]>
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> |
1da177e4 LT |
30 | #include <asm/uaccess.h> |
31 | #include <asm/unistd.h> | |
07b20889 | 32 | #include "pnode.h" |
948730b0 | 33 | #include "internal.h" |
1da177e4 | 34 | |
13f14b4d ED |
35 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
36 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
37 | ||
1da177e4 | 38 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
39 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
40 | ||
41 | static int event; | |
1da177e4 | 42 | |
fa3536cc | 43 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 44 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 45 | static struct rw_semaphore namespace_sem; |
1da177e4 | 46 | |
f87fd4c2 | 47 | /* /sys/fs */ |
00d26666 GKH |
48 | struct kobject *fs_kobj; |
49 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 50 | |
1da177e4 LT |
51 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
52 | { | |
b58fed8b RP |
53 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
54 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
55 | tmp = tmp + (tmp >> HASH_SHIFT); |
56 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
57 | } |
58 | ||
3d733633 DH |
59 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
60 | ||
1da177e4 LT |
61 | struct vfsmount *alloc_vfsmnt(const char *name) |
62 | { | |
c3762229 | 63 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 64 | if (mnt) { |
b58fed8b | 65 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
66 | INIT_LIST_HEAD(&mnt->mnt_hash); |
67 | INIT_LIST_HEAD(&mnt->mnt_child); | |
68 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
69 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 70 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 71 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
72 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
73 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
3d733633 | 74 | atomic_set(&mnt->__mnt_writers, 0); |
1da177e4 | 75 | if (name) { |
b58fed8b | 76 | int size = strlen(name) + 1; |
1da177e4 LT |
77 | char *newname = kmalloc(size, GFP_KERNEL); |
78 | if (newname) { | |
79 | memcpy(newname, name, size); | |
80 | mnt->mnt_devname = newname; | |
81 | } | |
82 | } | |
83 | } | |
84 | return mnt; | |
85 | } | |
86 | ||
3d733633 DH |
87 | /* |
88 | * Most r/o checks on a fs are for operations that take | |
89 | * discrete amounts of time, like a write() or unlink(). | |
90 | * We must keep track of when those operations start | |
91 | * (for permission checks) and when they end, so that | |
92 | * we can determine when writes are able to occur to | |
93 | * a filesystem. | |
94 | */ | |
95 | /* | |
96 | * __mnt_is_readonly: check whether a mount is read-only | |
97 | * @mnt: the mount to check for its write status | |
98 | * | |
99 | * This shouldn't be used directly ouside of the VFS. | |
100 | * It does not guarantee that the filesystem will stay | |
101 | * r/w, just that it is right *now*. This can not and | |
102 | * should not be used in place of IS_RDONLY(inode). | |
103 | * mnt_want/drop_write() will _keep_ the filesystem | |
104 | * r/w. | |
105 | */ | |
106 | int __mnt_is_readonly(struct vfsmount *mnt) | |
107 | { | |
2e4b7fcd DH |
108 | if (mnt->mnt_flags & MNT_READONLY) |
109 | return 1; | |
110 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
111 | return 1; | |
112 | return 0; | |
3d733633 DH |
113 | } |
114 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
115 | ||
116 | struct mnt_writer { | |
117 | /* | |
118 | * If holding multiple instances of this lock, they | |
119 | * must be ordered by cpu number. | |
120 | */ | |
121 | spinlock_t lock; | |
122 | struct lock_class_key lock_class; /* compiles out with !lockdep */ | |
123 | unsigned long count; | |
124 | struct vfsmount *mnt; | |
125 | } ____cacheline_aligned_in_smp; | |
126 | static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); | |
127 | ||
128 | static int __init init_mnt_writers(void) | |
129 | { | |
130 | int cpu; | |
131 | for_each_possible_cpu(cpu) { | |
132 | struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); | |
133 | spin_lock_init(&writer->lock); | |
134 | lockdep_set_class(&writer->lock, &writer->lock_class); | |
135 | writer->count = 0; | |
136 | } | |
137 | return 0; | |
138 | } | |
139 | fs_initcall(init_mnt_writers); | |
140 | ||
141 | static void unlock_mnt_writers(void) | |
142 | { | |
143 | int cpu; | |
144 | struct mnt_writer *cpu_writer; | |
145 | ||
146 | for_each_possible_cpu(cpu) { | |
147 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
148 | spin_unlock(&cpu_writer->lock); | |
149 | } | |
150 | } | |
151 | ||
152 | static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) | |
153 | { | |
154 | if (!cpu_writer->mnt) | |
155 | return; | |
156 | /* | |
157 | * This is in case anyone ever leaves an invalid, | |
158 | * old ->mnt and a count of 0. | |
159 | */ | |
160 | if (!cpu_writer->count) | |
161 | return; | |
162 | atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); | |
163 | cpu_writer->count = 0; | |
164 | } | |
165 | /* | |
166 | * must hold cpu_writer->lock | |
167 | */ | |
168 | static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, | |
169 | struct vfsmount *mnt) | |
170 | { | |
171 | if (cpu_writer->mnt == mnt) | |
172 | return; | |
173 | __clear_mnt_count(cpu_writer); | |
174 | cpu_writer->mnt = mnt; | |
175 | } | |
176 | ||
8366025e DH |
177 | /* |
178 | * Most r/o checks on a fs are for operations that take | |
179 | * discrete amounts of time, like a write() or unlink(). | |
180 | * We must keep track of when those operations start | |
181 | * (for permission checks) and when they end, so that | |
182 | * we can determine when writes are able to occur to | |
183 | * a filesystem. | |
184 | */ | |
185 | /** | |
186 | * mnt_want_write - get write access to a mount | |
187 | * @mnt: the mount on which to take a write | |
188 | * | |
189 | * This tells the low-level filesystem that a write is | |
190 | * about to be performed to it, and makes sure that | |
191 | * writes are allowed before returning success. When | |
192 | * the write operation is finished, mnt_drop_write() | |
193 | * must be called. This is effectively a refcount. | |
194 | */ | |
195 | int mnt_want_write(struct vfsmount *mnt) | |
196 | { | |
3d733633 DH |
197 | int ret = 0; |
198 | struct mnt_writer *cpu_writer; | |
199 | ||
200 | cpu_writer = &get_cpu_var(mnt_writers); | |
201 | spin_lock(&cpu_writer->lock); | |
202 | if (__mnt_is_readonly(mnt)) { | |
203 | ret = -EROFS; | |
204 | goto out; | |
205 | } | |
206 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
207 | cpu_writer->count++; | |
208 | out: | |
209 | spin_unlock(&cpu_writer->lock); | |
210 | put_cpu_var(mnt_writers); | |
211 | return ret; | |
8366025e DH |
212 | } |
213 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
214 | ||
3d733633 DH |
215 | static void lock_mnt_writers(void) |
216 | { | |
217 | int cpu; | |
218 | struct mnt_writer *cpu_writer; | |
219 | ||
220 | for_each_possible_cpu(cpu) { | |
221 | cpu_writer = &per_cpu(mnt_writers, cpu); | |
222 | spin_lock(&cpu_writer->lock); | |
223 | __clear_mnt_count(cpu_writer); | |
224 | cpu_writer->mnt = NULL; | |
225 | } | |
226 | } | |
227 | ||
228 | /* | |
229 | * These per-cpu write counts are not guaranteed to have | |
230 | * matched increments and decrements on any given cpu. | |
231 | * A file open()ed for write on one cpu and close()d on | |
232 | * another cpu will imbalance this count. Make sure it | |
233 | * does not get too far out of whack. | |
234 | */ | |
235 | static void handle_write_count_underflow(struct vfsmount *mnt) | |
236 | { | |
237 | if (atomic_read(&mnt->__mnt_writers) >= | |
238 | MNT_WRITER_UNDERFLOW_LIMIT) | |
239 | return; | |
240 | /* | |
241 | * It isn't necessary to hold all of the locks | |
242 | * at the same time, but doing it this way makes | |
243 | * us share a lot more code. | |
244 | */ | |
245 | lock_mnt_writers(); | |
246 | /* | |
247 | * vfsmount_lock is for mnt_flags. | |
248 | */ | |
249 | spin_lock(&vfsmount_lock); | |
250 | /* | |
251 | * If coalescing the per-cpu writer counts did not | |
252 | * get us back to a positive writer count, we have | |
253 | * a bug. | |
254 | */ | |
255 | if ((atomic_read(&mnt->__mnt_writers) < 0) && | |
256 | !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { | |
257 | printk(KERN_DEBUG "leak detected on mount(%p) writers " | |
258 | "count: %d\n", | |
259 | mnt, atomic_read(&mnt->__mnt_writers)); | |
260 | WARN_ON(1); | |
261 | /* use the flag to keep the dmesg spam down */ | |
262 | mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; | |
263 | } | |
264 | spin_unlock(&vfsmount_lock); | |
265 | unlock_mnt_writers(); | |
266 | } | |
267 | ||
8366025e DH |
268 | /** |
269 | * mnt_drop_write - give up write access to a mount | |
270 | * @mnt: the mount on which to give up write access | |
271 | * | |
272 | * Tells the low-level filesystem that we are done | |
273 | * performing writes to it. Must be matched with | |
274 | * mnt_want_write() call above. | |
275 | */ | |
276 | void mnt_drop_write(struct vfsmount *mnt) | |
277 | { | |
3d733633 DH |
278 | int must_check_underflow = 0; |
279 | struct mnt_writer *cpu_writer; | |
280 | ||
281 | cpu_writer = &get_cpu_var(mnt_writers); | |
282 | spin_lock(&cpu_writer->lock); | |
283 | ||
284 | use_cpu_writer_for_mount(cpu_writer, mnt); | |
285 | if (cpu_writer->count > 0) { | |
286 | cpu_writer->count--; | |
287 | } else { | |
288 | must_check_underflow = 1; | |
289 | atomic_dec(&mnt->__mnt_writers); | |
290 | } | |
291 | ||
292 | spin_unlock(&cpu_writer->lock); | |
293 | /* | |
294 | * Logically, we could call this each time, | |
295 | * but the __mnt_writers cacheline tends to | |
296 | * be cold, and makes this expensive. | |
297 | */ | |
298 | if (must_check_underflow) | |
299 | handle_write_count_underflow(mnt); | |
300 | /* | |
301 | * This could be done right after the spinlock | |
302 | * is taken because the spinlock keeps us on | |
303 | * the cpu, and disables preemption. However, | |
304 | * putting it here bounds the amount that | |
305 | * __mnt_writers can underflow. Without it, | |
306 | * we could theoretically wrap __mnt_writers. | |
307 | */ | |
308 | put_cpu_var(mnt_writers); | |
8366025e DH |
309 | } |
310 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
311 | ||
2e4b7fcd | 312 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 313 | { |
3d733633 DH |
314 | int ret = 0; |
315 | ||
316 | lock_mnt_writers(); | |
317 | /* | |
318 | * With all the locks held, this value is stable | |
319 | */ | |
320 | if (atomic_read(&mnt->__mnt_writers) > 0) { | |
321 | ret = -EBUSY; | |
322 | goto out; | |
323 | } | |
324 | /* | |
2e4b7fcd DH |
325 | * nobody can do a successful mnt_want_write() with all |
326 | * of the counts in MNT_DENIED_WRITE and the locks held. | |
3d733633 | 327 | */ |
2e4b7fcd DH |
328 | spin_lock(&vfsmount_lock); |
329 | if (!ret) | |
330 | mnt->mnt_flags |= MNT_READONLY; | |
331 | spin_unlock(&vfsmount_lock); | |
3d733633 DH |
332 | out: |
333 | unlock_mnt_writers(); | |
334 | return ret; | |
8366025e | 335 | } |
8366025e | 336 | |
2e4b7fcd DH |
337 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
338 | { | |
339 | spin_lock(&vfsmount_lock); | |
340 | mnt->mnt_flags &= ~MNT_READONLY; | |
341 | spin_unlock(&vfsmount_lock); | |
342 | } | |
343 | ||
454e2398 DH |
344 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
345 | { | |
346 | mnt->mnt_sb = sb; | |
347 | mnt->mnt_root = dget(sb->s_root); | |
348 | return 0; | |
349 | } | |
350 | ||
351 | EXPORT_SYMBOL(simple_set_mnt); | |
352 | ||
1da177e4 LT |
353 | void free_vfsmnt(struct vfsmount *mnt) |
354 | { | |
355 | kfree(mnt->mnt_devname); | |
356 | kmem_cache_free(mnt_cache, mnt); | |
357 | } | |
358 | ||
359 | /* | |
a05964f3 RP |
360 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
361 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 362 | */ |
a05964f3 RP |
363 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
364 | int dir) | |
1da177e4 | 365 | { |
b58fed8b RP |
366 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
367 | struct list_head *tmp = head; | |
1da177e4 LT |
368 | struct vfsmount *p, *found = NULL; |
369 | ||
1da177e4 | 370 | for (;;) { |
a05964f3 | 371 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
372 | p = NULL; |
373 | if (tmp == head) | |
374 | break; | |
375 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
376 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 377 | found = p; |
1da177e4 LT |
378 | break; |
379 | } | |
380 | } | |
1da177e4 LT |
381 | return found; |
382 | } | |
383 | ||
a05964f3 RP |
384 | /* |
385 | * lookup_mnt increments the ref count before returning | |
386 | * the vfsmount struct. | |
387 | */ | |
388 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
389 | { | |
390 | struct vfsmount *child_mnt; | |
391 | spin_lock(&vfsmount_lock); | |
392 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
393 | mntget(child_mnt); | |
394 | spin_unlock(&vfsmount_lock); | |
395 | return child_mnt; | |
396 | } | |
397 | ||
1da177e4 LT |
398 | static inline int check_mnt(struct vfsmount *mnt) |
399 | { | |
6b3286ed | 400 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
401 | } |
402 | ||
6b3286ed | 403 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
404 | { |
405 | if (ns) { | |
406 | ns->event = ++event; | |
407 | wake_up_interruptible(&ns->poll); | |
408 | } | |
409 | } | |
410 | ||
6b3286ed | 411 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
412 | { |
413 | if (ns && ns->event != event) { | |
414 | ns->event = event; | |
415 | wake_up_interruptible(&ns->poll); | |
416 | } | |
417 | } | |
418 | ||
1a390689 | 419 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 420 | { |
1a390689 AV |
421 | old_path->dentry = mnt->mnt_mountpoint; |
422 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
423 | mnt->mnt_parent = mnt; |
424 | mnt->mnt_mountpoint = mnt->mnt_root; | |
425 | list_del_init(&mnt->mnt_child); | |
426 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 427 | old_path->dentry->d_mounted--; |
1da177e4 LT |
428 | } |
429 | ||
b90fa9ae RP |
430 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
431 | struct vfsmount *child_mnt) | |
432 | { | |
433 | child_mnt->mnt_parent = mntget(mnt); | |
434 | child_mnt->mnt_mountpoint = dget(dentry); | |
435 | dentry->d_mounted++; | |
436 | } | |
437 | ||
1a390689 | 438 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 439 | { |
1a390689 | 440 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 441 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
442 | hash(path->mnt, path->dentry)); |
443 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
444 | } |
445 | ||
446 | /* | |
447 | * the caller must hold vfsmount_lock | |
448 | */ | |
449 | static void commit_tree(struct vfsmount *mnt) | |
450 | { | |
451 | struct vfsmount *parent = mnt->mnt_parent; | |
452 | struct vfsmount *m; | |
453 | LIST_HEAD(head); | |
6b3286ed | 454 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
455 | |
456 | BUG_ON(parent == mnt); | |
457 | ||
458 | list_add_tail(&head, &mnt->mnt_list); | |
459 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 460 | m->mnt_ns = n; |
b90fa9ae RP |
461 | list_splice(&head, n->list.prev); |
462 | ||
463 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
464 | hash(parent, mnt->mnt_mountpoint)); | |
465 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 466 | touch_mnt_namespace(n); |
1da177e4 LT |
467 | } |
468 | ||
469 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
470 | { | |
471 | struct list_head *next = p->mnt_mounts.next; | |
472 | if (next == &p->mnt_mounts) { | |
473 | while (1) { | |
474 | if (p == root) | |
475 | return NULL; | |
476 | next = p->mnt_child.next; | |
477 | if (next != &p->mnt_parent->mnt_mounts) | |
478 | break; | |
479 | p = p->mnt_parent; | |
480 | } | |
481 | } | |
482 | return list_entry(next, struct vfsmount, mnt_child); | |
483 | } | |
484 | ||
9676f0c6 RP |
485 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
486 | { | |
487 | struct list_head *prev = p->mnt_mounts.prev; | |
488 | while (prev != &p->mnt_mounts) { | |
489 | p = list_entry(prev, struct vfsmount, mnt_child); | |
490 | prev = p->mnt_mounts.prev; | |
491 | } | |
492 | return p; | |
493 | } | |
494 | ||
36341f64 RP |
495 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
496 | int flag) | |
1da177e4 LT |
497 | { |
498 | struct super_block *sb = old->mnt_sb; | |
499 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
500 | ||
501 | if (mnt) { | |
502 | mnt->mnt_flags = old->mnt_flags; | |
503 | atomic_inc(&sb->s_active); | |
504 | mnt->mnt_sb = sb; | |
505 | mnt->mnt_root = dget(root); | |
506 | mnt->mnt_mountpoint = mnt->mnt_root; | |
507 | mnt->mnt_parent = mnt; | |
b90fa9ae | 508 | |
5afe0022 RP |
509 | if (flag & CL_SLAVE) { |
510 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
511 | mnt->mnt_master = old; | |
512 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 513 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
514 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
515 | list_add(&mnt->mnt_share, &old->mnt_share); | |
516 | if (IS_MNT_SLAVE(old)) | |
517 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
518 | mnt->mnt_master = old->mnt_master; | |
519 | } | |
b90fa9ae RP |
520 | if (flag & CL_MAKE_SHARED) |
521 | set_mnt_shared(mnt); | |
1da177e4 LT |
522 | |
523 | /* stick the duplicate mount on the same expiry list | |
524 | * as the original if that was on one */ | |
36341f64 | 525 | if (flag & CL_EXPIRE) { |
36341f64 RP |
526 | if (!list_empty(&old->mnt_expire)) |
527 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 528 | } |
1da177e4 LT |
529 | } |
530 | return mnt; | |
531 | } | |
532 | ||
7b7b1ace | 533 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 | 534 | { |
3d733633 | 535 | int cpu; |
1da177e4 | 536 | struct super_block *sb = mnt->mnt_sb; |
3d733633 DH |
537 | /* |
538 | * We don't have to hold all of the locks at the | |
539 | * same time here because we know that we're the | |
540 | * last reference to mnt and that no new writers | |
541 | * can come in. | |
542 | */ | |
543 | for_each_possible_cpu(cpu) { | |
544 | struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); | |
545 | if (cpu_writer->mnt != mnt) | |
546 | continue; | |
547 | spin_lock(&cpu_writer->lock); | |
548 | atomic_add(cpu_writer->count, &mnt->__mnt_writers); | |
549 | cpu_writer->count = 0; | |
550 | /* | |
551 | * Might as well do this so that no one | |
552 | * ever sees the pointer and expects | |
553 | * it to be valid. | |
554 | */ | |
555 | cpu_writer->mnt = NULL; | |
556 | spin_unlock(&cpu_writer->lock); | |
557 | } | |
558 | /* | |
559 | * This probably indicates that somebody messed | |
560 | * up a mnt_want/drop_write() pair. If this | |
561 | * happens, the filesystem was probably unable | |
562 | * to make r/w->r/o transitions. | |
563 | */ | |
564 | WARN_ON(atomic_read(&mnt->__mnt_writers)); | |
1da177e4 LT |
565 | dput(mnt->mnt_root); |
566 | free_vfsmnt(mnt); | |
567 | deactivate_super(sb); | |
568 | } | |
569 | ||
7b7b1ace AV |
570 | void mntput_no_expire(struct vfsmount *mnt) |
571 | { | |
572 | repeat: | |
573 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
574 | if (likely(!mnt->mnt_pinned)) { | |
575 | spin_unlock(&vfsmount_lock); | |
576 | __mntput(mnt); | |
577 | return; | |
578 | } | |
579 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
580 | mnt->mnt_pinned = 0; | |
581 | spin_unlock(&vfsmount_lock); | |
582 | acct_auto_close_mnt(mnt); | |
583 | security_sb_umount_close(mnt); | |
584 | goto repeat; | |
585 | } | |
586 | } | |
587 | ||
588 | EXPORT_SYMBOL(mntput_no_expire); | |
589 | ||
590 | void mnt_pin(struct vfsmount *mnt) | |
591 | { | |
592 | spin_lock(&vfsmount_lock); | |
593 | mnt->mnt_pinned++; | |
594 | spin_unlock(&vfsmount_lock); | |
595 | } | |
596 | ||
597 | EXPORT_SYMBOL(mnt_pin); | |
598 | ||
599 | void mnt_unpin(struct vfsmount *mnt) | |
600 | { | |
601 | spin_lock(&vfsmount_lock); | |
602 | if (mnt->mnt_pinned) { | |
603 | atomic_inc(&mnt->mnt_count); | |
604 | mnt->mnt_pinned--; | |
605 | } | |
606 | spin_unlock(&vfsmount_lock); | |
607 | } | |
608 | ||
609 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 610 | |
b3b304a2 MS |
611 | static inline void mangle(struct seq_file *m, const char *s) |
612 | { | |
613 | seq_escape(m, s, " \t\n\\"); | |
614 | } | |
615 | ||
616 | /* | |
617 | * Simple .show_options callback for filesystems which don't want to | |
618 | * implement more complex mount option showing. | |
619 | * | |
620 | * See also save_mount_options(). | |
621 | */ | |
622 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
623 | { | |
624 | const char *options = mnt->mnt_sb->s_options; | |
625 | ||
626 | if (options != NULL && options[0]) { | |
627 | seq_putc(m, ','); | |
628 | mangle(m, options); | |
629 | } | |
630 | ||
631 | return 0; | |
632 | } | |
633 | EXPORT_SYMBOL(generic_show_options); | |
634 | ||
635 | /* | |
636 | * If filesystem uses generic_show_options(), this function should be | |
637 | * called from the fill_super() callback. | |
638 | * | |
639 | * The .remount_fs callback usually needs to be handled in a special | |
640 | * way, to make sure, that previous options are not overwritten if the | |
641 | * remount fails. | |
642 | * | |
643 | * Also note, that if the filesystem's .remount_fs function doesn't | |
644 | * reset all options to their default value, but changes only newly | |
645 | * given options, then the displayed options will not reflect reality | |
646 | * any more. | |
647 | */ | |
648 | void save_mount_options(struct super_block *sb, char *options) | |
649 | { | |
650 | kfree(sb->s_options); | |
651 | sb->s_options = kstrdup(options, GFP_KERNEL); | |
652 | } | |
653 | EXPORT_SYMBOL(save_mount_options); | |
654 | ||
1da177e4 LT |
655 | /* iterator */ |
656 | static void *m_start(struct seq_file *m, loff_t *pos) | |
657 | { | |
6b3286ed | 658 | struct mnt_namespace *n = m->private; |
1da177e4 | 659 | |
390c6843 | 660 | down_read(&namespace_sem); |
b0765fb8 | 661 | return seq_list_start(&n->list, *pos); |
1da177e4 LT |
662 | } |
663 | ||
664 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
665 | { | |
6b3286ed | 666 | struct mnt_namespace *n = m->private; |
b0765fb8 PE |
667 | |
668 | return seq_list_next(v, &n->list, pos); | |
1da177e4 LT |
669 | } |
670 | ||
671 | static void m_stop(struct seq_file *m, void *v) | |
672 | { | |
390c6843 | 673 | up_read(&namespace_sem); |
1da177e4 LT |
674 | } |
675 | ||
1da177e4 LT |
676 | static int show_vfsmnt(struct seq_file *m, void *v) |
677 | { | |
b0765fb8 | 678 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
1da177e4 LT |
679 | int err = 0; |
680 | static struct proc_fs_info { | |
681 | int flag; | |
682 | char *str; | |
683 | } fs_info[] = { | |
684 | { MS_SYNCHRONOUS, ",sync" }, | |
685 | { MS_DIRSYNC, ",dirsync" }, | |
686 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
687 | { 0, NULL } |
688 | }; | |
689 | static struct proc_fs_info mnt_info[] = { | |
690 | { MNT_NOSUID, ",nosuid" }, | |
691 | { MNT_NODEV, ",nodev" }, | |
692 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
693 | { MNT_NOATIME, ",noatime" }, |
694 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 695 | { MNT_RELATIME, ",relatime" }, |
1da177e4 LT |
696 | { 0, NULL } |
697 | }; | |
698 | struct proc_fs_info *fs_infop; | |
c32c2f63 | 699 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
700 | |
701 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
702 | seq_putc(m, ' '); | |
c32c2f63 | 703 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 LT |
704 | seq_putc(m, ' '); |
705 | mangle(m, mnt->mnt_sb->s_type->name); | |
79c0b2df MS |
706 | if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { |
707 | seq_putc(m, '.'); | |
708 | mangle(m, mnt->mnt_sb->s_subtype); | |
709 | } | |
2e4b7fcd | 710 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
1da177e4 LT |
711 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { |
712 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
713 | seq_puts(m, fs_infop->str); | |
714 | } | |
715 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
716 | if (mnt->mnt_flags & fs_infop->flag) | |
717 | seq_puts(m, fs_infop->str); | |
718 | } | |
719 | if (mnt->mnt_sb->s_op->show_options) | |
720 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
721 | seq_puts(m, " 0 0\n"); | |
722 | return err; | |
723 | } | |
724 | ||
725 | struct seq_operations mounts_op = { | |
726 | .start = m_start, | |
727 | .next = m_next, | |
728 | .stop = m_stop, | |
729 | .show = show_vfsmnt | |
730 | }; | |
731 | ||
b4629fe2 CL |
732 | static int show_vfsstat(struct seq_file *m, void *v) |
733 | { | |
b0765fb8 | 734 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 735 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
736 | int err = 0; |
737 | ||
738 | /* device */ | |
739 | if (mnt->mnt_devname) { | |
740 | seq_puts(m, "device "); | |
741 | mangle(m, mnt->mnt_devname); | |
742 | } else | |
743 | seq_puts(m, "no device"); | |
744 | ||
745 | /* mount point */ | |
746 | seq_puts(m, " mounted on "); | |
c32c2f63 | 747 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
748 | seq_putc(m, ' '); |
749 | ||
750 | /* file system type */ | |
751 | seq_puts(m, "with fstype "); | |
752 | mangle(m, mnt->mnt_sb->s_type->name); | |
753 | ||
754 | /* optional statistics */ | |
755 | if (mnt->mnt_sb->s_op->show_stats) { | |
756 | seq_putc(m, ' '); | |
757 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
758 | } | |
759 | ||
760 | seq_putc(m, '\n'); | |
761 | return err; | |
762 | } | |
763 | ||
764 | struct seq_operations mountstats_op = { | |
765 | .start = m_start, | |
766 | .next = m_next, | |
767 | .stop = m_stop, | |
768 | .show = show_vfsstat, | |
769 | }; | |
770 | ||
1da177e4 LT |
771 | /** |
772 | * may_umount_tree - check if a mount tree is busy | |
773 | * @mnt: root of mount tree | |
774 | * | |
775 | * This is called to check if a tree of mounts has any | |
776 | * open files, pwds, chroots or sub mounts that are | |
777 | * busy. | |
778 | */ | |
779 | int may_umount_tree(struct vfsmount *mnt) | |
780 | { | |
36341f64 RP |
781 | int actual_refs = 0; |
782 | int minimum_refs = 0; | |
783 | struct vfsmount *p; | |
1da177e4 LT |
784 | |
785 | spin_lock(&vfsmount_lock); | |
36341f64 | 786 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
787 | actual_refs += atomic_read(&p->mnt_count); |
788 | minimum_refs += 2; | |
1da177e4 LT |
789 | } |
790 | spin_unlock(&vfsmount_lock); | |
791 | ||
792 | if (actual_refs > minimum_refs) | |
e3474a8e | 793 | return 0; |
1da177e4 | 794 | |
e3474a8e | 795 | return 1; |
1da177e4 LT |
796 | } |
797 | ||
798 | EXPORT_SYMBOL(may_umount_tree); | |
799 | ||
800 | /** | |
801 | * may_umount - check if a mount point is busy | |
802 | * @mnt: root of mount | |
803 | * | |
804 | * This is called to check if a mount point has any | |
805 | * open files, pwds, chroots or sub mounts. If the | |
806 | * mount has sub mounts this will return busy | |
807 | * regardless of whether the sub mounts are busy. | |
808 | * | |
809 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
810 | * give false negatives. The main reason why it's here is that we need | |
811 | * a non-destructive way to look for easily umountable filesystems. | |
812 | */ | |
813 | int may_umount(struct vfsmount *mnt) | |
814 | { | |
e3474a8e | 815 | int ret = 1; |
a05964f3 RP |
816 | spin_lock(&vfsmount_lock); |
817 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 818 | ret = 0; |
a05964f3 RP |
819 | spin_unlock(&vfsmount_lock); |
820 | return ret; | |
1da177e4 LT |
821 | } |
822 | ||
823 | EXPORT_SYMBOL(may_umount); | |
824 | ||
b90fa9ae | 825 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
826 | { |
827 | struct vfsmount *mnt; | |
bf066c7d | 828 | while (!list_empty(head)) { |
b5e61818 | 829 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
830 | list_del_init(&mnt->mnt_hash); |
831 | if (mnt->mnt_parent != mnt) { | |
832 | struct dentry *dentry; | |
833 | struct vfsmount *m; | |
834 | spin_lock(&vfsmount_lock); | |
835 | dentry = mnt->mnt_mountpoint; | |
836 | m = mnt->mnt_parent; | |
837 | mnt->mnt_mountpoint = mnt->mnt_root; | |
838 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 839 | m->mnt_ghosts--; |
70fbcdf4 RP |
840 | spin_unlock(&vfsmount_lock); |
841 | dput(dentry); | |
842 | mntput(m); | |
843 | } | |
844 | mntput(mnt); | |
845 | } | |
846 | } | |
847 | ||
a05964f3 | 848 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
849 | { |
850 | struct vfsmount *p; | |
1da177e4 | 851 | |
1bfba4e8 AM |
852 | for (p = mnt; p; p = next_mnt(p, mnt)) |
853 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 854 | |
a05964f3 RP |
855 | if (propagate) |
856 | propagate_umount(kill); | |
857 | ||
70fbcdf4 RP |
858 | list_for_each_entry(p, kill, mnt_hash) { |
859 | list_del_init(&p->mnt_expire); | |
860 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
861 | __touch_mnt_namespace(p->mnt_ns); |
862 | p->mnt_ns = NULL; | |
70fbcdf4 | 863 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
864 | if (p->mnt_parent != p) { |
865 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 866 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 867 | } |
a05964f3 | 868 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
869 | } |
870 | } | |
871 | ||
c35038be AV |
872 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
873 | ||
1da177e4 LT |
874 | static int do_umount(struct vfsmount *mnt, int flags) |
875 | { | |
b58fed8b | 876 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 877 | int retval; |
70fbcdf4 | 878 | LIST_HEAD(umount_list); |
1da177e4 LT |
879 | |
880 | retval = security_sb_umount(mnt, flags); | |
881 | if (retval) | |
882 | return retval; | |
883 | ||
884 | /* | |
885 | * Allow userspace to request a mountpoint be expired rather than | |
886 | * unmounting unconditionally. Unmount only happens if: | |
887 | * (1) the mark is already set (the mark is cleared by mntput()) | |
888 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
889 | */ | |
890 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 891 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
892 | flags & (MNT_FORCE | MNT_DETACH)) |
893 | return -EINVAL; | |
894 | ||
895 | if (atomic_read(&mnt->mnt_count) != 2) | |
896 | return -EBUSY; | |
897 | ||
898 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
899 | return -EAGAIN; | |
900 | } | |
901 | ||
902 | /* | |
903 | * If we may have to abort operations to get out of this | |
904 | * mount, and they will themselves hold resources we must | |
905 | * allow the fs to do things. In the Unix tradition of | |
906 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
907 | * might fail to complete on the first run through as other tasks | |
908 | * must return, and the like. Thats for the mount program to worry | |
909 | * about for the moment. | |
910 | */ | |
911 | ||
912 | lock_kernel(); | |
8b512d9a TM |
913 | if (sb->s_op->umount_begin) |
914 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
915 | unlock_kernel(); |
916 | ||
917 | /* | |
918 | * No sense to grab the lock for this test, but test itself looks | |
919 | * somewhat bogus. Suggestions for better replacement? | |
920 | * Ho-hum... In principle, we might treat that as umount + switch | |
921 | * to rootfs. GC would eventually take care of the old vfsmount. | |
922 | * Actually it makes sense, especially if rootfs would contain a | |
923 | * /reboot - static binary that would close all descriptors and | |
924 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
925 | */ | |
6ac08c39 | 926 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
927 | /* |
928 | * Special case for "unmounting" root ... | |
929 | * we just try to remount it readonly. | |
930 | */ | |
931 | down_write(&sb->s_umount); | |
932 | if (!(sb->s_flags & MS_RDONLY)) { | |
933 | lock_kernel(); | |
934 | DQUOT_OFF(sb); | |
935 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
936 | unlock_kernel(); | |
937 | } | |
938 | up_write(&sb->s_umount); | |
939 | return retval; | |
940 | } | |
941 | ||
390c6843 | 942 | down_write(&namespace_sem); |
1da177e4 | 943 | spin_lock(&vfsmount_lock); |
5addc5dd | 944 | event++; |
1da177e4 | 945 | |
c35038be AV |
946 | if (!(flags & MNT_DETACH)) |
947 | shrink_submounts(mnt, &umount_list); | |
948 | ||
1da177e4 | 949 | retval = -EBUSY; |
a05964f3 | 950 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 951 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 952 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
953 | retval = 0; |
954 | } | |
955 | spin_unlock(&vfsmount_lock); | |
956 | if (retval) | |
957 | security_sb_umount_busy(mnt); | |
390c6843 | 958 | up_write(&namespace_sem); |
70fbcdf4 | 959 | release_mounts(&umount_list); |
1da177e4 LT |
960 | return retval; |
961 | } | |
962 | ||
963 | /* | |
964 | * Now umount can handle mount points as well as block devices. | |
965 | * This is important for filesystems which use unnamed block devices. | |
966 | * | |
967 | * We now support a flag for forced unmount like the other 'big iron' | |
968 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
969 | */ | |
970 | ||
971 | asmlinkage long sys_umount(char __user * name, int flags) | |
972 | { | |
973 | struct nameidata nd; | |
974 | int retval; | |
975 | ||
976 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
977 | if (retval) | |
978 | goto out; | |
979 | retval = -EINVAL; | |
4ac91378 | 980 | if (nd.path.dentry != nd.path.mnt->mnt_root) |
1da177e4 | 981 | goto dput_and_out; |
4ac91378 | 982 | if (!check_mnt(nd.path.mnt)) |
1da177e4 LT |
983 | goto dput_and_out; |
984 | ||
985 | retval = -EPERM; | |
986 | if (!capable(CAP_SYS_ADMIN)) | |
987 | goto dput_and_out; | |
988 | ||
4ac91378 | 989 | retval = do_umount(nd.path.mnt, flags); |
1da177e4 | 990 | dput_and_out: |
429731b1 | 991 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
4ac91378 JB |
992 | dput(nd.path.dentry); |
993 | mntput_no_expire(nd.path.mnt); | |
1da177e4 LT |
994 | out: |
995 | return retval; | |
996 | } | |
997 | ||
998 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
999 | ||
1000 | /* | |
b58fed8b | 1001 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1002 | */ |
1da177e4 LT |
1003 | asmlinkage long sys_oldumount(char __user * name) |
1004 | { | |
b58fed8b | 1005 | return sys_umount(name, 0); |
1da177e4 LT |
1006 | } |
1007 | ||
1008 | #endif | |
1009 | ||
1010 | static int mount_is_safe(struct nameidata *nd) | |
1011 | { | |
1012 | if (capable(CAP_SYS_ADMIN)) | |
1013 | return 0; | |
1014 | return -EPERM; | |
1015 | #ifdef notyet | |
4ac91378 | 1016 | if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) |
1da177e4 | 1017 | return -EPERM; |
4ac91378 JB |
1018 | if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { |
1019 | if (current->uid != nd->path.dentry->d_inode->i_uid) | |
1da177e4 LT |
1020 | return -EPERM; |
1021 | } | |
e4543edd | 1022 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
1023 | return -EPERM; |
1024 | return 0; | |
1025 | #endif | |
1026 | } | |
1027 | ||
b58fed8b | 1028 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
1029 | { |
1030 | while (1) { | |
1031 | if (d == dentry) | |
1032 | return 1; | |
1033 | if (d == NULL || d == d->d_parent) | |
1034 | return 0; | |
1035 | d = d->d_parent; | |
1036 | } | |
1037 | } | |
1038 | ||
b90fa9ae | 1039 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1040 | int flag) |
1da177e4 LT |
1041 | { |
1042 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1043 | struct path path; |
1da177e4 | 1044 | |
9676f0c6 RP |
1045 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1046 | return NULL; | |
1047 | ||
36341f64 | 1048 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1049 | if (!q) |
1050 | goto Enomem; | |
1051 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1052 | ||
1053 | p = mnt; | |
fdadd65f | 1054 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
1055 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
1056 | continue; | |
1057 | ||
1058 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1059 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1060 | s = skip_mnt_tree(s); | |
1061 | continue; | |
1062 | } | |
1da177e4 LT |
1063 | while (p != s->mnt_parent) { |
1064 | p = p->mnt_parent; | |
1065 | q = q->mnt_parent; | |
1066 | } | |
1067 | p = s; | |
1a390689 AV |
1068 | path.mnt = q; |
1069 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1070 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1071 | if (!q) |
1072 | goto Enomem; | |
1073 | spin_lock(&vfsmount_lock); | |
1074 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1075 | attach_mnt(q, &path); |
1da177e4 LT |
1076 | spin_unlock(&vfsmount_lock); |
1077 | } | |
1078 | } | |
1079 | return res; | |
b58fed8b | 1080 | Enomem: |
1da177e4 | 1081 | if (res) { |
70fbcdf4 | 1082 | LIST_HEAD(umount_list); |
1da177e4 | 1083 | spin_lock(&vfsmount_lock); |
a05964f3 | 1084 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1085 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1086 | release_mounts(&umount_list); |
1da177e4 LT |
1087 | } |
1088 | return NULL; | |
1089 | } | |
1090 | ||
8aec0809 AV |
1091 | struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) |
1092 | { | |
1093 | struct vfsmount *tree; | |
1a60a280 | 1094 | down_write(&namespace_sem); |
8aec0809 | 1095 | tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1096 | up_write(&namespace_sem); |
8aec0809 AV |
1097 | return tree; |
1098 | } | |
1099 | ||
1100 | void drop_collected_mounts(struct vfsmount *mnt) | |
1101 | { | |
1102 | LIST_HEAD(umount_list); | |
1a60a280 | 1103 | down_write(&namespace_sem); |
8aec0809 AV |
1104 | spin_lock(&vfsmount_lock); |
1105 | umount_tree(mnt, 0, &umount_list); | |
1106 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1107 | up_write(&namespace_sem); |
8aec0809 AV |
1108 | release_mounts(&umount_list); |
1109 | } | |
1110 | ||
b90fa9ae RP |
1111 | /* |
1112 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1113 | * @nd : place the mount tree @source_mnt is attached |
1114 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1115 | * store the parent mount and mountpoint dentry. | |
1116 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1117 | * |
1118 | * NOTE: in the table below explains the semantics when a source mount | |
1119 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1120 | * --------------------------------------------------------------------------- |
1121 | * | BIND MOUNT OPERATION | | |
1122 | * |************************************************************************** | |
1123 | * | source-->| shared | private | slave | unbindable | | |
1124 | * | dest | | | | | | |
1125 | * | | | | | | | | |
1126 | * | v | | | | | | |
1127 | * |************************************************************************** | |
1128 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1129 | * | | | | | | | |
1130 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1131 | * *************************************************************************** | |
b90fa9ae RP |
1132 | * A bind operation clones the source mount and mounts the clone on the |
1133 | * destination mount. | |
1134 | * | |
1135 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1136 | * tree of the destination mount and the cloned mount is added to | |
1137 | * the peer group of the source mount. | |
1138 | * (+) the cloned mount is created under the destination mount and is marked | |
1139 | * as shared. The cloned mount is added to the peer group of the source | |
1140 | * mount. | |
5afe0022 RP |
1141 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1142 | * of the destination mount and the cloned mount is made slave | |
1143 | * of the same master as that of the source mount. The cloned mount | |
1144 | * is marked as 'shared and slave'. | |
1145 | * (*) the cloned mount is made a slave of the same master as that of the | |
1146 | * source mount. | |
1147 | * | |
9676f0c6 RP |
1148 | * --------------------------------------------------------------------------- |
1149 | * | MOVE MOUNT OPERATION | | |
1150 | * |************************************************************************** | |
1151 | * | source-->| shared | private | slave | unbindable | | |
1152 | * | dest | | | | | | |
1153 | * | | | | | | | | |
1154 | * | v | | | | | | |
1155 | * |************************************************************************** | |
1156 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1157 | * | | | | | | | |
1158 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1159 | * *************************************************************************** | |
5afe0022 RP |
1160 | * |
1161 | * (+) the mount is moved to the destination. And is then propagated to | |
1162 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1163 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1164 | * (+++) the mount is moved to the destination and is then propagated to |
1165 | * all the mounts belonging to the destination mount's propagation tree. | |
1166 | * the mount is marked as 'shared and slave'. | |
1167 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1168 | * |
1169 | * if the source mount is a tree, the operations explained above is | |
1170 | * applied to each mount in the tree. | |
1171 | * Must be called without spinlocks held, since this function can sleep | |
1172 | * in allocations. | |
1173 | */ | |
1174 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1175 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1176 | { |
1177 | LIST_HEAD(tree_list); | |
1a390689 AV |
1178 | struct vfsmount *dest_mnt = path->mnt; |
1179 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae RP |
1180 | struct vfsmount *child, *p; |
1181 | ||
1182 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
1183 | return -EINVAL; | |
1184 | ||
1185 | if (IS_MNT_SHARED(dest_mnt)) { | |
1186 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1187 | set_mnt_shared(p); | |
1188 | } | |
1189 | ||
1190 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1191 | if (parent_path) { |
1192 | detach_mnt(source_mnt, parent_path); | |
1193 | attach_mnt(source_mnt, path); | |
6b3286ed | 1194 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
21444403 RP |
1195 | } else { |
1196 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1197 | commit_tree(source_mnt); | |
1198 | } | |
b90fa9ae RP |
1199 | |
1200 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1201 | list_del_init(&child->mnt_hash); | |
1202 | commit_tree(child); | |
1203 | } | |
1204 | spin_unlock(&vfsmount_lock); | |
1205 | return 0; | |
1206 | } | |
1207 | ||
1da177e4 LT |
1208 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
1209 | { | |
1210 | int err; | |
1211 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1212 | return -EINVAL; | |
1213 | ||
4ac91378 | 1214 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1da177e4 LT |
1215 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1216 | return -ENOTDIR; | |
1217 | ||
1218 | err = -ENOENT; | |
4ac91378 JB |
1219 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1220 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1221 | goto out_unlock; |
1222 | ||
b5266eb4 | 1223 | err = security_sb_check_sb(mnt, &nd->path); |
1da177e4 LT |
1224 | if (err) |
1225 | goto out_unlock; | |
1226 | ||
1227 | err = -ENOENT; | |
4ac91378 | 1228 | if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) |
1a390689 | 1229 | err = attach_recursive_mnt(mnt, &nd->path, NULL); |
1da177e4 | 1230 | out_unlock: |
4ac91378 | 1231 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1232 | if (!err) |
b5266eb4 | 1233 | security_sb_post_addmount(mnt, &nd->path); |
1da177e4 LT |
1234 | return err; |
1235 | } | |
1236 | ||
07b20889 RP |
1237 | /* |
1238 | * recursively change the type of the mountpoint. | |
2dafe1c4 | 1239 | * noinline this do_mount helper to save do_mount stack space. |
07b20889 | 1240 | */ |
2dafe1c4 | 1241 | static noinline int do_change_type(struct nameidata *nd, int flag) |
07b20889 | 1242 | { |
4ac91378 | 1243 | struct vfsmount *m, *mnt = nd->path.mnt; |
07b20889 RP |
1244 | int recurse = flag & MS_REC; |
1245 | int type = flag & ~MS_REC; | |
1246 | ||
ee6f9582 MS |
1247 | if (!capable(CAP_SYS_ADMIN)) |
1248 | return -EPERM; | |
1249 | ||
4ac91378 | 1250 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
07b20889 RP |
1251 | return -EINVAL; |
1252 | ||
1253 | down_write(&namespace_sem); | |
1254 | spin_lock(&vfsmount_lock); | |
1255 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1256 | change_mnt_propagation(m, type); | |
1257 | spin_unlock(&vfsmount_lock); | |
1258 | up_write(&namespace_sem); | |
1259 | return 0; | |
1260 | } | |
1261 | ||
1da177e4 LT |
1262 | /* |
1263 | * do loopback mount. | |
2dafe1c4 | 1264 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1265 | */ |
2dafe1c4 ES |
1266 | static noinline int do_loopback(struct nameidata *nd, char *old_name, |
1267 | int recurse) | |
1da177e4 LT |
1268 | { |
1269 | struct nameidata old_nd; | |
1270 | struct vfsmount *mnt = NULL; | |
1271 | int err = mount_is_safe(nd); | |
1272 | if (err) | |
1273 | return err; | |
1274 | if (!old_name || !*old_name) | |
1275 | return -EINVAL; | |
1276 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1277 | if (err) | |
1278 | return err; | |
1279 | ||
390c6843 | 1280 | down_write(&namespace_sem); |
1da177e4 | 1281 | err = -EINVAL; |
4ac91378 JB |
1282 | if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) |
1283 | goto out; | |
9676f0c6 | 1284 | |
4ac91378 | 1285 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
ccd48bc7 | 1286 | goto out; |
1da177e4 | 1287 | |
ccd48bc7 AV |
1288 | err = -ENOMEM; |
1289 | if (recurse) | |
4ac91378 | 1290 | mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 | 1291 | else |
4ac91378 | 1292 | mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); |
ccd48bc7 AV |
1293 | |
1294 | if (!mnt) | |
1295 | goto out; | |
1296 | ||
ccd48bc7 AV |
1297 | err = graft_tree(mnt, nd); |
1298 | if (err) { | |
70fbcdf4 | 1299 | LIST_HEAD(umount_list); |
1da177e4 | 1300 | spin_lock(&vfsmount_lock); |
a05964f3 | 1301 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1302 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1303 | release_mounts(&umount_list); |
5b83d2c5 | 1304 | } |
1da177e4 | 1305 | |
ccd48bc7 | 1306 | out: |
390c6843 | 1307 | up_write(&namespace_sem); |
1d957f9b | 1308 | path_put(&old_nd.path); |
1da177e4 LT |
1309 | return err; |
1310 | } | |
1311 | ||
2e4b7fcd DH |
1312 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1313 | { | |
1314 | int error = 0; | |
1315 | int readonly_request = 0; | |
1316 | ||
1317 | if (ms_flags & MS_RDONLY) | |
1318 | readonly_request = 1; | |
1319 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1320 | return 0; | |
1321 | ||
1322 | if (readonly_request) | |
1323 | error = mnt_make_readonly(mnt); | |
1324 | else | |
1325 | __mnt_unmake_readonly(mnt); | |
1326 | return error; | |
1327 | } | |
1328 | ||
1da177e4 LT |
1329 | /* |
1330 | * change filesystem flags. dir should be a physical root of filesystem. | |
1331 | * If you've mounted a non-root directory somewhere and want to do remount | |
1332 | * on it - tough luck. | |
2dafe1c4 | 1333 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1334 | */ |
2dafe1c4 | 1335 | static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
1da177e4 LT |
1336 | void *data) |
1337 | { | |
1338 | int err; | |
4ac91378 | 1339 | struct super_block *sb = nd->path.mnt->mnt_sb; |
1da177e4 LT |
1340 | |
1341 | if (!capable(CAP_SYS_ADMIN)) | |
1342 | return -EPERM; | |
1343 | ||
4ac91378 | 1344 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1345 | return -EINVAL; |
1346 | ||
4ac91378 | 1347 | if (nd->path.dentry != nd->path.mnt->mnt_root) |
1da177e4 LT |
1348 | return -EINVAL; |
1349 | ||
1350 | down_write(&sb->s_umount); | |
2e4b7fcd DH |
1351 | if (flags & MS_BIND) |
1352 | err = change_mount_flags(nd->path.mnt, flags); | |
1353 | else | |
1354 | err = do_remount_sb(sb, flags, data, 0); | |
1da177e4 | 1355 | if (!err) |
4ac91378 | 1356 | nd->path.mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
1357 | up_write(&sb->s_umount); |
1358 | if (!err) | |
4ac91378 | 1359 | security_sb_post_remount(nd->path.mnt, flags, data); |
1da177e4 LT |
1360 | return err; |
1361 | } | |
1362 | ||
9676f0c6 RP |
1363 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1364 | { | |
1365 | struct vfsmount *p; | |
1366 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1367 | if (IS_MNT_UNBINDABLE(p)) | |
1368 | return 1; | |
1369 | } | |
1370 | return 0; | |
1371 | } | |
1372 | ||
2dafe1c4 ES |
1373 | /* |
1374 | * noinline this do_mount helper to save do_mount stack space. | |
1375 | */ | |
1376 | static noinline int do_move_mount(struct nameidata *nd, char *old_name) | |
1da177e4 | 1377 | { |
1a390689 AV |
1378 | struct nameidata old_nd; |
1379 | struct path parent_path; | |
1da177e4 LT |
1380 | struct vfsmount *p; |
1381 | int err = 0; | |
1382 | if (!capable(CAP_SYS_ADMIN)) | |
1383 | return -EPERM; | |
1384 | if (!old_name || !*old_name) | |
1385 | return -EINVAL; | |
1386 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
1387 | if (err) | |
1388 | return err; | |
1389 | ||
390c6843 | 1390 | down_write(&namespace_sem); |
4ac91378 JB |
1391 | while (d_mountpoint(nd->path.dentry) && |
1392 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1393 | ; |
1394 | err = -EINVAL; | |
4ac91378 | 1395 | if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) |
1da177e4 LT |
1396 | goto out; |
1397 | ||
1398 | err = -ENOENT; | |
4ac91378 JB |
1399 | mutex_lock(&nd->path.dentry->d_inode->i_mutex); |
1400 | if (IS_DEADDIR(nd->path.dentry->d_inode)) | |
1da177e4 LT |
1401 | goto out1; |
1402 | ||
4ac91378 | 1403 | if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) |
21444403 | 1404 | goto out1; |
1da177e4 LT |
1405 | |
1406 | err = -EINVAL; | |
4ac91378 | 1407 | if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) |
21444403 | 1408 | goto out1; |
1da177e4 | 1409 | |
4ac91378 | 1410 | if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) |
21444403 | 1411 | goto out1; |
1da177e4 | 1412 | |
4ac91378 JB |
1413 | if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != |
1414 | S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) | |
21444403 RP |
1415 | goto out1; |
1416 | /* | |
1417 | * Don't move a mount residing in a shared parent. | |
1418 | */ | |
4ac91378 JB |
1419 | if (old_nd.path.mnt->mnt_parent && |
1420 | IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) | |
21444403 | 1421 | goto out1; |
9676f0c6 RP |
1422 | /* |
1423 | * Don't move a mount tree containing unbindable mounts to a destination | |
1424 | * mount which is shared. | |
1425 | */ | |
4ac91378 JB |
1426 | if (IS_MNT_SHARED(nd->path.mnt) && |
1427 | tree_contains_unbindable(old_nd.path.mnt)) | |
9676f0c6 | 1428 | goto out1; |
1da177e4 | 1429 | err = -ELOOP; |
4ac91378 JB |
1430 | for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) |
1431 | if (p == old_nd.path.mnt) | |
21444403 | 1432 | goto out1; |
1da177e4 | 1433 | |
1a390689 | 1434 | err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); |
4ac91378 | 1435 | if (err) |
21444403 | 1436 | goto out1; |
1da177e4 LT |
1437 | |
1438 | /* if the mount is moved, it should no longer be expire | |
1439 | * automatically */ | |
4ac91378 | 1440 | list_del_init(&old_nd.path.mnt->mnt_expire); |
1da177e4 | 1441 | out1: |
4ac91378 | 1442 | mutex_unlock(&nd->path.dentry->d_inode->i_mutex); |
1da177e4 | 1443 | out: |
390c6843 | 1444 | up_write(&namespace_sem); |
1da177e4 | 1445 | if (!err) |
1a390689 | 1446 | path_put(&parent_path); |
1d957f9b | 1447 | path_put(&old_nd.path); |
1da177e4 LT |
1448 | return err; |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * create a new mount for userspace and request it to be added into the | |
1453 | * namespace's tree | |
2dafe1c4 | 1454 | * noinline this do_mount helper to save do_mount stack space. |
1da177e4 | 1455 | */ |
2dafe1c4 | 1456 | static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, |
1da177e4 LT |
1457 | int mnt_flags, char *name, void *data) |
1458 | { | |
1459 | struct vfsmount *mnt; | |
1460 | ||
1461 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1462 | return -EINVAL; | |
1463 | ||
1464 | /* we need capabilities... */ | |
1465 | if (!capable(CAP_SYS_ADMIN)) | |
1466 | return -EPERM; | |
1467 | ||
1468 | mnt = do_kern_mount(type, flags, name, data); | |
1469 | if (IS_ERR(mnt)) | |
1470 | return PTR_ERR(mnt); | |
1471 | ||
1472 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * add a mount into a namespace's mount tree | |
1477 | * - provide the option of adding the new mount to an expiration list | |
1478 | */ | |
1479 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1480 | int mnt_flags, struct list_head *fslist) | |
1481 | { | |
1482 | int err; | |
1483 | ||
390c6843 | 1484 | down_write(&namespace_sem); |
1da177e4 | 1485 | /* Something was mounted here while we slept */ |
4ac91378 JB |
1486 | while (d_mountpoint(nd->path.dentry) && |
1487 | follow_down(&nd->path.mnt, &nd->path.dentry)) | |
1da177e4 LT |
1488 | ; |
1489 | err = -EINVAL; | |
4ac91378 | 1490 | if (!check_mnt(nd->path.mnt)) |
1da177e4 LT |
1491 | goto unlock; |
1492 | ||
1493 | /* Refuse the same filesystem on the same mount point */ | |
1494 | err = -EBUSY; | |
4ac91378 JB |
1495 | if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && |
1496 | nd->path.mnt->mnt_root == nd->path.dentry) | |
1da177e4 LT |
1497 | goto unlock; |
1498 | ||
1499 | err = -EINVAL; | |
1500 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1501 | goto unlock; | |
1502 | ||
1503 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1504 | if ((err = graft_tree(newmnt, nd))) |
1505 | goto unlock; | |
1da177e4 | 1506 | |
6758f953 | 1507 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1508 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1509 | |
390c6843 | 1510 | up_write(&namespace_sem); |
5b83d2c5 | 1511 | return 0; |
1da177e4 LT |
1512 | |
1513 | unlock: | |
390c6843 | 1514 | up_write(&namespace_sem); |
1da177e4 LT |
1515 | mntput(newmnt); |
1516 | return err; | |
1517 | } | |
1518 | ||
1519 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1520 | ||
1521 | /* | |
1522 | * process a list of expirable mountpoints with the intent of discarding any | |
1523 | * mountpoints that aren't in use and haven't been touched since last we came | |
1524 | * here | |
1525 | */ | |
1526 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1527 | { | |
1da177e4 LT |
1528 | struct vfsmount *mnt, *next; |
1529 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1530 | LIST_HEAD(umounts); |
1da177e4 LT |
1531 | |
1532 | if (list_empty(mounts)) | |
1533 | return; | |
1534 | ||
bcc5c7d2 | 1535 | down_write(&namespace_sem); |
1da177e4 LT |
1536 | spin_lock(&vfsmount_lock); |
1537 | ||
1538 | /* extract from the expiration list every vfsmount that matches the | |
1539 | * following criteria: | |
1540 | * - only referenced by its parent vfsmount | |
1541 | * - still marked for expiry (marked on the last call here; marks are | |
1542 | * cleared by mntput()) | |
1543 | */ | |
55e700b9 | 1544 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1545 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1546 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1547 | continue; |
55e700b9 | 1548 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1549 | } |
bcc5c7d2 AV |
1550 | while (!list_empty(&graveyard)) { |
1551 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1552 | touch_mnt_namespace(mnt->mnt_ns); | |
1553 | umount_tree(mnt, 1, &umounts); | |
1554 | } | |
5528f911 | 1555 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1556 | up_write(&namespace_sem); |
1557 | ||
1558 | release_mounts(&umounts); | |
5528f911 TM |
1559 | } |
1560 | ||
1561 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1562 | ||
1563 | /* | |
1564 | * Ripoff of 'select_parent()' | |
1565 | * | |
1566 | * search the list of submounts for a given mountpoint, and move any | |
1567 | * shrinkable submounts to the 'graveyard' list. | |
1568 | */ | |
1569 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1570 | { | |
1571 | struct vfsmount *this_parent = parent; | |
1572 | struct list_head *next; | |
1573 | int found = 0; | |
1574 | ||
1575 | repeat: | |
1576 | next = this_parent->mnt_mounts.next; | |
1577 | resume: | |
1578 | while (next != &this_parent->mnt_mounts) { | |
1579 | struct list_head *tmp = next; | |
1580 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1581 | ||
1582 | next = tmp->next; | |
1583 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1584 | continue; |
5528f911 TM |
1585 | /* |
1586 | * Descend a level if the d_mounts list is non-empty. | |
1587 | */ | |
1588 | if (!list_empty(&mnt->mnt_mounts)) { | |
1589 | this_parent = mnt; | |
1590 | goto repeat; | |
1591 | } | |
1da177e4 | 1592 | |
5528f911 | 1593 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1594 | list_move_tail(&mnt->mnt_expire, graveyard); |
1595 | found++; | |
1596 | } | |
1da177e4 | 1597 | } |
5528f911 TM |
1598 | /* |
1599 | * All done at this level ... ascend and resume the search | |
1600 | */ | |
1601 | if (this_parent != parent) { | |
1602 | next = this_parent->mnt_child.next; | |
1603 | this_parent = this_parent->mnt_parent; | |
1604 | goto resume; | |
1605 | } | |
1606 | return found; | |
1607 | } | |
1608 | ||
1609 | /* | |
1610 | * process a list of expirable mountpoints with the intent of discarding any | |
1611 | * submounts of a specific parent mountpoint | |
1612 | */ | |
c35038be | 1613 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1614 | { |
1615 | LIST_HEAD(graveyard); | |
c35038be | 1616 | struct vfsmount *m; |
5528f911 | 1617 | |
5528f911 | 1618 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1619 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1620 | while (!list_empty(&graveyard)) { |
c35038be | 1621 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 AV |
1622 | mnt_expire); |
1623 | touch_mnt_namespace(mnt->mnt_ns); | |
c35038be | 1624 | umount_tree(mnt, 1, umounts); |
bcc5c7d2 AV |
1625 | } |
1626 | } | |
1da177e4 LT |
1627 | } |
1628 | ||
1da177e4 LT |
1629 | /* |
1630 | * Some copy_from_user() implementations do not return the exact number of | |
1631 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1632 | * Note that this function differs from copy_from_user() in that it will oops | |
1633 | * on bad values of `to', rather than returning a short copy. | |
1634 | */ | |
b58fed8b RP |
1635 | static long exact_copy_from_user(void *to, const void __user * from, |
1636 | unsigned long n) | |
1da177e4 LT |
1637 | { |
1638 | char *t = to; | |
1639 | const char __user *f = from; | |
1640 | char c; | |
1641 | ||
1642 | if (!access_ok(VERIFY_READ, from, n)) | |
1643 | return n; | |
1644 | ||
1645 | while (n) { | |
1646 | if (__get_user(c, f)) { | |
1647 | memset(t, 0, n); | |
1648 | break; | |
1649 | } | |
1650 | *t++ = c; | |
1651 | f++; | |
1652 | n--; | |
1653 | } | |
1654 | return n; | |
1655 | } | |
1656 | ||
b58fed8b | 1657 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1658 | { |
1659 | int i; | |
1660 | unsigned long page; | |
1661 | unsigned long size; | |
b58fed8b | 1662 | |
1da177e4 LT |
1663 | *where = 0; |
1664 | if (!data) | |
1665 | return 0; | |
1666 | ||
1667 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1668 | return -ENOMEM; | |
1669 | ||
1670 | /* We only care that *some* data at the address the user | |
1671 | * gave us is valid. Just in case, we'll zero | |
1672 | * the remainder of the page. | |
1673 | */ | |
1674 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1675 | size = TASK_SIZE - (unsigned long)data; | |
1676 | if (size > PAGE_SIZE) | |
1677 | size = PAGE_SIZE; | |
1678 | ||
1679 | i = size - exact_copy_from_user((void *)page, data, size); | |
1680 | if (!i) { | |
b58fed8b | 1681 | free_page(page); |
1da177e4 LT |
1682 | return -EFAULT; |
1683 | } | |
1684 | if (i != PAGE_SIZE) | |
1685 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1686 | *where = page; | |
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1692 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1693 | * | |
1694 | * data is a (void *) that can point to any structure up to | |
1695 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1696 | * information (or be NULL). | |
1697 | * | |
1698 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1699 | * When the flags word was introduced its top half was required | |
1700 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1701 | * Therefore, if this magic number is present, it carries no information | |
1702 | * and must be discarded. | |
1703 | */ | |
b58fed8b | 1704 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1705 | unsigned long flags, void *data_page) |
1706 | { | |
1707 | struct nameidata nd; | |
1708 | int retval = 0; | |
1709 | int mnt_flags = 0; | |
1710 | ||
1711 | /* Discard magic */ | |
1712 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1713 | flags &= ~MS_MGC_MSK; | |
1714 | ||
1715 | /* Basic sanity checks */ | |
1716 | ||
1717 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1718 | return -EINVAL; | |
1719 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1720 | return -EINVAL; | |
1721 | ||
1722 | if (data_page) | |
1723 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1724 | ||
1725 | /* Separate the per-mountpoint flags */ | |
1726 | if (flags & MS_NOSUID) | |
1727 | mnt_flags |= MNT_NOSUID; | |
1728 | if (flags & MS_NODEV) | |
1729 | mnt_flags |= MNT_NODEV; | |
1730 | if (flags & MS_NOEXEC) | |
1731 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1732 | if (flags & MS_NOATIME) |
1733 | mnt_flags |= MNT_NOATIME; | |
1734 | if (flags & MS_NODIRATIME) | |
1735 | mnt_flags |= MNT_NODIRATIME; | |
47ae32d6 VH |
1736 | if (flags & MS_RELATIME) |
1737 | mnt_flags |= MNT_RELATIME; | |
2e4b7fcd DH |
1738 | if (flags & MS_RDONLY) |
1739 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1740 | |
1741 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
8bf9725c | 1742 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); |
1da177e4 LT |
1743 | |
1744 | /* ... and get the mountpoint */ | |
1745 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1746 | if (retval) | |
1747 | return retval; | |
1748 | ||
b5266eb4 AV |
1749 | retval = security_sb_mount(dev_name, &nd.path, |
1750 | type_page, flags, data_page); | |
1da177e4 LT |
1751 | if (retval) |
1752 | goto dput_out; | |
1753 | ||
1754 | if (flags & MS_REMOUNT) | |
1755 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1756 | data_page); | |
1757 | else if (flags & MS_BIND) | |
eee391a6 | 1758 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1759 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1760 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1761 | else if (flags & MS_MOVE) |
1762 | retval = do_move_mount(&nd, dev_name); | |
1763 | else | |
1764 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1765 | dev_name, data_page); | |
1766 | dput_out: | |
1d957f9b | 1767 | path_put(&nd.path); |
1da177e4 LT |
1768 | return retval; |
1769 | } | |
1770 | ||
741a2951 JD |
1771 | /* |
1772 | * Allocate a new namespace structure and populate it with contents | |
1773 | * copied from the namespace of the passed in task structure. | |
1774 | */ | |
e3222c4e | 1775 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1776 | struct fs_struct *fs) |
1da177e4 | 1777 | { |
6b3286ed | 1778 | struct mnt_namespace *new_ns; |
1da177e4 | 1779 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; |
1da177e4 LT |
1780 | struct vfsmount *p, *q; |
1781 | ||
6b3286ed | 1782 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
1da177e4 | 1783 | if (!new_ns) |
467e9f4b | 1784 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1785 | |
1786 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1787 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1788 | init_waitqueue_head(&new_ns->poll); |
1789 | new_ns->event = 0; | |
1da177e4 | 1790 | |
390c6843 | 1791 | down_write(&namespace_sem); |
1da177e4 | 1792 | /* First pass: copy the tree topology */ |
6b3286ed | 1793 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1794 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1795 | if (!new_ns->root) { |
390c6843 | 1796 | up_write(&namespace_sem); |
1da177e4 | 1797 | kfree(new_ns); |
467e9f4b | 1798 | return ERR_PTR(-ENOMEM);; |
1da177e4 LT |
1799 | } |
1800 | spin_lock(&vfsmount_lock); | |
1801 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1802 | spin_unlock(&vfsmount_lock); | |
1803 | ||
1804 | /* | |
1805 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1806 | * as belonging to new namespace. We have already acquired a private | |
1807 | * fs_struct, so tsk->fs->lock is not needed. | |
1808 | */ | |
6b3286ed | 1809 | p = mnt_ns->root; |
1da177e4 LT |
1810 | q = new_ns->root; |
1811 | while (p) { | |
6b3286ed | 1812 | q->mnt_ns = new_ns; |
1da177e4 | 1813 | if (fs) { |
6ac08c39 | 1814 | if (p == fs->root.mnt) { |
1da177e4 | 1815 | rootmnt = p; |
6ac08c39 | 1816 | fs->root.mnt = mntget(q); |
1da177e4 | 1817 | } |
6ac08c39 | 1818 | if (p == fs->pwd.mnt) { |
1da177e4 | 1819 | pwdmnt = p; |
6ac08c39 | 1820 | fs->pwd.mnt = mntget(q); |
1da177e4 | 1821 | } |
6ac08c39 | 1822 | if (p == fs->altroot.mnt) { |
1da177e4 | 1823 | altrootmnt = p; |
6ac08c39 | 1824 | fs->altroot.mnt = mntget(q); |
1da177e4 LT |
1825 | } |
1826 | } | |
6b3286ed | 1827 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
1828 | q = next_mnt(q, new_ns->root); |
1829 | } | |
390c6843 | 1830 | up_write(&namespace_sem); |
1da177e4 | 1831 | |
1da177e4 LT |
1832 | if (rootmnt) |
1833 | mntput(rootmnt); | |
1834 | if (pwdmnt) | |
1835 | mntput(pwdmnt); | |
1836 | if (altrootmnt) | |
1837 | mntput(altrootmnt); | |
1838 | ||
741a2951 JD |
1839 | return new_ns; |
1840 | } | |
1841 | ||
213dd266 | 1842 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 1843 | struct fs_struct *new_fs) |
741a2951 | 1844 | { |
6b3286ed | 1845 | struct mnt_namespace *new_ns; |
741a2951 | 1846 | |
e3222c4e | 1847 | BUG_ON(!ns); |
6b3286ed | 1848 | get_mnt_ns(ns); |
741a2951 JD |
1849 | |
1850 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 1851 | return ns; |
741a2951 | 1852 | |
e3222c4e | 1853 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 1854 | |
6b3286ed | 1855 | put_mnt_ns(ns); |
e3222c4e | 1856 | return new_ns; |
1da177e4 LT |
1857 | } |
1858 | ||
1859 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1860 | char __user * type, unsigned long flags, | |
1861 | void __user * data) | |
1862 | { | |
1863 | int retval; | |
1864 | unsigned long data_page; | |
1865 | unsigned long type_page; | |
1866 | unsigned long dev_page; | |
1867 | char *dir_page; | |
1868 | ||
b58fed8b | 1869 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1870 | if (retval < 0) |
1871 | return retval; | |
1872 | ||
1873 | dir_page = getname(dir_name); | |
1874 | retval = PTR_ERR(dir_page); | |
1875 | if (IS_ERR(dir_page)) | |
1876 | goto out1; | |
1877 | ||
b58fed8b | 1878 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1879 | if (retval < 0) |
1880 | goto out2; | |
1881 | ||
b58fed8b | 1882 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1883 | if (retval < 0) |
1884 | goto out3; | |
1885 | ||
1886 | lock_kernel(); | |
b58fed8b RP |
1887 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1888 | flags, (void *)data_page); | |
1da177e4 LT |
1889 | unlock_kernel(); |
1890 | free_page(data_page); | |
1891 | ||
1892 | out3: | |
1893 | free_page(dev_page); | |
1894 | out2: | |
1895 | putname(dir_page); | |
1896 | out1: | |
1897 | free_page(type_page); | |
1898 | return retval; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1903 | * It can block. Requires the big lock held. | |
1904 | */ | |
ac748a09 | 1905 | void set_fs_root(struct fs_struct *fs, struct path *path) |
1da177e4 | 1906 | { |
6ac08c39 JB |
1907 | struct path old_root; |
1908 | ||
1da177e4 LT |
1909 | write_lock(&fs->lock); |
1910 | old_root = fs->root; | |
ac748a09 JB |
1911 | fs->root = *path; |
1912 | path_get(path); | |
1da177e4 | 1913 | write_unlock(&fs->lock); |
6ac08c39 JB |
1914 | if (old_root.dentry) |
1915 | path_put(&old_root); | |
1da177e4 LT |
1916 | } |
1917 | ||
1918 | /* | |
1919 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1920 | * It can block. Requires the big lock held. | |
1921 | */ | |
ac748a09 | 1922 | void set_fs_pwd(struct fs_struct *fs, struct path *path) |
1da177e4 | 1923 | { |
6ac08c39 | 1924 | struct path old_pwd; |
1da177e4 LT |
1925 | |
1926 | write_lock(&fs->lock); | |
1927 | old_pwd = fs->pwd; | |
ac748a09 JB |
1928 | fs->pwd = *path; |
1929 | path_get(path); | |
1da177e4 LT |
1930 | write_unlock(&fs->lock); |
1931 | ||
6ac08c39 JB |
1932 | if (old_pwd.dentry) |
1933 | path_put(&old_pwd); | |
1da177e4 LT |
1934 | } |
1935 | ||
1a390689 | 1936 | static void chroot_fs_refs(struct path *old_root, struct path *new_root) |
1da177e4 LT |
1937 | { |
1938 | struct task_struct *g, *p; | |
1939 | struct fs_struct *fs; | |
1940 | ||
1941 | read_lock(&tasklist_lock); | |
1942 | do_each_thread(g, p) { | |
1943 | task_lock(p); | |
1944 | fs = p->fs; | |
1945 | if (fs) { | |
1946 | atomic_inc(&fs->count); | |
1947 | task_unlock(p); | |
1a390689 AV |
1948 | if (fs->root.dentry == old_root->dentry |
1949 | && fs->root.mnt == old_root->mnt) | |
1950 | set_fs_root(fs, new_root); | |
1951 | if (fs->pwd.dentry == old_root->dentry | |
1952 | && fs->pwd.mnt == old_root->mnt) | |
1953 | set_fs_pwd(fs, new_root); | |
1da177e4 LT |
1954 | put_fs_struct(fs); |
1955 | } else | |
1956 | task_unlock(p); | |
1957 | } while_each_thread(g, p); | |
1958 | read_unlock(&tasklist_lock); | |
1959 | } | |
1960 | ||
1961 | /* | |
1962 | * pivot_root Semantics: | |
1963 | * Moves the root file system of the current process to the directory put_old, | |
1964 | * makes new_root as the new root file system of the current process, and sets | |
1965 | * root/cwd of all processes which had them on the current root to new_root. | |
1966 | * | |
1967 | * Restrictions: | |
1968 | * The new_root and put_old must be directories, and must not be on the | |
1969 | * same file system as the current process root. The put_old must be | |
1970 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1971 | * pointed to by put_old must yield the same directory as new_root. No other | |
1972 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1973 | * | |
4a0d11fa NB |
1974 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
1975 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1976 | * in this situation. | |
1977 | * | |
1da177e4 LT |
1978 | * Notes: |
1979 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1980 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1981 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1982 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1983 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1984 | * first. | |
1985 | */ | |
b58fed8b RP |
1986 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1987 | const char __user * put_old) | |
1da177e4 LT |
1988 | { |
1989 | struct vfsmount *tmp; | |
1a390689 AV |
1990 | struct nameidata new_nd, old_nd, user_nd; |
1991 | struct path parent_path, root_parent; | |
1da177e4 LT |
1992 | int error; |
1993 | ||
1994 | if (!capable(CAP_SYS_ADMIN)) | |
1995 | return -EPERM; | |
1996 | ||
1997 | lock_kernel(); | |
1998 | ||
b58fed8b RP |
1999 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
2000 | &new_nd); | |
1da177e4 LT |
2001 | if (error) |
2002 | goto out0; | |
2003 | error = -EINVAL; | |
4ac91378 | 2004 | if (!check_mnt(new_nd.path.mnt)) |
1da177e4 LT |
2005 | goto out1; |
2006 | ||
b58fed8b | 2007 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
2008 | if (error) |
2009 | goto out1; | |
2010 | ||
b5266eb4 | 2011 | error = security_sb_pivotroot(&old_nd.path, &new_nd.path); |
1da177e4 | 2012 | if (error) { |
1d957f9b | 2013 | path_put(&old_nd.path); |
1da177e4 LT |
2014 | goto out1; |
2015 | } | |
2016 | ||
2017 | read_lock(¤t->fs->lock); | |
6ac08c39 JB |
2018 | user_nd.path = current->fs->root; |
2019 | path_get(¤t->fs->root); | |
1da177e4 | 2020 | read_unlock(¤t->fs->lock); |
390c6843 | 2021 | down_write(&namespace_sem); |
4ac91378 | 2022 | mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); |
1da177e4 | 2023 | error = -EINVAL; |
4ac91378 JB |
2024 | if (IS_MNT_SHARED(old_nd.path.mnt) || |
2025 | IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || | |
2026 | IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) | |
21444403 | 2027 | goto out2; |
4ac91378 | 2028 | if (!check_mnt(user_nd.path.mnt)) |
1da177e4 LT |
2029 | goto out2; |
2030 | error = -ENOENT; | |
4ac91378 | 2031 | if (IS_DEADDIR(new_nd.path.dentry->d_inode)) |
1da177e4 | 2032 | goto out2; |
4ac91378 | 2033 | if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) |
1da177e4 | 2034 | goto out2; |
4ac91378 | 2035 | if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) |
1da177e4 LT |
2036 | goto out2; |
2037 | error = -EBUSY; | |
4ac91378 JB |
2038 | if (new_nd.path.mnt == user_nd.path.mnt || |
2039 | old_nd.path.mnt == user_nd.path.mnt) | |
1da177e4 LT |
2040 | goto out2; /* loop, on the same file system */ |
2041 | error = -EINVAL; | |
4ac91378 | 2042 | if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) |
1da177e4 | 2043 | goto out2; /* not a mountpoint */ |
4ac91378 | 2044 | if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) |
0bb6fcc1 | 2045 | goto out2; /* not attached */ |
4ac91378 | 2046 | if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) |
1da177e4 | 2047 | goto out2; /* not a mountpoint */ |
4ac91378 | 2048 | if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) |
0bb6fcc1 | 2049 | goto out2; /* not attached */ |
4ac91378 JB |
2050 | /* make sure we can reach put_old from new_root */ |
2051 | tmp = old_nd.path.mnt; | |
1da177e4 | 2052 | spin_lock(&vfsmount_lock); |
4ac91378 | 2053 | if (tmp != new_nd.path.mnt) { |
1da177e4 LT |
2054 | for (;;) { |
2055 | if (tmp->mnt_parent == tmp) | |
2056 | goto out3; /* already mounted on put_old */ | |
4ac91378 | 2057 | if (tmp->mnt_parent == new_nd.path.mnt) |
1da177e4 LT |
2058 | break; |
2059 | tmp = tmp->mnt_parent; | |
2060 | } | |
4ac91378 | 2061 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) |
1da177e4 | 2062 | goto out3; |
4ac91378 | 2063 | } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) |
1da177e4 | 2064 | goto out3; |
1a390689 | 2065 | detach_mnt(new_nd.path.mnt, &parent_path); |
4ac91378 JB |
2066 | detach_mnt(user_nd.path.mnt, &root_parent); |
2067 | /* mount old root on put_old */ | |
1a390689 | 2068 | attach_mnt(user_nd.path.mnt, &old_nd.path); |
4ac91378 JB |
2069 | /* mount new_root on / */ |
2070 | attach_mnt(new_nd.path.mnt, &root_parent); | |
6b3286ed | 2071 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2072 | spin_unlock(&vfsmount_lock); |
1a390689 | 2073 | chroot_fs_refs(&user_nd.path, &new_nd.path); |
b5266eb4 | 2074 | security_sb_post_pivotroot(&user_nd.path, &new_nd.path); |
1da177e4 | 2075 | error = 0; |
1a390689 AV |
2076 | path_put(&root_parent); |
2077 | path_put(&parent_path); | |
1da177e4 | 2078 | out2: |
4ac91378 | 2079 | mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); |
390c6843 | 2080 | up_write(&namespace_sem); |
1d957f9b JB |
2081 | path_put(&user_nd.path); |
2082 | path_put(&old_nd.path); | |
1da177e4 | 2083 | out1: |
1d957f9b | 2084 | path_put(&new_nd.path); |
1da177e4 LT |
2085 | out0: |
2086 | unlock_kernel(); | |
2087 | return error; | |
2088 | out3: | |
2089 | spin_unlock(&vfsmount_lock); | |
2090 | goto out2; | |
2091 | } | |
2092 | ||
2093 | static void __init init_mount_tree(void) | |
2094 | { | |
2095 | struct vfsmount *mnt; | |
6b3286ed | 2096 | struct mnt_namespace *ns; |
ac748a09 | 2097 | struct path root; |
1da177e4 LT |
2098 | |
2099 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2100 | if (IS_ERR(mnt)) | |
2101 | panic("Can't create rootfs"); | |
6b3286ed KK |
2102 | ns = kmalloc(sizeof(*ns), GFP_KERNEL); |
2103 | if (!ns) | |
1da177e4 | 2104 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2105 | atomic_set(&ns->count, 1); |
2106 | INIT_LIST_HEAD(&ns->list); | |
2107 | init_waitqueue_head(&ns->poll); | |
2108 | ns->event = 0; | |
2109 | list_add(&mnt->mnt_list, &ns->list); | |
2110 | ns->root = mnt; | |
2111 | mnt->mnt_ns = ns; | |
2112 | ||
2113 | init_task.nsproxy->mnt_ns = ns; | |
2114 | get_mnt_ns(ns); | |
2115 | ||
ac748a09 JB |
2116 | root.mnt = ns->root; |
2117 | root.dentry = ns->root->mnt_root; | |
2118 | ||
2119 | set_fs_pwd(current->fs, &root); | |
2120 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2121 | } |
2122 | ||
74bf17cf | 2123 | void __init mnt_init(void) |
1da177e4 | 2124 | { |
13f14b4d | 2125 | unsigned u; |
15a67dd8 | 2126 | int err; |
1da177e4 | 2127 | |
390c6843 RP |
2128 | init_rwsem(&namespace_sem); |
2129 | ||
1da177e4 | 2130 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2131 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2132 | |
b58fed8b | 2133 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2134 | |
2135 | if (!mount_hashtable) | |
2136 | panic("Failed to allocate mount hash table\n"); | |
2137 | ||
13f14b4d ED |
2138 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2139 | ||
2140 | for (u = 0; u < HASH_SIZE; u++) | |
2141 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2142 | |
15a67dd8 RD |
2143 | err = sysfs_init(); |
2144 | if (err) | |
2145 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
2146 | __FUNCTION__, err); | |
00d26666 GKH |
2147 | fs_kobj = kobject_create_and_add("fs", NULL); |
2148 | if (!fs_kobj) | |
2149 | printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); | |
1da177e4 LT |
2150 | init_rootfs(); |
2151 | init_mount_tree(); | |
2152 | } | |
2153 | ||
6b3286ed | 2154 | void __put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2155 | { |
6b3286ed | 2156 | struct vfsmount *root = ns->root; |
70fbcdf4 | 2157 | LIST_HEAD(umount_list); |
6b3286ed | 2158 | ns->root = NULL; |
1ce88cf4 | 2159 | spin_unlock(&vfsmount_lock); |
390c6843 | 2160 | down_write(&namespace_sem); |
1da177e4 | 2161 | spin_lock(&vfsmount_lock); |
a05964f3 | 2162 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2163 | spin_unlock(&vfsmount_lock); |
390c6843 | 2164 | up_write(&namespace_sem); |
70fbcdf4 | 2165 | release_mounts(&umount_list); |
6b3286ed | 2166 | kfree(ns); |
1da177e4 | 2167 | } |