source: src/linux/universal/linux-4.9/fs/namespace.c @ 31662

Last change on this file since 31662 was 31662, checked in by brainslayer, 4 months ago

use new squashfs in all kernels

File size: 84.6 KB
Line 
1/*
2 *  linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 *      Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11#include <linux/syscalls.h>
12#include <linux/export.h>
13#include <linux/capability.h>
14#include <linux/mnt_namespace.h>
15#include <linux/user_namespace.h>
16#include <linux/namei.h>
17#include <linux/security.h>
18#include <linux/idr.h>
19#include <linux/init.h>         /* init_rootfs */
20#include <linux/fs_struct.h>    /* get_fs_root et.al. */
21#include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
22#include <linux/uaccess.h>
23#include <linux/proc_ns.h>
24#include <linux/magic.h>
25#include <linux/bootmem.h>
26#include <linux/task_work.h>
27#include "pnode.h"
28#include "internal.h"
29
30/* Maximum number of mounts in a mount namespace */
31unsigned int sysctl_mount_max __read_mostly = 100000;
32
33static unsigned int m_hash_mask __read_mostly;
34static unsigned int m_hash_shift __read_mostly;
35static unsigned int mp_hash_mask __read_mostly;
36static unsigned int mp_hash_shift __read_mostly;
37
38static __initdata unsigned long mhash_entries;
39static int __init set_mhash_entries(char *str)
40{
41        if (!str)
42                return 0;
43        mhash_entries = simple_strtoul(str, &str, 0);
44        return 1;
45}
46__setup("mhash_entries=", set_mhash_entries);
47
48static __initdata unsigned long mphash_entries;
49static int __init set_mphash_entries(char *str)
50{
51        if (!str)
52                return 0;
53        mphash_entries = simple_strtoul(str, &str, 0);
54        return 1;
55}
56__setup("mphash_entries=", set_mphash_entries);
57
58static u64 event;
59static DEFINE_IDA(mnt_id_ida);
60static DEFINE_IDA(mnt_group_ida);
61static DEFINE_SPINLOCK(mnt_id_lock);
62static int mnt_id_start = 0;
63static int mnt_group_start = 1;
64
65static struct hlist_head *mount_hashtable __read_mostly;
66static struct hlist_head *mountpoint_hashtable __read_mostly;
67static struct kmem_cache *mnt_cache __read_mostly;
68static DECLARE_RWSEM(namespace_sem);
69
70/* /sys/fs */
71struct kobject *fs_kobj;
72EXPORT_SYMBOL_GPL(fs_kobj);
73
74/*
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
77 * up the tree.
78 *
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
81 */
82__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
83
84static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
85{
86        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
88        tmp = tmp + (tmp >> m_hash_shift);
89        return &mount_hashtable[tmp & m_hash_mask];
90}
91
92static inline struct hlist_head *mp_hash(struct dentry *dentry)
93{
94        unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95        tmp = tmp + (tmp >> mp_hash_shift);
96        return &mountpoint_hashtable[tmp & mp_hash_mask];
97}
98
99/*
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
102 */
103static int mnt_alloc_id(struct mount *mnt)
104{
105        int res;
106
107retry:
108        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
109        spin_lock(&mnt_id_lock);
110        res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
111        if (!res)
112                mnt_id_start = mnt->mnt_id + 1;
113        spin_unlock(&mnt_id_lock);
114        if (res == -EAGAIN)
115                goto retry;
116
117        return res;
118}
119
120static void mnt_free_id(struct mount *mnt)
121{
122        int id = mnt->mnt_id;
123        spin_lock(&mnt_id_lock);
124        ida_remove(&mnt_id_ida, id);
125        if (mnt_id_start > id)
126                mnt_id_start = id;
127        spin_unlock(&mnt_id_lock);
128}
129
130/*
131 * Allocate a new peer group ID
132 *
133 * mnt_group_ida is protected by namespace_sem
134 */
135static int mnt_alloc_group_id(struct mount *mnt)
136{
137        int res;
138
139        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
140                return -ENOMEM;
141
142        res = ida_get_new_above(&mnt_group_ida,
143                                mnt_group_start,
144                                &mnt->mnt_group_id);
145        if (!res)
146                mnt_group_start = mnt->mnt_group_id + 1;
147
148        return res;
149}
150
151/*
152 * Release a peer group ID
153 */
154void mnt_release_group_id(struct mount *mnt)
155{
156        int id = mnt->mnt_group_id;
157        ida_remove(&mnt_group_ida, id);
158        if (mnt_group_start > id)
159                mnt_group_start = id;
160        mnt->mnt_group_id = 0;
161}
162
163/*
164 * vfsmount lock must be held for read
165 */
166static inline void mnt_add_count(struct mount *mnt, int n)
167{
168#ifdef CONFIG_SMP
169        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
170#else
171        preempt_disable();
172        mnt->mnt_count += n;
173        preempt_enable();
174#endif
175}
176
177/*
178 * vfsmount lock must be held for write
179 */
180unsigned int mnt_get_count(struct mount *mnt)
181{
182#ifdef CONFIG_SMP
183        unsigned int count = 0;
184        int cpu;
185
186        for_each_possible_cpu(cpu) {
187                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
188        }
189
190        return count;
191#else
192        return mnt->mnt_count;
193#endif
194}
195
196static void drop_mountpoint(struct fs_pin *p)
197{
198        struct mount *m = container_of(p, struct mount, mnt_umount);
199        dput(m->mnt_ex_mountpoint);
200        pin_remove(p);
201        mntput(&m->mnt);
202}
203
204static struct mount *alloc_vfsmnt(const char *name)
205{
206        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207        if (mnt) {
208                int err;
209
210                err = mnt_alloc_id(mnt);
211                if (err)
212                        goto out_free_cache;
213
214                if (name) {
215                        mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
216                        if (!mnt->mnt_devname)
217                                goto out_free_id;
218                }
219
220#ifdef CONFIG_SMP
221                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
222                if (!mnt->mnt_pcp)
223                        goto out_free_devname;
224
225                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
226#else
227                mnt->mnt_count = 1;
228                mnt->mnt_writers = 0;
229#endif
230
231                INIT_HLIST_NODE(&mnt->mnt_hash);
232                INIT_LIST_HEAD(&mnt->mnt_child);
233                INIT_LIST_HEAD(&mnt->mnt_mounts);
234                INIT_LIST_HEAD(&mnt->mnt_list);
235                INIT_LIST_HEAD(&mnt->mnt_expire);
236                INIT_LIST_HEAD(&mnt->mnt_share);
237                INIT_LIST_HEAD(&mnt->mnt_slave_list);
238                INIT_LIST_HEAD(&mnt->mnt_slave);
239                INIT_HLIST_NODE(&mnt->mnt_mp_list);
240#ifdef CONFIG_FSNOTIFY
241                INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
242#endif
243                init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
244        }
245        return mnt;
246
247#ifdef CONFIG_SMP
248out_free_devname:
249        kfree_const(mnt->mnt_devname);
250#endif
251out_free_id:
252        mnt_free_id(mnt);
253out_free_cache:
254        kmem_cache_free(mnt_cache, mnt);
255        return NULL;
256}
257
258/*
259 * Most r/o checks on a fs are for operations that take
260 * discrete amounts of time, like a write() or unlink().
261 * We must keep track of when those operations start
262 * (for permission checks) and when they end, so that
263 * we can determine when writes are able to occur to
264 * a filesystem.
265 */
266/*
267 * __mnt_is_readonly: check whether a mount is read-only
268 * @mnt: the mount to check for its write status
269 *
270 * This shouldn't be used directly ouside of the VFS.
271 * It does not guarantee that the filesystem will stay
272 * r/w, just that it is right *now*.  This can not and
273 * should not be used in place of IS_RDONLY(inode).
274 * mnt_want/drop_write() will _keep_ the filesystem
275 * r/w.
276 */
277int __mnt_is_readonly(struct vfsmount *mnt)
278{
279        if (mnt->mnt_flags & MNT_READONLY)
280                return 1;
281        if (mnt->mnt_sb->s_flags & MS_RDONLY)
282                return 1;
283        return 0;
284}
285EXPORT_SYMBOL_GPL(__mnt_is_readonly);
286
287static inline void mnt_inc_writers(struct mount *mnt)
288{
289#ifdef CONFIG_SMP
290        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
291#else
292        mnt->mnt_writers++;
293#endif
294}
295
296static inline void mnt_dec_writers(struct mount *mnt)
297{
298#ifdef CONFIG_SMP
299        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
300#else
301        mnt->mnt_writers--;
302#endif
303}
304
305static unsigned int mnt_get_writers(struct mount *mnt)
306{
307#ifdef CONFIG_SMP
308        unsigned int count = 0;
309        int cpu;
310
311        for_each_possible_cpu(cpu) {
312                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
313        }
314
315        return count;
316#else
317        return mnt->mnt_writers;
318#endif
319}
320
321static int mnt_is_readonly(struct vfsmount *mnt)
322{
323        if (mnt->mnt_sb->s_readonly_remount)
324                return 1;
325        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
326        smp_rmb();
327        return __mnt_is_readonly(mnt);
328}
329
330/*
331 * Most r/o & frozen checks on a fs are for operations that take discrete
332 * amounts of time, like a write() or unlink().  We must keep track of when
333 * those operations start (for permission checks) and when they end, so that we
334 * can determine when writes are able to occur to a filesystem.
335 */
336/**
337 * __mnt_want_write - get write access to a mount without freeze protection
338 * @m: the mount on which to take a write
339 *
340 * This tells the low-level filesystem that a write is about to be performed to
341 * it, and makes sure that writes are allowed (mnt it read-write) before
342 * returning success. This operation does not protect against filesystem being
343 * frozen. When the write operation is finished, __mnt_drop_write() must be
344 * called. This is effectively a refcount.
345 */
346int __mnt_want_write(struct vfsmount *m)
347{
348        struct mount *mnt = real_mount(m);
349        int ret = 0;
350
351        preempt_disable();
352        mnt_inc_writers(mnt);
353        /*
354         * The store to mnt_inc_writers must be visible before we pass
355         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
356         * incremented count after it has set MNT_WRITE_HOLD.
357         */
358        smp_mb();
359        while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
360                cpu_relax();
361        /*
362         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
363         * be set to match its requirements. So we must not load that until
364         * MNT_WRITE_HOLD is cleared.
365         */
366        smp_rmb();
367        if (mnt_is_readonly(m)) {
368                mnt_dec_writers(mnt);
369                ret = -EROFS;
370        }
371        preempt_enable();
372
373        return ret;
374}
375
376/**
377 * mnt_want_write - get write access to a mount
378 * @m: the mount on which to take a write
379 *
380 * This tells the low-level filesystem that a write is about to be performed to
381 * it, and makes sure that writes are allowed (mount is read-write, filesystem
382 * is not frozen) before returning success.  When the write operation is
383 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
384 */
385int mnt_want_write(struct vfsmount *m)
386{
387        int ret;
388
389        sb_start_write(m->mnt_sb);
390        ret = __mnt_want_write(m);
391        if (ret)
392                sb_end_write(m->mnt_sb);
393        return ret;
394}
395EXPORT_SYMBOL_GPL(mnt_want_write);
396
397/**
398 * mnt_clone_write - get write access to a mount
399 * @mnt: the mount on which to take a write
400 *
401 * This is effectively like mnt_want_write, except
402 * it must only be used to take an extra write reference
403 * on a mountpoint that we already know has a write reference
404 * on it. This allows some optimisation.
405 *
406 * After finished, mnt_drop_write must be called as usual to
407 * drop the reference.
408 */
409int mnt_clone_write(struct vfsmount *mnt)
410{
411        /* superblock may be r/o */
412        if (__mnt_is_readonly(mnt))
413                return -EROFS;
414        preempt_disable();
415        mnt_inc_writers(real_mount(mnt));
416        preempt_enable();
417        return 0;
418}
419EXPORT_SYMBOL_GPL(mnt_clone_write);
420
421/**
422 * __mnt_want_write_file - get write access to a file's mount
423 * @file: the file who's mount on which to take a write
424 *
425 * This is like __mnt_want_write, but it takes a file and can
426 * do some optimisations if the file is open for write already
427 */
428int __mnt_want_write_file(struct file *file)
429{
430        if (!(file->f_mode & FMODE_WRITER))
431                return __mnt_want_write(file->f_path.mnt);
432        else
433                return mnt_clone_write(file->f_path.mnt);
434}
435
436/**
437 * mnt_want_write_file - get write access to a file's mount
438 * @file: the file who's mount on which to take a write
439 *
440 * This is like mnt_want_write, but it takes a file and can
441 * do some optimisations if the file is open for write already
442 */
443int mnt_want_write_file(struct file *file)
444{
445        int ret;
446
447        sb_start_write(file->f_path.mnt->mnt_sb);
448        ret = __mnt_want_write_file(file);
449        if (ret)
450                sb_end_write(file->f_path.mnt->mnt_sb);
451        return ret;
452}
453EXPORT_SYMBOL_GPL(mnt_want_write_file);
454
455/**
456 * __mnt_drop_write - give up write access to a mount
457 * @mnt: the mount on which to give up write access
458 *
459 * Tells the low-level filesystem that we are done
460 * performing writes to it.  Must be matched with
461 * __mnt_want_write() call above.
462 */
463void __mnt_drop_write(struct vfsmount *mnt)
464{
465        preempt_disable();
466        mnt_dec_writers(real_mount(mnt));
467        preempt_enable();
468}
469
470/**
471 * mnt_drop_write - give up write access to a mount
472 * @mnt: the mount on which to give up write access
473 *
474 * Tells the low-level filesystem that we are done performing writes to it and
475 * also allows filesystem to be frozen again.  Must be matched with
476 * mnt_want_write() call above.
477 */
478void mnt_drop_write(struct vfsmount *mnt)
479{
480        __mnt_drop_write(mnt);
481        sb_end_write(mnt->mnt_sb);
482}
483EXPORT_SYMBOL_GPL(mnt_drop_write);
484
485void __mnt_drop_write_file(struct file *file)
486{
487        __mnt_drop_write(file->f_path.mnt);
488}
489
490void mnt_drop_write_file(struct file *file)
491{
492        mnt_drop_write(file->f_path.mnt);
493}
494EXPORT_SYMBOL(mnt_drop_write_file);
495
496static int mnt_make_readonly(struct mount *mnt)
497{
498        int ret = 0;
499
500        lock_mount_hash();
501        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
502        /*
503         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
504         * should be visible before we do.
505         */
506        smp_mb();
507
508        /*
509         * With writers on hold, if this value is zero, then there are
510         * definitely no active writers (although held writers may subsequently
511         * increment the count, they'll have to wait, and decrement it after
512         * seeing MNT_READONLY).
513         *
514         * It is OK to have counter incremented on one CPU and decremented on
515         * another: the sum will add up correctly. The danger would be when we
516         * sum up each counter, if we read a counter before it is incremented,
517         * but then read another CPU's count which it has been subsequently
518         * decremented from -- we would see more decrements than we should.
519         * MNT_WRITE_HOLD protects against this scenario, because
520         * mnt_want_write first increments count, then smp_mb, then spins on
521         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
522         * we're counting up here.
523         */
524        if (mnt_get_writers(mnt) > 0)
525                ret = -EBUSY;
526        else
527                mnt->mnt.mnt_flags |= MNT_READONLY;
528        /*
529         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
530         * that become unheld will see MNT_READONLY.
531         */
532        smp_wmb();
533        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
534        unlock_mount_hash();
535        return ret;
536}
537
538static void __mnt_unmake_readonly(struct mount *mnt)
539{
540        lock_mount_hash();
541        mnt->mnt.mnt_flags &= ~MNT_READONLY;
542        unlock_mount_hash();
543}
544
545int sb_prepare_remount_readonly(struct super_block *sb)
546{
547        struct mount *mnt;
548        int err = 0;
549
550        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
551        if (atomic_long_read(&sb->s_remove_count))
552                return -EBUSY;
553
554        lock_mount_hash();
555        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
556                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
557                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
558                        smp_mb();
559                        if (mnt_get_writers(mnt) > 0) {
560                                err = -EBUSY;
561                                break;
562                        }
563                }
564        }
565        if (!err && atomic_long_read(&sb->s_remove_count))
566                err = -EBUSY;
567
568        if (!err) {
569                sb->s_readonly_remount = 1;
570                smp_wmb();
571        }
572        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
573                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
574                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
575        }
576        unlock_mount_hash();
577
578        return err;
579}
580
581static void free_vfsmnt(struct mount *mnt)
582{
583        kfree_const(mnt->mnt_devname);
584#ifdef CONFIG_SMP
585        free_percpu(mnt->mnt_pcp);
586#endif
587        kmem_cache_free(mnt_cache, mnt);
588}
589
590static void delayed_free_vfsmnt(struct rcu_head *head)
591{
592        free_vfsmnt(container_of(head, struct mount, mnt_rcu));
593}
594
595/* call under rcu_read_lock */
596int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
597{
598        struct mount *mnt;
599        if (read_seqretry(&mount_lock, seq))
600                return 1;
601        if (bastard == NULL)
602                return 0;
603        mnt = real_mount(bastard);
604        mnt_add_count(mnt, 1);
605        if (likely(!read_seqretry(&mount_lock, seq)))
606                return 0;
607        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
608                mnt_add_count(mnt, -1);
609                return 1;
610        }
611        return -1;
612}
613
614/* call under rcu_read_lock */
615bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
616{
617        int res = __legitimize_mnt(bastard, seq);
618        if (likely(!res))
619                return true;
620        if (unlikely(res < 0)) {
621                rcu_read_unlock();
622                mntput(bastard);
623                rcu_read_lock();
624        }
625        return false;
626}
627
628/*
629 * find the first mount at @dentry on vfsmount @mnt.
630 * call under rcu_read_lock()
631 */
632struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
633{
634        struct hlist_head *head = m_hash(mnt, dentry);
635        struct mount *p;
636
637        hlist_for_each_entry_rcu(p, head, mnt_hash)
638                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
639                        return p;
640        return NULL;
641}
642
643/*
644 * lookup_mnt - Return the first child mount mounted at path
645 *
646 * "First" means first mounted chronologically.  If you create the
647 * following mounts:
648 *
649 * mount /dev/sda1 /mnt
650 * mount /dev/sda2 /mnt
651 * mount /dev/sda3 /mnt
652 *
653 * Then lookup_mnt() on the base /mnt dentry in the root mount will
654 * return successively the root dentry and vfsmount of /dev/sda1, then
655 * /dev/sda2, then /dev/sda3, then NULL.
656 *
657 * lookup_mnt takes a reference to the found vfsmount.
658 */
659struct vfsmount *lookup_mnt(struct path *path)
660{
661        struct mount *child_mnt;
662        struct vfsmount *m;
663        unsigned seq;
664
665        rcu_read_lock();
666        do {
667                seq = read_seqbegin(&mount_lock);
668                child_mnt = __lookup_mnt(path->mnt, path->dentry);
669                m = child_mnt ? &child_mnt->mnt : NULL;
670        } while (!legitimize_mnt(m, seq));
671        rcu_read_unlock();
672        return m;
673}
674
675/*
676 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
677 *                         current mount namespace.
678 *
679 * The common case is dentries are not mountpoints at all and that
680 * test is handled inline.  For the slow case when we are actually
681 * dealing with a mountpoint of some kind, walk through all of the
682 * mounts in the current mount namespace and test to see if the dentry
683 * is a mountpoint.
684 *
685 * The mount_hashtable is not usable in the context because we
686 * need to identify all mounts that may be in the current mount
687 * namespace not just a mount that happens to have some specified
688 * parent mount.
689 */
690bool __is_local_mountpoint(struct dentry *dentry)
691{
692        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
693        struct mount *mnt;
694        bool is_covered = false;
695
696        if (!d_mountpoint(dentry))
697                goto out;
698
699        down_read(&namespace_sem);
700        list_for_each_entry(mnt, &ns->list, mnt_list) {
701                is_covered = (mnt->mnt_mountpoint == dentry);
702                if (is_covered)
703                        break;
704        }
705        up_read(&namespace_sem);
706out:
707        return is_covered;
708}
709
710static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
711{
712        struct hlist_head *chain = mp_hash(dentry);
713        struct mountpoint *mp;
714
715        hlist_for_each_entry(mp, chain, m_hash) {
716                if (mp->m_dentry == dentry) {
717                        /* might be worth a WARN_ON() */
718                        if (d_unlinked(dentry))
719                                return ERR_PTR(-ENOENT);
720                        mp->m_count++;
721                        return mp;
722                }
723        }
724        return NULL;
725}
726
727static struct mountpoint *get_mountpoint(struct dentry *dentry)
728{
729        struct mountpoint *mp, *new = NULL;
730        int ret;
731
732        if (d_mountpoint(dentry)) {
733mountpoint:
734                read_seqlock_excl(&mount_lock);
735                mp = lookup_mountpoint(dentry);
736                read_sequnlock_excl(&mount_lock);
737                if (mp)
738                        goto done;
739        }
740
741        if (!new)
742                new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
743        if (!new)
744                return ERR_PTR(-ENOMEM);
745
746
747        /* Exactly one processes may set d_mounted */
748        ret = d_set_mounted(dentry);
749
750        /* Someone else set d_mounted? */
751        if (ret == -EBUSY)
752                goto mountpoint;
753
754        /* The dentry is not available as a mountpoint? */
755        mp = ERR_PTR(ret);
756        if (ret)
757                goto done;
758
759        /* Add the new mountpoint to the hash table */
760        read_seqlock_excl(&mount_lock);
761        new->m_dentry = dentry;
762        new->m_count = 1;
763        hlist_add_head(&new->m_hash, mp_hash(dentry));
764        INIT_HLIST_HEAD(&new->m_list);
765        read_sequnlock_excl(&mount_lock);
766
767        mp = new;
768        new = NULL;
769done:
770        kfree(new);
771        return mp;
772}
773
774static void put_mountpoint(struct mountpoint *mp)
775{
776        if (!--mp->m_count) {
777                struct dentry *dentry = mp->m_dentry;
778                BUG_ON(!hlist_empty(&mp->m_list));
779                spin_lock(&dentry->d_lock);
780                dentry->d_flags &= ~DCACHE_MOUNTED;
781                spin_unlock(&dentry->d_lock);
782                hlist_del(&mp->m_hash);
783                kfree(mp);
784        }
785}
786
787static inline int check_mnt(struct mount *mnt)
788{
789        return mnt->mnt_ns == current->nsproxy->mnt_ns;
790}
791
792/*
793 * vfsmount lock must be held for write
794 */
795static void touch_mnt_namespace(struct mnt_namespace *ns)
796{
797        if (ns) {
798                ns->event = ++event;
799                wake_up_interruptible(&ns->poll);
800        }
801}
802
803/*
804 * vfsmount lock must be held for write
805 */
806static void __touch_mnt_namespace(struct mnt_namespace *ns)
807{
808        if (ns && ns->event != event) {
809                ns->event = event;
810                wake_up_interruptible(&ns->poll);
811        }
812}
813
814/*
815 * vfsmount lock must be held for write
816 */
817static void unhash_mnt(struct mount *mnt)
818{
819        mnt->mnt_parent = mnt;
820        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
821        list_del_init(&mnt->mnt_child);
822        hlist_del_init_rcu(&mnt->mnt_hash);
823        hlist_del_init(&mnt->mnt_mp_list);
824        put_mountpoint(mnt->mnt_mp);
825        mnt->mnt_mp = NULL;
826}
827
828/*
829 * vfsmount lock must be held for write
830 */
831static void detach_mnt(struct mount *mnt, struct path *old_path)
832{
833        old_path->dentry = mnt->mnt_mountpoint;
834        old_path->mnt = &mnt->mnt_parent->mnt;
835        unhash_mnt(mnt);
836}
837
838/*
839 * vfsmount lock must be held for write
840 */
841static void umount_mnt(struct mount *mnt)
842{
843        /* old mountpoint will be dropped when we can do that */
844        mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
845        unhash_mnt(mnt);
846}
847
848/*
849 * vfsmount lock must be held for write
850 */
851void mnt_set_mountpoint(struct mount *mnt,
852                        struct mountpoint *mp,
853                        struct mount *child_mnt)
854{
855        mp->m_count++;
856        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
857        child_mnt->mnt_mountpoint = dget(mp->m_dentry);
858        child_mnt->mnt_parent = mnt;
859        child_mnt->mnt_mp = mp;
860        hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
861}
862
863static void __attach_mnt(struct mount *mnt, struct mount *parent)
864{
865        hlist_add_head_rcu(&mnt->mnt_hash,
866                           m_hash(&parent->mnt, mnt->mnt_mountpoint));
867        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
868}
869
870/*
871 * vfsmount lock must be held for write
872 */
873static void attach_mnt(struct mount *mnt,
874                        struct mount *parent,
875                        struct mountpoint *mp)
876{
877        mnt_set_mountpoint(parent, mp, mnt);
878        __attach_mnt(mnt, parent);
879}
880
881void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
882{
883        struct mountpoint *old_mp = mnt->mnt_mp;
884        struct dentry *old_mountpoint = mnt->mnt_mountpoint;
885        struct mount *old_parent = mnt->mnt_parent;
886
887        list_del_init(&mnt->mnt_child);
888        hlist_del_init(&mnt->mnt_mp_list);
889        hlist_del_init_rcu(&mnt->mnt_hash);
890
891        attach_mnt(mnt, parent, mp);
892
893        put_mountpoint(old_mp);
894
895        /*
896         * Safely avoid even the suggestion this code might sleep or
897         * lock the mount hash by taking advantage of the knowledge that
898         * mnt_change_mountpoint will not release the final reference
899         * to a mountpoint.
900         *
901         * During mounting, the mount passed in as the parent mount will
902         * continue to use the old mountpoint and during unmounting, the
903         * old mountpoint will continue to exist until namespace_unlock,
904         * which happens well after mnt_change_mountpoint.
905         */
906        spin_lock(&old_mountpoint->d_lock);
907        old_mountpoint->d_lockref.count--;
908        spin_unlock(&old_mountpoint->d_lock);
909
910        mnt_add_count(old_parent, -1);
911}
912
913/*
914 * vfsmount lock must be held for write
915 */
916static void commit_tree(struct mount *mnt)
917{
918        struct mount *parent = mnt->mnt_parent;
919        struct mount *m;
920        LIST_HEAD(head);
921        struct mnt_namespace *n = parent->mnt_ns;
922
923        BUG_ON(parent == mnt);
924
925        list_add_tail(&head, &mnt->mnt_list);
926        list_for_each_entry(m, &head, mnt_list)
927                m->mnt_ns = n;
928
929        list_splice(&head, n->list.prev);
930
931        n->mounts += n->pending_mounts;
932        n->pending_mounts = 0;
933
934        __attach_mnt(mnt, parent);
935        touch_mnt_namespace(n);
936}
937
938static struct mount *next_mnt(struct mount *p, struct mount *root)
939{
940        struct list_head *next = p->mnt_mounts.next;
941        if (next == &p->mnt_mounts) {
942                while (1) {
943                        if (p == root)
944                                return NULL;
945                        next = p->mnt_child.next;
946                        if (next != &p->mnt_parent->mnt_mounts)
947                                break;
948                        p = p->mnt_parent;
949                }
950        }
951        return list_entry(next, struct mount, mnt_child);
952}
953
954static struct mount *skip_mnt_tree(struct mount *p)
955{
956        struct list_head *prev = p->mnt_mounts.prev;
957        while (prev != &p->mnt_mounts) {
958                p = list_entry(prev, struct mount, mnt_child);
959                prev = p->mnt_mounts.prev;
960        }
961        return p;
962}
963
964struct vfsmount *
965vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
966{
967        struct mount *mnt;
968        struct dentry *root;
969
970        if (!type)
971                return ERR_PTR(-ENODEV);
972
973        mnt = alloc_vfsmnt(name);
974        if (!mnt)
975                return ERR_PTR(-ENOMEM);
976
977        if (flags & MS_KERNMOUNT)
978                mnt->mnt.mnt_flags = MNT_INTERNAL;
979
980        root = mount_fs(type, flags, name, data);
981        if (IS_ERR(root)) {
982                mnt_free_id(mnt);
983                free_vfsmnt(mnt);
984                return ERR_CAST(root);
985        }
986
987        mnt->mnt.mnt_root = root;
988        mnt->mnt.mnt_sb = root->d_sb;
989        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
990        mnt->mnt_parent = mnt;
991        lock_mount_hash();
992        list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
993        unlock_mount_hash();
994        return &mnt->mnt;
995}
996EXPORT_SYMBOL_GPL(vfs_kern_mount);
997
998struct vfsmount *
999vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1000             const char *name, void *data)
1001{
1002        /* Until it is worked out how to pass the user namespace
1003         * through from the parent mount to the submount don't support
1004         * unprivileged mounts with submounts.
1005         */
1006        if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1007                return ERR_PTR(-EPERM);
1008
1009        return vfs_kern_mount(type, MS_SUBMOUNT, name, data);
1010}
1011EXPORT_SYMBOL_GPL(vfs_submount);
1012
1013static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1014                                        int flag)
1015{
1016        struct super_block *sb = old->mnt.mnt_sb;
1017        struct mount *mnt;
1018        int err;
1019
1020        mnt = alloc_vfsmnt(old->mnt_devname);
1021        if (!mnt)
1022                return ERR_PTR(-ENOMEM);
1023
1024        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1025                mnt->mnt_group_id = 0; /* not a peer of original */
1026        else
1027                mnt->mnt_group_id = old->mnt_group_id;
1028
1029        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1030                err = mnt_alloc_group_id(mnt);
1031                if (err)
1032                        goto out_free;
1033        }
1034
1035        mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1036        /* Don't allow unprivileged users to change mount flags */
1037        if (flag & CL_UNPRIVILEGED) {
1038                mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1039
1040                if (mnt->mnt.mnt_flags & MNT_READONLY)
1041                        mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1042
1043                if (mnt->mnt.mnt_flags & MNT_NODEV)
1044                        mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1045
1046                if (mnt->mnt.mnt_flags & MNT_NOSUID)
1047                        mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1048
1049                if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1050                        mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1051        }
1052
1053        /* Don't allow unprivileged users to reveal what is under a mount */
1054        if ((flag & CL_UNPRIVILEGED) &&
1055            (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1056                mnt->mnt.mnt_flags |= MNT_LOCKED;
1057
1058        atomic_inc(&sb->s_active);
1059        mnt->mnt.mnt_sb = sb;
1060        mnt->mnt.mnt_root = dget(root);
1061        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1062        mnt->mnt_parent = mnt;
1063        lock_mount_hash();
1064        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1065        unlock_mount_hash();
1066
1067        if ((flag & CL_SLAVE) ||
1068            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1069                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1070                mnt->mnt_master = old;
1071                CLEAR_MNT_SHARED(mnt);
1072        } else if (!(flag & CL_PRIVATE)) {
1073                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1074                        list_add(&mnt->mnt_share, &old->mnt_share);
1075                if (IS_MNT_SLAVE(old))
1076                        list_add(&mnt->mnt_slave, &old->mnt_slave);
1077                mnt->mnt_master = old->mnt_master;
1078        }
1079        if (flag & CL_MAKE_SHARED)
1080                set_mnt_shared(mnt);
1081
1082        /* stick the duplicate mount on the same expiry list
1083         * as the original if that was on one */
1084        if (flag & CL_EXPIRE) {
1085                if (!list_empty(&old->mnt_expire))
1086                        list_add(&mnt->mnt_expire, &old->mnt_expire);
1087        }
1088
1089        return mnt;
1090
1091 out_free:
1092        mnt_free_id(mnt);
1093        free_vfsmnt(mnt);
1094        return ERR_PTR(err);
1095}
1096
1097static void cleanup_mnt(struct mount *mnt)
1098{
1099        /*
1100         * This probably indicates that somebody messed
1101         * up a mnt_want/drop_write() pair.  If this
1102         * happens, the filesystem was probably unable
1103         * to make r/w->r/o transitions.
1104         */
1105        /*
1106         * The locking used to deal with mnt_count decrement provides barriers,
1107         * so mnt_get_writers() below is safe.
1108         */
1109        WARN_ON(mnt_get_writers(mnt));
1110        if (unlikely(mnt->mnt_pins.first))
1111                mnt_pin_kill(mnt);
1112        fsnotify_vfsmount_delete(&mnt->mnt);
1113        dput(mnt->mnt.mnt_root);
1114        deactivate_super(mnt->mnt.mnt_sb);
1115        mnt_free_id(mnt);
1116        call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1117}
1118
1119static void __cleanup_mnt(struct rcu_head *head)
1120{
1121        cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1122}
1123
1124static LLIST_HEAD(delayed_mntput_list);
1125static void delayed_mntput(struct work_struct *unused)
1126{
1127        struct llist_node *node = llist_del_all(&delayed_mntput_list);
1128        struct llist_node *next;
1129
1130        for (; node; node = next) {
1131                next = llist_next(node);
1132                cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1133        }
1134}
1135static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1136
1137static void mntput_no_expire(struct mount *mnt)
1138{
1139        rcu_read_lock();
1140        mnt_add_count(mnt, -1);
1141        if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1142                rcu_read_unlock();
1143                return;
1144        }
1145        lock_mount_hash();
1146        if (mnt_get_count(mnt)) {
1147                rcu_read_unlock();
1148                unlock_mount_hash();
1149                return;
1150        }
1151        if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1152                rcu_read_unlock();
1153                unlock_mount_hash();
1154                return;
1155        }
1156        mnt->mnt.mnt_flags |= MNT_DOOMED;
1157        rcu_read_unlock();
1158
1159        list_del(&mnt->mnt_instance);
1160
1161        if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1162                struct mount *p, *tmp;
1163                list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1164                        umount_mnt(p);
1165                }
1166        }
1167        unlock_mount_hash();
1168
1169        if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1170                struct task_struct *task = current;
1171                if (likely(!(task->flags & PF_KTHREAD))) {
1172                        init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1173                        if (!task_work_add(task, &mnt->mnt_rcu, true))
1174                                return;
1175                }
1176                if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1177                        schedule_delayed_work(&delayed_mntput_work, 1);
1178                return;
1179        }
1180        cleanup_mnt(mnt);
1181}
1182
1183void mntput(struct vfsmount *mnt)
1184{
1185        if (mnt) {
1186                struct mount *m = real_mount(mnt);
1187                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1188                if (unlikely(m->mnt_expiry_mark))
1189                        m->mnt_expiry_mark = 0;
1190                mntput_no_expire(m);
1191        }
1192}
1193EXPORT_SYMBOL(mntput);
1194
1195struct vfsmount *mntget(struct vfsmount *mnt)
1196{
1197        if (mnt)
1198                mnt_add_count(real_mount(mnt), 1);
1199        return mnt;
1200}
1201EXPORT_SYMBOL(mntget);
1202
1203struct vfsmount *mnt_clone_internal(struct path *path)
1204{
1205        struct mount *p;
1206        p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1207        if (IS_ERR(p))
1208                return ERR_CAST(p);
1209        p->mnt.mnt_flags |= MNT_INTERNAL;
1210        return &p->mnt;
1211}
1212
1213static inline void mangle(struct seq_file *m, const char *s)
1214{
1215        seq_escape(m, s, " \t\n\\");
1216}
1217
1218/*
1219 * Simple .show_options callback for filesystems which don't want to
1220 * implement more complex mount option showing.
1221 *
1222 * See also save_mount_options().
1223 */
1224int generic_show_options(struct seq_file *m, struct dentry *root)
1225{
1226        const char *options;
1227
1228        rcu_read_lock();
1229        options = rcu_dereference(root->d_sb->s_options);
1230
1231        if (options != NULL && options[0]) {
1232                seq_putc(m, ',');
1233                mangle(m, options);
1234        }
1235        rcu_read_unlock();
1236
1237        return 0;
1238}
1239EXPORT_SYMBOL(generic_show_options);
1240
1241/*
1242 * If filesystem uses generic_show_options(), this function should be
1243 * called from the fill_super() callback.
1244 *
1245 * The .remount_fs callback usually needs to be handled in a special
1246 * way, to make sure, that previous options are not overwritten if the
1247 * remount fails.
1248 *
1249 * Also note, that if the filesystem's .remount_fs function doesn't
1250 * reset all options to their default value, but changes only newly
1251 * given options, then the displayed options will not reflect reality
1252 * any more.
1253 */
1254void save_mount_options(struct super_block *sb, char *options)
1255{
1256        BUG_ON(sb->s_options);
1257        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1258}
1259EXPORT_SYMBOL(save_mount_options);
1260
1261void replace_mount_options(struct super_block *sb, char *options)
1262{
1263        char *old = sb->s_options;
1264        rcu_assign_pointer(sb->s_options, options);
1265        if (old) {
1266                synchronize_rcu();
1267                kfree(old);
1268        }
1269}
1270EXPORT_SYMBOL(replace_mount_options);
1271
1272#ifdef CONFIG_PROC_FS
1273/* iterator; we want it to have access to namespace_sem, thus here... */
1274static void *m_start(struct seq_file *m, loff_t *pos)
1275{
1276        struct proc_mounts *p = m->private;
1277
1278        down_read(&namespace_sem);
1279        if (p->cached_event == p->ns->event) {
1280                void *v = p->cached_mount;
1281                if (*pos == p->cached_index)
1282                        return v;
1283                if (*pos == p->cached_index + 1) {
1284                        v = seq_list_next(v, &p->ns->list, &p->cached_index);
1285                        return p->cached_mount = v;
1286                }
1287        }
1288
1289        p->cached_event = p->ns->event;
1290        p->cached_mount = seq_list_start(&p->ns->list, *pos);
1291        p->cached_index = *pos;
1292        return p->cached_mount;
1293}
1294
1295static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1296{
1297        struct proc_mounts *p = m->private;
1298
1299        p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1300        p->cached_index = *pos;
1301        return p->cached_mount;
1302}
1303
1304static void m_stop(struct seq_file *m, void *v)
1305{
1306        up_read(&namespace_sem);
1307}
1308
1309static int m_show(struct seq_file *m, void *v)
1310{
1311        struct proc_mounts *p = m->private;
1312        struct mount *r = list_entry(v, struct mount, mnt_list);
1313        return p->show(m, &r->mnt);
1314}
1315
1316const struct seq_operations mounts_op = {
1317        .start  = m_start,
1318        .next   = m_next,
1319        .stop   = m_stop,
1320        .show   = m_show,
1321};
1322#endif  /* CONFIG_PROC_FS */
1323
1324/**
1325 * may_umount_tree - check if a mount tree is busy
1326 * @mnt: root of mount tree
1327 *
1328 * This is called to check if a tree of mounts has any
1329 * open files, pwds, chroots or sub mounts that are
1330 * busy.
1331 */
1332int may_umount_tree(struct vfsmount *m)
1333{
1334        struct mount *mnt = real_mount(m);
1335        int actual_refs = 0;
1336        int minimum_refs = 0;
1337        struct mount *p;
1338        BUG_ON(!m);
1339
1340        /* write lock needed for mnt_get_count */
1341        lock_mount_hash();
1342        for (p = mnt; p; p = next_mnt(p, mnt)) {
1343                actual_refs += mnt_get_count(p);
1344                minimum_refs += 2;
1345        }
1346        unlock_mount_hash();
1347
1348        if (actual_refs > minimum_refs)
1349                return 0;
1350
1351        return 1;
1352}
1353
1354EXPORT_SYMBOL(may_umount_tree);
1355
1356/**
1357 * may_umount - check if a mount point is busy
1358 * @mnt: root of mount
1359 *
1360 * This is called to check if a mount point has any
1361 * open files, pwds, chroots or sub mounts. If the
1362 * mount has sub mounts this will return busy
1363 * regardless of whether the sub mounts are busy.
1364 *
1365 * Doesn't take quota and stuff into account. IOW, in some cases it will
1366 * give false negatives. The main reason why it's here is that we need
1367 * a non-destructive way to look for easily umountable filesystems.
1368 */
1369int may_umount(struct vfsmount *mnt)
1370{
1371        int ret = 1;
1372        down_read(&namespace_sem);
1373        lock_mount_hash();
1374        if (propagate_mount_busy(real_mount(mnt), 2))
1375                ret = 0;
1376        unlock_mount_hash();
1377        up_read(&namespace_sem);
1378        return ret;
1379}
1380
1381EXPORT_SYMBOL(may_umount);
1382
1383static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
1384
1385static void namespace_unlock(void)
1386{
1387        struct hlist_head head;
1388
1389        hlist_move_list(&unmounted, &head);
1390
1391        up_write(&namespace_sem);
1392
1393        if (likely(hlist_empty(&head)))
1394                return;
1395
1396        synchronize_rcu();
1397
1398        group_pin_kill(&head);
1399}
1400
1401static inline void namespace_lock(void)
1402{
1403        down_write(&namespace_sem);
1404}
1405
1406enum umount_tree_flags {
1407        UMOUNT_SYNC = 1,
1408        UMOUNT_PROPAGATE = 2,
1409        UMOUNT_CONNECTED = 4,
1410};
1411
1412static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1413{
1414        /* Leaving mounts connected is only valid for lazy umounts */
1415        if (how & UMOUNT_SYNC)
1416                return true;
1417
1418        /* A mount without a parent has nothing to be connected to */
1419        if (!mnt_has_parent(mnt))
1420                return true;
1421
1422        /* Because the reference counting rules change when mounts are
1423         * unmounted and connected, umounted mounts may not be
1424         * connected to mounted mounts.
1425         */
1426        if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1427                return true;
1428
1429        /* Has it been requested that the mount remain connected? */
1430        if (how & UMOUNT_CONNECTED)
1431                return false;
1432
1433        /* Is the mount locked such that it needs to remain connected? */
1434        if (IS_MNT_LOCKED(mnt))
1435                return false;
1436
1437        /* By default disconnect the mount */
1438        return true;
1439}
1440
1441/*
1442 * mount_lock must be held
1443 * namespace_sem must be held for write
1444 */
1445static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1446{
1447        LIST_HEAD(tmp_list);
1448        struct mount *p;
1449
1450        if (how & UMOUNT_PROPAGATE)
1451                propagate_mount_unlock(mnt);
1452
1453        /* Gather the mounts to umount */
1454        for (p = mnt; p; p = next_mnt(p, mnt)) {
1455                p->mnt.mnt_flags |= MNT_UMOUNT;
1456                list_move(&p->mnt_list, &tmp_list);
1457        }
1458
1459        /* Hide the mounts from mnt_mounts */
1460        list_for_each_entry(p, &tmp_list, mnt_list) {
1461                list_del_init(&p->mnt_child);
1462        }
1463
1464        /* Add propogated mounts to the tmp_list */
1465        if (how & UMOUNT_PROPAGATE)
1466                propagate_umount(&tmp_list);
1467
1468        while (!list_empty(&tmp_list)) {
1469                struct mnt_namespace *ns;
1470                bool disconnect;
1471                p = list_first_entry(&tmp_list, struct mount, mnt_list);
1472                list_del_init(&p->mnt_expire);
1473                list_del_init(&p->mnt_list);
1474                ns = p->mnt_ns;
1475                if (ns) {
1476                        ns->mounts--;
1477                        __touch_mnt_namespace(ns);
1478                }
1479                p->mnt_ns = NULL;
1480                if (how & UMOUNT_SYNC)
1481                        p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1482
1483                disconnect = disconnect_mount(p, how);
1484
1485                pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1486                                 disconnect ? &unmounted : NULL);
1487                if (mnt_has_parent(p)) {
1488                        mnt_add_count(p->mnt_parent, -1);
1489                        if (!disconnect) {
1490                                /* Don't forget about p */
1491                                list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1492                        } else {
1493                                umount_mnt(p);
1494                        }
1495                }
1496                change_mnt_propagation(p, MS_PRIVATE);
1497        }
1498}
1499
1500static void shrink_submounts(struct mount *mnt);
1501
1502static int do_umount(struct mount *mnt, int flags)
1503{
1504        struct super_block *sb = mnt->mnt.mnt_sb;
1505        int retval;
1506
1507        retval = security_sb_umount(&mnt->mnt, flags);
1508        if (retval)
1509                return retval;
1510
1511        /*
1512         * Allow userspace to request a mountpoint be expired rather than
1513         * unmounting unconditionally. Unmount only happens if:
1514         *  (1) the mark is already set (the mark is cleared by mntput())
1515         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1516         */
1517        if (flags & MNT_EXPIRE) {
1518                if (&mnt->mnt == current->fs->root.mnt ||
1519                    flags & (MNT_FORCE | MNT_DETACH))
1520                        return -EINVAL;
1521
1522                /*
1523                 * probably don't strictly need the lock here if we examined
1524                 * all race cases, but it's a slowpath.
1525                 */
1526                lock_mount_hash();
1527                if (mnt_get_count(mnt) != 2) {
1528                        unlock_mount_hash();
1529                        return -EBUSY;
1530                }
1531                unlock_mount_hash();
1532
1533                if (!xchg(&mnt->mnt_expiry_mark, 1))
1534                        return -EAGAIN;
1535        }
1536
1537        /*
1538         * If we may have to abort operations to get out of this
1539         * mount, and they will themselves hold resources we must
1540         * allow the fs to do things. In the Unix tradition of
1541         * 'Gee thats tricky lets do it in userspace' the umount_begin
1542         * might fail to complete on the first run through as other tasks
1543         * must return, and the like. Thats for the mount program to worry
1544         * about for the moment.
1545         */
1546
1547        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1548                sb->s_op->umount_begin(sb);
1549        }
1550
1551        /*
1552         * No sense to grab the lock for this test, but test itself looks
1553         * somewhat bogus. Suggestions for better replacement?
1554         * Ho-hum... In principle, we might treat that as umount + switch
1555         * to rootfs. GC would eventually take care of the old vfsmount.
1556         * Actually it makes sense, especially if rootfs would contain a
1557         * /reboot - static binary that would close all descriptors and
1558         * call reboot(9). Then init(8) could umount root and exec /reboot.
1559         */
1560        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1561                /*
1562                 * Special case for "unmounting" root ...
1563                 * we just try to remount it readonly.
1564                 */
1565                if (!capable(CAP_SYS_ADMIN))
1566                        return -EPERM;
1567                down_write(&sb->s_umount);
1568                if (!(sb->s_flags & MS_RDONLY))
1569                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1570                up_write(&sb->s_umount);
1571                return retval;
1572        }
1573
1574        namespace_lock();
1575        lock_mount_hash();
1576        event++;
1577
1578        if (flags & MNT_DETACH) {
1579                if (!list_empty(&mnt->mnt_list))
1580                        umount_tree(mnt, UMOUNT_PROPAGATE);
1581                retval = 0;
1582        } else {
1583                shrink_submounts(mnt);
1584                retval = -EBUSY;
1585                if (!propagate_mount_busy(mnt, 2)) {
1586                        if (!list_empty(&mnt->mnt_list))
1587                                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1588                        retval = 0;
1589                }
1590        }
1591        unlock_mount_hash();
1592        namespace_unlock();
1593        return retval;
1594}
1595
1596/*
1597 * __detach_mounts - lazily unmount all mounts on the specified dentry
1598 *
1599 * During unlink, rmdir, and d_drop it is possible to loose the path
1600 * to an existing mountpoint, and wind up leaking the mount.
1601 * detach_mounts allows lazily unmounting those mounts instead of
1602 * leaking them.
1603 *
1604 * The caller may hold dentry->d_inode->i_mutex.
1605 */
1606void __detach_mounts(struct dentry *dentry)
1607{
1608        struct mountpoint *mp;
1609        struct mount *mnt;
1610
1611        namespace_lock();
1612        lock_mount_hash();
1613        mp = lookup_mountpoint(dentry);
1614        if (IS_ERR_OR_NULL(mp))
1615                goto out_unlock;
1616
1617        event++;
1618        while (!hlist_empty(&mp->m_list)) {
1619                mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1620                if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1621                        hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1622                        umount_mnt(mnt);
1623                }
1624                else umount_tree(mnt, UMOUNT_CONNECTED);
1625        }
1626        put_mountpoint(mp);
1627out_unlock:
1628        unlock_mount_hash();
1629        namespace_unlock();
1630}
1631
1632/*
1633 * Is the caller allowed to modify his namespace?
1634 */
1635static inline bool may_mount(void)
1636{
1637        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1638}
1639
1640static inline bool may_mandlock(void)
1641{
1642#ifndef CONFIG_MANDATORY_FILE_LOCKING
1643        return false;
1644#endif
1645        return capable(CAP_SYS_ADMIN);
1646}
1647
1648/*
1649 * Now umount can handle mount points as well as block devices.
1650 * This is important for filesystems which use unnamed block devices.
1651 *
1652 * We now support a flag for forced unmount like the other 'big iron'
1653 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1654 */
1655
1656SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1657{
1658        struct path path;
1659        struct mount *mnt;
1660        int retval;
1661        int lookup_flags = 0;
1662
1663        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1664                return -EINVAL;
1665
1666        if (!may_mount())
1667                return -EPERM;
1668
1669        if (!(flags & UMOUNT_NOFOLLOW))
1670                lookup_flags |= LOOKUP_FOLLOW;
1671
1672        retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1673        if (retval)
1674                goto out;
1675        mnt = real_mount(path.mnt);
1676        retval = -EINVAL;
1677        if (path.dentry != path.mnt->mnt_root)
1678                goto dput_and_out;
1679        if (!check_mnt(mnt))
1680                goto dput_and_out;
1681        if (mnt->mnt.mnt_flags & MNT_LOCKED)
1682                goto dput_and_out;
1683        retval = -EPERM;
1684        if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1685                goto dput_and_out;
1686
1687        retval = do_umount(mnt, flags);
1688dput_and_out:
1689        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1690        dput(path.dentry);
1691        mntput_no_expire(mnt);
1692out:
1693        return retval;
1694}
1695
1696#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1697
1698/*
1699 *      The 2.0 compatible umount. No flags.
1700 */
1701SYSCALL_DEFINE1(oldumount, char __user *, name)
1702{
1703        return sys_umount(name, 0);
1704}
1705
1706#endif
1707
1708static bool is_mnt_ns_file(struct dentry *dentry)
1709{
1710        /* Is this a proxy for a mount namespace? */
1711        return dentry->d_op == &ns_dentry_operations &&
1712               dentry->d_fsdata == &mntns_operations;
1713}
1714
1715struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1716{
1717        return container_of(ns, struct mnt_namespace, ns);
1718}
1719
1720static bool mnt_ns_loop(struct dentry *dentry)
1721{
1722        /* Could bind mounting the mount namespace inode cause a
1723         * mount namespace loop?
1724         */
1725        struct mnt_namespace *mnt_ns;
1726        if (!is_mnt_ns_file(dentry))
1727                return false;
1728
1729        mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1730        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1731}
1732
1733struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1734                                        int flag)
1735{
1736        struct mount *res, *p, *q, *r, *parent;
1737
1738        if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1739                return ERR_PTR(-EINVAL);
1740
1741        if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1742                return ERR_PTR(-EINVAL);
1743
1744        res = q = clone_mnt(mnt, dentry, flag);
1745        if (IS_ERR(q))
1746                return q;
1747
1748        q->mnt_mountpoint = mnt->mnt_mountpoint;
1749
1750        p = mnt;
1751        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1752                struct mount *s;
1753                if (!is_subdir(r->mnt_mountpoint, dentry))
1754                        continue;
1755
1756                for (s = r; s; s = next_mnt(s, r)) {
1757                        if (!(flag & CL_COPY_UNBINDABLE) &&
1758                            IS_MNT_UNBINDABLE(s)) {
1759                                s = skip_mnt_tree(s);
1760                                continue;
1761                        }
1762                        if (!(flag & CL_COPY_MNT_NS_FILE) &&
1763                            is_mnt_ns_file(s->mnt.mnt_root)) {
1764                                s = skip_mnt_tree(s);
1765                                continue;
1766                        }
1767                        while (p != s->mnt_parent) {
1768                                p = p->mnt_parent;
1769                                q = q->mnt_parent;
1770                        }
1771                        p = s;
1772                        parent = q;
1773                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1774                        if (IS_ERR(q))
1775                                goto out;
1776                        lock_mount_hash();
1777                        list_add_tail(&q->mnt_list, &res->mnt_list);
1778                        attach_mnt(q, parent, p->mnt_mp);
1779                        unlock_mount_hash();
1780                }
1781        }
1782        return res;
1783out:
1784        if (res) {
1785                lock_mount_hash();
1786                umount_tree(res, UMOUNT_SYNC);
1787                unlock_mount_hash();
1788        }
1789        return q;
1790}
1791
1792/* Caller should check returned pointer for errors */
1793
1794struct vfsmount *collect_mounts(struct path *path)
1795{
1796        struct mount *tree;
1797        namespace_lock();
1798        if (!check_mnt(real_mount(path->mnt)))
1799                tree = ERR_PTR(-EINVAL);
1800        else
1801                tree = copy_tree(real_mount(path->mnt), path->dentry,
1802                                 CL_COPY_ALL | CL_PRIVATE);
1803        namespace_unlock();
1804        if (IS_ERR(tree))
1805                return ERR_CAST(tree);
1806        return &tree->mnt;
1807}
1808
1809void drop_collected_mounts(struct vfsmount *mnt)
1810{
1811        namespace_lock();
1812        lock_mount_hash();
1813        umount_tree(real_mount(mnt), UMOUNT_SYNC);
1814        unlock_mount_hash();
1815        namespace_unlock();
1816}
1817
1818/**
1819 * clone_private_mount - create a private clone of a path
1820 *
1821 * This creates a new vfsmount, which will be the clone of @path.  The new will
1822 * not be attached anywhere in the namespace and will be private (i.e. changes
1823 * to the originating mount won't be propagated into this).
1824 *
1825 * Release with mntput().
1826 */
1827struct vfsmount *clone_private_mount(struct path *path)
1828{
1829        struct mount *old_mnt = real_mount(path->mnt);
1830        struct mount *new_mnt;
1831
1832        if (IS_MNT_UNBINDABLE(old_mnt))
1833                return ERR_PTR(-EINVAL);
1834
1835        down_read(&namespace_sem);
1836        new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1837        up_read(&namespace_sem);
1838        if (IS_ERR(new_mnt))
1839                return ERR_CAST(new_mnt);
1840
1841        return &new_mnt->mnt;
1842}
1843EXPORT_SYMBOL_GPL(clone_private_mount);
1844
1845int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1846                   struct vfsmount *root)
1847{
1848        struct mount *mnt;
1849        int res = f(root, arg);
1850        if (res)
1851                return res;
1852        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1853                res = f(&mnt->mnt, arg);
1854                if (res)
1855                        return res;
1856        }
1857        return 0;
1858}
1859
1860static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1861{
1862        struct mount *p;
1863
1864        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1865                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1866                        mnt_release_group_id(p);
1867        }
1868}
1869
1870static int invent_group_ids(struct mount *mnt, bool recurse)
1871{
1872        struct mount *p;
1873
1874        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1875                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1876                        int err = mnt_alloc_group_id(p);
1877                        if (err) {
1878                                cleanup_group_ids(mnt, p);
1879                                return err;
1880                        }
1881                }
1882        }
1883
1884        return 0;
1885}
1886
1887int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1888{
1889        unsigned int max = READ_ONCE(sysctl_mount_max);
1890        unsigned int mounts = 0, old, pending, sum;
1891        struct mount *p;
1892
1893        for (p = mnt; p; p = next_mnt(p, mnt))
1894                mounts++;
1895
1896        old = ns->mounts;
1897        pending = ns->pending_mounts;
1898        sum = old + pending;
1899        if ((old > sum) ||
1900            (pending > sum) ||
1901            (max < sum) ||
1902            (mounts > (max - sum)))
1903                return -ENOSPC;
1904
1905        ns->pending_mounts = pending + mounts;
1906        return 0;
1907}
1908
1909/*
1910 *  @source_mnt : mount tree to be attached
1911 *  @nd         : place the mount tree @source_mnt is attached
1912 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1913 *                 store the parent mount and mountpoint dentry.
1914 *                 (done when source_mnt is moved)
1915 *
1916 *  NOTE: in the table below explains the semantics when a source mount
1917 *  of a given type is attached to a destination mount of a given type.
1918 * ---------------------------------------------------------------------------
1919 * |         BIND MOUNT OPERATION                                            |
1920 * |**************************************************************************
1921 * | source-->| shared        |       private  |       slave    | unbindable |
1922 * | dest     |               |                |                |            |
1923 * |   |      |               |                |                |            |
1924 * |   v      |               |                |                |            |
1925 * |**************************************************************************
1926 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1927 * |          |               |                |                |            |
1928 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1929 * ***************************************************************************
1930 * A bind operation clones the source mount and mounts the clone on the
1931 * destination mount.
1932 *
1933 * (++)  the cloned mount is propagated to all the mounts in the propagation
1934 *       tree of the destination mount and the cloned mount is added to
1935 *       the peer group of the source mount.
1936 * (+)   the cloned mount is created under the destination mount and is marked
1937 *       as shared. The cloned mount is added to the peer group of the source
1938 *       mount.
1939 * (+++) the mount is propagated to all the mounts in the propagation tree
1940 *       of the destination mount and the cloned mount is made slave
1941 *       of the same master as that of the source mount. The cloned mount
1942 *       is marked as 'shared and slave'.
1943 * (*)   the cloned mount is made a slave of the same master as that of the
1944 *       source mount.
1945 *
1946 * ---------------------------------------------------------------------------
1947 * |                    MOVE MOUNT OPERATION                                 |
1948 * |**************************************************************************
1949 * | source-->| shared        |       private  |       slave    | unbindable |
1950 * | dest     |               |                |                |            |
1951 * |   |      |               |                |                |            |
1952 * |   v      |               |                |                |            |
1953 * |**************************************************************************
1954 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1955 * |          |               |                |                |            |
1956 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1957 * ***************************************************************************
1958 *
1959 * (+)  the mount is moved to the destination. And is then propagated to
1960 *      all the mounts in the propagation tree of the destination mount.
1961 * (+*)  the mount is moved to the destination.
1962 * (+++)  the mount is moved to the destination and is then propagated to
1963 *      all the mounts belonging to the destination mount's propagation tree.
1964 *      the mount is marked as 'shared and slave'.
1965 * (*)  the mount continues to be a slave at the new location.
1966 *
1967 * if the source mount is a tree, the operations explained above is
1968 * applied to each mount in the tree.
1969 * Must be called without spinlocks held, since this function can sleep
1970 * in allocations.
1971 */
1972static int attach_recursive_mnt(struct mount *source_mnt,
1973                        struct mount *dest_mnt,
1974                        struct mountpoint *dest_mp,
1975                        struct path *parent_path)
1976{
1977        HLIST_HEAD(tree_list);
1978        struct mnt_namespace *ns = dest_mnt->mnt_ns;
1979        struct mountpoint *smp;
1980        struct mount *child, *p;
1981        struct hlist_node *n;
1982        int err;
1983
1984        /* Preallocate a mountpoint in case the new mounts need
1985         * to be tucked under other mounts.
1986         */
1987        smp = get_mountpoint(source_mnt->mnt.mnt_root);
1988        if (IS_ERR(smp))
1989                return PTR_ERR(smp);
1990
1991        /* Is there space to add these mounts to the mount namespace? */
1992        if (!parent_path) {
1993                err = count_mounts(ns, source_mnt);
1994                if (err)
1995                        goto out;
1996        }
1997
1998        if (IS_MNT_SHARED(dest_mnt)) {
1999                err = invent_group_ids(source_mnt, true);
2000                if (err)
2001                        goto out;
2002                err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2003                lock_mount_hash();
2004                if (err)
2005                        goto out_cleanup_ids;
2006                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2007                        set_mnt_shared(p);
2008        } else {
2009                lock_mount_hash();
2010        }
2011        if (parent_path) {
2012                detach_mnt(source_mnt, parent_path);
2013                attach_mnt(source_mnt, dest_mnt, dest_mp);
2014                touch_mnt_namespace(source_mnt->mnt_ns);
2015        } else {
2016                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2017                commit_tree(source_mnt);
2018        }
2019
2020        hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2021                struct mount *q;
2022                hlist_del_init(&child->mnt_hash);
2023                q = __lookup_mnt(&child->mnt_parent->mnt,
2024                                 child->mnt_mountpoint);
2025                if (q)
2026                        mnt_change_mountpoint(child, smp, q);
2027                commit_tree(child);
2028        }
2029        put_mountpoint(smp);
2030        unlock_mount_hash();
2031
2032        return 0;
2033
2034 out_cleanup_ids:
2035        while (!hlist_empty(&tree_list)) {
2036                child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2037                child->mnt_parent->mnt_ns->pending_mounts = 0;
2038                umount_tree(child, UMOUNT_SYNC);
2039        }
2040        unlock_mount_hash();
2041        cleanup_group_ids(source_mnt, NULL);
2042 out:
2043        ns->pending_mounts = 0;
2044
2045        read_seqlock_excl(&mount_lock);
2046        put_mountpoint(smp);
2047        read_sequnlock_excl(&mount_lock);
2048
2049        return err;
2050}
2051
2052static struct mountpoint *lock_mount(struct path *path)
2053{
2054        struct vfsmount *mnt;
2055        struct dentry *dentry = path->dentry;
2056retry:
2057        inode_lock(dentry->d_inode);
2058        if (unlikely(cant_mount(dentry))) {
2059                inode_unlock(dentry->d_inode);
2060                return ERR_PTR(-ENOENT);
2061        }
2062        namespace_lock();
2063        mnt = lookup_mnt(path);
2064        if (likely(!mnt)) {
2065                struct mountpoint *mp = get_mountpoint(dentry);
2066                if (IS_ERR(mp)) {
2067                        namespace_unlock();
2068                        inode_unlock(dentry->d_inode);
2069                        return mp;
2070                }
2071                return mp;
2072        }
2073        namespace_unlock();
2074        inode_unlock(path->dentry->d_inode);
2075        path_put(path);
2076        path->mnt = mnt;
2077        dentry = path->dentry = dget(mnt->mnt_root);
2078        goto retry;
2079}
2080
2081static void unlock_mount(struct mountpoint *where)
2082{
2083        struct dentry *dentry = where->m_dentry;
2084
2085        read_seqlock_excl(&mount_lock);
2086        put_mountpoint(where);
2087        read_sequnlock_excl(&mount_lock);
2088
2089        namespace_unlock();
2090        inode_unlock(dentry->d_inode);
2091}
2092
2093static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2094{
2095        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2096                return -EINVAL;
2097
2098        if (d_is_dir(mp->m_dentry) !=
2099              d_is_dir(mnt->mnt.mnt_root))
2100                return -ENOTDIR;
2101
2102        return attach_recursive_mnt(mnt, p, mp, NULL);
2103}
2104
2105/*
2106 * Sanity check the flags to change_mnt_propagation.
2107 */
2108
2109static int flags_to_propagation_type(int flags)
2110{
2111        int type = flags & ~(MS_REC | MS_SILENT);
2112
2113        /* Fail if any non-propagation flags are set */
2114        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2115                return 0;
2116        /* Only one propagation flag should be set */
2117        if (!is_power_of_2(type))
2118                return 0;
2119        return type;
2120}
2121
2122/*
2123 * recursively change the type of the mountpoint.
2124 */
2125static int do_change_type(struct path *path, int flag)
2126{
2127        struct mount *m;
2128        struct mount *mnt = real_mount(path->mnt);
2129        int recurse = flag & MS_REC;
2130        int type;
2131        int err = 0;
2132
2133        if (path->dentry != path->mnt->mnt_root)
2134                return -EINVAL;
2135
2136        type = flags_to_propagation_type(flag);
2137        if (!type)
2138                return -EINVAL;
2139
2140        namespace_lock();
2141        if (type == MS_SHARED) {
2142                err = invent_group_ids(mnt, recurse);
2143                if (err)
2144                        goto out_unlock;
2145        }
2146
2147        lock_mount_hash();
2148        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2149                change_mnt_propagation(m, type);
2150        unlock_mount_hash();
2151
2152 out_unlock:
2153        namespace_unlock();
2154        return err;
2155}
2156
2157static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2158{
2159        struct mount *child;
2160        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2161                if (!is_subdir(child->mnt_mountpoint, dentry))
2162                        continue;
2163
2164                if (child->mnt.mnt_flags & MNT_LOCKED)
2165                        return true;
2166        }
2167        return false;
2168}
2169
2170/*
2171 * do loopback mount.
2172 */
2173static int do_loopback(struct path *path, const char *old_name,
2174                                int recurse)
2175{
2176        struct path old_path;
2177        struct mount *mnt = NULL, *old, *parent;
2178        struct mountpoint *mp;
2179        int err;
2180        if (!old_name || !*old_name)
2181                return -EINVAL;
2182        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2183        if (err)
2184                return err;
2185
2186        err = -EINVAL;
2187        if (mnt_ns_loop(old_path.dentry))
2188                goto out;
2189
2190        mp = lock_mount(path);
2191        err = PTR_ERR(mp);
2192        if (IS_ERR(mp))
2193                goto out;
2194
2195        old = real_mount(old_path.mnt);
2196        parent = real_mount(path->mnt);
2197
2198        err = -EINVAL;
2199        if (IS_MNT_UNBINDABLE(old))
2200                goto out2;
2201
2202        if (!check_mnt(parent))
2203                goto out2;
2204
2205        if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2206                goto out2;
2207
2208        if (!recurse && has_locked_children(old, old_path.dentry))
2209                goto out2;
2210
2211        if (recurse)
2212                mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2213        else
2214                mnt = clone_mnt(old, old_path.dentry, 0);
2215
2216        if (IS_ERR(mnt)) {
2217                err = PTR_ERR(mnt);
2218                goto out2;
2219        }
2220
2221        mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2222
2223        err = graft_tree(mnt, parent, mp);
2224        if (err) {
2225                lock_mount_hash();
2226                umount_tree(mnt, UMOUNT_SYNC);
2227                unlock_mount_hash();
2228        }
2229out2:
2230        unlock_mount(mp);
2231out:
2232        path_put(&old_path);
2233        return err;
2234}
2235
2236static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2237{
2238        int error = 0;
2239        int readonly_request = 0;
2240
2241        if (ms_flags & MS_RDONLY)
2242                readonly_request = 1;
2243        if (readonly_request == __mnt_is_readonly(mnt))
2244                return 0;
2245
2246        if (readonly_request)
2247                error = mnt_make_readonly(real_mount(mnt));
2248        else
2249                __mnt_unmake_readonly(real_mount(mnt));
2250        return error;
2251}
2252
2253/*
2254 * change filesystem flags. dir should be a physical root of filesystem.
2255 * If you've mounted a non-root directory somewhere and want to do remount
2256 * on it - tough luck.
2257 */
2258static int do_remount(struct path *path, int flags, int mnt_flags,
2259                      void *data)
2260{
2261        int err;
2262        struct super_block *sb = path->mnt->mnt_sb;
2263        struct mount *mnt = real_mount(path->mnt);
2264
2265        if (!check_mnt(mnt))
2266                return -EINVAL;
2267
2268        if (path->dentry != path->mnt->mnt_root)
2269                return -EINVAL;
2270
2271        /* Don't allow changing of locked mnt flags.
2272         *
2273         * No locks need to be held here while testing the various
2274         * MNT_LOCK flags because those flags can never be cleared
2275         * once they are set.
2276         */
2277        if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2278            !(mnt_flags & MNT_READONLY)) {
2279                return -EPERM;
2280        }
2281        if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2282            !(mnt_flags & MNT_NODEV)) {
2283                return -EPERM;
2284        }
2285        if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2286            !(mnt_flags & MNT_NOSUID)) {
2287                return -EPERM;
2288        }
2289        if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2290            !(mnt_flags & MNT_NOEXEC)) {
2291                return -EPERM;
2292        }
2293        if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2294            ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2295                return -EPERM;
2296        }
2297
2298        err = security_sb_remount(sb, data);
2299        if (err)
2300                return err;
2301
2302        down_write(&sb->s_umount);
2303        if (flags & MS_BIND)
2304                err = change_mount_flags(path->mnt, flags);
2305        else if (!capable(CAP_SYS_ADMIN))
2306                err = -EPERM;
2307        else
2308                err = do_remount_sb(sb, flags, data, 0);
2309        if (!err) {
2310                lock_mount_hash();
2311                mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2312                mnt->mnt.mnt_flags = mnt_flags;
2313                touch_mnt_namespace(mnt->mnt_ns);
2314                unlock_mount_hash();
2315        }
2316        up_write(&sb->s_umount);
2317        return err;
2318}
2319
2320static inline int tree_contains_unbindable(struct mount *mnt)
2321{
2322        struct mount *p;
2323        for (p = mnt; p; p = next_mnt(p, mnt)) {
2324                if (IS_MNT_UNBINDABLE(p))
2325                        return 1;
2326        }
2327        return 0;
2328}
2329
2330static int do_move_mount(struct path *path, const char *old_name)
2331{
2332        struct path old_path, parent_path;
2333        struct mount *p;
2334        struct mount *old;
2335        struct mountpoint *mp;
2336        int err;
2337        if (!old_name || !*old_name)
2338                return -EINVAL;
2339        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2340        if (err)
2341                return err;
2342
2343        mp = lock_mount(path);
2344        err = PTR_ERR(mp);
2345        if (IS_ERR(mp))
2346                goto out;
2347
2348        old = real_mount(old_path.mnt);
2349        p = real_mount(path->mnt);
2350
2351        err = -EINVAL;
2352        if (!check_mnt(p) || !check_mnt(old))
2353                goto out1;
2354
2355        if (old->mnt.mnt_flags & MNT_LOCKED)
2356                goto out1;
2357
2358        err = -EINVAL;
2359        if (old_path.dentry != old_path.mnt->mnt_root)
2360                goto out1;
2361
2362        if (!mnt_has_parent(old))
2363                goto out1;
2364
2365        if (d_is_dir(path->dentry) !=
2366              d_is_dir(old_path.dentry))
2367                goto out1;
2368        /*
2369         * Don't move a mount residing in a shared parent.
2370         */
2371        if (IS_MNT_SHARED(old->mnt_parent))
2372                goto out1;
2373        /*
2374         * Don't move a mount tree containing unbindable mounts to a destination
2375         * mount which is shared.
2376         */
2377        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2378                goto out1;
2379        err = -ELOOP;
2380        for (; mnt_has_parent(p); p = p->mnt_parent)
2381                if (p == old)
2382                        goto out1;
2383
2384        err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2385        if (err)
2386                goto out1;
2387
2388        /* if the mount is moved, it should no longer be expire
2389         * automatically */
2390        list_del_init(&old->mnt_expire);
2391out1:
2392        unlock_mount(mp);
2393out:
2394        if (!err)
2395                path_put(&parent_path);
2396        path_put(&old_path);
2397        return err;
2398}
2399
2400static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2401{
2402        int err;
2403        const char *subtype = strchr(fstype, '.');
2404        if (subtype) {
2405                subtype++;
2406                err = -EINVAL;
2407                if (!subtype[0])
2408                        goto err;
2409        } else
2410                subtype = "";
2411
2412        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2413        err = -ENOMEM;
2414        if (!mnt->mnt_sb->s_subtype)
2415                goto err;
2416        return mnt;
2417
2418 err:
2419        mntput(mnt);
2420        return ERR_PTR(err);
2421}
2422
2423/*
2424 * add a mount into a namespace's mount tree
2425 */
2426static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2427{
2428        struct mountpoint *mp;
2429        struct mount *parent;
2430        int err;
2431
2432        mnt_flags &= ~MNT_INTERNAL_FLAGS;
2433
2434        mp = lock_mount(path);
2435        if (IS_ERR(mp))
2436                return PTR_ERR(mp);
2437
2438        parent = real_mount(path->mnt);
2439        err = -EINVAL;
2440        if (unlikely(!check_mnt(parent))) {
2441                /* that's acceptable only for automounts done in private ns */
2442                if (!(mnt_flags & MNT_SHRINKABLE))
2443                        goto unlock;
2444                /* ... and for those we'd better have mountpoint still alive */
2445                if (!parent->mnt_ns)
2446                        goto unlock;
2447        }
2448
2449        /* Refuse the same filesystem on the same mount point */
2450        err = -EBUSY;
2451        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2452            path->mnt->mnt_root == path->dentry)
2453                goto unlock;
2454
2455        err = -EINVAL;
2456        if (d_is_symlink(newmnt->mnt.mnt_root))
2457                goto unlock;
2458
2459        newmnt->mnt.mnt_flags = mnt_flags;
2460        err = graft_tree(newmnt, parent, mp);
2461
2462unlock:
2463        unlock_mount(mp);
2464        return err;
2465}
2466
2467static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2468
2469/*
2470 * create a new mount for userspace and request it to be added into the
2471 * namespace's tree
2472 */
2473static int do_new_mount(struct path *path, const char *fstype, int flags,
2474                        int mnt_flags, const char *name, void *data)
2475{
2476        struct file_system_type *type;
2477        struct vfsmount *mnt;
2478        int err;
2479
2480        if (!fstype)
2481                return -EINVAL;
2482
2483        type = get_fs_type(fstype);
2484        if (!type)
2485                return -ENODEV;
2486
2487        mnt = vfs_kern_mount(type, flags, name, data);
2488        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2489            !mnt->mnt_sb->s_subtype)
2490                mnt = fs_set_subtype(mnt, fstype);
2491
2492        put_filesystem(type);
2493        if (IS_ERR(mnt))
2494                return PTR_ERR(mnt);
2495
2496        if (mount_too_revealing(mnt, &mnt_flags)) {
2497                mntput(mnt);
2498                return -EPERM;
2499        }
2500
2501        err = do_add_mount(real_mount(mnt), path, mnt_flags);
2502        if (err)
2503                mntput(mnt);
2504        return err;
2505}
2506
2507int finish_automount(struct vfsmount *m, struct path *path)
2508{
2509        struct mount *mnt = real_mount(m);
2510        int err;
2511        /* The new mount record should have at least 2 refs to prevent it being
2512         * expired before we get a chance to add it
2513         */
2514        BUG_ON(mnt_get_count(mnt) < 2);
2515
2516        if (m->mnt_sb == path->mnt->mnt_sb &&
2517            m->mnt_root == path->dentry) {
2518                err = -ELOOP;
2519                goto fail;
2520        }
2521
2522        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2523        if (!err)
2524                return 0;
2525fail:
2526        /* remove m from any expiration list it may be on */
2527        if (!list_empty(&mnt->mnt_expire)) {
2528                namespace_lock();
2529                list_del_init(&mnt->mnt_expire);
2530                namespace_unlock();
2531        }
2532        mntput(m);
2533        mntput(m);
2534        return err;
2535}
2536
2537/**
2538 * mnt_set_expiry - Put a mount on an expiration list
2539 * @mnt: The mount to list.
2540 * @expiry_list: The list to add the mount to.
2541 */
2542void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2543{
2544        namespace_lock();
2545
2546        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2547
2548        namespace_unlock();
2549}
2550EXPORT_SYMBOL(mnt_set_expiry);
2551
2552/*
2553 * process a list of expirable mountpoints with the intent of discarding any
2554 * mountpoints that aren't in use and haven't been touched since last we came
2555 * here
2556 */
2557void mark_mounts_for_expiry(struct list_head *mounts)
2558{
2559        struct mount *mnt, *next;
2560        LIST_HEAD(graveyard);
2561
2562        if (list_empty(mounts))
2563                return;
2564
2565        namespace_lock();
2566        lock_mount_hash();
2567
2568        /* extract from the expiration list every vfsmount that matches the
2569         * following criteria:
2570         * - only referenced by its parent vfsmount
2571         * - still marked for expiry (marked on the last call here; marks are
2572         *   cleared by mntput())
2573         */
2574        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2575                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2576                        propagate_mount_busy(mnt, 1))
2577                        continue;
2578                list_move(&mnt->mnt_expire, &graveyard);
2579        }
2580        while (!list_empty(&graveyard)) {
2581                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2582                touch_mnt_namespace(mnt->mnt_ns);
2583                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2584        }
2585        unlock_mount_hash();
2586        namespace_unlock();
2587}
2588
2589EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2590
2591/*
2592 * Ripoff of 'select_parent()'
2593 *
2594 * search the list of submounts for a given mountpoint, and move any
2595 * shrinkable submounts to the 'graveyard' list.
2596 */
2597static int select_submounts(struct mount *parent, struct list_head *graveyard)
2598{
2599        struct mount *this_parent = parent;
2600        struct list_head *next;
2601        int found = 0;
2602
2603repeat:
2604        next = this_parent->mnt_mounts.next;
2605resume:
2606        while (next != &this_parent->mnt_mounts) {
2607                struct list_head *tmp = next;
2608                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2609
2610                next = tmp->next;
2611                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2612                        continue;
2613                /*
2614                 * Descend a level if the d_mounts list is non-empty.
2615                 */
2616                if (!list_empty(&mnt->mnt_mounts)) {
2617                        this_parent = mnt;
2618                        goto repeat;
2619                }
2620
2621                if (!propagate_mount_busy(mnt, 1)) {
2622                        list_move_tail(&mnt->mnt_expire, graveyard);
2623                        found++;
2624                }
2625        }
2626        /*
2627         * All done at this level ... ascend and resume the search
2628         */
2629        if (this_parent != parent) {
2630                next = this_parent->mnt_child.next;
2631                this_parent = this_parent->mnt_parent;
2632                goto resume;
2633        }
2634        return found;
2635}
2636
2637/*
2638 * process a list of expirable mountpoints with the intent of discarding any
2639 * submounts of a specific parent mountpoint
2640 *
2641 * mount_lock must be held for write
2642 */
2643static void shrink_submounts(struct mount *mnt)
2644{
2645        LIST_HEAD(graveyard);
2646        struct mount *m;
2647
2648        /* extract submounts of 'mountpoint' from the expiration list */
2649        while (select_submounts(mnt, &graveyard)) {
2650                while (!list_empty(&graveyard)) {
2651                        m = list_first_entry(&graveyard, struct mount,
2652                                                mnt_expire);
2653                        touch_mnt_namespace(m->mnt_ns);
2654                        umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2655                }
2656        }
2657}
2658
2659/*
2660 * Some copy_from_user() implementations do not return the exact number of
2661 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2662 * Note that this function differs from copy_from_user() in that it will oops
2663 * on bad values of `to', rather than returning a short copy.
2664 */
2665static long exact_copy_from_user(void *to, const void __user * from,
2666                                 unsigned long n)
2667{
2668        char *t = to;
2669        const char __user *f = from;
2670        char c;
2671
2672        if (!access_ok(VERIFY_READ, from, n))
2673                return n;
2674
2675        while (n) {
2676                if (__get_user(c, f)) {
2677                        memset(t, 0, n);
2678                        break;
2679                }
2680                *t++ = c;
2681                f++;
2682                n--;
2683        }
2684        return n;
2685}
2686
2687void *copy_mount_options(const void __user * data)
2688{
2689        int i;
2690        unsigned long size;
2691        char *copy;
2692
2693        if (!data)
2694                return NULL;
2695
2696        copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2697        if (!copy)
2698                return ERR_PTR(-ENOMEM);
2699
2700        /* We only care that *some* data at the address the user
2701         * gave us is valid.  Just in case, we'll zero
2702         * the remainder of the page.
2703         */
2704        /* copy_from_user cannot cross TASK_SIZE ! */
2705        size = TASK_SIZE - (unsigned long)data;
2706        if (size > PAGE_SIZE)
2707                size = PAGE_SIZE;
2708
2709        i = size - exact_copy_from_user(copy, data, size);
2710        if (!i) {
2711                kfree(copy);
2712                return ERR_PTR(-EFAULT);
2713        }
2714        if (i != PAGE_SIZE)
2715                memset(copy + i, 0, PAGE_SIZE - i);
2716        return copy;
2717}
2718
2719char *copy_mount_string(const void __user *data)
2720{
2721        return data ? strndup_user(data, PAGE_SIZE) : NULL;
2722}
2723
2724/*
2725 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2726 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2727 *
2728 * data is a (void *) that can point to any structure up to
2729 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2730 * information (or be NULL).
2731 *
2732 * Pre-0.97 versions of mount() didn't have a flags word.
2733 * When the flags word was introduced its top half was required
2734 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2735 * Therefore, if this magic number is present, it carries no information
2736 * and must be discarded.
2737 */
2738long do_mount(const char *dev_name, const char __user *dir_name,
2739                const char *type_page, unsigned long flags, void *data_page)
2740{
2741        struct path path;
2742        int retval = 0;
2743        int mnt_flags = 0;
2744
2745        /* Discard magic */
2746        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2747                flags &= ~MS_MGC_MSK;
2748
2749        /* Basic sanity checks */
2750        if (data_page)
2751                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2752
2753        /* ... and get the mountpoint */
2754        retval = user_path(dir_name, &path);
2755        if (retval)
2756                return retval;
2757
2758        retval = security_sb_mount(dev_name, &path,
2759                                   type_page, flags, data_page);
2760        if (!retval && !may_mount())
2761                retval = -EPERM;
2762        if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2763                retval = -EPERM;
2764        if (retval)
2765                goto dput_out;
2766
2767        /* Default to relatime unless overriden */
2768        if (!(flags & MS_NOATIME))
2769                mnt_flags |= MNT_RELATIME;
2770
2771        /* Separate the per-mountpoint flags */
2772        if (flags & MS_NOSUID)
2773                mnt_flags |= MNT_NOSUID;
2774        if (flags & MS_NODEV)
2775                mnt_flags |= MNT_NODEV;
2776        if (flags & MS_NOEXEC)
2777                mnt_flags |= MNT_NOEXEC;
2778        if (flags & MS_NOATIME)
2779                mnt_flags |= MNT_NOATIME;
2780        if (flags & MS_NODIRATIME)
2781                mnt_flags |= MNT_NODIRATIME;
2782        if (flags & MS_STRICTATIME)
2783                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2784        if (flags & MS_RDONLY)
2785                mnt_flags |= MNT_READONLY;
2786
2787        /* The default atime for remount is preservation */
2788        if ((flags & MS_REMOUNT) &&
2789            ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2790                       MS_STRICTATIME)) == 0)) {
2791                mnt_flags &= ~MNT_ATIME_MASK;
2792                mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2793        }
2794
2795        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2796                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2797                   MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT);
2798
2799        if (flags & MS_REMOUNT)
2800                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2801                                    data_page);
2802        else if (flags & MS_BIND)
2803                retval = do_loopback(&path, dev_name, flags & MS_REC);
2804        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2805                retval = do_change_type(&path, flags);
2806        else if (flags & MS_MOVE)
2807                retval = do_move_mount(&path, dev_name);
2808        else
2809                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2810                                      dev_name, data_page);
2811dput_out:
2812        path_put(&path);
2813        return retval;
2814}
2815
2816static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2817{
2818        return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2819}
2820
2821static void dec_mnt_namespaces(struct ucounts *ucounts)
2822{
2823        dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2824}
2825
2826static void free_mnt_ns(struct mnt_namespace *ns)
2827{
2828        ns_free_inum(&ns->ns);
2829        dec_mnt_namespaces(ns->ucounts);
2830        put_user_ns(ns->user_ns);
2831        kfree(ns);
2832}
2833
2834/*
2835 * Assign a sequence number so we can detect when we attempt to bind
2836 * mount a reference to an older mount namespace into the current
2837 * mount namespace, preventing reference counting loops.  A 64bit
2838 * number incrementing at 10Ghz will take 12,427 years to wrap which
2839 * is effectively never, so we can ignore the possibility.
2840 */
2841static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2842
2843static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2844{
2845        struct mnt_namespace *new_ns;
2846        struct ucounts *ucounts;
2847        int ret;
2848
2849        ucounts = inc_mnt_namespaces(user_ns);
2850        if (!ucounts)
2851                return ERR_PTR(-ENOSPC);
2852
2853        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2854        if (!new_ns) {
2855                dec_mnt_namespaces(ucounts);
2856                return ERR_PTR(-ENOMEM);
2857        }
2858        ret = ns_alloc_inum(&new_ns->ns);
2859        if (ret) {
2860                kfree(new_ns);
2861                dec_mnt_namespaces(ucounts);
2862                return ERR_PTR(ret);
2863        }
2864        new_ns->ns.ops = &mntns_operations;
2865        new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2866        atomic_set(&new_ns->count, 1);
2867        new_ns->root = NULL;
2868        INIT_LIST_HEAD(&new_ns->list);
2869        init_waitqueue_head(&new_ns->poll);
2870        new_ns->event = 0;
2871        new_ns->user_ns = get_user_ns(user_ns);
2872        new_ns->ucounts = ucounts;
2873        new_ns->mounts = 0;
2874        new_ns->pending_mounts = 0;
2875        return new_ns;
2876}
2877
2878__latent_entropy
2879struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2880                struct user_namespace *user_ns, struct fs_struct *new_fs)
2881{
2882        struct mnt_namespace *new_ns;
2883        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2884        struct mount *p, *q;
2885        struct mount *old;
2886        struct mount *new;
2887        int copy_flags;
2888
2889        BUG_ON(!ns);
2890
2891        if (likely(!(flags & CLONE_NEWNS))) {
2892                get_mnt_ns(ns);
2893                return ns;
2894        }
2895
2896        old = ns->root;
2897
2898        new_ns = alloc_mnt_ns(user_ns);
2899        if (IS_ERR(new_ns))
2900                return new_ns;
2901
2902        namespace_lock();
2903        /* First pass: copy the tree topology */
2904        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2905        if (user_ns != ns->user_ns)
2906                copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2907        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2908        if (IS_ERR(new)) {
2909                namespace_unlock();
2910                free_mnt_ns(new_ns);
2911                return ERR_CAST(new);
2912        }
2913        new_ns->root = new;
2914        list_add_tail(&new_ns->list, &new->mnt_list);
2915
2916        /*
2917         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2918         * as belonging to new namespace.  We have already acquired a private
2919         * fs_struct, so tsk->fs->lock is not needed.
2920         */
2921        p = old;
2922        q = new;
2923        while (p) {
2924                q->mnt_ns = new_ns;
2925                new_ns->mounts++;
2926                if (new_fs) {
2927                        if (&p->mnt == new_fs->root.mnt) {
2928                                new_fs->root.mnt = mntget(&q->mnt);
2929                                rootmnt = &p->mnt;
2930                        }
2931                        if (&p->mnt == new_fs->pwd.mnt) {
2932                                new_fs->pwd.mnt = mntget(&q->mnt);
2933                                pwdmnt = &p->mnt;
2934                        }
2935                }
2936                p = next_mnt(p, old);
2937                q = next_mnt(q, new);
2938                if (!q)
2939                        break;
2940                while (p->mnt.mnt_root != q->mnt.mnt_root)
2941                        p = next_mnt(p, old);
2942        }
2943        namespace_unlock();
2944
2945        if (rootmnt)
2946                mntput(rootmnt);
2947        if (pwdmnt)
2948                mntput(pwdmnt);
2949
2950        return new_ns;
2951}
2952
2953/**
2954 * create_mnt_ns - creates a private namespace and adds a root filesystem
2955 * @mnt: pointer to the new root filesystem mountpoint
2956 */
2957static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2958{
2959        struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2960        if (!IS_ERR(new_ns)) {
2961                struct mount *mnt = real_mount(m);
2962                mnt->mnt_ns = new_ns;
2963                new_ns->root = mnt;
2964                new_ns->mounts++;
2965                list_add(&mnt->mnt_list, &new_ns->list);
2966        } else {
2967                mntput(m);
2968        }
2969        return new_ns;
2970}
2971
2972struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2973{
2974        struct mnt_namespace *ns;
2975        struct super_block *s;
2976        struct path path;
2977        int err;
2978
2979        ns = create_mnt_ns(mnt);
2980        if (IS_ERR(ns))
2981                return ERR_CAST(ns);
2982
2983        err = vfs_path_lookup(mnt->mnt_root, mnt,
2984                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2985
2986        put_mnt_ns(ns);
2987
2988        if (err)
2989                return ERR_PTR(err);
2990
2991        /* trade a vfsmount reference for active sb one */
2992        s = path.mnt->mnt_sb;
2993        atomic_inc(&s->s_active);
2994        mntput(path.mnt);
2995        /* lock the sucker */
2996        down_write(&s->s_umount);
2997        /* ... and return the root of (sub)tree on it */
2998        return path.dentry;
2999}
3000EXPORT_SYMBOL(mount_subtree);
3001
3002SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3003                char __user *, type, unsigned long, flags, void __user *, data)
3004{
3005        int ret;
3006        char *kernel_type;
3007        char *kernel_dev;
3008        void *options;
3009
3010        kernel_type = copy_mount_string(type);
3011        ret = PTR_ERR(kernel_type);
3012        if (IS_ERR(kernel_type))
3013                goto out_type;
3014
3015        kernel_dev = copy_mount_string(dev_name);
3016        ret = PTR_ERR(kernel_dev);
3017        if (IS_ERR(kernel_dev))
3018                goto out_dev;
3019
3020        options = copy_mount_options(data);
3021        ret = PTR_ERR(options);
3022        if (IS_ERR(options))
3023                goto out_data;
3024
3025        ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3026
3027        kfree(options);
3028out_data:
3029        kfree(kernel_dev);
3030out_dev:
3031        kfree(kernel_type);
3032out_type:
3033        return ret;
3034}
3035
3036/*
3037 * Return true if path is reachable from root
3038 *
3039 * namespace_sem or mount_lock is held
3040 */
3041bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3042                         const struct path *root)
3043{
3044        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3045                dentry = mnt->mnt_mountpoint;
3046                mnt = mnt->mnt_parent;
3047        }
3048        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3049}
3050
3051bool path_is_under(struct path *path1, struct path *path2)
3052{
3053        bool res;
3054        read_seqlock_excl(&mount_lock);
3055        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3056        read_sequnlock_excl(&mount_lock);
3057        return res;
3058}
3059EXPORT_SYMBOL(path_is_under);
3060
3061/*
3062 * pivot_root Semantics:
3063 * Moves the root file system of the current process to the directory put_old,
3064 * makes new_root as the new root file system of the current process, and sets
3065 * root/cwd of all processes which had them on the current root to new_root.
3066 *
3067 * Restrictions:
3068 * The new_root and put_old must be directories, and  must not be on the
3069 * same file  system as the current process root. The put_old  must  be
3070 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3071 * pointed to by put_old must yield the same directory as new_root. No other
3072 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3073 *
3074 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3075 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3076 * in this situation.
3077 *
3078 * Notes:
3079 *  - we don't move root/cwd if they are not at the root (reason: if something
3080 *    cared enough to change them, it's probably wrong to force them elsewhere)
3081 *  - it's okay to pick a root that isn't the root of a file system, e.g.
3082 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3083 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3084 *    first.
3085 */
3086SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3087                const char __user *, put_old)
3088{
3089        struct path new, old, parent_path, root_parent, root;
3090        struct mount *new_mnt, *root_mnt, *old_mnt;
3091        struct mountpoint *old_mp, *root_mp;
3092        int error;
3093
3094        if (!may_mount())
3095                return -EPERM;
3096
3097        error = user_path_dir(new_root, &new);
3098        if (error)
3099                goto out0;
3100
3101        error = user_path_dir(put_old, &old);
3102        if (error)
3103                goto out1;
3104
3105        error = security_sb_pivotroot(&old, &new);
3106        if (error)
3107                goto out2;
3108
3109        get_fs_root(current->fs, &root);
3110        old_mp = lock_mount(&old);
3111        error = PTR_ERR(old_mp);
3112        if (IS_ERR(old_mp))
3113                goto out3;
3114
3115        error = -EINVAL;
3116        new_mnt = real_mount(new.mnt);
3117        root_mnt = real_mount(root.mnt);
3118        old_mnt = real_mount(old.mnt);
3119        if (IS_MNT_SHARED(old_mnt) ||
3120                IS_MNT_SHARED(new_mnt->mnt_parent) ||
3121                IS_MNT_SHARED(root_mnt->mnt_parent))
3122                goto out4;
3123        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3124                goto out4;
3125        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3126                goto out4;
3127        error = -ENOENT;
3128        if (d_unlinked(new.dentry))
3129                goto out4;
3130        error = -EBUSY;
3131        if (new_mnt == root_mnt || old_mnt == root_mnt)
3132                goto out4; /* loop, on the same file system  */
3133        error = -EINVAL;
3134        if (root.mnt->mnt_root != root.dentry)
3135                goto out4; /* not a mountpoint */
3136        if (!mnt_has_parent(root_mnt))
3137                goto out4; /* not attached */
3138        root_mp = root_mnt->mnt_mp;
3139        if (new.mnt->mnt_root != new.dentry)
3140                goto out4; /* not a mountpoint */
3141        if (!mnt_has_parent(new_mnt))
3142                goto out4; /* not attached */
3143        /* make sure we can reach put_old from new_root */
3144        if (!is_path_reachable(old_mnt, old.dentry, &new))
3145                goto out4;
3146        /* make certain new is below the root */
3147        if (!is_path_reachable(new_mnt, new.dentry, &root))
3148                goto out4;
3149        root_mp->m_count++; /* pin it so it won't go away */
3150        lock_mount_hash();
3151        detach_mnt(new_mnt, &parent_path);
3152        detach_mnt(root_mnt, &root_parent);
3153        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3154                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3155                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3156        }
3157        /* mount old root on put_old */
3158        attach_mnt(root_mnt, old_mnt, old_mp);
3159        /* mount new_root on / */
3160        attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3161        touch_mnt_namespace(current->nsproxy->mnt_ns);
3162        /* A moved mount should not expire automatically */
3163        list_del_init(&new_mnt->mnt_expire);
3164        put_mountpoint(root_mp);
3165        unlock_mount_hash();
3166        chroot_fs_refs(&root, &new);
3167        error = 0;
3168out4:
3169        unlock_mount(old_mp);
3170        if (!error) {
3171                path_put(&root_parent);
3172                path_put(&parent_path);
3173        }
3174out3:
3175        path_put(&root);
3176out2:
3177        path_put(&old);
3178out1:
3179        path_put(&new);
3180out0:
3181        return error;
3182}
3183
3184static void __init init_mount_tree(void)
3185{
3186        struct vfsmount *mnt;
3187        struct mnt_namespace *ns;
3188        struct path root;
3189        struct file_system_type *type;
3190
3191        type = get_fs_type("rootfs");
3192        if (!type)
3193                panic("Can't find rootfs type");
3194        mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3195        put_filesystem(type);
3196        if (IS_ERR(mnt))
3197                panic("Can't create rootfs");
3198
3199        ns = create_mnt_ns(mnt);
3200        if (IS_ERR(ns))
3201                panic("Can't allocate initial namespace");
3202
3203        init_task.nsproxy->mnt_ns = ns;
3204        get_mnt_ns(ns);
3205
3206        root.mnt = mnt;
3207        root.dentry = mnt->mnt_root;
3208        mnt->mnt_flags |= MNT_LOCKED;
3209
3210        set_fs_pwd(current->fs, &root);
3211        set_fs_root(current->fs, &root);
3212}
3213
3214void __init mnt_init(void)
3215{
3216        unsigned u;
3217        int err;
3218
3219        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3220                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3221
3222        mount_hashtable = alloc_large_system_hash("Mount-cache",
3223                                sizeof(struct hlist_head),
3224                                mhash_entries, 19,
3225                                0,
3226                                &m_hash_shift, &m_hash_mask, 0, 0);
3227        mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3228                                sizeof(struct hlist_head),
3229                                mphash_entries, 19,
3230                                0,
3231                                &mp_hash_shift, &mp_hash_mask, 0, 0);
3232
3233        if (!mount_hashtable || !mountpoint_hashtable)
3234                panic("Failed to allocate mount hash table\n");
3235
3236        for (u = 0; u <= m_hash_mask; u++)
3237                INIT_HLIST_HEAD(&mount_hashtable[u]);
3238        for (u = 0; u <= mp_hash_mask; u++)
3239                INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3240
3241        kernfs_init();
3242
3243        err = sysfs_init();
3244        if (err)
3245                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3246                        __func__, err);
3247        fs_kobj = kobject_create_and_add("fs", NULL);
3248        if (!fs_kobj)
3249                printk(KERN_WARNING "%s: kobj create error\n", __func__);
3250        init_rootfs();
3251        init_mount_tree();
3252}
3253
3254void put_mnt_ns(struct mnt_namespace *ns)
3255{
3256        if (!atomic_dec_and_test(&ns->count))
3257                return;
3258        drop_collected_mounts(&ns->root->mnt);
3259        free_mnt_ns(ns);
3260}
3261
3262struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3263{
3264        struct vfsmount *mnt;
3265        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3266        if (!IS_ERR(mnt)) {
3267                /*
3268                 * it is a longterm mount, don't release mnt until
3269                 * we unmount before file sys is unregistered
3270                */
3271                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3272        }
3273        return mnt;
3274}
3275EXPORT_SYMBOL_GPL(kern_mount_data);
3276
3277void kern_unmount(struct vfsmount *mnt)
3278{
3279        /* release long term mount so mount point can be released */
3280        if (!IS_ERR_OR_NULL(mnt)) {
3281                real_mount(mnt)->mnt_ns = NULL;
3282                synchronize_rcu();      /* yecchhh... */
3283                mntput(mnt);
3284        }
3285}
3286EXPORT_SYMBOL(kern_unmount);
3287
3288bool our_mnt(struct vfsmount *mnt)
3289{
3290        return check_mnt(real_mount(mnt));
3291}
3292
3293bool current_chrooted(void)
3294{
3295        /* Does the current process have a non-standard root */
3296        struct path ns_root;
3297        struct path fs_root;
3298        bool chrooted;
3299
3300        /* Find the namespace root */
3301        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3302        ns_root.dentry = ns_root.mnt->mnt_root;
3303        path_get(&ns_root);
3304        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3305                ;
3306
3307        get_fs_root(current->fs, &fs_root);
3308
3309        chrooted = !path_equal(&fs_root, &ns_root);
3310
3311        path_put(&fs_root);
3312        path_put(&ns_root);
3313
3314        return chrooted;
3315}
3316
3317static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3318                                int *new_mnt_flags)
3319{
3320        int new_flags = *new_mnt_flags;
3321        struct mount *mnt;
3322        bool visible = false;
3323
3324        down_read(&namespace_sem);
3325        list_for_each_entry(mnt, &ns->list, mnt_list) {
3326                struct mount *child;
3327                int mnt_flags;
3328
3329                if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3330                        continue;
3331
3332                /* This mount is not fully visible if it's root directory
3333                 * is not the root directory of the filesystem.
3334                 */
3335                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3336                        continue;
3337
3338                /* A local view of the mount flags */
3339                mnt_flags = mnt->mnt.mnt_flags;
3340
3341                /* Don't miss readonly hidden in the superblock flags */
3342                if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3343                        mnt_flags |= MNT_LOCK_READONLY;
3344
3345                /* Verify the mount flags are equal to or more permissive
3346                 * than the proposed new mount.
3347                 */
3348                if ((mnt_flags & MNT_LOCK_READONLY) &&
3349                    !(new_flags & MNT_READONLY))
3350                        continue;
3351                if ((mnt_flags & MNT_LOCK_ATIME) &&
3352                    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3353                        continue;
3354
3355                /* This mount is not fully visible if there are any
3356                 * locked child mounts that cover anything except for
3357                 * empty directories.
3358                 */
3359                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3360                        struct inode *inode = child->mnt_mountpoint->d_inode;
3361                        /* Only worry about locked mounts */
3362                        if (!(child->mnt.mnt_flags & MNT_LOCKED))
3363                                continue;
3364                        /* Is the directory permanetly empty? */
3365                        if (!is_empty_dir_inode(inode))
3366                                goto next;
3367                }
3368                /* Preserve the locked attributes */
3369                *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3370                                               MNT_LOCK_ATIME);
3371                visible = true;
3372                goto found;
3373        next:   ;
3374        }
3375found:
3376        up_read(&namespace_sem);
3377        return visible;
3378}
3379
3380static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3381{
3382        const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3383        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3384        unsigned long s_iflags;
3385
3386        if (ns->user_ns == &init_user_ns)
3387                return false;
3388
3389        /* Can this filesystem be too revealing? */
3390        s_iflags = mnt->mnt_sb->s_iflags;
3391        if (!(s_iflags & SB_I_USERNS_VISIBLE))
3392                return false;
3393
3394        if ((s_iflags & required_iflags) != required_iflags) {
3395                WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3396                          required_iflags);
3397                return true;
3398        }
3399
3400        return !mnt_already_visible(ns, mnt, new_mnt_flags);
3401}
3402
3403bool mnt_may_suid(struct vfsmount *mnt)
3404{
3405        /*
3406         * Foreign mounts (accessed via fchdir or through /proc
3407         * symlinks) are always treated as if they are nosuid.  This
3408         * prevents namespaces from trusting potentially unsafe
3409         * suid/sgid bits, file caps, or security labels that originate
3410         * in other namespaces.
3411         */
3412        return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3413               current_in_userns(mnt->mnt_sb->s_user_ns);
3414}
3415
3416static struct ns_common *mntns_get(struct task_struct *task)
3417{
3418        struct ns_common *ns = NULL;
3419        struct nsproxy *nsproxy;
3420
3421        task_lock(task);
3422        nsproxy = task->nsproxy;
3423        if (nsproxy) {
3424                ns = &nsproxy->mnt_ns->ns;
3425                get_mnt_ns(to_mnt_ns(ns));
3426        }
3427        task_unlock(task);
3428
3429        return ns;
3430}
3431
3432static void mntns_put(struct ns_common *ns)
3433{
3434        put_mnt_ns(to_mnt_ns(ns));
3435}
3436
3437static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3438{
3439        struct fs_struct *fs = current->fs;
3440        struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3441        struct path root;
3442
3443        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3444            !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3445            !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3446                return -EPERM;
3447
3448        if (fs->users != 1)
3449                return -EINVAL;
3450
3451        get_mnt_ns(mnt_ns);
3452        put_mnt_ns(nsproxy->mnt_ns);
3453        nsproxy->mnt_ns = mnt_ns;
3454
3455        /* Find the root */
3456        root.mnt    = &mnt_ns->root->mnt;
3457        root.dentry = mnt_ns->root->mnt.mnt_root;
3458        path_get(&root);
3459        while(d_mountpoint(root.dentry) && follow_down_one(&root))
3460                ;
3461
3462        /* Update the pwd and root */
3463        set_fs_pwd(fs, &root);
3464        set_fs_root(fs, &root);
3465
3466        path_put(&root);
3467        return 0;
3468}
3469
3470static struct user_namespace *mntns_owner(struct ns_common *ns)
3471{
3472        return to_mnt_ns(ns)->user_ns;
3473}
3474
3475const struct proc_ns_operations mntns_operations = {
3476        .name           = "mnt",
3477        .type           = CLONE_NEWNS,
3478        .get            = mntns_get,
3479        .put            = mntns_put,
3480        .install        = mntns_install,
3481        .owner          = mntns_owner,
3482};
Note: See TracBrowser for help on using the repository browser.