source: src/linux/universal/linux-4.4/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: 82.5 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
30static unsigned int m_hash_mask __read_mostly;
31static unsigned int m_hash_shift __read_mostly;
32static unsigned int mp_hash_mask __read_mostly;
33static unsigned int mp_hash_shift __read_mostly;
34
35static __initdata unsigned long mhash_entries;
36static int __init set_mhash_entries(char *str)
37{
38        if (!str)
39                return 0;
40        mhash_entries = simple_strtoul(str, &str, 0);
41        return 1;
42}
43__setup("mhash_entries=", set_mhash_entries);
44
45static __initdata unsigned long mphash_entries;
46static int __init set_mphash_entries(char *str)
47{
48        if (!str)
49                return 0;
50        mphash_entries = simple_strtoul(str, &str, 0);
51        return 1;
52}
53__setup("mphash_entries=", set_mphash_entries);
54
55static u64 event;
56static DEFINE_IDA(mnt_id_ida);
57static DEFINE_IDA(mnt_group_ida);
58static DEFINE_SPINLOCK(mnt_id_lock);
59static int mnt_id_start = 0;
60static int mnt_group_start = 1;
61
62static struct hlist_head *mount_hashtable __read_mostly;
63static struct hlist_head *mountpoint_hashtable __read_mostly;
64static struct kmem_cache *mnt_cache __read_mostly;
65static DECLARE_RWSEM(namespace_sem);
66
67/* /sys/fs */
68struct kobject *fs_kobj;
69EXPORT_SYMBOL_GPL(fs_kobj);
70
71/*
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
74 * up the tree.
75 *
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
78 */
79__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
80
81static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
82{
83        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85        tmp = tmp + (tmp >> m_hash_shift);
86        return &mount_hashtable[tmp & m_hash_mask];
87}
88
89static inline struct hlist_head *mp_hash(struct dentry *dentry)
90{
91        unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92        tmp = tmp + (tmp >> mp_hash_shift);
93        return &mountpoint_hashtable[tmp & mp_hash_mask];
94}
95
96/*
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
99 */
100static int mnt_alloc_id(struct mount *mnt)
101{
102        int res;
103
104retry:
105        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106        spin_lock(&mnt_id_lock);
107        res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
108        if (!res)
109                mnt_id_start = mnt->mnt_id + 1;
110        spin_unlock(&mnt_id_lock);
111        if (res == -EAGAIN)
112                goto retry;
113
114        return res;
115}
116
117static void mnt_free_id(struct mount *mnt)
118{
119        int id = mnt->mnt_id;
120        spin_lock(&mnt_id_lock);
121        ida_remove(&mnt_id_ida, id);
122        if (mnt_id_start > id)
123                mnt_id_start = id;
124        spin_unlock(&mnt_id_lock);
125}
126
127/*
128 * Allocate a new peer group ID
129 *
130 * mnt_group_ida is protected by namespace_sem
131 */
132static int mnt_alloc_group_id(struct mount *mnt)
133{
134        int res;
135
136        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
137                return -ENOMEM;
138
139        res = ida_get_new_above(&mnt_group_ida,
140                                mnt_group_start,
141                                &mnt->mnt_group_id);
142        if (!res)
143                mnt_group_start = mnt->mnt_group_id + 1;
144
145        return res;
146}
147
148/*
149 * Release a peer group ID
150 */
151void mnt_release_group_id(struct mount *mnt)
152{
153        int id = mnt->mnt_group_id;
154        ida_remove(&mnt_group_ida, id);
155        if (mnt_group_start > id)
156                mnt_group_start = id;
157        mnt->mnt_group_id = 0;
158}
159
160/*
161 * vfsmount lock must be held for read
162 */
163static inline void mnt_add_count(struct mount *mnt, int n)
164{
165#ifdef CONFIG_SMP
166        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
167#else
168        preempt_disable();
169        mnt->mnt_count += n;
170        preempt_enable();
171#endif
172}
173
174/*
175 * vfsmount lock must be held for write
176 */
177unsigned int mnt_get_count(struct mount *mnt)
178{
179#ifdef CONFIG_SMP
180        unsigned int count = 0;
181        int cpu;
182
183        for_each_possible_cpu(cpu) {
184                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
185        }
186
187        return count;
188#else
189        return mnt->mnt_count;
190#endif
191}
192
193static void drop_mountpoint(struct fs_pin *p)
194{
195        struct mount *m = container_of(p, struct mount, mnt_umount);
196        dput(m->mnt_ex_mountpoint);
197        pin_remove(p);
198        mntput(&m->mnt);
199}
200
201static struct mount *alloc_vfsmnt(const char *name)
202{
203        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
204        if (mnt) {
205                int err;
206
207                err = mnt_alloc_id(mnt);
208                if (err)
209                        goto out_free_cache;
210
211                if (name) {
212                        mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213                        if (!mnt->mnt_devname)
214                                goto out_free_id;
215                }
216
217#ifdef CONFIG_SMP
218                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219                if (!mnt->mnt_pcp)
220                        goto out_free_devname;
221
222                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
223#else
224                mnt->mnt_count = 1;
225                mnt->mnt_writers = 0;
226#endif
227
228                INIT_HLIST_NODE(&mnt->mnt_hash);
229                INIT_LIST_HEAD(&mnt->mnt_child);
230                INIT_LIST_HEAD(&mnt->mnt_mounts);
231                INIT_LIST_HEAD(&mnt->mnt_list);
232                INIT_LIST_HEAD(&mnt->mnt_expire);
233                INIT_LIST_HEAD(&mnt->mnt_share);
234                INIT_LIST_HEAD(&mnt->mnt_slave_list);
235                INIT_LIST_HEAD(&mnt->mnt_slave);
236                INIT_HLIST_NODE(&mnt->mnt_mp_list);
237#ifdef CONFIG_FSNOTIFY
238                INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239#endif
240                init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241        }
242        return mnt;
243
244#ifdef CONFIG_SMP
245out_free_devname:
246        kfree_const(mnt->mnt_devname);
247#endif
248out_free_id:
249        mnt_free_id(mnt);
250out_free_cache:
251        kmem_cache_free(mnt_cache, mnt);
252        return NULL;
253}
254
255/*
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
261 * a filesystem.
262 */
263/*
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
266 *
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*.  This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
272 * r/w.
273 */
274int __mnt_is_readonly(struct vfsmount *mnt)
275{
276        if (mnt->mnt_flags & MNT_READONLY)
277                return 1;
278        if (mnt->mnt_sb->s_flags & MS_RDONLY)
279                return 1;
280        return 0;
281}
282EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
284static inline void mnt_inc_writers(struct mount *mnt)
285{
286#ifdef CONFIG_SMP
287        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
288#else
289        mnt->mnt_writers++;
290#endif
291}
292
293static inline void mnt_dec_writers(struct mount *mnt)
294{
295#ifdef CONFIG_SMP
296        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
297#else
298        mnt->mnt_writers--;
299#endif
300}
301
302static unsigned int mnt_get_writers(struct mount *mnt)
303{
304#ifdef CONFIG_SMP
305        unsigned int count = 0;
306        int cpu;
307
308        for_each_possible_cpu(cpu) {
309                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310        }
311
312        return count;
313#else
314        return mnt->mnt_writers;
315#endif
316}
317
318static int mnt_is_readonly(struct vfsmount *mnt)
319{
320        if (mnt->mnt_sb->s_readonly_remount)
321                return 1;
322        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323        smp_rmb();
324        return __mnt_is_readonly(mnt);
325}
326
327/*
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink().  We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
332 */
333/**
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
336 *
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
342 */
343int __mnt_want_write(struct vfsmount *m)
344{
345        struct mount *mnt = real_mount(m);
346        int ret = 0;
347
348        preempt_disable();
349        mnt_inc_writers(mnt);
350        /*
351         * The store to mnt_inc_writers must be visible before we pass
352         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353         * incremented count after it has set MNT_WRITE_HOLD.
354         */
355        smp_mb();
356        while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
357                cpu_relax();
358        /*
359         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360         * be set to match its requirements. So we must not load that until
361         * MNT_WRITE_HOLD is cleared.
362         */
363        smp_rmb();
364        if (mnt_is_readonly(m)) {
365                mnt_dec_writers(mnt);
366                ret = -EROFS;
367        }
368        preempt_enable();
369
370        return ret;
371}
372
373/**
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
376 *
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success.  When the write operation is
380 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
381 */
382int mnt_want_write(struct vfsmount *m)
383{
384        int ret;
385
386        sb_start_write(m->mnt_sb);
387        ret = __mnt_want_write(m);
388        if (ret)
389                sb_end_write(m->mnt_sb);
390        return ret;
391}
392EXPORT_SYMBOL_GPL(mnt_want_write);
393
394/**
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
397 *
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
402 *
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
405 */
406int mnt_clone_write(struct vfsmount *mnt)
407{
408        /* superblock may be r/o */
409        if (__mnt_is_readonly(mnt))
410                return -EROFS;
411        preempt_disable();
412        mnt_inc_writers(real_mount(mnt));
413        preempt_enable();
414        return 0;
415}
416EXPORT_SYMBOL_GPL(mnt_clone_write);
417
418/**
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
421 *
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
424 */
425int __mnt_want_write_file(struct file *file)
426{
427        if (!(file->f_mode & FMODE_WRITER))
428                return __mnt_want_write(file->f_path.mnt);
429        else
430                return mnt_clone_write(file->f_path.mnt);
431}
432
433/**
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
436 *
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
439 */
440int mnt_want_write_file(struct file *file)
441{
442        int ret;
443
444        sb_start_write(file->f_path.mnt->mnt_sb);
445        ret = __mnt_want_write_file(file);
446        if (ret)
447                sb_end_write(file->f_path.mnt->mnt_sb);
448        return ret;
449}
450EXPORT_SYMBOL_GPL(mnt_want_write_file);
451
452/**
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
455 *
456 * Tells the low-level filesystem that we are done
457 * performing writes to it.  Must be matched with
458 * __mnt_want_write() call above.
459 */
460void __mnt_drop_write(struct vfsmount *mnt)
461{
462        preempt_disable();
463        mnt_dec_writers(real_mount(mnt));
464        preempt_enable();
465}
466
467/**
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
470 *
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again.  Must be matched with
473 * mnt_want_write() call above.
474 */
475void mnt_drop_write(struct vfsmount *mnt)
476{
477        __mnt_drop_write(mnt);
478        sb_end_write(mnt->mnt_sb);
479}
480EXPORT_SYMBOL_GPL(mnt_drop_write);
481
482void __mnt_drop_write_file(struct file *file)
483{
484        __mnt_drop_write(file->f_path.mnt);
485}
486
487void mnt_drop_write_file(struct file *file)
488{
489        mnt_drop_write(file->f_path.mnt);
490}
491EXPORT_SYMBOL(mnt_drop_write_file);
492
493static int mnt_make_readonly(struct mount *mnt)
494{
495        int ret = 0;
496
497        lock_mount_hash();
498        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
499        /*
500         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501         * should be visible before we do.
502         */
503        smp_mb();
504
505        /*
506         * With writers on hold, if this value is zero, then there are
507         * definitely no active writers (although held writers may subsequently
508         * increment the count, they'll have to wait, and decrement it after
509         * seeing MNT_READONLY).
510         *
511         * It is OK to have counter incremented on one CPU and decremented on
512         * another: the sum will add up correctly. The danger would be when we
513         * sum up each counter, if we read a counter before it is incremented,
514         * but then read another CPU's count which it has been subsequently
515         * decremented from -- we would see more decrements than we should.
516         * MNT_WRITE_HOLD protects against this scenario, because
517         * mnt_want_write first increments count, then smp_mb, then spins on
518         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519         * we're counting up here.
520         */
521        if (mnt_get_writers(mnt) > 0)
522                ret = -EBUSY;
523        else
524                mnt->mnt.mnt_flags |= MNT_READONLY;
525        /*
526         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527         * that become unheld will see MNT_READONLY.
528         */
529        smp_wmb();
530        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
531        unlock_mount_hash();
532        return ret;
533}
534
535static void __mnt_unmake_readonly(struct mount *mnt)
536{
537        lock_mount_hash();
538        mnt->mnt.mnt_flags &= ~MNT_READONLY;
539        unlock_mount_hash();
540}
541
542int sb_prepare_remount_readonly(struct super_block *sb)
543{
544        struct mount *mnt;
545        int err = 0;
546
547        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
548        if (atomic_long_read(&sb->s_remove_count))
549                return -EBUSY;
550
551        lock_mount_hash();
552        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555                        smp_mb();
556                        if (mnt_get_writers(mnt) > 0) {
557                                err = -EBUSY;
558                                break;
559                        }
560                }
561        }
562        if (!err && atomic_long_read(&sb->s_remove_count))
563                err = -EBUSY;
564
565        if (!err) {
566                sb->s_readonly_remount = 1;
567                smp_wmb();
568        }
569        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
572        }
573        unlock_mount_hash();
574
575        return err;
576}
577
578static void free_vfsmnt(struct mount *mnt)
579{
580        kfree_const(mnt->mnt_devname);
581#ifdef CONFIG_SMP
582        free_percpu(mnt->mnt_pcp);
583#endif
584        kmem_cache_free(mnt_cache, mnt);
585}
586
587static void delayed_free_vfsmnt(struct rcu_head *head)
588{
589        free_vfsmnt(container_of(head, struct mount, mnt_rcu));
590}
591
592/* call under rcu_read_lock */
593int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
594{
595        struct mount *mnt;
596        if (read_seqretry(&mount_lock, seq))
597                return 1;
598        if (bastard == NULL)
599                return 0;
600        mnt = real_mount(bastard);
601        mnt_add_count(mnt, 1);
602        if (likely(!read_seqretry(&mount_lock, seq)))
603                return 0;
604        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605                mnt_add_count(mnt, -1);
606                return 1;
607        }
608        return -1;
609}
610
611/* call under rcu_read_lock */
612bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
613{
614        int res = __legitimize_mnt(bastard, seq);
615        if (likely(!res))
616                return true;
617        if (unlikely(res < 0)) {
618                rcu_read_unlock();
619                mntput(bastard);
620                rcu_read_lock();
621        }
622        return false;
623}
624
625/*
626 * find the first mount at @dentry on vfsmount @mnt.
627 * call under rcu_read_lock()
628 */
629struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
630{
631        struct hlist_head *head = m_hash(mnt, dentry);
632        struct mount *p;
633
634        hlist_for_each_entry_rcu(p, head, mnt_hash)
635                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
636                        return p;
637        return NULL;
638}
639
640/*
641 * lookup_mnt - Return the first child mount mounted at path
642 *
643 * "First" means first mounted chronologically.  If you create the
644 * following mounts:
645 *
646 * mount /dev/sda1 /mnt
647 * mount /dev/sda2 /mnt
648 * mount /dev/sda3 /mnt
649 *
650 * Then lookup_mnt() on the base /mnt dentry in the root mount will
651 * return successively the root dentry and vfsmount of /dev/sda1, then
652 * /dev/sda2, then /dev/sda3, then NULL.
653 *
654 * lookup_mnt takes a reference to the found vfsmount.
655 */
656struct vfsmount *lookup_mnt(struct path *path)
657{
658        struct mount *child_mnt;
659        struct vfsmount *m;
660        unsigned seq;
661
662        rcu_read_lock();
663        do {
664                seq = read_seqbegin(&mount_lock);
665                child_mnt = __lookup_mnt(path->mnt, path->dentry);
666                m = child_mnt ? &child_mnt->mnt : NULL;
667        } while (!legitimize_mnt(m, seq));
668        rcu_read_unlock();
669        return m;
670}
671
672/*
673 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
674 *                         current mount namespace.
675 *
676 * The common case is dentries are not mountpoints at all and that
677 * test is handled inline.  For the slow case when we are actually
678 * dealing with a mountpoint of some kind, walk through all of the
679 * mounts in the current mount namespace and test to see if the dentry
680 * is a mountpoint.
681 *
682 * The mount_hashtable is not usable in the context because we
683 * need to identify all mounts that may be in the current mount
684 * namespace not just a mount that happens to have some specified
685 * parent mount.
686 */
687bool __is_local_mountpoint(struct dentry *dentry)
688{
689        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
690        struct mount *mnt;
691        bool is_covered = false;
692
693        if (!d_mountpoint(dentry))
694                goto out;
695
696        down_read(&namespace_sem);
697        list_for_each_entry(mnt, &ns->list, mnt_list) {
698                is_covered = (mnt->mnt_mountpoint == dentry);
699                if (is_covered)
700                        break;
701        }
702        up_read(&namespace_sem);
703out:
704        return is_covered;
705}
706
707static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
708{
709        struct hlist_head *chain = mp_hash(dentry);
710        struct mountpoint *mp;
711
712        hlist_for_each_entry(mp, chain, m_hash) {
713                if (mp->m_dentry == dentry) {
714                        /* might be worth a WARN_ON() */
715                        if (d_unlinked(dentry))
716                                return ERR_PTR(-ENOENT);
717                        mp->m_count++;
718                        return mp;
719                }
720        }
721        return NULL;
722}
723
724static struct mountpoint *get_mountpoint(struct dentry *dentry)
725{
726        struct mountpoint *mp, *new = NULL;
727        int ret;
728
729        if (d_mountpoint(dentry)) {
730mountpoint:
731                read_seqlock_excl(&mount_lock);
732                mp = lookup_mountpoint(dentry);
733                read_sequnlock_excl(&mount_lock);
734                if (mp)
735                        goto done;
736        }
737
738        if (!new)
739                new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
740        if (!new)
741                return ERR_PTR(-ENOMEM);
742
743
744        /* Exactly one processes may set d_mounted */
745        ret = d_set_mounted(dentry);
746
747        /* Someone else set d_mounted? */
748        if (ret == -EBUSY)
749                goto mountpoint;
750
751        /* The dentry is not available as a mountpoint? */
752        mp = ERR_PTR(ret);
753        if (ret)
754                goto done;
755
756        /* Add the new mountpoint to the hash table */
757        read_seqlock_excl(&mount_lock);
758        new->m_dentry = dentry;
759        new->m_count = 1;
760        hlist_add_head(&new->m_hash, mp_hash(dentry));
761        INIT_HLIST_HEAD(&new->m_list);
762        read_sequnlock_excl(&mount_lock);
763
764        mp = new;
765        new = NULL;
766done:
767        kfree(new);
768        return mp;
769}
770
771static void put_mountpoint(struct mountpoint *mp)
772{
773        if (!--mp->m_count) {
774                struct dentry *dentry = mp->m_dentry;
775                BUG_ON(!hlist_empty(&mp->m_list));
776                spin_lock(&dentry->d_lock);
777                dentry->d_flags &= ~DCACHE_MOUNTED;
778                spin_unlock(&dentry->d_lock);
779                hlist_del(&mp->m_hash);
780                kfree(mp);
781        }
782}
783
784static inline int check_mnt(struct mount *mnt)
785{
786        return mnt->mnt_ns == current->nsproxy->mnt_ns;
787}
788
789/*
790 * vfsmount lock must be held for write
791 */
792static void touch_mnt_namespace(struct mnt_namespace *ns)
793{
794        if (ns) {
795                ns->event = ++event;
796                wake_up_interruptible(&ns->poll);
797        }
798}
799
800/*
801 * vfsmount lock must be held for write
802 */
803static void __touch_mnt_namespace(struct mnt_namespace *ns)
804{
805        if (ns && ns->event != event) {
806                ns->event = event;
807                wake_up_interruptible(&ns->poll);
808        }
809}
810
811/*
812 * vfsmount lock must be held for write
813 */
814static void unhash_mnt(struct mount *mnt)
815{
816        mnt->mnt_parent = mnt;
817        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
818        list_del_init(&mnt->mnt_child);
819        hlist_del_init_rcu(&mnt->mnt_hash);
820        hlist_del_init(&mnt->mnt_mp_list);
821        put_mountpoint(mnt->mnt_mp);
822        mnt->mnt_mp = NULL;
823}
824
825/*
826 * vfsmount lock must be held for write
827 */
828static void detach_mnt(struct mount *mnt, struct path *old_path)
829{
830        old_path->dentry = mnt->mnt_mountpoint;
831        old_path->mnt = &mnt->mnt_parent->mnt;
832        unhash_mnt(mnt);
833}
834
835/*
836 * vfsmount lock must be held for write
837 */
838static void umount_mnt(struct mount *mnt)
839{
840        /* old mountpoint will be dropped when we can do that */
841        mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
842        unhash_mnt(mnt);
843}
844
845/*
846 * vfsmount lock must be held for write
847 */
848void mnt_set_mountpoint(struct mount *mnt,
849                        struct mountpoint *mp,
850                        struct mount *child_mnt)
851{
852        mp->m_count++;
853        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
854        child_mnt->mnt_mountpoint = dget(mp->m_dentry);
855        child_mnt->mnt_parent = mnt;
856        child_mnt->mnt_mp = mp;
857        hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
858}
859
860static void __attach_mnt(struct mount *mnt, struct mount *parent)
861{
862        hlist_add_head_rcu(&mnt->mnt_hash,
863                           m_hash(&parent->mnt, mnt->mnt_mountpoint));
864        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
865}
866
867/*
868 * vfsmount lock must be held for write
869 */
870static void attach_mnt(struct mount *mnt,
871                        struct mount *parent,
872                        struct mountpoint *mp)
873{
874        mnt_set_mountpoint(parent, mp, mnt);
875        __attach_mnt(mnt, parent);
876}
877
878void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
879{
880        struct mountpoint *old_mp = mnt->mnt_mp;
881        struct dentry *old_mountpoint = mnt->mnt_mountpoint;
882        struct mount *old_parent = mnt->mnt_parent;
883
884        list_del_init(&mnt->mnt_child);
885        hlist_del_init(&mnt->mnt_mp_list);
886        hlist_del_init_rcu(&mnt->mnt_hash);
887
888        attach_mnt(mnt, parent, mp);
889
890        put_mountpoint(old_mp);
891
892        /*
893         * Safely avoid even the suggestion this code might sleep or
894         * lock the mount hash by taking advantage of the knowledge that
895         * mnt_change_mountpoint will not release the final reference
896         * to a mountpoint.
897         *
898         * During mounting, the mount passed in as the parent mount will
899         * continue to use the old mountpoint and during unmounting, the
900         * old mountpoint will continue to exist until namespace_unlock,
901         * which happens well after mnt_change_mountpoint.
902         */
903        spin_lock(&old_mountpoint->d_lock);
904        old_mountpoint->d_lockref.count--;
905        spin_unlock(&old_mountpoint->d_lock);
906
907        mnt_add_count(old_parent, -1);
908}
909
910/*
911 * vfsmount lock must be held for write
912 */
913static void commit_tree(struct mount *mnt)
914{
915        struct mount *parent = mnt->mnt_parent;
916        struct mount *m;
917        LIST_HEAD(head);
918        struct mnt_namespace *n = parent->mnt_ns;
919
920        BUG_ON(parent == mnt);
921
922        list_add_tail(&head, &mnt->mnt_list);
923        list_for_each_entry(m, &head, mnt_list)
924                m->mnt_ns = n;
925
926        list_splice(&head, n->list.prev);
927
928        __attach_mnt(mnt, parent);
929        touch_mnt_namespace(n);
930}
931
932static struct mount *next_mnt(struct mount *p, struct mount *root)
933{
934        struct list_head *next = p->mnt_mounts.next;
935        if (next == &p->mnt_mounts) {
936                while (1) {
937                        if (p == root)
938                                return NULL;
939                        next = p->mnt_child.next;
940                        if (next != &p->mnt_parent->mnt_mounts)
941                                break;
942                        p = p->mnt_parent;
943                }
944        }
945        return list_entry(next, struct mount, mnt_child);
946}
947
948static struct mount *skip_mnt_tree(struct mount *p)
949{
950        struct list_head *prev = p->mnt_mounts.prev;
951        while (prev != &p->mnt_mounts) {
952                p = list_entry(prev, struct mount, mnt_child);
953                prev = p->mnt_mounts.prev;
954        }
955        return p;
956}
957
958struct vfsmount *
959vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
960{
961        struct mount *mnt;
962        struct dentry *root;
963
964        if (!type)
965                return ERR_PTR(-ENODEV);
966
967        mnt = alloc_vfsmnt(name);
968        if (!mnt)
969                return ERR_PTR(-ENOMEM);
970
971        if (flags & MS_KERNMOUNT)
972                mnt->mnt.mnt_flags = MNT_INTERNAL;
973
974        root = mount_fs(type, flags, name, data);
975        if (IS_ERR(root)) {
976                mnt_free_id(mnt);
977                free_vfsmnt(mnt);
978                return ERR_CAST(root);
979        }
980
981        mnt->mnt.mnt_root = root;
982        mnt->mnt.mnt_sb = root->d_sb;
983        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
984        mnt->mnt_parent = mnt;
985        lock_mount_hash();
986        list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
987        unlock_mount_hash();
988        return &mnt->mnt;
989}
990EXPORT_SYMBOL_GPL(vfs_kern_mount);
991
992static struct mount *clone_mnt(struct mount *old, struct dentry *root,
993                                        int flag)
994{
995        struct super_block *sb = old->mnt.mnt_sb;
996        struct mount *mnt;
997        int err;
998
999        mnt = alloc_vfsmnt(old->mnt_devname);
1000        if (!mnt)
1001                return ERR_PTR(-ENOMEM);
1002
1003        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1004                mnt->mnt_group_id = 0; /* not a peer of original */
1005        else
1006                mnt->mnt_group_id = old->mnt_group_id;
1007
1008        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1009                err = mnt_alloc_group_id(mnt);
1010                if (err)
1011                        goto out_free;
1012        }
1013
1014        mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1015        /* Don't allow unprivileged users to change mount flags */
1016        if (flag & CL_UNPRIVILEGED) {
1017                mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1018
1019                if (mnt->mnt.mnt_flags & MNT_READONLY)
1020                        mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1021
1022                if (mnt->mnt.mnt_flags & MNT_NODEV)
1023                        mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1024
1025                if (mnt->mnt.mnt_flags & MNT_NOSUID)
1026                        mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1027
1028                if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1029                        mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1030        }
1031
1032        /* Don't allow unprivileged users to reveal what is under a mount */
1033        if ((flag & CL_UNPRIVILEGED) &&
1034            (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1035                mnt->mnt.mnt_flags |= MNT_LOCKED;
1036
1037        atomic_inc(&sb->s_active);
1038        mnt->mnt.mnt_sb = sb;
1039        mnt->mnt.mnt_root = dget(root);
1040        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1041        mnt->mnt_parent = mnt;
1042        lock_mount_hash();
1043        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1044        unlock_mount_hash();
1045
1046        if ((flag & CL_SLAVE) ||
1047            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1048                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1049                mnt->mnt_master = old;
1050                CLEAR_MNT_SHARED(mnt);
1051        } else if (!(flag & CL_PRIVATE)) {
1052                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1053                        list_add(&mnt->mnt_share, &old->mnt_share);
1054                if (IS_MNT_SLAVE(old))
1055                        list_add(&mnt->mnt_slave, &old->mnt_slave);
1056                mnt->mnt_master = old->mnt_master;
1057        }
1058        if (flag & CL_MAKE_SHARED)
1059                set_mnt_shared(mnt);
1060
1061        /* stick the duplicate mount on the same expiry list
1062         * as the original if that was on one */
1063        if (flag & CL_EXPIRE) {
1064                if (!list_empty(&old->mnt_expire))
1065                        list_add(&mnt->mnt_expire, &old->mnt_expire);
1066        }
1067
1068        return mnt;
1069
1070 out_free:
1071        mnt_free_id(mnt);
1072        free_vfsmnt(mnt);
1073        return ERR_PTR(err);
1074}
1075
1076static void cleanup_mnt(struct mount *mnt)
1077{
1078        /*
1079         * This probably indicates that somebody messed
1080         * up a mnt_want/drop_write() pair.  If this
1081         * happens, the filesystem was probably unable
1082         * to make r/w->r/o transitions.
1083         */
1084        /*
1085         * The locking used to deal with mnt_count decrement provides barriers,
1086         * so mnt_get_writers() below is safe.
1087         */
1088        WARN_ON(mnt_get_writers(mnt));
1089        if (unlikely(mnt->mnt_pins.first))
1090                mnt_pin_kill(mnt);
1091        fsnotify_vfsmount_delete(&mnt->mnt);
1092        dput(mnt->mnt.mnt_root);
1093        deactivate_super(mnt->mnt.mnt_sb);
1094        mnt_free_id(mnt);
1095        call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1096}
1097
1098static void __cleanup_mnt(struct rcu_head *head)
1099{
1100        cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1101}
1102
1103static LLIST_HEAD(delayed_mntput_list);
1104static void delayed_mntput(struct work_struct *unused)
1105{
1106        struct llist_node *node = llist_del_all(&delayed_mntput_list);
1107        struct llist_node *next;
1108
1109        for (; node; node = next) {
1110                next = llist_next(node);
1111                cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1112        }
1113}
1114static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1115
1116static void mntput_no_expire(struct mount *mnt)
1117{
1118        rcu_read_lock();
1119        mnt_add_count(mnt, -1);
1120        if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1121                rcu_read_unlock();
1122                return;
1123        }
1124        lock_mount_hash();
1125        if (mnt_get_count(mnt)) {
1126                rcu_read_unlock();
1127                unlock_mount_hash();
1128                return;
1129        }
1130        if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1131                rcu_read_unlock();
1132                unlock_mount_hash();
1133                return;
1134        }
1135        mnt->mnt.mnt_flags |= MNT_DOOMED;
1136        rcu_read_unlock();
1137
1138        list_del(&mnt->mnt_instance);
1139
1140        if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1141                struct mount *p, *tmp;
1142                list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1143                        umount_mnt(p);
1144                }
1145        }
1146        unlock_mount_hash();
1147
1148        if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1149                struct task_struct *task = current;
1150                if (likely(!(task->flags & PF_KTHREAD))) {
1151                        init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1152                        if (!task_work_add(task, &mnt->mnt_rcu, true))
1153                                return;
1154                }
1155                if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1156                        schedule_delayed_work(&delayed_mntput_work, 1);
1157                return;
1158        }
1159        cleanup_mnt(mnt);
1160}
1161
1162void mntput(struct vfsmount *mnt)
1163{
1164        if (mnt) {
1165                struct mount *m = real_mount(mnt);
1166                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1167                if (unlikely(m->mnt_expiry_mark))
1168                        m->mnt_expiry_mark = 0;
1169                mntput_no_expire(m);
1170        }
1171}
1172EXPORT_SYMBOL(mntput);
1173
1174struct vfsmount *mntget(struct vfsmount *mnt)
1175{
1176        if (mnt)
1177                mnt_add_count(real_mount(mnt), 1);
1178        return mnt;
1179}
1180EXPORT_SYMBOL(mntget);
1181
1182struct vfsmount *mnt_clone_internal(struct path *path)
1183{
1184        struct mount *p;
1185        p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1186        if (IS_ERR(p))
1187                return ERR_CAST(p);
1188        p->mnt.mnt_flags |= MNT_INTERNAL;
1189        return &p->mnt;
1190}
1191
1192static inline void mangle(struct seq_file *m, const char *s)
1193{
1194        seq_escape(m, s, " \t\n\\");
1195}
1196
1197/*
1198 * Simple .show_options callback for filesystems which don't want to
1199 * implement more complex mount option showing.
1200 *
1201 * See also save_mount_options().
1202 */
1203int generic_show_options(struct seq_file *m, struct dentry *root)
1204{
1205        const char *options;
1206
1207        rcu_read_lock();
1208        options = rcu_dereference(root->d_sb->s_options);
1209
1210        if (options != NULL && options[0]) {
1211                seq_putc(m, ',');
1212                mangle(m, options);
1213        }
1214        rcu_read_unlock();
1215
1216        return 0;
1217}
1218EXPORT_SYMBOL(generic_show_options);
1219
1220/*
1221 * If filesystem uses generic_show_options(), this function should be
1222 * called from the fill_super() callback.
1223 *
1224 * The .remount_fs callback usually needs to be handled in a special
1225 * way, to make sure, that previous options are not overwritten if the
1226 * remount fails.
1227 *
1228 * Also note, that if the filesystem's .remount_fs function doesn't
1229 * reset all options to their default value, but changes only newly
1230 * given options, then the displayed options will not reflect reality
1231 * any more.
1232 */
1233void save_mount_options(struct super_block *sb, char *options)
1234{
1235        BUG_ON(sb->s_options);
1236        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1237}
1238EXPORT_SYMBOL(save_mount_options);
1239
1240void replace_mount_options(struct super_block *sb, char *options)
1241{
1242        char *old = sb->s_options;
1243        rcu_assign_pointer(sb->s_options, options);
1244        if (old) {
1245                synchronize_rcu();
1246                kfree(old);
1247        }
1248}
1249EXPORT_SYMBOL(replace_mount_options);
1250
1251#ifdef CONFIG_PROC_FS
1252/* iterator; we want it to have access to namespace_sem, thus here... */
1253static void *m_start(struct seq_file *m, loff_t *pos)
1254{
1255        struct proc_mounts *p = m->private;
1256
1257        down_read(&namespace_sem);
1258        if (p->cached_event == p->ns->event) {
1259                void *v = p->cached_mount;
1260                if (*pos == p->cached_index)
1261                        return v;
1262                if (*pos == p->cached_index + 1) {
1263                        v = seq_list_next(v, &p->ns->list, &p->cached_index);
1264                        return p->cached_mount = v;
1265                }
1266        }
1267
1268        p->cached_event = p->ns->event;
1269        p->cached_mount = seq_list_start(&p->ns->list, *pos);
1270        p->cached_index = *pos;
1271        return p->cached_mount;
1272}
1273
1274static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1275{
1276        struct proc_mounts *p = m->private;
1277
1278        p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1279        p->cached_index = *pos;
1280        return p->cached_mount;
1281}
1282
1283static void m_stop(struct seq_file *m, void *v)
1284{
1285        up_read(&namespace_sem);
1286}
1287
1288static int m_show(struct seq_file *m, void *v)
1289{
1290        struct proc_mounts *p = m->private;
1291        struct mount *r = list_entry(v, struct mount, mnt_list);
1292        return p->show(m, &r->mnt);
1293}
1294
1295const struct seq_operations mounts_op = {
1296        .start  = m_start,
1297        .next   = m_next,
1298        .stop   = m_stop,
1299        .show   = m_show,
1300};
1301#endif  /* CONFIG_PROC_FS */
1302
1303/**
1304 * may_umount_tree - check if a mount tree is busy
1305 * @mnt: root of mount tree
1306 *
1307 * This is called to check if a tree of mounts has any
1308 * open files, pwds, chroots or sub mounts that are
1309 * busy.
1310 */
1311int may_umount_tree(struct vfsmount *m)
1312{
1313        struct mount *mnt = real_mount(m);
1314        int actual_refs = 0;
1315        int minimum_refs = 0;
1316        struct mount *p;
1317        BUG_ON(!m);
1318
1319        /* write lock needed for mnt_get_count */
1320        lock_mount_hash();
1321        for (p = mnt; p; p = next_mnt(p, mnt)) {
1322                actual_refs += mnt_get_count(p);
1323                minimum_refs += 2;
1324        }
1325        unlock_mount_hash();
1326
1327        if (actual_refs > minimum_refs)
1328                return 0;
1329
1330        return 1;
1331}
1332
1333EXPORT_SYMBOL(may_umount_tree);
1334
1335/**
1336 * may_umount - check if a mount point is busy
1337 * @mnt: root of mount
1338 *
1339 * This is called to check if a mount point has any
1340 * open files, pwds, chroots or sub mounts. If the
1341 * mount has sub mounts this will return busy
1342 * regardless of whether the sub mounts are busy.
1343 *
1344 * Doesn't take quota and stuff into account. IOW, in some cases it will
1345 * give false negatives. The main reason why it's here is that we need
1346 * a non-destructive way to look for easily umountable filesystems.
1347 */
1348int may_umount(struct vfsmount *mnt)
1349{
1350        int ret = 1;
1351        down_read(&namespace_sem);
1352        lock_mount_hash();
1353        if (propagate_mount_busy(real_mount(mnt), 2))
1354                ret = 0;
1355        unlock_mount_hash();
1356        up_read(&namespace_sem);
1357        return ret;
1358}
1359
1360EXPORT_SYMBOL(may_umount);
1361
1362static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
1363
1364static void namespace_unlock(void)
1365{
1366        struct hlist_head head;
1367
1368        hlist_move_list(&unmounted, &head);
1369
1370        up_write(&namespace_sem);
1371
1372        if (likely(hlist_empty(&head)))
1373                return;
1374
1375        synchronize_rcu();
1376
1377        group_pin_kill(&head);
1378}
1379
1380static inline void namespace_lock(void)
1381{
1382        down_write(&namespace_sem);
1383}
1384
1385enum umount_tree_flags {
1386        UMOUNT_SYNC = 1,
1387        UMOUNT_PROPAGATE = 2,
1388        UMOUNT_CONNECTED = 4,
1389};
1390
1391static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1392{
1393        /* Leaving mounts connected is only valid for lazy umounts */
1394        if (how & UMOUNT_SYNC)
1395                return true;
1396
1397        /* A mount without a parent has nothing to be connected to */
1398        if (!mnt_has_parent(mnt))
1399                return true;
1400
1401        /* Because the reference counting rules change when mounts are
1402         * unmounted and connected, umounted mounts may not be
1403         * connected to mounted mounts.
1404         */
1405        if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1406                return true;
1407
1408        /* Has it been requested that the mount remain connected? */
1409        if (how & UMOUNT_CONNECTED)
1410                return false;
1411
1412        /* Is the mount locked such that it needs to remain connected? */
1413        if (IS_MNT_LOCKED(mnt))
1414                return false;
1415
1416        /* By default disconnect the mount */
1417        return true;
1418}
1419
1420/*
1421 * mount_lock must be held
1422 * namespace_sem must be held for write
1423 */
1424static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1425{
1426        LIST_HEAD(tmp_list);
1427        struct mount *p;
1428
1429        if (how & UMOUNT_PROPAGATE)
1430                propagate_mount_unlock(mnt);
1431
1432        /* Gather the mounts to umount */
1433        for (p = mnt; p; p = next_mnt(p, mnt)) {
1434                p->mnt.mnt_flags |= MNT_UMOUNT;
1435                list_move(&p->mnt_list, &tmp_list);
1436        }
1437
1438        /* Hide the mounts from mnt_mounts */
1439        list_for_each_entry(p, &tmp_list, mnt_list) {
1440                list_del_init(&p->mnt_child);
1441        }
1442
1443        /* Add propogated mounts to the tmp_list */
1444        if (how & UMOUNT_PROPAGATE)
1445                propagate_umount(&tmp_list);
1446
1447        while (!list_empty(&tmp_list)) {
1448                bool disconnect;
1449                p = list_first_entry(&tmp_list, struct mount, mnt_list);
1450                list_del_init(&p->mnt_expire);
1451                list_del_init(&p->mnt_list);
1452                __touch_mnt_namespace(p->mnt_ns);
1453                p->mnt_ns = NULL;
1454                if (how & UMOUNT_SYNC)
1455                        p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1456
1457                disconnect = disconnect_mount(p, how);
1458
1459                pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1460                                 disconnect ? &unmounted : NULL);
1461                if (mnt_has_parent(p)) {
1462                        mnt_add_count(p->mnt_parent, -1);
1463                        if (!disconnect) {
1464                                /* Don't forget about p */
1465                                list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1466                        } else {
1467                                umount_mnt(p);
1468                        }
1469                }
1470                change_mnt_propagation(p, MS_PRIVATE);
1471        }
1472}
1473
1474static void shrink_submounts(struct mount *mnt);
1475
1476static int do_umount(struct mount *mnt, int flags)
1477{
1478        struct super_block *sb = mnt->mnt.mnt_sb;
1479        int retval;
1480
1481        retval = security_sb_umount(&mnt->mnt, flags);
1482        if (retval)
1483                return retval;
1484
1485        /*
1486         * Allow userspace to request a mountpoint be expired rather than
1487         * unmounting unconditionally. Unmount only happens if:
1488         *  (1) the mark is already set (the mark is cleared by mntput())
1489         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1490         */
1491        if (flags & MNT_EXPIRE) {
1492                if (&mnt->mnt == current->fs->root.mnt ||
1493                    flags & (MNT_FORCE | MNT_DETACH))
1494                        return -EINVAL;
1495
1496                /*
1497                 * probably don't strictly need the lock here if we examined
1498                 * all race cases, but it's a slowpath.
1499                 */
1500                lock_mount_hash();
1501                if (mnt_get_count(mnt) != 2) {
1502                        unlock_mount_hash();
1503                        return -EBUSY;
1504                }
1505                unlock_mount_hash();
1506
1507                if (!xchg(&mnt->mnt_expiry_mark, 1))
1508                        return -EAGAIN;
1509        }
1510
1511        /*
1512         * If we may have to abort operations to get out of this
1513         * mount, and they will themselves hold resources we must
1514         * allow the fs to do things. In the Unix tradition of
1515         * 'Gee thats tricky lets do it in userspace' the umount_begin
1516         * might fail to complete on the first run through as other tasks
1517         * must return, and the like. Thats for the mount program to worry
1518         * about for the moment.
1519         */
1520
1521        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1522                sb->s_op->umount_begin(sb);
1523        }
1524
1525        /*
1526         * No sense to grab the lock for this test, but test itself looks
1527         * somewhat bogus. Suggestions for better replacement?
1528         * Ho-hum... In principle, we might treat that as umount + switch
1529         * to rootfs. GC would eventually take care of the old vfsmount.
1530         * Actually it makes sense, especially if rootfs would contain a
1531         * /reboot - static binary that would close all descriptors and
1532         * call reboot(9). Then init(8) could umount root and exec /reboot.
1533         */
1534        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1535                /*
1536                 * Special case for "unmounting" root ...
1537                 * we just try to remount it readonly.
1538                 */
1539                if (!capable(CAP_SYS_ADMIN))
1540                        return -EPERM;
1541                down_write(&sb->s_umount);
1542                if (!(sb->s_flags & MS_RDONLY))
1543                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1544                up_write(&sb->s_umount);
1545                return retval;
1546        }
1547
1548        namespace_lock();
1549        lock_mount_hash();
1550        event++;
1551
1552        if (flags & MNT_DETACH) {
1553                if (!list_empty(&mnt->mnt_list))
1554                        umount_tree(mnt, UMOUNT_PROPAGATE);
1555                retval = 0;
1556        } else {
1557                shrink_submounts(mnt);
1558                retval = -EBUSY;
1559                if (!propagate_mount_busy(mnt, 2)) {
1560                        if (!list_empty(&mnt->mnt_list))
1561                                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1562                        retval = 0;
1563                }
1564        }
1565        unlock_mount_hash();
1566        namespace_unlock();
1567        return retval;
1568}
1569
1570/*
1571 * __detach_mounts - lazily unmount all mounts on the specified dentry
1572 *
1573 * During unlink, rmdir, and d_drop it is possible to loose the path
1574 * to an existing mountpoint, and wind up leaking the mount.
1575 * detach_mounts allows lazily unmounting those mounts instead of
1576 * leaking them.
1577 *
1578 * The caller may hold dentry->d_inode->i_mutex.
1579 */
1580void __detach_mounts(struct dentry *dentry)
1581{
1582        struct mountpoint *mp;
1583        struct mount *mnt;
1584
1585        namespace_lock();
1586        lock_mount_hash();
1587        mp = lookup_mountpoint(dentry);
1588        if (IS_ERR_OR_NULL(mp))
1589                goto out_unlock;
1590
1591        event++;
1592        while (!hlist_empty(&mp->m_list)) {
1593                mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1594                if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1595                        hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1596                        umount_mnt(mnt);
1597                }
1598                else umount_tree(mnt, UMOUNT_CONNECTED);
1599        }
1600        put_mountpoint(mp);
1601out_unlock:
1602        unlock_mount_hash();
1603        namespace_unlock();
1604}
1605
1606/*
1607 * Is the caller allowed to modify his namespace?
1608 */
1609static inline bool may_mount(void)
1610{
1611        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1612}
1613
1614/*
1615 * Now umount can handle mount points as well as block devices.
1616 * This is important for filesystems which use unnamed block devices.
1617 *
1618 * We now support a flag for forced unmount like the other 'big iron'
1619 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1620 */
1621
1622SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1623{
1624        struct path path;
1625        struct mount *mnt;
1626        int retval;
1627        int lookup_flags = 0;
1628
1629        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1630                return -EINVAL;
1631
1632        if (!may_mount())
1633                return -EPERM;
1634
1635        if (!(flags & UMOUNT_NOFOLLOW))
1636                lookup_flags |= LOOKUP_FOLLOW;
1637
1638        retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1639        if (retval)
1640                goto out;
1641        mnt = real_mount(path.mnt);
1642        retval = -EINVAL;
1643        if (path.dentry != path.mnt->mnt_root)
1644                goto dput_and_out;
1645        if (!check_mnt(mnt))
1646                goto dput_and_out;
1647        if (mnt->mnt.mnt_flags & MNT_LOCKED)
1648                goto dput_and_out;
1649        retval = -EPERM;
1650        if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1651                goto dput_and_out;
1652
1653        retval = do_umount(mnt, flags);
1654dput_and_out:
1655        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1656        dput(path.dentry);
1657        mntput_no_expire(mnt);
1658out:
1659        return retval;
1660}
1661
1662#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1663
1664/*
1665 *      The 2.0 compatible umount. No flags.
1666 */
1667SYSCALL_DEFINE1(oldumount, char __user *, name)
1668{
1669        return sys_umount(name, 0);
1670}
1671
1672#endif
1673
1674static bool is_mnt_ns_file(struct dentry *dentry)
1675{
1676        /* Is this a proxy for a mount namespace? */
1677        return dentry->d_op == &ns_dentry_operations &&
1678               dentry->d_fsdata == &mntns_operations;
1679}
1680
1681struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1682{
1683        return container_of(ns, struct mnt_namespace, ns);
1684}
1685
1686static bool mnt_ns_loop(struct dentry *dentry)
1687{
1688        /* Could bind mounting the mount namespace inode cause a
1689         * mount namespace loop?
1690         */
1691        struct mnt_namespace *mnt_ns;
1692        if (!is_mnt_ns_file(dentry))
1693                return false;
1694
1695        mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1696        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1697}
1698
1699struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1700                                        int flag)
1701{
1702        struct mount *res, *p, *q, *r, *parent;
1703
1704        if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1705                return ERR_PTR(-EINVAL);
1706
1707        if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1708                return ERR_PTR(-EINVAL);
1709
1710        res = q = clone_mnt(mnt, dentry, flag);
1711        if (IS_ERR(q))
1712                return q;
1713
1714        q->mnt_mountpoint = mnt->mnt_mountpoint;
1715
1716        p = mnt;
1717        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1718                struct mount *s;
1719                if (!is_subdir(r->mnt_mountpoint, dentry))
1720                        continue;
1721
1722                for (s = r; s; s = next_mnt(s, r)) {
1723                        if (!(flag & CL_COPY_UNBINDABLE) &&
1724                            IS_MNT_UNBINDABLE(s)) {
1725                                s = skip_mnt_tree(s);
1726                                continue;
1727                        }
1728                        if (!(flag & CL_COPY_MNT_NS_FILE) &&
1729                            is_mnt_ns_file(s->mnt.mnt_root)) {
1730                                s = skip_mnt_tree(s);
1731                                continue;
1732                        }
1733                        while (p != s->mnt_parent) {
1734                                p = p->mnt_parent;
1735                                q = q->mnt_parent;
1736                        }
1737                        p = s;
1738                        parent = q;
1739                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1740                        if (IS_ERR(q))
1741                                goto out;
1742                        lock_mount_hash();
1743                        list_add_tail(&q->mnt_list, &res->mnt_list);
1744                        attach_mnt(q, parent, p->mnt_mp);
1745                        unlock_mount_hash();
1746                }
1747        }
1748        return res;
1749out:
1750        if (res) {
1751                lock_mount_hash();
1752                umount_tree(res, UMOUNT_SYNC);
1753                unlock_mount_hash();
1754        }
1755        return q;
1756}
1757
1758/* Caller should check returned pointer for errors */
1759
1760struct vfsmount *collect_mounts(struct path *path)
1761{
1762        struct mount *tree;
1763        namespace_lock();
1764        if (!check_mnt(real_mount(path->mnt)))
1765                tree = ERR_PTR(-EINVAL);
1766        else
1767                tree = copy_tree(real_mount(path->mnt), path->dentry,
1768                                 CL_COPY_ALL | CL_PRIVATE);
1769        namespace_unlock();
1770        if (IS_ERR(tree))
1771                return ERR_CAST(tree);
1772        return &tree->mnt;
1773}
1774
1775void drop_collected_mounts(struct vfsmount *mnt)
1776{
1777        namespace_lock();
1778        lock_mount_hash();
1779        umount_tree(real_mount(mnt), UMOUNT_SYNC);
1780        unlock_mount_hash();
1781        namespace_unlock();
1782}
1783
1784/**
1785 * clone_private_mount - create a private clone of a path
1786 *
1787 * This creates a new vfsmount, which will be the clone of @path.  The new will
1788 * not be attached anywhere in the namespace and will be private (i.e. changes
1789 * to the originating mount won't be propagated into this).
1790 *
1791 * Release with mntput().
1792 */
1793struct vfsmount *clone_private_mount(struct path *path)
1794{
1795        struct mount *old_mnt = real_mount(path->mnt);
1796        struct mount *new_mnt;
1797
1798        if (IS_MNT_UNBINDABLE(old_mnt))
1799                return ERR_PTR(-EINVAL);
1800
1801        down_read(&namespace_sem);
1802        new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1803        up_read(&namespace_sem);
1804        if (IS_ERR(new_mnt))
1805                return ERR_CAST(new_mnt);
1806
1807        return &new_mnt->mnt;
1808}
1809EXPORT_SYMBOL_GPL(clone_private_mount);
1810
1811int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1812                   struct vfsmount *root)
1813{
1814        struct mount *mnt;
1815        int res = f(root, arg);
1816        if (res)
1817                return res;
1818        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1819                res = f(&mnt->mnt, arg);
1820                if (res)
1821                        return res;
1822        }
1823        return 0;
1824}
1825
1826static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1827{
1828        struct mount *p;
1829
1830        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1831                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1832                        mnt_release_group_id(p);
1833        }
1834}
1835
1836static int invent_group_ids(struct mount *mnt, bool recurse)
1837{
1838        struct mount *p;
1839
1840        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1841                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1842                        int err = mnt_alloc_group_id(p);
1843                        if (err) {
1844                                cleanup_group_ids(mnt, p);
1845                                return err;
1846                        }
1847                }
1848        }
1849
1850        return 0;
1851}
1852
1853/*
1854 *  @source_mnt : mount tree to be attached
1855 *  @nd         : place the mount tree @source_mnt is attached
1856 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1857 *                 store the parent mount and mountpoint dentry.
1858 *                 (done when source_mnt is moved)
1859 *
1860 *  NOTE: in the table below explains the semantics when a source mount
1861 *  of a given type is attached to a destination mount of a given type.
1862 * ---------------------------------------------------------------------------
1863 * |         BIND MOUNT OPERATION                                            |
1864 * |**************************************************************************
1865 * | source-->| shared        |       private  |       slave    | unbindable |
1866 * | dest     |               |                |                |            |
1867 * |   |      |               |                |                |            |
1868 * |   v      |               |                |                |            |
1869 * |**************************************************************************
1870 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1871 * |          |               |                |                |            |
1872 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1873 * ***************************************************************************
1874 * A bind operation clones the source mount and mounts the clone on the
1875 * destination mount.
1876 *
1877 * (++)  the cloned mount is propagated to all the mounts in the propagation
1878 *       tree of the destination mount and the cloned mount is added to
1879 *       the peer group of the source mount.
1880 * (+)   the cloned mount is created under the destination mount and is marked
1881 *       as shared. The cloned mount is added to the peer group of the source
1882 *       mount.
1883 * (+++) the mount is propagated to all the mounts in the propagation tree
1884 *       of the destination mount and the cloned mount is made slave
1885 *       of the same master as that of the source mount. The cloned mount
1886 *       is marked as 'shared and slave'.
1887 * (*)   the cloned mount is made a slave of the same master as that of the
1888 *       source mount.
1889 *
1890 * ---------------------------------------------------------------------------
1891 * |                    MOVE MOUNT OPERATION                                 |
1892 * |**************************************************************************
1893 * | source-->| shared        |       private  |       slave    | unbindable |
1894 * | dest     |               |                |                |            |
1895 * |   |      |               |                |                |            |
1896 * |   v      |               |                |                |            |
1897 * |**************************************************************************
1898 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1899 * |          |               |                |                |            |
1900 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1901 * ***************************************************************************
1902 *
1903 * (+)  the mount is moved to the destination. And is then propagated to
1904 *      all the mounts in the propagation tree of the destination mount.
1905 * (+*)  the mount is moved to the destination.
1906 * (+++)  the mount is moved to the destination and is then propagated to
1907 *      all the mounts belonging to the destination mount's propagation tree.
1908 *      the mount is marked as 'shared and slave'.
1909 * (*)  the mount continues to be a slave at the new location.
1910 *
1911 * if the source mount is a tree, the operations explained above is
1912 * applied to each mount in the tree.
1913 * Must be called without spinlocks held, since this function can sleep
1914 * in allocations.
1915 */
1916static int attach_recursive_mnt(struct mount *source_mnt,
1917                        struct mount *dest_mnt,
1918                        struct mountpoint *dest_mp,
1919                        struct path *parent_path)
1920{
1921        HLIST_HEAD(tree_list);
1922        struct mountpoint *smp;
1923        struct mount *child, *p;
1924        struct hlist_node *n;
1925        int err;
1926
1927        /* Preallocate a mountpoint in case the new mounts need
1928         * to be tucked under other mounts.
1929         */
1930        smp = get_mountpoint(source_mnt->mnt.mnt_root);
1931        if (IS_ERR(smp))
1932                return PTR_ERR(smp);
1933
1934        if (IS_MNT_SHARED(dest_mnt)) {
1935                err = invent_group_ids(source_mnt, true);
1936                if (err)
1937                        goto out;
1938                err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1939                lock_mount_hash();
1940                if (err)
1941                        goto out_cleanup_ids;
1942                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1943                        set_mnt_shared(p);
1944        } else {
1945                lock_mount_hash();
1946        }
1947        if (parent_path) {
1948                detach_mnt(source_mnt, parent_path);
1949                attach_mnt(source_mnt, dest_mnt, dest_mp);
1950                touch_mnt_namespace(source_mnt->mnt_ns);
1951        } else {
1952                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1953                commit_tree(source_mnt);
1954        }
1955
1956        hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1957                struct mount *q;
1958                hlist_del_init(&child->mnt_hash);
1959                q = __lookup_mnt(&child->mnt_parent->mnt,
1960                                 child->mnt_mountpoint);
1961                if (q)
1962                        mnt_change_mountpoint(child, smp, q);
1963                commit_tree(child);
1964        }
1965        put_mountpoint(smp);
1966        unlock_mount_hash();
1967
1968        return 0;
1969
1970 out_cleanup_ids:
1971        while (!hlist_empty(&tree_list)) {
1972                child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1973                umount_tree(child, UMOUNT_SYNC);
1974        }
1975        unlock_mount_hash();
1976        cleanup_group_ids(source_mnt, NULL);
1977 out:
1978        read_seqlock_excl(&mount_lock);
1979        put_mountpoint(smp);
1980        read_sequnlock_excl(&mount_lock);
1981
1982        return err;
1983}
1984
1985static struct mountpoint *lock_mount(struct path *path)
1986{
1987        struct vfsmount *mnt;
1988        struct dentry *dentry = path->dentry;
1989retry:
1990        mutex_lock(&dentry->d_inode->i_mutex);
1991        if (unlikely(cant_mount(dentry))) {
1992                mutex_unlock(&dentry->d_inode->i_mutex);
1993                return ERR_PTR(-ENOENT);
1994        }
1995        namespace_lock();
1996        mnt = lookup_mnt(path);
1997        if (likely(!mnt)) {
1998                struct mountpoint *mp = get_mountpoint(dentry);
1999                if (IS_ERR(mp)) {
2000                        namespace_unlock();
2001                        mutex_unlock(&dentry->d_inode->i_mutex);
2002                        return mp;
2003                }
2004                return mp;
2005        }
2006        namespace_unlock();
2007        mutex_unlock(&path->dentry->d_inode->i_mutex);
2008        path_put(path);
2009        path->mnt = mnt;
2010        dentry = path->dentry = dget(mnt->mnt_root);
2011        goto retry;
2012}
2013
2014static void unlock_mount(struct mountpoint *where)
2015{
2016        struct dentry *dentry = where->m_dentry;
2017
2018        read_seqlock_excl(&mount_lock);
2019        put_mountpoint(where);
2020        read_sequnlock_excl(&mount_lock);
2021
2022        namespace_unlock();
2023        mutex_unlock(&dentry->d_inode->i_mutex);
2024}
2025
2026static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2027{
2028        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2029                return -EINVAL;
2030
2031        if (d_is_dir(mp->m_dentry) !=
2032              d_is_dir(mnt->mnt.mnt_root))
2033                return -ENOTDIR;
2034
2035        return attach_recursive_mnt(mnt, p, mp, NULL);
2036}
2037
2038/*
2039 * Sanity check the flags to change_mnt_propagation.
2040 */
2041
2042static int flags_to_propagation_type(int flags)
2043{
2044        int type = flags & ~(MS_REC | MS_SILENT);
2045
2046        /* Fail if any non-propagation flags are set */
2047        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2048                return 0;
2049        /* Only one propagation flag should be set */
2050        if (!is_power_of_2(type))
2051                return 0;
2052        return type;
2053}
2054
2055/*
2056 * recursively change the type of the mountpoint.
2057 */
2058static int do_change_type(struct path *path, int flag)
2059{
2060        struct mount *m;
2061        struct mount *mnt = real_mount(path->mnt);
2062        int recurse = flag & MS_REC;
2063        int type;
2064        int err = 0;
2065
2066        if (path->dentry != path->mnt->mnt_root)
2067                return -EINVAL;
2068
2069        type = flags_to_propagation_type(flag);
2070        if (!type)
2071                return -EINVAL;
2072
2073        namespace_lock();
2074        if (type == MS_SHARED) {
2075                err = invent_group_ids(mnt, recurse);
2076                if (err)
2077                        goto out_unlock;
2078        }
2079
2080        lock_mount_hash();
2081        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2082                change_mnt_propagation(m, type);
2083        unlock_mount_hash();
2084
2085 out_unlock:
2086        namespace_unlock();
2087        return err;
2088}
2089
2090static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2091{
2092        struct mount *child;
2093        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2094                if (!is_subdir(child->mnt_mountpoint, dentry))
2095                        continue;
2096
2097                if (child->mnt.mnt_flags & MNT_LOCKED)
2098                        return true;
2099        }
2100        return false;
2101}
2102
2103/*
2104 * do loopback mount.
2105 */
2106static int do_loopback(struct path *path, const char *old_name,
2107                                int recurse)
2108{
2109        struct path old_path;
2110        struct mount *mnt = NULL, *old, *parent;
2111        struct mountpoint *mp;
2112        int err;
2113        if (!old_name || !*old_name)
2114                return -EINVAL;
2115        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2116        if (err)
2117                return err;
2118
2119        err = -EINVAL;
2120        if (mnt_ns_loop(old_path.dentry))
2121                goto out;
2122
2123        mp = lock_mount(path);
2124        err = PTR_ERR(mp);
2125        if (IS_ERR(mp))
2126                goto out;
2127
2128        old = real_mount(old_path.mnt);
2129        parent = real_mount(path->mnt);
2130
2131        err = -EINVAL;
2132        if (IS_MNT_UNBINDABLE(old))
2133                goto out2;
2134
2135        if (!check_mnt(parent))
2136                goto out2;
2137
2138        if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2139                goto out2;
2140
2141        if (!recurse && has_locked_children(old, old_path.dentry))
2142                goto out2;
2143
2144        if (recurse)
2145                mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2146        else
2147                mnt = clone_mnt(old, old_path.dentry, 0);
2148
2149        if (IS_ERR(mnt)) {
2150                err = PTR_ERR(mnt);
2151                goto out2;
2152        }
2153
2154        mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2155
2156        err = graft_tree(mnt, parent, mp);
2157        if (err) {
2158                lock_mount_hash();
2159                umount_tree(mnt, UMOUNT_SYNC);
2160                unlock_mount_hash();
2161        }
2162out2:
2163        unlock_mount(mp);
2164out:
2165        path_put(&old_path);
2166        return err;
2167}
2168
2169static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2170{
2171        int error = 0;
2172        int readonly_request = 0;
2173
2174        if (ms_flags & MS_RDONLY)
2175                readonly_request = 1;
2176        if (readonly_request == __mnt_is_readonly(mnt))
2177                return 0;
2178
2179        if (readonly_request)
2180                error = mnt_make_readonly(real_mount(mnt));
2181        else
2182                __mnt_unmake_readonly(real_mount(mnt));
2183        return error;
2184}
2185
2186/*
2187 * change filesystem flags. dir should be a physical root of filesystem.
2188 * If you've mounted a non-root directory somewhere and want to do remount
2189 * on it - tough luck.
2190 */
2191static int do_remount(struct path *path, int flags, int mnt_flags,
2192                      void *data)
2193{
2194        int err;
2195        struct super_block *sb = path->mnt->mnt_sb;
2196        struct mount *mnt = real_mount(path->mnt);
2197
2198        if (!check_mnt(mnt))
2199                return -EINVAL;
2200
2201        if (path->dentry != path->mnt->mnt_root)
2202                return -EINVAL;
2203
2204        /* Don't allow changing of locked mnt flags.
2205         *
2206         * No locks need to be held here while testing the various
2207         * MNT_LOCK flags because those flags can never be cleared
2208         * once they are set.
2209         */
2210        if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2211            !(mnt_flags & MNT_READONLY)) {
2212                return -EPERM;
2213        }
2214        if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2215            !(mnt_flags & MNT_NODEV)) {
2216                /* Was the nodev implicitly added in mount? */
2217                if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2218                    !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2219                        mnt_flags |= MNT_NODEV;
2220                } else {
2221                        return -EPERM;
2222                }
2223        }
2224        if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2225            !(mnt_flags & MNT_NOSUID)) {
2226                return -EPERM;
2227        }
2228        if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2229            !(mnt_flags & MNT_NOEXEC)) {
2230                return -EPERM;
2231        }
2232        if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2233            ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2234                return -EPERM;
2235        }
2236
2237        err = security_sb_remount(sb, data);
2238        if (err)
2239                return err;
2240
2241        down_write(&sb->s_umount);
2242        if (flags & MS_BIND)
2243                err = change_mount_flags(path->mnt, flags);
2244        else if (!capable(CAP_SYS_ADMIN))
2245                err = -EPERM;
2246        else
2247                err = do_remount_sb(sb, flags, data, 0);
2248        if (!err) {
2249                lock_mount_hash();
2250                mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2251                mnt->mnt.mnt_flags = mnt_flags;
2252                touch_mnt_namespace(mnt->mnt_ns);
2253                unlock_mount_hash();
2254        }
2255        up_write(&sb->s_umount);
2256        return err;
2257}
2258
2259static inline int tree_contains_unbindable(struct mount *mnt)
2260{
2261        struct mount *p;
2262        for (p = mnt; p; p = next_mnt(p, mnt)) {
2263                if (IS_MNT_UNBINDABLE(p))
2264                        return 1;
2265        }
2266        return 0;
2267}
2268
2269static int do_move_mount(struct path *path, const char *old_name)
2270{
2271        struct path old_path, parent_path;
2272        struct mount *p;
2273        struct mount *old;
2274        struct mountpoint *mp;
2275        int err;
2276        if (!old_name || !*old_name)
2277                return -EINVAL;
2278        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2279        if (err)
2280                return err;
2281
2282        mp = lock_mount(path);
2283        err = PTR_ERR(mp);
2284        if (IS_ERR(mp))
2285                goto out;
2286
2287        old = real_mount(old_path.mnt);
2288        p = real_mount(path->mnt);
2289
2290        err = -EINVAL;
2291        if (!check_mnt(p) || !check_mnt(old))
2292                goto out1;
2293
2294        if (old->mnt.mnt_flags & MNT_LOCKED)
2295                goto out1;
2296
2297        err = -EINVAL;
2298        if (old_path.dentry != old_path.mnt->mnt_root)
2299                goto out1;
2300
2301        if (!mnt_has_parent(old))
2302                goto out1;
2303
2304        if (d_is_dir(path->dentry) !=
2305              d_is_dir(old_path.dentry))
2306                goto out1;
2307        /*
2308         * Don't move a mount residing in a shared parent.
2309         */
2310        if (IS_MNT_SHARED(old->mnt_parent))
2311                goto out1;
2312        /*
2313         * Don't move a mount tree containing unbindable mounts to a destination
2314         * mount which is shared.
2315         */
2316        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2317                goto out1;
2318        err = -ELOOP;
2319        for (; mnt_has_parent(p); p = p->mnt_parent)
2320                if (p == old)
2321                        goto out1;
2322
2323        err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2324        if (err)
2325                goto out1;
2326
2327        /* if the mount is moved, it should no longer be expire
2328         * automatically */
2329        list_del_init(&old->mnt_expire);
2330out1:
2331        unlock_mount(mp);
2332out:
2333        if (!err)
2334                path_put(&parent_path);
2335        path_put(&old_path);
2336        return err;
2337}
2338
2339static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2340{
2341        int err;
2342        const char *subtype = strchr(fstype, '.');
2343        if (subtype) {
2344                subtype++;
2345                err = -EINVAL;
2346                if (!subtype[0])
2347                        goto err;
2348        } else
2349                subtype = "";
2350
2351        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2352        err = -ENOMEM;
2353        if (!mnt->mnt_sb->s_subtype)
2354                goto err;
2355        return mnt;
2356
2357 err:
2358        mntput(mnt);
2359        return ERR_PTR(err);
2360}
2361
2362/*
2363 * add a mount into a namespace's mount tree
2364 */
2365static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2366{
2367        struct mountpoint *mp;
2368        struct mount *parent;
2369        int err;
2370
2371        mnt_flags &= ~MNT_INTERNAL_FLAGS;
2372
2373        mp = lock_mount(path);
2374        if (IS_ERR(mp))
2375                return PTR_ERR(mp);
2376
2377        parent = real_mount(path->mnt);
2378        err = -EINVAL;
2379        if (unlikely(!check_mnt(parent))) {
2380                /* that's acceptable only for automounts done in private ns */
2381                if (!(mnt_flags & MNT_SHRINKABLE))
2382                        goto unlock;
2383                /* ... and for those we'd better have mountpoint still alive */
2384                if (!parent->mnt_ns)
2385                        goto unlock;
2386        }
2387
2388        /* Refuse the same filesystem on the same mount point */
2389        err = -EBUSY;
2390        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2391            path->mnt->mnt_root == path->dentry)
2392                goto unlock;
2393
2394        err = -EINVAL;
2395        if (d_is_symlink(newmnt->mnt.mnt_root))
2396                goto unlock;
2397
2398        newmnt->mnt.mnt_flags = mnt_flags;
2399        err = graft_tree(newmnt, parent, mp);
2400
2401unlock:
2402        unlock_mount(mp);
2403        return err;
2404}
2405
2406static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2407
2408/*
2409 * create a new mount for userspace and request it to be added into the
2410 * namespace's tree
2411 */
2412static int do_new_mount(struct path *path, const char *fstype, int flags,
2413                        int mnt_flags, const char *name, void *data)
2414{
2415        struct file_system_type *type;
2416        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2417        struct vfsmount *mnt;
2418        int err;
2419
2420        if (!fstype)
2421                return -EINVAL;
2422
2423        type = get_fs_type(fstype);
2424        if (!type)
2425                return -ENODEV;
2426
2427        if (user_ns != &init_user_ns) {
2428                if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2429                        put_filesystem(type);
2430                        return -EPERM;
2431                }
2432                /* Only in special cases allow devices from mounts
2433                 * created outside the initial user namespace.
2434                 */
2435                if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2436                        flags |= MS_NODEV;
2437                        mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2438                }
2439                if (type->fs_flags & FS_USERNS_VISIBLE) {
2440                        if (!fs_fully_visible(type, &mnt_flags)) {
2441                                put_filesystem(type);
2442                                return -EPERM;
2443                        }
2444                }
2445        }
2446
2447        mnt = vfs_kern_mount(type, flags, name, data);
2448        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2449            !mnt->mnt_sb->s_subtype)
2450                mnt = fs_set_subtype(mnt, fstype);
2451
2452        put_filesystem(type);
2453        if (IS_ERR(mnt))
2454                return PTR_ERR(mnt);
2455
2456        err = do_add_mount(real_mount(mnt), path, mnt_flags);
2457        if (err)
2458                mntput(mnt);
2459        return err;
2460}
2461
2462int finish_automount(struct vfsmount *m, struct path *path)
2463{
2464        struct mount *mnt = real_mount(m);
2465        int err;
2466        /* The new mount record should have at least 2 refs to prevent it being
2467         * expired before we get a chance to add it
2468         */
2469        BUG_ON(mnt_get_count(mnt) < 2);
2470
2471        if (m->mnt_sb == path->mnt->mnt_sb &&
2472            m->mnt_root == path->dentry) {
2473                err = -ELOOP;
2474                goto fail;
2475        }
2476
2477        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2478        if (!err)
2479                return 0;
2480fail:
2481        /* remove m from any expiration list it may be on */
2482        if (!list_empty(&mnt->mnt_expire)) {
2483                namespace_lock();
2484                list_del_init(&mnt->mnt_expire);
2485                namespace_unlock();
2486        }
2487        mntput(m);
2488        mntput(m);
2489        return err;
2490}
2491
2492/**
2493 * mnt_set_expiry - Put a mount on an expiration list
2494 * @mnt: The mount to list.
2495 * @expiry_list: The list to add the mount to.
2496 */
2497void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2498{
2499        namespace_lock();
2500
2501        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2502
2503        namespace_unlock();
2504}
2505EXPORT_SYMBOL(mnt_set_expiry);
2506
2507/*
2508 * process a list of expirable mountpoints with the intent of discarding any
2509 * mountpoints that aren't in use and haven't been touched since last we came
2510 * here
2511 */
2512void mark_mounts_for_expiry(struct list_head *mounts)
2513{
2514        struct mount *mnt, *next;
2515        LIST_HEAD(graveyard);
2516
2517        if (list_empty(mounts))
2518                return;
2519
2520        namespace_lock();
2521        lock_mount_hash();
2522
2523        /* extract from the expiration list every vfsmount that matches the
2524         * following criteria:
2525         * - only referenced by its parent vfsmount
2526         * - still marked for expiry (marked on the last call here; marks are
2527         *   cleared by mntput())
2528         */
2529        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2530                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2531                        propagate_mount_busy(mnt, 1))
2532                        continue;
2533                list_move(&mnt->mnt_expire, &graveyard);
2534        }
2535        while (!list_empty(&graveyard)) {
2536                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2537                touch_mnt_namespace(mnt->mnt_ns);
2538                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2539        }
2540        unlock_mount_hash();
2541        namespace_unlock();
2542}
2543
2544EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2545
2546/*
2547 * Ripoff of 'select_parent()'
2548 *
2549 * search the list of submounts for a given mountpoint, and move any
2550 * shrinkable submounts to the 'graveyard' list.
2551 */
2552static int select_submounts(struct mount *parent, struct list_head *graveyard)
2553{
2554        struct mount *this_parent = parent;
2555        struct list_head *next;
2556        int found = 0;
2557
2558repeat:
2559        next = this_parent->mnt_mounts.next;
2560resume:
2561        while (next != &this_parent->mnt_mounts) {
2562                struct list_head *tmp = next;
2563                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2564
2565                next = tmp->next;
2566                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2567                        continue;
2568                /*
2569                 * Descend a level if the d_mounts list is non-empty.
2570                 */
2571                if (!list_empty(&mnt->mnt_mounts)) {
2572                        this_parent = mnt;
2573                        goto repeat;
2574                }
2575
2576                if (!propagate_mount_busy(mnt, 1)) {
2577                        list_move_tail(&mnt->mnt_expire, graveyard);
2578                        found++;
2579                }
2580        }
2581        /*
2582         * All done at this level ... ascend and resume the search
2583         */
2584        if (this_parent != parent) {
2585                next = this_parent->mnt_child.next;
2586                this_parent = this_parent->mnt_parent;
2587                goto resume;
2588        }
2589        return found;
2590}
2591
2592/*
2593 * process a list of expirable mountpoints with the intent of discarding any
2594 * submounts of a specific parent mountpoint
2595 *
2596 * mount_lock must be held for write
2597 */
2598static void shrink_submounts(struct mount *mnt)
2599{
2600        LIST_HEAD(graveyard);
2601        struct mount *m;
2602
2603        /* extract submounts of 'mountpoint' from the expiration list */
2604        while (select_submounts(mnt, &graveyard)) {
2605                while (!list_empty(&graveyard)) {
2606                        m = list_first_entry(&graveyard, struct mount,
2607                                                mnt_expire);
2608                        touch_mnt_namespace(m->mnt_ns);
2609                        umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2610                }
2611        }
2612}
2613
2614/*
2615 * Some copy_from_user() implementations do not return the exact number of
2616 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2617 * Note that this function differs from copy_from_user() in that it will oops
2618 * on bad values of `to', rather than returning a short copy.
2619 */
2620static long exact_copy_from_user(void *to, const void __user * from,
2621                                 unsigned long n)
2622{
2623        char *t = to;
2624        const char __user *f = from;
2625        char c;
2626
2627        if (!access_ok(VERIFY_READ, from, n))
2628                return n;
2629
2630        while (n) {
2631                if (__get_user(c, f)) {
2632                        memset(t, 0, n);
2633                        break;
2634                }
2635                *t++ = c;
2636                f++;
2637                n--;
2638        }
2639        return n;
2640}
2641
2642int copy_mount_options(const void __user * data, unsigned long *where)
2643{
2644        int i;
2645        unsigned long page;
2646        unsigned long size;
2647
2648        *where = 0;
2649        if (!data)
2650                return 0;
2651
2652        if (!(page = __get_free_page(GFP_KERNEL)))
2653                return -ENOMEM;
2654
2655        /* We only care that *some* data at the address the user
2656         * gave us is valid.  Just in case, we'll zero
2657         * the remainder of the page.
2658         */
2659        /* copy_from_user cannot cross TASK_SIZE ! */
2660        size = TASK_SIZE - (unsigned long)data;
2661        if (size > PAGE_SIZE)
2662                size = PAGE_SIZE;
2663
2664        i = size - exact_copy_from_user((void *)page, data, size);
2665        if (!i) {
2666                free_page(page);
2667                return -EFAULT;
2668        }
2669        if (i != PAGE_SIZE)
2670                memset((char *)page + i, 0, PAGE_SIZE - i);
2671        *where = page;
2672        return 0;
2673}
2674
2675char *copy_mount_string(const void __user *data)
2676{
2677        return data ? strndup_user(data, PAGE_SIZE) : NULL;
2678}
2679
2680/*
2681 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2682 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2683 *
2684 * data is a (void *) that can point to any structure up to
2685 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2686 * information (or be NULL).
2687 *
2688 * Pre-0.97 versions of mount() didn't have a flags word.
2689 * When the flags word was introduced its top half was required
2690 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2691 * Therefore, if this magic number is present, it carries no information
2692 * and must be discarded.
2693 */
2694long do_mount(const char *dev_name, const char __user *dir_name,
2695                const char *type_page, unsigned long flags, void *data_page)
2696{
2697        struct path path;
2698        int retval = 0;
2699        int mnt_flags = 0;
2700
2701        /* Discard magic */
2702        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2703                flags &= ~MS_MGC_MSK;
2704
2705        /* Basic sanity checks */
2706        if (data_page)
2707                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2708
2709        /* ... and get the mountpoint */
2710        retval = user_path(dir_name, &path);
2711        if (retval)
2712                return retval;
2713
2714        retval = security_sb_mount(dev_name, &path,
2715                                   type_page, flags, data_page);
2716        if (!retval && !may_mount())
2717                retval = -EPERM;
2718        if (retval)
2719                goto dput_out;
2720
2721        /* Default to relatime unless overriden */
2722        if (!(flags & MS_NOATIME))
2723                mnt_flags |= MNT_RELATIME;
2724
2725        /* Separate the per-mountpoint flags */
2726        if (flags & MS_NOSUID)
2727                mnt_flags |= MNT_NOSUID;
2728        if (flags & MS_NODEV)
2729                mnt_flags |= MNT_NODEV;
2730        if (flags & MS_NOEXEC)
2731                mnt_flags |= MNT_NOEXEC;
2732        if (flags & MS_NOATIME)
2733                mnt_flags |= MNT_NOATIME;
2734        if (flags & MS_NODIRATIME)
2735                mnt_flags |= MNT_NODIRATIME;
2736        if (flags & MS_STRICTATIME)
2737                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2738        if (flags & MS_RDONLY)
2739                mnt_flags |= MNT_READONLY;
2740
2741        /* The default atime for remount is preservation */
2742        if ((flags & MS_REMOUNT) &&
2743            ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2744                       MS_STRICTATIME)) == 0)) {
2745                mnt_flags &= ~MNT_ATIME_MASK;
2746                mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2747        }
2748
2749        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2750                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2751                   MS_STRICTATIME);
2752
2753        if (flags & MS_REMOUNT)
2754                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2755                                    data_page);
2756        else if (flags & MS_BIND)
2757                retval = do_loopback(&path, dev_name, flags & MS_REC);
2758        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2759                retval = do_change_type(&path, flags);
2760        else if (flags & MS_MOVE)
2761                retval = do_move_mount(&path, dev_name);
2762        else
2763                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2764                                      dev_name, data_page);
2765dput_out:
2766        path_put(&path);
2767        return retval;
2768}
2769
2770static void free_mnt_ns(struct mnt_namespace *ns)
2771{
2772        ns_free_inum(&ns->ns);
2773        put_user_ns(ns->user_ns);
2774        kfree(ns);
2775}
2776
2777/*
2778 * Assign a sequence number so we can detect when we attempt to bind
2779 * mount a reference to an older mount namespace into the current
2780 * mount namespace, preventing reference counting loops.  A 64bit
2781 * number incrementing at 10Ghz will take 12,427 years to wrap which
2782 * is effectively never, so we can ignore the possibility.
2783 */
2784static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2785
2786static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2787{
2788        struct mnt_namespace *new_ns;
2789        int ret;
2790
2791        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2792        if (!new_ns)
2793                return ERR_PTR(-ENOMEM);
2794        ret = ns_alloc_inum(&new_ns->ns);
2795        if (ret) {
2796                kfree(new_ns);
2797                return ERR_PTR(ret);
2798        }
2799        new_ns->ns.ops = &mntns_operations;
2800        new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2801        atomic_set(&new_ns->count, 1);
2802        new_ns->root = NULL;
2803        INIT_LIST_HEAD(&new_ns->list);
2804        init_waitqueue_head(&new_ns->poll);
2805        new_ns->event = 0;
2806        new_ns->user_ns = get_user_ns(user_ns);
2807        return new_ns;
2808}
2809
2810struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2811                struct user_namespace *user_ns, struct fs_struct *new_fs)
2812{
2813        struct mnt_namespace *new_ns;
2814        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2815        struct mount *p, *q;
2816        struct mount *old;
2817        struct mount *new;
2818        int copy_flags;
2819
2820        BUG_ON(!ns);
2821
2822        if (likely(!(flags & CLONE_NEWNS))) {
2823                get_mnt_ns(ns);
2824                return ns;
2825        }
2826
2827        old = ns->root;
2828
2829        new_ns = alloc_mnt_ns(user_ns);
2830        if (IS_ERR(new_ns))
2831                return new_ns;
2832
2833        namespace_lock();
2834        /* First pass: copy the tree topology */
2835        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2836        if (user_ns != ns->user_ns)
2837                copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2838        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2839        if (IS_ERR(new)) {
2840                namespace_unlock();
2841                free_mnt_ns(new_ns);
2842                return ERR_CAST(new);
2843        }
2844        new_ns->root = new;
2845        list_add_tail(&new_ns->list, &new->mnt_list);
2846
2847        /*
2848         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2849         * as belonging to new namespace.  We have already acquired a private
2850         * fs_struct, so tsk->fs->lock is not needed.
2851         */
2852        p = old;
2853        q = new;
2854        while (p) {
2855                q->mnt_ns = new_ns;
2856                if (new_fs) {
2857                        if (&p->mnt == new_fs->root.mnt) {
2858                                new_fs->root.mnt = mntget(&q->mnt);
2859                                rootmnt = &p->mnt;
2860                        }
2861                        if (&p->mnt == new_fs->pwd.mnt) {
2862                                new_fs->pwd.mnt = mntget(&q->mnt);
2863                                pwdmnt = &p->mnt;
2864                        }
2865                }
2866                p = next_mnt(p, old);
2867                q = next_mnt(q, new);
2868                if (!q)
2869                        break;
2870                while (p->mnt.mnt_root != q->mnt.mnt_root)
2871                        p = next_mnt(p, old);
2872        }
2873        namespace_unlock();
2874
2875        if (rootmnt)
2876                mntput(rootmnt);
2877        if (pwdmnt)
2878                mntput(pwdmnt);
2879
2880        return new_ns;
2881}
2882
2883/**
2884 * create_mnt_ns - creates a private namespace and adds a root filesystem
2885 * @mnt: pointer to the new root filesystem mountpoint
2886 */
2887static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2888{
2889        struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2890        if (!IS_ERR(new_ns)) {
2891                struct mount *mnt = real_mount(m);
2892                mnt->mnt_ns = new_ns;
2893                new_ns->root = mnt;
2894                list_add(&mnt->mnt_list, &new_ns->list);
2895        } else {
2896                mntput(m);
2897        }
2898        return new_ns;
2899}
2900
2901struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2902{
2903        struct mnt_namespace *ns;
2904        struct super_block *s;
2905        struct path path;
2906        int err;
2907
2908        ns = create_mnt_ns(mnt);
2909        if (IS_ERR(ns))
2910                return ERR_CAST(ns);
2911
2912        err = vfs_path_lookup(mnt->mnt_root, mnt,
2913                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2914
2915        put_mnt_ns(ns);
2916
2917        if (err)
2918                return ERR_PTR(err);
2919
2920        /* trade a vfsmount reference for active sb one */
2921        s = path.mnt->mnt_sb;
2922        atomic_inc(&s->s_active);
2923        mntput(path.mnt);
2924        /* lock the sucker */
2925        down_write(&s->s_umount);
2926        /* ... and return the root of (sub)tree on it */
2927        return path.dentry;
2928}
2929EXPORT_SYMBOL(mount_subtree);
2930
2931SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2932                char __user *, type, unsigned long, flags, void __user *, data)
2933{
2934        int ret;
2935        char *kernel_type;
2936        char *kernel_dev;
2937        unsigned long data_page;
2938
2939        kernel_type = copy_mount_string(type);
2940        ret = PTR_ERR(kernel_type);
2941        if (IS_ERR(kernel_type))
2942                goto out_type;
2943
2944        kernel_dev = copy_mount_string(dev_name);
2945        ret = PTR_ERR(kernel_dev);
2946        if (IS_ERR(kernel_dev))
2947                goto out_dev;
2948
2949        ret = copy_mount_options(data, &data_page);
2950        if (ret < 0)
2951                goto out_data;
2952
2953        ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2954                (void *) data_page);
2955
2956        free_page(data_page);
2957out_data:
2958        kfree(kernel_dev);
2959out_dev:
2960        kfree(kernel_type);
2961out_type:
2962        return ret;
2963}
2964
2965/*
2966 * Return true if path is reachable from root
2967 *
2968 * namespace_sem or mount_lock is held
2969 */
2970bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2971                         const struct path *root)
2972{
2973        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2974                dentry = mnt->mnt_mountpoint;
2975                mnt = mnt->mnt_parent;
2976        }
2977        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2978}
2979
2980int path_is_under(struct path *path1, struct path *path2)
2981{
2982        int res;
2983        read_seqlock_excl(&mount_lock);
2984        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2985        read_sequnlock_excl(&mount_lock);
2986        return res;
2987}
2988EXPORT_SYMBOL(path_is_under);
2989
2990/*
2991 * pivot_root Semantics:
2992 * Moves the root file system of the current process to the directory put_old,
2993 * makes new_root as the new root file system of the current process, and sets
2994 * root/cwd of all processes which had them on the current root to new_root.
2995 *
2996 * Restrictions:
2997 * The new_root and put_old must be directories, and  must not be on the
2998 * same file  system as the current process root. The put_old  must  be
2999 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3000 * pointed to by put_old must yield the same directory as new_root. No other
3001 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3002 *
3003 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3004 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3005 * in this situation.
3006 *
3007 * Notes:
3008 *  - we don't move root/cwd if they are not at the root (reason: if something
3009 *    cared enough to change them, it's probably wrong to force them elsewhere)
3010 *  - it's okay to pick a root that isn't the root of a file system, e.g.
3011 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3012 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3013 *    first.
3014 */
3015SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3016                const char __user *, put_old)
3017{
3018        struct path new, old, parent_path, root_parent, root;
3019        struct mount *new_mnt, *root_mnt, *old_mnt;
3020        struct mountpoint *old_mp, *root_mp;
3021        int error;
3022
3023        if (!may_mount())
3024                return -EPERM;
3025
3026        error = user_path_dir(new_root, &new);
3027        if (error)
3028                goto out0;
3029
3030        error = user_path_dir(put_old, &old);
3031        if (error)
3032                goto out1;
3033
3034        error = security_sb_pivotroot(&old, &new);
3035        if (error)
3036                goto out2;
3037
3038        get_fs_root(current->fs, &root);
3039        old_mp = lock_mount(&old);
3040        error = PTR_ERR(old_mp);
3041        if (IS_ERR(old_mp))
3042                goto out3;
3043
3044        error = -EINVAL;
3045        new_mnt = real_mount(new.mnt);
3046        root_mnt = real_mount(root.mnt);
3047        old_mnt = real_mount(old.mnt);
3048        if (IS_MNT_SHARED(old_mnt) ||
3049                IS_MNT_SHARED(new_mnt->mnt_parent) ||
3050                IS_MNT_SHARED(root_mnt->mnt_parent))
3051                goto out4;
3052        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3053                goto out4;
3054        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3055                goto out4;
3056        error = -ENOENT;
3057        if (d_unlinked(new.dentry))
3058                goto out4;
3059        error = -EBUSY;
3060        if (new_mnt == root_mnt || old_mnt == root_mnt)
3061                goto out4; /* loop, on the same file system  */
3062        error = -EINVAL;
3063        if (root.mnt->mnt_root != root.dentry)
3064                goto out4; /* not a mountpoint */
3065        if (!mnt_has_parent(root_mnt))
3066                goto out4; /* not attached */
3067        root_mp = root_mnt->mnt_mp;
3068        if (new.mnt->mnt_root != new.dentry)
3069                goto out4; /* not a mountpoint */
3070        if (!mnt_has_parent(new_mnt))
3071                goto out4; /* not attached */
3072        /* make sure we can reach put_old from new_root */
3073        if (!is_path_reachable(old_mnt, old.dentry, &new))
3074                goto out4;
3075        /* make certain new is below the root */
3076        if (!is_path_reachable(new_mnt, new.dentry, &root))
3077                goto out4;
3078        root_mp->m_count++; /* pin it so it won't go away */
3079        lock_mount_hash();
3080        detach_mnt(new_mnt, &parent_path);
3081        detach_mnt(root_mnt, &root_parent);
3082        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3083                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3084                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3085        }
3086        /* mount old root on put_old */
3087        attach_mnt(root_mnt, old_mnt, old_mp);
3088        /* mount new_root on / */
3089        attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3090        touch_mnt_namespace(current->nsproxy->mnt_ns);
3091        /* A moved mount should not expire automatically */
3092        list_del_init(&new_mnt->mnt_expire);
3093        put_mountpoint(root_mp);
3094        unlock_mount_hash();
3095        chroot_fs_refs(&root, &new);
3096        error = 0;
3097out4:
3098        unlock_mount(old_mp);
3099        if (!error) {
3100                path_put(&root_parent);
3101                path_put(&parent_path);
3102        }
3103out3:
3104        path_put(&root);
3105out2:
3106        path_put(&old);
3107out1:
3108        path_put(&new);
3109out0:
3110        return error;
3111}
3112
3113static void __init init_mount_tree(void)
3114{
3115        struct vfsmount *mnt;
3116        struct mnt_namespace *ns;
3117        struct path root;
3118        struct file_system_type *type;
3119
3120        type = get_fs_type("rootfs");
3121        if (!type)
3122                panic("Can't find rootfs type");
3123        mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3124        put_filesystem(type);
3125        if (IS_ERR(mnt))
3126                panic("Can't create rootfs");
3127
3128        ns = create_mnt_ns(mnt);
3129        if (IS_ERR(ns))
3130                panic("Can't allocate initial namespace");
3131
3132        init_task.nsproxy->mnt_ns = ns;
3133        get_mnt_ns(ns);
3134
3135        root.mnt = mnt;
3136        root.dentry = mnt->mnt_root;
3137        mnt->mnt_flags |= MNT_LOCKED;
3138
3139        set_fs_pwd(current->fs, &root);
3140        set_fs_root(current->fs, &root);
3141}
3142
3143void __init mnt_init(void)
3144{
3145        unsigned u;
3146        int err;
3147
3148        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3149                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3150
3151        mount_hashtable = alloc_large_system_hash("Mount-cache",
3152                                sizeof(struct hlist_head),
3153                                mhash_entries, 19,
3154                                0,
3155                                &m_hash_shift, &m_hash_mask, 0, 0);
3156        mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3157                                sizeof(struct hlist_head),
3158                                mphash_entries, 19,
3159                                0,
3160                                &mp_hash_shift, &mp_hash_mask, 0, 0);
3161
3162        if (!mount_hashtable || !mountpoint_hashtable)
3163                panic("Failed to allocate mount hash table\n");
3164
3165        for (u = 0; u <= m_hash_mask; u++)
3166                INIT_HLIST_HEAD(&mount_hashtable[u]);
3167        for (u = 0; u <= mp_hash_mask; u++)
3168                INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3169
3170        kernfs_init();
3171
3172        err = sysfs_init();
3173        if (err)
3174                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3175                        __func__, err);
3176        fs_kobj = kobject_create_and_add("fs", NULL);
3177        if (!fs_kobj)
3178                printk(KERN_WARNING "%s: kobj create error\n", __func__);
3179        init_rootfs();
3180        init_mount_tree();
3181}
3182
3183void put_mnt_ns(struct mnt_namespace *ns)
3184{
3185        if (!atomic_dec_and_test(&ns->count))
3186                return;
3187        drop_collected_mounts(&ns->root->mnt);
3188        free_mnt_ns(ns);
3189}
3190
3191struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3192{
3193        struct vfsmount *mnt;
3194        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3195        if (!IS_ERR(mnt)) {
3196                /*
3197                 * it is a longterm mount, don't release mnt until
3198                 * we unmount before file sys is unregistered
3199                */
3200                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3201        }
3202        return mnt;
3203}
3204EXPORT_SYMBOL_GPL(kern_mount_data);
3205
3206void kern_unmount(struct vfsmount *mnt)
3207{
3208        /* release long term mount so mount point can be released */
3209        if (!IS_ERR_OR_NULL(mnt)) {
3210                real_mount(mnt)->mnt_ns = NULL;
3211                synchronize_rcu();      /* yecchhh... */
3212                mntput(mnt);
3213        }
3214}
3215EXPORT_SYMBOL(kern_unmount);
3216
3217bool our_mnt(struct vfsmount *mnt)
3218{
3219        return check_mnt(real_mount(mnt));
3220}
3221
3222bool current_chrooted(void)
3223{
3224        /* Does the current process have a non-standard root */
3225        struct path ns_root;
3226        struct path fs_root;
3227        bool chrooted;
3228
3229        /* Find the namespace root */
3230        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3231        ns_root.dentry = ns_root.mnt->mnt_root;
3232        path_get(&ns_root);
3233        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3234                ;
3235
3236        get_fs_root(current->fs, &fs_root);
3237
3238        chrooted = !path_equal(&fs_root, &ns_root);
3239
3240        path_put(&fs_root);
3241        path_put(&ns_root);
3242
3243        return chrooted;
3244}
3245
3246static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3247{
3248        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3249        int new_flags = *new_mnt_flags;
3250        struct mount *mnt;
3251        bool visible = false;
3252
3253        if (unlikely(!ns))
3254                return false;
3255
3256        down_read(&namespace_sem);
3257        list_for_each_entry(mnt, &ns->list, mnt_list) {
3258                struct mount *child;
3259                int mnt_flags;
3260
3261                if (mnt->mnt.mnt_sb->s_type != type)
3262                        continue;
3263
3264                /* This mount is not fully visible if it's root directory
3265                 * is not the root directory of the filesystem.
3266                 */
3267                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3268                        continue;
3269
3270                /* Read the mount flags and filter out flags that
3271                 * may safely be ignored.
3272                 */
3273                mnt_flags = mnt->mnt.mnt_flags;
3274                if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3275                        mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3276
3277                /* Don't miss readonly hidden in the superblock flags */
3278                if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3279                        mnt_flags |= MNT_LOCK_READONLY;
3280
3281                /* Verify the mount flags are equal to or more permissive
3282                 * than the proposed new mount.
3283                 */
3284                if ((mnt_flags & MNT_LOCK_READONLY) &&
3285                    !(new_flags & MNT_READONLY))
3286                        continue;
3287                if ((mnt_flags & MNT_LOCK_NODEV) &&
3288                    !(new_flags & MNT_NODEV))
3289                        continue;
3290                if ((mnt_flags & MNT_LOCK_NOSUID) &&
3291                    !(new_flags & MNT_NOSUID))
3292                        continue;
3293                if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3294                    !(new_flags & MNT_NOEXEC))
3295                        continue;
3296                if ((mnt_flags & MNT_LOCK_ATIME) &&
3297                    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3298                        continue;
3299
3300                /* This mount is not fully visible if there are any
3301                 * locked child mounts that cover anything except for
3302                 * empty directories.
3303                 */
3304                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3305                        struct inode *inode = child->mnt_mountpoint->d_inode;
3306                        /* Only worry about locked mounts */
3307                        if (!(child->mnt.mnt_flags & MNT_LOCKED))
3308                                continue;
3309                        /* Is the directory permanetly empty? */
3310                        if (!is_empty_dir_inode(inode))
3311                                goto next;
3312                }
3313                /* Preserve the locked attributes */
3314                *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3315                                               MNT_LOCK_NODEV    | \
3316                                               MNT_LOCK_NOSUID   | \
3317                                               MNT_LOCK_NOEXEC   | \
3318                                               MNT_LOCK_ATIME);
3319                visible = true;
3320                goto found;
3321        next:   ;
3322        }
3323found:
3324        up_read(&namespace_sem);
3325        return visible;
3326}
3327
3328static struct ns_common *mntns_get(struct task_struct *task)
3329{
3330        struct ns_common *ns = NULL;
3331        struct nsproxy *nsproxy;
3332
3333        task_lock(task);
3334        nsproxy = task->nsproxy;
3335        if (nsproxy) {
3336                ns = &nsproxy->mnt_ns->ns;
3337                get_mnt_ns(to_mnt_ns(ns));
3338        }
3339        task_unlock(task);
3340
3341        return ns;
3342}
3343
3344static void mntns_put(struct ns_common *ns)
3345{
3346        put_mnt_ns(to_mnt_ns(ns));
3347}
3348
3349static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3350{
3351        struct fs_struct *fs = current->fs;
3352        struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3353        struct path root;
3354
3355        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3356            !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3357            !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3358                return -EPERM;
3359
3360        if (fs->users != 1)
3361                return -EINVAL;
3362
3363        get_mnt_ns(mnt_ns);
3364        put_mnt_ns(nsproxy->mnt_ns);
3365        nsproxy->mnt_ns = mnt_ns;
3366
3367        /* Find the root */
3368        root.mnt    = &mnt_ns->root->mnt;
3369        root.dentry = mnt_ns->root->mnt.mnt_root;
3370        path_get(&root);
3371        while(d_mountpoint(root.dentry) && follow_down_one(&root))
3372                ;
3373
3374        /* Update the pwd and root */
3375        set_fs_pwd(fs, &root);
3376        set_fs_root(fs, &root);
3377
3378        path_put(&root);
3379        return 0;
3380}
3381
3382const struct proc_ns_operations mntns_operations = {
3383        .name           = "mnt",
3384        .type           = CLONE_NEWNS,
3385        .get            = mntns_get,
3386        .put            = mntns_put,
3387        .install        = mntns_install,
3388};
Note: See TracBrowser for help on using the repository browser.