]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
9ea459e1 | 26 | #include <linux/task_work.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
d2921684 EB |
30 | /* Maximum number of mounts in a mount namespace */ |
31 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
32 | ||
0818bf27 AV |
33 | static unsigned int m_hash_mask __read_mostly; |
34 | static unsigned int m_hash_shift __read_mostly; | |
35 | static unsigned int mp_hash_mask __read_mostly; | |
36 | static unsigned int mp_hash_shift __read_mostly; | |
37 | ||
38 | static __initdata unsigned long mhash_entries; | |
39 | static int __init set_mhash_entries(char *str) | |
40 | { | |
41 | if (!str) | |
42 | return 0; | |
43 | mhash_entries = simple_strtoul(str, &str, 0); | |
44 | return 1; | |
45 | } | |
46 | __setup("mhash_entries=", set_mhash_entries); | |
47 | ||
48 | static __initdata unsigned long mphash_entries; | |
49 | static int __init set_mphash_entries(char *str) | |
50 | { | |
51 | if (!str) | |
52 | return 0; | |
53 | mphash_entries = simple_strtoul(str, &str, 0); | |
54 | return 1; | |
55 | } | |
56 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 57 | |
c7999c36 | 58 | static u64 event; |
73cd49ec | 59 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 60 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 61 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
62 | static int mnt_id_start = 0; |
63 | static int mnt_group_start = 1; | |
1da177e4 | 64 | |
38129a13 | 65 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 66 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 67 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 68 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 69 | |
f87fd4c2 | 70 | /* /sys/fs */ |
00d26666 GKH |
71 | struct kobject *fs_kobj; |
72 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 73 | |
99b7db7b NP |
74 | /* |
75 | * vfsmount lock may be taken for read to prevent changes to the | |
76 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
77 | * up the tree. | |
78 | * | |
79 | * It should be taken for write in all cases where the vfsmount | |
80 | * tree or hash is modified or when a vfsmount structure is modified. | |
81 | */ | |
48a066e7 | 82 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 83 | |
38129a13 | 84 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 85 | { |
b58fed8b RP |
86 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
87 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
88 | tmp = tmp + (tmp >> m_hash_shift); |
89 | return &mount_hashtable[tmp & m_hash_mask]; | |
90 | } | |
91 | ||
92 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
93 | { | |
94 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
95 | tmp = tmp + (tmp >> mp_hash_shift); | |
96 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
97 | } |
98 | ||
b105e270 | 99 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
100 | { |
101 | int res; | |
102 | ||
103 | retry: | |
104 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 105 | spin_lock(&mnt_id_lock); |
15169fe7 | 106 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 107 | if (!res) |
15169fe7 | 108 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 109 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
110 | if (res == -EAGAIN) |
111 | goto retry; | |
112 | ||
113 | return res; | |
114 | } | |
115 | ||
b105e270 | 116 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 117 | { |
15169fe7 | 118 | int id = mnt->mnt_id; |
99b7db7b | 119 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
120 | ida_remove(&mnt_id_ida, id); |
121 | if (mnt_id_start > id) | |
122 | mnt_id_start = id; | |
99b7db7b | 123 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
124 | } |
125 | ||
719f5d7f MS |
126 | /* |
127 | * Allocate a new peer group ID | |
128 | * | |
129 | * mnt_group_ida is protected by namespace_sem | |
130 | */ | |
4b8b21f4 | 131 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 132 | { |
f21f6220 AV |
133 | int res; |
134 | ||
719f5d7f MS |
135 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
136 | return -ENOMEM; | |
137 | ||
f21f6220 AV |
138 | res = ida_get_new_above(&mnt_group_ida, |
139 | mnt_group_start, | |
15169fe7 | 140 | &mnt->mnt_group_id); |
f21f6220 | 141 | if (!res) |
15169fe7 | 142 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
143 | |
144 | return res; | |
719f5d7f MS |
145 | } |
146 | ||
147 | /* | |
148 | * Release a peer group ID | |
149 | */ | |
4b8b21f4 | 150 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 151 | { |
15169fe7 | 152 | int id = mnt->mnt_group_id; |
f21f6220 AV |
153 | ida_remove(&mnt_group_ida, id); |
154 | if (mnt_group_start > id) | |
155 | mnt_group_start = id; | |
15169fe7 | 156 | mnt->mnt_group_id = 0; |
719f5d7f MS |
157 | } |
158 | ||
b3e19d92 NP |
159 | /* |
160 | * vfsmount lock must be held for read | |
161 | */ | |
83adc753 | 162 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
163 | { |
164 | #ifdef CONFIG_SMP | |
68e8a9fe | 165 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
166 | #else |
167 | preempt_disable(); | |
68e8a9fe | 168 | mnt->mnt_count += n; |
b3e19d92 NP |
169 | preempt_enable(); |
170 | #endif | |
171 | } | |
172 | ||
b3e19d92 NP |
173 | /* |
174 | * vfsmount lock must be held for write | |
175 | */ | |
83adc753 | 176 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
177 | { |
178 | #ifdef CONFIG_SMP | |
f03c6599 | 179 | unsigned int count = 0; |
b3e19d92 NP |
180 | int cpu; |
181 | ||
182 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 183 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
184 | } |
185 | ||
186 | return count; | |
187 | #else | |
68e8a9fe | 188 | return mnt->mnt_count; |
b3e19d92 NP |
189 | #endif |
190 | } | |
191 | ||
87b95ce0 AV |
192 | static void drop_mountpoint(struct fs_pin *p) |
193 | { | |
194 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
195 | dput(m->mnt_ex_mountpoint); | |
196 | pin_remove(p); | |
197 | mntput(&m->mnt); | |
198 | } | |
199 | ||
b105e270 | 200 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 201 | { |
c63181e6 AV |
202 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
203 | if (mnt) { | |
73cd49ec MS |
204 | int err; |
205 | ||
c63181e6 | 206 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
207 | if (err) |
208 | goto out_free_cache; | |
209 | ||
210 | if (name) { | |
fcc139ae | 211 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 212 | if (!mnt->mnt_devname) |
88b38782 | 213 | goto out_free_id; |
73cd49ec MS |
214 | } |
215 | ||
b3e19d92 | 216 | #ifdef CONFIG_SMP |
c63181e6 AV |
217 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
218 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
219 | goto out_free_devname; |
220 | ||
c63181e6 | 221 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 222 | #else |
c63181e6 AV |
223 | mnt->mnt_count = 1; |
224 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
225 | #endif |
226 | ||
38129a13 | 227 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
228 | INIT_LIST_HEAD(&mnt->mnt_child); |
229 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
230 | INIT_LIST_HEAD(&mnt->mnt_list); | |
231 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
232 | INIT_LIST_HEAD(&mnt->mnt_share); | |
233 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
234 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 235 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
236 | #ifdef CONFIG_FSNOTIFY |
237 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 238 | #endif |
87b95ce0 | 239 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 240 | } |
c63181e6 | 241 | return mnt; |
88b38782 | 242 | |
d3ef3d73 | 243 | #ifdef CONFIG_SMP |
244 | out_free_devname: | |
fcc139ae | 245 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 246 | #endif |
88b38782 | 247 | out_free_id: |
c63181e6 | 248 | mnt_free_id(mnt); |
88b38782 | 249 | out_free_cache: |
c63181e6 | 250 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 251 | return NULL; |
1da177e4 LT |
252 | } |
253 | ||
3d733633 DH |
254 | /* |
255 | * Most r/o checks on a fs are for operations that take | |
256 | * discrete amounts of time, like a write() or unlink(). | |
257 | * We must keep track of when those operations start | |
258 | * (for permission checks) and when they end, so that | |
259 | * we can determine when writes are able to occur to | |
260 | * a filesystem. | |
261 | */ | |
262 | /* | |
263 | * __mnt_is_readonly: check whether a mount is read-only | |
264 | * @mnt: the mount to check for its write status | |
265 | * | |
266 | * This shouldn't be used directly ouside of the VFS. | |
267 | * It does not guarantee that the filesystem will stay | |
268 | * r/w, just that it is right *now*. This can not and | |
269 | * should not be used in place of IS_RDONLY(inode). | |
270 | * mnt_want/drop_write() will _keep_ the filesystem | |
271 | * r/w. | |
272 | */ | |
273 | int __mnt_is_readonly(struct vfsmount *mnt) | |
274 | { | |
2e4b7fcd DH |
275 | if (mnt->mnt_flags & MNT_READONLY) |
276 | return 1; | |
277 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
278 | return 1; | |
279 | return 0; | |
3d733633 DH |
280 | } |
281 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
282 | ||
83adc753 | 283 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 284 | { |
285 | #ifdef CONFIG_SMP | |
68e8a9fe | 286 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 287 | #else |
68e8a9fe | 288 | mnt->mnt_writers++; |
d3ef3d73 | 289 | #endif |
290 | } | |
3d733633 | 291 | |
83adc753 | 292 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 293 | { |
d3ef3d73 | 294 | #ifdef CONFIG_SMP |
68e8a9fe | 295 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 296 | #else |
68e8a9fe | 297 | mnt->mnt_writers--; |
d3ef3d73 | 298 | #endif |
3d733633 | 299 | } |
3d733633 | 300 | |
83adc753 | 301 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 302 | { |
d3ef3d73 | 303 | #ifdef CONFIG_SMP |
304 | unsigned int count = 0; | |
3d733633 | 305 | int cpu; |
3d733633 DH |
306 | |
307 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 308 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 309 | } |
3d733633 | 310 | |
d3ef3d73 | 311 | return count; |
312 | #else | |
313 | return mnt->mnt_writers; | |
314 | #endif | |
3d733633 DH |
315 | } |
316 | ||
4ed5e82f MS |
317 | static int mnt_is_readonly(struct vfsmount *mnt) |
318 | { | |
319 | if (mnt->mnt_sb->s_readonly_remount) | |
320 | return 1; | |
321 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
322 | smp_rmb(); | |
323 | return __mnt_is_readonly(mnt); | |
324 | } | |
325 | ||
8366025e | 326 | /* |
eb04c282 JK |
327 | * Most r/o & frozen checks on a fs are for operations that take discrete |
328 | * amounts of time, like a write() or unlink(). We must keep track of when | |
329 | * those operations start (for permission checks) and when they end, so that we | |
330 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
331 | */ |
332 | /** | |
eb04c282 | 333 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 334 | * @m: the mount on which to take a write |
8366025e | 335 | * |
eb04c282 JK |
336 | * This tells the low-level filesystem that a write is about to be performed to |
337 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
338 | * returning success. This operation does not protect against filesystem being | |
339 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
340 | * called. This is effectively a refcount. | |
8366025e | 341 | */ |
eb04c282 | 342 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 343 | { |
83adc753 | 344 | struct mount *mnt = real_mount(m); |
3d733633 | 345 | int ret = 0; |
3d733633 | 346 | |
d3ef3d73 | 347 | preempt_disable(); |
c6653a83 | 348 | mnt_inc_writers(mnt); |
d3ef3d73 | 349 | /* |
c6653a83 | 350 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 351 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
352 | * incremented count after it has set MNT_WRITE_HOLD. | |
353 | */ | |
354 | smp_mb(); | |
1e75529e | 355 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 356 | cpu_relax(); |
357 | /* | |
358 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
359 | * be set to match its requirements. So we must not load that until | |
360 | * MNT_WRITE_HOLD is cleared. | |
361 | */ | |
362 | smp_rmb(); | |
4ed5e82f | 363 | if (mnt_is_readonly(m)) { |
c6653a83 | 364 | mnt_dec_writers(mnt); |
3d733633 | 365 | ret = -EROFS; |
3d733633 | 366 | } |
d3ef3d73 | 367 | preempt_enable(); |
eb04c282 JK |
368 | |
369 | return ret; | |
370 | } | |
371 | ||
372 | /** | |
373 | * mnt_want_write - get write access to a mount | |
374 | * @m: the mount on which to take a write | |
375 | * | |
376 | * This tells the low-level filesystem that a write is about to be performed to | |
377 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
378 | * is not frozen) before returning success. When the write operation is | |
379 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
380 | */ | |
381 | int mnt_want_write(struct vfsmount *m) | |
382 | { | |
383 | int ret; | |
384 | ||
385 | sb_start_write(m->mnt_sb); | |
386 | ret = __mnt_want_write(m); | |
387 | if (ret) | |
388 | sb_end_write(m->mnt_sb); | |
3d733633 | 389 | return ret; |
8366025e DH |
390 | } |
391 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
392 | ||
96029c4e | 393 | /** |
394 | * mnt_clone_write - get write access to a mount | |
395 | * @mnt: the mount on which to take a write | |
396 | * | |
397 | * This is effectively like mnt_want_write, except | |
398 | * it must only be used to take an extra write reference | |
399 | * on a mountpoint that we already know has a write reference | |
400 | * on it. This allows some optimisation. | |
401 | * | |
402 | * After finished, mnt_drop_write must be called as usual to | |
403 | * drop the reference. | |
404 | */ | |
405 | int mnt_clone_write(struct vfsmount *mnt) | |
406 | { | |
407 | /* superblock may be r/o */ | |
408 | if (__mnt_is_readonly(mnt)) | |
409 | return -EROFS; | |
410 | preempt_disable(); | |
83adc753 | 411 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 412 | preempt_enable(); |
413 | return 0; | |
414 | } | |
415 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
416 | ||
417 | /** | |
eb04c282 | 418 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 419 | * @file: the file who's mount on which to take a write |
420 | * | |
eb04c282 | 421 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 422 | * do some optimisations if the file is open for write already |
423 | */ | |
eb04c282 | 424 | int __mnt_want_write_file(struct file *file) |
96029c4e | 425 | { |
83f936c7 | 426 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 427 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 428 | else |
429 | return mnt_clone_write(file->f_path.mnt); | |
430 | } | |
eb04c282 JK |
431 | |
432 | /** | |
433 | * mnt_want_write_file - get write access to a file's mount | |
434 | * @file: the file who's mount on which to take a write | |
435 | * | |
436 | * This is like mnt_want_write, but it takes a file and can | |
437 | * do some optimisations if the file is open for write already | |
438 | */ | |
439 | int mnt_want_write_file(struct file *file) | |
440 | { | |
441 | int ret; | |
442 | ||
443 | sb_start_write(file->f_path.mnt->mnt_sb); | |
444 | ret = __mnt_want_write_file(file); | |
445 | if (ret) | |
446 | sb_end_write(file->f_path.mnt->mnt_sb); | |
447 | return ret; | |
448 | } | |
96029c4e | 449 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
450 | ||
8366025e | 451 | /** |
eb04c282 | 452 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
453 | * @mnt: the mount on which to give up write access |
454 | * | |
455 | * Tells the low-level filesystem that we are done | |
456 | * performing writes to it. Must be matched with | |
eb04c282 | 457 | * __mnt_want_write() call above. |
8366025e | 458 | */ |
eb04c282 | 459 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 460 | { |
d3ef3d73 | 461 | preempt_disable(); |
83adc753 | 462 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 463 | preempt_enable(); |
8366025e | 464 | } |
e14748e8 | 465 | EXPORT_SYMBOL_GPL(__mnt_drop_write); |
eb04c282 JK |
466 | |
467 | /** | |
468 | * mnt_drop_write - give up write access to a mount | |
469 | * @mnt: the mount on which to give up write access | |
470 | * | |
471 | * Tells the low-level filesystem that we are done performing writes to it and | |
472 | * also allows filesystem to be frozen again. Must be matched with | |
473 | * mnt_want_write() call above. | |
474 | */ | |
475 | void mnt_drop_write(struct vfsmount *mnt) | |
476 | { | |
477 | __mnt_drop_write(mnt); | |
478 | sb_end_write(mnt->mnt_sb); | |
479 | } | |
8366025e DH |
480 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
481 | ||
eb04c282 JK |
482 | void __mnt_drop_write_file(struct file *file) |
483 | { | |
484 | __mnt_drop_write(file->f_path.mnt); | |
485 | } | |
486 | ||
2a79f17e AV |
487 | void mnt_drop_write_file(struct file *file) |
488 | { | |
489 | mnt_drop_write(file->f_path.mnt); | |
490 | } | |
491 | EXPORT_SYMBOL(mnt_drop_write_file); | |
492 | ||
83adc753 | 493 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 494 | { |
3d733633 DH |
495 | int ret = 0; |
496 | ||
719ea2fb | 497 | lock_mount_hash(); |
83adc753 | 498 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 499 | /* |
d3ef3d73 | 500 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
501 | * should be visible before we do. | |
3d733633 | 502 | */ |
d3ef3d73 | 503 | smp_mb(); |
504 | ||
3d733633 | 505 | /* |
d3ef3d73 | 506 | * With writers on hold, if this value is zero, then there are |
507 | * definitely no active writers (although held writers may subsequently | |
508 | * increment the count, they'll have to wait, and decrement it after | |
509 | * seeing MNT_READONLY). | |
510 | * | |
511 | * It is OK to have counter incremented on one CPU and decremented on | |
512 | * another: the sum will add up correctly. The danger would be when we | |
513 | * sum up each counter, if we read a counter before it is incremented, | |
514 | * but then read another CPU's count which it has been subsequently | |
515 | * decremented from -- we would see more decrements than we should. | |
516 | * MNT_WRITE_HOLD protects against this scenario, because | |
517 | * mnt_want_write first increments count, then smp_mb, then spins on | |
518 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
519 | * we're counting up here. | |
3d733633 | 520 | */ |
c6653a83 | 521 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 522 | ret = -EBUSY; |
523 | else | |
83adc753 | 524 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 525 | /* |
526 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
527 | * that become unheld will see MNT_READONLY. | |
528 | */ | |
529 | smp_wmb(); | |
83adc753 | 530 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 531 | unlock_mount_hash(); |
3d733633 | 532 | return ret; |
8366025e | 533 | } |
8366025e | 534 | |
83adc753 | 535 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 536 | { |
719ea2fb | 537 | lock_mount_hash(); |
83adc753 | 538 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 539 | unlock_mount_hash(); |
2e4b7fcd DH |
540 | } |
541 | ||
4ed5e82f MS |
542 | int sb_prepare_remount_readonly(struct super_block *sb) |
543 | { | |
544 | struct mount *mnt; | |
545 | int err = 0; | |
546 | ||
8e8b8796 MS |
547 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
548 | if (atomic_long_read(&sb->s_remove_count)) | |
549 | return -EBUSY; | |
550 | ||
719ea2fb | 551 | lock_mount_hash(); |
4ed5e82f MS |
552 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
553 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
554 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
555 | smp_mb(); | |
556 | if (mnt_get_writers(mnt) > 0) { | |
557 | err = -EBUSY; | |
558 | break; | |
559 | } | |
560 | } | |
561 | } | |
8e8b8796 MS |
562 | if (!err && atomic_long_read(&sb->s_remove_count)) |
563 | err = -EBUSY; | |
564 | ||
4ed5e82f MS |
565 | if (!err) { |
566 | sb->s_readonly_remount = 1; | |
567 | smp_wmb(); | |
568 | } | |
569 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
570 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
571 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
572 | } | |
719ea2fb | 573 | unlock_mount_hash(); |
4ed5e82f MS |
574 | |
575 | return err; | |
576 | } | |
577 | ||
b105e270 | 578 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 579 | { |
fcc139ae | 580 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 581 | #ifdef CONFIG_SMP |
68e8a9fe | 582 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 583 | #endif |
b105e270 | 584 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
585 | } |
586 | ||
8ffcb32e DH |
587 | static void delayed_free_vfsmnt(struct rcu_head *head) |
588 | { | |
589 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
590 | } | |
591 | ||
48a066e7 | 592 | /* call under rcu_read_lock */ |
294d71ff | 593 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
594 | { |
595 | struct mount *mnt; | |
596 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 597 | return 1; |
48a066e7 | 598 | if (bastard == NULL) |
294d71ff | 599 | return 0; |
48a066e7 AV |
600 | mnt = real_mount(bastard); |
601 | mnt_add_count(mnt, 1); | |
602 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 603 | return 0; |
48a066e7 AV |
604 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
605 | mnt_add_count(mnt, -1); | |
294d71ff AV |
606 | return 1; |
607 | } | |
608 | return -1; | |
609 | } | |
610 | ||
611 | /* call under rcu_read_lock */ | |
612 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
613 | { | |
614 | int res = __legitimize_mnt(bastard, seq); | |
615 | if (likely(!res)) | |
616 | return true; | |
617 | if (unlikely(res < 0)) { | |
618 | rcu_read_unlock(); | |
619 | mntput(bastard); | |
620 | rcu_read_lock(); | |
48a066e7 | 621 | } |
48a066e7 AV |
622 | return false; |
623 | } | |
624 | ||
1da177e4 | 625 | /* |
474279dc | 626 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 627 | * call under rcu_read_lock() |
1da177e4 | 628 | */ |
474279dc | 629 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 630 | { |
38129a13 | 631 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
632 | struct mount *p; |
633 | ||
38129a13 | 634 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
635 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
636 | return p; | |
637 | return NULL; | |
638 | } | |
639 | ||
a05964f3 | 640 | /* |
f015f126 DH |
641 | * lookup_mnt - Return the first child mount mounted at path |
642 | * | |
643 | * "First" means first mounted chronologically. If you create the | |
644 | * following mounts: | |
645 | * | |
646 | * mount /dev/sda1 /mnt | |
647 | * mount /dev/sda2 /mnt | |
648 | * mount /dev/sda3 /mnt | |
649 | * | |
650 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
651 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
652 | * /dev/sda2, then /dev/sda3, then NULL. | |
653 | * | |
654 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 655 | */ |
ca71cf71 | 656 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 657 | { |
c7105365 | 658 | struct mount *child_mnt; |
48a066e7 AV |
659 | struct vfsmount *m; |
660 | unsigned seq; | |
99b7db7b | 661 | |
48a066e7 AV |
662 | rcu_read_lock(); |
663 | do { | |
664 | seq = read_seqbegin(&mount_lock); | |
665 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
666 | m = child_mnt ? &child_mnt->mnt : NULL; | |
667 | } while (!legitimize_mnt(m, seq)); | |
668 | rcu_read_unlock(); | |
669 | return m; | |
a05964f3 RP |
670 | } |
671 | ||
7af1364f EB |
672 | /* |
673 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
674 | * current mount namespace. | |
675 | * | |
676 | * The common case is dentries are not mountpoints at all and that | |
677 | * test is handled inline. For the slow case when we are actually | |
678 | * dealing with a mountpoint of some kind, walk through all of the | |
679 | * mounts in the current mount namespace and test to see if the dentry | |
680 | * is a mountpoint. | |
681 | * | |
682 | * The mount_hashtable is not usable in the context because we | |
683 | * need to identify all mounts that may be in the current mount | |
684 | * namespace not just a mount that happens to have some specified | |
685 | * parent mount. | |
686 | */ | |
687 | bool __is_local_mountpoint(struct dentry *dentry) | |
688 | { | |
689 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
690 | struct mount *mnt; | |
691 | bool is_covered = false; | |
692 | ||
693 | if (!d_mountpoint(dentry)) | |
694 | goto out; | |
695 | ||
696 | down_read(&namespace_sem); | |
697 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
698 | is_covered = (mnt->mnt_mountpoint == dentry); | |
699 | if (is_covered) | |
700 | break; | |
701 | } | |
702 | up_read(&namespace_sem); | |
703 | out: | |
704 | return is_covered; | |
705 | } | |
706 | ||
e2dfa935 | 707 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 708 | { |
0818bf27 | 709 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
710 | struct mountpoint *mp; |
711 | ||
0818bf27 | 712 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
713 | if (mp->m_dentry == dentry) { |
714 | /* might be worth a WARN_ON() */ | |
715 | if (d_unlinked(dentry)) | |
716 | return ERR_PTR(-ENOENT); | |
717 | mp->m_count++; | |
718 | return mp; | |
719 | } | |
720 | } | |
e2dfa935 EB |
721 | return NULL; |
722 | } | |
723 | ||
3895dbf8 | 724 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 725 | { |
3895dbf8 | 726 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 727 | int ret; |
84d17192 | 728 | |
3895dbf8 EB |
729 | if (d_mountpoint(dentry)) { |
730 | mountpoint: | |
731 | read_seqlock_excl(&mount_lock); | |
732 | mp = lookup_mountpoint(dentry); | |
733 | read_sequnlock_excl(&mount_lock); | |
734 | if (mp) | |
735 | goto done; | |
736 | } | |
737 | ||
738 | if (!new) | |
739 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
740 | if (!new) | |
84d17192 AV |
741 | return ERR_PTR(-ENOMEM); |
742 | ||
3895dbf8 EB |
743 | |
744 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 745 | ret = d_set_mounted(dentry); |
eed81007 | 746 | |
3895dbf8 EB |
747 | /* Someone else set d_mounted? */ |
748 | if (ret == -EBUSY) | |
749 | goto mountpoint; | |
750 | ||
751 | /* The dentry is not available as a mountpoint? */ | |
752 | mp = ERR_PTR(ret); | |
753 | if (ret) | |
754 | goto done; | |
755 | ||
756 | /* Add the new mountpoint to the hash table */ | |
757 | read_seqlock_excl(&mount_lock); | |
758 | new->m_dentry = dentry; | |
759 | new->m_count = 1; | |
760 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
761 | INIT_HLIST_HEAD(&new->m_list); | |
762 | read_sequnlock_excl(&mount_lock); | |
763 | ||
764 | mp = new; | |
765 | new = NULL; | |
766 | done: | |
767 | kfree(new); | |
84d17192 AV |
768 | return mp; |
769 | } | |
770 | ||
771 | static void put_mountpoint(struct mountpoint *mp) | |
772 | { | |
773 | if (!--mp->m_count) { | |
774 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 775 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
776 | spin_lock(&dentry->d_lock); |
777 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
778 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 779 | hlist_del(&mp->m_hash); |
84d17192 AV |
780 | kfree(mp); |
781 | } | |
782 | } | |
783 | ||
143c8c91 | 784 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 785 | { |
6b3286ed | 786 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
787 | } |
788 | ||
99b7db7b NP |
789 | /* |
790 | * vfsmount lock must be held for write | |
791 | */ | |
6b3286ed | 792 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
793 | { |
794 | if (ns) { | |
795 | ns->event = ++event; | |
796 | wake_up_interruptible(&ns->poll); | |
797 | } | |
798 | } | |
799 | ||
99b7db7b NP |
800 | /* |
801 | * vfsmount lock must be held for write | |
802 | */ | |
6b3286ed | 803 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
804 | { |
805 | if (ns && ns->event != event) { | |
806 | ns->event = event; | |
807 | wake_up_interruptible(&ns->poll); | |
808 | } | |
809 | } | |
810 | ||
99b7db7b NP |
811 | /* |
812 | * vfsmount lock must be held for write | |
813 | */ | |
7bdb11de | 814 | static void unhash_mnt(struct mount *mnt) |
419148da | 815 | { |
0714a533 | 816 | mnt->mnt_parent = mnt; |
a73324da | 817 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 818 | list_del_init(&mnt->mnt_child); |
38129a13 | 819 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 820 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
821 | put_mountpoint(mnt->mnt_mp); |
822 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
823 | } |
824 | ||
7bdb11de EB |
825 | /* |
826 | * vfsmount lock must be held for write | |
827 | */ | |
828 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
829 | { | |
830 | old_path->dentry = mnt->mnt_mountpoint; | |
831 | old_path->mnt = &mnt->mnt_parent->mnt; | |
832 | unhash_mnt(mnt); | |
833 | } | |
834 | ||
6a46c573 EB |
835 | /* |
836 | * vfsmount lock must be held for write | |
837 | */ | |
838 | static void umount_mnt(struct mount *mnt) | |
839 | { | |
840 | /* old mountpoint will be dropped when we can do that */ | |
841 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
842 | unhash_mnt(mnt); | |
843 | } | |
844 | ||
99b7db7b NP |
845 | /* |
846 | * vfsmount lock must be held for write | |
847 | */ | |
84d17192 AV |
848 | void mnt_set_mountpoint(struct mount *mnt, |
849 | struct mountpoint *mp, | |
44d964d6 | 850 | struct mount *child_mnt) |
b90fa9ae | 851 | { |
84d17192 | 852 | mp->m_count++; |
3a2393d7 | 853 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 854 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 855 | child_mnt->mnt_parent = mnt; |
84d17192 | 856 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 857 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
858 | } |
859 | ||
12766daa EB |
860 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
861 | { | |
862 | hlist_add_head_rcu(&mnt->mnt_hash, | |
863 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
864 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
865 | } | |
866 | ||
99b7db7b NP |
867 | /* |
868 | * vfsmount lock must be held for write | |
869 | */ | |
84d17192 AV |
870 | static void attach_mnt(struct mount *mnt, |
871 | struct mount *parent, | |
872 | struct mountpoint *mp) | |
1da177e4 | 873 | { |
84d17192 | 874 | mnt_set_mountpoint(parent, mp, mnt); |
12766daa | 875 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
876 | } |
877 | ||
12766daa | 878 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 879 | { |
12766daa EB |
880 | struct mountpoint *old_mp = mnt->mnt_mp; |
881 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
882 | struct mount *old_parent = mnt->mnt_parent; | |
883 | ||
884 | list_del_init(&mnt->mnt_child); | |
885 | hlist_del_init(&mnt->mnt_mp_list); | |
886 | hlist_del_init_rcu(&mnt->mnt_hash); | |
887 | ||
888 | attach_mnt(mnt, parent, mp); | |
889 | ||
890 | put_mountpoint(old_mp); | |
891 | ||
892 | /* | |
893 | * Safely avoid even the suggestion this code might sleep or | |
894 | * lock the mount hash by taking advantage of the knowledge that | |
895 | * mnt_change_mountpoint will not release the final reference | |
896 | * to a mountpoint. | |
897 | * | |
898 | * During mounting, the mount passed in as the parent mount will | |
899 | * continue to use the old mountpoint and during unmounting, the | |
900 | * old mountpoint will continue to exist until namespace_unlock, | |
901 | * which happens well after mnt_change_mountpoint. | |
902 | */ | |
903 | spin_lock(&old_mountpoint->d_lock); | |
904 | old_mountpoint->d_lockref.count--; | |
905 | spin_unlock(&old_mountpoint->d_lock); | |
906 | ||
907 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
908 | } |
909 | ||
b90fa9ae | 910 | /* |
99b7db7b | 911 | * vfsmount lock must be held for write |
b90fa9ae | 912 | */ |
12766daa | 913 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 914 | { |
0714a533 | 915 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 916 | struct mount *m; |
b90fa9ae | 917 | LIST_HEAD(head); |
143c8c91 | 918 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 919 | |
0714a533 | 920 | BUG_ON(parent == mnt); |
b90fa9ae | 921 | |
1a4eeaf2 | 922 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 923 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 924 | m->mnt_ns = n; |
f03c6599 | 925 | |
b90fa9ae RP |
926 | list_splice(&head, n->list.prev); |
927 | ||
d2921684 EB |
928 | n->mounts += n->pending_mounts; |
929 | n->pending_mounts = 0; | |
930 | ||
12766daa | 931 | __attach_mnt(mnt, parent); |
6b3286ed | 932 | touch_mnt_namespace(n); |
1da177e4 LT |
933 | } |
934 | ||
909b0a88 | 935 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 936 | { |
6b41d536 AV |
937 | struct list_head *next = p->mnt_mounts.next; |
938 | if (next == &p->mnt_mounts) { | |
1da177e4 | 939 | while (1) { |
909b0a88 | 940 | if (p == root) |
1da177e4 | 941 | return NULL; |
6b41d536 AV |
942 | next = p->mnt_child.next; |
943 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 944 | break; |
0714a533 | 945 | p = p->mnt_parent; |
1da177e4 LT |
946 | } |
947 | } | |
6b41d536 | 948 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
949 | } |
950 | ||
315fc83e | 951 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 952 | { |
6b41d536 AV |
953 | struct list_head *prev = p->mnt_mounts.prev; |
954 | while (prev != &p->mnt_mounts) { | |
955 | p = list_entry(prev, struct mount, mnt_child); | |
956 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
957 | } |
958 | return p; | |
959 | } | |
960 | ||
9d412a43 AV |
961 | struct vfsmount * |
962 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
963 | { | |
b105e270 | 964 | struct mount *mnt; |
9d412a43 AV |
965 | struct dentry *root; |
966 | ||
967 | if (!type) | |
968 | return ERR_PTR(-ENODEV); | |
969 | ||
970 | mnt = alloc_vfsmnt(name); | |
971 | if (!mnt) | |
972 | return ERR_PTR(-ENOMEM); | |
973 | ||
974 | if (flags & MS_KERNMOUNT) | |
b105e270 | 975 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
976 | |
977 | root = mount_fs(type, flags, name, data); | |
978 | if (IS_ERR(root)) { | |
8ffcb32e | 979 | mnt_free_id(mnt); |
9d412a43 AV |
980 | free_vfsmnt(mnt); |
981 | return ERR_CAST(root); | |
982 | } | |
983 | ||
b105e270 AV |
984 | mnt->mnt.mnt_root = root; |
985 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 986 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 987 | mnt->mnt_parent = mnt; |
719ea2fb | 988 | lock_mount_hash(); |
39f7c4db | 989 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 990 | unlock_mount_hash(); |
b105e270 | 991 | return &mnt->mnt; |
9d412a43 AV |
992 | } |
993 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
994 | ||
c8d3e3d5 EB |
995 | struct vfsmount * |
996 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
997 | const char *name, void *data) | |
998 | { | |
999 | /* Until it is worked out how to pass the user namespace | |
1000 | * through from the parent mount to the submount don't support | |
1001 | * unprivileged mounts with submounts. | |
1002 | */ | |
1003 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1004 | return ERR_PTR(-EPERM); | |
1005 | ||
1006 | return vfs_kern_mount(type, MS_SUBMOUNT, name, data); | |
1007 | } | |
1008 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1009 | ||
87129cc0 | 1010 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1011 | int flag) |
1da177e4 | 1012 | { |
87129cc0 | 1013 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1014 | struct mount *mnt; |
1015 | int err; | |
1da177e4 | 1016 | |
be34d1a3 DH |
1017 | mnt = alloc_vfsmnt(old->mnt_devname); |
1018 | if (!mnt) | |
1019 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1020 | |
7a472ef4 | 1021 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1022 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1023 | else | |
1024 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1025 | |
be34d1a3 DH |
1026 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1027 | err = mnt_alloc_group_id(mnt); | |
1028 | if (err) | |
1029 | goto out_free; | |
1da177e4 | 1030 | } |
be34d1a3 | 1031 | |
f2ebb3a9 | 1032 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 1033 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1034 | if (flag & CL_UNPRIVILEGED) { |
1035 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1036 | ||
1037 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1038 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1039 | ||
1040 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1041 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1042 | ||
1043 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1044 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1045 | ||
1046 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1047 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1048 | } | |
132c94e3 | 1049 | |
5ff9d8a6 | 1050 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1051 | if ((flag & CL_UNPRIVILEGED) && |
1052 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1053 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1054 | ||
be34d1a3 DH |
1055 | atomic_inc(&sb->s_active); |
1056 | mnt->mnt.mnt_sb = sb; | |
1057 | mnt->mnt.mnt_root = dget(root); | |
1058 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1059 | mnt->mnt_parent = mnt; | |
719ea2fb | 1060 | lock_mount_hash(); |
be34d1a3 | 1061 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1062 | unlock_mount_hash(); |
be34d1a3 | 1063 | |
7a472ef4 EB |
1064 | if ((flag & CL_SLAVE) || |
1065 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1066 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1067 | mnt->mnt_master = old; | |
1068 | CLEAR_MNT_SHARED(mnt); | |
1069 | } else if (!(flag & CL_PRIVATE)) { | |
1070 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1071 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1072 | if (IS_MNT_SLAVE(old)) | |
1073 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1074 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1075 | } else { |
1076 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1077 | } |
1078 | if (flag & CL_MAKE_SHARED) | |
1079 | set_mnt_shared(mnt); | |
1080 | ||
1081 | /* stick the duplicate mount on the same expiry list | |
1082 | * as the original if that was on one */ | |
1083 | if (flag & CL_EXPIRE) { | |
1084 | if (!list_empty(&old->mnt_expire)) | |
1085 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1086 | } | |
1087 | ||
cb338d06 | 1088 | return mnt; |
719f5d7f MS |
1089 | |
1090 | out_free: | |
8ffcb32e | 1091 | mnt_free_id(mnt); |
719f5d7f | 1092 | free_vfsmnt(mnt); |
be34d1a3 | 1093 | return ERR_PTR(err); |
1da177e4 LT |
1094 | } |
1095 | ||
9ea459e1 AV |
1096 | static void cleanup_mnt(struct mount *mnt) |
1097 | { | |
1098 | /* | |
1099 | * This probably indicates that somebody messed | |
1100 | * up a mnt_want/drop_write() pair. If this | |
1101 | * happens, the filesystem was probably unable | |
1102 | * to make r/w->r/o transitions. | |
1103 | */ | |
1104 | /* | |
1105 | * The locking used to deal with mnt_count decrement provides barriers, | |
1106 | * so mnt_get_writers() below is safe. | |
1107 | */ | |
1108 | WARN_ON(mnt_get_writers(mnt)); | |
1109 | if (unlikely(mnt->mnt_pins.first)) | |
1110 | mnt_pin_kill(mnt); | |
1111 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1112 | dput(mnt->mnt.mnt_root); | |
1113 | deactivate_super(mnt->mnt.mnt_sb); | |
1114 | mnt_free_id(mnt); | |
1115 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1116 | } | |
1117 | ||
1118 | static void __cleanup_mnt(struct rcu_head *head) | |
1119 | { | |
1120 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1121 | } | |
1122 | ||
1123 | static LLIST_HEAD(delayed_mntput_list); | |
1124 | static void delayed_mntput(struct work_struct *unused) | |
1125 | { | |
1126 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1127 | struct llist_node *next; | |
1128 | ||
1129 | for (; node; node = next) { | |
1130 | next = llist_next(node); | |
1131 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1132 | } | |
1133 | } | |
1134 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1135 | ||
900148dc | 1136 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1137 | { |
48a066e7 AV |
1138 | rcu_read_lock(); |
1139 | mnt_add_count(mnt, -1); | |
1140 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1141 | rcu_read_unlock(); | |
f03c6599 | 1142 | return; |
b3e19d92 | 1143 | } |
719ea2fb | 1144 | lock_mount_hash(); |
b3e19d92 | 1145 | if (mnt_get_count(mnt)) { |
48a066e7 | 1146 | rcu_read_unlock(); |
719ea2fb | 1147 | unlock_mount_hash(); |
99b7db7b NP |
1148 | return; |
1149 | } | |
48a066e7 AV |
1150 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1151 | rcu_read_unlock(); | |
1152 | unlock_mount_hash(); | |
1153 | return; | |
1154 | } | |
1155 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1156 | rcu_read_unlock(); | |
962830df | 1157 | |
39f7c4db | 1158 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1159 | |
1160 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1161 | struct mount *p, *tmp; | |
1162 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1163 | umount_mnt(p); | |
1164 | } | |
1165 | } | |
719ea2fb | 1166 | unlock_mount_hash(); |
649a795a | 1167 | |
9ea459e1 AV |
1168 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1169 | struct task_struct *task = current; | |
1170 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1171 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1172 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1173 | return; | |
1174 | } | |
1175 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1176 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1177 | return; | |
1178 | } | |
1179 | cleanup_mnt(mnt); | |
b3e19d92 | 1180 | } |
b3e19d92 NP |
1181 | |
1182 | void mntput(struct vfsmount *mnt) | |
1183 | { | |
1184 | if (mnt) { | |
863d684f | 1185 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1186 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1187 | if (unlikely(m->mnt_expiry_mark)) |
1188 | m->mnt_expiry_mark = 0; | |
1189 | mntput_no_expire(m); | |
b3e19d92 NP |
1190 | } |
1191 | } | |
1192 | EXPORT_SYMBOL(mntput); | |
1193 | ||
1194 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1195 | { | |
1196 | if (mnt) | |
83adc753 | 1197 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1198 | return mnt; |
1199 | } | |
1200 | EXPORT_SYMBOL(mntget); | |
1201 | ||
c6609c0a IK |
1202 | /* path_is_mountpoint() - Check if path is a mount in the current |
1203 | * namespace. | |
1204 | * | |
1205 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1206 | * not mounted in any namespace and that common case is handled inline. | |
1207 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1208 | * mounts using a given dentry in a different namespace. This function | |
1209 | * checks if the passed in path is a mountpoint rather than the dentry | |
1210 | * alone. | |
1211 | */ | |
1212 | bool path_is_mountpoint(const struct path *path) | |
1213 | { | |
1214 | unsigned seq; | |
1215 | bool res; | |
1216 | ||
1217 | if (!d_mountpoint(path->dentry)) | |
1218 | return false; | |
1219 | ||
1220 | rcu_read_lock(); | |
1221 | do { | |
1222 | seq = read_seqbegin(&mount_lock); | |
1223 | res = __path_is_mountpoint(path); | |
1224 | } while (read_seqretry(&mount_lock, seq)); | |
1225 | rcu_read_unlock(); | |
1226 | ||
1227 | return res; | |
1228 | } | |
1229 | EXPORT_SYMBOL(path_is_mountpoint); | |
1230 | ||
ca71cf71 | 1231 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1232 | { |
3064c356 AV |
1233 | struct mount *p; |
1234 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1235 | if (IS_ERR(p)) | |
1236 | return ERR_CAST(p); | |
1237 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1238 | return &p->mnt; | |
7b7b1ace | 1239 | } |
1da177e4 | 1240 | |
b3b304a2 MS |
1241 | static inline void mangle(struct seq_file *m, const char *s) |
1242 | { | |
1243 | seq_escape(m, s, " \t\n\\"); | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * Simple .show_options callback for filesystems which don't want to | |
1248 | * implement more complex mount option showing. | |
1249 | * | |
1250 | * See also save_mount_options(). | |
1251 | */ | |
34c80b1d | 1252 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1253 | { |
2a32cebd AV |
1254 | const char *options; |
1255 | ||
1256 | rcu_read_lock(); | |
34c80b1d | 1257 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1258 | |
1259 | if (options != NULL && options[0]) { | |
1260 | seq_putc(m, ','); | |
1261 | mangle(m, options); | |
1262 | } | |
2a32cebd | 1263 | rcu_read_unlock(); |
b3b304a2 MS |
1264 | |
1265 | return 0; | |
1266 | } | |
1267 | EXPORT_SYMBOL(generic_show_options); | |
1268 | ||
1269 | /* | |
1270 | * If filesystem uses generic_show_options(), this function should be | |
1271 | * called from the fill_super() callback. | |
1272 | * | |
1273 | * The .remount_fs callback usually needs to be handled in a special | |
1274 | * way, to make sure, that previous options are not overwritten if the | |
1275 | * remount fails. | |
1276 | * | |
1277 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1278 | * reset all options to their default value, but changes only newly | |
1279 | * given options, then the displayed options will not reflect reality | |
1280 | * any more. | |
1281 | */ | |
1282 | void save_mount_options(struct super_block *sb, char *options) | |
1283 | { | |
2a32cebd AV |
1284 | BUG_ON(sb->s_options); |
1285 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1286 | } |
1287 | EXPORT_SYMBOL(save_mount_options); | |
1288 | ||
2a32cebd AV |
1289 | void replace_mount_options(struct super_block *sb, char *options) |
1290 | { | |
1291 | char *old = sb->s_options; | |
1292 | rcu_assign_pointer(sb->s_options, options); | |
1293 | if (old) { | |
1294 | synchronize_rcu(); | |
1295 | kfree(old); | |
1296 | } | |
1297 | } | |
1298 | EXPORT_SYMBOL(replace_mount_options); | |
1299 | ||
a1a2c409 | 1300 | #ifdef CONFIG_PROC_FS |
0226f492 | 1301 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1302 | static void *m_start(struct seq_file *m, loff_t *pos) |
1303 | { | |
ede1bf0d | 1304 | struct proc_mounts *p = m->private; |
1da177e4 | 1305 | |
390c6843 | 1306 | down_read(&namespace_sem); |
c7999c36 AV |
1307 | if (p->cached_event == p->ns->event) { |
1308 | void *v = p->cached_mount; | |
1309 | if (*pos == p->cached_index) | |
1310 | return v; | |
1311 | if (*pos == p->cached_index + 1) { | |
1312 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1313 | return p->cached_mount = v; | |
1314 | } | |
1315 | } | |
1316 | ||
1317 | p->cached_event = p->ns->event; | |
1318 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1319 | p->cached_index = *pos; | |
1320 | return p->cached_mount; | |
1da177e4 LT |
1321 | } |
1322 | ||
1323 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1324 | { | |
ede1bf0d | 1325 | struct proc_mounts *p = m->private; |
b0765fb8 | 1326 | |
c7999c36 AV |
1327 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1328 | p->cached_index = *pos; | |
1329 | return p->cached_mount; | |
1da177e4 LT |
1330 | } |
1331 | ||
1332 | static void m_stop(struct seq_file *m, void *v) | |
1333 | { | |
390c6843 | 1334 | up_read(&namespace_sem); |
1da177e4 LT |
1335 | } |
1336 | ||
0226f492 | 1337 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1338 | { |
ede1bf0d | 1339 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1340 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1341 | return p->show(m, &r->mnt); |
1da177e4 LT |
1342 | } |
1343 | ||
a1a2c409 | 1344 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1345 | .start = m_start, |
1346 | .next = m_next, | |
1347 | .stop = m_stop, | |
0226f492 | 1348 | .show = m_show, |
b4629fe2 | 1349 | }; |
a1a2c409 | 1350 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1351 | |
1da177e4 LT |
1352 | /** |
1353 | * may_umount_tree - check if a mount tree is busy | |
1354 | * @mnt: root of mount tree | |
1355 | * | |
1356 | * This is called to check if a tree of mounts has any | |
1357 | * open files, pwds, chroots or sub mounts that are | |
1358 | * busy. | |
1359 | */ | |
909b0a88 | 1360 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1361 | { |
909b0a88 | 1362 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1363 | int actual_refs = 0; |
1364 | int minimum_refs = 0; | |
315fc83e | 1365 | struct mount *p; |
909b0a88 | 1366 | BUG_ON(!m); |
1da177e4 | 1367 | |
b3e19d92 | 1368 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1369 | lock_mount_hash(); |
909b0a88 | 1370 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1371 | actual_refs += mnt_get_count(p); |
1da177e4 | 1372 | minimum_refs += 2; |
1da177e4 | 1373 | } |
719ea2fb | 1374 | unlock_mount_hash(); |
1da177e4 LT |
1375 | |
1376 | if (actual_refs > minimum_refs) | |
e3474a8e | 1377 | return 0; |
1da177e4 | 1378 | |
e3474a8e | 1379 | return 1; |
1da177e4 LT |
1380 | } |
1381 | ||
1382 | EXPORT_SYMBOL(may_umount_tree); | |
1383 | ||
1384 | /** | |
1385 | * may_umount - check if a mount point is busy | |
1386 | * @mnt: root of mount | |
1387 | * | |
1388 | * This is called to check if a mount point has any | |
1389 | * open files, pwds, chroots or sub mounts. If the | |
1390 | * mount has sub mounts this will return busy | |
1391 | * regardless of whether the sub mounts are busy. | |
1392 | * | |
1393 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1394 | * give false negatives. The main reason why it's here is that we need | |
1395 | * a non-destructive way to look for easily umountable filesystems. | |
1396 | */ | |
1397 | int may_umount(struct vfsmount *mnt) | |
1398 | { | |
e3474a8e | 1399 | int ret = 1; |
8ad08d8a | 1400 | down_read(&namespace_sem); |
719ea2fb | 1401 | lock_mount_hash(); |
1ab59738 | 1402 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1403 | ret = 0; |
719ea2fb | 1404 | unlock_mount_hash(); |
8ad08d8a | 1405 | up_read(&namespace_sem); |
a05964f3 | 1406 | return ret; |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | EXPORT_SYMBOL(may_umount); | |
1410 | ||
38129a13 | 1411 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1412 | |
97216be0 | 1413 | static void namespace_unlock(void) |
70fbcdf4 | 1414 | { |
a3b3c562 | 1415 | struct hlist_head head; |
97216be0 | 1416 | |
a3b3c562 | 1417 | hlist_move_list(&unmounted, &head); |
97216be0 | 1418 | |
97216be0 AV |
1419 | up_write(&namespace_sem); |
1420 | ||
a3b3c562 EB |
1421 | if (likely(hlist_empty(&head))) |
1422 | return; | |
1423 | ||
48a066e7 AV |
1424 | synchronize_rcu(); |
1425 | ||
87b95ce0 | 1426 | group_pin_kill(&head); |
70fbcdf4 RP |
1427 | } |
1428 | ||
97216be0 | 1429 | static inline void namespace_lock(void) |
e3197d83 | 1430 | { |
97216be0 | 1431 | down_write(&namespace_sem); |
e3197d83 AV |
1432 | } |
1433 | ||
e819f152 EB |
1434 | enum umount_tree_flags { |
1435 | UMOUNT_SYNC = 1, | |
1436 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1437 | UMOUNT_CONNECTED = 4, |
e819f152 | 1438 | }; |
f2d0a123 EB |
1439 | |
1440 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1441 | { | |
1442 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1443 | if (how & UMOUNT_SYNC) | |
1444 | return true; | |
1445 | ||
1446 | /* A mount without a parent has nothing to be connected to */ | |
1447 | if (!mnt_has_parent(mnt)) | |
1448 | return true; | |
1449 | ||
1450 | /* Because the reference counting rules change when mounts are | |
1451 | * unmounted and connected, umounted mounts may not be | |
1452 | * connected to mounted mounts. | |
1453 | */ | |
1454 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1455 | return true; | |
1456 | ||
1457 | /* Has it been requested that the mount remain connected? */ | |
1458 | if (how & UMOUNT_CONNECTED) | |
1459 | return false; | |
1460 | ||
1461 | /* Is the mount locked such that it needs to remain connected? */ | |
1462 | if (IS_MNT_LOCKED(mnt)) | |
1463 | return false; | |
1464 | ||
1465 | /* By default disconnect the mount */ | |
1466 | return true; | |
1467 | } | |
1468 | ||
99b7db7b | 1469 | /* |
48a066e7 | 1470 | * mount_lock must be held |
99b7db7b NP |
1471 | * namespace_sem must be held for write |
1472 | */ | |
e819f152 | 1473 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1474 | { |
c003b26f | 1475 | LIST_HEAD(tmp_list); |
315fc83e | 1476 | struct mount *p; |
1da177e4 | 1477 | |
5d88457e EB |
1478 | if (how & UMOUNT_PROPAGATE) |
1479 | propagate_mount_unlock(mnt); | |
1480 | ||
c003b26f | 1481 | /* Gather the mounts to umount */ |
590ce4bc EB |
1482 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1483 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1484 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1485 | } |
1da177e4 | 1486 | |
411a938b | 1487 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1488 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1489 | list_del_init(&p->mnt_child); |
c003b26f | 1490 | } |
88b368f2 | 1491 | |
c003b26f | 1492 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1493 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1494 | propagate_umount(&tmp_list); |
a05964f3 | 1495 | |
c003b26f | 1496 | while (!list_empty(&tmp_list)) { |
d2921684 | 1497 | struct mnt_namespace *ns; |
ce07d891 | 1498 | bool disconnect; |
c003b26f | 1499 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1500 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1501 | list_del_init(&p->mnt_list); |
d2921684 EB |
1502 | ns = p->mnt_ns; |
1503 | if (ns) { | |
1504 | ns->mounts--; | |
1505 | __touch_mnt_namespace(ns); | |
1506 | } | |
143c8c91 | 1507 | p->mnt_ns = NULL; |
e819f152 | 1508 | if (how & UMOUNT_SYNC) |
48a066e7 | 1509 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1510 | |
f2d0a123 | 1511 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1512 | |
1513 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1514 | disconnect ? &unmounted : NULL); | |
676da58d | 1515 | if (mnt_has_parent(p)) { |
81b6b061 | 1516 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1517 | if (!disconnect) { |
1518 | /* Don't forget about p */ | |
1519 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1520 | } else { | |
1521 | umount_mnt(p); | |
1522 | } | |
7c4b93d8 | 1523 | } |
0f0afb1d | 1524 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1525 | } |
1526 | } | |
1527 | ||
b54b9be7 | 1528 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1529 | |
1ab59738 | 1530 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1531 | { |
1ab59738 | 1532 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1533 | int retval; |
1534 | ||
1ab59738 | 1535 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1536 | if (retval) |
1537 | return retval; | |
1538 | ||
1539 | /* | |
1540 | * Allow userspace to request a mountpoint be expired rather than | |
1541 | * unmounting unconditionally. Unmount only happens if: | |
1542 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1543 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1544 | */ | |
1545 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1546 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1547 | flags & (MNT_FORCE | MNT_DETACH)) |
1548 | return -EINVAL; | |
1549 | ||
b3e19d92 NP |
1550 | /* |
1551 | * probably don't strictly need the lock here if we examined | |
1552 | * all race cases, but it's a slowpath. | |
1553 | */ | |
719ea2fb | 1554 | lock_mount_hash(); |
83adc753 | 1555 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1556 | unlock_mount_hash(); |
1da177e4 | 1557 | return -EBUSY; |
b3e19d92 | 1558 | } |
719ea2fb | 1559 | unlock_mount_hash(); |
1da177e4 | 1560 | |
863d684f | 1561 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1562 | return -EAGAIN; |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * If we may have to abort operations to get out of this | |
1567 | * mount, and they will themselves hold resources we must | |
1568 | * allow the fs to do things. In the Unix tradition of | |
1569 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1570 | * might fail to complete on the first run through as other tasks | |
1571 | * must return, and the like. Thats for the mount program to worry | |
1572 | * about for the moment. | |
1573 | */ | |
1574 | ||
42faad99 | 1575 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1576 | sb->s_op->umount_begin(sb); |
42faad99 | 1577 | } |
1da177e4 LT |
1578 | |
1579 | /* | |
1580 | * No sense to grab the lock for this test, but test itself looks | |
1581 | * somewhat bogus. Suggestions for better replacement? | |
1582 | * Ho-hum... In principle, we might treat that as umount + switch | |
1583 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1584 | * Actually it makes sense, especially if rootfs would contain a | |
1585 | * /reboot - static binary that would close all descriptors and | |
1586 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1587 | */ | |
1ab59738 | 1588 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1589 | /* |
1590 | * Special case for "unmounting" root ... | |
1591 | * we just try to remount it readonly. | |
1592 | */ | |
f17a3091 | 1593 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1594 | return -EPERM; |
1da177e4 | 1595 | down_write(&sb->s_umount); |
4aa98cf7 | 1596 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1597 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1598 | up_write(&sb->s_umount); |
1599 | return retval; | |
1600 | } | |
1601 | ||
97216be0 | 1602 | namespace_lock(); |
719ea2fb | 1603 | lock_mount_hash(); |
5addc5dd | 1604 | event++; |
1da177e4 | 1605 | |
48a066e7 | 1606 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1607 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1608 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1609 | retval = 0; |
48a066e7 AV |
1610 | } else { |
1611 | shrink_submounts(mnt); | |
1612 | retval = -EBUSY; | |
1613 | if (!propagate_mount_busy(mnt, 2)) { | |
1614 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1615 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1616 | retval = 0; |
1617 | } | |
1da177e4 | 1618 | } |
719ea2fb | 1619 | unlock_mount_hash(); |
e3197d83 | 1620 | namespace_unlock(); |
1da177e4 LT |
1621 | return retval; |
1622 | } | |
1623 | ||
80b5dce8 EB |
1624 | /* |
1625 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1626 | * | |
1627 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1628 | * to an existing mountpoint, and wind up leaking the mount. | |
1629 | * detach_mounts allows lazily unmounting those mounts instead of | |
1630 | * leaking them. | |
1631 | * | |
1632 | * The caller may hold dentry->d_inode->i_mutex. | |
1633 | */ | |
1634 | void __detach_mounts(struct dentry *dentry) | |
1635 | { | |
1636 | struct mountpoint *mp; | |
1637 | struct mount *mnt; | |
1638 | ||
1639 | namespace_lock(); | |
3895dbf8 | 1640 | lock_mount_hash(); |
80b5dce8 | 1641 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1642 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1643 | goto out_unlock; |
1644 | ||
e06b933e | 1645 | event++; |
80b5dce8 EB |
1646 | while (!hlist_empty(&mp->m_list)) { |
1647 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1648 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1649 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1650 | umount_mnt(mnt); | |
ce07d891 | 1651 | } |
e0c9c0af | 1652 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1653 | } |
80b5dce8 EB |
1654 | put_mountpoint(mp); |
1655 | out_unlock: | |
3895dbf8 | 1656 | unlock_mount_hash(); |
80b5dce8 EB |
1657 | namespace_unlock(); |
1658 | } | |
1659 | ||
9b40bc90 AV |
1660 | /* |
1661 | * Is the caller allowed to modify his namespace? | |
1662 | */ | |
1663 | static inline bool may_mount(void) | |
1664 | { | |
1665 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1666 | } | |
1667 | ||
9e8925b6 JL |
1668 | static inline bool may_mandlock(void) |
1669 | { | |
1670 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1671 | return false; | |
1672 | #endif | |
95ace754 | 1673 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1674 | } |
1675 | ||
1da177e4 LT |
1676 | /* |
1677 | * Now umount can handle mount points as well as block devices. | |
1678 | * This is important for filesystems which use unnamed block devices. | |
1679 | * | |
1680 | * We now support a flag for forced unmount like the other 'big iron' | |
1681 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1682 | */ | |
1683 | ||
bdc480e3 | 1684 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1685 | { |
2d8f3038 | 1686 | struct path path; |
900148dc | 1687 | struct mount *mnt; |
1da177e4 | 1688 | int retval; |
db1f05bb | 1689 | int lookup_flags = 0; |
1da177e4 | 1690 | |
db1f05bb MS |
1691 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1692 | return -EINVAL; | |
1693 | ||
9b40bc90 AV |
1694 | if (!may_mount()) |
1695 | return -EPERM; | |
1696 | ||
db1f05bb MS |
1697 | if (!(flags & UMOUNT_NOFOLLOW)) |
1698 | lookup_flags |= LOOKUP_FOLLOW; | |
1699 | ||
197df04c | 1700 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1701 | if (retval) |
1702 | goto out; | |
900148dc | 1703 | mnt = real_mount(path.mnt); |
1da177e4 | 1704 | retval = -EINVAL; |
2d8f3038 | 1705 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1706 | goto dput_and_out; |
143c8c91 | 1707 | if (!check_mnt(mnt)) |
1da177e4 | 1708 | goto dput_and_out; |
5ff9d8a6 EB |
1709 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1710 | goto dput_and_out; | |
b2f5d4dc EB |
1711 | retval = -EPERM; |
1712 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1713 | goto dput_and_out; | |
1da177e4 | 1714 | |
900148dc | 1715 | retval = do_umount(mnt, flags); |
1da177e4 | 1716 | dput_and_out: |
429731b1 | 1717 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1718 | dput(path.dentry); |
900148dc | 1719 | mntput_no_expire(mnt); |
1da177e4 LT |
1720 | out: |
1721 | return retval; | |
1722 | } | |
1723 | ||
1724 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1725 | ||
1726 | /* | |
b58fed8b | 1727 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1728 | */ |
bdc480e3 | 1729 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1730 | { |
b58fed8b | 1731 | return sys_umount(name, 0); |
1da177e4 LT |
1732 | } |
1733 | ||
1734 | #endif | |
1735 | ||
4ce5d2b1 | 1736 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1737 | { |
4ce5d2b1 | 1738 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1739 | return dentry->d_op == &ns_dentry_operations && |
1740 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1741 | } |
1742 | ||
58be2825 AV |
1743 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1744 | { | |
1745 | return container_of(ns, struct mnt_namespace, ns); | |
1746 | } | |
1747 | ||
4ce5d2b1 EB |
1748 | static bool mnt_ns_loop(struct dentry *dentry) |
1749 | { | |
1750 | /* Could bind mounting the mount namespace inode cause a | |
1751 | * mount namespace loop? | |
1752 | */ | |
1753 | struct mnt_namespace *mnt_ns; | |
1754 | if (!is_mnt_ns_file(dentry)) | |
1755 | return false; | |
1756 | ||
f77c8014 | 1757 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1758 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1759 | } | |
1760 | ||
87129cc0 | 1761 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1762 | int flag) |
1da177e4 | 1763 | { |
84d17192 | 1764 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1765 | |
4ce5d2b1 EB |
1766 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1767 | return ERR_PTR(-EINVAL); | |
1768 | ||
1769 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1770 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1771 | |
36341f64 | 1772 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1773 | if (IS_ERR(q)) |
1774 | return q; | |
1775 | ||
a73324da | 1776 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1777 | |
1778 | p = mnt; | |
6b41d536 | 1779 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1780 | struct mount *s; |
7ec02ef1 | 1781 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1782 | continue; |
1783 | ||
909b0a88 | 1784 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1785 | if (!(flag & CL_COPY_UNBINDABLE) && |
1786 | IS_MNT_UNBINDABLE(s)) { | |
1787 | s = skip_mnt_tree(s); | |
1788 | continue; | |
1789 | } | |
1790 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1791 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1792 | s = skip_mnt_tree(s); |
1793 | continue; | |
1794 | } | |
0714a533 AV |
1795 | while (p != s->mnt_parent) { |
1796 | p = p->mnt_parent; | |
1797 | q = q->mnt_parent; | |
1da177e4 | 1798 | } |
87129cc0 | 1799 | p = s; |
84d17192 | 1800 | parent = q; |
87129cc0 | 1801 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1802 | if (IS_ERR(q)) |
1803 | goto out; | |
719ea2fb | 1804 | lock_mount_hash(); |
1a4eeaf2 | 1805 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12766daa | 1806 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1807 | unlock_mount_hash(); |
1da177e4 LT |
1808 | } |
1809 | } | |
1810 | return res; | |
be34d1a3 | 1811 | out: |
1da177e4 | 1812 | if (res) { |
719ea2fb | 1813 | lock_mount_hash(); |
e819f152 | 1814 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1815 | unlock_mount_hash(); |
1da177e4 | 1816 | } |
be34d1a3 | 1817 | return q; |
1da177e4 LT |
1818 | } |
1819 | ||
be34d1a3 DH |
1820 | /* Caller should check returned pointer for errors */ |
1821 | ||
ca71cf71 | 1822 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1823 | { |
cb338d06 | 1824 | struct mount *tree; |
97216be0 | 1825 | namespace_lock(); |
cd4a4017 EB |
1826 | if (!check_mnt(real_mount(path->mnt))) |
1827 | tree = ERR_PTR(-EINVAL); | |
1828 | else | |
1829 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1830 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1831 | namespace_unlock(); |
be34d1a3 | 1832 | if (IS_ERR(tree)) |
52e220d3 | 1833 | return ERR_CAST(tree); |
be34d1a3 | 1834 | return &tree->mnt; |
8aec0809 AV |
1835 | } |
1836 | ||
1837 | void drop_collected_mounts(struct vfsmount *mnt) | |
1838 | { | |
97216be0 | 1839 | namespace_lock(); |
719ea2fb | 1840 | lock_mount_hash(); |
e819f152 | 1841 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1842 | unlock_mount_hash(); |
3ab6abee | 1843 | namespace_unlock(); |
8aec0809 AV |
1844 | } |
1845 | ||
c771d683 MS |
1846 | /** |
1847 | * clone_private_mount - create a private clone of a path | |
1848 | * | |
1849 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1850 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1851 | * to the originating mount won't be propagated into this). | |
1852 | * | |
1853 | * Release with mntput(). | |
1854 | */ | |
ca71cf71 | 1855 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1856 | { |
1857 | struct mount *old_mnt = real_mount(path->mnt); | |
1858 | struct mount *new_mnt; | |
1859 | ||
1860 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1861 | return ERR_PTR(-EINVAL); | |
1862 | ||
c771d683 | 1863 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1864 | if (IS_ERR(new_mnt)) |
1865 | return ERR_CAST(new_mnt); | |
1866 | ||
1867 | return &new_mnt->mnt; | |
1868 | } | |
1869 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1870 | ||
1f707137 AV |
1871 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1872 | struct vfsmount *root) | |
1873 | { | |
1a4eeaf2 | 1874 | struct mount *mnt; |
1f707137 AV |
1875 | int res = f(root, arg); |
1876 | if (res) | |
1877 | return res; | |
1a4eeaf2 AV |
1878 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1879 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1880 | if (res) |
1881 | return res; | |
1882 | } | |
1883 | return 0; | |
1884 | } | |
e14748e8 | 1885 | EXPORT_SYMBOL_GPL(iterate_mounts); |
1f707137 | 1886 | |
4b8b21f4 | 1887 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1888 | { |
315fc83e | 1889 | struct mount *p; |
719f5d7f | 1890 | |
909b0a88 | 1891 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1892 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1893 | mnt_release_group_id(p); |
719f5d7f MS |
1894 | } |
1895 | } | |
1896 | ||
4b8b21f4 | 1897 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1898 | { |
315fc83e | 1899 | struct mount *p; |
719f5d7f | 1900 | |
909b0a88 | 1901 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1902 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1903 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1904 | if (err) { |
4b8b21f4 | 1905 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1906 | return err; |
1907 | } | |
1908 | } | |
1909 | } | |
1910 | ||
1911 | return 0; | |
1912 | } | |
1913 | ||
d2921684 EB |
1914 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1915 | { | |
1916 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1917 | unsigned int mounts = 0, old, pending, sum; | |
1918 | struct mount *p; | |
1919 | ||
1920 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1921 | mounts++; | |
1922 | ||
1923 | old = ns->mounts; | |
1924 | pending = ns->pending_mounts; | |
1925 | sum = old + pending; | |
1926 | if ((old > sum) || | |
1927 | (pending > sum) || | |
1928 | (max < sum) || | |
1929 | (mounts > (max - sum))) | |
1930 | return -ENOSPC; | |
1931 | ||
1932 | ns->pending_mounts = pending + mounts; | |
1933 | return 0; | |
1934 | } | |
1935 | ||
b90fa9ae RP |
1936 | /* |
1937 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1938 | * @nd : place the mount tree @source_mnt is attached |
1939 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1940 | * store the parent mount and mountpoint dentry. | |
1941 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1942 | * |
1943 | * NOTE: in the table below explains the semantics when a source mount | |
1944 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1945 | * --------------------------------------------------------------------------- |
1946 | * | BIND MOUNT OPERATION | | |
1947 | * |************************************************************************** | |
1948 | * | source-->| shared | private | slave | unbindable | | |
1949 | * | dest | | | | | | |
1950 | * | | | | | | | | |
1951 | * | v | | | | | | |
1952 | * |************************************************************************** | |
1953 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1954 | * | | | | | | | |
1955 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1956 | * *************************************************************************** | |
b90fa9ae RP |
1957 | * A bind operation clones the source mount and mounts the clone on the |
1958 | * destination mount. | |
1959 | * | |
1960 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1961 | * tree of the destination mount and the cloned mount is added to | |
1962 | * the peer group of the source mount. | |
1963 | * (+) the cloned mount is created under the destination mount and is marked | |
1964 | * as shared. The cloned mount is added to the peer group of the source | |
1965 | * mount. | |
5afe0022 RP |
1966 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1967 | * of the destination mount and the cloned mount is made slave | |
1968 | * of the same master as that of the source mount. The cloned mount | |
1969 | * is marked as 'shared and slave'. | |
1970 | * (*) the cloned mount is made a slave of the same master as that of the | |
1971 | * source mount. | |
1972 | * | |
9676f0c6 RP |
1973 | * --------------------------------------------------------------------------- |
1974 | * | MOVE MOUNT OPERATION | | |
1975 | * |************************************************************************** | |
1976 | * | source-->| shared | private | slave | unbindable | | |
1977 | * | dest | | | | | | |
1978 | * | | | | | | | | |
1979 | * | v | | | | | | |
1980 | * |************************************************************************** | |
1981 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1982 | * | | | | | | | |
1983 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1984 | * *************************************************************************** | |
5afe0022 RP |
1985 | * |
1986 | * (+) the mount is moved to the destination. And is then propagated to | |
1987 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1988 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1989 | * (+++) the mount is moved to the destination and is then propagated to |
1990 | * all the mounts belonging to the destination mount's propagation tree. | |
1991 | * the mount is marked as 'shared and slave'. | |
1992 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1993 | * |
1994 | * if the source mount is a tree, the operations explained above is | |
1995 | * applied to each mount in the tree. | |
1996 | * Must be called without spinlocks held, since this function can sleep | |
1997 | * in allocations. | |
1998 | */ | |
0fb54e50 | 1999 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2000 | struct mount *dest_mnt, |
2001 | struct mountpoint *dest_mp, | |
2002 | struct path *parent_path) | |
b90fa9ae | 2003 | { |
38129a13 | 2004 | HLIST_HEAD(tree_list); |
d2921684 | 2005 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
12766daa | 2006 | struct mountpoint *smp; |
315fc83e | 2007 | struct mount *child, *p; |
38129a13 | 2008 | struct hlist_node *n; |
719f5d7f | 2009 | int err; |
b90fa9ae | 2010 | |
12766daa EB |
2011 | /* Preallocate a mountpoint in case the new mounts need |
2012 | * to be tucked under other mounts. | |
2013 | */ | |
2014 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2015 | if (IS_ERR(smp)) | |
2016 | return PTR_ERR(smp); | |
2017 | ||
d2921684 EB |
2018 | /* Is there space to add these mounts to the mount namespace? */ |
2019 | if (!parent_path) { | |
2020 | err = count_mounts(ns, source_mnt); | |
2021 | if (err) | |
2022 | goto out; | |
2023 | } | |
2024 | ||
fc7be130 | 2025 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2026 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2027 | if (err) |
2028 | goto out; | |
0b1b901b | 2029 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2030 | lock_mount_hash(); |
0b1b901b AV |
2031 | if (err) |
2032 | goto out_cleanup_ids; | |
909b0a88 | 2033 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2034 | set_mnt_shared(p); |
0b1b901b AV |
2035 | } else { |
2036 | lock_mount_hash(); | |
b90fa9ae | 2037 | } |
1a390689 | 2038 | if (parent_path) { |
0fb54e50 | 2039 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2040 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2041 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2042 | } else { |
84d17192 | 2043 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
12766daa | 2044 | commit_tree(source_mnt); |
21444403 | 2045 | } |
b90fa9ae | 2046 | |
38129a13 | 2047 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2048 | struct mount *q; |
38129a13 | 2049 | hlist_del_init(&child->mnt_hash); |
12766daa EB |
2050 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2051 | child->mnt_mountpoint); | |
2052 | if (q) | |
2053 | mnt_change_mountpoint(child, smp, q); | |
2054 | commit_tree(child); | |
b90fa9ae | 2055 | } |
12766daa | 2056 | put_mountpoint(smp); |
719ea2fb | 2057 | unlock_mount_hash(); |
99b7db7b | 2058 | |
b90fa9ae | 2059 | return 0; |
719f5d7f MS |
2060 | |
2061 | out_cleanup_ids: | |
f2ebb3a9 AV |
2062 | while (!hlist_empty(&tree_list)) { |
2063 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2064 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2065 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2066 | } |
2067 | unlock_mount_hash(); | |
0b1b901b | 2068 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2069 | out: |
d2921684 | 2070 | ns->pending_mounts = 0; |
12766daa EB |
2071 | |
2072 | read_seqlock_excl(&mount_lock); | |
2073 | put_mountpoint(smp); | |
2074 | read_sequnlock_excl(&mount_lock); | |
2075 | ||
719f5d7f | 2076 | return err; |
b90fa9ae RP |
2077 | } |
2078 | ||
84d17192 | 2079 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2080 | { |
2081 | struct vfsmount *mnt; | |
84d17192 | 2082 | struct dentry *dentry = path->dentry; |
b12cea91 | 2083 | retry: |
5955102c | 2084 | inode_lock(dentry->d_inode); |
84d17192 | 2085 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2086 | inode_unlock(dentry->d_inode); |
84d17192 | 2087 | return ERR_PTR(-ENOENT); |
b12cea91 | 2088 | } |
97216be0 | 2089 | namespace_lock(); |
b12cea91 | 2090 | mnt = lookup_mnt(path); |
84d17192 | 2091 | if (likely(!mnt)) { |
3895dbf8 | 2092 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2093 | if (IS_ERR(mp)) { |
97216be0 | 2094 | namespace_unlock(); |
5955102c | 2095 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2096 | return mp; |
2097 | } | |
2098 | return mp; | |
2099 | } | |
97216be0 | 2100 | namespace_unlock(); |
5955102c | 2101 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2102 | path_put(path); |
2103 | path->mnt = mnt; | |
84d17192 | 2104 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2105 | goto retry; |
2106 | } | |
2107 | ||
84d17192 | 2108 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2109 | { |
84d17192 | 2110 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2111 | |
2112 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2113 | put_mountpoint(where); |
3895dbf8 EB |
2114 | read_sequnlock_excl(&mount_lock); |
2115 | ||
328e6d90 | 2116 | namespace_unlock(); |
5955102c | 2117 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2118 | } |
2119 | ||
84d17192 | 2120 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2121 | { |
95bc5f25 | 2122 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
2123 | return -EINVAL; |
2124 | ||
e36cb0b8 DH |
2125 | if (d_is_dir(mp->m_dentry) != |
2126 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2127 | return -ENOTDIR; |
2128 | ||
84d17192 | 2129 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2130 | } |
2131 | ||
7a2e8a8f VA |
2132 | /* |
2133 | * Sanity check the flags to change_mnt_propagation. | |
2134 | */ | |
2135 | ||
2136 | static int flags_to_propagation_type(int flags) | |
2137 | { | |
7c6e984d | 2138 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2139 | |
2140 | /* Fail if any non-propagation flags are set */ | |
2141 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2142 | return 0; | |
2143 | /* Only one propagation flag should be set */ | |
2144 | if (!is_power_of_2(type)) | |
2145 | return 0; | |
2146 | return type; | |
2147 | } | |
2148 | ||
07b20889 RP |
2149 | /* |
2150 | * recursively change the type of the mountpoint. | |
2151 | */ | |
0a0d8a46 | 2152 | static int do_change_type(struct path *path, int flag) |
07b20889 | 2153 | { |
315fc83e | 2154 | struct mount *m; |
4b8b21f4 | 2155 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 2156 | int recurse = flag & MS_REC; |
7a2e8a8f | 2157 | int type; |
719f5d7f | 2158 | int err = 0; |
07b20889 | 2159 | |
2d92ab3c | 2160 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2161 | return -EINVAL; |
2162 | ||
7a2e8a8f VA |
2163 | type = flags_to_propagation_type(flag); |
2164 | if (!type) | |
2165 | return -EINVAL; | |
2166 | ||
97216be0 | 2167 | namespace_lock(); |
719f5d7f MS |
2168 | if (type == MS_SHARED) { |
2169 | err = invent_group_ids(mnt, recurse); | |
2170 | if (err) | |
2171 | goto out_unlock; | |
2172 | } | |
2173 | ||
719ea2fb | 2174 | lock_mount_hash(); |
909b0a88 | 2175 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2176 | change_mnt_propagation(m, type); |
719ea2fb | 2177 | unlock_mount_hash(); |
719f5d7f MS |
2178 | |
2179 | out_unlock: | |
97216be0 | 2180 | namespace_unlock(); |
719f5d7f | 2181 | return err; |
07b20889 RP |
2182 | } |
2183 | ||
5ff9d8a6 EB |
2184 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2185 | { | |
2186 | struct mount *child; | |
2187 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2188 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2189 | continue; | |
2190 | ||
2191 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2192 | return true; | |
2193 | } | |
2194 | return false; | |
2195 | } | |
2196 | ||
1da177e4 LT |
2197 | /* |
2198 | * do loopback mount. | |
2199 | */ | |
808d4e3c | 2200 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2201 | int recurse) |
1da177e4 | 2202 | { |
2d92ab3c | 2203 | struct path old_path; |
84d17192 AV |
2204 | struct mount *mnt = NULL, *old, *parent; |
2205 | struct mountpoint *mp; | |
57eccb83 | 2206 | int err; |
1da177e4 LT |
2207 | if (!old_name || !*old_name) |
2208 | return -EINVAL; | |
815d405c | 2209 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2210 | if (err) |
2211 | return err; | |
2212 | ||
8823c079 | 2213 | err = -EINVAL; |
4ce5d2b1 | 2214 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2215 | goto out; |
2216 | ||
84d17192 AV |
2217 | mp = lock_mount(path); |
2218 | err = PTR_ERR(mp); | |
2219 | if (IS_ERR(mp)) | |
b12cea91 AV |
2220 | goto out; |
2221 | ||
87129cc0 | 2222 | old = real_mount(old_path.mnt); |
84d17192 | 2223 | parent = real_mount(path->mnt); |
87129cc0 | 2224 | |
1da177e4 | 2225 | err = -EINVAL; |
fc7be130 | 2226 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2227 | goto out2; |
9676f0c6 | 2228 | |
e149ed2b AV |
2229 | if (!check_mnt(parent)) |
2230 | goto out2; | |
2231 | ||
2232 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2233 | goto out2; |
1da177e4 | 2234 | |
5ff9d8a6 EB |
2235 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2236 | goto out2; | |
2237 | ||
ccd48bc7 | 2238 | if (recurse) |
4ce5d2b1 | 2239 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2240 | else |
87129cc0 | 2241 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2242 | |
be34d1a3 DH |
2243 | if (IS_ERR(mnt)) { |
2244 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2245 | goto out2; |
be34d1a3 | 2246 | } |
ccd48bc7 | 2247 | |
5ff9d8a6 EB |
2248 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2249 | ||
84d17192 | 2250 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2251 | if (err) { |
719ea2fb | 2252 | lock_mount_hash(); |
e819f152 | 2253 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2254 | unlock_mount_hash(); |
5b83d2c5 | 2255 | } |
b12cea91 | 2256 | out2: |
84d17192 | 2257 | unlock_mount(mp); |
ccd48bc7 | 2258 | out: |
2d92ab3c | 2259 | path_put(&old_path); |
1da177e4 LT |
2260 | return err; |
2261 | } | |
2262 | ||
2e4b7fcd DH |
2263 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2264 | { | |
2265 | int error = 0; | |
2266 | int readonly_request = 0; | |
2267 | ||
2268 | if (ms_flags & MS_RDONLY) | |
2269 | readonly_request = 1; | |
2270 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2271 | return 0; | |
2272 | ||
2273 | if (readonly_request) | |
83adc753 | 2274 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2275 | else |
83adc753 | 2276 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2277 | return error; |
2278 | } | |
2279 | ||
1da177e4 LT |
2280 | /* |
2281 | * change filesystem flags. dir should be a physical root of filesystem. | |
2282 | * If you've mounted a non-root directory somewhere and want to do remount | |
2283 | * on it - tough luck. | |
2284 | */ | |
0a0d8a46 | 2285 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2286 | void *data) |
2287 | { | |
2288 | int err; | |
2d92ab3c | 2289 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2290 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2291 | |
143c8c91 | 2292 | if (!check_mnt(mnt)) |
1da177e4 LT |
2293 | return -EINVAL; |
2294 | ||
2d92ab3c | 2295 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2296 | return -EINVAL; |
2297 | ||
07b64558 EB |
2298 | /* Don't allow changing of locked mnt flags. |
2299 | * | |
2300 | * No locks need to be held here while testing the various | |
2301 | * MNT_LOCK flags because those flags can never be cleared | |
2302 | * once they are set. | |
2303 | */ | |
2304 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2305 | !(mnt_flags & MNT_READONLY)) { | |
2306 | return -EPERM; | |
2307 | } | |
9566d674 EB |
2308 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2309 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2310 | return -EPERM; |
9566d674 EB |
2311 | } |
2312 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2313 | !(mnt_flags & MNT_NOSUID)) { | |
2314 | return -EPERM; | |
2315 | } | |
2316 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2317 | !(mnt_flags & MNT_NOEXEC)) { | |
2318 | return -EPERM; | |
2319 | } | |
2320 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2321 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2322 | return -EPERM; | |
2323 | } | |
2324 | ||
ff36fe2c EP |
2325 | err = security_sb_remount(sb, data); |
2326 | if (err) | |
2327 | return err; | |
2328 | ||
1da177e4 | 2329 | down_write(&sb->s_umount); |
2e4b7fcd | 2330 | if (flags & MS_BIND) |
2d92ab3c | 2331 | err = change_mount_flags(path->mnt, flags); |
f17a3091 | 2332 | else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
57eccb83 | 2333 | err = -EPERM; |
4aa98cf7 | 2334 | else |
2e4b7fcd | 2335 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2336 | if (!err) { |
719ea2fb | 2337 | lock_mount_hash(); |
a6138db8 | 2338 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2339 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2340 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2341 | unlock_mount_hash(); |
0e55a7cc | 2342 | } |
6339dab8 | 2343 | up_write(&sb->s_umount); |
1da177e4 LT |
2344 | return err; |
2345 | } | |
2346 | ||
cbbe362c | 2347 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2348 | { |
315fc83e | 2349 | struct mount *p; |
909b0a88 | 2350 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2351 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2352 | return 1; |
2353 | } | |
2354 | return 0; | |
2355 | } | |
2356 | ||
808d4e3c | 2357 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2358 | { |
2d92ab3c | 2359 | struct path old_path, parent_path; |
676da58d | 2360 | struct mount *p; |
0fb54e50 | 2361 | struct mount *old; |
84d17192 | 2362 | struct mountpoint *mp; |
57eccb83 | 2363 | int err; |
1da177e4 LT |
2364 | if (!old_name || !*old_name) |
2365 | return -EINVAL; | |
2d92ab3c | 2366 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2367 | if (err) |
2368 | return err; | |
2369 | ||
84d17192 AV |
2370 | mp = lock_mount(path); |
2371 | err = PTR_ERR(mp); | |
2372 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2373 | goto out; |
2374 | ||
143c8c91 | 2375 | old = real_mount(old_path.mnt); |
fc7be130 | 2376 | p = real_mount(path->mnt); |
143c8c91 | 2377 | |
1da177e4 | 2378 | err = -EINVAL; |
fc7be130 | 2379 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2380 | goto out1; |
2381 | ||
5ff9d8a6 EB |
2382 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2383 | goto out1; | |
2384 | ||
1da177e4 | 2385 | err = -EINVAL; |
2d92ab3c | 2386 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2387 | goto out1; |
1da177e4 | 2388 | |
676da58d | 2389 | if (!mnt_has_parent(old)) |
21444403 | 2390 | goto out1; |
1da177e4 | 2391 | |
e36cb0b8 DH |
2392 | if (d_is_dir(path->dentry) != |
2393 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2394 | goto out1; |
2395 | /* | |
2396 | * Don't move a mount residing in a shared parent. | |
2397 | */ | |
fc7be130 | 2398 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2399 | goto out1; |
9676f0c6 RP |
2400 | /* |
2401 | * Don't move a mount tree containing unbindable mounts to a destination | |
2402 | * mount which is shared. | |
2403 | */ | |
fc7be130 | 2404 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2405 | goto out1; |
1da177e4 | 2406 | err = -ELOOP; |
fc7be130 | 2407 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2408 | if (p == old) |
21444403 | 2409 | goto out1; |
1da177e4 | 2410 | |
84d17192 | 2411 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2412 | if (err) |
21444403 | 2413 | goto out1; |
1da177e4 LT |
2414 | |
2415 | /* if the mount is moved, it should no longer be expire | |
2416 | * automatically */ | |
6776db3d | 2417 | list_del_init(&old->mnt_expire); |
1da177e4 | 2418 | out1: |
84d17192 | 2419 | unlock_mount(mp); |
1da177e4 | 2420 | out: |
1da177e4 | 2421 | if (!err) |
1a390689 | 2422 | path_put(&parent_path); |
2d92ab3c | 2423 | path_put(&old_path); |
1da177e4 LT |
2424 | return err; |
2425 | } | |
2426 | ||
9d412a43 AV |
2427 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2428 | { | |
2429 | int err; | |
2430 | const char *subtype = strchr(fstype, '.'); | |
2431 | if (subtype) { | |
2432 | subtype++; | |
2433 | err = -EINVAL; | |
2434 | if (!subtype[0]) | |
2435 | goto err; | |
2436 | } else | |
2437 | subtype = ""; | |
2438 | ||
2439 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2440 | err = -ENOMEM; | |
2441 | if (!mnt->mnt_sb->s_subtype) | |
2442 | goto err; | |
2443 | return mnt; | |
2444 | ||
2445 | err: | |
2446 | mntput(mnt); | |
2447 | return ERR_PTR(err); | |
2448 | } | |
2449 | ||
9d412a43 AV |
2450 | /* |
2451 | * add a mount into a namespace's mount tree | |
2452 | */ | |
95bc5f25 | 2453 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2454 | { |
84d17192 AV |
2455 | struct mountpoint *mp; |
2456 | struct mount *parent; | |
9d412a43 AV |
2457 | int err; |
2458 | ||
f2ebb3a9 | 2459 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2460 | |
84d17192 AV |
2461 | mp = lock_mount(path); |
2462 | if (IS_ERR(mp)) | |
2463 | return PTR_ERR(mp); | |
9d412a43 | 2464 | |
84d17192 | 2465 | parent = real_mount(path->mnt); |
9d412a43 | 2466 | err = -EINVAL; |
84d17192 | 2467 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2468 | /* that's acceptable only for automounts done in private ns */ |
2469 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2470 | goto unlock; | |
2471 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2472 | if (!parent->mnt_ns) |
156cacb1 AV |
2473 | goto unlock; |
2474 | } | |
9d412a43 AV |
2475 | |
2476 | /* Refuse the same filesystem on the same mount point */ | |
2477 | err = -EBUSY; | |
95bc5f25 | 2478 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2479 | path->mnt->mnt_root == path->dentry) |
2480 | goto unlock; | |
2481 | ||
2482 | err = -EINVAL; | |
e36cb0b8 | 2483 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2484 | goto unlock; |
2485 | ||
95bc5f25 | 2486 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2487 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2488 | |
2489 | unlock: | |
84d17192 | 2490 | unlock_mount(mp); |
9d412a43 AV |
2491 | return err; |
2492 | } | |
b1e75df4 | 2493 | |
8654df4e | 2494 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2495 | |
1da177e4 LT |
2496 | /* |
2497 | * create a new mount for userspace and request it to be added into the | |
2498 | * namespace's tree | |
2499 | */ | |
0c55cfc4 | 2500 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2501 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2502 | { |
0c55cfc4 | 2503 | struct file_system_type *type; |
1da177e4 | 2504 | struct vfsmount *mnt; |
15f9a3f3 | 2505 | int err; |
1da177e4 | 2506 | |
0c55cfc4 | 2507 | if (!fstype) |
1da177e4 LT |
2508 | return -EINVAL; |
2509 | ||
0c55cfc4 EB |
2510 | type = get_fs_type(fstype); |
2511 | if (!type) | |
2512 | return -ENODEV; | |
2513 | ||
0c55cfc4 EB |
2514 | mnt = vfs_kern_mount(type, flags, name, data); |
2515 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2516 | !mnt->mnt_sb->s_subtype) | |
2517 | mnt = fs_set_subtype(mnt, fstype); | |
2518 | ||
2519 | put_filesystem(type); | |
1da177e4 LT |
2520 | if (IS_ERR(mnt)) |
2521 | return PTR_ERR(mnt); | |
2522 | ||
8654df4e EB |
2523 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2524 | mntput(mnt); | |
2525 | return -EPERM; | |
2526 | } | |
2527 | ||
95bc5f25 | 2528 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2529 | if (err) |
2530 | mntput(mnt); | |
2531 | return err; | |
1da177e4 LT |
2532 | } |
2533 | ||
19a167af AV |
2534 | int finish_automount(struct vfsmount *m, struct path *path) |
2535 | { | |
6776db3d | 2536 | struct mount *mnt = real_mount(m); |
19a167af AV |
2537 | int err; |
2538 | /* The new mount record should have at least 2 refs to prevent it being | |
2539 | * expired before we get a chance to add it | |
2540 | */ | |
6776db3d | 2541 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2542 | |
2543 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2544 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2545 | err = -ELOOP; |
2546 | goto fail; | |
19a167af AV |
2547 | } |
2548 | ||
95bc5f25 | 2549 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2550 | if (!err) |
2551 | return 0; | |
2552 | fail: | |
2553 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2554 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2555 | namespace_lock(); |
6776db3d | 2556 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2557 | namespace_unlock(); |
19a167af | 2558 | } |
b1e75df4 AV |
2559 | mntput(m); |
2560 | mntput(m); | |
19a167af AV |
2561 | return err; |
2562 | } | |
2563 | ||
ea5b778a DH |
2564 | /** |
2565 | * mnt_set_expiry - Put a mount on an expiration list | |
2566 | * @mnt: The mount to list. | |
2567 | * @expiry_list: The list to add the mount to. | |
2568 | */ | |
2569 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2570 | { | |
97216be0 | 2571 | namespace_lock(); |
ea5b778a | 2572 | |
6776db3d | 2573 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2574 | |
97216be0 | 2575 | namespace_unlock(); |
ea5b778a DH |
2576 | } |
2577 | EXPORT_SYMBOL(mnt_set_expiry); | |
2578 | ||
1da177e4 LT |
2579 | /* |
2580 | * process a list of expirable mountpoints with the intent of discarding any | |
2581 | * mountpoints that aren't in use and haven't been touched since last we came | |
2582 | * here | |
2583 | */ | |
2584 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2585 | { | |
761d5c38 | 2586 | struct mount *mnt, *next; |
1da177e4 LT |
2587 | LIST_HEAD(graveyard); |
2588 | ||
2589 | if (list_empty(mounts)) | |
2590 | return; | |
2591 | ||
97216be0 | 2592 | namespace_lock(); |
719ea2fb | 2593 | lock_mount_hash(); |
1da177e4 LT |
2594 | |
2595 | /* extract from the expiration list every vfsmount that matches the | |
2596 | * following criteria: | |
2597 | * - only referenced by its parent vfsmount | |
2598 | * - still marked for expiry (marked on the last call here; marks are | |
2599 | * cleared by mntput()) | |
2600 | */ | |
6776db3d | 2601 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2602 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2603 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2604 | continue; |
6776db3d | 2605 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2606 | } |
bcc5c7d2 | 2607 | while (!list_empty(&graveyard)) { |
6776db3d | 2608 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2609 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2610 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2611 | } |
719ea2fb | 2612 | unlock_mount_hash(); |
3ab6abee | 2613 | namespace_unlock(); |
5528f911 TM |
2614 | } |
2615 | ||
2616 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2617 | ||
2618 | /* | |
2619 | * Ripoff of 'select_parent()' | |
2620 | * | |
2621 | * search the list of submounts for a given mountpoint, and move any | |
2622 | * shrinkable submounts to the 'graveyard' list. | |
2623 | */ | |
692afc31 | 2624 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2625 | { |
692afc31 | 2626 | struct mount *this_parent = parent; |
5528f911 TM |
2627 | struct list_head *next; |
2628 | int found = 0; | |
2629 | ||
2630 | repeat: | |
6b41d536 | 2631 | next = this_parent->mnt_mounts.next; |
5528f911 | 2632 | resume: |
6b41d536 | 2633 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2634 | struct list_head *tmp = next; |
6b41d536 | 2635 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2636 | |
2637 | next = tmp->next; | |
692afc31 | 2638 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2639 | continue; |
5528f911 TM |
2640 | /* |
2641 | * Descend a level if the d_mounts list is non-empty. | |
2642 | */ | |
6b41d536 | 2643 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2644 | this_parent = mnt; |
2645 | goto repeat; | |
2646 | } | |
1da177e4 | 2647 | |
1ab59738 | 2648 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2649 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2650 | found++; |
2651 | } | |
1da177e4 | 2652 | } |
5528f911 TM |
2653 | /* |
2654 | * All done at this level ... ascend and resume the search | |
2655 | */ | |
2656 | if (this_parent != parent) { | |
6b41d536 | 2657 | next = this_parent->mnt_child.next; |
0714a533 | 2658 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2659 | goto resume; |
2660 | } | |
2661 | return found; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * process a list of expirable mountpoints with the intent of discarding any | |
2666 | * submounts of a specific parent mountpoint | |
99b7db7b | 2667 | * |
48a066e7 | 2668 | * mount_lock must be held for write |
5528f911 | 2669 | */ |
b54b9be7 | 2670 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2671 | { |
2672 | LIST_HEAD(graveyard); | |
761d5c38 | 2673 | struct mount *m; |
5528f911 | 2674 | |
5528f911 | 2675 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2676 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2677 | while (!list_empty(&graveyard)) { |
761d5c38 | 2678 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2679 | mnt_expire); |
143c8c91 | 2680 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2681 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2682 | } |
2683 | } | |
1da177e4 LT |
2684 | } |
2685 | ||
1da177e4 LT |
2686 | /* |
2687 | * Some copy_from_user() implementations do not return the exact number of | |
2688 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2689 | * Note that this function differs from copy_from_user() in that it will oops | |
2690 | * on bad values of `to', rather than returning a short copy. | |
2691 | */ | |
b58fed8b RP |
2692 | static long exact_copy_from_user(void *to, const void __user * from, |
2693 | unsigned long n) | |
1da177e4 LT |
2694 | { |
2695 | char *t = to; | |
2696 | const char __user *f = from; | |
2697 | char c; | |
2698 | ||
2699 | if (!access_ok(VERIFY_READ, from, n)) | |
2700 | return n; | |
2701 | ||
2702 | while (n) { | |
2703 | if (__get_user(c, f)) { | |
2704 | memset(t, 0, n); | |
2705 | break; | |
2706 | } | |
2707 | *t++ = c; | |
2708 | f++; | |
2709 | n--; | |
2710 | } | |
2711 | return n; | |
2712 | } | |
2713 | ||
b40ef869 | 2714 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2715 | { |
2716 | int i; | |
1da177e4 | 2717 | unsigned long size; |
b40ef869 | 2718 | char *copy; |
b58fed8b | 2719 | |
1da177e4 | 2720 | if (!data) |
b40ef869 | 2721 | return NULL; |
1da177e4 | 2722 | |
b40ef869 AV |
2723 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2724 | if (!copy) | |
2725 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2726 | |
2727 | /* We only care that *some* data at the address the user | |
2728 | * gave us is valid. Just in case, we'll zero | |
2729 | * the remainder of the page. | |
2730 | */ | |
2731 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2732 | size = TASK_SIZE - (unsigned long)data; | |
2733 | if (size > PAGE_SIZE) | |
2734 | size = PAGE_SIZE; | |
2735 | ||
b40ef869 | 2736 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2737 | if (!i) { |
b40ef869 AV |
2738 | kfree(copy); |
2739 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2740 | } |
2741 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2742 | memset(copy + i, 0, PAGE_SIZE - i); |
2743 | return copy; | |
1da177e4 LT |
2744 | } |
2745 | ||
b8850d1f | 2746 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2747 | { |
b8850d1f | 2748 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2749 | } |
2750 | ||
1da177e4 LT |
2751 | /* |
2752 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2753 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2754 | * | |
2755 | * data is a (void *) that can point to any structure up to | |
2756 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2757 | * information (or be NULL). | |
2758 | * | |
2759 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2760 | * When the flags word was introduced its top half was required | |
2761 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2762 | * Therefore, if this magic number is present, it carries no information | |
2763 | * and must be discarded. | |
2764 | */ | |
5e6123f3 | 2765 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2766 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2767 | { |
2d92ab3c | 2768 | struct path path; |
1da177e4 LT |
2769 | int retval = 0; |
2770 | int mnt_flags = 0; | |
2771 | ||
2772 | /* Discard magic */ | |
2773 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2774 | flags &= ~MS_MGC_MSK; | |
2775 | ||
2776 | /* Basic sanity checks */ | |
1da177e4 LT |
2777 | if (data_page) |
2778 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2779 | ||
a27ab9f2 | 2780 | /* ... and get the mountpoint */ |
5e6123f3 | 2781 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2782 | if (retval) |
2783 | return retval; | |
2784 | ||
2785 | retval = security_sb_mount(dev_name, &path, | |
2786 | type_page, flags, data_page); | |
0d5cadb8 AV |
2787 | if (!retval && !may_mount()) |
2788 | retval = -EPERM; | |
9e8925b6 JL |
2789 | if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) |
2790 | retval = -EPERM; | |
a27ab9f2 TH |
2791 | if (retval) |
2792 | goto dput_out; | |
2793 | ||
613cbe3d AK |
2794 | /* Default to relatime unless overriden */ |
2795 | if (!(flags & MS_NOATIME)) | |
2796 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2797 | |
1da177e4 LT |
2798 | /* Separate the per-mountpoint flags */ |
2799 | if (flags & MS_NOSUID) | |
2800 | mnt_flags |= MNT_NOSUID; | |
2801 | if (flags & MS_NODEV) | |
2802 | mnt_flags |= MNT_NODEV; | |
2803 | if (flags & MS_NOEXEC) | |
2804 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2805 | if (flags & MS_NOATIME) |
2806 | mnt_flags |= MNT_NOATIME; | |
2807 | if (flags & MS_NODIRATIME) | |
2808 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2809 | if (flags & MS_STRICTATIME) |
2810 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2811 | if (flags & MS_RDONLY) |
2812 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2813 | |
ffbc6f0e EB |
2814 | /* The default atime for remount is preservation */ |
2815 | if ((flags & MS_REMOUNT) && | |
2816 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2817 | MS_STRICTATIME)) == 0)) { | |
2818 | mnt_flags &= ~MNT_ATIME_MASK; | |
2819 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2820 | } | |
2821 | ||
7a4dec53 | 2822 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 | 2823 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
c8d3e3d5 | 2824 | MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT); |
1da177e4 | 2825 | |
1da177e4 | 2826 | if (flags & MS_REMOUNT) |
2d92ab3c | 2827 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2828 | data_page); |
2829 | else if (flags & MS_BIND) | |
2d92ab3c | 2830 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2831 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2832 | retval = do_change_type(&path, flags); |
1da177e4 | 2833 | else if (flags & MS_MOVE) |
2d92ab3c | 2834 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2835 | else |
2d92ab3c | 2836 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2837 | dev_name, data_page); |
2838 | dput_out: | |
2d92ab3c | 2839 | path_put(&path); |
1da177e4 LT |
2840 | return retval; |
2841 | } | |
2842 | ||
537f7ccb EB |
2843 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2844 | { | |
2845 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2846 | } | |
2847 | ||
2848 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2849 | { | |
2850 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2851 | } | |
2852 | ||
771b1371 EB |
2853 | static void free_mnt_ns(struct mnt_namespace *ns) |
2854 | { | |
6344c433 | 2855 | ns_free_inum(&ns->ns); |
537f7ccb | 2856 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2857 | put_user_ns(ns->user_ns); |
2858 | kfree(ns); | |
2859 | } | |
2860 | ||
8823c079 EB |
2861 | /* |
2862 | * Assign a sequence number so we can detect when we attempt to bind | |
2863 | * mount a reference to an older mount namespace into the current | |
2864 | * mount namespace, preventing reference counting loops. A 64bit | |
2865 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2866 | * is effectively never, so we can ignore the possibility. | |
2867 | */ | |
2868 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2869 | ||
771b1371 | 2870 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2871 | { |
2872 | struct mnt_namespace *new_ns; | |
537f7ccb | 2873 | struct ucounts *ucounts; |
98f842e6 | 2874 | int ret; |
cf8d2c11 | 2875 | |
537f7ccb EB |
2876 | ucounts = inc_mnt_namespaces(user_ns); |
2877 | if (!ucounts) | |
df75e774 | 2878 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2879 | |
cf8d2c11 | 2880 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2881 | if (!new_ns) { |
2882 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2883 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2884 | } |
6344c433 | 2885 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2886 | if (ret) { |
2887 | kfree(new_ns); | |
537f7ccb | 2888 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2889 | return ERR_PTR(ret); |
2890 | } | |
33c42940 | 2891 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2892 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2893 | atomic_set(&new_ns->count, 1); |
2894 | new_ns->root = NULL; | |
2895 | INIT_LIST_HEAD(&new_ns->list); | |
2896 | init_waitqueue_head(&new_ns->poll); | |
2897 | new_ns->event = 0; | |
771b1371 | 2898 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2899 | new_ns->ucounts = ucounts; |
d2921684 EB |
2900 | new_ns->mounts = 0; |
2901 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2902 | return new_ns; |
2903 | } | |
2904 | ||
0766f788 | 2905 | __latent_entropy |
9559f689 AV |
2906 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2907 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2908 | { |
6b3286ed | 2909 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2910 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2911 | struct mount *p, *q; |
9559f689 | 2912 | struct mount *old; |
cb338d06 | 2913 | struct mount *new; |
7a472ef4 | 2914 | int copy_flags; |
1da177e4 | 2915 | |
9559f689 AV |
2916 | BUG_ON(!ns); |
2917 | ||
2918 | if (likely(!(flags & CLONE_NEWNS))) { | |
2919 | get_mnt_ns(ns); | |
2920 | return ns; | |
2921 | } | |
2922 | ||
2923 | old = ns->root; | |
2924 | ||
771b1371 | 2925 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2926 | if (IS_ERR(new_ns)) |
2927 | return new_ns; | |
1da177e4 | 2928 | |
97216be0 | 2929 | namespace_lock(); |
1da177e4 | 2930 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2931 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2932 | if (user_ns != ns->user_ns) |
132c94e3 | 2933 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2934 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2935 | if (IS_ERR(new)) { |
328e6d90 | 2936 | namespace_unlock(); |
771b1371 | 2937 | free_mnt_ns(new_ns); |
be34d1a3 | 2938 | return ERR_CAST(new); |
1da177e4 | 2939 | } |
be08d6d2 | 2940 | new_ns->root = new; |
1a4eeaf2 | 2941 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2942 | |
2943 | /* | |
2944 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2945 | * as belonging to new namespace. We have already acquired a private | |
2946 | * fs_struct, so tsk->fs->lock is not needed. | |
2947 | */ | |
909b0a88 | 2948 | p = old; |
cb338d06 | 2949 | q = new; |
1da177e4 | 2950 | while (p) { |
143c8c91 | 2951 | q->mnt_ns = new_ns; |
d2921684 | 2952 | new_ns->mounts++; |
9559f689 AV |
2953 | if (new_fs) { |
2954 | if (&p->mnt == new_fs->root.mnt) { | |
2955 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2956 | rootmnt = &p->mnt; |
1da177e4 | 2957 | } |
9559f689 AV |
2958 | if (&p->mnt == new_fs->pwd.mnt) { |
2959 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2960 | pwdmnt = &p->mnt; |
1da177e4 | 2961 | } |
1da177e4 | 2962 | } |
909b0a88 AV |
2963 | p = next_mnt(p, old); |
2964 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2965 | if (!q) |
2966 | break; | |
2967 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2968 | p = next_mnt(p, old); | |
1da177e4 | 2969 | } |
328e6d90 | 2970 | namespace_unlock(); |
1da177e4 | 2971 | |
1da177e4 | 2972 | if (rootmnt) |
f03c6599 | 2973 | mntput(rootmnt); |
1da177e4 | 2974 | if (pwdmnt) |
f03c6599 | 2975 | mntput(pwdmnt); |
1da177e4 | 2976 | |
741a2951 | 2977 | return new_ns; |
1da177e4 LT |
2978 | } |
2979 | ||
cf8d2c11 TM |
2980 | /** |
2981 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2982 | * @mnt: pointer to the new root filesystem mountpoint | |
2983 | */ | |
1a4eeaf2 | 2984 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2985 | { |
771b1371 | 2986 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2987 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2988 | struct mount *mnt = real_mount(m); |
2989 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2990 | new_ns->root = mnt; |
d2921684 | 2991 | new_ns->mounts++; |
b1983cd8 | 2992 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2993 | } else { |
1a4eeaf2 | 2994 | mntput(m); |
cf8d2c11 TM |
2995 | } |
2996 | return new_ns; | |
2997 | } | |
cf8d2c11 | 2998 | |
ea441d11 AV |
2999 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3000 | { | |
3001 | struct mnt_namespace *ns; | |
d31da0f0 | 3002 | struct super_block *s; |
ea441d11 AV |
3003 | struct path path; |
3004 | int err; | |
3005 | ||
3006 | ns = create_mnt_ns(mnt); | |
3007 | if (IS_ERR(ns)) | |
3008 | return ERR_CAST(ns); | |
3009 | ||
3010 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3011 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3012 | ||
3013 | put_mnt_ns(ns); | |
3014 | ||
3015 | if (err) | |
3016 | return ERR_PTR(err); | |
3017 | ||
3018 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3019 | s = path.mnt->mnt_sb; |
3020 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3021 | mntput(path.mnt); |
3022 | /* lock the sucker */ | |
d31da0f0 | 3023 | down_write(&s->s_umount); |
ea441d11 AV |
3024 | /* ... and return the root of (sub)tree on it */ |
3025 | return path.dentry; | |
3026 | } | |
3027 | EXPORT_SYMBOL(mount_subtree); | |
3028 | ||
bdc480e3 HC |
3029 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3030 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3031 | { |
eca6f534 VN |
3032 | int ret; |
3033 | char *kernel_type; | |
eca6f534 | 3034 | char *kernel_dev; |
b40ef869 | 3035 | void *options; |
1da177e4 | 3036 | |
b8850d1f TG |
3037 | kernel_type = copy_mount_string(type); |
3038 | ret = PTR_ERR(kernel_type); | |
3039 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3040 | goto out_type; |
1da177e4 | 3041 | |
b8850d1f TG |
3042 | kernel_dev = copy_mount_string(dev_name); |
3043 | ret = PTR_ERR(kernel_dev); | |
3044 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3045 | goto out_dev; |
1da177e4 | 3046 | |
b40ef869 AV |
3047 | options = copy_mount_options(data); |
3048 | ret = PTR_ERR(options); | |
3049 | if (IS_ERR(options)) | |
eca6f534 | 3050 | goto out_data; |
1da177e4 | 3051 | |
b40ef869 | 3052 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3053 | |
b40ef869 | 3054 | kfree(options); |
eca6f534 VN |
3055 | out_data: |
3056 | kfree(kernel_dev); | |
3057 | out_dev: | |
eca6f534 VN |
3058 | kfree(kernel_type); |
3059 | out_type: | |
3060 | return ret; | |
1da177e4 LT |
3061 | } |
3062 | ||
afac7cba AV |
3063 | /* |
3064 | * Return true if path is reachable from root | |
3065 | * | |
48a066e7 | 3066 | * namespace_sem or mount_lock is held |
afac7cba | 3067 | */ |
643822b4 | 3068 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3069 | const struct path *root) |
3070 | { | |
643822b4 | 3071 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3072 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3073 | mnt = mnt->mnt_parent; |
afac7cba | 3074 | } |
643822b4 | 3075 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3076 | } |
3077 | ||
640eb7e7 | 3078 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3079 | { |
25ab4c9b | 3080 | bool res; |
48a066e7 | 3081 | read_seqlock_excl(&mount_lock); |
643822b4 | 3082 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3083 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3084 | return res; |
3085 | } | |
3086 | EXPORT_SYMBOL(path_is_under); | |
3087 | ||
1da177e4 LT |
3088 | /* |
3089 | * pivot_root Semantics: | |
3090 | * Moves the root file system of the current process to the directory put_old, | |
3091 | * makes new_root as the new root file system of the current process, and sets | |
3092 | * root/cwd of all processes which had them on the current root to new_root. | |
3093 | * | |
3094 | * Restrictions: | |
3095 | * The new_root and put_old must be directories, and must not be on the | |
3096 | * same file system as the current process root. The put_old must be | |
3097 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3098 | * pointed to by put_old must yield the same directory as new_root. No other | |
3099 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3100 | * | |
4a0d11fa NB |
3101 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3102 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3103 | * in this situation. | |
3104 | * | |
1da177e4 LT |
3105 | * Notes: |
3106 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3107 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3108 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3109 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3110 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3111 | * first. | |
3112 | */ | |
3480b257 HC |
3113 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3114 | const char __user *, put_old) | |
1da177e4 | 3115 | { |
2d8f3038 | 3116 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3117 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3118 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3119 | int error; |
3120 | ||
9b40bc90 | 3121 | if (!may_mount()) |
1da177e4 LT |
3122 | return -EPERM; |
3123 | ||
2d8f3038 | 3124 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3125 | if (error) |
3126 | goto out0; | |
1da177e4 | 3127 | |
2d8f3038 | 3128 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3129 | if (error) |
3130 | goto out1; | |
3131 | ||
2d8f3038 | 3132 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3133 | if (error) |
3134 | goto out2; | |
1da177e4 | 3135 | |
f7ad3c6b | 3136 | get_fs_root(current->fs, &root); |
84d17192 AV |
3137 | old_mp = lock_mount(&old); |
3138 | error = PTR_ERR(old_mp); | |
3139 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3140 | goto out3; |
3141 | ||
1da177e4 | 3142 | error = -EINVAL; |
419148da AV |
3143 | new_mnt = real_mount(new.mnt); |
3144 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3145 | old_mnt = real_mount(old.mnt); |
3146 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3147 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3148 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3149 | goto out4; |
143c8c91 | 3150 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3151 | goto out4; |
5ff9d8a6 EB |
3152 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3153 | goto out4; | |
1da177e4 | 3154 | error = -ENOENT; |
f3da392e | 3155 | if (d_unlinked(new.dentry)) |
b12cea91 | 3156 | goto out4; |
1da177e4 | 3157 | error = -EBUSY; |
84d17192 | 3158 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3159 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3160 | error = -EINVAL; |
8c3ee42e | 3161 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3162 | goto out4; /* not a mountpoint */ |
676da58d | 3163 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3164 | goto out4; /* not attached */ |
84d17192 | 3165 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3166 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3167 | goto out4; /* not a mountpoint */ |
676da58d | 3168 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3169 | goto out4; /* not attached */ |
4ac91378 | 3170 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3171 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3172 | goto out4; |
0d082601 EB |
3173 | /* make certain new is below the root */ |
3174 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3175 | goto out4; | |
84d17192 | 3176 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3177 | lock_mount_hash(); |
419148da AV |
3178 | detach_mnt(new_mnt, &parent_path); |
3179 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3180 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3181 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3182 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3183 | } | |
4ac91378 | 3184 | /* mount old root on put_old */ |
84d17192 | 3185 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3186 | /* mount new_root on / */ |
84d17192 | 3187 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3188 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3189 | /* A moved mount should not expire automatically */ |
3190 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3191 | put_mountpoint(root_mp); |
719ea2fb | 3192 | unlock_mount_hash(); |
2d8f3038 | 3193 | chroot_fs_refs(&root, &new); |
1da177e4 | 3194 | error = 0; |
b12cea91 | 3195 | out4: |
84d17192 | 3196 | unlock_mount(old_mp); |
b12cea91 AV |
3197 | if (!error) { |
3198 | path_put(&root_parent); | |
3199 | path_put(&parent_path); | |
3200 | } | |
3201 | out3: | |
8c3ee42e | 3202 | path_put(&root); |
b12cea91 | 3203 | out2: |
2d8f3038 | 3204 | path_put(&old); |
1da177e4 | 3205 | out1: |
2d8f3038 | 3206 | path_put(&new); |
1da177e4 | 3207 | out0: |
1da177e4 | 3208 | return error; |
1da177e4 LT |
3209 | } |
3210 | ||
3211 | static void __init init_mount_tree(void) | |
3212 | { | |
3213 | struct vfsmount *mnt; | |
6b3286ed | 3214 | struct mnt_namespace *ns; |
ac748a09 | 3215 | struct path root; |
0c55cfc4 | 3216 | struct file_system_type *type; |
1da177e4 | 3217 | |
0c55cfc4 EB |
3218 | type = get_fs_type("rootfs"); |
3219 | if (!type) | |
3220 | panic("Can't find rootfs type"); | |
3221 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3222 | put_filesystem(type); | |
1da177e4 LT |
3223 | if (IS_ERR(mnt)) |
3224 | panic("Can't create rootfs"); | |
b3e19d92 | 3225 | |
3b22edc5 TM |
3226 | ns = create_mnt_ns(mnt); |
3227 | if (IS_ERR(ns)) | |
1da177e4 | 3228 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3229 | |
3230 | init_task.nsproxy->mnt_ns = ns; | |
3231 | get_mnt_ns(ns); | |
3232 | ||
be08d6d2 AV |
3233 | root.mnt = mnt; |
3234 | root.dentry = mnt->mnt_root; | |
da362b09 | 3235 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3236 | |
3237 | set_fs_pwd(current->fs, &root); | |
3238 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3239 | } |
3240 | ||
74bf17cf | 3241 | void __init mnt_init(void) |
1da177e4 | 3242 | { |
13f14b4d | 3243 | unsigned u; |
15a67dd8 | 3244 | int err; |
1da177e4 | 3245 | |
7d6fec45 | 3246 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3247 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3248 | |
0818bf27 | 3249 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3250 | sizeof(struct hlist_head), |
0818bf27 AV |
3251 | mhash_entries, 19, |
3252 | 0, | |
3253 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3254 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3255 | sizeof(struct hlist_head), | |
3256 | mphash_entries, 19, | |
3257 | 0, | |
3258 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3259 | |
84d17192 | 3260 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3261 | panic("Failed to allocate mount hash table\n"); |
3262 | ||
0818bf27 | 3263 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3264 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3265 | for (u = 0; u <= mp_hash_mask; u++) |
3266 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3267 | |
4b93dc9b TH |
3268 | kernfs_init(); |
3269 | ||
15a67dd8 RD |
3270 | err = sysfs_init(); |
3271 | if (err) | |
3272 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3273 | __func__, err); |
00d26666 GKH |
3274 | fs_kobj = kobject_create_and_add("fs", NULL); |
3275 | if (!fs_kobj) | |
8e24eea7 | 3276 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3277 | init_rootfs(); |
3278 | init_mount_tree(); | |
3279 | } | |
3280 | ||
616511d0 | 3281 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3282 | { |
d498b25a | 3283 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3284 | return; |
7b00ed6f | 3285 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3286 | free_mnt_ns(ns); |
1da177e4 | 3287 | } |
9d412a43 AV |
3288 | |
3289 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3290 | { | |
423e0ab0 TC |
3291 | struct vfsmount *mnt; |
3292 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3293 | if (!IS_ERR(mnt)) { | |
3294 | /* | |
3295 | * it is a longterm mount, don't release mnt until | |
3296 | * we unmount before file sys is unregistered | |
3297 | */ | |
f7a99c5b | 3298 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3299 | } |
3300 | return mnt; | |
9d412a43 AV |
3301 | } |
3302 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3303 | |
3304 | void kern_unmount(struct vfsmount *mnt) | |
3305 | { | |
3306 | /* release long term mount so mount point can be released */ | |
3307 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3308 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3309 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3310 | mntput(mnt); |
3311 | } | |
3312 | } | |
3313 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3314 | |
3315 | bool our_mnt(struct vfsmount *mnt) | |
3316 | { | |
143c8c91 | 3317 | return check_mnt(real_mount(mnt)); |
02125a82 | 3318 | } |
8823c079 | 3319 | |
3151527e EB |
3320 | bool current_chrooted(void) |
3321 | { | |
3322 | /* Does the current process have a non-standard root */ | |
3323 | struct path ns_root; | |
3324 | struct path fs_root; | |
3325 | bool chrooted; | |
3326 | ||
3327 | /* Find the namespace root */ | |
3328 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3329 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3330 | path_get(&ns_root); | |
3331 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3332 | ; | |
3333 | ||
3334 | get_fs_root(current->fs, &fs_root); | |
3335 | ||
3336 | chrooted = !path_equal(&fs_root, &ns_root); | |
3337 | ||
3338 | path_put(&fs_root); | |
3339 | path_put(&ns_root); | |
3340 | ||
3341 | return chrooted; | |
3342 | } | |
3343 | ||
8654df4e EB |
3344 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3345 | int *new_mnt_flags) | |
87a8ebd6 | 3346 | { |
8c6cf9cc | 3347 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3348 | struct mount *mnt; |
e51db735 | 3349 | bool visible = false; |
87a8ebd6 | 3350 | |
44bb4385 | 3351 | down_read(&namespace_sem); |
87a8ebd6 | 3352 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3353 | struct mount *child; |
77b1a97d EB |
3354 | int mnt_flags; |
3355 | ||
8654df4e | 3356 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3357 | continue; |
3358 | ||
7e96c1b0 EB |
3359 | /* This mount is not fully visible if it's root directory |
3360 | * is not the root directory of the filesystem. | |
3361 | */ | |
3362 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3363 | continue; | |
3364 | ||
a1935c17 | 3365 | /* A local view of the mount flags */ |
77b1a97d | 3366 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3367 | |
695e9df0 EB |
3368 | /* Don't miss readonly hidden in the superblock flags */ |
3369 | if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) | |
3370 | mnt_flags |= MNT_LOCK_READONLY; | |
3371 | ||
8c6cf9cc EB |
3372 | /* Verify the mount flags are equal to or more permissive |
3373 | * than the proposed new mount. | |
3374 | */ | |
77b1a97d | 3375 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3376 | !(new_flags & MNT_READONLY)) |
3377 | continue; | |
77b1a97d EB |
3378 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3379 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3380 | continue; |
3381 | ||
ceeb0e5d EB |
3382 | /* This mount is not fully visible if there are any |
3383 | * locked child mounts that cover anything except for | |
3384 | * empty directories. | |
e51db735 EB |
3385 | */ |
3386 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3387 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3388 | /* Only worry about locked mounts */ |
d71ed6c9 | 3389 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3390 | continue; |
7236c85e EB |
3391 | /* Is the directory permanetly empty? */ |
3392 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3393 | goto next; |
87a8ebd6 | 3394 | } |
8c6cf9cc | 3395 | /* Preserve the locked attributes */ |
77b1a97d | 3396 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3397 | MNT_LOCK_ATIME); |
e51db735 EB |
3398 | visible = true; |
3399 | goto found; | |
3400 | next: ; | |
87a8ebd6 | 3401 | } |
e51db735 | 3402 | found: |
44bb4385 | 3403 | up_read(&namespace_sem); |
e51db735 | 3404 | return visible; |
87a8ebd6 EB |
3405 | } |
3406 | ||
8654df4e EB |
3407 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3408 | { | |
a1935c17 | 3409 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3410 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3411 | unsigned long s_iflags; | |
3412 | ||
3413 | if (ns->user_ns == &init_user_ns) | |
3414 | return false; | |
3415 | ||
3416 | /* Can this filesystem be too revealing? */ | |
3417 | s_iflags = mnt->mnt_sb->s_iflags; | |
3418 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3419 | return false; | |
3420 | ||
a1935c17 EB |
3421 | if ((s_iflags & required_iflags) != required_iflags) { |
3422 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3423 | required_iflags); | |
3424 | return true; | |
3425 | } | |
3426 | ||
8654df4e EB |
3427 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3428 | } | |
3429 | ||
380cf5ba AL |
3430 | bool mnt_may_suid(struct vfsmount *mnt) |
3431 | { | |
3432 | /* | |
3433 | * Foreign mounts (accessed via fchdir or through /proc | |
3434 | * symlinks) are always treated as if they are nosuid. This | |
3435 | * prevents namespaces from trusting potentially unsafe | |
3436 | * suid/sgid bits, file caps, or security labels that originate | |
3437 | * in other namespaces. | |
3438 | */ | |
3439 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3440 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3441 | } | |
3442 | ||
64964528 | 3443 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3444 | { |
58be2825 | 3445 | struct ns_common *ns = NULL; |
8823c079 EB |
3446 | struct nsproxy *nsproxy; |
3447 | ||
728dba3a EB |
3448 | task_lock(task); |
3449 | nsproxy = task->nsproxy; | |
8823c079 | 3450 | if (nsproxy) { |
58be2825 AV |
3451 | ns = &nsproxy->mnt_ns->ns; |
3452 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3453 | } |
728dba3a | 3454 | task_unlock(task); |
8823c079 EB |
3455 | |
3456 | return ns; | |
3457 | } | |
3458 | ||
64964528 | 3459 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3460 | { |
58be2825 | 3461 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3462 | } |
3463 | ||
64964528 | 3464 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3465 | { |
3466 | struct fs_struct *fs = current->fs; | |
58be2825 | 3467 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3468 | struct path root; |
3469 | ||
0c55cfc4 | 3470 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3471 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3472 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3473 | return -EPERM; |
8823c079 EB |
3474 | |
3475 | if (fs->users != 1) | |
3476 | return -EINVAL; | |
3477 | ||
3478 | get_mnt_ns(mnt_ns); | |
3479 | put_mnt_ns(nsproxy->mnt_ns); | |
3480 | nsproxy->mnt_ns = mnt_ns; | |
3481 | ||
3482 | /* Find the root */ | |
3483 | root.mnt = &mnt_ns->root->mnt; | |
3484 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3485 | path_get(&root); | |
3486 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3487 | ; | |
3488 | ||
3489 | /* Update the pwd and root */ | |
3490 | set_fs_pwd(fs, &root); | |
3491 | set_fs_root(fs, &root); | |
3492 | ||
3493 | path_put(&root); | |
3494 | return 0; | |
3495 | } | |
3496 | ||
bcac25a5 AV |
3497 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3498 | { | |
3499 | return to_mnt_ns(ns)->user_ns; | |
3500 | } | |
3501 | ||
8823c079 EB |
3502 | const struct proc_ns_operations mntns_operations = { |
3503 | .name = "mnt", | |
3504 | .type = CLONE_NEWNS, | |
3505 | .get = mntns_get, | |
3506 | .put = mntns_put, | |
3507 | .install = mntns_install, | |
bcac25a5 | 3508 | .owner = mntns_owner, |
8823c079 | 3509 | }; |