]>
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); |
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 | ||
1da177e4 LT |
86 | struct vfsmount *alloc_vfsmnt(const char *name) |
87 | { | |
c3762229 | 88 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 89 | if (mnt) { |
73cd49ec MS |
90 | int err; |
91 | ||
92 | err = mnt_alloc_id(mnt); | |
93 | if (err) { | |
94 | kmem_cache_free(mnt_cache, mnt); | |
95 | return NULL; | |
96 | } | |
97 | ||
b58fed8b | 98 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
99 | INIT_LIST_HEAD(&mnt->mnt_hash); |
100 | INIT_LIST_HEAD(&mnt->mnt_child); | |
101 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
102 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 103 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 104 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
105 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
106 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
3d733633 | 107 | atomic_set(&mnt->__mnt_writers, 0); |
1da177e4 | 108 | if (name) { |
b58fed8b | 109 | int size = strlen(name) + 1; |
1da177e4 LT |
110 | char *newname = kmalloc(size, GFP_KERNEL); |
111 | if (newname) { | |
112 | memcpy(newname, name, size); | |
113 | mnt->mnt_devname = newname; | |
114 | } | |
115 | } | |
116 | } | |
117 | return mnt; | |
118 | } | |
119 | ||
3d733633 DH |
120 | /* |
121 | * Most r/o checks on a fs are for operations that take | |
122 | * discrete amounts of time, like a write() or unlink(). | |
123 | * We must keep track of when those operations start | |
124 | * (for permission checks) and when they end, so that | |
125 | * we can determine when writes are able to occur to | |
126 | * a filesystem. | |
127 | */ | |
128 | /* | |
129 | * __mnt_is_readonly: check whether a mount is read-only | |
130 | * @mnt: the mount to check for its write status | |
131 | * | |
132 | * This shouldn't be used directly ouside of the VFS. | |
133 | * It does not guarantee that the filesystem will stay | |
134 | * r/w, just that it is right *now*. This can not and | |
135 | * should not be used in place of IS_RDONLY(inode). | |
136 | * mnt_want/drop_write() will _keep_ the filesystem | |
137 | * r/w. | |
138 | */ | |
139 | int __mnt_is_readonly(struct vfsmount *mnt) | |
140 | { | |
2e4b7fcd DH |
141 | if (mnt->mnt_flags & MNT_READONLY) |
142 | return 1; | |
143 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
144 | return 1; | |
145 | return 0; | |
3d733633 DH |
146 | } |
147 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
148 | ||
149 | struct mnt_writer { | |
150 | /* | |
151 | * If holding multiple instances of this lock, they | |
152 | * must be ordered by cpu number. | |
153 | */ | |
154 | spinlock_t lock; | |
155 | struct lock_class_key lock_class; /* compiles out with !lockdep */ | |
156 | unsigned long count; | |
157 | struct vfsmount *mnt; | |
158 | } ____cacheline_aligned_in_smp; | |
159 | static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); | |
160 | ||
161 | static int __init init_mnt_writers(void) | |
162 | { | |
163 | int cpu; | |
164 | for_each_possible_cpu(cpu) { | |
165 | struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); | |
166 | spin_lock_init(&writer->lock); | |
167 | lockdep_set_class(&writer->lock, &writer->lock_class); | |
168 | writer->count = 0; | |
169 | } | |
170 | return 0; | |
171 | } | |
172 | fs_initcall(init_mnt_writers); | |
173 | ||
174 | static void unlock_mnt_writers(void) | |
175 | { | |
176 | int cpu; | |
177 | struct mnt_writer *cpu_writer; | |
178 | ||
179 | for_each_possible_cpu(cpu) { | |
180 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
181 | spin_unlock(&cpu_writer->lock); | |
182 | } | |
183 | } | |
184 | ||
185 | static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) | |
186 | { | |
187 | if (!cpu_writer->mnt) | |
188 | return; | |
189 | /* | |
190 | * This is in case anyone ever leaves an invalid, | |
191 | * old ->mnt and a count of 0. | |
192 | */ | |
193 | if (!cpu_writer->count) | |
194 | return; | |
195 | atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); | |
196 | cpu_writer->count = 0; | |
197 | } | |
198 | /* | |
199 | * must hold cpu_writer->lock | |
200 | */ | |
201 | static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, | |
202 | struct vfsmount *mnt) | |
203 | { | |
204 | if (cpu_writer->mnt == mnt) | |
205 | return; | |
206 | __clear_mnt_count(cpu_writer); | |
207 | cpu_writer->mnt = mnt; | |
208 | } | |
209 | ||
8366025e DH |
210 | /* |
211 | * Most r/o checks on a fs are for operations that take | |
212 | * discrete amounts of time, like a write() or unlink(). | |
213 | * We must keep track of when those operations start | |
214 | * (for permission checks) and when they end, so that | |
215 | * we can determine when writes are able to occur to | |
216 | * a filesystem. | |
217 | */ | |
218 | /** | |
219 | * mnt_want_write - get write access to a mount | |
220 | * @mnt: the mount on which to take a write | |
221 | * | |
222 | * This tells the low-level filesystem that a write is | |
223 | * about to be performed to it, and makes sure that | |
224 | * writes are allowed before returning success. When | |
225 | * the write operation is finished, mnt_drop_write() | |
226 | * must be called. This is effectively a refcount. | |
227 | */ | |
228 | int mnt_want_write(struct vfsmount *mnt) | |
229 | { | |
3d733633 DH |
230 | int ret = 0; |
231 | struct mnt_writer *cpu_writer; | |
232 | ||
233 | cpu_writer = &get_cpu_var(mnt_writers); | |
234 | spin_lock(&cpu_writer->lock); | |
235 | if (__mnt_is_readonly(mnt)) { | |
236 | ret = -EROFS; | |
237 | goto out; | |
238 | } | |
239 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
240 | cpu_writer->count++; | |
241 | out: | |
242 | spin_unlock(&cpu_writer->lock); | |
243 | put_cpu_var(mnt_writers); | |
244 | return ret; | |
8366025e DH |
245 | } |
246 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
247 | ||
3d733633 DH |
248 | static void lock_mnt_writers(void) |
249 | { | |
250 | int cpu; | |
251 | struct mnt_writer *cpu_writer; | |
252 | ||
253 | for_each_possible_cpu(cpu) { | |
254 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
255 | spin_lock(&cpu_writer->lock); | |
256 | __clear_mnt_count(cpu_writer); | |
257 | cpu_writer->mnt = NULL; | |
258 | } | |
259 | } | |
260 | ||
261 | /* | |
262 | * These per-cpu write counts are not guaranteed to have | |
263 | * matched increments and decrements on any given cpu. | |
264 | * A file open()ed for write on one cpu and close()d on | |
265 | * another cpu will imbalance this count. Make sure it | |
266 | * does not get too far out of whack. | |
267 | */ | |
268 | static void handle_write_count_underflow(struct vfsmount *mnt) | |
269 | { | |
270 | if (atomic_read(&mnt->__mnt_writers) >= | |
271 | MNT_WRITER_UNDERFLOW_LIMIT) | |
272 | return; | |
273 | /* | |
274 | * It isn't necessary to hold all of the locks | |
275 | * at the same time, but doing it this way makes | |
276 | * us share a lot more code. | |
277 | */ | |
278 | lock_mnt_writers(); | |
279 | /* | |
280 | * vfsmount_lock is for mnt_flags. | |
281 | */ | |
282 | spin_lock(&vfsmount_lock); | |
283 | /* | |
284 | * If coalescing the per-cpu writer counts did not | |
285 | * get us back to a positive writer count, we have | |
286 | * a bug. | |
287 | */ | |
288 | if ((atomic_read(&mnt->__mnt_writers) < 0) && | |
289 | !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { | |
290 | printk(KERN_DEBUG "leak detected on mount(%p) writers " | |
291 | "count: %d\n", | |
292 | mnt, atomic_read(&mnt->__mnt_writers)); | |
293 | WARN_ON(1); | |
294 | /* use the flag to keep the dmesg spam down */ | |
295 | mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; | |
296 | } | |
297 | spin_unlock(&vfsmount_lock); | |
298 | unlock_mnt_writers(); | |
299 | } | |
300 | ||
8366025e DH |
301 | /** |
302 | * mnt_drop_write - give up write access to a mount | |
303 | * @mnt: the mount on which to give up write access | |
304 | * | |
305 | * Tells the low-level filesystem that we are done | |
306 | * performing writes to it. Must be matched with | |
307 | * mnt_want_write() call above. | |
308 | */ | |
309 | void mnt_drop_write(struct vfsmount *mnt) | |
310 | { | |
3d733633 DH |
311 | int must_check_underflow = 0; |
312 | struct mnt_writer *cpu_writer; | |
313 | ||
314 | cpu_writer = &get_cpu_var(mnt_writers); | |
315 | spin_lock(&cpu_writer->lock); | |
316 | ||
317 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
318 | if (cpu_writer->count > 0) { | |
319 | cpu_writer->count--; | |
320 | } else { | |
321 | must_check_underflow = 1; | |
322 | atomic_dec(&mnt->__mnt_writers); | |
323 | } | |
324 | ||
325 | spin_unlock(&cpu_writer->lock); | |
326 | /* | |
327 | * Logically, we could call this each time, | |
328 | * but the __mnt_writers cacheline tends to | |
329 | * be cold, and makes this expensive. | |
330 | */ | |
331 | if (must_check_underflow) | |
332 | handle_write_count_underflow(mnt); | |
333 | /* | |
334 | * This could be done right after the spinlock | |
335 | * is taken because the spinlock keeps us on | |
336 | * the cpu, and disables preemption. However, | |
337 | * putting it here bounds the amount that | |
338 | * __mnt_writers can underflow. Without it, | |
339 | * we could theoretically wrap __mnt_writers. | |
340 | */ | |
341 | put_cpu_var(mnt_writers); | |
8366025e DH |
342 | } |
343 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
344 | ||
2e4b7fcd | 345 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 346 | { |
3d733633 DH |
347 | int ret = 0; |
348 | ||
349 | lock_mnt_writers(); | |
350 | /* | |
351 | * With all the locks held, this value is stable | |
352 | */ | |
353 | if (atomic_read(&mnt->__mnt_writers) > 0) { | |
354 | ret = -EBUSY; | |
355 | goto out; | |
356 | } | |
357 | /* | |
2e4b7fcd DH |
358 | * nobody can do a successful mnt_want_write() with all |
359 | * of the counts in MNT_DENIED_WRITE and the locks held. | |
3d733633 | 360 | */ |
2e4b7fcd DH |
361 | spin_lock(&vfsmount_lock); |
362 | if (!ret) | |
363 | mnt->mnt_flags |= MNT_READONLY; | |
364 | spin_unlock(&vfsmount_lock); | |
3d733633 DH |
365 | out: |
366 | unlock_mnt_writers(); | |
367 | return ret; | |
8366025e | 368 | } |
8366025e | 369 | |
2e4b7fcd DH |
370 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
371 | { | |
372 | spin_lock(&vfsmount_lock); | |
373 | mnt->mnt_flags &= ~MNT_READONLY; | |
374 | spin_unlock(&vfsmount_lock); | |
375 | } | |
376 | ||
454e2398 DH |
377 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
378 | { | |
379 | mnt->mnt_sb = sb; | |
380 | mnt->mnt_root = dget(sb->s_root); | |
381 | return 0; | |
382 | } | |
383 | ||
384 | EXPORT_SYMBOL(simple_set_mnt); | |
385 | ||
1da177e4 LT |
386 | void free_vfsmnt(struct vfsmount *mnt) |
387 | { | |
388 | kfree(mnt->mnt_devname); | |
73cd49ec | 389 | mnt_free_id(mnt); |
1da177e4 LT |
390 | kmem_cache_free(mnt_cache, mnt); |
391 | } | |
392 | ||
393 | /* | |
a05964f3 RP |
394 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
395 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 396 | */ |
a05964f3 RP |
397 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
398 | int dir) | |
1da177e4 | 399 | { |
b58fed8b RP |
400 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
401 | struct list_head *tmp = head; | |
1da177e4 LT |
402 | struct vfsmount *p, *found = NULL; |
403 | ||
1da177e4 | 404 | for (;;) { |
a05964f3 | 405 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
406 | p = NULL; |
407 | if (tmp == head) | |
408 | break; | |
409 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
410 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 411 | found = p; |
1da177e4 LT |
412 | break; |
413 | } | |
414 | } | |
1da177e4 LT |
415 | return found; |
416 | } | |
417 | ||
a05964f3 RP |
418 | /* |
419 | * lookup_mnt increments the ref count before returning | |
420 | * the vfsmount struct. | |
421 | */ | |
422 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
423 | { | |
424 | struct vfsmount *child_mnt; | |
425 | spin_lock(&vfsmount_lock); | |
426 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
427 | mntget(child_mnt); | |
428 | spin_unlock(&vfsmount_lock); | |
429 | return child_mnt; | |
430 | } | |
431 | ||
1da177e4 LT |
432 | static inline int check_mnt(struct vfsmount *mnt) |
433 | { | |
6b3286ed | 434 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
435 | } |
436 | ||
6b3286ed | 437 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
438 | { |
439 | if (ns) { | |
440 | ns->event = ++event; | |
441 | wake_up_interruptible(&ns->poll); | |
442 | } | |
443 | } | |
444 | ||
6b3286ed | 445 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
446 | { |
447 | if (ns && ns->event != event) { | |
448 | ns->event = event; | |
449 | wake_up_interruptible(&ns->poll); | |
450 | } | |
451 | } | |
452 | ||
1a390689 | 453 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 454 | { |
1a390689 AV |
455 | old_path->dentry = mnt->mnt_mountpoint; |
456 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
457 | mnt->mnt_parent = mnt; |
458 | mnt->mnt_mountpoint = mnt->mnt_root; | |
459 | list_del_init(&mnt->mnt_child); | |
460 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 461 | old_path->dentry->d_mounted--; |
1da177e4 LT |
462 | } |
463 | ||
b90fa9ae RP |
464 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
465 | struct vfsmount *child_mnt) | |
466 | { | |
467 | child_mnt->mnt_parent = mntget(mnt); | |
468 | child_mnt->mnt_mountpoint = dget(dentry); | |
469 | dentry->d_mounted++; | |
470 | } | |
471 | ||
1a390689 | 472 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 473 | { |
1a390689 | 474 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 475 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
476 | hash(path->mnt, path->dentry)); |
477 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
478 | } |
479 | ||
480 | /* | |
481 | * the caller must hold vfsmount_lock | |
482 | */ | |
483 | static void commit_tree(struct vfsmount *mnt) | |
484 | { | |
485 | struct vfsmount *parent = mnt->mnt_parent; | |
486 | struct vfsmount *m; | |
487 | LIST_HEAD(head); | |
6b3286ed | 488 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
489 | |
490 | BUG_ON(parent == mnt); | |
491 | ||
492 | list_add_tail(&head, &mnt->mnt_list); | |
493 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 494 | m->mnt_ns = n; |
b90fa9ae RP |
495 | list_splice(&head, n->list.prev); |
496 | ||
497 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
498 | hash(parent, mnt->mnt_mountpoint)); | |
499 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 500 | touch_mnt_namespace(n); |
1da177e4 LT |
501 | } |
502 | ||
503 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
504 | { | |
505 | struct list_head *next = p->mnt_mounts.next; | |
506 | if (next == &p->mnt_mounts) { | |
507 | while (1) { | |
508 | if (p == root) | |
509 | return NULL; | |
510 | next = p->mnt_child.next; | |
511 | if (next != &p->mnt_parent->mnt_mounts) | |
512 | break; | |
513 | p = p->mnt_parent; | |
514 | } | |
515 | } | |
516 | return list_entry(next, struct vfsmount, mnt_child); | |
517 | } | |
518 | ||
9676f0c6 RP |
519 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
520 | { | |
521 | struct list_head *prev = p->mnt_mounts.prev; | |
522 | while (prev != &p->mnt_mounts) { | |
523 | p = list_entry(prev, struct vfsmount, mnt_child); | |
524 | prev = p->mnt_mounts.prev; | |
525 | } | |
526 | return p; | |
527 | } | |
528 | ||
36341f64 RP |
529 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
530 | int flag) | |
1da177e4 LT |
531 | { |
532 | struct super_block *sb = old->mnt_sb; | |
533 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
534 | ||
535 | if (mnt) { | |
536 | mnt->mnt_flags = old->mnt_flags; | |
537 | atomic_inc(&sb->s_active); | |
538 | mnt->mnt_sb = sb; | |
539 | mnt->mnt_root = dget(root); | |
540 | mnt->mnt_mountpoint = mnt->mnt_root; | |
541 | mnt->mnt_parent = mnt; | |
b90fa9ae | 542 | |
5afe0022 RP |
543 | if (flag & CL_SLAVE) { |
544 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
545 | mnt->mnt_master = old; | |
546 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 547 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
548 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
549 | list_add(&mnt->mnt_share, &old->mnt_share); | |
550 | if (IS_MNT_SLAVE(old)) | |
551 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
552 | mnt->mnt_master = old->mnt_master; | |
553 | } | |
b90fa9ae RP |
554 | if (flag & CL_MAKE_SHARED) |
555 | set_mnt_shared(mnt); | |
1da177e4 LT |
556 | |
557 | /* stick the duplicate mount on the same expiry list | |
558 | * as the original if that was on one */ | |
36341f64 | 559 | if (flag & CL_EXPIRE) { |
36341f64 RP |
560 | if (!list_empty(&old->mnt_expire)) |
561 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 562 | } |
1da177e4 LT |
563 | } |
564 | return mnt; | |
565 | } | |
566 | ||
7b7b1ace | 567 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 | 568 | { |
3d733633 | 569 | int cpu; |
1da177e4 | 570 | struct super_block *sb = mnt->mnt_sb; |
3d733633 DH |
571 | /* |
572 | * We don't have to hold all of the locks at the | |
573 | * same time here because we know that we're the | |
574 | * last reference to mnt and that no new writers | |
575 | * can come in. | |
576 | */ | |
577 | for_each_possible_cpu(cpu) { | |
578 | struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); | |
579 | if (cpu_writer->mnt != mnt) | |
580 | continue; | |
581 | spin_lock(&cpu_writer->lock); | |
582 | atomic_add(cpu_writer->count, &mnt->__mnt_writers); | |
583 | cpu_writer->count = 0; | |
584 | /* | |
585 | * Might as well do this so that no one | |
586 | * ever sees the pointer and expects | |
587 | * it to be valid. | |
588 | */ | |
589 | cpu_writer->mnt = NULL; | |
590 | spin_unlock(&cpu_writer->lock); | |
591 | } | |
592 | /* | |
593 | * This probably indicates that somebody messed | |
594 | * up a mnt_want/drop_write() pair. If this | |
595 | * happens, the filesystem was probably unable | |
596 | * to make r/w->r/o transitions. | |
597 | */ | |
598 | WARN_ON(atomic_read(&mnt->__mnt_writers)); | |
1da177e4 LT |
599 | dput(mnt->mnt_root); |
600 | free_vfsmnt(mnt); | |
601 | deactivate_super(sb); | |
602 | } | |
603 | ||
7b7b1ace AV |
604 | void mntput_no_expire(struct vfsmount *mnt) |
605 | { | |
606 | repeat: | |
607 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
608 | if (likely(!mnt->mnt_pinned)) { | |
609 | spin_unlock(&vfsmount_lock); | |
610 | __mntput(mnt); | |
611 | return; | |
612 | } | |
613 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
614 | mnt->mnt_pinned = 0; | |
615 | spin_unlock(&vfsmount_lock); | |
616 | acct_auto_close_mnt(mnt); | |
617 | security_sb_umount_close(mnt); | |
618 | goto repeat; | |
619 | } | |
620 | } | |
621 | ||
622 | EXPORT_SYMBOL(mntput_no_expire); | |
623 | ||
624 | void mnt_pin(struct vfsmount *mnt) | |
625 | { | |
626 | spin_lock(&vfsmount_lock); | |
627 | mnt->mnt_pinned++; | |
628 | spin_unlock(&vfsmount_lock); | |
629 | } | |
630 | ||
631 | EXPORT_SYMBOL(mnt_pin); | |
632 | ||
633 | void mnt_unpin(struct vfsmount *mnt) | |
634 | { | |
635 | spin_lock(&vfsmount_lock); | |
636 | if (mnt->mnt_pinned) { | |
637 | atomic_inc(&mnt->mnt_count); | |
638 | mnt->mnt_pinned--; | |
639 | } | |
640 | spin_unlock(&vfsmount_lock); | |
641 | } | |
642 | ||
643 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 644 | |
b3b304a2 MS |
645 | static inline void mangle(struct seq_file *m, const char *s) |
646 | { | |
647 | seq_escape(m, s, " \t\n\\"); | |
648 | } | |
649 | ||
650 | /* | |
651 | * Simple .show_options callback for filesystems which don't want to | |
652 | * implement more complex mount option showing. | |
653 | * | |
654 | * See also save_mount_options(). | |
655 | */ | |
656 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
657 | { | |
658 | const char *options = mnt->mnt_sb->s_options; | |
659 | ||
660 | if (options != NULL && options[0]) { | |
661 | seq_putc(m, ','); | |
662 | mangle(m, options); | |
663 | } | |
664 | ||
665 | return 0; | |
666 | } | |
667 | EXPORT_SYMBOL(generic_show_options); | |
668 | ||
669 | /* | |
670 | * If filesystem uses generic_show_options(), this function should be | |
671 | * called from the fill_super() callback. | |
672 | * | |
673 | * The .remount_fs callback usually needs to be handled in a special | |
674 | * way, to make sure, that previous options are not overwritten if the | |
675 | * remount fails. | |
676 | * | |
677 | * Also note, that if the filesystem's .remount_fs function doesn't | |
678 | * reset all options to their default value, but changes only newly | |
679 | * given options, then the displayed options will not reflect reality | |
680 | * any more. | |
681 | */ | |
682 | void save_mount_options(struct super_block *sb, char *options) | |
683 | { | |
684 | kfree(sb->s_options); | |
685 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
686 | } | |
687 | EXPORT_SYMBOL(save_mount_options); | |
688 | ||
1da177e4 LT |
689 | /* iterator */ |
690 | static void *m_start(struct seq_file *m, loff_t *pos) | |
691 | { | |
6b3286ed | 692 | struct mnt_namespace *n = m->private; |
1da177e4 | 693 | |
390c6843 | 694 | down_read(&namespace_sem); |
b0765fb8 | 695 | return seq_list_start(&n->list, *pos); |
1da177e4 LT |
696 | } |
697 | ||
698 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
699 | { | |
6b3286ed | 700 | struct mnt_namespace *n = m->private; |
b0765fb8 PE |
701 | |
702 | return seq_list_next(v, &n->list, pos); | |
1da177e4 LT |
703 | } |
704 | ||
705 | static void m_stop(struct seq_file *m, void *v) | |
706 | { | |
390c6843 | 707 | up_read(&namespace_sem); |
1da177e4 LT |
708 | } |
709 | ||
1da177e4 LT |
710 | static int show_vfsmnt(struct seq_file *m, void *v) |
711 | { | |
b0765fb8 | 712 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
1da177e4 LT |
713 | int err = 0; |
714 | static struct proc_fs_info { | |
715 | int flag; | |
716 | char *str; | |
717 | } fs_info[] = { | |
718 | { MS_SYNCHRONOUS, ",sync" }, | |
719 | { MS_DIRSYNC, ",dirsync" }, | |
720 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
721 | { 0, NULL } |
722 | }; | |
723 | static struct proc_fs_info mnt_info[] = { | |
724 | { MNT_NOSUID, ",nosuid" }, | |
725 | { MNT_NODEV, ",nodev" }, | |
726 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
727 | { MNT_NOATIME, ",noatime" }, |
728 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 729 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
730 | { 0, NULL } |
731 | }; | |
732 | struct proc_fs_info *fs_infop; | |
c32c2f63 | 733 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
734 | |
735 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
736 | seq_putc(m, ' '); | |
c32c2f63 | 737 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 LT |
738 | seq_putc(m, ' '); |
739 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
740 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { |
741 | seq_putc(m, '.'); | |
742 | mangle(m, mnt->mnt_sb->s_subtype); | |
743 | } | |
2e4b7fcd | 744 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
1da177e4 LT |
745 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { |
746 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
747 | seq_puts(m, fs_infop->str); | |
748 | } | |
749 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
750 | if (mnt->mnt_flags & fs_infop->flag) | |
751 | seq_puts(m, fs_infop->str); | |
752 | } | |
753 | if (mnt->mnt_sb->s_op->show_options) | |
754 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
755 | seq_puts(m, " 0 0\n"); | |
756 | return err; | |
757 | } | |
758 | ||
759 | struct seq_operations mounts_op = { | |
760 | .start = m_start, | |
761 | .next = m_next, | |
762 | .stop = m_stop, | |
763 | .show = show_vfsmnt | |
764 | }; | |
765 | ||
b4629fe2 CL |
766 | static int show_vfsstat(struct seq_file *m, void *v) |
767 | { | |
b0765fb8 | 768 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 769 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
770 | int err = 0; |
771 | ||
772 | /* device */ | |
773 | if (mnt->mnt_devname) { | |
774 | seq_puts(m, "device "); | |
775 | mangle(m, mnt->mnt_devname); | |
776 | } else | |
777 | seq_puts(m, "no device"); | |
778 | ||
779 | /* mount point */ | |
780 | seq_puts(m, " mounted on "); | |
c32c2f63 | 781 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
782 | seq_putc(m, ' '); |
783 | ||
784 | /* file system type */ | |
785 | seq_puts(m, "with fstype "); | |
786 | mangle(m, mnt->mnt_sb->s_type->name); | |
787 | ||
788 | /* optional statistics */ | |
789 | if (mnt->mnt_sb->s_op->show_stats) { | |
790 | seq_putc(m, ' '); | |
791 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
792 | } | |
793 | ||
794 | seq_putc(m, '\n'); | |
795 | return err; | |
796 | } | |
797 | ||
798 | struct seq_operations mountstats_op = { | |
799 | .start = m_start, | |
800 | .next = m_next, | |
801 | .stop = m_stop, | |
802 | .show = show_vfsstat, | |
803 | }; | |
804 | ||
1da177e4 LT |
805 | /** |
806 | * may_umount_tree - check if a mount tree is busy | |
807 | * @mnt: root of mount tree | |
808 | * | |
809 | * This is called to check if a tree of mounts has any | |
810 | * open files, pwds, chroots or sub mounts that are | |
811 | * busy. | |
812 | */ | |
813 | int may_umount_tree(struct vfsmount *mnt) | |
814 | { | |
36341f64 RP |
815 | int actual_refs = 0; |
816 | int minimum_refs = 0; | |
817 | struct vfsmount *p; | |
1da177e4 LT |
818 | |
819 | spin_lock(&vfsmount_lock); | |
36341f64 | 820 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
821 | actual_refs += atomic_read(&p->mnt_count); |
822 | minimum_refs += 2; | |
1da177e4 LT |
823 | } |
824 | spin_unlock(&vfsmount_lock); | |
825 | ||
826 | if (actual_refs > minimum_refs) | |
e3474a8e | 827 | return 0; |
1da177e4 | 828 | |
e3474a8e | 829 | return 1; |
1da177e4 LT |
830 | } |
831 | ||
832 | EXPORT_SYMBOL(may_umount_tree); | |
833 | ||
834 | /** | |
835 | * may_umount - check if a mount point is busy | |
836 | * @mnt: root of mount | |
837 | * | |
838 | * This is called to check if a mount point has any | |
839 | * open files, pwds, chroots or sub mounts. If the | |
840 | * mount has sub mounts this will return busy | |
841 | * regardless of whether the sub mounts are busy. | |
842 | * | |
843 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
844 | * give false negatives. The main reason why it's here is that we need | |
845 | * a non-destructive way to look for easily umountable filesystems. | |
846 | */ | |
847 | int may_umount(struct vfsmount *mnt) | |
848 | { | |
e3474a8e | 849 | int ret = 1; |
a05964f3 RP |
850 | spin_lock(&vfsmount_lock); |
851 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 852 | ret = 0; |
a05964f3 RP |
853 | spin_unlock(&vfsmount_lock); |
854 | return ret; | |
1da177e4 LT |
855 | } |
856 | ||
857 | EXPORT_SYMBOL(may_umount); | |
858 | ||
b90fa9ae | 859 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
860 | { |
861 | struct vfsmount *mnt; | |
bf066c7d | 862 | while (!list_empty(head)) { |
b5e61818 | 863 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
864 | list_del_init(&mnt->mnt_hash); |
865 | if (mnt->mnt_parent != mnt) { | |
866 | struct dentry *dentry; | |
867 | struct vfsmount *m; | |
868 | spin_lock(&vfsmount_lock); | |
869 | dentry = mnt->mnt_mountpoint; | |
870 | m = mnt->mnt_parent; | |
871 | mnt->mnt_mountpoint = mnt->mnt_root; | |
872 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 873 | m->mnt_ghosts--; |
70fbcdf4 RP |
874 | spin_unlock(&vfsmount_lock); |
875 | dput(dentry); | |
876 | mntput(m); | |
877 | } | |
878 | mntput(mnt); | |
879 | } | |
880 | } | |
881 | ||
a05964f3 | 882 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
883 | { |
884 | struct vfsmount *p; | |
1da177e4 | 885 | |
1bfba4e8 AM |
886 | for (p = mnt; p; p = next_mnt(p, mnt)) |
887 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 888 | |
a05964f3 RP |
889 | if (propagate) |
890 | propagate_umount(kill); | |
891 | ||
70fbcdf4 RP |
892 | list_for_each_entry(p, kill, mnt_hash) { |
893 | list_del_init(&p->mnt_expire); | |
894 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
895 | __touch_mnt_namespace(p->mnt_ns); |
896 | p->mnt_ns = NULL; | |
70fbcdf4 | 897 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
898 | if (p->mnt_parent != p) { |
899 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 900 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 901 | } |
a05964f3 | 902 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
903 | } |
904 | } | |
905 | ||
c35038be AV |
906 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
907 | ||
1da177e4 LT |
908 | static int do_umount(struct vfsmount *mnt, int flags) |
909 | { | |
b58fed8b | 910 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 911 | int retval; |
70fbcdf4 | 912 | LIST_HEAD(umount_list); |
1da177e4 LT |
913 | |
914 | retval = security_sb_umount(mnt, flags); | |
915 | if (retval) | |
916 | return retval; | |
917 | ||
918 | /* | |
919 | * Allow userspace to request a mountpoint be expired rather than | |
920 | * unmounting unconditionally. Unmount only happens if: | |
921 | * (1) the mark is already set (the mark is cleared by mntput()) | |
922 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
923 | */ | |
924 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 925 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
926 | flags & (MNT_FORCE | MNT_DETACH)) |
927 | return -EINVAL; | |
928 | ||
929 | if (atomic_read(&mnt->mnt_count) != 2) | |
930 | return -EBUSY; | |
931 | ||
932 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
933 | return -EAGAIN; | |
934 | } | |
935 | ||
936 | /* | |
937 | * If we may have to abort operations to get out of this | |
938 | * mount, and they will themselves hold resources we must | |
939 | * allow the fs to do things. In the Unix tradition of | |
940 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
941 | * might fail to complete on the first run through as other tasks | |
942 | * must return, and the like. Thats for the mount program to worry | |
943 | * about for the moment. | |
944 | */ | |
945 | ||
946 | lock_kernel(); | |
8b512d9a TM |
947 | if (sb->s_op->umount_begin) |
948 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
949 | unlock_kernel(); |
950 | ||
951 | /* | |
952 | * No sense to grab the lock for this test, but test itself looks | |
953 | * somewhat bogus. Suggestions for better replacement? | |
954 | * Ho-hum... In principle, we might treat that as umount + switch | |
955 | * to rootfs. GC would eventually take care of the old vfsmount. | |
956 | * Actually it makes sense, especially if rootfs would contain a | |
957 | * /reboot - static binary that would close all descriptors and | |
958 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
959 | */ | |
6ac08c39 | 960 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
961 | /* |
962 | * Special case for "unmounting" root ... | |
963 | * we just try to remount it readonly. | |
964 | */ | |
965 | down_write(&sb->s_umount); | |
966 | if (!(sb->s_flags & MS_RDONLY)) { | |
967 | lock_kernel(); | |
968 | DQUOT_OFF(sb); | |
969 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
970 | unlock_kernel(); | |
971 | } | |
972 | up_write(&sb->s_umount); | |
973 | return retval; | |
974 | } | |
975 | ||
390c6843 | 976 | down_write(&namespace_sem); |
1da177e4 | 977 | spin_lock(&vfsmount_lock); |
5addc5dd | 978 | event++; |
1da177e4 | 979 | |
c35038be AV |
980 | if (!(flags & MNT_DETACH)) |
981 | shrink_submounts(mnt, &umount_list); | |
982 | ||
1da177e4 | 983 | retval = -EBUSY; |
a05964f3 | 984 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 985 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 986 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
987 | retval = 0; |
988 | } | |
989 | spin_unlock(&vfsmount_lock); | |
990 | if (retval) | |
991 | security_sb_umount_busy(mnt); | |
390c6843 | 992 | up_write(&namespace_sem); |
70fbcdf4 | 993 | release_mounts(&umount_list); |
1da177e4 LT |
994 | return retval; |
995 | } | |
996 | ||
997 | /* | |
998 | * Now umount can handle mount points as well as block devices. | |
999 | * This is important for filesystems which use unnamed block devices. | |
1000 | * | |
1001 | * We now support a flag for forced unmount like the other 'big iron' | |
1002 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1003 | */ | |
1004 | ||
1005 | asmlinkage long sys_umount(char __user * name, int flags) | |
1006 | { | |
1007 | struct nameidata nd; | |
1008 | int retval; | |
1009 | ||
1010 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
1011 | if (retval) | |
1012 | goto out; | |
1013 | retval = -EINVAL; | |
4ac91378 | 1014 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 1015 | goto dput_and_out; |
4ac91378 | 1016 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
1017 | goto dput_and_out; |
1018 | ||
1019 | retval = -EPERM; | |
1020 | if (!capable(CAP_SYS_ADMIN)) | |
1021 | goto dput_and_out; | |
1022 | ||
4ac91378 | 1023 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 1024 | dput_and_out: |
429731b1 | 1025 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
1026 | dput(nd.path.dentry); |
1027 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
1028 | out: |
1029 | return retval; | |
1030 | } | |
1031 | ||
1032 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1033 | ||
1034 | /* | |
b58fed8b | 1035 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1036 | */ |
1da177e4 LT |
1037 | asmlinkage long sys_oldumount(char __user * name) |
1038 | { | |
b58fed8b | 1039 | return sys_umount(name, 0); |
1da177e4 LT |
1040 | } |
1041 | ||
1042 | #endif | |
1043 | ||
1044 | static int mount_is_safe(struct nameidata *nd) | |
1045 | { | |
1046 | if (capable(CAP_SYS_ADMIN)) | |
1047 | return 0; | |
1048 | return -EPERM; | |
1049 | #ifdef notyet | |
4ac91378 | 1050 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 1051 | return -EPERM; |
4ac91378 JB |
1052 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
1053 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
1054 | return -EPERM; |
1055 | } | |
e4543edd | 1056 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
1057 | return -EPERM; |
1058 | return 0; | |
1059 | #endif | |
1060 | } | |
1061 | ||
b58fed8b | 1062 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
1063 | { |
1064 | while (1) { | |
1065 | if (d == dentry) | |
1066 | return 1; | |
1067 | if (d == NULL || d == d->d_parent) | |
1068 | return 0; | |
1069 | d = d->d_parent; | |
1070 | } | |
1071 | } | |
1072 | ||
b90fa9ae | 1073 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1074 | int flag) |
1da177e4 LT |
1075 | { |
1076 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1077 | struct path path; |
1da177e4 | 1078 | |
9676f0c6 RP |
1079 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1080 | return NULL; | |
1081 | ||
36341f64 | 1082 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1083 | if (!q) |
1084 | goto Enomem; | |
1085 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1086 | ||
1087 | p = mnt; | |
fdadd65f | 1088 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
1089 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
1090 | continue; | |
1091 | ||
1092 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1093 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1094 | s = skip_mnt_tree(s); | |
1095 | continue; | |
1096 | } | |
1da177e4 LT |
1097 | while (p != s->mnt_parent) { |
1098 | p = p->mnt_parent; | |
1099 | q = q->mnt_parent; | |
1100 | } | |
1101 | p = s; | |
1a390689 AV |
1102 | path.mnt = q; |
1103 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1104 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1105 | if (!q) |
1106 | goto Enomem; | |
1107 | spin_lock(&vfsmount_lock); | |
1108 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1109 | attach_mnt(q, &path); |
1da177e4 LT |
1110 | spin_unlock(&vfsmount_lock); |
1111 | } | |
1112 | } | |
1113 | return res; | |
b58fed8b | 1114 | Enomem: |
1da177e4 | 1115 | if (res) { |
70fbcdf4 | 1116 | LIST_HEAD(umount_list); |
1da177e4 | 1117 | spin_lock(&vfsmount_lock); |
a05964f3 | 1118 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1119 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1120 | release_mounts(&umount_list); |
1da177e4 LT |
1121 | } |
1122 | return NULL; | |
1123 | } | |
1124 | ||
8aec0809 AV |
1125 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
1126 | { | |
1127 | struct vfsmount *tree; | |
1a60a280 | 1128 | down_write(&namespace_sem); |
8aec0809 | 1129 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1130 | up_write(&namespace_sem); |
8aec0809 AV |
1131 | return tree; |
1132 | } | |
1133 | ||
1134 | void drop_collected_mounts(struct vfsmount *mnt) | |
1135 | { | |
1136 | LIST_HEAD(umount_list); | |
1a60a280 | 1137 | down_write(&namespace_sem); |
8aec0809 AV |
1138 | spin_lock(&vfsmount_lock); |
1139 | umount_tree(mnt, 0, &umount_list); | |
1140 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1141 | up_write(&namespace_sem); |
8aec0809 AV |
1142 | release_mounts(&umount_list); |
1143 | } | |
1144 | ||
b90fa9ae RP |
1145 | /* |
1146 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1147 | * @nd : place the mount tree @source_mnt is attached |
1148 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1149 | * store the parent mount and mountpoint dentry. | |
1150 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1151 | * |
1152 | * NOTE: in the table below explains the semantics when a source mount | |
1153 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1154 | * --------------------------------------------------------------------------- |
1155 | * | BIND MOUNT OPERATION | | |
1156 | * |************************************************************************** | |
1157 | * | source-->| shared | private | slave | unbindable | | |
1158 | * | dest | | | | | | |
1159 | * | | | | | | | | |
1160 | * | v | | | | | | |
1161 | * |************************************************************************** | |
1162 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1163 | * | | | | | | | |
1164 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1165 | * *************************************************************************** | |
b90fa9ae RP |
1166 | * A bind operation clones the source mount and mounts the clone on the |
1167 | * destination mount. | |
1168 | * | |
1169 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1170 | * tree of the destination mount and the cloned mount is added to | |
1171 | * the peer group of the source mount. | |
1172 | * (+) the cloned mount is created under the destination mount and is marked | |
1173 | * as shared. The cloned mount is added to the peer group of the source | |
1174 | * mount. | |
5afe0022 RP |
1175 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1176 | * of the destination mount and the cloned mount is made slave | |
1177 | * of the same master as that of the source mount. The cloned mount | |
1178 | * is marked as 'shared and slave'. | |
1179 | * (*) the cloned mount is made a slave of the same master as that of the | |
1180 | * source mount. | |
1181 | * | |
9676f0c6 RP |
1182 | * --------------------------------------------------------------------------- |
1183 | * | MOVE MOUNT OPERATION | | |
1184 | * |************************************************************************** | |
1185 | * | source-->| shared | private | slave | unbindable | | |
1186 | * | dest | | | | | | |
1187 | * | | | | | | | | |
1188 | * | v | | | | | | |
1189 | * |************************************************************************** | |
1190 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1191 | * | | | | | | | |
1192 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1193 | * *************************************************************************** | |
5afe0022 RP |
1194 | * |
1195 | * (+) the mount is moved to the destination. And is then propagated to | |
1196 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1197 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1198 | * (+++) the mount is moved to the destination and is then propagated to |
1199 | * all the mounts belonging to the destination mount's propagation tree. | |
1200 | * the mount is marked as 'shared and slave'. | |
1201 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1202 | * |
1203 | * if the source mount is a tree, the operations explained above is | |
1204 | * applied to each mount in the tree. | |
1205 | * Must be called without spinlocks held, since this function can sleep | |
1206 | * in allocations. | |
1207 | */ | |
1208 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1209 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1210 | { |
1211 | LIST_HEAD(tree_list); | |
1a390689 AV |
1212 | struct vfsmount *dest_mnt = path->mnt; |
1213 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae RP |
1214 | struct vfsmount *child, *p; |
1215 | ||
1216 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
1217 | return -EINVAL; | |
1218 | ||
1219 | if (IS_MNT_SHARED(dest_mnt)) { | |
1220 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1221 | set_mnt_shared(p); | |
1222 | } | |
1223 | ||
1224 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1225 | if (parent_path) { |
1226 | detach_mnt(source_mnt, parent_path); | |
1227 | attach_mnt(source_mnt, path); | |
6b3286ed | 1228 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
1229 | } else { |
1230 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1231 | commit_tree(source_mnt); | |
1232 | } | |
b90fa9ae RP |
1233 | |
1234 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1235 | list_del_init(&child->mnt_hash); | |
1236 | commit_tree(child); | |
1237 | } | |
1238 | spin_unlock(&vfsmount_lock); | |
1239 | return 0; | |
1240 | } | |
1241 | ||
8c3ee42e | 1242 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1243 | { |
1244 | int err; | |
1245 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1246 | return -EINVAL; | |
1247 | ||
8c3ee42e | 1248 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1249 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1250 | return -ENOTDIR; | |
1251 | ||
1252 | err = -ENOENT; | |
8c3ee42e AV |
1253 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1254 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1255 | goto out_unlock; |
1256 | ||
8c3ee42e | 1257 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1258 | if (err) |
1259 | goto out_unlock; | |
1260 | ||
1261 | err = -ENOENT; | |
8c3ee42e AV |
1262 | if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) |
1263 | err = attach_recursive_mnt(mnt, path, NULL); | |
1da177e4 | 1264 | out_unlock: |
8c3ee42e | 1265 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1266 | if (!err) |
8c3ee42e | 1267 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1268 | return err; |
1269 | } | |
1270 | ||
07b20889 RP |
1271 | /* |
1272 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 1273 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 1274 | */ |
2dafe1c4 | 1275 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 1276 | { |
4ac91378 | 1277 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
1278 | int recurse = flag & MS_REC; |
1279 | int type = flag & ~MS_REC; | |
1280 | ||
ee6f9582 MS |
1281 | if (!capable(CAP_SYS_ADMIN)) |
1282 | return -EPERM; | |
1283 | ||
4ac91378 | 1284 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
1285 | return -EINVAL; |
1286 | ||
1287 | down_write(&namespace_sem); | |
1288 | spin_lock(&vfsmount_lock); | |
1289 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1290 | change_mnt_propagation(m, type); | |
1291 | spin_unlock(&vfsmount_lock); | |
1292 | up_write(&namespace_sem); | |
1293 | return 0; | |
1294 | } | |
1295 | ||
1da177e4 LT |
1296 | /* |
1297 | * do loopback mount. | |
2dafe1c4 | 1298 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1299 | */ |
2dafe1c4 ES |
1300 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
1301 | int recurse) | |
1da177e4 LT |
1302 | { |
1303 | struct nameidata old_nd; | |
1304 | struct vfsmount *mnt = NULL; | |
1305 | int err = mount_is_safe(nd); | |
1306 | if (err) | |
1307 | return err; | |
1308 | if (!old_name || !*old_name) | |
1309 | return -EINVAL; | |
1310 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1311 | if (err) | |
1312 | return err; | |
1313 | ||
390c6843 | 1314 | down_write(&namespace_sem); |
1da177e4 | 1315 | err = -EINVAL; |
4ac91378 JB |
1316 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
1317 | goto out; | |
9676f0c6 | 1318 | |
4ac91378 | 1319 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 1320 | goto out; |
1da177e4 | 1321 | |
ccd48bc7 AV |
1322 | err = -ENOMEM; |
1323 | if (recurse) | |
4ac91378 | 1324 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 1325 | else |
4ac91378 | 1326 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1327 | |
1328 | if (!mnt) | |
1329 | goto out; | |
1330 | ||
8c3ee42e | 1331 | err = graft_tree(mnt, &nd->path); |
ccd48bc7 | 1332 | if (err) { |
70fbcdf4 | 1333 | LIST_HEAD(umount_list); |
1da177e4 | 1334 | spin_lock(&vfsmount_lock); |
a05964f3 | 1335 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1336 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1337 | release_mounts(&umount_list); |
5b83d2c5 | 1338 | } |
1da177e4 | 1339 | |
ccd48bc7 | 1340 | out: |
390c6843 | 1341 | up_write(&namespace_sem); |
1d957f9b | 1342 | path_put(&old_nd.path); |
1da177e4 LT |
1343 | return err; |
1344 | } | |
1345 | ||
2e4b7fcd DH |
1346 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1347 | { | |
1348 | int error = 0; | |
1349 | int readonly_request = 0; | |
1350 | ||
1351 | if (ms_flags & MS_RDONLY) | |
1352 | readonly_request = 1; | |
1353 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1354 | return 0; | |
1355 | ||
1356 | if (readonly_request) | |
1357 | error = mnt_make_readonly(mnt); | |
1358 | else | |
1359 | __mnt_unmake_readonly(mnt); | |
1360 | return error; | |
1361 | } | |
1362 | ||
1da177e4 LT |
1363 | /* |
1364 | * change filesystem flags. dir should be a physical root of filesystem. | |
1365 | * If you've mounted a non-root directory somewhere and want to do remount | |
1366 | * on it - tough luck. | |
2dafe1c4 | 1367 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1368 | */ |
2dafe1c4 | 1369 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1370 | void *data) |
1371 | { | |
1372 | int err; | |
4ac91378 | 1373 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1374 | |
1375 | if (!capable(CAP_SYS_ADMIN)) | |
1376 | return -EPERM; | |
1377 | ||
4ac91378 | 1378 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1379 | return -EINVAL; |
1380 | ||
4ac91378 | 1381 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1382 | return -EINVAL; |
1383 | ||
1384 | down_write(&sb->s_umount); | |
2e4b7fcd DH |
1385 | if (flags & MS_BIND) |
1386 | err = change_mount_flags(nd->path.mnt, flags); | |
1387 | else | |
1388 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1389 | if (!err) |
4ac91378 | 1390 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1391 | up_write(&sb->s_umount); |
1392 | if (!err) | |
4ac91378 | 1393 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1394 | return err; |
1395 | } | |
1396 | ||
9676f0c6 RP |
1397 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1398 | { | |
1399 | struct vfsmount *p; | |
1400 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1401 | if (IS_MNT_UNBINDABLE(p)) | |
1402 | return 1; | |
1403 | } | |
1404 | return 0; | |
1405 | } | |
1406 | ||
2dafe1c4 ES |
1407 | /* |
1408 | * noinline this do_mount helper to save do_mount stack space. | |
1409 | */ | |
1410 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1411 | { |
1a390689 AV |
1412 | struct nameidata old_nd; |
1413 | struct path parent_path; | |
1da177e4 LT |
1414 | struct vfsmount *p; |
1415 | int err = 0; | |
1416 | if (!capable(CAP_SYS_ADMIN)) | |
1417 | return -EPERM; | |
1418 | if (!old_name || !*old_name) | |
1419 | return -EINVAL; | |
1420 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1421 | if (err) | |
1422 | return err; | |
1423 | ||
390c6843 | 1424 | down_write(&namespace_sem); |
4ac91378 JB |
1425 | while (d_mountpoint(nd->path.dentry) && |
1426 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1427 | ; |
1428 | err = -EINVAL; | |
4ac91378 | 1429 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1430 | goto out; |
1431 | ||
1432 | err = -ENOENT; | |
4ac91378 JB |
1433 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1434 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1435 | goto out1; |
1436 | ||
4ac91378 | 1437 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1438 | goto out1; |
1da177e4 LT |
1439 | |
1440 | err = -EINVAL; | |
4ac91378 | 1441 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1442 | goto out1; |
1da177e4 | 1443 | |
4ac91378 | 1444 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1445 | goto out1; |
1da177e4 | 1446 | |
4ac91378 JB |
1447 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1448 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1449 | goto out1; |
1450 | /* | |
1451 | * Don't move a mount residing in a shared parent. | |
1452 | */ | |
4ac91378 JB |
1453 | if (old_nd.path.mnt->mnt_parent && |
1454 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1455 | goto out1; |
9676f0c6 RP |
1456 | /* |
1457 | * Don't move a mount tree containing unbindable mounts to a destination | |
1458 | * mount which is shared. | |
1459 | */ | |
4ac91378 JB |
1460 | if (IS_MNT_SHARED(nd->path.mnt) && |
1461 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1462 | goto out1; |
1da177e4 | 1463 | err = -ELOOP; |
4ac91378 JB |
1464 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1465 | if (p == old_nd.path.mnt) | |
21444403 | 1466 | goto out1; |
1da177e4 | 1467 | |
1a390689 | 1468 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1469 | if (err) |
21444403 | 1470 | goto out1; |
1da177e4 LT |
1471 | |
1472 | /* if the mount is moved, it should no longer be expire | |
1473 | * automatically */ | |
4ac91378 | 1474 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 | 1475 | out1: |
4ac91378 | 1476 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1477 | out: |
390c6843 | 1478 | up_write(&namespace_sem); |
1da177e4 | 1479 | if (!err) |
1a390689 | 1480 | path_put(&parent_path); |
1d957f9b | 1481 | path_put(&old_nd.path); |
1da177e4 LT |
1482 | return err; |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * create a new mount for userspace and request it to be added into the | |
1487 | * namespace's tree | |
2dafe1c4 | 1488 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1489 | */ |
2dafe1c4 | 1490 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1491 | int mnt_flags, char *name, void *data) |
1492 | { | |
1493 | struct vfsmount *mnt; | |
1494 | ||
1495 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1496 | return -EINVAL; | |
1497 | ||
1498 | /* we need capabilities... */ | |
1499 | if (!capable(CAP_SYS_ADMIN)) | |
1500 | return -EPERM; | |
1501 | ||
1502 | mnt = do_kern_mount(type, flags, name, data); | |
1503 | if (IS_ERR(mnt)) | |
1504 | return PTR_ERR(mnt); | |
1505 | ||
1506 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * add a mount into a namespace's mount tree | |
1511 | * - provide the option of adding the new mount to an expiration list | |
1512 | */ | |
1513 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1514 | int mnt_flags, struct list_head *fslist) | |
1515 | { | |
1516 | int err; | |
1517 | ||
390c6843 | 1518 | down_write(&namespace_sem); |
1da177e4 | 1519 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1520 | while (d_mountpoint(nd->path.dentry) && |
1521 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1522 | ; |
1523 | err = -EINVAL; | |
4ac91378 | 1524 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1525 | goto unlock; |
1526 | ||
1527 | /* Refuse the same filesystem on the same mount point */ | |
1528 | err = -EBUSY; | |
4ac91378 JB |
1529 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1530 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1531 | goto unlock; |
1532 | ||
1533 | err = -EINVAL; | |
1534 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1535 | goto unlock; | |
1536 | ||
1537 | newmnt->mnt_flags = mnt_flags; | |
8c3ee42e | 1538 | if ((err = graft_tree(newmnt, &nd->path))) |
5b83d2c5 | 1539 | goto unlock; |
1da177e4 | 1540 | |
6758f953 | 1541 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1542 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1543 | |
390c6843 | 1544 | up_write(&namespace_sem); |
5b83d2c5 | 1545 | return 0; |
1da177e4 LT |
1546 | |
1547 | unlock: | |
390c6843 | 1548 | up_write(&namespace_sem); |
1da177e4 LT |
1549 | mntput(newmnt); |
1550 | return err; | |
1551 | } | |
1552 | ||
1553 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1554 | ||
1555 | /* | |
1556 | * process a list of expirable mountpoints with the intent of discarding any | |
1557 | * mountpoints that aren't in use and haven't been touched since last we came | |
1558 | * here | |
1559 | */ | |
1560 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1561 | { | |
1da177e4 LT |
1562 | struct vfsmount *mnt, *next; |
1563 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1564 | LIST_HEAD(umounts); |
1da177e4 LT |
1565 | |
1566 | if (list_empty(mounts)) | |
1567 | return; | |
1568 | ||
bcc5c7d2 | 1569 | down_write(&namespace_sem); |
1da177e4 LT |
1570 | spin_lock(&vfsmount_lock); |
1571 | ||
1572 | /* extract from the expiration list every vfsmount that matches the | |
1573 | * following criteria: | |
1574 | * - only referenced by its parent vfsmount | |
1575 | * - still marked for expiry (marked on the last call here; marks are | |
1576 | * cleared by mntput()) | |
1577 | */ | |
55e700b9 | 1578 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1579 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1580 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1581 | continue; |
55e700b9 | 1582 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1583 | } |
bcc5c7d2 AV |
1584 | while (!list_empty(&graveyard)) { |
1585 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1586 | touch_mnt_namespace(mnt->mnt_ns); | |
1587 | umount_tree(mnt, 1, &umounts); | |
1588 | } | |
5528f911 | 1589 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1590 | up_write(&namespace_sem); |
1591 | ||
1592 | release_mounts(&umounts); | |
5528f911 TM |
1593 | } |
1594 | ||
1595 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1596 | ||
1597 | /* | |
1598 | * Ripoff of 'select_parent()' | |
1599 | * | |
1600 | * search the list of submounts for a given mountpoint, and move any | |
1601 | * shrinkable submounts to the 'graveyard' list. | |
1602 | */ | |
1603 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1604 | { | |
1605 | struct vfsmount *this_parent = parent; | |
1606 | struct list_head *next; | |
1607 | int found = 0; | |
1608 | ||
1609 | repeat: | |
1610 | next = this_parent->mnt_mounts.next; | |
1611 | resume: | |
1612 | while (next != &this_parent->mnt_mounts) { | |
1613 | struct list_head *tmp = next; | |
1614 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1615 | ||
1616 | next = tmp->next; | |
1617 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1618 | continue; |
5528f911 TM |
1619 | /* |
1620 | * Descend a level if the d_mounts list is non-empty. | |
1621 | */ | |
1622 | if (!list_empty(&mnt->mnt_mounts)) { | |
1623 | this_parent = mnt; | |
1624 | goto repeat; | |
1625 | } | |
1da177e4 | 1626 | |
5528f911 | 1627 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1628 | list_move_tail(&mnt->mnt_expire, graveyard); |
1629 | found++; | |
1630 | } | |
1da177e4 | 1631 | } |
5528f911 TM |
1632 | /* |
1633 | * All done at this level ... ascend and resume the search | |
1634 | */ | |
1635 | if (this_parent != parent) { | |
1636 | next = this_parent->mnt_child.next; | |
1637 | this_parent = this_parent->mnt_parent; | |
1638 | goto resume; | |
1639 | } | |
1640 | return found; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * process a list of expirable mountpoints with the intent of discarding any | |
1645 | * submounts of a specific parent mountpoint | |
1646 | */ | |
c35038be | 1647 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1648 | { |
1649 | LIST_HEAD(graveyard); | |
c35038be | 1650 | struct vfsmount *m; |
5528f911 | 1651 | |
5528f911 | 1652 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1653 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1654 | while (!list_empty(&graveyard)) { |
c35038be | 1655 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 AV |
1656 | mnt_expire); |
1657 | touch_mnt_namespace(mnt->mnt_ns); | |
c35038be | 1658 | umount_tree(mnt, 1, umounts); |
bcc5c7d2 AV |
1659 | } |
1660 | } | |
1da177e4 LT |
1661 | } |
1662 | ||
1da177e4 LT |
1663 | /* |
1664 | * Some copy_from_user() implementations do not return the exact number of | |
1665 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1666 | * Note that this function differs from copy_from_user() in that it will oops | |
1667 | * on bad values of `to', rather than returning a short copy. | |
1668 | */ | |
b58fed8b RP |
1669 | static long exact_copy_from_user(void *to, const void __user * from, |
1670 | unsigned long n) | |
1da177e4 LT |
1671 | { |
1672 | char *t = to; | |
1673 | const char __user *f = from; | |
1674 | char c; | |
1675 | ||
1676 | if (!access_ok(VERIFY_READ, from, n)) | |
1677 | return n; | |
1678 | ||
1679 | while (n) { | |
1680 | if (__get_user(c, f)) { | |
1681 | memset(t, 0, n); | |
1682 | break; | |
1683 | } | |
1684 | *t++ = c; | |
1685 | f++; | |
1686 | n--; | |
1687 | } | |
1688 | return n; | |
1689 | } | |
1690 | ||
b58fed8b | 1691 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1692 | { |
1693 | int i; | |
1694 | unsigned long page; | |
1695 | unsigned long size; | |
b58fed8b | 1696 | |
1da177e4 LT |
1697 | *where = 0; |
1698 | if (!data) | |
1699 | return 0; | |
1700 | ||
1701 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1702 | return -ENOMEM; | |
1703 | ||
1704 | /* We only care that *some* data at the address the user | |
1705 | * gave us is valid. Just in case, we'll zero | |
1706 | * the remainder of the page. | |
1707 | */ | |
1708 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1709 | size = TASK_SIZE - (unsigned long)data; | |
1710 | if (size > PAGE_SIZE) | |
1711 | size = PAGE_SIZE; | |
1712 | ||
1713 | i = size - exact_copy_from_user((void *)page, data, size); | |
1714 | if (!i) { | |
b58fed8b | 1715 | free_page(page); |
1da177e4 LT |
1716 | return -EFAULT; |
1717 | } | |
1718 | if (i != PAGE_SIZE) | |
1719 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1720 | *where = page; | |
1721 | return 0; | |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1726 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1727 | * | |
1728 | * data is a (void *) that can point to any structure up to | |
1729 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1730 | * information (or be NULL). | |
1731 | * | |
1732 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1733 | * When the flags word was introduced its top half was required | |
1734 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1735 | * Therefore, if this magic number is present, it carries no information | |
1736 | * and must be discarded. | |
1737 | */ | |
b58fed8b | 1738 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1739 | unsigned long flags, void *data_page) |
1740 | { | |
1741 | struct nameidata nd; | |
1742 | int retval = 0; | |
1743 | int mnt_flags = 0; | |
1744 | ||
1745 | /* Discard magic */ | |
1746 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1747 | flags &= ~MS_MGC_MSK; | |
1748 | ||
1749 | /* Basic sanity checks */ | |
1750 | ||
1751 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1752 | return -EINVAL; | |
1753 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1754 | return -EINVAL; | |
1755 | ||
1756 | if (data_page) | |
1757 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1758 | ||
1759 | /* Separate the per-mountpoint flags */ | |
1760 | if (flags & MS_NOSUID) | |
1761 | mnt_flags |= MNT_NOSUID; | |
1762 | if (flags & MS_NODEV) | |
1763 | mnt_flags |= MNT_NODEV; | |
1764 | if (flags & MS_NOEXEC) | |
1765 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1766 | if (flags & MS_NOATIME) |
1767 | mnt_flags |= MNT_NOATIME; | |
1768 | if (flags & MS_NODIRATIME) | |
1769 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1770 | if (flags & MS_RELATIME) |
1771 | mnt_flags |= MNT_RELATIME; | |
2e4b7fcd DH |
1772 | if (flags & MS_RDONLY) |
1773 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1774 | |
1775 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1776 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1777 | |
1778 | /* ... and get the mountpoint */ | |
1779 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1780 | if (retval) | |
1781 | return retval; | |
1782 | ||
b5266eb4 AV |
1783 | retval = security_sb_mount(dev_name, &nd.path, |
1784 | type_page, flags, data_page); | |
1da177e4 LT |
1785 | if (retval) |
1786 | goto dput_out; | |
1787 | ||
1788 | if (flags & MS_REMOUNT) | |
1789 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1790 | data_page); | |
1791 | else if (flags & MS_BIND) | |
eee391a6 | 1792 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1793 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1794 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1795 | else if (flags & MS_MOVE) |
1796 | retval = do_move_mount(&nd, dev_name); | |
1797 | else | |
1798 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1799 | dev_name, data_page); | |
1800 | dput_out: | |
1d957f9b | 1801 | path_put(&nd.path); |
1da177e4 LT |
1802 | return retval; |
1803 | } | |
1804 | ||
741a2951 JD |
1805 | /* |
1806 | * Allocate a new namespace structure and populate it with contents | |
1807 | * copied from the namespace of the passed in task structure. | |
1808 | */ | |
e3222c4e | 1809 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1810 | struct fs_struct *fs) |
1da177e4 | 1811 | { |
6b3286ed | 1812 | struct mnt_namespace *new_ns; |
1da177e4 | 1813 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1814 | struct vfsmount *p, *q; |
1815 | ||
6b3286ed | 1816 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1817 | if (!new_ns) |
467e9f4b | 1818 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1819 | |
1820 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1821 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1822 | init_waitqueue_head(&new_ns->poll); |
1823 | new_ns->event = 0; | |
1da177e4 | 1824 | |
390c6843 | 1825 | down_write(&namespace_sem); |
1da177e4 | 1826 | /* First pass: copy the tree topology */ |
6b3286ed | 1827 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1828 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1829 | if (!new_ns->root) { |
390c6843 | 1830 | up_write(&namespace_sem); |
1da177e4 | 1831 | kfree(new_ns); |
467e9f4b | 1832 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1833 | } |
1834 | spin_lock(&vfsmount_lock); | |
1835 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1836 | spin_unlock(&vfsmount_lock); | |
1837 | ||
1838 | /* | |
1839 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1840 | * as belonging to new namespace. We have already acquired a private | |
1841 | * fs_struct, so tsk->fs->lock is not needed. | |
1842 | */ | |
6b3286ed | 1843 | p = mnt_ns->root; |
1da177e4 LT |
1844 | q = new_ns->root; |
1845 | while (p) { | |
6b3286ed | 1846 | q->mnt_ns = new_ns; |
1da177e4 | 1847 | if (fs) { |
6ac08c39 | 1848 | if (p == fs->root.mnt) { |
1da177e4 | 1849 | rootmnt = p; |
6ac08c39 | 1850 | fs->root.mnt = mntget(q); |
1da177e4 | 1851 | } |
6ac08c39 | 1852 | if (p == fs->pwd.mnt) { |
1da177e4 | 1853 | pwdmnt = p; |
6ac08c39 | 1854 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1855 | } |
6ac08c39 | 1856 | if (p == fs->altroot.mnt) { |
1da177e4 | 1857 | altrootmnt = p; |
6ac08c39 | 1858 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1859 | } |
1860 | } | |
6b3286ed | 1861 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1862 | q = next_mnt(q, new_ns->root); |
1863 | } | |
390c6843 | 1864 | up_write(&namespace_sem); |
1da177e4 | 1865 | |
1da177e4 LT |
1866 | if (rootmnt) |
1867 | mntput(rootmnt); | |
1868 | if (pwdmnt) | |
1869 | mntput(pwdmnt); | |
1870 | if (altrootmnt) | |
1871 | mntput(altrootmnt); | |
1872 | ||
741a2951 JD |
1873 | return new_ns; |
1874 | } | |
1875 | ||
213dd266 | 1876 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1877 | struct fs_struct *new_fs) |
741a2951 | 1878 | { |
6b3286ed | 1879 | struct mnt_namespace *new_ns; |
741a2951 | 1880 | |
e3222c4e | 1881 | BUG_ON(!ns); |
6b3286ed | 1882 | get_mnt_ns(ns); |
741a2951 JD |
1883 | |
1884 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1885 | return ns; |
741a2951 | 1886 | |
e3222c4e | 1887 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1888 | |
6b3286ed | 1889 | put_mnt_ns(ns); |
e3222c4e | 1890 | return new_ns; |
1da177e4 LT |
1891 | } |
1892 | ||
1893 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1894 | char __user * type, unsigned long flags, | |
1895 | void __user * data) | |
1896 | { | |
1897 | int retval; | |
1898 | unsigned long data_page; | |
1899 | unsigned long type_page; | |
1900 | unsigned long dev_page; | |
1901 | char *dir_page; | |
1902 | ||
b58fed8b | 1903 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1904 | if (retval < 0) |
1905 | return retval; | |
1906 | ||
1907 | dir_page = getname(dir_name); | |
1908 | retval = PTR_ERR(dir_page); | |
1909 | if (IS_ERR(dir_page)) | |
1910 | goto out1; | |
1911 | ||
b58fed8b | 1912 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1913 | if (retval < 0) |
1914 | goto out2; | |
1915 | ||
b58fed8b | 1916 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1917 | if (retval < 0) |
1918 | goto out3; | |
1919 | ||
1920 | lock_kernel(); | |
b58fed8b RP |
1921 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1922 | flags, (void *)data_page); | |
1da177e4 LT |
1923 | unlock_kernel(); |
1924 | free_page(data_page); | |
1925 | ||
1926 | out3: | |
1927 | free_page(dev_page); | |
1928 | out2: | |
1929 | putname(dir_page); | |
1930 | out1: | |
1931 | free_page(type_page); | |
1932 | return retval; | |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1937 | * It can block. Requires the big lock held. | |
1938 | */ | |
ac748a09 | 1939 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 1940 | { |
6ac08c39 JB |
1941 | struct path old_root; |
1942 | ||
1da177e4 LT |
1943 | write_lock(&fs->lock); |
1944 | old_root = fs->root; | |
ac748a09 JB |
1945 | fs->root = *path; |
1946 | path_get(path); | |
1da177e4 | 1947 | write_unlock(&fs->lock); |
6ac08c39 JB |
1948 | if (old_root.dentry) |
1949 | path_put(&old_root); | |
1da177e4 LT |
1950 | } |
1951 | ||
1952 | /* | |
1953 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1954 | * It can block. Requires the big lock held. | |
1955 | */ | |
ac748a09 | 1956 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 1957 | { |
6ac08c39 | 1958 | struct path old_pwd; |
1da177e4 LT |
1959 | |
1960 | write_lock(&fs->lock); | |
1961 | old_pwd = fs->pwd; | |
ac748a09 JB |
1962 | fs->pwd = *path; |
1963 | path_get(path); | |
1da177e4 LT |
1964 | write_unlock(&fs->lock); |
1965 | ||
6ac08c39 JB |
1966 | if (old_pwd.dentry) |
1967 | path_put(&old_pwd); | |
1da177e4 LT |
1968 | } |
1969 | ||
1a390689 | 1970 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
1971 | { |
1972 | struct task_struct *g, *p; | |
1973 | struct fs_struct *fs; | |
1974 | ||
1975 | read_lock(&tasklist_lock); | |
1976 | do_each_thread(g, p) { | |
1977 | task_lock(p); | |
1978 | fs = p->fs; | |
1979 | if (fs) { | |
1980 | atomic_inc(&fs->count); | |
1981 | task_unlock(p); | |
1a390689 AV |
1982 | if (fs->root.dentry == old_root->dentry |
1983 | && fs->root.mnt == old_root->mnt) | |
1984 | set_fs_root(fs, new_root); | |
1985 | if (fs->pwd.dentry == old_root->dentry | |
1986 | && fs->pwd.mnt == old_root->mnt) | |
1987 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
1988 | put_fs_struct(fs); |
1989 | } else | |
1990 | task_unlock(p); | |
1991 | } while_each_thread(g, p); | |
1992 | read_unlock(&tasklist_lock); | |
1993 | } | |
1994 | ||
1995 | /* | |
1996 | * pivot_root Semantics: | |
1997 | * Moves the root file system of the current process to the directory put_old, | |
1998 | * makes new_root as the new root file system of the current process, and sets | |
1999 | * root/cwd of all processes which had them on the current root to new_root. | |
2000 | * | |
2001 | * Restrictions: | |
2002 | * The new_root and put_old must be directories, and must not be on the | |
2003 | * same file system as the current process root. The put_old must be | |
2004 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2005 | * pointed to by put_old must yield the same directory as new_root. No other | |
2006 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2007 | * | |
4a0d11fa NB |
2008 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2009 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2010 | * in this situation. | |
2011 | * | |
1da177e4 LT |
2012 | * Notes: |
2013 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2014 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2015 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2016 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2017 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2018 | * first. | |
2019 | */ | |
b58fed8b RP |
2020 | asmlinkage long sys_pivot_root(const char __user * new_root, |
2021 | const char __user * put_old) | |
1da177e4 LT |
2022 | { |
2023 | struct vfsmount *tmp; | |
8c3ee42e AV |
2024 | struct nameidata new_nd, old_nd; |
2025 | struct path parent_path, root_parent, root; | |
1da177e4 LT |
2026 | int error; |
2027 | ||
2028 | if (!capable(CAP_SYS_ADMIN)) | |
2029 | return -EPERM; | |
2030 | ||
b58fed8b RP |
2031 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
2032 | &new_nd); | |
1da177e4 LT |
2033 | if (error) |
2034 | goto out0; | |
2035 | error = -EINVAL; | |
4ac91378 | 2036 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
2037 | goto out1; |
2038 | ||
b58fed8b | 2039 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
2040 | if (error) |
2041 | goto out1; | |
2042 | ||
b5266eb4 | 2043 | error = security_sb_pivotroot(&old_nd.path, &new_nd.path); |
1da177e4 | 2044 | if (error) { |
1d957f9b | 2045 | path_put(&old_nd.path); |
1da177e4 LT |
2046 | goto out1; |
2047 | } | |
2048 | ||
2049 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2050 | root = current->fs->root; |
6ac08c39 | 2051 | path_get(¤t->fs->root); |
1da177e4 | 2052 | read_unlock(¤t->fs->lock); |
390c6843 | 2053 | down_write(&namespace_sem); |
4ac91378 | 2054 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2055 | error = -EINVAL; |
4ac91378 JB |
2056 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
2057 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
8c3ee42e | 2058 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2059 | goto out2; |
8c3ee42e | 2060 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2061 | goto out2; |
2062 | error = -ENOENT; | |
4ac91378 | 2063 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 2064 | goto out2; |
4ac91378 | 2065 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 2066 | goto out2; |
4ac91378 | 2067 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
2068 | goto out2; |
2069 | error = -EBUSY; | |
8c3ee42e AV |
2070 | if (new_nd.path.mnt == root.mnt || |
2071 | old_nd.path.mnt == root.mnt) | |
1da177e4 LT |
2072 | goto out2; /* loop, on the same file system */ |
2073 | error = -EINVAL; | |
8c3ee42e | 2074 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2075 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2076 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2077 | goto out2; /* not attached */ |
4ac91378 | 2078 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 2079 | goto out2; /* not a mountpoint */ |
4ac91378 | 2080 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 2081 | goto out2; /* not attached */ |
4ac91378 JB |
2082 | /* make sure we can reach put_old from new_root */ |
2083 | tmp = old_nd.path.mnt; | |
1da177e4 | 2084 | spin_lock(&vfsmount_lock); |
4ac91378 | 2085 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
2086 | for (;;) { |
2087 | if (tmp->mnt_parent == tmp) | |
2088 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 2089 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
2090 | break; |
2091 | tmp = tmp->mnt_parent; | |
2092 | } | |
4ac91378 | 2093 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 2094 | goto out3; |
4ac91378 | 2095 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 2096 | goto out3; |
1a390689 | 2097 | detach_mnt(new_nd.path.mnt, &parent_path); |
8c3ee42e | 2098 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2099 | /* mount old root on put_old */ |
8c3ee42e | 2100 | attach_mnt(root.mnt, &old_nd.path); |
4ac91378 JB |
2101 | /* mount new_root on / */ |
2102 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 2103 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2104 | spin_unlock(&vfsmount_lock); |
8c3ee42e AV |
2105 | chroot_fs_refs(&root, &new_nd.path); |
2106 | security_sb_post_pivotroot(&root, &new_nd.path); | |
1da177e4 | 2107 | error = 0; |
1a390689 AV |
2108 | path_put(&root_parent); |
2109 | path_put(&parent_path); | |
1da177e4 | 2110 | out2: |
4ac91378 | 2111 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 2112 | up_write(&namespace_sem); |
8c3ee42e | 2113 | path_put(&root); |
1d957f9b | 2114 | path_put(&old_nd.path); |
1da177e4 | 2115 | out1: |
1d957f9b | 2116 | path_put(&new_nd.path); |
1da177e4 | 2117 | out0: |
1da177e4 LT |
2118 | return error; |
2119 | out3: | |
2120 | spin_unlock(&vfsmount_lock); | |
2121 | goto out2; | |
2122 | } | |
2123 | ||
2124 | static void __init init_mount_tree(void) | |
2125 | { | |
2126 | struct vfsmount *mnt; | |
6b3286ed | 2127 | struct mnt_namespace *ns; |
ac748a09 | 2128 | struct path root; |
1da177e4 LT |
2129 | |
2130 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2131 | if (IS_ERR(mnt)) | |
2132 | panic("Can't create rootfs"); | |
6b3286ed KK |
2133 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2134 | if (!ns) | |
1da177e4 | 2135 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2136 | atomic_set(&ns->count, 1); |
2137 | INIT_LIST_HEAD(&ns->list); | |
2138 | init_waitqueue_head(&ns->poll); | |
2139 | ns->event = 0; | |
2140 | list_add(&mnt->mnt_list, &ns->list); | |
2141 | ns->root = mnt; | |
2142 | mnt->mnt_ns = ns; | |
2143 | ||
2144 | init_task.nsproxy->mnt_ns = ns; | |
2145 | get_mnt_ns(ns); | |
2146 | ||
ac748a09 JB |
2147 | root.mnt = ns->root; |
2148 | root.dentry = ns->root->mnt_root; | |
2149 | ||
2150 | set_fs_pwd(current->fs, &root); | |
2151 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2152 | } |
2153 | ||
74bf17cf | 2154 | void __init mnt_init(void) |
1da177e4 | 2155 | { |
13f14b4d | 2156 | unsigned u; |
15a67dd8 | 2157 | int err; |
1da177e4 | 2158 | |
390c6843 RP |
2159 | init_rwsem(&namespace_sem); |
2160 | ||
1da177e4 | 2161 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2162 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2163 | |
b58fed8b | 2164 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2165 | |
2166 | if (!mount_hashtable) | |
2167 | panic("Failed to allocate mount hash table\n"); | |
2168 | ||
13f14b4d ED |
2169 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2170 | ||
2171 | for (u = 0; u < HASH_SIZE; u++) | |
2172 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2173 | |
15a67dd8 RD |
2174 | err = sysfs_init(); |
2175 | if (err) | |
2176 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
2177 | __FUNCTION__, err); | |
00d26666 GKH |
2178 | fs_kobj = kobject_create_and_add("fs", NULL); |
2179 | if (!fs_kobj) | |
2180 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
2181 | init_rootfs(); |
2182 | init_mount_tree(); | |
2183 | } | |
2184 | ||
6b3286ed | 2185 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2186 | { |
6b3286ed | 2187 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2188 | LIST_HEAD(umount_list); |
6b3286ed | 2189 | ns->root = NULL; |
1ce88cf4 | 2190 | spin_unlock(&vfsmount_lock); |
390c6843 | 2191 | down_write(&namespace_sem); |
1da177e4 | 2192 | spin_lock(&vfsmount_lock); |
a05964f3 | 2193 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2194 | spin_unlock(&vfsmount_lock); |
390c6843 | 2195 | up_write(&namespace_sem); |
70fbcdf4 | 2196 | release_mounts(&umount_list); |
6b3286ed | 2197 | kfree(ns); |
1da177e4 | 2198 | } |