 | down_write(&namespace_sem); |
1da177e4 | 1105 | spin_lock(&vfsmount_lock); |
5addc5dd | 1106 | event++; |
1da177e4 | 1107 | |
c35038be AV |
1108 | if (!(flags & MNT_DETACH)) |
1109 | shrink_submounts(mnt, &umount_list); | |
1110 | ||
1da177e4 | 1111 | retval = -EBUSY; |
a05964f3 | 1112 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1113 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1114 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1115 | retval = 0; |
1116 | } | |
1117 | spin_unlock(&vfsmount_lock); | |
1118 | if (retval) | |
1119 | security_sb_umount_busy(mnt); | |
390c6843 | 1120 | up_write(&namespace_sem); |
70fbcdf4 | 1121 | release_mounts(&umount_list); |
1da177e4 LT |
1122 | return retval; |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Now umount can handle mount points as well as block devices. | |
1127 | * This is important for filesystems which use unnamed block devices. | |
1128 | * | |
1129 | * We now support a flag for forced unmount like the other 'big iron' | |
1130 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1131 | */ | |
1132 | ||
bdc480e3 | 1133 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1134 | { |
2d8f3038 | 1135 | struct path path; |
1da177e4 LT |
1136 | int retval; |
1137 | ||
2d8f3038 | 1138 | retval = user_path(name, &path); |
1da177e4 LT |
1139 | if (retval) |
1140 | goto out; | |
1141 | retval = -EINVAL; | |
2d8f3038 | 1142 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1143 | goto dput_and_out; |
2d8f3038 | 1144 | if (!check_mnt(path.mnt)) |
1da177e4 LT |
1145 | goto dput_and_out; |
1146 | ||
1147 | retval = -EPERM; | |
1148 | if (!capable(CAP_SYS_ADMIN)) | |
1149 | goto dput_and_out; | |
1150 | ||
2d8f3038 | 1151 | retval = do_umount(path.mnt, flags); |
1da177e4 | 1152 | dput_and_out: |
429731b1 | 1153 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 AV |
1154 | dput(path.dentry); |
1155 | mntput_no_expire(path.mnt); | |
1da177e4 LT |
1156 | out: |
1157 | return retval; | |
1158 | } | |
1159 | ||
1160 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1161 | ||
1162 | /* | |
b58fed8b | 1163 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1164 | */ |
bdc480e3 | 1165 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1166 | { |
b58fed8b | 1167 | return sys_umount(name, 0); |
1da177e4 LT |
1168 | } |
1169 | ||
1170 | #endif | |
1171 | ||
2d92ab3c | 1172 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1173 | { |
1174 | if (capable(CAP_SYS_ADMIN)) | |
1175 | return 0; | |
1176 | return -EPERM; | |
1177 | #ifdef notyet | |
2d92ab3c | 1178 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1179 | return -EPERM; |
2d92ab3c | 1180 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1181 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1182 | return -EPERM; |
1183 | } | |
2d92ab3c | 1184 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1185 | return -EPERM; |
1186 | return 0; | |
1187 | #endif | |
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) { |
7ec02ef1 | 1206 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
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. | |
1430 | */ | |
0a0d8a46 | 1431 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1432 | { |
2d92ab3c | 1433 | struct vfsmount *m, *mnt = path->mnt; |
07b20889 RP |
1434 | int recurse = flag & MS_REC; |
1435 | int type = flag & ~MS_REC; | |
719f5d7f | 1436 | int err = 0; |
07b20889 | 1437 | |
ee6f9582 MS |
1438 | if (!capable(CAP_SYS_ADMIN)) |
1439 | return -EPERM; | |
1440 | ||
2d92ab3c | 1441 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1442 | return -EINVAL; |
1443 | ||
1444 | down_write(&namespace_sem); | |
719f5d7f MS |
1445 | if (type == MS_SHARED) { |
1446 | err = invent_group_ids(mnt, recurse); | |
1447 | if (err) | |
1448 | goto out_unlock; | |
1449 | } | |
1450 | ||
07b20889 RP |
1451 | spin_lock(&vfsmount_lock); |
1452 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1453 | change_mnt_propagation(m, type); | |
1454 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1455 | |
1456 | out_unlock: | |
07b20889 | 1457 | up_write(&namespace_sem); |
719f5d7f | 1458 | return err; |
07b20889 RP |
1459 | } |
1460 | ||
1da177e4 LT |
1461 | /* |
1462 | * do loopback mount. | |
1463 | */ | |
0a0d8a46 | 1464 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1465 | int recurse) |
1da177e4 | 1466 | { |
2d92ab3c | 1467 | struct path old_path; |
1da177e4 | 1468 | struct vfsmount *mnt = NULL; |
2d92ab3c | 1469 | int err = mount_is_safe(path); |
1da177e4 LT |
1470 | if (err) |
1471 | return err; | |
1472 | if (!old_name || !*old_name) | |
1473 | return -EINVAL; | |
2d92ab3c | 1474 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1475 | if (err) |
1476 | return err; | |
1477 | ||
390c6843 | 1478 | down_write(&namespace_sem); |
1da177e4 | 1479 | err = -EINVAL; |
2d92ab3c | 1480 | if (IS_MNT_UNBINDABLE(old_path.mnt)) |
4ac91378 | 1481 | goto out; |
9676f0c6 | 1482 | |
2d92ab3c | 1483 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
ccd48bc7 | 1484 | goto out; |
1da177e4 | 1485 | |
ccd48bc7 AV |
1486 | err = -ENOMEM; |
1487 | if (recurse) | |
2d92ab3c | 1488 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 | 1489 | else |
2d92ab3c | 1490 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 AV |
1491 | |
1492 | if (!mnt) | |
1493 | goto out; | |
1494 | ||
2d92ab3c | 1495 | err = graft_tree(mnt, path); |
ccd48bc7 | 1496 | if (err) { |
70fbcdf4 | 1497 | LIST_HEAD(umount_list); |
1da177e4 | 1498 | spin_lock(&vfsmount_lock); |
a05964f3 | 1499 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1500 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1501 | release_mounts(&umount_list); |
5b83d2c5 | 1502 | } |
1da177e4 | 1503 | |
ccd48bc7 | 1504 | out: |
390c6843 | 1505 | up_write(&namespace_sem); |
2d92ab3c | 1506 | path_put(&old_path); |
1da177e4 LT |
1507 | return err; |
1508 | } | |
1509 | ||
2e4b7fcd DH |
1510 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1511 | { | |
1512 | int error = 0; | |
1513 | int readonly_request = 0; | |
1514 | ||
1515 | if (ms_flags & MS_RDONLY) | |
1516 | readonly_request = 1; | |
1517 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1518 | return 0; | |
1519 | ||
1520 | if (readonly_request) | |
1521 | error = mnt_make_readonly(mnt); | |
1522 | else | |
1523 | __mnt_unmake_readonly(mnt); | |
1524 | return error; | |
1525 | } | |
1526 | ||
1da177e4 LT |
1527 | /* |
1528 | * change filesystem flags. dir should be a physical root of filesystem. | |
1529 | * If you've mounted a non-root directory somewhere and want to do remount | |
1530 | * on it - tough luck. | |
1531 | */ | |
0a0d8a46 | 1532 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1533 | void *data) |
1534 | { | |
1535 | int err; | |
2d92ab3c | 1536 | struct super_block *sb = path->mnt->mnt_sb; |
1da177e4 LT |
1537 | |
1538 | if (!capable(CAP_SYS_ADMIN)) | |
1539 | return -EPERM; | |
1540 | ||
2d92ab3c | 1541 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1542 | return -EINVAL; |
1543 | ||
2d92ab3c | 1544 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1545 | return -EINVAL; |
1546 | ||
1547 | down_write(&sb->s_umount); | |
2e4b7fcd | 1548 | if (flags & MS_BIND) |
2d92ab3c | 1549 | err = change_mount_flags(path->mnt, flags); |
2e4b7fcd DH |
1550 | else |
1551 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1552 | if (!err) |
2d92ab3c | 1553 | path->mnt->mnt_flags = mnt_flags; |
1da177e4 | 1554 | up_write(&sb->s_umount); |
0e55a7cc | 1555 | if (!err) { |
2d92ab3c | 1556 | security_sb_post_remount(path->mnt, flags, data); |
0e55a7cc DW |
1557 | |
1558 | spin_lock(&vfsmount_lock); | |
1559 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1560 | spin_unlock(&vfsmount_lock); | |
1561 | } | |
1da177e4 LT |
1562 | return err; |
1563 | } | |
1564 | ||
9676f0c6 RP |
1565 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1566 | { | |
1567 | struct vfsmount *p; | |
1568 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1569 | if (IS_MNT_UNBINDABLE(p)) | |
1570 | return 1; | |
1571 | } | |
1572 | return 0; | |
1573 | } | |
1574 | ||
0a0d8a46 | 1575 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1576 | { |
2d92ab3c | 1577 | struct path old_path, parent_path; |
1da177e4 LT |
1578 | struct vfsmount *p; |
1579 | int err = 0; | |
1580 | if (!capable(CAP_SYS_ADMIN)) | |
1581 | return -EPERM; | |
1582 | if (!old_name || !*old_name) | |
1583 | return -EINVAL; | |
2d92ab3c | 1584 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1585 | if (err) |
1586 | return err; | |
1587 | ||
390c6843 | 1588 | down_write(&namespace_sem); |
2d92ab3c AV |
1589 | while (d_mountpoint(path->dentry) && |
1590 | follow_down(&path->mnt, &path->dentry)) | |
1da177e4 LT |
1591 | ; |
1592 | err = -EINVAL; | |
2d92ab3c | 1593 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
1da177e4 LT |
1594 | goto out; |
1595 | ||
1596 | err = -ENOENT; | |
2d92ab3c AV |
1597 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1598 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1599 | goto out1; |
1600 | ||
2d92ab3c | 1601 | if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry)) |
21444403 | 1602 | goto out1; |
1da177e4 LT |
1603 | |
1604 | err = -EINVAL; | |
2d92ab3c | 1605 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1606 | goto out1; |
1da177e4 | 1607 | |
2d92ab3c | 1608 | if (old_path.mnt == old_path.mnt->mnt_parent) |
21444403 | 1609 | goto out1; |
1da177e4 | 1610 | |
2d92ab3c AV |
1611 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1612 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1613 | goto out1; |
1614 | /* | |
1615 | * Don't move a mount residing in a shared parent. | |
1616 | */ | |
2d92ab3c AV |
1617 | if (old_path.mnt->mnt_parent && |
1618 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
21444403 | 1619 | goto out1; |
9676f0c6 RP |
1620 | /* |
1621 | * Don't move a mount tree containing unbindable mounts to a destination | |
1622 | * mount which is shared. | |
1623 | */ | |
2d92ab3c AV |
1624 | if (IS_MNT_SHARED(path->mnt) && |
1625 | tree_contains_unbindable(old_path.mnt)) | |
9676f0c6 | 1626 | goto out1; |
1da177e4 | 1627 | err = -ELOOP; |
2d92ab3c AV |
1628 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1629 | if (p == old_path.mnt) | |
21444403 | 1630 | goto out1; |
1da177e4 | 1631 | |
2d92ab3c | 1632 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); |
4ac91378 | 1633 | if (err) |
21444403 | 1634 | goto out1; |
1da177e4 LT |
1635 | |
1636 | /* if the mount is moved, it should no longer be expire | |
1637 | * automatically */ | |
2d92ab3c | 1638 | list_del_init(&old_path.mnt->mnt_expire); |
1da177e4 | 1639 | out1: |
2d92ab3c | 1640 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1641 | out: |
390c6843 | 1642 | up_write(&namespace_sem); |
1da177e4 | 1643 | if (!err) |
1a390689 | 1644 | path_put(&parent_path); |
2d92ab3c | 1645 | path_put(&old_path); |
1da177e4 LT |
1646 | return err; |
1647 | } | |
1648 | ||
1649 | /* | |
1650 | * create a new mount for userspace and request it to be added into the | |
1651 | * namespace's tree | |
1652 | */ | |
0a0d8a46 | 1653 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1654 | int mnt_flags, char *name, void *data) |
1655 | { | |
1656 | struct vfsmount *mnt; | |
1657 | ||
1658 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1659 | return -EINVAL; | |
1660 | ||
1661 | /* we need capabilities... */ | |
1662 | if (!capable(CAP_SYS_ADMIN)) | |
1663 | return -EPERM; | |
1664 | ||
1665 | mnt = do_kern_mount(type, flags, name, data); | |
1666 | if (IS_ERR(mnt)) | |
1667 | return PTR_ERR(mnt); | |
1668 | ||
2d92ab3c | 1669 | return do_add_mount(mnt, path, mnt_flags, NULL); |
1da177e4 LT |
1670 | } |
1671 | ||
1672 | /* | |
1673 | * add a mount into a namespace's mount tree | |
1674 | * - provide the option of adding the new mount to an expiration list | |
1675 | */ | |
8d66bf54 | 1676 | int do_add_mount(struct vfsmount *newmnt, struct path *path, |
1da177e4 LT |
1677 | int mnt_flags, struct list_head *fslist) |
1678 | { | |
1679 | int err; | |
1680 | ||
390c6843 | 1681 | down_write(&namespace_sem); |
1da177e4 | 1682 | /* Something was mounted here while we slept */ |
8d66bf54 AV |
1683 | while (d_mountpoint(path->dentry) && |
1684 | follow_down(&path->mnt, &path->dentry)) | |
1da177e4 LT |
1685 | ; |
1686 | err = -EINVAL; | |
8d66bf54 | 1687 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1688 | goto unlock; |
1689 | ||
1690 | /* Refuse the same filesystem on the same mount point */ | |
1691 | err = -EBUSY; | |
8d66bf54 AV |
1692 | if (path->mnt->mnt_sb == newmnt->mnt_sb && |
1693 | path->mnt->mnt_root == path->dentry) | |
1da177e4 LT |
1694 | goto unlock; |
1695 | ||
1696 | err = -EINVAL; | |
1697 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1698 | goto unlock; | |
1699 | ||
1700 | newmnt->mnt_flags = mnt_flags; | |
8d66bf54 | 1701 | if ((err = graft_tree(newmnt, path))) |
5b83d2c5 | 1702 | goto unlock; |
1da177e4 | 1703 | |
6758f953 | 1704 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1705 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1706 | |
390c6843 | 1707 | up_write(&namespace_sem); |
5b83d2c5 | 1708 | return 0; |
1da177e4 LT |
1709 | |
1710 | unlock: | |
390c6843 | 1711 | up_write(&namespace_sem); |
1da177e4 LT |
1712 | mntput(newmnt); |
1713 | return err; | |
1714 | } | |
1715 | ||
1716 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1717 | ||
1718 | /* | |
1719 | * process a list of expirable mountpoints with the intent of discarding any | |
1720 | * mountpoints that aren't in use and haven't been touched since last we came | |
1721 | * here | |
1722 | */ | |
1723 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1724 | { | |
1da177e4 LT |
1725 | struct vfsmount *mnt, *next; |
1726 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1727 | LIST_HEAD(umounts); |
1da177e4 LT |
1728 | |
1729 | if (list_empty(mounts)) | |
1730 | return; | |
1731 | ||
bcc5c7d2 | 1732 | down_write(&namespace_sem); |
1da177e4 LT |
1733 | spin_lock(&vfsmount_lock); |
1734 | ||
1735 | /* extract from the expiration list every vfsmount that matches the | |
1736 | * following criteria: | |
1737 | * - only referenced by its parent vfsmount | |
1738 | * - still marked for expiry (marked on the last call here; marks are | |
1739 | * cleared by mntput()) | |
1740 | */ | |
55e700b9 | 1741 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1742 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1743 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1744 | continue; |
55e700b9 | 1745 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1746 | } |
bcc5c7d2 AV |
1747 | while (!list_empty(&graveyard)) { |
1748 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1749 | touch_mnt_namespace(mnt->mnt_ns); | |
1750 | umount_tree(mnt, 1, &umounts); | |
1751 | } | |
5528f911 | 1752 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1753 | up_write(&namespace_sem); |
1754 | ||
1755 | release_mounts(&umounts); | |
5528f911 TM |
1756 | } |
1757 | ||
1758 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1759 | ||
1760 | /* | |
1761 | * Ripoff of 'select_parent()' | |
1762 | * | |
1763 | * search the list of submounts for a given mountpoint, and move any | |
1764 | * shrinkable submounts to the 'graveyard' list. | |
1765 | */ | |
1766 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1767 | { | |
1768 | struct vfsmount *this_parent = parent; | |
1769 | struct list_head *next; | |
1770 | int found = 0; | |
1771 | ||
1772 | repeat: | |
1773 | next = this_parent->mnt_mounts.next; | |
1774 | resume: | |
1775 | while (next != &this_parent->mnt_mounts) { | |
1776 | struct list_head *tmp = next; | |
1777 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1778 | ||
1779 | next = tmp->next; | |
1780 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1781 | continue; |
5528f911 TM |
1782 | /* |
1783 | * Descend a level if the d_mounts list is non-empty. | |
1784 | */ | |
1785 | if (!list_empty(&mnt->mnt_mounts)) { | |
1786 | this_parent = mnt; | |
1787 | goto repeat; | |
1788 | } | |
1da177e4 | 1789 | |
5528f911 | 1790 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1791 | list_move_tail(&mnt->mnt_expire, graveyard); |
1792 | found++; | |
1793 | } | |
1da177e4 | 1794 | } |
5528f911 TM |
1795 | /* |
1796 | * All done at this level ... ascend and resume the search | |
1797 | */ | |
1798 | if (this_parent != parent) { | |
1799 | next = this_parent->mnt_child.next; | |
1800 | this_parent = this_parent->mnt_parent; | |
1801 | goto resume; | |
1802 | } | |
1803 | return found; | |
1804 | } | |
1805 | ||
1806 | /* | |
1807 | * process a list of expirable mountpoints with the intent of discarding any | |
1808 | * submounts of a specific parent mountpoint | |
1809 | */ | |
c35038be | 1810 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1811 | { |
1812 | LIST_HEAD(graveyard); | |
c35038be | 1813 | struct vfsmount *m; |
5528f911 | 1814 | |
5528f911 | 1815 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1816 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1817 | while (!list_empty(&graveyard)) { |
c35038be | 1818 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 | 1819 | mnt_expire); |
afef80b3 EB |
1820 | touch_mnt_namespace(m->mnt_ns); |
1821 | umount_tree(m, 1, umounts); | |
bcc5c7d2 AV |
1822 | } |
1823 | } | |
1da177e4 LT |
1824 | } |
1825 | ||
1da177e4 LT |
1826 | /* |
1827 | * Some copy_from_user() implementations do not return the exact number of | |
1828 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1829 | * Note that this function differs from copy_from_user() in that it will oops | |
1830 | * on bad values of `to', rather than returning a short copy. | |
1831 | */ | |
b58fed8b RP |
1832 | static long exact_copy_from_user(void *to, const void __user * from, |
1833 | unsigned long n) | |
1da177e4 LT |
1834 | { |
1835 | char *t = to; | |
1836 | const char __user *f = from; | |
1837 | char c; | |
1838 | ||
1839 | if (!access_ok(VERIFY_READ, from, n)) | |
1840 | return n; | |
1841 | ||
1842 | while (n) { | |
1843 | if (__get_user(c, f)) { | |
1844 | memset(t, 0, n); | |
1845 | break; | |
1846 | } | |
1847 | *t++ = c; | |
1848 | f++; | |
1849 | n--; | |
1850 | } | |
1851 | return n; | |
1852 | } | |
1853 | ||
b58fed8b | 1854 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1855 | { |
1856 | int i; | |
1857 | unsigned long page; | |
1858 | unsigned long size; | |
b58fed8b | 1859 | |
1da177e4 LT |
1860 | *where = 0; |
1861 | if (!data) | |
1862 | return 0; | |
1863 | ||
1864 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1865 | return -ENOMEM; | |
1866 | ||
1867 | /* We only care that *some* data at the address the user | |
1868 | * gave us is valid. Just in case, we'll zero | |
1869 | * the remainder of the page. | |
1870 | */ | |
1871 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1872 | size = TASK_SIZE - (unsigned long)data; | |
1873 | if (size > PAGE_SIZE) | |
1874 | size = PAGE_SIZE; | |
1875 | ||
1876 | i = size - exact_copy_from_user((void *)page, data, size); | |
1877 | if (!i) { | |
b58fed8b | 1878 | free_page(page); |
1da177e4 LT |
1879 | return -EFAULT; |
1880 | } | |
1881 | if (i != PAGE_SIZE) | |
1882 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1883 | *where = page; | |
1884 | return 0; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1889 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1890 | * | |
1891 | * data is a (void *) that can point to any structure up to | |
1892 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1893 | * information (or be NULL). | |
1894 | * | |
1895 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1896 | * When the flags word was introduced its top half was required | |
1897 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1898 | * Therefore, if this magic number is present, it carries no information | |
1899 | * and must be discarded. | |
1900 | */ | |
b58fed8b | 1901 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1902 | unsigned long flags, void *data_page) |
1903 | { | |
2d92ab3c | 1904 | struct path path; |
1da177e4 LT |
1905 | int retval = 0; |
1906 | int mnt_flags = 0; | |
1907 | ||
1908 | /* Discard magic */ | |
1909 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1910 | flags &= ~MS_MGC_MSK; | |
1911 | ||
1912 | /* Basic sanity checks */ | |
1913 | ||
1914 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1915 | return -EINVAL; | |
1916 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1917 | return -EINVAL; | |
1918 | ||
1919 | if (data_page) | |
1920 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1921 | ||
0a1c01c9 MG |
1922 | /* Default to relatime */ |
1923 | mnt_flags |= MNT_RELATIME; | |
1924 | ||
1da177e4 LT |
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; | |
d0adde57 MG |
1936 | if (flags & MS_STRICTATIME) |
1937 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
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 | | |
d0adde57 MG |
1942 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
1943 | MS_STRICTATIME); | |
1da177e4 LT |
1944 | |
1945 | /* ... and get the mountpoint */ | |
2d92ab3c | 1946 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1947 | if (retval) |
1948 | return retval; | |
1949 | ||
2d92ab3c | 1950 | retval = security_sb_mount(dev_name, &path, |
b5266eb4 | 1951 | type_page, flags, data_page); |
1da177e4 LT |
1952 | if (retval) |
1953 | goto dput_out; | |
1954 | ||
1955 | if (flags & MS_REMOUNT) | |
2d92ab3c | 1956 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
1957 | data_page); |
1958 | else if (flags & MS_BIND) | |
2d92ab3c | 1959 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 1960 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 1961 | retval = do_change_type(&path, flags); |
1da177e4 | 1962 | else if (flags & MS_MOVE) |
2d92ab3c | 1963 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 1964 | else |
2d92ab3c | 1965 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
1966 | dev_name, data_page); |
1967 | dput_out: | |
2d92ab3c | 1968 | path_put(&path); |
1da177e4 LT |
1969 | return retval; |
1970 | } | |
1971 | ||
741a2951 JD |
1972 | /* |
1973 | * Allocate a new namespace structure and populate it with contents | |
1974 | * copied from the namespace of the passed in task structure. | |
1975 | */ | |
e3222c4e | 1976 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1977 | struct fs_struct *fs) |
1da177e4 | 1978 | { |
6b3286ed | 1979 | struct mnt_namespace *new_ns; |
7f2da1e7 | 1980 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
1da177e4 LT |
1981 | struct vfsmount *p, *q; |
1982 | ||
6b3286ed | 1983 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1984 | if (!new_ns) |
467e9f4b | 1985 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1986 | |
1987 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1988 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1989 | init_waitqueue_head(&new_ns->poll); |
1990 | new_ns->event = 0; | |
1da177e4 | 1991 | |
390c6843 | 1992 | down_write(&namespace_sem); |
1da177e4 | 1993 | /* First pass: copy the tree topology */ |
6b3286ed | 1994 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1995 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1996 | if (!new_ns->root) { |
390c6843 | 1997 | up_write(&namespace_sem); |
1da177e4 | 1998 | kfree(new_ns); |
5cc4a034 | 1999 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
2000 | } |
2001 | spin_lock(&vfsmount_lock); | |
2002 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
2003 | spin_unlock(&vfsmount_lock); | |
2004 | ||
2005 | /* | |
2006 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2007 | * as belonging to new namespace. We have already acquired a private | |
2008 | * fs_struct, so tsk->fs->lock is not needed. | |
2009 | */ | |
6b3286ed | 2010 | p = mnt_ns->root; |
1da177e4 LT |
2011 | q = new_ns->root; |
2012 | while (p) { | |
6b3286ed | 2013 | q->mnt_ns = new_ns; |
1da177e4 | 2014 | if (fs) { |
6ac08c39 | 2015 | if (p == fs->root.mnt) { |
1da177e4 | 2016 | rootmnt = p; |
6ac08c39 | 2017 | fs->root.mnt = mntget(q); |
1da177e4 | 2018 | } |
6ac08c39 | 2019 | if (p == fs->pwd.mnt) { |
1da177e4 | 2020 | pwdmnt = p; |
6ac08c39 | 2021 | fs->pwd.mnt = mntget(q); |
1da177e4 | 2022 | } |
1da177e4 | 2023 | } |
6b3286ed | 2024 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
2025 | q = next_mnt(q, new_ns->root); |
2026 | } | |
390c6843 | 2027 | up_write(&namespace_sem); |
1da177e4 | 2028 | |
1da177e4 LT |
2029 | if (rootmnt) |
2030 | mntput(rootmnt); | |
2031 | if (pwdmnt) | |
2032 | mntput(pwdmnt); | |
1da177e4 | 2033 | |
741a2951 JD |
2034 | return new_ns; |
2035 | } | |
2036 | ||
213dd266 | 2037 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2038 | struct fs_struct *new_fs) |
741a2951 | 2039 | { |
6b3286ed | 2040 | struct mnt_namespace *new_ns; |
741a2951 | 2041 | |
e3222c4e | 2042 | BUG_ON(!ns); |
6b3286ed | 2043 | get_mnt_ns(ns); |
741a2951 JD |
2044 | |
2045 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2046 | return ns; |
741a2951 | 2047 | |
e3222c4e | 2048 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2049 | |
6b3286ed | 2050 | put_mnt_ns(ns); |
e3222c4e | 2051 | return new_ns; |
1da177e4 LT |
2052 | } |
2053 | ||
bdc480e3 HC |
2054 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2055 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 LT |
2056 | { |
2057 | int retval; | |
2058 | unsigned long data_page; | |
2059 | unsigned long type_page; | |
2060 | unsigned long dev_page; | |
2061 | char *dir_page; | |
2062 | ||
b58fed8b | 2063 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
2064 | if (retval < 0) |
2065 | return retval; | |
2066 | ||
2067 | dir_page = getname(dir_name); | |
2068 | retval = PTR_ERR(dir_page); | |
2069 | if (IS_ERR(dir_page)) | |
2070 | goto out1; | |
2071 | ||
b58fed8b | 2072 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2073 | if (retval < 0) |
2074 | goto out2; | |
2075 | ||
b58fed8b | 2076 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2077 | if (retval < 0) |
2078 | goto out3; | |
2079 | ||
2080 | lock_kernel(); | |
b58fed8b RP |
2081 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2082 | flags, (void *)data_page); | |
1da177e4 LT |
2083 | unlock_kernel(); |
2084 | free_page(data_page); | |
2085 | ||
2086 | out3: | |
2087 | free_page(dev_page); | |
2088 | out2: | |
2089 | putname(dir_page); | |
2090 | out1: | |
2091 | free_page(type_page); | |
2092 | return retval; | |
2093 | } | |
2094 | ||
1da177e4 LT |
2095 | /* |
2096 | * pivot_root Semantics: | |
2097 | * Moves the root file system of the current process to the directory put_old, | |
2098 | * makes new_root as the new root file system of the current process, and sets | |
2099 | * root/cwd of all processes which had them on the current root to new_root. | |
2100 | * | |
2101 | * Restrictions: | |
2102 | * The new_root and put_old must be directories, and must not be on the | |
2103 | * same file system as the current process root. The put_old must be | |
2104 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2105 | * pointed to by put_old must yield the same directory as new_root. No other | |
2106 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2107 | * | |
4a0d11fa NB |
2108 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2109 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2110 | * in this situation. | |
2111 | * | |
1da177e4 LT |
2112 | * Notes: |
2113 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2114 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2115 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2116 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2117 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2118 | * first. | |
2119 | */ | |
3480b257 HC |
2120 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2121 | const char __user *, put_old) | |
1da177e4 LT |
2122 | { |
2123 | struct vfsmount *tmp; | |
2d8f3038 | 2124 | struct path new, old, parent_path, root_parent, root; |
1da177e4 LT |
2125 | int error; |
2126 | ||
2127 | if (!capable(CAP_SYS_ADMIN)) | |
2128 | return -EPERM; | |
2129 | ||
2d8f3038 | 2130 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2131 | if (error) |
2132 | goto out0; | |
2133 | error = -EINVAL; | |
2d8f3038 | 2134 | if (!check_mnt(new.mnt)) |
1da177e4 LT |
2135 | goto out1; |
2136 | ||
2d8f3038 | 2137 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2138 | if (error) |
2139 | goto out1; | |
2140 | ||
2d8f3038 | 2141 | error = security_sb_pivotroot(&old, &new); |
1da177e4 | 2142 | if (error) { |
2d8f3038 | 2143 | path_put(&old); |
1da177e4 LT |
2144 | goto out1; |
2145 | } | |
2146 | ||
2147 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2148 | root = current->fs->root; |
6ac08c39 | 2149 | path_get(¤t->fs->root); |
1da177e4 | 2150 | read_unlock(¤t->fs->lock); |
390c6843 | 2151 | down_write(&namespace_sem); |
2d8f3038 | 2152 | mutex_lock(&old.dentry->d_inode->i_mutex); |
1da177e4 | 2153 | error = -EINVAL; |
2d8f3038 AV |
2154 | if (IS_MNT_SHARED(old.mnt) || |
2155 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
8c3ee42e | 2156 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2157 | goto out2; |
8c3ee42e | 2158 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2159 | goto out2; |
2160 | error = -ENOENT; | |
2d8f3038 | 2161 | if (IS_DEADDIR(new.dentry->d_inode)) |
1da177e4 | 2162 | goto out2; |
2d8f3038 | 2163 | if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry)) |
1da177e4 | 2164 | goto out2; |
2d8f3038 | 2165 | if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry)) |
1da177e4 LT |
2166 | goto out2; |
2167 | error = -EBUSY; | |
2d8f3038 AV |
2168 | if (new.mnt == root.mnt || |
2169 | old.mnt == root.mnt) | |
1da177e4 LT |
2170 | goto out2; /* loop, on the same file system */ |
2171 | error = -EINVAL; | |
8c3ee42e | 2172 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2173 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2174 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2175 | goto out2; /* not attached */ |
2d8f3038 | 2176 | if (new.mnt->mnt_root != new.dentry) |
1da177e4 | 2177 | goto out2; /* not a mountpoint */ |
2d8f3038 | 2178 | if (new.mnt->mnt_parent == new.mnt) |
0bb6fcc1 | 2179 | goto out2; /* not attached */ |
4ac91378 | 2180 | /* make sure we can reach put_old from new_root */ |
2d8f3038 | 2181 | tmp = old.mnt; |
1da177e4 | 2182 | spin_lock(&vfsmount_lock); |
2d8f3038 | 2183 | if (tmp != new.mnt) { |
1da177e4 LT |
2184 | for (;;) { |
2185 | if (tmp->mnt_parent == tmp) | |
2186 | goto out3; /* already mounted on put_old */ | |
2d8f3038 | 2187 | if (tmp->mnt_parent == new.mnt) |
1da177e4 LT |
2188 | break; |
2189 | tmp = tmp->mnt_parent; | |
2190 | } | |
2d8f3038 | 2191 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) |
1da177e4 | 2192 | goto out3; |
2d8f3038 | 2193 | } else if (!is_subdir(old.dentry, new.dentry)) |
1da177e4 | 2194 | goto out3; |
2d8f3038 | 2195 | detach_mnt(new.mnt, &parent_path); |
8c3ee42e | 2196 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2197 | /* mount old root on put_old */ |
2d8f3038 | 2198 | attach_mnt(root.mnt, &old); |
4ac91378 | 2199 | /* mount new_root on / */ |
2d8f3038 | 2200 | attach_mnt(new.mnt, &root_parent); |
6b3286ed | 2201 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2202 | spin_unlock(&vfsmount_lock); |
2d8f3038 AV |
2203 | chroot_fs_refs(&root, &new); |
2204 | security_sb_post_pivotroot(&root, &new); | |
1da177e4 | 2205 | error = 0; |
1a390689 AV |
2206 | path_put(&root_parent); |
2207 | path_put(&parent_path); | |
1da177e4 | 2208 | out2: |
2d8f3038 | 2209 | mutex_unlock(&old.dentry->d_inode->i_mutex); |
390c6843 | 2210 | up_write(&namespace_sem); |
8c3ee42e | 2211 | path_put(&root); |
2d8f3038 | 2212 | path_put(&old); |
1da177e4 | 2213 | out1: |
2d8f3038 | 2214 | path_put(&new); |
1da177e4 | 2215 | out0: |
1da177e4 LT |
2216 | return error; |
2217 | out3: | |
2218 | spin_unlock(&vfsmount_lock); | |
2219 | goto out2; | |
2220 | } | |
2221 | ||
2222 | static void __init init_mount_tree(void) | |
2223 | { | |
2224 | struct vfsmount *mnt; | |
6b3286ed | 2225 | struct mnt_namespace *ns; |
ac748a09 | 2226 | struct path root; |
1da177e4 LT |
2227 | |
2228 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2229 | if (IS_ERR(mnt)) | |
2230 | panic("Can't create rootfs"); | |
6b3286ed KK |
2231 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2232 | if (!ns) | |
1da177e4 | 2233 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2234 | atomic_set(&ns->count, 1); |
2235 | INIT_LIST_HEAD(&ns->list); | |
2236 | init_waitqueue_head(&ns->poll); | |
2237 | ns->event = 0; | |
2238 | list_add(&mnt->mnt_list, &ns->list); | |
2239 | ns->root = mnt; | |
2240 | mnt->mnt_ns = ns; | |
2241 | ||
2242 | init_task.nsproxy->mnt_ns = ns; | |
2243 | get_mnt_ns(ns); | |
2244 | ||
ac748a09 JB |
2245 | root.mnt = ns->root; |
2246 | root.dentry = ns->root->mnt_root; | |
2247 | ||
2248 | set_fs_pwd(current->fs, &root); | |
2249 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2250 | } |
2251 | ||
74bf17cf | 2252 | void __init mnt_init(void) |
1da177e4 | 2253 | { |
13f14b4d | 2254 | unsigned u; |
15a67dd8 | 2255 | int err; |
1da177e4 | 2256 | |
390c6843 RP |
2257 | init_rwsem(&namespace_sem); |
2258 | ||
1da177e4 | 2259 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2260 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2261 | |
b58fed8b | 2262 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2263 | |
2264 | if (!mount_hashtable) | |
2265 | panic("Failed to allocate mount hash table\n"); | |
2266 | ||
13f14b4d ED |
2267 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2268 | ||
2269 | for (u = 0; u < HASH_SIZE; u++) | |
2270 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2271 | |
15a67dd8 RD |
2272 | err = sysfs_init(); |
2273 | if (err) | |
2274 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2275 | __func__, err); |
00d26666 GKH |
2276 | fs_kobj = kobject_create_and_add("fs", NULL); |
2277 | if (!fs_kobj) | |
8e24eea7 | 2278 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2279 | init_rootfs(); |
2280 | init_mount_tree(); | |
2281 | } | |
2282 | ||
6b3286ed | 2283 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2284 | { |
6b3286ed | 2285 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2286 | LIST_HEAD(umount_list); |
6b3286ed | 2287 | ns->root = NULL; |
1ce88cf4 | 2288 | spin_unlock(&vfsmount_lock); |
390c6843 | 2289 | down_write(&namespace_sem); |
1da177e4 | 2290 | spin_lock(&vfsmount_lock); |
a05964f3 | 2291 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2292 | spin_unlock(&vfsmount_lock); |
390c6843 | 2293 | up_write(&namespace_sem); |
70fbcdf4 | 2294 | release_mounts(&umount_list); |
6b3286ed | 2295 | kfree(ns); |
1da177e4 | 2296 | } |