]>
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 LT |
17 | #include <linux/quotaops.h> |
18 | #include <linux/acct.h> | |
16f7e0fe | 19 | #include <linux/capability.h> |
3d733633 | 20 | #include <linux/cpumask.h> |
1da177e4 | 21 | #include <linux/module.h> |
f20a9ead | 22 | #include <linux/sysfs.h> |
1da177e4 | 23 | #include <linux/seq_file.h> |
6b3286ed | 24 | #include <linux/mnt_namespace.h> |
1da177e4 LT |
25 | #include <linux/namei.h> |
26 | #include <linux/security.h> | |
27 | #include <linux/mount.h> | |
07f3f05c | 28 | #include <linux/ramfs.h> |
13f14b4d | 29 | #include <linux/log2.h> |
73cd49ec | 30 | #include <linux/idr.h> |
1da177e4 LT |
31 | #include <asm/uaccess.h> |
32 | #include <asm/unistd.h> | |
07b20889 | 33 | #include "pnode.h" |
948730b0 | 34 | #include "internal.h" |
1da177e4 | 35 | |
13f14b4d ED |
36 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
37 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
38 | ||
1da177e4 | 39 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
40 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
41 | ||
42 | static int event; | |
73cd49ec | 43 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 44 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 45 | |
fa3536cc | 46 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 47 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 48 | static struct rw_semaphore namespace_sem; |
1da177e4 | 49 | |
f87fd4c2 | 50 | /* /sys/fs */ |
00d26666 GKH |
51 | struct kobject *fs_kobj; |
52 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 53 | |
1da177e4 LT |
54 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
55 | { | |
b58fed8b RP |
56 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
57 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
58 | tmp = tmp + (tmp >> HASH_SHIFT); |
59 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
60 | } |
61 | ||
3d733633 DH |
62 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
63 | ||
73cd49ec MS |
64 | /* allocation is serialized by namespace_sem */ |
65 | static int mnt_alloc_id(struct vfsmount *mnt) | |
66 | { | |
67 | int res; | |
68 | ||
69 | retry: | |
70 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
71 | spin_lock(&vfsmount_lock); | |
72 | res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); | |
73 | spin_unlock(&vfsmount_lock); | |
74 | if (res == -EAGAIN) | |
75 | goto retry; | |
76 | ||
77 | return res; | |
78 | } | |
79 | ||
80 | static void mnt_free_id(struct vfsmount *mnt) | |
81 | { | |
82 | spin_lock(&vfsmount_lock); | |
83 | ida_remove(&mnt_id_ida, mnt->mnt_id); | |
84 | spin_unlock(&vfsmount_lock); | |
85 | } | |
86 | ||
719f5d7f MS |
87 | /* |
88 | * Allocate a new peer group ID | |
89 | * | |
90 | * mnt_group_ida is protected by namespace_sem | |
91 | */ | |
92 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
93 | { | |
94 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) | |
95 | return -ENOMEM; | |
96 | ||
97 | return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id); | |
98 | } | |
99 | ||
100 | /* | |
101 | * Release a peer group ID | |
102 | */ | |
103 | void mnt_release_group_id(struct vfsmount *mnt) | |
104 | { | |
105 | ida_remove(&mnt_group_ida, mnt->mnt_group_id); | |
106 | mnt->mnt_group_id = 0; | |
107 | } | |
108 | ||
1da177e4 LT |
109 | struct vfsmount *alloc_vfsmnt(const char *name) |
110 | { | |
c3762229 | 111 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 112 | if (mnt) { |
73cd49ec MS |
113 | int err; |
114 | ||
115 | err = mnt_alloc_id(mnt); | |
116 | if (err) { | |
117 | kmem_cache_free(mnt_cache, mnt); | |
118 | return NULL; | |
119 | } | |
120 | ||
b58fed8b | 121 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
122 | INIT_LIST_HEAD(&mnt->mnt_hash); |
123 | INIT_LIST_HEAD(&mnt->mnt_child); | |
124 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
125 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 126 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 127 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
128 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
129 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
3d733633 | 130 | atomic_set(&mnt->__mnt_writers, 0); |
1da177e4 | 131 | if (name) { |
b58fed8b | 132 | int size = strlen(name) + 1; |
1da177e4 LT |
133 | char *newname = kmalloc(size, GFP_KERNEL); |
134 | if (newname) { | |
135 | memcpy(newname, name, size); | |
136 | mnt->mnt_devname = newname; | |
137 | } | |
138 | } | |
139 | } | |
140 | return mnt; | |
141 | } | |
142 | ||
3d733633 DH |
143 | /* |
144 | * Most r/o checks on a fs are for operations that take | |
145 | * discrete amounts of time, like a write() or unlink(). | |
146 | * We must keep track of when those operations start | |
147 | * (for permission checks) and when they end, so that | |
148 | * we can determine when writes are able to occur to | |
149 | * a filesystem. | |
150 | */ | |
151 | /* | |
152 | * __mnt_is_readonly: check whether a mount is read-only | |
153 | * @mnt: the mount to check for its write status | |
154 | * | |
155 | * This shouldn't be used directly ouside of the VFS. | |
156 | * It does not guarantee that the filesystem will stay | |
157 | * r/w, just that it is right *now*. This can not and | |
158 | * should not be used in place of IS_RDONLY(inode). | |
159 | * mnt_want/drop_write() will _keep_ the filesystem | |
160 | * r/w. | |
161 | */ | |
162 | int __mnt_is_readonly(struct vfsmount *mnt) | |
163 | { | |
2e4b7fcd DH |
164 | if (mnt->mnt_flags & MNT_READONLY) |
165 | return 1; | |
166 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
167 | return 1; | |
168 | return 0; | |
3d733633 DH |
169 | } |
170 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
171 | ||
172 | struct mnt_writer { | |
173 | /* | |
174 | * If holding multiple instances of this lock, they | |
175 | * must be ordered by cpu number. | |
176 | */ | |
177 | spinlock_t lock; | |
178 | struct lock_class_key lock_class; /* compiles out with !lockdep */ | |
179 | unsigned long count; | |
180 | struct vfsmount *mnt; | |
181 | } ____cacheline_aligned_in_smp; | |
182 | static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); | |
183 | ||
184 | static int __init init_mnt_writers(void) | |
185 | { | |
186 | int cpu; | |
187 | for_each_possible_cpu(cpu) { | |
188 | struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); | |
189 | spin_lock_init(&writer->lock); | |
190 | lockdep_set_class(&writer->lock, &writer->lock_class); | |
191 | writer->count = 0; | |
192 | } | |
193 | return 0; | |
194 | } | |
195 | fs_initcall(init_mnt_writers); | |
196 | ||
197 | static void unlock_mnt_writers(void) | |
198 | { | |
199 | int cpu; | |
200 | struct mnt_writer *cpu_writer; | |
201 | ||
202 | for_each_possible_cpu(cpu) { | |
203 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
204 | spin_unlock(&cpu_writer->lock); | |
205 | } | |
206 | } | |
207 | ||
208 | static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) | |
209 | { | |
210 | if (!cpu_writer->mnt) | |
211 | return; | |
212 | /* | |
213 | * This is in case anyone ever leaves an invalid, | |
214 | * old ->mnt and a count of 0. | |
215 | */ | |
216 | if (!cpu_writer->count) | |
217 | return; | |
218 | atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); | |
219 | cpu_writer->count = 0; | |
220 | } | |
221 | /* | |
222 | * must hold cpu_writer->lock | |
223 | */ | |
224 | static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, | |
225 | struct vfsmount *mnt) | |
226 | { | |
227 | if (cpu_writer->mnt == mnt) | |
228 | return; | |
229 | __clear_mnt_count(cpu_writer); | |
230 | cpu_writer->mnt = mnt; | |
231 | } | |
232 | ||
8366025e DH |
233 | /* |
234 | * Most r/o checks on a fs are for operations that take | |
235 | * discrete amounts of time, like a write() or unlink(). | |
236 | * We must keep track of when those operations start | |
237 | * (for permission checks) and when they end, so that | |
238 | * we can determine when writes are able to occur to | |
239 | * a filesystem. | |
240 | */ | |
241 | /** | |
242 | * mnt_want_write - get write access to a mount | |
243 | * @mnt: the mount on which to take a write | |
244 | * | |
245 | * This tells the low-level filesystem that a write is | |
246 | * about to be performed to it, and makes sure that | |
247 | * writes are allowed before returning success. When | |
248 | * the write operation is finished, mnt_drop_write() | |
249 | * must be called. This is effectively a refcount. | |
250 | */ | |
251 | int mnt_want_write(struct vfsmount *mnt) | |
252 | { | |
3d733633 DH |
253 | int ret = 0; |
254 | struct mnt_writer *cpu_writer; | |
255 | ||
256 | cpu_writer = &get_cpu_var(mnt_writers); | |
257 | spin_lock(&cpu_writer->lock); | |
258 | if (__mnt_is_readonly(mnt)) { | |
259 | ret = -EROFS; | |
260 | goto out; | |
261 | } | |
262 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
263 | cpu_writer->count++; | |
264 | out: | |
265 | spin_unlock(&cpu_writer->lock); | |
266 | put_cpu_var(mnt_writers); | |
267 | return ret; | |
8366025e DH |
268 | } |
269 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
270 | ||
3d733633 DH |
271 | static void lock_mnt_writers(void) |
272 | { | |
273 | int cpu; | |
274 | struct mnt_writer *cpu_writer; | |
275 | ||
276 | for_each_possible_cpu(cpu) { | |
277 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
278 | spin_lock(&cpu_writer->lock); | |
279 | __clear_mnt_count(cpu_writer); | |
280 | cpu_writer->mnt = NULL; | |
281 | } | |
282 | } | |
283 | ||
284 | /* | |
285 | * These per-cpu write counts are not guaranteed to have | |
286 | * matched increments and decrements on any given cpu. | |
287 | * A file open()ed for write on one cpu and close()d on | |
288 | * another cpu will imbalance this count. Make sure it | |
289 | * does not get too far out of whack. | |
290 | */ | |
291 | static void handle_write_count_underflow(struct vfsmount *mnt) | |
292 | { | |
293 | if (atomic_read(&mnt->__mnt_writers) >= | |
294 | MNT_WRITER_UNDERFLOW_LIMIT) | |
295 | return; | |
296 | /* | |
297 | * It isn't necessary to hold all of the locks | |
298 | * at the same time, but doing it this way makes | |
299 | * us share a lot more code. | |
300 | */ | |
301 | lock_mnt_writers(); | |
302 | /* | |
303 | * vfsmount_lock is for mnt_flags. | |
304 | */ | |
305 | spin_lock(&vfsmount_lock); | |
306 | /* | |
307 | * If coalescing the per-cpu writer counts did not | |
308 | * get us back to a positive writer count, we have | |
309 | * a bug. | |
310 | */ | |
311 | if ((atomic_read(&mnt->__mnt_writers) < 0) && | |
312 | !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { | |
313 | printk(KERN_DEBUG "leak detected on mount(%p) writers " | |
314 | "count: %d\n", | |
315 | mnt, atomic_read(&mnt->__mnt_writers)); | |
316 | WARN_ON(1); | |
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 | ||
454e2398 DH |
400 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
401 | { | |
402 | mnt->mnt_sb = sb; | |
403 | mnt->mnt_root = dget(sb->s_root); | |
404 | return 0; | |
405 | } | |
406 | ||
407 | EXPORT_SYMBOL(simple_set_mnt); | |
408 | ||
1da177e4 LT |
409 | void free_vfsmnt(struct vfsmount *mnt) |
410 | { | |
411 | kfree(mnt->mnt_devname); | |
73cd49ec | 412 | mnt_free_id(mnt); |
1da177e4 LT |
413 | kmem_cache_free(mnt_cache, mnt); |
414 | } | |
415 | ||
416 | /* | |
a05964f3 RP |
417 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
418 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 419 | */ |
a05964f3 RP |
420 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
421 | int dir) | |
1da177e4 | 422 | { |
b58fed8b RP |
423 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
424 | struct list_head *tmp = head; | |
1da177e4 LT |
425 | struct vfsmount *p, *found = NULL; |
426 | ||
1da177e4 | 427 | for (;;) { |
a05964f3 | 428 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
429 | p = NULL; |
430 | if (tmp == head) | |
431 | break; | |
432 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
433 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 434 | found = p; |
1da177e4 LT |
435 | break; |
436 | } | |
437 | } | |
1da177e4 LT |
438 | return found; |
439 | } | |
440 | ||
a05964f3 RP |
441 | /* |
442 | * lookup_mnt increments the ref count before returning | |
443 | * the vfsmount struct. | |
444 | */ | |
445 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
446 | { | |
447 | struct vfsmount *child_mnt; | |
448 | spin_lock(&vfsmount_lock); | |
449 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
450 | mntget(child_mnt); | |
451 | spin_unlock(&vfsmount_lock); | |
452 | return child_mnt; | |
453 | } | |
454 | ||
1da177e4 LT |
455 | static inline int check_mnt(struct vfsmount *mnt) |
456 | { | |
6b3286ed | 457 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
458 | } |
459 | ||
6b3286ed | 460 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
461 | { |
462 | if (ns) { | |
463 | ns->event = ++event; | |
464 | wake_up_interruptible(&ns->poll); | |
465 | } | |
466 | } | |
467 | ||
6b3286ed | 468 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
469 | { |
470 | if (ns && ns->event != event) { | |
471 | ns->event = event; | |
472 | wake_up_interruptible(&ns->poll); | |
473 | } | |
474 | } | |
475 | ||
1a390689 | 476 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 477 | { |
1a390689 AV |
478 | old_path->dentry = mnt->mnt_mountpoint; |
479 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
480 | mnt->mnt_parent = mnt; |
481 | mnt->mnt_mountpoint = mnt->mnt_root; | |
482 | list_del_init(&mnt->mnt_child); | |
483 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 484 | old_path->dentry->d_mounted--; |
1da177e4 LT |
485 | } |
486 | ||
b90fa9ae RP |
487 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
488 | struct vfsmount *child_mnt) | |
489 | { | |
490 | child_mnt->mnt_parent = mntget(mnt); | |
491 | child_mnt->mnt_mountpoint = dget(dentry); | |
492 | dentry->d_mounted++; | |
493 | } | |
494 | ||
1a390689 | 495 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 496 | { |
1a390689 | 497 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 498 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
499 | hash(path->mnt, path->dentry)); |
500 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
501 | } |
502 | ||
503 | /* | |
504 | * the caller must hold vfsmount_lock | |
505 | */ | |
506 | static void commit_tree(struct vfsmount *mnt) | |
507 | { | |
508 | struct vfsmount *parent = mnt->mnt_parent; | |
509 | struct vfsmount *m; | |
510 | LIST_HEAD(head); | |
6b3286ed | 511 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
512 | |
513 | BUG_ON(parent == mnt); | |
514 | ||
515 | list_add_tail(&head, &mnt->mnt_list); | |
516 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 517 | m->mnt_ns = n; |
b90fa9ae RP |
518 | list_splice(&head, n->list.prev); |
519 | ||
520 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
521 | hash(parent, mnt->mnt_mountpoint)); | |
522 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 523 | touch_mnt_namespace(n); |
1da177e4 LT |
524 | } |
525 | ||
526 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
527 | { | |
528 | struct list_head *next = p->mnt_mounts.next; | |
529 | if (next == &p->mnt_mounts) { | |
530 | while (1) { | |
531 | if (p == root) | |
532 | return NULL; | |
533 | next = p->mnt_child.next; | |
534 | if (next != &p->mnt_parent->mnt_mounts) | |
535 | break; | |
536 | p = p->mnt_parent; | |
537 | } | |
538 | } | |
539 | return list_entry(next, struct vfsmount, mnt_child); | |
540 | } | |
541 | ||
9676f0c6 RP |
542 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
543 | { | |
544 | struct list_head *prev = p->mnt_mounts.prev; | |
545 | while (prev != &p->mnt_mounts) { | |
546 | p = list_entry(prev, struct vfsmount, mnt_child); | |
547 | prev = p->mnt_mounts.prev; | |
548 | } | |
549 | return p; | |
550 | } | |
551 | ||
36341f64 RP |
552 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
553 | int flag) | |
1da177e4 LT |
554 | { |
555 | struct super_block *sb = old->mnt_sb; | |
556 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
557 | ||
558 | if (mnt) { | |
719f5d7f MS |
559 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
560 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
561 | else | |
562 | mnt->mnt_group_id = old->mnt_group_id; | |
563 | ||
564 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
565 | int err = mnt_alloc_group_id(mnt); | |
566 | if (err) | |
567 | goto out_free; | |
568 | } | |
569 | ||
1da177e4 LT |
570 | mnt->mnt_flags = old->mnt_flags; |
571 | atomic_inc(&sb->s_active); | |
572 | mnt->mnt_sb = sb; | |
573 | mnt->mnt_root = dget(root); | |
574 | mnt->mnt_mountpoint = mnt->mnt_root; | |
575 | mnt->mnt_parent = mnt; | |
b90fa9ae | 576 | |
5afe0022 RP |
577 | if (flag & CL_SLAVE) { |
578 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
579 | mnt->mnt_master = old; | |
580 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 581 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
582 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
583 | list_add(&mnt->mnt_share, &old->mnt_share); | |
584 | if (IS_MNT_SLAVE(old)) | |
585 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
586 | mnt->mnt_master = old->mnt_master; | |
587 | } | |
b90fa9ae RP |
588 | if (flag & CL_MAKE_SHARED) |
589 | set_mnt_shared(mnt); | |
1da177e4 LT |
590 | |
591 | /* stick the duplicate mount on the same expiry list | |
592 | * as the original if that was on one */ | |
36341f64 | 593 | if (flag & CL_EXPIRE) { |
36341f64 RP |
594 | if (!list_empty(&old->mnt_expire)) |
595 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 596 | } |
1da177e4 LT |
597 | } |
598 | return mnt; | |
719f5d7f MS |
599 | |
600 | out_free: | |
601 | free_vfsmnt(mnt); | |
602 | return NULL; | |
1da177e4 LT |
603 | } |
604 | ||
7b7b1ace | 605 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 | 606 | { |
3d733633 | 607 | int cpu; |
1da177e4 | 608 | struct super_block *sb = mnt->mnt_sb; |
3d733633 DH |
609 | /* |
610 | * We don't have to hold all of the locks at the | |
611 | * same time here because we know that we're the | |
612 | * last reference to mnt and that no new writers | |
613 | * can come in. | |
614 | */ | |
615 | for_each_possible_cpu(cpu) { | |
616 | struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); | |
617 | if (cpu_writer->mnt != mnt) | |
618 | continue; | |
619 | spin_lock(&cpu_writer->lock); | |
620 | atomic_add(cpu_writer->count, &mnt->__mnt_writers); | |
621 | cpu_writer->count = 0; | |
622 | /* | |
623 | * Might as well do this so that no one | |
624 | * ever sees the pointer and expects | |
625 | * it to be valid. | |
626 | */ | |
627 | cpu_writer->mnt = NULL; | |
628 | spin_unlock(&cpu_writer->lock); | |
629 | } | |
630 | /* | |
631 | * This probably indicates that somebody messed | |
632 | * up a mnt_want/drop_write() pair. If this | |
633 | * happens, the filesystem was probably unable | |
634 | * to make r/w->r/o transitions. | |
635 | */ | |
636 | WARN_ON(atomic_read(&mnt->__mnt_writers)); | |
1da177e4 LT |
637 | dput(mnt->mnt_root); |
638 | free_vfsmnt(mnt); | |
639 | deactivate_super(sb); | |
640 | } | |
641 | ||
7b7b1ace AV |
642 | void mntput_no_expire(struct vfsmount *mnt) |
643 | { | |
644 | repeat: | |
645 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
646 | if (likely(!mnt->mnt_pinned)) { | |
647 | spin_unlock(&vfsmount_lock); | |
648 | __mntput(mnt); | |
649 | return; | |
650 | } | |
651 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
652 | mnt->mnt_pinned = 0; | |
653 | spin_unlock(&vfsmount_lock); | |
654 | acct_auto_close_mnt(mnt); | |
655 | security_sb_umount_close(mnt); | |
656 | goto repeat; | |
657 | } | |
658 | } | |
659 | ||
660 | EXPORT_SYMBOL(mntput_no_expire); | |
661 | ||
662 | void mnt_pin(struct vfsmount *mnt) | |
663 | { | |
664 | spin_lock(&vfsmount_lock); | |
665 | mnt->mnt_pinned++; | |
666 | spin_unlock(&vfsmount_lock); | |
667 | } | |
668 | ||
669 | EXPORT_SYMBOL(mnt_pin); | |
670 | ||
671 | void mnt_unpin(struct vfsmount *mnt) | |
672 | { | |
673 | spin_lock(&vfsmount_lock); | |
674 | if (mnt->mnt_pinned) { | |
675 | atomic_inc(&mnt->mnt_count); | |
676 | mnt->mnt_pinned--; | |
677 | } | |
678 | spin_unlock(&vfsmount_lock); | |
679 | } | |
680 | ||
681 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 682 | |
b3b304a2 MS |
683 | static inline void mangle(struct seq_file *m, const char *s) |
684 | { | |
685 | seq_escape(m, s, " \t\n\\"); | |
686 | } | |
687 | ||
688 | /* | |
689 | * Simple .show_options callback for filesystems which don't want to | |
690 | * implement more complex mount option showing. | |
691 | * | |
692 | * See also save_mount_options(). | |
693 | */ | |
694 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
695 | { | |
696 | const char *options = mnt->mnt_sb->s_options; | |
697 | ||
698 | if (options != NULL && options[0]) { | |
699 | seq_putc(m, ','); | |
700 | mangle(m, options); | |
701 | } | |
702 | ||
703 | return 0; | |
704 | } | |
705 | EXPORT_SYMBOL(generic_show_options); | |
706 | ||
707 | /* | |
708 | * If filesystem uses generic_show_options(), this function should be | |
709 | * called from the fill_super() callback. | |
710 | * | |
711 | * The .remount_fs callback usually needs to be handled in a special | |
712 | * way, to make sure, that previous options are not overwritten if the | |
713 | * remount fails. | |
714 | * | |
715 | * Also note, that if the filesystem's .remount_fs function doesn't | |
716 | * reset all options to their default value, but changes only newly | |
717 | * given options, then the displayed options will not reflect reality | |
718 | * any more. | |
719 | */ | |
720 | void save_mount_options(struct super_block *sb, char *options) | |
721 | { | |
722 | kfree(sb->s_options); | |
723 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
724 | } | |
725 | EXPORT_SYMBOL(save_mount_options); | |
726 | ||
a1a2c409 | 727 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
728 | /* iterator */ |
729 | static void *m_start(struct seq_file *m, loff_t *pos) | |
730 | { | |
a1a2c409 | 731 | struct proc_mounts *p = m->private; |
1da177e4 | 732 | |
390c6843 | 733 | down_read(&namespace_sem); |
a1a2c409 | 734 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
735 | } |
736 | ||
737 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
738 | { | |
a1a2c409 | 739 | struct proc_mounts *p = m->private; |
b0765fb8 | 740 | |
a1a2c409 | 741 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
742 | } |
743 | ||
744 | static void m_stop(struct seq_file *m, void *v) | |
745 | { | |
390c6843 | 746 | up_read(&namespace_sem); |
1da177e4 LT |
747 | } |
748 | ||
1da177e4 LT |
749 | static int show_vfsmnt(struct seq_file *m, void *v) |
750 | { | |
b0765fb8 | 751 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
1da177e4 LT |
752 | int err = 0; |
753 | static struct proc_fs_info { | |
754 | int flag; | |
755 | char *str; | |
756 | } fs_info[] = { | |
757 | { MS_SYNCHRONOUS, ",sync" }, | |
758 | { MS_DIRSYNC, ",dirsync" }, | |
759 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
760 | { 0, NULL } |
761 | }; | |
762 | static struct proc_fs_info mnt_info[] = { | |
763 | { MNT_NOSUID, ",nosuid" }, | |
764 | { MNT_NODEV, ",nodev" }, | |
765 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
766 | { MNT_NOATIME, ",noatime" }, |
767 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 768 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
769 | { 0, NULL } |
770 | }; | |
771 | struct proc_fs_info *fs_infop; | |
c32c2f63 | 772 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
773 | |
774 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
775 | seq_putc(m, ' '); | |
c32c2f63 | 776 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 LT |
777 | seq_putc(m, ' '); |
778 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
779 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { |
780 | seq_putc(m, '.'); | |
781 | mangle(m, mnt->mnt_sb->s_subtype); | |
782 | } | |
2e4b7fcd | 783 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
1da177e4 LT |
784 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { |
785 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
786 | seq_puts(m, fs_infop->str); | |
787 | } | |
788 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
789 | if (mnt->mnt_flags & fs_infop->flag) | |
790 | seq_puts(m, fs_infop->str); | |
791 | } | |
792 | if (mnt->mnt_sb->s_op->show_options) | |
793 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
794 | seq_puts(m, " 0 0\n"); | |
795 | return err; | |
796 | } | |
797 | ||
a1a2c409 | 798 | const struct seq_operations mounts_op = { |
1da177e4 LT |
799 | .start = m_start, |
800 | .next = m_next, | |
801 | .stop = m_stop, | |
802 | .show = show_vfsmnt | |
803 | }; | |
804 | ||
b4629fe2 CL |
805 | static int show_vfsstat(struct seq_file *m, void *v) |
806 | { | |
b0765fb8 | 807 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 808 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
809 | int err = 0; |
810 | ||
811 | /* device */ | |
812 | if (mnt->mnt_devname) { | |
813 | seq_puts(m, "device "); | |
814 | mangle(m, mnt->mnt_devname); | |
815 | } else | |
816 | seq_puts(m, "no device"); | |
817 | ||
818 | /* mount point */ | |
819 | seq_puts(m, " mounted on "); | |
c32c2f63 | 820 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
821 | seq_putc(m, ' '); |
822 | ||
823 | /* file system type */ | |
824 | seq_puts(m, "with fstype "); | |
825 | mangle(m, mnt->mnt_sb->s_type->name); | |
826 | ||
827 | /* optional statistics */ | |
828 | if (mnt->mnt_sb->s_op->show_stats) { | |
829 | seq_putc(m, ' '); | |
830 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
831 | } | |
832 | ||
833 | seq_putc(m, '\n'); | |
834 | return err; | |
835 | } | |
836 | ||
a1a2c409 | 837 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
838 | .start = m_start, |
839 | .next = m_next, | |
840 | .stop = m_stop, | |
841 | .show = show_vfsstat, | |
842 | }; | |
a1a2c409 | 843 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 844 | |
1da177e4 LT |
845 | /** |
846 | * may_umount_tree - check if a mount tree is busy | |
847 | * @mnt: root of mount tree | |
848 | * | |
849 | * This is called to check if a tree of mounts has any | |
850 | * open files, pwds, chroots or sub mounts that are | |
851 | * busy. | |
852 | */ | |
853 | int may_umount_tree(struct vfsmount *mnt) | |
854 | { | |
36341f64 RP |
855 | int actual_refs = 0; |
856 | int minimum_refs = 0; | |
857 | struct vfsmount *p; | |
1da177e4 LT |
858 | |
859 | spin_lock(&vfsmount_lock); | |
36341f64 | 860 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
861 | actual_refs += atomic_read(&p->mnt_count); |
862 | minimum_refs += 2; | |
1da177e4 LT |
863 | } |
864 | spin_unlock(&vfsmount_lock); | |
865 | ||
866 | if (actual_refs > minimum_refs) | |
e3474a8e | 867 | return 0; |
1da177e4 | 868 | |
e3474a8e | 869 | return 1; |
1da177e4 LT |
870 | } |
871 | ||
872 | EXPORT_SYMBOL(may_umount_tree); | |
873 | ||
874 | /** | |
875 | * may_umount - check if a mount point is busy | |
876 | * @mnt: root of mount | |
877 | * | |
878 | * This is called to check if a mount point has any | |
879 | * open files, pwds, chroots or sub mounts. If the | |
880 | * mount has sub mounts this will return busy | |
881 | * regardless of whether the sub mounts are busy. | |
882 | * | |
883 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
884 | * give false negatives. The main reason why it's here is that we need | |
885 | * a non-destructive way to look for easily umountable filesystems. | |
886 | */ | |
887 | int may_umount(struct vfsmount *mnt) | |
888 | { | |
e3474a8e | 889 | int ret = 1; |
a05964f3 RP |
890 | spin_lock(&vfsmount_lock); |
891 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 892 | ret = 0; |
a05964f3 RP |
893 | spin_unlock(&vfsmount_lock); |
894 | return ret; | |
1da177e4 LT |
895 | } |
896 | ||
897 | EXPORT_SYMBOL(may_umount); | |
898 | ||
b90fa9ae | 899 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
900 | { |
901 | struct vfsmount *mnt; | |
bf066c7d | 902 | while (!list_empty(head)) { |
b5e61818 | 903 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
904 | list_del_init(&mnt->mnt_hash); |
905 | if (mnt->mnt_parent != mnt) { | |
906 | struct dentry *dentry; | |
907 | struct vfsmount *m; | |
908 | spin_lock(&vfsmount_lock); | |
909 | dentry = mnt->mnt_mountpoint; | |
910 | m = mnt->mnt_parent; | |
911 | mnt->mnt_mountpoint = mnt->mnt_root; | |
912 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 913 | m->mnt_ghosts--; |
70fbcdf4 RP |
914 | spin_unlock(&vfsmount_lock); |
915 | dput(dentry); | |
916 | mntput(m); | |
917 | } | |
918 | mntput(mnt); | |
919 | } | |
920 | } | |
921 | ||
a05964f3 | 922 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
923 | { |
924 | struct vfsmount *p; | |
1da177e4 | 925 | |
1bfba4e8 AM |
926 | for (p = mnt; p; p = next_mnt(p, mnt)) |
927 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 928 | |
a05964f3 RP |
929 | if (propagate) |
930 | propagate_umount(kill); | |
931 | ||
70fbcdf4 RP |
932 | list_for_each_entry(p, kill, mnt_hash) { |
933 | list_del_init(&p->mnt_expire); | |
934 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
935 | __touch_mnt_namespace(p->mnt_ns); |
936 | p->mnt_ns = NULL; | |
70fbcdf4 | 937 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
938 | if (p->mnt_parent != p) { |
939 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 940 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 941 | } |
a05964f3 | 942 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
943 | } |
944 | } | |
945 | ||
c35038be AV |
946 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
947 | ||
1da177e4 LT |
948 | static int do_umount(struct vfsmount *mnt, int flags) |
949 | { | |
b58fed8b | 950 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 951 | int retval; |
70fbcdf4 | 952 | LIST_HEAD(umount_list); |
1da177e4 LT |
953 | |
954 | retval = security_sb_umount(mnt, flags); | |
955 | if (retval) | |
956 | return retval; | |
957 | ||
958 | /* | |
959 | * Allow userspace to request a mountpoint be expired rather than | |
960 | * unmounting unconditionally. Unmount only happens if: | |
961 | * (1) the mark is already set (the mark is cleared by mntput()) | |
962 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
963 | */ | |
964 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 965 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
966 | flags & (MNT_FORCE | MNT_DETACH)) |
967 | return -EINVAL; | |
968 | ||
969 | if (atomic_read(&mnt->mnt_count) != 2) | |
970 | return -EBUSY; | |
971 | ||
972 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
973 | return -EAGAIN; | |
974 | } | |
975 | ||
976 | /* | |
977 | * If we may have to abort operations to get out of this | |
978 | * mount, and they will themselves hold resources we must | |
979 | * allow the fs to do things. In the Unix tradition of | |
980 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
981 | * might fail to complete on the first run through as other tasks | |
982 | * must return, and the like. Thats for the mount program to worry | |
983 | * about for the moment. | |
984 | */ | |
985 | ||
986 | lock_kernel(); | |
8b512d9a TM |
987 | if (sb->s_op->umount_begin) |
988 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
989 | unlock_kernel(); |
990 | ||
991 | /* | |
992 | * No sense to grab the lock for this test, but test itself looks | |
993 | * somewhat bogus. Suggestions for better replacement? | |
994 | * Ho-hum... In principle, we might treat that as umount + switch | |
995 | * to rootfs. GC would eventually take care of the old vfsmount. | |
996 | * Actually it makes sense, especially if rootfs would contain a | |
997 | * /reboot - static binary that would close all descriptors and | |
998 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
999 | */ | |
6ac08c39 | 1000 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1001 | /* |
1002 | * Special case for "unmounting" root ... | |
1003 | * we just try to remount it readonly. | |
1004 | */ | |
1005 | down_write(&sb->s_umount); | |
1006 | if (!(sb->s_flags & MS_RDONLY)) { | |
1007 | lock_kernel(); | |
1008 | DQUOT_OFF(sb); | |
1009 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
1010 | unlock_kernel(); | |
1011 | } | |
1012 | up_write(&sb->s_umount); | |
1013 | return retval; | |
1014 | } | |
1015 | ||
390c6843 | 1016 | down_write(&namespace_sem); |
1da177e4 | 1017 | spin_lock(&vfsmount_lock); |
5addc5dd | 1018 | event++; |
1da177e4 | 1019 | |
c35038be AV |
1020 | if (!(flags & MNT_DETACH)) |
1021 | shrink_submounts(mnt, &umount_list); | |
1022 | ||
1da177e4 | 1023 | retval = -EBUSY; |
a05964f3 | 1024 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1025 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1026 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1027 | retval = 0; |
1028 | } | |
1029 | spin_unlock(&vfsmount_lock); | |
1030 | if (retval) | |
1031 | security_sb_umount_busy(mnt); | |
390c6843 | 1032 | up_write(&namespace_sem); |
70fbcdf4 | 1033 | release_mounts(&umount_list); |
1da177e4 LT |
1034 | return retval; |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Now umount can handle mount points as well as block devices. | |
1039 | * This is important for filesystems which use unnamed block devices. | |
1040 | * | |
1041 | * We now support a flag for forced unmount like the other 'big iron' | |
1042 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1043 | */ | |
1044 | ||
1045 | asmlinkage long sys_umount(char __user * name, int flags) | |
1046 | { | |
1047 | struct nameidata nd; | |
1048 | int retval; | |
1049 | ||
1050 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
1051 | if (retval) | |
1052 | goto out; | |
1053 | retval = -EINVAL; | |
4ac91378 | 1054 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 1055 | goto dput_and_out; |
4ac91378 | 1056 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
1057 | goto dput_and_out; |
1058 | ||
1059 | retval = -EPERM; | |
1060 | if (!capable(CAP_SYS_ADMIN)) | |
1061 | goto dput_and_out; | |
1062 | ||
4ac91378 | 1063 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 1064 | dput_and_out: |
429731b1 | 1065 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
1066 | dput(nd.path.dentry); |
1067 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
1068 | out: |
1069 | return retval; | |
1070 | } | |
1071 | ||
1072 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1073 | ||
1074 | /* | |
b58fed8b | 1075 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1076 | */ |
1da177e4 LT |
1077 | asmlinkage long sys_oldumount(char __user * name) |
1078 | { | |
b58fed8b | 1079 | return sys_umount(name, 0); |
1da177e4 LT |
1080 | } |
1081 | ||
1082 | #endif | |
1083 | ||
1084 | static int mount_is_safe(struct nameidata *nd) | |
1085 | { | |
1086 | if (capable(CAP_SYS_ADMIN)) | |
1087 | return 0; | |
1088 | return -EPERM; | |
1089 | #ifdef notyet | |
4ac91378 | 1090 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 1091 | return -EPERM; |
4ac91378 JB |
1092 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
1093 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
1094 | return -EPERM; |
1095 | } | |
e4543edd | 1096 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
1097 | return -EPERM; |
1098 | return 0; | |
1099 | #endif | |
1100 | } | |
1101 | ||
b58fed8b | 1102 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
1103 | { |
1104 | while (1) { | |
1105 | if (d == dentry) | |
1106 | return 1; | |
1107 | if (d == NULL || d == d->d_parent) | |
1108 | return 0; | |
1109 | d = d->d_parent; | |
1110 | } | |
1111 | } | |
1112 | ||
b90fa9ae | 1113 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1114 | int flag) |
1da177e4 LT |
1115 | { |
1116 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1117 | struct path path; |
1da177e4 | 1118 | |
9676f0c6 RP |
1119 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1120 | return NULL; | |
1121 | ||
36341f64 | 1122 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1123 | if (!q) |
1124 | goto Enomem; | |
1125 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1126 | ||
1127 | p = mnt; | |
fdadd65f | 1128 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
1129 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
1130 | continue; | |
1131 | ||
1132 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1133 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1134 | s = skip_mnt_tree(s); | |
1135 | continue; | |
1136 | } | |
1da177e4 LT |
1137 | while (p != s->mnt_parent) { |
1138 | p = p->mnt_parent; | |
1139 | q = q->mnt_parent; | |
1140 | } | |
1141 | p = s; | |
1a390689 AV |
1142 | path.mnt = q; |
1143 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1144 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1145 | if (!q) |
1146 | goto Enomem; | |
1147 | spin_lock(&vfsmount_lock); | |
1148 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1149 | attach_mnt(q, &path); |
1da177e4 LT |
1150 | spin_unlock(&vfsmount_lock); |
1151 | } | |
1152 | } | |
1153 | return res; | |
b58fed8b | 1154 | Enomem: |
1da177e4 | 1155 | if (res) { |
70fbcdf4 | 1156 | LIST_HEAD(umount_list); |
1da177e4 | 1157 | spin_lock(&vfsmount_lock); |
a05964f3 | 1158 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1159 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1160 | release_mounts(&umount_list); |
1da177e4 LT |
1161 | } |
1162 | return NULL; | |
1163 | } | |
1164 | ||
8aec0809 AV |
1165 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
1166 | { | |
1167 | struct vfsmount *tree; | |
1a60a280 | 1168 | down_write(&namespace_sem); |
8aec0809 | 1169 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1170 | up_write(&namespace_sem); |
8aec0809 AV |
1171 | return tree; |
1172 | } | |
1173 | ||
1174 | void drop_collected_mounts(struct vfsmount *mnt) | |
1175 | { | |
1176 | LIST_HEAD(umount_list); | |
1a60a280 | 1177 | down_write(&namespace_sem); |
8aec0809 AV |
1178 | spin_lock(&vfsmount_lock); |
1179 | umount_tree(mnt, 0, &umount_list); | |
1180 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1181 | up_write(&namespace_sem); |
8aec0809 AV |
1182 | release_mounts(&umount_list); |
1183 | } | |
1184 | ||
719f5d7f MS |
1185 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1186 | { | |
1187 | struct vfsmount *p; | |
1188 | ||
1189 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1190 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1191 | mnt_release_group_id(p); | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1196 | { | |
1197 | struct vfsmount *p; | |
1198 | ||
1199 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1200 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1201 | int err = mnt_alloc_group_id(p); | |
1202 | if (err) { | |
1203 | cleanup_group_ids(mnt, p); | |
1204 | return err; | |
1205 | } | |
1206 | } | |
1207 | } | |
1208 | ||
1209 | return 0; | |
1210 | } | |
1211 | ||
b90fa9ae RP |
1212 | /* |
1213 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1214 | * @nd : place the mount tree @source_mnt is attached |
1215 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1216 | * store the parent mount and mountpoint dentry. | |
1217 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1218 | * |
1219 | * NOTE: in the table below explains the semantics when a source mount | |
1220 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1221 | * --------------------------------------------------------------------------- |
1222 | * | BIND MOUNT OPERATION | | |
1223 | * |************************************************************************** | |
1224 | * | source-->| shared | private | slave | unbindable | | |
1225 | * | dest | | | | | | |
1226 | * | | | | | | | | |
1227 | * | v | | | | | | |
1228 | * |************************************************************************** | |
1229 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1230 | * | | | | | | | |
1231 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1232 | * *************************************************************************** | |
b90fa9ae RP |
1233 | * A bind operation clones the source mount and mounts the clone on the |
1234 | * destination mount. | |
1235 | * | |
1236 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1237 | * tree of the destination mount and the cloned mount is added to | |
1238 | * the peer group of the source mount. | |
1239 | * (+) the cloned mount is created under the destination mount and is marked | |
1240 | * as shared. The cloned mount is added to the peer group of the source | |
1241 | * mount. | |
5afe0022 RP |
1242 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1243 | * of the destination mount and the cloned mount is made slave | |
1244 | * of the same master as that of the source mount. The cloned mount | |
1245 | * is marked as 'shared and slave'. | |
1246 | * (*) the cloned mount is made a slave of the same master as that of the | |
1247 | * source mount. | |
1248 | * | |
9676f0c6 RP |
1249 | * --------------------------------------------------------------------------- |
1250 | * | MOVE MOUNT OPERATION | | |
1251 | * |************************************************************************** | |
1252 | * | source-->| shared | private | slave | unbindable | | |
1253 | * | dest | | | | | | |
1254 | * | | | | | | | | |
1255 | * | v | | | | | | |
1256 | * |************************************************************************** | |
1257 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1258 | * | | | | | | | |
1259 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1260 | * *************************************************************************** | |
5afe0022 RP |
1261 | * |
1262 | * (+) the mount is moved to the destination. And is then propagated to | |
1263 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1264 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1265 | * (+++) the mount is moved to the destination and is then propagated to |
1266 | * all the mounts belonging to the destination mount's propagation tree. | |
1267 | * the mount is marked as 'shared and slave'. | |
1268 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1269 | * |
1270 | * if the source mount is a tree, the operations explained above is | |
1271 | * applied to each mount in the tree. | |
1272 | * Must be called without spinlocks held, since this function can sleep | |
1273 | * in allocations. | |
1274 | */ | |
1275 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1276 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1277 | { |
1278 | LIST_HEAD(tree_list); | |
1a390689 AV |
1279 | struct vfsmount *dest_mnt = path->mnt; |
1280 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1281 | struct vfsmount *child, *p; |
719f5d7f | 1282 | int err; |
b90fa9ae | 1283 | |
719f5d7f MS |
1284 | if (IS_MNT_SHARED(dest_mnt)) { |
1285 | err = invent_group_ids(source_mnt, true); | |
1286 | if (err) | |
1287 | goto out; | |
1288 | } | |
1289 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1290 | if (err) | |
1291 | goto out_cleanup_ids; | |
b90fa9ae RP |
1292 | |
1293 | if (IS_MNT_SHARED(dest_mnt)) { | |
1294 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1295 | set_mnt_shared(p); | |
1296 | } | |
1297 | ||
1298 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1299 | if (parent_path) { |
1300 | detach_mnt(source_mnt, parent_path); | |
1301 | attach_mnt(source_mnt, path); | |
6b3286ed | 1302 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
1303 | } else { |
1304 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1305 | commit_tree(source_mnt); | |
1306 | } | |
b90fa9ae RP |
1307 | |
1308 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1309 | list_del_init(&child->mnt_hash); | |
1310 | commit_tree(child); | |
1311 | } | |
1312 | spin_unlock(&vfsmount_lock); | |
1313 | return 0; | |
719f5d7f MS |
1314 | |
1315 | out_cleanup_ids: | |
1316 | if (IS_MNT_SHARED(dest_mnt)) | |
1317 | cleanup_group_ids(source_mnt, NULL); | |
1318 | out: | |
1319 | return err; | |
b90fa9ae RP |
1320 | } |
1321 | ||
8c3ee42e | 1322 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1323 | { |
1324 | int err; | |
1325 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1326 | return -EINVAL; | |
1327 | ||
8c3ee42e | 1328 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1329 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1330 | return -ENOTDIR; | |
1331 | ||
1332 | err = -ENOENT; | |
8c3ee42e AV |
1333 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1334 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1335 | goto out_unlock; |
1336 | ||
8c3ee42e | 1337 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1338 | if (err) |
1339 | goto out_unlock; | |
1340 | ||
1341 | err = -ENOENT; | |
8c3ee42e AV |
1342 | if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) |
1343 | err = attach_recursive_mnt(mnt, path, NULL); | |
1da177e4 | 1344 | out_unlock: |
8c3ee42e | 1345 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1346 | if (!err) |
8c3ee42e | 1347 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1348 | return err; |
1349 | } | |
1350 | ||
07b20889 RP |
1351 | /* |
1352 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 1353 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 1354 | */ |
2dafe1c4 | 1355 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 1356 | { |
4ac91378 | 1357 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
1358 | int recurse = flag & MS_REC; |
1359 | int type = flag & ~MS_REC; | |
719f5d7f | 1360 | int err = 0; |
07b20889 | 1361 | |
ee6f9582 MS |
1362 | if (!capable(CAP_SYS_ADMIN)) |
1363 | return -EPERM; | |
1364 | ||
4ac91378 | 1365 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
1366 | return -EINVAL; |
1367 | ||
1368 | down_write(&namespace_sem); | |
719f5d7f MS |
1369 | if (type == MS_SHARED) { |
1370 | err = invent_group_ids(mnt, recurse); | |
1371 | if (err) | |
1372 | goto out_unlock; | |
1373 | } | |
1374 | ||
07b20889 RP |
1375 | spin_lock(&vfsmount_lock); |
1376 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1377 | change_mnt_propagation(m, type); | |
1378 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1379 | |
1380 | out_unlock: | |
07b20889 | 1381 | up_write(&namespace_sem); |
719f5d7f | 1382 | return err; |
07b20889 RP |
1383 | } |
1384 | ||
1da177e4 LT |
1385 | /* |
1386 | * do loopback mount. | |
2dafe1c4 | 1387 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1388 | */ |
2dafe1c4 ES |
1389 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
1390 | int recurse) | |
1da177e4 LT |
1391 | { |
1392 | struct nameidata old_nd; | |
1393 | struct vfsmount *mnt = NULL; | |
1394 | int err = mount_is_safe(nd); | |
1395 | if (err) | |
1396 | return err; | |
1397 | if (!old_name || !*old_name) | |
1398 | return -EINVAL; | |
1399 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1400 | if (err) | |
1401 | return err; | |
1402 | ||
390c6843 | 1403 | down_write(&namespace_sem); |
1da177e4 | 1404 | err = -EINVAL; |
4ac91378 JB |
1405 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
1406 | goto out; | |
9676f0c6 | 1407 | |
4ac91378 | 1408 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 1409 | goto out; |
1da177e4 | 1410 | |
ccd48bc7 AV |
1411 | err = -ENOMEM; |
1412 | if (recurse) | |
4ac91378 | 1413 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 1414 | else |
4ac91378 | 1415 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1416 | |
1417 | if (!mnt) | |
1418 | goto out; | |
1419 | ||
8c3ee42e | 1420 | err = graft_tree(mnt, &nd->path); |
ccd48bc7 | 1421 | if (err) { |
70fbcdf4 | 1422 | LIST_HEAD(umount_list); |
1da177e4 | 1423 | spin_lock(&vfsmount_lock); |
a05964f3 | 1424 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1425 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1426 | release_mounts(&umount_list); |
5b83d2c5 | 1427 | } |
1da177e4 | 1428 | |
ccd48bc7 | 1429 | out: |
390c6843 | 1430 | up_write(&namespace_sem); |
1d957f9b | 1431 | path_put(&old_nd.path); |
1da177e4 LT |
1432 | return err; |
1433 | } | |
1434 | ||
2e4b7fcd DH |
1435 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1436 | { | |
1437 | int error = 0; | |
1438 | int readonly_request = 0; | |
1439 | ||
1440 | if (ms_flags & MS_RDONLY) | |
1441 | readonly_request = 1; | |
1442 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1443 | return 0; | |
1444 | ||
1445 | if (readonly_request) | |
1446 | error = mnt_make_readonly(mnt); | |
1447 | else | |
1448 | __mnt_unmake_readonly(mnt); | |
1449 | return error; | |
1450 | } | |
1451 | ||
1da177e4 LT |
1452 | /* |
1453 | * change filesystem flags. dir should be a physical root of filesystem. | |
1454 | * If you've mounted a non-root directory somewhere and want to do remount | |
1455 | * on it - tough luck. | |
2dafe1c4 | 1456 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1457 | */ |
2dafe1c4 | 1458 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1459 | void *data) |
1460 | { | |
1461 | int err; | |
4ac91378 | 1462 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1463 | |
1464 | if (!capable(CAP_SYS_ADMIN)) | |
1465 | return -EPERM; | |
1466 | ||
4ac91378 | 1467 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1468 | return -EINVAL; |
1469 | ||
4ac91378 | 1470 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1471 | return -EINVAL; |
1472 | ||
1473 | down_write(&sb->s_umount); | |
2e4b7fcd DH |
1474 | if (flags & MS_BIND) |
1475 | err = change_mount_flags(nd->path.mnt, flags); | |
1476 | else | |
1477 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1478 | if (!err) |
4ac91378 | 1479 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1480 | up_write(&sb->s_umount); |
1481 | if (!err) | |
4ac91378 | 1482 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1483 | return err; |
1484 | } | |
1485 | ||
9676f0c6 RP |
1486 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1487 | { | |
1488 | struct vfsmount *p; | |
1489 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1490 | if (IS_MNT_UNBINDABLE(p)) | |
1491 | return 1; | |
1492 | } | |
1493 | return 0; | |
1494 | } | |
1495 | ||
2dafe1c4 ES |
1496 | /* |
1497 | * noinline this do_mount helper to save do_mount stack space. | |
1498 | */ | |
1499 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1500 | { |
1a390689 AV |
1501 | struct nameidata old_nd; |
1502 | struct path parent_path; | |
1da177e4 LT |
1503 | struct vfsmount *p; |
1504 | int err = 0; | |
1505 | if (!capable(CAP_SYS_ADMIN)) | |
1506 | return -EPERM; | |
1507 | if (!old_name || !*old_name) | |
1508 | return -EINVAL; | |
1509 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1510 | if (err) | |
1511 | return err; | |
1512 | ||
390c6843 | 1513 | down_write(&namespace_sem); |
4ac91378 JB |
1514 | while (d_mountpoint(nd->path.dentry) && |
1515 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1516 | ; |
1517 | err = -EINVAL; | |
4ac91378 | 1518 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1519 | goto out; |
1520 | ||
1521 | err = -ENOENT; | |
4ac91378 JB |
1522 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1523 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1524 | goto out1; |
1525 | ||
4ac91378 | 1526 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1527 | goto out1; |
1da177e4 LT |
1528 | |
1529 | err = -EINVAL; | |
4ac91378 | 1530 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1531 | goto out1; |
1da177e4 | 1532 | |
4ac91378 | 1533 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1534 | goto out1; |
1da177e4 | 1535 | |
4ac91378 JB |
1536 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1537 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1538 | goto out1; |
1539 | /* | |
1540 | * Don't move a mount residing in a shared parent. | |
1541 | */ | |
4ac91378 JB |
1542 | if (old_nd.path.mnt->mnt_parent && |
1543 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1544 | goto out1; |
9676f0c6 RP |
1545 | /* |
1546 | * Don't move a mount tree containing unbindable mounts to a destination | |
1547 | * mount which is shared. | |
1548 | */ | |
4ac91378 JB |
1549 | if (IS_MNT_SHARED(nd->path.mnt) && |
1550 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1551 | goto out1; |
1da177e4 | 1552 | err = -ELOOP; |
4ac91378 JB |
1553 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1554 | if (p == old_nd.path.mnt) | |
21444403 | 1555 | goto out1; |
1da177e4 | 1556 | |
1a390689 | 1557 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1558 | if (err) |
21444403 | 1559 | goto out1; |
1da177e4 LT |
1560 | |
1561 | /* if the mount is moved, it should no longer be expire | |
1562 | * automatically */ | |
4ac91378 | 1563 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 | 1564 | out1: |
4ac91378 | 1565 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1566 | out: |
390c6843 | 1567 | up_write(&namespace_sem); |
1da177e4 | 1568 | if (!err) |
1a390689 | 1569 | path_put(&parent_path); |
1d957f9b | 1570 | path_put(&old_nd.path); |
1da177e4 LT |
1571 | return err; |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * create a new mount for userspace and request it to be added into the | |
1576 | * namespace's tree | |
2dafe1c4 | 1577 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1578 | */ |
2dafe1c4 | 1579 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1580 | int mnt_flags, char *name, void *data) |
1581 | { | |
1582 | struct vfsmount *mnt; | |
1583 | ||
1584 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1585 | return -EINVAL; | |
1586 | ||
1587 | /* we need capabilities... */ | |
1588 | if (!capable(CAP_SYS_ADMIN)) | |
1589 | return -EPERM; | |
1590 | ||
1591 | mnt = do_kern_mount(type, flags, name, data); | |
1592 | if (IS_ERR(mnt)) | |
1593 | return PTR_ERR(mnt); | |
1594 | ||
1595 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * add a mount into a namespace's mount tree | |
1600 | * - provide the option of adding the new mount to an expiration list | |
1601 | */ | |
1602 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1603 | int mnt_flags, struct list_head *fslist) | |
1604 | { | |
1605 | int err; | |
1606 | ||
390c6843 | 1607 | down_write(&namespace_sem); |
1da177e4 | 1608 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1609 | while (d_mountpoint(nd->path.dentry) && |
1610 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1611 | ; |
1612 | err = -EINVAL; | |
4ac91378 | 1613 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1614 | goto unlock; |
1615 | ||
1616 | /* Refuse the same filesystem on the same mount point */ | |
1617 | err = -EBUSY; | |
4ac91378 JB |
1618 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1619 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1620 | goto unlock; |
1621 | ||
1622 | err = -EINVAL; | |
1623 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1624 | goto unlock; | |
1625 | ||
1626 | newmnt->mnt_flags = mnt_flags; | |
8c3ee42e | 1627 | if ((err = graft_tree(newmnt, &nd->path))) |
5b83d2c5 | 1628 | goto unlock; |
1da177e4 | 1629 | |
6758f953 | 1630 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1631 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1632 | |
390c6843 | 1633 | up_write(&namespace_sem); |
5b83d2c5 | 1634 | return 0; |
1da177e4 LT |
1635 | |
1636 | unlock: | |
390c6843 | 1637 | up_write(&namespace_sem); |
1da177e4 LT |
1638 | mntput(newmnt); |
1639 | return err; | |
1640 | } | |
1641 | ||
1642 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1643 | ||
1644 | /* | |
1645 | * process a list of expirable mountpoints with the intent of discarding any | |
1646 | * mountpoints that aren't in use and haven't been touched since last we came | |
1647 | * here | |
1648 | */ | |
1649 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1650 | { | |
1da177e4 LT |
1651 | struct vfsmount *mnt, *next; |
1652 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1653 | LIST_HEAD(umounts); |
1da177e4 LT |
1654 | |
1655 | if (list_empty(mounts)) | |
1656 | return; | |
1657 | ||
bcc5c7d2 | 1658 | down_write(&namespace_sem); |
1da177e4 LT |
1659 | spin_lock(&vfsmount_lock); |
1660 | ||
1661 | /* extract from the expiration list every vfsmount that matches the | |
1662 | * following criteria: | |
1663 | * - only referenced by its parent vfsmount | |
1664 | * - still marked for expiry (marked on the last call here; marks are | |
1665 | * cleared by mntput()) | |
1666 | */ | |
55e700b9 | 1667 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1668 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1669 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1670 | continue; |
55e700b9 | 1671 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1672 | } |
bcc5c7d2 AV |
1673 | while (!list_empty(&graveyard)) { |
1674 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1675 | touch_mnt_namespace(mnt->mnt_ns); | |
1676 | umount_tree(mnt, 1, &umounts); | |
1677 | } | |
5528f911 | 1678 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1679 | up_write(&namespace_sem); |
1680 | ||
1681 | release_mounts(&umounts); | |
5528f911 TM |
1682 | } |
1683 | ||
1684 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1685 | ||
1686 | /* | |
1687 | * Ripoff of 'select_parent()' | |
1688 | * | |
1689 | * search the list of submounts for a given mountpoint, and move any | |
1690 | * shrinkable submounts to the 'graveyard' list. | |
1691 | */ | |
1692 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1693 | { | |
1694 | struct vfsmount *this_parent = parent; | |
1695 | struct list_head *next; | |
1696 | int found = 0; | |
1697 | ||
1698 | repeat: | |
1699 | next = this_parent->mnt_mounts.next; | |
1700 | resume: | |
1701 | while (next != &this_parent->mnt_mounts) { | |
1702 | struct list_head *tmp = next; | |
1703 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1704 | ||
1705 | next = tmp->next; | |
1706 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1707 | continue; |
5528f911 TM |
1708 | /* |
1709 | * Descend a level if the d_mounts list is non-empty. | |
1710 | */ | |
1711 | if (!list_empty(&mnt->mnt_mounts)) { | |
1712 | this_parent = mnt; | |
1713 | goto repeat; | |
1714 | } | |
1da177e4 | 1715 | |
5528f911 | 1716 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1717 | list_move_tail(&mnt->mnt_expire, graveyard); |
1718 | found++; | |
1719 | } | |
1da177e4 | 1720 | } |
5528f911 TM |
1721 | /* |
1722 | * All done at this level ... ascend and resume the search | |
1723 | */ | |
1724 | if (this_parent != parent) { | |
1725 | next = this_parent->mnt_child.next; | |
1726 | this_parent = this_parent->mnt_parent; | |
1727 | goto resume; | |
1728 | } | |
1729 | return found; | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * process a list of expirable mountpoints with the intent of discarding any | |
1734 | * submounts of a specific parent mountpoint | |
1735 | */ | |
c35038be | 1736 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1737 | { |
1738 | LIST_HEAD(graveyard); | |
c35038be | 1739 | struct vfsmount *m; |
5528f911 | 1740 | |
5528f911 | 1741 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1742 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1743 | while (!list_empty(&graveyard)) { |
c35038be | 1744 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 AV |
1745 | mnt_expire); |
1746 | touch_mnt_namespace(mnt->mnt_ns); | |
c35038be | 1747 | umount_tree(mnt, 1, umounts); |
bcc5c7d2 AV |
1748 | } |
1749 | } | |
1da177e4 LT |
1750 | } |
1751 | ||
1da177e4 LT |
1752 | /* |
1753 | * Some copy_from_user() implementations do not return the exact number of | |
1754 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1755 | * Note that this function differs from copy_from_user() in that it will oops | |
1756 | * on bad values of `to', rather than returning a short copy. | |
1757 | */ | |
b58fed8b RP |
1758 | static long exact_copy_from_user(void *to, const void __user * from, |
1759 | unsigned long n) | |
1da177e4 LT |
1760 | { |
1761 | char *t = to; | |
1762 | const char __user *f = from; | |
1763 | char c; | |
1764 | ||
1765 | if (!access_ok(VERIFY_READ, from, n)) | |
1766 | return n; | |
1767 | ||
1768 | while (n) { | |
1769 | if (__get_user(c, f)) { | |
1770 | memset(t, 0, n); | |
1771 | break; | |
1772 | } | |
1773 | *t++ = c; | |
1774 | f++; | |
1775 | n--; | |
1776 | } | |
1777 | return n; | |
1778 | } | |
1779 | ||
b58fed8b | 1780 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1781 | { |
1782 | int i; | |
1783 | unsigned long page; | |
1784 | unsigned long size; | |
b58fed8b | 1785 | |
1da177e4 LT |
1786 | *where = 0; |
1787 | if (!data) | |
1788 | return 0; | |
1789 | ||
1790 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1791 | return -ENOMEM; | |
1792 | ||
1793 | /* We only care that *some* data at the address the user | |
1794 | * gave us is valid. Just in case, we'll zero | |
1795 | * the remainder of the page. | |
1796 | */ | |
1797 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1798 | size = TASK_SIZE - (unsigned long)data; | |
1799 | if (size > PAGE_SIZE) | |
1800 | size = PAGE_SIZE; | |
1801 | ||
1802 | i = size - exact_copy_from_user((void *)page, data, size); | |
1803 | if (!i) { | |
b58fed8b | 1804 | free_page(page); |
1da177e4 LT |
1805 | return -EFAULT; |
1806 | } | |
1807 | if (i != PAGE_SIZE) | |
1808 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1809 | *where = page; | |
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | /* | |
1814 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1815 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1816 | * | |
1817 | * data is a (void *) that can point to any structure up to | |
1818 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1819 | * information (or be NULL). | |
1820 | * | |
1821 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1822 | * When the flags word was introduced its top half was required | |
1823 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1824 | * Therefore, if this magic number is present, it carries no information | |
1825 | * and must be discarded. | |
1826 | */ | |
b58fed8b | 1827 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1828 | unsigned long flags, void *data_page) |
1829 | { | |
1830 | struct nameidata nd; | |
1831 | int retval = 0; | |
1832 | int mnt_flags = 0; | |
1833 | ||
1834 | /* Discard magic */ | |
1835 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1836 | flags &= ~MS_MGC_MSK; | |
1837 | ||
1838 | /* Basic sanity checks */ | |
1839 | ||
1840 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1841 | return -EINVAL; | |
1842 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1843 | return -EINVAL; | |
1844 | ||
1845 | if (data_page) | |
1846 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1847 | ||
1848 | /* Separate the per-mountpoint flags */ | |
1849 | if (flags & MS_NOSUID) | |
1850 | mnt_flags |= MNT_NOSUID; | |
1851 | if (flags & MS_NODEV) | |
1852 | mnt_flags |= MNT_NODEV; | |
1853 | if (flags & MS_NOEXEC) | |
1854 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1855 | if (flags & MS_NOATIME) |
1856 | mnt_flags |= MNT_NOATIME; | |
1857 | if (flags & MS_NODIRATIME) | |
1858 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1859 | if (flags & MS_RELATIME) |
1860 | mnt_flags |= MNT_RELATIME; | |
2e4b7fcd DH |
1861 | if (flags & MS_RDONLY) |
1862 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1863 | |
1864 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1865 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1866 | |
1867 | /* ... and get the mountpoint */ | |
1868 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1869 | if (retval) | |
1870 | return retval; | |
1871 | ||
b5266eb4 AV |
1872 | retval = security_sb_mount(dev_name, &nd.path, |
1873 | type_page, flags, data_page); | |
1da177e4 LT |
1874 | if (retval) |
1875 | goto dput_out; | |
1876 | ||
1877 | if (flags & MS_REMOUNT) | |
1878 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1879 | data_page); | |
1880 | else if (flags & MS_BIND) | |
eee391a6 | 1881 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1882 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1883 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1884 | else if (flags & MS_MOVE) |
1885 | retval = do_move_mount(&nd, dev_name); | |
1886 | else | |
1887 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1888 | dev_name, data_page); | |
1889 | dput_out: | |
1d957f9b | 1890 | path_put(&nd.path); |
1da177e4 LT |
1891 | return retval; |
1892 | } | |
1893 | ||
741a2951 JD |
1894 | /* |
1895 | * Allocate a new namespace structure and populate it with contents | |
1896 | * copied from the namespace of the passed in task structure. | |
1897 | */ | |
e3222c4e | 1898 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1899 | struct fs_struct *fs) |
1da177e4 | 1900 | { |
6b3286ed | 1901 | struct mnt_namespace *new_ns; |
1da177e4 | 1902 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1903 | struct vfsmount *p, *q; |
1904 | ||
6b3286ed | 1905 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1906 | if (!new_ns) |
467e9f4b | 1907 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1908 | |
1909 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1910 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1911 | init_waitqueue_head(&new_ns->poll); |
1912 | new_ns->event = 0; | |
1da177e4 | 1913 | |
390c6843 | 1914 | down_write(&namespace_sem); |
1da177e4 | 1915 | /* First pass: copy the tree topology */ |
6b3286ed | 1916 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1917 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1918 | if (!new_ns->root) { |
390c6843 | 1919 | up_write(&namespace_sem); |
1da177e4 | 1920 | kfree(new_ns); |
467e9f4b | 1921 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1922 | } |
1923 | spin_lock(&vfsmount_lock); | |
1924 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1925 | spin_unlock(&vfsmount_lock); | |
1926 | ||
1927 | /* | |
1928 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1929 | * as belonging to new namespace. We have already acquired a private | |
1930 | * fs_struct, so tsk->fs->lock is not needed. | |
1931 | */ | |
6b3286ed | 1932 | p = mnt_ns->root; |
1da177e4 LT |
1933 | q = new_ns->root; |
1934 | while (p) { | |
6b3286ed | 1935 | q->mnt_ns = new_ns; |
1da177e4 | 1936 | if (fs) { |
6ac08c39 | 1937 | if (p == fs->root.mnt) { |
1da177e4 | 1938 | rootmnt = p; |
6ac08c39 | 1939 | fs->root.mnt = mntget(q); |
1da177e4 | 1940 | } |
6ac08c39 | 1941 | if (p == fs->pwd.mnt) { |
1da177e4 | 1942 | pwdmnt = p; |
6ac08c39 | 1943 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1944 | } |
6ac08c39 | 1945 | if (p == fs->altroot.mnt) { |
1da177e4 | 1946 | altrootmnt = p; |
6ac08c39 | 1947 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1948 | } |
1949 | } | |
6b3286ed | 1950 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1951 | q = next_mnt(q, new_ns->root); |
1952 | } | |
390c6843 | 1953 | up_write(&namespace_sem); |
1da177e4 | 1954 | |
1da177e4 LT |
1955 | if (rootmnt) |
1956 | mntput(rootmnt); | |
1957 | if (pwdmnt) | |
1958 | mntput(pwdmnt); | |
1959 | if (altrootmnt) | |
1960 | mntput(altrootmnt); | |
1961 | ||
741a2951 JD |
1962 | return new_ns; |
1963 | } | |
1964 | ||
213dd266 | 1965 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1966 | struct fs_struct *new_fs) |
741a2951 | 1967 | { |
6b3286ed | 1968 | struct mnt_namespace *new_ns; |
741a2951 | 1969 | |
e3222c4e | 1970 | BUG_ON(!ns); |
6b3286ed | 1971 | get_mnt_ns(ns); |
741a2951 JD |
1972 | |
1973 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1974 | return ns; |
741a2951 | 1975 | |
e3222c4e | 1976 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1977 | |
6b3286ed | 1978 | put_mnt_ns(ns); |
e3222c4e | 1979 | return new_ns; |
1da177e4 LT |
1980 | } |
1981 | ||
1982 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1983 | char __user * type, unsigned long flags, | |
1984 | void __user * data) | |
1985 | { | |
1986 | int retval; | |
1987 | unsigned long data_page; | |
1988 | unsigned long type_page; | |
1989 | unsigned long dev_page; | |
1990 | char *dir_page; | |
1991 | ||
b58fed8b | 1992 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1993 | if (retval < 0) |
1994 | return retval; | |
1995 | ||
1996 | dir_page = getname(dir_name); | |
1997 | retval = PTR_ERR(dir_page); | |
1998 | if (IS_ERR(dir_page)) | |
1999 | goto out1; | |
2000 | ||
b58fed8b | 2001 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2002 | if (retval < 0) |
2003 | goto out2; | |
2004 | ||
b58fed8b | 2005 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2006 | if (retval < 0) |
2007 | goto out3; | |
2008 | ||
2009 | lock_kernel(); | |
b58fed8b RP |
2010 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2011 | flags, (void *)data_page); | |
1da177e4 LT |
2012 | unlock_kernel(); |
2013 | free_page(data_page); | |
2014 | ||
2015 | out3: | |
2016 | free_page(dev_page); | |
2017 | out2: | |
2018 | putname(dir_page); | |
2019 | out1: | |
2020 | free_page(type_page); | |
2021 | return retval; | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
2026 | * It can block. Requires the big lock held. | |
2027 | */ | |
ac748a09 | 2028 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 2029 | { |
6ac08c39 JB |
2030 | struct path old_root; |
2031 | ||
1da177e4 LT |
2032 | write_lock(&fs->lock); |
2033 | old_root = fs->root; | |
ac748a09 JB |
2034 | fs->root = *path; |
2035 | path_get(path); | |
1da177e4 | 2036 | write_unlock(&fs->lock); |
6ac08c39 JB |
2037 | if (old_root.dentry) |
2038 | path_put(&old_root); | |
1da177e4 LT |
2039 | } |
2040 | ||
2041 | /* | |
2042 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
2043 | * It can block. Requires the big lock held. | |
2044 | */ | |
ac748a09 | 2045 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 2046 | { |
6ac08c39 | 2047 | struct path old_pwd; |
1da177e4 LT |
2048 | |
2049 | write_lock(&fs->lock); | |
2050 | old_pwd = fs->pwd; | |
ac748a09 JB |
2051 | fs->pwd = *path; |
2052 | path_get(path); | |
1da177e4 LT |
2053 | write_unlock(&fs->lock); |
2054 | ||
6ac08c39 JB |
2055 | if (old_pwd.dentry) |
2056 | path_put(&old_pwd); | |
1da177e4 LT |
2057 | } |
2058 | ||
1a390689 | 2059 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
2060 | { |
2061 | struct task_struct *g, *p; | |
2062 | struct fs_struct *fs; | |
2063 | ||
2064 | read_lock(&tasklist_lock); | |
2065 | do_each_thread(g, p) { | |
2066 | task_lock(p); | |
2067 | fs = p->fs; | |
2068 | if (fs) { | |
2069 | atomic_inc(&fs->count); | |
2070 | task_unlock(p); | |
1a390689 AV |
2071 | if (fs->root.dentry == old_root->dentry |
2072 | && fs->root.mnt == old_root->mnt) | |
2073 | set_fs_root(fs, new_root); | |
2074 | if (fs->pwd.dentry == old_root->dentry | |
2075 | && fs->pwd.mnt == old_root->mnt) | |
2076 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
2077 | put_fs_struct(fs); |
2078 | } else | |
2079 | task_unlock(p); | |
2080 | } while_each_thread(g, p); | |
2081 | read_unlock(&tasklist_lock); | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * pivot_root Semantics: | |
2086 | * Moves the root file system of the current process to the directory put_old, | |
2087 | * makes new_root as the new root file system of the current process, and sets | |
2088 | * root/cwd of all processes which had them on the current root to new_root. | |
2089 | * | |
2090 | * Restrictions: | |
2091 | * The new_root and put_old must be directories, and must not be on the | |
2092 | * same file system as the current process root. The put_old must be | |
2093 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2094 | * pointed to by put_old must yield the same directory as new_root. No other | |
2095 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2096 | * | |
4a0d11fa NB |
2097 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2098 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2099 | * in this situation. | |
2100 | * | |
1da177e4 LT |
2101 | * Notes: |
2102 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2103 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2104 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2105 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2106 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2107 | * first. | |
2108 | */ | |
b58fed8b RP |
2109 | asmlinkage long sys_pivot_root(const char __user * new_root, |
2110 | const char __user * put_old) | |
1da177e4 LT |
2111 | { |
2112 | struct vfsmount *tmp; | |
8c3ee42e AV |
2113 | struct nameidata new_nd, old_nd; |
2114 | struct path parent_path, root_parent, root; | |
1da177e4 LT |
2115 | int error; |
2116 | ||
2117 | if (!capable(CAP_SYS_ADMIN)) | |
2118 | return -EPERM; | |
2119 | ||
b58fed8b RP |
2120 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
2121 | &new_nd); | |
1da177e4 LT |
2122 | if (error) |
2123 | goto out0; | |
2124 | error = -EINVAL; | |
4ac91378 | 2125 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
2126 | goto out1; |
2127 | ||
b58fed8b | 2128 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
2129 | if (error) |
2130 | goto out1; | |
2131 | ||
b5266eb4 | 2132 | error = security_sb_pivotroot(&old_nd.path, &new_nd.path); |
1da177e4 | 2133 | if (error) { |
1d957f9b | 2134 | path_put(&old_nd.path); |
1da177e4 LT |
2135 | goto out1; |
2136 | } | |
2137 | ||
2138 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2139 | root = current->fs->root; |
6ac08c39 | 2140 | path_get(¤t->fs->root); |
1da177e4 | 2141 | read_unlock(¤t->fs->lock); |
390c6843 | 2142 | down_write(&namespace_sem); |
4ac91378 | 2143 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2144 | error = -EINVAL; |
4ac91378 JB |
2145 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
2146 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
8c3ee42e | 2147 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2148 | goto out2; |
8c3ee42e | 2149 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2150 | goto out2; |
2151 | error = -ENOENT; | |
4ac91378 | 2152 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 2153 | goto out2; |
4ac91378 | 2154 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 2155 | goto out2; |
4ac91378 | 2156 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
2157 | goto out2; |
2158 | error = -EBUSY; | |
8c3ee42e AV |
2159 | if (new_nd.path.mnt == root.mnt || |
2160 | old_nd.path.mnt == root.mnt) | |
1da177e4 LT |
2161 | goto out2; /* loop, on the same file system */ |
2162 | error = -EINVAL; | |
8c3ee42e | 2163 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2164 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2165 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2166 | goto out2; /* not attached */ |
4ac91378 | 2167 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 2168 | goto out2; /* not a mountpoint */ |
4ac91378 | 2169 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 2170 | goto out2; /* not attached */ |
4ac91378 JB |
2171 | /* make sure we can reach put_old from new_root */ |
2172 | tmp = old_nd.path.mnt; | |
1da177e4 | 2173 | spin_lock(&vfsmount_lock); |
4ac91378 | 2174 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
2175 | for (;;) { |
2176 | if (tmp->mnt_parent == tmp) | |
2177 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 2178 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
2179 | break; |
2180 | tmp = tmp->mnt_parent; | |
2181 | } | |
4ac91378 | 2182 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 2183 | goto out3; |
4ac91378 | 2184 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 2185 | goto out3; |
1a390689 | 2186 | detach_mnt(new_nd.path.mnt, &parent_path); |
8c3ee42e | 2187 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2188 | /* mount old root on put_old */ |
8c3ee42e | 2189 | attach_mnt(root.mnt, &old_nd.path); |
4ac91378 JB |
2190 | /* mount new_root on / */ |
2191 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 2192 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2193 | spin_unlock(&vfsmount_lock); |
8c3ee42e AV |
2194 | chroot_fs_refs(&root, &new_nd.path); |
2195 | security_sb_post_pivotroot(&root, &new_nd.path); | |
1da177e4 | 2196 | error = 0; |
1a390689 AV |
2197 | path_put(&root_parent); |
2198 | path_put(&parent_path); | |
1da177e4 | 2199 | out2: |
4ac91378 | 2200 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 2201 | up_write(&namespace_sem); |
8c3ee42e | 2202 | path_put(&root); |
1d957f9b | 2203 | path_put(&old_nd.path); |
1da177e4 | 2204 | out1: |
1d957f9b | 2205 | path_put(&new_nd.path); |
1da177e4 | 2206 | out0: |
1da177e4 LT |
2207 | return error; |
2208 | out3: | |
2209 | spin_unlock(&vfsmount_lock); | |
2210 | goto out2; | |
2211 | } | |
2212 | ||
2213 | static void __init init_mount_tree(void) | |
2214 | { | |
2215 | struct vfsmount *mnt; | |
6b3286ed | 2216 | struct mnt_namespace *ns; |
ac748a09 | 2217 | struct path root; |
1da177e4 LT |
2218 | |
2219 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2220 | if (IS_ERR(mnt)) | |
2221 | panic("Can't create rootfs"); | |
6b3286ed KK |
2222 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2223 | if (!ns) | |
1da177e4 | 2224 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2225 | atomic_set(&ns->count, 1); |
2226 | INIT_LIST_HEAD(&ns->list); | |
2227 | init_waitqueue_head(&ns->poll); | |
2228 | ns->event = 0; | |
2229 | list_add(&mnt->mnt_list, &ns->list); | |
2230 | ns->root = mnt; | |
2231 | mnt->mnt_ns = ns; | |
2232 | ||
2233 | init_task.nsproxy->mnt_ns = ns; | |
2234 | get_mnt_ns(ns); | |
2235 | ||
ac748a09 JB |
2236 | root.mnt = ns->root; |
2237 | root.dentry = ns->root->mnt_root; | |
2238 | ||
2239 | set_fs_pwd(current->fs, &root); | |
2240 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2241 | } |
2242 | ||
74bf17cf | 2243 | void __init mnt_init(void) |
1da177e4 | 2244 | { |
13f14b4d | 2245 | unsigned u; |
15a67dd8 | 2246 | int err; |
1da177e4 | 2247 | |
390c6843 RP |
2248 | init_rwsem(&namespace_sem); |
2249 | ||
1da177e4 | 2250 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2251 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2252 | |
b58fed8b | 2253 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2254 | |
2255 | if (!mount_hashtable) | |
2256 | panic("Failed to allocate mount hash table\n"); | |
2257 | ||
13f14b4d ED |
2258 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2259 | ||
2260 | for (u = 0; u < HASH_SIZE; u++) | |
2261 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2262 | |
15a67dd8 RD |
2263 | err = sysfs_init(); |
2264 | if (err) | |
2265 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
2266 | __FUNCTION__, err); | |
00d26666 GKH |
2267 | fs_kobj = kobject_create_and_add("fs", NULL); |
2268 | if (!fs_kobj) | |
2269 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
2270 | init_rootfs(); |
2271 | init_mount_tree(); | |
2272 | } | |
2273 | ||
6b3286ed | 2274 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2275 | { |
6b3286ed | 2276 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2277 | LIST_HEAD(umount_list); |
6b3286ed | 2278 | ns->root = NULL; |
1ce88cf4 | 2279 | spin_unlock(&vfsmount_lock); |
390c6843 | 2280 | down_write(&namespace_sem); |
1da177e4 | 2281 | spin_lock(&vfsmount_lock); |
a05964f3 | 2282 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2283 | spin_unlock(&vfsmount_lock); |
390c6843 | 2284 | up_write(&namespace_sem); |
70fbcdf4 | 2285 | release_mounts(&umount_list); |
6b3286ed | 2286 | kfree(ns); |
1da177e4 | 2287 | } |