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