Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
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/file.h>
24#include <linux/uaccess.h>
25#include <linux/proc_ns.h>
26#include <linux/magic.h>
27#include <linux/memblock.h>
28#include <linux/proc_fs.h>
29#include <linux/task_work.h>
30#include <linux/sched/task.h>
31#include <uapi/linux/mount.h>
32#include <linux/fs_context.h>
33#include <linux/shmem_fs.h>
34#include <linux/mnt_idmapping.h>
35#include <linux/nospec.h>
36
37#include "pnode.h"
38#include "internal.h"
39
40/* Maximum number of mounts in a mount namespace */
41static unsigned int sysctl_mount_max __read_mostly = 100000;
42
43static unsigned int m_hash_mask __ro_after_init;
44static unsigned int m_hash_shift __ro_after_init;
45static unsigned int mp_hash_mask __ro_after_init;
46static unsigned int mp_hash_shift __ro_after_init;
47
48static __initdata unsigned long mhash_entries;
49static int __init set_mhash_entries(char *str)
50{
51 if (!str)
52 return 0;
53 mhash_entries = simple_strtoul(str, &str, 0);
54 return 1;
55}
56__setup("mhash_entries=", set_mhash_entries);
57
58static __initdata unsigned long mphash_entries;
59static int __init set_mphash_entries(char *str)
60{
61 if (!str)
62 return 0;
63 mphash_entries = simple_strtoul(str, &str, 0);
64 return 1;
65}
66__setup("mphash_entries=", set_mphash_entries);
67
68static u64 event;
69static DEFINE_IDA(mnt_id_ida);
70static DEFINE_IDA(mnt_group_ida);
71
72/* Don't allow confusion with old 32bit mount ID */
73static atomic64_t mnt_id_ctr = ATOMIC64_INIT(1ULL << 32);
74
75static struct hlist_head *mount_hashtable __ro_after_init;
76static struct hlist_head *mountpoint_hashtable __ro_after_init;
77static struct kmem_cache *mnt_cache __ro_after_init;
78static DECLARE_RWSEM(namespace_sem);
79static HLIST_HEAD(unmounted); /* protected by namespace_sem */
80static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
81
82struct mount_kattr {
83 unsigned int attr_set;
84 unsigned int attr_clr;
85 unsigned int propagation;
86 unsigned int lookup_flags;
87 bool recurse;
88 struct user_namespace *mnt_userns;
89 struct mnt_idmap *mnt_idmap;
90};
91
92/* /sys/fs */
93struct kobject *fs_kobj __ro_after_init;
94EXPORT_SYMBOL_GPL(fs_kobj);
95
96/*
97 * vfsmount lock may be taken for read to prevent changes to the
98 * vfsmount hash, ie. during mountpoint lookups or walking back
99 * up the tree.
100 *
101 * It should be taken for write in all cases where the vfsmount
102 * tree or hash is modified or when a vfsmount structure is modified.
103 */
104__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
105
106static inline void lock_mount_hash(void)
107{
108 write_seqlock(&mount_lock);
109}
110
111static inline void unlock_mount_hash(void)
112{
113 write_sequnlock(&mount_lock);
114}
115
116static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
117{
118 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
119 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
120 tmp = tmp + (tmp >> m_hash_shift);
121 return &mount_hashtable[tmp & m_hash_mask];
122}
123
124static inline struct hlist_head *mp_hash(struct dentry *dentry)
125{
126 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
127 tmp = tmp + (tmp >> mp_hash_shift);
128 return &mountpoint_hashtable[tmp & mp_hash_mask];
129}
130
131static int mnt_alloc_id(struct mount *mnt)
132{
133 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
134
135 if (res < 0)
136 return res;
137 mnt->mnt_id = res;
138 mnt->mnt_id_unique = atomic64_inc_return(&mnt_id_ctr);
139 return 0;
140}
141
142static void mnt_free_id(struct mount *mnt)
143{
144 ida_free(&mnt_id_ida, mnt->mnt_id);
145}
146
147/*
148 * Allocate a new peer group ID
149 */
150static int mnt_alloc_group_id(struct mount *mnt)
151{
152 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
153
154 if (res < 0)
155 return res;
156 mnt->mnt_group_id = res;
157 return 0;
158}
159
160/*
161 * Release a peer group ID
162 */
163void mnt_release_group_id(struct mount *mnt)
164{
165 ida_free(&mnt_group_ida, mnt->mnt_group_id);
166 mnt->mnt_group_id = 0;
167}
168
169/*
170 * vfsmount lock must be held for read
171 */
172static inline void mnt_add_count(struct mount *mnt, int n)
173{
174#ifdef CONFIG_SMP
175 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
176#else
177 preempt_disable();
178 mnt->mnt_count += n;
179 preempt_enable();
180#endif
181}
182
183/*
184 * vfsmount lock must be held for write
185 */
186int mnt_get_count(struct mount *mnt)
187{
188#ifdef CONFIG_SMP
189 int count = 0;
190 int cpu;
191
192 for_each_possible_cpu(cpu) {
193 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
194 }
195
196 return count;
197#else
198 return mnt->mnt_count;
199#endif
200}
201
202static struct mount *alloc_vfsmnt(const char *name)
203{
204 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
205 if (mnt) {
206 int err;
207
208 err = mnt_alloc_id(mnt);
209 if (err)
210 goto out_free_cache;
211
212 if (name) {
213 mnt->mnt_devname = kstrdup_const(name,
214 GFP_KERNEL_ACCOUNT);
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_HLIST_HEAD(&mnt->mnt_stuck_children);
241 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
242 }
243 return mnt;
244
245#ifdef CONFIG_SMP
246out_free_devname:
247 kfree_const(mnt->mnt_devname);
248#endif
249out_free_id:
250 mnt_free_id(mnt);
251out_free_cache:
252 kmem_cache_free(mnt_cache, mnt);
253 return NULL;
254}
255
256/*
257 * Most r/o checks on a fs are for operations that take
258 * discrete amounts of time, like a write() or unlink().
259 * We must keep track of when those operations start
260 * (for permission checks) and when they end, so that
261 * we can determine when writes are able to occur to
262 * a filesystem.
263 */
264/*
265 * __mnt_is_readonly: check whether a mount is read-only
266 * @mnt: the mount to check for its write status
267 *
268 * This shouldn't be used directly ouside of the VFS.
269 * It does not guarantee that the filesystem will stay
270 * r/w, just that it is right *now*. This can not and
271 * should not be used in place of IS_RDONLY(inode).
272 * mnt_want/drop_write() will _keep_ the filesystem
273 * r/w.
274 */
275bool __mnt_is_readonly(struct vfsmount *mnt)
276{
277 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
278}
279EXPORT_SYMBOL_GPL(__mnt_is_readonly);
280
281static inline void mnt_inc_writers(struct mount *mnt)
282{
283#ifdef CONFIG_SMP
284 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
285#else
286 mnt->mnt_writers++;
287#endif
288}
289
290static inline void mnt_dec_writers(struct mount *mnt)
291{
292#ifdef CONFIG_SMP
293 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
294#else
295 mnt->mnt_writers--;
296#endif
297}
298
299static unsigned int mnt_get_writers(struct mount *mnt)
300{
301#ifdef CONFIG_SMP
302 unsigned int count = 0;
303 int cpu;
304
305 for_each_possible_cpu(cpu) {
306 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
307 }
308
309 return count;
310#else
311 return mnt->mnt_writers;
312#endif
313}
314
315static int mnt_is_readonly(struct vfsmount *mnt)
316{
317 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
318 return 1;
319 /*
320 * The barrier pairs with the barrier in sb_start_ro_state_change()
321 * making sure if we don't see s_readonly_remount set yet, we also will
322 * not see any superblock / mount flag changes done by remount.
323 * It also pairs with the barrier in sb_end_ro_state_change()
324 * assuring that if we see s_readonly_remount already cleared, we will
325 * see the values of superblock / mount flags updated by remount.
326 */
327 smp_rmb();
328 return __mnt_is_readonly(mnt);
329}
330
331/*
332 * Most r/o & frozen checks on a fs are for operations that take discrete
333 * amounts of time, like a write() or unlink(). We must keep track of when
334 * those operations start (for permission checks) and when they end, so that we
335 * can determine when writes are able to occur to a filesystem.
336 */
337/**
338 * mnt_get_write_access - get write access to a mount without freeze protection
339 * @m: the mount on which to take a write
340 *
341 * This tells the low-level filesystem that a write is about to be performed to
342 * it, and makes sure that writes are allowed (mnt it read-write) before
343 * returning success. This operation does not protect against filesystem being
344 * frozen. When the write operation is finished, mnt_put_write_access() must be
345 * called. This is effectively a refcount.
346 */
347int mnt_get_write_access(struct vfsmount *m)
348{
349 struct mount *mnt = real_mount(m);
350 int ret = 0;
351
352 preempt_disable();
353 mnt_inc_writers(mnt);
354 /*
355 * The store to mnt_inc_writers must be visible before we pass
356 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
357 * incremented count after it has set MNT_WRITE_HOLD.
358 */
359 smp_mb();
360 might_lock(&mount_lock.lock);
361 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
362 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
363 cpu_relax();
364 } else {
365 /*
366 * This prevents priority inversion, if the task
367 * setting MNT_WRITE_HOLD got preempted on a remote
368 * CPU, and it prevents life lock if the task setting
369 * MNT_WRITE_HOLD has a lower priority and is bound to
370 * the same CPU as the task that is spinning here.
371 */
372 preempt_enable();
373 lock_mount_hash();
374 unlock_mount_hash();
375 preempt_disable();
376 }
377 }
378 /*
379 * The barrier pairs with the barrier sb_start_ro_state_change() making
380 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
381 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
382 * mnt_is_readonly() and bail in case we are racing with remount
383 * read-only.
384 */
385 smp_rmb();
386 if (mnt_is_readonly(m)) {
387 mnt_dec_writers(mnt);
388 ret = -EROFS;
389 }
390 preempt_enable();
391
392 return ret;
393}
394EXPORT_SYMBOL_GPL(mnt_get_write_access);
395
396/**
397 * mnt_want_write - get write access to a mount
398 * @m: the mount on which to take a write
399 *
400 * This tells the low-level filesystem that a write is about to be performed to
401 * it, and makes sure that writes are allowed (mount is read-write, filesystem
402 * is not frozen) before returning success. When the write operation is
403 * finished, mnt_drop_write() must be called. This is effectively a refcount.
404 */
405int mnt_want_write(struct vfsmount *m)
406{
407 int ret;
408
409 sb_start_write(m->mnt_sb);
410 ret = mnt_get_write_access(m);
411 if (ret)
412 sb_end_write(m->mnt_sb);
413 return ret;
414}
415EXPORT_SYMBOL_GPL(mnt_want_write);
416
417/**
418 * mnt_get_write_access_file - get write access to a file's mount
419 * @file: the file who's mount on which to take a write
420 *
421 * This is like mnt_get_write_access, but if @file is already open for write it
422 * skips incrementing mnt_writers (since the open file already has a reference)
423 * and instead only does the check for emergency r/o remounts. This must be
424 * paired with mnt_put_write_access_file.
425 */
426int mnt_get_write_access_file(struct file *file)
427{
428 if (file->f_mode & FMODE_WRITER) {
429 /*
430 * Superblock may have become readonly while there are still
431 * writable fd's, e.g. due to a fs error with errors=remount-ro
432 */
433 if (__mnt_is_readonly(file->f_path.mnt))
434 return -EROFS;
435 return 0;
436 }
437 return mnt_get_write_access(file->f_path.mnt);
438}
439
440/**
441 * mnt_want_write_file - get write access to a file's mount
442 * @file: the file who's mount on which to take a write
443 *
444 * This is like mnt_want_write, but if the file is already open for writing it
445 * skips incrementing mnt_writers (since the open file already has a reference)
446 * and instead only does the freeze protection and the check for emergency r/o
447 * remounts. This must be paired with mnt_drop_write_file.
448 */
449int mnt_want_write_file(struct file *file)
450{
451 int ret;
452
453 sb_start_write(file_inode(file)->i_sb);
454 ret = mnt_get_write_access_file(file);
455 if (ret)
456 sb_end_write(file_inode(file)->i_sb);
457 return ret;
458}
459EXPORT_SYMBOL_GPL(mnt_want_write_file);
460
461/**
462 * mnt_put_write_access - give up write access to a mount
463 * @mnt: the mount on which to give up write access
464 *
465 * Tells the low-level filesystem that we are done
466 * performing writes to it. Must be matched with
467 * mnt_get_write_access() call above.
468 */
469void mnt_put_write_access(struct vfsmount *mnt)
470{
471 preempt_disable();
472 mnt_dec_writers(real_mount(mnt));
473 preempt_enable();
474}
475EXPORT_SYMBOL_GPL(mnt_put_write_access);
476
477/**
478 * mnt_drop_write - give up write access to a mount
479 * @mnt: the mount on which to give up write access
480 *
481 * Tells the low-level filesystem that we are done performing writes to it and
482 * also allows filesystem to be frozen again. Must be matched with
483 * mnt_want_write() call above.
484 */
485void mnt_drop_write(struct vfsmount *mnt)
486{
487 mnt_put_write_access(mnt);
488 sb_end_write(mnt->mnt_sb);
489}
490EXPORT_SYMBOL_GPL(mnt_drop_write);
491
492void mnt_put_write_access_file(struct file *file)
493{
494 if (!(file->f_mode & FMODE_WRITER))
495 mnt_put_write_access(file->f_path.mnt);
496}
497
498void mnt_drop_write_file(struct file *file)
499{
500 mnt_put_write_access_file(file);
501 sb_end_write(file_inode(file)->i_sb);
502}
503EXPORT_SYMBOL(mnt_drop_write_file);
504
505/**
506 * mnt_hold_writers - prevent write access to the given mount
507 * @mnt: mnt to prevent write access to
508 *
509 * Prevents write access to @mnt if there are no active writers for @mnt.
510 * This function needs to be called and return successfully before changing
511 * properties of @mnt that need to remain stable for callers with write access
512 * to @mnt.
513 *
514 * After this functions has been called successfully callers must pair it with
515 * a call to mnt_unhold_writers() in order to stop preventing write access to
516 * @mnt.
517 *
518 * Context: This function expects lock_mount_hash() to be held serializing
519 * setting MNT_WRITE_HOLD.
520 * Return: On success 0 is returned.
521 * On error, -EBUSY is returned.
522 */
523static inline int mnt_hold_writers(struct mount *mnt)
524{
525 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
526 /*
527 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
528 * should be visible before we do.
529 */
530 smp_mb();
531
532 /*
533 * With writers on hold, if this value is zero, then there are
534 * definitely no active writers (although held writers may subsequently
535 * increment the count, they'll have to wait, and decrement it after
536 * seeing MNT_READONLY).
537 *
538 * It is OK to have counter incremented on one CPU and decremented on
539 * another: the sum will add up correctly. The danger would be when we
540 * sum up each counter, if we read a counter before it is incremented,
541 * but then read another CPU's count which it has been subsequently
542 * decremented from -- we would see more decrements than we should.
543 * MNT_WRITE_HOLD protects against this scenario, because
544 * mnt_want_write first increments count, then smp_mb, then spins on
545 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
546 * we're counting up here.
547 */
548 if (mnt_get_writers(mnt) > 0)
549 return -EBUSY;
550
551 return 0;
552}
553
554/**
555 * mnt_unhold_writers - stop preventing write access to the given mount
556 * @mnt: mnt to stop preventing write access to
557 *
558 * Stop preventing write access to @mnt allowing callers to gain write access
559 * to @mnt again.
560 *
561 * This function can only be called after a successful call to
562 * mnt_hold_writers().
563 *
564 * Context: This function expects lock_mount_hash() to be held.
565 */
566static inline void mnt_unhold_writers(struct mount *mnt)
567{
568 /*
569 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
570 * that become unheld will see MNT_READONLY.
571 */
572 smp_wmb();
573 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
574}
575
576static int mnt_make_readonly(struct mount *mnt)
577{
578 int ret;
579
580 ret = mnt_hold_writers(mnt);
581 if (!ret)
582 mnt->mnt.mnt_flags |= MNT_READONLY;
583 mnt_unhold_writers(mnt);
584 return ret;
585}
586
587int sb_prepare_remount_readonly(struct super_block *sb)
588{
589 struct mount *mnt;
590 int err = 0;
591
592 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
593 if (atomic_long_read(&sb->s_remove_count))
594 return -EBUSY;
595
596 lock_mount_hash();
597 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
598 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
599 err = mnt_hold_writers(mnt);
600 if (err)
601 break;
602 }
603 }
604 if (!err && atomic_long_read(&sb->s_remove_count))
605 err = -EBUSY;
606
607 if (!err)
608 sb_start_ro_state_change(sb);
609 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
610 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
611 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
612 }
613 unlock_mount_hash();
614
615 return err;
616}
617
618static void free_vfsmnt(struct mount *mnt)
619{
620 mnt_idmap_put(mnt_idmap(&mnt->mnt));
621 kfree_const(mnt->mnt_devname);
622#ifdef CONFIG_SMP
623 free_percpu(mnt->mnt_pcp);
624#endif
625 kmem_cache_free(mnt_cache, mnt);
626}
627
628static void delayed_free_vfsmnt(struct rcu_head *head)
629{
630 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
631}
632
633/* call under rcu_read_lock */
634int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
635{
636 struct mount *mnt;
637 if (read_seqretry(&mount_lock, seq))
638 return 1;
639 if (bastard == NULL)
640 return 0;
641 mnt = real_mount(bastard);
642 mnt_add_count(mnt, 1);
643 smp_mb(); // see mntput_no_expire()
644 if (likely(!read_seqretry(&mount_lock, seq)))
645 return 0;
646 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
647 mnt_add_count(mnt, -1);
648 return 1;
649 }
650 lock_mount_hash();
651 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
652 mnt_add_count(mnt, -1);
653 unlock_mount_hash();
654 return 1;
655 }
656 unlock_mount_hash();
657 /* caller will mntput() */
658 return -1;
659}
660
661/* call under rcu_read_lock */
662static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
663{
664 int res = __legitimize_mnt(bastard, seq);
665 if (likely(!res))
666 return true;
667 if (unlikely(res < 0)) {
668 rcu_read_unlock();
669 mntput(bastard);
670 rcu_read_lock();
671 }
672 return false;
673}
674
675/**
676 * __lookup_mnt - find first child mount
677 * @mnt: parent mount
678 * @dentry: mountpoint
679 *
680 * If @mnt has a child mount @c mounted @dentry find and return it.
681 *
682 * Note that the child mount @c need not be unique. There are cases
683 * where shadow mounts are created. For example, during mount
684 * propagation when a source mount @mnt whose root got overmounted by a
685 * mount @o after path lookup but before @namespace_sem could be
686 * acquired gets copied and propagated. So @mnt gets copied including
687 * @o. When @mnt is propagated to a destination mount @d that already
688 * has another mount @n mounted at the same mountpoint then the source
689 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
690 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
691 * on @dentry.
692 *
693 * Return: The first child of @mnt mounted @dentry or NULL.
694 */
695struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
696{
697 struct hlist_head *head = m_hash(mnt, dentry);
698 struct mount *p;
699
700 hlist_for_each_entry_rcu(p, head, mnt_hash)
701 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
702 return p;
703 return NULL;
704}
705
706/*
707 * lookup_mnt - Return the first child mount mounted at path
708 *
709 * "First" means first mounted chronologically. If you create the
710 * following mounts:
711 *
712 * mount /dev/sda1 /mnt
713 * mount /dev/sda2 /mnt
714 * mount /dev/sda3 /mnt
715 *
716 * Then lookup_mnt() on the base /mnt dentry in the root mount will
717 * return successively the root dentry and vfsmount of /dev/sda1, then
718 * /dev/sda2, then /dev/sda3, then NULL.
719 *
720 * lookup_mnt takes a reference to the found vfsmount.
721 */
722struct vfsmount *lookup_mnt(const struct path *path)
723{
724 struct mount *child_mnt;
725 struct vfsmount *m;
726 unsigned seq;
727
728 rcu_read_lock();
729 do {
730 seq = read_seqbegin(&mount_lock);
731 child_mnt = __lookup_mnt(path->mnt, path->dentry);
732 m = child_mnt ? &child_mnt->mnt : NULL;
733 } while (!legitimize_mnt(m, seq));
734 rcu_read_unlock();
735 return m;
736}
737
738/*
739 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
740 * current mount namespace.
741 *
742 * The common case is dentries are not mountpoints at all and that
743 * test is handled inline. For the slow case when we are actually
744 * dealing with a mountpoint of some kind, walk through all of the
745 * mounts in the current mount namespace and test to see if the dentry
746 * is a mountpoint.
747 *
748 * The mount_hashtable is not usable in the context because we
749 * need to identify all mounts that may be in the current mount
750 * namespace not just a mount that happens to have some specified
751 * parent mount.
752 */
753bool __is_local_mountpoint(struct dentry *dentry)
754{
755 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
756 struct mount *mnt, *n;
757 bool is_covered = false;
758
759 down_read(&namespace_sem);
760 rbtree_postorder_for_each_entry_safe(mnt, n, &ns->mounts, mnt_node) {
761 is_covered = (mnt->mnt_mountpoint == dentry);
762 if (is_covered)
763 break;
764 }
765 up_read(&namespace_sem);
766
767 return is_covered;
768}
769
770static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
771{
772 struct hlist_head *chain = mp_hash(dentry);
773 struct mountpoint *mp;
774
775 hlist_for_each_entry(mp, chain, m_hash) {
776 if (mp->m_dentry == dentry) {
777 mp->m_count++;
778 return mp;
779 }
780 }
781 return NULL;
782}
783
784static struct mountpoint *get_mountpoint(struct dentry *dentry)
785{
786 struct mountpoint *mp, *new = NULL;
787 int ret;
788
789 if (d_mountpoint(dentry)) {
790 /* might be worth a WARN_ON() */
791 if (d_unlinked(dentry))
792 return ERR_PTR(-ENOENT);
793mountpoint:
794 read_seqlock_excl(&mount_lock);
795 mp = lookup_mountpoint(dentry);
796 read_sequnlock_excl(&mount_lock);
797 if (mp)
798 goto done;
799 }
800
801 if (!new)
802 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
803 if (!new)
804 return ERR_PTR(-ENOMEM);
805
806
807 /* Exactly one processes may set d_mounted */
808 ret = d_set_mounted(dentry);
809
810 /* Someone else set d_mounted? */
811 if (ret == -EBUSY)
812 goto mountpoint;
813
814 /* The dentry is not available as a mountpoint? */
815 mp = ERR_PTR(ret);
816 if (ret)
817 goto done;
818
819 /* Add the new mountpoint to the hash table */
820 read_seqlock_excl(&mount_lock);
821 new->m_dentry = dget(dentry);
822 new->m_count = 1;
823 hlist_add_head(&new->m_hash, mp_hash(dentry));
824 INIT_HLIST_HEAD(&new->m_list);
825 read_sequnlock_excl(&mount_lock);
826
827 mp = new;
828 new = NULL;
829done:
830 kfree(new);
831 return mp;
832}
833
834/*
835 * vfsmount lock must be held. Additionally, the caller is responsible
836 * for serializing calls for given disposal list.
837 */
838static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
839{
840 if (!--mp->m_count) {
841 struct dentry *dentry = mp->m_dentry;
842 BUG_ON(!hlist_empty(&mp->m_list));
843 spin_lock(&dentry->d_lock);
844 dentry->d_flags &= ~DCACHE_MOUNTED;
845 spin_unlock(&dentry->d_lock);
846 dput_to_list(dentry, list);
847 hlist_del(&mp->m_hash);
848 kfree(mp);
849 }
850}
851
852/* called with namespace_lock and vfsmount lock */
853static void put_mountpoint(struct mountpoint *mp)
854{
855 __put_mountpoint(mp, &ex_mountpoints);
856}
857
858static inline int check_mnt(struct mount *mnt)
859{
860 return mnt->mnt_ns == current->nsproxy->mnt_ns;
861}
862
863/*
864 * vfsmount lock must be held for write
865 */
866static void touch_mnt_namespace(struct mnt_namespace *ns)
867{
868 if (ns) {
869 ns->event = ++event;
870 wake_up_interruptible(&ns->poll);
871 }
872}
873
874/*
875 * vfsmount lock must be held for write
876 */
877static void __touch_mnt_namespace(struct mnt_namespace *ns)
878{
879 if (ns && ns->event != event) {
880 ns->event = event;
881 wake_up_interruptible(&ns->poll);
882 }
883}
884
885/*
886 * vfsmount lock must be held for write
887 */
888static struct mountpoint *unhash_mnt(struct mount *mnt)
889{
890 struct mountpoint *mp;
891 mnt->mnt_parent = mnt;
892 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
893 list_del_init(&mnt->mnt_child);
894 hlist_del_init_rcu(&mnt->mnt_hash);
895 hlist_del_init(&mnt->mnt_mp_list);
896 mp = mnt->mnt_mp;
897 mnt->mnt_mp = NULL;
898 return mp;
899}
900
901/*
902 * vfsmount lock must be held for write
903 */
904static void umount_mnt(struct mount *mnt)
905{
906 put_mountpoint(unhash_mnt(mnt));
907}
908
909/*
910 * vfsmount lock must be held for write
911 */
912void mnt_set_mountpoint(struct mount *mnt,
913 struct mountpoint *mp,
914 struct mount *child_mnt)
915{
916 mp->m_count++;
917 mnt_add_count(mnt, 1); /* essentially, that's mntget */
918 child_mnt->mnt_mountpoint = mp->m_dentry;
919 child_mnt->mnt_parent = mnt;
920 child_mnt->mnt_mp = mp;
921 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
922}
923
924/**
925 * mnt_set_mountpoint_beneath - mount a mount beneath another one
926 *
927 * @new_parent: the source mount
928 * @top_mnt: the mount beneath which @new_parent is mounted
929 * @new_mp: the new mountpoint of @top_mnt on @new_parent
930 *
931 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
932 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
933 * @new_mp. And mount @new_parent on the old parent and old
934 * mountpoint of @top_mnt.
935 *
936 * Context: This function expects namespace_lock() and lock_mount_hash()
937 * to have been acquired in that order.
938 */
939static void mnt_set_mountpoint_beneath(struct mount *new_parent,
940 struct mount *top_mnt,
941 struct mountpoint *new_mp)
942{
943 struct mount *old_top_parent = top_mnt->mnt_parent;
944 struct mountpoint *old_top_mp = top_mnt->mnt_mp;
945
946 mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
947 mnt_change_mountpoint(new_parent, new_mp, top_mnt);
948}
949
950
951static void __attach_mnt(struct mount *mnt, struct mount *parent)
952{
953 hlist_add_head_rcu(&mnt->mnt_hash,
954 m_hash(&parent->mnt, mnt->mnt_mountpoint));
955 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
956}
957
958/**
959 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
960 * list of child mounts
961 * @parent: the parent
962 * @mnt: the new mount
963 * @mp: the new mountpoint
964 * @beneath: whether to mount @mnt beneath or on top of @parent
965 *
966 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
967 * to @parent's child mount list and to @mount_hashtable.
968 *
969 * If @beneath is true, remove @mnt from its current parent and
970 * mountpoint and mount it on @mp on @parent, and mount @parent on the
971 * old parent and old mountpoint of @mnt. Finally, attach @parent to
972 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
973 *
974 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
975 * to the correct parent.
976 *
977 * Context: This function expects namespace_lock() and lock_mount_hash()
978 * to have been acquired in that order.
979 */
980static void attach_mnt(struct mount *mnt, struct mount *parent,
981 struct mountpoint *mp, bool beneath)
982{
983 if (beneath)
984 mnt_set_mountpoint_beneath(mnt, parent, mp);
985 else
986 mnt_set_mountpoint(parent, mp, mnt);
987 /*
988 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
989 * beneath @parent then @mnt will need to be attached to
990 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
991 * isn't the same mount as @parent.
992 */
993 __attach_mnt(mnt, mnt->mnt_parent);
994}
995
996void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
997{
998 struct mountpoint *old_mp = mnt->mnt_mp;
999 struct mount *old_parent = mnt->mnt_parent;
1000
1001 list_del_init(&mnt->mnt_child);
1002 hlist_del_init(&mnt->mnt_mp_list);
1003 hlist_del_init_rcu(&mnt->mnt_hash);
1004
1005 attach_mnt(mnt, parent, mp, false);
1006
1007 put_mountpoint(old_mp);
1008 mnt_add_count(old_parent, -1);
1009}
1010
1011static inline struct mount *node_to_mount(struct rb_node *node)
1012{
1013 return node ? rb_entry(node, struct mount, mnt_node) : NULL;
1014}
1015
1016static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
1017{
1018 struct rb_node **link = &ns->mounts.rb_node;
1019 struct rb_node *parent = NULL;
1020
1021 WARN_ON(mnt->mnt.mnt_flags & MNT_ONRB);
1022 mnt->mnt_ns = ns;
1023 while (*link) {
1024 parent = *link;
1025 if (mnt->mnt_id_unique < node_to_mount(parent)->mnt_id_unique)
1026 link = &parent->rb_left;
1027 else
1028 link = &parent->rb_right;
1029 }
1030 rb_link_node(&mnt->mnt_node, parent, link);
1031 rb_insert_color(&mnt->mnt_node, &ns->mounts);
1032 mnt->mnt.mnt_flags |= MNT_ONRB;
1033}
1034
1035/*
1036 * vfsmount lock must be held for write
1037 */
1038static void commit_tree(struct mount *mnt)
1039{
1040 struct mount *parent = mnt->mnt_parent;
1041 struct mount *m;
1042 LIST_HEAD(head);
1043 struct mnt_namespace *n = parent->mnt_ns;
1044
1045 BUG_ON(parent == mnt);
1046
1047 list_add_tail(&head, &mnt->mnt_list);
1048 while (!list_empty(&head)) {
1049 m = list_first_entry(&head, typeof(*m), mnt_list);
1050 list_del(&m->mnt_list);
1051
1052 mnt_add_to_ns(n, m);
1053 }
1054 n->nr_mounts += n->pending_mounts;
1055 n->pending_mounts = 0;
1056
1057 __attach_mnt(mnt, parent);
1058 touch_mnt_namespace(n);
1059}
1060
1061static struct mount *next_mnt(struct mount *p, struct mount *root)
1062{
1063 struct list_head *next = p->mnt_mounts.next;
1064 if (next == &p->mnt_mounts) {
1065 while (1) {
1066 if (p == root)
1067 return NULL;
1068 next = p->mnt_child.next;
1069 if (next != &p->mnt_parent->mnt_mounts)
1070 break;
1071 p = p->mnt_parent;
1072 }
1073 }
1074 return list_entry(next, struct mount, mnt_child);
1075}
1076
1077static struct mount *skip_mnt_tree(struct mount *p)
1078{
1079 struct list_head *prev = p->mnt_mounts.prev;
1080 while (prev != &p->mnt_mounts) {
1081 p = list_entry(prev, struct mount, mnt_child);
1082 prev = p->mnt_mounts.prev;
1083 }
1084 return p;
1085}
1086
1087/**
1088 * vfs_create_mount - Create a mount for a configured superblock
1089 * @fc: The configuration context with the superblock attached
1090 *
1091 * Create a mount to an already configured superblock. If necessary, the
1092 * caller should invoke vfs_get_tree() before calling this.
1093 *
1094 * Note that this does not attach the mount to anything.
1095 */
1096struct vfsmount *vfs_create_mount(struct fs_context *fc)
1097{
1098 struct mount *mnt;
1099
1100 if (!fc->root)
1101 return ERR_PTR(-EINVAL);
1102
1103 mnt = alloc_vfsmnt(fc->source ?: "none");
1104 if (!mnt)
1105 return ERR_PTR(-ENOMEM);
1106
1107 if (fc->sb_flags & SB_KERNMOUNT)
1108 mnt->mnt.mnt_flags = MNT_INTERNAL;
1109
1110 atomic_inc(&fc->root->d_sb->s_active);
1111 mnt->mnt.mnt_sb = fc->root->d_sb;
1112 mnt->mnt.mnt_root = dget(fc->root);
1113 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1114 mnt->mnt_parent = mnt;
1115
1116 lock_mount_hash();
1117 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1118 unlock_mount_hash();
1119 return &mnt->mnt;
1120}
1121EXPORT_SYMBOL(vfs_create_mount);
1122
1123struct vfsmount *fc_mount(struct fs_context *fc)
1124{
1125 int err = vfs_get_tree(fc);
1126 if (!err) {
1127 up_write(&fc->root->d_sb->s_umount);
1128 return vfs_create_mount(fc);
1129 }
1130 return ERR_PTR(err);
1131}
1132EXPORT_SYMBOL(fc_mount);
1133
1134struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1135 int flags, const char *name,
1136 void *data)
1137{
1138 struct fs_context *fc;
1139 struct vfsmount *mnt;
1140 int ret = 0;
1141
1142 if (!type)
1143 return ERR_PTR(-EINVAL);
1144
1145 fc = fs_context_for_mount(type, flags);
1146 if (IS_ERR(fc))
1147 return ERR_CAST(fc);
1148
1149 if (name)
1150 ret = vfs_parse_fs_string(fc, "source",
1151 name, strlen(name));
1152 if (!ret)
1153 ret = parse_monolithic_mount_data(fc, data);
1154 if (!ret)
1155 mnt = fc_mount(fc);
1156 else
1157 mnt = ERR_PTR(ret);
1158
1159 put_fs_context(fc);
1160 return mnt;
1161}
1162EXPORT_SYMBOL_GPL(vfs_kern_mount);
1163
1164struct vfsmount *
1165vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1166 const char *name, void *data)
1167{
1168 /* Until it is worked out how to pass the user namespace
1169 * through from the parent mount to the submount don't support
1170 * unprivileged mounts with submounts.
1171 */
1172 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1173 return ERR_PTR(-EPERM);
1174
1175 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1176}
1177EXPORT_SYMBOL_GPL(vfs_submount);
1178
1179static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1180 int flag)
1181{
1182 struct super_block *sb = old->mnt.mnt_sb;
1183 struct mount *mnt;
1184 int err;
1185
1186 mnt = alloc_vfsmnt(old->mnt_devname);
1187 if (!mnt)
1188 return ERR_PTR(-ENOMEM);
1189
1190 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1191 mnt->mnt_group_id = 0; /* not a peer of original */
1192 else
1193 mnt->mnt_group_id = old->mnt_group_id;
1194
1195 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1196 err = mnt_alloc_group_id(mnt);
1197 if (err)
1198 goto out_free;
1199 }
1200
1201 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1202 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_ONRB);
1203
1204 atomic_inc(&sb->s_active);
1205 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1206
1207 mnt->mnt.mnt_sb = sb;
1208 mnt->mnt.mnt_root = dget(root);
1209 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1210 mnt->mnt_parent = mnt;
1211 lock_mount_hash();
1212 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1213 unlock_mount_hash();
1214
1215 if ((flag & CL_SLAVE) ||
1216 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1217 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1218 mnt->mnt_master = old;
1219 CLEAR_MNT_SHARED(mnt);
1220 } else if (!(flag & CL_PRIVATE)) {
1221 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1222 list_add(&mnt->mnt_share, &old->mnt_share);
1223 if (IS_MNT_SLAVE(old))
1224 list_add(&mnt->mnt_slave, &old->mnt_slave);
1225 mnt->mnt_master = old->mnt_master;
1226 } else {
1227 CLEAR_MNT_SHARED(mnt);
1228 }
1229 if (flag & CL_MAKE_SHARED)
1230 set_mnt_shared(mnt);
1231
1232 /* stick the duplicate mount on the same expiry list
1233 * as the original if that was on one */
1234 if (flag & CL_EXPIRE) {
1235 if (!list_empty(&old->mnt_expire))
1236 list_add(&mnt->mnt_expire, &old->mnt_expire);
1237 }
1238
1239 return mnt;
1240
1241 out_free:
1242 mnt_free_id(mnt);
1243 free_vfsmnt(mnt);
1244 return ERR_PTR(err);
1245}
1246
1247static void cleanup_mnt(struct mount *mnt)
1248{
1249 struct hlist_node *p;
1250 struct mount *m;
1251 /*
1252 * The warning here probably indicates that somebody messed
1253 * up a mnt_want/drop_write() pair. If this happens, the
1254 * filesystem was probably unable to make r/w->r/o transitions.
1255 * The locking used to deal with mnt_count decrement provides barriers,
1256 * so mnt_get_writers() below is safe.
1257 */
1258 WARN_ON(mnt_get_writers(mnt));
1259 if (unlikely(mnt->mnt_pins.first))
1260 mnt_pin_kill(mnt);
1261 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1262 hlist_del(&m->mnt_umount);
1263 mntput(&m->mnt);
1264 }
1265 fsnotify_vfsmount_delete(&mnt->mnt);
1266 dput(mnt->mnt.mnt_root);
1267 deactivate_super(mnt->mnt.mnt_sb);
1268 mnt_free_id(mnt);
1269 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1270}
1271
1272static void __cleanup_mnt(struct rcu_head *head)
1273{
1274 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1275}
1276
1277static LLIST_HEAD(delayed_mntput_list);
1278static void delayed_mntput(struct work_struct *unused)
1279{
1280 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1281 struct mount *m, *t;
1282
1283 llist_for_each_entry_safe(m, t, node, mnt_llist)
1284 cleanup_mnt(m);
1285}
1286static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1287
1288static void mntput_no_expire(struct mount *mnt)
1289{
1290 LIST_HEAD(list);
1291 int count;
1292
1293 rcu_read_lock();
1294 if (likely(READ_ONCE(mnt->mnt_ns))) {
1295 /*
1296 * Since we don't do lock_mount_hash() here,
1297 * ->mnt_ns can change under us. However, if it's
1298 * non-NULL, then there's a reference that won't
1299 * be dropped until after an RCU delay done after
1300 * turning ->mnt_ns NULL. So if we observe it
1301 * non-NULL under rcu_read_lock(), the reference
1302 * we are dropping is not the final one.
1303 */
1304 mnt_add_count(mnt, -1);
1305 rcu_read_unlock();
1306 return;
1307 }
1308 lock_mount_hash();
1309 /*
1310 * make sure that if __legitimize_mnt() has not seen us grab
1311 * mount_lock, we'll see their refcount increment here.
1312 */
1313 smp_mb();
1314 mnt_add_count(mnt, -1);
1315 count = mnt_get_count(mnt);
1316 if (count != 0) {
1317 WARN_ON(count < 0);
1318 rcu_read_unlock();
1319 unlock_mount_hash();
1320 return;
1321 }
1322 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1323 rcu_read_unlock();
1324 unlock_mount_hash();
1325 return;
1326 }
1327 mnt->mnt.mnt_flags |= MNT_DOOMED;
1328 rcu_read_unlock();
1329
1330 list_del(&mnt->mnt_instance);
1331
1332 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1333 struct mount *p, *tmp;
1334 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1335 __put_mountpoint(unhash_mnt(p), &list);
1336 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1337 }
1338 }
1339 unlock_mount_hash();
1340 shrink_dentry_list(&list);
1341
1342 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1343 struct task_struct *task = current;
1344 if (likely(!(task->flags & PF_KTHREAD))) {
1345 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1346 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1347 return;
1348 }
1349 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1350 schedule_delayed_work(&delayed_mntput_work, 1);
1351 return;
1352 }
1353 cleanup_mnt(mnt);
1354}
1355
1356void mntput(struct vfsmount *mnt)
1357{
1358 if (mnt) {
1359 struct mount *m = real_mount(mnt);
1360 /* avoid cacheline pingpong */
1361 if (unlikely(m->mnt_expiry_mark))
1362 WRITE_ONCE(m->mnt_expiry_mark, 0);
1363 mntput_no_expire(m);
1364 }
1365}
1366EXPORT_SYMBOL(mntput);
1367
1368struct vfsmount *mntget(struct vfsmount *mnt)
1369{
1370 if (mnt)
1371 mnt_add_count(real_mount(mnt), 1);
1372 return mnt;
1373}
1374EXPORT_SYMBOL(mntget);
1375
1376/*
1377 * Make a mount point inaccessible to new lookups.
1378 * Because there may still be current users, the caller MUST WAIT
1379 * for an RCU grace period before destroying the mount point.
1380 */
1381void mnt_make_shortterm(struct vfsmount *mnt)
1382{
1383 if (mnt)
1384 real_mount(mnt)->mnt_ns = NULL;
1385}
1386
1387/**
1388 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1389 * @path: path to check
1390 *
1391 * d_mountpoint() can only be used reliably to establish if a dentry is
1392 * not mounted in any namespace and that common case is handled inline.
1393 * d_mountpoint() isn't aware of the possibility there may be multiple
1394 * mounts using a given dentry in a different namespace. This function
1395 * checks if the passed in path is a mountpoint rather than the dentry
1396 * alone.
1397 */
1398bool path_is_mountpoint(const struct path *path)
1399{
1400 unsigned seq;
1401 bool res;
1402
1403 if (!d_mountpoint(path->dentry))
1404 return false;
1405
1406 rcu_read_lock();
1407 do {
1408 seq = read_seqbegin(&mount_lock);
1409 res = __path_is_mountpoint(path);
1410 } while (read_seqretry(&mount_lock, seq));
1411 rcu_read_unlock();
1412
1413 return res;
1414}
1415EXPORT_SYMBOL(path_is_mountpoint);
1416
1417struct vfsmount *mnt_clone_internal(const struct path *path)
1418{
1419 struct mount *p;
1420 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1421 if (IS_ERR(p))
1422 return ERR_CAST(p);
1423 p->mnt.mnt_flags |= MNT_INTERNAL;
1424 return &p->mnt;
1425}
1426
1427/*
1428 * Returns the mount which either has the specified mnt_id, or has the next
1429 * smallest id afer the specified one.
1430 */
1431static struct mount *mnt_find_id_at(struct mnt_namespace *ns, u64 mnt_id)
1432{
1433 struct rb_node *node = ns->mounts.rb_node;
1434 struct mount *ret = NULL;
1435
1436 while (node) {
1437 struct mount *m = node_to_mount(node);
1438
1439 if (mnt_id <= m->mnt_id_unique) {
1440 ret = node_to_mount(node);
1441 if (mnt_id == m->mnt_id_unique)
1442 break;
1443 node = node->rb_left;
1444 } else {
1445 node = node->rb_right;
1446 }
1447 }
1448 return ret;
1449}
1450
1451#ifdef CONFIG_PROC_FS
1452
1453/* iterator; we want it to have access to namespace_sem, thus here... */
1454static void *m_start(struct seq_file *m, loff_t *pos)
1455{
1456 struct proc_mounts *p = m->private;
1457
1458 down_read(&namespace_sem);
1459
1460 return mnt_find_id_at(p->ns, *pos);
1461}
1462
1463static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1464{
1465 struct mount *next = NULL, *mnt = v;
1466 struct rb_node *node = rb_next(&mnt->mnt_node);
1467
1468 ++*pos;
1469 if (node) {
1470 next = node_to_mount(node);
1471 *pos = next->mnt_id_unique;
1472 }
1473 return next;
1474}
1475
1476static void m_stop(struct seq_file *m, void *v)
1477{
1478 up_read(&namespace_sem);
1479}
1480
1481static int m_show(struct seq_file *m, void *v)
1482{
1483 struct proc_mounts *p = m->private;
1484 struct mount *r = v;
1485 return p->show(m, &r->mnt);
1486}
1487
1488const struct seq_operations mounts_op = {
1489 .start = m_start,
1490 .next = m_next,
1491 .stop = m_stop,
1492 .show = m_show,
1493};
1494
1495#endif /* CONFIG_PROC_FS */
1496
1497/**
1498 * may_umount_tree - check if a mount tree is busy
1499 * @m: root of mount tree
1500 *
1501 * This is called to check if a tree of mounts has any
1502 * open files, pwds, chroots or sub mounts that are
1503 * busy.
1504 */
1505int may_umount_tree(struct vfsmount *m)
1506{
1507 struct mount *mnt = real_mount(m);
1508 int actual_refs = 0;
1509 int minimum_refs = 0;
1510 struct mount *p;
1511 BUG_ON(!m);
1512
1513 /* write lock needed for mnt_get_count */
1514 lock_mount_hash();
1515 for (p = mnt; p; p = next_mnt(p, mnt)) {
1516 actual_refs += mnt_get_count(p);
1517 minimum_refs += 2;
1518 }
1519 unlock_mount_hash();
1520
1521 if (actual_refs > minimum_refs)
1522 return 0;
1523
1524 return 1;
1525}
1526
1527EXPORT_SYMBOL(may_umount_tree);
1528
1529/**
1530 * may_umount - check if a mount point is busy
1531 * @mnt: root of mount
1532 *
1533 * This is called to check if a mount point has any
1534 * open files, pwds, chroots or sub mounts. If the
1535 * mount has sub mounts this will return busy
1536 * regardless of whether the sub mounts are busy.
1537 *
1538 * Doesn't take quota and stuff into account. IOW, in some cases it will
1539 * give false negatives. The main reason why it's here is that we need
1540 * a non-destructive way to look for easily umountable filesystems.
1541 */
1542int may_umount(struct vfsmount *mnt)
1543{
1544 int ret = 1;
1545 down_read(&namespace_sem);
1546 lock_mount_hash();
1547 if (propagate_mount_busy(real_mount(mnt), 2))
1548 ret = 0;
1549 unlock_mount_hash();
1550 up_read(&namespace_sem);
1551 return ret;
1552}
1553
1554EXPORT_SYMBOL(may_umount);
1555
1556static void namespace_unlock(void)
1557{
1558 struct hlist_head head;
1559 struct hlist_node *p;
1560 struct mount *m;
1561 LIST_HEAD(list);
1562
1563 hlist_move_list(&unmounted, &head);
1564 list_splice_init(&ex_mountpoints, &list);
1565
1566 up_write(&namespace_sem);
1567
1568 shrink_dentry_list(&list);
1569
1570 if (likely(hlist_empty(&head)))
1571 return;
1572
1573 synchronize_rcu_expedited();
1574
1575 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1576 hlist_del(&m->mnt_umount);
1577 mntput(&m->mnt);
1578 }
1579}
1580
1581static inline void namespace_lock(void)
1582{
1583 down_write(&namespace_sem);
1584}
1585
1586enum umount_tree_flags {
1587 UMOUNT_SYNC = 1,
1588 UMOUNT_PROPAGATE = 2,
1589 UMOUNT_CONNECTED = 4,
1590};
1591
1592static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1593{
1594 /* Leaving mounts connected is only valid for lazy umounts */
1595 if (how & UMOUNT_SYNC)
1596 return true;
1597
1598 /* A mount without a parent has nothing to be connected to */
1599 if (!mnt_has_parent(mnt))
1600 return true;
1601
1602 /* Because the reference counting rules change when mounts are
1603 * unmounted and connected, umounted mounts may not be
1604 * connected to mounted mounts.
1605 */
1606 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1607 return true;
1608
1609 /* Has it been requested that the mount remain connected? */
1610 if (how & UMOUNT_CONNECTED)
1611 return false;
1612
1613 /* Is the mount locked such that it needs to remain connected? */
1614 if (IS_MNT_LOCKED(mnt))
1615 return false;
1616
1617 /* By default disconnect the mount */
1618 return true;
1619}
1620
1621/*
1622 * mount_lock must be held
1623 * namespace_sem must be held for write
1624 */
1625static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1626{
1627 LIST_HEAD(tmp_list);
1628 struct mount *p;
1629
1630 if (how & UMOUNT_PROPAGATE)
1631 propagate_mount_unlock(mnt);
1632
1633 /* Gather the mounts to umount */
1634 for (p = mnt; p; p = next_mnt(p, mnt)) {
1635 p->mnt.mnt_flags |= MNT_UMOUNT;
1636 if (p->mnt.mnt_flags & MNT_ONRB)
1637 move_from_ns(p, &tmp_list);
1638 else
1639 list_move(&p->mnt_list, &tmp_list);
1640 }
1641
1642 /* Hide the mounts from mnt_mounts */
1643 list_for_each_entry(p, &tmp_list, mnt_list) {
1644 list_del_init(&p->mnt_child);
1645 }
1646
1647 /* Add propogated mounts to the tmp_list */
1648 if (how & UMOUNT_PROPAGATE)
1649 propagate_umount(&tmp_list);
1650
1651 while (!list_empty(&tmp_list)) {
1652 struct mnt_namespace *ns;
1653 bool disconnect;
1654 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1655 list_del_init(&p->mnt_expire);
1656 list_del_init(&p->mnt_list);
1657 ns = p->mnt_ns;
1658 if (ns) {
1659 ns->nr_mounts--;
1660 __touch_mnt_namespace(ns);
1661 }
1662 p->mnt_ns = NULL;
1663 if (how & UMOUNT_SYNC)
1664 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1665
1666 disconnect = disconnect_mount(p, how);
1667 if (mnt_has_parent(p)) {
1668 mnt_add_count(p->mnt_parent, -1);
1669 if (!disconnect) {
1670 /* Don't forget about p */
1671 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1672 } else {
1673 umount_mnt(p);
1674 }
1675 }
1676 change_mnt_propagation(p, MS_PRIVATE);
1677 if (disconnect)
1678 hlist_add_head(&p->mnt_umount, &unmounted);
1679 }
1680}
1681
1682static void shrink_submounts(struct mount *mnt);
1683
1684static int do_umount_root(struct super_block *sb)
1685{
1686 int ret = 0;
1687
1688 down_write(&sb->s_umount);
1689 if (!sb_rdonly(sb)) {
1690 struct fs_context *fc;
1691
1692 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1693 SB_RDONLY);
1694 if (IS_ERR(fc)) {
1695 ret = PTR_ERR(fc);
1696 } else {
1697 ret = parse_monolithic_mount_data(fc, NULL);
1698 if (!ret)
1699 ret = reconfigure_super(fc);
1700 put_fs_context(fc);
1701 }
1702 }
1703 up_write(&sb->s_umount);
1704 return ret;
1705}
1706
1707static int do_umount(struct mount *mnt, int flags)
1708{
1709 struct super_block *sb = mnt->mnt.mnt_sb;
1710 int retval;
1711
1712 retval = security_sb_umount(&mnt->mnt, flags);
1713 if (retval)
1714 return retval;
1715
1716 /*
1717 * Allow userspace to request a mountpoint be expired rather than
1718 * unmounting unconditionally. Unmount only happens if:
1719 * (1) the mark is already set (the mark is cleared by mntput())
1720 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1721 */
1722 if (flags & MNT_EXPIRE) {
1723 if (&mnt->mnt == current->fs->root.mnt ||
1724 flags & (MNT_FORCE | MNT_DETACH))
1725 return -EINVAL;
1726
1727 /*
1728 * probably don't strictly need the lock here if we examined
1729 * all race cases, but it's a slowpath.
1730 */
1731 lock_mount_hash();
1732 if (mnt_get_count(mnt) != 2) {
1733 unlock_mount_hash();
1734 return -EBUSY;
1735 }
1736 unlock_mount_hash();
1737
1738 if (!xchg(&mnt->mnt_expiry_mark, 1))
1739 return -EAGAIN;
1740 }
1741
1742 /*
1743 * If we may have to abort operations to get out of this
1744 * mount, and they will themselves hold resources we must
1745 * allow the fs to do things. In the Unix tradition of
1746 * 'Gee thats tricky lets do it in userspace' the umount_begin
1747 * might fail to complete on the first run through as other tasks
1748 * must return, and the like. Thats for the mount program to worry
1749 * about for the moment.
1750 */
1751
1752 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1753 sb->s_op->umount_begin(sb);
1754 }
1755
1756 /*
1757 * No sense to grab the lock for this test, but test itself looks
1758 * somewhat bogus. Suggestions for better replacement?
1759 * Ho-hum... In principle, we might treat that as umount + switch
1760 * to rootfs. GC would eventually take care of the old vfsmount.
1761 * Actually it makes sense, especially if rootfs would contain a
1762 * /reboot - static binary that would close all descriptors and
1763 * call reboot(9). Then init(8) could umount root and exec /reboot.
1764 */
1765 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1766 /*
1767 * Special case for "unmounting" root ...
1768 * we just try to remount it readonly.
1769 */
1770 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1771 return -EPERM;
1772 return do_umount_root(sb);
1773 }
1774
1775 namespace_lock();
1776 lock_mount_hash();
1777
1778 /* Recheck MNT_LOCKED with the locks held */
1779 retval = -EINVAL;
1780 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1781 goto out;
1782
1783 event++;
1784 if (flags & MNT_DETACH) {
1785 if (mnt->mnt.mnt_flags & MNT_ONRB ||
1786 !list_empty(&mnt->mnt_list))
1787 umount_tree(mnt, UMOUNT_PROPAGATE);
1788 retval = 0;
1789 } else {
1790 shrink_submounts(mnt);
1791 retval = -EBUSY;
1792 if (!propagate_mount_busy(mnt, 2)) {
1793 if (mnt->mnt.mnt_flags & MNT_ONRB ||
1794 !list_empty(&mnt->mnt_list))
1795 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1796 retval = 0;
1797 }
1798 }
1799out:
1800 unlock_mount_hash();
1801 namespace_unlock();
1802 return retval;
1803}
1804
1805/*
1806 * __detach_mounts - lazily unmount all mounts on the specified dentry
1807 *
1808 * During unlink, rmdir, and d_drop it is possible to loose the path
1809 * to an existing mountpoint, and wind up leaking the mount.
1810 * detach_mounts allows lazily unmounting those mounts instead of
1811 * leaking them.
1812 *
1813 * The caller may hold dentry->d_inode->i_mutex.
1814 */
1815void __detach_mounts(struct dentry *dentry)
1816{
1817 struct mountpoint *mp;
1818 struct mount *mnt;
1819
1820 namespace_lock();
1821 lock_mount_hash();
1822 mp = lookup_mountpoint(dentry);
1823 if (!mp)
1824 goto out_unlock;
1825
1826 event++;
1827 while (!hlist_empty(&mp->m_list)) {
1828 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1829 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1830 umount_mnt(mnt);
1831 hlist_add_head(&mnt->mnt_umount, &unmounted);
1832 }
1833 else umount_tree(mnt, UMOUNT_CONNECTED);
1834 }
1835 put_mountpoint(mp);
1836out_unlock:
1837 unlock_mount_hash();
1838 namespace_unlock();
1839}
1840
1841/*
1842 * Is the caller allowed to modify his namespace?
1843 */
1844bool may_mount(void)
1845{
1846 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1847}
1848
1849/**
1850 * path_mounted - check whether path is mounted
1851 * @path: path to check
1852 *
1853 * Determine whether @path refers to the root of a mount.
1854 *
1855 * Return: true if @path is the root of a mount, false if not.
1856 */
1857static inline bool path_mounted(const struct path *path)
1858{
1859 return path->mnt->mnt_root == path->dentry;
1860}
1861
1862static void warn_mandlock(void)
1863{
1864 pr_warn_once("=======================================================\n"
1865 "WARNING: The mand mount option has been deprecated and\n"
1866 " and is ignored by this kernel. Remove the mand\n"
1867 " option from the mount to silence this warning.\n"
1868 "=======================================================\n");
1869}
1870
1871static int can_umount(const struct path *path, int flags)
1872{
1873 struct mount *mnt = real_mount(path->mnt);
1874
1875 if (!may_mount())
1876 return -EPERM;
1877 if (!path_mounted(path))
1878 return -EINVAL;
1879 if (!check_mnt(mnt))
1880 return -EINVAL;
1881 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1882 return -EINVAL;
1883 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1884 return -EPERM;
1885 return 0;
1886}
1887
1888// caller is responsible for flags being sane
1889int path_umount(struct path *path, int flags)
1890{
1891 struct mount *mnt = real_mount(path->mnt);
1892 int ret;
1893
1894 ret = can_umount(path, flags);
1895 if (!ret)
1896 ret = do_umount(mnt, flags);
1897
1898 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1899 dput(path->dentry);
1900 mntput_no_expire(mnt);
1901 return ret;
1902}
1903
1904static int ksys_umount(char __user *name, int flags)
1905{
1906 int lookup_flags = LOOKUP_MOUNTPOINT;
1907 struct path path;
1908 int ret;
1909
1910 // basic validity checks done first
1911 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1912 return -EINVAL;
1913
1914 if (!(flags & UMOUNT_NOFOLLOW))
1915 lookup_flags |= LOOKUP_FOLLOW;
1916 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1917 if (ret)
1918 return ret;
1919 return path_umount(&path, flags);
1920}
1921
1922SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1923{
1924 return ksys_umount(name, flags);
1925}
1926
1927#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1928
1929/*
1930 * The 2.0 compatible umount. No flags.
1931 */
1932SYSCALL_DEFINE1(oldumount, char __user *, name)
1933{
1934 return ksys_umount(name, 0);
1935}
1936
1937#endif
1938
1939static bool is_mnt_ns_file(struct dentry *dentry)
1940{
1941 /* Is this a proxy for a mount namespace? */
1942 return dentry->d_op == &ns_dentry_operations &&
1943 dentry->d_fsdata == &mntns_operations;
1944}
1945
1946static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1947{
1948 return container_of(ns, struct mnt_namespace, ns);
1949}
1950
1951struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1952{
1953 return &mnt->ns;
1954}
1955
1956static bool mnt_ns_loop(struct dentry *dentry)
1957{
1958 /* Could bind mounting the mount namespace inode cause a
1959 * mount namespace loop?
1960 */
1961 struct mnt_namespace *mnt_ns;
1962 if (!is_mnt_ns_file(dentry))
1963 return false;
1964
1965 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1966 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1967}
1968
1969struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1970 int flag)
1971{
1972 struct mount *res, *p, *q, *r, *parent;
1973
1974 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1975 return ERR_PTR(-EINVAL);
1976
1977 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1978 return ERR_PTR(-EINVAL);
1979
1980 res = q = clone_mnt(mnt, dentry, flag);
1981 if (IS_ERR(q))
1982 return q;
1983
1984 q->mnt_mountpoint = mnt->mnt_mountpoint;
1985
1986 p = mnt;
1987 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1988 struct mount *s;
1989 if (!is_subdir(r->mnt_mountpoint, dentry))
1990 continue;
1991
1992 for (s = r; s; s = next_mnt(s, r)) {
1993 if (!(flag & CL_COPY_UNBINDABLE) &&
1994 IS_MNT_UNBINDABLE(s)) {
1995 if (s->mnt.mnt_flags & MNT_LOCKED) {
1996 /* Both unbindable and locked. */
1997 q = ERR_PTR(-EPERM);
1998 goto out;
1999 } else {
2000 s = skip_mnt_tree(s);
2001 continue;
2002 }
2003 }
2004 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2005 is_mnt_ns_file(s->mnt.mnt_root)) {
2006 s = skip_mnt_tree(s);
2007 continue;
2008 }
2009 while (p != s->mnt_parent) {
2010 p = p->mnt_parent;
2011 q = q->mnt_parent;
2012 }
2013 p = s;
2014 parent = q;
2015 q = clone_mnt(p, p->mnt.mnt_root, flag);
2016 if (IS_ERR(q))
2017 goto out;
2018 lock_mount_hash();
2019 list_add_tail(&q->mnt_list, &res->mnt_list);
2020 attach_mnt(q, parent, p->mnt_mp, false);
2021 unlock_mount_hash();
2022 }
2023 }
2024 return res;
2025out:
2026 if (res) {
2027 lock_mount_hash();
2028 umount_tree(res, UMOUNT_SYNC);
2029 unlock_mount_hash();
2030 }
2031 return q;
2032}
2033
2034/* Caller should check returned pointer for errors */
2035
2036struct vfsmount *collect_mounts(const struct path *path)
2037{
2038 struct mount *tree;
2039 namespace_lock();
2040 if (!check_mnt(real_mount(path->mnt)))
2041 tree = ERR_PTR(-EINVAL);
2042 else
2043 tree = copy_tree(real_mount(path->mnt), path->dentry,
2044 CL_COPY_ALL | CL_PRIVATE);
2045 namespace_unlock();
2046 if (IS_ERR(tree))
2047 return ERR_CAST(tree);
2048 return &tree->mnt;
2049}
2050
2051static void free_mnt_ns(struct mnt_namespace *);
2052static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2053
2054void dissolve_on_fput(struct vfsmount *mnt)
2055{
2056 struct mnt_namespace *ns;
2057 namespace_lock();
2058 lock_mount_hash();
2059 ns = real_mount(mnt)->mnt_ns;
2060 if (ns) {
2061 if (is_anon_ns(ns))
2062 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2063 else
2064 ns = NULL;
2065 }
2066 unlock_mount_hash();
2067 namespace_unlock();
2068 if (ns)
2069 free_mnt_ns(ns);
2070}
2071
2072void drop_collected_mounts(struct vfsmount *mnt)
2073{
2074 namespace_lock();
2075 lock_mount_hash();
2076 umount_tree(real_mount(mnt), 0);
2077 unlock_mount_hash();
2078 namespace_unlock();
2079}
2080
2081static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2082{
2083 struct mount *child;
2084
2085 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2086 if (!is_subdir(child->mnt_mountpoint, dentry))
2087 continue;
2088
2089 if (child->mnt.mnt_flags & MNT_LOCKED)
2090 return true;
2091 }
2092 return false;
2093}
2094
2095/**
2096 * clone_private_mount - create a private clone of a path
2097 * @path: path to clone
2098 *
2099 * This creates a new vfsmount, which will be the clone of @path. The new mount
2100 * will not be attached anywhere in the namespace and will be private (i.e.
2101 * changes to the originating mount won't be propagated into this).
2102 *
2103 * Release with mntput().
2104 */
2105struct vfsmount *clone_private_mount(const struct path *path)
2106{
2107 struct mount *old_mnt = real_mount(path->mnt);
2108 struct mount *new_mnt;
2109
2110 down_read(&namespace_sem);
2111 if (IS_MNT_UNBINDABLE(old_mnt))
2112 goto invalid;
2113
2114 if (!check_mnt(old_mnt))
2115 goto invalid;
2116
2117 if (has_locked_children(old_mnt, path->dentry))
2118 goto invalid;
2119
2120 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2121 up_read(&namespace_sem);
2122
2123 if (IS_ERR(new_mnt))
2124 return ERR_CAST(new_mnt);
2125
2126 /* Longterm mount to be removed by kern_unmount*() */
2127 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2128
2129 return &new_mnt->mnt;
2130
2131invalid:
2132 up_read(&namespace_sem);
2133 return ERR_PTR(-EINVAL);
2134}
2135EXPORT_SYMBOL_GPL(clone_private_mount);
2136
2137int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2138 struct vfsmount *root)
2139{
2140 struct mount *mnt;
2141 int res = f(root, arg);
2142 if (res)
2143 return res;
2144 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2145 res = f(&mnt->mnt, arg);
2146 if (res)
2147 return res;
2148 }
2149 return 0;
2150}
2151
2152static void lock_mnt_tree(struct mount *mnt)
2153{
2154 struct mount *p;
2155
2156 for (p = mnt; p; p = next_mnt(p, mnt)) {
2157 int flags = p->mnt.mnt_flags;
2158 /* Don't allow unprivileged users to change mount flags */
2159 flags |= MNT_LOCK_ATIME;
2160
2161 if (flags & MNT_READONLY)
2162 flags |= MNT_LOCK_READONLY;
2163
2164 if (flags & MNT_NODEV)
2165 flags |= MNT_LOCK_NODEV;
2166
2167 if (flags & MNT_NOSUID)
2168 flags |= MNT_LOCK_NOSUID;
2169
2170 if (flags & MNT_NOEXEC)
2171 flags |= MNT_LOCK_NOEXEC;
2172 /* Don't allow unprivileged users to reveal what is under a mount */
2173 if (list_empty(&p->mnt_expire))
2174 flags |= MNT_LOCKED;
2175 p->mnt.mnt_flags = flags;
2176 }
2177}
2178
2179static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2180{
2181 struct mount *p;
2182
2183 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2184 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2185 mnt_release_group_id(p);
2186 }
2187}
2188
2189static int invent_group_ids(struct mount *mnt, bool recurse)
2190{
2191 struct mount *p;
2192
2193 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2194 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2195 int err = mnt_alloc_group_id(p);
2196 if (err) {
2197 cleanup_group_ids(mnt, p);
2198 return err;
2199 }
2200 }
2201 }
2202
2203 return 0;
2204}
2205
2206int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2207{
2208 unsigned int max = READ_ONCE(sysctl_mount_max);
2209 unsigned int mounts = 0;
2210 struct mount *p;
2211
2212 if (ns->nr_mounts >= max)
2213 return -ENOSPC;
2214 max -= ns->nr_mounts;
2215 if (ns->pending_mounts >= max)
2216 return -ENOSPC;
2217 max -= ns->pending_mounts;
2218
2219 for (p = mnt; p; p = next_mnt(p, mnt))
2220 mounts++;
2221
2222 if (mounts > max)
2223 return -ENOSPC;
2224
2225 ns->pending_mounts += mounts;
2226 return 0;
2227}
2228
2229enum mnt_tree_flags_t {
2230 MNT_TREE_MOVE = BIT(0),
2231 MNT_TREE_BENEATH = BIT(1),
2232};
2233
2234/**
2235 * attach_recursive_mnt - attach a source mount tree
2236 * @source_mnt: mount tree to be attached
2237 * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath
2238 * @dest_mp: the mountpoint @source_mnt will be mounted at
2239 * @flags: modify how @source_mnt is supposed to be attached
2240 *
2241 * NOTE: in the table below explains the semantics when a source mount
2242 * of a given type is attached to a destination mount of a given type.
2243 * ---------------------------------------------------------------------------
2244 * | BIND MOUNT OPERATION |
2245 * |**************************************************************************
2246 * | source-->| shared | private | slave | unbindable |
2247 * | dest | | | | |
2248 * | | | | | | |
2249 * | v | | | | |
2250 * |**************************************************************************
2251 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2252 * | | | | | |
2253 * |non-shared| shared (+) | private | slave (*) | invalid |
2254 * ***************************************************************************
2255 * A bind operation clones the source mount and mounts the clone on the
2256 * destination mount.
2257 *
2258 * (++) the cloned mount is propagated to all the mounts in the propagation
2259 * tree of the destination mount and the cloned mount is added to
2260 * the peer group of the source mount.
2261 * (+) the cloned mount is created under the destination mount and is marked
2262 * as shared. The cloned mount is added to the peer group of the source
2263 * mount.
2264 * (+++) the mount is propagated to all the mounts in the propagation tree
2265 * of the destination mount and the cloned mount is made slave
2266 * of the same master as that of the source mount. The cloned mount
2267 * is marked as 'shared and slave'.
2268 * (*) the cloned mount is made a slave of the same master as that of the
2269 * source mount.
2270 *
2271 * ---------------------------------------------------------------------------
2272 * | MOVE MOUNT OPERATION |
2273 * |**************************************************************************
2274 * | source-->| shared | private | slave | unbindable |
2275 * | dest | | | | |
2276 * | | | | | | |
2277 * | v | | | | |
2278 * |**************************************************************************
2279 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2280 * | | | | | |
2281 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2282 * ***************************************************************************
2283 *
2284 * (+) the mount is moved to the destination. And is then propagated to
2285 * all the mounts in the propagation tree of the destination mount.
2286 * (+*) the mount is moved to the destination.
2287 * (+++) the mount is moved to the destination and is then propagated to
2288 * all the mounts belonging to the destination mount's propagation tree.
2289 * the mount is marked as 'shared and slave'.
2290 * (*) the mount continues to be a slave at the new location.
2291 *
2292 * if the source mount is a tree, the operations explained above is
2293 * applied to each mount in the tree.
2294 * Must be called without spinlocks held, since this function can sleep
2295 * in allocations.
2296 *
2297 * Context: The function expects namespace_lock() to be held.
2298 * Return: If @source_mnt was successfully attached 0 is returned.
2299 * Otherwise a negative error code is returned.
2300 */
2301static int attach_recursive_mnt(struct mount *source_mnt,
2302 struct mount *top_mnt,
2303 struct mountpoint *dest_mp,
2304 enum mnt_tree_flags_t flags)
2305{
2306 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2307 HLIST_HEAD(tree_list);
2308 struct mnt_namespace *ns = top_mnt->mnt_ns;
2309 struct mountpoint *smp;
2310 struct mount *child, *dest_mnt, *p;
2311 struct hlist_node *n;
2312 int err = 0;
2313 bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
2314
2315 /*
2316 * Preallocate a mountpoint in case the new mounts need to be
2317 * mounted beneath mounts on the same mountpoint.
2318 */
2319 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2320 if (IS_ERR(smp))
2321 return PTR_ERR(smp);
2322
2323 /* Is there space to add these mounts to the mount namespace? */
2324 if (!moving) {
2325 err = count_mounts(ns, source_mnt);
2326 if (err)
2327 goto out;
2328 }
2329
2330 if (beneath)
2331 dest_mnt = top_mnt->mnt_parent;
2332 else
2333 dest_mnt = top_mnt;
2334
2335 if (IS_MNT_SHARED(dest_mnt)) {
2336 err = invent_group_ids(source_mnt, true);
2337 if (err)
2338 goto out;
2339 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2340 }
2341 lock_mount_hash();
2342 if (err)
2343 goto out_cleanup_ids;
2344
2345 if (IS_MNT_SHARED(dest_mnt)) {
2346 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2347 set_mnt_shared(p);
2348 }
2349
2350 if (moving) {
2351 if (beneath)
2352 dest_mp = smp;
2353 unhash_mnt(source_mnt);
2354 attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
2355 touch_mnt_namespace(source_mnt->mnt_ns);
2356 } else {
2357 if (source_mnt->mnt_ns) {
2358 LIST_HEAD(head);
2359
2360 /* move from anon - the caller will destroy */
2361 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2362 move_from_ns(p, &head);
2363 list_del_init(&head);
2364 }
2365 if (beneath)
2366 mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
2367 else
2368 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2369 commit_tree(source_mnt);
2370 }
2371
2372 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2373 struct mount *q;
2374 hlist_del_init(&child->mnt_hash);
2375 q = __lookup_mnt(&child->mnt_parent->mnt,
2376 child->mnt_mountpoint);
2377 if (q)
2378 mnt_change_mountpoint(child, smp, q);
2379 /* Notice when we are propagating across user namespaces */
2380 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2381 lock_mnt_tree(child);
2382 child->mnt.mnt_flags &= ~MNT_LOCKED;
2383 commit_tree(child);
2384 }
2385 put_mountpoint(smp);
2386 unlock_mount_hash();
2387
2388 return 0;
2389
2390 out_cleanup_ids:
2391 while (!hlist_empty(&tree_list)) {
2392 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2393 child->mnt_parent->mnt_ns->pending_mounts = 0;
2394 umount_tree(child, UMOUNT_SYNC);
2395 }
2396 unlock_mount_hash();
2397 cleanup_group_ids(source_mnt, NULL);
2398 out:
2399 ns->pending_mounts = 0;
2400
2401 read_seqlock_excl(&mount_lock);
2402 put_mountpoint(smp);
2403 read_sequnlock_excl(&mount_lock);
2404
2405 return err;
2406}
2407
2408/**
2409 * do_lock_mount - lock mount and mountpoint
2410 * @path: target path
2411 * @beneath: whether the intention is to mount beneath @path
2412 *
2413 * Follow the mount stack on @path until the top mount @mnt is found. If
2414 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2415 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2416 * until nothing is stacked on top of it anymore.
2417 *
2418 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2419 * against concurrent removal of the new mountpoint from another mount
2420 * namespace.
2421 *
2422 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2423 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2424 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2425 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2426 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2427 * on top of it for @beneath.
2428 *
2429 * In addition, @beneath needs to make sure that @mnt hasn't been
2430 * unmounted or moved from its current mountpoint in between dropping
2431 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2432 * being unmounted would be detected later by e.g., calling
2433 * check_mnt(mnt) in the function it's called from. For the @beneath
2434 * case however, it's useful to detect it directly in do_lock_mount().
2435 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2436 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2437 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2438 *
2439 * Return: Either the target mountpoint on the top mount or the top
2440 * mount's mountpoint.
2441 */
2442static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
2443{
2444 struct vfsmount *mnt = path->mnt;
2445 struct dentry *dentry;
2446 struct mountpoint *mp = ERR_PTR(-ENOENT);
2447
2448 for (;;) {
2449 struct mount *m;
2450
2451 if (beneath) {
2452 m = real_mount(mnt);
2453 read_seqlock_excl(&mount_lock);
2454 dentry = dget(m->mnt_mountpoint);
2455 read_sequnlock_excl(&mount_lock);
2456 } else {
2457 dentry = path->dentry;
2458 }
2459
2460 inode_lock(dentry->d_inode);
2461 if (unlikely(cant_mount(dentry))) {
2462 inode_unlock(dentry->d_inode);
2463 goto out;
2464 }
2465
2466 namespace_lock();
2467
2468 if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) {
2469 namespace_unlock();
2470 inode_unlock(dentry->d_inode);
2471 goto out;
2472 }
2473
2474 mnt = lookup_mnt(path);
2475 if (likely(!mnt))
2476 break;
2477
2478 namespace_unlock();
2479 inode_unlock(dentry->d_inode);
2480 if (beneath)
2481 dput(dentry);
2482 path_put(path);
2483 path->mnt = mnt;
2484 path->dentry = dget(mnt->mnt_root);
2485 }
2486
2487 mp = get_mountpoint(dentry);
2488 if (IS_ERR(mp)) {
2489 namespace_unlock();
2490 inode_unlock(dentry->d_inode);
2491 }
2492
2493out:
2494 if (beneath)
2495 dput(dentry);
2496
2497 return mp;
2498}
2499
2500static inline struct mountpoint *lock_mount(struct path *path)
2501{
2502 return do_lock_mount(path, false);
2503}
2504
2505static void unlock_mount(struct mountpoint *where)
2506{
2507 struct dentry *dentry = where->m_dentry;
2508
2509 read_seqlock_excl(&mount_lock);
2510 put_mountpoint(where);
2511 read_sequnlock_excl(&mount_lock);
2512
2513 namespace_unlock();
2514 inode_unlock(dentry->d_inode);
2515}
2516
2517static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2518{
2519 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2520 return -EINVAL;
2521
2522 if (d_is_dir(mp->m_dentry) !=
2523 d_is_dir(mnt->mnt.mnt_root))
2524 return -ENOTDIR;
2525
2526 return attach_recursive_mnt(mnt, p, mp, 0);
2527}
2528
2529/*
2530 * Sanity check the flags to change_mnt_propagation.
2531 */
2532
2533static int flags_to_propagation_type(int ms_flags)
2534{
2535 int type = ms_flags & ~(MS_REC | MS_SILENT);
2536
2537 /* Fail if any non-propagation flags are set */
2538 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2539 return 0;
2540 /* Only one propagation flag should be set */
2541 if (!is_power_of_2(type))
2542 return 0;
2543 return type;
2544}
2545
2546/*
2547 * recursively change the type of the mountpoint.
2548 */
2549static int do_change_type(struct path *path, int ms_flags)
2550{
2551 struct mount *m;
2552 struct mount *mnt = real_mount(path->mnt);
2553 int recurse = ms_flags & MS_REC;
2554 int type;
2555 int err = 0;
2556
2557 if (!path_mounted(path))
2558 return -EINVAL;
2559
2560 type = flags_to_propagation_type(ms_flags);
2561 if (!type)
2562 return -EINVAL;
2563
2564 namespace_lock();
2565 if (type == MS_SHARED) {
2566 err = invent_group_ids(mnt, recurse);
2567 if (err)
2568 goto out_unlock;
2569 }
2570
2571 lock_mount_hash();
2572 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2573 change_mnt_propagation(m, type);
2574 unlock_mount_hash();
2575
2576 out_unlock:
2577 namespace_unlock();
2578 return err;
2579}
2580
2581static struct mount *__do_loopback(struct path *old_path, int recurse)
2582{
2583 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2584
2585 if (IS_MNT_UNBINDABLE(old))
2586 return mnt;
2587
2588 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2589 return mnt;
2590
2591 if (!recurse && has_locked_children(old, old_path->dentry))
2592 return mnt;
2593
2594 if (recurse)
2595 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2596 else
2597 mnt = clone_mnt(old, old_path->dentry, 0);
2598
2599 if (!IS_ERR(mnt))
2600 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2601
2602 return mnt;
2603}
2604
2605/*
2606 * do loopback mount.
2607 */
2608static int do_loopback(struct path *path, const char *old_name,
2609 int recurse)
2610{
2611 struct path old_path;
2612 struct mount *mnt = NULL, *parent;
2613 struct mountpoint *mp;
2614 int err;
2615 if (!old_name || !*old_name)
2616 return -EINVAL;
2617 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2618 if (err)
2619 return err;
2620
2621 err = -EINVAL;
2622 if (mnt_ns_loop(old_path.dentry))
2623 goto out;
2624
2625 mp = lock_mount(path);
2626 if (IS_ERR(mp)) {
2627 err = PTR_ERR(mp);
2628 goto out;
2629 }
2630
2631 parent = real_mount(path->mnt);
2632 if (!check_mnt(parent))
2633 goto out2;
2634
2635 mnt = __do_loopback(&old_path, recurse);
2636 if (IS_ERR(mnt)) {
2637 err = PTR_ERR(mnt);
2638 goto out2;
2639 }
2640
2641 err = graft_tree(mnt, parent, mp);
2642 if (err) {
2643 lock_mount_hash();
2644 umount_tree(mnt, UMOUNT_SYNC);
2645 unlock_mount_hash();
2646 }
2647out2:
2648 unlock_mount(mp);
2649out:
2650 path_put(&old_path);
2651 return err;
2652}
2653
2654static struct file *open_detached_copy(struct path *path, bool recursive)
2655{
2656 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2657 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2658 struct mount *mnt, *p;
2659 struct file *file;
2660
2661 if (IS_ERR(ns))
2662 return ERR_CAST(ns);
2663
2664 namespace_lock();
2665 mnt = __do_loopback(path, recursive);
2666 if (IS_ERR(mnt)) {
2667 namespace_unlock();
2668 free_mnt_ns(ns);
2669 return ERR_CAST(mnt);
2670 }
2671
2672 lock_mount_hash();
2673 for (p = mnt; p; p = next_mnt(p, mnt)) {
2674 mnt_add_to_ns(ns, p);
2675 ns->nr_mounts++;
2676 }
2677 ns->root = mnt;
2678 mntget(&mnt->mnt);
2679 unlock_mount_hash();
2680 namespace_unlock();
2681
2682 mntput(path->mnt);
2683 path->mnt = &mnt->mnt;
2684 file = dentry_open(path, O_PATH, current_cred());
2685 if (IS_ERR(file))
2686 dissolve_on_fput(path->mnt);
2687 else
2688 file->f_mode |= FMODE_NEED_UNMOUNT;
2689 return file;
2690}
2691
2692SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2693{
2694 struct file *file;
2695 struct path path;
2696 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2697 bool detached = flags & OPEN_TREE_CLONE;
2698 int error;
2699 int fd;
2700
2701 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2702
2703 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2704 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2705 OPEN_TREE_CLOEXEC))
2706 return -EINVAL;
2707
2708 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2709 return -EINVAL;
2710
2711 if (flags & AT_NO_AUTOMOUNT)
2712 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2713 if (flags & AT_SYMLINK_NOFOLLOW)
2714 lookup_flags &= ~LOOKUP_FOLLOW;
2715 if (flags & AT_EMPTY_PATH)
2716 lookup_flags |= LOOKUP_EMPTY;
2717
2718 if (detached && !may_mount())
2719 return -EPERM;
2720
2721 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2722 if (fd < 0)
2723 return fd;
2724
2725 error = user_path_at(dfd, filename, lookup_flags, &path);
2726 if (unlikely(error)) {
2727 file = ERR_PTR(error);
2728 } else {
2729 if (detached)
2730 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2731 else
2732 file = dentry_open(&path, O_PATH, current_cred());
2733 path_put(&path);
2734 }
2735 if (IS_ERR(file)) {
2736 put_unused_fd(fd);
2737 return PTR_ERR(file);
2738 }
2739 fd_install(fd, file);
2740 return fd;
2741}
2742
2743/*
2744 * Don't allow locked mount flags to be cleared.
2745 *
2746 * No locks need to be held here while testing the various MNT_LOCK
2747 * flags because those flags can never be cleared once they are set.
2748 */
2749static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2750{
2751 unsigned int fl = mnt->mnt.mnt_flags;
2752
2753 if ((fl & MNT_LOCK_READONLY) &&
2754 !(mnt_flags & MNT_READONLY))
2755 return false;
2756
2757 if ((fl & MNT_LOCK_NODEV) &&
2758 !(mnt_flags & MNT_NODEV))
2759 return false;
2760
2761 if ((fl & MNT_LOCK_NOSUID) &&
2762 !(mnt_flags & MNT_NOSUID))
2763 return false;
2764
2765 if ((fl & MNT_LOCK_NOEXEC) &&
2766 !(mnt_flags & MNT_NOEXEC))
2767 return false;
2768
2769 if ((fl & MNT_LOCK_ATIME) &&
2770 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2771 return false;
2772
2773 return true;
2774}
2775
2776static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2777{
2778 bool readonly_request = (mnt_flags & MNT_READONLY);
2779
2780 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2781 return 0;
2782
2783 if (readonly_request)
2784 return mnt_make_readonly(mnt);
2785
2786 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2787 return 0;
2788}
2789
2790static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2791{
2792 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2793 mnt->mnt.mnt_flags = mnt_flags;
2794 touch_mnt_namespace(mnt->mnt_ns);
2795}
2796
2797static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2798{
2799 struct super_block *sb = mnt->mnt_sb;
2800
2801 if (!__mnt_is_readonly(mnt) &&
2802 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2803 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2804 char *buf = (char *)__get_free_page(GFP_KERNEL);
2805 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2806
2807 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2808 sb->s_type->name,
2809 is_mounted(mnt) ? "remounted" : "mounted",
2810 mntpath, &sb->s_time_max,
2811 (unsigned long long)sb->s_time_max);
2812
2813 free_page((unsigned long)buf);
2814 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2815 }
2816}
2817
2818/*
2819 * Handle reconfiguration of the mountpoint only without alteration of the
2820 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2821 * to mount(2).
2822 */
2823static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2824{
2825 struct super_block *sb = path->mnt->mnt_sb;
2826 struct mount *mnt = real_mount(path->mnt);
2827 int ret;
2828
2829 if (!check_mnt(mnt))
2830 return -EINVAL;
2831
2832 if (!path_mounted(path))
2833 return -EINVAL;
2834
2835 if (!can_change_locked_flags(mnt, mnt_flags))
2836 return -EPERM;
2837
2838 /*
2839 * We're only checking whether the superblock is read-only not
2840 * changing it, so only take down_read(&sb->s_umount).
2841 */
2842 down_read(&sb->s_umount);
2843 lock_mount_hash();
2844 ret = change_mount_ro_state(mnt, mnt_flags);
2845 if (ret == 0)
2846 set_mount_attributes(mnt, mnt_flags);
2847 unlock_mount_hash();
2848 up_read(&sb->s_umount);
2849
2850 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2851
2852 return ret;
2853}
2854
2855/*
2856 * change filesystem flags. dir should be a physical root of filesystem.
2857 * If you've mounted a non-root directory somewhere and want to do remount
2858 * on it - tough luck.
2859 */
2860static int do_remount(struct path *path, int ms_flags, int sb_flags,
2861 int mnt_flags, void *data)
2862{
2863 int err;
2864 struct super_block *sb = path->mnt->mnt_sb;
2865 struct mount *mnt = real_mount(path->mnt);
2866 struct fs_context *fc;
2867
2868 if (!check_mnt(mnt))
2869 return -EINVAL;
2870
2871 if (!path_mounted(path))
2872 return -EINVAL;
2873
2874 if (!can_change_locked_flags(mnt, mnt_flags))
2875 return -EPERM;
2876
2877 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2878 if (IS_ERR(fc))
2879 return PTR_ERR(fc);
2880
2881 /*
2882 * Indicate to the filesystem that the remount request is coming
2883 * from the legacy mount system call.
2884 */
2885 fc->oldapi = true;
2886
2887 err = parse_monolithic_mount_data(fc, data);
2888 if (!err) {
2889 down_write(&sb->s_umount);
2890 err = -EPERM;
2891 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2892 err = reconfigure_super(fc);
2893 if (!err) {
2894 lock_mount_hash();
2895 set_mount_attributes(mnt, mnt_flags);
2896 unlock_mount_hash();
2897 }
2898 }
2899 up_write(&sb->s_umount);
2900 }
2901
2902 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2903
2904 put_fs_context(fc);
2905 return err;
2906}
2907
2908static inline int tree_contains_unbindable(struct mount *mnt)
2909{
2910 struct mount *p;
2911 for (p = mnt; p; p = next_mnt(p, mnt)) {
2912 if (IS_MNT_UNBINDABLE(p))
2913 return 1;
2914 }
2915 return 0;
2916}
2917
2918/*
2919 * Check that there aren't references to earlier/same mount namespaces in the
2920 * specified subtree. Such references can act as pins for mount namespaces
2921 * that aren't checked by the mount-cycle checking code, thereby allowing
2922 * cycles to be made.
2923 */
2924static bool check_for_nsfs_mounts(struct mount *subtree)
2925{
2926 struct mount *p;
2927 bool ret = false;
2928
2929 lock_mount_hash();
2930 for (p = subtree; p; p = next_mnt(p, subtree))
2931 if (mnt_ns_loop(p->mnt.mnt_root))
2932 goto out;
2933
2934 ret = true;
2935out:
2936 unlock_mount_hash();
2937 return ret;
2938}
2939
2940static int do_set_group(struct path *from_path, struct path *to_path)
2941{
2942 struct mount *from, *to;
2943 int err;
2944
2945 from = real_mount(from_path->mnt);
2946 to = real_mount(to_path->mnt);
2947
2948 namespace_lock();
2949
2950 err = -EINVAL;
2951 /* To and From must be mounted */
2952 if (!is_mounted(&from->mnt))
2953 goto out;
2954 if (!is_mounted(&to->mnt))
2955 goto out;
2956
2957 err = -EPERM;
2958 /* We should be allowed to modify mount namespaces of both mounts */
2959 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2960 goto out;
2961 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2962 goto out;
2963
2964 err = -EINVAL;
2965 /* To and From paths should be mount roots */
2966 if (!path_mounted(from_path))
2967 goto out;
2968 if (!path_mounted(to_path))
2969 goto out;
2970
2971 /* Setting sharing groups is only allowed across same superblock */
2972 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2973 goto out;
2974
2975 /* From mount root should be wider than To mount root */
2976 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2977 goto out;
2978
2979 /* From mount should not have locked children in place of To's root */
2980 if (has_locked_children(from, to->mnt.mnt_root))
2981 goto out;
2982
2983 /* Setting sharing groups is only allowed on private mounts */
2984 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2985 goto out;
2986
2987 /* From should not be private */
2988 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2989 goto out;
2990
2991 if (IS_MNT_SLAVE(from)) {
2992 struct mount *m = from->mnt_master;
2993
2994 list_add(&to->mnt_slave, &m->mnt_slave_list);
2995 to->mnt_master = m;
2996 }
2997
2998 if (IS_MNT_SHARED(from)) {
2999 to->mnt_group_id = from->mnt_group_id;
3000 list_add(&to->mnt_share, &from->mnt_share);
3001 lock_mount_hash();
3002 set_mnt_shared(to);
3003 unlock_mount_hash();
3004 }
3005
3006 err = 0;
3007out:
3008 namespace_unlock();
3009 return err;
3010}
3011
3012/**
3013 * path_overmounted - check if path is overmounted
3014 * @path: path to check
3015 *
3016 * Check if path is overmounted, i.e., if there's a mount on top of
3017 * @path->mnt with @path->dentry as mountpoint.
3018 *
3019 * Context: This function expects namespace_lock() to be held.
3020 * Return: If path is overmounted true is returned, false if not.
3021 */
3022static inline bool path_overmounted(const struct path *path)
3023{
3024 rcu_read_lock();
3025 if (unlikely(__lookup_mnt(path->mnt, path->dentry))) {
3026 rcu_read_unlock();
3027 return true;
3028 }
3029 rcu_read_unlock();
3030 return false;
3031}
3032
3033/**
3034 * can_move_mount_beneath - check that we can mount beneath the top mount
3035 * @from: mount to mount beneath
3036 * @to: mount under which to mount
3037 * @mp: mountpoint of @to
3038 *
3039 * - Make sure that @to->dentry is actually the root of a mount under
3040 * which we can mount another mount.
3041 * - Make sure that nothing can be mounted beneath the caller's current
3042 * root or the rootfs of the namespace.
3043 * - Make sure that the caller can unmount the topmost mount ensuring
3044 * that the caller could reveal the underlying mountpoint.
3045 * - Ensure that nothing has been mounted on top of @from before we
3046 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3047 * - Prevent mounting beneath a mount if the propagation relationship
3048 * between the source mount, parent mount, and top mount would lead to
3049 * nonsensical mount trees.
3050 *
3051 * Context: This function expects namespace_lock() to be held.
3052 * Return: On success 0, and on error a negative error code is returned.
3053 */
3054static int can_move_mount_beneath(const struct path *from,
3055 const struct path *to,
3056 const struct mountpoint *mp)
3057{
3058 struct mount *mnt_from = real_mount(from->mnt),
3059 *mnt_to = real_mount(to->mnt),
3060 *parent_mnt_to = mnt_to->mnt_parent;
3061
3062 if (!mnt_has_parent(mnt_to))
3063 return -EINVAL;
3064
3065 if (!path_mounted(to))
3066 return -EINVAL;
3067
3068 if (IS_MNT_LOCKED(mnt_to))
3069 return -EINVAL;
3070
3071 /* Avoid creating shadow mounts during mount propagation. */
3072 if (path_overmounted(from))
3073 return -EINVAL;
3074
3075 /*
3076 * Mounting beneath the rootfs only makes sense when the
3077 * semantics of pivot_root(".", ".") are used.
3078 */
3079 if (&mnt_to->mnt == current->fs->root.mnt)
3080 return -EINVAL;
3081 if (parent_mnt_to == current->nsproxy->mnt_ns->root)
3082 return -EINVAL;
3083
3084 for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
3085 if (p == mnt_to)
3086 return -EINVAL;
3087
3088 /*
3089 * If the parent mount propagates to the child mount this would
3090 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3091 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3092 * defeats the whole purpose of mounting beneath another mount.
3093 */
3094 if (propagation_would_overmount(parent_mnt_to, mnt_to, mp))
3095 return -EINVAL;
3096
3097 /*
3098 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3099 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3100 * Afterwards @mnt_from would be mounted on top of
3101 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3102 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3103 * already mounted on @mnt_from, @mnt_to would ultimately be
3104 * remounted on top of @c. Afterwards, @mnt_from would be
3105 * covered by a copy @c of @mnt_from and @c would be covered by
3106 * @mnt_from itself. This defeats the whole purpose of mounting
3107 * @mnt_from beneath @mnt_to.
3108 */
3109 if (propagation_would_overmount(parent_mnt_to, mnt_from, mp))
3110 return -EINVAL;
3111
3112 return 0;
3113}
3114
3115static int do_move_mount(struct path *old_path, struct path *new_path,
3116 bool beneath)
3117{
3118 struct mnt_namespace *ns;
3119 struct mount *p;
3120 struct mount *old;
3121 struct mount *parent;
3122 struct mountpoint *mp, *old_mp;
3123 int err;
3124 bool attached;
3125 enum mnt_tree_flags_t flags = 0;
3126
3127 mp = do_lock_mount(new_path, beneath);
3128 if (IS_ERR(mp))
3129 return PTR_ERR(mp);
3130
3131 old = real_mount(old_path->mnt);
3132 p = real_mount(new_path->mnt);
3133 parent = old->mnt_parent;
3134 attached = mnt_has_parent(old);
3135 if (attached)
3136 flags |= MNT_TREE_MOVE;
3137 old_mp = old->mnt_mp;
3138 ns = old->mnt_ns;
3139
3140 err = -EINVAL;
3141 /* The mountpoint must be in our namespace. */
3142 if (!check_mnt(p))
3143 goto out;
3144
3145 /* The thing moved must be mounted... */
3146 if (!is_mounted(&old->mnt))
3147 goto out;
3148
3149 /* ... and either ours or the root of anon namespace */
3150 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
3151 goto out;
3152
3153 if (old->mnt.mnt_flags & MNT_LOCKED)
3154 goto out;
3155
3156 if (!path_mounted(old_path))
3157 goto out;
3158
3159 if (d_is_dir(new_path->dentry) !=
3160 d_is_dir(old_path->dentry))
3161 goto out;
3162 /*
3163 * Don't move a mount residing in a shared parent.
3164 */
3165 if (attached && IS_MNT_SHARED(parent))
3166 goto out;
3167
3168 if (beneath) {
3169 err = can_move_mount_beneath(old_path, new_path, mp);
3170 if (err)
3171 goto out;
3172
3173 err = -EINVAL;
3174 p = p->mnt_parent;
3175 flags |= MNT_TREE_BENEATH;
3176 }
3177
3178 /*
3179 * Don't move a mount tree containing unbindable mounts to a destination
3180 * mount which is shared.
3181 */
3182 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
3183 goto out;
3184 err = -ELOOP;
3185 if (!check_for_nsfs_mounts(old))
3186 goto out;
3187 for (; mnt_has_parent(p); p = p->mnt_parent)
3188 if (p == old)
3189 goto out;
3190
3191 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
3192 if (err)
3193 goto out;
3194
3195 /* if the mount is moved, it should no longer be expire
3196 * automatically */
3197 list_del_init(&old->mnt_expire);
3198 if (attached)
3199 put_mountpoint(old_mp);
3200out:
3201 unlock_mount(mp);
3202 if (!err) {
3203 if (attached)
3204 mntput_no_expire(parent);
3205 else
3206 free_mnt_ns(ns);
3207 }
3208 return err;
3209}
3210
3211static int do_move_mount_old(struct path *path, const char *old_name)
3212{
3213 struct path old_path;
3214 int err;
3215
3216 if (!old_name || !*old_name)
3217 return -EINVAL;
3218
3219 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3220 if (err)
3221 return err;
3222
3223 err = do_move_mount(&old_path, path, false);
3224 path_put(&old_path);
3225 return err;
3226}
3227
3228/*
3229 * add a mount into a namespace's mount tree
3230 */
3231static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
3232 const struct path *path, int mnt_flags)
3233{
3234 struct mount *parent = real_mount(path->mnt);
3235
3236 mnt_flags &= ~MNT_INTERNAL_FLAGS;
3237
3238 if (unlikely(!check_mnt(parent))) {
3239 /* that's acceptable only for automounts done in private ns */
3240 if (!(mnt_flags & MNT_SHRINKABLE))
3241 return -EINVAL;
3242 /* ... and for those we'd better have mountpoint still alive */
3243 if (!parent->mnt_ns)
3244 return -EINVAL;
3245 }
3246
3247 /* Refuse the same filesystem on the same mount point */
3248 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
3249 return -EBUSY;
3250
3251 if (d_is_symlink(newmnt->mnt.mnt_root))
3252 return -EINVAL;
3253
3254 newmnt->mnt.mnt_flags = mnt_flags;
3255 return graft_tree(newmnt, parent, mp);
3256}
3257
3258static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3259
3260/*
3261 * Create a new mount using a superblock configuration and request it
3262 * be added to the namespace tree.
3263 */
3264static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3265 unsigned int mnt_flags)
3266{
3267 struct vfsmount *mnt;
3268 struct mountpoint *mp;
3269 struct super_block *sb = fc->root->d_sb;
3270 int error;
3271
3272 error = security_sb_kern_mount(sb);
3273 if (!error && mount_too_revealing(sb, &mnt_flags))
3274 error = -EPERM;
3275
3276 if (unlikely(error)) {
3277 fc_drop_locked(fc);
3278 return error;
3279 }
3280
3281 up_write(&sb->s_umount);
3282
3283 mnt = vfs_create_mount(fc);
3284 if (IS_ERR(mnt))
3285 return PTR_ERR(mnt);
3286
3287 mnt_warn_timestamp_expiry(mountpoint, mnt);
3288
3289 mp = lock_mount(mountpoint);
3290 if (IS_ERR(mp)) {
3291 mntput(mnt);
3292 return PTR_ERR(mp);
3293 }
3294 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3295 unlock_mount(mp);
3296 if (error < 0)
3297 mntput(mnt);
3298 return error;
3299}
3300
3301/*
3302 * create a new mount for userspace and request it to be added into the
3303 * namespace's tree
3304 */
3305static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3306 int mnt_flags, const char *name, void *data)
3307{
3308 struct file_system_type *type;
3309 struct fs_context *fc;
3310 const char *subtype = NULL;
3311 int err = 0;
3312
3313 if (!fstype)
3314 return -EINVAL;
3315
3316 type = get_fs_type(fstype);
3317 if (!type)
3318 return -ENODEV;
3319
3320 if (type->fs_flags & FS_HAS_SUBTYPE) {
3321 subtype = strchr(fstype, '.');
3322 if (subtype) {
3323 subtype++;
3324 if (!*subtype) {
3325 put_filesystem(type);
3326 return -EINVAL;
3327 }
3328 }
3329 }
3330
3331 fc = fs_context_for_mount(type, sb_flags);
3332 put_filesystem(type);
3333 if (IS_ERR(fc))
3334 return PTR_ERR(fc);
3335
3336 /*
3337 * Indicate to the filesystem that the mount request is coming
3338 * from the legacy mount system call.
3339 */
3340 fc->oldapi = true;
3341
3342 if (subtype)
3343 err = vfs_parse_fs_string(fc, "subtype",
3344 subtype, strlen(subtype));
3345 if (!err && name)
3346 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3347 if (!err)
3348 err = parse_monolithic_mount_data(fc, data);
3349 if (!err && !mount_capable(fc))
3350 err = -EPERM;
3351 if (!err)
3352 err = vfs_get_tree(fc);
3353 if (!err)
3354 err = do_new_mount_fc(fc, path, mnt_flags);
3355
3356 put_fs_context(fc);
3357 return err;
3358}
3359
3360int finish_automount(struct vfsmount *m, const struct path *path)
3361{
3362 struct dentry *dentry = path->dentry;
3363 struct mountpoint *mp;
3364 struct mount *mnt;
3365 int err;
3366
3367 if (!m)
3368 return 0;
3369 if (IS_ERR(m))
3370 return PTR_ERR(m);
3371
3372 mnt = real_mount(m);
3373 /* The new mount record should have at least 2 refs to prevent it being
3374 * expired before we get a chance to add it
3375 */
3376 BUG_ON(mnt_get_count(mnt) < 2);
3377
3378 if (m->mnt_sb == path->mnt->mnt_sb &&
3379 m->mnt_root == dentry) {
3380 err = -ELOOP;
3381 goto discard;
3382 }
3383
3384 /*
3385 * we don't want to use lock_mount() - in this case finding something
3386 * that overmounts our mountpoint to be means "quitely drop what we've
3387 * got", not "try to mount it on top".
3388 */
3389 inode_lock(dentry->d_inode);
3390 namespace_lock();
3391 if (unlikely(cant_mount(dentry))) {
3392 err = -ENOENT;
3393 goto discard_locked;
3394 }
3395 if (path_overmounted(path)) {
3396 err = 0;
3397 goto discard_locked;
3398 }
3399 mp = get_mountpoint(dentry);
3400 if (IS_ERR(mp)) {
3401 err = PTR_ERR(mp);
3402 goto discard_locked;
3403 }
3404
3405 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3406 unlock_mount(mp);
3407 if (unlikely(err))
3408 goto discard;
3409 mntput(m);
3410 return 0;
3411
3412discard_locked:
3413 namespace_unlock();
3414 inode_unlock(dentry->d_inode);
3415discard:
3416 /* remove m from any expiration list it may be on */
3417 if (!list_empty(&mnt->mnt_expire)) {
3418 namespace_lock();
3419 list_del_init(&mnt->mnt_expire);
3420 namespace_unlock();
3421 }
3422 mntput(m);
3423 mntput(m);
3424 return err;
3425}
3426
3427/**
3428 * mnt_set_expiry - Put a mount on an expiration list
3429 * @mnt: The mount to list.
3430 * @expiry_list: The list to add the mount to.
3431 */
3432void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3433{
3434 namespace_lock();
3435
3436 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3437
3438 namespace_unlock();
3439}
3440EXPORT_SYMBOL(mnt_set_expiry);
3441
3442/*
3443 * process a list of expirable mountpoints with the intent of discarding any
3444 * mountpoints that aren't in use and haven't been touched since last we came
3445 * here
3446 */
3447void mark_mounts_for_expiry(struct list_head *mounts)
3448{
3449 struct mount *mnt, *next;
3450 LIST_HEAD(graveyard);
3451
3452 if (list_empty(mounts))
3453 return;
3454
3455 namespace_lock();
3456 lock_mount_hash();
3457
3458 /* extract from the expiration list every vfsmount that matches the
3459 * following criteria:
3460 * - only referenced by its parent vfsmount
3461 * - still marked for expiry (marked on the last call here; marks are
3462 * cleared by mntput())
3463 */
3464 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3465 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3466 propagate_mount_busy(mnt, 1))
3467 continue;
3468 list_move(&mnt->mnt_expire, &graveyard);
3469 }
3470 while (!list_empty(&graveyard)) {
3471 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3472 touch_mnt_namespace(mnt->mnt_ns);
3473 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3474 }
3475 unlock_mount_hash();
3476 namespace_unlock();
3477}
3478
3479EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3480
3481/*
3482 * Ripoff of 'select_parent()'
3483 *
3484 * search the list of submounts for a given mountpoint, and move any
3485 * shrinkable submounts to the 'graveyard' list.
3486 */
3487static int select_submounts(struct mount *parent, struct list_head *graveyard)
3488{
3489 struct mount *this_parent = parent;
3490 struct list_head *next;
3491 int found = 0;
3492
3493repeat:
3494 next = this_parent->mnt_mounts.next;
3495resume:
3496 while (next != &this_parent->mnt_mounts) {
3497 struct list_head *tmp = next;
3498 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3499
3500 next = tmp->next;
3501 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3502 continue;
3503 /*
3504 * Descend a level if the d_mounts list is non-empty.
3505 */
3506 if (!list_empty(&mnt->mnt_mounts)) {
3507 this_parent = mnt;
3508 goto repeat;
3509 }
3510
3511 if (!propagate_mount_busy(mnt, 1)) {
3512 list_move_tail(&mnt->mnt_expire, graveyard);
3513 found++;
3514 }
3515 }
3516 /*
3517 * All done at this level ... ascend and resume the search
3518 */
3519 if (this_parent != parent) {
3520 next = this_parent->mnt_child.next;
3521 this_parent = this_parent->mnt_parent;
3522 goto resume;
3523 }
3524 return found;
3525}
3526
3527/*
3528 * process a list of expirable mountpoints with the intent of discarding any
3529 * submounts of a specific parent mountpoint
3530 *
3531 * mount_lock must be held for write
3532 */
3533static void shrink_submounts(struct mount *mnt)
3534{
3535 LIST_HEAD(graveyard);
3536 struct mount *m;
3537
3538 /* extract submounts of 'mountpoint' from the expiration list */
3539 while (select_submounts(mnt, &graveyard)) {
3540 while (!list_empty(&graveyard)) {
3541 m = list_first_entry(&graveyard, struct mount,
3542 mnt_expire);
3543 touch_mnt_namespace(m->mnt_ns);
3544 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3545 }
3546 }
3547}
3548
3549static void *copy_mount_options(const void __user * data)
3550{
3551 char *copy;
3552 unsigned left, offset;
3553
3554 if (!data)
3555 return NULL;
3556
3557 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3558 if (!copy)
3559 return ERR_PTR(-ENOMEM);
3560
3561 left = copy_from_user(copy, data, PAGE_SIZE);
3562
3563 /*
3564 * Not all architectures have an exact copy_from_user(). Resort to
3565 * byte at a time.
3566 */
3567 offset = PAGE_SIZE - left;
3568 while (left) {
3569 char c;
3570 if (get_user(c, (const char __user *)data + offset))
3571 break;
3572 copy[offset] = c;
3573 left--;
3574 offset++;
3575 }
3576
3577 if (left == PAGE_SIZE) {
3578 kfree(copy);
3579 return ERR_PTR(-EFAULT);
3580 }
3581
3582 return copy;
3583}
3584
3585static char *copy_mount_string(const void __user *data)
3586{
3587 return data ? strndup_user(data, PATH_MAX) : NULL;
3588}
3589
3590/*
3591 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3592 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3593 *
3594 * data is a (void *) that can point to any structure up to
3595 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3596 * information (or be NULL).
3597 *
3598 * Pre-0.97 versions of mount() didn't have a flags word.
3599 * When the flags word was introduced its top half was required
3600 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3601 * Therefore, if this magic number is present, it carries no information
3602 * and must be discarded.
3603 */
3604int path_mount(const char *dev_name, struct path *path,
3605 const char *type_page, unsigned long flags, void *data_page)
3606{
3607 unsigned int mnt_flags = 0, sb_flags;
3608 int ret;
3609
3610 /* Discard magic */
3611 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3612 flags &= ~MS_MGC_MSK;
3613
3614 /* Basic sanity checks */
3615 if (data_page)
3616 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3617
3618 if (flags & MS_NOUSER)
3619 return -EINVAL;
3620
3621 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3622 if (ret)
3623 return ret;
3624 if (!may_mount())
3625 return -EPERM;
3626 if (flags & SB_MANDLOCK)
3627 warn_mandlock();
3628
3629 /* Default to relatime unless overriden */
3630 if (!(flags & MS_NOATIME))
3631 mnt_flags |= MNT_RELATIME;
3632
3633 /* Separate the per-mountpoint flags */
3634 if (flags & MS_NOSUID)
3635 mnt_flags |= MNT_NOSUID;
3636 if (flags & MS_NODEV)
3637 mnt_flags |= MNT_NODEV;
3638 if (flags & MS_NOEXEC)
3639 mnt_flags |= MNT_NOEXEC;
3640 if (flags & MS_NOATIME)
3641 mnt_flags |= MNT_NOATIME;
3642 if (flags & MS_NODIRATIME)
3643 mnt_flags |= MNT_NODIRATIME;
3644 if (flags & MS_STRICTATIME)
3645 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3646 if (flags & MS_RDONLY)
3647 mnt_flags |= MNT_READONLY;
3648 if (flags & MS_NOSYMFOLLOW)
3649 mnt_flags |= MNT_NOSYMFOLLOW;
3650
3651 /* The default atime for remount is preservation */
3652 if ((flags & MS_REMOUNT) &&
3653 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3654 MS_STRICTATIME)) == 0)) {
3655 mnt_flags &= ~MNT_ATIME_MASK;
3656 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3657 }
3658
3659 sb_flags = flags & (SB_RDONLY |
3660 SB_SYNCHRONOUS |
3661 SB_MANDLOCK |
3662 SB_DIRSYNC |
3663 SB_SILENT |
3664 SB_POSIXACL |
3665 SB_LAZYTIME |
3666 SB_I_VERSION);
3667
3668 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3669 return do_reconfigure_mnt(path, mnt_flags);
3670 if (flags & MS_REMOUNT)
3671 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3672 if (flags & MS_BIND)
3673 return do_loopback(path, dev_name, flags & MS_REC);
3674 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3675 return do_change_type(path, flags);
3676 if (flags & MS_MOVE)
3677 return do_move_mount_old(path, dev_name);
3678
3679 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3680 data_page);
3681}
3682
3683long do_mount(const char *dev_name, const char __user *dir_name,
3684 const char *type_page, unsigned long flags, void *data_page)
3685{
3686 struct path path;
3687 int ret;
3688
3689 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3690 if (ret)
3691 return ret;
3692 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3693 path_put(&path);
3694 return ret;
3695}
3696
3697static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3698{
3699 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3700}
3701
3702static void dec_mnt_namespaces(struct ucounts *ucounts)
3703{
3704 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3705}
3706
3707static void free_mnt_ns(struct mnt_namespace *ns)
3708{
3709 if (!is_anon_ns(ns))
3710 ns_free_inum(&ns->ns);
3711 dec_mnt_namespaces(ns->ucounts);
3712 put_user_ns(ns->user_ns);
3713 kfree(ns);
3714}
3715
3716/*
3717 * Assign a sequence number so we can detect when we attempt to bind
3718 * mount a reference to an older mount namespace into the current
3719 * mount namespace, preventing reference counting loops. A 64bit
3720 * number incrementing at 10Ghz will take 12,427 years to wrap which
3721 * is effectively never, so we can ignore the possibility.
3722 */
3723static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3724
3725static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3726{
3727 struct mnt_namespace *new_ns;
3728 struct ucounts *ucounts;
3729 int ret;
3730
3731 ucounts = inc_mnt_namespaces(user_ns);
3732 if (!ucounts)
3733 return ERR_PTR(-ENOSPC);
3734
3735 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3736 if (!new_ns) {
3737 dec_mnt_namespaces(ucounts);
3738 return ERR_PTR(-ENOMEM);
3739 }
3740 if (!anon) {
3741 ret = ns_alloc_inum(&new_ns->ns);
3742 if (ret) {
3743 kfree(new_ns);
3744 dec_mnt_namespaces(ucounts);
3745 return ERR_PTR(ret);
3746 }
3747 }
3748 new_ns->ns.ops = &mntns_operations;
3749 if (!anon)
3750 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3751 refcount_set(&new_ns->ns.count, 1);
3752 new_ns->mounts = RB_ROOT;
3753 init_waitqueue_head(&new_ns->poll);
3754 new_ns->user_ns = get_user_ns(user_ns);
3755 new_ns->ucounts = ucounts;
3756 return new_ns;
3757}
3758
3759__latent_entropy
3760struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3761 struct user_namespace *user_ns, struct fs_struct *new_fs)
3762{
3763 struct mnt_namespace *new_ns;
3764 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3765 struct mount *p, *q;
3766 struct mount *old;
3767 struct mount *new;
3768 int copy_flags;
3769
3770 BUG_ON(!ns);
3771
3772 if (likely(!(flags & CLONE_NEWNS))) {
3773 get_mnt_ns(ns);
3774 return ns;
3775 }
3776
3777 old = ns->root;
3778
3779 new_ns = alloc_mnt_ns(user_ns, false);
3780 if (IS_ERR(new_ns))
3781 return new_ns;
3782
3783 namespace_lock();
3784 /* First pass: copy the tree topology */
3785 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3786 if (user_ns != ns->user_ns)
3787 copy_flags |= CL_SHARED_TO_SLAVE;
3788 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3789 if (IS_ERR(new)) {
3790 namespace_unlock();
3791 free_mnt_ns(new_ns);
3792 return ERR_CAST(new);
3793 }
3794 if (user_ns != ns->user_ns) {
3795 lock_mount_hash();
3796 lock_mnt_tree(new);
3797 unlock_mount_hash();
3798 }
3799 new_ns->root = new;
3800
3801 /*
3802 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3803 * as belonging to new namespace. We have already acquired a private
3804 * fs_struct, so tsk->fs->lock is not needed.
3805 */
3806 p = old;
3807 q = new;
3808 while (p) {
3809 mnt_add_to_ns(new_ns, q);
3810 new_ns->nr_mounts++;
3811 if (new_fs) {
3812 if (&p->mnt == new_fs->root.mnt) {
3813 new_fs->root.mnt = mntget(&q->mnt);
3814 rootmnt = &p->mnt;
3815 }
3816 if (&p->mnt == new_fs->pwd.mnt) {
3817 new_fs->pwd.mnt = mntget(&q->mnt);
3818 pwdmnt = &p->mnt;
3819 }
3820 }
3821 p = next_mnt(p, old);
3822 q = next_mnt(q, new);
3823 if (!q)
3824 break;
3825 // an mntns binding we'd skipped?
3826 while (p->mnt.mnt_root != q->mnt.mnt_root)
3827 p = next_mnt(skip_mnt_tree(p), old);
3828 }
3829 namespace_unlock();
3830
3831 if (rootmnt)
3832 mntput(rootmnt);
3833 if (pwdmnt)
3834 mntput(pwdmnt);
3835
3836 return new_ns;
3837}
3838
3839struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3840{
3841 struct mount *mnt = real_mount(m);
3842 struct mnt_namespace *ns;
3843 struct super_block *s;
3844 struct path path;
3845 int err;
3846
3847 ns = alloc_mnt_ns(&init_user_ns, true);
3848 if (IS_ERR(ns)) {
3849 mntput(m);
3850 return ERR_CAST(ns);
3851 }
3852 ns->root = mnt;
3853 ns->nr_mounts++;
3854 mnt_add_to_ns(ns, mnt);
3855
3856 err = vfs_path_lookup(m->mnt_root, m,
3857 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3858
3859 put_mnt_ns(ns);
3860
3861 if (err)
3862 return ERR_PTR(err);
3863
3864 /* trade a vfsmount reference for active sb one */
3865 s = path.mnt->mnt_sb;
3866 atomic_inc(&s->s_active);
3867 mntput(path.mnt);
3868 /* lock the sucker */
3869 down_write(&s->s_umount);
3870 /* ... and return the root of (sub)tree on it */
3871 return path.dentry;
3872}
3873EXPORT_SYMBOL(mount_subtree);
3874
3875SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3876 char __user *, type, unsigned long, flags, void __user *, data)
3877{
3878 int ret;
3879 char *kernel_type;
3880 char *kernel_dev;
3881 void *options;
3882
3883 kernel_type = copy_mount_string(type);
3884 ret = PTR_ERR(kernel_type);
3885 if (IS_ERR(kernel_type))
3886 goto out_type;
3887
3888 kernel_dev = copy_mount_string(dev_name);
3889 ret = PTR_ERR(kernel_dev);
3890 if (IS_ERR(kernel_dev))
3891 goto out_dev;
3892
3893 options = copy_mount_options(data);
3894 ret = PTR_ERR(options);
3895 if (IS_ERR(options))
3896 goto out_data;
3897
3898 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3899
3900 kfree(options);
3901out_data:
3902 kfree(kernel_dev);
3903out_dev:
3904 kfree(kernel_type);
3905out_type:
3906 return ret;
3907}
3908
3909#define FSMOUNT_VALID_FLAGS \
3910 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3911 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3912 MOUNT_ATTR_NOSYMFOLLOW)
3913
3914#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3915
3916#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3917 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3918
3919static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3920{
3921 unsigned int mnt_flags = 0;
3922
3923 if (attr_flags & MOUNT_ATTR_RDONLY)
3924 mnt_flags |= MNT_READONLY;
3925 if (attr_flags & MOUNT_ATTR_NOSUID)
3926 mnt_flags |= MNT_NOSUID;
3927 if (attr_flags & MOUNT_ATTR_NODEV)
3928 mnt_flags |= MNT_NODEV;
3929 if (attr_flags & MOUNT_ATTR_NOEXEC)
3930 mnt_flags |= MNT_NOEXEC;
3931 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3932 mnt_flags |= MNT_NODIRATIME;
3933 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3934 mnt_flags |= MNT_NOSYMFOLLOW;
3935
3936 return mnt_flags;
3937}
3938
3939/*
3940 * Create a kernel mount representation for a new, prepared superblock
3941 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3942 */
3943SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3944 unsigned int, attr_flags)
3945{
3946 struct mnt_namespace *ns;
3947 struct fs_context *fc;
3948 struct file *file;
3949 struct path newmount;
3950 struct mount *mnt;
3951 struct fd f;
3952 unsigned int mnt_flags = 0;
3953 long ret;
3954
3955 if (!may_mount())
3956 return -EPERM;
3957
3958 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3959 return -EINVAL;
3960
3961 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3962 return -EINVAL;
3963
3964 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3965
3966 switch (attr_flags & MOUNT_ATTR__ATIME) {
3967 case MOUNT_ATTR_STRICTATIME:
3968 break;
3969 case MOUNT_ATTR_NOATIME:
3970 mnt_flags |= MNT_NOATIME;
3971 break;
3972 case MOUNT_ATTR_RELATIME:
3973 mnt_flags |= MNT_RELATIME;
3974 break;
3975 default:
3976 return -EINVAL;
3977 }
3978
3979 f = fdget(fs_fd);
3980 if (!f.file)
3981 return -EBADF;
3982
3983 ret = -EINVAL;
3984 if (f.file->f_op != &fscontext_fops)
3985 goto err_fsfd;
3986
3987 fc = f.file->private_data;
3988
3989 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3990 if (ret < 0)
3991 goto err_fsfd;
3992
3993 /* There must be a valid superblock or we can't mount it */
3994 ret = -EINVAL;
3995 if (!fc->root)
3996 goto err_unlock;
3997
3998 ret = -EPERM;
3999 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
4000 pr_warn("VFS: Mount too revealing\n");
4001 goto err_unlock;
4002 }
4003
4004 ret = -EBUSY;
4005 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
4006 goto err_unlock;
4007
4008 if (fc->sb_flags & SB_MANDLOCK)
4009 warn_mandlock();
4010
4011 newmount.mnt = vfs_create_mount(fc);
4012 if (IS_ERR(newmount.mnt)) {
4013 ret = PTR_ERR(newmount.mnt);
4014 goto err_unlock;
4015 }
4016 newmount.dentry = dget(fc->root);
4017 newmount.mnt->mnt_flags = mnt_flags;
4018
4019 /* We've done the mount bit - now move the file context into more or
4020 * less the same state as if we'd done an fspick(). We don't want to
4021 * do any memory allocation or anything like that at this point as we
4022 * don't want to have to handle any errors incurred.
4023 */
4024 vfs_clean_context(fc);
4025
4026 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
4027 if (IS_ERR(ns)) {
4028 ret = PTR_ERR(ns);
4029 goto err_path;
4030 }
4031 mnt = real_mount(newmount.mnt);
4032 ns->root = mnt;
4033 ns->nr_mounts = 1;
4034 mnt_add_to_ns(ns, mnt);
4035 mntget(newmount.mnt);
4036
4037 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4038 * it, not just simply put it.
4039 */
4040 file = dentry_open(&newmount, O_PATH, fc->cred);
4041 if (IS_ERR(file)) {
4042 dissolve_on_fput(newmount.mnt);
4043 ret = PTR_ERR(file);
4044 goto err_path;
4045 }
4046 file->f_mode |= FMODE_NEED_UNMOUNT;
4047
4048 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
4049 if (ret >= 0)
4050 fd_install(ret, file);
4051 else
4052 fput(file);
4053
4054err_path:
4055 path_put(&newmount);
4056err_unlock:
4057 mutex_unlock(&fc->uapi_mutex);
4058err_fsfd:
4059 fdput(f);
4060 return ret;
4061}
4062
4063/*
4064 * Move a mount from one place to another. In combination with
4065 * fsopen()/fsmount() this is used to install a new mount and in combination
4066 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4067 * a mount subtree.
4068 *
4069 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4070 */
4071SYSCALL_DEFINE5(move_mount,
4072 int, from_dfd, const char __user *, from_pathname,
4073 int, to_dfd, const char __user *, to_pathname,
4074 unsigned int, flags)
4075{
4076 struct path from_path, to_path;
4077 unsigned int lflags;
4078 int ret = 0;
4079
4080 if (!may_mount())
4081 return -EPERM;
4082
4083 if (flags & ~MOVE_MOUNT__MASK)
4084 return -EINVAL;
4085
4086 if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) ==
4087 (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP))
4088 return -EINVAL;
4089
4090 /* If someone gives a pathname, they aren't permitted to move
4091 * from an fd that requires unmount as we can't get at the flag
4092 * to clear it afterwards.
4093 */
4094 lflags = 0;
4095 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4096 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4097 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4098
4099 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
4100 if (ret < 0)
4101 return ret;
4102
4103 lflags = 0;
4104 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4105 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4106 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4107
4108 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
4109 if (ret < 0)
4110 goto out_from;
4111
4112 ret = security_move_mount(&from_path, &to_path);
4113 if (ret < 0)
4114 goto out_to;
4115
4116 if (flags & MOVE_MOUNT_SET_GROUP)
4117 ret = do_set_group(&from_path, &to_path);
4118 else
4119 ret = do_move_mount(&from_path, &to_path,
4120 (flags & MOVE_MOUNT_BENEATH));
4121
4122out_to:
4123 path_put(&to_path);
4124out_from:
4125 path_put(&from_path);
4126 return ret;
4127}
4128
4129/*
4130 * Return true if path is reachable from root
4131 *
4132 * namespace_sem or mount_lock is held
4133 */
4134bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
4135 const struct path *root)
4136{
4137 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
4138 dentry = mnt->mnt_mountpoint;
4139 mnt = mnt->mnt_parent;
4140 }
4141 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
4142}
4143
4144bool path_is_under(const struct path *path1, const struct path *path2)
4145{
4146 bool res;
4147 read_seqlock_excl(&mount_lock);
4148 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
4149 read_sequnlock_excl(&mount_lock);
4150 return res;
4151}
4152EXPORT_SYMBOL(path_is_under);
4153
4154/*
4155 * pivot_root Semantics:
4156 * Moves the root file system of the current process to the directory put_old,
4157 * makes new_root as the new root file system of the current process, and sets
4158 * root/cwd of all processes which had them on the current root to new_root.
4159 *
4160 * Restrictions:
4161 * The new_root and put_old must be directories, and must not be on the
4162 * same file system as the current process root. The put_old must be
4163 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4164 * pointed to by put_old must yield the same directory as new_root. No other
4165 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4166 *
4167 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
4168 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4169 * in this situation.
4170 *
4171 * Notes:
4172 * - we don't move root/cwd if they are not at the root (reason: if something
4173 * cared enough to change them, it's probably wrong to force them elsewhere)
4174 * - it's okay to pick a root that isn't the root of a file system, e.g.
4175 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4176 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4177 * first.
4178 */
4179SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
4180 const char __user *, put_old)
4181{
4182 struct path new, old, root;
4183 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
4184 struct mountpoint *old_mp, *root_mp;
4185 int error;
4186
4187 if (!may_mount())
4188 return -EPERM;
4189
4190 error = user_path_at(AT_FDCWD, new_root,
4191 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
4192 if (error)
4193 goto out0;
4194
4195 error = user_path_at(AT_FDCWD, put_old,
4196 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
4197 if (error)
4198 goto out1;
4199
4200 error = security_sb_pivotroot(&old, &new);
4201 if (error)
4202 goto out2;
4203
4204 get_fs_root(current->fs, &root);
4205 old_mp = lock_mount(&old);
4206 error = PTR_ERR(old_mp);
4207 if (IS_ERR(old_mp))
4208 goto out3;
4209
4210 error = -EINVAL;
4211 new_mnt = real_mount(new.mnt);
4212 root_mnt = real_mount(root.mnt);
4213 old_mnt = real_mount(old.mnt);
4214 ex_parent = new_mnt->mnt_parent;
4215 root_parent = root_mnt->mnt_parent;
4216 if (IS_MNT_SHARED(old_mnt) ||
4217 IS_MNT_SHARED(ex_parent) ||
4218 IS_MNT_SHARED(root_parent))
4219 goto out4;
4220 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
4221 goto out4;
4222 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4223 goto out4;
4224 error = -ENOENT;
4225 if (d_unlinked(new.dentry))
4226 goto out4;
4227 error = -EBUSY;
4228 if (new_mnt == root_mnt || old_mnt == root_mnt)
4229 goto out4; /* loop, on the same file system */
4230 error = -EINVAL;
4231 if (!path_mounted(&root))
4232 goto out4; /* not a mountpoint */
4233 if (!mnt_has_parent(root_mnt))
4234 goto out4; /* not attached */
4235 if (!path_mounted(&new))
4236 goto out4; /* not a mountpoint */
4237 if (!mnt_has_parent(new_mnt))
4238 goto out4; /* not attached */
4239 /* make sure we can reach put_old from new_root */
4240 if (!is_path_reachable(old_mnt, old.dentry, &new))
4241 goto out4;
4242 /* make certain new is below the root */
4243 if (!is_path_reachable(new_mnt, new.dentry, &root))
4244 goto out4;
4245 lock_mount_hash();
4246 umount_mnt(new_mnt);
4247 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
4248 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4249 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4250 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4251 }
4252 /* mount old root on put_old */
4253 attach_mnt(root_mnt, old_mnt, old_mp, false);
4254 /* mount new_root on / */
4255 attach_mnt(new_mnt, root_parent, root_mp, false);
4256 mnt_add_count(root_parent, -1);
4257 touch_mnt_namespace(current->nsproxy->mnt_ns);
4258 /* A moved mount should not expire automatically */
4259 list_del_init(&new_mnt->mnt_expire);
4260 put_mountpoint(root_mp);
4261 unlock_mount_hash();
4262 chroot_fs_refs(&root, &new);
4263 error = 0;
4264out4:
4265 unlock_mount(old_mp);
4266 if (!error)
4267 mntput_no_expire(ex_parent);
4268out3:
4269 path_put(&root);
4270out2:
4271 path_put(&old);
4272out1:
4273 path_put(&new);
4274out0:
4275 return error;
4276}
4277
4278static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4279{
4280 unsigned int flags = mnt->mnt.mnt_flags;
4281
4282 /* flags to clear */
4283 flags &= ~kattr->attr_clr;
4284 /* flags to raise */
4285 flags |= kattr->attr_set;
4286
4287 return flags;
4288}
4289
4290static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4291{
4292 struct vfsmount *m = &mnt->mnt;
4293 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4294
4295 if (!kattr->mnt_idmap)
4296 return 0;
4297
4298 /*
4299 * Creating an idmapped mount with the filesystem wide idmapping
4300 * doesn't make sense so block that. We don't allow mushy semantics.
4301 */
4302 if (kattr->mnt_userns == m->mnt_sb->s_user_ns)
4303 return -EINVAL;
4304
4305 /*
4306 * Once a mount has been idmapped we don't allow it to change its
4307 * mapping. It makes things simpler and callers can just create
4308 * another bind-mount they can idmap if they want to.
4309 */
4310 if (is_idmapped_mnt(m))
4311 return -EPERM;
4312
4313 /* The underlying filesystem doesn't support idmapped mounts yet. */
4314 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4315 return -EINVAL;
4316
4317 /* We're not controlling the superblock. */
4318 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4319 return -EPERM;
4320
4321 /* Mount has already been visible in the filesystem hierarchy. */
4322 if (!is_anon_ns(mnt->mnt_ns))
4323 return -EINVAL;
4324
4325 return 0;
4326}
4327
4328/**
4329 * mnt_allow_writers() - check whether the attribute change allows writers
4330 * @kattr: the new mount attributes
4331 * @mnt: the mount to which @kattr will be applied
4332 *
4333 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4334 *
4335 * Return: true if writers need to be held, false if not
4336 */
4337static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4338 const struct mount *mnt)
4339{
4340 return (!(kattr->attr_set & MNT_READONLY) ||
4341 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4342 !kattr->mnt_idmap;
4343}
4344
4345static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4346{
4347 struct mount *m;
4348 int err;
4349
4350 for (m = mnt; m; m = next_mnt(m, mnt)) {
4351 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4352 err = -EPERM;
4353 break;
4354 }
4355
4356 err = can_idmap_mount(kattr, m);
4357 if (err)
4358 break;
4359
4360 if (!mnt_allow_writers(kattr, m)) {
4361 err = mnt_hold_writers(m);
4362 if (err)
4363 break;
4364 }
4365
4366 if (!kattr->recurse)
4367 return 0;
4368 }
4369
4370 if (err) {
4371 struct mount *p;
4372
4373 /*
4374 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4375 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4376 * mounts and needs to take care to include the first mount.
4377 */
4378 for (p = mnt; p; p = next_mnt(p, mnt)) {
4379 /* If we had to hold writers unblock them. */
4380 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4381 mnt_unhold_writers(p);
4382
4383 /*
4384 * We're done once the first mount we changed got
4385 * MNT_WRITE_HOLD unset.
4386 */
4387 if (p == m)
4388 break;
4389 }
4390 }
4391 return err;
4392}
4393
4394static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4395{
4396 if (!kattr->mnt_idmap)
4397 return;
4398
4399 /*
4400 * Pairs with smp_load_acquire() in mnt_idmap().
4401 *
4402 * Since we only allow a mount to change the idmapping once and
4403 * verified this in can_idmap_mount() we know that the mount has
4404 * @nop_mnt_idmap attached to it. So there's no need to drop any
4405 * references.
4406 */
4407 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4408}
4409
4410static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4411{
4412 struct mount *m;
4413
4414 for (m = mnt; m; m = next_mnt(m, mnt)) {
4415 unsigned int flags;
4416
4417 do_idmap_mount(kattr, m);
4418 flags = recalc_flags(kattr, m);
4419 WRITE_ONCE(m->mnt.mnt_flags, flags);
4420
4421 /* If we had to hold writers unblock them. */
4422 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4423 mnt_unhold_writers(m);
4424
4425 if (kattr->propagation)
4426 change_mnt_propagation(m, kattr->propagation);
4427 if (!kattr->recurse)
4428 break;
4429 }
4430 touch_mnt_namespace(mnt->mnt_ns);
4431}
4432
4433static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4434{
4435 struct mount *mnt = real_mount(path->mnt);
4436 int err = 0;
4437
4438 if (!path_mounted(path))
4439 return -EINVAL;
4440
4441 if (kattr->mnt_userns) {
4442 struct mnt_idmap *mnt_idmap;
4443
4444 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4445 if (IS_ERR(mnt_idmap))
4446 return PTR_ERR(mnt_idmap);
4447 kattr->mnt_idmap = mnt_idmap;
4448 }
4449
4450 if (kattr->propagation) {
4451 /*
4452 * Only take namespace_lock() if we're actually changing
4453 * propagation.
4454 */
4455 namespace_lock();
4456 if (kattr->propagation == MS_SHARED) {
4457 err = invent_group_ids(mnt, kattr->recurse);
4458 if (err) {
4459 namespace_unlock();
4460 return err;
4461 }
4462 }
4463 }
4464
4465 err = -EINVAL;
4466 lock_mount_hash();
4467
4468 /* Ensure that this isn't anything purely vfs internal. */
4469 if (!is_mounted(&mnt->mnt))
4470 goto out;
4471
4472 /*
4473 * If this is an attached mount make sure it's located in the callers
4474 * mount namespace. If it's not don't let the caller interact with it.
4475 *
4476 * If this mount doesn't have a parent it's most often simply a
4477 * detached mount with an anonymous mount namespace. IOW, something
4478 * that's simply not attached yet. But there are apparently also users
4479 * that do change mount properties on the rootfs itself. That obviously
4480 * neither has a parent nor is it a detached mount so we cannot
4481 * unconditionally check for detached mounts.
4482 */
4483 if ((mnt_has_parent(mnt) || !is_anon_ns(mnt->mnt_ns)) && !check_mnt(mnt))
4484 goto out;
4485
4486 /*
4487 * First, we get the mount tree in a shape where we can change mount
4488 * properties without failure. If we succeeded to do so we commit all
4489 * changes and if we failed we clean up.
4490 */
4491 err = mount_setattr_prepare(kattr, mnt);
4492 if (!err)
4493 mount_setattr_commit(kattr, mnt);
4494
4495out:
4496 unlock_mount_hash();
4497
4498 if (kattr->propagation) {
4499 if (err)
4500 cleanup_group_ids(mnt, NULL);
4501 namespace_unlock();
4502 }
4503
4504 return err;
4505}
4506
4507static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4508 struct mount_kattr *kattr, unsigned int flags)
4509{
4510 int err = 0;
4511 struct ns_common *ns;
4512 struct user_namespace *mnt_userns;
4513 struct fd f;
4514
4515 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4516 return 0;
4517
4518 /*
4519 * We currently do not support clearing an idmapped mount. If this ever
4520 * is a use-case we can revisit this but for now let's keep it simple
4521 * and not allow it.
4522 */
4523 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4524 return -EINVAL;
4525
4526 if (attr->userns_fd > INT_MAX)
4527 return -EINVAL;
4528
4529 f = fdget(attr->userns_fd);
4530 if (!f.file)
4531 return -EBADF;
4532
4533 if (!proc_ns_file(f.file)) {
4534 err = -EINVAL;
4535 goto out_fput;
4536 }
4537
4538 ns = get_proc_ns(file_inode(f.file));
4539 if (ns->ops->type != CLONE_NEWUSER) {
4540 err = -EINVAL;
4541 goto out_fput;
4542 }
4543
4544 /*
4545 * The initial idmapping cannot be used to create an idmapped
4546 * mount. We use the initial idmapping as an indicator of a mount
4547 * that is not idmapped. It can simply be passed into helpers that
4548 * are aware of idmapped mounts as a convenient shortcut. A user
4549 * can just create a dedicated identity mapping to achieve the same
4550 * result.
4551 */
4552 mnt_userns = container_of(ns, struct user_namespace, ns);
4553 if (mnt_userns == &init_user_ns) {
4554 err = -EPERM;
4555 goto out_fput;
4556 }
4557
4558 /* We're not controlling the target namespace. */
4559 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4560 err = -EPERM;
4561 goto out_fput;
4562 }
4563
4564 kattr->mnt_userns = get_user_ns(mnt_userns);
4565
4566out_fput:
4567 fdput(f);
4568 return err;
4569}
4570
4571static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4572 struct mount_kattr *kattr, unsigned int flags)
4573{
4574 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4575
4576 if (flags & AT_NO_AUTOMOUNT)
4577 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4578 if (flags & AT_SYMLINK_NOFOLLOW)
4579 lookup_flags &= ~LOOKUP_FOLLOW;
4580 if (flags & AT_EMPTY_PATH)
4581 lookup_flags |= LOOKUP_EMPTY;
4582
4583 *kattr = (struct mount_kattr) {
4584 .lookup_flags = lookup_flags,
4585 .recurse = !!(flags & AT_RECURSIVE),
4586 };
4587
4588 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4589 return -EINVAL;
4590 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4591 return -EINVAL;
4592 kattr->propagation = attr->propagation;
4593
4594 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4595 return -EINVAL;
4596
4597 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4598 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4599
4600 /*
4601 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4602 * users wanting to transition to a different atime setting cannot
4603 * simply specify the atime setting in @attr_set, but must also
4604 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4605 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4606 * @attr_clr and that @attr_set can't have any atime bits set if
4607 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4608 */
4609 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4610 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4611 return -EINVAL;
4612
4613 /*
4614 * Clear all previous time settings as they are mutually
4615 * exclusive.
4616 */
4617 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4618 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4619 case MOUNT_ATTR_RELATIME:
4620 kattr->attr_set |= MNT_RELATIME;
4621 break;
4622 case MOUNT_ATTR_NOATIME:
4623 kattr->attr_set |= MNT_NOATIME;
4624 break;
4625 case MOUNT_ATTR_STRICTATIME:
4626 break;
4627 default:
4628 return -EINVAL;
4629 }
4630 } else {
4631 if (attr->attr_set & MOUNT_ATTR__ATIME)
4632 return -EINVAL;
4633 }
4634
4635 return build_mount_idmapped(attr, usize, kattr, flags);
4636}
4637
4638static void finish_mount_kattr(struct mount_kattr *kattr)
4639{
4640 put_user_ns(kattr->mnt_userns);
4641 kattr->mnt_userns = NULL;
4642
4643 if (kattr->mnt_idmap)
4644 mnt_idmap_put(kattr->mnt_idmap);
4645}
4646
4647SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4648 unsigned int, flags, struct mount_attr __user *, uattr,
4649 size_t, usize)
4650{
4651 int err;
4652 struct path target;
4653 struct mount_attr attr;
4654 struct mount_kattr kattr;
4655
4656 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4657
4658 if (flags & ~(AT_EMPTY_PATH |
4659 AT_RECURSIVE |
4660 AT_SYMLINK_NOFOLLOW |
4661 AT_NO_AUTOMOUNT))
4662 return -EINVAL;
4663
4664 if (unlikely(usize > PAGE_SIZE))
4665 return -E2BIG;
4666 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4667 return -EINVAL;
4668
4669 if (!may_mount())
4670 return -EPERM;
4671
4672 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4673 if (err)
4674 return err;
4675
4676 /* Don't bother walking through the mounts if this is a nop. */
4677 if (attr.attr_set == 0 &&
4678 attr.attr_clr == 0 &&
4679 attr.propagation == 0)
4680 return 0;
4681
4682 err = build_mount_kattr(&attr, usize, &kattr, flags);
4683 if (err)
4684 return err;
4685
4686 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4687 if (!err) {
4688 err = do_mount_setattr(&target, &kattr);
4689 path_put(&target);
4690 }
4691 finish_mount_kattr(&kattr);
4692 return err;
4693}
4694
4695int show_path(struct seq_file *m, struct dentry *root)
4696{
4697 if (root->d_sb->s_op->show_path)
4698 return root->d_sb->s_op->show_path(m, root);
4699
4700 seq_dentry(m, root, " \t\n\\");
4701 return 0;
4702}
4703
4704static struct vfsmount *lookup_mnt_in_ns(u64 id, struct mnt_namespace *ns)
4705{
4706 struct mount *mnt = mnt_find_id_at(ns, id);
4707
4708 if (!mnt || mnt->mnt_id_unique != id)
4709 return NULL;
4710
4711 return &mnt->mnt;
4712}
4713
4714struct kstatmount {
4715 struct statmount __user *buf;
4716 size_t bufsize;
4717 struct vfsmount *mnt;
4718 u64 mask;
4719 struct path root;
4720 struct statmount sm;
4721 struct seq_file seq;
4722};
4723
4724static u64 mnt_to_attr_flags(struct vfsmount *mnt)
4725{
4726 unsigned int mnt_flags = READ_ONCE(mnt->mnt_flags);
4727 u64 attr_flags = 0;
4728
4729 if (mnt_flags & MNT_READONLY)
4730 attr_flags |= MOUNT_ATTR_RDONLY;
4731 if (mnt_flags & MNT_NOSUID)
4732 attr_flags |= MOUNT_ATTR_NOSUID;
4733 if (mnt_flags & MNT_NODEV)
4734 attr_flags |= MOUNT_ATTR_NODEV;
4735 if (mnt_flags & MNT_NOEXEC)
4736 attr_flags |= MOUNT_ATTR_NOEXEC;
4737 if (mnt_flags & MNT_NODIRATIME)
4738 attr_flags |= MOUNT_ATTR_NODIRATIME;
4739 if (mnt_flags & MNT_NOSYMFOLLOW)
4740 attr_flags |= MOUNT_ATTR_NOSYMFOLLOW;
4741
4742 if (mnt_flags & MNT_NOATIME)
4743 attr_flags |= MOUNT_ATTR_NOATIME;
4744 else if (mnt_flags & MNT_RELATIME)
4745 attr_flags |= MOUNT_ATTR_RELATIME;
4746 else
4747 attr_flags |= MOUNT_ATTR_STRICTATIME;
4748
4749 if (is_idmapped_mnt(mnt))
4750 attr_flags |= MOUNT_ATTR_IDMAP;
4751
4752 return attr_flags;
4753}
4754
4755static u64 mnt_to_propagation_flags(struct mount *m)
4756{
4757 u64 propagation = 0;
4758
4759 if (IS_MNT_SHARED(m))
4760 propagation |= MS_SHARED;
4761 if (IS_MNT_SLAVE(m))
4762 propagation |= MS_SLAVE;
4763 if (IS_MNT_UNBINDABLE(m))
4764 propagation |= MS_UNBINDABLE;
4765 if (!propagation)
4766 propagation |= MS_PRIVATE;
4767
4768 return propagation;
4769}
4770
4771static void statmount_sb_basic(struct kstatmount *s)
4772{
4773 struct super_block *sb = s->mnt->mnt_sb;
4774
4775 s->sm.mask |= STATMOUNT_SB_BASIC;
4776 s->sm.sb_dev_major = MAJOR(sb->s_dev);
4777 s->sm.sb_dev_minor = MINOR(sb->s_dev);
4778 s->sm.sb_magic = sb->s_magic;
4779 s->sm.sb_flags = sb->s_flags & (SB_RDONLY|SB_SYNCHRONOUS|SB_DIRSYNC|SB_LAZYTIME);
4780}
4781
4782static void statmount_mnt_basic(struct kstatmount *s)
4783{
4784 struct mount *m = real_mount(s->mnt);
4785
4786 s->sm.mask |= STATMOUNT_MNT_BASIC;
4787 s->sm.mnt_id = m->mnt_id_unique;
4788 s->sm.mnt_parent_id = m->mnt_parent->mnt_id_unique;
4789 s->sm.mnt_id_old = m->mnt_id;
4790 s->sm.mnt_parent_id_old = m->mnt_parent->mnt_id;
4791 s->sm.mnt_attr = mnt_to_attr_flags(&m->mnt);
4792 s->sm.mnt_propagation = mnt_to_propagation_flags(m);
4793 s->sm.mnt_peer_group = IS_MNT_SHARED(m) ? m->mnt_group_id : 0;
4794 s->sm.mnt_master = IS_MNT_SLAVE(m) ? m->mnt_master->mnt_group_id : 0;
4795}
4796
4797static void statmount_propagate_from(struct kstatmount *s)
4798{
4799 struct mount *m = real_mount(s->mnt);
4800
4801 s->sm.mask |= STATMOUNT_PROPAGATE_FROM;
4802 if (IS_MNT_SLAVE(m))
4803 s->sm.propagate_from = get_dominating_id(m, ¤t->fs->root);
4804}
4805
4806static int statmount_mnt_root(struct kstatmount *s, struct seq_file *seq)
4807{
4808 int ret;
4809 size_t start = seq->count;
4810
4811 ret = show_path(seq, s->mnt->mnt_root);
4812 if (ret)
4813 return ret;
4814
4815 if (unlikely(seq_has_overflowed(seq)))
4816 return -EAGAIN;
4817
4818 /*
4819 * Unescape the result. It would be better if supplied string was not
4820 * escaped in the first place, but that's a pretty invasive change.
4821 */
4822 seq->buf[seq->count] = '\0';
4823 seq->count = start;
4824 seq_commit(seq, string_unescape_inplace(seq->buf + start, UNESCAPE_OCTAL));
4825 return 0;
4826}
4827
4828static int statmount_mnt_point(struct kstatmount *s, struct seq_file *seq)
4829{
4830 struct vfsmount *mnt = s->mnt;
4831 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
4832 int err;
4833
4834 err = seq_path_root(seq, &mnt_path, &s->root, "");
4835 return err == SEQ_SKIP ? 0 : err;
4836}
4837
4838static int statmount_fs_type(struct kstatmount *s, struct seq_file *seq)
4839{
4840 struct super_block *sb = s->mnt->mnt_sb;
4841
4842 seq_puts(seq, sb->s_type->name);
4843 return 0;
4844}
4845
4846static int statmount_string(struct kstatmount *s, u64 flag)
4847{
4848 int ret;
4849 size_t kbufsize;
4850 struct seq_file *seq = &s->seq;
4851 struct statmount *sm = &s->sm;
4852
4853 switch (flag) {
4854 case STATMOUNT_FS_TYPE:
4855 sm->fs_type = seq->count;
4856 ret = statmount_fs_type(s, seq);
4857 break;
4858 case STATMOUNT_MNT_ROOT:
4859 sm->mnt_root = seq->count;
4860 ret = statmount_mnt_root(s, seq);
4861 break;
4862 case STATMOUNT_MNT_POINT:
4863 sm->mnt_point = seq->count;
4864 ret = statmount_mnt_point(s, seq);
4865 break;
4866 default:
4867 WARN_ON_ONCE(true);
4868 return -EINVAL;
4869 }
4870
4871 if (unlikely(check_add_overflow(sizeof(*sm), seq->count, &kbufsize)))
4872 return -EOVERFLOW;
4873 if (kbufsize >= s->bufsize)
4874 return -EOVERFLOW;
4875
4876 /* signal a retry */
4877 if (unlikely(seq_has_overflowed(seq)))
4878 return -EAGAIN;
4879
4880 if (ret)
4881 return ret;
4882
4883 seq->buf[seq->count++] = '\0';
4884 sm->mask |= flag;
4885 return 0;
4886}
4887
4888static int copy_statmount_to_user(struct kstatmount *s)
4889{
4890 struct statmount *sm = &s->sm;
4891 struct seq_file *seq = &s->seq;
4892 char __user *str = ((char __user *)s->buf) + sizeof(*sm);
4893 size_t copysize = min_t(size_t, s->bufsize, sizeof(*sm));
4894
4895 if (seq->count && copy_to_user(str, seq->buf, seq->count))
4896 return -EFAULT;
4897
4898 /* Return the number of bytes copied to the buffer */
4899 sm->size = copysize + seq->count;
4900 if (copy_to_user(s->buf, sm, copysize))
4901 return -EFAULT;
4902
4903 return 0;
4904}
4905
4906static int do_statmount(struct kstatmount *s)
4907{
4908 struct mount *m = real_mount(s->mnt);
4909 int err;
4910
4911 /*
4912 * Don't trigger audit denials. We just want to determine what
4913 * mounts to show users.
4914 */
4915 if (!is_path_reachable(m, m->mnt.mnt_root, &s->root) &&
4916 !ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN))
4917 return -EPERM;
4918
4919 err = security_sb_statfs(s->mnt->mnt_root);
4920 if (err)
4921 return err;
4922
4923 if (s->mask & STATMOUNT_SB_BASIC)
4924 statmount_sb_basic(s);
4925
4926 if (s->mask & STATMOUNT_MNT_BASIC)
4927 statmount_mnt_basic(s);
4928
4929 if (s->mask & STATMOUNT_PROPAGATE_FROM)
4930 statmount_propagate_from(s);
4931
4932 if (s->mask & STATMOUNT_FS_TYPE)
4933 err = statmount_string(s, STATMOUNT_FS_TYPE);
4934
4935 if (!err && s->mask & STATMOUNT_MNT_ROOT)
4936 err = statmount_string(s, STATMOUNT_MNT_ROOT);
4937
4938 if (!err && s->mask & STATMOUNT_MNT_POINT)
4939 err = statmount_string(s, STATMOUNT_MNT_POINT);
4940
4941 if (err)
4942 return err;
4943
4944 return 0;
4945}
4946
4947static inline bool retry_statmount(const long ret, size_t *seq_size)
4948{
4949 if (likely(ret != -EAGAIN))
4950 return false;
4951 if (unlikely(check_mul_overflow(*seq_size, 2, seq_size)))
4952 return false;
4953 if (unlikely(*seq_size > MAX_RW_COUNT))
4954 return false;
4955 return true;
4956}
4957
4958static int prepare_kstatmount(struct kstatmount *ks, struct mnt_id_req *kreq,
4959 struct statmount __user *buf, size_t bufsize,
4960 size_t seq_size)
4961{
4962 if (!access_ok(buf, bufsize))
4963 return -EFAULT;
4964
4965 memset(ks, 0, sizeof(*ks));
4966 ks->mask = kreq->param;
4967 ks->buf = buf;
4968 ks->bufsize = bufsize;
4969 ks->seq.size = seq_size;
4970 ks->seq.buf = kvmalloc(seq_size, GFP_KERNEL_ACCOUNT);
4971 if (!ks->seq.buf)
4972 return -ENOMEM;
4973 return 0;
4974}
4975
4976static int copy_mnt_id_req(const struct mnt_id_req __user *req,
4977 struct mnt_id_req *kreq)
4978{
4979 int ret;
4980 size_t usize;
4981
4982 BUILD_BUG_ON(sizeof(struct mnt_id_req) != MNT_ID_REQ_SIZE_VER0);
4983
4984 ret = get_user(usize, &req->size);
4985 if (ret)
4986 return -EFAULT;
4987 if (unlikely(usize > PAGE_SIZE))
4988 return -E2BIG;
4989 if (unlikely(usize < MNT_ID_REQ_SIZE_VER0))
4990 return -EINVAL;
4991 memset(kreq, 0, sizeof(*kreq));
4992 ret = copy_struct_from_user(kreq, sizeof(*kreq), req, usize);
4993 if (ret)
4994 return ret;
4995 if (kreq->spare != 0)
4996 return -EINVAL;
4997 return 0;
4998}
4999
5000SYSCALL_DEFINE4(statmount, const struct mnt_id_req __user *, req,
5001 struct statmount __user *, buf, size_t, bufsize,
5002 unsigned int, flags)
5003{
5004 struct vfsmount *mnt;
5005 struct mnt_id_req kreq;
5006 struct kstatmount ks;
5007 /* We currently support retrieval of 3 strings. */
5008 size_t seq_size = 3 * PATH_MAX;
5009 int ret;
5010
5011 if (flags)
5012 return -EINVAL;
5013
5014 ret = copy_mnt_id_req(req, &kreq);
5015 if (ret)
5016 return ret;
5017
5018retry:
5019 ret = prepare_kstatmount(&ks, &kreq, buf, bufsize, seq_size);
5020 if (ret)
5021 return ret;
5022
5023 down_read(&namespace_sem);
5024 mnt = lookup_mnt_in_ns(kreq.mnt_id, current->nsproxy->mnt_ns);
5025 if (!mnt) {
5026 up_read(&namespace_sem);
5027 kvfree(ks.seq.buf);
5028 return -ENOENT;
5029 }
5030
5031 ks.mnt = mnt;
5032 get_fs_root(current->fs, &ks.root);
5033 ret = do_statmount(&ks);
5034 path_put(&ks.root);
5035 up_read(&namespace_sem);
5036
5037 if (!ret)
5038 ret = copy_statmount_to_user(&ks);
5039 kvfree(ks.seq.buf);
5040 if (retry_statmount(ret, &seq_size))
5041 goto retry;
5042 return ret;
5043}
5044
5045static struct mount *listmnt_next(struct mount *curr)
5046{
5047 return node_to_mount(rb_next(&curr->mnt_node));
5048}
5049
5050static ssize_t do_listmount(struct mount *first, struct path *orig,
5051 u64 mnt_parent_id, u64 __user *mnt_ids,
5052 size_t nr_mnt_ids, const struct path *root)
5053{
5054 struct mount *r;
5055 ssize_t ret;
5056
5057 /*
5058 * Don't trigger audit denials. We just want to determine what
5059 * mounts to show users.
5060 */
5061 if (!is_path_reachable(real_mount(orig->mnt), orig->dentry, root) &&
5062 !ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN))
5063 return -EPERM;
5064
5065 ret = security_sb_statfs(orig->dentry);
5066 if (ret)
5067 return ret;
5068
5069 for (ret = 0, r = first; r && nr_mnt_ids; r = listmnt_next(r)) {
5070 if (r->mnt_id_unique == mnt_parent_id)
5071 continue;
5072 if (!is_path_reachable(r, r->mnt.mnt_root, orig))
5073 continue;
5074 if (put_user(r->mnt_id_unique, mnt_ids))
5075 return -EFAULT;
5076 mnt_ids++;
5077 nr_mnt_ids--;
5078 ret++;
5079 }
5080 return ret;
5081}
5082
5083SYSCALL_DEFINE4(listmount, const struct mnt_id_req __user *, req, u64 __user *,
5084 mnt_ids, size_t, nr_mnt_ids, unsigned int, flags)
5085{
5086 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
5087 struct mnt_id_req kreq;
5088 struct mount *first;
5089 struct path root, orig;
5090 u64 mnt_parent_id, last_mnt_id;
5091 const size_t maxcount = (size_t)-1 >> 3;
5092 ssize_t ret;
5093
5094 if (flags)
5095 return -EINVAL;
5096
5097 if (unlikely(nr_mnt_ids > maxcount))
5098 return -EFAULT;
5099
5100 if (!access_ok(mnt_ids, nr_mnt_ids * sizeof(*mnt_ids)))
5101 return -EFAULT;
5102
5103 ret = copy_mnt_id_req(req, &kreq);
5104 if (ret)
5105 return ret;
5106 mnt_parent_id = kreq.mnt_id;
5107 last_mnt_id = kreq.param;
5108
5109 down_read(&namespace_sem);
5110 get_fs_root(current->fs, &root);
5111 if (mnt_parent_id == LSMT_ROOT) {
5112 orig = root;
5113 } else {
5114 ret = -ENOENT;
5115 orig.mnt = lookup_mnt_in_ns(mnt_parent_id, ns);
5116 if (!orig.mnt)
5117 goto err;
5118 orig.dentry = orig.mnt->mnt_root;
5119 }
5120 if (!last_mnt_id)
5121 first = node_to_mount(rb_first(&ns->mounts));
5122 else
5123 first = mnt_find_id_at(ns, last_mnt_id + 1);
5124
5125 ret = do_listmount(first, &orig, mnt_parent_id, mnt_ids, nr_mnt_ids, &root);
5126err:
5127 path_put(&root);
5128 up_read(&namespace_sem);
5129 return ret;
5130}
5131
5132
5133static void __init init_mount_tree(void)
5134{
5135 struct vfsmount *mnt;
5136 struct mount *m;
5137 struct mnt_namespace *ns;
5138 struct path root;
5139
5140 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
5141 if (IS_ERR(mnt))
5142 panic("Can't create rootfs");
5143
5144 ns = alloc_mnt_ns(&init_user_ns, false);
5145 if (IS_ERR(ns))
5146 panic("Can't allocate initial namespace");
5147 m = real_mount(mnt);
5148 ns->root = m;
5149 ns->nr_mounts = 1;
5150 mnt_add_to_ns(ns, m);
5151 init_task.nsproxy->mnt_ns = ns;
5152 get_mnt_ns(ns);
5153
5154 root.mnt = mnt;
5155 root.dentry = mnt->mnt_root;
5156 mnt->mnt_flags |= MNT_LOCKED;
5157
5158 set_fs_pwd(current->fs, &root);
5159 set_fs_root(current->fs, &root);
5160}
5161
5162void __init mnt_init(void)
5163{
5164 int err;
5165
5166 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
5167 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
5168
5169 mount_hashtable = alloc_large_system_hash("Mount-cache",
5170 sizeof(struct hlist_head),
5171 mhash_entries, 19,
5172 HASH_ZERO,
5173 &m_hash_shift, &m_hash_mask, 0, 0);
5174 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
5175 sizeof(struct hlist_head),
5176 mphash_entries, 19,
5177 HASH_ZERO,
5178 &mp_hash_shift, &mp_hash_mask, 0, 0);
5179
5180 if (!mount_hashtable || !mountpoint_hashtable)
5181 panic("Failed to allocate mount hash table\n");
5182
5183 kernfs_init();
5184
5185 err = sysfs_init();
5186 if (err)
5187 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
5188 __func__, err);
5189 fs_kobj = kobject_create_and_add("fs", NULL);
5190 if (!fs_kobj)
5191 printk(KERN_WARNING "%s: kobj create error\n", __func__);
5192 shmem_init();
5193 init_rootfs();
5194 init_mount_tree();
5195}
5196
5197void put_mnt_ns(struct mnt_namespace *ns)
5198{
5199 if (!refcount_dec_and_test(&ns->ns.count))
5200 return;
5201 drop_collected_mounts(&ns->root->mnt);
5202 free_mnt_ns(ns);
5203}
5204
5205struct vfsmount *kern_mount(struct file_system_type *type)
5206{
5207 struct vfsmount *mnt;
5208 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
5209 if (!IS_ERR(mnt)) {
5210 /*
5211 * it is a longterm mount, don't release mnt until
5212 * we unmount before file sys is unregistered
5213 */
5214 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
5215 }
5216 return mnt;
5217}
5218EXPORT_SYMBOL_GPL(kern_mount);
5219
5220void kern_unmount(struct vfsmount *mnt)
5221{
5222 /* release long term mount so mount point can be released */
5223 if (!IS_ERR(mnt)) {
5224 mnt_make_shortterm(mnt);
5225 synchronize_rcu(); /* yecchhh... */
5226 mntput(mnt);
5227 }
5228}
5229EXPORT_SYMBOL(kern_unmount);
5230
5231void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
5232{
5233 unsigned int i;
5234
5235 for (i = 0; i < num; i++)
5236 mnt_make_shortterm(mnt[i]);
5237 synchronize_rcu_expedited();
5238 for (i = 0; i < num; i++)
5239 mntput(mnt[i]);
5240}
5241EXPORT_SYMBOL(kern_unmount_array);
5242
5243bool our_mnt(struct vfsmount *mnt)
5244{
5245 return check_mnt(real_mount(mnt));
5246}
5247
5248bool current_chrooted(void)
5249{
5250 /* Does the current process have a non-standard root */
5251 struct path ns_root;
5252 struct path fs_root;
5253 bool chrooted;
5254
5255 /* Find the namespace root */
5256 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
5257 ns_root.dentry = ns_root.mnt->mnt_root;
5258 path_get(&ns_root);
5259 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
5260 ;
5261
5262 get_fs_root(current->fs, &fs_root);
5263
5264 chrooted = !path_equal(&fs_root, &ns_root);
5265
5266 path_put(&fs_root);
5267 path_put(&ns_root);
5268
5269 return chrooted;
5270}
5271
5272static bool mnt_already_visible(struct mnt_namespace *ns,
5273 const struct super_block *sb,
5274 int *new_mnt_flags)
5275{
5276 int new_flags = *new_mnt_flags;
5277 struct mount *mnt, *n;
5278 bool visible = false;
5279
5280 down_read(&namespace_sem);
5281 rbtree_postorder_for_each_entry_safe(mnt, n, &ns->mounts, mnt_node) {
5282 struct mount *child;
5283 int mnt_flags;
5284
5285 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
5286 continue;
5287
5288 /* This mount is not fully visible if it's root directory
5289 * is not the root directory of the filesystem.
5290 */
5291 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
5292 continue;
5293
5294 /* A local view of the mount flags */
5295 mnt_flags = mnt->mnt.mnt_flags;
5296
5297 /* Don't miss readonly hidden in the superblock flags */
5298 if (sb_rdonly(mnt->mnt.mnt_sb))
5299 mnt_flags |= MNT_LOCK_READONLY;
5300
5301 /* Verify the mount flags are equal to or more permissive
5302 * than the proposed new mount.
5303 */
5304 if ((mnt_flags & MNT_LOCK_READONLY) &&
5305 !(new_flags & MNT_READONLY))
5306 continue;
5307 if ((mnt_flags & MNT_LOCK_ATIME) &&
5308 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
5309 continue;
5310
5311 /* This mount is not fully visible if there are any
5312 * locked child mounts that cover anything except for
5313 * empty directories.
5314 */
5315 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
5316 struct inode *inode = child->mnt_mountpoint->d_inode;
5317 /* Only worry about locked mounts */
5318 if (!(child->mnt.mnt_flags & MNT_LOCKED))
5319 continue;
5320 /* Is the directory permanetly empty? */
5321 if (!is_empty_dir_inode(inode))
5322 goto next;
5323 }
5324 /* Preserve the locked attributes */
5325 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
5326 MNT_LOCK_ATIME);
5327 visible = true;
5328 goto found;
5329 next: ;
5330 }
5331found:
5332 up_read(&namespace_sem);
5333 return visible;
5334}
5335
5336static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
5337{
5338 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
5339 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
5340 unsigned long s_iflags;
5341
5342 if (ns->user_ns == &init_user_ns)
5343 return false;
5344
5345 /* Can this filesystem be too revealing? */
5346 s_iflags = sb->s_iflags;
5347 if (!(s_iflags & SB_I_USERNS_VISIBLE))
5348 return false;
5349
5350 if ((s_iflags & required_iflags) != required_iflags) {
5351 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
5352 required_iflags);
5353 return true;
5354 }
5355
5356 return !mnt_already_visible(ns, sb, new_mnt_flags);
5357}
5358
5359bool mnt_may_suid(struct vfsmount *mnt)
5360{
5361 /*
5362 * Foreign mounts (accessed via fchdir or through /proc
5363 * symlinks) are always treated as if they are nosuid. This
5364 * prevents namespaces from trusting potentially unsafe
5365 * suid/sgid bits, file caps, or security labels that originate
5366 * in other namespaces.
5367 */
5368 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
5369 current_in_userns(mnt->mnt_sb->s_user_ns);
5370}
5371
5372static struct ns_common *mntns_get(struct task_struct *task)
5373{
5374 struct ns_common *ns = NULL;
5375 struct nsproxy *nsproxy;
5376
5377 task_lock(task);
5378 nsproxy = task->nsproxy;
5379 if (nsproxy) {
5380 ns = &nsproxy->mnt_ns->ns;
5381 get_mnt_ns(to_mnt_ns(ns));
5382 }
5383 task_unlock(task);
5384
5385 return ns;
5386}
5387
5388static void mntns_put(struct ns_common *ns)
5389{
5390 put_mnt_ns(to_mnt_ns(ns));
5391}
5392
5393static int mntns_install(struct nsset *nsset, struct ns_common *ns)
5394{
5395 struct nsproxy *nsproxy = nsset->nsproxy;
5396 struct fs_struct *fs = nsset->fs;
5397 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
5398 struct user_namespace *user_ns = nsset->cred->user_ns;
5399 struct path root;
5400 int err;
5401
5402 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
5403 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
5404 !ns_capable(user_ns, CAP_SYS_ADMIN))
5405 return -EPERM;
5406
5407 if (is_anon_ns(mnt_ns))
5408 return -EINVAL;
5409
5410 if (fs->users != 1)
5411 return -EINVAL;
5412
5413 get_mnt_ns(mnt_ns);
5414 old_mnt_ns = nsproxy->mnt_ns;
5415 nsproxy->mnt_ns = mnt_ns;
5416
5417 /* Find the root */
5418 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
5419 "/", LOOKUP_DOWN, &root);
5420 if (err) {
5421 /* revert to old namespace */
5422 nsproxy->mnt_ns = old_mnt_ns;
5423 put_mnt_ns(mnt_ns);
5424 return err;
5425 }
5426
5427 put_mnt_ns(old_mnt_ns);
5428
5429 /* Update the pwd and root */
5430 set_fs_pwd(fs, &root);
5431 set_fs_root(fs, &root);
5432
5433 path_put(&root);
5434 return 0;
5435}
5436
5437static struct user_namespace *mntns_owner(struct ns_common *ns)
5438{
5439 return to_mnt_ns(ns)->user_ns;
5440}
5441
5442const struct proc_ns_operations mntns_operations = {
5443 .name = "mnt",
5444 .type = CLONE_NEWNS,
5445 .get = mntns_get,
5446 .put = mntns_put,
5447 .install = mntns_install,
5448 .owner = mntns_owner,
5449};
5450
5451#ifdef CONFIG_SYSCTL
5452static struct ctl_table fs_namespace_sysctls[] = {
5453 {
5454 .procname = "mount-max",
5455 .data = &sysctl_mount_max,
5456 .maxlen = sizeof(unsigned int),
5457 .mode = 0644,
5458 .proc_handler = proc_dointvec_minmax,
5459 .extra1 = SYSCTL_ONE,
5460 },
5461};
5462
5463static int __init init_fs_namespace_sysctls(void)
5464{
5465 register_sysctl_init("fs", fs_namespace_sysctls);
5466 return 0;
5467}
5468fs_initcall(init_fs_namespace_sysctls);
5469
5470#endif /* CONFIG_SYSCTL */
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/slab.h>
13#include <linux/sched.h>
14#include <linux/spinlock.h>
15#include <linux/percpu.h>
16#include <linux/init.h>
17#include <linux/kernel.h>
18#include <linux/acct.h>
19#include <linux/capability.h>
20#include <linux/cpumask.h>
21#include <linux/module.h>
22#include <linux/sysfs.h>
23#include <linux/seq_file.h>
24#include <linux/mnt_namespace.h>
25#include <linux/namei.h>
26#include <linux/nsproxy.h>
27#include <linux/security.h>
28#include <linux/mount.h>
29#include <linux/ramfs.h>
30#include <linux/log2.h>
31#include <linux/idr.h>
32#include <linux/fs_struct.h>
33#include <linux/fsnotify.h>
34#include <asm/uaccess.h>
35#include <asm/unistd.h>
36#include "pnode.h"
37#include "internal.h"
38
39#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40#define HASH_SIZE (1UL << HASH_SHIFT)
41
42static int event;
43static DEFINE_IDA(mnt_id_ida);
44static DEFINE_IDA(mnt_group_ida);
45static DEFINE_SPINLOCK(mnt_id_lock);
46static int mnt_id_start = 0;
47static int mnt_group_start = 1;
48
49static struct list_head *mount_hashtable __read_mostly;
50static struct kmem_cache *mnt_cache __read_mostly;
51static struct rw_semaphore namespace_sem;
52
53/* /sys/fs */
54struct kobject *fs_kobj;
55EXPORT_SYMBOL_GPL(fs_kobj);
56
57/*
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
60 * up the tree.
61 *
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
64 */
65DEFINE_BRLOCK(vfsmount_lock);
66
67static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
68{
69 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
71 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
73}
74
75#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
76
77/*
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
80 */
81static int mnt_alloc_id(struct vfsmount *mnt)
82{
83 int res;
84
85retry:
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
87 spin_lock(&mnt_id_lock);
88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
89 if (!res)
90 mnt_id_start = mnt->mnt_id + 1;
91 spin_unlock(&mnt_id_lock);
92 if (res == -EAGAIN)
93 goto retry;
94
95 return res;
96}
97
98static void mnt_free_id(struct vfsmount *mnt)
99{
100 int id = mnt->mnt_id;
101 spin_lock(&mnt_id_lock);
102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
104 mnt_id_start = id;
105 spin_unlock(&mnt_id_lock);
106}
107
108/*
109 * Allocate a new peer group ID
110 *
111 * mnt_group_ida is protected by namespace_sem
112 */
113static int mnt_alloc_group_id(struct vfsmount *mnt)
114{
115 int res;
116
117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
118 return -ENOMEM;
119
120 res = ida_get_new_above(&mnt_group_ida,
121 mnt_group_start,
122 &mnt->mnt_group_id);
123 if (!res)
124 mnt_group_start = mnt->mnt_group_id + 1;
125
126 return res;
127}
128
129/*
130 * Release a peer group ID
131 */
132void mnt_release_group_id(struct vfsmount *mnt)
133{
134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
138 mnt->mnt_group_id = 0;
139}
140
141/*
142 * vfsmount lock must be held for read
143 */
144static inline void mnt_add_count(struct vfsmount *mnt, int n)
145{
146#ifdef CONFIG_SMP
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
148#else
149 preempt_disable();
150 mnt->mnt_count += n;
151 preempt_enable();
152#endif
153}
154
155static inline void mnt_set_count(struct vfsmount *mnt, int n)
156{
157#ifdef CONFIG_SMP
158 this_cpu_write(mnt->mnt_pcp->mnt_count, n);
159#else
160 mnt->mnt_count = n;
161#endif
162}
163
164/*
165 * vfsmount lock must be held for read
166 */
167static inline void mnt_inc_count(struct vfsmount *mnt)
168{
169 mnt_add_count(mnt, 1);
170}
171
172/*
173 * vfsmount lock must be held for read
174 */
175static inline void mnt_dec_count(struct vfsmount *mnt)
176{
177 mnt_add_count(mnt, -1);
178}
179
180/*
181 * vfsmount lock must be held for write
182 */
183unsigned int mnt_get_count(struct vfsmount *mnt)
184{
185#ifdef CONFIG_SMP
186 unsigned int count = 0;
187 int cpu;
188
189 for_each_possible_cpu(cpu) {
190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191 }
192
193 return count;
194#else
195 return mnt->mnt_count;
196#endif
197}
198
199static struct vfsmount *alloc_vfsmnt(const char *name)
200{
201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
202 if (mnt) {
203 int err;
204
205 err = mnt_alloc_id(mnt);
206 if (err)
207 goto out_free_cache;
208
209 if (name) {
210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211 if (!mnt->mnt_devname)
212 goto out_free_id;
213 }
214
215#ifdef CONFIG_SMP
216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 if (!mnt->mnt_pcp)
218 goto out_free_devname;
219
220 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
221#else
222 mnt->mnt_count = 1;
223 mnt->mnt_writers = 0;
224#endif
225
226 INIT_LIST_HEAD(&mnt->mnt_hash);
227 INIT_LIST_HEAD(&mnt->mnt_child);
228 INIT_LIST_HEAD(&mnt->mnt_mounts);
229 INIT_LIST_HEAD(&mnt->mnt_list);
230 INIT_LIST_HEAD(&mnt->mnt_expire);
231 INIT_LIST_HEAD(&mnt->mnt_share);
232 INIT_LIST_HEAD(&mnt->mnt_slave_list);
233 INIT_LIST_HEAD(&mnt->mnt_slave);
234#ifdef CONFIG_FSNOTIFY
235 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
236#endif
237 }
238 return mnt;
239
240#ifdef CONFIG_SMP
241out_free_devname:
242 kfree(mnt->mnt_devname);
243#endif
244out_free_id:
245 mnt_free_id(mnt);
246out_free_cache:
247 kmem_cache_free(mnt_cache, mnt);
248 return NULL;
249}
250
251/*
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
257 * a filesystem.
258 */
259/*
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
262 *
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
268 * r/w.
269 */
270int __mnt_is_readonly(struct vfsmount *mnt)
271{
272 if (mnt->mnt_flags & MNT_READONLY)
273 return 1;
274 if (mnt->mnt_sb->s_flags & MS_RDONLY)
275 return 1;
276 return 0;
277}
278EXPORT_SYMBOL_GPL(__mnt_is_readonly);
279
280static inline void mnt_inc_writers(struct vfsmount *mnt)
281{
282#ifdef CONFIG_SMP
283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
284#else
285 mnt->mnt_writers++;
286#endif
287}
288
289static inline void mnt_dec_writers(struct vfsmount *mnt)
290{
291#ifdef CONFIG_SMP
292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
293#else
294 mnt->mnt_writers--;
295#endif
296}
297
298static unsigned int mnt_get_writers(struct vfsmount *mnt)
299{
300#ifdef CONFIG_SMP
301 unsigned int count = 0;
302 int cpu;
303
304 for_each_possible_cpu(cpu) {
305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
306 }
307
308 return count;
309#else
310 return mnt->mnt_writers;
311#endif
312}
313
314/*
315 * Most r/o checks on a fs are for operations that take
316 * discrete amounts of time, like a write() or unlink().
317 * We must keep track of when those operations start
318 * (for permission checks) and when they end, so that
319 * we can determine when writes are able to occur to
320 * a filesystem.
321 */
322/**
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
325 *
326 * This tells the low-level filesystem that a write is
327 * about to be performed to it, and makes sure that
328 * writes are allowed before returning success. When
329 * the write operation is finished, mnt_drop_write()
330 * must be called. This is effectively a refcount.
331 */
332int mnt_want_write(struct vfsmount *mnt)
333{
334 int ret = 0;
335
336 preempt_disable();
337 mnt_inc_writers(mnt);
338 /*
339 * The store to mnt_inc_writers must be visible before we pass
340 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
341 * incremented count after it has set MNT_WRITE_HOLD.
342 */
343 smp_mb();
344 while (mnt->mnt_flags & MNT_WRITE_HOLD)
345 cpu_relax();
346 /*
347 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
348 * be set to match its requirements. So we must not load that until
349 * MNT_WRITE_HOLD is cleared.
350 */
351 smp_rmb();
352 if (__mnt_is_readonly(mnt)) {
353 mnt_dec_writers(mnt);
354 ret = -EROFS;
355 goto out;
356 }
357out:
358 preempt_enable();
359 return ret;
360}
361EXPORT_SYMBOL_GPL(mnt_want_write);
362
363/**
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
366 *
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
371 *
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
374 */
375int mnt_clone_write(struct vfsmount *mnt)
376{
377 /* superblock may be r/o */
378 if (__mnt_is_readonly(mnt))
379 return -EROFS;
380 preempt_disable();
381 mnt_inc_writers(mnt);
382 preempt_enable();
383 return 0;
384}
385EXPORT_SYMBOL_GPL(mnt_clone_write);
386
387/**
388 * mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
390 *
391 * This is like mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
393 */
394int mnt_want_write_file(struct file *file)
395{
396 struct inode *inode = file->f_dentry->d_inode;
397 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
398 return mnt_want_write(file->f_path.mnt);
399 else
400 return mnt_clone_write(file->f_path.mnt);
401}
402EXPORT_SYMBOL_GPL(mnt_want_write_file);
403
404/**
405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
407 *
408 * Tells the low-level filesystem that we are done
409 * performing writes to it. Must be matched with
410 * mnt_want_write() call above.
411 */
412void mnt_drop_write(struct vfsmount *mnt)
413{
414 preempt_disable();
415 mnt_dec_writers(mnt);
416 preempt_enable();
417}
418EXPORT_SYMBOL_GPL(mnt_drop_write);
419
420static int mnt_make_readonly(struct vfsmount *mnt)
421{
422 int ret = 0;
423
424 br_write_lock(vfsmount_lock);
425 mnt->mnt_flags |= MNT_WRITE_HOLD;
426 /*
427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428 * should be visible before we do.
429 */
430 smp_mb();
431
432 /*
433 * With writers on hold, if this value is zero, then there are
434 * definitely no active writers (although held writers may subsequently
435 * increment the count, they'll have to wait, and decrement it after
436 * seeing MNT_READONLY).
437 *
438 * It is OK to have counter incremented on one CPU and decremented on
439 * another: the sum will add up correctly. The danger would be when we
440 * sum up each counter, if we read a counter before it is incremented,
441 * but then read another CPU's count which it has been subsequently
442 * decremented from -- we would see more decrements than we should.
443 * MNT_WRITE_HOLD protects against this scenario, because
444 * mnt_want_write first increments count, then smp_mb, then spins on
445 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
446 * we're counting up here.
447 */
448 if (mnt_get_writers(mnt) > 0)
449 ret = -EBUSY;
450 else
451 mnt->mnt_flags |= MNT_READONLY;
452 /*
453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454 * that become unheld will see MNT_READONLY.
455 */
456 smp_wmb();
457 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
458 br_write_unlock(vfsmount_lock);
459 return ret;
460}
461
462static void __mnt_unmake_readonly(struct vfsmount *mnt)
463{
464 br_write_lock(vfsmount_lock);
465 mnt->mnt_flags &= ~MNT_READONLY;
466 br_write_unlock(vfsmount_lock);
467}
468
469static void free_vfsmnt(struct vfsmount *mnt)
470{
471 kfree(mnt->mnt_devname);
472 mnt_free_id(mnt);
473#ifdef CONFIG_SMP
474 free_percpu(mnt->mnt_pcp);
475#endif
476 kmem_cache_free(mnt_cache, mnt);
477}
478
479/*
480 * find the first or last mount at @dentry on vfsmount @mnt depending on
481 * @dir. If @dir is set return the first mount else return the last mount.
482 * vfsmount_lock must be held for read or write.
483 */
484struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
485 int dir)
486{
487 struct list_head *head = mount_hashtable + hash(mnt, dentry);
488 struct list_head *tmp = head;
489 struct vfsmount *p, *found = NULL;
490
491 for (;;) {
492 tmp = dir ? tmp->next : tmp->prev;
493 p = NULL;
494 if (tmp == head)
495 break;
496 p = list_entry(tmp, struct vfsmount, mnt_hash);
497 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
498 found = p;
499 break;
500 }
501 }
502 return found;
503}
504
505/*
506 * lookup_mnt increments the ref count before returning
507 * the vfsmount struct.
508 */
509struct vfsmount *lookup_mnt(struct path *path)
510{
511 struct vfsmount *child_mnt;
512
513 br_read_lock(vfsmount_lock);
514 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
515 mntget(child_mnt);
516 br_read_unlock(vfsmount_lock);
517 return child_mnt;
518}
519
520static inline int check_mnt(struct vfsmount *mnt)
521{
522 return mnt->mnt_ns == current->nsproxy->mnt_ns;
523}
524
525/*
526 * vfsmount lock must be held for write
527 */
528static void touch_mnt_namespace(struct mnt_namespace *ns)
529{
530 if (ns) {
531 ns->event = ++event;
532 wake_up_interruptible(&ns->poll);
533 }
534}
535
536/*
537 * vfsmount lock must be held for write
538 */
539static void __touch_mnt_namespace(struct mnt_namespace *ns)
540{
541 if (ns && ns->event != event) {
542 ns->event = event;
543 wake_up_interruptible(&ns->poll);
544 }
545}
546
547/*
548 * Clear dentry's mounted state if it has no remaining mounts.
549 * vfsmount_lock must be held for write.
550 */
551static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
552{
553 unsigned u;
554
555 for (u = 0; u < HASH_SIZE; u++) {
556 struct vfsmount *p;
557
558 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
559 if (p->mnt_mountpoint == dentry)
560 return;
561 }
562 }
563 spin_lock(&dentry->d_lock);
564 dentry->d_flags &= ~DCACHE_MOUNTED;
565 spin_unlock(&dentry->d_lock);
566}
567
568/*
569 * vfsmount lock must be held for write
570 */
571static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
572{
573 old_path->dentry = mnt->mnt_mountpoint;
574 old_path->mnt = mnt->mnt_parent;
575 mnt->mnt_parent = mnt;
576 mnt->mnt_mountpoint = mnt->mnt_root;
577 list_del_init(&mnt->mnt_child);
578 list_del_init(&mnt->mnt_hash);
579 dentry_reset_mounted(old_path->mnt, old_path->dentry);
580}
581
582/*
583 * vfsmount lock must be held for write
584 */
585void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
586 struct vfsmount *child_mnt)
587{
588 child_mnt->mnt_parent = mntget(mnt);
589 child_mnt->mnt_mountpoint = dget(dentry);
590 spin_lock(&dentry->d_lock);
591 dentry->d_flags |= DCACHE_MOUNTED;
592 spin_unlock(&dentry->d_lock);
593}
594
595/*
596 * vfsmount lock must be held for write
597 */
598static void attach_mnt(struct vfsmount *mnt, struct path *path)
599{
600 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
601 list_add_tail(&mnt->mnt_hash, mount_hashtable +
602 hash(path->mnt, path->dentry));
603 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
604}
605
606static inline void __mnt_make_longterm(struct vfsmount *mnt)
607{
608#ifdef CONFIG_SMP
609 atomic_inc(&mnt->mnt_longterm);
610#endif
611}
612
613/* needs vfsmount lock for write */
614static inline void __mnt_make_shortterm(struct vfsmount *mnt)
615{
616#ifdef CONFIG_SMP
617 atomic_dec(&mnt->mnt_longterm);
618#endif
619}
620
621/*
622 * vfsmount lock must be held for write
623 */
624static void commit_tree(struct vfsmount *mnt)
625{
626 struct vfsmount *parent = mnt->mnt_parent;
627 struct vfsmount *m;
628 LIST_HEAD(head);
629 struct mnt_namespace *n = parent->mnt_ns;
630
631 BUG_ON(parent == mnt);
632
633 list_add_tail(&head, &mnt->mnt_list);
634 list_for_each_entry(m, &head, mnt_list) {
635 m->mnt_ns = n;
636 __mnt_make_longterm(m);
637 }
638
639 list_splice(&head, n->list.prev);
640
641 list_add_tail(&mnt->mnt_hash, mount_hashtable +
642 hash(parent, mnt->mnt_mountpoint));
643 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
644 touch_mnt_namespace(n);
645}
646
647static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
648{
649 struct list_head *next = p->mnt_mounts.next;
650 if (next == &p->mnt_mounts) {
651 while (1) {
652 if (p == root)
653 return NULL;
654 next = p->mnt_child.next;
655 if (next != &p->mnt_parent->mnt_mounts)
656 break;
657 p = p->mnt_parent;
658 }
659 }
660 return list_entry(next, struct vfsmount, mnt_child);
661}
662
663static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
664{
665 struct list_head *prev = p->mnt_mounts.prev;
666 while (prev != &p->mnt_mounts) {
667 p = list_entry(prev, struct vfsmount, mnt_child);
668 prev = p->mnt_mounts.prev;
669 }
670 return p;
671}
672
673struct vfsmount *
674vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
675{
676 struct vfsmount *mnt;
677 struct dentry *root;
678
679 if (!type)
680 return ERR_PTR(-ENODEV);
681
682 mnt = alloc_vfsmnt(name);
683 if (!mnt)
684 return ERR_PTR(-ENOMEM);
685
686 if (flags & MS_KERNMOUNT)
687 mnt->mnt_flags = MNT_INTERNAL;
688
689 root = mount_fs(type, flags, name, data);
690 if (IS_ERR(root)) {
691 free_vfsmnt(mnt);
692 return ERR_CAST(root);
693 }
694
695 mnt->mnt_root = root;
696 mnt->mnt_sb = root->d_sb;
697 mnt->mnt_mountpoint = mnt->mnt_root;
698 mnt->mnt_parent = mnt;
699 return mnt;
700}
701EXPORT_SYMBOL_GPL(vfs_kern_mount);
702
703static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
704 int flag)
705{
706 struct super_block *sb = old->mnt_sb;
707 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
708
709 if (mnt) {
710 if (flag & (CL_SLAVE | CL_PRIVATE))
711 mnt->mnt_group_id = 0; /* not a peer of original */
712 else
713 mnt->mnt_group_id = old->mnt_group_id;
714
715 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
716 int err = mnt_alloc_group_id(mnt);
717 if (err)
718 goto out_free;
719 }
720
721 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
722 atomic_inc(&sb->s_active);
723 mnt->mnt_sb = sb;
724 mnt->mnt_root = dget(root);
725 mnt->mnt_mountpoint = mnt->mnt_root;
726 mnt->mnt_parent = mnt;
727
728 if (flag & CL_SLAVE) {
729 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
730 mnt->mnt_master = old;
731 CLEAR_MNT_SHARED(mnt);
732 } else if (!(flag & CL_PRIVATE)) {
733 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
734 list_add(&mnt->mnt_share, &old->mnt_share);
735 if (IS_MNT_SLAVE(old))
736 list_add(&mnt->mnt_slave, &old->mnt_slave);
737 mnt->mnt_master = old->mnt_master;
738 }
739 if (flag & CL_MAKE_SHARED)
740 set_mnt_shared(mnt);
741
742 /* stick the duplicate mount on the same expiry list
743 * as the original if that was on one */
744 if (flag & CL_EXPIRE) {
745 if (!list_empty(&old->mnt_expire))
746 list_add(&mnt->mnt_expire, &old->mnt_expire);
747 }
748 }
749 return mnt;
750
751 out_free:
752 free_vfsmnt(mnt);
753 return NULL;
754}
755
756static inline void mntfree(struct vfsmount *mnt)
757{
758 struct super_block *sb = mnt->mnt_sb;
759
760 /*
761 * This probably indicates that somebody messed
762 * up a mnt_want/drop_write() pair. If this
763 * happens, the filesystem was probably unable
764 * to make r/w->r/o transitions.
765 */
766 /*
767 * The locking used to deal with mnt_count decrement provides barriers,
768 * so mnt_get_writers() below is safe.
769 */
770 WARN_ON(mnt_get_writers(mnt));
771 fsnotify_vfsmount_delete(mnt);
772 dput(mnt->mnt_root);
773 free_vfsmnt(mnt);
774 deactivate_super(sb);
775}
776
777static void mntput_no_expire(struct vfsmount *mnt)
778{
779put_again:
780#ifdef CONFIG_SMP
781 br_read_lock(vfsmount_lock);
782 if (likely(atomic_read(&mnt->mnt_longterm))) {
783 mnt_dec_count(mnt);
784 br_read_unlock(vfsmount_lock);
785 return;
786 }
787 br_read_unlock(vfsmount_lock);
788
789 br_write_lock(vfsmount_lock);
790 mnt_dec_count(mnt);
791 if (mnt_get_count(mnt)) {
792 br_write_unlock(vfsmount_lock);
793 return;
794 }
795#else
796 mnt_dec_count(mnt);
797 if (likely(mnt_get_count(mnt)))
798 return;
799 br_write_lock(vfsmount_lock);
800#endif
801 if (unlikely(mnt->mnt_pinned)) {
802 mnt_add_count(mnt, mnt->mnt_pinned + 1);
803 mnt->mnt_pinned = 0;
804 br_write_unlock(vfsmount_lock);
805 acct_auto_close_mnt(mnt);
806 goto put_again;
807 }
808 br_write_unlock(vfsmount_lock);
809 mntfree(mnt);
810}
811
812void mntput(struct vfsmount *mnt)
813{
814 if (mnt) {
815 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
816 if (unlikely(mnt->mnt_expiry_mark))
817 mnt->mnt_expiry_mark = 0;
818 mntput_no_expire(mnt);
819 }
820}
821EXPORT_SYMBOL(mntput);
822
823struct vfsmount *mntget(struct vfsmount *mnt)
824{
825 if (mnt)
826 mnt_inc_count(mnt);
827 return mnt;
828}
829EXPORT_SYMBOL(mntget);
830
831void mnt_pin(struct vfsmount *mnt)
832{
833 br_write_lock(vfsmount_lock);
834 mnt->mnt_pinned++;
835 br_write_unlock(vfsmount_lock);
836}
837EXPORT_SYMBOL(mnt_pin);
838
839void mnt_unpin(struct vfsmount *mnt)
840{
841 br_write_lock(vfsmount_lock);
842 if (mnt->mnt_pinned) {
843 mnt_inc_count(mnt);
844 mnt->mnt_pinned--;
845 }
846 br_write_unlock(vfsmount_lock);
847}
848EXPORT_SYMBOL(mnt_unpin);
849
850static inline void mangle(struct seq_file *m, const char *s)
851{
852 seq_escape(m, s, " \t\n\\");
853}
854
855/*
856 * Simple .show_options callback for filesystems which don't want to
857 * implement more complex mount option showing.
858 *
859 * See also save_mount_options().
860 */
861int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
862{
863 const char *options;
864
865 rcu_read_lock();
866 options = rcu_dereference(mnt->mnt_sb->s_options);
867
868 if (options != NULL && options[0]) {
869 seq_putc(m, ',');
870 mangle(m, options);
871 }
872 rcu_read_unlock();
873
874 return 0;
875}
876EXPORT_SYMBOL(generic_show_options);
877
878/*
879 * If filesystem uses generic_show_options(), this function should be
880 * called from the fill_super() callback.
881 *
882 * The .remount_fs callback usually needs to be handled in a special
883 * way, to make sure, that previous options are not overwritten if the
884 * remount fails.
885 *
886 * Also note, that if the filesystem's .remount_fs function doesn't
887 * reset all options to their default value, but changes only newly
888 * given options, then the displayed options will not reflect reality
889 * any more.
890 */
891void save_mount_options(struct super_block *sb, char *options)
892{
893 BUG_ON(sb->s_options);
894 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
895}
896EXPORT_SYMBOL(save_mount_options);
897
898void replace_mount_options(struct super_block *sb, char *options)
899{
900 char *old = sb->s_options;
901 rcu_assign_pointer(sb->s_options, options);
902 if (old) {
903 synchronize_rcu();
904 kfree(old);
905 }
906}
907EXPORT_SYMBOL(replace_mount_options);
908
909#ifdef CONFIG_PROC_FS
910/* iterator */
911static void *m_start(struct seq_file *m, loff_t *pos)
912{
913 struct proc_mounts *p = m->private;
914
915 down_read(&namespace_sem);
916 return seq_list_start(&p->ns->list, *pos);
917}
918
919static void *m_next(struct seq_file *m, void *v, loff_t *pos)
920{
921 struct proc_mounts *p = m->private;
922
923 return seq_list_next(v, &p->ns->list, pos);
924}
925
926static void m_stop(struct seq_file *m, void *v)
927{
928 up_read(&namespace_sem);
929}
930
931int mnt_had_events(struct proc_mounts *p)
932{
933 struct mnt_namespace *ns = p->ns;
934 int res = 0;
935
936 br_read_lock(vfsmount_lock);
937 if (p->m.poll_event != ns->event) {
938 p->m.poll_event = ns->event;
939 res = 1;
940 }
941 br_read_unlock(vfsmount_lock);
942
943 return res;
944}
945
946struct proc_fs_info {
947 int flag;
948 const char *str;
949};
950
951static int show_sb_opts(struct seq_file *m, struct super_block *sb)
952{
953 static const struct proc_fs_info fs_info[] = {
954 { MS_SYNCHRONOUS, ",sync" },
955 { MS_DIRSYNC, ",dirsync" },
956 { MS_MANDLOCK, ",mand" },
957 { 0, NULL }
958 };
959 const struct proc_fs_info *fs_infop;
960
961 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
962 if (sb->s_flags & fs_infop->flag)
963 seq_puts(m, fs_infop->str);
964 }
965
966 return security_sb_show_options(m, sb);
967}
968
969static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
970{
971 static const struct proc_fs_info mnt_info[] = {
972 { MNT_NOSUID, ",nosuid" },
973 { MNT_NODEV, ",nodev" },
974 { MNT_NOEXEC, ",noexec" },
975 { MNT_NOATIME, ",noatime" },
976 { MNT_NODIRATIME, ",nodiratime" },
977 { MNT_RELATIME, ",relatime" },
978 { 0, NULL }
979 };
980 const struct proc_fs_info *fs_infop;
981
982 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
983 if (mnt->mnt_flags & fs_infop->flag)
984 seq_puts(m, fs_infop->str);
985 }
986}
987
988static void show_type(struct seq_file *m, struct super_block *sb)
989{
990 mangle(m, sb->s_type->name);
991 if (sb->s_subtype && sb->s_subtype[0]) {
992 seq_putc(m, '.');
993 mangle(m, sb->s_subtype);
994 }
995}
996
997static int show_vfsmnt(struct seq_file *m, void *v)
998{
999 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1000 int err = 0;
1001 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1002
1003 if (mnt->mnt_sb->s_op->show_devname) {
1004 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1005 if (err)
1006 goto out;
1007 } else {
1008 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1009 }
1010 seq_putc(m, ' ');
1011 seq_path(m, &mnt_path, " \t\n\\");
1012 seq_putc(m, ' ');
1013 show_type(m, mnt->mnt_sb);
1014 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1015 err = show_sb_opts(m, mnt->mnt_sb);
1016 if (err)
1017 goto out;
1018 show_mnt_opts(m, mnt);
1019 if (mnt->mnt_sb->s_op->show_options)
1020 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1021 seq_puts(m, " 0 0\n");
1022out:
1023 return err;
1024}
1025
1026const struct seq_operations mounts_op = {
1027 .start = m_start,
1028 .next = m_next,
1029 .stop = m_stop,
1030 .show = show_vfsmnt
1031};
1032
1033static int show_mountinfo(struct seq_file *m, void *v)
1034{
1035 struct proc_mounts *p = m->private;
1036 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1037 struct super_block *sb = mnt->mnt_sb;
1038 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1039 struct path root = p->root;
1040 int err = 0;
1041
1042 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1043 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1044 if (sb->s_op->show_path)
1045 err = sb->s_op->show_path(m, mnt);
1046 else
1047 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1048 if (err)
1049 goto out;
1050 seq_putc(m, ' ');
1051 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1052 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1053 /*
1054 * Mountpoint is outside root, discard that one. Ugly,
1055 * but less so than trying to do that in iterator in a
1056 * race-free way (due to renames).
1057 */
1058 return SEQ_SKIP;
1059 }
1060 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1061 show_mnt_opts(m, mnt);
1062
1063 /* Tagged fields ("foo:X" or "bar") */
1064 if (IS_MNT_SHARED(mnt))
1065 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1066 if (IS_MNT_SLAVE(mnt)) {
1067 int master = mnt->mnt_master->mnt_group_id;
1068 int dom = get_dominating_id(mnt, &p->root);
1069 seq_printf(m, " master:%i", master);
1070 if (dom && dom != master)
1071 seq_printf(m, " propagate_from:%i", dom);
1072 }
1073 if (IS_MNT_UNBINDABLE(mnt))
1074 seq_puts(m, " unbindable");
1075
1076 /* Filesystem specific data */
1077 seq_puts(m, " - ");
1078 show_type(m, sb);
1079 seq_putc(m, ' ');
1080 if (sb->s_op->show_devname)
1081 err = sb->s_op->show_devname(m, mnt);
1082 else
1083 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084 if (err)
1085 goto out;
1086 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1087 err = show_sb_opts(m, sb);
1088 if (err)
1089 goto out;
1090 if (sb->s_op->show_options)
1091 err = sb->s_op->show_options(m, mnt);
1092 seq_putc(m, '\n');
1093out:
1094 return err;
1095}
1096
1097const struct seq_operations mountinfo_op = {
1098 .start = m_start,
1099 .next = m_next,
1100 .stop = m_stop,
1101 .show = show_mountinfo,
1102};
1103
1104static int show_vfsstat(struct seq_file *m, void *v)
1105{
1106 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1107 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1108 int err = 0;
1109
1110 /* device */
1111 if (mnt->mnt_sb->s_op->show_devname) {
1112 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1113 } else {
1114 if (mnt->mnt_devname) {
1115 seq_puts(m, "device ");
1116 mangle(m, mnt->mnt_devname);
1117 } else
1118 seq_puts(m, "no device");
1119 }
1120
1121 /* mount point */
1122 seq_puts(m, " mounted on ");
1123 seq_path(m, &mnt_path, " \t\n\\");
1124 seq_putc(m, ' ');
1125
1126 /* file system type */
1127 seq_puts(m, "with fstype ");
1128 show_type(m, mnt->mnt_sb);
1129
1130 /* optional statistics */
1131 if (mnt->mnt_sb->s_op->show_stats) {
1132 seq_putc(m, ' ');
1133 if (!err)
1134 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1135 }
1136
1137 seq_putc(m, '\n');
1138 return err;
1139}
1140
1141const struct seq_operations mountstats_op = {
1142 .start = m_start,
1143 .next = m_next,
1144 .stop = m_stop,
1145 .show = show_vfsstat,
1146};
1147#endif /* CONFIG_PROC_FS */
1148
1149/**
1150 * may_umount_tree - check if a mount tree is busy
1151 * @mnt: root of mount tree
1152 *
1153 * This is called to check if a tree of mounts has any
1154 * open files, pwds, chroots or sub mounts that are
1155 * busy.
1156 */
1157int may_umount_tree(struct vfsmount *mnt)
1158{
1159 int actual_refs = 0;
1160 int minimum_refs = 0;
1161 struct vfsmount *p;
1162
1163 /* write lock needed for mnt_get_count */
1164 br_write_lock(vfsmount_lock);
1165 for (p = mnt; p; p = next_mnt(p, mnt)) {
1166 actual_refs += mnt_get_count(p);
1167 minimum_refs += 2;
1168 }
1169 br_write_unlock(vfsmount_lock);
1170
1171 if (actual_refs > minimum_refs)
1172 return 0;
1173
1174 return 1;
1175}
1176
1177EXPORT_SYMBOL(may_umount_tree);
1178
1179/**
1180 * may_umount - check if a mount point is busy
1181 * @mnt: root of mount
1182 *
1183 * This is called to check if a mount point has any
1184 * open files, pwds, chroots or sub mounts. If the
1185 * mount has sub mounts this will return busy
1186 * regardless of whether the sub mounts are busy.
1187 *
1188 * Doesn't take quota and stuff into account. IOW, in some cases it will
1189 * give false negatives. The main reason why it's here is that we need
1190 * a non-destructive way to look for easily umountable filesystems.
1191 */
1192int may_umount(struct vfsmount *mnt)
1193{
1194 int ret = 1;
1195 down_read(&namespace_sem);
1196 br_write_lock(vfsmount_lock);
1197 if (propagate_mount_busy(mnt, 2))
1198 ret = 0;
1199 br_write_unlock(vfsmount_lock);
1200 up_read(&namespace_sem);
1201 return ret;
1202}
1203
1204EXPORT_SYMBOL(may_umount);
1205
1206void release_mounts(struct list_head *head)
1207{
1208 struct vfsmount *mnt;
1209 while (!list_empty(head)) {
1210 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1211 list_del_init(&mnt->mnt_hash);
1212 if (mnt->mnt_parent != mnt) {
1213 struct dentry *dentry;
1214 struct vfsmount *m;
1215
1216 br_write_lock(vfsmount_lock);
1217 dentry = mnt->mnt_mountpoint;
1218 m = mnt->mnt_parent;
1219 mnt->mnt_mountpoint = mnt->mnt_root;
1220 mnt->mnt_parent = mnt;
1221 m->mnt_ghosts--;
1222 br_write_unlock(vfsmount_lock);
1223 dput(dentry);
1224 mntput(m);
1225 }
1226 mntput(mnt);
1227 }
1228}
1229
1230/*
1231 * vfsmount lock must be held for write
1232 * namespace_sem must be held for write
1233 */
1234void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1235{
1236 LIST_HEAD(tmp_list);
1237 struct vfsmount *p;
1238
1239 for (p = mnt; p; p = next_mnt(p, mnt))
1240 list_move(&p->mnt_hash, &tmp_list);
1241
1242 if (propagate)
1243 propagate_umount(&tmp_list);
1244
1245 list_for_each_entry(p, &tmp_list, mnt_hash) {
1246 list_del_init(&p->mnt_expire);
1247 list_del_init(&p->mnt_list);
1248 __touch_mnt_namespace(p->mnt_ns);
1249 p->mnt_ns = NULL;
1250 __mnt_make_shortterm(p);
1251 list_del_init(&p->mnt_child);
1252 if (p->mnt_parent != p) {
1253 p->mnt_parent->mnt_ghosts++;
1254 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1255 }
1256 change_mnt_propagation(p, MS_PRIVATE);
1257 }
1258 list_splice(&tmp_list, kill);
1259}
1260
1261static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1262
1263static int do_umount(struct vfsmount *mnt, int flags)
1264{
1265 struct super_block *sb = mnt->mnt_sb;
1266 int retval;
1267 LIST_HEAD(umount_list);
1268
1269 retval = security_sb_umount(mnt, flags);
1270 if (retval)
1271 return retval;
1272
1273 /*
1274 * Allow userspace to request a mountpoint be expired rather than
1275 * unmounting unconditionally. Unmount only happens if:
1276 * (1) the mark is already set (the mark is cleared by mntput())
1277 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1278 */
1279 if (flags & MNT_EXPIRE) {
1280 if (mnt == current->fs->root.mnt ||
1281 flags & (MNT_FORCE | MNT_DETACH))
1282 return -EINVAL;
1283
1284 /*
1285 * probably don't strictly need the lock here if we examined
1286 * all race cases, but it's a slowpath.
1287 */
1288 br_write_lock(vfsmount_lock);
1289 if (mnt_get_count(mnt) != 2) {
1290 br_write_unlock(vfsmount_lock);
1291 return -EBUSY;
1292 }
1293 br_write_unlock(vfsmount_lock);
1294
1295 if (!xchg(&mnt->mnt_expiry_mark, 1))
1296 return -EAGAIN;
1297 }
1298
1299 /*
1300 * If we may have to abort operations to get out of this
1301 * mount, and they will themselves hold resources we must
1302 * allow the fs to do things. In the Unix tradition of
1303 * 'Gee thats tricky lets do it in userspace' the umount_begin
1304 * might fail to complete on the first run through as other tasks
1305 * must return, and the like. Thats for the mount program to worry
1306 * about for the moment.
1307 */
1308
1309 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1310 sb->s_op->umount_begin(sb);
1311 }
1312
1313 /*
1314 * No sense to grab the lock for this test, but test itself looks
1315 * somewhat bogus. Suggestions for better replacement?
1316 * Ho-hum... In principle, we might treat that as umount + switch
1317 * to rootfs. GC would eventually take care of the old vfsmount.
1318 * Actually it makes sense, especially if rootfs would contain a
1319 * /reboot - static binary that would close all descriptors and
1320 * call reboot(9). Then init(8) could umount root and exec /reboot.
1321 */
1322 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1323 /*
1324 * Special case for "unmounting" root ...
1325 * we just try to remount it readonly.
1326 */
1327 down_write(&sb->s_umount);
1328 if (!(sb->s_flags & MS_RDONLY))
1329 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1330 up_write(&sb->s_umount);
1331 return retval;
1332 }
1333
1334 down_write(&namespace_sem);
1335 br_write_lock(vfsmount_lock);
1336 event++;
1337
1338 if (!(flags & MNT_DETACH))
1339 shrink_submounts(mnt, &umount_list);
1340
1341 retval = -EBUSY;
1342 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1343 if (!list_empty(&mnt->mnt_list))
1344 umount_tree(mnt, 1, &umount_list);
1345 retval = 0;
1346 }
1347 br_write_unlock(vfsmount_lock);
1348 up_write(&namespace_sem);
1349 release_mounts(&umount_list);
1350 return retval;
1351}
1352
1353/*
1354 * Now umount can handle mount points as well as block devices.
1355 * This is important for filesystems which use unnamed block devices.
1356 *
1357 * We now support a flag for forced unmount like the other 'big iron'
1358 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1359 */
1360
1361SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1362{
1363 struct path path;
1364 int retval;
1365 int lookup_flags = 0;
1366
1367 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1368 return -EINVAL;
1369
1370 if (!(flags & UMOUNT_NOFOLLOW))
1371 lookup_flags |= LOOKUP_FOLLOW;
1372
1373 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1374 if (retval)
1375 goto out;
1376 retval = -EINVAL;
1377 if (path.dentry != path.mnt->mnt_root)
1378 goto dput_and_out;
1379 if (!check_mnt(path.mnt))
1380 goto dput_and_out;
1381
1382 retval = -EPERM;
1383 if (!capable(CAP_SYS_ADMIN))
1384 goto dput_and_out;
1385
1386 retval = do_umount(path.mnt, flags);
1387dput_and_out:
1388 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1389 dput(path.dentry);
1390 mntput_no_expire(path.mnt);
1391out:
1392 return retval;
1393}
1394
1395#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1396
1397/*
1398 * The 2.0 compatible umount. No flags.
1399 */
1400SYSCALL_DEFINE1(oldumount, char __user *, name)
1401{
1402 return sys_umount(name, 0);
1403}
1404
1405#endif
1406
1407static int mount_is_safe(struct path *path)
1408{
1409 if (capable(CAP_SYS_ADMIN))
1410 return 0;
1411 return -EPERM;
1412#ifdef notyet
1413 if (S_ISLNK(path->dentry->d_inode->i_mode))
1414 return -EPERM;
1415 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1416 if (current_uid() != path->dentry->d_inode->i_uid)
1417 return -EPERM;
1418 }
1419 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1420 return -EPERM;
1421 return 0;
1422#endif
1423}
1424
1425struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1426 int flag)
1427{
1428 struct vfsmount *res, *p, *q, *r, *s;
1429 struct path path;
1430
1431 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1432 return NULL;
1433
1434 res = q = clone_mnt(mnt, dentry, flag);
1435 if (!q)
1436 goto Enomem;
1437 q->mnt_mountpoint = mnt->mnt_mountpoint;
1438
1439 p = mnt;
1440 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1441 if (!is_subdir(r->mnt_mountpoint, dentry))
1442 continue;
1443
1444 for (s = r; s; s = next_mnt(s, r)) {
1445 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1446 s = skip_mnt_tree(s);
1447 continue;
1448 }
1449 while (p != s->mnt_parent) {
1450 p = p->mnt_parent;
1451 q = q->mnt_parent;
1452 }
1453 p = s;
1454 path.mnt = q;
1455 path.dentry = p->mnt_mountpoint;
1456 q = clone_mnt(p, p->mnt_root, flag);
1457 if (!q)
1458 goto Enomem;
1459 br_write_lock(vfsmount_lock);
1460 list_add_tail(&q->mnt_list, &res->mnt_list);
1461 attach_mnt(q, &path);
1462 br_write_unlock(vfsmount_lock);
1463 }
1464 }
1465 return res;
1466Enomem:
1467 if (res) {
1468 LIST_HEAD(umount_list);
1469 br_write_lock(vfsmount_lock);
1470 umount_tree(res, 0, &umount_list);
1471 br_write_unlock(vfsmount_lock);
1472 release_mounts(&umount_list);
1473 }
1474 return NULL;
1475}
1476
1477struct vfsmount *collect_mounts(struct path *path)
1478{
1479 struct vfsmount *tree;
1480 down_write(&namespace_sem);
1481 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1482 up_write(&namespace_sem);
1483 return tree;
1484}
1485
1486void drop_collected_mounts(struct vfsmount *mnt)
1487{
1488 LIST_HEAD(umount_list);
1489 down_write(&namespace_sem);
1490 br_write_lock(vfsmount_lock);
1491 umount_tree(mnt, 0, &umount_list);
1492 br_write_unlock(vfsmount_lock);
1493 up_write(&namespace_sem);
1494 release_mounts(&umount_list);
1495}
1496
1497int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1498 struct vfsmount *root)
1499{
1500 struct vfsmount *mnt;
1501 int res = f(root, arg);
1502 if (res)
1503 return res;
1504 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1505 res = f(mnt, arg);
1506 if (res)
1507 return res;
1508 }
1509 return 0;
1510}
1511
1512static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1513{
1514 struct vfsmount *p;
1515
1516 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1517 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1518 mnt_release_group_id(p);
1519 }
1520}
1521
1522static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1523{
1524 struct vfsmount *p;
1525
1526 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1527 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1528 int err = mnt_alloc_group_id(p);
1529 if (err) {
1530 cleanup_group_ids(mnt, p);
1531 return err;
1532 }
1533 }
1534 }
1535
1536 return 0;
1537}
1538
1539/*
1540 * @source_mnt : mount tree to be attached
1541 * @nd : place the mount tree @source_mnt is attached
1542 * @parent_nd : if non-null, detach the source_mnt from its parent and
1543 * store the parent mount and mountpoint dentry.
1544 * (done when source_mnt is moved)
1545 *
1546 * NOTE: in the table below explains the semantics when a source mount
1547 * of a given type is attached to a destination mount of a given type.
1548 * ---------------------------------------------------------------------------
1549 * | BIND MOUNT OPERATION |
1550 * |**************************************************************************
1551 * | source-->| shared | private | slave | unbindable |
1552 * | dest | | | | |
1553 * | | | | | | |
1554 * | v | | | | |
1555 * |**************************************************************************
1556 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1557 * | | | | | |
1558 * |non-shared| shared (+) | private | slave (*) | invalid |
1559 * ***************************************************************************
1560 * A bind operation clones the source mount and mounts the clone on the
1561 * destination mount.
1562 *
1563 * (++) the cloned mount is propagated to all the mounts in the propagation
1564 * tree of the destination mount and the cloned mount is added to
1565 * the peer group of the source mount.
1566 * (+) the cloned mount is created under the destination mount and is marked
1567 * as shared. The cloned mount is added to the peer group of the source
1568 * mount.
1569 * (+++) the mount is propagated to all the mounts in the propagation tree
1570 * of the destination mount and the cloned mount is made slave
1571 * of the same master as that of the source mount. The cloned mount
1572 * is marked as 'shared and slave'.
1573 * (*) the cloned mount is made a slave of the same master as that of the
1574 * source mount.
1575 *
1576 * ---------------------------------------------------------------------------
1577 * | MOVE MOUNT OPERATION |
1578 * |**************************************************************************
1579 * | source-->| shared | private | slave | unbindable |
1580 * | dest | | | | |
1581 * | | | | | | |
1582 * | v | | | | |
1583 * |**************************************************************************
1584 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1585 * | | | | | |
1586 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1587 * ***************************************************************************
1588 *
1589 * (+) the mount is moved to the destination. And is then propagated to
1590 * all the mounts in the propagation tree of the destination mount.
1591 * (+*) the mount is moved to the destination.
1592 * (+++) the mount is moved to the destination and is then propagated to
1593 * all the mounts belonging to the destination mount's propagation tree.
1594 * the mount is marked as 'shared and slave'.
1595 * (*) the mount continues to be a slave at the new location.
1596 *
1597 * if the source mount is a tree, the operations explained above is
1598 * applied to each mount in the tree.
1599 * Must be called without spinlocks held, since this function can sleep
1600 * in allocations.
1601 */
1602static int attach_recursive_mnt(struct vfsmount *source_mnt,
1603 struct path *path, struct path *parent_path)
1604{
1605 LIST_HEAD(tree_list);
1606 struct vfsmount *dest_mnt = path->mnt;
1607 struct dentry *dest_dentry = path->dentry;
1608 struct vfsmount *child, *p;
1609 int err;
1610
1611 if (IS_MNT_SHARED(dest_mnt)) {
1612 err = invent_group_ids(source_mnt, true);
1613 if (err)
1614 goto out;
1615 }
1616 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1617 if (err)
1618 goto out_cleanup_ids;
1619
1620 br_write_lock(vfsmount_lock);
1621
1622 if (IS_MNT_SHARED(dest_mnt)) {
1623 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1624 set_mnt_shared(p);
1625 }
1626 if (parent_path) {
1627 detach_mnt(source_mnt, parent_path);
1628 attach_mnt(source_mnt, path);
1629 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1630 } else {
1631 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1632 commit_tree(source_mnt);
1633 }
1634
1635 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1636 list_del_init(&child->mnt_hash);
1637 commit_tree(child);
1638 }
1639 br_write_unlock(vfsmount_lock);
1640
1641 return 0;
1642
1643 out_cleanup_ids:
1644 if (IS_MNT_SHARED(dest_mnt))
1645 cleanup_group_ids(source_mnt, NULL);
1646 out:
1647 return err;
1648}
1649
1650static int lock_mount(struct path *path)
1651{
1652 struct vfsmount *mnt;
1653retry:
1654 mutex_lock(&path->dentry->d_inode->i_mutex);
1655 if (unlikely(cant_mount(path->dentry))) {
1656 mutex_unlock(&path->dentry->d_inode->i_mutex);
1657 return -ENOENT;
1658 }
1659 down_write(&namespace_sem);
1660 mnt = lookup_mnt(path);
1661 if (likely(!mnt))
1662 return 0;
1663 up_write(&namespace_sem);
1664 mutex_unlock(&path->dentry->d_inode->i_mutex);
1665 path_put(path);
1666 path->mnt = mnt;
1667 path->dentry = dget(mnt->mnt_root);
1668 goto retry;
1669}
1670
1671static void unlock_mount(struct path *path)
1672{
1673 up_write(&namespace_sem);
1674 mutex_unlock(&path->dentry->d_inode->i_mutex);
1675}
1676
1677static int graft_tree(struct vfsmount *mnt, struct path *path)
1678{
1679 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1680 return -EINVAL;
1681
1682 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1683 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1684 return -ENOTDIR;
1685
1686 if (d_unlinked(path->dentry))
1687 return -ENOENT;
1688
1689 return attach_recursive_mnt(mnt, path, NULL);
1690}
1691
1692/*
1693 * Sanity check the flags to change_mnt_propagation.
1694 */
1695
1696static int flags_to_propagation_type(int flags)
1697{
1698 int type = flags & ~(MS_REC | MS_SILENT);
1699
1700 /* Fail if any non-propagation flags are set */
1701 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1702 return 0;
1703 /* Only one propagation flag should be set */
1704 if (!is_power_of_2(type))
1705 return 0;
1706 return type;
1707}
1708
1709/*
1710 * recursively change the type of the mountpoint.
1711 */
1712static int do_change_type(struct path *path, int flag)
1713{
1714 struct vfsmount *m, *mnt = path->mnt;
1715 int recurse = flag & MS_REC;
1716 int type;
1717 int err = 0;
1718
1719 if (!capable(CAP_SYS_ADMIN))
1720 return -EPERM;
1721
1722 if (path->dentry != path->mnt->mnt_root)
1723 return -EINVAL;
1724
1725 type = flags_to_propagation_type(flag);
1726 if (!type)
1727 return -EINVAL;
1728
1729 down_write(&namespace_sem);
1730 if (type == MS_SHARED) {
1731 err = invent_group_ids(mnt, recurse);
1732 if (err)
1733 goto out_unlock;
1734 }
1735
1736 br_write_lock(vfsmount_lock);
1737 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1738 change_mnt_propagation(m, type);
1739 br_write_unlock(vfsmount_lock);
1740
1741 out_unlock:
1742 up_write(&namespace_sem);
1743 return err;
1744}
1745
1746/*
1747 * do loopback mount.
1748 */
1749static int do_loopback(struct path *path, char *old_name,
1750 int recurse)
1751{
1752 LIST_HEAD(umount_list);
1753 struct path old_path;
1754 struct vfsmount *mnt = NULL;
1755 int err = mount_is_safe(path);
1756 if (err)
1757 return err;
1758 if (!old_name || !*old_name)
1759 return -EINVAL;
1760 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1761 if (err)
1762 return err;
1763
1764 err = lock_mount(path);
1765 if (err)
1766 goto out;
1767
1768 err = -EINVAL;
1769 if (IS_MNT_UNBINDABLE(old_path.mnt))
1770 goto out2;
1771
1772 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1773 goto out2;
1774
1775 err = -ENOMEM;
1776 if (recurse)
1777 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1778 else
1779 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1780
1781 if (!mnt)
1782 goto out2;
1783
1784 err = graft_tree(mnt, path);
1785 if (err) {
1786 br_write_lock(vfsmount_lock);
1787 umount_tree(mnt, 0, &umount_list);
1788 br_write_unlock(vfsmount_lock);
1789 }
1790out2:
1791 unlock_mount(path);
1792 release_mounts(&umount_list);
1793out:
1794 path_put(&old_path);
1795 return err;
1796}
1797
1798static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1799{
1800 int error = 0;
1801 int readonly_request = 0;
1802
1803 if (ms_flags & MS_RDONLY)
1804 readonly_request = 1;
1805 if (readonly_request == __mnt_is_readonly(mnt))
1806 return 0;
1807
1808 if (readonly_request)
1809 error = mnt_make_readonly(mnt);
1810 else
1811 __mnt_unmake_readonly(mnt);
1812 return error;
1813}
1814
1815/*
1816 * change filesystem flags. dir should be a physical root of filesystem.
1817 * If you've mounted a non-root directory somewhere and want to do remount
1818 * on it - tough luck.
1819 */
1820static int do_remount(struct path *path, int flags, int mnt_flags,
1821 void *data)
1822{
1823 int err;
1824 struct super_block *sb = path->mnt->mnt_sb;
1825
1826 if (!capable(CAP_SYS_ADMIN))
1827 return -EPERM;
1828
1829 if (!check_mnt(path->mnt))
1830 return -EINVAL;
1831
1832 if (path->dentry != path->mnt->mnt_root)
1833 return -EINVAL;
1834
1835 err = security_sb_remount(sb, data);
1836 if (err)
1837 return err;
1838
1839 down_write(&sb->s_umount);
1840 if (flags & MS_BIND)
1841 err = change_mount_flags(path->mnt, flags);
1842 else
1843 err = do_remount_sb(sb, flags, data, 0);
1844 if (!err) {
1845 br_write_lock(vfsmount_lock);
1846 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1847 path->mnt->mnt_flags = mnt_flags;
1848 br_write_unlock(vfsmount_lock);
1849 }
1850 up_write(&sb->s_umount);
1851 if (!err) {
1852 br_write_lock(vfsmount_lock);
1853 touch_mnt_namespace(path->mnt->mnt_ns);
1854 br_write_unlock(vfsmount_lock);
1855 }
1856 return err;
1857}
1858
1859static inline int tree_contains_unbindable(struct vfsmount *mnt)
1860{
1861 struct vfsmount *p;
1862 for (p = mnt; p; p = next_mnt(p, mnt)) {
1863 if (IS_MNT_UNBINDABLE(p))
1864 return 1;
1865 }
1866 return 0;
1867}
1868
1869static int do_move_mount(struct path *path, char *old_name)
1870{
1871 struct path old_path, parent_path;
1872 struct vfsmount *p;
1873 int err = 0;
1874 if (!capable(CAP_SYS_ADMIN))
1875 return -EPERM;
1876 if (!old_name || !*old_name)
1877 return -EINVAL;
1878 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1879 if (err)
1880 return err;
1881
1882 err = lock_mount(path);
1883 if (err < 0)
1884 goto out;
1885
1886 err = -EINVAL;
1887 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1888 goto out1;
1889
1890 if (d_unlinked(path->dentry))
1891 goto out1;
1892
1893 err = -EINVAL;
1894 if (old_path.dentry != old_path.mnt->mnt_root)
1895 goto out1;
1896
1897 if (old_path.mnt == old_path.mnt->mnt_parent)
1898 goto out1;
1899
1900 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1901 S_ISDIR(old_path.dentry->d_inode->i_mode))
1902 goto out1;
1903 /*
1904 * Don't move a mount residing in a shared parent.
1905 */
1906 if (old_path.mnt->mnt_parent &&
1907 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1908 goto out1;
1909 /*
1910 * Don't move a mount tree containing unbindable mounts to a destination
1911 * mount which is shared.
1912 */
1913 if (IS_MNT_SHARED(path->mnt) &&
1914 tree_contains_unbindable(old_path.mnt))
1915 goto out1;
1916 err = -ELOOP;
1917 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1918 if (p == old_path.mnt)
1919 goto out1;
1920
1921 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1922 if (err)
1923 goto out1;
1924
1925 /* if the mount is moved, it should no longer be expire
1926 * automatically */
1927 list_del_init(&old_path.mnt->mnt_expire);
1928out1:
1929 unlock_mount(path);
1930out:
1931 if (!err)
1932 path_put(&parent_path);
1933 path_put(&old_path);
1934 return err;
1935}
1936
1937static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1938{
1939 int err;
1940 const char *subtype = strchr(fstype, '.');
1941 if (subtype) {
1942 subtype++;
1943 err = -EINVAL;
1944 if (!subtype[0])
1945 goto err;
1946 } else
1947 subtype = "";
1948
1949 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1950 err = -ENOMEM;
1951 if (!mnt->mnt_sb->s_subtype)
1952 goto err;
1953 return mnt;
1954
1955 err:
1956 mntput(mnt);
1957 return ERR_PTR(err);
1958}
1959
1960struct vfsmount *
1961do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1962{
1963 struct file_system_type *type = get_fs_type(fstype);
1964 struct vfsmount *mnt;
1965 if (!type)
1966 return ERR_PTR(-ENODEV);
1967 mnt = vfs_kern_mount(type, flags, name, data);
1968 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1969 !mnt->mnt_sb->s_subtype)
1970 mnt = fs_set_subtype(mnt, fstype);
1971 put_filesystem(type);
1972 return mnt;
1973}
1974EXPORT_SYMBOL_GPL(do_kern_mount);
1975
1976/*
1977 * add a mount into a namespace's mount tree
1978 */
1979static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1980{
1981 int err;
1982
1983 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1984
1985 err = lock_mount(path);
1986 if (err)
1987 return err;
1988
1989 err = -EINVAL;
1990 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1991 goto unlock;
1992
1993 /* Refuse the same filesystem on the same mount point */
1994 err = -EBUSY;
1995 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1996 path->mnt->mnt_root == path->dentry)
1997 goto unlock;
1998
1999 err = -EINVAL;
2000 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2001 goto unlock;
2002
2003 newmnt->mnt_flags = mnt_flags;
2004 err = graft_tree(newmnt, path);
2005
2006unlock:
2007 unlock_mount(path);
2008 return err;
2009}
2010
2011/*
2012 * create a new mount for userspace and request it to be added into the
2013 * namespace's tree
2014 */
2015static int do_new_mount(struct path *path, char *type, int flags,
2016 int mnt_flags, char *name, void *data)
2017{
2018 struct vfsmount *mnt;
2019 int err;
2020
2021 if (!type)
2022 return -EINVAL;
2023
2024 /* we need capabilities... */
2025 if (!capable(CAP_SYS_ADMIN))
2026 return -EPERM;
2027
2028 mnt = do_kern_mount(type, flags, name, data);
2029 if (IS_ERR(mnt))
2030 return PTR_ERR(mnt);
2031
2032 err = do_add_mount(mnt, path, mnt_flags);
2033 if (err)
2034 mntput(mnt);
2035 return err;
2036}
2037
2038int finish_automount(struct vfsmount *m, struct path *path)
2039{
2040 int err;
2041 /* The new mount record should have at least 2 refs to prevent it being
2042 * expired before we get a chance to add it
2043 */
2044 BUG_ON(mnt_get_count(m) < 2);
2045
2046 if (m->mnt_sb == path->mnt->mnt_sb &&
2047 m->mnt_root == path->dentry) {
2048 err = -ELOOP;
2049 goto fail;
2050 }
2051
2052 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2053 if (!err)
2054 return 0;
2055fail:
2056 /* remove m from any expiration list it may be on */
2057 if (!list_empty(&m->mnt_expire)) {
2058 down_write(&namespace_sem);
2059 br_write_lock(vfsmount_lock);
2060 list_del_init(&m->mnt_expire);
2061 br_write_unlock(vfsmount_lock);
2062 up_write(&namespace_sem);
2063 }
2064 mntput(m);
2065 mntput(m);
2066 return err;
2067}
2068
2069/**
2070 * mnt_set_expiry - Put a mount on an expiration list
2071 * @mnt: The mount to list.
2072 * @expiry_list: The list to add the mount to.
2073 */
2074void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2075{
2076 down_write(&namespace_sem);
2077 br_write_lock(vfsmount_lock);
2078
2079 list_add_tail(&mnt->mnt_expire, expiry_list);
2080
2081 br_write_unlock(vfsmount_lock);
2082 up_write(&namespace_sem);
2083}
2084EXPORT_SYMBOL(mnt_set_expiry);
2085
2086/*
2087 * process a list of expirable mountpoints with the intent of discarding any
2088 * mountpoints that aren't in use and haven't been touched since last we came
2089 * here
2090 */
2091void mark_mounts_for_expiry(struct list_head *mounts)
2092{
2093 struct vfsmount *mnt, *next;
2094 LIST_HEAD(graveyard);
2095 LIST_HEAD(umounts);
2096
2097 if (list_empty(mounts))
2098 return;
2099
2100 down_write(&namespace_sem);
2101 br_write_lock(vfsmount_lock);
2102
2103 /* extract from the expiration list every vfsmount that matches the
2104 * following criteria:
2105 * - only referenced by its parent vfsmount
2106 * - still marked for expiry (marked on the last call here; marks are
2107 * cleared by mntput())
2108 */
2109 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2110 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2111 propagate_mount_busy(mnt, 1))
2112 continue;
2113 list_move(&mnt->mnt_expire, &graveyard);
2114 }
2115 while (!list_empty(&graveyard)) {
2116 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2117 touch_mnt_namespace(mnt->mnt_ns);
2118 umount_tree(mnt, 1, &umounts);
2119 }
2120 br_write_unlock(vfsmount_lock);
2121 up_write(&namespace_sem);
2122
2123 release_mounts(&umounts);
2124}
2125
2126EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2127
2128/*
2129 * Ripoff of 'select_parent()'
2130 *
2131 * search the list of submounts for a given mountpoint, and move any
2132 * shrinkable submounts to the 'graveyard' list.
2133 */
2134static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2135{
2136 struct vfsmount *this_parent = parent;
2137 struct list_head *next;
2138 int found = 0;
2139
2140repeat:
2141 next = this_parent->mnt_mounts.next;
2142resume:
2143 while (next != &this_parent->mnt_mounts) {
2144 struct list_head *tmp = next;
2145 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2146
2147 next = tmp->next;
2148 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2149 continue;
2150 /*
2151 * Descend a level if the d_mounts list is non-empty.
2152 */
2153 if (!list_empty(&mnt->mnt_mounts)) {
2154 this_parent = mnt;
2155 goto repeat;
2156 }
2157
2158 if (!propagate_mount_busy(mnt, 1)) {
2159 list_move_tail(&mnt->mnt_expire, graveyard);
2160 found++;
2161 }
2162 }
2163 /*
2164 * All done at this level ... ascend and resume the search
2165 */
2166 if (this_parent != parent) {
2167 next = this_parent->mnt_child.next;
2168 this_parent = this_parent->mnt_parent;
2169 goto resume;
2170 }
2171 return found;
2172}
2173
2174/*
2175 * process a list of expirable mountpoints with the intent of discarding any
2176 * submounts of a specific parent mountpoint
2177 *
2178 * vfsmount_lock must be held for write
2179 */
2180static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2181{
2182 LIST_HEAD(graveyard);
2183 struct vfsmount *m;
2184
2185 /* extract submounts of 'mountpoint' from the expiration list */
2186 while (select_submounts(mnt, &graveyard)) {
2187 while (!list_empty(&graveyard)) {
2188 m = list_first_entry(&graveyard, struct vfsmount,
2189 mnt_expire);
2190 touch_mnt_namespace(m->mnt_ns);
2191 umount_tree(m, 1, umounts);
2192 }
2193 }
2194}
2195
2196/*
2197 * Some copy_from_user() implementations do not return the exact number of
2198 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2199 * Note that this function differs from copy_from_user() in that it will oops
2200 * on bad values of `to', rather than returning a short copy.
2201 */
2202static long exact_copy_from_user(void *to, const void __user * from,
2203 unsigned long n)
2204{
2205 char *t = to;
2206 const char __user *f = from;
2207 char c;
2208
2209 if (!access_ok(VERIFY_READ, from, n))
2210 return n;
2211
2212 while (n) {
2213 if (__get_user(c, f)) {
2214 memset(t, 0, n);
2215 break;
2216 }
2217 *t++ = c;
2218 f++;
2219 n--;
2220 }
2221 return n;
2222}
2223
2224int copy_mount_options(const void __user * data, unsigned long *where)
2225{
2226 int i;
2227 unsigned long page;
2228 unsigned long size;
2229
2230 *where = 0;
2231 if (!data)
2232 return 0;
2233
2234 if (!(page = __get_free_page(GFP_KERNEL)))
2235 return -ENOMEM;
2236
2237 /* We only care that *some* data at the address the user
2238 * gave us is valid. Just in case, we'll zero
2239 * the remainder of the page.
2240 */
2241 /* copy_from_user cannot cross TASK_SIZE ! */
2242 size = TASK_SIZE - (unsigned long)data;
2243 if (size > PAGE_SIZE)
2244 size = PAGE_SIZE;
2245
2246 i = size - exact_copy_from_user((void *)page, data, size);
2247 if (!i) {
2248 free_page(page);
2249 return -EFAULT;
2250 }
2251 if (i != PAGE_SIZE)
2252 memset((char *)page + i, 0, PAGE_SIZE - i);
2253 *where = page;
2254 return 0;
2255}
2256
2257int copy_mount_string(const void __user *data, char **where)
2258{
2259 char *tmp;
2260
2261 if (!data) {
2262 *where = NULL;
2263 return 0;
2264 }
2265
2266 tmp = strndup_user(data, PAGE_SIZE);
2267 if (IS_ERR(tmp))
2268 return PTR_ERR(tmp);
2269
2270 *where = tmp;
2271 return 0;
2272}
2273
2274/*
2275 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2276 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2277 *
2278 * data is a (void *) that can point to any structure up to
2279 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2280 * information (or be NULL).
2281 *
2282 * Pre-0.97 versions of mount() didn't have a flags word.
2283 * When the flags word was introduced its top half was required
2284 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2285 * Therefore, if this magic number is present, it carries no information
2286 * and must be discarded.
2287 */
2288long do_mount(char *dev_name, char *dir_name, char *type_page,
2289 unsigned long flags, void *data_page)
2290{
2291 struct path path;
2292 int retval = 0;
2293 int mnt_flags = 0;
2294
2295 /* Discard magic */
2296 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2297 flags &= ~MS_MGC_MSK;
2298
2299 /* Basic sanity checks */
2300
2301 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2302 return -EINVAL;
2303
2304 if (data_page)
2305 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2306
2307 /* ... and get the mountpoint */
2308 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2309 if (retval)
2310 return retval;
2311
2312 retval = security_sb_mount(dev_name, &path,
2313 type_page, flags, data_page);
2314 if (retval)
2315 goto dput_out;
2316
2317 /* Default to relatime unless overriden */
2318 if (!(flags & MS_NOATIME))
2319 mnt_flags |= MNT_RELATIME;
2320
2321 /* Separate the per-mountpoint flags */
2322 if (flags & MS_NOSUID)
2323 mnt_flags |= MNT_NOSUID;
2324 if (flags & MS_NODEV)
2325 mnt_flags |= MNT_NODEV;
2326 if (flags & MS_NOEXEC)
2327 mnt_flags |= MNT_NOEXEC;
2328 if (flags & MS_NOATIME)
2329 mnt_flags |= MNT_NOATIME;
2330 if (flags & MS_NODIRATIME)
2331 mnt_flags |= MNT_NODIRATIME;
2332 if (flags & MS_STRICTATIME)
2333 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2334 if (flags & MS_RDONLY)
2335 mnt_flags |= MNT_READONLY;
2336
2337 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2338 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2339 MS_STRICTATIME);
2340
2341 if (flags & MS_REMOUNT)
2342 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2343 data_page);
2344 else if (flags & MS_BIND)
2345 retval = do_loopback(&path, dev_name, flags & MS_REC);
2346 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2347 retval = do_change_type(&path, flags);
2348 else if (flags & MS_MOVE)
2349 retval = do_move_mount(&path, dev_name);
2350 else
2351 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2352 dev_name, data_page);
2353dput_out:
2354 path_put(&path);
2355 return retval;
2356}
2357
2358static struct mnt_namespace *alloc_mnt_ns(void)
2359{
2360 struct mnt_namespace *new_ns;
2361
2362 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2363 if (!new_ns)
2364 return ERR_PTR(-ENOMEM);
2365 atomic_set(&new_ns->count, 1);
2366 new_ns->root = NULL;
2367 INIT_LIST_HEAD(&new_ns->list);
2368 init_waitqueue_head(&new_ns->poll);
2369 new_ns->event = 0;
2370 return new_ns;
2371}
2372
2373void mnt_make_longterm(struct vfsmount *mnt)
2374{
2375 __mnt_make_longterm(mnt);
2376}
2377
2378void mnt_make_shortterm(struct vfsmount *mnt)
2379{
2380#ifdef CONFIG_SMP
2381 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2382 return;
2383 br_write_lock(vfsmount_lock);
2384 atomic_dec(&mnt->mnt_longterm);
2385 br_write_unlock(vfsmount_lock);
2386#endif
2387}
2388
2389/*
2390 * Allocate a new namespace structure and populate it with contents
2391 * copied from the namespace of the passed in task structure.
2392 */
2393static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2394 struct fs_struct *fs)
2395{
2396 struct mnt_namespace *new_ns;
2397 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2398 struct vfsmount *p, *q;
2399
2400 new_ns = alloc_mnt_ns();
2401 if (IS_ERR(new_ns))
2402 return new_ns;
2403
2404 down_write(&namespace_sem);
2405 /* First pass: copy the tree topology */
2406 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2407 CL_COPY_ALL | CL_EXPIRE);
2408 if (!new_ns->root) {
2409 up_write(&namespace_sem);
2410 kfree(new_ns);
2411 return ERR_PTR(-ENOMEM);
2412 }
2413 br_write_lock(vfsmount_lock);
2414 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2415 br_write_unlock(vfsmount_lock);
2416
2417 /*
2418 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2419 * as belonging to new namespace. We have already acquired a private
2420 * fs_struct, so tsk->fs->lock is not needed.
2421 */
2422 p = mnt_ns->root;
2423 q = new_ns->root;
2424 while (p) {
2425 q->mnt_ns = new_ns;
2426 __mnt_make_longterm(q);
2427 if (fs) {
2428 if (p == fs->root.mnt) {
2429 fs->root.mnt = mntget(q);
2430 __mnt_make_longterm(q);
2431 mnt_make_shortterm(p);
2432 rootmnt = p;
2433 }
2434 if (p == fs->pwd.mnt) {
2435 fs->pwd.mnt = mntget(q);
2436 __mnt_make_longterm(q);
2437 mnt_make_shortterm(p);
2438 pwdmnt = p;
2439 }
2440 }
2441 p = next_mnt(p, mnt_ns->root);
2442 q = next_mnt(q, new_ns->root);
2443 }
2444 up_write(&namespace_sem);
2445
2446 if (rootmnt)
2447 mntput(rootmnt);
2448 if (pwdmnt)
2449 mntput(pwdmnt);
2450
2451 return new_ns;
2452}
2453
2454struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2455 struct fs_struct *new_fs)
2456{
2457 struct mnt_namespace *new_ns;
2458
2459 BUG_ON(!ns);
2460 get_mnt_ns(ns);
2461
2462 if (!(flags & CLONE_NEWNS))
2463 return ns;
2464
2465 new_ns = dup_mnt_ns(ns, new_fs);
2466
2467 put_mnt_ns(ns);
2468 return new_ns;
2469}
2470
2471/**
2472 * create_mnt_ns - creates a private namespace and adds a root filesystem
2473 * @mnt: pointer to the new root filesystem mountpoint
2474 */
2475struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2476{
2477 struct mnt_namespace *new_ns;
2478
2479 new_ns = alloc_mnt_ns();
2480 if (!IS_ERR(new_ns)) {
2481 mnt->mnt_ns = new_ns;
2482 __mnt_make_longterm(mnt);
2483 new_ns->root = mnt;
2484 list_add(&new_ns->list, &new_ns->root->mnt_list);
2485 }
2486 return new_ns;
2487}
2488EXPORT_SYMBOL(create_mnt_ns);
2489
2490SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2491 char __user *, type, unsigned long, flags, void __user *, data)
2492{
2493 int ret;
2494 char *kernel_type;
2495 char *kernel_dir;
2496 char *kernel_dev;
2497 unsigned long data_page;
2498
2499 ret = copy_mount_string(type, &kernel_type);
2500 if (ret < 0)
2501 goto out_type;
2502
2503 kernel_dir = getname(dir_name);
2504 if (IS_ERR(kernel_dir)) {
2505 ret = PTR_ERR(kernel_dir);
2506 goto out_dir;
2507 }
2508
2509 ret = copy_mount_string(dev_name, &kernel_dev);
2510 if (ret < 0)
2511 goto out_dev;
2512
2513 ret = copy_mount_options(data, &data_page);
2514 if (ret < 0)
2515 goto out_data;
2516
2517 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2518 (void *) data_page);
2519
2520 free_page(data_page);
2521out_data:
2522 kfree(kernel_dev);
2523out_dev:
2524 putname(kernel_dir);
2525out_dir:
2526 kfree(kernel_type);
2527out_type:
2528 return ret;
2529}
2530
2531/*
2532 * pivot_root Semantics:
2533 * Moves the root file system of the current process to the directory put_old,
2534 * makes new_root as the new root file system of the current process, and sets
2535 * root/cwd of all processes which had them on the current root to new_root.
2536 *
2537 * Restrictions:
2538 * The new_root and put_old must be directories, and must not be on the
2539 * same file system as the current process root. The put_old must be
2540 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2541 * pointed to by put_old must yield the same directory as new_root. No other
2542 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2543 *
2544 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2545 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2546 * in this situation.
2547 *
2548 * Notes:
2549 * - we don't move root/cwd if they are not at the root (reason: if something
2550 * cared enough to change them, it's probably wrong to force them elsewhere)
2551 * - it's okay to pick a root that isn't the root of a file system, e.g.
2552 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2553 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2554 * first.
2555 */
2556SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2557 const char __user *, put_old)
2558{
2559 struct vfsmount *tmp;
2560 struct path new, old, parent_path, root_parent, root;
2561 int error;
2562
2563 if (!capable(CAP_SYS_ADMIN))
2564 return -EPERM;
2565
2566 error = user_path_dir(new_root, &new);
2567 if (error)
2568 goto out0;
2569
2570 error = user_path_dir(put_old, &old);
2571 if (error)
2572 goto out1;
2573
2574 error = security_sb_pivotroot(&old, &new);
2575 if (error)
2576 goto out2;
2577
2578 get_fs_root(current->fs, &root);
2579 error = lock_mount(&old);
2580 if (error)
2581 goto out3;
2582
2583 error = -EINVAL;
2584 if (IS_MNT_SHARED(old.mnt) ||
2585 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2586 IS_MNT_SHARED(root.mnt->mnt_parent))
2587 goto out4;
2588 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2589 goto out4;
2590 error = -ENOENT;
2591 if (d_unlinked(new.dentry))
2592 goto out4;
2593 if (d_unlinked(old.dentry))
2594 goto out4;
2595 error = -EBUSY;
2596 if (new.mnt == root.mnt ||
2597 old.mnt == root.mnt)
2598 goto out4; /* loop, on the same file system */
2599 error = -EINVAL;
2600 if (root.mnt->mnt_root != root.dentry)
2601 goto out4; /* not a mountpoint */
2602 if (root.mnt->mnt_parent == root.mnt)
2603 goto out4; /* not attached */
2604 if (new.mnt->mnt_root != new.dentry)
2605 goto out4; /* not a mountpoint */
2606 if (new.mnt->mnt_parent == new.mnt)
2607 goto out4; /* not attached */
2608 /* make sure we can reach put_old from new_root */
2609 tmp = old.mnt;
2610 if (tmp != new.mnt) {
2611 for (;;) {
2612 if (tmp->mnt_parent == tmp)
2613 goto out4; /* already mounted on put_old */
2614 if (tmp->mnt_parent == new.mnt)
2615 break;
2616 tmp = tmp->mnt_parent;
2617 }
2618 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2619 goto out4;
2620 } else if (!is_subdir(old.dentry, new.dentry))
2621 goto out4;
2622 br_write_lock(vfsmount_lock);
2623 detach_mnt(new.mnt, &parent_path);
2624 detach_mnt(root.mnt, &root_parent);
2625 /* mount old root on put_old */
2626 attach_mnt(root.mnt, &old);
2627 /* mount new_root on / */
2628 attach_mnt(new.mnt, &root_parent);
2629 touch_mnt_namespace(current->nsproxy->mnt_ns);
2630 br_write_unlock(vfsmount_lock);
2631 chroot_fs_refs(&root, &new);
2632 error = 0;
2633out4:
2634 unlock_mount(&old);
2635 if (!error) {
2636 path_put(&root_parent);
2637 path_put(&parent_path);
2638 }
2639out3:
2640 path_put(&root);
2641out2:
2642 path_put(&old);
2643out1:
2644 path_put(&new);
2645out0:
2646 return error;
2647}
2648
2649static void __init init_mount_tree(void)
2650{
2651 struct vfsmount *mnt;
2652 struct mnt_namespace *ns;
2653 struct path root;
2654
2655 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2656 if (IS_ERR(mnt))
2657 panic("Can't create rootfs");
2658
2659 ns = create_mnt_ns(mnt);
2660 if (IS_ERR(ns))
2661 panic("Can't allocate initial namespace");
2662
2663 init_task.nsproxy->mnt_ns = ns;
2664 get_mnt_ns(ns);
2665
2666 root.mnt = ns->root;
2667 root.dentry = ns->root->mnt_root;
2668
2669 set_fs_pwd(current->fs, &root);
2670 set_fs_root(current->fs, &root);
2671}
2672
2673void __init mnt_init(void)
2674{
2675 unsigned u;
2676 int err;
2677
2678 init_rwsem(&namespace_sem);
2679
2680 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2681 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2682
2683 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2684
2685 if (!mount_hashtable)
2686 panic("Failed to allocate mount hash table\n");
2687
2688 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2689
2690 for (u = 0; u < HASH_SIZE; u++)
2691 INIT_LIST_HEAD(&mount_hashtable[u]);
2692
2693 br_lock_init(vfsmount_lock);
2694
2695 err = sysfs_init();
2696 if (err)
2697 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2698 __func__, err);
2699 fs_kobj = kobject_create_and_add("fs", NULL);
2700 if (!fs_kobj)
2701 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2702 init_rootfs();
2703 init_mount_tree();
2704}
2705
2706void put_mnt_ns(struct mnt_namespace *ns)
2707{
2708 LIST_HEAD(umount_list);
2709
2710 if (!atomic_dec_and_test(&ns->count))
2711 return;
2712 down_write(&namespace_sem);
2713 br_write_lock(vfsmount_lock);
2714 umount_tree(ns->root, 0, &umount_list);
2715 br_write_unlock(vfsmount_lock);
2716 up_write(&namespace_sem);
2717 release_mounts(&umount_list);
2718 kfree(ns);
2719}
2720EXPORT_SYMBOL(put_mnt_ns);
2721
2722struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2723{
2724 struct vfsmount *mnt;
2725 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2726 if (!IS_ERR(mnt)) {
2727 /*
2728 * it is a longterm mount, don't release mnt until
2729 * we unmount before file sys is unregistered
2730 */
2731 mnt_make_longterm(mnt);
2732 }
2733 return mnt;
2734}
2735EXPORT_SYMBOL_GPL(kern_mount_data);
2736
2737void kern_unmount(struct vfsmount *mnt)
2738{
2739 /* release long term mount so mount point can be released */
2740 if (!IS_ERR_OR_NULL(mnt)) {
2741 mnt_make_shortterm(mnt);
2742 mntput(mnt);
2743 }
2744}
2745EXPORT_SYMBOL(kern_unmount);