]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
73cd49ec | 18 | #include <linux/idr.h> |
d10577a8 AV |
19 | #include <linux/acct.h> /* acct_auto_close_mnt */ |
20 | #include <linux/ramfs.h> /* init_rootfs */ | |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ | |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
8823c079 | 24 | #include <linux/proc_fs.h> |
07b20889 | 25 | #include "pnode.h" |
948730b0 | 26 | #include "internal.h" |
1da177e4 | 27 | |
13f14b4d ED |
28 | #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) |
29 | #define HASH_SIZE (1UL << HASH_SHIFT) | |
30 | ||
5addc5dd | 31 | static int event; |
73cd49ec | 32 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 33 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 34 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
35 | static int mnt_id_start = 0; |
36 | static int mnt_group_start = 1; | |
1da177e4 | 37 | |
fa3536cc | 38 | static struct list_head *mount_hashtable __read_mostly; |
e18b890b | 39 | static struct kmem_cache *mnt_cache __read_mostly; |
390c6843 | 40 | static struct rw_semaphore namespace_sem; |
1da177e4 | 41 | |
f87fd4c2 | 42 | /* /sys/fs */ |
00d26666 GKH |
43 | struct kobject *fs_kobj; |
44 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 45 | |
99b7db7b NP |
46 | /* |
47 | * vfsmount lock may be taken for read to prevent changes to the | |
48 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
49 | * up the tree. | |
50 | * | |
51 | * It should be taken for write in all cases where the vfsmount | |
52 | * tree or hash is modified or when a vfsmount structure is modified. | |
53 | */ | |
54 | DEFINE_BRLOCK(vfsmount_lock); | |
55 | ||
1da177e4 LT |
56 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
57 | { | |
b58fed8b RP |
58 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
59 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
13f14b4d ED |
60 | tmp = tmp + (tmp >> HASH_SHIFT); |
61 | return tmp & (HASH_SIZE - 1); | |
1da177e4 LT |
62 | } |
63 | ||
3d733633 DH |
64 | #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) |
65 | ||
99b7db7b NP |
66 | /* |
67 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
68 | * serialize with freeing. | |
69 | */ | |
b105e270 | 70 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec MS |
71 | { |
72 | int res; | |
73 | ||
74 | retry: | |
75 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 76 | spin_lock(&mnt_id_lock); |
15169fe7 | 77 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 78 | if (!res) |
15169fe7 | 79 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 80 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
81 | if (res == -EAGAIN) |
82 | goto retry; | |
83 | ||
84 | return res; | |
85 | } | |
86 | ||
b105e270 | 87 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 88 | { |
15169fe7 | 89 | int id = mnt->mnt_id; |
99b7db7b | 90 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
91 | ida_remove(&mnt_id_ida, id); |
92 | if (mnt_id_start > id) | |
93 | mnt_id_start = id; | |
99b7db7b | 94 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
95 | } |
96 | ||
719f5d7f MS |
97 | /* |
98 | * Allocate a new peer group ID | |
99 | * | |
100 | * mnt_group_ida is protected by namespace_sem | |
101 | */ | |
4b8b21f4 | 102 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 103 | { |
f21f6220 AV |
104 | int res; |
105 | ||
719f5d7f MS |
106 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
107 | return -ENOMEM; | |
108 | ||
f21f6220 AV |
109 | res = ida_get_new_above(&mnt_group_ida, |
110 | mnt_group_start, | |
15169fe7 | 111 | &mnt->mnt_group_id); |
f21f6220 | 112 | if (!res) |
15169fe7 | 113 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
114 | |
115 | return res; | |
719f5d7f MS |
116 | } |
117 | ||
118 | /* | |
119 | * Release a peer group ID | |
120 | */ | |
4b8b21f4 | 121 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 122 | { |
15169fe7 | 123 | int id = mnt->mnt_group_id; |
f21f6220 AV |
124 | ida_remove(&mnt_group_ida, id); |
125 | if (mnt_group_start > id) | |
126 | mnt_group_start = id; | |
15169fe7 | 127 | mnt->mnt_group_id = 0; |
719f5d7f MS |
128 | } |
129 | ||
b3e19d92 NP |
130 | /* |
131 | * vfsmount lock must be held for read | |
132 | */ | |
83adc753 | 133 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
134 | { |
135 | #ifdef CONFIG_SMP | |
68e8a9fe | 136 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
137 | #else |
138 | preempt_disable(); | |
68e8a9fe | 139 | mnt->mnt_count += n; |
b3e19d92 NP |
140 | preempt_enable(); |
141 | #endif | |
142 | } | |
143 | ||
b3e19d92 NP |
144 | /* |
145 | * vfsmount lock must be held for write | |
146 | */ | |
83adc753 | 147 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
148 | { |
149 | #ifdef CONFIG_SMP | |
f03c6599 | 150 | unsigned int count = 0; |
b3e19d92 NP |
151 | int cpu; |
152 | ||
153 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 154 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
155 | } |
156 | ||
157 | return count; | |
158 | #else | |
68e8a9fe | 159 | return mnt->mnt_count; |
b3e19d92 NP |
160 | #endif |
161 | } | |
162 | ||
b105e270 | 163 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 164 | { |
c63181e6 AV |
165 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
166 | if (mnt) { | |
73cd49ec MS |
167 | int err; |
168 | ||
c63181e6 | 169 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
170 | if (err) |
171 | goto out_free_cache; | |
172 | ||
173 | if (name) { | |
c63181e6 AV |
174 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); |
175 | if (!mnt->mnt_devname) | |
88b38782 | 176 | goto out_free_id; |
73cd49ec MS |
177 | } |
178 | ||
b3e19d92 | 179 | #ifdef CONFIG_SMP |
c63181e6 AV |
180 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
181 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
182 | goto out_free_devname; |
183 | ||
c63181e6 | 184 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 185 | #else |
c63181e6 AV |
186 | mnt->mnt_count = 1; |
187 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
188 | #endif |
189 | ||
c63181e6 AV |
190 | INIT_LIST_HEAD(&mnt->mnt_hash); |
191 | INIT_LIST_HEAD(&mnt->mnt_child); | |
192 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
193 | INIT_LIST_HEAD(&mnt->mnt_list); | |
194 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
195 | INIT_LIST_HEAD(&mnt->mnt_share); | |
196 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
197 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
2504c5d6 AG |
198 | #ifdef CONFIG_FSNOTIFY |
199 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
d3ef3d73 | 200 | #endif |
1da177e4 | 201 | } |
c63181e6 | 202 | return mnt; |
88b38782 | 203 | |
d3ef3d73 | 204 | #ifdef CONFIG_SMP |
205 | out_free_devname: | |
c63181e6 | 206 | kfree(mnt->mnt_devname); |
d3ef3d73 | 207 | #endif |
88b38782 | 208 | out_free_id: |
c63181e6 | 209 | mnt_free_id(mnt); |
88b38782 | 210 | out_free_cache: |
c63181e6 | 211 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 212 | return NULL; |
1da177e4 LT |
213 | } |
214 | ||
3d733633 DH |
215 | /* |
216 | * Most r/o checks on a fs are for operations that take | |
217 | * discrete amounts of time, like a write() or unlink(). | |
218 | * We must keep track of when those operations start | |
219 | * (for permission checks) and when they end, so that | |
220 | * we can determine when writes are able to occur to | |
221 | * a filesystem. | |
222 | */ | |
223 | /* | |
224 | * __mnt_is_readonly: check whether a mount is read-only | |
225 | * @mnt: the mount to check for its write status | |
226 | * | |
227 | * This shouldn't be used directly ouside of the VFS. | |
228 | * It does not guarantee that the filesystem will stay | |
229 | * r/w, just that it is right *now*. This can not and | |
230 | * should not be used in place of IS_RDONLY(inode). | |
231 | * mnt_want/drop_write() will _keep_ the filesystem | |
232 | * r/w. | |
233 | */ | |
234 | int __mnt_is_readonly(struct vfsmount *mnt) | |
235 | { | |
2e4b7fcd DH |
236 | if (mnt->mnt_flags & MNT_READONLY) |
237 | return 1; | |
238 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
239 | return 1; | |
240 | return 0; | |
3d733633 DH |
241 | } |
242 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
243 | ||
83adc753 | 244 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 245 | { |
246 | #ifdef CONFIG_SMP | |
68e8a9fe | 247 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 248 | #else |
68e8a9fe | 249 | mnt->mnt_writers++; |
d3ef3d73 | 250 | #endif |
251 | } | |
3d733633 | 252 | |
83adc753 | 253 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 254 | { |
d3ef3d73 | 255 | #ifdef CONFIG_SMP |
68e8a9fe | 256 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 257 | #else |
68e8a9fe | 258 | mnt->mnt_writers--; |
d3ef3d73 | 259 | #endif |
3d733633 | 260 | } |
3d733633 | 261 | |
83adc753 | 262 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 263 | { |
d3ef3d73 | 264 | #ifdef CONFIG_SMP |
265 | unsigned int count = 0; | |
3d733633 | 266 | int cpu; |
3d733633 DH |
267 | |
268 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 269 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 270 | } |
3d733633 | 271 | |
d3ef3d73 | 272 | return count; |
273 | #else | |
274 | return mnt->mnt_writers; | |
275 | #endif | |
3d733633 DH |
276 | } |
277 | ||
4ed5e82f MS |
278 | static int mnt_is_readonly(struct vfsmount *mnt) |
279 | { | |
280 | if (mnt->mnt_sb->s_readonly_remount) | |
281 | return 1; | |
282 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
283 | smp_rmb(); | |
284 | return __mnt_is_readonly(mnt); | |
285 | } | |
286 | ||
8366025e | 287 | /* |
eb04c282 JK |
288 | * Most r/o & frozen checks on a fs are for operations that take discrete |
289 | * amounts of time, like a write() or unlink(). We must keep track of when | |
290 | * those operations start (for permission checks) and when they end, so that we | |
291 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
292 | */ |
293 | /** | |
eb04c282 | 294 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 295 | * @m: the mount on which to take a write |
8366025e | 296 | * |
eb04c282 JK |
297 | * This tells the low-level filesystem that a write is about to be performed to |
298 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
299 | * returning success. This operation does not protect against filesystem being | |
300 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
301 | * called. This is effectively a refcount. | |
8366025e | 302 | */ |
eb04c282 | 303 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 304 | { |
83adc753 | 305 | struct mount *mnt = real_mount(m); |
3d733633 | 306 | int ret = 0; |
3d733633 | 307 | |
d3ef3d73 | 308 | preempt_disable(); |
c6653a83 | 309 | mnt_inc_writers(mnt); |
d3ef3d73 | 310 | /* |
c6653a83 | 311 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 312 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
313 | * incremented count after it has set MNT_WRITE_HOLD. | |
314 | */ | |
315 | smp_mb(); | |
83adc753 | 316 | while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) |
d3ef3d73 | 317 | cpu_relax(); |
318 | /* | |
319 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
320 | * be set to match its requirements. So we must not load that until | |
321 | * MNT_WRITE_HOLD is cleared. | |
322 | */ | |
323 | smp_rmb(); | |
4ed5e82f | 324 | if (mnt_is_readonly(m)) { |
c6653a83 | 325 | mnt_dec_writers(mnt); |
3d733633 | 326 | ret = -EROFS; |
3d733633 | 327 | } |
d3ef3d73 | 328 | preempt_enable(); |
eb04c282 JK |
329 | |
330 | return ret; | |
331 | } | |
332 | ||
333 | /** | |
334 | * mnt_want_write - get write access to a mount | |
335 | * @m: the mount on which to take a write | |
336 | * | |
337 | * This tells the low-level filesystem that a write is about to be performed to | |
338 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
339 | * is not frozen) before returning success. When the write operation is | |
340 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
341 | */ | |
342 | int mnt_want_write(struct vfsmount *m) | |
343 | { | |
344 | int ret; | |
345 | ||
346 | sb_start_write(m->mnt_sb); | |
347 | ret = __mnt_want_write(m); | |
348 | if (ret) | |
349 | sb_end_write(m->mnt_sb); | |
3d733633 | 350 | return ret; |
8366025e DH |
351 | } |
352 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
353 | ||
96029c4e | 354 | /** |
355 | * mnt_clone_write - get write access to a mount | |
356 | * @mnt: the mount on which to take a write | |
357 | * | |
358 | * This is effectively like mnt_want_write, except | |
359 | * it must only be used to take an extra write reference | |
360 | * on a mountpoint that we already know has a write reference | |
361 | * on it. This allows some optimisation. | |
362 | * | |
363 | * After finished, mnt_drop_write must be called as usual to | |
364 | * drop the reference. | |
365 | */ | |
366 | int mnt_clone_write(struct vfsmount *mnt) | |
367 | { | |
368 | /* superblock may be r/o */ | |
369 | if (__mnt_is_readonly(mnt)) | |
370 | return -EROFS; | |
371 | preempt_disable(); | |
83adc753 | 372 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 373 | preempt_enable(); |
374 | return 0; | |
375 | } | |
376 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
377 | ||
378 | /** | |
eb04c282 | 379 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 380 | * @file: the file who's mount on which to take a write |
381 | * | |
eb04c282 | 382 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 383 | * do some optimisations if the file is open for write already |
384 | */ | |
eb04c282 | 385 | int __mnt_want_write_file(struct file *file) |
96029c4e | 386 | { |
2d8dd38a | 387 | struct inode *inode = file->f_dentry->d_inode; |
eb04c282 | 388 | |
2d8dd38a | 389 | if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) |
eb04c282 | 390 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 391 | else |
392 | return mnt_clone_write(file->f_path.mnt); | |
393 | } | |
eb04c282 JK |
394 | |
395 | /** | |
396 | * mnt_want_write_file - get write access to a file's mount | |
397 | * @file: the file who's mount on which to take a write | |
398 | * | |
399 | * This is like mnt_want_write, but it takes a file and can | |
400 | * do some optimisations if the file is open for write already | |
401 | */ | |
402 | int mnt_want_write_file(struct file *file) | |
403 | { | |
404 | int ret; | |
405 | ||
406 | sb_start_write(file->f_path.mnt->mnt_sb); | |
407 | ret = __mnt_want_write_file(file); | |
408 | if (ret) | |
409 | sb_end_write(file->f_path.mnt->mnt_sb); | |
410 | return ret; | |
411 | } | |
96029c4e | 412 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
413 | ||
8366025e | 414 | /** |
eb04c282 | 415 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
416 | * @mnt: the mount on which to give up write access |
417 | * | |
418 | * Tells the low-level filesystem that we are done | |
419 | * performing writes to it. Must be matched with | |
eb04c282 | 420 | * __mnt_want_write() call above. |
8366025e | 421 | */ |
eb04c282 | 422 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 423 | { |
d3ef3d73 | 424 | preempt_disable(); |
83adc753 | 425 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 426 | preempt_enable(); |
8366025e | 427 | } |
eb04c282 JK |
428 | |
429 | /** | |
430 | * mnt_drop_write - give up write access to a mount | |
431 | * @mnt: the mount on which to give up write access | |
432 | * | |
433 | * Tells the low-level filesystem that we are done performing writes to it and | |
434 | * also allows filesystem to be frozen again. Must be matched with | |
435 | * mnt_want_write() call above. | |
436 | */ | |
437 | void mnt_drop_write(struct vfsmount *mnt) | |
438 | { | |
439 | __mnt_drop_write(mnt); | |
440 | sb_end_write(mnt->mnt_sb); | |
441 | } | |
8366025e DH |
442 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
443 | ||
eb04c282 JK |
444 | void __mnt_drop_write_file(struct file *file) |
445 | { | |
446 | __mnt_drop_write(file->f_path.mnt); | |
447 | } | |
448 | ||
2a79f17e AV |
449 | void mnt_drop_write_file(struct file *file) |
450 | { | |
451 | mnt_drop_write(file->f_path.mnt); | |
452 | } | |
453 | EXPORT_SYMBOL(mnt_drop_write_file); | |
454 | ||
83adc753 | 455 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 456 | { |
3d733633 DH |
457 | int ret = 0; |
458 | ||
962830df | 459 | br_write_lock(&vfsmount_lock); |
83adc753 | 460 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 461 | /* |
d3ef3d73 | 462 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
463 | * should be visible before we do. | |
3d733633 | 464 | */ |
d3ef3d73 | 465 | smp_mb(); |
466 | ||
3d733633 | 467 | /* |
d3ef3d73 | 468 | * With writers on hold, if this value is zero, then there are |
469 | * definitely no active writers (although held writers may subsequently | |
470 | * increment the count, they'll have to wait, and decrement it after | |
471 | * seeing MNT_READONLY). | |
472 | * | |
473 | * It is OK to have counter incremented on one CPU and decremented on | |
474 | * another: the sum will add up correctly. The danger would be when we | |
475 | * sum up each counter, if we read a counter before it is incremented, | |
476 | * but then read another CPU's count which it has been subsequently | |
477 | * decremented from -- we would see more decrements than we should. | |
478 | * MNT_WRITE_HOLD protects against this scenario, because | |
479 | * mnt_want_write first increments count, then smp_mb, then spins on | |
480 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
481 | * we're counting up here. | |
3d733633 | 482 | */ |
c6653a83 | 483 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 484 | ret = -EBUSY; |
485 | else | |
83adc753 | 486 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 487 | /* |
488 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
489 | * that become unheld will see MNT_READONLY. | |
490 | */ | |
491 | smp_wmb(); | |
83adc753 | 492 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
962830df | 493 | br_write_unlock(&vfsmount_lock); |
3d733633 | 494 | return ret; |
8366025e | 495 | } |
8366025e | 496 | |
83adc753 | 497 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 498 | { |
962830df | 499 | br_write_lock(&vfsmount_lock); |
83adc753 | 500 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
962830df | 501 | br_write_unlock(&vfsmount_lock); |
2e4b7fcd DH |
502 | } |
503 | ||
4ed5e82f MS |
504 | int sb_prepare_remount_readonly(struct super_block *sb) |
505 | { | |
506 | struct mount *mnt; | |
507 | int err = 0; | |
508 | ||
8e8b8796 MS |
509 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
510 | if (atomic_long_read(&sb->s_remove_count)) | |
511 | return -EBUSY; | |
512 | ||
962830df | 513 | br_write_lock(&vfsmount_lock); |
4ed5e82f MS |
514 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
515 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
516 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
517 | smp_mb(); | |
518 | if (mnt_get_writers(mnt) > 0) { | |
519 | err = -EBUSY; | |
520 | break; | |
521 | } | |
522 | } | |
523 | } | |
8e8b8796 MS |
524 | if (!err && atomic_long_read(&sb->s_remove_count)) |
525 | err = -EBUSY; | |
526 | ||
4ed5e82f MS |
527 | if (!err) { |
528 | sb->s_readonly_remount = 1; | |
529 | smp_wmb(); | |
530 | } | |
531 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
532 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
533 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
534 | } | |
962830df | 535 | br_write_unlock(&vfsmount_lock); |
4ed5e82f MS |
536 | |
537 | return err; | |
538 | } | |
539 | ||
b105e270 | 540 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 541 | { |
52ba1621 | 542 | kfree(mnt->mnt_devname); |
73cd49ec | 543 | mnt_free_id(mnt); |
d3ef3d73 | 544 | #ifdef CONFIG_SMP |
68e8a9fe | 545 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 546 | #endif |
b105e270 | 547 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
548 | } |
549 | ||
550 | /* | |
a05964f3 RP |
551 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
552 | * @dir. If @dir is set return the first mount else return the last mount. | |
99b7db7b | 553 | * vfsmount_lock must be held for read or write. |
1da177e4 | 554 | */ |
c7105365 | 555 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
a05964f3 | 556 | int dir) |
1da177e4 | 557 | { |
b58fed8b RP |
558 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
559 | struct list_head *tmp = head; | |
c7105365 | 560 | struct mount *p, *found = NULL; |
1da177e4 | 561 | |
1da177e4 | 562 | for (;;) { |
a05964f3 | 563 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
564 | p = NULL; |
565 | if (tmp == head) | |
566 | break; | |
1b8e5564 | 567 | p = list_entry(tmp, struct mount, mnt_hash); |
a73324da | 568 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { |
a05964f3 | 569 | found = p; |
1da177e4 LT |
570 | break; |
571 | } | |
572 | } | |
1da177e4 LT |
573 | return found; |
574 | } | |
575 | ||
a05964f3 | 576 | /* |
f015f126 DH |
577 | * lookup_mnt - Return the first child mount mounted at path |
578 | * | |
579 | * "First" means first mounted chronologically. If you create the | |
580 | * following mounts: | |
581 | * | |
582 | * mount /dev/sda1 /mnt | |
583 | * mount /dev/sda2 /mnt | |
584 | * mount /dev/sda3 /mnt | |
585 | * | |
586 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
587 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
588 | * /dev/sda2, then /dev/sda3, then NULL. | |
589 | * | |
590 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 591 | */ |
1c755af4 | 592 | struct vfsmount *lookup_mnt(struct path *path) |
a05964f3 | 593 | { |
c7105365 | 594 | struct mount *child_mnt; |
99b7db7b | 595 | |
962830df | 596 | br_read_lock(&vfsmount_lock); |
c7105365 AV |
597 | child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); |
598 | if (child_mnt) { | |
599 | mnt_add_count(child_mnt, 1); | |
962830df | 600 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
601 | return &child_mnt->mnt; |
602 | } else { | |
962830df | 603 | br_read_unlock(&vfsmount_lock); |
c7105365 AV |
604 | return NULL; |
605 | } | |
a05964f3 RP |
606 | } |
607 | ||
143c8c91 | 608 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 609 | { |
6b3286ed | 610 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
611 | } |
612 | ||
99b7db7b NP |
613 | /* |
614 | * vfsmount lock must be held for write | |
615 | */ | |
6b3286ed | 616 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
617 | { |
618 | if (ns) { | |
619 | ns->event = ++event; | |
620 | wake_up_interruptible(&ns->poll); | |
621 | } | |
622 | } | |
623 | ||
99b7db7b NP |
624 | /* |
625 | * vfsmount lock must be held for write | |
626 | */ | |
6b3286ed | 627 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
628 | { |
629 | if (ns && ns->event != event) { | |
630 | ns->event = event; | |
631 | wake_up_interruptible(&ns->poll); | |
632 | } | |
633 | } | |
634 | ||
5f57cbcc NP |
635 | /* |
636 | * Clear dentry's mounted state if it has no remaining mounts. | |
637 | * vfsmount_lock must be held for write. | |
638 | */ | |
aa0a4cf0 | 639 | static void dentry_reset_mounted(struct dentry *dentry) |
5f57cbcc NP |
640 | { |
641 | unsigned u; | |
642 | ||
643 | for (u = 0; u < HASH_SIZE; u++) { | |
d5e50f74 | 644 | struct mount *p; |
5f57cbcc | 645 | |
1b8e5564 | 646 | list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { |
a73324da | 647 | if (p->mnt_mountpoint == dentry) |
5f57cbcc NP |
648 | return; |
649 | } | |
650 | } | |
651 | spin_lock(&dentry->d_lock); | |
652 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
653 | spin_unlock(&dentry->d_lock); | |
654 | } | |
655 | ||
99b7db7b NP |
656 | /* |
657 | * vfsmount lock must be held for write | |
658 | */ | |
419148da AV |
659 | static void detach_mnt(struct mount *mnt, struct path *old_path) |
660 | { | |
a73324da | 661 | old_path->dentry = mnt->mnt_mountpoint; |
0714a533 AV |
662 | old_path->mnt = &mnt->mnt_parent->mnt; |
663 | mnt->mnt_parent = mnt; | |
a73324da | 664 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 665 | list_del_init(&mnt->mnt_child); |
1b8e5564 | 666 | list_del_init(&mnt->mnt_hash); |
aa0a4cf0 | 667 | dentry_reset_mounted(old_path->dentry); |
1da177e4 LT |
668 | } |
669 | ||
99b7db7b NP |
670 | /* |
671 | * vfsmount lock must be held for write | |
672 | */ | |
14cf1fa8 | 673 | void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry, |
44d964d6 | 674 | struct mount *child_mnt) |
b90fa9ae | 675 | { |
3a2393d7 | 676 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
a73324da | 677 | child_mnt->mnt_mountpoint = dget(dentry); |
3a2393d7 | 678 | child_mnt->mnt_parent = mnt; |
5f57cbcc NP |
679 | spin_lock(&dentry->d_lock); |
680 | dentry->d_flags |= DCACHE_MOUNTED; | |
681 | spin_unlock(&dentry->d_lock); | |
b90fa9ae RP |
682 | } |
683 | ||
99b7db7b NP |
684 | /* |
685 | * vfsmount lock must be held for write | |
686 | */ | |
419148da | 687 | static void attach_mnt(struct mount *mnt, struct path *path) |
1da177e4 | 688 | { |
14cf1fa8 | 689 | mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt); |
1b8e5564 | 690 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
1a390689 | 691 | hash(path->mnt, path->dentry)); |
6b41d536 | 692 | list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts); |
b90fa9ae RP |
693 | } |
694 | ||
695 | /* | |
99b7db7b | 696 | * vfsmount lock must be held for write |
b90fa9ae | 697 | */ |
4b2619a5 | 698 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 699 | { |
0714a533 | 700 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 701 | struct mount *m; |
b90fa9ae | 702 | LIST_HEAD(head); |
143c8c91 | 703 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 704 | |
0714a533 | 705 | BUG_ON(parent == mnt); |
b90fa9ae | 706 | |
1a4eeaf2 | 707 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 708 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 709 | m->mnt_ns = n; |
f03c6599 | 710 | |
b90fa9ae RP |
711 | list_splice(&head, n->list.prev); |
712 | ||
1b8e5564 | 713 | list_add_tail(&mnt->mnt_hash, mount_hashtable + |
a73324da | 714 | hash(&parent->mnt, mnt->mnt_mountpoint)); |
6b41d536 | 715 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); |
6b3286ed | 716 | touch_mnt_namespace(n); |
1da177e4 LT |
717 | } |
718 | ||
909b0a88 | 719 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 720 | { |
6b41d536 AV |
721 | struct list_head *next = p->mnt_mounts.next; |
722 | if (next == &p->mnt_mounts) { | |
1da177e4 | 723 | while (1) { |
909b0a88 | 724 | if (p == root) |
1da177e4 | 725 | return NULL; |
6b41d536 AV |
726 | next = p->mnt_child.next; |
727 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 728 | break; |
0714a533 | 729 | p = p->mnt_parent; |
1da177e4 LT |
730 | } |
731 | } | |
6b41d536 | 732 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
733 | } |
734 | ||
315fc83e | 735 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 736 | { |
6b41d536 AV |
737 | struct list_head *prev = p->mnt_mounts.prev; |
738 | while (prev != &p->mnt_mounts) { | |
739 | p = list_entry(prev, struct mount, mnt_child); | |
740 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
741 | } |
742 | return p; | |
743 | } | |
744 | ||
9d412a43 AV |
745 | struct vfsmount * |
746 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
747 | { | |
b105e270 | 748 | struct mount *mnt; |
9d412a43 AV |
749 | struct dentry *root; |
750 | ||
751 | if (!type) | |
752 | return ERR_PTR(-ENODEV); | |
753 | ||
754 | mnt = alloc_vfsmnt(name); | |
755 | if (!mnt) | |
756 | return ERR_PTR(-ENOMEM); | |
757 | ||
758 | if (flags & MS_KERNMOUNT) | |
b105e270 | 759 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
760 | |
761 | root = mount_fs(type, flags, name, data); | |
762 | if (IS_ERR(root)) { | |
763 | free_vfsmnt(mnt); | |
764 | return ERR_CAST(root); | |
765 | } | |
766 | ||
b105e270 AV |
767 | mnt->mnt.mnt_root = root; |
768 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 769 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 770 | mnt->mnt_parent = mnt; |
962830df | 771 | br_write_lock(&vfsmount_lock); |
39f7c4db | 772 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
962830df | 773 | br_write_unlock(&vfsmount_lock); |
b105e270 | 774 | return &mnt->mnt; |
9d412a43 AV |
775 | } |
776 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
777 | ||
87129cc0 | 778 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 779 | int flag) |
1da177e4 | 780 | { |
87129cc0 | 781 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
782 | struct mount *mnt; |
783 | int err; | |
1da177e4 | 784 | |
be34d1a3 DH |
785 | mnt = alloc_vfsmnt(old->mnt_devname); |
786 | if (!mnt) | |
787 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 788 | |
be34d1a3 DH |
789 | if (flag & (CL_SLAVE | CL_PRIVATE)) |
790 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
791 | else | |
792 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 793 | |
be34d1a3 DH |
794 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
795 | err = mnt_alloc_group_id(mnt); | |
796 | if (err) | |
797 | goto out_free; | |
1da177e4 | 798 | } |
be34d1a3 DH |
799 | |
800 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; | |
801 | atomic_inc(&sb->s_active); | |
802 | mnt->mnt.mnt_sb = sb; | |
803 | mnt->mnt.mnt_root = dget(root); | |
804 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
805 | mnt->mnt_parent = mnt; | |
806 | br_write_lock(&vfsmount_lock); | |
807 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
808 | br_write_unlock(&vfsmount_lock); | |
809 | ||
810 | if (flag & CL_SLAVE) { | |
811 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
812 | mnt->mnt_master = old; | |
813 | CLEAR_MNT_SHARED(mnt); | |
814 | } else if (!(flag & CL_PRIVATE)) { | |
815 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
816 | list_add(&mnt->mnt_share, &old->mnt_share); | |
817 | if (IS_MNT_SLAVE(old)) | |
818 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
819 | mnt->mnt_master = old->mnt_master; | |
820 | } | |
821 | if (flag & CL_MAKE_SHARED) | |
822 | set_mnt_shared(mnt); | |
823 | ||
824 | /* stick the duplicate mount on the same expiry list | |
825 | * as the original if that was on one */ | |
826 | if (flag & CL_EXPIRE) { | |
827 | if (!list_empty(&old->mnt_expire)) | |
828 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
829 | } | |
830 | ||
cb338d06 | 831 | return mnt; |
719f5d7f MS |
832 | |
833 | out_free: | |
834 | free_vfsmnt(mnt); | |
be34d1a3 | 835 | return ERR_PTR(err); |
1da177e4 LT |
836 | } |
837 | ||
83adc753 | 838 | static inline void mntfree(struct mount *mnt) |
1da177e4 | 839 | { |
83adc753 AV |
840 | struct vfsmount *m = &mnt->mnt; |
841 | struct super_block *sb = m->mnt_sb; | |
b3e19d92 | 842 | |
3d733633 DH |
843 | /* |
844 | * This probably indicates that somebody messed | |
845 | * up a mnt_want/drop_write() pair. If this | |
846 | * happens, the filesystem was probably unable | |
847 | * to make r/w->r/o transitions. | |
848 | */ | |
d3ef3d73 | 849 | /* |
b3e19d92 NP |
850 | * The locking used to deal with mnt_count decrement provides barriers, |
851 | * so mnt_get_writers() below is safe. | |
d3ef3d73 | 852 | */ |
c6653a83 | 853 | WARN_ON(mnt_get_writers(mnt)); |
83adc753 AV |
854 | fsnotify_vfsmount_delete(m); |
855 | dput(m->mnt_root); | |
856 | free_vfsmnt(mnt); | |
1da177e4 LT |
857 | deactivate_super(sb); |
858 | } | |
859 | ||
900148dc | 860 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 861 | { |
b3e19d92 | 862 | put_again: |
f03c6599 | 863 | #ifdef CONFIG_SMP |
962830df | 864 | br_read_lock(&vfsmount_lock); |
f7a99c5b AV |
865 | if (likely(mnt->mnt_ns)) { |
866 | /* shouldn't be the last one */ | |
aa9c0e07 | 867 | mnt_add_count(mnt, -1); |
962830df | 868 | br_read_unlock(&vfsmount_lock); |
f03c6599 | 869 | return; |
b3e19d92 | 870 | } |
962830df | 871 | br_read_unlock(&vfsmount_lock); |
b3e19d92 | 872 | |
962830df | 873 | br_write_lock(&vfsmount_lock); |
aa9c0e07 | 874 | mnt_add_count(mnt, -1); |
b3e19d92 | 875 | if (mnt_get_count(mnt)) { |
962830df | 876 | br_write_unlock(&vfsmount_lock); |
99b7db7b NP |
877 | return; |
878 | } | |
b3e19d92 | 879 | #else |
aa9c0e07 | 880 | mnt_add_count(mnt, -1); |
b3e19d92 | 881 | if (likely(mnt_get_count(mnt))) |
99b7db7b | 882 | return; |
962830df | 883 | br_write_lock(&vfsmount_lock); |
f03c6599 | 884 | #endif |
863d684f AV |
885 | if (unlikely(mnt->mnt_pinned)) { |
886 | mnt_add_count(mnt, mnt->mnt_pinned + 1); | |
887 | mnt->mnt_pinned = 0; | |
962830df | 888 | br_write_unlock(&vfsmount_lock); |
900148dc | 889 | acct_auto_close_mnt(&mnt->mnt); |
b3e19d92 | 890 | goto put_again; |
7b7b1ace | 891 | } |
962830df | 892 | |
39f7c4db | 893 | list_del(&mnt->mnt_instance); |
962830df | 894 | br_write_unlock(&vfsmount_lock); |
b3e19d92 NP |
895 | mntfree(mnt); |
896 | } | |
b3e19d92 NP |
897 | |
898 | void mntput(struct vfsmount *mnt) | |
899 | { | |
900 | if (mnt) { | |
863d684f | 901 | struct mount *m = real_mount(mnt); |
b3e19d92 | 902 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
903 | if (unlikely(m->mnt_expiry_mark)) |
904 | m->mnt_expiry_mark = 0; | |
905 | mntput_no_expire(m); | |
b3e19d92 NP |
906 | } |
907 | } | |
908 | EXPORT_SYMBOL(mntput); | |
909 | ||
910 | struct vfsmount *mntget(struct vfsmount *mnt) | |
911 | { | |
912 | if (mnt) | |
83adc753 | 913 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
914 | return mnt; |
915 | } | |
916 | EXPORT_SYMBOL(mntget); | |
917 | ||
7b7b1ace AV |
918 | void mnt_pin(struct vfsmount *mnt) |
919 | { | |
962830df | 920 | br_write_lock(&vfsmount_lock); |
863d684f | 921 | real_mount(mnt)->mnt_pinned++; |
962830df | 922 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 923 | } |
7b7b1ace AV |
924 | EXPORT_SYMBOL(mnt_pin); |
925 | ||
863d684f | 926 | void mnt_unpin(struct vfsmount *m) |
7b7b1ace | 927 | { |
863d684f | 928 | struct mount *mnt = real_mount(m); |
962830df | 929 | br_write_lock(&vfsmount_lock); |
7b7b1ace | 930 | if (mnt->mnt_pinned) { |
863d684f | 931 | mnt_add_count(mnt, 1); |
7b7b1ace AV |
932 | mnt->mnt_pinned--; |
933 | } | |
962830df | 934 | br_write_unlock(&vfsmount_lock); |
7b7b1ace | 935 | } |
7b7b1ace | 936 | EXPORT_SYMBOL(mnt_unpin); |
1da177e4 | 937 | |
b3b304a2 MS |
938 | static inline void mangle(struct seq_file *m, const char *s) |
939 | { | |
940 | seq_escape(m, s, " \t\n\\"); | |
941 | } | |
942 | ||
943 | /* | |
944 | * Simple .show_options callback for filesystems which don't want to | |
945 | * implement more complex mount option showing. | |
946 | * | |
947 | * See also save_mount_options(). | |
948 | */ | |
34c80b1d | 949 | int generic_show_options(struct seq_file *m, struct dentry *root) |
b3b304a2 | 950 | { |
2a32cebd AV |
951 | const char *options; |
952 | ||
953 | rcu_read_lock(); | |
34c80b1d | 954 | options = rcu_dereference(root->d_sb->s_options); |
b3b304a2 MS |
955 | |
956 | if (options != NULL && options[0]) { | |
957 | seq_putc(m, ','); | |
958 | mangle(m, options); | |
959 | } | |
2a32cebd | 960 | rcu_read_unlock(); |
b3b304a2 MS |
961 | |
962 | return 0; | |
963 | } | |
964 | EXPORT_SYMBOL(generic_show_options); | |
965 | ||
966 | /* | |
967 | * If filesystem uses generic_show_options(), this function should be | |
968 | * called from the fill_super() callback. | |
969 | * | |
970 | * The .remount_fs callback usually needs to be handled in a special | |
971 | * way, to make sure, that previous options are not overwritten if the | |
972 | * remount fails. | |
973 | * | |
974 | * Also note, that if the filesystem's .remount_fs function doesn't | |
975 | * reset all options to their default value, but changes only newly | |
976 | * given options, then the displayed options will not reflect reality | |
977 | * any more. | |
978 | */ | |
979 | void save_mount_options(struct super_block *sb, char *options) | |
980 | { | |
2a32cebd AV |
981 | BUG_ON(sb->s_options); |
982 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
b3b304a2 MS |
983 | } |
984 | EXPORT_SYMBOL(save_mount_options); | |
985 | ||
2a32cebd AV |
986 | void replace_mount_options(struct super_block *sb, char *options) |
987 | { | |
988 | char *old = sb->s_options; | |
989 | rcu_assign_pointer(sb->s_options, options); | |
990 | if (old) { | |
991 | synchronize_rcu(); | |
992 | kfree(old); | |
993 | } | |
994 | } | |
995 | EXPORT_SYMBOL(replace_mount_options); | |
996 | ||
a1a2c409 | 997 | #ifdef CONFIG_PROC_FS |
0226f492 | 998 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
999 | static void *m_start(struct seq_file *m, loff_t *pos) |
1000 | { | |
6ce6e24e | 1001 | struct proc_mounts *p = proc_mounts(m); |
1da177e4 | 1002 | |
390c6843 | 1003 | down_read(&namespace_sem); |
a1a2c409 | 1004 | return seq_list_start(&p->ns->list, *pos); |
1da177e4 LT |
1005 | } |
1006 | ||
1007 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1008 | { | |
6ce6e24e | 1009 | struct proc_mounts *p = proc_mounts(m); |
b0765fb8 | 1010 | |
a1a2c409 | 1011 | return seq_list_next(v, &p->ns->list, pos); |
1da177e4 LT |
1012 | } |
1013 | ||
1014 | static void m_stop(struct seq_file *m, void *v) | |
1015 | { | |
390c6843 | 1016 | up_read(&namespace_sem); |
1da177e4 LT |
1017 | } |
1018 | ||
0226f492 | 1019 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1020 | { |
6ce6e24e | 1021 | struct proc_mounts *p = proc_mounts(m); |
1a4eeaf2 | 1022 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1023 | return p->show(m, &r->mnt); |
1da177e4 LT |
1024 | } |
1025 | ||
a1a2c409 | 1026 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1027 | .start = m_start, |
1028 | .next = m_next, | |
1029 | .stop = m_stop, | |
0226f492 | 1030 | .show = m_show, |
b4629fe2 | 1031 | }; |
a1a2c409 | 1032 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1033 | |
1da177e4 LT |
1034 | /** |
1035 | * may_umount_tree - check if a mount tree is busy | |
1036 | * @mnt: root of mount tree | |
1037 | * | |
1038 | * This is called to check if a tree of mounts has any | |
1039 | * open files, pwds, chroots or sub mounts that are | |
1040 | * busy. | |
1041 | */ | |
909b0a88 | 1042 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1043 | { |
909b0a88 | 1044 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1045 | int actual_refs = 0; |
1046 | int minimum_refs = 0; | |
315fc83e | 1047 | struct mount *p; |
909b0a88 | 1048 | BUG_ON(!m); |
1da177e4 | 1049 | |
b3e19d92 | 1050 | /* write lock needed for mnt_get_count */ |
962830df | 1051 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1052 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1053 | actual_refs += mnt_get_count(p); |
1da177e4 | 1054 | minimum_refs += 2; |
1da177e4 | 1055 | } |
962830df | 1056 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1057 | |
1058 | if (actual_refs > minimum_refs) | |
e3474a8e | 1059 | return 0; |
1da177e4 | 1060 | |
e3474a8e | 1061 | return 1; |
1da177e4 LT |
1062 | } |
1063 | ||
1064 | EXPORT_SYMBOL(may_umount_tree); | |
1065 | ||
1066 | /** | |
1067 | * may_umount - check if a mount point is busy | |
1068 | * @mnt: root of mount | |
1069 | * | |
1070 | * This is called to check if a mount point has any | |
1071 | * open files, pwds, chroots or sub mounts. If the | |
1072 | * mount has sub mounts this will return busy | |
1073 | * regardless of whether the sub mounts are busy. | |
1074 | * | |
1075 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1076 | * give false negatives. The main reason why it's here is that we need | |
1077 | * a non-destructive way to look for easily umountable filesystems. | |
1078 | */ | |
1079 | int may_umount(struct vfsmount *mnt) | |
1080 | { | |
e3474a8e | 1081 | int ret = 1; |
8ad08d8a | 1082 | down_read(&namespace_sem); |
962830df | 1083 | br_write_lock(&vfsmount_lock); |
1ab59738 | 1084 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1085 | ret = 0; |
962830df | 1086 | br_write_unlock(&vfsmount_lock); |
8ad08d8a | 1087 | up_read(&namespace_sem); |
a05964f3 | 1088 | return ret; |
1da177e4 LT |
1089 | } |
1090 | ||
1091 | EXPORT_SYMBOL(may_umount); | |
1092 | ||
b90fa9ae | 1093 | void release_mounts(struct list_head *head) |
70fbcdf4 | 1094 | { |
d5e50f74 | 1095 | struct mount *mnt; |
bf066c7d | 1096 | while (!list_empty(head)) { |
1b8e5564 AV |
1097 | mnt = list_first_entry(head, struct mount, mnt_hash); |
1098 | list_del_init(&mnt->mnt_hash); | |
676da58d | 1099 | if (mnt_has_parent(mnt)) { |
70fbcdf4 | 1100 | struct dentry *dentry; |
863d684f | 1101 | struct mount *m; |
99b7db7b | 1102 | |
962830df | 1103 | br_write_lock(&vfsmount_lock); |
a73324da | 1104 | dentry = mnt->mnt_mountpoint; |
863d684f | 1105 | m = mnt->mnt_parent; |
a73324da | 1106 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1107 | mnt->mnt_parent = mnt; |
7c4b93d8 | 1108 | m->mnt_ghosts--; |
962830df | 1109 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1110 | dput(dentry); |
863d684f | 1111 | mntput(&m->mnt); |
70fbcdf4 | 1112 | } |
d5e50f74 | 1113 | mntput(&mnt->mnt); |
70fbcdf4 RP |
1114 | } |
1115 | } | |
1116 | ||
99b7db7b NP |
1117 | /* |
1118 | * vfsmount lock must be held for write | |
1119 | * namespace_sem must be held for write | |
1120 | */ | |
761d5c38 | 1121 | void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) |
1da177e4 | 1122 | { |
7b8a53fd | 1123 | LIST_HEAD(tmp_list); |
315fc83e | 1124 | struct mount *p; |
1da177e4 | 1125 | |
909b0a88 | 1126 | for (p = mnt; p; p = next_mnt(p, mnt)) |
1b8e5564 | 1127 | list_move(&p->mnt_hash, &tmp_list); |
1da177e4 | 1128 | |
a05964f3 | 1129 | if (propagate) |
7b8a53fd | 1130 | propagate_umount(&tmp_list); |
a05964f3 | 1131 | |
1b8e5564 | 1132 | list_for_each_entry(p, &tmp_list, mnt_hash) { |
6776db3d | 1133 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1134 | list_del_init(&p->mnt_list); |
143c8c91 AV |
1135 | __touch_mnt_namespace(p->mnt_ns); |
1136 | p->mnt_ns = NULL; | |
6b41d536 | 1137 | list_del_init(&p->mnt_child); |
676da58d | 1138 | if (mnt_has_parent(p)) { |
863d684f | 1139 | p->mnt_parent->mnt_ghosts++; |
a73324da | 1140 | dentry_reset_mounted(p->mnt_mountpoint); |
7c4b93d8 | 1141 | } |
0f0afb1d | 1142 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 | 1143 | } |
7b8a53fd | 1144 | list_splice(&tmp_list, kill); |
1da177e4 LT |
1145 | } |
1146 | ||
692afc31 | 1147 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts); |
c35038be | 1148 | |
1ab59738 | 1149 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1150 | { |
1ab59738 | 1151 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 | 1152 | int retval; |
70fbcdf4 | 1153 | LIST_HEAD(umount_list); |
1da177e4 | 1154 | |
1ab59738 | 1155 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1156 | if (retval) |
1157 | return retval; | |
1158 | ||
1159 | /* | |
1160 | * Allow userspace to request a mountpoint be expired rather than | |
1161 | * unmounting unconditionally. Unmount only happens if: | |
1162 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1163 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1164 | */ | |
1165 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1166 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1167 | flags & (MNT_FORCE | MNT_DETACH)) |
1168 | return -EINVAL; | |
1169 | ||
b3e19d92 NP |
1170 | /* |
1171 | * probably don't strictly need the lock here if we examined | |
1172 | * all race cases, but it's a slowpath. | |
1173 | */ | |
962830df | 1174 | br_write_lock(&vfsmount_lock); |
83adc753 | 1175 | if (mnt_get_count(mnt) != 2) { |
962830df | 1176 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1177 | return -EBUSY; |
b3e19d92 | 1178 | } |
962830df | 1179 | br_write_unlock(&vfsmount_lock); |
1da177e4 | 1180 | |
863d684f | 1181 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1182 | return -EAGAIN; |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * If we may have to abort operations to get out of this | |
1187 | * mount, and they will themselves hold resources we must | |
1188 | * allow the fs to do things. In the Unix tradition of | |
1189 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1190 | * might fail to complete on the first run through as other tasks | |
1191 | * must return, and the like. Thats for the mount program to worry | |
1192 | * about for the moment. | |
1193 | */ | |
1194 | ||
42faad99 | 1195 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1196 | sb->s_op->umount_begin(sb); |
42faad99 | 1197 | } |
1da177e4 LT |
1198 | |
1199 | /* | |
1200 | * No sense to grab the lock for this test, but test itself looks | |
1201 | * somewhat bogus. Suggestions for better replacement? | |
1202 | * Ho-hum... In principle, we might treat that as umount + switch | |
1203 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1204 | * Actually it makes sense, especially if rootfs would contain a | |
1205 | * /reboot - static binary that would close all descriptors and | |
1206 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1207 | */ | |
1ab59738 | 1208 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1209 | /* |
1210 | * Special case for "unmounting" root ... | |
1211 | * we just try to remount it readonly. | |
1212 | */ | |
1213 | down_write(&sb->s_umount); | |
4aa98cf7 | 1214 | if (!(sb->s_flags & MS_RDONLY)) |
1da177e4 | 1215 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); |
1da177e4 LT |
1216 | up_write(&sb->s_umount); |
1217 | return retval; | |
1218 | } | |
1219 | ||
390c6843 | 1220 | down_write(&namespace_sem); |
962830df | 1221 | br_write_lock(&vfsmount_lock); |
5addc5dd | 1222 | event++; |
1da177e4 | 1223 | |
c35038be | 1224 | if (!(flags & MNT_DETACH)) |
1ab59738 | 1225 | shrink_submounts(mnt, &umount_list); |
c35038be | 1226 | |
1da177e4 | 1227 | retval = -EBUSY; |
a05964f3 | 1228 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1a4eeaf2 | 1229 | if (!list_empty(&mnt->mnt_list)) |
1ab59738 | 1230 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
1231 | retval = 0; |
1232 | } | |
962830df | 1233 | br_write_unlock(&vfsmount_lock); |
390c6843 | 1234 | up_write(&namespace_sem); |
70fbcdf4 | 1235 | release_mounts(&umount_list); |
1da177e4 LT |
1236 | return retval; |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Now umount can handle mount points as well as block devices. | |
1241 | * This is important for filesystems which use unnamed block devices. | |
1242 | * | |
1243 | * We now support a flag for forced unmount like the other 'big iron' | |
1244 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1245 | */ | |
1246 | ||
bdc480e3 | 1247 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1248 | { |
2d8f3038 | 1249 | struct path path; |
900148dc | 1250 | struct mount *mnt; |
1da177e4 | 1251 | int retval; |
db1f05bb | 1252 | int lookup_flags = 0; |
1da177e4 | 1253 | |
db1f05bb MS |
1254 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1255 | return -EINVAL; | |
1256 | ||
1257 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1258 | lookup_flags |= LOOKUP_FOLLOW; | |
1259 | ||
1260 | retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1da177e4 LT |
1261 | if (retval) |
1262 | goto out; | |
900148dc | 1263 | mnt = real_mount(path.mnt); |
1da177e4 | 1264 | retval = -EINVAL; |
2d8f3038 | 1265 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1266 | goto dput_and_out; |
143c8c91 | 1267 | if (!check_mnt(mnt)) |
1da177e4 LT |
1268 | goto dput_and_out; |
1269 | ||
1270 | retval = -EPERM; | |
1271 | if (!capable(CAP_SYS_ADMIN)) | |
1272 | goto dput_and_out; | |
1273 | ||
900148dc | 1274 | retval = do_umount(mnt, flags); |
1da177e4 | 1275 | dput_and_out: |
429731b1 | 1276 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1277 | dput(path.dentry); |
900148dc | 1278 | mntput_no_expire(mnt); |
1da177e4 LT |
1279 | out: |
1280 | return retval; | |
1281 | } | |
1282 | ||
1283 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1284 | ||
1285 | /* | |
b58fed8b | 1286 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1287 | */ |
bdc480e3 | 1288 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1289 | { |
b58fed8b | 1290 | return sys_umount(name, 0); |
1da177e4 LT |
1291 | } |
1292 | ||
1293 | #endif | |
1294 | ||
2d92ab3c | 1295 | static int mount_is_safe(struct path *path) |
1da177e4 LT |
1296 | { |
1297 | if (capable(CAP_SYS_ADMIN)) | |
1298 | return 0; | |
1299 | return -EPERM; | |
1300 | #ifdef notyet | |
2d92ab3c | 1301 | if (S_ISLNK(path->dentry->d_inode->i_mode)) |
1da177e4 | 1302 | return -EPERM; |
2d92ab3c | 1303 | if (path->dentry->d_inode->i_mode & S_ISVTX) { |
da9592ed | 1304 | if (current_uid() != path->dentry->d_inode->i_uid) |
1da177e4 LT |
1305 | return -EPERM; |
1306 | } | |
2d92ab3c | 1307 | if (inode_permission(path->dentry->d_inode, MAY_WRITE)) |
1da177e4 LT |
1308 | return -EPERM; |
1309 | return 0; | |
1310 | #endif | |
1311 | } | |
1312 | ||
8823c079 EB |
1313 | static bool mnt_ns_loop(struct path *path) |
1314 | { | |
1315 | /* Could bind mounting the mount namespace inode cause a | |
1316 | * mount namespace loop? | |
1317 | */ | |
1318 | struct inode *inode = path->dentry->d_inode; | |
1319 | struct proc_inode *ei; | |
1320 | struct mnt_namespace *mnt_ns; | |
1321 | ||
1322 | if (!proc_ns_inode(inode)) | |
1323 | return false; | |
1324 | ||
1325 | ei = PROC_I(inode); | |
1326 | if (ei->ns_ops != &mntns_operations) | |
1327 | return false; | |
1328 | ||
1329 | mnt_ns = ei->ns; | |
1330 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; | |
1331 | } | |
1332 | ||
87129cc0 | 1333 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1334 | int flag) |
1da177e4 | 1335 | { |
a73324da | 1336 | struct mount *res, *p, *q, *r; |
1a390689 | 1337 | struct path path; |
1da177e4 | 1338 | |
fc7be130 | 1339 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
be34d1a3 | 1340 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1341 | |
36341f64 | 1342 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1343 | if (IS_ERR(q)) |
1344 | return q; | |
1345 | ||
a73324da | 1346 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1347 | |
1348 | p = mnt; | |
6b41d536 | 1349 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1350 | struct mount *s; |
7ec02ef1 | 1351 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1352 | continue; |
1353 | ||
909b0a88 | 1354 | for (s = r; s; s = next_mnt(s, r)) { |
fc7be130 | 1355 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
9676f0c6 RP |
1356 | s = skip_mnt_tree(s); |
1357 | continue; | |
1358 | } | |
0714a533 AV |
1359 | while (p != s->mnt_parent) { |
1360 | p = p->mnt_parent; | |
1361 | q = q->mnt_parent; | |
1da177e4 | 1362 | } |
87129cc0 | 1363 | p = s; |
cb338d06 | 1364 | path.mnt = &q->mnt; |
a73324da | 1365 | path.dentry = p->mnt_mountpoint; |
87129cc0 | 1366 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1367 | if (IS_ERR(q)) |
1368 | goto out; | |
962830df | 1369 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 1370 | list_add_tail(&q->mnt_list, &res->mnt_list); |
cb338d06 | 1371 | attach_mnt(q, &path); |
962830df | 1372 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
1373 | } |
1374 | } | |
1375 | return res; | |
be34d1a3 | 1376 | out: |
1da177e4 | 1377 | if (res) { |
70fbcdf4 | 1378 | LIST_HEAD(umount_list); |
962830df | 1379 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1380 | umount_tree(res, 0, &umount_list); |
962830df | 1381 | br_write_unlock(&vfsmount_lock); |
70fbcdf4 | 1382 | release_mounts(&umount_list); |
1da177e4 | 1383 | } |
be34d1a3 | 1384 | return q; |
1da177e4 LT |
1385 | } |
1386 | ||
be34d1a3 DH |
1387 | /* Caller should check returned pointer for errors */ |
1388 | ||
589ff870 | 1389 | struct vfsmount *collect_mounts(struct path *path) |
8aec0809 | 1390 | { |
cb338d06 | 1391 | struct mount *tree; |
1a60a280 | 1392 | down_write(&namespace_sem); |
87129cc0 AV |
1393 | tree = copy_tree(real_mount(path->mnt), path->dentry, |
1394 | CL_COPY_ALL | CL_PRIVATE); | |
1a60a280 | 1395 | up_write(&namespace_sem); |
be34d1a3 DH |
1396 | if (IS_ERR(tree)) |
1397 | return NULL; | |
1398 | return &tree->mnt; | |
8aec0809 AV |
1399 | } |
1400 | ||
1401 | void drop_collected_mounts(struct vfsmount *mnt) | |
1402 | { | |
1403 | LIST_HEAD(umount_list); | |
1a60a280 | 1404 | down_write(&namespace_sem); |
962830df | 1405 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1406 | umount_tree(real_mount(mnt), 0, &umount_list); |
962830df | 1407 | br_write_unlock(&vfsmount_lock); |
1a60a280 | 1408 | up_write(&namespace_sem); |
8aec0809 AV |
1409 | release_mounts(&umount_list); |
1410 | } | |
1411 | ||
1f707137 AV |
1412 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1413 | struct vfsmount *root) | |
1414 | { | |
1a4eeaf2 | 1415 | struct mount *mnt; |
1f707137 AV |
1416 | int res = f(root, arg); |
1417 | if (res) | |
1418 | return res; | |
1a4eeaf2 AV |
1419 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1420 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1421 | if (res) |
1422 | return res; | |
1423 | } | |
1424 | return 0; | |
1425 | } | |
1426 | ||
4b8b21f4 | 1427 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1428 | { |
315fc83e | 1429 | struct mount *p; |
719f5d7f | 1430 | |
909b0a88 | 1431 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1432 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1433 | mnt_release_group_id(p); |
719f5d7f MS |
1434 | } |
1435 | } | |
1436 | ||
4b8b21f4 | 1437 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1438 | { |
315fc83e | 1439 | struct mount *p; |
719f5d7f | 1440 | |
909b0a88 | 1441 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1442 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1443 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1444 | if (err) { |
4b8b21f4 | 1445 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1446 | return err; |
1447 | } | |
1448 | } | |
1449 | } | |
1450 | ||
1451 | return 0; | |
1452 | } | |
1453 | ||
b90fa9ae RP |
1454 | /* |
1455 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1456 | * @nd : place the mount tree @source_mnt is attached |
1457 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1458 | * store the parent mount and mountpoint dentry. | |
1459 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1460 | * |
1461 | * NOTE: in the table below explains the semantics when a source mount | |
1462 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1463 | * --------------------------------------------------------------------------- |
1464 | * | BIND MOUNT OPERATION | | |
1465 | * |************************************************************************** | |
1466 | * | source-->| shared | private | slave | unbindable | | |
1467 | * | dest | | | | | | |
1468 | * | | | | | | | | |
1469 | * | v | | | | | | |
1470 | * |************************************************************************** | |
1471 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1472 | * | | | | | | | |
1473 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1474 | * *************************************************************************** | |
b90fa9ae RP |
1475 | * A bind operation clones the source mount and mounts the clone on the |
1476 | * destination mount. | |
1477 | * | |
1478 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1479 | * tree of the destination mount and the cloned mount is added to | |
1480 | * the peer group of the source mount. | |
1481 | * (+) the cloned mount is created under the destination mount and is marked | |
1482 | * as shared. The cloned mount is added to the peer group of the source | |
1483 | * mount. | |
5afe0022 RP |
1484 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1485 | * of the destination mount and the cloned mount is made slave | |
1486 | * of the same master as that of the source mount. The cloned mount | |
1487 | * is marked as 'shared and slave'. | |
1488 | * (*) the cloned mount is made a slave of the same master as that of the | |
1489 | * source mount. | |
1490 | * | |
9676f0c6 RP |
1491 | * --------------------------------------------------------------------------- |
1492 | * | MOVE MOUNT OPERATION | | |
1493 | * |************************************************************************** | |
1494 | * | source-->| shared | private | slave | unbindable | | |
1495 | * | dest | | | | | | |
1496 | * | | | | | | | | |
1497 | * | v | | | | | | |
1498 | * |************************************************************************** | |
1499 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1500 | * | | | | | | | |
1501 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1502 | * *************************************************************************** | |
5afe0022 RP |
1503 | * |
1504 | * (+) the mount is moved to the destination. And is then propagated to | |
1505 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1506 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1507 | * (+++) the mount is moved to the destination and is then propagated to |
1508 | * all the mounts belonging to the destination mount's propagation tree. | |
1509 | * the mount is marked as 'shared and slave'. | |
1510 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1511 | * |
1512 | * if the source mount is a tree, the operations explained above is | |
1513 | * applied to each mount in the tree. | |
1514 | * Must be called without spinlocks held, since this function can sleep | |
1515 | * in allocations. | |
1516 | */ | |
0fb54e50 | 1517 | static int attach_recursive_mnt(struct mount *source_mnt, |
1a390689 | 1518 | struct path *path, struct path *parent_path) |
b90fa9ae RP |
1519 | { |
1520 | LIST_HEAD(tree_list); | |
a8d56d8e | 1521 | struct mount *dest_mnt = real_mount(path->mnt); |
1a390689 | 1522 | struct dentry *dest_dentry = path->dentry; |
315fc83e | 1523 | struct mount *child, *p; |
719f5d7f | 1524 | int err; |
b90fa9ae | 1525 | |
fc7be130 | 1526 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1527 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1528 | if (err) |
1529 | goto out; | |
1530 | } | |
a8d56d8e | 1531 | err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); |
719f5d7f MS |
1532 | if (err) |
1533 | goto out_cleanup_ids; | |
b90fa9ae | 1534 | |
962830df | 1535 | br_write_lock(&vfsmount_lock); |
df1a1ad2 | 1536 | |
fc7be130 | 1537 | if (IS_MNT_SHARED(dest_mnt)) { |
909b0a88 | 1538 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1539 | set_mnt_shared(p); |
b90fa9ae | 1540 | } |
1a390689 | 1541 | if (parent_path) { |
0fb54e50 AV |
1542 | detach_mnt(source_mnt, parent_path); |
1543 | attach_mnt(source_mnt, path); | |
143c8c91 | 1544 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1545 | } else { |
14cf1fa8 | 1546 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); |
0fb54e50 | 1547 | commit_tree(source_mnt); |
21444403 | 1548 | } |
b90fa9ae | 1549 | |
1b8e5564 AV |
1550 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { |
1551 | list_del_init(&child->mnt_hash); | |
4b2619a5 | 1552 | commit_tree(child); |
b90fa9ae | 1553 | } |
962830df | 1554 | br_write_unlock(&vfsmount_lock); |
99b7db7b | 1555 | |
b90fa9ae | 1556 | return 0; |
719f5d7f MS |
1557 | |
1558 | out_cleanup_ids: | |
fc7be130 | 1559 | if (IS_MNT_SHARED(dest_mnt)) |
0fb54e50 | 1560 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f MS |
1561 | out: |
1562 | return err; | |
b90fa9ae RP |
1563 | } |
1564 | ||
b12cea91 AV |
1565 | static int lock_mount(struct path *path) |
1566 | { | |
1567 | struct vfsmount *mnt; | |
1568 | retry: | |
1569 | mutex_lock(&path->dentry->d_inode->i_mutex); | |
1570 | if (unlikely(cant_mount(path->dentry))) { | |
1571 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1572 | return -ENOENT; | |
1573 | } | |
1574 | down_write(&namespace_sem); | |
1575 | mnt = lookup_mnt(path); | |
1576 | if (likely(!mnt)) | |
1577 | return 0; | |
1578 | up_write(&namespace_sem); | |
1579 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1580 | path_put(path); | |
1581 | path->mnt = mnt; | |
1582 | path->dentry = dget(mnt->mnt_root); | |
1583 | goto retry; | |
1584 | } | |
1585 | ||
1586 | static void unlock_mount(struct path *path) | |
1587 | { | |
1588 | up_write(&namespace_sem); | |
1589 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1590 | } | |
1591 | ||
95bc5f25 | 1592 | static int graft_tree(struct mount *mnt, struct path *path) |
1da177e4 | 1593 | { |
95bc5f25 | 1594 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) |
1da177e4 LT |
1595 | return -EINVAL; |
1596 | ||
8c3ee42e | 1597 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
95bc5f25 | 1598 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) |
1da177e4 LT |
1599 | return -ENOTDIR; |
1600 | ||
b12cea91 AV |
1601 | if (d_unlinked(path->dentry)) |
1602 | return -ENOENT; | |
1da177e4 | 1603 | |
95bc5f25 | 1604 | return attach_recursive_mnt(mnt, path, NULL); |
1da177e4 LT |
1605 | } |
1606 | ||
7a2e8a8f VA |
1607 | /* |
1608 | * Sanity check the flags to change_mnt_propagation. | |
1609 | */ | |
1610 | ||
1611 | static int flags_to_propagation_type(int flags) | |
1612 | { | |
7c6e984d | 1613 | int type = flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
1614 | |
1615 | /* Fail if any non-propagation flags are set */ | |
1616 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1617 | return 0; | |
1618 | /* Only one propagation flag should be set */ | |
1619 | if (!is_power_of_2(type)) | |
1620 | return 0; | |
1621 | return type; | |
1622 | } | |
1623 | ||
07b20889 RP |
1624 | /* |
1625 | * recursively change the type of the mountpoint. | |
1626 | */ | |
0a0d8a46 | 1627 | static int do_change_type(struct path *path, int flag) |
07b20889 | 1628 | { |
315fc83e | 1629 | struct mount *m; |
4b8b21f4 | 1630 | struct mount *mnt = real_mount(path->mnt); |
07b20889 | 1631 | int recurse = flag & MS_REC; |
7a2e8a8f | 1632 | int type; |
719f5d7f | 1633 | int err = 0; |
07b20889 | 1634 | |
ee6f9582 MS |
1635 | if (!capable(CAP_SYS_ADMIN)) |
1636 | return -EPERM; | |
1637 | ||
2d92ab3c | 1638 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
1639 | return -EINVAL; |
1640 | ||
7a2e8a8f VA |
1641 | type = flags_to_propagation_type(flag); |
1642 | if (!type) | |
1643 | return -EINVAL; | |
1644 | ||
07b20889 | 1645 | down_write(&namespace_sem); |
719f5d7f MS |
1646 | if (type == MS_SHARED) { |
1647 | err = invent_group_ids(mnt, recurse); | |
1648 | if (err) | |
1649 | goto out_unlock; | |
1650 | } | |
1651 | ||
962830df | 1652 | br_write_lock(&vfsmount_lock); |
909b0a88 | 1653 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 1654 | change_mnt_propagation(m, type); |
962830df | 1655 | br_write_unlock(&vfsmount_lock); |
719f5d7f MS |
1656 | |
1657 | out_unlock: | |
07b20889 | 1658 | up_write(&namespace_sem); |
719f5d7f | 1659 | return err; |
07b20889 RP |
1660 | } |
1661 | ||
1da177e4 LT |
1662 | /* |
1663 | * do loopback mount. | |
1664 | */ | |
808d4e3c | 1665 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 1666 | int recurse) |
1da177e4 | 1667 | { |
b12cea91 | 1668 | LIST_HEAD(umount_list); |
2d92ab3c | 1669 | struct path old_path; |
87129cc0 | 1670 | struct mount *mnt = NULL, *old; |
2d92ab3c | 1671 | int err = mount_is_safe(path); |
1da177e4 LT |
1672 | if (err) |
1673 | return err; | |
1674 | if (!old_name || !*old_name) | |
1675 | return -EINVAL; | |
815d405c | 1676 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
1677 | if (err) |
1678 | return err; | |
1679 | ||
8823c079 EB |
1680 | err = -EINVAL; |
1681 | if (mnt_ns_loop(&old_path)) | |
1682 | goto out; | |
1683 | ||
b12cea91 AV |
1684 | err = lock_mount(path); |
1685 | if (err) | |
1686 | goto out; | |
1687 | ||
87129cc0 AV |
1688 | old = real_mount(old_path.mnt); |
1689 | ||
1da177e4 | 1690 | err = -EINVAL; |
fc7be130 | 1691 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 1692 | goto out2; |
9676f0c6 | 1693 | |
143c8c91 | 1694 | if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old)) |
b12cea91 | 1695 | goto out2; |
1da177e4 | 1696 | |
ccd48bc7 | 1697 | if (recurse) |
87129cc0 | 1698 | mnt = copy_tree(old, old_path.dentry, 0); |
ccd48bc7 | 1699 | else |
87129cc0 | 1700 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 1701 | |
be34d1a3 DH |
1702 | if (IS_ERR(mnt)) { |
1703 | err = PTR_ERR(mnt); | |
1704 | goto out; | |
1705 | } | |
ccd48bc7 | 1706 | |
95bc5f25 | 1707 | err = graft_tree(mnt, path); |
ccd48bc7 | 1708 | if (err) { |
962830df | 1709 | br_write_lock(&vfsmount_lock); |
761d5c38 | 1710 | umount_tree(mnt, 0, &umount_list); |
962830df | 1711 | br_write_unlock(&vfsmount_lock); |
5b83d2c5 | 1712 | } |
b12cea91 AV |
1713 | out2: |
1714 | unlock_mount(path); | |
1715 | release_mounts(&umount_list); | |
ccd48bc7 | 1716 | out: |
2d92ab3c | 1717 | path_put(&old_path); |
1da177e4 LT |
1718 | return err; |
1719 | } | |
1720 | ||
2e4b7fcd DH |
1721 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
1722 | { | |
1723 | int error = 0; | |
1724 | int readonly_request = 0; | |
1725 | ||
1726 | if (ms_flags & MS_RDONLY) | |
1727 | readonly_request = 1; | |
1728 | if (readonly_request == __mnt_is_readonly(mnt)) | |
1729 | return 0; | |
1730 | ||
1731 | if (readonly_request) | |
83adc753 | 1732 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 1733 | else |
83adc753 | 1734 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
1735 | return error; |
1736 | } | |
1737 | ||
1da177e4 LT |
1738 | /* |
1739 | * change filesystem flags. dir should be a physical root of filesystem. | |
1740 | * If you've mounted a non-root directory somewhere and want to do remount | |
1741 | * on it - tough luck. | |
1742 | */ | |
0a0d8a46 | 1743 | static int do_remount(struct path *path, int flags, int mnt_flags, |
1da177e4 LT |
1744 | void *data) |
1745 | { | |
1746 | int err; | |
2d92ab3c | 1747 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 1748 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 LT |
1749 | |
1750 | if (!capable(CAP_SYS_ADMIN)) | |
1751 | return -EPERM; | |
1752 | ||
143c8c91 | 1753 | if (!check_mnt(mnt)) |
1da177e4 LT |
1754 | return -EINVAL; |
1755 | ||
2d92ab3c | 1756 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
1757 | return -EINVAL; |
1758 | ||
ff36fe2c EP |
1759 | err = security_sb_remount(sb, data); |
1760 | if (err) | |
1761 | return err; | |
1762 | ||
1da177e4 | 1763 | down_write(&sb->s_umount); |
2e4b7fcd | 1764 | if (flags & MS_BIND) |
2d92ab3c | 1765 | err = change_mount_flags(path->mnt, flags); |
4aa98cf7 | 1766 | else |
2e4b7fcd | 1767 | err = do_remount_sb(sb, flags, data, 0); |
7b43a79f | 1768 | if (!err) { |
962830df | 1769 | br_write_lock(&vfsmount_lock); |
143c8c91 AV |
1770 | mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; |
1771 | mnt->mnt.mnt_flags = mnt_flags; | |
962830df | 1772 | br_write_unlock(&vfsmount_lock); |
7b43a79f | 1773 | } |
1da177e4 | 1774 | up_write(&sb->s_umount); |
0e55a7cc | 1775 | if (!err) { |
962830df | 1776 | br_write_lock(&vfsmount_lock); |
143c8c91 | 1777 | touch_mnt_namespace(mnt->mnt_ns); |
962830df | 1778 | br_write_unlock(&vfsmount_lock); |
0e55a7cc | 1779 | } |
1da177e4 LT |
1780 | return err; |
1781 | } | |
1782 | ||
cbbe362c | 1783 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 1784 | { |
315fc83e | 1785 | struct mount *p; |
909b0a88 | 1786 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 1787 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
1788 | return 1; |
1789 | } | |
1790 | return 0; | |
1791 | } | |
1792 | ||
808d4e3c | 1793 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 1794 | { |
2d92ab3c | 1795 | struct path old_path, parent_path; |
676da58d | 1796 | struct mount *p; |
0fb54e50 | 1797 | struct mount *old; |
1da177e4 LT |
1798 | int err = 0; |
1799 | if (!capable(CAP_SYS_ADMIN)) | |
1800 | return -EPERM; | |
1801 | if (!old_name || !*old_name) | |
1802 | return -EINVAL; | |
2d92ab3c | 1803 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
1804 | if (err) |
1805 | return err; | |
1806 | ||
b12cea91 | 1807 | err = lock_mount(path); |
cc53ce53 DH |
1808 | if (err < 0) |
1809 | goto out; | |
1810 | ||
143c8c91 | 1811 | old = real_mount(old_path.mnt); |
fc7be130 | 1812 | p = real_mount(path->mnt); |
143c8c91 | 1813 | |
1da177e4 | 1814 | err = -EINVAL; |
fc7be130 | 1815 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
1816 | goto out1; |
1817 | ||
f3da392e | 1818 | if (d_unlinked(path->dentry)) |
21444403 | 1819 | goto out1; |
1da177e4 LT |
1820 | |
1821 | err = -EINVAL; | |
2d92ab3c | 1822 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 1823 | goto out1; |
1da177e4 | 1824 | |
676da58d | 1825 | if (!mnt_has_parent(old)) |
21444403 | 1826 | goto out1; |
1da177e4 | 1827 | |
2d92ab3c AV |
1828 | if (S_ISDIR(path->dentry->d_inode->i_mode) != |
1829 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
21444403 RP |
1830 | goto out1; |
1831 | /* | |
1832 | * Don't move a mount residing in a shared parent. | |
1833 | */ | |
fc7be130 | 1834 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 1835 | goto out1; |
9676f0c6 RP |
1836 | /* |
1837 | * Don't move a mount tree containing unbindable mounts to a destination | |
1838 | * mount which is shared. | |
1839 | */ | |
fc7be130 | 1840 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 1841 | goto out1; |
1da177e4 | 1842 | err = -ELOOP; |
fc7be130 | 1843 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 1844 | if (p == old) |
21444403 | 1845 | goto out1; |
1da177e4 | 1846 | |
0fb54e50 | 1847 | err = attach_recursive_mnt(old, path, &parent_path); |
4ac91378 | 1848 | if (err) |
21444403 | 1849 | goto out1; |
1da177e4 LT |
1850 | |
1851 | /* if the mount is moved, it should no longer be expire | |
1852 | * automatically */ | |
6776db3d | 1853 | list_del_init(&old->mnt_expire); |
1da177e4 | 1854 | out1: |
b12cea91 | 1855 | unlock_mount(path); |
1da177e4 | 1856 | out: |
1da177e4 | 1857 | if (!err) |
1a390689 | 1858 | path_put(&parent_path); |
2d92ab3c | 1859 | path_put(&old_path); |
1da177e4 LT |
1860 | return err; |
1861 | } | |
1862 | ||
9d412a43 AV |
1863 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
1864 | { | |
1865 | int err; | |
1866 | const char *subtype = strchr(fstype, '.'); | |
1867 | if (subtype) { | |
1868 | subtype++; | |
1869 | err = -EINVAL; | |
1870 | if (!subtype[0]) | |
1871 | goto err; | |
1872 | } else | |
1873 | subtype = ""; | |
1874 | ||
1875 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
1876 | err = -ENOMEM; | |
1877 | if (!mnt->mnt_sb->s_subtype) | |
1878 | goto err; | |
1879 | return mnt; | |
1880 | ||
1881 | err: | |
1882 | mntput(mnt); | |
1883 | return ERR_PTR(err); | |
1884 | } | |
1885 | ||
79e801a9 | 1886 | static struct vfsmount * |
9d412a43 AV |
1887 | do_kern_mount(const char *fstype, int flags, const char *name, void *data) |
1888 | { | |
1889 | struct file_system_type *type = get_fs_type(fstype); | |
1890 | struct vfsmount *mnt; | |
1891 | if (!type) | |
1892 | return ERR_PTR(-ENODEV); | |
1893 | mnt = vfs_kern_mount(type, flags, name, data); | |
1894 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
1895 | !mnt->mnt_sb->s_subtype) | |
1896 | mnt = fs_set_subtype(mnt, fstype); | |
1897 | put_filesystem(type); | |
1898 | return mnt; | |
1899 | } | |
9d412a43 AV |
1900 | |
1901 | /* | |
1902 | * add a mount into a namespace's mount tree | |
1903 | */ | |
95bc5f25 | 1904 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 AV |
1905 | { |
1906 | int err; | |
1907 | ||
1908 | mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); | |
1909 | ||
b12cea91 AV |
1910 | err = lock_mount(path); |
1911 | if (err) | |
1912 | return err; | |
9d412a43 AV |
1913 | |
1914 | err = -EINVAL; | |
156cacb1 AV |
1915 | if (unlikely(!check_mnt(real_mount(path->mnt)))) { |
1916 | /* that's acceptable only for automounts done in private ns */ | |
1917 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
1918 | goto unlock; | |
1919 | /* ... and for those we'd better have mountpoint still alive */ | |
1920 | if (!real_mount(path->mnt)->mnt_ns) | |
1921 | goto unlock; | |
1922 | } | |
9d412a43 AV |
1923 | |
1924 | /* Refuse the same filesystem on the same mount point */ | |
1925 | err = -EBUSY; | |
95bc5f25 | 1926 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
1927 | path->mnt->mnt_root == path->dentry) |
1928 | goto unlock; | |
1929 | ||
1930 | err = -EINVAL; | |
95bc5f25 | 1931 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) |
9d412a43 AV |
1932 | goto unlock; |
1933 | ||
95bc5f25 | 1934 | newmnt->mnt.mnt_flags = mnt_flags; |
9d412a43 AV |
1935 | err = graft_tree(newmnt, path); |
1936 | ||
1937 | unlock: | |
b12cea91 | 1938 | unlock_mount(path); |
9d412a43 AV |
1939 | return err; |
1940 | } | |
b1e75df4 | 1941 | |
1da177e4 LT |
1942 | /* |
1943 | * create a new mount for userspace and request it to be added into the | |
1944 | * namespace's tree | |
1945 | */ | |
808d4e3c AV |
1946 | static int do_new_mount(struct path *path, const char *type, int flags, |
1947 | int mnt_flags, const char *name, void *data) | |
1da177e4 LT |
1948 | { |
1949 | struct vfsmount *mnt; | |
15f9a3f3 | 1950 | int err; |
1da177e4 | 1951 | |
eca6f534 | 1952 | if (!type) |
1da177e4 LT |
1953 | return -EINVAL; |
1954 | ||
1955 | /* we need capabilities... */ | |
1956 | if (!capable(CAP_SYS_ADMIN)) | |
1957 | return -EPERM; | |
1958 | ||
1959 | mnt = do_kern_mount(type, flags, name, data); | |
1960 | if (IS_ERR(mnt)) | |
1961 | return PTR_ERR(mnt); | |
1962 | ||
95bc5f25 | 1963 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
1964 | if (err) |
1965 | mntput(mnt); | |
1966 | return err; | |
1da177e4 LT |
1967 | } |
1968 | ||
19a167af AV |
1969 | int finish_automount(struct vfsmount *m, struct path *path) |
1970 | { | |
6776db3d | 1971 | struct mount *mnt = real_mount(m); |
19a167af AV |
1972 | int err; |
1973 | /* The new mount record should have at least 2 refs to prevent it being | |
1974 | * expired before we get a chance to add it | |
1975 | */ | |
6776db3d | 1976 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
1977 | |
1978 | if (m->mnt_sb == path->mnt->mnt_sb && | |
1979 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
1980 | err = -ELOOP; |
1981 | goto fail; | |
19a167af AV |
1982 | } |
1983 | ||
95bc5f25 | 1984 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
1985 | if (!err) |
1986 | return 0; | |
1987 | fail: | |
1988 | /* remove m from any expiration list it may be on */ | |
6776db3d | 1989 | if (!list_empty(&mnt->mnt_expire)) { |
b1e75df4 | 1990 | down_write(&namespace_sem); |
962830df | 1991 | br_write_lock(&vfsmount_lock); |
6776db3d | 1992 | list_del_init(&mnt->mnt_expire); |
962830df | 1993 | br_write_unlock(&vfsmount_lock); |
b1e75df4 | 1994 | up_write(&namespace_sem); |
19a167af | 1995 | } |
b1e75df4 AV |
1996 | mntput(m); |
1997 | mntput(m); | |
19a167af AV |
1998 | return err; |
1999 | } | |
2000 | ||
ea5b778a DH |
2001 | /** |
2002 | * mnt_set_expiry - Put a mount on an expiration list | |
2003 | * @mnt: The mount to list. | |
2004 | * @expiry_list: The list to add the mount to. | |
2005 | */ | |
2006 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2007 | { | |
2008 | down_write(&namespace_sem); | |
962830df | 2009 | br_write_lock(&vfsmount_lock); |
ea5b778a | 2010 | |
6776db3d | 2011 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2012 | |
962830df | 2013 | br_write_unlock(&vfsmount_lock); |
ea5b778a DH |
2014 | up_write(&namespace_sem); |
2015 | } | |
2016 | EXPORT_SYMBOL(mnt_set_expiry); | |
2017 | ||
1da177e4 LT |
2018 | /* |
2019 | * process a list of expirable mountpoints with the intent of discarding any | |
2020 | * mountpoints that aren't in use and haven't been touched since last we came | |
2021 | * here | |
2022 | */ | |
2023 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2024 | { | |
761d5c38 | 2025 | struct mount *mnt, *next; |
1da177e4 | 2026 | LIST_HEAD(graveyard); |
bcc5c7d2 | 2027 | LIST_HEAD(umounts); |
1da177e4 LT |
2028 | |
2029 | if (list_empty(mounts)) | |
2030 | return; | |
2031 | ||
bcc5c7d2 | 2032 | down_write(&namespace_sem); |
962830df | 2033 | br_write_lock(&vfsmount_lock); |
1da177e4 LT |
2034 | |
2035 | /* extract from the expiration list every vfsmount that matches the | |
2036 | * following criteria: | |
2037 | * - only referenced by its parent vfsmount | |
2038 | * - still marked for expiry (marked on the last call here; marks are | |
2039 | * cleared by mntput()) | |
2040 | */ | |
6776db3d | 2041 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2042 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2043 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2044 | continue; |
6776db3d | 2045 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2046 | } |
bcc5c7d2 | 2047 | while (!list_empty(&graveyard)) { |
6776db3d | 2048 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2049 | touch_mnt_namespace(mnt->mnt_ns); |
bcc5c7d2 AV |
2050 | umount_tree(mnt, 1, &umounts); |
2051 | } | |
962830df | 2052 | br_write_unlock(&vfsmount_lock); |
bcc5c7d2 AV |
2053 | up_write(&namespace_sem); |
2054 | ||
2055 | release_mounts(&umounts); | |
5528f911 TM |
2056 | } |
2057 | ||
2058 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2059 | ||
2060 | /* | |
2061 | * Ripoff of 'select_parent()' | |
2062 | * | |
2063 | * search the list of submounts for a given mountpoint, and move any | |
2064 | * shrinkable submounts to the 'graveyard' list. | |
2065 | */ | |
692afc31 | 2066 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2067 | { |
692afc31 | 2068 | struct mount *this_parent = parent; |
5528f911 TM |
2069 | struct list_head *next; |
2070 | int found = 0; | |
2071 | ||
2072 | repeat: | |
6b41d536 | 2073 | next = this_parent->mnt_mounts.next; |
5528f911 | 2074 | resume: |
6b41d536 | 2075 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2076 | struct list_head *tmp = next; |
6b41d536 | 2077 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2078 | |
2079 | next = tmp->next; | |
692afc31 | 2080 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2081 | continue; |
5528f911 TM |
2082 | /* |
2083 | * Descend a level if the d_mounts list is non-empty. | |
2084 | */ | |
6b41d536 | 2085 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2086 | this_parent = mnt; |
2087 | goto repeat; | |
2088 | } | |
1da177e4 | 2089 | |
1ab59738 | 2090 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2091 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2092 | found++; |
2093 | } | |
1da177e4 | 2094 | } |
5528f911 TM |
2095 | /* |
2096 | * All done at this level ... ascend and resume the search | |
2097 | */ | |
2098 | if (this_parent != parent) { | |
6b41d536 | 2099 | next = this_parent->mnt_child.next; |
0714a533 | 2100 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2101 | goto resume; |
2102 | } | |
2103 | return found; | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * process a list of expirable mountpoints with the intent of discarding any | |
2108 | * submounts of a specific parent mountpoint | |
99b7db7b NP |
2109 | * |
2110 | * vfsmount_lock must be held for write | |
5528f911 | 2111 | */ |
692afc31 | 2112 | static void shrink_submounts(struct mount *mnt, struct list_head *umounts) |
5528f911 TM |
2113 | { |
2114 | LIST_HEAD(graveyard); | |
761d5c38 | 2115 | struct mount *m; |
5528f911 | 2116 | |
5528f911 | 2117 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2118 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2119 | while (!list_empty(&graveyard)) { |
761d5c38 | 2120 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2121 | mnt_expire); |
143c8c91 | 2122 | touch_mnt_namespace(m->mnt_ns); |
afef80b3 | 2123 | umount_tree(m, 1, umounts); |
bcc5c7d2 AV |
2124 | } |
2125 | } | |
1da177e4 LT |
2126 | } |
2127 | ||
1da177e4 LT |
2128 | /* |
2129 | * Some copy_from_user() implementations do not return the exact number of | |
2130 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2131 | * Note that this function differs from copy_from_user() in that it will oops | |
2132 | * on bad values of `to', rather than returning a short copy. | |
2133 | */ | |
b58fed8b RP |
2134 | static long exact_copy_from_user(void *to, const void __user * from, |
2135 | unsigned long n) | |
1da177e4 LT |
2136 | { |
2137 | char *t = to; | |
2138 | const char __user *f = from; | |
2139 | char c; | |
2140 | ||
2141 | if (!access_ok(VERIFY_READ, from, n)) | |
2142 | return n; | |
2143 | ||
2144 | while (n) { | |
2145 | if (__get_user(c, f)) { | |
2146 | memset(t, 0, n); | |
2147 | break; | |
2148 | } | |
2149 | *t++ = c; | |
2150 | f++; | |
2151 | n--; | |
2152 | } | |
2153 | return n; | |
2154 | } | |
2155 | ||
b58fed8b | 2156 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
2157 | { |
2158 | int i; | |
2159 | unsigned long page; | |
2160 | unsigned long size; | |
b58fed8b | 2161 | |
1da177e4 LT |
2162 | *where = 0; |
2163 | if (!data) | |
2164 | return 0; | |
2165 | ||
2166 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2167 | return -ENOMEM; | |
2168 | ||
2169 | /* We only care that *some* data at the address the user | |
2170 | * gave us is valid. Just in case, we'll zero | |
2171 | * the remainder of the page. | |
2172 | */ | |
2173 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2174 | size = TASK_SIZE - (unsigned long)data; | |
2175 | if (size > PAGE_SIZE) | |
2176 | size = PAGE_SIZE; | |
2177 | ||
2178 | i = size - exact_copy_from_user((void *)page, data, size); | |
2179 | if (!i) { | |
b58fed8b | 2180 | free_page(page); |
1da177e4 LT |
2181 | return -EFAULT; |
2182 | } | |
2183 | if (i != PAGE_SIZE) | |
2184 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2185 | *where = page; | |
2186 | return 0; | |
2187 | } | |
2188 | ||
eca6f534 VN |
2189 | int copy_mount_string(const void __user *data, char **where) |
2190 | { | |
2191 | char *tmp; | |
2192 | ||
2193 | if (!data) { | |
2194 | *where = NULL; | |
2195 | return 0; | |
2196 | } | |
2197 | ||
2198 | tmp = strndup_user(data, PAGE_SIZE); | |
2199 | if (IS_ERR(tmp)) | |
2200 | return PTR_ERR(tmp); | |
2201 | ||
2202 | *where = tmp; | |
2203 | return 0; | |
2204 | } | |
2205 | ||
1da177e4 LT |
2206 | /* |
2207 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2208 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2209 | * | |
2210 | * data is a (void *) that can point to any structure up to | |
2211 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2212 | * information (or be NULL). | |
2213 | * | |
2214 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2215 | * When the flags word was introduced its top half was required | |
2216 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2217 | * Therefore, if this magic number is present, it carries no information | |
2218 | * and must be discarded. | |
2219 | */ | |
808d4e3c AV |
2220 | long do_mount(const char *dev_name, const char *dir_name, |
2221 | const char *type_page, unsigned long flags, void *data_page) | |
1da177e4 | 2222 | { |
2d92ab3c | 2223 | struct path path; |
1da177e4 LT |
2224 | int retval = 0; |
2225 | int mnt_flags = 0; | |
2226 | ||
2227 | /* Discard magic */ | |
2228 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2229 | flags &= ~MS_MGC_MSK; | |
2230 | ||
2231 | /* Basic sanity checks */ | |
2232 | ||
2233 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
2234 | return -EINVAL; | |
1da177e4 LT |
2235 | |
2236 | if (data_page) | |
2237 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2238 | ||
a27ab9f2 TH |
2239 | /* ... and get the mountpoint */ |
2240 | retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); | |
2241 | if (retval) | |
2242 | return retval; | |
2243 | ||
2244 | retval = security_sb_mount(dev_name, &path, | |
2245 | type_page, flags, data_page); | |
2246 | if (retval) | |
2247 | goto dput_out; | |
2248 | ||
613cbe3d AK |
2249 | /* Default to relatime unless overriden */ |
2250 | if (!(flags & MS_NOATIME)) | |
2251 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2252 | |
1da177e4 LT |
2253 | /* Separate the per-mountpoint flags */ |
2254 | if (flags & MS_NOSUID) | |
2255 | mnt_flags |= MNT_NOSUID; | |
2256 | if (flags & MS_NODEV) | |
2257 | mnt_flags |= MNT_NODEV; | |
2258 | if (flags & MS_NOEXEC) | |
2259 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2260 | if (flags & MS_NOATIME) |
2261 | mnt_flags |= MNT_NOATIME; | |
2262 | if (flags & MS_NODIRATIME) | |
2263 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2264 | if (flags & MS_STRICTATIME) |
2265 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2e4b7fcd DH |
2266 | if (flags & MS_RDONLY) |
2267 | mnt_flags |= MNT_READONLY; | |
fc33a7bb | 2268 | |
7a4dec53 | 2269 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | |
d0adde57 MG |
2270 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | |
2271 | MS_STRICTATIME); | |
1da177e4 | 2272 | |
1da177e4 | 2273 | if (flags & MS_REMOUNT) |
2d92ab3c | 2274 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, |
1da177e4 LT |
2275 | data_page); |
2276 | else if (flags & MS_BIND) | |
2d92ab3c | 2277 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2278 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2279 | retval = do_change_type(&path, flags); |
1da177e4 | 2280 | else if (flags & MS_MOVE) |
2d92ab3c | 2281 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2282 | else |
2d92ab3c | 2283 | retval = do_new_mount(&path, type_page, flags, mnt_flags, |
1da177e4 LT |
2284 | dev_name, data_page); |
2285 | dput_out: | |
2d92ab3c | 2286 | path_put(&path); |
1da177e4 LT |
2287 | return retval; |
2288 | } | |
2289 | ||
771b1371 EB |
2290 | static void free_mnt_ns(struct mnt_namespace *ns) |
2291 | { | |
2292 | put_user_ns(ns->user_ns); | |
2293 | kfree(ns); | |
2294 | } | |
2295 | ||
8823c079 EB |
2296 | /* |
2297 | * Assign a sequence number so we can detect when we attempt to bind | |
2298 | * mount a reference to an older mount namespace into the current | |
2299 | * mount namespace, preventing reference counting loops. A 64bit | |
2300 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2301 | * is effectively never, so we can ignore the possibility. | |
2302 | */ | |
2303 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2304 | ||
771b1371 | 2305 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2306 | { |
2307 | struct mnt_namespace *new_ns; | |
2308 | ||
2309 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2310 | if (!new_ns) | |
2311 | return ERR_PTR(-ENOMEM); | |
8823c079 | 2312 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2313 | atomic_set(&new_ns->count, 1); |
2314 | new_ns->root = NULL; | |
2315 | INIT_LIST_HEAD(&new_ns->list); | |
2316 | init_waitqueue_head(&new_ns->poll); | |
2317 | new_ns->event = 0; | |
771b1371 | 2318 | new_ns->user_ns = get_user_ns(user_ns); |
cf8d2c11 TM |
2319 | return new_ns; |
2320 | } | |
2321 | ||
741a2951 JD |
2322 | /* |
2323 | * Allocate a new namespace structure and populate it with contents | |
2324 | * copied from the namespace of the passed in task structure. | |
2325 | */ | |
e3222c4e | 2326 | static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, |
771b1371 | 2327 | struct user_namespace *user_ns, struct fs_struct *fs) |
1da177e4 | 2328 | { |
6b3286ed | 2329 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2330 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2331 | struct mount *p, *q; |
be08d6d2 | 2332 | struct mount *old = mnt_ns->root; |
cb338d06 | 2333 | struct mount *new; |
1da177e4 | 2334 | |
771b1371 | 2335 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2336 | if (IS_ERR(new_ns)) |
2337 | return new_ns; | |
1da177e4 | 2338 | |
390c6843 | 2339 | down_write(&namespace_sem); |
1da177e4 | 2340 | /* First pass: copy the tree topology */ |
909b0a88 | 2341 | new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE); |
be34d1a3 | 2342 | if (IS_ERR(new)) { |
390c6843 | 2343 | up_write(&namespace_sem); |
771b1371 | 2344 | free_mnt_ns(new_ns); |
be34d1a3 | 2345 | return ERR_CAST(new); |
1da177e4 | 2346 | } |
be08d6d2 | 2347 | new_ns->root = new; |
962830df | 2348 | br_write_lock(&vfsmount_lock); |
1a4eeaf2 | 2349 | list_add_tail(&new_ns->list, &new->mnt_list); |
962830df | 2350 | br_write_unlock(&vfsmount_lock); |
1da177e4 LT |
2351 | |
2352 | /* | |
2353 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2354 | * as belonging to new namespace. We have already acquired a private | |
2355 | * fs_struct, so tsk->fs->lock is not needed. | |
2356 | */ | |
909b0a88 | 2357 | p = old; |
cb338d06 | 2358 | q = new; |
1da177e4 | 2359 | while (p) { |
143c8c91 | 2360 | q->mnt_ns = new_ns; |
1da177e4 | 2361 | if (fs) { |
315fc83e AV |
2362 | if (&p->mnt == fs->root.mnt) { |
2363 | fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2364 | rootmnt = &p->mnt; |
1da177e4 | 2365 | } |
315fc83e AV |
2366 | if (&p->mnt == fs->pwd.mnt) { |
2367 | fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2368 | pwdmnt = &p->mnt; |
1da177e4 | 2369 | } |
1da177e4 | 2370 | } |
909b0a88 AV |
2371 | p = next_mnt(p, old); |
2372 | q = next_mnt(q, new); | |
1da177e4 | 2373 | } |
390c6843 | 2374 | up_write(&namespace_sem); |
1da177e4 | 2375 | |
1da177e4 | 2376 | if (rootmnt) |
f03c6599 | 2377 | mntput(rootmnt); |
1da177e4 | 2378 | if (pwdmnt) |
f03c6599 | 2379 | mntput(pwdmnt); |
1da177e4 | 2380 | |
741a2951 JD |
2381 | return new_ns; |
2382 | } | |
2383 | ||
213dd266 | 2384 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
771b1371 | 2385 | struct user_namespace *user_ns, struct fs_struct *new_fs) |
741a2951 | 2386 | { |
6b3286ed | 2387 | struct mnt_namespace *new_ns; |
741a2951 | 2388 | |
e3222c4e | 2389 | BUG_ON(!ns); |
6b3286ed | 2390 | get_mnt_ns(ns); |
741a2951 JD |
2391 | |
2392 | if (!(flags & CLONE_NEWNS)) | |
e3222c4e | 2393 | return ns; |
741a2951 | 2394 | |
771b1371 | 2395 | new_ns = dup_mnt_ns(ns, user_ns, new_fs); |
741a2951 | 2396 | |
6b3286ed | 2397 | put_mnt_ns(ns); |
e3222c4e | 2398 | return new_ns; |
1da177e4 LT |
2399 | } |
2400 | ||
cf8d2c11 TM |
2401 | /** |
2402 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2403 | * @mnt: pointer to the new root filesystem mountpoint | |
2404 | */ | |
1a4eeaf2 | 2405 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2406 | { |
771b1371 | 2407 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2408 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2409 | struct mount *mnt = real_mount(m); |
2410 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2411 | new_ns->root = mnt; |
1a4eeaf2 | 2412 | list_add(&new_ns->list, &mnt->mnt_list); |
c1334495 | 2413 | } else { |
1a4eeaf2 | 2414 | mntput(m); |
cf8d2c11 TM |
2415 | } |
2416 | return new_ns; | |
2417 | } | |
cf8d2c11 | 2418 | |
ea441d11 AV |
2419 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
2420 | { | |
2421 | struct mnt_namespace *ns; | |
d31da0f0 | 2422 | struct super_block *s; |
ea441d11 AV |
2423 | struct path path; |
2424 | int err; | |
2425 | ||
2426 | ns = create_mnt_ns(mnt); | |
2427 | if (IS_ERR(ns)) | |
2428 | return ERR_CAST(ns); | |
2429 | ||
2430 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2431 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2432 | ||
2433 | put_mnt_ns(ns); | |
2434 | ||
2435 | if (err) | |
2436 | return ERR_PTR(err); | |
2437 | ||
2438 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
2439 | s = path.mnt->mnt_sb; |
2440 | atomic_inc(&s->s_active); | |
ea441d11 AV |
2441 | mntput(path.mnt); |
2442 | /* lock the sucker */ | |
d31da0f0 | 2443 | down_write(&s->s_umount); |
ea441d11 AV |
2444 | /* ... and return the root of (sub)tree on it */ |
2445 | return path.dentry; | |
2446 | } | |
2447 | EXPORT_SYMBOL(mount_subtree); | |
2448 | ||
bdc480e3 HC |
2449 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
2450 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 2451 | { |
eca6f534 VN |
2452 | int ret; |
2453 | char *kernel_type; | |
91a27b2a | 2454 | struct filename *kernel_dir; |
eca6f534 | 2455 | char *kernel_dev; |
1da177e4 | 2456 | unsigned long data_page; |
1da177e4 | 2457 | |
eca6f534 VN |
2458 | ret = copy_mount_string(type, &kernel_type); |
2459 | if (ret < 0) | |
2460 | goto out_type; | |
1da177e4 | 2461 | |
eca6f534 VN |
2462 | kernel_dir = getname(dir_name); |
2463 | if (IS_ERR(kernel_dir)) { | |
2464 | ret = PTR_ERR(kernel_dir); | |
2465 | goto out_dir; | |
2466 | } | |
1da177e4 | 2467 | |
eca6f534 VN |
2468 | ret = copy_mount_string(dev_name, &kernel_dev); |
2469 | if (ret < 0) | |
2470 | goto out_dev; | |
1da177e4 | 2471 | |
eca6f534 VN |
2472 | ret = copy_mount_options(data, &data_page); |
2473 | if (ret < 0) | |
2474 | goto out_data; | |
1da177e4 | 2475 | |
91a27b2a | 2476 | ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, |
eca6f534 | 2477 | (void *) data_page); |
1da177e4 | 2478 | |
eca6f534 VN |
2479 | free_page(data_page); |
2480 | out_data: | |
2481 | kfree(kernel_dev); | |
2482 | out_dev: | |
2483 | putname(kernel_dir); | |
2484 | out_dir: | |
2485 | kfree(kernel_type); | |
2486 | out_type: | |
2487 | return ret; | |
1da177e4 LT |
2488 | } |
2489 | ||
afac7cba AV |
2490 | /* |
2491 | * Return true if path is reachable from root | |
2492 | * | |
2493 | * namespace_sem or vfsmount_lock is held | |
2494 | */ | |
643822b4 | 2495 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
2496 | const struct path *root) |
2497 | { | |
643822b4 | 2498 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 2499 | dentry = mnt->mnt_mountpoint; |
0714a533 | 2500 | mnt = mnt->mnt_parent; |
afac7cba | 2501 | } |
643822b4 | 2502 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
2503 | } |
2504 | ||
2505 | int path_is_under(struct path *path1, struct path *path2) | |
2506 | { | |
2507 | int res; | |
962830df | 2508 | br_read_lock(&vfsmount_lock); |
643822b4 | 2509 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
962830df | 2510 | br_read_unlock(&vfsmount_lock); |
afac7cba AV |
2511 | return res; |
2512 | } | |
2513 | EXPORT_SYMBOL(path_is_under); | |
2514 | ||
1da177e4 LT |
2515 | /* |
2516 | * pivot_root Semantics: | |
2517 | * Moves the root file system of the current process to the directory put_old, | |
2518 | * makes new_root as the new root file system of the current process, and sets | |
2519 | * root/cwd of all processes which had them on the current root to new_root. | |
2520 | * | |
2521 | * Restrictions: | |
2522 | * The new_root and put_old must be directories, and must not be on the | |
2523 | * same file system as the current process root. The put_old must be | |
2524 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2525 | * pointed to by put_old must yield the same directory as new_root. No other | |
2526 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2527 | * | |
4a0d11fa NB |
2528 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
2529 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2530 | * in this situation. | |
2531 | * | |
1da177e4 LT |
2532 | * Notes: |
2533 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2534 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2535 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2536 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2537 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2538 | * first. | |
2539 | */ | |
3480b257 HC |
2540 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
2541 | const char __user *, put_old) | |
1da177e4 | 2542 | { |
2d8f3038 | 2543 | struct path new, old, parent_path, root_parent, root; |
419148da | 2544 | struct mount *new_mnt, *root_mnt; |
1da177e4 LT |
2545 | int error; |
2546 | ||
2547 | if (!capable(CAP_SYS_ADMIN)) | |
2548 | return -EPERM; | |
2549 | ||
2d8f3038 | 2550 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
2551 | if (error) |
2552 | goto out0; | |
1da177e4 | 2553 | |
2d8f3038 | 2554 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
2555 | if (error) |
2556 | goto out1; | |
2557 | ||
2d8f3038 | 2558 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
2559 | if (error) |
2560 | goto out2; | |
1da177e4 | 2561 | |
f7ad3c6b | 2562 | get_fs_root(current->fs, &root); |
b12cea91 AV |
2563 | error = lock_mount(&old); |
2564 | if (error) | |
2565 | goto out3; | |
2566 | ||
1da177e4 | 2567 | error = -EINVAL; |
419148da AV |
2568 | new_mnt = real_mount(new.mnt); |
2569 | root_mnt = real_mount(root.mnt); | |
fc7be130 AV |
2570 | if (IS_MNT_SHARED(real_mount(old.mnt)) || |
2571 | IS_MNT_SHARED(new_mnt->mnt_parent) || | |
2572 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 2573 | goto out4; |
143c8c91 | 2574 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 2575 | goto out4; |
1da177e4 | 2576 | error = -ENOENT; |
f3da392e | 2577 | if (d_unlinked(new.dentry)) |
b12cea91 | 2578 | goto out4; |
f3da392e | 2579 | if (d_unlinked(old.dentry)) |
b12cea91 | 2580 | goto out4; |
1da177e4 | 2581 | error = -EBUSY; |
2d8f3038 AV |
2582 | if (new.mnt == root.mnt || |
2583 | old.mnt == root.mnt) | |
b12cea91 | 2584 | goto out4; /* loop, on the same file system */ |
1da177e4 | 2585 | error = -EINVAL; |
8c3ee42e | 2586 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 2587 | goto out4; /* not a mountpoint */ |
676da58d | 2588 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 2589 | goto out4; /* not attached */ |
2d8f3038 | 2590 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 2591 | goto out4; /* not a mountpoint */ |
676da58d | 2592 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 2593 | goto out4; /* not attached */ |
4ac91378 | 2594 | /* make sure we can reach put_old from new_root */ |
643822b4 | 2595 | if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new)) |
b12cea91 | 2596 | goto out4; |
962830df | 2597 | br_write_lock(&vfsmount_lock); |
419148da AV |
2598 | detach_mnt(new_mnt, &parent_path); |
2599 | detach_mnt(root_mnt, &root_parent); | |
4ac91378 | 2600 | /* mount old root on put_old */ |
419148da | 2601 | attach_mnt(root_mnt, &old); |
4ac91378 | 2602 | /* mount new_root on / */ |
419148da | 2603 | attach_mnt(new_mnt, &root_parent); |
6b3286ed | 2604 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
962830df | 2605 | br_write_unlock(&vfsmount_lock); |
2d8f3038 | 2606 | chroot_fs_refs(&root, &new); |
1da177e4 | 2607 | error = 0; |
b12cea91 AV |
2608 | out4: |
2609 | unlock_mount(&old); | |
2610 | if (!error) { | |
2611 | path_put(&root_parent); | |
2612 | path_put(&parent_path); | |
2613 | } | |
2614 | out3: | |
8c3ee42e | 2615 | path_put(&root); |
b12cea91 | 2616 | out2: |
2d8f3038 | 2617 | path_put(&old); |
1da177e4 | 2618 | out1: |
2d8f3038 | 2619 | path_put(&new); |
1da177e4 | 2620 | out0: |
1da177e4 | 2621 | return error; |
1da177e4 LT |
2622 | } |
2623 | ||
2624 | static void __init init_mount_tree(void) | |
2625 | { | |
2626 | struct vfsmount *mnt; | |
6b3286ed | 2627 | struct mnt_namespace *ns; |
ac748a09 | 2628 | struct path root; |
1da177e4 LT |
2629 | |
2630 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
2631 | if (IS_ERR(mnt)) | |
2632 | panic("Can't create rootfs"); | |
b3e19d92 | 2633 | |
3b22edc5 TM |
2634 | ns = create_mnt_ns(mnt); |
2635 | if (IS_ERR(ns)) | |
1da177e4 | 2636 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
2637 | |
2638 | init_task.nsproxy->mnt_ns = ns; | |
2639 | get_mnt_ns(ns); | |
2640 | ||
be08d6d2 AV |
2641 | root.mnt = mnt; |
2642 | root.dentry = mnt->mnt_root; | |
ac748a09 JB |
2643 | |
2644 | set_fs_pwd(current->fs, &root); | |
2645 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
2646 | } |
2647 | ||
74bf17cf | 2648 | void __init mnt_init(void) |
1da177e4 | 2649 | { |
13f14b4d | 2650 | unsigned u; |
15a67dd8 | 2651 | int err; |
1da177e4 | 2652 | |
390c6843 RP |
2653 | init_rwsem(&namespace_sem); |
2654 | ||
7d6fec45 | 2655 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 2656 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 2657 | |
b58fed8b | 2658 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
2659 | |
2660 | if (!mount_hashtable) | |
2661 | panic("Failed to allocate mount hash table\n"); | |
2662 | ||
80cdc6da | 2663 | printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); |
13f14b4d ED |
2664 | |
2665 | for (u = 0; u < HASH_SIZE; u++) | |
2666 | INIT_LIST_HEAD(&mount_hashtable[u]); | |
1da177e4 | 2667 | |
962830df | 2668 | br_lock_init(&vfsmount_lock); |
99b7db7b | 2669 | |
15a67dd8 RD |
2670 | err = sysfs_init(); |
2671 | if (err) | |
2672 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 2673 | __func__, err); |
00d26666 GKH |
2674 | fs_kobj = kobject_create_and_add("fs", NULL); |
2675 | if (!fs_kobj) | |
8e24eea7 | 2676 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
2677 | init_rootfs(); |
2678 | init_mount_tree(); | |
2679 | } | |
2680 | ||
616511d0 | 2681 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 2682 | { |
70fbcdf4 | 2683 | LIST_HEAD(umount_list); |
616511d0 | 2684 | |
d498b25a | 2685 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 2686 | return; |
390c6843 | 2687 | down_write(&namespace_sem); |
962830df | 2688 | br_write_lock(&vfsmount_lock); |
be08d6d2 | 2689 | umount_tree(ns->root, 0, &umount_list); |
962830df | 2690 | br_write_unlock(&vfsmount_lock); |
390c6843 | 2691 | up_write(&namespace_sem); |
70fbcdf4 | 2692 | release_mounts(&umount_list); |
771b1371 | 2693 | free_mnt_ns(ns); |
1da177e4 | 2694 | } |
9d412a43 AV |
2695 | |
2696 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
2697 | { | |
423e0ab0 TC |
2698 | struct vfsmount *mnt; |
2699 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
2700 | if (!IS_ERR(mnt)) { | |
2701 | /* | |
2702 | * it is a longterm mount, don't release mnt until | |
2703 | * we unmount before file sys is unregistered | |
2704 | */ | |
f7a99c5b | 2705 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
2706 | } |
2707 | return mnt; | |
9d412a43 AV |
2708 | } |
2709 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
2710 | |
2711 | void kern_unmount(struct vfsmount *mnt) | |
2712 | { | |
2713 | /* release long term mount so mount point can be released */ | |
2714 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b AV |
2715 | br_write_lock(&vfsmount_lock); |
2716 | real_mount(mnt)->mnt_ns = NULL; | |
2717 | br_write_unlock(&vfsmount_lock); | |
423e0ab0 TC |
2718 | mntput(mnt); |
2719 | } | |
2720 | } | |
2721 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
2722 | |
2723 | bool our_mnt(struct vfsmount *mnt) | |
2724 | { | |
143c8c91 | 2725 | return check_mnt(real_mount(mnt)); |
02125a82 | 2726 | } |
8823c079 EB |
2727 | |
2728 | static void *mntns_get(struct task_struct *task) | |
2729 | { | |
2730 | struct mnt_namespace *ns = NULL; | |
2731 | struct nsproxy *nsproxy; | |
2732 | ||
2733 | rcu_read_lock(); | |
2734 | nsproxy = task_nsproxy(task); | |
2735 | if (nsproxy) { | |
2736 | ns = nsproxy->mnt_ns; | |
2737 | get_mnt_ns(ns); | |
2738 | } | |
2739 | rcu_read_unlock(); | |
2740 | ||
2741 | return ns; | |
2742 | } | |
2743 | ||
2744 | static void mntns_put(void *ns) | |
2745 | { | |
2746 | put_mnt_ns(ns); | |
2747 | } | |
2748 | ||
2749 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
2750 | { | |
2751 | struct fs_struct *fs = current->fs; | |
2752 | struct mnt_namespace *mnt_ns = ns; | |
2753 | struct path root; | |
2754 | ||
2755 | if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_CHROOT)) | |
2756 | return -EINVAL; | |
2757 | ||
2758 | if (fs->users != 1) | |
2759 | return -EINVAL; | |
2760 | ||
2761 | get_mnt_ns(mnt_ns); | |
2762 | put_mnt_ns(nsproxy->mnt_ns); | |
2763 | nsproxy->mnt_ns = mnt_ns; | |
2764 | ||
2765 | /* Find the root */ | |
2766 | root.mnt = &mnt_ns->root->mnt; | |
2767 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
2768 | path_get(&root); | |
2769 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
2770 | ; | |
2771 | ||
2772 | /* Update the pwd and root */ | |
2773 | set_fs_pwd(fs, &root); | |
2774 | set_fs_root(fs, &root); | |
2775 | ||
2776 | path_put(&root); | |
2777 | return 0; | |
2778 | } | |
2779 | ||
2780 | const struct proc_ns_operations mntns_operations = { | |
2781 | .name = "mnt", | |
2782 | .type = CLONE_NEWNS, | |
2783 | .get = mntns_get, | |
2784 | .put = mntns_put, | |
2785 | .install = mntns_install, | |
2786 | }; |