]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/namespace.c
Merge remote-tracking branches 'regulator/fix/as3722', 'regulator/fix/ltc3589' and...
[mirror_ubuntu-bionic-kernel.git] / fs / namespace.c
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/acct.h> /* acct_auto_close_mnt */
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
27 #include "pnode.h"
28 #include "internal.h"
29
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
34
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
37 {
38 if (!str)
39 return 0;
40 mhash_entries = simple_strtoul(str, &str, 0);
41 return 1;
42 }
43 __setup("mhash_entries=", set_mhash_entries);
44
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
47 {
48 if (!str)
49 return 0;
50 mphash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52 }
53 __setup("mphash_entries=", set_mphash_entries);
54
55 static u64 event;
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
61
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
66
67 /* /sys/fs */
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
70
71 /*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
80
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
82 {
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
87 }
88
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
90 {
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
94 }
95
96 /*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
100 static int mnt_alloc_id(struct mount *mnt)
101 {
102 int res;
103
104 retry:
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
108 if (!res)
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
111 if (res == -EAGAIN)
112 goto retry;
113
114 return res;
115 }
116
117 static void mnt_free_id(struct mount *mnt)
118 {
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
123 mnt_id_start = id;
124 spin_unlock(&mnt_id_lock);
125 }
126
127 /*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
132 static int mnt_alloc_group_id(struct mount *mnt)
133 {
134 int res;
135
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137 return -ENOMEM;
138
139 res = ida_get_new_above(&mnt_group_ida,
140 mnt_group_start,
141 &mnt->mnt_group_id);
142 if (!res)
143 mnt_group_start = mnt->mnt_group_id + 1;
144
145 return res;
146 }
147
148 /*
149 * Release a peer group ID
150 */
151 void mnt_release_group_id(struct mount *mnt)
152 {
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
158 }
159
160 /*
161 * vfsmount lock must be held for read
162 */
163 static inline void mnt_add_count(struct mount *mnt, int n)
164 {
165 #ifdef CONFIG_SMP
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
167 #else
168 preempt_disable();
169 mnt->mnt_count += n;
170 preempt_enable();
171 #endif
172 }
173
174 /*
175 * vfsmount lock must be held for write
176 */
177 unsigned int mnt_get_count(struct mount *mnt)
178 {
179 #ifdef CONFIG_SMP
180 unsigned int count = 0;
181 int cpu;
182
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
185 }
186
187 return count;
188 #else
189 return mnt->mnt_count;
190 #endif
191 }
192
193 static struct mount *alloc_vfsmnt(const char *name)
194 {
195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
196 if (mnt) {
197 int err;
198
199 err = mnt_alloc_id(mnt);
200 if (err)
201 goto out_free_cache;
202
203 if (name) {
204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
206 goto out_free_id;
207 }
208
209 #ifdef CONFIG_SMP
210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
211 if (!mnt->mnt_pcp)
212 goto out_free_devname;
213
214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
215 #else
216 mnt->mnt_count = 1;
217 mnt->mnt_writers = 0;
218 #endif
219
220 INIT_HLIST_NODE(&mnt->mnt_hash);
221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
228 #ifdef CONFIG_FSNOTIFY
229 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
230 #endif
231 }
232 return mnt;
233
234 #ifdef CONFIG_SMP
235 out_free_devname:
236 kfree(mnt->mnt_devname);
237 #endif
238 out_free_id:
239 mnt_free_id(mnt);
240 out_free_cache:
241 kmem_cache_free(mnt_cache, mnt);
242 return NULL;
243 }
244
245 /*
246 * Most r/o checks on a fs are for operations that take
247 * discrete amounts of time, like a write() or unlink().
248 * We must keep track of when those operations start
249 * (for permission checks) and when they end, so that
250 * we can determine when writes are able to occur to
251 * a filesystem.
252 */
253 /*
254 * __mnt_is_readonly: check whether a mount is read-only
255 * @mnt: the mount to check for its write status
256 *
257 * This shouldn't be used directly ouside of the VFS.
258 * It does not guarantee that the filesystem will stay
259 * r/w, just that it is right *now*. This can not and
260 * should not be used in place of IS_RDONLY(inode).
261 * mnt_want/drop_write() will _keep_ the filesystem
262 * r/w.
263 */
264 int __mnt_is_readonly(struct vfsmount *mnt)
265 {
266 if (mnt->mnt_flags & MNT_READONLY)
267 return 1;
268 if (mnt->mnt_sb->s_flags & MS_RDONLY)
269 return 1;
270 return 0;
271 }
272 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
273
274 static inline void mnt_inc_writers(struct mount *mnt)
275 {
276 #ifdef CONFIG_SMP
277 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
278 #else
279 mnt->mnt_writers++;
280 #endif
281 }
282
283 static inline void mnt_dec_writers(struct mount *mnt)
284 {
285 #ifdef CONFIG_SMP
286 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
287 #else
288 mnt->mnt_writers--;
289 #endif
290 }
291
292 static unsigned int mnt_get_writers(struct mount *mnt)
293 {
294 #ifdef CONFIG_SMP
295 unsigned int count = 0;
296 int cpu;
297
298 for_each_possible_cpu(cpu) {
299 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
300 }
301
302 return count;
303 #else
304 return mnt->mnt_writers;
305 #endif
306 }
307
308 static int mnt_is_readonly(struct vfsmount *mnt)
309 {
310 if (mnt->mnt_sb->s_readonly_remount)
311 return 1;
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
313 smp_rmb();
314 return __mnt_is_readonly(mnt);
315 }
316
317 /*
318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
322 */
323 /**
324 * __mnt_want_write - get write access to a mount without freeze protection
325 * @m: the mount on which to take a write
326 *
327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
332 */
333 int __mnt_want_write(struct vfsmount *m)
334 {
335 struct mount *mnt = real_mount(m);
336 int ret = 0;
337
338 preempt_disable();
339 mnt_inc_writers(mnt);
340 /*
341 * The store to mnt_inc_writers must be visible before we pass
342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
344 */
345 smp_mb();
346 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
347 cpu_relax();
348 /*
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
352 */
353 smp_rmb();
354 if (mnt_is_readonly(m)) {
355 mnt_dec_writers(mnt);
356 ret = -EROFS;
357 }
358 preempt_enable();
359
360 return ret;
361 }
362
363 /**
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
366 *
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
371 */
372 int mnt_want_write(struct vfsmount *m)
373 {
374 int ret;
375
376 sb_start_write(m->mnt_sb);
377 ret = __mnt_want_write(m);
378 if (ret)
379 sb_end_write(m->mnt_sb);
380 return ret;
381 }
382 EXPORT_SYMBOL_GPL(mnt_want_write);
383
384 /**
385 * mnt_clone_write - get write access to a mount
386 * @mnt: the mount on which to take a write
387 *
388 * This is effectively like mnt_want_write, except
389 * it must only be used to take an extra write reference
390 * on a mountpoint that we already know has a write reference
391 * on it. This allows some optimisation.
392 *
393 * After finished, mnt_drop_write must be called as usual to
394 * drop the reference.
395 */
396 int mnt_clone_write(struct vfsmount *mnt)
397 {
398 /* superblock may be r/o */
399 if (__mnt_is_readonly(mnt))
400 return -EROFS;
401 preempt_disable();
402 mnt_inc_writers(real_mount(mnt));
403 preempt_enable();
404 return 0;
405 }
406 EXPORT_SYMBOL_GPL(mnt_clone_write);
407
408 /**
409 * __mnt_want_write_file - get write access to a file's mount
410 * @file: the file who's mount on which to take a write
411 *
412 * This is like __mnt_want_write, but it takes a file and can
413 * do some optimisations if the file is open for write already
414 */
415 int __mnt_want_write_file(struct file *file)
416 {
417 if (!(file->f_mode & FMODE_WRITER))
418 return __mnt_want_write(file->f_path.mnt);
419 else
420 return mnt_clone_write(file->f_path.mnt);
421 }
422
423 /**
424 * mnt_want_write_file - get write access to a file's mount
425 * @file: the file who's mount on which to take a write
426 *
427 * This is like mnt_want_write, but it takes a file and can
428 * do some optimisations if the file is open for write already
429 */
430 int mnt_want_write_file(struct file *file)
431 {
432 int ret;
433
434 sb_start_write(file->f_path.mnt->mnt_sb);
435 ret = __mnt_want_write_file(file);
436 if (ret)
437 sb_end_write(file->f_path.mnt->mnt_sb);
438 return ret;
439 }
440 EXPORT_SYMBOL_GPL(mnt_want_write_file);
441
442 /**
443 * __mnt_drop_write - give up write access to a mount
444 * @mnt: the mount on which to give up write access
445 *
446 * Tells the low-level filesystem that we are done
447 * performing writes to it. Must be matched with
448 * __mnt_want_write() call above.
449 */
450 void __mnt_drop_write(struct vfsmount *mnt)
451 {
452 preempt_disable();
453 mnt_dec_writers(real_mount(mnt));
454 preempt_enable();
455 }
456
457 /**
458 * mnt_drop_write - give up write access to a mount
459 * @mnt: the mount on which to give up write access
460 *
461 * Tells the low-level filesystem that we are done performing writes to it and
462 * also allows filesystem to be frozen again. Must be matched with
463 * mnt_want_write() call above.
464 */
465 void mnt_drop_write(struct vfsmount *mnt)
466 {
467 __mnt_drop_write(mnt);
468 sb_end_write(mnt->mnt_sb);
469 }
470 EXPORT_SYMBOL_GPL(mnt_drop_write);
471
472 void __mnt_drop_write_file(struct file *file)
473 {
474 __mnt_drop_write(file->f_path.mnt);
475 }
476
477 void mnt_drop_write_file(struct file *file)
478 {
479 mnt_drop_write(file->f_path.mnt);
480 }
481 EXPORT_SYMBOL(mnt_drop_write_file);
482
483 static int mnt_make_readonly(struct mount *mnt)
484 {
485 int ret = 0;
486
487 lock_mount_hash();
488 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
489 /*
490 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
491 * should be visible before we do.
492 */
493 smp_mb();
494
495 /*
496 * With writers on hold, if this value is zero, then there are
497 * definitely no active writers (although held writers may subsequently
498 * increment the count, they'll have to wait, and decrement it after
499 * seeing MNT_READONLY).
500 *
501 * It is OK to have counter incremented on one CPU and decremented on
502 * another: the sum will add up correctly. The danger would be when we
503 * sum up each counter, if we read a counter before it is incremented,
504 * but then read another CPU's count which it has been subsequently
505 * decremented from -- we would see more decrements than we should.
506 * MNT_WRITE_HOLD protects against this scenario, because
507 * mnt_want_write first increments count, then smp_mb, then spins on
508 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
509 * we're counting up here.
510 */
511 if (mnt_get_writers(mnt) > 0)
512 ret = -EBUSY;
513 else
514 mnt->mnt.mnt_flags |= MNT_READONLY;
515 /*
516 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
517 * that become unheld will see MNT_READONLY.
518 */
519 smp_wmb();
520 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
521 unlock_mount_hash();
522 return ret;
523 }
524
525 static void __mnt_unmake_readonly(struct mount *mnt)
526 {
527 lock_mount_hash();
528 mnt->mnt.mnt_flags &= ~MNT_READONLY;
529 unlock_mount_hash();
530 }
531
532 int sb_prepare_remount_readonly(struct super_block *sb)
533 {
534 struct mount *mnt;
535 int err = 0;
536
537 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
538 if (atomic_long_read(&sb->s_remove_count))
539 return -EBUSY;
540
541 lock_mount_hash();
542 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
543 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
544 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
545 smp_mb();
546 if (mnt_get_writers(mnt) > 0) {
547 err = -EBUSY;
548 break;
549 }
550 }
551 }
552 if (!err && atomic_long_read(&sb->s_remove_count))
553 err = -EBUSY;
554
555 if (!err) {
556 sb->s_readonly_remount = 1;
557 smp_wmb();
558 }
559 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
560 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
561 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
562 }
563 unlock_mount_hash();
564
565 return err;
566 }
567
568 static void free_vfsmnt(struct mount *mnt)
569 {
570 kfree(mnt->mnt_devname);
571 #ifdef CONFIG_SMP
572 free_percpu(mnt->mnt_pcp);
573 #endif
574 kmem_cache_free(mnt_cache, mnt);
575 }
576
577 static void delayed_free_vfsmnt(struct rcu_head *head)
578 {
579 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
580 }
581
582 /* call under rcu_read_lock */
583 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
584 {
585 struct mount *mnt;
586 if (read_seqretry(&mount_lock, seq))
587 return false;
588 if (bastard == NULL)
589 return true;
590 mnt = real_mount(bastard);
591 mnt_add_count(mnt, 1);
592 if (likely(!read_seqretry(&mount_lock, seq)))
593 return true;
594 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
595 mnt_add_count(mnt, -1);
596 return false;
597 }
598 rcu_read_unlock();
599 mntput(bastard);
600 rcu_read_lock();
601 return false;
602 }
603
604 /*
605 * find the first mount at @dentry on vfsmount @mnt.
606 * call under rcu_read_lock()
607 */
608 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
609 {
610 struct hlist_head *head = m_hash(mnt, dentry);
611 struct mount *p;
612
613 hlist_for_each_entry_rcu(p, head, mnt_hash)
614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
615 return p;
616 return NULL;
617 }
618
619 /*
620 * find the last mount at @dentry on vfsmount @mnt.
621 * mount_lock must be held.
622 */
623 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
624 {
625 struct mount *p, *res;
626 res = p = __lookup_mnt(mnt, dentry);
627 if (!p)
628 goto out;
629 hlist_for_each_entry_continue(p, mnt_hash) {
630 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
631 break;
632 res = p;
633 }
634 out:
635 return res;
636 }
637
638 /*
639 * lookup_mnt - Return the first child mount mounted at path
640 *
641 * "First" means first mounted chronologically. If you create the
642 * following mounts:
643 *
644 * mount /dev/sda1 /mnt
645 * mount /dev/sda2 /mnt
646 * mount /dev/sda3 /mnt
647 *
648 * Then lookup_mnt() on the base /mnt dentry in the root mount will
649 * return successively the root dentry and vfsmount of /dev/sda1, then
650 * /dev/sda2, then /dev/sda3, then NULL.
651 *
652 * lookup_mnt takes a reference to the found vfsmount.
653 */
654 struct vfsmount *lookup_mnt(struct path *path)
655 {
656 struct mount *child_mnt;
657 struct vfsmount *m;
658 unsigned seq;
659
660 rcu_read_lock();
661 do {
662 seq = read_seqbegin(&mount_lock);
663 child_mnt = __lookup_mnt(path->mnt, path->dentry);
664 m = child_mnt ? &child_mnt->mnt : NULL;
665 } while (!legitimize_mnt(m, seq));
666 rcu_read_unlock();
667 return m;
668 }
669
670 static struct mountpoint *new_mountpoint(struct dentry *dentry)
671 {
672 struct hlist_head *chain = mp_hash(dentry);
673 struct mountpoint *mp;
674 int ret;
675
676 hlist_for_each_entry(mp, chain, m_hash) {
677 if (mp->m_dentry == dentry) {
678 /* might be worth a WARN_ON() */
679 if (d_unlinked(dentry))
680 return ERR_PTR(-ENOENT);
681 mp->m_count++;
682 return mp;
683 }
684 }
685
686 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
687 if (!mp)
688 return ERR_PTR(-ENOMEM);
689
690 ret = d_set_mounted(dentry);
691 if (ret) {
692 kfree(mp);
693 return ERR_PTR(ret);
694 }
695
696 mp->m_dentry = dentry;
697 mp->m_count = 1;
698 hlist_add_head(&mp->m_hash, chain);
699 return mp;
700 }
701
702 static void put_mountpoint(struct mountpoint *mp)
703 {
704 if (!--mp->m_count) {
705 struct dentry *dentry = mp->m_dentry;
706 spin_lock(&dentry->d_lock);
707 dentry->d_flags &= ~DCACHE_MOUNTED;
708 spin_unlock(&dentry->d_lock);
709 hlist_del(&mp->m_hash);
710 kfree(mp);
711 }
712 }
713
714 static inline int check_mnt(struct mount *mnt)
715 {
716 return mnt->mnt_ns == current->nsproxy->mnt_ns;
717 }
718
719 /*
720 * vfsmount lock must be held for write
721 */
722 static void touch_mnt_namespace(struct mnt_namespace *ns)
723 {
724 if (ns) {
725 ns->event = ++event;
726 wake_up_interruptible(&ns->poll);
727 }
728 }
729
730 /*
731 * vfsmount lock must be held for write
732 */
733 static void __touch_mnt_namespace(struct mnt_namespace *ns)
734 {
735 if (ns && ns->event != event) {
736 ns->event = event;
737 wake_up_interruptible(&ns->poll);
738 }
739 }
740
741 /*
742 * vfsmount lock must be held for write
743 */
744 static void detach_mnt(struct mount *mnt, struct path *old_path)
745 {
746 old_path->dentry = mnt->mnt_mountpoint;
747 old_path->mnt = &mnt->mnt_parent->mnt;
748 mnt->mnt_parent = mnt;
749 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
750 list_del_init(&mnt->mnt_child);
751 hlist_del_init_rcu(&mnt->mnt_hash);
752 put_mountpoint(mnt->mnt_mp);
753 mnt->mnt_mp = NULL;
754 }
755
756 /*
757 * vfsmount lock must be held for write
758 */
759 void mnt_set_mountpoint(struct mount *mnt,
760 struct mountpoint *mp,
761 struct mount *child_mnt)
762 {
763 mp->m_count++;
764 mnt_add_count(mnt, 1); /* essentially, that's mntget */
765 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
766 child_mnt->mnt_parent = mnt;
767 child_mnt->mnt_mp = mp;
768 }
769
770 /*
771 * vfsmount lock must be held for write
772 */
773 static void attach_mnt(struct mount *mnt,
774 struct mount *parent,
775 struct mountpoint *mp)
776 {
777 mnt_set_mountpoint(parent, mp, mnt);
778 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
779 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
780 }
781
782 /*
783 * vfsmount lock must be held for write
784 */
785 static void commit_tree(struct mount *mnt, struct mount *shadows)
786 {
787 struct mount *parent = mnt->mnt_parent;
788 struct mount *m;
789 LIST_HEAD(head);
790 struct mnt_namespace *n = parent->mnt_ns;
791
792 BUG_ON(parent == mnt);
793
794 list_add_tail(&head, &mnt->mnt_list);
795 list_for_each_entry(m, &head, mnt_list)
796 m->mnt_ns = n;
797
798 list_splice(&head, n->list.prev);
799
800 if (shadows)
801 hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash);
802 else
803 hlist_add_head_rcu(&mnt->mnt_hash,
804 m_hash(&parent->mnt, mnt->mnt_mountpoint));
805 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
806 touch_mnt_namespace(n);
807 }
808
809 static struct mount *next_mnt(struct mount *p, struct mount *root)
810 {
811 struct list_head *next = p->mnt_mounts.next;
812 if (next == &p->mnt_mounts) {
813 while (1) {
814 if (p == root)
815 return NULL;
816 next = p->mnt_child.next;
817 if (next != &p->mnt_parent->mnt_mounts)
818 break;
819 p = p->mnt_parent;
820 }
821 }
822 return list_entry(next, struct mount, mnt_child);
823 }
824
825 static struct mount *skip_mnt_tree(struct mount *p)
826 {
827 struct list_head *prev = p->mnt_mounts.prev;
828 while (prev != &p->mnt_mounts) {
829 p = list_entry(prev, struct mount, mnt_child);
830 prev = p->mnt_mounts.prev;
831 }
832 return p;
833 }
834
835 struct vfsmount *
836 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
837 {
838 struct mount *mnt;
839 struct dentry *root;
840
841 if (!type)
842 return ERR_PTR(-ENODEV);
843
844 mnt = alloc_vfsmnt(name);
845 if (!mnt)
846 return ERR_PTR(-ENOMEM);
847
848 if (flags & MS_KERNMOUNT)
849 mnt->mnt.mnt_flags = MNT_INTERNAL;
850
851 root = mount_fs(type, flags, name, data);
852 if (IS_ERR(root)) {
853 mnt_free_id(mnt);
854 free_vfsmnt(mnt);
855 return ERR_CAST(root);
856 }
857
858 mnt->mnt.mnt_root = root;
859 mnt->mnt.mnt_sb = root->d_sb;
860 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
861 mnt->mnt_parent = mnt;
862 lock_mount_hash();
863 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
864 unlock_mount_hash();
865 return &mnt->mnt;
866 }
867 EXPORT_SYMBOL_GPL(vfs_kern_mount);
868
869 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
870 int flag)
871 {
872 struct super_block *sb = old->mnt.mnt_sb;
873 struct mount *mnt;
874 int err;
875
876 mnt = alloc_vfsmnt(old->mnt_devname);
877 if (!mnt)
878 return ERR_PTR(-ENOMEM);
879
880 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
881 mnt->mnt_group_id = 0; /* not a peer of original */
882 else
883 mnt->mnt_group_id = old->mnt_group_id;
884
885 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
886 err = mnt_alloc_group_id(mnt);
887 if (err)
888 goto out_free;
889 }
890
891 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
892 /* Don't allow unprivileged users to change mount flags */
893 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
894 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
895
896 /* Don't allow unprivileged users to reveal what is under a mount */
897 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
898 mnt->mnt.mnt_flags |= MNT_LOCKED;
899
900 atomic_inc(&sb->s_active);
901 mnt->mnt.mnt_sb = sb;
902 mnt->mnt.mnt_root = dget(root);
903 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
904 mnt->mnt_parent = mnt;
905 lock_mount_hash();
906 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
907 unlock_mount_hash();
908
909 if ((flag & CL_SLAVE) ||
910 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
911 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
912 mnt->mnt_master = old;
913 CLEAR_MNT_SHARED(mnt);
914 } else if (!(flag & CL_PRIVATE)) {
915 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
916 list_add(&mnt->mnt_share, &old->mnt_share);
917 if (IS_MNT_SLAVE(old))
918 list_add(&mnt->mnt_slave, &old->mnt_slave);
919 mnt->mnt_master = old->mnt_master;
920 }
921 if (flag & CL_MAKE_SHARED)
922 set_mnt_shared(mnt);
923
924 /* stick the duplicate mount on the same expiry list
925 * as the original if that was on one */
926 if (flag & CL_EXPIRE) {
927 if (!list_empty(&old->mnt_expire))
928 list_add(&mnt->mnt_expire, &old->mnt_expire);
929 }
930
931 return mnt;
932
933 out_free:
934 mnt_free_id(mnt);
935 free_vfsmnt(mnt);
936 return ERR_PTR(err);
937 }
938
939 static void mntput_no_expire(struct mount *mnt)
940 {
941 put_again:
942 rcu_read_lock();
943 mnt_add_count(mnt, -1);
944 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
945 rcu_read_unlock();
946 return;
947 }
948 lock_mount_hash();
949 if (mnt_get_count(mnt)) {
950 rcu_read_unlock();
951 unlock_mount_hash();
952 return;
953 }
954 if (unlikely(mnt->mnt_pinned)) {
955 mnt_add_count(mnt, mnt->mnt_pinned + 1);
956 mnt->mnt_pinned = 0;
957 rcu_read_unlock();
958 unlock_mount_hash();
959 acct_auto_close_mnt(&mnt->mnt);
960 goto put_again;
961 }
962 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
963 rcu_read_unlock();
964 unlock_mount_hash();
965 return;
966 }
967 mnt->mnt.mnt_flags |= MNT_DOOMED;
968 rcu_read_unlock();
969
970 list_del(&mnt->mnt_instance);
971 unlock_mount_hash();
972
973 /*
974 * This probably indicates that somebody messed
975 * up a mnt_want/drop_write() pair. If this
976 * happens, the filesystem was probably unable
977 * to make r/w->r/o transitions.
978 */
979 /*
980 * The locking used to deal with mnt_count decrement provides barriers,
981 * so mnt_get_writers() below is safe.
982 */
983 WARN_ON(mnt_get_writers(mnt));
984 fsnotify_vfsmount_delete(&mnt->mnt);
985 dput(mnt->mnt.mnt_root);
986 deactivate_super(mnt->mnt.mnt_sb);
987 mnt_free_id(mnt);
988 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
989 }
990
991 void mntput(struct vfsmount *mnt)
992 {
993 if (mnt) {
994 struct mount *m = real_mount(mnt);
995 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
996 if (unlikely(m->mnt_expiry_mark))
997 m->mnt_expiry_mark = 0;
998 mntput_no_expire(m);
999 }
1000 }
1001 EXPORT_SYMBOL(mntput);
1002
1003 struct vfsmount *mntget(struct vfsmount *mnt)
1004 {
1005 if (mnt)
1006 mnt_add_count(real_mount(mnt), 1);
1007 return mnt;
1008 }
1009 EXPORT_SYMBOL(mntget);
1010
1011 void mnt_pin(struct vfsmount *mnt)
1012 {
1013 lock_mount_hash();
1014 real_mount(mnt)->mnt_pinned++;
1015 unlock_mount_hash();
1016 }
1017 EXPORT_SYMBOL(mnt_pin);
1018
1019 void mnt_unpin(struct vfsmount *m)
1020 {
1021 struct mount *mnt = real_mount(m);
1022 lock_mount_hash();
1023 if (mnt->mnt_pinned) {
1024 mnt_add_count(mnt, 1);
1025 mnt->mnt_pinned--;
1026 }
1027 unlock_mount_hash();
1028 }
1029 EXPORT_SYMBOL(mnt_unpin);
1030
1031 static inline void mangle(struct seq_file *m, const char *s)
1032 {
1033 seq_escape(m, s, " \t\n\\");
1034 }
1035
1036 /*
1037 * Simple .show_options callback for filesystems which don't want to
1038 * implement more complex mount option showing.
1039 *
1040 * See also save_mount_options().
1041 */
1042 int generic_show_options(struct seq_file *m, struct dentry *root)
1043 {
1044 const char *options;
1045
1046 rcu_read_lock();
1047 options = rcu_dereference(root->d_sb->s_options);
1048
1049 if (options != NULL && options[0]) {
1050 seq_putc(m, ',');
1051 mangle(m, options);
1052 }
1053 rcu_read_unlock();
1054
1055 return 0;
1056 }
1057 EXPORT_SYMBOL(generic_show_options);
1058
1059 /*
1060 * If filesystem uses generic_show_options(), this function should be
1061 * called from the fill_super() callback.
1062 *
1063 * The .remount_fs callback usually needs to be handled in a special
1064 * way, to make sure, that previous options are not overwritten if the
1065 * remount fails.
1066 *
1067 * Also note, that if the filesystem's .remount_fs function doesn't
1068 * reset all options to their default value, but changes only newly
1069 * given options, then the displayed options will not reflect reality
1070 * any more.
1071 */
1072 void save_mount_options(struct super_block *sb, char *options)
1073 {
1074 BUG_ON(sb->s_options);
1075 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1076 }
1077 EXPORT_SYMBOL(save_mount_options);
1078
1079 void replace_mount_options(struct super_block *sb, char *options)
1080 {
1081 char *old = sb->s_options;
1082 rcu_assign_pointer(sb->s_options, options);
1083 if (old) {
1084 synchronize_rcu();
1085 kfree(old);
1086 }
1087 }
1088 EXPORT_SYMBOL(replace_mount_options);
1089
1090 #ifdef CONFIG_PROC_FS
1091 /* iterator; we want it to have access to namespace_sem, thus here... */
1092 static void *m_start(struct seq_file *m, loff_t *pos)
1093 {
1094 struct proc_mounts *p = proc_mounts(m);
1095
1096 down_read(&namespace_sem);
1097 if (p->cached_event == p->ns->event) {
1098 void *v = p->cached_mount;
1099 if (*pos == p->cached_index)
1100 return v;
1101 if (*pos == p->cached_index + 1) {
1102 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1103 return p->cached_mount = v;
1104 }
1105 }
1106
1107 p->cached_event = p->ns->event;
1108 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1109 p->cached_index = *pos;
1110 return p->cached_mount;
1111 }
1112
1113 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1114 {
1115 struct proc_mounts *p = proc_mounts(m);
1116
1117 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1118 p->cached_index = *pos;
1119 return p->cached_mount;
1120 }
1121
1122 static void m_stop(struct seq_file *m, void *v)
1123 {
1124 up_read(&namespace_sem);
1125 }
1126
1127 static int m_show(struct seq_file *m, void *v)
1128 {
1129 struct proc_mounts *p = proc_mounts(m);
1130 struct mount *r = list_entry(v, struct mount, mnt_list);
1131 return p->show(m, &r->mnt);
1132 }
1133
1134 const struct seq_operations mounts_op = {
1135 .start = m_start,
1136 .next = m_next,
1137 .stop = m_stop,
1138 .show = m_show,
1139 };
1140 #endif /* CONFIG_PROC_FS */
1141
1142 /**
1143 * may_umount_tree - check if a mount tree is busy
1144 * @mnt: root of mount tree
1145 *
1146 * This is called to check if a tree of mounts has any
1147 * open files, pwds, chroots or sub mounts that are
1148 * busy.
1149 */
1150 int may_umount_tree(struct vfsmount *m)
1151 {
1152 struct mount *mnt = real_mount(m);
1153 int actual_refs = 0;
1154 int minimum_refs = 0;
1155 struct mount *p;
1156 BUG_ON(!m);
1157
1158 /* write lock needed for mnt_get_count */
1159 lock_mount_hash();
1160 for (p = mnt; p; p = next_mnt(p, mnt)) {
1161 actual_refs += mnt_get_count(p);
1162 minimum_refs += 2;
1163 }
1164 unlock_mount_hash();
1165
1166 if (actual_refs > minimum_refs)
1167 return 0;
1168
1169 return 1;
1170 }
1171
1172 EXPORT_SYMBOL(may_umount_tree);
1173
1174 /**
1175 * may_umount - check if a mount point is busy
1176 * @mnt: root of mount
1177 *
1178 * This is called to check if a mount point has any
1179 * open files, pwds, chroots or sub mounts. If the
1180 * mount has sub mounts this will return busy
1181 * regardless of whether the sub mounts are busy.
1182 *
1183 * Doesn't take quota and stuff into account. IOW, in some cases it will
1184 * give false negatives. The main reason why it's here is that we need
1185 * a non-destructive way to look for easily umountable filesystems.
1186 */
1187 int may_umount(struct vfsmount *mnt)
1188 {
1189 int ret = 1;
1190 down_read(&namespace_sem);
1191 lock_mount_hash();
1192 if (propagate_mount_busy(real_mount(mnt), 2))
1193 ret = 0;
1194 unlock_mount_hash();
1195 up_read(&namespace_sem);
1196 return ret;
1197 }
1198
1199 EXPORT_SYMBOL(may_umount);
1200
1201 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1202
1203 static void namespace_unlock(void)
1204 {
1205 struct mount *mnt;
1206 struct hlist_head head = unmounted;
1207
1208 if (likely(hlist_empty(&head))) {
1209 up_write(&namespace_sem);
1210 return;
1211 }
1212
1213 head.first->pprev = &head.first;
1214 INIT_HLIST_HEAD(&unmounted);
1215
1216 up_write(&namespace_sem);
1217
1218 synchronize_rcu();
1219
1220 while (!hlist_empty(&head)) {
1221 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1222 hlist_del_init(&mnt->mnt_hash);
1223 if (mnt->mnt_ex_mountpoint.mnt)
1224 path_put(&mnt->mnt_ex_mountpoint);
1225 mntput(&mnt->mnt);
1226 }
1227 }
1228
1229 static inline void namespace_lock(void)
1230 {
1231 down_write(&namespace_sem);
1232 }
1233
1234 /*
1235 * mount_lock must be held
1236 * namespace_sem must be held for write
1237 * how = 0 => just this tree, don't propagate
1238 * how = 1 => propagate; we know that nobody else has reference to any victims
1239 * how = 2 => lazy umount
1240 */
1241 void umount_tree(struct mount *mnt, int how)
1242 {
1243 HLIST_HEAD(tmp_list);
1244 struct mount *p;
1245 struct mount *last = NULL;
1246
1247 for (p = mnt; p; p = next_mnt(p, mnt)) {
1248 hlist_del_init_rcu(&p->mnt_hash);
1249 hlist_add_head(&p->mnt_hash, &tmp_list);
1250 }
1251
1252 if (how)
1253 propagate_umount(&tmp_list);
1254
1255 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
1256 list_del_init(&p->mnt_expire);
1257 list_del_init(&p->mnt_list);
1258 __touch_mnt_namespace(p->mnt_ns);
1259 p->mnt_ns = NULL;
1260 if (how < 2)
1261 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1262 list_del_init(&p->mnt_child);
1263 if (mnt_has_parent(p)) {
1264 put_mountpoint(p->mnt_mp);
1265 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1266 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1267 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1268 p->mnt_mountpoint = p->mnt.mnt_root;
1269 p->mnt_parent = p;
1270 p->mnt_mp = NULL;
1271 }
1272 change_mnt_propagation(p, MS_PRIVATE);
1273 last = p;
1274 }
1275 if (last) {
1276 last->mnt_hash.next = unmounted.first;
1277 unmounted.first = tmp_list.first;
1278 unmounted.first->pprev = &unmounted.first;
1279 }
1280 }
1281
1282 static void shrink_submounts(struct mount *mnt);
1283
1284 static int do_umount(struct mount *mnt, int flags)
1285 {
1286 struct super_block *sb = mnt->mnt.mnt_sb;
1287 int retval;
1288
1289 retval = security_sb_umount(&mnt->mnt, flags);
1290 if (retval)
1291 return retval;
1292
1293 /*
1294 * Allow userspace to request a mountpoint be expired rather than
1295 * unmounting unconditionally. Unmount only happens if:
1296 * (1) the mark is already set (the mark is cleared by mntput())
1297 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1298 */
1299 if (flags & MNT_EXPIRE) {
1300 if (&mnt->mnt == current->fs->root.mnt ||
1301 flags & (MNT_FORCE | MNT_DETACH))
1302 return -EINVAL;
1303
1304 /*
1305 * probably don't strictly need the lock here if we examined
1306 * all race cases, but it's a slowpath.
1307 */
1308 lock_mount_hash();
1309 if (mnt_get_count(mnt) != 2) {
1310 unlock_mount_hash();
1311 return -EBUSY;
1312 }
1313 unlock_mount_hash();
1314
1315 if (!xchg(&mnt->mnt_expiry_mark, 1))
1316 return -EAGAIN;
1317 }
1318
1319 /*
1320 * If we may have to abort operations to get out of this
1321 * mount, and they will themselves hold resources we must
1322 * allow the fs to do things. In the Unix tradition of
1323 * 'Gee thats tricky lets do it in userspace' the umount_begin
1324 * might fail to complete on the first run through as other tasks
1325 * must return, and the like. Thats for the mount program to worry
1326 * about for the moment.
1327 */
1328
1329 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1330 sb->s_op->umount_begin(sb);
1331 }
1332
1333 /*
1334 * No sense to grab the lock for this test, but test itself looks
1335 * somewhat bogus. Suggestions for better replacement?
1336 * Ho-hum... In principle, we might treat that as umount + switch
1337 * to rootfs. GC would eventually take care of the old vfsmount.
1338 * Actually it makes sense, especially if rootfs would contain a
1339 * /reboot - static binary that would close all descriptors and
1340 * call reboot(9). Then init(8) could umount root and exec /reboot.
1341 */
1342 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1343 /*
1344 * Special case for "unmounting" root ...
1345 * we just try to remount it readonly.
1346 */
1347 down_write(&sb->s_umount);
1348 if (!(sb->s_flags & MS_RDONLY))
1349 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1350 up_write(&sb->s_umount);
1351 return retval;
1352 }
1353
1354 namespace_lock();
1355 lock_mount_hash();
1356 event++;
1357
1358 if (flags & MNT_DETACH) {
1359 if (!list_empty(&mnt->mnt_list))
1360 umount_tree(mnt, 2);
1361 retval = 0;
1362 } else {
1363 shrink_submounts(mnt);
1364 retval = -EBUSY;
1365 if (!propagate_mount_busy(mnt, 2)) {
1366 if (!list_empty(&mnt->mnt_list))
1367 umount_tree(mnt, 1);
1368 retval = 0;
1369 }
1370 }
1371 unlock_mount_hash();
1372 namespace_unlock();
1373 return retval;
1374 }
1375
1376 /*
1377 * Is the caller allowed to modify his namespace?
1378 */
1379 static inline bool may_mount(void)
1380 {
1381 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1382 }
1383
1384 /*
1385 * Now umount can handle mount points as well as block devices.
1386 * This is important for filesystems which use unnamed block devices.
1387 *
1388 * We now support a flag for forced unmount like the other 'big iron'
1389 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1390 */
1391
1392 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1393 {
1394 struct path path;
1395 struct mount *mnt;
1396 int retval;
1397 int lookup_flags = 0;
1398
1399 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1400 return -EINVAL;
1401
1402 if (!may_mount())
1403 return -EPERM;
1404
1405 if (!(flags & UMOUNT_NOFOLLOW))
1406 lookup_flags |= LOOKUP_FOLLOW;
1407
1408 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1409 if (retval)
1410 goto out;
1411 mnt = real_mount(path.mnt);
1412 retval = -EINVAL;
1413 if (path.dentry != path.mnt->mnt_root)
1414 goto dput_and_out;
1415 if (!check_mnt(mnt))
1416 goto dput_and_out;
1417 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1418 goto dput_and_out;
1419
1420 retval = do_umount(mnt, flags);
1421 dput_and_out:
1422 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1423 dput(path.dentry);
1424 mntput_no_expire(mnt);
1425 out:
1426 return retval;
1427 }
1428
1429 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1430
1431 /*
1432 * The 2.0 compatible umount. No flags.
1433 */
1434 SYSCALL_DEFINE1(oldumount, char __user *, name)
1435 {
1436 return sys_umount(name, 0);
1437 }
1438
1439 #endif
1440
1441 static bool is_mnt_ns_file(struct dentry *dentry)
1442 {
1443 /* Is this a proxy for a mount namespace? */
1444 struct inode *inode = dentry->d_inode;
1445 struct proc_ns *ei;
1446
1447 if (!proc_ns_inode(inode))
1448 return false;
1449
1450 ei = get_proc_ns(inode);
1451 if (ei->ns_ops != &mntns_operations)
1452 return false;
1453
1454 return true;
1455 }
1456
1457 static bool mnt_ns_loop(struct dentry *dentry)
1458 {
1459 /* Could bind mounting the mount namespace inode cause a
1460 * mount namespace loop?
1461 */
1462 struct mnt_namespace *mnt_ns;
1463 if (!is_mnt_ns_file(dentry))
1464 return false;
1465
1466 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1467 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1468 }
1469
1470 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1471 int flag)
1472 {
1473 struct mount *res, *p, *q, *r, *parent;
1474
1475 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1476 return ERR_PTR(-EINVAL);
1477
1478 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1479 return ERR_PTR(-EINVAL);
1480
1481 res = q = clone_mnt(mnt, dentry, flag);
1482 if (IS_ERR(q))
1483 return q;
1484
1485 q->mnt.mnt_flags &= ~MNT_LOCKED;
1486 q->mnt_mountpoint = mnt->mnt_mountpoint;
1487
1488 p = mnt;
1489 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1490 struct mount *s;
1491 if (!is_subdir(r->mnt_mountpoint, dentry))
1492 continue;
1493
1494 for (s = r; s; s = next_mnt(s, r)) {
1495 if (!(flag & CL_COPY_UNBINDABLE) &&
1496 IS_MNT_UNBINDABLE(s)) {
1497 s = skip_mnt_tree(s);
1498 continue;
1499 }
1500 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1501 is_mnt_ns_file(s->mnt.mnt_root)) {
1502 s = skip_mnt_tree(s);
1503 continue;
1504 }
1505 while (p != s->mnt_parent) {
1506 p = p->mnt_parent;
1507 q = q->mnt_parent;
1508 }
1509 p = s;
1510 parent = q;
1511 q = clone_mnt(p, p->mnt.mnt_root, flag);
1512 if (IS_ERR(q))
1513 goto out;
1514 lock_mount_hash();
1515 list_add_tail(&q->mnt_list, &res->mnt_list);
1516 attach_mnt(q, parent, p->mnt_mp);
1517 unlock_mount_hash();
1518 }
1519 }
1520 return res;
1521 out:
1522 if (res) {
1523 lock_mount_hash();
1524 umount_tree(res, 0);
1525 unlock_mount_hash();
1526 }
1527 return q;
1528 }
1529
1530 /* Caller should check returned pointer for errors */
1531
1532 struct vfsmount *collect_mounts(struct path *path)
1533 {
1534 struct mount *tree;
1535 namespace_lock();
1536 tree = copy_tree(real_mount(path->mnt), path->dentry,
1537 CL_COPY_ALL | CL_PRIVATE);
1538 namespace_unlock();
1539 if (IS_ERR(tree))
1540 return ERR_CAST(tree);
1541 return &tree->mnt;
1542 }
1543
1544 void drop_collected_mounts(struct vfsmount *mnt)
1545 {
1546 namespace_lock();
1547 lock_mount_hash();
1548 umount_tree(real_mount(mnt), 0);
1549 unlock_mount_hash();
1550 namespace_unlock();
1551 }
1552
1553 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1554 struct vfsmount *root)
1555 {
1556 struct mount *mnt;
1557 int res = f(root, arg);
1558 if (res)
1559 return res;
1560 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1561 res = f(&mnt->mnt, arg);
1562 if (res)
1563 return res;
1564 }
1565 return 0;
1566 }
1567
1568 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1569 {
1570 struct mount *p;
1571
1572 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1573 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1574 mnt_release_group_id(p);
1575 }
1576 }
1577
1578 static int invent_group_ids(struct mount *mnt, bool recurse)
1579 {
1580 struct mount *p;
1581
1582 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1583 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1584 int err = mnt_alloc_group_id(p);
1585 if (err) {
1586 cleanup_group_ids(mnt, p);
1587 return err;
1588 }
1589 }
1590 }
1591
1592 return 0;
1593 }
1594
1595 /*
1596 * @source_mnt : mount tree to be attached
1597 * @nd : place the mount tree @source_mnt is attached
1598 * @parent_nd : if non-null, detach the source_mnt from its parent and
1599 * store the parent mount and mountpoint dentry.
1600 * (done when source_mnt is moved)
1601 *
1602 * NOTE: in the table below explains the semantics when a source mount
1603 * of a given type is attached to a destination mount of a given type.
1604 * ---------------------------------------------------------------------------
1605 * | BIND MOUNT OPERATION |
1606 * |**************************************************************************
1607 * | source-->| shared | private | slave | unbindable |
1608 * | dest | | | | |
1609 * | | | | | | |
1610 * | v | | | | |
1611 * |**************************************************************************
1612 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1613 * | | | | | |
1614 * |non-shared| shared (+) | private | slave (*) | invalid |
1615 * ***************************************************************************
1616 * A bind operation clones the source mount and mounts the clone on the
1617 * destination mount.
1618 *
1619 * (++) the cloned mount is propagated to all the mounts in the propagation
1620 * tree of the destination mount and the cloned mount is added to
1621 * the peer group of the source mount.
1622 * (+) the cloned mount is created under the destination mount and is marked
1623 * as shared. The cloned mount is added to the peer group of the source
1624 * mount.
1625 * (+++) the mount is propagated to all the mounts in the propagation tree
1626 * of the destination mount and the cloned mount is made slave
1627 * of the same master as that of the source mount. The cloned mount
1628 * is marked as 'shared and slave'.
1629 * (*) the cloned mount is made a slave of the same master as that of the
1630 * source mount.
1631 *
1632 * ---------------------------------------------------------------------------
1633 * | MOVE MOUNT OPERATION |
1634 * |**************************************************************************
1635 * | source-->| shared | private | slave | unbindable |
1636 * | dest | | | | |
1637 * | | | | | | |
1638 * | v | | | | |
1639 * |**************************************************************************
1640 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1641 * | | | | | |
1642 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1643 * ***************************************************************************
1644 *
1645 * (+) the mount is moved to the destination. And is then propagated to
1646 * all the mounts in the propagation tree of the destination mount.
1647 * (+*) the mount is moved to the destination.
1648 * (+++) the mount is moved to the destination and is then propagated to
1649 * all the mounts belonging to the destination mount's propagation tree.
1650 * the mount is marked as 'shared and slave'.
1651 * (*) the mount continues to be a slave at the new location.
1652 *
1653 * if the source mount is a tree, the operations explained above is
1654 * applied to each mount in the tree.
1655 * Must be called without spinlocks held, since this function can sleep
1656 * in allocations.
1657 */
1658 static int attach_recursive_mnt(struct mount *source_mnt,
1659 struct mount *dest_mnt,
1660 struct mountpoint *dest_mp,
1661 struct path *parent_path)
1662 {
1663 HLIST_HEAD(tree_list);
1664 struct mount *child, *p;
1665 struct hlist_node *n;
1666 int err;
1667
1668 if (IS_MNT_SHARED(dest_mnt)) {
1669 err = invent_group_ids(source_mnt, true);
1670 if (err)
1671 goto out;
1672 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1673 lock_mount_hash();
1674 if (err)
1675 goto out_cleanup_ids;
1676 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1677 set_mnt_shared(p);
1678 } else {
1679 lock_mount_hash();
1680 }
1681 if (parent_path) {
1682 detach_mnt(source_mnt, parent_path);
1683 attach_mnt(source_mnt, dest_mnt, dest_mp);
1684 touch_mnt_namespace(source_mnt->mnt_ns);
1685 } else {
1686 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1687 commit_tree(source_mnt, NULL);
1688 }
1689
1690 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1691 struct mount *q;
1692 hlist_del_init(&child->mnt_hash);
1693 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1694 child->mnt_mountpoint);
1695 commit_tree(child, q);
1696 }
1697 unlock_mount_hash();
1698
1699 return 0;
1700
1701 out_cleanup_ids:
1702 while (!hlist_empty(&tree_list)) {
1703 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1704 umount_tree(child, 0);
1705 }
1706 unlock_mount_hash();
1707 cleanup_group_ids(source_mnt, NULL);
1708 out:
1709 return err;
1710 }
1711
1712 static struct mountpoint *lock_mount(struct path *path)
1713 {
1714 struct vfsmount *mnt;
1715 struct dentry *dentry = path->dentry;
1716 retry:
1717 mutex_lock(&dentry->d_inode->i_mutex);
1718 if (unlikely(cant_mount(dentry))) {
1719 mutex_unlock(&dentry->d_inode->i_mutex);
1720 return ERR_PTR(-ENOENT);
1721 }
1722 namespace_lock();
1723 mnt = lookup_mnt(path);
1724 if (likely(!mnt)) {
1725 struct mountpoint *mp = new_mountpoint(dentry);
1726 if (IS_ERR(mp)) {
1727 namespace_unlock();
1728 mutex_unlock(&dentry->d_inode->i_mutex);
1729 return mp;
1730 }
1731 return mp;
1732 }
1733 namespace_unlock();
1734 mutex_unlock(&path->dentry->d_inode->i_mutex);
1735 path_put(path);
1736 path->mnt = mnt;
1737 dentry = path->dentry = dget(mnt->mnt_root);
1738 goto retry;
1739 }
1740
1741 static void unlock_mount(struct mountpoint *where)
1742 {
1743 struct dentry *dentry = where->m_dentry;
1744 put_mountpoint(where);
1745 namespace_unlock();
1746 mutex_unlock(&dentry->d_inode->i_mutex);
1747 }
1748
1749 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1750 {
1751 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1752 return -EINVAL;
1753
1754 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1755 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1756 return -ENOTDIR;
1757
1758 return attach_recursive_mnt(mnt, p, mp, NULL);
1759 }
1760
1761 /*
1762 * Sanity check the flags to change_mnt_propagation.
1763 */
1764
1765 static int flags_to_propagation_type(int flags)
1766 {
1767 int type = flags & ~(MS_REC | MS_SILENT);
1768
1769 /* Fail if any non-propagation flags are set */
1770 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1771 return 0;
1772 /* Only one propagation flag should be set */
1773 if (!is_power_of_2(type))
1774 return 0;
1775 return type;
1776 }
1777
1778 /*
1779 * recursively change the type of the mountpoint.
1780 */
1781 static int do_change_type(struct path *path, int flag)
1782 {
1783 struct mount *m;
1784 struct mount *mnt = real_mount(path->mnt);
1785 int recurse = flag & MS_REC;
1786 int type;
1787 int err = 0;
1788
1789 if (path->dentry != path->mnt->mnt_root)
1790 return -EINVAL;
1791
1792 type = flags_to_propagation_type(flag);
1793 if (!type)
1794 return -EINVAL;
1795
1796 namespace_lock();
1797 if (type == MS_SHARED) {
1798 err = invent_group_ids(mnt, recurse);
1799 if (err)
1800 goto out_unlock;
1801 }
1802
1803 lock_mount_hash();
1804 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1805 change_mnt_propagation(m, type);
1806 unlock_mount_hash();
1807
1808 out_unlock:
1809 namespace_unlock();
1810 return err;
1811 }
1812
1813 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1814 {
1815 struct mount *child;
1816 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1817 if (!is_subdir(child->mnt_mountpoint, dentry))
1818 continue;
1819
1820 if (child->mnt.mnt_flags & MNT_LOCKED)
1821 return true;
1822 }
1823 return false;
1824 }
1825
1826 /*
1827 * do loopback mount.
1828 */
1829 static int do_loopback(struct path *path, const char *old_name,
1830 int recurse)
1831 {
1832 struct path old_path;
1833 struct mount *mnt = NULL, *old, *parent;
1834 struct mountpoint *mp;
1835 int err;
1836 if (!old_name || !*old_name)
1837 return -EINVAL;
1838 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1839 if (err)
1840 return err;
1841
1842 err = -EINVAL;
1843 if (mnt_ns_loop(old_path.dentry))
1844 goto out;
1845
1846 mp = lock_mount(path);
1847 err = PTR_ERR(mp);
1848 if (IS_ERR(mp))
1849 goto out;
1850
1851 old = real_mount(old_path.mnt);
1852 parent = real_mount(path->mnt);
1853
1854 err = -EINVAL;
1855 if (IS_MNT_UNBINDABLE(old))
1856 goto out2;
1857
1858 if (!check_mnt(parent) || !check_mnt(old))
1859 goto out2;
1860
1861 if (!recurse && has_locked_children(old, old_path.dentry))
1862 goto out2;
1863
1864 if (recurse)
1865 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1866 else
1867 mnt = clone_mnt(old, old_path.dentry, 0);
1868
1869 if (IS_ERR(mnt)) {
1870 err = PTR_ERR(mnt);
1871 goto out2;
1872 }
1873
1874 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1875
1876 err = graft_tree(mnt, parent, mp);
1877 if (err) {
1878 lock_mount_hash();
1879 umount_tree(mnt, 0);
1880 unlock_mount_hash();
1881 }
1882 out2:
1883 unlock_mount(mp);
1884 out:
1885 path_put(&old_path);
1886 return err;
1887 }
1888
1889 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1890 {
1891 int error = 0;
1892 int readonly_request = 0;
1893
1894 if (ms_flags & MS_RDONLY)
1895 readonly_request = 1;
1896 if (readonly_request == __mnt_is_readonly(mnt))
1897 return 0;
1898
1899 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1900 return -EPERM;
1901
1902 if (readonly_request)
1903 error = mnt_make_readonly(real_mount(mnt));
1904 else
1905 __mnt_unmake_readonly(real_mount(mnt));
1906 return error;
1907 }
1908
1909 /*
1910 * change filesystem flags. dir should be a physical root of filesystem.
1911 * If you've mounted a non-root directory somewhere and want to do remount
1912 * on it - tough luck.
1913 */
1914 static int do_remount(struct path *path, int flags, int mnt_flags,
1915 void *data)
1916 {
1917 int err;
1918 struct super_block *sb = path->mnt->mnt_sb;
1919 struct mount *mnt = real_mount(path->mnt);
1920
1921 if (!check_mnt(mnt))
1922 return -EINVAL;
1923
1924 if (path->dentry != path->mnt->mnt_root)
1925 return -EINVAL;
1926
1927 err = security_sb_remount(sb, data);
1928 if (err)
1929 return err;
1930
1931 down_write(&sb->s_umount);
1932 if (flags & MS_BIND)
1933 err = change_mount_flags(path->mnt, flags);
1934 else if (!capable(CAP_SYS_ADMIN))
1935 err = -EPERM;
1936 else
1937 err = do_remount_sb(sb, flags, data, 0);
1938 if (!err) {
1939 lock_mount_hash();
1940 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1941 mnt->mnt.mnt_flags = mnt_flags;
1942 touch_mnt_namespace(mnt->mnt_ns);
1943 unlock_mount_hash();
1944 }
1945 up_write(&sb->s_umount);
1946 return err;
1947 }
1948
1949 static inline int tree_contains_unbindable(struct mount *mnt)
1950 {
1951 struct mount *p;
1952 for (p = mnt; p; p = next_mnt(p, mnt)) {
1953 if (IS_MNT_UNBINDABLE(p))
1954 return 1;
1955 }
1956 return 0;
1957 }
1958
1959 static int do_move_mount(struct path *path, const char *old_name)
1960 {
1961 struct path old_path, parent_path;
1962 struct mount *p;
1963 struct mount *old;
1964 struct mountpoint *mp;
1965 int err;
1966 if (!old_name || !*old_name)
1967 return -EINVAL;
1968 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1969 if (err)
1970 return err;
1971
1972 mp = lock_mount(path);
1973 err = PTR_ERR(mp);
1974 if (IS_ERR(mp))
1975 goto out;
1976
1977 old = real_mount(old_path.mnt);
1978 p = real_mount(path->mnt);
1979
1980 err = -EINVAL;
1981 if (!check_mnt(p) || !check_mnt(old))
1982 goto out1;
1983
1984 if (old->mnt.mnt_flags & MNT_LOCKED)
1985 goto out1;
1986
1987 err = -EINVAL;
1988 if (old_path.dentry != old_path.mnt->mnt_root)
1989 goto out1;
1990
1991 if (!mnt_has_parent(old))
1992 goto out1;
1993
1994 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1995 S_ISDIR(old_path.dentry->d_inode->i_mode))
1996 goto out1;
1997 /*
1998 * Don't move a mount residing in a shared parent.
1999 */
2000 if (IS_MNT_SHARED(old->mnt_parent))
2001 goto out1;
2002 /*
2003 * Don't move a mount tree containing unbindable mounts to a destination
2004 * mount which is shared.
2005 */
2006 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2007 goto out1;
2008 err = -ELOOP;
2009 for (; mnt_has_parent(p); p = p->mnt_parent)
2010 if (p == old)
2011 goto out1;
2012
2013 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2014 if (err)
2015 goto out1;
2016
2017 /* if the mount is moved, it should no longer be expire
2018 * automatically */
2019 list_del_init(&old->mnt_expire);
2020 out1:
2021 unlock_mount(mp);
2022 out:
2023 if (!err)
2024 path_put(&parent_path);
2025 path_put(&old_path);
2026 return err;
2027 }
2028
2029 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2030 {
2031 int err;
2032 const char *subtype = strchr(fstype, '.');
2033 if (subtype) {
2034 subtype++;
2035 err = -EINVAL;
2036 if (!subtype[0])
2037 goto err;
2038 } else
2039 subtype = "";
2040
2041 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2042 err = -ENOMEM;
2043 if (!mnt->mnt_sb->s_subtype)
2044 goto err;
2045 return mnt;
2046
2047 err:
2048 mntput(mnt);
2049 return ERR_PTR(err);
2050 }
2051
2052 /*
2053 * add a mount into a namespace's mount tree
2054 */
2055 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2056 {
2057 struct mountpoint *mp;
2058 struct mount *parent;
2059 int err;
2060
2061 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2062
2063 mp = lock_mount(path);
2064 if (IS_ERR(mp))
2065 return PTR_ERR(mp);
2066
2067 parent = real_mount(path->mnt);
2068 err = -EINVAL;
2069 if (unlikely(!check_mnt(parent))) {
2070 /* that's acceptable only for automounts done in private ns */
2071 if (!(mnt_flags & MNT_SHRINKABLE))
2072 goto unlock;
2073 /* ... and for those we'd better have mountpoint still alive */
2074 if (!parent->mnt_ns)
2075 goto unlock;
2076 }
2077
2078 /* Refuse the same filesystem on the same mount point */
2079 err = -EBUSY;
2080 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2081 path->mnt->mnt_root == path->dentry)
2082 goto unlock;
2083
2084 err = -EINVAL;
2085 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2086 goto unlock;
2087
2088 newmnt->mnt.mnt_flags = mnt_flags;
2089 err = graft_tree(newmnt, parent, mp);
2090
2091 unlock:
2092 unlock_mount(mp);
2093 return err;
2094 }
2095
2096 /*
2097 * create a new mount for userspace and request it to be added into the
2098 * namespace's tree
2099 */
2100 static int do_new_mount(struct path *path, const char *fstype, int flags,
2101 int mnt_flags, const char *name, void *data)
2102 {
2103 struct file_system_type *type;
2104 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2105 struct vfsmount *mnt;
2106 int err;
2107
2108 if (!fstype)
2109 return -EINVAL;
2110
2111 type = get_fs_type(fstype);
2112 if (!type)
2113 return -ENODEV;
2114
2115 if (user_ns != &init_user_ns) {
2116 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2117 put_filesystem(type);
2118 return -EPERM;
2119 }
2120 /* Only in special cases allow devices from mounts
2121 * created outside the initial user namespace.
2122 */
2123 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2124 flags |= MS_NODEV;
2125 mnt_flags |= MNT_NODEV;
2126 }
2127 }
2128
2129 mnt = vfs_kern_mount(type, flags, name, data);
2130 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2131 !mnt->mnt_sb->s_subtype)
2132 mnt = fs_set_subtype(mnt, fstype);
2133
2134 put_filesystem(type);
2135 if (IS_ERR(mnt))
2136 return PTR_ERR(mnt);
2137
2138 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2139 if (err)
2140 mntput(mnt);
2141 return err;
2142 }
2143
2144 int finish_automount(struct vfsmount *m, struct path *path)
2145 {
2146 struct mount *mnt = real_mount(m);
2147 int err;
2148 /* The new mount record should have at least 2 refs to prevent it being
2149 * expired before we get a chance to add it
2150 */
2151 BUG_ON(mnt_get_count(mnt) < 2);
2152
2153 if (m->mnt_sb == path->mnt->mnt_sb &&
2154 m->mnt_root == path->dentry) {
2155 err = -ELOOP;
2156 goto fail;
2157 }
2158
2159 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2160 if (!err)
2161 return 0;
2162 fail:
2163 /* remove m from any expiration list it may be on */
2164 if (!list_empty(&mnt->mnt_expire)) {
2165 namespace_lock();
2166 list_del_init(&mnt->mnt_expire);
2167 namespace_unlock();
2168 }
2169 mntput(m);
2170 mntput(m);
2171 return err;
2172 }
2173
2174 /**
2175 * mnt_set_expiry - Put a mount on an expiration list
2176 * @mnt: The mount to list.
2177 * @expiry_list: The list to add the mount to.
2178 */
2179 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2180 {
2181 namespace_lock();
2182
2183 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2184
2185 namespace_unlock();
2186 }
2187 EXPORT_SYMBOL(mnt_set_expiry);
2188
2189 /*
2190 * process a list of expirable mountpoints with the intent of discarding any
2191 * mountpoints that aren't in use and haven't been touched since last we came
2192 * here
2193 */
2194 void mark_mounts_for_expiry(struct list_head *mounts)
2195 {
2196 struct mount *mnt, *next;
2197 LIST_HEAD(graveyard);
2198
2199 if (list_empty(mounts))
2200 return;
2201
2202 namespace_lock();
2203 lock_mount_hash();
2204
2205 /* extract from the expiration list every vfsmount that matches the
2206 * following criteria:
2207 * - only referenced by its parent vfsmount
2208 * - still marked for expiry (marked on the last call here; marks are
2209 * cleared by mntput())
2210 */
2211 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2212 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2213 propagate_mount_busy(mnt, 1))
2214 continue;
2215 list_move(&mnt->mnt_expire, &graveyard);
2216 }
2217 while (!list_empty(&graveyard)) {
2218 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2219 touch_mnt_namespace(mnt->mnt_ns);
2220 umount_tree(mnt, 1);
2221 }
2222 unlock_mount_hash();
2223 namespace_unlock();
2224 }
2225
2226 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2227
2228 /*
2229 * Ripoff of 'select_parent()'
2230 *
2231 * search the list of submounts for a given mountpoint, and move any
2232 * shrinkable submounts to the 'graveyard' list.
2233 */
2234 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2235 {
2236 struct mount *this_parent = parent;
2237 struct list_head *next;
2238 int found = 0;
2239
2240 repeat:
2241 next = this_parent->mnt_mounts.next;
2242 resume:
2243 while (next != &this_parent->mnt_mounts) {
2244 struct list_head *tmp = next;
2245 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2246
2247 next = tmp->next;
2248 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2249 continue;
2250 /*
2251 * Descend a level if the d_mounts list is non-empty.
2252 */
2253 if (!list_empty(&mnt->mnt_mounts)) {
2254 this_parent = mnt;
2255 goto repeat;
2256 }
2257
2258 if (!propagate_mount_busy(mnt, 1)) {
2259 list_move_tail(&mnt->mnt_expire, graveyard);
2260 found++;
2261 }
2262 }
2263 /*
2264 * All done at this level ... ascend and resume the search
2265 */
2266 if (this_parent != parent) {
2267 next = this_parent->mnt_child.next;
2268 this_parent = this_parent->mnt_parent;
2269 goto resume;
2270 }
2271 return found;
2272 }
2273
2274 /*
2275 * process a list of expirable mountpoints with the intent of discarding any
2276 * submounts of a specific parent mountpoint
2277 *
2278 * mount_lock must be held for write
2279 */
2280 static void shrink_submounts(struct mount *mnt)
2281 {
2282 LIST_HEAD(graveyard);
2283 struct mount *m;
2284
2285 /* extract submounts of 'mountpoint' from the expiration list */
2286 while (select_submounts(mnt, &graveyard)) {
2287 while (!list_empty(&graveyard)) {
2288 m = list_first_entry(&graveyard, struct mount,
2289 mnt_expire);
2290 touch_mnt_namespace(m->mnt_ns);
2291 umount_tree(m, 1);
2292 }
2293 }
2294 }
2295
2296 /*
2297 * Some copy_from_user() implementations do not return the exact number of
2298 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2299 * Note that this function differs from copy_from_user() in that it will oops
2300 * on bad values of `to', rather than returning a short copy.
2301 */
2302 static long exact_copy_from_user(void *to, const void __user * from,
2303 unsigned long n)
2304 {
2305 char *t = to;
2306 const char __user *f = from;
2307 char c;
2308
2309 if (!access_ok(VERIFY_READ, from, n))
2310 return n;
2311
2312 while (n) {
2313 if (__get_user(c, f)) {
2314 memset(t, 0, n);
2315 break;
2316 }
2317 *t++ = c;
2318 f++;
2319 n--;
2320 }
2321 return n;
2322 }
2323
2324 int copy_mount_options(const void __user * data, unsigned long *where)
2325 {
2326 int i;
2327 unsigned long page;
2328 unsigned long size;
2329
2330 *where = 0;
2331 if (!data)
2332 return 0;
2333
2334 if (!(page = __get_free_page(GFP_KERNEL)))
2335 return -ENOMEM;
2336
2337 /* We only care that *some* data at the address the user
2338 * gave us is valid. Just in case, we'll zero
2339 * the remainder of the page.
2340 */
2341 /* copy_from_user cannot cross TASK_SIZE ! */
2342 size = TASK_SIZE - (unsigned long)data;
2343 if (size > PAGE_SIZE)
2344 size = PAGE_SIZE;
2345
2346 i = size - exact_copy_from_user((void *)page, data, size);
2347 if (!i) {
2348 free_page(page);
2349 return -EFAULT;
2350 }
2351 if (i != PAGE_SIZE)
2352 memset((char *)page + i, 0, PAGE_SIZE - i);
2353 *where = page;
2354 return 0;
2355 }
2356
2357 int copy_mount_string(const void __user *data, char **where)
2358 {
2359 char *tmp;
2360
2361 if (!data) {
2362 *where = NULL;
2363 return 0;
2364 }
2365
2366 tmp = strndup_user(data, PAGE_SIZE);
2367 if (IS_ERR(tmp))
2368 return PTR_ERR(tmp);
2369
2370 *where = tmp;
2371 return 0;
2372 }
2373
2374 /*
2375 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2376 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2377 *
2378 * data is a (void *) that can point to any structure up to
2379 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2380 * information (or be NULL).
2381 *
2382 * Pre-0.97 versions of mount() didn't have a flags word.
2383 * When the flags word was introduced its top half was required
2384 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2385 * Therefore, if this magic number is present, it carries no information
2386 * and must be discarded.
2387 */
2388 long do_mount(const char *dev_name, const char *dir_name,
2389 const char *type_page, unsigned long flags, void *data_page)
2390 {
2391 struct path path;
2392 int retval = 0;
2393 int mnt_flags = 0;
2394
2395 /* Discard magic */
2396 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2397 flags &= ~MS_MGC_MSK;
2398
2399 /* Basic sanity checks */
2400
2401 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2402 return -EINVAL;
2403
2404 if (data_page)
2405 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2406
2407 /* ... and get the mountpoint */
2408 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2409 if (retval)
2410 return retval;
2411
2412 retval = security_sb_mount(dev_name, &path,
2413 type_page, flags, data_page);
2414 if (!retval && !may_mount())
2415 retval = -EPERM;
2416 if (retval)
2417 goto dput_out;
2418
2419 /* Default to relatime unless overriden */
2420 if (!(flags & MS_NOATIME))
2421 mnt_flags |= MNT_RELATIME;
2422
2423 /* Separate the per-mountpoint flags */
2424 if (flags & MS_NOSUID)
2425 mnt_flags |= MNT_NOSUID;
2426 if (flags & MS_NODEV)
2427 mnt_flags |= MNT_NODEV;
2428 if (flags & MS_NOEXEC)
2429 mnt_flags |= MNT_NOEXEC;
2430 if (flags & MS_NOATIME)
2431 mnt_flags |= MNT_NOATIME;
2432 if (flags & MS_NODIRATIME)
2433 mnt_flags |= MNT_NODIRATIME;
2434 if (flags & MS_STRICTATIME)
2435 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2436 if (flags & MS_RDONLY)
2437 mnt_flags |= MNT_READONLY;
2438
2439 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2440 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2441 MS_STRICTATIME);
2442
2443 if (flags & MS_REMOUNT)
2444 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2445 data_page);
2446 else if (flags & MS_BIND)
2447 retval = do_loopback(&path, dev_name, flags & MS_REC);
2448 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2449 retval = do_change_type(&path, flags);
2450 else if (flags & MS_MOVE)
2451 retval = do_move_mount(&path, dev_name);
2452 else
2453 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2454 dev_name, data_page);
2455 dput_out:
2456 path_put(&path);
2457 return retval;
2458 }
2459
2460 static void free_mnt_ns(struct mnt_namespace *ns)
2461 {
2462 proc_free_inum(ns->proc_inum);
2463 put_user_ns(ns->user_ns);
2464 kfree(ns);
2465 }
2466
2467 /*
2468 * Assign a sequence number so we can detect when we attempt to bind
2469 * mount a reference to an older mount namespace into the current
2470 * mount namespace, preventing reference counting loops. A 64bit
2471 * number incrementing at 10Ghz will take 12,427 years to wrap which
2472 * is effectively never, so we can ignore the possibility.
2473 */
2474 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2475
2476 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2477 {
2478 struct mnt_namespace *new_ns;
2479 int ret;
2480
2481 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2482 if (!new_ns)
2483 return ERR_PTR(-ENOMEM);
2484 ret = proc_alloc_inum(&new_ns->proc_inum);
2485 if (ret) {
2486 kfree(new_ns);
2487 return ERR_PTR(ret);
2488 }
2489 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2490 atomic_set(&new_ns->count, 1);
2491 new_ns->root = NULL;
2492 INIT_LIST_HEAD(&new_ns->list);
2493 init_waitqueue_head(&new_ns->poll);
2494 new_ns->event = 0;
2495 new_ns->user_ns = get_user_ns(user_ns);
2496 return new_ns;
2497 }
2498
2499 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2500 struct user_namespace *user_ns, struct fs_struct *new_fs)
2501 {
2502 struct mnt_namespace *new_ns;
2503 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2504 struct mount *p, *q;
2505 struct mount *old;
2506 struct mount *new;
2507 int copy_flags;
2508
2509 BUG_ON(!ns);
2510
2511 if (likely(!(flags & CLONE_NEWNS))) {
2512 get_mnt_ns(ns);
2513 return ns;
2514 }
2515
2516 old = ns->root;
2517
2518 new_ns = alloc_mnt_ns(user_ns);
2519 if (IS_ERR(new_ns))
2520 return new_ns;
2521
2522 namespace_lock();
2523 /* First pass: copy the tree topology */
2524 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2525 if (user_ns != ns->user_ns)
2526 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2527 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2528 if (IS_ERR(new)) {
2529 namespace_unlock();
2530 free_mnt_ns(new_ns);
2531 return ERR_CAST(new);
2532 }
2533 new_ns->root = new;
2534 list_add_tail(&new_ns->list, &new->mnt_list);
2535
2536 /*
2537 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2538 * as belonging to new namespace. We have already acquired a private
2539 * fs_struct, so tsk->fs->lock is not needed.
2540 */
2541 p = old;
2542 q = new;
2543 while (p) {
2544 q->mnt_ns = new_ns;
2545 if (new_fs) {
2546 if (&p->mnt == new_fs->root.mnt) {
2547 new_fs->root.mnt = mntget(&q->mnt);
2548 rootmnt = &p->mnt;
2549 }
2550 if (&p->mnt == new_fs->pwd.mnt) {
2551 new_fs->pwd.mnt = mntget(&q->mnt);
2552 pwdmnt = &p->mnt;
2553 }
2554 }
2555 p = next_mnt(p, old);
2556 q = next_mnt(q, new);
2557 if (!q)
2558 break;
2559 while (p->mnt.mnt_root != q->mnt.mnt_root)
2560 p = next_mnt(p, old);
2561 }
2562 namespace_unlock();
2563
2564 if (rootmnt)
2565 mntput(rootmnt);
2566 if (pwdmnt)
2567 mntput(pwdmnt);
2568
2569 return new_ns;
2570 }
2571
2572 /**
2573 * create_mnt_ns - creates a private namespace and adds a root filesystem
2574 * @mnt: pointer to the new root filesystem mountpoint
2575 */
2576 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2577 {
2578 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2579 if (!IS_ERR(new_ns)) {
2580 struct mount *mnt = real_mount(m);
2581 mnt->mnt_ns = new_ns;
2582 new_ns->root = mnt;
2583 list_add(&mnt->mnt_list, &new_ns->list);
2584 } else {
2585 mntput(m);
2586 }
2587 return new_ns;
2588 }
2589
2590 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2591 {
2592 struct mnt_namespace *ns;
2593 struct super_block *s;
2594 struct path path;
2595 int err;
2596
2597 ns = create_mnt_ns(mnt);
2598 if (IS_ERR(ns))
2599 return ERR_CAST(ns);
2600
2601 err = vfs_path_lookup(mnt->mnt_root, mnt,
2602 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2603
2604 put_mnt_ns(ns);
2605
2606 if (err)
2607 return ERR_PTR(err);
2608
2609 /* trade a vfsmount reference for active sb one */
2610 s = path.mnt->mnt_sb;
2611 atomic_inc(&s->s_active);
2612 mntput(path.mnt);
2613 /* lock the sucker */
2614 down_write(&s->s_umount);
2615 /* ... and return the root of (sub)tree on it */
2616 return path.dentry;
2617 }
2618 EXPORT_SYMBOL(mount_subtree);
2619
2620 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2621 char __user *, type, unsigned long, flags, void __user *, data)
2622 {
2623 int ret;
2624 char *kernel_type;
2625 struct filename *kernel_dir;
2626 char *kernel_dev;
2627 unsigned long data_page;
2628
2629 ret = copy_mount_string(type, &kernel_type);
2630 if (ret < 0)
2631 goto out_type;
2632
2633 kernel_dir = getname(dir_name);
2634 if (IS_ERR(kernel_dir)) {
2635 ret = PTR_ERR(kernel_dir);
2636 goto out_dir;
2637 }
2638
2639 ret = copy_mount_string(dev_name, &kernel_dev);
2640 if (ret < 0)
2641 goto out_dev;
2642
2643 ret = copy_mount_options(data, &data_page);
2644 if (ret < 0)
2645 goto out_data;
2646
2647 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2648 (void *) data_page);
2649
2650 free_page(data_page);
2651 out_data:
2652 kfree(kernel_dev);
2653 out_dev:
2654 putname(kernel_dir);
2655 out_dir:
2656 kfree(kernel_type);
2657 out_type:
2658 return ret;
2659 }
2660
2661 /*
2662 * Return true if path is reachable from root
2663 *
2664 * namespace_sem or mount_lock is held
2665 */
2666 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2667 const struct path *root)
2668 {
2669 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2670 dentry = mnt->mnt_mountpoint;
2671 mnt = mnt->mnt_parent;
2672 }
2673 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2674 }
2675
2676 int path_is_under(struct path *path1, struct path *path2)
2677 {
2678 int res;
2679 read_seqlock_excl(&mount_lock);
2680 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2681 read_sequnlock_excl(&mount_lock);
2682 return res;
2683 }
2684 EXPORT_SYMBOL(path_is_under);
2685
2686 /*
2687 * pivot_root Semantics:
2688 * Moves the root file system of the current process to the directory put_old,
2689 * makes new_root as the new root file system of the current process, and sets
2690 * root/cwd of all processes which had them on the current root to new_root.
2691 *
2692 * Restrictions:
2693 * The new_root and put_old must be directories, and must not be on the
2694 * same file system as the current process root. The put_old must be
2695 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2696 * pointed to by put_old must yield the same directory as new_root. No other
2697 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2698 *
2699 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2700 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2701 * in this situation.
2702 *
2703 * Notes:
2704 * - we don't move root/cwd if they are not at the root (reason: if something
2705 * cared enough to change them, it's probably wrong to force them elsewhere)
2706 * - it's okay to pick a root that isn't the root of a file system, e.g.
2707 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2708 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2709 * first.
2710 */
2711 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2712 const char __user *, put_old)
2713 {
2714 struct path new, old, parent_path, root_parent, root;
2715 struct mount *new_mnt, *root_mnt, *old_mnt;
2716 struct mountpoint *old_mp, *root_mp;
2717 int error;
2718
2719 if (!may_mount())
2720 return -EPERM;
2721
2722 error = user_path_dir(new_root, &new);
2723 if (error)
2724 goto out0;
2725
2726 error = user_path_dir(put_old, &old);
2727 if (error)
2728 goto out1;
2729
2730 error = security_sb_pivotroot(&old, &new);
2731 if (error)
2732 goto out2;
2733
2734 get_fs_root(current->fs, &root);
2735 old_mp = lock_mount(&old);
2736 error = PTR_ERR(old_mp);
2737 if (IS_ERR(old_mp))
2738 goto out3;
2739
2740 error = -EINVAL;
2741 new_mnt = real_mount(new.mnt);
2742 root_mnt = real_mount(root.mnt);
2743 old_mnt = real_mount(old.mnt);
2744 if (IS_MNT_SHARED(old_mnt) ||
2745 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2746 IS_MNT_SHARED(root_mnt->mnt_parent))
2747 goto out4;
2748 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2749 goto out4;
2750 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2751 goto out4;
2752 error = -ENOENT;
2753 if (d_unlinked(new.dentry))
2754 goto out4;
2755 error = -EBUSY;
2756 if (new_mnt == root_mnt || old_mnt == root_mnt)
2757 goto out4; /* loop, on the same file system */
2758 error = -EINVAL;
2759 if (root.mnt->mnt_root != root.dentry)
2760 goto out4; /* not a mountpoint */
2761 if (!mnt_has_parent(root_mnt))
2762 goto out4; /* not attached */
2763 root_mp = root_mnt->mnt_mp;
2764 if (new.mnt->mnt_root != new.dentry)
2765 goto out4; /* not a mountpoint */
2766 if (!mnt_has_parent(new_mnt))
2767 goto out4; /* not attached */
2768 /* make sure we can reach put_old from new_root */
2769 if (!is_path_reachable(old_mnt, old.dentry, &new))
2770 goto out4;
2771 root_mp->m_count++; /* pin it so it won't go away */
2772 lock_mount_hash();
2773 detach_mnt(new_mnt, &parent_path);
2774 detach_mnt(root_mnt, &root_parent);
2775 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2776 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2777 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2778 }
2779 /* mount old root on put_old */
2780 attach_mnt(root_mnt, old_mnt, old_mp);
2781 /* mount new_root on / */
2782 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2783 touch_mnt_namespace(current->nsproxy->mnt_ns);
2784 unlock_mount_hash();
2785 chroot_fs_refs(&root, &new);
2786 put_mountpoint(root_mp);
2787 error = 0;
2788 out4:
2789 unlock_mount(old_mp);
2790 if (!error) {
2791 path_put(&root_parent);
2792 path_put(&parent_path);
2793 }
2794 out3:
2795 path_put(&root);
2796 out2:
2797 path_put(&old);
2798 out1:
2799 path_put(&new);
2800 out0:
2801 return error;
2802 }
2803
2804 static void __init init_mount_tree(void)
2805 {
2806 struct vfsmount *mnt;
2807 struct mnt_namespace *ns;
2808 struct path root;
2809 struct file_system_type *type;
2810
2811 type = get_fs_type("rootfs");
2812 if (!type)
2813 panic("Can't find rootfs type");
2814 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2815 put_filesystem(type);
2816 if (IS_ERR(mnt))
2817 panic("Can't create rootfs");
2818
2819 ns = create_mnt_ns(mnt);
2820 if (IS_ERR(ns))
2821 panic("Can't allocate initial namespace");
2822
2823 init_task.nsproxy->mnt_ns = ns;
2824 get_mnt_ns(ns);
2825
2826 root.mnt = mnt;
2827 root.dentry = mnt->mnt_root;
2828
2829 set_fs_pwd(current->fs, &root);
2830 set_fs_root(current->fs, &root);
2831 }
2832
2833 void __init mnt_init(void)
2834 {
2835 unsigned u;
2836 int err;
2837
2838 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2839 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2840
2841 mount_hashtable = alloc_large_system_hash("Mount-cache",
2842 sizeof(struct hlist_head),
2843 mhash_entries, 19,
2844 0,
2845 &m_hash_shift, &m_hash_mask, 0, 0);
2846 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
2847 sizeof(struct hlist_head),
2848 mphash_entries, 19,
2849 0,
2850 &mp_hash_shift, &mp_hash_mask, 0, 0);
2851
2852 if (!mount_hashtable || !mountpoint_hashtable)
2853 panic("Failed to allocate mount hash table\n");
2854
2855 for (u = 0; u <= m_hash_mask; u++)
2856 INIT_HLIST_HEAD(&mount_hashtable[u]);
2857 for (u = 0; u <= mp_hash_mask; u++)
2858 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
2859
2860 kernfs_init();
2861
2862 err = sysfs_init();
2863 if (err)
2864 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2865 __func__, err);
2866 fs_kobj = kobject_create_and_add("fs", NULL);
2867 if (!fs_kobj)
2868 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2869 init_rootfs();
2870 init_mount_tree();
2871 }
2872
2873 void put_mnt_ns(struct mnt_namespace *ns)
2874 {
2875 if (!atomic_dec_and_test(&ns->count))
2876 return;
2877 drop_collected_mounts(&ns->root->mnt);
2878 free_mnt_ns(ns);
2879 }
2880
2881 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2882 {
2883 struct vfsmount *mnt;
2884 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2885 if (!IS_ERR(mnt)) {
2886 /*
2887 * it is a longterm mount, don't release mnt until
2888 * we unmount before file sys is unregistered
2889 */
2890 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2891 }
2892 return mnt;
2893 }
2894 EXPORT_SYMBOL_GPL(kern_mount_data);
2895
2896 void kern_unmount(struct vfsmount *mnt)
2897 {
2898 /* release long term mount so mount point can be released */
2899 if (!IS_ERR_OR_NULL(mnt)) {
2900 real_mount(mnt)->mnt_ns = NULL;
2901 synchronize_rcu(); /* yecchhh... */
2902 mntput(mnt);
2903 }
2904 }
2905 EXPORT_SYMBOL(kern_unmount);
2906
2907 bool our_mnt(struct vfsmount *mnt)
2908 {
2909 return check_mnt(real_mount(mnt));
2910 }
2911
2912 bool current_chrooted(void)
2913 {
2914 /* Does the current process have a non-standard root */
2915 struct path ns_root;
2916 struct path fs_root;
2917 bool chrooted;
2918
2919 /* Find the namespace root */
2920 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2921 ns_root.dentry = ns_root.mnt->mnt_root;
2922 path_get(&ns_root);
2923 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2924 ;
2925
2926 get_fs_root(current->fs, &fs_root);
2927
2928 chrooted = !path_equal(&fs_root, &ns_root);
2929
2930 path_put(&fs_root);
2931 path_put(&ns_root);
2932
2933 return chrooted;
2934 }
2935
2936 bool fs_fully_visible(struct file_system_type *type)
2937 {
2938 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2939 struct mount *mnt;
2940 bool visible = false;
2941
2942 if (unlikely(!ns))
2943 return false;
2944
2945 down_read(&namespace_sem);
2946 list_for_each_entry(mnt, &ns->list, mnt_list) {
2947 struct mount *child;
2948 if (mnt->mnt.mnt_sb->s_type != type)
2949 continue;
2950
2951 /* This mount is not fully visible if there are any child mounts
2952 * that cover anything except for empty directories.
2953 */
2954 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2955 struct inode *inode = child->mnt_mountpoint->d_inode;
2956 if (!S_ISDIR(inode->i_mode))
2957 goto next;
2958 if (inode->i_nlink > 2)
2959 goto next;
2960 }
2961 visible = true;
2962 goto found;
2963 next: ;
2964 }
2965 found:
2966 up_read(&namespace_sem);
2967 return visible;
2968 }
2969
2970 static void *mntns_get(struct task_struct *task)
2971 {
2972 struct mnt_namespace *ns = NULL;
2973 struct nsproxy *nsproxy;
2974
2975 rcu_read_lock();
2976 nsproxy = task_nsproxy(task);
2977 if (nsproxy) {
2978 ns = nsproxy->mnt_ns;
2979 get_mnt_ns(ns);
2980 }
2981 rcu_read_unlock();
2982
2983 return ns;
2984 }
2985
2986 static void mntns_put(void *ns)
2987 {
2988 put_mnt_ns(ns);
2989 }
2990
2991 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2992 {
2993 struct fs_struct *fs = current->fs;
2994 struct mnt_namespace *mnt_ns = ns;
2995 struct path root;
2996
2997 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2998 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2999 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3000 return -EPERM;
3001
3002 if (fs->users != 1)
3003 return -EINVAL;
3004
3005 get_mnt_ns(mnt_ns);
3006 put_mnt_ns(nsproxy->mnt_ns);
3007 nsproxy->mnt_ns = mnt_ns;
3008
3009 /* Find the root */
3010 root.mnt = &mnt_ns->root->mnt;
3011 root.dentry = mnt_ns->root->mnt.mnt_root;
3012 path_get(&root);
3013 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3014 ;
3015
3016 /* Update the pwd and root */
3017 set_fs_pwd(fs, &root);
3018 set_fs_root(fs, &root);
3019
3020 path_put(&root);
3021 return 0;
3022 }
3023
3024 static unsigned int mntns_inum(void *ns)
3025 {
3026 struct mnt_namespace *mnt_ns = ns;
3027 return mnt_ns->proc_inum;
3028 }
3029
3030 const struct proc_ns_operations mntns_operations = {
3031 .name = "mnt",
3032 .type = CLONE_NEWNS,
3033 .get = mntns_get,
3034 .put = mntns_put,
3035 .install = mntns_install,
3036 .inum = mntns_inum,
3037 };