]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/namespace.c
drm/nouveau/i2c: Disable i2c bus access after ->fini()
[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/cred.h>
19 #include <linux/idr.h>
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 <linux/task_work.h>
28 #include <linux/sched/task.h>
29
30 #include "pnode.h"
31 #include "internal.h"
32
33 /* Maximum number of mounts in a mount namespace */
34 unsigned int sysctl_mount_max __read_mostly = 100000;
35
36 static unsigned int m_hash_mask __read_mostly;
37 static unsigned int m_hash_shift __read_mostly;
38 static unsigned int mp_hash_mask __read_mostly;
39 static unsigned int mp_hash_shift __read_mostly;
40
41 static __initdata unsigned long mhash_entries;
42 static int __init set_mhash_entries(char *str)
43 {
44 if (!str)
45 return 0;
46 mhash_entries = simple_strtoul(str, &str, 0);
47 return 1;
48 }
49 __setup("mhash_entries=", set_mhash_entries);
50
51 static __initdata unsigned long mphash_entries;
52 static int __init set_mphash_entries(char *str)
53 {
54 if (!str)
55 return 0;
56 mphash_entries = simple_strtoul(str, &str, 0);
57 return 1;
58 }
59 __setup("mphash_entries=", set_mphash_entries);
60
61 static u64 event;
62 static DEFINE_IDA(mnt_id_ida);
63 static DEFINE_IDA(mnt_group_ida);
64 static DEFINE_SPINLOCK(mnt_id_lock);
65 static int mnt_id_start = 0;
66 static int mnt_group_start = 1;
67
68 static struct hlist_head *mount_hashtable __read_mostly;
69 static struct hlist_head *mountpoint_hashtable __read_mostly;
70 static struct kmem_cache *mnt_cache __read_mostly;
71 static DECLARE_RWSEM(namespace_sem);
72
73 /* /sys/fs */
74 struct kobject *fs_kobj;
75 EXPORT_SYMBOL_GPL(fs_kobj);
76
77 /*
78 * vfsmount lock may be taken for read to prevent changes to the
79 * vfsmount hash, ie. during mountpoint lookups or walking back
80 * up the tree.
81 *
82 * It should be taken for write in all cases where the vfsmount
83 * tree or hash is modified or when a vfsmount structure is modified.
84 */
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
86
87 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
88 {
89 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
90 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
91 tmp = tmp + (tmp >> m_hash_shift);
92 return &mount_hashtable[tmp & m_hash_mask];
93 }
94
95 static inline struct hlist_head *mp_hash(struct dentry *dentry)
96 {
97 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
98 tmp = tmp + (tmp >> mp_hash_shift);
99 return &mountpoint_hashtable[tmp & mp_hash_mask];
100 }
101
102 static int mnt_alloc_id(struct mount *mnt)
103 {
104 int res;
105
106 retry:
107 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
108 spin_lock(&mnt_id_lock);
109 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
110 if (!res)
111 mnt_id_start = mnt->mnt_id + 1;
112 spin_unlock(&mnt_id_lock);
113 if (res == -EAGAIN)
114 goto retry;
115
116 return res;
117 }
118
119 static void mnt_free_id(struct mount *mnt)
120 {
121 int id = mnt->mnt_id;
122 spin_lock(&mnt_id_lock);
123 ida_remove(&mnt_id_ida, id);
124 if (mnt_id_start > id)
125 mnt_id_start = id;
126 spin_unlock(&mnt_id_lock);
127 }
128
129 /*
130 * Allocate a new peer group ID
131 *
132 * mnt_group_ida is protected by namespace_sem
133 */
134 static int mnt_alloc_group_id(struct mount *mnt)
135 {
136 int res;
137
138 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 return -ENOMEM;
140
141 res = ida_get_new_above(&mnt_group_ida,
142 mnt_group_start,
143 &mnt->mnt_group_id);
144 if (!res)
145 mnt_group_start = mnt->mnt_group_id + 1;
146
147 return res;
148 }
149
150 /*
151 * Release a peer group ID
152 */
153 void mnt_release_group_id(struct mount *mnt)
154 {
155 int id = mnt->mnt_group_id;
156 ida_remove(&mnt_group_ida, id);
157 if (mnt_group_start > id)
158 mnt_group_start = id;
159 mnt->mnt_group_id = 0;
160 }
161
162 /*
163 * vfsmount lock must be held for read
164 */
165 static inline void mnt_add_count(struct mount *mnt, int n)
166 {
167 #ifdef CONFIG_SMP
168 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
169 #else
170 preempt_disable();
171 mnt->mnt_count += n;
172 preempt_enable();
173 #endif
174 }
175
176 /*
177 * vfsmount lock must be held for write
178 */
179 unsigned int mnt_get_count(struct mount *mnt)
180 {
181 #ifdef CONFIG_SMP
182 unsigned int count = 0;
183 int cpu;
184
185 for_each_possible_cpu(cpu) {
186 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
187 }
188
189 return count;
190 #else
191 return mnt->mnt_count;
192 #endif
193 }
194
195 static void drop_mountpoint(struct fs_pin *p)
196 {
197 struct mount *m = container_of(p, struct mount, mnt_umount);
198 dput(m->mnt_ex_mountpoint);
199 pin_remove(p);
200 mntput(&m->mnt);
201 }
202
203 static struct mount *alloc_vfsmnt(const char *name)
204 {
205 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
206 if (mnt) {
207 int err;
208
209 err = mnt_alloc_id(mnt);
210 if (err)
211 goto out_free_cache;
212
213 if (name) {
214 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
215 if (!mnt->mnt_devname)
216 goto out_free_id;
217 }
218
219 #ifdef CONFIG_SMP
220 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 if (!mnt->mnt_pcp)
222 goto out_free_devname;
223
224 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225 #else
226 mnt->mnt_count = 1;
227 mnt->mnt_writers = 0;
228 #endif
229
230 INIT_HLIST_NODE(&mnt->mnt_hash);
231 INIT_LIST_HEAD(&mnt->mnt_child);
232 INIT_LIST_HEAD(&mnt->mnt_mounts);
233 INIT_LIST_HEAD(&mnt->mnt_list);
234 INIT_LIST_HEAD(&mnt->mnt_expire);
235 INIT_LIST_HEAD(&mnt->mnt_share);
236 INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 INIT_LIST_HEAD(&mnt->mnt_slave);
238 INIT_HLIST_NODE(&mnt->mnt_mp_list);
239 INIT_LIST_HEAD(&mnt->mnt_umounting);
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 */
274 int __mnt_is_readonly(struct vfsmount *mnt)
275 {
276 if (mnt->mnt_flags & MNT_READONLY)
277 return 1;
278 if (sb_rdonly(mnt->mnt_sb))
279 return 1;
280 return 0;
281 }
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
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
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
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
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 */
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 (READ_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 */
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 */
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 */
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_path - 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 *
440 * Called by the vfs for cases when we have an open file at hand, but will do an
441 * inode operation on it (important distinction for files opened on overlayfs,
442 * since the file operations will come from the real underlying file, while
443 * inode operations come from the overlay).
444 */
445 int mnt_want_write_file_path(struct file *file)
446 {
447 int ret;
448
449 sb_start_write(file->f_path.mnt->mnt_sb);
450 ret = __mnt_want_write_file(file);
451 if (ret)
452 sb_end_write(file->f_path.mnt->mnt_sb);
453 return ret;
454 }
455
456 static inline int may_write_real(struct file *file)
457 {
458 struct dentry *dentry = file->f_path.dentry;
459 struct dentry *upperdentry;
460
461 /* Writable file? */
462 if (file->f_mode & FMODE_WRITER)
463 return 0;
464
465 /* Not overlayfs? */
466 if (likely(!(dentry->d_flags & DCACHE_OP_REAL)))
467 return 0;
468
469 /* File refers to upper, writable layer? */
470 upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
471 if (upperdentry &&
472 (file_inode(file) == d_inode(upperdentry) ||
473 file_inode(file) == d_inode(dentry)))
474 return 0;
475
476 /* Lower layer: can't write to real file, sorry... */
477 return -EPERM;
478 }
479
480 /**
481 * mnt_want_write_file - get write access to a file's mount
482 * @file: the file who's mount on which to take a write
483 *
484 * This is like mnt_want_write, but it takes a file and can
485 * do some optimisations if the file is open for write already
486 *
487 * Mostly called by filesystems from their ioctl operation before performing
488 * modification. On overlayfs this needs to check if the file is on a read-only
489 * lower layer and deny access in that case.
490 */
491 int mnt_want_write_file(struct file *file)
492 {
493 int ret;
494
495 ret = may_write_real(file);
496 if (!ret) {
497 sb_start_write(file_inode(file)->i_sb);
498 ret = __mnt_want_write_file(file);
499 if (ret)
500 sb_end_write(file_inode(file)->i_sb);
501 }
502 return ret;
503 }
504 EXPORT_SYMBOL_GPL(mnt_want_write_file);
505
506 /**
507 * __mnt_drop_write - give up write access to a mount
508 * @mnt: the mount on which to give up write access
509 *
510 * Tells the low-level filesystem that we are done
511 * performing writes to it. Must be matched with
512 * __mnt_want_write() call above.
513 */
514 void __mnt_drop_write(struct vfsmount *mnt)
515 {
516 preempt_disable();
517 mnt_dec_writers(real_mount(mnt));
518 preempt_enable();
519 }
520 EXPORT_SYMBOL_GPL(__mnt_drop_write);
521
522 /**
523 * mnt_drop_write - give up write access to a mount
524 * @mnt: the mount on which to give up write access
525 *
526 * Tells the low-level filesystem that we are done performing writes to it and
527 * also allows filesystem to be frozen again. Must be matched with
528 * mnt_want_write() call above.
529 */
530 void mnt_drop_write(struct vfsmount *mnt)
531 {
532 __mnt_drop_write(mnt);
533 sb_end_write(mnt->mnt_sb);
534 }
535 EXPORT_SYMBOL_GPL(mnt_drop_write);
536
537 void __mnt_drop_write_file(struct file *file)
538 {
539 __mnt_drop_write(file->f_path.mnt);
540 }
541
542 void mnt_drop_write_file_path(struct file *file)
543 {
544 mnt_drop_write(file->f_path.mnt);
545 }
546
547 void mnt_drop_write_file(struct file *file)
548 {
549 __mnt_drop_write(file->f_path.mnt);
550 sb_end_write(file_inode(file)->i_sb);
551 }
552 EXPORT_SYMBOL(mnt_drop_write_file);
553
554 static int mnt_make_readonly(struct mount *mnt)
555 {
556 int ret = 0;
557
558 lock_mount_hash();
559 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
560 /*
561 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
562 * should be visible before we do.
563 */
564 smp_mb();
565
566 /*
567 * With writers on hold, if this value is zero, then there are
568 * definitely no active writers (although held writers may subsequently
569 * increment the count, they'll have to wait, and decrement it after
570 * seeing MNT_READONLY).
571 *
572 * It is OK to have counter incremented on one CPU and decremented on
573 * another: the sum will add up correctly. The danger would be when we
574 * sum up each counter, if we read a counter before it is incremented,
575 * but then read another CPU's count which it has been subsequently
576 * decremented from -- we would see more decrements than we should.
577 * MNT_WRITE_HOLD protects against this scenario, because
578 * mnt_want_write first increments count, then smp_mb, then spins on
579 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
580 * we're counting up here.
581 */
582 if (mnt_get_writers(mnt) > 0)
583 ret = -EBUSY;
584 else
585 mnt->mnt.mnt_flags |= MNT_READONLY;
586 /*
587 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
588 * that become unheld will see MNT_READONLY.
589 */
590 smp_wmb();
591 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
592 unlock_mount_hash();
593 return ret;
594 }
595
596 static void __mnt_unmake_readonly(struct mount *mnt)
597 {
598 lock_mount_hash();
599 mnt->mnt.mnt_flags &= ~MNT_READONLY;
600 unlock_mount_hash();
601 }
602
603 int sb_prepare_remount_readonly(struct super_block *sb)
604 {
605 struct mount *mnt;
606 int err = 0;
607
608 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
609 if (atomic_long_read(&sb->s_remove_count))
610 return -EBUSY;
611
612 lock_mount_hash();
613 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
614 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
615 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
616 smp_mb();
617 if (mnt_get_writers(mnt) > 0) {
618 err = -EBUSY;
619 break;
620 }
621 }
622 }
623 if (!err && atomic_long_read(&sb->s_remove_count))
624 err = -EBUSY;
625
626 if (!err) {
627 sb->s_readonly_remount = 1;
628 smp_wmb();
629 }
630 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
631 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
632 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
633 }
634 unlock_mount_hash();
635
636 return err;
637 }
638
639 static void free_vfsmnt(struct mount *mnt)
640 {
641 kfree_const(mnt->mnt_devname);
642 #ifdef CONFIG_SMP
643 free_percpu(mnt->mnt_pcp);
644 #endif
645 kmem_cache_free(mnt_cache, mnt);
646 }
647
648 static void delayed_free_vfsmnt(struct rcu_head *head)
649 {
650 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
651 }
652
653 /* call under rcu_read_lock */
654 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
655 {
656 struct mount *mnt;
657 if (read_seqretry(&mount_lock, seq))
658 return 1;
659 if (bastard == NULL)
660 return 0;
661 mnt = real_mount(bastard);
662 mnt_add_count(mnt, 1);
663 smp_mb(); // see mntput_no_expire()
664 if (likely(!read_seqretry(&mount_lock, seq)))
665 return 0;
666 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
667 mnt_add_count(mnt, -1);
668 return 1;
669 }
670 lock_mount_hash();
671 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
672 mnt_add_count(mnt, -1);
673 unlock_mount_hash();
674 return 1;
675 }
676 unlock_mount_hash();
677 /* caller will mntput() */
678 return -1;
679 }
680
681 /* call under rcu_read_lock */
682 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
683 {
684 int res = __legitimize_mnt(bastard, seq);
685 if (likely(!res))
686 return true;
687 if (unlikely(res < 0)) {
688 rcu_read_unlock();
689 mntput(bastard);
690 rcu_read_lock();
691 }
692 return false;
693 }
694
695 /*
696 * find the first mount at @dentry on vfsmount @mnt.
697 * call under rcu_read_lock()
698 */
699 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
700 {
701 struct hlist_head *head = m_hash(mnt, dentry);
702 struct mount *p;
703
704 hlist_for_each_entry_rcu(p, head, mnt_hash)
705 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
706 return p;
707 return NULL;
708 }
709
710 /*
711 * lookup_mnt - Return the first child mount mounted at path
712 *
713 * "First" means first mounted chronologically. If you create the
714 * following mounts:
715 *
716 * mount /dev/sda1 /mnt
717 * mount /dev/sda2 /mnt
718 * mount /dev/sda3 /mnt
719 *
720 * Then lookup_mnt() on the base /mnt dentry in the root mount will
721 * return successively the root dentry and vfsmount of /dev/sda1, then
722 * /dev/sda2, then /dev/sda3, then NULL.
723 *
724 * lookup_mnt takes a reference to the found vfsmount.
725 */
726 struct vfsmount *lookup_mnt(const struct path *path)
727 {
728 struct mount *child_mnt;
729 struct vfsmount *m;
730 unsigned seq;
731
732 rcu_read_lock();
733 do {
734 seq = read_seqbegin(&mount_lock);
735 child_mnt = __lookup_mnt(path->mnt, path->dentry);
736 m = child_mnt ? &child_mnt->mnt : NULL;
737 } while (!legitimize_mnt(m, seq));
738 rcu_read_unlock();
739 return m;
740 }
741
742 /*
743 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
744 * current mount namespace.
745 *
746 * The common case is dentries are not mountpoints at all and that
747 * test is handled inline. For the slow case when we are actually
748 * dealing with a mountpoint of some kind, walk through all of the
749 * mounts in the current mount namespace and test to see if the dentry
750 * is a mountpoint.
751 *
752 * The mount_hashtable is not usable in the context because we
753 * need to identify all mounts that may be in the current mount
754 * namespace not just a mount that happens to have some specified
755 * parent mount.
756 */
757 bool __is_local_mountpoint(struct dentry *dentry)
758 {
759 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
760 struct mount *mnt;
761 bool is_covered = false;
762
763 if (!d_mountpoint(dentry))
764 goto out;
765
766 down_read(&namespace_sem);
767 list_for_each_entry(mnt, &ns->list, mnt_list) {
768 is_covered = (mnt->mnt_mountpoint == dentry);
769 if (is_covered)
770 break;
771 }
772 up_read(&namespace_sem);
773 out:
774 return is_covered;
775 }
776
777 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
778 {
779 struct hlist_head *chain = mp_hash(dentry);
780 struct mountpoint *mp;
781
782 hlist_for_each_entry(mp, chain, m_hash) {
783 if (mp->m_dentry == dentry) {
784 mp->m_count++;
785 return mp;
786 }
787 }
788 return NULL;
789 }
790
791 static struct mountpoint *get_mountpoint(struct dentry *dentry)
792 {
793 struct mountpoint *mp, *new = NULL;
794 int ret;
795
796 if (d_mountpoint(dentry)) {
797 /* might be worth a WARN_ON() */
798 if (d_unlinked(dentry))
799 return ERR_PTR(-ENOENT);
800 mountpoint:
801 read_seqlock_excl(&mount_lock);
802 mp = lookup_mountpoint(dentry);
803 read_sequnlock_excl(&mount_lock);
804 if (mp)
805 goto done;
806 }
807
808 if (!new)
809 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
810 if (!new)
811 return ERR_PTR(-ENOMEM);
812
813
814 /* Exactly one processes may set d_mounted */
815 ret = d_set_mounted(dentry);
816
817 /* Someone else set d_mounted? */
818 if (ret == -EBUSY)
819 goto mountpoint;
820
821 /* The dentry is not available as a mountpoint? */
822 mp = ERR_PTR(ret);
823 if (ret)
824 goto done;
825
826 /* Add the new mountpoint to the hash table */
827 read_seqlock_excl(&mount_lock);
828 new->m_dentry = dentry;
829 new->m_count = 1;
830 hlist_add_head(&new->m_hash, mp_hash(dentry));
831 INIT_HLIST_HEAD(&new->m_list);
832 read_sequnlock_excl(&mount_lock);
833
834 mp = new;
835 new = NULL;
836 done:
837 kfree(new);
838 return mp;
839 }
840
841 static void put_mountpoint(struct mountpoint *mp)
842 {
843 if (!--mp->m_count) {
844 struct dentry *dentry = mp->m_dentry;
845 BUG_ON(!hlist_empty(&mp->m_list));
846 spin_lock(&dentry->d_lock);
847 dentry->d_flags &= ~DCACHE_MOUNTED;
848 spin_unlock(&dentry->d_lock);
849 hlist_del(&mp->m_hash);
850 kfree(mp);
851 }
852 }
853
854 static inline int check_mnt(struct mount *mnt)
855 {
856 return mnt->mnt_ns == current->nsproxy->mnt_ns;
857 }
858
859 /* for aufs, CONFIG_AUFS_BR_FUSE */
860 int is_current_mnt_ns(struct vfsmount *mnt)
861 {
862 return check_mnt(real_mount(mnt));
863 }
864 EXPORT_SYMBOL_GPL(is_current_mnt_ns);
865
866 /*
867 * vfsmount lock must be held for write
868 */
869 static void touch_mnt_namespace(struct mnt_namespace *ns)
870 {
871 if (ns) {
872 ns->event = ++event;
873 wake_up_interruptible(&ns->poll);
874 }
875 }
876
877 /*
878 * vfsmount lock must be held for write
879 */
880 static void __touch_mnt_namespace(struct mnt_namespace *ns)
881 {
882 if (ns && ns->event != event) {
883 ns->event = event;
884 wake_up_interruptible(&ns->poll);
885 }
886 }
887
888 /*
889 * vfsmount lock must be held for write
890 */
891 static void unhash_mnt(struct mount *mnt)
892 {
893 mnt->mnt_parent = mnt;
894 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
895 list_del_init(&mnt->mnt_child);
896 hlist_del_init_rcu(&mnt->mnt_hash);
897 hlist_del_init(&mnt->mnt_mp_list);
898 put_mountpoint(mnt->mnt_mp);
899 mnt->mnt_mp = NULL;
900 }
901
902 /*
903 * vfsmount lock must be held for write
904 */
905 static void detach_mnt(struct mount *mnt, struct path *old_path)
906 {
907 old_path->dentry = mnt->mnt_mountpoint;
908 old_path->mnt = &mnt->mnt_parent->mnt;
909 unhash_mnt(mnt);
910 }
911
912 /*
913 * vfsmount lock must be held for write
914 */
915 static void umount_mnt(struct mount *mnt)
916 {
917 /* old mountpoint will be dropped when we can do that */
918 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
919 unhash_mnt(mnt);
920 }
921
922 /*
923 * vfsmount lock must be held for write
924 */
925 void mnt_set_mountpoint(struct mount *mnt,
926 struct mountpoint *mp,
927 struct mount *child_mnt)
928 {
929 mp->m_count++;
930 mnt_add_count(mnt, 1); /* essentially, that's mntget */
931 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
932 child_mnt->mnt_parent = mnt;
933 child_mnt->mnt_mp = mp;
934 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
935 }
936
937 static void __attach_mnt(struct mount *mnt, struct mount *parent)
938 {
939 hlist_add_head_rcu(&mnt->mnt_hash,
940 m_hash(&parent->mnt, mnt->mnt_mountpoint));
941 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
942 }
943
944 /*
945 * vfsmount lock must be held for write
946 */
947 static void attach_mnt(struct mount *mnt,
948 struct mount *parent,
949 struct mountpoint *mp)
950 {
951 mnt_set_mountpoint(parent, mp, mnt);
952 __attach_mnt(mnt, parent);
953 }
954
955 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
956 {
957 struct mountpoint *old_mp = mnt->mnt_mp;
958 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
959 struct mount *old_parent = mnt->mnt_parent;
960
961 list_del_init(&mnt->mnt_child);
962 hlist_del_init(&mnt->mnt_mp_list);
963 hlist_del_init_rcu(&mnt->mnt_hash);
964
965 attach_mnt(mnt, parent, mp);
966
967 put_mountpoint(old_mp);
968
969 /*
970 * Safely avoid even the suggestion this code might sleep or
971 * lock the mount hash by taking advantage of the knowledge that
972 * mnt_change_mountpoint will not release the final reference
973 * to a mountpoint.
974 *
975 * During mounting, the mount passed in as the parent mount will
976 * continue to use the old mountpoint and during unmounting, the
977 * old mountpoint will continue to exist until namespace_unlock,
978 * which happens well after mnt_change_mountpoint.
979 */
980 spin_lock(&old_mountpoint->d_lock);
981 old_mountpoint->d_lockref.count--;
982 spin_unlock(&old_mountpoint->d_lock);
983
984 mnt_add_count(old_parent, -1);
985 }
986
987 /*
988 * vfsmount lock must be held for write
989 */
990 static void commit_tree(struct mount *mnt)
991 {
992 struct mount *parent = mnt->mnt_parent;
993 struct mount *m;
994 LIST_HEAD(head);
995 struct mnt_namespace *n = parent->mnt_ns;
996
997 BUG_ON(parent == mnt);
998
999 list_add_tail(&head, &mnt->mnt_list);
1000 list_for_each_entry(m, &head, mnt_list)
1001 m->mnt_ns = n;
1002
1003 list_splice(&head, n->list.prev);
1004
1005 n->mounts += n->pending_mounts;
1006 n->pending_mounts = 0;
1007
1008 __attach_mnt(mnt, parent);
1009 touch_mnt_namespace(n);
1010 }
1011
1012 static struct mount *next_mnt(struct mount *p, struct mount *root)
1013 {
1014 struct list_head *next = p->mnt_mounts.next;
1015 if (next == &p->mnt_mounts) {
1016 while (1) {
1017 if (p == root)
1018 return NULL;
1019 next = p->mnt_child.next;
1020 if (next != &p->mnt_parent->mnt_mounts)
1021 break;
1022 p = p->mnt_parent;
1023 }
1024 }
1025 return list_entry(next, struct mount, mnt_child);
1026 }
1027
1028 static struct mount *skip_mnt_tree(struct mount *p)
1029 {
1030 struct list_head *prev = p->mnt_mounts.prev;
1031 while (prev != &p->mnt_mounts) {
1032 p = list_entry(prev, struct mount, mnt_child);
1033 prev = p->mnt_mounts.prev;
1034 }
1035 return p;
1036 }
1037
1038 struct vfsmount *
1039 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
1040 {
1041 struct mount *mnt;
1042 struct dentry *root;
1043
1044 if (!type)
1045 return ERR_PTR(-ENODEV);
1046
1047 mnt = alloc_vfsmnt(name);
1048 if (!mnt)
1049 return ERR_PTR(-ENOMEM);
1050
1051 if (flags & SB_KERNMOUNT)
1052 mnt->mnt.mnt_flags = MNT_INTERNAL;
1053
1054 root = mount_fs(type, flags, name, data);
1055 if (IS_ERR(root)) {
1056 mnt_free_id(mnt);
1057 free_vfsmnt(mnt);
1058 return ERR_CAST(root);
1059 }
1060
1061 mnt->mnt.mnt_root = root;
1062 mnt->mnt.mnt_sb = root->d_sb;
1063 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1064 mnt->mnt_parent = mnt;
1065 lock_mount_hash();
1066 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1067 unlock_mount_hash();
1068 return &mnt->mnt;
1069 }
1070 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1071
1072 struct vfsmount *
1073 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1074 const char *name, void *data)
1075 {
1076 /* Until it is worked out how to pass the user namespace
1077 * through from the parent mount to the submount don't support
1078 * unprivileged mounts with submounts.
1079 */
1080 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1081 return ERR_PTR(-EPERM);
1082
1083 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1084 }
1085 EXPORT_SYMBOL_GPL(vfs_submount);
1086
1087 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1088 int flag)
1089 {
1090 struct super_block *sb = old->mnt.mnt_sb;
1091 struct mount *mnt;
1092 int err;
1093
1094 mnt = alloc_vfsmnt(old->mnt_devname);
1095 if (!mnt)
1096 return ERR_PTR(-ENOMEM);
1097
1098 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1099 mnt->mnt_group_id = 0; /* not a peer of original */
1100 else
1101 mnt->mnt_group_id = old->mnt_group_id;
1102
1103 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1104 err = mnt_alloc_group_id(mnt);
1105 if (err)
1106 goto out_free;
1107 }
1108
1109 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1110 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1111 /* Don't allow unprivileged users to change mount flags */
1112 if (flag & CL_UNPRIVILEGED) {
1113 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1114
1115 if (mnt->mnt.mnt_flags & MNT_READONLY)
1116 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1117
1118 if (mnt->mnt.mnt_flags & MNT_NODEV)
1119 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1120
1121 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1122 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1123
1124 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1125 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1126 }
1127
1128 /* Don't allow unprivileged users to reveal what is under a mount */
1129 if ((flag & CL_UNPRIVILEGED) &&
1130 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1131 mnt->mnt.mnt_flags |= MNT_LOCKED;
1132
1133 atomic_inc(&sb->s_active);
1134 mnt->mnt.mnt_sb = sb;
1135 mnt->mnt.mnt_root = dget(root);
1136 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1137 mnt->mnt_parent = mnt;
1138 lock_mount_hash();
1139 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1140 unlock_mount_hash();
1141
1142 if ((flag & CL_SLAVE) ||
1143 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1144 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1145 mnt->mnt_master = old;
1146 CLEAR_MNT_SHARED(mnt);
1147 } else if (!(flag & CL_PRIVATE)) {
1148 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1149 list_add(&mnt->mnt_share, &old->mnt_share);
1150 if (IS_MNT_SLAVE(old))
1151 list_add(&mnt->mnt_slave, &old->mnt_slave);
1152 mnt->mnt_master = old->mnt_master;
1153 } else {
1154 CLEAR_MNT_SHARED(mnt);
1155 }
1156 if (flag & CL_MAKE_SHARED)
1157 set_mnt_shared(mnt);
1158
1159 /* stick the duplicate mount on the same expiry list
1160 * as the original if that was on one */
1161 if (flag & CL_EXPIRE) {
1162 if (!list_empty(&old->mnt_expire))
1163 list_add(&mnt->mnt_expire, &old->mnt_expire);
1164 }
1165
1166 return mnt;
1167
1168 out_free:
1169 mnt_free_id(mnt);
1170 free_vfsmnt(mnt);
1171 return ERR_PTR(err);
1172 }
1173
1174 static void cleanup_mnt(struct mount *mnt)
1175 {
1176 /*
1177 * This probably indicates that somebody messed
1178 * up a mnt_want/drop_write() pair. If this
1179 * happens, the filesystem was probably unable
1180 * to make r/w->r/o transitions.
1181 */
1182 /*
1183 * The locking used to deal with mnt_count decrement provides barriers,
1184 * so mnt_get_writers() below is safe.
1185 */
1186 WARN_ON(mnt_get_writers(mnt));
1187 if (unlikely(mnt->mnt_pins.first))
1188 mnt_pin_kill(mnt);
1189 fsnotify_vfsmount_delete(&mnt->mnt);
1190 dput(mnt->mnt.mnt_root);
1191 deactivate_super(mnt->mnt.mnt_sb);
1192 mnt_free_id(mnt);
1193 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1194 }
1195
1196 static void __cleanup_mnt(struct rcu_head *head)
1197 {
1198 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1199 }
1200
1201 static LLIST_HEAD(delayed_mntput_list);
1202 static void delayed_mntput(struct work_struct *unused)
1203 {
1204 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1205 struct mount *m, *t;
1206
1207 llist_for_each_entry_safe(m, t, node, mnt_llist)
1208 cleanup_mnt(m);
1209 }
1210 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1211
1212 static void mntput_no_expire(struct mount *mnt)
1213 {
1214 rcu_read_lock();
1215 if (likely(READ_ONCE(mnt->mnt_ns))) {
1216 /*
1217 * Since we don't do lock_mount_hash() here,
1218 * ->mnt_ns can change under us. However, if it's
1219 * non-NULL, then there's a reference that won't
1220 * be dropped until after an RCU delay done after
1221 * turning ->mnt_ns NULL. So if we observe it
1222 * non-NULL under rcu_read_lock(), the reference
1223 * we are dropping is not the final one.
1224 */
1225 mnt_add_count(mnt, -1);
1226 rcu_read_unlock();
1227 return;
1228 }
1229 lock_mount_hash();
1230 /*
1231 * make sure that if __legitimize_mnt() has not seen us grab
1232 * mount_lock, we'll see their refcount increment here.
1233 */
1234 smp_mb();
1235 mnt_add_count(mnt, -1);
1236 if (mnt_get_count(mnt)) {
1237 rcu_read_unlock();
1238 unlock_mount_hash();
1239 return;
1240 }
1241 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1242 rcu_read_unlock();
1243 unlock_mount_hash();
1244 return;
1245 }
1246 mnt->mnt.mnt_flags |= MNT_DOOMED;
1247 rcu_read_unlock();
1248
1249 list_del(&mnt->mnt_instance);
1250
1251 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1252 struct mount *p, *tmp;
1253 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1254 umount_mnt(p);
1255 }
1256 }
1257 unlock_mount_hash();
1258
1259 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1260 struct task_struct *task = current;
1261 if (likely(!(task->flags & PF_KTHREAD))) {
1262 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1263 if (!task_work_add(task, &mnt->mnt_rcu, true))
1264 return;
1265 }
1266 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1267 schedule_delayed_work(&delayed_mntput_work, 1);
1268 return;
1269 }
1270 cleanup_mnt(mnt);
1271 }
1272
1273 void mntput(struct vfsmount *mnt)
1274 {
1275 if (mnt) {
1276 struct mount *m = real_mount(mnt);
1277 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1278 if (unlikely(m->mnt_expiry_mark))
1279 m->mnt_expiry_mark = 0;
1280 mntput_no_expire(m);
1281 }
1282 }
1283 EXPORT_SYMBOL(mntput);
1284
1285 struct vfsmount *mntget(struct vfsmount *mnt)
1286 {
1287 if (mnt)
1288 mnt_add_count(real_mount(mnt), 1);
1289 return mnt;
1290 }
1291 EXPORT_SYMBOL(mntget);
1292
1293 /* path_is_mountpoint() - Check if path is a mount in the current
1294 * namespace.
1295 *
1296 * d_mountpoint() can only be used reliably to establish if a dentry is
1297 * not mounted in any namespace and that common case is handled inline.
1298 * d_mountpoint() isn't aware of the possibility there may be multiple
1299 * mounts using a given dentry in a different namespace. This function
1300 * checks if the passed in path is a mountpoint rather than the dentry
1301 * alone.
1302 */
1303 bool path_is_mountpoint(const struct path *path)
1304 {
1305 unsigned seq;
1306 bool res;
1307
1308 if (!d_mountpoint(path->dentry))
1309 return false;
1310
1311 rcu_read_lock();
1312 do {
1313 seq = read_seqbegin(&mount_lock);
1314 res = __path_is_mountpoint(path);
1315 } while (read_seqretry(&mount_lock, seq));
1316 rcu_read_unlock();
1317
1318 return res;
1319 }
1320 EXPORT_SYMBOL(path_is_mountpoint);
1321
1322 struct vfsmount *mnt_clone_internal(const struct path *path)
1323 {
1324 struct mount *p;
1325 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1326 if (IS_ERR(p))
1327 return ERR_CAST(p);
1328 p->mnt.mnt_flags |= MNT_INTERNAL;
1329 return &p->mnt;
1330 }
1331
1332 #ifdef CONFIG_PROC_FS
1333 /* iterator; we want it to have access to namespace_sem, thus here... */
1334 static void *m_start(struct seq_file *m, loff_t *pos)
1335 {
1336 struct proc_mounts *p = m->private;
1337
1338 down_read(&namespace_sem);
1339 if (p->cached_event == p->ns->event) {
1340 void *v = p->cached_mount;
1341 if (*pos == p->cached_index)
1342 return v;
1343 if (*pos == p->cached_index + 1) {
1344 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1345 return p->cached_mount = v;
1346 }
1347 }
1348
1349 p->cached_event = p->ns->event;
1350 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1351 p->cached_index = *pos;
1352 return p->cached_mount;
1353 }
1354
1355 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1356 {
1357 struct proc_mounts *p = m->private;
1358
1359 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1360 p->cached_index = *pos;
1361 return p->cached_mount;
1362 }
1363
1364 static void m_stop(struct seq_file *m, void *v)
1365 {
1366 up_read(&namespace_sem);
1367 }
1368
1369 static int m_show(struct seq_file *m, void *v)
1370 {
1371 struct proc_mounts *p = m->private;
1372 struct mount *r = list_entry(v, struct mount, mnt_list);
1373 return p->show(m, &r->mnt);
1374 }
1375
1376 const struct seq_operations mounts_op = {
1377 .start = m_start,
1378 .next = m_next,
1379 .stop = m_stop,
1380 .show = m_show,
1381 };
1382 #endif /* CONFIG_PROC_FS */
1383
1384 /**
1385 * may_umount_tree - check if a mount tree is busy
1386 * @mnt: root of mount tree
1387 *
1388 * This is called to check if a tree of mounts has any
1389 * open files, pwds, chroots or sub mounts that are
1390 * busy.
1391 */
1392 int may_umount_tree(struct vfsmount *m)
1393 {
1394 struct mount *mnt = real_mount(m);
1395 int actual_refs = 0;
1396 int minimum_refs = 0;
1397 struct mount *p;
1398 BUG_ON(!m);
1399
1400 /* write lock needed for mnt_get_count */
1401 lock_mount_hash();
1402 for (p = mnt; p; p = next_mnt(p, mnt)) {
1403 actual_refs += mnt_get_count(p);
1404 minimum_refs += 2;
1405 }
1406 unlock_mount_hash();
1407
1408 if (actual_refs > minimum_refs)
1409 return 0;
1410
1411 return 1;
1412 }
1413
1414 EXPORT_SYMBOL(may_umount_tree);
1415
1416 /**
1417 * may_umount - check if a mount point is busy
1418 * @mnt: root of mount
1419 *
1420 * This is called to check if a mount point has any
1421 * open files, pwds, chroots or sub mounts. If the
1422 * mount has sub mounts this will return busy
1423 * regardless of whether the sub mounts are busy.
1424 *
1425 * Doesn't take quota and stuff into account. IOW, in some cases it will
1426 * give false negatives. The main reason why it's here is that we need
1427 * a non-destructive way to look for easily umountable filesystems.
1428 */
1429 int may_umount(struct vfsmount *mnt)
1430 {
1431 int ret = 1;
1432 down_read(&namespace_sem);
1433 lock_mount_hash();
1434 if (propagate_mount_busy(real_mount(mnt), 2))
1435 ret = 0;
1436 unlock_mount_hash();
1437 up_read(&namespace_sem);
1438 return ret;
1439 }
1440
1441 EXPORT_SYMBOL(may_umount);
1442
1443 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1444
1445 static void namespace_unlock(void)
1446 {
1447 struct hlist_head head;
1448
1449 hlist_move_list(&unmounted, &head);
1450
1451 up_write(&namespace_sem);
1452
1453 if (likely(hlist_empty(&head)))
1454 return;
1455
1456 synchronize_rcu();
1457
1458 group_pin_kill(&head);
1459 }
1460
1461 static inline void namespace_lock(void)
1462 {
1463 down_write(&namespace_sem);
1464 }
1465
1466 enum umount_tree_flags {
1467 UMOUNT_SYNC = 1,
1468 UMOUNT_PROPAGATE = 2,
1469 UMOUNT_CONNECTED = 4,
1470 };
1471
1472 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1473 {
1474 /* Leaving mounts connected is only valid for lazy umounts */
1475 if (how & UMOUNT_SYNC)
1476 return true;
1477
1478 /* A mount without a parent has nothing to be connected to */
1479 if (!mnt_has_parent(mnt))
1480 return true;
1481
1482 /* Because the reference counting rules change when mounts are
1483 * unmounted and connected, umounted mounts may not be
1484 * connected to mounted mounts.
1485 */
1486 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1487 return true;
1488
1489 /* Has it been requested that the mount remain connected? */
1490 if (how & UMOUNT_CONNECTED)
1491 return false;
1492
1493 /* Is the mount locked such that it needs to remain connected? */
1494 if (IS_MNT_LOCKED(mnt))
1495 return false;
1496
1497 /* By default disconnect the mount */
1498 return true;
1499 }
1500
1501 /*
1502 * mount_lock must be held
1503 * namespace_sem must be held for write
1504 */
1505 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1506 {
1507 LIST_HEAD(tmp_list);
1508 struct mount *p;
1509
1510 if (how & UMOUNT_PROPAGATE)
1511 propagate_mount_unlock(mnt);
1512
1513 /* Gather the mounts to umount */
1514 for (p = mnt; p; p = next_mnt(p, mnt)) {
1515 p->mnt.mnt_flags |= MNT_UMOUNT;
1516 list_move(&p->mnt_list, &tmp_list);
1517 }
1518
1519 /* Hide the mounts from mnt_mounts */
1520 list_for_each_entry(p, &tmp_list, mnt_list) {
1521 list_del_init(&p->mnt_child);
1522 }
1523
1524 /* Add propogated mounts to the tmp_list */
1525 if (how & UMOUNT_PROPAGATE)
1526 propagate_umount(&tmp_list);
1527
1528 while (!list_empty(&tmp_list)) {
1529 struct mnt_namespace *ns;
1530 bool disconnect;
1531 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1532 list_del_init(&p->mnt_expire);
1533 list_del_init(&p->mnt_list);
1534 ns = p->mnt_ns;
1535 if (ns) {
1536 ns->mounts--;
1537 __touch_mnt_namespace(ns);
1538 }
1539 p->mnt_ns = NULL;
1540 if (how & UMOUNT_SYNC)
1541 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1542
1543 disconnect = disconnect_mount(p, how);
1544
1545 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1546 disconnect ? &unmounted : NULL);
1547 if (mnt_has_parent(p)) {
1548 mnt_add_count(p->mnt_parent, -1);
1549 if (!disconnect) {
1550 /* Don't forget about p */
1551 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1552 } else {
1553 umount_mnt(p);
1554 }
1555 }
1556 change_mnt_propagation(p, MS_PRIVATE);
1557 }
1558 }
1559
1560 static void shrink_submounts(struct mount *mnt);
1561
1562 static int do_umount(struct mount *mnt, int flags)
1563 {
1564 struct super_block *sb = mnt->mnt.mnt_sb;
1565 int retval;
1566
1567 retval = security_sb_umount(&mnt->mnt, flags);
1568 if (retval)
1569 return retval;
1570
1571 /*
1572 * Allow userspace to request a mountpoint be expired rather than
1573 * unmounting unconditionally. Unmount only happens if:
1574 * (1) the mark is already set (the mark is cleared by mntput())
1575 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1576 */
1577 if (flags & MNT_EXPIRE) {
1578 if (&mnt->mnt == current->fs->root.mnt ||
1579 flags & (MNT_FORCE | MNT_DETACH))
1580 return -EINVAL;
1581
1582 /*
1583 * probably don't strictly need the lock here if we examined
1584 * all race cases, but it's a slowpath.
1585 */
1586 lock_mount_hash();
1587 if (mnt_get_count(mnt) != 2) {
1588 unlock_mount_hash();
1589 return -EBUSY;
1590 }
1591 unlock_mount_hash();
1592
1593 if (!xchg(&mnt->mnt_expiry_mark, 1))
1594 return -EAGAIN;
1595 }
1596
1597 /*
1598 * If we may have to abort operations to get out of this
1599 * mount, and they will themselves hold resources we must
1600 * allow the fs to do things. In the Unix tradition of
1601 * 'Gee thats tricky lets do it in userspace' the umount_begin
1602 * might fail to complete on the first run through as other tasks
1603 * must return, and the like. Thats for the mount program to worry
1604 * about for the moment.
1605 */
1606
1607 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1608 sb->s_op->umount_begin(sb);
1609 }
1610
1611 /*
1612 * No sense to grab the lock for this test, but test itself looks
1613 * somewhat bogus. Suggestions for better replacement?
1614 * Ho-hum... In principle, we might treat that as umount + switch
1615 * to rootfs. GC would eventually take care of the old vfsmount.
1616 * Actually it makes sense, especially if rootfs would contain a
1617 * /reboot - static binary that would close all descriptors and
1618 * call reboot(9). Then init(8) could umount root and exec /reboot.
1619 */
1620 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1621 /*
1622 * Special case for "unmounting" root ...
1623 * we just try to remount it readonly.
1624 */
1625 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1626 return -EPERM;
1627 down_write(&sb->s_umount);
1628 if (!sb_rdonly(sb))
1629 retval = do_remount_sb(sb, SB_RDONLY, NULL, 0);
1630 up_write(&sb->s_umount);
1631 return retval;
1632 }
1633
1634 namespace_lock();
1635 lock_mount_hash();
1636
1637 /* Recheck MNT_LOCKED with the locks held */
1638 retval = -EINVAL;
1639 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1640 goto out;
1641
1642 event++;
1643 if (flags & MNT_DETACH) {
1644 if (!list_empty(&mnt->mnt_list))
1645 umount_tree(mnt, UMOUNT_PROPAGATE);
1646 retval = 0;
1647 } else {
1648 shrink_submounts(mnt);
1649 retval = -EBUSY;
1650 if (!propagate_mount_busy(mnt, 2)) {
1651 if (!list_empty(&mnt->mnt_list))
1652 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1653 retval = 0;
1654 }
1655 }
1656 out:
1657 unlock_mount_hash();
1658 namespace_unlock();
1659 return retval;
1660 }
1661
1662 /*
1663 * __detach_mounts - lazily unmount all mounts on the specified dentry
1664 *
1665 * During unlink, rmdir, and d_drop it is possible to loose the path
1666 * to an existing mountpoint, and wind up leaking the mount.
1667 * detach_mounts allows lazily unmounting those mounts instead of
1668 * leaking them.
1669 *
1670 * The caller may hold dentry->d_inode->i_mutex.
1671 */
1672 void __detach_mounts(struct dentry *dentry)
1673 {
1674 struct mountpoint *mp;
1675 struct mount *mnt;
1676
1677 namespace_lock();
1678 lock_mount_hash();
1679 mp = lookup_mountpoint(dentry);
1680 if (IS_ERR_OR_NULL(mp))
1681 goto out_unlock;
1682
1683 event++;
1684 while (!hlist_empty(&mp->m_list)) {
1685 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1686 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1687 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1688 umount_mnt(mnt);
1689 }
1690 else umount_tree(mnt, UMOUNT_CONNECTED);
1691 }
1692 put_mountpoint(mp);
1693 out_unlock:
1694 unlock_mount_hash();
1695 namespace_unlock();
1696 }
1697
1698 /*
1699 * Is the caller allowed to modify his namespace?
1700 */
1701 static inline bool may_mount(void)
1702 {
1703 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1704 }
1705
1706 static inline bool may_mandlock(void)
1707 {
1708 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1709 return false;
1710 #endif
1711 return capable(CAP_SYS_ADMIN);
1712 }
1713
1714 /*
1715 * Now umount can handle mount points as well as block devices.
1716 * This is important for filesystems which use unnamed block devices.
1717 *
1718 * We now support a flag for forced unmount like the other 'big iron'
1719 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1720 */
1721
1722 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1723 {
1724 struct path path;
1725 struct mount *mnt;
1726 int retval;
1727 int lookup_flags = 0;
1728
1729 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1730 return -EINVAL;
1731
1732 if (!may_mount())
1733 return -EPERM;
1734
1735 if (!(flags & UMOUNT_NOFOLLOW))
1736 lookup_flags |= LOOKUP_FOLLOW;
1737
1738 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1739 if (retval)
1740 goto out;
1741 mnt = real_mount(path.mnt);
1742 retval = -EINVAL;
1743 if (path.dentry != path.mnt->mnt_root)
1744 goto dput_and_out;
1745 if (!check_mnt(mnt))
1746 goto dput_and_out;
1747 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1748 goto dput_and_out;
1749 retval = -EPERM;
1750 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1751 goto dput_and_out;
1752
1753 retval = do_umount(mnt, flags);
1754 dput_and_out:
1755 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1756 dput(path.dentry);
1757 mntput_no_expire(mnt);
1758 out:
1759 return retval;
1760 }
1761
1762 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1763
1764 /*
1765 * The 2.0 compatible umount. No flags.
1766 */
1767 SYSCALL_DEFINE1(oldumount, char __user *, name)
1768 {
1769 return sys_umount(name, 0);
1770 }
1771
1772 #endif
1773
1774 static bool is_mnt_ns_file(struct dentry *dentry)
1775 {
1776 /* Is this a proxy for a mount namespace? */
1777 return dentry->d_op == &ns_dentry_operations &&
1778 dentry->d_fsdata == &mntns_operations;
1779 }
1780
1781 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1782 {
1783 return container_of(ns, struct mnt_namespace, ns);
1784 }
1785
1786 static bool mnt_ns_loop(struct dentry *dentry)
1787 {
1788 /* Could bind mounting the mount namespace inode cause a
1789 * mount namespace loop?
1790 */
1791 struct mnt_namespace *mnt_ns;
1792 if (!is_mnt_ns_file(dentry))
1793 return false;
1794
1795 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1796 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1797 }
1798
1799 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1800 int flag)
1801 {
1802 struct mount *res, *p, *q, *r, *parent;
1803
1804 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1805 return ERR_PTR(-EINVAL);
1806
1807 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1808 return ERR_PTR(-EINVAL);
1809
1810 res = q = clone_mnt(mnt, dentry, flag);
1811 if (IS_ERR(q))
1812 return q;
1813
1814 q->mnt_mountpoint = mnt->mnt_mountpoint;
1815
1816 p = mnt;
1817 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1818 struct mount *s;
1819 if (!is_subdir(r->mnt_mountpoint, dentry))
1820 continue;
1821
1822 for (s = r; s; s = next_mnt(s, r)) {
1823 if (!(flag & CL_COPY_UNBINDABLE) &&
1824 IS_MNT_UNBINDABLE(s)) {
1825 if (s->mnt.mnt_flags & MNT_LOCKED) {
1826 /* Both unbindable and locked. */
1827 q = ERR_PTR(-EPERM);
1828 goto out;
1829 } else {
1830 s = skip_mnt_tree(s);
1831 continue;
1832 }
1833 }
1834 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1835 is_mnt_ns_file(s->mnt.mnt_root)) {
1836 s = skip_mnt_tree(s);
1837 continue;
1838 }
1839 while (p != s->mnt_parent) {
1840 p = p->mnt_parent;
1841 q = q->mnt_parent;
1842 }
1843 p = s;
1844 parent = q;
1845 q = clone_mnt(p, p->mnt.mnt_root, flag);
1846 if (IS_ERR(q))
1847 goto out;
1848 lock_mount_hash();
1849 list_add_tail(&q->mnt_list, &res->mnt_list);
1850 attach_mnt(q, parent, p->mnt_mp);
1851 unlock_mount_hash();
1852 }
1853 }
1854 return res;
1855 out:
1856 if (res) {
1857 lock_mount_hash();
1858 umount_tree(res, UMOUNT_SYNC);
1859 unlock_mount_hash();
1860 }
1861 return q;
1862 }
1863
1864 /* Caller should check returned pointer for errors */
1865
1866 struct vfsmount *collect_mounts(const struct path *path)
1867 {
1868 struct mount *tree;
1869 namespace_lock();
1870 if (!check_mnt(real_mount(path->mnt)))
1871 tree = ERR_PTR(-EINVAL);
1872 else
1873 tree = copy_tree(real_mount(path->mnt), path->dentry,
1874 CL_COPY_ALL | CL_PRIVATE);
1875 namespace_unlock();
1876 if (IS_ERR(tree))
1877 return ERR_CAST(tree);
1878 return &tree->mnt;
1879 }
1880
1881 void drop_collected_mounts(struct vfsmount *mnt)
1882 {
1883 namespace_lock();
1884 lock_mount_hash();
1885 umount_tree(real_mount(mnt), 0);
1886 unlock_mount_hash();
1887 namespace_unlock();
1888 }
1889
1890 /**
1891 * clone_private_mount - create a private clone of a path
1892 *
1893 * This creates a new vfsmount, which will be the clone of @path. The new will
1894 * not be attached anywhere in the namespace and will be private (i.e. changes
1895 * to the originating mount won't be propagated into this).
1896 *
1897 * Release with mntput().
1898 */
1899 struct vfsmount *clone_private_mount(const struct path *path)
1900 {
1901 struct mount *old_mnt = real_mount(path->mnt);
1902 struct mount *new_mnt;
1903
1904 if (IS_MNT_UNBINDABLE(old_mnt))
1905 return ERR_PTR(-EINVAL);
1906
1907 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1908 if (IS_ERR(new_mnt))
1909 return ERR_CAST(new_mnt);
1910
1911 return &new_mnt->mnt;
1912 }
1913 EXPORT_SYMBOL_GPL(clone_private_mount);
1914
1915 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1916 struct vfsmount *root)
1917 {
1918 struct mount *mnt;
1919 int res = f(root, arg);
1920 if (res)
1921 return res;
1922 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1923 res = f(&mnt->mnt, arg);
1924 if (res)
1925 return res;
1926 }
1927 return 0;
1928 }
1929 EXPORT_SYMBOL_GPL(iterate_mounts);
1930
1931 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1932 {
1933 struct mount *p;
1934
1935 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1936 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1937 mnt_release_group_id(p);
1938 }
1939 }
1940
1941 static int invent_group_ids(struct mount *mnt, bool recurse)
1942 {
1943 struct mount *p;
1944
1945 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1946 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1947 int err = mnt_alloc_group_id(p);
1948 if (err) {
1949 cleanup_group_ids(mnt, p);
1950 return err;
1951 }
1952 }
1953 }
1954
1955 return 0;
1956 }
1957
1958 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1959 {
1960 unsigned int max = READ_ONCE(sysctl_mount_max);
1961 unsigned int mounts = 0, old, pending, sum;
1962 struct mount *p;
1963
1964 for (p = mnt; p; p = next_mnt(p, mnt))
1965 mounts++;
1966
1967 old = ns->mounts;
1968 pending = ns->pending_mounts;
1969 sum = old + pending;
1970 if ((old > sum) ||
1971 (pending > sum) ||
1972 (max < sum) ||
1973 (mounts > (max - sum)))
1974 return -ENOSPC;
1975
1976 ns->pending_mounts = pending + mounts;
1977 return 0;
1978 }
1979
1980 /*
1981 * @source_mnt : mount tree to be attached
1982 * @nd : place the mount tree @source_mnt is attached
1983 * @parent_nd : if non-null, detach the source_mnt from its parent and
1984 * store the parent mount and mountpoint dentry.
1985 * (done when source_mnt is moved)
1986 *
1987 * NOTE: in the table below explains the semantics when a source mount
1988 * of a given type is attached to a destination mount of a given type.
1989 * ---------------------------------------------------------------------------
1990 * | BIND MOUNT OPERATION |
1991 * |**************************************************************************
1992 * | source-->| shared | private | slave | unbindable |
1993 * | dest | | | | |
1994 * | | | | | | |
1995 * | v | | | | |
1996 * |**************************************************************************
1997 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1998 * | | | | | |
1999 * |non-shared| shared (+) | private | slave (*) | invalid |
2000 * ***************************************************************************
2001 * A bind operation clones the source mount and mounts the clone on the
2002 * destination mount.
2003 *
2004 * (++) the cloned mount is propagated to all the mounts in the propagation
2005 * tree of the destination mount and the cloned mount is added to
2006 * the peer group of the source mount.
2007 * (+) the cloned mount is created under the destination mount and is marked
2008 * as shared. The cloned mount is added to the peer group of the source
2009 * mount.
2010 * (+++) the mount is propagated to all the mounts in the propagation tree
2011 * of the destination mount and the cloned mount is made slave
2012 * of the same master as that of the source mount. The cloned mount
2013 * is marked as 'shared and slave'.
2014 * (*) the cloned mount is made a slave of the same master as that of the
2015 * source mount.
2016 *
2017 * ---------------------------------------------------------------------------
2018 * | MOVE MOUNT OPERATION |
2019 * |**************************************************************************
2020 * | source-->| shared | private | slave | unbindable |
2021 * | dest | | | | |
2022 * | | | | | | |
2023 * | v | | | | |
2024 * |**************************************************************************
2025 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2026 * | | | | | |
2027 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2028 * ***************************************************************************
2029 *
2030 * (+) the mount is moved to the destination. And is then propagated to
2031 * all the mounts in the propagation tree of the destination mount.
2032 * (+*) the mount is moved to the destination.
2033 * (+++) the mount is moved to the destination and is then propagated to
2034 * all the mounts belonging to the destination mount's propagation tree.
2035 * the mount is marked as 'shared and slave'.
2036 * (*) the mount continues to be a slave at the new location.
2037 *
2038 * if the source mount is a tree, the operations explained above is
2039 * applied to each mount in the tree.
2040 * Must be called without spinlocks held, since this function can sleep
2041 * in allocations.
2042 */
2043 static int attach_recursive_mnt(struct mount *source_mnt,
2044 struct mount *dest_mnt,
2045 struct mountpoint *dest_mp,
2046 struct path *parent_path)
2047 {
2048 HLIST_HEAD(tree_list);
2049 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2050 struct mountpoint *smp;
2051 struct mount *child, *p;
2052 struct hlist_node *n;
2053 int err;
2054
2055 /* Preallocate a mountpoint in case the new mounts need
2056 * to be tucked under other mounts.
2057 */
2058 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2059 if (IS_ERR(smp))
2060 return PTR_ERR(smp);
2061
2062 /* Is there space to add these mounts to the mount namespace? */
2063 if (!parent_path) {
2064 err = count_mounts(ns, source_mnt);
2065 if (err)
2066 goto out;
2067 }
2068
2069 if (IS_MNT_SHARED(dest_mnt)) {
2070 err = invent_group_ids(source_mnt, true);
2071 if (err)
2072 goto out;
2073 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2074 lock_mount_hash();
2075 if (err)
2076 goto out_cleanup_ids;
2077 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2078 set_mnt_shared(p);
2079 } else {
2080 lock_mount_hash();
2081 }
2082 if (parent_path) {
2083 detach_mnt(source_mnt, parent_path);
2084 attach_mnt(source_mnt, dest_mnt, dest_mp);
2085 touch_mnt_namespace(source_mnt->mnt_ns);
2086 } else {
2087 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2088 commit_tree(source_mnt);
2089 }
2090
2091 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2092 struct mount *q;
2093 hlist_del_init(&child->mnt_hash);
2094 q = __lookup_mnt(&child->mnt_parent->mnt,
2095 child->mnt_mountpoint);
2096 if (q)
2097 mnt_change_mountpoint(child, smp, q);
2098 commit_tree(child);
2099 }
2100 put_mountpoint(smp);
2101 unlock_mount_hash();
2102
2103 return 0;
2104
2105 out_cleanup_ids:
2106 while (!hlist_empty(&tree_list)) {
2107 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2108 child->mnt_parent->mnt_ns->pending_mounts = 0;
2109 umount_tree(child, UMOUNT_SYNC);
2110 }
2111 unlock_mount_hash();
2112 cleanup_group_ids(source_mnt, NULL);
2113 out:
2114 ns->pending_mounts = 0;
2115
2116 read_seqlock_excl(&mount_lock);
2117 put_mountpoint(smp);
2118 read_sequnlock_excl(&mount_lock);
2119
2120 return err;
2121 }
2122
2123 static struct mountpoint *lock_mount(struct path *path)
2124 {
2125 struct vfsmount *mnt;
2126 struct dentry *dentry = path->dentry;
2127 retry:
2128 inode_lock(dentry->d_inode);
2129 if (unlikely(cant_mount(dentry))) {
2130 inode_unlock(dentry->d_inode);
2131 return ERR_PTR(-ENOENT);
2132 }
2133 namespace_lock();
2134 mnt = lookup_mnt(path);
2135 if (likely(!mnt)) {
2136 struct mountpoint *mp = get_mountpoint(dentry);
2137 if (IS_ERR(mp)) {
2138 namespace_unlock();
2139 inode_unlock(dentry->d_inode);
2140 return mp;
2141 }
2142 return mp;
2143 }
2144 namespace_unlock();
2145 inode_unlock(path->dentry->d_inode);
2146 path_put(path);
2147 path->mnt = mnt;
2148 dentry = path->dentry = dget(mnt->mnt_root);
2149 goto retry;
2150 }
2151
2152 static void unlock_mount(struct mountpoint *where)
2153 {
2154 struct dentry *dentry = where->m_dentry;
2155
2156 read_seqlock_excl(&mount_lock);
2157 put_mountpoint(where);
2158 read_sequnlock_excl(&mount_lock);
2159
2160 namespace_unlock();
2161 inode_unlock(dentry->d_inode);
2162 }
2163
2164 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2165 {
2166 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2167 return -EINVAL;
2168
2169 if (d_is_dir(mp->m_dentry) !=
2170 d_is_dir(mnt->mnt.mnt_root))
2171 return -ENOTDIR;
2172
2173 return attach_recursive_mnt(mnt, p, mp, NULL);
2174 }
2175
2176 /*
2177 * Sanity check the flags to change_mnt_propagation.
2178 */
2179
2180 static int flags_to_propagation_type(int ms_flags)
2181 {
2182 int type = ms_flags & ~(MS_REC | MS_SILENT);
2183
2184 /* Fail if any non-propagation flags are set */
2185 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2186 return 0;
2187 /* Only one propagation flag should be set */
2188 if (!is_power_of_2(type))
2189 return 0;
2190 return type;
2191 }
2192
2193 /*
2194 * recursively change the type of the mountpoint.
2195 */
2196 static int do_change_type(struct path *path, int ms_flags)
2197 {
2198 struct mount *m;
2199 struct mount *mnt = real_mount(path->mnt);
2200 int recurse = ms_flags & MS_REC;
2201 int type;
2202 int err = 0;
2203
2204 if (path->dentry != path->mnt->mnt_root)
2205 return -EINVAL;
2206
2207 type = flags_to_propagation_type(ms_flags);
2208 if (!type)
2209 return -EINVAL;
2210
2211 namespace_lock();
2212 if (type == MS_SHARED) {
2213 err = invent_group_ids(mnt, recurse);
2214 if (err)
2215 goto out_unlock;
2216 }
2217
2218 lock_mount_hash();
2219 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2220 change_mnt_propagation(m, type);
2221 unlock_mount_hash();
2222
2223 out_unlock:
2224 namespace_unlock();
2225 return err;
2226 }
2227
2228 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2229 {
2230 struct mount *child;
2231 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2232 if (!is_subdir(child->mnt_mountpoint, dentry))
2233 continue;
2234
2235 if (child->mnt.mnt_flags & MNT_LOCKED)
2236 return true;
2237 }
2238 return false;
2239 }
2240
2241 /*
2242 * do loopback mount.
2243 */
2244 static int do_loopback(struct path *path, const char *old_name,
2245 int recurse)
2246 {
2247 struct path old_path;
2248 struct mount *mnt = NULL, *old, *parent;
2249 struct mountpoint *mp;
2250 int err;
2251 if (!old_name || !*old_name)
2252 return -EINVAL;
2253 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2254 if (err)
2255 return err;
2256
2257 err = -EINVAL;
2258 if (mnt_ns_loop(old_path.dentry))
2259 goto out;
2260
2261 mp = lock_mount(path);
2262 err = PTR_ERR(mp);
2263 if (IS_ERR(mp))
2264 goto out;
2265
2266 old = real_mount(old_path.mnt);
2267 parent = real_mount(path->mnt);
2268
2269 err = -EINVAL;
2270 if (IS_MNT_UNBINDABLE(old))
2271 goto out2;
2272
2273 if (!check_mnt(parent))
2274 goto out2;
2275
2276 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2277 goto out2;
2278
2279 if (!recurse && has_locked_children(old, old_path.dentry))
2280 goto out2;
2281
2282 if (recurse)
2283 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2284 else
2285 mnt = clone_mnt(old, old_path.dentry, 0);
2286
2287 if (IS_ERR(mnt)) {
2288 err = PTR_ERR(mnt);
2289 goto out2;
2290 }
2291
2292 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2293
2294 err = graft_tree(mnt, parent, mp);
2295 if (err) {
2296 lock_mount_hash();
2297 umount_tree(mnt, UMOUNT_SYNC);
2298 unlock_mount_hash();
2299 }
2300 out2:
2301 unlock_mount(mp);
2302 out:
2303 path_put(&old_path);
2304 return err;
2305 }
2306
2307 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2308 {
2309 int error = 0;
2310 int readonly_request = 0;
2311
2312 if (ms_flags & MS_RDONLY)
2313 readonly_request = 1;
2314 if (readonly_request == __mnt_is_readonly(mnt))
2315 return 0;
2316
2317 if (readonly_request)
2318 error = mnt_make_readonly(real_mount(mnt));
2319 else
2320 __mnt_unmake_readonly(real_mount(mnt));
2321 return error;
2322 }
2323
2324 /*
2325 * change filesystem flags. dir should be a physical root of filesystem.
2326 * If you've mounted a non-root directory somewhere and want to do remount
2327 * on it - tough luck.
2328 */
2329 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2330 int mnt_flags, void *data)
2331 {
2332 int err;
2333 struct super_block *sb = path->mnt->mnt_sb;
2334 struct mount *mnt = real_mount(path->mnt);
2335
2336 if (!check_mnt(mnt))
2337 return -EINVAL;
2338
2339 if (path->dentry != path->mnt->mnt_root)
2340 return -EINVAL;
2341
2342 /* Don't allow changing of locked mnt flags.
2343 *
2344 * No locks need to be held here while testing the various
2345 * MNT_LOCK flags because those flags can never be cleared
2346 * once they are set.
2347 */
2348 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2349 !(mnt_flags & MNT_READONLY)) {
2350 return -EPERM;
2351 }
2352 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2353 !(mnt_flags & MNT_NODEV)) {
2354 return -EPERM;
2355 }
2356 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2357 !(mnt_flags & MNT_NOSUID)) {
2358 return -EPERM;
2359 }
2360 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2361 !(mnt_flags & MNT_NOEXEC)) {
2362 return -EPERM;
2363 }
2364 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2365 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2366 return -EPERM;
2367 }
2368
2369 err = security_sb_remount(sb, data);
2370 if (err)
2371 return err;
2372
2373 down_write(&sb->s_umount);
2374 if (ms_flags & MS_BIND)
2375 err = change_mount_flags(path->mnt, ms_flags);
2376 else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
2377 err = -EPERM;
2378 else
2379 err = do_remount_sb(sb, sb_flags, data, 0);
2380 if (!err) {
2381 lock_mount_hash();
2382 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2383 mnt->mnt.mnt_flags = mnt_flags;
2384 touch_mnt_namespace(mnt->mnt_ns);
2385 unlock_mount_hash();
2386 }
2387 up_write(&sb->s_umount);
2388 return err;
2389 }
2390
2391 static inline int tree_contains_unbindable(struct mount *mnt)
2392 {
2393 struct mount *p;
2394 for (p = mnt; p; p = next_mnt(p, mnt)) {
2395 if (IS_MNT_UNBINDABLE(p))
2396 return 1;
2397 }
2398 return 0;
2399 }
2400
2401 static int do_move_mount(struct path *path, const char *old_name)
2402 {
2403 struct path old_path, parent_path;
2404 struct mount *p;
2405 struct mount *old;
2406 struct mountpoint *mp;
2407 int err;
2408 if (!old_name || !*old_name)
2409 return -EINVAL;
2410 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2411 if (err)
2412 return err;
2413
2414 mp = lock_mount(path);
2415 err = PTR_ERR(mp);
2416 if (IS_ERR(mp))
2417 goto out;
2418
2419 old = real_mount(old_path.mnt);
2420 p = real_mount(path->mnt);
2421
2422 err = -EINVAL;
2423 if (!check_mnt(p) || !check_mnt(old))
2424 goto out1;
2425
2426 if (old->mnt.mnt_flags & MNT_LOCKED)
2427 goto out1;
2428
2429 err = -EINVAL;
2430 if (old_path.dentry != old_path.mnt->mnt_root)
2431 goto out1;
2432
2433 if (!mnt_has_parent(old))
2434 goto out1;
2435
2436 if (d_is_dir(path->dentry) !=
2437 d_is_dir(old_path.dentry))
2438 goto out1;
2439 /*
2440 * Don't move a mount residing in a shared parent.
2441 */
2442 if (IS_MNT_SHARED(old->mnt_parent))
2443 goto out1;
2444 /*
2445 * Don't move a mount tree containing unbindable mounts to a destination
2446 * mount which is shared.
2447 */
2448 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2449 goto out1;
2450 err = -ELOOP;
2451 for (; mnt_has_parent(p); p = p->mnt_parent)
2452 if (p == old)
2453 goto out1;
2454
2455 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2456 if (err)
2457 goto out1;
2458
2459 /* if the mount is moved, it should no longer be expire
2460 * automatically */
2461 list_del_init(&old->mnt_expire);
2462 out1:
2463 unlock_mount(mp);
2464 out:
2465 if (!err)
2466 path_put(&parent_path);
2467 path_put(&old_path);
2468 return err;
2469 }
2470
2471 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2472 {
2473 int err;
2474 const char *subtype = strchr(fstype, '.');
2475 if (subtype) {
2476 subtype++;
2477 err = -EINVAL;
2478 if (!subtype[0])
2479 goto err;
2480 } else
2481 subtype = "";
2482
2483 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2484 err = -ENOMEM;
2485 if (!mnt->mnt_sb->s_subtype)
2486 goto err;
2487 return mnt;
2488
2489 err:
2490 mntput(mnt);
2491 return ERR_PTR(err);
2492 }
2493
2494 /*
2495 * add a mount into a namespace's mount tree
2496 */
2497 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2498 {
2499 struct mountpoint *mp;
2500 struct mount *parent;
2501 int err;
2502
2503 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2504
2505 mp = lock_mount(path);
2506 if (IS_ERR(mp))
2507 return PTR_ERR(mp);
2508
2509 parent = real_mount(path->mnt);
2510 err = -EINVAL;
2511 if (unlikely(!check_mnt(parent))) {
2512 /* that's acceptable only for automounts done in private ns */
2513 if (!(mnt_flags & MNT_SHRINKABLE))
2514 goto unlock;
2515 /* ... and for those we'd better have mountpoint still alive */
2516 if (!parent->mnt_ns)
2517 goto unlock;
2518 }
2519
2520 /* Refuse the same filesystem on the same mount point */
2521 err = -EBUSY;
2522 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2523 path->mnt->mnt_root == path->dentry)
2524 goto unlock;
2525
2526 err = -EINVAL;
2527 if (d_is_symlink(newmnt->mnt.mnt_root))
2528 goto unlock;
2529
2530 newmnt->mnt.mnt_flags = mnt_flags;
2531 err = graft_tree(newmnt, parent, mp);
2532
2533 unlock:
2534 unlock_mount(mp);
2535 return err;
2536 }
2537
2538 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2539
2540 /*
2541 * create a new mount for userspace and request it to be added into the
2542 * namespace's tree
2543 */
2544 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2545 int mnt_flags, const char *name, void *data)
2546 {
2547 struct file_system_type *type;
2548 struct vfsmount *mnt;
2549 int err;
2550
2551 if (!fstype)
2552 return -EINVAL;
2553
2554 type = get_fs_type(fstype);
2555 if (!type)
2556 return -ENODEV;
2557
2558 mnt = vfs_kern_mount(type, sb_flags, name, data);
2559 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2560 !mnt->mnt_sb->s_subtype)
2561 mnt = fs_set_subtype(mnt, fstype);
2562
2563 put_filesystem(type);
2564 if (IS_ERR(mnt))
2565 return PTR_ERR(mnt);
2566
2567 if (mount_too_revealing(mnt, &mnt_flags)) {
2568 mntput(mnt);
2569 return -EPERM;
2570 }
2571
2572 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2573 if (err)
2574 mntput(mnt);
2575 return err;
2576 }
2577
2578 int finish_automount(struct vfsmount *m, struct path *path)
2579 {
2580 struct mount *mnt = real_mount(m);
2581 int err;
2582 /* The new mount record should have at least 2 refs to prevent it being
2583 * expired before we get a chance to add it
2584 */
2585 BUG_ON(mnt_get_count(mnt) < 2);
2586
2587 if (m->mnt_sb == path->mnt->mnt_sb &&
2588 m->mnt_root == path->dentry) {
2589 err = -ELOOP;
2590 goto fail;
2591 }
2592
2593 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2594 if (!err)
2595 return 0;
2596 fail:
2597 /* remove m from any expiration list it may be on */
2598 if (!list_empty(&mnt->mnt_expire)) {
2599 namespace_lock();
2600 list_del_init(&mnt->mnt_expire);
2601 namespace_unlock();
2602 }
2603 mntput(m);
2604 mntput(m);
2605 return err;
2606 }
2607
2608 /**
2609 * mnt_set_expiry - Put a mount on an expiration list
2610 * @mnt: The mount to list.
2611 * @expiry_list: The list to add the mount to.
2612 */
2613 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2614 {
2615 namespace_lock();
2616
2617 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2618
2619 namespace_unlock();
2620 }
2621 EXPORT_SYMBOL(mnt_set_expiry);
2622
2623 /*
2624 * process a list of expirable mountpoints with the intent of discarding any
2625 * mountpoints that aren't in use and haven't been touched since last we came
2626 * here
2627 */
2628 void mark_mounts_for_expiry(struct list_head *mounts)
2629 {
2630 struct mount *mnt, *next;
2631 LIST_HEAD(graveyard);
2632
2633 if (list_empty(mounts))
2634 return;
2635
2636 namespace_lock();
2637 lock_mount_hash();
2638
2639 /* extract from the expiration list every vfsmount that matches the
2640 * following criteria:
2641 * - only referenced by its parent vfsmount
2642 * - still marked for expiry (marked on the last call here; marks are
2643 * cleared by mntput())
2644 */
2645 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2646 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2647 propagate_mount_busy(mnt, 1))
2648 continue;
2649 list_move(&mnt->mnt_expire, &graveyard);
2650 }
2651 while (!list_empty(&graveyard)) {
2652 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2653 touch_mnt_namespace(mnt->mnt_ns);
2654 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2655 }
2656 unlock_mount_hash();
2657 namespace_unlock();
2658 }
2659
2660 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2661
2662 /*
2663 * Ripoff of 'select_parent()'
2664 *
2665 * search the list of submounts for a given mountpoint, and move any
2666 * shrinkable submounts to the 'graveyard' list.
2667 */
2668 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2669 {
2670 struct mount *this_parent = parent;
2671 struct list_head *next;
2672 int found = 0;
2673
2674 repeat:
2675 next = this_parent->mnt_mounts.next;
2676 resume:
2677 while (next != &this_parent->mnt_mounts) {
2678 struct list_head *tmp = next;
2679 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2680
2681 next = tmp->next;
2682 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2683 continue;
2684 /*
2685 * Descend a level if the d_mounts list is non-empty.
2686 */
2687 if (!list_empty(&mnt->mnt_mounts)) {
2688 this_parent = mnt;
2689 goto repeat;
2690 }
2691
2692 if (!propagate_mount_busy(mnt, 1)) {
2693 list_move_tail(&mnt->mnt_expire, graveyard);
2694 found++;
2695 }
2696 }
2697 /*
2698 * All done at this level ... ascend and resume the search
2699 */
2700 if (this_parent != parent) {
2701 next = this_parent->mnt_child.next;
2702 this_parent = this_parent->mnt_parent;
2703 goto resume;
2704 }
2705 return found;
2706 }
2707
2708 /*
2709 * process a list of expirable mountpoints with the intent of discarding any
2710 * submounts of a specific parent mountpoint
2711 *
2712 * mount_lock must be held for write
2713 */
2714 static void shrink_submounts(struct mount *mnt)
2715 {
2716 LIST_HEAD(graveyard);
2717 struct mount *m;
2718
2719 /* extract submounts of 'mountpoint' from the expiration list */
2720 while (select_submounts(mnt, &graveyard)) {
2721 while (!list_empty(&graveyard)) {
2722 m = list_first_entry(&graveyard, struct mount,
2723 mnt_expire);
2724 touch_mnt_namespace(m->mnt_ns);
2725 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2726 }
2727 }
2728 }
2729
2730 /*
2731 * Some copy_from_user() implementations do not return the exact number of
2732 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2733 * Note that this function differs from copy_from_user() in that it will oops
2734 * on bad values of `to', rather than returning a short copy.
2735 */
2736 static long exact_copy_from_user(void *to, const void __user * from,
2737 unsigned long n)
2738 {
2739 char *t = to;
2740 const char __user *f = from;
2741 char c;
2742
2743 if (!access_ok(VERIFY_READ, from, n))
2744 return n;
2745
2746 while (n) {
2747 if (__get_user(c, f)) {
2748 memset(t, 0, n);
2749 break;
2750 }
2751 *t++ = c;
2752 f++;
2753 n--;
2754 }
2755 return n;
2756 }
2757
2758 void *copy_mount_options(const void __user * data)
2759 {
2760 int i;
2761 unsigned long size;
2762 char *copy;
2763
2764 if (!data)
2765 return NULL;
2766
2767 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2768 if (!copy)
2769 return ERR_PTR(-ENOMEM);
2770
2771 /* We only care that *some* data at the address the user
2772 * gave us is valid. Just in case, we'll zero
2773 * the remainder of the page.
2774 */
2775 /* copy_from_user cannot cross TASK_SIZE ! */
2776 size = TASK_SIZE - (unsigned long)data;
2777 if (size > PAGE_SIZE)
2778 size = PAGE_SIZE;
2779
2780 i = size - exact_copy_from_user(copy, data, size);
2781 if (!i) {
2782 kfree(copy);
2783 return ERR_PTR(-EFAULT);
2784 }
2785 if (i != PAGE_SIZE)
2786 memset(copy + i, 0, PAGE_SIZE - i);
2787 return copy;
2788 }
2789
2790 char *copy_mount_string(const void __user *data)
2791 {
2792 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2793 }
2794
2795 /*
2796 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2797 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2798 *
2799 * data is a (void *) that can point to any structure up to
2800 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2801 * information (or be NULL).
2802 *
2803 * Pre-0.97 versions of mount() didn't have a flags word.
2804 * When the flags word was introduced its top half was required
2805 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2806 * Therefore, if this magic number is present, it carries no information
2807 * and must be discarded.
2808 */
2809 long do_mount(const char *dev_name, const char __user *dir_name,
2810 const char *type_page, unsigned long flags, void *data_page)
2811 {
2812 struct path path;
2813 unsigned int mnt_flags = 0, sb_flags;
2814 int retval = 0;
2815
2816 /* Discard magic */
2817 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2818 flags &= ~MS_MGC_MSK;
2819
2820 /* Basic sanity checks */
2821 if (data_page)
2822 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2823
2824 if (flags & MS_NOUSER)
2825 return -EINVAL;
2826
2827 /* ... and get the mountpoint */
2828 retval = user_path(dir_name, &path);
2829 if (retval)
2830 return retval;
2831
2832 retval = security_sb_mount(dev_name, &path,
2833 type_page, flags, data_page);
2834 if (!retval && !may_mount())
2835 retval = -EPERM;
2836 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
2837 retval = -EPERM;
2838 if (retval)
2839 goto dput_out;
2840
2841 /* Default to relatime unless overriden */
2842 if (!(flags & MS_NOATIME))
2843 mnt_flags |= MNT_RELATIME;
2844
2845 /* Separate the per-mountpoint flags */
2846 if (flags & MS_NOSUID)
2847 mnt_flags |= MNT_NOSUID;
2848 if (flags & MS_NODEV)
2849 mnt_flags |= MNT_NODEV;
2850 if (flags & MS_NOEXEC)
2851 mnt_flags |= MNT_NOEXEC;
2852 if (flags & MS_NOATIME)
2853 mnt_flags |= MNT_NOATIME;
2854 if (flags & MS_NODIRATIME)
2855 mnt_flags |= MNT_NODIRATIME;
2856 if (flags & MS_STRICTATIME)
2857 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2858 if (flags & MS_RDONLY)
2859 mnt_flags |= MNT_READONLY;
2860
2861 /* The default atime for remount is preservation */
2862 if ((flags & MS_REMOUNT) &&
2863 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2864 MS_STRICTATIME)) == 0)) {
2865 mnt_flags &= ~MNT_ATIME_MASK;
2866 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2867 }
2868
2869 sb_flags = flags & (SB_RDONLY |
2870 SB_SYNCHRONOUS |
2871 SB_MANDLOCK |
2872 SB_DIRSYNC |
2873 SB_SILENT |
2874 SB_POSIXACL |
2875 SB_LAZYTIME |
2876 SB_I_VERSION);
2877
2878 if (flags & MS_REMOUNT)
2879 retval = do_remount(&path, flags, sb_flags, mnt_flags,
2880 data_page);
2881 else if (flags & MS_BIND)
2882 retval = do_loopback(&path, dev_name, flags & MS_REC);
2883 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2884 retval = do_change_type(&path, flags);
2885 else if (flags & MS_MOVE)
2886 retval = do_move_mount(&path, dev_name);
2887 else
2888 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2889 dev_name, data_page);
2890 dput_out:
2891 path_put(&path);
2892 return retval;
2893 }
2894
2895 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2896 {
2897 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2898 }
2899
2900 static void dec_mnt_namespaces(struct ucounts *ucounts)
2901 {
2902 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2903 }
2904
2905 static void free_mnt_ns(struct mnt_namespace *ns)
2906 {
2907 ns_free_inum(&ns->ns);
2908 dec_mnt_namespaces(ns->ucounts);
2909 put_user_ns(ns->user_ns);
2910 kfree(ns);
2911 }
2912
2913 /*
2914 * Assign a sequence number so we can detect when we attempt to bind
2915 * mount a reference to an older mount namespace into the current
2916 * mount namespace, preventing reference counting loops. A 64bit
2917 * number incrementing at 10Ghz will take 12,427 years to wrap which
2918 * is effectively never, so we can ignore the possibility.
2919 */
2920 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2921
2922 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2923 {
2924 struct mnt_namespace *new_ns;
2925 struct ucounts *ucounts;
2926 int ret;
2927
2928 ucounts = inc_mnt_namespaces(user_ns);
2929 if (!ucounts)
2930 return ERR_PTR(-ENOSPC);
2931
2932 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2933 if (!new_ns) {
2934 dec_mnt_namespaces(ucounts);
2935 return ERR_PTR(-ENOMEM);
2936 }
2937 ret = ns_alloc_inum(&new_ns->ns);
2938 if (ret) {
2939 kfree(new_ns);
2940 dec_mnt_namespaces(ucounts);
2941 return ERR_PTR(ret);
2942 }
2943 new_ns->ns.ops = &mntns_operations;
2944 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2945 atomic_set(&new_ns->count, 1);
2946 new_ns->root = NULL;
2947 INIT_LIST_HEAD(&new_ns->list);
2948 init_waitqueue_head(&new_ns->poll);
2949 new_ns->event = 0;
2950 new_ns->user_ns = get_user_ns(user_ns);
2951 new_ns->ucounts = ucounts;
2952 new_ns->mounts = 0;
2953 new_ns->pending_mounts = 0;
2954 return new_ns;
2955 }
2956
2957 __latent_entropy
2958 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2959 struct user_namespace *user_ns, struct fs_struct *new_fs)
2960 {
2961 struct mnt_namespace *new_ns;
2962 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2963 struct mount *p, *q;
2964 struct mount *old;
2965 struct mount *new;
2966 int copy_flags;
2967
2968 BUG_ON(!ns);
2969
2970 if (likely(!(flags & CLONE_NEWNS))) {
2971 get_mnt_ns(ns);
2972 return ns;
2973 }
2974
2975 old = ns->root;
2976
2977 new_ns = alloc_mnt_ns(user_ns);
2978 if (IS_ERR(new_ns))
2979 return new_ns;
2980
2981 namespace_lock();
2982 /* First pass: copy the tree topology */
2983 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2984 if (user_ns != ns->user_ns)
2985 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2986 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2987 if (IS_ERR(new)) {
2988 namespace_unlock();
2989 free_mnt_ns(new_ns);
2990 return ERR_CAST(new);
2991 }
2992 new_ns->root = new;
2993 list_add_tail(&new_ns->list, &new->mnt_list);
2994
2995 /*
2996 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2997 * as belonging to new namespace. We have already acquired a private
2998 * fs_struct, so tsk->fs->lock is not needed.
2999 */
3000 p = old;
3001 q = new;
3002 while (p) {
3003 q->mnt_ns = new_ns;
3004 new_ns->mounts++;
3005 if (new_fs) {
3006 if (&p->mnt == new_fs->root.mnt) {
3007 new_fs->root.mnt = mntget(&q->mnt);
3008 rootmnt = &p->mnt;
3009 }
3010 if (&p->mnt == new_fs->pwd.mnt) {
3011 new_fs->pwd.mnt = mntget(&q->mnt);
3012 pwdmnt = &p->mnt;
3013 }
3014 }
3015 p = next_mnt(p, old);
3016 q = next_mnt(q, new);
3017 if (!q)
3018 break;
3019 while (p->mnt.mnt_root != q->mnt.mnt_root)
3020 p = next_mnt(p, old);
3021 }
3022 namespace_unlock();
3023
3024 if (rootmnt)
3025 mntput(rootmnt);
3026 if (pwdmnt)
3027 mntput(pwdmnt);
3028
3029 return new_ns;
3030 }
3031
3032 /**
3033 * create_mnt_ns - creates a private namespace and adds a root filesystem
3034 * @mnt: pointer to the new root filesystem mountpoint
3035 */
3036 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
3037 {
3038 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
3039 if (!IS_ERR(new_ns)) {
3040 struct mount *mnt = real_mount(m);
3041 mnt->mnt_ns = new_ns;
3042 new_ns->root = mnt;
3043 new_ns->mounts++;
3044 list_add(&mnt->mnt_list, &new_ns->list);
3045 } else {
3046 mntput(m);
3047 }
3048 return new_ns;
3049 }
3050
3051 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3052 {
3053 struct mnt_namespace *ns;
3054 struct super_block *s;
3055 struct path path;
3056 int err;
3057
3058 ns = create_mnt_ns(mnt);
3059 if (IS_ERR(ns))
3060 return ERR_CAST(ns);
3061
3062 err = vfs_path_lookup(mnt->mnt_root, mnt,
3063 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3064
3065 put_mnt_ns(ns);
3066
3067 if (err)
3068 return ERR_PTR(err);
3069
3070 /* trade a vfsmount reference for active sb one */
3071 s = path.mnt->mnt_sb;
3072 atomic_inc(&s->s_active);
3073 mntput(path.mnt);
3074 /* lock the sucker */
3075 down_write(&s->s_umount);
3076 /* ... and return the root of (sub)tree on it */
3077 return path.dentry;
3078 }
3079 EXPORT_SYMBOL(mount_subtree);
3080
3081 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3082 char __user *, type, unsigned long, flags, void __user *, data)
3083 {
3084 int ret;
3085 char *kernel_type;
3086 char *kernel_dev;
3087 void *options;
3088
3089 kernel_type = copy_mount_string(type);
3090 ret = PTR_ERR(kernel_type);
3091 if (IS_ERR(kernel_type))
3092 goto out_type;
3093
3094 kernel_dev = copy_mount_string(dev_name);
3095 ret = PTR_ERR(kernel_dev);
3096 if (IS_ERR(kernel_dev))
3097 goto out_dev;
3098
3099 options = copy_mount_options(data);
3100 ret = PTR_ERR(options);
3101 if (IS_ERR(options))
3102 goto out_data;
3103
3104 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3105
3106 kfree(options);
3107 out_data:
3108 kfree(kernel_dev);
3109 out_dev:
3110 kfree(kernel_type);
3111 out_type:
3112 return ret;
3113 }
3114
3115 /*
3116 * Return true if path is reachable from root
3117 *
3118 * namespace_sem or mount_lock is held
3119 */
3120 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3121 const struct path *root)
3122 {
3123 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3124 dentry = mnt->mnt_mountpoint;
3125 mnt = mnt->mnt_parent;
3126 }
3127 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3128 }
3129
3130 bool path_is_under(const struct path *path1, const struct path *path2)
3131 {
3132 bool res;
3133 read_seqlock_excl(&mount_lock);
3134 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3135 read_sequnlock_excl(&mount_lock);
3136 return res;
3137 }
3138 EXPORT_SYMBOL(path_is_under);
3139
3140 /*
3141 * pivot_root Semantics:
3142 * Moves the root file system of the current process to the directory put_old,
3143 * makes new_root as the new root file system of the current process, and sets
3144 * root/cwd of all processes which had them on the current root to new_root.
3145 *
3146 * Restrictions:
3147 * The new_root and put_old must be directories, and must not be on the
3148 * same file system as the current process root. The put_old must be
3149 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3150 * pointed to by put_old must yield the same directory as new_root. No other
3151 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3152 *
3153 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3154 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3155 * in this situation.
3156 *
3157 * Notes:
3158 * - we don't move root/cwd if they are not at the root (reason: if something
3159 * cared enough to change them, it's probably wrong to force them elsewhere)
3160 * - it's okay to pick a root that isn't the root of a file system, e.g.
3161 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3162 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3163 * first.
3164 */
3165 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3166 const char __user *, put_old)
3167 {
3168 struct path new, old, parent_path, root_parent, root;
3169 struct mount *new_mnt, *root_mnt, *old_mnt;
3170 struct mountpoint *old_mp, *root_mp;
3171 int error;
3172
3173 if (!may_mount())
3174 return -EPERM;
3175
3176 error = user_path_dir(new_root, &new);
3177 if (error)
3178 goto out0;
3179
3180 error = user_path_dir(put_old, &old);
3181 if (error)
3182 goto out1;
3183
3184 error = security_sb_pivotroot(&old, &new);
3185 if (error)
3186 goto out2;
3187
3188 get_fs_root(current->fs, &root);
3189 old_mp = lock_mount(&old);
3190 error = PTR_ERR(old_mp);
3191 if (IS_ERR(old_mp))
3192 goto out3;
3193
3194 error = -EINVAL;
3195 new_mnt = real_mount(new.mnt);
3196 root_mnt = real_mount(root.mnt);
3197 old_mnt = real_mount(old.mnt);
3198 if (IS_MNT_SHARED(old_mnt) ||
3199 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3200 IS_MNT_SHARED(root_mnt->mnt_parent))
3201 goto out4;
3202 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3203 goto out4;
3204 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3205 goto out4;
3206 error = -ENOENT;
3207 if (d_unlinked(new.dentry))
3208 goto out4;
3209 error = -EBUSY;
3210 if (new_mnt == root_mnt || old_mnt == root_mnt)
3211 goto out4; /* loop, on the same file system */
3212 error = -EINVAL;
3213 if (root.mnt->mnt_root != root.dentry)
3214 goto out4; /* not a mountpoint */
3215 if (!mnt_has_parent(root_mnt))
3216 goto out4; /* not attached */
3217 root_mp = root_mnt->mnt_mp;
3218 if (new.mnt->mnt_root != new.dentry)
3219 goto out4; /* not a mountpoint */
3220 if (!mnt_has_parent(new_mnt))
3221 goto out4; /* not attached */
3222 /* make sure we can reach put_old from new_root */
3223 if (!is_path_reachable(old_mnt, old.dentry, &new))
3224 goto out4;
3225 /* make certain new is below the root */
3226 if (!is_path_reachable(new_mnt, new.dentry, &root))
3227 goto out4;
3228 root_mp->m_count++; /* pin it so it won't go away */
3229 lock_mount_hash();
3230 detach_mnt(new_mnt, &parent_path);
3231 detach_mnt(root_mnt, &root_parent);
3232 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3233 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3234 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3235 }
3236 /* mount old root on put_old */
3237 attach_mnt(root_mnt, old_mnt, old_mp);
3238 /* mount new_root on / */
3239 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3240 touch_mnt_namespace(current->nsproxy->mnt_ns);
3241 /* A moved mount should not expire automatically */
3242 list_del_init(&new_mnt->mnt_expire);
3243 put_mountpoint(root_mp);
3244 unlock_mount_hash();
3245 chroot_fs_refs(&root, &new);
3246 error = 0;
3247 out4:
3248 unlock_mount(old_mp);
3249 if (!error) {
3250 path_put(&root_parent);
3251 path_put(&parent_path);
3252 }
3253 out3:
3254 path_put(&root);
3255 out2:
3256 path_put(&old);
3257 out1:
3258 path_put(&new);
3259 out0:
3260 return error;
3261 }
3262
3263 static void __init init_mount_tree(void)
3264 {
3265 struct vfsmount *mnt;
3266 struct mnt_namespace *ns;
3267 struct path root;
3268 struct file_system_type *type;
3269
3270 type = get_fs_type("rootfs");
3271 if (!type)
3272 panic("Can't find rootfs type");
3273 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3274 put_filesystem(type);
3275 if (IS_ERR(mnt))
3276 panic("Can't create rootfs");
3277
3278 ns = create_mnt_ns(mnt);
3279 if (IS_ERR(ns))
3280 panic("Can't allocate initial namespace");
3281
3282 init_task.nsproxy->mnt_ns = ns;
3283 get_mnt_ns(ns);
3284
3285 root.mnt = mnt;
3286 root.dentry = mnt->mnt_root;
3287 mnt->mnt_flags |= MNT_LOCKED;
3288
3289 set_fs_pwd(current->fs, &root);
3290 set_fs_root(current->fs, &root);
3291 }
3292
3293 void __init mnt_init(void)
3294 {
3295 int err;
3296
3297 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3298 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3299
3300 mount_hashtable = alloc_large_system_hash("Mount-cache",
3301 sizeof(struct hlist_head),
3302 mhash_entries, 19,
3303 HASH_ZERO,
3304 &m_hash_shift, &m_hash_mask, 0, 0);
3305 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3306 sizeof(struct hlist_head),
3307 mphash_entries, 19,
3308 HASH_ZERO,
3309 &mp_hash_shift, &mp_hash_mask, 0, 0);
3310
3311 if (!mount_hashtable || !mountpoint_hashtable)
3312 panic("Failed to allocate mount hash table\n");
3313
3314 kernfs_init();
3315
3316 err = sysfs_init();
3317 if (err)
3318 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3319 __func__, err);
3320 fs_kobj = kobject_create_and_add("fs", NULL);
3321 if (!fs_kobj)
3322 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3323 init_rootfs();
3324 init_mount_tree();
3325 }
3326
3327 void put_mnt_ns(struct mnt_namespace *ns)
3328 {
3329 if (!atomic_dec_and_test(&ns->count))
3330 return;
3331 drop_collected_mounts(&ns->root->mnt);
3332 free_mnt_ns(ns);
3333 }
3334
3335 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3336 {
3337 struct vfsmount *mnt;
3338 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data);
3339 if (!IS_ERR(mnt)) {
3340 /*
3341 * it is a longterm mount, don't release mnt until
3342 * we unmount before file sys is unregistered
3343 */
3344 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3345 }
3346 return mnt;
3347 }
3348 EXPORT_SYMBOL_GPL(kern_mount_data);
3349
3350 void kern_unmount(struct vfsmount *mnt)
3351 {
3352 /* release long term mount so mount point can be released */
3353 if (!IS_ERR_OR_NULL(mnt)) {
3354 real_mount(mnt)->mnt_ns = NULL;
3355 synchronize_rcu(); /* yecchhh... */
3356 mntput(mnt);
3357 }
3358 }
3359 EXPORT_SYMBOL(kern_unmount);
3360
3361 bool our_mnt(struct vfsmount *mnt)
3362 {
3363 return check_mnt(real_mount(mnt));
3364 }
3365
3366 bool current_chrooted(void)
3367 {
3368 /* Does the current process have a non-standard root */
3369 struct path ns_root;
3370 struct path fs_root;
3371 bool chrooted;
3372
3373 /* Find the namespace root */
3374 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3375 ns_root.dentry = ns_root.mnt->mnt_root;
3376 path_get(&ns_root);
3377 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3378 ;
3379
3380 get_fs_root(current->fs, &fs_root);
3381
3382 chrooted = !path_equal(&fs_root, &ns_root);
3383
3384 path_put(&fs_root);
3385 path_put(&ns_root);
3386
3387 return chrooted;
3388 }
3389
3390 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3391 int *new_mnt_flags)
3392 {
3393 int new_flags = *new_mnt_flags;
3394 struct mount *mnt;
3395 bool visible = false;
3396
3397 down_read(&namespace_sem);
3398 list_for_each_entry(mnt, &ns->list, mnt_list) {
3399 struct mount *child;
3400 int mnt_flags;
3401
3402 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3403 continue;
3404
3405 /* This mount is not fully visible if it's root directory
3406 * is not the root directory of the filesystem.
3407 */
3408 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3409 continue;
3410
3411 /* A local view of the mount flags */
3412 mnt_flags = mnt->mnt.mnt_flags;
3413
3414 /* Don't miss readonly hidden in the superblock flags */
3415 if (sb_rdonly(mnt->mnt.mnt_sb))
3416 mnt_flags |= MNT_LOCK_READONLY;
3417
3418 /* Verify the mount flags are equal to or more permissive
3419 * than the proposed new mount.
3420 */
3421 if ((mnt_flags & MNT_LOCK_READONLY) &&
3422 !(new_flags & MNT_READONLY))
3423 continue;
3424 if ((mnt_flags & MNT_LOCK_ATIME) &&
3425 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3426 continue;
3427
3428 /* This mount is not fully visible if there are any
3429 * locked child mounts that cover anything except for
3430 * empty directories.
3431 */
3432 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3433 struct inode *inode = child->mnt_mountpoint->d_inode;
3434 /* Only worry about locked mounts */
3435 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3436 continue;
3437 /* Is the directory permanetly empty? */
3438 if (!is_empty_dir_inode(inode))
3439 goto next;
3440 }
3441 /* Preserve the locked attributes */
3442 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3443 MNT_LOCK_ATIME);
3444 visible = true;
3445 goto found;
3446 next: ;
3447 }
3448 found:
3449 up_read(&namespace_sem);
3450 return visible;
3451 }
3452
3453 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3454 {
3455 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3456 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3457 unsigned long s_iflags;
3458
3459 if (ns->user_ns == &init_user_ns)
3460 return false;
3461
3462 /* Can this filesystem be too revealing? */
3463 s_iflags = mnt->mnt_sb->s_iflags;
3464 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3465 return false;
3466
3467 if ((s_iflags & required_iflags) != required_iflags) {
3468 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3469 required_iflags);
3470 return true;
3471 }
3472
3473 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3474 }
3475
3476 bool mnt_may_suid(struct vfsmount *mnt)
3477 {
3478 /*
3479 * Foreign mounts (accessed via fchdir or through /proc
3480 * symlinks) are always treated as if they are nosuid. This
3481 * prevents namespaces from trusting potentially unsafe
3482 * suid/sgid bits, file caps, or security labels that originate
3483 * in other namespaces.
3484 */
3485 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3486 current_in_userns(mnt->mnt_sb->s_user_ns);
3487 }
3488
3489 static struct ns_common *mntns_get(struct task_struct *task)
3490 {
3491 struct ns_common *ns = NULL;
3492 struct nsproxy *nsproxy;
3493
3494 task_lock(task);
3495 nsproxy = task->nsproxy;
3496 if (nsproxy) {
3497 ns = &nsproxy->mnt_ns->ns;
3498 get_mnt_ns(to_mnt_ns(ns));
3499 }
3500 task_unlock(task);
3501
3502 return ns;
3503 }
3504
3505 static void mntns_put(struct ns_common *ns)
3506 {
3507 put_mnt_ns(to_mnt_ns(ns));
3508 }
3509
3510 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3511 {
3512 struct fs_struct *fs = current->fs;
3513 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3514 struct path root;
3515 int err;
3516
3517 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3518 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3519 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3520 return -EPERM;
3521
3522 if (fs->users != 1)
3523 return -EINVAL;
3524
3525 get_mnt_ns(mnt_ns);
3526 old_mnt_ns = nsproxy->mnt_ns;
3527 nsproxy->mnt_ns = mnt_ns;
3528
3529 /* Find the root */
3530 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3531 "/", LOOKUP_DOWN, &root);
3532 if (err) {
3533 /* revert to old namespace */
3534 nsproxy->mnt_ns = old_mnt_ns;
3535 put_mnt_ns(mnt_ns);
3536 return err;
3537 }
3538
3539 put_mnt_ns(old_mnt_ns);
3540
3541 /* Update the pwd and root */
3542 set_fs_pwd(fs, &root);
3543 set_fs_root(fs, &root);
3544
3545 path_put(&root);
3546 return 0;
3547 }
3548
3549 static struct user_namespace *mntns_owner(struct ns_common *ns)
3550 {
3551 return to_mnt_ns(ns)->user_ns;
3552 }
3553
3554 const struct proc_ns_operations mntns_operations = {
3555 .name = "mnt",
3556 .type = CLONE_NEWNS,
3557 .get = mntns_get,
3558 .put = mntns_put,
3559 .install = mntns_install,
3560 .owner = mntns_owner,
3561 };