]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
9ea459e1 | 26 | #include <linux/task_work.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); | |
0a5eb7c8 | 228 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
229 | #ifdef CONFIG_FSNOTIFY |
230 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 231 | #endif |
1da177e4 | 232 | } |
c63181e6 | 233 | return mnt; |
88b38782 | 234 | |
d3ef3d73 | 235 | #ifdef CONFIG_SMP |
236 | out_free_devname: | |
c63181e6 | 237 | kfree(mnt->mnt_devname); |
d3ef3d73 | 238 | #endif |
88b38782 | 239 | out_free_id: |
c63181e6 | 240 | mnt_free_id(mnt); |
88b38782 | 241 | out_free_cache: |
c63181e6 | 242 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 243 | return NULL; |
1da177e4 LT |
244 | } |
245 | ||
3d733633 DH |
246 | /* |
247 | * Most r/o checks on a fs are for operations that take | |
248 | * discrete amounts of time, like a write() or unlink(). | |
249 | * We must keep track of when those operations start | |
250 | * (for permission checks) and when they end, so that | |
251 | * we can determine when writes are able to occur to | |
252 | * a filesystem. | |
253 | */ | |
254 | /* | |
255 | * __mnt_is_readonly: check whether a mount is read-only | |
256 | * @mnt: the mount to check for its write status | |
257 | * | |
258 | * This shouldn't be used directly ouside of the VFS. | |
259 | * It does not guarantee that the filesystem will stay | |
260 | * r/w, just that it is right *now*. This can not and | |
261 | * should not be used in place of IS_RDONLY(inode). | |
262 | * mnt_want/drop_write() will _keep_ the filesystem | |
263 | * r/w. | |
264 | */ | |
265 | int __mnt_is_readonly(struct vfsmount *mnt) | |
266 | { | |
2e4b7fcd DH |
267 | if (mnt->mnt_flags & MNT_READONLY) |
268 | return 1; | |
269 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
270 | return 1; | |
271 | return 0; | |
3d733633 DH |
272 | } |
273 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
274 | ||
83adc753 | 275 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 276 | { |
277 | #ifdef CONFIG_SMP | |
68e8a9fe | 278 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 279 | #else |
68e8a9fe | 280 | mnt->mnt_writers++; |
d3ef3d73 | 281 | #endif |
282 | } | |
3d733633 | 283 | |
83adc753 | 284 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 285 | { |
d3ef3d73 | 286 | #ifdef CONFIG_SMP |
68e8a9fe | 287 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 288 | #else |
68e8a9fe | 289 | mnt->mnt_writers--; |
d3ef3d73 | 290 | #endif |
3d733633 | 291 | } |
3d733633 | 292 | |
83adc753 | 293 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 294 | { |
d3ef3d73 | 295 | #ifdef CONFIG_SMP |
296 | unsigned int count = 0; | |
3d733633 | 297 | int cpu; |
3d733633 DH |
298 | |
299 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 300 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 301 | } |
3d733633 | 302 | |
d3ef3d73 | 303 | return count; |
304 | #else | |
305 | return mnt->mnt_writers; | |
306 | #endif | |
3d733633 DH |
307 | } |
308 | ||
4ed5e82f MS |
309 | static int mnt_is_readonly(struct vfsmount *mnt) |
310 | { | |
311 | if (mnt->mnt_sb->s_readonly_remount) | |
312 | return 1; | |
313 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
314 | smp_rmb(); | |
315 | return __mnt_is_readonly(mnt); | |
316 | } | |
317 | ||
8366025e | 318 | /* |
eb04c282 JK |
319 | * Most r/o & frozen checks on a fs are for operations that take discrete |
320 | * amounts of time, like a write() or unlink(). We must keep track of when | |
321 | * those operations start (for permission checks) and when they end, so that we | |
322 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
323 | */ |
324 | /** | |
eb04c282 | 325 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 326 | * @m: the mount on which to take a write |
8366025e | 327 | * |
eb04c282 JK |
328 | * This tells the low-level filesystem that a write is about to be performed to |
329 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
330 | * returning success. This operation does not protect against filesystem being | |
331 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
332 | * called. This is effectively a refcount. | |
8366025e | 333 | */ |
eb04c282 | 334 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 335 | { |
83adc753 | 336 | struct mount *mnt = real_mount(m); |
3d733633 | 337 | int ret = 0; |
3d733633 | 338 | |
d3ef3d73 | 339 | preempt_disable(); |
c6653a83 | 340 | mnt_inc_writers(mnt); |
d3ef3d73 | 341 | /* |
c6653a83 | 342 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 343 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
344 | * incremented count after it has set MNT_WRITE_HOLD. | |
345 | */ | |
346 | smp_mb(); | |
1e75529e | 347 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 348 | cpu_relax(); |
349 | /* | |
350 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
351 | * be set to match its requirements. So we must not load that until | |
352 | * MNT_WRITE_HOLD is cleared. | |
353 | */ | |
354 | smp_rmb(); | |
4ed5e82f | 355 | if (mnt_is_readonly(m)) { |
c6653a83 | 356 | mnt_dec_writers(mnt); |
3d733633 | 357 | ret = -EROFS; |
3d733633 | 358 | } |
d3ef3d73 | 359 | preempt_enable(); |
eb04c282 JK |
360 | |
361 | return ret; | |
362 | } | |
363 | ||
364 | /** | |
365 | * mnt_want_write - get write access to a mount | |
366 | * @m: the mount on which to take a write | |
367 | * | |
368 | * This tells the low-level filesystem that a write is about to be performed to | |
369 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
370 | * is not frozen) before returning success. When the write operation is | |
371 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
372 | */ | |
373 | int mnt_want_write(struct vfsmount *m) | |
374 | { | |
375 | int ret; | |
376 | ||
377 | sb_start_write(m->mnt_sb); | |
378 | ret = __mnt_want_write(m); | |
379 | if (ret) | |
380 | sb_end_write(m->mnt_sb); | |
3d733633 | 381 | return ret; |
8366025e DH |
382 | } |
383 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
384 | ||
96029c4e | 385 | /** |
386 | * mnt_clone_write - get write access to a mount | |
387 | * @mnt: the mount on which to take a write | |
388 | * | |
389 | * This is effectively like mnt_want_write, except | |
390 | * it must only be used to take an extra write reference | |
391 | * on a mountpoint that we already know has a write reference | |
392 | * on it. This allows some optimisation. | |
393 | * | |
394 | * After finished, mnt_drop_write must be called as usual to | |
395 | * drop the reference. | |
396 | */ | |
397 | int mnt_clone_write(struct vfsmount *mnt) | |
398 | { | |
399 | /* superblock may be r/o */ | |
400 | if (__mnt_is_readonly(mnt)) | |
401 | return -EROFS; | |
402 | preempt_disable(); | |
83adc753 | 403 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 404 | preempt_enable(); |
405 | return 0; | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
408 | ||
409 | /** | |
eb04c282 | 410 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 411 | * @file: the file who's mount on which to take a write |
412 | * | |
eb04c282 | 413 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 414 | * do some optimisations if the file is open for write already |
415 | */ | |
eb04c282 | 416 | int __mnt_want_write_file(struct file *file) |
96029c4e | 417 | { |
83f936c7 | 418 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 419 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 420 | else |
421 | return mnt_clone_write(file->f_path.mnt); | |
422 | } | |
eb04c282 JK |
423 | |
424 | /** | |
425 | * mnt_want_write_file - get write access to a file's mount | |
426 | * @file: the file who's mount on which to take a write | |
427 | * | |
428 | * This is like mnt_want_write, but it takes a file and can | |
429 | * do some optimisations if the file is open for write already | |
430 | */ | |
431 | int mnt_want_write_file(struct file *file) | |
432 | { | |
433 | int ret; | |
434 | ||
435 | sb_start_write(file->f_path.mnt->mnt_sb); | |
436 | ret = __mnt_want_write_file(file); | |
437 | if (ret) | |
438 | sb_end_write(file->f_path.mnt->mnt_sb); | |
439 | return ret; | |
440 | } | |
96029c4e | 441 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
442 | ||
8366025e | 443 | /** |
eb04c282 | 444 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
445 | * @mnt: the mount on which to give up write access |
446 | * | |
447 | * Tells the low-level filesystem that we are done | |
448 | * performing writes to it. Must be matched with | |
eb04c282 | 449 | * __mnt_want_write() call above. |
8366025e | 450 | */ |
eb04c282 | 451 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 452 | { |
d3ef3d73 | 453 | preempt_disable(); |
83adc753 | 454 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 455 | preempt_enable(); |
8366025e | 456 | } |
eb04c282 JK |
457 | |
458 | /** | |
459 | * mnt_drop_write - give up write access to a mount | |
460 | * @mnt: the mount on which to give up write access | |
461 | * | |
462 | * Tells the low-level filesystem that we are done performing writes to it and | |
463 | * also allows filesystem to be frozen again. Must be matched with | |
464 | * mnt_want_write() call above. | |
465 | */ | |
466 | void mnt_drop_write(struct vfsmount *mnt) | |
467 | { | |
468 | __mnt_drop_write(mnt); | |
469 | sb_end_write(mnt->mnt_sb); | |
470 | } | |
8366025e DH |
471 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
472 | ||
eb04c282 JK |
473 | void __mnt_drop_write_file(struct file *file) |
474 | { | |
475 | __mnt_drop_write(file->f_path.mnt); | |
476 | } | |
477 | ||
2a79f17e AV |
478 | void mnt_drop_write_file(struct file *file) |
479 | { | |
480 | mnt_drop_write(file->f_path.mnt); | |
481 | } | |
482 | EXPORT_SYMBOL(mnt_drop_write_file); | |
483 | ||
83adc753 | 484 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 485 | { |
3d733633 DH |
486 | int ret = 0; |
487 | ||
719ea2fb | 488 | lock_mount_hash(); |
83adc753 | 489 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 490 | /* |
d3ef3d73 | 491 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
492 | * should be visible before we do. | |
3d733633 | 493 | */ |
d3ef3d73 | 494 | smp_mb(); |
495 | ||
3d733633 | 496 | /* |
d3ef3d73 | 497 | * With writers on hold, if this value is zero, then there are |
498 | * definitely no active writers (although held writers may subsequently | |
499 | * increment the count, they'll have to wait, and decrement it after | |
500 | * seeing MNT_READONLY). | |
501 | * | |
502 | * It is OK to have counter incremented on one CPU and decremented on | |
503 | * another: the sum will add up correctly. The danger would be when we | |
504 | * sum up each counter, if we read a counter before it is incremented, | |
505 | * but then read another CPU's count which it has been subsequently | |
506 | * decremented from -- we would see more decrements than we should. | |
507 | * MNT_WRITE_HOLD protects against this scenario, because | |
508 | * mnt_want_write first increments count, then smp_mb, then spins on | |
509 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
510 | * we're counting up here. | |
3d733633 | 511 | */ |
c6653a83 | 512 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 513 | ret = -EBUSY; |
514 | else | |
83adc753 | 515 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 516 | /* |
517 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
518 | * that become unheld will see MNT_READONLY. | |
519 | */ | |
520 | smp_wmb(); | |
83adc753 | 521 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 522 | unlock_mount_hash(); |
3d733633 | 523 | return ret; |
8366025e | 524 | } |
8366025e | 525 | |
83adc753 | 526 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 527 | { |
719ea2fb | 528 | lock_mount_hash(); |
83adc753 | 529 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 530 | unlock_mount_hash(); |
2e4b7fcd DH |
531 | } |
532 | ||
4ed5e82f MS |
533 | int sb_prepare_remount_readonly(struct super_block *sb) |
534 | { | |
535 | struct mount *mnt; | |
536 | int err = 0; | |
537 | ||
8e8b8796 MS |
538 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
539 | if (atomic_long_read(&sb->s_remove_count)) | |
540 | return -EBUSY; | |
541 | ||
719ea2fb | 542 | lock_mount_hash(); |
4ed5e82f MS |
543 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
544 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
545 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
546 | smp_mb(); | |
547 | if (mnt_get_writers(mnt) > 0) { | |
548 | err = -EBUSY; | |
549 | break; | |
550 | } | |
551 | } | |
552 | } | |
8e8b8796 MS |
553 | if (!err && atomic_long_read(&sb->s_remove_count)) |
554 | err = -EBUSY; | |
555 | ||
4ed5e82f MS |
556 | if (!err) { |
557 | sb->s_readonly_remount = 1; | |
558 | smp_wmb(); | |
559 | } | |
560 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
561 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
562 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
563 | } | |
719ea2fb | 564 | unlock_mount_hash(); |
4ed5e82f MS |
565 | |
566 | return err; | |
567 | } | |
568 | ||
b105e270 | 569 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 570 | { |
52ba1621 | 571 | kfree(mnt->mnt_devname); |
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 | ||
8ffcb32e DH |
578 | static void delayed_free_vfsmnt(struct rcu_head *head) |
579 | { | |
580 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
581 | } | |
582 | ||
48a066e7 AV |
583 | /* call under rcu_read_lock */ |
584 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
585 | { | |
586 | struct mount *mnt; | |
587 | if (read_seqretry(&mount_lock, seq)) | |
588 | return false; | |
589 | if (bastard == NULL) | |
590 | return true; | |
591 | mnt = real_mount(bastard); | |
592 | mnt_add_count(mnt, 1); | |
593 | if (likely(!read_seqretry(&mount_lock, seq))) | |
594 | return true; | |
595 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
596 | mnt_add_count(mnt, -1); | |
597 | return false; | |
598 | } | |
599 | rcu_read_unlock(); | |
600 | mntput(bastard); | |
601 | rcu_read_lock(); | |
602 | return false; | |
603 | } | |
604 | ||
1da177e4 | 605 | /* |
474279dc | 606 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 607 | * call under rcu_read_lock() |
1da177e4 | 608 | */ |
474279dc | 609 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 610 | { |
38129a13 | 611 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
612 | struct mount *p; |
613 | ||
38129a13 | 614 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
615 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
616 | return p; | |
617 | return NULL; | |
618 | } | |
619 | ||
620 | /* | |
621 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 622 | * mount_lock must be held. |
474279dc AV |
623 | */ |
624 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
625 | { | |
38129a13 AV |
626 | struct mount *p, *res; |
627 | res = p = __lookup_mnt(mnt, dentry); | |
628 | if (!p) | |
629 | goto out; | |
630 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
631 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
632 | break; | |
633 | res = p; | |
634 | } | |
38129a13 | 635 | out: |
1d6a32ac | 636 | return res; |
1da177e4 LT |
637 | } |
638 | ||
a05964f3 | 639 | /* |
f015f126 DH |
640 | * lookup_mnt - Return the first child mount mounted at path |
641 | * | |
642 | * "First" means first mounted chronologically. If you create the | |
643 | * following mounts: | |
644 | * | |
645 | * mount /dev/sda1 /mnt | |
646 | * mount /dev/sda2 /mnt | |
647 | * mount /dev/sda3 /mnt | |
648 | * | |
649 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
650 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
651 | * /dev/sda2, then /dev/sda3, then NULL. | |
652 | * | |
653 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 654 | */ |
1c755af4 | 655 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 656 | { |
c7105365 | 657 | struct mount *child_mnt; |
48a066e7 AV |
658 | struct vfsmount *m; |
659 | unsigned seq; | |
99b7db7b | 660 | |
48a066e7 AV |
661 | rcu_read_lock(); |
662 | do { | |
663 | seq = read_seqbegin(&mount_lock); | |
664 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
665 | m = child_mnt ? &child_mnt->mnt : NULL; | |
666 | } while (!legitimize_mnt(m, seq)); | |
667 | rcu_read_unlock(); | |
668 | return m; | |
a05964f3 RP |
669 | } |
670 | ||
7af1364f EB |
671 | /* |
672 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
673 | * current mount namespace. | |
674 | * | |
675 | * The common case is dentries are not mountpoints at all and that | |
676 | * test is handled inline. For the slow case when we are actually | |
677 | * dealing with a mountpoint of some kind, walk through all of the | |
678 | * mounts in the current mount namespace and test to see if the dentry | |
679 | * is a mountpoint. | |
680 | * | |
681 | * The mount_hashtable is not usable in the context because we | |
682 | * need to identify all mounts that may be in the current mount | |
683 | * namespace not just a mount that happens to have some specified | |
684 | * parent mount. | |
685 | */ | |
686 | bool __is_local_mountpoint(struct dentry *dentry) | |
687 | { | |
688 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
689 | struct mount *mnt; | |
690 | bool is_covered = false; | |
691 | ||
692 | if (!d_mountpoint(dentry)) | |
693 | goto out; | |
694 | ||
695 | down_read(&namespace_sem); | |
696 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
697 | is_covered = (mnt->mnt_mountpoint == dentry); | |
698 | if (is_covered) | |
699 | break; | |
700 | } | |
701 | up_read(&namespace_sem); | |
702 | out: | |
703 | return is_covered; | |
704 | } | |
705 | ||
e2dfa935 | 706 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 707 | { |
0818bf27 | 708 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
709 | struct mountpoint *mp; |
710 | ||
0818bf27 | 711 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
712 | if (mp->m_dentry == dentry) { |
713 | /* might be worth a WARN_ON() */ | |
714 | if (d_unlinked(dentry)) | |
715 | return ERR_PTR(-ENOENT); | |
716 | mp->m_count++; | |
717 | return mp; | |
718 | } | |
719 | } | |
e2dfa935 EB |
720 | return NULL; |
721 | } | |
722 | ||
723 | static struct mountpoint *new_mountpoint(struct dentry *dentry) | |
724 | { | |
725 | struct hlist_head *chain = mp_hash(dentry); | |
726 | struct mountpoint *mp; | |
727 | int ret; | |
84d17192 AV |
728 | |
729 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
730 | if (!mp) | |
731 | return ERR_PTR(-ENOMEM); | |
732 | ||
eed81007 MS |
733 | ret = d_set_mounted(dentry); |
734 | if (ret) { | |
84d17192 | 735 | kfree(mp); |
eed81007 | 736 | return ERR_PTR(ret); |
84d17192 | 737 | } |
eed81007 | 738 | |
84d17192 AV |
739 | mp->m_dentry = dentry; |
740 | mp->m_count = 1; | |
0818bf27 | 741 | hlist_add_head(&mp->m_hash, chain); |
0a5eb7c8 | 742 | INIT_HLIST_HEAD(&mp->m_list); |
84d17192 AV |
743 | return mp; |
744 | } | |
745 | ||
746 | static void put_mountpoint(struct mountpoint *mp) | |
747 | { | |
748 | if (!--mp->m_count) { | |
749 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 750 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
751 | spin_lock(&dentry->d_lock); |
752 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
753 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 754 | hlist_del(&mp->m_hash); |
84d17192 AV |
755 | kfree(mp); |
756 | } | |
757 | } | |
758 | ||
143c8c91 | 759 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 760 | { |
6b3286ed | 761 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
762 | } |
763 | ||
99b7db7b NP |
764 | /* |
765 | * vfsmount lock must be held for write | |
766 | */ | |
6b3286ed | 767 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
768 | { |
769 | if (ns) { | |
770 | ns->event = ++event; | |
771 | wake_up_interruptible(&ns->poll); | |
772 | } | |
773 | } | |
774 | ||
99b7db7b NP |
775 | /* |
776 | * vfsmount lock must be held for write | |
777 | */ | |
6b3286ed | 778 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
779 | { |
780 | if (ns && ns->event != event) { | |
781 | ns->event = event; | |
782 | wake_up_interruptible(&ns->poll); | |
783 | } | |
784 | } | |
785 | ||
99b7db7b NP |
786 | /* |
787 | * vfsmount lock must be held for write | |
788 | */ | |
419148da AV |
789 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
790 | { | |
a73324da | 791 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
792 | old_path->mnt = &mnt->mnt_parent->mnt; |
793 | mnt->mnt_parent = mnt; | |
a73324da | 794 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 795 | list_del_init(&mnt->mnt_child); |
38129a13 | 796 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 797 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
798 | put_mountpoint(mnt->mnt_mp); |
799 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
800 | } |
801 | ||
99b7db7b NP |
802 | /* |
803 | * vfsmount lock must be held for write | |
804 | */ | |
84d17192 AV |
805 | void mnt_set_mountpoint(struct mount *mnt, |
806 | struct mountpoint *mp, | |
44d964d6 | 807 | struct mount *child_mnt) |
b90fa9ae | 808 | { |
84d17192 | 809 | mp->m_count++; |
3a2393d7 | 810 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 811 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 812 | child_mnt->mnt_parent = mnt; |
84d17192 | 813 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 814 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
815 | } |
816 | ||
99b7db7b NP |
817 | /* |
818 | * vfsmount lock must be held for write | |
819 | */ | |
84d17192 AV |
820 | static void attach_mnt(struct mount *mnt, |
821 | struct mount *parent, | |
822 | struct mountpoint *mp) | |
1da177e4 | 823 | { |
84d17192 | 824 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 825 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 826 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
827 | } |
828 | ||
12a5b529 AV |
829 | static void attach_shadowed(struct mount *mnt, |
830 | struct mount *parent, | |
831 | struct mount *shadows) | |
832 | { | |
833 | if (shadows) { | |
f6f99332 | 834 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
835 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
836 | } else { | |
837 | hlist_add_head_rcu(&mnt->mnt_hash, | |
838 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
839 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
840 | } | |
841 | } | |
842 | ||
b90fa9ae | 843 | /* |
99b7db7b | 844 | * vfsmount lock must be held for write |
b90fa9ae | 845 | */ |
1d6a32ac | 846 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 847 | { |
0714a533 | 848 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 849 | struct mount *m; |
b90fa9ae | 850 | LIST_HEAD(head); |
143c8c91 | 851 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 852 | |
0714a533 | 853 | BUG_ON(parent == mnt); |
b90fa9ae | 854 | |
1a4eeaf2 | 855 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 856 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 857 | m->mnt_ns = n; |
f03c6599 | 858 | |
b90fa9ae RP |
859 | list_splice(&head, n->list.prev); |
860 | ||
12a5b529 | 861 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 862 | touch_mnt_namespace(n); |
1da177e4 LT |
863 | } |
864 | ||
909b0a88 | 865 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 866 | { |
6b41d536 AV |
867 | struct list_head *next = p->mnt_mounts.next; |
868 | if (next == &p->mnt_mounts) { | |
1da177e4 | 869 | while (1) { |
909b0a88 | 870 | if (p == root) |
1da177e4 | 871 | return NULL; |
6b41d536 AV |
872 | next = p->mnt_child.next; |
873 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 874 | break; |
0714a533 | 875 | p = p->mnt_parent; |
1da177e4 LT |
876 | } |
877 | } | |
6b41d536 | 878 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
879 | } |
880 | ||
315fc83e | 881 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 882 | { |
6b41d536 AV |
883 | struct list_head *prev = p->mnt_mounts.prev; |
884 | while (prev != &p->mnt_mounts) { | |
885 | p = list_entry(prev, struct mount, mnt_child); | |
886 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
887 | } |
888 | return p; | |
889 | } | |
890 | ||
9d412a43 AV |
891 | struct vfsmount * |
892 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
893 | { | |
b105e270 | 894 | struct mount *mnt; |
9d412a43 AV |
895 | struct dentry *root; |
896 | ||
897 | if (!type) | |
898 | return ERR_PTR(-ENODEV); | |
899 | ||
900 | mnt = alloc_vfsmnt(name); | |
901 | if (!mnt) | |
902 | return ERR_PTR(-ENOMEM); | |
903 | ||
904 | if (flags & MS_KERNMOUNT) | |
b105e270 | 905 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
906 | |
907 | root = mount_fs(type, flags, name, data); | |
908 | if (IS_ERR(root)) { | |
8ffcb32e | 909 | mnt_free_id(mnt); |
9d412a43 AV |
910 | free_vfsmnt(mnt); |
911 | return ERR_CAST(root); | |
912 | } | |
913 | ||
b105e270 AV |
914 | mnt->mnt.mnt_root = root; |
915 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 916 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 917 | mnt->mnt_parent = mnt; |
719ea2fb | 918 | lock_mount_hash(); |
39f7c4db | 919 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 920 | unlock_mount_hash(); |
b105e270 | 921 | return &mnt->mnt; |
9d412a43 AV |
922 | } |
923 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
924 | ||
87129cc0 | 925 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 926 | int flag) |
1da177e4 | 927 | { |
87129cc0 | 928 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
929 | struct mount *mnt; |
930 | int err; | |
1da177e4 | 931 | |
be34d1a3 DH |
932 | mnt = alloc_vfsmnt(old->mnt_devname); |
933 | if (!mnt) | |
934 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 935 | |
7a472ef4 | 936 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
937 | mnt->mnt_group_id = 0; /* not a peer of original */ |
938 | else | |
939 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 940 | |
be34d1a3 DH |
941 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
942 | err = mnt_alloc_group_id(mnt); | |
943 | if (err) | |
944 | goto out_free; | |
1da177e4 | 945 | } |
be34d1a3 | 946 | |
f2ebb3a9 | 947 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 948 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
949 | if (flag & CL_UNPRIVILEGED) { |
950 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
951 | ||
952 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
953 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
954 | ||
955 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
956 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
957 | ||
958 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
959 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
960 | ||
961 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
962 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
963 | } | |
132c94e3 | 964 | |
5ff9d8a6 | 965 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
966 | if ((flag & CL_UNPRIVILEGED) && |
967 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
968 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
969 | ||
be34d1a3 DH |
970 | atomic_inc(&sb->s_active); |
971 | mnt->mnt.mnt_sb = sb; | |
972 | mnt->mnt.mnt_root = dget(root); | |
973 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
974 | mnt->mnt_parent = mnt; | |
719ea2fb | 975 | lock_mount_hash(); |
be34d1a3 | 976 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 977 | unlock_mount_hash(); |
be34d1a3 | 978 | |
7a472ef4 EB |
979 | if ((flag & CL_SLAVE) || |
980 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
981 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
982 | mnt->mnt_master = old; | |
983 | CLEAR_MNT_SHARED(mnt); | |
984 | } else if (!(flag & CL_PRIVATE)) { | |
985 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
986 | list_add(&mnt->mnt_share, &old->mnt_share); | |
987 | if (IS_MNT_SLAVE(old)) | |
988 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
989 | mnt->mnt_master = old->mnt_master; | |
990 | } | |
991 | if (flag & CL_MAKE_SHARED) | |
992 | set_mnt_shared(mnt); | |
993 | ||
994 | /* stick the duplicate mount on the same expiry list | |
995 | * as the original if that was on one */ | |
996 | if (flag & CL_EXPIRE) { | |
997 | if (!list_empty(&old->mnt_expire)) | |
998 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
999 | } | |
1000 | ||
cb338d06 | 1001 | return mnt; |
719f5d7f MS |
1002 | |
1003 | out_free: | |
8ffcb32e | 1004 | mnt_free_id(mnt); |
719f5d7f | 1005 | free_vfsmnt(mnt); |
be34d1a3 | 1006 | return ERR_PTR(err); |
1da177e4 LT |
1007 | } |
1008 | ||
9ea459e1 AV |
1009 | static void cleanup_mnt(struct mount *mnt) |
1010 | { | |
1011 | /* | |
1012 | * This probably indicates that somebody messed | |
1013 | * up a mnt_want/drop_write() pair. If this | |
1014 | * happens, the filesystem was probably unable | |
1015 | * to make r/w->r/o transitions. | |
1016 | */ | |
1017 | /* | |
1018 | * The locking used to deal with mnt_count decrement provides barriers, | |
1019 | * so mnt_get_writers() below is safe. | |
1020 | */ | |
1021 | WARN_ON(mnt_get_writers(mnt)); | |
1022 | if (unlikely(mnt->mnt_pins.first)) | |
1023 | mnt_pin_kill(mnt); | |
1024 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1025 | dput(mnt->mnt.mnt_root); | |
1026 | deactivate_super(mnt->mnt.mnt_sb); | |
1027 | mnt_free_id(mnt); | |
1028 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1029 | } | |
1030 | ||
1031 | static void __cleanup_mnt(struct rcu_head *head) | |
1032 | { | |
1033 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1034 | } | |
1035 | ||
1036 | static LLIST_HEAD(delayed_mntput_list); | |
1037 | static void delayed_mntput(struct work_struct *unused) | |
1038 | { | |
1039 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1040 | struct llist_node *next; | |
1041 | ||
1042 | for (; node; node = next) { | |
1043 | next = llist_next(node); | |
1044 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1045 | } | |
1046 | } | |
1047 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1048 | ||
900148dc | 1049 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1050 | { |
48a066e7 AV |
1051 | rcu_read_lock(); |
1052 | mnt_add_count(mnt, -1); | |
1053 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1054 | rcu_read_unlock(); | |
f03c6599 | 1055 | return; |
b3e19d92 | 1056 | } |
719ea2fb | 1057 | lock_mount_hash(); |
b3e19d92 | 1058 | if (mnt_get_count(mnt)) { |
48a066e7 | 1059 | rcu_read_unlock(); |
719ea2fb | 1060 | unlock_mount_hash(); |
99b7db7b NP |
1061 | return; |
1062 | } | |
48a066e7 AV |
1063 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1064 | rcu_read_unlock(); | |
1065 | unlock_mount_hash(); | |
1066 | return; | |
1067 | } | |
1068 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1069 | rcu_read_unlock(); | |
962830df | 1070 | |
39f7c4db | 1071 | list_del(&mnt->mnt_instance); |
719ea2fb | 1072 | unlock_mount_hash(); |
649a795a | 1073 | |
9ea459e1 AV |
1074 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1075 | struct task_struct *task = current; | |
1076 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1077 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1078 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1079 | return; | |
1080 | } | |
1081 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1082 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1083 | return; | |
1084 | } | |
1085 | cleanup_mnt(mnt); | |
b3e19d92 | 1086 | } |
b3e19d92 NP |
1087 | |
1088 | void mntput(struct vfsmount *mnt) | |
1089 | { | |
1090 | if (mnt) { | |
863d684f | 1091 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1092 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1093 | if (unlikely(m->mnt_expiry_mark)) |
1094 | m->mnt_expiry_mark = 0; | |
1095 | mntput_no_expire(m); | |
b3e19d92 NP |
1096 | } |
1097 | } | |
1098 | EXPORT_SYMBOL(mntput); | |
1099 | ||
1100 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1101 | { | |
1102 | if (mnt) | |
83adc753 | 1103 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1104 | return mnt; |
1105 | } | |
1106 | EXPORT_SYMBOL(mntget); | |
1107 | ||
3064c356 | 1108 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1109 | { |
3064c356 AV |
1110 | struct mount *p; |
1111 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1112 | if (IS_ERR(p)) | |
1113 | return ERR_CAST(p); | |
1114 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1115 | return &p->mnt; | |
7b7b1ace | 1116 | } |
1da177e4 | 1117 | |
b3b304a2 MS |
1118 | static inline void mangle(struct seq_file *m, const char *s) |
1119 | { | |
1120 | seq_escape(m, s, " \t\n\\"); | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * Simple .show_options callback for filesystems which don't want to | |
1125 | * implement more complex mount option showing. | |
1126 | * | |
1127 | * See also save_mount_options(). | |
1128 | */ | |
34c80b1d | 1129 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1130 | { |
2a32cebd AV |
1131 | const char *options; |
1132 | ||
1133 | rcu_read_lock(); | |
34c80b1d | 1134 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1135 | |
1136 | if (options != NULL && options[0]) { | |
1137 | seq_putc(m, ','); | |
1138 | mangle(m, options); | |
1139 | } | |
2a32cebd | 1140 | rcu_read_unlock(); |
b3b304a2 MS |
1141 | |
1142 | return 0; | |
1143 | } | |
1144 | EXPORT_SYMBOL(generic_show_options); | |
1145 | ||
1146 | /* | |
1147 | * If filesystem uses generic_show_options(), this function should be | |
1148 | * called from the fill_super() callback. | |
1149 | * | |
1150 | * The .remount_fs callback usually needs to be handled in a special | |
1151 | * way, to make sure, that previous options are not overwritten if the | |
1152 | * remount fails. | |
1153 | * | |
1154 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1155 | * reset all options to their default value, but changes only newly | |
1156 | * given options, then the displayed options will not reflect reality | |
1157 | * any more. | |
1158 | */ | |
1159 | void save_mount_options(struct super_block *sb, char *options) | |
1160 | { | |
2a32cebd AV |
1161 | BUG_ON(sb->s_options); |
1162 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1163 | } |
1164 | EXPORT_SYMBOL(save_mount_options); | |
1165 | ||
2a32cebd AV |
1166 | void replace_mount_options(struct super_block *sb, char *options) |
1167 | { | |
1168 | char *old = sb->s_options; | |
1169 | rcu_assign_pointer(sb->s_options, options); | |
1170 | if (old) { | |
1171 | synchronize_rcu(); | |
1172 | kfree(old); | |
1173 | } | |
1174 | } | |
1175 | EXPORT_SYMBOL(replace_mount_options); | |
1176 | ||
a1a2c409 | 1177 | #ifdef CONFIG_PROC_FS |
0226f492 | 1178 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1179 | static void *m_start(struct seq_file *m, loff_t *pos) |
1180 | { | |
6ce6e24e | 1181 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1182 | |
390c6843 | 1183 | down_read(&namespace_sem); |
c7999c36 AV |
1184 | if (p->cached_event == p->ns->event) { |
1185 | void *v = p->cached_mount; | |
1186 | if (*pos == p->cached_index) | |
1187 | return v; | |
1188 | if (*pos == p->cached_index + 1) { | |
1189 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1190 | return p->cached_mount = v; | |
1191 | } | |
1192 | } | |
1193 | ||
1194 | p->cached_event = p->ns->event; | |
1195 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1196 | p->cached_index = *pos; | |
1197 | return p->cached_mount; | |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1201 | { | |
6ce6e24e | 1202 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1203 | |
c7999c36 AV |
1204 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1205 | p->cached_index = *pos; | |
1206 | return p->cached_mount; | |
1da177e4 LT |
1207 | } |
1208 | ||
1209 | static void m_stop(struct seq_file *m, void *v) | |
1210 | { | |
390c6843 | 1211 | up_read(&namespace_sem); |
1da177e4 LT |
1212 | } |
1213 | ||
0226f492 | 1214 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1215 | { |
6ce6e24e | 1216 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1217 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1218 | return p->show(m, &r->mnt); |
1da177e4 LT |
1219 | } |
1220 | ||
a1a2c409 | 1221 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1222 | .start = m_start, |
1223 | .next = m_next, | |
1224 | .stop = m_stop, | |
0226f492 | 1225 | .show = m_show, |
b4629fe2 | 1226 | }; |
a1a2c409 | 1227 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1228 | |
1da177e4 LT |
1229 | /** |
1230 | * may_umount_tree - check if a mount tree is busy | |
1231 | * @mnt: root of mount tree | |
1232 | * | |
1233 | * This is called to check if a tree of mounts has any | |
1234 | * open files, pwds, chroots or sub mounts that are | |
1235 | * busy. | |
1236 | */ | |
909b0a88 | 1237 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1238 | { |
909b0a88 | 1239 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1240 | int actual_refs = 0; |
1241 | int minimum_refs = 0; | |
315fc83e | 1242 | struct mount *p; |
909b0a88 | 1243 | BUG_ON(!m); |
1da177e4 | 1244 | |
b3e19d92 | 1245 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1246 | lock_mount_hash(); |
909b0a88 | 1247 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1248 | actual_refs += mnt_get_count(p); |
1da177e4 | 1249 | minimum_refs += 2; |
1da177e4 | 1250 | } |
719ea2fb | 1251 | unlock_mount_hash(); |
1da177e4 LT |
1252 | |
1253 | if (actual_refs > minimum_refs) | |
e3474a8e | 1254 | return 0; |
1da177e4 | 1255 | |
e3474a8e | 1256 | return 1; |
1da177e4 LT |
1257 | } |
1258 | ||
1259 | EXPORT_SYMBOL(may_umount_tree); | |
1260 | ||
1261 | /** | |
1262 | * may_umount - check if a mount point is busy | |
1263 | * @mnt: root of mount | |
1264 | * | |
1265 | * This is called to check if a mount point has any | |
1266 | * open files, pwds, chroots or sub mounts. If the | |
1267 | * mount has sub mounts this will return busy | |
1268 | * regardless of whether the sub mounts are busy. | |
1269 | * | |
1270 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1271 | * give false negatives. The main reason why it's here is that we need | |
1272 | * a non-destructive way to look for easily umountable filesystems. | |
1273 | */ | |
1274 | int may_umount(struct vfsmount *mnt) | |
1275 | { | |
e3474a8e | 1276 | int ret = 1; |
8ad08d8a | 1277 | down_read(&namespace_sem); |
719ea2fb | 1278 | lock_mount_hash(); |
1ab59738 | 1279 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1280 | ret = 0; |
719ea2fb | 1281 | unlock_mount_hash(); |
8ad08d8a | 1282 | up_read(&namespace_sem); |
a05964f3 | 1283 | return ret; |
1da177e4 LT |
1284 | } |
1285 | ||
1286 | EXPORT_SYMBOL(may_umount); | |
1287 | ||
38129a13 | 1288 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1289 | |
97216be0 | 1290 | static void namespace_unlock(void) |
70fbcdf4 | 1291 | { |
d5e50f74 | 1292 | struct mount *mnt; |
38129a13 | 1293 | struct hlist_head head = unmounted; |
97216be0 | 1294 | |
38129a13 | 1295 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1296 | up_write(&namespace_sem); |
1297 | return; | |
1298 | } | |
1299 | ||
38129a13 AV |
1300 | head.first->pprev = &head.first; |
1301 | INIT_HLIST_HEAD(&unmounted); | |
1302 | ||
81b6b061 AV |
1303 | /* undo decrements we'd done in umount_tree() */ |
1304 | hlist_for_each_entry(mnt, &head, mnt_hash) | |
1305 | if (mnt->mnt_ex_mountpoint.mnt) | |
1306 | mntget(mnt->mnt_ex_mountpoint.mnt); | |
1307 | ||
97216be0 AV |
1308 | up_write(&namespace_sem); |
1309 | ||
48a066e7 AV |
1310 | synchronize_rcu(); |
1311 | ||
38129a13 AV |
1312 | while (!hlist_empty(&head)) { |
1313 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1314 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1315 | if (mnt->mnt_ex_mountpoint.mnt) |
1316 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1317 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1318 | } |
1319 | } | |
1320 | ||
97216be0 | 1321 | static inline void namespace_lock(void) |
e3197d83 | 1322 | { |
97216be0 | 1323 | down_write(&namespace_sem); |
e3197d83 AV |
1324 | } |
1325 | ||
99b7db7b | 1326 | /* |
48a066e7 | 1327 | * mount_lock must be held |
99b7db7b | 1328 | * namespace_sem must be held for write |
48a066e7 AV |
1329 | * how = 0 => just this tree, don't propagate |
1330 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1331 | * how = 2 => lazy umount | |
99b7db7b | 1332 | */ |
48a066e7 | 1333 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1334 | { |
38129a13 | 1335 | HLIST_HEAD(tmp_list); |
315fc83e | 1336 | struct mount *p; |
38129a13 | 1337 | struct mount *last = NULL; |
1da177e4 | 1338 | |
38129a13 AV |
1339 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1340 | hlist_del_init_rcu(&p->mnt_hash); | |
1341 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1342 | } | |
1da177e4 | 1343 | |
88b368f2 AV |
1344 | hlist_for_each_entry(p, &tmp_list, mnt_hash) |
1345 | list_del_init(&p->mnt_child); | |
1346 | ||
48a066e7 | 1347 | if (how) |
7b8a53fd | 1348 | propagate_umount(&tmp_list); |
a05964f3 | 1349 | |
38129a13 | 1350 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1351 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1352 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1353 | __touch_mnt_namespace(p->mnt_ns); |
1354 | p->mnt_ns = NULL; | |
48a066e7 AV |
1355 | if (how < 2) |
1356 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
676da58d | 1357 | if (mnt_has_parent(p)) { |
0a5eb7c8 | 1358 | hlist_del_init(&p->mnt_mp_list); |
84d17192 | 1359 | put_mountpoint(p->mnt_mp); |
81b6b061 | 1360 | mnt_add_count(p->mnt_parent, -1); |
aba809cf AV |
1361 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1362 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1363 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1364 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1365 | p->mnt_parent = p; | |
84d17192 | 1366 | p->mnt_mp = NULL; |
7c4b93d8 | 1367 | } |
0f0afb1d | 1368 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1369 | last = p; |
1370 | } | |
1371 | if (last) { | |
1372 | last->mnt_hash.next = unmounted.first; | |
c297abfd EB |
1373 | if (unmounted.first) |
1374 | unmounted.first->pprev = &last->mnt_hash.next; | |
38129a13 AV |
1375 | unmounted.first = tmp_list.first; |
1376 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1377 | } |
1378 | } | |
1379 | ||
b54b9be7 | 1380 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1381 | |
1ab59738 | 1382 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1383 | { |
1ab59738 | 1384 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1385 | int retval; |
1386 | ||
1ab59738 | 1387 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1388 | if (retval) |
1389 | return retval; | |
1390 | ||
1391 | /* | |
1392 | * Allow userspace to request a mountpoint be expired rather than | |
1393 | * unmounting unconditionally. Unmount only happens if: | |
1394 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1395 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1396 | */ | |
1397 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1398 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1399 | flags & (MNT_FORCE | MNT_DETACH)) |
1400 | return -EINVAL; | |
1401 | ||
b3e19d92 NP |
1402 | /* |
1403 | * probably don't strictly need the lock here if we examined | |
1404 | * all race cases, but it's a slowpath. | |
1405 | */ | |
719ea2fb | 1406 | lock_mount_hash(); |
83adc753 | 1407 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1408 | unlock_mount_hash(); |
1da177e4 | 1409 | return -EBUSY; |
b3e19d92 | 1410 | } |
719ea2fb | 1411 | unlock_mount_hash(); |
1da177e4 | 1412 | |
863d684f | 1413 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1414 | return -EAGAIN; |
1415 | } | |
1416 | ||
1417 | /* | |
1418 | * If we may have to abort operations to get out of this | |
1419 | * mount, and they will themselves hold resources we must | |
1420 | * allow the fs to do things. In the Unix tradition of | |
1421 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1422 | * might fail to complete on the first run through as other tasks | |
1423 | * must return, and the like. Thats for the mount program to worry | |
1424 | * about for the moment. | |
1425 | */ | |
1426 | ||
42faad99 | 1427 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1428 | sb->s_op->umount_begin(sb); |
42faad99 | 1429 | } |
1da177e4 LT |
1430 | |
1431 | /* | |
1432 | * No sense to grab the lock for this test, but test itself looks | |
1433 | * somewhat bogus. Suggestions for better replacement? | |
1434 | * Ho-hum... In principle, we might treat that as umount + switch | |
1435 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1436 | * Actually it makes sense, especially if rootfs would contain a | |
1437 | * /reboot - static binary that would close all descriptors and | |
1438 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1439 | */ | |
1ab59738 | 1440 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1441 | /* |
1442 | * Special case for "unmounting" root ... | |
1443 | * we just try to remount it readonly. | |
1444 | */ | |
a1480dcc AL |
1445 | if (!capable(CAP_SYS_ADMIN)) |
1446 | return -EPERM; | |
1da177e4 | 1447 | down_write(&sb->s_umount); |
4aa98cf7 | 1448 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1449 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1450 | up_write(&sb->s_umount); |
1451 | return retval; | |
1452 | } | |
1453 | ||
97216be0 | 1454 | namespace_lock(); |
719ea2fb | 1455 | lock_mount_hash(); |
5addc5dd | 1456 | event++; |
1da177e4 | 1457 | |
48a066e7 | 1458 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1459 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1460 | umount_tree(mnt, 2); |
1da177e4 | 1461 | retval = 0; |
48a066e7 AV |
1462 | } else { |
1463 | shrink_submounts(mnt); | |
1464 | retval = -EBUSY; | |
1465 | if (!propagate_mount_busy(mnt, 2)) { | |
1466 | if (!list_empty(&mnt->mnt_list)) | |
1467 | umount_tree(mnt, 1); | |
1468 | retval = 0; | |
1469 | } | |
1da177e4 | 1470 | } |
719ea2fb | 1471 | unlock_mount_hash(); |
e3197d83 | 1472 | namespace_unlock(); |
1da177e4 LT |
1473 | return retval; |
1474 | } | |
1475 | ||
80b5dce8 EB |
1476 | /* |
1477 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1478 | * | |
1479 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1480 | * to an existing mountpoint, and wind up leaking the mount. | |
1481 | * detach_mounts allows lazily unmounting those mounts instead of | |
1482 | * leaking them. | |
1483 | * | |
1484 | * The caller may hold dentry->d_inode->i_mutex. | |
1485 | */ | |
1486 | void __detach_mounts(struct dentry *dentry) | |
1487 | { | |
1488 | struct mountpoint *mp; | |
1489 | struct mount *mnt; | |
1490 | ||
1491 | namespace_lock(); | |
1492 | mp = lookup_mountpoint(dentry); | |
1493 | if (!mp) | |
1494 | goto out_unlock; | |
1495 | ||
1496 | lock_mount_hash(); | |
1497 | while (!hlist_empty(&mp->m_list)) { | |
1498 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
1499 | umount_tree(mnt, 2); | |
1500 | } | |
1501 | unlock_mount_hash(); | |
1502 | put_mountpoint(mp); | |
1503 | out_unlock: | |
1504 | namespace_unlock(); | |
1505 | } | |
1506 | ||
9b40bc90 AV |
1507 | /* |
1508 | * Is the caller allowed to modify his namespace? | |
1509 | */ | |
1510 | static inline bool may_mount(void) | |
1511 | { | |
1512 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1513 | } | |
1514 | ||
1da177e4 LT |
1515 | /* |
1516 | * Now umount can handle mount points as well as block devices. | |
1517 | * This is important for filesystems which use unnamed block devices. | |
1518 | * | |
1519 | * We now support a flag for forced unmount like the other 'big iron' | |
1520 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1521 | */ | |
1522 | ||
bdc480e3 | 1523 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1524 | { |
2d8f3038 | 1525 | struct path path; |
900148dc | 1526 | struct mount *mnt; |
1da177e4 | 1527 | int retval; |
db1f05bb | 1528 | int lookup_flags = 0; |
1da177e4 | 1529 | |
db1f05bb MS |
1530 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1531 | return -EINVAL; | |
1532 | ||
9b40bc90 AV |
1533 | if (!may_mount()) |
1534 | return -EPERM; | |
1535 | ||
db1f05bb MS |
1536 | if (!(flags & UMOUNT_NOFOLLOW)) |
1537 | lookup_flags |= LOOKUP_FOLLOW; | |
1538 | ||
197df04c | 1539 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1540 | if (retval) |
1541 | goto out; | |
900148dc | 1542 | mnt = real_mount(path.mnt); |
1da177e4 | 1543 | retval = -EINVAL; |
2d8f3038 | 1544 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1545 | goto dput_and_out; |
143c8c91 | 1546 | if (!check_mnt(mnt)) |
1da177e4 | 1547 | goto dput_and_out; |
5ff9d8a6 EB |
1548 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1549 | goto dput_and_out; | |
b2f5d4dc EB |
1550 | retval = -EPERM; |
1551 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1552 | goto dput_and_out; | |
1da177e4 | 1553 | |
900148dc | 1554 | retval = do_umount(mnt, flags); |
1da177e4 | 1555 | dput_and_out: |
429731b1 | 1556 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1557 | dput(path.dentry); |
900148dc | 1558 | mntput_no_expire(mnt); |
1da177e4 LT |
1559 | out: |
1560 | return retval; | |
1561 | } | |
1562 | ||
1563 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1564 | ||
1565 | /* | |
b58fed8b | 1566 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1567 | */ |
bdc480e3 | 1568 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1569 | { |
b58fed8b | 1570 | return sys_umount(name, 0); |
1da177e4 LT |
1571 | } |
1572 | ||
1573 | #endif | |
1574 | ||
4ce5d2b1 | 1575 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1576 | { |
4ce5d2b1 | 1577 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1578 | return dentry->d_op == &ns_dentry_operations && |
1579 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1580 | } |
1581 | ||
58be2825 AV |
1582 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1583 | { | |
1584 | return container_of(ns, struct mnt_namespace, ns); | |
1585 | } | |
1586 | ||
4ce5d2b1 EB |
1587 | static bool mnt_ns_loop(struct dentry *dentry) |
1588 | { | |
1589 | /* Could bind mounting the mount namespace inode cause a | |
1590 | * mount namespace loop? | |
1591 | */ | |
1592 | struct mnt_namespace *mnt_ns; | |
1593 | if (!is_mnt_ns_file(dentry)) | |
1594 | return false; | |
1595 | ||
f77c8014 | 1596 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1597 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1598 | } | |
1599 | ||
87129cc0 | 1600 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1601 | int flag) |
1da177e4 | 1602 | { |
84d17192 | 1603 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1604 | |
4ce5d2b1 EB |
1605 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1606 | return ERR_PTR(-EINVAL); | |
1607 | ||
1608 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1609 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1610 | |
36341f64 | 1611 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1612 | if (IS_ERR(q)) |
1613 | return q; | |
1614 | ||
a73324da | 1615 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1616 | |
1617 | p = mnt; | |
6b41d536 | 1618 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1619 | struct mount *s; |
7ec02ef1 | 1620 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1621 | continue; |
1622 | ||
909b0a88 | 1623 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1624 | struct mount *t = NULL; |
4ce5d2b1 EB |
1625 | if (!(flag & CL_COPY_UNBINDABLE) && |
1626 | IS_MNT_UNBINDABLE(s)) { | |
1627 | s = skip_mnt_tree(s); | |
1628 | continue; | |
1629 | } | |
1630 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1631 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1632 | s = skip_mnt_tree(s); |
1633 | continue; | |
1634 | } | |
0714a533 AV |
1635 | while (p != s->mnt_parent) { |
1636 | p = p->mnt_parent; | |
1637 | q = q->mnt_parent; | |
1da177e4 | 1638 | } |
87129cc0 | 1639 | p = s; |
84d17192 | 1640 | parent = q; |
87129cc0 | 1641 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1642 | if (IS_ERR(q)) |
1643 | goto out; | |
719ea2fb | 1644 | lock_mount_hash(); |
1a4eeaf2 | 1645 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1646 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1647 | if (!list_empty(&parent->mnt_mounts)) { | |
1648 | t = list_last_entry(&parent->mnt_mounts, | |
1649 | struct mount, mnt_child); | |
1650 | if (t->mnt_mp != p->mnt_mp) | |
1651 | t = NULL; | |
1652 | } | |
1653 | attach_shadowed(q, parent, t); | |
719ea2fb | 1654 | unlock_mount_hash(); |
1da177e4 LT |
1655 | } |
1656 | } | |
1657 | return res; | |
be34d1a3 | 1658 | out: |
1da177e4 | 1659 | if (res) { |
719ea2fb | 1660 | lock_mount_hash(); |
328e6d90 | 1661 | umount_tree(res, 0); |
719ea2fb | 1662 | unlock_mount_hash(); |
1da177e4 | 1663 | } |
be34d1a3 | 1664 | return q; |
1da177e4 LT |
1665 | } |
1666 | ||
be34d1a3 DH |
1667 | /* Caller should check returned pointer for errors */ |
1668 | ||
589ff870 | 1669 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1670 | { |
cb338d06 | 1671 | struct mount *tree; |
97216be0 | 1672 | namespace_lock(); |
87129cc0 AV |
1673 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1674 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1675 | namespace_unlock(); |
be34d1a3 | 1676 | if (IS_ERR(tree)) |
52e220d3 | 1677 | return ERR_CAST(tree); |
be34d1a3 | 1678 | return &tree->mnt; |
8aec0809 AV |
1679 | } |
1680 | ||
1681 | void drop_collected_mounts(struct vfsmount *mnt) | |
1682 | { | |
97216be0 | 1683 | namespace_lock(); |
719ea2fb | 1684 | lock_mount_hash(); |
328e6d90 | 1685 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1686 | unlock_mount_hash(); |
3ab6abee | 1687 | namespace_unlock(); |
8aec0809 AV |
1688 | } |
1689 | ||
c771d683 MS |
1690 | /** |
1691 | * clone_private_mount - create a private clone of a path | |
1692 | * | |
1693 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1694 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1695 | * to the originating mount won't be propagated into this). | |
1696 | * | |
1697 | * Release with mntput(). | |
1698 | */ | |
1699 | struct vfsmount *clone_private_mount(struct path *path) | |
1700 | { | |
1701 | struct mount *old_mnt = real_mount(path->mnt); | |
1702 | struct mount *new_mnt; | |
1703 | ||
1704 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1705 | return ERR_PTR(-EINVAL); | |
1706 | ||
1707 | down_read(&namespace_sem); | |
1708 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); | |
1709 | up_read(&namespace_sem); | |
1710 | if (IS_ERR(new_mnt)) | |
1711 | return ERR_CAST(new_mnt); | |
1712 | ||
1713 | return &new_mnt->mnt; | |
1714 | } | |
1715 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1716 | ||
1f707137 AV |
1717 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1718 | struct vfsmount *root) | |
1719 | { | |
1a4eeaf2 | 1720 | struct mount *mnt; |
1f707137 AV |
1721 | int res = f(root, arg); |
1722 | if (res) | |
1723 | return res; | |
1a4eeaf2 AV |
1724 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1725 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1726 | if (res) |
1727 | return res; | |
1728 | } | |
1729 | return 0; | |
1730 | } | |
1731 | ||
4b8b21f4 | 1732 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1733 | { |
315fc83e | 1734 | struct mount *p; |
719f5d7f | 1735 | |
909b0a88 | 1736 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1737 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1738 | mnt_release_group_id(p); |
719f5d7f MS |
1739 | } |
1740 | } | |
1741 | ||
4b8b21f4 | 1742 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1743 | { |
315fc83e | 1744 | struct mount *p; |
719f5d7f | 1745 | |
909b0a88 | 1746 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1747 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1748 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1749 | if (err) { |
4b8b21f4 | 1750 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1751 | return err; |
1752 | } | |
1753 | } | |
1754 | } | |
1755 | ||
1756 | return 0; | |
1757 | } | |
1758 | ||
b90fa9ae RP |
1759 | /* |
1760 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1761 | * @nd : place the mount tree @source_mnt is attached |
1762 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1763 | * store the parent mount and mountpoint dentry. | |
1764 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1765 | * |
1766 | * NOTE: in the table below explains the semantics when a source mount | |
1767 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1768 | * --------------------------------------------------------------------------- |
1769 | * | BIND MOUNT OPERATION | | |
1770 | * |************************************************************************** | |
1771 | * | source-->| shared | private | slave | unbindable | | |
1772 | * | dest | | | | | | |
1773 | * | | | | | | | | |
1774 | * | v | | | | | | |
1775 | * |************************************************************************** | |
1776 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1777 | * | | | | | | | |
1778 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1779 | * *************************************************************************** | |
b90fa9ae RP |
1780 | * A bind operation clones the source mount and mounts the clone on the |
1781 | * destination mount. | |
1782 | * | |
1783 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1784 | * tree of the destination mount and the cloned mount is added to | |
1785 | * the peer group of the source mount. | |
1786 | * (+) the cloned mount is created under the destination mount and is marked | |
1787 | * as shared. The cloned mount is added to the peer group of the source | |
1788 | * mount. | |
5afe0022 RP |
1789 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1790 | * of the destination mount and the cloned mount is made slave | |
1791 | * of the same master as that of the source mount. The cloned mount | |
1792 | * is marked as 'shared and slave'. | |
1793 | * (*) the cloned mount is made a slave of the same master as that of the | |
1794 | * source mount. | |
1795 | * | |
9676f0c6 RP |
1796 | * --------------------------------------------------------------------------- |
1797 | * | MOVE MOUNT OPERATION | | |
1798 | * |************************************************************************** | |
1799 | * | source-->| shared | private | slave | unbindable | | |
1800 | * | dest | | | | | | |
1801 | * | | | | | | | | |
1802 | * | v | | | | | | |
1803 | * |************************************************************************** | |
1804 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1805 | * | | | | | | | |
1806 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1807 | * *************************************************************************** | |
5afe0022 RP |
1808 | * |
1809 | * (+) the mount is moved to the destination. And is then propagated to | |
1810 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1811 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1812 | * (+++) the mount is moved to the destination and is then propagated to |
1813 | * all the mounts belonging to the destination mount's propagation tree. | |
1814 | * the mount is marked as 'shared and slave'. | |
1815 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1816 | * |
1817 | * if the source mount is a tree, the operations explained above is | |
1818 | * applied to each mount in the tree. | |
1819 | * Must be called without spinlocks held, since this function can sleep | |
1820 | * in allocations. | |
1821 | */ | |
0fb54e50 | 1822 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1823 | struct mount *dest_mnt, |
1824 | struct mountpoint *dest_mp, | |
1825 | struct path *parent_path) | |
b90fa9ae | 1826 | { |
38129a13 | 1827 | HLIST_HEAD(tree_list); |
315fc83e | 1828 | struct mount *child, *p; |
38129a13 | 1829 | struct hlist_node *n; |
719f5d7f | 1830 | int err; |
b90fa9ae | 1831 | |
fc7be130 | 1832 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1833 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1834 | if (err) |
1835 | goto out; | |
0b1b901b | 1836 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1837 | lock_mount_hash(); |
0b1b901b AV |
1838 | if (err) |
1839 | goto out_cleanup_ids; | |
909b0a88 | 1840 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1841 | set_mnt_shared(p); |
0b1b901b AV |
1842 | } else { |
1843 | lock_mount_hash(); | |
b90fa9ae | 1844 | } |
1a390689 | 1845 | if (parent_path) { |
0fb54e50 | 1846 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1847 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1848 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1849 | } else { |
84d17192 | 1850 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1851 | commit_tree(source_mnt, NULL); |
21444403 | 1852 | } |
b90fa9ae | 1853 | |
38129a13 | 1854 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1855 | struct mount *q; |
38129a13 | 1856 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1857 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1858 | child->mnt_mountpoint); | |
1859 | commit_tree(child, q); | |
b90fa9ae | 1860 | } |
719ea2fb | 1861 | unlock_mount_hash(); |
99b7db7b | 1862 | |
b90fa9ae | 1863 | return 0; |
719f5d7f MS |
1864 | |
1865 | out_cleanup_ids: | |
f2ebb3a9 AV |
1866 | while (!hlist_empty(&tree_list)) { |
1867 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1868 | umount_tree(child, 0); | |
1869 | } | |
1870 | unlock_mount_hash(); | |
0b1b901b | 1871 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1872 | out: |
1873 | return err; | |
b90fa9ae RP |
1874 | } |
1875 | ||
84d17192 | 1876 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1877 | { |
1878 | struct vfsmount *mnt; | |
84d17192 | 1879 | struct dentry *dentry = path->dentry; |
b12cea91 | 1880 | retry: |
84d17192 AV |
1881 | mutex_lock(&dentry->d_inode->i_mutex); |
1882 | if (unlikely(cant_mount(dentry))) { | |
1883 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1884 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1885 | } |
97216be0 | 1886 | namespace_lock(); |
b12cea91 | 1887 | mnt = lookup_mnt(path); |
84d17192 | 1888 | if (likely(!mnt)) { |
e2dfa935 EB |
1889 | struct mountpoint *mp = lookup_mountpoint(dentry); |
1890 | if (!mp) | |
1891 | mp = new_mountpoint(dentry); | |
84d17192 | 1892 | if (IS_ERR(mp)) { |
97216be0 | 1893 | namespace_unlock(); |
84d17192 AV |
1894 | mutex_unlock(&dentry->d_inode->i_mutex); |
1895 | return mp; | |
1896 | } | |
1897 | return mp; | |
1898 | } | |
97216be0 | 1899 | namespace_unlock(); |
b12cea91 AV |
1900 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1901 | path_put(path); | |
1902 | path->mnt = mnt; | |
84d17192 | 1903 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1904 | goto retry; |
1905 | } | |
1906 | ||
84d17192 | 1907 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1908 | { |
84d17192 AV |
1909 | struct dentry *dentry = where->m_dentry; |
1910 | put_mountpoint(where); | |
328e6d90 | 1911 | namespace_unlock(); |
84d17192 | 1912 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1913 | } |
1914 | ||
84d17192 | 1915 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1916 | { |
95bc5f25 | 1917 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1918 | return -EINVAL; |
1919 | ||
84d17192 | 1920 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1921 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1922 | return -ENOTDIR; |
1923 | ||
84d17192 | 1924 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1925 | } |
1926 | ||
7a2e8a8f VA |
1927 | /* |
1928 | * Sanity check the flags to change_mnt_propagation. | |
1929 | */ | |
1930 | ||
1931 | static int flags_to_propagation_type(int flags) | |
1932 | { | |
7c6e984d | 1933 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1934 | |
1935 | /* Fail if any non-propagation flags are set */ | |
1936 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1937 | return 0; | |
1938 | /* Only one propagation flag should be set */ | |
1939 | if (!is_power_of_2(type)) | |
1940 | return 0; | |
1941 | return type; | |
1942 | } | |
1943 | ||
07b20889 RP |
1944 | /* |
1945 | * recursively change the type of the mountpoint. | |
1946 | */ | |
0a0d8a46 | 1947 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1948 | { |
315fc83e | 1949 | struct mount *m; |
4b8b21f4 | 1950 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1951 | int recurse = flag & MS_REC; |
7a2e8a8f | 1952 | int type; |
719f5d7f | 1953 | int err = 0; |
07b20889 | 1954 | |
2d92ab3c | 1955 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1956 | return -EINVAL; |
1957 | ||
7a2e8a8f VA |
1958 | type = flags_to_propagation_type(flag); |
1959 | if (!type) | |
1960 | return -EINVAL; | |
1961 | ||
97216be0 | 1962 | namespace_lock(); |
719f5d7f MS |
1963 | if (type == MS_SHARED) { |
1964 | err = invent_group_ids(mnt, recurse); | |
1965 | if (err) | |
1966 | goto out_unlock; | |
1967 | } | |
1968 | ||
719ea2fb | 1969 | lock_mount_hash(); |
909b0a88 | 1970 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1971 | change_mnt_propagation(m, type); |
719ea2fb | 1972 | unlock_mount_hash(); |
719f5d7f MS |
1973 | |
1974 | out_unlock: | |
97216be0 | 1975 | namespace_unlock(); |
719f5d7f | 1976 | return err; |
07b20889 RP |
1977 | } |
1978 | ||
5ff9d8a6 EB |
1979 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1980 | { | |
1981 | struct mount *child; | |
1982 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1983 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1984 | continue; | |
1985 | ||
1986 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1987 | return true; | |
1988 | } | |
1989 | return false; | |
1990 | } | |
1991 | ||
1da177e4 LT |
1992 | /* |
1993 | * do loopback mount. | |
1994 | */ | |
808d4e3c | 1995 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1996 | int recurse) |
1da177e4 | 1997 | { |
2d92ab3c | 1998 | struct path old_path; |
84d17192 AV |
1999 | struct mount *mnt = NULL, *old, *parent; |
2000 | struct mountpoint *mp; | |
57eccb83 | 2001 | int err; |
1da177e4 LT |
2002 | if (!old_name || !*old_name) |
2003 | return -EINVAL; | |
815d405c | 2004 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2005 | if (err) |
2006 | return err; | |
2007 | ||
8823c079 | 2008 | err = -EINVAL; |
4ce5d2b1 | 2009 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2010 | goto out; |
2011 | ||
84d17192 AV |
2012 | mp = lock_mount(path); |
2013 | err = PTR_ERR(mp); | |
2014 | if (IS_ERR(mp)) | |
b12cea91 AV |
2015 | goto out; |
2016 | ||
87129cc0 | 2017 | old = real_mount(old_path.mnt); |
84d17192 | 2018 | parent = real_mount(path->mnt); |
87129cc0 | 2019 | |
1da177e4 | 2020 | err = -EINVAL; |
fc7be130 | 2021 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2022 | goto out2; |
9676f0c6 | 2023 | |
e149ed2b AV |
2024 | if (!check_mnt(parent)) |
2025 | goto out2; | |
2026 | ||
2027 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2028 | goto out2; |
1da177e4 | 2029 | |
5ff9d8a6 EB |
2030 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2031 | goto out2; | |
2032 | ||
ccd48bc7 | 2033 | if (recurse) |
4ce5d2b1 | 2034 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2035 | else |
87129cc0 | 2036 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2037 | |
be34d1a3 DH |
2038 | if (IS_ERR(mnt)) { |
2039 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2040 | goto out2; |
be34d1a3 | 2041 | } |
ccd48bc7 | 2042 | |
5ff9d8a6 EB |
2043 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2044 | ||
84d17192 | 2045 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2046 | if (err) { |
719ea2fb | 2047 | lock_mount_hash(); |
328e6d90 | 2048 | umount_tree(mnt, 0); |
719ea2fb | 2049 | unlock_mount_hash(); |
5b83d2c5 | 2050 | } |
b12cea91 | 2051 | out2: |
84d17192 | 2052 | unlock_mount(mp); |
ccd48bc7 | 2053 | out: |
2d92ab3c | 2054 | path_put(&old_path); |
1da177e4 LT |
2055 | return err; |
2056 | } | |
2057 | ||
2e4b7fcd DH |
2058 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2059 | { | |
2060 | int error = 0; | |
2061 | int readonly_request = 0; | |
2062 | ||
2063 | if (ms_flags & MS_RDONLY) | |
2064 | readonly_request = 1; | |
2065 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2066 | return 0; | |
2067 | ||
2068 | if (readonly_request) | |
83adc753 | 2069 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2070 | else |
83adc753 | 2071 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2072 | return error; |
2073 | } | |
2074 | ||
1da177e4 LT |
2075 | /* |
2076 | * change filesystem flags. dir should be a physical root of filesystem. | |
2077 | * If you've mounted a non-root directory somewhere and want to do remount | |
2078 | * on it - tough luck. | |
2079 | */ | |
0a0d8a46 | 2080 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2081 | void *data) |
2082 | { | |
2083 | int err; | |
2d92ab3c | 2084 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2085 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2086 | |
143c8c91 | 2087 | if (!check_mnt(mnt)) |
1da177e4 LT |
2088 | return -EINVAL; |
2089 | ||
2d92ab3c | 2090 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2091 | return -EINVAL; |
2092 | ||
07b64558 EB |
2093 | /* Don't allow changing of locked mnt flags. |
2094 | * | |
2095 | * No locks need to be held here while testing the various | |
2096 | * MNT_LOCK flags because those flags can never be cleared | |
2097 | * once they are set. | |
2098 | */ | |
2099 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2100 | !(mnt_flags & MNT_READONLY)) { | |
2101 | return -EPERM; | |
2102 | } | |
9566d674 EB |
2103 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2104 | !(mnt_flags & MNT_NODEV)) { | |
3e186641 EB |
2105 | /* Was the nodev implicitly added in mount? */ |
2106 | if ((mnt->mnt_ns->user_ns != &init_user_ns) && | |
2107 | !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2108 | mnt_flags |= MNT_NODEV; | |
2109 | } else { | |
2110 | return -EPERM; | |
2111 | } | |
9566d674 EB |
2112 | } |
2113 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2114 | !(mnt_flags & MNT_NOSUID)) { | |
2115 | return -EPERM; | |
2116 | } | |
2117 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2118 | !(mnt_flags & MNT_NOEXEC)) { | |
2119 | return -EPERM; | |
2120 | } | |
2121 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2122 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2123 | return -EPERM; | |
2124 | } | |
2125 | ||
ff36fe2c EP |
2126 | err = security_sb_remount(sb, data); |
2127 | if (err) | |
2128 | return err; | |
2129 | ||
1da177e4 | 2130 | down_write(&sb->s_umount); |
2e4b7fcd | 2131 | if (flags & MS_BIND) |
2d92ab3c | 2132 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
2133 | else if (!capable(CAP_SYS_ADMIN)) |
2134 | err = -EPERM; | |
4aa98cf7 | 2135 | else |
2e4b7fcd | 2136 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2137 | if (!err) { |
719ea2fb | 2138 | lock_mount_hash(); |
a6138db8 | 2139 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2140 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2141 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2142 | unlock_mount_hash(); |
0e55a7cc | 2143 | } |
6339dab8 | 2144 | up_write(&sb->s_umount); |
1da177e4 LT |
2145 | return err; |
2146 | } | |
2147 | ||
cbbe362c | 2148 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2149 | { |
315fc83e | 2150 | struct mount *p; |
909b0a88 | 2151 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2152 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2153 | return 1; |
2154 | } | |
2155 | return 0; | |
2156 | } | |
2157 | ||
808d4e3c | 2158 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2159 | { |
2d92ab3c | 2160 | struct path old_path, parent_path; |
676da58d | 2161 | struct mount *p; |
0fb54e50 | 2162 | struct mount *old; |
84d17192 | 2163 | struct mountpoint *mp; |
57eccb83 | 2164 | int err; |
1da177e4 LT |
2165 | if (!old_name || !*old_name) |
2166 | return -EINVAL; | |
2d92ab3c | 2167 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2168 | if (err) |
2169 | return err; | |
2170 | ||
84d17192 AV |
2171 | mp = lock_mount(path); |
2172 | err = PTR_ERR(mp); | |
2173 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2174 | goto out; |
2175 | ||
143c8c91 | 2176 | old = real_mount(old_path.mnt); |
fc7be130 | 2177 | p = real_mount(path->mnt); |
143c8c91 | 2178 | |
1da177e4 | 2179 | err = -EINVAL; |
fc7be130 | 2180 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2181 | goto out1; |
2182 | ||
5ff9d8a6 EB |
2183 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2184 | goto out1; | |
2185 | ||
1da177e4 | 2186 | err = -EINVAL; |
2d92ab3c | 2187 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2188 | goto out1; |
1da177e4 | 2189 | |
676da58d | 2190 | if (!mnt_has_parent(old)) |
21444403 | 2191 | goto out1; |
1da177e4 | 2192 | |
2d92ab3c AV |
2193 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
2194 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
2195 | goto out1; |
2196 | /* | |
2197 | * Don't move a mount residing in a shared parent. | |
2198 | */ | |
fc7be130 | 2199 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2200 | goto out1; |
9676f0c6 RP |
2201 | /* |
2202 | * Don't move a mount tree containing unbindable mounts to a destination | |
2203 | * mount which is shared. | |
2204 | */ | |
fc7be130 | 2205 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2206 | goto out1; |
1da177e4 | 2207 | err = -ELOOP; |
fc7be130 | 2208 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2209 | if (p == old) |
21444403 | 2210 | goto out1; |
1da177e4 | 2211 | |
84d17192 | 2212 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2213 | if (err) |
21444403 | 2214 | goto out1; |
1da177e4 LT |
2215 | |
2216 | /* if the mount is moved, it should no longer be expire | |
2217 | * automatically */ | |
6776db3d | 2218 | list_del_init(&old->mnt_expire); |
1da177e4 | 2219 | out1: |
84d17192 | 2220 | unlock_mount(mp); |
1da177e4 | 2221 | out: |
1da177e4 | 2222 | if (!err) |
1a390689 | 2223 | path_put(&parent_path); |
2d92ab3c | 2224 | path_put(&old_path); |
1da177e4 LT |
2225 | return err; |
2226 | } | |
2227 | ||
9d412a43 AV |
2228 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2229 | { | |
2230 | int err; | |
2231 | const char *subtype = strchr(fstype, '.'); | |
2232 | if (subtype) { | |
2233 | subtype++; | |
2234 | err = -EINVAL; | |
2235 | if (!subtype[0]) | |
2236 | goto err; | |
2237 | } else | |
2238 | subtype = ""; | |
2239 | ||
2240 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2241 | err = -ENOMEM; | |
2242 | if (!mnt->mnt_sb->s_subtype) | |
2243 | goto err; | |
2244 | return mnt; | |
2245 | ||
2246 | err: | |
2247 | mntput(mnt); | |
2248 | return ERR_PTR(err); | |
2249 | } | |
2250 | ||
9d412a43 AV |
2251 | /* |
2252 | * add a mount into a namespace's mount tree | |
2253 | */ | |
95bc5f25 | 2254 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2255 | { |
84d17192 AV |
2256 | struct mountpoint *mp; |
2257 | struct mount *parent; | |
9d412a43 AV |
2258 | int err; |
2259 | ||
f2ebb3a9 | 2260 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2261 | |
84d17192 AV |
2262 | mp = lock_mount(path); |
2263 | if (IS_ERR(mp)) | |
2264 | return PTR_ERR(mp); | |
9d412a43 | 2265 | |
84d17192 | 2266 | parent = real_mount(path->mnt); |
9d412a43 | 2267 | err = -EINVAL; |
84d17192 | 2268 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2269 | /* that's acceptable only for automounts done in private ns */ |
2270 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2271 | goto unlock; | |
2272 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2273 | if (!parent->mnt_ns) |
156cacb1 AV |
2274 | goto unlock; |
2275 | } | |
9d412a43 AV |
2276 | |
2277 | /* Refuse the same filesystem on the same mount point */ | |
2278 | err = -EBUSY; | |
95bc5f25 | 2279 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2280 | path->mnt->mnt_root == path->dentry) |
2281 | goto unlock; | |
2282 | ||
2283 | err = -EINVAL; | |
95bc5f25 | 2284 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2285 | goto unlock; |
2286 | ||
95bc5f25 | 2287 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2288 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2289 | |
2290 | unlock: | |
84d17192 | 2291 | unlock_mount(mp); |
9d412a43 AV |
2292 | return err; |
2293 | } | |
b1e75df4 | 2294 | |
1da177e4 LT |
2295 | /* |
2296 | * create a new mount for userspace and request it to be added into the | |
2297 | * namespace's tree | |
2298 | */ | |
0c55cfc4 | 2299 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2300 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2301 | { |
0c55cfc4 | 2302 | struct file_system_type *type; |
9b40bc90 | 2303 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2304 | struct vfsmount *mnt; |
15f9a3f3 | 2305 | int err; |
1da177e4 | 2306 | |
0c55cfc4 | 2307 | if (!fstype) |
1da177e4 LT |
2308 | return -EINVAL; |
2309 | ||
0c55cfc4 EB |
2310 | type = get_fs_type(fstype); |
2311 | if (!type) | |
2312 | return -ENODEV; | |
2313 | ||
2314 | if (user_ns != &init_user_ns) { | |
2315 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2316 | put_filesystem(type); | |
2317 | return -EPERM; | |
2318 | } | |
2319 | /* Only in special cases allow devices from mounts | |
2320 | * created outside the initial user namespace. | |
2321 | */ | |
2322 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2323 | flags |= MS_NODEV; | |
9566d674 | 2324 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 EB |
2325 | } |
2326 | } | |
2327 | ||
2328 | mnt = vfs_kern_mount(type, flags, name, data); | |
2329 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2330 | !mnt->mnt_sb->s_subtype) | |
2331 | mnt = fs_set_subtype(mnt, fstype); | |
2332 | ||
2333 | put_filesystem(type); | |
1da177e4 LT |
2334 | if (IS_ERR(mnt)) |
2335 | return PTR_ERR(mnt); | |
2336 | ||
95bc5f25 | 2337 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2338 | if (err) |
2339 | mntput(mnt); | |
2340 | return err; | |
1da177e4 LT |
2341 | } |
2342 | ||
19a167af AV |
2343 | int finish_automount(struct vfsmount *m, struct path *path) |
2344 | { | |
6776db3d | 2345 | struct mount *mnt = real_mount(m); |
19a167af AV |
2346 | int err; |
2347 | /* The new mount record should have at least 2 refs to prevent it being | |
2348 | * expired before we get a chance to add it | |
2349 | */ | |
6776db3d | 2350 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2351 | |
2352 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2353 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2354 | err = -ELOOP; |
2355 | goto fail; | |
19a167af AV |
2356 | } |
2357 | ||
95bc5f25 | 2358 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2359 | if (!err) |
2360 | return 0; | |
2361 | fail: | |
2362 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2363 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2364 | namespace_lock(); |
6776db3d | 2365 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2366 | namespace_unlock(); |
19a167af | 2367 | } |
b1e75df4 AV |
2368 | mntput(m); |
2369 | mntput(m); | |
19a167af AV |
2370 | return err; |
2371 | } | |
2372 | ||
ea5b778a DH |
2373 | /** |
2374 | * mnt_set_expiry - Put a mount on an expiration list | |
2375 | * @mnt: The mount to list. | |
2376 | * @expiry_list: The list to add the mount to. | |
2377 | */ | |
2378 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2379 | { | |
97216be0 | 2380 | namespace_lock(); |
ea5b778a | 2381 | |
6776db3d | 2382 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2383 | |
97216be0 | 2384 | namespace_unlock(); |
ea5b778a DH |
2385 | } |
2386 | EXPORT_SYMBOL(mnt_set_expiry); | |
2387 | ||
1da177e4 LT |
2388 | /* |
2389 | * process a list of expirable mountpoints with the intent of discarding any | |
2390 | * mountpoints that aren't in use and haven't been touched since last we came | |
2391 | * here | |
2392 | */ | |
2393 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2394 | { | |
761d5c38 | 2395 | struct mount *mnt, *next; |
1da177e4 LT |
2396 | LIST_HEAD(graveyard); |
2397 | ||
2398 | if (list_empty(mounts)) | |
2399 | return; | |
2400 | ||
97216be0 | 2401 | namespace_lock(); |
719ea2fb | 2402 | lock_mount_hash(); |
1da177e4 LT |
2403 | |
2404 | /* extract from the expiration list every vfsmount that matches the | |
2405 | * following criteria: | |
2406 | * - only referenced by its parent vfsmount | |
2407 | * - still marked for expiry (marked on the last call here; marks are | |
2408 | * cleared by mntput()) | |
2409 | */ | |
6776db3d | 2410 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2411 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2412 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2413 | continue; |
6776db3d | 2414 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2415 | } |
bcc5c7d2 | 2416 | while (!list_empty(&graveyard)) { |
6776db3d | 2417 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2418 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2419 | umount_tree(mnt, 1); |
bcc5c7d2 | 2420 | } |
719ea2fb | 2421 | unlock_mount_hash(); |
3ab6abee | 2422 | namespace_unlock(); |
5528f911 TM |
2423 | } |
2424 | ||
2425 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2426 | ||
2427 | /* | |
2428 | * Ripoff of 'select_parent()' | |
2429 | * | |
2430 | * search the list of submounts for a given mountpoint, and move any | |
2431 | * shrinkable submounts to the 'graveyard' list. | |
2432 | */ | |
692afc31 | 2433 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2434 | { |
692afc31 | 2435 | struct mount *this_parent = parent; |
5528f911 TM |
2436 | struct list_head *next; |
2437 | int found = 0; | |
2438 | ||
2439 | repeat: | |
6b41d536 | 2440 | next = this_parent->mnt_mounts.next; |
5528f911 | 2441 | resume: |
6b41d536 | 2442 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2443 | struct list_head *tmp = next; |
6b41d536 | 2444 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2445 | |
2446 | next = tmp->next; | |
692afc31 | 2447 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2448 | continue; |
5528f911 TM |
2449 | /* |
2450 | * Descend a level if the d_mounts list is non-empty. | |
2451 | */ | |
6b41d536 | 2452 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2453 | this_parent = mnt; |
2454 | goto repeat; | |
2455 | } | |
1da177e4 | 2456 | |
1ab59738 | 2457 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2458 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2459 | found++; |
2460 | } | |
1da177e4 | 2461 | } |
5528f911 TM |
2462 | /* |
2463 | * All done at this level ... ascend and resume the search | |
2464 | */ | |
2465 | if (this_parent != parent) { | |
6b41d536 | 2466 | next = this_parent->mnt_child.next; |
0714a533 | 2467 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2468 | goto resume; |
2469 | } | |
2470 | return found; | |
2471 | } | |
2472 | ||
2473 | /* | |
2474 | * process a list of expirable mountpoints with the intent of discarding any | |
2475 | * submounts of a specific parent mountpoint | |
99b7db7b | 2476 | * |
48a066e7 | 2477 | * mount_lock must be held for write |
5528f911 | 2478 | */ |
b54b9be7 | 2479 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2480 | { |
2481 | LIST_HEAD(graveyard); | |
761d5c38 | 2482 | struct mount *m; |
5528f911 | 2483 | |
5528f911 | 2484 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2485 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2486 | while (!list_empty(&graveyard)) { |
761d5c38 | 2487 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2488 | mnt_expire); |
143c8c91 | 2489 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2490 | umount_tree(m, 1); |
bcc5c7d2 AV |
2491 | } |
2492 | } | |
1da177e4 LT |
2493 | } |
2494 | ||
1da177e4 LT |
2495 | /* |
2496 | * Some copy_from_user() implementations do not return the exact number of | |
2497 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2498 | * Note that this function differs from copy_from_user() in that it will oops | |
2499 | * on bad values of `to', rather than returning a short copy. | |
2500 | */ | |
b58fed8b RP |
2501 | static long exact_copy_from_user(void *to, const void __user * from, |
2502 | unsigned long n) | |
1da177e4 LT |
2503 | { |
2504 | char *t = to; | |
2505 | const char __user *f = from; | |
2506 | char c; | |
2507 | ||
2508 | if (!access_ok(VERIFY_READ, from, n)) | |
2509 | return n; | |
2510 | ||
2511 | while (n) { | |
2512 | if (__get_user(c, f)) { | |
2513 | memset(t, 0, n); | |
2514 | break; | |
2515 | } | |
2516 | *t++ = c; | |
2517 | f++; | |
2518 | n--; | |
2519 | } | |
2520 | return n; | |
2521 | } | |
2522 | ||
b58fed8b | 2523 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2524 | { |
2525 | int i; | |
2526 | unsigned long page; | |
2527 | unsigned long size; | |
b58fed8b | 2528 | |
1da177e4 LT |
2529 | *where = 0; |
2530 | if (!data) | |
2531 | return 0; | |
2532 | ||
2533 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2534 | return -ENOMEM; | |
2535 | ||
2536 | /* We only care that *some* data at the address the user | |
2537 | * gave us is valid. Just in case, we'll zero | |
2538 | * the remainder of the page. | |
2539 | */ | |
2540 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2541 | size = TASK_SIZE - (unsigned long)data; | |
2542 | if (size > PAGE_SIZE) | |
2543 | size = PAGE_SIZE; | |
2544 | ||
2545 | i = size - exact_copy_from_user((void *)page, data, size); | |
2546 | if (!i) { | |
b58fed8b | 2547 | free_page(page); |
1da177e4 LT |
2548 | return -EFAULT; |
2549 | } | |
2550 | if (i != PAGE_SIZE) | |
2551 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2552 | *where = page; | |
2553 | return 0; | |
2554 | } | |
2555 | ||
b8850d1f | 2556 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2557 | { |
b8850d1f | 2558 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2559 | } |
2560 | ||
1da177e4 LT |
2561 | /* |
2562 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2563 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2564 | * | |
2565 | * data is a (void *) that can point to any structure up to | |
2566 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2567 | * information (or be NULL). | |
2568 | * | |
2569 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2570 | * When the flags word was introduced its top half was required | |
2571 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2572 | * Therefore, if this magic number is present, it carries no information | |
2573 | * and must be discarded. | |
2574 | */ | |
5e6123f3 | 2575 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2576 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2577 | { |
2d92ab3c | 2578 | struct path path; |
1da177e4 LT |
2579 | int retval = 0; |
2580 | int mnt_flags = 0; | |
2581 | ||
2582 | /* Discard magic */ | |
2583 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2584 | flags &= ~MS_MGC_MSK; | |
2585 | ||
2586 | /* Basic sanity checks */ | |
1da177e4 LT |
2587 | if (data_page) |
2588 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2589 | ||
a27ab9f2 | 2590 | /* ... and get the mountpoint */ |
5e6123f3 | 2591 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2592 | if (retval) |
2593 | return retval; | |
2594 | ||
2595 | retval = security_sb_mount(dev_name, &path, | |
2596 | type_page, flags, data_page); | |
0d5cadb8 AV |
2597 | if (!retval && !may_mount()) |
2598 | retval = -EPERM; | |
a27ab9f2 TH |
2599 | if (retval) |
2600 | goto dput_out; | |
2601 | ||
613cbe3d AK |
2602 | /* Default to relatime unless overriden */ |
2603 | if (!(flags & MS_NOATIME)) | |
2604 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2605 | |
1da177e4 LT |
2606 | /* Separate the per-mountpoint flags */ |
2607 | if (flags & MS_NOSUID) | |
2608 | mnt_flags |= MNT_NOSUID; | |
2609 | if (flags & MS_NODEV) | |
2610 | mnt_flags |= MNT_NODEV; | |
2611 | if (flags & MS_NOEXEC) | |
2612 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2613 | if (flags & MS_NOATIME) |
2614 | mnt_flags |= MNT_NOATIME; | |
2615 | if (flags & MS_NODIRATIME) | |
2616 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2617 | if (flags & MS_STRICTATIME) |
2618 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2619 | if (flags & MS_RDONLY) |
2620 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2621 | |
ffbc6f0e EB |
2622 | /* The default atime for remount is preservation */ |
2623 | if ((flags & MS_REMOUNT) && | |
2624 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2625 | MS_STRICTATIME)) == 0)) { | |
2626 | mnt_flags &= ~MNT_ATIME_MASK; | |
2627 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2628 | } | |
2629 | ||
7a4dec53 | 2630 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2631 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2632 | MS_STRICTATIME); | |
1da177e4 | 2633 | |
1da177e4 | 2634 | if (flags & MS_REMOUNT) |
2d92ab3c | 2635 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2636 | data_page); |
2637 | else if (flags & MS_BIND) | |
2d92ab3c | 2638 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2639 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2640 | retval = do_change_type(&path, flags); |
1da177e4 | 2641 | else if (flags & MS_MOVE) |
2d92ab3c | 2642 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2643 | else |
2d92ab3c | 2644 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2645 | dev_name, data_page); |
2646 | dput_out: | |
2d92ab3c | 2647 | path_put(&path); |
1da177e4 LT |
2648 | return retval; |
2649 | } | |
2650 | ||
771b1371 EB |
2651 | static void free_mnt_ns(struct mnt_namespace *ns) |
2652 | { | |
6344c433 | 2653 | ns_free_inum(&ns->ns); |
771b1371 EB |
2654 | put_user_ns(ns->user_ns); |
2655 | kfree(ns); | |
2656 | } | |
2657 | ||
8823c079 EB |
2658 | /* |
2659 | * Assign a sequence number so we can detect when we attempt to bind | |
2660 | * mount a reference to an older mount namespace into the current | |
2661 | * mount namespace, preventing reference counting loops. A 64bit | |
2662 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2663 | * is effectively never, so we can ignore the possibility. | |
2664 | */ | |
2665 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2666 | ||
771b1371 | 2667 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2668 | { |
2669 | struct mnt_namespace *new_ns; | |
98f842e6 | 2670 | int ret; |
cf8d2c11 TM |
2671 | |
2672 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2673 | if (!new_ns) | |
2674 | return ERR_PTR(-ENOMEM); | |
6344c433 | 2675 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2676 | if (ret) { |
2677 | kfree(new_ns); | |
2678 | return ERR_PTR(ret); | |
2679 | } | |
33c42940 | 2680 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2681 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2682 | atomic_set(&new_ns->count, 1); |
2683 | new_ns->root = NULL; | |
2684 | INIT_LIST_HEAD(&new_ns->list); | |
2685 | init_waitqueue_head(&new_ns->poll); | |
2686 | new_ns->event = 0; | |
771b1371 | 2687 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2688 | return new_ns; |
2689 | } | |
2690 | ||
9559f689 AV |
2691 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2692 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2693 | { |
6b3286ed | 2694 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2695 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2696 | struct mount *p, *q; |
9559f689 | 2697 | struct mount *old; |
cb338d06 | 2698 | struct mount *new; |
7a472ef4 | 2699 | int copy_flags; |
1da177e4 | 2700 | |
9559f689 AV |
2701 | BUG_ON(!ns); |
2702 | ||
2703 | if (likely(!(flags & CLONE_NEWNS))) { | |
2704 | get_mnt_ns(ns); | |
2705 | return ns; | |
2706 | } | |
2707 | ||
2708 | old = ns->root; | |
2709 | ||
771b1371 | 2710 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2711 | if (IS_ERR(new_ns)) |
2712 | return new_ns; | |
1da177e4 | 2713 | |
97216be0 | 2714 | namespace_lock(); |
1da177e4 | 2715 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2716 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2717 | if (user_ns != ns->user_ns) |
132c94e3 | 2718 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2719 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2720 | if (IS_ERR(new)) { |
328e6d90 | 2721 | namespace_unlock(); |
771b1371 | 2722 | free_mnt_ns(new_ns); |
be34d1a3 | 2723 | return ERR_CAST(new); |
1da177e4 | 2724 | } |
be08d6d2 | 2725 | new_ns->root = new; |
1a4eeaf2 | 2726 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2727 | |
2728 | /* | |
2729 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2730 | * as belonging to new namespace. We have already acquired a private | |
2731 | * fs_struct, so tsk->fs->lock is not needed. | |
2732 | */ | |
909b0a88 | 2733 | p = old; |
cb338d06 | 2734 | q = new; |
1da177e4 | 2735 | while (p) { |
143c8c91 | 2736 | q->mnt_ns = new_ns; |
9559f689 AV |
2737 | if (new_fs) { |
2738 | if (&p->mnt == new_fs->root.mnt) { | |
2739 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2740 | rootmnt = &p->mnt; |
1da177e4 | 2741 | } |
9559f689 AV |
2742 | if (&p->mnt == new_fs->pwd.mnt) { |
2743 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2744 | pwdmnt = &p->mnt; |
1da177e4 | 2745 | } |
1da177e4 | 2746 | } |
909b0a88 AV |
2747 | p = next_mnt(p, old); |
2748 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2749 | if (!q) |
2750 | break; | |
2751 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2752 | p = next_mnt(p, old); | |
1da177e4 | 2753 | } |
328e6d90 | 2754 | namespace_unlock(); |
1da177e4 | 2755 | |
1da177e4 | 2756 | if (rootmnt) |
f03c6599 | 2757 | mntput(rootmnt); |
1da177e4 | 2758 | if (pwdmnt) |
f03c6599 | 2759 | mntput(pwdmnt); |
1da177e4 | 2760 | |
741a2951 | 2761 | return new_ns; |
1da177e4 LT |
2762 | } |
2763 | ||
cf8d2c11 TM |
2764 | /** |
2765 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2766 | * @mnt: pointer to the new root filesystem mountpoint | |
2767 | */ | |
1a4eeaf2 | 2768 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2769 | { |
771b1371 | 2770 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2771 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2772 | struct mount *mnt = real_mount(m); |
2773 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2774 | new_ns->root = mnt; |
b1983cd8 | 2775 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2776 | } else { |
1a4eeaf2 | 2777 | mntput(m); |
cf8d2c11 TM |
2778 | } |
2779 | return new_ns; | |
2780 | } | |
cf8d2c11 | 2781 | |
ea441d11 AV |
2782 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2783 | { | |
2784 | struct mnt_namespace *ns; | |
d31da0f0 | 2785 | struct super_block *s; |
ea441d11 AV |
2786 | struct path path; |
2787 | int err; | |
2788 | ||
2789 | ns = create_mnt_ns(mnt); | |
2790 | if (IS_ERR(ns)) | |
2791 | return ERR_CAST(ns); | |
2792 | ||
2793 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2794 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2795 | ||
2796 | put_mnt_ns(ns); | |
2797 | ||
2798 | if (err) | |
2799 | return ERR_PTR(err); | |
2800 | ||
2801 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2802 | s = path.mnt->mnt_sb; |
2803 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2804 | mntput(path.mnt); |
2805 | /* lock the sucker */ | |
d31da0f0 | 2806 | down_write(&s->s_umount); |
ea441d11 AV |
2807 | /* ... and return the root of (sub)tree on it */ |
2808 | return path.dentry; | |
2809 | } | |
2810 | EXPORT_SYMBOL(mount_subtree); | |
2811 | ||
bdc480e3 HC |
2812 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2813 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2814 | { |
eca6f534 VN |
2815 | int ret; |
2816 | char *kernel_type; | |
eca6f534 | 2817 | char *kernel_dev; |
1da177e4 | 2818 | unsigned long data_page; |
1da177e4 | 2819 | |
b8850d1f TG |
2820 | kernel_type = copy_mount_string(type); |
2821 | ret = PTR_ERR(kernel_type); | |
2822 | if (IS_ERR(kernel_type)) | |
eca6f534 | 2823 | goto out_type; |
1da177e4 | 2824 | |
b8850d1f TG |
2825 | kernel_dev = copy_mount_string(dev_name); |
2826 | ret = PTR_ERR(kernel_dev); | |
2827 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 2828 | goto out_dev; |
1da177e4 | 2829 | |
eca6f534 VN |
2830 | ret = copy_mount_options(data, &data_page); |
2831 | if (ret < 0) | |
2832 | goto out_data; | |
1da177e4 | 2833 | |
5e6123f3 | 2834 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, |
eca6f534 | 2835 | (void *) data_page); |
1da177e4 | 2836 | |
eca6f534 VN |
2837 | free_page(data_page); |
2838 | out_data: | |
2839 | kfree(kernel_dev); | |
2840 | out_dev: | |
eca6f534 VN |
2841 | kfree(kernel_type); |
2842 | out_type: | |
2843 | return ret; | |
1da177e4 LT |
2844 | } |
2845 | ||
afac7cba AV |
2846 | /* |
2847 | * Return true if path is reachable from root | |
2848 | * | |
48a066e7 | 2849 | * namespace_sem or mount_lock is held |
afac7cba | 2850 | */ |
643822b4 | 2851 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2852 | const struct path *root) |
2853 | { | |
643822b4 | 2854 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2855 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2856 | mnt = mnt->mnt_parent; |
afac7cba | 2857 | } |
643822b4 | 2858 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2859 | } |
2860 | ||
2861 | int path_is_under(struct path *path1, struct path *path2) | |
2862 | { | |
2863 | int res; | |
48a066e7 | 2864 | read_seqlock_excl(&mount_lock); |
643822b4 | 2865 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2866 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2867 | return res; |
2868 | } | |
2869 | EXPORT_SYMBOL(path_is_under); | |
2870 | ||
1da177e4 LT |
2871 | /* |
2872 | * pivot_root Semantics: | |
2873 | * Moves the root file system of the current process to the directory put_old, | |
2874 | * makes new_root as the new root file system of the current process, and sets | |
2875 | * root/cwd of all processes which had them on the current root to new_root. | |
2876 | * | |
2877 | * Restrictions: | |
2878 | * The new_root and put_old must be directories, and must not be on the | |
2879 | * same file system as the current process root. The put_old must be | |
2880 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2881 | * pointed to by put_old must yield the same directory as new_root. No other | |
2882 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2883 | * | |
4a0d11fa NB |
2884 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2885 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2886 | * in this situation. | |
2887 | * | |
1da177e4 LT |
2888 | * Notes: |
2889 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2890 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2891 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2892 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2893 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2894 | * first. | |
2895 | */ | |
3480b257 HC |
2896 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2897 | const char __user *, put_old) | |
1da177e4 | 2898 | { |
2d8f3038 | 2899 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2900 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2901 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2902 | int error; |
2903 | ||
9b40bc90 | 2904 | if (!may_mount()) |
1da177e4 LT |
2905 | return -EPERM; |
2906 | ||
2d8f3038 | 2907 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2908 | if (error) |
2909 | goto out0; | |
1da177e4 | 2910 | |
2d8f3038 | 2911 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2912 | if (error) |
2913 | goto out1; | |
2914 | ||
2d8f3038 | 2915 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2916 | if (error) |
2917 | goto out2; | |
1da177e4 | 2918 | |
f7ad3c6b | 2919 | get_fs_root(current->fs, &root); |
84d17192 AV |
2920 | old_mp = lock_mount(&old); |
2921 | error = PTR_ERR(old_mp); | |
2922 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2923 | goto out3; |
2924 | ||
1da177e4 | 2925 | error = -EINVAL; |
419148da AV |
2926 | new_mnt = real_mount(new.mnt); |
2927 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2928 | old_mnt = real_mount(old.mnt); |
2929 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2930 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2931 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2932 | goto out4; |
143c8c91 | 2933 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2934 | goto out4; |
5ff9d8a6 EB |
2935 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2936 | goto out4; | |
1da177e4 | 2937 | error = -ENOENT; |
f3da392e | 2938 | if (d_unlinked(new.dentry)) |
b12cea91 | 2939 | goto out4; |
1da177e4 | 2940 | error = -EBUSY; |
84d17192 | 2941 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2942 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2943 | error = -EINVAL; |
8c3ee42e | 2944 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2945 | goto out4; /* not a mountpoint */ |
676da58d | 2946 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2947 | goto out4; /* not attached */ |
84d17192 | 2948 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2949 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2950 | goto out4; /* not a mountpoint */ |
676da58d | 2951 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2952 | goto out4; /* not attached */ |
4ac91378 | 2953 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2954 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2955 | goto out4; |
0d082601 EB |
2956 | /* make certain new is below the root */ |
2957 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
2958 | goto out4; | |
84d17192 | 2959 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2960 | lock_mount_hash(); |
419148da AV |
2961 | detach_mnt(new_mnt, &parent_path); |
2962 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2963 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2964 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2965 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2966 | } | |
4ac91378 | 2967 | /* mount old root on put_old */ |
84d17192 | 2968 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2969 | /* mount new_root on / */ |
84d17192 | 2970 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2971 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
2972 | /* A moved mount should not expire automatically */ |
2973 | list_del_init(&new_mnt->mnt_expire); | |
719ea2fb | 2974 | unlock_mount_hash(); |
2d8f3038 | 2975 | chroot_fs_refs(&root, &new); |
84d17192 | 2976 | put_mountpoint(root_mp); |
1da177e4 | 2977 | error = 0; |
b12cea91 | 2978 | out4: |
84d17192 | 2979 | unlock_mount(old_mp); |
b12cea91 AV |
2980 | if (!error) { |
2981 | path_put(&root_parent); | |
2982 | path_put(&parent_path); | |
2983 | } | |
2984 | out3: | |
8c3ee42e | 2985 | path_put(&root); |
b12cea91 | 2986 | out2: |
2d8f3038 | 2987 | path_put(&old); |
1da177e4 | 2988 | out1: |
2d8f3038 | 2989 | path_put(&new); |
1da177e4 | 2990 | out0: |
1da177e4 | 2991 | return error; |
1da177e4 LT |
2992 | } |
2993 | ||
2994 | static void __init init_mount_tree(void) | |
2995 | { | |
2996 | struct vfsmount *mnt; | |
6b3286ed | 2997 | struct mnt_namespace *ns; |
ac748a09 | 2998 | struct path root; |
0c55cfc4 | 2999 | struct file_system_type *type; |
1da177e4 | 3000 | |
0c55cfc4 EB |
3001 | type = get_fs_type("rootfs"); |
3002 | if (!type) | |
3003 | panic("Can't find rootfs type"); | |
3004 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3005 | put_filesystem(type); | |
1da177e4 LT |
3006 | if (IS_ERR(mnt)) |
3007 | panic("Can't create rootfs"); | |
b3e19d92 | 3008 | |
3b22edc5 TM |
3009 | ns = create_mnt_ns(mnt); |
3010 | if (IS_ERR(ns)) | |
1da177e4 | 3011 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3012 | |
3013 | init_task.nsproxy->mnt_ns = ns; | |
3014 | get_mnt_ns(ns); | |
3015 | ||
be08d6d2 AV |
3016 | root.mnt = mnt; |
3017 | root.dentry = mnt->mnt_root; | |
da362b09 | 3018 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3019 | |
3020 | set_fs_pwd(current->fs, &root); | |
3021 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3022 | } |
3023 | ||
74bf17cf | 3024 | void __init mnt_init(void) |
1da177e4 | 3025 | { |
13f14b4d | 3026 | unsigned u; |
15a67dd8 | 3027 | int err; |
1da177e4 | 3028 | |
7d6fec45 | 3029 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3030 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3031 | |
0818bf27 | 3032 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3033 | sizeof(struct hlist_head), |
0818bf27 AV |
3034 | mhash_entries, 19, |
3035 | 0, | |
3036 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3037 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3038 | sizeof(struct hlist_head), | |
3039 | mphash_entries, 19, | |
3040 | 0, | |
3041 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3042 | |
84d17192 | 3043 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3044 | panic("Failed to allocate mount hash table\n"); |
3045 | ||
0818bf27 | 3046 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3047 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3048 | for (u = 0; u <= mp_hash_mask; u++) |
3049 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3050 | |
4b93dc9b TH |
3051 | kernfs_init(); |
3052 | ||
15a67dd8 RD |
3053 | err = sysfs_init(); |
3054 | if (err) | |
3055 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3056 | __func__, err); |
00d26666 GKH |
3057 | fs_kobj = kobject_create_and_add("fs", NULL); |
3058 | if (!fs_kobj) | |
8e24eea7 | 3059 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3060 | init_rootfs(); |
3061 | init_mount_tree(); | |
3062 | } | |
3063 | ||
616511d0 | 3064 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3065 | { |
d498b25a | 3066 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3067 | return; |
7b00ed6f | 3068 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3069 | free_mnt_ns(ns); |
1da177e4 | 3070 | } |
9d412a43 AV |
3071 | |
3072 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3073 | { | |
423e0ab0 TC |
3074 | struct vfsmount *mnt; |
3075 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3076 | if (!IS_ERR(mnt)) { | |
3077 | /* | |
3078 | * it is a longterm mount, don't release mnt until | |
3079 | * we unmount before file sys is unregistered | |
3080 | */ | |
f7a99c5b | 3081 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3082 | } |
3083 | return mnt; | |
9d412a43 AV |
3084 | } |
3085 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3086 | |
3087 | void kern_unmount(struct vfsmount *mnt) | |
3088 | { | |
3089 | /* release long term mount so mount point can be released */ | |
3090 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3091 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3092 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3093 | mntput(mnt); |
3094 | } | |
3095 | } | |
3096 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3097 | |
3098 | bool our_mnt(struct vfsmount *mnt) | |
3099 | { | |
143c8c91 | 3100 | return check_mnt(real_mount(mnt)); |
02125a82 | 3101 | } |
8823c079 | 3102 | |
3151527e EB |
3103 | bool current_chrooted(void) |
3104 | { | |
3105 | /* Does the current process have a non-standard root */ | |
3106 | struct path ns_root; | |
3107 | struct path fs_root; | |
3108 | bool chrooted; | |
3109 | ||
3110 | /* Find the namespace root */ | |
3111 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3112 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3113 | path_get(&ns_root); | |
3114 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3115 | ; | |
3116 | ||
3117 | get_fs_root(current->fs, &fs_root); | |
3118 | ||
3119 | chrooted = !path_equal(&fs_root, &ns_root); | |
3120 | ||
3121 | path_put(&fs_root); | |
3122 | path_put(&ns_root); | |
3123 | ||
3124 | return chrooted; | |
3125 | } | |
3126 | ||
e51db735 | 3127 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
3128 | { |
3129 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
3130 | struct mount *mnt; | |
e51db735 | 3131 | bool visible = false; |
87a8ebd6 | 3132 | |
e51db735 EB |
3133 | if (unlikely(!ns)) |
3134 | return false; | |
3135 | ||
44bb4385 | 3136 | down_read(&namespace_sem); |
87a8ebd6 | 3137 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
3138 | struct mount *child; |
3139 | if (mnt->mnt.mnt_sb->s_type != type) | |
3140 | continue; | |
3141 | ||
3142 | /* This mount is not fully visible if there are any child mounts | |
3143 | * that cover anything except for empty directories. | |
3144 | */ | |
3145 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3146 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
3147 | if (!S_ISDIR(inode->i_mode)) | |
3148 | goto next; | |
41301ae7 | 3149 | if (inode->i_nlink > 2) |
e51db735 | 3150 | goto next; |
87a8ebd6 | 3151 | } |
e51db735 EB |
3152 | visible = true; |
3153 | goto found; | |
3154 | next: ; | |
87a8ebd6 | 3155 | } |
e51db735 | 3156 | found: |
44bb4385 | 3157 | up_read(&namespace_sem); |
e51db735 | 3158 | return visible; |
87a8ebd6 EB |
3159 | } |
3160 | ||
64964528 | 3161 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3162 | { |
58be2825 | 3163 | struct ns_common *ns = NULL; |
8823c079 EB |
3164 | struct nsproxy *nsproxy; |
3165 | ||
728dba3a EB |
3166 | task_lock(task); |
3167 | nsproxy = task->nsproxy; | |
8823c079 | 3168 | if (nsproxy) { |
58be2825 AV |
3169 | ns = &nsproxy->mnt_ns->ns; |
3170 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3171 | } |
728dba3a | 3172 | task_unlock(task); |
8823c079 EB |
3173 | |
3174 | return ns; | |
3175 | } | |
3176 | ||
64964528 | 3177 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3178 | { |
58be2825 | 3179 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3180 | } |
3181 | ||
64964528 | 3182 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3183 | { |
3184 | struct fs_struct *fs = current->fs; | |
58be2825 | 3185 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3186 | struct path root; |
3187 | ||
0c55cfc4 | 3188 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3189 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3190 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3191 | return -EPERM; |
8823c079 EB |
3192 | |
3193 | if (fs->users != 1) | |
3194 | return -EINVAL; | |
3195 | ||
3196 | get_mnt_ns(mnt_ns); | |
3197 | put_mnt_ns(nsproxy->mnt_ns); | |
3198 | nsproxy->mnt_ns = mnt_ns; | |
3199 | ||
3200 | /* Find the root */ | |
3201 | root.mnt = &mnt_ns->root->mnt; | |
3202 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3203 | path_get(&root); | |
3204 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3205 | ; | |
3206 | ||
3207 | /* Update the pwd and root */ | |
3208 | set_fs_pwd(fs, &root); | |
3209 | set_fs_root(fs, &root); | |
3210 | ||
3211 | path_put(&root); | |
3212 | return 0; | |
3213 | } | |
3214 | ||
3215 | const struct proc_ns_operations mntns_operations = { | |
3216 | .name = "mnt", | |
3217 | .type = CLONE_NEWNS, | |
3218 | .get = mntns_get, | |
3219 | .put = mntns_put, | |
3220 | .install = mntns_install, | |
3221 | }; |