]>
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 | |
fd4b5fa6 EB |
30 | /* Maximum number of mounts in a mount namespace */ |
31 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
32 | ||
0818bf27 AV |
33 | static unsigned int m_hash_mask __read_mostly; |
34 | static unsigned int m_hash_shift __read_mostly; | |
35 | static unsigned int mp_hash_mask __read_mostly; | |
36 | static unsigned int mp_hash_shift __read_mostly; | |
37 | ||
38 | static __initdata unsigned long mhash_entries; | |
39 | static int __init set_mhash_entries(char *str) | |
40 | { | |
41 | if (!str) | |
42 | return 0; | |
43 | mhash_entries = simple_strtoul(str, &str, 0); | |
44 | return 1; | |
45 | } | |
46 | __setup("mhash_entries=", set_mhash_entries); | |
47 | ||
48 | static __initdata unsigned long mphash_entries; | |
49 | static int __init set_mphash_entries(char *str) | |
50 | { | |
51 | if (!str) | |
52 | return 0; | |
53 | mphash_entries = simple_strtoul(str, &str, 0); | |
54 | return 1; | |
55 | } | |
56 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 57 | |
c7999c36 | 58 | static u64 event; |
73cd49ec | 59 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 60 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 61 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
62 | static int mnt_id_start = 0; |
63 | static int mnt_group_start = 1; | |
1da177e4 | 64 | |
38129a13 | 65 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 66 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 67 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 68 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 69 | |
f87fd4c2 | 70 | /* /sys/fs */ |
00d26666 GKH |
71 | struct kobject *fs_kobj; |
72 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 73 | |
99b7db7b NP |
74 | /* |
75 | * vfsmount lock may be taken for read to prevent changes to the | |
76 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
77 | * up the tree. | |
78 | * | |
79 | * It should be taken for write in all cases where the vfsmount | |
80 | * tree or hash is modified or when a vfsmount structure is modified. | |
81 | */ | |
48a066e7 | 82 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 83 | |
38129a13 | 84 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 85 | { |
b58fed8b RP |
86 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
87 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
88 | tmp = tmp + (tmp >> m_hash_shift); |
89 | return &mount_hashtable[tmp & m_hash_mask]; | |
90 | } | |
91 | ||
92 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
93 | { | |
94 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
95 | tmp = tmp + (tmp >> mp_hash_shift); | |
96 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
97 | } |
98 | ||
99b7db7b NP |
99 | /* |
100 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
101 | * serialize with freeing. | |
102 | */ | |
b105e270 | 103 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
104 | { |
105 | int res; | |
106 | ||
107 | retry: | |
108 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 109 | spin_lock(&mnt_id_lock); |
15169fe7 | 110 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 111 | if (!res) |
15169fe7 | 112 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 113 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
114 | if (res == -EAGAIN) |
115 | goto retry; | |
116 | ||
117 | return res; | |
118 | } | |
119 | ||
b105e270 | 120 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 121 | { |
15169fe7 | 122 | int id = mnt->mnt_id; |
99b7db7b | 123 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
124 | ida_remove(&mnt_id_ida, id); |
125 | if (mnt_id_start > id) | |
126 | mnt_id_start = id; | |
99b7db7b | 127 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
128 | } |
129 | ||
719f5d7f MS |
130 | /* |
131 | * Allocate a new peer group ID | |
132 | * | |
133 | * mnt_group_ida is protected by namespace_sem | |
134 | */ | |
4b8b21f4 | 135 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 136 | { |
f21f6220 AV |
137 | int res; |
138 | ||
719f5d7f MS |
139 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
140 | return -ENOMEM; | |
141 | ||
f21f6220 AV |
142 | res = ida_get_new_above(&mnt_group_ida, |
143 | mnt_group_start, | |
15169fe7 | 144 | &mnt->mnt_group_id); |
f21f6220 | 145 | if (!res) |
15169fe7 | 146 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
147 | |
148 | return res; | |
719f5d7f MS |
149 | } |
150 | ||
151 | /* | |
152 | * Release a peer group ID | |
153 | */ | |
4b8b21f4 | 154 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 155 | { |
15169fe7 | 156 | int id = mnt->mnt_group_id; |
f21f6220 AV |
157 | ida_remove(&mnt_group_ida, id); |
158 | if (mnt_group_start > id) | |
159 | mnt_group_start = id; | |
15169fe7 | 160 | mnt->mnt_group_id = 0; |
719f5d7f MS |
161 | } |
162 | ||
b3e19d92 NP |
163 | /* |
164 | * vfsmount lock must be held for read | |
165 | */ | |
83adc753 | 166 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
167 | { |
168 | #ifdef CONFIG_SMP | |
68e8a9fe | 169 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
170 | #else |
171 | preempt_disable(); | |
68e8a9fe | 172 | mnt->mnt_count += n; |
b3e19d92 NP |
173 | preempt_enable(); |
174 | #endif | |
175 | } | |
176 | ||
b3e19d92 NP |
177 | /* |
178 | * vfsmount lock must be held for write | |
179 | */ | |
83adc753 | 180 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
181 | { |
182 | #ifdef CONFIG_SMP | |
f03c6599 | 183 | unsigned int count = 0; |
b3e19d92 NP |
184 | int cpu; |
185 | ||
186 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 187 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
188 | } |
189 | ||
190 | return count; | |
191 | #else | |
68e8a9fe | 192 | return mnt->mnt_count; |
b3e19d92 NP |
193 | #endif |
194 | } | |
195 | ||
87b95ce0 AV |
196 | static void drop_mountpoint(struct fs_pin *p) |
197 | { | |
198 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
199 | dput(m->mnt_ex_mountpoint); | |
200 | pin_remove(p); | |
201 | mntput(&m->mnt); | |
202 | } | |
203 | ||
b105e270 | 204 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 205 | { |
c63181e6 AV |
206 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
207 | if (mnt) { | |
73cd49ec MS |
208 | int err; |
209 | ||
c63181e6 | 210 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
211 | if (err) |
212 | goto out_free_cache; | |
213 | ||
214 | if (name) { | |
fcc139ae | 215 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 216 | if (!mnt->mnt_devname) |
88b38782 | 217 | goto out_free_id; |
73cd49ec MS |
218 | } |
219 | ||
b3e19d92 | 220 | #ifdef CONFIG_SMP |
c63181e6 AV |
221 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
222 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
223 | goto out_free_devname; |
224 | ||
c63181e6 | 225 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 226 | #else |
c63181e6 AV |
227 | mnt->mnt_count = 1; |
228 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
229 | #endif |
230 | ||
38129a13 | 231 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
232 | INIT_LIST_HEAD(&mnt->mnt_child); |
233 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
234 | INIT_LIST_HEAD(&mnt->mnt_list); | |
235 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
236 | INIT_LIST_HEAD(&mnt->mnt_share); | |
237 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
238 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 239 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
240 | #ifdef CONFIG_FSNOTIFY |
241 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 242 | #endif |
87b95ce0 | 243 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 244 | } |
c63181e6 | 245 | return mnt; |
88b38782 | 246 | |
d3ef3d73 | 247 | #ifdef CONFIG_SMP |
248 | out_free_devname: | |
fcc139ae | 249 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 250 | #endif |
88b38782 | 251 | out_free_id: |
c63181e6 | 252 | mnt_free_id(mnt); |
88b38782 | 253 | out_free_cache: |
c63181e6 | 254 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 255 | return NULL; |
1da177e4 LT |
256 | } |
257 | ||
3d733633 DH |
258 | /* |
259 | * Most r/o checks on a fs are for operations that take | |
260 | * discrete amounts of time, like a write() or unlink(). | |
261 | * We must keep track of when those operations start | |
262 | * (for permission checks) and when they end, so that | |
263 | * we can determine when writes are able to occur to | |
264 | * a filesystem. | |
265 | */ | |
266 | /* | |
267 | * __mnt_is_readonly: check whether a mount is read-only | |
268 | * @mnt: the mount to check for its write status | |
269 | * | |
270 | * This shouldn't be used directly ouside of the VFS. | |
271 | * It does not guarantee that the filesystem will stay | |
272 | * r/w, just that it is right *now*. This can not and | |
273 | * should not be used in place of IS_RDONLY(inode). | |
274 | * mnt_want/drop_write() will _keep_ the filesystem | |
275 | * r/w. | |
276 | */ | |
277 | int __mnt_is_readonly(struct vfsmount *mnt) | |
278 | { | |
2e4b7fcd DH |
279 | if (mnt->mnt_flags & MNT_READONLY) |
280 | return 1; | |
281 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
282 | return 1; | |
283 | return 0; | |
3d733633 DH |
284 | } |
285 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
286 | ||
83adc753 | 287 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 288 | { |
289 | #ifdef CONFIG_SMP | |
68e8a9fe | 290 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 291 | #else |
68e8a9fe | 292 | mnt->mnt_writers++; |
d3ef3d73 | 293 | #endif |
294 | } | |
3d733633 | 295 | |
83adc753 | 296 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 297 | { |
d3ef3d73 | 298 | #ifdef CONFIG_SMP |
68e8a9fe | 299 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 300 | #else |
68e8a9fe | 301 | mnt->mnt_writers--; |
d3ef3d73 | 302 | #endif |
3d733633 | 303 | } |
3d733633 | 304 | |
83adc753 | 305 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 306 | { |
d3ef3d73 | 307 | #ifdef CONFIG_SMP |
308 | unsigned int count = 0; | |
3d733633 | 309 | int cpu; |
3d733633 DH |
310 | |
311 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 312 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 313 | } |
3d733633 | 314 | |
d3ef3d73 | 315 | return count; |
316 | #else | |
317 | return mnt->mnt_writers; | |
318 | #endif | |
3d733633 DH |
319 | } |
320 | ||
4ed5e82f MS |
321 | static int mnt_is_readonly(struct vfsmount *mnt) |
322 | { | |
323 | if (mnt->mnt_sb->s_readonly_remount) | |
324 | return 1; | |
325 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
326 | smp_rmb(); | |
327 | return __mnt_is_readonly(mnt); | |
328 | } | |
329 | ||
8366025e | 330 | /* |
eb04c282 JK |
331 | * Most r/o & frozen checks on a fs are for operations that take discrete |
332 | * amounts of time, like a write() or unlink(). We must keep track of when | |
333 | * those operations start (for permission checks) and when they end, so that we | |
334 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
335 | */ |
336 | /** | |
eb04c282 | 337 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 338 | * @m: the mount on which to take a write |
8366025e | 339 | * |
eb04c282 JK |
340 | * This tells the low-level filesystem that a write is about to be performed to |
341 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
342 | * returning success. This operation does not protect against filesystem being | |
343 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
344 | * called. This is effectively a refcount. | |
8366025e | 345 | */ |
eb04c282 | 346 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 347 | { |
83adc753 | 348 | struct mount *mnt = real_mount(m); |
3d733633 | 349 | int ret = 0; |
3d733633 | 350 | |
d3ef3d73 | 351 | preempt_disable(); |
c6653a83 | 352 | mnt_inc_writers(mnt); |
d3ef3d73 | 353 | /* |
c6653a83 | 354 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 355 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
356 | * incremented count after it has set MNT_WRITE_HOLD. | |
357 | */ | |
358 | smp_mb(); | |
1e75529e | 359 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 360 | cpu_relax(); |
361 | /* | |
362 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
363 | * be set to match its requirements. So we must not load that until | |
364 | * MNT_WRITE_HOLD is cleared. | |
365 | */ | |
366 | smp_rmb(); | |
4ed5e82f | 367 | if (mnt_is_readonly(m)) { |
c6653a83 | 368 | mnt_dec_writers(mnt); |
3d733633 | 369 | ret = -EROFS; |
3d733633 | 370 | } |
d3ef3d73 | 371 | preempt_enable(); |
eb04c282 JK |
372 | |
373 | return ret; | |
374 | } | |
375 | ||
376 | /** | |
377 | * mnt_want_write - get write access to a mount | |
378 | * @m: the mount on which to take a write | |
379 | * | |
380 | * This tells the low-level filesystem that a write is about to be performed to | |
381 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
382 | * is not frozen) before returning success. When the write operation is | |
383 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
384 | */ | |
385 | int mnt_want_write(struct vfsmount *m) | |
386 | { | |
387 | int ret; | |
388 | ||
389 | sb_start_write(m->mnt_sb); | |
390 | ret = __mnt_want_write(m); | |
391 | if (ret) | |
392 | sb_end_write(m->mnt_sb); | |
3d733633 | 393 | return ret; |
8366025e DH |
394 | } |
395 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
396 | ||
96029c4e | 397 | /** |
398 | * mnt_clone_write - get write access to a mount | |
399 | * @mnt: the mount on which to take a write | |
400 | * | |
401 | * This is effectively like mnt_want_write, except | |
402 | * it must only be used to take an extra write reference | |
403 | * on a mountpoint that we already know has a write reference | |
404 | * on it. This allows some optimisation. | |
405 | * | |
406 | * After finished, mnt_drop_write must be called as usual to | |
407 | * drop the reference. | |
408 | */ | |
409 | int mnt_clone_write(struct vfsmount *mnt) | |
410 | { | |
411 | /* superblock may be r/o */ | |
412 | if (__mnt_is_readonly(mnt)) | |
413 | return -EROFS; | |
414 | preempt_disable(); | |
83adc753 | 415 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 416 | preempt_enable(); |
417 | return 0; | |
418 | } | |
419 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
420 | ||
421 | /** | |
eb04c282 | 422 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 423 | * @file: the file who's mount on which to take a write |
424 | * | |
eb04c282 | 425 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 426 | * do some optimisations if the file is open for write already |
427 | */ | |
eb04c282 | 428 | int __mnt_want_write_file(struct file *file) |
96029c4e | 429 | { |
83f936c7 | 430 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 431 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 432 | else |
433 | return mnt_clone_write(file->f_path.mnt); | |
434 | } | |
eb04c282 JK |
435 | |
436 | /** | |
437 | * mnt_want_write_file - get write access to a file's mount | |
438 | * @file: the file who's mount on which to take a write | |
439 | * | |
440 | * This is like mnt_want_write, but it takes a file and can | |
441 | * do some optimisations if the file is open for write already | |
442 | */ | |
443 | int mnt_want_write_file(struct file *file) | |
444 | { | |
445 | int ret; | |
446 | ||
447 | sb_start_write(file->f_path.mnt->mnt_sb); | |
448 | ret = __mnt_want_write_file(file); | |
449 | if (ret) | |
450 | sb_end_write(file->f_path.mnt->mnt_sb); | |
451 | return ret; | |
452 | } | |
96029c4e | 453 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
454 | ||
8366025e | 455 | /** |
eb04c282 | 456 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
457 | * @mnt: the mount on which to give up write access |
458 | * | |
459 | * Tells the low-level filesystem that we are done | |
460 | * performing writes to it. Must be matched with | |
eb04c282 | 461 | * __mnt_want_write() call above. |
8366025e | 462 | */ |
eb04c282 | 463 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 464 | { |
d3ef3d73 | 465 | preempt_disable(); |
83adc753 | 466 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 467 | preempt_enable(); |
8366025e | 468 | } |
dee5220c | 469 | EXPORT_SYMBOL_GPL(__mnt_drop_write); |
eb04c282 JK |
470 | |
471 | /** | |
472 | * mnt_drop_write - give up write access to a mount | |
473 | * @mnt: the mount on which to give up write access | |
474 | * | |
475 | * Tells the low-level filesystem that we are done performing writes to it and | |
476 | * also allows filesystem to be frozen again. Must be matched with | |
477 | * mnt_want_write() call above. | |
478 | */ | |
479 | void mnt_drop_write(struct vfsmount *mnt) | |
480 | { | |
481 | __mnt_drop_write(mnt); | |
482 | sb_end_write(mnt->mnt_sb); | |
483 | } | |
8366025e DH |
484 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
485 | ||
eb04c282 JK |
486 | void __mnt_drop_write_file(struct file *file) |
487 | { | |
488 | __mnt_drop_write(file->f_path.mnt); | |
489 | } | |
490 | ||
2a79f17e AV |
491 | void mnt_drop_write_file(struct file *file) |
492 | { | |
493 | mnt_drop_write(file->f_path.mnt); | |
494 | } | |
495 | EXPORT_SYMBOL(mnt_drop_write_file); | |
496 | ||
83adc753 | 497 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 498 | { |
3d733633 DH |
499 | int ret = 0; |
500 | ||
719ea2fb | 501 | lock_mount_hash(); |
83adc753 | 502 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 503 | /* |
d3ef3d73 | 504 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
505 | * should be visible before we do. | |
3d733633 | 506 | */ |
d3ef3d73 | 507 | smp_mb(); |
508 | ||
3d733633 | 509 | /* |
d3ef3d73 | 510 | * With writers on hold, if this value is zero, then there are |
511 | * definitely no active writers (although held writers may subsequently | |
512 | * increment the count, they'll have to wait, and decrement it after | |
513 | * seeing MNT_READONLY). | |
514 | * | |
515 | * It is OK to have counter incremented on one CPU and decremented on | |
516 | * another: the sum will add up correctly. The danger would be when we | |
517 | * sum up each counter, if we read a counter before it is incremented, | |
518 | * but then read another CPU's count which it has been subsequently | |
519 | * decremented from -- we would see more decrements than we should. | |
520 | * MNT_WRITE_HOLD protects against this scenario, because | |
521 | * mnt_want_write first increments count, then smp_mb, then spins on | |
522 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
523 | * we're counting up here. | |
3d733633 | 524 | */ |
c6653a83 | 525 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 526 | ret = -EBUSY; |
527 | else | |
83adc753 | 528 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 529 | /* |
530 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
531 | * that become unheld will see MNT_READONLY. | |
532 | */ | |
533 | smp_wmb(); | |
83adc753 | 534 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 535 | unlock_mount_hash(); |
3d733633 | 536 | return ret; |
8366025e | 537 | } |
8366025e | 538 | |
83adc753 | 539 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 540 | { |
719ea2fb | 541 | lock_mount_hash(); |
83adc753 | 542 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 543 | unlock_mount_hash(); |
2e4b7fcd DH |
544 | } |
545 | ||
4ed5e82f MS |
546 | int sb_prepare_remount_readonly(struct super_block *sb) |
547 | { | |
548 | struct mount *mnt; | |
549 | int err = 0; | |
550 | ||
8e8b8796 MS |
551 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
552 | if (atomic_long_read(&sb->s_remove_count)) | |
553 | return -EBUSY; | |
554 | ||
719ea2fb | 555 | lock_mount_hash(); |
4ed5e82f MS |
556 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
557 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
558 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
559 | smp_mb(); | |
560 | if (mnt_get_writers(mnt) > 0) { | |
561 | err = -EBUSY; | |
562 | break; | |
563 | } | |
564 | } | |
565 | } | |
8e8b8796 MS |
566 | if (!err && atomic_long_read(&sb->s_remove_count)) |
567 | err = -EBUSY; | |
568 | ||
4ed5e82f MS |
569 | if (!err) { |
570 | sb->s_readonly_remount = 1; | |
571 | smp_wmb(); | |
572 | } | |
573 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
574 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
575 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
576 | } | |
719ea2fb | 577 | unlock_mount_hash(); |
4ed5e82f MS |
578 | |
579 | return err; | |
580 | } | |
581 | ||
b105e270 | 582 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 583 | { |
fcc139ae | 584 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 585 | #ifdef CONFIG_SMP |
68e8a9fe | 586 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 587 | #endif |
b105e270 | 588 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
589 | } |
590 | ||
8ffcb32e DH |
591 | static void delayed_free_vfsmnt(struct rcu_head *head) |
592 | { | |
593 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
594 | } | |
595 | ||
48a066e7 | 596 | /* call under rcu_read_lock */ |
294d71ff | 597 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
598 | { |
599 | struct mount *mnt; | |
600 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 601 | return 1; |
48a066e7 | 602 | if (bastard == NULL) |
294d71ff | 603 | return 0; |
48a066e7 AV |
604 | mnt = real_mount(bastard); |
605 | mnt_add_count(mnt, 1); | |
606 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 607 | return 0; |
48a066e7 AV |
608 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
609 | mnt_add_count(mnt, -1); | |
294d71ff AV |
610 | return 1; |
611 | } | |
612 | return -1; | |
613 | } | |
614 | ||
615 | /* call under rcu_read_lock */ | |
616 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
617 | { | |
618 | int res = __legitimize_mnt(bastard, seq); | |
619 | if (likely(!res)) | |
620 | return true; | |
621 | if (unlikely(res < 0)) { | |
622 | rcu_read_unlock(); | |
623 | mntput(bastard); | |
624 | rcu_read_lock(); | |
48a066e7 | 625 | } |
48a066e7 AV |
626 | return false; |
627 | } | |
628 | ||
1da177e4 | 629 | /* |
474279dc | 630 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 631 | * call under rcu_read_lock() |
1da177e4 | 632 | */ |
474279dc | 633 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 634 | { |
38129a13 | 635 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
636 | struct mount *p; |
637 | ||
38129a13 | 638 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
639 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
640 | return p; | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | /* | |
645 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 646 | * mount_lock must be held. |
474279dc AV |
647 | */ |
648 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
649 | { | |
411a938b EB |
650 | struct mount *p, *res = NULL; |
651 | p = __lookup_mnt(mnt, dentry); | |
38129a13 AV |
652 | if (!p) |
653 | goto out; | |
411a938b EB |
654 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
655 | res = p; | |
38129a13 | 656 | hlist_for_each_entry_continue(p, mnt_hash) { |
1d6a32ac AV |
657 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
658 | break; | |
411a938b EB |
659 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
660 | res = p; | |
1d6a32ac | 661 | } |
38129a13 | 662 | out: |
1d6a32ac | 663 | return res; |
1da177e4 LT |
664 | } |
665 | ||
a05964f3 | 666 | /* |
f015f126 DH |
667 | * lookup_mnt - Return the first child mount mounted at path |
668 | * | |
669 | * "First" means first mounted chronologically. If you create the | |
670 | * following mounts: | |
671 | * | |
672 | * mount /dev/sda1 /mnt | |
673 | * mount /dev/sda2 /mnt | |
674 | * mount /dev/sda3 /mnt | |
675 | * | |
676 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
677 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
678 | * /dev/sda2, then /dev/sda3, then NULL. | |
679 | * | |
680 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 681 | */ |
1c755af4 | 682 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 683 | { |
c7105365 | 684 | struct mount *child_mnt; |
48a066e7 AV |
685 | struct vfsmount *m; |
686 | unsigned seq; | |
99b7db7b | 687 | |
48a066e7 AV |
688 | rcu_read_lock(); |
689 | do { | |
690 | seq = read_seqbegin(&mount_lock); | |
691 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
692 | m = child_mnt ? &child_mnt->mnt : NULL; | |
693 | } while (!legitimize_mnt(m, seq)); | |
694 | rcu_read_unlock(); | |
695 | return m; | |
a05964f3 RP |
696 | } |
697 | ||
7af1364f EB |
698 | /* |
699 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
700 | * current mount namespace. | |
701 | * | |
702 | * The common case is dentries are not mountpoints at all and that | |
703 | * test is handled inline. For the slow case when we are actually | |
704 | * dealing with a mountpoint of some kind, walk through all of the | |
705 | * mounts in the current mount namespace and test to see if the dentry | |
706 | * is a mountpoint. | |
707 | * | |
708 | * The mount_hashtable is not usable in the context because we | |
709 | * need to identify all mounts that may be in the current mount | |
710 | * namespace not just a mount that happens to have some specified | |
711 | * parent mount. | |
712 | */ | |
713 | bool __is_local_mountpoint(struct dentry *dentry) | |
714 | { | |
715 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
716 | struct mount *mnt; | |
717 | bool is_covered = false; | |
718 | ||
719 | if (!d_mountpoint(dentry)) | |
720 | goto out; | |
721 | ||
722 | down_read(&namespace_sem); | |
723 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
724 | is_covered = (mnt->mnt_mountpoint == dentry); | |
725 | if (is_covered) | |
726 | break; | |
727 | } | |
728 | up_read(&namespace_sem); | |
729 | out: | |
730 | return is_covered; | |
731 | } | |
732 | ||
e2dfa935 | 733 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 734 | { |
0818bf27 | 735 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
736 | struct mountpoint *mp; |
737 | ||
0818bf27 | 738 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
739 | if (mp->m_dentry == dentry) { |
740 | /* might be worth a WARN_ON() */ | |
741 | if (d_unlinked(dentry)) | |
742 | return ERR_PTR(-ENOENT); | |
743 | mp->m_count++; | |
744 | return mp; | |
745 | } | |
746 | } | |
e2dfa935 EB |
747 | return NULL; |
748 | } | |
749 | ||
e178bd11 | 750 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 751 | { |
e178bd11 | 752 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 753 | int ret; |
84d17192 | 754 | |
e178bd11 EB |
755 | if (d_mountpoint(dentry)) { |
756 | mountpoint: | |
757 | read_seqlock_excl(&mount_lock); | |
758 | mp = lookup_mountpoint(dentry); | |
759 | read_sequnlock_excl(&mount_lock); | |
760 | if (mp) | |
761 | goto done; | |
762 | } | |
763 | ||
764 | if (!new) | |
765 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
766 | if (!new) | |
84d17192 AV |
767 | return ERR_PTR(-ENOMEM); |
768 | ||
e178bd11 EB |
769 | |
770 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 771 | ret = d_set_mounted(dentry); |
eed81007 | 772 | |
e178bd11 EB |
773 | /* Someone else set d_mounted? */ |
774 | if (ret == -EBUSY) | |
775 | goto mountpoint; | |
776 | ||
777 | /* The dentry is not available as a mountpoint? */ | |
778 | mp = ERR_PTR(ret); | |
779 | if (ret) | |
780 | goto done; | |
781 | ||
782 | /* Add the new mountpoint to the hash table */ | |
783 | read_seqlock_excl(&mount_lock); | |
784 | new->m_dentry = dentry; | |
785 | new->m_count = 1; | |
786 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
787 | INIT_HLIST_HEAD(&new->m_list); | |
788 | read_sequnlock_excl(&mount_lock); | |
789 | ||
790 | mp = new; | |
791 | new = NULL; | |
792 | done: | |
793 | kfree(new); | |
84d17192 AV |
794 | return mp; |
795 | } | |
796 | ||
797 | static void put_mountpoint(struct mountpoint *mp) | |
798 | { | |
799 | if (!--mp->m_count) { | |
800 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 801 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
802 | spin_lock(&dentry->d_lock); |
803 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
804 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 805 | hlist_del(&mp->m_hash); |
84d17192 AV |
806 | kfree(mp); |
807 | } | |
808 | } | |
809 | ||
143c8c91 | 810 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 811 | { |
6b3286ed | 812 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
813 | } |
814 | ||
99b7db7b NP |
815 | /* |
816 | * vfsmount lock must be held for write | |
817 | */ | |
6b3286ed | 818 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
819 | { |
820 | if (ns) { | |
821 | ns->event = ++event; | |
822 | wake_up_interruptible(&ns->poll); | |
823 | } | |
824 | } | |
825 | ||
99b7db7b NP |
826 | /* |
827 | * vfsmount lock must be held for write | |
828 | */ | |
6b3286ed | 829 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
830 | { |
831 | if (ns && ns->event != event) { | |
832 | ns->event = event; | |
833 | wake_up_interruptible(&ns->poll); | |
834 | } | |
835 | } | |
836 | ||
99b7db7b NP |
837 | /* |
838 | * vfsmount lock must be held for write | |
839 | */ | |
7bdb11de | 840 | static void unhash_mnt(struct mount *mnt) |
419148da | 841 | { |
0714a533 | 842 | mnt->mnt_parent = mnt; |
a73324da | 843 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 844 | list_del_init(&mnt->mnt_child); |
38129a13 | 845 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 846 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
847 | put_mountpoint(mnt->mnt_mp); |
848 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
849 | } |
850 | ||
7bdb11de EB |
851 | /* |
852 | * vfsmount lock must be held for write | |
853 | */ | |
854 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
855 | { | |
856 | old_path->dentry = mnt->mnt_mountpoint; | |
857 | old_path->mnt = &mnt->mnt_parent->mnt; | |
858 | unhash_mnt(mnt); | |
859 | } | |
860 | ||
6a46c573 EB |
861 | /* |
862 | * vfsmount lock must be held for write | |
863 | */ | |
864 | static void umount_mnt(struct mount *mnt) | |
865 | { | |
866 | /* old mountpoint will be dropped when we can do that */ | |
867 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
868 | unhash_mnt(mnt); | |
869 | } | |
870 | ||
99b7db7b NP |
871 | /* |
872 | * vfsmount lock must be held for write | |
873 | */ | |
84d17192 AV |
874 | void mnt_set_mountpoint(struct mount *mnt, |
875 | struct mountpoint *mp, | |
44d964d6 | 876 | struct mount *child_mnt) |
b90fa9ae | 877 | { |
84d17192 | 878 | mp->m_count++; |
3a2393d7 | 879 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 880 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 881 | child_mnt->mnt_parent = mnt; |
84d17192 | 882 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 883 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
884 | } |
885 | ||
99b7db7b NP |
886 | /* |
887 | * vfsmount lock must be held for write | |
888 | */ | |
84d17192 AV |
889 | static void attach_mnt(struct mount *mnt, |
890 | struct mount *parent, | |
891 | struct mountpoint *mp) | |
1da177e4 | 892 | { |
84d17192 | 893 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 894 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 895 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
896 | } |
897 | ||
12a5b529 AV |
898 | static void attach_shadowed(struct mount *mnt, |
899 | struct mount *parent, | |
900 | struct mount *shadows) | |
901 | { | |
902 | if (shadows) { | |
f6f99332 | 903 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
904 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
905 | } else { | |
906 | hlist_add_head_rcu(&mnt->mnt_hash, | |
907 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
908 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
909 | } | |
910 | } | |
911 | ||
b90fa9ae | 912 | /* |
99b7db7b | 913 | * vfsmount lock must be held for write |
b90fa9ae | 914 | */ |
1d6a32ac | 915 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 916 | { |
0714a533 | 917 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 918 | struct mount *m; |
b90fa9ae | 919 | LIST_HEAD(head); |
143c8c91 | 920 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 921 | |
0714a533 | 922 | BUG_ON(parent == mnt); |
b90fa9ae | 923 | |
1a4eeaf2 | 924 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 925 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 926 | m->mnt_ns = n; |
f03c6599 | 927 | |
b90fa9ae RP |
928 | list_splice(&head, n->list.prev); |
929 | ||
fd4b5fa6 EB |
930 | n->mounts += n->pending_mounts; |
931 | n->pending_mounts = 0; | |
932 | ||
12a5b529 | 933 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 934 | touch_mnt_namespace(n); |
1da177e4 LT |
935 | } |
936 | ||
909b0a88 | 937 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 938 | { |
6b41d536 AV |
939 | struct list_head *next = p->mnt_mounts.next; |
940 | if (next == &p->mnt_mounts) { | |
1da177e4 | 941 | while (1) { |
909b0a88 | 942 | if (p == root) |
1da177e4 | 943 | return NULL; |
6b41d536 AV |
944 | next = p->mnt_child.next; |
945 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 946 | break; |
0714a533 | 947 | p = p->mnt_parent; |
1da177e4 LT |
948 | } |
949 | } | |
6b41d536 | 950 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
951 | } |
952 | ||
315fc83e | 953 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 954 | { |
6b41d536 AV |
955 | struct list_head *prev = p->mnt_mounts.prev; |
956 | while (prev != &p->mnt_mounts) { | |
957 | p = list_entry(prev, struct mount, mnt_child); | |
958 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
959 | } |
960 | return p; | |
961 | } | |
962 | ||
9d412a43 AV |
963 | struct vfsmount * |
964 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
965 | { | |
b105e270 | 966 | struct mount *mnt; |
9d412a43 AV |
967 | struct dentry *root; |
968 | ||
969 | if (!type) | |
970 | return ERR_PTR(-ENODEV); | |
971 | ||
972 | mnt = alloc_vfsmnt(name); | |
973 | if (!mnt) | |
974 | return ERR_PTR(-ENOMEM); | |
975 | ||
976 | if (flags & MS_KERNMOUNT) | |
b105e270 | 977 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
978 | |
979 | root = mount_fs(type, flags, name, data); | |
980 | if (IS_ERR(root)) { | |
8ffcb32e | 981 | mnt_free_id(mnt); |
9d412a43 AV |
982 | free_vfsmnt(mnt); |
983 | return ERR_CAST(root); | |
984 | } | |
985 | ||
b105e270 AV |
986 | mnt->mnt.mnt_root = root; |
987 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 988 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 989 | mnt->mnt_parent = mnt; |
719ea2fb | 990 | lock_mount_hash(); |
39f7c4db | 991 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 992 | unlock_mount_hash(); |
b105e270 | 993 | return &mnt->mnt; |
9d412a43 AV |
994 | } |
995 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
996 | ||
28ba5876 EB |
997 | struct vfsmount * |
998 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
999 | const char *name, void *data) | |
1000 | { | |
1001 | /* Until it is worked out how to pass the user namespace | |
1002 | * through from the parent mount to the submount don't support | |
1003 | * unprivileged mounts with submounts. | |
1004 | */ | |
1005 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1006 | return ERR_PTR(-EPERM); | |
1007 | ||
1008 | return vfs_kern_mount(type, MS_SUBMOUNT, name, data); | |
1009 | } | |
1010 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1011 | ||
87129cc0 | 1012 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1013 | int flag) |
1da177e4 | 1014 | { |
87129cc0 | 1015 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1016 | struct mount *mnt; |
1017 | int err; | |
1da177e4 | 1018 | |
be34d1a3 DH |
1019 | mnt = alloc_vfsmnt(old->mnt_devname); |
1020 | if (!mnt) | |
1021 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1022 | |
7a472ef4 | 1023 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1024 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1025 | else | |
1026 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1027 | |
be34d1a3 DH |
1028 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1029 | err = mnt_alloc_group_id(mnt); | |
1030 | if (err) | |
1031 | goto out_free; | |
1da177e4 | 1032 | } |
be34d1a3 | 1033 | |
f2ebb3a9 | 1034 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 1035 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1036 | if (flag & CL_UNPRIVILEGED) { |
1037 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1038 | ||
1039 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1040 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1041 | ||
1042 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1043 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1044 | ||
1045 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1046 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1047 | ||
1048 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1049 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1050 | } | |
132c94e3 | 1051 | |
5ff9d8a6 | 1052 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1053 | if ((flag & CL_UNPRIVILEGED) && |
1054 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1055 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1056 | ||
be34d1a3 DH |
1057 | atomic_inc(&sb->s_active); |
1058 | mnt->mnt.mnt_sb = sb; | |
1059 | mnt->mnt.mnt_root = dget(root); | |
1060 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1061 | mnt->mnt_parent = mnt; | |
719ea2fb | 1062 | lock_mount_hash(); |
be34d1a3 | 1063 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1064 | unlock_mount_hash(); |
be34d1a3 | 1065 | |
7a472ef4 EB |
1066 | if ((flag & CL_SLAVE) || |
1067 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1068 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1069 | mnt->mnt_master = old; | |
1070 | CLEAR_MNT_SHARED(mnt); | |
1071 | } else if (!(flag & CL_PRIVATE)) { | |
1072 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1073 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1074 | if (IS_MNT_SLAVE(old)) | |
1075 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1076 | mnt->mnt_master = old->mnt_master; | |
1077 | } | |
1078 | if (flag & CL_MAKE_SHARED) | |
1079 | set_mnt_shared(mnt); | |
1080 | ||
1081 | /* stick the duplicate mount on the same expiry list | |
1082 | * as the original if that was on one */ | |
1083 | if (flag & CL_EXPIRE) { | |
1084 | if (!list_empty(&old->mnt_expire)) | |
1085 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1086 | } | |
1087 | ||
cb338d06 | 1088 | return mnt; |
719f5d7f MS |
1089 | |
1090 | out_free: | |
8ffcb32e | 1091 | mnt_free_id(mnt); |
719f5d7f | 1092 | free_vfsmnt(mnt); |
be34d1a3 | 1093 | return ERR_PTR(err); |
1da177e4 LT |
1094 | } |
1095 | ||
9ea459e1 AV |
1096 | static void cleanup_mnt(struct mount *mnt) |
1097 | { | |
1098 | /* | |
1099 | * This probably indicates that somebody messed | |
1100 | * up a mnt_want/drop_write() pair. If this | |
1101 | * happens, the filesystem was probably unable | |
1102 | * to make r/w->r/o transitions. | |
1103 | */ | |
1104 | /* | |
1105 | * The locking used to deal with mnt_count decrement provides barriers, | |
1106 | * so mnt_get_writers() below is safe. | |
1107 | */ | |
1108 | WARN_ON(mnt_get_writers(mnt)); | |
1109 | if (unlikely(mnt->mnt_pins.first)) | |
1110 | mnt_pin_kill(mnt); | |
1111 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1112 | dput(mnt->mnt.mnt_root); | |
1113 | deactivate_super(mnt->mnt.mnt_sb); | |
1114 | mnt_free_id(mnt); | |
1115 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1116 | } | |
1117 | ||
1118 | static void __cleanup_mnt(struct rcu_head *head) | |
1119 | { | |
1120 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1121 | } | |
1122 | ||
1123 | static LLIST_HEAD(delayed_mntput_list); | |
1124 | static void delayed_mntput(struct work_struct *unused) | |
1125 | { | |
1126 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1127 | struct llist_node *next; | |
1128 | ||
1129 | for (; node; node = next) { | |
1130 | next = llist_next(node); | |
1131 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1132 | } | |
1133 | } | |
1134 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1135 | ||
900148dc | 1136 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1137 | { |
48a066e7 AV |
1138 | rcu_read_lock(); |
1139 | mnt_add_count(mnt, -1); | |
1140 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1141 | rcu_read_unlock(); | |
f03c6599 | 1142 | return; |
b3e19d92 | 1143 | } |
719ea2fb | 1144 | lock_mount_hash(); |
b3e19d92 | 1145 | if (mnt_get_count(mnt)) { |
48a066e7 | 1146 | rcu_read_unlock(); |
719ea2fb | 1147 | unlock_mount_hash(); |
99b7db7b NP |
1148 | return; |
1149 | } | |
48a066e7 AV |
1150 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1151 | rcu_read_unlock(); | |
1152 | unlock_mount_hash(); | |
1153 | return; | |
1154 | } | |
1155 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1156 | rcu_read_unlock(); | |
962830df | 1157 | |
39f7c4db | 1158 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1159 | |
1160 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1161 | struct mount *p, *tmp; | |
1162 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1163 | umount_mnt(p); | |
1164 | } | |
1165 | } | |
719ea2fb | 1166 | unlock_mount_hash(); |
649a795a | 1167 | |
9ea459e1 AV |
1168 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1169 | struct task_struct *task = current; | |
1170 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1171 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1172 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1173 | return; | |
1174 | } | |
1175 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1176 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1177 | return; | |
1178 | } | |
1179 | cleanup_mnt(mnt); | |
b3e19d92 | 1180 | } |
b3e19d92 NP |
1181 | |
1182 | void mntput(struct vfsmount *mnt) | |
1183 | { | |
1184 | if (mnt) { | |
863d684f | 1185 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1186 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1187 | if (unlikely(m->mnt_expiry_mark)) |
1188 | m->mnt_expiry_mark = 0; | |
1189 | mntput_no_expire(m); | |
b3e19d92 NP |
1190 | } |
1191 | } | |
1192 | EXPORT_SYMBOL(mntput); | |
1193 | ||
1194 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1195 | { | |
1196 | if (mnt) | |
83adc753 | 1197 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1198 | return mnt; |
1199 | } | |
1200 | EXPORT_SYMBOL(mntget); | |
1201 | ||
3064c356 | 1202 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1203 | { |
3064c356 AV |
1204 | struct mount *p; |
1205 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1206 | if (IS_ERR(p)) | |
1207 | return ERR_CAST(p); | |
1208 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1209 | return &p->mnt; | |
7b7b1ace | 1210 | } |
1da177e4 | 1211 | |
b3b304a2 MS |
1212 | static inline void mangle(struct seq_file *m, const char *s) |
1213 | { | |
1214 | seq_escape(m, s, " \t\n\\"); | |
1215 | } | |
1216 | ||
1217 | /* | |
1218 | * Simple .show_options callback for filesystems which don't want to | |
1219 | * implement more complex mount option showing. | |
1220 | * | |
1221 | * See also save_mount_options(). | |
1222 | */ | |
34c80b1d | 1223 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1224 | { |
2a32cebd AV |
1225 | const char *options; |
1226 | ||
1227 | rcu_read_lock(); | |
34c80b1d | 1228 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1229 | |
1230 | if (options != NULL && options[0]) { | |
1231 | seq_putc(m, ','); | |
1232 | mangle(m, options); | |
1233 | } | |
2a32cebd | 1234 | rcu_read_unlock(); |
b3b304a2 MS |
1235 | |
1236 | return 0; | |
1237 | } | |
1238 | EXPORT_SYMBOL(generic_show_options); | |
1239 | ||
1240 | /* | |
1241 | * If filesystem uses generic_show_options(), this function should be | |
1242 | * called from the fill_super() callback. | |
1243 | * | |
1244 | * The .remount_fs callback usually needs to be handled in a special | |
1245 | * way, to make sure, that previous options are not overwritten if the | |
1246 | * remount fails. | |
1247 | * | |
1248 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1249 | * reset all options to their default value, but changes only newly | |
1250 | * given options, then the displayed options will not reflect reality | |
1251 | * any more. | |
1252 | */ | |
1253 | void save_mount_options(struct super_block *sb, char *options) | |
1254 | { | |
2a32cebd AV |
1255 | BUG_ON(sb->s_options); |
1256 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1257 | } |
1258 | EXPORT_SYMBOL(save_mount_options); | |
1259 | ||
2a32cebd AV |
1260 | void replace_mount_options(struct super_block *sb, char *options) |
1261 | { | |
1262 | char *old = sb->s_options; | |
1263 | rcu_assign_pointer(sb->s_options, options); | |
1264 | if (old) { | |
1265 | synchronize_rcu(); | |
1266 | kfree(old); | |
1267 | } | |
1268 | } | |
1269 | EXPORT_SYMBOL(replace_mount_options); | |
1270 | ||
a1a2c409 | 1271 | #ifdef CONFIG_PROC_FS |
0226f492 | 1272 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1273 | static void *m_start(struct seq_file *m, loff_t *pos) |
1274 | { | |
ede1bf0d | 1275 | struct proc_mounts *p = m->private; |
1da177e4 | 1276 | |
390c6843 | 1277 | down_read(&namespace_sem); |
c7999c36 AV |
1278 | if (p->cached_event == p->ns->event) { |
1279 | void *v = p->cached_mount; | |
1280 | if (*pos == p->cached_index) | |
1281 | return v; | |
1282 | if (*pos == p->cached_index + 1) { | |
1283 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1284 | return p->cached_mount = v; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | p->cached_event = p->ns->event; | |
1289 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1290 | p->cached_index = *pos; | |
1291 | return p->cached_mount; | |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1295 | { | |
ede1bf0d | 1296 | struct proc_mounts *p = m->private; |
b0765fb8 | 1297 | |
c7999c36 AV |
1298 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1299 | p->cached_index = *pos; | |
1300 | return p->cached_mount; | |
1da177e4 LT |
1301 | } |
1302 | ||
1303 | static void m_stop(struct seq_file *m, void *v) | |
1304 | { | |
390c6843 | 1305 | up_read(&namespace_sem); |
1da177e4 LT |
1306 | } |
1307 | ||
0226f492 | 1308 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1309 | { |
ede1bf0d | 1310 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1311 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1312 | return p->show(m, &r->mnt); |
1da177e4 LT |
1313 | } |
1314 | ||
a1a2c409 | 1315 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1316 | .start = m_start, |
1317 | .next = m_next, | |
1318 | .stop = m_stop, | |
0226f492 | 1319 | .show = m_show, |
b4629fe2 | 1320 | }; |
a1a2c409 | 1321 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1322 | |
1da177e4 LT |
1323 | /** |
1324 | * may_umount_tree - check if a mount tree is busy | |
1325 | * @mnt: root of mount tree | |
1326 | * | |
1327 | * This is called to check if a tree of mounts has any | |
1328 | * open files, pwds, chroots or sub mounts that are | |
1329 | * busy. | |
1330 | */ | |
909b0a88 | 1331 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1332 | { |
909b0a88 | 1333 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1334 | int actual_refs = 0; |
1335 | int minimum_refs = 0; | |
315fc83e | 1336 | struct mount *p; |
909b0a88 | 1337 | BUG_ON(!m); |
1da177e4 | 1338 | |
b3e19d92 | 1339 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1340 | lock_mount_hash(); |
909b0a88 | 1341 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1342 | actual_refs += mnt_get_count(p); |
1da177e4 | 1343 | minimum_refs += 2; |
1da177e4 | 1344 | } |
719ea2fb | 1345 | unlock_mount_hash(); |
1da177e4 LT |
1346 | |
1347 | if (actual_refs > minimum_refs) | |
e3474a8e | 1348 | return 0; |
1da177e4 | 1349 | |
e3474a8e | 1350 | return 1; |
1da177e4 LT |
1351 | } |
1352 | ||
1353 | EXPORT_SYMBOL(may_umount_tree); | |
1354 | ||
1355 | /** | |
1356 | * may_umount - check if a mount point is busy | |
1357 | * @mnt: root of mount | |
1358 | * | |
1359 | * This is called to check if a mount point has any | |
1360 | * open files, pwds, chroots or sub mounts. If the | |
1361 | * mount has sub mounts this will return busy | |
1362 | * regardless of whether the sub mounts are busy. | |
1363 | * | |
1364 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1365 | * give false negatives. The main reason why it's here is that we need | |
1366 | * a non-destructive way to look for easily umountable filesystems. | |
1367 | */ | |
1368 | int may_umount(struct vfsmount *mnt) | |
1369 | { | |
e3474a8e | 1370 | int ret = 1; |
8ad08d8a | 1371 | down_read(&namespace_sem); |
719ea2fb | 1372 | lock_mount_hash(); |
1ab59738 | 1373 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1374 | ret = 0; |
719ea2fb | 1375 | unlock_mount_hash(); |
8ad08d8a | 1376 | up_read(&namespace_sem); |
a05964f3 | 1377 | return ret; |
1da177e4 LT |
1378 | } |
1379 | ||
1380 | EXPORT_SYMBOL(may_umount); | |
1381 | ||
38129a13 | 1382 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1383 | |
97216be0 | 1384 | static void namespace_unlock(void) |
70fbcdf4 | 1385 | { |
a3b3c562 | 1386 | struct hlist_head head; |
97216be0 | 1387 | |
a3b3c562 | 1388 | hlist_move_list(&unmounted, &head); |
97216be0 | 1389 | |
97216be0 AV |
1390 | up_write(&namespace_sem); |
1391 | ||
a3b3c562 EB |
1392 | if (likely(hlist_empty(&head))) |
1393 | return; | |
1394 | ||
48a066e7 AV |
1395 | synchronize_rcu(); |
1396 | ||
87b95ce0 | 1397 | group_pin_kill(&head); |
70fbcdf4 RP |
1398 | } |
1399 | ||
97216be0 | 1400 | static inline void namespace_lock(void) |
e3197d83 | 1401 | { |
97216be0 | 1402 | down_write(&namespace_sem); |
e3197d83 AV |
1403 | } |
1404 | ||
e819f152 EB |
1405 | enum umount_tree_flags { |
1406 | UMOUNT_SYNC = 1, | |
1407 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1408 | UMOUNT_CONNECTED = 4, |
e819f152 | 1409 | }; |
f2d0a123 EB |
1410 | |
1411 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1412 | { | |
1413 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1414 | if (how & UMOUNT_SYNC) | |
1415 | return true; | |
1416 | ||
1417 | /* A mount without a parent has nothing to be connected to */ | |
1418 | if (!mnt_has_parent(mnt)) | |
1419 | return true; | |
1420 | ||
1421 | /* Because the reference counting rules change when mounts are | |
1422 | * unmounted and connected, umounted mounts may not be | |
1423 | * connected to mounted mounts. | |
1424 | */ | |
1425 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1426 | return true; | |
1427 | ||
1428 | /* Has it been requested that the mount remain connected? */ | |
1429 | if (how & UMOUNT_CONNECTED) | |
1430 | return false; | |
1431 | ||
1432 | /* Is the mount locked such that it needs to remain connected? */ | |
1433 | if (IS_MNT_LOCKED(mnt)) | |
1434 | return false; | |
1435 | ||
1436 | /* By default disconnect the mount */ | |
1437 | return true; | |
1438 | } | |
1439 | ||
99b7db7b | 1440 | /* |
48a066e7 | 1441 | * mount_lock must be held |
99b7db7b NP |
1442 | * namespace_sem must be held for write |
1443 | */ | |
e819f152 | 1444 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1445 | { |
c003b26f | 1446 | LIST_HEAD(tmp_list); |
315fc83e | 1447 | struct mount *p; |
1da177e4 | 1448 | |
5d88457e EB |
1449 | if (how & UMOUNT_PROPAGATE) |
1450 | propagate_mount_unlock(mnt); | |
1451 | ||
c003b26f | 1452 | /* Gather the mounts to umount */ |
590ce4bc EB |
1453 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1454 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1455 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1456 | } |
1da177e4 | 1457 | |
411a938b | 1458 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1459 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1460 | list_del_init(&p->mnt_child); |
c003b26f | 1461 | } |
88b368f2 | 1462 | |
c003b26f | 1463 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1464 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1465 | propagate_umount(&tmp_list); |
a05964f3 | 1466 | |
c003b26f | 1467 | while (!list_empty(&tmp_list)) { |
fd4b5fa6 | 1468 | struct mnt_namespace *ns; |
ce07d891 | 1469 | bool disconnect; |
c003b26f | 1470 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1471 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1472 | list_del_init(&p->mnt_list); |
fd4b5fa6 EB |
1473 | ns = p->mnt_ns; |
1474 | if (ns) { | |
1475 | ns->mounts--; | |
1476 | __touch_mnt_namespace(ns); | |
1477 | } | |
143c8c91 | 1478 | p->mnt_ns = NULL; |
e819f152 | 1479 | if (how & UMOUNT_SYNC) |
48a066e7 | 1480 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1481 | |
f2d0a123 | 1482 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1483 | |
1484 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1485 | disconnect ? &unmounted : NULL); | |
676da58d | 1486 | if (mnt_has_parent(p)) { |
81b6b061 | 1487 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1488 | if (!disconnect) { |
1489 | /* Don't forget about p */ | |
1490 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1491 | } else { | |
1492 | umount_mnt(p); | |
1493 | } | |
7c4b93d8 | 1494 | } |
0f0afb1d | 1495 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1496 | } |
1497 | } | |
1498 | ||
b54b9be7 | 1499 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1500 | |
1ab59738 | 1501 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1502 | { |
1ab59738 | 1503 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1504 | int retval; |
1505 | ||
1ab59738 | 1506 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1507 | if (retval) |
1508 | return retval; | |
1509 | ||
1510 | /* | |
1511 | * Allow userspace to request a mountpoint be expired rather than | |
1512 | * unmounting unconditionally. Unmount only happens if: | |
1513 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1514 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1515 | */ | |
1516 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1517 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1518 | flags & (MNT_FORCE | MNT_DETACH)) |
1519 | return -EINVAL; | |
1520 | ||
b3e19d92 NP |
1521 | /* |
1522 | * probably don't strictly need the lock here if we examined | |
1523 | * all race cases, but it's a slowpath. | |
1524 | */ | |
719ea2fb | 1525 | lock_mount_hash(); |
83adc753 | 1526 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1527 | unlock_mount_hash(); |
1da177e4 | 1528 | return -EBUSY; |
b3e19d92 | 1529 | } |
719ea2fb | 1530 | unlock_mount_hash(); |
1da177e4 | 1531 | |
863d684f | 1532 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1533 | return -EAGAIN; |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * If we may have to abort operations to get out of this | |
1538 | * mount, and they will themselves hold resources we must | |
1539 | * allow the fs to do things. In the Unix tradition of | |
1540 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1541 | * might fail to complete on the first run through as other tasks | |
1542 | * must return, and the like. Thats for the mount program to worry | |
1543 | * about for the moment. | |
1544 | */ | |
1545 | ||
42faad99 | 1546 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1547 | sb->s_op->umount_begin(sb); |
42faad99 | 1548 | } |
1da177e4 LT |
1549 | |
1550 | /* | |
1551 | * No sense to grab the lock for this test, but test itself looks | |
1552 | * somewhat bogus. Suggestions for better replacement? | |
1553 | * Ho-hum... In principle, we might treat that as umount + switch | |
1554 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1555 | * Actually it makes sense, especially if rootfs would contain a | |
1556 | * /reboot - static binary that would close all descriptors and | |
1557 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1558 | */ | |
1ab59738 | 1559 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1560 | /* |
1561 | * Special case for "unmounting" root ... | |
1562 | * we just try to remount it readonly. | |
1563 | */ | |
05b459a0 | 1564 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1565 | return -EPERM; |
1da177e4 | 1566 | down_write(&sb->s_umount); |
4aa98cf7 | 1567 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1568 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1569 | up_write(&sb->s_umount); |
1570 | return retval; | |
1571 | } | |
1572 | ||
97216be0 | 1573 | namespace_lock(); |
719ea2fb | 1574 | lock_mount_hash(); |
5addc5dd | 1575 | event++; |
1da177e4 | 1576 | |
48a066e7 | 1577 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1578 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1579 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1580 | retval = 0; |
48a066e7 AV |
1581 | } else { |
1582 | shrink_submounts(mnt); | |
1583 | retval = -EBUSY; | |
1584 | if (!propagate_mount_busy(mnt, 2)) { | |
1585 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1586 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1587 | retval = 0; |
1588 | } | |
1da177e4 | 1589 | } |
719ea2fb | 1590 | unlock_mount_hash(); |
e3197d83 | 1591 | namespace_unlock(); |
1da177e4 LT |
1592 | return retval; |
1593 | } | |
1594 | ||
80b5dce8 EB |
1595 | /* |
1596 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1597 | * | |
1598 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1599 | * to an existing mountpoint, and wind up leaking the mount. | |
1600 | * detach_mounts allows lazily unmounting those mounts instead of | |
1601 | * leaking them. | |
1602 | * | |
1603 | * The caller may hold dentry->d_inode->i_mutex. | |
1604 | */ | |
1605 | void __detach_mounts(struct dentry *dentry) | |
1606 | { | |
1607 | struct mountpoint *mp; | |
1608 | struct mount *mnt; | |
1609 | ||
1610 | namespace_lock(); | |
e178bd11 | 1611 | lock_mount_hash(); |
80b5dce8 | 1612 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1613 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1614 | goto out_unlock; |
1615 | ||
fe27d92f | 1616 | event++; |
80b5dce8 EB |
1617 | while (!hlist_empty(&mp->m_list)) { |
1618 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1619 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1620 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1621 | umount_mnt(mnt); | |
ce07d891 | 1622 | } |
e0c9c0af | 1623 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1624 | } |
80b5dce8 EB |
1625 | put_mountpoint(mp); |
1626 | out_unlock: | |
e178bd11 | 1627 | unlock_mount_hash(); |
80b5dce8 EB |
1628 | namespace_unlock(); |
1629 | } | |
1630 | ||
9b40bc90 AV |
1631 | /* |
1632 | * Is the caller allowed to modify his namespace? | |
1633 | */ | |
1634 | static inline bool may_mount(void) | |
1635 | { | |
1636 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1637 | } | |
1638 | ||
1da177e4 LT |
1639 | /* |
1640 | * Now umount can handle mount points as well as block devices. | |
1641 | * This is important for filesystems which use unnamed block devices. | |
1642 | * | |
1643 | * We now support a flag for forced unmount like the other 'big iron' | |
1644 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1645 | */ | |
1646 | ||
bdc480e3 | 1647 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1648 | { |
2d8f3038 | 1649 | struct path path; |
900148dc | 1650 | struct mount *mnt; |
1da177e4 | 1651 | int retval; |
db1f05bb | 1652 | int lookup_flags = 0; |
1da177e4 | 1653 | |
db1f05bb MS |
1654 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1655 | return -EINVAL; | |
1656 | ||
9b40bc90 AV |
1657 | if (!may_mount()) |
1658 | return -EPERM; | |
1659 | ||
db1f05bb MS |
1660 | if (!(flags & UMOUNT_NOFOLLOW)) |
1661 | lookup_flags |= LOOKUP_FOLLOW; | |
1662 | ||
197df04c | 1663 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1664 | if (retval) |
1665 | goto out; | |
900148dc | 1666 | mnt = real_mount(path.mnt); |
1da177e4 | 1667 | retval = -EINVAL; |
2d8f3038 | 1668 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1669 | goto dput_and_out; |
143c8c91 | 1670 | if (!check_mnt(mnt)) |
1da177e4 | 1671 | goto dput_and_out; |
5ff9d8a6 EB |
1672 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1673 | goto dput_and_out; | |
b2f5d4dc EB |
1674 | retval = -EPERM; |
1675 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1676 | goto dput_and_out; | |
1da177e4 | 1677 | |
900148dc | 1678 | retval = do_umount(mnt, flags); |
1da177e4 | 1679 | dput_and_out: |
429731b1 | 1680 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1681 | dput(path.dentry); |
900148dc | 1682 | mntput_no_expire(mnt); |
1da177e4 LT |
1683 | out: |
1684 | return retval; | |
1685 | } | |
1686 | ||
1687 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1688 | ||
1689 | /* | |
b58fed8b | 1690 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1691 | */ |
bdc480e3 | 1692 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1693 | { |
b58fed8b | 1694 | return sys_umount(name, 0); |
1da177e4 LT |
1695 | } |
1696 | ||
1697 | #endif | |
1698 | ||
4ce5d2b1 | 1699 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1700 | { |
4ce5d2b1 | 1701 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1702 | return dentry->d_op == &ns_dentry_operations && |
1703 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1704 | } |
1705 | ||
58be2825 AV |
1706 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1707 | { | |
1708 | return container_of(ns, struct mnt_namespace, ns); | |
1709 | } | |
1710 | ||
4ce5d2b1 EB |
1711 | static bool mnt_ns_loop(struct dentry *dentry) |
1712 | { | |
1713 | /* Could bind mounting the mount namespace inode cause a | |
1714 | * mount namespace loop? | |
1715 | */ | |
1716 | struct mnt_namespace *mnt_ns; | |
1717 | if (!is_mnt_ns_file(dentry)) | |
1718 | return false; | |
1719 | ||
f77c8014 | 1720 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1721 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1722 | } | |
1723 | ||
87129cc0 | 1724 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1725 | int flag) |
1da177e4 | 1726 | { |
84d17192 | 1727 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1728 | |
4ce5d2b1 EB |
1729 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1730 | return ERR_PTR(-EINVAL); | |
1731 | ||
1732 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1733 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1734 | |
36341f64 | 1735 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1736 | if (IS_ERR(q)) |
1737 | return q; | |
1738 | ||
a73324da | 1739 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1740 | |
1741 | p = mnt; | |
6b41d536 | 1742 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1743 | struct mount *s; |
7ec02ef1 | 1744 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1745 | continue; |
1746 | ||
909b0a88 | 1747 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1748 | struct mount *t = NULL; |
4ce5d2b1 EB |
1749 | if (!(flag & CL_COPY_UNBINDABLE) && |
1750 | IS_MNT_UNBINDABLE(s)) { | |
1751 | s = skip_mnt_tree(s); | |
1752 | continue; | |
1753 | } | |
1754 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1755 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1756 | s = skip_mnt_tree(s); |
1757 | continue; | |
1758 | } | |
0714a533 AV |
1759 | while (p != s->mnt_parent) { |
1760 | p = p->mnt_parent; | |
1761 | q = q->mnt_parent; | |
1da177e4 | 1762 | } |
87129cc0 | 1763 | p = s; |
84d17192 | 1764 | parent = q; |
87129cc0 | 1765 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1766 | if (IS_ERR(q)) |
1767 | goto out; | |
719ea2fb | 1768 | lock_mount_hash(); |
1a4eeaf2 | 1769 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1770 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1771 | if (!list_empty(&parent->mnt_mounts)) { | |
1772 | t = list_last_entry(&parent->mnt_mounts, | |
1773 | struct mount, mnt_child); | |
1774 | if (t->mnt_mp != p->mnt_mp) | |
1775 | t = NULL; | |
1776 | } | |
1777 | attach_shadowed(q, parent, t); | |
719ea2fb | 1778 | unlock_mount_hash(); |
1da177e4 LT |
1779 | } |
1780 | } | |
1781 | return res; | |
be34d1a3 | 1782 | out: |
1da177e4 | 1783 | if (res) { |
719ea2fb | 1784 | lock_mount_hash(); |
e819f152 | 1785 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1786 | unlock_mount_hash(); |
1da177e4 | 1787 | } |
be34d1a3 | 1788 | return q; |
1da177e4 LT |
1789 | } |
1790 | ||
be34d1a3 DH |
1791 | /* Caller should check returned pointer for errors */ |
1792 | ||
589ff870 | 1793 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1794 | { |
cb338d06 | 1795 | struct mount *tree; |
97216be0 | 1796 | namespace_lock(); |
cd4a4017 EB |
1797 | if (!check_mnt(real_mount(path->mnt))) |
1798 | tree = ERR_PTR(-EINVAL); | |
1799 | else | |
1800 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1801 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1802 | namespace_unlock(); |
be34d1a3 | 1803 | if (IS_ERR(tree)) |
52e220d3 | 1804 | return ERR_CAST(tree); |
be34d1a3 | 1805 | return &tree->mnt; |
8aec0809 AV |
1806 | } |
1807 | ||
1808 | void drop_collected_mounts(struct vfsmount *mnt) | |
1809 | { | |
97216be0 | 1810 | namespace_lock(); |
719ea2fb | 1811 | lock_mount_hash(); |
e819f152 | 1812 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1813 | unlock_mount_hash(); |
3ab6abee | 1814 | namespace_unlock(); |
8aec0809 AV |
1815 | } |
1816 | ||
c771d683 MS |
1817 | /** |
1818 | * clone_private_mount - create a private clone of a path | |
1819 | * | |
1820 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1821 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1822 | * to the originating mount won't be propagated into this). | |
1823 | * | |
1824 | * Release with mntput(). | |
1825 | */ | |
1826 | struct vfsmount *clone_private_mount(struct path *path) | |
1827 | { | |
1828 | struct mount *old_mnt = real_mount(path->mnt); | |
1829 | struct mount *new_mnt; | |
1830 | ||
1831 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1832 | return ERR_PTR(-EINVAL); | |
1833 | ||
1834 | down_read(&namespace_sem); | |
1835 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); | |
1836 | up_read(&namespace_sem); | |
1837 | if (IS_ERR(new_mnt)) | |
1838 | return ERR_CAST(new_mnt); | |
1839 | ||
1840 | return &new_mnt->mnt; | |
1841 | } | |
1842 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1843 | ||
1f707137 AV |
1844 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1845 | struct vfsmount *root) | |
1846 | { | |
1a4eeaf2 | 1847 | struct mount *mnt; |
1f707137 AV |
1848 | int res = f(root, arg); |
1849 | if (res) | |
1850 | return res; | |
1a4eeaf2 AV |
1851 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1852 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1853 | if (res) |
1854 | return res; | |
1855 | } | |
1856 | return 0; | |
1857 | } | |
dee5220c | 1858 | EXPORT_SYMBOL(iterate_mounts); |
1f707137 | 1859 | |
4b8b21f4 | 1860 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1861 | { |
315fc83e | 1862 | struct mount *p; |
719f5d7f | 1863 | |
909b0a88 | 1864 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1865 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1866 | mnt_release_group_id(p); |
719f5d7f MS |
1867 | } |
1868 | } | |
1869 | ||
4b8b21f4 | 1870 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1871 | { |
315fc83e | 1872 | struct mount *p; |
719f5d7f | 1873 | |
909b0a88 | 1874 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1875 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1876 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1877 | if (err) { |
4b8b21f4 | 1878 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1879 | return err; |
1880 | } | |
1881 | } | |
1882 | } | |
1883 | ||
1884 | return 0; | |
1885 | } | |
1886 | ||
fd4b5fa6 EB |
1887 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1888 | { | |
1889 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1890 | unsigned int mounts = 0, old, pending, sum; | |
1891 | struct mount *p; | |
1892 | ||
1893 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1894 | mounts++; | |
1895 | ||
1896 | old = ns->mounts; | |
1897 | pending = ns->pending_mounts; | |
1898 | sum = old + pending; | |
1899 | if ((old > sum) || | |
1900 | (pending > sum) || | |
1901 | (max < sum) || | |
1902 | (mounts > (max - sum))) | |
1903 | return -ENOSPC; | |
1904 | ||
1905 | ns->pending_mounts = pending + mounts; | |
1906 | return 0; | |
1907 | } | |
1908 | ||
b90fa9ae RP |
1909 | /* |
1910 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1911 | * @nd : place the mount tree @source_mnt is attached |
1912 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1913 | * store the parent mount and mountpoint dentry. | |
1914 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1915 | * |
1916 | * NOTE: in the table below explains the semantics when a source mount | |
1917 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1918 | * --------------------------------------------------------------------------- |
1919 | * | BIND MOUNT OPERATION | | |
1920 | * |************************************************************************** | |
1921 | * | source-->| shared | private | slave | unbindable | | |
1922 | * | dest | | | | | | |
1923 | * | | | | | | | | |
1924 | * | v | | | | | | |
1925 | * |************************************************************************** | |
1926 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1927 | * | | | | | | | |
1928 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1929 | * *************************************************************************** | |
b90fa9ae RP |
1930 | * A bind operation clones the source mount and mounts the clone on the |
1931 | * destination mount. | |
1932 | * | |
1933 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1934 | * tree of the destination mount and the cloned mount is added to | |
1935 | * the peer group of the source mount. | |
1936 | * (+) the cloned mount is created under the destination mount and is marked | |
1937 | * as shared. The cloned mount is added to the peer group of the source | |
1938 | * mount. | |
5afe0022 RP |
1939 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1940 | * of the destination mount and the cloned mount is made slave | |
1941 | * of the same master as that of the source mount. The cloned mount | |
1942 | * is marked as 'shared and slave'. | |
1943 | * (*) the cloned mount is made a slave of the same master as that of the | |
1944 | * source mount. | |
1945 | * | |
9676f0c6 RP |
1946 | * --------------------------------------------------------------------------- |
1947 | * | MOVE MOUNT OPERATION | | |
1948 | * |************************************************************************** | |
1949 | * | source-->| shared | private | slave | unbindable | | |
1950 | * | dest | | | | | | |
1951 | * | | | | | | | | |
1952 | * | v | | | | | | |
1953 | * |************************************************************************** | |
1954 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1955 | * | | | | | | | |
1956 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1957 | * *************************************************************************** | |
5afe0022 RP |
1958 | * |
1959 | * (+) the mount is moved to the destination. And is then propagated to | |
1960 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1961 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1962 | * (+++) the mount is moved to the destination and is then propagated to |
1963 | * all the mounts belonging to the destination mount's propagation tree. | |
1964 | * the mount is marked as 'shared and slave'. | |
1965 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1966 | * |
1967 | * if the source mount is a tree, the operations explained above is | |
1968 | * applied to each mount in the tree. | |
1969 | * Must be called without spinlocks held, since this function can sleep | |
1970 | * in allocations. | |
1971 | */ | |
0fb54e50 | 1972 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1973 | struct mount *dest_mnt, |
1974 | struct mountpoint *dest_mp, | |
1975 | struct path *parent_path) | |
b90fa9ae | 1976 | { |
38129a13 | 1977 | HLIST_HEAD(tree_list); |
fd4b5fa6 | 1978 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
315fc83e | 1979 | struct mount *child, *p; |
38129a13 | 1980 | struct hlist_node *n; |
719f5d7f | 1981 | int err; |
b90fa9ae | 1982 | |
fd4b5fa6 EB |
1983 | /* Is there space to add these mounts to the mount namespace? */ |
1984 | if (!parent_path) { | |
1985 | err = count_mounts(ns, source_mnt); | |
1986 | if (err) | |
1987 | goto out; | |
1988 | } | |
1989 | ||
fc7be130 | 1990 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1991 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1992 | if (err) |
1993 | goto out; | |
0b1b901b | 1994 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1995 | lock_mount_hash(); |
0b1b901b AV |
1996 | if (err) |
1997 | goto out_cleanup_ids; | |
909b0a88 | 1998 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1999 | set_mnt_shared(p); |
0b1b901b AV |
2000 | } else { |
2001 | lock_mount_hash(); | |
b90fa9ae | 2002 | } |
1a390689 | 2003 | if (parent_path) { |
0fb54e50 | 2004 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2005 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2006 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2007 | } else { |
84d17192 | 2008 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 2009 | commit_tree(source_mnt, NULL); |
21444403 | 2010 | } |
b90fa9ae | 2011 | |
38129a13 | 2012 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2013 | struct mount *q; |
38129a13 | 2014 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
2015 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
2016 | child->mnt_mountpoint); | |
2017 | commit_tree(child, q); | |
b90fa9ae | 2018 | } |
719ea2fb | 2019 | unlock_mount_hash(); |
99b7db7b | 2020 | |
b90fa9ae | 2021 | return 0; |
719f5d7f MS |
2022 | |
2023 | out_cleanup_ids: | |
f2ebb3a9 AV |
2024 | while (!hlist_empty(&tree_list)) { |
2025 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
fd4b5fa6 | 2026 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2027 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2028 | } |
2029 | unlock_mount_hash(); | |
0b1b901b | 2030 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2031 | out: |
fd4b5fa6 | 2032 | ns->pending_mounts = 0; |
719f5d7f | 2033 | return err; |
b90fa9ae RP |
2034 | } |
2035 | ||
84d17192 | 2036 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2037 | { |
2038 | struct vfsmount *mnt; | |
84d17192 | 2039 | struct dentry *dentry = path->dentry; |
b12cea91 | 2040 | retry: |
84d17192 AV |
2041 | mutex_lock(&dentry->d_inode->i_mutex); |
2042 | if (unlikely(cant_mount(dentry))) { | |
2043 | mutex_unlock(&dentry->d_inode->i_mutex); | |
2044 | return ERR_PTR(-ENOENT); | |
b12cea91 | 2045 | } |
97216be0 | 2046 | namespace_lock(); |
b12cea91 | 2047 | mnt = lookup_mnt(path); |
84d17192 | 2048 | if (likely(!mnt)) { |
e178bd11 | 2049 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2050 | if (IS_ERR(mp)) { |
97216be0 | 2051 | namespace_unlock(); |
84d17192 AV |
2052 | mutex_unlock(&dentry->d_inode->i_mutex); |
2053 | return mp; | |
2054 | } | |
2055 | return mp; | |
2056 | } | |
97216be0 | 2057 | namespace_unlock(); |
b12cea91 AV |
2058 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
2059 | path_put(path); | |
2060 | path->mnt = mnt; | |
84d17192 | 2061 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2062 | goto retry; |
2063 | } | |
2064 | ||
84d17192 | 2065 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2066 | { |
84d17192 | 2067 | struct dentry *dentry = where->m_dentry; |
e178bd11 EB |
2068 | |
2069 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2070 | put_mountpoint(where); |
e178bd11 EB |
2071 | read_sequnlock_excl(&mount_lock); |
2072 | ||
328e6d90 | 2073 | namespace_unlock(); |
84d17192 | 2074 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
2075 | } |
2076 | ||
84d17192 | 2077 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2078 | { |
95bc5f25 | 2079 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
2080 | return -EINVAL; |
2081 | ||
e36cb0b8 DH |
2082 | if (d_is_dir(mp->m_dentry) != |
2083 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2084 | return -ENOTDIR; |
2085 | ||
84d17192 | 2086 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2087 | } |
2088 | ||
7a2e8a8f VA |
2089 | /* |
2090 | * Sanity check the flags to change_mnt_propagation. | |
2091 | */ | |
2092 | ||
2093 | static int flags_to_propagation_type(int flags) | |
2094 | { | |
7c6e984d | 2095 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2096 | |
2097 | /* Fail if any non-propagation flags are set */ | |
2098 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2099 | return 0; | |
2100 | /* Only one propagation flag should be set */ | |
2101 | if (!is_power_of_2(type)) | |
2102 | return 0; | |
2103 | return type; | |
2104 | } | |
2105 | ||
07b20889 RP |
2106 | /* |
2107 | * recursively change the type of the mountpoint. | |
2108 | */ | |
0a0d8a46 | 2109 | static int do_change_type(struct path *path, int flag) |
07b20889 | 2110 | { |
315fc83e | 2111 | struct mount *m; |
4b8b21f4 | 2112 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 2113 | int recurse = flag & MS_REC; |
7a2e8a8f | 2114 | int type; |
719f5d7f | 2115 | int err = 0; |
07b20889 | 2116 | |
2d92ab3c | 2117 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2118 | return -EINVAL; |
2119 | ||
7a2e8a8f VA |
2120 | type = flags_to_propagation_type(flag); |
2121 | if (!type) | |
2122 | return -EINVAL; | |
2123 | ||
97216be0 | 2124 | namespace_lock(); |
719f5d7f MS |
2125 | if (type == MS_SHARED) { |
2126 | err = invent_group_ids(mnt, recurse); | |
2127 | if (err) | |
2128 | goto out_unlock; | |
2129 | } | |
2130 | ||
719ea2fb | 2131 | lock_mount_hash(); |
909b0a88 | 2132 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2133 | change_mnt_propagation(m, type); |
719ea2fb | 2134 | unlock_mount_hash(); |
719f5d7f MS |
2135 | |
2136 | out_unlock: | |
97216be0 | 2137 | namespace_unlock(); |
719f5d7f | 2138 | return err; |
07b20889 RP |
2139 | } |
2140 | ||
5ff9d8a6 EB |
2141 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2142 | { | |
2143 | struct mount *child; | |
2144 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2145 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2146 | continue; | |
2147 | ||
2148 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2149 | return true; | |
2150 | } | |
2151 | return false; | |
2152 | } | |
2153 | ||
1da177e4 LT |
2154 | /* |
2155 | * do loopback mount. | |
2156 | */ | |
808d4e3c | 2157 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2158 | int recurse) |
1da177e4 | 2159 | { |
2d92ab3c | 2160 | struct path old_path; |
84d17192 AV |
2161 | struct mount *mnt = NULL, *old, *parent; |
2162 | struct mountpoint *mp; | |
57eccb83 | 2163 | int err; |
1da177e4 LT |
2164 | if (!old_name || !*old_name) |
2165 | return -EINVAL; | |
815d405c | 2166 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2167 | if (err) |
2168 | return err; | |
2169 | ||
8823c079 | 2170 | err = -EINVAL; |
4ce5d2b1 | 2171 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2172 | goto out; |
2173 | ||
84d17192 AV |
2174 | mp = lock_mount(path); |
2175 | err = PTR_ERR(mp); | |
2176 | if (IS_ERR(mp)) | |
b12cea91 AV |
2177 | goto out; |
2178 | ||
87129cc0 | 2179 | old = real_mount(old_path.mnt); |
84d17192 | 2180 | parent = real_mount(path->mnt); |
87129cc0 | 2181 | |
1da177e4 | 2182 | err = -EINVAL; |
fc7be130 | 2183 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2184 | goto out2; |
9676f0c6 | 2185 | |
e149ed2b AV |
2186 | if (!check_mnt(parent)) |
2187 | goto out2; | |
2188 | ||
2189 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2190 | goto out2; |
1da177e4 | 2191 | |
5ff9d8a6 EB |
2192 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2193 | goto out2; | |
2194 | ||
ccd48bc7 | 2195 | if (recurse) |
4ce5d2b1 | 2196 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2197 | else |
87129cc0 | 2198 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2199 | |
be34d1a3 DH |
2200 | if (IS_ERR(mnt)) { |
2201 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2202 | goto out2; |
be34d1a3 | 2203 | } |
ccd48bc7 | 2204 | |
5ff9d8a6 EB |
2205 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2206 | ||
84d17192 | 2207 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2208 | if (err) { |
719ea2fb | 2209 | lock_mount_hash(); |
e819f152 | 2210 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2211 | unlock_mount_hash(); |
5b83d2c5 | 2212 | } |
b12cea91 | 2213 | out2: |
84d17192 | 2214 | unlock_mount(mp); |
ccd48bc7 | 2215 | out: |
2d92ab3c | 2216 | path_put(&old_path); |
1da177e4 LT |
2217 | return err; |
2218 | } | |
2219 | ||
2e4b7fcd DH |
2220 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2221 | { | |
2222 | int error = 0; | |
2223 | int readonly_request = 0; | |
2224 | ||
2225 | if (ms_flags & MS_RDONLY) | |
2226 | readonly_request = 1; | |
2227 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2228 | return 0; | |
2229 | ||
2230 | if (readonly_request) | |
83adc753 | 2231 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2232 | else |
83adc753 | 2233 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2234 | return error; |
2235 | } | |
2236 | ||
1da177e4 LT |
2237 | /* |
2238 | * change filesystem flags. dir should be a physical root of filesystem. | |
2239 | * If you've mounted a non-root directory somewhere and want to do remount | |
2240 | * on it - tough luck. | |
2241 | */ | |
0a0d8a46 | 2242 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2243 | void *data) |
2244 | { | |
2245 | int err; | |
2d92ab3c | 2246 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2247 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2248 | |
143c8c91 | 2249 | if (!check_mnt(mnt)) |
1da177e4 LT |
2250 | return -EINVAL; |
2251 | ||
2d92ab3c | 2252 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2253 | return -EINVAL; |
2254 | ||
07b64558 EB |
2255 | /* Don't allow changing of locked mnt flags. |
2256 | * | |
2257 | * No locks need to be held here while testing the various | |
2258 | * MNT_LOCK flags because those flags can never be cleared | |
2259 | * once they are set. | |
2260 | */ | |
2261 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2262 | !(mnt_flags & MNT_READONLY)) { | |
2263 | return -EPERM; | |
2264 | } | |
9566d674 EB |
2265 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2266 | !(mnt_flags & MNT_NODEV)) { | |
3e186641 EB |
2267 | /* Was the nodev implicitly added in mount? */ |
2268 | if ((mnt->mnt_ns->user_ns != &init_user_ns) && | |
2269 | !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2270 | mnt_flags |= MNT_NODEV; | |
2271 | } else { | |
2272 | return -EPERM; | |
2273 | } | |
9566d674 EB |
2274 | } |
2275 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2276 | !(mnt_flags & MNT_NOSUID)) { | |
2277 | return -EPERM; | |
2278 | } | |
2279 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2280 | !(mnt_flags & MNT_NOEXEC)) { | |
2281 | return -EPERM; | |
2282 | } | |
2283 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2284 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2285 | return -EPERM; | |
2286 | } | |
2287 | ||
ff36fe2c EP |
2288 | err = security_sb_remount(sb, data); |
2289 | if (err) | |
2290 | return err; | |
2291 | ||
1da177e4 | 2292 | down_write(&sb->s_umount); |
2e4b7fcd | 2293 | if (flags & MS_BIND) |
2d92ab3c | 2294 | err = change_mount_flags(path->mnt, flags); |
05b459a0 | 2295 | else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
57eccb83 | 2296 | err = -EPERM; |
4aa98cf7 | 2297 | else |
2e4b7fcd | 2298 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2299 | if (!err) { |
719ea2fb | 2300 | lock_mount_hash(); |
a6138db8 | 2301 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2302 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2303 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2304 | unlock_mount_hash(); |
0e55a7cc | 2305 | } |
6339dab8 | 2306 | up_write(&sb->s_umount); |
1da177e4 LT |
2307 | return err; |
2308 | } | |
2309 | ||
cbbe362c | 2310 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2311 | { |
315fc83e | 2312 | struct mount *p; |
909b0a88 | 2313 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2314 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2315 | return 1; |
2316 | } | |
2317 | return 0; | |
2318 | } | |
2319 | ||
808d4e3c | 2320 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2321 | { |
2d92ab3c | 2322 | struct path old_path, parent_path; |
676da58d | 2323 | struct mount *p; |
0fb54e50 | 2324 | struct mount *old; |
84d17192 | 2325 | struct mountpoint *mp; |
57eccb83 | 2326 | int err; |
1da177e4 LT |
2327 | if (!old_name || !*old_name) |
2328 | return -EINVAL; | |
2d92ab3c | 2329 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2330 | if (err) |
2331 | return err; | |
2332 | ||
84d17192 AV |
2333 | mp = lock_mount(path); |
2334 | err = PTR_ERR(mp); | |
2335 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2336 | goto out; |
2337 | ||
143c8c91 | 2338 | old = real_mount(old_path.mnt); |
fc7be130 | 2339 | p = real_mount(path->mnt); |
143c8c91 | 2340 | |
1da177e4 | 2341 | err = -EINVAL; |
fc7be130 | 2342 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2343 | goto out1; |
2344 | ||
5ff9d8a6 EB |
2345 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2346 | goto out1; | |
2347 | ||
1da177e4 | 2348 | err = -EINVAL; |
2d92ab3c | 2349 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2350 | goto out1; |
1da177e4 | 2351 | |
676da58d | 2352 | if (!mnt_has_parent(old)) |
21444403 | 2353 | goto out1; |
1da177e4 | 2354 | |
e36cb0b8 DH |
2355 | if (d_is_dir(path->dentry) != |
2356 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2357 | goto out1; |
2358 | /* | |
2359 | * Don't move a mount residing in a shared parent. | |
2360 | */ | |
fc7be130 | 2361 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2362 | goto out1; |
9676f0c6 RP |
2363 | /* |
2364 | * Don't move a mount tree containing unbindable mounts to a destination | |
2365 | * mount which is shared. | |
2366 | */ | |
fc7be130 | 2367 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2368 | goto out1; |
1da177e4 | 2369 | err = -ELOOP; |
fc7be130 | 2370 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2371 | if (p == old) |
21444403 | 2372 | goto out1; |
1da177e4 | 2373 | |
84d17192 | 2374 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2375 | if (err) |
21444403 | 2376 | goto out1; |
1da177e4 LT |
2377 | |
2378 | /* if the mount is moved, it should no longer be expire | |
2379 | * automatically */ | |
6776db3d | 2380 | list_del_init(&old->mnt_expire); |
1da177e4 | 2381 | out1: |
84d17192 | 2382 | unlock_mount(mp); |
1da177e4 | 2383 | out: |
1da177e4 | 2384 | if (!err) |
1a390689 | 2385 | path_put(&parent_path); |
2d92ab3c | 2386 | path_put(&old_path); |
1da177e4 LT |
2387 | return err; |
2388 | } | |
2389 | ||
9d412a43 AV |
2390 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2391 | { | |
2392 | int err; | |
2393 | const char *subtype = strchr(fstype, '.'); | |
2394 | if (subtype) { | |
2395 | subtype++; | |
2396 | err = -EINVAL; | |
2397 | if (!subtype[0]) | |
2398 | goto err; | |
2399 | } else | |
2400 | subtype = ""; | |
2401 | ||
2402 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2403 | err = -ENOMEM; | |
2404 | if (!mnt->mnt_sb->s_subtype) | |
2405 | goto err; | |
2406 | return mnt; | |
2407 | ||
2408 | err: | |
2409 | mntput(mnt); | |
2410 | return ERR_PTR(err); | |
2411 | } | |
2412 | ||
9d412a43 AV |
2413 | /* |
2414 | * add a mount into a namespace's mount tree | |
2415 | */ | |
95bc5f25 | 2416 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2417 | { |
84d17192 AV |
2418 | struct mountpoint *mp; |
2419 | struct mount *parent; | |
9d412a43 AV |
2420 | int err; |
2421 | ||
f2ebb3a9 | 2422 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2423 | |
84d17192 AV |
2424 | mp = lock_mount(path); |
2425 | if (IS_ERR(mp)) | |
2426 | return PTR_ERR(mp); | |
9d412a43 | 2427 | |
84d17192 | 2428 | parent = real_mount(path->mnt); |
9d412a43 | 2429 | err = -EINVAL; |
84d17192 | 2430 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2431 | /* that's acceptable only for automounts done in private ns */ |
2432 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2433 | goto unlock; | |
2434 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2435 | if (!parent->mnt_ns) |
156cacb1 AV |
2436 | goto unlock; |
2437 | } | |
9d412a43 AV |
2438 | |
2439 | /* Refuse the same filesystem on the same mount point */ | |
2440 | err = -EBUSY; | |
95bc5f25 | 2441 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2442 | path->mnt->mnt_root == path->dentry) |
2443 | goto unlock; | |
2444 | ||
2445 | err = -EINVAL; | |
e36cb0b8 | 2446 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2447 | goto unlock; |
2448 | ||
95bc5f25 | 2449 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2450 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2451 | |
2452 | unlock: | |
84d17192 | 2453 | unlock_mount(mp); |
9d412a43 AV |
2454 | return err; |
2455 | } | |
b1e75df4 | 2456 | |
8c6cf9cc | 2457 | static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags); |
1b852bce | 2458 | |
1da177e4 LT |
2459 | /* |
2460 | * create a new mount for userspace and request it to be added into the | |
2461 | * namespace's tree | |
2462 | */ | |
0c55cfc4 | 2463 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2464 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2465 | { |
0c55cfc4 | 2466 | struct file_system_type *type; |
9b40bc90 | 2467 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2468 | struct vfsmount *mnt; |
15f9a3f3 | 2469 | int err; |
1da177e4 | 2470 | |
b6500fda SF |
2471 | if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN)) |
2472 | return -EPERM; | |
2473 | ||
0c55cfc4 | 2474 | if (!fstype) |
1da177e4 LT |
2475 | return -EINVAL; |
2476 | ||
0c55cfc4 EB |
2477 | type = get_fs_type(fstype); |
2478 | if (!type) | |
2479 | return -ENODEV; | |
2480 | ||
2481 | if (user_ns != &init_user_ns) { | |
0c55cfc4 EB |
2482 | /* Only in special cases allow devices from mounts |
2483 | * created outside the initial user namespace. | |
2484 | */ | |
2485 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2486 | flags |= MS_NODEV; | |
9566d674 | 2487 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 | 2488 | } |
1b852bce | 2489 | if (type->fs_flags & FS_USERNS_VISIBLE) { |
2bf68fcb EB |
2490 | if (!fs_fully_visible(type, &mnt_flags)) { |
2491 | put_filesystem(type); | |
1b852bce | 2492 | return -EPERM; |
2bf68fcb | 2493 | } |
1b852bce | 2494 | } |
0c55cfc4 EB |
2495 | } |
2496 | ||
2497 | mnt = vfs_kern_mount(type, flags, name, data); | |
2498 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2499 | !mnt->mnt_sb->s_subtype) | |
2500 | mnt = fs_set_subtype(mnt, fstype); | |
2501 | ||
2502 | put_filesystem(type); | |
1da177e4 LT |
2503 | if (IS_ERR(mnt)) |
2504 | return PTR_ERR(mnt); | |
2505 | ||
95bc5f25 | 2506 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2507 | if (err) |
2508 | mntput(mnt); | |
2509 | return err; | |
1da177e4 LT |
2510 | } |
2511 | ||
19a167af AV |
2512 | int finish_automount(struct vfsmount *m, struct path *path) |
2513 | { | |
6776db3d | 2514 | struct mount *mnt = real_mount(m); |
19a167af AV |
2515 | int err; |
2516 | /* The new mount record should have at least 2 refs to prevent it being | |
2517 | * expired before we get a chance to add it | |
2518 | */ | |
6776db3d | 2519 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2520 | |
2521 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2522 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2523 | err = -ELOOP; |
2524 | goto fail; | |
19a167af AV |
2525 | } |
2526 | ||
95bc5f25 | 2527 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2528 | if (!err) |
2529 | return 0; | |
2530 | fail: | |
2531 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2532 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2533 | namespace_lock(); |
6776db3d | 2534 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2535 | namespace_unlock(); |
19a167af | 2536 | } |
b1e75df4 AV |
2537 | mntput(m); |
2538 | mntput(m); | |
19a167af AV |
2539 | return err; |
2540 | } | |
2541 | ||
ea5b778a DH |
2542 | /** |
2543 | * mnt_set_expiry - Put a mount on an expiration list | |
2544 | * @mnt: The mount to list. | |
2545 | * @expiry_list: The list to add the mount to. | |
2546 | */ | |
2547 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2548 | { | |
97216be0 | 2549 | namespace_lock(); |
ea5b778a | 2550 | |
6776db3d | 2551 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2552 | |
97216be0 | 2553 | namespace_unlock(); |
ea5b778a DH |
2554 | } |
2555 | EXPORT_SYMBOL(mnt_set_expiry); | |
2556 | ||
1da177e4 LT |
2557 | /* |
2558 | * process a list of expirable mountpoints with the intent of discarding any | |
2559 | * mountpoints that aren't in use and haven't been touched since last we came | |
2560 | * here | |
2561 | */ | |
2562 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2563 | { | |
761d5c38 | 2564 | struct mount *mnt, *next; |
1da177e4 LT |
2565 | LIST_HEAD(graveyard); |
2566 | ||
2567 | if (list_empty(mounts)) | |
2568 | return; | |
2569 | ||
97216be0 | 2570 | namespace_lock(); |
719ea2fb | 2571 | lock_mount_hash(); |
1da177e4 LT |
2572 | |
2573 | /* extract from the expiration list every vfsmount that matches the | |
2574 | * following criteria: | |
2575 | * - only referenced by its parent vfsmount | |
2576 | * - still marked for expiry (marked on the last call here; marks are | |
2577 | * cleared by mntput()) | |
2578 | */ | |
6776db3d | 2579 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2580 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2581 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2582 | continue; |
6776db3d | 2583 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2584 | } |
bcc5c7d2 | 2585 | while (!list_empty(&graveyard)) { |
6776db3d | 2586 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2587 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2588 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2589 | } |
719ea2fb | 2590 | unlock_mount_hash(); |
3ab6abee | 2591 | namespace_unlock(); |
5528f911 TM |
2592 | } |
2593 | ||
2594 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2595 | ||
2596 | /* | |
2597 | * Ripoff of 'select_parent()' | |
2598 | * | |
2599 | * search the list of submounts for a given mountpoint, and move any | |
2600 | * shrinkable submounts to the 'graveyard' list. | |
2601 | */ | |
692afc31 | 2602 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2603 | { |
692afc31 | 2604 | struct mount *this_parent = parent; |
5528f911 TM |
2605 | struct list_head *next; |
2606 | int found = 0; | |
2607 | ||
2608 | repeat: | |
6b41d536 | 2609 | next = this_parent->mnt_mounts.next; |
5528f911 | 2610 | resume: |
6b41d536 | 2611 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2612 | struct list_head *tmp = next; |
6b41d536 | 2613 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2614 | |
2615 | next = tmp->next; | |
692afc31 | 2616 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2617 | continue; |
5528f911 TM |
2618 | /* |
2619 | * Descend a level if the d_mounts list is non-empty. | |
2620 | */ | |
6b41d536 | 2621 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2622 | this_parent = mnt; |
2623 | goto repeat; | |
2624 | } | |
1da177e4 | 2625 | |
1ab59738 | 2626 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2627 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2628 | found++; |
2629 | } | |
1da177e4 | 2630 | } |
5528f911 TM |
2631 | /* |
2632 | * All done at this level ... ascend and resume the search | |
2633 | */ | |
2634 | if (this_parent != parent) { | |
6b41d536 | 2635 | next = this_parent->mnt_child.next; |
0714a533 | 2636 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2637 | goto resume; |
2638 | } | |
2639 | return found; | |
2640 | } | |
2641 | ||
2642 | /* | |
2643 | * process a list of expirable mountpoints with the intent of discarding any | |
2644 | * submounts of a specific parent mountpoint | |
99b7db7b | 2645 | * |
48a066e7 | 2646 | * mount_lock must be held for write |
5528f911 | 2647 | */ |
b54b9be7 | 2648 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2649 | { |
2650 | LIST_HEAD(graveyard); | |
761d5c38 | 2651 | struct mount *m; |
5528f911 | 2652 | |
5528f911 | 2653 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2654 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2655 | while (!list_empty(&graveyard)) { |
761d5c38 | 2656 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2657 | mnt_expire); |
143c8c91 | 2658 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2659 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2660 | } |
2661 | } | |
1da177e4 LT |
2662 | } |
2663 | ||
1da177e4 LT |
2664 | /* |
2665 | * Some copy_from_user() implementations do not return the exact number of | |
2666 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2667 | * Note that this function differs from copy_from_user() in that it will oops | |
2668 | * on bad values of `to', rather than returning a short copy. | |
2669 | */ | |
b58fed8b RP |
2670 | static long exact_copy_from_user(void *to, const void __user * from, |
2671 | unsigned long n) | |
1da177e4 LT |
2672 | { |
2673 | char *t = to; | |
2674 | const char __user *f = from; | |
2675 | char c; | |
2676 | ||
2677 | if (!access_ok(VERIFY_READ, from, n)) | |
2678 | return n; | |
2679 | ||
2680 | while (n) { | |
2681 | if (__get_user(c, f)) { | |
2682 | memset(t, 0, n); | |
2683 | break; | |
2684 | } | |
2685 | *t++ = c; | |
2686 | f++; | |
2687 | n--; | |
2688 | } | |
2689 | return n; | |
2690 | } | |
2691 | ||
b58fed8b | 2692 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2693 | { |
2694 | int i; | |
2695 | unsigned long page; | |
2696 | unsigned long size; | |
b58fed8b | 2697 | |
1da177e4 LT |
2698 | *where = 0; |
2699 | if (!data) | |
2700 | return 0; | |
2701 | ||
2702 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2703 | return -ENOMEM; | |
2704 | ||
2705 | /* We only care that *some* data at the address the user | |
2706 | * gave us is valid. Just in case, we'll zero | |
2707 | * the remainder of the page. | |
2708 | */ | |
2709 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2710 | size = TASK_SIZE - (unsigned long)data; | |
2711 | if (size > PAGE_SIZE) | |
2712 | size = PAGE_SIZE; | |
2713 | ||
2714 | i = size - exact_copy_from_user((void *)page, data, size); | |
2715 | if (!i) { | |
b58fed8b | 2716 | free_page(page); |
1da177e4 LT |
2717 | return -EFAULT; |
2718 | } | |
2719 | if (i != PAGE_SIZE) | |
2720 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2721 | *where = page; | |
2722 | return 0; | |
2723 | } | |
2724 | ||
b8850d1f | 2725 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2726 | { |
b8850d1f | 2727 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2728 | } |
2729 | ||
1da177e4 LT |
2730 | /* |
2731 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2732 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2733 | * | |
2734 | * data is a (void *) that can point to any structure up to | |
2735 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2736 | * information (or be NULL). | |
2737 | * | |
2738 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2739 | * When the flags word was introduced its top half was required | |
2740 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2741 | * Therefore, if this magic number is present, it carries no information | |
2742 | * and must be discarded. | |
2743 | */ | |
5e6123f3 | 2744 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2745 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2746 | { |
2d92ab3c | 2747 | struct path path; |
1da177e4 LT |
2748 | int retval = 0; |
2749 | int mnt_flags = 0; | |
2750 | ||
2751 | /* Discard magic */ | |
2752 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2753 | flags &= ~MS_MGC_MSK; | |
2754 | ||
2755 | /* Basic sanity checks */ | |
1da177e4 LT |
2756 | if (data_page) |
2757 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2758 | ||
a27ab9f2 | 2759 | /* ... and get the mountpoint */ |
5e6123f3 | 2760 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2761 | if (retval) |
2762 | return retval; | |
2763 | ||
2764 | retval = security_sb_mount(dev_name, &path, | |
2765 | type_page, flags, data_page); | |
0d5cadb8 AV |
2766 | if (!retval && !may_mount()) |
2767 | retval = -EPERM; | |
a27ab9f2 TH |
2768 | if (retval) |
2769 | goto dput_out; | |
2770 | ||
613cbe3d AK |
2771 | /* Default to relatime unless overriden */ |
2772 | if (!(flags & MS_NOATIME)) | |
2773 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2774 | |
1da177e4 LT |
2775 | /* Separate the per-mountpoint flags */ |
2776 | if (flags & MS_NOSUID) | |
2777 | mnt_flags |= MNT_NOSUID; | |
2778 | if (flags & MS_NODEV) | |
2779 | mnt_flags |= MNT_NODEV; | |
2780 | if (flags & MS_NOEXEC) | |
2781 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2782 | if (flags & MS_NOATIME) |
2783 | mnt_flags |= MNT_NOATIME; | |
2784 | if (flags & MS_NODIRATIME) | |
2785 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2786 | if (flags & MS_STRICTATIME) |
2787 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2788 | if (flags & MS_RDONLY) |
2789 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2790 | |
ffbc6f0e EB |
2791 | /* The default atime for remount is preservation */ |
2792 | if ((flags & MS_REMOUNT) && | |
2793 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2794 | MS_STRICTATIME)) == 0)) { | |
2795 | mnt_flags &= ~MNT_ATIME_MASK; | |
2796 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2797 | } | |
2798 | ||
7a4dec53 | 2799 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 | 2800 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
28ba5876 | 2801 | MS_STRICTATIME | MS_SUBMOUNT); |
1da177e4 | 2802 | |
1da177e4 | 2803 | if (flags & MS_REMOUNT) |
2d92ab3c | 2804 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2805 | data_page); |
2806 | else if (flags & MS_BIND) | |
2d92ab3c | 2807 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2808 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2809 | retval = do_change_type(&path, flags); |
1da177e4 | 2810 | else if (flags & MS_MOVE) |
2d92ab3c | 2811 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2812 | else |
2d92ab3c | 2813 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2814 | dev_name, data_page); |
2815 | dput_out: | |
2d92ab3c | 2816 | path_put(&path); |
1da177e4 LT |
2817 | return retval; |
2818 | } | |
2819 | ||
771b1371 EB |
2820 | static void free_mnt_ns(struct mnt_namespace *ns) |
2821 | { | |
6344c433 | 2822 | ns_free_inum(&ns->ns); |
771b1371 EB |
2823 | put_user_ns(ns->user_ns); |
2824 | kfree(ns); | |
2825 | } | |
2826 | ||
8823c079 EB |
2827 | /* |
2828 | * Assign a sequence number so we can detect when we attempt to bind | |
2829 | * mount a reference to an older mount namespace into the current | |
2830 | * mount namespace, preventing reference counting loops. A 64bit | |
2831 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2832 | * is effectively never, so we can ignore the possibility. | |
2833 | */ | |
2834 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2835 | ||
771b1371 | 2836 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2837 | { |
2838 | struct mnt_namespace *new_ns; | |
98f842e6 | 2839 | int ret; |
cf8d2c11 TM |
2840 | |
2841 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2842 | if (!new_ns) | |
2843 | return ERR_PTR(-ENOMEM); | |
6344c433 | 2844 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2845 | if (ret) { |
2846 | kfree(new_ns); | |
2847 | return ERR_PTR(ret); | |
2848 | } | |
33c42940 | 2849 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2850 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2851 | atomic_set(&new_ns->count, 1); |
2852 | new_ns->root = NULL; | |
2853 | INIT_LIST_HEAD(&new_ns->list); | |
2854 | init_waitqueue_head(&new_ns->poll); | |
2855 | new_ns->event = 0; | |
771b1371 | 2856 | new_ns->user_ns = get_user_ns(user_ns); |
fd4b5fa6 EB |
2857 | new_ns->mounts = 0; |
2858 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2859 | return new_ns; |
2860 | } | |
2861 | ||
9559f689 AV |
2862 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2863 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2864 | { |
6b3286ed | 2865 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2866 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2867 | struct mount *p, *q; |
9559f689 | 2868 | struct mount *old; |
cb338d06 | 2869 | struct mount *new; |
7a472ef4 | 2870 | int copy_flags; |
1da177e4 | 2871 | |
9559f689 AV |
2872 | BUG_ON(!ns); |
2873 | ||
2874 | if (likely(!(flags & CLONE_NEWNS))) { | |
2875 | get_mnt_ns(ns); | |
2876 | return ns; | |
2877 | } | |
2878 | ||
2879 | old = ns->root; | |
2880 | ||
771b1371 | 2881 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2882 | if (IS_ERR(new_ns)) |
2883 | return new_ns; | |
1da177e4 | 2884 | |
97216be0 | 2885 | namespace_lock(); |
1da177e4 | 2886 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2887 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2888 | if (user_ns != ns->user_ns) |
132c94e3 | 2889 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2890 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2891 | if (IS_ERR(new)) { |
328e6d90 | 2892 | namespace_unlock(); |
771b1371 | 2893 | free_mnt_ns(new_ns); |
be34d1a3 | 2894 | return ERR_CAST(new); |
1da177e4 | 2895 | } |
be08d6d2 | 2896 | new_ns->root = new; |
1a4eeaf2 | 2897 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2898 | |
2899 | /* | |
2900 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2901 | * as belonging to new namespace. We have already acquired a private | |
2902 | * fs_struct, so tsk->fs->lock is not needed. | |
2903 | */ | |
909b0a88 | 2904 | p = old; |
cb338d06 | 2905 | q = new; |
1da177e4 | 2906 | while (p) { |
143c8c91 | 2907 | q->mnt_ns = new_ns; |
fd4b5fa6 | 2908 | new_ns->mounts++; |
9559f689 AV |
2909 | if (new_fs) { |
2910 | if (&p->mnt == new_fs->root.mnt) { | |
2911 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2912 | rootmnt = &p->mnt; |
1da177e4 | 2913 | } |
9559f689 AV |
2914 | if (&p->mnt == new_fs->pwd.mnt) { |
2915 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2916 | pwdmnt = &p->mnt; |
1da177e4 | 2917 | } |
1da177e4 | 2918 | } |
909b0a88 AV |
2919 | p = next_mnt(p, old); |
2920 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2921 | if (!q) |
2922 | break; | |
2923 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2924 | p = next_mnt(p, old); | |
1da177e4 | 2925 | } |
328e6d90 | 2926 | namespace_unlock(); |
1da177e4 | 2927 | |
1da177e4 | 2928 | if (rootmnt) |
f03c6599 | 2929 | mntput(rootmnt); |
1da177e4 | 2930 | if (pwdmnt) |
f03c6599 | 2931 | mntput(pwdmnt); |
1da177e4 | 2932 | |
741a2951 | 2933 | return new_ns; |
1da177e4 LT |
2934 | } |
2935 | ||
cf8d2c11 TM |
2936 | /** |
2937 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2938 | * @mnt: pointer to the new root filesystem mountpoint | |
2939 | */ | |
1a4eeaf2 | 2940 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2941 | { |
771b1371 | 2942 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2943 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2944 | struct mount *mnt = real_mount(m); |
2945 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2946 | new_ns->root = mnt; |
fd4b5fa6 | 2947 | new_ns->mounts++; |
b1983cd8 | 2948 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2949 | } else { |
1a4eeaf2 | 2950 | mntput(m); |
cf8d2c11 TM |
2951 | } |
2952 | return new_ns; | |
2953 | } | |
cf8d2c11 | 2954 | |
ea441d11 AV |
2955 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2956 | { | |
2957 | struct mnt_namespace *ns; | |
d31da0f0 | 2958 | struct super_block *s; |
ea441d11 AV |
2959 | struct path path; |
2960 | int err; | |
2961 | ||
2962 | ns = create_mnt_ns(mnt); | |
2963 | if (IS_ERR(ns)) | |
2964 | return ERR_CAST(ns); | |
2965 | ||
2966 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2967 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2968 | ||
2969 | put_mnt_ns(ns); | |
2970 | ||
2971 | if (err) | |
2972 | return ERR_PTR(err); | |
2973 | ||
2974 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2975 | s = path.mnt->mnt_sb; |
2976 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2977 | mntput(path.mnt); |
2978 | /* lock the sucker */ | |
d31da0f0 | 2979 | down_write(&s->s_umount); |
ea441d11 AV |
2980 | /* ... and return the root of (sub)tree on it */ |
2981 | return path.dentry; | |
2982 | } | |
2983 | EXPORT_SYMBOL(mount_subtree); | |
2984 | ||
bdc480e3 HC |
2985 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2986 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2987 | { |
eca6f534 VN |
2988 | int ret; |
2989 | char *kernel_type; | |
eca6f534 | 2990 | char *kernel_dev; |
1da177e4 | 2991 | unsigned long data_page; |
1da177e4 | 2992 | |
b8850d1f TG |
2993 | kernel_type = copy_mount_string(type); |
2994 | ret = PTR_ERR(kernel_type); | |
2995 | if (IS_ERR(kernel_type)) | |
eca6f534 | 2996 | goto out_type; |
1da177e4 | 2997 | |
b8850d1f TG |
2998 | kernel_dev = copy_mount_string(dev_name); |
2999 | ret = PTR_ERR(kernel_dev); | |
3000 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3001 | goto out_dev; |
1da177e4 | 3002 | |
eca6f534 VN |
3003 | ret = copy_mount_options(data, &data_page); |
3004 | if (ret < 0) | |
3005 | goto out_data; | |
1da177e4 | 3006 | |
5e6123f3 | 3007 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, |
eca6f534 | 3008 | (void *) data_page); |
1da177e4 | 3009 | |
eca6f534 VN |
3010 | free_page(data_page); |
3011 | out_data: | |
3012 | kfree(kernel_dev); | |
3013 | out_dev: | |
eca6f534 VN |
3014 | kfree(kernel_type); |
3015 | out_type: | |
3016 | return ret; | |
1da177e4 LT |
3017 | } |
3018 | ||
afac7cba AV |
3019 | /* |
3020 | * Return true if path is reachable from root | |
3021 | * | |
48a066e7 | 3022 | * namespace_sem or mount_lock is held |
afac7cba | 3023 | */ |
643822b4 | 3024 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3025 | const struct path *root) |
3026 | { | |
643822b4 | 3027 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3028 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3029 | mnt = mnt->mnt_parent; |
afac7cba | 3030 | } |
643822b4 | 3031 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3032 | } |
3033 | ||
3034 | int path_is_under(struct path *path1, struct path *path2) | |
3035 | { | |
3036 | int res; | |
48a066e7 | 3037 | read_seqlock_excl(&mount_lock); |
643822b4 | 3038 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3039 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3040 | return res; |
3041 | } | |
3042 | EXPORT_SYMBOL(path_is_under); | |
3043 | ||
1da177e4 LT |
3044 | /* |
3045 | * pivot_root Semantics: | |
3046 | * Moves the root file system of the current process to the directory put_old, | |
3047 | * makes new_root as the new root file system of the current process, and sets | |
3048 | * root/cwd of all processes which had them on the current root to new_root. | |
3049 | * | |
3050 | * Restrictions: | |
3051 | * The new_root and put_old must be directories, and must not be on the | |
3052 | * same file system as the current process root. The put_old must be | |
3053 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3054 | * pointed to by put_old must yield the same directory as new_root. No other | |
3055 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3056 | * | |
4a0d11fa NB |
3057 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3058 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3059 | * in this situation. | |
3060 | * | |
1da177e4 LT |
3061 | * Notes: |
3062 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3063 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3064 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3065 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3066 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3067 | * first. | |
3068 | */ | |
3480b257 HC |
3069 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3070 | const char __user *, put_old) | |
1da177e4 | 3071 | { |
2d8f3038 | 3072 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3073 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3074 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3075 | int error; |
3076 | ||
9b40bc90 | 3077 | if (!may_mount()) |
1da177e4 LT |
3078 | return -EPERM; |
3079 | ||
2d8f3038 | 3080 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3081 | if (error) |
3082 | goto out0; | |
1da177e4 | 3083 | |
2d8f3038 | 3084 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3085 | if (error) |
3086 | goto out1; | |
3087 | ||
2d8f3038 | 3088 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3089 | if (error) |
3090 | goto out2; | |
1da177e4 | 3091 | |
f7ad3c6b | 3092 | get_fs_root(current->fs, &root); |
84d17192 AV |
3093 | old_mp = lock_mount(&old); |
3094 | error = PTR_ERR(old_mp); | |
3095 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3096 | goto out3; |
3097 | ||
1da177e4 | 3098 | error = -EINVAL; |
419148da AV |
3099 | new_mnt = real_mount(new.mnt); |
3100 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3101 | old_mnt = real_mount(old.mnt); |
3102 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3103 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3104 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3105 | goto out4; |
143c8c91 | 3106 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3107 | goto out4; |
5ff9d8a6 EB |
3108 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3109 | goto out4; | |
1da177e4 | 3110 | error = -ENOENT; |
f3da392e | 3111 | if (d_unlinked(new.dentry)) |
b12cea91 | 3112 | goto out4; |
1da177e4 | 3113 | error = -EBUSY; |
84d17192 | 3114 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3115 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3116 | error = -EINVAL; |
8c3ee42e | 3117 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3118 | goto out4; /* not a mountpoint */ |
676da58d | 3119 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3120 | goto out4; /* not attached */ |
84d17192 | 3121 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3122 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3123 | goto out4; /* not a mountpoint */ |
676da58d | 3124 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3125 | goto out4; /* not attached */ |
4ac91378 | 3126 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3127 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3128 | goto out4; |
0d082601 EB |
3129 | /* make certain new is below the root */ |
3130 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3131 | goto out4; | |
84d17192 | 3132 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3133 | lock_mount_hash(); |
419148da AV |
3134 | detach_mnt(new_mnt, &parent_path); |
3135 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3136 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3137 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3138 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3139 | } | |
4ac91378 | 3140 | /* mount old root on put_old */ |
84d17192 | 3141 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3142 | /* mount new_root on / */ |
84d17192 | 3143 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3144 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3145 | /* A moved mount should not expire automatically */ |
3146 | list_del_init(&new_mnt->mnt_expire); | |
e178bd11 | 3147 | put_mountpoint(root_mp); |
719ea2fb | 3148 | unlock_mount_hash(); |
2d8f3038 | 3149 | chroot_fs_refs(&root, &new); |
1da177e4 | 3150 | error = 0; |
b12cea91 | 3151 | out4: |
84d17192 | 3152 | unlock_mount(old_mp); |
b12cea91 AV |
3153 | if (!error) { |
3154 | path_put(&root_parent); | |
3155 | path_put(&parent_path); | |
3156 | } | |
3157 | out3: | |
8c3ee42e | 3158 | path_put(&root); |
b12cea91 | 3159 | out2: |
2d8f3038 | 3160 | path_put(&old); |
1da177e4 | 3161 | out1: |
2d8f3038 | 3162 | path_put(&new); |
1da177e4 | 3163 | out0: |
1da177e4 | 3164 | return error; |
1da177e4 LT |
3165 | } |
3166 | ||
3167 | static void __init init_mount_tree(void) | |
3168 | { | |
3169 | struct vfsmount *mnt; | |
6b3286ed | 3170 | struct mnt_namespace *ns; |
ac748a09 | 3171 | struct path root; |
0c55cfc4 | 3172 | struct file_system_type *type; |
1da177e4 | 3173 | |
0c55cfc4 EB |
3174 | type = get_fs_type("rootfs"); |
3175 | if (!type) | |
3176 | panic("Can't find rootfs type"); | |
3177 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3178 | put_filesystem(type); | |
1da177e4 LT |
3179 | if (IS_ERR(mnt)) |
3180 | panic("Can't create rootfs"); | |
b3e19d92 | 3181 | |
3b22edc5 TM |
3182 | ns = create_mnt_ns(mnt); |
3183 | if (IS_ERR(ns)) | |
1da177e4 | 3184 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3185 | |
3186 | init_task.nsproxy->mnt_ns = ns; | |
3187 | get_mnt_ns(ns); | |
3188 | ||
be08d6d2 AV |
3189 | root.mnt = mnt; |
3190 | root.dentry = mnt->mnt_root; | |
da362b09 | 3191 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3192 | |
3193 | set_fs_pwd(current->fs, &root); | |
3194 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3195 | } |
3196 | ||
74bf17cf | 3197 | void __init mnt_init(void) |
1da177e4 | 3198 | { |
13f14b4d | 3199 | unsigned u; |
15a67dd8 | 3200 | int err; |
1da177e4 | 3201 | |
7d6fec45 | 3202 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3203 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3204 | |
0818bf27 | 3205 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3206 | sizeof(struct hlist_head), |
0818bf27 AV |
3207 | mhash_entries, 19, |
3208 | 0, | |
3209 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3210 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3211 | sizeof(struct hlist_head), | |
3212 | mphash_entries, 19, | |
3213 | 0, | |
3214 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3215 | |
84d17192 | 3216 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3217 | panic("Failed to allocate mount hash table\n"); |
3218 | ||
0818bf27 | 3219 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3220 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3221 | for (u = 0; u <= mp_hash_mask; u++) |
3222 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3223 | |
4b93dc9b TH |
3224 | kernfs_init(); |
3225 | ||
15a67dd8 RD |
3226 | err = sysfs_init(); |
3227 | if (err) | |
3228 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3229 | __func__, err); |
00d26666 GKH |
3230 | fs_kobj = kobject_create_and_add("fs", NULL); |
3231 | if (!fs_kobj) | |
8e24eea7 | 3232 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3233 | init_rootfs(); |
3234 | init_mount_tree(); | |
3235 | } | |
3236 | ||
616511d0 | 3237 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3238 | { |
d498b25a | 3239 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3240 | return; |
7b00ed6f | 3241 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3242 | free_mnt_ns(ns); |
1da177e4 | 3243 | } |
9d412a43 AV |
3244 | |
3245 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3246 | { | |
423e0ab0 TC |
3247 | struct vfsmount *mnt; |
3248 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3249 | if (!IS_ERR(mnt)) { | |
3250 | /* | |
3251 | * it is a longterm mount, don't release mnt until | |
3252 | * we unmount before file sys is unregistered | |
3253 | */ | |
f7a99c5b | 3254 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3255 | } |
3256 | return mnt; | |
9d412a43 AV |
3257 | } |
3258 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3259 | |
3260 | void kern_unmount(struct vfsmount *mnt) | |
3261 | { | |
3262 | /* release long term mount so mount point can be released */ | |
3263 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3264 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3265 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3266 | mntput(mnt); |
3267 | } | |
3268 | } | |
3269 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3270 | |
3271 | bool our_mnt(struct vfsmount *mnt) | |
3272 | { | |
143c8c91 | 3273 | return check_mnt(real_mount(mnt)); |
02125a82 | 3274 | } |
8823c079 | 3275 | |
3151527e EB |
3276 | bool current_chrooted(void) |
3277 | { | |
3278 | /* Does the current process have a non-standard root */ | |
3279 | struct path ns_root; | |
3280 | struct path fs_root; | |
3281 | bool chrooted; | |
3282 | ||
3283 | /* Find the namespace root */ | |
3284 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3285 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3286 | path_get(&ns_root); | |
3287 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3288 | ; | |
3289 | ||
3290 | get_fs_root(current->fs, &fs_root); | |
3291 | ||
3292 | chrooted = !path_equal(&fs_root, &ns_root); | |
3293 | ||
3294 | path_put(&fs_root); | |
3295 | path_put(&ns_root); | |
3296 | ||
3297 | return chrooted; | |
3298 | } | |
3299 | ||
8c6cf9cc | 3300 | static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags) |
87a8ebd6 EB |
3301 | { |
3302 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
8c6cf9cc | 3303 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3304 | struct mount *mnt; |
e51db735 | 3305 | bool visible = false; |
87a8ebd6 | 3306 | |
e51db735 EB |
3307 | if (unlikely(!ns)) |
3308 | return false; | |
3309 | ||
44bb4385 | 3310 | down_read(&namespace_sem); |
87a8ebd6 | 3311 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3312 | struct mount *child; |
77b1a97d EB |
3313 | int mnt_flags; |
3314 | ||
e51db735 EB |
3315 | if (mnt->mnt.mnt_sb->s_type != type) |
3316 | continue; | |
3317 | ||
7e96c1b0 EB |
3318 | /* This mount is not fully visible if it's root directory |
3319 | * is not the root directory of the filesystem. | |
3320 | */ | |
3321 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3322 | continue; | |
3323 | ||
77b1a97d EB |
3324 | /* Read the mount flags and filter out flags that |
3325 | * may safely be ignored. | |
3326 | */ | |
3327 | mnt_flags = mnt->mnt.mnt_flags; | |
3328 | if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC) | |
3329 | mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC); | |
06f412ab EB |
3330 | |
3331 | /* Don't miss readonly hidden in the superblock flags */ | |
3332 | if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) | |
3333 | mnt_flags |= MNT_LOCK_READONLY; | |
77b1a97d | 3334 | |
8c6cf9cc EB |
3335 | /* Verify the mount flags are equal to or more permissive |
3336 | * than the proposed new mount. | |
3337 | */ | |
77b1a97d | 3338 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3339 | !(new_flags & MNT_READONLY)) |
3340 | continue; | |
77b1a97d | 3341 | if ((mnt_flags & MNT_LOCK_NODEV) && |
8c6cf9cc EB |
3342 | !(new_flags & MNT_NODEV)) |
3343 | continue; | |
77b1a97d EB |
3344 | if ((mnt_flags & MNT_LOCK_NOSUID) && |
3345 | !(new_flags & MNT_NOSUID)) | |
3346 | continue; | |
3347 | if ((mnt_flags & MNT_LOCK_NOEXEC) && | |
3348 | !(new_flags & MNT_NOEXEC)) | |
3349 | continue; | |
3350 | if ((mnt_flags & MNT_LOCK_ATIME) && | |
3351 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3352 | continue; |
3353 | ||
ceeb0e5d EB |
3354 | /* This mount is not fully visible if there are any |
3355 | * locked child mounts that cover anything except for | |
3356 | * empty directories. | |
e51db735 EB |
3357 | */ |
3358 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3359 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3360 | /* Only worry about locked mounts */ |
314c8817 | 3361 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3362 | continue; |
7236c85e EB |
3363 | /* Is the directory permanetly empty? */ |
3364 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3365 | goto next; |
87a8ebd6 | 3366 | } |
8c6cf9cc | 3367 | /* Preserve the locked attributes */ |
77b1a97d EB |
3368 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
3369 | MNT_LOCK_NODEV | \ | |
3370 | MNT_LOCK_NOSUID | \ | |
3371 | MNT_LOCK_NOEXEC | \ | |
3372 | MNT_LOCK_ATIME); | |
e51db735 EB |
3373 | visible = true; |
3374 | goto found; | |
3375 | next: ; | |
87a8ebd6 | 3376 | } |
e51db735 | 3377 | found: |
44bb4385 | 3378 | up_read(&namespace_sem); |
e51db735 | 3379 | return visible; |
87a8ebd6 EB |
3380 | } |
3381 | ||
895b5840 AL |
3382 | bool mnt_may_suid(struct vfsmount *mnt) |
3383 | { | |
3384 | /* | |
3385 | * Foreign mounts (accessed via fchdir or through /proc | |
3386 | * symlinks) are always treated as if they are nosuid. This | |
3387 | * prevents namespaces from trusting potentially unsafe | |
3388 | * suid/sgid bits, file caps, or security labels that originate | |
3389 | * in other namespaces. | |
3390 | */ | |
3391 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
2c95625c | 3392 | current_in_userns(mnt->mnt_sb->s_user_ns); |
895b5840 AL |
3393 | } |
3394 | ||
64964528 | 3395 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3396 | { |
58be2825 | 3397 | struct ns_common *ns = NULL; |
8823c079 EB |
3398 | struct nsproxy *nsproxy; |
3399 | ||
728dba3a EB |
3400 | task_lock(task); |
3401 | nsproxy = task->nsproxy; | |
8823c079 | 3402 | if (nsproxy) { |
58be2825 AV |
3403 | ns = &nsproxy->mnt_ns->ns; |
3404 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3405 | } |
728dba3a | 3406 | task_unlock(task); |
8823c079 EB |
3407 | |
3408 | return ns; | |
3409 | } | |
3410 | ||
64964528 | 3411 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3412 | { |
58be2825 | 3413 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3414 | } |
3415 | ||
64964528 | 3416 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3417 | { |
3418 | struct fs_struct *fs = current->fs; | |
58be2825 | 3419 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3420 | struct path root; |
3421 | ||
0c55cfc4 | 3422 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3423 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3424 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3425 | return -EPERM; |
8823c079 EB |
3426 | |
3427 | if (fs->users != 1) | |
3428 | return -EINVAL; | |
3429 | ||
3430 | get_mnt_ns(mnt_ns); | |
3431 | put_mnt_ns(nsproxy->mnt_ns); | |
3432 | nsproxy->mnt_ns = mnt_ns; | |
3433 | ||
3434 | /* Find the root */ | |
3435 | root.mnt = &mnt_ns->root->mnt; | |
3436 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3437 | path_get(&root); | |
3438 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3439 | ; | |
3440 | ||
3441 | /* Update the pwd and root */ | |
3442 | set_fs_pwd(fs, &root); | |
3443 | set_fs_root(fs, &root); | |
3444 | ||
3445 | path_put(&root); | |
3446 | return 0; | |
3447 | } | |
3448 | ||
3449 | const struct proc_ns_operations mntns_operations = { | |
3450 | .name = "mnt", | |
3451 | .type = CLONE_NEWNS, | |
3452 | .get = mntns_get, | |
3453 | .put = mntns_put, | |
3454 | .install = mntns_install, | |
3455 | }; |