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;
set_mhash_entries(char * str)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;
set_mphash_entries(char * str)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
m_hash(struct vfsmount * mnt,struct dentry * dentry)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
mp_hash(struct dentry * dentry)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 */
mnt_alloc_id(struct mount * mnt)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
mnt_free_id(struct mount * mnt)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 */
mnt_alloc_group_id(struct mount * mnt)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 */
mnt_release_group_id(struct mount * mnt)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 */
mnt_add_count(struct mount * mnt,int n)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 */
mnt_get_count(struct mount * mnt)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
drop_mountpoint(struct fs_pin * p)193 static void drop_mountpoint(struct fs_pin *p)
194 {
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
197 pin_remove(p);
198 mntput(&m->mnt);
199 }
200
alloc_vfsmnt(const char * name)201 static struct mount *alloc_vfsmnt(const char *name)
202 {
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
204 if (mnt) {
205 int err;
206
207 err = mnt_alloc_id(mnt);
208 if (err)
209 goto out_free_cache;
210
211 if (name) {
212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 if (!mnt->mnt_devname)
214 goto out_free_id;
215 }
216
217 #ifdef CONFIG_SMP
218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219 if (!mnt->mnt_pcp)
220 goto out_free_devname;
221
222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
223 #else
224 mnt->mnt_count = 1;
225 mnt->mnt_writers = 0;
226 #endif
227
228 INIT_HLIST_NODE(&mnt->mnt_hash);
229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239 #endif
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241 }
242 return mnt;
243
244 #ifdef CONFIG_SMP
245 out_free_devname:
246 kfree_const(mnt->mnt_devname);
247 #endif
248 out_free_id:
249 mnt_free_id(mnt);
250 out_free_cache:
251 kmem_cache_free(mnt_cache, mnt);
252 return NULL;
253 }
254
255 /*
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
261 * a filesystem.
262 */
263 /*
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
266 *
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
272 * r/w.
273 */
__mnt_is_readonly(struct vfsmount * mnt)274 int __mnt_is_readonly(struct vfsmount *mnt)
275 {
276 if (mnt->mnt_flags & MNT_READONLY)
277 return 1;
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
279 return 1;
280 return 0;
281 }
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
mnt_inc_writers(struct mount * mnt)284 static inline void mnt_inc_writers(struct mount *mnt)
285 {
286 #ifdef CONFIG_SMP
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
288 #else
289 mnt->mnt_writers++;
290 #endif
291 }
292
mnt_dec_writers(struct mount * mnt)293 static inline void mnt_dec_writers(struct mount *mnt)
294 {
295 #ifdef CONFIG_SMP
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
297 #else
298 mnt->mnt_writers--;
299 #endif
300 }
301
mnt_get_writers(struct mount * mnt)302 static unsigned int mnt_get_writers(struct mount *mnt)
303 {
304 #ifdef CONFIG_SMP
305 unsigned int count = 0;
306 int cpu;
307
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 }
311
312 return count;
313 #else
314 return mnt->mnt_writers;
315 #endif
316 }
317
mnt_is_readonly(struct vfsmount * mnt)318 static int mnt_is_readonly(struct vfsmount *mnt)
319 {
320 if (mnt->mnt_sb->s_readonly_remount)
321 return 1;
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 smp_rmb();
324 return __mnt_is_readonly(mnt);
325 }
326
327 /*
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
332 */
333 /**
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
336 *
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
342 */
__mnt_want_write(struct vfsmount * m)343 int __mnt_want_write(struct vfsmount *m)
344 {
345 struct mount *mnt = real_mount(m);
346 int ret = 0;
347
348 preempt_disable();
349 mnt_inc_writers(mnt);
350 /*
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
354 */
355 smp_mb();
356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
357 cpu_relax();
358 /*
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
362 */
363 smp_rmb();
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
366 ret = -EROFS;
367 }
368 preempt_enable();
369
370 return ret;
371 }
372
373 /**
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
376 *
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
381 */
mnt_want_write(struct vfsmount * m)382 int mnt_want_write(struct vfsmount *m)
383 {
384 int ret;
385
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
388 if (ret)
389 sb_end_write(m->mnt_sb);
390 return ret;
391 }
392 EXPORT_SYMBOL_GPL(mnt_want_write);
393
394 /**
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
397 *
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
402 *
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
405 */
mnt_clone_write(struct vfsmount * mnt)406 int mnt_clone_write(struct vfsmount *mnt)
407 {
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
410 return -EROFS;
411 preempt_disable();
412 mnt_inc_writers(real_mount(mnt));
413 preempt_enable();
414 return 0;
415 }
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
417
418 /**
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
421 *
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
424 */
__mnt_want_write_file(struct file * file)425 int __mnt_want_write_file(struct file *file)
426 {
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
429 else
430 return mnt_clone_write(file->f_path.mnt);
431 }
432
433 /**
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
436 *
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
439 */
mnt_want_write_file(struct file * file)440 int mnt_want_write_file(struct file *file)
441 {
442 int ret;
443
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
446 if (ret)
447 sb_end_write(file->f_path.mnt->mnt_sb);
448 return ret;
449 }
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
451
452 /**
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
455 *
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
459 */
__mnt_drop_write(struct vfsmount * mnt)460 void __mnt_drop_write(struct vfsmount *mnt)
461 {
462 preempt_disable();
463 mnt_dec_writers(real_mount(mnt));
464 preempt_enable();
465 }
466
467 /**
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
470 *
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
474 */
mnt_drop_write(struct vfsmount * mnt)475 void mnt_drop_write(struct vfsmount *mnt)
476 {
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
479 }
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
481
__mnt_drop_write_file(struct file * file)482 void __mnt_drop_write_file(struct file *file)
483 {
484 __mnt_drop_write(file->f_path.mnt);
485 }
486
mnt_drop_write_file(struct file * file)487 void mnt_drop_write_file(struct file *file)
488 {
489 mnt_drop_write(file->f_path.mnt);
490 }
491 EXPORT_SYMBOL(mnt_drop_write_file);
492
mnt_make_readonly(struct mount * mnt)493 static int mnt_make_readonly(struct mount *mnt)
494 {
495 int ret = 0;
496
497 lock_mount_hash();
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
499 /*
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
502 */
503 smp_mb();
504
505 /*
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
510 *
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
520 */
521 if (mnt_get_writers(mnt) > 0)
522 ret = -EBUSY;
523 else
524 mnt->mnt.mnt_flags |= MNT_READONLY;
525 /*
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
528 */
529 smp_wmb();
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
531 unlock_mount_hash();
532 return ret;
533 }
534
__mnt_unmake_readonly(struct mount * mnt)535 static void __mnt_unmake_readonly(struct mount *mnt)
536 {
537 lock_mount_hash();
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
539 unlock_mount_hash();
540 }
541
sb_prepare_remount_readonly(struct super_block * sb)542 int sb_prepare_remount_readonly(struct super_block *sb)
543 {
544 struct mount *mnt;
545 int err = 0;
546
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
549 return -EBUSY;
550
551 lock_mount_hash();
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 smp_mb();
556 if (mnt_get_writers(mnt) > 0) {
557 err = -EBUSY;
558 break;
559 }
560 }
561 }
562 if (!err && atomic_long_read(&sb->s_remove_count))
563 err = -EBUSY;
564
565 if (!err) {
566 sb->s_readonly_remount = 1;
567 smp_wmb();
568 }
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572 }
573 unlock_mount_hash();
574
575 return err;
576 }
577
free_vfsmnt(struct mount * mnt)578 static void free_vfsmnt(struct mount *mnt)
579 {
580 kfree_const(mnt->mnt_devname);
581 #ifdef CONFIG_SMP
582 free_percpu(mnt->mnt_pcp);
583 #endif
584 kmem_cache_free(mnt_cache, mnt);
585 }
586
delayed_free_vfsmnt(struct rcu_head * head)587 static void delayed_free_vfsmnt(struct rcu_head *head)
588 {
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590 }
591
592 /* call under rcu_read_lock */
legitimize_mnt(struct vfsmount * bastard,unsigned seq)593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
594 {
595 struct mount *mnt;
596 if (read_seqretry(&mount_lock, seq))
597 return false;
598 if (bastard == NULL)
599 return true;
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
603 return true;
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
606 return false;
607 }
608 rcu_read_unlock();
609 mntput(bastard);
610 rcu_read_lock();
611 return false;
612 }
613
614 /*
615 * find the first mount at @dentry on vfsmount @mnt.
616 * call under rcu_read_lock()
617 */
__lookup_mnt(struct vfsmount * mnt,struct dentry * dentry)618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
619 {
620 struct hlist_head *head = m_hash(mnt, dentry);
621 struct mount *p;
622
623 hlist_for_each_entry_rcu(p, head, mnt_hash)
624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
625 return p;
626 return NULL;
627 }
628
629 /*
630 * find the last mount at @dentry on vfsmount @mnt.
631 * mount_lock must be held.
632 */
__lookup_mnt_last(struct vfsmount * mnt,struct dentry * dentry)633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
634 {
635 struct mount *p, *res = NULL;
636 p = __lookup_mnt(mnt, dentry);
637 if (!p)
638 goto out;
639 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
640 res = p;
641 hlist_for_each_entry_continue(p, mnt_hash) {
642 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
643 break;
644 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
645 res = p;
646 }
647 out:
648 return res;
649 }
650
651 /*
652 * lookup_mnt - Return the first child mount mounted at path
653 *
654 * "First" means first mounted chronologically. If you create the
655 * following mounts:
656 *
657 * mount /dev/sda1 /mnt
658 * mount /dev/sda2 /mnt
659 * mount /dev/sda3 /mnt
660 *
661 * Then lookup_mnt() on the base /mnt dentry in the root mount will
662 * return successively the root dentry and vfsmount of /dev/sda1, then
663 * /dev/sda2, then /dev/sda3, then NULL.
664 *
665 * lookup_mnt takes a reference to the found vfsmount.
666 */
lookup_mnt(struct path * path)667 struct vfsmount *lookup_mnt(struct path *path)
668 {
669 struct mount *child_mnt;
670 struct vfsmount *m;
671 unsigned seq;
672
673 rcu_read_lock();
674 do {
675 seq = read_seqbegin(&mount_lock);
676 child_mnt = __lookup_mnt(path->mnt, path->dentry);
677 m = child_mnt ? &child_mnt->mnt : NULL;
678 } while (!legitimize_mnt(m, seq));
679 rcu_read_unlock();
680 return m;
681 }
682
683 /*
684 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685 * current mount namespace.
686 *
687 * The common case is dentries are not mountpoints at all and that
688 * test is handled inline. For the slow case when we are actually
689 * dealing with a mountpoint of some kind, walk through all of the
690 * mounts in the current mount namespace and test to see if the dentry
691 * is a mountpoint.
692 *
693 * The mount_hashtable is not usable in the context because we
694 * need to identify all mounts that may be in the current mount
695 * namespace not just a mount that happens to have some specified
696 * parent mount.
697 */
__is_local_mountpoint(struct dentry * dentry)698 bool __is_local_mountpoint(struct dentry *dentry)
699 {
700 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
701 struct mount *mnt;
702 bool is_covered = false;
703
704 if (!d_mountpoint(dentry))
705 goto out;
706
707 down_read(&namespace_sem);
708 list_for_each_entry(mnt, &ns->list, mnt_list) {
709 is_covered = (mnt->mnt_mountpoint == dentry);
710 if (is_covered)
711 break;
712 }
713 up_read(&namespace_sem);
714 out:
715 return is_covered;
716 }
717
lookup_mountpoint(struct dentry * dentry)718 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
719 {
720 struct hlist_head *chain = mp_hash(dentry);
721 struct mountpoint *mp;
722
723 hlist_for_each_entry(mp, chain, m_hash) {
724 if (mp->m_dentry == dentry) {
725 /* might be worth a WARN_ON() */
726 if (d_unlinked(dentry))
727 return ERR_PTR(-ENOENT);
728 mp->m_count++;
729 return mp;
730 }
731 }
732 return NULL;
733 }
734
new_mountpoint(struct dentry * dentry)735 static struct mountpoint *new_mountpoint(struct dentry *dentry)
736 {
737 struct hlist_head *chain = mp_hash(dentry);
738 struct mountpoint *mp;
739 int ret;
740
741 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
742 if (!mp)
743 return ERR_PTR(-ENOMEM);
744
745 ret = d_set_mounted(dentry);
746 if (ret) {
747 kfree(mp);
748 return ERR_PTR(ret);
749 }
750
751 mp->m_dentry = dentry;
752 mp->m_count = 1;
753 hlist_add_head(&mp->m_hash, chain);
754 INIT_HLIST_HEAD(&mp->m_list);
755 return mp;
756 }
757
put_mountpoint(struct mountpoint * mp)758 static void put_mountpoint(struct mountpoint *mp)
759 {
760 if (!--mp->m_count) {
761 struct dentry *dentry = mp->m_dentry;
762 BUG_ON(!hlist_empty(&mp->m_list));
763 spin_lock(&dentry->d_lock);
764 dentry->d_flags &= ~DCACHE_MOUNTED;
765 spin_unlock(&dentry->d_lock);
766 hlist_del(&mp->m_hash);
767 kfree(mp);
768 }
769 }
770
check_mnt(struct mount * mnt)771 static inline int check_mnt(struct mount *mnt)
772 {
773 return mnt->mnt_ns == current->nsproxy->mnt_ns;
774 }
775
776 /*
777 * vfsmount lock must be held for write
778 */
touch_mnt_namespace(struct mnt_namespace * ns)779 static void touch_mnt_namespace(struct mnt_namespace *ns)
780 {
781 if (ns) {
782 ns->event = ++event;
783 wake_up_interruptible(&ns->poll);
784 }
785 }
786
787 /*
788 * vfsmount lock must be held for write
789 */
__touch_mnt_namespace(struct mnt_namespace * ns)790 static void __touch_mnt_namespace(struct mnt_namespace *ns)
791 {
792 if (ns && ns->event != event) {
793 ns->event = event;
794 wake_up_interruptible(&ns->poll);
795 }
796 }
797
798 /*
799 * vfsmount lock must be held for write
800 */
unhash_mnt(struct mount * mnt)801 static void unhash_mnt(struct mount *mnt)
802 {
803 mnt->mnt_parent = mnt;
804 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
805 list_del_init(&mnt->mnt_child);
806 hlist_del_init_rcu(&mnt->mnt_hash);
807 hlist_del_init(&mnt->mnt_mp_list);
808 put_mountpoint(mnt->mnt_mp);
809 mnt->mnt_mp = NULL;
810 }
811
812 /*
813 * vfsmount lock must be held for write
814 */
detach_mnt(struct mount * mnt,struct path * old_path)815 static void detach_mnt(struct mount *mnt, struct path *old_path)
816 {
817 old_path->dentry = mnt->mnt_mountpoint;
818 old_path->mnt = &mnt->mnt_parent->mnt;
819 unhash_mnt(mnt);
820 }
821
822 /*
823 * vfsmount lock must be held for write
824 */
umount_mnt(struct mount * mnt)825 static void umount_mnt(struct mount *mnt)
826 {
827 /* old mountpoint will be dropped when we can do that */
828 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
829 unhash_mnt(mnt);
830 }
831
832 /*
833 * vfsmount lock must be held for write
834 */
mnt_set_mountpoint(struct mount * mnt,struct mountpoint * mp,struct mount * child_mnt)835 void mnt_set_mountpoint(struct mount *mnt,
836 struct mountpoint *mp,
837 struct mount *child_mnt)
838 {
839 mp->m_count++;
840 mnt_add_count(mnt, 1); /* essentially, that's mntget */
841 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
842 child_mnt->mnt_parent = mnt;
843 child_mnt->mnt_mp = mp;
844 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
845 }
846
847 /*
848 * vfsmount lock must be held for write
849 */
attach_mnt(struct mount * mnt,struct mount * parent,struct mountpoint * mp)850 static void attach_mnt(struct mount *mnt,
851 struct mount *parent,
852 struct mountpoint *mp)
853 {
854 mnt_set_mountpoint(parent, mp, mnt);
855 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
856 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
857 }
858
attach_shadowed(struct mount * mnt,struct mount * parent,struct mount * shadows)859 static void attach_shadowed(struct mount *mnt,
860 struct mount *parent,
861 struct mount *shadows)
862 {
863 if (shadows) {
864 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
865 list_add(&mnt->mnt_child, &shadows->mnt_child);
866 } else {
867 hlist_add_head_rcu(&mnt->mnt_hash,
868 m_hash(&parent->mnt, mnt->mnt_mountpoint));
869 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
870 }
871 }
872
873 /*
874 * vfsmount lock must be held for write
875 */
commit_tree(struct mount * mnt,struct mount * shadows)876 static void commit_tree(struct mount *mnt, struct mount *shadows)
877 {
878 struct mount *parent = mnt->mnt_parent;
879 struct mount *m;
880 LIST_HEAD(head);
881 struct mnt_namespace *n = parent->mnt_ns;
882
883 BUG_ON(parent == mnt);
884
885 list_add_tail(&head, &mnt->mnt_list);
886 list_for_each_entry(m, &head, mnt_list)
887 m->mnt_ns = n;
888
889 list_splice(&head, n->list.prev);
890
891 attach_shadowed(mnt, parent, shadows);
892 touch_mnt_namespace(n);
893 }
894
next_mnt(struct mount * p,struct mount * root)895 static struct mount *next_mnt(struct mount *p, struct mount *root)
896 {
897 struct list_head *next = p->mnt_mounts.next;
898 if (next == &p->mnt_mounts) {
899 while (1) {
900 if (p == root)
901 return NULL;
902 next = p->mnt_child.next;
903 if (next != &p->mnt_parent->mnt_mounts)
904 break;
905 p = p->mnt_parent;
906 }
907 }
908 return list_entry(next, struct mount, mnt_child);
909 }
910
skip_mnt_tree(struct mount * p)911 static struct mount *skip_mnt_tree(struct mount *p)
912 {
913 struct list_head *prev = p->mnt_mounts.prev;
914 while (prev != &p->mnt_mounts) {
915 p = list_entry(prev, struct mount, mnt_child);
916 prev = p->mnt_mounts.prev;
917 }
918 return p;
919 }
920
921 struct vfsmount *
vfs_kern_mount(struct file_system_type * type,int flags,const char * name,void * data)922 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
923 {
924 struct mount *mnt;
925 struct dentry *root;
926
927 if (!type)
928 return ERR_PTR(-ENODEV);
929
930 mnt = alloc_vfsmnt(name);
931 if (!mnt)
932 return ERR_PTR(-ENOMEM);
933
934 if (flags & MS_KERNMOUNT)
935 mnt->mnt.mnt_flags = MNT_INTERNAL;
936
937 root = mount_fs(type, flags, name, data);
938 if (IS_ERR(root)) {
939 mnt_free_id(mnt);
940 free_vfsmnt(mnt);
941 return ERR_CAST(root);
942 }
943
944 mnt->mnt.mnt_root = root;
945 mnt->mnt.mnt_sb = root->d_sb;
946 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
947 mnt->mnt_parent = mnt;
948 lock_mount_hash();
949 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
950 unlock_mount_hash();
951 return &mnt->mnt;
952 }
953 EXPORT_SYMBOL_GPL(vfs_kern_mount);
954
clone_mnt(struct mount * old,struct dentry * root,int flag)955 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
956 int flag)
957 {
958 struct super_block *sb = old->mnt.mnt_sb;
959 struct mount *mnt;
960 int err;
961
962 mnt = alloc_vfsmnt(old->mnt_devname);
963 if (!mnt)
964 return ERR_PTR(-ENOMEM);
965
966 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
967 mnt->mnt_group_id = 0; /* not a peer of original */
968 else
969 mnt->mnt_group_id = old->mnt_group_id;
970
971 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
972 err = mnt_alloc_group_id(mnt);
973 if (err)
974 goto out_free;
975 }
976
977 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
978 /* Don't allow unprivileged users to change mount flags */
979 if (flag & CL_UNPRIVILEGED) {
980 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
981
982 if (mnt->mnt.mnt_flags & MNT_READONLY)
983 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
984
985 if (mnt->mnt.mnt_flags & MNT_NODEV)
986 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
987
988 if (mnt->mnt.mnt_flags & MNT_NOSUID)
989 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
990
991 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
992 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
993 }
994
995 /* Don't allow unprivileged users to reveal what is under a mount */
996 if ((flag & CL_UNPRIVILEGED) &&
997 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
998 mnt->mnt.mnt_flags |= MNT_LOCKED;
999
1000 atomic_inc(&sb->s_active);
1001 mnt->mnt.mnt_sb = sb;
1002 mnt->mnt.mnt_root = dget(root);
1003 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1004 mnt->mnt_parent = mnt;
1005 lock_mount_hash();
1006 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1007 unlock_mount_hash();
1008
1009 if ((flag & CL_SLAVE) ||
1010 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1011 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1012 mnt->mnt_master = old;
1013 CLEAR_MNT_SHARED(mnt);
1014 } else if (!(flag & CL_PRIVATE)) {
1015 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1016 list_add(&mnt->mnt_share, &old->mnt_share);
1017 if (IS_MNT_SLAVE(old))
1018 list_add(&mnt->mnt_slave, &old->mnt_slave);
1019 mnt->mnt_master = old->mnt_master;
1020 }
1021 if (flag & CL_MAKE_SHARED)
1022 set_mnt_shared(mnt);
1023
1024 /* stick the duplicate mount on the same expiry list
1025 * as the original if that was on one */
1026 if (flag & CL_EXPIRE) {
1027 if (!list_empty(&old->mnt_expire))
1028 list_add(&mnt->mnt_expire, &old->mnt_expire);
1029 }
1030
1031 return mnt;
1032
1033 out_free:
1034 mnt_free_id(mnt);
1035 free_vfsmnt(mnt);
1036 return ERR_PTR(err);
1037 }
1038
cleanup_mnt(struct mount * mnt)1039 static void cleanup_mnt(struct mount *mnt)
1040 {
1041 /*
1042 * This probably indicates that somebody messed
1043 * up a mnt_want/drop_write() pair. If this
1044 * happens, the filesystem was probably unable
1045 * to make r/w->r/o transitions.
1046 */
1047 /*
1048 * The locking used to deal with mnt_count decrement provides barriers,
1049 * so mnt_get_writers() below is safe.
1050 */
1051 WARN_ON(mnt_get_writers(mnt));
1052 if (unlikely(mnt->mnt_pins.first))
1053 mnt_pin_kill(mnt);
1054 fsnotify_vfsmount_delete(&mnt->mnt);
1055 dput(mnt->mnt.mnt_root);
1056 deactivate_super(mnt->mnt.mnt_sb);
1057 mnt_free_id(mnt);
1058 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1059 }
1060
__cleanup_mnt(struct rcu_head * head)1061 static void __cleanup_mnt(struct rcu_head *head)
1062 {
1063 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1064 }
1065
1066 static LLIST_HEAD(delayed_mntput_list);
delayed_mntput(struct work_struct * unused)1067 static void delayed_mntput(struct work_struct *unused)
1068 {
1069 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1070 struct llist_node *next;
1071
1072 for (; node; node = next) {
1073 next = llist_next(node);
1074 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1075 }
1076 }
1077 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1078
mntput_no_expire(struct mount * mnt)1079 static void mntput_no_expire(struct mount *mnt)
1080 {
1081 rcu_read_lock();
1082 mnt_add_count(mnt, -1);
1083 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1084 rcu_read_unlock();
1085 return;
1086 }
1087 lock_mount_hash();
1088 if (mnt_get_count(mnt)) {
1089 rcu_read_unlock();
1090 unlock_mount_hash();
1091 return;
1092 }
1093 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1094 rcu_read_unlock();
1095 unlock_mount_hash();
1096 return;
1097 }
1098 mnt->mnt.mnt_flags |= MNT_DOOMED;
1099 rcu_read_unlock();
1100
1101 list_del(&mnt->mnt_instance);
1102
1103 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1104 struct mount *p, *tmp;
1105 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1106 umount_mnt(p);
1107 }
1108 }
1109 unlock_mount_hash();
1110
1111 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1112 struct task_struct *task = current;
1113 if (likely(!(task->flags & PF_KTHREAD))) {
1114 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1115 if (!task_work_add(task, &mnt->mnt_rcu, true))
1116 return;
1117 }
1118 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1119 schedule_delayed_work(&delayed_mntput_work, 1);
1120 return;
1121 }
1122 cleanup_mnt(mnt);
1123 }
1124
mntput(struct vfsmount * mnt)1125 void mntput(struct vfsmount *mnt)
1126 {
1127 if (mnt) {
1128 struct mount *m = real_mount(mnt);
1129 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1130 if (unlikely(m->mnt_expiry_mark))
1131 m->mnt_expiry_mark = 0;
1132 mntput_no_expire(m);
1133 }
1134 }
1135 EXPORT_SYMBOL(mntput);
1136
mntget(struct vfsmount * mnt)1137 struct vfsmount *mntget(struct vfsmount *mnt)
1138 {
1139 if (mnt)
1140 mnt_add_count(real_mount(mnt), 1);
1141 return mnt;
1142 }
1143 EXPORT_SYMBOL(mntget);
1144
mnt_clone_internal(struct path * path)1145 struct vfsmount *mnt_clone_internal(struct path *path)
1146 {
1147 struct mount *p;
1148 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1149 if (IS_ERR(p))
1150 return ERR_CAST(p);
1151 p->mnt.mnt_flags |= MNT_INTERNAL;
1152 return &p->mnt;
1153 }
1154
mangle(struct seq_file * m,const char * s)1155 static inline void mangle(struct seq_file *m, const char *s)
1156 {
1157 seq_escape(m, s, " \t\n\\");
1158 }
1159
1160 /*
1161 * Simple .show_options callback for filesystems which don't want to
1162 * implement more complex mount option showing.
1163 *
1164 * See also save_mount_options().
1165 */
generic_show_options(struct seq_file * m,struct dentry * root)1166 int generic_show_options(struct seq_file *m, struct dentry *root)
1167 {
1168 const char *options;
1169
1170 rcu_read_lock();
1171 options = rcu_dereference(root->d_sb->s_options);
1172
1173 if (options != NULL && options[0]) {
1174 seq_putc(m, ',');
1175 mangle(m, options);
1176 }
1177 rcu_read_unlock();
1178
1179 return 0;
1180 }
1181 EXPORT_SYMBOL(generic_show_options);
1182
1183 /*
1184 * If filesystem uses generic_show_options(), this function should be
1185 * called from the fill_super() callback.
1186 *
1187 * The .remount_fs callback usually needs to be handled in a special
1188 * way, to make sure, that previous options are not overwritten if the
1189 * remount fails.
1190 *
1191 * Also note, that if the filesystem's .remount_fs function doesn't
1192 * reset all options to their default value, but changes only newly
1193 * given options, then the displayed options will not reflect reality
1194 * any more.
1195 */
save_mount_options(struct super_block * sb,char * options)1196 void save_mount_options(struct super_block *sb, char *options)
1197 {
1198 BUG_ON(sb->s_options);
1199 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1200 }
1201 EXPORT_SYMBOL(save_mount_options);
1202
replace_mount_options(struct super_block * sb,char * options)1203 void replace_mount_options(struct super_block *sb, char *options)
1204 {
1205 char *old = sb->s_options;
1206 rcu_assign_pointer(sb->s_options, options);
1207 if (old) {
1208 synchronize_rcu();
1209 kfree(old);
1210 }
1211 }
1212 EXPORT_SYMBOL(replace_mount_options);
1213
1214 #ifdef CONFIG_PROC_FS
1215 /* iterator; we want it to have access to namespace_sem, thus here... */
m_start(struct seq_file * m,loff_t * pos)1216 static void *m_start(struct seq_file *m, loff_t *pos)
1217 {
1218 struct proc_mounts *p = proc_mounts(m);
1219
1220 down_read(&namespace_sem);
1221 if (p->cached_event == p->ns->event) {
1222 void *v = p->cached_mount;
1223 if (*pos == p->cached_index)
1224 return v;
1225 if (*pos == p->cached_index + 1) {
1226 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1227 return p->cached_mount = v;
1228 }
1229 }
1230
1231 p->cached_event = p->ns->event;
1232 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1233 p->cached_index = *pos;
1234 return p->cached_mount;
1235 }
1236
m_next(struct seq_file * m,void * v,loff_t * pos)1237 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1238 {
1239 struct proc_mounts *p = proc_mounts(m);
1240
1241 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1242 p->cached_index = *pos;
1243 return p->cached_mount;
1244 }
1245
m_stop(struct seq_file * m,void * v)1246 static void m_stop(struct seq_file *m, void *v)
1247 {
1248 up_read(&namespace_sem);
1249 }
1250
m_show(struct seq_file * m,void * v)1251 static int m_show(struct seq_file *m, void *v)
1252 {
1253 struct proc_mounts *p = proc_mounts(m);
1254 struct mount *r = list_entry(v, struct mount, mnt_list);
1255 return p->show(m, &r->mnt);
1256 }
1257
1258 const struct seq_operations mounts_op = {
1259 .start = m_start,
1260 .next = m_next,
1261 .stop = m_stop,
1262 .show = m_show,
1263 };
1264 #endif /* CONFIG_PROC_FS */
1265
1266 /**
1267 * may_umount_tree - check if a mount tree is busy
1268 * @mnt: root of mount tree
1269 *
1270 * This is called to check if a tree of mounts has any
1271 * open files, pwds, chroots or sub mounts that are
1272 * busy.
1273 */
may_umount_tree(struct vfsmount * m)1274 int may_umount_tree(struct vfsmount *m)
1275 {
1276 struct mount *mnt = real_mount(m);
1277 int actual_refs = 0;
1278 int minimum_refs = 0;
1279 struct mount *p;
1280 BUG_ON(!m);
1281
1282 /* write lock needed for mnt_get_count */
1283 lock_mount_hash();
1284 for (p = mnt; p; p = next_mnt(p, mnt)) {
1285 actual_refs += mnt_get_count(p);
1286 minimum_refs += 2;
1287 }
1288 unlock_mount_hash();
1289
1290 if (actual_refs > minimum_refs)
1291 return 0;
1292
1293 return 1;
1294 }
1295
1296 EXPORT_SYMBOL(may_umount_tree);
1297
1298 /**
1299 * may_umount - check if a mount point is busy
1300 * @mnt: root of mount
1301 *
1302 * This is called to check if a mount point has any
1303 * open files, pwds, chroots or sub mounts. If the
1304 * mount has sub mounts this will return busy
1305 * regardless of whether the sub mounts are busy.
1306 *
1307 * Doesn't take quota and stuff into account. IOW, in some cases it will
1308 * give false negatives. The main reason why it's here is that we need
1309 * a non-destructive way to look for easily umountable filesystems.
1310 */
may_umount(struct vfsmount * mnt)1311 int may_umount(struct vfsmount *mnt)
1312 {
1313 int ret = 1;
1314 down_read(&namespace_sem);
1315 lock_mount_hash();
1316 if (propagate_mount_busy(real_mount(mnt), 2))
1317 ret = 0;
1318 unlock_mount_hash();
1319 up_read(&namespace_sem);
1320 return ret;
1321 }
1322
1323 EXPORT_SYMBOL(may_umount);
1324
1325 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1326
namespace_unlock(void)1327 static void namespace_unlock(void)
1328 {
1329 struct hlist_head head;
1330
1331 hlist_move_list(&unmounted, &head);
1332
1333 up_write(&namespace_sem);
1334
1335 if (likely(hlist_empty(&head)))
1336 return;
1337
1338 synchronize_rcu();
1339
1340 group_pin_kill(&head);
1341 }
1342
namespace_lock(void)1343 static inline void namespace_lock(void)
1344 {
1345 down_write(&namespace_sem);
1346 }
1347
1348 enum umount_tree_flags {
1349 UMOUNT_SYNC = 1,
1350 UMOUNT_PROPAGATE = 2,
1351 UMOUNT_CONNECTED = 4,
1352 };
1353
disconnect_mount(struct mount * mnt,enum umount_tree_flags how)1354 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1355 {
1356 /* Leaving mounts connected is only valid for lazy umounts */
1357 if (how & UMOUNT_SYNC)
1358 return true;
1359
1360 /* A mount without a parent has nothing to be connected to */
1361 if (!mnt_has_parent(mnt))
1362 return true;
1363
1364 /* Because the reference counting rules change when mounts are
1365 * unmounted and connected, umounted mounts may not be
1366 * connected to mounted mounts.
1367 */
1368 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1369 return true;
1370
1371 /* Has it been requested that the mount remain connected? */
1372 if (how & UMOUNT_CONNECTED)
1373 return false;
1374
1375 /* Is the mount locked such that it needs to remain connected? */
1376 if (IS_MNT_LOCKED(mnt))
1377 return false;
1378
1379 /* By default disconnect the mount */
1380 return true;
1381 }
1382
1383 /*
1384 * mount_lock must be held
1385 * namespace_sem must be held for write
1386 */
umount_tree(struct mount * mnt,enum umount_tree_flags how)1387 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1388 {
1389 LIST_HEAD(tmp_list);
1390 struct mount *p;
1391
1392 if (how & UMOUNT_PROPAGATE)
1393 propagate_mount_unlock(mnt);
1394
1395 /* Gather the mounts to umount */
1396 for (p = mnt; p; p = next_mnt(p, mnt)) {
1397 p->mnt.mnt_flags |= MNT_UMOUNT;
1398 list_move(&p->mnt_list, &tmp_list);
1399 }
1400
1401 /* Hide the mounts from mnt_mounts */
1402 list_for_each_entry(p, &tmp_list, mnt_list) {
1403 list_del_init(&p->mnt_child);
1404 }
1405
1406 /* Add propogated mounts to the tmp_list */
1407 if (how & UMOUNT_PROPAGATE)
1408 propagate_umount(&tmp_list);
1409
1410 while (!list_empty(&tmp_list)) {
1411 bool disconnect;
1412 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1413 list_del_init(&p->mnt_expire);
1414 list_del_init(&p->mnt_list);
1415 __touch_mnt_namespace(p->mnt_ns);
1416 p->mnt_ns = NULL;
1417 if (how & UMOUNT_SYNC)
1418 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1419
1420 disconnect = disconnect_mount(p, how);
1421
1422 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1423 disconnect ? &unmounted : NULL);
1424 if (mnt_has_parent(p)) {
1425 mnt_add_count(p->mnt_parent, -1);
1426 if (!disconnect) {
1427 /* Don't forget about p */
1428 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1429 } else {
1430 umount_mnt(p);
1431 }
1432 }
1433 change_mnt_propagation(p, MS_PRIVATE);
1434 }
1435 }
1436
1437 static void shrink_submounts(struct mount *mnt);
1438
do_umount(struct mount * mnt,int flags)1439 static int do_umount(struct mount *mnt, int flags)
1440 {
1441 struct super_block *sb = mnt->mnt.mnt_sb;
1442 int retval;
1443
1444 retval = security_sb_umount(&mnt->mnt, flags);
1445 if (retval)
1446 return retval;
1447
1448 /*
1449 * Allow userspace to request a mountpoint be expired rather than
1450 * unmounting unconditionally. Unmount only happens if:
1451 * (1) the mark is already set (the mark is cleared by mntput())
1452 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1453 */
1454 if (flags & MNT_EXPIRE) {
1455 if (&mnt->mnt == current->fs->root.mnt ||
1456 flags & (MNT_FORCE | MNT_DETACH))
1457 return -EINVAL;
1458
1459 /*
1460 * probably don't strictly need the lock here if we examined
1461 * all race cases, but it's a slowpath.
1462 */
1463 lock_mount_hash();
1464 if (mnt_get_count(mnt) != 2) {
1465 unlock_mount_hash();
1466 return -EBUSY;
1467 }
1468 unlock_mount_hash();
1469
1470 if (!xchg(&mnt->mnt_expiry_mark, 1))
1471 return -EAGAIN;
1472 }
1473
1474 /*
1475 * If we may have to abort operations to get out of this
1476 * mount, and they will themselves hold resources we must
1477 * allow the fs to do things. In the Unix tradition of
1478 * 'Gee thats tricky lets do it in userspace' the umount_begin
1479 * might fail to complete on the first run through as other tasks
1480 * must return, and the like. Thats for the mount program to worry
1481 * about for the moment.
1482 */
1483
1484 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1485 sb->s_op->umount_begin(sb);
1486 }
1487
1488 /*
1489 * No sense to grab the lock for this test, but test itself looks
1490 * somewhat bogus. Suggestions for better replacement?
1491 * Ho-hum... In principle, we might treat that as umount + switch
1492 * to rootfs. GC would eventually take care of the old vfsmount.
1493 * Actually it makes sense, especially if rootfs would contain a
1494 * /reboot - static binary that would close all descriptors and
1495 * call reboot(9). Then init(8) could umount root and exec /reboot.
1496 */
1497 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1498 /*
1499 * Special case for "unmounting" root ...
1500 * we just try to remount it readonly.
1501 */
1502 if (!capable(CAP_SYS_ADMIN))
1503 return -EPERM;
1504 down_write(&sb->s_umount);
1505 if (!(sb->s_flags & MS_RDONLY))
1506 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1507 up_write(&sb->s_umount);
1508 return retval;
1509 }
1510
1511 namespace_lock();
1512 lock_mount_hash();
1513 event++;
1514
1515 if (flags & MNT_DETACH) {
1516 if (!list_empty(&mnt->mnt_list))
1517 umount_tree(mnt, UMOUNT_PROPAGATE);
1518 retval = 0;
1519 } else {
1520 shrink_submounts(mnt);
1521 retval = -EBUSY;
1522 if (!propagate_mount_busy(mnt, 2)) {
1523 if (!list_empty(&mnt->mnt_list))
1524 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1525 retval = 0;
1526 }
1527 }
1528 unlock_mount_hash();
1529 namespace_unlock();
1530 return retval;
1531 }
1532
1533 /*
1534 * __detach_mounts - lazily unmount all mounts on the specified dentry
1535 *
1536 * During unlink, rmdir, and d_drop it is possible to loose the path
1537 * to an existing mountpoint, and wind up leaking the mount.
1538 * detach_mounts allows lazily unmounting those mounts instead of
1539 * leaking them.
1540 *
1541 * The caller may hold dentry->d_inode->i_mutex.
1542 */
__detach_mounts(struct dentry * dentry)1543 void __detach_mounts(struct dentry *dentry)
1544 {
1545 struct mountpoint *mp;
1546 struct mount *mnt;
1547
1548 namespace_lock();
1549 mp = lookup_mountpoint(dentry);
1550 if (IS_ERR_OR_NULL(mp))
1551 goto out_unlock;
1552
1553 lock_mount_hash();
1554 while (!hlist_empty(&mp->m_list)) {
1555 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1556 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1557 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1558 umount_mnt(mnt);
1559 }
1560 else umount_tree(mnt, UMOUNT_CONNECTED);
1561 }
1562 unlock_mount_hash();
1563 put_mountpoint(mp);
1564 out_unlock:
1565 namespace_unlock();
1566 }
1567
1568 /*
1569 * Is the caller allowed to modify his namespace?
1570 */
may_mount(void)1571 static inline bool may_mount(void)
1572 {
1573 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1574 }
1575
1576 /*
1577 * Now umount can handle mount points as well as block devices.
1578 * This is important for filesystems which use unnamed block devices.
1579 *
1580 * We now support a flag for forced unmount like the other 'big iron'
1581 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1582 */
1583
SYSCALL_DEFINE2(umount,char __user *,name,int,flags)1584 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1585 {
1586 struct path path;
1587 struct mount *mnt;
1588 int retval;
1589 int lookup_flags = 0;
1590
1591 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1592 return -EINVAL;
1593
1594 if (!may_mount())
1595 return -EPERM;
1596
1597 if (!(flags & UMOUNT_NOFOLLOW))
1598 lookup_flags |= LOOKUP_FOLLOW;
1599
1600 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1601 if (retval)
1602 goto out;
1603 mnt = real_mount(path.mnt);
1604 retval = -EINVAL;
1605 if (path.dentry != path.mnt->mnt_root)
1606 goto dput_and_out;
1607 if (!check_mnt(mnt))
1608 goto dput_and_out;
1609 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1610 goto dput_and_out;
1611 retval = -EPERM;
1612 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1613 goto dput_and_out;
1614
1615 retval = do_umount(mnt, flags);
1616 dput_and_out:
1617 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1618 dput(path.dentry);
1619 mntput_no_expire(mnt);
1620 out:
1621 return retval;
1622 }
1623
1624 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1625
1626 /*
1627 * The 2.0 compatible umount. No flags.
1628 */
SYSCALL_DEFINE1(oldumount,char __user *,name)1629 SYSCALL_DEFINE1(oldumount, char __user *, name)
1630 {
1631 return sys_umount(name, 0);
1632 }
1633
1634 #endif
1635
is_mnt_ns_file(struct dentry * dentry)1636 static bool is_mnt_ns_file(struct dentry *dentry)
1637 {
1638 /* Is this a proxy for a mount namespace? */
1639 return dentry->d_op == &ns_dentry_operations &&
1640 dentry->d_fsdata == &mntns_operations;
1641 }
1642
to_mnt_ns(struct ns_common * ns)1643 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1644 {
1645 return container_of(ns, struct mnt_namespace, ns);
1646 }
1647
mnt_ns_loop(struct dentry * dentry)1648 static bool mnt_ns_loop(struct dentry *dentry)
1649 {
1650 /* Could bind mounting the mount namespace inode cause a
1651 * mount namespace loop?
1652 */
1653 struct mnt_namespace *mnt_ns;
1654 if (!is_mnt_ns_file(dentry))
1655 return false;
1656
1657 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1658 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1659 }
1660
copy_tree(struct mount * mnt,struct dentry * dentry,int flag)1661 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1662 int flag)
1663 {
1664 struct mount *res, *p, *q, *r, *parent;
1665
1666 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1667 return ERR_PTR(-EINVAL);
1668
1669 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1670 return ERR_PTR(-EINVAL);
1671
1672 res = q = clone_mnt(mnt, dentry, flag);
1673 if (IS_ERR(q))
1674 return q;
1675
1676 q->mnt_mountpoint = mnt->mnt_mountpoint;
1677
1678 p = mnt;
1679 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1680 struct mount *s;
1681 if (!is_subdir(r->mnt_mountpoint, dentry))
1682 continue;
1683
1684 for (s = r; s; s = next_mnt(s, r)) {
1685 struct mount *t = NULL;
1686 if (!(flag & CL_COPY_UNBINDABLE) &&
1687 IS_MNT_UNBINDABLE(s)) {
1688 s = skip_mnt_tree(s);
1689 continue;
1690 }
1691 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1692 is_mnt_ns_file(s->mnt.mnt_root)) {
1693 s = skip_mnt_tree(s);
1694 continue;
1695 }
1696 while (p != s->mnt_parent) {
1697 p = p->mnt_parent;
1698 q = q->mnt_parent;
1699 }
1700 p = s;
1701 parent = q;
1702 q = clone_mnt(p, p->mnt.mnt_root, flag);
1703 if (IS_ERR(q))
1704 goto out;
1705 lock_mount_hash();
1706 list_add_tail(&q->mnt_list, &res->mnt_list);
1707 mnt_set_mountpoint(parent, p->mnt_mp, q);
1708 if (!list_empty(&parent->mnt_mounts)) {
1709 t = list_last_entry(&parent->mnt_mounts,
1710 struct mount, mnt_child);
1711 if (t->mnt_mp != p->mnt_mp)
1712 t = NULL;
1713 }
1714 attach_shadowed(q, parent, t);
1715 unlock_mount_hash();
1716 }
1717 }
1718 return res;
1719 out:
1720 if (res) {
1721 lock_mount_hash();
1722 umount_tree(res, UMOUNT_SYNC);
1723 unlock_mount_hash();
1724 }
1725 return q;
1726 }
1727
1728 /* Caller should check returned pointer for errors */
1729
collect_mounts(struct path * path)1730 struct vfsmount *collect_mounts(struct path *path)
1731 {
1732 struct mount *tree;
1733 namespace_lock();
1734 if (!check_mnt(real_mount(path->mnt)))
1735 tree = ERR_PTR(-EINVAL);
1736 else
1737 tree = copy_tree(real_mount(path->mnt), path->dentry,
1738 CL_COPY_ALL | CL_PRIVATE);
1739 namespace_unlock();
1740 if (IS_ERR(tree))
1741 return ERR_CAST(tree);
1742 return &tree->mnt;
1743 }
1744
drop_collected_mounts(struct vfsmount * mnt)1745 void drop_collected_mounts(struct vfsmount *mnt)
1746 {
1747 namespace_lock();
1748 lock_mount_hash();
1749 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1750 unlock_mount_hash();
1751 namespace_unlock();
1752 }
1753
1754 /**
1755 * clone_private_mount - create a private clone of a path
1756 *
1757 * This creates a new vfsmount, which will be the clone of @path. The new will
1758 * not be attached anywhere in the namespace and will be private (i.e. changes
1759 * to the originating mount won't be propagated into this).
1760 *
1761 * Release with mntput().
1762 */
clone_private_mount(struct path * path)1763 struct vfsmount *clone_private_mount(struct path *path)
1764 {
1765 struct mount *old_mnt = real_mount(path->mnt);
1766 struct mount *new_mnt;
1767
1768 if (IS_MNT_UNBINDABLE(old_mnt))
1769 return ERR_PTR(-EINVAL);
1770
1771 down_read(&namespace_sem);
1772 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1773 up_read(&namespace_sem);
1774 if (IS_ERR(new_mnt))
1775 return ERR_CAST(new_mnt);
1776
1777 return &new_mnt->mnt;
1778 }
1779 EXPORT_SYMBOL_GPL(clone_private_mount);
1780
iterate_mounts(int (* f)(struct vfsmount *,void *),void * arg,struct vfsmount * root)1781 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1782 struct vfsmount *root)
1783 {
1784 struct mount *mnt;
1785 int res = f(root, arg);
1786 if (res)
1787 return res;
1788 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1789 res = f(&mnt->mnt, arg);
1790 if (res)
1791 return res;
1792 }
1793 return 0;
1794 }
1795
cleanup_group_ids(struct mount * mnt,struct mount * end)1796 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1797 {
1798 struct mount *p;
1799
1800 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1801 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1802 mnt_release_group_id(p);
1803 }
1804 }
1805
invent_group_ids(struct mount * mnt,bool recurse)1806 static int invent_group_ids(struct mount *mnt, bool recurse)
1807 {
1808 struct mount *p;
1809
1810 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1811 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1812 int err = mnt_alloc_group_id(p);
1813 if (err) {
1814 cleanup_group_ids(mnt, p);
1815 return err;
1816 }
1817 }
1818 }
1819
1820 return 0;
1821 }
1822
1823 /*
1824 * @source_mnt : mount tree to be attached
1825 * @nd : place the mount tree @source_mnt is attached
1826 * @parent_nd : if non-null, detach the source_mnt from its parent and
1827 * store the parent mount and mountpoint dentry.
1828 * (done when source_mnt is moved)
1829 *
1830 * NOTE: in the table below explains the semantics when a source mount
1831 * of a given type is attached to a destination mount of a given type.
1832 * ---------------------------------------------------------------------------
1833 * | BIND MOUNT OPERATION |
1834 * |**************************************************************************
1835 * | source-->| shared | private | slave | unbindable |
1836 * | dest | | | | |
1837 * | | | | | | |
1838 * | v | | | | |
1839 * |**************************************************************************
1840 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1841 * | | | | | |
1842 * |non-shared| shared (+) | private | slave (*) | invalid |
1843 * ***************************************************************************
1844 * A bind operation clones the source mount and mounts the clone on the
1845 * destination mount.
1846 *
1847 * (++) the cloned mount is propagated to all the mounts in the propagation
1848 * tree of the destination mount and the cloned mount is added to
1849 * the peer group of the source mount.
1850 * (+) the cloned mount is created under the destination mount and is marked
1851 * as shared. The cloned mount is added to the peer group of the source
1852 * mount.
1853 * (+++) the mount is propagated to all the mounts in the propagation tree
1854 * of the destination mount and the cloned mount is made slave
1855 * of the same master as that of the source mount. The cloned mount
1856 * is marked as 'shared and slave'.
1857 * (*) the cloned mount is made a slave of the same master as that of the
1858 * source mount.
1859 *
1860 * ---------------------------------------------------------------------------
1861 * | MOVE MOUNT OPERATION |
1862 * |**************************************************************************
1863 * | source-->| shared | private | slave | unbindable |
1864 * | dest | | | | |
1865 * | | | | | | |
1866 * | v | | | | |
1867 * |**************************************************************************
1868 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1869 * | | | | | |
1870 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1871 * ***************************************************************************
1872 *
1873 * (+) the mount is moved to the destination. And is then propagated to
1874 * all the mounts in the propagation tree of the destination mount.
1875 * (+*) the mount is moved to the destination.
1876 * (+++) the mount is moved to the destination and is then propagated to
1877 * all the mounts belonging to the destination mount's propagation tree.
1878 * the mount is marked as 'shared and slave'.
1879 * (*) the mount continues to be a slave at the new location.
1880 *
1881 * if the source mount is a tree, the operations explained above is
1882 * applied to each mount in the tree.
1883 * Must be called without spinlocks held, since this function can sleep
1884 * in allocations.
1885 */
attach_recursive_mnt(struct mount * source_mnt,struct mount * dest_mnt,struct mountpoint * dest_mp,struct path * parent_path)1886 static int attach_recursive_mnt(struct mount *source_mnt,
1887 struct mount *dest_mnt,
1888 struct mountpoint *dest_mp,
1889 struct path *parent_path)
1890 {
1891 HLIST_HEAD(tree_list);
1892 struct mount *child, *p;
1893 struct hlist_node *n;
1894 int err;
1895
1896 if (IS_MNT_SHARED(dest_mnt)) {
1897 err = invent_group_ids(source_mnt, true);
1898 if (err)
1899 goto out;
1900 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1901 lock_mount_hash();
1902 if (err)
1903 goto out_cleanup_ids;
1904 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1905 set_mnt_shared(p);
1906 } else {
1907 lock_mount_hash();
1908 }
1909 if (parent_path) {
1910 detach_mnt(source_mnt, parent_path);
1911 attach_mnt(source_mnt, dest_mnt, dest_mp);
1912 touch_mnt_namespace(source_mnt->mnt_ns);
1913 } else {
1914 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1915 commit_tree(source_mnt, NULL);
1916 }
1917
1918 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1919 struct mount *q;
1920 hlist_del_init(&child->mnt_hash);
1921 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1922 child->mnt_mountpoint);
1923 commit_tree(child, q);
1924 }
1925 unlock_mount_hash();
1926
1927 return 0;
1928
1929 out_cleanup_ids:
1930 while (!hlist_empty(&tree_list)) {
1931 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1932 umount_tree(child, UMOUNT_SYNC);
1933 }
1934 unlock_mount_hash();
1935 cleanup_group_ids(source_mnt, NULL);
1936 out:
1937 return err;
1938 }
1939
lock_mount(struct path * path)1940 static struct mountpoint *lock_mount(struct path *path)
1941 {
1942 struct vfsmount *mnt;
1943 struct dentry *dentry = path->dentry;
1944 retry:
1945 mutex_lock(&dentry->d_inode->i_mutex);
1946 if (unlikely(cant_mount(dentry))) {
1947 mutex_unlock(&dentry->d_inode->i_mutex);
1948 return ERR_PTR(-ENOENT);
1949 }
1950 namespace_lock();
1951 mnt = lookup_mnt(path);
1952 if (likely(!mnt)) {
1953 struct mountpoint *mp = lookup_mountpoint(dentry);
1954 if (!mp)
1955 mp = new_mountpoint(dentry);
1956 if (IS_ERR(mp)) {
1957 namespace_unlock();
1958 mutex_unlock(&dentry->d_inode->i_mutex);
1959 return mp;
1960 }
1961 return mp;
1962 }
1963 namespace_unlock();
1964 mutex_unlock(&path->dentry->d_inode->i_mutex);
1965 path_put(path);
1966 path->mnt = mnt;
1967 dentry = path->dentry = dget(mnt->mnt_root);
1968 goto retry;
1969 }
1970
unlock_mount(struct mountpoint * where)1971 static void unlock_mount(struct mountpoint *where)
1972 {
1973 struct dentry *dentry = where->m_dentry;
1974 put_mountpoint(where);
1975 namespace_unlock();
1976 mutex_unlock(&dentry->d_inode->i_mutex);
1977 }
1978
graft_tree(struct mount * mnt,struct mount * p,struct mountpoint * mp)1979 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1980 {
1981 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1982 return -EINVAL;
1983
1984 if (d_is_dir(mp->m_dentry) !=
1985 d_is_dir(mnt->mnt.mnt_root))
1986 return -ENOTDIR;
1987
1988 return attach_recursive_mnt(mnt, p, mp, NULL);
1989 }
1990
1991 /*
1992 * Sanity check the flags to change_mnt_propagation.
1993 */
1994
flags_to_propagation_type(int flags)1995 static int flags_to_propagation_type(int flags)
1996 {
1997 int type = flags & ~(MS_REC | MS_SILENT);
1998
1999 /* Fail if any non-propagation flags are set */
2000 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2001 return 0;
2002 /* Only one propagation flag should be set */
2003 if (!is_power_of_2(type))
2004 return 0;
2005 return type;
2006 }
2007
2008 /*
2009 * recursively change the type of the mountpoint.
2010 */
do_change_type(struct path * path,int flag)2011 static int do_change_type(struct path *path, int flag)
2012 {
2013 struct mount *m;
2014 struct mount *mnt = real_mount(path->mnt);
2015 int recurse = flag & MS_REC;
2016 int type;
2017 int err = 0;
2018
2019 if (path->dentry != path->mnt->mnt_root)
2020 return -EINVAL;
2021
2022 type = flags_to_propagation_type(flag);
2023 if (!type)
2024 return -EINVAL;
2025
2026 namespace_lock();
2027 if (type == MS_SHARED) {
2028 err = invent_group_ids(mnt, recurse);
2029 if (err)
2030 goto out_unlock;
2031 }
2032
2033 lock_mount_hash();
2034 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2035 change_mnt_propagation(m, type);
2036 unlock_mount_hash();
2037
2038 out_unlock:
2039 namespace_unlock();
2040 return err;
2041 }
2042
has_locked_children(struct mount * mnt,struct dentry * dentry)2043 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2044 {
2045 struct mount *child;
2046 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2047 if (!is_subdir(child->mnt_mountpoint, dentry))
2048 continue;
2049
2050 if (child->mnt.mnt_flags & MNT_LOCKED)
2051 return true;
2052 }
2053 return false;
2054 }
2055
2056 /*
2057 * do loopback mount.
2058 */
do_loopback(struct path * path,const char * old_name,int recurse)2059 static int do_loopback(struct path *path, const char *old_name,
2060 int recurse)
2061 {
2062 struct path old_path;
2063 struct mount *mnt = NULL, *old, *parent;
2064 struct mountpoint *mp;
2065 int err;
2066 if (!old_name || !*old_name)
2067 return -EINVAL;
2068 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2069 if (err)
2070 return err;
2071
2072 err = -EINVAL;
2073 if (mnt_ns_loop(old_path.dentry))
2074 goto out;
2075
2076 mp = lock_mount(path);
2077 err = PTR_ERR(mp);
2078 if (IS_ERR(mp))
2079 goto out;
2080
2081 old = real_mount(old_path.mnt);
2082 parent = real_mount(path->mnt);
2083
2084 err = -EINVAL;
2085 if (IS_MNT_UNBINDABLE(old))
2086 goto out2;
2087
2088 if (!check_mnt(parent))
2089 goto out2;
2090
2091 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2092 goto out2;
2093
2094 if (!recurse && has_locked_children(old, old_path.dentry))
2095 goto out2;
2096
2097 if (recurse)
2098 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2099 else
2100 mnt = clone_mnt(old, old_path.dentry, 0);
2101
2102 if (IS_ERR(mnt)) {
2103 err = PTR_ERR(mnt);
2104 goto out2;
2105 }
2106
2107 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2108
2109 err = graft_tree(mnt, parent, mp);
2110 if (err) {
2111 lock_mount_hash();
2112 umount_tree(mnt, UMOUNT_SYNC);
2113 unlock_mount_hash();
2114 }
2115 out2:
2116 unlock_mount(mp);
2117 out:
2118 path_put(&old_path);
2119 return err;
2120 }
2121
change_mount_flags(struct vfsmount * mnt,int ms_flags)2122 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2123 {
2124 int error = 0;
2125 int readonly_request = 0;
2126
2127 if (ms_flags & MS_RDONLY)
2128 readonly_request = 1;
2129 if (readonly_request == __mnt_is_readonly(mnt))
2130 return 0;
2131
2132 if (readonly_request)
2133 error = mnt_make_readonly(real_mount(mnt));
2134 else
2135 __mnt_unmake_readonly(real_mount(mnt));
2136 return error;
2137 }
2138
2139 /*
2140 * change filesystem flags. dir should be a physical root of filesystem.
2141 * If you've mounted a non-root directory somewhere and want to do remount
2142 * on it - tough luck.
2143 */
do_remount(struct path * path,int flags,int mnt_flags,void * data)2144 static int do_remount(struct path *path, int flags, int mnt_flags,
2145 void *data)
2146 {
2147 int err;
2148 struct super_block *sb = path->mnt->mnt_sb;
2149 struct mount *mnt = real_mount(path->mnt);
2150
2151 if (!check_mnt(mnt))
2152 return -EINVAL;
2153
2154 if (path->dentry != path->mnt->mnt_root)
2155 return -EINVAL;
2156
2157 /* Don't allow changing of locked mnt flags.
2158 *
2159 * No locks need to be held here while testing the various
2160 * MNT_LOCK flags because those flags can never be cleared
2161 * once they are set.
2162 */
2163 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2164 !(mnt_flags & MNT_READONLY)) {
2165 return -EPERM;
2166 }
2167 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2168 !(mnt_flags & MNT_NODEV)) {
2169 /* Was the nodev implicitly added in mount? */
2170 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2171 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2172 mnt_flags |= MNT_NODEV;
2173 } else {
2174 return -EPERM;
2175 }
2176 }
2177 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2178 !(mnt_flags & MNT_NOSUID)) {
2179 return -EPERM;
2180 }
2181 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2182 !(mnt_flags & MNT_NOEXEC)) {
2183 return -EPERM;
2184 }
2185 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2186 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2187 return -EPERM;
2188 }
2189
2190 err = security_sb_remount(sb, data);
2191 if (err)
2192 return err;
2193
2194 down_write(&sb->s_umount);
2195 if (flags & MS_BIND)
2196 err = change_mount_flags(path->mnt, flags);
2197 else if (!capable(CAP_SYS_ADMIN))
2198 err = -EPERM;
2199 else
2200 err = do_remount_sb(sb, flags, data, 0);
2201 if (!err) {
2202 lock_mount_hash();
2203 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2204 mnt->mnt.mnt_flags = mnt_flags;
2205 touch_mnt_namespace(mnt->mnt_ns);
2206 unlock_mount_hash();
2207 }
2208 up_write(&sb->s_umount);
2209 return err;
2210 }
2211
tree_contains_unbindable(struct mount * mnt)2212 static inline int tree_contains_unbindable(struct mount *mnt)
2213 {
2214 struct mount *p;
2215 for (p = mnt; p; p = next_mnt(p, mnt)) {
2216 if (IS_MNT_UNBINDABLE(p))
2217 return 1;
2218 }
2219 return 0;
2220 }
2221
do_move_mount(struct path * path,const char * old_name)2222 static int do_move_mount(struct path *path, const char *old_name)
2223 {
2224 struct path old_path, parent_path;
2225 struct mount *p;
2226 struct mount *old;
2227 struct mountpoint *mp;
2228 int err;
2229 if (!old_name || !*old_name)
2230 return -EINVAL;
2231 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2232 if (err)
2233 return err;
2234
2235 mp = lock_mount(path);
2236 err = PTR_ERR(mp);
2237 if (IS_ERR(mp))
2238 goto out;
2239
2240 old = real_mount(old_path.mnt);
2241 p = real_mount(path->mnt);
2242
2243 err = -EINVAL;
2244 if (!check_mnt(p) || !check_mnt(old))
2245 goto out1;
2246
2247 if (old->mnt.mnt_flags & MNT_LOCKED)
2248 goto out1;
2249
2250 err = -EINVAL;
2251 if (old_path.dentry != old_path.mnt->mnt_root)
2252 goto out1;
2253
2254 if (!mnt_has_parent(old))
2255 goto out1;
2256
2257 if (d_is_dir(path->dentry) !=
2258 d_is_dir(old_path.dentry))
2259 goto out1;
2260 /*
2261 * Don't move a mount residing in a shared parent.
2262 */
2263 if (IS_MNT_SHARED(old->mnt_parent))
2264 goto out1;
2265 /*
2266 * Don't move a mount tree containing unbindable mounts to a destination
2267 * mount which is shared.
2268 */
2269 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2270 goto out1;
2271 err = -ELOOP;
2272 for (; mnt_has_parent(p); p = p->mnt_parent)
2273 if (p == old)
2274 goto out1;
2275
2276 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2277 if (err)
2278 goto out1;
2279
2280 /* if the mount is moved, it should no longer be expire
2281 * automatically */
2282 list_del_init(&old->mnt_expire);
2283 out1:
2284 unlock_mount(mp);
2285 out:
2286 if (!err)
2287 path_put(&parent_path);
2288 path_put(&old_path);
2289 return err;
2290 }
2291
fs_set_subtype(struct vfsmount * mnt,const char * fstype)2292 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2293 {
2294 int err;
2295 const char *subtype = strchr(fstype, '.');
2296 if (subtype) {
2297 subtype++;
2298 err = -EINVAL;
2299 if (!subtype[0])
2300 goto err;
2301 } else
2302 subtype = "";
2303
2304 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2305 err = -ENOMEM;
2306 if (!mnt->mnt_sb->s_subtype)
2307 goto err;
2308 return mnt;
2309
2310 err:
2311 mntput(mnt);
2312 return ERR_PTR(err);
2313 }
2314
2315 /*
2316 * add a mount into a namespace's mount tree
2317 */
do_add_mount(struct mount * newmnt,struct path * path,int mnt_flags)2318 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2319 {
2320 struct mountpoint *mp;
2321 struct mount *parent;
2322 int err;
2323
2324 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2325
2326 mp = lock_mount(path);
2327 if (IS_ERR(mp))
2328 return PTR_ERR(mp);
2329
2330 parent = real_mount(path->mnt);
2331 err = -EINVAL;
2332 if (unlikely(!check_mnt(parent))) {
2333 /* that's acceptable only for automounts done in private ns */
2334 if (!(mnt_flags & MNT_SHRINKABLE))
2335 goto unlock;
2336 /* ... and for those we'd better have mountpoint still alive */
2337 if (!parent->mnt_ns)
2338 goto unlock;
2339 }
2340
2341 /* Refuse the same filesystem on the same mount point */
2342 err = -EBUSY;
2343 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2344 path->mnt->mnt_root == path->dentry)
2345 goto unlock;
2346
2347 err = -EINVAL;
2348 if (d_is_symlink(newmnt->mnt.mnt_root))
2349 goto unlock;
2350
2351 newmnt->mnt.mnt_flags = mnt_flags;
2352 err = graft_tree(newmnt, parent, mp);
2353
2354 unlock:
2355 unlock_mount(mp);
2356 return err;
2357 }
2358
2359 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2360
2361 /*
2362 * create a new mount for userspace and request it to be added into the
2363 * namespace's tree
2364 */
do_new_mount(struct path * path,const char * fstype,int flags,int mnt_flags,const char * name,void * data)2365 static int do_new_mount(struct path *path, const char *fstype, int flags,
2366 int mnt_flags, const char *name, void *data)
2367 {
2368 struct file_system_type *type;
2369 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2370 struct vfsmount *mnt;
2371 int err;
2372
2373 if (!fstype)
2374 return -EINVAL;
2375
2376 type = get_fs_type(fstype);
2377 if (!type)
2378 return -ENODEV;
2379
2380 if (user_ns != &init_user_ns) {
2381 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2382 put_filesystem(type);
2383 return -EPERM;
2384 }
2385 /* Only in special cases allow devices from mounts
2386 * created outside the initial user namespace.
2387 */
2388 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2389 flags |= MS_NODEV;
2390 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2391 }
2392 if (type->fs_flags & FS_USERNS_VISIBLE) {
2393 if (!fs_fully_visible(type, &mnt_flags)) {
2394 put_filesystem(type);
2395 return -EPERM;
2396 }
2397 }
2398 }
2399
2400 mnt = vfs_kern_mount(type, flags, name, data);
2401 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2402 !mnt->mnt_sb->s_subtype)
2403 mnt = fs_set_subtype(mnt, fstype);
2404
2405 put_filesystem(type);
2406 if (IS_ERR(mnt))
2407 return PTR_ERR(mnt);
2408
2409 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2410 if (err)
2411 mntput(mnt);
2412 return err;
2413 }
2414
finish_automount(struct vfsmount * m,struct path * path)2415 int finish_automount(struct vfsmount *m, struct path *path)
2416 {
2417 struct mount *mnt = real_mount(m);
2418 int err;
2419 /* The new mount record should have at least 2 refs to prevent it being
2420 * expired before we get a chance to add it
2421 */
2422 BUG_ON(mnt_get_count(mnt) < 2);
2423
2424 if (m->mnt_sb == path->mnt->mnt_sb &&
2425 m->mnt_root == path->dentry) {
2426 err = -ELOOP;
2427 goto fail;
2428 }
2429
2430 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2431 if (!err)
2432 return 0;
2433 fail:
2434 /* remove m from any expiration list it may be on */
2435 if (!list_empty(&mnt->mnt_expire)) {
2436 namespace_lock();
2437 list_del_init(&mnt->mnt_expire);
2438 namespace_unlock();
2439 }
2440 mntput(m);
2441 mntput(m);
2442 return err;
2443 }
2444
2445 /**
2446 * mnt_set_expiry - Put a mount on an expiration list
2447 * @mnt: The mount to list.
2448 * @expiry_list: The list to add the mount to.
2449 */
mnt_set_expiry(struct vfsmount * mnt,struct list_head * expiry_list)2450 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2451 {
2452 namespace_lock();
2453
2454 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2455
2456 namespace_unlock();
2457 }
2458 EXPORT_SYMBOL(mnt_set_expiry);
2459
2460 /*
2461 * process a list of expirable mountpoints with the intent of discarding any
2462 * mountpoints that aren't in use and haven't been touched since last we came
2463 * here
2464 */
mark_mounts_for_expiry(struct list_head * mounts)2465 void mark_mounts_for_expiry(struct list_head *mounts)
2466 {
2467 struct mount *mnt, *next;
2468 LIST_HEAD(graveyard);
2469
2470 if (list_empty(mounts))
2471 return;
2472
2473 namespace_lock();
2474 lock_mount_hash();
2475
2476 /* extract from the expiration list every vfsmount that matches the
2477 * following criteria:
2478 * - only referenced by its parent vfsmount
2479 * - still marked for expiry (marked on the last call here; marks are
2480 * cleared by mntput())
2481 */
2482 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2483 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2484 propagate_mount_busy(mnt, 1))
2485 continue;
2486 list_move(&mnt->mnt_expire, &graveyard);
2487 }
2488 while (!list_empty(&graveyard)) {
2489 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2490 touch_mnt_namespace(mnt->mnt_ns);
2491 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2492 }
2493 unlock_mount_hash();
2494 namespace_unlock();
2495 }
2496
2497 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2498
2499 /*
2500 * Ripoff of 'select_parent()'
2501 *
2502 * search the list of submounts for a given mountpoint, and move any
2503 * shrinkable submounts to the 'graveyard' list.
2504 */
select_submounts(struct mount * parent,struct list_head * graveyard)2505 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2506 {
2507 struct mount *this_parent = parent;
2508 struct list_head *next;
2509 int found = 0;
2510
2511 repeat:
2512 next = this_parent->mnt_mounts.next;
2513 resume:
2514 while (next != &this_parent->mnt_mounts) {
2515 struct list_head *tmp = next;
2516 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2517
2518 next = tmp->next;
2519 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2520 continue;
2521 /*
2522 * Descend a level if the d_mounts list is non-empty.
2523 */
2524 if (!list_empty(&mnt->mnt_mounts)) {
2525 this_parent = mnt;
2526 goto repeat;
2527 }
2528
2529 if (!propagate_mount_busy(mnt, 1)) {
2530 list_move_tail(&mnt->mnt_expire, graveyard);
2531 found++;
2532 }
2533 }
2534 /*
2535 * All done at this level ... ascend and resume the search
2536 */
2537 if (this_parent != parent) {
2538 next = this_parent->mnt_child.next;
2539 this_parent = this_parent->mnt_parent;
2540 goto resume;
2541 }
2542 return found;
2543 }
2544
2545 /*
2546 * process a list of expirable mountpoints with the intent of discarding any
2547 * submounts of a specific parent mountpoint
2548 *
2549 * mount_lock must be held for write
2550 */
shrink_submounts(struct mount * mnt)2551 static void shrink_submounts(struct mount *mnt)
2552 {
2553 LIST_HEAD(graveyard);
2554 struct mount *m;
2555
2556 /* extract submounts of 'mountpoint' from the expiration list */
2557 while (select_submounts(mnt, &graveyard)) {
2558 while (!list_empty(&graveyard)) {
2559 m = list_first_entry(&graveyard, struct mount,
2560 mnt_expire);
2561 touch_mnt_namespace(m->mnt_ns);
2562 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2563 }
2564 }
2565 }
2566
2567 /*
2568 * Some copy_from_user() implementations do not return the exact number of
2569 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2570 * Note that this function differs from copy_from_user() in that it will oops
2571 * on bad values of `to', rather than returning a short copy.
2572 */
exact_copy_from_user(void * to,const void __user * from,unsigned long n)2573 static long exact_copy_from_user(void *to, const void __user * from,
2574 unsigned long n)
2575 {
2576 char *t = to;
2577 const char __user *f = from;
2578 char c;
2579
2580 if (!access_ok(VERIFY_READ, from, n))
2581 return n;
2582
2583 while (n) {
2584 if (__get_user(c, f)) {
2585 memset(t, 0, n);
2586 break;
2587 }
2588 *t++ = c;
2589 f++;
2590 n--;
2591 }
2592 return n;
2593 }
2594
copy_mount_options(const void __user * data,unsigned long * where)2595 int copy_mount_options(const void __user * data, unsigned long *where)
2596 {
2597 int i;
2598 unsigned long page;
2599 unsigned long size;
2600
2601 *where = 0;
2602 if (!data)
2603 return 0;
2604
2605 if (!(page = __get_free_page(GFP_KERNEL)))
2606 return -ENOMEM;
2607
2608 /* We only care that *some* data at the address the user
2609 * gave us is valid. Just in case, we'll zero
2610 * the remainder of the page.
2611 */
2612 /* copy_from_user cannot cross TASK_SIZE ! */
2613 size = TASK_SIZE - (unsigned long)data;
2614 if (size > PAGE_SIZE)
2615 size = PAGE_SIZE;
2616
2617 i = size - exact_copy_from_user((void *)page, data, size);
2618 if (!i) {
2619 free_page(page);
2620 return -EFAULT;
2621 }
2622 if (i != PAGE_SIZE)
2623 memset((char *)page + i, 0, PAGE_SIZE - i);
2624 *where = page;
2625 return 0;
2626 }
2627
copy_mount_string(const void __user * data)2628 char *copy_mount_string(const void __user *data)
2629 {
2630 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2631 }
2632
2633 /*
2634 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2635 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2636 *
2637 * data is a (void *) that can point to any structure up to
2638 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2639 * information (or be NULL).
2640 *
2641 * Pre-0.97 versions of mount() didn't have a flags word.
2642 * When the flags word was introduced its top half was required
2643 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2644 * Therefore, if this magic number is present, it carries no information
2645 * and must be discarded.
2646 */
do_mount(const char * dev_name,const char __user * dir_name,const char * type_page,unsigned long flags,void * data_page)2647 long do_mount(const char *dev_name, const char __user *dir_name,
2648 const char *type_page, unsigned long flags, void *data_page)
2649 {
2650 struct path path;
2651 int retval = 0;
2652 int mnt_flags = 0;
2653
2654 /* Discard magic */
2655 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2656 flags &= ~MS_MGC_MSK;
2657
2658 /* Basic sanity checks */
2659 if (data_page)
2660 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2661
2662 /* ... and get the mountpoint */
2663 retval = user_path(dir_name, &path);
2664 if (retval)
2665 return retval;
2666
2667 retval = security_sb_mount(dev_name, &path,
2668 type_page, flags, data_page);
2669 if (!retval && !may_mount())
2670 retval = -EPERM;
2671 if (retval)
2672 goto dput_out;
2673
2674 /* Default to relatime unless overriden */
2675 if (!(flags & MS_NOATIME))
2676 mnt_flags |= MNT_RELATIME;
2677
2678 /* Separate the per-mountpoint flags */
2679 if (flags & MS_NOSUID)
2680 mnt_flags |= MNT_NOSUID;
2681 if (flags & MS_NODEV)
2682 mnt_flags |= MNT_NODEV;
2683 if (flags & MS_NOEXEC)
2684 mnt_flags |= MNT_NOEXEC;
2685 if (flags & MS_NOATIME)
2686 mnt_flags |= MNT_NOATIME;
2687 if (flags & MS_NODIRATIME)
2688 mnt_flags |= MNT_NODIRATIME;
2689 if (flags & MS_STRICTATIME)
2690 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2691 if (flags & MS_RDONLY)
2692 mnt_flags |= MNT_READONLY;
2693
2694 /* The default atime for remount is preservation */
2695 if ((flags & MS_REMOUNT) &&
2696 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2697 MS_STRICTATIME)) == 0)) {
2698 mnt_flags &= ~MNT_ATIME_MASK;
2699 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2700 }
2701
2702 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2703 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2704 MS_STRICTATIME);
2705
2706 if (flags & MS_REMOUNT)
2707 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2708 data_page);
2709 else if (flags & MS_BIND)
2710 retval = do_loopback(&path, dev_name, flags & MS_REC);
2711 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2712 retval = do_change_type(&path, flags);
2713 else if (flags & MS_MOVE)
2714 retval = do_move_mount(&path, dev_name);
2715 else
2716 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2717 dev_name, data_page);
2718 dput_out:
2719 path_put(&path);
2720 return retval;
2721 }
2722
free_mnt_ns(struct mnt_namespace * ns)2723 static void free_mnt_ns(struct mnt_namespace *ns)
2724 {
2725 ns_free_inum(&ns->ns);
2726 put_user_ns(ns->user_ns);
2727 kfree(ns);
2728 }
2729
2730 /*
2731 * Assign a sequence number so we can detect when we attempt to bind
2732 * mount a reference to an older mount namespace into the current
2733 * mount namespace, preventing reference counting loops. A 64bit
2734 * number incrementing at 10Ghz will take 12,427 years to wrap which
2735 * is effectively never, so we can ignore the possibility.
2736 */
2737 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2738
alloc_mnt_ns(struct user_namespace * user_ns)2739 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2740 {
2741 struct mnt_namespace *new_ns;
2742 int ret;
2743
2744 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2745 if (!new_ns)
2746 return ERR_PTR(-ENOMEM);
2747 ret = ns_alloc_inum(&new_ns->ns);
2748 if (ret) {
2749 kfree(new_ns);
2750 return ERR_PTR(ret);
2751 }
2752 new_ns->ns.ops = &mntns_operations;
2753 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2754 atomic_set(&new_ns->count, 1);
2755 new_ns->root = NULL;
2756 INIT_LIST_HEAD(&new_ns->list);
2757 init_waitqueue_head(&new_ns->poll);
2758 new_ns->event = 0;
2759 new_ns->user_ns = get_user_ns(user_ns);
2760 return new_ns;
2761 }
2762
copy_mnt_ns(unsigned long flags,struct mnt_namespace * ns,struct user_namespace * user_ns,struct fs_struct * new_fs)2763 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2764 struct user_namespace *user_ns, struct fs_struct *new_fs)
2765 {
2766 struct mnt_namespace *new_ns;
2767 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2768 struct mount *p, *q;
2769 struct mount *old;
2770 struct mount *new;
2771 int copy_flags;
2772
2773 BUG_ON(!ns);
2774
2775 if (likely(!(flags & CLONE_NEWNS))) {
2776 get_mnt_ns(ns);
2777 return ns;
2778 }
2779
2780 old = ns->root;
2781
2782 new_ns = alloc_mnt_ns(user_ns);
2783 if (IS_ERR(new_ns))
2784 return new_ns;
2785
2786 namespace_lock();
2787 /* First pass: copy the tree topology */
2788 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2789 if (user_ns != ns->user_ns)
2790 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2791 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2792 if (IS_ERR(new)) {
2793 namespace_unlock();
2794 free_mnt_ns(new_ns);
2795 return ERR_CAST(new);
2796 }
2797 new_ns->root = new;
2798 list_add_tail(&new_ns->list, &new->mnt_list);
2799
2800 /*
2801 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2802 * as belonging to new namespace. We have already acquired a private
2803 * fs_struct, so tsk->fs->lock is not needed.
2804 */
2805 p = old;
2806 q = new;
2807 while (p) {
2808 q->mnt_ns = new_ns;
2809 if (new_fs) {
2810 if (&p->mnt == new_fs->root.mnt) {
2811 new_fs->root.mnt = mntget(&q->mnt);
2812 rootmnt = &p->mnt;
2813 }
2814 if (&p->mnt == new_fs->pwd.mnt) {
2815 new_fs->pwd.mnt = mntget(&q->mnt);
2816 pwdmnt = &p->mnt;
2817 }
2818 }
2819 p = next_mnt(p, old);
2820 q = next_mnt(q, new);
2821 if (!q)
2822 break;
2823 while (p->mnt.mnt_root != q->mnt.mnt_root)
2824 p = next_mnt(p, old);
2825 }
2826 namespace_unlock();
2827
2828 if (rootmnt)
2829 mntput(rootmnt);
2830 if (pwdmnt)
2831 mntput(pwdmnt);
2832
2833 return new_ns;
2834 }
2835
2836 /**
2837 * create_mnt_ns - creates a private namespace and adds a root filesystem
2838 * @mnt: pointer to the new root filesystem mountpoint
2839 */
create_mnt_ns(struct vfsmount * m)2840 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2841 {
2842 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2843 if (!IS_ERR(new_ns)) {
2844 struct mount *mnt = real_mount(m);
2845 mnt->mnt_ns = new_ns;
2846 new_ns->root = mnt;
2847 list_add(&mnt->mnt_list, &new_ns->list);
2848 } else {
2849 mntput(m);
2850 }
2851 return new_ns;
2852 }
2853
mount_subtree(struct vfsmount * mnt,const char * name)2854 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2855 {
2856 struct mnt_namespace *ns;
2857 struct super_block *s;
2858 struct path path;
2859 int err;
2860
2861 ns = create_mnt_ns(mnt);
2862 if (IS_ERR(ns))
2863 return ERR_CAST(ns);
2864
2865 err = vfs_path_lookup(mnt->mnt_root, mnt,
2866 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2867
2868 put_mnt_ns(ns);
2869
2870 if (err)
2871 return ERR_PTR(err);
2872
2873 /* trade a vfsmount reference for active sb one */
2874 s = path.mnt->mnt_sb;
2875 atomic_inc(&s->s_active);
2876 mntput(path.mnt);
2877 /* lock the sucker */
2878 down_write(&s->s_umount);
2879 /* ... and return the root of (sub)tree on it */
2880 return path.dentry;
2881 }
2882 EXPORT_SYMBOL(mount_subtree);
2883
SYSCALL_DEFINE5(mount,char __user *,dev_name,char __user *,dir_name,char __user *,type,unsigned long,flags,void __user *,data)2884 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2885 char __user *, type, unsigned long, flags, void __user *, data)
2886 {
2887 int ret;
2888 char *kernel_type;
2889 char *kernel_dev;
2890 unsigned long data_page;
2891
2892 kernel_type = copy_mount_string(type);
2893 ret = PTR_ERR(kernel_type);
2894 if (IS_ERR(kernel_type))
2895 goto out_type;
2896
2897 kernel_dev = copy_mount_string(dev_name);
2898 ret = PTR_ERR(kernel_dev);
2899 if (IS_ERR(kernel_dev))
2900 goto out_dev;
2901
2902 ret = copy_mount_options(data, &data_page);
2903 if (ret < 0)
2904 goto out_data;
2905
2906 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2907 (void *) data_page);
2908
2909 free_page(data_page);
2910 out_data:
2911 kfree(kernel_dev);
2912 out_dev:
2913 kfree(kernel_type);
2914 out_type:
2915 return ret;
2916 }
2917
2918 /*
2919 * Return true if path is reachable from root
2920 *
2921 * namespace_sem or mount_lock is held
2922 */
is_path_reachable(struct mount * mnt,struct dentry * dentry,const struct path * root)2923 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2924 const struct path *root)
2925 {
2926 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2927 dentry = mnt->mnt_mountpoint;
2928 mnt = mnt->mnt_parent;
2929 }
2930 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2931 }
2932
path_is_under(struct path * path1,struct path * path2)2933 int path_is_under(struct path *path1, struct path *path2)
2934 {
2935 int res;
2936 read_seqlock_excl(&mount_lock);
2937 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2938 read_sequnlock_excl(&mount_lock);
2939 return res;
2940 }
2941 EXPORT_SYMBOL(path_is_under);
2942
2943 /*
2944 * pivot_root Semantics:
2945 * Moves the root file system of the current process to the directory put_old,
2946 * makes new_root as the new root file system of the current process, and sets
2947 * root/cwd of all processes which had them on the current root to new_root.
2948 *
2949 * Restrictions:
2950 * The new_root and put_old must be directories, and must not be on the
2951 * same file system as the current process root. The put_old must be
2952 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2953 * pointed to by put_old must yield the same directory as new_root. No other
2954 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2955 *
2956 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2957 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2958 * in this situation.
2959 *
2960 * Notes:
2961 * - we don't move root/cwd if they are not at the root (reason: if something
2962 * cared enough to change them, it's probably wrong to force them elsewhere)
2963 * - it's okay to pick a root that isn't the root of a file system, e.g.
2964 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2965 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2966 * first.
2967 */
SYSCALL_DEFINE2(pivot_root,const char __user *,new_root,const char __user *,put_old)2968 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2969 const char __user *, put_old)
2970 {
2971 struct path new, old, parent_path, root_parent, root;
2972 struct mount *new_mnt, *root_mnt, *old_mnt;
2973 struct mountpoint *old_mp, *root_mp;
2974 int error;
2975
2976 if (!may_mount())
2977 return -EPERM;
2978
2979 error = user_path_dir(new_root, &new);
2980 if (error)
2981 goto out0;
2982
2983 error = user_path_dir(put_old, &old);
2984 if (error)
2985 goto out1;
2986
2987 error = security_sb_pivotroot(&old, &new);
2988 if (error)
2989 goto out2;
2990
2991 get_fs_root(current->fs, &root);
2992 old_mp = lock_mount(&old);
2993 error = PTR_ERR(old_mp);
2994 if (IS_ERR(old_mp))
2995 goto out3;
2996
2997 error = -EINVAL;
2998 new_mnt = real_mount(new.mnt);
2999 root_mnt = real_mount(root.mnt);
3000 old_mnt = real_mount(old.mnt);
3001 if (IS_MNT_SHARED(old_mnt) ||
3002 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3003 IS_MNT_SHARED(root_mnt->mnt_parent))
3004 goto out4;
3005 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3006 goto out4;
3007 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3008 goto out4;
3009 error = -ENOENT;
3010 if (d_unlinked(new.dentry))
3011 goto out4;
3012 error = -EBUSY;
3013 if (new_mnt == root_mnt || old_mnt == root_mnt)
3014 goto out4; /* loop, on the same file system */
3015 error = -EINVAL;
3016 if (root.mnt->mnt_root != root.dentry)
3017 goto out4; /* not a mountpoint */
3018 if (!mnt_has_parent(root_mnt))
3019 goto out4; /* not attached */
3020 root_mp = root_mnt->mnt_mp;
3021 if (new.mnt->mnt_root != new.dentry)
3022 goto out4; /* not a mountpoint */
3023 if (!mnt_has_parent(new_mnt))
3024 goto out4; /* not attached */
3025 /* make sure we can reach put_old from new_root */
3026 if (!is_path_reachable(old_mnt, old.dentry, &new))
3027 goto out4;
3028 /* make certain new is below the root */
3029 if (!is_path_reachable(new_mnt, new.dentry, &root))
3030 goto out4;
3031 root_mp->m_count++; /* pin it so it won't go away */
3032 lock_mount_hash();
3033 detach_mnt(new_mnt, &parent_path);
3034 detach_mnt(root_mnt, &root_parent);
3035 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3036 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3037 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3038 }
3039 /* mount old root on put_old */
3040 attach_mnt(root_mnt, old_mnt, old_mp);
3041 /* mount new_root on / */
3042 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3043 touch_mnt_namespace(current->nsproxy->mnt_ns);
3044 /* A moved mount should not expire automatically */
3045 list_del_init(&new_mnt->mnt_expire);
3046 unlock_mount_hash();
3047 chroot_fs_refs(&root, &new);
3048 put_mountpoint(root_mp);
3049 error = 0;
3050 out4:
3051 unlock_mount(old_mp);
3052 if (!error) {
3053 path_put(&root_parent);
3054 path_put(&parent_path);
3055 }
3056 out3:
3057 path_put(&root);
3058 out2:
3059 path_put(&old);
3060 out1:
3061 path_put(&new);
3062 out0:
3063 return error;
3064 }
3065
init_mount_tree(void)3066 static void __init init_mount_tree(void)
3067 {
3068 struct vfsmount *mnt;
3069 struct mnt_namespace *ns;
3070 struct path root;
3071 struct file_system_type *type;
3072
3073 type = get_fs_type("rootfs");
3074 if (!type)
3075 panic("Can't find rootfs type");
3076 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3077 put_filesystem(type);
3078 if (IS_ERR(mnt))
3079 panic("Can't create rootfs");
3080
3081 ns = create_mnt_ns(mnt);
3082 if (IS_ERR(ns))
3083 panic("Can't allocate initial namespace");
3084
3085 init_task.nsproxy->mnt_ns = ns;
3086 get_mnt_ns(ns);
3087
3088 root.mnt = mnt;
3089 root.dentry = mnt->mnt_root;
3090 mnt->mnt_flags |= MNT_LOCKED;
3091
3092 set_fs_pwd(current->fs, &root);
3093 set_fs_root(current->fs, &root);
3094 }
3095
mnt_init(void)3096 void __init mnt_init(void)
3097 {
3098 unsigned u;
3099 int err;
3100
3101 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3102 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3103
3104 mount_hashtable = alloc_large_system_hash("Mount-cache",
3105 sizeof(struct hlist_head),
3106 mhash_entries, 19,
3107 0,
3108 &m_hash_shift, &m_hash_mask, 0, 0);
3109 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3110 sizeof(struct hlist_head),
3111 mphash_entries, 19,
3112 0,
3113 &mp_hash_shift, &mp_hash_mask, 0, 0);
3114
3115 if (!mount_hashtable || !mountpoint_hashtable)
3116 panic("Failed to allocate mount hash table\n");
3117
3118 for (u = 0; u <= m_hash_mask; u++)
3119 INIT_HLIST_HEAD(&mount_hashtable[u]);
3120 for (u = 0; u <= mp_hash_mask; u++)
3121 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3122
3123 kernfs_init();
3124
3125 err = sysfs_init();
3126 if (err)
3127 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3128 __func__, err);
3129 fs_kobj = kobject_create_and_add("fs", NULL);
3130 if (!fs_kobj)
3131 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3132 init_rootfs();
3133 init_mount_tree();
3134 }
3135
put_mnt_ns(struct mnt_namespace * ns)3136 void put_mnt_ns(struct mnt_namespace *ns)
3137 {
3138 if (!atomic_dec_and_test(&ns->count))
3139 return;
3140 drop_collected_mounts(&ns->root->mnt);
3141 free_mnt_ns(ns);
3142 }
3143
kern_mount_data(struct file_system_type * type,void * data)3144 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3145 {
3146 struct vfsmount *mnt;
3147 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3148 if (!IS_ERR(mnt)) {
3149 /*
3150 * it is a longterm mount, don't release mnt until
3151 * we unmount before file sys is unregistered
3152 */
3153 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3154 }
3155 return mnt;
3156 }
3157 EXPORT_SYMBOL_GPL(kern_mount_data);
3158
kern_unmount(struct vfsmount * mnt)3159 void kern_unmount(struct vfsmount *mnt)
3160 {
3161 /* release long term mount so mount point can be released */
3162 if (!IS_ERR_OR_NULL(mnt)) {
3163 real_mount(mnt)->mnt_ns = NULL;
3164 synchronize_rcu(); /* yecchhh... */
3165 mntput(mnt);
3166 }
3167 }
3168 EXPORT_SYMBOL(kern_unmount);
3169
our_mnt(struct vfsmount * mnt)3170 bool our_mnt(struct vfsmount *mnt)
3171 {
3172 return check_mnt(real_mount(mnt));
3173 }
3174
current_chrooted(void)3175 bool current_chrooted(void)
3176 {
3177 /* Does the current process have a non-standard root */
3178 struct path ns_root;
3179 struct path fs_root;
3180 bool chrooted;
3181
3182 /* Find the namespace root */
3183 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3184 ns_root.dentry = ns_root.mnt->mnt_root;
3185 path_get(&ns_root);
3186 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3187 ;
3188
3189 get_fs_root(current->fs, &fs_root);
3190
3191 chrooted = !path_equal(&fs_root, &ns_root);
3192
3193 path_put(&fs_root);
3194 path_put(&ns_root);
3195
3196 return chrooted;
3197 }
3198
fs_fully_visible(struct file_system_type * type,int * new_mnt_flags)3199 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3200 {
3201 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3202 int new_flags = *new_mnt_flags;
3203 struct mount *mnt;
3204 bool visible = false;
3205
3206 if (unlikely(!ns))
3207 return false;
3208
3209 down_read(&namespace_sem);
3210 list_for_each_entry(mnt, &ns->list, mnt_list) {
3211 struct mount *child;
3212 if (mnt->mnt.mnt_sb->s_type != type)
3213 continue;
3214
3215 /* This mount is not fully visible if it's root directory
3216 * is not the root directory of the filesystem.
3217 */
3218 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3219 continue;
3220
3221 /* Verify the mount flags are equal to or more permissive
3222 * than the proposed new mount.
3223 */
3224 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
3225 !(new_flags & MNT_READONLY))
3226 continue;
3227 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
3228 !(new_flags & MNT_NODEV))
3229 continue;
3230 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
3231 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3232 continue;
3233
3234 /* This mount is not fully visible if there are any
3235 * locked child mounts that cover anything except for
3236 * empty directories.
3237 */
3238 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3239 struct inode *inode = child->mnt_mountpoint->d_inode;
3240 /* Only worry about locked mounts */
3241 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3242 continue;
3243 /* Is the directory permanetly empty? */
3244 if (!is_empty_dir_inode(inode))
3245 goto next;
3246 }
3247 /* Preserve the locked attributes */
3248 *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
3249 MNT_LOCK_NODEV | \
3250 MNT_LOCK_ATIME);
3251 visible = true;
3252 goto found;
3253 next: ;
3254 }
3255 found:
3256 up_read(&namespace_sem);
3257 return visible;
3258 }
3259
mntns_get(struct task_struct * task)3260 static struct ns_common *mntns_get(struct task_struct *task)
3261 {
3262 struct ns_common *ns = NULL;
3263 struct nsproxy *nsproxy;
3264
3265 task_lock(task);
3266 nsproxy = task->nsproxy;
3267 if (nsproxy) {
3268 ns = &nsproxy->mnt_ns->ns;
3269 get_mnt_ns(to_mnt_ns(ns));
3270 }
3271 task_unlock(task);
3272
3273 return ns;
3274 }
3275
mntns_put(struct ns_common * ns)3276 static void mntns_put(struct ns_common *ns)
3277 {
3278 put_mnt_ns(to_mnt_ns(ns));
3279 }
3280
mntns_install(struct nsproxy * nsproxy,struct ns_common * ns)3281 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3282 {
3283 struct fs_struct *fs = current->fs;
3284 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3285 struct path root;
3286
3287 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3288 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3289 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3290 return -EPERM;
3291
3292 if (fs->users != 1)
3293 return -EINVAL;
3294
3295 get_mnt_ns(mnt_ns);
3296 put_mnt_ns(nsproxy->mnt_ns);
3297 nsproxy->mnt_ns = mnt_ns;
3298
3299 /* Find the root */
3300 root.mnt = &mnt_ns->root->mnt;
3301 root.dentry = mnt_ns->root->mnt.mnt_root;
3302 path_get(&root);
3303 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3304 ;
3305
3306 /* Update the pwd and root */
3307 set_fs_pwd(fs, &root);
3308 set_fs_root(fs, &root);
3309
3310 path_put(&root);
3311 return 0;
3312 }
3313
3314 const struct proc_ns_operations mntns_operations = {
3315 .name = "mnt",
3316 .type = CLONE_NEWNS,
3317 .get = mntns_get,
3318 .put = mntns_put,
3319 .install = mntns_install,
3320 };
3321