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> |
d10577a8 | 19 | #include <linux/acct.h> /* acct_auto_close_mnt */ |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
0818bf27 | 26 | #include <linux/bootmem.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
0818bf27 AV |
30 | static unsigned int m_hash_mask __read_mostly; |
31 | static unsigned int m_hash_shift __read_mostly; | |
32 | static unsigned int mp_hash_mask __read_mostly; | |
33 | static unsigned int mp_hash_shift __read_mostly; | |
34 | ||
35 | static __initdata unsigned long mhash_entries; | |
36 | static int __init set_mhash_entries(char *str) | |
37 | { | |
38 | if (!str) | |
39 | return 0; | |
40 | mhash_entries = simple_strtoul(str, &str, 0); | |
41 | return 1; | |
42 | } | |
43 | __setup("mhash_entries=", set_mhash_entries); | |
44 | ||
45 | static __initdata unsigned long mphash_entries; | |
46 | static int __init set_mphash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mphash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 54 | |
c7999c36 | 55 | static u64 event; |
73cd49ec | 56 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 57 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 58 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
59 | static int mnt_id_start = 0; |
60 | static int mnt_group_start = 1; | |
1da177e4 | 61 | |
38129a13 | 62 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 63 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 64 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 65 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 66 | |
f87fd4c2 | 67 | /* /sys/fs */ |
00d26666 GKH |
68 | struct kobject *fs_kobj; |
69 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 70 | |
99b7db7b NP |
71 | /* |
72 | * vfsmount lock may be taken for read to prevent changes to the | |
73 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
74 | * up the tree. | |
75 | * | |
76 | * It should be taken for write in all cases where the vfsmount | |
77 | * tree or hash is modified or when a vfsmount structure is modified. | |
78 | */ | |
48a066e7 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 80 | |
38129a13 | 81 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 82 | { |
b58fed8b RP |
83 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
84 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
85 | tmp = tmp + (tmp >> m_hash_shift); |
86 | return &mount_hashtable[tmp & m_hash_mask]; | |
87 | } | |
88 | ||
89 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
90 | { | |
91 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
92 | tmp = tmp + (tmp >> mp_hash_shift); | |
93 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
94 | } |
95 | ||
99b7db7b NP |
96 | /* |
97 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
98 | * serialize with freeing. | |
99 | */ | |
b105e270 | 100 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
101 | { |
102 | int res; | |
103 | ||
104 | retry: | |
105 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 106 | spin_lock(&mnt_id_lock); |
15169fe7 | 107 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 108 | if (!res) |
15169fe7 | 109 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 110 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
111 | if (res == -EAGAIN) |
112 | goto retry; | |
113 | ||
114 | return res; | |
115 | } | |
116 | ||
b105e270 | 117 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 118 | { |
15169fe7 | 119 | int id = mnt->mnt_id; |
99b7db7b | 120 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
121 | ida_remove(&mnt_id_ida, id); |
122 | if (mnt_id_start > id) | |
123 | mnt_id_start = id; | |
99b7db7b | 124 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
125 | } |
126 | ||
719f5d7f MS |
127 | /* |
128 | * Allocate a new peer group ID | |
129 | * | |
130 | * mnt_group_ida is protected by namespace_sem | |
131 | */ | |
4b8b21f4 | 132 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 133 | { |
f21f6220 AV |
134 | int res; |
135 | ||
719f5d7f MS |
136 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
137 | return -ENOMEM; | |
138 | ||
f21f6220 AV |
139 | res = ida_get_new_above(&mnt_group_ida, |
140 | mnt_group_start, | |
15169fe7 | 141 | &mnt->mnt_group_id); |
f21f6220 | 142 | if (!res) |
15169fe7 | 143 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
144 | |
145 | return res; | |
719f5d7f MS |
146 | } |
147 | ||
148 | /* | |
149 | * Release a peer group ID | |
150 | */ | |
4b8b21f4 | 151 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 152 | { |
15169fe7 | 153 | int id = mnt->mnt_group_id; |
f21f6220 AV |
154 | ida_remove(&mnt_group_ida, id); |
155 | if (mnt_group_start > id) | |
156 | mnt_group_start = id; | |
15169fe7 | 157 | mnt->mnt_group_id = 0; |
719f5d7f MS |
158 | } |
159 | ||
b3e19d92 NP |
160 | /* |
161 | * vfsmount lock must be held for read | |
162 | */ | |
83adc753 | 163 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
164 | { |
165 | #ifdef CONFIG_SMP | |
68e8a9fe | 166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
167 | #else |
168 | preempt_disable(); | |
68e8a9fe | 169 | mnt->mnt_count += n; |
b3e19d92 NP |
170 | preempt_enable(); |
171 | #endif | |
172 | } | |
173 | ||
b3e19d92 NP |
174 | /* |
175 | * vfsmount lock must be held for write | |
176 | */ | |
83adc753 | 177 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
178 | { |
179 | #ifdef CONFIG_SMP | |
f03c6599 | 180 | unsigned int count = 0; |
b3e19d92 NP |
181 | int cpu; |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
185 | } |
186 | ||
187 | return count; | |
188 | #else | |
68e8a9fe | 189 | return mnt->mnt_count; |
b3e19d92 NP |
190 | #endif |
191 | } | |
192 | ||
b105e270 | 193 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 194 | { |
c63181e6 AV |
195 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
196 | if (mnt) { | |
73cd49ec MS |
197 | int err; |
198 | ||
c63181e6 | 199 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
200 | if (err) |
201 | goto out_free_cache; | |
202 | ||
203 | if (name) { | |
c63181e6 AV |
204 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
205 | if (!mnt->mnt_devname) | |
88b38782 | 206 | goto out_free_id; |
73cd49ec MS |
207 | } |
208 | ||
b3e19d92 | 209 | #ifdef CONFIG_SMP |
c63181e6 AV |
210 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
211 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
212 | goto out_free_devname; |
213 | ||
c63181e6 | 214 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 215 | #else |
c63181e6 AV |
216 | mnt->mnt_count = 1; |
217 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
218 | #endif |
219 | ||
38129a13 | 220 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
221 | INIT_LIST_HEAD(&mnt->mnt_child); |
222 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
223 | INIT_LIST_HEAD(&mnt->mnt_list); | |
224 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
225 | INIT_LIST_HEAD(&mnt->mnt_share); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
227 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
228 | #ifdef CONFIG_FSNOTIFY |
229 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 230 | #endif |
1da177e4 | 231 | } |
c63181e6 | 232 | return mnt; |
88b38782 | 233 | |
d3ef3d73 | 234 | #ifdef CONFIG_SMP |
235 | out_free_devname: | |
c63181e6 | 236 | kfree(mnt->mnt_devname); |
d3ef3d73 | 237 | #endif |
88b38782 | 238 | out_free_id: |
c63181e6 | 239 | mnt_free_id(mnt); |
88b38782 | 240 | out_free_cache: |
c63181e6 | 241 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 242 | return NULL; |
1da177e4 LT |
243 | } |
244 | ||
3d733633 DH |
245 | /* |
246 | * Most r/o checks on a fs are for operations that take | |
247 | * discrete amounts of time, like a write() or unlink(). | |
248 | * We must keep track of when those operations start | |
249 | * (for permission checks) and when they end, so that | |
250 | * we can determine when writes are able to occur to | |
251 | * a filesystem. | |
252 | */ | |
253 | /* | |
254 | * __mnt_is_readonly: check whether a mount is read-only | |
255 | * @mnt: the mount to check for its write status | |
256 | * | |
257 | * This shouldn't be used directly ouside of the VFS. | |
258 | * It does not guarantee that the filesystem will stay | |
259 | * r/w, just that it is right *now*. This can not and | |
260 | * should not be used in place of IS_RDONLY(inode). | |
261 | * mnt_want/drop_write() will _keep_ the filesystem | |
262 | * r/w. | |
263 | */ | |
264 | int __mnt_is_readonly(struct vfsmount *mnt) | |
265 | { | |
2e4b7fcd DH |
266 | if (mnt->mnt_flags & MNT_READONLY) |
267 | return 1; | |
268 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
269 | return 1; | |
270 | return 0; | |
3d733633 DH |
271 | } |
272 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
273 | ||
83adc753 | 274 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 275 | { |
276 | #ifdef CONFIG_SMP | |
68e8a9fe | 277 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 278 | #else |
68e8a9fe | 279 | mnt->mnt_writers++; |
d3ef3d73 | 280 | #endif |
281 | } | |
3d733633 | 282 | |
83adc753 | 283 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 284 | { |
d3ef3d73 | 285 | #ifdef CONFIG_SMP |
68e8a9fe | 286 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 287 | #else |
68e8a9fe | 288 | mnt->mnt_writers--; |
d3ef3d73 | 289 | #endif |
3d733633 | 290 | } |
3d733633 | 291 | |
83adc753 | 292 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 293 | { |
d3ef3d73 | 294 | #ifdef CONFIG_SMP |
295 | unsigned int count = 0; | |
3d733633 | 296 | int cpu; |
3d733633 DH |
297 | |
298 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 299 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 300 | } |
3d733633 | 301 | |
d3ef3d73 | 302 | return count; |
303 | #else | |
304 | return mnt->mnt_writers; | |
305 | #endif | |
3d733633 DH |
306 | } |
307 | ||
4ed5e82f MS |
308 | static int mnt_is_readonly(struct vfsmount *mnt) |
309 | { | |
310 | if (mnt->mnt_sb->s_readonly_remount) | |
311 | return 1; | |
312 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
313 | smp_rmb(); | |
314 | return __mnt_is_readonly(mnt); | |
315 | } | |
316 | ||
8366025e | 317 | /* |
eb04c282 JK |
318 | * Most r/o & frozen checks on a fs are for operations that take discrete |
319 | * amounts of time, like a write() or unlink(). We must keep track of when | |
320 | * those operations start (for permission checks) and when they end, so that we | |
321 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
322 | */ |
323 | /** | |
eb04c282 | 324 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 325 | * @m: the mount on which to take a write |
8366025e | 326 | * |
eb04c282 JK |
327 | * This tells the low-level filesystem that a write is about to be performed to |
328 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
329 | * returning success. This operation does not protect against filesystem being | |
330 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
331 | * called. This is effectively a refcount. | |
8366025e | 332 | */ |
eb04c282 | 333 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 334 | { |
83adc753 | 335 | struct mount *mnt = real_mount(m); |
3d733633 | 336 | int ret = 0; |
3d733633 | 337 | |
d3ef3d73 | 338 | preempt_disable(); |
c6653a83 | 339 | mnt_inc_writers(mnt); |
d3ef3d73 | 340 | /* |
c6653a83 | 341 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 342 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
343 | * incremented count after it has set MNT_WRITE_HOLD. | |
344 | */ | |
345 | smp_mb(); | |
1e75529e | 346 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 347 | cpu_relax(); |
348 | /* | |
349 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
350 | * be set to match its requirements. So we must not load that until | |
351 | * MNT_WRITE_HOLD is cleared. | |
352 | */ | |
353 | smp_rmb(); | |
4ed5e82f | 354 | if (mnt_is_readonly(m)) { |
c6653a83 | 355 | mnt_dec_writers(mnt); |
3d733633 | 356 | ret = -EROFS; |
3d733633 | 357 | } |
d3ef3d73 | 358 | preempt_enable(); |
eb04c282 JK |
359 | |
360 | return ret; | |
361 | } | |
362 | ||
363 | /** | |
364 | * mnt_want_write - get write access to a mount | |
365 | * @m: the mount on which to take a write | |
366 | * | |
367 | * This tells the low-level filesystem that a write is about to be performed to | |
368 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
369 | * is not frozen) before returning success. When the write operation is | |
370 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
371 | */ | |
372 | int mnt_want_write(struct vfsmount *m) | |
373 | { | |
374 | int ret; | |
375 | ||
376 | sb_start_write(m->mnt_sb); | |
377 | ret = __mnt_want_write(m); | |
378 | if (ret) | |
379 | sb_end_write(m->mnt_sb); | |
3d733633 | 380 | return ret; |
8366025e DH |
381 | } |
382 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
383 | ||
96029c4e | 384 | /** |
385 | * mnt_clone_write - get write access to a mount | |
386 | * @mnt: the mount on which to take a write | |
387 | * | |
388 | * This is effectively like mnt_want_write, except | |
389 | * it must only be used to take an extra write reference | |
390 | * on a mountpoint that we already know has a write reference | |
391 | * on it. This allows some optimisation. | |
392 | * | |
393 | * After finished, mnt_drop_write must be called as usual to | |
394 | * drop the reference. | |
395 | */ | |
396 | int mnt_clone_write(struct vfsmount *mnt) | |
397 | { | |
398 | /* superblock may be r/o */ | |
399 | if (__mnt_is_readonly(mnt)) | |
400 | return -EROFS; | |
401 | preempt_disable(); | |
83adc753 | 402 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 403 | preempt_enable(); |
404 | return 0; | |
405 | } | |
406 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
407 | ||
408 | /** | |
eb04c282 | 409 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 410 | * @file: the file who's mount on which to take a write |
411 | * | |
eb04c282 | 412 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 413 | * do some optimisations if the file is open for write already |
414 | */ | |
eb04c282 | 415 | int __mnt_want_write_file(struct file *file) |
96029c4e | 416 | { |
83f936c7 | 417 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 418 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 419 | else |
420 | return mnt_clone_write(file->f_path.mnt); | |
421 | } | |
eb04c282 JK |
422 | |
423 | /** | |
424 | * mnt_want_write_file - get write access to a file's mount | |
425 | * @file: the file who's mount on which to take a write | |
426 | * | |
427 | * This is like mnt_want_write, but it takes a file and can | |
428 | * do some optimisations if the file is open for write already | |
429 | */ | |
430 | int mnt_want_write_file(struct file *file) | |
431 | { | |
432 | int ret; | |
433 | ||
434 | sb_start_write(file->f_path.mnt->mnt_sb); | |
435 | ret = __mnt_want_write_file(file); | |
436 | if (ret) | |
437 | sb_end_write(file->f_path.mnt->mnt_sb); | |
438 | return ret; | |
439 | } | |
96029c4e | 440 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
441 | ||
8366025e | 442 | /** |
eb04c282 | 443 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
444 | * @mnt: the mount on which to give up write access |
445 | * | |
446 | * Tells the low-level filesystem that we are done | |
447 | * performing writes to it. Must be matched with | |
eb04c282 | 448 | * __mnt_want_write() call above. |
8366025e | 449 | */ |
eb04c282 | 450 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 451 | { |
d3ef3d73 | 452 | preempt_disable(); |
83adc753 | 453 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 454 | preempt_enable(); |
8366025e | 455 | } |
eb04c282 JK |
456 | |
457 | /** | |
458 | * mnt_drop_write - give up write access to a mount | |
459 | * @mnt: the mount on which to give up write access | |
460 | * | |
461 | * Tells the low-level filesystem that we are done performing writes to it and | |
462 | * also allows filesystem to be frozen again. Must be matched with | |
463 | * mnt_want_write() call above. | |
464 | */ | |
465 | void mnt_drop_write(struct vfsmount *mnt) | |
466 | { | |
467 | __mnt_drop_write(mnt); | |
468 | sb_end_write(mnt->mnt_sb); | |
469 | } | |
8366025e DH |
470 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
471 | ||
eb04c282 JK |
472 | void __mnt_drop_write_file(struct file *file) |
473 | { | |
474 | __mnt_drop_write(file->f_path.mnt); | |
475 | } | |
476 | ||
2a79f17e AV |
477 | void mnt_drop_write_file(struct file *file) |
478 | { | |
479 | mnt_drop_write(file->f_path.mnt); | |
480 | } | |
481 | EXPORT_SYMBOL(mnt_drop_write_file); | |
482 | ||
83adc753 | 483 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 484 | { |
3d733633 DH |
485 | int ret = 0; |
486 | ||
719ea2fb | 487 | lock_mount_hash(); |
83adc753 | 488 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 489 | /* |
d3ef3d73 | 490 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
491 | * should be visible before we do. | |
3d733633 | 492 | */ |
d3ef3d73 | 493 | smp_mb(); |
494 | ||
3d733633 | 495 | /* |
d3ef3d73 | 496 | * With writers on hold, if this value is zero, then there are |
497 | * definitely no active writers (although held writers may subsequently | |
498 | * increment the count, they'll have to wait, and decrement it after | |
499 | * seeing MNT_READONLY). | |
500 | * | |
501 | * It is OK to have counter incremented on one CPU and decremented on | |
502 | * another: the sum will add up correctly. The danger would be when we | |
503 | * sum up each counter, if we read a counter before it is incremented, | |
504 | * but then read another CPU's count which it has been subsequently | |
505 | * decremented from -- we would see more decrements than we should. | |
506 | * MNT_WRITE_HOLD protects against this scenario, because | |
507 | * mnt_want_write first increments count, then smp_mb, then spins on | |
508 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
509 | * we're counting up here. | |
3d733633 | 510 | */ |
c6653a83 | 511 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 512 | ret = -EBUSY; |
513 | else | |
83adc753 | 514 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 515 | /* |
516 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
517 | * that become unheld will see MNT_READONLY. | |
518 | */ | |
519 | smp_wmb(); | |
83adc753 | 520 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 521 | unlock_mount_hash(); |
3d733633 | 522 | return ret; |
8366025e | 523 | } |
8366025e | 524 | |
83adc753 | 525 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 526 | { |
719ea2fb | 527 | lock_mount_hash(); |
83adc753 | 528 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 529 | unlock_mount_hash(); |
2e4b7fcd DH |
530 | } |
531 | ||
4ed5e82f MS |
532 | int sb_prepare_remount_readonly(struct super_block *sb) |
533 | { | |
534 | struct mount *mnt; | |
535 | int err = 0; | |
536 | ||
8e8b8796 MS |
537 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
538 | if (atomic_long_read(&sb->s_remove_count)) | |
539 | return -EBUSY; | |
540 | ||
719ea2fb | 541 | lock_mount_hash(); |
4ed5e82f MS |
542 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
543 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
544 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
545 | smp_mb(); | |
546 | if (mnt_get_writers(mnt) > 0) { | |
547 | err = -EBUSY; | |
548 | break; | |
549 | } | |
550 | } | |
551 | } | |
8e8b8796 MS |
552 | if (!err && atomic_long_read(&sb->s_remove_count)) |
553 | err = -EBUSY; | |
554 | ||
4ed5e82f MS |
555 | if (!err) { |
556 | sb->s_readonly_remount = 1; | |
557 | smp_wmb(); | |
558 | } | |
559 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
560 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
561 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
562 | } | |
719ea2fb | 563 | unlock_mount_hash(); |
4ed5e82f MS |
564 | |
565 | return err; | |
566 | } | |
567 | ||
b105e270 | 568 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 569 | { |
52ba1621 | 570 | kfree(mnt->mnt_devname); |
73cd49ec | 571 | mnt_free_id(mnt); |
d3ef3d73 | 572 | #ifdef CONFIG_SMP |
68e8a9fe | 573 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 574 | #endif |
b105e270 | 575 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
576 | } |
577 | ||
48a066e7 AV |
578 | /* call under rcu_read_lock */ |
579 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
580 | { | |
581 | struct mount *mnt; | |
582 | if (read_seqretry(&mount_lock, seq)) | |
583 | return false; | |
584 | if (bastard == NULL) | |
585 | return true; | |
586 | mnt = real_mount(bastard); | |
587 | mnt_add_count(mnt, 1); | |
588 | if (likely(!read_seqretry(&mount_lock, seq))) | |
589 | return true; | |
590 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
591 | mnt_add_count(mnt, -1); | |
592 | return false; | |
593 | } | |
594 | rcu_read_unlock(); | |
595 | mntput(bastard); | |
596 | rcu_read_lock(); | |
597 | return false; | |
598 | } | |
599 | ||
1da177e4 | 600 | /* |
474279dc | 601 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 602 | * call under rcu_read_lock() |
1da177e4 | 603 | */ |
474279dc | 604 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 605 | { |
38129a13 | 606 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
607 | struct mount *p; |
608 | ||
38129a13 | 609 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
610 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
611 | return p; | |
612 | return NULL; | |
613 | } | |
614 | ||
615 | /* | |
616 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 617 | * mount_lock must be held. |
474279dc AV |
618 | */ |
619 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
620 | { | |
38129a13 AV |
621 | struct mount *p, *res; |
622 | res = p = __lookup_mnt(mnt, dentry); | |
623 | if (!p) | |
624 | goto out; | |
625 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
626 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
627 | break; | |
628 | res = p; | |
629 | } | |
38129a13 | 630 | out: |
1d6a32ac | 631 | return res; |
1da177e4 LT |
632 | } |
633 | ||
a05964f3 | 634 | /* |
f015f126 DH |
635 | * lookup_mnt - Return the first child mount mounted at path |
636 | * | |
637 | * "First" means first mounted chronologically. If you create the | |
638 | * following mounts: | |
639 | * | |
640 | * mount /dev/sda1 /mnt | |
641 | * mount /dev/sda2 /mnt | |
642 | * mount /dev/sda3 /mnt | |
643 | * | |
644 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
645 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
646 | * /dev/sda2, then /dev/sda3, then NULL. | |
647 | * | |
648 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 649 | */ |
1c755af4 | 650 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 651 | { |
c7105365 | 652 | struct mount *child_mnt; |
48a066e7 AV |
653 | struct vfsmount *m; |
654 | unsigned seq; | |
99b7db7b | 655 | |
48a066e7 AV |
656 | rcu_read_lock(); |
657 | do { | |
658 | seq = read_seqbegin(&mount_lock); | |
659 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
660 | m = child_mnt ? &child_mnt->mnt : NULL; | |
661 | } while (!legitimize_mnt(m, seq)); | |
662 | rcu_read_unlock(); | |
663 | return m; | |
a05964f3 RP |
664 | } |
665 | ||
84d17192 AV |
666 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
667 | { | |
0818bf27 | 668 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 | 669 | struct mountpoint *mp; |
eed81007 | 670 | int ret; |
84d17192 | 671 | |
0818bf27 | 672 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
673 | if (mp->m_dentry == dentry) { |
674 | /* might be worth a WARN_ON() */ | |
675 | if (d_unlinked(dentry)) | |
676 | return ERR_PTR(-ENOENT); | |
677 | mp->m_count++; | |
678 | return mp; | |
679 | } | |
680 | } | |
681 | ||
682 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
683 | if (!mp) | |
684 | return ERR_PTR(-ENOMEM); | |
685 | ||
eed81007 MS |
686 | ret = d_set_mounted(dentry); |
687 | if (ret) { | |
84d17192 | 688 | kfree(mp); |
eed81007 | 689 | return ERR_PTR(ret); |
84d17192 | 690 | } |
eed81007 | 691 | |
84d17192 AV |
692 | mp->m_dentry = dentry; |
693 | mp->m_count = 1; | |
0818bf27 | 694 | hlist_add_head(&mp->m_hash, chain); |
84d17192 AV |
695 | return mp; |
696 | } | |
697 | ||
698 | static void put_mountpoint(struct mountpoint *mp) | |
699 | { | |
700 | if (!--mp->m_count) { | |
701 | struct dentry *dentry = mp->m_dentry; | |
702 | spin_lock(&dentry->d_lock); | |
703 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
704 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 705 | hlist_del(&mp->m_hash); |
84d17192 AV |
706 | kfree(mp); |
707 | } | |
708 | } | |
709 | ||
143c8c91 | 710 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 711 | { |
6b3286ed | 712 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
713 | } |
714 | ||
99b7db7b NP |
715 | /* |
716 | * vfsmount lock must be held for write | |
717 | */ | |
6b3286ed | 718 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
719 | { |
720 | if (ns) { | |
721 | ns->event = ++event; | |
722 | wake_up_interruptible(&ns->poll); | |
723 | } | |
724 | } | |
725 | ||
99b7db7b NP |
726 | /* |
727 | * vfsmount lock must be held for write | |
728 | */ | |
6b3286ed | 729 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
730 | { |
731 | if (ns && ns->event != event) { | |
732 | ns->event = event; | |
733 | wake_up_interruptible(&ns->poll); | |
734 | } | |
735 | } | |
736 | ||
99b7db7b NP |
737 | /* |
738 | * vfsmount lock must be held for write | |
739 | */ | |
419148da AV |
740 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
741 | { | |
a73324da | 742 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
743 | old_path->mnt = &mnt->mnt_parent->mnt; |
744 | mnt->mnt_parent = mnt; | |
a73324da | 745 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 746 | list_del_init(&mnt->mnt_child); |
38129a13 | 747 | hlist_del_init_rcu(&mnt->mnt_hash); |
84d17192 AV |
748 | put_mountpoint(mnt->mnt_mp); |
749 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
750 | } |
751 | ||
99b7db7b NP |
752 | /* |
753 | * vfsmount lock must be held for write | |
754 | */ | |
84d17192 AV |
755 | void mnt_set_mountpoint(struct mount *mnt, |
756 | struct mountpoint *mp, | |
44d964d6 | 757 | struct mount *child_mnt) |
b90fa9ae | 758 | { |
84d17192 | 759 | mp->m_count++; |
3a2393d7 | 760 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 761 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 762 | child_mnt->mnt_parent = mnt; |
84d17192 | 763 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
764 | } |
765 | ||
99b7db7b NP |
766 | /* |
767 | * vfsmount lock must be held for write | |
768 | */ | |
84d17192 AV |
769 | static void attach_mnt(struct mount *mnt, |
770 | struct mount *parent, | |
771 | struct mountpoint *mp) | |
1da177e4 | 772 | { |
84d17192 | 773 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 774 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 775 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
776 | } |
777 | ||
778 | /* | |
99b7db7b | 779 | * vfsmount lock must be held for write |
b90fa9ae | 780 | */ |
1d6a32ac | 781 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 782 | { |
0714a533 | 783 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 784 | struct mount *m; |
b90fa9ae | 785 | LIST_HEAD(head); |
143c8c91 | 786 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 787 | |
0714a533 | 788 | BUG_ON(parent == mnt); |
b90fa9ae | 789 | |
1a4eeaf2 | 790 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 791 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 792 | m->mnt_ns = n; |
f03c6599 | 793 | |
b90fa9ae RP |
794 | list_splice(&head, n->list.prev); |
795 | ||
1d6a32ac | 796 | if (shadows) |
38129a13 | 797 | hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash); |
1d6a32ac | 798 | else |
38129a13 | 799 | hlist_add_head_rcu(&mnt->mnt_hash, |
0818bf27 | 800 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 801 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 802 | touch_mnt_namespace(n); |
1da177e4 LT |
803 | } |
804 | ||
909b0a88 | 805 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 806 | { |
6b41d536 AV |
807 | struct list_head *next = p->mnt_mounts.next; |
808 | if (next == &p->mnt_mounts) { | |
1da177e4 | 809 | while (1) { |
909b0a88 | 810 | if (p == root) |
1da177e4 | 811 | return NULL; |
6b41d536 AV |
812 | next = p->mnt_child.next; |
813 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 814 | break; |
0714a533 | 815 | p = p->mnt_parent; |
1da177e4 LT |
816 | } |
817 | } | |
6b41d536 | 818 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
819 | } |
820 | ||
315fc83e | 821 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 822 | { |
6b41d536 AV |
823 | struct list_head *prev = p->mnt_mounts.prev; |
824 | while (prev != &p->mnt_mounts) { | |
825 | p = list_entry(prev, struct mount, mnt_child); | |
826 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
827 | } |
828 | return p; | |
829 | } | |
830 | ||
9d412a43 AV |
831 | struct vfsmount * |
832 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
833 | { | |
b105e270 | 834 | struct mount *mnt; |
9d412a43 AV |
835 | struct dentry *root; |
836 | ||
837 | if (!type) | |
838 | return ERR_PTR(-ENODEV); | |
839 | ||
840 | mnt = alloc_vfsmnt(name); | |
841 | if (!mnt) | |
842 | return ERR_PTR(-ENOMEM); | |
843 | ||
844 | if (flags & MS_KERNMOUNT) | |
b105e270 | 845 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
846 | |
847 | root = mount_fs(type, flags, name, data); | |
848 | if (IS_ERR(root)) { | |
849 | free_vfsmnt(mnt); | |
850 | return ERR_CAST(root); | |
851 | } | |
852 | ||
b105e270 AV |
853 | mnt->mnt.mnt_root = root; |
854 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 855 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 856 | mnt->mnt_parent = mnt; |
719ea2fb | 857 | lock_mount_hash(); |
39f7c4db | 858 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 859 | unlock_mount_hash(); |
b105e270 | 860 | return &mnt->mnt; |
9d412a43 AV |
861 | } |
862 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
863 | ||
87129cc0 | 864 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 865 | int flag) |
1da177e4 | 866 | { |
87129cc0 | 867 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
868 | struct mount *mnt; |
869 | int err; | |
1da177e4 | 870 | |
be34d1a3 DH |
871 | mnt = alloc_vfsmnt(old->mnt_devname); |
872 | if (!mnt) | |
873 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 874 | |
7a472ef4 | 875 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
876 | mnt->mnt_group_id = 0; /* not a peer of original */ |
877 | else | |
878 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 879 | |
be34d1a3 DH |
880 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
881 | err = mnt_alloc_group_id(mnt); | |
882 | if (err) | |
883 | goto out_free; | |
1da177e4 | 884 | } |
be34d1a3 | 885 | |
f2ebb3a9 | 886 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 EB |
887 | /* Don't allow unprivileged users to change mount flags */ |
888 | if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) | |
889 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
890 | ||
5ff9d8a6 EB |
891 | /* Don't allow unprivileged users to reveal what is under a mount */ |
892 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
893 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
894 | ||
be34d1a3 DH |
895 | atomic_inc(&sb->s_active); |
896 | mnt->mnt.mnt_sb = sb; | |
897 | mnt->mnt.mnt_root = dget(root); | |
898 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
899 | mnt->mnt_parent = mnt; | |
719ea2fb | 900 | lock_mount_hash(); |
be34d1a3 | 901 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 902 | unlock_mount_hash(); |
be34d1a3 | 903 | |
7a472ef4 EB |
904 | if ((flag & CL_SLAVE) || |
905 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
906 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
907 | mnt->mnt_master = old; | |
908 | CLEAR_MNT_SHARED(mnt); | |
909 | } else if (!(flag & CL_PRIVATE)) { | |
910 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
911 | list_add(&mnt->mnt_share, &old->mnt_share); | |
912 | if (IS_MNT_SLAVE(old)) | |
913 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
914 | mnt->mnt_master = old->mnt_master; | |
915 | } | |
916 | if (flag & CL_MAKE_SHARED) | |
917 | set_mnt_shared(mnt); | |
918 | ||
919 | /* stick the duplicate mount on the same expiry list | |
920 | * as the original if that was on one */ | |
921 | if (flag & CL_EXPIRE) { | |
922 | if (!list_empty(&old->mnt_expire)) | |
923 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
924 | } | |
925 | ||
cb338d06 | 926 | return mnt; |
719f5d7f MS |
927 | |
928 | out_free: | |
929 | free_vfsmnt(mnt); | |
be34d1a3 | 930 | return ERR_PTR(err); |
1da177e4 LT |
931 | } |
932 | ||
48a066e7 AV |
933 | static void delayed_free(struct rcu_head *head) |
934 | { | |
935 | struct mount *mnt = container_of(head, struct mount, mnt_rcu); | |
936 | kfree(mnt->mnt_devname); | |
937 | #ifdef CONFIG_SMP | |
938 | free_percpu(mnt->mnt_pcp); | |
939 | #endif | |
940 | kmem_cache_free(mnt_cache, mnt); | |
941 | } | |
942 | ||
900148dc | 943 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 944 | { |
b3e19d92 | 945 | put_again: |
48a066e7 AV |
946 | rcu_read_lock(); |
947 | mnt_add_count(mnt, -1); | |
948 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
949 | rcu_read_unlock(); | |
f03c6599 | 950 | return; |
b3e19d92 | 951 | } |
719ea2fb | 952 | lock_mount_hash(); |
b3e19d92 | 953 | if (mnt_get_count(mnt)) { |
48a066e7 | 954 | rcu_read_unlock(); |
719ea2fb | 955 | unlock_mount_hash(); |
99b7db7b NP |
956 | return; |
957 | } | |
863d684f AV |
958 | if (unlikely(mnt->mnt_pinned)) { |
959 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
960 | mnt->mnt_pinned = 0; | |
48a066e7 | 961 | rcu_read_unlock(); |
719ea2fb | 962 | unlock_mount_hash(); |
900148dc | 963 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 964 | goto put_again; |
7b7b1ace | 965 | } |
48a066e7 AV |
966 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
967 | rcu_read_unlock(); | |
968 | unlock_mount_hash(); | |
969 | return; | |
970 | } | |
971 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
972 | rcu_read_unlock(); | |
962830df | 973 | |
39f7c4db | 974 | list_del(&mnt->mnt_instance); |
719ea2fb | 975 | unlock_mount_hash(); |
649a795a AV |
976 | |
977 | /* | |
978 | * This probably indicates that somebody messed | |
979 | * up a mnt_want/drop_write() pair. If this | |
980 | * happens, the filesystem was probably unable | |
981 | * to make r/w->r/o transitions. | |
982 | */ | |
983 | /* | |
984 | * The locking used to deal with mnt_count decrement provides barriers, | |
985 | * so mnt_get_writers() below is safe. | |
986 | */ | |
987 | WARN_ON(mnt_get_writers(mnt)); | |
988 | fsnotify_vfsmount_delete(&mnt->mnt); | |
989 | dput(mnt->mnt.mnt_root); | |
990 | deactivate_super(mnt->mnt.mnt_sb); | |
48a066e7 AV |
991 | mnt_free_id(mnt); |
992 | call_rcu(&mnt->mnt_rcu, delayed_free); | |
b3e19d92 | 993 | } |
b3e19d92 NP |
994 | |
995 | void mntput(struct vfsmount *mnt) | |
996 | { | |
997 | if (mnt) { | |
863d684f | 998 | struct mount *m = real_mount(mnt); |
b3e19d92 | 999 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1000 | if (unlikely(m->mnt_expiry_mark)) |
1001 | m->mnt_expiry_mark = 0; | |
1002 | mntput_no_expire(m); | |
b3e19d92 NP |
1003 | } |
1004 | } | |
1005 | EXPORT_SYMBOL(mntput); | |
1006 | ||
1007 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1008 | { | |
1009 | if (mnt) | |
83adc753 | 1010 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1011 | return mnt; |
1012 | } | |
1013 | EXPORT_SYMBOL(mntget); | |
1014 | ||
7b7b1ace AV |
1015 | void mnt_pin(struct vfsmount *mnt) |
1016 | { | |
719ea2fb | 1017 | lock_mount_hash(); |
863d684f | 1018 | real_mount(mnt)->mnt_pinned++; |
719ea2fb | 1019 | unlock_mount_hash(); |
7b7b1ace | 1020 | } |
7b7b1ace AV |
1021 | EXPORT_SYMBOL(mnt_pin); |
1022 | ||
863d684f | 1023 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 1024 | { |
863d684f | 1025 | struct mount *mnt = real_mount(m); |
719ea2fb | 1026 | lock_mount_hash(); |
7b7b1ace | 1027 | if (mnt->mnt_pinned) { |
863d684f | 1028 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
1029 | mnt->mnt_pinned--; |
1030 | } | |
719ea2fb | 1031 | unlock_mount_hash(); |
7b7b1ace | 1032 | } |
7b7b1ace | 1033 | EXPORT_SYMBOL(mnt_unpin); |
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 | ||
97216be0 AV |
1220 | up_write(&namespace_sem); |
1221 | ||
48a066e7 AV |
1222 | synchronize_rcu(); |
1223 | ||
38129a13 AV |
1224 | while (!hlist_empty(&head)) { |
1225 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1226 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1227 | if (mnt->mnt_ex_mountpoint.mnt) |
1228 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1229 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1230 | } |
1231 | } | |
1232 | ||
97216be0 | 1233 | static inline void namespace_lock(void) |
e3197d83 | 1234 | { |
97216be0 | 1235 | down_write(&namespace_sem); |
e3197d83 AV |
1236 | } |
1237 | ||
99b7db7b | 1238 | /* |
48a066e7 | 1239 | * mount_lock must be held |
99b7db7b | 1240 | * namespace_sem must be held for write |
48a066e7 AV |
1241 | * how = 0 => just this tree, don't propagate |
1242 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1243 | * how = 2 => lazy umount | |
99b7db7b | 1244 | */ |
48a066e7 | 1245 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1246 | { |
38129a13 | 1247 | HLIST_HEAD(tmp_list); |
315fc83e | 1248 | struct mount *p; |
38129a13 | 1249 | struct mount *last = NULL; |
1da177e4 | 1250 | |
38129a13 AV |
1251 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1252 | hlist_del_init_rcu(&p->mnt_hash); | |
1253 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1254 | } | |
1da177e4 | 1255 | |
48a066e7 | 1256 | if (how) |
7b8a53fd | 1257 | propagate_umount(&tmp_list); |
a05964f3 | 1258 | |
38129a13 | 1259 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1260 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1261 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1262 | __touch_mnt_namespace(p->mnt_ns); |
1263 | p->mnt_ns = NULL; | |
48a066e7 AV |
1264 | if (how < 2) |
1265 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
6b41d536 | 1266 | list_del_init(&p->mnt_child); |
676da58d | 1267 | if (mnt_has_parent(p)) { |
84d17192 | 1268 | put_mountpoint(p->mnt_mp); |
aba809cf AV |
1269 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1270 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1271 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1272 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1273 | p->mnt_parent = p; | |
84d17192 | 1274 | p->mnt_mp = NULL; |
7c4b93d8 | 1275 | } |
0f0afb1d | 1276 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1277 | last = p; |
1278 | } | |
1279 | if (last) { | |
1280 | last->mnt_hash.next = unmounted.first; | |
1281 | unmounted.first = tmp_list.first; | |
1282 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1283 | } |
1284 | } | |
1285 | ||
b54b9be7 | 1286 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1287 | |
1ab59738 | 1288 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1289 | { |
1ab59738 | 1290 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1291 | int retval; |
1292 | ||
1ab59738 | 1293 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1294 | if (retval) |
1295 | return retval; | |
1296 | ||
1297 | /* | |
1298 | * Allow userspace to request a mountpoint be expired rather than | |
1299 | * unmounting unconditionally. Unmount only happens if: | |
1300 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1301 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1302 | */ | |
1303 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1304 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1305 | flags & (MNT_FORCE | MNT_DETACH)) |
1306 | return -EINVAL; | |
1307 | ||
b3e19d92 NP |
1308 | /* |
1309 | * probably don't strictly need the lock here if we examined | |
1310 | * all race cases, but it's a slowpath. | |
1311 | */ | |
719ea2fb | 1312 | lock_mount_hash(); |
83adc753 | 1313 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1314 | unlock_mount_hash(); |
1da177e4 | 1315 | return -EBUSY; |
b3e19d92 | 1316 | } |
719ea2fb | 1317 | unlock_mount_hash(); |
1da177e4 | 1318 | |
863d684f | 1319 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1320 | return -EAGAIN; |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * If we may have to abort operations to get out of this | |
1325 | * mount, and they will themselves hold resources we must | |
1326 | * allow the fs to do things. In the Unix tradition of | |
1327 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1328 | * might fail to complete on the first run through as other tasks | |
1329 | * must return, and the like. Thats for the mount program to worry | |
1330 | * about for the moment. | |
1331 | */ | |
1332 | ||
42faad99 | 1333 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1334 | sb->s_op->umount_begin(sb); |
42faad99 | 1335 | } |
1da177e4 LT |
1336 | |
1337 | /* | |
1338 | * No sense to grab the lock for this test, but test itself looks | |
1339 | * somewhat bogus. Suggestions for better replacement? | |
1340 | * Ho-hum... In principle, we might treat that as umount + switch | |
1341 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1342 | * Actually it makes sense, especially if rootfs would contain a | |
1343 | * /reboot - static binary that would close all descriptors and | |
1344 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1345 | */ | |
1ab59738 | 1346 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1347 | /* |
1348 | * Special case for "unmounting" root ... | |
1349 | * we just try to remount it readonly. | |
1350 | */ | |
1351 | down_write(&sb->s_umount); | |
4aa98cf7 | 1352 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1353 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1354 | up_write(&sb->s_umount); |
1355 | return retval; | |
1356 | } | |
1357 | ||
97216be0 | 1358 | namespace_lock(); |
719ea2fb | 1359 | lock_mount_hash(); |
5addc5dd | 1360 | event++; |
1da177e4 | 1361 | |
48a066e7 | 1362 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1363 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1364 | umount_tree(mnt, 2); |
1da177e4 | 1365 | retval = 0; |
48a066e7 AV |
1366 | } else { |
1367 | shrink_submounts(mnt); | |
1368 | retval = -EBUSY; | |
1369 | if (!propagate_mount_busy(mnt, 2)) { | |
1370 | if (!list_empty(&mnt->mnt_list)) | |
1371 | umount_tree(mnt, 1); | |
1372 | retval = 0; | |
1373 | } | |
1da177e4 | 1374 | } |
719ea2fb | 1375 | unlock_mount_hash(); |
e3197d83 | 1376 | namespace_unlock(); |
1da177e4 LT |
1377 | return retval; |
1378 | } | |
1379 | ||
9b40bc90 AV |
1380 | /* |
1381 | * Is the caller allowed to modify his namespace? | |
1382 | */ | |
1383 | static inline bool may_mount(void) | |
1384 | { | |
1385 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1386 | } | |
1387 | ||
1da177e4 LT |
1388 | /* |
1389 | * Now umount can handle mount points as well as block devices. | |
1390 | * This is important for filesystems which use unnamed block devices. | |
1391 | * | |
1392 | * We now support a flag for forced unmount like the other 'big iron' | |
1393 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1394 | */ | |
1395 | ||
bdc480e3 | 1396 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1397 | { |
2d8f3038 | 1398 | struct path path; |
900148dc | 1399 | struct mount *mnt; |
1da177e4 | 1400 | int retval; |
db1f05bb | 1401 | int lookup_flags = 0; |
1da177e4 | 1402 | |
db1f05bb MS |
1403 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1404 | return -EINVAL; | |
1405 | ||
9b40bc90 AV |
1406 | if (!may_mount()) |
1407 | return -EPERM; | |
1408 | ||
db1f05bb MS |
1409 | if (!(flags & UMOUNT_NOFOLLOW)) |
1410 | lookup_flags |= LOOKUP_FOLLOW; | |
1411 | ||
197df04c | 1412 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1413 | if (retval) |
1414 | goto out; | |
900148dc | 1415 | mnt = real_mount(path.mnt); |
1da177e4 | 1416 | retval = -EINVAL; |
2d8f3038 | 1417 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1418 | goto dput_and_out; |
143c8c91 | 1419 | if (!check_mnt(mnt)) |
1da177e4 | 1420 | goto dput_and_out; |
5ff9d8a6 EB |
1421 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1422 | goto dput_and_out; | |
1da177e4 | 1423 | |
900148dc | 1424 | retval = do_umount(mnt, flags); |
1da177e4 | 1425 | dput_and_out: |
429731b1 | 1426 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1427 | dput(path.dentry); |
900148dc | 1428 | mntput_no_expire(mnt); |
1da177e4 LT |
1429 | out: |
1430 | return retval; | |
1431 | } | |
1432 | ||
1433 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1434 | ||
1435 | /* | |
b58fed8b | 1436 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1437 | */ |
bdc480e3 | 1438 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1439 | { |
b58fed8b | 1440 | return sys_umount(name, 0); |
1da177e4 LT |
1441 | } |
1442 | ||
1443 | #endif | |
1444 | ||
4ce5d2b1 | 1445 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1446 | { |
4ce5d2b1 EB |
1447 | /* Is this a proxy for a mount namespace? */ |
1448 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1449 | struct proc_ns *ei; |
8823c079 EB |
1450 | |
1451 | if (!proc_ns_inode(inode)) | |
1452 | return false; | |
1453 | ||
0bb80f24 | 1454 | ei = get_proc_ns(inode); |
8823c079 EB |
1455 | if (ei->ns_ops != &mntns_operations) |
1456 | return false; | |
1457 | ||
4ce5d2b1 EB |
1458 | return true; |
1459 | } | |
1460 | ||
1461 | static bool mnt_ns_loop(struct dentry *dentry) | |
1462 | { | |
1463 | /* Could bind mounting the mount namespace inode cause a | |
1464 | * mount namespace loop? | |
1465 | */ | |
1466 | struct mnt_namespace *mnt_ns; | |
1467 | if (!is_mnt_ns_file(dentry)) | |
1468 | return false; | |
1469 | ||
1470 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1471 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1472 | } | |
1473 | ||
87129cc0 | 1474 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1475 | int flag) |
1da177e4 | 1476 | { |
84d17192 | 1477 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1478 | |
4ce5d2b1 EB |
1479 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1480 | return ERR_PTR(-EINVAL); | |
1481 | ||
1482 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1483 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1484 | |
36341f64 | 1485 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1486 | if (IS_ERR(q)) |
1487 | return q; | |
1488 | ||
5ff9d8a6 | 1489 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1490 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1491 | |
1492 | p = mnt; | |
6b41d536 | 1493 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1494 | struct mount *s; |
7ec02ef1 | 1495 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1496 | continue; |
1497 | ||
909b0a88 | 1498 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1499 | if (!(flag & CL_COPY_UNBINDABLE) && |
1500 | IS_MNT_UNBINDABLE(s)) { | |
1501 | s = skip_mnt_tree(s); | |
1502 | continue; | |
1503 | } | |
1504 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1505 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1506 | s = skip_mnt_tree(s); |
1507 | continue; | |
1508 | } | |
0714a533 AV |
1509 | while (p != s->mnt_parent) { |
1510 | p = p->mnt_parent; | |
1511 | q = q->mnt_parent; | |
1da177e4 | 1512 | } |
87129cc0 | 1513 | p = s; |
84d17192 | 1514 | parent = q; |
87129cc0 | 1515 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1516 | if (IS_ERR(q)) |
1517 | goto out; | |
719ea2fb | 1518 | lock_mount_hash(); |
1a4eeaf2 | 1519 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1520 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1521 | unlock_mount_hash(); |
1da177e4 LT |
1522 | } |
1523 | } | |
1524 | return res; | |
be34d1a3 | 1525 | out: |
1da177e4 | 1526 | if (res) { |
719ea2fb | 1527 | lock_mount_hash(); |
328e6d90 | 1528 | umount_tree(res, 0); |
719ea2fb | 1529 | unlock_mount_hash(); |
1da177e4 | 1530 | } |
be34d1a3 | 1531 | return q; |
1da177e4 LT |
1532 | } |
1533 | ||
be34d1a3 DH |
1534 | /* Caller should check returned pointer for errors */ |
1535 | ||
589ff870 | 1536 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1537 | { |
cb338d06 | 1538 | struct mount *tree; |
97216be0 | 1539 | namespace_lock(); |
87129cc0 AV |
1540 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1541 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1542 | namespace_unlock(); |
be34d1a3 | 1543 | if (IS_ERR(tree)) |
52e220d3 | 1544 | return ERR_CAST(tree); |
be34d1a3 | 1545 | return &tree->mnt; |
8aec0809 AV |
1546 | } |
1547 | ||
1548 | void drop_collected_mounts(struct vfsmount *mnt) | |
1549 | { | |
97216be0 | 1550 | namespace_lock(); |
719ea2fb | 1551 | lock_mount_hash(); |
328e6d90 | 1552 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1553 | unlock_mount_hash(); |
3ab6abee | 1554 | namespace_unlock(); |
8aec0809 AV |
1555 | } |
1556 | ||
1f707137 AV |
1557 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1558 | struct vfsmount *root) | |
1559 | { | |
1a4eeaf2 | 1560 | struct mount *mnt; |
1f707137 AV |
1561 | int res = f(root, arg); |
1562 | if (res) | |
1563 | return res; | |
1a4eeaf2 AV |
1564 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1565 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1566 | if (res) |
1567 | return res; | |
1568 | } | |
1569 | return 0; | |
1570 | } | |
1571 | ||
4b8b21f4 | 1572 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1573 | { |
315fc83e | 1574 | struct mount *p; |
719f5d7f | 1575 | |
909b0a88 | 1576 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1577 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1578 | mnt_release_group_id(p); |
719f5d7f MS |
1579 | } |
1580 | } | |
1581 | ||
4b8b21f4 | 1582 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1583 | { |
315fc83e | 1584 | struct mount *p; |
719f5d7f | 1585 | |
909b0a88 | 1586 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1587 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1588 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1589 | if (err) { |
4b8b21f4 | 1590 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1591 | return err; |
1592 | } | |
1593 | } | |
1594 | } | |
1595 | ||
1596 | return 0; | |
1597 | } | |
1598 | ||
b90fa9ae RP |
1599 | /* |
1600 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1601 | * @nd : place the mount tree @source_mnt is attached |
1602 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1603 | * store the parent mount and mountpoint dentry. | |
1604 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1605 | * |
1606 | * NOTE: in the table below explains the semantics when a source mount | |
1607 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1608 | * --------------------------------------------------------------------------- |
1609 | * | BIND MOUNT OPERATION | | |
1610 | * |************************************************************************** | |
1611 | * | source-->| shared | private | slave | unbindable | | |
1612 | * | dest | | | | | | |
1613 | * | | | | | | | | |
1614 | * | v | | | | | | |
1615 | * |************************************************************************** | |
1616 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1617 | * | | | | | | | |
1618 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1619 | * *************************************************************************** | |
b90fa9ae RP |
1620 | * A bind operation clones the source mount and mounts the clone on the |
1621 | * destination mount. | |
1622 | * | |
1623 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1624 | * tree of the destination mount and the cloned mount is added to | |
1625 | * the peer group of the source mount. | |
1626 | * (+) the cloned mount is created under the destination mount and is marked | |
1627 | * as shared. The cloned mount is added to the peer group of the source | |
1628 | * mount. | |
5afe0022 RP |
1629 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1630 | * of the destination mount and the cloned mount is made slave | |
1631 | * of the same master as that of the source mount. The cloned mount | |
1632 | * is marked as 'shared and slave'. | |
1633 | * (*) the cloned mount is made a slave of the same master as that of the | |
1634 | * source mount. | |
1635 | * | |
9676f0c6 RP |
1636 | * --------------------------------------------------------------------------- |
1637 | * | MOVE MOUNT OPERATION | | |
1638 | * |************************************************************************** | |
1639 | * | source-->| shared | private | slave | unbindable | | |
1640 | * | dest | | | | | | |
1641 | * | | | | | | | | |
1642 | * | v | | | | | | |
1643 | * |************************************************************************** | |
1644 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1645 | * | | | | | | | |
1646 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1647 | * *************************************************************************** | |
5afe0022 RP |
1648 | * |
1649 | * (+) the mount is moved to the destination. And is then propagated to | |
1650 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1651 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1652 | * (+++) the mount is moved to the destination and is then propagated to |
1653 | * all the mounts belonging to the destination mount's propagation tree. | |
1654 | * the mount is marked as 'shared and slave'. | |
1655 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1656 | * |
1657 | * if the source mount is a tree, the operations explained above is | |
1658 | * applied to each mount in the tree. | |
1659 | * Must be called without spinlocks held, since this function can sleep | |
1660 | * in allocations. | |
1661 | */ | |
0fb54e50 | 1662 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1663 | struct mount *dest_mnt, |
1664 | struct mountpoint *dest_mp, | |
1665 | struct path *parent_path) | |
b90fa9ae | 1666 | { |
38129a13 | 1667 | HLIST_HEAD(tree_list); |
315fc83e | 1668 | struct mount *child, *p; |
38129a13 | 1669 | struct hlist_node *n; |
719f5d7f | 1670 | int err; |
b90fa9ae | 1671 | |
fc7be130 | 1672 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1673 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1674 | if (err) |
1675 | goto out; | |
0b1b901b | 1676 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1677 | lock_mount_hash(); |
0b1b901b AV |
1678 | if (err) |
1679 | goto out_cleanup_ids; | |
909b0a88 | 1680 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1681 | set_mnt_shared(p); |
0b1b901b AV |
1682 | } else { |
1683 | lock_mount_hash(); | |
b90fa9ae | 1684 | } |
1a390689 | 1685 | if (parent_path) { |
0fb54e50 | 1686 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1687 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1688 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1689 | } else { |
84d17192 | 1690 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1691 | commit_tree(source_mnt, NULL); |
21444403 | 1692 | } |
b90fa9ae | 1693 | |
38129a13 | 1694 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1695 | struct mount *q; |
38129a13 | 1696 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1697 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1698 | child->mnt_mountpoint); | |
1699 | commit_tree(child, q); | |
b90fa9ae | 1700 | } |
719ea2fb | 1701 | unlock_mount_hash(); |
99b7db7b | 1702 | |
b90fa9ae | 1703 | return 0; |
719f5d7f MS |
1704 | |
1705 | out_cleanup_ids: | |
f2ebb3a9 AV |
1706 | while (!hlist_empty(&tree_list)) { |
1707 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1708 | umount_tree(child, 0); | |
1709 | } | |
1710 | unlock_mount_hash(); | |
0b1b901b | 1711 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1712 | out: |
1713 | return err; | |
b90fa9ae RP |
1714 | } |
1715 | ||
84d17192 | 1716 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1717 | { |
1718 | struct vfsmount *mnt; | |
84d17192 | 1719 | struct dentry *dentry = path->dentry; |
b12cea91 | 1720 | retry: |
84d17192 AV |
1721 | mutex_lock(&dentry->d_inode->i_mutex); |
1722 | if (unlikely(cant_mount(dentry))) { | |
1723 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1724 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1725 | } |
97216be0 | 1726 | namespace_lock(); |
b12cea91 | 1727 | mnt = lookup_mnt(path); |
84d17192 AV |
1728 | if (likely(!mnt)) { |
1729 | struct mountpoint *mp = new_mountpoint(dentry); | |
1730 | if (IS_ERR(mp)) { | |
97216be0 | 1731 | namespace_unlock(); |
84d17192 AV |
1732 | mutex_unlock(&dentry->d_inode->i_mutex); |
1733 | return mp; | |
1734 | } | |
1735 | return mp; | |
1736 | } | |
97216be0 | 1737 | namespace_unlock(); |
b12cea91 AV |
1738 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1739 | path_put(path); | |
1740 | path->mnt = mnt; | |
84d17192 | 1741 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1742 | goto retry; |
1743 | } | |
1744 | ||
84d17192 | 1745 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1746 | { |
84d17192 AV |
1747 | struct dentry *dentry = where->m_dentry; |
1748 | put_mountpoint(where); | |
328e6d90 | 1749 | namespace_unlock(); |
84d17192 | 1750 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1751 | } |
1752 | ||
84d17192 | 1753 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1754 | { |
95bc5f25 | 1755 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1756 | return -EINVAL; |
1757 | ||
84d17192 | 1758 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1759 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1760 | return -ENOTDIR; |
1761 | ||
84d17192 | 1762 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1763 | } |
1764 | ||
7a2e8a8f VA |
1765 | /* |
1766 | * Sanity check the flags to change_mnt_propagation. | |
1767 | */ | |
1768 | ||
1769 | static int flags_to_propagation_type(int flags) | |
1770 | { | |
7c6e984d | 1771 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1772 | |
1773 | /* Fail if any non-propagation flags are set */ | |
1774 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1775 | return 0; | |
1776 | /* Only one propagation flag should be set */ | |
1777 | if (!is_power_of_2(type)) | |
1778 | return 0; | |
1779 | return type; | |
1780 | } | |
1781 | ||
07b20889 RP |
1782 | /* |
1783 | * recursively change the type of the mountpoint. | |
1784 | */ | |
0a0d8a46 | 1785 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1786 | { |
315fc83e | 1787 | struct mount *m; |
4b8b21f4 | 1788 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1789 | int recurse = flag & MS_REC; |
7a2e8a8f | 1790 | int type; |
719f5d7f | 1791 | int err = 0; |
07b20889 | 1792 | |
2d92ab3c | 1793 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1794 | return -EINVAL; |
1795 | ||
7a2e8a8f VA |
1796 | type = flags_to_propagation_type(flag); |
1797 | if (!type) | |
1798 | return -EINVAL; | |
1799 | ||
97216be0 | 1800 | namespace_lock(); |
719f5d7f MS |
1801 | if (type == MS_SHARED) { |
1802 | err = invent_group_ids(mnt, recurse); | |
1803 | if (err) | |
1804 | goto out_unlock; | |
1805 | } | |
1806 | ||
719ea2fb | 1807 | lock_mount_hash(); |
909b0a88 | 1808 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1809 | change_mnt_propagation(m, type); |
719ea2fb | 1810 | unlock_mount_hash(); |
719f5d7f MS |
1811 | |
1812 | out_unlock: | |
97216be0 | 1813 | namespace_unlock(); |
719f5d7f | 1814 | return err; |
07b20889 RP |
1815 | } |
1816 | ||
5ff9d8a6 EB |
1817 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1818 | { | |
1819 | struct mount *child; | |
1820 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1821 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1822 | continue; | |
1823 | ||
1824 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1825 | return true; | |
1826 | } | |
1827 | return false; | |
1828 | } | |
1829 | ||
1da177e4 LT |
1830 | /* |
1831 | * do loopback mount. | |
1832 | */ | |
808d4e3c | 1833 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1834 | int recurse) |
1da177e4 | 1835 | { |
2d92ab3c | 1836 | struct path old_path; |
84d17192 AV |
1837 | struct mount *mnt = NULL, *old, *parent; |
1838 | struct mountpoint *mp; | |
57eccb83 | 1839 | int err; |
1da177e4 LT |
1840 | if (!old_name || !*old_name) |
1841 | return -EINVAL; | |
815d405c | 1842 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1843 | if (err) |
1844 | return err; | |
1845 | ||
8823c079 | 1846 | err = -EINVAL; |
4ce5d2b1 | 1847 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1848 | goto out; |
1849 | ||
84d17192 AV |
1850 | mp = lock_mount(path); |
1851 | err = PTR_ERR(mp); | |
1852 | if (IS_ERR(mp)) | |
b12cea91 AV |
1853 | goto out; |
1854 | ||
87129cc0 | 1855 | old = real_mount(old_path.mnt); |
84d17192 | 1856 | parent = real_mount(path->mnt); |
87129cc0 | 1857 | |
1da177e4 | 1858 | err = -EINVAL; |
fc7be130 | 1859 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1860 | goto out2; |
9676f0c6 | 1861 | |
84d17192 | 1862 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1863 | goto out2; |
1da177e4 | 1864 | |
5ff9d8a6 EB |
1865 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1866 | goto out2; | |
1867 | ||
ccd48bc7 | 1868 | if (recurse) |
4ce5d2b1 | 1869 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1870 | else |
87129cc0 | 1871 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1872 | |
be34d1a3 DH |
1873 | if (IS_ERR(mnt)) { |
1874 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1875 | goto out2; |
be34d1a3 | 1876 | } |
ccd48bc7 | 1877 | |
5ff9d8a6 EB |
1878 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1879 | ||
84d17192 | 1880 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1881 | if (err) { |
719ea2fb | 1882 | lock_mount_hash(); |
328e6d90 | 1883 | umount_tree(mnt, 0); |
719ea2fb | 1884 | unlock_mount_hash(); |
5b83d2c5 | 1885 | } |
b12cea91 | 1886 | out2: |
84d17192 | 1887 | unlock_mount(mp); |
ccd48bc7 | 1888 | out: |
2d92ab3c | 1889 | path_put(&old_path); |
1da177e4 LT |
1890 | return err; |
1891 | } | |
1892 | ||
2e4b7fcd DH |
1893 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1894 | { | |
1895 | int error = 0; | |
1896 | int readonly_request = 0; | |
1897 | ||
1898 | if (ms_flags & MS_RDONLY) | |
1899 | readonly_request = 1; | |
1900 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1901 | return 0; | |
1902 | ||
90563b19 EB |
1903 | if (mnt->mnt_flags & MNT_LOCK_READONLY) |
1904 | return -EPERM; | |
1905 | ||
2e4b7fcd | 1906 | if (readonly_request) |
83adc753 | 1907 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1908 | else |
83adc753 | 1909 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1910 | return error; |
1911 | } | |
1912 | ||
1da177e4 LT |
1913 | /* |
1914 | * change filesystem flags. dir should be a physical root of filesystem. | |
1915 | * If you've mounted a non-root directory somewhere and want to do remount | |
1916 | * on it - tough luck. | |
1917 | */ | |
0a0d8a46 | 1918 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1919 | void *data) |
1920 | { | |
1921 | int err; | |
2d92ab3c | 1922 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1923 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1924 | |
143c8c91 | 1925 | if (!check_mnt(mnt)) |
1da177e4 LT |
1926 | return -EINVAL; |
1927 | ||
2d92ab3c | 1928 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1929 | return -EINVAL; |
1930 | ||
ff36fe2c EP |
1931 | err = security_sb_remount(sb, data); |
1932 | if (err) | |
1933 | return err; | |
1934 | ||
1da177e4 | 1935 | down_write(&sb->s_umount); |
2e4b7fcd | 1936 | if (flags & MS_BIND) |
2d92ab3c | 1937 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1938 | else if (!capable(CAP_SYS_ADMIN)) |
1939 | err = -EPERM; | |
4aa98cf7 | 1940 | else |
2e4b7fcd | 1941 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1942 | if (!err) { |
719ea2fb | 1943 | lock_mount_hash(); |
143c8c91 AV |
1944 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1945 | mnt->mnt.mnt_flags = mnt_flags; | |
143c8c91 | 1946 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 1947 | unlock_mount_hash(); |
0e55a7cc | 1948 | } |
6339dab8 | 1949 | up_write(&sb->s_umount); |
1da177e4 LT |
1950 | return err; |
1951 | } | |
1952 | ||
cbbe362c | 1953 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1954 | { |
315fc83e | 1955 | struct mount *p; |
909b0a88 | 1956 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1957 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1958 | return 1; |
1959 | } | |
1960 | return 0; | |
1961 | } | |
1962 | ||
808d4e3c | 1963 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1964 | { |
2d92ab3c | 1965 | struct path old_path, parent_path; |
676da58d | 1966 | struct mount *p; |
0fb54e50 | 1967 | struct mount *old; |
84d17192 | 1968 | struct mountpoint *mp; |
57eccb83 | 1969 | int err; |
1da177e4 LT |
1970 | if (!old_name || !*old_name) |
1971 | return -EINVAL; | |
2d92ab3c | 1972 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1973 | if (err) |
1974 | return err; | |
1975 | ||
84d17192 AV |
1976 | mp = lock_mount(path); |
1977 | err = PTR_ERR(mp); | |
1978 | if (IS_ERR(mp)) | |
cc53ce53 DH |
1979 | goto out; |
1980 | ||
143c8c91 | 1981 | old = real_mount(old_path.mnt); |
fc7be130 | 1982 | p = real_mount(path->mnt); |
143c8c91 | 1983 | |
1da177e4 | 1984 | err = -EINVAL; |
fc7be130 | 1985 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1986 | goto out1; |
1987 | ||
5ff9d8a6 EB |
1988 | if (old->mnt.mnt_flags & MNT_LOCKED) |
1989 | goto out1; | |
1990 | ||
1da177e4 | 1991 | err = -EINVAL; |
2d92ab3c | 1992 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1993 | goto out1; |
1da177e4 | 1994 | |
676da58d | 1995 | if (!mnt_has_parent(old)) |
21444403 | 1996 | goto out1; |
1da177e4 | 1997 | |
2d92ab3c AV |
1998 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1999 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
2000 | goto out1; |
2001 | /* | |
2002 | * Don't move a mount residing in a shared parent. | |
2003 | */ | |
fc7be130 | 2004 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2005 | goto out1; |
9676f0c6 RP |
2006 | /* |
2007 | * Don't move a mount tree containing unbindable mounts to a destination | |
2008 | * mount which is shared. | |
2009 | */ | |
fc7be130 | 2010 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2011 | goto out1; |
1da177e4 | 2012 | err = -ELOOP; |
fc7be130 | 2013 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2014 | if (p == old) |
21444403 | 2015 | goto out1; |
1da177e4 | 2016 | |
84d17192 | 2017 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2018 | if (err) |
21444403 | 2019 | goto out1; |
1da177e4 LT |
2020 | |
2021 | /* if the mount is moved, it should no longer be expire | |
2022 | * automatically */ | |
6776db3d | 2023 | list_del_init(&old->mnt_expire); |
1da177e4 | 2024 | out1: |
84d17192 | 2025 | unlock_mount(mp); |
1da177e4 | 2026 | out: |
1da177e4 | 2027 | if (!err) |
1a390689 | 2028 | path_put(&parent_path); |
2d92ab3c | 2029 | path_put(&old_path); |
1da177e4 LT |
2030 | return err; |
2031 | } | |
2032 | ||
9d412a43 AV |
2033 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2034 | { | |
2035 | int err; | |
2036 | const char *subtype = strchr(fstype, '.'); | |
2037 | if (subtype) { | |
2038 | subtype++; | |
2039 | err = -EINVAL; | |
2040 | if (!subtype[0]) | |
2041 | goto err; | |
2042 | } else | |
2043 | subtype = ""; | |
2044 | ||
2045 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2046 | err = -ENOMEM; | |
2047 | if (!mnt->mnt_sb->s_subtype) | |
2048 | goto err; | |
2049 | return mnt; | |
2050 | ||
2051 | err: | |
2052 | mntput(mnt); | |
2053 | return ERR_PTR(err); | |
2054 | } | |
2055 | ||
9d412a43 AV |
2056 | /* |
2057 | * add a mount into a namespace's mount tree | |
2058 | */ | |
95bc5f25 | 2059 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2060 | { |
84d17192 AV |
2061 | struct mountpoint *mp; |
2062 | struct mount *parent; | |
9d412a43 AV |
2063 | int err; |
2064 | ||
f2ebb3a9 | 2065 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2066 | |
84d17192 AV |
2067 | mp = lock_mount(path); |
2068 | if (IS_ERR(mp)) | |
2069 | return PTR_ERR(mp); | |
9d412a43 | 2070 | |
84d17192 | 2071 | parent = real_mount(path->mnt); |
9d412a43 | 2072 | err = -EINVAL; |
84d17192 | 2073 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2074 | /* that's acceptable only for automounts done in private ns */ |
2075 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2076 | goto unlock; | |
2077 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2078 | if (!parent->mnt_ns) |
156cacb1 AV |
2079 | goto unlock; |
2080 | } | |
9d412a43 AV |
2081 | |
2082 | /* Refuse the same filesystem on the same mount point */ | |
2083 | err = -EBUSY; | |
95bc5f25 | 2084 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2085 | path->mnt->mnt_root == path->dentry) |
2086 | goto unlock; | |
2087 | ||
2088 | err = -EINVAL; | |
95bc5f25 | 2089 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2090 | goto unlock; |
2091 | ||
95bc5f25 | 2092 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2093 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2094 | |
2095 | unlock: | |
84d17192 | 2096 | unlock_mount(mp); |
9d412a43 AV |
2097 | return err; |
2098 | } | |
b1e75df4 | 2099 | |
1da177e4 LT |
2100 | /* |
2101 | * create a new mount for userspace and request it to be added into the | |
2102 | * namespace's tree | |
2103 | */ | |
0c55cfc4 | 2104 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2105 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2106 | { |
0c55cfc4 | 2107 | struct file_system_type *type; |
9b40bc90 | 2108 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2109 | struct vfsmount *mnt; |
15f9a3f3 | 2110 | int err; |
1da177e4 | 2111 | |
0c55cfc4 | 2112 | if (!fstype) |
1da177e4 LT |
2113 | return -EINVAL; |
2114 | ||
0c55cfc4 EB |
2115 | type = get_fs_type(fstype); |
2116 | if (!type) | |
2117 | return -ENODEV; | |
2118 | ||
2119 | if (user_ns != &init_user_ns) { | |
2120 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2121 | put_filesystem(type); | |
2122 | return -EPERM; | |
2123 | } | |
2124 | /* Only in special cases allow devices from mounts | |
2125 | * created outside the initial user namespace. | |
2126 | */ | |
2127 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2128 | flags |= MS_NODEV; | |
2129 | mnt_flags |= MNT_NODEV; | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | mnt = vfs_kern_mount(type, flags, name, data); | |
2134 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2135 | !mnt->mnt_sb->s_subtype) | |
2136 | mnt = fs_set_subtype(mnt, fstype); | |
2137 | ||
2138 | put_filesystem(type); | |
1da177e4 LT |
2139 | if (IS_ERR(mnt)) |
2140 | return PTR_ERR(mnt); | |
2141 | ||
95bc5f25 | 2142 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2143 | if (err) |
2144 | mntput(mnt); | |
2145 | return err; | |
1da177e4 LT |
2146 | } |
2147 | ||
19a167af AV |
2148 | int finish_automount(struct vfsmount *m, struct path *path) |
2149 | { | |
6776db3d | 2150 | struct mount *mnt = real_mount(m); |
19a167af AV |
2151 | int err; |
2152 | /* The new mount record should have at least 2 refs to prevent it being | |
2153 | * expired before we get a chance to add it | |
2154 | */ | |
6776db3d | 2155 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2156 | |
2157 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2158 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2159 | err = -ELOOP; |
2160 | goto fail; | |
19a167af AV |
2161 | } |
2162 | ||
95bc5f25 | 2163 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2164 | if (!err) |
2165 | return 0; | |
2166 | fail: | |
2167 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2168 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2169 | namespace_lock(); |
6776db3d | 2170 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2171 | namespace_unlock(); |
19a167af | 2172 | } |
b1e75df4 AV |
2173 | mntput(m); |
2174 | mntput(m); | |
19a167af AV |
2175 | return err; |
2176 | } | |
2177 | ||
ea5b778a DH |
2178 | /** |
2179 | * mnt_set_expiry - Put a mount on an expiration list | |
2180 | * @mnt: The mount to list. | |
2181 | * @expiry_list: The list to add the mount to. | |
2182 | */ | |
2183 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2184 | { | |
97216be0 | 2185 | namespace_lock(); |
ea5b778a | 2186 | |
6776db3d | 2187 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2188 | |
97216be0 | 2189 | namespace_unlock(); |
ea5b778a DH |
2190 | } |
2191 | EXPORT_SYMBOL(mnt_set_expiry); | |
2192 | ||
1da177e4 LT |
2193 | /* |
2194 | * process a list of expirable mountpoints with the intent of discarding any | |
2195 | * mountpoints that aren't in use and haven't been touched since last we came | |
2196 | * here | |
2197 | */ | |
2198 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2199 | { | |
761d5c38 | 2200 | struct mount *mnt, *next; |
1da177e4 LT |
2201 | LIST_HEAD(graveyard); |
2202 | ||
2203 | if (list_empty(mounts)) | |
2204 | return; | |
2205 | ||
97216be0 | 2206 | namespace_lock(); |
719ea2fb | 2207 | lock_mount_hash(); |
1da177e4 LT |
2208 | |
2209 | /* extract from the expiration list every vfsmount that matches the | |
2210 | * following criteria: | |
2211 | * - only referenced by its parent vfsmount | |
2212 | * - still marked for expiry (marked on the last call here; marks are | |
2213 | * cleared by mntput()) | |
2214 | */ | |
6776db3d | 2215 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2216 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2217 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2218 | continue; |
6776db3d | 2219 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2220 | } |
bcc5c7d2 | 2221 | while (!list_empty(&graveyard)) { |
6776db3d | 2222 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2223 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2224 | umount_tree(mnt, 1); |
bcc5c7d2 | 2225 | } |
719ea2fb | 2226 | unlock_mount_hash(); |
3ab6abee | 2227 | namespace_unlock(); |
5528f911 TM |
2228 | } |
2229 | ||
2230 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2231 | ||
2232 | /* | |
2233 | * Ripoff of 'select_parent()' | |
2234 | * | |
2235 | * search the list of submounts for a given mountpoint, and move any | |
2236 | * shrinkable submounts to the 'graveyard' list. | |
2237 | */ | |
692afc31 | 2238 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2239 | { |
692afc31 | 2240 | struct mount *this_parent = parent; |
5528f911 TM |
2241 | struct list_head *next; |
2242 | int found = 0; | |
2243 | ||
2244 | repeat: | |
6b41d536 | 2245 | next = this_parent->mnt_mounts.next; |
5528f911 | 2246 | resume: |
6b41d536 | 2247 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2248 | struct list_head *tmp = next; |
6b41d536 | 2249 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2250 | |
2251 | next = tmp->next; | |
692afc31 | 2252 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2253 | continue; |
5528f911 TM |
2254 | /* |
2255 | * Descend a level if the d_mounts list is non-empty. | |
2256 | */ | |
6b41d536 | 2257 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2258 | this_parent = mnt; |
2259 | goto repeat; | |
2260 | } | |
1da177e4 | 2261 | |
1ab59738 | 2262 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2263 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2264 | found++; |
2265 | } | |
1da177e4 | 2266 | } |
5528f911 TM |
2267 | /* |
2268 | * All done at this level ... ascend and resume the search | |
2269 | */ | |
2270 | if (this_parent != parent) { | |
6b41d536 | 2271 | next = this_parent->mnt_child.next; |
0714a533 | 2272 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2273 | goto resume; |
2274 | } | |
2275 | return found; | |
2276 | } | |
2277 | ||
2278 | /* | |
2279 | * process a list of expirable mountpoints with the intent of discarding any | |
2280 | * submounts of a specific parent mountpoint | |
99b7db7b | 2281 | * |
48a066e7 | 2282 | * mount_lock must be held for write |
5528f911 | 2283 | */ |
b54b9be7 | 2284 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2285 | { |
2286 | LIST_HEAD(graveyard); | |
761d5c38 | 2287 | struct mount *m; |
5528f911 | 2288 | |
5528f911 | 2289 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2290 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2291 | while (!list_empty(&graveyard)) { |
761d5c38 | 2292 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2293 | mnt_expire); |
143c8c91 | 2294 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2295 | umount_tree(m, 1); |
bcc5c7d2 AV |
2296 | } |
2297 | } | |
1da177e4 LT |
2298 | } |
2299 | ||
1da177e4 LT |
2300 | /* |
2301 | * Some copy_from_user() implementations do not return the exact number of | |
2302 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2303 | * Note that this function differs from copy_from_user() in that it will oops | |
2304 | * on bad values of `to', rather than returning a short copy. | |
2305 | */ | |
b58fed8b RP |
2306 | static long exact_copy_from_user(void *to, const void __user * from, |
2307 | unsigned long n) | |
1da177e4 LT |
2308 | { |
2309 | char *t = to; | |
2310 | const char __user *f = from; | |
2311 | char c; | |
2312 | ||
2313 | if (!access_ok(VERIFY_READ, from, n)) | |
2314 | return n; | |
2315 | ||
2316 | while (n) { | |
2317 | if (__get_user(c, f)) { | |
2318 | memset(t, 0, n); | |
2319 | break; | |
2320 | } | |
2321 | *t++ = c; | |
2322 | f++; | |
2323 | n--; | |
2324 | } | |
2325 | return n; | |
2326 | } | |
2327 | ||
b58fed8b | 2328 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2329 | { |
2330 | int i; | |
2331 | unsigned long page; | |
2332 | unsigned long size; | |
b58fed8b | 2333 | |
1da177e4 LT |
2334 | *where = 0; |
2335 | if (!data) | |
2336 | return 0; | |
2337 | ||
2338 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2339 | return -ENOMEM; | |
2340 | ||
2341 | /* We only care that *some* data at the address the user | |
2342 | * gave us is valid. Just in case, we'll zero | |
2343 | * the remainder of the page. | |
2344 | */ | |
2345 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2346 | size = TASK_SIZE - (unsigned long)data; | |
2347 | if (size > PAGE_SIZE) | |
2348 | size = PAGE_SIZE; | |
2349 | ||
2350 | i = size - exact_copy_from_user((void *)page, data, size); | |
2351 | if (!i) { | |
b58fed8b | 2352 | free_page(page); |
1da177e4 LT |
2353 | return -EFAULT; |
2354 | } | |
2355 | if (i != PAGE_SIZE) | |
2356 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2357 | *where = page; | |
2358 | return 0; | |
2359 | } | |
2360 | ||
eca6f534 VN |
2361 | int copy_mount_string(const void __user *data, char **where) |
2362 | { | |
2363 | char *tmp; | |
2364 | ||
2365 | if (!data) { | |
2366 | *where = NULL; | |
2367 | return 0; | |
2368 | } | |
2369 | ||
2370 | tmp = strndup_user(data, PAGE_SIZE); | |
2371 | if (IS_ERR(tmp)) | |
2372 | return PTR_ERR(tmp); | |
2373 | ||
2374 | *where = tmp; | |
2375 | return 0; | |
2376 | } | |
2377 | ||
1da177e4 LT |
2378 | /* |
2379 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2380 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2381 | * | |
2382 | * data is a (void *) that can point to any structure up to | |
2383 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2384 | * information (or be NULL). | |
2385 | * | |
2386 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2387 | * When the flags word was introduced its top half was required | |
2388 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2389 | * Therefore, if this magic number is present, it carries no information | |
2390 | * and must be discarded. | |
2391 | */ | |
808d4e3c AV |
2392 | long do_mount(const char *dev_name, const char *dir_name, |
2393 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2394 | { |
2d92ab3c | 2395 | struct path path; |
1da177e4 LT |
2396 | int retval = 0; |
2397 | int mnt_flags = 0; | |
2398 | ||
2399 | /* Discard magic */ | |
2400 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2401 | flags &= ~MS_MGC_MSK; | |
2402 | ||
2403 | /* Basic sanity checks */ | |
2404 | ||
2405 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2406 | return -EINVAL; | |
1da177e4 LT |
2407 | |
2408 | if (data_page) | |
2409 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2410 | ||
a27ab9f2 TH |
2411 | /* ... and get the mountpoint */ |
2412 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2413 | if (retval) | |
2414 | return retval; | |
2415 | ||
2416 | retval = security_sb_mount(dev_name, &path, | |
2417 | type_page, flags, data_page); | |
0d5cadb8 AV |
2418 | if (!retval && !may_mount()) |
2419 | retval = -EPERM; | |
a27ab9f2 TH |
2420 | if (retval) |
2421 | goto dput_out; | |
2422 | ||
613cbe3d AK |
2423 | /* Default to relatime unless overriden */ |
2424 | if (!(flags & MS_NOATIME)) | |
2425 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2426 | |
1da177e4 LT |
2427 | /* Separate the per-mountpoint flags */ |
2428 | if (flags & MS_NOSUID) | |
2429 | mnt_flags |= MNT_NOSUID; | |
2430 | if (flags & MS_NODEV) | |
2431 | mnt_flags |= MNT_NODEV; | |
2432 | if (flags & MS_NOEXEC) | |
2433 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2434 | if (flags & MS_NOATIME) |
2435 | mnt_flags |= MNT_NOATIME; | |
2436 | if (flags & MS_NODIRATIME) | |
2437 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2438 | if (flags & MS_STRICTATIME) |
2439 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2440 | if (flags & MS_RDONLY) |
2441 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2442 | |
7a4dec53 | 2443 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2444 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2445 | MS_STRICTATIME); | |
1da177e4 | 2446 | |
1da177e4 | 2447 | if (flags & MS_REMOUNT) |
2d92ab3c | 2448 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2449 | data_page); |
2450 | else if (flags & MS_BIND) | |
2d92ab3c | 2451 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2452 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2453 | retval = do_change_type(&path, flags); |
1da177e4 | 2454 | else if (flags & MS_MOVE) |
2d92ab3c | 2455 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2456 | else |
2d92ab3c | 2457 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2458 | dev_name, data_page); |
2459 | dput_out: | |
2d92ab3c | 2460 | path_put(&path); |
1da177e4 LT |
2461 | return retval; |
2462 | } | |
2463 | ||
771b1371 EB |
2464 | static void free_mnt_ns(struct mnt_namespace *ns) |
2465 | { | |
98f842e6 | 2466 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2467 | put_user_ns(ns->user_ns); |
2468 | kfree(ns); | |
2469 | } | |
2470 | ||
8823c079 EB |
2471 | /* |
2472 | * Assign a sequence number so we can detect when we attempt to bind | |
2473 | * mount a reference to an older mount namespace into the current | |
2474 | * mount namespace, preventing reference counting loops. A 64bit | |
2475 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2476 | * is effectively never, so we can ignore the possibility. | |
2477 | */ | |
2478 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2479 | ||
771b1371 | 2480 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2481 | { |
2482 | struct mnt_namespace *new_ns; | |
98f842e6 | 2483 | int ret; |
cf8d2c11 TM |
2484 | |
2485 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2486 | if (!new_ns) | |
2487 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2488 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2489 | if (ret) { | |
2490 | kfree(new_ns); | |
2491 | return ERR_PTR(ret); | |
2492 | } | |
8823c079 | 2493 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2494 | atomic_set(&new_ns->count, 1); |
2495 | new_ns->root = NULL; | |
2496 | INIT_LIST_HEAD(&new_ns->list); | |
2497 | init_waitqueue_head(&new_ns->poll); | |
2498 | new_ns->event = 0; | |
771b1371 | 2499 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2500 | return new_ns; |
2501 | } | |
2502 | ||
9559f689 AV |
2503 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2504 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2505 | { |
6b3286ed | 2506 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2507 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2508 | struct mount *p, *q; |
9559f689 | 2509 | struct mount *old; |
cb338d06 | 2510 | struct mount *new; |
7a472ef4 | 2511 | int copy_flags; |
1da177e4 | 2512 | |
9559f689 AV |
2513 | BUG_ON(!ns); |
2514 | ||
2515 | if (likely(!(flags & CLONE_NEWNS))) { | |
2516 | get_mnt_ns(ns); | |
2517 | return ns; | |
2518 | } | |
2519 | ||
2520 | old = ns->root; | |
2521 | ||
771b1371 | 2522 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2523 | if (IS_ERR(new_ns)) |
2524 | return new_ns; | |
1da177e4 | 2525 | |
97216be0 | 2526 | namespace_lock(); |
1da177e4 | 2527 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2528 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2529 | if (user_ns != ns->user_ns) |
132c94e3 | 2530 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2531 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2532 | if (IS_ERR(new)) { |
328e6d90 | 2533 | namespace_unlock(); |
771b1371 | 2534 | free_mnt_ns(new_ns); |
be34d1a3 | 2535 | return ERR_CAST(new); |
1da177e4 | 2536 | } |
be08d6d2 | 2537 | new_ns->root = new; |
1a4eeaf2 | 2538 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2539 | |
2540 | /* | |
2541 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2542 | * as belonging to new namespace. We have already acquired a private | |
2543 | * fs_struct, so tsk->fs->lock is not needed. | |
2544 | */ | |
909b0a88 | 2545 | p = old; |
cb338d06 | 2546 | q = new; |
1da177e4 | 2547 | while (p) { |
143c8c91 | 2548 | q->mnt_ns = new_ns; |
9559f689 AV |
2549 | if (new_fs) { |
2550 | if (&p->mnt == new_fs->root.mnt) { | |
2551 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2552 | rootmnt = &p->mnt; |
1da177e4 | 2553 | } |
9559f689 AV |
2554 | if (&p->mnt == new_fs->pwd.mnt) { |
2555 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2556 | pwdmnt = &p->mnt; |
1da177e4 | 2557 | } |
1da177e4 | 2558 | } |
909b0a88 AV |
2559 | p = next_mnt(p, old); |
2560 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2561 | if (!q) |
2562 | break; | |
2563 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2564 | p = next_mnt(p, old); | |
1da177e4 | 2565 | } |
328e6d90 | 2566 | namespace_unlock(); |
1da177e4 | 2567 | |
1da177e4 | 2568 | if (rootmnt) |
f03c6599 | 2569 | mntput(rootmnt); |
1da177e4 | 2570 | if (pwdmnt) |
f03c6599 | 2571 | mntput(pwdmnt); |
1da177e4 | 2572 | |
741a2951 | 2573 | return new_ns; |
1da177e4 LT |
2574 | } |
2575 | ||
cf8d2c11 TM |
2576 | /** |
2577 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2578 | * @mnt: pointer to the new root filesystem mountpoint | |
2579 | */ | |
1a4eeaf2 | 2580 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2581 | { |
771b1371 | 2582 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2583 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2584 | struct mount *mnt = real_mount(m); |
2585 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2586 | new_ns->root = mnt; |
b1983cd8 | 2587 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2588 | } else { |
1a4eeaf2 | 2589 | mntput(m); |
cf8d2c11 TM |
2590 | } |
2591 | return new_ns; | |
2592 | } | |
cf8d2c11 | 2593 | |
ea441d11 AV |
2594 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2595 | { | |
2596 | struct mnt_namespace *ns; | |
d31da0f0 | 2597 | struct super_block *s; |
ea441d11 AV |
2598 | struct path path; |
2599 | int err; | |
2600 | ||
2601 | ns = create_mnt_ns(mnt); | |
2602 | if (IS_ERR(ns)) | |
2603 | return ERR_CAST(ns); | |
2604 | ||
2605 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2606 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2607 | ||
2608 | put_mnt_ns(ns); | |
2609 | ||
2610 | if (err) | |
2611 | return ERR_PTR(err); | |
2612 | ||
2613 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2614 | s = path.mnt->mnt_sb; |
2615 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2616 | mntput(path.mnt); |
2617 | /* lock the sucker */ | |
d31da0f0 | 2618 | down_write(&s->s_umount); |
ea441d11 AV |
2619 | /* ... and return the root of (sub)tree on it */ |
2620 | return path.dentry; | |
2621 | } | |
2622 | EXPORT_SYMBOL(mount_subtree); | |
2623 | ||
bdc480e3 HC |
2624 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2625 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2626 | { |
eca6f534 VN |
2627 | int ret; |
2628 | char *kernel_type; | |
91a27b2a | 2629 | struct filename *kernel_dir; |
eca6f534 | 2630 | char *kernel_dev; |
1da177e4 | 2631 | unsigned long data_page; |
1da177e4 | 2632 | |
eca6f534 VN |
2633 | ret = copy_mount_string(type, &kernel_type); |
2634 | if (ret < 0) | |
2635 | goto out_type; | |
1da177e4 | 2636 | |
eca6f534 VN |
2637 | kernel_dir = getname(dir_name); |
2638 | if (IS_ERR(kernel_dir)) { | |
2639 | ret = PTR_ERR(kernel_dir); | |
2640 | goto out_dir; | |
2641 | } | |
1da177e4 | 2642 | |
eca6f534 VN |
2643 | ret = copy_mount_string(dev_name, &kernel_dev); |
2644 | if (ret < 0) | |
2645 | goto out_dev; | |
1da177e4 | 2646 | |
eca6f534 VN |
2647 | ret = copy_mount_options(data, &data_page); |
2648 | if (ret < 0) | |
2649 | goto out_data; | |
1da177e4 | 2650 | |
91a27b2a | 2651 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2652 | (void *) data_page); |
1da177e4 | 2653 | |
eca6f534 VN |
2654 | free_page(data_page); |
2655 | out_data: | |
2656 | kfree(kernel_dev); | |
2657 | out_dev: | |
2658 | putname(kernel_dir); | |
2659 | out_dir: | |
2660 | kfree(kernel_type); | |
2661 | out_type: | |
2662 | return ret; | |
1da177e4 LT |
2663 | } |
2664 | ||
afac7cba AV |
2665 | /* |
2666 | * Return true if path is reachable from root | |
2667 | * | |
48a066e7 | 2668 | * namespace_sem or mount_lock is held |
afac7cba | 2669 | */ |
643822b4 | 2670 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2671 | const struct path *root) |
2672 | { | |
643822b4 | 2673 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2674 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2675 | mnt = mnt->mnt_parent; |
afac7cba | 2676 | } |
643822b4 | 2677 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2678 | } |
2679 | ||
2680 | int path_is_under(struct path *path1, struct path *path2) | |
2681 | { | |
2682 | int res; | |
48a066e7 | 2683 | read_seqlock_excl(&mount_lock); |
643822b4 | 2684 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2685 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2686 | return res; |
2687 | } | |
2688 | EXPORT_SYMBOL(path_is_under); | |
2689 | ||
1da177e4 LT |
2690 | /* |
2691 | * pivot_root Semantics: | |
2692 | * Moves the root file system of the current process to the directory put_old, | |
2693 | * makes new_root as the new root file system of the current process, and sets | |
2694 | * root/cwd of all processes which had them on the current root to new_root. | |
2695 | * | |
2696 | * Restrictions: | |
2697 | * The new_root and put_old must be directories, and must not be on the | |
2698 | * same file system as the current process root. The put_old must be | |
2699 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2700 | * pointed to by put_old must yield the same directory as new_root. No other | |
2701 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2702 | * | |
4a0d11fa NB |
2703 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2704 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2705 | * in this situation. | |
2706 | * | |
1da177e4 LT |
2707 | * Notes: |
2708 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2709 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2710 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2711 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2712 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2713 | * first. | |
2714 | */ | |
3480b257 HC |
2715 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2716 | const char __user *, put_old) | |
1da177e4 | 2717 | { |
2d8f3038 | 2718 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2719 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2720 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2721 | int error; |
2722 | ||
9b40bc90 | 2723 | if (!may_mount()) |
1da177e4 LT |
2724 | return -EPERM; |
2725 | ||
2d8f3038 | 2726 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2727 | if (error) |
2728 | goto out0; | |
1da177e4 | 2729 | |
2d8f3038 | 2730 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2731 | if (error) |
2732 | goto out1; | |
2733 | ||
2d8f3038 | 2734 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2735 | if (error) |
2736 | goto out2; | |
1da177e4 | 2737 | |
f7ad3c6b | 2738 | get_fs_root(current->fs, &root); |
84d17192 AV |
2739 | old_mp = lock_mount(&old); |
2740 | error = PTR_ERR(old_mp); | |
2741 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2742 | goto out3; |
2743 | ||
1da177e4 | 2744 | error = -EINVAL; |
419148da AV |
2745 | new_mnt = real_mount(new.mnt); |
2746 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2747 | old_mnt = real_mount(old.mnt); |
2748 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2749 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2750 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2751 | goto out4; |
143c8c91 | 2752 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2753 | goto out4; |
5ff9d8a6 EB |
2754 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2755 | goto out4; | |
1da177e4 | 2756 | error = -ENOENT; |
f3da392e | 2757 | if (d_unlinked(new.dentry)) |
b12cea91 | 2758 | goto out4; |
1da177e4 | 2759 | error = -EBUSY; |
84d17192 | 2760 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2761 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2762 | error = -EINVAL; |
8c3ee42e | 2763 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2764 | goto out4; /* not a mountpoint */ |
676da58d | 2765 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2766 | goto out4; /* not attached */ |
84d17192 | 2767 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2768 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2769 | goto out4; /* not a mountpoint */ |
676da58d | 2770 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2771 | goto out4; /* not attached */ |
4ac91378 | 2772 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2773 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2774 | goto out4; |
84d17192 | 2775 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2776 | lock_mount_hash(); |
419148da AV |
2777 | detach_mnt(new_mnt, &parent_path); |
2778 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2779 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2780 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2781 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2782 | } | |
4ac91378 | 2783 | /* mount old root on put_old */ |
84d17192 | 2784 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2785 | /* mount new_root on / */ |
84d17192 | 2786 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2787 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2788 | unlock_mount_hash(); |
2d8f3038 | 2789 | chroot_fs_refs(&root, &new); |
84d17192 | 2790 | put_mountpoint(root_mp); |
1da177e4 | 2791 | error = 0; |
b12cea91 | 2792 | out4: |
84d17192 | 2793 | unlock_mount(old_mp); |
b12cea91 AV |
2794 | if (!error) { |
2795 | path_put(&root_parent); | |
2796 | path_put(&parent_path); | |
2797 | } | |
2798 | out3: | |
8c3ee42e | 2799 | path_put(&root); |
b12cea91 | 2800 | out2: |
2d8f3038 | 2801 | path_put(&old); |
1da177e4 | 2802 | out1: |
2d8f3038 | 2803 | path_put(&new); |
1da177e4 | 2804 | out0: |
1da177e4 | 2805 | return error; |
1da177e4 LT |
2806 | } |
2807 | ||
2808 | static void __init init_mount_tree(void) | |
2809 | { | |
2810 | struct vfsmount *mnt; | |
6b3286ed | 2811 | struct mnt_namespace *ns; |
ac748a09 | 2812 | struct path root; |
0c55cfc4 | 2813 | struct file_system_type *type; |
1da177e4 | 2814 | |
0c55cfc4 EB |
2815 | type = get_fs_type("rootfs"); |
2816 | if (!type) | |
2817 | panic("Can't find rootfs type"); | |
2818 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2819 | put_filesystem(type); | |
1da177e4 LT |
2820 | if (IS_ERR(mnt)) |
2821 | panic("Can't create rootfs"); | |
b3e19d92 | 2822 | |
3b22edc5 TM |
2823 | ns = create_mnt_ns(mnt); |
2824 | if (IS_ERR(ns)) | |
1da177e4 | 2825 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2826 | |
2827 | init_task.nsproxy->mnt_ns = ns; | |
2828 | get_mnt_ns(ns); | |
2829 | ||
be08d6d2 AV |
2830 | root.mnt = mnt; |
2831 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2832 | |
2833 | set_fs_pwd(current->fs, &root); | |
2834 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2835 | } |
2836 | ||
74bf17cf | 2837 | void __init mnt_init(void) |
1da177e4 | 2838 | { |
13f14b4d | 2839 | unsigned u; |
15a67dd8 | 2840 | int err; |
1da177e4 | 2841 | |
7d6fec45 | 2842 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2843 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2844 | |
0818bf27 | 2845 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2846 | sizeof(struct hlist_head), |
0818bf27 AV |
2847 | mhash_entries, 19, |
2848 | 0, | |
2849 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2850 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2851 | sizeof(struct hlist_head), | |
2852 | mphash_entries, 19, | |
2853 | 0, | |
2854 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2855 | |
84d17192 | 2856 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2857 | panic("Failed to allocate mount hash table\n"); |
2858 | ||
0818bf27 | 2859 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2860 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2861 | for (u = 0; u <= mp_hash_mask; u++) |
2862 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2863 | |
4b93dc9b TH |
2864 | kernfs_init(); |
2865 | ||
15a67dd8 RD |
2866 | err = sysfs_init(); |
2867 | if (err) | |
2868 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2869 | __func__, err); |
00d26666 GKH |
2870 | fs_kobj = kobject_create_and_add("fs", NULL); |
2871 | if (!fs_kobj) | |
8e24eea7 | 2872 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2873 | init_rootfs(); |
2874 | init_mount_tree(); | |
2875 | } | |
2876 | ||
616511d0 | 2877 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2878 | { |
d498b25a | 2879 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2880 | return; |
7b00ed6f | 2881 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2882 | free_mnt_ns(ns); |
1da177e4 | 2883 | } |
9d412a43 AV |
2884 | |
2885 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2886 | { | |
423e0ab0 TC |
2887 | struct vfsmount *mnt; |
2888 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2889 | if (!IS_ERR(mnt)) { | |
2890 | /* | |
2891 | * it is a longterm mount, don't release mnt until | |
2892 | * we unmount before file sys is unregistered | |
2893 | */ | |
f7a99c5b | 2894 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2895 | } |
2896 | return mnt; | |
9d412a43 AV |
2897 | } |
2898 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2899 | |
2900 | void kern_unmount(struct vfsmount *mnt) | |
2901 | { | |
2902 | /* release long term mount so mount point can be released */ | |
2903 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2904 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2905 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2906 | mntput(mnt); |
2907 | } | |
2908 | } | |
2909 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2910 | |
2911 | bool our_mnt(struct vfsmount *mnt) | |
2912 | { | |
143c8c91 | 2913 | return check_mnt(real_mount(mnt)); |
02125a82 | 2914 | } |
8823c079 | 2915 | |
3151527e EB |
2916 | bool current_chrooted(void) |
2917 | { | |
2918 | /* Does the current process have a non-standard root */ | |
2919 | struct path ns_root; | |
2920 | struct path fs_root; | |
2921 | bool chrooted; | |
2922 | ||
2923 | /* Find the namespace root */ | |
2924 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2925 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2926 | path_get(&ns_root); | |
2927 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2928 | ; | |
2929 | ||
2930 | get_fs_root(current->fs, &fs_root); | |
2931 | ||
2932 | chrooted = !path_equal(&fs_root, &ns_root); | |
2933 | ||
2934 | path_put(&fs_root); | |
2935 | path_put(&ns_root); | |
2936 | ||
2937 | return chrooted; | |
2938 | } | |
2939 | ||
e51db735 | 2940 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
2941 | { |
2942 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2943 | struct mount *mnt; | |
e51db735 | 2944 | bool visible = false; |
87a8ebd6 | 2945 | |
e51db735 EB |
2946 | if (unlikely(!ns)) |
2947 | return false; | |
2948 | ||
44bb4385 | 2949 | down_read(&namespace_sem); |
87a8ebd6 | 2950 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
2951 | struct mount *child; |
2952 | if (mnt->mnt.mnt_sb->s_type != type) | |
2953 | continue; | |
2954 | ||
2955 | /* This mount is not fully visible if there are any child mounts | |
2956 | * that cover anything except for empty directories. | |
2957 | */ | |
2958 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2959 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
2960 | if (!S_ISDIR(inode->i_mode)) | |
2961 | goto next; | |
41301ae7 | 2962 | if (inode->i_nlink > 2) |
e51db735 | 2963 | goto next; |
87a8ebd6 | 2964 | } |
e51db735 EB |
2965 | visible = true; |
2966 | goto found; | |
2967 | next: ; | |
87a8ebd6 | 2968 | } |
e51db735 | 2969 | found: |
44bb4385 | 2970 | up_read(&namespace_sem); |
e51db735 | 2971 | return visible; |
87a8ebd6 EB |
2972 | } |
2973 | ||
8823c079 EB |
2974 | static void *mntns_get(struct task_struct *task) |
2975 | { | |
2976 | struct mnt_namespace *ns = NULL; | |
2977 | struct nsproxy *nsproxy; | |
2978 | ||
2979 | rcu_read_lock(); | |
2980 | nsproxy = task_nsproxy(task); | |
2981 | if (nsproxy) { | |
2982 | ns = nsproxy->mnt_ns; | |
2983 | get_mnt_ns(ns); | |
2984 | } | |
2985 | rcu_read_unlock(); | |
2986 | ||
2987 | return ns; | |
2988 | } | |
2989 | ||
2990 | static void mntns_put(void *ns) | |
2991 | { | |
2992 | put_mnt_ns(ns); | |
2993 | } | |
2994 | ||
2995 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2996 | { | |
2997 | struct fs_struct *fs = current->fs; | |
2998 | struct mnt_namespace *mnt_ns = ns; | |
2999 | struct path root; | |
3000 | ||
0c55cfc4 | 3001 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3002 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3003 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3004 | return -EPERM; |
8823c079 EB |
3005 | |
3006 | if (fs->users != 1) | |
3007 | return -EINVAL; | |
3008 | ||
3009 | get_mnt_ns(mnt_ns); | |
3010 | put_mnt_ns(nsproxy->mnt_ns); | |
3011 | nsproxy->mnt_ns = mnt_ns; | |
3012 | ||
3013 | /* Find the root */ | |
3014 | root.mnt = &mnt_ns->root->mnt; | |
3015 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3016 | path_get(&root); | |
3017 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3018 | ; | |
3019 | ||
3020 | /* Update the pwd and root */ | |
3021 | set_fs_pwd(fs, &root); | |
3022 | set_fs_root(fs, &root); | |
3023 | ||
3024 | path_put(&root); | |
3025 | return 0; | |
3026 | } | |
3027 | ||
98f842e6 EB |
3028 | static unsigned int mntns_inum(void *ns) |
3029 | { | |
3030 | struct mnt_namespace *mnt_ns = ns; | |
3031 | return mnt_ns->proc_inum; | |
3032 | } | |
3033 | ||
8823c079 EB |
3034 | const struct proc_ns_operations mntns_operations = { |
3035 | .name = "mnt", | |
3036 | .type = CLONE_NEWNS, | |
3037 | .get = mntns_get, | |
3038 | .put = mntns_put, | |
3039 | .install = mntns_install, | |
98f842e6 | 3040 | .inum = mntns_inum, |
8823c079 | 3041 | }; |