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 | |
5addc5dd | 55 | static int 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 | { |
496ad9aa | 417 | struct inode *inode = file_inode(file); |
eb04c282 | 418 | |
2d8dd38a | 419 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) |
eb04c282 | 420 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 421 | else |
422 | return mnt_clone_write(file->f_path.mnt); | |
423 | } | |
eb04c282 JK |
424 | |
425 | /** | |
426 | * mnt_want_write_file - get write access to a file's mount | |
427 | * @file: the file who's mount on which to take a write | |
428 | * | |
429 | * This is like mnt_want_write, but it takes a file and can | |
430 | * do some optimisations if the file is open for write already | |
431 | */ | |
432 | int mnt_want_write_file(struct file *file) | |
433 | { | |
434 | int ret; | |
435 | ||
436 | sb_start_write(file->f_path.mnt->mnt_sb); | |
437 | ret = __mnt_want_write_file(file); | |
438 | if (ret) | |
439 | sb_end_write(file->f_path.mnt->mnt_sb); | |
440 | return ret; | |
441 | } | |
96029c4e | 442 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
443 | ||
8366025e | 444 | /** |
eb04c282 | 445 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
446 | * @mnt: the mount on which to give up write access |
447 | * | |
448 | * Tells the low-level filesystem that we are done | |
449 | * performing writes to it. Must be matched with | |
eb04c282 | 450 | * __mnt_want_write() call above. |
8366025e | 451 | */ |
eb04c282 | 452 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 453 | { |
d3ef3d73 | 454 | preempt_disable(); |
83adc753 | 455 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 456 | preempt_enable(); |
8366025e | 457 | } |
eb04c282 JK |
458 | |
459 | /** | |
460 | * mnt_drop_write - give up write access to a mount | |
461 | * @mnt: the mount on which to give up write access | |
462 | * | |
463 | * Tells the low-level filesystem that we are done performing writes to it and | |
464 | * also allows filesystem to be frozen again. Must be matched with | |
465 | * mnt_want_write() call above. | |
466 | */ | |
467 | void mnt_drop_write(struct vfsmount *mnt) | |
468 | { | |
469 | __mnt_drop_write(mnt); | |
470 | sb_end_write(mnt->mnt_sb); | |
471 | } | |
8366025e DH |
472 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
473 | ||
eb04c282 JK |
474 | void __mnt_drop_write_file(struct file *file) |
475 | { | |
476 | __mnt_drop_write(file->f_path.mnt); | |
477 | } | |
478 | ||
2a79f17e AV |
479 | void mnt_drop_write_file(struct file *file) |
480 | { | |
481 | mnt_drop_write(file->f_path.mnt); | |
482 | } | |
483 | EXPORT_SYMBOL(mnt_drop_write_file); | |
484 | ||
83adc753 | 485 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 486 | { |
3d733633 DH |
487 | int ret = 0; |
488 | ||
719ea2fb | 489 | lock_mount_hash(); |
83adc753 | 490 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 491 | /* |
d3ef3d73 | 492 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
493 | * should be visible before we do. | |
3d733633 | 494 | */ |
d3ef3d73 | 495 | smp_mb(); |
496 | ||
3d733633 | 497 | /* |
d3ef3d73 | 498 | * With writers on hold, if this value is zero, then there are |
499 | * definitely no active writers (although held writers may subsequently | |
500 | * increment the count, they'll have to wait, and decrement it after | |
501 | * seeing MNT_READONLY). | |
502 | * | |
503 | * It is OK to have counter incremented on one CPU and decremented on | |
504 | * another: the sum will add up correctly. The danger would be when we | |
505 | * sum up each counter, if we read a counter before it is incremented, | |
506 | * but then read another CPU's count which it has been subsequently | |
507 | * decremented from -- we would see more decrements than we should. | |
508 | * MNT_WRITE_HOLD protects against this scenario, because | |
509 | * mnt_want_write first increments count, then smp_mb, then spins on | |
510 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
511 | * we're counting up here. | |
3d733633 | 512 | */ |
c6653a83 | 513 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 514 | ret = -EBUSY; |
515 | else | |
83adc753 | 516 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 517 | /* |
518 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
519 | * that become unheld will see MNT_READONLY. | |
520 | */ | |
521 | smp_wmb(); | |
83adc753 | 522 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 523 | unlock_mount_hash(); |
3d733633 | 524 | return ret; |
8366025e | 525 | } |
8366025e | 526 | |
83adc753 | 527 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 528 | { |
719ea2fb | 529 | lock_mount_hash(); |
83adc753 | 530 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 531 | unlock_mount_hash(); |
2e4b7fcd DH |
532 | } |
533 | ||
4ed5e82f MS |
534 | int sb_prepare_remount_readonly(struct super_block *sb) |
535 | { | |
536 | struct mount *mnt; | |
537 | int err = 0; | |
538 | ||
8e8b8796 MS |
539 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
540 | if (atomic_long_read(&sb->s_remove_count)) | |
541 | return -EBUSY; | |
542 | ||
719ea2fb | 543 | lock_mount_hash(); |
4ed5e82f MS |
544 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
545 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
546 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
547 | smp_mb(); | |
548 | if (mnt_get_writers(mnt) > 0) { | |
549 | err = -EBUSY; | |
550 | break; | |
551 | } | |
552 | } | |
553 | } | |
8e8b8796 MS |
554 | if (!err && atomic_long_read(&sb->s_remove_count)) |
555 | err = -EBUSY; | |
556 | ||
4ed5e82f MS |
557 | if (!err) { |
558 | sb->s_readonly_remount = 1; | |
559 | smp_wmb(); | |
560 | } | |
561 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
562 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
563 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
564 | } | |
719ea2fb | 565 | unlock_mount_hash(); |
4ed5e82f MS |
566 | |
567 | return err; | |
568 | } | |
569 | ||
b105e270 | 570 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 571 | { |
52ba1621 | 572 | kfree(mnt->mnt_devname); |
73cd49ec | 573 | mnt_free_id(mnt); |
d3ef3d73 | 574 | #ifdef CONFIG_SMP |
68e8a9fe | 575 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 576 | #endif |
b105e270 | 577 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
578 | } |
579 | ||
48a066e7 AV |
580 | /* call under rcu_read_lock */ |
581 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
582 | { | |
583 | struct mount *mnt; | |
584 | if (read_seqretry(&mount_lock, seq)) | |
585 | return false; | |
586 | if (bastard == NULL) | |
587 | return true; | |
588 | mnt = real_mount(bastard); | |
589 | mnt_add_count(mnt, 1); | |
590 | if (likely(!read_seqretry(&mount_lock, seq))) | |
591 | return true; | |
592 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
593 | mnt_add_count(mnt, -1); | |
594 | return false; | |
595 | } | |
596 | rcu_read_unlock(); | |
597 | mntput(bastard); | |
598 | rcu_read_lock(); | |
599 | return false; | |
600 | } | |
601 | ||
1da177e4 | 602 | /* |
474279dc | 603 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 604 | * call under rcu_read_lock() |
1da177e4 | 605 | */ |
474279dc | 606 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 607 | { |
38129a13 | 608 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
609 | struct mount *p; |
610 | ||
38129a13 | 611 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
612 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
613 | return p; | |
614 | return NULL; | |
615 | } | |
616 | ||
617 | /* | |
618 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 619 | * mount_lock must be held. |
474279dc AV |
620 | */ |
621 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
622 | { | |
38129a13 AV |
623 | struct mount *p, *res; |
624 | res = p = __lookup_mnt(mnt, dentry); | |
625 | if (!p) | |
626 | goto out; | |
627 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
628 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
629 | break; | |
630 | res = p; | |
631 | } | |
38129a13 | 632 | out: |
1d6a32ac | 633 | return res; |
1da177e4 LT |
634 | } |
635 | ||
a05964f3 | 636 | /* |
f015f126 DH |
637 | * lookup_mnt - Return the first child mount mounted at path |
638 | * | |
639 | * "First" means first mounted chronologically. If you create the | |
640 | * following mounts: | |
641 | * | |
642 | * mount /dev/sda1 /mnt | |
643 | * mount /dev/sda2 /mnt | |
644 | * mount /dev/sda3 /mnt | |
645 | * | |
646 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
647 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
648 | * /dev/sda2, then /dev/sda3, then NULL. | |
649 | * | |
650 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 651 | */ |
1c755af4 | 652 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 653 | { |
c7105365 | 654 | struct mount *child_mnt; |
48a066e7 AV |
655 | struct vfsmount *m; |
656 | unsigned seq; | |
99b7db7b | 657 | |
48a066e7 AV |
658 | rcu_read_lock(); |
659 | do { | |
660 | seq = read_seqbegin(&mount_lock); | |
661 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
662 | m = child_mnt ? &child_mnt->mnt : NULL; | |
663 | } while (!legitimize_mnt(m, seq)); | |
664 | rcu_read_unlock(); | |
665 | return m; | |
a05964f3 RP |
666 | } |
667 | ||
84d17192 AV |
668 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
669 | { | |
0818bf27 | 670 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 | 671 | struct mountpoint *mp; |
eed81007 | 672 | int ret; |
84d17192 | 673 | |
0818bf27 | 674 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
675 | if (mp->m_dentry == dentry) { |
676 | /* might be worth a WARN_ON() */ | |
677 | if (d_unlinked(dentry)) | |
678 | return ERR_PTR(-ENOENT); | |
679 | mp->m_count++; | |
680 | return mp; | |
681 | } | |
682 | } | |
683 | ||
684 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
685 | if (!mp) | |
686 | return ERR_PTR(-ENOMEM); | |
687 | ||
eed81007 MS |
688 | ret = d_set_mounted(dentry); |
689 | if (ret) { | |
84d17192 | 690 | kfree(mp); |
eed81007 | 691 | return ERR_PTR(ret); |
84d17192 | 692 | } |
eed81007 | 693 | |
84d17192 AV |
694 | mp->m_dentry = dentry; |
695 | mp->m_count = 1; | |
0818bf27 | 696 | hlist_add_head(&mp->m_hash, chain); |
84d17192 AV |
697 | return mp; |
698 | } | |
699 | ||
700 | static void put_mountpoint(struct mountpoint *mp) | |
701 | { | |
702 | if (!--mp->m_count) { | |
703 | struct dentry *dentry = mp->m_dentry; | |
704 | spin_lock(&dentry->d_lock); | |
705 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
706 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 707 | hlist_del(&mp->m_hash); |
84d17192 AV |
708 | kfree(mp); |
709 | } | |
710 | } | |
711 | ||
143c8c91 | 712 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 713 | { |
6b3286ed | 714 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
715 | } |
716 | ||
99b7db7b NP |
717 | /* |
718 | * vfsmount lock must be held for write | |
719 | */ | |
6b3286ed | 720 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
721 | { |
722 | if (ns) { | |
723 | ns->event = ++event; | |
724 | wake_up_interruptible(&ns->poll); | |
725 | } | |
726 | } | |
727 | ||
99b7db7b NP |
728 | /* |
729 | * vfsmount lock must be held for write | |
730 | */ | |
6b3286ed | 731 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
732 | { |
733 | if (ns && ns->event != event) { | |
734 | ns->event = event; | |
735 | wake_up_interruptible(&ns->poll); | |
736 | } | |
737 | } | |
738 | ||
99b7db7b NP |
739 | /* |
740 | * vfsmount lock must be held for write | |
741 | */ | |
419148da AV |
742 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
743 | { | |
a73324da | 744 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
745 | old_path->mnt = &mnt->mnt_parent->mnt; |
746 | mnt->mnt_parent = mnt; | |
a73324da | 747 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 748 | list_del_init(&mnt->mnt_child); |
38129a13 | 749 | hlist_del_init_rcu(&mnt->mnt_hash); |
84d17192 AV |
750 | put_mountpoint(mnt->mnt_mp); |
751 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
752 | } |
753 | ||
99b7db7b NP |
754 | /* |
755 | * vfsmount lock must be held for write | |
756 | */ | |
84d17192 AV |
757 | void mnt_set_mountpoint(struct mount *mnt, |
758 | struct mountpoint *mp, | |
44d964d6 | 759 | struct mount *child_mnt) |
b90fa9ae | 760 | { |
84d17192 | 761 | mp->m_count++; |
3a2393d7 | 762 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 763 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 764 | child_mnt->mnt_parent = mnt; |
84d17192 | 765 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
766 | } |
767 | ||
99b7db7b NP |
768 | /* |
769 | * vfsmount lock must be held for write | |
770 | */ | |
84d17192 AV |
771 | static void attach_mnt(struct mount *mnt, |
772 | struct mount *parent, | |
773 | struct mountpoint *mp) | |
1da177e4 | 774 | { |
84d17192 | 775 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 776 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 777 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
778 | } |
779 | ||
780 | /* | |
99b7db7b | 781 | * vfsmount lock must be held for write |
b90fa9ae | 782 | */ |
1d6a32ac | 783 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 784 | { |
0714a533 | 785 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 786 | struct mount *m; |
b90fa9ae | 787 | LIST_HEAD(head); |
143c8c91 | 788 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 789 | |
0714a533 | 790 | BUG_ON(parent == mnt); |
b90fa9ae | 791 | |
1a4eeaf2 | 792 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 793 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 794 | m->mnt_ns = n; |
f03c6599 | 795 | |
b90fa9ae RP |
796 | list_splice(&head, n->list.prev); |
797 | ||
1d6a32ac | 798 | if (shadows) |
38129a13 | 799 | hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash); |
1d6a32ac | 800 | else |
38129a13 | 801 | hlist_add_head_rcu(&mnt->mnt_hash, |
0818bf27 | 802 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 803 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 804 | touch_mnt_namespace(n); |
1da177e4 LT |
805 | } |
806 | ||
909b0a88 | 807 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 808 | { |
6b41d536 AV |
809 | struct list_head *next = p->mnt_mounts.next; |
810 | if (next == &p->mnt_mounts) { | |
1da177e4 | 811 | while (1) { |
909b0a88 | 812 | if (p == root) |
1da177e4 | 813 | return NULL; |
6b41d536 AV |
814 | next = p->mnt_child.next; |
815 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 816 | break; |
0714a533 | 817 | p = p->mnt_parent; |
1da177e4 LT |
818 | } |
819 | } | |
6b41d536 | 820 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
821 | } |
822 | ||
315fc83e | 823 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 824 | { |
6b41d536 AV |
825 | struct list_head *prev = p->mnt_mounts.prev; |
826 | while (prev != &p->mnt_mounts) { | |
827 | p = list_entry(prev, struct mount, mnt_child); | |
828 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
829 | } |
830 | return p; | |
831 | } | |
832 | ||
9d412a43 AV |
833 | struct vfsmount * |
834 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
835 | { | |
b105e270 | 836 | struct mount *mnt; |
9d412a43 AV |
837 | struct dentry *root; |
838 | ||
839 | if (!type) | |
840 | return ERR_PTR(-ENODEV); | |
841 | ||
842 | mnt = alloc_vfsmnt(name); | |
843 | if (!mnt) | |
844 | return ERR_PTR(-ENOMEM); | |
845 | ||
846 | if (flags & MS_KERNMOUNT) | |
b105e270 | 847 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
848 | |
849 | root = mount_fs(type, flags, name, data); | |
850 | if (IS_ERR(root)) { | |
851 | free_vfsmnt(mnt); | |
852 | return ERR_CAST(root); | |
853 | } | |
854 | ||
b105e270 AV |
855 | mnt->mnt.mnt_root = root; |
856 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 857 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 858 | mnt->mnt_parent = mnt; |
719ea2fb | 859 | lock_mount_hash(); |
39f7c4db | 860 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 861 | unlock_mount_hash(); |
b105e270 | 862 | return &mnt->mnt; |
9d412a43 AV |
863 | } |
864 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
865 | ||
87129cc0 | 866 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 867 | int flag) |
1da177e4 | 868 | { |
87129cc0 | 869 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
870 | struct mount *mnt; |
871 | int err; | |
1da177e4 | 872 | |
be34d1a3 DH |
873 | mnt = alloc_vfsmnt(old->mnt_devname); |
874 | if (!mnt) | |
875 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 876 | |
7a472ef4 | 877 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
878 | mnt->mnt_group_id = 0; /* not a peer of original */ |
879 | else | |
880 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 881 | |
be34d1a3 DH |
882 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
883 | err = mnt_alloc_group_id(mnt); | |
884 | if (err) | |
885 | goto out_free; | |
1da177e4 | 886 | } |
be34d1a3 DH |
887 | |
888 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
132c94e3 EB |
889 | /* Don't allow unprivileged users to change mount flags */ |
890 | if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) | |
891 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
892 | ||
5ff9d8a6 EB |
893 | /* Don't allow unprivileged users to reveal what is under a mount */ |
894 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
895 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
896 | ||
be34d1a3 DH |
897 | atomic_inc(&sb->s_active); |
898 | mnt->mnt.mnt_sb = sb; | |
899 | mnt->mnt.mnt_root = dget(root); | |
900 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
901 | mnt->mnt_parent = mnt; | |
719ea2fb | 902 | lock_mount_hash(); |
be34d1a3 | 903 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 904 | unlock_mount_hash(); |
be34d1a3 | 905 | |
7a472ef4 EB |
906 | if ((flag & CL_SLAVE) || |
907 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
908 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
909 | mnt->mnt_master = old; | |
910 | CLEAR_MNT_SHARED(mnt); | |
911 | } else if (!(flag & CL_PRIVATE)) { | |
912 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
913 | list_add(&mnt->mnt_share, &old->mnt_share); | |
914 | if (IS_MNT_SLAVE(old)) | |
915 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
916 | mnt->mnt_master = old->mnt_master; | |
917 | } | |
918 | if (flag & CL_MAKE_SHARED) | |
919 | set_mnt_shared(mnt); | |
920 | ||
921 | /* stick the duplicate mount on the same expiry list | |
922 | * as the original if that was on one */ | |
923 | if (flag & CL_EXPIRE) { | |
924 | if (!list_empty(&old->mnt_expire)) | |
925 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
926 | } | |
927 | ||
cb338d06 | 928 | return mnt; |
719f5d7f MS |
929 | |
930 | out_free: | |
931 | free_vfsmnt(mnt); | |
be34d1a3 | 932 | return ERR_PTR(err); |
1da177e4 LT |
933 | } |
934 | ||
48a066e7 AV |
935 | static void delayed_free(struct rcu_head *head) |
936 | { | |
937 | struct mount *mnt = container_of(head, struct mount, mnt_rcu); | |
938 | kfree(mnt->mnt_devname); | |
939 | #ifdef CONFIG_SMP | |
940 | free_percpu(mnt->mnt_pcp); | |
941 | #endif | |
942 | kmem_cache_free(mnt_cache, mnt); | |
943 | } | |
944 | ||
900148dc | 945 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 946 | { |
b3e19d92 | 947 | put_again: |
48a066e7 AV |
948 | rcu_read_lock(); |
949 | mnt_add_count(mnt, -1); | |
950 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
951 | rcu_read_unlock(); | |
f03c6599 | 952 | return; |
b3e19d92 | 953 | } |
719ea2fb | 954 | lock_mount_hash(); |
b3e19d92 | 955 | if (mnt_get_count(mnt)) { |
48a066e7 | 956 | rcu_read_unlock(); |
719ea2fb | 957 | unlock_mount_hash(); |
99b7db7b NP |
958 | return; |
959 | } | |
863d684f AV |
960 | if (unlikely(mnt->mnt_pinned)) { |
961 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
962 | mnt->mnt_pinned = 0; | |
48a066e7 | 963 | rcu_read_unlock(); |
719ea2fb | 964 | unlock_mount_hash(); |
900148dc | 965 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 966 | goto put_again; |
7b7b1ace | 967 | } |
48a066e7 AV |
968 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
969 | rcu_read_unlock(); | |
970 | unlock_mount_hash(); | |
971 | return; | |
972 | } | |
973 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
974 | rcu_read_unlock(); | |
962830df | 975 | |
39f7c4db | 976 | list_del(&mnt->mnt_instance); |
719ea2fb | 977 | unlock_mount_hash(); |
649a795a AV |
978 | |
979 | /* | |
980 | * This probably indicates that somebody messed | |
981 | * up a mnt_want/drop_write() pair. If this | |
982 | * happens, the filesystem was probably unable | |
983 | * to make r/w->r/o transitions. | |
984 | */ | |
985 | /* | |
986 | * The locking used to deal with mnt_count decrement provides barriers, | |
987 | * so mnt_get_writers() below is safe. | |
988 | */ | |
989 | WARN_ON(mnt_get_writers(mnt)); | |
990 | fsnotify_vfsmount_delete(&mnt->mnt); | |
991 | dput(mnt->mnt.mnt_root); | |
992 | deactivate_super(mnt->mnt.mnt_sb); | |
48a066e7 AV |
993 | mnt_free_id(mnt); |
994 | call_rcu(&mnt->mnt_rcu, delayed_free); | |
b3e19d92 | 995 | } |
b3e19d92 NP |
996 | |
997 | void mntput(struct vfsmount *mnt) | |
998 | { | |
999 | if (mnt) { | |
863d684f | 1000 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1001 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1002 | if (unlikely(m->mnt_expiry_mark)) |
1003 | m->mnt_expiry_mark = 0; | |
1004 | mntput_no_expire(m); | |
b3e19d92 NP |
1005 | } |
1006 | } | |
1007 | EXPORT_SYMBOL(mntput); | |
1008 | ||
1009 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1010 | { | |
1011 | if (mnt) | |
83adc753 | 1012 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1013 | return mnt; |
1014 | } | |
1015 | EXPORT_SYMBOL(mntget); | |
1016 | ||
7b7b1ace AV |
1017 | void mnt_pin(struct vfsmount *mnt) |
1018 | { | |
719ea2fb | 1019 | lock_mount_hash(); |
863d684f | 1020 | real_mount(mnt)->mnt_pinned++; |
719ea2fb | 1021 | unlock_mount_hash(); |
7b7b1ace | 1022 | } |
7b7b1ace AV |
1023 | EXPORT_SYMBOL(mnt_pin); |
1024 | ||
863d684f | 1025 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 1026 | { |
863d684f | 1027 | struct mount *mnt = real_mount(m); |
719ea2fb | 1028 | lock_mount_hash(); |
7b7b1ace | 1029 | if (mnt->mnt_pinned) { |
863d684f | 1030 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
1031 | mnt->mnt_pinned--; |
1032 | } | |
719ea2fb | 1033 | unlock_mount_hash(); |
7b7b1ace | 1034 | } |
7b7b1ace | 1035 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 1036 | |
b3b304a2 MS |
1037 | static inline void mangle(struct seq_file *m, const char *s) |
1038 | { | |
1039 | seq_escape(m, s, " \t\n\\"); | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Simple .show_options callback for filesystems which don't want to | |
1044 | * implement more complex mount option showing. | |
1045 | * | |
1046 | * See also save_mount_options(). | |
1047 | */ | |
34c80b1d | 1048 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1049 | { |
2a32cebd AV |
1050 | const char *options; |
1051 | ||
1052 | rcu_read_lock(); | |
34c80b1d | 1053 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1054 | |
1055 | if (options != NULL && options[0]) { | |
1056 | seq_putc(m, ','); | |
1057 | mangle(m, options); | |
1058 | } | |
2a32cebd | 1059 | rcu_read_unlock(); |
b3b304a2 MS |
1060 | |
1061 | return 0; | |
1062 | } | |
1063 | EXPORT_SYMBOL(generic_show_options); | |
1064 | ||
1065 | /* | |
1066 | * If filesystem uses generic_show_options(), this function should be | |
1067 | * called from the fill_super() callback. | |
1068 | * | |
1069 | * The .remount_fs callback usually needs to be handled in a special | |
1070 | * way, to make sure, that previous options are not overwritten if the | |
1071 | * remount fails. | |
1072 | * | |
1073 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1074 | * reset all options to their default value, but changes only newly | |
1075 | * given options, then the displayed options will not reflect reality | |
1076 | * any more. | |
1077 | */ | |
1078 | void save_mount_options(struct super_block *sb, char *options) | |
1079 | { | |
2a32cebd AV |
1080 | BUG_ON(sb->s_options); |
1081 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1082 | } |
1083 | EXPORT_SYMBOL(save_mount_options); | |
1084 | ||
2a32cebd AV |
1085 | void replace_mount_options(struct super_block *sb, char *options) |
1086 | { | |
1087 | char *old = sb->s_options; | |
1088 | rcu_assign_pointer(sb->s_options, options); | |
1089 | if (old) { | |
1090 | synchronize_rcu(); | |
1091 | kfree(old); | |
1092 | } | |
1093 | } | |
1094 | EXPORT_SYMBOL(replace_mount_options); | |
1095 | ||
a1a2c409 | 1096 | #ifdef CONFIG_PROC_FS |
0226f492 | 1097 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1098 | static void *m_start(struct seq_file *m, loff_t *pos) |
1099 | { | |
6ce6e24e | 1100 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1101 | |
390c6843 | 1102 | down_read(&namespace_sem); |
a1a2c409 | 1103 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
1104 | } |
1105 | ||
1106 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1107 | { | |
6ce6e24e | 1108 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1109 | |
a1a2c409 | 1110 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
1111 | } |
1112 | ||
1113 | static void m_stop(struct seq_file *m, void *v) | |
1114 | { | |
390c6843 | 1115 | up_read(&namespace_sem); |
1da177e4 LT |
1116 | } |
1117 | ||
0226f492 | 1118 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1119 | { |
6ce6e24e | 1120 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1121 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1122 | return p->show(m, &r->mnt); |
1da177e4 LT |
1123 | } |
1124 | ||
a1a2c409 | 1125 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1126 | .start = m_start, |
1127 | .next = m_next, | |
1128 | .stop = m_stop, | |
0226f492 | 1129 | .show = m_show, |
b4629fe2 | 1130 | }; |
a1a2c409 | 1131 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1132 | |
1da177e4 LT |
1133 | /** |
1134 | * may_umount_tree - check if a mount tree is busy | |
1135 | * @mnt: root of mount tree | |
1136 | * | |
1137 | * This is called to check if a tree of mounts has any | |
1138 | * open files, pwds, chroots or sub mounts that are | |
1139 | * busy. | |
1140 | */ | |
909b0a88 | 1141 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1142 | { |
909b0a88 | 1143 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1144 | int actual_refs = 0; |
1145 | int minimum_refs = 0; | |
315fc83e | 1146 | struct mount *p; |
909b0a88 | 1147 | BUG_ON(!m); |
1da177e4 | 1148 | |
b3e19d92 | 1149 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1150 | lock_mount_hash(); |
909b0a88 | 1151 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1152 | actual_refs += mnt_get_count(p); |
1da177e4 | 1153 | minimum_refs += 2; |
1da177e4 | 1154 | } |
719ea2fb | 1155 | unlock_mount_hash(); |
1da177e4 LT |
1156 | |
1157 | if (actual_refs > minimum_refs) | |
e3474a8e | 1158 | return 0; |
1da177e4 | 1159 | |
e3474a8e | 1160 | return 1; |
1da177e4 LT |
1161 | } |
1162 | ||
1163 | EXPORT_SYMBOL(may_umount_tree); | |
1164 | ||
1165 | /** | |
1166 | * may_umount - check if a mount point is busy | |
1167 | * @mnt: root of mount | |
1168 | * | |
1169 | * This is called to check if a mount point has any | |
1170 | * open files, pwds, chroots or sub mounts. If the | |
1171 | * mount has sub mounts this will return busy | |
1172 | * regardless of whether the sub mounts are busy. | |
1173 | * | |
1174 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1175 | * give false negatives. The main reason why it's here is that we need | |
1176 | * a non-destructive way to look for easily umountable filesystems. | |
1177 | */ | |
1178 | int may_umount(struct vfsmount *mnt) | |
1179 | { | |
e3474a8e | 1180 | int ret = 1; |
8ad08d8a | 1181 | down_read(&namespace_sem); |
719ea2fb | 1182 | lock_mount_hash(); |
1ab59738 | 1183 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1184 | ret = 0; |
719ea2fb | 1185 | unlock_mount_hash(); |
8ad08d8a | 1186 | up_read(&namespace_sem); |
a05964f3 | 1187 | return ret; |
1da177e4 LT |
1188 | } |
1189 | ||
1190 | EXPORT_SYMBOL(may_umount); | |
1191 | ||
38129a13 | 1192 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1193 | |
97216be0 | 1194 | static void namespace_unlock(void) |
70fbcdf4 | 1195 | { |
d5e50f74 | 1196 | struct mount *mnt; |
38129a13 | 1197 | struct hlist_head head = unmounted; |
97216be0 | 1198 | |
38129a13 | 1199 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1200 | up_write(&namespace_sem); |
1201 | return; | |
1202 | } | |
1203 | ||
38129a13 AV |
1204 | head.first->pprev = &head.first; |
1205 | INIT_HLIST_HEAD(&unmounted); | |
1206 | ||
97216be0 AV |
1207 | up_write(&namespace_sem); |
1208 | ||
48a066e7 AV |
1209 | synchronize_rcu(); |
1210 | ||
38129a13 AV |
1211 | while (!hlist_empty(&head)) { |
1212 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1213 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1214 | if (mnt->mnt_ex_mountpoint.mnt) |
1215 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1216 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1217 | } |
1218 | } | |
1219 | ||
97216be0 | 1220 | static inline void namespace_lock(void) |
e3197d83 | 1221 | { |
97216be0 | 1222 | down_write(&namespace_sem); |
e3197d83 AV |
1223 | } |
1224 | ||
99b7db7b | 1225 | /* |
48a066e7 | 1226 | * mount_lock must be held |
99b7db7b | 1227 | * namespace_sem must be held for write |
48a066e7 AV |
1228 | * how = 0 => just this tree, don't propagate |
1229 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1230 | * how = 2 => lazy umount | |
99b7db7b | 1231 | */ |
48a066e7 | 1232 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1233 | { |
38129a13 | 1234 | HLIST_HEAD(tmp_list); |
315fc83e | 1235 | struct mount *p; |
38129a13 | 1236 | struct mount *last = NULL; |
1da177e4 | 1237 | |
38129a13 AV |
1238 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1239 | hlist_del_init_rcu(&p->mnt_hash); | |
1240 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1241 | } | |
1da177e4 | 1242 | |
48a066e7 | 1243 | if (how) |
7b8a53fd | 1244 | propagate_umount(&tmp_list); |
a05964f3 | 1245 | |
38129a13 | 1246 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1247 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1248 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1249 | __touch_mnt_namespace(p->mnt_ns); |
1250 | p->mnt_ns = NULL; | |
48a066e7 AV |
1251 | if (how < 2) |
1252 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
6b41d536 | 1253 | list_del_init(&p->mnt_child); |
676da58d | 1254 | if (mnt_has_parent(p)) { |
84d17192 | 1255 | put_mountpoint(p->mnt_mp); |
aba809cf AV |
1256 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1257 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1258 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1259 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1260 | p->mnt_parent = p; | |
84d17192 | 1261 | p->mnt_mp = NULL; |
7c4b93d8 | 1262 | } |
0f0afb1d | 1263 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1264 | last = p; |
1265 | } | |
1266 | if (last) { | |
1267 | last->mnt_hash.next = unmounted.first; | |
1268 | unmounted.first = tmp_list.first; | |
1269 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1270 | } |
1271 | } | |
1272 | ||
b54b9be7 | 1273 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1274 | |
1ab59738 | 1275 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1276 | { |
1ab59738 | 1277 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1278 | int retval; |
1279 | ||
1ab59738 | 1280 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1281 | if (retval) |
1282 | return retval; | |
1283 | ||
1284 | /* | |
1285 | * Allow userspace to request a mountpoint be expired rather than | |
1286 | * unmounting unconditionally. Unmount only happens if: | |
1287 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1288 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1289 | */ | |
1290 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1291 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1292 | flags & (MNT_FORCE | MNT_DETACH)) |
1293 | return -EINVAL; | |
1294 | ||
b3e19d92 NP |
1295 | /* |
1296 | * probably don't strictly need the lock here if we examined | |
1297 | * all race cases, but it's a slowpath. | |
1298 | */ | |
719ea2fb | 1299 | lock_mount_hash(); |
83adc753 | 1300 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1301 | unlock_mount_hash(); |
1da177e4 | 1302 | return -EBUSY; |
b3e19d92 | 1303 | } |
719ea2fb | 1304 | unlock_mount_hash(); |
1da177e4 | 1305 | |
863d684f | 1306 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1307 | return -EAGAIN; |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * If we may have to abort operations to get out of this | |
1312 | * mount, and they will themselves hold resources we must | |
1313 | * allow the fs to do things. In the Unix tradition of | |
1314 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1315 | * might fail to complete on the first run through as other tasks | |
1316 | * must return, and the like. Thats for the mount program to worry | |
1317 | * about for the moment. | |
1318 | */ | |
1319 | ||
42faad99 | 1320 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1321 | sb->s_op->umount_begin(sb); |
42faad99 | 1322 | } |
1da177e4 LT |
1323 | |
1324 | /* | |
1325 | * No sense to grab the lock for this test, but test itself looks | |
1326 | * somewhat bogus. Suggestions for better replacement? | |
1327 | * Ho-hum... In principle, we might treat that as umount + switch | |
1328 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1329 | * Actually it makes sense, especially if rootfs would contain a | |
1330 | * /reboot - static binary that would close all descriptors and | |
1331 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1332 | */ | |
1ab59738 | 1333 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1334 | /* |
1335 | * Special case for "unmounting" root ... | |
1336 | * we just try to remount it readonly. | |
1337 | */ | |
1338 | down_write(&sb->s_umount); | |
4aa98cf7 | 1339 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1340 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1341 | up_write(&sb->s_umount); |
1342 | return retval; | |
1343 | } | |
1344 | ||
97216be0 | 1345 | namespace_lock(); |
719ea2fb | 1346 | lock_mount_hash(); |
5addc5dd | 1347 | event++; |
1da177e4 | 1348 | |
48a066e7 | 1349 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1350 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1351 | umount_tree(mnt, 2); |
1da177e4 | 1352 | retval = 0; |
48a066e7 AV |
1353 | } else { |
1354 | shrink_submounts(mnt); | |
1355 | retval = -EBUSY; | |
1356 | if (!propagate_mount_busy(mnt, 2)) { | |
1357 | if (!list_empty(&mnt->mnt_list)) | |
1358 | umount_tree(mnt, 1); | |
1359 | retval = 0; | |
1360 | } | |
1da177e4 | 1361 | } |
719ea2fb | 1362 | unlock_mount_hash(); |
e3197d83 | 1363 | namespace_unlock(); |
1da177e4 LT |
1364 | return retval; |
1365 | } | |
1366 | ||
9b40bc90 AV |
1367 | /* |
1368 | * Is the caller allowed to modify his namespace? | |
1369 | */ | |
1370 | static inline bool may_mount(void) | |
1371 | { | |
1372 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1373 | } | |
1374 | ||
1da177e4 LT |
1375 | /* |
1376 | * Now umount can handle mount points as well as block devices. | |
1377 | * This is important for filesystems which use unnamed block devices. | |
1378 | * | |
1379 | * We now support a flag for forced unmount like the other 'big iron' | |
1380 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1381 | */ | |
1382 | ||
bdc480e3 | 1383 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1384 | { |
2d8f3038 | 1385 | struct path path; |
900148dc | 1386 | struct mount *mnt; |
1da177e4 | 1387 | int retval; |
db1f05bb | 1388 | int lookup_flags = 0; |
1da177e4 | 1389 | |
db1f05bb MS |
1390 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1391 | return -EINVAL; | |
1392 | ||
9b40bc90 AV |
1393 | if (!may_mount()) |
1394 | return -EPERM; | |
1395 | ||
db1f05bb MS |
1396 | if (!(flags & UMOUNT_NOFOLLOW)) |
1397 | lookup_flags |= LOOKUP_FOLLOW; | |
1398 | ||
197df04c | 1399 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1400 | if (retval) |
1401 | goto out; | |
900148dc | 1402 | mnt = real_mount(path.mnt); |
1da177e4 | 1403 | retval = -EINVAL; |
2d8f3038 | 1404 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1405 | goto dput_and_out; |
143c8c91 | 1406 | if (!check_mnt(mnt)) |
1da177e4 | 1407 | goto dput_and_out; |
5ff9d8a6 EB |
1408 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1409 | goto dput_and_out; | |
1da177e4 | 1410 | |
900148dc | 1411 | retval = do_umount(mnt, flags); |
1da177e4 | 1412 | dput_and_out: |
429731b1 | 1413 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1414 | dput(path.dentry); |
900148dc | 1415 | mntput_no_expire(mnt); |
1da177e4 LT |
1416 | out: |
1417 | return retval; | |
1418 | } | |
1419 | ||
1420 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1421 | ||
1422 | /* | |
b58fed8b | 1423 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1424 | */ |
bdc480e3 | 1425 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1426 | { |
b58fed8b | 1427 | return sys_umount(name, 0); |
1da177e4 LT |
1428 | } |
1429 | ||
1430 | #endif | |
1431 | ||
4ce5d2b1 | 1432 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1433 | { |
4ce5d2b1 EB |
1434 | /* Is this a proxy for a mount namespace? */ |
1435 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1436 | struct proc_ns *ei; |
8823c079 EB |
1437 | |
1438 | if (!proc_ns_inode(inode)) | |
1439 | return false; | |
1440 | ||
0bb80f24 | 1441 | ei = get_proc_ns(inode); |
8823c079 EB |
1442 | if (ei->ns_ops != &mntns_operations) |
1443 | return false; | |
1444 | ||
4ce5d2b1 EB |
1445 | return true; |
1446 | } | |
1447 | ||
1448 | static bool mnt_ns_loop(struct dentry *dentry) | |
1449 | { | |
1450 | /* Could bind mounting the mount namespace inode cause a | |
1451 | * mount namespace loop? | |
1452 | */ | |
1453 | struct mnt_namespace *mnt_ns; | |
1454 | if (!is_mnt_ns_file(dentry)) | |
1455 | return false; | |
1456 | ||
1457 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1458 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1459 | } | |
1460 | ||
87129cc0 | 1461 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1462 | int flag) |
1da177e4 | 1463 | { |
84d17192 | 1464 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1465 | |
4ce5d2b1 EB |
1466 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1467 | return ERR_PTR(-EINVAL); | |
1468 | ||
1469 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1470 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1471 | |
36341f64 | 1472 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1473 | if (IS_ERR(q)) |
1474 | return q; | |
1475 | ||
5ff9d8a6 | 1476 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1477 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1478 | |
1479 | p = mnt; | |
6b41d536 | 1480 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1481 | struct mount *s; |
7ec02ef1 | 1482 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1483 | continue; |
1484 | ||
909b0a88 | 1485 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1486 | if (!(flag & CL_COPY_UNBINDABLE) && |
1487 | IS_MNT_UNBINDABLE(s)) { | |
1488 | s = skip_mnt_tree(s); | |
1489 | continue; | |
1490 | } | |
1491 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1492 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1493 | s = skip_mnt_tree(s); |
1494 | continue; | |
1495 | } | |
0714a533 AV |
1496 | while (p != s->mnt_parent) { |
1497 | p = p->mnt_parent; | |
1498 | q = q->mnt_parent; | |
1da177e4 | 1499 | } |
87129cc0 | 1500 | p = s; |
84d17192 | 1501 | parent = q; |
87129cc0 | 1502 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1503 | if (IS_ERR(q)) |
1504 | goto out; | |
719ea2fb | 1505 | lock_mount_hash(); |
1a4eeaf2 | 1506 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1507 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1508 | unlock_mount_hash(); |
1da177e4 LT |
1509 | } |
1510 | } | |
1511 | return res; | |
be34d1a3 | 1512 | out: |
1da177e4 | 1513 | if (res) { |
719ea2fb | 1514 | lock_mount_hash(); |
328e6d90 | 1515 | umount_tree(res, 0); |
719ea2fb | 1516 | unlock_mount_hash(); |
1da177e4 | 1517 | } |
be34d1a3 | 1518 | return q; |
1da177e4 LT |
1519 | } |
1520 | ||
be34d1a3 DH |
1521 | /* Caller should check returned pointer for errors */ |
1522 | ||
589ff870 | 1523 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1524 | { |
cb338d06 | 1525 | struct mount *tree; |
97216be0 | 1526 | namespace_lock(); |
87129cc0 AV |
1527 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1528 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1529 | namespace_unlock(); |
be34d1a3 | 1530 | if (IS_ERR(tree)) |
52e220d3 | 1531 | return ERR_CAST(tree); |
be34d1a3 | 1532 | return &tree->mnt; |
8aec0809 AV |
1533 | } |
1534 | ||
1535 | void drop_collected_mounts(struct vfsmount *mnt) | |
1536 | { | |
97216be0 | 1537 | namespace_lock(); |
719ea2fb | 1538 | lock_mount_hash(); |
328e6d90 | 1539 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1540 | unlock_mount_hash(); |
3ab6abee | 1541 | namespace_unlock(); |
8aec0809 AV |
1542 | } |
1543 | ||
1f707137 AV |
1544 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1545 | struct vfsmount *root) | |
1546 | { | |
1a4eeaf2 | 1547 | struct mount *mnt; |
1f707137 AV |
1548 | int res = f(root, arg); |
1549 | if (res) | |
1550 | return res; | |
1a4eeaf2 AV |
1551 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1552 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1553 | if (res) |
1554 | return res; | |
1555 | } | |
1556 | return 0; | |
1557 | } | |
1558 | ||
4b8b21f4 | 1559 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1560 | { |
315fc83e | 1561 | struct mount *p; |
719f5d7f | 1562 | |
909b0a88 | 1563 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1564 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1565 | mnt_release_group_id(p); |
719f5d7f MS |
1566 | } |
1567 | } | |
1568 | ||
4b8b21f4 | 1569 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1570 | { |
315fc83e | 1571 | struct mount *p; |
719f5d7f | 1572 | |
909b0a88 | 1573 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1574 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1575 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1576 | if (err) { |
4b8b21f4 | 1577 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1578 | return err; |
1579 | } | |
1580 | } | |
1581 | } | |
1582 | ||
1583 | return 0; | |
1584 | } | |
1585 | ||
b90fa9ae RP |
1586 | /* |
1587 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1588 | * @nd : place the mount tree @source_mnt is attached |
1589 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1590 | * store the parent mount and mountpoint dentry. | |
1591 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1592 | * |
1593 | * NOTE: in the table below explains the semantics when a source mount | |
1594 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1595 | * --------------------------------------------------------------------------- |
1596 | * | BIND MOUNT OPERATION | | |
1597 | * |************************************************************************** | |
1598 | * | source-->| shared | private | slave | unbindable | | |
1599 | * | dest | | | | | | |
1600 | * | | | | | | | | |
1601 | * | v | | | | | | |
1602 | * |************************************************************************** | |
1603 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1604 | * | | | | | | | |
1605 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1606 | * *************************************************************************** | |
b90fa9ae RP |
1607 | * A bind operation clones the source mount and mounts the clone on the |
1608 | * destination mount. | |
1609 | * | |
1610 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1611 | * tree of the destination mount and the cloned mount is added to | |
1612 | * the peer group of the source mount. | |
1613 | * (+) the cloned mount is created under the destination mount and is marked | |
1614 | * as shared. The cloned mount is added to the peer group of the source | |
1615 | * mount. | |
5afe0022 RP |
1616 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1617 | * of the destination mount and the cloned mount is made slave | |
1618 | * of the same master as that of the source mount. The cloned mount | |
1619 | * is marked as 'shared and slave'. | |
1620 | * (*) the cloned mount is made a slave of the same master as that of the | |
1621 | * source mount. | |
1622 | * | |
9676f0c6 RP |
1623 | * --------------------------------------------------------------------------- |
1624 | * | MOVE MOUNT OPERATION | | |
1625 | * |************************************************************************** | |
1626 | * | source-->| shared | private | slave | unbindable | | |
1627 | * | dest | | | | | | |
1628 | * | | | | | | | | |
1629 | * | v | | | | | | |
1630 | * |************************************************************************** | |
1631 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1632 | * | | | | | | | |
1633 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1634 | * *************************************************************************** | |
5afe0022 RP |
1635 | * |
1636 | * (+) the mount is moved to the destination. And is then propagated to | |
1637 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1638 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1639 | * (+++) the mount is moved to the destination and is then propagated to |
1640 | * all the mounts belonging to the destination mount's propagation tree. | |
1641 | * the mount is marked as 'shared and slave'. | |
1642 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1643 | * |
1644 | * if the source mount is a tree, the operations explained above is | |
1645 | * applied to each mount in the tree. | |
1646 | * Must be called without spinlocks held, since this function can sleep | |
1647 | * in allocations. | |
1648 | */ | |
0fb54e50 | 1649 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1650 | struct mount *dest_mnt, |
1651 | struct mountpoint *dest_mp, | |
1652 | struct path *parent_path) | |
b90fa9ae | 1653 | { |
38129a13 | 1654 | HLIST_HEAD(tree_list); |
315fc83e | 1655 | struct mount *child, *p; |
38129a13 | 1656 | struct hlist_node *n; |
719f5d7f | 1657 | int err; |
b90fa9ae | 1658 | |
fc7be130 | 1659 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1660 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1661 | if (err) |
1662 | goto out; | |
0b1b901b AV |
1663 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
1664 | if (err) | |
1665 | goto out_cleanup_ids; | |
1666 | lock_mount_hash(); | |
909b0a88 | 1667 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1668 | set_mnt_shared(p); |
0b1b901b AV |
1669 | } else { |
1670 | lock_mount_hash(); | |
b90fa9ae | 1671 | } |
1a390689 | 1672 | if (parent_path) { |
0fb54e50 | 1673 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1674 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1675 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1676 | } else { |
84d17192 | 1677 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1678 | commit_tree(source_mnt, NULL); |
21444403 | 1679 | } |
b90fa9ae | 1680 | |
38129a13 | 1681 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1682 | struct mount *q; |
38129a13 | 1683 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1684 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1685 | child->mnt_mountpoint); | |
1686 | commit_tree(child, q); | |
b90fa9ae | 1687 | } |
719ea2fb | 1688 | unlock_mount_hash(); |
99b7db7b | 1689 | |
b90fa9ae | 1690 | return 0; |
719f5d7f MS |
1691 | |
1692 | out_cleanup_ids: | |
0b1b901b | 1693 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1694 | out: |
1695 | return err; | |
b90fa9ae RP |
1696 | } |
1697 | ||
84d17192 | 1698 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1699 | { |
1700 | struct vfsmount *mnt; | |
84d17192 | 1701 | struct dentry *dentry = path->dentry; |
b12cea91 | 1702 | retry: |
84d17192 AV |
1703 | mutex_lock(&dentry->d_inode->i_mutex); |
1704 | if (unlikely(cant_mount(dentry))) { | |
1705 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1706 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1707 | } |
97216be0 | 1708 | namespace_lock(); |
b12cea91 | 1709 | mnt = lookup_mnt(path); |
84d17192 AV |
1710 | if (likely(!mnt)) { |
1711 | struct mountpoint *mp = new_mountpoint(dentry); | |
1712 | if (IS_ERR(mp)) { | |
97216be0 | 1713 | namespace_unlock(); |
84d17192 AV |
1714 | mutex_unlock(&dentry->d_inode->i_mutex); |
1715 | return mp; | |
1716 | } | |
1717 | return mp; | |
1718 | } | |
97216be0 | 1719 | namespace_unlock(); |
b12cea91 AV |
1720 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1721 | path_put(path); | |
1722 | path->mnt = mnt; | |
84d17192 | 1723 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1724 | goto retry; |
1725 | } | |
1726 | ||
84d17192 | 1727 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1728 | { |
84d17192 AV |
1729 | struct dentry *dentry = where->m_dentry; |
1730 | put_mountpoint(where); | |
328e6d90 | 1731 | namespace_unlock(); |
84d17192 | 1732 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1733 | } |
1734 | ||
84d17192 | 1735 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1736 | { |
95bc5f25 | 1737 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1738 | return -EINVAL; |
1739 | ||
84d17192 | 1740 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1741 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1742 | return -ENOTDIR; |
1743 | ||
84d17192 | 1744 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1745 | } |
1746 | ||
7a2e8a8f VA |
1747 | /* |
1748 | * Sanity check the flags to change_mnt_propagation. | |
1749 | */ | |
1750 | ||
1751 | static int flags_to_propagation_type(int flags) | |
1752 | { | |
7c6e984d | 1753 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1754 | |
1755 | /* Fail if any non-propagation flags are set */ | |
1756 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1757 | return 0; | |
1758 | /* Only one propagation flag should be set */ | |
1759 | if (!is_power_of_2(type)) | |
1760 | return 0; | |
1761 | return type; | |
1762 | } | |
1763 | ||
07b20889 RP |
1764 | /* |
1765 | * recursively change the type of the mountpoint. | |
1766 | */ | |
0a0d8a46 | 1767 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1768 | { |
315fc83e | 1769 | struct mount *m; |
4b8b21f4 | 1770 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1771 | int recurse = flag & MS_REC; |
7a2e8a8f | 1772 | int type; |
719f5d7f | 1773 | int err = 0; |
07b20889 | 1774 | |
2d92ab3c | 1775 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1776 | return -EINVAL; |
1777 | ||
7a2e8a8f VA |
1778 | type = flags_to_propagation_type(flag); |
1779 | if (!type) | |
1780 | return -EINVAL; | |
1781 | ||
97216be0 | 1782 | namespace_lock(); |
719f5d7f MS |
1783 | if (type == MS_SHARED) { |
1784 | err = invent_group_ids(mnt, recurse); | |
1785 | if (err) | |
1786 | goto out_unlock; | |
1787 | } | |
1788 | ||
719ea2fb | 1789 | lock_mount_hash(); |
909b0a88 | 1790 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1791 | change_mnt_propagation(m, type); |
719ea2fb | 1792 | unlock_mount_hash(); |
719f5d7f MS |
1793 | |
1794 | out_unlock: | |
97216be0 | 1795 | namespace_unlock(); |
719f5d7f | 1796 | return err; |
07b20889 RP |
1797 | } |
1798 | ||
5ff9d8a6 EB |
1799 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1800 | { | |
1801 | struct mount *child; | |
1802 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1803 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1804 | continue; | |
1805 | ||
1806 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1807 | return true; | |
1808 | } | |
1809 | return false; | |
1810 | } | |
1811 | ||
1da177e4 LT |
1812 | /* |
1813 | * do loopback mount. | |
1814 | */ | |
808d4e3c | 1815 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1816 | int recurse) |
1da177e4 | 1817 | { |
2d92ab3c | 1818 | struct path old_path; |
84d17192 AV |
1819 | struct mount *mnt = NULL, *old, *parent; |
1820 | struct mountpoint *mp; | |
57eccb83 | 1821 | int err; |
1da177e4 LT |
1822 | if (!old_name || !*old_name) |
1823 | return -EINVAL; | |
815d405c | 1824 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1825 | if (err) |
1826 | return err; | |
1827 | ||
8823c079 | 1828 | err = -EINVAL; |
4ce5d2b1 | 1829 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1830 | goto out; |
1831 | ||
84d17192 AV |
1832 | mp = lock_mount(path); |
1833 | err = PTR_ERR(mp); | |
1834 | if (IS_ERR(mp)) | |
b12cea91 AV |
1835 | goto out; |
1836 | ||
87129cc0 | 1837 | old = real_mount(old_path.mnt); |
84d17192 | 1838 | parent = real_mount(path->mnt); |
87129cc0 | 1839 | |
1da177e4 | 1840 | err = -EINVAL; |
fc7be130 | 1841 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1842 | goto out2; |
9676f0c6 | 1843 | |
84d17192 | 1844 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1845 | goto out2; |
1da177e4 | 1846 | |
5ff9d8a6 EB |
1847 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1848 | goto out2; | |
1849 | ||
ccd48bc7 | 1850 | if (recurse) |
4ce5d2b1 | 1851 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1852 | else |
87129cc0 | 1853 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1854 | |
be34d1a3 DH |
1855 | if (IS_ERR(mnt)) { |
1856 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1857 | goto out2; |
be34d1a3 | 1858 | } |
ccd48bc7 | 1859 | |
5ff9d8a6 EB |
1860 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1861 | ||
84d17192 | 1862 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1863 | if (err) { |
719ea2fb | 1864 | lock_mount_hash(); |
328e6d90 | 1865 | umount_tree(mnt, 0); |
719ea2fb | 1866 | unlock_mount_hash(); |
5b83d2c5 | 1867 | } |
b12cea91 | 1868 | out2: |
84d17192 | 1869 | unlock_mount(mp); |
ccd48bc7 | 1870 | out: |
2d92ab3c | 1871 | path_put(&old_path); |
1da177e4 LT |
1872 | return err; |
1873 | } | |
1874 | ||
2e4b7fcd DH |
1875 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1876 | { | |
1877 | int error = 0; | |
1878 | int readonly_request = 0; | |
1879 | ||
1880 | if (ms_flags & MS_RDONLY) | |
1881 | readonly_request = 1; | |
1882 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1883 | return 0; | |
1884 | ||
90563b19 EB |
1885 | if (mnt->mnt_flags & MNT_LOCK_READONLY) |
1886 | return -EPERM; | |
1887 | ||
2e4b7fcd | 1888 | if (readonly_request) |
83adc753 | 1889 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1890 | else |
83adc753 | 1891 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1892 | return error; |
1893 | } | |
1894 | ||
1da177e4 LT |
1895 | /* |
1896 | * change filesystem flags. dir should be a physical root of filesystem. | |
1897 | * If you've mounted a non-root directory somewhere and want to do remount | |
1898 | * on it - tough luck. | |
1899 | */ | |
0a0d8a46 | 1900 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1901 | void *data) |
1902 | { | |
1903 | int err; | |
2d92ab3c | 1904 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1905 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1906 | |
143c8c91 | 1907 | if (!check_mnt(mnt)) |
1da177e4 LT |
1908 | return -EINVAL; |
1909 | ||
2d92ab3c | 1910 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1911 | return -EINVAL; |
1912 | ||
ff36fe2c EP |
1913 | err = security_sb_remount(sb, data); |
1914 | if (err) | |
1915 | return err; | |
1916 | ||
1da177e4 | 1917 | down_write(&sb->s_umount); |
2e4b7fcd | 1918 | if (flags & MS_BIND) |
2d92ab3c | 1919 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1920 | else if (!capable(CAP_SYS_ADMIN)) |
1921 | err = -EPERM; | |
4aa98cf7 | 1922 | else |
2e4b7fcd | 1923 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1924 | if (!err) { |
719ea2fb | 1925 | lock_mount_hash(); |
143c8c91 AV |
1926 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1927 | mnt->mnt.mnt_flags = mnt_flags; | |
143c8c91 | 1928 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 1929 | unlock_mount_hash(); |
0e55a7cc | 1930 | } |
6339dab8 | 1931 | up_write(&sb->s_umount); |
1da177e4 LT |
1932 | return err; |
1933 | } | |
1934 | ||
cbbe362c | 1935 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1936 | { |
315fc83e | 1937 | struct mount *p; |
909b0a88 | 1938 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1939 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1940 | return 1; |
1941 | } | |
1942 | return 0; | |
1943 | } | |
1944 | ||
808d4e3c | 1945 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1946 | { |
2d92ab3c | 1947 | struct path old_path, parent_path; |
676da58d | 1948 | struct mount *p; |
0fb54e50 | 1949 | struct mount *old; |
84d17192 | 1950 | struct mountpoint *mp; |
57eccb83 | 1951 | int err; |
1da177e4 LT |
1952 | if (!old_name || !*old_name) |
1953 | return -EINVAL; | |
2d92ab3c | 1954 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1955 | if (err) |
1956 | return err; | |
1957 | ||
84d17192 AV |
1958 | mp = lock_mount(path); |
1959 | err = PTR_ERR(mp); | |
1960 | if (IS_ERR(mp)) | |
cc53ce53 DH |
1961 | goto out; |
1962 | ||
143c8c91 | 1963 | old = real_mount(old_path.mnt); |
fc7be130 | 1964 | p = real_mount(path->mnt); |
143c8c91 | 1965 | |
1da177e4 | 1966 | err = -EINVAL; |
fc7be130 | 1967 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1968 | goto out1; |
1969 | ||
5ff9d8a6 EB |
1970 | if (old->mnt.mnt_flags & MNT_LOCKED) |
1971 | goto out1; | |
1972 | ||
1da177e4 | 1973 | err = -EINVAL; |
2d92ab3c | 1974 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1975 | goto out1; |
1da177e4 | 1976 | |
676da58d | 1977 | if (!mnt_has_parent(old)) |
21444403 | 1978 | goto out1; |
1da177e4 | 1979 | |
2d92ab3c AV |
1980 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1981 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1982 | goto out1; |
1983 | /* | |
1984 | * Don't move a mount residing in a shared parent. | |
1985 | */ | |
fc7be130 | 1986 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1987 | goto out1; |
9676f0c6 RP |
1988 | /* |
1989 | * Don't move a mount tree containing unbindable mounts to a destination | |
1990 | * mount which is shared. | |
1991 | */ | |
fc7be130 | 1992 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1993 | goto out1; |
1da177e4 | 1994 | err = -ELOOP; |
fc7be130 | 1995 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1996 | if (p == old) |
21444403 | 1997 | goto out1; |
1da177e4 | 1998 | |
84d17192 | 1999 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2000 | if (err) |
21444403 | 2001 | goto out1; |
1da177e4 LT |
2002 | |
2003 | /* if the mount is moved, it should no longer be expire | |
2004 | * automatically */ | |
6776db3d | 2005 | list_del_init(&old->mnt_expire); |
1da177e4 | 2006 | out1: |
84d17192 | 2007 | unlock_mount(mp); |
1da177e4 | 2008 | out: |
1da177e4 | 2009 | if (!err) |
1a390689 | 2010 | path_put(&parent_path); |
2d92ab3c | 2011 | path_put(&old_path); |
1da177e4 LT |
2012 | return err; |
2013 | } | |
2014 | ||
9d412a43 AV |
2015 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2016 | { | |
2017 | int err; | |
2018 | const char *subtype = strchr(fstype, '.'); | |
2019 | if (subtype) { | |
2020 | subtype++; | |
2021 | err = -EINVAL; | |
2022 | if (!subtype[0]) | |
2023 | goto err; | |
2024 | } else | |
2025 | subtype = ""; | |
2026 | ||
2027 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2028 | err = -ENOMEM; | |
2029 | if (!mnt->mnt_sb->s_subtype) | |
2030 | goto err; | |
2031 | return mnt; | |
2032 | ||
2033 | err: | |
2034 | mntput(mnt); | |
2035 | return ERR_PTR(err); | |
2036 | } | |
2037 | ||
9d412a43 AV |
2038 | /* |
2039 | * add a mount into a namespace's mount tree | |
2040 | */ | |
95bc5f25 | 2041 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2042 | { |
84d17192 AV |
2043 | struct mountpoint *mp; |
2044 | struct mount *parent; | |
9d412a43 AV |
2045 | int err; |
2046 | ||
48a066e7 | 2047 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT); |
9d412a43 | 2048 | |
84d17192 AV |
2049 | mp = lock_mount(path); |
2050 | if (IS_ERR(mp)) | |
2051 | return PTR_ERR(mp); | |
9d412a43 | 2052 | |
84d17192 | 2053 | parent = real_mount(path->mnt); |
9d412a43 | 2054 | err = -EINVAL; |
84d17192 | 2055 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2056 | /* that's acceptable only for automounts done in private ns */ |
2057 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2058 | goto unlock; | |
2059 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2060 | if (!parent->mnt_ns) |
156cacb1 AV |
2061 | goto unlock; |
2062 | } | |
9d412a43 AV |
2063 | |
2064 | /* Refuse the same filesystem on the same mount point */ | |
2065 | err = -EBUSY; | |
95bc5f25 | 2066 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2067 | path->mnt->mnt_root == path->dentry) |
2068 | goto unlock; | |
2069 | ||
2070 | err = -EINVAL; | |
95bc5f25 | 2071 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2072 | goto unlock; |
2073 | ||
95bc5f25 | 2074 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2075 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2076 | |
2077 | unlock: | |
84d17192 | 2078 | unlock_mount(mp); |
9d412a43 AV |
2079 | return err; |
2080 | } | |
b1e75df4 | 2081 | |
1da177e4 LT |
2082 | /* |
2083 | * create a new mount for userspace and request it to be added into the | |
2084 | * namespace's tree | |
2085 | */ | |
0c55cfc4 | 2086 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2087 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2088 | { |
0c55cfc4 | 2089 | struct file_system_type *type; |
9b40bc90 | 2090 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2091 | struct vfsmount *mnt; |
15f9a3f3 | 2092 | int err; |
1da177e4 | 2093 | |
0c55cfc4 | 2094 | if (!fstype) |
1da177e4 LT |
2095 | return -EINVAL; |
2096 | ||
0c55cfc4 EB |
2097 | type = get_fs_type(fstype); |
2098 | if (!type) | |
2099 | return -ENODEV; | |
2100 | ||
2101 | if (user_ns != &init_user_ns) { | |
2102 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2103 | put_filesystem(type); | |
2104 | return -EPERM; | |
2105 | } | |
2106 | /* Only in special cases allow devices from mounts | |
2107 | * created outside the initial user namespace. | |
2108 | */ | |
2109 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2110 | flags |= MS_NODEV; | |
2111 | mnt_flags |= MNT_NODEV; | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | mnt = vfs_kern_mount(type, flags, name, data); | |
2116 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2117 | !mnt->mnt_sb->s_subtype) | |
2118 | mnt = fs_set_subtype(mnt, fstype); | |
2119 | ||
2120 | put_filesystem(type); | |
1da177e4 LT |
2121 | if (IS_ERR(mnt)) |
2122 | return PTR_ERR(mnt); | |
2123 | ||
95bc5f25 | 2124 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2125 | if (err) |
2126 | mntput(mnt); | |
2127 | return err; | |
1da177e4 LT |
2128 | } |
2129 | ||
19a167af AV |
2130 | int finish_automount(struct vfsmount *m, struct path *path) |
2131 | { | |
6776db3d | 2132 | struct mount *mnt = real_mount(m); |
19a167af AV |
2133 | int err; |
2134 | /* The new mount record should have at least 2 refs to prevent it being | |
2135 | * expired before we get a chance to add it | |
2136 | */ | |
6776db3d | 2137 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2138 | |
2139 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2140 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2141 | err = -ELOOP; |
2142 | goto fail; | |
19a167af AV |
2143 | } |
2144 | ||
95bc5f25 | 2145 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2146 | if (!err) |
2147 | return 0; | |
2148 | fail: | |
2149 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2150 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2151 | namespace_lock(); |
6776db3d | 2152 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2153 | namespace_unlock(); |
19a167af | 2154 | } |
b1e75df4 AV |
2155 | mntput(m); |
2156 | mntput(m); | |
19a167af AV |
2157 | return err; |
2158 | } | |
2159 | ||
ea5b778a DH |
2160 | /** |
2161 | * mnt_set_expiry - Put a mount on an expiration list | |
2162 | * @mnt: The mount to list. | |
2163 | * @expiry_list: The list to add the mount to. | |
2164 | */ | |
2165 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2166 | { | |
97216be0 | 2167 | namespace_lock(); |
ea5b778a | 2168 | |
6776db3d | 2169 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2170 | |
97216be0 | 2171 | namespace_unlock(); |
ea5b778a DH |
2172 | } |
2173 | EXPORT_SYMBOL(mnt_set_expiry); | |
2174 | ||
1da177e4 LT |
2175 | /* |
2176 | * process a list of expirable mountpoints with the intent of discarding any | |
2177 | * mountpoints that aren't in use and haven't been touched since last we came | |
2178 | * here | |
2179 | */ | |
2180 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2181 | { | |
761d5c38 | 2182 | struct mount *mnt, *next; |
1da177e4 LT |
2183 | LIST_HEAD(graveyard); |
2184 | ||
2185 | if (list_empty(mounts)) | |
2186 | return; | |
2187 | ||
97216be0 | 2188 | namespace_lock(); |
719ea2fb | 2189 | lock_mount_hash(); |
1da177e4 LT |
2190 | |
2191 | /* extract from the expiration list every vfsmount that matches the | |
2192 | * following criteria: | |
2193 | * - only referenced by its parent vfsmount | |
2194 | * - still marked for expiry (marked on the last call here; marks are | |
2195 | * cleared by mntput()) | |
2196 | */ | |
6776db3d | 2197 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2198 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2199 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2200 | continue; |
6776db3d | 2201 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2202 | } |
bcc5c7d2 | 2203 | while (!list_empty(&graveyard)) { |
6776db3d | 2204 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2205 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2206 | umount_tree(mnt, 1); |
bcc5c7d2 | 2207 | } |
719ea2fb | 2208 | unlock_mount_hash(); |
3ab6abee | 2209 | namespace_unlock(); |
5528f911 TM |
2210 | } |
2211 | ||
2212 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2213 | ||
2214 | /* | |
2215 | * Ripoff of 'select_parent()' | |
2216 | * | |
2217 | * search the list of submounts for a given mountpoint, and move any | |
2218 | * shrinkable submounts to the 'graveyard' list. | |
2219 | */ | |
692afc31 | 2220 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2221 | { |
692afc31 | 2222 | struct mount *this_parent = parent; |
5528f911 TM |
2223 | struct list_head *next; |
2224 | int found = 0; | |
2225 | ||
2226 | repeat: | |
6b41d536 | 2227 | next = this_parent->mnt_mounts.next; |
5528f911 | 2228 | resume: |
6b41d536 | 2229 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2230 | struct list_head *tmp = next; |
6b41d536 | 2231 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2232 | |
2233 | next = tmp->next; | |
692afc31 | 2234 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2235 | continue; |
5528f911 TM |
2236 | /* |
2237 | * Descend a level if the d_mounts list is non-empty. | |
2238 | */ | |
6b41d536 | 2239 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2240 | this_parent = mnt; |
2241 | goto repeat; | |
2242 | } | |
1da177e4 | 2243 | |
1ab59738 | 2244 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2245 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2246 | found++; |
2247 | } | |
1da177e4 | 2248 | } |
5528f911 TM |
2249 | /* |
2250 | * All done at this level ... ascend and resume the search | |
2251 | */ | |
2252 | if (this_parent != parent) { | |
6b41d536 | 2253 | next = this_parent->mnt_child.next; |
0714a533 | 2254 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2255 | goto resume; |
2256 | } | |
2257 | return found; | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * process a list of expirable mountpoints with the intent of discarding any | |
2262 | * submounts of a specific parent mountpoint | |
99b7db7b | 2263 | * |
48a066e7 | 2264 | * mount_lock must be held for write |
5528f911 | 2265 | */ |
b54b9be7 | 2266 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2267 | { |
2268 | LIST_HEAD(graveyard); | |
761d5c38 | 2269 | struct mount *m; |
5528f911 | 2270 | |
5528f911 | 2271 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2272 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2273 | while (!list_empty(&graveyard)) { |
761d5c38 | 2274 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2275 | mnt_expire); |
143c8c91 | 2276 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2277 | umount_tree(m, 1); |
bcc5c7d2 AV |
2278 | } |
2279 | } | |
1da177e4 LT |
2280 | } |
2281 | ||
1da177e4 LT |
2282 | /* |
2283 | * Some copy_from_user() implementations do not return the exact number of | |
2284 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2285 | * Note that this function differs from copy_from_user() in that it will oops | |
2286 | * on bad values of `to', rather than returning a short copy. | |
2287 | */ | |
b58fed8b RP |
2288 | static long exact_copy_from_user(void *to, const void __user * from, |
2289 | unsigned long n) | |
1da177e4 LT |
2290 | { |
2291 | char *t = to; | |
2292 | const char __user *f = from; | |
2293 | char c; | |
2294 | ||
2295 | if (!access_ok(VERIFY_READ, from, n)) | |
2296 | return n; | |
2297 | ||
2298 | while (n) { | |
2299 | if (__get_user(c, f)) { | |
2300 | memset(t, 0, n); | |
2301 | break; | |
2302 | } | |
2303 | *t++ = c; | |
2304 | f++; | |
2305 | n--; | |
2306 | } | |
2307 | return n; | |
2308 | } | |
2309 | ||
b58fed8b | 2310 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2311 | { |
2312 | int i; | |
2313 | unsigned long page; | |
2314 | unsigned long size; | |
b58fed8b | 2315 | |
1da177e4 LT |
2316 | *where = 0; |
2317 | if (!data) | |
2318 | return 0; | |
2319 | ||
2320 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2321 | return -ENOMEM; | |
2322 | ||
2323 | /* We only care that *some* data at the address the user | |
2324 | * gave us is valid. Just in case, we'll zero | |
2325 | * the remainder of the page. | |
2326 | */ | |
2327 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2328 | size = TASK_SIZE - (unsigned long)data; | |
2329 | if (size > PAGE_SIZE) | |
2330 | size = PAGE_SIZE; | |
2331 | ||
2332 | i = size - exact_copy_from_user((void *)page, data, size); | |
2333 | if (!i) { | |
b58fed8b | 2334 | free_page(page); |
1da177e4 LT |
2335 | return -EFAULT; |
2336 | } | |
2337 | if (i != PAGE_SIZE) | |
2338 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2339 | *where = page; | |
2340 | return 0; | |
2341 | } | |
2342 | ||
eca6f534 VN |
2343 | int copy_mount_string(const void __user *data, char **where) |
2344 | { | |
2345 | char *tmp; | |
2346 | ||
2347 | if (!data) { | |
2348 | *where = NULL; | |
2349 | return 0; | |
2350 | } | |
2351 | ||
2352 | tmp = strndup_user(data, PAGE_SIZE); | |
2353 | if (IS_ERR(tmp)) | |
2354 | return PTR_ERR(tmp); | |
2355 | ||
2356 | *where = tmp; | |
2357 | return 0; | |
2358 | } | |
2359 | ||
1da177e4 LT |
2360 | /* |
2361 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2362 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2363 | * | |
2364 | * data is a (void *) that can point to any structure up to | |
2365 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2366 | * information (or be NULL). | |
2367 | * | |
2368 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2369 | * When the flags word was introduced its top half was required | |
2370 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2371 | * Therefore, if this magic number is present, it carries no information | |
2372 | * and must be discarded. | |
2373 | */ | |
808d4e3c AV |
2374 | long do_mount(const char *dev_name, const char *dir_name, |
2375 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2376 | { |
2d92ab3c | 2377 | struct path path; |
1da177e4 LT |
2378 | int retval = 0; |
2379 | int mnt_flags = 0; | |
2380 | ||
2381 | /* Discard magic */ | |
2382 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2383 | flags &= ~MS_MGC_MSK; | |
2384 | ||
2385 | /* Basic sanity checks */ | |
2386 | ||
2387 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2388 | return -EINVAL; | |
1da177e4 LT |
2389 | |
2390 | if (data_page) | |
2391 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2392 | ||
a27ab9f2 TH |
2393 | /* ... and get the mountpoint */ |
2394 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2395 | if (retval) | |
2396 | return retval; | |
2397 | ||
2398 | retval = security_sb_mount(dev_name, &path, | |
2399 | type_page, flags, data_page); | |
0d5cadb8 AV |
2400 | if (!retval && !may_mount()) |
2401 | retval = -EPERM; | |
a27ab9f2 TH |
2402 | if (retval) |
2403 | goto dput_out; | |
2404 | ||
613cbe3d AK |
2405 | /* Default to relatime unless overriden */ |
2406 | if (!(flags & MS_NOATIME)) | |
2407 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2408 | |
1da177e4 LT |
2409 | /* Separate the per-mountpoint flags */ |
2410 | if (flags & MS_NOSUID) | |
2411 | mnt_flags |= MNT_NOSUID; | |
2412 | if (flags & MS_NODEV) | |
2413 | mnt_flags |= MNT_NODEV; | |
2414 | if (flags & MS_NOEXEC) | |
2415 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2416 | if (flags & MS_NOATIME) |
2417 | mnt_flags |= MNT_NOATIME; | |
2418 | if (flags & MS_NODIRATIME) | |
2419 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2420 | if (flags & MS_STRICTATIME) |
2421 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2422 | if (flags & MS_RDONLY) |
2423 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2424 | |
7a4dec53 | 2425 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2426 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2427 | MS_STRICTATIME); | |
1da177e4 | 2428 | |
1da177e4 | 2429 | if (flags & MS_REMOUNT) |
2d92ab3c | 2430 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2431 | data_page); |
2432 | else if (flags & MS_BIND) | |
2d92ab3c | 2433 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2434 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2435 | retval = do_change_type(&path, flags); |
1da177e4 | 2436 | else if (flags & MS_MOVE) |
2d92ab3c | 2437 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2438 | else |
2d92ab3c | 2439 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2440 | dev_name, data_page); |
2441 | dput_out: | |
2d92ab3c | 2442 | path_put(&path); |
1da177e4 LT |
2443 | return retval; |
2444 | } | |
2445 | ||
771b1371 EB |
2446 | static void free_mnt_ns(struct mnt_namespace *ns) |
2447 | { | |
98f842e6 | 2448 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2449 | put_user_ns(ns->user_ns); |
2450 | kfree(ns); | |
2451 | } | |
2452 | ||
8823c079 EB |
2453 | /* |
2454 | * Assign a sequence number so we can detect when we attempt to bind | |
2455 | * mount a reference to an older mount namespace into the current | |
2456 | * mount namespace, preventing reference counting loops. A 64bit | |
2457 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2458 | * is effectively never, so we can ignore the possibility. | |
2459 | */ | |
2460 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2461 | ||
771b1371 | 2462 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2463 | { |
2464 | struct mnt_namespace *new_ns; | |
98f842e6 | 2465 | int ret; |
cf8d2c11 TM |
2466 | |
2467 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2468 | if (!new_ns) | |
2469 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2470 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2471 | if (ret) { | |
2472 | kfree(new_ns); | |
2473 | return ERR_PTR(ret); | |
2474 | } | |
8823c079 | 2475 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2476 | atomic_set(&new_ns->count, 1); |
2477 | new_ns->root = NULL; | |
2478 | INIT_LIST_HEAD(&new_ns->list); | |
2479 | init_waitqueue_head(&new_ns->poll); | |
2480 | new_ns->event = 0; | |
771b1371 | 2481 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2482 | return new_ns; |
2483 | } | |
2484 | ||
9559f689 AV |
2485 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2486 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2487 | { |
6b3286ed | 2488 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2489 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2490 | struct mount *p, *q; |
9559f689 | 2491 | struct mount *old; |
cb338d06 | 2492 | struct mount *new; |
7a472ef4 | 2493 | int copy_flags; |
1da177e4 | 2494 | |
9559f689 AV |
2495 | BUG_ON(!ns); |
2496 | ||
2497 | if (likely(!(flags & CLONE_NEWNS))) { | |
2498 | get_mnt_ns(ns); | |
2499 | return ns; | |
2500 | } | |
2501 | ||
2502 | old = ns->root; | |
2503 | ||
771b1371 | 2504 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2505 | if (IS_ERR(new_ns)) |
2506 | return new_ns; | |
1da177e4 | 2507 | |
97216be0 | 2508 | namespace_lock(); |
1da177e4 | 2509 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2510 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2511 | if (user_ns != ns->user_ns) |
132c94e3 | 2512 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2513 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2514 | if (IS_ERR(new)) { |
328e6d90 | 2515 | namespace_unlock(); |
771b1371 | 2516 | free_mnt_ns(new_ns); |
be34d1a3 | 2517 | return ERR_CAST(new); |
1da177e4 | 2518 | } |
be08d6d2 | 2519 | new_ns->root = new; |
1a4eeaf2 | 2520 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2521 | |
2522 | /* | |
2523 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2524 | * as belonging to new namespace. We have already acquired a private | |
2525 | * fs_struct, so tsk->fs->lock is not needed. | |
2526 | */ | |
909b0a88 | 2527 | p = old; |
cb338d06 | 2528 | q = new; |
1da177e4 | 2529 | while (p) { |
143c8c91 | 2530 | q->mnt_ns = new_ns; |
9559f689 AV |
2531 | if (new_fs) { |
2532 | if (&p->mnt == new_fs->root.mnt) { | |
2533 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2534 | rootmnt = &p->mnt; |
1da177e4 | 2535 | } |
9559f689 AV |
2536 | if (&p->mnt == new_fs->pwd.mnt) { |
2537 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2538 | pwdmnt = &p->mnt; |
1da177e4 | 2539 | } |
1da177e4 | 2540 | } |
909b0a88 AV |
2541 | p = next_mnt(p, old); |
2542 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2543 | if (!q) |
2544 | break; | |
2545 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2546 | p = next_mnt(p, old); | |
1da177e4 | 2547 | } |
328e6d90 | 2548 | namespace_unlock(); |
1da177e4 | 2549 | |
1da177e4 | 2550 | if (rootmnt) |
f03c6599 | 2551 | mntput(rootmnt); |
1da177e4 | 2552 | if (pwdmnt) |
f03c6599 | 2553 | mntput(pwdmnt); |
1da177e4 | 2554 | |
741a2951 | 2555 | return new_ns; |
1da177e4 LT |
2556 | } |
2557 | ||
cf8d2c11 TM |
2558 | /** |
2559 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2560 | * @mnt: pointer to the new root filesystem mountpoint | |
2561 | */ | |
1a4eeaf2 | 2562 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2563 | { |
771b1371 | 2564 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2565 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2566 | struct mount *mnt = real_mount(m); |
2567 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2568 | new_ns->root = mnt; |
b1983cd8 | 2569 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2570 | } else { |
1a4eeaf2 | 2571 | mntput(m); |
cf8d2c11 TM |
2572 | } |
2573 | return new_ns; | |
2574 | } | |
cf8d2c11 | 2575 | |
ea441d11 AV |
2576 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2577 | { | |
2578 | struct mnt_namespace *ns; | |
d31da0f0 | 2579 | struct super_block *s; |
ea441d11 AV |
2580 | struct path path; |
2581 | int err; | |
2582 | ||
2583 | ns = create_mnt_ns(mnt); | |
2584 | if (IS_ERR(ns)) | |
2585 | return ERR_CAST(ns); | |
2586 | ||
2587 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2588 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2589 | ||
2590 | put_mnt_ns(ns); | |
2591 | ||
2592 | if (err) | |
2593 | return ERR_PTR(err); | |
2594 | ||
2595 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2596 | s = path.mnt->mnt_sb; |
2597 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2598 | mntput(path.mnt); |
2599 | /* lock the sucker */ | |
d31da0f0 | 2600 | down_write(&s->s_umount); |
ea441d11 AV |
2601 | /* ... and return the root of (sub)tree on it */ |
2602 | return path.dentry; | |
2603 | } | |
2604 | EXPORT_SYMBOL(mount_subtree); | |
2605 | ||
bdc480e3 HC |
2606 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2607 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2608 | { |
eca6f534 VN |
2609 | int ret; |
2610 | char *kernel_type; | |
91a27b2a | 2611 | struct filename *kernel_dir; |
eca6f534 | 2612 | char *kernel_dev; |
1da177e4 | 2613 | unsigned long data_page; |
1da177e4 | 2614 | |
eca6f534 VN |
2615 | ret = copy_mount_string(type, &kernel_type); |
2616 | if (ret < 0) | |
2617 | goto out_type; | |
1da177e4 | 2618 | |
eca6f534 VN |
2619 | kernel_dir = getname(dir_name); |
2620 | if (IS_ERR(kernel_dir)) { | |
2621 | ret = PTR_ERR(kernel_dir); | |
2622 | goto out_dir; | |
2623 | } | |
1da177e4 | 2624 | |
eca6f534 VN |
2625 | ret = copy_mount_string(dev_name, &kernel_dev); |
2626 | if (ret < 0) | |
2627 | goto out_dev; | |
1da177e4 | 2628 | |
eca6f534 VN |
2629 | ret = copy_mount_options(data, &data_page); |
2630 | if (ret < 0) | |
2631 | goto out_data; | |
1da177e4 | 2632 | |
91a27b2a | 2633 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2634 | (void *) data_page); |
1da177e4 | 2635 | |
eca6f534 VN |
2636 | free_page(data_page); |
2637 | out_data: | |
2638 | kfree(kernel_dev); | |
2639 | out_dev: | |
2640 | putname(kernel_dir); | |
2641 | out_dir: | |
2642 | kfree(kernel_type); | |
2643 | out_type: | |
2644 | return ret; | |
1da177e4 LT |
2645 | } |
2646 | ||
afac7cba AV |
2647 | /* |
2648 | * Return true if path is reachable from root | |
2649 | * | |
48a066e7 | 2650 | * namespace_sem or mount_lock is held |
afac7cba | 2651 | */ |
643822b4 | 2652 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2653 | const struct path *root) |
2654 | { | |
643822b4 | 2655 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2656 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2657 | mnt = mnt->mnt_parent; |
afac7cba | 2658 | } |
643822b4 | 2659 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2660 | } |
2661 | ||
2662 | int path_is_under(struct path *path1, struct path *path2) | |
2663 | { | |
2664 | int res; | |
48a066e7 | 2665 | read_seqlock_excl(&mount_lock); |
643822b4 | 2666 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2667 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2668 | return res; |
2669 | } | |
2670 | EXPORT_SYMBOL(path_is_under); | |
2671 | ||
1da177e4 LT |
2672 | /* |
2673 | * pivot_root Semantics: | |
2674 | * Moves the root file system of the current process to the directory put_old, | |
2675 | * makes new_root as the new root file system of the current process, and sets | |
2676 | * root/cwd of all processes which had them on the current root to new_root. | |
2677 | * | |
2678 | * Restrictions: | |
2679 | * The new_root and put_old must be directories, and must not be on the | |
2680 | * same file system as the current process root. The put_old must be | |
2681 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2682 | * pointed to by put_old must yield the same directory as new_root. No other | |
2683 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2684 | * | |
4a0d11fa NB |
2685 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2686 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2687 | * in this situation. | |
2688 | * | |
1da177e4 LT |
2689 | * Notes: |
2690 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2691 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2692 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2693 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2694 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2695 | * first. | |
2696 | */ | |
3480b257 HC |
2697 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2698 | const char __user *, put_old) | |
1da177e4 | 2699 | { |
2d8f3038 | 2700 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2701 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2702 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2703 | int error; |
2704 | ||
9b40bc90 | 2705 | if (!may_mount()) |
1da177e4 LT |
2706 | return -EPERM; |
2707 | ||
2d8f3038 | 2708 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2709 | if (error) |
2710 | goto out0; | |
1da177e4 | 2711 | |
2d8f3038 | 2712 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2713 | if (error) |
2714 | goto out1; | |
2715 | ||
2d8f3038 | 2716 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2717 | if (error) |
2718 | goto out2; | |
1da177e4 | 2719 | |
f7ad3c6b | 2720 | get_fs_root(current->fs, &root); |
84d17192 AV |
2721 | old_mp = lock_mount(&old); |
2722 | error = PTR_ERR(old_mp); | |
2723 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2724 | goto out3; |
2725 | ||
1da177e4 | 2726 | error = -EINVAL; |
419148da AV |
2727 | new_mnt = real_mount(new.mnt); |
2728 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2729 | old_mnt = real_mount(old.mnt); |
2730 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2731 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2732 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2733 | goto out4; |
143c8c91 | 2734 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2735 | goto out4; |
5ff9d8a6 EB |
2736 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2737 | goto out4; | |
1da177e4 | 2738 | error = -ENOENT; |
f3da392e | 2739 | if (d_unlinked(new.dentry)) |
b12cea91 | 2740 | goto out4; |
1da177e4 | 2741 | error = -EBUSY; |
84d17192 | 2742 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2743 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2744 | error = -EINVAL; |
8c3ee42e | 2745 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2746 | goto out4; /* not a mountpoint */ |
676da58d | 2747 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2748 | goto out4; /* not attached */ |
84d17192 | 2749 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2750 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2751 | goto out4; /* not a mountpoint */ |
676da58d | 2752 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2753 | goto out4; /* not attached */ |
4ac91378 | 2754 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2755 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2756 | goto out4; |
84d17192 | 2757 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2758 | lock_mount_hash(); |
419148da AV |
2759 | detach_mnt(new_mnt, &parent_path); |
2760 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2761 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2762 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2763 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2764 | } | |
4ac91378 | 2765 | /* mount old root on put_old */ |
84d17192 | 2766 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2767 | /* mount new_root on / */ |
84d17192 | 2768 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2769 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2770 | unlock_mount_hash(); |
2d8f3038 | 2771 | chroot_fs_refs(&root, &new); |
84d17192 | 2772 | put_mountpoint(root_mp); |
1da177e4 | 2773 | error = 0; |
b12cea91 | 2774 | out4: |
84d17192 | 2775 | unlock_mount(old_mp); |
b12cea91 AV |
2776 | if (!error) { |
2777 | path_put(&root_parent); | |
2778 | path_put(&parent_path); | |
2779 | } | |
2780 | out3: | |
8c3ee42e | 2781 | path_put(&root); |
b12cea91 | 2782 | out2: |
2d8f3038 | 2783 | path_put(&old); |
1da177e4 | 2784 | out1: |
2d8f3038 | 2785 | path_put(&new); |
1da177e4 | 2786 | out0: |
1da177e4 | 2787 | return error; |
1da177e4 LT |
2788 | } |
2789 | ||
2790 | static void __init init_mount_tree(void) | |
2791 | { | |
2792 | struct vfsmount *mnt; | |
6b3286ed | 2793 | struct mnt_namespace *ns; |
ac748a09 | 2794 | struct path root; |
0c55cfc4 | 2795 | struct file_system_type *type; |
1da177e4 | 2796 | |
0c55cfc4 EB |
2797 | type = get_fs_type("rootfs"); |
2798 | if (!type) | |
2799 | panic("Can't find rootfs type"); | |
2800 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2801 | put_filesystem(type); | |
1da177e4 LT |
2802 | if (IS_ERR(mnt)) |
2803 | panic("Can't create rootfs"); | |
b3e19d92 | 2804 | |
3b22edc5 TM |
2805 | ns = create_mnt_ns(mnt); |
2806 | if (IS_ERR(ns)) | |
1da177e4 | 2807 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2808 | |
2809 | init_task.nsproxy->mnt_ns = ns; | |
2810 | get_mnt_ns(ns); | |
2811 | ||
be08d6d2 AV |
2812 | root.mnt = mnt; |
2813 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2814 | |
2815 | set_fs_pwd(current->fs, &root); | |
2816 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2817 | } |
2818 | ||
74bf17cf | 2819 | void __init mnt_init(void) |
1da177e4 | 2820 | { |
13f14b4d | 2821 | unsigned u; |
15a67dd8 | 2822 | int err; |
1da177e4 | 2823 | |
7d6fec45 | 2824 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2825 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2826 | |
0818bf27 | 2827 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2828 | sizeof(struct hlist_head), |
0818bf27 AV |
2829 | mhash_entries, 19, |
2830 | 0, | |
2831 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2832 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2833 | sizeof(struct hlist_head), | |
2834 | mphash_entries, 19, | |
2835 | 0, | |
2836 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2837 | |
84d17192 | 2838 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2839 | panic("Failed to allocate mount hash table\n"); |
2840 | ||
0818bf27 | 2841 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2842 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2843 | for (u = 0; u <= mp_hash_mask; u++) |
2844 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2845 | |
4b93dc9b TH |
2846 | kernfs_init(); |
2847 | ||
15a67dd8 RD |
2848 | err = sysfs_init(); |
2849 | if (err) | |
2850 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2851 | __func__, err); |
00d26666 GKH |
2852 | fs_kobj = kobject_create_and_add("fs", NULL); |
2853 | if (!fs_kobj) | |
8e24eea7 | 2854 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2855 | init_rootfs(); |
2856 | init_mount_tree(); | |
2857 | } | |
2858 | ||
616511d0 | 2859 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2860 | { |
d498b25a | 2861 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2862 | return; |
7b00ed6f | 2863 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2864 | free_mnt_ns(ns); |
1da177e4 | 2865 | } |
9d412a43 AV |
2866 | |
2867 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2868 | { | |
423e0ab0 TC |
2869 | struct vfsmount *mnt; |
2870 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2871 | if (!IS_ERR(mnt)) { | |
2872 | /* | |
2873 | * it is a longterm mount, don't release mnt until | |
2874 | * we unmount before file sys is unregistered | |
2875 | */ | |
f7a99c5b | 2876 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2877 | } |
2878 | return mnt; | |
9d412a43 AV |
2879 | } |
2880 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2881 | |
2882 | void kern_unmount(struct vfsmount *mnt) | |
2883 | { | |
2884 | /* release long term mount so mount point can be released */ | |
2885 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2886 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2887 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2888 | mntput(mnt); |
2889 | } | |
2890 | } | |
2891 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2892 | |
2893 | bool our_mnt(struct vfsmount *mnt) | |
2894 | { | |
143c8c91 | 2895 | return check_mnt(real_mount(mnt)); |
02125a82 | 2896 | } |
8823c079 | 2897 | |
3151527e EB |
2898 | bool current_chrooted(void) |
2899 | { | |
2900 | /* Does the current process have a non-standard root */ | |
2901 | struct path ns_root; | |
2902 | struct path fs_root; | |
2903 | bool chrooted; | |
2904 | ||
2905 | /* Find the namespace root */ | |
2906 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2907 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2908 | path_get(&ns_root); | |
2909 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2910 | ; | |
2911 | ||
2912 | get_fs_root(current->fs, &fs_root); | |
2913 | ||
2914 | chrooted = !path_equal(&fs_root, &ns_root); | |
2915 | ||
2916 | path_put(&fs_root); | |
2917 | path_put(&ns_root); | |
2918 | ||
2919 | return chrooted; | |
2920 | } | |
2921 | ||
e51db735 | 2922 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
2923 | { |
2924 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2925 | struct mount *mnt; | |
e51db735 | 2926 | bool visible = false; |
87a8ebd6 | 2927 | |
e51db735 EB |
2928 | if (unlikely(!ns)) |
2929 | return false; | |
2930 | ||
44bb4385 | 2931 | down_read(&namespace_sem); |
87a8ebd6 | 2932 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
2933 | struct mount *child; |
2934 | if (mnt->mnt.mnt_sb->s_type != type) | |
2935 | continue; | |
2936 | ||
2937 | /* This mount is not fully visible if there are any child mounts | |
2938 | * that cover anything except for empty directories. | |
2939 | */ | |
2940 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2941 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
2942 | if (!S_ISDIR(inode->i_mode)) | |
2943 | goto next; | |
41301ae7 | 2944 | if (inode->i_nlink > 2) |
e51db735 | 2945 | goto next; |
87a8ebd6 | 2946 | } |
e51db735 EB |
2947 | visible = true; |
2948 | goto found; | |
2949 | next: ; | |
87a8ebd6 | 2950 | } |
e51db735 | 2951 | found: |
44bb4385 | 2952 | up_read(&namespace_sem); |
e51db735 | 2953 | return visible; |
87a8ebd6 EB |
2954 | } |
2955 | ||
8823c079 EB |
2956 | static void *mntns_get(struct task_struct *task) |
2957 | { | |
2958 | struct mnt_namespace *ns = NULL; | |
2959 | struct nsproxy *nsproxy; | |
2960 | ||
2961 | rcu_read_lock(); | |
2962 | nsproxy = task_nsproxy(task); | |
2963 | if (nsproxy) { | |
2964 | ns = nsproxy->mnt_ns; | |
2965 | get_mnt_ns(ns); | |
2966 | } | |
2967 | rcu_read_unlock(); | |
2968 | ||
2969 | return ns; | |
2970 | } | |
2971 | ||
2972 | static void mntns_put(void *ns) | |
2973 | { | |
2974 | put_mnt_ns(ns); | |
2975 | } | |
2976 | ||
2977 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2978 | { | |
2979 | struct fs_struct *fs = current->fs; | |
2980 | struct mnt_namespace *mnt_ns = ns; | |
2981 | struct path root; | |
2982 | ||
0c55cfc4 | 2983 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
2984 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
2985 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 2986 | return -EPERM; |
8823c079 EB |
2987 | |
2988 | if (fs->users != 1) | |
2989 | return -EINVAL; | |
2990 | ||
2991 | get_mnt_ns(mnt_ns); | |
2992 | put_mnt_ns(nsproxy->mnt_ns); | |
2993 | nsproxy->mnt_ns = mnt_ns; | |
2994 | ||
2995 | /* Find the root */ | |
2996 | root.mnt = &mnt_ns->root->mnt; | |
2997 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2998 | path_get(&root); | |
2999 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3000 | ; | |
3001 | ||
3002 | /* Update the pwd and root */ | |
3003 | set_fs_pwd(fs, &root); | |
3004 | set_fs_root(fs, &root); | |
3005 | ||
3006 | path_put(&root); | |
3007 | return 0; | |
3008 | } | |
3009 | ||
98f842e6 EB |
3010 | static unsigned int mntns_inum(void *ns) |
3011 | { | |
3012 | struct mnt_namespace *mnt_ns = ns; | |
3013 | return mnt_ns->proc_inum; | |
3014 | } | |
3015 | ||
8823c079 EB |
3016 | const struct proc_ns_operations mntns_operations = { |
3017 | .name = "mnt", | |
3018 | .type = CLONE_NEWNS, | |
3019 | .get = mntns_get, | |
3020 | .put = mntns_put, | |
3021 | .install = mntns_install, | |
98f842e6 | 3022 | .inum = mntns_inum, |
8823c079 | 3023 | }; |