]>
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> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
07b20889 | 26 | #include "pnode.h" |
948730b0 | 27 | #include "internal.h" |
1da177e4 | 28 | |
13f14b4d ED |
29 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
30 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
31 | ||
5addc5dd | 32 | static int event; |
73cd49ec | 33 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 34 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 35 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
36 | static int mnt_id_start = 0; |
37 | static int mnt_group_start = 1; | |
1da177e4 | 38 | |
fa3536cc | 39 | static struct list_head *mount_hashtable __read_mostly; |
84d17192 | 40 | static struct list_head *mountpoint_hashtable __read_mostly; |
e18b890b | 41 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 42 | static struct rw_semaphore namespace_sem; |
1da177e4 | 43 | |
f87fd4c2 | 44 | /* /sys/fs */ |
00d26666 GKH |
45 | struct kobject *fs_kobj; |
46 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 47 | |
99b7db7b NP |
48 | /* |
49 | * vfsmount lock may be taken for read to prevent changes to the | |
50 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
51 | * up the tree. | |
52 | * | |
53 | * It should be taken for write in all cases where the vfsmount | |
54 | * tree or hash is modified or when a vfsmount structure is modified. | |
55 | */ | |
56 | DEFINE_BRLOCK(vfsmount_lock); | |
57 | ||
1da177e4 LT |
58 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
59 | { | |
b58fed8b RP |
60 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
61 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
62 | tmp = tmp + (tmp >> HASH_SHIFT); |
63 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
64 | } |
65 | ||
3d733633 DH |
66 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
67 | ||
99b7db7b NP |
68 | /* |
69 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
70 | * serialize with freeing. | |
71 | */ | |
b105e270 | 72 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
73 | { |
74 | int res; | |
75 | ||
76 | retry: | |
77 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 78 | spin_lock(&mnt_id_lock); |
15169fe7 | 79 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 80 | if (!res) |
15169fe7 | 81 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 82 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
83 | if (res == -EAGAIN) |
84 | goto retry; | |
85 | ||
86 | return res; | |
87 | } | |
88 | ||
b105e270 | 89 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 90 | { |
15169fe7 | 91 | int id = mnt->mnt_id; |
99b7db7b | 92 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
93 | ida_remove(&mnt_id_ida, id); |
94 | if (mnt_id_start > id) | |
95 | mnt_id_start = id; | |
99b7db7b | 96 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
97 | } |
98 | ||
719f5d7f MS |
99 | /* |
100 | * Allocate a new peer group ID | |
101 | * | |
102 | * mnt_group_ida is protected by namespace_sem | |
103 | */ | |
4b8b21f4 | 104 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 105 | { |
f21f6220 AV |
106 | int res; |
107 | ||
719f5d7f MS |
108 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
109 | return -ENOMEM; | |
110 | ||
f21f6220 AV |
111 | res = ida_get_new_above(&mnt_group_ida, |
112 | mnt_group_start, | |
15169fe7 | 113 | &mnt->mnt_group_id); |
f21f6220 | 114 | if (!res) |
15169fe7 | 115 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
116 | |
117 | return res; | |
719f5d7f MS |
118 | } |
119 | ||
120 | /* | |
121 | * Release a peer group ID | |
122 | */ | |
4b8b21f4 | 123 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 124 | { |
15169fe7 | 125 | int id = mnt->mnt_group_id; |
f21f6220 AV |
126 | ida_remove(&mnt_group_ida, id); |
127 | if (mnt_group_start > id) | |
128 | mnt_group_start = id; | |
15169fe7 | 129 | mnt->mnt_group_id = 0; |
719f5d7f MS |
130 | } |
131 | ||
b3e19d92 NP |
132 | /* |
133 | * vfsmount lock must be held for read | |
134 | */ | |
83adc753 | 135 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
136 | { |
137 | #ifdef CONFIG_SMP | |
68e8a9fe | 138 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
139 | #else |
140 | preempt_disable(); | |
68e8a9fe | 141 | mnt->mnt_count += n; |
b3e19d92 NP |
142 | preempt_enable(); |
143 | #endif | |
144 | } | |
145 | ||
b3e19d92 NP |
146 | /* |
147 | * vfsmount lock must be held for write | |
148 | */ | |
83adc753 | 149 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
150 | { |
151 | #ifdef CONFIG_SMP | |
f03c6599 | 152 | unsigned int count = 0; |
b3e19d92 NP |
153 | int cpu; |
154 | ||
155 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 156 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
157 | } |
158 | ||
159 | return count; | |
160 | #else | |
68e8a9fe | 161 | return mnt->mnt_count; |
b3e19d92 NP |
162 | #endif |
163 | } | |
164 | ||
b105e270 | 165 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 166 | { |
c63181e6 AV |
167 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
168 | if (mnt) { | |
73cd49ec MS |
169 | int err; |
170 | ||
c63181e6 | 171 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
172 | if (err) |
173 | goto out_free_cache; | |
174 | ||
175 | if (name) { | |
c63181e6 AV |
176 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
177 | if (!mnt->mnt_devname) | |
88b38782 | 178 | goto out_free_id; |
73cd49ec MS |
179 | } |
180 | ||
b3e19d92 | 181 | #ifdef CONFIG_SMP |
c63181e6 AV |
182 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
183 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
184 | goto out_free_devname; |
185 | ||
c63181e6 | 186 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 187 | #else |
c63181e6 AV |
188 | mnt->mnt_count = 1; |
189 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
190 | #endif |
191 | ||
c63181e6 AV |
192 | INIT_LIST_HEAD(&mnt->mnt_hash); |
193 | INIT_LIST_HEAD(&mnt->mnt_child); | |
194 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
195 | INIT_LIST_HEAD(&mnt->mnt_list); | |
196 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
197 | INIT_LIST_HEAD(&mnt->mnt_share); | |
198 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
199 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
200 | #ifdef CONFIG_FSNOTIFY |
201 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 202 | #endif |
1da177e4 | 203 | } |
c63181e6 | 204 | return mnt; |
88b38782 | 205 | |
d3ef3d73 | 206 | #ifdef CONFIG_SMP |
207 | out_free_devname: | |
c63181e6 | 208 | kfree(mnt->mnt_devname); |
d3ef3d73 | 209 | #endif |
88b38782 | 210 | out_free_id: |
c63181e6 | 211 | mnt_free_id(mnt); |
88b38782 | 212 | out_free_cache: |
c63181e6 | 213 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 214 | return NULL; |
1da177e4 LT |
215 | } |
216 | ||
3d733633 DH |
217 | /* |
218 | * Most r/o checks on a fs are for operations that take | |
219 | * discrete amounts of time, like a write() or unlink(). | |
220 | * We must keep track of when those operations start | |
221 | * (for permission checks) and when they end, so that | |
222 | * we can determine when writes are able to occur to | |
223 | * a filesystem. | |
224 | */ | |
225 | /* | |
226 | * __mnt_is_readonly: check whether a mount is read-only | |
227 | * @mnt: the mount to check for its write status | |
228 | * | |
229 | * This shouldn't be used directly ouside of the VFS. | |
230 | * It does not guarantee that the filesystem will stay | |
231 | * r/w, just that it is right *now*. This can not and | |
232 | * should not be used in place of IS_RDONLY(inode). | |
233 | * mnt_want/drop_write() will _keep_ the filesystem | |
234 | * r/w. | |
235 | */ | |
236 | int __mnt_is_readonly(struct vfsmount *mnt) | |
237 | { | |
2e4b7fcd DH |
238 | if (mnt->mnt_flags & MNT_READONLY) |
239 | return 1; | |
240 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
241 | return 1; | |
242 | return 0; | |
3d733633 DH |
243 | } |
244 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
245 | ||
83adc753 | 246 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 247 | { |
248 | #ifdef CONFIG_SMP | |
68e8a9fe | 249 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 250 | #else |
68e8a9fe | 251 | mnt->mnt_writers++; |
d3ef3d73 | 252 | #endif |
253 | } | |
3d733633 | 254 | |
83adc753 | 255 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 256 | { |
d3ef3d73 | 257 | #ifdef CONFIG_SMP |
68e8a9fe | 258 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 259 | #else |
68e8a9fe | 260 | mnt->mnt_writers--; |
d3ef3d73 | 261 | #endif |
3d733633 | 262 | } |
3d733633 | 263 | |
83adc753 | 264 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 265 | { |
d3ef3d73 | 266 | #ifdef CONFIG_SMP |
267 | unsigned int count = 0; | |
3d733633 | 268 | int cpu; |
3d733633 DH |
269 | |
270 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 271 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 272 | } |
3d733633 | 273 | |
d3ef3d73 | 274 | return count; |
275 | #else | |
276 | return mnt->mnt_writers; | |
277 | #endif | |
3d733633 DH |
278 | } |
279 | ||
4ed5e82f MS |
280 | static int mnt_is_readonly(struct vfsmount *mnt) |
281 | { | |
282 | if (mnt->mnt_sb->s_readonly_remount) | |
283 | return 1; | |
284 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
285 | smp_rmb(); | |
286 | return __mnt_is_readonly(mnt); | |
287 | } | |
288 | ||
8366025e | 289 | /* |
eb04c282 JK |
290 | * Most r/o & frozen checks on a fs are for operations that take discrete |
291 | * amounts of time, like a write() or unlink(). We must keep track of when | |
292 | * those operations start (for permission checks) and when they end, so that we | |
293 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
294 | */ |
295 | /** | |
eb04c282 | 296 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 297 | * @m: the mount on which to take a write |
8366025e | 298 | * |
eb04c282 JK |
299 | * This tells the low-level filesystem that a write is about to be performed to |
300 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
301 | * returning success. This operation does not protect against filesystem being | |
302 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
303 | * called. This is effectively a refcount. | |
8366025e | 304 | */ |
eb04c282 | 305 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 306 | { |
83adc753 | 307 | struct mount *mnt = real_mount(m); |
3d733633 | 308 | int ret = 0; |
3d733633 | 309 | |
d3ef3d73 | 310 | preempt_disable(); |
c6653a83 | 311 | mnt_inc_writers(mnt); |
d3ef3d73 | 312 | /* |
c6653a83 | 313 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 314 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
315 | * incremented count after it has set MNT_WRITE_HOLD. | |
316 | */ | |
317 | smp_mb(); | |
1e75529e | 318 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 319 | cpu_relax(); |
320 | /* | |
321 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
322 | * be set to match its requirements. So we must not load that until | |
323 | * MNT_WRITE_HOLD is cleared. | |
324 | */ | |
325 | smp_rmb(); | |
4ed5e82f | 326 | if (mnt_is_readonly(m)) { |
c6653a83 | 327 | mnt_dec_writers(mnt); |
3d733633 | 328 | ret = -EROFS; |
3d733633 | 329 | } |
d3ef3d73 | 330 | preempt_enable(); |
eb04c282 JK |
331 | |
332 | return ret; | |
333 | } | |
334 | ||
335 | /** | |
336 | * mnt_want_write - get write access to a mount | |
337 | * @m: the mount on which to take a write | |
338 | * | |
339 | * This tells the low-level filesystem that a write is about to be performed to | |
340 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
341 | * is not frozen) before returning success. When the write operation is | |
342 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
343 | */ | |
344 | int mnt_want_write(struct vfsmount *m) | |
345 | { | |
346 | int ret; | |
347 | ||
348 | sb_start_write(m->mnt_sb); | |
349 | ret = __mnt_want_write(m); | |
350 | if (ret) | |
351 | sb_end_write(m->mnt_sb); | |
3d733633 | 352 | return ret; |
8366025e DH |
353 | } |
354 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
355 | ||
96029c4e | 356 | /** |
357 | * mnt_clone_write - get write access to a mount | |
358 | * @mnt: the mount on which to take a write | |
359 | * | |
360 | * This is effectively like mnt_want_write, except | |
361 | * it must only be used to take an extra write reference | |
362 | * on a mountpoint that we already know has a write reference | |
363 | * on it. This allows some optimisation. | |
364 | * | |
365 | * After finished, mnt_drop_write must be called as usual to | |
366 | * drop the reference. | |
367 | */ | |
368 | int mnt_clone_write(struct vfsmount *mnt) | |
369 | { | |
370 | /* superblock may be r/o */ | |
371 | if (__mnt_is_readonly(mnt)) | |
372 | return -EROFS; | |
373 | preempt_disable(); | |
83adc753 | 374 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 375 | preempt_enable(); |
376 | return 0; | |
377 | } | |
378 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
379 | ||
380 | /** | |
eb04c282 | 381 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 382 | * @file: the file who's mount on which to take a write |
383 | * | |
eb04c282 | 384 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 385 | * do some optimisations if the file is open for write already |
386 | */ | |
eb04c282 | 387 | int __mnt_want_write_file(struct file *file) |
96029c4e | 388 | { |
496ad9aa | 389 | struct inode *inode = file_inode(file); |
eb04c282 | 390 | |
2d8dd38a | 391 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) |
eb04c282 | 392 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 393 | else |
394 | return mnt_clone_write(file->f_path.mnt); | |
395 | } | |
eb04c282 JK |
396 | |
397 | /** | |
398 | * mnt_want_write_file - get write access to a file's mount | |
399 | * @file: the file who's mount on which to take a write | |
400 | * | |
401 | * This is like mnt_want_write, but it takes a file and can | |
402 | * do some optimisations if the file is open for write already | |
403 | */ | |
404 | int mnt_want_write_file(struct file *file) | |
405 | { | |
406 | int ret; | |
407 | ||
408 | sb_start_write(file->f_path.mnt->mnt_sb); | |
409 | ret = __mnt_want_write_file(file); | |
410 | if (ret) | |
411 | sb_end_write(file->f_path.mnt->mnt_sb); | |
412 | return ret; | |
413 | } | |
96029c4e | 414 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
415 | ||
8366025e | 416 | /** |
eb04c282 | 417 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
418 | * @mnt: the mount on which to give up write access |
419 | * | |
420 | * Tells the low-level filesystem that we are done | |
421 | * performing writes to it. Must be matched with | |
eb04c282 | 422 | * __mnt_want_write() call above. |
8366025e | 423 | */ |
eb04c282 | 424 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 425 | { |
d3ef3d73 | 426 | preempt_disable(); |
83adc753 | 427 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 428 | preempt_enable(); |
8366025e | 429 | } |
eb04c282 JK |
430 | |
431 | /** | |
432 | * mnt_drop_write - give up write access to a mount | |
433 | * @mnt: the mount on which to give up write access | |
434 | * | |
435 | * Tells the low-level filesystem that we are done performing writes to it and | |
436 | * also allows filesystem to be frozen again. Must be matched with | |
437 | * mnt_want_write() call above. | |
438 | */ | |
439 | void mnt_drop_write(struct vfsmount *mnt) | |
440 | { | |
441 | __mnt_drop_write(mnt); | |
442 | sb_end_write(mnt->mnt_sb); | |
443 | } | |
8366025e DH |
444 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
445 | ||
eb04c282 JK |
446 | void __mnt_drop_write_file(struct file *file) |
447 | { | |
448 | __mnt_drop_write(file->f_path.mnt); | |
449 | } | |
450 | ||
2a79f17e AV |
451 | void mnt_drop_write_file(struct file *file) |
452 | { | |
453 | mnt_drop_write(file->f_path.mnt); | |
454 | } | |
455 | EXPORT_SYMBOL(mnt_drop_write_file); | |
456 | ||
83adc753 | 457 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 458 | { |
3d733633 DH |
459 | int ret = 0; |
460 | ||
962830df | 461 | br_write_lock(&vfsmount_lock); |
83adc753 | 462 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 463 | /* |
d3ef3d73 | 464 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
465 | * should be visible before we do. | |
3d733633 | 466 | */ |
d3ef3d73 | 467 | smp_mb(); |
468 | ||
3d733633 | 469 | /* |
d3ef3d73 | 470 | * With writers on hold, if this value is zero, then there are |
471 | * definitely no active writers (although held writers may subsequently | |
472 | * increment the count, they'll have to wait, and decrement it after | |
473 | * seeing MNT_READONLY). | |
474 | * | |
475 | * It is OK to have counter incremented on one CPU and decremented on | |
476 | * another: the sum will add up correctly. The danger would be when we | |
477 | * sum up each counter, if we read a counter before it is incremented, | |
478 | * but then read another CPU's count which it has been subsequently | |
479 | * decremented from -- we would see more decrements than we should. | |
480 | * MNT_WRITE_HOLD protects against this scenario, because | |
481 | * mnt_want_write first increments count, then smp_mb, then spins on | |
482 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
483 | * we're counting up here. | |
3d733633 | 484 | */ |
c6653a83 | 485 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 486 | ret = -EBUSY; |
487 | else | |
83adc753 | 488 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 489 | /* |
490 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
491 | * that become unheld will see MNT_READONLY. | |
492 | */ | |
493 | smp_wmb(); | |
83adc753 | 494 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
962830df | 495 | br_write_unlock(&vfsmount_lock); |
3d733633 | 496 | return ret; |
8366025e | 497 | } |
8366025e | 498 | |
83adc753 | 499 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 500 | { |
962830df | 501 | br_write_lock(&vfsmount_lock); |
83adc753 | 502 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
962830df | 503 | br_write_unlock(&vfsmount_lock); |
2e4b7fcd DH |
504 | } |
505 | ||
4ed5e82f MS |
506 | int sb_prepare_remount_readonly(struct super_block *sb) |
507 | { | |
508 | struct mount *mnt; | |
509 | int err = 0; | |
510 | ||
8e8b8796 MS |
511 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
512 | if (atomic_long_read(&sb->s_remove_count)) | |
513 | return -EBUSY; | |
514 | ||
962830df | 515 | br_write_lock(&vfsmount_lock); |
4ed5e82f MS |
516 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
517 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
518 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
519 | smp_mb(); | |
520 | if (mnt_get_writers(mnt) > 0) { | |
521 | err = -EBUSY; | |
522 | break; | |
523 | } | |
524 | } | |
525 | } | |
8e8b8796 MS |
526 | if (!err && atomic_long_read(&sb->s_remove_count)) |
527 | err = -EBUSY; | |
528 | ||
4ed5e82f MS |
529 | if (!err) { |
530 | sb->s_readonly_remount = 1; | |
531 | smp_wmb(); | |
532 | } | |
533 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
534 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
535 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
536 | } | |
962830df | 537 | br_write_unlock(&vfsmount_lock); |
4ed5e82f MS |
538 | |
539 | return err; | |
540 | } | |
541 | ||
b105e270 | 542 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 543 | { |
52ba1621 | 544 | kfree(mnt->mnt_devname); |
73cd49ec | 545 | mnt_free_id(mnt); |
d3ef3d73 | 546 | #ifdef CONFIG_SMP |
68e8a9fe | 547 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 548 | #endif |
b105e270 | 549 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
550 | } |
551 | ||
552 | /* | |
a05964f3 RP |
553 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
554 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 555 | * vfsmount_lock must be held for read or write. |
1da177e4 | 556 | */ |
c7105365 | 557 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 558 | int dir) |
1da177e4 | 559 | { |
b58fed8b RP |
560 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
561 | struct list_head *tmp = head; | |
c7105365 | 562 | struct mount *p, *found = NULL; |
1da177e4 | 563 | |
1da177e4 | 564 | for (;;) { |
a05964f3 | 565 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
566 | p = NULL; |
567 | if (tmp == head) | |
568 | break; | |
1b8e5564 | 569 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 570 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 571 | found = p; |
1da177e4 LT |
572 | break; |
573 | } | |
574 | } | |
1da177e4 LT |
575 | return found; |
576 | } | |
577 | ||
a05964f3 | 578 | /* |
f015f126 DH |
579 | * lookup_mnt - Return the first child mount mounted at path |
580 | * | |
581 | * "First" means first mounted chronologically. If you create the | |
582 | * following mounts: | |
583 | * | |
584 | * mount /dev/sda1 /mnt | |
585 | * mount /dev/sda2 /mnt | |
586 | * mount /dev/sda3 /mnt | |
587 | * | |
588 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
589 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
590 | * /dev/sda2, then /dev/sda3, then NULL. | |
591 | * | |
592 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 593 | */ |
1c755af4 | 594 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 595 | { |
c7105365 | 596 | struct mount *child_mnt; |
99b7db7b | 597 | |
962830df | 598 | br_read_lock(&vfsmount_lock); |
c7105365 AV |
599 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
600 | if (child_mnt) { | |
601 | mnt_add_count(child_mnt, 1); | |
962830df | 602 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
603 | return &child_mnt->mnt; |
604 | } else { | |
962830df | 605 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
606 | return NULL; |
607 | } | |
a05964f3 RP |
608 | } |
609 | ||
84d17192 AV |
610 | static struct mountpoint *new_mountpoint(struct dentry *dentry) |
611 | { | |
612 | struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry); | |
613 | struct mountpoint *mp; | |
614 | ||
615 | list_for_each_entry(mp, chain, m_hash) { | |
616 | if (mp->m_dentry == dentry) { | |
617 | /* might be worth a WARN_ON() */ | |
618 | if (d_unlinked(dentry)) | |
619 | return ERR_PTR(-ENOENT); | |
620 | mp->m_count++; | |
621 | return mp; | |
622 | } | |
623 | } | |
624 | ||
625 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
626 | if (!mp) | |
627 | return ERR_PTR(-ENOMEM); | |
628 | ||
629 | spin_lock(&dentry->d_lock); | |
630 | if (d_unlinked(dentry)) { | |
631 | spin_unlock(&dentry->d_lock); | |
632 | kfree(mp); | |
633 | return ERR_PTR(-ENOENT); | |
634 | } | |
635 | dentry->d_flags |= DCACHE_MOUNTED; | |
636 | spin_unlock(&dentry->d_lock); | |
637 | mp->m_dentry = dentry; | |
638 | mp->m_count = 1; | |
639 | list_add(&mp->m_hash, chain); | |
640 | return mp; | |
641 | } | |
642 | ||
643 | static void put_mountpoint(struct mountpoint *mp) | |
644 | { | |
645 | if (!--mp->m_count) { | |
646 | struct dentry *dentry = mp->m_dentry; | |
647 | spin_lock(&dentry->d_lock); | |
648 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
649 | spin_unlock(&dentry->d_lock); | |
650 | list_del(&mp->m_hash); | |
651 | kfree(mp); | |
652 | } | |
653 | } | |
654 | ||
143c8c91 | 655 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 656 | { |
6b3286ed | 657 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
658 | } |
659 | ||
99b7db7b NP |
660 | /* |
661 | * vfsmount lock must be held for write | |
662 | */ | |
6b3286ed | 663 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
664 | { |
665 | if (ns) { | |
666 | ns->event = ++event; | |
667 | wake_up_interruptible(&ns->poll); | |
668 | } | |
669 | } | |
670 | ||
99b7db7b NP |
671 | /* |
672 | * vfsmount lock must be held for write | |
673 | */ | |
6b3286ed | 674 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
675 | { |
676 | if (ns && ns->event != event) { | |
677 | ns->event = event; | |
678 | wake_up_interruptible(&ns->poll); | |
679 | } | |
680 | } | |
681 | ||
99b7db7b NP |
682 | /* |
683 | * vfsmount lock must be held for write | |
684 | */ | |
419148da AV |
685 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
686 | { | |
a73324da | 687 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
688 | old_path->mnt = &mnt->mnt_parent->mnt; |
689 | mnt->mnt_parent = mnt; | |
a73324da | 690 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 691 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 692 | list_del_init(&mnt->mnt_hash); |
84d17192 AV |
693 | put_mountpoint(mnt->mnt_mp); |
694 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
695 | } |
696 | ||
99b7db7b NP |
697 | /* |
698 | * vfsmount lock must be held for write | |
699 | */ | |
84d17192 AV |
700 | void mnt_set_mountpoint(struct mount *mnt, |
701 | struct mountpoint *mp, | |
44d964d6 | 702 | struct mount *child_mnt) |
b90fa9ae | 703 | { |
84d17192 | 704 | mp->m_count++; |
3a2393d7 | 705 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 706 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 707 | child_mnt->mnt_parent = mnt; |
84d17192 | 708 | child_mnt->mnt_mp = mp; |
b90fa9ae RP |
709 | } |
710 | ||
99b7db7b NP |
711 | /* |
712 | * vfsmount lock must be held for write | |
713 | */ | |
84d17192 AV |
714 | static void attach_mnt(struct mount *mnt, |
715 | struct mount *parent, | |
716 | struct mountpoint *mp) | |
1da177e4 | 717 | { |
84d17192 | 718 | mnt_set_mountpoint(parent, mp, mnt); |
1b8e5564 | 719 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
84d17192 AV |
720 | hash(&parent->mnt, mp->m_dentry)); |
721 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
b90fa9ae RP |
722 | } |
723 | ||
724 | /* | |
99b7db7b | 725 | * vfsmount lock must be held for write |
b90fa9ae | 726 | */ |
4b2619a5 | 727 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 728 | { |
0714a533 | 729 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 730 | struct mount *m; |
b90fa9ae | 731 | LIST_HEAD(head); |
143c8c91 | 732 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 733 | |
0714a533 | 734 | BUG_ON(parent == mnt); |
b90fa9ae | 735 | |
1a4eeaf2 | 736 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 737 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 738 | m->mnt_ns = n; |
f03c6599 | 739 | |
b90fa9ae RP |
740 | list_splice(&head, n->list.prev); |
741 | ||
1b8e5564 | 742 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 743 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 744 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 745 | touch_mnt_namespace(n); |
1da177e4 LT |
746 | } |
747 | ||
909b0a88 | 748 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 749 | { |
6b41d536 AV |
750 | struct list_head *next = p->mnt_mounts.next; |
751 | if (next == &p->mnt_mounts) { | |
1da177e4 | 752 | while (1) { |
909b0a88 | 753 | if (p == root) |
1da177e4 | 754 | return NULL; |
6b41d536 AV |
755 | next = p->mnt_child.next; |
756 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 757 | break; |
0714a533 | 758 | p = p->mnt_parent; |
1da177e4 LT |
759 | } |
760 | } | |
6b41d536 | 761 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
762 | } |
763 | ||
315fc83e | 764 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 765 | { |
6b41d536 AV |
766 | struct list_head *prev = p->mnt_mounts.prev; |
767 | while (prev != &p->mnt_mounts) { | |
768 | p = list_entry(prev, struct mount, mnt_child); | |
769 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
770 | } |
771 | return p; | |
772 | } | |
773 | ||
9d412a43 AV |
774 | struct vfsmount * |
775 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
776 | { | |
b105e270 | 777 | struct mount *mnt; |
9d412a43 AV |
778 | struct dentry *root; |
779 | ||
780 | if (!type) | |
781 | return ERR_PTR(-ENODEV); | |
782 | ||
783 | mnt = alloc_vfsmnt(name); | |
784 | if (!mnt) | |
785 | return ERR_PTR(-ENOMEM); | |
786 | ||
787 | if (flags & MS_KERNMOUNT) | |
b105e270 | 788 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
789 | |
790 | root = mount_fs(type, flags, name, data); | |
791 | if (IS_ERR(root)) { | |
792 | free_vfsmnt(mnt); | |
793 | return ERR_CAST(root); | |
794 | } | |
795 | ||
b105e270 AV |
796 | mnt->mnt.mnt_root = root; |
797 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 798 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 799 | mnt->mnt_parent = mnt; |
962830df | 800 | br_write_lock(&vfsmount_lock); |
39f7c4db | 801 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
962830df | 802 | br_write_unlock(&vfsmount_lock); |
b105e270 | 803 | return &mnt->mnt; |
9d412a43 AV |
804 | } |
805 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
806 | ||
87129cc0 | 807 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 808 | int flag) |
1da177e4 | 809 | { |
87129cc0 | 810 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
811 | struct mount *mnt; |
812 | int err; | |
1da177e4 | 813 | |
be34d1a3 DH |
814 | mnt = alloc_vfsmnt(old->mnt_devname); |
815 | if (!mnt) | |
816 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 817 | |
7a472ef4 | 818 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
819 | mnt->mnt_group_id = 0; /* not a peer of original */ |
820 | else | |
821 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 822 | |
be34d1a3 DH |
823 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
824 | err = mnt_alloc_group_id(mnt); | |
825 | if (err) | |
826 | goto out_free; | |
1da177e4 | 827 | } |
be34d1a3 DH |
828 | |
829 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
132c94e3 EB |
830 | /* Don't allow unprivileged users to change mount flags */ |
831 | if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) | |
832 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
833 | ||
5ff9d8a6 EB |
834 | /* Don't allow unprivileged users to reveal what is under a mount */ |
835 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
836 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
837 | ||
be34d1a3 DH |
838 | atomic_inc(&sb->s_active); |
839 | mnt->mnt.mnt_sb = sb; | |
840 | mnt->mnt.mnt_root = dget(root); | |
841 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
842 | mnt->mnt_parent = mnt; | |
843 | br_write_lock(&vfsmount_lock); | |
844 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
845 | br_write_unlock(&vfsmount_lock); | |
846 | ||
7a472ef4 EB |
847 | if ((flag & CL_SLAVE) || |
848 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
849 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
850 | mnt->mnt_master = old; | |
851 | CLEAR_MNT_SHARED(mnt); | |
852 | } else if (!(flag & CL_PRIVATE)) { | |
853 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
854 | list_add(&mnt->mnt_share, &old->mnt_share); | |
855 | if (IS_MNT_SLAVE(old)) | |
856 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
857 | mnt->mnt_master = old->mnt_master; | |
858 | } | |
859 | if (flag & CL_MAKE_SHARED) | |
860 | set_mnt_shared(mnt); | |
861 | ||
862 | /* stick the duplicate mount on the same expiry list | |
863 | * as the original if that was on one */ | |
864 | if (flag & CL_EXPIRE) { | |
865 | if (!list_empty(&old->mnt_expire)) | |
866 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
867 | } | |
868 | ||
cb338d06 | 869 | return mnt; |
719f5d7f MS |
870 | |
871 | out_free: | |
872 | free_vfsmnt(mnt); | |
be34d1a3 | 873 | return ERR_PTR(err); |
1da177e4 LT |
874 | } |
875 | ||
83adc753 | 876 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 877 | { |
83adc753 AV |
878 | struct vfsmount *m = &mnt->mnt; |
879 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 880 | |
3d733633 DH |
881 | /* |
882 | * This probably indicates that somebody messed | |
883 | * up a mnt_want/drop_write() pair. If this | |
884 | * happens, the filesystem was probably unable | |
885 | * to make r/w->r/o transitions. | |
886 | */ | |
d3ef3d73 | 887 | /* |
b3e19d92 NP |
888 | * The locking used to deal with mnt_count decrement provides barriers, |
889 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 890 | */ |
c6653a83 | 891 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
892 | fsnotify_vfsmount_delete(m); |
893 | dput(m->mnt_root); | |
894 | free_vfsmnt(mnt); | |
1da177e4 LT |
895 | deactivate_super(sb); |
896 | } | |
897 | ||
900148dc | 898 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 899 | { |
b3e19d92 | 900 | put_again: |
f03c6599 | 901 | #ifdef CONFIG_SMP |
962830df | 902 | br_read_lock(&vfsmount_lock); |
f7a99c5b AV |
903 | if (likely(mnt->mnt_ns)) { |
904 | /* shouldn't be the last one */ | |
aa9c0e07 | 905 | mnt_add_count(mnt, -1); |
962830df | 906 | br_read_unlock(&vfsmount_lock); |
f03c6599 | 907 | return; |
b3e19d92 | 908 | } |
962830df | 909 | br_read_unlock(&vfsmount_lock); |
b3e19d92 | 910 | |
962830df | 911 | br_write_lock(&vfsmount_lock); |
aa9c0e07 | 912 | mnt_add_count(mnt, -1); |
b3e19d92 | 913 | if (mnt_get_count(mnt)) { |
962830df | 914 | br_write_unlock(&vfsmount_lock); |
99b7db7b NP |
915 | return; |
916 | } | |
b3e19d92 | 917 | #else |
aa9c0e07 | 918 | mnt_add_count(mnt, -1); |
b3e19d92 | 919 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 920 | return; |
962830df | 921 | br_write_lock(&vfsmount_lock); |
f03c6599 | 922 | #endif |
863d684f AV |
923 | if (unlikely(mnt->mnt_pinned)) { |
924 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
925 | mnt->mnt_pinned = 0; | |
962830df | 926 | br_write_unlock(&vfsmount_lock); |
900148dc | 927 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 928 | goto put_again; |
7b7b1ace | 929 | } |
962830df | 930 | |
39f7c4db | 931 | list_del(&mnt->mnt_instance); |
962830df | 932 | br_write_unlock(&vfsmount_lock); |
b3e19d92 NP |
933 | mntfree(mnt); |
934 | } | |
b3e19d92 NP |
935 | |
936 | void mntput(struct vfsmount *mnt) | |
937 | { | |
938 | if (mnt) { | |
863d684f | 939 | struct mount *m = real_mount(mnt); |
b3e19d92 | 940 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
941 | if (unlikely(m->mnt_expiry_mark)) |
942 | m->mnt_expiry_mark = 0; | |
943 | mntput_no_expire(m); | |
b3e19d92 NP |
944 | } |
945 | } | |
946 | EXPORT_SYMBOL(mntput); | |
947 | ||
948 | struct vfsmount *mntget(struct vfsmount *mnt) | |
949 | { | |
950 | if (mnt) | |
83adc753 | 951 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
952 | return mnt; |
953 | } | |
954 | EXPORT_SYMBOL(mntget); | |
955 | ||
7b7b1ace AV |
956 | void mnt_pin(struct vfsmount *mnt) |
957 | { | |
962830df | 958 | br_write_lock(&vfsmount_lock); |
863d684f | 959 | real_mount(mnt)->mnt_pinned++; |
962830df | 960 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 961 | } |
7b7b1ace AV |
962 | EXPORT_SYMBOL(mnt_pin); |
963 | ||
863d684f | 964 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 965 | { |
863d684f | 966 | struct mount *mnt = real_mount(m); |
962830df | 967 | br_write_lock(&vfsmount_lock); |
7b7b1ace | 968 | if (mnt->mnt_pinned) { |
863d684f | 969 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
970 | mnt->mnt_pinned--; |
971 | } | |
962830df | 972 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 973 | } |
7b7b1ace | 974 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 975 | |
b3b304a2 MS |
976 | static inline void mangle(struct seq_file *m, const char *s) |
977 | { | |
978 | seq_escape(m, s, " \t\n\\"); | |
979 | } | |
980 | ||
981 | /* | |
982 | * Simple .show_options callback for filesystems which don't want to | |
983 | * implement more complex mount option showing. | |
984 | * | |
985 | * See also save_mount_options(). | |
986 | */ | |
34c80b1d | 987 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 988 | { |
2a32cebd AV |
989 | const char *options; |
990 | ||
991 | rcu_read_lock(); | |
34c80b1d | 992 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
993 | |
994 | if (options != NULL && options[0]) { | |
995 | seq_putc(m, ','); | |
996 | mangle(m, options); | |
997 | } | |
2a32cebd | 998 | rcu_read_unlock(); |
b3b304a2 MS |
999 | |
1000 | return 0; | |
1001 | } | |
1002 | EXPORT_SYMBOL(generic_show_options); | |
1003 | ||
1004 | /* | |
1005 | * If filesystem uses generic_show_options(), this function should be | |
1006 | * called from the fill_super() callback. | |
1007 | * | |
1008 | * The .remount_fs callback usually needs to be handled in a special | |
1009 | * way, to make sure, that previous options are not overwritten if the | |
1010 | * remount fails. | |
1011 | * | |
1012 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1013 | * reset all options to their default value, but changes only newly | |
1014 | * given options, then the displayed options will not reflect reality | |
1015 | * any more. | |
1016 | */ | |
1017 | void save_mount_options(struct super_block *sb, char *options) | |
1018 | { | |
2a32cebd AV |
1019 | BUG_ON(sb->s_options); |
1020 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
1021 | } |
1022 | EXPORT_SYMBOL(save_mount_options); | |
1023 | ||
2a32cebd AV |
1024 | void replace_mount_options(struct super_block *sb, char *options) |
1025 | { | |
1026 | char *old = sb->s_options; | |
1027 | rcu_assign_pointer(sb->s_options, options); | |
1028 | if (old) { | |
1029 | synchronize_rcu(); | |
1030 | kfree(old); | |
1031 | } | |
1032 | } | |
1033 | EXPORT_SYMBOL(replace_mount_options); | |
1034 | ||
a1a2c409 | 1035 | #ifdef CONFIG_PROC_FS |
0226f492 | 1036 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1037 | static void *m_start(struct seq_file *m, loff_t *pos) |
1038 | { | |
6ce6e24e | 1039 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1040 | |
390c6843 | 1041 | down_read(&namespace_sem); |
a1a2c409 | 1042 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1046 | { | |
6ce6e24e | 1047 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1048 | |
a1a2c409 | 1049 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
1050 | } |
1051 | ||
1052 | static void m_stop(struct seq_file *m, void *v) | |
1053 | { | |
390c6843 | 1054 | up_read(&namespace_sem); |
1da177e4 LT |
1055 | } |
1056 | ||
0226f492 | 1057 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1058 | { |
6ce6e24e | 1059 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1060 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1061 | return p->show(m, &r->mnt); |
1da177e4 LT |
1062 | } |
1063 | ||
a1a2c409 | 1064 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1065 | .start = m_start, |
1066 | .next = m_next, | |
1067 | .stop = m_stop, | |
0226f492 | 1068 | .show = m_show, |
b4629fe2 | 1069 | }; |
a1a2c409 | 1070 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1071 | |
1da177e4 LT |
1072 | /** |
1073 | * may_umount_tree - check if a mount tree is busy | |
1074 | * @mnt: root of mount tree | |
1075 | * | |
1076 | * This is called to check if a tree of mounts has any | |
1077 | * open files, pwds, chroots or sub mounts that are | |
1078 | * busy. | |
1079 | */ | |
909b0a88 | 1080 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1081 | { |
909b0a88 | 1082 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1083 | int actual_refs = 0; |
1084 | int minimum_refs = 0; | |
315fc83e | 1085 | struct mount *p; |
909b0a88 | 1086 | BUG_ON(!m); |
1da177e4 | 1087 | |
b3e19d92 | 1088 | /* write lock needed for mnt_get_count */ |
962830df | 1089 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1090 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1091 | actual_refs += mnt_get_count(p); |
1da177e4 | 1092 | minimum_refs += 2; |
1da177e4 | 1093 | } |
962830df | 1094 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1095 | |
1096 | if (actual_refs > minimum_refs) | |
e3474a8e | 1097 | return 0; |
1da177e4 | 1098 | |
e3474a8e | 1099 | return 1; |
1da177e4 LT |
1100 | } |
1101 | ||
1102 | EXPORT_SYMBOL(may_umount_tree); | |
1103 | ||
1104 | /** | |
1105 | * may_umount - check if a mount point is busy | |
1106 | * @mnt: root of mount | |
1107 | * | |
1108 | * This is called to check if a mount point has any | |
1109 | * open files, pwds, chroots or sub mounts. If the | |
1110 | * mount has sub mounts this will return busy | |
1111 | * regardless of whether the sub mounts are busy. | |
1112 | * | |
1113 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1114 | * give false negatives. The main reason why it's here is that we need | |
1115 | * a non-destructive way to look for easily umountable filesystems. | |
1116 | */ | |
1117 | int may_umount(struct vfsmount *mnt) | |
1118 | { | |
e3474a8e | 1119 | int ret = 1; |
8ad08d8a | 1120 | down_read(&namespace_sem); |
962830df | 1121 | br_write_lock(&vfsmount_lock); |
1ab59738 | 1122 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1123 | ret = 0; |
962830df | 1124 | br_write_unlock(&vfsmount_lock); |
8ad08d8a | 1125 | up_read(&namespace_sem); |
a05964f3 | 1126 | return ret; |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | EXPORT_SYMBOL(may_umount); | |
1130 | ||
e3197d83 AV |
1131 | static LIST_HEAD(unmounted); /* protected by namespace_sem */ |
1132 | ||
97216be0 | 1133 | static void namespace_unlock(void) |
70fbcdf4 | 1134 | { |
d5e50f74 | 1135 | struct mount *mnt; |
97216be0 AV |
1136 | LIST_HEAD(head); |
1137 | ||
1138 | if (likely(list_empty(&unmounted))) { | |
1139 | up_write(&namespace_sem); | |
1140 | return; | |
1141 | } | |
1142 | ||
1143 | list_splice_init(&unmounted, &head); | |
1144 | up_write(&namespace_sem); | |
1145 | ||
1146 | while (!list_empty(&head)) { | |
1147 | mnt = list_first_entry(&head, struct mount, mnt_hash); | |
1b8e5564 | 1148 | list_del_init(&mnt->mnt_hash); |
676da58d | 1149 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 1150 | struct dentry *dentry; |
863d684f | 1151 | struct mount *m; |
99b7db7b | 1152 | |
962830df | 1153 | br_write_lock(&vfsmount_lock); |
a73324da | 1154 | dentry = mnt->mnt_mountpoint; |
863d684f | 1155 | m = mnt->mnt_parent; |
a73324da | 1156 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1157 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1158 | m->mnt_ghosts--; |
962830df | 1159 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1160 | dput(dentry); |
863d684f | 1161 | mntput(&m->mnt); |
70fbcdf4 | 1162 | } |
d5e50f74 | 1163 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1164 | } |
1165 | } | |
1166 | ||
97216be0 | 1167 | static inline void namespace_lock(void) |
e3197d83 | 1168 | { |
97216be0 | 1169 | down_write(&namespace_sem); |
e3197d83 AV |
1170 | } |
1171 | ||
99b7db7b NP |
1172 | /* |
1173 | * vfsmount lock must be held for write | |
1174 | * namespace_sem must be held for write | |
1175 | */ | |
328e6d90 | 1176 | void umount_tree(struct mount *mnt, int propagate) |
1da177e4 | 1177 | { |
7b8a53fd | 1178 | LIST_HEAD(tmp_list); |
315fc83e | 1179 | struct mount *p; |
1da177e4 | 1180 | |
909b0a88 | 1181 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1182 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1183 | |
a05964f3 | 1184 | if (propagate) |
7b8a53fd | 1185 | propagate_umount(&tmp_list); |
a05964f3 | 1186 | |
1b8e5564 | 1187 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1188 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1189 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1190 | __touch_mnt_namespace(p->mnt_ns); |
1191 | p->mnt_ns = NULL; | |
6b41d536 | 1192 | list_del_init(&p->mnt_child); |
676da58d | 1193 | if (mnt_has_parent(p)) { |
863d684f | 1194 | p->mnt_parent->mnt_ghosts++; |
84d17192 AV |
1195 | put_mountpoint(p->mnt_mp); |
1196 | p->mnt_mp = NULL; | |
7c4b93d8 | 1197 | } |
0f0afb1d | 1198 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1199 | } |
328e6d90 | 1200 | list_splice(&tmp_list, &unmounted); |
1da177e4 LT |
1201 | } |
1202 | ||
b54b9be7 | 1203 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1204 | |
1ab59738 | 1205 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1206 | { |
1ab59738 | 1207 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1208 | int retval; |
1209 | ||
1ab59738 | 1210 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1211 | if (retval) |
1212 | return retval; | |
1213 | ||
1214 | /* | |
1215 | * Allow userspace to request a mountpoint be expired rather than | |
1216 | * unmounting unconditionally. Unmount only happens if: | |
1217 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1218 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1219 | */ | |
1220 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1221 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1222 | flags & (MNT_FORCE | MNT_DETACH)) |
1223 | return -EINVAL; | |
1224 | ||
b3e19d92 NP |
1225 | /* |
1226 | * probably don't strictly need the lock here if we examined | |
1227 | * all race cases, but it's a slowpath. | |
1228 | */ | |
962830df | 1229 | br_write_lock(&vfsmount_lock); |
83adc753 | 1230 | if (mnt_get_count(mnt) != 2) { |
962830df | 1231 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1232 | return -EBUSY; |
b3e19d92 | 1233 | } |
962830df | 1234 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1235 | |
863d684f | 1236 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1237 | return -EAGAIN; |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * If we may have to abort operations to get out of this | |
1242 | * mount, and they will themselves hold resources we must | |
1243 | * allow the fs to do things. In the Unix tradition of | |
1244 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1245 | * might fail to complete on the first run through as other tasks | |
1246 | * must return, and the like. Thats for the mount program to worry | |
1247 | * about for the moment. | |
1248 | */ | |
1249 | ||
42faad99 | 1250 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1251 | sb->s_op->umount_begin(sb); |
42faad99 | 1252 | } |
1da177e4 LT |
1253 | |
1254 | /* | |
1255 | * No sense to grab the lock for this test, but test itself looks | |
1256 | * somewhat bogus. Suggestions for better replacement? | |
1257 | * Ho-hum... In principle, we might treat that as umount + switch | |
1258 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1259 | * Actually it makes sense, especially if rootfs would contain a | |
1260 | * /reboot - static binary that would close all descriptors and | |
1261 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1262 | */ | |
1ab59738 | 1263 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1264 | /* |
1265 | * Special case for "unmounting" root ... | |
1266 | * we just try to remount it readonly. | |
1267 | */ | |
1268 | down_write(&sb->s_umount); | |
4aa98cf7 | 1269 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1270 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1271 | up_write(&sb->s_umount); |
1272 | return retval; | |
1273 | } | |
1274 | ||
97216be0 | 1275 | namespace_lock(); |
962830df | 1276 | br_write_lock(&vfsmount_lock); |
5addc5dd | 1277 | event++; |
1da177e4 | 1278 | |
c35038be | 1279 | if (!(flags & MNT_DETACH)) |
b54b9be7 | 1280 | shrink_submounts(mnt); |
c35038be | 1281 | |
1da177e4 | 1282 | retval = -EBUSY; |
a05964f3 | 1283 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1284 | if (!list_empty(&mnt->mnt_list)) |
328e6d90 | 1285 | umount_tree(mnt, 1); |
1da177e4 LT |
1286 | retval = 0; |
1287 | } | |
962830df | 1288 | br_write_unlock(&vfsmount_lock); |
e3197d83 | 1289 | namespace_unlock(); |
1da177e4 LT |
1290 | return retval; |
1291 | } | |
1292 | ||
9b40bc90 AV |
1293 | /* |
1294 | * Is the caller allowed to modify his namespace? | |
1295 | */ | |
1296 | static inline bool may_mount(void) | |
1297 | { | |
1298 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1299 | } | |
1300 | ||
1da177e4 LT |
1301 | /* |
1302 | * Now umount can handle mount points as well as block devices. | |
1303 | * This is important for filesystems which use unnamed block devices. | |
1304 | * | |
1305 | * We now support a flag for forced unmount like the other 'big iron' | |
1306 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1307 | */ | |
1308 | ||
bdc480e3 | 1309 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1310 | { |
2d8f3038 | 1311 | struct path path; |
900148dc | 1312 | struct mount *mnt; |
1da177e4 | 1313 | int retval; |
db1f05bb | 1314 | int lookup_flags = 0; |
1da177e4 | 1315 | |
db1f05bb MS |
1316 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1317 | return -EINVAL; | |
1318 | ||
9b40bc90 AV |
1319 | if (!may_mount()) |
1320 | return -EPERM; | |
1321 | ||
db1f05bb MS |
1322 | if (!(flags & UMOUNT_NOFOLLOW)) |
1323 | lookup_flags |= LOOKUP_FOLLOW; | |
1324 | ||
1325 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1326 | if (retval) |
1327 | goto out; | |
900148dc | 1328 | mnt = real_mount(path.mnt); |
1da177e4 | 1329 | retval = -EINVAL; |
2d8f3038 | 1330 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1331 | goto dput_and_out; |
143c8c91 | 1332 | if (!check_mnt(mnt)) |
1da177e4 | 1333 | goto dput_and_out; |
5ff9d8a6 EB |
1334 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1335 | goto dput_and_out; | |
1da177e4 | 1336 | |
900148dc | 1337 | retval = do_umount(mnt, flags); |
1da177e4 | 1338 | dput_and_out: |
429731b1 | 1339 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1340 | dput(path.dentry); |
900148dc | 1341 | mntput_no_expire(mnt); |
1da177e4 LT |
1342 | out: |
1343 | return retval; | |
1344 | } | |
1345 | ||
1346 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1347 | ||
1348 | /* | |
b58fed8b | 1349 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1350 | */ |
bdc480e3 | 1351 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1352 | { |
b58fed8b | 1353 | return sys_umount(name, 0); |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | #endif | |
1357 | ||
4ce5d2b1 | 1358 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1359 | { |
4ce5d2b1 EB |
1360 | /* Is this a proxy for a mount namespace? */ |
1361 | struct inode *inode = dentry->d_inode; | |
0bb80f24 | 1362 | struct proc_ns *ei; |
8823c079 EB |
1363 | |
1364 | if (!proc_ns_inode(inode)) | |
1365 | return false; | |
1366 | ||
0bb80f24 | 1367 | ei = get_proc_ns(inode); |
8823c079 EB |
1368 | if (ei->ns_ops != &mntns_operations) |
1369 | return false; | |
1370 | ||
4ce5d2b1 EB |
1371 | return true; |
1372 | } | |
1373 | ||
1374 | static bool mnt_ns_loop(struct dentry *dentry) | |
1375 | { | |
1376 | /* Could bind mounting the mount namespace inode cause a | |
1377 | * mount namespace loop? | |
1378 | */ | |
1379 | struct mnt_namespace *mnt_ns; | |
1380 | if (!is_mnt_ns_file(dentry)) | |
1381 | return false; | |
1382 | ||
1383 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
8823c079 EB |
1384 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1385 | } | |
1386 | ||
87129cc0 | 1387 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1388 | int flag) |
1da177e4 | 1389 | { |
84d17192 | 1390 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1391 | |
4ce5d2b1 EB |
1392 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1393 | return ERR_PTR(-EINVAL); | |
1394 | ||
1395 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1396 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1397 | |
36341f64 | 1398 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1399 | if (IS_ERR(q)) |
1400 | return q; | |
1401 | ||
5ff9d8a6 | 1402 | q->mnt.mnt_flags &= ~MNT_LOCKED; |
a73324da | 1403 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1404 | |
1405 | p = mnt; | |
6b41d536 | 1406 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1407 | struct mount *s; |
7ec02ef1 | 1408 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1409 | continue; |
1410 | ||
909b0a88 | 1411 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1412 | if (!(flag & CL_COPY_UNBINDABLE) && |
1413 | IS_MNT_UNBINDABLE(s)) { | |
1414 | s = skip_mnt_tree(s); | |
1415 | continue; | |
1416 | } | |
1417 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1418 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1419 | s = skip_mnt_tree(s); |
1420 | continue; | |
1421 | } | |
0714a533 AV |
1422 | while (p != s->mnt_parent) { |
1423 | p = p->mnt_parent; | |
1424 | q = q->mnt_parent; | |
1da177e4 | 1425 | } |
87129cc0 | 1426 | p = s; |
84d17192 | 1427 | parent = q; |
87129cc0 | 1428 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1429 | if (IS_ERR(q)) |
1430 | goto out; | |
962830df | 1431 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 1432 | list_add_tail(&q->mnt_list, &res->mnt_list); |
84d17192 | 1433 | attach_mnt(q, parent, p->mnt_mp); |
962830df | 1434 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1435 | } |
1436 | } | |
1437 | return res; | |
be34d1a3 | 1438 | out: |
1da177e4 | 1439 | if (res) { |
962830df | 1440 | br_write_lock(&vfsmount_lock); |
328e6d90 | 1441 | umount_tree(res, 0); |
962830df | 1442 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1443 | } |
be34d1a3 | 1444 | return q; |
1da177e4 LT |
1445 | } |
1446 | ||
be34d1a3 DH |
1447 | /* Caller should check returned pointer for errors */ |
1448 | ||
589ff870 | 1449 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1450 | { |
cb338d06 | 1451 | struct mount *tree; |
97216be0 | 1452 | namespace_lock(); |
87129cc0 AV |
1453 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1454 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1455 | namespace_unlock(); |
be34d1a3 DH |
1456 | if (IS_ERR(tree)) |
1457 | return NULL; | |
1458 | return &tree->mnt; | |
8aec0809 AV |
1459 | } |
1460 | ||
1461 | void drop_collected_mounts(struct vfsmount *mnt) | |
1462 | { | |
97216be0 | 1463 | namespace_lock(); |
962830df | 1464 | br_write_lock(&vfsmount_lock); |
328e6d90 | 1465 | umount_tree(real_mount(mnt), 0); |
962830df | 1466 | br_write_unlock(&vfsmount_lock); |
3ab6abee | 1467 | namespace_unlock(); |
8aec0809 AV |
1468 | } |
1469 | ||
1f707137 AV |
1470 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1471 | struct vfsmount *root) | |
1472 | { | |
1a4eeaf2 | 1473 | struct mount *mnt; |
1f707137 AV |
1474 | int res = f(root, arg); |
1475 | if (res) | |
1476 | return res; | |
1a4eeaf2 AV |
1477 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1478 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1479 | if (res) |
1480 | return res; | |
1481 | } | |
1482 | return 0; | |
1483 | } | |
1484 | ||
4b8b21f4 | 1485 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1486 | { |
315fc83e | 1487 | struct mount *p; |
719f5d7f | 1488 | |
909b0a88 | 1489 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1490 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1491 | mnt_release_group_id(p); |
719f5d7f MS |
1492 | } |
1493 | } | |
1494 | ||
4b8b21f4 | 1495 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1496 | { |
315fc83e | 1497 | struct mount *p; |
719f5d7f | 1498 | |
909b0a88 | 1499 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1500 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1501 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1502 | if (err) { |
4b8b21f4 | 1503 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1504 | return err; |
1505 | } | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | return 0; | |
1510 | } | |
1511 | ||
b90fa9ae RP |
1512 | /* |
1513 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1514 | * @nd : place the mount tree @source_mnt is attached |
1515 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1516 | * store the parent mount and mountpoint dentry. | |
1517 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1518 | * |
1519 | * NOTE: in the table below explains the semantics when a source mount | |
1520 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1521 | * --------------------------------------------------------------------------- |
1522 | * | BIND MOUNT OPERATION | | |
1523 | * |************************************************************************** | |
1524 | * | source-->| shared | private | slave | unbindable | | |
1525 | * | dest | | | | | | |
1526 | * | | | | | | | | |
1527 | * | v | | | | | | |
1528 | * |************************************************************************** | |
1529 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1530 | * | | | | | | | |
1531 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1532 | * *************************************************************************** | |
b90fa9ae RP |
1533 | * A bind operation clones the source mount and mounts the clone on the |
1534 | * destination mount. | |
1535 | * | |
1536 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1537 | * tree of the destination mount and the cloned mount is added to | |
1538 | * the peer group of the source mount. | |
1539 | * (+) the cloned mount is created under the destination mount and is marked | |
1540 | * as shared. The cloned mount is added to the peer group of the source | |
1541 | * mount. | |
5afe0022 RP |
1542 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1543 | * of the destination mount and the cloned mount is made slave | |
1544 | * of the same master as that of the source mount. The cloned mount | |
1545 | * is marked as 'shared and slave'. | |
1546 | * (*) the cloned mount is made a slave of the same master as that of the | |
1547 | * source mount. | |
1548 | * | |
9676f0c6 RP |
1549 | * --------------------------------------------------------------------------- |
1550 | * | MOVE MOUNT OPERATION | | |
1551 | * |************************************************************************** | |
1552 | * | source-->| shared | private | slave | unbindable | | |
1553 | * | dest | | | | | | |
1554 | * | | | | | | | | |
1555 | * | v | | | | | | |
1556 | * |************************************************************************** | |
1557 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1558 | * | | | | | | | |
1559 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1560 | * *************************************************************************** | |
5afe0022 RP |
1561 | * |
1562 | * (+) the mount is moved to the destination. And is then propagated to | |
1563 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1564 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1565 | * (+++) the mount is moved to the destination and is then propagated to |
1566 | * all the mounts belonging to the destination mount's propagation tree. | |
1567 | * the mount is marked as 'shared and slave'. | |
1568 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1569 | * |
1570 | * if the source mount is a tree, the operations explained above is | |
1571 | * applied to each mount in the tree. | |
1572 | * Must be called without spinlocks held, since this function can sleep | |
1573 | * in allocations. | |
1574 | */ | |
0fb54e50 | 1575 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1576 | struct mount *dest_mnt, |
1577 | struct mountpoint *dest_mp, | |
1578 | struct path *parent_path) | |
b90fa9ae RP |
1579 | { |
1580 | LIST_HEAD(tree_list); | |
315fc83e | 1581 | struct mount *child, *p; |
719f5d7f | 1582 | int err; |
b90fa9ae | 1583 | |
fc7be130 | 1584 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1585 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1586 | if (err) |
1587 | goto out; | |
1588 | } | |
84d17192 | 1589 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
719f5d7f MS |
1590 | if (err) |
1591 | goto out_cleanup_ids; | |
b90fa9ae | 1592 | |
962830df | 1593 | br_write_lock(&vfsmount_lock); |
df1a1ad2 | 1594 | |
fc7be130 | 1595 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1596 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1597 | set_mnt_shared(p); |
b90fa9ae | 1598 | } |
1a390689 | 1599 | if (parent_path) { |
0fb54e50 | 1600 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1601 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1602 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1603 | } else { |
84d17192 | 1604 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
0fb54e50 | 1605 | commit_tree(source_mnt); |
21444403 | 1606 | } |
b90fa9ae | 1607 | |
1b8e5564 AV |
1608 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1609 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1610 | commit_tree(child); |
b90fa9ae | 1611 | } |
962830df | 1612 | br_write_unlock(&vfsmount_lock); |
99b7db7b | 1613 | |
b90fa9ae | 1614 | return 0; |
719f5d7f MS |
1615 | |
1616 | out_cleanup_ids: | |
fc7be130 | 1617 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1618 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1619 | out: |
1620 | return err; | |
b90fa9ae RP |
1621 | } |
1622 | ||
84d17192 | 1623 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
1624 | { |
1625 | struct vfsmount *mnt; | |
84d17192 | 1626 | struct dentry *dentry = path->dentry; |
b12cea91 | 1627 | retry: |
84d17192 AV |
1628 | mutex_lock(&dentry->d_inode->i_mutex); |
1629 | if (unlikely(cant_mount(dentry))) { | |
1630 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1631 | return ERR_PTR(-ENOENT); | |
b12cea91 | 1632 | } |
97216be0 | 1633 | namespace_lock(); |
b12cea91 | 1634 | mnt = lookup_mnt(path); |
84d17192 AV |
1635 | if (likely(!mnt)) { |
1636 | struct mountpoint *mp = new_mountpoint(dentry); | |
1637 | if (IS_ERR(mp)) { | |
97216be0 | 1638 | namespace_unlock(); |
84d17192 AV |
1639 | mutex_unlock(&dentry->d_inode->i_mutex); |
1640 | return mp; | |
1641 | } | |
1642 | return mp; | |
1643 | } | |
97216be0 | 1644 | namespace_unlock(); |
b12cea91 AV |
1645 | mutex_unlock(&path->dentry->d_inode->i_mutex); |
1646 | path_put(path); | |
1647 | path->mnt = mnt; | |
84d17192 | 1648 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
1649 | goto retry; |
1650 | } | |
1651 | ||
84d17192 | 1652 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 1653 | { |
84d17192 AV |
1654 | struct dentry *dentry = where->m_dentry; |
1655 | put_mountpoint(where); | |
328e6d90 | 1656 | namespace_unlock(); |
84d17192 | 1657 | mutex_unlock(&dentry->d_inode->i_mutex); |
b12cea91 AV |
1658 | } |
1659 | ||
84d17192 | 1660 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 1661 | { |
95bc5f25 | 1662 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1663 | return -EINVAL; |
1664 | ||
84d17192 | 1665 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != |
95bc5f25 | 1666 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1667 | return -ENOTDIR; |
1668 | ||
84d17192 | 1669 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
1670 | } |
1671 | ||
7a2e8a8f VA |
1672 | /* |
1673 | * Sanity check the flags to change_mnt_propagation. | |
1674 | */ | |
1675 | ||
1676 | static int flags_to_propagation_type(int flags) | |
1677 | { | |
7c6e984d | 1678 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1679 | |
1680 | /* Fail if any non-propagation flags are set */ | |
1681 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1682 | return 0; | |
1683 | /* Only one propagation flag should be set */ | |
1684 | if (!is_power_of_2(type)) | |
1685 | return 0; | |
1686 | return type; | |
1687 | } | |
1688 | ||
07b20889 RP |
1689 | /* |
1690 | * recursively change the type of the mountpoint. | |
1691 | */ | |
0a0d8a46 | 1692 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1693 | { |
315fc83e | 1694 | struct mount *m; |
4b8b21f4 | 1695 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1696 | int recurse = flag & MS_REC; |
7a2e8a8f | 1697 | int type; |
719f5d7f | 1698 | int err = 0; |
07b20889 | 1699 | |
2d92ab3c | 1700 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1701 | return -EINVAL; |
1702 | ||
7a2e8a8f VA |
1703 | type = flags_to_propagation_type(flag); |
1704 | if (!type) | |
1705 | return -EINVAL; | |
1706 | ||
97216be0 | 1707 | namespace_lock(); |
719f5d7f MS |
1708 | if (type == MS_SHARED) { |
1709 | err = invent_group_ids(mnt, recurse); | |
1710 | if (err) | |
1711 | goto out_unlock; | |
1712 | } | |
1713 | ||
962830df | 1714 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1715 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1716 | change_mnt_propagation(m, type); |
962830df | 1717 | br_write_unlock(&vfsmount_lock); |
719f5d7f MS |
1718 | |
1719 | out_unlock: | |
97216be0 | 1720 | namespace_unlock(); |
719f5d7f | 1721 | return err; |
07b20889 RP |
1722 | } |
1723 | ||
5ff9d8a6 EB |
1724 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1725 | { | |
1726 | struct mount *child; | |
1727 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1728 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1729 | continue; | |
1730 | ||
1731 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1732 | return true; | |
1733 | } | |
1734 | return false; | |
1735 | } | |
1736 | ||
1da177e4 LT |
1737 | /* |
1738 | * do loopback mount. | |
1739 | */ | |
808d4e3c | 1740 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1741 | int recurse) |
1da177e4 | 1742 | { |
2d92ab3c | 1743 | struct path old_path; |
84d17192 AV |
1744 | struct mount *mnt = NULL, *old, *parent; |
1745 | struct mountpoint *mp; | |
57eccb83 | 1746 | int err; |
1da177e4 LT |
1747 | if (!old_name || !*old_name) |
1748 | return -EINVAL; | |
815d405c | 1749 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1750 | if (err) |
1751 | return err; | |
1752 | ||
8823c079 | 1753 | err = -EINVAL; |
4ce5d2b1 | 1754 | if (mnt_ns_loop(old_path.dentry)) |
8823c079 EB |
1755 | goto out; |
1756 | ||
84d17192 AV |
1757 | mp = lock_mount(path); |
1758 | err = PTR_ERR(mp); | |
1759 | if (IS_ERR(mp)) | |
b12cea91 AV |
1760 | goto out; |
1761 | ||
87129cc0 | 1762 | old = real_mount(old_path.mnt); |
84d17192 | 1763 | parent = real_mount(path->mnt); |
87129cc0 | 1764 | |
1da177e4 | 1765 | err = -EINVAL; |
fc7be130 | 1766 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1767 | goto out2; |
9676f0c6 | 1768 | |
84d17192 | 1769 | if (!check_mnt(parent) || !check_mnt(old)) |
b12cea91 | 1770 | goto out2; |
1da177e4 | 1771 | |
5ff9d8a6 EB |
1772 | if (!recurse && has_locked_children(old, old_path.dentry)) |
1773 | goto out2; | |
1774 | ||
ccd48bc7 | 1775 | if (recurse) |
4ce5d2b1 | 1776 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 1777 | else |
87129cc0 | 1778 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1779 | |
be34d1a3 DH |
1780 | if (IS_ERR(mnt)) { |
1781 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 1782 | goto out2; |
be34d1a3 | 1783 | } |
ccd48bc7 | 1784 | |
5ff9d8a6 EB |
1785 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
1786 | ||
84d17192 | 1787 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 1788 | if (err) { |
962830df | 1789 | br_write_lock(&vfsmount_lock); |
328e6d90 | 1790 | umount_tree(mnt, 0); |
962830df | 1791 | br_write_unlock(&vfsmount_lock); |
5b83d2c5 | 1792 | } |
b12cea91 | 1793 | out2: |
84d17192 | 1794 | unlock_mount(mp); |
ccd48bc7 | 1795 | out: |
2d92ab3c | 1796 | path_put(&old_path); |
1da177e4 LT |
1797 | return err; |
1798 | } | |
1799 | ||
2e4b7fcd DH |
1800 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1801 | { | |
1802 | int error = 0; | |
1803 | int readonly_request = 0; | |
1804 | ||
1805 | if (ms_flags & MS_RDONLY) | |
1806 | readonly_request = 1; | |
1807 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1808 | return 0; | |
1809 | ||
90563b19 EB |
1810 | if (mnt->mnt_flags & MNT_LOCK_READONLY) |
1811 | return -EPERM; | |
1812 | ||
2e4b7fcd | 1813 | if (readonly_request) |
83adc753 | 1814 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1815 | else |
83adc753 | 1816 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1817 | return error; |
1818 | } | |
1819 | ||
1da177e4 LT |
1820 | /* |
1821 | * change filesystem flags. dir should be a physical root of filesystem. | |
1822 | * If you've mounted a non-root directory somewhere and want to do remount | |
1823 | * on it - tough luck. | |
1824 | */ | |
0a0d8a46 | 1825 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1826 | void *data) |
1827 | { | |
1828 | int err; | |
2d92ab3c | 1829 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1830 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1831 | |
143c8c91 | 1832 | if (!check_mnt(mnt)) |
1da177e4 LT |
1833 | return -EINVAL; |
1834 | ||
2d92ab3c | 1835 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1836 | return -EINVAL; |
1837 | ||
ff36fe2c EP |
1838 | err = security_sb_remount(sb, data); |
1839 | if (err) | |
1840 | return err; | |
1841 | ||
1da177e4 | 1842 | down_write(&sb->s_umount); |
2e4b7fcd | 1843 | if (flags & MS_BIND) |
2d92ab3c | 1844 | err = change_mount_flags(path->mnt, flags); |
57eccb83 AV |
1845 | else if (!capable(CAP_SYS_ADMIN)) |
1846 | err = -EPERM; | |
4aa98cf7 | 1847 | else |
2e4b7fcd | 1848 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1849 | if (!err) { |
962830df | 1850 | br_write_lock(&vfsmount_lock); |
143c8c91 AV |
1851 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1852 | mnt->mnt.mnt_flags = mnt_flags; | |
962830df | 1853 | br_write_unlock(&vfsmount_lock); |
7b43a79f | 1854 | } |
1da177e4 | 1855 | up_write(&sb->s_umount); |
0e55a7cc | 1856 | if (!err) { |
962830df | 1857 | br_write_lock(&vfsmount_lock); |
143c8c91 | 1858 | touch_mnt_namespace(mnt->mnt_ns); |
962830df | 1859 | br_write_unlock(&vfsmount_lock); |
0e55a7cc | 1860 | } |
1da177e4 LT |
1861 | return err; |
1862 | } | |
1863 | ||
cbbe362c | 1864 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1865 | { |
315fc83e | 1866 | struct mount *p; |
909b0a88 | 1867 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1868 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1869 | return 1; |
1870 | } | |
1871 | return 0; | |
1872 | } | |
1873 | ||
808d4e3c | 1874 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1875 | { |
2d92ab3c | 1876 | struct path old_path, parent_path; |
676da58d | 1877 | struct mount *p; |
0fb54e50 | 1878 | struct mount *old; |
84d17192 | 1879 | struct mountpoint *mp; |
57eccb83 | 1880 | int err; |
1da177e4 LT |
1881 | if (!old_name || !*old_name) |
1882 | return -EINVAL; | |
2d92ab3c | 1883 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1884 | if (err) |
1885 | return err; | |
1886 | ||
84d17192 AV |
1887 | mp = lock_mount(path); |
1888 | err = PTR_ERR(mp); | |
1889 | if (IS_ERR(mp)) | |
cc53ce53 DH |
1890 | goto out; |
1891 | ||
143c8c91 | 1892 | old = real_mount(old_path.mnt); |
fc7be130 | 1893 | p = real_mount(path->mnt); |
143c8c91 | 1894 | |
1da177e4 | 1895 | err = -EINVAL; |
fc7be130 | 1896 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1897 | goto out1; |
1898 | ||
5ff9d8a6 EB |
1899 | if (old->mnt.mnt_flags & MNT_LOCKED) |
1900 | goto out1; | |
1901 | ||
1da177e4 | 1902 | err = -EINVAL; |
2d92ab3c | 1903 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1904 | goto out1; |
1da177e4 | 1905 | |
676da58d | 1906 | if (!mnt_has_parent(old)) |
21444403 | 1907 | goto out1; |
1da177e4 | 1908 | |
2d92ab3c AV |
1909 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1910 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1911 | goto out1; |
1912 | /* | |
1913 | * Don't move a mount residing in a shared parent. | |
1914 | */ | |
fc7be130 | 1915 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1916 | goto out1; |
9676f0c6 RP |
1917 | /* |
1918 | * Don't move a mount tree containing unbindable mounts to a destination | |
1919 | * mount which is shared. | |
1920 | */ | |
fc7be130 | 1921 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1922 | goto out1; |
1da177e4 | 1923 | err = -ELOOP; |
fc7be130 | 1924 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1925 | if (p == old) |
21444403 | 1926 | goto out1; |
1da177e4 | 1927 | |
84d17192 | 1928 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 1929 | if (err) |
21444403 | 1930 | goto out1; |
1da177e4 LT |
1931 | |
1932 | /* if the mount is moved, it should no longer be expire | |
1933 | * automatically */ | |
6776db3d | 1934 | list_del_init(&old->mnt_expire); |
1da177e4 | 1935 | out1: |
84d17192 | 1936 | unlock_mount(mp); |
1da177e4 | 1937 | out: |
1da177e4 | 1938 | if (!err) |
1a390689 | 1939 | path_put(&parent_path); |
2d92ab3c | 1940 | path_put(&old_path); |
1da177e4 LT |
1941 | return err; |
1942 | } | |
1943 | ||
9d412a43 AV |
1944 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1945 | { | |
1946 | int err; | |
1947 | const char *subtype = strchr(fstype, '.'); | |
1948 | if (subtype) { | |
1949 | subtype++; | |
1950 | err = -EINVAL; | |
1951 | if (!subtype[0]) | |
1952 | goto err; | |
1953 | } else | |
1954 | subtype = ""; | |
1955 | ||
1956 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1957 | err = -ENOMEM; | |
1958 | if (!mnt->mnt_sb->s_subtype) | |
1959 | goto err; | |
1960 | return mnt; | |
1961 | ||
1962 | err: | |
1963 | mntput(mnt); | |
1964 | return ERR_PTR(err); | |
1965 | } | |
1966 | ||
9d412a43 AV |
1967 | /* |
1968 | * add a mount into a namespace's mount tree | |
1969 | */ | |
95bc5f25 | 1970 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 1971 | { |
84d17192 AV |
1972 | struct mountpoint *mp; |
1973 | struct mount *parent; | |
9d412a43 AV |
1974 | int err; |
1975 | ||
1976 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1977 | ||
84d17192 AV |
1978 | mp = lock_mount(path); |
1979 | if (IS_ERR(mp)) | |
1980 | return PTR_ERR(mp); | |
9d412a43 | 1981 | |
84d17192 | 1982 | parent = real_mount(path->mnt); |
9d412a43 | 1983 | err = -EINVAL; |
84d17192 | 1984 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
1985 | /* that's acceptable only for automounts done in private ns */ |
1986 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
1987 | goto unlock; | |
1988 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 1989 | if (!parent->mnt_ns) |
156cacb1 AV |
1990 | goto unlock; |
1991 | } | |
9d412a43 AV |
1992 | |
1993 | /* Refuse the same filesystem on the same mount point */ | |
1994 | err = -EBUSY; | |
95bc5f25 | 1995 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1996 | path->mnt->mnt_root == path->dentry) |
1997 | goto unlock; | |
1998 | ||
1999 | err = -EINVAL; | |
95bc5f25 | 2000 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
2001 | goto unlock; |
2002 | ||
95bc5f25 | 2003 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2004 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2005 | |
2006 | unlock: | |
84d17192 | 2007 | unlock_mount(mp); |
9d412a43 AV |
2008 | return err; |
2009 | } | |
b1e75df4 | 2010 | |
1da177e4 LT |
2011 | /* |
2012 | * create a new mount for userspace and request it to be added into the | |
2013 | * namespace's tree | |
2014 | */ | |
0c55cfc4 | 2015 | static int do_new_mount(struct path *path, const char *fstype, int flags, |
808d4e3c | 2016 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2017 | { |
0c55cfc4 | 2018 | struct file_system_type *type; |
9b40bc90 | 2019 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
1da177e4 | 2020 | struct vfsmount *mnt; |
15f9a3f3 | 2021 | int err; |
1da177e4 | 2022 | |
0c55cfc4 | 2023 | if (!fstype) |
1da177e4 LT |
2024 | return -EINVAL; |
2025 | ||
0c55cfc4 EB |
2026 | type = get_fs_type(fstype); |
2027 | if (!type) | |
2028 | return -ENODEV; | |
2029 | ||
2030 | if (user_ns != &init_user_ns) { | |
2031 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2032 | put_filesystem(type); | |
2033 | return -EPERM; | |
2034 | } | |
2035 | /* Only in special cases allow devices from mounts | |
2036 | * created outside the initial user namespace. | |
2037 | */ | |
2038 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2039 | flags |= MS_NODEV; | |
2040 | mnt_flags |= MNT_NODEV; | |
2041 | } | |
2042 | } | |
2043 | ||
2044 | mnt = vfs_kern_mount(type, flags, name, data); | |
2045 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2046 | !mnt->mnt_sb->s_subtype) | |
2047 | mnt = fs_set_subtype(mnt, fstype); | |
2048 | ||
2049 | put_filesystem(type); | |
1da177e4 LT |
2050 | if (IS_ERR(mnt)) |
2051 | return PTR_ERR(mnt); | |
2052 | ||
95bc5f25 | 2053 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2054 | if (err) |
2055 | mntput(mnt); | |
2056 | return err; | |
1da177e4 LT |
2057 | } |
2058 | ||
19a167af AV |
2059 | int finish_automount(struct vfsmount *m, struct path *path) |
2060 | { | |
6776db3d | 2061 | struct mount *mnt = real_mount(m); |
19a167af AV |
2062 | int err; |
2063 | /* The new mount record should have at least 2 refs to prevent it being | |
2064 | * expired before we get a chance to add it | |
2065 | */ | |
6776db3d | 2066 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2067 | |
2068 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2069 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2070 | err = -ELOOP; |
2071 | goto fail; | |
19a167af AV |
2072 | } |
2073 | ||
95bc5f25 | 2074 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2075 | if (!err) |
2076 | return 0; | |
2077 | fail: | |
2078 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2079 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2080 | namespace_lock(); |
962830df | 2081 | br_write_lock(&vfsmount_lock); |
6776db3d | 2082 | list_del_init(&mnt->mnt_expire); |
962830df | 2083 | br_write_unlock(&vfsmount_lock); |
97216be0 | 2084 | namespace_unlock(); |
19a167af | 2085 | } |
b1e75df4 AV |
2086 | mntput(m); |
2087 | mntput(m); | |
19a167af AV |
2088 | return err; |
2089 | } | |
2090 | ||
ea5b778a DH |
2091 | /** |
2092 | * mnt_set_expiry - Put a mount on an expiration list | |
2093 | * @mnt: The mount to list. | |
2094 | * @expiry_list: The list to add the mount to. | |
2095 | */ | |
2096 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2097 | { | |
97216be0 | 2098 | namespace_lock(); |
962830df | 2099 | br_write_lock(&vfsmount_lock); |
ea5b778a | 2100 | |
6776db3d | 2101 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2102 | |
962830df | 2103 | br_write_unlock(&vfsmount_lock); |
97216be0 | 2104 | namespace_unlock(); |
ea5b778a DH |
2105 | } |
2106 | EXPORT_SYMBOL(mnt_set_expiry); | |
2107 | ||
1da177e4 LT |
2108 | /* |
2109 | * process a list of expirable mountpoints with the intent of discarding any | |
2110 | * mountpoints that aren't in use and haven't been touched since last we came | |
2111 | * here | |
2112 | */ | |
2113 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2114 | { | |
761d5c38 | 2115 | struct mount *mnt, *next; |
1da177e4 LT |
2116 | LIST_HEAD(graveyard); |
2117 | ||
2118 | if (list_empty(mounts)) | |
2119 | return; | |
2120 | ||
97216be0 | 2121 | namespace_lock(); |
962830df | 2122 | br_write_lock(&vfsmount_lock); |
1da177e4 LT |
2123 | |
2124 | /* extract from the expiration list every vfsmount that matches the | |
2125 | * following criteria: | |
2126 | * - only referenced by its parent vfsmount | |
2127 | * - still marked for expiry (marked on the last call here; marks are | |
2128 | * cleared by mntput()) | |
2129 | */ | |
6776db3d | 2130 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2131 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2132 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2133 | continue; |
6776db3d | 2134 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2135 | } |
bcc5c7d2 | 2136 | while (!list_empty(&graveyard)) { |
6776db3d | 2137 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2138 | touch_mnt_namespace(mnt->mnt_ns); |
328e6d90 | 2139 | umount_tree(mnt, 1); |
bcc5c7d2 | 2140 | } |
962830df | 2141 | br_write_unlock(&vfsmount_lock); |
3ab6abee | 2142 | namespace_unlock(); |
5528f911 TM |
2143 | } |
2144 | ||
2145 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2146 | ||
2147 | /* | |
2148 | * Ripoff of 'select_parent()' | |
2149 | * | |
2150 | * search the list of submounts for a given mountpoint, and move any | |
2151 | * shrinkable submounts to the 'graveyard' list. | |
2152 | */ | |
692afc31 | 2153 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2154 | { |
692afc31 | 2155 | struct mount *this_parent = parent; |
5528f911 TM |
2156 | struct list_head *next; |
2157 | int found = 0; | |
2158 | ||
2159 | repeat: | |
6b41d536 | 2160 | next = this_parent->mnt_mounts.next; |
5528f911 | 2161 | resume: |
6b41d536 | 2162 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2163 | struct list_head *tmp = next; |
6b41d536 | 2164 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2165 | |
2166 | next = tmp->next; | |
692afc31 | 2167 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2168 | continue; |
5528f911 TM |
2169 | /* |
2170 | * Descend a level if the d_mounts list is non-empty. | |
2171 | */ | |
6b41d536 | 2172 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2173 | this_parent = mnt; |
2174 | goto repeat; | |
2175 | } | |
1da177e4 | 2176 | |
1ab59738 | 2177 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2178 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2179 | found++; |
2180 | } | |
1da177e4 | 2181 | } |
5528f911 TM |
2182 | /* |
2183 | * All done at this level ... ascend and resume the search | |
2184 | */ | |
2185 | if (this_parent != parent) { | |
6b41d536 | 2186 | next = this_parent->mnt_child.next; |
0714a533 | 2187 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2188 | goto resume; |
2189 | } | |
2190 | return found; | |
2191 | } | |
2192 | ||
2193 | /* | |
2194 | * process a list of expirable mountpoints with the intent of discarding any | |
2195 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
2196 | * |
2197 | * vfsmount_lock must be held for write | |
5528f911 | 2198 | */ |
b54b9be7 | 2199 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2200 | { |
2201 | LIST_HEAD(graveyard); | |
761d5c38 | 2202 | struct mount *m; |
5528f911 | 2203 | |
5528f911 | 2204 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2205 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2206 | while (!list_empty(&graveyard)) { |
761d5c38 | 2207 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2208 | mnt_expire); |
143c8c91 | 2209 | touch_mnt_namespace(m->mnt_ns); |
328e6d90 | 2210 | umount_tree(m, 1); |
bcc5c7d2 AV |
2211 | } |
2212 | } | |
1da177e4 LT |
2213 | } |
2214 | ||
1da177e4 LT |
2215 | /* |
2216 | * Some copy_from_user() implementations do not return the exact number of | |
2217 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2218 | * Note that this function differs from copy_from_user() in that it will oops | |
2219 | * on bad values of `to', rather than returning a short copy. | |
2220 | */ | |
b58fed8b RP |
2221 | static long exact_copy_from_user(void *to, const void __user * from, |
2222 | unsigned long n) | |
1da177e4 LT |
2223 | { |
2224 | char *t = to; | |
2225 | const char __user *f = from; | |
2226 | char c; | |
2227 | ||
2228 | if (!access_ok(VERIFY_READ, from, n)) | |
2229 | return n; | |
2230 | ||
2231 | while (n) { | |
2232 | if (__get_user(c, f)) { | |
2233 | memset(t, 0, n); | |
2234 | break; | |
2235 | } | |
2236 | *t++ = c; | |
2237 | f++; | |
2238 | n--; | |
2239 | } | |
2240 | return n; | |
2241 | } | |
2242 | ||
b58fed8b | 2243 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2244 | { |
2245 | int i; | |
2246 | unsigned long page; | |
2247 | unsigned long size; | |
b58fed8b | 2248 | |
1da177e4 LT |
2249 | *where = 0; |
2250 | if (!data) | |
2251 | return 0; | |
2252 | ||
2253 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2254 | return -ENOMEM; | |
2255 | ||
2256 | /* We only care that *some* data at the address the user | |
2257 | * gave us is valid. Just in case, we'll zero | |
2258 | * the remainder of the page. | |
2259 | */ | |
2260 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2261 | size = TASK_SIZE - (unsigned long)data; | |
2262 | if (size > PAGE_SIZE) | |
2263 | size = PAGE_SIZE; | |
2264 | ||
2265 | i = size - exact_copy_from_user((void *)page, data, size); | |
2266 | if (!i) { | |
b58fed8b | 2267 | free_page(page); |
1da177e4 LT |
2268 | return -EFAULT; |
2269 | } | |
2270 | if (i != PAGE_SIZE) | |
2271 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2272 | *where = page; | |
2273 | return 0; | |
2274 | } | |
2275 | ||
eca6f534 VN |
2276 | int copy_mount_string(const void __user *data, char **where) |
2277 | { | |
2278 | char *tmp; | |
2279 | ||
2280 | if (!data) { | |
2281 | *where = NULL; | |
2282 | return 0; | |
2283 | } | |
2284 | ||
2285 | tmp = strndup_user(data, PAGE_SIZE); | |
2286 | if (IS_ERR(tmp)) | |
2287 | return PTR_ERR(tmp); | |
2288 | ||
2289 | *where = tmp; | |
2290 | return 0; | |
2291 | } | |
2292 | ||
1da177e4 LT |
2293 | /* |
2294 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2295 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2296 | * | |
2297 | * data is a (void *) that can point to any structure up to | |
2298 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2299 | * information (or be NULL). | |
2300 | * | |
2301 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2302 | * When the flags word was introduced its top half was required | |
2303 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2304 | * Therefore, if this magic number is present, it carries no information | |
2305 | * and must be discarded. | |
2306 | */ | |
808d4e3c AV |
2307 | long do_mount(const char *dev_name, const char *dir_name, |
2308 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2309 | { |
2d92ab3c | 2310 | struct path path; |
1da177e4 LT |
2311 | int retval = 0; |
2312 | int mnt_flags = 0; | |
2313 | ||
2314 | /* Discard magic */ | |
2315 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2316 | flags &= ~MS_MGC_MSK; | |
2317 | ||
2318 | /* Basic sanity checks */ | |
2319 | ||
2320 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2321 | return -EINVAL; | |
1da177e4 LT |
2322 | |
2323 | if (data_page) | |
2324 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2325 | ||
a27ab9f2 TH |
2326 | /* ... and get the mountpoint */ |
2327 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2328 | if (retval) | |
2329 | return retval; | |
2330 | ||
2331 | retval = security_sb_mount(dev_name, &path, | |
2332 | type_page, flags, data_page); | |
0d5cadb8 AV |
2333 | if (!retval && !may_mount()) |
2334 | retval = -EPERM; | |
a27ab9f2 TH |
2335 | if (retval) |
2336 | goto dput_out; | |
2337 | ||
613cbe3d AK |
2338 | /* Default to relatime unless overriden */ |
2339 | if (!(flags & MS_NOATIME)) | |
2340 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2341 | |
1da177e4 LT |
2342 | /* Separate the per-mountpoint flags */ |
2343 | if (flags & MS_NOSUID) | |
2344 | mnt_flags |= MNT_NOSUID; | |
2345 | if (flags & MS_NODEV) | |
2346 | mnt_flags |= MNT_NODEV; | |
2347 | if (flags & MS_NOEXEC) | |
2348 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2349 | if (flags & MS_NOATIME) |
2350 | mnt_flags |= MNT_NOATIME; | |
2351 | if (flags & MS_NODIRATIME) | |
2352 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2353 | if (flags & MS_STRICTATIME) |
2354 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2355 | if (flags & MS_RDONLY) |
2356 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2357 | |
7a4dec53 | 2358 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2359 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2360 | MS_STRICTATIME); | |
1da177e4 | 2361 | |
1da177e4 | 2362 | if (flags & MS_REMOUNT) |
2d92ab3c | 2363 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2364 | data_page); |
2365 | else if (flags & MS_BIND) | |
2d92ab3c | 2366 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2367 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2368 | retval = do_change_type(&path, flags); |
1da177e4 | 2369 | else if (flags & MS_MOVE) |
2d92ab3c | 2370 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2371 | else |
2d92ab3c | 2372 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2373 | dev_name, data_page); |
2374 | dput_out: | |
2d92ab3c | 2375 | path_put(&path); |
1da177e4 LT |
2376 | return retval; |
2377 | } | |
2378 | ||
771b1371 EB |
2379 | static void free_mnt_ns(struct mnt_namespace *ns) |
2380 | { | |
98f842e6 | 2381 | proc_free_inum(ns->proc_inum); |
771b1371 EB |
2382 | put_user_ns(ns->user_ns); |
2383 | kfree(ns); | |
2384 | } | |
2385 | ||
8823c079 EB |
2386 | /* |
2387 | * Assign a sequence number so we can detect when we attempt to bind | |
2388 | * mount a reference to an older mount namespace into the current | |
2389 | * mount namespace, preventing reference counting loops. A 64bit | |
2390 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2391 | * is effectively never, so we can ignore the possibility. | |
2392 | */ | |
2393 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2394 | ||
771b1371 | 2395 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2396 | { |
2397 | struct mnt_namespace *new_ns; | |
98f842e6 | 2398 | int ret; |
cf8d2c11 TM |
2399 | |
2400 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2401 | if (!new_ns) | |
2402 | return ERR_PTR(-ENOMEM); | |
98f842e6 EB |
2403 | ret = proc_alloc_inum(&new_ns->proc_inum); |
2404 | if (ret) { | |
2405 | kfree(new_ns); | |
2406 | return ERR_PTR(ret); | |
2407 | } | |
8823c079 | 2408 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2409 | atomic_set(&new_ns->count, 1); |
2410 | new_ns->root = NULL; | |
2411 | INIT_LIST_HEAD(&new_ns->list); | |
2412 | init_waitqueue_head(&new_ns->poll); | |
2413 | new_ns->event = 0; | |
771b1371 | 2414 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2415 | return new_ns; |
2416 | } | |
2417 | ||
741a2951 JD |
2418 | /* |
2419 | * Allocate a new namespace structure and populate it with contents | |
2420 | * copied from the namespace of the passed in task structure. | |
2421 | */ | |
e3222c4e | 2422 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
771b1371 | 2423 | struct user_namespace *user_ns, struct fs_struct *fs) |
1da177e4 | 2424 | { |
6b3286ed | 2425 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2426 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2427 | struct mount *p, *q; |
be08d6d2 | 2428 | struct mount *old = mnt_ns->root; |
cb338d06 | 2429 | struct mount *new; |
7a472ef4 | 2430 | int copy_flags; |
1da177e4 | 2431 | |
771b1371 | 2432 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2433 | if (IS_ERR(new_ns)) |
2434 | return new_ns; | |
1da177e4 | 2435 | |
97216be0 | 2436 | namespace_lock(); |
1da177e4 | 2437 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2438 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
7a472ef4 | 2439 | if (user_ns != mnt_ns->user_ns) |
132c94e3 | 2440 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2441 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2442 | if (IS_ERR(new)) { |
328e6d90 | 2443 | namespace_unlock(); |
771b1371 | 2444 | free_mnt_ns(new_ns); |
be34d1a3 | 2445 | return ERR_CAST(new); |
1da177e4 | 2446 | } |
be08d6d2 | 2447 | new_ns->root = new; |
962830df | 2448 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 2449 | list_add_tail(&new_ns->list, &new->mnt_list); |
962830df | 2450 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
2451 | |
2452 | /* | |
2453 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2454 | * as belonging to new namespace. We have already acquired a private | |
2455 | * fs_struct, so tsk->fs->lock is not needed. | |
2456 | */ | |
909b0a88 | 2457 | p = old; |
cb338d06 | 2458 | q = new; |
1da177e4 | 2459 | while (p) { |
143c8c91 | 2460 | q->mnt_ns = new_ns; |
1da177e4 | 2461 | if (fs) { |
315fc83e AV |
2462 | if (&p->mnt == fs->root.mnt) { |
2463 | fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2464 | rootmnt = &p->mnt; |
1da177e4 | 2465 | } |
315fc83e AV |
2466 | if (&p->mnt == fs->pwd.mnt) { |
2467 | fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2468 | pwdmnt = &p->mnt; |
1da177e4 | 2469 | } |
1da177e4 | 2470 | } |
909b0a88 AV |
2471 | p = next_mnt(p, old); |
2472 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2473 | if (!q) |
2474 | break; | |
2475 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2476 | p = next_mnt(p, old); | |
1da177e4 | 2477 | } |
328e6d90 | 2478 | namespace_unlock(); |
1da177e4 | 2479 | |
1da177e4 | 2480 | if (rootmnt) |
f03c6599 | 2481 | mntput(rootmnt); |
1da177e4 | 2482 | if (pwdmnt) |
f03c6599 | 2483 | mntput(pwdmnt); |
1da177e4 | 2484 | |
741a2951 JD |
2485 | return new_ns; |
2486 | } | |
2487 | ||
213dd266 | 2488 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
771b1371 | 2489 | struct user_namespace *user_ns, struct fs_struct *new_fs) |
741a2951 | 2490 | { |
6b3286ed | 2491 | struct mnt_namespace *new_ns; |
741a2951 | 2492 | |
e3222c4e | 2493 | BUG_ON(!ns); |
6b3286ed | 2494 | get_mnt_ns(ns); |
741a2951 JD |
2495 | |
2496 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2497 | return ns; |
741a2951 | 2498 | |
771b1371 | 2499 | new_ns = dup_mnt_ns(ns, user_ns, new_fs); |
741a2951 | 2500 | |
6b3286ed | 2501 | put_mnt_ns(ns); |
e3222c4e | 2502 | return new_ns; |
1da177e4 LT |
2503 | } |
2504 | ||
cf8d2c11 TM |
2505 | /** |
2506 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2507 | * @mnt: pointer to the new root filesystem mountpoint | |
2508 | */ | |
1a4eeaf2 | 2509 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2510 | { |
771b1371 | 2511 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2512 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2513 | struct mount *mnt = real_mount(m); |
2514 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2515 | new_ns->root = mnt; |
b1983cd8 | 2516 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2517 | } else { |
1a4eeaf2 | 2518 | mntput(m); |
cf8d2c11 TM |
2519 | } |
2520 | return new_ns; | |
2521 | } | |
cf8d2c11 | 2522 | |
ea441d11 AV |
2523 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2524 | { | |
2525 | struct mnt_namespace *ns; | |
d31da0f0 | 2526 | struct super_block *s; |
ea441d11 AV |
2527 | struct path path; |
2528 | int err; | |
2529 | ||
2530 | ns = create_mnt_ns(mnt); | |
2531 | if (IS_ERR(ns)) | |
2532 | return ERR_CAST(ns); | |
2533 | ||
2534 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2535 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2536 | ||
2537 | put_mnt_ns(ns); | |
2538 | ||
2539 | if (err) | |
2540 | return ERR_PTR(err); | |
2541 | ||
2542 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2543 | s = path.mnt->mnt_sb; |
2544 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2545 | mntput(path.mnt); |
2546 | /* lock the sucker */ | |
d31da0f0 | 2547 | down_write(&s->s_umount); |
ea441d11 AV |
2548 | /* ... and return the root of (sub)tree on it */ |
2549 | return path.dentry; | |
2550 | } | |
2551 | EXPORT_SYMBOL(mount_subtree); | |
2552 | ||
bdc480e3 HC |
2553 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2554 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2555 | { |
eca6f534 VN |
2556 | int ret; |
2557 | char *kernel_type; | |
91a27b2a | 2558 | struct filename *kernel_dir; |
eca6f534 | 2559 | char *kernel_dev; |
1da177e4 | 2560 | unsigned long data_page; |
1da177e4 | 2561 | |
eca6f534 VN |
2562 | ret = copy_mount_string(type, &kernel_type); |
2563 | if (ret < 0) | |
2564 | goto out_type; | |
1da177e4 | 2565 | |
eca6f534 VN |
2566 | kernel_dir = getname(dir_name); |
2567 | if (IS_ERR(kernel_dir)) { | |
2568 | ret = PTR_ERR(kernel_dir); | |
2569 | goto out_dir; | |
2570 | } | |
1da177e4 | 2571 | |
eca6f534 VN |
2572 | ret = copy_mount_string(dev_name, &kernel_dev); |
2573 | if (ret < 0) | |
2574 | goto out_dev; | |
1da177e4 | 2575 | |
eca6f534 VN |
2576 | ret = copy_mount_options(data, &data_page); |
2577 | if (ret < 0) | |
2578 | goto out_data; | |
1da177e4 | 2579 | |
91a27b2a | 2580 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2581 | (void *) data_page); |
1da177e4 | 2582 | |
eca6f534 VN |
2583 | free_page(data_page); |
2584 | out_data: | |
2585 | kfree(kernel_dev); | |
2586 | out_dev: | |
2587 | putname(kernel_dir); | |
2588 | out_dir: | |
2589 | kfree(kernel_type); | |
2590 | out_type: | |
2591 | return ret; | |
1da177e4 LT |
2592 | } |
2593 | ||
afac7cba AV |
2594 | /* |
2595 | * Return true if path is reachable from root | |
2596 | * | |
2597 | * namespace_sem or vfsmount_lock is held | |
2598 | */ | |
643822b4 | 2599 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2600 | const struct path *root) |
2601 | { | |
643822b4 | 2602 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2603 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2604 | mnt = mnt->mnt_parent; |
afac7cba | 2605 | } |
643822b4 | 2606 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2607 | } |
2608 | ||
2609 | int path_is_under(struct path *path1, struct path *path2) | |
2610 | { | |
2611 | int res; | |
962830df | 2612 | br_read_lock(&vfsmount_lock); |
643822b4 | 2613 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
962830df | 2614 | br_read_unlock(&vfsmount_lock); |
afac7cba AV |
2615 | return res; |
2616 | } | |
2617 | EXPORT_SYMBOL(path_is_under); | |
2618 | ||
1da177e4 LT |
2619 | /* |
2620 | * pivot_root Semantics: | |
2621 | * Moves the root file system of the current process to the directory put_old, | |
2622 | * makes new_root as the new root file system of the current process, and sets | |
2623 | * root/cwd of all processes which had them on the current root to new_root. | |
2624 | * | |
2625 | * Restrictions: | |
2626 | * The new_root and put_old must be directories, and must not be on the | |
2627 | * same file system as the current process root. The put_old must be | |
2628 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2629 | * pointed to by put_old must yield the same directory as new_root. No other | |
2630 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2631 | * | |
4a0d11fa NB |
2632 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2633 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2634 | * in this situation. | |
2635 | * | |
1da177e4 LT |
2636 | * Notes: |
2637 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2638 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2639 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2640 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2641 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2642 | * first. | |
2643 | */ | |
3480b257 HC |
2644 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2645 | const char __user *, put_old) | |
1da177e4 | 2646 | { |
2d8f3038 | 2647 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
2648 | struct mount *new_mnt, *root_mnt, *old_mnt; |
2649 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
2650 | int error; |
2651 | ||
9b40bc90 | 2652 | if (!may_mount()) |
1da177e4 LT |
2653 | return -EPERM; |
2654 | ||
2d8f3038 | 2655 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2656 | if (error) |
2657 | goto out0; | |
1da177e4 | 2658 | |
2d8f3038 | 2659 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2660 | if (error) |
2661 | goto out1; | |
2662 | ||
2d8f3038 | 2663 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2664 | if (error) |
2665 | goto out2; | |
1da177e4 | 2666 | |
f7ad3c6b | 2667 | get_fs_root(current->fs, &root); |
84d17192 AV |
2668 | old_mp = lock_mount(&old); |
2669 | error = PTR_ERR(old_mp); | |
2670 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
2671 | goto out3; |
2672 | ||
1da177e4 | 2673 | error = -EINVAL; |
419148da AV |
2674 | new_mnt = real_mount(new.mnt); |
2675 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
2676 | old_mnt = real_mount(old.mnt); |
2677 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
2678 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
2679 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2680 | goto out4; |
143c8c91 | 2681 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2682 | goto out4; |
5ff9d8a6 EB |
2683 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
2684 | goto out4; | |
1da177e4 | 2685 | error = -ENOENT; |
f3da392e | 2686 | if (d_unlinked(new.dentry)) |
b12cea91 | 2687 | goto out4; |
1da177e4 | 2688 | error = -EBUSY; |
84d17192 | 2689 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 2690 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2691 | error = -EINVAL; |
8c3ee42e | 2692 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2693 | goto out4; /* not a mountpoint */ |
676da58d | 2694 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2695 | goto out4; /* not attached */ |
84d17192 | 2696 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 2697 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2698 | goto out4; /* not a mountpoint */ |
676da58d | 2699 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2700 | goto out4; /* not attached */ |
4ac91378 | 2701 | /* make sure we can reach put_old from new_root */ |
84d17192 | 2702 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 2703 | goto out4; |
84d17192 | 2704 | root_mp->m_count++; /* pin it so it won't go away */ |
962830df | 2705 | br_write_lock(&vfsmount_lock); |
419148da AV |
2706 | detach_mnt(new_mnt, &parent_path); |
2707 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
2708 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
2709 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2710 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2711 | } | |
4ac91378 | 2712 | /* mount old root on put_old */ |
84d17192 | 2713 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 2714 | /* mount new_root on / */ |
84d17192 | 2715 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 2716 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
962830df | 2717 | br_write_unlock(&vfsmount_lock); |
2d8f3038 | 2718 | chroot_fs_refs(&root, &new); |
84d17192 | 2719 | put_mountpoint(root_mp); |
1da177e4 | 2720 | error = 0; |
b12cea91 | 2721 | out4: |
84d17192 | 2722 | unlock_mount(old_mp); |
b12cea91 AV |
2723 | if (!error) { |
2724 | path_put(&root_parent); | |
2725 | path_put(&parent_path); | |
2726 | } | |
2727 | out3: | |
8c3ee42e | 2728 | path_put(&root); |
b12cea91 | 2729 | out2: |
2d8f3038 | 2730 | path_put(&old); |
1da177e4 | 2731 | out1: |
2d8f3038 | 2732 | path_put(&new); |
1da177e4 | 2733 | out0: |
1da177e4 | 2734 | return error; |
1da177e4 LT |
2735 | } |
2736 | ||
2737 | static void __init init_mount_tree(void) | |
2738 | { | |
2739 | struct vfsmount *mnt; | |
6b3286ed | 2740 | struct mnt_namespace *ns; |
ac748a09 | 2741 | struct path root; |
0c55cfc4 | 2742 | struct file_system_type *type; |
1da177e4 | 2743 | |
0c55cfc4 EB |
2744 | type = get_fs_type("rootfs"); |
2745 | if (!type) | |
2746 | panic("Can't find rootfs type"); | |
2747 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
2748 | put_filesystem(type); | |
1da177e4 LT |
2749 | if (IS_ERR(mnt)) |
2750 | panic("Can't create rootfs"); | |
b3e19d92 | 2751 | |
3b22edc5 TM |
2752 | ns = create_mnt_ns(mnt); |
2753 | if (IS_ERR(ns)) | |
1da177e4 | 2754 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2755 | |
2756 | init_task.nsproxy->mnt_ns = ns; | |
2757 | get_mnt_ns(ns); | |
2758 | ||
be08d6d2 AV |
2759 | root.mnt = mnt; |
2760 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2761 | |
2762 | set_fs_pwd(current->fs, &root); | |
2763 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2764 | } |
2765 | ||
74bf17cf | 2766 | void __init mnt_init(void) |
1da177e4 | 2767 | { |
13f14b4d | 2768 | unsigned u; |
15a67dd8 | 2769 | int err; |
1da177e4 | 2770 | |
390c6843 RP |
2771 | init_rwsem(&namespace_sem); |
2772 | ||
7d6fec45 | 2773 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2774 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2775 | |
b58fed8b | 2776 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
84d17192 | 2777 | mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 | 2778 | |
84d17192 | 2779 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
2780 | panic("Failed to allocate mount hash table\n"); |
2781 | ||
80cdc6da | 2782 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2783 | |
2784 | for (u = 0; u < HASH_SIZE; u++) | |
2785 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
84d17192 AV |
2786 | for (u = 0; u < HASH_SIZE; u++) |
2787 | INIT_LIST_HEAD(&mountpoint_hashtable[u]); | |
1da177e4 | 2788 | |
962830df | 2789 | br_lock_init(&vfsmount_lock); |
99b7db7b | 2790 | |
15a67dd8 RD |
2791 | err = sysfs_init(); |
2792 | if (err) | |
2793 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2794 | __func__, err); |
00d26666 GKH |
2795 | fs_kobj = kobject_create_and_add("fs", NULL); |
2796 | if (!fs_kobj) | |
8e24eea7 | 2797 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2798 | init_rootfs(); |
2799 | init_mount_tree(); | |
2800 | } | |
2801 | ||
616511d0 | 2802 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2803 | { |
d498b25a | 2804 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2805 | return; |
97216be0 | 2806 | namespace_lock(); |
962830df | 2807 | br_write_lock(&vfsmount_lock); |
328e6d90 | 2808 | umount_tree(ns->root, 0); |
962830df | 2809 | br_write_unlock(&vfsmount_lock); |
3ab6abee | 2810 | namespace_unlock(); |
771b1371 | 2811 | free_mnt_ns(ns); |
1da177e4 | 2812 | } |
9d412a43 AV |
2813 | |
2814 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2815 | { | |
423e0ab0 TC |
2816 | struct vfsmount *mnt; |
2817 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2818 | if (!IS_ERR(mnt)) { | |
2819 | /* | |
2820 | * it is a longterm mount, don't release mnt until | |
2821 | * we unmount before file sys is unregistered | |
2822 | */ | |
f7a99c5b | 2823 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2824 | } |
2825 | return mnt; | |
9d412a43 AV |
2826 | } |
2827 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2828 | |
2829 | void kern_unmount(struct vfsmount *mnt) | |
2830 | { | |
2831 | /* release long term mount so mount point can be released */ | |
2832 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b AV |
2833 | br_write_lock(&vfsmount_lock); |
2834 | real_mount(mnt)->mnt_ns = NULL; | |
2835 | br_write_unlock(&vfsmount_lock); | |
423e0ab0 TC |
2836 | mntput(mnt); |
2837 | } | |
2838 | } | |
2839 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2840 | |
2841 | bool our_mnt(struct vfsmount *mnt) | |
2842 | { | |
143c8c91 | 2843 | return check_mnt(real_mount(mnt)); |
02125a82 | 2844 | } |
8823c079 | 2845 | |
3151527e EB |
2846 | bool current_chrooted(void) |
2847 | { | |
2848 | /* Does the current process have a non-standard root */ | |
2849 | struct path ns_root; | |
2850 | struct path fs_root; | |
2851 | bool chrooted; | |
2852 | ||
2853 | /* Find the namespace root */ | |
2854 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
2855 | ns_root.dentry = ns_root.mnt->mnt_root; | |
2856 | path_get(&ns_root); | |
2857 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
2858 | ; | |
2859 | ||
2860 | get_fs_root(current->fs, &fs_root); | |
2861 | ||
2862 | chrooted = !path_equal(&fs_root, &ns_root); | |
2863 | ||
2864 | path_put(&fs_root); | |
2865 | path_put(&ns_root); | |
2866 | ||
2867 | return chrooted; | |
2868 | } | |
2869 | ||
87a8ebd6 EB |
2870 | void update_mnt_policy(struct user_namespace *userns) |
2871 | { | |
2872 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
2873 | struct mount *mnt; | |
2874 | ||
2875 | down_read(&namespace_sem); | |
2876 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
2877 | switch (mnt->mnt.mnt_sb->s_magic) { | |
2878 | case SYSFS_MAGIC: | |
2879 | userns->may_mount_sysfs = true; | |
2880 | break; | |
2881 | case PROC_SUPER_MAGIC: | |
2882 | userns->may_mount_proc = true; | |
2883 | break; | |
2884 | } | |
2885 | if (userns->may_mount_sysfs && userns->may_mount_proc) | |
2886 | break; | |
2887 | } | |
2888 | up_read(&namespace_sem); | |
2889 | } | |
2890 | ||
8823c079 EB |
2891 | static void *mntns_get(struct task_struct *task) |
2892 | { | |
2893 | struct mnt_namespace *ns = NULL; | |
2894 | struct nsproxy *nsproxy; | |
2895 | ||
2896 | rcu_read_lock(); | |
2897 | nsproxy = task_nsproxy(task); | |
2898 | if (nsproxy) { | |
2899 | ns = nsproxy->mnt_ns; | |
2900 | get_mnt_ns(ns); | |
2901 | } | |
2902 | rcu_read_unlock(); | |
2903 | ||
2904 | return ns; | |
2905 | } | |
2906 | ||
2907 | static void mntns_put(void *ns) | |
2908 | { | |
2909 | put_mnt_ns(ns); | |
2910 | } | |
2911 | ||
2912 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2913 | { | |
2914 | struct fs_struct *fs = current->fs; | |
2915 | struct mnt_namespace *mnt_ns = ns; | |
2916 | struct path root; | |
2917 | ||
0c55cfc4 | 2918 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
5e4a0847 EB |
2919 | !nsown_capable(CAP_SYS_CHROOT) || |
2920 | !nsown_capable(CAP_SYS_ADMIN)) | |
ae11e0f1 | 2921 | return -EPERM; |
8823c079 EB |
2922 | |
2923 | if (fs->users != 1) | |
2924 | return -EINVAL; | |
2925 | ||
2926 | get_mnt_ns(mnt_ns); | |
2927 | put_mnt_ns(nsproxy->mnt_ns); | |
2928 | nsproxy->mnt_ns = mnt_ns; | |
2929 | ||
2930 | /* Find the root */ | |
2931 | root.mnt = &mnt_ns->root->mnt; | |
2932 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2933 | path_get(&root); | |
2934 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
2935 | ; | |
2936 | ||
2937 | /* Update the pwd and root */ | |
2938 | set_fs_pwd(fs, &root); | |
2939 | set_fs_root(fs, &root); | |
2940 | ||
2941 | path_put(&root); | |
2942 | return 0; | |
2943 | } | |
2944 | ||
98f842e6 EB |
2945 | static unsigned int mntns_inum(void *ns) |
2946 | { | |
2947 | struct mnt_namespace *mnt_ns = ns; | |
2948 | return mnt_ns->proc_inum; | |
2949 | } | |
2950 | ||
8823c079 EB |
2951 | const struct proc_ns_operations mntns_operations = { |
2952 | .name = "mnt", | |
2953 | .type = CLONE_NEWNS, | |
2954 | .get = mntns_get, | |
2955 | .put = mntns_put, | |
2956 | .install = mntns_install, | |
98f842e6 | 2957 | .inum = mntns_inum, |
8823c079 | 2958 | }; |