]>
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 | 24 | #include <linux/namei.h> |
b43f3cbd | 25 | #include <linux/nsproxy.h> |
1da177e4 LT |
26 | #include <linux/security.h> |
27 | #include <linux/mount.h> | |
07f3f05c | 28 | #include <linux/ramfs.h> |
13f14b4d | 29 | #include <linux/log2.h> |
73cd49ec | 30 | #include <linux/idr.h> |
5ad4e53b | 31 | #include <linux/fs_struct.h> |
1da177e4 LT |
32 | #include <asm/uaccess.h> |
33 | #include <asm/unistd.h> | |
07b20889 | 34 | #include "pnode.h" |
948730b0 | 35 | #include "internal.h" |
1da177e4 | 36 | |
13f14b4d ED |
37 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
38 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
39 | ||
1da177e4 | 40 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
41 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
42 | ||
43 | static int event; | |
73cd49ec | 44 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 45 | static DEFINE_IDA(mnt_group_ida); |
f21f6220 AV |
46 | static int mnt_id_start = 0; |
47 | static int mnt_group_start = 1; | |
1da177e4 | 48 | |
fa3536cc | 49 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 50 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 51 | static struct rw_semaphore namespace_sem; |
1da177e4 | 52 | |
f87fd4c2 | 53 | /* /sys/fs */ |
00d26666 GKH |
54 | struct kobject *fs_kobj; |
55 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 56 | |
1da177e4 LT |
57 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
58 | { | |
b58fed8b RP |
59 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
60 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
61 | tmp = tmp + (tmp >> HASH_SHIFT); |
62 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
63 | } |
64 | ||
3d733633 DH |
65 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
66 | ||
73cd49ec MS |
67 | /* allocation is serialized by namespace_sem */ |
68 | static int mnt_alloc_id(struct vfsmount *mnt) | |
69 | { | |
70 | int res; | |
71 | ||
72 | retry: | |
73 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
74 | spin_lock(&vfsmount_lock); | |
f21f6220 AV |
75 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
76 | if (!res) | |
77 | mnt_id_start = mnt->mnt_id + 1; | |
73cd49ec MS |
78 | spin_unlock(&vfsmount_lock); |
79 | if (res == -EAGAIN) | |
80 | goto retry; | |
81 | ||
82 | return res; | |
83 | } | |
84 | ||
85 | static void mnt_free_id(struct vfsmount *mnt) | |
86 | { | |
f21f6220 | 87 | int id = mnt->mnt_id; |
73cd49ec | 88 | spin_lock(&vfsmount_lock); |
f21f6220 AV |
89 | ida_remove(&mnt_id_ida, id); |
90 | if (mnt_id_start > id) | |
91 | mnt_id_start = id; | |
73cd49ec MS |
92 | spin_unlock(&vfsmount_lock); |
93 | } | |
94 | ||
719f5d7f MS |
95 | /* |
96 | * Allocate a new peer group ID | |
97 | * | |
98 | * mnt_group_ida is protected by namespace_sem | |
99 | */ | |
100 | static int mnt_alloc_group_id(struct vfsmount *mnt) | |
101 | { | |
f21f6220 AV |
102 | int res; |
103 | ||
719f5d7f MS |
104 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
105 | return -ENOMEM; | |
106 | ||
f21f6220 AV |
107 | res = ida_get_new_above(&mnt_group_ida, |
108 | mnt_group_start, | |
109 | &mnt->mnt_group_id); | |
110 | if (!res) | |
111 | mnt_group_start = mnt->mnt_group_id + 1; | |
112 | ||
113 | return res; | |
719f5d7f MS |
114 | } |
115 | ||
116 | /* | |
117 | * Release a peer group ID | |
118 | */ | |
119 | void mnt_release_group_id(struct vfsmount *mnt) | |
120 | { | |
f21f6220 AV |
121 | int id = mnt->mnt_group_id; |
122 | ida_remove(&mnt_group_ida, id); | |
123 | if (mnt_group_start > id) | |
124 | mnt_group_start = id; | |
719f5d7f MS |
125 | mnt->mnt_group_id = 0; |
126 | } | |
127 | ||
1da177e4 LT |
128 | struct vfsmount *alloc_vfsmnt(const char *name) |
129 | { | |
c3762229 | 130 | struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
1da177e4 | 131 | if (mnt) { |
73cd49ec MS |
132 | int err; |
133 | ||
134 | err = mnt_alloc_id(mnt); | |
88b38782 LZ |
135 | if (err) |
136 | goto out_free_cache; | |
137 | ||
138 | if (name) { | |
139 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
140 | if (!mnt->mnt_devname) | |
141 | goto out_free_id; | |
73cd49ec MS |
142 | } |
143 | ||
b58fed8b | 144 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
145 | INIT_LIST_HEAD(&mnt->mnt_hash); |
146 | INIT_LIST_HEAD(&mnt->mnt_child); | |
147 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
148 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 149 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 150 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
151 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
152 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
d3ef3d73 | 153 | #ifdef CONFIG_SMP |
154 | mnt->mnt_writers = alloc_percpu(int); | |
155 | if (!mnt->mnt_writers) | |
156 | goto out_free_devname; | |
157 | #else | |
158 | mnt->mnt_writers = 0; | |
159 | #endif | |
1da177e4 LT |
160 | } |
161 | return mnt; | |
88b38782 | 162 | |
d3ef3d73 | 163 | #ifdef CONFIG_SMP |
164 | out_free_devname: | |
165 | kfree(mnt->mnt_devname); | |
166 | #endif | |
88b38782 LZ |
167 | out_free_id: |
168 | mnt_free_id(mnt); | |
169 | out_free_cache: | |
170 | kmem_cache_free(mnt_cache, mnt); | |
171 | return NULL; | |
1da177e4 LT |
172 | } |
173 | ||
3d733633 DH |
174 | /* |
175 | * Most r/o checks on a fs are for operations that take | |
176 | * discrete amounts of time, like a write() or unlink(). | |
177 | * We must keep track of when those operations start | |
178 | * (for permission checks) and when they end, so that | |
179 | * we can determine when writes are able to occur to | |
180 | * a filesystem. | |
181 | */ | |
182 | /* | |
183 | * __mnt_is_readonly: check whether a mount is read-only | |
184 | * @mnt: the mount to check for its write status | |
185 | * | |
186 | * This shouldn't be used directly ouside of the VFS. | |
187 | * It does not guarantee that the filesystem will stay | |
188 | * r/w, just that it is right *now*. This can not and | |
189 | * should not be used in place of IS_RDONLY(inode). | |
190 | * mnt_want/drop_write() will _keep_ the filesystem | |
191 | * r/w. | |
192 | */ | |
193 | int __mnt_is_readonly(struct vfsmount *mnt) | |
194 | { | |
2e4b7fcd DH |
195 | if (mnt->mnt_flags & MNT_READONLY) |
196 | return 1; | |
197 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
198 | return 1; | |
199 | return 0; | |
3d733633 DH |
200 | } |
201 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
202 | ||
d3ef3d73 | 203 | static inline void inc_mnt_writers(struct vfsmount *mnt) |
204 | { | |
205 | #ifdef CONFIG_SMP | |
206 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++; | |
207 | #else | |
208 | mnt->mnt_writers++; | |
209 | #endif | |
210 | } | |
3d733633 | 211 | |
d3ef3d73 | 212 | static inline void dec_mnt_writers(struct vfsmount *mnt) |
3d733633 | 213 | { |
d3ef3d73 | 214 | #ifdef CONFIG_SMP |
215 | (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--; | |
216 | #else | |
217 | mnt->mnt_writers--; | |
218 | #endif | |
3d733633 | 219 | } |
3d733633 | 220 | |
d3ef3d73 | 221 | static unsigned int count_mnt_writers(struct vfsmount *mnt) |
3d733633 | 222 | { |
d3ef3d73 | 223 | #ifdef CONFIG_SMP |
224 | unsigned int count = 0; | |
3d733633 | 225 | int cpu; |
3d733633 DH |
226 | |
227 | for_each_possible_cpu(cpu) { | |
d3ef3d73 | 228 | count += *per_cpu_ptr(mnt->mnt_writers, cpu); |
3d733633 | 229 | } |
3d733633 | 230 | |
d3ef3d73 | 231 | return count; |
232 | #else | |
233 | return mnt->mnt_writers; | |
234 | #endif | |
3d733633 DH |
235 | } |
236 | ||
8366025e DH |
237 | /* |
238 | * Most r/o checks on a fs are for operations that take | |
239 | * discrete amounts of time, like a write() or unlink(). | |
240 | * We must keep track of when those operations start | |
241 | * (for permission checks) and when they end, so that | |
242 | * we can determine when writes are able to occur to | |
243 | * a filesystem. | |
244 | */ | |
245 | /** | |
246 | * mnt_want_write - get write access to a mount | |
247 | * @mnt: the mount on which to take a write | |
248 | * | |
249 | * This tells the low-level filesystem that a write is | |
250 | * about to be performed to it, and makes sure that | |
251 | * writes are allowed before returning success. When | |
252 | * the write operation is finished, mnt_drop_write() | |
253 | * must be called. This is effectively a refcount. | |
254 | */ | |
255 | int mnt_want_write(struct vfsmount *mnt) | |
256 | { | |
3d733633 | 257 | int ret = 0; |
3d733633 | 258 | |
d3ef3d73 | 259 | preempt_disable(); |
260 | inc_mnt_writers(mnt); | |
261 | /* | |
262 | * The store to inc_mnt_writers must be visible before we pass | |
263 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our | |
264 | * incremented count after it has set MNT_WRITE_HOLD. | |
265 | */ | |
266 | smp_mb(); | |
267 | while (mnt->mnt_flags & MNT_WRITE_HOLD) | |
268 | cpu_relax(); | |
269 | /* | |
270 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
271 | * be set to match its requirements. So we must not load that until | |
272 | * MNT_WRITE_HOLD is cleared. | |
273 | */ | |
274 | smp_rmb(); | |
3d733633 | 275 | if (__mnt_is_readonly(mnt)) { |
d3ef3d73 | 276 | dec_mnt_writers(mnt); |
3d733633 DH |
277 | ret = -EROFS; |
278 | goto out; | |
279 | } | |
3d733633 | 280 | out: |
d3ef3d73 | 281 | preempt_enable(); |
3d733633 | 282 | return ret; |
8366025e DH |
283 | } |
284 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
285 | ||
96029c4e | 286 | /** |
287 | * mnt_clone_write - get write access to a mount | |
288 | * @mnt: the mount on which to take a write | |
289 | * | |
290 | * This is effectively like mnt_want_write, except | |
291 | * it must only be used to take an extra write reference | |
292 | * on a mountpoint that we already know has a write reference | |
293 | * on it. This allows some optimisation. | |
294 | * | |
295 | * After finished, mnt_drop_write must be called as usual to | |
296 | * drop the reference. | |
297 | */ | |
298 | int mnt_clone_write(struct vfsmount *mnt) | |
299 | { | |
300 | /* superblock may be r/o */ | |
301 | if (__mnt_is_readonly(mnt)) | |
302 | return -EROFS; | |
303 | preempt_disable(); | |
304 | inc_mnt_writers(mnt); | |
305 | preempt_enable(); | |
306 | return 0; | |
307 | } | |
308 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
309 | ||
310 | /** | |
311 | * mnt_want_write_file - get write access to a file's mount | |
312 | * @file: the file who's mount on which to take a write | |
313 | * | |
314 | * This is like mnt_want_write, but it takes a file and can | |
315 | * do some optimisations if the file is open for write already | |
316 | */ | |
317 | int mnt_want_write_file(struct file *file) | |
318 | { | |
319 | if (!(file->f_mode & FMODE_WRITE)) | |
320 | return mnt_want_write(file->f_path.mnt); | |
321 | else | |
322 | return mnt_clone_write(file->f_path.mnt); | |
323 | } | |
324 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
325 | ||
8366025e DH |
326 | /** |
327 | * mnt_drop_write - give up write access to a mount | |
328 | * @mnt: the mount on which to give up write access | |
329 | * | |
330 | * Tells the low-level filesystem that we are done | |
331 | * performing writes to it. Must be matched with | |
332 | * mnt_want_write() call above. | |
333 | */ | |
334 | void mnt_drop_write(struct vfsmount *mnt) | |
335 | { | |
d3ef3d73 | 336 | preempt_disable(); |
337 | dec_mnt_writers(mnt); | |
338 | preempt_enable(); | |
8366025e DH |
339 | } |
340 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
341 | ||
2e4b7fcd | 342 | static int mnt_make_readonly(struct vfsmount *mnt) |
8366025e | 343 | { |
3d733633 DH |
344 | int ret = 0; |
345 | ||
d3ef3d73 | 346 | spin_lock(&vfsmount_lock); |
347 | mnt->mnt_flags |= MNT_WRITE_HOLD; | |
3d733633 | 348 | /* |
d3ef3d73 | 349 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
350 | * should be visible before we do. | |
3d733633 | 351 | */ |
d3ef3d73 | 352 | smp_mb(); |
353 | ||
3d733633 | 354 | /* |
d3ef3d73 | 355 | * With writers on hold, if this value is zero, then there are |
356 | * definitely no active writers (although held writers may subsequently | |
357 | * increment the count, they'll have to wait, and decrement it after | |
358 | * seeing MNT_READONLY). | |
359 | * | |
360 | * It is OK to have counter incremented on one CPU and decremented on | |
361 | * another: the sum will add up correctly. The danger would be when we | |
362 | * sum up each counter, if we read a counter before it is incremented, | |
363 | * but then read another CPU's count which it has been subsequently | |
364 | * decremented from -- we would see more decrements than we should. | |
365 | * MNT_WRITE_HOLD protects against this scenario, because | |
366 | * mnt_want_write first increments count, then smp_mb, then spins on | |
367 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
368 | * we're counting up here. | |
3d733633 | 369 | */ |
d3ef3d73 | 370 | if (count_mnt_writers(mnt) > 0) |
371 | ret = -EBUSY; | |
372 | else | |
2e4b7fcd | 373 | mnt->mnt_flags |= MNT_READONLY; |
d3ef3d73 | 374 | /* |
375 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
376 | * that become unheld will see MNT_READONLY. | |
377 | */ | |
378 | smp_wmb(); | |
379 | mnt->mnt_flags &= ~MNT_WRITE_HOLD; | |
2e4b7fcd | 380 | spin_unlock(&vfsmount_lock); |
3d733633 | 381 | return ret; |
8366025e | 382 | } |
8366025e | 383 | |
2e4b7fcd DH |
384 | static void __mnt_unmake_readonly(struct vfsmount *mnt) |
385 | { | |
386 | spin_lock(&vfsmount_lock); | |
387 | mnt->mnt_flags &= ~MNT_READONLY; | |
388 | spin_unlock(&vfsmount_lock); | |
389 | } | |
390 | ||
a3ec947c | 391 | void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
454e2398 DH |
392 | { |
393 | mnt->mnt_sb = sb; | |
394 | mnt->mnt_root = dget(sb->s_root); | |
454e2398 DH |
395 | } |
396 | ||
397 | EXPORT_SYMBOL(simple_set_mnt); | |
398 | ||
1da177e4 LT |
399 | void free_vfsmnt(struct vfsmount *mnt) |
400 | { | |
401 | kfree(mnt->mnt_devname); | |
73cd49ec | 402 | mnt_free_id(mnt); |
d3ef3d73 | 403 | #ifdef CONFIG_SMP |
404 | free_percpu(mnt->mnt_writers); | |
405 | #endif | |
1da177e4 LT |
406 | kmem_cache_free(mnt_cache, mnt); |
407 | } | |
408 | ||
409 | /* | |
a05964f3 RP |
410 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
411 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 412 | */ |
a05964f3 RP |
413 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
414 | int dir) | |
1da177e4 | 415 | { |
b58fed8b RP |
416 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
417 | struct list_head *tmp = head; | |
1da177e4 LT |
418 | struct vfsmount *p, *found = NULL; |
419 | ||
1da177e4 | 420 | for (;;) { |
a05964f3 | 421 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
422 | p = NULL; |
423 | if (tmp == head) | |
424 | break; | |
425 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
426 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 427 | found = p; |
1da177e4 LT |
428 | break; |
429 | } | |
430 | } | |
1da177e4 LT |
431 | return found; |
432 | } | |
433 | ||
a05964f3 RP |
434 | /* |
435 | * lookup_mnt increments the ref count before returning | |
436 | * the vfsmount struct. | |
437 | */ | |
1c755af4 | 438 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 RP |
439 | { |
440 | struct vfsmount *child_mnt; | |
441 | spin_lock(&vfsmount_lock); | |
1c755af4 | 442 | if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) |
a05964f3 RP |
443 | mntget(child_mnt); |
444 | spin_unlock(&vfsmount_lock); | |
445 | return child_mnt; | |
446 | } | |
447 | ||
1da177e4 LT |
448 | static inline int check_mnt(struct vfsmount *mnt) |
449 | { | |
6b3286ed | 450 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
451 | } |
452 | ||
6b3286ed | 453 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
454 | { |
455 | if (ns) { | |
456 | ns->event = ++event; | |
457 | wake_up_interruptible(&ns->poll); | |
458 | } | |
459 | } | |
460 | ||
6b3286ed | 461 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
462 | { |
463 | if (ns && ns->event != event) { | |
464 | ns->event = event; | |
465 | wake_up_interruptible(&ns->poll); | |
466 | } | |
467 | } | |
468 | ||
1a390689 | 469 | static void detach_mnt(struct vfsmount *mnt, struct path *old_path) |
1da177e4 | 470 | { |
1a390689 AV |
471 | old_path->dentry = mnt->mnt_mountpoint; |
472 | old_path->mnt = mnt->mnt_parent; | |
1da177e4 LT |
473 | mnt->mnt_parent = mnt; |
474 | mnt->mnt_mountpoint = mnt->mnt_root; | |
475 | list_del_init(&mnt->mnt_child); | |
476 | list_del_init(&mnt->mnt_hash); | |
1a390689 | 477 | old_path->dentry->d_mounted--; |
1da177e4 LT |
478 | } |
479 | ||
b90fa9ae RP |
480 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
481 | struct vfsmount *child_mnt) | |
482 | { | |
483 | child_mnt->mnt_parent = mntget(mnt); | |
484 | child_mnt->mnt_mountpoint = dget(dentry); | |
485 | dentry->d_mounted++; | |
486 | } | |
487 | ||
1a390689 | 488 | static void attach_mnt(struct vfsmount *mnt, struct path *path) |
1da177e4 | 489 | { |
1a390689 | 490 | mnt_set_mountpoint(path->mnt, path->dentry, mnt); |
b90fa9ae | 491 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 AV |
492 | hash(path->mnt, path->dentry)); |
493 | list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); | |
b90fa9ae RP |
494 | } |
495 | ||
496 | /* | |
497 | * the caller must hold vfsmount_lock | |
498 | */ | |
499 | static void commit_tree(struct vfsmount *mnt) | |
500 | { | |
501 | struct vfsmount *parent = mnt->mnt_parent; | |
502 | struct vfsmount *m; | |
503 | LIST_HEAD(head); | |
6b3286ed | 504 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae RP |
505 | |
506 | BUG_ON(parent == mnt); | |
507 | ||
508 | list_add_tail(&head, &mnt->mnt_list); | |
509 | list_for_each_entry(m, &head, mnt_list) | |
6b3286ed | 510 | m->mnt_ns = n; |
b90fa9ae RP |
511 | list_splice(&head, n->list.prev); |
512 | ||
513 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
514 | hash(parent, mnt->mnt_mountpoint)); | |
515 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
6b3286ed | 516 | touch_mnt_namespace(n); |
1da177e4 LT |
517 | } |
518 | ||
519 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
520 | { | |
521 | struct list_head *next = p->mnt_mounts.next; | |
522 | if (next == &p->mnt_mounts) { | |
523 | while (1) { | |
524 | if (p == root) | |
525 | return NULL; | |
526 | next = p->mnt_child.next; | |
527 | if (next != &p->mnt_parent->mnt_mounts) | |
528 | break; | |
529 | p = p->mnt_parent; | |
530 | } | |
531 | } | |
532 | return list_entry(next, struct vfsmount, mnt_child); | |
533 | } | |
534 | ||
9676f0c6 RP |
535 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
536 | { | |
537 | struct list_head *prev = p->mnt_mounts.prev; | |
538 | while (prev != &p->mnt_mounts) { | |
539 | p = list_entry(prev, struct vfsmount, mnt_child); | |
540 | prev = p->mnt_mounts.prev; | |
541 | } | |
542 | return p; | |
543 | } | |
544 | ||
36341f64 RP |
545 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
546 | int flag) | |
1da177e4 LT |
547 | { |
548 | struct super_block *sb = old->mnt_sb; | |
549 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
550 | ||
551 | if (mnt) { | |
719f5d7f MS |
552 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
553 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
554 | else | |
555 | mnt->mnt_group_id = old->mnt_group_id; | |
556 | ||
557 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
558 | int err = mnt_alloc_group_id(mnt); | |
559 | if (err) | |
560 | goto out_free; | |
561 | } | |
562 | ||
1da177e4 LT |
563 | mnt->mnt_flags = old->mnt_flags; |
564 | atomic_inc(&sb->s_active); | |
565 | mnt->mnt_sb = sb; | |
566 | mnt->mnt_root = dget(root); | |
567 | mnt->mnt_mountpoint = mnt->mnt_root; | |
568 | mnt->mnt_parent = mnt; | |
b90fa9ae | 569 | |
5afe0022 RP |
570 | if (flag & CL_SLAVE) { |
571 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
572 | mnt->mnt_master = old; | |
573 | CLEAR_MNT_SHARED(mnt); | |
8aec0809 | 574 | } else if (!(flag & CL_PRIVATE)) { |
5afe0022 RP |
575 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) |
576 | list_add(&mnt->mnt_share, &old->mnt_share); | |
577 | if (IS_MNT_SLAVE(old)) | |
578 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
579 | mnt->mnt_master = old->mnt_master; | |
580 | } | |
b90fa9ae RP |
581 | if (flag & CL_MAKE_SHARED) |
582 | set_mnt_shared(mnt); | |
1da177e4 LT |
583 | |
584 | /* stick the duplicate mount on the same expiry list | |
585 | * as the original if that was on one */ | |
36341f64 | 586 | if (flag & CL_EXPIRE) { |
36341f64 RP |
587 | if (!list_empty(&old->mnt_expire)) |
588 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 589 | } |
1da177e4 LT |
590 | } |
591 | return mnt; | |
719f5d7f MS |
592 | |
593 | out_free: | |
594 | free_vfsmnt(mnt); | |
595 | return NULL; | |
1da177e4 LT |
596 | } |
597 | ||
7b7b1ace | 598 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
599 | { |
600 | struct super_block *sb = mnt->mnt_sb; | |
3d733633 DH |
601 | /* |
602 | * This probably indicates that somebody messed | |
603 | * up a mnt_want/drop_write() pair. If this | |
604 | * happens, the filesystem was probably unable | |
605 | * to make r/w->r/o transitions. | |
606 | */ | |
d3ef3d73 | 607 | /* |
608 | * atomic_dec_and_lock() used to deal with ->mnt_count decrements | |
609 | * provides barriers, so count_mnt_writers() below is safe. AV | |
610 | */ | |
611 | WARN_ON(count_mnt_writers(mnt)); | |
1da177e4 LT |
612 | dput(mnt->mnt_root); |
613 | free_vfsmnt(mnt); | |
614 | deactivate_super(sb); | |
615 | } | |
616 | ||
7b7b1ace AV |
617 | void mntput_no_expire(struct vfsmount *mnt) |
618 | { | |
619 | repeat: | |
620 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
621 | if (likely(!mnt->mnt_pinned)) { | |
622 | spin_unlock(&vfsmount_lock); | |
623 | __mntput(mnt); | |
624 | return; | |
625 | } | |
626 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
627 | mnt->mnt_pinned = 0; | |
628 | spin_unlock(&vfsmount_lock); | |
629 | acct_auto_close_mnt(mnt); | |
630 | security_sb_umount_close(mnt); | |
631 | goto repeat; | |
632 | } | |
633 | } | |
634 | ||
635 | EXPORT_SYMBOL(mntput_no_expire); | |
636 | ||
637 | void mnt_pin(struct vfsmount *mnt) | |
638 | { | |
639 | spin_lock(&vfsmount_lock); | |
640 | mnt->mnt_pinned++; | |
641 | spin_unlock(&vfsmount_lock); | |
642 | } | |
643 | ||
644 | EXPORT_SYMBOL(mnt_pin); | |
645 | ||
646 | void mnt_unpin(struct vfsmount *mnt) | |
647 | { | |
648 | spin_lock(&vfsmount_lock); | |
649 | if (mnt->mnt_pinned) { | |
650 | atomic_inc(&mnt->mnt_count); | |
651 | mnt->mnt_pinned--; | |
652 | } | |
653 | spin_unlock(&vfsmount_lock); | |
654 | } | |
655 | ||
656 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 | 657 | |
b3b304a2 MS |
658 | static inline void mangle(struct seq_file *m, const char *s) |
659 | { | |
660 | seq_escape(m, s, " \t\n\\"); | |
661 | } | |
662 | ||
663 | /* | |
664 | * Simple .show_options callback for filesystems which don't want to | |
665 | * implement more complex mount option showing. | |
666 | * | |
667 | * See also save_mount_options(). | |
668 | */ | |
669 | int generic_show_options(struct seq_file *m, struct vfsmount *mnt) | |
670 | { | |
2a32cebd AV |
671 | const char *options; |
672 | ||
673 | rcu_read_lock(); | |
674 | options = rcu_dereference(mnt->mnt_sb->s_options); | |
b3b304a2 MS |
675 | |
676 | if (options != NULL && options[0]) { | |
677 | seq_putc(m, ','); | |
678 | mangle(m, options); | |
679 | } | |
2a32cebd | 680 | rcu_read_unlock(); |
b3b304a2 MS |
681 | |
682 | return 0; | |
683 | } | |
684 | EXPORT_SYMBOL(generic_show_options); | |
685 | ||
686 | /* | |
687 | * If filesystem uses generic_show_options(), this function should be | |
688 | * called from the fill_super() callback. | |
689 | * | |
690 | * The .remount_fs callback usually needs to be handled in a special | |
691 | * way, to make sure, that previous options are not overwritten if the | |
692 | * remount fails. | |
693 | * | |
694 | * Also note, that if the filesystem's .remount_fs function doesn't | |
695 | * reset all options to their default value, but changes only newly | |
696 | * given options, then the displayed options will not reflect reality | |
697 | * any more. | |
698 | */ | |
699 | void save_mount_options(struct super_block *sb, char *options) | |
700 | { | |
2a32cebd AV |
701 | BUG_ON(sb->s_options); |
702 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
703 | } |
704 | EXPORT_SYMBOL(save_mount_options); | |
705 | ||
2a32cebd AV |
706 | void replace_mount_options(struct super_block *sb, char *options) |
707 | { | |
708 | char *old = sb->s_options; | |
709 | rcu_assign_pointer(sb->s_options, options); | |
710 | if (old) { | |
711 | synchronize_rcu(); | |
712 | kfree(old); | |
713 | } | |
714 | } | |
715 | EXPORT_SYMBOL(replace_mount_options); | |
716 | ||
a1a2c409 | 717 | #ifdef CONFIG_PROC_FS |
1da177e4 LT |
718 | /* iterator */ |
719 | static void *m_start(struct seq_file *m, loff_t *pos) | |
720 | { | |
a1a2c409 | 721 | struct proc_mounts *p = m->private; |
1da177e4 | 722 | |
390c6843 | 723 | down_read(&namespace_sem); |
a1a2c409 | 724 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
725 | } |
726 | ||
727 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
728 | { | |
a1a2c409 | 729 | struct proc_mounts *p = m->private; |
b0765fb8 | 730 | |
a1a2c409 | 731 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
732 | } |
733 | ||
734 | static void m_stop(struct seq_file *m, void *v) | |
735 | { | |
390c6843 | 736 | up_read(&namespace_sem); |
1da177e4 LT |
737 | } |
738 | ||
2d4d4864 RP |
739 | struct proc_fs_info { |
740 | int flag; | |
741 | const char *str; | |
742 | }; | |
743 | ||
2069f457 | 744 | static int show_sb_opts(struct seq_file *m, struct super_block *sb) |
1da177e4 | 745 | { |
2d4d4864 | 746 | static const struct proc_fs_info fs_info[] = { |
1da177e4 LT |
747 | { MS_SYNCHRONOUS, ",sync" }, |
748 | { MS_DIRSYNC, ",dirsync" }, | |
749 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
750 | { 0, NULL } |
751 | }; | |
2d4d4864 RP |
752 | const struct proc_fs_info *fs_infop; |
753 | ||
754 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
755 | if (sb->s_flags & fs_infop->flag) | |
756 | seq_puts(m, fs_infop->str); | |
757 | } | |
2069f457 EP |
758 | |
759 | return security_sb_show_options(m, sb); | |
2d4d4864 RP |
760 | } |
761 | ||
762 | static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) | |
763 | { | |
764 | static const struct proc_fs_info mnt_info[] = { | |
1da177e4 LT |
765 | { MNT_NOSUID, ",nosuid" }, |
766 | { MNT_NODEV, ",nodev" }, | |
767 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
768 | { MNT_NOATIME, ",noatime" }, |
769 | { MNT_NODIRATIME, ",nodiratime" }, | |
47ae32d6 | 770 | { MNT_RELATIME, ",relatime" }, |
d0adde57 | 771 | { MNT_STRICTATIME, ",strictatime" }, |
1da177e4 LT |
772 | { 0, NULL } |
773 | }; | |
2d4d4864 RP |
774 | const struct proc_fs_info *fs_infop; |
775 | ||
776 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
777 | if (mnt->mnt_flags & fs_infop->flag) | |
778 | seq_puts(m, fs_infop->str); | |
779 | } | |
780 | } | |
781 | ||
782 | static void show_type(struct seq_file *m, struct super_block *sb) | |
783 | { | |
784 | mangle(m, sb->s_type->name); | |
785 | if (sb->s_subtype && sb->s_subtype[0]) { | |
786 | seq_putc(m, '.'); | |
787 | mangle(m, sb->s_subtype); | |
788 | } | |
789 | } | |
790 | ||
791 | static int show_vfsmnt(struct seq_file *m, void *v) | |
792 | { | |
793 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
794 | int err = 0; | |
c32c2f63 | 795 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
1da177e4 LT |
796 | |
797 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
798 | seq_putc(m, ' '); | |
c32c2f63 | 799 | seq_path(m, &mnt_path, " \t\n\\"); |
1da177e4 | 800 | seq_putc(m, ' '); |
2d4d4864 | 801 | show_type(m, mnt->mnt_sb); |
2e4b7fcd | 802 | seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); |
2069f457 EP |
803 | err = show_sb_opts(m, mnt->mnt_sb); |
804 | if (err) | |
805 | goto out; | |
2d4d4864 | 806 | show_mnt_opts(m, mnt); |
1da177e4 LT |
807 | if (mnt->mnt_sb->s_op->show_options) |
808 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
809 | seq_puts(m, " 0 0\n"); | |
2069f457 | 810 | out: |
1da177e4 LT |
811 | return err; |
812 | } | |
813 | ||
a1a2c409 | 814 | const struct seq_operations mounts_op = { |
1da177e4 LT |
815 | .start = m_start, |
816 | .next = m_next, | |
817 | .stop = m_stop, | |
818 | .show = show_vfsmnt | |
819 | }; | |
820 | ||
2d4d4864 RP |
821 | static int show_mountinfo(struct seq_file *m, void *v) |
822 | { | |
823 | struct proc_mounts *p = m->private; | |
824 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); | |
825 | struct super_block *sb = mnt->mnt_sb; | |
826 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; | |
827 | struct path root = p->root; | |
828 | int err = 0; | |
829 | ||
830 | seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, | |
831 | MAJOR(sb->s_dev), MINOR(sb->s_dev)); | |
832 | seq_dentry(m, mnt->mnt_root, " \t\n\\"); | |
833 | seq_putc(m, ' '); | |
834 | seq_path_root(m, &mnt_path, &root, " \t\n\\"); | |
835 | if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { | |
836 | /* | |
837 | * Mountpoint is outside root, discard that one. Ugly, | |
838 | * but less so than trying to do that in iterator in a | |
839 | * race-free way (due to renames). | |
840 | */ | |
841 | return SEQ_SKIP; | |
842 | } | |
843 | seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); | |
844 | show_mnt_opts(m, mnt); | |
845 | ||
846 | /* Tagged fields ("foo:X" or "bar") */ | |
847 | if (IS_MNT_SHARED(mnt)) | |
848 | seq_printf(m, " shared:%i", mnt->mnt_group_id); | |
97e7e0f7 MS |
849 | if (IS_MNT_SLAVE(mnt)) { |
850 | int master = mnt->mnt_master->mnt_group_id; | |
851 | int dom = get_dominating_id(mnt, &p->root); | |
852 | seq_printf(m, " master:%i", master); | |
853 | if (dom && dom != master) | |
854 | seq_printf(m, " propagate_from:%i", dom); | |
855 | } | |
2d4d4864 RP |
856 | if (IS_MNT_UNBINDABLE(mnt)) |
857 | seq_puts(m, " unbindable"); | |
858 | ||
859 | /* Filesystem specific data */ | |
860 | seq_puts(m, " - "); | |
861 | show_type(m, sb); | |
862 | seq_putc(m, ' '); | |
863 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
864 | seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
2069f457 EP |
865 | err = show_sb_opts(m, sb); |
866 | if (err) | |
867 | goto out; | |
2d4d4864 RP |
868 | if (sb->s_op->show_options) |
869 | err = sb->s_op->show_options(m, mnt); | |
870 | seq_putc(m, '\n'); | |
2069f457 | 871 | out: |
2d4d4864 RP |
872 | return err; |
873 | } | |
874 | ||
875 | const struct seq_operations mountinfo_op = { | |
876 | .start = m_start, | |
877 | .next = m_next, | |
878 | .stop = m_stop, | |
879 | .show = show_mountinfo, | |
880 | }; | |
881 | ||
b4629fe2 CL |
882 | static int show_vfsstat(struct seq_file *m, void *v) |
883 | { | |
b0765fb8 | 884 | struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); |
c32c2f63 | 885 | struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; |
b4629fe2 CL |
886 | int err = 0; |
887 | ||
888 | /* device */ | |
889 | if (mnt->mnt_devname) { | |
890 | seq_puts(m, "device "); | |
891 | mangle(m, mnt->mnt_devname); | |
892 | } else | |
893 | seq_puts(m, "no device"); | |
894 | ||
895 | /* mount point */ | |
896 | seq_puts(m, " mounted on "); | |
c32c2f63 | 897 | seq_path(m, &mnt_path, " \t\n\\"); |
b4629fe2 CL |
898 | seq_putc(m, ' '); |
899 | ||
900 | /* file system type */ | |
901 | seq_puts(m, "with fstype "); | |
2d4d4864 | 902 | show_type(m, mnt->mnt_sb); |
b4629fe2 CL |
903 | |
904 | /* optional statistics */ | |
905 | if (mnt->mnt_sb->s_op->show_stats) { | |
906 | seq_putc(m, ' '); | |
907 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
908 | } | |
909 | ||
910 | seq_putc(m, '\n'); | |
911 | return err; | |
912 | } | |
913 | ||
a1a2c409 | 914 | const struct seq_operations mountstats_op = { |
b4629fe2 CL |
915 | .start = m_start, |
916 | .next = m_next, | |
917 | .stop = m_stop, | |
918 | .show = show_vfsstat, | |
919 | }; | |
a1a2c409 | 920 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 921 | |
1da177e4 LT |
922 | /** |
923 | * may_umount_tree - check if a mount tree is busy | |
924 | * @mnt: root of mount tree | |
925 | * | |
926 | * This is called to check if a tree of mounts has any | |
927 | * open files, pwds, chroots or sub mounts that are | |
928 | * busy. | |
929 | */ | |
930 | int may_umount_tree(struct vfsmount *mnt) | |
931 | { | |
36341f64 RP |
932 | int actual_refs = 0; |
933 | int minimum_refs = 0; | |
934 | struct vfsmount *p; | |
1da177e4 LT |
935 | |
936 | spin_lock(&vfsmount_lock); | |
36341f64 | 937 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
938 | actual_refs += atomic_read(&p->mnt_count); |
939 | minimum_refs += 2; | |
1da177e4 LT |
940 | } |
941 | spin_unlock(&vfsmount_lock); | |
942 | ||
943 | if (actual_refs > minimum_refs) | |
e3474a8e | 944 | return 0; |
1da177e4 | 945 | |
e3474a8e | 946 | return 1; |
1da177e4 LT |
947 | } |
948 | ||
949 | EXPORT_SYMBOL(may_umount_tree); | |
950 | ||
951 | /** | |
952 | * may_umount - check if a mount point is busy | |
953 | * @mnt: root of mount | |
954 | * | |
955 | * This is called to check if a mount point has any | |
956 | * open files, pwds, chroots or sub mounts. If the | |
957 | * mount has sub mounts this will return busy | |
958 | * regardless of whether the sub mounts are busy. | |
959 | * | |
960 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
961 | * give false negatives. The main reason why it's here is that we need | |
962 | * a non-destructive way to look for easily umountable filesystems. | |
963 | */ | |
964 | int may_umount(struct vfsmount *mnt) | |
965 | { | |
e3474a8e | 966 | int ret = 1; |
a05964f3 RP |
967 | spin_lock(&vfsmount_lock); |
968 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 969 | ret = 0; |
a05964f3 RP |
970 | spin_unlock(&vfsmount_lock); |
971 | return ret; | |
1da177e4 LT |
972 | } |
973 | ||
974 | EXPORT_SYMBOL(may_umount); | |
975 | ||
b90fa9ae | 976 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
977 | { |
978 | struct vfsmount *mnt; | |
bf066c7d | 979 | while (!list_empty(head)) { |
b5e61818 | 980 | mnt = list_first_entry(head, struct vfsmount, mnt_hash); |
70fbcdf4 RP |
981 | list_del_init(&mnt->mnt_hash); |
982 | if (mnt->mnt_parent != mnt) { | |
983 | struct dentry *dentry; | |
984 | struct vfsmount *m; | |
985 | spin_lock(&vfsmount_lock); | |
986 | dentry = mnt->mnt_mountpoint; | |
987 | m = mnt->mnt_parent; | |
988 | mnt->mnt_mountpoint = mnt->mnt_root; | |
989 | mnt->mnt_parent = mnt; | |
7c4b93d8 | 990 | m->mnt_ghosts--; |
70fbcdf4 RP |
991 | spin_unlock(&vfsmount_lock); |
992 | dput(dentry); | |
993 | mntput(m); | |
994 | } | |
995 | mntput(mnt); | |
996 | } | |
997 | } | |
998 | ||
a05964f3 | 999 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
1000 | { |
1001 | struct vfsmount *p; | |
1da177e4 | 1002 | |
1bfba4e8 AM |
1003 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1004 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 1005 | |
a05964f3 RP |
1006 | if (propagate) |
1007 | propagate_umount(kill); | |
1008 | ||
70fbcdf4 RP |
1009 | list_for_each_entry(p, kill, mnt_hash) { |
1010 | list_del_init(&p->mnt_expire); | |
1011 | list_del_init(&p->mnt_list); | |
6b3286ed KK |
1012 | __touch_mnt_namespace(p->mnt_ns); |
1013 | p->mnt_ns = NULL; | |
70fbcdf4 | 1014 | list_del_init(&p->mnt_child); |
7c4b93d8 AV |
1015 | if (p->mnt_parent != p) { |
1016 | p->mnt_parent->mnt_ghosts++; | |
f30ac319 | 1017 | p->mnt_mountpoint->d_mounted--; |
7c4b93d8 | 1018 | } |
a05964f3 | 1019 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1020 | } |
1021 | } | |
1022 | ||
c35038be AV |
1023 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); |
1024 | ||
1da177e4 LT |
1025 | static int do_umount(struct vfsmount *mnt, int flags) |
1026 | { | |
b58fed8b | 1027 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 1028 | int retval; |
70fbcdf4 | 1029 | LIST_HEAD(umount_list); |
1da177e4 LT |
1030 | |
1031 | retval = security_sb_umount(mnt, flags); | |
1032 | if (retval) | |
1033 | return retval; | |
1034 | ||
1035 | /* | |
1036 | * Allow userspace to request a mountpoint be expired rather than | |
1037 | * unmounting unconditionally. Unmount only happens if: | |
1038 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1039 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1040 | */ | |
1041 | if (flags & MNT_EXPIRE) { | |
6ac08c39 | 1042 | if (mnt == current->fs->root.mnt || |
1da177e4 LT |
1043 | flags & (MNT_FORCE | MNT_DETACH)) |
1044 | return -EINVAL; | |
1045 | ||
1046 | if (atomic_read(&mnt->mnt_count) != 2) | |
1047 | return -EBUSY; | |
1048 | ||
1049 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1050 | return -EAGAIN; | |
1051 | } | |
1052 | ||
1053 | /* | |
1054 | * If we may have to abort operations to get out of this | |
1055 | * mount, and they will themselves hold resources we must | |
1056 | * allow the fs to do things. In the Unix tradition of | |
1057 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1058 | * might fail to complete on the first run through as other tasks | |
1059 | * must return, and the like. Thats for the mount program to worry | |
1060 | * about for the moment. | |
1061 | */ | |
1062 | ||
42faad99 | 1063 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1064 | sb->s_op->umount_begin(sb); |
42faad99 | 1065 | } |
1da177e4 LT |
1066 | |
1067 | /* | |
1068 | * No sense to grab the lock for this test, but test itself looks | |
1069 | * somewhat bogus. Suggestions for better replacement? | |
1070 | * Ho-hum... In principle, we might treat that as umount + switch | |
1071 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1072 | * Actually it makes sense, especially if rootfs would contain a | |
1073 | * /reboot - static binary that would close all descriptors and | |
1074 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1075 | */ | |
6ac08c39 | 1076 | if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1077 | /* |
1078 | * Special case for "unmounting" root ... | |
1079 | * we just try to remount it readonly. | |
1080 | */ | |
1081 | down_write(&sb->s_umount); | |
4aa98cf7 | 1082 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1083 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1084 | up_write(&sb->s_umount); |
1085 | return retval; | |
1086 | } | |
1087 | ||
390c6843 | 1088 | down_write(&namespace_sem); |
1da177e4 | 1089 | spin_lock(&vfsmount_lock); |
5addc5dd | 1090 | event++; |
1da177e4 | 1091 | |
c35038be AV |
1092 | if (!(flags & MNT_DETACH)) |
1093 | shrink_submounts(mnt, &umount_list); | |
1094 | ||
1da177e4 | 1095 | retval = -EBUSY; |
a05964f3 | 1096 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 1097 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 1098 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1099 | retval = 0; |
1100 | } | |
1101 | spin_unlock(&vfsmount_lock); | |
1102 | if (retval) | |
1103 | security_sb_umount_busy(mnt); | |
390c6843 | 1104 | up_write(&namespace_sem); |
70fbcdf4 | 1105 | release_mounts(&umount_list); |
1da177e4 LT |
1106 | return retval; |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Now umount can handle mount points as well as block devices. | |
1111 | * This is important for filesystems which use unnamed block devices. | |
1112 | * | |
1113 | * We now support a flag for forced unmount like the other 'big iron' | |
1114 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1115 | */ | |
1116 | ||
bdc480e3 | 1117 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1118 | { |
2d8f3038 | 1119 | struct path path; |
1da177e4 LT |
1120 | int retval; |
1121 | ||
2d8f3038 | 1122 | retval = user_path(name, &path); |
1da177e4 LT |
1123 | if (retval) |
1124 | goto out; | |
1125 | retval = -EINVAL; | |
2d8f3038 | 1126 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1127 | goto dput_and_out; |
2d8f3038 | 1128 | if (!check_mnt(path.mnt)) |
1da177e4 LT |
1129 | goto dput_and_out; |
1130 | ||
1131 | retval = -EPERM; | |
1132 | if (!capable(CAP_SYS_ADMIN)) | |
1133 | goto dput_and_out; | |
1134 | ||
2d8f3038 | 1135 | retval = do_umount(path.mnt, flags); |
1da177e4 | 1136 | dput_and_out: |
429731b1 | 1137 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 AV |
1138 | dput(path.dentry); |
1139 | mntput_no_expire(path.mnt); | |
1da177e4 LT |
1140 | out: |
1141 | return retval; | |
1142 | } | |
1143 | ||
1144 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1145 | ||
1146 | /* | |
b58fed8b | 1147 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1148 | */ |
bdc480e3 | 1149 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1150 | { |
b58fed8b | 1151 | return sys_umount(name, 0); |
1da177e4 LT |
1152 | } |
1153 | ||
1154 | #endif | |
1155 | ||
2d92ab3c | 1156 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1157 | { |
1158 | if (capable(CAP_SYS_ADMIN)) | |
1159 | return 0; | |
1160 | return -EPERM; | |
1161 | #ifdef notyet | |
2d92ab3c | 1162 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1163 | return -EPERM; |
2d92ab3c | 1164 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1165 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1166 | return -EPERM; |
1167 | } | |
2d92ab3c | 1168 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1169 | return -EPERM; |
1170 | return 0; | |
1171 | #endif | |
1172 | } | |
1173 | ||
b90fa9ae | 1174 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 1175 | int flag) |
1da177e4 LT |
1176 | { |
1177 | struct vfsmount *res, *p, *q, *r, *s; | |
1a390689 | 1178 | struct path path; |
1da177e4 | 1179 | |
9676f0c6 RP |
1180 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
1181 | return NULL; | |
1182 | ||
36341f64 | 1183 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1184 | if (!q) |
1185 | goto Enomem; | |
1186 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1187 | ||
1188 | p = mnt; | |
fdadd65f | 1189 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
7ec02ef1 | 1190 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1191 | continue; |
1192 | ||
1193 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
1194 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
1195 | s = skip_mnt_tree(s); | |
1196 | continue; | |
1197 | } | |
1da177e4 LT |
1198 | while (p != s->mnt_parent) { |
1199 | p = p->mnt_parent; | |
1200 | q = q->mnt_parent; | |
1201 | } | |
1202 | p = s; | |
1a390689 AV |
1203 | path.mnt = q; |
1204 | path.dentry = p->mnt_mountpoint; | |
36341f64 | 1205 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
1206 | if (!q) |
1207 | goto Enomem; | |
1208 | spin_lock(&vfsmount_lock); | |
1209 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1a390689 | 1210 | attach_mnt(q, &path); |
1da177e4 LT |
1211 | spin_unlock(&vfsmount_lock); |
1212 | } | |
1213 | } | |
1214 | return res; | |
b58fed8b | 1215 | Enomem: |
1da177e4 | 1216 | if (res) { |
70fbcdf4 | 1217 | LIST_HEAD(umount_list); |
1da177e4 | 1218 | spin_lock(&vfsmount_lock); |
a05964f3 | 1219 | umount_tree(res, 0, &umount_list); |
1da177e4 | 1220 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1221 | release_mounts(&umount_list); |
1da177e4 LT |
1222 | } |
1223 | return NULL; | |
1224 | } | |
1225 | ||
589ff870 | 1226 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 AV |
1227 | { |
1228 | struct vfsmount *tree; | |
1a60a280 | 1229 | down_write(&namespace_sem); |
589ff870 | 1230 | tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); |
1a60a280 | 1231 | up_write(&namespace_sem); |
8aec0809 AV |
1232 | return tree; |
1233 | } | |
1234 | ||
1235 | void drop_collected_mounts(struct vfsmount *mnt) | |
1236 | { | |
1237 | LIST_HEAD(umount_list); | |
1a60a280 | 1238 | down_write(&namespace_sem); |
8aec0809 AV |
1239 | spin_lock(&vfsmount_lock); |
1240 | umount_tree(mnt, 0, &umount_list); | |
1241 | spin_unlock(&vfsmount_lock); | |
1a60a280 | 1242 | up_write(&namespace_sem); |
8aec0809 AV |
1243 | release_mounts(&umount_list); |
1244 | } | |
1245 | ||
719f5d7f MS |
1246 | static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) |
1247 | { | |
1248 | struct vfsmount *p; | |
1249 | ||
1250 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1251 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1252 | mnt_release_group_id(p); | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | static int invent_group_ids(struct vfsmount *mnt, bool recurse) | |
1257 | { | |
1258 | struct vfsmount *p; | |
1259 | ||
1260 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1261 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1262 | int err = mnt_alloc_group_id(p); | |
1263 | if (err) { | |
1264 | cleanup_group_ids(mnt, p); | |
1265 | return err; | |
1266 | } | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | return 0; | |
1271 | } | |
1272 | ||
b90fa9ae RP |
1273 | /* |
1274 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1275 | * @nd : place the mount tree @source_mnt is attached |
1276 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1277 | * store the parent mount and mountpoint dentry. | |
1278 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1279 | * |
1280 | * NOTE: in the table below explains the semantics when a source mount | |
1281 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1282 | * --------------------------------------------------------------------------- |
1283 | * | BIND MOUNT OPERATION | | |
1284 | * |************************************************************************** | |
1285 | * | source-->| shared | private | slave | unbindable | | |
1286 | * | dest | | | | | | |
1287 | * | | | | | | | | |
1288 | * | v | | | | | | |
1289 | * |************************************************************************** | |
1290 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1291 | * | | | | | | | |
1292 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1293 | * *************************************************************************** | |
b90fa9ae RP |
1294 | * A bind operation clones the source mount and mounts the clone on the |
1295 | * destination mount. | |
1296 | * | |
1297 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1298 | * tree of the destination mount and the cloned mount is added to | |
1299 | * the peer group of the source mount. | |
1300 | * (+) the cloned mount is created under the destination mount and is marked | |
1301 | * as shared. The cloned mount is added to the peer group of the source | |
1302 | * mount. | |
5afe0022 RP |
1303 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1304 | * of the destination mount and the cloned mount is made slave | |
1305 | * of the same master as that of the source mount. The cloned mount | |
1306 | * is marked as 'shared and slave'. | |
1307 | * (*) the cloned mount is made a slave of the same master as that of the | |
1308 | * source mount. | |
1309 | * | |
9676f0c6 RP |
1310 | * --------------------------------------------------------------------------- |
1311 | * | MOVE MOUNT OPERATION | | |
1312 | * |************************************************************************** | |
1313 | * | source-->| shared | private | slave | unbindable | | |
1314 | * | dest | | | | | | |
1315 | * | | | | | | | | |
1316 | * | v | | | | | | |
1317 | * |************************************************************************** | |
1318 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1319 | * | | | | | | | |
1320 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1321 | * *************************************************************************** | |
5afe0022 RP |
1322 | * |
1323 | * (+) the mount is moved to the destination. And is then propagated to | |
1324 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1325 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1326 | * (+++) the mount is moved to the destination and is then propagated to |
1327 | * all the mounts belonging to the destination mount's propagation tree. | |
1328 | * the mount is marked as 'shared and slave'. | |
1329 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1330 | * |
1331 | * if the source mount is a tree, the operations explained above is | |
1332 | * applied to each mount in the tree. | |
1333 | * Must be called without spinlocks held, since this function can sleep | |
1334 | * in allocations. | |
1335 | */ | |
1336 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
1a390689 | 1337 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1338 | { |
1339 | LIST_HEAD(tree_list); | |
1a390689 AV |
1340 | struct vfsmount *dest_mnt = path->mnt; |
1341 | struct dentry *dest_dentry = path->dentry; | |
b90fa9ae | 1342 | struct vfsmount *child, *p; |
719f5d7f | 1343 | int err; |
b90fa9ae | 1344 | |
719f5d7f MS |
1345 | if (IS_MNT_SHARED(dest_mnt)) { |
1346 | err = invent_group_ids(source_mnt, true); | |
1347 | if (err) | |
1348 | goto out; | |
1349 | } | |
1350 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); | |
1351 | if (err) | |
1352 | goto out_cleanup_ids; | |
b90fa9ae RP |
1353 | |
1354 | if (IS_MNT_SHARED(dest_mnt)) { | |
1355 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1356 | set_mnt_shared(p); | |
1357 | } | |
1358 | ||
1359 | spin_lock(&vfsmount_lock); | |
1a390689 AV |
1360 | if (parent_path) { |
1361 | detach_mnt(source_mnt, parent_path); | |
1362 | attach_mnt(source_mnt, path); | |
e5d67f07 | 1363 | touch_mnt_namespace(parent_path->mnt->mnt_ns); |
21444403 RP |
1364 | } else { |
1365 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
1366 | commit_tree(source_mnt); | |
1367 | } | |
b90fa9ae RP |
1368 | |
1369 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
1370 | list_del_init(&child->mnt_hash); | |
1371 | commit_tree(child); | |
1372 | } | |
1373 | spin_unlock(&vfsmount_lock); | |
1374 | return 0; | |
719f5d7f MS |
1375 | |
1376 | out_cleanup_ids: | |
1377 | if (IS_MNT_SHARED(dest_mnt)) | |
1378 | cleanup_group_ids(source_mnt, NULL); | |
1379 | out: | |
1380 | return err; | |
b90fa9ae RP |
1381 | } |
1382 | ||
8c3ee42e | 1383 | static int graft_tree(struct vfsmount *mnt, struct path *path) |
1da177e4 LT |
1384 | { |
1385 | int err; | |
1386 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
1387 | return -EINVAL; | |
1388 | ||
8c3ee42e | 1389 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1da177e4 LT |
1390 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) |
1391 | return -ENOTDIR; | |
1392 | ||
1393 | err = -ENOENT; | |
8c3ee42e AV |
1394 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1395 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1396 | goto out_unlock; |
1397 | ||
8c3ee42e | 1398 | err = security_sb_check_sb(mnt, path); |
1da177e4 LT |
1399 | if (err) |
1400 | goto out_unlock; | |
1401 | ||
1402 | err = -ENOENT; | |
f3da392e | 1403 | if (!d_unlinked(path->dentry)) |
8c3ee42e | 1404 | err = attach_recursive_mnt(mnt, path, NULL); |
1da177e4 | 1405 | out_unlock: |
8c3ee42e | 1406 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1407 | if (!err) |
8c3ee42e | 1408 | security_sb_post_addmount(mnt, path); |
1da177e4 LT |
1409 | return err; |
1410 | } | |
1411 | ||
07b20889 RP |
1412 | /* |
1413 | * recursively change the type of the mountpoint. | |
1414 | */ | |
0a0d8a46 | 1415 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1416 | { |
2d92ab3c | 1417 | struct vfsmount *m, *mnt = path->mnt; |
07b20889 RP |
1418 | int recurse = flag & MS_REC; |
1419 | int type = flag & ~MS_REC; | |
719f5d7f | 1420 | int err = 0; |
07b20889 | 1421 | |
ee6f9582 MS |
1422 | if (!capable(CAP_SYS_ADMIN)) |
1423 | return -EPERM; | |
1424 | ||
2d92ab3c | 1425 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1426 | return -EINVAL; |
1427 | ||
1428 | down_write(&namespace_sem); | |
719f5d7f MS |
1429 | if (type == MS_SHARED) { |
1430 | err = invent_group_ids(mnt, recurse); | |
1431 | if (err) | |
1432 | goto out_unlock; | |
1433 | } | |
1434 | ||
07b20889 RP |
1435 | spin_lock(&vfsmount_lock); |
1436 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1437 | change_mnt_propagation(m, type); | |
1438 | spin_unlock(&vfsmount_lock); | |
719f5d7f MS |
1439 | |
1440 | out_unlock: | |
07b20889 | 1441 | up_write(&namespace_sem); |
719f5d7f | 1442 | return err; |
07b20889 RP |
1443 | } |
1444 | ||
1da177e4 LT |
1445 | /* |
1446 | * do loopback mount. | |
1447 | */ | |
0a0d8a46 | 1448 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1449 | int recurse) |
1da177e4 | 1450 | { |
2d92ab3c | 1451 | struct path old_path; |
1da177e4 | 1452 | struct vfsmount *mnt = NULL; |
2d92ab3c | 1453 | int err = mount_is_safe(path); |
1da177e4 LT |
1454 | if (err) |
1455 | return err; | |
1456 | if (!old_name || !*old_name) | |
1457 | return -EINVAL; | |
2d92ab3c | 1458 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1459 | if (err) |
1460 | return err; | |
1461 | ||
390c6843 | 1462 | down_write(&namespace_sem); |
1da177e4 | 1463 | err = -EINVAL; |
2d92ab3c | 1464 | if (IS_MNT_UNBINDABLE(old_path.mnt)) |
4ac91378 | 1465 | goto out; |
9676f0c6 | 1466 | |
2d92ab3c | 1467 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
ccd48bc7 | 1468 | goto out; |
1da177e4 | 1469 | |
ccd48bc7 AV |
1470 | err = -ENOMEM; |
1471 | if (recurse) | |
2d92ab3c | 1472 | mnt = copy_tree(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 | 1473 | else |
2d92ab3c | 1474 | mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); |
ccd48bc7 AV |
1475 | |
1476 | if (!mnt) | |
1477 | goto out; | |
1478 | ||
2d92ab3c | 1479 | err = graft_tree(mnt, path); |
ccd48bc7 | 1480 | if (err) { |
70fbcdf4 | 1481 | LIST_HEAD(umount_list); |
1da177e4 | 1482 | spin_lock(&vfsmount_lock); |
a05964f3 | 1483 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 1484 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 1485 | release_mounts(&umount_list); |
5b83d2c5 | 1486 | } |
1da177e4 | 1487 | |
ccd48bc7 | 1488 | out: |
390c6843 | 1489 | up_write(&namespace_sem); |
2d92ab3c | 1490 | path_put(&old_path); |
1da177e4 LT |
1491 | return err; |
1492 | } | |
1493 | ||
2e4b7fcd DH |
1494 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1495 | { | |
1496 | int error = 0; | |
1497 | int readonly_request = 0; | |
1498 | ||
1499 | if (ms_flags & MS_RDONLY) | |
1500 | readonly_request = 1; | |
1501 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1502 | return 0; | |
1503 | ||
1504 | if (readonly_request) | |
1505 | error = mnt_make_readonly(mnt); | |
1506 | else | |
1507 | __mnt_unmake_readonly(mnt); | |
1508 | return error; | |
1509 | } | |
1510 | ||
1da177e4 LT |
1511 | /* |
1512 | * change filesystem flags. dir should be a physical root of filesystem. | |
1513 | * If you've mounted a non-root directory somewhere and want to do remount | |
1514 | * on it - tough luck. | |
1515 | */ | |
0a0d8a46 | 1516 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1517 | void *data) |
1518 | { | |
1519 | int err; | |
2d92ab3c | 1520 | struct super_block *sb = path->mnt->mnt_sb; |
1da177e4 LT |
1521 | |
1522 | if (!capable(CAP_SYS_ADMIN)) | |
1523 | return -EPERM; | |
1524 | ||
2d92ab3c | 1525 | if (!check_mnt(path->mnt)) |
1da177e4 LT |
1526 | return -EINVAL; |
1527 | ||
2d92ab3c | 1528 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1529 | return -EINVAL; |
1530 | ||
1531 | down_write(&sb->s_umount); | |
2e4b7fcd | 1532 | if (flags & MS_BIND) |
2d92ab3c | 1533 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1534 | else |
2e4b7fcd | 1535 | err = do_remount_sb(sb, flags, data, 0); |
1da177e4 | 1536 | if (!err) |
2d92ab3c | 1537 | path->mnt->mnt_flags = mnt_flags; |
1da177e4 | 1538 | up_write(&sb->s_umount); |
0e55a7cc | 1539 | if (!err) { |
2d92ab3c | 1540 | security_sb_post_remount(path->mnt, flags, data); |
0e55a7cc DW |
1541 | |
1542 | spin_lock(&vfsmount_lock); | |
1543 | touch_mnt_namespace(path->mnt->mnt_ns); | |
1544 | spin_unlock(&vfsmount_lock); | |
1545 | } | |
1da177e4 LT |
1546 | return err; |
1547 | } | |
1548 | ||
9676f0c6 RP |
1549 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
1550 | { | |
1551 | struct vfsmount *p; | |
1552 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1553 | if (IS_MNT_UNBINDABLE(p)) | |
1554 | return 1; | |
1555 | } | |
1556 | return 0; | |
1557 | } | |
1558 | ||
0a0d8a46 | 1559 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1560 | { |
2d92ab3c | 1561 | struct path old_path, parent_path; |
1da177e4 LT |
1562 | struct vfsmount *p; |
1563 | int err = 0; | |
1564 | if (!capable(CAP_SYS_ADMIN)) | |
1565 | return -EPERM; | |
1566 | if (!old_name || !*old_name) | |
1567 | return -EINVAL; | |
2d92ab3c | 1568 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1569 | if (err) |
1570 | return err; | |
1571 | ||
390c6843 | 1572 | down_write(&namespace_sem); |
2d92ab3c | 1573 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1574 | follow_down(path)) |
1da177e4 LT |
1575 | ; |
1576 | err = -EINVAL; | |
2d92ab3c | 1577 | if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) |
1da177e4 LT |
1578 | goto out; |
1579 | ||
1580 | err = -ENOENT; | |
2d92ab3c AV |
1581 | mutex_lock(&path->dentry->d_inode->i_mutex); |
1582 | if (IS_DEADDIR(path->dentry->d_inode)) | |
1da177e4 LT |
1583 | goto out1; |
1584 | ||
f3da392e | 1585 | if (d_unlinked(path->dentry)) |
21444403 | 1586 | goto out1; |
1da177e4 LT |
1587 | |
1588 | err = -EINVAL; | |
2d92ab3c | 1589 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1590 | goto out1; |
1da177e4 | 1591 | |
2d92ab3c | 1592 | if (old_path.mnt == old_path.mnt->mnt_parent) |
21444403 | 1593 | goto out1; |
1da177e4 | 1594 | |
2d92ab3c AV |
1595 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1596 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1597 | goto out1; |
1598 | /* | |
1599 | * Don't move a mount residing in a shared parent. | |
1600 | */ | |
2d92ab3c AV |
1601 | if (old_path.mnt->mnt_parent && |
1602 | IS_MNT_SHARED(old_path.mnt->mnt_parent)) | |
21444403 | 1603 | goto out1; |
9676f0c6 RP |
1604 | /* |
1605 | * Don't move a mount tree containing unbindable mounts to a destination | |
1606 | * mount which is shared. | |
1607 | */ | |
2d92ab3c AV |
1608 | if (IS_MNT_SHARED(path->mnt) && |
1609 | tree_contains_unbindable(old_path.mnt)) | |
9676f0c6 | 1610 | goto out1; |
1da177e4 | 1611 | err = -ELOOP; |
2d92ab3c AV |
1612 | for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1613 | if (p == old_path.mnt) | |
21444403 | 1614 | goto out1; |
1da177e4 | 1615 | |
2d92ab3c | 1616 | err = attach_recursive_mnt(old_path.mnt, path, &parent_path); |
4ac91378 | 1617 | if (err) |
21444403 | 1618 | goto out1; |
1da177e4 LT |
1619 | |
1620 | /* if the mount is moved, it should no longer be expire | |
1621 | * automatically */ | |
2d92ab3c | 1622 | list_del_init(&old_path.mnt->mnt_expire); |
1da177e4 | 1623 | out1: |
2d92ab3c | 1624 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1da177e4 | 1625 | out: |
390c6843 | 1626 | up_write(&namespace_sem); |
1da177e4 | 1627 | if (!err) |
1a390689 | 1628 | path_put(&parent_path); |
2d92ab3c | 1629 | path_put(&old_path); |
1da177e4 LT |
1630 | return err; |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * create a new mount for userspace and request it to be added into the | |
1635 | * namespace's tree | |
1636 | */ | |
0a0d8a46 | 1637 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1638 | int mnt_flags, char *name, void *data) |
1639 | { | |
1640 | struct vfsmount *mnt; | |
1641 | ||
1642 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1643 | return -EINVAL; | |
1644 | ||
1645 | /* we need capabilities... */ | |
1646 | if (!capable(CAP_SYS_ADMIN)) | |
1647 | return -EPERM; | |
1648 | ||
7f78d4cd | 1649 | lock_kernel(); |
1da177e4 | 1650 | mnt = do_kern_mount(type, flags, name, data); |
7f78d4cd | 1651 | unlock_kernel(); |
1da177e4 LT |
1652 | if (IS_ERR(mnt)) |
1653 | return PTR_ERR(mnt); | |
1654 | ||
2d92ab3c | 1655 | return do_add_mount(mnt, path, mnt_flags, NULL); |
1da177e4 LT |
1656 | } |
1657 | ||
1658 | /* | |
1659 | * add a mount into a namespace's mount tree | |
1660 | * - provide the option of adding the new mount to an expiration list | |
1661 | */ | |
8d66bf54 | 1662 | int do_add_mount(struct vfsmount *newmnt, struct path *path, |
1da177e4 LT |
1663 | int mnt_flags, struct list_head *fslist) |
1664 | { | |
1665 | int err; | |
1666 | ||
390c6843 | 1667 | down_write(&namespace_sem); |
1da177e4 | 1668 | /* Something was mounted here while we slept */ |
8d66bf54 | 1669 | while (d_mountpoint(path->dentry) && |
9393bd07 | 1670 | follow_down(path)) |
1da177e4 LT |
1671 | ; |
1672 | err = -EINVAL; | |
dd5cae6e | 1673 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) |
1da177e4 LT |
1674 | goto unlock; |
1675 | ||
1676 | /* Refuse the same filesystem on the same mount point */ | |
1677 | err = -EBUSY; | |
8d66bf54 AV |
1678 | if (path->mnt->mnt_sb == newmnt->mnt_sb && |
1679 | path->mnt->mnt_root == path->dentry) | |
1da177e4 LT |
1680 | goto unlock; |
1681 | ||
1682 | err = -EINVAL; | |
1683 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1684 | goto unlock; | |
1685 | ||
1686 | newmnt->mnt_flags = mnt_flags; | |
8d66bf54 | 1687 | if ((err = graft_tree(newmnt, path))) |
5b83d2c5 | 1688 | goto unlock; |
1da177e4 | 1689 | |
6758f953 | 1690 | if (fslist) /* add to the specified expiration list */ |
55e700b9 | 1691 | list_add_tail(&newmnt->mnt_expire, fslist); |
6758f953 | 1692 | |
390c6843 | 1693 | up_write(&namespace_sem); |
5b83d2c5 | 1694 | return 0; |
1da177e4 LT |
1695 | |
1696 | unlock: | |
390c6843 | 1697 | up_write(&namespace_sem); |
1da177e4 LT |
1698 | mntput(newmnt); |
1699 | return err; | |
1700 | } | |
1701 | ||
1702 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1703 | ||
1704 | /* | |
1705 | * process a list of expirable mountpoints with the intent of discarding any | |
1706 | * mountpoints that aren't in use and haven't been touched since last we came | |
1707 | * here | |
1708 | */ | |
1709 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1710 | { | |
1da177e4 LT |
1711 | struct vfsmount *mnt, *next; |
1712 | LIST_HEAD(graveyard); | |
bcc5c7d2 | 1713 | LIST_HEAD(umounts); |
1da177e4 LT |
1714 | |
1715 | if (list_empty(mounts)) | |
1716 | return; | |
1717 | ||
bcc5c7d2 | 1718 | down_write(&namespace_sem); |
1da177e4 LT |
1719 | spin_lock(&vfsmount_lock); |
1720 | ||
1721 | /* extract from the expiration list every vfsmount that matches the | |
1722 | * following criteria: | |
1723 | * - only referenced by its parent vfsmount | |
1724 | * - still marked for expiry (marked on the last call here; marks are | |
1725 | * cleared by mntput()) | |
1726 | */ | |
55e700b9 | 1727 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 | 1728 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
bcc5c7d2 | 1729 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1730 | continue; |
55e700b9 | 1731 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1732 | } |
bcc5c7d2 AV |
1733 | while (!list_empty(&graveyard)) { |
1734 | mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); | |
1735 | touch_mnt_namespace(mnt->mnt_ns); | |
1736 | umount_tree(mnt, 1, &umounts); | |
1737 | } | |
5528f911 | 1738 | spin_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1739 | up_write(&namespace_sem); |
1740 | ||
1741 | release_mounts(&umounts); | |
5528f911 TM |
1742 | } |
1743 | ||
1744 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1745 | ||
1746 | /* | |
1747 | * Ripoff of 'select_parent()' | |
1748 | * | |
1749 | * search the list of submounts for a given mountpoint, and move any | |
1750 | * shrinkable submounts to the 'graveyard' list. | |
1751 | */ | |
1752 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1753 | { | |
1754 | struct vfsmount *this_parent = parent; | |
1755 | struct list_head *next; | |
1756 | int found = 0; | |
1757 | ||
1758 | repeat: | |
1759 | next = this_parent->mnt_mounts.next; | |
1760 | resume: | |
1761 | while (next != &this_parent->mnt_mounts) { | |
1762 | struct list_head *tmp = next; | |
1763 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1764 | ||
1765 | next = tmp->next; | |
1766 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1767 | continue; |
5528f911 TM |
1768 | /* |
1769 | * Descend a level if the d_mounts list is non-empty. | |
1770 | */ | |
1771 | if (!list_empty(&mnt->mnt_mounts)) { | |
1772 | this_parent = mnt; | |
1773 | goto repeat; | |
1774 | } | |
1da177e4 | 1775 | |
5528f911 | 1776 | if (!propagate_mount_busy(mnt, 1)) { |
5528f911 TM |
1777 | list_move_tail(&mnt->mnt_expire, graveyard); |
1778 | found++; | |
1779 | } | |
1da177e4 | 1780 | } |
5528f911 TM |
1781 | /* |
1782 | * All done at this level ... ascend and resume the search | |
1783 | */ | |
1784 | if (this_parent != parent) { | |
1785 | next = this_parent->mnt_child.next; | |
1786 | this_parent = this_parent->mnt_parent; | |
1787 | goto resume; | |
1788 | } | |
1789 | return found; | |
1790 | } | |
1791 | ||
1792 | /* | |
1793 | * process a list of expirable mountpoints with the intent of discarding any | |
1794 | * submounts of a specific parent mountpoint | |
1795 | */ | |
c35038be | 1796 | static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) |
5528f911 TM |
1797 | { |
1798 | LIST_HEAD(graveyard); | |
c35038be | 1799 | struct vfsmount *m; |
5528f911 | 1800 | |
5528f911 | 1801 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1802 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1803 | while (!list_empty(&graveyard)) { |
c35038be | 1804 | m = list_first_entry(&graveyard, struct vfsmount, |
bcc5c7d2 | 1805 | mnt_expire); |
afef80b3 EB |
1806 | touch_mnt_namespace(m->mnt_ns); |
1807 | umount_tree(m, 1, umounts); | |
bcc5c7d2 AV |
1808 | } |
1809 | } | |
1da177e4 LT |
1810 | } |
1811 | ||
1da177e4 LT |
1812 | /* |
1813 | * Some copy_from_user() implementations do not return the exact number of | |
1814 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1815 | * Note that this function differs from copy_from_user() in that it will oops | |
1816 | * on bad values of `to', rather than returning a short copy. | |
1817 | */ | |
b58fed8b RP |
1818 | static long exact_copy_from_user(void *to, const void __user * from, |
1819 | unsigned long n) | |
1da177e4 LT |
1820 | { |
1821 | char *t = to; | |
1822 | const char __user *f = from; | |
1823 | char c; | |
1824 | ||
1825 | if (!access_ok(VERIFY_READ, from, n)) | |
1826 | return n; | |
1827 | ||
1828 | while (n) { | |
1829 | if (__get_user(c, f)) { | |
1830 | memset(t, 0, n); | |
1831 | break; | |
1832 | } | |
1833 | *t++ = c; | |
1834 | f++; | |
1835 | n--; | |
1836 | } | |
1837 | return n; | |
1838 | } | |
1839 | ||
b58fed8b | 1840 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1841 | { |
1842 | int i; | |
1843 | unsigned long page; | |
1844 | unsigned long size; | |
b58fed8b | 1845 | |
1da177e4 LT |
1846 | *where = 0; |
1847 | if (!data) | |
1848 | return 0; | |
1849 | ||
1850 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1851 | return -ENOMEM; | |
1852 | ||
1853 | /* We only care that *some* data at the address the user | |
1854 | * gave us is valid. Just in case, we'll zero | |
1855 | * the remainder of the page. | |
1856 | */ | |
1857 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1858 | size = TASK_SIZE - (unsigned long)data; | |
1859 | if (size > PAGE_SIZE) | |
1860 | size = PAGE_SIZE; | |
1861 | ||
1862 | i = size - exact_copy_from_user((void *)page, data, size); | |
1863 | if (!i) { | |
b58fed8b | 1864 | free_page(page); |
1da177e4 LT |
1865 | return -EFAULT; |
1866 | } | |
1867 | if (i != PAGE_SIZE) | |
1868 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1869 | *where = page; | |
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | /* | |
1874 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1875 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1876 | * | |
1877 | * data is a (void *) that can point to any structure up to | |
1878 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1879 | * information (or be NULL). | |
1880 | * | |
1881 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1882 | * When the flags word was introduced its top half was required | |
1883 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1884 | * Therefore, if this magic number is present, it carries no information | |
1885 | * and must be discarded. | |
1886 | */ | |
b58fed8b | 1887 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1888 | unsigned long flags, void *data_page) |
1889 | { | |
2d92ab3c | 1890 | struct path path; |
1da177e4 LT |
1891 | int retval = 0; |
1892 | int mnt_flags = 0; | |
1893 | ||
1894 | /* Discard magic */ | |
1895 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1896 | flags &= ~MS_MGC_MSK; | |
1897 | ||
1898 | /* Basic sanity checks */ | |
1899 | ||
1900 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1901 | return -EINVAL; | |
1902 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1903 | return -EINVAL; | |
1904 | ||
1905 | if (data_page) | |
1906 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1907 | ||
613cbe3d AK |
1908 | /* Default to relatime unless overriden */ |
1909 | if (!(flags & MS_NOATIME)) | |
1910 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 1911 | |
1da177e4 LT |
1912 | /* Separate the per-mountpoint flags */ |
1913 | if (flags & MS_NOSUID) | |
1914 | mnt_flags |= MNT_NOSUID; | |
1915 | if (flags & MS_NODEV) | |
1916 | mnt_flags |= MNT_NODEV; | |
1917 | if (flags & MS_NOEXEC) | |
1918 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1919 | if (flags & MS_NOATIME) |
1920 | mnt_flags |= MNT_NOATIME; | |
1921 | if (flags & MS_NODIRATIME) | |
1922 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
1923 | if (flags & MS_STRICTATIME) |
1924 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
1925 | if (flags & MS_RDONLY) |
1926 | mnt_flags |= MNT_READONLY; | |
fc33a7bb CH |
1927 | |
1928 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
d0adde57 MG |
1929 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
1930 | MS_STRICTATIME); | |
1da177e4 LT |
1931 | |
1932 | /* ... and get the mountpoint */ | |
2d92ab3c | 1933 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1934 | if (retval) |
1935 | return retval; | |
1936 | ||
2d92ab3c | 1937 | retval = security_sb_mount(dev_name, &path, |
b5266eb4 | 1938 | type_page, flags, data_page); |
1da177e4 LT |
1939 | if (retval) |
1940 | goto dput_out; | |
1941 | ||
1942 | if (flags & MS_REMOUNT) | |
2d92ab3c | 1943 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
1944 | data_page); |
1945 | else if (flags & MS_BIND) | |
2d92ab3c | 1946 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 1947 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 1948 | retval = do_change_type(&path, flags); |
1da177e4 | 1949 | else if (flags & MS_MOVE) |
2d92ab3c | 1950 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 1951 | else |
2d92ab3c | 1952 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
1953 | dev_name, data_page); |
1954 | dput_out: | |
2d92ab3c | 1955 | path_put(&path); |
1da177e4 LT |
1956 | return retval; |
1957 | } | |
1958 | ||
cf8d2c11 TM |
1959 | static struct mnt_namespace *alloc_mnt_ns(void) |
1960 | { | |
1961 | struct mnt_namespace *new_ns; | |
1962 | ||
1963 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
1964 | if (!new_ns) | |
1965 | return ERR_PTR(-ENOMEM); | |
1966 | atomic_set(&new_ns->count, 1); | |
1967 | new_ns->root = NULL; | |
1968 | INIT_LIST_HEAD(&new_ns->list); | |
1969 | init_waitqueue_head(&new_ns->poll); | |
1970 | new_ns->event = 0; | |
1971 | return new_ns; | |
1972 | } | |
1973 | ||
741a2951 JD |
1974 | /* |
1975 | * Allocate a new namespace structure and populate it with contents | |
1976 | * copied from the namespace of the passed in task structure. | |
1977 | */ | |
e3222c4e | 1978 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 1979 | struct fs_struct *fs) |
1da177e4 | 1980 | { |
6b3286ed | 1981 | struct mnt_namespace *new_ns; |
7f2da1e7 | 1982 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
1da177e4 LT |
1983 | struct vfsmount *p, *q; |
1984 | ||
cf8d2c11 TM |
1985 | new_ns = alloc_mnt_ns(); |
1986 | if (IS_ERR(new_ns)) | |
1987 | return new_ns; | |
1da177e4 | 1988 | |
390c6843 | 1989 | down_write(&namespace_sem); |
1da177e4 | 1990 | /* First pass: copy the tree topology */ |
6b3286ed | 1991 | new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, |
9676f0c6 | 1992 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1993 | if (!new_ns->root) { |
390c6843 | 1994 | up_write(&namespace_sem); |
1da177e4 | 1995 | kfree(new_ns); |
5cc4a034 | 1996 | return ERR_PTR(-ENOMEM); |
1da177e4 LT |
1997 | } |
1998 | spin_lock(&vfsmount_lock); | |
1999 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
2000 | spin_unlock(&vfsmount_lock); | |
2001 | ||
2002 | /* | |
2003 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2004 | * as belonging to new namespace. We have already acquired a private | |
2005 | * fs_struct, so tsk->fs->lock is not needed. | |
2006 | */ | |
6b3286ed | 2007 | p = mnt_ns->root; |
1da177e4 LT |
2008 | q = new_ns->root; |
2009 | while (p) { | |
6b3286ed | 2010 | q->mnt_ns = new_ns; |
1da177e4 | 2011 | if (fs) { |
6ac08c39 | 2012 | if (p == fs->root.mnt) { |
1da177e4 | 2013 | rootmnt = p; |
6ac08c39 | 2014 | fs->root.mnt = mntget(q); |
1da177e4 | 2015 | } |
6ac08c39 | 2016 | if (p == fs->pwd.mnt) { |
1da177e4 | 2017 | pwdmnt = p; |
6ac08c39 | 2018 | fs->pwd.mnt = mntget(q); |
1da177e4 | 2019 | } |
1da177e4 | 2020 | } |
6b3286ed | 2021 | p = next_mnt(p, mnt_ns->root); |
1da177e4 LT |
2022 | q = next_mnt(q, new_ns->root); |
2023 | } | |
390c6843 | 2024 | up_write(&namespace_sem); |
1da177e4 | 2025 | |
1da177e4 LT |
2026 | if (rootmnt) |
2027 | mntput(rootmnt); | |
2028 | if (pwdmnt) | |
2029 | mntput(pwdmnt); | |
1da177e4 | 2030 | |
741a2951 JD |
2031 | return new_ns; |
2032 | } | |
2033 | ||
213dd266 | 2034 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2035 | struct fs_struct *new_fs) |
741a2951 | 2036 | { |
6b3286ed | 2037 | struct mnt_namespace *new_ns; |
741a2951 | 2038 | |
e3222c4e | 2039 | BUG_ON(!ns); |
6b3286ed | 2040 | get_mnt_ns(ns); |
741a2951 JD |
2041 | |
2042 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2043 | return ns; |
741a2951 | 2044 | |
e3222c4e | 2045 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2046 | |
6b3286ed | 2047 | put_mnt_ns(ns); |
e3222c4e | 2048 | return new_ns; |
1da177e4 LT |
2049 | } |
2050 | ||
cf8d2c11 TM |
2051 | /** |
2052 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2053 | * @mnt: pointer to the new root filesystem mountpoint | |
2054 | */ | |
2055 | struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) | |
2056 | { | |
2057 | struct mnt_namespace *new_ns; | |
2058 | ||
2059 | new_ns = alloc_mnt_ns(); | |
2060 | if (!IS_ERR(new_ns)) { | |
2061 | mnt->mnt_ns = new_ns; | |
2062 | new_ns->root = mnt; | |
2063 | list_add(&new_ns->list, &new_ns->root->mnt_list); | |
2064 | } | |
2065 | return new_ns; | |
2066 | } | |
2067 | EXPORT_SYMBOL(create_mnt_ns); | |
2068 | ||
bdc480e3 HC |
2069 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2070 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 LT |
2071 | { |
2072 | int retval; | |
2073 | unsigned long data_page; | |
2074 | unsigned long type_page; | |
2075 | unsigned long dev_page; | |
2076 | char *dir_page; | |
2077 | ||
b58fed8b | 2078 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
2079 | if (retval < 0) |
2080 | return retval; | |
2081 | ||
2082 | dir_page = getname(dir_name); | |
2083 | retval = PTR_ERR(dir_page); | |
2084 | if (IS_ERR(dir_page)) | |
2085 | goto out1; | |
2086 | ||
b58fed8b | 2087 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
2088 | if (retval < 0) |
2089 | goto out2; | |
2090 | ||
b58fed8b | 2091 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
2092 | if (retval < 0) |
2093 | goto out3; | |
2094 | ||
b58fed8b RP |
2095 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
2096 | flags, (void *)data_page); | |
1da177e4 LT |
2097 | free_page(data_page); |
2098 | ||
2099 | out3: | |
2100 | free_page(dev_page); | |
2101 | out2: | |
2102 | putname(dir_page); | |
2103 | out1: | |
2104 | free_page(type_page); | |
2105 | return retval; | |
2106 | } | |
2107 | ||
1da177e4 LT |
2108 | /* |
2109 | * pivot_root Semantics: | |
2110 | * Moves the root file system of the current process to the directory put_old, | |
2111 | * makes new_root as the new root file system of the current process, and sets | |
2112 | * root/cwd of all processes which had them on the current root to new_root. | |
2113 | * | |
2114 | * Restrictions: | |
2115 | * The new_root and put_old must be directories, and must not be on the | |
2116 | * same file system as the current process root. The put_old must be | |
2117 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2118 | * pointed to by put_old must yield the same directory as new_root. No other | |
2119 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2120 | * | |
4a0d11fa NB |
2121 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2122 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2123 | * in this situation. | |
2124 | * | |
1da177e4 LT |
2125 | * Notes: |
2126 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2127 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2128 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2129 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2130 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2131 | * first. | |
2132 | */ | |
3480b257 HC |
2133 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2134 | const char __user *, put_old) | |
1da177e4 LT |
2135 | { |
2136 | struct vfsmount *tmp; | |
2d8f3038 | 2137 | struct path new, old, parent_path, root_parent, root; |
1da177e4 LT |
2138 | int error; |
2139 | ||
2140 | if (!capable(CAP_SYS_ADMIN)) | |
2141 | return -EPERM; | |
2142 | ||
2d8f3038 | 2143 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2144 | if (error) |
2145 | goto out0; | |
2146 | error = -EINVAL; | |
2d8f3038 | 2147 | if (!check_mnt(new.mnt)) |
1da177e4 LT |
2148 | goto out1; |
2149 | ||
2d8f3038 | 2150 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2151 | if (error) |
2152 | goto out1; | |
2153 | ||
2d8f3038 | 2154 | error = security_sb_pivotroot(&old, &new); |
1da177e4 | 2155 | if (error) { |
2d8f3038 | 2156 | path_put(&old); |
1da177e4 LT |
2157 | goto out1; |
2158 | } | |
2159 | ||
2160 | read_lock(¤t->fs->lock); | |
8c3ee42e | 2161 | root = current->fs->root; |
6ac08c39 | 2162 | path_get(¤t->fs->root); |
1da177e4 | 2163 | read_unlock(¤t->fs->lock); |
390c6843 | 2164 | down_write(&namespace_sem); |
2d8f3038 | 2165 | mutex_lock(&old.dentry->d_inode->i_mutex); |
1da177e4 | 2166 | error = -EINVAL; |
2d8f3038 AV |
2167 | if (IS_MNT_SHARED(old.mnt) || |
2168 | IS_MNT_SHARED(new.mnt->mnt_parent) || | |
8c3ee42e | 2169 | IS_MNT_SHARED(root.mnt->mnt_parent)) |
21444403 | 2170 | goto out2; |
8c3ee42e | 2171 | if (!check_mnt(root.mnt)) |
1da177e4 LT |
2172 | goto out2; |
2173 | error = -ENOENT; | |
2d8f3038 | 2174 | if (IS_DEADDIR(new.dentry->d_inode)) |
1da177e4 | 2175 | goto out2; |
f3da392e | 2176 | if (d_unlinked(new.dentry)) |
1da177e4 | 2177 | goto out2; |
f3da392e | 2178 | if (d_unlinked(old.dentry)) |
1da177e4 LT |
2179 | goto out2; |
2180 | error = -EBUSY; | |
2d8f3038 AV |
2181 | if (new.mnt == root.mnt || |
2182 | old.mnt == root.mnt) | |
1da177e4 LT |
2183 | goto out2; /* loop, on the same file system */ |
2184 | error = -EINVAL; | |
8c3ee42e | 2185 | if (root.mnt->mnt_root != root.dentry) |
1da177e4 | 2186 | goto out2; /* not a mountpoint */ |
8c3ee42e | 2187 | if (root.mnt->mnt_parent == root.mnt) |
0bb6fcc1 | 2188 | goto out2; /* not attached */ |
2d8f3038 | 2189 | if (new.mnt->mnt_root != new.dentry) |
1da177e4 | 2190 | goto out2; /* not a mountpoint */ |
2d8f3038 | 2191 | if (new.mnt->mnt_parent == new.mnt) |
0bb6fcc1 | 2192 | goto out2; /* not attached */ |
4ac91378 | 2193 | /* make sure we can reach put_old from new_root */ |
2d8f3038 | 2194 | tmp = old.mnt; |
1da177e4 | 2195 | spin_lock(&vfsmount_lock); |
2d8f3038 | 2196 | if (tmp != new.mnt) { |
1da177e4 LT |
2197 | for (;;) { |
2198 | if (tmp->mnt_parent == tmp) | |
2199 | goto out3; /* already mounted on put_old */ | |
2d8f3038 | 2200 | if (tmp->mnt_parent == new.mnt) |
1da177e4 LT |
2201 | break; |
2202 | tmp = tmp->mnt_parent; | |
2203 | } | |
2d8f3038 | 2204 | if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) |
1da177e4 | 2205 | goto out3; |
2d8f3038 | 2206 | } else if (!is_subdir(old.dentry, new.dentry)) |
1da177e4 | 2207 | goto out3; |
2d8f3038 | 2208 | detach_mnt(new.mnt, &parent_path); |
8c3ee42e | 2209 | detach_mnt(root.mnt, &root_parent); |
4ac91378 | 2210 | /* mount old root on put_old */ |
2d8f3038 | 2211 | attach_mnt(root.mnt, &old); |
4ac91378 | 2212 | /* mount new_root on / */ |
2d8f3038 | 2213 | attach_mnt(new.mnt, &root_parent); |
6b3286ed | 2214 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
1da177e4 | 2215 | spin_unlock(&vfsmount_lock); |
2d8f3038 AV |
2216 | chroot_fs_refs(&root, &new); |
2217 | security_sb_post_pivotroot(&root, &new); | |
1da177e4 | 2218 | error = 0; |
1a390689 AV |
2219 | path_put(&root_parent); |
2220 | path_put(&parent_path); | |
1da177e4 | 2221 | out2: |
2d8f3038 | 2222 | mutex_unlock(&old.dentry->d_inode->i_mutex); |
390c6843 | 2223 | up_write(&namespace_sem); |
8c3ee42e | 2224 | path_put(&root); |
2d8f3038 | 2225 | path_put(&old); |
1da177e4 | 2226 | out1: |
2d8f3038 | 2227 | path_put(&new); |
1da177e4 | 2228 | out0: |
1da177e4 LT |
2229 | return error; |
2230 | out3: | |
2231 | spin_unlock(&vfsmount_lock); | |
2232 | goto out2; | |
2233 | } | |
2234 | ||
2235 | static void __init init_mount_tree(void) | |
2236 | { | |
2237 | struct vfsmount *mnt; | |
6b3286ed | 2238 | struct mnt_namespace *ns; |
ac748a09 | 2239 | struct path root; |
1da177e4 LT |
2240 | |
2241 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2242 | if (IS_ERR(mnt)) | |
2243 | panic("Can't create rootfs"); | |
3b22edc5 TM |
2244 | ns = create_mnt_ns(mnt); |
2245 | if (IS_ERR(ns)) | |
1da177e4 | 2246 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2247 | |
2248 | init_task.nsproxy->mnt_ns = ns; | |
2249 | get_mnt_ns(ns); | |
2250 | ||
ac748a09 JB |
2251 | root.mnt = ns->root; |
2252 | root.dentry = ns->root->mnt_root; | |
2253 | ||
2254 | set_fs_pwd(current->fs, &root); | |
2255 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2256 | } |
2257 | ||
74bf17cf | 2258 | void __init mnt_init(void) |
1da177e4 | 2259 | { |
13f14b4d | 2260 | unsigned u; |
15a67dd8 | 2261 | int err; |
1da177e4 | 2262 | |
390c6843 RP |
2263 | init_rwsem(&namespace_sem); |
2264 | ||
1da177e4 | 2265 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
20c2df83 | 2266 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2267 | |
b58fed8b | 2268 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2269 | |
2270 | if (!mount_hashtable) | |
2271 | panic("Failed to allocate mount hash table\n"); | |
2272 | ||
13f14b4d ED |
2273 | printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); |
2274 | ||
2275 | for (u = 0; u < HASH_SIZE; u++) | |
2276 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2277 | |
15a67dd8 RD |
2278 | err = sysfs_init(); |
2279 | if (err) | |
2280 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2281 | __func__, err); |
00d26666 GKH |
2282 | fs_kobj = kobject_create_and_add("fs", NULL); |
2283 | if (!fs_kobj) | |
8e24eea7 | 2284 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2285 | init_rootfs(); |
2286 | init_mount_tree(); | |
2287 | } | |
2288 | ||
616511d0 | 2289 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2290 | { |
616511d0 | 2291 | struct vfsmount *root; |
70fbcdf4 | 2292 | LIST_HEAD(umount_list); |
616511d0 TM |
2293 | |
2294 | if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock)) | |
2295 | return; | |
2296 | root = ns->root; | |
6b3286ed | 2297 | ns->root = NULL; |
1ce88cf4 | 2298 | spin_unlock(&vfsmount_lock); |
390c6843 | 2299 | down_write(&namespace_sem); |
1da177e4 | 2300 | spin_lock(&vfsmount_lock); |
a05964f3 | 2301 | umount_tree(root, 0, &umount_list); |
1da177e4 | 2302 | spin_unlock(&vfsmount_lock); |
390c6843 | 2303 | up_write(&namespace_sem); |
70fbcdf4 | 2304 | release_mounts(&umount_list); |
6b3286ed | 2305 | kfree(ns); |
1da177e4 | 2306 | } |
cf8d2c11 | 2307 | EXPORT_SYMBOL(put_mnt_ns); |