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 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
1da177e4 LT |
15 | #include <linux/namei.h> |
16 | #include <linux/security.h> | |
73cd49ec | 17 | #include <linux/idr.h> |
d10577a8 AV |
18 | #include <linux/acct.h> /* acct_auto_close_mnt */ |
19 | #include <linux/ramfs.h> /* init_rootfs */ | |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ | |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
07b20889 | 23 | #include "pnode.h" |
948730b0 | 24 | #include "internal.h" |
1da177e4 | 25 | |
13f14b4d ED |
26 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
27 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
28 | ||
5addc5dd | 29 | static int event; |
73cd49ec | 30 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 31 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 32 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
33 | static int mnt_id_start = 0; |
34 | static int mnt_group_start = 1; | |
1da177e4 | 35 | |
fa3536cc | 36 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 37 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 38 | static struct rw_semaphore namespace_sem; |
1da177e4 | 39 | |
f87fd4c2 | 40 | /* /sys/fs */ |
00d26666 GKH |
41 | struct kobject *fs_kobj; |
42 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 43 | |
99b7db7b NP |
44 | /* |
45 | * vfsmount lock may be taken for read to prevent changes to the | |
46 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
47 | * up the tree. | |
48 | * | |
49 | * It should be taken for write in all cases where the vfsmount | |
50 | * tree or hash is modified or when a vfsmount structure is modified. | |
51 | */ | |
52 | DEFINE_BRLOCK(vfsmount_lock); | |
53 | ||
1da177e4 LT |
54 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
55 | { | |
b58fed8b RP |
56 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
57 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
58 | tmp = tmp + (tmp >> HASH_SHIFT); |
59 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
60 | } |
61 | ||
3d733633 DH |
62 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
63 | ||
99b7db7b NP |
64 | /* |
65 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
66 | * serialize with freeing. | |
67 | */ | |
b105e270 | 68 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
69 | { |
70 | int res; | |
71 | ||
72 | retry: | |
73 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 74 | spin_lock(&mnt_id_lock); |
15169fe7 | 75 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 76 | if (!res) |
15169fe7 | 77 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 78 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
79 | if (res == -EAGAIN) |
80 | goto retry; | |
81 | ||
82 | return res; | |
83 | } | |
84 | ||
b105e270 | 85 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 86 | { |
15169fe7 | 87 | int id = mnt->mnt_id; |
99b7db7b | 88 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
89 | ida_remove(&mnt_id_ida, id); |
90 | if (mnt_id_start > id) | |
91 | mnt_id_start = id; | |
99b7db7b | 92 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
93 | } |
94 | ||
719f5d7f MS |
95 | /* |
96 | * Allocate a new peer group ID | |
97 | * | |
98 | * mnt_group_ida is protected by namespace_sem | |
99 | */ | |
4b8b21f4 | 100 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 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, | |
15169fe7 | 109 | &mnt->mnt_group_id); |
f21f6220 | 110 | if (!res) |
15169fe7 | 111 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
112 | |
113 | return res; | |
719f5d7f MS |
114 | } |
115 | ||
116 | /* | |
117 | * Release a peer group ID | |
118 | */ | |
4b8b21f4 | 119 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 120 | { |
15169fe7 | 121 | int id = mnt->mnt_group_id; |
f21f6220 AV |
122 | ida_remove(&mnt_group_ida, id); |
123 | if (mnt_group_start > id) | |
124 | mnt_group_start = id; | |
15169fe7 | 125 | mnt->mnt_group_id = 0; |
719f5d7f MS |
126 | } |
127 | ||
b3e19d92 NP |
128 | /* |
129 | * vfsmount lock must be held for read | |
130 | */ | |
83adc753 | 131 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
132 | { |
133 | #ifdef CONFIG_SMP | |
68e8a9fe | 134 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
135 | #else |
136 | preempt_disable(); | |
68e8a9fe | 137 | mnt->mnt_count += n; |
b3e19d92 NP |
138 | preempt_enable(); |
139 | #endif | |
140 | } | |
141 | ||
b3e19d92 NP |
142 | /* |
143 | * vfsmount lock must be held for write | |
144 | */ | |
83adc753 | 145 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
146 | { |
147 | #ifdef CONFIG_SMP | |
f03c6599 | 148 | unsigned int count = 0; |
b3e19d92 NP |
149 | int cpu; |
150 | ||
151 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 152 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
153 | } |
154 | ||
155 | return count; | |
156 | #else | |
68e8a9fe | 157 | return mnt->mnt_count; |
b3e19d92 NP |
158 | #endif |
159 | } | |
160 | ||
b105e270 | 161 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 162 | { |
c63181e6 AV |
163 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
164 | if (mnt) { | |
73cd49ec MS |
165 | int err; |
166 | ||
c63181e6 | 167 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
168 | if (err) |
169 | goto out_free_cache; | |
170 | ||
171 | if (name) { | |
c63181e6 AV |
172 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
173 | if (!mnt->mnt_devname) | |
88b38782 | 174 | goto out_free_id; |
73cd49ec MS |
175 | } |
176 | ||
b3e19d92 | 177 | #ifdef CONFIG_SMP |
c63181e6 AV |
178 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
179 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
180 | goto out_free_devname; |
181 | ||
c63181e6 | 182 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 183 | #else |
c63181e6 AV |
184 | mnt->mnt_count = 1; |
185 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
186 | #endif |
187 | ||
c63181e6 AV |
188 | INIT_LIST_HEAD(&mnt->mnt_hash); |
189 | INIT_LIST_HEAD(&mnt->mnt_child); | |
190 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
191 | INIT_LIST_HEAD(&mnt->mnt_list); | |
192 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
193 | INIT_LIST_HEAD(&mnt->mnt_share); | |
194 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
195 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
196 | #ifdef CONFIG_FSNOTIFY |
197 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 198 | #endif |
1da177e4 | 199 | } |
c63181e6 | 200 | return mnt; |
88b38782 | 201 | |
d3ef3d73 | 202 | #ifdef CONFIG_SMP |
203 | out_free_devname: | |
c63181e6 | 204 | kfree(mnt->mnt_devname); |
d3ef3d73 | 205 | #endif |
88b38782 | 206 | out_free_id: |
c63181e6 | 207 | mnt_free_id(mnt); |
88b38782 | 208 | out_free_cache: |
c63181e6 | 209 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 210 | return NULL; |
1da177e4 LT |
211 | } |
212 | ||
3d733633 DH |
213 | /* |
214 | * Most r/o checks on a fs are for operations that take | |
215 | * discrete amounts of time, like a write() or unlink(). | |
216 | * We must keep track of when those operations start | |
217 | * (for permission checks) and when they end, so that | |
218 | * we can determine when writes are able to occur to | |
219 | * a filesystem. | |
220 | */ | |
221 | /* | |
222 | * __mnt_is_readonly: check whether a mount is read-only | |
223 | * @mnt: the mount to check for its write status | |
224 | * | |
225 | * This shouldn't be used directly ouside of the VFS. | |
226 | * It does not guarantee that the filesystem will stay | |
227 | * r/w, just that it is right *now*. This can not and | |
228 | * should not be used in place of IS_RDONLY(inode). | |
229 | * mnt_want/drop_write() will _keep_ the filesystem | |
230 | * r/w. | |
231 | */ | |
232 | int __mnt_is_readonly(struct vfsmount *mnt) | |
233 | { | |
2e4b7fcd DH |
234 | if (mnt->mnt_flags & MNT_READONLY) |
235 | return 1; | |
236 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
237 | return 1; | |
238 | return 0; | |
3d733633 DH |
239 | } |
240 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
241 | ||
83adc753 | 242 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 243 | { |
244 | #ifdef CONFIG_SMP | |
68e8a9fe | 245 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 246 | #else |
68e8a9fe | 247 | mnt->mnt_writers++; |
d3ef3d73 | 248 | #endif |
249 | } | |
3d733633 | 250 | |
83adc753 | 251 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 252 | { |
d3ef3d73 | 253 | #ifdef CONFIG_SMP |
68e8a9fe | 254 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 255 | #else |
68e8a9fe | 256 | mnt->mnt_writers--; |
d3ef3d73 | 257 | #endif |
3d733633 | 258 | } |
3d733633 | 259 | |
83adc753 | 260 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 261 | { |
d3ef3d73 | 262 | #ifdef CONFIG_SMP |
263 | unsigned int count = 0; | |
3d733633 | 264 | int cpu; |
3d733633 DH |
265 | |
266 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 267 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 268 | } |
3d733633 | 269 | |
d3ef3d73 | 270 | return count; |
271 | #else | |
272 | return mnt->mnt_writers; | |
273 | #endif | |
3d733633 DH |
274 | } |
275 | ||
8366025e DH |
276 | /* |
277 | * Most r/o checks on a fs are for operations that take | |
278 | * discrete amounts of time, like a write() or unlink(). | |
279 | * We must keep track of when those operations start | |
280 | * (for permission checks) and when they end, so that | |
281 | * we can determine when writes are able to occur to | |
282 | * a filesystem. | |
283 | */ | |
284 | /** | |
285 | * mnt_want_write - get write access to a mount | |
83adc753 | 286 | * @m: the mount on which to take a write |
8366025e DH |
287 | * |
288 | * This tells the low-level filesystem that a write is | |
289 | * about to be performed to it, and makes sure that | |
290 | * writes are allowed before returning success. When | |
291 | * the write operation is finished, mnt_drop_write() | |
292 | * must be called. This is effectively a refcount. | |
293 | */ | |
83adc753 | 294 | int mnt_want_write(struct vfsmount *m) |
8366025e | 295 | { |
83adc753 | 296 | struct mount *mnt = real_mount(m); |
3d733633 | 297 | int ret = 0; |
3d733633 | 298 | |
d3ef3d73 | 299 | preempt_disable(); |
c6653a83 | 300 | mnt_inc_writers(mnt); |
d3ef3d73 | 301 | /* |
c6653a83 | 302 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 303 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
304 | * incremented count after it has set MNT_WRITE_HOLD. | |
305 | */ | |
306 | smp_mb(); | |
83adc753 | 307 | while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) |
d3ef3d73 | 308 | cpu_relax(); |
309 | /* | |
310 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
311 | * be set to match its requirements. So we must not load that until | |
312 | * MNT_WRITE_HOLD is cleared. | |
313 | */ | |
314 | smp_rmb(); | |
83adc753 | 315 | if (__mnt_is_readonly(m)) { |
c6653a83 | 316 | mnt_dec_writers(mnt); |
3d733633 DH |
317 | ret = -EROFS; |
318 | goto out; | |
319 | } | |
3d733633 | 320 | out: |
d3ef3d73 | 321 | preempt_enable(); |
3d733633 | 322 | return ret; |
8366025e DH |
323 | } |
324 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
325 | ||
96029c4e | 326 | /** |
327 | * mnt_clone_write - get write access to a mount | |
328 | * @mnt: the mount on which to take a write | |
329 | * | |
330 | * This is effectively like mnt_want_write, except | |
331 | * it must only be used to take an extra write reference | |
332 | * on a mountpoint that we already know has a write reference | |
333 | * on it. This allows some optimisation. | |
334 | * | |
335 | * After finished, mnt_drop_write must be called as usual to | |
336 | * drop the reference. | |
337 | */ | |
338 | int mnt_clone_write(struct vfsmount *mnt) | |
339 | { | |
340 | /* superblock may be r/o */ | |
341 | if (__mnt_is_readonly(mnt)) | |
342 | return -EROFS; | |
343 | preempt_disable(); | |
83adc753 | 344 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 345 | preempt_enable(); |
346 | return 0; | |
347 | } | |
348 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
349 | ||
350 | /** | |
351 | * mnt_want_write_file - get write access to a file's mount | |
352 | * @file: the file who's mount on which to take a write | |
353 | * | |
354 | * This is like mnt_want_write, but it takes a file and can | |
355 | * do some optimisations if the file is open for write already | |
356 | */ | |
357 | int mnt_want_write_file(struct file *file) | |
358 | { | |
2d8dd38a OH |
359 | struct inode *inode = file->f_dentry->d_inode; |
360 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) | |
96029c4e | 361 | return mnt_want_write(file->f_path.mnt); |
362 | else | |
363 | return mnt_clone_write(file->f_path.mnt); | |
364 | } | |
365 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
366 | ||
8366025e DH |
367 | /** |
368 | * mnt_drop_write - give up write access to a mount | |
369 | * @mnt: the mount on which to give up write access | |
370 | * | |
371 | * Tells the low-level filesystem that we are done | |
372 | * performing writes to it. Must be matched with | |
373 | * mnt_want_write() call above. | |
374 | */ | |
375 | void mnt_drop_write(struct vfsmount *mnt) | |
376 | { | |
d3ef3d73 | 377 | preempt_disable(); |
83adc753 | 378 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 379 | preempt_enable(); |
8366025e DH |
380 | } |
381 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
382 | ||
2a79f17e AV |
383 | void mnt_drop_write_file(struct file *file) |
384 | { | |
385 | mnt_drop_write(file->f_path.mnt); | |
386 | } | |
387 | EXPORT_SYMBOL(mnt_drop_write_file); | |
388 | ||
83adc753 | 389 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 390 | { |
3d733633 DH |
391 | int ret = 0; |
392 | ||
99b7db7b | 393 | br_write_lock(vfsmount_lock); |
83adc753 | 394 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 395 | /* |
d3ef3d73 | 396 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
397 | * should be visible before we do. | |
3d733633 | 398 | */ |
d3ef3d73 | 399 | smp_mb(); |
400 | ||
3d733633 | 401 | /* |
d3ef3d73 | 402 | * With writers on hold, if this value is zero, then there are |
403 | * definitely no active writers (although held writers may subsequently | |
404 | * increment the count, they'll have to wait, and decrement it after | |
405 | * seeing MNT_READONLY). | |
406 | * | |
407 | * It is OK to have counter incremented on one CPU and decremented on | |
408 | * another: the sum will add up correctly. The danger would be when we | |
409 | * sum up each counter, if we read a counter before it is incremented, | |
410 | * but then read another CPU's count which it has been subsequently | |
411 | * decremented from -- we would see more decrements than we should. | |
412 | * MNT_WRITE_HOLD protects against this scenario, because | |
413 | * mnt_want_write first increments count, then smp_mb, then spins on | |
414 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
415 | * we're counting up here. | |
3d733633 | 416 | */ |
c6653a83 | 417 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 418 | ret = -EBUSY; |
419 | else | |
83adc753 | 420 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 421 | /* |
422 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
423 | * that become unheld will see MNT_READONLY. | |
424 | */ | |
425 | smp_wmb(); | |
83adc753 | 426 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
99b7db7b | 427 | br_write_unlock(vfsmount_lock); |
3d733633 | 428 | return ret; |
8366025e | 429 | } |
8366025e | 430 | |
83adc753 | 431 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 432 | { |
99b7db7b | 433 | br_write_lock(vfsmount_lock); |
83adc753 | 434 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
99b7db7b | 435 | br_write_unlock(vfsmount_lock); |
2e4b7fcd DH |
436 | } |
437 | ||
b105e270 | 438 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 439 | { |
52ba1621 | 440 | kfree(mnt->mnt_devname); |
73cd49ec | 441 | mnt_free_id(mnt); |
d3ef3d73 | 442 | #ifdef CONFIG_SMP |
68e8a9fe | 443 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 444 | #endif |
b105e270 | 445 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
446 | } |
447 | ||
448 | /* | |
a05964f3 RP |
449 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
450 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 451 | * vfsmount_lock must be held for read or write. |
1da177e4 | 452 | */ |
c7105365 | 453 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 454 | int dir) |
1da177e4 | 455 | { |
b58fed8b RP |
456 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
457 | struct list_head *tmp = head; | |
c7105365 | 458 | struct mount *p, *found = NULL; |
1da177e4 | 459 | |
1da177e4 | 460 | for (;;) { |
a05964f3 | 461 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
462 | p = NULL; |
463 | if (tmp == head) | |
464 | break; | |
1b8e5564 | 465 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 466 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 467 | found = p; |
1da177e4 LT |
468 | break; |
469 | } | |
470 | } | |
1da177e4 LT |
471 | return found; |
472 | } | |
473 | ||
a05964f3 RP |
474 | /* |
475 | * lookup_mnt increments the ref count before returning | |
476 | * the vfsmount struct. | |
477 | */ | |
1c755af4 | 478 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 479 | { |
c7105365 | 480 | struct mount *child_mnt; |
99b7db7b NP |
481 | |
482 | br_read_lock(vfsmount_lock); | |
c7105365 AV |
483 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
484 | if (child_mnt) { | |
485 | mnt_add_count(child_mnt, 1); | |
486 | br_read_unlock(vfsmount_lock); | |
487 | return &child_mnt->mnt; | |
488 | } else { | |
489 | br_read_unlock(vfsmount_lock); | |
490 | return NULL; | |
491 | } | |
a05964f3 RP |
492 | } |
493 | ||
143c8c91 | 494 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 495 | { |
6b3286ed | 496 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
497 | } |
498 | ||
99b7db7b NP |
499 | /* |
500 | * vfsmount lock must be held for write | |
501 | */ | |
6b3286ed | 502 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
503 | { |
504 | if (ns) { | |
505 | ns->event = ++event; | |
506 | wake_up_interruptible(&ns->poll); | |
507 | } | |
508 | } | |
509 | ||
99b7db7b NP |
510 | /* |
511 | * vfsmount lock must be held for write | |
512 | */ | |
6b3286ed | 513 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
514 | { |
515 | if (ns && ns->event != event) { | |
516 | ns->event = event; | |
517 | wake_up_interruptible(&ns->poll); | |
518 | } | |
519 | } | |
520 | ||
5f57cbcc NP |
521 | /* |
522 | * Clear dentry's mounted state if it has no remaining mounts. | |
523 | * vfsmount_lock must be held for write. | |
524 | */ | |
aa0a4cf0 | 525 | static void dentry_reset_mounted(struct dentry *dentry) |
5f57cbcc NP |
526 | { |
527 | unsigned u; | |
528 | ||
529 | for (u = 0; u < HASH_SIZE; u++) { | |
d5e50f74 | 530 | struct mount *p; |
5f57cbcc | 531 | |
1b8e5564 | 532 | list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { |
a73324da | 533 | if (p->mnt_mountpoint == dentry) |
5f57cbcc NP |
534 | return; |
535 | } | |
536 | } | |
537 | spin_lock(&dentry->d_lock); | |
538 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
539 | spin_unlock(&dentry->d_lock); | |
540 | } | |
541 | ||
99b7db7b NP |
542 | /* |
543 | * vfsmount lock must be held for write | |
544 | */ | |
419148da AV |
545 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
546 | { | |
a73324da | 547 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
548 | old_path->mnt = &mnt->mnt_parent->mnt; |
549 | mnt->mnt_parent = mnt; | |
a73324da | 550 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 551 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 552 | list_del_init(&mnt->mnt_hash); |
aa0a4cf0 | 553 | dentry_reset_mounted(old_path->dentry); |
1da177e4 LT |
554 | } |
555 | ||
99b7db7b NP |
556 | /* |
557 | * vfsmount lock must be held for write | |
558 | */ | |
14cf1fa8 | 559 | void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry, |
44d964d6 | 560 | struct mount *child_mnt) |
b90fa9ae | 561 | { |
3a2393d7 | 562 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
a73324da | 563 | child_mnt->mnt_mountpoint = dget(dentry); |
3a2393d7 | 564 | child_mnt->mnt_parent = mnt; |
5f57cbcc NP |
565 | spin_lock(&dentry->d_lock); |
566 | dentry->d_flags |= DCACHE_MOUNTED; | |
567 | spin_unlock(&dentry->d_lock); | |
b90fa9ae RP |
568 | } |
569 | ||
99b7db7b NP |
570 | /* |
571 | * vfsmount lock must be held for write | |
572 | */ | |
419148da | 573 | static void attach_mnt(struct mount *mnt, struct path *path) |
1da177e4 | 574 | { |
14cf1fa8 | 575 | mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt); |
1b8e5564 | 576 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 | 577 | hash(path->mnt, path->dentry)); |
6b41d536 | 578 | list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts); |
b90fa9ae RP |
579 | } |
580 | ||
83adc753 | 581 | static inline void __mnt_make_longterm(struct mount *mnt) |
7e3d0eb0 AV |
582 | { |
583 | #ifdef CONFIG_SMP | |
68e8a9fe | 584 | atomic_inc(&mnt->mnt_longterm); |
7e3d0eb0 AV |
585 | #endif |
586 | } | |
587 | ||
588 | /* needs vfsmount lock for write */ | |
83adc753 | 589 | static inline void __mnt_make_shortterm(struct mount *mnt) |
7e3d0eb0 AV |
590 | { |
591 | #ifdef CONFIG_SMP | |
68e8a9fe | 592 | atomic_dec(&mnt->mnt_longterm); |
7e3d0eb0 AV |
593 | #endif |
594 | } | |
595 | ||
b90fa9ae | 596 | /* |
99b7db7b | 597 | * vfsmount lock must be held for write |
b90fa9ae | 598 | */ |
4b2619a5 | 599 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 600 | { |
0714a533 | 601 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 602 | struct mount *m; |
b90fa9ae | 603 | LIST_HEAD(head); |
143c8c91 | 604 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 605 | |
0714a533 | 606 | BUG_ON(parent == mnt); |
b90fa9ae | 607 | |
1a4eeaf2 AV |
608 | list_add_tail(&head, &mnt->mnt_list); |
609 | list_for_each_entry(m, &head, mnt_list) { | |
143c8c91 | 610 | m->mnt_ns = n; |
7e3d0eb0 | 611 | __mnt_make_longterm(m); |
f03c6599 AV |
612 | } |
613 | ||
b90fa9ae RP |
614 | list_splice(&head, n->list.prev); |
615 | ||
1b8e5564 | 616 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 617 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 618 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 619 | touch_mnt_namespace(n); |
1da177e4 LT |
620 | } |
621 | ||
909b0a88 | 622 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 623 | { |
6b41d536 AV |
624 | struct list_head *next = p->mnt_mounts.next; |
625 | if (next == &p->mnt_mounts) { | |
1da177e4 | 626 | while (1) { |
909b0a88 | 627 | if (p == root) |
1da177e4 | 628 | return NULL; |
6b41d536 AV |
629 | next = p->mnt_child.next; |
630 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 631 | break; |
0714a533 | 632 | p = p->mnt_parent; |
1da177e4 LT |
633 | } |
634 | } | |
6b41d536 | 635 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
636 | } |
637 | ||
315fc83e | 638 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 639 | { |
6b41d536 AV |
640 | struct list_head *prev = p->mnt_mounts.prev; |
641 | while (prev != &p->mnt_mounts) { | |
642 | p = list_entry(prev, struct mount, mnt_child); | |
643 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
644 | } |
645 | return p; | |
646 | } | |
647 | ||
9d412a43 AV |
648 | struct vfsmount * |
649 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
650 | { | |
b105e270 | 651 | struct mount *mnt; |
9d412a43 AV |
652 | struct dentry *root; |
653 | ||
654 | if (!type) | |
655 | return ERR_PTR(-ENODEV); | |
656 | ||
657 | mnt = alloc_vfsmnt(name); | |
658 | if (!mnt) | |
659 | return ERR_PTR(-ENOMEM); | |
660 | ||
661 | if (flags & MS_KERNMOUNT) | |
b105e270 | 662 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
663 | |
664 | root = mount_fs(type, flags, name, data); | |
665 | if (IS_ERR(root)) { | |
666 | free_vfsmnt(mnt); | |
667 | return ERR_CAST(root); | |
668 | } | |
669 | ||
b105e270 AV |
670 | mnt->mnt.mnt_root = root; |
671 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 672 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 673 | mnt->mnt_parent = mnt; |
39f7c4db MS |
674 | br_write_lock(vfsmount_lock); |
675 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); | |
676 | br_write_unlock(vfsmount_lock); | |
b105e270 | 677 | return &mnt->mnt; |
9d412a43 AV |
678 | } |
679 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
680 | ||
87129cc0 | 681 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 682 | int flag) |
1da177e4 | 683 | { |
87129cc0 | 684 | struct super_block *sb = old->mnt.mnt_sb; |
52ba1621 | 685 | struct mount *mnt = alloc_vfsmnt(old->mnt_devname); |
1da177e4 LT |
686 | |
687 | if (mnt) { | |
719f5d7f | 688 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
15169fe7 | 689 | mnt->mnt_group_id = 0; /* not a peer of original */ |
719f5d7f | 690 | else |
15169fe7 | 691 | mnt->mnt_group_id = old->mnt_group_id; |
719f5d7f | 692 | |
15169fe7 | 693 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
b105e270 | 694 | int err = mnt_alloc_group_id(mnt); |
719f5d7f MS |
695 | if (err) |
696 | goto out_free; | |
697 | } | |
698 | ||
87129cc0 | 699 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; |
1da177e4 | 700 | atomic_inc(&sb->s_active); |
b105e270 AV |
701 | mnt->mnt.mnt_sb = sb; |
702 | mnt->mnt.mnt_root = dget(root); | |
a73324da | 703 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 704 | mnt->mnt_parent = mnt; |
39f7c4db MS |
705 | br_write_lock(vfsmount_lock); |
706 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
707 | br_write_unlock(vfsmount_lock); | |
b90fa9ae | 708 | |
5afe0022 | 709 | if (flag & CL_SLAVE) { |
6776db3d | 710 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
32301920 | 711 | mnt->mnt_master = old; |
fc7be130 | 712 | CLEAR_MNT_SHARED(mnt); |
8aec0809 | 713 | } else if (!(flag & CL_PRIVATE)) { |
fc7be130 | 714 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) |
6776db3d | 715 | list_add(&mnt->mnt_share, &old->mnt_share); |
d10e8def | 716 | if (IS_MNT_SLAVE(old)) |
6776db3d | 717 | list_add(&mnt->mnt_slave, &old->mnt_slave); |
d10e8def | 718 | mnt->mnt_master = old->mnt_master; |
5afe0022 | 719 | } |
b90fa9ae | 720 | if (flag & CL_MAKE_SHARED) |
0f0afb1d | 721 | set_mnt_shared(mnt); |
1da177e4 LT |
722 | |
723 | /* stick the duplicate mount on the same expiry list | |
724 | * as the original if that was on one */ | |
36341f64 | 725 | if (flag & CL_EXPIRE) { |
6776db3d AV |
726 | if (!list_empty(&old->mnt_expire)) |
727 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
36341f64 | 728 | } |
1da177e4 | 729 | } |
cb338d06 | 730 | return mnt; |
719f5d7f MS |
731 | |
732 | out_free: | |
733 | free_vfsmnt(mnt); | |
734 | return NULL; | |
1da177e4 LT |
735 | } |
736 | ||
83adc753 | 737 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 738 | { |
83adc753 AV |
739 | struct vfsmount *m = &mnt->mnt; |
740 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 741 | |
3d733633 DH |
742 | /* |
743 | * This probably indicates that somebody messed | |
744 | * up a mnt_want/drop_write() pair. If this | |
745 | * happens, the filesystem was probably unable | |
746 | * to make r/w->r/o transitions. | |
747 | */ | |
d3ef3d73 | 748 | /* |
b3e19d92 NP |
749 | * The locking used to deal with mnt_count decrement provides barriers, |
750 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 751 | */ |
c6653a83 | 752 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
753 | fsnotify_vfsmount_delete(m); |
754 | dput(m->mnt_root); | |
755 | free_vfsmnt(mnt); | |
1da177e4 LT |
756 | deactivate_super(sb); |
757 | } | |
758 | ||
900148dc | 759 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 760 | { |
b3e19d92 | 761 | put_again: |
f03c6599 AV |
762 | #ifdef CONFIG_SMP |
763 | br_read_lock(vfsmount_lock); | |
68e8a9fe | 764 | if (likely(atomic_read(&mnt->mnt_longterm))) { |
aa9c0e07 | 765 | mnt_add_count(mnt, -1); |
b3e19d92 | 766 | br_read_unlock(vfsmount_lock); |
f03c6599 | 767 | return; |
b3e19d92 | 768 | } |
f03c6599 | 769 | br_read_unlock(vfsmount_lock); |
b3e19d92 | 770 | |
99b7db7b | 771 | br_write_lock(vfsmount_lock); |
aa9c0e07 | 772 | mnt_add_count(mnt, -1); |
b3e19d92 | 773 | if (mnt_get_count(mnt)) { |
99b7db7b NP |
774 | br_write_unlock(vfsmount_lock); |
775 | return; | |
776 | } | |
b3e19d92 | 777 | #else |
aa9c0e07 | 778 | mnt_add_count(mnt, -1); |
b3e19d92 | 779 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 780 | return; |
b3e19d92 | 781 | br_write_lock(vfsmount_lock); |
f03c6599 | 782 | #endif |
863d684f AV |
783 | if (unlikely(mnt->mnt_pinned)) { |
784 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
785 | mnt->mnt_pinned = 0; | |
b3e19d92 | 786 | br_write_unlock(vfsmount_lock); |
900148dc | 787 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 788 | goto put_again; |
7b7b1ace | 789 | } |
39f7c4db | 790 | list_del(&mnt->mnt_instance); |
99b7db7b | 791 | br_write_unlock(vfsmount_lock); |
b3e19d92 NP |
792 | mntfree(mnt); |
793 | } | |
b3e19d92 NP |
794 | |
795 | void mntput(struct vfsmount *mnt) | |
796 | { | |
797 | if (mnt) { | |
863d684f | 798 | struct mount *m = real_mount(mnt); |
b3e19d92 | 799 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
800 | if (unlikely(m->mnt_expiry_mark)) |
801 | m->mnt_expiry_mark = 0; | |
802 | mntput_no_expire(m); | |
b3e19d92 NP |
803 | } |
804 | } | |
805 | EXPORT_SYMBOL(mntput); | |
806 | ||
807 | struct vfsmount *mntget(struct vfsmount *mnt) | |
808 | { | |
809 | if (mnt) | |
83adc753 | 810 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
811 | return mnt; |
812 | } | |
813 | EXPORT_SYMBOL(mntget); | |
814 | ||
7b7b1ace AV |
815 | void mnt_pin(struct vfsmount *mnt) |
816 | { | |
99b7db7b | 817 | br_write_lock(vfsmount_lock); |
863d684f | 818 | real_mount(mnt)->mnt_pinned++; |
99b7db7b | 819 | br_write_unlock(vfsmount_lock); |
7b7b1ace | 820 | } |
7b7b1ace AV |
821 | EXPORT_SYMBOL(mnt_pin); |
822 | ||
863d684f | 823 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 824 | { |
863d684f | 825 | struct mount *mnt = real_mount(m); |
99b7db7b | 826 | br_write_lock(vfsmount_lock); |
7b7b1ace | 827 | if (mnt->mnt_pinned) { |
863d684f | 828 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
829 | mnt->mnt_pinned--; |
830 | } | |
99b7db7b | 831 | br_write_unlock(vfsmount_lock); |
7b7b1ace | 832 | } |
7b7b1ace | 833 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 834 | |
b3b304a2 MS |
835 | static inline void mangle(struct seq_file *m, const char *s) |
836 | { | |
837 | seq_escape(m, s, " \t\n\\"); | |
838 | } | |
839 | ||
840 | /* | |
841 | * Simple .show_options callback for filesystems which don't want to | |
842 | * implement more complex mount option showing. | |
843 | * | |
844 | * See also save_mount_options(). | |
845 | */ | |
34c80b1d | 846 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 847 | { |
2a32cebd AV |
848 | const char *options; |
849 | ||
850 | rcu_read_lock(); | |
34c80b1d | 851 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
852 | |
853 | if (options != NULL && options[0]) { | |
854 | seq_putc(m, ','); | |
855 | mangle(m, options); | |
856 | } | |
2a32cebd | 857 | rcu_read_unlock(); |
b3b304a2 MS |
858 | |
859 | return 0; | |
860 | } | |
861 | EXPORT_SYMBOL(generic_show_options); | |
862 | ||
863 | /* | |
864 | * If filesystem uses generic_show_options(), this function should be | |
865 | * called from the fill_super() callback. | |
866 | * | |
867 | * The .remount_fs callback usually needs to be handled in a special | |
868 | * way, to make sure, that previous options are not overwritten if the | |
869 | * remount fails. | |
870 | * | |
871 | * Also note, that if the filesystem's .remount_fs function doesn't | |
872 | * reset all options to their default value, but changes only newly | |
873 | * given options, then the displayed options will not reflect reality | |
874 | * any more. | |
875 | */ | |
876 | void save_mount_options(struct super_block *sb, char *options) | |
877 | { | |
2a32cebd AV |
878 | BUG_ON(sb->s_options); |
879 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
880 | } |
881 | EXPORT_SYMBOL(save_mount_options); | |
882 | ||
2a32cebd AV |
883 | void replace_mount_options(struct super_block *sb, char *options) |
884 | { | |
885 | char *old = sb->s_options; | |
886 | rcu_assign_pointer(sb->s_options, options); | |
887 | if (old) { | |
888 | synchronize_rcu(); | |
889 | kfree(old); | |
890 | } | |
891 | } | |
892 | EXPORT_SYMBOL(replace_mount_options); | |
893 | ||
a1a2c409 | 894 | #ifdef CONFIG_PROC_FS |
0226f492 | 895 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
896 | static void *m_start(struct seq_file *m, loff_t *pos) |
897 | { | |
0226f492 | 898 | struct proc_mounts *p = container_of(m, struct proc_mounts, m); |
1da177e4 | 899 | |
390c6843 | 900 | down_read(&namespace_sem); |
a1a2c409 | 901 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
902 | } |
903 | ||
904 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
905 | { | |
0226f492 | 906 | struct proc_mounts *p = container_of(m, struct proc_mounts, m); |
b0765fb8 | 907 | |
a1a2c409 | 908 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
909 | } |
910 | ||
911 | static void m_stop(struct seq_file *m, void *v) | |
912 | { | |
390c6843 | 913 | up_read(&namespace_sem); |
1da177e4 LT |
914 | } |
915 | ||
0226f492 | 916 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 917 | { |
0226f492 | 918 | struct proc_mounts *p = container_of(m, struct proc_mounts, m); |
1a4eeaf2 | 919 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 920 | return p->show(m, &r->mnt); |
1da177e4 LT |
921 | } |
922 | ||
a1a2c409 | 923 | const struct seq_operations mounts_op = { |
1da177e4 LT |
924 | .start = m_start, |
925 | .next = m_next, | |
926 | .stop = m_stop, | |
0226f492 | 927 | .show = m_show, |
b4629fe2 | 928 | }; |
a1a2c409 | 929 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 930 | |
1da177e4 LT |
931 | /** |
932 | * may_umount_tree - check if a mount tree is busy | |
933 | * @mnt: root of mount tree | |
934 | * | |
935 | * This is called to check if a tree of mounts has any | |
936 | * open files, pwds, chroots or sub mounts that are | |
937 | * busy. | |
938 | */ | |
909b0a88 | 939 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 940 | { |
909b0a88 | 941 | struct mount *mnt = real_mount(m); |
36341f64 RP |
942 | int actual_refs = 0; |
943 | int minimum_refs = 0; | |
315fc83e | 944 | struct mount *p; |
909b0a88 | 945 | BUG_ON(!m); |
1da177e4 | 946 | |
b3e19d92 NP |
947 | /* write lock needed for mnt_get_count */ |
948 | br_write_lock(vfsmount_lock); | |
909b0a88 | 949 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 950 | actual_refs += mnt_get_count(p); |
1da177e4 | 951 | minimum_refs += 2; |
1da177e4 | 952 | } |
b3e19d92 | 953 | br_write_unlock(vfsmount_lock); |
1da177e4 LT |
954 | |
955 | if (actual_refs > minimum_refs) | |
e3474a8e | 956 | return 0; |
1da177e4 | 957 | |
e3474a8e | 958 | return 1; |
1da177e4 LT |
959 | } |
960 | ||
961 | EXPORT_SYMBOL(may_umount_tree); | |
962 | ||
963 | /** | |
964 | * may_umount - check if a mount point is busy | |
965 | * @mnt: root of mount | |
966 | * | |
967 | * This is called to check if a mount point has any | |
968 | * open files, pwds, chroots or sub mounts. If the | |
969 | * mount has sub mounts this will return busy | |
970 | * regardless of whether the sub mounts are busy. | |
971 | * | |
972 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
973 | * give false negatives. The main reason why it's here is that we need | |
974 | * a non-destructive way to look for easily umountable filesystems. | |
975 | */ | |
976 | int may_umount(struct vfsmount *mnt) | |
977 | { | |
e3474a8e | 978 | int ret = 1; |
8ad08d8a | 979 | down_read(&namespace_sem); |
b3e19d92 | 980 | br_write_lock(vfsmount_lock); |
1ab59738 | 981 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 982 | ret = 0; |
b3e19d92 | 983 | br_write_unlock(vfsmount_lock); |
8ad08d8a | 984 | up_read(&namespace_sem); |
a05964f3 | 985 | return ret; |
1da177e4 LT |
986 | } |
987 | ||
988 | EXPORT_SYMBOL(may_umount); | |
989 | ||
b90fa9ae | 990 | void release_mounts(struct list_head *head) |
70fbcdf4 | 991 | { |
d5e50f74 | 992 | struct mount *mnt; |
bf066c7d | 993 | while (!list_empty(head)) { |
1b8e5564 AV |
994 | mnt = list_first_entry(head, struct mount, mnt_hash); |
995 | list_del_init(&mnt->mnt_hash); | |
676da58d | 996 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 997 | struct dentry *dentry; |
863d684f | 998 | struct mount *m; |
99b7db7b NP |
999 | |
1000 | br_write_lock(vfsmount_lock); | |
a73324da | 1001 | dentry = mnt->mnt_mountpoint; |
863d684f | 1002 | m = mnt->mnt_parent; |
a73324da | 1003 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1004 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1005 | m->mnt_ghosts--; |
99b7db7b | 1006 | br_write_unlock(vfsmount_lock); |
70fbcdf4 | 1007 | dput(dentry); |
863d684f | 1008 | mntput(&m->mnt); |
70fbcdf4 | 1009 | } |
d5e50f74 | 1010 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1011 | } |
1012 | } | |
1013 | ||
99b7db7b NP |
1014 | /* |
1015 | * vfsmount lock must be held for write | |
1016 | * namespace_sem must be held for write | |
1017 | */ | |
761d5c38 | 1018 | void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) |
1da177e4 | 1019 | { |
7b8a53fd | 1020 | LIST_HEAD(tmp_list); |
315fc83e | 1021 | struct mount *p; |
1da177e4 | 1022 | |
909b0a88 | 1023 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1024 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1025 | |
a05964f3 | 1026 | if (propagate) |
7b8a53fd | 1027 | propagate_umount(&tmp_list); |
a05964f3 | 1028 | |
1b8e5564 | 1029 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1030 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1031 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1032 | __touch_mnt_namespace(p->mnt_ns); |
1033 | p->mnt_ns = NULL; | |
83adc753 | 1034 | __mnt_make_shortterm(p); |
6b41d536 | 1035 | list_del_init(&p->mnt_child); |
676da58d | 1036 | if (mnt_has_parent(p)) { |
863d684f | 1037 | p->mnt_parent->mnt_ghosts++; |
a73324da | 1038 | dentry_reset_mounted(p->mnt_mountpoint); |
7c4b93d8 | 1039 | } |
0f0afb1d | 1040 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1041 | } |
7b8a53fd | 1042 | list_splice(&tmp_list, kill); |
1da177e4 LT |
1043 | } |
1044 | ||
692afc31 | 1045 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts); |
c35038be | 1046 | |
1ab59738 | 1047 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1048 | { |
1ab59738 | 1049 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 | 1050 | int retval; |
70fbcdf4 | 1051 | LIST_HEAD(umount_list); |
1da177e4 | 1052 | |
1ab59738 | 1053 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1054 | if (retval) |
1055 | return retval; | |
1056 | ||
1057 | /* | |
1058 | * Allow userspace to request a mountpoint be expired rather than | |
1059 | * unmounting unconditionally. Unmount only happens if: | |
1060 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1061 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1062 | */ | |
1063 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1064 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1065 | flags & (MNT_FORCE | MNT_DETACH)) |
1066 | return -EINVAL; | |
1067 | ||
b3e19d92 NP |
1068 | /* |
1069 | * probably don't strictly need the lock here if we examined | |
1070 | * all race cases, but it's a slowpath. | |
1071 | */ | |
1072 | br_write_lock(vfsmount_lock); | |
83adc753 | 1073 | if (mnt_get_count(mnt) != 2) { |
bf9faa2a | 1074 | br_write_unlock(vfsmount_lock); |
1da177e4 | 1075 | return -EBUSY; |
b3e19d92 NP |
1076 | } |
1077 | br_write_unlock(vfsmount_lock); | |
1da177e4 | 1078 | |
863d684f | 1079 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1080 | return -EAGAIN; |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * If we may have to abort operations to get out of this | |
1085 | * mount, and they will themselves hold resources we must | |
1086 | * allow the fs to do things. In the Unix tradition of | |
1087 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1088 | * might fail to complete on the first run through as other tasks | |
1089 | * must return, and the like. Thats for the mount program to worry | |
1090 | * about for the moment. | |
1091 | */ | |
1092 | ||
42faad99 | 1093 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1094 | sb->s_op->umount_begin(sb); |
42faad99 | 1095 | } |
1da177e4 LT |
1096 | |
1097 | /* | |
1098 | * No sense to grab the lock for this test, but test itself looks | |
1099 | * somewhat bogus. Suggestions for better replacement? | |
1100 | * Ho-hum... In principle, we might treat that as umount + switch | |
1101 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1102 | * Actually it makes sense, especially if rootfs would contain a | |
1103 | * /reboot - static binary that would close all descriptors and | |
1104 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1105 | */ | |
1ab59738 | 1106 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1107 | /* |
1108 | * Special case for "unmounting" root ... | |
1109 | * we just try to remount it readonly. | |
1110 | */ | |
1111 | down_write(&sb->s_umount); | |
4aa98cf7 | 1112 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1113 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1114 | up_write(&sb->s_umount); |
1115 | return retval; | |
1116 | } | |
1117 | ||
390c6843 | 1118 | down_write(&namespace_sem); |
99b7db7b | 1119 | br_write_lock(vfsmount_lock); |
5addc5dd | 1120 | event++; |
1da177e4 | 1121 | |
c35038be | 1122 | if (!(flags & MNT_DETACH)) |
1ab59738 | 1123 | shrink_submounts(mnt, &umount_list); |
c35038be | 1124 | |
1da177e4 | 1125 | retval = -EBUSY; |
a05964f3 | 1126 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1127 | if (!list_empty(&mnt->mnt_list)) |
1ab59738 | 1128 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1129 | retval = 0; |
1130 | } | |
99b7db7b | 1131 | br_write_unlock(vfsmount_lock); |
390c6843 | 1132 | up_write(&namespace_sem); |
70fbcdf4 | 1133 | release_mounts(&umount_list); |
1da177e4 LT |
1134 | return retval; |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Now umount can handle mount points as well as block devices. | |
1139 | * This is important for filesystems which use unnamed block devices. | |
1140 | * | |
1141 | * We now support a flag for forced unmount like the other 'big iron' | |
1142 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1143 | */ | |
1144 | ||
bdc480e3 | 1145 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1146 | { |
2d8f3038 | 1147 | struct path path; |
900148dc | 1148 | struct mount *mnt; |
1da177e4 | 1149 | int retval; |
db1f05bb | 1150 | int lookup_flags = 0; |
1da177e4 | 1151 | |
db1f05bb MS |
1152 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1153 | return -EINVAL; | |
1154 | ||
1155 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1156 | lookup_flags |= LOOKUP_FOLLOW; | |
1157 | ||
1158 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1159 | if (retval) |
1160 | goto out; | |
900148dc | 1161 | mnt = real_mount(path.mnt); |
1da177e4 | 1162 | retval = -EINVAL; |
2d8f3038 | 1163 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1164 | goto dput_and_out; |
143c8c91 | 1165 | if (!check_mnt(mnt)) |
1da177e4 LT |
1166 | goto dput_and_out; |
1167 | ||
1168 | retval = -EPERM; | |
1169 | if (!capable(CAP_SYS_ADMIN)) | |
1170 | goto dput_and_out; | |
1171 | ||
900148dc | 1172 | retval = do_umount(mnt, flags); |
1da177e4 | 1173 | dput_and_out: |
429731b1 | 1174 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1175 | dput(path.dentry); |
900148dc | 1176 | mntput_no_expire(mnt); |
1da177e4 LT |
1177 | out: |
1178 | return retval; | |
1179 | } | |
1180 | ||
1181 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1182 | ||
1183 | /* | |
b58fed8b | 1184 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1185 | */ |
bdc480e3 | 1186 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1187 | { |
b58fed8b | 1188 | return sys_umount(name, 0); |
1da177e4 LT |
1189 | } |
1190 | ||
1191 | #endif | |
1192 | ||
2d92ab3c | 1193 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1194 | { |
1195 | if (capable(CAP_SYS_ADMIN)) | |
1196 | return 0; | |
1197 | return -EPERM; | |
1198 | #ifdef notyet | |
2d92ab3c | 1199 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1200 | return -EPERM; |
2d92ab3c | 1201 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1202 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1203 | return -EPERM; |
1204 | } | |
2d92ab3c | 1205 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1206 | return -EPERM; |
1207 | return 0; | |
1208 | #endif | |
1209 | } | |
1210 | ||
87129cc0 | 1211 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1212 | int flag) |
1da177e4 | 1213 | { |
a73324da | 1214 | struct mount *res, *p, *q, *r; |
1a390689 | 1215 | struct path path; |
1da177e4 | 1216 | |
fc7be130 | 1217 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
9676f0c6 RP |
1218 | return NULL; |
1219 | ||
36341f64 | 1220 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
1221 | if (!q) |
1222 | goto Enomem; | |
a73324da | 1223 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1224 | |
1225 | p = mnt; | |
6b41d536 | 1226 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1227 | struct mount *s; |
7ec02ef1 | 1228 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1229 | continue; |
1230 | ||
909b0a88 | 1231 | for (s = r; s; s = next_mnt(s, r)) { |
fc7be130 | 1232 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
9676f0c6 RP |
1233 | s = skip_mnt_tree(s); |
1234 | continue; | |
1235 | } | |
0714a533 AV |
1236 | while (p != s->mnt_parent) { |
1237 | p = p->mnt_parent; | |
1238 | q = q->mnt_parent; | |
1da177e4 | 1239 | } |
87129cc0 | 1240 | p = s; |
cb338d06 | 1241 | path.mnt = &q->mnt; |
a73324da | 1242 | path.dentry = p->mnt_mountpoint; |
87129cc0 | 1243 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
1da177e4 LT |
1244 | if (!q) |
1245 | goto Enomem; | |
99b7db7b | 1246 | br_write_lock(vfsmount_lock); |
1a4eeaf2 | 1247 | list_add_tail(&q->mnt_list, &res->mnt_list); |
cb338d06 | 1248 | attach_mnt(q, &path); |
99b7db7b | 1249 | br_write_unlock(vfsmount_lock); |
1da177e4 LT |
1250 | } |
1251 | } | |
1252 | return res; | |
b58fed8b | 1253 | Enomem: |
1da177e4 | 1254 | if (res) { |
70fbcdf4 | 1255 | LIST_HEAD(umount_list); |
99b7db7b | 1256 | br_write_lock(vfsmount_lock); |
761d5c38 | 1257 | umount_tree(res, 0, &umount_list); |
99b7db7b | 1258 | br_write_unlock(vfsmount_lock); |
70fbcdf4 | 1259 | release_mounts(&umount_list); |
1da177e4 LT |
1260 | } |
1261 | return NULL; | |
1262 | } | |
1263 | ||
589ff870 | 1264 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1265 | { |
cb338d06 | 1266 | struct mount *tree; |
1a60a280 | 1267 | down_write(&namespace_sem); |
87129cc0 AV |
1268 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1269 | CL_COPY_ALL | CL_PRIVATE); | |
1a60a280 | 1270 | up_write(&namespace_sem); |
cb338d06 | 1271 | return tree ? &tree->mnt : NULL; |
8aec0809 AV |
1272 | } |
1273 | ||
1274 | void drop_collected_mounts(struct vfsmount *mnt) | |
1275 | { | |
1276 | LIST_HEAD(umount_list); | |
1a60a280 | 1277 | down_write(&namespace_sem); |
99b7db7b | 1278 | br_write_lock(vfsmount_lock); |
761d5c38 | 1279 | umount_tree(real_mount(mnt), 0, &umount_list); |
99b7db7b | 1280 | br_write_unlock(vfsmount_lock); |
1a60a280 | 1281 | up_write(&namespace_sem); |
8aec0809 AV |
1282 | release_mounts(&umount_list); |
1283 | } | |
1284 | ||
1f707137 AV |
1285 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1286 | struct vfsmount *root) | |
1287 | { | |
1a4eeaf2 | 1288 | struct mount *mnt; |
1f707137 AV |
1289 | int res = f(root, arg); |
1290 | if (res) | |
1291 | return res; | |
1a4eeaf2 AV |
1292 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1293 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1294 | if (res) |
1295 | return res; | |
1296 | } | |
1297 | return 0; | |
1298 | } | |
1299 | ||
4b8b21f4 | 1300 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1301 | { |
315fc83e | 1302 | struct mount *p; |
719f5d7f | 1303 | |
909b0a88 | 1304 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1305 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1306 | mnt_release_group_id(p); |
719f5d7f MS |
1307 | } |
1308 | } | |
1309 | ||
4b8b21f4 | 1310 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1311 | { |
315fc83e | 1312 | struct mount *p; |
719f5d7f | 1313 | |
909b0a88 | 1314 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1315 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1316 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1317 | if (err) { |
4b8b21f4 | 1318 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1319 | return err; |
1320 | } | |
1321 | } | |
1322 | } | |
1323 | ||
1324 | return 0; | |
1325 | } | |
1326 | ||
b90fa9ae RP |
1327 | /* |
1328 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1329 | * @nd : place the mount tree @source_mnt is attached |
1330 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1331 | * store the parent mount and mountpoint dentry. | |
1332 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1333 | * |
1334 | * NOTE: in the table below explains the semantics when a source mount | |
1335 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1336 | * --------------------------------------------------------------------------- |
1337 | * | BIND MOUNT OPERATION | | |
1338 | * |************************************************************************** | |
1339 | * | source-->| shared | private | slave | unbindable | | |
1340 | * | dest | | | | | | |
1341 | * | | | | | | | | |
1342 | * | v | | | | | | |
1343 | * |************************************************************************** | |
1344 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1345 | * | | | | | | | |
1346 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1347 | * *************************************************************************** | |
b90fa9ae RP |
1348 | * A bind operation clones the source mount and mounts the clone on the |
1349 | * destination mount. | |
1350 | * | |
1351 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1352 | * tree of the destination mount and the cloned mount is added to | |
1353 | * the peer group of the source mount. | |
1354 | * (+) the cloned mount is created under the destination mount and is marked | |
1355 | * as shared. The cloned mount is added to the peer group of the source | |
1356 | * mount. | |
5afe0022 RP |
1357 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1358 | * of the destination mount and the cloned mount is made slave | |
1359 | * of the same master as that of the source mount. The cloned mount | |
1360 | * is marked as 'shared and slave'. | |
1361 | * (*) the cloned mount is made a slave of the same master as that of the | |
1362 | * source mount. | |
1363 | * | |
9676f0c6 RP |
1364 | * --------------------------------------------------------------------------- |
1365 | * | MOVE MOUNT OPERATION | | |
1366 | * |************************************************************************** | |
1367 | * | source-->| shared | private | slave | unbindable | | |
1368 | * | dest | | | | | | |
1369 | * | | | | | | | | |
1370 | * | v | | | | | | |
1371 | * |************************************************************************** | |
1372 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1373 | * | | | | | | | |
1374 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1375 | * *************************************************************************** | |
5afe0022 RP |
1376 | * |
1377 | * (+) the mount is moved to the destination. And is then propagated to | |
1378 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1379 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1380 | * (+++) the mount is moved to the destination and is then propagated to |
1381 | * all the mounts belonging to the destination mount's propagation tree. | |
1382 | * the mount is marked as 'shared and slave'. | |
1383 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1384 | * |
1385 | * if the source mount is a tree, the operations explained above is | |
1386 | * applied to each mount in the tree. | |
1387 | * Must be called without spinlocks held, since this function can sleep | |
1388 | * in allocations. | |
1389 | */ | |
0fb54e50 | 1390 | static int attach_recursive_mnt(struct mount *source_mnt, |
1a390689 | 1391 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1392 | { |
1393 | LIST_HEAD(tree_list); | |
a8d56d8e | 1394 | struct mount *dest_mnt = real_mount(path->mnt); |
1a390689 | 1395 | struct dentry *dest_dentry = path->dentry; |
315fc83e | 1396 | struct mount *child, *p; |
719f5d7f | 1397 | int err; |
b90fa9ae | 1398 | |
fc7be130 | 1399 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1400 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1401 | if (err) |
1402 | goto out; | |
1403 | } | |
a8d56d8e | 1404 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); |
719f5d7f MS |
1405 | if (err) |
1406 | goto out_cleanup_ids; | |
b90fa9ae | 1407 | |
99b7db7b | 1408 | br_write_lock(vfsmount_lock); |
df1a1ad2 | 1409 | |
fc7be130 | 1410 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1411 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1412 | set_mnt_shared(p); |
b90fa9ae | 1413 | } |
1a390689 | 1414 | if (parent_path) { |
0fb54e50 AV |
1415 | detach_mnt(source_mnt, parent_path); |
1416 | attach_mnt(source_mnt, path); | |
143c8c91 | 1417 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1418 | } else { |
14cf1fa8 | 1419 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); |
0fb54e50 | 1420 | commit_tree(source_mnt); |
21444403 | 1421 | } |
b90fa9ae | 1422 | |
1b8e5564 AV |
1423 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1424 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1425 | commit_tree(child); |
b90fa9ae | 1426 | } |
99b7db7b NP |
1427 | br_write_unlock(vfsmount_lock); |
1428 | ||
b90fa9ae | 1429 | return 0; |
719f5d7f MS |
1430 | |
1431 | out_cleanup_ids: | |
fc7be130 | 1432 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1433 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1434 | out: |
1435 | return err; | |
b90fa9ae RP |
1436 | } |
1437 | ||
b12cea91 AV |
1438 | static int lock_mount(struct path *path) |
1439 | { | |
1440 | struct vfsmount *mnt; | |
1441 | retry: | |
1442 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1443 | if (unlikely(cant_mount(path->dentry))) { | |
1444 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1445 | return -ENOENT; | |
1446 | } | |
1447 | down_write(&namespace_sem); | |
1448 | mnt = lookup_mnt(path); | |
1449 | if (likely(!mnt)) | |
1450 | return 0; | |
1451 | up_write(&namespace_sem); | |
1452 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1453 | path_put(path); | |
1454 | path->mnt = mnt; | |
1455 | path->dentry = dget(mnt->mnt_root); | |
1456 | goto retry; | |
1457 | } | |
1458 | ||
1459 | static void unlock_mount(struct path *path) | |
1460 | { | |
1461 | up_write(&namespace_sem); | |
1462 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1463 | } | |
1464 | ||
95bc5f25 | 1465 | static int graft_tree(struct mount *mnt, struct path *path) |
1da177e4 | 1466 | { |
95bc5f25 | 1467 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1468 | return -EINVAL; |
1469 | ||
8c3ee42e | 1470 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
95bc5f25 | 1471 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1472 | return -ENOTDIR; |
1473 | ||
b12cea91 AV |
1474 | if (d_unlinked(path->dentry)) |
1475 | return -ENOENT; | |
1da177e4 | 1476 | |
95bc5f25 | 1477 | return attach_recursive_mnt(mnt, path, NULL); |
1da177e4 LT |
1478 | } |
1479 | ||
7a2e8a8f VA |
1480 | /* |
1481 | * Sanity check the flags to change_mnt_propagation. | |
1482 | */ | |
1483 | ||
1484 | static int flags_to_propagation_type(int flags) | |
1485 | { | |
7c6e984d | 1486 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1487 | |
1488 | /* Fail if any non-propagation flags are set */ | |
1489 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1490 | return 0; | |
1491 | /* Only one propagation flag should be set */ | |
1492 | if (!is_power_of_2(type)) | |
1493 | return 0; | |
1494 | return type; | |
1495 | } | |
1496 | ||
07b20889 RP |
1497 | /* |
1498 | * recursively change the type of the mountpoint. | |
1499 | */ | |
0a0d8a46 | 1500 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1501 | { |
315fc83e | 1502 | struct mount *m; |
4b8b21f4 | 1503 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1504 | int recurse = flag & MS_REC; |
7a2e8a8f | 1505 | int type; |
719f5d7f | 1506 | int err = 0; |
07b20889 | 1507 | |
ee6f9582 MS |
1508 | if (!capable(CAP_SYS_ADMIN)) |
1509 | return -EPERM; | |
1510 | ||
2d92ab3c | 1511 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1512 | return -EINVAL; |
1513 | ||
7a2e8a8f VA |
1514 | type = flags_to_propagation_type(flag); |
1515 | if (!type) | |
1516 | return -EINVAL; | |
1517 | ||
07b20889 | 1518 | down_write(&namespace_sem); |
719f5d7f MS |
1519 | if (type == MS_SHARED) { |
1520 | err = invent_group_ids(mnt, recurse); | |
1521 | if (err) | |
1522 | goto out_unlock; | |
1523 | } | |
1524 | ||
99b7db7b | 1525 | br_write_lock(vfsmount_lock); |
909b0a88 | 1526 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1527 | change_mnt_propagation(m, type); |
99b7db7b | 1528 | br_write_unlock(vfsmount_lock); |
719f5d7f MS |
1529 | |
1530 | out_unlock: | |
07b20889 | 1531 | up_write(&namespace_sem); |
719f5d7f | 1532 | return err; |
07b20889 RP |
1533 | } |
1534 | ||
1da177e4 LT |
1535 | /* |
1536 | * do loopback mount. | |
1537 | */ | |
0a0d8a46 | 1538 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1539 | int recurse) |
1da177e4 | 1540 | { |
b12cea91 | 1541 | LIST_HEAD(umount_list); |
2d92ab3c | 1542 | struct path old_path; |
87129cc0 | 1543 | struct mount *mnt = NULL, *old; |
2d92ab3c | 1544 | int err = mount_is_safe(path); |
1da177e4 LT |
1545 | if (err) |
1546 | return err; | |
1547 | if (!old_name || !*old_name) | |
1548 | return -EINVAL; | |
815d405c | 1549 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1550 | if (err) |
1551 | return err; | |
1552 | ||
b12cea91 AV |
1553 | err = lock_mount(path); |
1554 | if (err) | |
1555 | goto out; | |
1556 | ||
87129cc0 AV |
1557 | old = real_mount(old_path.mnt); |
1558 | ||
1da177e4 | 1559 | err = -EINVAL; |
fc7be130 | 1560 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1561 | goto out2; |
9676f0c6 | 1562 | |
143c8c91 | 1563 | if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old)) |
b12cea91 | 1564 | goto out2; |
1da177e4 | 1565 | |
ccd48bc7 AV |
1566 | err = -ENOMEM; |
1567 | if (recurse) | |
87129cc0 | 1568 | mnt = copy_tree(old, old_path.dentry, 0); |
ccd48bc7 | 1569 | else |
87129cc0 | 1570 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 AV |
1571 | |
1572 | if (!mnt) | |
b12cea91 | 1573 | goto out2; |
ccd48bc7 | 1574 | |
95bc5f25 | 1575 | err = graft_tree(mnt, path); |
ccd48bc7 | 1576 | if (err) { |
99b7db7b | 1577 | br_write_lock(vfsmount_lock); |
761d5c38 | 1578 | umount_tree(mnt, 0, &umount_list); |
99b7db7b | 1579 | br_write_unlock(vfsmount_lock); |
5b83d2c5 | 1580 | } |
b12cea91 AV |
1581 | out2: |
1582 | unlock_mount(path); | |
1583 | release_mounts(&umount_list); | |
ccd48bc7 | 1584 | out: |
2d92ab3c | 1585 | path_put(&old_path); |
1da177e4 LT |
1586 | return err; |
1587 | } | |
1588 | ||
2e4b7fcd DH |
1589 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1590 | { | |
1591 | int error = 0; | |
1592 | int readonly_request = 0; | |
1593 | ||
1594 | if (ms_flags & MS_RDONLY) | |
1595 | readonly_request = 1; | |
1596 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1597 | return 0; | |
1598 | ||
1599 | if (readonly_request) | |
83adc753 | 1600 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1601 | else |
83adc753 | 1602 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1603 | return error; |
1604 | } | |
1605 | ||
1da177e4 LT |
1606 | /* |
1607 | * change filesystem flags. dir should be a physical root of filesystem. | |
1608 | * If you've mounted a non-root directory somewhere and want to do remount | |
1609 | * on it - tough luck. | |
1610 | */ | |
0a0d8a46 | 1611 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1612 | void *data) |
1613 | { | |
1614 | int err; | |
2d92ab3c | 1615 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1616 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 LT |
1617 | |
1618 | if (!capable(CAP_SYS_ADMIN)) | |
1619 | return -EPERM; | |
1620 | ||
143c8c91 | 1621 | if (!check_mnt(mnt)) |
1da177e4 LT |
1622 | return -EINVAL; |
1623 | ||
2d92ab3c | 1624 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1625 | return -EINVAL; |
1626 | ||
ff36fe2c EP |
1627 | err = security_sb_remount(sb, data); |
1628 | if (err) | |
1629 | return err; | |
1630 | ||
1da177e4 | 1631 | down_write(&sb->s_umount); |
2e4b7fcd | 1632 | if (flags & MS_BIND) |
2d92ab3c | 1633 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1634 | else |
2e4b7fcd | 1635 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1636 | if (!err) { |
99b7db7b | 1637 | br_write_lock(vfsmount_lock); |
143c8c91 AV |
1638 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1639 | mnt->mnt.mnt_flags = mnt_flags; | |
99b7db7b | 1640 | br_write_unlock(vfsmount_lock); |
7b43a79f | 1641 | } |
1da177e4 | 1642 | up_write(&sb->s_umount); |
0e55a7cc | 1643 | if (!err) { |
99b7db7b | 1644 | br_write_lock(vfsmount_lock); |
143c8c91 | 1645 | touch_mnt_namespace(mnt->mnt_ns); |
99b7db7b | 1646 | br_write_unlock(vfsmount_lock); |
0e55a7cc | 1647 | } |
1da177e4 LT |
1648 | return err; |
1649 | } | |
1650 | ||
cbbe362c | 1651 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1652 | { |
315fc83e | 1653 | struct mount *p; |
909b0a88 | 1654 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1655 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1656 | return 1; |
1657 | } | |
1658 | return 0; | |
1659 | } | |
1660 | ||
0a0d8a46 | 1661 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1662 | { |
2d92ab3c | 1663 | struct path old_path, parent_path; |
676da58d | 1664 | struct mount *p; |
0fb54e50 | 1665 | struct mount *old; |
1da177e4 LT |
1666 | int err = 0; |
1667 | if (!capable(CAP_SYS_ADMIN)) | |
1668 | return -EPERM; | |
1669 | if (!old_name || !*old_name) | |
1670 | return -EINVAL; | |
2d92ab3c | 1671 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1672 | if (err) |
1673 | return err; | |
1674 | ||
b12cea91 | 1675 | err = lock_mount(path); |
cc53ce53 DH |
1676 | if (err < 0) |
1677 | goto out; | |
1678 | ||
143c8c91 | 1679 | old = real_mount(old_path.mnt); |
fc7be130 | 1680 | p = real_mount(path->mnt); |
143c8c91 | 1681 | |
1da177e4 | 1682 | err = -EINVAL; |
fc7be130 | 1683 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1684 | goto out1; |
1685 | ||
f3da392e | 1686 | if (d_unlinked(path->dentry)) |
21444403 | 1687 | goto out1; |
1da177e4 LT |
1688 | |
1689 | err = -EINVAL; | |
2d92ab3c | 1690 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1691 | goto out1; |
1da177e4 | 1692 | |
676da58d | 1693 | if (!mnt_has_parent(old)) |
21444403 | 1694 | goto out1; |
1da177e4 | 1695 | |
2d92ab3c AV |
1696 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1697 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1698 | goto out1; |
1699 | /* | |
1700 | * Don't move a mount residing in a shared parent. | |
1701 | */ | |
fc7be130 | 1702 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1703 | goto out1; |
9676f0c6 RP |
1704 | /* |
1705 | * Don't move a mount tree containing unbindable mounts to a destination | |
1706 | * mount which is shared. | |
1707 | */ | |
fc7be130 | 1708 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1709 | goto out1; |
1da177e4 | 1710 | err = -ELOOP; |
fc7be130 | 1711 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1712 | if (p == old) |
21444403 | 1713 | goto out1; |
1da177e4 | 1714 | |
0fb54e50 | 1715 | err = attach_recursive_mnt(old, path, &parent_path); |
4ac91378 | 1716 | if (err) |
21444403 | 1717 | goto out1; |
1da177e4 LT |
1718 | |
1719 | /* if the mount is moved, it should no longer be expire | |
1720 | * automatically */ | |
6776db3d | 1721 | list_del_init(&old->mnt_expire); |
1da177e4 | 1722 | out1: |
b12cea91 | 1723 | unlock_mount(path); |
1da177e4 | 1724 | out: |
1da177e4 | 1725 | if (!err) |
1a390689 | 1726 | path_put(&parent_path); |
2d92ab3c | 1727 | path_put(&old_path); |
1da177e4 LT |
1728 | return err; |
1729 | } | |
1730 | ||
9d412a43 AV |
1731 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1732 | { | |
1733 | int err; | |
1734 | const char *subtype = strchr(fstype, '.'); | |
1735 | if (subtype) { | |
1736 | subtype++; | |
1737 | err = -EINVAL; | |
1738 | if (!subtype[0]) | |
1739 | goto err; | |
1740 | } else | |
1741 | subtype = ""; | |
1742 | ||
1743 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1744 | err = -ENOMEM; | |
1745 | if (!mnt->mnt_sb->s_subtype) | |
1746 | goto err; | |
1747 | return mnt; | |
1748 | ||
1749 | err: | |
1750 | mntput(mnt); | |
1751 | return ERR_PTR(err); | |
1752 | } | |
1753 | ||
79e801a9 | 1754 | static struct vfsmount * |
9d412a43 AV |
1755 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) |
1756 | { | |
1757 | struct file_system_type *type = get_fs_type(fstype); | |
1758 | struct vfsmount *mnt; | |
1759 | if (!type) | |
1760 | return ERR_PTR(-ENODEV); | |
1761 | mnt = vfs_kern_mount(type, flags, name, data); | |
1762 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1763 | !mnt->mnt_sb->s_subtype) | |
1764 | mnt = fs_set_subtype(mnt, fstype); | |
1765 | put_filesystem(type); | |
1766 | return mnt; | |
1767 | } | |
9d412a43 AV |
1768 | |
1769 | /* | |
1770 | * add a mount into a namespace's mount tree | |
1771 | */ | |
95bc5f25 | 1772 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 AV |
1773 | { |
1774 | int err; | |
1775 | ||
1776 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1777 | ||
b12cea91 AV |
1778 | err = lock_mount(path); |
1779 | if (err) | |
1780 | return err; | |
9d412a43 AV |
1781 | |
1782 | err = -EINVAL; | |
143c8c91 | 1783 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt))) |
9d412a43 AV |
1784 | goto unlock; |
1785 | ||
1786 | /* Refuse the same filesystem on the same mount point */ | |
1787 | err = -EBUSY; | |
95bc5f25 | 1788 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1789 | path->mnt->mnt_root == path->dentry) |
1790 | goto unlock; | |
1791 | ||
1792 | err = -EINVAL; | |
95bc5f25 | 1793 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
1794 | goto unlock; |
1795 | ||
95bc5f25 | 1796 | newmnt->mnt.mnt_flags = mnt_flags; |
9d412a43 AV |
1797 | err = graft_tree(newmnt, path); |
1798 | ||
1799 | unlock: | |
b12cea91 | 1800 | unlock_mount(path); |
9d412a43 AV |
1801 | return err; |
1802 | } | |
b1e75df4 | 1803 | |
1da177e4 LT |
1804 | /* |
1805 | * create a new mount for userspace and request it to be added into the | |
1806 | * namespace's tree | |
1807 | */ | |
0a0d8a46 | 1808 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1809 | int mnt_flags, char *name, void *data) |
1810 | { | |
1811 | struct vfsmount *mnt; | |
15f9a3f3 | 1812 | int err; |
1da177e4 | 1813 | |
eca6f534 | 1814 | if (!type) |
1da177e4 LT |
1815 | return -EINVAL; |
1816 | ||
1817 | /* we need capabilities... */ | |
1818 | if (!capable(CAP_SYS_ADMIN)) | |
1819 | return -EPERM; | |
1820 | ||
1821 | mnt = do_kern_mount(type, flags, name, data); | |
1822 | if (IS_ERR(mnt)) | |
1823 | return PTR_ERR(mnt); | |
1824 | ||
95bc5f25 | 1825 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
1826 | if (err) |
1827 | mntput(mnt); | |
1828 | return err; | |
1da177e4 LT |
1829 | } |
1830 | ||
19a167af AV |
1831 | int finish_automount(struct vfsmount *m, struct path *path) |
1832 | { | |
6776db3d | 1833 | struct mount *mnt = real_mount(m); |
19a167af AV |
1834 | int err; |
1835 | /* The new mount record should have at least 2 refs to prevent it being | |
1836 | * expired before we get a chance to add it | |
1837 | */ | |
6776db3d | 1838 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
1839 | |
1840 | if (m->mnt_sb == path->mnt->mnt_sb && | |
1841 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
1842 | err = -ELOOP; |
1843 | goto fail; | |
19a167af AV |
1844 | } |
1845 | ||
95bc5f25 | 1846 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
1847 | if (!err) |
1848 | return 0; | |
1849 | fail: | |
1850 | /* remove m from any expiration list it may be on */ | |
6776db3d | 1851 | if (!list_empty(&mnt->mnt_expire)) { |
b1e75df4 AV |
1852 | down_write(&namespace_sem); |
1853 | br_write_lock(vfsmount_lock); | |
6776db3d | 1854 | list_del_init(&mnt->mnt_expire); |
b1e75df4 AV |
1855 | br_write_unlock(vfsmount_lock); |
1856 | up_write(&namespace_sem); | |
19a167af | 1857 | } |
b1e75df4 AV |
1858 | mntput(m); |
1859 | mntput(m); | |
19a167af AV |
1860 | return err; |
1861 | } | |
1862 | ||
ea5b778a DH |
1863 | /** |
1864 | * mnt_set_expiry - Put a mount on an expiration list | |
1865 | * @mnt: The mount to list. | |
1866 | * @expiry_list: The list to add the mount to. | |
1867 | */ | |
1868 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
1869 | { | |
1870 | down_write(&namespace_sem); | |
1871 | br_write_lock(vfsmount_lock); | |
1872 | ||
6776db3d | 1873 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a DH |
1874 | |
1875 | br_write_unlock(vfsmount_lock); | |
1876 | up_write(&namespace_sem); | |
1877 | } | |
1878 | EXPORT_SYMBOL(mnt_set_expiry); | |
1879 | ||
1da177e4 LT |
1880 | /* |
1881 | * process a list of expirable mountpoints with the intent of discarding any | |
1882 | * mountpoints that aren't in use and haven't been touched since last we came | |
1883 | * here | |
1884 | */ | |
1885 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1886 | { | |
761d5c38 | 1887 | struct mount *mnt, *next; |
1da177e4 | 1888 | LIST_HEAD(graveyard); |
bcc5c7d2 | 1889 | LIST_HEAD(umounts); |
1da177e4 LT |
1890 | |
1891 | if (list_empty(mounts)) | |
1892 | return; | |
1893 | ||
bcc5c7d2 | 1894 | down_write(&namespace_sem); |
99b7db7b | 1895 | br_write_lock(vfsmount_lock); |
1da177e4 LT |
1896 | |
1897 | /* extract from the expiration list every vfsmount that matches the | |
1898 | * following criteria: | |
1899 | * - only referenced by its parent vfsmount | |
1900 | * - still marked for expiry (marked on the last call here; marks are | |
1901 | * cleared by mntput()) | |
1902 | */ | |
6776db3d | 1903 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 1904 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 1905 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1906 | continue; |
6776db3d | 1907 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1908 | } |
bcc5c7d2 | 1909 | while (!list_empty(&graveyard)) { |
6776db3d | 1910 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 1911 | touch_mnt_namespace(mnt->mnt_ns); |
bcc5c7d2 AV |
1912 | umount_tree(mnt, 1, &umounts); |
1913 | } | |
99b7db7b | 1914 | br_write_unlock(vfsmount_lock); |
bcc5c7d2 AV |
1915 | up_write(&namespace_sem); |
1916 | ||
1917 | release_mounts(&umounts); | |
5528f911 TM |
1918 | } |
1919 | ||
1920 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1921 | ||
1922 | /* | |
1923 | * Ripoff of 'select_parent()' | |
1924 | * | |
1925 | * search the list of submounts for a given mountpoint, and move any | |
1926 | * shrinkable submounts to the 'graveyard' list. | |
1927 | */ | |
692afc31 | 1928 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 1929 | { |
692afc31 | 1930 | struct mount *this_parent = parent; |
5528f911 TM |
1931 | struct list_head *next; |
1932 | int found = 0; | |
1933 | ||
1934 | repeat: | |
6b41d536 | 1935 | next = this_parent->mnt_mounts.next; |
5528f911 | 1936 | resume: |
6b41d536 | 1937 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 1938 | struct list_head *tmp = next; |
6b41d536 | 1939 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
1940 | |
1941 | next = tmp->next; | |
692afc31 | 1942 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 1943 | continue; |
5528f911 TM |
1944 | /* |
1945 | * Descend a level if the d_mounts list is non-empty. | |
1946 | */ | |
6b41d536 | 1947 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
1948 | this_parent = mnt; |
1949 | goto repeat; | |
1950 | } | |
1da177e4 | 1951 | |
1ab59738 | 1952 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 1953 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
1954 | found++; |
1955 | } | |
1da177e4 | 1956 | } |
5528f911 TM |
1957 | /* |
1958 | * All done at this level ... ascend and resume the search | |
1959 | */ | |
1960 | if (this_parent != parent) { | |
6b41d536 | 1961 | next = this_parent->mnt_child.next; |
0714a533 | 1962 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
1963 | goto resume; |
1964 | } | |
1965 | return found; | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * process a list of expirable mountpoints with the intent of discarding any | |
1970 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
1971 | * |
1972 | * vfsmount_lock must be held for write | |
5528f911 | 1973 | */ |
692afc31 | 1974 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts) |
5528f911 TM |
1975 | { |
1976 | LIST_HEAD(graveyard); | |
761d5c38 | 1977 | struct mount *m; |
5528f911 | 1978 | |
5528f911 | 1979 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 1980 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 1981 | while (!list_empty(&graveyard)) { |
761d5c38 | 1982 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 1983 | mnt_expire); |
143c8c91 | 1984 | touch_mnt_namespace(m->mnt_ns); |
afef80b3 | 1985 | umount_tree(m, 1, umounts); |
bcc5c7d2 AV |
1986 | } |
1987 | } | |
1da177e4 LT |
1988 | } |
1989 | ||
1da177e4 LT |
1990 | /* |
1991 | * Some copy_from_user() implementations do not return the exact number of | |
1992 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1993 | * Note that this function differs from copy_from_user() in that it will oops | |
1994 | * on bad values of `to', rather than returning a short copy. | |
1995 | */ | |
b58fed8b RP |
1996 | static long exact_copy_from_user(void *to, const void __user * from, |
1997 | unsigned long n) | |
1da177e4 LT |
1998 | { |
1999 | char *t = to; | |
2000 | const char __user *f = from; | |
2001 | char c; | |
2002 | ||
2003 | if (!access_ok(VERIFY_READ, from, n)) | |
2004 | return n; | |
2005 | ||
2006 | while (n) { | |
2007 | if (__get_user(c, f)) { | |
2008 | memset(t, 0, n); | |
2009 | break; | |
2010 | } | |
2011 | *t++ = c; | |
2012 | f++; | |
2013 | n--; | |
2014 | } | |
2015 | return n; | |
2016 | } | |
2017 | ||
b58fed8b | 2018 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2019 | { |
2020 | int i; | |
2021 | unsigned long page; | |
2022 | unsigned long size; | |
b58fed8b | 2023 | |
1da177e4 LT |
2024 | *where = 0; |
2025 | if (!data) | |
2026 | return 0; | |
2027 | ||
2028 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2029 | return -ENOMEM; | |
2030 | ||
2031 | /* We only care that *some* data at the address the user | |
2032 | * gave us is valid. Just in case, we'll zero | |
2033 | * the remainder of the page. | |
2034 | */ | |
2035 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2036 | size = TASK_SIZE - (unsigned long)data; | |
2037 | if (size > PAGE_SIZE) | |
2038 | size = PAGE_SIZE; | |
2039 | ||
2040 | i = size - exact_copy_from_user((void *)page, data, size); | |
2041 | if (!i) { | |
b58fed8b | 2042 | free_page(page); |
1da177e4 LT |
2043 | return -EFAULT; |
2044 | } | |
2045 | if (i != PAGE_SIZE) | |
2046 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2047 | *where = page; | |
2048 | return 0; | |
2049 | } | |
2050 | ||
eca6f534 VN |
2051 | int copy_mount_string(const void __user *data, char **where) |
2052 | { | |
2053 | char *tmp; | |
2054 | ||
2055 | if (!data) { | |
2056 | *where = NULL; | |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | tmp = strndup_user(data, PAGE_SIZE); | |
2061 | if (IS_ERR(tmp)) | |
2062 | return PTR_ERR(tmp); | |
2063 | ||
2064 | *where = tmp; | |
2065 | return 0; | |
2066 | } | |
2067 | ||
1da177e4 LT |
2068 | /* |
2069 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2070 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2071 | * | |
2072 | * data is a (void *) that can point to any structure up to | |
2073 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2074 | * information (or be NULL). | |
2075 | * | |
2076 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2077 | * When the flags word was introduced its top half was required | |
2078 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2079 | * Therefore, if this magic number is present, it carries no information | |
2080 | * and must be discarded. | |
2081 | */ | |
b58fed8b | 2082 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
2083 | unsigned long flags, void *data_page) |
2084 | { | |
2d92ab3c | 2085 | struct path path; |
1da177e4 LT |
2086 | int retval = 0; |
2087 | int mnt_flags = 0; | |
2088 | ||
2089 | /* Discard magic */ | |
2090 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2091 | flags &= ~MS_MGC_MSK; | |
2092 | ||
2093 | /* Basic sanity checks */ | |
2094 | ||
2095 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2096 | return -EINVAL; | |
1da177e4 LT |
2097 | |
2098 | if (data_page) | |
2099 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2100 | ||
a27ab9f2 TH |
2101 | /* ... and get the mountpoint */ |
2102 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2103 | if (retval) | |
2104 | return retval; | |
2105 | ||
2106 | retval = security_sb_mount(dev_name, &path, | |
2107 | type_page, flags, data_page); | |
2108 | if (retval) | |
2109 | goto dput_out; | |
2110 | ||
613cbe3d AK |
2111 | /* Default to relatime unless overriden */ |
2112 | if (!(flags & MS_NOATIME)) | |
2113 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2114 | |
1da177e4 LT |
2115 | /* Separate the per-mountpoint flags */ |
2116 | if (flags & MS_NOSUID) | |
2117 | mnt_flags |= MNT_NOSUID; | |
2118 | if (flags & MS_NODEV) | |
2119 | mnt_flags |= MNT_NODEV; | |
2120 | if (flags & MS_NOEXEC) | |
2121 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2122 | if (flags & MS_NOATIME) |
2123 | mnt_flags |= MNT_NOATIME; | |
2124 | if (flags & MS_NODIRATIME) | |
2125 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2126 | if (flags & MS_STRICTATIME) |
2127 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2128 | if (flags & MS_RDONLY) |
2129 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2130 | |
7a4dec53 | 2131 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2132 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2133 | MS_STRICTATIME); | |
1da177e4 | 2134 | |
1da177e4 | 2135 | if (flags & MS_REMOUNT) |
2d92ab3c | 2136 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2137 | data_page); |
2138 | else if (flags & MS_BIND) | |
2d92ab3c | 2139 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2140 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2141 | retval = do_change_type(&path, flags); |
1da177e4 | 2142 | else if (flags & MS_MOVE) |
2d92ab3c | 2143 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2144 | else |
2d92ab3c | 2145 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2146 | dev_name, data_page); |
2147 | dput_out: | |
2d92ab3c | 2148 | path_put(&path); |
1da177e4 LT |
2149 | return retval; |
2150 | } | |
2151 | ||
cf8d2c11 TM |
2152 | static struct mnt_namespace *alloc_mnt_ns(void) |
2153 | { | |
2154 | struct mnt_namespace *new_ns; | |
2155 | ||
2156 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2157 | if (!new_ns) | |
2158 | return ERR_PTR(-ENOMEM); | |
2159 | atomic_set(&new_ns->count, 1); | |
2160 | new_ns->root = NULL; | |
2161 | INIT_LIST_HEAD(&new_ns->list); | |
2162 | init_waitqueue_head(&new_ns->poll); | |
2163 | new_ns->event = 0; | |
2164 | return new_ns; | |
2165 | } | |
2166 | ||
f03c6599 AV |
2167 | void mnt_make_longterm(struct vfsmount *mnt) |
2168 | { | |
83adc753 | 2169 | __mnt_make_longterm(real_mount(mnt)); |
f03c6599 AV |
2170 | } |
2171 | ||
83adc753 | 2172 | void mnt_make_shortterm(struct vfsmount *m) |
f03c6599 | 2173 | { |
7e3d0eb0 | 2174 | #ifdef CONFIG_SMP |
83adc753 | 2175 | struct mount *mnt = real_mount(m); |
68e8a9fe | 2176 | if (atomic_add_unless(&mnt->mnt_longterm, -1, 1)) |
f03c6599 AV |
2177 | return; |
2178 | br_write_lock(vfsmount_lock); | |
68e8a9fe | 2179 | atomic_dec(&mnt->mnt_longterm); |
f03c6599 | 2180 | br_write_unlock(vfsmount_lock); |
7e3d0eb0 | 2181 | #endif |
f03c6599 AV |
2182 | } |
2183 | ||
741a2951 JD |
2184 | /* |
2185 | * Allocate a new namespace structure and populate it with contents | |
2186 | * copied from the namespace of the passed in task structure. | |
2187 | */ | |
e3222c4e | 2188 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 2189 | struct fs_struct *fs) |
1da177e4 | 2190 | { |
6b3286ed | 2191 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2192 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2193 | struct mount *p, *q; |
be08d6d2 | 2194 | struct mount *old = mnt_ns->root; |
cb338d06 | 2195 | struct mount *new; |
1da177e4 | 2196 | |
cf8d2c11 TM |
2197 | new_ns = alloc_mnt_ns(); |
2198 | if (IS_ERR(new_ns)) | |
2199 | return new_ns; | |
1da177e4 | 2200 | |
390c6843 | 2201 | down_write(&namespace_sem); |
1da177e4 | 2202 | /* First pass: copy the tree topology */ |
909b0a88 | 2203 | new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE); |
cb338d06 | 2204 | if (!new) { |
390c6843 | 2205 | up_write(&namespace_sem); |
1da177e4 | 2206 | kfree(new_ns); |
5cc4a034 | 2207 | return ERR_PTR(-ENOMEM); |
1da177e4 | 2208 | } |
be08d6d2 | 2209 | new_ns->root = new; |
99b7db7b | 2210 | br_write_lock(vfsmount_lock); |
1a4eeaf2 | 2211 | list_add_tail(&new_ns->list, &new->mnt_list); |
99b7db7b | 2212 | br_write_unlock(vfsmount_lock); |
1da177e4 LT |
2213 | |
2214 | /* | |
2215 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2216 | * as belonging to new namespace. We have already acquired a private | |
2217 | * fs_struct, so tsk->fs->lock is not needed. | |
2218 | */ | |
909b0a88 | 2219 | p = old; |
cb338d06 | 2220 | q = new; |
1da177e4 | 2221 | while (p) { |
143c8c91 | 2222 | q->mnt_ns = new_ns; |
83adc753 | 2223 | __mnt_make_longterm(q); |
1da177e4 | 2224 | if (fs) { |
315fc83e AV |
2225 | if (&p->mnt == fs->root.mnt) { |
2226 | fs->root.mnt = mntget(&q->mnt); | |
83adc753 | 2227 | __mnt_make_longterm(q); |
315fc83e AV |
2228 | mnt_make_shortterm(&p->mnt); |
2229 | rootmnt = &p->mnt; | |
1da177e4 | 2230 | } |
315fc83e AV |
2231 | if (&p->mnt == fs->pwd.mnt) { |
2232 | fs->pwd.mnt = mntget(&q->mnt); | |
83adc753 | 2233 | __mnt_make_longterm(q); |
315fc83e AV |
2234 | mnt_make_shortterm(&p->mnt); |
2235 | pwdmnt = &p->mnt; | |
1da177e4 | 2236 | } |
1da177e4 | 2237 | } |
909b0a88 AV |
2238 | p = next_mnt(p, old); |
2239 | q = next_mnt(q, new); | |
1da177e4 | 2240 | } |
390c6843 | 2241 | up_write(&namespace_sem); |
1da177e4 | 2242 | |
1da177e4 | 2243 | if (rootmnt) |
f03c6599 | 2244 | mntput(rootmnt); |
1da177e4 | 2245 | if (pwdmnt) |
f03c6599 | 2246 | mntput(pwdmnt); |
1da177e4 | 2247 | |
741a2951 JD |
2248 | return new_ns; |
2249 | } | |
2250 | ||
213dd266 | 2251 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2252 | struct fs_struct *new_fs) |
741a2951 | 2253 | { |
6b3286ed | 2254 | struct mnt_namespace *new_ns; |
741a2951 | 2255 | |
e3222c4e | 2256 | BUG_ON(!ns); |
6b3286ed | 2257 | get_mnt_ns(ns); |
741a2951 JD |
2258 | |
2259 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2260 | return ns; |
741a2951 | 2261 | |
e3222c4e | 2262 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2263 | |
6b3286ed | 2264 | put_mnt_ns(ns); |
e3222c4e | 2265 | return new_ns; |
1da177e4 LT |
2266 | } |
2267 | ||
cf8d2c11 TM |
2268 | /** |
2269 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2270 | * @mnt: pointer to the new root filesystem mountpoint | |
2271 | */ | |
1a4eeaf2 | 2272 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2273 | { |
1a4eeaf2 | 2274 | struct mnt_namespace *new_ns = alloc_mnt_ns(); |
cf8d2c11 | 2275 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2276 | struct mount *mnt = real_mount(m); |
2277 | mnt->mnt_ns = new_ns; | |
2278 | __mnt_make_longterm(mnt); | |
be08d6d2 | 2279 | new_ns->root = mnt; |
1a4eeaf2 | 2280 | list_add(&new_ns->list, &mnt->mnt_list); |
c1334495 | 2281 | } else { |
1a4eeaf2 | 2282 | mntput(m); |
cf8d2c11 TM |
2283 | } |
2284 | return new_ns; | |
2285 | } | |
cf8d2c11 | 2286 | |
ea441d11 AV |
2287 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2288 | { | |
2289 | struct mnt_namespace *ns; | |
d31da0f0 | 2290 | struct super_block *s; |
ea441d11 AV |
2291 | struct path path; |
2292 | int err; | |
2293 | ||
2294 | ns = create_mnt_ns(mnt); | |
2295 | if (IS_ERR(ns)) | |
2296 | return ERR_CAST(ns); | |
2297 | ||
2298 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2299 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2300 | ||
2301 | put_mnt_ns(ns); | |
2302 | ||
2303 | if (err) | |
2304 | return ERR_PTR(err); | |
2305 | ||
2306 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2307 | s = path.mnt->mnt_sb; |
2308 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2309 | mntput(path.mnt); |
2310 | /* lock the sucker */ | |
d31da0f0 | 2311 | down_write(&s->s_umount); |
ea441d11 AV |
2312 | /* ... and return the root of (sub)tree on it */ |
2313 | return path.dentry; | |
2314 | } | |
2315 | EXPORT_SYMBOL(mount_subtree); | |
2316 | ||
bdc480e3 HC |
2317 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2318 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2319 | { |
eca6f534 VN |
2320 | int ret; |
2321 | char *kernel_type; | |
2322 | char *kernel_dir; | |
2323 | char *kernel_dev; | |
1da177e4 | 2324 | unsigned long data_page; |
1da177e4 | 2325 | |
eca6f534 VN |
2326 | ret = copy_mount_string(type, &kernel_type); |
2327 | if (ret < 0) | |
2328 | goto out_type; | |
1da177e4 | 2329 | |
eca6f534 VN |
2330 | kernel_dir = getname(dir_name); |
2331 | if (IS_ERR(kernel_dir)) { | |
2332 | ret = PTR_ERR(kernel_dir); | |
2333 | goto out_dir; | |
2334 | } | |
1da177e4 | 2335 | |
eca6f534 VN |
2336 | ret = copy_mount_string(dev_name, &kernel_dev); |
2337 | if (ret < 0) | |
2338 | goto out_dev; | |
1da177e4 | 2339 | |
eca6f534 VN |
2340 | ret = copy_mount_options(data, &data_page); |
2341 | if (ret < 0) | |
2342 | goto out_data; | |
1da177e4 | 2343 | |
eca6f534 VN |
2344 | ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, |
2345 | (void *) data_page); | |
1da177e4 | 2346 | |
eca6f534 VN |
2347 | free_page(data_page); |
2348 | out_data: | |
2349 | kfree(kernel_dev); | |
2350 | out_dev: | |
2351 | putname(kernel_dir); | |
2352 | out_dir: | |
2353 | kfree(kernel_type); | |
2354 | out_type: | |
2355 | return ret; | |
1da177e4 LT |
2356 | } |
2357 | ||
afac7cba AV |
2358 | /* |
2359 | * Return true if path is reachable from root | |
2360 | * | |
2361 | * namespace_sem or vfsmount_lock is held | |
2362 | */ | |
643822b4 | 2363 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2364 | const struct path *root) |
2365 | { | |
643822b4 | 2366 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2367 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2368 | mnt = mnt->mnt_parent; |
afac7cba | 2369 | } |
643822b4 | 2370 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2371 | } |
2372 | ||
2373 | int path_is_under(struct path *path1, struct path *path2) | |
2374 | { | |
2375 | int res; | |
2376 | br_read_lock(vfsmount_lock); | |
643822b4 | 2377 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
afac7cba AV |
2378 | br_read_unlock(vfsmount_lock); |
2379 | return res; | |
2380 | } | |
2381 | EXPORT_SYMBOL(path_is_under); | |
2382 | ||
1da177e4 LT |
2383 | /* |
2384 | * pivot_root Semantics: | |
2385 | * Moves the root file system of the current process to the directory put_old, | |
2386 | * makes new_root as the new root file system of the current process, and sets | |
2387 | * root/cwd of all processes which had them on the current root to new_root. | |
2388 | * | |
2389 | * Restrictions: | |
2390 | * The new_root and put_old must be directories, and must not be on the | |
2391 | * same file system as the current process root. The put_old must be | |
2392 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2393 | * pointed to by put_old must yield the same directory as new_root. No other | |
2394 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2395 | * | |
4a0d11fa NB |
2396 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2397 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2398 | * in this situation. | |
2399 | * | |
1da177e4 LT |
2400 | * Notes: |
2401 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2402 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2403 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2404 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2405 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2406 | * first. | |
2407 | */ | |
3480b257 HC |
2408 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2409 | const char __user *, put_old) | |
1da177e4 | 2410 | { |
2d8f3038 | 2411 | struct path new, old, parent_path, root_parent, root; |
419148da | 2412 | struct mount *new_mnt, *root_mnt; |
1da177e4 LT |
2413 | int error; |
2414 | ||
2415 | if (!capable(CAP_SYS_ADMIN)) | |
2416 | return -EPERM; | |
2417 | ||
2d8f3038 | 2418 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2419 | if (error) |
2420 | goto out0; | |
1da177e4 | 2421 | |
2d8f3038 | 2422 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2423 | if (error) |
2424 | goto out1; | |
2425 | ||
2d8f3038 | 2426 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2427 | if (error) |
2428 | goto out2; | |
1da177e4 | 2429 | |
f7ad3c6b | 2430 | get_fs_root(current->fs, &root); |
b12cea91 AV |
2431 | error = lock_mount(&old); |
2432 | if (error) | |
2433 | goto out3; | |
2434 | ||
1da177e4 | 2435 | error = -EINVAL; |
419148da AV |
2436 | new_mnt = real_mount(new.mnt); |
2437 | root_mnt = real_mount(root.mnt); | |
fc7be130 AV |
2438 | if (IS_MNT_SHARED(real_mount(old.mnt)) || |
2439 | IS_MNT_SHARED(new_mnt->mnt_parent) || | |
2440 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2441 | goto out4; |
143c8c91 | 2442 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2443 | goto out4; |
1da177e4 | 2444 | error = -ENOENT; |
f3da392e | 2445 | if (d_unlinked(new.dentry)) |
b12cea91 | 2446 | goto out4; |
f3da392e | 2447 | if (d_unlinked(old.dentry)) |
b12cea91 | 2448 | goto out4; |
1da177e4 | 2449 | error = -EBUSY; |
2d8f3038 AV |
2450 | if (new.mnt == root.mnt || |
2451 | old.mnt == root.mnt) | |
b12cea91 | 2452 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2453 | error = -EINVAL; |
8c3ee42e | 2454 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2455 | goto out4; /* not a mountpoint */ |
676da58d | 2456 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2457 | goto out4; /* not attached */ |
2d8f3038 | 2458 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2459 | goto out4; /* not a mountpoint */ |
676da58d | 2460 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2461 | goto out4; /* not attached */ |
4ac91378 | 2462 | /* make sure we can reach put_old from new_root */ |
643822b4 | 2463 | if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new)) |
b12cea91 | 2464 | goto out4; |
27cb1572 | 2465 | br_write_lock(vfsmount_lock); |
419148da AV |
2466 | detach_mnt(new_mnt, &parent_path); |
2467 | detach_mnt(root_mnt, &root_parent); | |
4ac91378 | 2468 | /* mount old root on put_old */ |
419148da | 2469 | attach_mnt(root_mnt, &old); |
4ac91378 | 2470 | /* mount new_root on / */ |
419148da | 2471 | attach_mnt(new_mnt, &root_parent); |
6b3286ed | 2472 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
99b7db7b | 2473 | br_write_unlock(vfsmount_lock); |
2d8f3038 | 2474 | chroot_fs_refs(&root, &new); |
1da177e4 | 2475 | error = 0; |
b12cea91 AV |
2476 | out4: |
2477 | unlock_mount(&old); | |
2478 | if (!error) { | |
2479 | path_put(&root_parent); | |
2480 | path_put(&parent_path); | |
2481 | } | |
2482 | out3: | |
8c3ee42e | 2483 | path_put(&root); |
b12cea91 | 2484 | out2: |
2d8f3038 | 2485 | path_put(&old); |
1da177e4 | 2486 | out1: |
2d8f3038 | 2487 | path_put(&new); |
1da177e4 | 2488 | out0: |
1da177e4 | 2489 | return error; |
1da177e4 LT |
2490 | } |
2491 | ||
2492 | static void __init init_mount_tree(void) | |
2493 | { | |
2494 | struct vfsmount *mnt; | |
6b3286ed | 2495 | struct mnt_namespace *ns; |
ac748a09 | 2496 | struct path root; |
1da177e4 LT |
2497 | |
2498 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2499 | if (IS_ERR(mnt)) | |
2500 | panic("Can't create rootfs"); | |
b3e19d92 | 2501 | |
3b22edc5 TM |
2502 | ns = create_mnt_ns(mnt); |
2503 | if (IS_ERR(ns)) | |
1da177e4 | 2504 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2505 | |
2506 | init_task.nsproxy->mnt_ns = ns; | |
2507 | get_mnt_ns(ns); | |
2508 | ||
be08d6d2 AV |
2509 | root.mnt = mnt; |
2510 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2511 | |
2512 | set_fs_pwd(current->fs, &root); | |
2513 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2514 | } |
2515 | ||
74bf17cf | 2516 | void __init mnt_init(void) |
1da177e4 | 2517 | { |
13f14b4d | 2518 | unsigned u; |
15a67dd8 | 2519 | int err; |
1da177e4 | 2520 | |
390c6843 RP |
2521 | init_rwsem(&namespace_sem); |
2522 | ||
7d6fec45 | 2523 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2524 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2525 | |
b58fed8b | 2526 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2527 | |
2528 | if (!mount_hashtable) | |
2529 | panic("Failed to allocate mount hash table\n"); | |
2530 | ||
80cdc6da | 2531 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2532 | |
2533 | for (u = 0; u < HASH_SIZE; u++) | |
2534 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2535 | |
99b7db7b NP |
2536 | br_lock_init(vfsmount_lock); |
2537 | ||
15a67dd8 RD |
2538 | err = sysfs_init(); |
2539 | if (err) | |
2540 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2541 | __func__, err); |
00d26666 GKH |
2542 | fs_kobj = kobject_create_and_add("fs", NULL); |
2543 | if (!fs_kobj) | |
8e24eea7 | 2544 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2545 | init_rootfs(); |
2546 | init_mount_tree(); | |
2547 | } | |
2548 | ||
616511d0 | 2549 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2550 | { |
70fbcdf4 | 2551 | LIST_HEAD(umount_list); |
616511d0 | 2552 | |
d498b25a | 2553 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2554 | return; |
390c6843 | 2555 | down_write(&namespace_sem); |
99b7db7b | 2556 | br_write_lock(vfsmount_lock); |
be08d6d2 | 2557 | umount_tree(ns->root, 0, &umount_list); |
99b7db7b | 2558 | br_write_unlock(vfsmount_lock); |
390c6843 | 2559 | up_write(&namespace_sem); |
70fbcdf4 | 2560 | release_mounts(&umount_list); |
6b3286ed | 2561 | kfree(ns); |
1da177e4 | 2562 | } |
9d412a43 AV |
2563 | |
2564 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2565 | { | |
423e0ab0 TC |
2566 | struct vfsmount *mnt; |
2567 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2568 | if (!IS_ERR(mnt)) { | |
2569 | /* | |
2570 | * it is a longterm mount, don't release mnt until | |
2571 | * we unmount before file sys is unregistered | |
2572 | */ | |
2573 | mnt_make_longterm(mnt); | |
2574 | } | |
2575 | return mnt; | |
9d412a43 AV |
2576 | } |
2577 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2578 | |
2579 | void kern_unmount(struct vfsmount *mnt) | |
2580 | { | |
2581 | /* release long term mount so mount point can be released */ | |
2582 | if (!IS_ERR_OR_NULL(mnt)) { | |
2583 | mnt_make_shortterm(mnt); | |
2584 | mntput(mnt); | |
2585 | } | |
2586 | } | |
2587 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2588 | |
2589 | bool our_mnt(struct vfsmount *mnt) | |
2590 | { | |
143c8c91 | 2591 | return check_mnt(real_mount(mnt)); |
02125a82 | 2592 | } |