]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
d10577a8 AV |
19 | #include <linux/acct.h> /* acct_auto_close_mnt */ |
20 | #include <linux/ramfs.h> /* init_rootfs */ | |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ | |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
8823c079 | 24 | #include <linux/proc_fs.h> |
07b20889 | 25 | #include "pnode.h" |
948730b0 | 26 | #include "internal.h" |
1da177e4 | 27 | |
13f14b4d ED |
28 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
29 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
30 | ||
5addc5dd | 31 | static int event; |
73cd49ec | 32 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 33 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 34 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
35 | static int mnt_id_start = 0; |
36 | static int mnt_group_start = 1; | |
1da177e4 | 37 | |
fa3536cc | 38 | static struct list_head *mount_hashtable __read_mostly; |
84d17192 | 39 | static struct list_head *mountpoint_hashtable __read_mostly; |
e18b890b | 40 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 41 | static struct rw_semaphore namespace_sem; |
1da177e4 | 42 | |
f87fd4c2 | 43 | /* /sys/fs */ |
00d26666 GKH |
44 | struct kobject *fs_kobj; |
45 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 46 | |
99b7db7b NP |
47 | /* |
48 | * vfsmount lock may be taken for read to prevent changes to the | |
49 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
50 | * up the tree. | |
51 | * | |
52 | * It should be taken for write in all cases where the vfsmount | |
53 | * tree or hash is modified or when a vfsmount structure is modified. | |
54 | */ | |
55 | DEFINE_BRLOCK(vfsmount_lock); | |
56 | ||
1da177e4 LT |
57 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
58 | { | |
b58fed8b RP |
59 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
60 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
61 | tmp = tmp + (tmp >> HASH_SHIFT); |
62 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
63 | } |
64 | ||
3d733633 DH |
65 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
66 | ||
99b7db7b NP |
67 | /* |
68 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
69 | * serialize with freeing. | |
70 | */ | |
b105e270 | 71 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
72 | { |
73 | int res; | |
74 | ||
75 | retry: | |
76 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 77 | spin_lock(&mnt_id_lock); |
15169fe7 | 78 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 79 | if (!res) |
15169fe7 | 80 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 81 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
82 | if (res == -EAGAIN) |
83 | goto retry; | |
84 | ||
85 | return res; | |
86 | } | |
87 | ||
b105e270 | 88 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 89 | { |
15169fe7 | 90 | int id = mnt->mnt_id; |
99b7db7b | 91 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
92 | ida_remove(&mnt_id_ida, id); |
93 | if (mnt_id_start > id) | |
94 | mnt_id_start = id; | |
99b7db7b | 95 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
96 | } |
97 | ||
719f5d7f MS |
98 | /* |
99 | * Allocate a new peer group ID | |
100 | * | |
101 | * mnt_group_ida is protected by namespace_sem | |
102 | */ | |
4b8b21f4 | 103 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 104 | { |
f21f6220 AV |
105 | int res; |
106 | ||
719f5d7f MS |
107 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
108 | return -ENOMEM; | |
109 | ||
f21f6220 AV |
110 | res = ida_get_new_above(&mnt_group_ida, |
111 | mnt_group_start, | |
15169fe7 | 112 | &mnt->mnt_group_id); |
f21f6220 | 113 | if (!res) |
15169fe7 | 114 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
115 | |
116 | return res; | |
719f5d7f MS |
117 | } |
118 | ||
119 | /* | |
120 | * Release a peer group ID | |
121 | */ | |
4b8b21f4 | 122 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 123 | { |
15169fe7 | 124 | int id = mnt->mnt_group_id; |
f21f6220 AV |
125 | ida_remove(&mnt_group_ida, id); |
126 | if (mnt_group_start > id) | |
127 | mnt_group_start = id; | |
15169fe7 | 128 | mnt->mnt_group_id = 0; |
719f5d7f MS |
129 | } |
130 | ||
b3e19d92 NP |
131 | /* |
132 | * vfsmount lock must be held for read | |
133 | */ | |
83adc753 | 134 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
135 | { |
136 | #ifdef CONFIG_SMP | |
68e8a9fe | 137 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
138 | #else |
139 | preempt_disable(); | |
68e8a9fe | 140 | mnt->mnt_count += n; |
b3e19d92 NP |
141 | preempt_enable(); |
142 | #endif | |
143 | } | |
144 | ||
b3e19d92 NP |
145 | /* |
146 | * vfsmount lock must be held for write | |
147 | */ | |
83adc753 | 148 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
149 | { |
150 | #ifdef CONFIG_SMP | |
f03c6599 | 151 | unsigned int count = 0; |
b3e19d92 NP |
152 | int cpu; |
153 | ||
154 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 155 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
156 | } |
157 | ||
158 | return count; | |
159 | #else | |
68e8a9fe | 160 | return mnt->mnt_count; |
b3e19d92 NP |
161 | #endif |
162 | } | |
163 | ||
b105e270 | 164 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 165 | { |
c63181e6 AV |
166 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
167 | if (mnt) { | |
73cd49ec MS |
168 | int err; |
169 | ||
c63181e6 | 170 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
171 | if (err) |
172 | goto out_free_cache; | |
173 | ||
174 | if (name) { | |
c63181e6 AV |
175 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
176 | if (!mnt->mnt_devname) | |
88b38782 | 177 | goto out_free_id; |
73cd49ec MS |
178 | } |
179 | ||
b3e19d92 | 180 | #ifdef CONFIG_SMP |
c63181e6 AV |
181 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
182 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
183 | goto out_free_devname; |
184 | ||
c63181e6 | 185 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 186 | #else |
c63181e6 AV |
187 | mnt->mnt_count = 1; |
188 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
189 | #endif |
190 | ||
c63181e6 AV |
191 | INIT_LIST_HEAD(&mnt->mnt_hash); |
192 | INIT_LIST_HEAD(&mnt->mnt_child); | |
193 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
194 | INIT_LIST_HEAD(&mnt->mnt_list); | |
195 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
196 | INIT_LIST_HEAD(&mnt->mnt_share); | |
197 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
198 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
199 | #ifdef CONFIG_FSNOTIFY |
200 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 201 | #endif |
1da177e4 | 202 | } |
c63181e6 | 203 | return mnt; |
88b38782 | 204 | |
d3ef3d73 NP |
205 | #ifdef CONFIG_SMP |
206 | out_free_devname: | |
c63181e6 | 207 | kfree(mnt->mnt_devname); |
d3ef3d73 | 208 | #endif |
88b38782 | 209 | out_free_id: |
c63181e6 | 210 | mnt_free_id(mnt); |
88b38782 | 211 | out_free_cache: |
c63181e6 | 212 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 213 | return NULL; |
1da177e4 LT |
214 | } |
215 | ||
3d733633 DH |
216 | /* |
217 | * Most r/o checks on a fs are for operations that take | |
218 | * discrete amounts of time, like a write() or unlink(). | |
219 | * We must keep track of when those operations start | |
220 | * (for permission checks) and when they end, so that | |
221 | * we can determine when writes are able to occur to | |
222 | * a filesystem. | |
223 | */ | |
224 | /* | |
225 | * __mnt_is_readonly: check whether a mount is read-only | |
226 | * @mnt: the mount to check for its write status | |
227 | * | |
228 | * This shouldn't be used directly ouside of the VFS. | |
229 | * It does not guarantee that the filesystem will stay | |
230 | * r/w, just that it is right *now*. This can not and | |
231 | * should not be used in place of IS_RDONLY(inode). | |
232 | * mnt_want/drop_write() will _keep_ the filesystem | |
233 | * r/w. | |
234 | */ | |
235 | int __mnt_is_readonly(struct vfsmount *mnt) | |
236 | { | |
2e4b7fcd DH |
237 | if (mnt->mnt_flags & MNT_READONLY) |
238 | return 1; | |
239 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
240 | return 1; | |
241 | return 0; | |
3d733633 DH |
242 | } |
243 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
244 | ||
83adc753 | 245 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
246 | { |
247 | #ifdef CONFIG_SMP | |
68e8a9fe | 248 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 249 | #else |
68e8a9fe | 250 | mnt->mnt_writers++; |
d3ef3d73 NP |
251 | #endif |
252 | } | |
3d733633 | 253 | |
83adc753 | 254 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 255 | { |
d3ef3d73 | 256 | #ifdef CONFIG_SMP |
68e8a9fe | 257 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 258 | #else |
68e8a9fe | 259 | mnt->mnt_writers--; |
d3ef3d73 | 260 | #endif |
3d733633 | 261 | } |
3d733633 | 262 | |
83adc753 | 263 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 264 | { |
d3ef3d73 NP |
265 | #ifdef CONFIG_SMP |
266 | unsigned int count = 0; | |
3d733633 | 267 | int cpu; |
3d733633 DH |
268 | |
269 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 270 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 271 | } |
3d733633 | 272 | |
d3ef3d73 NP |
273 | return count; |
274 | #else | |
275 | return mnt->mnt_writers; | |
276 | #endif | |
3d733633 DH |
277 | } |
278 | ||
4ed5e82f MS |
279 | static int mnt_is_readonly(struct vfsmount *mnt) |
280 | { | |
281 | if (mnt->mnt_sb->s_readonly_remount) | |
282 | return 1; | |
283 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
284 | smp_rmb(); | |
285 | return __mnt_is_readonly(mnt); | |
286 | } | |
287 | ||
8366025e | 288 | /* |
eb04c282 JK |
289 | * Most r/o & frozen checks on a fs are for operations that take discrete |
290 | * amounts of time, like a write() or unlink(). We must keep track of when | |
291 | * those operations start (for permission checks) and when they end, so that we | |
292 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
293 | */ |
294 | /** | |
eb04c282 | 295 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 296 | * @m: the mount on which to take a write |
8366025e | 297 | * |
eb04c282 JK |
298 | * This tells the low-level filesystem that a write is about to be performed to |
299 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
300 | * returning success. This operation does not protect against filesystem being | |
301 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
302 | * called. This is effectively a refcount. | |
8366025e | 303 | */ |
eb04c282 | 304 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 305 | { |
83adc753 | 306 | struct mount *mnt = real_mount(m); |
3d733633 | 307 | int ret = 0; |
3d733633 | 308 | |
d3ef3d73 | 309 | preempt_disable(); |
c6653a83 | 310 | mnt_inc_writers(mnt); |
d3ef3d73 | 311 | /* |
c6653a83 | 312 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
313 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
314 | * incremented count after it has set MNT_WRITE_HOLD. | |
315 | */ | |
316 | smp_mb(); | |
1e75529e | 317 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 NP |
318 | cpu_relax(); |
319 | /* | |
320 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
321 | * be set to match its requirements. So we must not load that until | |
322 | * MNT_WRITE_HOLD is cleared. | |
323 | */ | |
324 | smp_rmb(); | |
4ed5e82f | 325 | if (mnt_is_readonly(m)) { |
c6653a83 | 326 | mnt_dec_writers(mnt); |
3d733633 | 327 | ret = -EROFS; |
3d733633 | 328 | } |
d3ef3d73 | 329 | preempt_enable(); |
eb04c282 JK |
330 | |
331 | return ret; | |
332 | } | |
333 | ||
334 | /** | |
335 | * mnt_want_write - get write access to a mount | |
336 | * @m: the mount on which to take a write | |
337 | * | |
338 | * This tells the low-level filesystem that a write is about to be performed to | |
339 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
340 | * is not frozen) before returning success. When the write operation is | |
341 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
342 | */ | |
343 | int mnt_want_write(struct vfsmount *m) | |
344 | { | |
345 | int ret; | |
346 | ||
347 | sb_start_write(m->mnt_sb); | |
348 | ret = __mnt_want_write(m); | |
349 | if (ret) | |
350 | sb_end_write(m->mnt_sb); | |
3d733633 | 351 | return ret; |
8366025e DH |
352 | } |
353 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
354 | ||
96029c4e NP |
355 | /** |
356 | * mnt_clone_write - get write access to a mount | |
357 | * @mnt: the mount on which to take a write | |
358 | * | |
359 | * This is effectively like mnt_want_write, except | |
360 | * it must only be used to take an extra write reference | |
361 | * on a mountpoint that we already know has a write reference | |
362 | * on it. This allows some optimisation. | |
363 | * | |
364 | * After finished, mnt_drop_write must be called as usual to | |
365 | * drop the reference. | |
366 | */ | |
367 | int mnt_clone_write(struct vfsmount *mnt) | |
368 | { | |
369 | /* superblock may be r/o */ | |
370 | if (__mnt_is_readonly(mnt)) | |
371 | return -EROFS; | |
372 | preempt_disable(); | |
83adc753 | 373 | mnt_inc_writers(real_mount(mnt)); |
96029c4e NP |
374 | preempt_enable(); |
375 | return 0; | |
376 | } | |
377 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
378 | ||
379 | /** | |
eb04c282 | 380 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
381 | * @file: the file who's mount on which to take a write |
382 | * | |
eb04c282 | 383 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e NP |
384 | * do some optimisations if the file is open for write already |
385 | */ | |
eb04c282 | 386 | int __mnt_want_write_file(struct file *file) |
96029c4e | 387 | { |
496ad9aa | 388 | struct inode *inode = file_inode(file); |
eb04c282 | 389 | |
2d8dd38a | 390 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) |
eb04c282 | 391 | return __mnt_want_write(file->f_path.mnt); |
96029c4e NP |
392 | else |
393 | return mnt_clone_write(file->f_path.mnt); | |
394 | } | |
eb04c282 JK |
395 | |
396 | /** | |
397 | * mnt_want_write_file - get write access to a file's mount | |
398 | * @file: the file who's mount on which to take a write | |
399 | * | |
400 | * This is like mnt_want_write, but it takes a file and can | |
401 | * do some optimisations if the file is open for write already | |
402 | */ | |
403 | int mnt_want_write_file(struct file *file) | |
404 | { | |
405 | int ret; | |
406 | ||
407 | sb_start_write(file->f_path.mnt->mnt_sb); | |
408 | ret = __mnt_want_write_file(file); | |
409 | if (ret) | |
410 | sb_end_write(file->f_path.mnt->mnt_sb); | |
411 | return ret; | |
412 | } | |
96029c4e NP |
413 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
414 | ||
8366025e | 415 | /** |
eb04c282 | 416 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
417 | * @mnt: the mount on which to give up write access |
418 | * | |
419 | * Tells the low-level filesystem that we are done | |
420 | * performing writes to it. Must be matched with | |
eb04c282 | 421 | * __mnt_want_write() call above. |
8366025e | 422 | */ |
eb04c282 | 423 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 424 | { |
d3ef3d73 | 425 | preempt_disable(); |
83adc753 | 426 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 427 | preempt_enable(); |
8366025e | 428 | } |
eb04c282 JK |
429 | |
430 | /** | |
431 | * mnt_drop_write - give up write access to a mount | |
432 | * @mnt: the mount on which to give up write access | |
433 | * | |
434 | * Tells the low-level filesystem that we are done performing writes to it and | |
435 | * also allows filesystem to be frozen again. Must be matched with | |
436 | * mnt_want_write() call above. | |
437 | */ | |
438 | void mnt_drop_write(struct vfsmount *mnt) | |
439 | { | |
440 | __mnt_drop_write(mnt); | |
441 | sb_end_write(mnt->mnt_sb); | |
442 | } | |
8366025e DH |
443 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
444 | ||
eb04c282 JK |
445 | void __mnt_drop_write_file(struct file *file) |
446 | { | |
447 | __mnt_drop_write(file->f_path.mnt); | |
448 | } | |
449 | ||
2a79f17e AV |
450 | void mnt_drop_write_file(struct file *file) |
451 | { | |
452 | mnt_drop_write(file->f_path.mnt); | |
453 | } | |
454 | EXPORT_SYMBOL(mnt_drop_write_file); | |
455 | ||
83adc753 | 456 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 457 | { |
3d733633 DH |
458 | int ret = 0; |
459 | ||
962830df | 460 | br_write_lock(&vfsmount_lock); |
83adc753 | 461 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 462 | /* |
d3ef3d73 NP |
463 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
464 | * should be visible before we do. | |
3d733633 | 465 | */ |
d3ef3d73 NP |
466 | smp_mb(); |
467 | ||
3d733633 | 468 | /* |
d3ef3d73 NP |
469 | * With writers on hold, if this value is zero, then there are |
470 | * definitely no active writers (although held writers may subsequently | |
471 | * increment the count, they'll have to wait, and decrement it after | |
472 | * seeing MNT_READONLY). | |
473 | * | |
474 | * It is OK to have counter incremented on one CPU and decremented on | |
475 | * another: the sum will add up correctly. The danger would be when we | |
476 | * sum up each counter, if we read a counter before it is incremented, | |
477 | * but then read another CPU's count which it has been subsequently | |
478 | * decremented from -- we would see more decrements than we should. | |
479 | * MNT_WRITE_HOLD protects against this scenario, because | |
480 | * mnt_want_write first increments count, then smp_mb, then spins on | |
481 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
482 | * we're counting up here. | |
3d733633 | 483 | */ |
c6653a83 | 484 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 NP |
485 | ret = -EBUSY; |
486 | else | |
83adc753 | 487 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 NP |
488 | /* |
489 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
490 | * that become unheld will see MNT_READONLY. | |
491 | */ | |
492 | smp_wmb(); | |
83adc753 | 493 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
962830df | 494 | br_write_unlock(&vfsmount_lock); |
3d733633 | 495 | return ret; |
8366025e | 496 | } |
8366025e | 497 | |
83adc753 | 498 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 499 | { |
962830df | 500 | br_write_lock(&vfsmount_lock); |
83adc753 | 501 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
962830df | 502 | br_write_unlock(&vfsmount_lock); |
2e4b7fcd DH |
503 | } |
504 | ||
4ed5e82f MS |
505 | int sb_prepare_remount_readonly(struct super_block *sb) |
506 | { | |
507 | struct mount *mnt; | |
508 | int err = 0; | |
509 | ||
8e8b8796 MS |
510 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
511 | if (atomic_long_read(&sb->s_remove_count)) | |
512 | return -EBUSY; | |
513 | ||
962830df | 514 | br_write_lock(&vfsmount_lock); |
4ed5e82f MS |
515 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
516 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
517 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
518 | smp_mb(); | |
519 | if (mnt_get_writers(mnt) > 0) { | |
520 | err = -EBUSY; | |
521 | break; | |
522 | } | |
523 | } | |
524 | } | |
8e8b8796 MS |
525 | if (!err && atomic_long_read(&sb->s_remove_count)) |
526 | err = -EBUSY; | |
527 | ||
4ed5e82f MS |
528 | if (!err) { |
529 | sb->s_readonly_remount = 1; | |
530 | smp_wmb(); | |
531 | } | |
532 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
533 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
534 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
535 | } | |
962830df | 536 | br_write_unlock(&vfsmount_lock); |
4ed5e82f MS |
537 | |
538 | return err; | |
539 | } | |
540 | ||
b105e270 | 541 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 542 | { |
52ba1621 | 543 | kfree(mnt->mnt_devname); |
73cd49ec | 544 | mnt_free_id(mnt); |
d3ef3d73 | 545 | #ifdef CONFIG_SMP |
68e8a9fe | 546 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 547 | #endif |
b105e270 | 548 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
549 | } |
550 | ||
551 | /* | |
a05964f3 RP |
552 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
553 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 554 | * vfsmount_lock must be held for read or write. |
1da177e4 | 555 | */ |
c7105365 | 556 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 557 | int dir) |
1da177e4 | 558 | { |
b58fed8b RP |
559 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
560 | struct list_head *tmp = head; | |
c7105365 | 561 | struct mount *p, *found = NULL; |
1da177e4 | 562 | |
1da177e4 | 563 | for (;;) { |
a05964f3 | 564 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
565 | p = NULL; |
566 | if (tmp == head) | |
567 | break; | |
1b8e5564 | 568 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 569 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 570 | found = p; |
1da177e4 LT |
571 | break; |
572 | } | |
573 | } | |
1da177e4 LT |
574 | return found; |
575 | } | |
576 | ||
a05964f3 | 577 | /* |
f015f126 DH |
578 | * lookup_mnt - Return the first child mount mounted at path |
579 | * | |
580 | * "First" means first mounted chronologically. If you create the | |
581 | * following mounts: | |
582 | * | |
583 | * mount /dev/sda1 /mnt | |
584 | * mount /dev/sda2 /mnt | |
585 | * mount /dev/sda3 /mnt | |
586 | * | |
587 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
588 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
589 | * /dev/sda2, then /dev/sda3, then NULL. | |
590 | * | |
591 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 592 | */ |
1c755af4 | 593 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 594 | { |
c7105365 | 595 | struct mount *child_mnt; |
99b7db7b | 596 | |
962830df | 597 | br_read_lock(&vfsmount_lock); |
c7105365 AV |
598 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
599 | if (child_mnt) { | |
600 | mnt_add_count(child_mnt, 1); | |
962830df | 601 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
602 | return &child_mnt->mnt; |
603 | } else { | |
962830df | 604 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
605 | return NULL; |
606 | } | |
a05964f3 RP |
607 | } |
608 | ||
84d17192 AV |
609 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
610 | { | |
611 | struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry); | |
612 | struct mountpoint *mp; | |
613 | ||
614 | list_for_each_entry(mp, chain, m_hash) { | |
615 | if (mp->m_dentry == dentry) { | |
616 | /* might be worth a WARN_ON() */ | |
617 | if (d_unlinked(dentry)) | |
618 | return ERR_PTR(-ENOENT); | |
619 | mp->m_count++; | |
620 | return mp; | |
621 | } | |
622 | } | |
623 | ||
624 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
625 | if (!mp) | |
626 | return ERR_PTR(-ENOMEM); | |
627 | ||
628 | spin_lock(&dentry->d_lock); | |
629 | if (d_unlinked(dentry)) { | |
630 | spin_unlock(&dentry->d_lock); | |
631 | kfree(mp); | |
632 | return ERR_PTR(-ENOENT); | |
633 | } | |
634 | dentry->d_flags |= DCACHE_MOUNTED; | |
635 | spin_unlock(&dentry->d_lock); | |
636 | mp->m_dentry = dentry; | |
637 | mp->m_count = 1; | |
638 | list_add(&mp->m_hash, chain); | |
639 | return mp; | |
640 | } | |
641 | ||
642 | static void put_mountpoint(struct mountpoint *mp) | |
643 | { | |
644 | if (!--mp->m_count) { | |
645 | struct dentry *dentry = mp->m_dentry; | |
646 | spin_lock(&dentry->d_lock); | |
647 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
648 | spin_unlock(&dentry->d_lock); | |
649 | list_del(&mp->m_hash); | |
650 | kfree(mp); | |
651 | } | |
652 | } | |
653 | ||
143c8c91 | 654 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 655 | { |
6b3286ed | 656 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
657 | } |
658 | ||
99b7db7b NP |
659 | /* |
660 | * vfsmount lock must be held for write | |
661 | */ | |
6b3286ed | 662 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
663 | { |
664 | if (ns) { | |
665 | ns->event = ++event; | |
666 | wake_up_interruptible(&ns->poll); | |
667 | } | |
668 | } | |
669 | ||
99b7db7b NP |
670 | /* |
671 | * vfsmount lock must be held for write | |
672 | */ | |
6b3286ed | 673 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
674 | { |
675 | if (ns && ns->event != event) { | |
676 | ns->event = event; | |
677 | wake_up_interruptible(&ns->poll); | |
678 | } | |
679 | } | |
680 | ||
99b7db7b NP |
681 | /* |
682 | * vfsmount lock must be held for write | |
683 | */ | |
419148da AV |
684 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
685 | { | |
a73324da | 686 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
687 | old_path->mnt = &mnt->mnt_parent->mnt; |
688 | mnt->mnt_parent = mnt; | |
a73324da | 689 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 690 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 691 | list_del_init(&mnt->mnt_hash); |
84d17192 AV |
692 | put_mountpoint(mnt->mnt_mp); |
693 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
694 | } |
695 | ||
99b7db7b NP |
696 | /* |
697 | * vfsmount lock must be held for write | |
698 | */ | |
84d17192 AV |
699 | void mnt_set_mountpoint(struct mount *mnt, |
700 | struct mountpoint *mp, | |
44d964d6 | 701 | struct mount *child_mnt) |
b90fa9ae | 702 | { |
84d17192 | 703 | mp->m_count++; |
3a2393d7 | 704 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 705 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 706 | child_mnt->mnt_parent = mnt; |
84d17192 | 707 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
708 | } |
709 | ||
99b7db7b NP |
710 | /* |
711 | * vfsmount lock must be held for write | |
712 | */ | |
84d17192 AV |
713 | static void attach_mnt(struct mount *mnt, |
714 | struct mount *parent, | |
715 | struct mountpoint *mp) | |
1da177e4 | 716 | { |
84d17192 | 717 | mnt_set_mountpoint(parent, mp, mnt); |
1b8e5564 | 718 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
84d17192 AV |
719 | hash(&parent->mnt, mp->m_dentry)); |
720 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
b90fa9ae RP |
721 | } |
722 | ||
723 | /* | |
99b7db7b | 724 | * vfsmount lock must be held for write |
b90fa9ae | 725 | */ |
4b2619a5 | 726 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 727 | { |
0714a533 | 728 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 729 | struct mount *m; |
b90fa9ae | 730 | LIST_HEAD(head); |
143c8c91 | 731 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 732 | |
0714a533 | 733 | BUG_ON(parent == mnt); |
b90fa9ae | 734 | |
1a4eeaf2 | 735 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 736 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 737 | m->mnt_ns = n; |
f03c6599 | 738 | |
b90fa9ae RP |
739 | list_splice(&head, n->list.prev); |
740 | ||
1b8e5564 | 741 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 742 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 743 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 744 | touch_mnt_namespace(n); |
1da177e4 LT |
745 | } |
746 | ||
909b0a88 | 747 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 748 | { |
6b41d536 AV |
749 | struct list_head *next = p->mnt_mounts.next; |
750 | if (next == &p->mnt_mounts) { | |
1da177e4 | 751 | while (1) { |
909b0a88 | 752 | if (p == root) |
1da177e4 | 753 | return NULL; |
6b41d536 AV |
754 | next = p->mnt_child.next; |
755 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 756 | break; |
0714a533 | 757 | p = p->mnt_parent; |
1da177e4 LT |
758 | } |
759 | } | |
6b41d536 | 760 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
761 | } |
762 | ||
315fc83e | 763 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 764 | { |
6b41d536 AV |
765 | struct list_head *prev = p->mnt_mounts.prev; |
766 | while (prev != &p->mnt_mounts) { | |
767 | p = list_entry(prev, struct mount, mnt_child); | |
768 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
769 | } |
770 | return p; | |
771 | } | |
772 | ||
9d412a43 AV |
773 | struct vfsmount * |
774 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
775 | { | |
b105e270 | 776 | struct mount *mnt; |
9d412a43 AV |
777 | struct dentry *root; |
778 | ||
779 | if (!type) | |
780 | return ERR_PTR(-ENODEV); | |
781 | ||
782 | mnt = alloc_vfsmnt(name); | |
783 | if (!mnt) | |
784 | return ERR_PTR(-ENOMEM); | |
785 | ||
786 | if (flags & MS_KERNMOUNT) | |
b105e270 | 787 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
788 | |
789 | root = mount_fs(type, flags, name, data); | |
790 | if (IS_ERR(root)) { | |
791 | free_vfsmnt(mnt); | |
792 | return ERR_CAST(root); | |
793 | } | |
794 | ||
b105e270 AV |
795 | mnt->mnt.mnt_root = root; |
796 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 797 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 798 | mnt->mnt_parent = mnt; |
962830df | 799 | br_write_lock(&vfsmount_lock); |
39f7c4db | 800 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
962830df | 801 | br_write_unlock(&vfsmount_lock); |
b105e270 | 802 | return &mnt->mnt; |
9d412a43 AV |
803 | } |
804 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
805 | ||
87129cc0 | 806 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 807 | int flag) |
1da177e4 | 808 | { |
87129cc0 | 809 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
810 | struct mount *mnt; |
811 | int err; | |
1da177e4 | 812 | |
be34d1a3 DH |
813 | mnt = alloc_vfsmnt(old->mnt_devname); |
814 | if (!mnt) | |
815 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 816 | |
7a472ef4 | 817 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
818 | mnt->mnt_group_id = 0; /* not a peer of original */ |
819 | else | |
820 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 821 | |
be34d1a3 DH |
822 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
823 | err = mnt_alloc_group_id(mnt); | |
824 | if (err) | |
825 | goto out_free; | |
1da177e4 | 826 | } |
be34d1a3 DH |
827 | |
828 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
829 | atomic_inc(&sb->s_active); | |
830 | mnt->mnt.mnt_sb = sb; | |
831 | mnt->mnt.mnt_root = dget(root); | |
832 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
833 | mnt->mnt_parent = mnt; | |
834 | br_write_lock(&vfsmount_lock); | |
835 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
836 | br_write_unlock(&vfsmount_lock); | |
837 | ||
7a472ef4 EB |
838 | if ((flag & CL_SLAVE) || |
839 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
840 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
841 | mnt->mnt_master = old; | |
842 | CLEAR_MNT_SHARED(mnt); | |
843 | } else if (!(flag & CL_PRIVATE)) { | |
844 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
845 | list_add(&mnt->mnt_share, &old->mnt_share); | |
846 | if (IS_MNT_SLAVE(old)) | |
847 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
848 | mnt->mnt_master = old->mnt_master; | |
849 | } | |
850 | if (flag & CL_MAKE_SHARED) | |
851 | set_mnt_shared(mnt); | |
852 | ||
853 | /* stick the duplicate mount on the same expiry list | |
854 | * as the original if that was on one */ | |
855 | if (flag & CL_EXPIRE) { | |
856 | if (!list_empty(&old->mnt_expire)) | |
857 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
858 | } | |
859 | ||
cb338d06 | 860 | return mnt; |
719f5d7f MS |
861 | |
862 | out_free: | |
863 | free_vfsmnt(mnt); | |
be34d1a3 | 864 | return ERR_PTR(err); |
1da177e4 LT |
865 | } |
866 | ||
83adc753 | 867 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 868 | { |
83adc753 AV |
869 | struct vfsmount *m = &mnt->mnt; |
870 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 871 | |
3d733633 DH |
872 | /* |
873 | * This probably indicates that somebody messed | |
874 | * up a mnt_want/drop_write() pair. If this | |
875 | * happens, the filesystem was probably unable | |
876 | * to make r/w->r/o transitions. | |
877 | */ | |
d3ef3d73 | 878 | /* |
b3e19d92 NP |
879 | * The locking used to deal with mnt_count decrement provides barriers, |
880 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 881 | */ |
c6653a83 | 882 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
883 | fsnotify_vfsmount_delete(m); |
884 | dput(m->mnt_root); | |
885 | free_vfsmnt(mnt); | |
1da177e4 LT |
886 | deactivate_super(sb); |
887 | } | |
888 | ||
900148dc | 889 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 890 | { |
b3e19d92 | 891 | put_again: |
f03c6599 | 892 | #ifdef CONFIG_SMP |
962830df | 893 | br_read_lock(&vfsmount_lock); |
f7a99c5b AV |
894 | if (likely(mnt->mnt_ns)) { |
895 | /* shouldn't be the last one */ | |
aa9c0e07 | 896 | mnt_add_count(mnt, -1); |
962830df | 897 | br_read_unlock(&vfsmount_lock); |
f03c6599 | 898 | return; |
b3e19d92 | 899 | } |
962830df | 900 | br_read_unlock(&vfsmount_lock); |
b3e19d92 | 901 | |
962830df | 902 | br_write_lock(&vfsmount_lock); |
aa9c0e07 | 903 | mnt_add_count(mnt, -1); |
b3e19d92 | 904 | if (mnt_get_count(mnt)) { |
962830df | 905 | br_write_unlock(&vfsmount_lock); |
99b7db7b NP |
906 | return; |
907 | } | |
b3e19d92 | 908 | #else |
aa9c0e07 | 909 | mnt_add_count(mnt, -1); |
b3e19d92 | 910 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 911 | return; |
962830df | 912 | br_write_lock(&vfsmount_lock); |
f03c6599 | 913 | #endif |
863d684f AV |
914 | if (unlikely(mnt->mnt_pinned)) { |
915 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
916 | mnt->mnt_pinned = 0; | |
962830df | 917 | br_write_unlock(&vfsmount_lock); |
900148dc | 918 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 919 | goto put_again; |
7b7b1ace | 920 | } |
962830df | 921 | |
39f7c4db | 922 | list_del(&mnt->mnt_instance); |
962830df | 923 | br_write_unlock(&vfsmount_lock); |
b3e19d92 NP |
924 | mntfree(mnt); |
925 | } | |
b3e19d92 NP |
926 | |
927 | void mntput(struct vfsmount *mnt) | |
928 | { | |
929 | if (mnt) { | |
863d684f | 930 | struct mount *m = real_mount(mnt); |
b3e19d92 | 931 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
932 | if (unlikely(m->mnt_expiry_mark)) |
933 | m->mnt_expiry_mark = 0; | |
934 | mntput_no_expire(m); | |
b3e19d92 NP |
935 | } |
936 | } | |
937 | EXPORT_SYMBOL(mntput); | |
938 | ||
939 | struct vfsmount *mntget(struct vfsmount *mnt) | |
940 | { | |
941 | if (mnt) | |
83adc753 | 942 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
943 | return mnt; |
944 | } | |
945 | EXPORT_SYMBOL(mntget); | |
946 | ||
7b7b1ace AV |
947 | void mnt_pin(struct vfsmount *mnt) |
948 | { | |
962830df | 949 | br_write_lock(&vfsmount_lock); |
863d684f | 950 | real_mount(mnt)->mnt_pinned++; |
962830df | 951 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 952 | } |
7b7b1ace AV |
953 | EXPORT_SYMBOL(mnt_pin); |
954 | ||
863d684f | 955 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 956 | { |
863d684f | 957 | struct mount *mnt = real_mount(m); |
962830df | 958 | br_write_lock(&vfsmount_lock); |
7b7b1ace | 959 | if (mnt->mnt_pinned) { |
863d684f | 960 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
961 | mnt->mnt_pinned--; |
962 | } | |
962830df | 963 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 964 | } |
7b7b1ace | 965 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 966 | |
b3b304a2 MS |
967 | static inline void mangle(struct seq_file *m, const char *s) |
968 | { | |
969 | seq_escape(m, s, " \t\n\\"); | |
970 | } | |
971 | ||
972 | /* | |
973 | * Simple .show_options callback for filesystems which don't want to | |
974 | * implement more complex mount option showing. | |
975 | * | |
976 | * See also save_mount_options(). | |
977 | */ | |
34c80b1d | 978 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 979 | { |
2a32cebd AV |
980 | const char *options; |
981 | ||
982 | rcu_read_lock(); | |
34c80b1d | 983 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
984 | |
985 | if (options != NULL && options[0]) { | |
986 | seq_putc(m, ','); | |
987 | mangle(m, options); | |
988 | } | |
2a32cebd | 989 | rcu_read_unlock(); |
b3b304a2 MS |
990 | |
991 | return 0; | |
992 | } | |
993 | EXPORT_SYMBOL(generic_show_options); | |
994 | ||
995 | /* | |
996 | * If filesystem uses generic_show_options(), this function should be | |
997 | * called from the fill_super() callback. | |
998 | * | |
999 | * The .remount_fs callback usually needs to be handled in a special | |
1000 | * way, to make sure, that previous options are not overwritten if the | |
1001 | * remount fails. | |
1002 | * | |
1003 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1004 | * reset all options to their default value, but changes only newly | |
1005 | * given options, then the displayed options will not reflect reality | |
1006 | * any more. | |
1007 | */ | |
1008 | void save_mount_options(struct super_block *sb, char *options) | |
1009 | { | |
2a32cebd AV |
1010 | BUG_ON(sb->s_options); |
1011 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1012 | } |
1013 | EXPORT_SYMBOL(save_mount_options); | |
1014 | ||
2a32cebd AV |
1015 | void replace_mount_options(struct super_block *sb, char *options) |
1016 | { | |
1017 | char *old = sb->s_options; | |
1018 | rcu_assign_pointer(sb->s_options, options); | |
1019 | if (old) { | |
1020 | synchronize_rcu(); | |
1021 | kfree(old); | |
1022 | } | |
1023 | } | |
1024 | EXPORT_SYMBOL(replace_mount_options); | |
1025 | ||
a1a2c409 | 1026 | #ifdef CONFIG_PROC_FS |
0226f492 | 1027 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1028 | static void *m_start(struct seq_file *m, loff_t *pos) |
1029 | { | |
6ce6e24e | 1030 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1031 | |
390c6843 | 1032 | down_read(&namespace_sem); |
a1a2c409 | 1033 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
1034 | } |
1035 | ||
1036 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1037 | { | |
6ce6e24e | 1038 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1039 | |
a1a2c409 | 1040 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
1041 | } |
1042 | ||
1043 | static void m_stop(struct seq_file *m, void *v) | |
1044 | { | |
390c6843 | 1045 | up_read(&namespace_sem); |
1da177e4 LT |
1046 | } |
1047 | ||
0226f492 | 1048 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1049 | { |
6ce6e24e | 1050 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1051 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1052 | return p->show(m, &r->mnt); |
1da177e4 LT |
1053 | } |
1054 | ||
a1a2c409 | 1055 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1056 | .start = m_start, |
1057 | .next = m_next, | |
1058 | .stop = m_stop, | |
0226f492 | 1059 | .show = m_show, |
b4629fe2 | 1060 | }; |
a1a2c409 | 1061 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1062 | |
1da177e4 LT |
1063 | /** |
1064 | * may_umount_tree - check if a mount tree is busy | |
1065 | * @mnt: root of mount tree | |
1066 | * | |
1067 | * This is called to check if a tree of mounts has any | |
1068 | * open files, pwds, chroots or sub mounts that are | |
1069 | * busy. | |
1070 | */ | |
909b0a88 | 1071 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1072 | { |
909b0a88 | 1073 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1074 | int actual_refs = 0; |
1075 | int minimum_refs = 0; | |
315fc83e | 1076 | struct mount *p; |
909b0a88 | 1077 | BUG_ON(!m); |
1da177e4 | 1078 | |
b3e19d92 | 1079 | /* write lock needed for mnt_get_count */ |
962830df | 1080 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1081 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1082 | actual_refs += mnt_get_count(p); |
1da177e4 | 1083 | minimum_refs += 2; |
1da177e4 | 1084 | } |
962830df | 1085 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1086 | |
1087 | if (actual_refs > minimum_refs) | |
e3474a8e | 1088 | return 0; |
1da177e4 | 1089 | |
e3474a8e | 1090 | return 1; |
1da177e4 LT |
1091 | } |
1092 | ||
1093 | EXPORT_SYMBOL(may_umount_tree); | |
1094 | ||
1095 | /** | |
1096 | * may_umount - check if a mount point is busy | |
1097 | * @mnt: root of mount | |
1098 | * | |
1099 | * This is called to check if a mount point has any | |
1100 | * open files, pwds, chroots or sub mounts. If the | |
1101 | * mount has sub mounts this will return busy | |
1102 | * regardless of whether the sub mounts are busy. | |
1103 | * | |
1104 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1105 | * give false negatives. The main reason why it's here is that we need | |
1106 | * a non-destructive way to look for easily umountable filesystems. | |
1107 | */ | |
1108 | int may_umount(struct vfsmount *mnt) | |
1109 | { | |
e3474a8e | 1110 | int ret = 1; |
8ad08d8a | 1111 | down_read(&namespace_sem); |
962830df | 1112 | br_write_lock(&vfsmount_lock); |
1ab59738 | 1113 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1114 | ret = 0; |
962830df | 1115 | br_write_unlock(&vfsmount_lock); |
8ad08d8a | 1116 | up_read(&namespace_sem); |
a05964f3 | 1117 | return ret; |
1da177e4 LT |
1118 | } |
1119 | ||
1120 | EXPORT_SYMBOL(may_umount); | |
1121 | ||
b90fa9ae | 1122 | void release_mounts(struct list_head *head) |
70fbcdf4 | 1123 | { |
d5e50f74 | 1124 | struct mount *mnt; |
bf066c7d | 1125 | while (!list_empty(head)) { |
1b8e5564 AV |
1126 | mnt = list_first_entry(head, struct mount, mnt_hash); |
1127 | list_del_init(&mnt->mnt_hash); | |
676da58d | 1128 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 1129 | struct dentry *dentry; |
863d684f | 1130 | struct mount *m; |
99b7db7b | 1131 | |
962830df | 1132 | br_write_lock(&vfsmount_lock); |
a73324da | 1133 | dentry = mnt->mnt_mountpoint; |
863d684f | 1134 | m = mnt->mnt_parent; |
a73324da | 1135 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1136 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1137 | m->mnt_ghosts--; |
962830df | 1138 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1139 | dput(dentry); |
863d684f | 1140 | mntput(&m->mnt); |
70fbcdf4 | 1141 | } |
d5e50f74 | 1142 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1143 | } |
1144 | } | |
1145 | ||
99b7db7b NP |
1146 | /* |
1147 | * vfsmount lock must be held for write | |
1148 | * namespace_sem must be held for write | |
1149 | */ | |
761d5c38 | 1150 | void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) |
1da177e4 | 1151 | { |
7b8a53fd | 1152 | LIST_HEAD(tmp_list); |
315fc83e | 1153 | struct mount *p; |
1da177e4 | 1154 | |
909b0a88 | 1155 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1156 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1157 | |
a05964f3 | 1158 | if (propagate) |
7b8a53fd | 1159 | propagate_umount(&tmp_list); |
a05964f3 | 1160 | |
1b8e5564 | 1161 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1162 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1163 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1164 | __touch_mnt_namespace(p->mnt_ns); |
1165 | p->mnt_ns = NULL; | |
6b41d536 | 1166 | list_del_init(&p->mnt_child); |
676da58d | 1167 | if (mnt_has_parent(p)) { |
863d684f | 1168 | p->mnt_parent->mnt_ghosts++; |
84d17192 AV |
1169 | put_mountpoint(p->mnt_mp); |
1170 | p->mnt_mp = NULL; | |
7c4b93d8 | 1171 | } |
0f0afb1d | 1172 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1173 | } |
7b8a53fd | 1174 | list_splice(&tmp_list, kill); |
1da177e4 LT |
1175 | } |
1176 | ||
692afc31 | 1177 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts); |
c35038be | 1178 | |
1ab59738 | 1179 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1180 | { |
1ab59738 | 1181 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 | 1182 | int retval; |
70fbcdf4 | 1183 | LIST_HEAD(umount_list); |
1da177e4 | 1184 | |
1ab59738 | 1185 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1186 | if (retval) |
1187 | return retval; | |
1188 | ||
1189 | /* | |
1190 | * Allow userspace to request a mountpoint be expired rather than | |
1191 | * unmounting unconditionally. Unmount only happens if: | |
1192 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1193 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1194 | */ | |
1195 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1196 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1197 | flags & (MNT_FORCE | MNT_DETACH)) |
1198 | return -EINVAL; | |
1199 | ||
b3e19d92 NP |
1200 | /* |
1201 | * probably don't strictly need the lock here if we examined | |
1202 | * all race cases, but it's a slowpath. | |
1203 | */ | |
962830df | 1204 | br_write_lock(&vfsmount_lock); |
83adc753 | 1205 | if (mnt_get_count(mnt) != 2) { |
962830df | 1206 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1207 | return -EBUSY; |
b3e19d92 | 1208 | } |
962830df | 1209 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1210 | |
863d684f | 1211 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1212 | return -EAGAIN; |
1213 | } | |
1214 | ||
1215 | /* | |
1216 | * If we may have to abort operations to get out of this | |
1217 | * mount, and they will themselves hold resources we must | |
1218 | * allow the fs to do things. In the Unix tradition of | |
1219 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1220 | * might fail to complete on the first run through as other tasks | |
1221 | * must return, and the like. Thats for the mount program to worry | |
1222 | * about for the moment. | |
1223 | */ | |
1224 | ||
42faad99 | 1225 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1226 | sb->s_op->umount_begin(sb); |
42faad99 | 1227 | } |
1da177e4 LT |
1228 | |
1229 | /* | |
1230 | * No sense to grab the lock for this test, but test itself looks | |
1231 | * somewhat bogus. Suggestions for better replacement? | |
1232 | * Ho-hum... In principle, we might treat that as umount + switch | |
1233 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1234 | * Actually it makes sense, especially if rootfs would contain a | |
1235 | * /reboot - static binary that would close all descriptors and | |
1236 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1237 | */ | |
1ab59738 | 1238 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1239 | /* |
1240 | * Special case for "unmounting" root ... | |
1241 | * we just try to remount it readonly. | |
1242 | */ | |
1243 | down_write(&sb->s_umount); | |
4aa98cf7 | 1244 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1245 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1246 | up_write(&sb->s_umount); |
1247 | return retval; | |
1248 | } | |
1249 | ||
390c6843 | 1250 | down_write(&namespace_sem); |
962830df | 1251 | br_write_lock(&vfsmount_lock); |
5addc5dd | 1252 | event++; |
1da177e4 | 1253 | |
c35038be | 1254 | if (!(flags & MNT_DETACH)) |
1ab59738 | 1255 | shrink_submounts(mnt, &umount_list); |
c35038be | 1256 | |
1da177e4 | 1257 | retval = -EBUSY; |
a05964f3 | 1258 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1259 | if (!list_empty(&mnt->mnt_list)) |
1ab59738 | 1260 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1261 | retval = 0; |
1262 | } | |
962830df | 1263 | br_write_unlock(&vfsmount_lock); |
390c6843 | 1264 | up_write(&namespace_sem); |
70fbcdf4 | 1265 | release_mounts(&umount_list); |
1da177e4 LT |
1266 | return retval; |
1267 | } | |
1268 | ||
9b40bc90 AV |
1269 | /* |
1270 | * Is the caller allowed to modify his namespace? | |
1271 | */ | |
1272 | static inline bool may_mount(void) | |
1273 | { | |
1274 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1275 | } | |
1276 | ||
1da177e4 LT |
1277 | /* |
1278 | * Now umount can handle mount points as well as block devices. | |
1279 | * This is important for filesystems which use unnamed block devices. | |
1280 | * | |
1281 | * We now support a flag for forced unmount like the other 'big iron' | |
1282 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1283 | */ | |
1284 | ||
bdc480e3 | 1285 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1286 | { |
2d8f3038 | 1287 | struct path path; |
900148dc | 1288 | struct mount *mnt; |
1da177e4 | 1289 | int retval; |
db1f05bb | 1290 | int lookup_flags = 0; |
1da177e4 | 1291 | |
db1f05bb MS |
1292 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1293 | return -EINVAL; | |
1294 | ||
9b40bc90 AV |
1295 | if (!may_mount()) |
1296 | return -EPERM; | |
1297 | ||
db1f05bb MS |
1298 | if (!(flags & UMOUNT_NOFOLLOW)) |
1299 | lookup_flags |= LOOKUP_FOLLOW; | |
1300 | ||
1301 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1302 | if (retval) |
1303 | goto out; | |
900148dc | 1304 | mnt = real_mount(path.mnt); |
1da177e4 | 1305 | retval = -EINVAL; |
2d8f3038 | 1306 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1307 | goto dput_and_out; |
143c8c91 | 1308 | if (!check_mnt(mnt)) |
1da177e4 LT |
1309 | goto dput_and_out; |
1310 | ||
900148dc | 1311 | retval = do_umount(mnt, flags); |
1da177e4 | 1312 | dput_and_out: |
429731b1 | 1313 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1314 | dput(path.dentry); |
900148dc | 1315 | mntput_no_expire(mnt); |
1da177e4 LT |
1316 | out: |
1317 | return retval; | |
1318 | } | |
1319 | ||
1320 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1321 | ||
1322 | /* | |
b58fed8b | 1323 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1324 | */ |
bdc480e3 | 1325 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1326 | { |
b58fed8b | 1327 | return sys_umount(name, 0); |
1da177e4 LT |
1328 | } |
1329 | ||
1330 | #endif | |
1331 | ||
8823c079 EB |
1332 | static bool mnt_ns_loop(struct path *path) |
1333 | { | |
1334 | /* Could bind mounting the mount namespace inode cause a | |
1335 | * mount namespace loop? | |
1336 | */ | |
1337 | struct inode *inode = path->dentry->d_inode; | |
1338 | struct proc_inode *ei; | |
1339 | struct mnt_namespace *mnt_ns; | |
1340 | ||
1341 | if (!proc_ns_inode(inode)) | |
1342 | return false; | |
1343 | ||
1344 | ei = PROC_I(inode); | |
1345 | if (ei->ns_ops != &mntns_operations) | |
1346 | return false; | |
1347 | ||
1348 | mnt_ns = ei->ns; | |
1349 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; | |
1350 | } | |
1351 | ||
87129cc0 | 1352 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1353 | int flag) |
1da177e4 | 1354 | { |
84d17192 | 1355 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1356 | |
fc7be130 | 1357 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
be34d1a3 | 1358 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1359 | |
36341f64 | 1360 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1361 | if (IS_ERR(q)) |
1362 | return q; | |
1363 | ||
a73324da | 1364 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1365 | |
1366 | p = mnt; | |
6b41d536 | 1367 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1368 | struct mount *s; |
7ec02ef1 | 1369 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1370 | continue; |
1371 | ||
909b0a88 | 1372 | for (s = r; s; s = next_mnt(s, r)) { |
fc7be130 | 1373 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
9676f0c6 RP |
1374 | s = skip_mnt_tree(s); |
1375 | continue; | |
1376 | } | |
0714a533 AV |
1377 | while (p != s->mnt_parent) { |
1378 | p = p->mnt_parent; | |
1379 | q = q->mnt_parent; | |
1da177e4 | 1380 | } |
87129cc0 | 1381 | p = s; |
84d17192 | 1382 | parent = q; |
87129cc0 | 1383 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1384 | if (IS_ERR(q)) |
1385 | goto out; | |
962830df | 1386 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 1387 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1388 | attach_mnt(q, parent, p->mnt_mp); |
962830df | 1389 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1390 | } |
1391 | } | |
1392 | return res; | |
be34d1a3 | 1393 | out: |
1da177e4 | 1394 | if (res) { |
70fbcdf4 | 1395 | LIST_HEAD(umount_list); |
962830df | 1396 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1397 | umount_tree(res, 0, &umount_list); |
962830df | 1398 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1399 | release_mounts(&umount_list); |
1da177e4 | 1400 | } |
be34d1a3 | 1401 | return q; |
1da177e4 LT |
1402 | } |
1403 | ||
be34d1a3 DH |
1404 | /* Caller should check returned pointer for errors */ |
1405 | ||
589ff870 | 1406 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1407 | { |
cb338d06 | 1408 | struct mount *tree; |
1a60a280 | 1409 | down_write(&namespace_sem); |
87129cc0 AV |
1410 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1411 | CL_COPY_ALL | CL_PRIVATE); | |
1a60a280 | 1412 | up_write(&namespace_sem); |
be34d1a3 DH |
1413 | if (IS_ERR(tree)) |
1414 | return NULL; | |
1415 | return &tree->mnt; | |
8aec0809 AV |
1416 | } |
1417 | ||
1418 | void drop_collected_mounts(struct vfsmount *mnt) | |
1419 | { | |
1420 | LIST_HEAD(umount_list); | |
1a60a280 | 1421 | down_write(&namespace_sem); |
962830df | 1422 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1423 | umount_tree(real_mount(mnt), 0, &umount_list); |
962830df | 1424 | br_write_unlock(&vfsmount_lock); |
1a60a280 | 1425 | up_write(&namespace_sem); |
8aec0809 AV |
1426 | release_mounts(&umount_list); |
1427 | } | |
1428 | ||
1f707137 AV |
1429 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1430 | struct vfsmount *root) | |
1431 | { | |
1a4eeaf2 | 1432 | struct mount *mnt; |
1f707137 AV |
1433 | int res = f(root, arg); |
1434 | if (res) | |
1435 | return res; | |
1a4eeaf2 AV |
1436 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1437 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1438 | if (res) |
1439 | return res; | |
1440 | } | |
1441 | return 0; | |
1442 | } | |
1443 | ||
4b8b21f4 | 1444 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1445 | { |
315fc83e | 1446 | struct mount *p; |
719f5d7f | 1447 | |
909b0a88 | 1448 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1449 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1450 | mnt_release_group_id(p); |
719f5d7f MS |
1451 | } |
1452 | } | |
1453 | ||
4b8b21f4 | 1454 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1455 | { |
315fc83e | 1456 | struct mount *p; |
719f5d7f | 1457 | |
909b0a88 | 1458 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1459 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1460 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1461 | if (err) { |
4b8b21f4 | 1462 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1463 | return err; |
1464 | } | |
1465 | } | |
1466 | } | |
1467 | ||
1468 | return 0; | |
1469 | } | |
1470 | ||
b90fa9ae RP |
1471 | /* |
1472 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1473 | * @nd : place the mount tree @source_mnt is attached |
1474 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1475 | * store the parent mount and mountpoint dentry. | |
1476 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1477 | * |
1478 | * NOTE: in the table below explains the semantics when a source mount | |
1479 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1480 | * --------------------------------------------------------------------------- |
1481 | * | BIND MOUNT OPERATION | | |
1482 | * |************************************************************************** | |
1483 | * | source-->| shared | private | slave | unbindable | | |
1484 | * | dest | | | | | | |
1485 | * | | | | | | | | |
1486 | * | v | | | | | | |
1487 | * |************************************************************************** | |
1488 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1489 | * | | | | | | | |
1490 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1491 | * *************************************************************************** | |
b90fa9ae RP |
1492 | * A bind operation clones the source mount and mounts the clone on the |
1493 | * destination mount. | |
1494 | * | |
1495 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1496 | * tree of the destination mount and the cloned mount is added to | |
1497 | * the peer group of the source mount. | |
1498 | * (+) the cloned mount is created under the destination mount and is marked | |
1499 | * as shared. The cloned mount is added to the peer group of the source | |
1500 | * mount. | |
5afe0022 RP |
1501 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1502 | * of the destination mount and the cloned mount is made slave | |
1503 | * of the same master as that of the source mount. The cloned mount | |
1504 | * is marked as 'shared and slave'. | |
1505 | * (*) the cloned mount is made a slave of the same master as that of the | |
1506 | * source mount. | |
1507 | * | |
9676f0c6 RP |
1508 | * --------------------------------------------------------------------------- |
1509 | * | MOVE MOUNT OPERATION | | |
1510 | * |************************************************************************** | |
1511 | * | source-->| shared | private | slave | unbindable | | |
1512 | * | dest | | | | | | |
1513 | * | | | | | | | | |
1514 | * | v | | | | | | |
1515 | * |************************************************************************** | |
1516 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1517 | * | | | | | | | |
1518 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1519 | * *************************************************************************** | |
5afe0022 RP |
1520 | * |
1521 | * (+) the mount is moved to the destination. And is then propagated to | |
1522 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1523 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1524 | * (+++) the mount is moved to the destination and is then propagated to |
1525 | * all the mounts belonging to the destination mount's propagation tree. | |
1526 | * the mount is marked as 'shared and slave'. | |
1527 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1528 | * |
1529 | * if the source mount is a tree, the operations explained above is | |
1530 | * applied to each mount in the tree. | |
1531 | * Must be called without spinlocks held, since this function can sleep | |
1532 | * in allocations. | |
1533 | */ | |
0fb54e50 | 1534 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1535 | struct mount *dest_mnt, |
1536 | struct mountpoint *dest_mp, | |
1537 | struct path *parent_path) | |
b90fa9ae RP |
1538 | { |
1539 | LIST_HEAD(tree_list); | |
315fc83e | 1540 | struct mount *child, *p; |
719f5d7f | 1541 | int err; |
b90fa9ae | 1542 | |
fc7be130 | 1543 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1544 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1545 | if (err) |
1546 | goto out; | |
1547 | } | |
84d17192 | 1548 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
719f5d7f MS |
1549 | if (err) |
1550 | goto out_cleanup_ids; | |
b90fa9ae | 1551 | |
962830df | 1552 | br_write_lock(&vfsmount_lock); |
df1a1ad2 | 1553 | |
fc7be130 | 1554 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1555 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1556 | set_mnt_shared(p); |
b90fa9ae | 1557 | } |
1a390689 | 1558 | if (parent_path) { |
0fb54e50 | 1559 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1560 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1561 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1562 | } else { |
84d17192 | 1563 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
0fb54e50 | 1564 | commit_tree(source_mnt); |
21444403 | 1565 | } |
b90fa9ae | 1566 | |
1b8e5564 AV |
1567 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1568 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1569 | commit_tree(child); |
b90fa9ae | 1570 | } |
962830df | 1571 | br_write_unlock(&vfsmount_lock); |
99b7db7b | 1572 | |
b90fa9ae | 1573 | return 0; |
719f5d7f MS |
1574 | |
1575 | out_cleanup_ids: | |
fc7be130 | 1576 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1577 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1578 | out: |
1579 | return err; | |
b90fa9ae RP |
1580 | } |
1581 | ||
84d17192 | 1582 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1583 | { |
1584 | struct vfsmount *mnt; | |
84d17192 | 1585 | struct dentry *dentry = path->dentry; |
b12cea91 | 1586 | retry: |
84d17192 AV |
1587 | mutex_lock(&dentry->d_inode->i_mutex); |
1588 | if (unlikely(cant_mount(dentry))) { | |
1589 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1590 | return ERR_PTR(-ENOENT); | |
b12cea91 AV |
1591 | } |
1592 | down_write(&namespace_sem); | |
1593 | mnt = lookup_mnt(path); | |
84d17192 AV |
1594 | if (likely(!mnt)) { |
1595 | struct mountpoint *mp = new_mountpoint(dentry); | |
1596 | if (IS_ERR(mp)) { | |
1597 | up_write(&namespace_sem); | |
1598 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1599 | return mp; | |
1600 | } | |
1601 | return mp; | |
1602 | } | |
b12cea91 AV |
1603 | up_write(&namespace_sem); |
1604 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1605 | path_put(path); | |
1606 | path->mnt = mnt; | |
84d17192 | 1607 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1608 | goto retry; |
1609 | } | |
1610 | ||
84d17192 | 1611 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1612 | { |
84d17192 AV |
1613 | struct dentry *dentry = where->m_dentry; |
1614 | put_mountpoint(where); | |
b12cea91 | 1615 | up_write(&namespace_sem); |
84d17192 | 1616 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1617 | } |
1618 | ||
84d17192 | 1619 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1620 | { |
95bc5f25 | 1621 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1622 | return -EINVAL; |
1623 | ||
84d17192 | 1624 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1625 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1626 | return -ENOTDIR; |
1627 | ||
84d17192 | 1628 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1629 | } |
1630 | ||
7a2e8a8f VA |
1631 | /* |
1632 | * Sanity check the flags to change_mnt_propagation. | |
1633 | */ | |
1634 | ||
1635 | static int flags_to_propagation_type(int flags) | |
1636 | { | |
7c6e984d | 1637 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1638 | |
1639 | /* Fail if any non-propagation flags are set */ | |
1640 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1641 | return 0; | |
1642 | /* Only one propagation flag should be set */ | |
1643 | if (!is_power_of_2(type)) | |
1644 | return 0; | |
1645 | return type; | |
1646 | } | |
1647 | ||
07b20889 RP |
1648 | /* |
1649 | * recursively change the type of the mountpoint. | |
1650 | */ | |
0a0d8a46 | 1651 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1652 | { |
315fc83e | 1653 | struct mount *m; |
4b8b21f4 | 1654 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1655 | int recurse = flag & MS_REC; |
7a2e8a8f | 1656 | int type; |
719f5d7f | 1657 | int err = 0; |
07b20889 | 1658 | |
2d92ab3c | 1659 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1660 | return -EINVAL; |
1661 | ||
7a2e8a8f VA |
1662 | type = flags_to_propagation_type(flag); |
1663 | if (!type) | |
1664 | return -EINVAL; | |
1665 | ||
07b20889 | 1666 | down_write(&namespace_sem); |
719f5d7f MS |
1667 | if (type == MS_SHARED) { |
1668 | err = invent_group_ids(mnt, recurse); | |
1669 | if (err) | |
1670 | goto out_unlock; | |
1671 | } | |
1672 | ||
962830df | 1673 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1674 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1675 | change_mnt_propagation(m, type); |
962830df | 1676 | br_write_unlock(&vfsmount_lock); |
719f5d7f MS |
1677 | |
1678 | out_unlock: | |
07b20889 | 1679 | up_write(&namespace_sem); |
719f5d7f | 1680 | return err; |
07b20889 RP |
1681 | } |
1682 | ||
1da177e4 LT |
1683 | /* |
1684 | * do loopback mount. | |
1685 | */ | |
808d4e3c | 1686 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1687 | int recurse) |
1da177e4 | 1688 | { |
b12cea91 | 1689 | LIST_HEAD(umount_list); |
2d92ab3c | 1690 | struct path old_path; |
84d17192 AV |
1691 | struct mount *mnt = NULL, *old, *parent; |
1692 | struct mountpoint *mp; | |
57eccb83 | 1693 | int err; |
1da177e4 LT |
1694 | if (!old_name || !*old_name) |
1695 | return -EINVAL; | |
815d405c | 1696 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1697 | if (err) |
1698 | return err; | |
1699 | ||
8823c079 EB |
1700 | err = -EINVAL; |
1701 | if (mnt_ns_loop(&old_path)) | |
1702 | goto out; | |
1703 | ||
84d17192 AV |
1704 | mp = lock_mount(path); |
1705 | err = PTR_ERR(mp); | |
1706 | if (IS_ERR(mp)) | |
b12cea91 AV |
1707 | goto out; |
1708 | ||
87129cc0 | 1709 | old = real_mount(old_path.mnt); |
84d17192 | 1710 | parent = real_mount(path->mnt); |
87129cc0 | 1711 | |
1da177e4 | 1712 | err = -EINVAL; |
fc7be130 | 1713 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1714 | goto out2; |
9676f0c6 | 1715 | |
84d17192 | 1716 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1717 | goto out2; |
1da177e4 | 1718 | |
ccd48bc7 | 1719 | if (recurse) |
87129cc0 | 1720 | mnt = copy_tree(old, old_path.dentry, 0); |
ccd48bc7 | 1721 | else |
87129cc0 | 1722 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1723 | |
be34d1a3 DH |
1724 | if (IS_ERR(mnt)) { |
1725 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1726 | goto out2; |
be34d1a3 | 1727 | } |
ccd48bc7 | 1728 | |
84d17192 | 1729 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1730 | if (err) { |
962830df | 1731 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1732 | umount_tree(mnt, 0, &umount_list); |
962830df | 1733 | br_write_unlock(&vfsmount_lock); |
5b83d2c5 | 1734 | } |
b12cea91 | 1735 | out2: |
84d17192 | 1736 | unlock_mount(mp); |
b12cea91 | 1737 | release_mounts(&umount_list); |
ccd48bc7 | 1738 | out: |
2d92ab3c | 1739 | path_put(&old_path); |
1da177e4 LT |
1740 | return err; |
1741 | } | |
1742 | ||
2e4b7fcd DH |
1743 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1744 | { | |
1745 | int error = 0; | |
1746 | int readonly_request = 0; | |
1747 | ||
1748 | if (ms_flags & MS_RDONLY) | |
1749 | readonly_request = 1; | |
1750 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1751 | return 0; | |
1752 | ||
1753 | if (readonly_request) | |
83adc753 | 1754 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1755 | else |
83adc753 | 1756 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1757 | return error; |
1758 | } | |
1759 | ||
1da177e4 LT |
1760 | /* |
1761 | * change filesystem flags. dir should be a physical root of filesystem. | |
1762 | * If you've mounted a non-root directory somewhere and want to do remount | |
1763 | * on it - tough luck. | |
1764 | */ | |
0a0d8a46 | 1765 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1766 | void *data) |
1767 | { | |
1768 | int err; | |
2d92ab3c | 1769 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1770 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1771 | |
143c8c91 | 1772 | if (!check_mnt(mnt)) |
1da177e4 LT |
1773 | return -EINVAL; |
1774 | ||
2d92ab3c | 1775 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1776 | return -EINVAL; |
1777 | ||
ff36fe2c EP |
1778 | err = security_sb_remount(sb, data); |
1779 | if (err) | |
1780 | return err; | |
1781 | ||
1da177e4 | 1782 | down_write(&sb->s_umount); |
2e4b7fcd | 1783 | if (flags & MS_BIND) |
2d92ab3c | 1784 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1785 | else if (!capable(CAP_SYS_ADMIN)) |
1786 | err = -EPERM; | |
4aa98cf7 | 1787 | else |
2e4b7fcd | 1788 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1789 | if (!err) { |
962830df | 1790 | br_write_lock(&vfsmount_lock); |
143c8c91 AV |
1791 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1792 | mnt->mnt.mnt_flags = mnt_flags; | |
962830df | 1793 | br_write_unlock(&vfsmount_lock); |
7b43a79f | 1794 | } |
1da177e4 | 1795 | up_write(&sb->s_umount); |
0e55a7cc | 1796 | if (!err) { |
962830df | 1797 | br_write_lock(&vfsmount_lock); |
143c8c91 | 1798 | touch_mnt_namespace(mnt->mnt_ns); |
962830df | 1799 | br_write_unlock(&vfsmount_lock); |
0e55a7cc | 1800 | } |
1da177e4 LT |
1801 | return err; |
1802 | } | |
1803 | ||
cbbe362c | 1804 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1805 | { |
315fc83e | 1806 | struct mount *p; |
909b0a88 | 1807 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1808 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1809 | return 1; |
1810 | } | |
1811 | return 0; | |
1812 | } | |
1813 | ||
808d4e3c | 1814 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1815 | { |
2d92ab3c | 1816 | struct path old_path, parent_path; |
676da58d | 1817 | struct mount *p; |
0fb54e50 | 1818 | struct mount *old; |
84d17192 | 1819 | struct mountpoint *mp; |
57eccb83 | 1820 | int err; |
1da177e4 LT |
1821 | if (!old_name || !*old_name) |
1822 | return -EINVAL; | |
2d92ab3c | 1823 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1824 | if (err) |
1825 | return err; | |
1826 | ||
84d17192 AV |
1827 | mp = lock_mount(path); |
1828 | err = PTR_ERR(mp); | |
1829 | if (IS_ERR(mp)) | |
cc53ce53 DH |
1830 | goto out; |
1831 | ||
143c8c91 | 1832 | old = real_mount(old_path.mnt); |
fc7be130 | 1833 | p = real_mount(path->mnt); |
143c8c91 | 1834 | |
1da177e4 | 1835 | err = -EINVAL; |
fc7be130 | 1836 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1837 | goto out1; |
1838 | ||
1da177e4 | 1839 | err = -EINVAL; |
2d92ab3c | 1840 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1841 | goto out1; |
1da177e4 | 1842 | |
676da58d | 1843 | if (!mnt_has_parent(old)) |
21444403 | 1844 | goto out1; |
1da177e4 | 1845 | |
2d92ab3c AV |
1846 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1847 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1848 | goto out1; |
1849 | /* | |
1850 | * Don't move a mount residing in a shared parent. | |
1851 | */ | |
fc7be130 | 1852 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1853 | goto out1; |
9676f0c6 RP |
1854 | /* |
1855 | * Don't move a mount tree containing unbindable mounts to a destination | |
1856 | * mount which is shared. | |
1857 | */ | |
fc7be130 | 1858 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1859 | goto out1; |
1da177e4 | 1860 | err = -ELOOP; |
fc7be130 | 1861 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1862 | if (p == old) |
21444403 | 1863 | goto out1; |
1da177e4 | 1864 | |
84d17192 | 1865 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 1866 | if (err) |
21444403 | 1867 | goto out1; |
1da177e4 LT |
1868 | |
1869 | /* if the mount is moved, it should no longer be expire | |
1870 | * automatically */ | |
6776db3d | 1871 | list_del_init(&old->mnt_expire); |
1da177e4 | 1872 | out1: |
84d17192 | 1873 | unlock_mount(mp); |
1da177e4 | 1874 | out: |
1da177e4 | 1875 | if (!err) |
1a390689 | 1876 | path_put(&parent_path); |
2d92ab3c | 1877 | path_put(&old_path); |
1da177e4 LT |
1878 | return err; |
1879 | } | |
1880 | ||
9d412a43 AV |
1881 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1882 | { | |
1883 | int err; | |
1884 | const char *subtype = strchr(fstype, '.'); | |
1885 | if (subtype) { | |
1886 | subtype++; | |
1887 | err = -EINVAL; | |
1888 | if (!subtype[0]) | |
1889 | goto err; | |
1890 | } else | |
1891 | subtype = ""; | |
1892 | ||
1893 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1894 | err = -ENOMEM; | |
1895 | if (!mnt->mnt_sb->s_subtype) | |
1896 | goto err; | |
1897 | return mnt; | |
1898 | ||
1899 | err: | |
1900 | mntput(mnt); | |
1901 | return ERR_PTR(err); | |
1902 | } | |
1903 | ||
9d412a43 AV |
1904 | /* |
1905 | * add a mount into a namespace's mount tree | |
1906 | */ | |
95bc5f25 | 1907 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 1908 | { |
84d17192 AV |
1909 | struct mountpoint *mp; |
1910 | struct mount *parent; | |
9d412a43 AV |
1911 | int err; |
1912 | ||
1913 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1914 | ||
84d17192 AV |
1915 | mp = lock_mount(path); |
1916 | if (IS_ERR(mp)) | |
1917 | return PTR_ERR(mp); | |
9d412a43 | 1918 | |
84d17192 | 1919 | parent = real_mount(path->mnt); |
9d412a43 | 1920 | err = -EINVAL; |
84d17192 | 1921 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
1922 | /* that's acceptable only for automounts done in private ns */ |
1923 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
1924 | goto unlock; | |
1925 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 1926 | if (!parent->mnt_ns) |
156cacb1 AV |
1927 | goto unlock; |
1928 | } | |
9d412a43 AV |
1929 | |
1930 | /* Refuse the same filesystem on the same mount point */ | |
1931 | err = -EBUSY; | |
95bc5f25 | 1932 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1933 | path->mnt->mnt_root == path->dentry) |
1934 | goto unlock; | |
1935 | ||
1936 | err = -EINVAL; | |
95bc5f25 | 1937 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
1938 | goto unlock; |
1939 | ||
95bc5f25 | 1940 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 1941 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
1942 | |
1943 | unlock: | |
84d17192 | 1944 | unlock_mount(mp); |
9d412a43 AV |
1945 | return err; |
1946 | } | |
b1e75df4 | 1947 | |
1da177e4 LT |
1948 | /* |
1949 | * create a new mount for userspace and request it to be added into the | |
1950 | * namespace's tree | |
1951 | */ | |
0c55cfc4 | 1952 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 1953 | int mnt_flags, const char *name, void *data) |
1da177e4 | 1954 | { |
0c55cfc4 | 1955 | struct file_system_type *type; |
9b40bc90 | 1956 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 1957 | struct vfsmount *mnt; |
15f9a3f3 | 1958 | int err; |
1da177e4 | 1959 | |
0c55cfc4 | 1960 | if (!fstype) |
1da177e4 LT |
1961 | return -EINVAL; |
1962 | ||
0c55cfc4 EB |
1963 | type = get_fs_type(fstype); |
1964 | if (!type) | |
1965 | return -ENODEV; | |
1966 | ||
1967 | if (user_ns != &init_user_ns) { | |
1968 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
1969 | put_filesystem(type); | |
1970 | return -EPERM; | |
1971 | } | |
1972 | /* Only in special cases allow devices from mounts | |
1973 | * created outside the initial user namespace. | |
1974 | */ | |
1975 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
1976 | flags |= MS_NODEV; | |
1977 | mnt_flags |= MNT_NODEV; | |
1978 | } | |
1979 | } | |
1980 | ||
1981 | mnt = vfs_kern_mount(type, flags, name, data); | |
1982 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1983 | !mnt->mnt_sb->s_subtype) | |
1984 | mnt = fs_set_subtype(mnt, fstype); | |
1985 | ||
1986 | put_filesystem(type); | |
1da177e4 LT |
1987 | if (IS_ERR(mnt)) |
1988 | return PTR_ERR(mnt); | |
1989 | ||
95bc5f25 | 1990 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
1991 | if (err) |
1992 | mntput(mnt); | |
1993 | return err; | |
1da177e4 LT |
1994 | } |
1995 | ||
19a167af AV |
1996 | int finish_automount(struct vfsmount *m, struct path *path) |
1997 | { | |
6776db3d | 1998 | struct mount *mnt = real_mount(m); |
19a167af AV |
1999 | int err; |
2000 | /* The new mount record should have at least 2 refs to prevent it being | |
2001 | * expired before we get a chance to add it | |
2002 | */ | |
6776db3d | 2003 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2004 | |
2005 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2006 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2007 | err = -ELOOP; |
2008 | goto fail; | |
19a167af AV |
2009 | } |
2010 | ||
95bc5f25 | 2011 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2012 | if (!err) |
2013 | return 0; | |
2014 | fail: | |
2015 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2016 | if (!list_empty(&mnt->mnt_expire)) { |
b1e75df4 | 2017 | down_write(&namespace_sem); |
962830df | 2018 | br_write_lock(&vfsmount_lock); |
6776db3d | 2019 | list_del_init(&mnt->mnt_expire); |
962830df | 2020 | br_write_unlock(&vfsmount_lock); |
b1e75df4 | 2021 | up_write(&namespace_sem); |
19a167af | 2022 | } |
b1e75df4 AV |
2023 | mntput(m); |
2024 | mntput(m); | |
19a167af AV |
2025 | return err; |
2026 | } | |
2027 | ||
ea5b778a DH |
2028 | /** |
2029 | * mnt_set_expiry - Put a mount on an expiration list | |
2030 | * @mnt: The mount to list. | |
2031 | * @expiry_list: The list to add the mount to. | |
2032 | */ | |
2033 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2034 | { | |
2035 | down_write(&namespace_sem); | |
962830df | 2036 | br_write_lock(&vfsmount_lock); |
ea5b778a | 2037 | |
6776db3d | 2038 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2039 | |
962830df | 2040 | br_write_unlock(&vfsmount_lock); |
ea5b778a DH |
2041 | up_write(&namespace_sem); |
2042 | } | |
2043 | EXPORT_SYMBOL(mnt_set_expiry); | |
2044 | ||
1da177e4 LT |
2045 | /* |
2046 | * process a list of expirable mountpoints with the intent of discarding any | |
2047 | * mountpoints that aren't in use and haven't been touched since last we came | |
2048 | * here | |
2049 | */ | |
2050 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2051 | { | |
761d5c38 | 2052 | struct mount *mnt, *next; |
1da177e4 | 2053 | LIST_HEAD(graveyard); |
bcc5c7d2 | 2054 | LIST_HEAD(umounts); |
1da177e4 LT |
2055 | |
2056 | if (list_empty(mounts)) | |
2057 | return; | |
2058 | ||
bcc5c7d2 | 2059 | down_write(&namespace_sem); |
962830df | 2060 | br_write_lock(&vfsmount_lock); |
1da177e4 LT |
2061 | |
2062 | /* extract from the expiration list every vfsmount that matches the | |
2063 | * following criteria: | |
2064 | * - only referenced by its parent vfsmount | |
2065 | * - still marked for expiry (marked on the last call here; marks are | |
2066 | * cleared by mntput()) | |
2067 | */ | |
6776db3d | 2068 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2069 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2070 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2071 | continue; |
6776db3d | 2072 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2073 | } |
bcc5c7d2 | 2074 | while (!list_empty(&graveyard)) { |
6776db3d | 2075 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2076 | touch_mnt_namespace(mnt->mnt_ns); |
bcc5c7d2 AV |
2077 | umount_tree(mnt, 1, &umounts); |
2078 | } | |
962830df | 2079 | br_write_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
2080 | up_write(&namespace_sem); |
2081 | ||
2082 | release_mounts(&umounts); | |
5528f911 TM |
2083 | } |
2084 | ||
2085 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2086 | ||
2087 | /* | |
2088 | * Ripoff of 'select_parent()' | |
2089 | * | |
2090 | * search the list of submounts for a given mountpoint, and move any | |
2091 | * shrinkable submounts to the 'graveyard' list. | |
2092 | */ | |
692afc31 | 2093 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2094 | { |
692afc31 | 2095 | struct mount *this_parent = parent; |
5528f911 TM |
2096 | struct list_head *next; |
2097 | int found = 0; | |
2098 | ||
2099 | repeat: | |
6b41d536 | 2100 | next = this_parent->mnt_mounts.next; |
5528f911 | 2101 | resume: |
6b41d536 | 2102 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2103 | struct list_head *tmp = next; |
6b41d536 | 2104 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2105 | |
2106 | next = tmp->next; | |
692afc31 | 2107 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2108 | continue; |
5528f911 TM |
2109 | /* |
2110 | * Descend a level if the d_mounts list is non-empty. | |
2111 | */ | |
6b41d536 | 2112 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2113 | this_parent = mnt; |
2114 | goto repeat; | |
2115 | } | |
1da177e4 | 2116 | |
1ab59738 | 2117 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2118 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2119 | found++; |
2120 | } | |
1da177e4 | 2121 | } |
5528f911 TM |
2122 | /* |
2123 | * All done at this level ... ascend and resume the search | |
2124 | */ | |
2125 | if (this_parent != parent) { | |
6b41d536 | 2126 | next = this_parent->mnt_child.next; |
0714a533 | 2127 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2128 | goto resume; |
2129 | } | |
2130 | return found; | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * process a list of expirable mountpoints with the intent of discarding any | |
2135 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
2136 | * |
2137 | * vfsmount_lock must be held for write | |
5528f911 | 2138 | */ |
692afc31 | 2139 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts) |
5528f911 TM |
2140 | { |
2141 | LIST_HEAD(graveyard); | |
761d5c38 | 2142 | struct mount *m; |
5528f911 | 2143 | |
5528f911 | 2144 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2145 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2146 | while (!list_empty(&graveyard)) { |
761d5c38 | 2147 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2148 | mnt_expire); |
143c8c91 | 2149 | touch_mnt_namespace(m->mnt_ns); |
afef80b3 | 2150 | umount_tree(m, 1, umounts); |
bcc5c7d2 AV |
2151 | } |
2152 | } | |
1da177e4 LT |
2153 | } |
2154 | ||
1da177e4 LT |
2155 | /* |
2156 | * Some copy_from_user() implementations do not return the exact number of | |
2157 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2158 | * Note that this function differs from copy_from_user() in that it will oops | |
2159 | * on bad values of `to', rather than returning a short copy. | |
2160 | */ | |
b58fed8b RP |
2161 | static long exact_copy_from_user(void *to, const void __user * from, |
2162 | unsigned long n) | |
1da177e4 LT |
2163 | { |
2164 | char *t = to; | |
2165 | const char __user *f = from; | |
2166 | char c; | |
2167 | ||
2168 | if (!access_ok(VERIFY_READ, from, n)) | |
2169 | return n; | |
2170 | ||
2171 | while (n) { | |
2172 | if (__get_user(c, f)) { | |
2173 | memset(t, 0, n); | |
2174 | break; | |
2175 | } | |
2176 | *t++ = c; | |
2177 | f++; | |
2178 | n--; | |
2179 | } | |
2180 | return n; | |
2181 | } | |
2182 | ||
b58fed8b | 2183 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2184 | { |
2185 | int i; | |
2186 | unsigned long page; | |
2187 | unsigned long size; | |
b58fed8b | 2188 | |
1da177e4 LT |
2189 | *where = 0; |
2190 | if (!data) | |
2191 | return 0; | |
2192 | ||
2193 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2194 | return -ENOMEM; | |
2195 | ||
2196 | /* We only care that *some* data at the address the user | |
2197 | * gave us is valid. Just in case, we'll zero | |
2198 | * the remainder of the page. | |
2199 | */ | |
2200 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2201 | size = TASK_SIZE - (unsigned long)data; | |
2202 | if (size > PAGE_SIZE) | |
2203 | size = PAGE_SIZE; | |
2204 | ||
2205 | i = size - exact_copy_from_user((void *)page, data, size); | |
2206 | if (!i) { | |
b58fed8b | 2207 | free_page(page); |
1da177e4 LT |
2208 | return -EFAULT; |
2209 | } | |
2210 | if (i != PAGE_SIZE) | |
2211 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2212 | *where = page; | |
2213 | return 0; | |
2214 | } | |
2215 | ||
eca6f534 VN |
2216 | int copy_mount_string(const void __user *data, char **where) |
2217 | { | |
2218 | char *tmp; | |
2219 | ||
2220 | if (!data) { | |
2221 | *where = NULL; | |
2222 | return 0; | |
2223 | } | |
2224 | ||
2225 | tmp = strndup_user(data, PAGE_SIZE); | |
2226 | if (IS_ERR(tmp)) | |
2227 | return PTR_ERR(tmp); | |
2228 | ||
2229 | *where = tmp; | |
2230 | return 0; | |
2231 | } | |
2232 | ||
1da177e4 LT |
2233 | /* |
2234 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2235 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2236 | * | |
2237 | * data is a (void *) that can point to any structure up to | |
2238 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2239 | * information (or be NULL). | |
2240 | * | |
2241 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2242 | * When the flags word was introduced its top half was required | |
2243 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2244 | * Therefore, if this magic number is present, it carries no information | |
2245 | * and must be discarded. | |
2246 | */ | |
808d4e3c AV |
2247 | long do_mount(const char *dev_name, const char *dir_name, |
2248 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2249 | { |
2d92ab3c | 2250 | struct path path; |
1da177e4 LT |
2251 | int retval = 0; |
2252 | int mnt_flags = 0; | |
2253 | ||
2254 | /* Discard magic */ | |
2255 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2256 | flags &= ~MS_MGC_MSK; | |
2257 | ||
2258 | /* Basic sanity checks */ | |
2259 | ||
2260 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2261 | return -EINVAL; | |
1da177e4 LT |
2262 | |
2263 | if (data_page) | |
2264 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2265 | ||
a27ab9f2 TH |
2266 | /* ... and get the mountpoint */ |
2267 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2268 | if (retval) | |
2269 | return retval; | |
2270 | ||
2271 | retval = security_sb_mount(dev_name, &path, | |
2272 | type_page, flags, data_page); | |
2273 | if (retval) | |
2274 | goto dput_out; | |
2275 | ||
57eccb83 AV |
2276 | if (!may_mount()) |
2277 | return -EPERM; | |
2278 | ||
613cbe3d AK |
2279 | /* Default to relatime unless overriden */ |
2280 | if (!(flags & MS_NOATIME)) | |
2281 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2282 | |
1da177e4 LT |
2283 | /* Separate the per-mountpoint flags */ |
2284 | if (flags & MS_NOSUID) | |
2285 | mnt_flags |= MNT_NOSUID; | |
2286 | if (flags & MS_NODEV) | |
2287 | mnt_flags |= MNT_NODEV; | |
2288 | if (flags & MS_NOEXEC) | |
2289 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2290 | if (flags & MS_NOATIME) |
2291 | mnt_flags |= MNT_NOATIME; | |
2292 | if (flags & MS_NODIRATIME) | |
2293 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2294 | if (flags & MS_STRICTATIME) |
2295 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2296 | if (flags & MS_RDONLY) |
2297 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2298 | |
7a4dec53 | 2299 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2300 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2301 | MS_STRICTATIME); | |
1da177e4 | 2302 | |
1da177e4 | 2303 | if (flags & MS_REMOUNT) |
2d92ab3c | 2304 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2305 | data_page); |
2306 | else if (flags & MS_BIND) | |
2d92ab3c | 2307 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2308 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2309 | retval = do_change_type(&path, flags); |
1da177e4 | 2310 | else if (flags & MS_MOVE) |
2d92ab3c | 2311 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2312 | else |
2d92ab3c | 2313 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2314 | dev_name, data_page); |
2315 | dput_out: | |
2d92ab3c | 2316 | path_put(&path); |
1da177e4 LT |
2317 | return retval; |
2318 | } | |
2319 | ||
771b1371 EB |
2320 | static void free_mnt_ns(struct mnt_namespace *ns) |
2321 | { | |
98f842e6 | 2322 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2323 | put_user_ns(ns->user_ns); |
2324 | kfree(ns); | |
2325 | } | |
2326 | ||
8823c079 EB |
2327 | /* |
2328 | * Assign a sequence number so we can detect when we attempt to bind | |
2329 | * mount a reference to an older mount namespace into the current | |
2330 | * mount namespace, preventing reference counting loops. A 64bit | |
2331 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2332 | * is effectively never, so we can ignore the possibility. | |
2333 | */ | |
2334 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2335 | ||
771b1371 | 2336 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2337 | { |
2338 | struct mnt_namespace *new_ns; | |
98f842e6 | 2339 | int ret; |
cf8d2c11 TM |
2340 | |
2341 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2342 | if (!new_ns) | |
2343 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2344 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2345 | if (ret) { | |
2346 | kfree(new_ns); | |
2347 | return ERR_PTR(ret); | |
2348 | } | |
8823c079 | 2349 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2350 | atomic_set(&new_ns->count, 1); |
2351 | new_ns->root = NULL; | |
2352 | INIT_LIST_HEAD(&new_ns->list); | |
2353 | init_waitqueue_head(&new_ns->poll); | |
2354 | new_ns->event = 0; | |
771b1371 | 2355 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2356 | return new_ns; |
2357 | } | |
2358 | ||
741a2951 JD |
2359 | /* |
2360 | * Allocate a new namespace structure and populate it with contents | |
2361 | * copied from the namespace of the passed in task structure. | |
2362 | */ | |
e3222c4e | 2363 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
771b1371 | 2364 | struct user_namespace *user_ns, struct fs_struct *fs) |
1da177e4 | 2365 | { |
6b3286ed | 2366 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2367 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2368 | struct mount *p, *q; |
be08d6d2 | 2369 | struct mount *old = mnt_ns->root; |
cb338d06 | 2370 | struct mount *new; |
7a472ef4 | 2371 | int copy_flags; |
1da177e4 | 2372 | |
771b1371 | 2373 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2374 | if (IS_ERR(new_ns)) |
2375 | return new_ns; | |
1da177e4 | 2376 | |
390c6843 | 2377 | down_write(&namespace_sem); |
1da177e4 | 2378 | /* First pass: copy the tree topology */ |
7a472ef4 EB |
2379 | copy_flags = CL_COPY_ALL | CL_EXPIRE; |
2380 | if (user_ns != mnt_ns->user_ns) | |
2381 | copy_flags |= CL_SHARED_TO_SLAVE; | |
2382 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); | |
be34d1a3 | 2383 | if (IS_ERR(new)) { |
390c6843 | 2384 | up_write(&namespace_sem); |
771b1371 | 2385 | free_mnt_ns(new_ns); |
be34d1a3 | 2386 | return ERR_CAST(new); |
1da177e4 | 2387 | } |
be08d6d2 | 2388 | new_ns->root = new; |
962830df | 2389 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 2390 | list_add_tail(&new_ns->list, &new->mnt_list); |
962830df | 2391 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
2392 | |
2393 | /* | |
2394 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2395 | * as belonging to new namespace. We have already acquired a private | |
2396 | * fs_struct, so tsk->fs->lock is not needed. | |
2397 | */ | |
909b0a88 | 2398 | p = old; |
cb338d06 | 2399 | q = new; |
1da177e4 | 2400 | while (p) { |
143c8c91 | 2401 | q->mnt_ns = new_ns; |
1da177e4 | 2402 | if (fs) { |
315fc83e AV |
2403 | if (&p->mnt == fs->root.mnt) { |
2404 | fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2405 | rootmnt = &p->mnt; |
1da177e4 | 2406 | } |
315fc83e AV |
2407 | if (&p->mnt == fs->pwd.mnt) { |
2408 | fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2409 | pwdmnt = &p->mnt; |
1da177e4 | 2410 | } |
1da177e4 | 2411 | } |
909b0a88 AV |
2412 | p = next_mnt(p, old); |
2413 | q = next_mnt(q, new); | |
1da177e4 | 2414 | } |
390c6843 | 2415 | up_write(&namespace_sem); |
1da177e4 | 2416 | |
1da177e4 | 2417 | if (rootmnt) |
f03c6599 | 2418 | mntput(rootmnt); |
1da177e4 | 2419 | if (pwdmnt) |
f03c6599 | 2420 | mntput(pwdmnt); |
1da177e4 | 2421 | |
741a2951 JD |
2422 | return new_ns; |
2423 | } | |
2424 | ||
213dd266 | 2425 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
771b1371 | 2426 | struct user_namespace *user_ns, struct fs_struct *new_fs) |
741a2951 | 2427 | { |
6b3286ed | 2428 | struct mnt_namespace *new_ns; |
741a2951 | 2429 | |
e3222c4e | 2430 | BUG_ON(!ns); |
6b3286ed | 2431 | get_mnt_ns(ns); |
741a2951 JD |
2432 | |
2433 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2434 | return ns; |
741a2951 | 2435 | |
771b1371 | 2436 | new_ns = dup_mnt_ns(ns, user_ns, new_fs); |
741a2951 | 2437 | |
6b3286ed | 2438 | put_mnt_ns(ns); |
e3222c4e | 2439 | return new_ns; |
1da177e4 LT |
2440 | } |
2441 | ||
cf8d2c11 TM |
2442 | /** |
2443 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2444 | * @mnt: pointer to the new root filesystem mountpoint | |
2445 | */ | |
1a4eeaf2 | 2446 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2447 | { |
771b1371 | 2448 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2449 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2450 | struct mount *mnt = real_mount(m); |
2451 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2452 | new_ns->root = mnt; |
1a4eeaf2 | 2453 | list_add(&new_ns->list, &mnt->mnt_list); |
c1334495 | 2454 | } else { |
1a4eeaf2 | 2455 | mntput(m); |
cf8d2c11 TM |
2456 | } |
2457 | return new_ns; | |
2458 | } | |
cf8d2c11 | 2459 | |
ea441d11 AV |
2460 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2461 | { | |
2462 | struct mnt_namespace *ns; | |
d31da0f0 | 2463 | struct super_block *s; |
ea441d11 AV |
2464 | struct path path; |
2465 | int err; | |
2466 | ||
2467 | ns = create_mnt_ns(mnt); | |
2468 | if (IS_ERR(ns)) | |
2469 | return ERR_CAST(ns); | |
2470 | ||
2471 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2472 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2473 | ||
2474 | put_mnt_ns(ns); | |
2475 | ||
2476 | if (err) | |
2477 | return ERR_PTR(err); | |
2478 | ||
2479 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2480 | s = path.mnt->mnt_sb; |
2481 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2482 | mntput(path.mnt); |
2483 | /* lock the sucker */ | |
d31da0f0 | 2484 | down_write(&s->s_umount); |
ea441d11 AV |
2485 | /* ... and return the root of (sub)tree on it */ |
2486 | return path.dentry; | |
2487 | } | |
2488 | EXPORT_SYMBOL(mount_subtree); | |
2489 | ||
bdc480e3 HC |
2490 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2491 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2492 | { |
eca6f534 VN |
2493 | int ret; |
2494 | char *kernel_type; | |
91a27b2a | 2495 | struct filename *kernel_dir; |
eca6f534 | 2496 | char *kernel_dev; |
1da177e4 | 2497 | unsigned long data_page; |
1da177e4 | 2498 | |
eca6f534 VN |
2499 | ret = copy_mount_string(type, &kernel_type); |
2500 | if (ret < 0) | |
2501 | goto out_type; | |
1da177e4 | 2502 | |
eca6f534 VN |
2503 | kernel_dir = getname(dir_name); |
2504 | if (IS_ERR(kernel_dir)) { | |
2505 | ret = PTR_ERR(kernel_dir); | |
2506 | goto out_dir; | |
2507 | } | |
1da177e4 | 2508 | |
eca6f534 VN |
2509 | ret = copy_mount_string(dev_name, &kernel_dev); |
2510 | if (ret < 0) | |
2511 | goto out_dev; | |
1da177e4 | 2512 | |
eca6f534 VN |
2513 | ret = copy_mount_options(data, &data_page); |
2514 | if (ret < 0) | |
2515 | goto out_data; | |
1da177e4 | 2516 | |
91a27b2a | 2517 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2518 | (void *) data_page); |
1da177e4 | 2519 | |
eca6f534 VN |
2520 | free_page(data_page); |
2521 | out_data: | |
2522 | kfree(kernel_dev); | |
2523 | out_dev: | |
2524 | putname(kernel_dir); | |
2525 | out_dir: | |
2526 | kfree(kernel_type); | |
2527 | out_type: | |
2528 | return ret; | |
1da177e4 LT |
2529 | } |
2530 | ||
afac7cba AV |
2531 | /* |
2532 | * Return true if path is reachable from root | |
2533 | * | |
2534 | * namespace_sem or vfsmount_lock is held | |
2535 | */ | |
643822b4 | 2536 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2537 | const struct path *root) |
2538 | { | |
643822b4 | 2539 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2540 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2541 | mnt = mnt->mnt_parent; |
afac7cba | 2542 | } |
643822b4 | 2543 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2544 | } |
2545 | ||
2546 | int path_is_under(struct path *path1, struct path *path2) | |
2547 | { | |
2548 | int res; | |
962830df | 2549 | br_read_lock(&vfsmount_lock); |
643822b4 | 2550 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
962830df | 2551 | br_read_unlock(&vfsmount_lock); |
afac7cba AV |
2552 | return res; |
2553 | } | |
2554 | EXPORT_SYMBOL(path_is_under); | |
2555 | ||
1da177e4 LT |
2556 | /* |
2557 | * pivot_root Semantics: | |
2558 | * Moves the root file system of the current process to the directory put_old, | |
2559 | * makes new_root as the new root file system of the current process, and sets | |
2560 | * root/cwd of all processes which had them on the current root to new_root. | |
2561 | * | |
2562 | * Restrictions: | |
2563 | * The new_root and put_old must be directories, and must not be on the | |
2564 | * same file system as the current process root. The put_old must be | |
2565 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2566 | * pointed to by put_old must yield the same directory as new_root. No other | |
2567 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2568 | * | |
4a0d11fa NB |
2569 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2570 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2571 | * in this situation. | |
2572 | * | |
1da177e4 LT |
2573 | * Notes: |
2574 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2575 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2576 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2577 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2578 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2579 | * first. | |
2580 | */ | |
3480b257 HC |
2581 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2582 | const char __user *, put_old) | |
1da177e4 | 2583 | { |
2d8f3038 | 2584 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2585 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2586 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2587 | int error; |
2588 | ||
9b40bc90 | 2589 | if (!may_mount()) |
1da177e4 LT |
2590 | return -EPERM; |
2591 | ||
2d8f3038 | 2592 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2593 | if (error) |
2594 | goto out0; | |
1da177e4 | 2595 | |
2d8f3038 | 2596 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2597 | if (error) |
2598 | goto out1; | |
2599 | ||
2d8f3038 | 2600 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2601 | if (error) |
2602 | goto out2; | |
1da177e4 | 2603 | |
f7ad3c6b | 2604 | get_fs_root(current->fs, &root); |
84d17192 AV |
2605 | old_mp = lock_mount(&old); |
2606 | error = PTR_ERR(old_mp); | |
2607 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2608 | goto out3; |
2609 | ||
1da177e4 | 2610 | error = -EINVAL; |
419148da AV |
2611 | new_mnt = real_mount(new.mnt); |
2612 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2613 | old_mnt = real_mount(old.mnt); |
2614 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2615 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2616 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2617 | goto out4; |
143c8c91 | 2618 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2619 | goto out4; |
1da177e4 | 2620 | error = -ENOENT; |
f3da392e | 2621 | if (d_unlinked(new.dentry)) |
b12cea91 | 2622 | goto out4; |
1da177e4 | 2623 | error = -EBUSY; |
84d17192 | 2624 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2625 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2626 | error = -EINVAL; |
8c3ee42e | 2627 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2628 | goto out4; /* not a mountpoint */ |
676da58d | 2629 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2630 | goto out4; /* not attached */ |
84d17192 | 2631 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2632 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2633 | goto out4; /* not a mountpoint */ |
676da58d | 2634 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2635 | goto out4; /* not attached */ |
4ac91378 | 2636 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2637 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2638 | goto out4; |
84d17192 | 2639 | root_mp->m_count++; /* pin it so it won't go away */ |
962830df | 2640 | br_write_lock(&vfsmount_lock); |
419148da AV |
2641 | detach_mnt(new_mnt, &parent_path); |
2642 | detach_mnt(root_mnt, &root_parent); | |
4ac91378 | 2643 | /* mount old root on put_old */ |
84d17192 | 2644 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2645 | /* mount new_root on / */ |
84d17192 | 2646 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2647 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
962830df | 2648 | br_write_unlock(&vfsmount_lock); |
2d8f3038 | 2649 | chroot_fs_refs(&root, &new); |
84d17192 | 2650 | put_mountpoint(root_mp); |
1da177e4 | 2651 | error = 0; |
b12cea91 | 2652 | out4: |
84d17192 | 2653 | unlock_mount(old_mp); |
b12cea91 AV |
2654 | if (!error) { |
2655 | path_put(&root_parent); | |
2656 | path_put(&parent_path); | |
2657 | } | |
2658 | out3: | |
8c3ee42e | 2659 | path_put(&root); |
b12cea91 | 2660 | out2: |
2d8f3038 | 2661 | path_put(&old); |
1da177e4 | 2662 | out1: |
2d8f3038 | 2663 | path_put(&new); |
1da177e4 | 2664 | out0: |
1da177e4 | 2665 | return error; |
1da177e4 LT |
2666 | } |
2667 | ||
2668 | static void __init init_mount_tree(void) | |
2669 | { | |
2670 | struct vfsmount *mnt; | |
6b3286ed | 2671 | struct mnt_namespace *ns; |
ac748a09 | 2672 | struct path root; |
0c55cfc4 | 2673 | struct file_system_type *type; |
1da177e4 | 2674 | |
0c55cfc4 EB |
2675 | type = get_fs_type("rootfs"); |
2676 | if (!type) | |
2677 | panic("Can't find rootfs type"); | |
2678 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2679 | put_filesystem(type); | |
1da177e4 LT |
2680 | if (IS_ERR(mnt)) |
2681 | panic("Can't create rootfs"); | |
b3e19d92 | 2682 | |
3b22edc5 TM |
2683 | ns = create_mnt_ns(mnt); |
2684 | if (IS_ERR(ns)) | |
1da177e4 | 2685 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2686 | |
2687 | init_task.nsproxy->mnt_ns = ns; | |
2688 | get_mnt_ns(ns); | |
2689 | ||
be08d6d2 AV |
2690 | root.mnt = mnt; |
2691 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2692 | |
2693 | set_fs_pwd(current->fs, &root); | |
2694 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2695 | } |
2696 | ||
74bf17cf | 2697 | void __init mnt_init(void) |
1da177e4 | 2698 | { |
13f14b4d | 2699 | unsigned u; |
15a67dd8 | 2700 | int err; |
1da177e4 | 2701 | |
390c6843 RP |
2702 | init_rwsem(&namespace_sem); |
2703 | ||
7d6fec45 | 2704 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2705 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2706 | |
b58fed8b | 2707 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
84d17192 | 2708 | mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 | 2709 | |
84d17192 | 2710 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2711 | panic("Failed to allocate mount hash table\n"); |
2712 | ||
80cdc6da | 2713 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2714 | |
2715 | for (u = 0; u < HASH_SIZE; u++) | |
2716 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
84d17192 AV |
2717 | for (u = 0; u < HASH_SIZE; u++) |
2718 | INIT_LIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2719 | |
962830df | 2720 | br_lock_init(&vfsmount_lock); |
99b7db7b | 2721 | |
15a67dd8 RD |
2722 | err = sysfs_init(); |
2723 | if (err) | |
2724 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2725 | __func__, err); |
00d26666 GKH |
2726 | fs_kobj = kobject_create_and_add("fs", NULL); |
2727 | if (!fs_kobj) | |
8e24eea7 | 2728 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2729 | init_rootfs(); |
2730 | init_mount_tree(); | |
2731 | } | |
2732 | ||
616511d0 | 2733 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2734 | { |
70fbcdf4 | 2735 | LIST_HEAD(umount_list); |
616511d0 | 2736 | |
d498b25a | 2737 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2738 | return; |
390c6843 | 2739 | down_write(&namespace_sem); |
962830df | 2740 | br_write_lock(&vfsmount_lock); |
be08d6d2 | 2741 | umount_tree(ns->root, 0, &umount_list); |
962830df | 2742 | br_write_unlock(&vfsmount_lock); |
390c6843 | 2743 | up_write(&namespace_sem); |
70fbcdf4 | 2744 | release_mounts(&umount_list); |
771b1371 | 2745 | free_mnt_ns(ns); |
1da177e4 | 2746 | } |
9d412a43 AV |
2747 | |
2748 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2749 | { | |
423e0ab0 TC |
2750 | struct vfsmount *mnt; |
2751 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2752 | if (!IS_ERR(mnt)) { | |
2753 | /* | |
2754 | * it is a longterm mount, don't release mnt until | |
2755 | * we unmount before file sys is unregistered | |
2756 | */ | |
f7a99c5b | 2757 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2758 | } |
2759 | return mnt; | |
9d412a43 AV |
2760 | } |
2761 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2762 | |
2763 | void kern_unmount(struct vfsmount *mnt) | |
2764 | { | |
2765 | /* release long term mount so mount point can be released */ | |
2766 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b AV |
2767 | br_write_lock(&vfsmount_lock); |
2768 | real_mount(mnt)->mnt_ns = NULL; | |
2769 | br_write_unlock(&vfsmount_lock); | |
423e0ab0 TC |
2770 | mntput(mnt); |
2771 | } | |
2772 | } | |
2773 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2774 | |
2775 | bool our_mnt(struct vfsmount *mnt) | |
2776 | { | |
143c8c91 | 2777 | return check_mnt(real_mount(mnt)); |
02125a82 | 2778 | } |
8823c079 EB |
2779 | |
2780 | static void *mntns_get(struct task_struct *task) | |
2781 | { | |
2782 | struct mnt_namespace *ns = NULL; | |
2783 | struct nsproxy *nsproxy; | |
2784 | ||
2785 | rcu_read_lock(); | |
2786 | nsproxy = task_nsproxy(task); | |
2787 | if (nsproxy) { | |
2788 | ns = nsproxy->mnt_ns; | |
2789 | get_mnt_ns(ns); | |
2790 | } | |
2791 | rcu_read_unlock(); | |
2792 | ||
2793 | return ns; | |
2794 | } | |
2795 | ||
2796 | static void mntns_put(void *ns) | |
2797 | { | |
2798 | put_mnt_ns(ns); | |
2799 | } | |
2800 | ||
2801 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2802 | { | |
2803 | struct fs_struct *fs = current->fs; | |
2804 | struct mnt_namespace *mnt_ns = ns; | |
2805 | struct path root; | |
2806 | ||
0c55cfc4 | 2807 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
5e4a0847 EB |
2808 | !nsown_capable(CAP_SYS_CHROOT) || |
2809 | !nsown_capable(CAP_SYS_ADMIN)) | |
ae11e0f1 | 2810 | return -EPERM; |
8823c079 EB |
2811 | |
2812 | if (fs->users != 1) | |
2813 | return -EINVAL; | |
2814 | ||
2815 | get_mnt_ns(mnt_ns); | |
2816 | put_mnt_ns(nsproxy->mnt_ns); | |
2817 | nsproxy->mnt_ns = mnt_ns; | |
2818 | ||
2819 | /* Find the root */ | |
2820 | root.mnt = &mnt_ns->root->mnt; | |
2821 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2822 | path_get(&root); | |
2823 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
2824 | ; | |
2825 | ||
2826 | /* Update the pwd and root */ | |
2827 | set_fs_pwd(fs, &root); | |
2828 | set_fs_root(fs, &root); | |
2829 | ||
2830 | path_put(&root); | |
2831 | return 0; | |
2832 | } | |
2833 | ||
98f842e6 EB |
2834 | static unsigned int mntns_inum(void *ns) |
2835 | { | |
2836 | struct mnt_namespace *mnt_ns = ns; | |
2837 | return mnt_ns->proc_inum; | |
2838 | } | |
2839 | ||
8823c079 EB |
2840 | const struct proc_ns_operations mntns_operations = { |
2841 | .name = "mnt", | |
2842 | .type = CLONE_NEWNS, | |
2843 | .get = mntns_get, | |
2844 | .put = mntns_put, | |
2845 | .install = mntns_install, | |
98f842e6 | 2846 | .inum = mntns_inum, |
8823c079 | 2847 | }; |