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> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
07b20889 | 26 | #include "pnode.h" |
948730b0 | 27 | #include "internal.h" |
1da177e4 | 28 | |
0818bf27 AV |
29 | static unsigned int m_hash_mask __read_mostly; |
30 | static unsigned int m_hash_shift __read_mostly; | |
31 | static unsigned int mp_hash_mask __read_mostly; | |
32 | static unsigned int mp_hash_shift __read_mostly; | |
33 | ||
34 | static __initdata unsigned long mhash_entries; | |
35 | static int __init set_mhash_entries(char *str) | |
36 | { | |
37 | if (!str) | |
38 | return 0; | |
39 | mhash_entries = simple_strtoul(str, &str, 0); | |
40 | return 1; | |
41 | } | |
42 | __setup("mhash_entries=", set_mhash_entries); | |
43 | ||
44 | static __initdata unsigned long mphash_entries; | |
45 | static int __init set_mphash_entries(char *str) | |
46 | { | |
47 | if (!str) | |
48 | return 0; | |
49 | mphash_entries = simple_strtoul(str, &str, 0); | |
50 | return 1; | |
51 | } | |
52 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 53 | |
c7999c36 | 54 | static u64 event; |
73cd49ec | 55 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 56 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 57 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
58 | static int mnt_id_start = 0; |
59 | static int mnt_group_start = 1; | |
1da177e4 | 60 | |
38129a13 | 61 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 62 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 63 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 64 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 65 | |
f87fd4c2 | 66 | /* /sys/fs */ |
00d26666 GKH |
67 | struct kobject *fs_kobj; |
68 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 69 | |
99b7db7b NP |
70 | /* |
71 | * vfsmount lock may be taken for read to prevent changes to the | |
72 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
73 | * up the tree. | |
74 | * | |
75 | * It should be taken for write in all cases where the vfsmount | |
76 | * tree or hash is modified or when a vfsmount structure is modified. | |
77 | */ | |
48a066e7 | 78 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 79 | |
38129a13 | 80 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 81 | { |
b58fed8b RP |
82 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
83 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
84 | tmp = tmp + (tmp >> m_hash_shift); |
85 | return &mount_hashtable[tmp & m_hash_mask]; | |
86 | } | |
87 | ||
88 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
89 | { | |
90 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
91 | tmp = tmp + (tmp >> mp_hash_shift); | |
92 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
93 | } |
94 | ||
99b7db7b NP |
95 | /* |
96 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
97 | * serialize with freeing. | |
98 | */ | |
b105e270 | 99 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
100 | { |
101 | int res; | |
102 | ||
103 | retry: | |
104 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 105 | spin_lock(&mnt_id_lock); |
15169fe7 | 106 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 107 | if (!res) |
15169fe7 | 108 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 109 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
110 | if (res == -EAGAIN) |
111 | goto retry; | |
112 | ||
113 | return res; | |
114 | } | |
115 | ||
b105e270 | 116 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 117 | { |
15169fe7 | 118 | int id = mnt->mnt_id; |
99b7db7b | 119 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
120 | ida_remove(&mnt_id_ida, id); |
121 | if (mnt_id_start > id) | |
122 | mnt_id_start = id; | |
99b7db7b | 123 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
124 | } |
125 | ||
719f5d7f MS |
126 | /* |
127 | * Allocate a new peer group ID | |
128 | * | |
129 | * mnt_group_ida is protected by namespace_sem | |
130 | */ | |
4b8b21f4 | 131 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 132 | { |
f21f6220 AV |
133 | int res; |
134 | ||
719f5d7f MS |
135 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
136 | return -ENOMEM; | |
137 | ||
f21f6220 AV |
138 | res = ida_get_new_above(&mnt_group_ida, |
139 | mnt_group_start, | |
15169fe7 | 140 | &mnt->mnt_group_id); |
f21f6220 | 141 | if (!res) |
15169fe7 | 142 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
143 | |
144 | return res; | |
719f5d7f MS |
145 | } |
146 | ||
147 | /* | |
148 | * Release a peer group ID | |
149 | */ | |
4b8b21f4 | 150 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 151 | { |
15169fe7 | 152 | int id = mnt->mnt_group_id; |
f21f6220 AV |
153 | ida_remove(&mnt_group_ida, id); |
154 | if (mnt_group_start > id) | |
155 | mnt_group_start = id; | |
15169fe7 | 156 | mnt->mnt_group_id = 0; |
719f5d7f MS |
157 | } |
158 | ||
b3e19d92 NP |
159 | /* |
160 | * vfsmount lock must be held for read | |
161 | */ | |
83adc753 | 162 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
163 | { |
164 | #ifdef CONFIG_SMP | |
68e8a9fe | 165 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
166 | #else |
167 | preempt_disable(); | |
68e8a9fe | 168 | mnt->mnt_count += n; |
b3e19d92 NP |
169 | preempt_enable(); |
170 | #endif | |
171 | } | |
172 | ||
b3e19d92 NP |
173 | /* |
174 | * vfsmount lock must be held for write | |
175 | */ | |
83adc753 | 176 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
177 | { |
178 | #ifdef CONFIG_SMP | |
f03c6599 | 179 | unsigned int count = 0; |
b3e19d92 NP |
180 | int cpu; |
181 | ||
182 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 183 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
184 | } |
185 | ||
186 | return count; | |
187 | #else | |
68e8a9fe | 188 | return mnt->mnt_count; |
b3e19d92 NP |
189 | #endif |
190 | } | |
191 | ||
b105e270 | 192 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 193 | { |
c63181e6 AV |
194 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
195 | if (mnt) { | |
73cd49ec MS |
196 | int err; |
197 | ||
c63181e6 | 198 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
199 | if (err) |
200 | goto out_free_cache; | |
201 | ||
202 | if (name) { | |
c63181e6 AV |
203 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
204 | if (!mnt->mnt_devname) | |
88b38782 | 205 | goto out_free_id; |
73cd49ec MS |
206 | } |
207 | ||
b3e19d92 | 208 | #ifdef CONFIG_SMP |
c63181e6 AV |
209 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
210 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
211 | goto out_free_devname; |
212 | ||
c63181e6 | 213 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 214 | #else |
c63181e6 AV |
215 | mnt->mnt_count = 1; |
216 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
217 | #endif |
218 | ||
38129a13 | 219 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
220 | INIT_LIST_HEAD(&mnt->mnt_child); |
221 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
222 | INIT_LIST_HEAD(&mnt->mnt_list); | |
223 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
224 | INIT_LIST_HEAD(&mnt->mnt_share); | |
225 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
227 | #ifdef CONFIG_FSNOTIFY |
228 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 229 | #endif |
1da177e4 | 230 | } |
c63181e6 | 231 | return mnt; |
88b38782 | 232 | |
d3ef3d73 | 233 | #ifdef CONFIG_SMP |
234 | out_free_devname: | |
c63181e6 | 235 | kfree(mnt->mnt_devname); |
d3ef3d73 | 236 | #endif |
88b38782 | 237 | out_free_id: |
c63181e6 | 238 | mnt_free_id(mnt); |
88b38782 | 239 | out_free_cache: |
c63181e6 | 240 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 241 | return NULL; |
1da177e4 LT |
242 | } |
243 | ||
3d733633 DH |
244 | /* |
245 | * Most r/o checks on a fs are for operations that take | |
246 | * discrete amounts of time, like a write() or unlink(). | |
247 | * We must keep track of when those operations start | |
248 | * (for permission checks) and when they end, so that | |
249 | * we can determine when writes are able to occur to | |
250 | * a filesystem. | |
251 | */ | |
252 | /* | |
253 | * __mnt_is_readonly: check whether a mount is read-only | |
254 | * @mnt: the mount to check for its write status | |
255 | * | |
256 | * This shouldn't be used directly ouside of the VFS. | |
257 | * It does not guarantee that the filesystem will stay | |
258 | * r/w, just that it is right *now*. This can not and | |
259 | * should not be used in place of IS_RDONLY(inode). | |
260 | * mnt_want/drop_write() will _keep_ the filesystem | |
261 | * r/w. | |
262 | */ | |
263 | int __mnt_is_readonly(struct vfsmount *mnt) | |
264 | { | |
2e4b7fcd DH |
265 | if (mnt->mnt_flags & MNT_READONLY) |
266 | return 1; | |
267 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
268 | return 1; | |
269 | return 0; | |
3d733633 DH |
270 | } |
271 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
272 | ||
83adc753 | 273 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 274 | { |
275 | #ifdef CONFIG_SMP | |
68e8a9fe | 276 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 277 | #else |
68e8a9fe | 278 | mnt->mnt_writers++; |
d3ef3d73 | 279 | #endif |
280 | } | |
3d733633 | 281 | |
83adc753 | 282 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 283 | { |
d3ef3d73 | 284 | #ifdef CONFIG_SMP |
68e8a9fe | 285 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 286 | #else |
68e8a9fe | 287 | mnt->mnt_writers--; |
d3ef3d73 | 288 | #endif |
3d733633 | 289 | } |
3d733633 | 290 | |
83adc753 | 291 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 292 | { |
d3ef3d73 | 293 | #ifdef CONFIG_SMP |
294 | unsigned int count = 0; | |
3d733633 | 295 | int cpu; |
3d733633 DH |
296 | |
297 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 298 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 299 | } |
3d733633 | 300 | |
d3ef3d73 | 301 | return count; |
302 | #else | |
303 | return mnt->mnt_writers; | |
304 | #endif | |
3d733633 DH |
305 | } |
306 | ||
4ed5e82f MS |
307 | static int mnt_is_readonly(struct vfsmount *mnt) |
308 | { | |
309 | if (mnt->mnt_sb->s_readonly_remount) | |
310 | return 1; | |
311 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
312 | smp_rmb(); | |
313 | return __mnt_is_readonly(mnt); | |
314 | } | |
315 | ||
8366025e | 316 | /* |
eb04c282 JK |
317 | * Most r/o & frozen checks on a fs are for operations that take discrete |
318 | * amounts of time, like a write() or unlink(). We must keep track of when | |
319 | * those operations start (for permission checks) and when they end, so that we | |
320 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
321 | */ |
322 | /** | |
eb04c282 | 323 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 324 | * @m: the mount on which to take a write |
8366025e | 325 | * |
eb04c282 JK |
326 | * This tells the low-level filesystem that a write is about to be performed to |
327 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
328 | * returning success. This operation does not protect against filesystem being | |
329 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
330 | * called. This is effectively a refcount. | |
8366025e | 331 | */ |
eb04c282 | 332 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 333 | { |
83adc753 | 334 | struct mount *mnt = real_mount(m); |
3d733633 | 335 | int ret = 0; |
3d733633 | 336 | |
d3ef3d73 | 337 | preempt_disable(); |
c6653a83 | 338 | mnt_inc_writers(mnt); |
d3ef3d73 | 339 | /* |
c6653a83 | 340 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 341 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
342 | * incremented count after it has set MNT_WRITE_HOLD. | |
343 | */ | |
344 | smp_mb(); | |
1e75529e | 345 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 346 | cpu_relax(); |
347 | /* | |
348 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
349 | * be set to match its requirements. So we must not load that until | |
350 | * MNT_WRITE_HOLD is cleared. | |
351 | */ | |
352 | smp_rmb(); | |
4ed5e82f | 353 | if (mnt_is_readonly(m)) { |
c6653a83 | 354 | mnt_dec_writers(mnt); |
3d733633 | 355 | ret = -EROFS; |
3d733633 | 356 | } |
d3ef3d73 | 357 | preempt_enable(); |
eb04c282 JK |
358 | |
359 | return ret; | |
360 | } | |
361 | ||
362 | /** | |
363 | * mnt_want_write - get write access to a mount | |
364 | * @m: the mount on which to take a write | |
365 | * | |
366 | * This tells the low-level filesystem that a write is about to be performed to | |
367 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
368 | * is not frozen) before returning success. When the write operation is | |
369 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
370 | */ | |
371 | int mnt_want_write(struct vfsmount *m) | |
372 | { | |
373 | int ret; | |
374 | ||
375 | sb_start_write(m->mnt_sb); | |
376 | ret = __mnt_want_write(m); | |
377 | if (ret) | |
378 | sb_end_write(m->mnt_sb); | |
3d733633 | 379 | return ret; |
8366025e DH |
380 | } |
381 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
382 | ||
96029c4e | 383 | /** |
384 | * mnt_clone_write - get write access to a mount | |
385 | * @mnt: the mount on which to take a write | |
386 | * | |
387 | * This is effectively like mnt_want_write, except | |
388 | * it must only be used to take an extra write reference | |
389 | * on a mountpoint that we already know has a write reference | |
390 | * on it. This allows some optimisation. | |
391 | * | |
392 | * After finished, mnt_drop_write must be called as usual to | |
393 | * drop the reference. | |
394 | */ | |
395 | int mnt_clone_write(struct vfsmount *mnt) | |
396 | { | |
397 | /* superblock may be r/o */ | |
398 | if (__mnt_is_readonly(mnt)) | |
399 | return -EROFS; | |
400 | preempt_disable(); | |
83adc753 | 401 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 402 | preempt_enable(); |
403 | return 0; | |
404 | } | |
405 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
406 | ||
407 | /** | |
eb04c282 | 408 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 409 | * @file: the file who's mount on which to take a write |
410 | * | |
eb04c282 | 411 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 412 | * do some optimisations if the file is open for write already |
413 | */ | |
eb04c282 | 414 | int __mnt_want_write_file(struct file *file) |
96029c4e | 415 | { |
83f936c7 | 416 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 417 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 418 | else |
419 | return mnt_clone_write(file->f_path.mnt); | |
420 | } | |
eb04c282 JK |
421 | |
422 | /** | |
423 | * mnt_want_write_file - get write access to a file's mount | |
424 | * @file: the file who's mount on which to take a write | |
425 | * | |
426 | * This is like mnt_want_write, but it takes a file and can | |
427 | * do some optimisations if the file is open for write already | |
428 | */ | |
429 | int mnt_want_write_file(struct file *file) | |
430 | { | |
431 | int ret; | |
432 | ||
433 | sb_start_write(file->f_path.mnt->mnt_sb); | |
434 | ret = __mnt_want_write_file(file); | |
435 | if (ret) | |
436 | sb_end_write(file->f_path.mnt->mnt_sb); | |
437 | return ret; | |
438 | } | |
96029c4e | 439 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
440 | ||
8366025e | 441 | /** |
eb04c282 | 442 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
443 | * @mnt: the mount on which to give up write access |
444 | * | |
445 | * Tells the low-level filesystem that we are done | |
446 | * performing writes to it. Must be matched with | |
eb04c282 | 447 | * __mnt_want_write() call above. |
8366025e | 448 | */ |
eb04c282 | 449 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 450 | { |
d3ef3d73 | 451 | preempt_disable(); |
83adc753 | 452 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 453 | preempt_enable(); |
8366025e | 454 | } |
eb04c282 JK |
455 | |
456 | /** | |
457 | * mnt_drop_write - give up write access to a mount | |
458 | * @mnt: the mount on which to give up write access | |
459 | * | |
460 | * Tells the low-level filesystem that we are done performing writes to it and | |
461 | * also allows filesystem to be frozen again. Must be matched with | |
462 | * mnt_want_write() call above. | |
463 | */ | |
464 | void mnt_drop_write(struct vfsmount *mnt) | |
465 | { | |
466 | __mnt_drop_write(mnt); | |
467 | sb_end_write(mnt->mnt_sb); | |
468 | } | |
8366025e DH |
469 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
470 | ||
eb04c282 JK |
471 | void __mnt_drop_write_file(struct file *file) |
472 | { | |
473 | __mnt_drop_write(file->f_path.mnt); | |
474 | } | |
475 | ||
2a79f17e AV |
476 | void mnt_drop_write_file(struct file *file) |
477 | { | |
478 | mnt_drop_write(file->f_path.mnt); | |
479 | } | |
480 | EXPORT_SYMBOL(mnt_drop_write_file); | |
481 | ||
83adc753 | 482 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 483 | { |
3d733633 DH |
484 | int ret = 0; |
485 | ||
719ea2fb | 486 | lock_mount_hash(); |
83adc753 | 487 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 488 | /* |
d3ef3d73 | 489 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
490 | * should be visible before we do. | |
3d733633 | 491 | */ |
d3ef3d73 | 492 | smp_mb(); |
493 | ||
3d733633 | 494 | /* |
d3ef3d73 | 495 | * With writers on hold, if this value is zero, then there are |
496 | * definitely no active writers (although held writers may subsequently | |
497 | * increment the count, they'll have to wait, and decrement it after | |
498 | * seeing MNT_READONLY). | |
499 | * | |
500 | * It is OK to have counter incremented on one CPU and decremented on | |
501 | * another: the sum will add up correctly. The danger would be when we | |
502 | * sum up each counter, if we read a counter before it is incremented, | |
503 | * but then read another CPU's count which it has been subsequently | |
504 | * decremented from -- we would see more decrements than we should. | |
505 | * MNT_WRITE_HOLD protects against this scenario, because | |
506 | * mnt_want_write first increments count, then smp_mb, then spins on | |
507 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
508 | * we're counting up here. | |
3d733633 | 509 | */ |
c6653a83 | 510 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 511 | ret = -EBUSY; |
512 | else | |
83adc753 | 513 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 514 | /* |
515 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
516 | * that become unheld will see MNT_READONLY. | |
517 | */ | |
518 | smp_wmb(); | |
83adc753 | 519 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 520 | unlock_mount_hash(); |
3d733633 | 521 | return ret; |
8366025e | 522 | } |
8366025e | 523 | |
83adc753 | 524 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 525 | { |
719ea2fb | 526 | lock_mount_hash(); |
83adc753 | 527 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 528 | unlock_mount_hash(); |
2e4b7fcd DH |
529 | } |
530 | ||
4ed5e82f MS |
531 | int sb_prepare_remount_readonly(struct super_block *sb) |
532 | { | |
533 | struct mount *mnt; | |
534 | int err = 0; | |
535 | ||
8e8b8796 MS |
536 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
537 | if (atomic_long_read(&sb->s_remove_count)) | |
538 | return -EBUSY; | |
539 | ||
719ea2fb | 540 | lock_mount_hash(); |
4ed5e82f MS |
541 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
542 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
543 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
544 | smp_mb(); | |
545 | if (mnt_get_writers(mnt) > 0) { | |
546 | err = -EBUSY; | |
547 | break; | |
548 | } | |
549 | } | |
550 | } | |
8e8b8796 MS |
551 | if (!err && atomic_long_read(&sb->s_remove_count)) |
552 | err = -EBUSY; | |
553 | ||
4ed5e82f MS |
554 | if (!err) { |
555 | sb->s_readonly_remount = 1; | |
556 | smp_wmb(); | |
557 | } | |
558 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
559 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
560 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
561 | } | |
719ea2fb | 562 | unlock_mount_hash(); |
4ed5e82f MS |
563 | |
564 | return err; | |
565 | } | |
566 | ||
b105e270 | 567 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 568 | { |
52ba1621 | 569 | kfree(mnt->mnt_devname); |
d3ef3d73 | 570 | #ifdef CONFIG_SMP |
68e8a9fe | 571 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 572 | #endif |
b105e270 | 573 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
574 | } |
575 | ||
8ffcb32e DH |
576 | static void delayed_free_vfsmnt(struct rcu_head *head) |
577 | { | |
578 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
579 | } | |
580 | ||
48a066e7 AV |
581 | /* call under rcu_read_lock */ |
582 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
583 | { | |
584 | struct mount *mnt; | |
585 | if (read_seqretry(&mount_lock, seq)) | |
586 | return false; | |
587 | if (bastard == NULL) | |
588 | return true; | |
589 | mnt = real_mount(bastard); | |
590 | mnt_add_count(mnt, 1); | |
591 | if (likely(!read_seqretry(&mount_lock, seq))) | |
592 | return true; | |
593 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
594 | mnt_add_count(mnt, -1); | |
595 | return false; | |
596 | } | |
597 | rcu_read_unlock(); | |
598 | mntput(bastard); | |
599 | rcu_read_lock(); | |
600 | return false; | |
601 | } | |
602 | ||
1da177e4 | 603 | /* |
474279dc | 604 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 605 | * call under rcu_read_lock() |
1da177e4 | 606 | */ |
474279dc | 607 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 608 | { |
38129a13 | 609 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
610 | struct mount *p; |
611 | ||
38129a13 | 612 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
613 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
614 | return p; | |
615 | return NULL; | |
616 | } | |
617 | ||
618 | /* | |
619 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 620 | * mount_lock must be held. |
474279dc AV |
621 | */ |
622 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
623 | { | |
38129a13 AV |
624 | struct mount *p, *res; |
625 | res = p = __lookup_mnt(mnt, dentry); | |
626 | if (!p) | |
627 | goto out; | |
628 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
629 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
630 | break; | |
631 | res = p; | |
632 | } | |
38129a13 | 633 | out: |
1d6a32ac | 634 | return res; |
1da177e4 LT |
635 | } |
636 | ||
a05964f3 | 637 | /* |
f015f126 DH |
638 | * lookup_mnt - Return the first child mount mounted at path |
639 | * | |
640 | * "First" means first mounted chronologically. If you create the | |
641 | * following mounts: | |
642 | * | |
643 | * mount /dev/sda1 /mnt | |
644 | * mount /dev/sda2 /mnt | |
645 | * mount /dev/sda3 /mnt | |
646 | * | |
647 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
648 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
649 | * /dev/sda2, then /dev/sda3, then NULL. | |
650 | * | |
651 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 652 | */ |
1c755af4 | 653 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 654 | { |
c7105365 | 655 | struct mount *child_mnt; |
48a066e7 AV |
656 | struct vfsmount *m; |
657 | unsigned seq; | |
99b7db7b | 658 | |
48a066e7 AV |
659 | rcu_read_lock(); |
660 | do { | |
661 | seq = read_seqbegin(&mount_lock); | |
662 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
663 | m = child_mnt ? &child_mnt->mnt : NULL; | |
664 | } while (!legitimize_mnt(m, seq)); | |
665 | rcu_read_unlock(); | |
666 | return m; | |
a05964f3 RP |
667 | } |
668 | ||
84d17192 AV |
669 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
670 | { | |
0818bf27 | 671 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 | 672 | struct mountpoint *mp; |
eed81007 | 673 | int ret; |
84d17192 | 674 | |
0818bf27 | 675 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
676 | if (mp->m_dentry == dentry) { |
677 | /* might be worth a WARN_ON() */ | |
678 | if (d_unlinked(dentry)) | |
679 | return ERR_PTR(-ENOENT); | |
680 | mp->m_count++; | |
681 | return mp; | |
682 | } | |
683 | } | |
684 | ||
685 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
686 | if (!mp) | |
687 | return ERR_PTR(-ENOMEM); | |
688 | ||
eed81007 MS |
689 | ret = d_set_mounted(dentry); |
690 | if (ret) { | |
84d17192 | 691 | kfree(mp); |
eed81007 | 692 | return ERR_PTR(ret); |
84d17192 | 693 | } |
eed81007 | 694 | |
84d17192 AV |
695 | mp->m_dentry = dentry; |
696 | mp->m_count = 1; | |
0818bf27 | 697 | hlist_add_head(&mp->m_hash, chain); |
84d17192 AV |
698 | return mp; |
699 | } | |
700 | ||
701 | static void put_mountpoint(struct mountpoint *mp) | |
702 | { | |
703 | if (!--mp->m_count) { | |
704 | struct dentry *dentry = mp->m_dentry; | |
705 | spin_lock(&dentry->d_lock); | |
706 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
707 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 708 | hlist_del(&mp->m_hash); |
84d17192 AV |
709 | kfree(mp); |
710 | } | |
711 | } | |
712 | ||
143c8c91 | 713 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 714 | { |
6b3286ed | 715 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
716 | } |
717 | ||
99b7db7b NP |
718 | /* |
719 | * vfsmount lock must be held for write | |
720 | */ | |
6b3286ed | 721 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
722 | { |
723 | if (ns) { | |
724 | ns->event = ++event; | |
725 | wake_up_interruptible(&ns->poll); | |
726 | } | |
727 | } | |
728 | ||
99b7db7b NP |
729 | /* |
730 | * vfsmount lock must be held for write | |
731 | */ | |
6b3286ed | 732 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
733 | { |
734 | if (ns && ns->event != event) { | |
735 | ns->event = event; | |
736 | wake_up_interruptible(&ns->poll); | |
737 | } | |
738 | } | |
739 | ||
99b7db7b NP |
740 | /* |
741 | * vfsmount lock must be held for write | |
742 | */ | |
419148da AV |
743 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
744 | { | |
a73324da | 745 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
746 | old_path->mnt = &mnt->mnt_parent->mnt; |
747 | mnt->mnt_parent = mnt; | |
a73324da | 748 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 749 | list_del_init(&mnt->mnt_child); |
38129a13 | 750 | hlist_del_init_rcu(&mnt->mnt_hash); |
84d17192 AV |
751 | put_mountpoint(mnt->mnt_mp); |
752 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
753 | } |
754 | ||
99b7db7b NP |
755 | /* |
756 | * vfsmount lock must be held for write | |
757 | */ | |
84d17192 AV |
758 | void mnt_set_mountpoint(struct mount *mnt, |
759 | struct mountpoint *mp, | |
44d964d6 | 760 | struct mount *child_mnt) |
b90fa9ae | 761 | { |
84d17192 | 762 | mp->m_count++; |
3a2393d7 | 763 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 764 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 765 | child_mnt->mnt_parent = mnt; |
84d17192 | 766 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
767 | } |
768 | ||
99b7db7b NP |
769 | /* |
770 | * vfsmount lock must be held for write | |
771 | */ | |
84d17192 AV |
772 | static void attach_mnt(struct mount *mnt, |
773 | struct mount *parent, | |
774 | struct mountpoint *mp) | |
1da177e4 | 775 | { |
84d17192 | 776 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 777 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 778 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
779 | } |
780 | ||
781 | /* | |
99b7db7b | 782 | * vfsmount lock must be held for write |
b90fa9ae | 783 | */ |
1d6a32ac | 784 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 785 | { |
0714a533 | 786 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 787 | struct mount *m; |
b90fa9ae | 788 | LIST_HEAD(head); |
143c8c91 | 789 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 790 | |
0714a533 | 791 | BUG_ON(parent == mnt); |
b90fa9ae | 792 | |
1a4eeaf2 | 793 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 794 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 795 | m->mnt_ns = n; |
f03c6599 | 796 | |
b90fa9ae RP |
797 | list_splice(&head, n->list.prev); |
798 | ||
1d6a32ac | 799 | if (shadows) |
38129a13 | 800 | hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash); |
1d6a32ac | 801 | else |
38129a13 | 802 | hlist_add_head_rcu(&mnt->mnt_hash, |
0818bf27 | 803 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 804 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 805 | touch_mnt_namespace(n); |
1da177e4 LT |
806 | } |
807 | ||
909b0a88 | 808 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 809 | { |
6b41d536 AV |
810 | struct list_head *next = p->mnt_mounts.next; |
811 | if (next == &p->mnt_mounts) { | |
1da177e4 | 812 | while (1) { |
909b0a88 | 813 | if (p == root) |
1da177e4 | 814 | return NULL; |
6b41d536 AV |
815 | next = p->mnt_child.next; |
816 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 817 | break; |
0714a533 | 818 | p = p->mnt_parent; |
1da177e4 LT |
819 | } |
820 | } | |
6b41d536 | 821 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
822 | } |
823 | ||
315fc83e | 824 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 825 | { |
6b41d536 AV |
826 | struct list_head *prev = p->mnt_mounts.prev; |
827 | while (prev != &p->mnt_mounts) { | |
828 | p = list_entry(prev, struct mount, mnt_child); | |
829 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
830 | } |
831 | return p; | |
832 | } | |
833 | ||
9d412a43 AV |
834 | struct vfsmount * |
835 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
836 | { | |
b105e270 | 837 | struct mount *mnt; |
9d412a43 AV |
838 | struct dentry *root; |
839 | ||
840 | if (!type) | |
841 | return ERR_PTR(-ENODEV); | |
842 | ||
843 | mnt = alloc_vfsmnt(name); | |
844 | if (!mnt) | |
845 | return ERR_PTR(-ENOMEM); | |
846 | ||
847 | if (flags & MS_KERNMOUNT) | |
b105e270 | 848 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
849 | |
850 | root = mount_fs(type, flags, name, data); | |
851 | if (IS_ERR(root)) { | |
8ffcb32e | 852 | mnt_free_id(mnt); |
9d412a43 AV |
853 | free_vfsmnt(mnt); |
854 | return ERR_CAST(root); | |
855 | } | |
856 | ||
b105e270 AV |
857 | mnt->mnt.mnt_root = root; |
858 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 859 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 860 | mnt->mnt_parent = mnt; |
719ea2fb | 861 | lock_mount_hash(); |
39f7c4db | 862 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 863 | unlock_mount_hash(); |
b105e270 | 864 | return &mnt->mnt; |
9d412a43 AV |
865 | } |
866 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
867 | ||
87129cc0 | 868 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 869 | int flag) |
1da177e4 | 870 | { |
87129cc0 | 871 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
872 | struct mount *mnt; |
873 | int err; | |
1da177e4 | 874 | |
be34d1a3 DH |
875 | mnt = alloc_vfsmnt(old->mnt_devname); |
876 | if (!mnt) | |
877 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 878 | |
7a472ef4 | 879 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
880 | mnt->mnt_group_id = 0; /* not a peer of original */ |
881 | else | |
882 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 883 | |
be34d1a3 DH |
884 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
885 | err = mnt_alloc_group_id(mnt); | |
886 | if (err) | |
887 | goto out_free; | |
1da177e4 | 888 | } |
be34d1a3 | 889 | |
f2ebb3a9 | 890 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 EB |
891 | /* Don't allow unprivileged users to change mount flags */ |
892 | if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) | |
893 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
894 | ||
5ff9d8a6 EB |
895 | /* Don't allow unprivileged users to reveal what is under a mount */ |
896 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
897 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
898 | ||
be34d1a3 DH |
899 | atomic_inc(&sb->s_active); |
900 | mnt->mnt.mnt_sb = sb; | |
901 | mnt->mnt.mnt_root = dget(root); | |
902 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
903 | mnt->mnt_parent = mnt; | |
719ea2fb | 904 | lock_mount_hash(); |
be34d1a3 | 905 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 906 | unlock_mount_hash(); |
be34d1a3 | 907 | |
7a472ef4 EB |
908 | if ((flag & CL_SLAVE) || |
909 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
910 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
911 | mnt->mnt_master = old; | |
912 | CLEAR_MNT_SHARED(mnt); | |
913 | } else if (!(flag & CL_PRIVATE)) { | |
914 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
915 | list_add(&mnt->mnt_share, &old->mnt_share); | |
916 | if (IS_MNT_SLAVE(old)) | |
917 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
918 | mnt->mnt_master = old->mnt_master; | |
919 | } | |
920 | if (flag & CL_MAKE_SHARED) | |
921 | set_mnt_shared(mnt); | |
922 | ||
923 | /* stick the duplicate mount on the same expiry list | |
924 | * as the original if that was on one */ | |
925 | if (flag & CL_EXPIRE) { | |
926 | if (!list_empty(&old->mnt_expire)) | |
927 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
928 | } | |
929 | ||
cb338d06 | 930 | return mnt; |
719f5d7f MS |
931 | |
932 | out_free: | |
8ffcb32e | 933 | mnt_free_id(mnt); |
719f5d7f | 934 | free_vfsmnt(mnt); |
be34d1a3 | 935 | return ERR_PTR(err); |
1da177e4 LT |
936 | } |
937 | ||
900148dc | 938 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 939 | { |
48a066e7 AV |
940 | rcu_read_lock(); |
941 | mnt_add_count(mnt, -1); | |
942 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
943 | rcu_read_unlock(); | |
f03c6599 | 944 | return; |
b3e19d92 | 945 | } |
719ea2fb | 946 | lock_mount_hash(); |
b3e19d92 | 947 | if (mnt_get_count(mnt)) { |
48a066e7 | 948 | rcu_read_unlock(); |
719ea2fb | 949 | unlock_mount_hash(); |
99b7db7b NP |
950 | return; |
951 | } | |
48a066e7 AV |
952 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
953 | rcu_read_unlock(); | |
954 | unlock_mount_hash(); | |
955 | return; | |
956 | } | |
957 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
958 | rcu_read_unlock(); | |
962830df | 959 | |
39f7c4db | 960 | list_del(&mnt->mnt_instance); |
719ea2fb | 961 | unlock_mount_hash(); |
649a795a AV |
962 | |
963 | /* | |
964 | * This probably indicates that somebody messed | |
965 | * up a mnt_want/drop_write() pair. If this | |
966 | * happens, the filesystem was probably unable | |
967 | * to make r/w->r/o transitions. | |
968 | */ | |
969 | /* | |
970 | * The locking used to deal with mnt_count decrement provides barriers, | |
971 | * so mnt_get_writers() below is safe. | |
972 | */ | |
973 | WARN_ON(mnt_get_writers(mnt)); | |
3064c356 AV |
974 | if (unlikely(mnt->mnt_pins.first)) |
975 | mnt_pin_kill(mnt); | |
649a795a AV |
976 | fsnotify_vfsmount_delete(&mnt->mnt); |
977 | dput(mnt->mnt.mnt_root); | |
978 | deactivate_super(mnt->mnt.mnt_sb); | |
48a066e7 | 979 | mnt_free_id(mnt); |
8ffcb32e | 980 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); |
b3e19d92 | 981 | } |
b3e19d92 NP |
982 | |
983 | void mntput(struct vfsmount *mnt) | |
984 | { | |
985 | if (mnt) { | |
863d684f | 986 | struct mount *m = real_mount(mnt); |
b3e19d92 | 987 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
988 | if (unlikely(m->mnt_expiry_mark)) |
989 | m->mnt_expiry_mark = 0; | |
990 | mntput_no_expire(m); | |
b3e19d92 NP |
991 | } |
992 | } | |
993 | EXPORT_SYMBOL(mntput); | |
994 | ||
995 | struct vfsmount *mntget(struct vfsmount *mnt) | |
996 | { | |
997 | if (mnt) | |
83adc753 | 998 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
999 | return mnt; |
1000 | } | |
1001 | EXPORT_SYMBOL(mntget); | |
1002 | ||
3064c356 | 1003 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1004 | { |
3064c356 AV |
1005 | struct mount *p; |
1006 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1007 | if (IS_ERR(p)) | |
1008 | return ERR_CAST(p); | |
1009 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1010 | return &p->mnt; | |
7b7b1ace | 1011 | } |
1da177e4 | 1012 | |
b3b304a2 MS |
1013 | static inline void mangle(struct seq_file *m, const char *s) |
1014 | { | |
1015 | seq_escape(m, s, " \t\n\\"); | |
1016 | } | |
1017 | ||
1018 | /* | |
1019 | * Simple .show_options callback for filesystems which don't want to | |
1020 | * implement more complex mount option showing. | |
1021 | * | |
1022 | * See also save_mount_options(). | |
1023 | */ | |
34c80b1d | 1024 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1025 | { |
2a32cebd AV |
1026 | const char *options; |
1027 | ||
1028 | rcu_read_lock(); | |
34c80b1d | 1029 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1030 | |
1031 | if (options != NULL && options[0]) { | |
1032 | seq_putc(m, ','); | |
1033 | mangle(m, options); | |
1034 | } | |
2a32cebd | 1035 | rcu_read_unlock(); |
b3b304a2 MS |
1036 | |
1037 | return 0; | |
1038 | } | |
1039 | EXPORT_SYMBOL(generic_show_options); | |
1040 | ||
1041 | /* | |
1042 | * If filesystem uses generic_show_options(), this function should be | |
1043 | * called from the fill_super() callback. | |
1044 | * | |
1045 | * The .remount_fs callback usually needs to be handled in a special | |
1046 | * way, to make sure, that previous options are not overwritten if the | |
1047 | * remount fails. | |
1048 | * | |
1049 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1050 | * reset all options to their default value, but changes only newly | |
1051 | * given options, then the displayed options will not reflect reality | |
1052 | * any more. | |
1053 | */ | |
1054 | void save_mount_options(struct super_block *sb, char *options) | |
1055 | { | |
2a32cebd AV |
1056 | BUG_ON(sb->s_options); |
1057 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1058 | } |
1059 | EXPORT_SYMBOL(save_mount_options); | |
1060 | ||
2a32cebd AV |
1061 | void replace_mount_options(struct super_block *sb, char *options) |
1062 | { | |
1063 | char *old = sb->s_options; | |
1064 | rcu_assign_pointer(sb->s_options, options); | |
1065 | if (old) { | |
1066 | synchronize_rcu(); | |
1067 | kfree(old); | |
1068 | } | |
1069 | } | |
1070 | EXPORT_SYMBOL(replace_mount_options); | |
1071 | ||
a1a2c409 | 1072 | #ifdef CONFIG_PROC_FS |
0226f492 | 1073 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1074 | static void *m_start(struct seq_file *m, loff_t *pos) |
1075 | { | |
6ce6e24e | 1076 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1077 | |
390c6843 | 1078 | down_read(&namespace_sem); |
c7999c36 AV |
1079 | if (p->cached_event == p->ns->event) { |
1080 | void *v = p->cached_mount; | |
1081 | if (*pos == p->cached_index) | |
1082 | return v; | |
1083 | if (*pos == p->cached_index + 1) { | |
1084 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1085 | return p->cached_mount = v; | |
1086 | } | |
1087 | } | |
1088 | ||
1089 | p->cached_event = p->ns->event; | |
1090 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1091 | p->cached_index = *pos; | |
1092 | return p->cached_mount; | |
1da177e4 LT |
1093 | } |
1094 | ||
1095 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1096 | { | |
6ce6e24e | 1097 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1098 | |
c7999c36 AV |
1099 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1100 | p->cached_index = *pos; | |
1101 | return p->cached_mount; | |
1da177e4 LT |
1102 | } |
1103 | ||
1104 | static void m_stop(struct seq_file *m, void *v) | |
1105 | { | |
390c6843 | 1106 | up_read(&namespace_sem); |
1da177e4 LT |
1107 | } |
1108 | ||
0226f492 | 1109 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1110 | { |
6ce6e24e | 1111 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1112 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1113 | return p->show(m, &r->mnt); |
1da177e4 LT |
1114 | } |
1115 | ||
a1a2c409 | 1116 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1117 | .start = m_start, |
1118 | .next = m_next, | |
1119 | .stop = m_stop, | |
0226f492 | 1120 | .show = m_show, |
b4629fe2 | 1121 | }; |
a1a2c409 | 1122 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1123 | |
1da177e4 LT |
1124 | /** |
1125 | * may_umount_tree - check if a mount tree is busy | |
1126 | * @mnt: root of mount tree | |
1127 | * | |
1128 | * This is called to check if a tree of mounts has any | |
1129 | * open files, pwds, chroots or sub mounts that are | |
1130 | * busy. | |
1131 | */ | |
909b0a88 | 1132 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1133 | { |
909b0a88 | 1134 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1135 | int actual_refs = 0; |
1136 | int minimum_refs = 0; | |
315fc83e | 1137 | struct mount *p; |
909b0a88 | 1138 | BUG_ON(!m); |
1da177e4 | 1139 | |
b3e19d92 | 1140 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1141 | lock_mount_hash(); |
909b0a88 | 1142 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1143 | actual_refs += mnt_get_count(p); |
1da177e4 | 1144 | minimum_refs += 2; |
1da177e4 | 1145 | } |
719ea2fb | 1146 | unlock_mount_hash(); |
1da177e4 LT |
1147 | |
1148 | if (actual_refs > minimum_refs) | |
e3474a8e | 1149 | return 0; |
1da177e4 | 1150 | |
e3474a8e | 1151 | return 1; |
1da177e4 LT |
1152 | } |
1153 | ||
1154 | EXPORT_SYMBOL(may_umount_tree); | |
1155 | ||
1156 | /** | |
1157 | * may_umount - check if a mount point is busy | |
1158 | * @mnt: root of mount | |
1159 | * | |
1160 | * This is called to check if a mount point has any | |
1161 | * open files, pwds, chroots or sub mounts. If the | |
1162 | * mount has sub mounts this will return busy | |
1163 | * regardless of whether the sub mounts are busy. | |
1164 | * | |
1165 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1166 | * give false negatives. The main reason why it's here is that we need | |
1167 | * a non-destructive way to look for easily umountable filesystems. | |
1168 | */ | |
1169 | int may_umount(struct vfsmount *mnt) | |
1170 | { | |
e3474a8e | 1171 | int ret = 1; |
8ad08d8a | 1172 | down_read(&namespace_sem); |
719ea2fb | 1173 | lock_mount_hash(); |
1ab59738 | 1174 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1175 | ret = 0; |
719ea2fb | 1176 | unlock_mount_hash(); |
8ad08d8a | 1177 | up_read(&namespace_sem); |
a05964f3 | 1178 | return ret; |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | EXPORT_SYMBOL(may_umount); | |
1182 | ||
38129a13 | 1183 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1184 | |
97216be0 | 1185 | static void namespace_unlock(void) |
70fbcdf4 | 1186 | { |
d5e50f74 | 1187 | struct mount *mnt; |
38129a13 | 1188 | struct hlist_head head = unmounted; |
97216be0 | 1189 | |
38129a13 | 1190 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1191 | up_write(&namespace_sem); |
1192 | return; | |
1193 | } | |
1194 | ||
38129a13 AV |
1195 | head.first->pprev = &head.first; |
1196 | INIT_HLIST_HEAD(&unmounted); | |
1197 | ||
97216be0 AV |
1198 | up_write(&namespace_sem); |
1199 | ||
48a066e7 AV |
1200 | synchronize_rcu(); |
1201 | ||
38129a13 AV |
1202 | while (!hlist_empty(&head)) { |
1203 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1204 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1205 | if (mnt->mnt_ex_mountpoint.mnt) |
1206 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1207 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1208 | } |
1209 | } | |
1210 | ||
97216be0 | 1211 | static inline void namespace_lock(void) |
e3197d83 | 1212 | { |
97216be0 | 1213 | down_write(&namespace_sem); |
e3197d83 AV |
1214 | } |
1215 | ||
99b7db7b | 1216 | /* |
48a066e7 | 1217 | * mount_lock must be held |
99b7db7b | 1218 | * namespace_sem must be held for write |
48a066e7 AV |
1219 | * how = 0 => just this tree, don't propagate |
1220 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1221 | * how = 2 => lazy umount | |
99b7db7b | 1222 | */ |
48a066e7 | 1223 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1224 | { |
38129a13 | 1225 | HLIST_HEAD(tmp_list); |
315fc83e | 1226 | struct mount *p; |
38129a13 | 1227 | struct mount *last = NULL; |
1da177e4 | 1228 | |
38129a13 AV |
1229 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1230 | hlist_del_init_rcu(&p->mnt_hash); | |
1231 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1232 | } | |
1da177e4 | 1233 | |
48a066e7 | 1234 | if (how) |
7b8a53fd | 1235 | propagate_umount(&tmp_list); |
a05964f3 | 1236 | |
38129a13 | 1237 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1238 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1239 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1240 | __touch_mnt_namespace(p->mnt_ns); |
1241 | p->mnt_ns = NULL; | |
48a066e7 AV |
1242 | if (how < 2) |
1243 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
6b41d536 | 1244 | list_del_init(&p->mnt_child); |
676da58d | 1245 | if (mnt_has_parent(p)) { |
84d17192 | 1246 | put_mountpoint(p->mnt_mp); |
aba809cf AV |
1247 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1248 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1249 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1250 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1251 | p->mnt_parent = p; | |
84d17192 | 1252 | p->mnt_mp = NULL; |
7c4b93d8 | 1253 | } |
0f0afb1d | 1254 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1255 | last = p; |
1256 | } | |
1257 | if (last) { | |
1258 | last->mnt_hash.next = unmounted.first; | |
1259 | unmounted.first = tmp_list.first; | |
1260 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1261 | } |
1262 | } | |
1263 | ||
b54b9be7 | 1264 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1265 | |
1ab59738 | 1266 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1267 | { |
1ab59738 | 1268 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1269 | int retval; |
1270 | ||
1ab59738 | 1271 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1272 | if (retval) |
1273 | return retval; | |
1274 | ||
1275 | /* | |
1276 | * Allow userspace to request a mountpoint be expired rather than | |
1277 | * unmounting unconditionally. Unmount only happens if: | |
1278 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1279 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1280 | */ | |
1281 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1282 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1283 | flags & (MNT_FORCE | MNT_DETACH)) |
1284 | return -EINVAL; | |
1285 | ||
b3e19d92 NP |
1286 | /* |
1287 | * probably don't strictly need the lock here if we examined | |
1288 | * all race cases, but it's a slowpath. | |
1289 | */ | |
719ea2fb | 1290 | lock_mount_hash(); |
83adc753 | 1291 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1292 | unlock_mount_hash(); |
1da177e4 | 1293 | return -EBUSY; |
b3e19d92 | 1294 | } |
719ea2fb | 1295 | unlock_mount_hash(); |
1da177e4 | 1296 | |
863d684f | 1297 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1298 | return -EAGAIN; |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * If we may have to abort operations to get out of this | |
1303 | * mount, and they will themselves hold resources we must | |
1304 | * allow the fs to do things. In the Unix tradition of | |
1305 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1306 | * might fail to complete on the first run through as other tasks | |
1307 | * must return, and the like. Thats for the mount program to worry | |
1308 | * about for the moment. | |
1309 | */ | |
1310 | ||
42faad99 | 1311 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1312 | sb->s_op->umount_begin(sb); |
42faad99 | 1313 | } |
1da177e4 LT |
1314 | |
1315 | /* | |
1316 | * No sense to grab the lock for this test, but test itself looks | |
1317 | * somewhat bogus. Suggestions for better replacement? | |
1318 | * Ho-hum... In principle, we might treat that as umount + switch | |
1319 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1320 | * Actually it makes sense, especially if rootfs would contain a | |
1321 | * /reboot - static binary that would close all descriptors and | |
1322 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1323 | */ | |
1ab59738 | 1324 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1325 | /* |
1326 | * Special case for "unmounting" root ... | |
1327 | * we just try to remount it readonly. | |
1328 | */ | |
1329 | down_write(&sb->s_umount); | |
4aa98cf7 | 1330 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1331 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1332 | up_write(&sb->s_umount); |
1333 | return retval; | |
1334 | } | |
1335 | ||
97216be0 | 1336 | namespace_lock(); |
719ea2fb | 1337 | lock_mount_hash(); |
5addc5dd | 1338 | event++; |
1da177e4 | 1339 | |
48a066e7 | 1340 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1341 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1342 | umount_tree(mnt, 2); |
1da177e4 | 1343 | retval = 0; |
48a066e7 AV |
1344 | } else { |
1345 | shrink_submounts(mnt); | |
1346 | retval = -EBUSY; | |
1347 | if (!propagate_mount_busy(mnt, 2)) { | |
1348 | if (!list_empty(&mnt->mnt_list)) | |
1349 | umount_tree(mnt, 1); | |
1350 | retval = 0; | |
1351 | } | |
1da177e4 | 1352 | } |
719ea2fb | 1353 | unlock_mount_hash(); |
e3197d83 | 1354 | namespace_unlock(); |
1da177e4 LT |
1355 | return retval; |
1356 | } | |
1357 | ||
9b40bc90 AV |
1358 | /* |
1359 | * Is the caller allowed to modify his namespace? | |
1360 | */ | |
1361 | static inline bool may_mount(void) | |
1362 | { | |
1363 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1364 | } | |
1365 | ||
1da177e4 LT |
1366 | /* |
1367 | * Now umount can handle mount points as well as block devices. | |
1368 | * This is important for filesystems which use unnamed block devices. | |
1369 | * | |
1370 | * We now support a flag for forced unmount like the other 'big iron' | |
1371 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1372 | */ | |
1373 | ||
bdc480e3 | 1374 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1375 | { |
2d8f3038 | 1376 | struct path path; |
900148dc | 1377 | struct mount *mnt; |
1da177e4 | 1378 | int retval; |
db1f05bb | 1379 | int lookup_flags = 0; |
1da177e4 | 1380 | |
db1f05bb MS |
1381 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1382 | return -EINVAL; | |
1383 | ||
9b40bc90 AV |
1384 | if (!may_mount()) |
1385 | return -EPERM; | |
1386 | ||
db1f05bb MS |
1387 | if (!(flags & UMOUNT_NOFOLLOW)) |
1388 | lookup_flags |= LOOKUP_FOLLOW; | |
1389 | ||
197df04c | 1390 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1391 | if (retval) |
1392 | goto out; | |
900148dc | 1393 | mnt = real_mount(path.mnt); |
1da177e4 | 1394 | retval = -EINVAL; |
2d8f3038 | 1395 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1396 | goto dput_and_out; |
143c8c91 | 1397 | if (!check_mnt(mnt)) |
1da177e4 | 1398 | goto dput_and_out; |
5ff9d8a6 EB |
1399 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1400 | goto dput_and_out; | |
1da177e4 | 1401 | |
900148dc | 1402 | retval = do_umount(mnt, flags); |
1da177e4 | 1403 | dput_and_out: |
429731b1 | 1404 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1405 | dput(path.dentry); |
900148dc | 1406 | mntput_no_expire(mnt); |
1da177e4 LT |
1407 | out: |
1408 | return retval; | |
1409 | } | |
1410 | ||
1411 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1412 | ||
1413 | /* | |
b58fed8b | 1414 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1415 | */ |
bdc480e3 | 1416 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1417 | { |
b58fed8b | 1418 | return sys_umount(name, 0); |
1da177e4 LT |
1419 | } |
1420 | ||
1421 | #endif | |
1422 | ||
4ce5d2b1 | 1423 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1424 | { |
4ce5d2b1 EB |
1425 | /* Is this a proxy for a mount namespace? */ |
1426 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1427 | struct proc_ns *ei; |
8823c079 EB |
1428 | |
1429 | if (!proc_ns_inode(inode)) | |
1430 | return false; | |
1431 | ||
0bb80f24 | 1432 | ei = get_proc_ns(inode); |
8823c079 EB |
1433 | if (ei->ns_ops != &mntns_operations) |
1434 | return false; | |
1435 | ||
4ce5d2b1 EB |
1436 | return true; |
1437 | } | |
1438 | ||
1439 | static bool mnt_ns_loop(struct dentry *dentry) | |
1440 | { | |
1441 | /* Could bind mounting the mount namespace inode cause a | |
1442 | * mount namespace loop? | |
1443 | */ | |
1444 | struct mnt_namespace *mnt_ns; | |
1445 | if (!is_mnt_ns_file(dentry)) | |
1446 | return false; | |
1447 | ||
1448 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1449 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1450 | } | |
1451 | ||
87129cc0 | 1452 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1453 | int flag) |
1da177e4 | 1454 | { |
84d17192 | 1455 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1456 | |
4ce5d2b1 EB |
1457 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1458 | return ERR_PTR(-EINVAL); | |
1459 | ||
1460 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1461 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1462 | |
36341f64 | 1463 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1464 | if (IS_ERR(q)) |
1465 | return q; | |
1466 | ||
5ff9d8a6 | 1467 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1468 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1469 | |
1470 | p = mnt; | |
6b41d536 | 1471 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1472 | struct mount *s; |
7ec02ef1 | 1473 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1474 | continue; |
1475 | ||
909b0a88 | 1476 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1477 | if (!(flag & CL_COPY_UNBINDABLE) && |
1478 | IS_MNT_UNBINDABLE(s)) { | |
1479 | s = skip_mnt_tree(s); | |
1480 | continue; | |
1481 | } | |
1482 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1483 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1484 | s = skip_mnt_tree(s); |
1485 | continue; | |
1486 | } | |
0714a533 AV |
1487 | while (p != s->mnt_parent) { |
1488 | p = p->mnt_parent; | |
1489 | q = q->mnt_parent; | |
1da177e4 | 1490 | } |
87129cc0 | 1491 | p = s; |
84d17192 | 1492 | parent = q; |
87129cc0 | 1493 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1494 | if (IS_ERR(q)) |
1495 | goto out; | |
719ea2fb | 1496 | lock_mount_hash(); |
1a4eeaf2 | 1497 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1498 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1499 | unlock_mount_hash(); |
1da177e4 LT |
1500 | } |
1501 | } | |
1502 | return res; | |
be34d1a3 | 1503 | out: |
1da177e4 | 1504 | if (res) { |
719ea2fb | 1505 | lock_mount_hash(); |
328e6d90 | 1506 | umount_tree(res, 0); |
719ea2fb | 1507 | unlock_mount_hash(); |
1da177e4 | 1508 | } |
be34d1a3 | 1509 | return q; |
1da177e4 LT |
1510 | } |
1511 | ||
be34d1a3 DH |
1512 | /* Caller should check returned pointer for errors */ |
1513 | ||
589ff870 | 1514 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1515 | { |
cb338d06 | 1516 | struct mount *tree; |
97216be0 | 1517 | namespace_lock(); |
87129cc0 AV |
1518 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1519 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1520 | namespace_unlock(); |
be34d1a3 | 1521 | if (IS_ERR(tree)) |
52e220d3 | 1522 | return ERR_CAST(tree); |
be34d1a3 | 1523 | return &tree->mnt; |
8aec0809 AV |
1524 | } |
1525 | ||
1526 | void drop_collected_mounts(struct vfsmount *mnt) | |
1527 | { | |
97216be0 | 1528 | namespace_lock(); |
719ea2fb | 1529 | lock_mount_hash(); |
328e6d90 | 1530 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1531 | unlock_mount_hash(); |
3ab6abee | 1532 | namespace_unlock(); |
8aec0809 AV |
1533 | } |
1534 | ||
1f707137 AV |
1535 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1536 | struct vfsmount *root) | |
1537 | { | |
1a4eeaf2 | 1538 | struct mount *mnt; |
1f707137 AV |
1539 | int res = f(root, arg); |
1540 | if (res) | |
1541 | return res; | |
1a4eeaf2 AV |
1542 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1543 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1544 | if (res) |
1545 | return res; | |
1546 | } | |
1547 | return 0; | |
1548 | } | |
1549 | ||
4b8b21f4 | 1550 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1551 | { |
315fc83e | 1552 | struct mount *p; |
719f5d7f | 1553 | |
909b0a88 | 1554 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1555 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1556 | mnt_release_group_id(p); |
719f5d7f MS |
1557 | } |
1558 | } | |
1559 | ||
4b8b21f4 | 1560 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1561 | { |
315fc83e | 1562 | struct mount *p; |
719f5d7f | 1563 | |
909b0a88 | 1564 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1565 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1566 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1567 | if (err) { |
4b8b21f4 | 1568 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1569 | return err; |
1570 | } | |
1571 | } | |
1572 | } | |
1573 | ||
1574 | return 0; | |
1575 | } | |
1576 | ||
b90fa9ae RP |
1577 | /* |
1578 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1579 | * @nd : place the mount tree @source_mnt is attached |
1580 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1581 | * store the parent mount and mountpoint dentry. | |
1582 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1583 | * |
1584 | * NOTE: in the table below explains the semantics when a source mount | |
1585 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1586 | * --------------------------------------------------------------------------- |
1587 | * | BIND MOUNT OPERATION | | |
1588 | * |************************************************************************** | |
1589 | * | source-->| shared | private | slave | unbindable | | |
1590 | * | dest | | | | | | |
1591 | * | | | | | | | | |
1592 | * | v | | | | | | |
1593 | * |************************************************************************** | |
1594 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1595 | * | | | | | | | |
1596 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1597 | * *************************************************************************** | |
b90fa9ae RP |
1598 | * A bind operation clones the source mount and mounts the clone on the |
1599 | * destination mount. | |
1600 | * | |
1601 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1602 | * tree of the destination mount and the cloned mount is added to | |
1603 | * the peer group of the source mount. | |
1604 | * (+) the cloned mount is created under the destination mount and is marked | |
1605 | * as shared. The cloned mount is added to the peer group of the source | |
1606 | * mount. | |
5afe0022 RP |
1607 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1608 | * of the destination mount and the cloned mount is made slave | |
1609 | * of the same master as that of the source mount. The cloned mount | |
1610 | * is marked as 'shared and slave'. | |
1611 | * (*) the cloned mount is made a slave of the same master as that of the | |
1612 | * source mount. | |
1613 | * | |
9676f0c6 RP |
1614 | * --------------------------------------------------------------------------- |
1615 | * | MOVE MOUNT OPERATION | | |
1616 | * |************************************************************************** | |
1617 | * | source-->| shared | private | slave | unbindable | | |
1618 | * | dest | | | | | | |
1619 | * | | | | | | | | |
1620 | * | v | | | | | | |
1621 | * |************************************************************************** | |
1622 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1623 | * | | | | | | | |
1624 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1625 | * *************************************************************************** | |
5afe0022 RP |
1626 | * |
1627 | * (+) the mount is moved to the destination. And is then propagated to | |
1628 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1629 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1630 | * (+++) the mount is moved to the destination and is then propagated to |
1631 | * all the mounts belonging to the destination mount's propagation tree. | |
1632 | * the mount is marked as 'shared and slave'. | |
1633 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1634 | * |
1635 | * if the source mount is a tree, the operations explained above is | |
1636 | * applied to each mount in the tree. | |
1637 | * Must be called without spinlocks held, since this function can sleep | |
1638 | * in allocations. | |
1639 | */ | |
0fb54e50 | 1640 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1641 | struct mount *dest_mnt, |
1642 | struct mountpoint *dest_mp, | |
1643 | struct path *parent_path) | |
b90fa9ae | 1644 | { |
38129a13 | 1645 | HLIST_HEAD(tree_list); |
315fc83e | 1646 | struct mount *child, *p; |
38129a13 | 1647 | struct hlist_node *n; |
719f5d7f | 1648 | int err; |
b90fa9ae | 1649 | |
fc7be130 | 1650 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1651 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1652 | if (err) |
1653 | goto out; | |
0b1b901b | 1654 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1655 | lock_mount_hash(); |
0b1b901b AV |
1656 | if (err) |
1657 | goto out_cleanup_ids; | |
909b0a88 | 1658 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1659 | set_mnt_shared(p); |
0b1b901b AV |
1660 | } else { |
1661 | lock_mount_hash(); | |
b90fa9ae | 1662 | } |
1a390689 | 1663 | if (parent_path) { |
0fb54e50 | 1664 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1665 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1666 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1667 | } else { |
84d17192 | 1668 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1669 | commit_tree(source_mnt, NULL); |
21444403 | 1670 | } |
b90fa9ae | 1671 | |
38129a13 | 1672 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1673 | struct mount *q; |
38129a13 | 1674 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1675 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1676 | child->mnt_mountpoint); | |
1677 | commit_tree(child, q); | |
b90fa9ae | 1678 | } |
719ea2fb | 1679 | unlock_mount_hash(); |
99b7db7b | 1680 | |
b90fa9ae | 1681 | return 0; |
719f5d7f MS |
1682 | |
1683 | out_cleanup_ids: | |
f2ebb3a9 AV |
1684 | while (!hlist_empty(&tree_list)) { |
1685 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1686 | umount_tree(child, 0); | |
1687 | } | |
1688 | unlock_mount_hash(); | |
0b1b901b | 1689 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1690 | out: |
1691 | return err; | |
b90fa9ae RP |
1692 | } |
1693 | ||
84d17192 | 1694 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1695 | { |
1696 | struct vfsmount *mnt; | |
84d17192 | 1697 | struct dentry *dentry = path->dentry; |
b12cea91 | 1698 | retry: |
84d17192 AV |
1699 | mutex_lock(&dentry->d_inode->i_mutex); |
1700 | if (unlikely(cant_mount(dentry))) { | |
1701 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1702 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1703 | } |
97216be0 | 1704 | namespace_lock(); |
b12cea91 | 1705 | mnt = lookup_mnt(path); |
84d17192 AV |
1706 | if (likely(!mnt)) { |
1707 | struct mountpoint *mp = new_mountpoint(dentry); | |
1708 | if (IS_ERR(mp)) { | |
97216be0 | 1709 | namespace_unlock(); |
84d17192 AV |
1710 | mutex_unlock(&dentry->d_inode->i_mutex); |
1711 | return mp; | |
1712 | } | |
1713 | return mp; | |
1714 | } | |
97216be0 | 1715 | namespace_unlock(); |
b12cea91 AV |
1716 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1717 | path_put(path); | |
1718 | path->mnt = mnt; | |
84d17192 | 1719 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1720 | goto retry; |
1721 | } | |
1722 | ||
84d17192 | 1723 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1724 | { |
84d17192 AV |
1725 | struct dentry *dentry = where->m_dentry; |
1726 | put_mountpoint(where); | |
328e6d90 | 1727 | namespace_unlock(); |
84d17192 | 1728 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1729 | } |
1730 | ||
84d17192 | 1731 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1732 | { |
95bc5f25 | 1733 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1734 | return -EINVAL; |
1735 | ||
84d17192 | 1736 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1737 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1738 | return -ENOTDIR; |
1739 | ||
84d17192 | 1740 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1741 | } |
1742 | ||
7a2e8a8f VA |
1743 | /* |
1744 | * Sanity check the flags to change_mnt_propagation. | |
1745 | */ | |
1746 | ||
1747 | static int flags_to_propagation_type(int flags) | |
1748 | { | |
7c6e984d | 1749 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1750 | |
1751 | /* Fail if any non-propagation flags are set */ | |
1752 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1753 | return 0; | |
1754 | /* Only one propagation flag should be set */ | |
1755 | if (!is_power_of_2(type)) | |
1756 | return 0; | |
1757 | return type; | |
1758 | } | |
1759 | ||
07b20889 RP |
1760 | /* |
1761 | * recursively change the type of the mountpoint. | |
1762 | */ | |
0a0d8a46 | 1763 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1764 | { |
315fc83e | 1765 | struct mount *m; |
4b8b21f4 | 1766 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1767 | int recurse = flag & MS_REC; |
7a2e8a8f | 1768 | int type; |
719f5d7f | 1769 | int err = 0; |
07b20889 | 1770 | |
2d92ab3c | 1771 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1772 | return -EINVAL; |
1773 | ||
7a2e8a8f VA |
1774 | type = flags_to_propagation_type(flag); |
1775 | if (!type) | |
1776 | return -EINVAL; | |
1777 | ||
97216be0 | 1778 | namespace_lock(); |
719f5d7f MS |
1779 | if (type == MS_SHARED) { |
1780 | err = invent_group_ids(mnt, recurse); | |
1781 | if (err) | |
1782 | goto out_unlock; | |
1783 | } | |
1784 | ||
719ea2fb | 1785 | lock_mount_hash(); |
909b0a88 | 1786 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1787 | change_mnt_propagation(m, type); |
719ea2fb | 1788 | unlock_mount_hash(); |
719f5d7f MS |
1789 | |
1790 | out_unlock: | |
97216be0 | 1791 | namespace_unlock(); |
719f5d7f | 1792 | return err; |
07b20889 RP |
1793 | } |
1794 | ||
5ff9d8a6 EB |
1795 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1796 | { | |
1797 | struct mount *child; | |
1798 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1799 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1800 | continue; | |
1801 | ||
1802 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1803 | return true; | |
1804 | } | |
1805 | return false; | |
1806 | } | |
1807 | ||
1da177e4 LT |
1808 | /* |
1809 | * do loopback mount. | |
1810 | */ | |
808d4e3c | 1811 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1812 | int recurse) |
1da177e4 | 1813 | { |
2d92ab3c | 1814 | struct path old_path; |
84d17192 AV |
1815 | struct mount *mnt = NULL, *old, *parent; |
1816 | struct mountpoint *mp; | |
57eccb83 | 1817 | int err; |
1da177e4 LT |
1818 | if (!old_name || !*old_name) |
1819 | return -EINVAL; | |
815d405c | 1820 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1821 | if (err) |
1822 | return err; | |
1823 | ||
8823c079 | 1824 | err = -EINVAL; |
4ce5d2b1 | 1825 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1826 | goto out; |
1827 | ||
84d17192 AV |
1828 | mp = lock_mount(path); |
1829 | err = PTR_ERR(mp); | |
1830 | if (IS_ERR(mp)) | |
b12cea91 AV |
1831 | goto out; |
1832 | ||
87129cc0 | 1833 | old = real_mount(old_path.mnt); |
84d17192 | 1834 | parent = real_mount(path->mnt); |
87129cc0 | 1835 | |
1da177e4 | 1836 | err = -EINVAL; |
fc7be130 | 1837 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1838 | goto out2; |
9676f0c6 | 1839 | |
84d17192 | 1840 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1841 | goto out2; |
1da177e4 | 1842 | |
5ff9d8a6 EB |
1843 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1844 | goto out2; | |
1845 | ||
ccd48bc7 | 1846 | if (recurse) |
4ce5d2b1 | 1847 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1848 | else |
87129cc0 | 1849 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1850 | |
be34d1a3 DH |
1851 | if (IS_ERR(mnt)) { |
1852 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1853 | goto out2; |
be34d1a3 | 1854 | } |
ccd48bc7 | 1855 | |
5ff9d8a6 EB |
1856 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1857 | ||
84d17192 | 1858 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1859 | if (err) { |
719ea2fb | 1860 | lock_mount_hash(); |
328e6d90 | 1861 | umount_tree(mnt, 0); |
719ea2fb | 1862 | unlock_mount_hash(); |
5b83d2c5 | 1863 | } |
b12cea91 | 1864 | out2: |
84d17192 | 1865 | unlock_mount(mp); |
ccd48bc7 | 1866 | out: |
2d92ab3c | 1867 | path_put(&old_path); |
1da177e4 LT |
1868 | return err; |
1869 | } | |
1870 | ||
2e4b7fcd DH |
1871 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1872 | { | |
1873 | int error = 0; | |
1874 | int readonly_request = 0; | |
1875 | ||
1876 | if (ms_flags & MS_RDONLY) | |
1877 | readonly_request = 1; | |
1878 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1879 | return 0; | |
1880 | ||
90563b19 EB |
1881 | if (mnt->mnt_flags & MNT_LOCK_READONLY) |
1882 | return -EPERM; | |
1883 | ||
2e4b7fcd | 1884 | if (readonly_request) |
83adc753 | 1885 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1886 | else |
83adc753 | 1887 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1888 | return error; |
1889 | } | |
1890 | ||
1da177e4 LT |
1891 | /* |
1892 | * change filesystem flags. dir should be a physical root of filesystem. | |
1893 | * If you've mounted a non-root directory somewhere and want to do remount | |
1894 | * on it - tough luck. | |
1895 | */ | |
0a0d8a46 | 1896 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1897 | void *data) |
1898 | { | |
1899 | int err; | |
2d92ab3c | 1900 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1901 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1902 | |
143c8c91 | 1903 | if (!check_mnt(mnt)) |
1da177e4 LT |
1904 | return -EINVAL; |
1905 | ||
2d92ab3c | 1906 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1907 | return -EINVAL; |
1908 | ||
ff36fe2c EP |
1909 | err = security_sb_remount(sb, data); |
1910 | if (err) | |
1911 | return err; | |
1912 | ||
1da177e4 | 1913 | down_write(&sb->s_umount); |
2e4b7fcd | 1914 | if (flags & MS_BIND) |
2d92ab3c | 1915 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1916 | else if (!capable(CAP_SYS_ADMIN)) |
1917 | err = -EPERM; | |
4aa98cf7 | 1918 | else |
2e4b7fcd | 1919 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1920 | if (!err) { |
719ea2fb | 1921 | lock_mount_hash(); |
143c8c91 AV |
1922 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1923 | mnt->mnt.mnt_flags = mnt_flags; | |
143c8c91 | 1924 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 1925 | unlock_mount_hash(); |
0e55a7cc | 1926 | } |
6339dab8 | 1927 | up_write(&sb->s_umount); |
1da177e4 LT |
1928 | return err; |
1929 | } | |
1930 | ||
cbbe362c | 1931 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1932 | { |
315fc83e | 1933 | struct mount *p; |
909b0a88 | 1934 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1935 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1936 | return 1; |
1937 | } | |
1938 | return 0; | |
1939 | } | |
1940 | ||
808d4e3c | 1941 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1942 | { |
2d92ab3c | 1943 | struct path old_path, parent_path; |
676da58d | 1944 | struct mount *p; |
0fb54e50 | 1945 | struct mount *old; |
84d17192 | 1946 | struct mountpoint *mp; |
57eccb83 | 1947 | int err; |
1da177e4 LT |
1948 | if (!old_name || !*old_name) |
1949 | return -EINVAL; | |
2d92ab3c | 1950 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1951 | if (err) |
1952 | return err; | |
1953 | ||
84d17192 AV |
1954 | mp = lock_mount(path); |
1955 | err = PTR_ERR(mp); | |
1956 | if (IS_ERR(mp)) | |
cc53ce53 DH |
1957 | goto out; |
1958 | ||
143c8c91 | 1959 | old = real_mount(old_path.mnt); |
fc7be130 | 1960 | p = real_mount(path->mnt); |
143c8c91 | 1961 | |
1da177e4 | 1962 | err = -EINVAL; |
fc7be130 | 1963 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1964 | goto out1; |
1965 | ||
5ff9d8a6 EB |
1966 | if (old->mnt.mnt_flags & MNT_LOCKED) |
1967 | goto out1; | |
1968 | ||
1da177e4 | 1969 | err = -EINVAL; |
2d92ab3c | 1970 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1971 | goto out1; |
1da177e4 | 1972 | |
676da58d | 1973 | if (!mnt_has_parent(old)) |
21444403 | 1974 | goto out1; |
1da177e4 | 1975 | |
2d92ab3c AV |
1976 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1977 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1978 | goto out1; |
1979 | /* | |
1980 | * Don't move a mount residing in a shared parent. | |
1981 | */ | |
fc7be130 | 1982 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1983 | goto out1; |
9676f0c6 RP |
1984 | /* |
1985 | * Don't move a mount tree containing unbindable mounts to a destination | |
1986 | * mount which is shared. | |
1987 | */ | |
fc7be130 | 1988 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1989 | goto out1; |
1da177e4 | 1990 | err = -ELOOP; |
fc7be130 | 1991 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1992 | if (p == old) |
21444403 | 1993 | goto out1; |
1da177e4 | 1994 | |
84d17192 | 1995 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 1996 | if (err) |
21444403 | 1997 | goto out1; |
1da177e4 LT |
1998 | |
1999 | /* if the mount is moved, it should no longer be expire | |
2000 | * automatically */ | |
6776db3d | 2001 | list_del_init(&old->mnt_expire); |
1da177e4 | 2002 | out1: |
84d17192 | 2003 | unlock_mount(mp); |
1da177e4 | 2004 | out: |
1da177e4 | 2005 | if (!err) |
1a390689 | 2006 | path_put(&parent_path); |
2d92ab3c | 2007 | path_put(&old_path); |
1da177e4 LT |
2008 | return err; |
2009 | } | |
2010 | ||
9d412a43 AV |
2011 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2012 | { | |
2013 | int err; | |
2014 | const char *subtype = strchr(fstype, '.'); | |
2015 | if (subtype) { | |
2016 | subtype++; | |
2017 | err = -EINVAL; | |
2018 | if (!subtype[0]) | |
2019 | goto err; | |
2020 | } else | |
2021 | subtype = ""; | |
2022 | ||
2023 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2024 | err = -ENOMEM; | |
2025 | if (!mnt->mnt_sb->s_subtype) | |
2026 | goto err; | |
2027 | return mnt; | |
2028 | ||
2029 | err: | |
2030 | mntput(mnt); | |
2031 | return ERR_PTR(err); | |
2032 | } | |
2033 | ||
9d412a43 AV |
2034 | /* |
2035 | * add a mount into a namespace's mount tree | |
2036 | */ | |
95bc5f25 | 2037 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2038 | { |
84d17192 AV |
2039 | struct mountpoint *mp; |
2040 | struct mount *parent; | |
9d412a43 AV |
2041 | int err; |
2042 | ||
f2ebb3a9 | 2043 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2044 | |
84d17192 AV |
2045 | mp = lock_mount(path); |
2046 | if (IS_ERR(mp)) | |
2047 | return PTR_ERR(mp); | |
9d412a43 | 2048 | |
84d17192 | 2049 | parent = real_mount(path->mnt); |
9d412a43 | 2050 | err = -EINVAL; |
84d17192 | 2051 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2052 | /* that's acceptable only for automounts done in private ns */ |
2053 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2054 | goto unlock; | |
2055 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2056 | if (!parent->mnt_ns) |
156cacb1 AV |
2057 | goto unlock; |
2058 | } | |
9d412a43 AV |
2059 | |
2060 | /* Refuse the same filesystem on the same mount point */ | |
2061 | err = -EBUSY; | |
95bc5f25 | 2062 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2063 | path->mnt->mnt_root == path->dentry) |
2064 | goto unlock; | |
2065 | ||
2066 | err = -EINVAL; | |
95bc5f25 | 2067 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2068 | goto unlock; |
2069 | ||
95bc5f25 | 2070 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2071 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2072 | |
2073 | unlock: | |
84d17192 | 2074 | unlock_mount(mp); |
9d412a43 AV |
2075 | return err; |
2076 | } | |
b1e75df4 | 2077 | |
1da177e4 LT |
2078 | /* |
2079 | * create a new mount for userspace and request it to be added into the | |
2080 | * namespace's tree | |
2081 | */ | |
0c55cfc4 | 2082 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2083 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2084 | { |
0c55cfc4 | 2085 | struct file_system_type *type; |
9b40bc90 | 2086 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2087 | struct vfsmount *mnt; |
15f9a3f3 | 2088 | int err; |
1da177e4 | 2089 | |
0c55cfc4 | 2090 | if (!fstype) |
1da177e4 LT |
2091 | return -EINVAL; |
2092 | ||
0c55cfc4 EB |
2093 | type = get_fs_type(fstype); |
2094 | if (!type) | |
2095 | return -ENODEV; | |
2096 | ||
2097 | if (user_ns != &init_user_ns) { | |
2098 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2099 | put_filesystem(type); | |
2100 | return -EPERM; | |
2101 | } | |
2102 | /* Only in special cases allow devices from mounts | |
2103 | * created outside the initial user namespace. | |
2104 | */ | |
2105 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2106 | flags |= MS_NODEV; | |
2107 | mnt_flags |= MNT_NODEV; | |
2108 | } | |
2109 | } | |
2110 | ||
2111 | mnt = vfs_kern_mount(type, flags, name, data); | |
2112 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2113 | !mnt->mnt_sb->s_subtype) | |
2114 | mnt = fs_set_subtype(mnt, fstype); | |
2115 | ||
2116 | put_filesystem(type); | |
1da177e4 LT |
2117 | if (IS_ERR(mnt)) |
2118 | return PTR_ERR(mnt); | |
2119 | ||
95bc5f25 | 2120 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2121 | if (err) |
2122 | mntput(mnt); | |
2123 | return err; | |
1da177e4 LT |
2124 | } |
2125 | ||
19a167af AV |
2126 | int finish_automount(struct vfsmount *m, struct path *path) |
2127 | { | |
6776db3d | 2128 | struct mount *mnt = real_mount(m); |
19a167af AV |
2129 | int err; |
2130 | /* The new mount record should have at least 2 refs to prevent it being | |
2131 | * expired before we get a chance to add it | |
2132 | */ | |
6776db3d | 2133 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2134 | |
2135 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2136 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2137 | err = -ELOOP; |
2138 | goto fail; | |
19a167af AV |
2139 | } |
2140 | ||
95bc5f25 | 2141 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2142 | if (!err) |
2143 | return 0; | |
2144 | fail: | |
2145 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2146 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2147 | namespace_lock(); |
6776db3d | 2148 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2149 | namespace_unlock(); |
19a167af | 2150 | } |
b1e75df4 AV |
2151 | mntput(m); |
2152 | mntput(m); | |
19a167af AV |
2153 | return err; |
2154 | } | |
2155 | ||
ea5b778a DH |
2156 | /** |
2157 | * mnt_set_expiry - Put a mount on an expiration list | |
2158 | * @mnt: The mount to list. | |
2159 | * @expiry_list: The list to add the mount to. | |
2160 | */ | |
2161 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2162 | { | |
97216be0 | 2163 | namespace_lock(); |
ea5b778a | 2164 | |
6776db3d | 2165 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2166 | |
97216be0 | 2167 | namespace_unlock(); |
ea5b778a DH |
2168 | } |
2169 | EXPORT_SYMBOL(mnt_set_expiry); | |
2170 | ||
1da177e4 LT |
2171 | /* |
2172 | * process a list of expirable mountpoints with the intent of discarding any | |
2173 | * mountpoints that aren't in use and haven't been touched since last we came | |
2174 | * here | |
2175 | */ | |
2176 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2177 | { | |
761d5c38 | 2178 | struct mount *mnt, *next; |
1da177e4 LT |
2179 | LIST_HEAD(graveyard); |
2180 | ||
2181 | if (list_empty(mounts)) | |
2182 | return; | |
2183 | ||
97216be0 | 2184 | namespace_lock(); |
719ea2fb | 2185 | lock_mount_hash(); |
1da177e4 LT |
2186 | |
2187 | /* extract from the expiration list every vfsmount that matches the | |
2188 | * following criteria: | |
2189 | * - only referenced by its parent vfsmount | |
2190 | * - still marked for expiry (marked on the last call here; marks are | |
2191 | * cleared by mntput()) | |
2192 | */ | |
6776db3d | 2193 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2194 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2195 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2196 | continue; |
6776db3d | 2197 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2198 | } |
bcc5c7d2 | 2199 | while (!list_empty(&graveyard)) { |
6776db3d | 2200 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2201 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2202 | umount_tree(mnt, 1); |
bcc5c7d2 | 2203 | } |
719ea2fb | 2204 | unlock_mount_hash(); |
3ab6abee | 2205 | namespace_unlock(); |
5528f911 TM |
2206 | } |
2207 | ||
2208 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2209 | ||
2210 | /* | |
2211 | * Ripoff of 'select_parent()' | |
2212 | * | |
2213 | * search the list of submounts for a given mountpoint, and move any | |
2214 | * shrinkable submounts to the 'graveyard' list. | |
2215 | */ | |
692afc31 | 2216 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2217 | { |
692afc31 | 2218 | struct mount *this_parent = parent; |
5528f911 TM |
2219 | struct list_head *next; |
2220 | int found = 0; | |
2221 | ||
2222 | repeat: | |
6b41d536 | 2223 | next = this_parent->mnt_mounts.next; |
5528f911 | 2224 | resume: |
6b41d536 | 2225 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2226 | struct list_head *tmp = next; |
6b41d536 | 2227 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2228 | |
2229 | next = tmp->next; | |
692afc31 | 2230 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2231 | continue; |
5528f911 TM |
2232 | /* |
2233 | * Descend a level if the d_mounts list is non-empty. | |
2234 | */ | |
6b41d536 | 2235 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2236 | this_parent = mnt; |
2237 | goto repeat; | |
2238 | } | |
1da177e4 | 2239 | |
1ab59738 | 2240 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2241 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2242 | found++; |
2243 | } | |
1da177e4 | 2244 | } |
5528f911 TM |
2245 | /* |
2246 | * All done at this level ... ascend and resume the search | |
2247 | */ | |
2248 | if (this_parent != parent) { | |
6b41d536 | 2249 | next = this_parent->mnt_child.next; |
0714a533 | 2250 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2251 | goto resume; |
2252 | } | |
2253 | return found; | |
2254 | } | |
2255 | ||
2256 | /* | |
2257 | * process a list of expirable mountpoints with the intent of discarding any | |
2258 | * submounts of a specific parent mountpoint | |
99b7db7b | 2259 | * |
48a066e7 | 2260 | * mount_lock must be held for write |
5528f911 | 2261 | */ |
b54b9be7 | 2262 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2263 | { |
2264 | LIST_HEAD(graveyard); | |
761d5c38 | 2265 | struct mount *m; |
5528f911 | 2266 | |
5528f911 | 2267 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2268 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2269 | while (!list_empty(&graveyard)) { |
761d5c38 | 2270 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2271 | mnt_expire); |
143c8c91 | 2272 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2273 | umount_tree(m, 1); |
bcc5c7d2 AV |
2274 | } |
2275 | } | |
1da177e4 LT |
2276 | } |
2277 | ||
1da177e4 LT |
2278 | /* |
2279 | * Some copy_from_user() implementations do not return the exact number of | |
2280 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2281 | * Note that this function differs from copy_from_user() in that it will oops | |
2282 | * on bad values of `to', rather than returning a short copy. | |
2283 | */ | |
b58fed8b RP |
2284 | static long exact_copy_from_user(void *to, const void __user * from, |
2285 | unsigned long n) | |
1da177e4 LT |
2286 | { |
2287 | char *t = to; | |
2288 | const char __user *f = from; | |
2289 | char c; | |
2290 | ||
2291 | if (!access_ok(VERIFY_READ, from, n)) | |
2292 | return n; | |
2293 | ||
2294 | while (n) { | |
2295 | if (__get_user(c, f)) { | |
2296 | memset(t, 0, n); | |
2297 | break; | |
2298 | } | |
2299 | *t++ = c; | |
2300 | f++; | |
2301 | n--; | |
2302 | } | |
2303 | return n; | |
2304 | } | |
2305 | ||
b58fed8b | 2306 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2307 | { |
2308 | int i; | |
2309 | unsigned long page; | |
2310 | unsigned long size; | |
b58fed8b | 2311 | |
1da177e4 LT |
2312 | *where = 0; |
2313 | if (!data) | |
2314 | return 0; | |
2315 | ||
2316 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2317 | return -ENOMEM; | |
2318 | ||
2319 | /* We only care that *some* data at the address the user | |
2320 | * gave us is valid. Just in case, we'll zero | |
2321 | * the remainder of the page. | |
2322 | */ | |
2323 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2324 | size = TASK_SIZE - (unsigned long)data; | |
2325 | if (size > PAGE_SIZE) | |
2326 | size = PAGE_SIZE; | |
2327 | ||
2328 | i = size - exact_copy_from_user((void *)page, data, size); | |
2329 | if (!i) { | |
b58fed8b | 2330 | free_page(page); |
1da177e4 LT |
2331 | return -EFAULT; |
2332 | } | |
2333 | if (i != PAGE_SIZE) | |
2334 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2335 | *where = page; | |
2336 | return 0; | |
2337 | } | |
2338 | ||
eca6f534 VN |
2339 | int copy_mount_string(const void __user *data, char **where) |
2340 | { | |
2341 | char *tmp; | |
2342 | ||
2343 | if (!data) { | |
2344 | *where = NULL; | |
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | tmp = strndup_user(data, PAGE_SIZE); | |
2349 | if (IS_ERR(tmp)) | |
2350 | return PTR_ERR(tmp); | |
2351 | ||
2352 | *where = tmp; | |
2353 | return 0; | |
2354 | } | |
2355 | ||
1da177e4 LT |
2356 | /* |
2357 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2358 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2359 | * | |
2360 | * data is a (void *) that can point to any structure up to | |
2361 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2362 | * information (or be NULL). | |
2363 | * | |
2364 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2365 | * When the flags word was introduced its top half was required | |
2366 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2367 | * Therefore, if this magic number is present, it carries no information | |
2368 | * and must be discarded. | |
2369 | */ | |
808d4e3c AV |
2370 | long do_mount(const char *dev_name, const char *dir_name, |
2371 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2372 | { |
2d92ab3c | 2373 | struct path path; |
1da177e4 LT |
2374 | int retval = 0; |
2375 | int mnt_flags = 0; | |
2376 | ||
2377 | /* Discard magic */ | |
2378 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2379 | flags &= ~MS_MGC_MSK; | |
2380 | ||
2381 | /* Basic sanity checks */ | |
2382 | ||
2383 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2384 | return -EINVAL; | |
1da177e4 LT |
2385 | |
2386 | if (data_page) | |
2387 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2388 | ||
a27ab9f2 TH |
2389 | /* ... and get the mountpoint */ |
2390 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2391 | if (retval) | |
2392 | return retval; | |
2393 | ||
2394 | retval = security_sb_mount(dev_name, &path, | |
2395 | type_page, flags, data_page); | |
0d5cadb8 AV |
2396 | if (!retval && !may_mount()) |
2397 | retval = -EPERM; | |
a27ab9f2 TH |
2398 | if (retval) |
2399 | goto dput_out; | |
2400 | ||
613cbe3d AK |
2401 | /* Default to relatime unless overriden */ |
2402 | if (!(flags & MS_NOATIME)) | |
2403 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2404 | |
1da177e4 LT |
2405 | /* Separate the per-mountpoint flags */ |
2406 | if (flags & MS_NOSUID) | |
2407 | mnt_flags |= MNT_NOSUID; | |
2408 | if (flags & MS_NODEV) | |
2409 | mnt_flags |= MNT_NODEV; | |
2410 | if (flags & MS_NOEXEC) | |
2411 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2412 | if (flags & MS_NOATIME) |
2413 | mnt_flags |= MNT_NOATIME; | |
2414 | if (flags & MS_NODIRATIME) | |
2415 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2416 | if (flags & MS_STRICTATIME) |
2417 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2418 | if (flags & MS_RDONLY) |
2419 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2420 | |
7a4dec53 | 2421 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2422 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2423 | MS_STRICTATIME); | |
1da177e4 | 2424 | |
1da177e4 | 2425 | if (flags & MS_REMOUNT) |
2d92ab3c | 2426 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2427 | data_page); |
2428 | else if (flags & MS_BIND) | |
2d92ab3c | 2429 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2430 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2431 | retval = do_change_type(&path, flags); |
1da177e4 | 2432 | else if (flags & MS_MOVE) |
2d92ab3c | 2433 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2434 | else |
2d92ab3c | 2435 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2436 | dev_name, data_page); |
2437 | dput_out: | |
2d92ab3c | 2438 | path_put(&path); |
1da177e4 LT |
2439 | return retval; |
2440 | } | |
2441 | ||
771b1371 EB |
2442 | static void free_mnt_ns(struct mnt_namespace *ns) |
2443 | { | |
98f842e6 | 2444 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2445 | put_user_ns(ns->user_ns); |
2446 | kfree(ns); | |
2447 | } | |
2448 | ||
8823c079 EB |
2449 | /* |
2450 | * Assign a sequence number so we can detect when we attempt to bind | |
2451 | * mount a reference to an older mount namespace into the current | |
2452 | * mount namespace, preventing reference counting loops. A 64bit | |
2453 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2454 | * is effectively never, so we can ignore the possibility. | |
2455 | */ | |
2456 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2457 | ||
771b1371 | 2458 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2459 | { |
2460 | struct mnt_namespace *new_ns; | |
98f842e6 | 2461 | int ret; |
cf8d2c11 TM |
2462 | |
2463 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2464 | if (!new_ns) | |
2465 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2466 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2467 | if (ret) { | |
2468 | kfree(new_ns); | |
2469 | return ERR_PTR(ret); | |
2470 | } | |
8823c079 | 2471 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2472 | atomic_set(&new_ns->count, 1); |
2473 | new_ns->root = NULL; | |
2474 | INIT_LIST_HEAD(&new_ns->list); | |
2475 | init_waitqueue_head(&new_ns->poll); | |
2476 | new_ns->event = 0; | |
771b1371 | 2477 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2478 | return new_ns; |
2479 | } | |
2480 | ||
9559f689 AV |
2481 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2482 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2483 | { |
6b3286ed | 2484 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2485 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2486 | struct mount *p, *q; |
9559f689 | 2487 | struct mount *old; |
cb338d06 | 2488 | struct mount *new; |
7a472ef4 | 2489 | int copy_flags; |
1da177e4 | 2490 | |
9559f689 AV |
2491 | BUG_ON(!ns); |
2492 | ||
2493 | if (likely(!(flags & CLONE_NEWNS))) { | |
2494 | get_mnt_ns(ns); | |
2495 | return ns; | |
2496 | } | |
2497 | ||
2498 | old = ns->root; | |
2499 | ||
771b1371 | 2500 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2501 | if (IS_ERR(new_ns)) |
2502 | return new_ns; | |
1da177e4 | 2503 | |
97216be0 | 2504 | namespace_lock(); |
1da177e4 | 2505 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2506 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2507 | if (user_ns != ns->user_ns) |
132c94e3 | 2508 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2509 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2510 | if (IS_ERR(new)) { |
328e6d90 | 2511 | namespace_unlock(); |
771b1371 | 2512 | free_mnt_ns(new_ns); |
be34d1a3 | 2513 | return ERR_CAST(new); |
1da177e4 | 2514 | } |
be08d6d2 | 2515 | new_ns->root = new; |
1a4eeaf2 | 2516 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2517 | |
2518 | /* | |
2519 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2520 | * as belonging to new namespace. We have already acquired a private | |
2521 | * fs_struct, so tsk->fs->lock is not needed. | |
2522 | */ | |
909b0a88 | 2523 | p = old; |
cb338d06 | 2524 | q = new; |
1da177e4 | 2525 | while (p) { |
143c8c91 | 2526 | q->mnt_ns = new_ns; |
9559f689 AV |
2527 | if (new_fs) { |
2528 | if (&p->mnt == new_fs->root.mnt) { | |
2529 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2530 | rootmnt = &p->mnt; |
1da177e4 | 2531 | } |
9559f689 AV |
2532 | if (&p->mnt == new_fs->pwd.mnt) { |
2533 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2534 | pwdmnt = &p->mnt; |
1da177e4 | 2535 | } |
1da177e4 | 2536 | } |
909b0a88 AV |
2537 | p = next_mnt(p, old); |
2538 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2539 | if (!q) |
2540 | break; | |
2541 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2542 | p = next_mnt(p, old); | |
1da177e4 | 2543 | } |
328e6d90 | 2544 | namespace_unlock(); |
1da177e4 | 2545 | |
1da177e4 | 2546 | if (rootmnt) |
f03c6599 | 2547 | mntput(rootmnt); |
1da177e4 | 2548 | if (pwdmnt) |
f03c6599 | 2549 | mntput(pwdmnt); |
1da177e4 | 2550 | |
741a2951 | 2551 | return new_ns; |
1da177e4 LT |
2552 | } |
2553 | ||
cf8d2c11 TM |
2554 | /** |
2555 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2556 | * @mnt: pointer to the new root filesystem mountpoint | |
2557 | */ | |
1a4eeaf2 | 2558 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2559 | { |
771b1371 | 2560 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2561 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2562 | struct mount *mnt = real_mount(m); |
2563 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2564 | new_ns->root = mnt; |
b1983cd8 | 2565 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2566 | } else { |
1a4eeaf2 | 2567 | mntput(m); |
cf8d2c11 TM |
2568 | } |
2569 | return new_ns; | |
2570 | } | |
cf8d2c11 | 2571 | |
ea441d11 AV |
2572 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2573 | { | |
2574 | struct mnt_namespace *ns; | |
d31da0f0 | 2575 | struct super_block *s; |
ea441d11 AV |
2576 | struct path path; |
2577 | int err; | |
2578 | ||
2579 | ns = create_mnt_ns(mnt); | |
2580 | if (IS_ERR(ns)) | |
2581 | return ERR_CAST(ns); | |
2582 | ||
2583 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2584 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2585 | ||
2586 | put_mnt_ns(ns); | |
2587 | ||
2588 | if (err) | |
2589 | return ERR_PTR(err); | |
2590 | ||
2591 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2592 | s = path.mnt->mnt_sb; |
2593 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2594 | mntput(path.mnt); |
2595 | /* lock the sucker */ | |
d31da0f0 | 2596 | down_write(&s->s_umount); |
ea441d11 AV |
2597 | /* ... and return the root of (sub)tree on it */ |
2598 | return path.dentry; | |
2599 | } | |
2600 | EXPORT_SYMBOL(mount_subtree); | |
2601 | ||
bdc480e3 HC |
2602 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2603 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2604 | { |
eca6f534 VN |
2605 | int ret; |
2606 | char *kernel_type; | |
91a27b2a | 2607 | struct filename *kernel_dir; |
eca6f534 | 2608 | char *kernel_dev; |
1da177e4 | 2609 | unsigned long data_page; |
1da177e4 | 2610 | |
eca6f534 VN |
2611 | ret = copy_mount_string(type, &kernel_type); |
2612 | if (ret < 0) | |
2613 | goto out_type; | |
1da177e4 | 2614 | |
eca6f534 VN |
2615 | kernel_dir = getname(dir_name); |
2616 | if (IS_ERR(kernel_dir)) { | |
2617 | ret = PTR_ERR(kernel_dir); | |
2618 | goto out_dir; | |
2619 | } | |
1da177e4 | 2620 | |
eca6f534 VN |
2621 | ret = copy_mount_string(dev_name, &kernel_dev); |
2622 | if (ret < 0) | |
2623 | goto out_dev; | |
1da177e4 | 2624 | |
eca6f534 VN |
2625 | ret = copy_mount_options(data, &data_page); |
2626 | if (ret < 0) | |
2627 | goto out_data; | |
1da177e4 | 2628 | |
91a27b2a | 2629 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2630 | (void *) data_page); |
1da177e4 | 2631 | |
eca6f534 VN |
2632 | free_page(data_page); |
2633 | out_data: | |
2634 | kfree(kernel_dev); | |
2635 | out_dev: | |
2636 | putname(kernel_dir); | |
2637 | out_dir: | |
2638 | kfree(kernel_type); | |
2639 | out_type: | |
2640 | return ret; | |
1da177e4 LT |
2641 | } |
2642 | ||
afac7cba AV |
2643 | /* |
2644 | * Return true if path is reachable from root | |
2645 | * | |
48a066e7 | 2646 | * namespace_sem or mount_lock is held |
afac7cba | 2647 | */ |
643822b4 | 2648 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2649 | const struct path *root) |
2650 | { | |
643822b4 | 2651 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2652 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2653 | mnt = mnt->mnt_parent; |
afac7cba | 2654 | } |
643822b4 | 2655 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2656 | } |
2657 | ||
2658 | int path_is_under(struct path *path1, struct path *path2) | |
2659 | { | |
2660 | int res; | |
48a066e7 | 2661 | read_seqlock_excl(&mount_lock); |
643822b4 | 2662 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2663 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2664 | return res; |
2665 | } | |
2666 | EXPORT_SYMBOL(path_is_under); | |
2667 | ||
1da177e4 LT |
2668 | /* |
2669 | * pivot_root Semantics: | |
2670 | * Moves the root file system of the current process to the directory put_old, | |
2671 | * makes new_root as the new root file system of the current process, and sets | |
2672 | * root/cwd of all processes which had them on the current root to new_root. | |
2673 | * | |
2674 | * Restrictions: | |
2675 | * The new_root and put_old must be directories, and must not be on the | |
2676 | * same file system as the current process root. The put_old must be | |
2677 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2678 | * pointed to by put_old must yield the same directory as new_root. No other | |
2679 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2680 | * | |
4a0d11fa NB |
2681 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2682 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2683 | * in this situation. | |
2684 | * | |
1da177e4 LT |
2685 | * Notes: |
2686 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2687 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2688 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2689 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2690 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2691 | * first. | |
2692 | */ | |
3480b257 HC |
2693 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2694 | const char __user *, put_old) | |
1da177e4 | 2695 | { |
2d8f3038 | 2696 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2697 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2698 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2699 | int error; |
2700 | ||
9b40bc90 | 2701 | if (!may_mount()) |
1da177e4 LT |
2702 | return -EPERM; |
2703 | ||
2d8f3038 | 2704 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2705 | if (error) |
2706 | goto out0; | |
1da177e4 | 2707 | |
2d8f3038 | 2708 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2709 | if (error) |
2710 | goto out1; | |
2711 | ||
2d8f3038 | 2712 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2713 | if (error) |
2714 | goto out2; | |
1da177e4 | 2715 | |
f7ad3c6b | 2716 | get_fs_root(current->fs, &root); |
84d17192 AV |
2717 | old_mp = lock_mount(&old); |
2718 | error = PTR_ERR(old_mp); | |
2719 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2720 | goto out3; |
2721 | ||
1da177e4 | 2722 | error = -EINVAL; |
419148da AV |
2723 | new_mnt = real_mount(new.mnt); |
2724 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2725 | old_mnt = real_mount(old.mnt); |
2726 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2727 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2728 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2729 | goto out4; |
143c8c91 | 2730 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2731 | goto out4; |
5ff9d8a6 EB |
2732 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2733 | goto out4; | |
1da177e4 | 2734 | error = -ENOENT; |
f3da392e | 2735 | if (d_unlinked(new.dentry)) |
b12cea91 | 2736 | goto out4; |
1da177e4 | 2737 | error = -EBUSY; |
84d17192 | 2738 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2739 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2740 | error = -EINVAL; |
8c3ee42e | 2741 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2742 | goto out4; /* not a mountpoint */ |
676da58d | 2743 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2744 | goto out4; /* not attached */ |
84d17192 | 2745 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2746 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2747 | goto out4; /* not a mountpoint */ |
676da58d | 2748 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2749 | goto out4; /* not attached */ |
4ac91378 | 2750 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2751 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2752 | goto out4; |
84d17192 | 2753 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2754 | lock_mount_hash(); |
419148da AV |
2755 | detach_mnt(new_mnt, &parent_path); |
2756 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2757 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2758 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2759 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2760 | } | |
4ac91378 | 2761 | /* mount old root on put_old */ |
84d17192 | 2762 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2763 | /* mount new_root on / */ |
84d17192 | 2764 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2765 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2766 | unlock_mount_hash(); |
2d8f3038 | 2767 | chroot_fs_refs(&root, &new); |
84d17192 | 2768 | put_mountpoint(root_mp); |
1da177e4 | 2769 | error = 0; |
b12cea91 | 2770 | out4: |
84d17192 | 2771 | unlock_mount(old_mp); |
b12cea91 AV |
2772 | if (!error) { |
2773 | path_put(&root_parent); | |
2774 | path_put(&parent_path); | |
2775 | } | |
2776 | out3: | |
8c3ee42e | 2777 | path_put(&root); |
b12cea91 | 2778 | out2: |
2d8f3038 | 2779 | path_put(&old); |
1da177e4 | 2780 | out1: |
2d8f3038 | 2781 | path_put(&new); |
1da177e4 | 2782 | out0: |
1da177e4 | 2783 | return error; |
1da177e4 LT |
2784 | } |
2785 | ||
2786 | static void __init init_mount_tree(void) | |
2787 | { | |
2788 | struct vfsmount *mnt; | |
6b3286ed | 2789 | struct mnt_namespace *ns; |
ac748a09 | 2790 | struct path root; |
0c55cfc4 | 2791 | struct file_system_type *type; |
1da177e4 | 2792 | |
0c55cfc4 EB |
2793 | type = get_fs_type("rootfs"); |
2794 | if (!type) | |
2795 | panic("Can't find rootfs type"); | |
2796 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2797 | put_filesystem(type); | |
1da177e4 LT |
2798 | if (IS_ERR(mnt)) |
2799 | panic("Can't create rootfs"); | |
b3e19d92 | 2800 | |
3b22edc5 TM |
2801 | ns = create_mnt_ns(mnt); |
2802 | if (IS_ERR(ns)) | |
1da177e4 | 2803 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2804 | |
2805 | init_task.nsproxy->mnt_ns = ns; | |
2806 | get_mnt_ns(ns); | |
2807 | ||
be08d6d2 AV |
2808 | root.mnt = mnt; |
2809 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2810 | |
2811 | set_fs_pwd(current->fs, &root); | |
2812 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2813 | } |
2814 | ||
74bf17cf | 2815 | void __init mnt_init(void) |
1da177e4 | 2816 | { |
13f14b4d | 2817 | unsigned u; |
15a67dd8 | 2818 | int err; |
1da177e4 | 2819 | |
7d6fec45 | 2820 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2821 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2822 | |
0818bf27 | 2823 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2824 | sizeof(struct hlist_head), |
0818bf27 AV |
2825 | mhash_entries, 19, |
2826 | 0, | |
2827 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2828 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2829 | sizeof(struct hlist_head), | |
2830 | mphash_entries, 19, | |
2831 | 0, | |
2832 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2833 | |
84d17192 | 2834 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2835 | panic("Failed to allocate mount hash table\n"); |
2836 | ||
0818bf27 | 2837 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2838 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2839 | for (u = 0; u <= mp_hash_mask; u++) |
2840 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2841 | |
4b93dc9b TH |
2842 | kernfs_init(); |
2843 | ||
15a67dd8 RD |
2844 | err = sysfs_init(); |
2845 | if (err) | |
2846 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2847 | __func__, err); |
00d26666 GKH |
2848 | fs_kobj = kobject_create_and_add("fs", NULL); |
2849 | if (!fs_kobj) | |
8e24eea7 | 2850 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2851 | init_rootfs(); |
2852 | init_mount_tree(); | |
2853 | } | |
2854 | ||
616511d0 | 2855 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2856 | { |
d498b25a | 2857 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2858 | return; |
7b00ed6f | 2859 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2860 | free_mnt_ns(ns); |
1da177e4 | 2861 | } |
9d412a43 AV |
2862 | |
2863 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2864 | { | |
423e0ab0 TC |
2865 | struct vfsmount *mnt; |
2866 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2867 | if (!IS_ERR(mnt)) { | |
2868 | /* | |
2869 | * it is a longterm mount, don't release mnt until | |
2870 | * we unmount before file sys is unregistered | |
2871 | */ | |
f7a99c5b | 2872 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2873 | } |
2874 | return mnt; | |
9d412a43 AV |
2875 | } |
2876 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2877 | |
2878 | void kern_unmount(struct vfsmount *mnt) | |
2879 | { | |
2880 | /* release long term mount so mount point can be released */ | |
2881 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2882 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2883 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2884 | mntput(mnt); |
2885 | } | |
2886 | } | |
2887 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2888 | |
2889 | bool our_mnt(struct vfsmount *mnt) | |
2890 | { | |
143c8c91 | 2891 | return check_mnt(real_mount(mnt)); |
02125a82 | 2892 | } |
8823c079 | 2893 | |
3151527e EB |
2894 | bool current_chrooted(void) |
2895 | { | |
2896 | /* Does the current process have a non-standard root */ | |
2897 | struct path ns_root; | |
2898 | struct path fs_root; | |
2899 | bool chrooted; | |
2900 | ||
2901 | /* Find the namespace root */ | |
2902 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2903 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2904 | path_get(&ns_root); | |
2905 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2906 | ; | |
2907 | ||
2908 | get_fs_root(current->fs, &fs_root); | |
2909 | ||
2910 | chrooted = !path_equal(&fs_root, &ns_root); | |
2911 | ||
2912 | path_put(&fs_root); | |
2913 | path_put(&ns_root); | |
2914 | ||
2915 | return chrooted; | |
2916 | } | |
2917 | ||
e51db735 | 2918 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
2919 | { |
2920 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2921 | struct mount *mnt; | |
e51db735 | 2922 | bool visible = false; |
87a8ebd6 | 2923 | |
e51db735 EB |
2924 | if (unlikely(!ns)) |
2925 | return false; | |
2926 | ||
44bb4385 | 2927 | down_read(&namespace_sem); |
87a8ebd6 | 2928 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
2929 | struct mount *child; |
2930 | if (mnt->mnt.mnt_sb->s_type != type) | |
2931 | continue; | |
2932 | ||
2933 | /* This mount is not fully visible if there are any child mounts | |
2934 | * that cover anything except for empty directories. | |
2935 | */ | |
2936 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2937 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
2938 | if (!S_ISDIR(inode->i_mode)) | |
2939 | goto next; | |
41301ae7 | 2940 | if (inode->i_nlink > 2) |
e51db735 | 2941 | goto next; |
87a8ebd6 | 2942 | } |
e51db735 EB |
2943 | visible = true; |
2944 | goto found; | |
2945 | next: ; | |
87a8ebd6 | 2946 | } |
e51db735 | 2947 | found: |
44bb4385 | 2948 | up_read(&namespace_sem); |
e51db735 | 2949 | return visible; |
87a8ebd6 EB |
2950 | } |
2951 | ||
8823c079 EB |
2952 | static void *mntns_get(struct task_struct *task) |
2953 | { | |
2954 | struct mnt_namespace *ns = NULL; | |
2955 | struct nsproxy *nsproxy; | |
2956 | ||
2957 | rcu_read_lock(); | |
2958 | nsproxy = task_nsproxy(task); | |
2959 | if (nsproxy) { | |
2960 | ns = nsproxy->mnt_ns; | |
2961 | get_mnt_ns(ns); | |
2962 | } | |
2963 | rcu_read_unlock(); | |
2964 | ||
2965 | return ns; | |
2966 | } | |
2967 | ||
2968 | static void mntns_put(void *ns) | |
2969 | { | |
2970 | put_mnt_ns(ns); | |
2971 | } | |
2972 | ||
2973 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2974 | { | |
2975 | struct fs_struct *fs = current->fs; | |
2976 | struct mnt_namespace *mnt_ns = ns; | |
2977 | struct path root; | |
2978 | ||
0c55cfc4 | 2979 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
2980 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
2981 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 2982 | return -EPERM; |
8823c079 EB |
2983 | |
2984 | if (fs->users != 1) | |
2985 | return -EINVAL; | |
2986 | ||
2987 | get_mnt_ns(mnt_ns); | |
2988 | put_mnt_ns(nsproxy->mnt_ns); | |
2989 | nsproxy->mnt_ns = mnt_ns; | |
2990 | ||
2991 | /* Find the root */ | |
2992 | root.mnt = &mnt_ns->root->mnt; | |
2993 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2994 | path_get(&root); | |
2995 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
2996 | ; | |
2997 | ||
2998 | /* Update the pwd and root */ | |
2999 | set_fs_pwd(fs, &root); | |
3000 | set_fs_root(fs, &root); | |
3001 | ||
3002 | path_put(&root); | |
3003 | return 0; | |
3004 | } | |
3005 | ||
98f842e6 EB |
3006 | static unsigned int mntns_inum(void *ns) |
3007 | { | |
3008 | struct mnt_namespace *mnt_ns = ns; | |
3009 | return mnt_ns->proc_inum; | |
3010 | } | |
3011 | ||
8823c079 EB |
3012 | const struct proc_ns_operations mntns_operations = { |
3013 | .name = "mnt", | |
3014 | .type = CLONE_NEWNS, | |
3015 | .get = mntns_get, | |
3016 | .put = mntns_put, | |
3017 | .install = mntns_install, | |
98f842e6 | 3018 | .inum = mntns_inum, |
8823c079 | 3019 | }; |