]>
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> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
0818bf27 | 26 | #include <linux/bootmem.h> |
9ea459e1 | 27 | #include <linux/task_work.h> |
9164bb4a IM |
28 | #include <linux/sched/task.h> |
29 | ||
07b20889 | 30 | #include "pnode.h" |
948730b0 | 31 | #include "internal.h" |
1da177e4 | 32 | |
d2921684 EB |
33 | /* Maximum number of mounts in a mount namespace */ |
34 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
35 | ||
0818bf27 AV |
36 | static unsigned int m_hash_mask __read_mostly; |
37 | static unsigned int m_hash_shift __read_mostly; | |
38 | static unsigned int mp_hash_mask __read_mostly; | |
39 | static unsigned int mp_hash_shift __read_mostly; | |
40 | ||
41 | static __initdata unsigned long mhash_entries; | |
42 | static int __init set_mhash_entries(char *str) | |
43 | { | |
44 | if (!str) | |
45 | return 0; | |
46 | mhash_entries = simple_strtoul(str, &str, 0); | |
47 | return 1; | |
48 | } | |
49 | __setup("mhash_entries=", set_mhash_entries); | |
50 | ||
51 | static __initdata unsigned long mphash_entries; | |
52 | static int __init set_mphash_entries(char *str) | |
53 | { | |
54 | if (!str) | |
55 | return 0; | |
56 | mphash_entries = simple_strtoul(str, &str, 0); | |
57 | return 1; | |
58 | } | |
59 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 60 | |
c7999c36 | 61 | static u64 event; |
73cd49ec | 62 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 63 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 64 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
65 | static int mnt_id_start = 0; |
66 | static int mnt_group_start = 1; | |
1da177e4 | 67 | |
38129a13 | 68 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 69 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 70 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 71 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 72 | |
f87fd4c2 | 73 | /* /sys/fs */ |
00d26666 GKH |
74 | struct kobject *fs_kobj; |
75 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 76 | |
99b7db7b NP |
77 | /* |
78 | * vfsmount lock may be taken for read to prevent changes to the | |
79 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
80 | * up the tree. | |
81 | * | |
82 | * It should be taken for write in all cases where the vfsmount | |
83 | * tree or hash is modified or when a vfsmount structure is modified. | |
84 | */ | |
48a066e7 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 86 | |
38129a13 | 87 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 88 | { |
b58fed8b RP |
89 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
90 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
91 | tmp = tmp + (tmp >> m_hash_shift); |
92 | return &mount_hashtable[tmp & m_hash_mask]; | |
93 | } | |
94 | ||
95 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
96 | { | |
97 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
98 | tmp = tmp + (tmp >> mp_hash_shift); | |
99 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
100 | } |
101 | ||
b105e270 | 102 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
103 | { |
104 | int res; | |
105 | ||
106 | retry: | |
107 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 108 | spin_lock(&mnt_id_lock); |
15169fe7 | 109 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 110 | if (!res) |
15169fe7 | 111 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 112 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
113 | if (res == -EAGAIN) |
114 | goto retry; | |
115 | ||
116 | return res; | |
117 | } | |
118 | ||
b105e270 | 119 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 120 | { |
15169fe7 | 121 | int id = mnt->mnt_id; |
99b7db7b | 122 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
123 | ida_remove(&mnt_id_ida, id); |
124 | if (mnt_id_start > id) | |
125 | mnt_id_start = id; | |
99b7db7b | 126 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
127 | } |
128 | ||
719f5d7f MS |
129 | /* |
130 | * Allocate a new peer group ID | |
131 | * | |
132 | * mnt_group_ida is protected by namespace_sem | |
133 | */ | |
4b8b21f4 | 134 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 135 | { |
f21f6220 AV |
136 | int res; |
137 | ||
719f5d7f MS |
138 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
139 | return -ENOMEM; | |
140 | ||
f21f6220 AV |
141 | res = ida_get_new_above(&mnt_group_ida, |
142 | mnt_group_start, | |
15169fe7 | 143 | &mnt->mnt_group_id); |
f21f6220 | 144 | if (!res) |
15169fe7 | 145 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
146 | |
147 | return res; | |
719f5d7f MS |
148 | } |
149 | ||
150 | /* | |
151 | * Release a peer group ID | |
152 | */ | |
4b8b21f4 | 153 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 154 | { |
15169fe7 | 155 | int id = mnt->mnt_group_id; |
f21f6220 AV |
156 | ida_remove(&mnt_group_ida, id); |
157 | if (mnt_group_start > id) | |
158 | mnt_group_start = id; | |
15169fe7 | 159 | mnt->mnt_group_id = 0; |
719f5d7f MS |
160 | } |
161 | ||
b3e19d92 NP |
162 | /* |
163 | * vfsmount lock must be held for read | |
164 | */ | |
83adc753 | 165 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
166 | { |
167 | #ifdef CONFIG_SMP | |
68e8a9fe | 168 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
169 | #else |
170 | preempt_disable(); | |
68e8a9fe | 171 | mnt->mnt_count += n; |
b3e19d92 NP |
172 | preempt_enable(); |
173 | #endif | |
174 | } | |
175 | ||
b3e19d92 NP |
176 | /* |
177 | * vfsmount lock must be held for write | |
178 | */ | |
83adc753 | 179 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
180 | { |
181 | #ifdef CONFIG_SMP | |
f03c6599 | 182 | unsigned int count = 0; |
b3e19d92 NP |
183 | int cpu; |
184 | ||
185 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 186 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
187 | } |
188 | ||
189 | return count; | |
190 | #else | |
68e8a9fe | 191 | return mnt->mnt_count; |
b3e19d92 NP |
192 | #endif |
193 | } | |
194 | ||
87b95ce0 AV |
195 | static void drop_mountpoint(struct fs_pin *p) |
196 | { | |
197 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
198 | dput(m->mnt_ex_mountpoint); | |
199 | pin_remove(p); | |
200 | mntput(&m->mnt); | |
201 | } | |
202 | ||
b105e270 | 203 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 204 | { |
c63181e6 AV |
205 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
206 | if (mnt) { | |
73cd49ec MS |
207 | int err; |
208 | ||
c63181e6 | 209 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
210 | if (err) |
211 | goto out_free_cache; | |
212 | ||
213 | if (name) { | |
fcc139ae | 214 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 215 | if (!mnt->mnt_devname) |
88b38782 | 216 | goto out_free_id; |
73cd49ec MS |
217 | } |
218 | ||
b3e19d92 | 219 | #ifdef CONFIG_SMP |
c63181e6 AV |
220 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
221 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
222 | goto out_free_devname; |
223 | ||
c63181e6 | 224 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 225 | #else |
c63181e6 AV |
226 | mnt->mnt_count = 1; |
227 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
228 | #endif |
229 | ||
38129a13 | 230 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
231 | INIT_LIST_HEAD(&mnt->mnt_child); |
232 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
233 | INIT_LIST_HEAD(&mnt->mnt_list); | |
234 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
235 | INIT_LIST_HEAD(&mnt->mnt_share); | |
236 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
237 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 238 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
239 | #ifdef CONFIG_FSNOTIFY |
240 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 241 | #endif |
87b95ce0 | 242 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 243 | } |
c63181e6 | 244 | return mnt; |
88b38782 | 245 | |
d3ef3d73 | 246 | #ifdef CONFIG_SMP |
247 | out_free_devname: | |
fcc139ae | 248 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 249 | #endif |
88b38782 | 250 | out_free_id: |
c63181e6 | 251 | mnt_free_id(mnt); |
88b38782 | 252 | out_free_cache: |
c63181e6 | 253 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 254 | return NULL; |
1da177e4 LT |
255 | } |
256 | ||
3d733633 DH |
257 | /* |
258 | * Most r/o checks on a fs are for operations that take | |
259 | * discrete amounts of time, like a write() or unlink(). | |
260 | * We must keep track of when those operations start | |
261 | * (for permission checks) and when they end, so that | |
262 | * we can determine when writes are able to occur to | |
263 | * a filesystem. | |
264 | */ | |
265 | /* | |
266 | * __mnt_is_readonly: check whether a mount is read-only | |
267 | * @mnt: the mount to check for its write status | |
268 | * | |
269 | * This shouldn't be used directly ouside of the VFS. | |
270 | * It does not guarantee that the filesystem will stay | |
271 | * r/w, just that it is right *now*. This can not and | |
272 | * should not be used in place of IS_RDONLY(inode). | |
273 | * mnt_want/drop_write() will _keep_ the filesystem | |
274 | * r/w. | |
275 | */ | |
276 | int __mnt_is_readonly(struct vfsmount *mnt) | |
277 | { | |
2e4b7fcd DH |
278 | if (mnt->mnt_flags & MNT_READONLY) |
279 | return 1; | |
280 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
281 | return 1; | |
282 | return 0; | |
3d733633 DH |
283 | } |
284 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
285 | ||
83adc753 | 286 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 287 | { |
288 | #ifdef CONFIG_SMP | |
68e8a9fe | 289 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 290 | #else |
68e8a9fe | 291 | mnt->mnt_writers++; |
d3ef3d73 | 292 | #endif |
293 | } | |
3d733633 | 294 | |
83adc753 | 295 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 296 | { |
d3ef3d73 | 297 | #ifdef CONFIG_SMP |
68e8a9fe | 298 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 299 | #else |
68e8a9fe | 300 | mnt->mnt_writers--; |
d3ef3d73 | 301 | #endif |
3d733633 | 302 | } |
3d733633 | 303 | |
83adc753 | 304 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 305 | { |
d3ef3d73 | 306 | #ifdef CONFIG_SMP |
307 | unsigned int count = 0; | |
3d733633 | 308 | int cpu; |
3d733633 DH |
309 | |
310 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 311 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 312 | } |
3d733633 | 313 | |
d3ef3d73 | 314 | return count; |
315 | #else | |
316 | return mnt->mnt_writers; | |
317 | #endif | |
3d733633 DH |
318 | } |
319 | ||
4ed5e82f MS |
320 | static int mnt_is_readonly(struct vfsmount *mnt) |
321 | { | |
322 | if (mnt->mnt_sb->s_readonly_remount) | |
323 | return 1; | |
324 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
325 | smp_rmb(); | |
326 | return __mnt_is_readonly(mnt); | |
327 | } | |
328 | ||
8366025e | 329 | /* |
eb04c282 JK |
330 | * Most r/o & frozen checks on a fs are for operations that take discrete |
331 | * amounts of time, like a write() or unlink(). We must keep track of when | |
332 | * those operations start (for permission checks) and when they end, so that we | |
333 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
334 | */ |
335 | /** | |
eb04c282 | 336 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 337 | * @m: the mount on which to take a write |
8366025e | 338 | * |
eb04c282 JK |
339 | * This tells the low-level filesystem that a write is about to be performed to |
340 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
341 | * returning success. This operation does not protect against filesystem being | |
342 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
343 | * called. This is effectively a refcount. | |
8366025e | 344 | */ |
eb04c282 | 345 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 346 | { |
83adc753 | 347 | struct mount *mnt = real_mount(m); |
3d733633 | 348 | int ret = 0; |
3d733633 | 349 | |
d3ef3d73 | 350 | preempt_disable(); |
c6653a83 | 351 | mnt_inc_writers(mnt); |
d3ef3d73 | 352 | /* |
c6653a83 | 353 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 354 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
355 | * incremented count after it has set MNT_WRITE_HOLD. | |
356 | */ | |
357 | smp_mb(); | |
1e75529e | 358 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 359 | cpu_relax(); |
360 | /* | |
361 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
362 | * be set to match its requirements. So we must not load that until | |
363 | * MNT_WRITE_HOLD is cleared. | |
364 | */ | |
365 | smp_rmb(); | |
4ed5e82f | 366 | if (mnt_is_readonly(m)) { |
c6653a83 | 367 | mnt_dec_writers(mnt); |
3d733633 | 368 | ret = -EROFS; |
3d733633 | 369 | } |
d3ef3d73 | 370 | preempt_enable(); |
eb04c282 JK |
371 | |
372 | return ret; | |
373 | } | |
374 | ||
375 | /** | |
376 | * mnt_want_write - get write access to a mount | |
377 | * @m: the mount on which to take a write | |
378 | * | |
379 | * This tells the low-level filesystem that a write is about to be performed to | |
380 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
381 | * is not frozen) before returning success. When the write operation is | |
382 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
383 | */ | |
384 | int mnt_want_write(struct vfsmount *m) | |
385 | { | |
386 | int ret; | |
387 | ||
388 | sb_start_write(m->mnt_sb); | |
389 | ret = __mnt_want_write(m); | |
390 | if (ret) | |
391 | sb_end_write(m->mnt_sb); | |
3d733633 | 392 | return ret; |
8366025e DH |
393 | } |
394 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
395 | ||
96029c4e | 396 | /** |
397 | * mnt_clone_write - get write access to a mount | |
398 | * @mnt: the mount on which to take a write | |
399 | * | |
400 | * This is effectively like mnt_want_write, except | |
401 | * it must only be used to take an extra write reference | |
402 | * on a mountpoint that we already know has a write reference | |
403 | * on it. This allows some optimisation. | |
404 | * | |
405 | * After finished, mnt_drop_write must be called as usual to | |
406 | * drop the reference. | |
407 | */ | |
408 | int mnt_clone_write(struct vfsmount *mnt) | |
409 | { | |
410 | /* superblock may be r/o */ | |
411 | if (__mnt_is_readonly(mnt)) | |
412 | return -EROFS; | |
413 | preempt_disable(); | |
83adc753 | 414 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 415 | preempt_enable(); |
416 | return 0; | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
419 | ||
420 | /** | |
eb04c282 | 421 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 422 | * @file: the file who's mount on which to take a write |
423 | * | |
eb04c282 | 424 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 425 | * do some optimisations if the file is open for write already |
426 | */ | |
eb04c282 | 427 | int __mnt_want_write_file(struct file *file) |
96029c4e | 428 | { |
83f936c7 | 429 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 430 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 431 | else |
432 | return mnt_clone_write(file->f_path.mnt); | |
433 | } | |
eb04c282 JK |
434 | |
435 | /** | |
436 | * mnt_want_write_file - get write access to a file's mount | |
437 | * @file: the file who's mount on which to take a write | |
438 | * | |
439 | * This is like mnt_want_write, but it takes a file and can | |
440 | * do some optimisations if the file is open for write already | |
441 | */ | |
442 | int mnt_want_write_file(struct file *file) | |
443 | { | |
444 | int ret; | |
445 | ||
446 | sb_start_write(file->f_path.mnt->mnt_sb); | |
447 | ret = __mnt_want_write_file(file); | |
448 | if (ret) | |
449 | sb_end_write(file->f_path.mnt->mnt_sb); | |
450 | return ret; | |
451 | } | |
96029c4e | 452 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
453 | ||
8366025e | 454 | /** |
eb04c282 | 455 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
456 | * @mnt: the mount on which to give up write access |
457 | * | |
458 | * Tells the low-level filesystem that we are done | |
459 | * performing writes to it. Must be matched with | |
eb04c282 | 460 | * __mnt_want_write() call above. |
8366025e | 461 | */ |
eb04c282 | 462 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 463 | { |
d3ef3d73 | 464 | preempt_disable(); |
83adc753 | 465 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 466 | preempt_enable(); |
8366025e | 467 | } |
eb04c282 JK |
468 | |
469 | /** | |
470 | * mnt_drop_write - give up write access to a mount | |
471 | * @mnt: the mount on which to give up write access | |
472 | * | |
473 | * Tells the low-level filesystem that we are done performing writes to it and | |
474 | * also allows filesystem to be frozen again. Must be matched with | |
475 | * mnt_want_write() call above. | |
476 | */ | |
477 | void mnt_drop_write(struct vfsmount *mnt) | |
478 | { | |
479 | __mnt_drop_write(mnt); | |
480 | sb_end_write(mnt->mnt_sb); | |
481 | } | |
8366025e DH |
482 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
483 | ||
eb04c282 JK |
484 | void __mnt_drop_write_file(struct file *file) |
485 | { | |
486 | __mnt_drop_write(file->f_path.mnt); | |
487 | } | |
488 | ||
2a79f17e AV |
489 | void mnt_drop_write_file(struct file *file) |
490 | { | |
491 | mnt_drop_write(file->f_path.mnt); | |
492 | } | |
493 | EXPORT_SYMBOL(mnt_drop_write_file); | |
494 | ||
83adc753 | 495 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 496 | { |
3d733633 DH |
497 | int ret = 0; |
498 | ||
719ea2fb | 499 | lock_mount_hash(); |
83adc753 | 500 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 501 | /* |
d3ef3d73 | 502 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
503 | * should be visible before we do. | |
3d733633 | 504 | */ |
d3ef3d73 | 505 | smp_mb(); |
506 | ||
3d733633 | 507 | /* |
d3ef3d73 | 508 | * With writers on hold, if this value is zero, then there are |
509 | * definitely no active writers (although held writers may subsequently | |
510 | * increment the count, they'll have to wait, and decrement it after | |
511 | * seeing MNT_READONLY). | |
512 | * | |
513 | * It is OK to have counter incremented on one CPU and decremented on | |
514 | * another: the sum will add up correctly. The danger would be when we | |
515 | * sum up each counter, if we read a counter before it is incremented, | |
516 | * but then read another CPU's count which it has been subsequently | |
517 | * decremented from -- we would see more decrements than we should. | |
518 | * MNT_WRITE_HOLD protects against this scenario, because | |
519 | * mnt_want_write first increments count, then smp_mb, then spins on | |
520 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
521 | * we're counting up here. | |
3d733633 | 522 | */ |
c6653a83 | 523 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 524 | ret = -EBUSY; |
525 | else | |
83adc753 | 526 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 527 | /* |
528 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
529 | * that become unheld will see MNT_READONLY. | |
530 | */ | |
531 | smp_wmb(); | |
83adc753 | 532 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 533 | unlock_mount_hash(); |
3d733633 | 534 | return ret; |
8366025e | 535 | } |
8366025e | 536 | |
83adc753 | 537 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 538 | { |
719ea2fb | 539 | lock_mount_hash(); |
83adc753 | 540 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 541 | unlock_mount_hash(); |
2e4b7fcd DH |
542 | } |
543 | ||
4ed5e82f MS |
544 | int sb_prepare_remount_readonly(struct super_block *sb) |
545 | { | |
546 | struct mount *mnt; | |
547 | int err = 0; | |
548 | ||
8e8b8796 MS |
549 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
550 | if (atomic_long_read(&sb->s_remove_count)) | |
551 | return -EBUSY; | |
552 | ||
719ea2fb | 553 | lock_mount_hash(); |
4ed5e82f MS |
554 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
555 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
556 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
557 | smp_mb(); | |
558 | if (mnt_get_writers(mnt) > 0) { | |
559 | err = -EBUSY; | |
560 | break; | |
561 | } | |
562 | } | |
563 | } | |
8e8b8796 MS |
564 | if (!err && atomic_long_read(&sb->s_remove_count)) |
565 | err = -EBUSY; | |
566 | ||
4ed5e82f MS |
567 | if (!err) { |
568 | sb->s_readonly_remount = 1; | |
569 | smp_wmb(); | |
570 | } | |
571 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
572 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
573 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
574 | } | |
719ea2fb | 575 | unlock_mount_hash(); |
4ed5e82f MS |
576 | |
577 | return err; | |
578 | } | |
579 | ||
b105e270 | 580 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 581 | { |
fcc139ae | 582 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 583 | #ifdef CONFIG_SMP |
68e8a9fe | 584 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 585 | #endif |
b105e270 | 586 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
587 | } |
588 | ||
8ffcb32e DH |
589 | static void delayed_free_vfsmnt(struct rcu_head *head) |
590 | { | |
591 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
592 | } | |
593 | ||
48a066e7 | 594 | /* call under rcu_read_lock */ |
294d71ff | 595 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
596 | { |
597 | struct mount *mnt; | |
598 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 599 | return 1; |
48a066e7 | 600 | if (bastard == NULL) |
294d71ff | 601 | return 0; |
48a066e7 AV |
602 | mnt = real_mount(bastard); |
603 | mnt_add_count(mnt, 1); | |
604 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 605 | return 0; |
48a066e7 AV |
606 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
607 | mnt_add_count(mnt, -1); | |
294d71ff AV |
608 | return 1; |
609 | } | |
610 | return -1; | |
611 | } | |
612 | ||
613 | /* call under rcu_read_lock */ | |
614 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
615 | { | |
616 | int res = __legitimize_mnt(bastard, seq); | |
617 | if (likely(!res)) | |
618 | return true; | |
619 | if (unlikely(res < 0)) { | |
620 | rcu_read_unlock(); | |
621 | mntput(bastard); | |
622 | rcu_read_lock(); | |
48a066e7 | 623 | } |
48a066e7 AV |
624 | return false; |
625 | } | |
626 | ||
1da177e4 | 627 | /* |
474279dc | 628 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 629 | * call under rcu_read_lock() |
1da177e4 | 630 | */ |
474279dc | 631 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 632 | { |
38129a13 | 633 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
634 | struct mount *p; |
635 | ||
38129a13 | 636 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
637 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
638 | return p; | |
639 | return NULL; | |
640 | } | |
641 | ||
a05964f3 | 642 | /* |
f015f126 DH |
643 | * lookup_mnt - Return the first child mount mounted at path |
644 | * | |
645 | * "First" means first mounted chronologically. If you create the | |
646 | * following mounts: | |
647 | * | |
648 | * mount /dev/sda1 /mnt | |
649 | * mount /dev/sda2 /mnt | |
650 | * mount /dev/sda3 /mnt | |
651 | * | |
652 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
653 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
654 | * /dev/sda2, then /dev/sda3, then NULL. | |
655 | * | |
656 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 657 | */ |
ca71cf71 | 658 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 659 | { |
c7105365 | 660 | struct mount *child_mnt; |
48a066e7 AV |
661 | struct vfsmount *m; |
662 | unsigned seq; | |
99b7db7b | 663 | |
48a066e7 AV |
664 | rcu_read_lock(); |
665 | do { | |
666 | seq = read_seqbegin(&mount_lock); | |
667 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
668 | m = child_mnt ? &child_mnt->mnt : NULL; | |
669 | } while (!legitimize_mnt(m, seq)); | |
670 | rcu_read_unlock(); | |
671 | return m; | |
a05964f3 RP |
672 | } |
673 | ||
7af1364f EB |
674 | /* |
675 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
676 | * current mount namespace. | |
677 | * | |
678 | * The common case is dentries are not mountpoints at all and that | |
679 | * test is handled inline. For the slow case when we are actually | |
680 | * dealing with a mountpoint of some kind, walk through all of the | |
681 | * mounts in the current mount namespace and test to see if the dentry | |
682 | * is a mountpoint. | |
683 | * | |
684 | * The mount_hashtable is not usable in the context because we | |
685 | * need to identify all mounts that may be in the current mount | |
686 | * namespace not just a mount that happens to have some specified | |
687 | * parent mount. | |
688 | */ | |
689 | bool __is_local_mountpoint(struct dentry *dentry) | |
690 | { | |
691 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
692 | struct mount *mnt; | |
693 | bool is_covered = false; | |
694 | ||
695 | if (!d_mountpoint(dentry)) | |
696 | goto out; | |
697 | ||
698 | down_read(&namespace_sem); | |
699 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
700 | is_covered = (mnt->mnt_mountpoint == dentry); | |
701 | if (is_covered) | |
702 | break; | |
703 | } | |
704 | up_read(&namespace_sem); | |
705 | out: | |
706 | return is_covered; | |
707 | } | |
708 | ||
e2dfa935 | 709 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 710 | { |
0818bf27 | 711 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
712 | struct mountpoint *mp; |
713 | ||
0818bf27 | 714 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
715 | if (mp->m_dentry == dentry) { |
716 | /* might be worth a WARN_ON() */ | |
717 | if (d_unlinked(dentry)) | |
718 | return ERR_PTR(-ENOENT); | |
719 | mp->m_count++; | |
720 | return mp; | |
721 | } | |
722 | } | |
e2dfa935 EB |
723 | return NULL; |
724 | } | |
725 | ||
3895dbf8 | 726 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 727 | { |
3895dbf8 | 728 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 729 | int ret; |
84d17192 | 730 | |
3895dbf8 EB |
731 | if (d_mountpoint(dentry)) { |
732 | mountpoint: | |
733 | read_seqlock_excl(&mount_lock); | |
734 | mp = lookup_mountpoint(dentry); | |
735 | read_sequnlock_excl(&mount_lock); | |
736 | if (mp) | |
737 | goto done; | |
738 | } | |
739 | ||
740 | if (!new) | |
741 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
742 | if (!new) | |
84d17192 AV |
743 | return ERR_PTR(-ENOMEM); |
744 | ||
3895dbf8 EB |
745 | |
746 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 747 | ret = d_set_mounted(dentry); |
eed81007 | 748 | |
3895dbf8 EB |
749 | /* Someone else set d_mounted? */ |
750 | if (ret == -EBUSY) | |
751 | goto mountpoint; | |
752 | ||
753 | /* The dentry is not available as a mountpoint? */ | |
754 | mp = ERR_PTR(ret); | |
755 | if (ret) | |
756 | goto done; | |
757 | ||
758 | /* Add the new mountpoint to the hash table */ | |
759 | read_seqlock_excl(&mount_lock); | |
760 | new->m_dentry = dentry; | |
761 | new->m_count = 1; | |
762 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
763 | INIT_HLIST_HEAD(&new->m_list); | |
764 | read_sequnlock_excl(&mount_lock); | |
765 | ||
766 | mp = new; | |
767 | new = NULL; | |
768 | done: | |
769 | kfree(new); | |
84d17192 AV |
770 | return mp; |
771 | } | |
772 | ||
773 | static void put_mountpoint(struct mountpoint *mp) | |
774 | { | |
775 | if (!--mp->m_count) { | |
776 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 777 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
778 | spin_lock(&dentry->d_lock); |
779 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
780 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 781 | hlist_del(&mp->m_hash); |
84d17192 AV |
782 | kfree(mp); |
783 | } | |
784 | } | |
785 | ||
143c8c91 | 786 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 787 | { |
6b3286ed | 788 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
789 | } |
790 | ||
99b7db7b NP |
791 | /* |
792 | * vfsmount lock must be held for write | |
793 | */ | |
6b3286ed | 794 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
795 | { |
796 | if (ns) { | |
797 | ns->event = ++event; | |
798 | wake_up_interruptible(&ns->poll); | |
799 | } | |
800 | } | |
801 | ||
99b7db7b NP |
802 | /* |
803 | * vfsmount lock must be held for write | |
804 | */ | |
6b3286ed | 805 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
806 | { |
807 | if (ns && ns->event != event) { | |
808 | ns->event = event; | |
809 | wake_up_interruptible(&ns->poll); | |
810 | } | |
811 | } | |
812 | ||
99b7db7b NP |
813 | /* |
814 | * vfsmount lock must be held for write | |
815 | */ | |
7bdb11de | 816 | static void unhash_mnt(struct mount *mnt) |
419148da | 817 | { |
0714a533 | 818 | mnt->mnt_parent = mnt; |
a73324da | 819 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 820 | list_del_init(&mnt->mnt_child); |
38129a13 | 821 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 822 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
823 | put_mountpoint(mnt->mnt_mp); |
824 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
825 | } |
826 | ||
7bdb11de EB |
827 | /* |
828 | * vfsmount lock must be held for write | |
829 | */ | |
830 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
831 | { | |
832 | old_path->dentry = mnt->mnt_mountpoint; | |
833 | old_path->mnt = &mnt->mnt_parent->mnt; | |
834 | unhash_mnt(mnt); | |
835 | } | |
836 | ||
6a46c573 EB |
837 | /* |
838 | * vfsmount lock must be held for write | |
839 | */ | |
840 | static void umount_mnt(struct mount *mnt) | |
841 | { | |
842 | /* old mountpoint will be dropped when we can do that */ | |
843 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
844 | unhash_mnt(mnt); | |
845 | } | |
846 | ||
99b7db7b NP |
847 | /* |
848 | * vfsmount lock must be held for write | |
849 | */ | |
84d17192 AV |
850 | void mnt_set_mountpoint(struct mount *mnt, |
851 | struct mountpoint *mp, | |
44d964d6 | 852 | struct mount *child_mnt) |
b90fa9ae | 853 | { |
84d17192 | 854 | mp->m_count++; |
3a2393d7 | 855 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 856 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 857 | child_mnt->mnt_parent = mnt; |
84d17192 | 858 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 859 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
860 | } |
861 | ||
1064f874 EB |
862 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
863 | { | |
864 | hlist_add_head_rcu(&mnt->mnt_hash, | |
865 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
866 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
867 | } | |
868 | ||
99b7db7b NP |
869 | /* |
870 | * vfsmount lock must be held for write | |
871 | */ | |
84d17192 AV |
872 | static void attach_mnt(struct mount *mnt, |
873 | struct mount *parent, | |
874 | struct mountpoint *mp) | |
1da177e4 | 875 | { |
84d17192 | 876 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 877 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
878 | } |
879 | ||
1064f874 | 880 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 881 | { |
1064f874 EB |
882 | struct mountpoint *old_mp = mnt->mnt_mp; |
883 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
884 | struct mount *old_parent = mnt->mnt_parent; | |
885 | ||
886 | list_del_init(&mnt->mnt_child); | |
887 | hlist_del_init(&mnt->mnt_mp_list); | |
888 | hlist_del_init_rcu(&mnt->mnt_hash); | |
889 | ||
890 | attach_mnt(mnt, parent, mp); | |
891 | ||
892 | put_mountpoint(old_mp); | |
893 | ||
894 | /* | |
895 | * Safely avoid even the suggestion this code might sleep or | |
896 | * lock the mount hash by taking advantage of the knowledge that | |
897 | * mnt_change_mountpoint will not release the final reference | |
898 | * to a mountpoint. | |
899 | * | |
900 | * During mounting, the mount passed in as the parent mount will | |
901 | * continue to use the old mountpoint and during unmounting, the | |
902 | * old mountpoint will continue to exist until namespace_unlock, | |
903 | * which happens well after mnt_change_mountpoint. | |
904 | */ | |
905 | spin_lock(&old_mountpoint->d_lock); | |
906 | old_mountpoint->d_lockref.count--; | |
907 | spin_unlock(&old_mountpoint->d_lock); | |
908 | ||
909 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
910 | } |
911 | ||
b90fa9ae | 912 | /* |
99b7db7b | 913 | * vfsmount lock must be held for write |
b90fa9ae | 914 | */ |
1064f874 | 915 | static void commit_tree(struct mount *mnt) |
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 | ||
d2921684 EB |
930 | n->mounts += n->pending_mounts; |
931 | n->pending_mounts = 0; | |
932 | ||
1064f874 | 933 | __attach_mnt(mnt, parent); |
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 | ||
93faccbb 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; | |
5235d448 AV |
1077 | } else { |
1078 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1079 | } |
1080 | if (flag & CL_MAKE_SHARED) | |
1081 | set_mnt_shared(mnt); | |
1082 | ||
1083 | /* stick the duplicate mount on the same expiry list | |
1084 | * as the original if that was on one */ | |
1085 | if (flag & CL_EXPIRE) { | |
1086 | if (!list_empty(&old->mnt_expire)) | |
1087 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1088 | } | |
1089 | ||
cb338d06 | 1090 | return mnt; |
719f5d7f MS |
1091 | |
1092 | out_free: | |
8ffcb32e | 1093 | mnt_free_id(mnt); |
719f5d7f | 1094 | free_vfsmnt(mnt); |
be34d1a3 | 1095 | return ERR_PTR(err); |
1da177e4 LT |
1096 | } |
1097 | ||
9ea459e1 AV |
1098 | static void cleanup_mnt(struct mount *mnt) |
1099 | { | |
1100 | /* | |
1101 | * This probably indicates that somebody messed | |
1102 | * up a mnt_want/drop_write() pair. If this | |
1103 | * happens, the filesystem was probably unable | |
1104 | * to make r/w->r/o transitions. | |
1105 | */ | |
1106 | /* | |
1107 | * The locking used to deal with mnt_count decrement provides barriers, | |
1108 | * so mnt_get_writers() below is safe. | |
1109 | */ | |
1110 | WARN_ON(mnt_get_writers(mnt)); | |
1111 | if (unlikely(mnt->mnt_pins.first)) | |
1112 | mnt_pin_kill(mnt); | |
1113 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1114 | dput(mnt->mnt.mnt_root); | |
1115 | deactivate_super(mnt->mnt.mnt_sb); | |
1116 | mnt_free_id(mnt); | |
1117 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1118 | } | |
1119 | ||
1120 | static void __cleanup_mnt(struct rcu_head *head) | |
1121 | { | |
1122 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1123 | } | |
1124 | ||
1125 | static LLIST_HEAD(delayed_mntput_list); | |
1126 | static void delayed_mntput(struct work_struct *unused) | |
1127 | { | |
1128 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1129 | struct llist_node *next; | |
1130 | ||
1131 | for (; node; node = next) { | |
1132 | next = llist_next(node); | |
1133 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1134 | } | |
1135 | } | |
1136 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1137 | ||
900148dc | 1138 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1139 | { |
48a066e7 AV |
1140 | rcu_read_lock(); |
1141 | mnt_add_count(mnt, -1); | |
1142 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1143 | rcu_read_unlock(); | |
f03c6599 | 1144 | return; |
b3e19d92 | 1145 | } |
719ea2fb | 1146 | lock_mount_hash(); |
b3e19d92 | 1147 | if (mnt_get_count(mnt)) { |
48a066e7 | 1148 | rcu_read_unlock(); |
719ea2fb | 1149 | unlock_mount_hash(); |
99b7db7b NP |
1150 | return; |
1151 | } | |
48a066e7 AV |
1152 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1153 | rcu_read_unlock(); | |
1154 | unlock_mount_hash(); | |
1155 | return; | |
1156 | } | |
1157 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1158 | rcu_read_unlock(); | |
962830df | 1159 | |
39f7c4db | 1160 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1161 | |
1162 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1163 | struct mount *p, *tmp; | |
1164 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1165 | umount_mnt(p); | |
1166 | } | |
1167 | } | |
719ea2fb | 1168 | unlock_mount_hash(); |
649a795a | 1169 | |
9ea459e1 AV |
1170 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1171 | struct task_struct *task = current; | |
1172 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1173 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1174 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1175 | return; | |
1176 | } | |
1177 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1178 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1179 | return; | |
1180 | } | |
1181 | cleanup_mnt(mnt); | |
b3e19d92 | 1182 | } |
b3e19d92 NP |
1183 | |
1184 | void mntput(struct vfsmount *mnt) | |
1185 | { | |
1186 | if (mnt) { | |
863d684f | 1187 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1188 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1189 | if (unlikely(m->mnt_expiry_mark)) |
1190 | m->mnt_expiry_mark = 0; | |
1191 | mntput_no_expire(m); | |
b3e19d92 NP |
1192 | } |
1193 | } | |
1194 | EXPORT_SYMBOL(mntput); | |
1195 | ||
1196 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1197 | { | |
1198 | if (mnt) | |
83adc753 | 1199 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1200 | return mnt; |
1201 | } | |
1202 | EXPORT_SYMBOL(mntget); | |
1203 | ||
c6609c0a IK |
1204 | /* path_is_mountpoint() - Check if path is a mount in the current |
1205 | * namespace. | |
1206 | * | |
1207 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1208 | * not mounted in any namespace and that common case is handled inline. | |
1209 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1210 | * mounts using a given dentry in a different namespace. This function | |
1211 | * checks if the passed in path is a mountpoint rather than the dentry | |
1212 | * alone. | |
1213 | */ | |
1214 | bool path_is_mountpoint(const struct path *path) | |
1215 | { | |
1216 | unsigned seq; | |
1217 | bool res; | |
1218 | ||
1219 | if (!d_mountpoint(path->dentry)) | |
1220 | return false; | |
1221 | ||
1222 | rcu_read_lock(); | |
1223 | do { | |
1224 | seq = read_seqbegin(&mount_lock); | |
1225 | res = __path_is_mountpoint(path); | |
1226 | } while (read_seqretry(&mount_lock, seq)); | |
1227 | rcu_read_unlock(); | |
1228 | ||
1229 | return res; | |
1230 | } | |
1231 | EXPORT_SYMBOL(path_is_mountpoint); | |
1232 | ||
ca71cf71 | 1233 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1234 | { |
3064c356 AV |
1235 | struct mount *p; |
1236 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1237 | if (IS_ERR(p)) | |
1238 | return ERR_CAST(p); | |
1239 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1240 | return &p->mnt; | |
7b7b1ace | 1241 | } |
1da177e4 | 1242 | |
b3b304a2 MS |
1243 | static inline void mangle(struct seq_file *m, const char *s) |
1244 | { | |
1245 | seq_escape(m, s, " \t\n\\"); | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Simple .show_options callback for filesystems which don't want to | |
1250 | * implement more complex mount option showing. | |
1251 | * | |
1252 | * See also save_mount_options(). | |
1253 | */ | |
34c80b1d | 1254 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1255 | { |
2a32cebd AV |
1256 | const char *options; |
1257 | ||
1258 | rcu_read_lock(); | |
34c80b1d | 1259 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1260 | |
1261 | if (options != NULL && options[0]) { | |
1262 | seq_putc(m, ','); | |
1263 | mangle(m, options); | |
1264 | } | |
2a32cebd | 1265 | rcu_read_unlock(); |
b3b304a2 MS |
1266 | |
1267 | return 0; | |
1268 | } | |
1269 | EXPORT_SYMBOL(generic_show_options); | |
1270 | ||
1271 | /* | |
1272 | * If filesystem uses generic_show_options(), this function should be | |
1273 | * called from the fill_super() callback. | |
1274 | * | |
1275 | * The .remount_fs callback usually needs to be handled in a special | |
1276 | * way, to make sure, that previous options are not overwritten if the | |
1277 | * remount fails. | |
1278 | * | |
1279 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1280 | * reset all options to their default value, but changes only newly | |
1281 | * given options, then the displayed options will not reflect reality | |
1282 | * any more. | |
1283 | */ | |
1284 | void save_mount_options(struct super_block *sb, char *options) | |
1285 | { | |
2a32cebd AV |
1286 | BUG_ON(sb->s_options); |
1287 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1288 | } |
1289 | EXPORT_SYMBOL(save_mount_options); | |
1290 | ||
2a32cebd AV |
1291 | void replace_mount_options(struct super_block *sb, char *options) |
1292 | { | |
1293 | char *old = sb->s_options; | |
1294 | rcu_assign_pointer(sb->s_options, options); | |
1295 | if (old) { | |
1296 | synchronize_rcu(); | |
1297 | kfree(old); | |
1298 | } | |
1299 | } | |
1300 | EXPORT_SYMBOL(replace_mount_options); | |
1301 | ||
a1a2c409 | 1302 | #ifdef CONFIG_PROC_FS |
0226f492 | 1303 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1304 | static void *m_start(struct seq_file *m, loff_t *pos) |
1305 | { | |
ede1bf0d | 1306 | struct proc_mounts *p = m->private; |
1da177e4 | 1307 | |
390c6843 | 1308 | down_read(&namespace_sem); |
c7999c36 AV |
1309 | if (p->cached_event == p->ns->event) { |
1310 | void *v = p->cached_mount; | |
1311 | if (*pos == p->cached_index) | |
1312 | return v; | |
1313 | if (*pos == p->cached_index + 1) { | |
1314 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1315 | return p->cached_mount = v; | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | p->cached_event = p->ns->event; | |
1320 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1321 | p->cached_index = *pos; | |
1322 | return p->cached_mount; | |
1da177e4 LT |
1323 | } |
1324 | ||
1325 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1326 | { | |
ede1bf0d | 1327 | struct proc_mounts *p = m->private; |
b0765fb8 | 1328 | |
c7999c36 AV |
1329 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1330 | p->cached_index = *pos; | |
1331 | return p->cached_mount; | |
1da177e4 LT |
1332 | } |
1333 | ||
1334 | static void m_stop(struct seq_file *m, void *v) | |
1335 | { | |
390c6843 | 1336 | up_read(&namespace_sem); |
1da177e4 LT |
1337 | } |
1338 | ||
0226f492 | 1339 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1340 | { |
ede1bf0d | 1341 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1342 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1343 | return p->show(m, &r->mnt); |
1da177e4 LT |
1344 | } |
1345 | ||
a1a2c409 | 1346 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1347 | .start = m_start, |
1348 | .next = m_next, | |
1349 | .stop = m_stop, | |
0226f492 | 1350 | .show = m_show, |
b4629fe2 | 1351 | }; |
a1a2c409 | 1352 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1353 | |
1da177e4 LT |
1354 | /** |
1355 | * may_umount_tree - check if a mount tree is busy | |
1356 | * @mnt: root of mount tree | |
1357 | * | |
1358 | * This is called to check if a tree of mounts has any | |
1359 | * open files, pwds, chroots or sub mounts that are | |
1360 | * busy. | |
1361 | */ | |
909b0a88 | 1362 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1363 | { |
909b0a88 | 1364 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1365 | int actual_refs = 0; |
1366 | int minimum_refs = 0; | |
315fc83e | 1367 | struct mount *p; |
909b0a88 | 1368 | BUG_ON(!m); |
1da177e4 | 1369 | |
b3e19d92 | 1370 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1371 | lock_mount_hash(); |
909b0a88 | 1372 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1373 | actual_refs += mnt_get_count(p); |
1da177e4 | 1374 | minimum_refs += 2; |
1da177e4 | 1375 | } |
719ea2fb | 1376 | unlock_mount_hash(); |
1da177e4 LT |
1377 | |
1378 | if (actual_refs > minimum_refs) | |
e3474a8e | 1379 | return 0; |
1da177e4 | 1380 | |
e3474a8e | 1381 | return 1; |
1da177e4 LT |
1382 | } |
1383 | ||
1384 | EXPORT_SYMBOL(may_umount_tree); | |
1385 | ||
1386 | /** | |
1387 | * may_umount - check if a mount point is busy | |
1388 | * @mnt: root of mount | |
1389 | * | |
1390 | * This is called to check if a mount point has any | |
1391 | * open files, pwds, chroots or sub mounts. If the | |
1392 | * mount has sub mounts this will return busy | |
1393 | * regardless of whether the sub mounts are busy. | |
1394 | * | |
1395 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1396 | * give false negatives. The main reason why it's here is that we need | |
1397 | * a non-destructive way to look for easily umountable filesystems. | |
1398 | */ | |
1399 | int may_umount(struct vfsmount *mnt) | |
1400 | { | |
e3474a8e | 1401 | int ret = 1; |
8ad08d8a | 1402 | down_read(&namespace_sem); |
719ea2fb | 1403 | lock_mount_hash(); |
1ab59738 | 1404 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1405 | ret = 0; |
719ea2fb | 1406 | unlock_mount_hash(); |
8ad08d8a | 1407 | up_read(&namespace_sem); |
a05964f3 | 1408 | return ret; |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | EXPORT_SYMBOL(may_umount); | |
1412 | ||
38129a13 | 1413 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1414 | |
97216be0 | 1415 | static void namespace_unlock(void) |
70fbcdf4 | 1416 | { |
a3b3c562 | 1417 | struct hlist_head head; |
97216be0 | 1418 | |
a3b3c562 | 1419 | hlist_move_list(&unmounted, &head); |
97216be0 | 1420 | |
97216be0 AV |
1421 | up_write(&namespace_sem); |
1422 | ||
a3b3c562 EB |
1423 | if (likely(hlist_empty(&head))) |
1424 | return; | |
1425 | ||
48a066e7 AV |
1426 | synchronize_rcu(); |
1427 | ||
87b95ce0 | 1428 | group_pin_kill(&head); |
70fbcdf4 RP |
1429 | } |
1430 | ||
97216be0 | 1431 | static inline void namespace_lock(void) |
e3197d83 | 1432 | { |
97216be0 | 1433 | down_write(&namespace_sem); |
e3197d83 AV |
1434 | } |
1435 | ||
e819f152 EB |
1436 | enum umount_tree_flags { |
1437 | UMOUNT_SYNC = 1, | |
1438 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1439 | UMOUNT_CONNECTED = 4, |
e819f152 | 1440 | }; |
f2d0a123 EB |
1441 | |
1442 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1443 | { | |
1444 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1445 | if (how & UMOUNT_SYNC) | |
1446 | return true; | |
1447 | ||
1448 | /* A mount without a parent has nothing to be connected to */ | |
1449 | if (!mnt_has_parent(mnt)) | |
1450 | return true; | |
1451 | ||
1452 | /* Because the reference counting rules change when mounts are | |
1453 | * unmounted and connected, umounted mounts may not be | |
1454 | * connected to mounted mounts. | |
1455 | */ | |
1456 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1457 | return true; | |
1458 | ||
1459 | /* Has it been requested that the mount remain connected? */ | |
1460 | if (how & UMOUNT_CONNECTED) | |
1461 | return false; | |
1462 | ||
1463 | /* Is the mount locked such that it needs to remain connected? */ | |
1464 | if (IS_MNT_LOCKED(mnt)) | |
1465 | return false; | |
1466 | ||
1467 | /* By default disconnect the mount */ | |
1468 | return true; | |
1469 | } | |
1470 | ||
99b7db7b | 1471 | /* |
48a066e7 | 1472 | * mount_lock must be held |
99b7db7b NP |
1473 | * namespace_sem must be held for write |
1474 | */ | |
e819f152 | 1475 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1476 | { |
c003b26f | 1477 | LIST_HEAD(tmp_list); |
315fc83e | 1478 | struct mount *p; |
1da177e4 | 1479 | |
5d88457e EB |
1480 | if (how & UMOUNT_PROPAGATE) |
1481 | propagate_mount_unlock(mnt); | |
1482 | ||
c003b26f | 1483 | /* Gather the mounts to umount */ |
590ce4bc EB |
1484 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1485 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1486 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1487 | } |
1da177e4 | 1488 | |
411a938b | 1489 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1490 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1491 | list_del_init(&p->mnt_child); |
c003b26f | 1492 | } |
88b368f2 | 1493 | |
c003b26f | 1494 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1495 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1496 | propagate_umount(&tmp_list); |
a05964f3 | 1497 | |
c003b26f | 1498 | while (!list_empty(&tmp_list)) { |
d2921684 | 1499 | struct mnt_namespace *ns; |
ce07d891 | 1500 | bool disconnect; |
c003b26f | 1501 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1502 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1503 | list_del_init(&p->mnt_list); |
d2921684 EB |
1504 | ns = p->mnt_ns; |
1505 | if (ns) { | |
1506 | ns->mounts--; | |
1507 | __touch_mnt_namespace(ns); | |
1508 | } | |
143c8c91 | 1509 | p->mnt_ns = NULL; |
e819f152 | 1510 | if (how & UMOUNT_SYNC) |
48a066e7 | 1511 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1512 | |
f2d0a123 | 1513 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1514 | |
1515 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1516 | disconnect ? &unmounted : NULL); | |
676da58d | 1517 | if (mnt_has_parent(p)) { |
81b6b061 | 1518 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1519 | if (!disconnect) { |
1520 | /* Don't forget about p */ | |
1521 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1522 | } else { | |
1523 | umount_mnt(p); | |
1524 | } | |
7c4b93d8 | 1525 | } |
0f0afb1d | 1526 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1527 | } |
1528 | } | |
1529 | ||
b54b9be7 | 1530 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1531 | |
1ab59738 | 1532 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1533 | { |
1ab59738 | 1534 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1535 | int retval; |
1536 | ||
1ab59738 | 1537 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1538 | if (retval) |
1539 | return retval; | |
1540 | ||
1541 | /* | |
1542 | * Allow userspace to request a mountpoint be expired rather than | |
1543 | * unmounting unconditionally. Unmount only happens if: | |
1544 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1545 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1546 | */ | |
1547 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1548 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1549 | flags & (MNT_FORCE | MNT_DETACH)) |
1550 | return -EINVAL; | |
1551 | ||
b3e19d92 NP |
1552 | /* |
1553 | * probably don't strictly need the lock here if we examined | |
1554 | * all race cases, but it's a slowpath. | |
1555 | */ | |
719ea2fb | 1556 | lock_mount_hash(); |
83adc753 | 1557 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1558 | unlock_mount_hash(); |
1da177e4 | 1559 | return -EBUSY; |
b3e19d92 | 1560 | } |
719ea2fb | 1561 | unlock_mount_hash(); |
1da177e4 | 1562 | |
863d684f | 1563 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1564 | return -EAGAIN; |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * If we may have to abort operations to get out of this | |
1569 | * mount, and they will themselves hold resources we must | |
1570 | * allow the fs to do things. In the Unix tradition of | |
1571 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1572 | * might fail to complete on the first run through as other tasks | |
1573 | * must return, and the like. Thats for the mount program to worry | |
1574 | * about for the moment. | |
1575 | */ | |
1576 | ||
42faad99 | 1577 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1578 | sb->s_op->umount_begin(sb); |
42faad99 | 1579 | } |
1da177e4 LT |
1580 | |
1581 | /* | |
1582 | * No sense to grab the lock for this test, but test itself looks | |
1583 | * somewhat bogus. Suggestions for better replacement? | |
1584 | * Ho-hum... In principle, we might treat that as umount + switch | |
1585 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1586 | * Actually it makes sense, especially if rootfs would contain a | |
1587 | * /reboot - static binary that would close all descriptors and | |
1588 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1589 | */ | |
1ab59738 | 1590 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1591 | /* |
1592 | * Special case for "unmounting" root ... | |
1593 | * we just try to remount it readonly. | |
1594 | */ | |
a1480dcc AL |
1595 | if (!capable(CAP_SYS_ADMIN)) |
1596 | return -EPERM; | |
1da177e4 | 1597 | down_write(&sb->s_umount); |
4aa98cf7 | 1598 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1599 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1600 | up_write(&sb->s_umount); |
1601 | return retval; | |
1602 | } | |
1603 | ||
97216be0 | 1604 | namespace_lock(); |
719ea2fb | 1605 | lock_mount_hash(); |
5addc5dd | 1606 | event++; |
1da177e4 | 1607 | |
48a066e7 | 1608 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1609 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1610 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1611 | retval = 0; |
48a066e7 AV |
1612 | } else { |
1613 | shrink_submounts(mnt); | |
1614 | retval = -EBUSY; | |
1615 | if (!propagate_mount_busy(mnt, 2)) { | |
1616 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1617 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1618 | retval = 0; |
1619 | } | |
1da177e4 | 1620 | } |
719ea2fb | 1621 | unlock_mount_hash(); |
e3197d83 | 1622 | namespace_unlock(); |
1da177e4 LT |
1623 | return retval; |
1624 | } | |
1625 | ||
80b5dce8 EB |
1626 | /* |
1627 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1628 | * | |
1629 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1630 | * to an existing mountpoint, and wind up leaking the mount. | |
1631 | * detach_mounts allows lazily unmounting those mounts instead of | |
1632 | * leaking them. | |
1633 | * | |
1634 | * The caller may hold dentry->d_inode->i_mutex. | |
1635 | */ | |
1636 | void __detach_mounts(struct dentry *dentry) | |
1637 | { | |
1638 | struct mountpoint *mp; | |
1639 | struct mount *mnt; | |
1640 | ||
1641 | namespace_lock(); | |
3895dbf8 | 1642 | lock_mount_hash(); |
80b5dce8 | 1643 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1644 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1645 | goto out_unlock; |
1646 | ||
e06b933e | 1647 | event++; |
80b5dce8 EB |
1648 | while (!hlist_empty(&mp->m_list)) { |
1649 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1650 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1651 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1652 | umount_mnt(mnt); | |
ce07d891 | 1653 | } |
e0c9c0af | 1654 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1655 | } |
80b5dce8 EB |
1656 | put_mountpoint(mp); |
1657 | out_unlock: | |
3895dbf8 | 1658 | unlock_mount_hash(); |
80b5dce8 EB |
1659 | namespace_unlock(); |
1660 | } | |
1661 | ||
9b40bc90 AV |
1662 | /* |
1663 | * Is the caller allowed to modify his namespace? | |
1664 | */ | |
1665 | static inline bool may_mount(void) | |
1666 | { | |
1667 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1668 | } | |
1669 | ||
9e8925b6 JL |
1670 | static inline bool may_mandlock(void) |
1671 | { | |
1672 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1673 | return false; | |
1674 | #endif | |
95ace754 | 1675 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1676 | } |
1677 | ||
1da177e4 LT |
1678 | /* |
1679 | * Now umount can handle mount points as well as block devices. | |
1680 | * This is important for filesystems which use unnamed block devices. | |
1681 | * | |
1682 | * We now support a flag for forced unmount like the other 'big iron' | |
1683 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1684 | */ | |
1685 | ||
bdc480e3 | 1686 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1687 | { |
2d8f3038 | 1688 | struct path path; |
900148dc | 1689 | struct mount *mnt; |
1da177e4 | 1690 | int retval; |
db1f05bb | 1691 | int lookup_flags = 0; |
1da177e4 | 1692 | |
db1f05bb MS |
1693 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1694 | return -EINVAL; | |
1695 | ||
9b40bc90 AV |
1696 | if (!may_mount()) |
1697 | return -EPERM; | |
1698 | ||
db1f05bb MS |
1699 | if (!(flags & UMOUNT_NOFOLLOW)) |
1700 | lookup_flags |= LOOKUP_FOLLOW; | |
1701 | ||
197df04c | 1702 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1703 | if (retval) |
1704 | goto out; | |
900148dc | 1705 | mnt = real_mount(path.mnt); |
1da177e4 | 1706 | retval = -EINVAL; |
2d8f3038 | 1707 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1708 | goto dput_and_out; |
143c8c91 | 1709 | if (!check_mnt(mnt)) |
1da177e4 | 1710 | goto dput_and_out; |
5ff9d8a6 EB |
1711 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1712 | goto dput_and_out; | |
b2f5d4dc EB |
1713 | retval = -EPERM; |
1714 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1715 | goto dput_and_out; | |
1da177e4 | 1716 | |
900148dc | 1717 | retval = do_umount(mnt, flags); |
1da177e4 | 1718 | dput_and_out: |
429731b1 | 1719 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1720 | dput(path.dentry); |
900148dc | 1721 | mntput_no_expire(mnt); |
1da177e4 LT |
1722 | out: |
1723 | return retval; | |
1724 | } | |
1725 | ||
1726 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1727 | ||
1728 | /* | |
b58fed8b | 1729 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1730 | */ |
bdc480e3 | 1731 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1732 | { |
b58fed8b | 1733 | return sys_umount(name, 0); |
1da177e4 LT |
1734 | } |
1735 | ||
1736 | #endif | |
1737 | ||
4ce5d2b1 | 1738 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1739 | { |
4ce5d2b1 | 1740 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1741 | return dentry->d_op == &ns_dentry_operations && |
1742 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1743 | } |
1744 | ||
58be2825 AV |
1745 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1746 | { | |
1747 | return container_of(ns, struct mnt_namespace, ns); | |
1748 | } | |
1749 | ||
4ce5d2b1 EB |
1750 | static bool mnt_ns_loop(struct dentry *dentry) |
1751 | { | |
1752 | /* Could bind mounting the mount namespace inode cause a | |
1753 | * mount namespace loop? | |
1754 | */ | |
1755 | struct mnt_namespace *mnt_ns; | |
1756 | if (!is_mnt_ns_file(dentry)) | |
1757 | return false; | |
1758 | ||
f77c8014 | 1759 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1760 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1761 | } | |
1762 | ||
87129cc0 | 1763 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1764 | int flag) |
1da177e4 | 1765 | { |
84d17192 | 1766 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1767 | |
4ce5d2b1 EB |
1768 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1769 | return ERR_PTR(-EINVAL); | |
1770 | ||
1771 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1772 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1773 | |
36341f64 | 1774 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1775 | if (IS_ERR(q)) |
1776 | return q; | |
1777 | ||
a73324da | 1778 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1779 | |
1780 | p = mnt; | |
6b41d536 | 1781 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1782 | struct mount *s; |
7ec02ef1 | 1783 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1784 | continue; |
1785 | ||
909b0a88 | 1786 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1787 | if (!(flag & CL_COPY_UNBINDABLE) && |
1788 | IS_MNT_UNBINDABLE(s)) { | |
1789 | s = skip_mnt_tree(s); | |
1790 | continue; | |
1791 | } | |
1792 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1793 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1794 | s = skip_mnt_tree(s); |
1795 | continue; | |
1796 | } | |
0714a533 AV |
1797 | while (p != s->mnt_parent) { |
1798 | p = p->mnt_parent; | |
1799 | q = q->mnt_parent; | |
1da177e4 | 1800 | } |
87129cc0 | 1801 | p = s; |
84d17192 | 1802 | parent = q; |
87129cc0 | 1803 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1804 | if (IS_ERR(q)) |
1805 | goto out; | |
719ea2fb | 1806 | lock_mount_hash(); |
1a4eeaf2 | 1807 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1808 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1809 | unlock_mount_hash(); |
1da177e4 LT |
1810 | } |
1811 | } | |
1812 | return res; | |
be34d1a3 | 1813 | out: |
1da177e4 | 1814 | if (res) { |
719ea2fb | 1815 | lock_mount_hash(); |
e819f152 | 1816 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1817 | unlock_mount_hash(); |
1da177e4 | 1818 | } |
be34d1a3 | 1819 | return q; |
1da177e4 LT |
1820 | } |
1821 | ||
be34d1a3 DH |
1822 | /* Caller should check returned pointer for errors */ |
1823 | ||
ca71cf71 | 1824 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1825 | { |
cb338d06 | 1826 | struct mount *tree; |
97216be0 | 1827 | namespace_lock(); |
cd4a4017 EB |
1828 | if (!check_mnt(real_mount(path->mnt))) |
1829 | tree = ERR_PTR(-EINVAL); | |
1830 | else | |
1831 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1832 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1833 | namespace_unlock(); |
be34d1a3 | 1834 | if (IS_ERR(tree)) |
52e220d3 | 1835 | return ERR_CAST(tree); |
be34d1a3 | 1836 | return &tree->mnt; |
8aec0809 AV |
1837 | } |
1838 | ||
1839 | void drop_collected_mounts(struct vfsmount *mnt) | |
1840 | { | |
97216be0 | 1841 | namespace_lock(); |
719ea2fb | 1842 | lock_mount_hash(); |
e819f152 | 1843 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1844 | unlock_mount_hash(); |
3ab6abee | 1845 | namespace_unlock(); |
8aec0809 AV |
1846 | } |
1847 | ||
c771d683 MS |
1848 | /** |
1849 | * clone_private_mount - create a private clone of a path | |
1850 | * | |
1851 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1852 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1853 | * to the originating mount won't be propagated into this). | |
1854 | * | |
1855 | * Release with mntput(). | |
1856 | */ | |
ca71cf71 | 1857 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1858 | { |
1859 | struct mount *old_mnt = real_mount(path->mnt); | |
1860 | struct mount *new_mnt; | |
1861 | ||
1862 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1863 | return ERR_PTR(-EINVAL); | |
1864 | ||
c771d683 | 1865 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1866 | if (IS_ERR(new_mnt)) |
1867 | return ERR_CAST(new_mnt); | |
1868 | ||
1869 | return &new_mnt->mnt; | |
1870 | } | |
1871 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1872 | ||
1f707137 AV |
1873 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1874 | struct vfsmount *root) | |
1875 | { | |
1a4eeaf2 | 1876 | struct mount *mnt; |
1f707137 AV |
1877 | int res = f(root, arg); |
1878 | if (res) | |
1879 | return res; | |
1a4eeaf2 AV |
1880 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1881 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1882 | if (res) |
1883 | return res; | |
1884 | } | |
1885 | return 0; | |
1886 | } | |
1887 | ||
4b8b21f4 | 1888 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1889 | { |
315fc83e | 1890 | struct mount *p; |
719f5d7f | 1891 | |
909b0a88 | 1892 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1893 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1894 | mnt_release_group_id(p); |
719f5d7f MS |
1895 | } |
1896 | } | |
1897 | ||
4b8b21f4 | 1898 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1899 | { |
315fc83e | 1900 | struct mount *p; |
719f5d7f | 1901 | |
909b0a88 | 1902 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1903 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1904 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1905 | if (err) { |
4b8b21f4 | 1906 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1907 | return err; |
1908 | } | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | return 0; | |
1913 | } | |
1914 | ||
d2921684 EB |
1915 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1916 | { | |
1917 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1918 | unsigned int mounts = 0, old, pending, sum; | |
1919 | struct mount *p; | |
1920 | ||
1921 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1922 | mounts++; | |
1923 | ||
1924 | old = ns->mounts; | |
1925 | pending = ns->pending_mounts; | |
1926 | sum = old + pending; | |
1927 | if ((old > sum) || | |
1928 | (pending > sum) || | |
1929 | (max < sum) || | |
1930 | (mounts > (max - sum))) | |
1931 | return -ENOSPC; | |
1932 | ||
1933 | ns->pending_mounts = pending + mounts; | |
1934 | return 0; | |
1935 | } | |
1936 | ||
b90fa9ae RP |
1937 | /* |
1938 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1939 | * @nd : place the mount tree @source_mnt is attached |
1940 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1941 | * store the parent mount and mountpoint dentry. | |
1942 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1943 | * |
1944 | * NOTE: in the table below explains the semantics when a source mount | |
1945 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1946 | * --------------------------------------------------------------------------- |
1947 | * | BIND 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 (*) | invalid | | |
1957 | * *************************************************************************** | |
b90fa9ae RP |
1958 | * A bind operation clones the source mount and mounts the clone on the |
1959 | * destination mount. | |
1960 | * | |
1961 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1962 | * tree of the destination mount and the cloned mount is added to | |
1963 | * the peer group of the source mount. | |
1964 | * (+) the cloned mount is created under the destination mount and is marked | |
1965 | * as shared. The cloned mount is added to the peer group of the source | |
1966 | * mount. | |
5afe0022 RP |
1967 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1968 | * of the destination mount and the cloned mount is made slave | |
1969 | * of the same master as that of the source mount. The cloned mount | |
1970 | * is marked as 'shared and slave'. | |
1971 | * (*) the cloned mount is made a slave of the same master as that of the | |
1972 | * source mount. | |
1973 | * | |
9676f0c6 RP |
1974 | * --------------------------------------------------------------------------- |
1975 | * | MOVE MOUNT OPERATION | | |
1976 | * |************************************************************************** | |
1977 | * | source-->| shared | private | slave | unbindable | | |
1978 | * | dest | | | | | | |
1979 | * | | | | | | | | |
1980 | * | v | | | | | | |
1981 | * |************************************************************************** | |
1982 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1983 | * | | | | | | | |
1984 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1985 | * *************************************************************************** | |
5afe0022 RP |
1986 | * |
1987 | * (+) the mount is moved to the destination. And is then propagated to | |
1988 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1989 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1990 | * (+++) the mount is moved to the destination and is then propagated to |
1991 | * all the mounts belonging to the destination mount's propagation tree. | |
1992 | * the mount is marked as 'shared and slave'. | |
1993 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1994 | * |
1995 | * if the source mount is a tree, the operations explained above is | |
1996 | * applied to each mount in the tree. | |
1997 | * Must be called without spinlocks held, since this function can sleep | |
1998 | * in allocations. | |
1999 | */ | |
0fb54e50 | 2000 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2001 | struct mount *dest_mnt, |
2002 | struct mountpoint *dest_mp, | |
2003 | struct path *parent_path) | |
b90fa9ae | 2004 | { |
38129a13 | 2005 | HLIST_HEAD(tree_list); |
d2921684 | 2006 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2007 | struct mountpoint *smp; |
315fc83e | 2008 | struct mount *child, *p; |
38129a13 | 2009 | struct hlist_node *n; |
719f5d7f | 2010 | int err; |
b90fa9ae | 2011 | |
1064f874 EB |
2012 | /* Preallocate a mountpoint in case the new mounts need |
2013 | * to be tucked under other mounts. | |
2014 | */ | |
2015 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2016 | if (IS_ERR(smp)) | |
2017 | return PTR_ERR(smp); | |
2018 | ||
d2921684 EB |
2019 | /* Is there space to add these mounts to the mount namespace? */ |
2020 | if (!parent_path) { | |
2021 | err = count_mounts(ns, source_mnt); | |
2022 | if (err) | |
2023 | goto out; | |
2024 | } | |
2025 | ||
fc7be130 | 2026 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2027 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2028 | if (err) |
2029 | goto out; | |
0b1b901b | 2030 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2031 | lock_mount_hash(); |
0b1b901b AV |
2032 | if (err) |
2033 | goto out_cleanup_ids; | |
909b0a88 | 2034 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2035 | set_mnt_shared(p); |
0b1b901b AV |
2036 | } else { |
2037 | lock_mount_hash(); | |
b90fa9ae | 2038 | } |
1a390689 | 2039 | if (parent_path) { |
0fb54e50 | 2040 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2041 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2042 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2043 | } else { |
84d17192 | 2044 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2045 | commit_tree(source_mnt); |
21444403 | 2046 | } |
b90fa9ae | 2047 | |
38129a13 | 2048 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2049 | struct mount *q; |
38129a13 | 2050 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2051 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2052 | child->mnt_mountpoint); | |
2053 | if (q) | |
2054 | mnt_change_mountpoint(child, smp, q); | |
2055 | commit_tree(child); | |
b90fa9ae | 2056 | } |
1064f874 | 2057 | put_mountpoint(smp); |
719ea2fb | 2058 | unlock_mount_hash(); |
99b7db7b | 2059 | |
b90fa9ae | 2060 | return 0; |
719f5d7f MS |
2061 | |
2062 | out_cleanup_ids: | |
f2ebb3a9 AV |
2063 | while (!hlist_empty(&tree_list)) { |
2064 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2065 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2066 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2067 | } |
2068 | unlock_mount_hash(); | |
0b1b901b | 2069 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2070 | out: |
d2921684 | 2071 | ns->pending_mounts = 0; |
1064f874 EB |
2072 | |
2073 | read_seqlock_excl(&mount_lock); | |
2074 | put_mountpoint(smp); | |
2075 | read_sequnlock_excl(&mount_lock); | |
2076 | ||
719f5d7f | 2077 | return err; |
b90fa9ae RP |
2078 | } |
2079 | ||
84d17192 | 2080 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2081 | { |
2082 | struct vfsmount *mnt; | |
84d17192 | 2083 | struct dentry *dentry = path->dentry; |
b12cea91 | 2084 | retry: |
5955102c | 2085 | inode_lock(dentry->d_inode); |
84d17192 | 2086 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2087 | inode_unlock(dentry->d_inode); |
84d17192 | 2088 | return ERR_PTR(-ENOENT); |
b12cea91 | 2089 | } |
97216be0 | 2090 | namespace_lock(); |
b12cea91 | 2091 | mnt = lookup_mnt(path); |
84d17192 | 2092 | if (likely(!mnt)) { |
3895dbf8 | 2093 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2094 | if (IS_ERR(mp)) { |
97216be0 | 2095 | namespace_unlock(); |
5955102c | 2096 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2097 | return mp; |
2098 | } | |
2099 | return mp; | |
2100 | } | |
97216be0 | 2101 | namespace_unlock(); |
5955102c | 2102 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2103 | path_put(path); |
2104 | path->mnt = mnt; | |
84d17192 | 2105 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2106 | goto retry; |
2107 | } | |
2108 | ||
84d17192 | 2109 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2110 | { |
84d17192 | 2111 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2112 | |
2113 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2114 | put_mountpoint(where); |
3895dbf8 EB |
2115 | read_sequnlock_excl(&mount_lock); |
2116 | ||
328e6d90 | 2117 | namespace_unlock(); |
5955102c | 2118 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2119 | } |
2120 | ||
84d17192 | 2121 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2122 | { |
95bc5f25 | 2123 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
2124 | return -EINVAL; |
2125 | ||
e36cb0b8 DH |
2126 | if (d_is_dir(mp->m_dentry) != |
2127 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2128 | return -ENOTDIR; |
2129 | ||
84d17192 | 2130 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2131 | } |
2132 | ||
7a2e8a8f VA |
2133 | /* |
2134 | * Sanity check the flags to change_mnt_propagation. | |
2135 | */ | |
2136 | ||
2137 | static int flags_to_propagation_type(int flags) | |
2138 | { | |
7c6e984d | 2139 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2140 | |
2141 | /* Fail if any non-propagation flags are set */ | |
2142 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2143 | return 0; | |
2144 | /* Only one propagation flag should be set */ | |
2145 | if (!is_power_of_2(type)) | |
2146 | return 0; | |
2147 | return type; | |
2148 | } | |
2149 | ||
07b20889 RP |
2150 | /* |
2151 | * recursively change the type of the mountpoint. | |
2152 | */ | |
0a0d8a46 | 2153 | static int do_change_type(struct path *path, int flag) |
07b20889 | 2154 | { |
315fc83e | 2155 | struct mount *m; |
4b8b21f4 | 2156 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 2157 | int recurse = flag & MS_REC; |
7a2e8a8f | 2158 | int type; |
719f5d7f | 2159 | int err = 0; |
07b20889 | 2160 | |
2d92ab3c | 2161 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2162 | return -EINVAL; |
2163 | ||
7a2e8a8f VA |
2164 | type = flags_to_propagation_type(flag); |
2165 | if (!type) | |
2166 | return -EINVAL; | |
2167 | ||
97216be0 | 2168 | namespace_lock(); |
719f5d7f MS |
2169 | if (type == MS_SHARED) { |
2170 | err = invent_group_ids(mnt, recurse); | |
2171 | if (err) | |
2172 | goto out_unlock; | |
2173 | } | |
2174 | ||
719ea2fb | 2175 | lock_mount_hash(); |
909b0a88 | 2176 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2177 | change_mnt_propagation(m, type); |
719ea2fb | 2178 | unlock_mount_hash(); |
719f5d7f MS |
2179 | |
2180 | out_unlock: | |
97216be0 | 2181 | namespace_unlock(); |
719f5d7f | 2182 | return err; |
07b20889 RP |
2183 | } |
2184 | ||
5ff9d8a6 EB |
2185 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2186 | { | |
2187 | struct mount *child; | |
2188 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2189 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2190 | continue; | |
2191 | ||
2192 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2193 | return true; | |
2194 | } | |
2195 | return false; | |
2196 | } | |
2197 | ||
1da177e4 LT |
2198 | /* |
2199 | * do loopback mount. | |
2200 | */ | |
808d4e3c | 2201 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2202 | int recurse) |
1da177e4 | 2203 | { |
2d92ab3c | 2204 | struct path old_path; |
84d17192 AV |
2205 | struct mount *mnt = NULL, *old, *parent; |
2206 | struct mountpoint *mp; | |
57eccb83 | 2207 | int err; |
1da177e4 LT |
2208 | if (!old_name || !*old_name) |
2209 | return -EINVAL; | |
815d405c | 2210 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2211 | if (err) |
2212 | return err; | |
2213 | ||
8823c079 | 2214 | err = -EINVAL; |
4ce5d2b1 | 2215 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2216 | goto out; |
2217 | ||
84d17192 AV |
2218 | mp = lock_mount(path); |
2219 | err = PTR_ERR(mp); | |
2220 | if (IS_ERR(mp)) | |
b12cea91 AV |
2221 | goto out; |
2222 | ||
87129cc0 | 2223 | old = real_mount(old_path.mnt); |
84d17192 | 2224 | parent = real_mount(path->mnt); |
87129cc0 | 2225 | |
1da177e4 | 2226 | err = -EINVAL; |
fc7be130 | 2227 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2228 | goto out2; |
9676f0c6 | 2229 | |
e149ed2b AV |
2230 | if (!check_mnt(parent)) |
2231 | goto out2; | |
2232 | ||
2233 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2234 | goto out2; |
1da177e4 | 2235 | |
5ff9d8a6 EB |
2236 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2237 | goto out2; | |
2238 | ||
ccd48bc7 | 2239 | if (recurse) |
4ce5d2b1 | 2240 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2241 | else |
87129cc0 | 2242 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2243 | |
be34d1a3 DH |
2244 | if (IS_ERR(mnt)) { |
2245 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2246 | goto out2; |
be34d1a3 | 2247 | } |
ccd48bc7 | 2248 | |
5ff9d8a6 EB |
2249 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2250 | ||
84d17192 | 2251 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2252 | if (err) { |
719ea2fb | 2253 | lock_mount_hash(); |
e819f152 | 2254 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2255 | unlock_mount_hash(); |
5b83d2c5 | 2256 | } |
b12cea91 | 2257 | out2: |
84d17192 | 2258 | unlock_mount(mp); |
ccd48bc7 | 2259 | out: |
2d92ab3c | 2260 | path_put(&old_path); |
1da177e4 LT |
2261 | return err; |
2262 | } | |
2263 | ||
2e4b7fcd DH |
2264 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2265 | { | |
2266 | int error = 0; | |
2267 | int readonly_request = 0; | |
2268 | ||
2269 | if (ms_flags & MS_RDONLY) | |
2270 | readonly_request = 1; | |
2271 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2272 | return 0; | |
2273 | ||
2274 | if (readonly_request) | |
83adc753 | 2275 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2276 | else |
83adc753 | 2277 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2278 | return error; |
2279 | } | |
2280 | ||
1da177e4 LT |
2281 | /* |
2282 | * change filesystem flags. dir should be a physical root of filesystem. | |
2283 | * If you've mounted a non-root directory somewhere and want to do remount | |
2284 | * on it - tough luck. | |
2285 | */ | |
0a0d8a46 | 2286 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2287 | void *data) |
2288 | { | |
2289 | int err; | |
2d92ab3c | 2290 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2291 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2292 | |
143c8c91 | 2293 | if (!check_mnt(mnt)) |
1da177e4 LT |
2294 | return -EINVAL; |
2295 | ||
2d92ab3c | 2296 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2297 | return -EINVAL; |
2298 | ||
07b64558 EB |
2299 | /* Don't allow changing of locked mnt flags. |
2300 | * | |
2301 | * No locks need to be held here while testing the various | |
2302 | * MNT_LOCK flags because those flags can never be cleared | |
2303 | * once they are set. | |
2304 | */ | |
2305 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2306 | !(mnt_flags & MNT_READONLY)) { | |
2307 | return -EPERM; | |
2308 | } | |
9566d674 EB |
2309 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2310 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2311 | return -EPERM; |
9566d674 EB |
2312 | } |
2313 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2314 | !(mnt_flags & MNT_NOSUID)) { | |
2315 | return -EPERM; | |
2316 | } | |
2317 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2318 | !(mnt_flags & MNT_NOEXEC)) { | |
2319 | return -EPERM; | |
2320 | } | |
2321 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2322 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2323 | return -EPERM; | |
2324 | } | |
2325 | ||
ff36fe2c EP |
2326 | err = security_sb_remount(sb, data); |
2327 | if (err) | |
2328 | return err; | |
2329 | ||
1da177e4 | 2330 | down_write(&sb->s_umount); |
2e4b7fcd | 2331 | if (flags & MS_BIND) |
2d92ab3c | 2332 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
2333 | else if (!capable(CAP_SYS_ADMIN)) |
2334 | err = -EPERM; | |
4aa98cf7 | 2335 | else |
2e4b7fcd | 2336 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2337 | if (!err) { |
719ea2fb | 2338 | lock_mount_hash(); |
a6138db8 | 2339 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2340 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2341 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2342 | unlock_mount_hash(); |
0e55a7cc | 2343 | } |
6339dab8 | 2344 | up_write(&sb->s_umount); |
1da177e4 LT |
2345 | return err; |
2346 | } | |
2347 | ||
cbbe362c | 2348 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2349 | { |
315fc83e | 2350 | struct mount *p; |
909b0a88 | 2351 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2352 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2353 | return 1; |
2354 | } | |
2355 | return 0; | |
2356 | } | |
2357 | ||
808d4e3c | 2358 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2359 | { |
2d92ab3c | 2360 | struct path old_path, parent_path; |
676da58d | 2361 | struct mount *p; |
0fb54e50 | 2362 | struct mount *old; |
84d17192 | 2363 | struct mountpoint *mp; |
57eccb83 | 2364 | int err; |
1da177e4 LT |
2365 | if (!old_name || !*old_name) |
2366 | return -EINVAL; | |
2d92ab3c | 2367 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2368 | if (err) |
2369 | return err; | |
2370 | ||
84d17192 AV |
2371 | mp = lock_mount(path); |
2372 | err = PTR_ERR(mp); | |
2373 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2374 | goto out; |
2375 | ||
143c8c91 | 2376 | old = real_mount(old_path.mnt); |
fc7be130 | 2377 | p = real_mount(path->mnt); |
143c8c91 | 2378 | |
1da177e4 | 2379 | err = -EINVAL; |
fc7be130 | 2380 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2381 | goto out1; |
2382 | ||
5ff9d8a6 EB |
2383 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2384 | goto out1; | |
2385 | ||
1da177e4 | 2386 | err = -EINVAL; |
2d92ab3c | 2387 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2388 | goto out1; |
1da177e4 | 2389 | |
676da58d | 2390 | if (!mnt_has_parent(old)) |
21444403 | 2391 | goto out1; |
1da177e4 | 2392 | |
e36cb0b8 DH |
2393 | if (d_is_dir(path->dentry) != |
2394 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2395 | goto out1; |
2396 | /* | |
2397 | * Don't move a mount residing in a shared parent. | |
2398 | */ | |
fc7be130 | 2399 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2400 | goto out1; |
9676f0c6 RP |
2401 | /* |
2402 | * Don't move a mount tree containing unbindable mounts to a destination | |
2403 | * mount which is shared. | |
2404 | */ | |
fc7be130 | 2405 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2406 | goto out1; |
1da177e4 | 2407 | err = -ELOOP; |
fc7be130 | 2408 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2409 | if (p == old) |
21444403 | 2410 | goto out1; |
1da177e4 | 2411 | |
84d17192 | 2412 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2413 | if (err) |
21444403 | 2414 | goto out1; |
1da177e4 LT |
2415 | |
2416 | /* if the mount is moved, it should no longer be expire | |
2417 | * automatically */ | |
6776db3d | 2418 | list_del_init(&old->mnt_expire); |
1da177e4 | 2419 | out1: |
84d17192 | 2420 | unlock_mount(mp); |
1da177e4 | 2421 | out: |
1da177e4 | 2422 | if (!err) |
1a390689 | 2423 | path_put(&parent_path); |
2d92ab3c | 2424 | path_put(&old_path); |
1da177e4 LT |
2425 | return err; |
2426 | } | |
2427 | ||
9d412a43 AV |
2428 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2429 | { | |
2430 | int err; | |
2431 | const char *subtype = strchr(fstype, '.'); | |
2432 | if (subtype) { | |
2433 | subtype++; | |
2434 | err = -EINVAL; | |
2435 | if (!subtype[0]) | |
2436 | goto err; | |
2437 | } else | |
2438 | subtype = ""; | |
2439 | ||
2440 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2441 | err = -ENOMEM; | |
2442 | if (!mnt->mnt_sb->s_subtype) | |
2443 | goto err; | |
2444 | return mnt; | |
2445 | ||
2446 | err: | |
2447 | mntput(mnt); | |
2448 | return ERR_PTR(err); | |
2449 | } | |
2450 | ||
9d412a43 AV |
2451 | /* |
2452 | * add a mount into a namespace's mount tree | |
2453 | */ | |
95bc5f25 | 2454 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2455 | { |
84d17192 AV |
2456 | struct mountpoint *mp; |
2457 | struct mount *parent; | |
9d412a43 AV |
2458 | int err; |
2459 | ||
f2ebb3a9 | 2460 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2461 | |
84d17192 AV |
2462 | mp = lock_mount(path); |
2463 | if (IS_ERR(mp)) | |
2464 | return PTR_ERR(mp); | |
9d412a43 | 2465 | |
84d17192 | 2466 | parent = real_mount(path->mnt); |
9d412a43 | 2467 | err = -EINVAL; |
84d17192 | 2468 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2469 | /* that's acceptable only for automounts done in private ns */ |
2470 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2471 | goto unlock; | |
2472 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2473 | if (!parent->mnt_ns) |
156cacb1 AV |
2474 | goto unlock; |
2475 | } | |
9d412a43 AV |
2476 | |
2477 | /* Refuse the same filesystem on the same mount point */ | |
2478 | err = -EBUSY; | |
95bc5f25 | 2479 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2480 | path->mnt->mnt_root == path->dentry) |
2481 | goto unlock; | |
2482 | ||
2483 | err = -EINVAL; | |
e36cb0b8 | 2484 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2485 | goto unlock; |
2486 | ||
95bc5f25 | 2487 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2488 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2489 | |
2490 | unlock: | |
84d17192 | 2491 | unlock_mount(mp); |
9d412a43 AV |
2492 | return err; |
2493 | } | |
b1e75df4 | 2494 | |
8654df4e | 2495 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2496 | |
1da177e4 LT |
2497 | /* |
2498 | * create a new mount for userspace and request it to be added into the | |
2499 | * namespace's tree | |
2500 | */ | |
0c55cfc4 | 2501 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2502 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2503 | { |
0c55cfc4 | 2504 | struct file_system_type *type; |
1da177e4 | 2505 | struct vfsmount *mnt; |
15f9a3f3 | 2506 | int err; |
1da177e4 | 2507 | |
0c55cfc4 | 2508 | if (!fstype) |
1da177e4 LT |
2509 | return -EINVAL; |
2510 | ||
0c55cfc4 EB |
2511 | type = get_fs_type(fstype); |
2512 | if (!type) | |
2513 | return -ENODEV; | |
2514 | ||
0c55cfc4 EB |
2515 | mnt = vfs_kern_mount(type, flags, name, data); |
2516 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2517 | !mnt->mnt_sb->s_subtype) | |
2518 | mnt = fs_set_subtype(mnt, fstype); | |
2519 | ||
2520 | put_filesystem(type); | |
1da177e4 LT |
2521 | if (IS_ERR(mnt)) |
2522 | return PTR_ERR(mnt); | |
2523 | ||
8654df4e EB |
2524 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2525 | mntput(mnt); | |
2526 | return -EPERM; | |
2527 | } | |
2528 | ||
95bc5f25 | 2529 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2530 | if (err) |
2531 | mntput(mnt); | |
2532 | return err; | |
1da177e4 LT |
2533 | } |
2534 | ||
19a167af AV |
2535 | int finish_automount(struct vfsmount *m, struct path *path) |
2536 | { | |
6776db3d | 2537 | struct mount *mnt = real_mount(m); |
19a167af AV |
2538 | int err; |
2539 | /* The new mount record should have at least 2 refs to prevent it being | |
2540 | * expired before we get a chance to add it | |
2541 | */ | |
6776db3d | 2542 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2543 | |
2544 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2545 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2546 | err = -ELOOP; |
2547 | goto fail; | |
19a167af AV |
2548 | } |
2549 | ||
95bc5f25 | 2550 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2551 | if (!err) |
2552 | return 0; | |
2553 | fail: | |
2554 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2555 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2556 | namespace_lock(); |
6776db3d | 2557 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2558 | namespace_unlock(); |
19a167af | 2559 | } |
b1e75df4 AV |
2560 | mntput(m); |
2561 | mntput(m); | |
19a167af AV |
2562 | return err; |
2563 | } | |
2564 | ||
ea5b778a DH |
2565 | /** |
2566 | * mnt_set_expiry - Put a mount on an expiration list | |
2567 | * @mnt: The mount to list. | |
2568 | * @expiry_list: The list to add the mount to. | |
2569 | */ | |
2570 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2571 | { | |
97216be0 | 2572 | namespace_lock(); |
ea5b778a | 2573 | |
6776db3d | 2574 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2575 | |
97216be0 | 2576 | namespace_unlock(); |
ea5b778a DH |
2577 | } |
2578 | EXPORT_SYMBOL(mnt_set_expiry); | |
2579 | ||
1da177e4 LT |
2580 | /* |
2581 | * process a list of expirable mountpoints with the intent of discarding any | |
2582 | * mountpoints that aren't in use and haven't been touched since last we came | |
2583 | * here | |
2584 | */ | |
2585 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2586 | { | |
761d5c38 | 2587 | struct mount *mnt, *next; |
1da177e4 LT |
2588 | LIST_HEAD(graveyard); |
2589 | ||
2590 | if (list_empty(mounts)) | |
2591 | return; | |
2592 | ||
97216be0 | 2593 | namespace_lock(); |
719ea2fb | 2594 | lock_mount_hash(); |
1da177e4 LT |
2595 | |
2596 | /* extract from the expiration list every vfsmount that matches the | |
2597 | * following criteria: | |
2598 | * - only referenced by its parent vfsmount | |
2599 | * - still marked for expiry (marked on the last call here; marks are | |
2600 | * cleared by mntput()) | |
2601 | */ | |
6776db3d | 2602 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2603 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2604 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2605 | continue; |
6776db3d | 2606 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2607 | } |
bcc5c7d2 | 2608 | while (!list_empty(&graveyard)) { |
6776db3d | 2609 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2610 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2611 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2612 | } |
719ea2fb | 2613 | unlock_mount_hash(); |
3ab6abee | 2614 | namespace_unlock(); |
5528f911 TM |
2615 | } |
2616 | ||
2617 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2618 | ||
2619 | /* | |
2620 | * Ripoff of 'select_parent()' | |
2621 | * | |
2622 | * search the list of submounts for a given mountpoint, and move any | |
2623 | * shrinkable submounts to the 'graveyard' list. | |
2624 | */ | |
692afc31 | 2625 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2626 | { |
692afc31 | 2627 | struct mount *this_parent = parent; |
5528f911 TM |
2628 | struct list_head *next; |
2629 | int found = 0; | |
2630 | ||
2631 | repeat: | |
6b41d536 | 2632 | next = this_parent->mnt_mounts.next; |
5528f911 | 2633 | resume: |
6b41d536 | 2634 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2635 | struct list_head *tmp = next; |
6b41d536 | 2636 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2637 | |
2638 | next = tmp->next; | |
692afc31 | 2639 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2640 | continue; |
5528f911 TM |
2641 | /* |
2642 | * Descend a level if the d_mounts list is non-empty. | |
2643 | */ | |
6b41d536 | 2644 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2645 | this_parent = mnt; |
2646 | goto repeat; | |
2647 | } | |
1da177e4 | 2648 | |
1ab59738 | 2649 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2650 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2651 | found++; |
2652 | } | |
1da177e4 | 2653 | } |
5528f911 TM |
2654 | /* |
2655 | * All done at this level ... ascend and resume the search | |
2656 | */ | |
2657 | if (this_parent != parent) { | |
6b41d536 | 2658 | next = this_parent->mnt_child.next; |
0714a533 | 2659 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2660 | goto resume; |
2661 | } | |
2662 | return found; | |
2663 | } | |
2664 | ||
2665 | /* | |
2666 | * process a list of expirable mountpoints with the intent of discarding any | |
2667 | * submounts of a specific parent mountpoint | |
99b7db7b | 2668 | * |
48a066e7 | 2669 | * mount_lock must be held for write |
5528f911 | 2670 | */ |
b54b9be7 | 2671 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2672 | { |
2673 | LIST_HEAD(graveyard); | |
761d5c38 | 2674 | struct mount *m; |
5528f911 | 2675 | |
5528f911 | 2676 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2677 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2678 | while (!list_empty(&graveyard)) { |
761d5c38 | 2679 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2680 | mnt_expire); |
143c8c91 | 2681 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2682 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2683 | } |
2684 | } | |
1da177e4 LT |
2685 | } |
2686 | ||
1da177e4 LT |
2687 | /* |
2688 | * Some copy_from_user() implementations do not return the exact number of | |
2689 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2690 | * Note that this function differs from copy_from_user() in that it will oops | |
2691 | * on bad values of `to', rather than returning a short copy. | |
2692 | */ | |
b58fed8b RP |
2693 | static long exact_copy_from_user(void *to, const void __user * from, |
2694 | unsigned long n) | |
1da177e4 LT |
2695 | { |
2696 | char *t = to; | |
2697 | const char __user *f = from; | |
2698 | char c; | |
2699 | ||
2700 | if (!access_ok(VERIFY_READ, from, n)) | |
2701 | return n; | |
2702 | ||
2703 | while (n) { | |
2704 | if (__get_user(c, f)) { | |
2705 | memset(t, 0, n); | |
2706 | break; | |
2707 | } | |
2708 | *t++ = c; | |
2709 | f++; | |
2710 | n--; | |
2711 | } | |
2712 | return n; | |
2713 | } | |
2714 | ||
b40ef869 | 2715 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2716 | { |
2717 | int i; | |
1da177e4 | 2718 | unsigned long size; |
b40ef869 | 2719 | char *copy; |
b58fed8b | 2720 | |
1da177e4 | 2721 | if (!data) |
b40ef869 | 2722 | return NULL; |
1da177e4 | 2723 | |
b40ef869 AV |
2724 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2725 | if (!copy) | |
2726 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2727 | |
2728 | /* We only care that *some* data at the address the user | |
2729 | * gave us is valid. Just in case, we'll zero | |
2730 | * the remainder of the page. | |
2731 | */ | |
2732 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2733 | size = TASK_SIZE - (unsigned long)data; | |
2734 | if (size > PAGE_SIZE) | |
2735 | size = PAGE_SIZE; | |
2736 | ||
b40ef869 | 2737 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2738 | if (!i) { |
b40ef869 AV |
2739 | kfree(copy); |
2740 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2741 | } |
2742 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2743 | memset(copy + i, 0, PAGE_SIZE - i); |
2744 | return copy; | |
1da177e4 LT |
2745 | } |
2746 | ||
b8850d1f | 2747 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2748 | { |
b8850d1f | 2749 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2750 | } |
2751 | ||
1da177e4 LT |
2752 | /* |
2753 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2754 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2755 | * | |
2756 | * data is a (void *) that can point to any structure up to | |
2757 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2758 | * information (or be NULL). | |
2759 | * | |
2760 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2761 | * When the flags word was introduced its top half was required | |
2762 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2763 | * Therefore, if this magic number is present, it carries no information | |
2764 | * and must be discarded. | |
2765 | */ | |
5e6123f3 | 2766 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2767 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2768 | { |
2d92ab3c | 2769 | struct path path; |
1da177e4 LT |
2770 | int retval = 0; |
2771 | int mnt_flags = 0; | |
2772 | ||
2773 | /* Discard magic */ | |
2774 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2775 | flags &= ~MS_MGC_MSK; | |
2776 | ||
2777 | /* Basic sanity checks */ | |
1da177e4 LT |
2778 | if (data_page) |
2779 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2780 | ||
a27ab9f2 | 2781 | /* ... and get the mountpoint */ |
5e6123f3 | 2782 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2783 | if (retval) |
2784 | return retval; | |
2785 | ||
2786 | retval = security_sb_mount(dev_name, &path, | |
2787 | type_page, flags, data_page); | |
0d5cadb8 AV |
2788 | if (!retval && !may_mount()) |
2789 | retval = -EPERM; | |
9e8925b6 JL |
2790 | if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) |
2791 | retval = -EPERM; | |
a27ab9f2 TH |
2792 | if (retval) |
2793 | goto dput_out; | |
2794 | ||
613cbe3d AK |
2795 | /* Default to relatime unless overriden */ |
2796 | if (!(flags & MS_NOATIME)) | |
2797 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2798 | |
1da177e4 LT |
2799 | /* Separate the per-mountpoint flags */ |
2800 | if (flags & MS_NOSUID) | |
2801 | mnt_flags |= MNT_NOSUID; | |
2802 | if (flags & MS_NODEV) | |
2803 | mnt_flags |= MNT_NODEV; | |
2804 | if (flags & MS_NOEXEC) | |
2805 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2806 | if (flags & MS_NOATIME) |
2807 | mnt_flags |= MNT_NOATIME; | |
2808 | if (flags & MS_NODIRATIME) | |
2809 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2810 | if (flags & MS_STRICTATIME) |
2811 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2812 | if (flags & MS_RDONLY) |
2813 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2814 | |
ffbc6f0e EB |
2815 | /* The default atime for remount is preservation */ |
2816 | if ((flags & MS_REMOUNT) && | |
2817 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2818 | MS_STRICTATIME)) == 0)) { | |
2819 | mnt_flags &= ~MNT_ATIME_MASK; | |
2820 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2821 | } | |
2822 | ||
7a4dec53 | 2823 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 | 2824 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
93faccbb | 2825 | MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT); |
1da177e4 | 2826 | |
1da177e4 | 2827 | if (flags & MS_REMOUNT) |
2d92ab3c | 2828 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2829 | data_page); |
2830 | else if (flags & MS_BIND) | |
2d92ab3c | 2831 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2832 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2833 | retval = do_change_type(&path, flags); |
1da177e4 | 2834 | else if (flags & MS_MOVE) |
2d92ab3c | 2835 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2836 | else |
2d92ab3c | 2837 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2838 | dev_name, data_page); |
2839 | dput_out: | |
2d92ab3c | 2840 | path_put(&path); |
1da177e4 LT |
2841 | return retval; |
2842 | } | |
2843 | ||
537f7ccb EB |
2844 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2845 | { | |
2846 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2847 | } | |
2848 | ||
2849 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2850 | { | |
2851 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2852 | } | |
2853 | ||
771b1371 EB |
2854 | static void free_mnt_ns(struct mnt_namespace *ns) |
2855 | { | |
6344c433 | 2856 | ns_free_inum(&ns->ns); |
537f7ccb | 2857 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2858 | put_user_ns(ns->user_ns); |
2859 | kfree(ns); | |
2860 | } | |
2861 | ||
8823c079 EB |
2862 | /* |
2863 | * Assign a sequence number so we can detect when we attempt to bind | |
2864 | * mount a reference to an older mount namespace into the current | |
2865 | * mount namespace, preventing reference counting loops. A 64bit | |
2866 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2867 | * is effectively never, so we can ignore the possibility. | |
2868 | */ | |
2869 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2870 | ||
771b1371 | 2871 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2872 | { |
2873 | struct mnt_namespace *new_ns; | |
537f7ccb | 2874 | struct ucounts *ucounts; |
98f842e6 | 2875 | int ret; |
cf8d2c11 | 2876 | |
537f7ccb EB |
2877 | ucounts = inc_mnt_namespaces(user_ns); |
2878 | if (!ucounts) | |
df75e774 | 2879 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2880 | |
cf8d2c11 | 2881 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2882 | if (!new_ns) { |
2883 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2884 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2885 | } |
6344c433 | 2886 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2887 | if (ret) { |
2888 | kfree(new_ns); | |
537f7ccb | 2889 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2890 | return ERR_PTR(ret); |
2891 | } | |
33c42940 | 2892 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2893 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2894 | atomic_set(&new_ns->count, 1); |
2895 | new_ns->root = NULL; | |
2896 | INIT_LIST_HEAD(&new_ns->list); | |
2897 | init_waitqueue_head(&new_ns->poll); | |
2898 | new_ns->event = 0; | |
771b1371 | 2899 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2900 | new_ns->ucounts = ucounts; |
d2921684 EB |
2901 | new_ns->mounts = 0; |
2902 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2903 | return new_ns; |
2904 | } | |
2905 | ||
0766f788 | 2906 | __latent_entropy |
9559f689 AV |
2907 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2908 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2909 | { |
6b3286ed | 2910 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2911 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2912 | struct mount *p, *q; |
9559f689 | 2913 | struct mount *old; |
cb338d06 | 2914 | struct mount *new; |
7a472ef4 | 2915 | int copy_flags; |
1da177e4 | 2916 | |
9559f689 AV |
2917 | BUG_ON(!ns); |
2918 | ||
2919 | if (likely(!(flags & CLONE_NEWNS))) { | |
2920 | get_mnt_ns(ns); | |
2921 | return ns; | |
2922 | } | |
2923 | ||
2924 | old = ns->root; | |
2925 | ||
771b1371 | 2926 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2927 | if (IS_ERR(new_ns)) |
2928 | return new_ns; | |
1da177e4 | 2929 | |
97216be0 | 2930 | namespace_lock(); |
1da177e4 | 2931 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2932 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2933 | if (user_ns != ns->user_ns) |
132c94e3 | 2934 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2935 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2936 | if (IS_ERR(new)) { |
328e6d90 | 2937 | namespace_unlock(); |
771b1371 | 2938 | free_mnt_ns(new_ns); |
be34d1a3 | 2939 | return ERR_CAST(new); |
1da177e4 | 2940 | } |
be08d6d2 | 2941 | new_ns->root = new; |
1a4eeaf2 | 2942 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2943 | |
2944 | /* | |
2945 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2946 | * as belonging to new namespace. We have already acquired a private | |
2947 | * fs_struct, so tsk->fs->lock is not needed. | |
2948 | */ | |
909b0a88 | 2949 | p = old; |
cb338d06 | 2950 | q = new; |
1da177e4 | 2951 | while (p) { |
143c8c91 | 2952 | q->mnt_ns = new_ns; |
d2921684 | 2953 | new_ns->mounts++; |
9559f689 AV |
2954 | if (new_fs) { |
2955 | if (&p->mnt == new_fs->root.mnt) { | |
2956 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2957 | rootmnt = &p->mnt; |
1da177e4 | 2958 | } |
9559f689 AV |
2959 | if (&p->mnt == new_fs->pwd.mnt) { |
2960 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2961 | pwdmnt = &p->mnt; |
1da177e4 | 2962 | } |
1da177e4 | 2963 | } |
909b0a88 AV |
2964 | p = next_mnt(p, old); |
2965 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2966 | if (!q) |
2967 | break; | |
2968 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2969 | p = next_mnt(p, old); | |
1da177e4 | 2970 | } |
328e6d90 | 2971 | namespace_unlock(); |
1da177e4 | 2972 | |
1da177e4 | 2973 | if (rootmnt) |
f03c6599 | 2974 | mntput(rootmnt); |
1da177e4 | 2975 | if (pwdmnt) |
f03c6599 | 2976 | mntput(pwdmnt); |
1da177e4 | 2977 | |
741a2951 | 2978 | return new_ns; |
1da177e4 LT |
2979 | } |
2980 | ||
cf8d2c11 TM |
2981 | /** |
2982 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2983 | * @mnt: pointer to the new root filesystem mountpoint | |
2984 | */ | |
1a4eeaf2 | 2985 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2986 | { |
771b1371 | 2987 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2988 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2989 | struct mount *mnt = real_mount(m); |
2990 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2991 | new_ns->root = mnt; |
d2921684 | 2992 | new_ns->mounts++; |
b1983cd8 | 2993 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2994 | } else { |
1a4eeaf2 | 2995 | mntput(m); |
cf8d2c11 TM |
2996 | } |
2997 | return new_ns; | |
2998 | } | |
cf8d2c11 | 2999 | |
ea441d11 AV |
3000 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3001 | { | |
3002 | struct mnt_namespace *ns; | |
d31da0f0 | 3003 | struct super_block *s; |
ea441d11 AV |
3004 | struct path path; |
3005 | int err; | |
3006 | ||
3007 | ns = create_mnt_ns(mnt); | |
3008 | if (IS_ERR(ns)) | |
3009 | return ERR_CAST(ns); | |
3010 | ||
3011 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3012 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3013 | ||
3014 | put_mnt_ns(ns); | |
3015 | ||
3016 | if (err) | |
3017 | return ERR_PTR(err); | |
3018 | ||
3019 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3020 | s = path.mnt->mnt_sb; |
3021 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3022 | mntput(path.mnt); |
3023 | /* lock the sucker */ | |
d31da0f0 | 3024 | down_write(&s->s_umount); |
ea441d11 AV |
3025 | /* ... and return the root of (sub)tree on it */ |
3026 | return path.dentry; | |
3027 | } | |
3028 | EXPORT_SYMBOL(mount_subtree); | |
3029 | ||
bdc480e3 HC |
3030 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3031 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3032 | { |
eca6f534 VN |
3033 | int ret; |
3034 | char *kernel_type; | |
eca6f534 | 3035 | char *kernel_dev; |
b40ef869 | 3036 | void *options; |
1da177e4 | 3037 | |
b8850d1f TG |
3038 | kernel_type = copy_mount_string(type); |
3039 | ret = PTR_ERR(kernel_type); | |
3040 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3041 | goto out_type; |
1da177e4 | 3042 | |
b8850d1f TG |
3043 | kernel_dev = copy_mount_string(dev_name); |
3044 | ret = PTR_ERR(kernel_dev); | |
3045 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3046 | goto out_dev; |
1da177e4 | 3047 | |
b40ef869 AV |
3048 | options = copy_mount_options(data); |
3049 | ret = PTR_ERR(options); | |
3050 | if (IS_ERR(options)) | |
eca6f534 | 3051 | goto out_data; |
1da177e4 | 3052 | |
b40ef869 | 3053 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3054 | |
b40ef869 | 3055 | kfree(options); |
eca6f534 VN |
3056 | out_data: |
3057 | kfree(kernel_dev); | |
3058 | out_dev: | |
eca6f534 VN |
3059 | kfree(kernel_type); |
3060 | out_type: | |
3061 | return ret; | |
1da177e4 LT |
3062 | } |
3063 | ||
afac7cba AV |
3064 | /* |
3065 | * Return true if path is reachable from root | |
3066 | * | |
48a066e7 | 3067 | * namespace_sem or mount_lock is held |
afac7cba | 3068 | */ |
643822b4 | 3069 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3070 | const struct path *root) |
3071 | { | |
643822b4 | 3072 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3073 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3074 | mnt = mnt->mnt_parent; |
afac7cba | 3075 | } |
643822b4 | 3076 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3077 | } |
3078 | ||
640eb7e7 | 3079 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3080 | { |
25ab4c9b | 3081 | bool res; |
48a066e7 | 3082 | read_seqlock_excl(&mount_lock); |
643822b4 | 3083 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3084 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3085 | return res; |
3086 | } | |
3087 | EXPORT_SYMBOL(path_is_under); | |
3088 | ||
1da177e4 LT |
3089 | /* |
3090 | * pivot_root Semantics: | |
3091 | * Moves the root file system of the current process to the directory put_old, | |
3092 | * makes new_root as the new root file system of the current process, and sets | |
3093 | * root/cwd of all processes which had them on the current root to new_root. | |
3094 | * | |
3095 | * Restrictions: | |
3096 | * The new_root and put_old must be directories, and must not be on the | |
3097 | * same file system as the current process root. The put_old must be | |
3098 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3099 | * pointed to by put_old must yield the same directory as new_root. No other | |
3100 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3101 | * | |
4a0d11fa NB |
3102 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3103 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3104 | * in this situation. | |
3105 | * | |
1da177e4 LT |
3106 | * Notes: |
3107 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3108 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3109 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3110 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3111 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3112 | * first. | |
3113 | */ | |
3480b257 HC |
3114 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3115 | const char __user *, put_old) | |
1da177e4 | 3116 | { |
2d8f3038 | 3117 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3118 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3119 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3120 | int error; |
3121 | ||
9b40bc90 | 3122 | if (!may_mount()) |
1da177e4 LT |
3123 | return -EPERM; |
3124 | ||
2d8f3038 | 3125 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3126 | if (error) |
3127 | goto out0; | |
1da177e4 | 3128 | |
2d8f3038 | 3129 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3130 | if (error) |
3131 | goto out1; | |
3132 | ||
2d8f3038 | 3133 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3134 | if (error) |
3135 | goto out2; | |
1da177e4 | 3136 | |
f7ad3c6b | 3137 | get_fs_root(current->fs, &root); |
84d17192 AV |
3138 | old_mp = lock_mount(&old); |
3139 | error = PTR_ERR(old_mp); | |
3140 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3141 | goto out3; |
3142 | ||
1da177e4 | 3143 | error = -EINVAL; |
419148da AV |
3144 | new_mnt = real_mount(new.mnt); |
3145 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3146 | old_mnt = real_mount(old.mnt); |
3147 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3148 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3149 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3150 | goto out4; |
143c8c91 | 3151 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3152 | goto out4; |
5ff9d8a6 EB |
3153 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3154 | goto out4; | |
1da177e4 | 3155 | error = -ENOENT; |
f3da392e | 3156 | if (d_unlinked(new.dentry)) |
b12cea91 | 3157 | goto out4; |
1da177e4 | 3158 | error = -EBUSY; |
84d17192 | 3159 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3160 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3161 | error = -EINVAL; |
8c3ee42e | 3162 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3163 | goto out4; /* not a mountpoint */ |
676da58d | 3164 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3165 | goto out4; /* not attached */ |
84d17192 | 3166 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3167 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3168 | goto out4; /* not a mountpoint */ |
676da58d | 3169 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3170 | goto out4; /* not attached */ |
4ac91378 | 3171 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3172 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3173 | goto out4; |
0d082601 EB |
3174 | /* make certain new is below the root */ |
3175 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3176 | goto out4; | |
84d17192 | 3177 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3178 | lock_mount_hash(); |
419148da AV |
3179 | detach_mnt(new_mnt, &parent_path); |
3180 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3181 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3182 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3183 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3184 | } | |
4ac91378 | 3185 | /* mount old root on put_old */ |
84d17192 | 3186 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3187 | /* mount new_root on / */ |
84d17192 | 3188 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3189 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3190 | /* A moved mount should not expire automatically */ |
3191 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3192 | put_mountpoint(root_mp); |
719ea2fb | 3193 | unlock_mount_hash(); |
2d8f3038 | 3194 | chroot_fs_refs(&root, &new); |
1da177e4 | 3195 | error = 0; |
b12cea91 | 3196 | out4: |
84d17192 | 3197 | unlock_mount(old_mp); |
b12cea91 AV |
3198 | if (!error) { |
3199 | path_put(&root_parent); | |
3200 | path_put(&parent_path); | |
3201 | } | |
3202 | out3: | |
8c3ee42e | 3203 | path_put(&root); |
b12cea91 | 3204 | out2: |
2d8f3038 | 3205 | path_put(&old); |
1da177e4 | 3206 | out1: |
2d8f3038 | 3207 | path_put(&new); |
1da177e4 | 3208 | out0: |
1da177e4 | 3209 | return error; |
1da177e4 LT |
3210 | } |
3211 | ||
3212 | static void __init init_mount_tree(void) | |
3213 | { | |
3214 | struct vfsmount *mnt; | |
6b3286ed | 3215 | struct mnt_namespace *ns; |
ac748a09 | 3216 | struct path root; |
0c55cfc4 | 3217 | struct file_system_type *type; |
1da177e4 | 3218 | |
0c55cfc4 EB |
3219 | type = get_fs_type("rootfs"); |
3220 | if (!type) | |
3221 | panic("Can't find rootfs type"); | |
3222 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3223 | put_filesystem(type); | |
1da177e4 LT |
3224 | if (IS_ERR(mnt)) |
3225 | panic("Can't create rootfs"); | |
b3e19d92 | 3226 | |
3b22edc5 TM |
3227 | ns = create_mnt_ns(mnt); |
3228 | if (IS_ERR(ns)) | |
1da177e4 | 3229 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3230 | |
3231 | init_task.nsproxy->mnt_ns = ns; | |
3232 | get_mnt_ns(ns); | |
3233 | ||
be08d6d2 AV |
3234 | root.mnt = mnt; |
3235 | root.dentry = mnt->mnt_root; | |
da362b09 | 3236 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3237 | |
3238 | set_fs_pwd(current->fs, &root); | |
3239 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3240 | } |
3241 | ||
74bf17cf | 3242 | void __init mnt_init(void) |
1da177e4 | 3243 | { |
13f14b4d | 3244 | unsigned u; |
15a67dd8 | 3245 | int err; |
1da177e4 | 3246 | |
7d6fec45 | 3247 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3248 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3249 | |
0818bf27 | 3250 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3251 | sizeof(struct hlist_head), |
0818bf27 AV |
3252 | mhash_entries, 19, |
3253 | 0, | |
3254 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3255 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3256 | sizeof(struct hlist_head), | |
3257 | mphash_entries, 19, | |
3258 | 0, | |
3259 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3260 | |
84d17192 | 3261 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3262 | panic("Failed to allocate mount hash table\n"); |
3263 | ||
0818bf27 | 3264 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3265 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3266 | for (u = 0; u <= mp_hash_mask; u++) |
3267 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3268 | |
4b93dc9b TH |
3269 | kernfs_init(); |
3270 | ||
15a67dd8 RD |
3271 | err = sysfs_init(); |
3272 | if (err) | |
3273 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3274 | __func__, err); |
00d26666 GKH |
3275 | fs_kobj = kobject_create_and_add("fs", NULL); |
3276 | if (!fs_kobj) | |
8e24eea7 | 3277 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3278 | init_rootfs(); |
3279 | init_mount_tree(); | |
3280 | } | |
3281 | ||
616511d0 | 3282 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3283 | { |
d498b25a | 3284 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3285 | return; |
7b00ed6f | 3286 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3287 | free_mnt_ns(ns); |
1da177e4 | 3288 | } |
9d412a43 AV |
3289 | |
3290 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3291 | { | |
423e0ab0 TC |
3292 | struct vfsmount *mnt; |
3293 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3294 | if (!IS_ERR(mnt)) { | |
3295 | /* | |
3296 | * it is a longterm mount, don't release mnt until | |
3297 | * we unmount before file sys is unregistered | |
3298 | */ | |
f7a99c5b | 3299 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3300 | } |
3301 | return mnt; | |
9d412a43 AV |
3302 | } |
3303 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3304 | |
3305 | void kern_unmount(struct vfsmount *mnt) | |
3306 | { | |
3307 | /* release long term mount so mount point can be released */ | |
3308 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3309 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3310 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3311 | mntput(mnt); |
3312 | } | |
3313 | } | |
3314 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3315 | |
3316 | bool our_mnt(struct vfsmount *mnt) | |
3317 | { | |
143c8c91 | 3318 | return check_mnt(real_mount(mnt)); |
02125a82 | 3319 | } |
8823c079 | 3320 | |
3151527e EB |
3321 | bool current_chrooted(void) |
3322 | { | |
3323 | /* Does the current process have a non-standard root */ | |
3324 | struct path ns_root; | |
3325 | struct path fs_root; | |
3326 | bool chrooted; | |
3327 | ||
3328 | /* Find the namespace root */ | |
3329 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3330 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3331 | path_get(&ns_root); | |
3332 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3333 | ; | |
3334 | ||
3335 | get_fs_root(current->fs, &fs_root); | |
3336 | ||
3337 | chrooted = !path_equal(&fs_root, &ns_root); | |
3338 | ||
3339 | path_put(&fs_root); | |
3340 | path_put(&ns_root); | |
3341 | ||
3342 | return chrooted; | |
3343 | } | |
3344 | ||
8654df4e EB |
3345 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3346 | int *new_mnt_flags) | |
87a8ebd6 | 3347 | { |
8c6cf9cc | 3348 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3349 | struct mount *mnt; |
e51db735 | 3350 | bool visible = false; |
87a8ebd6 | 3351 | |
44bb4385 | 3352 | down_read(&namespace_sem); |
87a8ebd6 | 3353 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3354 | struct mount *child; |
77b1a97d EB |
3355 | int mnt_flags; |
3356 | ||
8654df4e | 3357 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3358 | continue; |
3359 | ||
7e96c1b0 EB |
3360 | /* This mount is not fully visible if it's root directory |
3361 | * is not the root directory of the filesystem. | |
3362 | */ | |
3363 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3364 | continue; | |
3365 | ||
a1935c17 | 3366 | /* A local view of the mount flags */ |
77b1a97d | 3367 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3368 | |
695e9df0 EB |
3369 | /* Don't miss readonly hidden in the superblock flags */ |
3370 | if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) | |
3371 | mnt_flags |= MNT_LOCK_READONLY; | |
3372 | ||
8c6cf9cc EB |
3373 | /* Verify the mount flags are equal to or more permissive |
3374 | * than the proposed new mount. | |
3375 | */ | |
77b1a97d | 3376 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3377 | !(new_flags & MNT_READONLY)) |
3378 | continue; | |
77b1a97d EB |
3379 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3380 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3381 | continue; |
3382 | ||
ceeb0e5d EB |
3383 | /* This mount is not fully visible if there are any |
3384 | * locked child mounts that cover anything except for | |
3385 | * empty directories. | |
e51db735 EB |
3386 | */ |
3387 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3388 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3389 | /* Only worry about locked mounts */ |
d71ed6c9 | 3390 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3391 | continue; |
7236c85e EB |
3392 | /* Is the directory permanetly empty? */ |
3393 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3394 | goto next; |
87a8ebd6 | 3395 | } |
8c6cf9cc | 3396 | /* Preserve the locked attributes */ |
77b1a97d | 3397 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3398 | MNT_LOCK_ATIME); |
e51db735 EB |
3399 | visible = true; |
3400 | goto found; | |
3401 | next: ; | |
87a8ebd6 | 3402 | } |
e51db735 | 3403 | found: |
44bb4385 | 3404 | up_read(&namespace_sem); |
e51db735 | 3405 | return visible; |
87a8ebd6 EB |
3406 | } |
3407 | ||
8654df4e EB |
3408 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3409 | { | |
a1935c17 | 3410 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3411 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3412 | unsigned long s_iflags; | |
3413 | ||
3414 | if (ns->user_ns == &init_user_ns) | |
3415 | return false; | |
3416 | ||
3417 | /* Can this filesystem be too revealing? */ | |
3418 | s_iflags = mnt->mnt_sb->s_iflags; | |
3419 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3420 | return false; | |
3421 | ||
a1935c17 EB |
3422 | if ((s_iflags & required_iflags) != required_iflags) { |
3423 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3424 | required_iflags); | |
3425 | return true; | |
3426 | } | |
3427 | ||
8654df4e EB |
3428 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3429 | } | |
3430 | ||
380cf5ba AL |
3431 | bool mnt_may_suid(struct vfsmount *mnt) |
3432 | { | |
3433 | /* | |
3434 | * Foreign mounts (accessed via fchdir or through /proc | |
3435 | * symlinks) are always treated as if they are nosuid. This | |
3436 | * prevents namespaces from trusting potentially unsafe | |
3437 | * suid/sgid bits, file caps, or security labels that originate | |
3438 | * in other namespaces. | |
3439 | */ | |
3440 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3441 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3442 | } | |
3443 | ||
64964528 | 3444 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3445 | { |
58be2825 | 3446 | struct ns_common *ns = NULL; |
8823c079 EB |
3447 | struct nsproxy *nsproxy; |
3448 | ||
728dba3a EB |
3449 | task_lock(task); |
3450 | nsproxy = task->nsproxy; | |
8823c079 | 3451 | if (nsproxy) { |
58be2825 AV |
3452 | ns = &nsproxy->mnt_ns->ns; |
3453 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3454 | } |
728dba3a | 3455 | task_unlock(task); |
8823c079 EB |
3456 | |
3457 | return ns; | |
3458 | } | |
3459 | ||
64964528 | 3460 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3461 | { |
58be2825 | 3462 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3463 | } |
3464 | ||
64964528 | 3465 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3466 | { |
3467 | struct fs_struct *fs = current->fs; | |
58be2825 | 3468 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3469 | struct path root; |
3470 | ||
0c55cfc4 | 3471 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3472 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3473 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3474 | return -EPERM; |
8823c079 EB |
3475 | |
3476 | if (fs->users != 1) | |
3477 | return -EINVAL; | |
3478 | ||
3479 | get_mnt_ns(mnt_ns); | |
3480 | put_mnt_ns(nsproxy->mnt_ns); | |
3481 | nsproxy->mnt_ns = mnt_ns; | |
3482 | ||
3483 | /* Find the root */ | |
3484 | root.mnt = &mnt_ns->root->mnt; | |
3485 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3486 | path_get(&root); | |
3487 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3488 | ; | |
3489 | ||
3490 | /* Update the pwd and root */ | |
3491 | set_fs_pwd(fs, &root); | |
3492 | set_fs_root(fs, &root); | |
3493 | ||
3494 | path_put(&root); | |
3495 | return 0; | |
3496 | } | |
3497 | ||
bcac25a5 AV |
3498 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3499 | { | |
3500 | return to_mnt_ns(ns)->user_ns; | |
3501 | } | |
3502 | ||
8823c079 EB |
3503 | const struct proc_ns_operations mntns_operations = { |
3504 | .name = "mnt", | |
3505 | .type = CLONE_NEWNS, | |
3506 | .get = mntns_get, | |
3507 | .put = mntns_put, | |
3508 | .install = mntns_install, | |
bcac25a5 | 3509 | .owner = mntns_owner, |
8823c079 | 3510 | }; |