Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
57f150a5 | 19 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
0bb80f24 | 23 | #include <linux/proc_ns.h> |
20b4fb48 | 24 | #include <linux/magic.h> |
0818bf27 | 25 | #include <linux/bootmem.h> |
9ea459e1 | 26 | #include <linux/task_work.h> |
07b20889 | 27 | #include "pnode.h" |
948730b0 | 28 | #include "internal.h" |
1da177e4 | 29 | |
0818bf27 AV |
30 | static unsigned int m_hash_mask __read_mostly; |
31 | static unsigned int m_hash_shift __read_mostly; | |
32 | static unsigned int mp_hash_mask __read_mostly; | |
33 | static unsigned int mp_hash_shift __read_mostly; | |
34 | ||
35 | static __initdata unsigned long mhash_entries; | |
36 | static int __init set_mhash_entries(char *str) | |
37 | { | |
38 | if (!str) | |
39 | return 0; | |
40 | mhash_entries = simple_strtoul(str, &str, 0); | |
41 | return 1; | |
42 | } | |
43 | __setup("mhash_entries=", set_mhash_entries); | |
44 | ||
45 | static __initdata unsigned long mphash_entries; | |
46 | static int __init set_mphash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mphash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 54 | |
c7999c36 | 55 | static u64 event; |
73cd49ec | 56 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 57 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 58 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
59 | static int mnt_id_start = 0; |
60 | static int mnt_group_start = 1; | |
1da177e4 | 61 | |
38129a13 | 62 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 63 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 64 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 65 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 66 | |
f87fd4c2 | 67 | /* /sys/fs */ |
00d26666 GKH |
68 | struct kobject *fs_kobj; |
69 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 70 | |
99b7db7b NP |
71 | /* |
72 | * vfsmount lock may be taken for read to prevent changes to the | |
73 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
74 | * up the tree. | |
75 | * | |
76 | * It should be taken for write in all cases where the vfsmount | |
77 | * tree or hash is modified or when a vfsmount structure is modified. | |
78 | */ | |
48a066e7 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 80 | |
38129a13 | 81 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 82 | { |
b58fed8b RP |
83 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
84 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
85 | tmp = tmp + (tmp >> m_hash_shift); |
86 | return &mount_hashtable[tmp & m_hash_mask]; | |
87 | } | |
88 | ||
89 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
90 | { | |
91 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
92 | tmp = tmp + (tmp >> mp_hash_shift); | |
93 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
94 | } |
95 | ||
99b7db7b NP |
96 | /* |
97 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
98 | * serialize with freeing. | |
99 | */ | |
b105e270 | 100 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
101 | { |
102 | int res; | |
103 | ||
104 | retry: | |
105 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 106 | spin_lock(&mnt_id_lock); |
15169fe7 | 107 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 108 | if (!res) |
15169fe7 | 109 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 110 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
111 | if (res == -EAGAIN) |
112 | goto retry; | |
113 | ||
114 | return res; | |
115 | } | |
116 | ||
b105e270 | 117 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 118 | { |
15169fe7 | 119 | int id = mnt->mnt_id; |
99b7db7b | 120 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
121 | ida_remove(&mnt_id_ida, id); |
122 | if (mnt_id_start > id) | |
123 | mnt_id_start = id; | |
99b7db7b | 124 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
125 | } |
126 | ||
719f5d7f MS |
127 | /* |
128 | * Allocate a new peer group ID | |
129 | * | |
130 | * mnt_group_ida is protected by namespace_sem | |
131 | */ | |
4b8b21f4 | 132 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 133 | { |
f21f6220 AV |
134 | int res; |
135 | ||
719f5d7f MS |
136 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
137 | return -ENOMEM; | |
138 | ||
f21f6220 AV |
139 | res = ida_get_new_above(&mnt_group_ida, |
140 | mnt_group_start, | |
15169fe7 | 141 | &mnt->mnt_group_id); |
f21f6220 | 142 | if (!res) |
15169fe7 | 143 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
144 | |
145 | return res; | |
719f5d7f MS |
146 | } |
147 | ||
148 | /* | |
149 | * Release a peer group ID | |
150 | */ | |
4b8b21f4 | 151 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 152 | { |
15169fe7 | 153 | int id = mnt->mnt_group_id; |
f21f6220 AV |
154 | ida_remove(&mnt_group_ida, id); |
155 | if (mnt_group_start > id) | |
156 | mnt_group_start = id; | |
15169fe7 | 157 | mnt->mnt_group_id = 0; |
719f5d7f MS |
158 | } |
159 | ||
b3e19d92 NP |
160 | /* |
161 | * vfsmount lock must be held for read | |
162 | */ | |
83adc753 | 163 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
164 | { |
165 | #ifdef CONFIG_SMP | |
68e8a9fe | 166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
167 | #else |
168 | preempt_disable(); | |
68e8a9fe | 169 | mnt->mnt_count += n; |
b3e19d92 NP |
170 | preempt_enable(); |
171 | #endif | |
172 | } | |
173 | ||
b3e19d92 NP |
174 | /* |
175 | * vfsmount lock must be held for write | |
176 | */ | |
83adc753 | 177 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
178 | { |
179 | #ifdef CONFIG_SMP | |
f03c6599 | 180 | unsigned int count = 0; |
b3e19d92 NP |
181 | int cpu; |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
185 | } |
186 | ||
187 | return count; | |
188 | #else | |
68e8a9fe | 189 | return mnt->mnt_count; |
b3e19d92 NP |
190 | #endif |
191 | } | |
192 | ||
b105e270 | 193 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 194 | { |
c63181e6 AV |
195 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
196 | if (mnt) { | |
73cd49ec MS |
197 | int err; |
198 | ||
c63181e6 | 199 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
200 | if (err) |
201 | goto out_free_cache; | |
202 | ||
203 | if (name) { | |
c63181e6 AV |
204 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
205 | if (!mnt->mnt_devname) | |
88b38782 | 206 | goto out_free_id; |
73cd49ec MS |
207 | } |
208 | ||
b3e19d92 | 209 | #ifdef CONFIG_SMP |
c63181e6 AV |
210 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
211 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
212 | goto out_free_devname; |
213 | ||
c63181e6 | 214 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 215 | #else |
c63181e6 AV |
216 | mnt->mnt_count = 1; |
217 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
218 | #endif |
219 | ||
38129a13 | 220 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
221 | INIT_LIST_HEAD(&mnt->mnt_child); |
222 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
223 | INIT_LIST_HEAD(&mnt->mnt_list); | |
224 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
225 | INIT_LIST_HEAD(&mnt->mnt_share); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
227 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
228 | #ifdef CONFIG_FSNOTIFY |
229 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 230 | #endif |
1da177e4 | 231 | } |
c63181e6 | 232 | return mnt; |
88b38782 | 233 | |
d3ef3d73 | 234 | #ifdef CONFIG_SMP |
235 | out_free_devname: | |
c63181e6 | 236 | kfree(mnt->mnt_devname); |
d3ef3d73 | 237 | #endif |
88b38782 | 238 | out_free_id: |
c63181e6 | 239 | mnt_free_id(mnt); |
88b38782 | 240 | out_free_cache: |
c63181e6 | 241 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 242 | return NULL; |
1da177e4 LT |
243 | } |
244 | ||
3d733633 DH |
245 | /* |
246 | * Most r/o checks on a fs are for operations that take | |
247 | * discrete amounts of time, like a write() or unlink(). | |
248 | * We must keep track of when those operations start | |
249 | * (for permission checks) and when they end, so that | |
250 | * we can determine when writes are able to occur to | |
251 | * a filesystem. | |
252 | */ | |
253 | /* | |
254 | * __mnt_is_readonly: check whether a mount is read-only | |
255 | * @mnt: the mount to check for its write status | |
256 | * | |
257 | * This shouldn't be used directly ouside of the VFS. | |
258 | * It does not guarantee that the filesystem will stay | |
259 | * r/w, just that it is right *now*. This can not and | |
260 | * should not be used in place of IS_RDONLY(inode). | |
261 | * mnt_want/drop_write() will _keep_ the filesystem | |
262 | * r/w. | |
263 | */ | |
264 | int __mnt_is_readonly(struct vfsmount *mnt) | |
265 | { | |
2e4b7fcd DH |
266 | if (mnt->mnt_flags & MNT_READONLY) |
267 | return 1; | |
268 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
269 | return 1; | |
270 | return 0; | |
3d733633 DH |
271 | } |
272 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
273 | ||
83adc753 | 274 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 275 | { |
276 | #ifdef CONFIG_SMP | |
68e8a9fe | 277 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 278 | #else |
68e8a9fe | 279 | mnt->mnt_writers++; |
d3ef3d73 | 280 | #endif |
281 | } | |
3d733633 | 282 | |
83adc753 | 283 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 284 | { |
d3ef3d73 | 285 | #ifdef CONFIG_SMP |
68e8a9fe | 286 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 287 | #else |
68e8a9fe | 288 | mnt->mnt_writers--; |
d3ef3d73 | 289 | #endif |
3d733633 | 290 | } |
3d733633 | 291 | |
83adc753 | 292 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 293 | { |
d3ef3d73 | 294 | #ifdef CONFIG_SMP |
295 | unsigned int count = 0; | |
3d733633 | 296 | int cpu; |
3d733633 DH |
297 | |
298 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 299 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 300 | } |
3d733633 | 301 | |
d3ef3d73 | 302 | return count; |
303 | #else | |
304 | return mnt->mnt_writers; | |
305 | #endif | |
3d733633 DH |
306 | } |
307 | ||
4ed5e82f MS |
308 | static int mnt_is_readonly(struct vfsmount *mnt) |
309 | { | |
310 | if (mnt->mnt_sb->s_readonly_remount) | |
311 | return 1; | |
312 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
313 | smp_rmb(); | |
314 | return __mnt_is_readonly(mnt); | |
315 | } | |
316 | ||
8366025e | 317 | /* |
eb04c282 JK |
318 | * Most r/o & frozen checks on a fs are for operations that take discrete |
319 | * amounts of time, like a write() or unlink(). We must keep track of when | |
320 | * those operations start (for permission checks) and when they end, so that we | |
321 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
322 | */ |
323 | /** | |
eb04c282 | 324 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 325 | * @m: the mount on which to take a write |
8366025e | 326 | * |
eb04c282 JK |
327 | * This tells the low-level filesystem that a write is about to be performed to |
328 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
329 | * returning success. This operation does not protect against filesystem being | |
330 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
331 | * called. This is effectively a refcount. | |
8366025e | 332 | */ |
eb04c282 | 333 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 334 | { |
83adc753 | 335 | struct mount *mnt = real_mount(m); |
3d733633 | 336 | int ret = 0; |
3d733633 | 337 | |
d3ef3d73 | 338 | preempt_disable(); |
c6653a83 | 339 | mnt_inc_writers(mnt); |
d3ef3d73 | 340 | /* |
c6653a83 | 341 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 342 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
343 | * incremented count after it has set MNT_WRITE_HOLD. | |
344 | */ | |
345 | smp_mb(); | |
1e75529e | 346 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 347 | cpu_relax(); |
348 | /* | |
349 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
350 | * be set to match its requirements. So we must not load that until | |
351 | * MNT_WRITE_HOLD is cleared. | |
352 | */ | |
353 | smp_rmb(); | |
4ed5e82f | 354 | if (mnt_is_readonly(m)) { |
c6653a83 | 355 | mnt_dec_writers(mnt); |
3d733633 | 356 | ret = -EROFS; |
3d733633 | 357 | } |
d3ef3d73 | 358 | preempt_enable(); |
eb04c282 JK |
359 | |
360 | return ret; | |
361 | } | |
362 | ||
363 | /** | |
364 | * mnt_want_write - get write access to a mount | |
365 | * @m: the mount on which to take a write | |
366 | * | |
367 | * This tells the low-level filesystem that a write is about to be performed to | |
368 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
369 | * is not frozen) before returning success. When the write operation is | |
370 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
371 | */ | |
372 | int mnt_want_write(struct vfsmount *m) | |
373 | { | |
374 | int ret; | |
375 | ||
376 | sb_start_write(m->mnt_sb); | |
377 | ret = __mnt_want_write(m); | |
378 | if (ret) | |
379 | sb_end_write(m->mnt_sb); | |
3d733633 | 380 | return ret; |
8366025e DH |
381 | } |
382 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
383 | ||
96029c4e | 384 | /** |
385 | * mnt_clone_write - get write access to a mount | |
386 | * @mnt: the mount on which to take a write | |
387 | * | |
388 | * This is effectively like mnt_want_write, except | |
389 | * it must only be used to take an extra write reference | |
390 | * on a mountpoint that we already know has a write reference | |
391 | * on it. This allows some optimisation. | |
392 | * | |
393 | * After finished, mnt_drop_write must be called as usual to | |
394 | * drop the reference. | |
395 | */ | |
396 | int mnt_clone_write(struct vfsmount *mnt) | |
397 | { | |
398 | /* superblock may be r/o */ | |
399 | if (__mnt_is_readonly(mnt)) | |
400 | return -EROFS; | |
401 | preempt_disable(); | |
83adc753 | 402 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 403 | preempt_enable(); |
404 | return 0; | |
405 | } | |
406 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
407 | ||
408 | /** | |
eb04c282 | 409 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 410 | * @file: the file who's mount on which to take a write |
411 | * | |
eb04c282 | 412 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 413 | * do some optimisations if the file is open for write already |
414 | */ | |
eb04c282 | 415 | int __mnt_want_write_file(struct file *file) |
96029c4e | 416 | { |
83f936c7 | 417 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 418 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 419 | else |
420 | return mnt_clone_write(file->f_path.mnt); | |
421 | } | |
eb04c282 JK |
422 | |
423 | /** | |
424 | * mnt_want_write_file - get write access to a file's mount | |
425 | * @file: the file who's mount on which to take a write | |
426 | * | |
427 | * This is like mnt_want_write, but it takes a file and can | |
428 | * do some optimisations if the file is open for write already | |
429 | */ | |
430 | int mnt_want_write_file(struct file *file) | |
431 | { | |
432 | int ret; | |
433 | ||
434 | sb_start_write(file->f_path.mnt->mnt_sb); | |
435 | ret = __mnt_want_write_file(file); | |
436 | if (ret) | |
437 | sb_end_write(file->f_path.mnt->mnt_sb); | |
438 | return ret; | |
439 | } | |
96029c4e | 440 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
441 | ||
8366025e | 442 | /** |
eb04c282 | 443 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
444 | * @mnt: the mount on which to give up write access |
445 | * | |
446 | * Tells the low-level filesystem that we are done | |
447 | * performing writes to it. Must be matched with | |
eb04c282 | 448 | * __mnt_want_write() call above. |
8366025e | 449 | */ |
eb04c282 | 450 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 451 | { |
d3ef3d73 | 452 | preempt_disable(); |
83adc753 | 453 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 454 | preempt_enable(); |
8366025e | 455 | } |
eb04c282 JK |
456 | |
457 | /** | |
458 | * mnt_drop_write - give up write access to a mount | |
459 | * @mnt: the mount on which to give up write access | |
460 | * | |
461 | * Tells the low-level filesystem that we are done performing writes to it and | |
462 | * also allows filesystem to be frozen again. Must be matched with | |
463 | * mnt_want_write() call above. | |
464 | */ | |
465 | void mnt_drop_write(struct vfsmount *mnt) | |
466 | { | |
467 | __mnt_drop_write(mnt); | |
468 | sb_end_write(mnt->mnt_sb); | |
469 | } | |
8366025e DH |
470 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
471 | ||
eb04c282 JK |
472 | void __mnt_drop_write_file(struct file *file) |
473 | { | |
474 | __mnt_drop_write(file->f_path.mnt); | |
475 | } | |
476 | ||
2a79f17e AV |
477 | void mnt_drop_write_file(struct file *file) |
478 | { | |
479 | mnt_drop_write(file->f_path.mnt); | |
480 | } | |
481 | EXPORT_SYMBOL(mnt_drop_write_file); | |
482 | ||
83adc753 | 483 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 484 | { |
3d733633 DH |
485 | int ret = 0; |
486 | ||
719ea2fb | 487 | lock_mount_hash(); |
83adc753 | 488 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 489 | /* |
d3ef3d73 | 490 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
491 | * should be visible before we do. | |
3d733633 | 492 | */ |
d3ef3d73 | 493 | smp_mb(); |
494 | ||
3d733633 | 495 | /* |
d3ef3d73 | 496 | * With writers on hold, if this value is zero, then there are |
497 | * definitely no active writers (although held writers may subsequently | |
498 | * increment the count, they'll have to wait, and decrement it after | |
499 | * seeing MNT_READONLY). | |
500 | * | |
501 | * It is OK to have counter incremented on one CPU and decremented on | |
502 | * another: the sum will add up correctly. The danger would be when we | |
503 | * sum up each counter, if we read a counter before it is incremented, | |
504 | * but then read another CPU's count which it has been subsequently | |
505 | * decremented from -- we would see more decrements than we should. | |
506 | * MNT_WRITE_HOLD protects against this scenario, because | |
507 | * mnt_want_write first increments count, then smp_mb, then spins on | |
508 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
509 | * we're counting up here. | |
3d733633 | 510 | */ |
c6653a83 | 511 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 512 | ret = -EBUSY; |
513 | else | |
83adc753 | 514 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 515 | /* |
516 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
517 | * that become unheld will see MNT_READONLY. | |
518 | */ | |
519 | smp_wmb(); | |
83adc753 | 520 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 521 | unlock_mount_hash(); |
3d733633 | 522 | return ret; |
8366025e | 523 | } |
8366025e | 524 | |
83adc753 | 525 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 526 | { |
719ea2fb | 527 | lock_mount_hash(); |
83adc753 | 528 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 529 | unlock_mount_hash(); |
2e4b7fcd DH |
530 | } |
531 | ||
4ed5e82f MS |
532 | int sb_prepare_remount_readonly(struct super_block *sb) |
533 | { | |
534 | struct mount *mnt; | |
535 | int err = 0; | |
536 | ||
8e8b8796 MS |
537 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
538 | if (atomic_long_read(&sb->s_remove_count)) | |
539 | return -EBUSY; | |
540 | ||
719ea2fb | 541 | lock_mount_hash(); |
4ed5e82f MS |
542 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
543 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
544 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
545 | smp_mb(); | |
546 | if (mnt_get_writers(mnt) > 0) { | |
547 | err = -EBUSY; | |
548 | break; | |
549 | } | |
550 | } | |
551 | } | |
8e8b8796 MS |
552 | if (!err && atomic_long_read(&sb->s_remove_count)) |
553 | err = -EBUSY; | |
554 | ||
4ed5e82f MS |
555 | if (!err) { |
556 | sb->s_readonly_remount = 1; | |
557 | smp_wmb(); | |
558 | } | |
559 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
560 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
561 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
562 | } | |
719ea2fb | 563 | unlock_mount_hash(); |
4ed5e82f MS |
564 | |
565 | return err; | |
566 | } | |
567 | ||
b105e270 | 568 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 569 | { |
52ba1621 | 570 | kfree(mnt->mnt_devname); |
d3ef3d73 | 571 | #ifdef CONFIG_SMP |
68e8a9fe | 572 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 573 | #endif |
b105e270 | 574 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
575 | } |
576 | ||
8ffcb32e DH |
577 | static void delayed_free_vfsmnt(struct rcu_head *head) |
578 | { | |
579 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
580 | } | |
581 | ||
48a066e7 AV |
582 | /* call under rcu_read_lock */ |
583 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
584 | { | |
585 | struct mount *mnt; | |
586 | if (read_seqretry(&mount_lock, seq)) | |
587 | return false; | |
588 | if (bastard == NULL) | |
589 | return true; | |
590 | mnt = real_mount(bastard); | |
591 | mnt_add_count(mnt, 1); | |
592 | if (likely(!read_seqretry(&mount_lock, seq))) | |
593 | return true; | |
594 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
595 | mnt_add_count(mnt, -1); | |
596 | return false; | |
597 | } | |
598 | rcu_read_unlock(); | |
599 | mntput(bastard); | |
600 | rcu_read_lock(); | |
601 | return false; | |
602 | } | |
603 | ||
1da177e4 | 604 | /* |
474279dc | 605 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 606 | * call under rcu_read_lock() |
1da177e4 | 607 | */ |
474279dc | 608 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 609 | { |
38129a13 | 610 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
611 | struct mount *p; |
612 | ||
38129a13 | 613 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
614 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
615 | return p; | |
616 | return NULL; | |
617 | } | |
618 | ||
619 | /* | |
620 | * find the last mount at @dentry on vfsmount @mnt. | |
48a066e7 | 621 | * mount_lock must be held. |
474279dc AV |
622 | */ |
623 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
624 | { | |
38129a13 AV |
625 | struct mount *p, *res; |
626 | res = p = __lookup_mnt(mnt, dentry); | |
627 | if (!p) | |
628 | goto out; | |
629 | hlist_for_each_entry_continue(p, mnt_hash) { | |
1d6a32ac AV |
630 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) |
631 | break; | |
632 | res = p; | |
633 | } | |
38129a13 | 634 | out: |
1d6a32ac | 635 | return res; |
1da177e4 LT |
636 | } |
637 | ||
a05964f3 | 638 | /* |
f015f126 DH |
639 | * lookup_mnt - Return the first child mount mounted at path |
640 | * | |
641 | * "First" means first mounted chronologically. If you create the | |
642 | * following mounts: | |
643 | * | |
644 | * mount /dev/sda1 /mnt | |
645 | * mount /dev/sda2 /mnt | |
646 | * mount /dev/sda3 /mnt | |
647 | * | |
648 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
649 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
650 | * /dev/sda2, then /dev/sda3, then NULL. | |
651 | * | |
652 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 653 | */ |
1c755af4 | 654 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 655 | { |
c7105365 | 656 | struct mount *child_mnt; |
48a066e7 AV |
657 | struct vfsmount *m; |
658 | unsigned seq; | |
99b7db7b | 659 | |
48a066e7 AV |
660 | rcu_read_lock(); |
661 | do { | |
662 | seq = read_seqbegin(&mount_lock); | |
663 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
664 | m = child_mnt ? &child_mnt->mnt : NULL; | |
665 | } while (!legitimize_mnt(m, seq)); | |
666 | rcu_read_unlock(); | |
667 | return m; | |
a05964f3 RP |
668 | } |
669 | ||
84d17192 AV |
670 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
671 | { | |
0818bf27 | 672 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 | 673 | struct mountpoint *mp; |
eed81007 | 674 | int ret; |
84d17192 | 675 | |
0818bf27 | 676 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
677 | if (mp->m_dentry == dentry) { |
678 | /* might be worth a WARN_ON() */ | |
679 | if (d_unlinked(dentry)) | |
680 | return ERR_PTR(-ENOENT); | |
681 | mp->m_count++; | |
682 | return mp; | |
683 | } | |
684 | } | |
685 | ||
686 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
687 | if (!mp) | |
688 | return ERR_PTR(-ENOMEM); | |
689 | ||
eed81007 MS |
690 | ret = d_set_mounted(dentry); |
691 | if (ret) { | |
84d17192 | 692 | kfree(mp); |
eed81007 | 693 | return ERR_PTR(ret); |
84d17192 | 694 | } |
eed81007 | 695 | |
84d17192 AV |
696 | mp->m_dentry = dentry; |
697 | mp->m_count = 1; | |
0818bf27 | 698 | hlist_add_head(&mp->m_hash, chain); |
84d17192 AV |
699 | return mp; |
700 | } | |
701 | ||
702 | static void put_mountpoint(struct mountpoint *mp) | |
703 | { | |
704 | if (!--mp->m_count) { | |
705 | struct dentry *dentry = mp->m_dentry; | |
706 | spin_lock(&dentry->d_lock); | |
707 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
708 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 709 | hlist_del(&mp->m_hash); |
84d17192 AV |
710 | kfree(mp); |
711 | } | |
712 | } | |
713 | ||
143c8c91 | 714 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 715 | { |
6b3286ed | 716 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
717 | } |
718 | ||
99b7db7b NP |
719 | /* |
720 | * vfsmount lock must be held for write | |
721 | */ | |
6b3286ed | 722 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
723 | { |
724 | if (ns) { | |
725 | ns->event = ++event; | |
726 | wake_up_interruptible(&ns->poll); | |
727 | } | |
728 | } | |
729 | ||
99b7db7b NP |
730 | /* |
731 | * vfsmount lock must be held for write | |
732 | */ | |
6b3286ed | 733 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
734 | { |
735 | if (ns && ns->event != event) { | |
736 | ns->event = event; | |
737 | wake_up_interruptible(&ns->poll); | |
738 | } | |
739 | } | |
740 | ||
99b7db7b NP |
741 | /* |
742 | * vfsmount lock must be held for write | |
743 | */ | |
419148da AV |
744 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
745 | { | |
a73324da | 746 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
747 | old_path->mnt = &mnt->mnt_parent->mnt; |
748 | mnt->mnt_parent = mnt; | |
a73324da | 749 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 750 | list_del_init(&mnt->mnt_child); |
38129a13 | 751 | hlist_del_init_rcu(&mnt->mnt_hash); |
84d17192 AV |
752 | put_mountpoint(mnt->mnt_mp); |
753 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
754 | } |
755 | ||
99b7db7b NP |
756 | /* |
757 | * vfsmount lock must be held for write | |
758 | */ | |
84d17192 AV |
759 | void mnt_set_mountpoint(struct mount *mnt, |
760 | struct mountpoint *mp, | |
44d964d6 | 761 | struct mount *child_mnt) |
b90fa9ae | 762 | { |
84d17192 | 763 | mp->m_count++; |
3a2393d7 | 764 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 765 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 766 | child_mnt->mnt_parent = mnt; |
84d17192 | 767 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
768 | } |
769 | ||
99b7db7b NP |
770 | /* |
771 | * vfsmount lock must be held for write | |
772 | */ | |
84d17192 AV |
773 | static void attach_mnt(struct mount *mnt, |
774 | struct mount *parent, | |
775 | struct mountpoint *mp) | |
1da177e4 | 776 | { |
84d17192 | 777 | mnt_set_mountpoint(parent, mp, mnt); |
38129a13 | 778 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); |
84d17192 | 779 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
b90fa9ae RP |
780 | } |
781 | ||
12a5b529 AV |
782 | static void attach_shadowed(struct mount *mnt, |
783 | struct mount *parent, | |
784 | struct mount *shadows) | |
785 | { | |
786 | if (shadows) { | |
f6f99332 | 787 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); |
12a5b529 AV |
788 | list_add(&mnt->mnt_child, &shadows->mnt_child); |
789 | } else { | |
790 | hlist_add_head_rcu(&mnt->mnt_hash, | |
791 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
792 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
793 | } | |
794 | } | |
795 | ||
b90fa9ae | 796 | /* |
99b7db7b | 797 | * vfsmount lock must be held for write |
b90fa9ae | 798 | */ |
1d6a32ac | 799 | static void commit_tree(struct mount *mnt, struct mount *shadows) |
b90fa9ae | 800 | { |
0714a533 | 801 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 802 | struct mount *m; |
b90fa9ae | 803 | LIST_HEAD(head); |
143c8c91 | 804 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 805 | |
0714a533 | 806 | BUG_ON(parent == mnt); |
b90fa9ae | 807 | |
1a4eeaf2 | 808 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 809 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 810 | m->mnt_ns = n; |
f03c6599 | 811 | |
b90fa9ae RP |
812 | list_splice(&head, n->list.prev); |
813 | ||
12a5b529 | 814 | attach_shadowed(mnt, parent, shadows); |
6b3286ed | 815 | touch_mnt_namespace(n); |
1da177e4 LT |
816 | } |
817 | ||
909b0a88 | 818 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 819 | { |
6b41d536 AV |
820 | struct list_head *next = p->mnt_mounts.next; |
821 | if (next == &p->mnt_mounts) { | |
1da177e4 | 822 | while (1) { |
909b0a88 | 823 | if (p == root) |
1da177e4 | 824 | return NULL; |
6b41d536 AV |
825 | next = p->mnt_child.next; |
826 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 827 | break; |
0714a533 | 828 | p = p->mnt_parent; |
1da177e4 LT |
829 | } |
830 | } | |
6b41d536 | 831 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
832 | } |
833 | ||
315fc83e | 834 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 835 | { |
6b41d536 AV |
836 | struct list_head *prev = p->mnt_mounts.prev; |
837 | while (prev != &p->mnt_mounts) { | |
838 | p = list_entry(prev, struct mount, mnt_child); | |
839 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
840 | } |
841 | return p; | |
842 | } | |
843 | ||
9d412a43 AV |
844 | struct vfsmount * |
845 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
846 | { | |
b105e270 | 847 | struct mount *mnt; |
9d412a43 AV |
848 | struct dentry *root; |
849 | ||
850 | if (!type) | |
851 | return ERR_PTR(-ENODEV); | |
852 | ||
853 | mnt = alloc_vfsmnt(name); | |
854 | if (!mnt) | |
855 | return ERR_PTR(-ENOMEM); | |
856 | ||
857 | if (flags & MS_KERNMOUNT) | |
b105e270 | 858 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
859 | |
860 | root = mount_fs(type, flags, name, data); | |
861 | if (IS_ERR(root)) { | |
8ffcb32e | 862 | mnt_free_id(mnt); |
9d412a43 AV |
863 | free_vfsmnt(mnt); |
864 | return ERR_CAST(root); | |
865 | } | |
866 | ||
b105e270 AV |
867 | mnt->mnt.mnt_root = root; |
868 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 869 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 870 | mnt->mnt_parent = mnt; |
719ea2fb | 871 | lock_mount_hash(); |
39f7c4db | 872 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 873 | unlock_mount_hash(); |
b105e270 | 874 | return &mnt->mnt; |
9d412a43 AV |
875 | } |
876 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
877 | ||
87129cc0 | 878 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 879 | int flag) |
1da177e4 | 880 | { |
87129cc0 | 881 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
882 | struct mount *mnt; |
883 | int err; | |
1da177e4 | 884 | |
be34d1a3 DH |
885 | mnt = alloc_vfsmnt(old->mnt_devname); |
886 | if (!mnt) | |
887 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 888 | |
7a472ef4 | 889 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
890 | mnt->mnt_group_id = 0; /* not a peer of original */ |
891 | else | |
892 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 893 | |
be34d1a3 DH |
894 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
895 | err = mnt_alloc_group_id(mnt); | |
896 | if (err) | |
897 | goto out_free; | |
1da177e4 | 898 | } |
be34d1a3 | 899 | |
f2ebb3a9 | 900 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 901 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
902 | if (flag & CL_UNPRIVILEGED) { |
903 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
904 | ||
905 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
906 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
907 | ||
908 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
909 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
910 | ||
911 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
912 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
913 | ||
914 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
915 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
916 | } | |
132c94e3 | 917 | |
5ff9d8a6 EB |
918 | /* Don't allow unprivileged users to reveal what is under a mount */ |
919 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
920 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
921 | ||
be34d1a3 DH |
922 | atomic_inc(&sb->s_active); |
923 | mnt->mnt.mnt_sb = sb; | |
924 | mnt->mnt.mnt_root = dget(root); | |
925 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
926 | mnt->mnt_parent = mnt; | |
719ea2fb | 927 | lock_mount_hash(); |
be34d1a3 | 928 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 929 | unlock_mount_hash(); |
be34d1a3 | 930 | |
7a472ef4 EB |
931 | if ((flag & CL_SLAVE) || |
932 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
933 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
934 | mnt->mnt_master = old; | |
935 | CLEAR_MNT_SHARED(mnt); | |
936 | } else if (!(flag & CL_PRIVATE)) { | |
937 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
938 | list_add(&mnt->mnt_share, &old->mnt_share); | |
939 | if (IS_MNT_SLAVE(old)) | |
940 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
941 | mnt->mnt_master = old->mnt_master; | |
942 | } | |
943 | if (flag & CL_MAKE_SHARED) | |
944 | set_mnt_shared(mnt); | |
945 | ||
946 | /* stick the duplicate mount on the same expiry list | |
947 | * as the original if that was on one */ | |
948 | if (flag & CL_EXPIRE) { | |
949 | if (!list_empty(&old->mnt_expire)) | |
950 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
951 | } | |
952 | ||
cb338d06 | 953 | return mnt; |
719f5d7f MS |
954 | |
955 | out_free: | |
8ffcb32e | 956 | mnt_free_id(mnt); |
719f5d7f | 957 | free_vfsmnt(mnt); |
be34d1a3 | 958 | return ERR_PTR(err); |
1da177e4 LT |
959 | } |
960 | ||
9ea459e1 AV |
961 | static void cleanup_mnt(struct mount *mnt) |
962 | { | |
963 | /* | |
964 | * This probably indicates that somebody messed | |
965 | * up a mnt_want/drop_write() pair. If this | |
966 | * happens, the filesystem was probably unable | |
967 | * to make r/w->r/o transitions. | |
968 | */ | |
969 | /* | |
970 | * The locking used to deal with mnt_count decrement provides barriers, | |
971 | * so mnt_get_writers() below is safe. | |
972 | */ | |
973 | WARN_ON(mnt_get_writers(mnt)); | |
974 | if (unlikely(mnt->mnt_pins.first)) | |
975 | mnt_pin_kill(mnt); | |
976 | fsnotify_vfsmount_delete(&mnt->mnt); | |
977 | dput(mnt->mnt.mnt_root); | |
978 | deactivate_super(mnt->mnt.mnt_sb); | |
979 | mnt_free_id(mnt); | |
980 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
981 | } | |
982 | ||
983 | static void __cleanup_mnt(struct rcu_head *head) | |
984 | { | |
985 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
986 | } | |
987 | ||
988 | static LLIST_HEAD(delayed_mntput_list); | |
989 | static void delayed_mntput(struct work_struct *unused) | |
990 | { | |
991 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
992 | struct llist_node *next; | |
993 | ||
994 | for (; node; node = next) { | |
995 | next = llist_next(node); | |
996 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
997 | } | |
998 | } | |
999 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1000 | ||
900148dc | 1001 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1002 | { |
48a066e7 AV |
1003 | rcu_read_lock(); |
1004 | mnt_add_count(mnt, -1); | |
1005 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1006 | rcu_read_unlock(); | |
f03c6599 | 1007 | return; |
b3e19d92 | 1008 | } |
719ea2fb | 1009 | lock_mount_hash(); |
b3e19d92 | 1010 | if (mnt_get_count(mnt)) { |
48a066e7 | 1011 | rcu_read_unlock(); |
719ea2fb | 1012 | unlock_mount_hash(); |
99b7db7b NP |
1013 | return; |
1014 | } | |
48a066e7 AV |
1015 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1016 | rcu_read_unlock(); | |
1017 | unlock_mount_hash(); | |
1018 | return; | |
1019 | } | |
1020 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1021 | rcu_read_unlock(); | |
962830df | 1022 | |
39f7c4db | 1023 | list_del(&mnt->mnt_instance); |
719ea2fb | 1024 | unlock_mount_hash(); |
649a795a | 1025 | |
9ea459e1 AV |
1026 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1027 | struct task_struct *task = current; | |
1028 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1029 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1030 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1031 | return; | |
1032 | } | |
1033 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1034 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1035 | return; | |
1036 | } | |
1037 | cleanup_mnt(mnt); | |
b3e19d92 | 1038 | } |
b3e19d92 NP |
1039 | |
1040 | void mntput(struct vfsmount *mnt) | |
1041 | { | |
1042 | if (mnt) { | |
863d684f | 1043 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1044 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1045 | if (unlikely(m->mnt_expiry_mark)) |
1046 | m->mnt_expiry_mark = 0; | |
1047 | mntput_no_expire(m); | |
b3e19d92 NP |
1048 | } |
1049 | } | |
1050 | EXPORT_SYMBOL(mntput); | |
1051 | ||
1052 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1053 | { | |
1054 | if (mnt) | |
83adc753 | 1055 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1056 | return mnt; |
1057 | } | |
1058 | EXPORT_SYMBOL(mntget); | |
1059 | ||
3064c356 | 1060 | struct vfsmount *mnt_clone_internal(struct path *path) |
7b7b1ace | 1061 | { |
3064c356 AV |
1062 | struct mount *p; |
1063 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1064 | if (IS_ERR(p)) | |
1065 | return ERR_CAST(p); | |
1066 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1067 | return &p->mnt; | |
7b7b1ace | 1068 | } |
1da177e4 | 1069 | |
b3b304a2 MS |
1070 | static inline void mangle(struct seq_file *m, const char *s) |
1071 | { | |
1072 | seq_escape(m, s, " \t\n\\"); | |
1073 | } | |
1074 | ||
1075 | /* | |
1076 | * Simple .show_options callback for filesystems which don't want to | |
1077 | * implement more complex mount option showing. | |
1078 | * | |
1079 | * See also save_mount_options(). | |
1080 | */ | |
34c80b1d | 1081 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 1082 | { |
2a32cebd AV |
1083 | const char *options; |
1084 | ||
1085 | rcu_read_lock(); | |
34c80b1d | 1086 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
1087 | |
1088 | if (options != NULL && options[0]) { | |
1089 | seq_putc(m, ','); | |
1090 | mangle(m, options); | |
1091 | } | |
2a32cebd | 1092 | rcu_read_unlock(); |
b3b304a2 MS |
1093 | |
1094 | return 0; | |
1095 | } | |
1096 | EXPORT_SYMBOL(generic_show_options); | |
1097 | ||
1098 | /* | |
1099 | * If filesystem uses generic_show_options(), this function should be | |
1100 | * called from the fill_super() callback. | |
1101 | * | |
1102 | * The .remount_fs callback usually needs to be handled in a special | |
1103 | * way, to make sure, that previous options are not overwritten if the | |
1104 | * remount fails. | |
1105 | * | |
1106 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1107 | * reset all options to their default value, but changes only newly | |
1108 | * given options, then the displayed options will not reflect reality | |
1109 | * any more. | |
1110 | */ | |
1111 | void save_mount_options(struct super_block *sb, char *options) | |
1112 | { | |
2a32cebd AV |
1113 | BUG_ON(sb->s_options); |
1114 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1115 | } |
1116 | EXPORT_SYMBOL(save_mount_options); | |
1117 | ||
2a32cebd AV |
1118 | void replace_mount_options(struct super_block *sb, char *options) |
1119 | { | |
1120 | char *old = sb->s_options; | |
1121 | rcu_assign_pointer(sb->s_options, options); | |
1122 | if (old) { | |
1123 | synchronize_rcu(); | |
1124 | kfree(old); | |
1125 | } | |
1126 | } | |
1127 | EXPORT_SYMBOL(replace_mount_options); | |
1128 | ||
a1a2c409 | 1129 | #ifdef CONFIG_PROC_FS |
0226f492 | 1130 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1131 | static void *m_start(struct seq_file *m, loff_t *pos) |
1132 | { | |
6ce6e24e | 1133 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1134 | |
390c6843 | 1135 | down_read(&namespace_sem); |
c7999c36 AV |
1136 | if (p->cached_event == p->ns->event) { |
1137 | void *v = p->cached_mount; | |
1138 | if (*pos == p->cached_index) | |
1139 | return v; | |
1140 | if (*pos == p->cached_index + 1) { | |
1141 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1142 | return p->cached_mount = v; | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | p->cached_event = p->ns->event; | |
1147 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1148 | p->cached_index = *pos; | |
1149 | return p->cached_mount; | |
1da177e4 LT |
1150 | } |
1151 | ||
1152 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1153 | { | |
6ce6e24e | 1154 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1155 | |
c7999c36 AV |
1156 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1157 | p->cached_index = *pos; | |
1158 | return p->cached_mount; | |
1da177e4 LT |
1159 | } |
1160 | ||
1161 | static void m_stop(struct seq_file *m, void *v) | |
1162 | { | |
390c6843 | 1163 | up_read(&namespace_sem); |
1da177e4 LT |
1164 | } |
1165 | ||
0226f492 | 1166 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1167 | { |
6ce6e24e | 1168 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1169 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1170 | return p->show(m, &r->mnt); |
1da177e4 LT |
1171 | } |
1172 | ||
a1a2c409 | 1173 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1174 | .start = m_start, |
1175 | .next = m_next, | |
1176 | .stop = m_stop, | |
0226f492 | 1177 | .show = m_show, |
b4629fe2 | 1178 | }; |
a1a2c409 | 1179 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1180 | |
1da177e4 LT |
1181 | /** |
1182 | * may_umount_tree - check if a mount tree is busy | |
1183 | * @mnt: root of mount tree | |
1184 | * | |
1185 | * This is called to check if a tree of mounts has any | |
1186 | * open files, pwds, chroots or sub mounts that are | |
1187 | * busy. | |
1188 | */ | |
909b0a88 | 1189 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1190 | { |
909b0a88 | 1191 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1192 | int actual_refs = 0; |
1193 | int minimum_refs = 0; | |
315fc83e | 1194 | struct mount *p; |
909b0a88 | 1195 | BUG_ON(!m); |
1da177e4 | 1196 | |
b3e19d92 | 1197 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1198 | lock_mount_hash(); |
909b0a88 | 1199 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1200 | actual_refs += mnt_get_count(p); |
1da177e4 | 1201 | minimum_refs += 2; |
1da177e4 | 1202 | } |
719ea2fb | 1203 | unlock_mount_hash(); |
1da177e4 LT |
1204 | |
1205 | if (actual_refs > minimum_refs) | |
e3474a8e | 1206 | return 0; |
1da177e4 | 1207 | |
e3474a8e | 1208 | return 1; |
1da177e4 LT |
1209 | } |
1210 | ||
1211 | EXPORT_SYMBOL(may_umount_tree); | |
1212 | ||
1213 | /** | |
1214 | * may_umount - check if a mount point is busy | |
1215 | * @mnt: root of mount | |
1216 | * | |
1217 | * This is called to check if a mount point has any | |
1218 | * open files, pwds, chroots or sub mounts. If the | |
1219 | * mount has sub mounts this will return busy | |
1220 | * regardless of whether the sub mounts are busy. | |
1221 | * | |
1222 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1223 | * give false negatives. The main reason why it's here is that we need | |
1224 | * a non-destructive way to look for easily umountable filesystems. | |
1225 | */ | |
1226 | int may_umount(struct vfsmount *mnt) | |
1227 | { | |
e3474a8e | 1228 | int ret = 1; |
8ad08d8a | 1229 | down_read(&namespace_sem); |
719ea2fb | 1230 | lock_mount_hash(); |
1ab59738 | 1231 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1232 | ret = 0; |
719ea2fb | 1233 | unlock_mount_hash(); |
8ad08d8a | 1234 | up_read(&namespace_sem); |
a05964f3 | 1235 | return ret; |
1da177e4 LT |
1236 | } |
1237 | ||
1238 | EXPORT_SYMBOL(may_umount); | |
1239 | ||
38129a13 | 1240 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1241 | |
97216be0 | 1242 | static void namespace_unlock(void) |
70fbcdf4 | 1243 | { |
d5e50f74 | 1244 | struct mount *mnt; |
38129a13 | 1245 | struct hlist_head head = unmounted; |
97216be0 | 1246 | |
38129a13 | 1247 | if (likely(hlist_empty(&head))) { |
97216be0 AV |
1248 | up_write(&namespace_sem); |
1249 | return; | |
1250 | } | |
1251 | ||
38129a13 AV |
1252 | head.first->pprev = &head.first; |
1253 | INIT_HLIST_HEAD(&unmounted); | |
1254 | ||
81b6b061 AV |
1255 | /* undo decrements we'd done in umount_tree() */ |
1256 | hlist_for_each_entry(mnt, &head, mnt_hash) | |
1257 | if (mnt->mnt_ex_mountpoint.mnt) | |
1258 | mntget(mnt->mnt_ex_mountpoint.mnt); | |
1259 | ||
97216be0 AV |
1260 | up_write(&namespace_sem); |
1261 | ||
48a066e7 AV |
1262 | synchronize_rcu(); |
1263 | ||
38129a13 AV |
1264 | while (!hlist_empty(&head)) { |
1265 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1266 | hlist_del_init(&mnt->mnt_hash); | |
aba809cf AV |
1267 | if (mnt->mnt_ex_mountpoint.mnt) |
1268 | path_put(&mnt->mnt_ex_mountpoint); | |
d5e50f74 | 1269 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1270 | } |
1271 | } | |
1272 | ||
97216be0 | 1273 | static inline void namespace_lock(void) |
e3197d83 | 1274 | { |
97216be0 | 1275 | down_write(&namespace_sem); |
e3197d83 AV |
1276 | } |
1277 | ||
99b7db7b | 1278 | /* |
48a066e7 | 1279 | * mount_lock must be held |
99b7db7b | 1280 | * namespace_sem must be held for write |
48a066e7 AV |
1281 | * how = 0 => just this tree, don't propagate |
1282 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1283 | * how = 2 => lazy umount | |
99b7db7b | 1284 | */ |
48a066e7 | 1285 | void umount_tree(struct mount *mnt, int how) |
1da177e4 | 1286 | { |
38129a13 | 1287 | HLIST_HEAD(tmp_list); |
315fc83e | 1288 | struct mount *p; |
38129a13 | 1289 | struct mount *last = NULL; |
1da177e4 | 1290 | |
38129a13 AV |
1291 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1292 | hlist_del_init_rcu(&p->mnt_hash); | |
1293 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1294 | } | |
1da177e4 | 1295 | |
88b368f2 AV |
1296 | hlist_for_each_entry(p, &tmp_list, mnt_hash) |
1297 | list_del_init(&p->mnt_child); | |
1298 | ||
48a066e7 | 1299 | if (how) |
7b8a53fd | 1300 | propagate_umount(&tmp_list); |
a05964f3 | 1301 | |
38129a13 | 1302 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1303 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1304 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1305 | __touch_mnt_namespace(p->mnt_ns); |
1306 | p->mnt_ns = NULL; | |
48a066e7 AV |
1307 | if (how < 2) |
1308 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
676da58d | 1309 | if (mnt_has_parent(p)) { |
84d17192 | 1310 | put_mountpoint(p->mnt_mp); |
81b6b061 | 1311 | mnt_add_count(p->mnt_parent, -1); |
aba809cf AV |
1312 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ |
1313 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1314 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1315 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1316 | p->mnt_parent = p; | |
84d17192 | 1317 | p->mnt_mp = NULL; |
7c4b93d8 | 1318 | } |
0f0afb1d | 1319 | change_mnt_propagation(p, MS_PRIVATE); |
38129a13 AV |
1320 | last = p; |
1321 | } | |
1322 | if (last) { | |
1323 | last->mnt_hash.next = unmounted.first; | |
1324 | unmounted.first = tmp_list.first; | |
1325 | unmounted.first->pprev = &unmounted.first; | |
1da177e4 LT |
1326 | } |
1327 | } | |
1328 | ||
b54b9be7 | 1329 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1330 | |
1ab59738 | 1331 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1332 | { |
1ab59738 | 1333 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1334 | int retval; |
1335 | ||
1ab59738 | 1336 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1337 | if (retval) |
1338 | return retval; | |
1339 | ||
1340 | /* | |
1341 | * Allow userspace to request a mountpoint be expired rather than | |
1342 | * unmounting unconditionally. Unmount only happens if: | |
1343 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1344 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1345 | */ | |
1346 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1347 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1348 | flags & (MNT_FORCE | MNT_DETACH)) |
1349 | return -EINVAL; | |
1350 | ||
b3e19d92 NP |
1351 | /* |
1352 | * probably don't strictly need the lock here if we examined | |
1353 | * all race cases, but it's a slowpath. | |
1354 | */ | |
719ea2fb | 1355 | lock_mount_hash(); |
83adc753 | 1356 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1357 | unlock_mount_hash(); |
1da177e4 | 1358 | return -EBUSY; |
b3e19d92 | 1359 | } |
719ea2fb | 1360 | unlock_mount_hash(); |
1da177e4 | 1361 | |
863d684f | 1362 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1363 | return -EAGAIN; |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * If we may have to abort operations to get out of this | |
1368 | * mount, and they will themselves hold resources we must | |
1369 | * allow the fs to do things. In the Unix tradition of | |
1370 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1371 | * might fail to complete on the first run through as other tasks | |
1372 | * must return, and the like. Thats for the mount program to worry | |
1373 | * about for the moment. | |
1374 | */ | |
1375 | ||
42faad99 | 1376 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1377 | sb->s_op->umount_begin(sb); |
42faad99 | 1378 | } |
1da177e4 LT |
1379 | |
1380 | /* | |
1381 | * No sense to grab the lock for this test, but test itself looks | |
1382 | * somewhat bogus. Suggestions for better replacement? | |
1383 | * Ho-hum... In principle, we might treat that as umount + switch | |
1384 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1385 | * Actually it makes sense, especially if rootfs would contain a | |
1386 | * /reboot - static binary that would close all descriptors and | |
1387 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1388 | */ | |
1ab59738 | 1389 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1390 | /* |
1391 | * Special case for "unmounting" root ... | |
1392 | * we just try to remount it readonly. | |
1393 | */ | |
1394 | down_write(&sb->s_umount); | |
4aa98cf7 | 1395 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1396 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1397 | up_write(&sb->s_umount); |
1398 | return retval; | |
1399 | } | |
1400 | ||
97216be0 | 1401 | namespace_lock(); |
719ea2fb | 1402 | lock_mount_hash(); |
5addc5dd | 1403 | event++; |
1da177e4 | 1404 | |
48a066e7 | 1405 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1406 | if (!list_empty(&mnt->mnt_list)) |
48a066e7 | 1407 | umount_tree(mnt, 2); |
1da177e4 | 1408 | retval = 0; |
48a066e7 AV |
1409 | } else { |
1410 | shrink_submounts(mnt); | |
1411 | retval = -EBUSY; | |
1412 | if (!propagate_mount_busy(mnt, 2)) { | |
1413 | if (!list_empty(&mnt->mnt_list)) | |
1414 | umount_tree(mnt, 1); | |
1415 | retval = 0; | |
1416 | } | |
1da177e4 | 1417 | } |
719ea2fb | 1418 | unlock_mount_hash(); |
e3197d83 | 1419 | namespace_unlock(); |
1da177e4 LT |
1420 | return retval; |
1421 | } | |
1422 | ||
9b40bc90 AV |
1423 | /* |
1424 | * Is the caller allowed to modify his namespace? | |
1425 | */ | |
1426 | static inline bool may_mount(void) | |
1427 | { | |
1428 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1429 | } | |
1430 | ||
1da177e4 LT |
1431 | /* |
1432 | * Now umount can handle mount points as well as block devices. | |
1433 | * This is important for filesystems which use unnamed block devices. | |
1434 | * | |
1435 | * We now support a flag for forced unmount like the other 'big iron' | |
1436 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1437 | */ | |
1438 | ||
bdc480e3 | 1439 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1440 | { |
2d8f3038 | 1441 | struct path path; |
900148dc | 1442 | struct mount *mnt; |
1da177e4 | 1443 | int retval; |
db1f05bb | 1444 | int lookup_flags = 0; |
1da177e4 | 1445 | |
db1f05bb MS |
1446 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1447 | return -EINVAL; | |
1448 | ||
9b40bc90 AV |
1449 | if (!may_mount()) |
1450 | return -EPERM; | |
1451 | ||
db1f05bb MS |
1452 | if (!(flags & UMOUNT_NOFOLLOW)) |
1453 | lookup_flags |= LOOKUP_FOLLOW; | |
1454 | ||
197df04c | 1455 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1456 | if (retval) |
1457 | goto out; | |
900148dc | 1458 | mnt = real_mount(path.mnt); |
1da177e4 | 1459 | retval = -EINVAL; |
2d8f3038 | 1460 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1461 | goto dput_and_out; |
143c8c91 | 1462 | if (!check_mnt(mnt)) |
1da177e4 | 1463 | goto dput_and_out; |
5ff9d8a6 EB |
1464 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1465 | goto dput_and_out; | |
1da177e4 | 1466 | |
900148dc | 1467 | retval = do_umount(mnt, flags); |
1da177e4 | 1468 | dput_and_out: |
429731b1 | 1469 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1470 | dput(path.dentry); |
900148dc | 1471 | mntput_no_expire(mnt); |
1da177e4 LT |
1472 | out: |
1473 | return retval; | |
1474 | } | |
1475 | ||
1476 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1477 | ||
1478 | /* | |
b58fed8b | 1479 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1480 | */ |
bdc480e3 | 1481 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1482 | { |
b58fed8b | 1483 | return sys_umount(name, 0); |
1da177e4 LT |
1484 | } |
1485 | ||
1486 | #endif | |
1487 | ||
4ce5d2b1 | 1488 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1489 | { |
4ce5d2b1 EB |
1490 | /* Is this a proxy for a mount namespace? */ |
1491 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1492 | struct proc_ns *ei; |
8823c079 EB |
1493 | |
1494 | if (!proc_ns_inode(inode)) | |
1495 | return false; | |
1496 | ||
0bb80f24 | 1497 | ei = get_proc_ns(inode); |
8823c079 EB |
1498 | if (ei->ns_ops != &mntns_operations) |
1499 | return false; | |
1500 | ||
4ce5d2b1 EB |
1501 | return true; |
1502 | } | |
1503 | ||
1504 | static bool mnt_ns_loop(struct dentry *dentry) | |
1505 | { | |
1506 | /* Could bind mounting the mount namespace inode cause a | |
1507 | * mount namespace loop? | |
1508 | */ | |
1509 | struct mnt_namespace *mnt_ns; | |
1510 | if (!is_mnt_ns_file(dentry)) | |
1511 | return false; | |
1512 | ||
1513 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1514 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1515 | } | |
1516 | ||
87129cc0 | 1517 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1518 | int flag) |
1da177e4 | 1519 | { |
84d17192 | 1520 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1521 | |
4ce5d2b1 EB |
1522 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1523 | return ERR_PTR(-EINVAL); | |
1524 | ||
1525 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1526 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1527 | |
36341f64 | 1528 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1529 | if (IS_ERR(q)) |
1530 | return q; | |
1531 | ||
5ff9d8a6 | 1532 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1533 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1534 | |
1535 | p = mnt; | |
6b41d536 | 1536 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1537 | struct mount *s; |
7ec02ef1 | 1538 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1539 | continue; |
1540 | ||
909b0a88 | 1541 | for (s = r; s; s = next_mnt(s, r)) { |
12a5b529 | 1542 | struct mount *t = NULL; |
4ce5d2b1 EB |
1543 | if (!(flag & CL_COPY_UNBINDABLE) && |
1544 | IS_MNT_UNBINDABLE(s)) { | |
1545 | s = skip_mnt_tree(s); | |
1546 | continue; | |
1547 | } | |
1548 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1549 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1550 | s = skip_mnt_tree(s); |
1551 | continue; | |
1552 | } | |
0714a533 AV |
1553 | while (p != s->mnt_parent) { |
1554 | p = p->mnt_parent; | |
1555 | q = q->mnt_parent; | |
1da177e4 | 1556 | } |
87129cc0 | 1557 | p = s; |
84d17192 | 1558 | parent = q; |
87129cc0 | 1559 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1560 | if (IS_ERR(q)) |
1561 | goto out; | |
719ea2fb | 1562 | lock_mount_hash(); |
1a4eeaf2 | 1563 | list_add_tail(&q->mnt_list, &res->mnt_list); |
12a5b529 AV |
1564 | mnt_set_mountpoint(parent, p->mnt_mp, q); |
1565 | if (!list_empty(&parent->mnt_mounts)) { | |
1566 | t = list_last_entry(&parent->mnt_mounts, | |
1567 | struct mount, mnt_child); | |
1568 | if (t->mnt_mp != p->mnt_mp) | |
1569 | t = NULL; | |
1570 | } | |
1571 | attach_shadowed(q, parent, t); | |
719ea2fb | 1572 | unlock_mount_hash(); |
1da177e4 LT |
1573 | } |
1574 | } | |
1575 | return res; | |
be34d1a3 | 1576 | out: |
1da177e4 | 1577 | if (res) { |
719ea2fb | 1578 | lock_mount_hash(); |
328e6d90 | 1579 | umount_tree(res, 0); |
719ea2fb | 1580 | unlock_mount_hash(); |
1da177e4 | 1581 | } |
be34d1a3 | 1582 | return q; |
1da177e4 LT |
1583 | } |
1584 | ||
be34d1a3 DH |
1585 | /* Caller should check returned pointer for errors */ |
1586 | ||
589ff870 | 1587 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1588 | { |
cb338d06 | 1589 | struct mount *tree; |
97216be0 | 1590 | namespace_lock(); |
87129cc0 AV |
1591 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1592 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1593 | namespace_unlock(); |
be34d1a3 | 1594 | if (IS_ERR(tree)) |
52e220d3 | 1595 | return ERR_CAST(tree); |
be34d1a3 | 1596 | return &tree->mnt; |
8aec0809 AV |
1597 | } |
1598 | ||
1599 | void drop_collected_mounts(struct vfsmount *mnt) | |
1600 | { | |
97216be0 | 1601 | namespace_lock(); |
719ea2fb | 1602 | lock_mount_hash(); |
328e6d90 | 1603 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1604 | unlock_mount_hash(); |
3ab6abee | 1605 | namespace_unlock(); |
8aec0809 AV |
1606 | } |
1607 | ||
1f707137 AV |
1608 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1609 | struct vfsmount *root) | |
1610 | { | |
1a4eeaf2 | 1611 | struct mount *mnt; |
1f707137 AV |
1612 | int res = f(root, arg); |
1613 | if (res) | |
1614 | return res; | |
1a4eeaf2 AV |
1615 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1616 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1617 | if (res) |
1618 | return res; | |
1619 | } | |
1620 | return 0; | |
1621 | } | |
1622 | ||
4b8b21f4 | 1623 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1624 | { |
315fc83e | 1625 | struct mount *p; |
719f5d7f | 1626 | |
909b0a88 | 1627 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1628 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1629 | mnt_release_group_id(p); |
719f5d7f MS |
1630 | } |
1631 | } | |
1632 | ||
4b8b21f4 | 1633 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1634 | { |
315fc83e | 1635 | struct mount *p; |
719f5d7f | 1636 | |
909b0a88 | 1637 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1638 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1639 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1640 | if (err) { |
4b8b21f4 | 1641 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1642 | return err; |
1643 | } | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | return 0; | |
1648 | } | |
1649 | ||
b90fa9ae RP |
1650 | /* |
1651 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1652 | * @nd : place the mount tree @source_mnt is attached |
1653 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1654 | * store the parent mount and mountpoint dentry. | |
1655 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1656 | * |
1657 | * NOTE: in the table below explains the semantics when a source mount | |
1658 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1659 | * --------------------------------------------------------------------------- |
1660 | * | BIND MOUNT OPERATION | | |
1661 | * |************************************************************************** | |
1662 | * | source-->| shared | private | slave | unbindable | | |
1663 | * | dest | | | | | | |
1664 | * | | | | | | | | |
1665 | * | v | | | | | | |
1666 | * |************************************************************************** | |
1667 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1668 | * | | | | | | | |
1669 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1670 | * *************************************************************************** | |
b90fa9ae RP |
1671 | * A bind operation clones the source mount and mounts the clone on the |
1672 | * destination mount. | |
1673 | * | |
1674 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1675 | * tree of the destination mount and the cloned mount is added to | |
1676 | * the peer group of the source mount. | |
1677 | * (+) the cloned mount is created under the destination mount and is marked | |
1678 | * as shared. The cloned mount is added to the peer group of the source | |
1679 | * mount. | |
5afe0022 RP |
1680 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1681 | * of the destination mount and the cloned mount is made slave | |
1682 | * of the same master as that of the source mount. The cloned mount | |
1683 | * is marked as 'shared and slave'. | |
1684 | * (*) the cloned mount is made a slave of the same master as that of the | |
1685 | * source mount. | |
1686 | * | |
9676f0c6 RP |
1687 | * --------------------------------------------------------------------------- |
1688 | * | MOVE MOUNT OPERATION | | |
1689 | * |************************************************************************** | |
1690 | * | source-->| shared | private | slave | unbindable | | |
1691 | * | dest | | | | | | |
1692 | * | | | | | | | | |
1693 | * | v | | | | | | |
1694 | * |************************************************************************** | |
1695 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1696 | * | | | | | | | |
1697 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1698 | * *************************************************************************** | |
5afe0022 RP |
1699 | * |
1700 | * (+) the mount is moved to the destination. And is then propagated to | |
1701 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1702 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1703 | * (+++) the mount is moved to the destination and is then propagated to |
1704 | * all the mounts belonging to the destination mount's propagation tree. | |
1705 | * the mount is marked as 'shared and slave'. | |
1706 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1707 | * |
1708 | * if the source mount is a tree, the operations explained above is | |
1709 | * applied to each mount in the tree. | |
1710 | * Must be called without spinlocks held, since this function can sleep | |
1711 | * in allocations. | |
1712 | */ | |
0fb54e50 | 1713 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1714 | struct mount *dest_mnt, |
1715 | struct mountpoint *dest_mp, | |
1716 | struct path *parent_path) | |
b90fa9ae | 1717 | { |
38129a13 | 1718 | HLIST_HEAD(tree_list); |
315fc83e | 1719 | struct mount *child, *p; |
38129a13 | 1720 | struct hlist_node *n; |
719f5d7f | 1721 | int err; |
b90fa9ae | 1722 | |
fc7be130 | 1723 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1724 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1725 | if (err) |
1726 | goto out; | |
0b1b901b | 1727 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1728 | lock_mount_hash(); |
0b1b901b AV |
1729 | if (err) |
1730 | goto out_cleanup_ids; | |
909b0a88 | 1731 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1732 | set_mnt_shared(p); |
0b1b901b AV |
1733 | } else { |
1734 | lock_mount_hash(); | |
b90fa9ae | 1735 | } |
1a390689 | 1736 | if (parent_path) { |
0fb54e50 | 1737 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1738 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1739 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1740 | } else { |
84d17192 | 1741 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1d6a32ac | 1742 | commit_tree(source_mnt, NULL); |
21444403 | 1743 | } |
b90fa9ae | 1744 | |
38129a13 | 1745 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 1746 | struct mount *q; |
38129a13 | 1747 | hlist_del_init(&child->mnt_hash); |
1d6a32ac AV |
1748 | q = __lookup_mnt_last(&child->mnt_parent->mnt, |
1749 | child->mnt_mountpoint); | |
1750 | commit_tree(child, q); | |
b90fa9ae | 1751 | } |
719ea2fb | 1752 | unlock_mount_hash(); |
99b7db7b | 1753 | |
b90fa9ae | 1754 | return 0; |
719f5d7f MS |
1755 | |
1756 | out_cleanup_ids: | |
f2ebb3a9 AV |
1757 | while (!hlist_empty(&tree_list)) { |
1758 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1759 | umount_tree(child, 0); | |
1760 | } | |
1761 | unlock_mount_hash(); | |
0b1b901b | 1762 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1763 | out: |
1764 | return err; | |
b90fa9ae RP |
1765 | } |
1766 | ||
84d17192 | 1767 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1768 | { |
1769 | struct vfsmount *mnt; | |
84d17192 | 1770 | struct dentry *dentry = path->dentry; |
b12cea91 | 1771 | retry: |
84d17192 AV |
1772 | mutex_lock(&dentry->d_inode->i_mutex); |
1773 | if (unlikely(cant_mount(dentry))) { | |
1774 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1775 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1776 | } |
97216be0 | 1777 | namespace_lock(); |
b12cea91 | 1778 | mnt = lookup_mnt(path); |
84d17192 AV |
1779 | if (likely(!mnt)) { |
1780 | struct mountpoint *mp = new_mountpoint(dentry); | |
1781 | if (IS_ERR(mp)) { | |
97216be0 | 1782 | namespace_unlock(); |
84d17192 AV |
1783 | mutex_unlock(&dentry->d_inode->i_mutex); |
1784 | return mp; | |
1785 | } | |
1786 | return mp; | |
1787 | } | |
97216be0 | 1788 | namespace_unlock(); |
b12cea91 AV |
1789 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1790 | path_put(path); | |
1791 | path->mnt = mnt; | |
84d17192 | 1792 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1793 | goto retry; |
1794 | } | |
1795 | ||
84d17192 | 1796 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1797 | { |
84d17192 AV |
1798 | struct dentry *dentry = where->m_dentry; |
1799 | put_mountpoint(where); | |
328e6d90 | 1800 | namespace_unlock(); |
84d17192 | 1801 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1802 | } |
1803 | ||
84d17192 | 1804 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1805 | { |
95bc5f25 | 1806 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1807 | return -EINVAL; |
1808 | ||
84d17192 | 1809 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1810 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1811 | return -ENOTDIR; |
1812 | ||
84d17192 | 1813 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1814 | } |
1815 | ||
7a2e8a8f VA |
1816 | /* |
1817 | * Sanity check the flags to change_mnt_propagation. | |
1818 | */ | |
1819 | ||
1820 | static int flags_to_propagation_type(int flags) | |
1821 | { | |
7c6e984d | 1822 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1823 | |
1824 | /* Fail if any non-propagation flags are set */ | |
1825 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1826 | return 0; | |
1827 | /* Only one propagation flag should be set */ | |
1828 | if (!is_power_of_2(type)) | |
1829 | return 0; | |
1830 | return type; | |
1831 | } | |
1832 | ||
07b20889 RP |
1833 | /* |
1834 | * recursively change the type of the mountpoint. | |
1835 | */ | |
0a0d8a46 | 1836 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1837 | { |
315fc83e | 1838 | struct mount *m; |
4b8b21f4 | 1839 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1840 | int recurse = flag & MS_REC; |
7a2e8a8f | 1841 | int type; |
719f5d7f | 1842 | int err = 0; |
07b20889 | 1843 | |
2d92ab3c | 1844 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1845 | return -EINVAL; |
1846 | ||
7a2e8a8f VA |
1847 | type = flags_to_propagation_type(flag); |
1848 | if (!type) | |
1849 | return -EINVAL; | |
1850 | ||
97216be0 | 1851 | namespace_lock(); |
719f5d7f MS |
1852 | if (type == MS_SHARED) { |
1853 | err = invent_group_ids(mnt, recurse); | |
1854 | if (err) | |
1855 | goto out_unlock; | |
1856 | } | |
1857 | ||
719ea2fb | 1858 | lock_mount_hash(); |
909b0a88 | 1859 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1860 | change_mnt_propagation(m, type); |
719ea2fb | 1861 | unlock_mount_hash(); |
719f5d7f MS |
1862 | |
1863 | out_unlock: | |
97216be0 | 1864 | namespace_unlock(); |
719f5d7f | 1865 | return err; |
07b20889 RP |
1866 | } |
1867 | ||
5ff9d8a6 EB |
1868 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1869 | { | |
1870 | struct mount *child; | |
1871 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1872 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1873 | continue; | |
1874 | ||
1875 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1876 | return true; | |
1877 | } | |
1878 | return false; | |
1879 | } | |
1880 | ||
1da177e4 LT |
1881 | /* |
1882 | * do loopback mount. | |
1883 | */ | |
808d4e3c | 1884 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1885 | int recurse) |
1da177e4 | 1886 | { |
2d92ab3c | 1887 | struct path old_path; |
84d17192 AV |
1888 | struct mount *mnt = NULL, *old, *parent; |
1889 | struct mountpoint *mp; | |
57eccb83 | 1890 | int err; |
1da177e4 LT |
1891 | if (!old_name || !*old_name) |
1892 | return -EINVAL; | |
815d405c | 1893 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1894 | if (err) |
1895 | return err; | |
1896 | ||
8823c079 | 1897 | err = -EINVAL; |
4ce5d2b1 | 1898 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1899 | goto out; |
1900 | ||
84d17192 AV |
1901 | mp = lock_mount(path); |
1902 | err = PTR_ERR(mp); | |
1903 | if (IS_ERR(mp)) | |
b12cea91 AV |
1904 | goto out; |
1905 | ||
87129cc0 | 1906 | old = real_mount(old_path.mnt); |
84d17192 | 1907 | parent = real_mount(path->mnt); |
87129cc0 | 1908 | |
1da177e4 | 1909 | err = -EINVAL; |
fc7be130 | 1910 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1911 | goto out2; |
9676f0c6 | 1912 | |
84d17192 | 1913 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1914 | goto out2; |
1da177e4 | 1915 | |
5ff9d8a6 EB |
1916 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1917 | goto out2; | |
1918 | ||
ccd48bc7 | 1919 | if (recurse) |
4ce5d2b1 | 1920 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1921 | else |
87129cc0 | 1922 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1923 | |
be34d1a3 DH |
1924 | if (IS_ERR(mnt)) { |
1925 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1926 | goto out2; |
be34d1a3 | 1927 | } |
ccd48bc7 | 1928 | |
5ff9d8a6 EB |
1929 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1930 | ||
84d17192 | 1931 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1932 | if (err) { |
719ea2fb | 1933 | lock_mount_hash(); |
328e6d90 | 1934 | umount_tree(mnt, 0); |
719ea2fb | 1935 | unlock_mount_hash(); |
5b83d2c5 | 1936 | } |
b12cea91 | 1937 | out2: |
84d17192 | 1938 | unlock_mount(mp); |
ccd48bc7 | 1939 | out: |
2d92ab3c | 1940 | path_put(&old_path); |
1da177e4 LT |
1941 | return err; |
1942 | } | |
1943 | ||
2e4b7fcd DH |
1944 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1945 | { | |
1946 | int error = 0; | |
1947 | int readonly_request = 0; | |
1948 | ||
1949 | if (ms_flags & MS_RDONLY) | |
1950 | readonly_request = 1; | |
1951 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1952 | return 0; | |
1953 | ||
1954 | if (readonly_request) | |
83adc753 | 1955 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1956 | else |
83adc753 | 1957 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1958 | return error; |
1959 | } | |
1960 | ||
1da177e4 LT |
1961 | /* |
1962 | * change filesystem flags. dir should be a physical root of filesystem. | |
1963 | * If you've mounted a non-root directory somewhere and want to do remount | |
1964 | * on it - tough luck. | |
1965 | */ | |
0a0d8a46 | 1966 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1967 | void *data) |
1968 | { | |
1969 | int err; | |
2d92ab3c | 1970 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1971 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1972 | |
143c8c91 | 1973 | if (!check_mnt(mnt)) |
1da177e4 LT |
1974 | return -EINVAL; |
1975 | ||
2d92ab3c | 1976 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1977 | return -EINVAL; |
1978 | ||
07b64558 EB |
1979 | /* Don't allow changing of locked mnt flags. |
1980 | * | |
1981 | * No locks need to be held here while testing the various | |
1982 | * MNT_LOCK flags because those flags can never be cleared | |
1983 | * once they are set. | |
1984 | */ | |
1985 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
1986 | !(mnt_flags & MNT_READONLY)) { | |
1987 | return -EPERM; | |
1988 | } | |
9566d674 EB |
1989 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
1990 | !(mnt_flags & MNT_NODEV)) { | |
1991 | return -EPERM; | |
1992 | } | |
1993 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
1994 | !(mnt_flags & MNT_NOSUID)) { | |
1995 | return -EPERM; | |
1996 | } | |
1997 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
1998 | !(mnt_flags & MNT_NOEXEC)) { | |
1999 | return -EPERM; | |
2000 | } | |
2001 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2002 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2003 | return -EPERM; | |
2004 | } | |
2005 | ||
ff36fe2c EP |
2006 | err = security_sb_remount(sb, data); |
2007 | if (err) | |
2008 | return err; | |
2009 | ||
1da177e4 | 2010 | down_write(&sb->s_umount); |
2e4b7fcd | 2011 | if (flags & MS_BIND) |
2d92ab3c | 2012 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
2013 | else if (!capable(CAP_SYS_ADMIN)) |
2014 | err = -EPERM; | |
4aa98cf7 | 2015 | else |
2e4b7fcd | 2016 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 2017 | if (!err) { |
719ea2fb | 2018 | lock_mount_hash(); |
a6138db8 | 2019 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2020 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2021 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2022 | unlock_mount_hash(); |
0e55a7cc | 2023 | } |
6339dab8 | 2024 | up_write(&sb->s_umount); |
1da177e4 LT |
2025 | return err; |
2026 | } | |
2027 | ||
cbbe362c | 2028 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2029 | { |
315fc83e | 2030 | struct mount *p; |
909b0a88 | 2031 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2032 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2033 | return 1; |
2034 | } | |
2035 | return 0; | |
2036 | } | |
2037 | ||
808d4e3c | 2038 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2039 | { |
2d92ab3c | 2040 | struct path old_path, parent_path; |
676da58d | 2041 | struct mount *p; |
0fb54e50 | 2042 | struct mount *old; |
84d17192 | 2043 | struct mountpoint *mp; |
57eccb83 | 2044 | int err; |
1da177e4 LT |
2045 | if (!old_name || !*old_name) |
2046 | return -EINVAL; | |
2d92ab3c | 2047 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2048 | if (err) |
2049 | return err; | |
2050 | ||
84d17192 AV |
2051 | mp = lock_mount(path); |
2052 | err = PTR_ERR(mp); | |
2053 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2054 | goto out; |
2055 | ||
143c8c91 | 2056 | old = real_mount(old_path.mnt); |
fc7be130 | 2057 | p = real_mount(path->mnt); |
143c8c91 | 2058 | |
1da177e4 | 2059 | err = -EINVAL; |
fc7be130 | 2060 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2061 | goto out1; |
2062 | ||
5ff9d8a6 EB |
2063 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2064 | goto out1; | |
2065 | ||
1da177e4 | 2066 | err = -EINVAL; |
2d92ab3c | 2067 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2068 | goto out1; |
1da177e4 | 2069 | |
676da58d | 2070 | if (!mnt_has_parent(old)) |
21444403 | 2071 | goto out1; |
1da177e4 | 2072 | |
2d92ab3c AV |
2073 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
2074 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
2075 | goto out1; |
2076 | /* | |
2077 | * Don't move a mount residing in a shared parent. | |
2078 | */ | |
fc7be130 | 2079 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2080 | goto out1; |
9676f0c6 RP |
2081 | /* |
2082 | * Don't move a mount tree containing unbindable mounts to a destination | |
2083 | * mount which is shared. | |
2084 | */ | |
fc7be130 | 2085 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2086 | goto out1; |
1da177e4 | 2087 | err = -ELOOP; |
fc7be130 | 2088 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2089 | if (p == old) |
21444403 | 2090 | goto out1; |
1da177e4 | 2091 | |
84d17192 | 2092 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2093 | if (err) |
21444403 | 2094 | goto out1; |
1da177e4 LT |
2095 | |
2096 | /* if the mount is moved, it should no longer be expire | |
2097 | * automatically */ | |
6776db3d | 2098 | list_del_init(&old->mnt_expire); |
1da177e4 | 2099 | out1: |
84d17192 | 2100 | unlock_mount(mp); |
1da177e4 | 2101 | out: |
1da177e4 | 2102 | if (!err) |
1a390689 | 2103 | path_put(&parent_path); |
2d92ab3c | 2104 | path_put(&old_path); |
1da177e4 LT |
2105 | return err; |
2106 | } | |
2107 | ||
9d412a43 AV |
2108 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2109 | { | |
2110 | int err; | |
2111 | const char *subtype = strchr(fstype, '.'); | |
2112 | if (subtype) { | |
2113 | subtype++; | |
2114 | err = -EINVAL; | |
2115 | if (!subtype[0]) | |
2116 | goto err; | |
2117 | } else | |
2118 | subtype = ""; | |
2119 | ||
2120 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2121 | err = -ENOMEM; | |
2122 | if (!mnt->mnt_sb->s_subtype) | |
2123 | goto err; | |
2124 | return mnt; | |
2125 | ||
2126 | err: | |
2127 | mntput(mnt); | |
2128 | return ERR_PTR(err); | |
2129 | } | |
2130 | ||
9d412a43 AV |
2131 | /* |
2132 | * add a mount into a namespace's mount tree | |
2133 | */ | |
95bc5f25 | 2134 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2135 | { |
84d17192 AV |
2136 | struct mountpoint *mp; |
2137 | struct mount *parent; | |
9d412a43 AV |
2138 | int err; |
2139 | ||
f2ebb3a9 | 2140 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2141 | |
84d17192 AV |
2142 | mp = lock_mount(path); |
2143 | if (IS_ERR(mp)) | |
2144 | return PTR_ERR(mp); | |
9d412a43 | 2145 | |
84d17192 | 2146 | parent = real_mount(path->mnt); |
9d412a43 | 2147 | err = -EINVAL; |
84d17192 | 2148 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2149 | /* that's acceptable only for automounts done in private ns */ |
2150 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2151 | goto unlock; | |
2152 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2153 | if (!parent->mnt_ns) |
156cacb1 AV |
2154 | goto unlock; |
2155 | } | |
9d412a43 AV |
2156 | |
2157 | /* Refuse the same filesystem on the same mount point */ | |
2158 | err = -EBUSY; | |
95bc5f25 | 2159 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2160 | path->mnt->mnt_root == path->dentry) |
2161 | goto unlock; | |
2162 | ||
2163 | err = -EINVAL; | |
95bc5f25 | 2164 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2165 | goto unlock; |
2166 | ||
95bc5f25 | 2167 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2168 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2169 | |
2170 | unlock: | |
84d17192 | 2171 | unlock_mount(mp); |
9d412a43 AV |
2172 | return err; |
2173 | } | |
b1e75df4 | 2174 | |
1da177e4 LT |
2175 | /* |
2176 | * create a new mount for userspace and request it to be added into the | |
2177 | * namespace's tree | |
2178 | */ | |
0c55cfc4 | 2179 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2180 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2181 | { |
0c55cfc4 | 2182 | struct file_system_type *type; |
9b40bc90 | 2183 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2184 | struct vfsmount *mnt; |
15f9a3f3 | 2185 | int err; |
1da177e4 | 2186 | |
0c55cfc4 | 2187 | if (!fstype) |
1da177e4 LT |
2188 | return -EINVAL; |
2189 | ||
0c55cfc4 EB |
2190 | type = get_fs_type(fstype); |
2191 | if (!type) | |
2192 | return -ENODEV; | |
2193 | ||
2194 | if (user_ns != &init_user_ns) { | |
2195 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2196 | put_filesystem(type); | |
2197 | return -EPERM; | |
2198 | } | |
2199 | /* Only in special cases allow devices from mounts | |
2200 | * created outside the initial user namespace. | |
2201 | */ | |
2202 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2203 | flags |= MS_NODEV; | |
9566d674 | 2204 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; |
0c55cfc4 EB |
2205 | } |
2206 | } | |
2207 | ||
2208 | mnt = vfs_kern_mount(type, flags, name, data); | |
2209 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2210 | !mnt->mnt_sb->s_subtype) | |
2211 | mnt = fs_set_subtype(mnt, fstype); | |
2212 | ||
2213 | put_filesystem(type); | |
1da177e4 LT |
2214 | if (IS_ERR(mnt)) |
2215 | return PTR_ERR(mnt); | |
2216 | ||
95bc5f25 | 2217 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2218 | if (err) |
2219 | mntput(mnt); | |
2220 | return err; | |
1da177e4 LT |
2221 | } |
2222 | ||
19a167af AV |
2223 | int finish_automount(struct vfsmount *m, struct path *path) |
2224 | { | |
6776db3d | 2225 | struct mount *mnt = real_mount(m); |
19a167af AV |
2226 | int err; |
2227 | /* The new mount record should have at least 2 refs to prevent it being | |
2228 | * expired before we get a chance to add it | |
2229 | */ | |
6776db3d | 2230 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2231 | |
2232 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2233 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2234 | err = -ELOOP; |
2235 | goto fail; | |
19a167af AV |
2236 | } |
2237 | ||
95bc5f25 | 2238 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2239 | if (!err) |
2240 | return 0; | |
2241 | fail: | |
2242 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2243 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2244 | namespace_lock(); |
6776db3d | 2245 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2246 | namespace_unlock(); |
19a167af | 2247 | } |
b1e75df4 AV |
2248 | mntput(m); |
2249 | mntput(m); | |
19a167af AV |
2250 | return err; |
2251 | } | |
2252 | ||
ea5b778a DH |
2253 | /** |
2254 | * mnt_set_expiry - Put a mount on an expiration list | |
2255 | * @mnt: The mount to list. | |
2256 | * @expiry_list: The list to add the mount to. | |
2257 | */ | |
2258 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2259 | { | |
97216be0 | 2260 | namespace_lock(); |
ea5b778a | 2261 | |
6776db3d | 2262 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2263 | |
97216be0 | 2264 | namespace_unlock(); |
ea5b778a DH |
2265 | } |
2266 | EXPORT_SYMBOL(mnt_set_expiry); | |
2267 | ||
1da177e4 LT |
2268 | /* |
2269 | * process a list of expirable mountpoints with the intent of discarding any | |
2270 | * mountpoints that aren't in use and haven't been touched since last we came | |
2271 | * here | |
2272 | */ | |
2273 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2274 | { | |
761d5c38 | 2275 | struct mount *mnt, *next; |
1da177e4 LT |
2276 | LIST_HEAD(graveyard); |
2277 | ||
2278 | if (list_empty(mounts)) | |
2279 | return; | |
2280 | ||
97216be0 | 2281 | namespace_lock(); |
719ea2fb | 2282 | lock_mount_hash(); |
1da177e4 LT |
2283 | |
2284 | /* extract from the expiration list every vfsmount that matches the | |
2285 | * following criteria: | |
2286 | * - only referenced by its parent vfsmount | |
2287 | * - still marked for expiry (marked on the last call here; marks are | |
2288 | * cleared by mntput()) | |
2289 | */ | |
6776db3d | 2290 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2291 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2292 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2293 | continue; |
6776db3d | 2294 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2295 | } |
bcc5c7d2 | 2296 | while (!list_empty(&graveyard)) { |
6776db3d | 2297 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2298 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2299 | umount_tree(mnt, 1); |
bcc5c7d2 | 2300 | } |
719ea2fb | 2301 | unlock_mount_hash(); |
3ab6abee | 2302 | namespace_unlock(); |
5528f911 TM |
2303 | } |
2304 | ||
2305 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2306 | ||
2307 | /* | |
2308 | * Ripoff of 'select_parent()' | |
2309 | * | |
2310 | * search the list of submounts for a given mountpoint, and move any | |
2311 | * shrinkable submounts to the 'graveyard' list. | |
2312 | */ | |
692afc31 | 2313 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2314 | { |
692afc31 | 2315 | struct mount *this_parent = parent; |
5528f911 TM |
2316 | struct list_head *next; |
2317 | int found = 0; | |
2318 | ||
2319 | repeat: | |
6b41d536 | 2320 | next = this_parent->mnt_mounts.next; |
5528f911 | 2321 | resume: |
6b41d536 | 2322 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2323 | struct list_head *tmp = next; |
6b41d536 | 2324 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2325 | |
2326 | next = tmp->next; | |
692afc31 | 2327 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2328 | continue; |
5528f911 TM |
2329 | /* |
2330 | * Descend a level if the d_mounts list is non-empty. | |
2331 | */ | |
6b41d536 | 2332 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2333 | this_parent = mnt; |
2334 | goto repeat; | |
2335 | } | |
1da177e4 | 2336 | |
1ab59738 | 2337 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2338 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2339 | found++; |
2340 | } | |
1da177e4 | 2341 | } |
5528f911 TM |
2342 | /* |
2343 | * All done at this level ... ascend and resume the search | |
2344 | */ | |
2345 | if (this_parent != parent) { | |
6b41d536 | 2346 | next = this_parent->mnt_child.next; |
0714a533 | 2347 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2348 | goto resume; |
2349 | } | |
2350 | return found; | |
2351 | } | |
2352 | ||
2353 | /* | |
2354 | * process a list of expirable mountpoints with the intent of discarding any | |
2355 | * submounts of a specific parent mountpoint | |
99b7db7b | 2356 | * |
48a066e7 | 2357 | * mount_lock must be held for write |
5528f911 | 2358 | */ |
b54b9be7 | 2359 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2360 | { |
2361 | LIST_HEAD(graveyard); | |
761d5c38 | 2362 | struct mount *m; |
5528f911 | 2363 | |
5528f911 | 2364 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2365 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2366 | while (!list_empty(&graveyard)) { |
761d5c38 | 2367 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2368 | mnt_expire); |
143c8c91 | 2369 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2370 | umount_tree(m, 1); |
bcc5c7d2 AV |
2371 | } |
2372 | } | |
1da177e4 LT |
2373 | } |
2374 | ||
1da177e4 LT |
2375 | /* |
2376 | * Some copy_from_user() implementations do not return the exact number of | |
2377 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2378 | * Note that this function differs from copy_from_user() in that it will oops | |
2379 | * on bad values of `to', rather than returning a short copy. | |
2380 | */ | |
b58fed8b RP |
2381 | static long exact_copy_from_user(void *to, const void __user * from, |
2382 | unsigned long n) | |
1da177e4 LT |
2383 | { |
2384 | char *t = to; | |
2385 | const char __user *f = from; | |
2386 | char c; | |
2387 | ||
2388 | if (!access_ok(VERIFY_READ, from, n)) | |
2389 | return n; | |
2390 | ||
2391 | while (n) { | |
2392 | if (__get_user(c, f)) { | |
2393 | memset(t, 0, n); | |
2394 | break; | |
2395 | } | |
2396 | *t++ = c; | |
2397 | f++; | |
2398 | n--; | |
2399 | } | |
2400 | return n; | |
2401 | } | |
2402 | ||
b58fed8b | 2403 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2404 | { |
2405 | int i; | |
2406 | unsigned long page; | |
2407 | unsigned long size; | |
b58fed8b | 2408 | |
1da177e4 LT |
2409 | *where = 0; |
2410 | if (!data) | |
2411 | return 0; | |
2412 | ||
2413 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2414 | return -ENOMEM; | |
2415 | ||
2416 | /* We only care that *some* data at the address the user | |
2417 | * gave us is valid. Just in case, we'll zero | |
2418 | * the remainder of the page. | |
2419 | */ | |
2420 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2421 | size = TASK_SIZE - (unsigned long)data; | |
2422 | if (size > PAGE_SIZE) | |
2423 | size = PAGE_SIZE; | |
2424 | ||
2425 | i = size - exact_copy_from_user((void *)page, data, size); | |
2426 | if (!i) { | |
b58fed8b | 2427 | free_page(page); |
1da177e4 LT |
2428 | return -EFAULT; |
2429 | } | |
2430 | if (i != PAGE_SIZE) | |
2431 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2432 | *where = page; | |
2433 | return 0; | |
2434 | } | |
2435 | ||
eca6f534 VN |
2436 | int copy_mount_string(const void __user *data, char **where) |
2437 | { | |
2438 | char *tmp; | |
2439 | ||
2440 | if (!data) { | |
2441 | *where = NULL; | |
2442 | return 0; | |
2443 | } | |
2444 | ||
2445 | tmp = strndup_user(data, PAGE_SIZE); | |
2446 | if (IS_ERR(tmp)) | |
2447 | return PTR_ERR(tmp); | |
2448 | ||
2449 | *where = tmp; | |
2450 | return 0; | |
2451 | } | |
2452 | ||
1da177e4 LT |
2453 | /* |
2454 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2455 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2456 | * | |
2457 | * data is a (void *) that can point to any structure up to | |
2458 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2459 | * information (or be NULL). | |
2460 | * | |
2461 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2462 | * When the flags word was introduced its top half was required | |
2463 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2464 | * Therefore, if this magic number is present, it carries no information | |
2465 | * and must be discarded. | |
2466 | */ | |
808d4e3c AV |
2467 | long do_mount(const char *dev_name, const char *dir_name, |
2468 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2469 | { |
2d92ab3c | 2470 | struct path path; |
1da177e4 LT |
2471 | int retval = 0; |
2472 | int mnt_flags = 0; | |
2473 | ||
2474 | /* Discard magic */ | |
2475 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2476 | flags &= ~MS_MGC_MSK; | |
2477 | ||
2478 | /* Basic sanity checks */ | |
2479 | ||
2480 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2481 | return -EINVAL; | |
1da177e4 LT |
2482 | |
2483 | if (data_page) | |
2484 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2485 | ||
a27ab9f2 TH |
2486 | /* ... and get the mountpoint */ |
2487 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2488 | if (retval) | |
2489 | return retval; | |
2490 | ||
2491 | retval = security_sb_mount(dev_name, &path, | |
2492 | type_page, flags, data_page); | |
0d5cadb8 AV |
2493 | if (!retval && !may_mount()) |
2494 | retval = -EPERM; | |
a27ab9f2 TH |
2495 | if (retval) |
2496 | goto dput_out; | |
2497 | ||
613cbe3d AK |
2498 | /* Default to relatime unless overriden */ |
2499 | if (!(flags & MS_NOATIME)) | |
2500 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2501 | |
1da177e4 LT |
2502 | /* Separate the per-mountpoint flags */ |
2503 | if (flags & MS_NOSUID) | |
2504 | mnt_flags |= MNT_NOSUID; | |
2505 | if (flags & MS_NODEV) | |
2506 | mnt_flags |= MNT_NODEV; | |
2507 | if (flags & MS_NOEXEC) | |
2508 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2509 | if (flags & MS_NOATIME) |
2510 | mnt_flags |= MNT_NOATIME; | |
2511 | if (flags & MS_NODIRATIME) | |
2512 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2513 | if (flags & MS_STRICTATIME) |
2514 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2515 | if (flags & MS_RDONLY) |
2516 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2517 | |
ffbc6f0e EB |
2518 | /* The default atime for remount is preservation */ |
2519 | if ((flags & MS_REMOUNT) && | |
2520 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2521 | MS_STRICTATIME)) == 0)) { | |
2522 | mnt_flags &= ~MNT_ATIME_MASK; | |
2523 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2524 | } | |
2525 | ||
7a4dec53 | 2526 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2527 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2528 | MS_STRICTATIME); | |
1da177e4 | 2529 | |
1da177e4 | 2530 | if (flags & MS_REMOUNT) |
2d92ab3c | 2531 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2532 | data_page); |
2533 | else if (flags & MS_BIND) | |
2d92ab3c | 2534 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2535 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2536 | retval = do_change_type(&path, flags); |
1da177e4 | 2537 | else if (flags & MS_MOVE) |
2d92ab3c | 2538 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2539 | else |
2d92ab3c | 2540 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2541 | dev_name, data_page); |
2542 | dput_out: | |
2d92ab3c | 2543 | path_put(&path); |
1da177e4 LT |
2544 | return retval; |
2545 | } | |
2546 | ||
771b1371 EB |
2547 | static void free_mnt_ns(struct mnt_namespace *ns) |
2548 | { | |
98f842e6 | 2549 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2550 | put_user_ns(ns->user_ns); |
2551 | kfree(ns); | |
2552 | } | |
2553 | ||
8823c079 EB |
2554 | /* |
2555 | * Assign a sequence number so we can detect when we attempt to bind | |
2556 | * mount a reference to an older mount namespace into the current | |
2557 | * mount namespace, preventing reference counting loops. A 64bit | |
2558 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2559 | * is effectively never, so we can ignore the possibility. | |
2560 | */ | |
2561 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2562 | ||
771b1371 | 2563 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2564 | { |
2565 | struct mnt_namespace *new_ns; | |
98f842e6 | 2566 | int ret; |
cf8d2c11 TM |
2567 | |
2568 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2569 | if (!new_ns) | |
2570 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2571 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2572 | if (ret) { | |
2573 | kfree(new_ns); | |
2574 | return ERR_PTR(ret); | |
2575 | } | |
8823c079 | 2576 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2577 | atomic_set(&new_ns->count, 1); |
2578 | new_ns->root = NULL; | |
2579 | INIT_LIST_HEAD(&new_ns->list); | |
2580 | init_waitqueue_head(&new_ns->poll); | |
2581 | new_ns->event = 0; | |
771b1371 | 2582 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2583 | return new_ns; |
2584 | } | |
2585 | ||
9559f689 AV |
2586 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2587 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2588 | { |
6b3286ed | 2589 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2590 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2591 | struct mount *p, *q; |
9559f689 | 2592 | struct mount *old; |
cb338d06 | 2593 | struct mount *new; |
7a472ef4 | 2594 | int copy_flags; |
1da177e4 | 2595 | |
9559f689 AV |
2596 | BUG_ON(!ns); |
2597 | ||
2598 | if (likely(!(flags & CLONE_NEWNS))) { | |
2599 | get_mnt_ns(ns); | |
2600 | return ns; | |
2601 | } | |
2602 | ||
2603 | old = ns->root; | |
2604 | ||
771b1371 | 2605 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2606 | if (IS_ERR(new_ns)) |
2607 | return new_ns; | |
1da177e4 | 2608 | |
97216be0 | 2609 | namespace_lock(); |
1da177e4 | 2610 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2611 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2612 | if (user_ns != ns->user_ns) |
132c94e3 | 2613 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2614 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2615 | if (IS_ERR(new)) { |
328e6d90 | 2616 | namespace_unlock(); |
771b1371 | 2617 | free_mnt_ns(new_ns); |
be34d1a3 | 2618 | return ERR_CAST(new); |
1da177e4 | 2619 | } |
be08d6d2 | 2620 | new_ns->root = new; |
1a4eeaf2 | 2621 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2622 | |
2623 | /* | |
2624 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2625 | * as belonging to new namespace. We have already acquired a private | |
2626 | * fs_struct, so tsk->fs->lock is not needed. | |
2627 | */ | |
909b0a88 | 2628 | p = old; |
cb338d06 | 2629 | q = new; |
1da177e4 | 2630 | while (p) { |
143c8c91 | 2631 | q->mnt_ns = new_ns; |
9559f689 AV |
2632 | if (new_fs) { |
2633 | if (&p->mnt == new_fs->root.mnt) { | |
2634 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2635 | rootmnt = &p->mnt; |
1da177e4 | 2636 | } |
9559f689 AV |
2637 | if (&p->mnt == new_fs->pwd.mnt) { |
2638 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2639 | pwdmnt = &p->mnt; |
1da177e4 | 2640 | } |
1da177e4 | 2641 | } |
909b0a88 AV |
2642 | p = next_mnt(p, old); |
2643 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2644 | if (!q) |
2645 | break; | |
2646 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2647 | p = next_mnt(p, old); | |
1da177e4 | 2648 | } |
328e6d90 | 2649 | namespace_unlock(); |
1da177e4 | 2650 | |
1da177e4 | 2651 | if (rootmnt) |
f03c6599 | 2652 | mntput(rootmnt); |
1da177e4 | 2653 | if (pwdmnt) |
f03c6599 | 2654 | mntput(pwdmnt); |
1da177e4 | 2655 | |
741a2951 | 2656 | return new_ns; |
1da177e4 LT |
2657 | } |
2658 | ||
cf8d2c11 TM |
2659 | /** |
2660 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2661 | * @mnt: pointer to the new root filesystem mountpoint | |
2662 | */ | |
1a4eeaf2 | 2663 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2664 | { |
771b1371 | 2665 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2666 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2667 | struct mount *mnt = real_mount(m); |
2668 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2669 | new_ns->root = mnt; |
b1983cd8 | 2670 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2671 | } else { |
1a4eeaf2 | 2672 | mntput(m); |
cf8d2c11 TM |
2673 | } |
2674 | return new_ns; | |
2675 | } | |
cf8d2c11 | 2676 | |
ea441d11 AV |
2677 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2678 | { | |
2679 | struct mnt_namespace *ns; | |
d31da0f0 | 2680 | struct super_block *s; |
ea441d11 AV |
2681 | struct path path; |
2682 | int err; | |
2683 | ||
2684 | ns = create_mnt_ns(mnt); | |
2685 | if (IS_ERR(ns)) | |
2686 | return ERR_CAST(ns); | |
2687 | ||
2688 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2689 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2690 | ||
2691 | put_mnt_ns(ns); | |
2692 | ||
2693 | if (err) | |
2694 | return ERR_PTR(err); | |
2695 | ||
2696 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2697 | s = path.mnt->mnt_sb; |
2698 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2699 | mntput(path.mnt); |
2700 | /* lock the sucker */ | |
d31da0f0 | 2701 | down_write(&s->s_umount); |
ea441d11 AV |
2702 | /* ... and return the root of (sub)tree on it */ |
2703 | return path.dentry; | |
2704 | } | |
2705 | EXPORT_SYMBOL(mount_subtree); | |
2706 | ||
bdc480e3 HC |
2707 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2708 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2709 | { |
eca6f534 VN |
2710 | int ret; |
2711 | char *kernel_type; | |
91a27b2a | 2712 | struct filename *kernel_dir; |
eca6f534 | 2713 | char *kernel_dev; |
1da177e4 | 2714 | unsigned long data_page; |
1da177e4 | 2715 | |
eca6f534 VN |
2716 | ret = copy_mount_string(type, &kernel_type); |
2717 | if (ret < 0) | |
2718 | goto out_type; | |
1da177e4 | 2719 | |
eca6f534 VN |
2720 | kernel_dir = getname(dir_name); |
2721 | if (IS_ERR(kernel_dir)) { | |
2722 | ret = PTR_ERR(kernel_dir); | |
2723 | goto out_dir; | |
2724 | } | |
1da177e4 | 2725 | |
eca6f534 VN |
2726 | ret = copy_mount_string(dev_name, &kernel_dev); |
2727 | if (ret < 0) | |
2728 | goto out_dev; | |
1da177e4 | 2729 | |
eca6f534 VN |
2730 | ret = copy_mount_options(data, &data_page); |
2731 | if (ret < 0) | |
2732 | goto out_data; | |
1da177e4 | 2733 | |
91a27b2a | 2734 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2735 | (void *) data_page); |
1da177e4 | 2736 | |
eca6f534 VN |
2737 | free_page(data_page); |
2738 | out_data: | |
2739 | kfree(kernel_dev); | |
2740 | out_dev: | |
2741 | putname(kernel_dir); | |
2742 | out_dir: | |
2743 | kfree(kernel_type); | |
2744 | out_type: | |
2745 | return ret; | |
1da177e4 LT |
2746 | } |
2747 | ||
afac7cba AV |
2748 | /* |
2749 | * Return true if path is reachable from root | |
2750 | * | |
48a066e7 | 2751 | * namespace_sem or mount_lock is held |
afac7cba | 2752 | */ |
643822b4 | 2753 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2754 | const struct path *root) |
2755 | { | |
643822b4 | 2756 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2757 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2758 | mnt = mnt->mnt_parent; |
afac7cba | 2759 | } |
643822b4 | 2760 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2761 | } |
2762 | ||
2763 | int path_is_under(struct path *path1, struct path *path2) | |
2764 | { | |
2765 | int res; | |
48a066e7 | 2766 | read_seqlock_excl(&mount_lock); |
643822b4 | 2767 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 2768 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
2769 | return res; |
2770 | } | |
2771 | EXPORT_SYMBOL(path_is_under); | |
2772 | ||
1da177e4 LT |
2773 | /* |
2774 | * pivot_root Semantics: | |
2775 | * Moves the root file system of the current process to the directory put_old, | |
2776 | * makes new_root as the new root file system of the current process, and sets | |
2777 | * root/cwd of all processes which had them on the current root to new_root. | |
2778 | * | |
2779 | * Restrictions: | |
2780 | * The new_root and put_old must be directories, and must not be on the | |
2781 | * same file system as the current process root. The put_old must be | |
2782 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2783 | * pointed to by put_old must yield the same directory as new_root. No other | |
2784 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2785 | * | |
4a0d11fa NB |
2786 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2787 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2788 | * in this situation. | |
2789 | * | |
1da177e4 LT |
2790 | * Notes: |
2791 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2792 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2793 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2794 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2795 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2796 | * first. | |
2797 | */ | |
3480b257 HC |
2798 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2799 | const char __user *, put_old) | |
1da177e4 | 2800 | { |
2d8f3038 | 2801 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2802 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2803 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2804 | int error; |
2805 | ||
9b40bc90 | 2806 | if (!may_mount()) |
1da177e4 LT |
2807 | return -EPERM; |
2808 | ||
2d8f3038 | 2809 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2810 | if (error) |
2811 | goto out0; | |
1da177e4 | 2812 | |
2d8f3038 | 2813 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2814 | if (error) |
2815 | goto out1; | |
2816 | ||
2d8f3038 | 2817 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2818 | if (error) |
2819 | goto out2; | |
1da177e4 | 2820 | |
f7ad3c6b | 2821 | get_fs_root(current->fs, &root); |
84d17192 AV |
2822 | old_mp = lock_mount(&old); |
2823 | error = PTR_ERR(old_mp); | |
2824 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2825 | goto out3; |
2826 | ||
1da177e4 | 2827 | error = -EINVAL; |
419148da AV |
2828 | new_mnt = real_mount(new.mnt); |
2829 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2830 | old_mnt = real_mount(old.mnt); |
2831 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2832 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2833 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2834 | goto out4; |
143c8c91 | 2835 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2836 | goto out4; |
5ff9d8a6 EB |
2837 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2838 | goto out4; | |
1da177e4 | 2839 | error = -ENOENT; |
f3da392e | 2840 | if (d_unlinked(new.dentry)) |
b12cea91 | 2841 | goto out4; |
1da177e4 | 2842 | error = -EBUSY; |
84d17192 | 2843 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2844 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2845 | error = -EINVAL; |
8c3ee42e | 2846 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2847 | goto out4; /* not a mountpoint */ |
676da58d | 2848 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2849 | goto out4; /* not attached */ |
84d17192 | 2850 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2851 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2852 | goto out4; /* not a mountpoint */ |
676da58d | 2853 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2854 | goto out4; /* not attached */ |
4ac91378 | 2855 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2856 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2857 | goto out4; |
84d17192 | 2858 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 2859 | lock_mount_hash(); |
419148da AV |
2860 | detach_mnt(new_mnt, &parent_path); |
2861 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2862 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2863 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2864 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2865 | } | |
4ac91378 | 2866 | /* mount old root on put_old */ |
84d17192 | 2867 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2868 | /* mount new_root on / */ |
84d17192 | 2869 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2870 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
719ea2fb | 2871 | unlock_mount_hash(); |
2d8f3038 | 2872 | chroot_fs_refs(&root, &new); |
84d17192 | 2873 | put_mountpoint(root_mp); |
1da177e4 | 2874 | error = 0; |
b12cea91 | 2875 | out4: |
84d17192 | 2876 | unlock_mount(old_mp); |
b12cea91 AV |
2877 | if (!error) { |
2878 | path_put(&root_parent); | |
2879 | path_put(&parent_path); | |
2880 | } | |
2881 | out3: | |
8c3ee42e | 2882 | path_put(&root); |
b12cea91 | 2883 | out2: |
2d8f3038 | 2884 | path_put(&old); |
1da177e4 | 2885 | out1: |
2d8f3038 | 2886 | path_put(&new); |
1da177e4 | 2887 | out0: |
1da177e4 | 2888 | return error; |
1da177e4 LT |
2889 | } |
2890 | ||
2891 | static void __init init_mount_tree(void) | |
2892 | { | |
2893 | struct vfsmount *mnt; | |
6b3286ed | 2894 | struct mnt_namespace *ns; |
ac748a09 | 2895 | struct path root; |
0c55cfc4 | 2896 | struct file_system_type *type; |
1da177e4 | 2897 | |
0c55cfc4 EB |
2898 | type = get_fs_type("rootfs"); |
2899 | if (!type) | |
2900 | panic("Can't find rootfs type"); | |
2901 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2902 | put_filesystem(type); | |
1da177e4 LT |
2903 | if (IS_ERR(mnt)) |
2904 | panic("Can't create rootfs"); | |
b3e19d92 | 2905 | |
3b22edc5 TM |
2906 | ns = create_mnt_ns(mnt); |
2907 | if (IS_ERR(ns)) | |
1da177e4 | 2908 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2909 | |
2910 | init_task.nsproxy->mnt_ns = ns; | |
2911 | get_mnt_ns(ns); | |
2912 | ||
be08d6d2 AV |
2913 | root.mnt = mnt; |
2914 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2915 | |
2916 | set_fs_pwd(current->fs, &root); | |
2917 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2918 | } |
2919 | ||
74bf17cf | 2920 | void __init mnt_init(void) |
1da177e4 | 2921 | { |
13f14b4d | 2922 | unsigned u; |
15a67dd8 | 2923 | int err; |
1da177e4 | 2924 | |
7d6fec45 | 2925 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2926 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2927 | |
0818bf27 | 2928 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 2929 | sizeof(struct hlist_head), |
0818bf27 AV |
2930 | mhash_entries, 19, |
2931 | 0, | |
2932 | &m_hash_shift, &m_hash_mask, 0, 0); | |
2933 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
2934 | sizeof(struct hlist_head), | |
2935 | mphash_entries, 19, | |
2936 | 0, | |
2937 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
1da177e4 | 2938 | |
84d17192 | 2939 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2940 | panic("Failed to allocate mount hash table\n"); |
2941 | ||
0818bf27 | 2942 | for (u = 0; u <= m_hash_mask; u++) |
38129a13 | 2943 | INIT_HLIST_HEAD(&mount_hashtable[u]); |
0818bf27 AV |
2944 | for (u = 0; u <= mp_hash_mask; u++) |
2945 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2946 | |
4b93dc9b TH |
2947 | kernfs_init(); |
2948 | ||
15a67dd8 RD |
2949 | err = sysfs_init(); |
2950 | if (err) | |
2951 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2952 | __func__, err); |
00d26666 GKH |
2953 | fs_kobj = kobject_create_and_add("fs", NULL); |
2954 | if (!fs_kobj) | |
8e24eea7 | 2955 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2956 | init_rootfs(); |
2957 | init_mount_tree(); | |
2958 | } | |
2959 | ||
616511d0 | 2960 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2961 | { |
d498b25a | 2962 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2963 | return; |
7b00ed6f | 2964 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 2965 | free_mnt_ns(ns); |
1da177e4 | 2966 | } |
9d412a43 AV |
2967 | |
2968 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2969 | { | |
423e0ab0 TC |
2970 | struct vfsmount *mnt; |
2971 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2972 | if (!IS_ERR(mnt)) { | |
2973 | /* | |
2974 | * it is a longterm mount, don't release mnt until | |
2975 | * we unmount before file sys is unregistered | |
2976 | */ | |
f7a99c5b | 2977 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2978 | } |
2979 | return mnt; | |
9d412a43 AV |
2980 | } |
2981 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2982 | |
2983 | void kern_unmount(struct vfsmount *mnt) | |
2984 | { | |
2985 | /* release long term mount so mount point can be released */ | |
2986 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 2987 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 2988 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
2989 | mntput(mnt); |
2990 | } | |
2991 | } | |
2992 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2993 | |
2994 | bool our_mnt(struct vfsmount *mnt) | |
2995 | { | |
143c8c91 | 2996 | return check_mnt(real_mount(mnt)); |
02125a82 | 2997 | } |
8823c079 | 2998 | |
3151527e EB |
2999 | bool current_chrooted(void) |
3000 | { | |
3001 | /* Does the current process have a non-standard root */ | |
3002 | struct path ns_root; | |
3003 | struct path fs_root; | |
3004 | bool chrooted; | |
3005 | ||
3006 | /* Find the namespace root */ | |
3007 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3008 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3009 | path_get(&ns_root); | |
3010 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3011 | ; | |
3012 | ||
3013 | get_fs_root(current->fs, &fs_root); | |
3014 | ||
3015 | chrooted = !path_equal(&fs_root, &ns_root); | |
3016 | ||
3017 | path_put(&fs_root); | |
3018 | path_put(&ns_root); | |
3019 | ||
3020 | return chrooted; | |
3021 | } | |
3022 | ||
e51db735 | 3023 | bool fs_fully_visible(struct file_system_type *type) |
87a8ebd6 EB |
3024 | { |
3025 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
3026 | struct mount *mnt; | |
e51db735 | 3027 | bool visible = false; |
87a8ebd6 | 3028 | |
e51db735 EB |
3029 | if (unlikely(!ns)) |
3030 | return false; | |
3031 | ||
44bb4385 | 3032 | down_read(&namespace_sem); |
87a8ebd6 | 3033 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 EB |
3034 | struct mount *child; |
3035 | if (mnt->mnt.mnt_sb->s_type != type) | |
3036 | continue; | |
3037 | ||
3038 | /* This mount is not fully visible if there are any child mounts | |
3039 | * that cover anything except for empty directories. | |
3040 | */ | |
3041 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3042 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
3043 | if (!S_ISDIR(inode->i_mode)) | |
3044 | goto next; | |
41301ae7 | 3045 | if (inode->i_nlink > 2) |
e51db735 | 3046 | goto next; |
87a8ebd6 | 3047 | } |
e51db735 EB |
3048 | visible = true; |
3049 | goto found; | |
3050 | next: ; | |
87a8ebd6 | 3051 | } |
e51db735 | 3052 | found: |
44bb4385 | 3053 | up_read(&namespace_sem); |
e51db735 | 3054 | return visible; |
87a8ebd6 EB |
3055 | } |
3056 | ||
8823c079 EB |
3057 | static void *mntns_get(struct task_struct *task) |
3058 | { | |
3059 | struct mnt_namespace *ns = NULL; | |
3060 | struct nsproxy *nsproxy; | |
3061 | ||
728dba3a EB |
3062 | task_lock(task); |
3063 | nsproxy = task->nsproxy; | |
8823c079 EB |
3064 | if (nsproxy) { |
3065 | ns = nsproxy->mnt_ns; | |
3066 | get_mnt_ns(ns); | |
3067 | } | |
728dba3a | 3068 | task_unlock(task); |
8823c079 EB |
3069 | |
3070 | return ns; | |
3071 | } | |
3072 | ||
3073 | static void mntns_put(void *ns) | |
3074 | { | |
3075 | put_mnt_ns(ns); | |
3076 | } | |
3077 | ||
3078 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
3079 | { | |
3080 | struct fs_struct *fs = current->fs; | |
3081 | struct mnt_namespace *mnt_ns = ns; | |
3082 | struct path root; | |
3083 | ||
0c55cfc4 | 3084 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3085 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3086 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3087 | return -EPERM; |
8823c079 EB |
3088 | |
3089 | if (fs->users != 1) | |
3090 | return -EINVAL; | |
3091 | ||
3092 | get_mnt_ns(mnt_ns); | |
3093 | put_mnt_ns(nsproxy->mnt_ns); | |
3094 | nsproxy->mnt_ns = mnt_ns; | |
3095 | ||
3096 | /* Find the root */ | |
3097 | root.mnt = &mnt_ns->root->mnt; | |
3098 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3099 | path_get(&root); | |
3100 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3101 | ; | |
3102 | ||
3103 | /* Update the pwd and root */ | |
3104 | set_fs_pwd(fs, &root); | |
3105 | set_fs_root(fs, &root); | |
3106 | ||
3107 | path_put(&root); | |
3108 | return 0; | |
3109 | } | |
3110 | ||
98f842e6 EB |
3111 | static unsigned int mntns_inum(void *ns) |
3112 | { | |
3113 | struct mnt_namespace *mnt_ns = ns; | |
3114 | return mnt_ns->proc_inum; | |
3115 | } | |
3116 | ||
8823c079 EB |
3117 | const struct proc_ns_operations mntns_operations = { |
3118 | .name = "mnt", | |
3119 | .type = CLONE_NEWNS, | |
3120 | .get = mntns_get, | |
3121 | .put = mntns_put, | |
3122 | .install = mntns_install, | |
98f842e6 | 3123 | .inum = mntns_inum, |
8823c079 | 3124 | }; |