Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 LT |
11 | #include <linux/syscalls.h> |
12 | #include <linux/slab.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/smp_lock.h> | |
15 | #include <linux/init.h> | |
15a67dd8 | 16 | #include <linux/kernel.h> |
1da177e4 | 17 | #include <linux/acct.h> |
16f7e0fe | 18 | #include <linux/capability.h> |
3d733633 | 19 | #include <linux/cpumask.h> |
1da177e4 | 20 | #include <linux/module.h> |
f20a9ead | 21 | #include <linux/sysfs.h> |
1da177e4 | 22 | #include <linux/seq_file.h> |
6b3286ed | 23 | #include <linux/mnt_namespace.h> |
1da177e4 LT |
24 | #include <linux/namei.h> |
25 | #include <linux/security.h> | |
26 | #include <linux/mount.h> | |
07f3f05c | 27 | #include <linux/ramfs.h> |
13f14b4d | 28 | #include <linux/log2.h> |
73cd49ec | 29 | #include <linux/idr.h> |
5ad4e53b | 30 | #include <linux/fs_struct.h> |
1da177e4 LT |
31 | #include <asm/uaccess.h> |
32 | #include <asm/unistd.h> | |
07b20889 | 33 | #include "pnode.h" |
948730b0 | 34 | #include "internal.h" |
1da177e4 | 35 | |
13f14b4d ED |
36 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
37 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
38 | ||
1da177e4 | 39 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
40 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
41 | ||
42 | static int event; | |
73cd49ec | 43 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 44 | static DEFINE_IDA(mnt_group_ida); |
f21f6220 AV |
45 | static int mnt_id_start = 0; |
46 | static int mnt_group_start = 1; | |
1da177e4 | 47 | |
fa3536cc | 48 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 49 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 50 | static struct rw_semaphore namespace_sem; |
1da177e4 | 51 | |
f87fd4c2 | 52 | /* /sys/fs */ |
00d26666 GKH |
53 | struct kobject *fs_kobj; |
54 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 55 | |
1da177e4 LT |
56 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
57 | { | |
b58fed8b RP |
58 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
59 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
60 | tmp = tmp + (tmp >> HASH_SHIFT); |
61 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
62 | } |
63 | ||
3d733633 DH |
64 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
65 | ||
73cd49ec MS |
66 | /* allocation is serialized by namespace_sem */ |
67 | static int mnt_alloc_id(struct vfsmount *mnt) | |
68 | { | |
69 | int res; | |
70 | ||
71 | retry: | |
72 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
73 | spin_lock(&vfsmount_lock); | |
f21f6220 AV |
74 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
75 | if (!res) | |
76 | mnt_id_start = mnt->mnt_id + 1; | |
73cd49ec MS |
77 | spin_unlock(&vfsmount_lock); |
78 | if (res == -EAGAIN) | |
79 | goto retry; | |
80 | ||
81 | return res; | |
82 | } | |
83 | ||
84 | static void mnt_free_id(struct vfsmount *mnt) | |
85 | { | |
f21f6220 | 86 | int id = mnt->mnt_id; |
73cd49ec | 87 | spin_lock(&vfsmount_lock); |
f21f6220 AV |
88 | ida_remove(&mnt_id_ida, id); |
89 | if (mnt_id_start > id) | |
90 | mnt_id_start = id; | |
73cd49ec MS |
91 | spin_unlock(&vfsmount_lock); |
92 | } | |
93 | ||
719f5d7f MS |
94 | /* |
95 | * Allocate a new peer group ID | |
96 | * | |
97 | * mnt_group_ida is protected by namespace_sem | |
98 | */ | |
99 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
100 | { | |
f21f6220 AV |
101 | int res; |
102 | ||
719f5d7f MS |
103 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
104 | return -ENOMEM; | |
105 | ||
f21f6220 AV |
106 | res = ida_get_new_above(&mnt_group_ida, |
107 | mnt_group_start, | |
108 | &mnt->mnt_group_id); | |
109 | if (!res) | |
110 | mnt_group_start = mnt->mnt_group_id + 1; | |
111 | ||
112 | return res; | |
719f5d7f MS |
113 | } |
114 | ||
115 | /* | |
116 | * Release a peer group ID | |
117 | */ | |
118 | void mnt_release_group_id(struct vfsmount *mnt) | |
119 | { | |
f21f6220 AV |
120 | int id = mnt->mnt_group_id; |
121 | ida_remove(&mnt_group_ida, id); | |
122 | if (mnt_group_start > id) | |
123 | mnt_group_start = id; | |
719f5d7f MS |
124 | mnt->mnt_group_id = 0; |
125 | } | |
126 | ||
1da177e4 LT |
127 | struct vfsmount *alloc_vfsmnt(const char *name) |
128 | { | |
c3762229 | 129 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 130 | if (mnt) { |
73cd49ec MS |
131 | int err; |
132 | ||
133 | err = mnt_alloc_id(mnt); | |
88b38782 LZ |
134 | if (err) |
135 | goto out_free_cache; | |
136 | ||
137 | if (name) { | |
138 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
139 | if (!mnt->mnt_devname) | |
140 | goto out_free_id; | |
73cd49ec MS |
141 | } |
142 | ||
b58fed8b | 143 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
144 | INIT_LIST_HEAD(&mnt->mnt_hash); |
145 | INIT_LIST_HEAD(&mnt->mnt_child); | |
146 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
147 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 148 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 149 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
150 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
151 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
d3ef3d73 | 152 | #ifdef CONFIG_SMP |
153 | mnt->mnt_writers = alloc_percpu(int); | |
154 | if (!mnt->mnt_writers) | |
155 | goto out_free_devname; | |
156 | #else | |
157 | mnt->mnt_writers = 0; | |
158 | #endif | |
1da177e4 LT |
159 | } |
160 | return mnt; | |
88b38782 | 161 | |
d3ef3d73 | 162 | #ifdef CONFIG_SMP |
163 | out_free_devname: | |
164 | kfree(mnt->mnt_devname); | |
165 | #endif | |
88b38782 LZ |
166 | out_free_id: |
167 | mnt_free_id(mnt); | |
168 | out_free_cache: | |
169 | kmem_cache_free(mnt_cache, mnt); | |
170 | return NULL; | |
1da177e4 LT |
171 | } |
172 | ||
3d733633 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_is_readonly: check whether a mount is read-only | |
183 | * @mnt: the mount to check for its write status | |
184 | * | |
185 | * This shouldn't be used directly ouside of the VFS. | |
186 | * It does not guarantee that the filesystem will stay | |
187 | * r/w, just that it is right *now*. This can not and | |
188 | * should not be used in place of IS_RDONLY(inode). | |
189 | * mnt_want/drop_write() will _keep_ the filesystem | |
190 | * r/w. | |
191 | */ | |
192 | int __mnt_is_readonly(struct vfsmount *mnt) | |
193 | { | |
2e4b7fcd DH |
194 | if (mnt->mnt_flags & MNT_READONLY) |
195 | return 1; | |
196 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
197 | return 1; | |
198 | return 0; | |
3d733633 DH |
199 | } |
200 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
201 | ||
d3ef3d73 | 202 | static inline void inc_mnt_writers(struct vfsmount *mnt) |
203 | { | |
204 | #ifdef CONFIG_SMP | |
205 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++; | |
206 | #else | |
207 | mnt->mnt_writers++; | |
208 | #endif | |
209 | } | |
3d733633 | 210 | |
d3ef3d73 | 211 | static inline void dec_mnt_writers(struct vfsmount *mnt) |
3d733633 | 212 | { |
d3ef3d73 | 213 | #ifdef CONFIG_SMP |
214 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--; | |
215 | #else | |
216 | mnt->mnt_writers--; | |
217 | #endif | |
3d733633 | 218 | } |
3d733633 | 219 | |
d3ef3d73 | 220 | static unsigned int count_mnt_writers(struct vfsmount *mnt) |
3d733633 | 221 | { |
d3ef3d73 | 222 | #ifdef CONFIG_SMP |
223 | unsigned int count = 0; | |
3d733633 | 224 | int cpu; |
3d733633 DH |
225 | |
226 | for_each_possible_cpu(cpu) { | |
d3ef3d73 | 227 | count += *per_cpu_ptr(mnt->mnt_writers, cpu); |
3d733633 | 228 | } |
3d733633 | 229 | |
d3ef3d73 | 230 | return count; |
231 | #else | |
232 | return mnt->mnt_writers; | |
233 | #endif | |
3d733633 DH |
234 | } |
235 | ||
8366025e DH |
236 | /* |
237 | * Most r/o checks on a fs are for operations that take | |
238 | * discrete amounts of time, like a write() or unlink(). | |
239 | * We must keep track of when those operations start | |
240 | * (for permission checks) and when they end, so that | |
241 | * we can determine when writes are able to occur to | |
242 | * a filesystem. | |
243 | */ | |
244 | /** | |
245 | * mnt_want_write - get write access to a mount | |
246 | * @mnt: the mount on which to take a write | |
247 | * | |
248 | * This tells the low-level filesystem that a write is | |
249 | * about to be performed to it, and makes sure that | |
250 | * writes are allowed before returning success. When | |
251 | * the write operation is finished, mnt_drop_write() | |
252 | * must be called. This is effectively a refcount. | |
253 | */ | |
254 | int mnt_want_write(struct vfsmount *mnt) | |
255 | { | |
3d733633 | 256 | int ret = 0; |
3d733633 | 257 | |
d3ef3d73 | 258 | preempt_disable(); |
259 | inc_mnt_writers(mnt); | |
260 | /* | |
261 | * The store to inc_mnt_writers must be visible before we pass | |
262 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our | |
263 | * incremented count after it has set MNT_WRITE_HOLD. | |
264 | */ | |
265 | smp_mb(); | |
266 | while (mnt->mnt_flags & MNT_WRITE_HOLD) | |
267 | cpu_relax(); | |
268 | /* | |
269 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
270 | * be set to match its requirements. So we must not load that until | |
271 | * MNT_WRITE_HOLD is cleared. | |
272 | */ | |
273 | smp_rmb(); | |
3d733633 | 274 | if (__mnt_is_readonly(mnt)) { |
d3ef3d73 | 275 | dec_mnt_writers(mnt); |
3d733633 DH |
276 | ret = -EROFS; |
277 | goto out; | |
278 | } | |
3d733633 | 279 | out: |
d3ef3d73 | 280 | preempt_enable(); |
3d733633 | 281 | return ret; |
8366025e DH |
282 | } |
283 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
284 | ||
96029c4e | 285 | /** |
286 | * mnt_clone_write - get write access to a mount | |
287 | * @mnt: the mount on which to take a write | |
288 | * | |
289 | * This is effectively like mnt_want_write, except | |
290 | * it must only be used to take an extra write reference | |
291 | * on a mountpoint that we already know has a write reference | |
292 | * on it. This allows some optimisation. | |
293 | * | |
294 | * After finished, mnt_drop_write must be called as usual to | |
295 | * drop the reference. | |
296 | */ | |
297 | int mnt_clone_write(struct vfsmount *mnt) | |
298 | { | |
299 | /* superblock may be r/o */ | |
300 | if (__mnt_is_readonly(mnt)) | |
301 | return -EROFS; | |
302 | preempt_disable(); | |
303 | inc_mnt_writers(mnt); | |
304 | preempt_enable(); | |
305 | return 0; | |
306 | } | |
307 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
308 | ||
309 | /** | |
310 | * mnt_want_write_file - get write access to a file's mount | |
311 | * @file: the file who's mount on which to take a write | |
312 | * | |
313 | * This is like mnt_want_write, but it takes a file and can | |
314 | * do some optimisations if the file is open for write already | |
315 | */ | |
316 | int mnt_want_write_file(struct file *file) | |
317 | { | |
318 | if (!(file->f_mode & FMODE_WRITE)) | |
319 | return mnt_want_write(file->f_path.mnt); | |
320 | else | |
321 | return mnt_clone_write(file->f_path.mnt); | |
322 | } | |
323 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
324 | ||
8366025e DH |
325 | /** |
326 | * mnt_drop_write - give up write access to a mount | |
327 | * @mnt: the mount on which to give up write access | |
328 | * | |
329 | * Tells the low-level filesystem that we are done | |
330 | * performing writes to it. Must be matched with | |
331 | * mnt_want_write() call above. | |
332 | */ | |
333 | void mnt_drop_write(struct vfsmount *mnt) | |
334 | { | |
d3ef3d73 | 335 | preempt_disable(); |
336 | dec_mnt_writers(mnt); | |
337 | preempt_enable(); | |
8366025e DH |
338 | } |
339 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
340 | ||
2e4b7fcd | 341 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 342 | { |
3d733633 DH |
343 | int ret = 0; |
344 | ||
d3ef3d73 | 345 | spin_lock(&vfsmount_lock); |
346 | mnt->mnt_flags |= MNT_WRITE_HOLD; | |
3d733633 | 347 | /* |
d3ef3d73 | 348 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
349 | * should be visible before we do. | |
3d733633 | 350 | */ |
d3ef3d73 | 351 | smp_mb(); |
352 | ||
3d733633 | 353 | /* |
d3ef3d73 | 354 | * With writers on hold, if this value is zero, then there are |
355 | * definitely no active writers (although held writers may subsequently | |
356 | * increment the count, they'll have to wait, and decrement it after | |
357 | * seeing MNT_READONLY). | |
358 | * | |
359 | * It is OK to have counter incremented on one CPU and decremented on | |
360 | * another: the sum will add up correctly. The danger would be when we | |
361 | * sum up each counter, if we read a counter before it is incremented, | |
362 | * but then read another CPU's count which it has been subsequently | |
363 | * decremented from -- we would see more decrements than we should. | |
364 | * MNT_WRITE_HOLD protects against this scenario, because | |
365 | * mnt_want_write first increments count, then smp_mb, then spins on | |
366 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
367 | * we're counting up here. | |
3d733633 | 368 | */ |
d3ef3d73 | 369 | if (count_mnt_writers(mnt) > 0) |
370 | ret = -EBUSY; | |
371 | else | |
2e4b7fcd | 372 | mnt->mnt_flags |= MNT_READONLY; |
d3ef3d73 | 373 | /* |
374 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
375 | * that become unheld will see MNT_READONLY. | |
376 | */ | |
377 | smp_wmb(); | |
378 | mnt->mnt_flags &= ~MNT_WRITE_HOLD; | |
2e4b7fcd | 379 | spin_unlock(&vfsmount_lock); |
3d733633 | 380 | return ret; |
8366025e | 381 | } |
8366025e | 382 | |
2e4b7fcd DH |
383 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
384 | { | |
385 | spin_lock(&vfsmount_lock); | |
386 | mnt->mnt_flags &= ~MNT_READONLY; | |
387 | spin_unlock(&vfsmount_lock); | |
388 | } | |
389 | ||
a3ec947c | 390 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
454e2398 DH |
391 | { |
392 | mnt->mnt_sb = sb; | |
393 | mnt->mnt_root = dget(sb->s_root); | |
454e2398 DH |
394 | } |
395 | ||
396 | EXPORT_SYMBOL(simple_set_mnt); | |
397 | ||
1da177e4 LT |
398 | void free_vfsmnt(struct vfsmount *mnt) |
399 | { | |
400 | kfree(mnt->mnt_devname); | |
73cd49ec | 401 | mnt_free_id(mnt); |
d3ef3d73 | 402 | #ifdef CONFIG_SMP |
403 | free_percpu(mnt->mnt_writers); | |
404 | #endif | |
1da177e4 LT |
405 | kmem_cache_free(mnt_cache, mnt); |
406 | } | |
407 | ||
408 | /* | |
a05964f3 RP |
409 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
410 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 411 | */ |
a05964f3 RP |
412 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
413 | int dir) | |
1da177e4 | 414 | { |
b58fed8b RP |
415 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
416 | struct list_head *tmp = head; | |
1da177e4 LT |
417 | struct vfsmount *p, *found = NULL; |
418 | ||
1da177e4 | 419 | for (;;) { |
a05964f3 | 420 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
421 | p = NULL; |
422 | if (tmp == head) | |
423 | break; | |
424 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
425 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 426 | found = p; |
1da177e4 LT |
427 | break; |
428 | } | |
429 | } | |
1da177e4 LT |
430 | return found; |
431 | } | |
432 | ||
a05964f3 RP |
433 | /* |
434 | * lookup_mnt increments the ref count before returning | |
435 | * the vfsmount struct. | |
436 | */ | |
1c755af4 | 437 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 RP |
438 | { |
439 | struct vfsmount *child_mnt; | |
440 | spin_lock(&vfsmount_lock); | |
1c755af4 | 441 | if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) |
a05964f3 RP |
442 | mntget(child_mnt); |
443 | spin_unlock(&vfsmount_lock); | |
444 | return child_mnt; | |
445 | } | |
446 | ||
1da177e4 LT |
447 | static inline int check_mnt(struct vfsmount *mnt) |
448 | { | |
6b3286ed | 449 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
450 | } |
451 | ||
6b3286ed | 452 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
453 | { |
454 | if (ns) { | |
455 | ns->event = ++event; | |
456 | wake_up_interruptible(&ns->poll); | |
457 | } | |
458 | } | |
459 | ||
6b3286ed | 460 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
461 | { |
462 | if (ns && ns->event != event) { | |
463 | ns->event = event; | |
464 | wake_up_interruptible(&ns->poll); | |
465 | } | |
466 | } | |
467 | ||
1a390689 | 468 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 469 | { |
1a390689 AV |
470 | old_path->dentry = mnt->mnt_mountpoint; |
471 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
472 | mnt->mnt_parent = mnt; |
473 | mnt->mnt_mountpoint = mnt->mnt_root; | |
474 | list_del_init(&mnt->mnt_child); | |
475 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 476 | old_path->dentry->d_mounted--; |
1da177e4 LT |
477 | } |
478 | ||
b90fa9ae RP |
479 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
480 | struct vfsmount *child_mnt) | |
481 | { | |
482 | child_mnt->mnt_parent = mntget(mnt); | |
483 | child_mnt->mnt_mountpoint = dget(dentry); | |
484 | dentry->d_mounted++; | |
485 | } | |
486 | ||
1a390689 | 487 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 488 | { |
1a390689 | 489 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 490 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
491 | hash(path->mnt, path->dentry)); |
492 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
493 | } |
494 | ||
495 | /* | |
496 | * the caller must hold vfsmount_lock | |
497 | */ | |
498 | static void commit_tree(struct vfsmount *mnt) | |
499 | { | |
500 | struct vfsmount *parent = mnt->mnt_parent; | |
501 | struct vfsmount *m; | |
502 | LIST_HEAD(head); | |
6b3286ed | 503 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
504 | |
505 | BUG_ON(parent == mnt); | |
506 | ||
507 | list_add_tail(&head, &mnt->mnt_list); | |
508 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 509 | m->mnt_ns = n; |
b90fa9ae RP |
510 | list_splice(&head, n->list.prev); |
511 | ||
512 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
513 | hash(parent, mnt->mnt_mountpoint)); | |
514 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 515 | touch_mnt_namespace(n); |
1da177e4 LT |
516 | } |
517 | ||
518 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
519 | { | |
520 | struct list_head *next = p->mnt_mounts.next; | |
521 | if (next == &p->mnt_mounts) { | |
522 | while (1) { | |
523 | if (p == root) | |
524 | return NULL; | |
525 | next = p->mnt_child.next; | |
526 | if (next != &p->mnt_parent->mnt_mounts) | |
527 | break; | |
528 | p = p->mnt_parent; | |
529 | } | |
530 | } | |
531 | return list_entry(next, struct vfsmount, mnt_child); | |
532 | } | |
533 | ||
9676f0c6 RP |
534 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
535 | { | |
536 | struct list_head *prev = p->mnt_mounts.prev; | |
537 | while (prev != &p->mnt_mounts) { | |
538 | p = list_entry(prev, struct vfsmount, mnt_child); | |
539 | prev = p->mnt_mounts.prev; | |
540 | } | |
541 | return p; | |
542 | } | |
543 | ||
36341f64 RP |
544 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
545 | int flag) | |
1da177e4 LT |
546 | { |
547 | struct super_block *sb = old->mnt_sb; | |
548 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
549 | ||
550 | if (mnt) { | |
719f5d7f MS |
551 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
552 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
553 | else | |
554 | mnt->mnt_group_id = old->mnt_group_id; | |
555 | ||
556 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
557 | int err = mnt_alloc_group_id(mnt); | |
558 | if (err) | |
559 | goto out_free; | |
560 | } | |
561 | ||
1da177e4 LT |
562 | mnt->mnt_flags = old->mnt_flags; |
563 | atomic_inc(&sb->s_active); | |
564 | mnt->mnt_sb = sb; | |
565 | mnt->mnt_root = dget(root); | |
566 | mnt->mnt_mountpoint = mnt->mnt_root; | |
567 | mnt->mnt_parent = mnt; | |
b90fa9ae | 568 | |
5afe0022 RP |
569 | if (flag & CL_SLAVE) { |
570 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
571 | mnt->mnt_master = old; | |
572 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 573 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
574 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
575 | list_add(&mnt->mnt_share, &old->mnt_share); | |
576 | if (IS_MNT_SLAVE(old)) | |
577 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
578 | mnt->mnt_master = old->mnt_master; | |
579 | } | |
b90fa9ae RP |
580 | if (flag & CL_MAKE_SHARED) |
581 | set_mnt_shared(mnt); | |
1da177e4 LT |
582 | |
583 | /* stick the duplicate mount on the same expiry list | |
584 | * as the original if that was on one */ | |
36341f64 | 585 | if (flag & CL_EXPIRE) { |
36341f64 RP |
586 | if (!list_empty(&old->mnt_expire)) |
587 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 588 | } |
1da177e4 LT |
589 | } |
590 | return mnt; | |
719f5d7f MS |
591 | |
592 | out_free: | |
593 | free_vfsmnt(mnt); | |
594 | return NULL; | |
1da177e4 LT |
595 | } |
596 | ||
7b7b1ace | 597 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
598 | { |
599 | struct super_block *sb = mnt->mnt_sb; | |
3d733633 DH |
600 | /* |
601 | * This probably indicates that somebody messed | |
602 | * up a mnt_want/drop_write() pair. If this | |
603 | * happens, the filesystem was probably unable | |
604 | * to make r/w->r/o transitions. | |
605 | */ | |
d3ef3d73 | 606 | /* |
607 | * atomic_dec_and_lock() used to deal with ->mnt_count decrements | |
608 | * provides barriers, so count_mnt_writers() below is safe. AV | |
609 | */ | |
610 | WARN_ON(count_mnt_writers(mnt)); | |
1da177e4 LT |
611 | dput(mnt->mnt_root); |
612 | free_vfsmnt(mnt); | |
613 | deactivate_super(sb); | |
614 | } | |
615 | ||
7b7b1ace AV |
616 | void mntput_no_expire(struct vfsmount *mnt) |
617 | { | |
618 | repeat: | |
619 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
620 | if (likely(!mnt->mnt_pinned)) { | |
621 | spin_unlock(&vfsmount_lock); | |
622 | __mntput(mnt); | |
623 | return; | |
624 | } | |
625 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
626 | mnt->mnt_pinned = 0; | |
627 | spin_unlock(&vfsmount_lock); | |
628 | acct_auto_close_mnt(mnt); | |
629 | security_sb_umount_close(mnt); | |
630 | goto repeat; | |
631 | } | |
632 | } | |
633 | ||
634 | EXPORT_SYMBOL(mntput_no_expire); | |
635 | ||
636 | void mnt_pin(struct vfsmount *mnt) | |
637 | { | |
638 | spin_lock(&vfsmount_lock); | |
639 | mnt->mnt_pinned++; | |
640 | spin_unlock(&vfsmount_lock); | |
641 | } | |
642 | ||
643 | EXPORT_SYMBOL(mnt_pin); | |
644 | ||
645 | void mnt_unpin(struct vfsmount *mnt) | |
646 | { | |
647 | spin_lock(&vfsmount_lock); | |
648 | if (mnt->mnt_pinned) { | |
649 | atomic_inc(&mnt->mnt_count); | |
650 | mnt->mnt_pinned--; | |
651 | } | |
652 | spin_unlock(&vfsmount_lock); | |
653 | } | |
654 | ||
655 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 656 | |
b3b304a2 MS |
657 | static inline void mangle(struct seq_file *m, const char *s) |
658 | { | |
659 | seq_escape(m, s, " \t\n\\"); | |
660 | } | |
661 | ||
662 | /* | |
663 | * Simple .show_options callback for filesystems which don't want to | |
664 | * implement more complex mount option showing. | |
665 | * | |
666 | * See also save_mount_options(). | |
667 | */ | |
668 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
669 | { | |
2a32cebd AV |
670 | const char *options; |
671 | ||
672 | rcu_read_lock(); | |
673 | options = rcu_dereference(mnt->mnt_sb->s_options); | |
b3b304a2 MS |
674 | |
675 | if (options != NULL && options[0]) { | |
676 | seq_putc(m, ','); | |
677 | mangle(m, options); | |
678 | } | |
2a32cebd | 679 | rcu_read_unlock(); |
b3b304a2 MS |
680 | |
681 | return 0; | |
682 | } | |
683 | EXPORT_SYMBOL(generic_show_options); | |
684 | ||
685 | /* | |
686 | * If filesystem uses generic_show_options(), this function should be | |
687 | * called from the fill_super() callback. | |
688 | * | |
689 | * The .remount_fs callback usually needs to be handled in a special | |
690 | * way, to make sure, that previous options are not overwritten if the | |
691 | * remount fails. | |
692 | * | |
693 | * Also note, that if the filesystem's .remount_fs function doesn't | |
694 | * reset all options to their default value, but changes only newly | |
695 | * given options, then the displayed options will not reflect reality | |
696 | * any more. | |
697 | */ | |
698 | void save_mount_options(struct super_block *sb, char *options) | |
699 | { | |
2a32cebd AV |
700 | BUG_ON(sb->s_options); |
701 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
702 | } |
703 | EXPORT_SYMBOL(save_mount_options); | |
704 | ||
2a32cebd AV |
705 | void replace_mount_options(struct super_block *sb, char *options) |
706 | { | |
707 | char *old = sb->s_options; | |
708 | rcu_assign_pointer(sb->s_options, options); | |
709 | if (old) { | |
710 | synchronize_rcu(); | |
711 | kfree(old); | |
712 | } | |
713 | } | |
714 | EXPORT_SYMBOL(replace_mount_options); | |
715 | ||
a1a2c409 | 716 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
717 | /* iterator */ |
718 | static void *m_start(struct seq_file *m, loff_t *pos) | |
719 | { | |
a1a2c409 | 720 | struct proc_mounts *p = m->private; |
1da177e4 | 721 | |
390c6843 | 722 | down_read(&namespace_sem); |
a1a2c409 | 723 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
724 | } |
725 | ||
726 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
727 | { | |
a1a2c409 | 728 | struct proc_mounts *p = m->private; |
b0765fb8 | 729 | |
a1a2c409 | 730 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
731 | } |
732 | ||
733 | static void m_stop(struct seq_file *m, void *v) | |
734 | { | |
390c6843 | 735 | up_read(&namespace_sem); |
1da177e4 LT |
736 | } |
737 | ||
2d4d4864 RP |
738 | struct proc_fs_info { |
739 | int flag; | |
740 | const char *str; | |
741 | }; | |
742 | ||
2069f457 | 743 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) |
1da177e4 | 744 | { |
2d4d4864 | 745 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
746 | { MS_SYNCHRONOUS, ",sync" }, |
747 | { MS_DIRSYNC, ",dirsync" }, | |
748 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
749 | { 0, NULL } |
750 | }; | |
2d4d4864 RP |
751 | const struct proc_fs_info *fs_infop; |
752 | ||
753 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
754 | if (sb->s_flags & fs_infop->flag) | |
755 | seq_puts(m, fs_infop->str); | |
756 | } | |
2069f457 EP |
757 | |
758 | return security_sb_show_options(m, sb); | |
2d4d4864 RP |
759 | } |
760 | ||
761 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
762 | { | |
763 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
764 | { MNT_NOSUID, ",nosuid" }, |
765 | { MNT_NODEV, ",nodev" }, | |
766 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
767 | { MNT_NOATIME, ",noatime" }, |
768 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 769 | { MNT_RELATIME, ",relatime" }, |
d0adde57 | 770 | { MNT_STRICTATIME, ",strictatime" }, |
1da177e4 LT |
771 | { 0, NULL } |
772 | }; | |
2d4d4864 RP |
773 | const struct proc_fs_info *fs_infop; |
774 | ||
775 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
776 | if (mnt->mnt_flags & fs_infop->flag) | |
777 | seq_puts(m, fs_infop->str); | |
778 | } | |
779 | } | |
780 | ||
781 | static void show_type(struct seq_file *m, struct super_block *sb) | |
782 | { | |
783 | mangle(m, sb->s_type->name); | |
784 | if (sb->s_subtype && sb->s_subtype[0]) { | |
785 | seq_putc(m, '.'); | |
786 | mangle(m, sb->s_subtype); | |
787 | } | |
788 | } | |
789 | ||
790 | static int show_vfsmnt(struct seq_file *m, void *v) | |
791 | { | |
792 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
793 | int err = 0; | |
c32c2f63 | 794 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
795 | |
796 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
797 | seq_putc(m, ' '); | |
c32c2f63 | 798 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 799 | seq_putc(m, ' '); |
2d4d4864 | 800 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 801 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2069f457 EP |
802 | err = show_sb_opts(m, mnt->mnt_sb); |
803 | if (err) | |
804 | goto out; | |
2d4d4864 | 805 | show_mnt_opts(m, mnt); |
1da177e4 LT |
806 | if (mnt->mnt_sb->s_op->show_options) |
807 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
808 | seq_puts(m, " 0 0\n"); | |
2069f457 | 809 | out: |
1da177e4 LT |
810 | return err; |
811 | } | |
812 | ||
a1a2c409 | 813 | const struct seq_operations mounts_op = { |
1da177e4 LT |
814 | .start = m_start, |
815 | .next = m_next, | |
816 | .stop = m_stop, | |
817 | .show = show_vfsmnt | |
818 | }; | |
819 | ||
2d4d4864 RP |
820 | static int show_mountinfo(struct seq_file *m, void *v) |
821 | { | |
822 | struct proc_mounts *p = m->private; | |
823 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
824 | struct super_block *sb = mnt->mnt_sb; | |
825 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
826 | struct path root = p->root; | |
827 | int err = 0; | |
828 | ||
829 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
830 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
831 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
832 | seq_putc(m, ' '); | |
833 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
834 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
835 | /* | |
836 | * Mountpoint is outside root, discard that one. Ugly, | |
837 | * but less so than trying to do that in iterator in a | |
838 | * race-free way (due to renames). | |
839 | */ | |
840 | return SEQ_SKIP; | |
841 | } | |
842 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
843 | show_mnt_opts(m, mnt); | |
844 | ||
845 | /* Tagged fields ("foo:X" or "bar") */ | |
846 | if (IS_MNT_SHARED(mnt)) | |
847 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
848 | if (IS_MNT_SLAVE(mnt)) { |
849 | int master = mnt->mnt_master->mnt_group_id; | |
850 | int dom = get_dominating_id(mnt, &p->root); | |
851 | seq_printf(m, " master:%i", master); | |
852 | if (dom && dom != master) | |
853 | seq_printf(m, " propagate_from:%i", dom); | |
854 | } | |
2d4d4864 RP |
855 | if (IS_MNT_UNBINDABLE(mnt)) |
856 | seq_puts(m, " unbindable"); | |
857 | ||
858 | /* Filesystem specific data */ | |
859 | seq_puts(m, " - "); | |
860 | show_type(m, sb); | |
861 | seq_putc(m, ' '); | |
862 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
863 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
2069f457 EP |
864 | err = show_sb_opts(m, sb); |
865 | if (err) | |
866 | goto out; | |
2d4d4864 RP |
867 | if (sb->s_op->show_options) |
868 | err = sb->s_op->show_options(m, mnt); | |
869 | seq_putc(m, '\n'); | |
2069f457 | 870 | out: |
2d4d4864 RP |
871 | return err; |
872 | } | |
873 | ||
874 | const struct seq_operations mountinfo_op = { | |
875 | .start = m_start, | |
876 | .next = m_next, | |
877 | .stop = m_stop, | |
878 | .show = show_mountinfo, | |
879 | }; | |
880 | ||
b4629fe2 CL |
881 | static int show_vfsstat(struct seq_file *m, void *v) |
882 | { | |
b0765fb8 | 883 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 884 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
885 | int err = 0; |
886 | ||
887 | /* device */ | |
888 | if (mnt->mnt_devname) { | |
889 | seq_puts(m, "device "); | |
890 | mangle(m, mnt->mnt_devname); | |
891 | } else | |
892 | seq_puts(m, "no device"); | |
893 | ||
894 | /* mount point */ | |
895 | seq_puts(m, " mounted on "); | |
c32c2f63 | 896 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
897 | seq_putc(m, ' '); |
898 | ||
899 | /* file system type */ | |
900 | seq_puts(m, "with fstype "); | |
2d4d4864 | 901 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
902 | |
903 | /* optional statistics */ | |
904 | if (mnt->mnt_sb->s_op->show_stats) { | |
905 | seq_putc(m, ' '); | |
906 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
907 | } | |
908 | ||
909 | seq_putc(m, '\n'); | |
910 | return err; | |
911 | } | |
912 | ||
a1a2c409 | 913 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
914 | .start = m_start, |
915 | .next = m_next, | |
916 | .stop = m_stop, | |
917 | .show = show_vfsstat, | |
918 | }; | |
a1a2c409 | 919 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 920 | |
1da177e4 LT |
921 | /** |
922 | * may_umount_tree - check if a mount tree is busy | |
923 | * @mnt: root of mount tree | |
924 | * | |
925 | * This is called to check if a tree of mounts has any | |
926 | * open files, pwds, chroots or sub mounts that are | |
927 | * busy. | |
928 | */ | |
929 | int may_umount_tree(struct vfsmount *mnt) | |
930 | { | |
36341f64 RP |
931 | int actual_refs = 0; |
932 | int minimum_refs = 0; | |
933 | struct vfsmount *p; | |
1da177e4 LT |
934 | |
935 | spin_lock(&vfsmount_lock); | |
36341f64 | 936 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
937 | actual_refs += atomic_read(&p->mnt_count); |
938 | minimum_refs += 2; | |
1da177e4 LT |
939 | } |
940 | spin_unlock(&vfsmount_lock); | |
941 | ||
942 | if (actual_refs > minimum_refs) | |
e3474a8e | 943 | return 0; |
1da177e4 | 944 | |
e3474a8e | 945 | return 1; |
1da177e4 LT |
946 | } |
947 | ||
948 | EXPORT_SYMBOL(may_umount_tree); | |
949 | ||
950 | /** | |
951 | * may_umount - check if a mount point is busy | |
952 | * @mnt: root of mount | |
953 | * | |
954 | * This is called to check if a mount point has any | |
955 | * open files, pwds, chroots or sub mounts. If the | |
956 | * mount has sub mounts this will return busy | |
957 | * regardless of whether the sub mounts are busy. | |
958 | * | |
959 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
960 | * give false negatives. The main reason why it's here is that we need | |
961 | * a non-destructive way to look for easily umountable filesystems. | |
962 | */ | |
963 | int may_umount(struct vfsmount *mnt) | |
964 | { | |
e3474a8e | 965 | int ret = 1; |
a05964f3 RP |
966 | spin_lock(&vfsmount_lock); |
967 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 968 | ret = 0; |
a05964f3 RP |
969 | spin_unlock(&vfsmount_lock); |
970 | return ret; | |
1da177e4 LT |
971 | } |
972 | ||
973 | EXPORT_SYMBOL(may_umount); | |
974 | ||
b90fa9ae | 975 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
976 | { |
977 | struct vfsmount *mnt; | |
bf066c7d | 978 | while (!list_empty(head)) { |
b5e61818 | 979 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
980 | list_del_init(&mnt->mnt_hash); |
981 | if (mnt->mnt_parent != mnt) { | |
982 | struct dentry *dentry; | |
983 | struct vfsmount *m; | |
984 | spin_lock(&vfsmount_lock); | |
985 | dentry = mnt->mnt_mountpoint; | |
986 | m = mnt->mnt_parent; | |
987 | mnt->mnt_mountpoint = mnt->mnt_root; | |
988 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 989 | m->mnt_ghosts--; |
70fbcdf4 RP |
990 | spin_unlock(&vfsmount_lock); |
991 | dput(dentry); | |
992 | mntput(m); | |
993 | } | |
994 | mntput(mnt); | |
995 | } | |
996 | } | |
997 | ||
a05964f3 | 998 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
999 | { |
1000 | struct vfsmount *p; | |
1da177e4 | 1001 | |
1bfba4e8 AM |
1002 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1003 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 1004 | |
a05964f3 RP |
1005 | if (propagate) |
1006 | propagate_umount(kill); | |
1007 | ||
70fbcdf4 RP |
1008 | list_for_each_entry(p, kill, mnt_hash) { |
1009 | list_del_init(&p->mnt_expire); | |
1010 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
1011 | __touch_mnt_namespace(p->mnt_ns); |
1012 | p->mnt_ns = NULL; | |
70fbcdf4 | 1013 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
1014 | if (p->mnt_parent != p) { |
1015 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 1016 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 1017 | } |
a05964f3 | 1018 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1019 | } |
1020 | } | |
1021 | ||
c35038be AV |
1022 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
1023 | ||
1da177e4 LT |
1024 | static int do_umount(struct vfsmount *mnt, int flags) |
1025 | { | |
b58fed8b | 1026 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 1027 | int retval; |
70fbcdf4 | 1028 | LIST_HEAD(umount_list); |
1da177e4 LT |
1029 | |
1030 | retval = security_sb_umount(mnt, flags); | |
1031 | if (retval) | |
1032 | return retval; | |
1033 | ||
1034 | /* | |
1035 | * Allow userspace to request a mountpoint be expired rather than | |
1036 | * unmounting unconditionally. Unmount only happens if: | |
1037 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1038 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1039 | */ | |
1040 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 1041 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
1042 | flags & (MNT_FORCE | MNT_DETACH)) |
1043 | return -EINVAL; | |
1044 | ||
1045 | if (atomic_read(&mnt->mnt_count) != 2) | |
1046 | return -EBUSY; | |
1047 | ||
1048 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1049 | return -EAGAIN; | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * If we may have to abort operations to get out of this | |
1054 | * mount, and they will themselves hold resources we must | |
1055 | * allow the fs to do things. In the Unix tradition of | |
1056 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1057 | * might fail to complete on the first run through as other tasks | |
1058 | * must return, and the like. Thats for the mount program to worry | |
1059 | * about for the moment. | |
1060 | */ | |
1061 | ||
42faad99 | 1062 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1063 | sb->s_op->umount_begin(sb); |
42faad99 | 1064 | } |
1da177e4 LT |
1065 | |
1066 | /* | |
1067 | * No sense to grab the lock for this test, but test itself looks | |
1068 | * somewhat bogus. Suggestions for better replacement? | |
1069 | * Ho-hum... In principle, we might treat that as umount + switch | |
1070 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1071 | * Actually it makes sense, especially if rootfs would contain a | |
1072 | * /reboot - static binary that would close all descriptors and | |
1073 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1074 | */ | |
6ac08c39 | 1075 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1076 | /* |
1077 | * Special case for "unmounting" root ... | |
1078 | * we just try to remount it readonly. | |
1079 | */ | |
1080 | down_write(&sb->s_umount); | |
4aa98cf7 | 1081 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1082 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1083 | up_write(&sb->s_umount); |
1084 | return retval; | |
1085 | } | |
1086 | ||
390c6843 | 1087 | down_write(&namespace_sem); |
1da177e4 | 1088 | spin_lock(&vfsmount_lock); |
5addc5dd | 1089 | event++; |
1da177e4 | 1090 | |
c35038be AV |
1091 | if (!(flags & MNT_DETACH)) |
1092 | shrink_submounts(mnt, &umount_list); | |
1093 | ||
1da177e4 | 1094 | retval = -EBUSY; |
a05964f3 | 1095 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1096 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1097 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1098 | retval = 0; |
1099 | } | |
1100 | spin_unlock(&vfsmount_lock); | |
1101 | if (retval) | |
1102 | security_sb_umount_busy(mnt); | |
390c6843 | 1103 | up_write(&namespace_sem); |
70fbcdf4 | 1104 | release_mounts(&umount_list); |
1da177e4 LT |
1105 | return retval; |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Now umount can handle mount points as well as block devices. | |
1110 | * This is important for filesystems which use unnamed block devices. | |
1111 | * | |
1112 | * We now support a flag for forced unmount like the other 'big iron' | |
1113 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1114 | */ | |
1115 | ||
bdc480e3 | 1116 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1117 | { |
2d8f3038 | 1118 | struct path path; |
1da177e4 LT |
1119 | int retval; |
1120 | ||
2d8f3038 | 1121 | retval = user_path(name, &path); |
1da177e4 LT |
1122 | if (retval) |
1123 | goto out; | |
1124 | retval = -EINVAL; | |
2d8f3038 | 1125 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1126 | goto dput_and_out; |
2d8f3038 | 1127 | if (!check_mnt(path.mnt)) |
1da177e4 LT |
1128 | goto dput_and_out; |
1129 | ||
1130 | retval = -EPERM; | |
1131 | if (!capable(CAP_SYS_ADMIN)) | |
1132 | goto dput_and_out; | |
1133 | ||
2d8f3038 | 1134 | retval = do_umount(path.mnt, flags); |
1da177e4 | 1135 | dput_and_out: |
429731b1 | 1136 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 AV |
1137 | dput(path.dentry); |
1138 | mntput_no_expire(path.mnt); | |
1da177e4 LT |
1139 | out: |
1140 | return retval; | |
1141 | } | |
1142 | ||
1143 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1144 | ||
1145 | /* | |
b58fed8b | 1146 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1147 | */ |
bdc480e3 | 1148 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1149 | { |
b58fed8b | 1150 | return sys_umount(name, 0); |
1da177e4 LT |
1151 | } |
1152 | ||
1153 | #endif | |
1154 | ||
2d92ab3c | 1155 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1156 | { |
1157 | if (capable(CAP_SYS_ADMIN)) | |
1158 | return 0; | |
1159 | return -EPERM; | |
1160 | #ifdef notyet | |
2d92ab3c | 1161 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1162 | return -EPERM; |
2d92ab3c | 1163 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1164 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1165 | return -EPERM; |
1166 | } | |
2d92ab3c | 1167 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1168 | return -EPERM; |
1169 | return 0; | |
1170 | #endif | |
1171 | } | |
1172 | ||
b90fa9ae | 1173 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1174 | int flag) |
1da177e4 LT |
1175 | { |
1176 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1177 | struct path path; |
1da177e4 | 1178 | |
9676f0c6 RP |
1179 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1180 | return NULL; | |
1181 | ||
36341f64 | 1182 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1183 | if (!q) |
1184 | goto Enomem; | |
1185 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1186 | ||
1187 | p = mnt; | |
fdadd65f | 1188 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
7ec02ef1 | 1189 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1190 | continue; |
1191 | ||
1192 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1193 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1194 | s = skip_mnt_tree(s); | |
1195 | continue; | |
1196 | } | |
1da177e4 LT |
1197 | while (p != s->mnt_parent) { |
1198 | p = p->mnt_parent; | |
1199 | q = q->mnt_parent; | |
1200 | } | |
1201 | p = s; | |
1a390689 AV |
1202 | path.mnt = q; |
1203 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1204 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1205 | if (!q) |
1206 | goto Enomem; | |
1207 | spin_lock(&vfsmount_lock); | |
1208 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1209 | attach_mnt(q, &path); |
1da177e4 LT |
1210 | spin_unlock(&vfsmount_lock); |
1211 | } | |
1212 | } | |
1213 | return res; | |
b58fed8b | 1214 | Enomem: |
1da177e4 | 1215 | if (res) { |
70fbcdf4 | 1216 | LIST_HEAD(umount_list); |
1da177e4 | 1217 | spin_lock(&vfsmount_lock); |
a05964f3 | 1218 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1219 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1220 | release_mounts(&umount_list); |
1da177e4 LT |
1221 | } |
1222 | return NULL; | |
1223 | } | |
1224 | ||
589ff870 | 1225 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 AV |
1226 | { |
1227 | struct vfsmount *tree; | |
1a60a280 | 1228 | down_write(&namespace_sem); |
589ff870 | 1229 | tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1230 | up_write(&namespace_sem); |
8aec0809 AV |
1231 | return tree; |
1232 | } | |
1233 | ||
1234 | void drop_collected_mounts(struct vfsmount *mnt) | |
1235 | { | |
1236 | LIST_HEAD(umount_list); | |
1a60a280 | 1237 | down_write(&namespace_sem); |
8aec0809 AV |
1238 | spin_lock(&vfsmount_lock); |
1239 | umount_tree(mnt, 0, &umount_list); | |
1240 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1241 | up_write(&namespace_sem); |
8aec0809 AV |
1242 | release_mounts(&umount_list); |
1243 | } | |
1244 | ||
719f5d7f MS |
1245 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1246 | { | |
1247 | struct vfsmount *p; | |
1248 | ||
1249 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1250 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1251 | mnt_release_group_id(p); | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1256 | { | |
1257 | struct vfsmount *p; | |
1258 | ||
1259 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1260 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1261 | int err = mnt_alloc_group_id(p); | |
1262 | if (err) { | |
1263 | cleanup_group_ids(mnt, p); | |
1264 | return err; | |
1265 | } | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | return 0; | |
1270 | } | |
1271 | ||
b90fa9ae RP |
1272 | /* |
1273 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1274 | * @nd : place the mount tree @source_mnt is attached |
1275 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1276 | * store the parent mount and mountpoint dentry. | |
1277 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1278 | * |
1279 | * NOTE: in the table below explains the semantics when a source mount | |
1280 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1281 | * --------------------------------------------------------------------------- |
1282 | * | BIND MOUNT OPERATION | | |
1283 | * |************************************************************************** | |
1284 | * | source-->| shared | private | slave | unbindable | | |
1285 | * | dest | | | | | | |
1286 | * | | | | | | | | |
1287 | * | v | | | | | | |
1288 | * |************************************************************************** | |
1289 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1290 | * | | | | | | | |
1291 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1292 | * *************************************************************************** | |
b90fa9ae RP |
1293 | * A bind operation clones the source mount and mounts the clone on the |
1294 | * destination mount. | |
1295 | * | |
1296 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1297 | * tree of the destination mount and the cloned mount is added to | |
1298 | * the peer group of the source mount. | |
1299 | * (+) the cloned mount is created under the destination mount and is marked | |
1300 | * as shared. The cloned mount is added to the peer group of the source | |
1301 | * mount. | |
5afe0022 RP |
1302 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1303 | * of the destination mount and the cloned mount is made slave | |
1304 | * of the same master as that of the source mount. The cloned mount | |
1305 | * is marked as 'shared and slave'. | |
1306 | * (*) the cloned mount is made a slave of the same master as that of the | |
1307 | * source mount. | |
1308 | * | |
9676f0c6 RP |
1309 | * --------------------------------------------------------------------------- |
1310 | * | MOVE MOUNT OPERATION | | |
1311 | * |************************************************************************** | |
1312 | * | source-->| shared | private | slave | unbindable | | |
1313 | * | dest | | | | | | |
1314 | * | | | | | | | | |
1315 | * | v | | | | | | |
1316 | * |************************************************************************** | |
1317 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1318 | * | | | | | | | |
1319 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1320 | * *************************************************************************** | |
5afe0022 RP |
1321 | * |
1322 | * (+) the mount is moved to the destination. And is then propagated to | |
1323 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1324 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1325 | * (+++) the mount is moved to the destination and is then propagated to |
1326 | * all the mounts belonging to the destination mount's propagation tree. | |
1327 | * the mount is marked as 'shared and slave'. | |
1328 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1329 | * |
1330 | * if the source mount is a tree, the operations explained above is | |
1331 | * applied to each mount in the tree. | |
1332 | * Must be called without spinlocks held, since this function can sleep | |
1333 | * in allocations. | |
1334 | */ | |
1335 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1336 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1337 | { |
1338 | LIST_HEAD(tree_list); | |
1a390689 AV |
1339 | struct vfsmount *dest_mnt = path->mnt; |
1340 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1341 | struct vfsmount *child, *p; |
719f5d7f | 1342 | int err; |
b90fa9ae | 1343 | |
719f5d7f MS |
1344 | if (IS_MNT_SHARED(dest_mnt)) { |
1345 | err = invent_group_ids(source_mnt, true); | |
1346 | if (err) | |
1347 | goto out; | |
1348 | } | |
1349 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1350 | if (err) | |
1351 | goto out_cleanup_ids; | |
b90fa9ae RP |
1352 | |
1353 | if (IS_MNT_SHARED(dest_mnt)) { | |
1354 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1355 | set_mnt_shared(p); | |
1356 | } | |
1357 | ||
1358 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1359 | if (parent_path) { |
1360 | detach_mnt(source_mnt, parent_path); | |
1361 | attach_mnt(source_mnt, path); | |
e5d67f07 | 1362 | touch_mnt_namespace(parent_path->mnt->mnt_ns); |
21444403 RP |
1363 | } else { |
1364 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1365 | commit_tree(source_mnt); | |
1366 | } | |
b90fa9ae RP |
1367 | |
1368 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1369 | list_del_init(&child->mnt_hash); | |
1370 | commit_tree(child); | |
1371 | } | |
1372 | spin_unlock(&vfsmount_lock); | |
1373 | return 0; | |
719f5d7f MS |
1374 | |
1375 | out_cleanup_ids: | |
1376 | if (IS_MNT_SHARED(dest_mnt)) | |
1377 | cleanup_group_ids(source_mnt, NULL); | |
1378 | out: | |
1379 | return err; | |
b90fa9ae RP |
1380 | } |
1381 | ||
8c3ee42e | 1382 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1383 | { |
1384 | int err; | |
1385 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1386 | return -EINVAL; | |
1387 | ||
8c3ee42e | 1388 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1389 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1390 | return -ENOTDIR; | |
1391 | ||
1392 | err = -ENOENT; | |
8c3ee42e AV |
1393 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1394 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1395 | goto out_unlock; |
1396 | ||
8c3ee42e | 1397 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1398 | if (err) |
1399 | goto out_unlock; | |
1400 | ||
1401 | err = -ENOENT; | |
f3da392e | 1402 | if (!d_unlinked(path->dentry)) |
8c3ee42e | 1403 | err = attach_recursive_mnt(mnt, path, NULL); |
1da177e4 | 1404 | out_unlock: |
8c3ee42e | 1405 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1406 | if (!err) |
8c3ee42e | 1407 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1408 | return err; |
1409 | } | |
1410 | ||
07b20889 RP |
1411 | /* |
1412 | * recursively change the type of the mountpoint. | |
1413 | */ | |
0a0d8a46 | 1414 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1415 | { |
2d92ab3c | 1416 | struct vfsmount *m, *mnt = path->mnt; |
07b20889 RP |
1417 | int recurse = flag & MS_REC; |
1418 | int type = flag & ~MS_REC; | |
719f5d7f | 1419 | int err = 0; |
07b20889 | 1420 | |
ee6f9582 MS |
1421 | if (!capable(CAP_SYS_ADMIN)) |
1422 | return -EPERM; | |
1423 | ||
2d92ab3c | 1424 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1425 | return -EINVAL; |
1426 | ||
1427 | down_write(&namespace_sem); | |
719f5d7f MS |
1428 | if (type == MS_SHARED) { |
1429 | err = invent_group_ids(mnt, recurse); | |
1430 | if (err) | |
1431 | goto out_unlock; | |
1432 | } | |
1433 | ||
07b20889 RP |
1434 | spin_lock(&vfsmount_lock); |
1435 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1436 | change_mnt_propagation(m, type); | |
1437 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1438 | |
1439 | out_unlock: | |
07b20889 | 1440 | up_write(&namespace_sem); |
719f5d7f | 1441 | return err; |
07b20889 RP |
1442 | } |
1443 | ||
1da177e4 LT |
1444 | /* |
1445 | * do loopback mount. | |
1446 | */ | |
0a0d8a46 | 1447 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1448 | int recurse) |
1da177e4 | 1449 | { |
2d92ab3c | 1450 | struct path old_path; |
1da177e4 | 1451 | struct vfsmount *mnt = NULL; |
2d92ab3c | 1452 | int err = mount_is_safe(path); |
1da177e4 LT |
1453 | if (err) |
1454 | return err; | |
1455 | if (!old_name || !*old_name) | |
1456 | return -EINVAL; | |
2d92ab3c | 1457 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1458 | if (err) |
1459 | return err; | |
1460 | ||
390c6843 | 1461 | down_write(&namespace_sem); |
1da177e4 | 1462 | err = -EINVAL; |
2d92ab3c | 1463 | if (IS_MNT_UNBINDABLE(old_path.mnt)) |
4ac91378 | 1464 | goto out; |
9676f0c6 | 1465 | |
2d92ab3c | 1466 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
ccd48bc7 | 1467 | goto out; |
1da177e4 | 1468 | |
ccd48bc7 AV |
1469 | err = -ENOMEM; |
1470 | if (recurse) | |
2d92ab3c | 1471 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 | 1472 | else |
2d92ab3c | 1473 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 AV |
1474 | |
1475 | if (!mnt) | |
1476 | goto out; | |
1477 | ||
2d92ab3c | 1478 | err = graft_tree(mnt, path); |
ccd48bc7 | 1479 | if (err) { |
70fbcdf4 | 1480 | LIST_HEAD(umount_list); |
1da177e4 | 1481 | spin_lock(&vfsmount_lock); |
a05964f3 | 1482 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1483 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1484 | release_mounts(&umount_list); |
5b83d2c5 | 1485 | } |
1da177e4 | 1486 | |
ccd48bc7 | 1487 | out: |
390c6843 | 1488 | up_write(&namespace_sem); |
2d92ab3c | 1489 | path_put(&old_path); |
1da177e4 LT |
1490 | return err; |
1491 | } | |
1492 | ||
2e4b7fcd DH |
1493 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1494 | { | |
1495 | int error = 0; | |
1496 | int readonly_request = 0; | |
1497 | ||
1498 | if (ms_flags & MS_RDONLY) | |
1499 | readonly_request = 1; | |
1500 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1501 | return 0; | |
1502 | ||
1503 | if (readonly_request) | |
1504 | error = mnt_make_readonly(mnt); | |
1505 | else | |
1506 | __mnt_unmake_readonly(mnt); | |
1507 | return error; | |
1508 | } | |
1509 | ||
1da177e4 LT |
1510 | /* |
1511 | * change filesystem flags. dir should be a physical root of filesystem. | |
1512 | * If you've mounted a non-root directory somewhere and want to do remount | |
1513 | * on it - tough luck. | |
1514 | */ | |
0a0d8a46 | 1515 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1516 | void *data) |
1517 | { | |
1518 | int err; | |
2d92ab3c | 1519 | struct super_block *sb = path->mnt->mnt_sb; |
1da177e4 LT |
1520 | |
1521 | if (!capable(CAP_SYS_ADMIN)) | |
1522 | return -EPERM; | |
1523 | ||
2d92ab3c | 1524 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1525 | return -EINVAL; |
1526 | ||
2d92ab3c | 1527 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1528 | return -EINVAL; |
1529 | ||
1530 | down_write(&sb->s_umount); | |
2e4b7fcd | 1531 | if (flags & MS_BIND) |
2d92ab3c | 1532 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1533 | else |
2e4b7fcd | 1534 | err = do_remount_sb(sb, flags, data, 0); |
1da177e4 | 1535 | if (!err) |
2d92ab3c | 1536 | path->mnt->mnt_flags = mnt_flags; |
1da177e4 | 1537 | up_write(&sb->s_umount); |
0e55a7cc | 1538 | if (!err) { |
2d92ab3c | 1539 | security_sb_post_remount(path->mnt, flags, data); |
0e55a7cc DW |
1540 | |
1541 | spin_lock(&vfsmount_lock); | |
1542 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1543 | spin_unlock(&vfsmount_lock); | |
1544 | } | |
1da177e4 LT |
1545 | return err; |
1546 | } | |
1547 | ||
9676f0c6 RP |
1548 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1549 | { | |
1550 | struct vfsmount *p; | |
1551 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1552 | if (IS_MNT_UNBINDABLE(p)) | |
1553 | return 1; | |
1554 | } | |
1555 | return 0; | |
1556 | } | |
1557 | ||
0a0d8a46 | 1558 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1559 | { |
2d92ab3c | 1560 | struct path old_path, parent_path; |
1da177e4 LT |
1561 | struct vfsmount *p; |
1562 | int err = 0; | |
1563 | if (!capable(CAP_SYS_ADMIN)) | |
1564 | return -EPERM; | |
1565 | if (!old_name || !*old_name) | |
1566 | return -EINVAL; | |
2d92ab3c | 1567 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1568 | if (err) |
1569 | return err; | |
1570 | ||
390c6843 | 1571 | down_write(&namespace_sem); |
2d92ab3c | 1572 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1573 | follow_down(path)) |
1da177e4 LT |
1574 | ; |
1575 | err = -EINVAL; | |
2d92ab3c | 1576 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
1da177e4 LT |
1577 | goto out; |
1578 | ||
1579 | err = -ENOENT; | |
2d92ab3c AV |
1580 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1581 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1582 | goto out1; |
1583 | ||
f3da392e | 1584 | if (d_unlinked(path->dentry)) |
21444403 | 1585 | goto out1; |
1da177e4 LT |
1586 | |
1587 | err = -EINVAL; | |
2d92ab3c | 1588 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1589 | goto out1; |
1da177e4 | 1590 | |
2d92ab3c | 1591 | if (old_path.mnt == old_path.mnt->mnt_parent) |
21444403 | 1592 | goto out1; |
1da177e4 | 1593 | |
2d92ab3c AV |
1594 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1595 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1596 | goto out1; |
1597 | /* | |
1598 | * Don't move a mount residing in a shared parent. | |
1599 | */ | |
2d92ab3c AV |
1600 | if (old_path.mnt->mnt_parent && |
1601 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
21444403 | 1602 | goto out1; |
9676f0c6 RP |
1603 | /* |
1604 | * Don't move a mount tree containing unbindable mounts to a destination | |
1605 | * mount which is shared. | |
1606 | */ | |
2d92ab3c AV |
1607 | if (IS_MNT_SHARED(path->mnt) && |
1608 | tree_contains_unbindable(old_path.mnt)) | |
9676f0c6 | 1609 | goto out1; |
1da177e4 | 1610 | err = -ELOOP; |
2d92ab3c AV |
1611 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1612 | if (p == old_path.mnt) | |
21444403 | 1613 | goto out1; |
1da177e4 | 1614 | |
2d92ab3c | 1615 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); |
4ac91378 | 1616 | if (err) |
21444403 | 1617 | goto out1; |
1da177e4 LT |
1618 | |
1619 | /* if the mount is moved, it should no longer be expire | |
1620 | * automatically */ | |
2d92ab3c | 1621 | list_del_init(&old_path.mnt->mnt_expire); |
1da177e4 | 1622 | out1: |
2d92ab3c | 1623 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1624 | out: |
390c6843 | 1625 | up_write(&namespace_sem); |
1da177e4 | 1626 | if (!err) |
1a390689 | 1627 | path_put(&parent_path); |
2d92ab3c | 1628 | path_put(&old_path); |
1da177e4 LT |
1629 | return err; |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * create a new mount for userspace and request it to be added into the | |
1634 | * namespace's tree | |
1635 | */ | |
0a0d8a46 | 1636 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1637 | int mnt_flags, char *name, void *data) |
1638 | { | |
1639 | struct vfsmount *mnt; | |
1640 | ||
1641 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1642 | return -EINVAL; | |
1643 | ||
1644 | /* we need capabilities... */ | |
1645 | if (!capable(CAP_SYS_ADMIN)) | |
1646 | return -EPERM; | |
1647 | ||
7f78d4cd | 1648 | lock_kernel(); |
1da177e4 | 1649 | mnt = do_kern_mount(type, flags, name, data); |
7f78d4cd | 1650 | unlock_kernel(); |
1da177e4 LT |
1651 | if (IS_ERR(mnt)) |
1652 | return PTR_ERR(mnt); | |
1653 | ||
2d92ab3c | 1654 | return do_add_mount(mnt, path, mnt_flags, NULL); |
1da177e4 LT |
1655 | } |
1656 | ||
1657 | /* | |
1658 | * add a mount into a namespace's mount tree | |
1659 | * - provide the option of adding the new mount to an expiration list | |
1660 | */ | |
8d66bf54 | 1661 | int do_add_mount(struct vfsmount *newmnt, struct path *path, |
1da177e4 LT |
1662 | int mnt_flags, struct list_head *fslist) |
1663 | { | |
1664 | int err; | |
1665 | ||
390c6843 | 1666 | down_write(&namespace_sem); |
1da177e4 | 1667 | /* Something was mounted here while we slept */ |
8d66bf54 | 1668 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1669 | follow_down(path)) |
1da177e4 LT |
1670 | ; |
1671 | err = -EINVAL; | |
dd5cae6e | 1672 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) |
1da177e4 LT |
1673 | goto unlock; |
1674 | ||
1675 | /* Refuse the same filesystem on the same mount point */ | |
1676 | err = -EBUSY; | |
8d66bf54 AV |
1677 | if (path->mnt->mnt_sb == newmnt->mnt_sb && |
1678 | path->mnt->mnt_root == path->dentry) | |
1da177e4 LT |
1679 | goto unlock; |
1680 | ||
1681 | err = -EINVAL; | |
1682 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1683 | goto unlock; | |
1684 | ||
1685 | newmnt->mnt_flags = mnt_flags; | |
8d66bf54 | 1686 | if ((err = graft_tree(newmnt, path))) |
5b83d2c5 | 1687 | goto unlock; |
1da177e4 | 1688 | |
6758f953 | 1689 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1690 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1691 | |
390c6843 | 1692 | up_write(&namespace_sem); |
5b83d2c5 | 1693 | return 0; |
1da177e4 LT |
1694 | |
1695 | unlock: | |
390c6843 | 1696 | up_write(&namespace_sem); |
1da177e4 LT |
1697 | mntput(newmnt); |
1698 | return err; | |
1699 | } | |
1700 | ||
1701 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1702 | ||
1703 | /* | |
1704 | * process a list of expirable mountpoints with the intent of discarding any | |
1705 | * mountpoints that aren't in use and haven't been touched since last we came | |
1706 | * here | |
1707 | */ | |
1708 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1709 | { | |
1da177e4 LT |
1710 | struct vfsmount *mnt, *next; |
1711 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1712 | LIST_HEAD(umounts); |
1da177e4 LT |
1713 | |
1714 | if (list_empty(mounts)) | |
1715 | return; | |
1716 | ||
bcc5c7d2 | 1717 | down_write(&namespace_sem); |
1da177e4 LT |
1718 | spin_lock(&vfsmount_lock); |
1719 | ||
1720 | /* extract from the expiration list every vfsmount that matches the | |
1721 | * following criteria: | |
1722 | * - only referenced by its parent vfsmount | |
1723 | * - still marked for expiry (marked on the last call here; marks are | |
1724 | * cleared by mntput()) | |
1725 | */ | |
55e700b9 | 1726 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1727 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1728 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1729 | continue; |
55e700b9 | 1730 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1731 | } |
bcc5c7d2 AV |
1732 | while (!list_empty(&graveyard)) { |
1733 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1734 | touch_mnt_namespace(mnt->mnt_ns); | |
1735 | umount_tree(mnt, 1, &umounts); | |
1736 | } | |
5528f911 | 1737 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1738 | up_write(&namespace_sem); |
1739 | ||
1740 | release_mounts(&umounts); | |
5528f911 TM |
1741 | } |
1742 | ||
1743 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1744 | ||
1745 | /* | |
1746 | * Ripoff of 'select_parent()' | |
1747 | * | |
1748 | * search the list of submounts for a given mountpoint, and move any | |
1749 | * shrinkable submounts to the 'graveyard' list. | |
1750 | */ | |
1751 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1752 | { | |
1753 | struct vfsmount *this_parent = parent; | |
1754 | struct list_head *next; | |
1755 | int found = 0; | |
1756 | ||
1757 | repeat: | |
1758 | next = this_parent->mnt_mounts.next; | |
1759 | resume: | |
1760 | while (next != &this_parent->mnt_mounts) { | |
1761 | struct list_head *tmp = next; | |
1762 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1763 | ||
1764 | next = tmp->next; | |
1765 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1766 | continue; |
5528f911 TM |
1767 | /* |
1768 | * Descend a level if the d_mounts list is non-empty. | |
1769 | */ | |
1770 | if (!list_empty(&mnt->mnt_mounts)) { | |
1771 | this_parent = mnt; | |
1772 | goto repeat; | |
1773 | } | |
1da177e4 | 1774 | |
5528f911 | 1775 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1776 | list_move_tail(&mnt->mnt_expire, graveyard); |
1777 | found++; | |
1778 | } | |
1da177e4 | 1779 | } |
5528f911 TM |
1780 | /* |
1781 | * All done at this level ... ascend and resume the search | |
1782 | */ | |
1783 | if (this_parent != parent) { | |
1784 | next = this_parent->mnt_child.next; | |
1785 | this_parent = this_parent->mnt_parent; | |
1786 | goto resume; | |
1787 | } | |
1788 | return found; | |
1789 | } | |
1790 | ||
1791 | /* | |
1792 | * process a list of expirable mountpoints with the intent of discarding any | |
1793 | * submounts of a specific parent mountpoint | |
1794 | */ | |
c35038be | 1795 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1796 | { |
1797 | LIST_HEAD(graveyard); | |
c35038be | 1798 | struct vfsmount *m; |
5528f911 | 1799 | |
5528f911 | 1800 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1801 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1802 | while (!list_empty(&graveyard)) { |
c35038be | 1803 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 | 1804 | mnt_expire); |
afef80b3 EB |
1805 | touch_mnt_namespace(m->mnt_ns); |
1806 | umount_tree(m, 1, umounts); | |
bcc5c7d2 AV |
1807 | } |
1808 | } | |
1da177e4 LT |
1809 | } |
1810 | ||
1da177e4 LT |
1811 | /* |
1812 | * Some copy_from_user() implementations do not return the exact number of | |
1813 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1814 | * Note that this function differs from copy_from_user() in that it will oops | |
1815 | * on bad values of `to', rather than returning a short copy. | |
1816 | */ | |
b58fed8b RP |
1817 | static long exact_copy_from_user(void *to, const void __user * from, |
1818 | unsigned long n) | |
1da177e4 LT |
1819 | { |
1820 | char *t = to; | |
1821 | const char __user *f = from; | |
1822 | char c; | |
1823 | ||
1824 | if (!access_ok(VERIFY_READ, from, n)) | |
1825 | return n; | |
1826 | ||
1827 | while (n) { | |
1828 | if (__get_user(c, f)) { | |
1829 | memset(t, 0, n); | |
1830 | break; | |
1831 | } | |
1832 | *t++ = c; | |
1833 | f++; | |
1834 | n--; | |
1835 | } | |
1836 | return n; | |
1837 | } | |
1838 | ||
b58fed8b | 1839 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1840 | { |
1841 | int i; | |
1842 | unsigned long page; | |
1843 | unsigned long size; | |
b58fed8b | 1844 | |
1da177e4 LT |
1845 | *where = 0; |
1846 | if (!data) | |
1847 | return 0; | |
1848 | ||
1849 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1850 | return -ENOMEM; | |
1851 | ||
1852 | /* We only care that *some* data at the address the user | |
1853 | * gave us is valid. Just in case, we'll zero | |
1854 | * the remainder of the page. | |
1855 | */ | |
1856 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1857 | size = TASK_SIZE - (unsigned long)data; | |
1858 | if (size > PAGE_SIZE) | |
1859 | size = PAGE_SIZE; | |
1860 | ||
1861 | i = size - exact_copy_from_user((void *)page, data, size); | |
1862 | if (!i) { | |
b58fed8b | 1863 | free_page(page); |
1da177e4 LT |
1864 | return -EFAULT; |
1865 | } | |
1866 | if (i != PAGE_SIZE) | |
1867 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1868 | *where = page; | |
1869 | return 0; | |
1870 | } | |
1871 | ||
1872 | /* | |
1873 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1874 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1875 | * | |
1876 | * data is a (void *) that can point to any structure up to | |
1877 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1878 | * information (or be NULL). | |
1879 | * | |
1880 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1881 | * When the flags word was introduced its top half was required | |
1882 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1883 | * Therefore, if this magic number is present, it carries no information | |
1884 | * and must be discarded. | |
1885 | */ | |
b58fed8b | 1886 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1887 | unsigned long flags, void *data_page) |
1888 | { | |
2d92ab3c | 1889 | struct path path; |
1da177e4 LT |
1890 | int retval = 0; |
1891 | int mnt_flags = 0; | |
1892 | ||
1893 | /* Discard magic */ | |
1894 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1895 | flags &= ~MS_MGC_MSK; | |
1896 | ||
1897 | /* Basic sanity checks */ | |
1898 | ||
1899 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1900 | return -EINVAL; | |
1901 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1902 | return -EINVAL; | |
1903 | ||
1904 | if (data_page) | |
1905 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1906 | ||
613cbe3d AK |
1907 | /* Default to relatime unless overriden */ |
1908 | if (!(flags & MS_NOATIME)) | |
1909 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 1910 | |
1da177e4 LT |
1911 | /* Separate the per-mountpoint flags */ |
1912 | if (flags & MS_NOSUID) | |
1913 | mnt_flags |= MNT_NOSUID; | |
1914 | if (flags & MS_NODEV) | |
1915 | mnt_flags |= MNT_NODEV; | |
1916 | if (flags & MS_NOEXEC) | |
1917 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1918 | if (flags & MS_NOATIME) |
1919 | mnt_flags |= MNT_NOATIME; | |
1920 | if (flags & MS_NODIRATIME) | |
1921 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
1922 | if (flags & MS_STRICTATIME) |
1923 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
1924 | if (flags & MS_RDONLY) |
1925 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1926 | |
1927 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
d0adde57 MG |
1928 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
1929 | MS_STRICTATIME); | |
1da177e4 LT |
1930 | |
1931 | /* ... and get the mountpoint */ | |
2d92ab3c | 1932 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1933 | if (retval) |
1934 | return retval; | |
1935 | ||
2d92ab3c | 1936 | retval = security_sb_mount(dev_name, &path, |
b5266eb4 | 1937 | type_page, flags, data_page); |
1da177e4 LT |
1938 | if (retval) |
1939 | goto dput_out; | |
1940 | ||
1941 | if (flags & MS_REMOUNT) | |
2d92ab3c | 1942 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
1943 | data_page); |
1944 | else if (flags & MS_BIND) | |
2d92ab3c | 1945 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 1946 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 1947 | retval = do_change_type(&path, flags); |
1da177e4 | 1948 | else if (flags & MS_MOVE) |
2d92ab3c | 1949 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 1950 | else |
2d92ab3c | 1951 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
1952 | dev_name, data_page); |
1953 | dput_out: | |
2d92ab3c | 1954 | path_put(&path); |
1da177e4 LT |
1955 | return retval; |
1956 | } | |
1957 | ||
cf8d2c11 TM |
1958 | static struct mnt_namespace *alloc_mnt_ns(void) |
1959 | { | |
1960 | struct mnt_namespace *new_ns; | |
1961 | ||
1962 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
1963 | if (!new_ns) | |
1964 | return ERR_PTR(-ENOMEM); | |
1965 | atomic_set(&new_ns->count, 1); | |
1966 | new_ns->root = NULL; | |
1967 | INIT_LIST_HEAD(&new_ns->list); | |
1968 | init_waitqueue_head(&new_ns->poll); | |
1969 | new_ns->event = 0; | |
1970 | return new_ns; | |
1971 | } | |
1972 | ||
741a2951 JD |
1973 | /* |
1974 | * Allocate a new namespace structure and populate it with contents | |
1975 | * copied from the namespace of the passed in task structure. | |
1976 | */ | |
e3222c4e | 1977 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1978 | struct fs_struct *fs) |
1da177e4 | 1979 | { |
6b3286ed | 1980 | struct mnt_namespace *new_ns; |
7f2da1e7 | 1981 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
1da177e4 LT |
1982 | struct vfsmount *p, *q; |
1983 | ||
cf8d2c11 TM |
1984 | new_ns = alloc_mnt_ns(); |
1985 | if (IS_ERR(new_ns)) | |
1986 | return new_ns; | |
1da177e4 | 1987 | |
390c6843 | 1988 | down_write(&namespace_sem); |
1da177e4 | 1989 | /* First pass: copy the tree topology */ |
6b3286ed | 1990 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1991 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1992 | if (!new_ns->root) { |
390c6843 | 1993 | up_write(&namespace_sem); |
1da177e4 | 1994 | kfree(new_ns); |
5cc4a034 | 1995 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1996 | } |
1997 | spin_lock(&vfsmount_lock); | |
1998 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1999 | spin_unlock(&vfsmount_lock); | |
2000 | ||
2001 | /* | |
2002 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2003 | * as belonging to new namespace. We have already acquired a private | |
2004 | * fs_struct, so tsk->fs->lock is not needed. | |
2005 | */ | |
6b3286ed | 2006 | p = mnt_ns->root; |
1da177e4 LT |
2007 | q = new_ns->root; |
2008 | while (p) { | |
6b3286ed | 2009 | q->mnt_ns = new_ns; |
1da177e4 | 2010 | if (fs) { |
6ac08c39 | 2011 | if (p == fs->root.mnt) { |
1da177e4 | 2012 | rootmnt = p; |
6ac08c39 | 2013 | fs->root.mnt = mntget(q); |
1da177e4 | 2014 | } |
6ac08c39 | 2015 | if (p == fs->pwd.mnt) { |
1da177e4 | 2016 | pwdmnt = p; |
6ac08c39 | 2017 | fs->pwd.mnt = mntget(q); |
1da177e4 | 2018 | } |
1da177e4 | 2019 | } |
6b3286ed | 2020 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
2021 | q = next_mnt(q, new_ns->root); |
2022 | } | |
390c6843 | 2023 | up_write(&namespace_sem); |
1da177e4 | 2024 | |
1da177e4 LT |
2025 | if (rootmnt) |
2026 | mntput(rootmnt); | |
2027 | if (pwdmnt) | |
2028 | mntput(pwdmnt); | |
1da177e4 | 2029 | |
741a2951 JD |
2030 | return new_ns; |
2031 | } | |
2032 | ||
213dd266 | 2033 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2034 | struct fs_struct *new_fs) |
741a2951 | 2035 | { |
6b3286ed | 2036 | struct mnt_namespace *new_ns; |
741a2951 | 2037 | |
e3222c4e | 2038 | BUG_ON(!ns); |
6b3286ed | 2039 | get_mnt_ns(ns); |
741a2951 JD |
2040 | |
2041 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2042 | return ns; |
741a2951 | 2043 | |
e3222c4e | 2044 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2045 | |
6b3286ed | 2046 | put_mnt_ns(ns); |
e3222c4e | 2047 | return new_ns; |
1da177e4 LT |
2048 | } |
2049 | ||
cf8d2c11 TM |
2050 | /** |
2051 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2052 | * @mnt: pointer to the new root filesystem mountpoint | |
2053 | */ | |
2054 | struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) | |
2055 | { | |
2056 | struct mnt_namespace *new_ns; | |
2057 | ||
2058 | new_ns = alloc_mnt_ns(); | |
2059 | if (!IS_ERR(new_ns)) { | |
2060 | mnt->mnt_ns = new_ns; | |
2061 | new_ns->root = mnt; | |
2062 | list_add(&new_ns->list, &new_ns->root->mnt_list); | |
2063 | } | |
2064 | return new_ns; | |
2065 | } | |
2066 | EXPORT_SYMBOL(create_mnt_ns); | |
2067 | ||
bdc480e3 HC |
2068 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2069 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 LT |
2070 | { |
2071 | int retval; | |
2072 | unsigned long data_page; | |
2073 | unsigned long type_page; | |
2074 | unsigned long dev_page; | |
2075 | char *dir_page; | |
2076 | ||
b58fed8b | 2077 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
2078 | if (retval < 0) |
2079 | return retval; | |
2080 | ||
2081 | dir_page = getname(dir_name); | |
2082 | retval = PTR_ERR(dir_page); | |
2083 | if (IS_ERR(dir_page)) | |
2084 | goto out1; | |
2085 | ||
b58fed8b | 2086 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2087 | if (retval < 0) |
2088 | goto out2; | |
2089 | ||
b58fed8b | 2090 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2091 | if (retval < 0) |
2092 | goto out3; | |
2093 | ||
b58fed8b RP |
2094 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2095 | flags, (void *)data_page); | |
1da177e4 LT |
2096 | free_page(data_page); |
2097 | ||
2098 | out3: | |
2099 | free_page(dev_page); | |
2100 | out2: | |
2101 | putname(dir_page); | |
2102 | out1: | |
2103 | free_page(type_page); | |
2104 | return retval; | |
2105 | } | |
2106 | ||
1da177e4 LT |
2107 | /* |
2108 | * pivot_root Semantics: | |
2109 | * Moves the root file system of the current process to the directory put_old, | |
2110 | * makes new_root as the new root file system of the current process, and sets | |
2111 | * root/cwd of all processes which had them on the current root to new_root. | |
2112 | * | |
2113 | * Restrictions: | |
2114 | * The new_root and put_old must be directories, and must not be on the | |
2115 | * same file system as the current process root. The put_old must be | |
2116 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2117 | * pointed to by put_old must yield the same directory as new_root. No other | |
2118 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2119 | * | |
4a0d11fa NB |
2120 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2121 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2122 | * in this situation. | |
2123 | * | |
1da177e4 LT |
2124 | * Notes: |
2125 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2126 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2127 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2128 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2129 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2130 | * first. | |
2131 | */ | |
3480b257 HC |
2132 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2133 | const char __user *, put_old) | |
1da177e4 LT |
2134 | { |
2135 | struct vfsmount *tmp; | |
2d8f3038 | 2136 | struct path new, old, parent_path, root_parent, root; |
1da177e4 LT |
2137 | int error; |
2138 | ||
2139 | if (!capable(CAP_SYS_ADMIN)) | |
2140 | return -EPERM; | |
2141 | ||
2d8f3038 | 2142 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2143 | if (error) |
2144 | goto out0; | |
2145 | error = -EINVAL; | |
2d8f3038 | 2146 | if (!check_mnt(new.mnt)) |
1da177e4 LT |
2147 | goto out1; |
2148 | ||
2d8f3038 | 2149 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2150 | if (error) |
2151 | goto out1; | |
2152 | ||
2d8f3038 | 2153 | error = security_sb_pivotroot(&old, &new); |
1da177e4 | 2154 | if (error) { |
2d8f3038 | 2155 | path_put(&old); |
1da177e4 LT |
2156 | goto out1; |
2157 | } | |
2158 | ||
2159 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2160 | root = current->fs->root; |
6ac08c39 | 2161 | path_get(¤t->fs->root); |
1da177e4 | 2162 | read_unlock(¤t->fs->lock); |
390c6843 | 2163 | down_write(&namespace_sem); |
2d8f3038 | 2164 | mutex_lock(&old.dentry->d_inode->i_mutex); |
1da177e4 | 2165 | error = -EINVAL; |
2d8f3038 AV |
2166 | if (IS_MNT_SHARED(old.mnt) || |
2167 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
8c3ee42e | 2168 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2169 | goto out2; |
8c3ee42e | 2170 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2171 | goto out2; |
2172 | error = -ENOENT; | |
2d8f3038 | 2173 | if (IS_DEADDIR(new.dentry->d_inode)) |
1da177e4 | 2174 | goto out2; |
f3da392e | 2175 | if (d_unlinked(new.dentry)) |
1da177e4 | 2176 | goto out2; |
f3da392e | 2177 | if (d_unlinked(old.dentry)) |
1da177e4 LT |
2178 | goto out2; |
2179 | error = -EBUSY; | |
2d8f3038 AV |
2180 | if (new.mnt == root.mnt || |
2181 | old.mnt == root.mnt) | |
1da177e4 LT |
2182 | goto out2; /* loop, on the same file system */ |
2183 | error = -EINVAL; | |
8c3ee42e | 2184 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2185 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2186 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2187 | goto out2; /* not attached */ |
2d8f3038 | 2188 | if (new.mnt->mnt_root != new.dentry) |
1da177e4 | 2189 | goto out2; /* not a mountpoint */ |
2d8f3038 | 2190 | if (new.mnt->mnt_parent == new.mnt) |
0bb6fcc1 | 2191 | goto out2; /* not attached */ |
4ac91378 | 2192 | /* make sure we can reach put_old from new_root */ |
2d8f3038 | 2193 | tmp = old.mnt; |
1da177e4 | 2194 | spin_lock(&vfsmount_lock); |
2d8f3038 | 2195 | if (tmp != new.mnt) { |
1da177e4 LT |
2196 | for (;;) { |
2197 | if (tmp->mnt_parent == tmp) | |
2198 | goto out3; /* already mounted on put_old */ | |
2d8f3038 | 2199 | if (tmp->mnt_parent == new.mnt) |
1da177e4 LT |
2200 | break; |
2201 | tmp = tmp->mnt_parent; | |
2202 | } | |
2d8f3038 | 2203 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) |
1da177e4 | 2204 | goto out3; |
2d8f3038 | 2205 | } else if (!is_subdir(old.dentry, new.dentry)) |
1da177e4 | 2206 | goto out3; |
2d8f3038 | 2207 | detach_mnt(new.mnt, &parent_path); |
8c3ee42e | 2208 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2209 | /* mount old root on put_old */ |
2d8f3038 | 2210 | attach_mnt(root.mnt, &old); |
4ac91378 | 2211 | /* mount new_root on / */ |
2d8f3038 | 2212 | attach_mnt(new.mnt, &root_parent); |
6b3286ed | 2213 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2214 | spin_unlock(&vfsmount_lock); |
2d8f3038 AV |
2215 | chroot_fs_refs(&root, &new); |
2216 | security_sb_post_pivotroot(&root, &new); | |
1da177e4 | 2217 | error = 0; |
1a390689 AV |
2218 | path_put(&root_parent); |
2219 | path_put(&parent_path); | |
1da177e4 | 2220 | out2: |
2d8f3038 | 2221 | mutex_unlock(&old.dentry->d_inode->i_mutex); |
390c6843 | 2222 | up_write(&namespace_sem); |
8c3ee42e | 2223 | path_put(&root); |
2d8f3038 | 2224 | path_put(&old); |
1da177e4 | 2225 | out1: |
2d8f3038 | 2226 | path_put(&new); |
1da177e4 | 2227 | out0: |
1da177e4 LT |
2228 | return error; |
2229 | out3: | |
2230 | spin_unlock(&vfsmount_lock); | |
2231 | goto out2; | |
2232 | } | |
2233 | ||
2234 | static void __init init_mount_tree(void) | |
2235 | { | |
2236 | struct vfsmount *mnt; | |
6b3286ed | 2237 | struct mnt_namespace *ns; |
ac748a09 | 2238 | struct path root; |
1da177e4 LT |
2239 | |
2240 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2241 | if (IS_ERR(mnt)) | |
2242 | panic("Can't create rootfs"); | |
3b22edc5 TM |
2243 | ns = create_mnt_ns(mnt); |
2244 | if (IS_ERR(ns)) | |
1da177e4 | 2245 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2246 | |
2247 | init_task.nsproxy->mnt_ns = ns; | |
2248 | get_mnt_ns(ns); | |
2249 | ||
ac748a09 JB |
2250 | root.mnt = ns->root; |
2251 | root.dentry = ns->root->mnt_root; | |
2252 | ||
2253 | set_fs_pwd(current->fs, &root); | |
2254 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2255 | } |
2256 | ||
74bf17cf | 2257 | void __init mnt_init(void) |
1da177e4 | 2258 | { |
13f14b4d | 2259 | unsigned u; |
15a67dd8 | 2260 | int err; |
1da177e4 | 2261 | |
390c6843 RP |
2262 | init_rwsem(&namespace_sem); |
2263 | ||
1da177e4 | 2264 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2265 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2266 | |
b58fed8b | 2267 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2268 | |
2269 | if (!mount_hashtable) | |
2270 | panic("Failed to allocate mount hash table\n"); | |
2271 | ||
13f14b4d ED |
2272 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2273 | ||
2274 | for (u = 0; u < HASH_SIZE; u++) | |
2275 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2276 | |
15a67dd8 RD |
2277 | err = sysfs_init(); |
2278 | if (err) | |
2279 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2280 | __func__, err); |
00d26666 GKH |
2281 | fs_kobj = kobject_create_and_add("fs", NULL); |
2282 | if (!fs_kobj) | |
8e24eea7 | 2283 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2284 | init_rootfs(); |
2285 | init_mount_tree(); | |
2286 | } | |
2287 | ||
616511d0 | 2288 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2289 | { |
616511d0 | 2290 | struct vfsmount *root; |
70fbcdf4 | 2291 | LIST_HEAD(umount_list); |
616511d0 TM |
2292 | |
2293 | if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock)) | |
2294 | return; | |
2295 | root = ns->root; | |
6b3286ed | 2296 | ns->root = NULL; |
1ce88cf4 | 2297 | spin_unlock(&vfsmount_lock); |
390c6843 | 2298 | down_write(&namespace_sem); |
1da177e4 | 2299 | spin_lock(&vfsmount_lock); |
a05964f3 | 2300 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2301 | spin_unlock(&vfsmount_lock); |
390c6843 | 2302 | up_write(&namespace_sem); |
70fbcdf4 | 2303 | release_mounts(&umount_list); |
6b3286ed | 2304 | kfree(ns); |
1da177e4 | 2305 | } |
cf8d2c11 | 2306 | EXPORT_SYMBOL(put_mnt_ns); |