]>
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 | ||
4ed5e82f MS |
276 | static int mnt_is_readonly(struct vfsmount *mnt) |
277 | { | |
278 | if (mnt->mnt_sb->s_readonly_remount) | |
279 | return 1; | |
280 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
281 | smp_rmb(); | |
282 | return __mnt_is_readonly(mnt); | |
283 | } | |
284 | ||
8366025e DH |
285 | /* |
286 | * Most r/o checks on a fs are for operations that take | |
287 | * discrete amounts of time, like a write() or unlink(). | |
288 | * We must keep track of when those operations start | |
289 | * (for permission checks) and when they end, so that | |
290 | * we can determine when writes are able to occur to | |
291 | * a filesystem. | |
292 | */ | |
293 | /** | |
294 | * mnt_want_write - get write access to a mount | |
83adc753 | 295 | * @m: the mount on which to take a write |
8366025e DH |
296 | * |
297 | * This tells the low-level filesystem that a write is | |
298 | * about to be performed to it, and makes sure that | |
299 | * writes are allowed before returning success. When | |
300 | * the write operation is finished, mnt_drop_write() | |
301 | * must be called. This is effectively a refcount. | |
302 | */ | |
83adc753 | 303 | int mnt_want_write(struct vfsmount *m) |
8366025e | 304 | { |
83adc753 | 305 | struct mount *mnt = real_mount(m); |
3d733633 | 306 | int ret = 0; |
3d733633 | 307 | |
d3ef3d73 | 308 | preempt_disable(); |
c6653a83 | 309 | mnt_inc_writers(mnt); |
d3ef3d73 | 310 | /* |
c6653a83 | 311 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 312 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
313 | * incremented count after it has set MNT_WRITE_HOLD. | |
314 | */ | |
315 | smp_mb(); | |
83adc753 | 316 | while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) |
d3ef3d73 | 317 | cpu_relax(); |
318 | /* | |
319 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
320 | * be set to match its requirements. So we must not load that until | |
321 | * MNT_WRITE_HOLD is cleared. | |
322 | */ | |
323 | smp_rmb(); | |
4ed5e82f | 324 | if (mnt_is_readonly(m)) { |
c6653a83 | 325 | mnt_dec_writers(mnt); |
3d733633 | 326 | ret = -EROFS; |
3d733633 | 327 | } |
d3ef3d73 | 328 | preempt_enable(); |
3d733633 | 329 | return ret; |
8366025e DH |
330 | } |
331 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
332 | ||
96029c4e | 333 | /** |
334 | * mnt_clone_write - get write access to a mount | |
335 | * @mnt: the mount on which to take a write | |
336 | * | |
337 | * This is effectively like mnt_want_write, except | |
338 | * it must only be used to take an extra write reference | |
339 | * on a mountpoint that we already know has a write reference | |
340 | * on it. This allows some optimisation. | |
341 | * | |
342 | * After finished, mnt_drop_write must be called as usual to | |
343 | * drop the reference. | |
344 | */ | |
345 | int mnt_clone_write(struct vfsmount *mnt) | |
346 | { | |
347 | /* superblock may be r/o */ | |
348 | if (__mnt_is_readonly(mnt)) | |
349 | return -EROFS; | |
350 | preempt_disable(); | |
83adc753 | 351 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 352 | preempt_enable(); |
353 | return 0; | |
354 | } | |
355 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
356 | ||
357 | /** | |
358 | * mnt_want_write_file - get write access to a file's mount | |
359 | * @file: the file who's mount on which to take a write | |
360 | * | |
361 | * This is like mnt_want_write, but it takes a file and can | |
362 | * do some optimisations if the file is open for write already | |
363 | */ | |
364 | int mnt_want_write_file(struct file *file) | |
365 | { | |
2d8dd38a OH |
366 | struct inode *inode = file->f_dentry->d_inode; |
367 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) | |
96029c4e | 368 | return mnt_want_write(file->f_path.mnt); |
369 | else | |
370 | return mnt_clone_write(file->f_path.mnt); | |
371 | } | |
372 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
373 | ||
8366025e DH |
374 | /** |
375 | * mnt_drop_write - give up write access to a mount | |
376 | * @mnt: the mount on which to give up write access | |
377 | * | |
378 | * Tells the low-level filesystem that we are done | |
379 | * performing writes to it. Must be matched with | |
380 | * mnt_want_write() call above. | |
381 | */ | |
382 | void mnt_drop_write(struct vfsmount *mnt) | |
383 | { | |
d3ef3d73 | 384 | preempt_disable(); |
83adc753 | 385 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 386 | preempt_enable(); |
8366025e DH |
387 | } |
388 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
389 | ||
2a79f17e AV |
390 | void mnt_drop_write_file(struct file *file) |
391 | { | |
392 | mnt_drop_write(file->f_path.mnt); | |
393 | } | |
394 | EXPORT_SYMBOL(mnt_drop_write_file); | |
395 | ||
83adc753 | 396 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 397 | { |
3d733633 DH |
398 | int ret = 0; |
399 | ||
962830df | 400 | br_write_lock(&vfsmount_lock); |
83adc753 | 401 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 402 | /* |
d3ef3d73 | 403 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
404 | * should be visible before we do. | |
3d733633 | 405 | */ |
d3ef3d73 | 406 | smp_mb(); |
407 | ||
3d733633 | 408 | /* |
d3ef3d73 | 409 | * With writers on hold, if this value is zero, then there are |
410 | * definitely no active writers (although held writers may subsequently | |
411 | * increment the count, they'll have to wait, and decrement it after | |
412 | * seeing MNT_READONLY). | |
413 | * | |
414 | * It is OK to have counter incremented on one CPU and decremented on | |
415 | * another: the sum will add up correctly. The danger would be when we | |
416 | * sum up each counter, if we read a counter before it is incremented, | |
417 | * but then read another CPU's count which it has been subsequently | |
418 | * decremented from -- we would see more decrements than we should. | |
419 | * MNT_WRITE_HOLD protects against this scenario, because | |
420 | * mnt_want_write first increments count, then smp_mb, then spins on | |
421 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
422 | * we're counting up here. | |
3d733633 | 423 | */ |
c6653a83 | 424 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 425 | ret = -EBUSY; |
426 | else | |
83adc753 | 427 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 428 | /* |
429 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
430 | * that become unheld will see MNT_READONLY. | |
431 | */ | |
432 | smp_wmb(); | |
83adc753 | 433 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
962830df | 434 | br_write_unlock(&vfsmount_lock); |
3d733633 | 435 | return ret; |
8366025e | 436 | } |
8366025e | 437 | |
83adc753 | 438 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 439 | { |
962830df | 440 | br_write_lock(&vfsmount_lock); |
83adc753 | 441 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
962830df | 442 | br_write_unlock(&vfsmount_lock); |
2e4b7fcd DH |
443 | } |
444 | ||
4ed5e82f MS |
445 | int sb_prepare_remount_readonly(struct super_block *sb) |
446 | { | |
447 | struct mount *mnt; | |
448 | int err = 0; | |
449 | ||
8e8b8796 MS |
450 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
451 | if (atomic_long_read(&sb->s_remove_count)) | |
452 | return -EBUSY; | |
453 | ||
962830df | 454 | br_write_lock(&vfsmount_lock); |
4ed5e82f MS |
455 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
456 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
457 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
458 | smp_mb(); | |
459 | if (mnt_get_writers(mnt) > 0) { | |
460 | err = -EBUSY; | |
461 | break; | |
462 | } | |
463 | } | |
464 | } | |
8e8b8796 MS |
465 | if (!err && atomic_long_read(&sb->s_remove_count)) |
466 | err = -EBUSY; | |
467 | ||
4ed5e82f MS |
468 | if (!err) { |
469 | sb->s_readonly_remount = 1; | |
470 | smp_wmb(); | |
471 | } | |
472 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
473 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
474 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
475 | } | |
962830df | 476 | br_write_unlock(&vfsmount_lock); |
4ed5e82f MS |
477 | |
478 | return err; | |
479 | } | |
480 | ||
b105e270 | 481 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 482 | { |
52ba1621 | 483 | kfree(mnt->mnt_devname); |
73cd49ec | 484 | mnt_free_id(mnt); |
d3ef3d73 | 485 | #ifdef CONFIG_SMP |
68e8a9fe | 486 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 487 | #endif |
b105e270 | 488 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
489 | } |
490 | ||
491 | /* | |
a05964f3 RP |
492 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
493 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 494 | * vfsmount_lock must be held for read or write. |
1da177e4 | 495 | */ |
c7105365 | 496 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 497 | int dir) |
1da177e4 | 498 | { |
b58fed8b RP |
499 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
500 | struct list_head *tmp = head; | |
c7105365 | 501 | struct mount *p, *found = NULL; |
1da177e4 | 502 | |
1da177e4 | 503 | for (;;) { |
a05964f3 | 504 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
505 | p = NULL; |
506 | if (tmp == head) | |
507 | break; | |
1b8e5564 | 508 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 509 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 510 | found = p; |
1da177e4 LT |
511 | break; |
512 | } | |
513 | } | |
1da177e4 LT |
514 | return found; |
515 | } | |
516 | ||
a05964f3 | 517 | /* |
f015f126 DH |
518 | * lookup_mnt - Return the first child mount mounted at path |
519 | * | |
520 | * "First" means first mounted chronologically. If you create the | |
521 | * following mounts: | |
522 | * | |
523 | * mount /dev/sda1 /mnt | |
524 | * mount /dev/sda2 /mnt | |
525 | * mount /dev/sda3 /mnt | |
526 | * | |
527 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
528 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
529 | * /dev/sda2, then /dev/sda3, then NULL. | |
530 | * | |
531 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 532 | */ |
1c755af4 | 533 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 534 | { |
c7105365 | 535 | struct mount *child_mnt; |
99b7db7b | 536 | |
962830df | 537 | br_read_lock(&vfsmount_lock); |
c7105365 AV |
538 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
539 | if (child_mnt) { | |
540 | mnt_add_count(child_mnt, 1); | |
962830df | 541 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
542 | return &child_mnt->mnt; |
543 | } else { | |
962830df | 544 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
545 | return NULL; |
546 | } | |
a05964f3 RP |
547 | } |
548 | ||
143c8c91 | 549 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 550 | { |
6b3286ed | 551 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
552 | } |
553 | ||
99b7db7b NP |
554 | /* |
555 | * vfsmount lock must be held for write | |
556 | */ | |
6b3286ed | 557 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
558 | { |
559 | if (ns) { | |
560 | ns->event = ++event; | |
561 | wake_up_interruptible(&ns->poll); | |
562 | } | |
563 | } | |
564 | ||
99b7db7b NP |
565 | /* |
566 | * vfsmount lock must be held for write | |
567 | */ | |
6b3286ed | 568 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
569 | { |
570 | if (ns && ns->event != event) { | |
571 | ns->event = event; | |
572 | wake_up_interruptible(&ns->poll); | |
573 | } | |
574 | } | |
575 | ||
5f57cbcc NP |
576 | /* |
577 | * Clear dentry's mounted state if it has no remaining mounts. | |
578 | * vfsmount_lock must be held for write. | |
579 | */ | |
aa0a4cf0 | 580 | static void dentry_reset_mounted(struct dentry *dentry) |
5f57cbcc NP |
581 | { |
582 | unsigned u; | |
583 | ||
584 | for (u = 0; u < HASH_SIZE; u++) { | |
d5e50f74 | 585 | struct mount *p; |
5f57cbcc | 586 | |
1b8e5564 | 587 | list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { |
a73324da | 588 | if (p->mnt_mountpoint == dentry) |
5f57cbcc NP |
589 | return; |
590 | } | |
591 | } | |
592 | spin_lock(&dentry->d_lock); | |
593 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
594 | spin_unlock(&dentry->d_lock); | |
595 | } | |
596 | ||
99b7db7b NP |
597 | /* |
598 | * vfsmount lock must be held for write | |
599 | */ | |
419148da AV |
600 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
601 | { | |
a73324da | 602 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
603 | old_path->mnt = &mnt->mnt_parent->mnt; |
604 | mnt->mnt_parent = mnt; | |
a73324da | 605 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 606 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 607 | list_del_init(&mnt->mnt_hash); |
aa0a4cf0 | 608 | dentry_reset_mounted(old_path->dentry); |
1da177e4 LT |
609 | } |
610 | ||
99b7db7b NP |
611 | /* |
612 | * vfsmount lock must be held for write | |
613 | */ | |
14cf1fa8 | 614 | void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry, |
44d964d6 | 615 | struct mount *child_mnt) |
b90fa9ae | 616 | { |
3a2393d7 | 617 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
a73324da | 618 | child_mnt->mnt_mountpoint = dget(dentry); |
3a2393d7 | 619 | child_mnt->mnt_parent = mnt; |
5f57cbcc NP |
620 | spin_lock(&dentry->d_lock); |
621 | dentry->d_flags |= DCACHE_MOUNTED; | |
622 | spin_unlock(&dentry->d_lock); | |
b90fa9ae RP |
623 | } |
624 | ||
99b7db7b NP |
625 | /* |
626 | * vfsmount lock must be held for write | |
627 | */ | |
419148da | 628 | static void attach_mnt(struct mount *mnt, struct path *path) |
1da177e4 | 629 | { |
14cf1fa8 | 630 | mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt); |
1b8e5564 | 631 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 | 632 | hash(path->mnt, path->dentry)); |
6b41d536 | 633 | list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts); |
b90fa9ae RP |
634 | } |
635 | ||
636 | /* | |
99b7db7b | 637 | * vfsmount lock must be held for write |
b90fa9ae | 638 | */ |
4b2619a5 | 639 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 640 | { |
0714a533 | 641 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 642 | struct mount *m; |
b90fa9ae | 643 | LIST_HEAD(head); |
143c8c91 | 644 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 645 | |
0714a533 | 646 | BUG_ON(parent == mnt); |
b90fa9ae | 647 | |
1a4eeaf2 | 648 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 649 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 650 | m->mnt_ns = n; |
f03c6599 | 651 | |
b90fa9ae RP |
652 | list_splice(&head, n->list.prev); |
653 | ||
1b8e5564 | 654 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 655 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 656 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 657 | touch_mnt_namespace(n); |
1da177e4 LT |
658 | } |
659 | ||
909b0a88 | 660 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 661 | { |
6b41d536 AV |
662 | struct list_head *next = p->mnt_mounts.next; |
663 | if (next == &p->mnt_mounts) { | |
1da177e4 | 664 | while (1) { |
909b0a88 | 665 | if (p == root) |
1da177e4 | 666 | return NULL; |
6b41d536 AV |
667 | next = p->mnt_child.next; |
668 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 669 | break; |
0714a533 | 670 | p = p->mnt_parent; |
1da177e4 LT |
671 | } |
672 | } | |
6b41d536 | 673 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
674 | } |
675 | ||
315fc83e | 676 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 677 | { |
6b41d536 AV |
678 | struct list_head *prev = p->mnt_mounts.prev; |
679 | while (prev != &p->mnt_mounts) { | |
680 | p = list_entry(prev, struct mount, mnt_child); | |
681 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
682 | } |
683 | return p; | |
684 | } | |
685 | ||
9d412a43 AV |
686 | struct vfsmount * |
687 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
688 | { | |
b105e270 | 689 | struct mount *mnt; |
9d412a43 AV |
690 | struct dentry *root; |
691 | ||
692 | if (!type) | |
693 | return ERR_PTR(-ENODEV); | |
694 | ||
695 | mnt = alloc_vfsmnt(name); | |
696 | if (!mnt) | |
697 | return ERR_PTR(-ENOMEM); | |
698 | ||
699 | if (flags & MS_KERNMOUNT) | |
b105e270 | 700 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
701 | |
702 | root = mount_fs(type, flags, name, data); | |
703 | if (IS_ERR(root)) { | |
704 | free_vfsmnt(mnt); | |
705 | return ERR_CAST(root); | |
706 | } | |
707 | ||
b105e270 AV |
708 | mnt->mnt.mnt_root = root; |
709 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 710 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 711 | mnt->mnt_parent = mnt; |
962830df | 712 | br_write_lock(&vfsmount_lock); |
39f7c4db | 713 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
962830df | 714 | br_write_unlock(&vfsmount_lock); |
b105e270 | 715 | return &mnt->mnt; |
9d412a43 AV |
716 | } |
717 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
718 | ||
87129cc0 | 719 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 720 | int flag) |
1da177e4 | 721 | { |
87129cc0 | 722 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
723 | struct mount *mnt; |
724 | int err; | |
1da177e4 | 725 | |
be34d1a3 DH |
726 | mnt = alloc_vfsmnt(old->mnt_devname); |
727 | if (!mnt) | |
728 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 729 | |
be34d1a3 DH |
730 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
731 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
732 | else | |
733 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 734 | |
be34d1a3 DH |
735 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
736 | err = mnt_alloc_group_id(mnt); | |
737 | if (err) | |
738 | goto out_free; | |
1da177e4 | 739 | } |
be34d1a3 DH |
740 | |
741 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
742 | atomic_inc(&sb->s_active); | |
743 | mnt->mnt.mnt_sb = sb; | |
744 | mnt->mnt.mnt_root = dget(root); | |
745 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
746 | mnt->mnt_parent = mnt; | |
747 | br_write_lock(&vfsmount_lock); | |
748 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
749 | br_write_unlock(&vfsmount_lock); | |
750 | ||
751 | if (flag & CL_SLAVE) { | |
752 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
753 | mnt->mnt_master = old; | |
754 | CLEAR_MNT_SHARED(mnt); | |
755 | } else if (!(flag & CL_PRIVATE)) { | |
756 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
757 | list_add(&mnt->mnt_share, &old->mnt_share); | |
758 | if (IS_MNT_SLAVE(old)) | |
759 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
760 | mnt->mnt_master = old->mnt_master; | |
761 | } | |
762 | if (flag & CL_MAKE_SHARED) | |
763 | set_mnt_shared(mnt); | |
764 | ||
765 | /* stick the duplicate mount on the same expiry list | |
766 | * as the original if that was on one */ | |
767 | if (flag & CL_EXPIRE) { | |
768 | if (!list_empty(&old->mnt_expire)) | |
769 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
770 | } | |
771 | ||
cb338d06 | 772 | return mnt; |
719f5d7f MS |
773 | |
774 | out_free: | |
775 | free_vfsmnt(mnt); | |
be34d1a3 | 776 | return ERR_PTR(err); |
1da177e4 LT |
777 | } |
778 | ||
83adc753 | 779 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 780 | { |
83adc753 AV |
781 | struct vfsmount *m = &mnt->mnt; |
782 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 783 | |
3d733633 DH |
784 | /* |
785 | * This probably indicates that somebody messed | |
786 | * up a mnt_want/drop_write() pair. If this | |
787 | * happens, the filesystem was probably unable | |
788 | * to make r/w->r/o transitions. | |
789 | */ | |
d3ef3d73 | 790 | /* |
b3e19d92 NP |
791 | * The locking used to deal with mnt_count decrement provides barriers, |
792 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 793 | */ |
c6653a83 | 794 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
795 | fsnotify_vfsmount_delete(m); |
796 | dput(m->mnt_root); | |
797 | free_vfsmnt(mnt); | |
1da177e4 LT |
798 | deactivate_super(sb); |
799 | } | |
800 | ||
900148dc | 801 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 802 | { |
b3e19d92 | 803 | put_again: |
f03c6599 | 804 | #ifdef CONFIG_SMP |
962830df | 805 | br_read_lock(&vfsmount_lock); |
f7a99c5b AV |
806 | if (likely(mnt->mnt_ns)) { |
807 | /* shouldn't be the last one */ | |
aa9c0e07 | 808 | mnt_add_count(mnt, -1); |
962830df | 809 | br_read_unlock(&vfsmount_lock); |
f03c6599 | 810 | return; |
b3e19d92 | 811 | } |
962830df | 812 | br_read_unlock(&vfsmount_lock); |
b3e19d92 | 813 | |
962830df | 814 | br_write_lock(&vfsmount_lock); |
aa9c0e07 | 815 | mnt_add_count(mnt, -1); |
b3e19d92 | 816 | if (mnt_get_count(mnt)) { |
962830df | 817 | br_write_unlock(&vfsmount_lock); |
99b7db7b NP |
818 | return; |
819 | } | |
b3e19d92 | 820 | #else |
aa9c0e07 | 821 | mnt_add_count(mnt, -1); |
b3e19d92 | 822 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 823 | return; |
962830df | 824 | br_write_lock(&vfsmount_lock); |
f03c6599 | 825 | #endif |
863d684f AV |
826 | if (unlikely(mnt->mnt_pinned)) { |
827 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
828 | mnt->mnt_pinned = 0; | |
962830df | 829 | br_write_unlock(&vfsmount_lock); |
900148dc | 830 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 831 | goto put_again; |
7b7b1ace | 832 | } |
962830df | 833 | |
39f7c4db | 834 | list_del(&mnt->mnt_instance); |
962830df | 835 | br_write_unlock(&vfsmount_lock); |
b3e19d92 NP |
836 | mntfree(mnt); |
837 | } | |
b3e19d92 NP |
838 | |
839 | void mntput(struct vfsmount *mnt) | |
840 | { | |
841 | if (mnt) { | |
863d684f | 842 | struct mount *m = real_mount(mnt); |
b3e19d92 | 843 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
844 | if (unlikely(m->mnt_expiry_mark)) |
845 | m->mnt_expiry_mark = 0; | |
846 | mntput_no_expire(m); | |
b3e19d92 NP |
847 | } |
848 | } | |
849 | EXPORT_SYMBOL(mntput); | |
850 | ||
851 | struct vfsmount *mntget(struct vfsmount *mnt) | |
852 | { | |
853 | if (mnt) | |
83adc753 | 854 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
855 | return mnt; |
856 | } | |
857 | EXPORT_SYMBOL(mntget); | |
858 | ||
7b7b1ace AV |
859 | void mnt_pin(struct vfsmount *mnt) |
860 | { | |
962830df | 861 | br_write_lock(&vfsmount_lock); |
863d684f | 862 | real_mount(mnt)->mnt_pinned++; |
962830df | 863 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 864 | } |
7b7b1ace AV |
865 | EXPORT_SYMBOL(mnt_pin); |
866 | ||
863d684f | 867 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 868 | { |
863d684f | 869 | struct mount *mnt = real_mount(m); |
962830df | 870 | br_write_lock(&vfsmount_lock); |
7b7b1ace | 871 | if (mnt->mnt_pinned) { |
863d684f | 872 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
873 | mnt->mnt_pinned--; |
874 | } | |
962830df | 875 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 876 | } |
7b7b1ace | 877 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 878 | |
b3b304a2 MS |
879 | static inline void mangle(struct seq_file *m, const char *s) |
880 | { | |
881 | seq_escape(m, s, " \t\n\\"); | |
882 | } | |
883 | ||
884 | /* | |
885 | * Simple .show_options callback for filesystems which don't want to | |
886 | * implement more complex mount option showing. | |
887 | * | |
888 | * See also save_mount_options(). | |
889 | */ | |
34c80b1d | 890 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 891 | { |
2a32cebd AV |
892 | const char *options; |
893 | ||
894 | rcu_read_lock(); | |
34c80b1d | 895 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
896 | |
897 | if (options != NULL && options[0]) { | |
898 | seq_putc(m, ','); | |
899 | mangle(m, options); | |
900 | } | |
2a32cebd | 901 | rcu_read_unlock(); |
b3b304a2 MS |
902 | |
903 | return 0; | |
904 | } | |
905 | EXPORT_SYMBOL(generic_show_options); | |
906 | ||
907 | /* | |
908 | * If filesystem uses generic_show_options(), this function should be | |
909 | * called from the fill_super() callback. | |
910 | * | |
911 | * The .remount_fs callback usually needs to be handled in a special | |
912 | * way, to make sure, that previous options are not overwritten if the | |
913 | * remount fails. | |
914 | * | |
915 | * Also note, that if the filesystem's .remount_fs function doesn't | |
916 | * reset all options to their default value, but changes only newly | |
917 | * given options, then the displayed options will not reflect reality | |
918 | * any more. | |
919 | */ | |
920 | void save_mount_options(struct super_block *sb, char *options) | |
921 | { | |
2a32cebd AV |
922 | BUG_ON(sb->s_options); |
923 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
924 | } |
925 | EXPORT_SYMBOL(save_mount_options); | |
926 | ||
2a32cebd AV |
927 | void replace_mount_options(struct super_block *sb, char *options) |
928 | { | |
929 | char *old = sb->s_options; | |
930 | rcu_assign_pointer(sb->s_options, options); | |
931 | if (old) { | |
932 | synchronize_rcu(); | |
933 | kfree(old); | |
934 | } | |
935 | } | |
936 | EXPORT_SYMBOL(replace_mount_options); | |
937 | ||
a1a2c409 | 938 | #ifdef CONFIG_PROC_FS |
0226f492 | 939 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
940 | static void *m_start(struct seq_file *m, loff_t *pos) |
941 | { | |
6ce6e24e | 942 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 943 | |
390c6843 | 944 | down_read(&namespace_sem); |
a1a2c409 | 945 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
946 | } |
947 | ||
948 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
949 | { | |
6ce6e24e | 950 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 951 | |
a1a2c409 | 952 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
953 | } |
954 | ||
955 | static void m_stop(struct seq_file *m, void *v) | |
956 | { | |
390c6843 | 957 | up_read(&namespace_sem); |
1da177e4 LT |
958 | } |
959 | ||
0226f492 | 960 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 961 | { |
6ce6e24e | 962 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 963 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 964 | return p->show(m, &r->mnt); |
1da177e4 LT |
965 | } |
966 | ||
a1a2c409 | 967 | const struct seq_operations mounts_op = { |
1da177e4 LT |
968 | .start = m_start, |
969 | .next = m_next, | |
970 | .stop = m_stop, | |
0226f492 | 971 | .show = m_show, |
b4629fe2 | 972 | }; |
a1a2c409 | 973 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 974 | |
1da177e4 LT |
975 | /** |
976 | * may_umount_tree - check if a mount tree is busy | |
977 | * @mnt: root of mount tree | |
978 | * | |
979 | * This is called to check if a tree of mounts has any | |
980 | * open files, pwds, chroots or sub mounts that are | |
981 | * busy. | |
982 | */ | |
909b0a88 | 983 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 984 | { |
909b0a88 | 985 | struct mount *mnt = real_mount(m); |
36341f64 RP |
986 | int actual_refs = 0; |
987 | int minimum_refs = 0; | |
315fc83e | 988 | struct mount *p; |
909b0a88 | 989 | BUG_ON(!m); |
1da177e4 | 990 | |
b3e19d92 | 991 | /* write lock needed for mnt_get_count */ |
962830df | 992 | br_write_lock(&vfsmount_lock); |
909b0a88 | 993 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 994 | actual_refs += mnt_get_count(p); |
1da177e4 | 995 | minimum_refs += 2; |
1da177e4 | 996 | } |
962830df | 997 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
998 | |
999 | if (actual_refs > minimum_refs) | |
e3474a8e | 1000 | return 0; |
1da177e4 | 1001 | |
e3474a8e | 1002 | return 1; |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | EXPORT_SYMBOL(may_umount_tree); | |
1006 | ||
1007 | /** | |
1008 | * may_umount - check if a mount point is busy | |
1009 | * @mnt: root of mount | |
1010 | * | |
1011 | * This is called to check if a mount point has any | |
1012 | * open files, pwds, chroots or sub mounts. If the | |
1013 | * mount has sub mounts this will return busy | |
1014 | * regardless of whether the sub mounts are busy. | |
1015 | * | |
1016 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1017 | * give false negatives. The main reason why it's here is that we need | |
1018 | * a non-destructive way to look for easily umountable filesystems. | |
1019 | */ | |
1020 | int may_umount(struct vfsmount *mnt) | |
1021 | { | |
e3474a8e | 1022 | int ret = 1; |
8ad08d8a | 1023 | down_read(&namespace_sem); |
962830df | 1024 | br_write_lock(&vfsmount_lock); |
1ab59738 | 1025 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1026 | ret = 0; |
962830df | 1027 | br_write_unlock(&vfsmount_lock); |
8ad08d8a | 1028 | up_read(&namespace_sem); |
a05964f3 | 1029 | return ret; |
1da177e4 LT |
1030 | } |
1031 | ||
1032 | EXPORT_SYMBOL(may_umount); | |
1033 | ||
b90fa9ae | 1034 | void release_mounts(struct list_head *head) |
70fbcdf4 | 1035 | { |
d5e50f74 | 1036 | struct mount *mnt; |
bf066c7d | 1037 | while (!list_empty(head)) { |
1b8e5564 AV |
1038 | mnt = list_first_entry(head, struct mount, mnt_hash); |
1039 | list_del_init(&mnt->mnt_hash); | |
676da58d | 1040 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 1041 | struct dentry *dentry; |
863d684f | 1042 | struct mount *m; |
99b7db7b | 1043 | |
962830df | 1044 | br_write_lock(&vfsmount_lock); |
a73324da | 1045 | dentry = mnt->mnt_mountpoint; |
863d684f | 1046 | m = mnt->mnt_parent; |
a73324da | 1047 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1048 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1049 | m->mnt_ghosts--; |
962830df | 1050 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1051 | dput(dentry); |
863d684f | 1052 | mntput(&m->mnt); |
70fbcdf4 | 1053 | } |
d5e50f74 | 1054 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1055 | } |
1056 | } | |
1057 | ||
99b7db7b NP |
1058 | /* |
1059 | * vfsmount lock must be held for write | |
1060 | * namespace_sem must be held for write | |
1061 | */ | |
761d5c38 | 1062 | void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) |
1da177e4 | 1063 | { |
7b8a53fd | 1064 | LIST_HEAD(tmp_list); |
315fc83e | 1065 | struct mount *p; |
1da177e4 | 1066 | |
909b0a88 | 1067 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1068 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1069 | |
a05964f3 | 1070 | if (propagate) |
7b8a53fd | 1071 | propagate_umount(&tmp_list); |
a05964f3 | 1072 | |
1b8e5564 | 1073 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1074 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1075 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1076 | __touch_mnt_namespace(p->mnt_ns); |
1077 | p->mnt_ns = NULL; | |
6b41d536 | 1078 | list_del_init(&p->mnt_child); |
676da58d | 1079 | if (mnt_has_parent(p)) { |
863d684f | 1080 | p->mnt_parent->mnt_ghosts++; |
a73324da | 1081 | dentry_reset_mounted(p->mnt_mountpoint); |
7c4b93d8 | 1082 | } |
0f0afb1d | 1083 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1084 | } |
7b8a53fd | 1085 | list_splice(&tmp_list, kill); |
1da177e4 LT |
1086 | } |
1087 | ||
692afc31 | 1088 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts); |
c35038be | 1089 | |
1ab59738 | 1090 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1091 | { |
1ab59738 | 1092 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 | 1093 | int retval; |
70fbcdf4 | 1094 | LIST_HEAD(umount_list); |
1da177e4 | 1095 | |
1ab59738 | 1096 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1097 | if (retval) |
1098 | return retval; | |
1099 | ||
1100 | /* | |
1101 | * Allow userspace to request a mountpoint be expired rather than | |
1102 | * unmounting unconditionally. Unmount only happens if: | |
1103 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1104 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1105 | */ | |
1106 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1107 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1108 | flags & (MNT_FORCE | MNT_DETACH)) |
1109 | return -EINVAL; | |
1110 | ||
b3e19d92 NP |
1111 | /* |
1112 | * probably don't strictly need the lock here if we examined | |
1113 | * all race cases, but it's a slowpath. | |
1114 | */ | |
962830df | 1115 | br_write_lock(&vfsmount_lock); |
83adc753 | 1116 | if (mnt_get_count(mnt) != 2) { |
962830df | 1117 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1118 | return -EBUSY; |
b3e19d92 | 1119 | } |
962830df | 1120 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1121 | |
863d684f | 1122 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1123 | return -EAGAIN; |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * If we may have to abort operations to get out of this | |
1128 | * mount, and they will themselves hold resources we must | |
1129 | * allow the fs to do things. In the Unix tradition of | |
1130 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1131 | * might fail to complete on the first run through as other tasks | |
1132 | * must return, and the like. Thats for the mount program to worry | |
1133 | * about for the moment. | |
1134 | */ | |
1135 | ||
42faad99 | 1136 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1137 | sb->s_op->umount_begin(sb); |
42faad99 | 1138 | } |
1da177e4 LT |
1139 | |
1140 | /* | |
1141 | * No sense to grab the lock for this test, but test itself looks | |
1142 | * somewhat bogus. Suggestions for better replacement? | |
1143 | * Ho-hum... In principle, we might treat that as umount + switch | |
1144 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1145 | * Actually it makes sense, especially if rootfs would contain a | |
1146 | * /reboot - static binary that would close all descriptors and | |
1147 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1148 | */ | |
1ab59738 | 1149 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1150 | /* |
1151 | * Special case for "unmounting" root ... | |
1152 | * we just try to remount it readonly. | |
1153 | */ | |
1154 | down_write(&sb->s_umount); | |
4aa98cf7 | 1155 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1156 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1157 | up_write(&sb->s_umount); |
1158 | return retval; | |
1159 | } | |
1160 | ||
390c6843 | 1161 | down_write(&namespace_sem); |
962830df | 1162 | br_write_lock(&vfsmount_lock); |
5addc5dd | 1163 | event++; |
1da177e4 | 1164 | |
c35038be | 1165 | if (!(flags & MNT_DETACH)) |
1ab59738 | 1166 | shrink_submounts(mnt, &umount_list); |
c35038be | 1167 | |
1da177e4 | 1168 | retval = -EBUSY; |
a05964f3 | 1169 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1170 | if (!list_empty(&mnt->mnt_list)) |
1ab59738 | 1171 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1172 | retval = 0; |
1173 | } | |
962830df | 1174 | br_write_unlock(&vfsmount_lock); |
390c6843 | 1175 | up_write(&namespace_sem); |
70fbcdf4 | 1176 | release_mounts(&umount_list); |
1da177e4 LT |
1177 | return retval; |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * Now umount can handle mount points as well as block devices. | |
1182 | * This is important for filesystems which use unnamed block devices. | |
1183 | * | |
1184 | * We now support a flag for forced unmount like the other 'big iron' | |
1185 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1186 | */ | |
1187 | ||
bdc480e3 | 1188 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1189 | { |
2d8f3038 | 1190 | struct path path; |
900148dc | 1191 | struct mount *mnt; |
1da177e4 | 1192 | int retval; |
db1f05bb | 1193 | int lookup_flags = 0; |
1da177e4 | 1194 | |
db1f05bb MS |
1195 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1196 | return -EINVAL; | |
1197 | ||
1198 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1199 | lookup_flags |= LOOKUP_FOLLOW; | |
1200 | ||
1201 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1202 | if (retval) |
1203 | goto out; | |
900148dc | 1204 | mnt = real_mount(path.mnt); |
1da177e4 | 1205 | retval = -EINVAL; |
2d8f3038 | 1206 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1207 | goto dput_and_out; |
143c8c91 | 1208 | if (!check_mnt(mnt)) |
1da177e4 LT |
1209 | goto dput_and_out; |
1210 | ||
1211 | retval = -EPERM; | |
1212 | if (!capable(CAP_SYS_ADMIN)) | |
1213 | goto dput_and_out; | |
1214 | ||
900148dc | 1215 | retval = do_umount(mnt, flags); |
1da177e4 | 1216 | dput_and_out: |
429731b1 | 1217 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1218 | dput(path.dentry); |
900148dc | 1219 | mntput_no_expire(mnt); |
1da177e4 LT |
1220 | out: |
1221 | return retval; | |
1222 | } | |
1223 | ||
1224 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1225 | ||
1226 | /* | |
b58fed8b | 1227 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1228 | */ |
bdc480e3 | 1229 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1230 | { |
b58fed8b | 1231 | return sys_umount(name, 0); |
1da177e4 LT |
1232 | } |
1233 | ||
1234 | #endif | |
1235 | ||
2d92ab3c | 1236 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1237 | { |
1238 | if (capable(CAP_SYS_ADMIN)) | |
1239 | return 0; | |
1240 | return -EPERM; | |
1241 | #ifdef notyet | |
2d92ab3c | 1242 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1243 | return -EPERM; |
2d92ab3c | 1244 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1245 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1246 | return -EPERM; |
1247 | } | |
2d92ab3c | 1248 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1249 | return -EPERM; |
1250 | return 0; | |
1251 | #endif | |
1252 | } | |
1253 | ||
87129cc0 | 1254 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1255 | int flag) |
1da177e4 | 1256 | { |
a73324da | 1257 | struct mount *res, *p, *q, *r; |
1a390689 | 1258 | struct path path; |
1da177e4 | 1259 | |
fc7be130 | 1260 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
be34d1a3 | 1261 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1262 | |
36341f64 | 1263 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1264 | if (IS_ERR(q)) |
1265 | return q; | |
1266 | ||
a73324da | 1267 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1268 | |
1269 | p = mnt; | |
6b41d536 | 1270 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1271 | struct mount *s; |
7ec02ef1 | 1272 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1273 | continue; |
1274 | ||
909b0a88 | 1275 | for (s = r; s; s = next_mnt(s, r)) { |
fc7be130 | 1276 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
9676f0c6 RP |
1277 | s = skip_mnt_tree(s); |
1278 | continue; | |
1279 | } | |
0714a533 AV |
1280 | while (p != s->mnt_parent) { |
1281 | p = p->mnt_parent; | |
1282 | q = q->mnt_parent; | |
1da177e4 | 1283 | } |
87129cc0 | 1284 | p = s; |
cb338d06 | 1285 | path.mnt = &q->mnt; |
a73324da | 1286 | path.dentry = p->mnt_mountpoint; |
87129cc0 | 1287 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1288 | if (IS_ERR(q)) |
1289 | goto out; | |
962830df | 1290 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 1291 | list_add_tail(&q->mnt_list, &res->mnt_list); |
cb338d06 | 1292 | attach_mnt(q, &path); |
962830df | 1293 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1294 | } |
1295 | } | |
1296 | return res; | |
be34d1a3 | 1297 | out: |
1da177e4 | 1298 | if (res) { |
70fbcdf4 | 1299 | LIST_HEAD(umount_list); |
962830df | 1300 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1301 | umount_tree(res, 0, &umount_list); |
962830df | 1302 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1303 | release_mounts(&umount_list); |
1da177e4 | 1304 | } |
be34d1a3 | 1305 | return q; |
1da177e4 LT |
1306 | } |
1307 | ||
be34d1a3 DH |
1308 | /* Caller should check returned pointer for errors */ |
1309 | ||
589ff870 | 1310 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1311 | { |
cb338d06 | 1312 | struct mount *tree; |
1a60a280 | 1313 | down_write(&namespace_sem); |
87129cc0 AV |
1314 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1315 | CL_COPY_ALL | CL_PRIVATE); | |
1a60a280 | 1316 | up_write(&namespace_sem); |
be34d1a3 DH |
1317 | if (IS_ERR(tree)) |
1318 | return NULL; | |
1319 | return &tree->mnt; | |
8aec0809 AV |
1320 | } |
1321 | ||
1322 | void drop_collected_mounts(struct vfsmount *mnt) | |
1323 | { | |
1324 | LIST_HEAD(umount_list); | |
1a60a280 | 1325 | down_write(&namespace_sem); |
962830df | 1326 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1327 | umount_tree(real_mount(mnt), 0, &umount_list); |
962830df | 1328 | br_write_unlock(&vfsmount_lock); |
1a60a280 | 1329 | up_write(&namespace_sem); |
8aec0809 AV |
1330 | release_mounts(&umount_list); |
1331 | } | |
1332 | ||
1f707137 AV |
1333 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1334 | struct vfsmount *root) | |
1335 | { | |
1a4eeaf2 | 1336 | struct mount *mnt; |
1f707137 AV |
1337 | int res = f(root, arg); |
1338 | if (res) | |
1339 | return res; | |
1a4eeaf2 AV |
1340 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1341 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1342 | if (res) |
1343 | return res; | |
1344 | } | |
1345 | return 0; | |
1346 | } | |
1347 | ||
4b8b21f4 | 1348 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1349 | { |
315fc83e | 1350 | struct mount *p; |
719f5d7f | 1351 | |
909b0a88 | 1352 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1353 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1354 | mnt_release_group_id(p); |
719f5d7f MS |
1355 | } |
1356 | } | |
1357 | ||
4b8b21f4 | 1358 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1359 | { |
315fc83e | 1360 | struct mount *p; |
719f5d7f | 1361 | |
909b0a88 | 1362 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1363 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1364 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1365 | if (err) { |
4b8b21f4 | 1366 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1367 | return err; |
1368 | } | |
1369 | } | |
1370 | } | |
1371 | ||
1372 | return 0; | |
1373 | } | |
1374 | ||
b90fa9ae RP |
1375 | /* |
1376 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1377 | * @nd : place the mount tree @source_mnt is attached |
1378 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1379 | * store the parent mount and mountpoint dentry. | |
1380 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1381 | * |
1382 | * NOTE: in the table below explains the semantics when a source mount | |
1383 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1384 | * --------------------------------------------------------------------------- |
1385 | * | BIND MOUNT OPERATION | | |
1386 | * |************************************************************************** | |
1387 | * | source-->| shared | private | slave | unbindable | | |
1388 | * | dest | | | | | | |
1389 | * | | | | | | | | |
1390 | * | v | | | | | | |
1391 | * |************************************************************************** | |
1392 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1393 | * | | | | | | | |
1394 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1395 | * *************************************************************************** | |
b90fa9ae RP |
1396 | * A bind operation clones the source mount and mounts the clone on the |
1397 | * destination mount. | |
1398 | * | |
1399 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1400 | * tree of the destination mount and the cloned mount is added to | |
1401 | * the peer group of the source mount. | |
1402 | * (+) the cloned mount is created under the destination mount and is marked | |
1403 | * as shared. The cloned mount is added to the peer group of the source | |
1404 | * mount. | |
5afe0022 RP |
1405 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1406 | * of the destination mount and the cloned mount is made slave | |
1407 | * of the same master as that of the source mount. The cloned mount | |
1408 | * is marked as 'shared and slave'. | |
1409 | * (*) the cloned mount is made a slave of the same master as that of the | |
1410 | * source mount. | |
1411 | * | |
9676f0c6 RP |
1412 | * --------------------------------------------------------------------------- |
1413 | * | MOVE MOUNT OPERATION | | |
1414 | * |************************************************************************** | |
1415 | * | source-->| shared | private | slave | unbindable | | |
1416 | * | dest | | | | | | |
1417 | * | | | | | | | | |
1418 | * | v | | | | | | |
1419 | * |************************************************************************** | |
1420 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1421 | * | | | | | | | |
1422 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1423 | * *************************************************************************** | |
5afe0022 RP |
1424 | * |
1425 | * (+) the mount is moved to the destination. And is then propagated to | |
1426 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1427 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1428 | * (+++) the mount is moved to the destination and is then propagated to |
1429 | * all the mounts belonging to the destination mount's propagation tree. | |
1430 | * the mount is marked as 'shared and slave'. | |
1431 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1432 | * |
1433 | * if the source mount is a tree, the operations explained above is | |
1434 | * applied to each mount in the tree. | |
1435 | * Must be called without spinlocks held, since this function can sleep | |
1436 | * in allocations. | |
1437 | */ | |
0fb54e50 | 1438 | static int attach_recursive_mnt(struct mount *source_mnt, |
1a390689 | 1439 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1440 | { |
1441 | LIST_HEAD(tree_list); | |
a8d56d8e | 1442 | struct mount *dest_mnt = real_mount(path->mnt); |
1a390689 | 1443 | struct dentry *dest_dentry = path->dentry; |
315fc83e | 1444 | struct mount *child, *p; |
719f5d7f | 1445 | int err; |
b90fa9ae | 1446 | |
fc7be130 | 1447 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1448 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1449 | if (err) |
1450 | goto out; | |
1451 | } | |
a8d56d8e | 1452 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); |
719f5d7f MS |
1453 | if (err) |
1454 | goto out_cleanup_ids; | |
b90fa9ae | 1455 | |
962830df | 1456 | br_write_lock(&vfsmount_lock); |
df1a1ad2 | 1457 | |
fc7be130 | 1458 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1459 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1460 | set_mnt_shared(p); |
b90fa9ae | 1461 | } |
1a390689 | 1462 | if (parent_path) { |
0fb54e50 AV |
1463 | detach_mnt(source_mnt, parent_path); |
1464 | attach_mnt(source_mnt, path); | |
143c8c91 | 1465 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1466 | } else { |
14cf1fa8 | 1467 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); |
0fb54e50 | 1468 | commit_tree(source_mnt); |
21444403 | 1469 | } |
b90fa9ae | 1470 | |
1b8e5564 AV |
1471 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1472 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1473 | commit_tree(child); |
b90fa9ae | 1474 | } |
962830df | 1475 | br_write_unlock(&vfsmount_lock); |
99b7db7b | 1476 | |
b90fa9ae | 1477 | return 0; |
719f5d7f MS |
1478 | |
1479 | out_cleanup_ids: | |
fc7be130 | 1480 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1481 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1482 | out: |
1483 | return err; | |
b90fa9ae RP |
1484 | } |
1485 | ||
b12cea91 AV |
1486 | static int lock_mount(struct path *path) |
1487 | { | |
1488 | struct vfsmount *mnt; | |
1489 | retry: | |
1490 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1491 | if (unlikely(cant_mount(path->dentry))) { | |
1492 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1493 | return -ENOENT; | |
1494 | } | |
1495 | down_write(&namespace_sem); | |
1496 | mnt = lookup_mnt(path); | |
1497 | if (likely(!mnt)) | |
1498 | return 0; | |
1499 | up_write(&namespace_sem); | |
1500 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1501 | path_put(path); | |
1502 | path->mnt = mnt; | |
1503 | path->dentry = dget(mnt->mnt_root); | |
1504 | goto retry; | |
1505 | } | |
1506 | ||
1507 | static void unlock_mount(struct path *path) | |
1508 | { | |
1509 | up_write(&namespace_sem); | |
1510 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1511 | } | |
1512 | ||
95bc5f25 | 1513 | static int graft_tree(struct mount *mnt, struct path *path) |
1da177e4 | 1514 | { |
95bc5f25 | 1515 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1516 | return -EINVAL; |
1517 | ||
8c3ee42e | 1518 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
95bc5f25 | 1519 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1520 | return -ENOTDIR; |
1521 | ||
b12cea91 AV |
1522 | if (d_unlinked(path->dentry)) |
1523 | return -ENOENT; | |
1da177e4 | 1524 | |
95bc5f25 | 1525 | return attach_recursive_mnt(mnt, path, NULL); |
1da177e4 LT |
1526 | } |
1527 | ||
7a2e8a8f VA |
1528 | /* |
1529 | * Sanity check the flags to change_mnt_propagation. | |
1530 | */ | |
1531 | ||
1532 | static int flags_to_propagation_type(int flags) | |
1533 | { | |
7c6e984d | 1534 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1535 | |
1536 | /* Fail if any non-propagation flags are set */ | |
1537 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1538 | return 0; | |
1539 | /* Only one propagation flag should be set */ | |
1540 | if (!is_power_of_2(type)) | |
1541 | return 0; | |
1542 | return type; | |
1543 | } | |
1544 | ||
07b20889 RP |
1545 | /* |
1546 | * recursively change the type of the mountpoint. | |
1547 | */ | |
0a0d8a46 | 1548 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1549 | { |
315fc83e | 1550 | struct mount *m; |
4b8b21f4 | 1551 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1552 | int recurse = flag & MS_REC; |
7a2e8a8f | 1553 | int type; |
719f5d7f | 1554 | int err = 0; |
07b20889 | 1555 | |
ee6f9582 MS |
1556 | if (!capable(CAP_SYS_ADMIN)) |
1557 | return -EPERM; | |
1558 | ||
2d92ab3c | 1559 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1560 | return -EINVAL; |
1561 | ||
7a2e8a8f VA |
1562 | type = flags_to_propagation_type(flag); |
1563 | if (!type) | |
1564 | return -EINVAL; | |
1565 | ||
07b20889 | 1566 | down_write(&namespace_sem); |
719f5d7f MS |
1567 | if (type == MS_SHARED) { |
1568 | err = invent_group_ids(mnt, recurse); | |
1569 | if (err) | |
1570 | goto out_unlock; | |
1571 | } | |
1572 | ||
962830df | 1573 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1574 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1575 | change_mnt_propagation(m, type); |
962830df | 1576 | br_write_unlock(&vfsmount_lock); |
719f5d7f MS |
1577 | |
1578 | out_unlock: | |
07b20889 | 1579 | up_write(&namespace_sem); |
719f5d7f | 1580 | return err; |
07b20889 RP |
1581 | } |
1582 | ||
1da177e4 LT |
1583 | /* |
1584 | * do loopback mount. | |
1585 | */ | |
0a0d8a46 | 1586 | static int do_loopback(struct path *path, char *old_name, |
2dafe1c4 | 1587 | int recurse) |
1da177e4 | 1588 | { |
b12cea91 | 1589 | LIST_HEAD(umount_list); |
2d92ab3c | 1590 | struct path old_path; |
87129cc0 | 1591 | struct mount *mnt = NULL, *old; |
2d92ab3c | 1592 | int err = mount_is_safe(path); |
1da177e4 LT |
1593 | if (err) |
1594 | return err; | |
1595 | if (!old_name || !*old_name) | |
1596 | return -EINVAL; | |
815d405c | 1597 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1598 | if (err) |
1599 | return err; | |
1600 | ||
b12cea91 AV |
1601 | err = lock_mount(path); |
1602 | if (err) | |
1603 | goto out; | |
1604 | ||
87129cc0 AV |
1605 | old = real_mount(old_path.mnt); |
1606 | ||
1da177e4 | 1607 | err = -EINVAL; |
fc7be130 | 1608 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1609 | goto out2; |
9676f0c6 | 1610 | |
143c8c91 | 1611 | if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old)) |
b12cea91 | 1612 | goto out2; |
1da177e4 | 1613 | |
ccd48bc7 | 1614 | if (recurse) |
87129cc0 | 1615 | mnt = copy_tree(old, old_path.dentry, 0); |
ccd48bc7 | 1616 | else |
87129cc0 | 1617 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1618 | |
be34d1a3 DH |
1619 | if (IS_ERR(mnt)) { |
1620 | err = PTR_ERR(mnt); | |
1621 | goto out; | |
1622 | } | |
ccd48bc7 | 1623 | |
95bc5f25 | 1624 | err = graft_tree(mnt, path); |
ccd48bc7 | 1625 | if (err) { |
962830df | 1626 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1627 | umount_tree(mnt, 0, &umount_list); |
962830df | 1628 | br_write_unlock(&vfsmount_lock); |
5b83d2c5 | 1629 | } |
b12cea91 AV |
1630 | out2: |
1631 | unlock_mount(path); | |
1632 | release_mounts(&umount_list); | |
ccd48bc7 | 1633 | out: |
2d92ab3c | 1634 | path_put(&old_path); |
1da177e4 LT |
1635 | return err; |
1636 | } | |
1637 | ||
2e4b7fcd DH |
1638 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1639 | { | |
1640 | int error = 0; | |
1641 | int readonly_request = 0; | |
1642 | ||
1643 | if (ms_flags & MS_RDONLY) | |
1644 | readonly_request = 1; | |
1645 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1646 | return 0; | |
1647 | ||
1648 | if (readonly_request) | |
83adc753 | 1649 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1650 | else |
83adc753 | 1651 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1652 | return error; |
1653 | } | |
1654 | ||
1da177e4 LT |
1655 | /* |
1656 | * change filesystem flags. dir should be a physical root of filesystem. | |
1657 | * If you've mounted a non-root directory somewhere and want to do remount | |
1658 | * on it - tough luck. | |
1659 | */ | |
0a0d8a46 | 1660 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1661 | void *data) |
1662 | { | |
1663 | int err; | |
2d92ab3c | 1664 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1665 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 LT |
1666 | |
1667 | if (!capable(CAP_SYS_ADMIN)) | |
1668 | return -EPERM; | |
1669 | ||
143c8c91 | 1670 | if (!check_mnt(mnt)) |
1da177e4 LT |
1671 | return -EINVAL; |
1672 | ||
2d92ab3c | 1673 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1674 | return -EINVAL; |
1675 | ||
ff36fe2c EP |
1676 | err = security_sb_remount(sb, data); |
1677 | if (err) | |
1678 | return err; | |
1679 | ||
1da177e4 | 1680 | down_write(&sb->s_umount); |
2e4b7fcd | 1681 | if (flags & MS_BIND) |
2d92ab3c | 1682 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1683 | else |
2e4b7fcd | 1684 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1685 | if (!err) { |
962830df | 1686 | br_write_lock(&vfsmount_lock); |
143c8c91 AV |
1687 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1688 | mnt->mnt.mnt_flags = mnt_flags; | |
962830df | 1689 | br_write_unlock(&vfsmount_lock); |
7b43a79f | 1690 | } |
1da177e4 | 1691 | up_write(&sb->s_umount); |
0e55a7cc | 1692 | if (!err) { |
962830df | 1693 | br_write_lock(&vfsmount_lock); |
143c8c91 | 1694 | touch_mnt_namespace(mnt->mnt_ns); |
962830df | 1695 | br_write_unlock(&vfsmount_lock); |
0e55a7cc | 1696 | } |
1da177e4 LT |
1697 | return err; |
1698 | } | |
1699 | ||
cbbe362c | 1700 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1701 | { |
315fc83e | 1702 | struct mount *p; |
909b0a88 | 1703 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1704 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1705 | return 1; |
1706 | } | |
1707 | return 0; | |
1708 | } | |
1709 | ||
0a0d8a46 | 1710 | static int do_move_mount(struct path *path, char *old_name) |
1da177e4 | 1711 | { |
2d92ab3c | 1712 | struct path old_path, parent_path; |
676da58d | 1713 | struct mount *p; |
0fb54e50 | 1714 | struct mount *old; |
1da177e4 LT |
1715 | int err = 0; |
1716 | if (!capable(CAP_SYS_ADMIN)) | |
1717 | return -EPERM; | |
1718 | if (!old_name || !*old_name) | |
1719 | return -EINVAL; | |
2d92ab3c | 1720 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1721 | if (err) |
1722 | return err; | |
1723 | ||
b12cea91 | 1724 | err = lock_mount(path); |
cc53ce53 DH |
1725 | if (err < 0) |
1726 | goto out; | |
1727 | ||
143c8c91 | 1728 | old = real_mount(old_path.mnt); |
fc7be130 | 1729 | p = real_mount(path->mnt); |
143c8c91 | 1730 | |
1da177e4 | 1731 | err = -EINVAL; |
fc7be130 | 1732 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1733 | goto out1; |
1734 | ||
f3da392e | 1735 | if (d_unlinked(path->dentry)) |
21444403 | 1736 | goto out1; |
1da177e4 LT |
1737 | |
1738 | err = -EINVAL; | |
2d92ab3c | 1739 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1740 | goto out1; |
1da177e4 | 1741 | |
676da58d | 1742 | if (!mnt_has_parent(old)) |
21444403 | 1743 | goto out1; |
1da177e4 | 1744 | |
2d92ab3c AV |
1745 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1746 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1747 | goto out1; |
1748 | /* | |
1749 | * Don't move a mount residing in a shared parent. | |
1750 | */ | |
fc7be130 | 1751 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1752 | goto out1; |
9676f0c6 RP |
1753 | /* |
1754 | * Don't move a mount tree containing unbindable mounts to a destination | |
1755 | * mount which is shared. | |
1756 | */ | |
fc7be130 | 1757 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1758 | goto out1; |
1da177e4 | 1759 | err = -ELOOP; |
fc7be130 | 1760 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1761 | if (p == old) |
21444403 | 1762 | goto out1; |
1da177e4 | 1763 | |
0fb54e50 | 1764 | err = attach_recursive_mnt(old, path, &parent_path); |
4ac91378 | 1765 | if (err) |
21444403 | 1766 | goto out1; |
1da177e4 LT |
1767 | |
1768 | /* if the mount is moved, it should no longer be expire | |
1769 | * automatically */ | |
6776db3d | 1770 | list_del_init(&old->mnt_expire); |
1da177e4 | 1771 | out1: |
b12cea91 | 1772 | unlock_mount(path); |
1da177e4 | 1773 | out: |
1da177e4 | 1774 | if (!err) |
1a390689 | 1775 | path_put(&parent_path); |
2d92ab3c | 1776 | path_put(&old_path); |
1da177e4 LT |
1777 | return err; |
1778 | } | |
1779 | ||
9d412a43 AV |
1780 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1781 | { | |
1782 | int err; | |
1783 | const char *subtype = strchr(fstype, '.'); | |
1784 | if (subtype) { | |
1785 | subtype++; | |
1786 | err = -EINVAL; | |
1787 | if (!subtype[0]) | |
1788 | goto err; | |
1789 | } else | |
1790 | subtype = ""; | |
1791 | ||
1792 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1793 | err = -ENOMEM; | |
1794 | if (!mnt->mnt_sb->s_subtype) | |
1795 | goto err; | |
1796 | return mnt; | |
1797 | ||
1798 | err: | |
1799 | mntput(mnt); | |
1800 | return ERR_PTR(err); | |
1801 | } | |
1802 | ||
79e801a9 | 1803 | static struct vfsmount * |
9d412a43 AV |
1804 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) |
1805 | { | |
1806 | struct file_system_type *type = get_fs_type(fstype); | |
1807 | struct vfsmount *mnt; | |
1808 | if (!type) | |
1809 | return ERR_PTR(-ENODEV); | |
1810 | mnt = vfs_kern_mount(type, flags, name, data); | |
1811 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1812 | !mnt->mnt_sb->s_subtype) | |
1813 | mnt = fs_set_subtype(mnt, fstype); | |
1814 | put_filesystem(type); | |
1815 | return mnt; | |
1816 | } | |
9d412a43 AV |
1817 | |
1818 | /* | |
1819 | * add a mount into a namespace's mount tree | |
1820 | */ | |
95bc5f25 | 1821 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 AV |
1822 | { |
1823 | int err; | |
1824 | ||
1825 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1826 | ||
b12cea91 AV |
1827 | err = lock_mount(path); |
1828 | if (err) | |
1829 | return err; | |
9d412a43 AV |
1830 | |
1831 | err = -EINVAL; | |
143c8c91 | 1832 | if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt))) |
9d412a43 AV |
1833 | goto unlock; |
1834 | ||
1835 | /* Refuse the same filesystem on the same mount point */ | |
1836 | err = -EBUSY; | |
95bc5f25 | 1837 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1838 | path->mnt->mnt_root == path->dentry) |
1839 | goto unlock; | |
1840 | ||
1841 | err = -EINVAL; | |
95bc5f25 | 1842 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
1843 | goto unlock; |
1844 | ||
95bc5f25 | 1845 | newmnt->mnt.mnt_flags = mnt_flags; |
9d412a43 AV |
1846 | err = graft_tree(newmnt, path); |
1847 | ||
1848 | unlock: | |
b12cea91 | 1849 | unlock_mount(path); |
9d412a43 AV |
1850 | return err; |
1851 | } | |
b1e75df4 | 1852 | |
1da177e4 LT |
1853 | /* |
1854 | * create a new mount for userspace and request it to be added into the | |
1855 | * namespace's tree | |
1856 | */ | |
0a0d8a46 | 1857 | static int do_new_mount(struct path *path, char *type, int flags, |
1da177e4 LT |
1858 | int mnt_flags, char *name, void *data) |
1859 | { | |
1860 | struct vfsmount *mnt; | |
15f9a3f3 | 1861 | int err; |
1da177e4 | 1862 | |
eca6f534 | 1863 | if (!type) |
1da177e4 LT |
1864 | return -EINVAL; |
1865 | ||
1866 | /* we need capabilities... */ | |
1867 | if (!capable(CAP_SYS_ADMIN)) | |
1868 | return -EPERM; | |
1869 | ||
1870 | mnt = do_kern_mount(type, flags, name, data); | |
1871 | if (IS_ERR(mnt)) | |
1872 | return PTR_ERR(mnt); | |
1873 | ||
95bc5f25 | 1874 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
1875 | if (err) |
1876 | mntput(mnt); | |
1877 | return err; | |
1da177e4 LT |
1878 | } |
1879 | ||
19a167af AV |
1880 | int finish_automount(struct vfsmount *m, struct path *path) |
1881 | { | |
6776db3d | 1882 | struct mount *mnt = real_mount(m); |
19a167af AV |
1883 | int err; |
1884 | /* The new mount record should have at least 2 refs to prevent it being | |
1885 | * expired before we get a chance to add it | |
1886 | */ | |
6776db3d | 1887 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
1888 | |
1889 | if (m->mnt_sb == path->mnt->mnt_sb && | |
1890 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
1891 | err = -ELOOP; |
1892 | goto fail; | |
19a167af AV |
1893 | } |
1894 | ||
95bc5f25 | 1895 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
1896 | if (!err) |
1897 | return 0; | |
1898 | fail: | |
1899 | /* remove m from any expiration list it may be on */ | |
6776db3d | 1900 | if (!list_empty(&mnt->mnt_expire)) { |
b1e75df4 | 1901 | down_write(&namespace_sem); |
962830df | 1902 | br_write_lock(&vfsmount_lock); |
6776db3d | 1903 | list_del_init(&mnt->mnt_expire); |
962830df | 1904 | br_write_unlock(&vfsmount_lock); |
b1e75df4 | 1905 | up_write(&namespace_sem); |
19a167af | 1906 | } |
b1e75df4 AV |
1907 | mntput(m); |
1908 | mntput(m); | |
19a167af AV |
1909 | return err; |
1910 | } | |
1911 | ||
ea5b778a DH |
1912 | /** |
1913 | * mnt_set_expiry - Put a mount on an expiration list | |
1914 | * @mnt: The mount to list. | |
1915 | * @expiry_list: The list to add the mount to. | |
1916 | */ | |
1917 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
1918 | { | |
1919 | down_write(&namespace_sem); | |
962830df | 1920 | br_write_lock(&vfsmount_lock); |
ea5b778a | 1921 | |
6776db3d | 1922 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 1923 | |
962830df | 1924 | br_write_unlock(&vfsmount_lock); |
ea5b778a DH |
1925 | up_write(&namespace_sem); |
1926 | } | |
1927 | EXPORT_SYMBOL(mnt_set_expiry); | |
1928 | ||
1da177e4 LT |
1929 | /* |
1930 | * process a list of expirable mountpoints with the intent of discarding any | |
1931 | * mountpoints that aren't in use and haven't been touched since last we came | |
1932 | * here | |
1933 | */ | |
1934 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1935 | { | |
761d5c38 | 1936 | struct mount *mnt, *next; |
1da177e4 | 1937 | LIST_HEAD(graveyard); |
bcc5c7d2 | 1938 | LIST_HEAD(umounts); |
1da177e4 LT |
1939 | |
1940 | if (list_empty(mounts)) | |
1941 | return; | |
1942 | ||
bcc5c7d2 | 1943 | down_write(&namespace_sem); |
962830df | 1944 | br_write_lock(&vfsmount_lock); |
1da177e4 LT |
1945 | |
1946 | /* extract from the expiration list every vfsmount that matches the | |
1947 | * following criteria: | |
1948 | * - only referenced by its parent vfsmount | |
1949 | * - still marked for expiry (marked on the last call here; marks are | |
1950 | * cleared by mntput()) | |
1951 | */ | |
6776db3d | 1952 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 1953 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 1954 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 1955 | continue; |
6776db3d | 1956 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 1957 | } |
bcc5c7d2 | 1958 | while (!list_empty(&graveyard)) { |
6776db3d | 1959 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 1960 | touch_mnt_namespace(mnt->mnt_ns); |
bcc5c7d2 AV |
1961 | umount_tree(mnt, 1, &umounts); |
1962 | } | |
962830df | 1963 | br_write_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
1964 | up_write(&namespace_sem); |
1965 | ||
1966 | release_mounts(&umounts); | |
5528f911 TM |
1967 | } |
1968 | ||
1969 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1970 | ||
1971 | /* | |
1972 | * Ripoff of 'select_parent()' | |
1973 | * | |
1974 | * search the list of submounts for a given mountpoint, and move any | |
1975 | * shrinkable submounts to the 'graveyard' list. | |
1976 | */ | |
692afc31 | 1977 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 1978 | { |
692afc31 | 1979 | struct mount *this_parent = parent; |
5528f911 TM |
1980 | struct list_head *next; |
1981 | int found = 0; | |
1982 | ||
1983 | repeat: | |
6b41d536 | 1984 | next = this_parent->mnt_mounts.next; |
5528f911 | 1985 | resume: |
6b41d536 | 1986 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 1987 | struct list_head *tmp = next; |
6b41d536 | 1988 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
1989 | |
1990 | next = tmp->next; | |
692afc31 | 1991 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 1992 | continue; |
5528f911 TM |
1993 | /* |
1994 | * Descend a level if the d_mounts list is non-empty. | |
1995 | */ | |
6b41d536 | 1996 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
1997 | this_parent = mnt; |
1998 | goto repeat; | |
1999 | } | |
1da177e4 | 2000 | |
1ab59738 | 2001 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2002 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2003 | found++; |
2004 | } | |
1da177e4 | 2005 | } |
5528f911 TM |
2006 | /* |
2007 | * All done at this level ... ascend and resume the search | |
2008 | */ | |
2009 | if (this_parent != parent) { | |
6b41d536 | 2010 | next = this_parent->mnt_child.next; |
0714a533 | 2011 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2012 | goto resume; |
2013 | } | |
2014 | return found; | |
2015 | } | |
2016 | ||
2017 | /* | |
2018 | * process a list of expirable mountpoints with the intent of discarding any | |
2019 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
2020 | * |
2021 | * vfsmount_lock must be held for write | |
5528f911 | 2022 | */ |
692afc31 | 2023 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts) |
5528f911 TM |
2024 | { |
2025 | LIST_HEAD(graveyard); | |
761d5c38 | 2026 | struct mount *m; |
5528f911 | 2027 | |
5528f911 | 2028 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2029 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2030 | while (!list_empty(&graveyard)) { |
761d5c38 | 2031 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2032 | mnt_expire); |
143c8c91 | 2033 | touch_mnt_namespace(m->mnt_ns); |
afef80b3 | 2034 | umount_tree(m, 1, umounts); |
bcc5c7d2 AV |
2035 | } |
2036 | } | |
1da177e4 LT |
2037 | } |
2038 | ||
1da177e4 LT |
2039 | /* |
2040 | * Some copy_from_user() implementations do not return the exact number of | |
2041 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2042 | * Note that this function differs from copy_from_user() in that it will oops | |
2043 | * on bad values of `to', rather than returning a short copy. | |
2044 | */ | |
b58fed8b RP |
2045 | static long exact_copy_from_user(void *to, const void __user * from, |
2046 | unsigned long n) | |
1da177e4 LT |
2047 | { |
2048 | char *t = to; | |
2049 | const char __user *f = from; | |
2050 | char c; | |
2051 | ||
2052 | if (!access_ok(VERIFY_READ, from, n)) | |
2053 | return n; | |
2054 | ||
2055 | while (n) { | |
2056 | if (__get_user(c, f)) { | |
2057 | memset(t, 0, n); | |
2058 | break; | |
2059 | } | |
2060 | *t++ = c; | |
2061 | f++; | |
2062 | n--; | |
2063 | } | |
2064 | return n; | |
2065 | } | |
2066 | ||
b58fed8b | 2067 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2068 | { |
2069 | int i; | |
2070 | unsigned long page; | |
2071 | unsigned long size; | |
b58fed8b | 2072 | |
1da177e4 LT |
2073 | *where = 0; |
2074 | if (!data) | |
2075 | return 0; | |
2076 | ||
2077 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2078 | return -ENOMEM; | |
2079 | ||
2080 | /* We only care that *some* data at the address the user | |
2081 | * gave us is valid. Just in case, we'll zero | |
2082 | * the remainder of the page. | |
2083 | */ | |
2084 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2085 | size = TASK_SIZE - (unsigned long)data; | |
2086 | if (size > PAGE_SIZE) | |
2087 | size = PAGE_SIZE; | |
2088 | ||
2089 | i = size - exact_copy_from_user((void *)page, data, size); | |
2090 | if (!i) { | |
b58fed8b | 2091 | free_page(page); |
1da177e4 LT |
2092 | return -EFAULT; |
2093 | } | |
2094 | if (i != PAGE_SIZE) | |
2095 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2096 | *where = page; | |
2097 | return 0; | |
2098 | } | |
2099 | ||
eca6f534 VN |
2100 | int copy_mount_string(const void __user *data, char **where) |
2101 | { | |
2102 | char *tmp; | |
2103 | ||
2104 | if (!data) { | |
2105 | *where = NULL; | |
2106 | return 0; | |
2107 | } | |
2108 | ||
2109 | tmp = strndup_user(data, PAGE_SIZE); | |
2110 | if (IS_ERR(tmp)) | |
2111 | return PTR_ERR(tmp); | |
2112 | ||
2113 | *where = tmp; | |
2114 | return 0; | |
2115 | } | |
2116 | ||
1da177e4 LT |
2117 | /* |
2118 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2119 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2120 | * | |
2121 | * data is a (void *) that can point to any structure up to | |
2122 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2123 | * information (or be NULL). | |
2124 | * | |
2125 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2126 | * When the flags word was introduced its top half was required | |
2127 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2128 | * Therefore, if this magic number is present, it carries no information | |
2129 | * and must be discarded. | |
2130 | */ | |
b58fed8b | 2131 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
2132 | unsigned long flags, void *data_page) |
2133 | { | |
2d92ab3c | 2134 | struct path path; |
1da177e4 LT |
2135 | int retval = 0; |
2136 | int mnt_flags = 0; | |
2137 | ||
2138 | /* Discard magic */ | |
2139 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2140 | flags &= ~MS_MGC_MSK; | |
2141 | ||
2142 | /* Basic sanity checks */ | |
2143 | ||
2144 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2145 | return -EINVAL; | |
1da177e4 LT |
2146 | |
2147 | if (data_page) | |
2148 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2149 | ||
a27ab9f2 TH |
2150 | /* ... and get the mountpoint */ |
2151 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2152 | if (retval) | |
2153 | return retval; | |
2154 | ||
2155 | retval = security_sb_mount(dev_name, &path, | |
2156 | type_page, flags, data_page); | |
2157 | if (retval) | |
2158 | goto dput_out; | |
2159 | ||
613cbe3d AK |
2160 | /* Default to relatime unless overriden */ |
2161 | if (!(flags & MS_NOATIME)) | |
2162 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2163 | |
1da177e4 LT |
2164 | /* Separate the per-mountpoint flags */ |
2165 | if (flags & MS_NOSUID) | |
2166 | mnt_flags |= MNT_NOSUID; | |
2167 | if (flags & MS_NODEV) | |
2168 | mnt_flags |= MNT_NODEV; | |
2169 | if (flags & MS_NOEXEC) | |
2170 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2171 | if (flags & MS_NOATIME) |
2172 | mnt_flags |= MNT_NOATIME; | |
2173 | if (flags & MS_NODIRATIME) | |
2174 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2175 | if (flags & MS_STRICTATIME) |
2176 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2177 | if (flags & MS_RDONLY) |
2178 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2179 | |
7a4dec53 | 2180 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2181 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2182 | MS_STRICTATIME); | |
1da177e4 | 2183 | |
1da177e4 | 2184 | if (flags & MS_REMOUNT) |
2d92ab3c | 2185 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2186 | data_page); |
2187 | else if (flags & MS_BIND) | |
2d92ab3c | 2188 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2189 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2190 | retval = do_change_type(&path, flags); |
1da177e4 | 2191 | else if (flags & MS_MOVE) |
2d92ab3c | 2192 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2193 | else |
2d92ab3c | 2194 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2195 | dev_name, data_page); |
2196 | dput_out: | |
2d92ab3c | 2197 | path_put(&path); |
1da177e4 LT |
2198 | return retval; |
2199 | } | |
2200 | ||
cf8d2c11 TM |
2201 | static struct mnt_namespace *alloc_mnt_ns(void) |
2202 | { | |
2203 | struct mnt_namespace *new_ns; | |
2204 | ||
2205 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2206 | if (!new_ns) | |
2207 | return ERR_PTR(-ENOMEM); | |
2208 | atomic_set(&new_ns->count, 1); | |
2209 | new_ns->root = NULL; | |
2210 | INIT_LIST_HEAD(&new_ns->list); | |
2211 | init_waitqueue_head(&new_ns->poll); | |
2212 | new_ns->event = 0; | |
2213 | return new_ns; | |
2214 | } | |
2215 | ||
741a2951 JD |
2216 | /* |
2217 | * Allocate a new namespace structure and populate it with contents | |
2218 | * copied from the namespace of the passed in task structure. | |
2219 | */ | |
e3222c4e | 2220 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 2221 | struct fs_struct *fs) |
1da177e4 | 2222 | { |
6b3286ed | 2223 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2224 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2225 | struct mount *p, *q; |
be08d6d2 | 2226 | struct mount *old = mnt_ns->root; |
cb338d06 | 2227 | struct mount *new; |
1da177e4 | 2228 | |
cf8d2c11 TM |
2229 | new_ns = alloc_mnt_ns(); |
2230 | if (IS_ERR(new_ns)) | |
2231 | return new_ns; | |
1da177e4 | 2232 | |
390c6843 | 2233 | down_write(&namespace_sem); |
1da177e4 | 2234 | /* First pass: copy the tree topology */ |
909b0a88 | 2235 | new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE); |
be34d1a3 | 2236 | if (IS_ERR(new)) { |
390c6843 | 2237 | up_write(&namespace_sem); |
1da177e4 | 2238 | kfree(new_ns); |
be34d1a3 | 2239 | return ERR_CAST(new); |
1da177e4 | 2240 | } |
be08d6d2 | 2241 | new_ns->root = new; |
962830df | 2242 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 2243 | list_add_tail(&new_ns->list, &new->mnt_list); |
962830df | 2244 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
2245 | |
2246 | /* | |
2247 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2248 | * as belonging to new namespace. We have already acquired a private | |
2249 | * fs_struct, so tsk->fs->lock is not needed. | |
2250 | */ | |
909b0a88 | 2251 | p = old; |
cb338d06 | 2252 | q = new; |
1da177e4 | 2253 | while (p) { |
143c8c91 | 2254 | q->mnt_ns = new_ns; |
1da177e4 | 2255 | if (fs) { |
315fc83e AV |
2256 | if (&p->mnt == fs->root.mnt) { |
2257 | fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2258 | rootmnt = &p->mnt; |
1da177e4 | 2259 | } |
315fc83e AV |
2260 | if (&p->mnt == fs->pwd.mnt) { |
2261 | fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2262 | pwdmnt = &p->mnt; |
1da177e4 | 2263 | } |
1da177e4 | 2264 | } |
909b0a88 AV |
2265 | p = next_mnt(p, old); |
2266 | q = next_mnt(q, new); | |
1da177e4 | 2267 | } |
390c6843 | 2268 | up_write(&namespace_sem); |
1da177e4 | 2269 | |
1da177e4 | 2270 | if (rootmnt) |
f03c6599 | 2271 | mntput(rootmnt); |
1da177e4 | 2272 | if (pwdmnt) |
f03c6599 | 2273 | mntput(pwdmnt); |
1da177e4 | 2274 | |
741a2951 JD |
2275 | return new_ns; |
2276 | } | |
2277 | ||
213dd266 | 2278 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2279 | struct fs_struct *new_fs) |
741a2951 | 2280 | { |
6b3286ed | 2281 | struct mnt_namespace *new_ns; |
741a2951 | 2282 | |
e3222c4e | 2283 | BUG_ON(!ns); |
6b3286ed | 2284 | get_mnt_ns(ns); |
741a2951 JD |
2285 | |
2286 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2287 | return ns; |
741a2951 | 2288 | |
e3222c4e | 2289 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2290 | |
6b3286ed | 2291 | put_mnt_ns(ns); |
e3222c4e | 2292 | return new_ns; |
1da177e4 LT |
2293 | } |
2294 | ||
cf8d2c11 TM |
2295 | /** |
2296 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2297 | * @mnt: pointer to the new root filesystem mountpoint | |
2298 | */ | |
1a4eeaf2 | 2299 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2300 | { |
1a4eeaf2 | 2301 | struct mnt_namespace *new_ns = alloc_mnt_ns(); |
cf8d2c11 | 2302 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2303 | struct mount *mnt = real_mount(m); |
2304 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2305 | new_ns->root = mnt; |
1a4eeaf2 | 2306 | list_add(&new_ns->list, &mnt->mnt_list); |
c1334495 | 2307 | } else { |
1a4eeaf2 | 2308 | mntput(m); |
cf8d2c11 TM |
2309 | } |
2310 | return new_ns; | |
2311 | } | |
cf8d2c11 | 2312 | |
ea441d11 AV |
2313 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2314 | { | |
2315 | struct mnt_namespace *ns; | |
d31da0f0 | 2316 | struct super_block *s; |
ea441d11 AV |
2317 | struct path path; |
2318 | int err; | |
2319 | ||
2320 | ns = create_mnt_ns(mnt); | |
2321 | if (IS_ERR(ns)) | |
2322 | return ERR_CAST(ns); | |
2323 | ||
2324 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2325 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2326 | ||
2327 | put_mnt_ns(ns); | |
2328 | ||
2329 | if (err) | |
2330 | return ERR_PTR(err); | |
2331 | ||
2332 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2333 | s = path.mnt->mnt_sb; |
2334 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2335 | mntput(path.mnt); |
2336 | /* lock the sucker */ | |
d31da0f0 | 2337 | down_write(&s->s_umount); |
ea441d11 AV |
2338 | /* ... and return the root of (sub)tree on it */ |
2339 | return path.dentry; | |
2340 | } | |
2341 | EXPORT_SYMBOL(mount_subtree); | |
2342 | ||
bdc480e3 HC |
2343 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2344 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2345 | { |
eca6f534 VN |
2346 | int ret; |
2347 | char *kernel_type; | |
2348 | char *kernel_dir; | |
2349 | char *kernel_dev; | |
1da177e4 | 2350 | unsigned long data_page; |
1da177e4 | 2351 | |
eca6f534 VN |
2352 | ret = copy_mount_string(type, &kernel_type); |
2353 | if (ret < 0) | |
2354 | goto out_type; | |
1da177e4 | 2355 | |
eca6f534 VN |
2356 | kernel_dir = getname(dir_name); |
2357 | if (IS_ERR(kernel_dir)) { | |
2358 | ret = PTR_ERR(kernel_dir); | |
2359 | goto out_dir; | |
2360 | } | |
1da177e4 | 2361 | |
eca6f534 VN |
2362 | ret = copy_mount_string(dev_name, &kernel_dev); |
2363 | if (ret < 0) | |
2364 | goto out_dev; | |
1da177e4 | 2365 | |
eca6f534 VN |
2366 | ret = copy_mount_options(data, &data_page); |
2367 | if (ret < 0) | |
2368 | goto out_data; | |
1da177e4 | 2369 | |
eca6f534 VN |
2370 | ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, |
2371 | (void *) data_page); | |
1da177e4 | 2372 | |
eca6f534 VN |
2373 | free_page(data_page); |
2374 | out_data: | |
2375 | kfree(kernel_dev); | |
2376 | out_dev: | |
2377 | putname(kernel_dir); | |
2378 | out_dir: | |
2379 | kfree(kernel_type); | |
2380 | out_type: | |
2381 | return ret; | |
1da177e4 LT |
2382 | } |
2383 | ||
afac7cba AV |
2384 | /* |
2385 | * Return true if path is reachable from root | |
2386 | * | |
2387 | * namespace_sem or vfsmount_lock is held | |
2388 | */ | |
643822b4 | 2389 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2390 | const struct path *root) |
2391 | { | |
643822b4 | 2392 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2393 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2394 | mnt = mnt->mnt_parent; |
afac7cba | 2395 | } |
643822b4 | 2396 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2397 | } |
2398 | ||
2399 | int path_is_under(struct path *path1, struct path *path2) | |
2400 | { | |
2401 | int res; | |
962830df | 2402 | br_read_lock(&vfsmount_lock); |
643822b4 | 2403 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
962830df | 2404 | br_read_unlock(&vfsmount_lock); |
afac7cba AV |
2405 | return res; |
2406 | } | |
2407 | EXPORT_SYMBOL(path_is_under); | |
2408 | ||
1da177e4 LT |
2409 | /* |
2410 | * pivot_root Semantics: | |
2411 | * Moves the root file system of the current process to the directory put_old, | |
2412 | * makes new_root as the new root file system of the current process, and sets | |
2413 | * root/cwd of all processes which had them on the current root to new_root. | |
2414 | * | |
2415 | * Restrictions: | |
2416 | * The new_root and put_old must be directories, and must not be on the | |
2417 | * same file system as the current process root. The put_old must be | |
2418 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2419 | * pointed to by put_old must yield the same directory as new_root. No other | |
2420 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2421 | * | |
4a0d11fa NB |
2422 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2423 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2424 | * in this situation. | |
2425 | * | |
1da177e4 LT |
2426 | * Notes: |
2427 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2428 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2429 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2430 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2431 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2432 | * first. | |
2433 | */ | |
3480b257 HC |
2434 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2435 | const char __user *, put_old) | |
1da177e4 | 2436 | { |
2d8f3038 | 2437 | struct path new, old, parent_path, root_parent, root; |
419148da | 2438 | struct mount *new_mnt, *root_mnt; |
1da177e4 LT |
2439 | int error; |
2440 | ||
2441 | if (!capable(CAP_SYS_ADMIN)) | |
2442 | return -EPERM; | |
2443 | ||
2d8f3038 | 2444 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2445 | if (error) |
2446 | goto out0; | |
1da177e4 | 2447 | |
2d8f3038 | 2448 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2449 | if (error) |
2450 | goto out1; | |
2451 | ||
2d8f3038 | 2452 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2453 | if (error) |
2454 | goto out2; | |
1da177e4 | 2455 | |
f7ad3c6b | 2456 | get_fs_root(current->fs, &root); |
b12cea91 AV |
2457 | error = lock_mount(&old); |
2458 | if (error) | |
2459 | goto out3; | |
2460 | ||
1da177e4 | 2461 | error = -EINVAL; |
419148da AV |
2462 | new_mnt = real_mount(new.mnt); |
2463 | root_mnt = real_mount(root.mnt); | |
fc7be130 AV |
2464 | if (IS_MNT_SHARED(real_mount(old.mnt)) || |
2465 | IS_MNT_SHARED(new_mnt->mnt_parent) || | |
2466 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2467 | goto out4; |
143c8c91 | 2468 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2469 | goto out4; |
1da177e4 | 2470 | error = -ENOENT; |
f3da392e | 2471 | if (d_unlinked(new.dentry)) |
b12cea91 | 2472 | goto out4; |
f3da392e | 2473 | if (d_unlinked(old.dentry)) |
b12cea91 | 2474 | goto out4; |
1da177e4 | 2475 | error = -EBUSY; |
2d8f3038 AV |
2476 | if (new.mnt == root.mnt || |
2477 | old.mnt == root.mnt) | |
b12cea91 | 2478 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2479 | error = -EINVAL; |
8c3ee42e | 2480 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2481 | goto out4; /* not a mountpoint */ |
676da58d | 2482 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2483 | goto out4; /* not attached */ |
2d8f3038 | 2484 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2485 | goto out4; /* not a mountpoint */ |
676da58d | 2486 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2487 | goto out4; /* not attached */ |
4ac91378 | 2488 | /* make sure we can reach put_old from new_root */ |
643822b4 | 2489 | if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new)) |
b12cea91 | 2490 | goto out4; |
962830df | 2491 | br_write_lock(&vfsmount_lock); |
419148da AV |
2492 | detach_mnt(new_mnt, &parent_path); |
2493 | detach_mnt(root_mnt, &root_parent); | |
4ac91378 | 2494 | /* mount old root on put_old */ |
419148da | 2495 | attach_mnt(root_mnt, &old); |
4ac91378 | 2496 | /* mount new_root on / */ |
419148da | 2497 | attach_mnt(new_mnt, &root_parent); |
6b3286ed | 2498 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
962830df | 2499 | br_write_unlock(&vfsmount_lock); |
2d8f3038 | 2500 | chroot_fs_refs(&root, &new); |
1da177e4 | 2501 | error = 0; |
b12cea91 AV |
2502 | out4: |
2503 | unlock_mount(&old); | |
2504 | if (!error) { | |
2505 | path_put(&root_parent); | |
2506 | path_put(&parent_path); | |
2507 | } | |
2508 | out3: | |
8c3ee42e | 2509 | path_put(&root); |
b12cea91 | 2510 | out2: |
2d8f3038 | 2511 | path_put(&old); |
1da177e4 | 2512 | out1: |
2d8f3038 | 2513 | path_put(&new); |
1da177e4 | 2514 | out0: |
1da177e4 | 2515 | return error; |
1da177e4 LT |
2516 | } |
2517 | ||
2518 | static void __init init_mount_tree(void) | |
2519 | { | |
2520 | struct vfsmount *mnt; | |
6b3286ed | 2521 | struct mnt_namespace *ns; |
ac748a09 | 2522 | struct path root; |
1da177e4 LT |
2523 | |
2524 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2525 | if (IS_ERR(mnt)) | |
2526 | panic("Can't create rootfs"); | |
b3e19d92 | 2527 | |
3b22edc5 TM |
2528 | ns = create_mnt_ns(mnt); |
2529 | if (IS_ERR(ns)) | |
1da177e4 | 2530 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2531 | |
2532 | init_task.nsproxy->mnt_ns = ns; | |
2533 | get_mnt_ns(ns); | |
2534 | ||
be08d6d2 AV |
2535 | root.mnt = mnt; |
2536 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2537 | |
2538 | set_fs_pwd(current->fs, &root); | |
2539 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2540 | } |
2541 | ||
74bf17cf | 2542 | void __init mnt_init(void) |
1da177e4 | 2543 | { |
13f14b4d | 2544 | unsigned u; |
15a67dd8 | 2545 | int err; |
1da177e4 | 2546 | |
390c6843 RP |
2547 | init_rwsem(&namespace_sem); |
2548 | ||
7d6fec45 | 2549 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2550 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2551 | |
b58fed8b | 2552 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2553 | |
2554 | if (!mount_hashtable) | |
2555 | panic("Failed to allocate mount hash table\n"); | |
2556 | ||
80cdc6da | 2557 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2558 | |
2559 | for (u = 0; u < HASH_SIZE; u++) | |
2560 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2561 | |
962830df | 2562 | br_lock_init(&vfsmount_lock); |
99b7db7b | 2563 | |
15a67dd8 RD |
2564 | err = sysfs_init(); |
2565 | if (err) | |
2566 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2567 | __func__, err); |
00d26666 GKH |
2568 | fs_kobj = kobject_create_and_add("fs", NULL); |
2569 | if (!fs_kobj) | |
8e24eea7 | 2570 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2571 | init_rootfs(); |
2572 | init_mount_tree(); | |
2573 | } | |
2574 | ||
616511d0 | 2575 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2576 | { |
70fbcdf4 | 2577 | LIST_HEAD(umount_list); |
616511d0 | 2578 | |
d498b25a | 2579 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2580 | return; |
390c6843 | 2581 | down_write(&namespace_sem); |
962830df | 2582 | br_write_lock(&vfsmount_lock); |
be08d6d2 | 2583 | umount_tree(ns->root, 0, &umount_list); |
962830df | 2584 | br_write_unlock(&vfsmount_lock); |
390c6843 | 2585 | up_write(&namespace_sem); |
70fbcdf4 | 2586 | release_mounts(&umount_list); |
6b3286ed | 2587 | kfree(ns); |
1da177e4 | 2588 | } |
9d412a43 AV |
2589 | |
2590 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2591 | { | |
423e0ab0 TC |
2592 | struct vfsmount *mnt; |
2593 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2594 | if (!IS_ERR(mnt)) { | |
2595 | /* | |
2596 | * it is a longterm mount, don't release mnt until | |
2597 | * we unmount before file sys is unregistered | |
2598 | */ | |
f7a99c5b | 2599 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2600 | } |
2601 | return mnt; | |
9d412a43 AV |
2602 | } |
2603 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2604 | |
2605 | void kern_unmount(struct vfsmount *mnt) | |
2606 | { | |
2607 | /* release long term mount so mount point can be released */ | |
2608 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b AV |
2609 | br_write_lock(&vfsmount_lock); |
2610 | real_mount(mnt)->mnt_ns = NULL; | |
2611 | br_write_unlock(&vfsmount_lock); | |
423e0ab0 TC |
2612 | mntput(mnt); |
2613 | } | |
2614 | } | |
2615 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2616 | |
2617 | bool our_mnt(struct vfsmount *mnt) | |
2618 | { | |
143c8c91 | 2619 | return check_mnt(real_mount(mnt)); |
02125a82 | 2620 | } |