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 | ||
12a5b529 AV |
781 | static void attach_shadowed(struct mount *mnt, |
782 | struct mount *parent, | |
783 | struct mount *shadows) | |
784 | { | |
785 | if (shadows) { | |
f6f99332 | 786 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
787 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
788 | } else { | |
789 | hlist_add_head_rcu(&mnt->mnt_hash, | |
790 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
791 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
792 | } | |
793 | } | |
794 | ||
b90fa9ae | 795 | /* |
99b7db7b | 796 | * vfsmount lock must be held for write |
b90fa9ae | 797 | */ |
1d6a32ac | 798 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 799 | { |
0714a533 | 800 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 801 | struct mount *m; |
b90fa9ae | 802 | LIST_HEAD(head); |
143c8c91 | 803 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 804 | |
0714a533 | 805 | BUG_ON(parent == mnt); |
b90fa9ae | 806 | |
1a4eeaf2 | 807 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 808 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 809 | m->mnt_ns = n; |
f03c6599 | 810 | |
b90fa9ae RP |
811 | list_splice(&head, n->list.prev); |
812 | ||
12a5b529 | 813 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 814 | touch_mnt_namespace(n); |
1da177e4 LT |
815 | } |
816 | ||
909b0a88 | 817 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 818 | { |
6b41d536 AV |
819 | struct list_head *next = p->mnt_mounts.next; |
820 | if (next == &p->mnt_mounts) { | |
1da177e4 | 821 | while (1) { |
909b0a88 | 822 | if (p == root) |
1da177e4 | 823 | return NULL; |
6b41d536 AV |
824 | next = p->mnt_child.next; |
825 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 826 | break; |
0714a533 | 827 | p = p->mnt_parent; |
1da177e4 LT |
828 | } |
829 | } | |
6b41d536 | 830 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
831 | } |
832 | ||
315fc83e | 833 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 834 | { |
6b41d536 AV |
835 | struct list_head *prev = p->mnt_mounts.prev; |
836 | while (prev != &p->mnt_mounts) { | |
837 | p = list_entry(prev, struct mount, mnt_child); | |
838 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
839 | } |
840 | return p; | |
841 | } | |
842 | ||
9d412a43 AV |
843 | struct vfsmount * |
844 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
845 | { | |
b105e270 | 846 | struct mount *mnt; |
9d412a43 AV |
847 | struct dentry *root; |
848 | ||
849 | if (!type) | |
850 | return ERR_PTR(-ENODEV); | |
851 | ||
852 | mnt = alloc_vfsmnt(name); | |
853 | if (!mnt) | |
854 | return ERR_PTR(-ENOMEM); | |
855 | ||
856 | if (flags & MS_KERNMOUNT) | |
b105e270 | 857 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
858 | |
859 | root = mount_fs(type, flags, name, data); | |
860 | if (IS_ERR(root)) { | |
8ffcb32e | 861 | mnt_free_id(mnt); |
9d412a43 AV |
862 | free_vfsmnt(mnt); |
863 | return ERR_CAST(root); | |
864 | } | |
865 | ||
b105e270 AV |
866 | mnt->mnt.mnt_root = root; |
867 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 868 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 869 | mnt->mnt_parent = mnt; |
719ea2fb | 870 | lock_mount_hash(); |
39f7c4db | 871 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 872 | unlock_mount_hash(); |
b105e270 | 873 | return &mnt->mnt; |
9d412a43 AV |
874 | } |
875 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
876 | ||
87129cc0 | 877 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 878 | int flag) |
1da177e4 | 879 | { |
87129cc0 | 880 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
881 | struct mount *mnt; |
882 | int err; | |
1da177e4 | 883 | |
be34d1a3 DH |
884 | mnt = alloc_vfsmnt(old->mnt_devname); |
885 | if (!mnt) | |
886 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 887 | |
7a472ef4 | 888 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
889 | mnt->mnt_group_id = 0; /* not a peer of original */ |
890 | else | |
891 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 892 | |
be34d1a3 DH |
893 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
894 | err = mnt_alloc_group_id(mnt); | |
895 | if (err) | |
896 | goto out_free; | |
1da177e4 | 897 | } |
be34d1a3 | 898 | |
f2ebb3a9 | 899 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 900 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
901 | if (flag & CL_UNPRIVILEGED) { |
902 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
903 | ||
904 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
905 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
906 | ||
907 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
908 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
909 | ||
910 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
911 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
912 | ||
913 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
914 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
915 | } | |
132c94e3 | 916 | |
5ff9d8a6 EB |
917 | /* Don't allow unprivileged users to reveal what is under a mount */ |
918 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
919 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
920 | ||
be34d1a3 DH |
921 | atomic_inc(&sb->s_active); |
922 | mnt->mnt.mnt_sb = sb; | |
923 | mnt->mnt.mnt_root = dget(root); | |
924 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
925 | mnt->mnt_parent = mnt; | |
719ea2fb | 926 | lock_mount_hash(); |
be34d1a3 | 927 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 928 | unlock_mount_hash(); |
be34d1a3 | 929 | |
7a472ef4 EB |
930 | if ((flag & CL_SLAVE) || |
931 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
932 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
933 | mnt->mnt_master = old; | |
934 | CLEAR_MNT_SHARED(mnt); | |
935 | } else if (!(flag & CL_PRIVATE)) { | |
936 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
937 | list_add(&mnt->mnt_share, &old->mnt_share); | |
938 | if (IS_MNT_SLAVE(old)) | |
939 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
940 | mnt->mnt_master = old->mnt_master; | |
941 | } | |
942 | if (flag & CL_MAKE_SHARED) | |
943 | set_mnt_shared(mnt); | |
944 | ||
945 | /* stick the duplicate mount on the same expiry list | |
946 | * as the original if that was on one */ | |
947 | if (flag & CL_EXPIRE) { | |
948 | if (!list_empty(&old->mnt_expire)) | |
949 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
950 | } | |
951 | ||
cb338d06 | 952 | return mnt; |
719f5d7f MS |
953 | |
954 | out_free: | |
8ffcb32e | 955 | mnt_free_id(mnt); |
719f5d7f | 956 | free_vfsmnt(mnt); |
be34d1a3 | 957 | return ERR_PTR(err); |
1da177e4 LT |
958 | } |
959 | ||
900148dc | 960 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 961 | { |
48a066e7 AV |
962 | rcu_read_lock(); |
963 | mnt_add_count(mnt, -1); | |
964 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
965 | rcu_read_unlock(); | |
f03c6599 | 966 | return; |
b3e19d92 | 967 | } |
719ea2fb | 968 | lock_mount_hash(); |
b3e19d92 | 969 | if (mnt_get_count(mnt)) { |
48a066e7 | 970 | rcu_read_unlock(); |
719ea2fb | 971 | unlock_mount_hash(); |
99b7db7b NP |
972 | return; |
973 | } | |
48a066e7 AV |
974 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
975 | rcu_read_unlock(); | |
976 | unlock_mount_hash(); | |
977 | return; | |
978 | } | |
979 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
980 | rcu_read_unlock(); | |
962830df | 981 | |
39f7c4db | 982 | list_del(&mnt->mnt_instance); |
719ea2fb | 983 | unlock_mount_hash(); |
649a795a AV |
984 | |
985 | /* | |
986 | * This probably indicates that somebody messed | |
987 | * up a mnt_want/drop_write() pair. If this | |
988 | * happens, the filesystem was probably unable | |
989 | * to make r/w->r/o transitions. | |
990 | */ | |
991 | /* | |
992 | * The locking used to deal with mnt_count decrement provides barriers, | |
993 | * so mnt_get_writers() below is safe. | |
994 | */ | |
995 | WARN_ON(mnt_get_writers(mnt)); | |
3064c356 AV |
996 | if (unlikely(mnt->mnt_pins.first)) |
997 | mnt_pin_kill(mnt); | |
649a795a AV |
998 | fsnotify_vfsmount_delete(&mnt->mnt); |
999 | dput(mnt->mnt.mnt_root); | |
1000 | deactivate_super(mnt->mnt.mnt_sb); | |
48a066e7 | 1001 | mnt_free_id(mnt); |
8ffcb32e | 1002 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); |
b3e19d92 | 1003 | } |
b3e19d92 NP |
1004 | |
1005 | void mntput(struct vfsmount *mnt) | |
1006 | { | |
1007 | if (mnt) { | |
863d684f | 1008 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1009 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1010 | if (unlikely(m->mnt_expiry_mark)) |
1011 | m->mnt_expiry_mark = 0; | |
1012 | mntput_no_expire(m); | |
b3e19d92 NP |
1013 | } |
1014 | } | |
1015 | EXPORT_SYMBOL(mntput); | |
1016 | ||
1017 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1018 | { | |
1019 | if (mnt) | |
83adc753 | 1020 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1021 | return mnt; |
1022 | } | |
1023 | EXPORT_SYMBOL(mntget); | |
1024 | ||
3064c356 | 1025 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1026 | { |
3064c356 AV |
1027 | struct mount *p; |
1028 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1029 | if (IS_ERR(p)) | |
1030 | return ERR_CAST(p); | |
1031 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1032 | return &p->mnt; | |
7b7b1ace | 1033 | } |
1da177e4 | 1034 | |
b3b304a2 MS |
1035 | static inline void mangle(struct seq_file *m, const char *s) |
1036 | { | |
1037 | seq_escape(m, s, " \t\n\\"); | |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * Simple .show_options callback for filesystems which don't want to | |
1042 | * implement more complex mount option showing. | |
1043 | * | |
1044 | * See also save_mount_options(). | |
1045 | */ | |
34c80b1d | 1046 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1047 | { |
2a32cebd AV |
1048 | const char *options; |
1049 | ||
1050 | rcu_read_lock(); | |
34c80b1d | 1051 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1052 | |
1053 | if (options != NULL && options[0]) { | |
1054 | seq_putc(m, ','); | |
1055 | mangle(m, options); | |
1056 | } | |
2a32cebd | 1057 | rcu_read_unlock(); |
b3b304a2 MS |
1058 | |
1059 | return 0; | |
1060 | } | |
1061 | EXPORT_SYMBOL(generic_show_options); | |
1062 | ||
1063 | /* | |
1064 | * If filesystem uses generic_show_options(), this function should be | |
1065 | * called from the fill_super() callback. | |
1066 | * | |
1067 | * The .remount_fs callback usually needs to be handled in a special | |
1068 | * way, to make sure, that previous options are not overwritten if the | |
1069 | * remount fails. | |
1070 | * | |
1071 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1072 | * reset all options to their default value, but changes only newly | |
1073 | * given options, then the displayed options will not reflect reality | |
1074 | * any more. | |
1075 | */ | |
1076 | void save_mount_options(struct super_block *sb, char *options) | |
1077 | { | |
2a32cebd AV |
1078 | BUG_ON(sb->s_options); |
1079 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1080 | } |
1081 | EXPORT_SYMBOL(save_mount_options); | |
1082 | ||
2a32cebd AV |
1083 | void replace_mount_options(struct super_block *sb, char *options) |
1084 | { | |
1085 | char *old = sb->s_options; | |
1086 | rcu_assign_pointer(sb->s_options, options); | |
1087 | if (old) { | |
1088 | synchronize_rcu(); | |
1089 | kfree(old); | |
1090 | } | |
1091 | } | |
1092 | EXPORT_SYMBOL(replace_mount_options); | |
1093 | ||
a1a2c409 | 1094 | #ifdef CONFIG_PROC_FS |
0226f492 | 1095 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1096 | static void *m_start(struct seq_file *m, loff_t *pos) |
1097 | { | |
6ce6e24e | 1098 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1099 | |
390c6843 | 1100 | down_read(&namespace_sem); |
c7999c36 AV |
1101 | if (p->cached_event == p->ns->event) { |
1102 | void *v = p->cached_mount; | |
1103 | if (*pos == p->cached_index) | |
1104 | return v; | |
1105 | if (*pos == p->cached_index + 1) { | |
1106 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1107 | return p->cached_mount = v; | |
1108 | } | |
1109 | } | |
1110 | ||
1111 | p->cached_event = p->ns->event; | |
1112 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1113 | p->cached_index = *pos; | |
1114 | return p->cached_mount; | |
1da177e4 LT |
1115 | } |
1116 | ||
1117 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1118 | { | |
6ce6e24e | 1119 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1120 | |
c7999c36 AV |
1121 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1122 | p->cached_index = *pos; | |
1123 | return p->cached_mount; | |
1da177e4 LT |
1124 | } |
1125 | ||
1126 | static void m_stop(struct seq_file *m, void *v) | |
1127 | { | |
390c6843 | 1128 | up_read(&namespace_sem); |
1da177e4 LT |
1129 | } |
1130 | ||
0226f492 | 1131 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1132 | { |
6ce6e24e | 1133 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1134 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1135 | return p->show(m, &r->mnt); |
1da177e4 LT |
1136 | } |
1137 | ||
a1a2c409 | 1138 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1139 | .start = m_start, |
1140 | .next = m_next, | |
1141 | .stop = m_stop, | |
0226f492 | 1142 | .show = m_show, |
b4629fe2 | 1143 | }; |
a1a2c409 | 1144 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1145 | |
1da177e4 LT |
1146 | /** |
1147 | * may_umount_tree - check if a mount tree is busy | |
1148 | * @mnt: root of mount tree | |
1149 | * | |
1150 | * This is called to check if a tree of mounts has any | |
1151 | * open files, pwds, chroots or sub mounts that are | |
1152 | * busy. | |
1153 | */ | |
909b0a88 | 1154 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1155 | { |
909b0a88 | 1156 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1157 | int actual_refs = 0; |
1158 | int minimum_refs = 0; | |
315fc83e | 1159 | struct mount *p; |
909b0a88 | 1160 | BUG_ON(!m); |
1da177e4 | 1161 | |
b3e19d92 | 1162 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1163 | lock_mount_hash(); |
909b0a88 | 1164 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1165 | actual_refs += mnt_get_count(p); |
1da177e4 | 1166 | minimum_refs += 2; |
1da177e4 | 1167 | } |
719ea2fb | 1168 | unlock_mount_hash(); |
1da177e4 LT |
1169 | |
1170 | if (actual_refs > minimum_refs) | |
e3474a8e | 1171 | return 0; |
1da177e4 | 1172 | |
e3474a8e | 1173 | return 1; |
1da177e4 LT |
1174 | } |
1175 | ||
1176 | EXPORT_SYMBOL(may_umount_tree); | |
1177 | ||
1178 | /** | |
1179 | * may_umount - check if a mount point is busy | |
1180 | * @mnt: root of mount | |
1181 | * | |
1182 | * This is called to check if a mount point has any | |
1183 | * open files, pwds, chroots or sub mounts. If the | |
1184 | * mount has sub mounts this will return busy | |
1185 | * regardless of whether the sub mounts are busy. | |
1186 | * | |
1187 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1188 | * give false negatives. The main reason why it's here is that we need | |
1189 | * a non-destructive way to look for easily umountable filesystems. | |
1190 | */ | |
1191 | int may_umount(struct vfsmount *mnt) | |
1192 | { | |
e3474a8e | 1193 | int ret = 1; |
8ad08d8a | 1194 | down_read(&namespace_sem); |
719ea2fb | 1195 | lock_mount_hash(); |
1ab59738 | 1196 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1197 | ret = 0; |
719ea2fb | 1198 | unlock_mount_hash(); |
8ad08d8a | 1199 | up_read(&namespace_sem); |
a05964f3 | 1200 | return ret; |
1da177e4 LT |
1201 | } |
1202 | ||
1203 | EXPORT_SYMBOL(may_umount); | |
1204 | ||
38129a13 | 1205 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1206 | |
97216be0 | 1207 | static void namespace_unlock(void) |
70fbcdf4 | 1208 | { |
d5e50f74 | 1209 | struct mount *mnt; |
38129a13 | 1210 | struct hlist_head head = unmounted; |
97216be0 | 1211 | |
38129a13 | 1212 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1213 | up_write(&namespace_sem); |
1214 | return; | |
1215 | } | |
1216 | ||
38129a13 AV |
1217 | head.first->pprev = &head.first; |
1218 | INIT_HLIST_HEAD(&unmounted); | |
1219 | ||
81b6b061 AV |
1220 | /* undo decrements we'd done in umount_tree() */ |
1221 | hlist_for_each_entry(mnt, &head, mnt_hash) | |
1222 | if (mnt->mnt_ex_mountpoint.mnt) | |
1223 | mntget(mnt->mnt_ex_mountpoint.mnt); | |
1224 | ||
97216be0 AV |
1225 | up_write(&namespace_sem); |
1226 | ||
48a066e7 AV |
1227 | synchronize_rcu(); |
1228 | ||
38129a13 AV |
1229 | while (!hlist_empty(&head)) { |
1230 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1231 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1232 | if (mnt->mnt_ex_mountpoint.mnt) |
1233 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1234 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1235 | } |
1236 | } | |
1237 | ||
97216be0 | 1238 | static inline void namespace_lock(void) |
e3197d83 | 1239 | { |
97216be0 | 1240 | down_write(&namespace_sem); |
e3197d83 AV |
1241 | } |
1242 | ||
99b7db7b | 1243 | /* |
48a066e7 | 1244 | * mount_lock must be held |
99b7db7b | 1245 | * namespace_sem must be held for write |
48a066e7 AV |
1246 | * how = 0 => just this tree, don't propagate |
1247 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1248 | * how = 2 => lazy umount | |
99b7db7b | 1249 | */ |
48a066e7 | 1250 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1251 | { |
38129a13 | 1252 | HLIST_HEAD(tmp_list); |
315fc83e | 1253 | struct mount *p; |
38129a13 | 1254 | struct mount *last = NULL; |
1da177e4 | 1255 | |
38129a13 AV |
1256 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1257 | hlist_del_init_rcu(&p->mnt_hash); | |
1258 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1259 | } | |
1da177e4 | 1260 | |
88b368f2 AV |
1261 | hlist_for_each_entry(p, &tmp_list, mnt_hash) |
1262 | list_del_init(&p->mnt_child); | |
1263 | ||
48a066e7 | 1264 | if (how) |
7b8a53fd | 1265 | propagate_umount(&tmp_list); |
a05964f3 | 1266 | |
38129a13 | 1267 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1268 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1269 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1270 | __touch_mnt_namespace(p->mnt_ns); |
1271 | p->mnt_ns = NULL; | |
48a066e7 AV |
1272 | if (how < 2) |
1273 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
676da58d | 1274 | if (mnt_has_parent(p)) { |
84d17192 | 1275 | put_mountpoint(p->mnt_mp); |
81b6b061 | 1276 | mnt_add_count(p->mnt_parent, -1); |
aba809cf AV |
1277 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1278 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1279 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1280 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1281 | p->mnt_parent = p; | |
84d17192 | 1282 | p->mnt_mp = NULL; |
7c4b93d8 | 1283 | } |
0f0afb1d | 1284 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1285 | last = p; |
1286 | } | |
1287 | if (last) { | |
1288 | last->mnt_hash.next = unmounted.first; | |
1289 | unmounted.first = tmp_list.first; | |
1290 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1291 | } |
1292 | } | |
1293 | ||
b54b9be7 | 1294 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1295 | |
1ab59738 | 1296 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1297 | { |
1ab59738 | 1298 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1299 | int retval; |
1300 | ||
1ab59738 | 1301 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1302 | if (retval) |
1303 | return retval; | |
1304 | ||
1305 | /* | |
1306 | * Allow userspace to request a mountpoint be expired rather than | |
1307 | * unmounting unconditionally. Unmount only happens if: | |
1308 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1309 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1310 | */ | |
1311 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1312 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1313 | flags & (MNT_FORCE | MNT_DETACH)) |
1314 | return -EINVAL; | |
1315 | ||
b3e19d92 NP |
1316 | /* |
1317 | * probably don't strictly need the lock here if we examined | |
1318 | * all race cases, but it's a slowpath. | |
1319 | */ | |
719ea2fb | 1320 | lock_mount_hash(); |
83adc753 | 1321 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1322 | unlock_mount_hash(); |
1da177e4 | 1323 | return -EBUSY; |
b3e19d92 | 1324 | } |
719ea2fb | 1325 | unlock_mount_hash(); |
1da177e4 | 1326 | |
863d684f | 1327 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1328 | return -EAGAIN; |
1329 | } | |
1330 | ||
1331 | /* | |
1332 | * If we may have to abort operations to get out of this | |
1333 | * mount, and they will themselves hold resources we must | |
1334 | * allow the fs to do things. In the Unix tradition of | |
1335 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1336 | * might fail to complete on the first run through as other tasks | |
1337 | * must return, and the like. Thats for the mount program to worry | |
1338 | * about for the moment. | |
1339 | */ | |
1340 | ||
42faad99 | 1341 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1342 | sb->s_op->umount_begin(sb); |
42faad99 | 1343 | } |
1da177e4 LT |
1344 | |
1345 | /* | |
1346 | * No sense to grab the lock for this test, but test itself looks | |
1347 | * somewhat bogus. Suggestions for better replacement? | |
1348 | * Ho-hum... In principle, we might treat that as umount + switch | |
1349 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1350 | * Actually it makes sense, especially if rootfs would contain a | |
1351 | * /reboot - static binary that would close all descriptors and | |
1352 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1353 | */ | |
1ab59738 | 1354 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1355 | /* |
1356 | * Special case for "unmounting" root ... | |
1357 | * we just try to remount it readonly. | |
1358 | */ | |
1359 | down_write(&sb->s_umount); | |
4aa98cf7 | 1360 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1361 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1362 | up_write(&sb->s_umount); |
1363 | return retval; | |
1364 | } | |
1365 | ||
97216be0 | 1366 | namespace_lock(); |
719ea2fb | 1367 | lock_mount_hash(); |
5addc5dd | 1368 | event++; |
1da177e4 | 1369 | |
48a066e7 | 1370 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1371 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1372 | umount_tree(mnt, 2); |
1da177e4 | 1373 | retval = 0; |
48a066e7 AV |
1374 | } else { |
1375 | shrink_submounts(mnt); | |
1376 | retval = -EBUSY; | |
1377 | if (!propagate_mount_busy(mnt, 2)) { | |
1378 | if (!list_empty(&mnt->mnt_list)) | |
1379 | umount_tree(mnt, 1); | |
1380 | retval = 0; | |
1381 | } | |
1da177e4 | 1382 | } |
719ea2fb | 1383 | unlock_mount_hash(); |
e3197d83 | 1384 | namespace_unlock(); |
1da177e4 LT |
1385 | return retval; |
1386 | } | |
1387 | ||
9b40bc90 AV |
1388 | /* |
1389 | * Is the caller allowed to modify his namespace? | |
1390 | */ | |
1391 | static inline bool may_mount(void) | |
1392 | { | |
1393 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1394 | } | |
1395 | ||
1da177e4 LT |
1396 | /* |
1397 | * Now umount can handle mount points as well as block devices. | |
1398 | * This is important for filesystems which use unnamed block devices. | |
1399 | * | |
1400 | * We now support a flag for forced unmount like the other 'big iron' | |
1401 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1402 | */ | |
1403 | ||
bdc480e3 | 1404 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1405 | { |
2d8f3038 | 1406 | struct path path; |
900148dc | 1407 | struct mount *mnt; |
1da177e4 | 1408 | int retval; |
db1f05bb | 1409 | int lookup_flags = 0; |
1da177e4 | 1410 | |
db1f05bb MS |
1411 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1412 | return -EINVAL; | |
1413 | ||
9b40bc90 AV |
1414 | if (!may_mount()) |
1415 | return -EPERM; | |
1416 | ||
db1f05bb MS |
1417 | if (!(flags & UMOUNT_NOFOLLOW)) |
1418 | lookup_flags |= LOOKUP_FOLLOW; | |
1419 | ||
197df04c | 1420 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1421 | if (retval) |
1422 | goto out; | |
900148dc | 1423 | mnt = real_mount(path.mnt); |
1da177e4 | 1424 | retval = -EINVAL; |
2d8f3038 | 1425 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1426 | goto dput_and_out; |
143c8c91 | 1427 | if (!check_mnt(mnt)) |
1da177e4 | 1428 | goto dput_and_out; |
5ff9d8a6 EB |
1429 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1430 | goto dput_and_out; | |
1da177e4 | 1431 | |
900148dc | 1432 | retval = do_umount(mnt, flags); |
1da177e4 | 1433 | dput_and_out: |
429731b1 | 1434 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1435 | dput(path.dentry); |
900148dc | 1436 | mntput_no_expire(mnt); |
1da177e4 LT |
1437 | out: |
1438 | return retval; | |
1439 | } | |
1440 | ||
1441 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1442 | ||
1443 | /* | |
b58fed8b | 1444 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1445 | */ |
bdc480e3 | 1446 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1447 | { |
b58fed8b | 1448 | return sys_umount(name, 0); |
1da177e4 LT |
1449 | } |
1450 | ||
1451 | #endif | |
1452 | ||
4ce5d2b1 | 1453 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1454 | { |
4ce5d2b1 EB |
1455 | /* Is this a proxy for a mount namespace? */ |
1456 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1457 | struct proc_ns *ei; |
8823c079 EB |
1458 | |
1459 | if (!proc_ns_inode(inode)) | |
1460 | return false; | |
1461 | ||
0bb80f24 | 1462 | ei = get_proc_ns(inode); |
8823c079 EB |
1463 | if (ei->ns_ops != &mntns_operations) |
1464 | return false; | |
1465 | ||
4ce5d2b1 EB |
1466 | return true; |
1467 | } | |
1468 | ||
1469 | static bool mnt_ns_loop(struct dentry *dentry) | |
1470 | { | |
1471 | /* Could bind mounting the mount namespace inode cause a | |
1472 | * mount namespace loop? | |
1473 | */ | |
1474 | struct mnt_namespace *mnt_ns; | |
1475 | if (!is_mnt_ns_file(dentry)) | |
1476 | return false; | |
1477 | ||
1478 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1479 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1480 | } | |
1481 | ||
87129cc0 | 1482 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1483 | int flag) |
1da177e4 | 1484 | { |
84d17192 | 1485 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1486 | |
4ce5d2b1 EB |
1487 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1488 | return ERR_PTR(-EINVAL); | |
1489 | ||
1490 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1491 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1492 | |
36341f64 | 1493 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1494 | if (IS_ERR(q)) |
1495 | return q; | |
1496 | ||
5ff9d8a6 | 1497 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1498 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1499 | |
1500 | p = mnt; | |
6b41d536 | 1501 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1502 | struct mount *s; |
7ec02ef1 | 1503 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1504 | continue; |
1505 | ||
909b0a88 | 1506 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1507 | struct mount *t = NULL; |
4ce5d2b1 EB |
1508 | if (!(flag & CL_COPY_UNBINDABLE) && |
1509 | IS_MNT_UNBINDABLE(s)) { | |
1510 | s = skip_mnt_tree(s); | |
1511 | continue; | |
1512 | } | |
1513 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1514 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1515 | s = skip_mnt_tree(s); |
1516 | continue; | |
1517 | } | |
0714a533 AV |
1518 | while (p != s->mnt_parent) { |
1519 | p = p->mnt_parent; | |
1520 | q = q->mnt_parent; | |
1da177e4 | 1521 | } |
87129cc0 | 1522 | p = s; |
84d17192 | 1523 | parent = q; |
87129cc0 | 1524 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1525 | if (IS_ERR(q)) |
1526 | goto out; | |
719ea2fb | 1527 | lock_mount_hash(); |
1a4eeaf2 | 1528 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1529 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1530 | if (!list_empty(&parent->mnt_mounts)) { | |
1531 | t = list_last_entry(&parent->mnt_mounts, | |
1532 | struct mount, mnt_child); | |
1533 | if (t->mnt_mp != p->mnt_mp) | |
1534 | t = NULL; | |
1535 | } | |
1536 | attach_shadowed(q, parent, t); | |
719ea2fb | 1537 | unlock_mount_hash(); |
1da177e4 LT |
1538 | } |
1539 | } | |
1540 | return res; | |
be34d1a3 | 1541 | out: |
1da177e4 | 1542 | if (res) { |
719ea2fb | 1543 | lock_mount_hash(); |
328e6d90 | 1544 | umount_tree(res, 0); |
719ea2fb | 1545 | unlock_mount_hash(); |
1da177e4 | 1546 | } |
be34d1a3 | 1547 | return q; |
1da177e4 LT |
1548 | } |
1549 | ||
be34d1a3 DH |
1550 | /* Caller should check returned pointer for errors */ |
1551 | ||
589ff870 | 1552 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1553 | { |
cb338d06 | 1554 | struct mount *tree; |
97216be0 | 1555 | namespace_lock(); |
87129cc0 AV |
1556 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1557 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1558 | namespace_unlock(); |
be34d1a3 | 1559 | if (IS_ERR(tree)) |
52e220d3 | 1560 | return ERR_CAST(tree); |
be34d1a3 | 1561 | return &tree->mnt; |
8aec0809 AV |
1562 | } |
1563 | ||
1564 | void drop_collected_mounts(struct vfsmount *mnt) | |
1565 | { | |
97216be0 | 1566 | namespace_lock(); |
719ea2fb | 1567 | lock_mount_hash(); |
328e6d90 | 1568 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1569 | unlock_mount_hash(); |
3ab6abee | 1570 | namespace_unlock(); |
8aec0809 AV |
1571 | } |
1572 | ||
1f707137 AV |
1573 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1574 | struct vfsmount *root) | |
1575 | { | |
1a4eeaf2 | 1576 | struct mount *mnt; |
1f707137 AV |
1577 | int res = f(root, arg); |
1578 | if (res) | |
1579 | return res; | |
1a4eeaf2 AV |
1580 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1581 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1582 | if (res) |
1583 | return res; | |
1584 | } | |
1585 | return 0; | |
1586 | } | |
1587 | ||
4b8b21f4 | 1588 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1589 | { |
315fc83e | 1590 | struct mount *p; |
719f5d7f | 1591 | |
909b0a88 | 1592 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1593 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1594 | mnt_release_group_id(p); |
719f5d7f MS |
1595 | } |
1596 | } | |
1597 | ||
4b8b21f4 | 1598 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1599 | { |
315fc83e | 1600 | struct mount *p; |
719f5d7f | 1601 | |
909b0a88 | 1602 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1603 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1604 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1605 | if (err) { |
4b8b21f4 | 1606 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1607 | return err; |
1608 | } | |
1609 | } | |
1610 | } | |
1611 | ||
1612 | return 0; | |
1613 | } | |
1614 | ||
b90fa9ae RP |
1615 | /* |
1616 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1617 | * @nd : place the mount tree @source_mnt is attached |
1618 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1619 | * store the parent mount and mountpoint dentry. | |
1620 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1621 | * |
1622 | * NOTE: in the table below explains the semantics when a source mount | |
1623 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1624 | * --------------------------------------------------------------------------- |
1625 | * | BIND MOUNT OPERATION | | |
1626 | * |************************************************************************** | |
1627 | * | source-->| shared | private | slave | unbindable | | |
1628 | * | dest | | | | | | |
1629 | * | | | | | | | | |
1630 | * | v | | | | | | |
1631 | * |************************************************************************** | |
1632 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1633 | * | | | | | | | |
1634 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1635 | * *************************************************************************** | |
b90fa9ae RP |
1636 | * A bind operation clones the source mount and mounts the clone on the |
1637 | * destination mount. | |
1638 | * | |
1639 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1640 | * tree of the destination mount and the cloned mount is added to | |
1641 | * the peer group of the source mount. | |
1642 | * (+) the cloned mount is created under the destination mount and is marked | |
1643 | * as shared. The cloned mount is added to the peer group of the source | |
1644 | * mount. | |
5afe0022 RP |
1645 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1646 | * of the destination mount and the cloned mount is made slave | |
1647 | * of the same master as that of the source mount. The cloned mount | |
1648 | * is marked as 'shared and slave'. | |
1649 | * (*) the cloned mount is made a slave of the same master as that of the | |
1650 | * source mount. | |
1651 | * | |
9676f0c6 RP |
1652 | * --------------------------------------------------------------------------- |
1653 | * | MOVE MOUNT OPERATION | | |
1654 | * |************************************************************************** | |
1655 | * | source-->| shared | private | slave | unbindable | | |
1656 | * | dest | | | | | | |
1657 | * | | | | | | | | |
1658 | * | v | | | | | | |
1659 | * |************************************************************************** | |
1660 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1661 | * | | | | | | | |
1662 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1663 | * *************************************************************************** | |
5afe0022 RP |
1664 | * |
1665 | * (+) the mount is moved to the destination. And is then propagated to | |
1666 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1667 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1668 | * (+++) the mount is moved to the destination and is then propagated to |
1669 | * all the mounts belonging to the destination mount's propagation tree. | |
1670 | * the mount is marked as 'shared and slave'. | |
1671 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1672 | * |
1673 | * if the source mount is a tree, the operations explained above is | |
1674 | * applied to each mount in the tree. | |
1675 | * Must be called without spinlocks held, since this function can sleep | |
1676 | * in allocations. | |
1677 | */ | |
0fb54e50 | 1678 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1679 | struct mount *dest_mnt, |
1680 | struct mountpoint *dest_mp, | |
1681 | struct path *parent_path) | |
b90fa9ae | 1682 | { |
38129a13 | 1683 | HLIST_HEAD(tree_list); |
315fc83e | 1684 | struct mount *child, *p; |
38129a13 | 1685 | struct hlist_node *n; |
719f5d7f | 1686 | int err; |
b90fa9ae | 1687 | |
fc7be130 | 1688 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1689 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1690 | if (err) |
1691 | goto out; | |
0b1b901b | 1692 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1693 | lock_mount_hash(); |
0b1b901b AV |
1694 | if (err) |
1695 | goto out_cleanup_ids; | |
909b0a88 | 1696 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1697 | set_mnt_shared(p); |
0b1b901b AV |
1698 | } else { |
1699 | lock_mount_hash(); | |
b90fa9ae | 1700 | } |
1a390689 | 1701 | if (parent_path) { |
0fb54e50 | 1702 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1703 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1704 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1705 | } else { |
84d17192 | 1706 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1707 | commit_tree(source_mnt, NULL); |
21444403 | 1708 | } |
b90fa9ae | 1709 | |
38129a13 | 1710 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1711 | struct mount *q; |
38129a13 | 1712 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1713 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1714 | child->mnt_mountpoint); | |
1715 | commit_tree(child, q); | |
b90fa9ae | 1716 | } |
719ea2fb | 1717 | unlock_mount_hash(); |
99b7db7b | 1718 | |
b90fa9ae | 1719 | return 0; |
719f5d7f MS |
1720 | |
1721 | out_cleanup_ids: | |
f2ebb3a9 AV |
1722 | while (!hlist_empty(&tree_list)) { |
1723 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1724 | umount_tree(child, 0); | |
1725 | } | |
1726 | unlock_mount_hash(); | |
0b1b901b | 1727 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1728 | out: |
1729 | return err; | |
b90fa9ae RP |
1730 | } |
1731 | ||
84d17192 | 1732 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1733 | { |
1734 | struct vfsmount *mnt; | |
84d17192 | 1735 | struct dentry *dentry = path->dentry; |
b12cea91 | 1736 | retry: |
84d17192 AV |
1737 | mutex_lock(&dentry->d_inode->i_mutex); |
1738 | if (unlikely(cant_mount(dentry))) { | |
1739 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1740 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1741 | } |
97216be0 | 1742 | namespace_lock(); |
b12cea91 | 1743 | mnt = lookup_mnt(path); |
84d17192 AV |
1744 | if (likely(!mnt)) { |
1745 | struct mountpoint *mp = new_mountpoint(dentry); | |
1746 | if (IS_ERR(mp)) { | |
97216be0 | 1747 | namespace_unlock(); |
84d17192 AV |
1748 | mutex_unlock(&dentry->d_inode->i_mutex); |
1749 | return mp; | |
1750 | } | |
1751 | return mp; | |
1752 | } | |
97216be0 | 1753 | namespace_unlock(); |
b12cea91 AV |
1754 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1755 | path_put(path); | |
1756 | path->mnt = mnt; | |
84d17192 | 1757 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1758 | goto retry; |
1759 | } | |
1760 | ||
84d17192 | 1761 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1762 | { |
84d17192 AV |
1763 | struct dentry *dentry = where->m_dentry; |
1764 | put_mountpoint(where); | |
328e6d90 | 1765 | namespace_unlock(); |
84d17192 | 1766 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1767 | } |
1768 | ||
84d17192 | 1769 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1770 | { |
95bc5f25 | 1771 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1772 | return -EINVAL; |
1773 | ||
84d17192 | 1774 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1775 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1776 | return -ENOTDIR; |
1777 | ||
84d17192 | 1778 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1779 | } |
1780 | ||
7a2e8a8f VA |
1781 | /* |
1782 | * Sanity check the flags to change_mnt_propagation. | |
1783 | */ | |
1784 | ||
1785 | static int flags_to_propagation_type(int flags) | |
1786 | { | |
7c6e984d | 1787 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1788 | |
1789 | /* Fail if any non-propagation flags are set */ | |
1790 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1791 | return 0; | |
1792 | /* Only one propagation flag should be set */ | |
1793 | if (!is_power_of_2(type)) | |
1794 | return 0; | |
1795 | return type; | |
1796 | } | |
1797 | ||
07b20889 RP |
1798 | /* |
1799 | * recursively change the type of the mountpoint. | |
1800 | */ | |
0a0d8a46 | 1801 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1802 | { |
315fc83e | 1803 | struct mount *m; |
4b8b21f4 | 1804 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1805 | int recurse = flag & MS_REC; |
7a2e8a8f | 1806 | int type; |
719f5d7f | 1807 | int err = 0; |
07b20889 | 1808 | |
2d92ab3c | 1809 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1810 | return -EINVAL; |
1811 | ||
7a2e8a8f VA |
1812 | type = flags_to_propagation_type(flag); |
1813 | if (!type) | |
1814 | return -EINVAL; | |
1815 | ||
97216be0 | 1816 | namespace_lock(); |
719f5d7f MS |
1817 | if (type == MS_SHARED) { |
1818 | err = invent_group_ids(mnt, recurse); | |
1819 | if (err) | |
1820 | goto out_unlock; | |
1821 | } | |
1822 | ||
719ea2fb | 1823 | lock_mount_hash(); |
909b0a88 | 1824 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1825 | change_mnt_propagation(m, type); |
719ea2fb | 1826 | unlock_mount_hash(); |
719f5d7f MS |
1827 | |
1828 | out_unlock: | |
97216be0 | 1829 | namespace_unlock(); |
719f5d7f | 1830 | return err; |
07b20889 RP |
1831 | } |
1832 | ||
5ff9d8a6 EB |
1833 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1834 | { | |
1835 | struct mount *child; | |
1836 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1837 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1838 | continue; | |
1839 | ||
1840 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1841 | return true; | |
1842 | } | |
1843 | return false; | |
1844 | } | |
1845 | ||
1da177e4 LT |
1846 | /* |
1847 | * do loopback mount. | |
1848 | */ | |
808d4e3c | 1849 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1850 | int recurse) |
1da177e4 | 1851 | { |
2d92ab3c | 1852 | struct path old_path; |
84d17192 AV |
1853 | struct mount *mnt = NULL, *old, *parent; |
1854 | struct mountpoint *mp; | |
57eccb83 | 1855 | int err; |
1da177e4 LT |
1856 | if (!old_name || !*old_name) |
1857 | return -EINVAL; | |
815d405c | 1858 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1859 | if (err) |
1860 | return err; | |
1861 | ||
8823c079 | 1862 | err = -EINVAL; |
4ce5d2b1 | 1863 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1864 | goto out; |
1865 | ||
84d17192 AV |
1866 | mp = lock_mount(path); |
1867 | err = PTR_ERR(mp); | |
1868 | if (IS_ERR(mp)) | |
b12cea91 AV |
1869 | goto out; |
1870 | ||
87129cc0 | 1871 | old = real_mount(old_path.mnt); |
84d17192 | 1872 | parent = real_mount(path->mnt); |
87129cc0 | 1873 | |
1da177e4 | 1874 | err = -EINVAL; |
fc7be130 | 1875 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1876 | goto out2; |
9676f0c6 | 1877 | |
84d17192 | 1878 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1879 | goto out2; |
1da177e4 | 1880 | |
5ff9d8a6 EB |
1881 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1882 | goto out2; | |
1883 | ||
ccd48bc7 | 1884 | if (recurse) |
4ce5d2b1 | 1885 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1886 | else |
87129cc0 | 1887 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1888 | |
be34d1a3 DH |
1889 | if (IS_ERR(mnt)) { |
1890 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1891 | goto out2; |
be34d1a3 | 1892 | } |
ccd48bc7 | 1893 | |
5ff9d8a6 EB |
1894 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1895 | ||
84d17192 | 1896 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1897 | if (err) { |
719ea2fb | 1898 | lock_mount_hash(); |
328e6d90 | 1899 | umount_tree(mnt, 0); |
719ea2fb | 1900 | unlock_mount_hash(); |
5b83d2c5 | 1901 | } |
b12cea91 | 1902 | out2: |
84d17192 | 1903 | unlock_mount(mp); |
ccd48bc7 | 1904 | out: |
2d92ab3c | 1905 | path_put(&old_path); |
1da177e4 LT |
1906 | return err; |
1907 | } | |
1908 | ||
2e4b7fcd DH |
1909 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1910 | { | |
1911 | int error = 0; | |
1912 | int readonly_request = 0; | |
1913 | ||
1914 | if (ms_flags & MS_RDONLY) | |
1915 | readonly_request = 1; | |
1916 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1917 | return 0; | |
1918 | ||
1919 | if (readonly_request) | |
83adc753 | 1920 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1921 | else |
83adc753 | 1922 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1923 | return error; |
1924 | } | |
1925 | ||
1da177e4 LT |
1926 | /* |
1927 | * change filesystem flags. dir should be a physical root of filesystem. | |
1928 | * If you've mounted a non-root directory somewhere and want to do remount | |
1929 | * on it - tough luck. | |
1930 | */ | |
0a0d8a46 | 1931 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1932 | void *data) |
1933 | { | |
1934 | int err; | |
2d92ab3c | 1935 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1936 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1937 | |
143c8c91 | 1938 | if (!check_mnt(mnt)) |
1da177e4 LT |
1939 | return -EINVAL; |
1940 | ||
2d92ab3c | 1941 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1942 | return -EINVAL; |
1943 | ||
07b64558 EB |
1944 | /* Don't allow changing of locked mnt flags. |
1945 | * | |
1946 | * No locks need to be held here while testing the various | |
1947 | * MNT_LOCK flags because those flags can never be cleared | |
1948 | * once they are set. | |
1949 | */ | |
1950 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
1951 | !(mnt_flags & MNT_READONLY)) { | |
1952 | return -EPERM; | |
1953 | } | |
9566d674 EB |
1954 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
1955 | !(mnt_flags & MNT_NODEV)) { | |
1956 | return -EPERM; | |
1957 | } | |
1958 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
1959 | !(mnt_flags & MNT_NOSUID)) { | |
1960 | return -EPERM; | |
1961 | } | |
1962 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
1963 | !(mnt_flags & MNT_NOEXEC)) { | |
1964 | return -EPERM; | |
1965 | } | |
1966 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
1967 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
1968 | return -EPERM; | |
1969 | } | |
1970 | ||
ff36fe2c EP |
1971 | err = security_sb_remount(sb, data); |
1972 | if (err) | |
1973 | return err; | |
1974 | ||
1da177e4 | 1975 | down_write(&sb->s_umount); |
2e4b7fcd | 1976 | if (flags & MS_BIND) |
2d92ab3c | 1977 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1978 | else if (!capable(CAP_SYS_ADMIN)) |
1979 | err = -EPERM; | |
4aa98cf7 | 1980 | else |
2e4b7fcd | 1981 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1982 | if (!err) { |
719ea2fb | 1983 | lock_mount_hash(); |
a6138db8 | 1984 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 1985 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 1986 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 1987 | unlock_mount_hash(); |
0e55a7cc | 1988 | } |
6339dab8 | 1989 | up_write(&sb->s_umount); |
1da177e4 LT |
1990 | return err; |
1991 | } | |
1992 | ||
cbbe362c | 1993 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1994 | { |
315fc83e | 1995 | struct mount *p; |
909b0a88 | 1996 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1997 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1998 | return 1; |
1999 | } | |
2000 | return 0; | |
2001 | } | |
2002 | ||
808d4e3c | 2003 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2004 | { |
2d92ab3c | 2005 | struct path old_path, parent_path; |
676da58d | 2006 | struct mount *p; |
0fb54e50 | 2007 | struct mount *old; |
84d17192 | 2008 | struct mountpoint *mp; |
57eccb83 | 2009 | int err; |
1da177e4 LT |
2010 | if (!old_name || !*old_name) |
2011 | return -EINVAL; | |
2d92ab3c | 2012 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2013 | if (err) |
2014 | return err; | |
2015 | ||
84d17192 AV |
2016 | mp = lock_mount(path); |
2017 | err = PTR_ERR(mp); | |
2018 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2019 | goto out; |
2020 | ||
143c8c91 | 2021 | old = real_mount(old_path.mnt); |
fc7be130 | 2022 | p = real_mount(path->mnt); |
143c8c91 | 2023 | |
1da177e4 | 2024 | err = -EINVAL; |
fc7be130 | 2025 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2026 | goto out1; |
2027 | ||
5ff9d8a6 EB |
2028 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2029 | goto out1; | |
2030 | ||
1da177e4 | 2031 | err = -EINVAL; |
2d92ab3c | 2032 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2033 | goto out1; |
1da177e4 | 2034 | |
676da58d | 2035 | if (!mnt_has_parent(old)) |
21444403 | 2036 | goto out1; |
1da177e4 | 2037 | |
2d92ab3c AV |
2038 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
2039 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
2040 | goto out1; |
2041 | /* | |
2042 | * Don't move a mount residing in a shared parent. | |
2043 | */ | |
fc7be130 | 2044 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2045 | goto out1; |
9676f0c6 RP |
2046 | /* |
2047 | * Don't move a mount tree containing unbindable mounts to a destination | |
2048 | * mount which is shared. | |
2049 | */ | |
fc7be130 | 2050 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2051 | goto out1; |
1da177e4 | 2052 | err = -ELOOP; |
fc7be130 | 2053 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2054 | if (p == old) |
21444403 | 2055 | goto out1; |
1da177e4 | 2056 | |
84d17192 | 2057 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2058 | if (err) |
21444403 | 2059 | goto out1; |
1da177e4 LT |
2060 | |
2061 | /* if the mount is moved, it should no longer be expire | |
2062 | * automatically */ | |
6776db3d | 2063 | list_del_init(&old->mnt_expire); |
1da177e4 | 2064 | out1: |
84d17192 | 2065 | unlock_mount(mp); |
1da177e4 | 2066 | out: |
1da177e4 | 2067 | if (!err) |
1a390689 | 2068 | path_put(&parent_path); |
2d92ab3c | 2069 | path_put(&old_path); |
1da177e4 LT |
2070 | return err; |
2071 | } | |
2072 | ||
9d412a43 AV |
2073 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2074 | { | |
2075 | int err; | |
2076 | const char *subtype = strchr(fstype, '.'); | |
2077 | if (subtype) { | |
2078 | subtype++; | |
2079 | err = -EINVAL; | |
2080 | if (!subtype[0]) | |
2081 | goto err; | |
2082 | } else | |
2083 | subtype = ""; | |
2084 | ||
2085 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2086 | err = -ENOMEM; | |
2087 | if (!mnt->mnt_sb->s_subtype) | |
2088 | goto err; | |
2089 | return mnt; | |
2090 | ||
2091 | err: | |
2092 | mntput(mnt); | |
2093 | return ERR_PTR(err); | |
2094 | } | |
2095 | ||
9d412a43 AV |
2096 | /* |
2097 | * add a mount into a namespace's mount tree | |
2098 | */ | |
95bc5f25 | 2099 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2100 | { |
84d17192 AV |
2101 | struct mountpoint *mp; |
2102 | struct mount *parent; | |
9d412a43 AV |
2103 | int err; |
2104 | ||
f2ebb3a9 | 2105 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2106 | |
84d17192 AV |
2107 | mp = lock_mount(path); |
2108 | if (IS_ERR(mp)) | |
2109 | return PTR_ERR(mp); | |
9d412a43 | 2110 | |
84d17192 | 2111 | parent = real_mount(path->mnt); |
9d412a43 | 2112 | err = -EINVAL; |
84d17192 | 2113 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2114 | /* that's acceptable only for automounts done in private ns */ |
2115 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2116 | goto unlock; | |
2117 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2118 | if (!parent->mnt_ns) |
156cacb1 AV |
2119 | goto unlock; |
2120 | } | |
9d412a43 AV |
2121 | |
2122 | /* Refuse the same filesystem on the same mount point */ | |
2123 | err = -EBUSY; | |
95bc5f25 | 2124 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2125 | path->mnt->mnt_root == path->dentry) |
2126 | goto unlock; | |
2127 | ||
2128 | err = -EINVAL; | |
95bc5f25 | 2129 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2130 | goto unlock; |
2131 | ||
95bc5f25 | 2132 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2133 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2134 | |
2135 | unlock: | |
84d17192 | 2136 | unlock_mount(mp); |
9d412a43 AV |
2137 | return err; |
2138 | } | |
b1e75df4 | 2139 | |
1da177e4 LT |
2140 | /* |
2141 | * create a new mount for userspace and request it to be added into the | |
2142 | * namespace's tree | |
2143 | */ | |
0c55cfc4 | 2144 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2145 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2146 | { |
0c55cfc4 | 2147 | struct file_system_type *type; |
9b40bc90 | 2148 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2149 | struct vfsmount *mnt; |
15f9a3f3 | 2150 | int err; |
1da177e4 | 2151 | |
0c55cfc4 | 2152 | if (!fstype) |
1da177e4 LT |
2153 | return -EINVAL; |
2154 | ||
0c55cfc4 EB |
2155 | type = get_fs_type(fstype); |
2156 | if (!type) | |
2157 | return -ENODEV; | |
2158 | ||
2159 | if (user_ns != &init_user_ns) { | |
2160 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2161 | put_filesystem(type); | |
2162 | return -EPERM; | |
2163 | } | |
2164 | /* Only in special cases allow devices from mounts | |
2165 | * created outside the initial user namespace. | |
2166 | */ | |
2167 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2168 | flags |= MS_NODEV; | |
9566d674 | 2169 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 EB |
2170 | } |
2171 | } | |
2172 | ||
2173 | mnt = vfs_kern_mount(type, flags, name, data); | |
2174 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2175 | !mnt->mnt_sb->s_subtype) | |
2176 | mnt = fs_set_subtype(mnt, fstype); | |
2177 | ||
2178 | put_filesystem(type); | |
1da177e4 LT |
2179 | if (IS_ERR(mnt)) |
2180 | return PTR_ERR(mnt); | |
2181 | ||
95bc5f25 | 2182 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2183 | if (err) |
2184 | mntput(mnt); | |
2185 | return err; | |
1da177e4 LT |
2186 | } |
2187 | ||
19a167af AV |
2188 | int finish_automount(struct vfsmount *m, struct path *path) |
2189 | { | |
6776db3d | 2190 | struct mount *mnt = real_mount(m); |
19a167af AV |
2191 | int err; |
2192 | /* The new mount record should have at least 2 refs to prevent it being | |
2193 | * expired before we get a chance to add it | |
2194 | */ | |
6776db3d | 2195 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2196 | |
2197 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2198 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2199 | err = -ELOOP; |
2200 | goto fail; | |
19a167af AV |
2201 | } |
2202 | ||
95bc5f25 | 2203 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2204 | if (!err) |
2205 | return 0; | |
2206 | fail: | |
2207 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2208 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2209 | namespace_lock(); |
6776db3d | 2210 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2211 | namespace_unlock(); |
19a167af | 2212 | } |
b1e75df4 AV |
2213 | mntput(m); |
2214 | mntput(m); | |
19a167af AV |
2215 | return err; |
2216 | } | |
2217 | ||
ea5b778a DH |
2218 | /** |
2219 | * mnt_set_expiry - Put a mount on an expiration list | |
2220 | * @mnt: The mount to list. | |
2221 | * @expiry_list: The list to add the mount to. | |
2222 | */ | |
2223 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2224 | { | |
97216be0 | 2225 | namespace_lock(); |
ea5b778a | 2226 | |
6776db3d | 2227 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2228 | |
97216be0 | 2229 | namespace_unlock(); |
ea5b778a DH |
2230 | } |
2231 | EXPORT_SYMBOL(mnt_set_expiry); | |
2232 | ||
1da177e4 LT |
2233 | /* |
2234 | * process a list of expirable mountpoints with the intent of discarding any | |
2235 | * mountpoints that aren't in use and haven't been touched since last we came | |
2236 | * here | |
2237 | */ | |
2238 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2239 | { | |
761d5c38 | 2240 | struct mount *mnt, *next; |
1da177e4 LT |
2241 | LIST_HEAD(graveyard); |
2242 | ||
2243 | if (list_empty(mounts)) | |
2244 | return; | |
2245 | ||
97216be0 | 2246 | namespace_lock(); |
719ea2fb | 2247 | lock_mount_hash(); |
1da177e4 LT |
2248 | |
2249 | /* extract from the expiration list every vfsmount that matches the | |
2250 | * following criteria: | |
2251 | * - only referenced by its parent vfsmount | |
2252 | * - still marked for expiry (marked on the last call here; marks are | |
2253 | * cleared by mntput()) | |
2254 | */ | |
6776db3d | 2255 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2256 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2257 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2258 | continue; |
6776db3d | 2259 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2260 | } |
bcc5c7d2 | 2261 | while (!list_empty(&graveyard)) { |
6776db3d | 2262 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2263 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2264 | umount_tree(mnt, 1); |
bcc5c7d2 | 2265 | } |
719ea2fb | 2266 | unlock_mount_hash(); |
3ab6abee | 2267 | namespace_unlock(); |
5528f911 TM |
2268 | } |
2269 | ||
2270 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2271 | ||
2272 | /* | |
2273 | * Ripoff of 'select_parent()' | |
2274 | * | |
2275 | * search the list of submounts for a given mountpoint, and move any | |
2276 | * shrinkable submounts to the 'graveyard' list. | |
2277 | */ | |
692afc31 | 2278 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2279 | { |
692afc31 | 2280 | struct mount *this_parent = parent; |
5528f911 TM |
2281 | struct list_head *next; |
2282 | int found = 0; | |
2283 | ||
2284 | repeat: | |
6b41d536 | 2285 | next = this_parent->mnt_mounts.next; |
5528f911 | 2286 | resume: |
6b41d536 | 2287 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2288 | struct list_head *tmp = next; |
6b41d536 | 2289 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2290 | |
2291 | next = tmp->next; | |
692afc31 | 2292 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2293 | continue; |
5528f911 TM |
2294 | /* |
2295 | * Descend a level if the d_mounts list is non-empty. | |
2296 | */ | |
6b41d536 | 2297 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2298 | this_parent = mnt; |
2299 | goto repeat; | |
2300 | } | |
1da177e4 | 2301 | |
1ab59738 | 2302 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2303 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2304 | found++; |
2305 | } | |
1da177e4 | 2306 | } |
5528f911 TM |
2307 | /* |
2308 | * All done at this level ... ascend and resume the search | |
2309 | */ | |
2310 | if (this_parent != parent) { | |
6b41d536 | 2311 | next = this_parent->mnt_child.next; |
0714a533 | 2312 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2313 | goto resume; |
2314 | } | |
2315 | return found; | |
2316 | } | |
2317 | ||
2318 | /* | |
2319 | * process a list of expirable mountpoints with the intent of discarding any | |
2320 | * submounts of a specific parent mountpoint | |
99b7db7b | 2321 | * |
48a066e7 | 2322 | * mount_lock must be held for write |
5528f911 | 2323 | */ |
b54b9be7 | 2324 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2325 | { |
2326 | LIST_HEAD(graveyard); | |
761d5c38 | 2327 | struct mount *m; |
5528f911 | 2328 | |
5528f911 | 2329 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2330 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2331 | while (!list_empty(&graveyard)) { |
761d5c38 | 2332 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2333 | mnt_expire); |
143c8c91 | 2334 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2335 | umount_tree(m, 1); |
bcc5c7d2 AV |
2336 | } |
2337 | } | |
1da177e4 LT |
2338 | } |
2339 | ||
1da177e4 LT |
2340 | /* |
2341 | * Some copy_from_user() implementations do not return the exact number of | |
2342 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2343 | * Note that this function differs from copy_from_user() in that it will oops | |
2344 | * on bad values of `to', rather than returning a short copy. | |
2345 | */ | |
b58fed8b RP |
2346 | static long exact_copy_from_user(void *to, const void __user * from, |
2347 | unsigned long n) | |
1da177e4 LT |
2348 | { |
2349 | char *t = to; | |
2350 | const char __user *f = from; | |
2351 | char c; | |
2352 | ||
2353 | if (!access_ok(VERIFY_READ, from, n)) | |
2354 | return n; | |
2355 | ||
2356 | while (n) { | |
2357 | if (__get_user(c, f)) { | |
2358 | memset(t, 0, n); | |
2359 | break; | |
2360 | } | |
2361 | *t++ = c; | |
2362 | f++; | |
2363 | n--; | |
2364 | } | |
2365 | return n; | |
2366 | } | |
2367 | ||
b58fed8b | 2368 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2369 | { |
2370 | int i; | |
2371 | unsigned long page; | |
2372 | unsigned long size; | |
b58fed8b | 2373 | |
1da177e4 LT |
2374 | *where = 0; |
2375 | if (!data) | |
2376 | return 0; | |
2377 | ||
2378 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2379 | return -ENOMEM; | |
2380 | ||
2381 | /* We only care that *some* data at the address the user | |
2382 | * gave us is valid. Just in case, we'll zero | |
2383 | * the remainder of the page. | |
2384 | */ | |
2385 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2386 | size = TASK_SIZE - (unsigned long)data; | |
2387 | if (size > PAGE_SIZE) | |
2388 | size = PAGE_SIZE; | |
2389 | ||
2390 | i = size - exact_copy_from_user((void *)page, data, size); | |
2391 | if (!i) { | |
b58fed8b | 2392 | free_page(page); |
1da177e4 LT |
2393 | return -EFAULT; |
2394 | } | |
2395 | if (i != PAGE_SIZE) | |
2396 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2397 | *where = page; | |
2398 | return 0; | |
2399 | } | |
2400 | ||
eca6f534 VN |
2401 | int copy_mount_string(const void __user *data, char **where) |
2402 | { | |
2403 | char *tmp; | |
2404 | ||
2405 | if (!data) { | |
2406 | *where = NULL; | |
2407 | return 0; | |
2408 | } | |
2409 | ||
2410 | tmp = strndup_user(data, PAGE_SIZE); | |
2411 | if (IS_ERR(tmp)) | |
2412 | return PTR_ERR(tmp); | |
2413 | ||
2414 | *where = tmp; | |
2415 | return 0; | |
2416 | } | |
2417 | ||
1da177e4 LT |
2418 | /* |
2419 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2420 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2421 | * | |
2422 | * data is a (void *) that can point to any structure up to | |
2423 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2424 | * information (or be NULL). | |
2425 | * | |
2426 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2427 | * When the flags word was introduced its top half was required | |
2428 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2429 | * Therefore, if this magic number is present, it carries no information | |
2430 | * and must be discarded. | |
2431 | */ | |
808d4e3c AV |
2432 | long do_mount(const char *dev_name, const char *dir_name, |
2433 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2434 | { |
2d92ab3c | 2435 | struct path path; |
1da177e4 LT |
2436 | int retval = 0; |
2437 | int mnt_flags = 0; | |
2438 | ||
2439 | /* Discard magic */ | |
2440 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2441 | flags &= ~MS_MGC_MSK; | |
2442 | ||
2443 | /* Basic sanity checks */ | |
2444 | ||
2445 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2446 | return -EINVAL; | |
1da177e4 LT |
2447 | |
2448 | if (data_page) | |
2449 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2450 | ||
a27ab9f2 TH |
2451 | /* ... and get the mountpoint */ |
2452 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2453 | if (retval) | |
2454 | return retval; | |
2455 | ||
2456 | retval = security_sb_mount(dev_name, &path, | |
2457 | type_page, flags, data_page); | |
0d5cadb8 AV |
2458 | if (!retval && !may_mount()) |
2459 | retval = -EPERM; | |
a27ab9f2 TH |
2460 | if (retval) |
2461 | goto dput_out; | |
2462 | ||
613cbe3d AK |
2463 | /* Default to relatime unless overriden */ |
2464 | if (!(flags & MS_NOATIME)) | |
2465 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2466 | |
1da177e4 LT |
2467 | /* Separate the per-mountpoint flags */ |
2468 | if (flags & MS_NOSUID) | |
2469 | mnt_flags |= MNT_NOSUID; | |
2470 | if (flags & MS_NODEV) | |
2471 | mnt_flags |= MNT_NODEV; | |
2472 | if (flags & MS_NOEXEC) | |
2473 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2474 | if (flags & MS_NOATIME) |
2475 | mnt_flags |= MNT_NOATIME; | |
2476 | if (flags & MS_NODIRATIME) | |
2477 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2478 | if (flags & MS_STRICTATIME) |
2479 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2480 | if (flags & MS_RDONLY) |
2481 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2482 | |
ffbc6f0e EB |
2483 | /* The default atime for remount is preservation */ |
2484 | if ((flags & MS_REMOUNT) && | |
2485 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2486 | MS_STRICTATIME)) == 0)) { | |
2487 | mnt_flags &= ~MNT_ATIME_MASK; | |
2488 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2489 | } | |
2490 | ||
7a4dec53 | 2491 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2492 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2493 | MS_STRICTATIME); | |
1da177e4 | 2494 | |
1da177e4 | 2495 | if (flags & MS_REMOUNT) |
2d92ab3c | 2496 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2497 | data_page); |
2498 | else if (flags & MS_BIND) | |
2d92ab3c | 2499 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2500 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2501 | retval = do_change_type(&path, flags); |
1da177e4 | 2502 | else if (flags & MS_MOVE) |
2d92ab3c | 2503 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2504 | else |
2d92ab3c | 2505 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2506 | dev_name, data_page); |
2507 | dput_out: | |
2d92ab3c | 2508 | path_put(&path); |
1da177e4 LT |
2509 | return retval; |
2510 | } | |
2511 | ||
771b1371 EB |
2512 | static void free_mnt_ns(struct mnt_namespace *ns) |
2513 | { | |
98f842e6 | 2514 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2515 | put_user_ns(ns->user_ns); |
2516 | kfree(ns); | |
2517 | } | |
2518 | ||
8823c079 EB |
2519 | /* |
2520 | * Assign a sequence number so we can detect when we attempt to bind | |
2521 | * mount a reference to an older mount namespace into the current | |
2522 | * mount namespace, preventing reference counting loops. A 64bit | |
2523 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2524 | * is effectively never, so we can ignore the possibility. | |
2525 | */ | |
2526 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2527 | ||
771b1371 | 2528 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2529 | { |
2530 | struct mnt_namespace *new_ns; | |
98f842e6 | 2531 | int ret; |
cf8d2c11 TM |
2532 | |
2533 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2534 | if (!new_ns) | |
2535 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2536 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2537 | if (ret) { | |
2538 | kfree(new_ns); | |
2539 | return ERR_PTR(ret); | |
2540 | } | |
8823c079 | 2541 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2542 | atomic_set(&new_ns->count, 1); |
2543 | new_ns->root = NULL; | |
2544 | INIT_LIST_HEAD(&new_ns->list); | |
2545 | init_waitqueue_head(&new_ns->poll); | |
2546 | new_ns->event = 0; | |
771b1371 | 2547 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2548 | return new_ns; |
2549 | } | |
2550 | ||
9559f689 AV |
2551 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2552 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2553 | { |
6b3286ed | 2554 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2555 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2556 | struct mount *p, *q; |
9559f689 | 2557 | struct mount *old; |
cb338d06 | 2558 | struct mount *new; |
7a472ef4 | 2559 | int copy_flags; |
1da177e4 | 2560 | |
9559f689 AV |
2561 | BUG_ON(!ns); |
2562 | ||
2563 | if (likely(!(flags & CLONE_NEWNS))) { | |
2564 | get_mnt_ns(ns); | |
2565 | return ns; | |
2566 | } | |
2567 | ||
2568 | old = ns->root; | |
2569 | ||
771b1371 | 2570 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2571 | if (IS_ERR(new_ns)) |
2572 | return new_ns; | |
1da177e4 | 2573 | |
97216be0 | 2574 | namespace_lock(); |
1da177e4 | 2575 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2576 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2577 | if (user_ns != ns->user_ns) |
132c94e3 | 2578 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2579 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2580 | if (IS_ERR(new)) { |
328e6d90 | 2581 | namespace_unlock(); |
771b1371 | 2582 | free_mnt_ns(new_ns); |
be34d1a3 | 2583 | return ERR_CAST(new); |
1da177e4 | 2584 | } |
be08d6d2 | 2585 | new_ns->root = new; |
1a4eeaf2 | 2586 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2587 | |
2588 | /* | |
2589 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2590 | * as belonging to new namespace. We have already acquired a private | |
2591 | * fs_struct, so tsk->fs->lock is not needed. | |
2592 | */ | |
909b0a88 | 2593 | p = old; |
cb338d06 | 2594 | q = new; |
1da177e4 | 2595 | while (p) { |
143c8c91 | 2596 | q->mnt_ns = new_ns; |
9559f689 AV |
2597 | if (new_fs) { |
2598 | if (&p->mnt == new_fs->root.mnt) { | |
2599 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2600 | rootmnt = &p->mnt; |
1da177e4 | 2601 | } |
9559f689 AV |
2602 | if (&p->mnt == new_fs->pwd.mnt) { |
2603 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2604 | pwdmnt = &p->mnt; |
1da177e4 | 2605 | } |
1da177e4 | 2606 | } |
909b0a88 AV |
2607 | p = next_mnt(p, old); |
2608 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2609 | if (!q) |
2610 | break; | |
2611 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2612 | p = next_mnt(p, old); | |
1da177e4 | 2613 | } |
328e6d90 | 2614 | namespace_unlock(); |
1da177e4 | 2615 | |
1da177e4 | 2616 | if (rootmnt) |
f03c6599 | 2617 | mntput(rootmnt); |
1da177e4 | 2618 | if (pwdmnt) |
f03c6599 | 2619 | mntput(pwdmnt); |
1da177e4 | 2620 | |
741a2951 | 2621 | return new_ns; |
1da177e4 LT |
2622 | } |
2623 | ||
cf8d2c11 TM |
2624 | /** |
2625 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2626 | * @mnt: pointer to the new root filesystem mountpoint | |
2627 | */ | |
1a4eeaf2 | 2628 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2629 | { |
771b1371 | 2630 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2631 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2632 | struct mount *mnt = real_mount(m); |
2633 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2634 | new_ns->root = mnt; |
b1983cd8 | 2635 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2636 | } else { |
1a4eeaf2 | 2637 | mntput(m); |
cf8d2c11 TM |
2638 | } |
2639 | return new_ns; | |
2640 | } | |
cf8d2c11 | 2641 | |
ea441d11 AV |
2642 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2643 | { | |
2644 | struct mnt_namespace *ns; | |
d31da0f0 | 2645 | struct super_block *s; |
ea441d11 AV |
2646 | struct path path; |
2647 | int err; | |
2648 | ||
2649 | ns = create_mnt_ns(mnt); | |
2650 | if (IS_ERR(ns)) | |
2651 | return ERR_CAST(ns); | |
2652 | ||
2653 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2654 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2655 | ||
2656 | put_mnt_ns(ns); | |
2657 | ||
2658 | if (err) | |
2659 | return ERR_PTR(err); | |
2660 | ||
2661 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2662 | s = path.mnt->mnt_sb; |
2663 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2664 | mntput(path.mnt); |
2665 | /* lock the sucker */ | |
d31da0f0 | 2666 | down_write(&s->s_umount); |
ea441d11 AV |
2667 | /* ... and return the root of (sub)tree on it */ |
2668 | return path.dentry; | |
2669 | } | |
2670 | EXPORT_SYMBOL(mount_subtree); | |
2671 | ||
bdc480e3 HC |
2672 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2673 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2674 | { |
eca6f534 VN |
2675 | int ret; |
2676 | char *kernel_type; | |
91a27b2a | 2677 | struct filename *kernel_dir; |
eca6f534 | 2678 | char *kernel_dev; |
1da177e4 | 2679 | unsigned long data_page; |
1da177e4 | 2680 | |
eca6f534 VN |
2681 | ret = copy_mount_string(type, &kernel_type); |
2682 | if (ret < 0) | |
2683 | goto out_type; | |
1da177e4 | 2684 | |
eca6f534 VN |
2685 | kernel_dir = getname(dir_name); |
2686 | if (IS_ERR(kernel_dir)) { | |
2687 | ret = PTR_ERR(kernel_dir); | |
2688 | goto out_dir; | |
2689 | } | |
1da177e4 | 2690 | |
eca6f534 VN |
2691 | ret = copy_mount_string(dev_name, &kernel_dev); |
2692 | if (ret < 0) | |
2693 | goto out_dev; | |
1da177e4 | 2694 | |
eca6f534 VN |
2695 | ret = copy_mount_options(data, &data_page); |
2696 | if (ret < 0) | |
2697 | goto out_data; | |
1da177e4 | 2698 | |
91a27b2a | 2699 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2700 | (void *) data_page); |
1da177e4 | 2701 | |
eca6f534 VN |
2702 | free_page(data_page); |
2703 | out_data: | |
2704 | kfree(kernel_dev); | |
2705 | out_dev: | |
2706 | putname(kernel_dir); | |
2707 | out_dir: | |
2708 | kfree(kernel_type); | |
2709 | out_type: | |
2710 | return ret; | |
1da177e4 LT |
2711 | } |
2712 | ||
afac7cba AV |
2713 | /* |
2714 | * Return true if path is reachable from root | |
2715 | * | |
48a066e7 | 2716 | * namespace_sem or mount_lock is held |
afac7cba | 2717 | */ |
643822b4 | 2718 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2719 | const struct path *root) |
2720 | { | |
643822b4 | 2721 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2722 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2723 | mnt = mnt->mnt_parent; |
afac7cba | 2724 | } |
643822b4 | 2725 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2726 | } |
2727 | ||
2728 | int path_is_under(struct path *path1, struct path *path2) | |
2729 | { | |
2730 | int res; | |
48a066e7 | 2731 | read_seqlock_excl(&mount_lock); |
643822b4 | 2732 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2733 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2734 | return res; |
2735 | } | |
2736 | EXPORT_SYMBOL(path_is_under); | |
2737 | ||
1da177e4 LT |
2738 | /* |
2739 | * pivot_root Semantics: | |
2740 | * Moves the root file system of the current process to the directory put_old, | |
2741 | * makes new_root as the new root file system of the current process, and sets | |
2742 | * root/cwd of all processes which had them on the current root to new_root. | |
2743 | * | |
2744 | * Restrictions: | |
2745 | * The new_root and put_old must be directories, and must not be on the | |
2746 | * same file system as the current process root. The put_old must be | |
2747 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2748 | * pointed to by put_old must yield the same directory as new_root. No other | |
2749 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2750 | * | |
4a0d11fa NB |
2751 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2752 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2753 | * in this situation. | |
2754 | * | |
1da177e4 LT |
2755 | * Notes: |
2756 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2757 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2758 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2759 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2760 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2761 | * first. | |
2762 | */ | |
3480b257 HC |
2763 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2764 | const char __user *, put_old) | |
1da177e4 | 2765 | { |
2d8f3038 | 2766 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2767 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2768 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2769 | int error; |
2770 | ||
9b40bc90 | 2771 | if (!may_mount()) |
1da177e4 LT |
2772 | return -EPERM; |
2773 | ||
2d8f3038 | 2774 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2775 | if (error) |
2776 | goto out0; | |
1da177e4 | 2777 | |
2d8f3038 | 2778 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2779 | if (error) |
2780 | goto out1; | |
2781 | ||
2d8f3038 | 2782 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2783 | if (error) |
2784 | goto out2; | |
1da177e4 | 2785 | |
f7ad3c6b | 2786 | get_fs_root(current->fs, &root); |
84d17192 AV |
2787 | old_mp = lock_mount(&old); |
2788 | error = PTR_ERR(old_mp); | |
2789 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2790 | goto out3; |
2791 | ||
1da177e4 | 2792 | error = -EINVAL; |
419148da AV |
2793 | new_mnt = real_mount(new.mnt); |
2794 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2795 | old_mnt = real_mount(old.mnt); |
2796 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2797 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2798 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2799 | goto out4; |
143c8c91 | 2800 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2801 | goto out4; |
5ff9d8a6 EB |
2802 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2803 | goto out4; | |
1da177e4 | 2804 | error = -ENOENT; |
f3da392e | 2805 | if (d_unlinked(new.dentry)) |
b12cea91 | 2806 | goto out4; |
1da177e4 | 2807 | error = -EBUSY; |
84d17192 | 2808 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2809 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2810 | error = -EINVAL; |
8c3ee42e | 2811 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2812 | goto out4; /* not a mountpoint */ |
676da58d | 2813 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2814 | goto out4; /* not attached */ |
84d17192 | 2815 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2816 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2817 | goto out4; /* not a mountpoint */ |
676da58d | 2818 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2819 | goto out4; /* not attached */ |
4ac91378 | 2820 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2821 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2822 | goto out4; |
84d17192 | 2823 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2824 | lock_mount_hash(); |
419148da AV |
2825 | detach_mnt(new_mnt, &parent_path); |
2826 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2827 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2828 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2829 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2830 | } | |
4ac91378 | 2831 | /* mount old root on put_old */ |
84d17192 | 2832 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2833 | /* mount new_root on / */ |
84d17192 | 2834 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2835 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2836 | unlock_mount_hash(); |
2d8f3038 | 2837 | chroot_fs_refs(&root, &new); |
84d17192 | 2838 | put_mountpoint(root_mp); |
1da177e4 | 2839 | error = 0; |
b12cea91 | 2840 | out4: |
84d17192 | 2841 | unlock_mount(old_mp); |
b12cea91 AV |
2842 | if (!error) { |
2843 | path_put(&root_parent); | |
2844 | path_put(&parent_path); | |
2845 | } | |
2846 | out3: | |
8c3ee42e | 2847 | path_put(&root); |
b12cea91 | 2848 | out2: |
2d8f3038 | 2849 | path_put(&old); |
1da177e4 | 2850 | out1: |
2d8f3038 | 2851 | path_put(&new); |
1da177e4 | 2852 | out0: |
1da177e4 | 2853 | return error; |
1da177e4 LT |
2854 | } |
2855 | ||
2856 | static void __init init_mount_tree(void) | |
2857 | { | |
2858 | struct vfsmount *mnt; | |
6b3286ed | 2859 | struct mnt_namespace *ns; |
ac748a09 | 2860 | struct path root; |
0c55cfc4 | 2861 | struct file_system_type *type; |
1da177e4 | 2862 | |
0c55cfc4 EB |
2863 | type = get_fs_type("rootfs"); |
2864 | if (!type) | |
2865 | panic("Can't find rootfs type"); | |
2866 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2867 | put_filesystem(type); | |
1da177e4 LT |
2868 | if (IS_ERR(mnt)) |
2869 | panic("Can't create rootfs"); | |
b3e19d92 | 2870 | |
3b22edc5 TM |
2871 | ns = create_mnt_ns(mnt); |
2872 | if (IS_ERR(ns)) | |
1da177e4 | 2873 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2874 | |
2875 | init_task.nsproxy->mnt_ns = ns; | |
2876 | get_mnt_ns(ns); | |
2877 | ||
be08d6d2 AV |
2878 | root.mnt = mnt; |
2879 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2880 | |
2881 | set_fs_pwd(current->fs, &root); | |
2882 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2883 | } |
2884 | ||
74bf17cf | 2885 | void __init mnt_init(void) |
1da177e4 | 2886 | { |
13f14b4d | 2887 | unsigned u; |
15a67dd8 | 2888 | int err; |
1da177e4 | 2889 | |
7d6fec45 | 2890 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2891 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2892 | |
0818bf27 | 2893 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2894 | sizeof(struct hlist_head), |
0818bf27 AV |
2895 | mhash_entries, 19, |
2896 | 0, | |
2897 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2898 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2899 | sizeof(struct hlist_head), | |
2900 | mphash_entries, 19, | |
2901 | 0, | |
2902 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2903 | |
84d17192 | 2904 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2905 | panic("Failed to allocate mount hash table\n"); |
2906 | ||
0818bf27 | 2907 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2908 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2909 | for (u = 0; u <= mp_hash_mask; u++) |
2910 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2911 | |
4b93dc9b TH |
2912 | kernfs_init(); |
2913 | ||
15a67dd8 RD |
2914 | err = sysfs_init(); |
2915 | if (err) | |
2916 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2917 | __func__, err); |
00d26666 GKH |
2918 | fs_kobj = kobject_create_and_add("fs", NULL); |
2919 | if (!fs_kobj) | |
8e24eea7 | 2920 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2921 | init_rootfs(); |
2922 | init_mount_tree(); | |
2923 | } | |
2924 | ||
616511d0 | 2925 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2926 | { |
d498b25a | 2927 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2928 | return; |
7b00ed6f | 2929 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2930 | free_mnt_ns(ns); |
1da177e4 | 2931 | } |
9d412a43 AV |
2932 | |
2933 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2934 | { | |
423e0ab0 TC |
2935 | struct vfsmount *mnt; |
2936 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2937 | if (!IS_ERR(mnt)) { | |
2938 | /* | |
2939 | * it is a longterm mount, don't release mnt until | |
2940 | * we unmount before file sys is unregistered | |
2941 | */ | |
f7a99c5b | 2942 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2943 | } |
2944 | return mnt; | |
9d412a43 AV |
2945 | } |
2946 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2947 | |
2948 | void kern_unmount(struct vfsmount *mnt) | |
2949 | { | |
2950 | /* release long term mount so mount point can be released */ | |
2951 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2952 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2953 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2954 | mntput(mnt); |
2955 | } | |
2956 | } | |
2957 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2958 | |
2959 | bool our_mnt(struct vfsmount *mnt) | |
2960 | { | |
143c8c91 | 2961 | return check_mnt(real_mount(mnt)); |
02125a82 | 2962 | } |
8823c079 | 2963 | |
3151527e EB |
2964 | bool current_chrooted(void) |
2965 | { | |
2966 | /* Does the current process have a non-standard root */ | |
2967 | struct path ns_root; | |
2968 | struct path fs_root; | |
2969 | bool chrooted; | |
2970 | ||
2971 | /* Find the namespace root */ | |
2972 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2973 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2974 | path_get(&ns_root); | |
2975 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2976 | ; | |
2977 | ||
2978 | get_fs_root(current->fs, &fs_root); | |
2979 | ||
2980 | chrooted = !path_equal(&fs_root, &ns_root); | |
2981 | ||
2982 | path_put(&fs_root); | |
2983 | path_put(&ns_root); | |
2984 | ||
2985 | return chrooted; | |
2986 | } | |
2987 | ||
e51db735 | 2988 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
2989 | { |
2990 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2991 | struct mount *mnt; | |
e51db735 | 2992 | bool visible = false; |
87a8ebd6 | 2993 | |
e51db735 EB |
2994 | if (unlikely(!ns)) |
2995 | return false; | |
2996 | ||
44bb4385 | 2997 | down_read(&namespace_sem); |
87a8ebd6 | 2998 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
2999 | struct mount *child; |
3000 | if (mnt->mnt.mnt_sb->s_type != type) | |
3001 | continue; | |
3002 | ||
3003 | /* This mount is not fully visible if there are any child mounts | |
3004 | * that cover anything except for empty directories. | |
3005 | */ | |
3006 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3007 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
3008 | if (!S_ISDIR(inode->i_mode)) | |
3009 | goto next; | |
41301ae7 | 3010 | if (inode->i_nlink > 2) |
e51db735 | 3011 | goto next; |
87a8ebd6 | 3012 | } |
e51db735 EB |
3013 | visible = true; |
3014 | goto found; | |
3015 | next: ; | |
87a8ebd6 | 3016 | } |
e51db735 | 3017 | found: |
44bb4385 | 3018 | up_read(&namespace_sem); |
e51db735 | 3019 | return visible; |
87a8ebd6 EB |
3020 | } |
3021 | ||
8823c079 EB |
3022 | static void *mntns_get(struct task_struct *task) |
3023 | { | |
3024 | struct mnt_namespace *ns = NULL; | |
3025 | struct nsproxy *nsproxy; | |
3026 | ||
728dba3a EB |
3027 | task_lock(task); |
3028 | nsproxy = task->nsproxy; | |
8823c079 EB |
3029 | if (nsproxy) { |
3030 | ns = nsproxy->mnt_ns; | |
3031 | get_mnt_ns(ns); | |
3032 | } | |
728dba3a | 3033 | task_unlock(task); |
8823c079 EB |
3034 | |
3035 | return ns; | |
3036 | } | |
3037 | ||
3038 | static void mntns_put(void *ns) | |
3039 | { | |
3040 | put_mnt_ns(ns); | |
3041 | } | |
3042 | ||
3043 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
3044 | { | |
3045 | struct fs_struct *fs = current->fs; | |
3046 | struct mnt_namespace *mnt_ns = ns; | |
3047 | struct path root; | |
3048 | ||
0c55cfc4 | 3049 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3050 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3051 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3052 | return -EPERM; |
8823c079 EB |
3053 | |
3054 | if (fs->users != 1) | |
3055 | return -EINVAL; | |
3056 | ||
3057 | get_mnt_ns(mnt_ns); | |
3058 | put_mnt_ns(nsproxy->mnt_ns); | |
3059 | nsproxy->mnt_ns = mnt_ns; | |
3060 | ||
3061 | /* Find the root */ | |
3062 | root.mnt = &mnt_ns->root->mnt; | |
3063 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3064 | path_get(&root); | |
3065 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3066 | ; | |
3067 | ||
3068 | /* Update the pwd and root */ | |
3069 | set_fs_pwd(fs, &root); | |
3070 | set_fs_root(fs, &root); | |
3071 | ||
3072 | path_put(&root); | |
3073 | return 0; | |
3074 | } | |
3075 | ||
98f842e6 EB |
3076 | static unsigned int mntns_inum(void *ns) |
3077 | { | |
3078 | struct mnt_namespace *mnt_ns = ns; | |
3079 | return mnt_ns->proc_inum; | |
3080 | } | |
3081 | ||
8823c079 EB |
3082 | const struct proc_ns_operations mntns_operations = { |
3083 | .name = "mnt", | |
3084 | .type = CLONE_NEWNS, | |
3085 | .get = mntns_get, | |
3086 | .put = mntns_put, | |
3087 | .install = mntns_install, | |
98f842e6 | 3088 | .inum = mntns_inum, |
8823c079 | 3089 | }; |