]>
Commit | Line | Data |
---|---|---|
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 | ||
11 | #include <linux/syscalls.h> | |
12 | #include <linux/export.h> | |
13 | #include <linux/capability.h> | |
14 | #include <linux/mnt_namespace.h> | |
15 | #include <linux/user_namespace.h> | |
16 | #include <linux/namei.h> | |
17 | #include <linux/security.h> | |
18 | #include <linux/idr.h> | |
19 | #include <linux/init.h> /* init_rootfs */ | |
20 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ | |
21 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
22 | #include <linux/uaccess.h> | |
23 | #include <linux/proc_ns.h> | |
24 | #include <linux/magic.h> | |
25 | #include <linux/bootmem.h> | |
26 | #include <linux/task_work.h> | |
27 | #include "pnode.h" | |
28 | #include "internal.h" | |
29 | ||
30 | static unsigned int m_hash_mask __read_mostly; | |
31 | static unsigned int m_hash_shift __read_mostly; | |
32 | static unsigned int mp_hash_mask __read_mostly; | |
33 | static unsigned int mp_hash_shift __read_mostly; | |
34 | ||
35 | static __initdata unsigned long mhash_entries; | |
36 | static int __init set_mhash_entries(char *str) | |
37 | { | |
38 | if (!str) | |
39 | return 0; | |
40 | mhash_entries = simple_strtoul(str, &str, 0); | |
41 | return 1; | |
42 | } | |
43 | __setup("mhash_entries=", set_mhash_entries); | |
44 | ||
45 | static __initdata unsigned long mphash_entries; | |
46 | static int __init set_mphash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mphash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mphash_entries=", set_mphash_entries); | |
54 | ||
55 | static u64 event; | |
56 | static DEFINE_IDA(mnt_id_ida); | |
57 | static DEFINE_IDA(mnt_group_ida); | |
58 | static DEFINE_SPINLOCK(mnt_id_lock); | |
59 | static int mnt_id_start = 0; | |
60 | static int mnt_group_start = 1; | |
61 | ||
62 | static struct hlist_head *mount_hashtable __read_mostly; | |
63 | static struct hlist_head *mountpoint_hashtable __read_mostly; | |
64 | static struct kmem_cache *mnt_cache __read_mostly; | |
65 | static DECLARE_RWSEM(namespace_sem); | |
66 | ||
67 | /* /sys/fs */ | |
68 | struct kobject *fs_kobj; | |
69 | EXPORT_SYMBOL_GPL(fs_kobj); | |
70 | ||
71 | /* | |
72 | * vfsmount lock may be taken for read to prevent changes to the | |
73 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
74 | * up the tree. | |
75 | * | |
76 | * It should be taken for write in all cases where the vfsmount | |
77 | * tree or hash is modified or when a vfsmount structure is modified. | |
78 | */ | |
79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); | |
80 | ||
81 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) | |
82 | { | |
83 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); | |
84 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
85 | tmp = tmp + (tmp >> m_hash_shift); | |
86 | return &mount_hashtable[tmp & m_hash_mask]; | |
87 | } | |
88 | ||
89 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
90 | { | |
91 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
92 | tmp = tmp + (tmp >> mp_hash_shift); | |
93 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
94 | } | |
95 | ||
96 | /* | |
97 | * allocation is serialized by namespace_sem, but we need the spinlock to | |
98 | * serialize with freeing. | |
99 | */ | |
100 | static int mnt_alloc_id(struct mount *mnt) | |
101 | { | |
102 | int res; | |
103 | ||
104 | retry: | |
105 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
106 | spin_lock(&mnt_id_lock); | |
107 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); | |
108 | if (!res) | |
109 | mnt_id_start = mnt->mnt_id + 1; | |
110 | spin_unlock(&mnt_id_lock); | |
111 | if (res == -EAGAIN) | |
112 | goto retry; | |
113 | ||
114 | return res; | |
115 | } | |
116 | ||
117 | static void mnt_free_id(struct mount *mnt) | |
118 | { | |
119 | int id = mnt->mnt_id; | |
120 | spin_lock(&mnt_id_lock); | |
121 | ida_remove(&mnt_id_ida, id); | |
122 | if (mnt_id_start > id) | |
123 | mnt_id_start = id; | |
124 | spin_unlock(&mnt_id_lock); | |
125 | } | |
126 | ||
127 | /* | |
128 | * Allocate a new peer group ID | |
129 | * | |
130 | * mnt_group_ida is protected by namespace_sem | |
131 | */ | |
132 | static int mnt_alloc_group_id(struct mount *mnt) | |
133 | { | |
134 | int res; | |
135 | ||
136 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) | |
137 | return -ENOMEM; | |
138 | ||
139 | res = ida_get_new_above(&mnt_group_ida, | |
140 | mnt_group_start, | |
141 | &mnt->mnt_group_id); | |
142 | if (!res) | |
143 | mnt_group_start = mnt->mnt_group_id + 1; | |
144 | ||
145 | return res; | |
146 | } | |
147 | ||
148 | /* | |
149 | * Release a peer group ID | |
150 | */ | |
151 | void mnt_release_group_id(struct mount *mnt) | |
152 | { | |
153 | int id = mnt->mnt_group_id; | |
154 | ida_remove(&mnt_group_ida, id); | |
155 | if (mnt_group_start > id) | |
156 | mnt_group_start = id; | |
157 | mnt->mnt_group_id = 0; | |
158 | } | |
159 | ||
160 | /* | |
161 | * vfsmount lock must be held for read | |
162 | */ | |
163 | static inline void mnt_add_count(struct mount *mnt, int n) | |
164 | { | |
165 | #ifdef CONFIG_SMP | |
166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); | |
167 | #else | |
168 | preempt_disable(); | |
169 | mnt->mnt_count += n; | |
170 | preempt_enable(); | |
171 | #endif | |
172 | } | |
173 | ||
174 | /* | |
175 | * vfsmount lock must be held for write | |
176 | */ | |
177 | unsigned int mnt_get_count(struct mount *mnt) | |
178 | { | |
179 | #ifdef CONFIG_SMP | |
180 | unsigned int count = 0; | |
181 | int cpu; | |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; | |
185 | } | |
186 | ||
187 | return count; | |
188 | #else | |
189 | return mnt->mnt_count; | |
190 | #endif | |
191 | } | |
192 | ||
193 | static struct mount *alloc_vfsmnt(const char *name) | |
194 | { | |
195 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); | |
196 | if (mnt) { | |
197 | int err; | |
198 | ||
199 | err = mnt_alloc_id(mnt); | |
200 | if (err) | |
201 | goto out_free_cache; | |
202 | ||
203 | if (name) { | |
204 | mnt->mnt_devname = kstrdup(name, GFP_KERNEL); | |
205 | if (!mnt->mnt_devname) | |
206 | goto out_free_id; | |
207 | } | |
208 | ||
209 | #ifdef CONFIG_SMP | |
210 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); | |
211 | if (!mnt->mnt_pcp) | |
212 | goto out_free_devname; | |
213 | ||
214 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); | |
215 | #else | |
216 | mnt->mnt_count = 1; | |
217 | mnt->mnt_writers = 0; | |
218 | #endif | |
219 | ||
220 | INIT_HLIST_NODE(&mnt->mnt_hash); | |
221 | INIT_LIST_HEAD(&mnt->mnt_child); | |
222 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
223 | INIT_LIST_HEAD(&mnt->mnt_list); | |
224 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
225 | INIT_LIST_HEAD(&mnt->mnt_share); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
227 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
228 | INIT_HLIST_NODE(&mnt->mnt_mp_list); | |
229 | #ifdef CONFIG_FSNOTIFY | |
230 | INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); | |
231 | #endif | |
232 | } | |
233 | return mnt; | |
234 | ||
235 | #ifdef CONFIG_SMP | |
236 | out_free_devname: | |
237 | kfree(mnt->mnt_devname); | |
238 | #endif | |
239 | out_free_id: | |
240 | mnt_free_id(mnt); | |
241 | out_free_cache: | |
242 | kmem_cache_free(mnt_cache, mnt); | |
243 | return NULL; | |
244 | } | |
245 | ||
246 | /* | |
247 | * Most r/o checks on a fs are for operations that take | |
248 | * discrete amounts of time, like a write() or unlink(). | |
249 | * We must keep track of when those operations start | |
250 | * (for permission checks) and when they end, so that | |
251 | * we can determine when writes are able to occur to | |
252 | * a filesystem. | |
253 | */ | |
254 | /* | |
255 | * __mnt_is_readonly: check whether a mount is read-only | |
256 | * @mnt: the mount to check for its write status | |
257 | * | |
258 | * This shouldn't be used directly ouside of the VFS. | |
259 | * It does not guarantee that the filesystem will stay | |
260 | * r/w, just that it is right *now*. This can not and | |
261 | * should not be used in place of IS_RDONLY(inode). | |
262 | * mnt_want/drop_write() will _keep_ the filesystem | |
263 | * r/w. | |
264 | */ | |
265 | int __mnt_is_readonly(struct vfsmount *mnt) | |
266 | { | |
267 | if (mnt->mnt_flags & MNT_READONLY) | |
268 | return 1; | |
269 | if (mnt->mnt_sb->s_flags & MS_RDONLY) | |
270 | return 1; | |
271 | return 0; | |
272 | } | |
273 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
274 | ||
275 | static inline void mnt_inc_writers(struct mount *mnt) | |
276 | { | |
277 | #ifdef CONFIG_SMP | |
278 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); | |
279 | #else | |
280 | mnt->mnt_writers++; | |
281 | #endif | |
282 | } | |
283 | ||
284 | static inline void mnt_dec_writers(struct mount *mnt) | |
285 | { | |
286 | #ifdef CONFIG_SMP | |
287 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); | |
288 | #else | |
289 | mnt->mnt_writers--; | |
290 | #endif | |
291 | } | |
292 | ||
293 | static unsigned int mnt_get_writers(struct mount *mnt) | |
294 | { | |
295 | #ifdef CONFIG_SMP | |
296 | unsigned int count = 0; | |
297 | int cpu; | |
298 | ||
299 | for_each_possible_cpu(cpu) { | |
300 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; | |
301 | } | |
302 | ||
303 | return count; | |
304 | #else | |
305 | return mnt->mnt_writers; | |
306 | #endif | |
307 | } | |
308 | ||
309 | static int mnt_is_readonly(struct vfsmount *mnt) | |
310 | { | |
311 | if (mnt->mnt_sb->s_readonly_remount) | |
312 | return 1; | |
313 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
314 | smp_rmb(); | |
315 | return __mnt_is_readonly(mnt); | |
316 | } | |
317 | ||
318 | /* | |
319 | * Most r/o & frozen checks on a fs are for operations that take discrete | |
320 | * amounts of time, like a write() or unlink(). We must keep track of when | |
321 | * those operations start (for permission checks) and when they end, so that we | |
322 | * can determine when writes are able to occur to a filesystem. | |
323 | */ | |
324 | /** | |
325 | * __mnt_want_write - get write access to a mount without freeze protection | |
326 | * @m: the mount on which to take a write | |
327 | * | |
328 | * This tells the low-level filesystem that a write is about to be performed to | |
329 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
330 | * returning success. This operation does not protect against filesystem being | |
331 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
332 | * called. This is effectively a refcount. | |
333 | */ | |
334 | int __mnt_want_write(struct vfsmount *m) | |
335 | { | |
336 | struct mount *mnt = real_mount(m); | |
337 | int ret = 0; | |
338 | ||
339 | preempt_disable(); | |
340 | mnt_inc_writers(mnt); | |
341 | /* | |
342 | * The store to mnt_inc_writers must be visible before we pass | |
343 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our | |
344 | * incremented count after it has set MNT_WRITE_HOLD. | |
345 | */ | |
346 | smp_mb(); | |
347 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) | |
348 | cpu_relax(); | |
349 | /* | |
350 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
351 | * be set to match its requirements. So we must not load that until | |
352 | * MNT_WRITE_HOLD is cleared. | |
353 | */ | |
354 | smp_rmb(); | |
355 | if (mnt_is_readonly(m)) { | |
356 | mnt_dec_writers(mnt); | |
357 | ret = -EROFS; | |
358 | } | |
359 | preempt_enable(); | |
360 | ||
361 | return ret; | |
362 | } | |
363 | ||
364 | /** | |
365 | * mnt_want_write - get write access to a mount | |
366 | * @m: the mount on which to take a write | |
367 | * | |
368 | * This tells the low-level filesystem that a write is about to be performed to | |
369 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
370 | * is not frozen) before returning success. When the write operation is | |
371 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
372 | */ | |
373 | int mnt_want_write(struct vfsmount *m) | |
374 | { | |
375 | int ret; | |
376 | ||
377 | sb_start_write(m->mnt_sb); | |
378 | ret = __mnt_want_write(m); | |
379 | if (ret) | |
380 | sb_end_write(m->mnt_sb); | |
381 | return ret; | |
382 | } | |
383 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
384 | ||
385 | /** | |
386 | * mnt_clone_write - get write access to a mount | |
387 | * @mnt: the mount on which to take a write | |
388 | * | |
389 | * This is effectively like mnt_want_write, except | |
390 | * it must only be used to take an extra write reference | |
391 | * on a mountpoint that we already know has a write reference | |
392 | * on it. This allows some optimisation. | |
393 | * | |
394 | * After finished, mnt_drop_write must be called as usual to | |
395 | * drop the reference. | |
396 | */ | |
397 | int mnt_clone_write(struct vfsmount *mnt) | |
398 | { | |
399 | /* superblock may be r/o */ | |
400 | if (__mnt_is_readonly(mnt)) | |
401 | return -EROFS; | |
402 | preempt_disable(); | |
403 | mnt_inc_writers(real_mount(mnt)); | |
404 | preempt_enable(); | |
405 | return 0; | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
408 | ||
409 | /** | |
410 | * __mnt_want_write_file - get write access to a file's mount | |
411 | * @file: the file who's mount on which to take a write | |
412 | * | |
413 | * This is like __mnt_want_write, but it takes a file and can | |
414 | * do some optimisations if the file is open for write already | |
415 | */ | |
416 | int __mnt_want_write_file(struct file *file) | |
417 | { | |
418 | if (!(file->f_mode & FMODE_WRITER)) | |
419 | return __mnt_want_write(file->f_path.mnt); | |
420 | else | |
421 | return mnt_clone_write(file->f_path.mnt); | |
422 | } | |
423 | ||
424 | /** | |
425 | * mnt_want_write_file - get write access to a file's mount | |
426 | * @file: the file who's mount on which to take a write | |
427 | * | |
428 | * This is like mnt_want_write, but it takes a file and can | |
429 | * do some optimisations if the file is open for write already | |
430 | */ | |
431 | int mnt_want_write_file(struct file *file) | |
432 | { | |
433 | int ret; | |
434 | ||
435 | sb_start_write(file->f_path.mnt->mnt_sb); | |
436 | ret = __mnt_want_write_file(file); | |
437 | if (ret) | |
438 | sb_end_write(file->f_path.mnt->mnt_sb); | |
439 | return ret; | |
440 | } | |
441 | EXPORT_SYMBOL_GPL(mnt_want_write_file); | |
442 | ||
443 | /** | |
444 | * __mnt_drop_write - give up write access to a mount | |
445 | * @mnt: the mount on which to give up write access | |
446 | * | |
447 | * Tells the low-level filesystem that we are done | |
448 | * performing writes to it. Must be matched with | |
449 | * __mnt_want_write() call above. | |
450 | */ | |
451 | void __mnt_drop_write(struct vfsmount *mnt) | |
452 | { | |
453 | preempt_disable(); | |
454 | mnt_dec_writers(real_mount(mnt)); | |
455 | preempt_enable(); | |
456 | } | |
457 | ||
458 | /** | |
459 | * mnt_drop_write - give up write access to a mount | |
460 | * @mnt: the mount on which to give up write access | |
461 | * | |
462 | * Tells the low-level filesystem that we are done performing writes to it and | |
463 | * also allows filesystem to be frozen again. Must be matched with | |
464 | * mnt_want_write() call above. | |
465 | */ | |
466 | void mnt_drop_write(struct vfsmount *mnt) | |
467 | { | |
468 | __mnt_drop_write(mnt); | |
469 | sb_end_write(mnt->mnt_sb); | |
470 | } | |
471 | EXPORT_SYMBOL_GPL(mnt_drop_write); | |
472 | ||
473 | void __mnt_drop_write_file(struct file *file) | |
474 | { | |
475 | __mnt_drop_write(file->f_path.mnt); | |
476 | } | |
477 | ||
478 | void mnt_drop_write_file(struct file *file) | |
479 | { | |
480 | mnt_drop_write(file->f_path.mnt); | |
481 | } | |
482 | EXPORT_SYMBOL(mnt_drop_write_file); | |
483 | ||
484 | static int mnt_make_readonly(struct mount *mnt) | |
485 | { | |
486 | int ret = 0; | |
487 | ||
488 | lock_mount_hash(); | |
489 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
490 | /* | |
491 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store | |
492 | * should be visible before we do. | |
493 | */ | |
494 | smp_mb(); | |
495 | ||
496 | /* | |
497 | * With writers on hold, if this value is zero, then there are | |
498 | * definitely no active writers (although held writers may subsequently | |
499 | * increment the count, they'll have to wait, and decrement it after | |
500 | * seeing MNT_READONLY). | |
501 | * | |
502 | * It is OK to have counter incremented on one CPU and decremented on | |
503 | * another: the sum will add up correctly. The danger would be when we | |
504 | * sum up each counter, if we read a counter before it is incremented, | |
505 | * but then read another CPU's count which it has been subsequently | |
506 | * decremented from -- we would see more decrements than we should. | |
507 | * MNT_WRITE_HOLD protects against this scenario, because | |
508 | * mnt_want_write first increments count, then smp_mb, then spins on | |
509 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
510 | * we're counting up here. | |
511 | */ | |
512 | if (mnt_get_writers(mnt) > 0) | |
513 | ret = -EBUSY; | |
514 | else | |
515 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
516 | /* | |
517 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
518 | * that become unheld will see MNT_READONLY. | |
519 | */ | |
520 | smp_wmb(); | |
521 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
522 | unlock_mount_hash(); | |
523 | return ret; | |
524 | } | |
525 | ||
526 | static void __mnt_unmake_readonly(struct mount *mnt) | |
527 | { | |
528 | lock_mount_hash(); | |
529 | mnt->mnt.mnt_flags &= ~MNT_READONLY; | |
530 | unlock_mount_hash(); | |
531 | } | |
532 | ||
533 | int sb_prepare_remount_readonly(struct super_block *sb) | |
534 | { | |
535 | struct mount *mnt; | |
536 | int err = 0; | |
537 | ||
538 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ | |
539 | if (atomic_long_read(&sb->s_remove_count)) | |
540 | return -EBUSY; | |
541 | ||
542 | lock_mount_hash(); | |
543 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
544 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
545 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
546 | smp_mb(); | |
547 | if (mnt_get_writers(mnt) > 0) { | |
548 | err = -EBUSY; | |
549 | break; | |
550 | } | |
551 | } | |
552 | } | |
553 | if (!err && atomic_long_read(&sb->s_remove_count)) | |
554 | err = -EBUSY; | |
555 | ||
556 | if (!err) { | |
557 | sb->s_readonly_remount = 1; | |
558 | smp_wmb(); | |
559 | } | |
560 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
561 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
562 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
563 | } | |
564 | unlock_mount_hash(); | |
565 | ||
566 | return err; | |
567 | } | |
568 | ||
569 | static void free_vfsmnt(struct mount *mnt) | |
570 | { | |
571 | kfree(mnt->mnt_devname); | |
572 | #ifdef CONFIG_SMP | |
573 | free_percpu(mnt->mnt_pcp); | |
574 | #endif | |
575 | kmem_cache_free(mnt_cache, mnt); | |
576 | } | |
577 | ||
578 | static void delayed_free_vfsmnt(struct rcu_head *head) | |
579 | { | |
580 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
581 | } | |
582 | ||
583 | /* call under rcu_read_lock */ | |
584 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
585 | { | |
586 | struct mount *mnt; | |
587 | if (read_seqretry(&mount_lock, seq)) | |
588 | return false; | |
589 | if (bastard == NULL) | |
590 | return true; | |
591 | mnt = real_mount(bastard); | |
592 | mnt_add_count(mnt, 1); | |
593 | if (likely(!read_seqretry(&mount_lock, seq))) | |
594 | return true; | |
595 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { | |
596 | mnt_add_count(mnt, -1); | |
597 | return false; | |
598 | } | |
599 | rcu_read_unlock(); | |
600 | mntput(bastard); | |
601 | rcu_read_lock(); | |
602 | return false; | |
603 | } | |
604 | ||
605 | /* | |
606 | * find the first mount at @dentry on vfsmount @mnt. | |
607 | * call under rcu_read_lock() | |
608 | */ | |
609 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
610 | { | |
611 | struct hlist_head *head = m_hash(mnt, dentry); | |
612 | struct mount *p; | |
613 | ||
614 | hlist_for_each_entry_rcu(p, head, mnt_hash) | |
615 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) | |
616 | return p; | |
617 | return NULL; | |
618 | } | |
619 | ||
620 | /* | |
621 | * find the last mount at @dentry on vfsmount @mnt. | |
622 | * mount_lock must be held. | |
623 | */ | |
624 | struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) | |
625 | { | |
626 | struct mount *p, *res; | |
627 | res = p = __lookup_mnt(mnt, dentry); | |
628 | if (!p) | |
629 | goto out; | |
630 | hlist_for_each_entry_continue(p, mnt_hash) { | |
631 | if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) | |
632 | break; | |
633 | res = p; | |
634 | } | |
635 | out: | |
636 | return res; | |
637 | } | |
638 | ||
639 | /* | |
640 | * lookup_mnt - Return the first child mount mounted at path | |
641 | * | |
642 | * "First" means first mounted chronologically. If you create the | |
643 | * following mounts: | |
644 | * | |
645 | * mount /dev/sda1 /mnt | |
646 | * mount /dev/sda2 /mnt | |
647 | * mount /dev/sda3 /mnt | |
648 | * | |
649 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
650 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
651 | * /dev/sda2, then /dev/sda3, then NULL. | |
652 | * | |
653 | * lookup_mnt takes a reference to the found vfsmount. | |
654 | */ | |
655 | struct vfsmount *lookup_mnt(struct path *path) | |
656 | { | |
657 | struct mount *child_mnt; | |
658 | struct vfsmount *m; | |
659 | unsigned seq; | |
660 | ||
661 | rcu_read_lock(); | |
662 | do { | |
663 | seq = read_seqbegin(&mount_lock); | |
664 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
665 | m = child_mnt ? &child_mnt->mnt : NULL; | |
666 | } while (!legitimize_mnt(m, seq)); | |
667 | rcu_read_unlock(); | |
668 | return m; | |
669 | } | |
670 | ||
671 | /* | |
672 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
673 | * current mount namespace. | |
674 | * | |
675 | * The common case is dentries are not mountpoints at all and that | |
676 | * test is handled inline. For the slow case when we are actually | |
677 | * dealing with a mountpoint of some kind, walk through all of the | |
678 | * mounts in the current mount namespace and test to see if the dentry | |
679 | * is a mountpoint. | |
680 | * | |
681 | * The mount_hashtable is not usable in the context because we | |
682 | * need to identify all mounts that may be in the current mount | |
683 | * namespace not just a mount that happens to have some specified | |
684 | * parent mount. | |
685 | */ | |
686 | bool __is_local_mountpoint(struct dentry *dentry) | |
687 | { | |
688 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
689 | struct mount *mnt; | |
690 | bool is_covered = false; | |
691 | ||
692 | if (!d_mountpoint(dentry)) | |
693 | goto out; | |
694 | ||
695 | down_read(&namespace_sem); | |
696 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
697 | is_covered = (mnt->mnt_mountpoint == dentry); | |
698 | if (is_covered) | |
699 | break; | |
700 | } | |
701 | up_read(&namespace_sem); | |
702 | out: | |
703 | return is_covered; | |
704 | } | |
705 | ||
706 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) | |
707 | { | |
708 | struct hlist_head *chain = mp_hash(dentry); | |
709 | struct mountpoint *mp; | |
710 | ||
711 | hlist_for_each_entry(mp, chain, m_hash) { | |
712 | if (mp->m_dentry == dentry) { | |
713 | /* might be worth a WARN_ON() */ | |
714 | if (d_unlinked(dentry)) | |
715 | return ERR_PTR(-ENOENT); | |
716 | mp->m_count++; | |
717 | return mp; | |
718 | } | |
719 | } | |
720 | return NULL; | |
721 | } | |
722 | ||
723 | static struct mountpoint *new_mountpoint(struct dentry *dentry) | |
724 | { | |
725 | struct hlist_head *chain = mp_hash(dentry); | |
726 | struct mountpoint *mp; | |
727 | int ret; | |
728 | ||
729 | mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
730 | if (!mp) | |
731 | return ERR_PTR(-ENOMEM); | |
732 | ||
733 | ret = d_set_mounted(dentry); | |
734 | if (ret) { | |
735 | kfree(mp); | |
736 | return ERR_PTR(ret); | |
737 | } | |
738 | ||
739 | mp->m_dentry = dentry; | |
740 | mp->m_count = 1; | |
741 | hlist_add_head(&mp->m_hash, chain); | |
742 | INIT_HLIST_HEAD(&mp->m_list); | |
743 | return mp; | |
744 | } | |
745 | ||
746 | static void put_mountpoint(struct mountpoint *mp) | |
747 | { | |
748 | if (!--mp->m_count) { | |
749 | struct dentry *dentry = mp->m_dentry; | |
750 | BUG_ON(!hlist_empty(&mp->m_list)); | |
751 | spin_lock(&dentry->d_lock); | |
752 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
753 | spin_unlock(&dentry->d_lock); | |
754 | hlist_del(&mp->m_hash); | |
755 | kfree(mp); | |
756 | } | |
757 | } | |
758 | ||
759 | static inline int check_mnt(struct mount *mnt) | |
760 | { | |
761 | return mnt->mnt_ns == current->nsproxy->mnt_ns; | |
762 | } | |
763 | ||
764 | /* | |
765 | * vfsmount lock must be held for write | |
766 | */ | |
767 | static void touch_mnt_namespace(struct mnt_namespace *ns) | |
768 | { | |
769 | if (ns) { | |
770 | ns->event = ++event; | |
771 | wake_up_interruptible(&ns->poll); | |
772 | } | |
773 | } | |
774 | ||
775 | /* | |
776 | * vfsmount lock must be held for write | |
777 | */ | |
778 | static void __touch_mnt_namespace(struct mnt_namespace *ns) | |
779 | { | |
780 | if (ns && ns->event != event) { | |
781 | ns->event = event; | |
782 | wake_up_interruptible(&ns->poll); | |
783 | } | |
784 | } | |
785 | ||
786 | /* | |
787 | * vfsmount lock must be held for write | |
788 | */ | |
789 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
790 | { | |
791 | old_path->dentry = mnt->mnt_mountpoint; | |
792 | old_path->mnt = &mnt->mnt_parent->mnt; | |
793 | mnt->mnt_parent = mnt; | |
794 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
795 | list_del_init(&mnt->mnt_child); | |
796 | hlist_del_init_rcu(&mnt->mnt_hash); | |
797 | hlist_del_init(&mnt->mnt_mp_list); | |
798 | put_mountpoint(mnt->mnt_mp); | |
799 | mnt->mnt_mp = NULL; | |
800 | } | |
801 | ||
802 | /* | |
803 | * vfsmount lock must be held for write | |
804 | */ | |
805 | void mnt_set_mountpoint(struct mount *mnt, | |
806 | struct mountpoint *mp, | |
807 | struct mount *child_mnt) | |
808 | { | |
809 | mp->m_count++; | |
810 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ | |
811 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); | |
812 | child_mnt->mnt_parent = mnt; | |
813 | child_mnt->mnt_mp = mp; | |
814 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); | |
815 | } | |
816 | ||
817 | /* | |
818 | * vfsmount lock must be held for write | |
819 | */ | |
820 | static void attach_mnt(struct mount *mnt, | |
821 | struct mount *parent, | |
822 | struct mountpoint *mp) | |
823 | { | |
824 | mnt_set_mountpoint(parent, mp, mnt); | |
825 | hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); | |
826 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
827 | } | |
828 | ||
829 | static void attach_shadowed(struct mount *mnt, | |
830 | struct mount *parent, | |
831 | struct mount *shadows) | |
832 | { | |
833 | if (shadows) { | |
834 | hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); | |
835 | list_add(&mnt->mnt_child, &shadows->mnt_child); | |
836 | } else { | |
837 | hlist_add_head_rcu(&mnt->mnt_hash, | |
838 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
839 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
840 | } | |
841 | } | |
842 | ||
843 | /* | |
844 | * vfsmount lock must be held for write | |
845 | */ | |
846 | static void commit_tree(struct mount *mnt, struct mount *shadows) | |
847 | { | |
848 | struct mount *parent = mnt->mnt_parent; | |
849 | struct mount *m; | |
850 | LIST_HEAD(head); | |
851 | struct mnt_namespace *n = parent->mnt_ns; | |
852 | ||
853 | BUG_ON(parent == mnt); | |
854 | ||
855 | list_add_tail(&head, &mnt->mnt_list); | |
856 | list_for_each_entry(m, &head, mnt_list) | |
857 | m->mnt_ns = n; | |
858 | ||
859 | list_splice(&head, n->list.prev); | |
860 | ||
861 | attach_shadowed(mnt, parent, shadows); | |
862 | touch_mnt_namespace(n); | |
863 | } | |
864 | ||
865 | static struct mount *next_mnt(struct mount *p, struct mount *root) | |
866 | { | |
867 | struct list_head *next = p->mnt_mounts.next; | |
868 | if (next == &p->mnt_mounts) { | |
869 | while (1) { | |
870 | if (p == root) | |
871 | return NULL; | |
872 | next = p->mnt_child.next; | |
873 | if (next != &p->mnt_parent->mnt_mounts) | |
874 | break; | |
875 | p = p->mnt_parent; | |
876 | } | |
877 | } | |
878 | return list_entry(next, struct mount, mnt_child); | |
879 | } | |
880 | ||
881 | static struct mount *skip_mnt_tree(struct mount *p) | |
882 | { | |
883 | struct list_head *prev = p->mnt_mounts.prev; | |
884 | while (prev != &p->mnt_mounts) { | |
885 | p = list_entry(prev, struct mount, mnt_child); | |
886 | prev = p->mnt_mounts.prev; | |
887 | } | |
888 | return p; | |
889 | } | |
890 | ||
891 | struct vfsmount * | |
892 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
893 | { | |
894 | struct mount *mnt; | |
895 | struct dentry *root; | |
896 | ||
897 | if (!type) | |
898 | return ERR_PTR(-ENODEV); | |
899 | ||
900 | mnt = alloc_vfsmnt(name); | |
901 | if (!mnt) | |
902 | return ERR_PTR(-ENOMEM); | |
903 | ||
904 | if (flags & MS_KERNMOUNT) | |
905 | mnt->mnt.mnt_flags = MNT_INTERNAL; | |
906 | ||
907 | root = mount_fs(type, flags, name, data); | |
908 | if (IS_ERR(root)) { | |
909 | mnt_free_id(mnt); | |
910 | free_vfsmnt(mnt); | |
911 | return ERR_CAST(root); | |
912 | } | |
913 | ||
914 | mnt->mnt.mnt_root = root; | |
915 | mnt->mnt.mnt_sb = root->d_sb; | |
916 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
917 | mnt->mnt_parent = mnt; | |
918 | lock_mount_hash(); | |
919 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); | |
920 | unlock_mount_hash(); | |
921 | return &mnt->mnt; | |
922 | } | |
923 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
924 | ||
925 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, | |
926 | int flag) | |
927 | { | |
928 | struct super_block *sb = old->mnt.mnt_sb; | |
929 | struct mount *mnt; | |
930 | int err; | |
931 | ||
932 | mnt = alloc_vfsmnt(old->mnt_devname); | |
933 | if (!mnt) | |
934 | return ERR_PTR(-ENOMEM); | |
935 | ||
936 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) | |
937 | mnt->mnt_group_id = 0; /* not a peer of original */ | |
938 | else | |
939 | mnt->mnt_group_id = old->mnt_group_id; | |
940 | ||
941 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { | |
942 | err = mnt_alloc_group_id(mnt); | |
943 | if (err) | |
944 | goto out_free; | |
945 | } | |
946 | ||
947 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); | |
948 | /* Don't allow unprivileged users to change mount flags */ | |
949 | if (flag & CL_UNPRIVILEGED) { | |
950 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
951 | ||
952 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
953 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
954 | ||
955 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
956 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
957 | ||
958 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
959 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
960 | ||
961 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
962 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
963 | } | |
964 | ||
965 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
966 | if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) | |
967 | mnt->mnt.mnt_flags |= MNT_LOCKED; | |
968 | ||
969 | atomic_inc(&sb->s_active); | |
970 | mnt->mnt.mnt_sb = sb; | |
971 | mnt->mnt.mnt_root = dget(root); | |
972 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
973 | mnt->mnt_parent = mnt; | |
974 | lock_mount_hash(); | |
975 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); | |
976 | unlock_mount_hash(); | |
977 | ||
978 | if ((flag & CL_SLAVE) || | |
979 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
980 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
981 | mnt->mnt_master = old; | |
982 | CLEAR_MNT_SHARED(mnt); | |
983 | } else if (!(flag & CL_PRIVATE)) { | |
984 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
985 | list_add(&mnt->mnt_share, &old->mnt_share); | |
986 | if (IS_MNT_SLAVE(old)) | |
987 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
988 | mnt->mnt_master = old->mnt_master; | |
989 | } | |
990 | if (flag & CL_MAKE_SHARED) | |
991 | set_mnt_shared(mnt); | |
992 | ||
993 | /* stick the duplicate mount on the same expiry list | |
994 | * as the original if that was on one */ | |
995 | if (flag & CL_EXPIRE) { | |
996 | if (!list_empty(&old->mnt_expire)) | |
997 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
998 | } | |
999 | ||
1000 | return mnt; | |
1001 | ||
1002 | out_free: | |
1003 | mnt_free_id(mnt); | |
1004 | free_vfsmnt(mnt); | |
1005 | return ERR_PTR(err); | |
1006 | } | |
1007 | ||
1008 | static void cleanup_mnt(struct mount *mnt) | |
1009 | { | |
1010 | /* | |
1011 | * This probably indicates that somebody messed | |
1012 | * up a mnt_want/drop_write() pair. If this | |
1013 | * happens, the filesystem was probably unable | |
1014 | * to make r/w->r/o transitions. | |
1015 | */ | |
1016 | /* | |
1017 | * The locking used to deal with mnt_count decrement provides barriers, | |
1018 | * so mnt_get_writers() below is safe. | |
1019 | */ | |
1020 | WARN_ON(mnt_get_writers(mnt)); | |
1021 | if (unlikely(mnt->mnt_pins.first)) | |
1022 | mnt_pin_kill(mnt); | |
1023 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1024 | dput(mnt->mnt.mnt_root); | |
1025 | deactivate_super(mnt->mnt.mnt_sb); | |
1026 | mnt_free_id(mnt); | |
1027 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1028 | } | |
1029 | ||
1030 | static void __cleanup_mnt(struct rcu_head *head) | |
1031 | { | |
1032 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1033 | } | |
1034 | ||
1035 | static LLIST_HEAD(delayed_mntput_list); | |
1036 | static void delayed_mntput(struct work_struct *unused) | |
1037 | { | |
1038 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
1039 | struct llist_node *next; | |
1040 | ||
1041 | for (; node; node = next) { | |
1042 | next = llist_next(node); | |
1043 | cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); | |
1044 | } | |
1045 | } | |
1046 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1047 | ||
1048 | static void mntput_no_expire(struct mount *mnt) | |
1049 | { | |
1050 | rcu_read_lock(); | |
1051 | mnt_add_count(mnt, -1); | |
1052 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1053 | rcu_read_unlock(); | |
1054 | return; | |
1055 | } | |
1056 | lock_mount_hash(); | |
1057 | if (mnt_get_count(mnt)) { | |
1058 | rcu_read_unlock(); | |
1059 | unlock_mount_hash(); | |
1060 | return; | |
1061 | } | |
1062 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { | |
1063 | rcu_read_unlock(); | |
1064 | unlock_mount_hash(); | |
1065 | return; | |
1066 | } | |
1067 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1068 | rcu_read_unlock(); | |
1069 | ||
1070 | list_del(&mnt->mnt_instance); | |
1071 | unlock_mount_hash(); | |
1072 | ||
1073 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { | |
1074 | struct task_struct *task = current; | |
1075 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1076 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1077 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1078 | return; | |
1079 | } | |
1080 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1081 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1082 | return; | |
1083 | } | |
1084 | cleanup_mnt(mnt); | |
1085 | } | |
1086 | ||
1087 | void mntput(struct vfsmount *mnt) | |
1088 | { | |
1089 | if (mnt) { | |
1090 | struct mount *m = real_mount(mnt); | |
1091 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ | |
1092 | if (unlikely(m->mnt_expiry_mark)) | |
1093 | m->mnt_expiry_mark = 0; | |
1094 | mntput_no_expire(m); | |
1095 | } | |
1096 | } | |
1097 | EXPORT_SYMBOL(mntput); | |
1098 | ||
1099 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1100 | { | |
1101 | if (mnt) | |
1102 | mnt_add_count(real_mount(mnt), 1); | |
1103 | return mnt; | |
1104 | } | |
1105 | EXPORT_SYMBOL(mntget); | |
1106 | ||
1107 | struct vfsmount *mnt_clone_internal(struct path *path) | |
1108 | { | |
1109 | struct mount *p; | |
1110 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1111 | if (IS_ERR(p)) | |
1112 | return ERR_CAST(p); | |
1113 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1114 | return &p->mnt; | |
1115 | } | |
1116 | ||
1117 | static inline void mangle(struct seq_file *m, const char *s) | |
1118 | { | |
1119 | seq_escape(m, s, " \t\n\\"); | |
1120 | } | |
1121 | ||
1122 | /* | |
1123 | * Simple .show_options callback for filesystems which don't want to | |
1124 | * implement more complex mount option showing. | |
1125 | * | |
1126 | * See also save_mount_options(). | |
1127 | */ | |
1128 | int generic_show_options(struct seq_file *m, struct dentry *root) | |
1129 | { | |
1130 | const char *options; | |
1131 | ||
1132 | rcu_read_lock(); | |
1133 | options = rcu_dereference(root->d_sb->s_options); | |
1134 | ||
1135 | if (options != NULL && options[0]) { | |
1136 | seq_putc(m, ','); | |
1137 | mangle(m, options); | |
1138 | } | |
1139 | rcu_read_unlock(); | |
1140 | ||
1141 | return 0; | |
1142 | } | |
1143 | EXPORT_SYMBOL(generic_show_options); | |
1144 | ||
1145 | /* | |
1146 | * If filesystem uses generic_show_options(), this function should be | |
1147 | * called from the fill_super() callback. | |
1148 | * | |
1149 | * The .remount_fs callback usually needs to be handled in a special | |
1150 | * way, to make sure, that previous options are not overwritten if the | |
1151 | * remount fails. | |
1152 | * | |
1153 | * Also note, that if the filesystem's .remount_fs function doesn't | |
1154 | * reset all options to their default value, but changes only newly | |
1155 | * given options, then the displayed options will not reflect reality | |
1156 | * any more. | |
1157 | */ | |
1158 | void save_mount_options(struct super_block *sb, char *options) | |
1159 | { | |
1160 | BUG_ON(sb->s_options); | |
1161 | rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); | |
1162 | } | |
1163 | EXPORT_SYMBOL(save_mount_options); | |
1164 | ||
1165 | void replace_mount_options(struct super_block *sb, char *options) | |
1166 | { | |
1167 | char *old = sb->s_options; | |
1168 | rcu_assign_pointer(sb->s_options, options); | |
1169 | if (old) { | |
1170 | synchronize_rcu(); | |
1171 | kfree(old); | |
1172 | } | |
1173 | } | |
1174 | EXPORT_SYMBOL(replace_mount_options); | |
1175 | ||
1176 | #ifdef CONFIG_PROC_FS | |
1177 | /* iterator; we want it to have access to namespace_sem, thus here... */ | |
1178 | static void *m_start(struct seq_file *m, loff_t *pos) | |
1179 | { | |
1180 | struct proc_mounts *p = proc_mounts(m); | |
1181 | ||
1182 | down_read(&namespace_sem); | |
1183 | if (p->cached_event == p->ns->event) { | |
1184 | void *v = p->cached_mount; | |
1185 | if (*pos == p->cached_index) | |
1186 | return v; | |
1187 | if (*pos == p->cached_index + 1) { | |
1188 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1189 | return p->cached_mount = v; | |
1190 | } | |
1191 | } | |
1192 | ||
1193 | p->cached_event = p->ns->event; | |
1194 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1195 | p->cached_index = *pos; | |
1196 | return p->cached_mount; | |
1197 | } | |
1198 | ||
1199 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1200 | { | |
1201 | struct proc_mounts *p = proc_mounts(m); | |
1202 | ||
1203 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); | |
1204 | p->cached_index = *pos; | |
1205 | return p->cached_mount; | |
1206 | } | |
1207 | ||
1208 | static void m_stop(struct seq_file *m, void *v) | |
1209 | { | |
1210 | up_read(&namespace_sem); | |
1211 | } | |
1212 | ||
1213 | static int m_show(struct seq_file *m, void *v) | |
1214 | { | |
1215 | struct proc_mounts *p = proc_mounts(m); | |
1216 | struct mount *r = list_entry(v, struct mount, mnt_list); | |
1217 | return p->show(m, &r->mnt); | |
1218 | } | |
1219 | ||
1220 | const struct seq_operations mounts_op = { | |
1221 | .start = m_start, | |
1222 | .next = m_next, | |
1223 | .stop = m_stop, | |
1224 | .show = m_show, | |
1225 | }; | |
1226 | #endif /* CONFIG_PROC_FS */ | |
1227 | ||
1228 | /** | |
1229 | * may_umount_tree - check if a mount tree is busy | |
1230 | * @mnt: root of mount tree | |
1231 | * | |
1232 | * This is called to check if a tree of mounts has any | |
1233 | * open files, pwds, chroots or sub mounts that are | |
1234 | * busy. | |
1235 | */ | |
1236 | int may_umount_tree(struct vfsmount *m) | |
1237 | { | |
1238 | struct mount *mnt = real_mount(m); | |
1239 | int actual_refs = 0; | |
1240 | int minimum_refs = 0; | |
1241 | struct mount *p; | |
1242 | BUG_ON(!m); | |
1243 | ||
1244 | /* write lock needed for mnt_get_count */ | |
1245 | lock_mount_hash(); | |
1246 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1247 | actual_refs += mnt_get_count(p); | |
1248 | minimum_refs += 2; | |
1249 | } | |
1250 | unlock_mount_hash(); | |
1251 | ||
1252 | if (actual_refs > minimum_refs) | |
1253 | return 0; | |
1254 | ||
1255 | return 1; | |
1256 | } | |
1257 | ||
1258 | EXPORT_SYMBOL(may_umount_tree); | |
1259 | ||
1260 | /** | |
1261 | * may_umount - check if a mount point is busy | |
1262 | * @mnt: root of mount | |
1263 | * | |
1264 | * This is called to check if a mount point has any | |
1265 | * open files, pwds, chroots or sub mounts. If the | |
1266 | * mount has sub mounts this will return busy | |
1267 | * regardless of whether the sub mounts are busy. | |
1268 | * | |
1269 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1270 | * give false negatives. The main reason why it's here is that we need | |
1271 | * a non-destructive way to look for easily umountable filesystems. | |
1272 | */ | |
1273 | int may_umount(struct vfsmount *mnt) | |
1274 | { | |
1275 | int ret = 1; | |
1276 | down_read(&namespace_sem); | |
1277 | lock_mount_hash(); | |
1278 | if (propagate_mount_busy(real_mount(mnt), 2)) | |
1279 | ret = 0; | |
1280 | unlock_mount_hash(); | |
1281 | up_read(&namespace_sem); | |
1282 | return ret; | |
1283 | } | |
1284 | ||
1285 | EXPORT_SYMBOL(may_umount); | |
1286 | ||
1287 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ | |
1288 | ||
1289 | static void namespace_unlock(void) | |
1290 | { | |
1291 | struct mount *mnt; | |
1292 | struct hlist_head head = unmounted; | |
1293 | ||
1294 | if (likely(hlist_empty(&head))) { | |
1295 | up_write(&namespace_sem); | |
1296 | return; | |
1297 | } | |
1298 | ||
1299 | head.first->pprev = &head.first; | |
1300 | INIT_HLIST_HEAD(&unmounted); | |
1301 | ||
1302 | /* undo decrements we'd done in umount_tree() */ | |
1303 | hlist_for_each_entry(mnt, &head, mnt_hash) | |
1304 | if (mnt->mnt_ex_mountpoint.mnt) | |
1305 | mntget(mnt->mnt_ex_mountpoint.mnt); | |
1306 | ||
1307 | up_write(&namespace_sem); | |
1308 | ||
1309 | synchronize_rcu(); | |
1310 | ||
1311 | while (!hlist_empty(&head)) { | |
1312 | mnt = hlist_entry(head.first, struct mount, mnt_hash); | |
1313 | hlist_del_init(&mnt->mnt_hash); | |
1314 | if (mnt->mnt_ex_mountpoint.mnt) | |
1315 | path_put(&mnt->mnt_ex_mountpoint); | |
1316 | mntput(&mnt->mnt); | |
1317 | } | |
1318 | } | |
1319 | ||
1320 | static inline void namespace_lock(void) | |
1321 | { | |
1322 | down_write(&namespace_sem); | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * mount_lock must be held | |
1327 | * namespace_sem must be held for write | |
1328 | * how = 0 => just this tree, don't propagate | |
1329 | * how = 1 => propagate; we know that nobody else has reference to any victims | |
1330 | * how = 2 => lazy umount | |
1331 | */ | |
1332 | void umount_tree(struct mount *mnt, int how) | |
1333 | { | |
1334 | HLIST_HEAD(tmp_list); | |
1335 | struct mount *p; | |
1336 | struct mount *last = NULL; | |
1337 | ||
1338 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1339 | hlist_del_init_rcu(&p->mnt_hash); | |
1340 | hlist_add_head(&p->mnt_hash, &tmp_list); | |
1341 | } | |
1342 | ||
1343 | hlist_for_each_entry(p, &tmp_list, mnt_hash) | |
1344 | list_del_init(&p->mnt_child); | |
1345 | ||
1346 | if (how) | |
1347 | propagate_umount(&tmp_list); | |
1348 | ||
1349 | hlist_for_each_entry(p, &tmp_list, mnt_hash) { | |
1350 | list_del_init(&p->mnt_expire); | |
1351 | list_del_init(&p->mnt_list); | |
1352 | __touch_mnt_namespace(p->mnt_ns); | |
1353 | p->mnt_ns = NULL; | |
1354 | if (how < 2) | |
1355 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; | |
1356 | if (mnt_has_parent(p)) { | |
1357 | hlist_del_init(&p->mnt_mp_list); | |
1358 | put_mountpoint(p->mnt_mp); | |
1359 | mnt_add_count(p->mnt_parent, -1); | |
1360 | /* move the reference to mountpoint into ->mnt_ex_mountpoint */ | |
1361 | p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; | |
1362 | p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; | |
1363 | p->mnt_mountpoint = p->mnt.mnt_root; | |
1364 | p->mnt_parent = p; | |
1365 | p->mnt_mp = NULL; | |
1366 | } | |
1367 | change_mnt_propagation(p, MS_PRIVATE); | |
1368 | last = p; | |
1369 | } | |
1370 | if (last) { | |
1371 | last->mnt_hash.next = unmounted.first; | |
1372 | unmounted.first = tmp_list.first; | |
1373 | unmounted.first->pprev = &unmounted.first; | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | static void shrink_submounts(struct mount *mnt); | |
1378 | ||
1379 | static int do_umount(struct mount *mnt, int flags) | |
1380 | { | |
1381 | struct super_block *sb = mnt->mnt.mnt_sb; | |
1382 | int retval; | |
1383 | ||
1384 | retval = security_sb_umount(&mnt->mnt, flags); | |
1385 | if (retval) | |
1386 | return retval; | |
1387 | ||
1388 | /* | |
1389 | * Allow userspace to request a mountpoint be expired rather than | |
1390 | * unmounting unconditionally. Unmount only happens if: | |
1391 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1392 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1393 | */ | |
1394 | if (flags & MNT_EXPIRE) { | |
1395 | if (&mnt->mnt == current->fs->root.mnt || | |
1396 | flags & (MNT_FORCE | MNT_DETACH)) | |
1397 | return -EINVAL; | |
1398 | ||
1399 | /* | |
1400 | * probably don't strictly need the lock here if we examined | |
1401 | * all race cases, but it's a slowpath. | |
1402 | */ | |
1403 | lock_mount_hash(); | |
1404 | if (mnt_get_count(mnt) != 2) { | |
1405 | unlock_mount_hash(); | |
1406 | return -EBUSY; | |
1407 | } | |
1408 | unlock_mount_hash(); | |
1409 | ||
1410 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
1411 | return -EAGAIN; | |
1412 | } | |
1413 | ||
1414 | /* | |
1415 | * If we may have to abort operations to get out of this | |
1416 | * mount, and they will themselves hold resources we must | |
1417 | * allow the fs to do things. In the Unix tradition of | |
1418 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1419 | * might fail to complete on the first run through as other tasks | |
1420 | * must return, and the like. Thats for the mount program to worry | |
1421 | * about for the moment. | |
1422 | */ | |
1423 | ||
1424 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { | |
1425 | sb->s_op->umount_begin(sb); | |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * No sense to grab the lock for this test, but test itself looks | |
1430 | * somewhat bogus. Suggestions for better replacement? | |
1431 | * Ho-hum... In principle, we might treat that as umount + switch | |
1432 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1433 | * Actually it makes sense, especially if rootfs would contain a | |
1434 | * /reboot - static binary that would close all descriptors and | |
1435 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1436 | */ | |
1437 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { | |
1438 | /* | |
1439 | * Special case for "unmounting" root ... | |
1440 | * we just try to remount it readonly. | |
1441 | */ | |
1442 | if (!capable(CAP_SYS_ADMIN)) | |
1443 | return -EPERM; | |
1444 | down_write(&sb->s_umount); | |
1445 | if (!(sb->s_flags & MS_RDONLY)) | |
1446 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
1447 | up_write(&sb->s_umount); | |
1448 | return retval; | |
1449 | } | |
1450 | ||
1451 | namespace_lock(); | |
1452 | lock_mount_hash(); | |
1453 | event++; | |
1454 | ||
1455 | if (flags & MNT_DETACH) { | |
1456 | if (!list_empty(&mnt->mnt_list)) | |
1457 | umount_tree(mnt, 2); | |
1458 | retval = 0; | |
1459 | } else { | |
1460 | shrink_submounts(mnt); | |
1461 | retval = -EBUSY; | |
1462 | if (!propagate_mount_busy(mnt, 2)) { | |
1463 | if (!list_empty(&mnt->mnt_list)) | |
1464 | umount_tree(mnt, 1); | |
1465 | retval = 0; | |
1466 | } | |
1467 | } | |
1468 | unlock_mount_hash(); | |
1469 | namespace_unlock(); | |
1470 | return retval; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1475 | * | |
1476 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1477 | * to an existing mountpoint, and wind up leaking the mount. | |
1478 | * detach_mounts allows lazily unmounting those mounts instead of | |
1479 | * leaking them. | |
1480 | * | |
1481 | * The caller may hold dentry->d_inode->i_mutex. | |
1482 | */ | |
1483 | void __detach_mounts(struct dentry *dentry) | |
1484 | { | |
1485 | struct mountpoint *mp; | |
1486 | struct mount *mnt; | |
1487 | ||
1488 | namespace_lock(); | |
1489 | mp = lookup_mountpoint(dentry); | |
1490 | if (!mp) | |
1491 | goto out_unlock; | |
1492 | ||
1493 | lock_mount_hash(); | |
1494 | while (!hlist_empty(&mp->m_list)) { | |
1495 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
1496 | umount_tree(mnt, 2); | |
1497 | } | |
1498 | unlock_mount_hash(); | |
1499 | put_mountpoint(mp); | |
1500 | out_unlock: | |
1501 | namespace_unlock(); | |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Is the caller allowed to modify his namespace? | |
1506 | */ | |
1507 | static inline bool may_mount(void) | |
1508 | { | |
1509 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Now umount can handle mount points as well as block devices. | |
1514 | * This is important for filesystems which use unnamed block devices. | |
1515 | * | |
1516 | * We now support a flag for forced unmount like the other 'big iron' | |
1517 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1518 | */ | |
1519 | ||
1520 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) | |
1521 | { | |
1522 | struct path path; | |
1523 | struct mount *mnt; | |
1524 | int retval; | |
1525 | int lookup_flags = 0; | |
1526 | ||
1527 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1528 | return -EINVAL; | |
1529 | ||
1530 | if (!may_mount()) | |
1531 | return -EPERM; | |
1532 | ||
1533 | if (!(flags & UMOUNT_NOFOLLOW)) | |
1534 | lookup_flags |= LOOKUP_FOLLOW; | |
1535 | ||
1536 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); | |
1537 | if (retval) | |
1538 | goto out; | |
1539 | mnt = real_mount(path.mnt); | |
1540 | retval = -EINVAL; | |
1541 | if (path.dentry != path.mnt->mnt_root) | |
1542 | goto dput_and_out; | |
1543 | if (!check_mnt(mnt)) | |
1544 | goto dput_and_out; | |
1545 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1546 | goto dput_and_out; | |
1547 | retval = -EPERM; | |
1548 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1549 | goto dput_and_out; | |
1550 | ||
1551 | retval = do_umount(mnt, flags); | |
1552 | dput_and_out: | |
1553 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ | |
1554 | dput(path.dentry); | |
1555 | mntput_no_expire(mnt); | |
1556 | out: | |
1557 | return retval; | |
1558 | } | |
1559 | ||
1560 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1561 | ||
1562 | /* | |
1563 | * The 2.0 compatible umount. No flags. | |
1564 | */ | |
1565 | SYSCALL_DEFINE1(oldumount, char __user *, name) | |
1566 | { | |
1567 | return sys_umount(name, 0); | |
1568 | } | |
1569 | ||
1570 | #endif | |
1571 | ||
1572 | static bool is_mnt_ns_file(struct dentry *dentry) | |
1573 | { | |
1574 | /* Is this a proxy for a mount namespace? */ | |
1575 | struct inode *inode = dentry->d_inode; | |
1576 | struct proc_ns *ei; | |
1577 | ||
1578 | if (!proc_ns_inode(inode)) | |
1579 | return false; | |
1580 | ||
1581 | ei = get_proc_ns(inode); | |
1582 | if (ei->ns_ops != &mntns_operations) | |
1583 | return false; | |
1584 | ||
1585 | return true; | |
1586 | } | |
1587 | ||
1588 | static bool mnt_ns_loop(struct dentry *dentry) | |
1589 | { | |
1590 | /* Could bind mounting the mount namespace inode cause a | |
1591 | * mount namespace loop? | |
1592 | */ | |
1593 | struct mnt_namespace *mnt_ns; | |
1594 | if (!is_mnt_ns_file(dentry)) | |
1595 | return false; | |
1596 | ||
1597 | mnt_ns = get_proc_ns(dentry->d_inode)->ns; | |
1598 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; | |
1599 | } | |
1600 | ||
1601 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, | |
1602 | int flag) | |
1603 | { | |
1604 | struct mount *res, *p, *q, *r, *parent; | |
1605 | ||
1606 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) | |
1607 | return ERR_PTR(-EINVAL); | |
1608 | ||
1609 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
1610 | return ERR_PTR(-EINVAL); | |
1611 | ||
1612 | res = q = clone_mnt(mnt, dentry, flag); | |
1613 | if (IS_ERR(q)) | |
1614 | return q; | |
1615 | ||
1616 | q->mnt.mnt_flags &= ~MNT_LOCKED; | |
1617 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
1618 | ||
1619 | p = mnt; | |
1620 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { | |
1621 | struct mount *s; | |
1622 | if (!is_subdir(r->mnt_mountpoint, dentry)) | |
1623 | continue; | |
1624 | ||
1625 | for (s = r; s; s = next_mnt(s, r)) { | |
1626 | struct mount *t = NULL; | |
1627 | if (!(flag & CL_COPY_UNBINDABLE) && | |
1628 | IS_MNT_UNBINDABLE(s)) { | |
1629 | s = skip_mnt_tree(s); | |
1630 | continue; | |
1631 | } | |
1632 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1633 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
1634 | s = skip_mnt_tree(s); | |
1635 | continue; | |
1636 | } | |
1637 | while (p != s->mnt_parent) { | |
1638 | p = p->mnt_parent; | |
1639 | q = q->mnt_parent; | |
1640 | } | |
1641 | p = s; | |
1642 | parent = q; | |
1643 | q = clone_mnt(p, p->mnt.mnt_root, flag); | |
1644 | if (IS_ERR(q)) | |
1645 | goto out; | |
1646 | lock_mount_hash(); | |
1647 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
1648 | mnt_set_mountpoint(parent, p->mnt_mp, q); | |
1649 | if (!list_empty(&parent->mnt_mounts)) { | |
1650 | t = list_last_entry(&parent->mnt_mounts, | |
1651 | struct mount, mnt_child); | |
1652 | if (t->mnt_mp != p->mnt_mp) | |
1653 | t = NULL; | |
1654 | } | |
1655 | attach_shadowed(q, parent, t); | |
1656 | unlock_mount_hash(); | |
1657 | } | |
1658 | } | |
1659 | return res; | |
1660 | out: | |
1661 | if (res) { | |
1662 | lock_mount_hash(); | |
1663 | umount_tree(res, 0); | |
1664 | unlock_mount_hash(); | |
1665 | } | |
1666 | return q; | |
1667 | } | |
1668 | ||
1669 | /* Caller should check returned pointer for errors */ | |
1670 | ||
1671 | struct vfsmount *collect_mounts(struct path *path) | |
1672 | { | |
1673 | struct mount *tree; | |
1674 | namespace_lock(); | |
1675 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1676 | CL_COPY_ALL | CL_PRIVATE); | |
1677 | namespace_unlock(); | |
1678 | if (IS_ERR(tree)) | |
1679 | return ERR_CAST(tree); | |
1680 | return &tree->mnt; | |
1681 | } | |
1682 | ||
1683 | void drop_collected_mounts(struct vfsmount *mnt) | |
1684 | { | |
1685 | namespace_lock(); | |
1686 | lock_mount_hash(); | |
1687 | umount_tree(real_mount(mnt), 0); | |
1688 | unlock_mount_hash(); | |
1689 | namespace_unlock(); | |
1690 | } | |
1691 | ||
1692 | /** | |
1693 | * clone_private_mount - create a private clone of a path | |
1694 | * | |
1695 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1696 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1697 | * to the originating mount won't be propagated into this). | |
1698 | * | |
1699 | * Release with mntput(). | |
1700 | */ | |
1701 | struct vfsmount *clone_private_mount(struct path *path) | |
1702 | { | |
1703 | struct mount *old_mnt = real_mount(path->mnt); | |
1704 | struct mount *new_mnt; | |
1705 | ||
1706 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1707 | return ERR_PTR(-EINVAL); | |
1708 | ||
1709 | down_read(&namespace_sem); | |
1710 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); | |
1711 | up_read(&namespace_sem); | |
1712 | if (IS_ERR(new_mnt)) | |
1713 | return ERR_CAST(new_mnt); | |
1714 | ||
1715 | return &new_mnt->mnt; | |
1716 | } | |
1717 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1718 | ||
1719 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, | |
1720 | struct vfsmount *root) | |
1721 | { | |
1722 | struct mount *mnt; | |
1723 | int res = f(root, arg); | |
1724 | if (res) | |
1725 | return res; | |
1726 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { | |
1727 | res = f(&mnt->mnt, arg); | |
1728 | if (res) | |
1729 | return res; | |
1730 | } | |
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) | |
1735 | { | |
1736 | struct mount *p; | |
1737 | ||
1738 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { | |
1739 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) | |
1740 | mnt_release_group_id(p); | |
1741 | } | |
1742 | } | |
1743 | ||
1744 | static int invent_group_ids(struct mount *mnt, bool recurse) | |
1745 | { | |
1746 | struct mount *p; | |
1747 | ||
1748 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { | |
1749 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { | |
1750 | int err = mnt_alloc_group_id(p); | |
1751 | if (err) { | |
1752 | cleanup_group_ids(mnt, p); | |
1753 | return err; | |
1754 | } | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | return 0; | |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * @source_mnt : mount tree to be attached | |
1763 | * @nd : place the mount tree @source_mnt is attached | |
1764 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1765 | * store the parent mount and mountpoint dentry. | |
1766 | * (done when source_mnt is moved) | |
1767 | * | |
1768 | * NOTE: in the table below explains the semantics when a source mount | |
1769 | * of a given type is attached to a destination mount of a given type. | |
1770 | * --------------------------------------------------------------------------- | |
1771 | * | BIND MOUNT OPERATION | | |
1772 | * |************************************************************************** | |
1773 | * | source-->| shared | private | slave | unbindable | | |
1774 | * | dest | | | | | | |
1775 | * | | | | | | | | |
1776 | * | v | | | | | | |
1777 | * |************************************************************************** | |
1778 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1779 | * | | | | | | | |
1780 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1781 | * *************************************************************************** | |
1782 | * A bind operation clones the source mount and mounts the clone on the | |
1783 | * destination mount. | |
1784 | * | |
1785 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1786 | * tree of the destination mount and the cloned mount is added to | |
1787 | * the peer group of the source mount. | |
1788 | * (+) the cloned mount is created under the destination mount and is marked | |
1789 | * as shared. The cloned mount is added to the peer group of the source | |
1790 | * mount. | |
1791 | * (+++) the mount is propagated to all the mounts in the propagation tree | |
1792 | * of the destination mount and the cloned mount is made slave | |
1793 | * of the same master as that of the source mount. The cloned mount | |
1794 | * is marked as 'shared and slave'. | |
1795 | * (*) the cloned mount is made a slave of the same master as that of the | |
1796 | * source mount. | |
1797 | * | |
1798 | * --------------------------------------------------------------------------- | |
1799 | * | MOVE MOUNT OPERATION | | |
1800 | * |************************************************************************** | |
1801 | * | source-->| shared | private | slave | unbindable | | |
1802 | * | dest | | | | | | |
1803 | * | | | | | | | | |
1804 | * | v | | | | | | |
1805 | * |************************************************************************** | |
1806 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1807 | * | | | | | | | |
1808 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1809 | * *************************************************************************** | |
1810 | * | |
1811 | * (+) the mount is moved to the destination. And is then propagated to | |
1812 | * all the mounts in the propagation tree of the destination mount. | |
1813 | * (+*) the mount is moved to the destination. | |
1814 | * (+++) the mount is moved to the destination and is then propagated to | |
1815 | * all the mounts belonging to the destination mount's propagation tree. | |
1816 | * the mount is marked as 'shared and slave'. | |
1817 | * (*) the mount continues to be a slave at the new location. | |
1818 | * | |
1819 | * if the source mount is a tree, the operations explained above is | |
1820 | * applied to each mount in the tree. | |
1821 | * Must be called without spinlocks held, since this function can sleep | |
1822 | * in allocations. | |
1823 | */ | |
1824 | static int attach_recursive_mnt(struct mount *source_mnt, | |
1825 | struct mount *dest_mnt, | |
1826 | struct mountpoint *dest_mp, | |
1827 | struct path *parent_path) | |
1828 | { | |
1829 | HLIST_HEAD(tree_list); | |
1830 | struct mount *child, *p; | |
1831 | struct hlist_node *n; | |
1832 | int err; | |
1833 | ||
1834 | if (IS_MNT_SHARED(dest_mnt)) { | |
1835 | err = invent_group_ids(source_mnt, true); | |
1836 | if (err) | |
1837 | goto out; | |
1838 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); | |
1839 | lock_mount_hash(); | |
1840 | if (err) | |
1841 | goto out_cleanup_ids; | |
1842 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
1843 | set_mnt_shared(p); | |
1844 | } else { | |
1845 | lock_mount_hash(); | |
1846 | } | |
1847 | if (parent_path) { | |
1848 | detach_mnt(source_mnt, parent_path); | |
1849 | attach_mnt(source_mnt, dest_mnt, dest_mp); | |
1850 | touch_mnt_namespace(source_mnt->mnt_ns); | |
1851 | } else { | |
1852 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); | |
1853 | commit_tree(source_mnt, NULL); | |
1854 | } | |
1855 | ||
1856 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { | |
1857 | struct mount *q; | |
1858 | hlist_del_init(&child->mnt_hash); | |
1859 | q = __lookup_mnt_last(&child->mnt_parent->mnt, | |
1860 | child->mnt_mountpoint); | |
1861 | commit_tree(child, q); | |
1862 | } | |
1863 | unlock_mount_hash(); | |
1864 | ||
1865 | return 0; | |
1866 | ||
1867 | out_cleanup_ids: | |
1868 | while (!hlist_empty(&tree_list)) { | |
1869 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
1870 | umount_tree(child, 0); | |
1871 | } | |
1872 | unlock_mount_hash(); | |
1873 | cleanup_group_ids(source_mnt, NULL); | |
1874 | out: | |
1875 | return err; | |
1876 | } | |
1877 | ||
1878 | static struct mountpoint *lock_mount(struct path *path) | |
1879 | { | |
1880 | struct vfsmount *mnt; | |
1881 | struct dentry *dentry = path->dentry; | |
1882 | retry: | |
1883 | mutex_lock(&dentry->d_inode->i_mutex); | |
1884 | if (unlikely(cant_mount(dentry))) { | |
1885 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1886 | return ERR_PTR(-ENOENT); | |
1887 | } | |
1888 | namespace_lock(); | |
1889 | mnt = lookup_mnt(path); | |
1890 | if (likely(!mnt)) { | |
1891 | struct mountpoint *mp = lookup_mountpoint(dentry); | |
1892 | if (!mp) | |
1893 | mp = new_mountpoint(dentry); | |
1894 | if (IS_ERR(mp)) { | |
1895 | namespace_unlock(); | |
1896 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1897 | return mp; | |
1898 | } | |
1899 | return mp; | |
1900 | } | |
1901 | namespace_unlock(); | |
1902 | mutex_unlock(&path->dentry->d_inode->i_mutex); | |
1903 | path_put(path); | |
1904 | path->mnt = mnt; | |
1905 | dentry = path->dentry = dget(mnt->mnt_root); | |
1906 | goto retry; | |
1907 | } | |
1908 | ||
1909 | static void unlock_mount(struct mountpoint *where) | |
1910 | { | |
1911 | struct dentry *dentry = where->m_dentry; | |
1912 | put_mountpoint(where); | |
1913 | namespace_unlock(); | |
1914 | mutex_unlock(&dentry->d_inode->i_mutex); | |
1915 | } | |
1916 | ||
1917 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) | |
1918 | { | |
1919 | if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) | |
1920 | return -EINVAL; | |
1921 | ||
1922 | if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != | |
1923 | S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) | |
1924 | return -ENOTDIR; | |
1925 | ||
1926 | return attach_recursive_mnt(mnt, p, mp, NULL); | |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * Sanity check the flags to change_mnt_propagation. | |
1931 | */ | |
1932 | ||
1933 | static int flags_to_propagation_type(int flags) | |
1934 | { | |
1935 | int type = flags & ~(MS_REC | MS_SILENT); | |
1936 | ||
1937 | /* Fail if any non-propagation flags are set */ | |
1938 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
1939 | return 0; | |
1940 | /* Only one propagation flag should be set */ | |
1941 | if (!is_power_of_2(type)) | |
1942 | return 0; | |
1943 | return type; | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * recursively change the type of the mountpoint. | |
1948 | */ | |
1949 | static int do_change_type(struct path *path, int flag) | |
1950 | { | |
1951 | struct mount *m; | |
1952 | struct mount *mnt = real_mount(path->mnt); | |
1953 | int recurse = flag & MS_REC; | |
1954 | int type; | |
1955 | int err = 0; | |
1956 | ||
1957 | if (path->dentry != path->mnt->mnt_root) | |
1958 | return -EINVAL; | |
1959 | ||
1960 | type = flags_to_propagation_type(flag); | |
1961 | if (!type) | |
1962 | return -EINVAL; | |
1963 | ||
1964 | namespace_lock(); | |
1965 | if (type == MS_SHARED) { | |
1966 | err = invent_group_ids(mnt, recurse); | |
1967 | if (err) | |
1968 | goto out_unlock; | |
1969 | } | |
1970 | ||
1971 | lock_mount_hash(); | |
1972 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
1973 | change_mnt_propagation(m, type); | |
1974 | unlock_mount_hash(); | |
1975 | ||
1976 | out_unlock: | |
1977 | namespace_unlock(); | |
1978 | return err; | |
1979 | } | |
1980 | ||
1981 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) | |
1982 | { | |
1983 | struct mount *child; | |
1984 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1985 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1986 | continue; | |
1987 | ||
1988 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1989 | return true; | |
1990 | } | |
1991 | return false; | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * do loopback mount. | |
1996 | */ | |
1997 | static int do_loopback(struct path *path, const char *old_name, | |
1998 | int recurse) | |
1999 | { | |
2000 | struct path old_path; | |
2001 | struct mount *mnt = NULL, *old, *parent; | |
2002 | struct mountpoint *mp; | |
2003 | int err; | |
2004 | if (!old_name || !*old_name) | |
2005 | return -EINVAL; | |
2006 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); | |
2007 | if (err) | |
2008 | return err; | |
2009 | ||
2010 | err = -EINVAL; | |
2011 | if (mnt_ns_loop(old_path.dentry)) | |
2012 | goto out; | |
2013 | ||
2014 | mp = lock_mount(path); | |
2015 | err = PTR_ERR(mp); | |
2016 | if (IS_ERR(mp)) | |
2017 | goto out; | |
2018 | ||
2019 | old = real_mount(old_path.mnt); | |
2020 | parent = real_mount(path->mnt); | |
2021 | ||
2022 | err = -EINVAL; | |
2023 | if (IS_MNT_UNBINDABLE(old)) | |
2024 | goto out2; | |
2025 | ||
2026 | if (!check_mnt(parent) || !check_mnt(old)) | |
2027 | goto out2; | |
2028 | ||
2029 | if (!recurse && has_locked_children(old, old_path.dentry)) | |
2030 | goto out2; | |
2031 | ||
2032 | if (recurse) | |
2033 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); | |
2034 | else | |
2035 | mnt = clone_mnt(old, old_path.dentry, 0); | |
2036 | ||
2037 | if (IS_ERR(mnt)) { | |
2038 | err = PTR_ERR(mnt); | |
2039 | goto out2; | |
2040 | } | |
2041 | ||
2042 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2043 | ||
2044 | err = graft_tree(mnt, parent, mp); | |
2045 | if (err) { | |
2046 | lock_mount_hash(); | |
2047 | umount_tree(mnt, 0); | |
2048 | unlock_mount_hash(); | |
2049 | } | |
2050 | out2: | |
2051 | unlock_mount(mp); | |
2052 | out: | |
2053 | path_put(&old_path); | |
2054 | return err; | |
2055 | } | |
2056 | ||
2057 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) | |
2058 | { | |
2059 | int error = 0; | |
2060 | int readonly_request = 0; | |
2061 | ||
2062 | if (ms_flags & MS_RDONLY) | |
2063 | readonly_request = 1; | |
2064 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2065 | return 0; | |
2066 | ||
2067 | if (readonly_request) | |
2068 | error = mnt_make_readonly(real_mount(mnt)); | |
2069 | else | |
2070 | __mnt_unmake_readonly(real_mount(mnt)); | |
2071 | return error; | |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * change filesystem flags. dir should be a physical root of filesystem. | |
2076 | * If you've mounted a non-root directory somewhere and want to do remount | |
2077 | * on it - tough luck. | |
2078 | */ | |
2079 | static int do_remount(struct path *path, int flags, int mnt_flags, | |
2080 | void *data) | |
2081 | { | |
2082 | int err; | |
2083 | struct super_block *sb = path->mnt->mnt_sb; | |
2084 | struct mount *mnt = real_mount(path->mnt); | |
2085 | ||
2086 | if (!check_mnt(mnt)) | |
2087 | return -EINVAL; | |
2088 | ||
2089 | if (path->dentry != path->mnt->mnt_root) | |
2090 | return -EINVAL; | |
2091 | ||
2092 | /* Don't allow changing of locked mnt flags. | |
2093 | * | |
2094 | * No locks need to be held here while testing the various | |
2095 | * MNT_LOCK flags because those flags can never be cleared | |
2096 | * once they are set. | |
2097 | */ | |
2098 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2099 | !(mnt_flags & MNT_READONLY)) { | |
2100 | return -EPERM; | |
2101 | } | |
2102 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && | |
2103 | !(mnt_flags & MNT_NODEV)) { | |
2104 | /* Was the nodev implicitly added in mount? */ | |
2105 | if ((mnt->mnt_ns->user_ns != &init_user_ns) && | |
2106 | !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2107 | mnt_flags |= MNT_NODEV; | |
2108 | } else { | |
2109 | return -EPERM; | |
2110 | } | |
2111 | } | |
2112 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2113 | !(mnt_flags & MNT_NOSUID)) { | |
2114 | return -EPERM; | |
2115 | } | |
2116 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2117 | !(mnt_flags & MNT_NOEXEC)) { | |
2118 | return -EPERM; | |
2119 | } | |
2120 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2121 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2122 | return -EPERM; | |
2123 | } | |
2124 | ||
2125 | err = security_sb_remount(sb, data); | |
2126 | if (err) | |
2127 | return err; | |
2128 | ||
2129 | down_write(&sb->s_umount); | |
2130 | if (flags & MS_BIND) | |
2131 | err = change_mount_flags(path->mnt, flags); | |
2132 | else if (!capable(CAP_SYS_ADMIN)) | |
2133 | err = -EPERM; | |
2134 | else | |
2135 | err = do_remount_sb(sb, flags, data, 0); | |
2136 | if (!err) { | |
2137 | lock_mount_hash(); | |
2138 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; | |
2139 | mnt->mnt.mnt_flags = mnt_flags; | |
2140 | touch_mnt_namespace(mnt->mnt_ns); | |
2141 | unlock_mount_hash(); | |
2142 | } | |
2143 | up_write(&sb->s_umount); | |
2144 | return err; | |
2145 | } | |
2146 | ||
2147 | static inline int tree_contains_unbindable(struct mount *mnt) | |
2148 | { | |
2149 | struct mount *p; | |
2150 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2151 | if (IS_MNT_UNBINDABLE(p)) | |
2152 | return 1; | |
2153 | } | |
2154 | return 0; | |
2155 | } | |
2156 | ||
2157 | static int do_move_mount(struct path *path, const char *old_name) | |
2158 | { | |
2159 | struct path old_path, parent_path; | |
2160 | struct mount *p; | |
2161 | struct mount *old; | |
2162 | struct mountpoint *mp; | |
2163 | int err; | |
2164 | if (!old_name || !*old_name) | |
2165 | return -EINVAL; | |
2166 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2167 | if (err) | |
2168 | return err; | |
2169 | ||
2170 | mp = lock_mount(path); | |
2171 | err = PTR_ERR(mp); | |
2172 | if (IS_ERR(mp)) | |
2173 | goto out; | |
2174 | ||
2175 | old = real_mount(old_path.mnt); | |
2176 | p = real_mount(path->mnt); | |
2177 | ||
2178 | err = -EINVAL; | |
2179 | if (!check_mnt(p) || !check_mnt(old)) | |
2180 | goto out1; | |
2181 | ||
2182 | if (old->mnt.mnt_flags & MNT_LOCKED) | |
2183 | goto out1; | |
2184 | ||
2185 | err = -EINVAL; | |
2186 | if (old_path.dentry != old_path.mnt->mnt_root) | |
2187 | goto out1; | |
2188 | ||
2189 | if (!mnt_has_parent(old)) | |
2190 | goto out1; | |
2191 | ||
2192 | if (S_ISDIR(path->dentry->d_inode->i_mode) != | |
2193 | S_ISDIR(old_path.dentry->d_inode->i_mode)) | |
2194 | goto out1; | |
2195 | /* | |
2196 | * Don't move a mount residing in a shared parent. | |
2197 | */ | |
2198 | if (IS_MNT_SHARED(old->mnt_parent)) | |
2199 | goto out1; | |
2200 | /* | |
2201 | * Don't move a mount tree containing unbindable mounts to a destination | |
2202 | * mount which is shared. | |
2203 | */ | |
2204 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) | |
2205 | goto out1; | |
2206 | err = -ELOOP; | |
2207 | for (; mnt_has_parent(p); p = p->mnt_parent) | |
2208 | if (p == old) | |
2209 | goto out1; | |
2210 | ||
2211 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); | |
2212 | if (err) | |
2213 | goto out1; | |
2214 | ||
2215 | /* if the mount is moved, it should no longer be expire | |
2216 | * automatically */ | |
2217 | list_del_init(&old->mnt_expire); | |
2218 | out1: | |
2219 | unlock_mount(mp); | |
2220 | out: | |
2221 | if (!err) | |
2222 | path_put(&parent_path); | |
2223 | path_put(&old_path); | |
2224 | return err; | |
2225 | } | |
2226 | ||
2227 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) | |
2228 | { | |
2229 | int err; | |
2230 | const char *subtype = strchr(fstype, '.'); | |
2231 | if (subtype) { | |
2232 | subtype++; | |
2233 | err = -EINVAL; | |
2234 | if (!subtype[0]) | |
2235 | goto err; | |
2236 | } else | |
2237 | subtype = ""; | |
2238 | ||
2239 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2240 | err = -ENOMEM; | |
2241 | if (!mnt->mnt_sb->s_subtype) | |
2242 | goto err; | |
2243 | return mnt; | |
2244 | ||
2245 | err: | |
2246 | mntput(mnt); | |
2247 | return ERR_PTR(err); | |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * add a mount into a namespace's mount tree | |
2252 | */ | |
2253 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) | |
2254 | { | |
2255 | struct mountpoint *mp; | |
2256 | struct mount *parent; | |
2257 | int err; | |
2258 | ||
2259 | mnt_flags &= ~MNT_INTERNAL_FLAGS; | |
2260 | ||
2261 | mp = lock_mount(path); | |
2262 | if (IS_ERR(mp)) | |
2263 | return PTR_ERR(mp); | |
2264 | ||
2265 | parent = real_mount(path->mnt); | |
2266 | err = -EINVAL; | |
2267 | if (unlikely(!check_mnt(parent))) { | |
2268 | /* that's acceptable only for automounts done in private ns */ | |
2269 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2270 | goto unlock; | |
2271 | /* ... and for those we'd better have mountpoint still alive */ | |
2272 | if (!parent->mnt_ns) | |
2273 | goto unlock; | |
2274 | } | |
2275 | ||
2276 | /* Refuse the same filesystem on the same mount point */ | |
2277 | err = -EBUSY; | |
2278 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && | |
2279 | path->mnt->mnt_root == path->dentry) | |
2280 | goto unlock; | |
2281 | ||
2282 | err = -EINVAL; | |
2283 | if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) | |
2284 | goto unlock; | |
2285 | ||
2286 | newmnt->mnt.mnt_flags = mnt_flags; | |
2287 | err = graft_tree(newmnt, parent, mp); | |
2288 | ||
2289 | unlock: | |
2290 | unlock_mount(mp); | |
2291 | return err; | |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * create a new mount for userspace and request it to be added into the | |
2296 | * namespace's tree | |
2297 | */ | |
2298 | static int do_new_mount(struct path *path, const char *fstype, int flags, | |
2299 | int mnt_flags, const char *name, void *data) | |
2300 | { | |
2301 | struct file_system_type *type; | |
2302 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2303 | struct vfsmount *mnt; | |
2304 | int err; | |
2305 | ||
2306 | if (!fstype) | |
2307 | return -EINVAL; | |
2308 | ||
2309 | type = get_fs_type(fstype); | |
2310 | if (!type) | |
2311 | return -ENODEV; | |
2312 | ||
2313 | if (user_ns != &init_user_ns) { | |
2314 | if (!(type->fs_flags & FS_USERNS_MOUNT)) { | |
2315 | put_filesystem(type); | |
2316 | return -EPERM; | |
2317 | } | |
2318 | /* Only in special cases allow devices from mounts | |
2319 | * created outside the initial user namespace. | |
2320 | */ | |
2321 | if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { | |
2322 | flags |= MS_NODEV; | |
2323 | mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; | |
2324 | } | |
2325 | } | |
2326 | ||
2327 | mnt = vfs_kern_mount(type, flags, name, data); | |
2328 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && | |
2329 | !mnt->mnt_sb->s_subtype) | |
2330 | mnt = fs_set_subtype(mnt, fstype); | |
2331 | ||
2332 | put_filesystem(type); | |
2333 | if (IS_ERR(mnt)) | |
2334 | return PTR_ERR(mnt); | |
2335 | ||
2336 | err = do_add_mount(real_mount(mnt), path, mnt_flags); | |
2337 | if (err) | |
2338 | mntput(mnt); | |
2339 | return err; | |
2340 | } | |
2341 | ||
2342 | int finish_automount(struct vfsmount *m, struct path *path) | |
2343 | { | |
2344 | struct mount *mnt = real_mount(m); | |
2345 | int err; | |
2346 | /* The new mount record should have at least 2 refs to prevent it being | |
2347 | * expired before we get a chance to add it | |
2348 | */ | |
2349 | BUG_ON(mnt_get_count(mnt) < 2); | |
2350 | ||
2351 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2352 | m->mnt_root == path->dentry) { | |
2353 | err = -ELOOP; | |
2354 | goto fail; | |
2355 | } | |
2356 | ||
2357 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); | |
2358 | if (!err) | |
2359 | return 0; | |
2360 | fail: | |
2361 | /* remove m from any expiration list it may be on */ | |
2362 | if (!list_empty(&mnt->mnt_expire)) { | |
2363 | namespace_lock(); | |
2364 | list_del_init(&mnt->mnt_expire); | |
2365 | namespace_unlock(); | |
2366 | } | |
2367 | mntput(m); | |
2368 | mntput(m); | |
2369 | return err; | |
2370 | } | |
2371 | ||
2372 | /** | |
2373 | * mnt_set_expiry - Put a mount on an expiration list | |
2374 | * @mnt: The mount to list. | |
2375 | * @expiry_list: The list to add the mount to. | |
2376 | */ | |
2377 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2378 | { | |
2379 | namespace_lock(); | |
2380 | ||
2381 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); | |
2382 | ||
2383 | namespace_unlock(); | |
2384 | } | |
2385 | EXPORT_SYMBOL(mnt_set_expiry); | |
2386 | ||
2387 | /* | |
2388 | * process a list of expirable mountpoints with the intent of discarding any | |
2389 | * mountpoints that aren't in use and haven't been touched since last we came | |
2390 | * here | |
2391 | */ | |
2392 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2393 | { | |
2394 | struct mount *mnt, *next; | |
2395 | LIST_HEAD(graveyard); | |
2396 | ||
2397 | if (list_empty(mounts)) | |
2398 | return; | |
2399 | ||
2400 | namespace_lock(); | |
2401 | lock_mount_hash(); | |
2402 | ||
2403 | /* extract from the expiration list every vfsmount that matches the | |
2404 | * following criteria: | |
2405 | * - only referenced by its parent vfsmount | |
2406 | * - still marked for expiry (marked on the last call here; marks are | |
2407 | * cleared by mntput()) | |
2408 | */ | |
2409 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { | |
2410 | if (!xchg(&mnt->mnt_expiry_mark, 1) || | |
2411 | propagate_mount_busy(mnt, 1)) | |
2412 | continue; | |
2413 | list_move(&mnt->mnt_expire, &graveyard); | |
2414 | } | |
2415 | while (!list_empty(&graveyard)) { | |
2416 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); | |
2417 | touch_mnt_namespace(mnt->mnt_ns); | |
2418 | umount_tree(mnt, 1); | |
2419 | } | |
2420 | unlock_mount_hash(); | |
2421 | namespace_unlock(); | |
2422 | } | |
2423 | ||
2424 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2425 | ||
2426 | /* | |
2427 | * Ripoff of 'select_parent()' | |
2428 | * | |
2429 | * search the list of submounts for a given mountpoint, and move any | |
2430 | * shrinkable submounts to the 'graveyard' list. | |
2431 | */ | |
2432 | static int select_submounts(struct mount *parent, struct list_head *graveyard) | |
2433 | { | |
2434 | struct mount *this_parent = parent; | |
2435 | struct list_head *next; | |
2436 | int found = 0; | |
2437 | ||
2438 | repeat: | |
2439 | next = this_parent->mnt_mounts.next; | |
2440 | resume: | |
2441 | while (next != &this_parent->mnt_mounts) { | |
2442 | struct list_head *tmp = next; | |
2443 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); | |
2444 | ||
2445 | next = tmp->next; | |
2446 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) | |
2447 | continue; | |
2448 | /* | |
2449 | * Descend a level if the d_mounts list is non-empty. | |
2450 | */ | |
2451 | if (!list_empty(&mnt->mnt_mounts)) { | |
2452 | this_parent = mnt; | |
2453 | goto repeat; | |
2454 | } | |
2455 | ||
2456 | if (!propagate_mount_busy(mnt, 1)) { | |
2457 | list_move_tail(&mnt->mnt_expire, graveyard); | |
2458 | found++; | |
2459 | } | |
2460 | } | |
2461 | /* | |
2462 | * All done at this level ... ascend and resume the search | |
2463 | */ | |
2464 | if (this_parent != parent) { | |
2465 | next = this_parent->mnt_child.next; | |
2466 | this_parent = this_parent->mnt_parent; | |
2467 | goto resume; | |
2468 | } | |
2469 | return found; | |
2470 | } | |
2471 | ||
2472 | /* | |
2473 | * process a list of expirable mountpoints with the intent of discarding any | |
2474 | * submounts of a specific parent mountpoint | |
2475 | * | |
2476 | * mount_lock must be held for write | |
2477 | */ | |
2478 | static void shrink_submounts(struct mount *mnt) | |
2479 | { | |
2480 | LIST_HEAD(graveyard); | |
2481 | struct mount *m; | |
2482 | ||
2483 | /* extract submounts of 'mountpoint' from the expiration list */ | |
2484 | while (select_submounts(mnt, &graveyard)) { | |
2485 | while (!list_empty(&graveyard)) { | |
2486 | m = list_first_entry(&graveyard, struct mount, | |
2487 | mnt_expire); | |
2488 | touch_mnt_namespace(m->mnt_ns); | |
2489 | umount_tree(m, 1); | |
2490 | } | |
2491 | } | |
2492 | } | |
2493 | ||
2494 | /* | |
2495 | * Some copy_from_user() implementations do not return the exact number of | |
2496 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2497 | * Note that this function differs from copy_from_user() in that it will oops | |
2498 | * on bad values of `to', rather than returning a short copy. | |
2499 | */ | |
2500 | static long exact_copy_from_user(void *to, const void __user * from, | |
2501 | unsigned long n) | |
2502 | { | |
2503 | char *t = to; | |
2504 | const char __user *f = from; | |
2505 | char c; | |
2506 | ||
2507 | if (!access_ok(VERIFY_READ, from, n)) | |
2508 | return n; | |
2509 | ||
2510 | while (n) { | |
2511 | if (__get_user(c, f)) { | |
2512 | memset(t, 0, n); | |
2513 | break; | |
2514 | } | |
2515 | *t++ = c; | |
2516 | f++; | |
2517 | n--; | |
2518 | } | |
2519 | return n; | |
2520 | } | |
2521 | ||
2522 | int copy_mount_options(const void __user * data, unsigned long *where) | |
2523 | { | |
2524 | int i; | |
2525 | unsigned long page; | |
2526 | unsigned long size; | |
2527 | ||
2528 | *where = 0; | |
2529 | if (!data) | |
2530 | return 0; | |
2531 | ||
2532 | if (!(page = __get_free_page(GFP_KERNEL))) | |
2533 | return -ENOMEM; | |
2534 | ||
2535 | /* We only care that *some* data at the address the user | |
2536 | * gave us is valid. Just in case, we'll zero | |
2537 | * the remainder of the page. | |
2538 | */ | |
2539 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2540 | size = TASK_SIZE - (unsigned long)data; | |
2541 | if (size > PAGE_SIZE) | |
2542 | size = PAGE_SIZE; | |
2543 | ||
2544 | i = size - exact_copy_from_user((void *)page, data, size); | |
2545 | if (!i) { | |
2546 | free_page(page); | |
2547 | return -EFAULT; | |
2548 | } | |
2549 | if (i != PAGE_SIZE) | |
2550 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
2551 | *where = page; | |
2552 | return 0; | |
2553 | } | |
2554 | ||
2555 | char *copy_mount_string(const void __user *data) | |
2556 | { | |
2557 | return data ? strndup_user(data, PAGE_SIZE) : NULL; | |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2562 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2563 | * | |
2564 | * data is a (void *) that can point to any structure up to | |
2565 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2566 | * information (or be NULL). | |
2567 | * | |
2568 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2569 | * When the flags word was introduced its top half was required | |
2570 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2571 | * Therefore, if this magic number is present, it carries no information | |
2572 | * and must be discarded. | |
2573 | */ | |
2574 | long do_mount(const char *dev_name, const char __user *dir_name, | |
2575 | const char *type_page, unsigned long flags, void *data_page) | |
2576 | { | |
2577 | struct path path; | |
2578 | int retval = 0; | |
2579 | int mnt_flags = 0; | |
2580 | ||
2581 | /* Discard magic */ | |
2582 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2583 | flags &= ~MS_MGC_MSK; | |
2584 | ||
2585 | /* Basic sanity checks */ | |
2586 | if (data_page) | |
2587 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2588 | ||
2589 | /* ... and get the mountpoint */ | |
2590 | retval = user_path(dir_name, &path); | |
2591 | if (retval) | |
2592 | return retval; | |
2593 | ||
2594 | retval = security_sb_mount(dev_name, &path, | |
2595 | type_page, flags, data_page); | |
2596 | if (!retval && !may_mount()) | |
2597 | retval = -EPERM; | |
2598 | if (retval) | |
2599 | goto dput_out; | |
2600 | ||
2601 | /* Default to relatime unless overriden */ | |
2602 | if (!(flags & MS_NOATIME)) | |
2603 | mnt_flags |= MNT_RELATIME; | |
2604 | ||
2605 | /* Separate the per-mountpoint flags */ | |
2606 | if (flags & MS_NOSUID) | |
2607 | mnt_flags |= MNT_NOSUID; | |
2608 | if (flags & MS_NODEV) | |
2609 | mnt_flags |= MNT_NODEV; | |
2610 | if (flags & MS_NOEXEC) | |
2611 | mnt_flags |= MNT_NOEXEC; | |
2612 | if (flags & MS_NOATIME) | |
2613 | mnt_flags |= MNT_NOATIME; | |
2614 | if (flags & MS_NODIRATIME) | |
2615 | mnt_flags |= MNT_NODIRATIME; | |
2616 | if (flags & MS_STRICTATIME) | |
2617 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
2618 | if (flags & MS_RDONLY) | |
2619 | mnt_flags |= MNT_READONLY; | |
2620 | ||
2621 | /* The default atime for remount is preservation */ | |
2622 | if ((flags & MS_REMOUNT) && | |
2623 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2624 | MS_STRICTATIME)) == 0)) { | |
2625 | mnt_flags &= ~MNT_ATIME_MASK; | |
2626 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2627 | } | |
2628 | ||
2629 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | | |
2630 | MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | | |
2631 | MS_STRICTATIME); | |
2632 | ||
2633 | if (flags & MS_REMOUNT) | |
2634 | retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, | |
2635 | data_page); | |
2636 | else if (flags & MS_BIND) | |
2637 | retval = do_loopback(&path, dev_name, flags & MS_REC); | |
2638 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2639 | retval = do_change_type(&path, flags); | |
2640 | else if (flags & MS_MOVE) | |
2641 | retval = do_move_mount(&path, dev_name); | |
2642 | else | |
2643 | retval = do_new_mount(&path, type_page, flags, mnt_flags, | |
2644 | dev_name, data_page); | |
2645 | dput_out: | |
2646 | path_put(&path); | |
2647 | return retval; | |
2648 | } | |
2649 | ||
2650 | static void free_mnt_ns(struct mnt_namespace *ns) | |
2651 | { | |
2652 | proc_free_inum(ns->proc_inum); | |
2653 | put_user_ns(ns->user_ns); | |
2654 | kfree(ns); | |
2655 | } | |
2656 | ||
2657 | /* | |
2658 | * Assign a sequence number so we can detect when we attempt to bind | |
2659 | * mount a reference to an older mount namespace into the current | |
2660 | * mount namespace, preventing reference counting loops. A 64bit | |
2661 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2662 | * is effectively never, so we can ignore the possibility. | |
2663 | */ | |
2664 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2665 | ||
2666 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) | |
2667 | { | |
2668 | struct mnt_namespace *new_ns; | |
2669 | int ret; | |
2670 | ||
2671 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); | |
2672 | if (!new_ns) | |
2673 | return ERR_PTR(-ENOMEM); | |
2674 | ret = proc_alloc_inum(&new_ns->proc_inum); | |
2675 | if (ret) { | |
2676 | kfree(new_ns); | |
2677 | return ERR_PTR(ret); | |
2678 | } | |
2679 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
2680 | atomic_set(&new_ns->count, 1); | |
2681 | new_ns->root = NULL; | |
2682 | INIT_LIST_HEAD(&new_ns->list); | |
2683 | init_waitqueue_head(&new_ns->poll); | |
2684 | new_ns->event = 0; | |
2685 | new_ns->user_ns = get_user_ns(user_ns); | |
2686 | return new_ns; | |
2687 | } | |
2688 | ||
2689 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, | |
2690 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
2691 | { | |
2692 | struct mnt_namespace *new_ns; | |
2693 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; | |
2694 | struct mount *p, *q; | |
2695 | struct mount *old; | |
2696 | struct mount *new; | |
2697 | int copy_flags; | |
2698 | ||
2699 | BUG_ON(!ns); | |
2700 | ||
2701 | if (likely(!(flags & CLONE_NEWNS))) { | |
2702 | get_mnt_ns(ns); | |
2703 | return ns; | |
2704 | } | |
2705 | ||
2706 | old = ns->root; | |
2707 | ||
2708 | new_ns = alloc_mnt_ns(user_ns); | |
2709 | if (IS_ERR(new_ns)) | |
2710 | return new_ns; | |
2711 | ||
2712 | namespace_lock(); | |
2713 | /* First pass: copy the tree topology */ | |
2714 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; | |
2715 | if (user_ns != ns->user_ns) | |
2716 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; | |
2717 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); | |
2718 | if (IS_ERR(new)) { | |
2719 | namespace_unlock(); | |
2720 | free_mnt_ns(new_ns); | |
2721 | return ERR_CAST(new); | |
2722 | } | |
2723 | new_ns->root = new; | |
2724 | list_add_tail(&new_ns->list, &new->mnt_list); | |
2725 | ||
2726 | /* | |
2727 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2728 | * as belonging to new namespace. We have already acquired a private | |
2729 | * fs_struct, so tsk->fs->lock is not needed. | |
2730 | */ | |
2731 | p = old; | |
2732 | q = new; | |
2733 | while (p) { | |
2734 | q->mnt_ns = new_ns; | |
2735 | if (new_fs) { | |
2736 | if (&p->mnt == new_fs->root.mnt) { | |
2737 | new_fs->root.mnt = mntget(&q->mnt); | |
2738 | rootmnt = &p->mnt; | |
2739 | } | |
2740 | if (&p->mnt == new_fs->pwd.mnt) { | |
2741 | new_fs->pwd.mnt = mntget(&q->mnt); | |
2742 | pwdmnt = &p->mnt; | |
2743 | } | |
2744 | } | |
2745 | p = next_mnt(p, old); | |
2746 | q = next_mnt(q, new); | |
2747 | if (!q) | |
2748 | break; | |
2749 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2750 | p = next_mnt(p, old); | |
2751 | } | |
2752 | namespace_unlock(); | |
2753 | ||
2754 | if (rootmnt) | |
2755 | mntput(rootmnt); | |
2756 | if (pwdmnt) | |
2757 | mntput(pwdmnt); | |
2758 | ||
2759 | return new_ns; | |
2760 | } | |
2761 | ||
2762 | /** | |
2763 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2764 | * @mnt: pointer to the new root filesystem mountpoint | |
2765 | */ | |
2766 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) | |
2767 | { | |
2768 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); | |
2769 | if (!IS_ERR(new_ns)) { | |
2770 | struct mount *mnt = real_mount(m); | |
2771 | mnt->mnt_ns = new_ns; | |
2772 | new_ns->root = mnt; | |
2773 | list_add(&mnt->mnt_list, &new_ns->list); | |
2774 | } else { | |
2775 | mntput(m); | |
2776 | } | |
2777 | return new_ns; | |
2778 | } | |
2779 | ||
2780 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) | |
2781 | { | |
2782 | struct mnt_namespace *ns; | |
2783 | struct super_block *s; | |
2784 | struct path path; | |
2785 | int err; | |
2786 | ||
2787 | ns = create_mnt_ns(mnt); | |
2788 | if (IS_ERR(ns)) | |
2789 | return ERR_CAST(ns); | |
2790 | ||
2791 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
2792 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
2793 | ||
2794 | put_mnt_ns(ns); | |
2795 | ||
2796 | if (err) | |
2797 | return ERR_PTR(err); | |
2798 | ||
2799 | /* trade a vfsmount reference for active sb one */ | |
2800 | s = path.mnt->mnt_sb; | |
2801 | atomic_inc(&s->s_active); | |
2802 | mntput(path.mnt); | |
2803 | /* lock the sucker */ | |
2804 | down_write(&s->s_umount); | |
2805 | /* ... and return the root of (sub)tree on it */ | |
2806 | return path.dentry; | |
2807 | } | |
2808 | EXPORT_SYMBOL(mount_subtree); | |
2809 | ||
2810 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, | |
2811 | char __user *, type, unsigned long, flags, void __user *, data) | |
2812 | { | |
2813 | int ret; | |
2814 | char *kernel_type; | |
2815 | char *kernel_dev; | |
2816 | unsigned long data_page; | |
2817 | ||
2818 | kernel_type = copy_mount_string(type); | |
2819 | ret = PTR_ERR(kernel_type); | |
2820 | if (IS_ERR(kernel_type)) | |
2821 | goto out_type; | |
2822 | ||
2823 | kernel_dev = copy_mount_string(dev_name); | |
2824 | ret = PTR_ERR(kernel_dev); | |
2825 | if (IS_ERR(kernel_dev)) | |
2826 | goto out_dev; | |
2827 | ||
2828 | ret = copy_mount_options(data, &data_page); | |
2829 | if (ret < 0) | |
2830 | goto out_data; | |
2831 | ||
2832 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, | |
2833 | (void *) data_page); | |
2834 | ||
2835 | free_page(data_page); | |
2836 | out_data: | |
2837 | kfree(kernel_dev); | |
2838 | out_dev: | |
2839 | kfree(kernel_type); | |
2840 | out_type: | |
2841 | return ret; | |
2842 | } | |
2843 | ||
2844 | /* | |
2845 | * Return true if path is reachable from root | |
2846 | * | |
2847 | * namespace_sem or mount_lock is held | |
2848 | */ | |
2849 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, | |
2850 | const struct path *root) | |
2851 | { | |
2852 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { | |
2853 | dentry = mnt->mnt_mountpoint; | |
2854 | mnt = mnt->mnt_parent; | |
2855 | } | |
2856 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); | |
2857 | } | |
2858 | ||
2859 | int path_is_under(struct path *path1, struct path *path2) | |
2860 | { | |
2861 | int res; | |
2862 | read_seqlock_excl(&mount_lock); | |
2863 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); | |
2864 | read_sequnlock_excl(&mount_lock); | |
2865 | return res; | |
2866 | } | |
2867 | EXPORT_SYMBOL(path_is_under); | |
2868 | ||
2869 | /* | |
2870 | * pivot_root Semantics: | |
2871 | * Moves the root file system of the current process to the directory put_old, | |
2872 | * makes new_root as the new root file system of the current process, and sets | |
2873 | * root/cwd of all processes which had them on the current root to new_root. | |
2874 | * | |
2875 | * Restrictions: | |
2876 | * The new_root and put_old must be directories, and must not be on the | |
2877 | * same file system as the current process root. The put_old must be | |
2878 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
2879 | * pointed to by put_old must yield the same directory as new_root. No other | |
2880 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
2881 | * | |
2882 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. | |
2883 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
2884 | * in this situation. | |
2885 | * | |
2886 | * Notes: | |
2887 | * - we don't move root/cwd if they are not at the root (reason: if something | |
2888 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
2889 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
2890 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
2891 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
2892 | * first. | |
2893 | */ | |
2894 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, | |
2895 | const char __user *, put_old) | |
2896 | { | |
2897 | struct path new, old, parent_path, root_parent, root; | |
2898 | struct mount *new_mnt, *root_mnt, *old_mnt; | |
2899 | struct mountpoint *old_mp, *root_mp; | |
2900 | int error; | |
2901 | ||
2902 | if (!may_mount()) | |
2903 | return -EPERM; | |
2904 | ||
2905 | error = user_path_dir(new_root, &new); | |
2906 | if (error) | |
2907 | goto out0; | |
2908 | ||
2909 | error = user_path_dir(put_old, &old); | |
2910 | if (error) | |
2911 | goto out1; | |
2912 | ||
2913 | error = security_sb_pivotroot(&old, &new); | |
2914 | if (error) | |
2915 | goto out2; | |
2916 | ||
2917 | get_fs_root(current->fs, &root); | |
2918 | old_mp = lock_mount(&old); | |
2919 | error = PTR_ERR(old_mp); | |
2920 | if (IS_ERR(old_mp)) | |
2921 | goto out3; | |
2922 | ||
2923 | error = -EINVAL; | |
2924 | new_mnt = real_mount(new.mnt); | |
2925 | root_mnt = real_mount(root.mnt); | |
2926 | old_mnt = real_mount(old.mnt); | |
2927 | if (IS_MNT_SHARED(old_mnt) || | |
2928 | IS_MNT_SHARED(new_mnt->mnt_parent) || | |
2929 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
2930 | goto out4; | |
2931 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) | |
2932 | goto out4; | |
2933 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) | |
2934 | goto out4; | |
2935 | error = -ENOENT; | |
2936 | if (d_unlinked(new.dentry)) | |
2937 | goto out4; | |
2938 | error = -EBUSY; | |
2939 | if (new_mnt == root_mnt || old_mnt == root_mnt) | |
2940 | goto out4; /* loop, on the same file system */ | |
2941 | error = -EINVAL; | |
2942 | if (root.mnt->mnt_root != root.dentry) | |
2943 | goto out4; /* not a mountpoint */ | |
2944 | if (!mnt_has_parent(root_mnt)) | |
2945 | goto out4; /* not attached */ | |
2946 | root_mp = root_mnt->mnt_mp; | |
2947 | if (new.mnt->mnt_root != new.dentry) | |
2948 | goto out4; /* not a mountpoint */ | |
2949 | if (!mnt_has_parent(new_mnt)) | |
2950 | goto out4; /* not attached */ | |
2951 | /* make sure we can reach put_old from new_root */ | |
2952 | if (!is_path_reachable(old_mnt, old.dentry, &new)) | |
2953 | goto out4; | |
2954 | /* make certain new is below the root */ | |
2955 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
2956 | goto out4; | |
2957 | root_mp->m_count++; /* pin it so it won't go away */ | |
2958 | lock_mount_hash(); | |
2959 | detach_mnt(new_mnt, &parent_path); | |
2960 | detach_mnt(root_mnt, &root_parent); | |
2961 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { | |
2962 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
2963 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2964 | } | |
2965 | /* mount old root on put_old */ | |
2966 | attach_mnt(root_mnt, old_mnt, old_mp); | |
2967 | /* mount new_root on / */ | |
2968 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); | |
2969 | touch_mnt_namespace(current->nsproxy->mnt_ns); | |
2970 | unlock_mount_hash(); | |
2971 | chroot_fs_refs(&root, &new); | |
2972 | put_mountpoint(root_mp); | |
2973 | error = 0; | |
2974 | out4: | |
2975 | unlock_mount(old_mp); | |
2976 | if (!error) { | |
2977 | path_put(&root_parent); | |
2978 | path_put(&parent_path); | |
2979 | } | |
2980 | out3: | |
2981 | path_put(&root); | |
2982 | out2: | |
2983 | path_put(&old); | |
2984 | out1: | |
2985 | path_put(&new); | |
2986 | out0: | |
2987 | return error; | |
2988 | } | |
2989 | ||
2990 | static void __init init_mount_tree(void) | |
2991 | { | |
2992 | struct vfsmount *mnt; | |
2993 | struct mnt_namespace *ns; | |
2994 | struct path root; | |
2995 | struct file_system_type *type; | |
2996 | ||
2997 | type = get_fs_type("rootfs"); | |
2998 | if (!type) | |
2999 | panic("Can't find rootfs type"); | |
3000 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3001 | put_filesystem(type); | |
3002 | if (IS_ERR(mnt)) | |
3003 | panic("Can't create rootfs"); | |
3004 | ||
3005 | ns = create_mnt_ns(mnt); | |
3006 | if (IS_ERR(ns)) | |
3007 | panic("Can't allocate initial namespace"); | |
3008 | ||
3009 | init_task.nsproxy->mnt_ns = ns; | |
3010 | get_mnt_ns(ns); | |
3011 | ||
3012 | root.mnt = mnt; | |
3013 | root.dentry = mnt->mnt_root; | |
3014 | ||
3015 | set_fs_pwd(current->fs, &root); | |
3016 | set_fs_root(current->fs, &root); | |
3017 | } | |
3018 | ||
3019 | void __init mnt_init(void) | |
3020 | { | |
3021 | unsigned u; | |
3022 | int err; | |
3023 | ||
3024 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), | |
3025 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3026 | ||
3027 | mount_hashtable = alloc_large_system_hash("Mount-cache", | |
3028 | sizeof(struct hlist_head), | |
3029 | mhash_entries, 19, | |
3030 | 0, | |
3031 | &m_hash_shift, &m_hash_mask, 0, 0); | |
3032 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3033 | sizeof(struct hlist_head), | |
3034 | mphash_entries, 19, | |
3035 | 0, | |
3036 | &mp_hash_shift, &mp_hash_mask, 0, 0); | |
3037 | ||
3038 | if (!mount_hashtable || !mountpoint_hashtable) | |
3039 | panic("Failed to allocate mount hash table\n"); | |
3040 | ||
3041 | for (u = 0; u <= m_hash_mask; u++) | |
3042 | INIT_HLIST_HEAD(&mount_hashtable[u]); | |
3043 | for (u = 0; u <= mp_hash_mask; u++) | |
3044 | INIT_HLIST_HEAD(&mountpoint_hashtable[u]); | |
3045 | ||
3046 | kernfs_init(); | |
3047 | ||
3048 | err = sysfs_init(); | |
3049 | if (err) | |
3050 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
3051 | __func__, err); | |
3052 | fs_kobj = kobject_create_and_add("fs", NULL); | |
3053 | if (!fs_kobj) | |
3054 | printk(KERN_WARNING "%s: kobj create error\n", __func__); | |
3055 | init_rootfs(); | |
3056 | init_mount_tree(); | |
3057 | } | |
3058 | ||
3059 | void put_mnt_ns(struct mnt_namespace *ns) | |
3060 | { | |
3061 | if (!atomic_dec_and_test(&ns->count)) | |
3062 | return; | |
3063 | drop_collected_mounts(&ns->root->mnt); | |
3064 | free_mnt_ns(ns); | |
3065 | } | |
3066 | ||
3067 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3068 | { | |
3069 | struct vfsmount *mnt; | |
3070 | mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); | |
3071 | if (!IS_ERR(mnt)) { | |
3072 | /* | |
3073 | * it is a longterm mount, don't release mnt until | |
3074 | * we unmount before file sys is unregistered | |
3075 | */ | |
3076 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; | |
3077 | } | |
3078 | return mnt; | |
3079 | } | |
3080 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
3081 | ||
3082 | void kern_unmount(struct vfsmount *mnt) | |
3083 | { | |
3084 | /* release long term mount so mount point can be released */ | |
3085 | if (!IS_ERR_OR_NULL(mnt)) { | |
3086 | real_mount(mnt)->mnt_ns = NULL; | |
3087 | synchronize_rcu(); /* yecchhh... */ | |
3088 | mntput(mnt); | |
3089 | } | |
3090 | } | |
3091 | EXPORT_SYMBOL(kern_unmount); | |
3092 | ||
3093 | bool our_mnt(struct vfsmount *mnt) | |
3094 | { | |
3095 | return check_mnt(real_mount(mnt)); | |
3096 | } | |
3097 | ||
3098 | bool current_chrooted(void) | |
3099 | { | |
3100 | /* Does the current process have a non-standard root */ | |
3101 | struct path ns_root; | |
3102 | struct path fs_root; | |
3103 | bool chrooted; | |
3104 | ||
3105 | /* Find the namespace root */ | |
3106 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3107 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3108 | path_get(&ns_root); | |
3109 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3110 | ; | |
3111 | ||
3112 | get_fs_root(current->fs, &fs_root); | |
3113 | ||
3114 | chrooted = !path_equal(&fs_root, &ns_root); | |
3115 | ||
3116 | path_put(&fs_root); | |
3117 | path_put(&ns_root); | |
3118 | ||
3119 | return chrooted; | |
3120 | } | |
3121 | ||
3122 | bool fs_fully_visible(struct file_system_type *type) | |
3123 | { | |
3124 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
3125 | struct mount *mnt; | |
3126 | bool visible = false; | |
3127 | ||
3128 | if (unlikely(!ns)) | |
3129 | return false; | |
3130 | ||
3131 | down_read(&namespace_sem); | |
3132 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
3133 | struct mount *child; | |
3134 | if (mnt->mnt.mnt_sb->s_type != type) | |
3135 | continue; | |
3136 | ||
3137 | /* This mount is not fully visible if there are any child mounts | |
3138 | * that cover anything except for empty directories. | |
3139 | */ | |
3140 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3141 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
3142 | if (!S_ISDIR(inode->i_mode)) | |
3143 | goto next; | |
3144 | if (inode->i_nlink > 2) | |
3145 | goto next; | |
3146 | } | |
3147 | visible = true; | |
3148 | goto found; | |
3149 | next: ; | |
3150 | } | |
3151 | found: | |
3152 | up_read(&namespace_sem); | |
3153 | return visible; | |
3154 | } | |
3155 | ||
3156 | static void *mntns_get(struct task_struct *task) | |
3157 | { | |
3158 | struct mnt_namespace *ns = NULL; | |
3159 | struct nsproxy *nsproxy; | |
3160 | ||
3161 | task_lock(task); | |
3162 | nsproxy = task->nsproxy; | |
3163 | if (nsproxy) { | |
3164 | ns = nsproxy->mnt_ns; | |
3165 | get_mnt_ns(ns); | |
3166 | } | |
3167 | task_unlock(task); | |
3168 | ||
3169 | return ns; | |
3170 | } | |
3171 | ||
3172 | static void mntns_put(void *ns) | |
3173 | { | |
3174 | put_mnt_ns(ns); | |
3175 | } | |
3176 | ||
3177 | static int mntns_install(struct nsproxy *nsproxy, void *ns) | |
3178 | { | |
3179 | struct fs_struct *fs = current->fs; | |
3180 | struct mnt_namespace *mnt_ns = ns; | |
3181 | struct path root; | |
3182 | ||
3183 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || | |
3184 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || | |
3185 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
3186 | return -EPERM; | |
3187 | ||
3188 | if (fs->users != 1) | |
3189 | return -EINVAL; | |
3190 | ||
3191 | get_mnt_ns(mnt_ns); | |
3192 | put_mnt_ns(nsproxy->mnt_ns); | |
3193 | nsproxy->mnt_ns = mnt_ns; | |
3194 | ||
3195 | /* Find the root */ | |
3196 | root.mnt = &mnt_ns->root->mnt; | |
3197 | root.dentry = mnt_ns->root->mnt.mnt_root; | |
3198 | path_get(&root); | |
3199 | while(d_mountpoint(root.dentry) && follow_down_one(&root)) | |
3200 | ; | |
3201 | ||
3202 | /* Update the pwd and root */ | |
3203 | set_fs_pwd(fs, &root); | |
3204 | set_fs_root(fs, &root); | |
3205 | ||
3206 | path_put(&root); | |
3207 | return 0; | |
3208 | } | |
3209 | ||
3210 | static unsigned int mntns_inum(void *ns) | |
3211 | { | |
3212 | struct mnt_namespace *mnt_ns = ns; | |
3213 | return mnt_ns->proc_inum; | |
3214 | } | |
3215 | ||
3216 | const struct proc_ns_operations mntns_operations = { | |
3217 | .name = "mnt", | |
3218 | .type = CLONE_NEWNS, | |
3219 | .get = mntns_get, | |
3220 | .put = mntns_put, | |
3221 | .install = mntns_install, | |
3222 | .inum = mntns_inum, | |
3223 | }; |