Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
9ea459e1 | 26 | #include <linux/task_work.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
0818bf27 AV |
30 | static unsigned int m_hash_mask __read_mostly; |
31 | static unsigned int m_hash_shift __read_mostly; | |
32 | static unsigned int mp_hash_mask __read_mostly; | |
33 | static unsigned int mp_hash_shift __read_mostly; | |
34 | ||
35 | static __initdata unsigned long mhash_entries; | |
36 | static int __init set_mhash_entries(char *str) | |
37 | { | |
38 | if (!str) | |
39 | return 0; | |
40 | mhash_entries = simple_strtoul(str, &str, 0); | |
41 | return 1; | |
42 | } | |
43 | __setup("mhash_entries=", set_mhash_entries); | |
44 | ||
45 | static __initdata unsigned long mphash_entries; | |
46 | static int __init set_mphash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mphash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 54 | |
c7999c36 | 55 | static u64 event; |
73cd49ec | 56 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 57 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 58 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
59 | static int mnt_id_start = 0; |
60 | static int mnt_group_start = 1; | |
1da177e4 | 61 | |
38129a13 | 62 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 63 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 64 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 65 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 66 | |
f87fd4c2 | 67 | /* /sys/fs */ |
00d26666 GKH |
68 | struct kobject *fs_kobj; |
69 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 70 | |
99b7db7b NP |
71 | /* |
72 | * vfsmount lock may be taken for read to prevent changes to the | |
73 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
74 | * up the tree. | |
75 | * | |
76 | * It should be taken for write in all cases where the vfsmount | |
77 | * tree or hash is modified or when a vfsmount structure is modified. | |
78 | */ | |
48a066e7 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 80 | |
38129a13 | 81 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 82 | { |
b58fed8b RP |
83 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
84 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
85 | tmp = tmp + (tmp >> m_hash_shift); |
86 | return &mount_hashtable[tmp & m_hash_mask]; | |
87 | } | |
88 | ||
89 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
90 | { | |
91 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
92 | tmp = tmp + (tmp >> mp_hash_shift); | |
93 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
94 | } |
95 | ||
99b7db7b NP |
96 | /* |
97 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
98 | * serialize with freeing. | |
99 | */ | |
b105e270 | 100 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
101 | { |
102 | int res; | |
103 | ||
104 | retry: | |
105 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 106 | spin_lock(&mnt_id_lock); |
15169fe7 | 107 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 108 | if (!res) |
15169fe7 | 109 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 110 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
111 | if (res == -EAGAIN) |
112 | goto retry; | |
113 | ||
114 | return res; | |
115 | } | |
116 | ||
b105e270 | 117 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 118 | { |
15169fe7 | 119 | int id = mnt->mnt_id; |
99b7db7b | 120 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
121 | ida_remove(&mnt_id_ida, id); |
122 | if (mnt_id_start > id) | |
123 | mnt_id_start = id; | |
99b7db7b | 124 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
125 | } |
126 | ||
719f5d7f MS |
127 | /* |
128 | * Allocate a new peer group ID | |
129 | * | |
130 | * mnt_group_ida is protected by namespace_sem | |
131 | */ | |
4b8b21f4 | 132 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 133 | { |
f21f6220 AV |
134 | int res; |
135 | ||
719f5d7f MS |
136 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
137 | return -ENOMEM; | |
138 | ||
f21f6220 AV |
139 | res = ida_get_new_above(&mnt_group_ida, |
140 | mnt_group_start, | |
15169fe7 | 141 | &mnt->mnt_group_id); |
f21f6220 | 142 | if (!res) |
15169fe7 | 143 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
144 | |
145 | return res; | |
719f5d7f MS |
146 | } |
147 | ||
148 | /* | |
149 | * Release a peer group ID | |
150 | */ | |
4b8b21f4 | 151 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 152 | { |
15169fe7 | 153 | int id = mnt->mnt_group_id; |
f21f6220 AV |
154 | ida_remove(&mnt_group_ida, id); |
155 | if (mnt_group_start > id) | |
156 | mnt_group_start = id; | |
15169fe7 | 157 | mnt->mnt_group_id = 0; |
719f5d7f MS |
158 | } |
159 | ||
b3e19d92 NP |
160 | /* |
161 | * vfsmount lock must be held for read | |
162 | */ | |
83adc753 | 163 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
164 | { |
165 | #ifdef CONFIG_SMP | |
68e8a9fe | 166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
167 | #else |
168 | preempt_disable(); | |
68e8a9fe | 169 | mnt->mnt_count += n; |
b3e19d92 NP |
170 | preempt_enable(); |
171 | #endif | |
172 | } | |
173 | ||
b3e19d92 NP |
174 | /* |
175 | * vfsmount lock must be held for write | |
176 | */ | |
83adc753 | 177 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
178 | { |
179 | #ifdef CONFIG_SMP | |
f03c6599 | 180 | unsigned int count = 0; |
b3e19d92 NP |
181 | int cpu; |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
185 | } |
186 | ||
187 | return count; | |
188 | #else | |
68e8a9fe | 189 | return mnt->mnt_count; |
b3e19d92 NP |
190 | #endif |
191 | } | |
192 | ||
87b95ce0 AV |
193 | static void drop_mountpoint(struct fs_pin *p) |
194 | { | |
195 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
196 | dput(m->mnt_ex_mountpoint); | |
197 | pin_remove(p); | |
198 | mntput(&m->mnt); | |
199 | } | |
200 | ||
b105e270 | 201 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 202 | { |
c63181e6 AV |
203 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
204 | if (mnt) { | |
73cd49ec MS |
205 | int err; |
206 | ||
c63181e6 | 207 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
208 | if (err) |
209 | goto out_free_cache; | |
210 | ||
211 | if (name) { | |
fcc139ae | 212 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 213 | if (!mnt->mnt_devname) |
88b38782 | 214 | goto out_free_id; |
73cd49ec MS |
215 | } |
216 | ||
b3e19d92 | 217 | #ifdef CONFIG_SMP |
c63181e6 AV |
218 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
219 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
220 | goto out_free_devname; |
221 | ||
c63181e6 | 222 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 223 | #else |
c63181e6 AV |
224 | mnt->mnt_count = 1; |
225 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
226 | #endif |
227 | ||
38129a13 | 228 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
229 | INIT_LIST_HEAD(&mnt->mnt_child); |
230 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
231 | INIT_LIST_HEAD(&mnt->mnt_list); | |
232 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
233 | INIT_LIST_HEAD(&mnt->mnt_share); | |
234 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
235 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 236 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
2504c5d6 AG |
237 | #ifdef CONFIG_FSNOTIFY |
238 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 239 | #endif |
87b95ce0 | 240 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 241 | } |
c63181e6 | 242 | return mnt; |
88b38782 | 243 | |
d3ef3d73 | 244 | #ifdef CONFIG_SMP |
245 | out_free_devname: | |
fcc139ae | 246 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 247 | #endif |
88b38782 | 248 | out_free_id: |
c63181e6 | 249 | mnt_free_id(mnt); |
88b38782 | 250 | out_free_cache: |
c63181e6 | 251 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 252 | return NULL; |
1da177e4 LT |
253 | } |
254 | ||
3d733633 DH |
255 | /* |
256 | * Most r/o checks on a fs are for operations that take | |
257 | * discrete amounts of time, like a write() or unlink(). | |
258 | * We must keep track of when those operations start | |
259 | * (for permission checks) and when they end, so that | |
260 | * we can determine when writes are able to occur to | |
261 | * a filesystem. | |
262 | */ | |
263 | /* | |
264 | * __mnt_is_readonly: check whether a mount is read-only | |
265 | * @mnt: the mount to check for its write status | |
266 | * | |
267 | * This shouldn't be used directly ouside of the VFS. | |
268 | * It does not guarantee that the filesystem will stay | |
269 | * r/w, just that it is right *now*. This can not and | |
270 | * should not be used in place of IS_RDONLY(inode). | |
271 | * mnt_want/drop_write() will _keep_ the filesystem | |
272 | * r/w. | |
273 | */ | |
274 | int __mnt_is_readonly(struct vfsmount *mnt) | |
275 | { | |
2e4b7fcd DH |
276 | if (mnt->mnt_flags & MNT_READONLY) |
277 | return 1; | |
278 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
279 | return 1; | |
280 | return 0; | |
3d733633 DH |
281 | } |
282 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
283 | ||
83adc753 | 284 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 285 | { |
286 | #ifdef CONFIG_SMP | |
68e8a9fe | 287 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 288 | #else |
68e8a9fe | 289 | mnt->mnt_writers++; |
d3ef3d73 | 290 | #endif |
291 | } | |
3d733633 | 292 | |
83adc753 | 293 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 294 | { |
d3ef3d73 | 295 | #ifdef CONFIG_SMP |
68e8a9fe | 296 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 297 | #else |
68e8a9fe | 298 | mnt->mnt_writers--; |
d3ef3d73 | 299 | #endif |
3d733633 | 300 | } |
3d733633 | 301 | |
83adc753 | 302 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 303 | { |
d3ef3d73 | 304 | #ifdef CONFIG_SMP |
305 | unsigned int count = 0; | |
3d733633 | 306 | int cpu; |
3d733633 DH |
307 | |
308 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 309 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 310 | } |
3d733633 | 311 | |
d3ef3d73 | 312 | return count; |
313 | #else | |
314 | return mnt->mnt_writers; | |
315 | #endif | |
3d733633 DH |
316 | } |
317 | ||
4ed5e82f MS |
318 | static int mnt_is_readonly(struct vfsmount *mnt) |
319 | { | |
320 | if (mnt->mnt_sb->s_readonly_remount) | |
321 | return 1; | |
322 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
323 | smp_rmb(); | |
324 | return __mnt_is_readonly(mnt); | |
325 | } | |
326 | ||
8366025e | 327 | /* |
eb04c282 JK |
328 | * Most r/o & frozen checks on a fs are for operations that take discrete |
329 | * amounts of time, like a write() or unlink(). We must keep track of when | |
330 | * those operations start (for permission checks) and when they end, so that we | |
331 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
332 | */ |
333 | /** | |
eb04c282 | 334 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 335 | * @m: the mount on which to take a write |
8366025e | 336 | * |
eb04c282 JK |
337 | * This tells the low-level filesystem that a write is about to be performed to |
338 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
339 | * returning success. This operation does not protect against filesystem being | |
340 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
341 | * called. This is effectively a refcount. | |
8366025e | 342 | */ |
eb04c282 | 343 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 344 | { |
83adc753 | 345 | struct mount *mnt = real_mount(m); |
3d733633 | 346 | int ret = 0; |
3d733633 | 347 | |
d3ef3d73 | 348 | preempt_disable(); |
c6653a83 | 349 | mnt_inc_writers(mnt); |
d3ef3d73 | 350 | /* |
c6653a83 | 351 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 352 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
353 | * incremented count after it has set MNT_WRITE_HOLD. | |
354 | */ | |
355 | smp_mb(); | |
1e75529e | 356 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 357 | cpu_relax(); |
358 | /* | |
359 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
360 | * be set to match its requirements. So we must not load that until | |
361 | * MNT_WRITE_HOLD is cleared. | |
362 | */ | |
363 | smp_rmb(); | |
4ed5e82f | 364 | if (mnt_is_readonly(m)) { |
c6653a83 | 365 | mnt_dec_writers(mnt); |
3d733633 | 366 | ret = -EROFS; |
3d733633 | 367 | } |
d3ef3d73 | 368 | preempt_enable(); |
eb04c282 JK |
369 | |
370 | return ret; | |
371 | } | |
372 | ||
373 | /** | |
374 | * mnt_want_write - get write access to a mount | |
375 | * @m: the mount on which to take a write | |
376 | * | |
377 | * This tells the low-level filesystem that a write is about to be performed to | |
378 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
379 | * is not frozen) before returning success. When the write operation is | |
380 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
381 | */ | |
382 | int mnt_want_write(struct vfsmount *m) | |
383 | { | |
384 | int ret; | |
385 | ||
386 | sb_start_write(m->mnt_sb); | |
387 | ret = __mnt_want_write(m); | |
388 | if (ret) | |
389 | sb_end_write(m->mnt_sb); | |
3d733633 | 390 | return ret; |
8366025e DH |
391 | } |
392 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
393 | ||
96029c4e | 394 | /** |
395 | * mnt_clone_write - get write access to a mount | |
396 | * @mnt: the mount on which to take a write | |
397 | * | |
398 | * This is effectively like mnt_want_write, except | |
399 | * it must only be used to take an extra write reference | |
400 | * on a mountpoint that we already know has a write reference | |
401 | * on it. This allows some optimisation. | |
402 | * | |
403 | * After finished, mnt_drop_write must be called as usual to | |
404 | * drop the reference. | |
405 | */ | |
406 | int mnt_clone_write(struct vfsmount *mnt) | |
407 | { | |
408 | /* superblock may be r/o */ | |
409 | if (__mnt_is_readonly(mnt)) | |
410 | return -EROFS; | |
411 | preempt_disable(); | |
83adc753 | 412 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 413 | preempt_enable(); |
414 | return 0; | |
415 | } | |
416 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
417 | ||
418 | /** | |
eb04c282 | 419 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 420 | * @file: the file who's mount on which to take a write |
421 | * | |
eb04c282 | 422 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 423 | * do some optimisations if the file is open for write already |
424 | */ | |
eb04c282 | 425 | int __mnt_want_write_file(struct file *file) |
96029c4e | 426 | { |
83f936c7 | 427 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 428 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 429 | else |
430 | return mnt_clone_write(file->f_path.mnt); | |
431 | } | |
eb04c282 JK |
432 | |
433 | /** | |
434 | * mnt_want_write_file - get write access to a file's mount | |
435 | * @file: the file who's mount on which to take a write | |
436 | * | |
437 | * This is like mnt_want_write, but it takes a file and can | |
438 | * do some optimisations if the file is open for write already | |
439 | */ | |
440 | int mnt_want_write_file(struct file *file) | |
441 | { | |
442 | int ret; | |
443 | ||
444 | sb_start_write(file->f_path.mnt->mnt_sb); | |
445 | ret = __mnt_want_write_file(file); | |
446 | if (ret) | |
447 | sb_end_write(file->f_path.mnt->mnt_sb); | |
448 | return ret; | |
449 | } | |
96029c4e | 450 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
451 | ||
8366025e | 452 | /** |
eb04c282 | 453 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
454 | * @mnt: the mount on which to give up write access |
455 | * | |
456 | * Tells the low-level filesystem that we are done | |
457 | * performing writes to it. Must be matched with | |
eb04c282 | 458 | * __mnt_want_write() call above. |
8366025e | 459 | */ |
eb04c282 | 460 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 461 | { |
d3ef3d73 | 462 | preempt_disable(); |
83adc753 | 463 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 464 | preempt_enable(); |
8366025e | 465 | } |
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 AV |
592 | /* call under rcu_read_lock */ |
593 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
594 | { | |
595 | struct mount *mnt; | |
596 | if (read_seqretry(&mount_lock, seq)) | |
597 | return false; | |
598 | if (bastard == NULL) | |
599 | return true; | |
600 | mnt = real_mount(bastard); | |
601 | mnt_add_count(mnt, 1); | |
602 | if (likely(!read_seqretry(&mount_lock, seq))) | |
603 | return true; | |
604 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
605 | mnt_add_count(mnt, -1); | |
606 | return false; | |
607 | } | |
608 | rcu_read_unlock(); | |
609 | mntput(bastard); | |
610 | rcu_read_lock(); | |
611 | return false; | |
612 | } | |
613 | ||
1da177e4 | 614 | /* |
474279dc | 615 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 616 | * call under rcu_read_lock() |
1da177e4 | 617 | */ |
474279dc | 618 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 619 | { |
38129a13 | 620 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
621 | struct mount *p; |
622 | ||
38129a13 | 623 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
624 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
625 | return p; | |
626 | return NULL; | |
627 | } | |
628 | ||
629 | /* | |
630 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 631 | * mount_lock must be held. |
474279dc AV |
632 | */ |
633 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
634 | { | |
411a938b EB |
635 | struct mount *p, *res = NULL; |
636 | p = __lookup_mnt(mnt, dentry); | |
38129a13 AV |
637 | if (!p) |
638 | goto out; | |
411a938b EB |
639 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
640 | res = p; | |
38129a13 | 641 | hlist_for_each_entry_continue(p, mnt_hash) { |
1d6a32ac AV |
642 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
643 | break; | |
411a938b EB |
644 | if (!(p->mnt.mnt_flags & MNT_UMOUNT)) |
645 | res = p; | |
1d6a32ac | 646 | } |
38129a13 | 647 | out: |
1d6a32ac | 648 | return res; |
1da177e4 LT |
649 | } |
650 | ||
a05964f3 | 651 | /* |
f015f126 DH |
652 | * lookup_mnt - Return the first child mount mounted at path |
653 | * | |
654 | * "First" means first mounted chronologically. If you create the | |
655 | * following mounts: | |
656 | * | |
657 | * mount /dev/sda1 /mnt | |
658 | * mount /dev/sda2 /mnt | |
659 | * mount /dev/sda3 /mnt | |
660 | * | |
661 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
662 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
663 | * /dev/sda2, then /dev/sda3, then NULL. | |
664 | * | |
665 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 666 | */ |
1c755af4 | 667 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 668 | { |
c7105365 | 669 | struct mount *child_mnt; |
48a066e7 AV |
670 | struct vfsmount *m; |
671 | unsigned seq; | |
99b7db7b | 672 | |
48a066e7 AV |
673 | rcu_read_lock(); |
674 | do { | |
675 | seq = read_seqbegin(&mount_lock); | |
676 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
677 | m = child_mnt ? &child_mnt->mnt : NULL; | |
678 | } while (!legitimize_mnt(m, seq)); | |
679 | rcu_read_unlock(); | |
680 | return m; | |
a05964f3 RP |
681 | } |
682 | ||
7af1364f EB |
683 | /* |
684 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
685 | * current mount namespace. | |
686 | * | |
687 | * The common case is dentries are not mountpoints at all and that | |
688 | * test is handled inline. For the slow case when we are actually | |
689 | * dealing with a mountpoint of some kind, walk through all of the | |
690 | * mounts in the current mount namespace and test to see if the dentry | |
691 | * is a mountpoint. | |
692 | * | |
693 | * The mount_hashtable is not usable in the context because we | |
694 | * need to identify all mounts that may be in the current mount | |
695 | * namespace not just a mount that happens to have some specified | |
696 | * parent mount. | |
697 | */ | |
698 | bool __is_local_mountpoint(struct dentry *dentry) | |
699 | { | |
700 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
701 | struct mount *mnt; | |
702 | bool is_covered = false; | |
703 | ||
704 | if (!d_mountpoint(dentry)) | |
705 | goto out; | |
706 | ||
707 | down_read(&namespace_sem); | |
708 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
709 | is_covered = (mnt->mnt_mountpoint == dentry); | |
710 | if (is_covered) | |
711 | break; | |
712 | } | |
713 | up_read(&namespace_sem); | |
714 | out: | |
715 | return is_covered; | |
716 | } | |
717 | ||
e2dfa935 | 718 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 719 | { |
0818bf27 | 720 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
721 | struct mountpoint *mp; |
722 | ||
0818bf27 | 723 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
724 | if (mp->m_dentry == dentry) { |
725 | /* might be worth a WARN_ON() */ | |
726 | if (d_unlinked(dentry)) | |
727 | return ERR_PTR(-ENOENT); | |
728 | mp->m_count++; | |
729 | return mp; | |
730 | } | |
731 | } | |
e2dfa935 EB |
732 | return NULL; |
733 | } | |
734 | ||
735 | static struct mountpoint *new_mountpoint(struct dentry *dentry) | |
736 | { | |
737 | struct hlist_head *chain = mp_hash(dentry); | |
738 | struct mountpoint *mp; | |
739 | int ret; | |
84d17192 AV |
740 | |
741 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
742 | if (!mp) | |
743 | return ERR_PTR(-ENOMEM); | |
744 | ||
eed81007 MS |
745 | ret = d_set_mounted(dentry); |
746 | if (ret) { | |
84d17192 | 747 | kfree(mp); |
eed81007 | 748 | return ERR_PTR(ret); |
84d17192 | 749 | } |
eed81007 | 750 | |
84d17192 AV |
751 | mp->m_dentry = dentry; |
752 | mp->m_count = 1; | |
0818bf27 | 753 | hlist_add_head(&mp->m_hash, chain); |
0a5eb7c8 | 754 | INIT_HLIST_HEAD(&mp->m_list); |
84d17192 AV |
755 | return mp; |
756 | } | |
757 | ||
758 | static void put_mountpoint(struct mountpoint *mp) | |
759 | { | |
760 | if (!--mp->m_count) { | |
761 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 762 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
763 | spin_lock(&dentry->d_lock); |
764 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
765 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 766 | hlist_del(&mp->m_hash); |
84d17192 AV |
767 | kfree(mp); |
768 | } | |
769 | } | |
770 | ||
143c8c91 | 771 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 772 | { |
6b3286ed | 773 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
774 | } |
775 | ||
99b7db7b NP |
776 | /* |
777 | * vfsmount lock must be held for write | |
778 | */ | |
6b3286ed | 779 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
780 | { |
781 | if (ns) { | |
782 | ns->event = ++event; | |
783 | wake_up_interruptible(&ns->poll); | |
784 | } | |
785 | } | |
786 | ||
99b7db7b NP |
787 | /* |
788 | * vfsmount lock must be held for write | |
789 | */ | |
6b3286ed | 790 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
791 | { |
792 | if (ns && ns->event != event) { | |
793 | ns->event = event; | |
794 | wake_up_interruptible(&ns->poll); | |
795 | } | |
796 | } | |
797 | ||
99b7db7b NP |
798 | /* |
799 | * vfsmount lock must be held for write | |
800 | */ | |
7bdb11de | 801 | static void unhash_mnt(struct mount *mnt) |
419148da | 802 | { |
0714a533 | 803 | mnt->mnt_parent = mnt; |
a73324da | 804 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 805 | list_del_init(&mnt->mnt_child); |
38129a13 | 806 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 807 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
808 | put_mountpoint(mnt->mnt_mp); |
809 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
810 | } |
811 | ||
7bdb11de EB |
812 | /* |
813 | * vfsmount lock must be held for write | |
814 | */ | |
815 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
816 | { | |
817 | old_path->dentry = mnt->mnt_mountpoint; | |
818 | old_path->mnt = &mnt->mnt_parent->mnt; | |
819 | unhash_mnt(mnt); | |
820 | } | |
821 | ||
99b7db7b NP |
822 | /* |
823 | * vfsmount lock must be held for write | |
824 | */ | |
84d17192 AV |
825 | void mnt_set_mountpoint(struct mount *mnt, |
826 | struct mountpoint *mp, | |
44d964d6 | 827 | struct mount *child_mnt) |
b90fa9ae | 828 | { |
84d17192 | 829 | mp->m_count++; |
3a2393d7 | 830 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 831 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 832 | child_mnt->mnt_parent = mnt; |
84d17192 | 833 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 834 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
835 | } |
836 | ||
99b7db7b NP |
837 | /* |
838 | * vfsmount lock must be held for write | |
839 | */ | |
84d17192 AV |
840 | static void attach_mnt(struct mount *mnt, |
841 | struct mount *parent, | |
842 | struct mountpoint *mp) | |
1da177e4 | 843 | { |
84d17192 | 844 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 845 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 846 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
847 | } |
848 | ||
12a5b529 AV |
849 | static void attach_shadowed(struct mount *mnt, |
850 | struct mount *parent, | |
851 | struct mount *shadows) | |
852 | { | |
853 | if (shadows) { | |
f6f99332 | 854 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
855 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
856 | } else { | |
857 | hlist_add_head_rcu(&mnt->mnt_hash, | |
858 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
859 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
860 | } | |
861 | } | |
862 | ||
b90fa9ae | 863 | /* |
99b7db7b | 864 | * vfsmount lock must be held for write |
b90fa9ae | 865 | */ |
1d6a32ac | 866 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 867 | { |
0714a533 | 868 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 869 | struct mount *m; |
b90fa9ae | 870 | LIST_HEAD(head); |
143c8c91 | 871 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 872 | |
0714a533 | 873 | BUG_ON(parent == mnt); |
b90fa9ae | 874 | |
1a4eeaf2 | 875 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 876 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 877 | m->mnt_ns = n; |
f03c6599 | 878 | |
b90fa9ae RP |
879 | list_splice(&head, n->list.prev); |
880 | ||
12a5b529 | 881 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 882 | touch_mnt_namespace(n); |
1da177e4 LT |
883 | } |
884 | ||
909b0a88 | 885 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 886 | { |
6b41d536 AV |
887 | struct list_head *next = p->mnt_mounts.next; |
888 | if (next == &p->mnt_mounts) { | |
1da177e4 | 889 | while (1) { |
909b0a88 | 890 | if (p == root) |
1da177e4 | 891 | return NULL; |
6b41d536 AV |
892 | next = p->mnt_child.next; |
893 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 894 | break; |
0714a533 | 895 | p = p->mnt_parent; |
1da177e4 LT |
896 | } |
897 | } | |
6b41d536 | 898 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
899 | } |
900 | ||
315fc83e | 901 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 902 | { |
6b41d536 AV |
903 | struct list_head *prev = p->mnt_mounts.prev; |
904 | while (prev != &p->mnt_mounts) { | |
905 | p = list_entry(prev, struct mount, mnt_child); | |
906 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
907 | } |
908 | return p; | |
909 | } | |
910 | ||
9d412a43 AV |
911 | struct vfsmount * |
912 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
913 | { | |
b105e270 | 914 | struct mount *mnt; |
9d412a43 AV |
915 | struct dentry *root; |
916 | ||
917 | if (!type) | |
918 | return ERR_PTR(-ENODEV); | |
919 | ||
920 | mnt = alloc_vfsmnt(name); | |
921 | if (!mnt) | |
922 | return ERR_PTR(-ENOMEM); | |
923 | ||
924 | if (flags & MS_KERNMOUNT) | |
b105e270 | 925 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
926 | |
927 | root = mount_fs(type, flags, name, data); | |
928 | if (IS_ERR(root)) { | |
8ffcb32e | 929 | mnt_free_id(mnt); |
9d412a43 AV |
930 | free_vfsmnt(mnt); |
931 | return ERR_CAST(root); | |
932 | } | |
933 | ||
b105e270 AV |
934 | mnt->mnt.mnt_root = root; |
935 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 936 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 937 | mnt->mnt_parent = mnt; |
719ea2fb | 938 | lock_mount_hash(); |
39f7c4db | 939 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 940 | unlock_mount_hash(); |
b105e270 | 941 | return &mnt->mnt; |
9d412a43 AV |
942 | } |
943 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
944 | ||
87129cc0 | 945 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 946 | int flag) |
1da177e4 | 947 | { |
87129cc0 | 948 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
949 | struct mount *mnt; |
950 | int err; | |
1da177e4 | 951 | |
be34d1a3 DH |
952 | mnt = alloc_vfsmnt(old->mnt_devname); |
953 | if (!mnt) | |
954 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 955 | |
7a472ef4 | 956 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
957 | mnt->mnt_group_id = 0; /* not a peer of original */ |
958 | else | |
959 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 960 | |
be34d1a3 DH |
961 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
962 | err = mnt_alloc_group_id(mnt); | |
963 | if (err) | |
964 | goto out_free; | |
1da177e4 | 965 | } |
be34d1a3 | 966 | |
f2ebb3a9 | 967 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 968 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
969 | if (flag & CL_UNPRIVILEGED) { |
970 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
971 | ||
972 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
973 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
974 | ||
975 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
976 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
977 | ||
978 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
979 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
980 | ||
981 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
982 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
983 | } | |
132c94e3 | 984 | |
5ff9d8a6 | 985 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
986 | if ((flag & CL_UNPRIVILEGED) && |
987 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
988 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
989 | ||
be34d1a3 DH |
990 | atomic_inc(&sb->s_active); |
991 | mnt->mnt.mnt_sb = sb; | |
992 | mnt->mnt.mnt_root = dget(root); | |
993 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
994 | mnt->mnt_parent = mnt; | |
719ea2fb | 995 | lock_mount_hash(); |
be34d1a3 | 996 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 997 | unlock_mount_hash(); |
be34d1a3 | 998 | |
7a472ef4 EB |
999 | if ((flag & CL_SLAVE) || |
1000 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1001 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1002 | mnt->mnt_master = old; | |
1003 | CLEAR_MNT_SHARED(mnt); | |
1004 | } else if (!(flag & CL_PRIVATE)) { | |
1005 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1006 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1007 | if (IS_MNT_SLAVE(old)) | |
1008 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1009 | mnt->mnt_master = old->mnt_master; | |
1010 | } | |
1011 | if (flag & CL_MAKE_SHARED) | |
1012 | set_mnt_shared(mnt); | |
1013 | ||
1014 | /* stick the duplicate mount on the same expiry list | |
1015 | * as the original if that was on one */ | |
1016 | if (flag & CL_EXPIRE) { | |
1017 | if (!list_empty(&old->mnt_expire)) | |
1018 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1019 | } | |
1020 | ||
cb338d06 | 1021 | return mnt; |
719f5d7f MS |
1022 | |
1023 | out_free: | |
8ffcb32e | 1024 | mnt_free_id(mnt); |
719f5d7f | 1025 | free_vfsmnt(mnt); |
be34d1a3 | 1026 | return ERR_PTR(err); |
1da177e4 LT |
1027 | } |
1028 | ||
9ea459e1 AV |
1029 | static void cleanup_mnt(struct mount *mnt) |
1030 | { | |
1031 | /* | |
1032 | * This probably indicates that somebody messed | |
1033 | * up a mnt_want/drop_write() pair. If this | |
1034 | * happens, the filesystem was probably unable | |
1035 | * to make r/w->r/o transitions. | |
1036 | */ | |
1037 | /* | |
1038 | * The locking used to deal with mnt_count decrement provides barriers, | |
1039 | * so mnt_get_writers() below is safe. | |
1040 | */ | |
1041 | WARN_ON(mnt_get_writers(mnt)); | |
1042 | if (unlikely(mnt->mnt_pins.first)) | |
1043 | mnt_pin_kill(mnt); | |
1044 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1045 | dput(mnt->mnt.mnt_root); | |
1046 | deactivate_super(mnt->mnt.mnt_sb); | |
1047 | mnt_free_id(mnt); | |
1048 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1049 | } | |
1050 | ||
1051 | static void __cleanup_mnt(struct rcu_head *head) | |
1052 | { | |
1053 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1054 | } | |
1055 | ||
1056 | static LLIST_HEAD(delayed_mntput_list); | |
1057 | static void delayed_mntput(struct work_struct *unused) | |
1058 | { | |
1059 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1060 | struct llist_node *next; | |
1061 | ||
1062 | for (; node; node = next) { | |
1063 | next = llist_next(node); | |
1064 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1065 | } | |
1066 | } | |
1067 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1068 | ||
900148dc | 1069 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1070 | { |
48a066e7 AV |
1071 | rcu_read_lock(); |
1072 | mnt_add_count(mnt, -1); | |
1073 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1074 | rcu_read_unlock(); | |
f03c6599 | 1075 | return; |
b3e19d92 | 1076 | } |
719ea2fb | 1077 | lock_mount_hash(); |
b3e19d92 | 1078 | if (mnt_get_count(mnt)) { |
48a066e7 | 1079 | rcu_read_unlock(); |
719ea2fb | 1080 | unlock_mount_hash(); |
99b7db7b NP |
1081 | return; |
1082 | } | |
48a066e7 AV |
1083 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1084 | rcu_read_unlock(); | |
1085 | unlock_mount_hash(); | |
1086 | return; | |
1087 | } | |
1088 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1089 | rcu_read_unlock(); | |
962830df | 1090 | |
39f7c4db | 1091 | list_del(&mnt->mnt_instance); |
719ea2fb | 1092 | unlock_mount_hash(); |
649a795a | 1093 | |
9ea459e1 AV |
1094 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1095 | struct task_struct *task = current; | |
1096 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1097 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1098 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1099 | return; | |
1100 | } | |
1101 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1102 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1103 | return; | |
1104 | } | |
1105 | cleanup_mnt(mnt); | |
b3e19d92 | 1106 | } |
b3e19d92 NP |
1107 | |
1108 | void mntput(struct vfsmount *mnt) | |
1109 | { | |
1110 | if (mnt) { | |
863d684f | 1111 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1112 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1113 | if (unlikely(m->mnt_expiry_mark)) |
1114 | m->mnt_expiry_mark = 0; | |
1115 | mntput_no_expire(m); | |
b3e19d92 NP |
1116 | } |
1117 | } | |
1118 | EXPORT_SYMBOL(mntput); | |
1119 | ||
1120 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1121 | { | |
1122 | if (mnt) | |
83adc753 | 1123 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1124 | return mnt; |
1125 | } | |
1126 | EXPORT_SYMBOL(mntget); | |
1127 | ||
3064c356 | 1128 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1129 | { |
3064c356 AV |
1130 | struct mount *p; |
1131 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1132 | if (IS_ERR(p)) | |
1133 | return ERR_CAST(p); | |
1134 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1135 | return &p->mnt; | |
7b7b1ace | 1136 | } |
1da177e4 | 1137 | |
b3b304a2 MS |
1138 | static inline void mangle(struct seq_file *m, const char *s) |
1139 | { | |
1140 | seq_escape(m, s, " \t\n\\"); | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * Simple .show_options callback for filesystems which don't want to | |
1145 | * implement more complex mount option showing. | |
1146 | * | |
1147 | * See also save_mount_options(). | |
1148 | */ | |
34c80b1d | 1149 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1150 | { |
2a32cebd AV |
1151 | const char *options; |
1152 | ||
1153 | rcu_read_lock(); | |
34c80b1d | 1154 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1155 | |
1156 | if (options != NULL && options[0]) { | |
1157 | seq_putc(m, ','); | |
1158 | mangle(m, options); | |
1159 | } | |
2a32cebd | 1160 | rcu_read_unlock(); |
b3b304a2 MS |
1161 | |
1162 | return 0; | |
1163 | } | |
1164 | EXPORT_SYMBOL(generic_show_options); | |
1165 | ||
1166 | /* | |
1167 | * If filesystem uses generic_show_options(), this function should be | |
1168 | * called from the fill_super() callback. | |
1169 | * | |
1170 | * The .remount_fs callback usually needs to be handled in a special | |
1171 | * way, to make sure, that previous options are not overwritten if the | |
1172 | * remount fails. | |
1173 | * | |
1174 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1175 | * reset all options to their default value, but changes only newly | |
1176 | * given options, then the displayed options will not reflect reality | |
1177 | * any more. | |
1178 | */ | |
1179 | void save_mount_options(struct super_block *sb, char *options) | |
1180 | { | |
2a32cebd AV |
1181 | BUG_ON(sb->s_options); |
1182 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1183 | } |
1184 | EXPORT_SYMBOL(save_mount_options); | |
1185 | ||
2a32cebd AV |
1186 | void replace_mount_options(struct super_block *sb, char *options) |
1187 | { | |
1188 | char *old = sb->s_options; | |
1189 | rcu_assign_pointer(sb->s_options, options); | |
1190 | if (old) { | |
1191 | synchronize_rcu(); | |
1192 | kfree(old); | |
1193 | } | |
1194 | } | |
1195 | EXPORT_SYMBOL(replace_mount_options); | |
1196 | ||
a1a2c409 | 1197 | #ifdef CONFIG_PROC_FS |
0226f492 | 1198 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1199 | static void *m_start(struct seq_file *m, loff_t *pos) |
1200 | { | |
6ce6e24e | 1201 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1202 | |
390c6843 | 1203 | down_read(&namespace_sem); |
c7999c36 AV |
1204 | if (p->cached_event == p->ns->event) { |
1205 | void *v = p->cached_mount; | |
1206 | if (*pos == p->cached_index) | |
1207 | return v; | |
1208 | if (*pos == p->cached_index + 1) { | |
1209 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1210 | return p->cached_mount = v; | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | p->cached_event = p->ns->event; | |
1215 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1216 | p->cached_index = *pos; | |
1217 | return p->cached_mount; | |
1da177e4 LT |
1218 | } |
1219 | ||
1220 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1221 | { | |
6ce6e24e | 1222 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1223 | |
c7999c36 AV |
1224 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1225 | p->cached_index = *pos; | |
1226 | return p->cached_mount; | |
1da177e4 LT |
1227 | } |
1228 | ||
1229 | static void m_stop(struct seq_file *m, void *v) | |
1230 | { | |
390c6843 | 1231 | up_read(&namespace_sem); |
1da177e4 LT |
1232 | } |
1233 | ||
0226f492 | 1234 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1235 | { |
6ce6e24e | 1236 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1237 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1238 | return p->show(m, &r->mnt); |
1da177e4 LT |
1239 | } |
1240 | ||
a1a2c409 | 1241 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1242 | .start = m_start, |
1243 | .next = m_next, | |
1244 | .stop = m_stop, | |
0226f492 | 1245 | .show = m_show, |
b4629fe2 | 1246 | }; |
a1a2c409 | 1247 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1248 | |
1da177e4 LT |
1249 | /** |
1250 | * may_umount_tree - check if a mount tree is busy | |
1251 | * @mnt: root of mount tree | |
1252 | * | |
1253 | * This is called to check if a tree of mounts has any | |
1254 | * open files, pwds, chroots or sub mounts that are | |
1255 | * busy. | |
1256 | */ | |
909b0a88 | 1257 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1258 | { |
909b0a88 | 1259 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1260 | int actual_refs = 0; |
1261 | int minimum_refs = 0; | |
315fc83e | 1262 | struct mount *p; |
909b0a88 | 1263 | BUG_ON(!m); |
1da177e4 | 1264 | |
b3e19d92 | 1265 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1266 | lock_mount_hash(); |
909b0a88 | 1267 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1268 | actual_refs += mnt_get_count(p); |
1da177e4 | 1269 | minimum_refs += 2; |
1da177e4 | 1270 | } |
719ea2fb | 1271 | unlock_mount_hash(); |
1da177e4 LT |
1272 | |
1273 | if (actual_refs > minimum_refs) | |
e3474a8e | 1274 | return 0; |
1da177e4 | 1275 | |
e3474a8e | 1276 | return 1; |
1da177e4 LT |
1277 | } |
1278 | ||
1279 | EXPORT_SYMBOL(may_umount_tree); | |
1280 | ||
1281 | /** | |
1282 | * may_umount - check if a mount point is busy | |
1283 | * @mnt: root of mount | |
1284 | * | |
1285 | * This is called to check if a mount point has any | |
1286 | * open files, pwds, chroots or sub mounts. If the | |
1287 | * mount has sub mounts this will return busy | |
1288 | * regardless of whether the sub mounts are busy. | |
1289 | * | |
1290 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1291 | * give false negatives. The main reason why it's here is that we need | |
1292 | * a non-destructive way to look for easily umountable filesystems. | |
1293 | */ | |
1294 | int may_umount(struct vfsmount *mnt) | |
1295 | { | |
e3474a8e | 1296 | int ret = 1; |
8ad08d8a | 1297 | down_read(&namespace_sem); |
719ea2fb | 1298 | lock_mount_hash(); |
1ab59738 | 1299 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1300 | ret = 0; |
719ea2fb | 1301 | unlock_mount_hash(); |
8ad08d8a | 1302 | up_read(&namespace_sem); |
a05964f3 | 1303 | return ret; |
1da177e4 LT |
1304 | } |
1305 | ||
1306 | EXPORT_SYMBOL(may_umount); | |
1307 | ||
38129a13 | 1308 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1309 | |
97216be0 | 1310 | static void namespace_unlock(void) |
70fbcdf4 | 1311 | { |
a3b3c562 | 1312 | struct hlist_head head; |
97216be0 | 1313 | |
a3b3c562 | 1314 | hlist_move_list(&unmounted, &head); |
97216be0 | 1315 | |
97216be0 AV |
1316 | up_write(&namespace_sem); |
1317 | ||
a3b3c562 EB |
1318 | if (likely(hlist_empty(&head))) |
1319 | return; | |
1320 | ||
48a066e7 AV |
1321 | synchronize_rcu(); |
1322 | ||
87b95ce0 | 1323 | group_pin_kill(&head); |
70fbcdf4 RP |
1324 | } |
1325 | ||
97216be0 | 1326 | static inline void namespace_lock(void) |
e3197d83 | 1327 | { |
97216be0 | 1328 | down_write(&namespace_sem); |
e3197d83 AV |
1329 | } |
1330 | ||
e819f152 EB |
1331 | enum umount_tree_flags { |
1332 | UMOUNT_SYNC = 1, | |
1333 | UMOUNT_PROPAGATE = 2, | |
1334 | }; | |
99b7db7b | 1335 | /* |
48a066e7 | 1336 | * mount_lock must be held |
99b7db7b NP |
1337 | * namespace_sem must be held for write |
1338 | */ | |
e819f152 | 1339 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1340 | { |
c003b26f | 1341 | LIST_HEAD(tmp_list); |
315fc83e | 1342 | struct mount *p; |
1da177e4 | 1343 | |
5d88457e EB |
1344 | if (how & UMOUNT_PROPAGATE) |
1345 | propagate_mount_unlock(mnt); | |
1346 | ||
c003b26f | 1347 | /* Gather the mounts to umount */ |
590ce4bc EB |
1348 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1349 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1350 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1351 | } |
1da177e4 | 1352 | |
411a938b | 1353 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1354 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1355 | list_del_init(&p->mnt_child); |
c003b26f | 1356 | } |
88b368f2 | 1357 | |
c003b26f | 1358 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1359 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1360 | propagate_umount(&tmp_list); |
a05964f3 | 1361 | |
c003b26f EB |
1362 | while (!list_empty(&tmp_list)) { |
1363 | p = list_first_entry(&tmp_list, struct mount, mnt_list); | |
6776db3d | 1364 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1365 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1366 | __touch_mnt_namespace(p->mnt_ns); |
1367 | p->mnt_ns = NULL; | |
e819f152 | 1368 | if (how & UMOUNT_SYNC) |
48a066e7 | 1369 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 AV |
1370 | |
1371 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, &unmounted); | |
676da58d | 1372 | if (mnt_has_parent(p)) { |
81b6b061 | 1373 | mnt_add_count(p->mnt_parent, -1); |
87b95ce0 AV |
1374 | /* old mountpoint will be dropped when we can do that */ |
1375 | p->mnt_ex_mountpoint = p->mnt_mountpoint; | |
7bdb11de | 1376 | unhash_mnt(p); |
7c4b93d8 | 1377 | } |
0f0afb1d | 1378 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1379 | } |
1380 | } | |
1381 | ||
b54b9be7 | 1382 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1383 | |
1ab59738 | 1384 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1385 | { |
1ab59738 | 1386 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1387 | int retval; |
1388 | ||
1ab59738 | 1389 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1390 | if (retval) |
1391 | return retval; | |
1392 | ||
1393 | /* | |
1394 | * Allow userspace to request a mountpoint be expired rather than | |
1395 | * unmounting unconditionally. Unmount only happens if: | |
1396 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1397 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1398 | */ | |
1399 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1400 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1401 | flags & (MNT_FORCE | MNT_DETACH)) |
1402 | return -EINVAL; | |
1403 | ||
b3e19d92 NP |
1404 | /* |
1405 | * probably don't strictly need the lock here if we examined | |
1406 | * all race cases, but it's a slowpath. | |
1407 | */ | |
719ea2fb | 1408 | lock_mount_hash(); |
83adc753 | 1409 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1410 | unlock_mount_hash(); |
1da177e4 | 1411 | return -EBUSY; |
b3e19d92 | 1412 | } |
719ea2fb | 1413 | unlock_mount_hash(); |
1da177e4 | 1414 | |
863d684f | 1415 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1416 | return -EAGAIN; |
1417 | } | |
1418 | ||
1419 | /* | |
1420 | * If we may have to abort operations to get out of this | |
1421 | * mount, and they will themselves hold resources we must | |
1422 | * allow the fs to do things. In the Unix tradition of | |
1423 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1424 | * might fail to complete on the first run through as other tasks | |
1425 | * must return, and the like. Thats for the mount program to worry | |
1426 | * about for the moment. | |
1427 | */ | |
1428 | ||
42faad99 | 1429 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1430 | sb->s_op->umount_begin(sb); |
42faad99 | 1431 | } |
1da177e4 LT |
1432 | |
1433 | /* | |
1434 | * No sense to grab the lock for this test, but test itself looks | |
1435 | * somewhat bogus. Suggestions for better replacement? | |
1436 | * Ho-hum... In principle, we might treat that as umount + switch | |
1437 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1438 | * Actually it makes sense, especially if rootfs would contain a | |
1439 | * /reboot - static binary that would close all descriptors and | |
1440 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1441 | */ | |
1ab59738 | 1442 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1443 | /* |
1444 | * Special case for "unmounting" root ... | |
1445 | * we just try to remount it readonly. | |
1446 | */ | |
a1480dcc AL |
1447 | if (!capable(CAP_SYS_ADMIN)) |
1448 | return -EPERM; | |
1da177e4 | 1449 | down_write(&sb->s_umount); |
4aa98cf7 | 1450 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1451 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1452 | up_write(&sb->s_umount); |
1453 | return retval; | |
1454 | } | |
1455 | ||
97216be0 | 1456 | namespace_lock(); |
719ea2fb | 1457 | lock_mount_hash(); |
5addc5dd | 1458 | event++; |
1da177e4 | 1459 | |
48a066e7 | 1460 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1461 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1462 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1463 | retval = 0; |
48a066e7 AV |
1464 | } else { |
1465 | shrink_submounts(mnt); | |
1466 | retval = -EBUSY; | |
1467 | if (!propagate_mount_busy(mnt, 2)) { | |
1468 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1469 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1470 | retval = 0; |
1471 | } | |
1da177e4 | 1472 | } |
719ea2fb | 1473 | unlock_mount_hash(); |
e3197d83 | 1474 | namespace_unlock(); |
1da177e4 LT |
1475 | return retval; |
1476 | } | |
1477 | ||
80b5dce8 EB |
1478 | /* |
1479 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1480 | * | |
1481 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1482 | * to an existing mountpoint, and wind up leaking the mount. | |
1483 | * detach_mounts allows lazily unmounting those mounts instead of | |
1484 | * leaking them. | |
1485 | * | |
1486 | * The caller may hold dentry->d_inode->i_mutex. | |
1487 | */ | |
1488 | void __detach_mounts(struct dentry *dentry) | |
1489 | { | |
1490 | struct mountpoint *mp; | |
1491 | struct mount *mnt; | |
1492 | ||
1493 | namespace_lock(); | |
1494 | mp = lookup_mountpoint(dentry); | |
1495 | if (!mp) | |
1496 | goto out_unlock; | |
1497 | ||
1498 | lock_mount_hash(); | |
1499 | while (!hlist_empty(&mp->m_list)) { | |
1500 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
8318e667 | 1501 | umount_tree(mnt, 0); |
80b5dce8 EB |
1502 | } |
1503 | unlock_mount_hash(); | |
1504 | put_mountpoint(mp); | |
1505 | out_unlock: | |
1506 | namespace_unlock(); | |
1507 | } | |
1508 | ||
9b40bc90 AV |
1509 | /* |
1510 | * Is the caller allowed to modify his namespace? | |
1511 | */ | |
1512 | static inline bool may_mount(void) | |
1513 | { | |
1514 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1515 | } | |
1516 | ||
1da177e4 LT |
1517 | /* |
1518 | * Now umount can handle mount points as well as block devices. | |
1519 | * This is important for filesystems which use unnamed block devices. | |
1520 | * | |
1521 | * We now support a flag for forced unmount like the other 'big iron' | |
1522 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1523 | */ | |
1524 | ||
bdc480e3 | 1525 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1526 | { |
2d8f3038 | 1527 | struct path path; |
900148dc | 1528 | struct mount *mnt; |
1da177e4 | 1529 | int retval; |
db1f05bb | 1530 | int lookup_flags = 0; |
1da177e4 | 1531 | |
db1f05bb MS |
1532 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1533 | return -EINVAL; | |
1534 | ||
9b40bc90 AV |
1535 | if (!may_mount()) |
1536 | return -EPERM; | |
1537 | ||
db1f05bb MS |
1538 | if (!(flags & UMOUNT_NOFOLLOW)) |
1539 | lookup_flags |= LOOKUP_FOLLOW; | |
1540 | ||
197df04c | 1541 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1542 | if (retval) |
1543 | goto out; | |
900148dc | 1544 | mnt = real_mount(path.mnt); |
1da177e4 | 1545 | retval = -EINVAL; |
2d8f3038 | 1546 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1547 | goto dput_and_out; |
143c8c91 | 1548 | if (!check_mnt(mnt)) |
1da177e4 | 1549 | goto dput_and_out; |
5ff9d8a6 EB |
1550 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1551 | goto dput_and_out; | |
b2f5d4dc EB |
1552 | retval = -EPERM; |
1553 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1554 | goto dput_and_out; | |
1da177e4 | 1555 | |
900148dc | 1556 | retval = do_umount(mnt, flags); |
1da177e4 | 1557 | dput_and_out: |
429731b1 | 1558 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1559 | dput(path.dentry); |
900148dc | 1560 | mntput_no_expire(mnt); |
1da177e4 LT |
1561 | out: |
1562 | return retval; | |
1563 | } | |
1564 | ||
1565 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1566 | ||
1567 | /* | |
b58fed8b | 1568 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1569 | */ |
bdc480e3 | 1570 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1571 | { |
b58fed8b | 1572 | return sys_umount(name, 0); |
1da177e4 LT |
1573 | } |
1574 | ||
1575 | #endif | |
1576 | ||
4ce5d2b1 | 1577 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1578 | { |
4ce5d2b1 | 1579 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1580 | return dentry->d_op == &ns_dentry_operations && |
1581 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1582 | } |
1583 | ||
58be2825 AV |
1584 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1585 | { | |
1586 | return container_of(ns, struct mnt_namespace, ns); | |
1587 | } | |
1588 | ||
4ce5d2b1 EB |
1589 | static bool mnt_ns_loop(struct dentry *dentry) |
1590 | { | |
1591 | /* Could bind mounting the mount namespace inode cause a | |
1592 | * mount namespace loop? | |
1593 | */ | |
1594 | struct mnt_namespace *mnt_ns; | |
1595 | if (!is_mnt_ns_file(dentry)) | |
1596 | return false; | |
1597 | ||
f77c8014 | 1598 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1599 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1600 | } | |
1601 | ||
87129cc0 | 1602 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1603 | int flag) |
1da177e4 | 1604 | { |
84d17192 | 1605 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1606 | |
4ce5d2b1 EB |
1607 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1608 | return ERR_PTR(-EINVAL); | |
1609 | ||
1610 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1611 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1612 | |
36341f64 | 1613 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1614 | if (IS_ERR(q)) |
1615 | return q; | |
1616 | ||
a73324da | 1617 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1618 | |
1619 | p = mnt; | |
6b41d536 | 1620 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1621 | struct mount *s; |
7ec02ef1 | 1622 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1623 | continue; |
1624 | ||
909b0a88 | 1625 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1626 | struct mount *t = NULL; |
4ce5d2b1 EB |
1627 | if (!(flag & CL_COPY_UNBINDABLE) && |
1628 | IS_MNT_UNBINDABLE(s)) { | |
1629 | s = skip_mnt_tree(s); | |
1630 | continue; | |
1631 | } | |
1632 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1633 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1634 | s = skip_mnt_tree(s); |
1635 | continue; | |
1636 | } | |
0714a533 AV |
1637 | while (p != s->mnt_parent) { |
1638 | p = p->mnt_parent; | |
1639 | q = q->mnt_parent; | |
1da177e4 | 1640 | } |
87129cc0 | 1641 | p = s; |
84d17192 | 1642 | parent = q; |
87129cc0 | 1643 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1644 | if (IS_ERR(q)) |
1645 | goto out; | |
719ea2fb | 1646 | lock_mount_hash(); |
1a4eeaf2 | 1647 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1648 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1649 | if (!list_empty(&parent->mnt_mounts)) { | |
1650 | t = list_last_entry(&parent->mnt_mounts, | |
1651 | struct mount, mnt_child); | |
1652 | if (t->mnt_mp != p->mnt_mp) | |
1653 | t = NULL; | |
1654 | } | |
1655 | attach_shadowed(q, parent, t); | |
719ea2fb | 1656 | unlock_mount_hash(); |
1da177e4 LT |
1657 | } |
1658 | } | |
1659 | return res; | |
be34d1a3 | 1660 | out: |
1da177e4 | 1661 | if (res) { |
719ea2fb | 1662 | lock_mount_hash(); |
e819f152 | 1663 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1664 | unlock_mount_hash(); |
1da177e4 | 1665 | } |
be34d1a3 | 1666 | return q; |
1da177e4 LT |
1667 | } |
1668 | ||
be34d1a3 DH |
1669 | /* Caller should check returned pointer for errors */ |
1670 | ||
589ff870 | 1671 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1672 | { |
cb338d06 | 1673 | struct mount *tree; |
97216be0 | 1674 | namespace_lock(); |
cd4a4017 EB |
1675 | if (!check_mnt(real_mount(path->mnt))) |
1676 | tree = ERR_PTR(-EINVAL); | |
1677 | else | |
1678 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1679 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1680 | namespace_unlock(); |
be34d1a3 | 1681 | if (IS_ERR(tree)) |
52e220d3 | 1682 | return ERR_CAST(tree); |
be34d1a3 | 1683 | return &tree->mnt; |
8aec0809 AV |
1684 | } |
1685 | ||
1686 | void drop_collected_mounts(struct vfsmount *mnt) | |
1687 | { | |
97216be0 | 1688 | namespace_lock(); |
719ea2fb | 1689 | lock_mount_hash(); |
e819f152 | 1690 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1691 | unlock_mount_hash(); |
3ab6abee | 1692 | namespace_unlock(); |
8aec0809 AV |
1693 | } |
1694 | ||
c771d683 MS |
1695 | /** |
1696 | * clone_private_mount - create a private clone of a path | |
1697 | * | |
1698 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1699 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1700 | * to the originating mount won't be propagated into this). | |
1701 | * | |
1702 | * Release with mntput(). | |
1703 | */ | |
1704 | struct vfsmount *clone_private_mount(struct path *path) | |
1705 | { | |
1706 | struct mount *old_mnt = real_mount(path->mnt); | |
1707 | struct mount *new_mnt; | |
1708 | ||
1709 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1710 | return ERR_PTR(-EINVAL); | |
1711 | ||
1712 | down_read(&namespace_sem); | |
1713 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); | |
1714 | up_read(&namespace_sem); | |
1715 | if (IS_ERR(new_mnt)) | |
1716 | return ERR_CAST(new_mnt); | |
1717 | ||
1718 | return &new_mnt->mnt; | |
1719 | } | |
1720 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1721 | ||
1f707137 AV |
1722 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1723 | struct vfsmount *root) | |
1724 | { | |
1a4eeaf2 | 1725 | struct mount *mnt; |
1f707137 AV |
1726 | int res = f(root, arg); |
1727 | if (res) | |
1728 | return res; | |
1a4eeaf2 AV |
1729 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1730 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1731 | if (res) |
1732 | return res; | |
1733 | } | |
1734 | return 0; | |
1735 | } | |
1736 | ||
4b8b21f4 | 1737 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1738 | { |
315fc83e | 1739 | struct mount *p; |
719f5d7f | 1740 | |
909b0a88 | 1741 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1742 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1743 | mnt_release_group_id(p); |
719f5d7f MS |
1744 | } |
1745 | } | |
1746 | ||
4b8b21f4 | 1747 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1748 | { |
315fc83e | 1749 | struct mount *p; |
719f5d7f | 1750 | |
909b0a88 | 1751 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1752 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1753 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1754 | if (err) { |
4b8b21f4 | 1755 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1756 | return err; |
1757 | } | |
1758 | } | |
1759 | } | |
1760 | ||
1761 | return 0; | |
1762 | } | |
1763 | ||
b90fa9ae RP |
1764 | /* |
1765 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1766 | * @nd : place the mount tree @source_mnt is attached |
1767 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1768 | * store the parent mount and mountpoint dentry. | |
1769 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1770 | * |
1771 | * NOTE: in the table below explains the semantics when a source mount | |
1772 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1773 | * --------------------------------------------------------------------------- |
1774 | * | BIND MOUNT OPERATION | | |
1775 | * |************************************************************************** | |
1776 | * | source-->| shared | private | slave | unbindable | | |
1777 | * | dest | | | | | | |
1778 | * | | | | | | | | |
1779 | * | v | | | | | | |
1780 | * |************************************************************************** | |
1781 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1782 | * | | | | | | | |
1783 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1784 | * *************************************************************************** | |
b90fa9ae RP |
1785 | * A bind operation clones the source mount and mounts the clone on the |
1786 | * destination mount. | |
1787 | * | |
1788 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1789 | * tree of the destination mount and the cloned mount is added to | |
1790 | * the peer group of the source mount. | |
1791 | * (+) the cloned mount is created under the destination mount and is marked | |
1792 | * as shared. The cloned mount is added to the peer group of the source | |
1793 | * mount. | |
5afe0022 RP |
1794 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1795 | * of the destination mount and the cloned mount is made slave | |
1796 | * of the same master as that of the source mount. The cloned mount | |
1797 | * is marked as 'shared and slave'. | |
1798 | * (*) the cloned mount is made a slave of the same master as that of the | |
1799 | * source mount. | |
1800 | * | |
9676f0c6 RP |
1801 | * --------------------------------------------------------------------------- |
1802 | * | MOVE MOUNT OPERATION | | |
1803 | * |************************************************************************** | |
1804 | * | source-->| shared | private | slave | unbindable | | |
1805 | * | dest | | | | | | |
1806 | * | | | | | | | | |
1807 | * | v | | | | | | |
1808 | * |************************************************************************** | |
1809 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1810 | * | | | | | | | |
1811 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1812 | * *************************************************************************** | |
5afe0022 RP |
1813 | * |
1814 | * (+) the mount is moved to the destination. And is then propagated to | |
1815 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1816 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1817 | * (+++) the mount is moved to the destination and is then propagated to |
1818 | * all the mounts belonging to the destination mount's propagation tree. | |
1819 | * the mount is marked as 'shared and slave'. | |
1820 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1821 | * |
1822 | * if the source mount is a tree, the operations explained above is | |
1823 | * applied to each mount in the tree. | |
1824 | * Must be called without spinlocks held, since this function can sleep | |
1825 | * in allocations. | |
1826 | */ | |
0fb54e50 | 1827 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1828 | struct mount *dest_mnt, |
1829 | struct mountpoint *dest_mp, | |
1830 | struct path *parent_path) | |
b90fa9ae | 1831 | { |
38129a13 | 1832 | HLIST_HEAD(tree_list); |
315fc83e | 1833 | struct mount *child, *p; |
38129a13 | 1834 | struct hlist_node *n; |
719f5d7f | 1835 | int err; |
b90fa9ae | 1836 | |
fc7be130 | 1837 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1838 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1839 | if (err) |
1840 | goto out; | |
0b1b901b | 1841 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1842 | lock_mount_hash(); |
0b1b901b AV |
1843 | if (err) |
1844 | goto out_cleanup_ids; | |
909b0a88 | 1845 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1846 | set_mnt_shared(p); |
0b1b901b AV |
1847 | } else { |
1848 | lock_mount_hash(); | |
b90fa9ae | 1849 | } |
1a390689 | 1850 | if (parent_path) { |
0fb54e50 | 1851 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1852 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1853 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1854 | } else { |
84d17192 | 1855 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1856 | commit_tree(source_mnt, NULL); |
21444403 | 1857 | } |
b90fa9ae | 1858 | |
38129a13 | 1859 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1860 | struct mount *q; |
38129a13 | 1861 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1862 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1863 | child->mnt_mountpoint); | |
1864 | commit_tree(child, q); | |
b90fa9ae | 1865 | } |
719ea2fb | 1866 | unlock_mount_hash(); |
99b7db7b | 1867 | |
b90fa9ae | 1868 | return 0; |
719f5d7f MS |
1869 | |
1870 | out_cleanup_ids: | |
f2ebb3a9 AV |
1871 | while (!hlist_empty(&tree_list)) { |
1872 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
e819f152 | 1873 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
1874 | } |
1875 | unlock_mount_hash(); | |
0b1b901b | 1876 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1877 | out: |
1878 | return err; | |
b90fa9ae RP |
1879 | } |
1880 | ||
84d17192 | 1881 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1882 | { |
1883 | struct vfsmount *mnt; | |
84d17192 | 1884 | struct dentry *dentry = path->dentry; |
b12cea91 | 1885 | retry: |
84d17192 AV |
1886 | mutex_lock(&dentry->d_inode->i_mutex); |
1887 | if (unlikely(cant_mount(dentry))) { | |
1888 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1889 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1890 | } |
97216be0 | 1891 | namespace_lock(); |
b12cea91 | 1892 | mnt = lookup_mnt(path); |
84d17192 | 1893 | if (likely(!mnt)) { |
e2dfa935 EB |
1894 | struct mountpoint *mp = lookup_mountpoint(dentry); |
1895 | if (!mp) | |
1896 | mp = new_mountpoint(dentry); | |
84d17192 | 1897 | if (IS_ERR(mp)) { |
97216be0 | 1898 | namespace_unlock(); |
84d17192 AV |
1899 | mutex_unlock(&dentry->d_inode->i_mutex); |
1900 | return mp; | |
1901 | } | |
1902 | return mp; | |
1903 | } | |
97216be0 | 1904 | namespace_unlock(); |
b12cea91 AV |
1905 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1906 | path_put(path); | |
1907 | path->mnt = mnt; | |
84d17192 | 1908 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1909 | goto retry; |
1910 | } | |
1911 | ||
84d17192 | 1912 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1913 | { |
84d17192 AV |
1914 | struct dentry *dentry = where->m_dentry; |
1915 | put_mountpoint(where); | |
328e6d90 | 1916 | namespace_unlock(); |
84d17192 | 1917 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1918 | } |
1919 | ||
84d17192 | 1920 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1921 | { |
95bc5f25 | 1922 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1923 | return -EINVAL; |
1924 | ||
e36cb0b8 DH |
1925 | if (d_is_dir(mp->m_dentry) != |
1926 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
1927 | return -ENOTDIR; |
1928 | ||
84d17192 | 1929 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1930 | } |
1931 | ||
7a2e8a8f VA |
1932 | /* |
1933 | * Sanity check the flags to change_mnt_propagation. | |
1934 | */ | |
1935 | ||
1936 | static int flags_to_propagation_type(int flags) | |
1937 | { | |
7c6e984d | 1938 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1939 | |
1940 | /* Fail if any non-propagation flags are set */ | |
1941 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1942 | return 0; | |
1943 | /* Only one propagation flag should be set */ | |
1944 | if (!is_power_of_2(type)) | |
1945 | return 0; | |
1946 | return type; | |
1947 | } | |
1948 | ||
07b20889 RP |
1949 | /* |
1950 | * recursively change the type of the mountpoint. | |
1951 | */ | |
0a0d8a46 | 1952 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1953 | { |
315fc83e | 1954 | struct mount *m; |
4b8b21f4 | 1955 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1956 | int recurse = flag & MS_REC; |
7a2e8a8f | 1957 | int type; |
719f5d7f | 1958 | int err = 0; |
07b20889 | 1959 | |
2d92ab3c | 1960 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1961 | return -EINVAL; |
1962 | ||
7a2e8a8f VA |
1963 | type = flags_to_propagation_type(flag); |
1964 | if (!type) | |
1965 | return -EINVAL; | |
1966 | ||
97216be0 | 1967 | namespace_lock(); |
719f5d7f MS |
1968 | if (type == MS_SHARED) { |
1969 | err = invent_group_ids(mnt, recurse); | |
1970 | if (err) | |
1971 | goto out_unlock; | |
1972 | } | |
1973 | ||
719ea2fb | 1974 | lock_mount_hash(); |
909b0a88 | 1975 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1976 | change_mnt_propagation(m, type); |
719ea2fb | 1977 | unlock_mount_hash(); |
719f5d7f MS |
1978 | |
1979 | out_unlock: | |
97216be0 | 1980 | namespace_unlock(); |
719f5d7f | 1981 | return err; |
07b20889 RP |
1982 | } |
1983 | ||
5ff9d8a6 EB |
1984 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1985 | { | |
1986 | struct mount *child; | |
1987 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1988 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1989 | continue; | |
1990 | ||
1991 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1992 | return true; | |
1993 | } | |
1994 | return false; | |
1995 | } | |
1996 | ||
1da177e4 LT |
1997 | /* |
1998 | * do loopback mount. | |
1999 | */ | |
808d4e3c | 2000 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2001 | int recurse) |
1da177e4 | 2002 | { |
2d92ab3c | 2003 | struct path old_path; |
84d17192 AV |
2004 | struct mount *mnt = NULL, *old, *parent; |
2005 | struct mountpoint *mp; | |
57eccb83 | 2006 | int err; |
1da177e4 LT |
2007 | if (!old_name || !*old_name) |
2008 | return -EINVAL; | |
815d405c | 2009 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2010 | if (err) |
2011 | return err; | |
2012 | ||
8823c079 | 2013 | err = -EINVAL; |
4ce5d2b1 | 2014 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
2015 | goto out; |
2016 | ||
84d17192 AV |
2017 | mp = lock_mount(path); |
2018 | err = PTR_ERR(mp); | |
2019 | if (IS_ERR(mp)) | |
b12cea91 AV |
2020 | goto out; |
2021 | ||
87129cc0 | 2022 | old = real_mount(old_path.mnt); |
84d17192 | 2023 | parent = real_mount(path->mnt); |
87129cc0 | 2024 | |
1da177e4 | 2025 | err = -EINVAL; |
fc7be130 | 2026 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2027 | goto out2; |
9676f0c6 | 2028 | |
e149ed2b AV |
2029 | if (!check_mnt(parent)) |
2030 | goto out2; | |
2031 | ||
2032 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2033 | goto out2; |
1da177e4 | 2034 | |
5ff9d8a6 EB |
2035 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2036 | goto out2; | |
2037 | ||
ccd48bc7 | 2038 | if (recurse) |
4ce5d2b1 | 2039 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2040 | else |
87129cc0 | 2041 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2042 | |
be34d1a3 DH |
2043 | if (IS_ERR(mnt)) { |
2044 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2045 | goto out2; |
be34d1a3 | 2046 | } |
ccd48bc7 | 2047 | |
5ff9d8a6 EB |
2048 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2049 | ||
84d17192 | 2050 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2051 | if (err) { |
719ea2fb | 2052 | lock_mount_hash(); |
e819f152 | 2053 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2054 | unlock_mount_hash(); |
5b83d2c5 | 2055 | } |
b12cea91 | 2056 | out2: |
84d17192 | 2057 | unlock_mount(mp); |
ccd48bc7 | 2058 | out: |
2d92ab3c | 2059 | path_put(&old_path); |
1da177e4 LT |
2060 | return err; |
2061 | } | |
2062 | ||
2e4b7fcd DH |
2063 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2064 | { | |
2065 | int error = 0; | |
2066 | int readonly_request = 0; | |
2067 | ||
2068 | if (ms_flags & MS_RDONLY) | |
2069 | readonly_request = 1; | |
2070 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2071 | return 0; | |
2072 | ||
2073 | if (readonly_request) | |
83adc753 | 2074 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2075 | else |
83adc753 | 2076 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2077 | return error; |
2078 | } | |
2079 | ||
1da177e4 LT |
2080 | /* |
2081 | * change filesystem flags. dir should be a physical root of filesystem. | |
2082 | * If you've mounted a non-root directory somewhere and want to do remount | |
2083 | * on it - tough luck. | |
2084 | */ | |
0a0d8a46 | 2085 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
2086 | void *data) |
2087 | { | |
2088 | int err; | |
2d92ab3c | 2089 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2090 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2091 | |
143c8c91 | 2092 | if (!check_mnt(mnt)) |
1da177e4 LT |
2093 | return -EINVAL; |
2094 | ||
2d92ab3c | 2095 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2096 | return -EINVAL; |
2097 | ||
07b64558 EB |
2098 | /* Don't allow changing of locked mnt flags. |
2099 | * | |
2100 | * No locks need to be held here while testing the various | |
2101 | * MNT_LOCK flags because those flags can never be cleared | |
2102 | * once they are set. | |
2103 | */ | |
2104 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2105 | !(mnt_flags & MNT_READONLY)) { | |
2106 | return -EPERM; | |
2107 | } | |
9566d674 EB |
2108 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2109 | !(mnt_flags & MNT_NODEV)) { | |
3e186641 EB |
2110 | /* Was the nodev implicitly added in mount? */ |
2111 | if ((mnt->mnt_ns->user_ns != &init_user_ns) && | |
2112 | !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2113 | mnt_flags |= MNT_NODEV; | |
2114 | } else { | |
2115 | return -EPERM; | |
2116 | } | |
9566d674 EB |
2117 | } |
2118 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2119 | !(mnt_flags & MNT_NOSUID)) { | |
2120 | return -EPERM; | |
2121 | } | |
2122 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2123 | !(mnt_flags & MNT_NOEXEC)) { | |
2124 | return -EPERM; | |
2125 | } | |
2126 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2127 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2128 | return -EPERM; | |
2129 | } | |
2130 | ||
ff36fe2c EP |
2131 | err = security_sb_remount(sb, data); |
2132 | if (err) | |
2133 | return err; | |
2134 | ||
1da177e4 | 2135 | down_write(&sb->s_umount); |
2e4b7fcd | 2136 | if (flags & MS_BIND) |
2d92ab3c | 2137 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
2138 | else if (!capable(CAP_SYS_ADMIN)) |
2139 | err = -EPERM; | |
4aa98cf7 | 2140 | else |
2e4b7fcd | 2141 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2142 | if (!err) { |
719ea2fb | 2143 | lock_mount_hash(); |
a6138db8 | 2144 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2145 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2146 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2147 | unlock_mount_hash(); |
0e55a7cc | 2148 | } |
6339dab8 | 2149 | up_write(&sb->s_umount); |
1da177e4 LT |
2150 | return err; |
2151 | } | |
2152 | ||
cbbe362c | 2153 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2154 | { |
315fc83e | 2155 | struct mount *p; |
909b0a88 | 2156 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2157 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2158 | return 1; |
2159 | } | |
2160 | return 0; | |
2161 | } | |
2162 | ||
808d4e3c | 2163 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2164 | { |
2d92ab3c | 2165 | struct path old_path, parent_path; |
676da58d | 2166 | struct mount *p; |
0fb54e50 | 2167 | struct mount *old; |
84d17192 | 2168 | struct mountpoint *mp; |
57eccb83 | 2169 | int err; |
1da177e4 LT |
2170 | if (!old_name || !*old_name) |
2171 | return -EINVAL; | |
2d92ab3c | 2172 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2173 | if (err) |
2174 | return err; | |
2175 | ||
84d17192 AV |
2176 | mp = lock_mount(path); |
2177 | err = PTR_ERR(mp); | |
2178 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2179 | goto out; |
2180 | ||
143c8c91 | 2181 | old = real_mount(old_path.mnt); |
fc7be130 | 2182 | p = real_mount(path->mnt); |
143c8c91 | 2183 | |
1da177e4 | 2184 | err = -EINVAL; |
fc7be130 | 2185 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2186 | goto out1; |
2187 | ||
5ff9d8a6 EB |
2188 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2189 | goto out1; | |
2190 | ||
1da177e4 | 2191 | err = -EINVAL; |
2d92ab3c | 2192 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2193 | goto out1; |
1da177e4 | 2194 | |
676da58d | 2195 | if (!mnt_has_parent(old)) |
21444403 | 2196 | goto out1; |
1da177e4 | 2197 | |
e36cb0b8 DH |
2198 | if (d_is_dir(path->dentry) != |
2199 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2200 | goto out1; |
2201 | /* | |
2202 | * Don't move a mount residing in a shared parent. | |
2203 | */ | |
fc7be130 | 2204 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2205 | goto out1; |
9676f0c6 RP |
2206 | /* |
2207 | * Don't move a mount tree containing unbindable mounts to a destination | |
2208 | * mount which is shared. | |
2209 | */ | |
fc7be130 | 2210 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2211 | goto out1; |
1da177e4 | 2212 | err = -ELOOP; |
fc7be130 | 2213 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2214 | if (p == old) |
21444403 | 2215 | goto out1; |
1da177e4 | 2216 | |
84d17192 | 2217 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2218 | if (err) |
21444403 | 2219 | goto out1; |
1da177e4 LT |
2220 | |
2221 | /* if the mount is moved, it should no longer be expire | |
2222 | * automatically */ | |
6776db3d | 2223 | list_del_init(&old->mnt_expire); |
1da177e4 | 2224 | out1: |
84d17192 | 2225 | unlock_mount(mp); |
1da177e4 | 2226 | out: |
1da177e4 | 2227 | if (!err) |
1a390689 | 2228 | path_put(&parent_path); |
2d92ab3c | 2229 | path_put(&old_path); |
1da177e4 LT |
2230 | return err; |
2231 | } | |
2232 | ||
9d412a43 AV |
2233 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2234 | { | |
2235 | int err; | |
2236 | const char *subtype = strchr(fstype, '.'); | |
2237 | if (subtype) { | |
2238 | subtype++; | |
2239 | err = -EINVAL; | |
2240 | if (!subtype[0]) | |
2241 | goto err; | |
2242 | } else | |
2243 | subtype = ""; | |
2244 | ||
2245 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2246 | err = -ENOMEM; | |
2247 | if (!mnt->mnt_sb->s_subtype) | |
2248 | goto err; | |
2249 | return mnt; | |
2250 | ||
2251 | err: | |
2252 | mntput(mnt); | |
2253 | return ERR_PTR(err); | |
2254 | } | |
2255 | ||
9d412a43 AV |
2256 | /* |
2257 | * add a mount into a namespace's mount tree | |
2258 | */ | |
95bc5f25 | 2259 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2260 | { |
84d17192 AV |
2261 | struct mountpoint *mp; |
2262 | struct mount *parent; | |
9d412a43 AV |
2263 | int err; |
2264 | ||
f2ebb3a9 | 2265 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2266 | |
84d17192 AV |
2267 | mp = lock_mount(path); |
2268 | if (IS_ERR(mp)) | |
2269 | return PTR_ERR(mp); | |
9d412a43 | 2270 | |
84d17192 | 2271 | parent = real_mount(path->mnt); |
9d412a43 | 2272 | err = -EINVAL; |
84d17192 | 2273 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2274 | /* that's acceptable only for automounts done in private ns */ |
2275 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2276 | goto unlock; | |
2277 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2278 | if (!parent->mnt_ns) |
156cacb1 AV |
2279 | goto unlock; |
2280 | } | |
9d412a43 AV |
2281 | |
2282 | /* Refuse the same filesystem on the same mount point */ | |
2283 | err = -EBUSY; | |
95bc5f25 | 2284 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2285 | path->mnt->mnt_root == path->dentry) |
2286 | goto unlock; | |
2287 | ||
2288 | err = -EINVAL; | |
e36cb0b8 | 2289 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2290 | goto unlock; |
2291 | ||
95bc5f25 | 2292 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2293 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2294 | |
2295 | unlock: | |
84d17192 | 2296 | unlock_mount(mp); |
9d412a43 AV |
2297 | return err; |
2298 | } | |
b1e75df4 | 2299 | |
1da177e4 LT |
2300 | /* |
2301 | * create a new mount for userspace and request it to be added into the | |
2302 | * namespace's tree | |
2303 | */ | |
0c55cfc4 | 2304 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2305 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2306 | { |
0c55cfc4 | 2307 | struct file_system_type *type; |
9b40bc90 | 2308 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2309 | struct vfsmount *mnt; |
15f9a3f3 | 2310 | int err; |
1da177e4 | 2311 | |
0c55cfc4 | 2312 | if (!fstype) |
1da177e4 LT |
2313 | return -EINVAL; |
2314 | ||
0c55cfc4 EB |
2315 | type = get_fs_type(fstype); |
2316 | if (!type) | |
2317 | return -ENODEV; | |
2318 | ||
2319 | if (user_ns != &init_user_ns) { | |
2320 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2321 | put_filesystem(type); | |
2322 | return -EPERM; | |
2323 | } | |
2324 | /* Only in special cases allow devices from mounts | |
2325 | * created outside the initial user namespace. | |
2326 | */ | |
2327 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2328 | flags |= MS_NODEV; | |
9566d674 | 2329 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 EB |
2330 | } |
2331 | } | |
2332 | ||
2333 | mnt = vfs_kern_mount(type, flags, name, data); | |
2334 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2335 | !mnt->mnt_sb->s_subtype) | |
2336 | mnt = fs_set_subtype(mnt, fstype); | |
2337 | ||
2338 | put_filesystem(type); | |
1da177e4 LT |
2339 | if (IS_ERR(mnt)) |
2340 | return PTR_ERR(mnt); | |
2341 | ||
95bc5f25 | 2342 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2343 | if (err) |
2344 | mntput(mnt); | |
2345 | return err; | |
1da177e4 LT |
2346 | } |
2347 | ||
19a167af AV |
2348 | int finish_automount(struct vfsmount *m, struct path *path) |
2349 | { | |
6776db3d | 2350 | struct mount *mnt = real_mount(m); |
19a167af AV |
2351 | int err; |
2352 | /* The new mount record should have at least 2 refs to prevent it being | |
2353 | * expired before we get a chance to add it | |
2354 | */ | |
6776db3d | 2355 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2356 | |
2357 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2358 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2359 | err = -ELOOP; |
2360 | goto fail; | |
19a167af AV |
2361 | } |
2362 | ||
95bc5f25 | 2363 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2364 | if (!err) |
2365 | return 0; | |
2366 | fail: | |
2367 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2368 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2369 | namespace_lock(); |
6776db3d | 2370 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2371 | namespace_unlock(); |
19a167af | 2372 | } |
b1e75df4 AV |
2373 | mntput(m); |
2374 | mntput(m); | |
19a167af AV |
2375 | return err; |
2376 | } | |
2377 | ||
ea5b778a DH |
2378 | /** |
2379 | * mnt_set_expiry - Put a mount on an expiration list | |
2380 | * @mnt: The mount to list. | |
2381 | * @expiry_list: The list to add the mount to. | |
2382 | */ | |
2383 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2384 | { | |
97216be0 | 2385 | namespace_lock(); |
ea5b778a | 2386 | |
6776db3d | 2387 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2388 | |
97216be0 | 2389 | namespace_unlock(); |
ea5b778a DH |
2390 | } |
2391 | EXPORT_SYMBOL(mnt_set_expiry); | |
2392 | ||
1da177e4 LT |
2393 | /* |
2394 | * process a list of expirable mountpoints with the intent of discarding any | |
2395 | * mountpoints that aren't in use and haven't been touched since last we came | |
2396 | * here | |
2397 | */ | |
2398 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2399 | { | |
761d5c38 | 2400 | struct mount *mnt, *next; |
1da177e4 LT |
2401 | LIST_HEAD(graveyard); |
2402 | ||
2403 | if (list_empty(mounts)) | |
2404 | return; | |
2405 | ||
97216be0 | 2406 | namespace_lock(); |
719ea2fb | 2407 | lock_mount_hash(); |
1da177e4 LT |
2408 | |
2409 | /* extract from the expiration list every vfsmount that matches the | |
2410 | * following criteria: | |
2411 | * - only referenced by its parent vfsmount | |
2412 | * - still marked for expiry (marked on the last call here; marks are | |
2413 | * cleared by mntput()) | |
2414 | */ | |
6776db3d | 2415 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2416 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2417 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2418 | continue; |
6776db3d | 2419 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2420 | } |
bcc5c7d2 | 2421 | while (!list_empty(&graveyard)) { |
6776db3d | 2422 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2423 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2424 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2425 | } |
719ea2fb | 2426 | unlock_mount_hash(); |
3ab6abee | 2427 | namespace_unlock(); |
5528f911 TM |
2428 | } |
2429 | ||
2430 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2431 | ||
2432 | /* | |
2433 | * Ripoff of 'select_parent()' | |
2434 | * | |
2435 | * search the list of submounts for a given mountpoint, and move any | |
2436 | * shrinkable submounts to the 'graveyard' list. | |
2437 | */ | |
692afc31 | 2438 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2439 | { |
692afc31 | 2440 | struct mount *this_parent = parent; |
5528f911 TM |
2441 | struct list_head *next; |
2442 | int found = 0; | |
2443 | ||
2444 | repeat: | |
6b41d536 | 2445 | next = this_parent->mnt_mounts.next; |
5528f911 | 2446 | resume: |
6b41d536 | 2447 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2448 | struct list_head *tmp = next; |
6b41d536 | 2449 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2450 | |
2451 | next = tmp->next; | |
692afc31 | 2452 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2453 | continue; |
5528f911 TM |
2454 | /* |
2455 | * Descend a level if the d_mounts list is non-empty. | |
2456 | */ | |
6b41d536 | 2457 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2458 | this_parent = mnt; |
2459 | goto repeat; | |
2460 | } | |
1da177e4 | 2461 | |
1ab59738 | 2462 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2463 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2464 | found++; |
2465 | } | |
1da177e4 | 2466 | } |
5528f911 TM |
2467 | /* |
2468 | * All done at this level ... ascend and resume the search | |
2469 | */ | |
2470 | if (this_parent != parent) { | |
6b41d536 | 2471 | next = this_parent->mnt_child.next; |
0714a533 | 2472 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2473 | goto resume; |
2474 | } | |
2475 | return found; | |
2476 | } | |
2477 | ||
2478 | /* | |
2479 | * process a list of expirable mountpoints with the intent of discarding any | |
2480 | * submounts of a specific parent mountpoint | |
99b7db7b | 2481 | * |
48a066e7 | 2482 | * mount_lock must be held for write |
5528f911 | 2483 | */ |
b54b9be7 | 2484 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2485 | { |
2486 | LIST_HEAD(graveyard); | |
761d5c38 | 2487 | struct mount *m; |
5528f911 | 2488 | |
5528f911 | 2489 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2490 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2491 | while (!list_empty(&graveyard)) { |
761d5c38 | 2492 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2493 | mnt_expire); |
143c8c91 | 2494 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2495 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2496 | } |
2497 | } | |
1da177e4 LT |
2498 | } |
2499 | ||
1da177e4 LT |
2500 | /* |
2501 | * Some copy_from_user() implementations do not return the exact number of | |
2502 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2503 | * Note that this function differs from copy_from_user() in that it will oops | |
2504 | * on bad values of `to', rather than returning a short copy. | |
2505 | */ | |
b58fed8b RP |
2506 | static long exact_copy_from_user(void *to, const void __user * from, |
2507 | unsigned long n) | |
1da177e4 LT |
2508 | { |
2509 | char *t = to; | |
2510 | const char __user *f = from; | |
2511 | char c; | |
2512 | ||
2513 | if (!access_ok(VERIFY_READ, from, n)) | |
2514 | return n; | |
2515 | ||
2516 | while (n) { | |
2517 | if (__get_user(c, f)) { | |
2518 | memset(t, 0, n); | |
2519 | break; | |
2520 | } | |
2521 | *t++ = c; | |
2522 | f++; | |
2523 | n--; | |
2524 | } | |
2525 | return n; | |
2526 | } | |
2527 | ||
b58fed8b | 2528 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2529 | { |
2530 | int i; | |
2531 | unsigned long page; | |
2532 | unsigned long size; | |
b58fed8b | 2533 | |
1da177e4 LT |
2534 | *where = 0; |
2535 | if (!data) | |
2536 | return 0; | |
2537 | ||
2538 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2539 | return -ENOMEM; | |
2540 | ||
2541 | /* We only care that *some* data at the address the user | |
2542 | * gave us is valid. Just in case, we'll zero | |
2543 | * the remainder of the page. | |
2544 | */ | |
2545 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2546 | size = TASK_SIZE - (unsigned long)data; | |
2547 | if (size > PAGE_SIZE) | |
2548 | size = PAGE_SIZE; | |
2549 | ||
2550 | i = size - exact_copy_from_user((void *)page, data, size); | |
2551 | if (!i) { | |
b58fed8b | 2552 | free_page(page); |
1da177e4 LT |
2553 | return -EFAULT; |
2554 | } | |
2555 | if (i != PAGE_SIZE) | |
2556 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2557 | *where = page; | |
2558 | return 0; | |
2559 | } | |
2560 | ||
b8850d1f | 2561 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2562 | { |
b8850d1f | 2563 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2564 | } |
2565 | ||
1da177e4 LT |
2566 | /* |
2567 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2568 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2569 | * | |
2570 | * data is a (void *) that can point to any structure up to | |
2571 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2572 | * information (or be NULL). | |
2573 | * | |
2574 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2575 | * When the flags word was introduced its top half was required | |
2576 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2577 | * Therefore, if this magic number is present, it carries no information | |
2578 | * and must be discarded. | |
2579 | */ | |
5e6123f3 | 2580 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2581 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2582 | { |
2d92ab3c | 2583 | struct path path; |
1da177e4 LT |
2584 | int retval = 0; |
2585 | int mnt_flags = 0; | |
2586 | ||
2587 | /* Discard magic */ | |
2588 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2589 | flags &= ~MS_MGC_MSK; | |
2590 | ||
2591 | /* Basic sanity checks */ | |
1da177e4 LT |
2592 | if (data_page) |
2593 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2594 | ||
a27ab9f2 | 2595 | /* ... and get the mountpoint */ |
5e6123f3 | 2596 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2597 | if (retval) |
2598 | return retval; | |
2599 | ||
2600 | retval = security_sb_mount(dev_name, &path, | |
2601 | type_page, flags, data_page); | |
0d5cadb8 AV |
2602 | if (!retval && !may_mount()) |
2603 | retval = -EPERM; | |
a27ab9f2 TH |
2604 | if (retval) |
2605 | goto dput_out; | |
2606 | ||
613cbe3d AK |
2607 | /* Default to relatime unless overriden */ |
2608 | if (!(flags & MS_NOATIME)) | |
2609 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2610 | |
1da177e4 LT |
2611 | /* Separate the per-mountpoint flags */ |
2612 | if (flags & MS_NOSUID) | |
2613 | mnt_flags |= MNT_NOSUID; | |
2614 | if (flags & MS_NODEV) | |
2615 | mnt_flags |= MNT_NODEV; | |
2616 | if (flags & MS_NOEXEC) | |
2617 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2618 | if (flags & MS_NOATIME) |
2619 | mnt_flags |= MNT_NOATIME; | |
2620 | if (flags & MS_NODIRATIME) | |
2621 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2622 | if (flags & MS_STRICTATIME) |
2623 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2624 | if (flags & MS_RDONLY) |
2625 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2626 | |
ffbc6f0e EB |
2627 | /* The default atime for remount is preservation */ |
2628 | if ((flags & MS_REMOUNT) && | |
2629 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2630 | MS_STRICTATIME)) == 0)) { | |
2631 | mnt_flags &= ~MNT_ATIME_MASK; | |
2632 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2633 | } | |
2634 | ||
7a4dec53 | 2635 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2636 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2637 | MS_STRICTATIME); | |
1da177e4 | 2638 | |
1da177e4 | 2639 | if (flags & MS_REMOUNT) |
2d92ab3c | 2640 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2641 | data_page); |
2642 | else if (flags & MS_BIND) | |
2d92ab3c | 2643 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2644 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2645 | retval = do_change_type(&path, flags); |
1da177e4 | 2646 | else if (flags & MS_MOVE) |
2d92ab3c | 2647 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2648 | else |
2d92ab3c | 2649 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2650 | dev_name, data_page); |
2651 | dput_out: | |
2d92ab3c | 2652 | path_put(&path); |
1da177e4 LT |
2653 | return retval; |
2654 | } | |
2655 | ||
771b1371 EB |
2656 | static void free_mnt_ns(struct mnt_namespace *ns) |
2657 | { | |
6344c433 | 2658 | ns_free_inum(&ns->ns); |
771b1371 EB |
2659 | put_user_ns(ns->user_ns); |
2660 | kfree(ns); | |
2661 | } | |
2662 | ||
8823c079 EB |
2663 | /* |
2664 | * Assign a sequence number so we can detect when we attempt to bind | |
2665 | * mount a reference to an older mount namespace into the current | |
2666 | * mount namespace, preventing reference counting loops. A 64bit | |
2667 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2668 | * is effectively never, so we can ignore the possibility. | |
2669 | */ | |
2670 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2671 | ||
771b1371 | 2672 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2673 | { |
2674 | struct mnt_namespace *new_ns; | |
98f842e6 | 2675 | int ret; |
cf8d2c11 TM |
2676 | |
2677 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2678 | if (!new_ns) | |
2679 | return ERR_PTR(-ENOMEM); | |
6344c433 | 2680 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2681 | if (ret) { |
2682 | kfree(new_ns); | |
2683 | return ERR_PTR(ret); | |
2684 | } | |
33c42940 | 2685 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2686 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2687 | atomic_set(&new_ns->count, 1); |
2688 | new_ns->root = NULL; | |
2689 | INIT_LIST_HEAD(&new_ns->list); | |
2690 | init_waitqueue_head(&new_ns->poll); | |
2691 | new_ns->event = 0; | |
771b1371 | 2692 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2693 | return new_ns; |
2694 | } | |
2695 | ||
9559f689 AV |
2696 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2697 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2698 | { |
6b3286ed | 2699 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2700 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2701 | struct mount *p, *q; |
9559f689 | 2702 | struct mount *old; |
cb338d06 | 2703 | struct mount *new; |
7a472ef4 | 2704 | int copy_flags; |
1da177e4 | 2705 | |
9559f689 AV |
2706 | BUG_ON(!ns); |
2707 | ||
2708 | if (likely(!(flags & CLONE_NEWNS))) { | |
2709 | get_mnt_ns(ns); | |
2710 | return ns; | |
2711 | } | |
2712 | ||
2713 | old = ns->root; | |
2714 | ||
771b1371 | 2715 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2716 | if (IS_ERR(new_ns)) |
2717 | return new_ns; | |
1da177e4 | 2718 | |
97216be0 | 2719 | namespace_lock(); |
1da177e4 | 2720 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2721 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2722 | if (user_ns != ns->user_ns) |
132c94e3 | 2723 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2724 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2725 | if (IS_ERR(new)) { |
328e6d90 | 2726 | namespace_unlock(); |
771b1371 | 2727 | free_mnt_ns(new_ns); |
be34d1a3 | 2728 | return ERR_CAST(new); |
1da177e4 | 2729 | } |
be08d6d2 | 2730 | new_ns->root = new; |
1a4eeaf2 | 2731 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2732 | |
2733 | /* | |
2734 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2735 | * as belonging to new namespace. We have already acquired a private | |
2736 | * fs_struct, so tsk->fs->lock is not needed. | |
2737 | */ | |
909b0a88 | 2738 | p = old; |
cb338d06 | 2739 | q = new; |
1da177e4 | 2740 | while (p) { |
143c8c91 | 2741 | q->mnt_ns = new_ns; |
9559f689 AV |
2742 | if (new_fs) { |
2743 | if (&p->mnt == new_fs->root.mnt) { | |
2744 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2745 | rootmnt = &p->mnt; |
1da177e4 | 2746 | } |
9559f689 AV |
2747 | if (&p->mnt == new_fs->pwd.mnt) { |
2748 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2749 | pwdmnt = &p->mnt; |
1da177e4 | 2750 | } |
1da177e4 | 2751 | } |
909b0a88 AV |
2752 | p = next_mnt(p, old); |
2753 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2754 | if (!q) |
2755 | break; | |
2756 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2757 | p = next_mnt(p, old); | |
1da177e4 | 2758 | } |
328e6d90 | 2759 | namespace_unlock(); |
1da177e4 | 2760 | |
1da177e4 | 2761 | if (rootmnt) |
f03c6599 | 2762 | mntput(rootmnt); |
1da177e4 | 2763 | if (pwdmnt) |
f03c6599 | 2764 | mntput(pwdmnt); |
1da177e4 | 2765 | |
741a2951 | 2766 | return new_ns; |
1da177e4 LT |
2767 | } |
2768 | ||
cf8d2c11 TM |
2769 | /** |
2770 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2771 | * @mnt: pointer to the new root filesystem mountpoint | |
2772 | */ | |
1a4eeaf2 | 2773 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2774 | { |
771b1371 | 2775 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2776 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2777 | struct mount *mnt = real_mount(m); |
2778 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2779 | new_ns->root = mnt; |
b1983cd8 | 2780 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2781 | } else { |
1a4eeaf2 | 2782 | mntput(m); |
cf8d2c11 TM |
2783 | } |
2784 | return new_ns; | |
2785 | } | |
cf8d2c11 | 2786 | |
ea441d11 AV |
2787 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2788 | { | |
2789 | struct mnt_namespace *ns; | |
d31da0f0 | 2790 | struct super_block *s; |
ea441d11 AV |
2791 | struct path path; |
2792 | int err; | |
2793 | ||
2794 | ns = create_mnt_ns(mnt); | |
2795 | if (IS_ERR(ns)) | |
2796 | return ERR_CAST(ns); | |
2797 | ||
2798 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2799 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2800 | ||
2801 | put_mnt_ns(ns); | |
2802 | ||
2803 | if (err) | |
2804 | return ERR_PTR(err); | |
2805 | ||
2806 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2807 | s = path.mnt->mnt_sb; |
2808 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2809 | mntput(path.mnt); |
2810 | /* lock the sucker */ | |
d31da0f0 | 2811 | down_write(&s->s_umount); |
ea441d11 AV |
2812 | /* ... and return the root of (sub)tree on it */ |
2813 | return path.dentry; | |
2814 | } | |
2815 | EXPORT_SYMBOL(mount_subtree); | |
2816 | ||
bdc480e3 HC |
2817 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2818 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2819 | { |
eca6f534 VN |
2820 | int ret; |
2821 | char *kernel_type; | |
eca6f534 | 2822 | char *kernel_dev; |
1da177e4 | 2823 | unsigned long data_page; |
1da177e4 | 2824 | |
b8850d1f TG |
2825 | kernel_type = copy_mount_string(type); |
2826 | ret = PTR_ERR(kernel_type); | |
2827 | if (IS_ERR(kernel_type)) | |
eca6f534 | 2828 | goto out_type; |
1da177e4 | 2829 | |
b8850d1f TG |
2830 | kernel_dev = copy_mount_string(dev_name); |
2831 | ret = PTR_ERR(kernel_dev); | |
2832 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 2833 | goto out_dev; |
1da177e4 | 2834 | |
eca6f534 VN |
2835 | ret = copy_mount_options(data, &data_page); |
2836 | if (ret < 0) | |
2837 | goto out_data; | |
1da177e4 | 2838 | |
5e6123f3 | 2839 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, |
eca6f534 | 2840 | (void *) data_page); |
1da177e4 | 2841 | |
eca6f534 VN |
2842 | free_page(data_page); |
2843 | out_data: | |
2844 | kfree(kernel_dev); | |
2845 | out_dev: | |
eca6f534 VN |
2846 | kfree(kernel_type); |
2847 | out_type: | |
2848 | return ret; | |
1da177e4 LT |
2849 | } |
2850 | ||
afac7cba AV |
2851 | /* |
2852 | * Return true if path is reachable from root | |
2853 | * | |
48a066e7 | 2854 | * namespace_sem or mount_lock is held |
afac7cba | 2855 | */ |
643822b4 | 2856 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2857 | const struct path *root) |
2858 | { | |
643822b4 | 2859 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2860 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2861 | mnt = mnt->mnt_parent; |
afac7cba | 2862 | } |
643822b4 | 2863 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2864 | } |
2865 | ||
2866 | int path_is_under(struct path *path1, struct path *path2) | |
2867 | { | |
2868 | int res; | |
48a066e7 | 2869 | read_seqlock_excl(&mount_lock); |
643822b4 | 2870 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2871 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2872 | return res; |
2873 | } | |
2874 | EXPORT_SYMBOL(path_is_under); | |
2875 | ||
1da177e4 LT |
2876 | /* |
2877 | * pivot_root Semantics: | |
2878 | * Moves the root file system of the current process to the directory put_old, | |
2879 | * makes new_root as the new root file system of the current process, and sets | |
2880 | * root/cwd of all processes which had them on the current root to new_root. | |
2881 | * | |
2882 | * Restrictions: | |
2883 | * The new_root and put_old must be directories, and must not be on the | |
2884 | * same file system as the current process root. The put_old must be | |
2885 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2886 | * pointed to by put_old must yield the same directory as new_root. No other | |
2887 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2888 | * | |
4a0d11fa NB |
2889 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2890 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2891 | * in this situation. | |
2892 | * | |
1da177e4 LT |
2893 | * Notes: |
2894 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2895 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2896 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2897 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2898 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2899 | * first. | |
2900 | */ | |
3480b257 HC |
2901 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2902 | const char __user *, put_old) | |
1da177e4 | 2903 | { |
2d8f3038 | 2904 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2905 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2906 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2907 | int error; |
2908 | ||
9b40bc90 | 2909 | if (!may_mount()) |
1da177e4 LT |
2910 | return -EPERM; |
2911 | ||
2d8f3038 | 2912 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2913 | if (error) |
2914 | goto out0; | |
1da177e4 | 2915 | |
2d8f3038 | 2916 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2917 | if (error) |
2918 | goto out1; | |
2919 | ||
2d8f3038 | 2920 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2921 | if (error) |
2922 | goto out2; | |
1da177e4 | 2923 | |
f7ad3c6b | 2924 | get_fs_root(current->fs, &root); |
84d17192 AV |
2925 | old_mp = lock_mount(&old); |
2926 | error = PTR_ERR(old_mp); | |
2927 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2928 | goto out3; |
2929 | ||
1da177e4 | 2930 | error = -EINVAL; |
419148da AV |
2931 | new_mnt = real_mount(new.mnt); |
2932 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2933 | old_mnt = real_mount(old.mnt); |
2934 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2935 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2936 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2937 | goto out4; |
143c8c91 | 2938 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2939 | goto out4; |
5ff9d8a6 EB |
2940 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2941 | goto out4; | |
1da177e4 | 2942 | error = -ENOENT; |
f3da392e | 2943 | if (d_unlinked(new.dentry)) |
b12cea91 | 2944 | goto out4; |
1da177e4 | 2945 | error = -EBUSY; |
84d17192 | 2946 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2947 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2948 | error = -EINVAL; |
8c3ee42e | 2949 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2950 | goto out4; /* not a mountpoint */ |
676da58d | 2951 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2952 | goto out4; /* not attached */ |
84d17192 | 2953 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2954 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2955 | goto out4; /* not a mountpoint */ |
676da58d | 2956 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2957 | goto out4; /* not attached */ |
4ac91378 | 2958 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2959 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2960 | goto out4; |
0d082601 EB |
2961 | /* make certain new is below the root */ |
2962 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
2963 | goto out4; | |
84d17192 | 2964 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2965 | lock_mount_hash(); |
419148da AV |
2966 | detach_mnt(new_mnt, &parent_path); |
2967 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2968 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2969 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2970 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2971 | } | |
4ac91378 | 2972 | /* mount old root on put_old */ |
84d17192 | 2973 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2974 | /* mount new_root on / */ |
84d17192 | 2975 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2976 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
2977 | /* A moved mount should not expire automatically */ |
2978 | list_del_init(&new_mnt->mnt_expire); | |
719ea2fb | 2979 | unlock_mount_hash(); |
2d8f3038 | 2980 | chroot_fs_refs(&root, &new); |
84d17192 | 2981 | put_mountpoint(root_mp); |
1da177e4 | 2982 | error = 0; |
b12cea91 | 2983 | out4: |
84d17192 | 2984 | unlock_mount(old_mp); |
b12cea91 AV |
2985 | if (!error) { |
2986 | path_put(&root_parent); | |
2987 | path_put(&parent_path); | |
2988 | } | |
2989 | out3: | |
8c3ee42e | 2990 | path_put(&root); |
b12cea91 | 2991 | out2: |
2d8f3038 | 2992 | path_put(&old); |
1da177e4 | 2993 | out1: |
2d8f3038 | 2994 | path_put(&new); |
1da177e4 | 2995 | out0: |
1da177e4 | 2996 | return error; |
1da177e4 LT |
2997 | } |
2998 | ||
2999 | static void __init init_mount_tree(void) | |
3000 | { | |
3001 | struct vfsmount *mnt; | |
6b3286ed | 3002 | struct mnt_namespace *ns; |
ac748a09 | 3003 | struct path root; |
0c55cfc4 | 3004 | struct file_system_type *type; |
1da177e4 | 3005 | |
0c55cfc4 EB |
3006 | type = get_fs_type("rootfs"); |
3007 | if (!type) | |
3008 | panic("Can't find rootfs type"); | |
3009 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3010 | put_filesystem(type); | |
1da177e4 LT |
3011 | if (IS_ERR(mnt)) |
3012 | panic("Can't create rootfs"); | |
b3e19d92 | 3013 | |
3b22edc5 TM |
3014 | ns = create_mnt_ns(mnt); |
3015 | if (IS_ERR(ns)) | |
1da177e4 | 3016 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3017 | |
3018 | init_task.nsproxy->mnt_ns = ns; | |
3019 | get_mnt_ns(ns); | |
3020 | ||
be08d6d2 AV |
3021 | root.mnt = mnt; |
3022 | root.dentry = mnt->mnt_root; | |
da362b09 | 3023 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3024 | |
3025 | set_fs_pwd(current->fs, &root); | |
3026 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3027 | } |
3028 | ||
74bf17cf | 3029 | void __init mnt_init(void) |
1da177e4 | 3030 | { |
13f14b4d | 3031 | unsigned u; |
15a67dd8 | 3032 | int err; |
1da177e4 | 3033 | |
7d6fec45 | 3034 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3035 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3036 | |
0818bf27 | 3037 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3038 | sizeof(struct hlist_head), |
0818bf27 AV |
3039 | mhash_entries, 19, |
3040 | 0, | |
3041 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3042 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3043 | sizeof(struct hlist_head), | |
3044 | mphash_entries, 19, | |
3045 | 0, | |
3046 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 3047 | |
84d17192 | 3048 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3049 | panic("Failed to allocate mount hash table\n"); |
3050 | ||
0818bf27 | 3051 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 3052 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
3053 | for (u = 0; u <= mp_hash_mask; u++) |
3054 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 3055 | |
4b93dc9b TH |
3056 | kernfs_init(); |
3057 | ||
15a67dd8 RD |
3058 | err = sysfs_init(); |
3059 | if (err) | |
3060 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3061 | __func__, err); |
00d26666 GKH |
3062 | fs_kobj = kobject_create_and_add("fs", NULL); |
3063 | if (!fs_kobj) | |
8e24eea7 | 3064 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3065 | init_rootfs(); |
3066 | init_mount_tree(); | |
3067 | } | |
3068 | ||
616511d0 | 3069 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3070 | { |
d498b25a | 3071 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3072 | return; |
7b00ed6f | 3073 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3074 | free_mnt_ns(ns); |
1da177e4 | 3075 | } |
9d412a43 AV |
3076 | |
3077 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3078 | { | |
423e0ab0 TC |
3079 | struct vfsmount *mnt; |
3080 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3081 | if (!IS_ERR(mnt)) { | |
3082 | /* | |
3083 | * it is a longterm mount, don't release mnt until | |
3084 | * we unmount before file sys is unregistered | |
3085 | */ | |
f7a99c5b | 3086 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3087 | } |
3088 | return mnt; | |
9d412a43 AV |
3089 | } |
3090 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3091 | |
3092 | void kern_unmount(struct vfsmount *mnt) | |
3093 | { | |
3094 | /* release long term mount so mount point can be released */ | |
3095 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3096 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3097 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3098 | mntput(mnt); |
3099 | } | |
3100 | } | |
3101 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3102 | |
3103 | bool our_mnt(struct vfsmount *mnt) | |
3104 | { | |
143c8c91 | 3105 | return check_mnt(real_mount(mnt)); |
02125a82 | 3106 | } |
8823c079 | 3107 | |
3151527e EB |
3108 | bool current_chrooted(void) |
3109 | { | |
3110 | /* Does the current process have a non-standard root */ | |
3111 | struct path ns_root; | |
3112 | struct path fs_root; | |
3113 | bool chrooted; | |
3114 | ||
3115 | /* Find the namespace root */ | |
3116 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3117 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3118 | path_get(&ns_root); | |
3119 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3120 | ; | |
3121 | ||
3122 | get_fs_root(current->fs, &fs_root); | |
3123 | ||
3124 | chrooted = !path_equal(&fs_root, &ns_root); | |
3125 | ||
3126 | path_put(&fs_root); | |
3127 | path_put(&ns_root); | |
3128 | ||
3129 | return chrooted; | |
3130 | } | |
3131 | ||
e51db735 | 3132 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
3133 | { |
3134 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
3135 | struct mount *mnt; | |
e51db735 | 3136 | bool visible = false; |
87a8ebd6 | 3137 | |
e51db735 EB |
3138 | if (unlikely(!ns)) |
3139 | return false; | |
3140 | ||
44bb4385 | 3141 | down_read(&namespace_sem); |
87a8ebd6 | 3142 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
3143 | struct mount *child; |
3144 | if (mnt->mnt.mnt_sb->s_type != type) | |
3145 | continue; | |
3146 | ||
3147 | /* This mount is not fully visible if there are any child mounts | |
3148 | * that cover anything except for empty directories. | |
3149 | */ | |
3150 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3151 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
3152 | if (!S_ISDIR(inode->i_mode)) | |
3153 | goto next; | |
41301ae7 | 3154 | if (inode->i_nlink > 2) |
e51db735 | 3155 | goto next; |
87a8ebd6 | 3156 | } |
e51db735 EB |
3157 | visible = true; |
3158 | goto found; | |
3159 | next: ; | |
87a8ebd6 | 3160 | } |
e51db735 | 3161 | found: |
44bb4385 | 3162 | up_read(&namespace_sem); |
e51db735 | 3163 | return visible; |
87a8ebd6 EB |
3164 | } |
3165 | ||
64964528 | 3166 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3167 | { |
58be2825 | 3168 | struct ns_common *ns = NULL; |
8823c079 EB |
3169 | struct nsproxy *nsproxy; |
3170 | ||
728dba3a EB |
3171 | task_lock(task); |
3172 | nsproxy = task->nsproxy; | |
8823c079 | 3173 | if (nsproxy) { |
58be2825 AV |
3174 | ns = &nsproxy->mnt_ns->ns; |
3175 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3176 | } |
728dba3a | 3177 | task_unlock(task); |
8823c079 EB |
3178 | |
3179 | return ns; | |
3180 | } | |
3181 | ||
64964528 | 3182 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3183 | { |
58be2825 | 3184 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3185 | } |
3186 | ||
64964528 | 3187 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3188 | { |
3189 | struct fs_struct *fs = current->fs; | |
58be2825 | 3190 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns); |
8823c079 EB |
3191 | struct path root; |
3192 | ||
0c55cfc4 | 3193 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3194 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3195 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3196 | return -EPERM; |
8823c079 EB |
3197 | |
3198 | if (fs->users != 1) | |
3199 | return -EINVAL; | |
3200 | ||
3201 | get_mnt_ns(mnt_ns); | |
3202 | put_mnt_ns(nsproxy->mnt_ns); | |
3203 | nsproxy->mnt_ns = mnt_ns; | |
3204 | ||
3205 | /* Find the root */ | |
3206 | root.mnt = &mnt_ns->root->mnt; | |
3207 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3208 | path_get(&root); | |
3209 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3210 | ; | |
3211 | ||
3212 | /* Update the pwd and root */ | |
3213 | set_fs_pwd(fs, &root); | |
3214 | set_fs_root(fs, &root); | |
3215 | ||
3216 | path_put(&root); | |
3217 | return 0; | |
3218 | } | |
3219 | ||
3220 | const struct proc_ns_operations mntns_operations = { | |
3221 | .name = "mnt", | |
3222 | .type = CLONE_NEWNS, | |
3223 | .get = mntns_get, | |
3224 | .put = mntns_put, | |
3225 | .install = mntns_install, | |
3226 | }; |