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> |
d10577a8 | 19 | #include <linux/acct.h> /* acct_auto_close_mnt */ |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
0818bf27 | 26 | #include <linux/bootmem.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
0818bf27 AV |
30 | static unsigned int m_hash_mask __read_mostly; |
31 | static unsigned int m_hash_shift __read_mostly; | |
32 | static unsigned int mp_hash_mask __read_mostly; | |
33 | static unsigned int mp_hash_shift __read_mostly; | |
34 | ||
35 | static __initdata unsigned long mhash_entries; | |
36 | static int __init set_mhash_entries(char *str) | |
37 | { | |
38 | if (!str) | |
39 | return 0; | |
40 | mhash_entries = simple_strtoul(str, &str, 0); | |
41 | return 1; | |
42 | } | |
43 | __setup("mhash_entries=", set_mhash_entries); | |
44 | ||
45 | static __initdata unsigned long mphash_entries; | |
46 | static int __init set_mphash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mphash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 54 | |
c7999c36 | 55 | static u64 event; |
73cd49ec | 56 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 57 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 58 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
59 | static int mnt_id_start = 0; |
60 | static int mnt_group_start = 1; | |
1da177e4 | 61 | |
38129a13 | 62 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 63 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 64 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 65 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 66 | |
f87fd4c2 | 67 | /* /sys/fs */ |
00d26666 GKH |
68 | struct kobject *fs_kobj; |
69 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 70 | |
99b7db7b NP |
71 | /* |
72 | * vfsmount lock may be taken for read to prevent changes to the | |
73 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
74 | * up the tree. | |
75 | * | |
76 | * It should be taken for write in all cases where the vfsmount | |
77 | * tree or hash is modified or when a vfsmount structure is modified. | |
78 | */ | |
48a066e7 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 80 | |
38129a13 | 81 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 82 | { |
b58fed8b RP |
83 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
84 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
85 | tmp = tmp + (tmp >> m_hash_shift); |
86 | return &mount_hashtable[tmp & m_hash_mask]; | |
87 | } | |
88 | ||
89 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
90 | { | |
91 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
92 | tmp = tmp + (tmp >> mp_hash_shift); | |
93 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
94 | } |
95 | ||
99b7db7b NP |
96 | /* |
97 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
98 | * serialize with freeing. | |
99 | */ | |
b105e270 | 100 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
101 | { |
102 | int res; | |
103 | ||
104 | retry: | |
105 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 106 | spin_lock(&mnt_id_lock); |
15169fe7 | 107 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 108 | if (!res) |
15169fe7 | 109 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 110 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
111 | if (res == -EAGAIN) |
112 | goto retry; | |
113 | ||
114 | return res; | |
115 | } | |
116 | ||
b105e270 | 117 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 118 | { |
15169fe7 | 119 | int id = mnt->mnt_id; |
99b7db7b | 120 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
121 | ida_remove(&mnt_id_ida, id); |
122 | if (mnt_id_start > id) | |
123 | mnt_id_start = id; | |
99b7db7b | 124 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
125 | } |
126 | ||
719f5d7f MS |
127 | /* |
128 | * Allocate a new peer group ID | |
129 | * | |
130 | * mnt_group_ida is protected by namespace_sem | |
131 | */ | |
4b8b21f4 | 132 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 133 | { |
f21f6220 AV |
134 | int res; |
135 | ||
719f5d7f MS |
136 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
137 | return -ENOMEM; | |
138 | ||
f21f6220 AV |
139 | res = ida_get_new_above(&mnt_group_ida, |
140 | mnt_group_start, | |
15169fe7 | 141 | &mnt->mnt_group_id); |
f21f6220 | 142 | if (!res) |
15169fe7 | 143 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
144 | |
145 | return res; | |
719f5d7f MS |
146 | } |
147 | ||
148 | /* | |
149 | * Release a peer group ID | |
150 | */ | |
4b8b21f4 | 151 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 152 | { |
15169fe7 | 153 | int id = mnt->mnt_group_id; |
f21f6220 AV |
154 | ida_remove(&mnt_group_ida, id); |
155 | if (mnt_group_start > id) | |
156 | mnt_group_start = id; | |
15169fe7 | 157 | mnt->mnt_group_id = 0; |
719f5d7f MS |
158 | } |
159 | ||
b3e19d92 NP |
160 | /* |
161 | * vfsmount lock must be held for read | |
162 | */ | |
83adc753 | 163 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
164 | { |
165 | #ifdef CONFIG_SMP | |
68e8a9fe | 166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
167 | #else |
168 | preempt_disable(); | |
68e8a9fe | 169 | mnt->mnt_count += n; |
b3e19d92 NP |
170 | preempt_enable(); |
171 | #endif | |
172 | } | |
173 | ||
b3e19d92 NP |
174 | /* |
175 | * vfsmount lock must be held for write | |
176 | */ | |
83adc753 | 177 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
178 | { |
179 | #ifdef CONFIG_SMP | |
f03c6599 | 180 | unsigned int count = 0; |
b3e19d92 NP |
181 | int cpu; |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
185 | } |
186 | ||
187 | return count; | |
188 | #else | |
68e8a9fe | 189 | return mnt->mnt_count; |
b3e19d92 NP |
190 | #endif |
191 | } | |
192 | ||
b105e270 | 193 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 194 | { |
c63181e6 AV |
195 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
196 | if (mnt) { | |
73cd49ec MS |
197 | int err; |
198 | ||
c63181e6 | 199 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
200 | if (err) |
201 | goto out_free_cache; | |
202 | ||
203 | if (name) { | |
c63181e6 AV |
204 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
205 | if (!mnt->mnt_devname) | |
88b38782 | 206 | goto out_free_id; |
73cd49ec MS |
207 | } |
208 | ||
b3e19d92 | 209 | #ifdef CONFIG_SMP |
c63181e6 AV |
210 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
211 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
212 | goto out_free_devname; |
213 | ||
c63181e6 | 214 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 215 | #else |
c63181e6 AV |
216 | mnt->mnt_count = 1; |
217 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
218 | #endif |
219 | ||
38129a13 | 220 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
221 | INIT_LIST_HEAD(&mnt->mnt_child); |
222 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
223 | INIT_LIST_HEAD(&mnt->mnt_list); | |
224 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
225 | INIT_LIST_HEAD(&mnt->mnt_share); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
227 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
228 | #ifdef CONFIG_FSNOTIFY |
229 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 230 | #endif |
1da177e4 | 231 | } |
c63181e6 | 232 | return mnt; |
88b38782 | 233 | |
d3ef3d73 | 234 | #ifdef CONFIG_SMP |
235 | out_free_devname: | |
c63181e6 | 236 | kfree(mnt->mnt_devname); |
d3ef3d73 | 237 | #endif |
88b38782 | 238 | out_free_id: |
c63181e6 | 239 | mnt_free_id(mnt); |
88b38782 | 240 | out_free_cache: |
c63181e6 | 241 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 242 | return NULL; |
1da177e4 LT |
243 | } |
244 | ||
3d733633 DH |
245 | /* |
246 | * Most r/o checks on a fs are for operations that take | |
247 | * discrete amounts of time, like a write() or unlink(). | |
248 | * We must keep track of when those operations start | |
249 | * (for permission checks) and when they end, so that | |
250 | * we can determine when writes are able to occur to | |
251 | * a filesystem. | |
252 | */ | |
253 | /* | |
254 | * __mnt_is_readonly: check whether a mount is read-only | |
255 | * @mnt: the mount to check for its write status | |
256 | * | |
257 | * This shouldn't be used directly ouside of the VFS. | |
258 | * It does not guarantee that the filesystem will stay | |
259 | * r/w, just that it is right *now*. This can not and | |
260 | * should not be used in place of IS_RDONLY(inode). | |
261 | * mnt_want/drop_write() will _keep_ the filesystem | |
262 | * r/w. | |
263 | */ | |
264 | int __mnt_is_readonly(struct vfsmount *mnt) | |
265 | { | |
2e4b7fcd DH |
266 | if (mnt->mnt_flags & MNT_READONLY) |
267 | return 1; | |
268 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
269 | return 1; | |
270 | return 0; | |
3d733633 DH |
271 | } |
272 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
273 | ||
83adc753 | 274 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 275 | { |
276 | #ifdef CONFIG_SMP | |
68e8a9fe | 277 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 278 | #else |
68e8a9fe | 279 | mnt->mnt_writers++; |
d3ef3d73 | 280 | #endif |
281 | } | |
3d733633 | 282 | |
83adc753 | 283 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 284 | { |
d3ef3d73 | 285 | #ifdef CONFIG_SMP |
68e8a9fe | 286 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 287 | #else |
68e8a9fe | 288 | mnt->mnt_writers--; |
d3ef3d73 | 289 | #endif |
3d733633 | 290 | } |
3d733633 | 291 | |
83adc753 | 292 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 293 | { |
d3ef3d73 | 294 | #ifdef CONFIG_SMP |
295 | unsigned int count = 0; | |
3d733633 | 296 | int cpu; |
3d733633 DH |
297 | |
298 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 299 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 300 | } |
3d733633 | 301 | |
d3ef3d73 | 302 | return count; |
303 | #else | |
304 | return mnt->mnt_writers; | |
305 | #endif | |
3d733633 DH |
306 | } |
307 | ||
4ed5e82f MS |
308 | static int mnt_is_readonly(struct vfsmount *mnt) |
309 | { | |
310 | if (mnt->mnt_sb->s_readonly_remount) | |
311 | return 1; | |
312 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
313 | smp_rmb(); | |
314 | return __mnt_is_readonly(mnt); | |
315 | } | |
316 | ||
8366025e | 317 | /* |
eb04c282 JK |
318 | * Most r/o & frozen checks on a fs are for operations that take discrete |
319 | * amounts of time, like a write() or unlink(). We must keep track of when | |
320 | * those operations start (for permission checks) and when they end, so that we | |
321 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
322 | */ |
323 | /** | |
eb04c282 | 324 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 325 | * @m: the mount on which to take a write |
8366025e | 326 | * |
eb04c282 JK |
327 | * This tells the low-level filesystem that a write is about to be performed to |
328 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
329 | * returning success. This operation does not protect against filesystem being | |
330 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
331 | * called. This is effectively a refcount. | |
8366025e | 332 | */ |
eb04c282 | 333 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 334 | { |
83adc753 | 335 | struct mount *mnt = real_mount(m); |
3d733633 | 336 | int ret = 0; |
3d733633 | 337 | |
d3ef3d73 | 338 | preempt_disable(); |
c6653a83 | 339 | mnt_inc_writers(mnt); |
d3ef3d73 | 340 | /* |
c6653a83 | 341 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 342 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
343 | * incremented count after it has set MNT_WRITE_HOLD. | |
344 | */ | |
345 | smp_mb(); | |
1e75529e | 346 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 347 | cpu_relax(); |
348 | /* | |
349 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
350 | * be set to match its requirements. So we must not load that until | |
351 | * MNT_WRITE_HOLD is cleared. | |
352 | */ | |
353 | smp_rmb(); | |
4ed5e82f | 354 | if (mnt_is_readonly(m)) { |
c6653a83 | 355 | mnt_dec_writers(mnt); |
3d733633 | 356 | ret = -EROFS; |
3d733633 | 357 | } |
d3ef3d73 | 358 | preempt_enable(); |
eb04c282 JK |
359 | |
360 | return ret; | |
361 | } | |
362 | ||
363 | /** | |
364 | * mnt_want_write - get write access to a mount | |
365 | * @m: the mount on which to take a write | |
366 | * | |
367 | * This tells the low-level filesystem that a write is about to be performed to | |
368 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
369 | * is not frozen) before returning success. When the write operation is | |
370 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
371 | */ | |
372 | int mnt_want_write(struct vfsmount *m) | |
373 | { | |
374 | int ret; | |
375 | ||
376 | sb_start_write(m->mnt_sb); | |
377 | ret = __mnt_want_write(m); | |
378 | if (ret) | |
379 | sb_end_write(m->mnt_sb); | |
3d733633 | 380 | return ret; |
8366025e DH |
381 | } |
382 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
383 | ||
96029c4e | 384 | /** |
385 | * mnt_clone_write - get write access to a mount | |
386 | * @mnt: the mount on which to take a write | |
387 | * | |
388 | * This is effectively like mnt_want_write, except | |
389 | * it must only be used to take an extra write reference | |
390 | * on a mountpoint that we already know has a write reference | |
391 | * on it. This allows some optimisation. | |
392 | * | |
393 | * After finished, mnt_drop_write must be called as usual to | |
394 | * drop the reference. | |
395 | */ | |
396 | int mnt_clone_write(struct vfsmount *mnt) | |
397 | { | |
398 | /* superblock may be r/o */ | |
399 | if (__mnt_is_readonly(mnt)) | |
400 | return -EROFS; | |
401 | preempt_disable(); | |
83adc753 | 402 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 403 | preempt_enable(); |
404 | return 0; | |
405 | } | |
406 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
407 | ||
408 | /** | |
eb04c282 | 409 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 410 | * @file: the file who's mount on which to take a write |
411 | * | |
eb04c282 | 412 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 413 | * do some optimisations if the file is open for write already |
414 | */ | |
eb04c282 | 415 | int __mnt_want_write_file(struct file *file) |
96029c4e | 416 | { |
83f936c7 | 417 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 418 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 419 | else |
420 | return mnt_clone_write(file->f_path.mnt); | |
421 | } | |
eb04c282 JK |
422 | |
423 | /** | |
424 | * mnt_want_write_file - get write access to a file's mount | |
425 | * @file: the file who's mount on which to take a write | |
426 | * | |
427 | * This is like mnt_want_write, but it takes a file and can | |
428 | * do some optimisations if the file is open for write already | |
429 | */ | |
430 | int mnt_want_write_file(struct file *file) | |
431 | { | |
432 | int ret; | |
433 | ||
434 | sb_start_write(file->f_path.mnt->mnt_sb); | |
435 | ret = __mnt_want_write_file(file); | |
436 | if (ret) | |
437 | sb_end_write(file->f_path.mnt->mnt_sb); | |
438 | return ret; | |
439 | } | |
96029c4e | 440 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
441 | ||
8366025e | 442 | /** |
eb04c282 | 443 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
444 | * @mnt: the mount on which to give up write access |
445 | * | |
446 | * Tells the low-level filesystem that we are done | |
447 | * performing writes to it. Must be matched with | |
eb04c282 | 448 | * __mnt_want_write() call above. |
8366025e | 449 | */ |
eb04c282 | 450 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 451 | { |
d3ef3d73 | 452 | preempt_disable(); |
83adc753 | 453 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 454 | preempt_enable(); |
8366025e | 455 | } |
eb04c282 JK |
456 | |
457 | /** | |
458 | * mnt_drop_write - give up write access to a mount | |
459 | * @mnt: the mount on which to give up write access | |
460 | * | |
461 | * Tells the low-level filesystem that we are done performing writes to it and | |
462 | * also allows filesystem to be frozen again. Must be matched with | |
463 | * mnt_want_write() call above. | |
464 | */ | |
465 | void mnt_drop_write(struct vfsmount *mnt) | |
466 | { | |
467 | __mnt_drop_write(mnt); | |
468 | sb_end_write(mnt->mnt_sb); | |
469 | } | |
8366025e DH |
470 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
471 | ||
eb04c282 JK |
472 | void __mnt_drop_write_file(struct file *file) |
473 | { | |
474 | __mnt_drop_write(file->f_path.mnt); | |
475 | } | |
476 | ||
2a79f17e AV |
477 | void mnt_drop_write_file(struct file *file) |
478 | { | |
479 | mnt_drop_write(file->f_path.mnt); | |
480 | } | |
481 | EXPORT_SYMBOL(mnt_drop_write_file); | |
482 | ||
83adc753 | 483 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 484 | { |
3d733633 DH |
485 | int ret = 0; |
486 | ||
719ea2fb | 487 | lock_mount_hash(); |
83adc753 | 488 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 489 | /* |
d3ef3d73 | 490 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
491 | * should be visible before we do. | |
3d733633 | 492 | */ |
d3ef3d73 | 493 | smp_mb(); |
494 | ||
3d733633 | 495 | /* |
d3ef3d73 | 496 | * With writers on hold, if this value is zero, then there are |
497 | * definitely no active writers (although held writers may subsequently | |
498 | * increment the count, they'll have to wait, and decrement it after | |
499 | * seeing MNT_READONLY). | |
500 | * | |
501 | * It is OK to have counter incremented on one CPU and decremented on | |
502 | * another: the sum will add up correctly. The danger would be when we | |
503 | * sum up each counter, if we read a counter before it is incremented, | |
504 | * but then read another CPU's count which it has been subsequently | |
505 | * decremented from -- we would see more decrements than we should. | |
506 | * MNT_WRITE_HOLD protects against this scenario, because | |
507 | * mnt_want_write first increments count, then smp_mb, then spins on | |
508 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
509 | * we're counting up here. | |
3d733633 | 510 | */ |
c6653a83 | 511 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 512 | ret = -EBUSY; |
513 | else | |
83adc753 | 514 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 515 | /* |
516 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
517 | * that become unheld will see MNT_READONLY. | |
518 | */ | |
519 | smp_wmb(); | |
83adc753 | 520 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 521 | unlock_mount_hash(); |
3d733633 | 522 | return ret; |
8366025e | 523 | } |
8366025e | 524 | |
83adc753 | 525 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 526 | { |
719ea2fb | 527 | lock_mount_hash(); |
83adc753 | 528 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 529 | unlock_mount_hash(); |
2e4b7fcd DH |
530 | } |
531 | ||
4ed5e82f MS |
532 | int sb_prepare_remount_readonly(struct super_block *sb) |
533 | { | |
534 | struct mount *mnt; | |
535 | int err = 0; | |
536 | ||
8e8b8796 MS |
537 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
538 | if (atomic_long_read(&sb->s_remove_count)) | |
539 | return -EBUSY; | |
540 | ||
719ea2fb | 541 | lock_mount_hash(); |
4ed5e82f MS |
542 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
543 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
544 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
545 | smp_mb(); | |
546 | if (mnt_get_writers(mnt) > 0) { | |
547 | err = -EBUSY; | |
548 | break; | |
549 | } | |
550 | } | |
551 | } | |
8e8b8796 MS |
552 | if (!err && atomic_long_read(&sb->s_remove_count)) |
553 | err = -EBUSY; | |
554 | ||
4ed5e82f MS |
555 | if (!err) { |
556 | sb->s_readonly_remount = 1; | |
557 | smp_wmb(); | |
558 | } | |
559 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
560 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
561 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
562 | } | |
719ea2fb | 563 | unlock_mount_hash(); |
4ed5e82f MS |
564 | |
565 | return err; | |
566 | } | |
567 | ||
b105e270 | 568 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 569 | { |
52ba1621 | 570 | kfree(mnt->mnt_devname); |
d3ef3d73 | 571 | #ifdef CONFIG_SMP |
68e8a9fe | 572 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 573 | #endif |
b105e270 | 574 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
575 | } |
576 | ||
8ffcb32e DH |
577 | static void delayed_free_vfsmnt(struct rcu_head *head) |
578 | { | |
579 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
580 | } | |
581 | ||
48a066e7 AV |
582 | /* call under rcu_read_lock */ |
583 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
584 | { | |
585 | struct mount *mnt; | |
586 | if (read_seqretry(&mount_lock, seq)) | |
587 | return false; | |
588 | if (bastard == NULL) | |
589 | return true; | |
590 | mnt = real_mount(bastard); | |
591 | mnt_add_count(mnt, 1); | |
592 | if (likely(!read_seqretry(&mount_lock, seq))) | |
593 | return true; | |
594 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
595 | mnt_add_count(mnt, -1); | |
596 | return false; | |
597 | } | |
598 | rcu_read_unlock(); | |
599 | mntput(bastard); | |
600 | rcu_read_lock(); | |
601 | return false; | |
602 | } | |
603 | ||
1da177e4 | 604 | /* |
474279dc | 605 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 606 | * call under rcu_read_lock() |
1da177e4 | 607 | */ |
474279dc | 608 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 609 | { |
38129a13 | 610 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
611 | struct mount *p; |
612 | ||
38129a13 | 613 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
614 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
615 | return p; | |
616 | return NULL; | |
617 | } | |
618 | ||
619 | /* | |
620 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 621 | * mount_lock must be held. |
474279dc AV |
622 | */ |
623 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
624 | { | |
38129a13 AV |
625 | struct mount *p, *res; |
626 | res = p = __lookup_mnt(mnt, dentry); | |
627 | if (!p) | |
628 | goto out; | |
629 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
630 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
631 | break; | |
632 | res = p; | |
633 | } | |
38129a13 | 634 | out: |
1d6a32ac | 635 | return res; |
1da177e4 LT |
636 | } |
637 | ||
a05964f3 | 638 | /* |
f015f126 DH |
639 | * lookup_mnt - Return the first child mount mounted at path |
640 | * | |
641 | * "First" means first mounted chronologically. If you create the | |
642 | * following mounts: | |
643 | * | |
644 | * mount /dev/sda1 /mnt | |
645 | * mount /dev/sda2 /mnt | |
646 | * mount /dev/sda3 /mnt | |
647 | * | |
648 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
649 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
650 | * /dev/sda2, then /dev/sda3, then NULL. | |
651 | * | |
652 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 653 | */ |
1c755af4 | 654 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 655 | { |
c7105365 | 656 | struct mount *child_mnt; |
48a066e7 AV |
657 | struct vfsmount *m; |
658 | unsigned seq; | |
99b7db7b | 659 | |
48a066e7 AV |
660 | rcu_read_lock(); |
661 | do { | |
662 | seq = read_seqbegin(&mount_lock); | |
663 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
664 | m = child_mnt ? &child_mnt->mnt : NULL; | |
665 | } while (!legitimize_mnt(m, seq)); | |
666 | rcu_read_unlock(); | |
667 | return m; | |
a05964f3 RP |
668 | } |
669 | ||
84d17192 AV |
670 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
671 | { | |
0818bf27 | 672 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 | 673 | struct mountpoint *mp; |
eed81007 | 674 | int ret; |
84d17192 | 675 | |
0818bf27 | 676 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
677 | if (mp->m_dentry == dentry) { |
678 | /* might be worth a WARN_ON() */ | |
679 | if (d_unlinked(dentry)) | |
680 | return ERR_PTR(-ENOENT); | |
681 | mp->m_count++; | |
682 | return mp; | |
683 | } | |
684 | } | |
685 | ||
686 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
687 | if (!mp) | |
688 | return ERR_PTR(-ENOMEM); | |
689 | ||
eed81007 MS |
690 | ret = d_set_mounted(dentry); |
691 | if (ret) { | |
84d17192 | 692 | kfree(mp); |
eed81007 | 693 | return ERR_PTR(ret); |
84d17192 | 694 | } |
eed81007 | 695 | |
84d17192 AV |
696 | mp->m_dentry = dentry; |
697 | mp->m_count = 1; | |
0818bf27 | 698 | hlist_add_head(&mp->m_hash, chain); |
84d17192 AV |
699 | return mp; |
700 | } | |
701 | ||
702 | static void put_mountpoint(struct mountpoint *mp) | |
703 | { | |
704 | if (!--mp->m_count) { | |
705 | struct dentry *dentry = mp->m_dentry; | |
706 | spin_lock(&dentry->d_lock); | |
707 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
708 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 709 | hlist_del(&mp->m_hash); |
84d17192 AV |
710 | kfree(mp); |
711 | } | |
712 | } | |
713 | ||
143c8c91 | 714 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 715 | { |
6b3286ed | 716 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
717 | } |
718 | ||
99b7db7b NP |
719 | /* |
720 | * vfsmount lock must be held for write | |
721 | */ | |
6b3286ed | 722 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
723 | { |
724 | if (ns) { | |
725 | ns->event = ++event; | |
726 | wake_up_interruptible(&ns->poll); | |
727 | } | |
728 | } | |
729 | ||
99b7db7b NP |
730 | /* |
731 | * vfsmount lock must be held for write | |
732 | */ | |
6b3286ed | 733 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
734 | { |
735 | if (ns && ns->event != event) { | |
736 | ns->event = event; | |
737 | wake_up_interruptible(&ns->poll); | |
738 | } | |
739 | } | |
740 | ||
99b7db7b NP |
741 | /* |
742 | * vfsmount lock must be held for write | |
743 | */ | |
419148da AV |
744 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
745 | { | |
a73324da | 746 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
747 | old_path->mnt = &mnt->mnt_parent->mnt; |
748 | mnt->mnt_parent = mnt; | |
a73324da | 749 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 750 | list_del_init(&mnt->mnt_child); |
38129a13 | 751 | hlist_del_init_rcu(&mnt->mnt_hash); |
84d17192 AV |
752 | put_mountpoint(mnt->mnt_mp); |
753 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
754 | } |
755 | ||
99b7db7b NP |
756 | /* |
757 | * vfsmount lock must be held for write | |
758 | */ | |
84d17192 AV |
759 | void mnt_set_mountpoint(struct mount *mnt, |
760 | struct mountpoint *mp, | |
44d964d6 | 761 | struct mount *child_mnt) |
b90fa9ae | 762 | { |
84d17192 | 763 | mp->m_count++; |
3a2393d7 | 764 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 765 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 766 | child_mnt->mnt_parent = mnt; |
84d17192 | 767 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
768 | } |
769 | ||
99b7db7b NP |
770 | /* |
771 | * vfsmount lock must be held for write | |
772 | */ | |
84d17192 AV |
773 | static void attach_mnt(struct mount *mnt, |
774 | struct mount *parent, | |
775 | struct mountpoint *mp) | |
1da177e4 | 776 | { |
84d17192 | 777 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 778 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 779 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
780 | } |
781 | ||
782 | /* | |
99b7db7b | 783 | * vfsmount lock must be held for write |
b90fa9ae | 784 | */ |
1d6a32ac | 785 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 786 | { |
0714a533 | 787 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 788 | struct mount *m; |
b90fa9ae | 789 | LIST_HEAD(head); |
143c8c91 | 790 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 791 | |
0714a533 | 792 | BUG_ON(parent == mnt); |
b90fa9ae | 793 | |
1a4eeaf2 | 794 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 795 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 796 | m->mnt_ns = n; |
f03c6599 | 797 | |
b90fa9ae RP |
798 | list_splice(&head, n->list.prev); |
799 | ||
1d6a32ac | 800 | if (shadows) |
38129a13 | 801 | hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash); |
1d6a32ac | 802 | else |
38129a13 | 803 | hlist_add_head_rcu(&mnt->mnt_hash, |
0818bf27 | 804 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 805 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 806 | touch_mnt_namespace(n); |
1da177e4 LT |
807 | } |
808 | ||
909b0a88 | 809 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 810 | { |
6b41d536 AV |
811 | struct list_head *next = p->mnt_mounts.next; |
812 | if (next == &p->mnt_mounts) { | |
1da177e4 | 813 | while (1) { |
909b0a88 | 814 | if (p == root) |
1da177e4 | 815 | return NULL; |
6b41d536 AV |
816 | next = p->mnt_child.next; |
817 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 818 | break; |
0714a533 | 819 | p = p->mnt_parent; |
1da177e4 LT |
820 | } |
821 | } | |
6b41d536 | 822 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
823 | } |
824 | ||
315fc83e | 825 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 826 | { |
6b41d536 AV |
827 | struct list_head *prev = p->mnt_mounts.prev; |
828 | while (prev != &p->mnt_mounts) { | |
829 | p = list_entry(prev, struct mount, mnt_child); | |
830 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
831 | } |
832 | return p; | |
833 | } | |
834 | ||
9d412a43 AV |
835 | struct vfsmount * |
836 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
837 | { | |
b105e270 | 838 | struct mount *mnt; |
9d412a43 AV |
839 | struct dentry *root; |
840 | ||
841 | if (!type) | |
842 | return ERR_PTR(-ENODEV); | |
843 | ||
844 | mnt = alloc_vfsmnt(name); | |
845 | if (!mnt) | |
846 | return ERR_PTR(-ENOMEM); | |
847 | ||
848 | if (flags & MS_KERNMOUNT) | |
b105e270 | 849 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
850 | |
851 | root = mount_fs(type, flags, name, data); | |
852 | if (IS_ERR(root)) { | |
8ffcb32e | 853 | mnt_free_id(mnt); |
9d412a43 AV |
854 | free_vfsmnt(mnt); |
855 | return ERR_CAST(root); | |
856 | } | |
857 | ||
b105e270 AV |
858 | mnt->mnt.mnt_root = root; |
859 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 860 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 861 | mnt->mnt_parent = mnt; |
719ea2fb | 862 | lock_mount_hash(); |
39f7c4db | 863 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 864 | unlock_mount_hash(); |
b105e270 | 865 | return &mnt->mnt; |
9d412a43 AV |
866 | } |
867 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
868 | ||
87129cc0 | 869 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 870 | int flag) |
1da177e4 | 871 | { |
87129cc0 | 872 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
873 | struct mount *mnt; |
874 | int err; | |
1da177e4 | 875 | |
be34d1a3 DH |
876 | mnt = alloc_vfsmnt(old->mnt_devname); |
877 | if (!mnt) | |
878 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 879 | |
7a472ef4 | 880 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
881 | mnt->mnt_group_id = 0; /* not a peer of original */ |
882 | else | |
883 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 884 | |
be34d1a3 DH |
885 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
886 | err = mnt_alloc_group_id(mnt); | |
887 | if (err) | |
888 | goto out_free; | |
1da177e4 | 889 | } |
be34d1a3 | 890 | |
f2ebb3a9 | 891 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 892 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
893 | if (flag & CL_UNPRIVILEGED) { |
894 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
895 | ||
896 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
897 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
898 | ||
899 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
900 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
901 | ||
902 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
903 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
904 | ||
905 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
906 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
907 | } | |
132c94e3 | 908 | |
5ff9d8a6 EB |
909 | /* Don't allow unprivileged users to reveal what is under a mount */ |
910 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
911 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
912 | ||
be34d1a3 DH |
913 | atomic_inc(&sb->s_active); |
914 | mnt->mnt.mnt_sb = sb; | |
915 | mnt->mnt.mnt_root = dget(root); | |
916 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
917 | mnt->mnt_parent = mnt; | |
719ea2fb | 918 | lock_mount_hash(); |
be34d1a3 | 919 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 920 | unlock_mount_hash(); |
be34d1a3 | 921 | |
7a472ef4 EB |
922 | if ((flag & CL_SLAVE) || |
923 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
924 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
925 | mnt->mnt_master = old; | |
926 | CLEAR_MNT_SHARED(mnt); | |
927 | } else if (!(flag & CL_PRIVATE)) { | |
928 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
929 | list_add(&mnt->mnt_share, &old->mnt_share); | |
930 | if (IS_MNT_SLAVE(old)) | |
931 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
932 | mnt->mnt_master = old->mnt_master; | |
933 | } | |
934 | if (flag & CL_MAKE_SHARED) | |
935 | set_mnt_shared(mnt); | |
936 | ||
937 | /* stick the duplicate mount on the same expiry list | |
938 | * as the original if that was on one */ | |
939 | if (flag & CL_EXPIRE) { | |
940 | if (!list_empty(&old->mnt_expire)) | |
941 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
942 | } | |
943 | ||
cb338d06 | 944 | return mnt; |
719f5d7f MS |
945 | |
946 | out_free: | |
8ffcb32e | 947 | mnt_free_id(mnt); |
719f5d7f | 948 | free_vfsmnt(mnt); |
be34d1a3 | 949 | return ERR_PTR(err); |
1da177e4 LT |
950 | } |
951 | ||
900148dc | 952 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 953 | { |
b3e19d92 | 954 | put_again: |
48a066e7 AV |
955 | rcu_read_lock(); |
956 | mnt_add_count(mnt, -1); | |
957 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
958 | rcu_read_unlock(); | |
f03c6599 | 959 | return; |
b3e19d92 | 960 | } |
719ea2fb | 961 | lock_mount_hash(); |
b3e19d92 | 962 | if (mnt_get_count(mnt)) { |
48a066e7 | 963 | rcu_read_unlock(); |
719ea2fb | 964 | unlock_mount_hash(); |
99b7db7b NP |
965 | return; |
966 | } | |
863d684f AV |
967 | if (unlikely(mnt->mnt_pinned)) { |
968 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
969 | mnt->mnt_pinned = 0; | |
48a066e7 | 970 | rcu_read_unlock(); |
719ea2fb | 971 | unlock_mount_hash(); |
900148dc | 972 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 973 | goto put_again; |
7b7b1ace | 974 | } |
48a066e7 AV |
975 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
976 | rcu_read_unlock(); | |
977 | unlock_mount_hash(); | |
978 | return; | |
979 | } | |
980 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
981 | rcu_read_unlock(); | |
962830df | 982 | |
39f7c4db | 983 | list_del(&mnt->mnt_instance); |
719ea2fb | 984 | unlock_mount_hash(); |
649a795a AV |
985 | |
986 | /* | |
987 | * This probably indicates that somebody messed | |
988 | * up a mnt_want/drop_write() pair. If this | |
989 | * happens, the filesystem was probably unable | |
990 | * to make r/w->r/o transitions. | |
991 | */ | |
992 | /* | |
993 | * The locking used to deal with mnt_count decrement provides barriers, | |
994 | * so mnt_get_writers() below is safe. | |
995 | */ | |
996 | WARN_ON(mnt_get_writers(mnt)); | |
997 | fsnotify_vfsmount_delete(&mnt->mnt); | |
998 | dput(mnt->mnt.mnt_root); | |
999 | deactivate_super(mnt->mnt.mnt_sb); | |
48a066e7 | 1000 | mnt_free_id(mnt); |
8ffcb32e | 1001 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); |
b3e19d92 | 1002 | } |
b3e19d92 NP |
1003 | |
1004 | void mntput(struct vfsmount *mnt) | |
1005 | { | |
1006 | if (mnt) { | |
863d684f | 1007 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1008 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1009 | if (unlikely(m->mnt_expiry_mark)) |
1010 | m->mnt_expiry_mark = 0; | |
1011 | mntput_no_expire(m); | |
b3e19d92 NP |
1012 | } |
1013 | } | |
1014 | EXPORT_SYMBOL(mntput); | |
1015 | ||
1016 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1017 | { | |
1018 | if (mnt) | |
83adc753 | 1019 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1020 | return mnt; |
1021 | } | |
1022 | EXPORT_SYMBOL(mntget); | |
1023 | ||
7b7b1ace AV |
1024 | void mnt_pin(struct vfsmount *mnt) |
1025 | { | |
719ea2fb | 1026 | lock_mount_hash(); |
863d684f | 1027 | real_mount(mnt)->mnt_pinned++; |
719ea2fb | 1028 | unlock_mount_hash(); |
7b7b1ace | 1029 | } |
7b7b1ace AV |
1030 | EXPORT_SYMBOL(mnt_pin); |
1031 | ||
863d684f | 1032 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 1033 | { |
863d684f | 1034 | struct mount *mnt = real_mount(m); |
719ea2fb | 1035 | lock_mount_hash(); |
7b7b1ace | 1036 | if (mnt->mnt_pinned) { |
863d684f | 1037 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
1038 | mnt->mnt_pinned--; |
1039 | } | |
719ea2fb | 1040 | unlock_mount_hash(); |
7b7b1ace | 1041 | } |
7b7b1ace | 1042 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 1043 | |
b3b304a2 MS |
1044 | static inline void mangle(struct seq_file *m, const char *s) |
1045 | { | |
1046 | seq_escape(m, s, " \t\n\\"); | |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Simple .show_options callback for filesystems which don't want to | |
1051 | * implement more complex mount option showing. | |
1052 | * | |
1053 | * See also save_mount_options(). | |
1054 | */ | |
34c80b1d | 1055 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1056 | { |
2a32cebd AV |
1057 | const char *options; |
1058 | ||
1059 | rcu_read_lock(); | |
34c80b1d | 1060 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1061 | |
1062 | if (options != NULL && options[0]) { | |
1063 | seq_putc(m, ','); | |
1064 | mangle(m, options); | |
1065 | } | |
2a32cebd | 1066 | rcu_read_unlock(); |
b3b304a2 MS |
1067 | |
1068 | return 0; | |
1069 | } | |
1070 | EXPORT_SYMBOL(generic_show_options); | |
1071 | ||
1072 | /* | |
1073 | * If filesystem uses generic_show_options(), this function should be | |
1074 | * called from the fill_super() callback. | |
1075 | * | |
1076 | * The .remount_fs callback usually needs to be handled in a special | |
1077 | * way, to make sure, that previous options are not overwritten if the | |
1078 | * remount fails. | |
1079 | * | |
1080 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1081 | * reset all options to their default value, but changes only newly | |
1082 | * given options, then the displayed options will not reflect reality | |
1083 | * any more. | |
1084 | */ | |
1085 | void save_mount_options(struct super_block *sb, char *options) | |
1086 | { | |
2a32cebd AV |
1087 | BUG_ON(sb->s_options); |
1088 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1089 | } |
1090 | EXPORT_SYMBOL(save_mount_options); | |
1091 | ||
2a32cebd AV |
1092 | void replace_mount_options(struct super_block *sb, char *options) |
1093 | { | |
1094 | char *old = sb->s_options; | |
1095 | rcu_assign_pointer(sb->s_options, options); | |
1096 | if (old) { | |
1097 | synchronize_rcu(); | |
1098 | kfree(old); | |
1099 | } | |
1100 | } | |
1101 | EXPORT_SYMBOL(replace_mount_options); | |
1102 | ||
a1a2c409 | 1103 | #ifdef CONFIG_PROC_FS |
0226f492 | 1104 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1105 | static void *m_start(struct seq_file *m, loff_t *pos) |
1106 | { | |
6ce6e24e | 1107 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1108 | |
390c6843 | 1109 | down_read(&namespace_sem); |
c7999c36 AV |
1110 | if (p->cached_event == p->ns->event) { |
1111 | void *v = p->cached_mount; | |
1112 | if (*pos == p->cached_index) | |
1113 | return v; | |
1114 | if (*pos == p->cached_index + 1) { | |
1115 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1116 | return p->cached_mount = v; | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | p->cached_event = p->ns->event; | |
1121 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1122 | p->cached_index = *pos; | |
1123 | return p->cached_mount; | |
1da177e4 LT |
1124 | } |
1125 | ||
1126 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1127 | { | |
6ce6e24e | 1128 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1129 | |
c7999c36 AV |
1130 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1131 | p->cached_index = *pos; | |
1132 | return p->cached_mount; | |
1da177e4 LT |
1133 | } |
1134 | ||
1135 | static void m_stop(struct seq_file *m, void *v) | |
1136 | { | |
390c6843 | 1137 | up_read(&namespace_sem); |
1da177e4 LT |
1138 | } |
1139 | ||
0226f492 | 1140 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1141 | { |
6ce6e24e | 1142 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1143 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1144 | return p->show(m, &r->mnt); |
1da177e4 LT |
1145 | } |
1146 | ||
a1a2c409 | 1147 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1148 | .start = m_start, |
1149 | .next = m_next, | |
1150 | .stop = m_stop, | |
0226f492 | 1151 | .show = m_show, |
b4629fe2 | 1152 | }; |
a1a2c409 | 1153 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1154 | |
1da177e4 LT |
1155 | /** |
1156 | * may_umount_tree - check if a mount tree is busy | |
1157 | * @mnt: root of mount tree | |
1158 | * | |
1159 | * This is called to check if a tree of mounts has any | |
1160 | * open files, pwds, chroots or sub mounts that are | |
1161 | * busy. | |
1162 | */ | |
909b0a88 | 1163 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1164 | { |
909b0a88 | 1165 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1166 | int actual_refs = 0; |
1167 | int minimum_refs = 0; | |
315fc83e | 1168 | struct mount *p; |
909b0a88 | 1169 | BUG_ON(!m); |
1da177e4 | 1170 | |
b3e19d92 | 1171 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1172 | lock_mount_hash(); |
909b0a88 | 1173 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1174 | actual_refs += mnt_get_count(p); |
1da177e4 | 1175 | minimum_refs += 2; |
1da177e4 | 1176 | } |
719ea2fb | 1177 | unlock_mount_hash(); |
1da177e4 LT |
1178 | |
1179 | if (actual_refs > minimum_refs) | |
e3474a8e | 1180 | return 0; |
1da177e4 | 1181 | |
e3474a8e | 1182 | return 1; |
1da177e4 LT |
1183 | } |
1184 | ||
1185 | EXPORT_SYMBOL(may_umount_tree); | |
1186 | ||
1187 | /** | |
1188 | * may_umount - check if a mount point is busy | |
1189 | * @mnt: root of mount | |
1190 | * | |
1191 | * This is called to check if a mount point has any | |
1192 | * open files, pwds, chroots or sub mounts. If the | |
1193 | * mount has sub mounts this will return busy | |
1194 | * regardless of whether the sub mounts are busy. | |
1195 | * | |
1196 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1197 | * give false negatives. The main reason why it's here is that we need | |
1198 | * a non-destructive way to look for easily umountable filesystems. | |
1199 | */ | |
1200 | int may_umount(struct vfsmount *mnt) | |
1201 | { | |
e3474a8e | 1202 | int ret = 1; |
8ad08d8a | 1203 | down_read(&namespace_sem); |
719ea2fb | 1204 | lock_mount_hash(); |
1ab59738 | 1205 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1206 | ret = 0; |
719ea2fb | 1207 | unlock_mount_hash(); |
8ad08d8a | 1208 | up_read(&namespace_sem); |
a05964f3 | 1209 | return ret; |
1da177e4 LT |
1210 | } |
1211 | ||
1212 | EXPORT_SYMBOL(may_umount); | |
1213 | ||
38129a13 | 1214 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1215 | |
97216be0 | 1216 | static void namespace_unlock(void) |
70fbcdf4 | 1217 | { |
d5e50f74 | 1218 | struct mount *mnt; |
38129a13 | 1219 | struct hlist_head head = unmounted; |
97216be0 | 1220 | |
38129a13 | 1221 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1222 | up_write(&namespace_sem); |
1223 | return; | |
1224 | } | |
1225 | ||
38129a13 AV |
1226 | head.first->pprev = &head.first; |
1227 | INIT_HLIST_HEAD(&unmounted); | |
1228 | ||
97216be0 AV |
1229 | up_write(&namespace_sem); |
1230 | ||
48a066e7 AV |
1231 | synchronize_rcu(); |
1232 | ||
38129a13 AV |
1233 | while (!hlist_empty(&head)) { |
1234 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1235 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1236 | if (mnt->mnt_ex_mountpoint.mnt) |
1237 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1238 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1239 | } |
1240 | } | |
1241 | ||
97216be0 | 1242 | static inline void namespace_lock(void) |
e3197d83 | 1243 | { |
97216be0 | 1244 | down_write(&namespace_sem); |
e3197d83 AV |
1245 | } |
1246 | ||
99b7db7b | 1247 | /* |
48a066e7 | 1248 | * mount_lock must be held |
99b7db7b | 1249 | * namespace_sem must be held for write |
48a066e7 AV |
1250 | * how = 0 => just this tree, don't propagate |
1251 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1252 | * how = 2 => lazy umount | |
99b7db7b | 1253 | */ |
48a066e7 | 1254 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1255 | { |
38129a13 | 1256 | HLIST_HEAD(tmp_list); |
315fc83e | 1257 | struct mount *p; |
38129a13 | 1258 | struct mount *last = NULL; |
1da177e4 | 1259 | |
38129a13 AV |
1260 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1261 | hlist_del_init_rcu(&p->mnt_hash); | |
1262 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1263 | } | |
1da177e4 | 1264 | |
48a066e7 | 1265 | if (how) |
7b8a53fd | 1266 | propagate_umount(&tmp_list); |
a05964f3 | 1267 | |
38129a13 | 1268 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1269 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1270 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1271 | __touch_mnt_namespace(p->mnt_ns); |
1272 | p->mnt_ns = NULL; | |
48a066e7 AV |
1273 | if (how < 2) |
1274 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
6b41d536 | 1275 | list_del_init(&p->mnt_child); |
676da58d | 1276 | if (mnt_has_parent(p)) { |
84d17192 | 1277 | put_mountpoint(p->mnt_mp); |
aba809cf AV |
1278 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1279 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1280 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1281 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1282 | p->mnt_parent = p; | |
84d17192 | 1283 | p->mnt_mp = NULL; |
7c4b93d8 | 1284 | } |
0f0afb1d | 1285 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1286 | last = p; |
1287 | } | |
1288 | if (last) { | |
1289 | last->mnt_hash.next = unmounted.first; | |
1290 | unmounted.first = tmp_list.first; | |
1291 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1292 | } |
1293 | } | |
1294 | ||
b54b9be7 | 1295 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1296 | |
1ab59738 | 1297 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1298 | { |
1ab59738 | 1299 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1300 | int retval; |
1301 | ||
1ab59738 | 1302 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1303 | if (retval) |
1304 | return retval; | |
1305 | ||
1306 | /* | |
1307 | * Allow userspace to request a mountpoint be expired rather than | |
1308 | * unmounting unconditionally. Unmount only happens if: | |
1309 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1310 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1311 | */ | |
1312 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1313 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1314 | flags & (MNT_FORCE | MNT_DETACH)) |
1315 | return -EINVAL; | |
1316 | ||
b3e19d92 NP |
1317 | /* |
1318 | * probably don't strictly need the lock here if we examined | |
1319 | * all race cases, but it's a slowpath. | |
1320 | */ | |
719ea2fb | 1321 | lock_mount_hash(); |
83adc753 | 1322 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1323 | unlock_mount_hash(); |
1da177e4 | 1324 | return -EBUSY; |
b3e19d92 | 1325 | } |
719ea2fb | 1326 | unlock_mount_hash(); |
1da177e4 | 1327 | |
863d684f | 1328 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1329 | return -EAGAIN; |
1330 | } | |
1331 | ||
1332 | /* | |
1333 | * If we may have to abort operations to get out of this | |
1334 | * mount, and they will themselves hold resources we must | |
1335 | * allow the fs to do things. In the Unix tradition of | |
1336 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1337 | * might fail to complete on the first run through as other tasks | |
1338 | * must return, and the like. Thats for the mount program to worry | |
1339 | * about for the moment. | |
1340 | */ | |
1341 | ||
42faad99 | 1342 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1343 | sb->s_op->umount_begin(sb); |
42faad99 | 1344 | } |
1da177e4 LT |
1345 | |
1346 | /* | |
1347 | * No sense to grab the lock for this test, but test itself looks | |
1348 | * somewhat bogus. Suggestions for better replacement? | |
1349 | * Ho-hum... In principle, we might treat that as umount + switch | |
1350 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1351 | * Actually it makes sense, especially if rootfs would contain a | |
1352 | * /reboot - static binary that would close all descriptors and | |
1353 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1354 | */ | |
1ab59738 | 1355 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1356 | /* |
1357 | * Special case for "unmounting" root ... | |
1358 | * we just try to remount it readonly. | |
1359 | */ | |
1360 | down_write(&sb->s_umount); | |
4aa98cf7 | 1361 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1362 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1363 | up_write(&sb->s_umount); |
1364 | return retval; | |
1365 | } | |
1366 | ||
97216be0 | 1367 | namespace_lock(); |
719ea2fb | 1368 | lock_mount_hash(); |
5addc5dd | 1369 | event++; |
1da177e4 | 1370 | |
48a066e7 | 1371 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1372 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1373 | umount_tree(mnt, 2); |
1da177e4 | 1374 | retval = 0; |
48a066e7 AV |
1375 | } else { |
1376 | shrink_submounts(mnt); | |
1377 | retval = -EBUSY; | |
1378 | if (!propagate_mount_busy(mnt, 2)) { | |
1379 | if (!list_empty(&mnt->mnt_list)) | |
1380 | umount_tree(mnt, 1); | |
1381 | retval = 0; | |
1382 | } | |
1da177e4 | 1383 | } |
719ea2fb | 1384 | unlock_mount_hash(); |
e3197d83 | 1385 | namespace_unlock(); |
1da177e4 LT |
1386 | return retval; |
1387 | } | |
1388 | ||
9b40bc90 AV |
1389 | /* |
1390 | * Is the caller allowed to modify his namespace? | |
1391 | */ | |
1392 | static inline bool may_mount(void) | |
1393 | { | |
1394 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1395 | } | |
1396 | ||
1da177e4 LT |
1397 | /* |
1398 | * Now umount can handle mount points as well as block devices. | |
1399 | * This is important for filesystems which use unnamed block devices. | |
1400 | * | |
1401 | * We now support a flag for forced unmount like the other 'big iron' | |
1402 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1403 | */ | |
1404 | ||
bdc480e3 | 1405 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1406 | { |
2d8f3038 | 1407 | struct path path; |
900148dc | 1408 | struct mount *mnt; |
1da177e4 | 1409 | int retval; |
db1f05bb | 1410 | int lookup_flags = 0; |
1da177e4 | 1411 | |
db1f05bb MS |
1412 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1413 | return -EINVAL; | |
1414 | ||
9b40bc90 AV |
1415 | if (!may_mount()) |
1416 | return -EPERM; | |
1417 | ||
db1f05bb MS |
1418 | if (!(flags & UMOUNT_NOFOLLOW)) |
1419 | lookup_flags |= LOOKUP_FOLLOW; | |
1420 | ||
197df04c | 1421 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1422 | if (retval) |
1423 | goto out; | |
900148dc | 1424 | mnt = real_mount(path.mnt); |
1da177e4 | 1425 | retval = -EINVAL; |
2d8f3038 | 1426 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1427 | goto dput_and_out; |
143c8c91 | 1428 | if (!check_mnt(mnt)) |
1da177e4 | 1429 | goto dput_and_out; |
5ff9d8a6 EB |
1430 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1431 | goto dput_and_out; | |
1da177e4 | 1432 | |
900148dc | 1433 | retval = do_umount(mnt, flags); |
1da177e4 | 1434 | dput_and_out: |
429731b1 | 1435 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1436 | dput(path.dentry); |
900148dc | 1437 | mntput_no_expire(mnt); |
1da177e4 LT |
1438 | out: |
1439 | return retval; | |
1440 | } | |
1441 | ||
1442 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1443 | ||
1444 | /* | |
b58fed8b | 1445 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1446 | */ |
bdc480e3 | 1447 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1448 | { |
b58fed8b | 1449 | return sys_umount(name, 0); |
1da177e4 LT |
1450 | } |
1451 | ||
1452 | #endif | |
1453 | ||
4ce5d2b1 | 1454 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1455 | { |
4ce5d2b1 EB |
1456 | /* Is this a proxy for a mount namespace? */ |
1457 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1458 | struct proc_ns *ei; |
8823c079 EB |
1459 | |
1460 | if (!proc_ns_inode(inode)) | |
1461 | return false; | |
1462 | ||
0bb80f24 | 1463 | ei = get_proc_ns(inode); |
8823c079 EB |
1464 | if (ei->ns_ops != &mntns_operations) |
1465 | return false; | |
1466 | ||
4ce5d2b1 EB |
1467 | return true; |
1468 | } | |
1469 | ||
1470 | static bool mnt_ns_loop(struct dentry *dentry) | |
1471 | { | |
1472 | /* Could bind mounting the mount namespace inode cause a | |
1473 | * mount namespace loop? | |
1474 | */ | |
1475 | struct mnt_namespace *mnt_ns; | |
1476 | if (!is_mnt_ns_file(dentry)) | |
1477 | return false; | |
1478 | ||
1479 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1480 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1481 | } | |
1482 | ||
87129cc0 | 1483 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1484 | int flag) |
1da177e4 | 1485 | { |
84d17192 | 1486 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1487 | |
4ce5d2b1 EB |
1488 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1489 | return ERR_PTR(-EINVAL); | |
1490 | ||
1491 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1492 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1493 | |
36341f64 | 1494 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1495 | if (IS_ERR(q)) |
1496 | return q; | |
1497 | ||
5ff9d8a6 | 1498 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1499 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1500 | |
1501 | p = mnt; | |
6b41d536 | 1502 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1503 | struct mount *s; |
7ec02ef1 | 1504 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1505 | continue; |
1506 | ||
909b0a88 | 1507 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1508 | if (!(flag & CL_COPY_UNBINDABLE) && |
1509 | IS_MNT_UNBINDABLE(s)) { | |
1510 | s = skip_mnt_tree(s); | |
1511 | continue; | |
1512 | } | |
1513 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1514 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1515 | s = skip_mnt_tree(s); |
1516 | continue; | |
1517 | } | |
0714a533 AV |
1518 | while (p != s->mnt_parent) { |
1519 | p = p->mnt_parent; | |
1520 | q = q->mnt_parent; | |
1da177e4 | 1521 | } |
87129cc0 | 1522 | p = s; |
84d17192 | 1523 | parent = q; |
87129cc0 | 1524 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1525 | if (IS_ERR(q)) |
1526 | goto out; | |
719ea2fb | 1527 | lock_mount_hash(); |
1a4eeaf2 | 1528 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1529 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1530 | unlock_mount_hash(); |
1da177e4 LT |
1531 | } |
1532 | } | |
1533 | return res; | |
be34d1a3 | 1534 | out: |
1da177e4 | 1535 | if (res) { |
719ea2fb | 1536 | lock_mount_hash(); |
328e6d90 | 1537 | umount_tree(res, 0); |
719ea2fb | 1538 | unlock_mount_hash(); |
1da177e4 | 1539 | } |
be34d1a3 | 1540 | return q; |
1da177e4 LT |
1541 | } |
1542 | ||
be34d1a3 DH |
1543 | /* Caller should check returned pointer for errors */ |
1544 | ||
589ff870 | 1545 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1546 | { |
cb338d06 | 1547 | struct mount *tree; |
97216be0 | 1548 | namespace_lock(); |
87129cc0 AV |
1549 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1550 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1551 | namespace_unlock(); |
be34d1a3 | 1552 | if (IS_ERR(tree)) |
52e220d3 | 1553 | return ERR_CAST(tree); |
be34d1a3 | 1554 | return &tree->mnt; |
8aec0809 AV |
1555 | } |
1556 | ||
1557 | void drop_collected_mounts(struct vfsmount *mnt) | |
1558 | { | |
97216be0 | 1559 | namespace_lock(); |
719ea2fb | 1560 | lock_mount_hash(); |
328e6d90 | 1561 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1562 | unlock_mount_hash(); |
3ab6abee | 1563 | namespace_unlock(); |
8aec0809 AV |
1564 | } |
1565 | ||
1f707137 AV |
1566 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1567 | struct vfsmount *root) | |
1568 | { | |
1a4eeaf2 | 1569 | struct mount *mnt; |
1f707137 AV |
1570 | int res = f(root, arg); |
1571 | if (res) | |
1572 | return res; | |
1a4eeaf2 AV |
1573 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1574 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1575 | if (res) |
1576 | return res; | |
1577 | } | |
1578 | return 0; | |
1579 | } | |
1580 | ||
4b8b21f4 | 1581 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1582 | { |
315fc83e | 1583 | struct mount *p; |
719f5d7f | 1584 | |
909b0a88 | 1585 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1586 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1587 | mnt_release_group_id(p); |
719f5d7f MS |
1588 | } |
1589 | } | |
1590 | ||
4b8b21f4 | 1591 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1592 | { |
315fc83e | 1593 | struct mount *p; |
719f5d7f | 1594 | |
909b0a88 | 1595 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1596 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1597 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1598 | if (err) { |
4b8b21f4 | 1599 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1600 | return err; |
1601 | } | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | return 0; | |
1606 | } | |
1607 | ||
b90fa9ae RP |
1608 | /* |
1609 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1610 | * @nd : place the mount tree @source_mnt is attached |
1611 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1612 | * store the parent mount and mountpoint dentry. | |
1613 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1614 | * |
1615 | * NOTE: in the table below explains the semantics when a source mount | |
1616 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1617 | * --------------------------------------------------------------------------- |
1618 | * | BIND MOUNT OPERATION | | |
1619 | * |************************************************************************** | |
1620 | * | source-->| shared | private | slave | unbindable | | |
1621 | * | dest | | | | | | |
1622 | * | | | | | | | | |
1623 | * | v | | | | | | |
1624 | * |************************************************************************** | |
1625 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1626 | * | | | | | | | |
1627 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1628 | * *************************************************************************** | |
b90fa9ae RP |
1629 | * A bind operation clones the source mount and mounts the clone on the |
1630 | * destination mount. | |
1631 | * | |
1632 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1633 | * tree of the destination mount and the cloned mount is added to | |
1634 | * the peer group of the source mount. | |
1635 | * (+) the cloned mount is created under the destination mount and is marked | |
1636 | * as shared. The cloned mount is added to the peer group of the source | |
1637 | * mount. | |
5afe0022 RP |
1638 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1639 | * of the destination mount and the cloned mount is made slave | |
1640 | * of the same master as that of the source mount. The cloned mount | |
1641 | * is marked as 'shared and slave'. | |
1642 | * (*) the cloned mount is made a slave of the same master as that of the | |
1643 | * source mount. | |
1644 | * | |
9676f0c6 RP |
1645 | * --------------------------------------------------------------------------- |
1646 | * | MOVE MOUNT OPERATION | | |
1647 | * |************************************************************************** | |
1648 | * | source-->| shared | private | slave | unbindable | | |
1649 | * | dest | | | | | | |
1650 | * | | | | | | | | |
1651 | * | v | | | | | | |
1652 | * |************************************************************************** | |
1653 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1654 | * | | | | | | | |
1655 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1656 | * *************************************************************************** | |
5afe0022 RP |
1657 | * |
1658 | * (+) the mount is moved to the destination. And is then propagated to | |
1659 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1660 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1661 | * (+++) the mount is moved to the destination and is then propagated to |
1662 | * all the mounts belonging to the destination mount's propagation tree. | |
1663 | * the mount is marked as 'shared and slave'. | |
1664 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1665 | * |
1666 | * if the source mount is a tree, the operations explained above is | |
1667 | * applied to each mount in the tree. | |
1668 | * Must be called without spinlocks held, since this function can sleep | |
1669 | * in allocations. | |
1670 | */ | |
0fb54e50 | 1671 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1672 | struct mount *dest_mnt, |
1673 | struct mountpoint *dest_mp, | |
1674 | struct path *parent_path) | |
b90fa9ae | 1675 | { |
38129a13 | 1676 | HLIST_HEAD(tree_list); |
315fc83e | 1677 | struct mount *child, *p; |
38129a13 | 1678 | struct hlist_node *n; |
719f5d7f | 1679 | int err; |
b90fa9ae | 1680 | |
fc7be130 | 1681 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1682 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1683 | if (err) |
1684 | goto out; | |
0b1b901b | 1685 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1686 | lock_mount_hash(); |
0b1b901b AV |
1687 | if (err) |
1688 | goto out_cleanup_ids; | |
909b0a88 | 1689 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1690 | set_mnt_shared(p); |
0b1b901b AV |
1691 | } else { |
1692 | lock_mount_hash(); | |
b90fa9ae | 1693 | } |
1a390689 | 1694 | if (parent_path) { |
0fb54e50 | 1695 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1696 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1697 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1698 | } else { |
84d17192 | 1699 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1700 | commit_tree(source_mnt, NULL); |
21444403 | 1701 | } |
b90fa9ae | 1702 | |
38129a13 | 1703 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1704 | struct mount *q; |
38129a13 | 1705 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1706 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1707 | child->mnt_mountpoint); | |
1708 | commit_tree(child, q); | |
b90fa9ae | 1709 | } |
719ea2fb | 1710 | unlock_mount_hash(); |
99b7db7b | 1711 | |
b90fa9ae | 1712 | return 0; |
719f5d7f MS |
1713 | |
1714 | out_cleanup_ids: | |
f2ebb3a9 AV |
1715 | while (!hlist_empty(&tree_list)) { |
1716 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1717 | umount_tree(child, 0); | |
1718 | } | |
1719 | unlock_mount_hash(); | |
0b1b901b | 1720 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1721 | out: |
1722 | return err; | |
b90fa9ae RP |
1723 | } |
1724 | ||
84d17192 | 1725 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1726 | { |
1727 | struct vfsmount *mnt; | |
84d17192 | 1728 | struct dentry *dentry = path->dentry; |
b12cea91 | 1729 | retry: |
84d17192 AV |
1730 | mutex_lock(&dentry->d_inode->i_mutex); |
1731 | if (unlikely(cant_mount(dentry))) { | |
1732 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1733 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1734 | } |
97216be0 | 1735 | namespace_lock(); |
b12cea91 | 1736 | mnt = lookup_mnt(path); |
84d17192 AV |
1737 | if (likely(!mnt)) { |
1738 | struct mountpoint *mp = new_mountpoint(dentry); | |
1739 | if (IS_ERR(mp)) { | |
97216be0 | 1740 | namespace_unlock(); |
84d17192 AV |
1741 | mutex_unlock(&dentry->d_inode->i_mutex); |
1742 | return mp; | |
1743 | } | |
1744 | return mp; | |
1745 | } | |
97216be0 | 1746 | namespace_unlock(); |
b12cea91 AV |
1747 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1748 | path_put(path); | |
1749 | path->mnt = mnt; | |
84d17192 | 1750 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1751 | goto retry; |
1752 | } | |
1753 | ||
84d17192 | 1754 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1755 | { |
84d17192 AV |
1756 | struct dentry *dentry = where->m_dentry; |
1757 | put_mountpoint(where); | |
328e6d90 | 1758 | namespace_unlock(); |
84d17192 | 1759 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1760 | } |
1761 | ||
84d17192 | 1762 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1763 | { |
95bc5f25 | 1764 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1765 | return -EINVAL; |
1766 | ||
84d17192 | 1767 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1768 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1769 | return -ENOTDIR; |
1770 | ||
84d17192 | 1771 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1772 | } |
1773 | ||
7a2e8a8f VA |
1774 | /* |
1775 | * Sanity check the flags to change_mnt_propagation. | |
1776 | */ | |
1777 | ||
1778 | static int flags_to_propagation_type(int flags) | |
1779 | { | |
7c6e984d | 1780 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1781 | |
1782 | /* Fail if any non-propagation flags are set */ | |
1783 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1784 | return 0; | |
1785 | /* Only one propagation flag should be set */ | |
1786 | if (!is_power_of_2(type)) | |
1787 | return 0; | |
1788 | return type; | |
1789 | } | |
1790 | ||
07b20889 RP |
1791 | /* |
1792 | * recursively change the type of the mountpoint. | |
1793 | */ | |
0a0d8a46 | 1794 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1795 | { |
315fc83e | 1796 | struct mount *m; |
4b8b21f4 | 1797 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1798 | int recurse = flag & MS_REC; |
7a2e8a8f | 1799 | int type; |
719f5d7f | 1800 | int err = 0; |
07b20889 | 1801 | |
2d92ab3c | 1802 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1803 | return -EINVAL; |
1804 | ||
7a2e8a8f VA |
1805 | type = flags_to_propagation_type(flag); |
1806 | if (!type) | |
1807 | return -EINVAL; | |
1808 | ||
97216be0 | 1809 | namespace_lock(); |
719f5d7f MS |
1810 | if (type == MS_SHARED) { |
1811 | err = invent_group_ids(mnt, recurse); | |
1812 | if (err) | |
1813 | goto out_unlock; | |
1814 | } | |
1815 | ||
719ea2fb | 1816 | lock_mount_hash(); |
909b0a88 | 1817 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1818 | change_mnt_propagation(m, type); |
719ea2fb | 1819 | unlock_mount_hash(); |
719f5d7f MS |
1820 | |
1821 | out_unlock: | |
97216be0 | 1822 | namespace_unlock(); |
719f5d7f | 1823 | return err; |
07b20889 RP |
1824 | } |
1825 | ||
5ff9d8a6 EB |
1826 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1827 | { | |
1828 | struct mount *child; | |
1829 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1830 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1831 | continue; | |
1832 | ||
1833 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1834 | return true; | |
1835 | } | |
1836 | return false; | |
1837 | } | |
1838 | ||
1da177e4 LT |
1839 | /* |
1840 | * do loopback mount. | |
1841 | */ | |
808d4e3c | 1842 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1843 | int recurse) |
1da177e4 | 1844 | { |
2d92ab3c | 1845 | struct path old_path; |
84d17192 AV |
1846 | struct mount *mnt = NULL, *old, *parent; |
1847 | struct mountpoint *mp; | |
57eccb83 | 1848 | int err; |
1da177e4 LT |
1849 | if (!old_name || !*old_name) |
1850 | return -EINVAL; | |
815d405c | 1851 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1852 | if (err) |
1853 | return err; | |
1854 | ||
8823c079 | 1855 | err = -EINVAL; |
4ce5d2b1 | 1856 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1857 | goto out; |
1858 | ||
84d17192 AV |
1859 | mp = lock_mount(path); |
1860 | err = PTR_ERR(mp); | |
1861 | if (IS_ERR(mp)) | |
b12cea91 AV |
1862 | goto out; |
1863 | ||
87129cc0 | 1864 | old = real_mount(old_path.mnt); |
84d17192 | 1865 | parent = real_mount(path->mnt); |
87129cc0 | 1866 | |
1da177e4 | 1867 | err = -EINVAL; |
fc7be130 | 1868 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1869 | goto out2; |
9676f0c6 | 1870 | |
84d17192 | 1871 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1872 | goto out2; |
1da177e4 | 1873 | |
5ff9d8a6 EB |
1874 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1875 | goto out2; | |
1876 | ||
ccd48bc7 | 1877 | if (recurse) |
4ce5d2b1 | 1878 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1879 | else |
87129cc0 | 1880 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1881 | |
be34d1a3 DH |
1882 | if (IS_ERR(mnt)) { |
1883 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1884 | goto out2; |
be34d1a3 | 1885 | } |
ccd48bc7 | 1886 | |
5ff9d8a6 EB |
1887 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1888 | ||
84d17192 | 1889 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1890 | if (err) { |
719ea2fb | 1891 | lock_mount_hash(); |
328e6d90 | 1892 | umount_tree(mnt, 0); |
719ea2fb | 1893 | unlock_mount_hash(); |
5b83d2c5 | 1894 | } |
b12cea91 | 1895 | out2: |
84d17192 | 1896 | unlock_mount(mp); |
ccd48bc7 | 1897 | out: |
2d92ab3c | 1898 | path_put(&old_path); |
1da177e4 LT |
1899 | return err; |
1900 | } | |
1901 | ||
2e4b7fcd DH |
1902 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1903 | { | |
1904 | int error = 0; | |
1905 | int readonly_request = 0; | |
1906 | ||
1907 | if (ms_flags & MS_RDONLY) | |
1908 | readonly_request = 1; | |
1909 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1910 | return 0; | |
1911 | ||
1912 | if (readonly_request) | |
83adc753 | 1913 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1914 | else |
83adc753 | 1915 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1916 | return error; |
1917 | } | |
1918 | ||
1da177e4 LT |
1919 | /* |
1920 | * change filesystem flags. dir should be a physical root of filesystem. | |
1921 | * If you've mounted a non-root directory somewhere and want to do remount | |
1922 | * on it - tough luck. | |
1923 | */ | |
0a0d8a46 | 1924 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1925 | void *data) |
1926 | { | |
1927 | int err; | |
2d92ab3c | 1928 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1929 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1930 | |
143c8c91 | 1931 | if (!check_mnt(mnt)) |
1da177e4 LT |
1932 | return -EINVAL; |
1933 | ||
2d92ab3c | 1934 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1935 | return -EINVAL; |
1936 | ||
07b64558 EB |
1937 | /* Don't allow changing of locked mnt flags. |
1938 | * | |
1939 | * No locks need to be held here while testing the various | |
1940 | * MNT_LOCK flags because those flags can never be cleared | |
1941 | * once they are set. | |
1942 | */ | |
1943 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
1944 | !(mnt_flags & MNT_READONLY)) { | |
1945 | return -EPERM; | |
1946 | } | |
9566d674 EB |
1947 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
1948 | !(mnt_flags & MNT_NODEV)) { | |
1949 | return -EPERM; | |
1950 | } | |
1951 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
1952 | !(mnt_flags & MNT_NOSUID)) { | |
1953 | return -EPERM; | |
1954 | } | |
1955 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
1956 | !(mnt_flags & MNT_NOEXEC)) { | |
1957 | return -EPERM; | |
1958 | } | |
1959 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
1960 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
1961 | return -EPERM; | |
1962 | } | |
1963 | ||
ff36fe2c EP |
1964 | err = security_sb_remount(sb, data); |
1965 | if (err) | |
1966 | return err; | |
1967 | ||
1da177e4 | 1968 | down_write(&sb->s_umount); |
2e4b7fcd | 1969 | if (flags & MS_BIND) |
2d92ab3c | 1970 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1971 | else if (!capable(CAP_SYS_ADMIN)) |
1972 | err = -EPERM; | |
4aa98cf7 | 1973 | else |
2e4b7fcd | 1974 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1975 | if (!err) { |
719ea2fb | 1976 | lock_mount_hash(); |
a6138db8 | 1977 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 1978 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 1979 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 1980 | unlock_mount_hash(); |
0e55a7cc | 1981 | } |
6339dab8 | 1982 | up_write(&sb->s_umount); |
1da177e4 LT |
1983 | return err; |
1984 | } | |
1985 | ||
cbbe362c | 1986 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1987 | { |
315fc83e | 1988 | struct mount *p; |
909b0a88 | 1989 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1990 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1991 | return 1; |
1992 | } | |
1993 | return 0; | |
1994 | } | |
1995 | ||
808d4e3c | 1996 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1997 | { |
2d92ab3c | 1998 | struct path old_path, parent_path; |
676da58d | 1999 | struct mount *p; |
0fb54e50 | 2000 | struct mount *old; |
84d17192 | 2001 | struct mountpoint *mp; |
57eccb83 | 2002 | int err; |
1da177e4 LT |
2003 | if (!old_name || !*old_name) |
2004 | return -EINVAL; | |
2d92ab3c | 2005 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2006 | if (err) |
2007 | return err; | |
2008 | ||
84d17192 AV |
2009 | mp = lock_mount(path); |
2010 | err = PTR_ERR(mp); | |
2011 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2012 | goto out; |
2013 | ||
143c8c91 | 2014 | old = real_mount(old_path.mnt); |
fc7be130 | 2015 | p = real_mount(path->mnt); |
143c8c91 | 2016 | |
1da177e4 | 2017 | err = -EINVAL; |
fc7be130 | 2018 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2019 | goto out1; |
2020 | ||
5ff9d8a6 EB |
2021 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2022 | goto out1; | |
2023 | ||
1da177e4 | 2024 | err = -EINVAL; |
2d92ab3c | 2025 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2026 | goto out1; |
1da177e4 | 2027 | |
676da58d | 2028 | if (!mnt_has_parent(old)) |
21444403 | 2029 | goto out1; |
1da177e4 | 2030 | |
2d92ab3c AV |
2031 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
2032 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
2033 | goto out1; |
2034 | /* | |
2035 | * Don't move a mount residing in a shared parent. | |
2036 | */ | |
fc7be130 | 2037 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2038 | goto out1; |
9676f0c6 RP |
2039 | /* |
2040 | * Don't move a mount tree containing unbindable mounts to a destination | |
2041 | * mount which is shared. | |
2042 | */ | |
fc7be130 | 2043 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2044 | goto out1; |
1da177e4 | 2045 | err = -ELOOP; |
fc7be130 | 2046 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2047 | if (p == old) |
21444403 | 2048 | goto out1; |
1da177e4 | 2049 | |
84d17192 | 2050 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2051 | if (err) |
21444403 | 2052 | goto out1; |
1da177e4 LT |
2053 | |
2054 | /* if the mount is moved, it should no longer be expire | |
2055 | * automatically */ | |
6776db3d | 2056 | list_del_init(&old->mnt_expire); |
1da177e4 | 2057 | out1: |
84d17192 | 2058 | unlock_mount(mp); |
1da177e4 | 2059 | out: |
1da177e4 | 2060 | if (!err) |
1a390689 | 2061 | path_put(&parent_path); |
2d92ab3c | 2062 | path_put(&old_path); |
1da177e4 LT |
2063 | return err; |
2064 | } | |
2065 | ||
9d412a43 AV |
2066 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2067 | { | |
2068 | int err; | |
2069 | const char *subtype = strchr(fstype, '.'); | |
2070 | if (subtype) { | |
2071 | subtype++; | |
2072 | err = -EINVAL; | |
2073 | if (!subtype[0]) | |
2074 | goto err; | |
2075 | } else | |
2076 | subtype = ""; | |
2077 | ||
2078 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2079 | err = -ENOMEM; | |
2080 | if (!mnt->mnt_sb->s_subtype) | |
2081 | goto err; | |
2082 | return mnt; | |
2083 | ||
2084 | err: | |
2085 | mntput(mnt); | |
2086 | return ERR_PTR(err); | |
2087 | } | |
2088 | ||
9d412a43 AV |
2089 | /* |
2090 | * add a mount into a namespace's mount tree | |
2091 | */ | |
95bc5f25 | 2092 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2093 | { |
84d17192 AV |
2094 | struct mountpoint *mp; |
2095 | struct mount *parent; | |
9d412a43 AV |
2096 | int err; |
2097 | ||
f2ebb3a9 | 2098 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2099 | |
84d17192 AV |
2100 | mp = lock_mount(path); |
2101 | if (IS_ERR(mp)) | |
2102 | return PTR_ERR(mp); | |
9d412a43 | 2103 | |
84d17192 | 2104 | parent = real_mount(path->mnt); |
9d412a43 | 2105 | err = -EINVAL; |
84d17192 | 2106 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2107 | /* that's acceptable only for automounts done in private ns */ |
2108 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2109 | goto unlock; | |
2110 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2111 | if (!parent->mnt_ns) |
156cacb1 AV |
2112 | goto unlock; |
2113 | } | |
9d412a43 AV |
2114 | |
2115 | /* Refuse the same filesystem on the same mount point */ | |
2116 | err = -EBUSY; | |
95bc5f25 | 2117 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2118 | path->mnt->mnt_root == path->dentry) |
2119 | goto unlock; | |
2120 | ||
2121 | err = -EINVAL; | |
95bc5f25 | 2122 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2123 | goto unlock; |
2124 | ||
95bc5f25 | 2125 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2126 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2127 | |
2128 | unlock: | |
84d17192 | 2129 | unlock_mount(mp); |
9d412a43 AV |
2130 | return err; |
2131 | } | |
b1e75df4 | 2132 | |
1da177e4 LT |
2133 | /* |
2134 | * create a new mount for userspace and request it to be added into the | |
2135 | * namespace's tree | |
2136 | */ | |
0c55cfc4 | 2137 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2138 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2139 | { |
0c55cfc4 | 2140 | struct file_system_type *type; |
9b40bc90 | 2141 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2142 | struct vfsmount *mnt; |
15f9a3f3 | 2143 | int err; |
1da177e4 | 2144 | |
0c55cfc4 | 2145 | if (!fstype) |
1da177e4 LT |
2146 | return -EINVAL; |
2147 | ||
0c55cfc4 EB |
2148 | type = get_fs_type(fstype); |
2149 | if (!type) | |
2150 | return -ENODEV; | |
2151 | ||
2152 | if (user_ns != &init_user_ns) { | |
2153 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2154 | put_filesystem(type); | |
2155 | return -EPERM; | |
2156 | } | |
2157 | /* Only in special cases allow devices from mounts | |
2158 | * created outside the initial user namespace. | |
2159 | */ | |
2160 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2161 | flags |= MS_NODEV; | |
9566d674 | 2162 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 EB |
2163 | } |
2164 | } | |
2165 | ||
2166 | mnt = vfs_kern_mount(type, flags, name, data); | |
2167 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2168 | !mnt->mnt_sb->s_subtype) | |
2169 | mnt = fs_set_subtype(mnt, fstype); | |
2170 | ||
2171 | put_filesystem(type); | |
1da177e4 LT |
2172 | if (IS_ERR(mnt)) |
2173 | return PTR_ERR(mnt); | |
2174 | ||
95bc5f25 | 2175 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2176 | if (err) |
2177 | mntput(mnt); | |
2178 | return err; | |
1da177e4 LT |
2179 | } |
2180 | ||
19a167af AV |
2181 | int finish_automount(struct vfsmount *m, struct path *path) |
2182 | { | |
6776db3d | 2183 | struct mount *mnt = real_mount(m); |
19a167af AV |
2184 | int err; |
2185 | /* The new mount record should have at least 2 refs to prevent it being | |
2186 | * expired before we get a chance to add it | |
2187 | */ | |
6776db3d | 2188 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2189 | |
2190 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2191 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2192 | err = -ELOOP; |
2193 | goto fail; | |
19a167af AV |
2194 | } |
2195 | ||
95bc5f25 | 2196 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2197 | if (!err) |
2198 | return 0; | |
2199 | fail: | |
2200 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2201 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2202 | namespace_lock(); |
6776db3d | 2203 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2204 | namespace_unlock(); |
19a167af | 2205 | } |
b1e75df4 AV |
2206 | mntput(m); |
2207 | mntput(m); | |
19a167af AV |
2208 | return err; |
2209 | } | |
2210 | ||
ea5b778a DH |
2211 | /** |
2212 | * mnt_set_expiry - Put a mount on an expiration list | |
2213 | * @mnt: The mount to list. | |
2214 | * @expiry_list: The list to add the mount to. | |
2215 | */ | |
2216 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2217 | { | |
97216be0 | 2218 | namespace_lock(); |
ea5b778a | 2219 | |
6776db3d | 2220 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2221 | |
97216be0 | 2222 | namespace_unlock(); |
ea5b778a DH |
2223 | } |
2224 | EXPORT_SYMBOL(mnt_set_expiry); | |
2225 | ||
1da177e4 LT |
2226 | /* |
2227 | * process a list of expirable mountpoints with the intent of discarding any | |
2228 | * mountpoints that aren't in use and haven't been touched since last we came | |
2229 | * here | |
2230 | */ | |
2231 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2232 | { | |
761d5c38 | 2233 | struct mount *mnt, *next; |
1da177e4 LT |
2234 | LIST_HEAD(graveyard); |
2235 | ||
2236 | if (list_empty(mounts)) | |
2237 | return; | |
2238 | ||
97216be0 | 2239 | namespace_lock(); |
719ea2fb | 2240 | lock_mount_hash(); |
1da177e4 LT |
2241 | |
2242 | /* extract from the expiration list every vfsmount that matches the | |
2243 | * following criteria: | |
2244 | * - only referenced by its parent vfsmount | |
2245 | * - still marked for expiry (marked on the last call here; marks are | |
2246 | * cleared by mntput()) | |
2247 | */ | |
6776db3d | 2248 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2249 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2250 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2251 | continue; |
6776db3d | 2252 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2253 | } |
bcc5c7d2 | 2254 | while (!list_empty(&graveyard)) { |
6776db3d | 2255 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2256 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2257 | umount_tree(mnt, 1); |
bcc5c7d2 | 2258 | } |
719ea2fb | 2259 | unlock_mount_hash(); |
3ab6abee | 2260 | namespace_unlock(); |
5528f911 TM |
2261 | } |
2262 | ||
2263 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2264 | ||
2265 | /* | |
2266 | * Ripoff of 'select_parent()' | |
2267 | * | |
2268 | * search the list of submounts for a given mountpoint, and move any | |
2269 | * shrinkable submounts to the 'graveyard' list. | |
2270 | */ | |
692afc31 | 2271 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2272 | { |
692afc31 | 2273 | struct mount *this_parent = parent; |
5528f911 TM |
2274 | struct list_head *next; |
2275 | int found = 0; | |
2276 | ||
2277 | repeat: | |
6b41d536 | 2278 | next = this_parent->mnt_mounts.next; |
5528f911 | 2279 | resume: |
6b41d536 | 2280 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2281 | struct list_head *tmp = next; |
6b41d536 | 2282 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2283 | |
2284 | next = tmp->next; | |
692afc31 | 2285 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2286 | continue; |
5528f911 TM |
2287 | /* |
2288 | * Descend a level if the d_mounts list is non-empty. | |
2289 | */ | |
6b41d536 | 2290 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2291 | this_parent = mnt; |
2292 | goto repeat; | |
2293 | } | |
1da177e4 | 2294 | |
1ab59738 | 2295 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2296 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2297 | found++; |
2298 | } | |
1da177e4 | 2299 | } |
5528f911 TM |
2300 | /* |
2301 | * All done at this level ... ascend and resume the search | |
2302 | */ | |
2303 | if (this_parent != parent) { | |
6b41d536 | 2304 | next = this_parent->mnt_child.next; |
0714a533 | 2305 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2306 | goto resume; |
2307 | } | |
2308 | return found; | |
2309 | } | |
2310 | ||
2311 | /* | |
2312 | * process a list of expirable mountpoints with the intent of discarding any | |
2313 | * submounts of a specific parent mountpoint | |
99b7db7b | 2314 | * |
48a066e7 | 2315 | * mount_lock must be held for write |
5528f911 | 2316 | */ |
b54b9be7 | 2317 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2318 | { |
2319 | LIST_HEAD(graveyard); | |
761d5c38 | 2320 | struct mount *m; |
5528f911 | 2321 | |
5528f911 | 2322 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2323 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2324 | while (!list_empty(&graveyard)) { |
761d5c38 | 2325 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2326 | mnt_expire); |
143c8c91 | 2327 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2328 | umount_tree(m, 1); |
bcc5c7d2 AV |
2329 | } |
2330 | } | |
1da177e4 LT |
2331 | } |
2332 | ||
1da177e4 LT |
2333 | /* |
2334 | * Some copy_from_user() implementations do not return the exact number of | |
2335 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2336 | * Note that this function differs from copy_from_user() in that it will oops | |
2337 | * on bad values of `to', rather than returning a short copy. | |
2338 | */ | |
b58fed8b RP |
2339 | static long exact_copy_from_user(void *to, const void __user * from, |
2340 | unsigned long n) | |
1da177e4 LT |
2341 | { |
2342 | char *t = to; | |
2343 | const char __user *f = from; | |
2344 | char c; | |
2345 | ||
2346 | if (!access_ok(VERIFY_READ, from, n)) | |
2347 | return n; | |
2348 | ||
2349 | while (n) { | |
2350 | if (__get_user(c, f)) { | |
2351 | memset(t, 0, n); | |
2352 | break; | |
2353 | } | |
2354 | *t++ = c; | |
2355 | f++; | |
2356 | n--; | |
2357 | } | |
2358 | return n; | |
2359 | } | |
2360 | ||
b58fed8b | 2361 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2362 | { |
2363 | int i; | |
2364 | unsigned long page; | |
2365 | unsigned long size; | |
b58fed8b | 2366 | |
1da177e4 LT |
2367 | *where = 0; |
2368 | if (!data) | |
2369 | return 0; | |
2370 | ||
2371 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2372 | return -ENOMEM; | |
2373 | ||
2374 | /* We only care that *some* data at the address the user | |
2375 | * gave us is valid. Just in case, we'll zero | |
2376 | * the remainder of the page. | |
2377 | */ | |
2378 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2379 | size = TASK_SIZE - (unsigned long)data; | |
2380 | if (size > PAGE_SIZE) | |
2381 | size = PAGE_SIZE; | |
2382 | ||
2383 | i = size - exact_copy_from_user((void *)page, data, size); | |
2384 | if (!i) { | |
b58fed8b | 2385 | free_page(page); |
1da177e4 LT |
2386 | return -EFAULT; |
2387 | } | |
2388 | if (i != PAGE_SIZE) | |
2389 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2390 | *where = page; | |
2391 | return 0; | |
2392 | } | |
2393 | ||
eca6f534 VN |
2394 | int copy_mount_string(const void __user *data, char **where) |
2395 | { | |
2396 | char *tmp; | |
2397 | ||
2398 | if (!data) { | |
2399 | *where = NULL; | |
2400 | return 0; | |
2401 | } | |
2402 | ||
2403 | tmp = strndup_user(data, PAGE_SIZE); | |
2404 | if (IS_ERR(tmp)) | |
2405 | return PTR_ERR(tmp); | |
2406 | ||
2407 | *where = tmp; | |
2408 | return 0; | |
2409 | } | |
2410 | ||
1da177e4 LT |
2411 | /* |
2412 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2413 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2414 | * | |
2415 | * data is a (void *) that can point to any structure up to | |
2416 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2417 | * information (or be NULL). | |
2418 | * | |
2419 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2420 | * When the flags word was introduced its top half was required | |
2421 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2422 | * Therefore, if this magic number is present, it carries no information | |
2423 | * and must be discarded. | |
2424 | */ | |
808d4e3c AV |
2425 | long do_mount(const char *dev_name, const char *dir_name, |
2426 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2427 | { |
2d92ab3c | 2428 | struct path path; |
1da177e4 LT |
2429 | int retval = 0; |
2430 | int mnt_flags = 0; | |
2431 | ||
2432 | /* Discard magic */ | |
2433 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2434 | flags &= ~MS_MGC_MSK; | |
2435 | ||
2436 | /* Basic sanity checks */ | |
2437 | ||
2438 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2439 | return -EINVAL; | |
1da177e4 LT |
2440 | |
2441 | if (data_page) | |
2442 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2443 | ||
a27ab9f2 TH |
2444 | /* ... and get the mountpoint */ |
2445 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2446 | if (retval) | |
2447 | return retval; | |
2448 | ||
2449 | retval = security_sb_mount(dev_name, &path, | |
2450 | type_page, flags, data_page); | |
0d5cadb8 AV |
2451 | if (!retval && !may_mount()) |
2452 | retval = -EPERM; | |
a27ab9f2 TH |
2453 | if (retval) |
2454 | goto dput_out; | |
2455 | ||
613cbe3d AK |
2456 | /* Default to relatime unless overriden */ |
2457 | if (!(flags & MS_NOATIME)) | |
2458 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2459 | |
1da177e4 LT |
2460 | /* Separate the per-mountpoint flags */ |
2461 | if (flags & MS_NOSUID) | |
2462 | mnt_flags |= MNT_NOSUID; | |
2463 | if (flags & MS_NODEV) | |
2464 | mnt_flags |= MNT_NODEV; | |
2465 | if (flags & MS_NOEXEC) | |
2466 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2467 | if (flags & MS_NOATIME) |
2468 | mnt_flags |= MNT_NOATIME; | |
2469 | if (flags & MS_NODIRATIME) | |
2470 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2471 | if (flags & MS_STRICTATIME) |
2472 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2473 | if (flags & MS_RDONLY) |
2474 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2475 | |
7a4dec53 | 2476 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2477 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2478 | MS_STRICTATIME); | |
1da177e4 | 2479 | |
1da177e4 | 2480 | if (flags & MS_REMOUNT) |
2d92ab3c | 2481 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2482 | data_page); |
2483 | else if (flags & MS_BIND) | |
2d92ab3c | 2484 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2485 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2486 | retval = do_change_type(&path, flags); |
1da177e4 | 2487 | else if (flags & MS_MOVE) |
2d92ab3c | 2488 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2489 | else |
2d92ab3c | 2490 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2491 | dev_name, data_page); |
2492 | dput_out: | |
2d92ab3c | 2493 | path_put(&path); |
1da177e4 LT |
2494 | return retval; |
2495 | } | |
2496 | ||
771b1371 EB |
2497 | static void free_mnt_ns(struct mnt_namespace *ns) |
2498 | { | |
98f842e6 | 2499 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2500 | put_user_ns(ns->user_ns); |
2501 | kfree(ns); | |
2502 | } | |
2503 | ||
8823c079 EB |
2504 | /* |
2505 | * Assign a sequence number so we can detect when we attempt to bind | |
2506 | * mount a reference to an older mount namespace into the current | |
2507 | * mount namespace, preventing reference counting loops. A 64bit | |
2508 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2509 | * is effectively never, so we can ignore the possibility. | |
2510 | */ | |
2511 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2512 | ||
771b1371 | 2513 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2514 | { |
2515 | struct mnt_namespace *new_ns; | |
98f842e6 | 2516 | int ret; |
cf8d2c11 TM |
2517 | |
2518 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2519 | if (!new_ns) | |
2520 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2521 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2522 | if (ret) { | |
2523 | kfree(new_ns); | |
2524 | return ERR_PTR(ret); | |
2525 | } | |
8823c079 | 2526 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2527 | atomic_set(&new_ns->count, 1); |
2528 | new_ns->root = NULL; | |
2529 | INIT_LIST_HEAD(&new_ns->list); | |
2530 | init_waitqueue_head(&new_ns->poll); | |
2531 | new_ns->event = 0; | |
771b1371 | 2532 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2533 | return new_ns; |
2534 | } | |
2535 | ||
9559f689 AV |
2536 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2537 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2538 | { |
6b3286ed | 2539 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2540 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2541 | struct mount *p, *q; |
9559f689 | 2542 | struct mount *old; |
cb338d06 | 2543 | struct mount *new; |
7a472ef4 | 2544 | int copy_flags; |
1da177e4 | 2545 | |
9559f689 AV |
2546 | BUG_ON(!ns); |
2547 | ||
2548 | if (likely(!(flags & CLONE_NEWNS))) { | |
2549 | get_mnt_ns(ns); | |
2550 | return ns; | |
2551 | } | |
2552 | ||
2553 | old = ns->root; | |
2554 | ||
771b1371 | 2555 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2556 | if (IS_ERR(new_ns)) |
2557 | return new_ns; | |
1da177e4 | 2558 | |
97216be0 | 2559 | namespace_lock(); |
1da177e4 | 2560 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2561 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2562 | if (user_ns != ns->user_ns) |
132c94e3 | 2563 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2564 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2565 | if (IS_ERR(new)) { |
328e6d90 | 2566 | namespace_unlock(); |
771b1371 | 2567 | free_mnt_ns(new_ns); |
be34d1a3 | 2568 | return ERR_CAST(new); |
1da177e4 | 2569 | } |
be08d6d2 | 2570 | new_ns->root = new; |
1a4eeaf2 | 2571 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2572 | |
2573 | /* | |
2574 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2575 | * as belonging to new namespace. We have already acquired a private | |
2576 | * fs_struct, so tsk->fs->lock is not needed. | |
2577 | */ | |
909b0a88 | 2578 | p = old; |
cb338d06 | 2579 | q = new; |
1da177e4 | 2580 | while (p) { |
143c8c91 | 2581 | q->mnt_ns = new_ns; |
9559f689 AV |
2582 | if (new_fs) { |
2583 | if (&p->mnt == new_fs->root.mnt) { | |
2584 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2585 | rootmnt = &p->mnt; |
1da177e4 | 2586 | } |
9559f689 AV |
2587 | if (&p->mnt == new_fs->pwd.mnt) { |
2588 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2589 | pwdmnt = &p->mnt; |
1da177e4 | 2590 | } |
1da177e4 | 2591 | } |
909b0a88 AV |
2592 | p = next_mnt(p, old); |
2593 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2594 | if (!q) |
2595 | break; | |
2596 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2597 | p = next_mnt(p, old); | |
1da177e4 | 2598 | } |
328e6d90 | 2599 | namespace_unlock(); |
1da177e4 | 2600 | |
1da177e4 | 2601 | if (rootmnt) |
f03c6599 | 2602 | mntput(rootmnt); |
1da177e4 | 2603 | if (pwdmnt) |
f03c6599 | 2604 | mntput(pwdmnt); |
1da177e4 | 2605 | |
741a2951 | 2606 | return new_ns; |
1da177e4 LT |
2607 | } |
2608 | ||
cf8d2c11 TM |
2609 | /** |
2610 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2611 | * @mnt: pointer to the new root filesystem mountpoint | |
2612 | */ | |
1a4eeaf2 | 2613 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2614 | { |
771b1371 | 2615 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2616 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2617 | struct mount *mnt = real_mount(m); |
2618 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2619 | new_ns->root = mnt; |
b1983cd8 | 2620 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2621 | } else { |
1a4eeaf2 | 2622 | mntput(m); |
cf8d2c11 TM |
2623 | } |
2624 | return new_ns; | |
2625 | } | |
cf8d2c11 | 2626 | |
ea441d11 AV |
2627 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2628 | { | |
2629 | struct mnt_namespace *ns; | |
d31da0f0 | 2630 | struct super_block *s; |
ea441d11 AV |
2631 | struct path path; |
2632 | int err; | |
2633 | ||
2634 | ns = create_mnt_ns(mnt); | |
2635 | if (IS_ERR(ns)) | |
2636 | return ERR_CAST(ns); | |
2637 | ||
2638 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2639 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2640 | ||
2641 | put_mnt_ns(ns); | |
2642 | ||
2643 | if (err) | |
2644 | return ERR_PTR(err); | |
2645 | ||
2646 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2647 | s = path.mnt->mnt_sb; |
2648 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2649 | mntput(path.mnt); |
2650 | /* lock the sucker */ | |
d31da0f0 | 2651 | down_write(&s->s_umount); |
ea441d11 AV |
2652 | /* ... and return the root of (sub)tree on it */ |
2653 | return path.dentry; | |
2654 | } | |
2655 | EXPORT_SYMBOL(mount_subtree); | |
2656 | ||
bdc480e3 HC |
2657 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2658 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2659 | { |
eca6f534 VN |
2660 | int ret; |
2661 | char *kernel_type; | |
91a27b2a | 2662 | struct filename *kernel_dir; |
eca6f534 | 2663 | char *kernel_dev; |
1da177e4 | 2664 | unsigned long data_page; |
1da177e4 | 2665 | |
eca6f534 VN |
2666 | ret = copy_mount_string(type, &kernel_type); |
2667 | if (ret < 0) | |
2668 | goto out_type; | |
1da177e4 | 2669 | |
eca6f534 VN |
2670 | kernel_dir = getname(dir_name); |
2671 | if (IS_ERR(kernel_dir)) { | |
2672 | ret = PTR_ERR(kernel_dir); | |
2673 | goto out_dir; | |
2674 | } | |
1da177e4 | 2675 | |
eca6f534 VN |
2676 | ret = copy_mount_string(dev_name, &kernel_dev); |
2677 | if (ret < 0) | |
2678 | goto out_dev; | |
1da177e4 | 2679 | |
eca6f534 VN |
2680 | ret = copy_mount_options(data, &data_page); |
2681 | if (ret < 0) | |
2682 | goto out_data; | |
1da177e4 | 2683 | |
91a27b2a | 2684 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2685 | (void *) data_page); |
1da177e4 | 2686 | |
eca6f534 VN |
2687 | free_page(data_page); |
2688 | out_data: | |
2689 | kfree(kernel_dev); | |
2690 | out_dev: | |
2691 | putname(kernel_dir); | |
2692 | out_dir: | |
2693 | kfree(kernel_type); | |
2694 | out_type: | |
2695 | return ret; | |
1da177e4 LT |
2696 | } |
2697 | ||
afac7cba AV |
2698 | /* |
2699 | * Return true if path is reachable from root | |
2700 | * | |
48a066e7 | 2701 | * namespace_sem or mount_lock is held |
afac7cba | 2702 | */ |
643822b4 | 2703 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2704 | const struct path *root) |
2705 | { | |
643822b4 | 2706 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2707 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2708 | mnt = mnt->mnt_parent; |
afac7cba | 2709 | } |
643822b4 | 2710 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2711 | } |
2712 | ||
2713 | int path_is_under(struct path *path1, struct path *path2) | |
2714 | { | |
2715 | int res; | |
48a066e7 | 2716 | read_seqlock_excl(&mount_lock); |
643822b4 | 2717 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2718 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2719 | return res; |
2720 | } | |
2721 | EXPORT_SYMBOL(path_is_under); | |
2722 | ||
1da177e4 LT |
2723 | /* |
2724 | * pivot_root Semantics: | |
2725 | * Moves the root file system of the current process to the directory put_old, | |
2726 | * makes new_root as the new root file system of the current process, and sets | |
2727 | * root/cwd of all processes which had them on the current root to new_root. | |
2728 | * | |
2729 | * Restrictions: | |
2730 | * The new_root and put_old must be directories, and must not be on the | |
2731 | * same file system as the current process root. The put_old must be | |
2732 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2733 | * pointed to by put_old must yield the same directory as new_root. No other | |
2734 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2735 | * | |
4a0d11fa NB |
2736 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2737 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2738 | * in this situation. | |
2739 | * | |
1da177e4 LT |
2740 | * Notes: |
2741 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2742 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2743 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2744 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2745 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2746 | * first. | |
2747 | */ | |
3480b257 HC |
2748 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2749 | const char __user *, put_old) | |
1da177e4 | 2750 | { |
2d8f3038 | 2751 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2752 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2753 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2754 | int error; |
2755 | ||
9b40bc90 | 2756 | if (!may_mount()) |
1da177e4 LT |
2757 | return -EPERM; |
2758 | ||
2d8f3038 | 2759 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2760 | if (error) |
2761 | goto out0; | |
1da177e4 | 2762 | |
2d8f3038 | 2763 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2764 | if (error) |
2765 | goto out1; | |
2766 | ||
2d8f3038 | 2767 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2768 | if (error) |
2769 | goto out2; | |
1da177e4 | 2770 | |
f7ad3c6b | 2771 | get_fs_root(current->fs, &root); |
84d17192 AV |
2772 | old_mp = lock_mount(&old); |
2773 | error = PTR_ERR(old_mp); | |
2774 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2775 | goto out3; |
2776 | ||
1da177e4 | 2777 | error = -EINVAL; |
419148da AV |
2778 | new_mnt = real_mount(new.mnt); |
2779 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2780 | old_mnt = real_mount(old.mnt); |
2781 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2782 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2783 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2784 | goto out4; |
143c8c91 | 2785 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2786 | goto out4; |
5ff9d8a6 EB |
2787 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2788 | goto out4; | |
1da177e4 | 2789 | error = -ENOENT; |
f3da392e | 2790 | if (d_unlinked(new.dentry)) |
b12cea91 | 2791 | goto out4; |
1da177e4 | 2792 | error = -EBUSY; |
84d17192 | 2793 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2794 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2795 | error = -EINVAL; |
8c3ee42e | 2796 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2797 | goto out4; /* not a mountpoint */ |
676da58d | 2798 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2799 | goto out4; /* not attached */ |
84d17192 | 2800 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2801 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2802 | goto out4; /* not a mountpoint */ |
676da58d | 2803 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2804 | goto out4; /* not attached */ |
4ac91378 | 2805 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2806 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2807 | goto out4; |
84d17192 | 2808 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2809 | lock_mount_hash(); |
419148da AV |
2810 | detach_mnt(new_mnt, &parent_path); |
2811 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2812 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2813 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2814 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2815 | } | |
4ac91378 | 2816 | /* mount old root on put_old */ |
84d17192 | 2817 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2818 | /* mount new_root on / */ |
84d17192 | 2819 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2820 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2821 | unlock_mount_hash(); |
2d8f3038 | 2822 | chroot_fs_refs(&root, &new); |
84d17192 | 2823 | put_mountpoint(root_mp); |
1da177e4 | 2824 | error = 0; |
b12cea91 | 2825 | out4: |
84d17192 | 2826 | unlock_mount(old_mp); |
b12cea91 AV |
2827 | if (!error) { |
2828 | path_put(&root_parent); | |
2829 | path_put(&parent_path); | |
2830 | } | |
2831 | out3: | |
8c3ee42e | 2832 | path_put(&root); |
b12cea91 | 2833 | out2: |
2d8f3038 | 2834 | path_put(&old); |
1da177e4 | 2835 | out1: |
2d8f3038 | 2836 | path_put(&new); |
1da177e4 | 2837 | out0: |
1da177e4 | 2838 | return error; |
1da177e4 LT |
2839 | } |
2840 | ||
2841 | static void __init init_mount_tree(void) | |
2842 | { | |
2843 | struct vfsmount *mnt; | |
6b3286ed | 2844 | struct mnt_namespace *ns; |
ac748a09 | 2845 | struct path root; |
0c55cfc4 | 2846 | struct file_system_type *type; |
1da177e4 | 2847 | |
0c55cfc4 EB |
2848 | type = get_fs_type("rootfs"); |
2849 | if (!type) | |
2850 | panic("Can't find rootfs type"); | |
2851 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2852 | put_filesystem(type); | |
1da177e4 LT |
2853 | if (IS_ERR(mnt)) |
2854 | panic("Can't create rootfs"); | |
b3e19d92 | 2855 | |
3b22edc5 TM |
2856 | ns = create_mnt_ns(mnt); |
2857 | if (IS_ERR(ns)) | |
1da177e4 | 2858 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2859 | |
2860 | init_task.nsproxy->mnt_ns = ns; | |
2861 | get_mnt_ns(ns); | |
2862 | ||
be08d6d2 AV |
2863 | root.mnt = mnt; |
2864 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2865 | |
2866 | set_fs_pwd(current->fs, &root); | |
2867 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2868 | } |
2869 | ||
74bf17cf | 2870 | void __init mnt_init(void) |
1da177e4 | 2871 | { |
13f14b4d | 2872 | unsigned u; |
15a67dd8 | 2873 | int err; |
1da177e4 | 2874 | |
7d6fec45 | 2875 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2876 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2877 | |
0818bf27 | 2878 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2879 | sizeof(struct hlist_head), |
0818bf27 AV |
2880 | mhash_entries, 19, |
2881 | 0, | |
2882 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2883 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2884 | sizeof(struct hlist_head), | |
2885 | mphash_entries, 19, | |
2886 | 0, | |
2887 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2888 | |
84d17192 | 2889 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2890 | panic("Failed to allocate mount hash table\n"); |
2891 | ||
0818bf27 | 2892 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2893 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2894 | for (u = 0; u <= mp_hash_mask; u++) |
2895 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2896 | |
4b93dc9b TH |
2897 | kernfs_init(); |
2898 | ||
15a67dd8 RD |
2899 | err = sysfs_init(); |
2900 | if (err) | |
2901 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2902 | __func__, err); |
00d26666 GKH |
2903 | fs_kobj = kobject_create_and_add("fs", NULL); |
2904 | if (!fs_kobj) | |
8e24eea7 | 2905 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2906 | init_rootfs(); |
2907 | init_mount_tree(); | |
2908 | } | |
2909 | ||
616511d0 | 2910 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2911 | { |
d498b25a | 2912 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2913 | return; |
7b00ed6f | 2914 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2915 | free_mnt_ns(ns); |
1da177e4 | 2916 | } |
9d412a43 AV |
2917 | |
2918 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2919 | { | |
423e0ab0 TC |
2920 | struct vfsmount *mnt; |
2921 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2922 | if (!IS_ERR(mnt)) { | |
2923 | /* | |
2924 | * it is a longterm mount, don't release mnt until | |
2925 | * we unmount before file sys is unregistered | |
2926 | */ | |
f7a99c5b | 2927 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2928 | } |
2929 | return mnt; | |
9d412a43 AV |
2930 | } |
2931 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2932 | |
2933 | void kern_unmount(struct vfsmount *mnt) | |
2934 | { | |
2935 | /* release long term mount so mount point can be released */ | |
2936 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2937 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2938 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2939 | mntput(mnt); |
2940 | } | |
2941 | } | |
2942 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2943 | |
2944 | bool our_mnt(struct vfsmount *mnt) | |
2945 | { | |
143c8c91 | 2946 | return check_mnt(real_mount(mnt)); |
02125a82 | 2947 | } |
8823c079 | 2948 | |
3151527e EB |
2949 | bool current_chrooted(void) |
2950 | { | |
2951 | /* Does the current process have a non-standard root */ | |
2952 | struct path ns_root; | |
2953 | struct path fs_root; | |
2954 | bool chrooted; | |
2955 | ||
2956 | /* Find the namespace root */ | |
2957 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2958 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2959 | path_get(&ns_root); | |
2960 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2961 | ; | |
2962 | ||
2963 | get_fs_root(current->fs, &fs_root); | |
2964 | ||
2965 | chrooted = !path_equal(&fs_root, &ns_root); | |
2966 | ||
2967 | path_put(&fs_root); | |
2968 | path_put(&ns_root); | |
2969 | ||
2970 | return chrooted; | |
2971 | } | |
2972 | ||
e51db735 | 2973 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
2974 | { |
2975 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2976 | struct mount *mnt; | |
e51db735 | 2977 | bool visible = false; |
87a8ebd6 | 2978 | |
e51db735 EB |
2979 | if (unlikely(!ns)) |
2980 | return false; | |
2981 | ||
44bb4385 | 2982 | down_read(&namespace_sem); |
87a8ebd6 | 2983 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
2984 | struct mount *child; |
2985 | if (mnt->mnt.mnt_sb->s_type != type) | |
2986 | continue; | |
2987 | ||
2988 | /* This mount is not fully visible if there are any child mounts | |
2989 | * that cover anything except for empty directories. | |
2990 | */ | |
2991 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2992 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
2993 | if (!S_ISDIR(inode->i_mode)) | |
2994 | goto next; | |
41301ae7 | 2995 | if (inode->i_nlink > 2) |
e51db735 | 2996 | goto next; |
87a8ebd6 | 2997 | } |
e51db735 EB |
2998 | visible = true; |
2999 | goto found; | |
3000 | next: ; | |
87a8ebd6 | 3001 | } |
e51db735 | 3002 | found: |
44bb4385 | 3003 | up_read(&namespace_sem); |
e51db735 | 3004 | return visible; |
87a8ebd6 EB |
3005 | } |
3006 | ||
8823c079 EB |
3007 | static void *mntns_get(struct task_struct *task) |
3008 | { | |
3009 | struct mnt_namespace *ns = NULL; | |
3010 | struct nsproxy *nsproxy; | |
3011 | ||
728dba3a EB |
3012 | task_lock(task); |
3013 | nsproxy = task->nsproxy; | |
8823c079 EB |
3014 | if (nsproxy) { |
3015 | ns = nsproxy->mnt_ns; | |
3016 | get_mnt_ns(ns); | |
3017 | } | |
728dba3a | 3018 | task_unlock(task); |
8823c079 EB |
3019 | |
3020 | return ns; | |
3021 | } | |
3022 | ||
3023 | static void mntns_put(void *ns) | |
3024 | { | |
3025 | put_mnt_ns(ns); | |
3026 | } | |
3027 | ||
3028 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
3029 | { | |
3030 | struct fs_struct *fs = current->fs; | |
3031 | struct mnt_namespace *mnt_ns = ns; | |
3032 | struct path root; | |
3033 | ||
0c55cfc4 | 3034 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3035 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3036 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3037 | return -EPERM; |
8823c079 EB |
3038 | |
3039 | if (fs->users != 1) | |
3040 | return -EINVAL; | |
3041 | ||
3042 | get_mnt_ns(mnt_ns); | |
3043 | put_mnt_ns(nsproxy->mnt_ns); | |
3044 | nsproxy->mnt_ns = mnt_ns; | |
3045 | ||
3046 | /* Find the root */ | |
3047 | root.mnt = &mnt_ns->root->mnt; | |
3048 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3049 | path_get(&root); | |
3050 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3051 | ; | |
3052 | ||
3053 | /* Update the pwd and root */ | |
3054 | set_fs_pwd(fs, &root); | |
3055 | set_fs_root(fs, &root); | |
3056 | ||
3057 | path_put(&root); | |
3058 | return 0; | |
3059 | } | |
3060 | ||
98f842e6 EB |
3061 | static unsigned int mntns_inum(void *ns) |
3062 | { | |
3063 | struct mnt_namespace *mnt_ns = ns; | |
3064 | return mnt_ns->proc_inum; | |
3065 | } | |
3066 | ||
8823c079 EB |
3067 | const struct proc_ns_operations mntns_operations = { |
3068 | .name = "mnt", | |
3069 | .type = CLONE_NEWNS, | |
3070 | .get = mntns_get, | |
3071 | .put = mntns_put, | |
3072 | .install = mntns_install, | |
98f842e6 | 3073 | .inum = mntns_inum, |
8823c079 | 3074 | }; |