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> | |
8823c079 | 23 | #include <linux/proc_fs.h> |
07b20889 | 24 | #include "pnode.h" |
948730b0 | 25 | #include "internal.h" |
1da177e4 | 26 | |
13f14b4d ED |
27 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
28 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
29 | ||
5addc5dd | 30 | static int event; |
73cd49ec | 31 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 32 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 33 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
34 | static int mnt_id_start = 0; |
35 | static int mnt_group_start = 1; | |
1da177e4 | 36 | |
fa3536cc | 37 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 38 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 39 | static struct rw_semaphore namespace_sem; |
1da177e4 | 40 | |
f87fd4c2 | 41 | /* /sys/fs */ |
00d26666 GKH |
42 | struct kobject *fs_kobj; |
43 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 44 | |
99b7db7b NP |
45 | /* |
46 | * vfsmount lock may be taken for read to prevent changes to the | |
47 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
48 | * up the tree. | |
49 | * | |
50 | * It should be taken for write in all cases where the vfsmount | |
51 | * tree or hash is modified or when a vfsmount structure is modified. | |
52 | */ | |
53 | DEFINE_BRLOCK(vfsmount_lock); | |
54 | ||
1da177e4 LT |
55 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
56 | { | |
b58fed8b RP |
57 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
58 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
59 | tmp = tmp + (tmp >> HASH_SHIFT); |
60 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
61 | } |
62 | ||
3d733633 DH |
63 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
64 | ||
99b7db7b NP |
65 | /* |
66 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
67 | * serialize with freeing. | |
68 | */ | |
b105e270 | 69 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
70 | { |
71 | int res; | |
72 | ||
73 | retry: | |
74 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 75 | spin_lock(&mnt_id_lock); |
15169fe7 | 76 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 77 | if (!res) |
15169fe7 | 78 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 79 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
80 | if (res == -EAGAIN) |
81 | goto retry; | |
82 | ||
83 | return res; | |
84 | } | |
85 | ||
b105e270 | 86 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 87 | { |
15169fe7 | 88 | int id = mnt->mnt_id; |
99b7db7b | 89 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
90 | ida_remove(&mnt_id_ida, id); |
91 | if (mnt_id_start > id) | |
92 | mnt_id_start = id; | |
99b7db7b | 93 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
94 | } |
95 | ||
719f5d7f MS |
96 | /* |
97 | * Allocate a new peer group ID | |
98 | * | |
99 | * mnt_group_ida is protected by namespace_sem | |
100 | */ | |
4b8b21f4 | 101 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 102 | { |
f21f6220 AV |
103 | int res; |
104 | ||
719f5d7f MS |
105 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
106 | return -ENOMEM; | |
107 | ||
f21f6220 AV |
108 | res = ida_get_new_above(&mnt_group_ida, |
109 | mnt_group_start, | |
15169fe7 | 110 | &mnt->mnt_group_id); |
f21f6220 | 111 | if (!res) |
15169fe7 | 112 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
113 | |
114 | return res; | |
719f5d7f MS |
115 | } |
116 | ||
117 | /* | |
118 | * Release a peer group ID | |
119 | */ | |
4b8b21f4 | 120 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 121 | { |
15169fe7 | 122 | int id = mnt->mnt_group_id; |
f21f6220 AV |
123 | ida_remove(&mnt_group_ida, id); |
124 | if (mnt_group_start > id) | |
125 | mnt_group_start = id; | |
15169fe7 | 126 | mnt->mnt_group_id = 0; |
719f5d7f MS |
127 | } |
128 | ||
b3e19d92 NP |
129 | /* |
130 | * vfsmount lock must be held for read | |
131 | */ | |
83adc753 | 132 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
133 | { |
134 | #ifdef CONFIG_SMP | |
68e8a9fe | 135 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
136 | #else |
137 | preempt_disable(); | |
68e8a9fe | 138 | mnt->mnt_count += n; |
b3e19d92 NP |
139 | preempt_enable(); |
140 | #endif | |
141 | } | |
142 | ||
b3e19d92 NP |
143 | /* |
144 | * vfsmount lock must be held for write | |
145 | */ | |
83adc753 | 146 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
147 | { |
148 | #ifdef CONFIG_SMP | |
f03c6599 | 149 | unsigned int count = 0; |
b3e19d92 NP |
150 | int cpu; |
151 | ||
152 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 153 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
154 | } |
155 | ||
156 | return count; | |
157 | #else | |
68e8a9fe | 158 | return mnt->mnt_count; |
b3e19d92 NP |
159 | #endif |
160 | } | |
161 | ||
b105e270 | 162 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 163 | { |
c63181e6 AV |
164 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
165 | if (mnt) { | |
73cd49ec MS |
166 | int err; |
167 | ||
c63181e6 | 168 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
169 | if (err) |
170 | goto out_free_cache; | |
171 | ||
172 | if (name) { | |
c63181e6 AV |
173 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
174 | if (!mnt->mnt_devname) | |
88b38782 | 175 | goto out_free_id; |
73cd49ec MS |
176 | } |
177 | ||
b3e19d92 | 178 | #ifdef CONFIG_SMP |
c63181e6 AV |
179 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
180 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
181 | goto out_free_devname; |
182 | ||
c63181e6 | 183 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 184 | #else |
c63181e6 AV |
185 | mnt->mnt_count = 1; |
186 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
187 | #endif |
188 | ||
c63181e6 AV |
189 | INIT_LIST_HEAD(&mnt->mnt_hash); |
190 | INIT_LIST_HEAD(&mnt->mnt_child); | |
191 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
192 | INIT_LIST_HEAD(&mnt->mnt_list); | |
193 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
194 | INIT_LIST_HEAD(&mnt->mnt_share); | |
195 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
196 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
197 | #ifdef CONFIG_FSNOTIFY |
198 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 199 | #endif |
1da177e4 | 200 | } |
c63181e6 | 201 | return mnt; |
88b38782 | 202 | |
d3ef3d73 | 203 | #ifdef CONFIG_SMP |
204 | out_free_devname: | |
c63181e6 | 205 | kfree(mnt->mnt_devname); |
d3ef3d73 | 206 | #endif |
88b38782 | 207 | out_free_id: |
c63181e6 | 208 | mnt_free_id(mnt); |
88b38782 | 209 | out_free_cache: |
c63181e6 | 210 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 211 | return NULL; |
1da177e4 LT |
212 | } |
213 | ||
3d733633 DH |
214 | /* |
215 | * Most r/o checks on a fs are for operations that take | |
216 | * discrete amounts of time, like a write() or unlink(). | |
217 | * We must keep track of when those operations start | |
218 | * (for permission checks) and when they end, so that | |
219 | * we can determine when writes are able to occur to | |
220 | * a filesystem. | |
221 | */ | |
222 | /* | |
223 | * __mnt_is_readonly: check whether a mount is read-only | |
224 | * @mnt: the mount to check for its write status | |
225 | * | |
226 | * This shouldn't be used directly ouside of the VFS. | |
227 | * It does not guarantee that the filesystem will stay | |
228 | * r/w, just that it is right *now*. This can not and | |
229 | * should not be used in place of IS_RDONLY(inode). | |
230 | * mnt_want/drop_write() will _keep_ the filesystem | |
231 | * r/w. | |
232 | */ | |
233 | int __mnt_is_readonly(struct vfsmount *mnt) | |
234 | { | |
2e4b7fcd DH |
235 | if (mnt->mnt_flags & MNT_READONLY) |
236 | return 1; | |
237 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
238 | return 1; | |
239 | return 0; | |
3d733633 DH |
240 | } |
241 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
242 | ||
83adc753 | 243 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 244 | { |
245 | #ifdef CONFIG_SMP | |
68e8a9fe | 246 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 247 | #else |
68e8a9fe | 248 | mnt->mnt_writers++; |
d3ef3d73 | 249 | #endif |
250 | } | |
3d733633 | 251 | |
83adc753 | 252 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 253 | { |
d3ef3d73 | 254 | #ifdef CONFIG_SMP |
68e8a9fe | 255 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 256 | #else |
68e8a9fe | 257 | mnt->mnt_writers--; |
d3ef3d73 | 258 | #endif |
3d733633 | 259 | } |
3d733633 | 260 | |
83adc753 | 261 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 262 | { |
d3ef3d73 | 263 | #ifdef CONFIG_SMP |
264 | unsigned int count = 0; | |
3d733633 | 265 | int cpu; |
3d733633 DH |
266 | |
267 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 268 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 269 | } |
3d733633 | 270 | |
d3ef3d73 | 271 | return count; |
272 | #else | |
273 | return mnt->mnt_writers; | |
274 | #endif | |
3d733633 DH |
275 | } |
276 | ||
4ed5e82f MS |
277 | static int mnt_is_readonly(struct vfsmount *mnt) |
278 | { | |
279 | if (mnt->mnt_sb->s_readonly_remount) | |
280 | return 1; | |
281 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
282 | smp_rmb(); | |
283 | return __mnt_is_readonly(mnt); | |
284 | } | |
285 | ||
8366025e | 286 | /* |
eb04c282 JK |
287 | * Most r/o & frozen checks on a fs are for operations that take discrete |
288 | * amounts of time, like a write() or unlink(). We must keep track of when | |
289 | * those operations start (for permission checks) and when they end, so that we | |
290 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
291 | */ |
292 | /** | |
eb04c282 | 293 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 294 | * @m: the mount on which to take a write |
8366025e | 295 | * |
eb04c282 JK |
296 | * This tells the low-level filesystem that a write is about to be performed to |
297 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
298 | * returning success. This operation does not protect against filesystem being | |
299 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
300 | * called. This is effectively a refcount. | |
8366025e | 301 | */ |
eb04c282 | 302 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 303 | { |
83adc753 | 304 | struct mount *mnt = real_mount(m); |
3d733633 | 305 | int ret = 0; |
3d733633 | 306 | |
d3ef3d73 | 307 | preempt_disable(); |
c6653a83 | 308 | mnt_inc_writers(mnt); |
d3ef3d73 | 309 | /* |
c6653a83 | 310 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 311 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
312 | * incremented count after it has set MNT_WRITE_HOLD. | |
313 | */ | |
314 | smp_mb(); | |
83adc753 | 315 | while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) |
d3ef3d73 | 316 | cpu_relax(); |
317 | /* | |
318 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
319 | * be set to match its requirements. So we must not load that until | |
320 | * MNT_WRITE_HOLD is cleared. | |
321 | */ | |
322 | smp_rmb(); | |
4ed5e82f | 323 | if (mnt_is_readonly(m)) { |
c6653a83 | 324 | mnt_dec_writers(mnt); |
3d733633 | 325 | ret = -EROFS; |
3d733633 | 326 | } |
d3ef3d73 | 327 | preempt_enable(); |
eb04c282 JK |
328 | |
329 | return ret; | |
330 | } | |
331 | ||
332 | /** | |
333 | * mnt_want_write - get write access to a mount | |
334 | * @m: the mount on which to take a write | |
335 | * | |
336 | * This tells the low-level filesystem that a write is about to be performed to | |
337 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
338 | * is not frozen) before returning success. When the write operation is | |
339 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
340 | */ | |
341 | int mnt_want_write(struct vfsmount *m) | |
342 | { | |
343 | int ret; | |
344 | ||
345 | sb_start_write(m->mnt_sb); | |
346 | ret = __mnt_want_write(m); | |
347 | if (ret) | |
348 | sb_end_write(m->mnt_sb); | |
3d733633 | 349 | return ret; |
8366025e DH |
350 | } |
351 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
352 | ||
96029c4e | 353 | /** |
354 | * mnt_clone_write - get write access to a mount | |
355 | * @mnt: the mount on which to take a write | |
356 | * | |
357 | * This is effectively like mnt_want_write, except | |
358 | * it must only be used to take an extra write reference | |
359 | * on a mountpoint that we already know has a write reference | |
360 | * on it. This allows some optimisation. | |
361 | * | |
362 | * After finished, mnt_drop_write must be called as usual to | |
363 | * drop the reference. | |
364 | */ | |
365 | int mnt_clone_write(struct vfsmount *mnt) | |
366 | { | |
367 | /* superblock may be r/o */ | |
368 | if (__mnt_is_readonly(mnt)) | |
369 | return -EROFS; | |
370 | preempt_disable(); | |
83adc753 | 371 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 372 | preempt_enable(); |
373 | return 0; | |
374 | } | |
375 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
376 | ||
377 | /** | |
eb04c282 | 378 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 379 | * @file: the file who's mount on which to take a write |
380 | * | |
eb04c282 | 381 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 382 | * do some optimisations if the file is open for write already |
383 | */ | |
eb04c282 | 384 | int __mnt_want_write_file(struct file *file) |
96029c4e | 385 | { |
2d8dd38a | 386 | struct inode *inode = file->f_dentry->d_inode; |
eb04c282 | 387 | |
2d8dd38a | 388 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) |
eb04c282 | 389 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 390 | else |
391 | return mnt_clone_write(file->f_path.mnt); | |
392 | } | |
eb04c282 JK |
393 | |
394 | /** | |
395 | * mnt_want_write_file - get write access to a file's mount | |
396 | * @file: the file who's mount on which to take a write | |
397 | * | |
398 | * This is like mnt_want_write, but it takes a file and can | |
399 | * do some optimisations if the file is open for write already | |
400 | */ | |
401 | int mnt_want_write_file(struct file *file) | |
402 | { | |
403 | int ret; | |
404 | ||
405 | sb_start_write(file->f_path.mnt->mnt_sb); | |
406 | ret = __mnt_want_write_file(file); | |
407 | if (ret) | |
408 | sb_end_write(file->f_path.mnt->mnt_sb); | |
409 | return ret; | |
410 | } | |
96029c4e | 411 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
412 | ||
8366025e | 413 | /** |
eb04c282 | 414 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
415 | * @mnt: the mount on which to give up write access |
416 | * | |
417 | * Tells the low-level filesystem that we are done | |
418 | * performing writes to it. Must be matched with | |
eb04c282 | 419 | * __mnt_want_write() call above. |
8366025e | 420 | */ |
eb04c282 | 421 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 422 | { |
d3ef3d73 | 423 | preempt_disable(); |
83adc753 | 424 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 425 | preempt_enable(); |
8366025e | 426 | } |
eb04c282 JK |
427 | |
428 | /** | |
429 | * mnt_drop_write - give up write access to a mount | |
430 | * @mnt: the mount on which to give up write access | |
431 | * | |
432 | * Tells the low-level filesystem that we are done performing writes to it and | |
433 | * also allows filesystem to be frozen again. Must be matched with | |
434 | * mnt_want_write() call above. | |
435 | */ | |
436 | void mnt_drop_write(struct vfsmount *mnt) | |
437 | { | |
438 | __mnt_drop_write(mnt); | |
439 | sb_end_write(mnt->mnt_sb); | |
440 | } | |
8366025e DH |
441 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
442 | ||
eb04c282 JK |
443 | void __mnt_drop_write_file(struct file *file) |
444 | { | |
445 | __mnt_drop_write(file->f_path.mnt); | |
446 | } | |
447 | ||
2a79f17e AV |
448 | void mnt_drop_write_file(struct file *file) |
449 | { | |
450 | mnt_drop_write(file->f_path.mnt); | |
451 | } | |
452 | EXPORT_SYMBOL(mnt_drop_write_file); | |
453 | ||
83adc753 | 454 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 455 | { |
3d733633 DH |
456 | int ret = 0; |
457 | ||
962830df | 458 | br_write_lock(&vfsmount_lock); |
83adc753 | 459 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 460 | /* |
d3ef3d73 | 461 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
462 | * should be visible before we do. | |
3d733633 | 463 | */ |
d3ef3d73 | 464 | smp_mb(); |
465 | ||
3d733633 | 466 | /* |
d3ef3d73 | 467 | * With writers on hold, if this value is zero, then there are |
468 | * definitely no active writers (although held writers may subsequently | |
469 | * increment the count, they'll have to wait, and decrement it after | |
470 | * seeing MNT_READONLY). | |
471 | * | |
472 | * It is OK to have counter incremented on one CPU and decremented on | |
473 | * another: the sum will add up correctly. The danger would be when we | |
474 | * sum up each counter, if we read a counter before it is incremented, | |
475 | * but then read another CPU's count which it has been subsequently | |
476 | * decremented from -- we would see more decrements than we should. | |
477 | * MNT_WRITE_HOLD protects against this scenario, because | |
478 | * mnt_want_write first increments count, then smp_mb, then spins on | |
479 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
480 | * we're counting up here. | |
3d733633 | 481 | */ |
c6653a83 | 482 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 483 | ret = -EBUSY; |
484 | else | |
83adc753 | 485 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 486 | /* |
487 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
488 | * that become unheld will see MNT_READONLY. | |
489 | */ | |
490 | smp_wmb(); | |
83adc753 | 491 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
962830df | 492 | br_write_unlock(&vfsmount_lock); |
3d733633 | 493 | return ret; |
8366025e | 494 | } |
8366025e | 495 | |
83adc753 | 496 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 497 | { |
962830df | 498 | br_write_lock(&vfsmount_lock); |
83adc753 | 499 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
962830df | 500 | br_write_unlock(&vfsmount_lock); |
2e4b7fcd DH |
501 | } |
502 | ||
4ed5e82f MS |
503 | int sb_prepare_remount_readonly(struct super_block *sb) |
504 | { | |
505 | struct mount *mnt; | |
506 | int err = 0; | |
507 | ||
8e8b8796 MS |
508 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
509 | if (atomic_long_read(&sb->s_remove_count)) | |
510 | return -EBUSY; | |
511 | ||
962830df | 512 | br_write_lock(&vfsmount_lock); |
4ed5e82f MS |
513 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
514 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
515 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
516 | smp_mb(); | |
517 | if (mnt_get_writers(mnt) > 0) { | |
518 | err = -EBUSY; | |
519 | break; | |
520 | } | |
521 | } | |
522 | } | |
8e8b8796 MS |
523 | if (!err && atomic_long_read(&sb->s_remove_count)) |
524 | err = -EBUSY; | |
525 | ||
4ed5e82f MS |
526 | if (!err) { |
527 | sb->s_readonly_remount = 1; | |
528 | smp_wmb(); | |
529 | } | |
530 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
531 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
532 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
533 | } | |
962830df | 534 | br_write_unlock(&vfsmount_lock); |
4ed5e82f MS |
535 | |
536 | return err; | |
537 | } | |
538 | ||
b105e270 | 539 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 540 | { |
52ba1621 | 541 | kfree(mnt->mnt_devname); |
73cd49ec | 542 | mnt_free_id(mnt); |
d3ef3d73 | 543 | #ifdef CONFIG_SMP |
68e8a9fe | 544 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 545 | #endif |
b105e270 | 546 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
547 | } |
548 | ||
549 | /* | |
a05964f3 RP |
550 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
551 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 552 | * vfsmount_lock must be held for read or write. |
1da177e4 | 553 | */ |
c7105365 | 554 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 555 | int dir) |
1da177e4 | 556 | { |
b58fed8b RP |
557 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
558 | struct list_head *tmp = head; | |
c7105365 | 559 | struct mount *p, *found = NULL; |
1da177e4 | 560 | |
1da177e4 | 561 | for (;;) { |
a05964f3 | 562 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
563 | p = NULL; |
564 | if (tmp == head) | |
565 | break; | |
1b8e5564 | 566 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 567 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 568 | found = p; |
1da177e4 LT |
569 | break; |
570 | } | |
571 | } | |
1da177e4 LT |
572 | return found; |
573 | } | |
574 | ||
a05964f3 | 575 | /* |
f015f126 DH |
576 | * lookup_mnt - Return the first child mount mounted at path |
577 | * | |
578 | * "First" means first mounted chronologically. If you create the | |
579 | * following mounts: | |
580 | * | |
581 | * mount /dev/sda1 /mnt | |
582 | * mount /dev/sda2 /mnt | |
583 | * mount /dev/sda3 /mnt | |
584 | * | |
585 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
586 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
587 | * /dev/sda2, then /dev/sda3, then NULL. | |
588 | * | |
589 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 590 | */ |
1c755af4 | 591 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 592 | { |
c7105365 | 593 | struct mount *child_mnt; |
99b7db7b | 594 | |
962830df | 595 | br_read_lock(&vfsmount_lock); |
c7105365 AV |
596 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
597 | if (child_mnt) { | |
598 | mnt_add_count(child_mnt, 1); | |
962830df | 599 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
600 | return &child_mnt->mnt; |
601 | } else { | |
962830df | 602 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
603 | return NULL; |
604 | } | |
a05964f3 RP |
605 | } |
606 | ||
143c8c91 | 607 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 608 | { |
6b3286ed | 609 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
610 | } |
611 | ||
99b7db7b NP |
612 | /* |
613 | * vfsmount lock must be held for write | |
614 | */ | |
6b3286ed | 615 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
616 | { |
617 | if (ns) { | |
618 | ns->event = ++event; | |
619 | wake_up_interruptible(&ns->poll); | |
620 | } | |
621 | } | |
622 | ||
99b7db7b NP |
623 | /* |
624 | * vfsmount lock must be held for write | |
625 | */ | |
6b3286ed | 626 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
627 | { |
628 | if (ns && ns->event != event) { | |
629 | ns->event = event; | |
630 | wake_up_interruptible(&ns->poll); | |
631 | } | |
632 | } | |
633 | ||
5f57cbcc NP |
634 | /* |
635 | * Clear dentry's mounted state if it has no remaining mounts. | |
636 | * vfsmount_lock must be held for write. | |
637 | */ | |
aa0a4cf0 | 638 | static void dentry_reset_mounted(struct dentry *dentry) |
5f57cbcc NP |
639 | { |
640 | unsigned u; | |
641 | ||
642 | for (u = 0; u < HASH_SIZE; u++) { | |
d5e50f74 | 643 | struct mount *p; |
5f57cbcc | 644 | |
1b8e5564 | 645 | list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { |
a73324da | 646 | if (p->mnt_mountpoint == dentry) |
5f57cbcc NP |
647 | return; |
648 | } | |
649 | } | |
650 | spin_lock(&dentry->d_lock); | |
651 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
652 | spin_unlock(&dentry->d_lock); | |
653 | } | |
654 | ||
99b7db7b NP |
655 | /* |
656 | * vfsmount lock must be held for write | |
657 | */ | |
419148da AV |
658 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
659 | { | |
a73324da | 660 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
661 | old_path->mnt = &mnt->mnt_parent->mnt; |
662 | mnt->mnt_parent = mnt; | |
a73324da | 663 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 664 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 665 | list_del_init(&mnt->mnt_hash); |
aa0a4cf0 | 666 | dentry_reset_mounted(old_path->dentry); |
1da177e4 LT |
667 | } |
668 | ||
99b7db7b NP |
669 | /* |
670 | * vfsmount lock must be held for write | |
671 | */ | |
14cf1fa8 | 672 | void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry, |
44d964d6 | 673 | struct mount *child_mnt) |
b90fa9ae | 674 | { |
3a2393d7 | 675 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
a73324da | 676 | child_mnt->mnt_mountpoint = dget(dentry); |
3a2393d7 | 677 | child_mnt->mnt_parent = mnt; |
5f57cbcc NP |
678 | spin_lock(&dentry->d_lock); |
679 | dentry->d_flags |= DCACHE_MOUNTED; | |
680 | spin_unlock(&dentry->d_lock); | |
b90fa9ae RP |
681 | } |
682 | ||
99b7db7b NP |
683 | /* |
684 | * vfsmount lock must be held for write | |
685 | */ | |
419148da | 686 | static void attach_mnt(struct mount *mnt, struct path *path) |
1da177e4 | 687 | { |
14cf1fa8 | 688 | mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt); |
1b8e5564 | 689 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 | 690 | hash(path->mnt, path->dentry)); |
6b41d536 | 691 | list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts); |
b90fa9ae RP |
692 | } |
693 | ||
694 | /* | |
99b7db7b | 695 | * vfsmount lock must be held for write |
b90fa9ae | 696 | */ |
4b2619a5 | 697 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 698 | { |
0714a533 | 699 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 700 | struct mount *m; |
b90fa9ae | 701 | LIST_HEAD(head); |
143c8c91 | 702 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 703 | |
0714a533 | 704 | BUG_ON(parent == mnt); |
b90fa9ae | 705 | |
1a4eeaf2 | 706 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 707 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 708 | m->mnt_ns = n; |
f03c6599 | 709 | |
b90fa9ae RP |
710 | list_splice(&head, n->list.prev); |
711 | ||
1b8e5564 | 712 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 713 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 714 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 715 | touch_mnt_namespace(n); |
1da177e4 LT |
716 | } |
717 | ||
909b0a88 | 718 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 719 | { |
6b41d536 AV |
720 | struct list_head *next = p->mnt_mounts.next; |
721 | if (next == &p->mnt_mounts) { | |
1da177e4 | 722 | while (1) { |
909b0a88 | 723 | if (p == root) |
1da177e4 | 724 | return NULL; |
6b41d536 AV |
725 | next = p->mnt_child.next; |
726 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 727 | break; |
0714a533 | 728 | p = p->mnt_parent; |
1da177e4 LT |
729 | } |
730 | } | |
6b41d536 | 731 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
732 | } |
733 | ||
315fc83e | 734 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 735 | { |
6b41d536 AV |
736 | struct list_head *prev = p->mnt_mounts.prev; |
737 | while (prev != &p->mnt_mounts) { | |
738 | p = list_entry(prev, struct mount, mnt_child); | |
739 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
740 | } |
741 | return p; | |
742 | } | |
743 | ||
9d412a43 AV |
744 | struct vfsmount * |
745 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
746 | { | |
b105e270 | 747 | struct mount *mnt; |
9d412a43 AV |
748 | struct dentry *root; |
749 | ||
750 | if (!type) | |
751 | return ERR_PTR(-ENODEV); | |
752 | ||
753 | mnt = alloc_vfsmnt(name); | |
754 | if (!mnt) | |
755 | return ERR_PTR(-ENOMEM); | |
756 | ||
757 | if (flags & MS_KERNMOUNT) | |
b105e270 | 758 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
759 | |
760 | root = mount_fs(type, flags, name, data); | |
761 | if (IS_ERR(root)) { | |
762 | free_vfsmnt(mnt); | |
763 | return ERR_CAST(root); | |
764 | } | |
765 | ||
b105e270 AV |
766 | mnt->mnt.mnt_root = root; |
767 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 768 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 769 | mnt->mnt_parent = mnt; |
962830df | 770 | br_write_lock(&vfsmount_lock); |
39f7c4db | 771 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
962830df | 772 | br_write_unlock(&vfsmount_lock); |
b105e270 | 773 | return &mnt->mnt; |
9d412a43 AV |
774 | } |
775 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
776 | ||
87129cc0 | 777 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 778 | int flag) |
1da177e4 | 779 | { |
87129cc0 | 780 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
781 | struct mount *mnt; |
782 | int err; | |
1da177e4 | 783 | |
be34d1a3 DH |
784 | mnt = alloc_vfsmnt(old->mnt_devname); |
785 | if (!mnt) | |
786 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 787 | |
be34d1a3 DH |
788 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
789 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
790 | else | |
791 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 792 | |
be34d1a3 DH |
793 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
794 | err = mnt_alloc_group_id(mnt); | |
795 | if (err) | |
796 | goto out_free; | |
1da177e4 | 797 | } |
be34d1a3 DH |
798 | |
799 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
800 | atomic_inc(&sb->s_active); | |
801 | mnt->mnt.mnt_sb = sb; | |
802 | mnt->mnt.mnt_root = dget(root); | |
803 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
804 | mnt->mnt_parent = mnt; | |
805 | br_write_lock(&vfsmount_lock); | |
806 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
807 | br_write_unlock(&vfsmount_lock); | |
808 | ||
809 | if (flag & CL_SLAVE) { | |
810 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
811 | mnt->mnt_master = old; | |
812 | CLEAR_MNT_SHARED(mnt); | |
813 | } else if (!(flag & CL_PRIVATE)) { | |
814 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
815 | list_add(&mnt->mnt_share, &old->mnt_share); | |
816 | if (IS_MNT_SLAVE(old)) | |
817 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
818 | mnt->mnt_master = old->mnt_master; | |
819 | } | |
820 | if (flag & CL_MAKE_SHARED) | |
821 | set_mnt_shared(mnt); | |
822 | ||
823 | /* stick the duplicate mount on the same expiry list | |
824 | * as the original if that was on one */ | |
825 | if (flag & CL_EXPIRE) { | |
826 | if (!list_empty(&old->mnt_expire)) | |
827 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
828 | } | |
829 | ||
cb338d06 | 830 | return mnt; |
719f5d7f MS |
831 | |
832 | out_free: | |
833 | free_vfsmnt(mnt); | |
be34d1a3 | 834 | return ERR_PTR(err); |
1da177e4 LT |
835 | } |
836 | ||
83adc753 | 837 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 838 | { |
83adc753 AV |
839 | struct vfsmount *m = &mnt->mnt; |
840 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 841 | |
3d733633 DH |
842 | /* |
843 | * This probably indicates that somebody messed | |
844 | * up a mnt_want/drop_write() pair. If this | |
845 | * happens, the filesystem was probably unable | |
846 | * to make r/w->r/o transitions. | |
847 | */ | |
d3ef3d73 | 848 | /* |
b3e19d92 NP |
849 | * The locking used to deal with mnt_count decrement provides barriers, |
850 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 851 | */ |
c6653a83 | 852 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
853 | fsnotify_vfsmount_delete(m); |
854 | dput(m->mnt_root); | |
855 | free_vfsmnt(mnt); | |
1da177e4 LT |
856 | deactivate_super(sb); |
857 | } | |
858 | ||
900148dc | 859 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 860 | { |
b3e19d92 | 861 | put_again: |
f03c6599 | 862 | #ifdef CONFIG_SMP |
962830df | 863 | br_read_lock(&vfsmount_lock); |
f7a99c5b AV |
864 | if (likely(mnt->mnt_ns)) { |
865 | /* shouldn't be the last one */ | |
aa9c0e07 | 866 | mnt_add_count(mnt, -1); |
962830df | 867 | br_read_unlock(&vfsmount_lock); |
f03c6599 | 868 | return; |
b3e19d92 | 869 | } |
962830df | 870 | br_read_unlock(&vfsmount_lock); |
b3e19d92 | 871 | |
962830df | 872 | br_write_lock(&vfsmount_lock); |
aa9c0e07 | 873 | mnt_add_count(mnt, -1); |
b3e19d92 | 874 | if (mnt_get_count(mnt)) { |
962830df | 875 | br_write_unlock(&vfsmount_lock); |
99b7db7b NP |
876 | return; |
877 | } | |
b3e19d92 | 878 | #else |
aa9c0e07 | 879 | mnt_add_count(mnt, -1); |
b3e19d92 | 880 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 881 | return; |
962830df | 882 | br_write_lock(&vfsmount_lock); |
f03c6599 | 883 | #endif |
863d684f AV |
884 | if (unlikely(mnt->mnt_pinned)) { |
885 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
886 | mnt->mnt_pinned = 0; | |
962830df | 887 | br_write_unlock(&vfsmount_lock); |
900148dc | 888 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 889 | goto put_again; |
7b7b1ace | 890 | } |
962830df | 891 | |
39f7c4db | 892 | list_del(&mnt->mnt_instance); |
962830df | 893 | br_write_unlock(&vfsmount_lock); |
b3e19d92 NP |
894 | mntfree(mnt); |
895 | } | |
b3e19d92 NP |
896 | |
897 | void mntput(struct vfsmount *mnt) | |
898 | { | |
899 | if (mnt) { | |
863d684f | 900 | struct mount *m = real_mount(mnt); |
b3e19d92 | 901 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
902 | if (unlikely(m->mnt_expiry_mark)) |
903 | m->mnt_expiry_mark = 0; | |
904 | mntput_no_expire(m); | |
b3e19d92 NP |
905 | } |
906 | } | |
907 | EXPORT_SYMBOL(mntput); | |
908 | ||
909 | struct vfsmount *mntget(struct vfsmount *mnt) | |
910 | { | |
911 | if (mnt) | |
83adc753 | 912 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
913 | return mnt; |
914 | } | |
915 | EXPORT_SYMBOL(mntget); | |
916 | ||
7b7b1ace AV |
917 | void mnt_pin(struct vfsmount *mnt) |
918 | { | |
962830df | 919 | br_write_lock(&vfsmount_lock); |
863d684f | 920 | real_mount(mnt)->mnt_pinned++; |
962830df | 921 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 922 | } |
7b7b1ace AV |
923 | EXPORT_SYMBOL(mnt_pin); |
924 | ||
863d684f | 925 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 926 | { |
863d684f | 927 | struct mount *mnt = real_mount(m); |
962830df | 928 | br_write_lock(&vfsmount_lock); |
7b7b1ace | 929 | if (mnt->mnt_pinned) { |
863d684f | 930 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
931 | mnt->mnt_pinned--; |
932 | } | |
962830df | 933 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 934 | } |
7b7b1ace | 935 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 936 | |
b3b304a2 MS |
937 | static inline void mangle(struct seq_file *m, const char *s) |
938 | { | |
939 | seq_escape(m, s, " \t\n\\"); | |
940 | } | |
941 | ||
942 | /* | |
943 | * Simple .show_options callback for filesystems which don't want to | |
944 | * implement more complex mount option showing. | |
945 | * | |
946 | * See also save_mount_options(). | |
947 | */ | |
34c80b1d | 948 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 949 | { |
2a32cebd AV |
950 | const char *options; |
951 | ||
952 | rcu_read_lock(); | |
34c80b1d | 953 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
954 | |
955 | if (options != NULL && options[0]) { | |
956 | seq_putc(m, ','); | |
957 | mangle(m, options); | |
958 | } | |
2a32cebd | 959 | rcu_read_unlock(); |
b3b304a2 MS |
960 | |
961 | return 0; | |
962 | } | |
963 | EXPORT_SYMBOL(generic_show_options); | |
964 | ||
965 | /* | |
966 | * If filesystem uses generic_show_options(), this function should be | |
967 | * called from the fill_super() callback. | |
968 | * | |
969 | * The .remount_fs callback usually needs to be handled in a special | |
970 | * way, to make sure, that previous options are not overwritten if the | |
971 | * remount fails. | |
972 | * | |
973 | * Also note, that if the filesystem's .remount_fs function doesn't | |
974 | * reset all options to their default value, but changes only newly | |
975 | * given options, then the displayed options will not reflect reality | |
976 | * any more. | |
977 | */ | |
978 | void save_mount_options(struct super_block *sb, char *options) | |
979 | { | |
2a32cebd AV |
980 | BUG_ON(sb->s_options); |
981 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
982 | } |
983 | EXPORT_SYMBOL(save_mount_options); | |
984 | ||
2a32cebd AV |
985 | void replace_mount_options(struct super_block *sb, char *options) |
986 | { | |
987 | char *old = sb->s_options; | |
988 | rcu_assign_pointer(sb->s_options, options); | |
989 | if (old) { | |
990 | synchronize_rcu(); | |
991 | kfree(old); | |
992 | } | |
993 | } | |
994 | EXPORT_SYMBOL(replace_mount_options); | |
995 | ||
a1a2c409 | 996 | #ifdef CONFIG_PROC_FS |
0226f492 | 997 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
998 | static void *m_start(struct seq_file *m, loff_t *pos) |
999 | { | |
6ce6e24e | 1000 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1001 | |
390c6843 | 1002 | down_read(&namespace_sem); |
a1a2c409 | 1003 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
1004 | } |
1005 | ||
1006 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1007 | { | |
6ce6e24e | 1008 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1009 | |
a1a2c409 | 1010 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
1011 | } |
1012 | ||
1013 | static void m_stop(struct seq_file *m, void *v) | |
1014 | { | |
390c6843 | 1015 | up_read(&namespace_sem); |
1da177e4 LT |
1016 | } |
1017 | ||
0226f492 | 1018 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1019 | { |
6ce6e24e | 1020 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1021 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1022 | return p->show(m, &r->mnt); |
1da177e4 LT |
1023 | } |
1024 | ||
a1a2c409 | 1025 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1026 | .start = m_start, |
1027 | .next = m_next, | |
1028 | .stop = m_stop, | |
0226f492 | 1029 | .show = m_show, |
b4629fe2 | 1030 | }; |
a1a2c409 | 1031 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1032 | |
1da177e4 LT |
1033 | /** |
1034 | * may_umount_tree - check if a mount tree is busy | |
1035 | * @mnt: root of mount tree | |
1036 | * | |
1037 | * This is called to check if a tree of mounts has any | |
1038 | * open files, pwds, chroots or sub mounts that are | |
1039 | * busy. | |
1040 | */ | |
909b0a88 | 1041 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1042 | { |
909b0a88 | 1043 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1044 | int actual_refs = 0; |
1045 | int minimum_refs = 0; | |
315fc83e | 1046 | struct mount *p; |
909b0a88 | 1047 | BUG_ON(!m); |
1da177e4 | 1048 | |
b3e19d92 | 1049 | /* write lock needed for mnt_get_count */ |
962830df | 1050 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1051 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1052 | actual_refs += mnt_get_count(p); |
1da177e4 | 1053 | minimum_refs += 2; |
1da177e4 | 1054 | } |
962830df | 1055 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1056 | |
1057 | if (actual_refs > minimum_refs) | |
e3474a8e | 1058 | return 0; |
1da177e4 | 1059 | |
e3474a8e | 1060 | return 1; |
1da177e4 LT |
1061 | } |
1062 | ||
1063 | EXPORT_SYMBOL(may_umount_tree); | |
1064 | ||
1065 | /** | |
1066 | * may_umount - check if a mount point is busy | |
1067 | * @mnt: root of mount | |
1068 | * | |
1069 | * This is called to check if a mount point has any | |
1070 | * open files, pwds, chroots or sub mounts. If the | |
1071 | * mount has sub mounts this will return busy | |
1072 | * regardless of whether the sub mounts are busy. | |
1073 | * | |
1074 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1075 | * give false negatives. The main reason why it's here is that we need | |
1076 | * a non-destructive way to look for easily umountable filesystems. | |
1077 | */ | |
1078 | int may_umount(struct vfsmount *mnt) | |
1079 | { | |
e3474a8e | 1080 | int ret = 1; |
8ad08d8a | 1081 | down_read(&namespace_sem); |
962830df | 1082 | br_write_lock(&vfsmount_lock); |
1ab59738 | 1083 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1084 | ret = 0; |
962830df | 1085 | br_write_unlock(&vfsmount_lock); |
8ad08d8a | 1086 | up_read(&namespace_sem); |
a05964f3 | 1087 | return ret; |
1da177e4 LT |
1088 | } |
1089 | ||
1090 | EXPORT_SYMBOL(may_umount); | |
1091 | ||
b90fa9ae | 1092 | void release_mounts(struct list_head *head) |
70fbcdf4 | 1093 | { |
d5e50f74 | 1094 | struct mount *mnt; |
bf066c7d | 1095 | while (!list_empty(head)) { |
1b8e5564 AV |
1096 | mnt = list_first_entry(head, struct mount, mnt_hash); |
1097 | list_del_init(&mnt->mnt_hash); | |
676da58d | 1098 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 1099 | struct dentry *dentry; |
863d684f | 1100 | struct mount *m; |
99b7db7b | 1101 | |
962830df | 1102 | br_write_lock(&vfsmount_lock); |
a73324da | 1103 | dentry = mnt->mnt_mountpoint; |
863d684f | 1104 | m = mnt->mnt_parent; |
a73324da | 1105 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1106 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1107 | m->mnt_ghosts--; |
962830df | 1108 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1109 | dput(dentry); |
863d684f | 1110 | mntput(&m->mnt); |
70fbcdf4 | 1111 | } |
d5e50f74 | 1112 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1113 | } |
1114 | } | |
1115 | ||
99b7db7b NP |
1116 | /* |
1117 | * vfsmount lock must be held for write | |
1118 | * namespace_sem must be held for write | |
1119 | */ | |
761d5c38 | 1120 | void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) |
1da177e4 | 1121 | { |
7b8a53fd | 1122 | LIST_HEAD(tmp_list); |
315fc83e | 1123 | struct mount *p; |
1da177e4 | 1124 | |
909b0a88 | 1125 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1126 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1127 | |
a05964f3 | 1128 | if (propagate) |
7b8a53fd | 1129 | propagate_umount(&tmp_list); |
a05964f3 | 1130 | |
1b8e5564 | 1131 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1132 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1133 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1134 | __touch_mnt_namespace(p->mnt_ns); |
1135 | p->mnt_ns = NULL; | |
6b41d536 | 1136 | list_del_init(&p->mnt_child); |
676da58d | 1137 | if (mnt_has_parent(p)) { |
863d684f | 1138 | p->mnt_parent->mnt_ghosts++; |
a73324da | 1139 | dentry_reset_mounted(p->mnt_mountpoint); |
7c4b93d8 | 1140 | } |
0f0afb1d | 1141 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1142 | } |
7b8a53fd | 1143 | list_splice(&tmp_list, kill); |
1da177e4 LT |
1144 | } |
1145 | ||
692afc31 | 1146 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts); |
c35038be | 1147 | |
1ab59738 | 1148 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1149 | { |
1ab59738 | 1150 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 | 1151 | int retval; |
70fbcdf4 | 1152 | LIST_HEAD(umount_list); |
1da177e4 | 1153 | |
1ab59738 | 1154 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1155 | if (retval) |
1156 | return retval; | |
1157 | ||
1158 | /* | |
1159 | * Allow userspace to request a mountpoint be expired rather than | |
1160 | * unmounting unconditionally. Unmount only happens if: | |
1161 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1162 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1163 | */ | |
1164 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1165 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1166 | flags & (MNT_FORCE | MNT_DETACH)) |
1167 | return -EINVAL; | |
1168 | ||
b3e19d92 NP |
1169 | /* |
1170 | * probably don't strictly need the lock here if we examined | |
1171 | * all race cases, but it's a slowpath. | |
1172 | */ | |
962830df | 1173 | br_write_lock(&vfsmount_lock); |
83adc753 | 1174 | if (mnt_get_count(mnt) != 2) { |
962830df | 1175 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1176 | return -EBUSY; |
b3e19d92 | 1177 | } |
962830df | 1178 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1179 | |
863d684f | 1180 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1181 | return -EAGAIN; |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * If we may have to abort operations to get out of this | |
1186 | * mount, and they will themselves hold resources we must | |
1187 | * allow the fs to do things. In the Unix tradition of | |
1188 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1189 | * might fail to complete on the first run through as other tasks | |
1190 | * must return, and the like. Thats for the mount program to worry | |
1191 | * about for the moment. | |
1192 | */ | |
1193 | ||
42faad99 | 1194 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1195 | sb->s_op->umount_begin(sb); |
42faad99 | 1196 | } |
1da177e4 LT |
1197 | |
1198 | /* | |
1199 | * No sense to grab the lock for this test, but test itself looks | |
1200 | * somewhat bogus. Suggestions for better replacement? | |
1201 | * Ho-hum... In principle, we might treat that as umount + switch | |
1202 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1203 | * Actually it makes sense, especially if rootfs would contain a | |
1204 | * /reboot - static binary that would close all descriptors and | |
1205 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1206 | */ | |
1ab59738 | 1207 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1208 | /* |
1209 | * Special case for "unmounting" root ... | |
1210 | * we just try to remount it readonly. | |
1211 | */ | |
1212 | down_write(&sb->s_umount); | |
4aa98cf7 | 1213 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1214 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1215 | up_write(&sb->s_umount); |
1216 | return retval; | |
1217 | } | |
1218 | ||
390c6843 | 1219 | down_write(&namespace_sem); |
962830df | 1220 | br_write_lock(&vfsmount_lock); |
5addc5dd | 1221 | event++; |
1da177e4 | 1222 | |
c35038be | 1223 | if (!(flags & MNT_DETACH)) |
1ab59738 | 1224 | shrink_submounts(mnt, &umount_list); |
c35038be | 1225 | |
1da177e4 | 1226 | retval = -EBUSY; |
a05964f3 | 1227 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1228 | if (!list_empty(&mnt->mnt_list)) |
1ab59738 | 1229 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1230 | retval = 0; |
1231 | } | |
962830df | 1232 | br_write_unlock(&vfsmount_lock); |
390c6843 | 1233 | up_write(&namespace_sem); |
70fbcdf4 | 1234 | release_mounts(&umount_list); |
1da177e4 LT |
1235 | return retval; |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * Now umount can handle mount points as well as block devices. | |
1240 | * This is important for filesystems which use unnamed block devices. | |
1241 | * | |
1242 | * We now support a flag for forced unmount like the other 'big iron' | |
1243 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1244 | */ | |
1245 | ||
bdc480e3 | 1246 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1247 | { |
2d8f3038 | 1248 | struct path path; |
900148dc | 1249 | struct mount *mnt; |
1da177e4 | 1250 | int retval; |
db1f05bb | 1251 | int lookup_flags = 0; |
1da177e4 | 1252 | |
db1f05bb MS |
1253 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1254 | return -EINVAL; | |
1255 | ||
1256 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1257 | lookup_flags |= LOOKUP_FOLLOW; | |
1258 | ||
1259 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1260 | if (retval) |
1261 | goto out; | |
900148dc | 1262 | mnt = real_mount(path.mnt); |
1da177e4 | 1263 | retval = -EINVAL; |
2d8f3038 | 1264 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1265 | goto dput_and_out; |
143c8c91 | 1266 | if (!check_mnt(mnt)) |
1da177e4 LT |
1267 | goto dput_and_out; |
1268 | ||
1269 | retval = -EPERM; | |
1270 | if (!capable(CAP_SYS_ADMIN)) | |
1271 | goto dput_and_out; | |
1272 | ||
900148dc | 1273 | retval = do_umount(mnt, flags); |
1da177e4 | 1274 | dput_and_out: |
429731b1 | 1275 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1276 | dput(path.dentry); |
900148dc | 1277 | mntput_no_expire(mnt); |
1da177e4 LT |
1278 | out: |
1279 | return retval; | |
1280 | } | |
1281 | ||
1282 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1283 | ||
1284 | /* | |
b58fed8b | 1285 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1286 | */ |
bdc480e3 | 1287 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1288 | { |
b58fed8b | 1289 | return sys_umount(name, 0); |
1da177e4 LT |
1290 | } |
1291 | ||
1292 | #endif | |
1293 | ||
2d92ab3c | 1294 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1295 | { |
1296 | if (capable(CAP_SYS_ADMIN)) | |
1297 | return 0; | |
1298 | return -EPERM; | |
1299 | #ifdef notyet | |
2d92ab3c | 1300 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1301 | return -EPERM; |
2d92ab3c | 1302 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1303 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1304 | return -EPERM; |
1305 | } | |
2d92ab3c | 1306 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1307 | return -EPERM; |
1308 | return 0; | |
1309 | #endif | |
1310 | } | |
1311 | ||
8823c079 EB |
1312 | static bool mnt_ns_loop(struct path *path) |
1313 | { | |
1314 | /* Could bind mounting the mount namespace inode cause a | |
1315 | * mount namespace loop? | |
1316 | */ | |
1317 | struct inode *inode = path->dentry->d_inode; | |
1318 | struct proc_inode *ei; | |
1319 | struct mnt_namespace *mnt_ns; | |
1320 | ||
1321 | if (!proc_ns_inode(inode)) | |
1322 | return false; | |
1323 | ||
1324 | ei = PROC_I(inode); | |
1325 | if (ei->ns_ops != &mntns_operations) | |
1326 | return false; | |
1327 | ||
1328 | mnt_ns = ei->ns; | |
1329 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; | |
1330 | } | |
1331 | ||
87129cc0 | 1332 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1333 | int flag) |
1da177e4 | 1334 | { |
a73324da | 1335 | struct mount *res, *p, *q, *r; |
1a390689 | 1336 | struct path path; |
1da177e4 | 1337 | |
fc7be130 | 1338 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
be34d1a3 | 1339 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1340 | |
36341f64 | 1341 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1342 | if (IS_ERR(q)) |
1343 | return q; | |
1344 | ||
a73324da | 1345 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1346 | |
1347 | p = mnt; | |
6b41d536 | 1348 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1349 | struct mount *s; |
7ec02ef1 | 1350 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1351 | continue; |
1352 | ||
909b0a88 | 1353 | for (s = r; s; s = next_mnt(s, r)) { |
fc7be130 | 1354 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
9676f0c6 RP |
1355 | s = skip_mnt_tree(s); |
1356 | continue; | |
1357 | } | |
0714a533 AV |
1358 | while (p != s->mnt_parent) { |
1359 | p = p->mnt_parent; | |
1360 | q = q->mnt_parent; | |
1da177e4 | 1361 | } |
87129cc0 | 1362 | p = s; |
cb338d06 | 1363 | path.mnt = &q->mnt; |
a73324da | 1364 | path.dentry = p->mnt_mountpoint; |
87129cc0 | 1365 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1366 | if (IS_ERR(q)) |
1367 | goto out; | |
962830df | 1368 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 1369 | list_add_tail(&q->mnt_list, &res->mnt_list); |
cb338d06 | 1370 | attach_mnt(q, &path); |
962830df | 1371 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1372 | } |
1373 | } | |
1374 | return res; | |
be34d1a3 | 1375 | out: |
1da177e4 | 1376 | if (res) { |
70fbcdf4 | 1377 | LIST_HEAD(umount_list); |
962830df | 1378 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1379 | umount_tree(res, 0, &umount_list); |
962830df | 1380 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1381 | release_mounts(&umount_list); |
1da177e4 | 1382 | } |
be34d1a3 | 1383 | return q; |
1da177e4 LT |
1384 | } |
1385 | ||
be34d1a3 DH |
1386 | /* Caller should check returned pointer for errors */ |
1387 | ||
589ff870 | 1388 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1389 | { |
cb338d06 | 1390 | struct mount *tree; |
1a60a280 | 1391 | down_write(&namespace_sem); |
87129cc0 AV |
1392 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1393 | CL_COPY_ALL | CL_PRIVATE); | |
1a60a280 | 1394 | up_write(&namespace_sem); |
be34d1a3 DH |
1395 | if (IS_ERR(tree)) |
1396 | return NULL; | |
1397 | return &tree->mnt; | |
8aec0809 AV |
1398 | } |
1399 | ||
1400 | void drop_collected_mounts(struct vfsmount *mnt) | |
1401 | { | |
1402 | LIST_HEAD(umount_list); | |
1a60a280 | 1403 | down_write(&namespace_sem); |
962830df | 1404 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1405 | umount_tree(real_mount(mnt), 0, &umount_list); |
962830df | 1406 | br_write_unlock(&vfsmount_lock); |
1a60a280 | 1407 | up_write(&namespace_sem); |
8aec0809 AV |
1408 | release_mounts(&umount_list); |
1409 | } | |
1410 | ||
1f707137 AV |
1411 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1412 | struct vfsmount *root) | |
1413 | { | |
1a4eeaf2 | 1414 | struct mount *mnt; |
1f707137 AV |
1415 | int res = f(root, arg); |
1416 | if (res) | |
1417 | return res; | |
1a4eeaf2 AV |
1418 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1419 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1420 | if (res) |
1421 | return res; | |
1422 | } | |
1423 | return 0; | |
1424 | } | |
1425 | ||
4b8b21f4 | 1426 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1427 | { |
315fc83e | 1428 | struct mount *p; |
719f5d7f | 1429 | |
909b0a88 | 1430 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1431 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1432 | mnt_release_group_id(p); |
719f5d7f MS |
1433 | } |
1434 | } | |
1435 | ||
4b8b21f4 | 1436 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1437 | { |
315fc83e | 1438 | struct mount *p; |
719f5d7f | 1439 | |
909b0a88 | 1440 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1441 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1442 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1443 | if (err) { |
4b8b21f4 | 1444 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1445 | return err; |
1446 | } | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | return 0; | |
1451 | } | |
1452 | ||
b90fa9ae RP |
1453 | /* |
1454 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1455 | * @nd : place the mount tree @source_mnt is attached |
1456 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1457 | * store the parent mount and mountpoint dentry. | |
1458 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1459 | * |
1460 | * NOTE: in the table below explains the semantics when a source mount | |
1461 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1462 | * --------------------------------------------------------------------------- |
1463 | * | BIND MOUNT OPERATION | | |
1464 | * |************************************************************************** | |
1465 | * | source-->| shared | private | slave | unbindable | | |
1466 | * | dest | | | | | | |
1467 | * | | | | | | | | |
1468 | * | v | | | | | | |
1469 | * |************************************************************************** | |
1470 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1471 | * | | | | | | | |
1472 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1473 | * *************************************************************************** | |
b90fa9ae RP |
1474 | * A bind operation clones the source mount and mounts the clone on the |
1475 | * destination mount. | |
1476 | * | |
1477 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1478 | * tree of the destination mount and the cloned mount is added to | |
1479 | * the peer group of the source mount. | |
1480 | * (+) the cloned mount is created under the destination mount and is marked | |
1481 | * as shared. The cloned mount is added to the peer group of the source | |
1482 | * mount. | |
5afe0022 RP |
1483 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1484 | * of the destination mount and the cloned mount is made slave | |
1485 | * of the same master as that of the source mount. The cloned mount | |
1486 | * is marked as 'shared and slave'. | |
1487 | * (*) the cloned mount is made a slave of the same master as that of the | |
1488 | * source mount. | |
1489 | * | |
9676f0c6 RP |
1490 | * --------------------------------------------------------------------------- |
1491 | * | MOVE MOUNT OPERATION | | |
1492 | * |************************************************************************** | |
1493 | * | source-->| shared | private | slave | unbindable | | |
1494 | * | dest | | | | | | |
1495 | * | | | | | | | | |
1496 | * | v | | | | | | |
1497 | * |************************************************************************** | |
1498 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1499 | * | | | | | | | |
1500 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1501 | * *************************************************************************** | |
5afe0022 RP |
1502 | * |
1503 | * (+) the mount is moved to the destination. And is then propagated to | |
1504 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1505 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1506 | * (+++) the mount is moved to the destination and is then propagated to |
1507 | * all the mounts belonging to the destination mount's propagation tree. | |
1508 | * the mount is marked as 'shared and slave'. | |
1509 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1510 | * |
1511 | * if the source mount is a tree, the operations explained above is | |
1512 | * applied to each mount in the tree. | |
1513 | * Must be called without spinlocks held, since this function can sleep | |
1514 | * in allocations. | |
1515 | */ | |
0fb54e50 | 1516 | static int attach_recursive_mnt(struct mount *source_mnt, |
1a390689 | 1517 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1518 | { |
1519 | LIST_HEAD(tree_list); | |
a8d56d8e | 1520 | struct mount *dest_mnt = real_mount(path->mnt); |
1a390689 | 1521 | struct dentry *dest_dentry = path->dentry; |
315fc83e | 1522 | struct mount *child, *p; |
719f5d7f | 1523 | int err; |
b90fa9ae | 1524 | |
fc7be130 | 1525 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1526 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1527 | if (err) |
1528 | goto out; | |
1529 | } | |
a8d56d8e | 1530 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); |
719f5d7f MS |
1531 | if (err) |
1532 | goto out_cleanup_ids; | |
b90fa9ae | 1533 | |
962830df | 1534 | br_write_lock(&vfsmount_lock); |
df1a1ad2 | 1535 | |
fc7be130 | 1536 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1537 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1538 | set_mnt_shared(p); |
b90fa9ae | 1539 | } |
1a390689 | 1540 | if (parent_path) { |
0fb54e50 AV |
1541 | detach_mnt(source_mnt, parent_path); |
1542 | attach_mnt(source_mnt, path); | |
143c8c91 | 1543 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1544 | } else { |
14cf1fa8 | 1545 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); |
0fb54e50 | 1546 | commit_tree(source_mnt); |
21444403 | 1547 | } |
b90fa9ae | 1548 | |
1b8e5564 AV |
1549 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1550 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1551 | commit_tree(child); |
b90fa9ae | 1552 | } |
962830df | 1553 | br_write_unlock(&vfsmount_lock); |
99b7db7b | 1554 | |
b90fa9ae | 1555 | return 0; |
719f5d7f MS |
1556 | |
1557 | out_cleanup_ids: | |
fc7be130 | 1558 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1559 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1560 | out: |
1561 | return err; | |
b90fa9ae RP |
1562 | } |
1563 | ||
b12cea91 AV |
1564 | static int lock_mount(struct path *path) |
1565 | { | |
1566 | struct vfsmount *mnt; | |
1567 | retry: | |
1568 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1569 | if (unlikely(cant_mount(path->dentry))) { | |
1570 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1571 | return -ENOENT; | |
1572 | } | |
1573 | down_write(&namespace_sem); | |
1574 | mnt = lookup_mnt(path); | |
1575 | if (likely(!mnt)) | |
1576 | return 0; | |
1577 | up_write(&namespace_sem); | |
1578 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1579 | path_put(path); | |
1580 | path->mnt = mnt; | |
1581 | path->dentry = dget(mnt->mnt_root); | |
1582 | goto retry; | |
1583 | } | |
1584 | ||
1585 | static void unlock_mount(struct path *path) | |
1586 | { | |
1587 | up_write(&namespace_sem); | |
1588 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1589 | } | |
1590 | ||
95bc5f25 | 1591 | static int graft_tree(struct mount *mnt, struct path *path) |
1da177e4 | 1592 | { |
95bc5f25 | 1593 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1594 | return -EINVAL; |
1595 | ||
8c3ee42e | 1596 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
95bc5f25 | 1597 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1598 | return -ENOTDIR; |
1599 | ||
b12cea91 AV |
1600 | if (d_unlinked(path->dentry)) |
1601 | return -ENOENT; | |
1da177e4 | 1602 | |
95bc5f25 | 1603 | return attach_recursive_mnt(mnt, path, NULL); |
1da177e4 LT |
1604 | } |
1605 | ||
7a2e8a8f VA |
1606 | /* |
1607 | * Sanity check the flags to change_mnt_propagation. | |
1608 | */ | |
1609 | ||
1610 | static int flags_to_propagation_type(int flags) | |
1611 | { | |
7c6e984d | 1612 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1613 | |
1614 | /* Fail if any non-propagation flags are set */ | |
1615 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1616 | return 0; | |
1617 | /* Only one propagation flag should be set */ | |
1618 | if (!is_power_of_2(type)) | |
1619 | return 0; | |
1620 | return type; | |
1621 | } | |
1622 | ||
07b20889 RP |
1623 | /* |
1624 | * recursively change the type of the mountpoint. | |
1625 | */ | |
0a0d8a46 | 1626 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1627 | { |
315fc83e | 1628 | struct mount *m; |
4b8b21f4 | 1629 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1630 | int recurse = flag & MS_REC; |
7a2e8a8f | 1631 | int type; |
719f5d7f | 1632 | int err = 0; |
07b20889 | 1633 | |
ee6f9582 MS |
1634 | if (!capable(CAP_SYS_ADMIN)) |
1635 | return -EPERM; | |
1636 | ||
2d92ab3c | 1637 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1638 | return -EINVAL; |
1639 | ||
7a2e8a8f VA |
1640 | type = flags_to_propagation_type(flag); |
1641 | if (!type) | |
1642 | return -EINVAL; | |
1643 | ||
07b20889 | 1644 | down_write(&namespace_sem); |
719f5d7f MS |
1645 | if (type == MS_SHARED) { |
1646 | err = invent_group_ids(mnt, recurse); | |
1647 | if (err) | |
1648 | goto out_unlock; | |
1649 | } | |
1650 | ||
962830df | 1651 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1652 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1653 | change_mnt_propagation(m, type); |
962830df | 1654 | br_write_unlock(&vfsmount_lock); |
719f5d7f MS |
1655 | |
1656 | out_unlock: | |
07b20889 | 1657 | up_write(&namespace_sem); |
719f5d7f | 1658 | return err; |
07b20889 RP |
1659 | } |
1660 | ||
1da177e4 LT |
1661 | /* |
1662 | * do loopback mount. | |
1663 | */ | |
808d4e3c | 1664 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1665 | int recurse) |
1da177e4 | 1666 | { |
b12cea91 | 1667 | LIST_HEAD(umount_list); |
2d92ab3c | 1668 | struct path old_path; |
87129cc0 | 1669 | struct mount *mnt = NULL, *old; |
2d92ab3c | 1670 | int err = mount_is_safe(path); |
1da177e4 LT |
1671 | if (err) |
1672 | return err; | |
1673 | if (!old_name || !*old_name) | |
1674 | return -EINVAL; | |
815d405c | 1675 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1676 | if (err) |
1677 | return err; | |
1678 | ||
8823c079 EB |
1679 | err = -EINVAL; |
1680 | if (mnt_ns_loop(&old_path)) | |
1681 | goto out; | |
1682 | ||
b12cea91 AV |
1683 | err = lock_mount(path); |
1684 | if (err) | |
1685 | goto out; | |
1686 | ||
87129cc0 AV |
1687 | old = real_mount(old_path.mnt); |
1688 | ||
1da177e4 | 1689 | err = -EINVAL; |
fc7be130 | 1690 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1691 | goto out2; |
9676f0c6 | 1692 | |
143c8c91 | 1693 | if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old)) |
b12cea91 | 1694 | goto out2; |
1da177e4 | 1695 | |
ccd48bc7 | 1696 | if (recurse) |
87129cc0 | 1697 | mnt = copy_tree(old, old_path.dentry, 0); |
ccd48bc7 | 1698 | else |
87129cc0 | 1699 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1700 | |
be34d1a3 DH |
1701 | if (IS_ERR(mnt)) { |
1702 | err = PTR_ERR(mnt); | |
1703 | goto out; | |
1704 | } | |
ccd48bc7 | 1705 | |
95bc5f25 | 1706 | err = graft_tree(mnt, path); |
ccd48bc7 | 1707 | if (err) { |
962830df | 1708 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1709 | umount_tree(mnt, 0, &umount_list); |
962830df | 1710 | br_write_unlock(&vfsmount_lock); |
5b83d2c5 | 1711 | } |
b12cea91 AV |
1712 | out2: |
1713 | unlock_mount(path); | |
1714 | release_mounts(&umount_list); | |
ccd48bc7 | 1715 | out: |
2d92ab3c | 1716 | path_put(&old_path); |
1da177e4 LT |
1717 | return err; |
1718 | } | |
1719 | ||
2e4b7fcd DH |
1720 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1721 | { | |
1722 | int error = 0; | |
1723 | int readonly_request = 0; | |
1724 | ||
1725 | if (ms_flags & MS_RDONLY) | |
1726 | readonly_request = 1; | |
1727 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1728 | return 0; | |
1729 | ||
1730 | if (readonly_request) | |
83adc753 | 1731 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1732 | else |
83adc753 | 1733 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1734 | return error; |
1735 | } | |
1736 | ||
1da177e4 LT |
1737 | /* |
1738 | * change filesystem flags. dir should be a physical root of filesystem. | |
1739 | * If you've mounted a non-root directory somewhere and want to do remount | |
1740 | * on it - tough luck. | |
1741 | */ | |
0a0d8a46 | 1742 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1743 | void *data) |
1744 | { | |
1745 | int err; | |
2d92ab3c | 1746 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1747 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 LT |
1748 | |
1749 | if (!capable(CAP_SYS_ADMIN)) | |
1750 | return -EPERM; | |
1751 | ||
143c8c91 | 1752 | if (!check_mnt(mnt)) |
1da177e4 LT |
1753 | return -EINVAL; |
1754 | ||
2d92ab3c | 1755 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1756 | return -EINVAL; |
1757 | ||
ff36fe2c EP |
1758 | err = security_sb_remount(sb, data); |
1759 | if (err) | |
1760 | return err; | |
1761 | ||
1da177e4 | 1762 | down_write(&sb->s_umount); |
2e4b7fcd | 1763 | if (flags & MS_BIND) |
2d92ab3c | 1764 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1765 | else |
2e4b7fcd | 1766 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1767 | if (!err) { |
962830df | 1768 | br_write_lock(&vfsmount_lock); |
143c8c91 AV |
1769 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1770 | mnt->mnt.mnt_flags = mnt_flags; | |
962830df | 1771 | br_write_unlock(&vfsmount_lock); |
7b43a79f | 1772 | } |
1da177e4 | 1773 | up_write(&sb->s_umount); |
0e55a7cc | 1774 | if (!err) { |
962830df | 1775 | br_write_lock(&vfsmount_lock); |
143c8c91 | 1776 | touch_mnt_namespace(mnt->mnt_ns); |
962830df | 1777 | br_write_unlock(&vfsmount_lock); |
0e55a7cc | 1778 | } |
1da177e4 LT |
1779 | return err; |
1780 | } | |
1781 | ||
cbbe362c | 1782 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1783 | { |
315fc83e | 1784 | struct mount *p; |
909b0a88 | 1785 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1786 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1787 | return 1; |
1788 | } | |
1789 | return 0; | |
1790 | } | |
1791 | ||
808d4e3c | 1792 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1793 | { |
2d92ab3c | 1794 | struct path old_path, parent_path; |
676da58d | 1795 | struct mount *p; |
0fb54e50 | 1796 | struct mount *old; |
1da177e4 LT |
1797 | int err = 0; |
1798 | if (!capable(CAP_SYS_ADMIN)) | |
1799 | return -EPERM; | |
1800 | if (!old_name || !*old_name) | |
1801 | return -EINVAL; | |
2d92ab3c | 1802 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1803 | if (err) |
1804 | return err; | |
1805 | ||
b12cea91 | 1806 | err = lock_mount(path); |
cc53ce53 DH |
1807 | if (err < 0) |
1808 | goto out; | |
1809 | ||
143c8c91 | 1810 | old = real_mount(old_path.mnt); |
fc7be130 | 1811 | p = real_mount(path->mnt); |
143c8c91 | 1812 | |
1da177e4 | 1813 | err = -EINVAL; |
fc7be130 | 1814 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1815 | goto out1; |
1816 | ||
f3da392e | 1817 | if (d_unlinked(path->dentry)) |
21444403 | 1818 | goto out1; |
1da177e4 LT |
1819 | |
1820 | err = -EINVAL; | |
2d92ab3c | 1821 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1822 | goto out1; |
1da177e4 | 1823 | |
676da58d | 1824 | if (!mnt_has_parent(old)) |
21444403 | 1825 | goto out1; |
1da177e4 | 1826 | |
2d92ab3c AV |
1827 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1828 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1829 | goto out1; |
1830 | /* | |
1831 | * Don't move a mount residing in a shared parent. | |
1832 | */ | |
fc7be130 | 1833 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1834 | goto out1; |
9676f0c6 RP |
1835 | /* |
1836 | * Don't move a mount tree containing unbindable mounts to a destination | |
1837 | * mount which is shared. | |
1838 | */ | |
fc7be130 | 1839 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1840 | goto out1; |
1da177e4 | 1841 | err = -ELOOP; |
fc7be130 | 1842 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1843 | if (p == old) |
21444403 | 1844 | goto out1; |
1da177e4 | 1845 | |
0fb54e50 | 1846 | err = attach_recursive_mnt(old, path, &parent_path); |
4ac91378 | 1847 | if (err) |
21444403 | 1848 | goto out1; |
1da177e4 LT |
1849 | |
1850 | /* if the mount is moved, it should no longer be expire | |
1851 | * automatically */ | |
6776db3d | 1852 | list_del_init(&old->mnt_expire); |
1da177e4 | 1853 | out1: |
b12cea91 | 1854 | unlock_mount(path); |
1da177e4 | 1855 | out: |
1da177e4 | 1856 | if (!err) |
1a390689 | 1857 | path_put(&parent_path); |
2d92ab3c | 1858 | path_put(&old_path); |
1da177e4 LT |
1859 | return err; |
1860 | } | |
1861 | ||
9d412a43 AV |
1862 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1863 | { | |
1864 | int err; | |
1865 | const char *subtype = strchr(fstype, '.'); | |
1866 | if (subtype) { | |
1867 | subtype++; | |
1868 | err = -EINVAL; | |
1869 | if (!subtype[0]) | |
1870 | goto err; | |
1871 | } else | |
1872 | subtype = ""; | |
1873 | ||
1874 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1875 | err = -ENOMEM; | |
1876 | if (!mnt->mnt_sb->s_subtype) | |
1877 | goto err; | |
1878 | return mnt; | |
1879 | ||
1880 | err: | |
1881 | mntput(mnt); | |
1882 | return ERR_PTR(err); | |
1883 | } | |
1884 | ||
79e801a9 | 1885 | static struct vfsmount * |
9d412a43 AV |
1886 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) |
1887 | { | |
1888 | struct file_system_type *type = get_fs_type(fstype); | |
1889 | struct vfsmount *mnt; | |
1890 | if (!type) | |
1891 | return ERR_PTR(-ENODEV); | |
1892 | mnt = vfs_kern_mount(type, flags, name, data); | |
1893 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1894 | !mnt->mnt_sb->s_subtype) | |
1895 | mnt = fs_set_subtype(mnt, fstype); | |
1896 | put_filesystem(type); | |
1897 | return mnt; | |
1898 | } | |
9d412a43 AV |
1899 | |
1900 | /* | |
1901 | * add a mount into a namespace's mount tree | |
1902 | */ | |
95bc5f25 | 1903 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 AV |
1904 | { |
1905 | int err; | |
1906 | ||
1907 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1908 | ||
b12cea91 AV |
1909 | err = lock_mount(path); |
1910 | if (err) | |
1911 | return err; | |
9d412a43 AV |
1912 | |
1913 | err = -EINVAL; | |
156cacb1 AV |
1914 | if (unlikely(!check_mnt(real_mount(path->mnt)))) { |
1915 | /* that's acceptable only for automounts done in private ns */ | |
1916 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
1917 | goto unlock; | |
1918 | /* ... and for those we'd better have mountpoint still alive */ | |
1919 | if (!real_mount(path->mnt)->mnt_ns) | |
1920 | goto unlock; | |
1921 | } | |
9d412a43 AV |
1922 | |
1923 | /* Refuse the same filesystem on the same mount point */ | |
1924 | err = -EBUSY; | |
95bc5f25 | 1925 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1926 | path->mnt->mnt_root == path->dentry) |
1927 | goto unlock; | |
1928 | ||
1929 | err = -EINVAL; | |
95bc5f25 | 1930 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
1931 | goto unlock; |
1932 | ||
95bc5f25 | 1933 | newmnt->mnt.mnt_flags = mnt_flags; |
9d412a43 AV |
1934 | err = graft_tree(newmnt, path); |
1935 | ||
1936 | unlock: | |
b12cea91 | 1937 | unlock_mount(path); |
9d412a43 AV |
1938 | return err; |
1939 | } | |
b1e75df4 | 1940 | |
1da177e4 LT |
1941 | /* |
1942 | * create a new mount for userspace and request it to be added into the | |
1943 | * namespace's tree | |
1944 | */ | |
808d4e3c AV |
1945 | static int do_new_mount(struct path *path, const char *type, int flags, |
1946 | int mnt_flags, const char *name, void *data) | |
1da177e4 LT |
1947 | { |
1948 | struct vfsmount *mnt; | |
15f9a3f3 | 1949 | int err; |
1da177e4 | 1950 | |
eca6f534 | 1951 | if (!type) |
1da177e4 LT |
1952 | return -EINVAL; |
1953 | ||
1954 | /* we need capabilities... */ | |
1955 | if (!capable(CAP_SYS_ADMIN)) | |
1956 | return -EPERM; | |
1957 | ||
1958 | mnt = do_kern_mount(type, flags, name, data); | |
1959 | if (IS_ERR(mnt)) | |
1960 | return PTR_ERR(mnt); | |
1961 | ||
95bc5f25 | 1962 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
1963 | if (err) |
1964 | mntput(mnt); | |
1965 | return err; | |
1da177e4 LT |
1966 | } |
1967 | ||
19a167af AV |
1968 | int finish_automount(struct vfsmount *m, struct path *path) |
1969 | { | |
6776db3d | 1970 | struct mount *mnt = real_mount(m); |
19a167af AV |
1971 | int err; |
1972 | /* The new mount record should have at least 2 refs to prevent it being | |
1973 | * expired before we get a chance to add it | |
1974 | */ | |
6776db3d | 1975 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
1976 | |
1977 | if (m->mnt_sb == path->mnt->mnt_sb && | |
1978 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
1979 | err = -ELOOP; |
1980 | goto fail; | |
19a167af AV |
1981 | } |
1982 | ||
95bc5f25 | 1983 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
1984 | if (!err) |
1985 | return 0; | |
1986 | fail: | |
1987 | /* remove m from any expiration list it may be on */ | |
6776db3d | 1988 | if (!list_empty(&mnt->mnt_expire)) { |
b1e75df4 | 1989 | down_write(&namespace_sem); |
962830df | 1990 | br_write_lock(&vfsmount_lock); |
6776db3d | 1991 | list_del_init(&mnt->mnt_expire); |
962830df | 1992 | br_write_unlock(&vfsmount_lock); |
b1e75df4 | 1993 | up_write(&namespace_sem); |
19a167af | 1994 | } |
b1e75df4 AV |
1995 | mntput(m); |
1996 | mntput(m); | |
19a167af AV |
1997 | return err; |
1998 | } | |
1999 | ||
ea5b778a DH |
2000 | /** |
2001 | * mnt_set_expiry - Put a mount on an expiration list | |
2002 | * @mnt: The mount to list. | |
2003 | * @expiry_list: The list to add the mount to. | |
2004 | */ | |
2005 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2006 | { | |
2007 | down_write(&namespace_sem); | |
962830df | 2008 | br_write_lock(&vfsmount_lock); |
ea5b778a | 2009 | |
6776db3d | 2010 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2011 | |
962830df | 2012 | br_write_unlock(&vfsmount_lock); |
ea5b778a DH |
2013 | up_write(&namespace_sem); |
2014 | } | |
2015 | EXPORT_SYMBOL(mnt_set_expiry); | |
2016 | ||
1da177e4 LT |
2017 | /* |
2018 | * process a list of expirable mountpoints with the intent of discarding any | |
2019 | * mountpoints that aren't in use and haven't been touched since last we came | |
2020 | * here | |
2021 | */ | |
2022 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2023 | { | |
761d5c38 | 2024 | struct mount *mnt, *next; |
1da177e4 | 2025 | LIST_HEAD(graveyard); |
bcc5c7d2 | 2026 | LIST_HEAD(umounts); |
1da177e4 LT |
2027 | |
2028 | if (list_empty(mounts)) | |
2029 | return; | |
2030 | ||
bcc5c7d2 | 2031 | down_write(&namespace_sem); |
962830df | 2032 | br_write_lock(&vfsmount_lock); |
1da177e4 LT |
2033 | |
2034 | /* extract from the expiration list every vfsmount that matches the | |
2035 | * following criteria: | |
2036 | * - only referenced by its parent vfsmount | |
2037 | * - still marked for expiry (marked on the last call here; marks are | |
2038 | * cleared by mntput()) | |
2039 | */ | |
6776db3d | 2040 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2041 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2042 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2043 | continue; |
6776db3d | 2044 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2045 | } |
bcc5c7d2 | 2046 | while (!list_empty(&graveyard)) { |
6776db3d | 2047 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2048 | touch_mnt_namespace(mnt->mnt_ns); |
bcc5c7d2 AV |
2049 | umount_tree(mnt, 1, &umounts); |
2050 | } | |
962830df | 2051 | br_write_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
2052 | up_write(&namespace_sem); |
2053 | ||
2054 | release_mounts(&umounts); | |
5528f911 TM |
2055 | } |
2056 | ||
2057 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2058 | ||
2059 | /* | |
2060 | * Ripoff of 'select_parent()' | |
2061 | * | |
2062 | * search the list of submounts for a given mountpoint, and move any | |
2063 | * shrinkable submounts to the 'graveyard' list. | |
2064 | */ | |
692afc31 | 2065 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2066 | { |
692afc31 | 2067 | struct mount *this_parent = parent; |
5528f911 TM |
2068 | struct list_head *next; |
2069 | int found = 0; | |
2070 | ||
2071 | repeat: | |
6b41d536 | 2072 | next = this_parent->mnt_mounts.next; |
5528f911 | 2073 | resume: |
6b41d536 | 2074 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2075 | struct list_head *tmp = next; |
6b41d536 | 2076 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2077 | |
2078 | next = tmp->next; | |
692afc31 | 2079 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2080 | continue; |
5528f911 TM |
2081 | /* |
2082 | * Descend a level if the d_mounts list is non-empty. | |
2083 | */ | |
6b41d536 | 2084 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2085 | this_parent = mnt; |
2086 | goto repeat; | |
2087 | } | |
1da177e4 | 2088 | |
1ab59738 | 2089 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2090 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2091 | found++; |
2092 | } | |
1da177e4 | 2093 | } |
5528f911 TM |
2094 | /* |
2095 | * All done at this level ... ascend and resume the search | |
2096 | */ | |
2097 | if (this_parent != parent) { | |
6b41d536 | 2098 | next = this_parent->mnt_child.next; |
0714a533 | 2099 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2100 | goto resume; |
2101 | } | |
2102 | return found; | |
2103 | } | |
2104 | ||
2105 | /* | |
2106 | * process a list of expirable mountpoints with the intent of discarding any | |
2107 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
2108 | * |
2109 | * vfsmount_lock must be held for write | |
5528f911 | 2110 | */ |
692afc31 | 2111 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts) |
5528f911 TM |
2112 | { |
2113 | LIST_HEAD(graveyard); | |
761d5c38 | 2114 | struct mount *m; |
5528f911 | 2115 | |
5528f911 | 2116 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2117 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2118 | while (!list_empty(&graveyard)) { |
761d5c38 | 2119 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2120 | mnt_expire); |
143c8c91 | 2121 | touch_mnt_namespace(m->mnt_ns); |
afef80b3 | 2122 | umount_tree(m, 1, umounts); |
bcc5c7d2 AV |
2123 | } |
2124 | } | |
1da177e4 LT |
2125 | } |
2126 | ||
1da177e4 LT |
2127 | /* |
2128 | * Some copy_from_user() implementations do not return the exact number of | |
2129 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2130 | * Note that this function differs from copy_from_user() in that it will oops | |
2131 | * on bad values of `to', rather than returning a short copy. | |
2132 | */ | |
b58fed8b RP |
2133 | static long exact_copy_from_user(void *to, const void __user * from, |
2134 | unsigned long n) | |
1da177e4 LT |
2135 | { |
2136 | char *t = to; | |
2137 | const char __user *f = from; | |
2138 | char c; | |
2139 | ||
2140 | if (!access_ok(VERIFY_READ, from, n)) | |
2141 | return n; | |
2142 | ||
2143 | while (n) { | |
2144 | if (__get_user(c, f)) { | |
2145 | memset(t, 0, n); | |
2146 | break; | |
2147 | } | |
2148 | *t++ = c; | |
2149 | f++; | |
2150 | n--; | |
2151 | } | |
2152 | return n; | |
2153 | } | |
2154 | ||
b58fed8b | 2155 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2156 | { |
2157 | int i; | |
2158 | unsigned long page; | |
2159 | unsigned long size; | |
b58fed8b | 2160 | |
1da177e4 LT |
2161 | *where = 0; |
2162 | if (!data) | |
2163 | return 0; | |
2164 | ||
2165 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2166 | return -ENOMEM; | |
2167 | ||
2168 | /* We only care that *some* data at the address the user | |
2169 | * gave us is valid. Just in case, we'll zero | |
2170 | * the remainder of the page. | |
2171 | */ | |
2172 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2173 | size = TASK_SIZE - (unsigned long)data; | |
2174 | if (size > PAGE_SIZE) | |
2175 | size = PAGE_SIZE; | |
2176 | ||
2177 | i = size - exact_copy_from_user((void *)page, data, size); | |
2178 | if (!i) { | |
b58fed8b | 2179 | free_page(page); |
1da177e4 LT |
2180 | return -EFAULT; |
2181 | } | |
2182 | if (i != PAGE_SIZE) | |
2183 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2184 | *where = page; | |
2185 | return 0; | |
2186 | } | |
2187 | ||
eca6f534 VN |
2188 | int copy_mount_string(const void __user *data, char **where) |
2189 | { | |
2190 | char *tmp; | |
2191 | ||
2192 | if (!data) { | |
2193 | *where = NULL; | |
2194 | return 0; | |
2195 | } | |
2196 | ||
2197 | tmp = strndup_user(data, PAGE_SIZE); | |
2198 | if (IS_ERR(tmp)) | |
2199 | return PTR_ERR(tmp); | |
2200 | ||
2201 | *where = tmp; | |
2202 | return 0; | |
2203 | } | |
2204 | ||
1da177e4 LT |
2205 | /* |
2206 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2207 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2208 | * | |
2209 | * data is a (void *) that can point to any structure up to | |
2210 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2211 | * information (or be NULL). | |
2212 | * | |
2213 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2214 | * When the flags word was introduced its top half was required | |
2215 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2216 | * Therefore, if this magic number is present, it carries no information | |
2217 | * and must be discarded. | |
2218 | */ | |
808d4e3c AV |
2219 | long do_mount(const char *dev_name, const char *dir_name, |
2220 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2221 | { |
2d92ab3c | 2222 | struct path path; |
1da177e4 LT |
2223 | int retval = 0; |
2224 | int mnt_flags = 0; | |
2225 | ||
2226 | /* Discard magic */ | |
2227 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2228 | flags &= ~MS_MGC_MSK; | |
2229 | ||
2230 | /* Basic sanity checks */ | |
2231 | ||
2232 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2233 | return -EINVAL; | |
1da177e4 LT |
2234 | |
2235 | if (data_page) | |
2236 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2237 | ||
a27ab9f2 TH |
2238 | /* ... and get the mountpoint */ |
2239 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2240 | if (retval) | |
2241 | return retval; | |
2242 | ||
2243 | retval = security_sb_mount(dev_name, &path, | |
2244 | type_page, flags, data_page); | |
2245 | if (retval) | |
2246 | goto dput_out; | |
2247 | ||
613cbe3d AK |
2248 | /* Default to relatime unless overriden */ |
2249 | if (!(flags & MS_NOATIME)) | |
2250 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2251 | |
1da177e4 LT |
2252 | /* Separate the per-mountpoint flags */ |
2253 | if (flags & MS_NOSUID) | |
2254 | mnt_flags |= MNT_NOSUID; | |
2255 | if (flags & MS_NODEV) | |
2256 | mnt_flags |= MNT_NODEV; | |
2257 | if (flags & MS_NOEXEC) | |
2258 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2259 | if (flags & MS_NOATIME) |
2260 | mnt_flags |= MNT_NOATIME; | |
2261 | if (flags & MS_NODIRATIME) | |
2262 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2263 | if (flags & MS_STRICTATIME) |
2264 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2265 | if (flags & MS_RDONLY) |
2266 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2267 | |
7a4dec53 | 2268 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2269 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2270 | MS_STRICTATIME); | |
1da177e4 | 2271 | |
1da177e4 | 2272 | if (flags & MS_REMOUNT) |
2d92ab3c | 2273 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2274 | data_page); |
2275 | else if (flags & MS_BIND) | |
2d92ab3c | 2276 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2277 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2278 | retval = do_change_type(&path, flags); |
1da177e4 | 2279 | else if (flags & MS_MOVE) |
2d92ab3c | 2280 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2281 | else |
2d92ab3c | 2282 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2283 | dev_name, data_page); |
2284 | dput_out: | |
2d92ab3c | 2285 | path_put(&path); |
1da177e4 LT |
2286 | return retval; |
2287 | } | |
2288 | ||
8823c079 EB |
2289 | /* |
2290 | * Assign a sequence number so we can detect when we attempt to bind | |
2291 | * mount a reference to an older mount namespace into the current | |
2292 | * mount namespace, preventing reference counting loops. A 64bit | |
2293 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2294 | * is effectively never, so we can ignore the possibility. | |
2295 | */ | |
2296 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2297 | ||
cf8d2c11 TM |
2298 | static struct mnt_namespace *alloc_mnt_ns(void) |
2299 | { | |
2300 | struct mnt_namespace *new_ns; | |
2301 | ||
2302 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2303 | if (!new_ns) | |
2304 | return ERR_PTR(-ENOMEM); | |
8823c079 | 2305 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2306 | atomic_set(&new_ns->count, 1); |
2307 | new_ns->root = NULL; | |
2308 | INIT_LIST_HEAD(&new_ns->list); | |
2309 | init_waitqueue_head(&new_ns->poll); | |
2310 | new_ns->event = 0; | |
2311 | return new_ns; | |
2312 | } | |
2313 | ||
741a2951 JD |
2314 | /* |
2315 | * Allocate a new namespace structure and populate it with contents | |
2316 | * copied from the namespace of the passed in task structure. | |
2317 | */ | |
e3222c4e | 2318 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
6b3286ed | 2319 | struct fs_struct *fs) |
1da177e4 | 2320 | { |
6b3286ed | 2321 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2322 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2323 | struct mount *p, *q; |
be08d6d2 | 2324 | struct mount *old = mnt_ns->root; |
cb338d06 | 2325 | struct mount *new; |
1da177e4 | 2326 | |
cf8d2c11 TM |
2327 | new_ns = alloc_mnt_ns(); |
2328 | if (IS_ERR(new_ns)) | |
2329 | return new_ns; | |
1da177e4 | 2330 | |
390c6843 | 2331 | down_write(&namespace_sem); |
1da177e4 | 2332 | /* First pass: copy the tree topology */ |
909b0a88 | 2333 | new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE); |
be34d1a3 | 2334 | if (IS_ERR(new)) { |
390c6843 | 2335 | up_write(&namespace_sem); |
1da177e4 | 2336 | kfree(new_ns); |
be34d1a3 | 2337 | return ERR_CAST(new); |
1da177e4 | 2338 | } |
be08d6d2 | 2339 | new_ns->root = new; |
962830df | 2340 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 2341 | list_add_tail(&new_ns->list, &new->mnt_list); |
962830df | 2342 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
2343 | |
2344 | /* | |
2345 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2346 | * as belonging to new namespace. We have already acquired a private | |
2347 | * fs_struct, so tsk->fs->lock is not needed. | |
2348 | */ | |
909b0a88 | 2349 | p = old; |
cb338d06 | 2350 | q = new; |
1da177e4 | 2351 | while (p) { |
143c8c91 | 2352 | q->mnt_ns = new_ns; |
1da177e4 | 2353 | if (fs) { |
315fc83e AV |
2354 | if (&p->mnt == fs->root.mnt) { |
2355 | fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2356 | rootmnt = &p->mnt; |
1da177e4 | 2357 | } |
315fc83e AV |
2358 | if (&p->mnt == fs->pwd.mnt) { |
2359 | fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2360 | pwdmnt = &p->mnt; |
1da177e4 | 2361 | } |
1da177e4 | 2362 | } |
909b0a88 AV |
2363 | p = next_mnt(p, old); |
2364 | q = next_mnt(q, new); | |
1da177e4 | 2365 | } |
390c6843 | 2366 | up_write(&namespace_sem); |
1da177e4 | 2367 | |
1da177e4 | 2368 | if (rootmnt) |
f03c6599 | 2369 | mntput(rootmnt); |
1da177e4 | 2370 | if (pwdmnt) |
f03c6599 | 2371 | mntput(pwdmnt); |
1da177e4 | 2372 | |
741a2951 JD |
2373 | return new_ns; |
2374 | } | |
2375 | ||
213dd266 | 2376 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
e3222c4e | 2377 | struct fs_struct *new_fs) |
741a2951 | 2378 | { |
6b3286ed | 2379 | struct mnt_namespace *new_ns; |
741a2951 | 2380 | |
e3222c4e | 2381 | BUG_ON(!ns); |
6b3286ed | 2382 | get_mnt_ns(ns); |
741a2951 JD |
2383 | |
2384 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2385 | return ns; |
741a2951 | 2386 | |
e3222c4e | 2387 | new_ns = dup_mnt_ns(ns, new_fs); |
741a2951 | 2388 | |
6b3286ed | 2389 | put_mnt_ns(ns); |
e3222c4e | 2390 | return new_ns; |
1da177e4 LT |
2391 | } |
2392 | ||
cf8d2c11 TM |
2393 | /** |
2394 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2395 | * @mnt: pointer to the new root filesystem mountpoint | |
2396 | */ | |
1a4eeaf2 | 2397 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2398 | { |
1a4eeaf2 | 2399 | struct mnt_namespace *new_ns = alloc_mnt_ns(); |
cf8d2c11 | 2400 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2401 | struct mount *mnt = real_mount(m); |
2402 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2403 | new_ns->root = mnt; |
1a4eeaf2 | 2404 | list_add(&new_ns->list, &mnt->mnt_list); |
c1334495 | 2405 | } else { |
1a4eeaf2 | 2406 | mntput(m); |
cf8d2c11 TM |
2407 | } |
2408 | return new_ns; | |
2409 | } | |
cf8d2c11 | 2410 | |
ea441d11 AV |
2411 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2412 | { | |
2413 | struct mnt_namespace *ns; | |
d31da0f0 | 2414 | struct super_block *s; |
ea441d11 AV |
2415 | struct path path; |
2416 | int err; | |
2417 | ||
2418 | ns = create_mnt_ns(mnt); | |
2419 | if (IS_ERR(ns)) | |
2420 | return ERR_CAST(ns); | |
2421 | ||
2422 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2423 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2424 | ||
2425 | put_mnt_ns(ns); | |
2426 | ||
2427 | if (err) | |
2428 | return ERR_PTR(err); | |
2429 | ||
2430 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2431 | s = path.mnt->mnt_sb; |
2432 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2433 | mntput(path.mnt); |
2434 | /* lock the sucker */ | |
d31da0f0 | 2435 | down_write(&s->s_umount); |
ea441d11 AV |
2436 | /* ... and return the root of (sub)tree on it */ |
2437 | return path.dentry; | |
2438 | } | |
2439 | EXPORT_SYMBOL(mount_subtree); | |
2440 | ||
bdc480e3 HC |
2441 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2442 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2443 | { |
eca6f534 VN |
2444 | int ret; |
2445 | char *kernel_type; | |
91a27b2a | 2446 | struct filename *kernel_dir; |
eca6f534 | 2447 | char *kernel_dev; |
1da177e4 | 2448 | unsigned long data_page; |
1da177e4 | 2449 | |
eca6f534 VN |
2450 | ret = copy_mount_string(type, &kernel_type); |
2451 | if (ret < 0) | |
2452 | goto out_type; | |
1da177e4 | 2453 | |
eca6f534 VN |
2454 | kernel_dir = getname(dir_name); |
2455 | if (IS_ERR(kernel_dir)) { | |
2456 | ret = PTR_ERR(kernel_dir); | |
2457 | goto out_dir; | |
2458 | } | |
1da177e4 | 2459 | |
eca6f534 VN |
2460 | ret = copy_mount_string(dev_name, &kernel_dev); |
2461 | if (ret < 0) | |
2462 | goto out_dev; | |
1da177e4 | 2463 | |
eca6f534 VN |
2464 | ret = copy_mount_options(data, &data_page); |
2465 | if (ret < 0) | |
2466 | goto out_data; | |
1da177e4 | 2467 | |
91a27b2a | 2468 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2469 | (void *) data_page); |
1da177e4 | 2470 | |
eca6f534 VN |
2471 | free_page(data_page); |
2472 | out_data: | |
2473 | kfree(kernel_dev); | |
2474 | out_dev: | |
2475 | putname(kernel_dir); | |
2476 | out_dir: | |
2477 | kfree(kernel_type); | |
2478 | out_type: | |
2479 | return ret; | |
1da177e4 LT |
2480 | } |
2481 | ||
afac7cba AV |
2482 | /* |
2483 | * Return true if path is reachable from root | |
2484 | * | |
2485 | * namespace_sem or vfsmount_lock is held | |
2486 | */ | |
643822b4 | 2487 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2488 | const struct path *root) |
2489 | { | |
643822b4 | 2490 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2491 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2492 | mnt = mnt->mnt_parent; |
afac7cba | 2493 | } |
643822b4 | 2494 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2495 | } |
2496 | ||
2497 | int path_is_under(struct path *path1, struct path *path2) | |
2498 | { | |
2499 | int res; | |
962830df | 2500 | br_read_lock(&vfsmount_lock); |
643822b4 | 2501 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
962830df | 2502 | br_read_unlock(&vfsmount_lock); |
afac7cba AV |
2503 | return res; |
2504 | } | |
2505 | EXPORT_SYMBOL(path_is_under); | |
2506 | ||
1da177e4 LT |
2507 | /* |
2508 | * pivot_root Semantics: | |
2509 | * Moves the root file system of the current process to the directory put_old, | |
2510 | * makes new_root as the new root file system of the current process, and sets | |
2511 | * root/cwd of all processes which had them on the current root to new_root. | |
2512 | * | |
2513 | * Restrictions: | |
2514 | * The new_root and put_old must be directories, and must not be on the | |
2515 | * same file system as the current process root. The put_old must be | |
2516 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2517 | * pointed to by put_old must yield the same directory as new_root. No other | |
2518 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2519 | * | |
4a0d11fa NB |
2520 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2521 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2522 | * in this situation. | |
2523 | * | |
1da177e4 LT |
2524 | * Notes: |
2525 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2526 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2527 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2528 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2529 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2530 | * first. | |
2531 | */ | |
3480b257 HC |
2532 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2533 | const char __user *, put_old) | |
1da177e4 | 2534 | { |
2d8f3038 | 2535 | struct path new, old, parent_path, root_parent, root; |
419148da | 2536 | struct mount *new_mnt, *root_mnt; |
1da177e4 LT |
2537 | int error; |
2538 | ||
2539 | if (!capable(CAP_SYS_ADMIN)) | |
2540 | return -EPERM; | |
2541 | ||
2d8f3038 | 2542 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2543 | if (error) |
2544 | goto out0; | |
1da177e4 | 2545 | |
2d8f3038 | 2546 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2547 | if (error) |
2548 | goto out1; | |
2549 | ||
2d8f3038 | 2550 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2551 | if (error) |
2552 | goto out2; | |
1da177e4 | 2553 | |
f7ad3c6b | 2554 | get_fs_root(current->fs, &root); |
b12cea91 AV |
2555 | error = lock_mount(&old); |
2556 | if (error) | |
2557 | goto out3; | |
2558 | ||
1da177e4 | 2559 | error = -EINVAL; |
419148da AV |
2560 | new_mnt = real_mount(new.mnt); |
2561 | root_mnt = real_mount(root.mnt); | |
fc7be130 AV |
2562 | if (IS_MNT_SHARED(real_mount(old.mnt)) || |
2563 | IS_MNT_SHARED(new_mnt->mnt_parent) || | |
2564 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2565 | goto out4; |
143c8c91 | 2566 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2567 | goto out4; |
1da177e4 | 2568 | error = -ENOENT; |
f3da392e | 2569 | if (d_unlinked(new.dentry)) |
b12cea91 | 2570 | goto out4; |
f3da392e | 2571 | if (d_unlinked(old.dentry)) |
b12cea91 | 2572 | goto out4; |
1da177e4 | 2573 | error = -EBUSY; |
2d8f3038 AV |
2574 | if (new.mnt == root.mnt || |
2575 | old.mnt == root.mnt) | |
b12cea91 | 2576 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2577 | error = -EINVAL; |
8c3ee42e | 2578 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2579 | goto out4; /* not a mountpoint */ |
676da58d | 2580 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2581 | goto out4; /* not attached */ |
2d8f3038 | 2582 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2583 | goto out4; /* not a mountpoint */ |
676da58d | 2584 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2585 | goto out4; /* not attached */ |
4ac91378 | 2586 | /* make sure we can reach put_old from new_root */ |
643822b4 | 2587 | if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new)) |
b12cea91 | 2588 | goto out4; |
962830df | 2589 | br_write_lock(&vfsmount_lock); |
419148da AV |
2590 | detach_mnt(new_mnt, &parent_path); |
2591 | detach_mnt(root_mnt, &root_parent); | |
4ac91378 | 2592 | /* mount old root on put_old */ |
419148da | 2593 | attach_mnt(root_mnt, &old); |
4ac91378 | 2594 | /* mount new_root on / */ |
419148da | 2595 | attach_mnt(new_mnt, &root_parent); |
6b3286ed | 2596 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
962830df | 2597 | br_write_unlock(&vfsmount_lock); |
2d8f3038 | 2598 | chroot_fs_refs(&root, &new); |
1da177e4 | 2599 | error = 0; |
b12cea91 AV |
2600 | out4: |
2601 | unlock_mount(&old); | |
2602 | if (!error) { | |
2603 | path_put(&root_parent); | |
2604 | path_put(&parent_path); | |
2605 | } | |
2606 | out3: | |
8c3ee42e | 2607 | path_put(&root); |
b12cea91 | 2608 | out2: |
2d8f3038 | 2609 | path_put(&old); |
1da177e4 | 2610 | out1: |
2d8f3038 | 2611 | path_put(&new); |
1da177e4 | 2612 | out0: |
1da177e4 | 2613 | return error; |
1da177e4 LT |
2614 | } |
2615 | ||
2616 | static void __init init_mount_tree(void) | |
2617 | { | |
2618 | struct vfsmount *mnt; | |
6b3286ed | 2619 | struct mnt_namespace *ns; |
ac748a09 | 2620 | struct path root; |
1da177e4 LT |
2621 | |
2622 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2623 | if (IS_ERR(mnt)) | |
2624 | panic("Can't create rootfs"); | |
b3e19d92 | 2625 | |
3b22edc5 TM |
2626 | ns = create_mnt_ns(mnt); |
2627 | if (IS_ERR(ns)) | |
1da177e4 | 2628 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2629 | |
2630 | init_task.nsproxy->mnt_ns = ns; | |
2631 | get_mnt_ns(ns); | |
2632 | ||
be08d6d2 AV |
2633 | root.mnt = mnt; |
2634 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2635 | |
2636 | set_fs_pwd(current->fs, &root); | |
2637 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2638 | } |
2639 | ||
74bf17cf | 2640 | void __init mnt_init(void) |
1da177e4 | 2641 | { |
13f14b4d | 2642 | unsigned u; |
15a67dd8 | 2643 | int err; |
1da177e4 | 2644 | |
390c6843 RP |
2645 | init_rwsem(&namespace_sem); |
2646 | ||
7d6fec45 | 2647 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2648 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2649 | |
b58fed8b | 2650 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2651 | |
2652 | if (!mount_hashtable) | |
2653 | panic("Failed to allocate mount hash table\n"); | |
2654 | ||
80cdc6da | 2655 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2656 | |
2657 | for (u = 0; u < HASH_SIZE; u++) | |
2658 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2659 | |
962830df | 2660 | br_lock_init(&vfsmount_lock); |
99b7db7b | 2661 | |
15a67dd8 RD |
2662 | err = sysfs_init(); |
2663 | if (err) | |
2664 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2665 | __func__, err); |
00d26666 GKH |
2666 | fs_kobj = kobject_create_and_add("fs", NULL); |
2667 | if (!fs_kobj) | |
8e24eea7 | 2668 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2669 | init_rootfs(); |
2670 | init_mount_tree(); | |
2671 | } | |
2672 | ||
616511d0 | 2673 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2674 | { |
70fbcdf4 | 2675 | LIST_HEAD(umount_list); |
616511d0 | 2676 | |
d498b25a | 2677 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2678 | return; |
390c6843 | 2679 | down_write(&namespace_sem); |
962830df | 2680 | br_write_lock(&vfsmount_lock); |
be08d6d2 | 2681 | umount_tree(ns->root, 0, &umount_list); |
962830df | 2682 | br_write_unlock(&vfsmount_lock); |
390c6843 | 2683 | up_write(&namespace_sem); |
70fbcdf4 | 2684 | release_mounts(&umount_list); |
6b3286ed | 2685 | kfree(ns); |
1da177e4 | 2686 | } |
9d412a43 AV |
2687 | |
2688 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2689 | { | |
423e0ab0 TC |
2690 | struct vfsmount *mnt; |
2691 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2692 | if (!IS_ERR(mnt)) { | |
2693 | /* | |
2694 | * it is a longterm mount, don't release mnt until | |
2695 | * we unmount before file sys is unregistered | |
2696 | */ | |
f7a99c5b | 2697 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2698 | } |
2699 | return mnt; | |
9d412a43 AV |
2700 | } |
2701 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2702 | |
2703 | void kern_unmount(struct vfsmount *mnt) | |
2704 | { | |
2705 | /* release long term mount so mount point can be released */ | |
2706 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b AV |
2707 | br_write_lock(&vfsmount_lock); |
2708 | real_mount(mnt)->mnt_ns = NULL; | |
2709 | br_write_unlock(&vfsmount_lock); | |
423e0ab0 TC |
2710 | mntput(mnt); |
2711 | } | |
2712 | } | |
2713 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2714 | |
2715 | bool our_mnt(struct vfsmount *mnt) | |
2716 | { | |
143c8c91 | 2717 | return check_mnt(real_mount(mnt)); |
02125a82 | 2718 | } |
8823c079 EB |
2719 | |
2720 | static void *mntns_get(struct task_struct *task) | |
2721 | { | |
2722 | struct mnt_namespace *ns = NULL; | |
2723 | struct nsproxy *nsproxy; | |
2724 | ||
2725 | rcu_read_lock(); | |
2726 | nsproxy = task_nsproxy(task); | |
2727 | if (nsproxy) { | |
2728 | ns = nsproxy->mnt_ns; | |
2729 | get_mnt_ns(ns); | |
2730 | } | |
2731 | rcu_read_unlock(); | |
2732 | ||
2733 | return ns; | |
2734 | } | |
2735 | ||
2736 | static void mntns_put(void *ns) | |
2737 | { | |
2738 | put_mnt_ns(ns); | |
2739 | } | |
2740 | ||
2741 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2742 | { | |
2743 | struct fs_struct *fs = current->fs; | |
2744 | struct mnt_namespace *mnt_ns = ns; | |
2745 | struct path root; | |
2746 | ||
2747 | if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_CHROOT)) | |
2748 | return -EINVAL; | |
2749 | ||
2750 | if (fs->users != 1) | |
2751 | return -EINVAL; | |
2752 | ||
2753 | get_mnt_ns(mnt_ns); | |
2754 | put_mnt_ns(nsproxy->mnt_ns); | |
2755 | nsproxy->mnt_ns = mnt_ns; | |
2756 | ||
2757 | /* Find the root */ | |
2758 | root.mnt = &mnt_ns->root->mnt; | |
2759 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2760 | path_get(&root); | |
2761 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
2762 | ; | |
2763 | ||
2764 | /* Update the pwd and root */ | |
2765 | set_fs_pwd(fs, &root); | |
2766 | set_fs_root(fs, &root); | |
2767 | ||
2768 | path_put(&root); | |
2769 | return 0; | |
2770 | } | |
2771 | ||
2772 | const struct proc_ns_operations mntns_operations = { | |
2773 | .name = "mnt", | |
2774 | .type = CLONE_NEWNS, | |
2775 | .get = mntns_get, | |
2776 | .put = mntns_put, | |
2777 | .install = mntns_install, | |
2778 | }; |