Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/fs/namespace.c

  1 /*
  2  *  linux/fs/namespace.c
  3  *
  4  * (C) Copyright Al Viro 2000, 2001
  5  *      Released under GPL v2.
  6  *
  7  * Based on code from fs/super.c, copyright Linus Torvalds and others.
  8  * Heavily rewritten.
  9  */
 10 
 11 #include <linux/syscalls.h>
 12 #include <linux/export.h>
 13 #include <linux/capability.h>
 14 #include <linux/mnt_namespace.h>
 15 #include <linux/user_namespace.h>
 16 #include <linux/namei.h>
 17 #include <linux/security.h>
 18 #include <linux/idr.h>
 19 #include <linux/init.h>         /* init_rootfs */
 20 #include <linux/fs_struct.h>    /* get_fs_root et.al. */
 21 #include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
 22 #include <linux/uaccess.h>
 23 #include <linux/proc_ns.h>
 24 #include <linux/magic.h>
 25 #include <linux/bootmem.h>
 26 #include <linux/task_work.h>
 27 #include "pnode.h"
 28 #include "internal.h"
 29 
 30 /* Maximum number of mounts in a mount namespace */
 31 unsigned int sysctl_mount_max __read_mostly = 100000;
 32 
 33 static unsigned int m_hash_mask __read_mostly;
 34 static unsigned int m_hash_shift __read_mostly;
 35 static unsigned int mp_hash_mask __read_mostly;
 36 static unsigned int mp_hash_shift __read_mostly;
 37 
 38 static __initdata unsigned long mhash_entries;
 39 static int __init set_mhash_entries(char *str)
 40 {
 41         if (!str)
 42                 return 0;
 43         mhash_entries = simple_strtoul(str, &str, 0);
 44         return 1;
 45 }
 46 __setup("mhash_entries=", set_mhash_entries);
 47 
 48 static __initdata unsigned long mphash_entries;
 49 static int __init set_mphash_entries(char *str)
 50 {
 51         if (!str)
 52                 return 0;
 53         mphash_entries = simple_strtoul(str, &str, 0);
 54         return 1;
 55 }
 56 __setup("mphash_entries=", set_mphash_entries);
 57 
 58 static u64 event;
 59 static DEFINE_IDA(mnt_id_ida);
 60 static DEFINE_IDA(mnt_group_ida);
 61 static DEFINE_SPINLOCK(mnt_id_lock);
 62 static int mnt_id_start = 0;
 63 static int mnt_group_start = 1;
 64 
 65 static struct hlist_head *mount_hashtable __read_mostly;
 66 static struct hlist_head *mountpoint_hashtable __read_mostly;
 67 static struct kmem_cache *mnt_cache __read_mostly;
 68 static DECLARE_RWSEM(namespace_sem);
 69 
 70 /* /sys/fs */
 71 struct kobject *fs_kobj;
 72 EXPORT_SYMBOL_GPL(fs_kobj);
 73 
 74 /*
 75  * vfsmount lock may be taken for read to prevent changes to the
 76  * vfsmount hash, ie. during mountpoint lookups or walking back
 77  * up the tree.
 78  *
 79  * It should be taken for write in all cases where the vfsmount
 80  * tree or hash is modified or when a vfsmount structure is modified.
 81  */
 82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
 83 
 84 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
 85 {
 86         unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
 87         tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
 88         tmp = tmp + (tmp >> m_hash_shift);
 89         return &mount_hashtable[tmp & m_hash_mask];
 90 }
 91 
 92 static inline struct hlist_head *mp_hash(struct dentry *dentry)
 93 {
 94         unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
 95         tmp = tmp + (tmp >> mp_hash_shift);
 96         return &mountpoint_hashtable[tmp & mp_hash_mask];
 97 }
 98 
 99 static int mnt_alloc_id(struct mount *mnt)
100 {
101         int res;
102 
103 retry:
104         ida_pre_get(&mnt_id_ida, GFP_KERNEL);
105         spin_lock(&mnt_id_lock);
106         res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
107         if (!res)
108                 mnt_id_start = mnt->mnt_id + 1;
109         spin_unlock(&mnt_id_lock);
110         if (res == -EAGAIN)
111                 goto retry;
112 
113         return res;
114 }
115 
116 static void mnt_free_id(struct mount *mnt)
117 {
118         int id = mnt->mnt_id;
119         spin_lock(&mnt_id_lock);
120         ida_remove(&mnt_id_ida, id);
121         if (mnt_id_start > id)
122                 mnt_id_start = id;
123         spin_unlock(&mnt_id_lock);
124 }
125 
126 /*
127  * Allocate a new peer group ID
128  *
129  * mnt_group_ida is protected by namespace_sem
130  */
131 static int mnt_alloc_group_id(struct mount *mnt)
132 {
133         int res;
134 
135         if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
136                 return -ENOMEM;
137 
138         res = ida_get_new_above(&mnt_group_ida,
139                                 mnt_group_start,
140                                 &mnt->mnt_group_id);
141         if (!res)
142                 mnt_group_start = mnt->mnt_group_id + 1;
143 
144         return res;
145 }
146 
147 /*
148  * Release a peer group ID
149  */
150 void mnt_release_group_id(struct mount *mnt)
151 {
152         int id = mnt->mnt_group_id;
153         ida_remove(&mnt_group_ida, id);
154         if (mnt_group_start > id)
155                 mnt_group_start = id;
156         mnt->mnt_group_id = 0;
157 }
158 
159 /*
160  * vfsmount lock must be held for read
161  */
162 static inline void mnt_add_count(struct mount *mnt, int n)
163 {
164 #ifdef CONFIG_SMP
165         this_cpu_add(mnt->mnt_pcp->mnt_count, n);
166 #else
167         preempt_disable();
168         mnt->mnt_count += n;
169         preempt_enable();
170 #endif
171 }
172 
173 /*
174  * vfsmount lock must be held for write
175  */
176 unsigned int mnt_get_count(struct mount *mnt)
177 {
178 #ifdef CONFIG_SMP
179         unsigned int count = 0;
180         int cpu;
181 
182         for_each_possible_cpu(cpu) {
183                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
184         }
185 
186         return count;
187 #else
188         return mnt->mnt_count;
189 #endif
190 }
191 
192 static void drop_mountpoint(struct fs_pin *p)
193 {
194         struct mount *m = container_of(p, struct mount, mnt_umount);
195         dput(m->mnt_ex_mountpoint);
196         pin_remove(p);
197         mntput(&m->mnt);
198 }
199 
200 static struct mount *alloc_vfsmnt(const char *name)
201 {
202         struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
203         if (mnt) {
204                 int err;
205 
206                 err = mnt_alloc_id(mnt);
207                 if (err)
208                         goto out_free_cache;
209 
210                 if (name) {
211                         mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
212                         if (!mnt->mnt_devname)
213                                 goto out_free_id;
214                 }
215 
216 #ifdef CONFIG_SMP
217                 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
218                 if (!mnt->mnt_pcp)
219                         goto out_free_devname;
220 
221                 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
222 #else
223                 mnt->mnt_count = 1;
224                 mnt->mnt_writers = 0;
225 #endif
226 
227                 INIT_HLIST_NODE(&mnt->mnt_hash);
228                 INIT_LIST_HEAD(&mnt->mnt_child);
229                 INIT_LIST_HEAD(&mnt->mnt_mounts);
230                 INIT_LIST_HEAD(&mnt->mnt_list);
231                 INIT_LIST_HEAD(&mnt->mnt_expire);
232                 INIT_LIST_HEAD(&mnt->mnt_share);
233                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
234                 INIT_LIST_HEAD(&mnt->mnt_slave);
235                 INIT_HLIST_NODE(&mnt->mnt_mp_list);
236 #ifdef CONFIG_FSNOTIFY
237                 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
238 #endif
239                 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
240         }
241         return mnt;
242 
243 #ifdef CONFIG_SMP
244 out_free_devname:
245         kfree_const(mnt->mnt_devname);
246 #endif
247 out_free_id:
248         mnt_free_id(mnt);
249 out_free_cache:
250         kmem_cache_free(mnt_cache, mnt);
251         return NULL;
252 }
253 
254 /*
255  * Most r/o checks on a fs are for operations that take
256  * discrete amounts of time, like a write() or unlink().
257  * We must keep track of when those operations start
258  * (for permission checks) and when they end, so that
259  * we can determine when writes are able to occur to
260  * a filesystem.
261  */
262 /*
263  * __mnt_is_readonly: check whether a mount is read-only
264  * @mnt: the mount to check for its write status
265  *
266  * This shouldn't be used directly ouside of the VFS.
267  * It does not guarantee that the filesystem will stay
268  * r/w, just that it is right *now*.  This can not and
269  * should not be used in place of IS_RDONLY(inode).
270  * mnt_want/drop_write() will _keep_ the filesystem
271  * r/w.
272  */
273 int __mnt_is_readonly(struct vfsmount *mnt)
274 {
275         if (mnt->mnt_flags & MNT_READONLY)
276                 return 1;
277         if (mnt->mnt_sb->s_flags & MS_RDONLY)
278                 return 1;
279         return 0;
280 }
281 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
282 
283 static inline void mnt_inc_writers(struct mount *mnt)
284 {
285 #ifdef CONFIG_SMP
286         this_cpu_inc(mnt->mnt_pcp->mnt_writers);
287 #else
288         mnt->mnt_writers++;
289 #endif
290 }
291 
292 static inline void mnt_dec_writers(struct mount *mnt)
293 {
294 #ifdef CONFIG_SMP
295         this_cpu_dec(mnt->mnt_pcp->mnt_writers);
296 #else
297         mnt->mnt_writers--;
298 #endif
299 }
300 
301 static unsigned int mnt_get_writers(struct mount *mnt)
302 {
303 #ifdef CONFIG_SMP
304         unsigned int count = 0;
305         int cpu;
306 
307         for_each_possible_cpu(cpu) {
308                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
309         }
310 
311         return count;
312 #else
313         return mnt->mnt_writers;
314 #endif
315 }
316 
317 static int mnt_is_readonly(struct vfsmount *mnt)
318 {
319         if (mnt->mnt_sb->s_readonly_remount)
320                 return 1;
321         /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
322         smp_rmb();
323         return __mnt_is_readonly(mnt);
324 }
325 
326 /*
327  * Most r/o & frozen checks on a fs are for operations that take discrete
328  * amounts of time, like a write() or unlink().  We must keep track of when
329  * those operations start (for permission checks) and when they end, so that we
330  * can determine when writes are able to occur to a filesystem.
331  */
332 /**
333  * __mnt_want_write - get write access to a mount without freeze protection
334  * @m: the mount on which to take a write
335  *
336  * This tells the low-level filesystem that a write is about to be performed to
337  * it, and makes sure that writes are allowed (mnt it read-write) before
338  * returning success. This operation does not protect against filesystem being
339  * frozen. When the write operation is finished, __mnt_drop_write() must be
340  * called. This is effectively a refcount.
341  */
342 int __mnt_want_write(struct vfsmount *m)
343 {
344         struct mount *mnt = real_mount(m);
345         int ret = 0;
346 
347         preempt_disable();
348         mnt_inc_writers(mnt);
349         /*
350          * The store to mnt_inc_writers must be visible before we pass
351          * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352          * incremented count after it has set MNT_WRITE_HOLD.
353          */
354         smp_mb();
355         while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
356                 cpu_relax();
357         /*
358          * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
359          * be set to match its requirements. So we must not load that until
360          * MNT_WRITE_HOLD is cleared.
361          */
362         smp_rmb();
363         if (mnt_is_readonly(m)) {
364                 mnt_dec_writers(mnt);
365                 ret = -EROFS;
366         }
367         preempt_enable();
368 
369         return ret;
370 }
371 
372 /**
373  * mnt_want_write - get write access to a mount
374  * @m: the mount on which to take a write
375  *
376  * This tells the low-level filesystem that a write is about to be performed to
377  * it, and makes sure that writes are allowed (mount is read-write, filesystem
378  * is not frozen) before returning success.  When the write operation is
379  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
380  */
381 int mnt_want_write(struct vfsmount *m)
382 {
383         int ret;
384 
385         sb_start_write(m->mnt_sb);
386         ret = __mnt_want_write(m);
387         if (ret)
388                 sb_end_write(m->mnt_sb);
389         return ret;
390 }
391 EXPORT_SYMBOL_GPL(mnt_want_write);
392 
393 /**
394  * mnt_clone_write - get write access to a mount
395  * @mnt: the mount on which to take a write
396  *
397  * This is effectively like mnt_want_write, except
398  * it must only be used to take an extra write reference
399  * on a mountpoint that we already know has a write reference
400  * on it. This allows some optimisation.
401  *
402  * After finished, mnt_drop_write must be called as usual to
403  * drop the reference.
404  */
405 int mnt_clone_write(struct vfsmount *mnt)
406 {
407         /* superblock may be r/o */
408         if (__mnt_is_readonly(mnt))
409                 return -EROFS;
410         preempt_disable();
411         mnt_inc_writers(real_mount(mnt));
412         preempt_enable();
413         return 0;
414 }
415 EXPORT_SYMBOL_GPL(mnt_clone_write);
416 
417 /**
418  * __mnt_want_write_file - get write access to a file's mount
419  * @file: the file who's mount on which to take a write
420  *
421  * This is like __mnt_want_write, but it takes a file and can
422  * do some optimisations if the file is open for write already
423  */
424 int __mnt_want_write_file(struct file *file)
425 {
426         if (!(file->f_mode & FMODE_WRITER))
427                 return __mnt_want_write(file->f_path.mnt);
428         else
429                 return mnt_clone_write(file->f_path.mnt);
430 }
431 
432 /**
433  * mnt_want_write_file - get write access to a file's mount
434  * @file: the file who's mount on which to take a write
435  *
436  * This is like mnt_want_write, but it takes a file and can
437  * do some optimisations if the file is open for write already
438  */
439 int mnt_want_write_file(struct file *file)
440 {
441         int ret;
442 
443         sb_start_write(file->f_path.mnt->mnt_sb);
444         ret = __mnt_want_write_file(file);
445         if (ret)
446                 sb_end_write(file->f_path.mnt->mnt_sb);
447         return ret;
448 }
449 EXPORT_SYMBOL_GPL(mnt_want_write_file);
450 
451 /**
452  * __mnt_drop_write - give up write access to a mount
453  * @mnt: the mount on which to give up write access
454  *
455  * Tells the low-level filesystem that we are done
456  * performing writes to it.  Must be matched with
457  * __mnt_want_write() call above.
458  */
459 void __mnt_drop_write(struct vfsmount *mnt)
460 {
461         preempt_disable();
462         mnt_dec_writers(real_mount(mnt));
463         preempt_enable();
464 }
465 
466 /**
467  * mnt_drop_write - give up write access to a mount
468  * @mnt: the mount on which to give up write access
469  *
470  * Tells the low-level filesystem that we are done performing writes to it and
471  * also allows filesystem to be frozen again.  Must be matched with
472  * mnt_want_write() call above.
473  */
474 void mnt_drop_write(struct vfsmount *mnt)
475 {
476         __mnt_drop_write(mnt);
477         sb_end_write(mnt->mnt_sb);
478 }
479 EXPORT_SYMBOL_GPL(mnt_drop_write);
480 
481 void __mnt_drop_write_file(struct file *file)
482 {
483         __mnt_drop_write(file->f_path.mnt);
484 }
485 
486 void mnt_drop_write_file(struct file *file)
487 {
488         mnt_drop_write(file->f_path.mnt);
489 }
490 EXPORT_SYMBOL(mnt_drop_write_file);
491 
492 static int mnt_make_readonly(struct mount *mnt)
493 {
494         int ret = 0;
495 
496         lock_mount_hash();
497         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
498         /*
499          * After storing MNT_WRITE_HOLD, we'll read the counters. This store
500          * should be visible before we do.
501          */
502         smp_mb();
503 
504         /*
505          * With writers on hold, if this value is zero, then there are
506          * definitely no active writers (although held writers may subsequently
507          * increment the count, they'll have to wait, and decrement it after
508          * seeing MNT_READONLY).
509          *
510          * It is OK to have counter incremented on one CPU and decremented on
511          * another: the sum will add up correctly. The danger would be when we
512          * sum up each counter, if we read a counter before it is incremented,
513          * but then read another CPU's count which it has been subsequently
514          * decremented from -- we would see more decrements than we should.
515          * MNT_WRITE_HOLD protects against this scenario, because
516          * mnt_want_write first increments count, then smp_mb, then spins on
517          * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
518          * we're counting up here.
519          */
520         if (mnt_get_writers(mnt) > 0)
521                 ret = -EBUSY;
522         else
523                 mnt->mnt.mnt_flags |= MNT_READONLY;
524         /*
525          * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
526          * that become unheld will see MNT_READONLY.
527          */
528         smp_wmb();
529         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
530         unlock_mount_hash();
531         return ret;
532 }
533 
534 static void __mnt_unmake_readonly(struct mount *mnt)
535 {
536         lock_mount_hash();
537         mnt->mnt.mnt_flags &= ~MNT_READONLY;
538         unlock_mount_hash();
539 }
540 
541 int sb_prepare_remount_readonly(struct super_block *sb)
542 {
543         struct mount *mnt;
544         int err = 0;
545 
546         /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
547         if (atomic_long_read(&sb->s_remove_count))
548                 return -EBUSY;
549 
550         lock_mount_hash();
551         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
552                 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
553                         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
554                         smp_mb();
555                         if (mnt_get_writers(mnt) > 0) {
556                                 err = -EBUSY;
557                                 break;
558                         }
559                 }
560         }
561         if (!err && atomic_long_read(&sb->s_remove_count))
562                 err = -EBUSY;
563 
564         if (!err) {
565                 sb->s_readonly_remount = 1;
566                 smp_wmb();
567         }
568         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
569                 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
570                         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
571         }
572         unlock_mount_hash();
573 
574         return err;
575 }
576 
577 static void free_vfsmnt(struct mount *mnt)
578 {
579         kfree_const(mnt->mnt_devname);
580 #ifdef CONFIG_SMP
581         free_percpu(mnt->mnt_pcp);
582 #endif
583         kmem_cache_free(mnt_cache, mnt);
584 }
585 
586 static void delayed_free_vfsmnt(struct rcu_head *head)
587 {
588         free_vfsmnt(container_of(head, struct mount, mnt_rcu));
589 }
590 
591 /* call under rcu_read_lock */
592 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
593 {
594         struct mount *mnt;
595         if (read_seqretry(&mount_lock, seq))
596                 return 1;
597         if (bastard == NULL)
598                 return 0;
599         mnt = real_mount(bastard);
600         mnt_add_count(mnt, 1);
601         if (likely(!read_seqretry(&mount_lock, seq)))
602                 return 0;
603         if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
604                 mnt_add_count(mnt, -1);
605                 return 1;
606         }
607         return -1;
608 }
609 
610 /* call under rcu_read_lock */
611 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
612 {
613         int res = __legitimize_mnt(bastard, seq);
614         if (likely(!res))
615                 return true;
616         if (unlikely(res < 0)) {
617                 rcu_read_unlock();
618                 mntput(bastard);
619                 rcu_read_lock();
620         }
621         return false;
622 }
623 
624 /*
625  * find the first mount at @dentry on vfsmount @mnt.
626  * call under rcu_read_lock()
627  */
628 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
629 {
630         struct hlist_head *head = m_hash(mnt, dentry);
631         struct mount *p;
632 
633         hlist_for_each_entry_rcu(p, head, mnt_hash)
634                 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
635                         return p;
636         return NULL;
637 }
638 
639 /*
640  * find the last mount at @dentry on vfsmount @mnt.
641  * mount_lock must be held.
642  */
643 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
644 {
645         struct mount *p, *res = NULL;
646         p = __lookup_mnt(mnt, dentry);
647         if (!p)
648                 goto out;
649         if (!(p->mnt.mnt_flags & MNT_UMOUNT))
650                 res = p;
651         hlist_for_each_entry_continue(p, mnt_hash) {
652                 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
653                         break;
654                 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
655                         res = p;
656         }
657 out:
658         return res;
659 }
660 
661 /*
662  * lookup_mnt - Return the first child mount mounted at path
663  *
664  * "First" means first mounted chronologically.  If you create the
665  * following mounts:
666  *
667  * mount /dev/sda1 /mnt
668  * mount /dev/sda2 /mnt
669  * mount /dev/sda3 /mnt
670  *
671  * Then lookup_mnt() on the base /mnt dentry in the root mount will
672  * return successively the root dentry and vfsmount of /dev/sda1, then
673  * /dev/sda2, then /dev/sda3, then NULL.
674  *
675  * lookup_mnt takes a reference to the found vfsmount.
676  */
677 struct vfsmount *lookup_mnt(const struct path *path)
678 {
679         struct mount *child_mnt;
680         struct vfsmount *m;
681         unsigned seq;
682 
683         rcu_read_lock();
684         do {
685                 seq = read_seqbegin(&mount_lock);
686                 child_mnt = __lookup_mnt(path->mnt, path->dentry);
687                 m = child_mnt ? &child_mnt->mnt : NULL;
688         } while (!legitimize_mnt(m, seq));
689         rcu_read_unlock();
690         return m;
691 }
692 
693 /*
694  * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
695  *                         current mount namespace.
696  *
697  * The common case is dentries are not mountpoints at all and that
698  * test is handled inline.  For the slow case when we are actually
699  * dealing with a mountpoint of some kind, walk through all of the
700  * mounts in the current mount namespace and test to see if the dentry
701  * is a mountpoint.
702  *
703  * The mount_hashtable is not usable in the context because we
704  * need to identify all mounts that may be in the current mount
705  * namespace not just a mount that happens to have some specified
706  * parent mount.
707  */
708 bool __is_local_mountpoint(struct dentry *dentry)
709 {
710         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
711         struct mount *mnt;
712         bool is_covered = false;
713 
714         if (!d_mountpoint(dentry))
715                 goto out;
716 
717         down_read(&namespace_sem);
718         list_for_each_entry(mnt, &ns->list, mnt_list) {
719                 is_covered = (mnt->mnt_mountpoint == dentry);
720                 if (is_covered)
721                         break;
722         }
723         up_read(&namespace_sem);
724 out:
725         return is_covered;
726 }
727 
728 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
729 {
730         struct hlist_head *chain = mp_hash(dentry);
731         struct mountpoint *mp;
732 
733         hlist_for_each_entry(mp, chain, m_hash) {
734                 if (mp->m_dentry == dentry) {
735                         /* might be worth a WARN_ON() */
736                         if (d_unlinked(dentry))
737                                 return ERR_PTR(-ENOENT);
738                         mp->m_count++;
739                         return mp;
740                 }
741         }
742         return NULL;
743 }
744 
745 static struct mountpoint *get_mountpoint(struct dentry *dentry)
746 {
747         struct mountpoint *mp, *new = NULL;
748         int ret;
749 
750         if (d_mountpoint(dentry)) {
751 mountpoint:
752                 read_seqlock_excl(&mount_lock);
753                 mp = lookup_mountpoint(dentry);
754                 read_sequnlock_excl(&mount_lock);
755                 if (mp)
756                         goto done;
757         }
758 
759         if (!new)
760                 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
761         if (!new)
762                 return ERR_PTR(-ENOMEM);
763 
764 
765         /* Exactly one processes may set d_mounted */
766         ret = d_set_mounted(dentry);
767 
768         /* Someone else set d_mounted? */
769         if (ret == -EBUSY)
770                 goto mountpoint;
771 
772         /* The dentry is not available as a mountpoint? */
773         mp = ERR_PTR(ret);
774         if (ret)
775                 goto done;
776 
777         /* Add the new mountpoint to the hash table */
778         read_seqlock_excl(&mount_lock);
779         new->m_dentry = dentry;
780         new->m_count = 1;
781         hlist_add_head(&new->m_hash, mp_hash(dentry));
782         INIT_HLIST_HEAD(&new->m_list);
783         read_sequnlock_excl(&mount_lock);
784 
785         mp = new;
786         new = NULL;
787 done:
788         kfree(new);
789         return mp;
790 }
791 
792 static void put_mountpoint(struct mountpoint *mp)
793 {
794         if (!--mp->m_count) {
795                 struct dentry *dentry = mp->m_dentry;
796                 BUG_ON(!hlist_empty(&mp->m_list));
797                 spin_lock(&dentry->d_lock);
798                 dentry->d_flags &= ~DCACHE_MOUNTED;
799                 spin_unlock(&dentry->d_lock);
800                 hlist_del(&mp->m_hash);
801                 kfree(mp);
802         }
803 }
804 
805 static inline int check_mnt(struct mount *mnt)
806 {
807         return mnt->mnt_ns == current->nsproxy->mnt_ns;
808 }
809 
810 /*
811  * vfsmount lock must be held for write
812  */
813 static void touch_mnt_namespace(struct mnt_namespace *ns)
814 {
815         if (ns) {
816                 ns->event = ++event;
817                 wake_up_interruptible(&ns->poll);
818         }
819 }
820 
821 /*
822  * vfsmount lock must be held for write
823  */
824 static void __touch_mnt_namespace(struct mnt_namespace *ns)
825 {
826         if (ns && ns->event != event) {
827                 ns->event = event;
828                 wake_up_interruptible(&ns->poll);
829         }
830 }
831 
832 /*
833  * vfsmount lock must be held for write
834  */
835 static void unhash_mnt(struct mount *mnt)
836 {
837         mnt->mnt_parent = mnt;
838         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
839         list_del_init(&mnt->mnt_child);
840         hlist_del_init_rcu(&mnt->mnt_hash);
841         hlist_del_init(&mnt->mnt_mp_list);
842         put_mountpoint(mnt->mnt_mp);
843         mnt->mnt_mp = NULL;
844 }
845 
846 /*
847  * vfsmount lock must be held for write
848  */
849 static void detach_mnt(struct mount *mnt, struct path *old_path)
850 {
851         old_path->dentry = mnt->mnt_mountpoint;
852         old_path->mnt = &mnt->mnt_parent->mnt;
853         unhash_mnt(mnt);
854 }
855 
856 /*
857  * vfsmount lock must be held for write
858  */
859 static void umount_mnt(struct mount *mnt)
860 {
861         /* old mountpoint will be dropped when we can do that */
862         mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
863         unhash_mnt(mnt);
864 }
865 
866 /*
867  * vfsmount lock must be held for write
868  */
869 void mnt_set_mountpoint(struct mount *mnt,
870                         struct mountpoint *mp,
871                         struct mount *child_mnt)
872 {
873         mp->m_count++;
874         mnt_add_count(mnt, 1);  /* essentially, that's mntget */
875         child_mnt->mnt_mountpoint = dget(mp->m_dentry);
876         child_mnt->mnt_parent = mnt;
877         child_mnt->mnt_mp = mp;
878         hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
879 }
880 
881 /*
882  * vfsmount lock must be held for write
883  */
884 static void attach_mnt(struct mount *mnt,
885                         struct mount *parent,
886                         struct mountpoint *mp)
887 {
888         mnt_set_mountpoint(parent, mp, mnt);
889         hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
890         list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
891 }
892 
893 static void attach_shadowed(struct mount *mnt,
894                         struct mount *parent,
895                         struct mount *shadows)
896 {
897         if (shadows) {
898                 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
899                 list_add(&mnt->mnt_child, &shadows->mnt_child);
900         } else {
901                 hlist_add_head_rcu(&mnt->mnt_hash,
902                                 m_hash(&parent->mnt, mnt->mnt_mountpoint));
903                 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
904         }
905 }
906 
907 /*
908  * vfsmount lock must be held for write
909  */
910 static void commit_tree(struct mount *mnt, struct mount *shadows)
911 {
912         struct mount *parent = mnt->mnt_parent;
913         struct mount *m;
914         LIST_HEAD(head);
915         struct mnt_namespace *n = parent->mnt_ns;
916 
917         BUG_ON(parent == mnt);
918 
919         list_add_tail(&head, &mnt->mnt_list);
920         list_for_each_entry(m, &head, mnt_list)
921                 m->mnt_ns = n;
922 
923         list_splice(&head, n->list.prev);
924 
925         n->mounts += n->pending_mounts;
926         n->pending_mounts = 0;
927 
928         attach_shadowed(mnt, parent, shadows);
929         touch_mnt_namespace(n);
930 }
931 
932 static 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 
948 static 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 
958 struct vfsmount *
959 vfs_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 }
990 EXPORT_SYMBOL_GPL(vfs_kern_mount);
991 
992 static 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         } else {
1058                 CLEAR_MNT_SHARED(mnt);
1059         }
1060         if (flag & CL_MAKE_SHARED)
1061                 set_mnt_shared(mnt);
1062 
1063         /* stick the duplicate mount on the same expiry list
1064          * as the original if that was on one */
1065         if (flag & CL_EXPIRE) {
1066                 if (!list_empty(&old->mnt_expire))
1067                         list_add(&mnt->mnt_expire, &old->mnt_expire);
1068         }
1069 
1070         return mnt;
1071 
1072  out_free:
1073         mnt_free_id(mnt);
1074         free_vfsmnt(mnt);
1075         return ERR_PTR(err);
1076 }
1077 
1078 static void cleanup_mnt(struct mount *mnt)
1079 {
1080         /*
1081          * This probably indicates that somebody messed
1082          * up a mnt_want/drop_write() pair.  If this
1083          * happens, the filesystem was probably unable
1084          * to make r/w->r/o transitions.
1085          */
1086         /*
1087          * The locking used to deal with mnt_count decrement provides barriers,
1088          * so mnt_get_writers() below is safe.
1089          */
1090         WARN_ON(mnt_get_writers(mnt));
1091         if (unlikely(mnt->mnt_pins.first))
1092                 mnt_pin_kill(mnt);
1093         fsnotify_vfsmount_delete(&mnt->mnt);
1094         dput(mnt->mnt.mnt_root);
1095         deactivate_super(mnt->mnt.mnt_sb);
1096         mnt_free_id(mnt);
1097         call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1098 }
1099 
1100 static void __cleanup_mnt(struct rcu_head *head)
1101 {
1102         cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1103 }
1104 
1105 static LLIST_HEAD(delayed_mntput_list);
1106 static void delayed_mntput(struct work_struct *unused)
1107 {
1108         struct llist_node *node = llist_del_all(&delayed_mntput_list);
1109         struct llist_node *next;
1110 
1111         for (; node; node = next) {
1112                 next = llist_next(node);
1113                 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1114         }
1115 }
1116 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1117 
1118 static void mntput_no_expire(struct mount *mnt)
1119 {
1120         rcu_read_lock();
1121         mnt_add_count(mnt, -1);
1122         if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1123                 rcu_read_unlock();
1124                 return;
1125         }
1126         lock_mount_hash();
1127         if (mnt_get_count(mnt)) {
1128                 rcu_read_unlock();
1129                 unlock_mount_hash();
1130                 return;
1131         }
1132         if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1133                 rcu_read_unlock();
1134                 unlock_mount_hash();
1135                 return;
1136         }
1137         mnt->mnt.mnt_flags |= MNT_DOOMED;
1138         rcu_read_unlock();
1139 
1140         list_del(&mnt->mnt_instance);
1141 
1142         if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1143                 struct mount *p, *tmp;
1144                 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1145                         umount_mnt(p);
1146                 }
1147         }
1148         unlock_mount_hash();
1149 
1150         if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1151                 struct task_struct *task = current;
1152                 if (likely(!(task->flags & PF_KTHREAD))) {
1153                         init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1154                         if (!task_work_add(task, &mnt->mnt_rcu, true))
1155                                 return;
1156                 }
1157                 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1158                         schedule_delayed_work(&delayed_mntput_work, 1);
1159                 return;
1160         }
1161         cleanup_mnt(mnt);
1162 }
1163 
1164 void mntput(struct vfsmount *mnt)
1165 {
1166         if (mnt) {
1167                 struct mount *m = real_mount(mnt);
1168                 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1169                 if (unlikely(m->mnt_expiry_mark))
1170                         m->mnt_expiry_mark = 0;
1171                 mntput_no_expire(m);
1172         }
1173 }
1174 EXPORT_SYMBOL(mntput);
1175 
1176 struct vfsmount *mntget(struct vfsmount *mnt)
1177 {
1178         if (mnt)
1179                 mnt_add_count(real_mount(mnt), 1);
1180         return mnt;
1181 }
1182 EXPORT_SYMBOL(mntget);
1183 
1184 /* path_is_mountpoint() - Check if path is a mount in the current
1185  *                          namespace.
1186  *
1187  *  d_mountpoint() can only be used reliably to establish if a dentry is
1188  *  not mounted in any namespace and that common case is handled inline.
1189  *  d_mountpoint() isn't aware of the possibility there may be multiple
1190  *  mounts using a given dentry in a different namespace. This function
1191  *  checks if the passed in path is a mountpoint rather than the dentry
1192  *  alone.
1193  */
1194 bool path_is_mountpoint(const struct path *path)
1195 {
1196         unsigned seq;
1197         bool res;
1198 
1199         if (!d_mountpoint(path->dentry))
1200                 return false;
1201 
1202         rcu_read_lock();
1203         do {
1204                 seq = read_seqbegin(&mount_lock);
1205                 res = __path_is_mountpoint(path);
1206         } while (read_seqretry(&mount_lock, seq));
1207         rcu_read_unlock();
1208 
1209         return res;
1210 }
1211 EXPORT_SYMBOL(path_is_mountpoint);
1212 
1213 struct vfsmount *mnt_clone_internal(const struct path *path)
1214 {
1215         struct mount *p;
1216         p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1217         if (IS_ERR(p))
1218                 return ERR_CAST(p);
1219         p->mnt.mnt_flags |= MNT_INTERNAL;
1220         return &p->mnt;
1221 }
1222 
1223 static inline void mangle(struct seq_file *m, const char *s)
1224 {
1225         seq_escape(m, s, " \t\n\\");
1226 }
1227 
1228 /*
1229  * Simple .show_options callback for filesystems which don't want to
1230  * implement more complex mount option showing.
1231  *
1232  * See also save_mount_options().
1233  */
1234 int generic_show_options(struct seq_file *m, struct dentry *root)
1235 {
1236         const char *options;
1237 
1238         rcu_read_lock();
1239         options = rcu_dereference(root->d_sb->s_options);
1240 
1241         if (options != NULL && options[0]) {
1242                 seq_putc(m, ',');
1243                 mangle(m, options);
1244         }
1245         rcu_read_unlock();
1246 
1247         return 0;
1248 }
1249 EXPORT_SYMBOL(generic_show_options);
1250 
1251 /*
1252  * If filesystem uses generic_show_options(), this function should be
1253  * called from the fill_super() callback.
1254  *
1255  * The .remount_fs callback usually needs to be handled in a special
1256  * way, to make sure, that previous options are not overwritten if the
1257  * remount fails.
1258  *
1259  * Also note, that if the filesystem's .remount_fs function doesn't
1260  * reset all options to their default value, but changes only newly
1261  * given options, then the displayed options will not reflect reality
1262  * any more.
1263  */
1264 void save_mount_options(struct super_block *sb, char *options)
1265 {
1266         BUG_ON(sb->s_options);
1267         rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1268 }
1269 EXPORT_SYMBOL(save_mount_options);
1270 
1271 void replace_mount_options(struct super_block *sb, char *options)
1272 {
1273         char *old = sb->s_options;
1274         rcu_assign_pointer(sb->s_options, options);
1275         if (old) {
1276                 synchronize_rcu();
1277                 kfree(old);
1278         }
1279 }
1280 EXPORT_SYMBOL(replace_mount_options);
1281 
1282 #ifdef CONFIG_PROC_FS
1283 /* iterator; we want it to have access to namespace_sem, thus here... */
1284 static void *m_start(struct seq_file *m, loff_t *pos)
1285 {
1286         struct proc_mounts *p = m->private;
1287 
1288         down_read(&namespace_sem);
1289         if (p->cached_event == p->ns->event) {
1290                 void *v = p->cached_mount;
1291                 if (*pos == p->cached_index)
1292                         return v;
1293                 if (*pos == p->cached_index + 1) {
1294                         v = seq_list_next(v, &p->ns->list, &p->cached_index);
1295                         return p->cached_mount = v;
1296                 }
1297         }
1298 
1299         p->cached_event = p->ns->event;
1300         p->cached_mount = seq_list_start(&p->ns->list, *pos);
1301         p->cached_index = *pos;
1302         return p->cached_mount;
1303 }
1304 
1305 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1306 {
1307         struct proc_mounts *p = m->private;
1308 
1309         p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1310         p->cached_index = *pos;
1311         return p->cached_mount;
1312 }
1313 
1314 static void m_stop(struct seq_file *m, void *v)
1315 {
1316         up_read(&namespace_sem);
1317 }
1318 
1319 static int m_show(struct seq_file *m, void *v)
1320 {
1321         struct proc_mounts *p = m->private;
1322         struct mount *r = list_entry(v, struct mount, mnt_list);
1323         return p->show(m, &r->mnt);
1324 }
1325 
1326 const struct seq_operations mounts_op = {
1327         .start  = m_start,
1328         .next   = m_next,
1329         .stop   = m_stop,
1330         .show   = m_show,
1331 };
1332 #endif  /* CONFIG_PROC_FS */
1333 
1334 /**
1335  * may_umount_tree - check if a mount tree is busy
1336  * @mnt: root of mount tree
1337  *
1338  * This is called to check if a tree of mounts has any
1339  * open files, pwds, chroots or sub mounts that are
1340  * busy.
1341  */
1342 int may_umount_tree(struct vfsmount *m)
1343 {
1344         struct mount *mnt = real_mount(m);
1345         int actual_refs = 0;
1346         int minimum_refs = 0;
1347         struct mount *p;
1348         BUG_ON(!m);
1349 
1350         /* write lock needed for mnt_get_count */
1351         lock_mount_hash();
1352         for (p = mnt; p; p = next_mnt(p, mnt)) {
1353                 actual_refs += mnt_get_count(p);
1354                 minimum_refs += 2;
1355         }
1356         unlock_mount_hash();
1357 
1358         if (actual_refs > minimum_refs)
1359                 return 0;
1360 
1361         return 1;
1362 }
1363 
1364 EXPORT_SYMBOL(may_umount_tree);
1365 
1366 /**
1367  * may_umount - check if a mount point is busy
1368  * @mnt: root of mount
1369  *
1370  * This is called to check if a mount point has any
1371  * open files, pwds, chroots or sub mounts. If the
1372  * mount has sub mounts this will return busy
1373  * regardless of whether the sub mounts are busy.
1374  *
1375  * Doesn't take quota and stuff into account. IOW, in some cases it will
1376  * give false negatives. The main reason why it's here is that we need
1377  * a non-destructive way to look for easily umountable filesystems.
1378  */
1379 int may_umount(struct vfsmount *mnt)
1380 {
1381         int ret = 1;
1382         down_read(&namespace_sem);
1383         lock_mount_hash();
1384         if (propagate_mount_busy(real_mount(mnt), 2))
1385                 ret = 0;
1386         unlock_mount_hash();
1387         up_read(&namespace_sem);
1388         return ret;
1389 }
1390 
1391 EXPORT_SYMBOL(may_umount);
1392 
1393 static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
1394 
1395 static void namespace_unlock(void)
1396 {
1397         struct hlist_head head;
1398 
1399         hlist_move_list(&unmounted, &head);
1400 
1401         up_write(&namespace_sem);
1402 
1403         if (likely(hlist_empty(&head)))
1404                 return;
1405 
1406         synchronize_rcu();
1407 
1408         group_pin_kill(&head);
1409 }
1410 
1411 static inline void namespace_lock(void)
1412 {
1413         down_write(&namespace_sem);
1414 }
1415 
1416 enum umount_tree_flags {
1417         UMOUNT_SYNC = 1,
1418         UMOUNT_PROPAGATE = 2,
1419         UMOUNT_CONNECTED = 4,
1420 };
1421 
1422 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1423 {
1424         /* Leaving mounts connected is only valid for lazy umounts */
1425         if (how & UMOUNT_SYNC)
1426                 return true;
1427 
1428         /* A mount without a parent has nothing to be connected to */
1429         if (!mnt_has_parent(mnt))
1430                 return true;
1431 
1432         /* Because the reference counting rules change when mounts are
1433          * unmounted and connected, umounted mounts may not be
1434          * connected to mounted mounts.
1435          */
1436         if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1437                 return true;
1438 
1439         /* Has it been requested that the mount remain connected? */
1440         if (how & UMOUNT_CONNECTED)
1441                 return false;
1442 
1443         /* Is the mount locked such that it needs to remain connected? */
1444         if (IS_MNT_LOCKED(mnt))
1445                 return false;
1446 
1447         /* By default disconnect the mount */
1448         return true;
1449 }
1450 
1451 /*
1452  * mount_lock must be held
1453  * namespace_sem must be held for write
1454  */
1455 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1456 {
1457         LIST_HEAD(tmp_list);
1458         struct mount *p;
1459 
1460         if (how & UMOUNT_PROPAGATE)
1461                 propagate_mount_unlock(mnt);
1462 
1463         /* Gather the mounts to umount */
1464         for (p = mnt; p; p = next_mnt(p, mnt)) {
1465                 p->mnt.mnt_flags |= MNT_UMOUNT;
1466                 list_move(&p->mnt_list, &tmp_list);
1467         }
1468 
1469         /* Hide the mounts from mnt_mounts */
1470         list_for_each_entry(p, &tmp_list, mnt_list) {
1471                 list_del_init(&p->mnt_child);
1472         }
1473 
1474         /* Add propogated mounts to the tmp_list */
1475         if (how & UMOUNT_PROPAGATE)
1476                 propagate_umount(&tmp_list);
1477 
1478         while (!list_empty(&tmp_list)) {
1479                 struct mnt_namespace *ns;
1480                 bool disconnect;
1481                 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1482                 list_del_init(&p->mnt_expire);
1483                 list_del_init(&p->mnt_list);
1484                 ns = p->mnt_ns;
1485                 if (ns) {
1486                         ns->mounts--;
1487                         __touch_mnt_namespace(ns);
1488                 }
1489                 p->mnt_ns = NULL;
1490                 if (how & UMOUNT_SYNC)
1491                         p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1492 
1493                 disconnect = disconnect_mount(p, how);
1494 
1495                 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1496                                  disconnect ? &unmounted : NULL);
1497                 if (mnt_has_parent(p)) {
1498                         mnt_add_count(p->mnt_parent, -1);
1499                         if (!disconnect) {
1500                                 /* Don't forget about p */
1501                                 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1502                         } else {
1503                                 umount_mnt(p);
1504                         }
1505                 }
1506                 change_mnt_propagation(p, MS_PRIVATE);
1507         }
1508 }
1509 
1510 static void shrink_submounts(struct mount *mnt);
1511 
1512 static int do_umount(struct mount *mnt, int flags)
1513 {
1514         struct super_block *sb = mnt->mnt.mnt_sb;
1515         int retval;
1516 
1517         retval = security_sb_umount(&mnt->mnt, flags);
1518         if (retval)
1519                 return retval;
1520 
1521         /*
1522          * Allow userspace to request a mountpoint be expired rather than
1523          * unmounting unconditionally. Unmount only happens if:
1524          *  (1) the mark is already set (the mark is cleared by mntput())
1525          *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1526          */
1527         if (flags & MNT_EXPIRE) {
1528                 if (&mnt->mnt == current->fs->root.mnt ||
1529                     flags & (MNT_FORCE | MNT_DETACH))
1530                         return -EINVAL;
1531 
1532                 /*
1533                  * probably don't strictly need the lock here if we examined
1534                  * all race cases, but it's a slowpath.
1535                  */
1536                 lock_mount_hash();
1537                 if (mnt_get_count(mnt) != 2) {
1538                         unlock_mount_hash();
1539                         return -EBUSY;
1540                 }
1541                 unlock_mount_hash();
1542 
1543                 if (!xchg(&mnt->mnt_expiry_mark, 1))
1544                         return -EAGAIN;
1545         }
1546 
1547         /*
1548          * If we may have to abort operations to get out of this
1549          * mount, and they will themselves hold resources we must
1550          * allow the fs to do things. In the Unix tradition of
1551          * 'Gee thats tricky lets do it in userspace' the umount_begin
1552          * might fail to complete on the first run through as other tasks
1553          * must return, and the like. Thats for the mount program to worry
1554          * about for the moment.
1555          */
1556 
1557         if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1558                 sb->s_op->umount_begin(sb);
1559         }
1560 
1561         /*
1562          * No sense to grab the lock for this test, but test itself looks
1563          * somewhat bogus. Suggestions for better replacement?
1564          * Ho-hum... In principle, we might treat that as umount + switch
1565          * to rootfs. GC would eventually take care of the old vfsmount.
1566          * Actually it makes sense, especially if rootfs would contain a
1567          * /reboot - static binary that would close all descriptors and
1568          * call reboot(9). Then init(8) could umount root and exec /reboot.
1569          */
1570         if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1571                 /*
1572                  * Special case for "unmounting" root ...
1573                  * we just try to remount it readonly.
1574                  */
1575                 if (!capable(CAP_SYS_ADMIN))
1576                         return -EPERM;
1577                 down_write(&sb->s_umount);
1578                 if (!(sb->s_flags & MS_RDONLY))
1579                         retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1580                 up_write(&sb->s_umount);
1581                 return retval;
1582         }
1583 
1584         namespace_lock();
1585         lock_mount_hash();
1586         event++;
1587 
1588         if (flags & MNT_DETACH) {
1589                 if (!list_empty(&mnt->mnt_list))
1590                         umount_tree(mnt, UMOUNT_PROPAGATE);
1591                 retval = 0;
1592         } else {
1593                 shrink_submounts(mnt);
1594                 retval = -EBUSY;
1595                 if (!propagate_mount_busy(mnt, 2)) {
1596                         if (!list_empty(&mnt->mnt_list))
1597                                 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1598                         retval = 0;
1599                 }
1600         }
1601         unlock_mount_hash();
1602         namespace_unlock();
1603         return retval;
1604 }
1605 
1606 /*
1607  * __detach_mounts - lazily unmount all mounts on the specified dentry
1608  *
1609  * During unlink, rmdir, and d_drop it is possible to loose the path
1610  * to an existing mountpoint, and wind up leaking the mount.
1611  * detach_mounts allows lazily unmounting those mounts instead of
1612  * leaking them.
1613  *
1614  * The caller may hold dentry->d_inode->i_mutex.
1615  */
1616 void __detach_mounts(struct dentry *dentry)
1617 {
1618         struct mountpoint *mp;
1619         struct mount *mnt;
1620 
1621         namespace_lock();
1622         lock_mount_hash();
1623         mp = lookup_mountpoint(dentry);
1624         if (IS_ERR_OR_NULL(mp))
1625                 goto out_unlock;
1626 
1627         event++;
1628         while (!hlist_empty(&mp->m_list)) {
1629                 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1630                 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1631                         hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1632                         umount_mnt(mnt);
1633                 }
1634                 else umount_tree(mnt, UMOUNT_CONNECTED);
1635         }
1636         put_mountpoint(mp);
1637 out_unlock:
1638         unlock_mount_hash();
1639         namespace_unlock();
1640 }
1641 
1642 /* 
1643  * Is the caller allowed to modify his namespace?
1644  */
1645 static inline bool may_mount(void)
1646 {
1647         return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1648 }
1649 
1650 static inline bool may_mandlock(void)
1651 {
1652 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1653         return false;
1654 #endif
1655         return capable(CAP_SYS_ADMIN);
1656 }
1657 
1658 /*
1659  * Now umount can handle mount points as well as block devices.
1660  * This is important for filesystems which use unnamed block devices.
1661  *
1662  * We now support a flag for forced unmount like the other 'big iron'
1663  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1664  */
1665 
1666 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1667 {
1668         struct path path;
1669         struct mount *mnt;
1670         int retval;
1671         int lookup_flags = 0;
1672 
1673         if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1674                 return -EINVAL;
1675 
1676         if (!may_mount())
1677                 return -EPERM;
1678 
1679         if (!(flags & UMOUNT_NOFOLLOW))
1680                 lookup_flags |= LOOKUP_FOLLOW;
1681 
1682         retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1683         if (retval)
1684                 goto out;
1685         mnt = real_mount(path.mnt);
1686         retval = -EINVAL;
1687         if (path.dentry != path.mnt->mnt_root)
1688                 goto dput_and_out;
1689         if (!check_mnt(mnt))
1690                 goto dput_and_out;
1691         if (mnt->mnt.mnt_flags & MNT_LOCKED)
1692                 goto dput_and_out;
1693         retval = -EPERM;
1694         if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1695                 goto dput_and_out;
1696 
1697         retval = do_umount(mnt, flags);
1698 dput_and_out:
1699         /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1700         dput(path.dentry);
1701         mntput_no_expire(mnt);
1702 out:
1703         return retval;
1704 }
1705 
1706 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1707 
1708 /*
1709  *      The 2.0 compatible umount. No flags.
1710  */
1711 SYSCALL_DEFINE1(oldumount, char __user *, name)
1712 {
1713         return sys_umount(name, 0);
1714 }
1715 
1716 #endif
1717 
1718 static bool is_mnt_ns_file(struct dentry *dentry)
1719 {
1720         /* Is this a proxy for a mount namespace? */
1721         return dentry->d_op == &ns_dentry_operations &&
1722                dentry->d_fsdata == &mntns_operations;
1723 }
1724 
1725 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1726 {
1727         return container_of(ns, struct mnt_namespace, ns);
1728 }
1729 
1730 static bool mnt_ns_loop(struct dentry *dentry)
1731 {
1732         /* Could bind mounting the mount namespace inode cause a
1733          * mount namespace loop?
1734          */
1735         struct mnt_namespace *mnt_ns;
1736         if (!is_mnt_ns_file(dentry))
1737                 return false;
1738 
1739         mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1740         return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1741 }
1742 
1743 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1744                                         int flag)
1745 {
1746         struct mount *res, *p, *q, *r, *parent;
1747 
1748         if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1749                 return ERR_PTR(-EINVAL);
1750 
1751         if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1752                 return ERR_PTR(-EINVAL);
1753 
1754         res = q = clone_mnt(mnt, dentry, flag);
1755         if (IS_ERR(q))
1756                 return q;
1757 
1758         q->mnt_mountpoint = mnt->mnt_mountpoint;
1759 
1760         p = mnt;
1761         list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1762                 struct mount *s;
1763                 if (!is_subdir(r->mnt_mountpoint, dentry))
1764                         continue;
1765 
1766                 for (s = r; s; s = next_mnt(s, r)) {
1767                         struct mount *t = NULL;
1768                         if (!(flag & CL_COPY_UNBINDABLE) &&
1769                             IS_MNT_UNBINDABLE(s)) {
1770                                 s = skip_mnt_tree(s);
1771                                 continue;
1772                         }
1773                         if (!(flag & CL_COPY_MNT_NS_FILE) &&
1774                             is_mnt_ns_file(s->mnt.mnt_root)) {
1775                                 s = skip_mnt_tree(s);
1776                                 continue;
1777                         }
1778                         while (p != s->mnt_parent) {
1779                                 p = p->mnt_parent;
1780                                 q = q->mnt_parent;
1781                         }
1782                         p = s;
1783                         parent = q;
1784                         q = clone_mnt(p, p->mnt.mnt_root, flag);
1785                         if (IS_ERR(q))
1786                                 goto out;
1787                         lock_mount_hash();
1788                         list_add_tail(&q->mnt_list, &res->mnt_list);
1789                         mnt_set_mountpoint(parent, p->mnt_mp, q);
1790                         if (!list_empty(&parent->mnt_mounts)) {
1791                                 t = list_last_entry(&parent->mnt_mounts,
1792                                         struct mount, mnt_child);
1793                                 if (t->mnt_mp != p->mnt_mp)
1794                                         t = NULL;
1795                         }
1796                         attach_shadowed(q, parent, t);
1797                         unlock_mount_hash();
1798                 }
1799         }
1800         return res;
1801 out:
1802         if (res) {
1803                 lock_mount_hash();
1804                 umount_tree(res, UMOUNT_SYNC);
1805                 unlock_mount_hash();
1806         }
1807         return q;
1808 }
1809 
1810 /* Caller should check returned pointer for errors */
1811 
1812 struct vfsmount *collect_mounts(const struct path *path)
1813 {
1814         struct mount *tree;
1815         namespace_lock();
1816         if (!check_mnt(real_mount(path->mnt)))
1817                 tree = ERR_PTR(-EINVAL);
1818         else
1819                 tree = copy_tree(real_mount(path->mnt), path->dentry,
1820                                  CL_COPY_ALL | CL_PRIVATE);
1821         namespace_unlock();
1822         if (IS_ERR(tree))
1823                 return ERR_CAST(tree);
1824         return &tree->mnt;
1825 }
1826 
1827 void drop_collected_mounts(struct vfsmount *mnt)
1828 {
1829         namespace_lock();
1830         lock_mount_hash();
1831         umount_tree(real_mount(mnt), UMOUNT_SYNC);
1832         unlock_mount_hash();
1833         namespace_unlock();
1834 }
1835 
1836 /**
1837  * clone_private_mount - create a private clone of a path
1838  *
1839  * This creates a new vfsmount, which will be the clone of @path.  The new will
1840  * not be attached anywhere in the namespace and will be private (i.e. changes
1841  * to the originating mount won't be propagated into this).
1842  *
1843  * Release with mntput().
1844  */
1845 struct vfsmount *clone_private_mount(const struct path *path)
1846 {
1847         struct mount *old_mnt = real_mount(path->mnt);
1848         struct mount *new_mnt;
1849 
1850         if (IS_MNT_UNBINDABLE(old_mnt))
1851                 return ERR_PTR(-EINVAL);
1852 
1853         new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1854         if (IS_ERR(new_mnt))
1855                 return ERR_CAST(new_mnt);
1856 
1857         return &new_mnt->mnt;
1858 }
1859 EXPORT_SYMBOL_GPL(clone_private_mount);
1860 
1861 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1862                    struct vfsmount *root)
1863 {
1864         struct mount *mnt;
1865         int res = f(root, arg);
1866         if (res)
1867                 return res;
1868         list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1869                 res = f(&mnt->mnt, arg);
1870                 if (res)
1871                         return res;
1872         }
1873         return 0;
1874 }
1875 
1876 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1877 {
1878         struct mount *p;
1879 
1880         for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1881                 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1882                         mnt_release_group_id(p);
1883         }
1884 }
1885 
1886 static int invent_group_ids(struct mount *mnt, bool recurse)
1887 {
1888         struct mount *p;
1889 
1890         for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1891                 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1892                         int err = mnt_alloc_group_id(p);
1893                         if (err) {
1894                                 cleanup_group_ids(mnt, p);
1895                                 return err;
1896                         }
1897                 }
1898         }
1899 
1900         return 0;
1901 }
1902 
1903 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1904 {
1905         unsigned int max = READ_ONCE(sysctl_mount_max);
1906         unsigned int mounts = 0, old, pending, sum;
1907         struct mount *p;
1908 
1909         for (p = mnt; p; p = next_mnt(p, mnt))
1910                 mounts++;
1911 
1912         old = ns->mounts;
1913         pending = ns->pending_mounts;
1914         sum = old + pending;
1915         if ((old > sum) ||
1916             (pending > sum) ||
1917             (max < sum) ||
1918             (mounts > (max - sum)))
1919                 return -ENOSPC;
1920 
1921         ns->pending_mounts = pending + mounts;
1922         return 0;
1923 }
1924 
1925 /*
1926  *  @source_mnt : mount tree to be attached
1927  *  @nd         : place the mount tree @source_mnt is attached
1928  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1929  *                 store the parent mount and mountpoint dentry.
1930  *                 (done when source_mnt is moved)
1931  *
1932  *  NOTE: in the table below explains the semantics when a source mount
1933  *  of a given type is attached to a destination mount of a given type.
1934  * ---------------------------------------------------------------------------
1935  * |         BIND MOUNT OPERATION                                            |
1936  * |**************************************************************************
1937  * | source-->| shared        |       private  |       slave    | unbindable |
1938  * | dest     |               |                |                |            |
1939  * |   |      |               |                |                |            |
1940  * |   v      |               |                |                |            |
1941  * |**************************************************************************
1942  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1943  * |          |               |                |                |            |
1944  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1945  * ***************************************************************************
1946  * A bind operation clones the source mount and mounts the clone on the
1947  * destination mount.
1948  *
1949  * (++)  the cloned mount is propagated to all the mounts in the propagation
1950  *       tree of the destination mount and the cloned mount is added to
1951  *       the peer group of the source mount.
1952  * (+)   the cloned mount is created under the destination mount and is marked
1953  *       as shared. The cloned mount is added to the peer group of the source
1954  *       mount.
1955  * (+++) the mount is propagated to all the mounts in the propagation tree
1956  *       of the destination mount and the cloned mount is made slave
1957  *       of the same master as that of the source mount. The cloned mount
1958  *       is marked as 'shared and slave'.
1959  * (*)   the cloned mount is made a slave of the same master as that of the
1960  *       source mount.
1961  *
1962  * ---------------------------------------------------------------------------
1963  * |                    MOVE MOUNT OPERATION                                 |
1964  * |**************************************************************************
1965  * | source-->| shared        |       private  |       slave    | unbindable |
1966  * | dest     |               |                |                |            |
1967  * |   |      |               |                |                |            |
1968  * |   v      |               |                |                |            |
1969  * |**************************************************************************
1970  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1971  * |          |               |                |                |            |
1972  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1973  * ***************************************************************************
1974  *
1975  * (+)  the mount is moved to the destination. And is then propagated to
1976  *      all the mounts in the propagation tree of the destination mount.
1977  * (+*)  the mount is moved to the destination.
1978  * (+++)  the mount is moved to the destination and is then propagated to
1979  *      all the mounts belonging to the destination mount's propagation tree.
1980  *      the mount is marked as 'shared and slave'.
1981  * (*)  the mount continues to be a slave at the new location.
1982  *
1983  * if the source mount is a tree, the operations explained above is
1984  * applied to each mount in the tree.
1985  * Must be called without spinlocks held, since this function can sleep
1986  * in allocations.
1987  */
1988 static int attach_recursive_mnt(struct mount *source_mnt,
1989                         struct mount *dest_mnt,
1990                         struct mountpoint *dest_mp,
1991                         struct path *parent_path)
1992 {
1993         HLIST_HEAD(tree_list);
1994         struct mnt_namespace *ns = dest_mnt->mnt_ns;
1995         struct mount *child, *p;
1996         struct hlist_node *n;
1997         int err;
1998 
1999         /* Is there space to add these mounts to the mount namespace? */
2000         if (!parent_path) {
2001                 err = count_mounts(ns, source_mnt);
2002                 if (err)
2003                         goto out;
2004         }
2005 
2006         if (IS_MNT_SHARED(dest_mnt)) {
2007                 err = invent_group_ids(source_mnt, true);
2008                 if (err)
2009                         goto out;
2010                 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2011                 lock_mount_hash();
2012                 if (err)
2013                         goto out_cleanup_ids;
2014                 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2015                         set_mnt_shared(p);
2016         } else {
2017                 lock_mount_hash();
2018         }
2019         if (parent_path) {
2020                 detach_mnt(source_mnt, parent_path);
2021                 attach_mnt(source_mnt, dest_mnt, dest_mp);
2022                 touch_mnt_namespace(source_mnt->mnt_ns);
2023         } else {
2024                 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2025                 commit_tree(source_mnt, NULL);
2026         }
2027 
2028         hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2029                 struct mount *q;
2030                 hlist_del_init(&child->mnt_hash);
2031                 q = __lookup_mnt_last(&child->mnt_parent->mnt,
2032                                       child->mnt_mountpoint);
2033                 commit_tree(child, q);
2034         }
2035         unlock_mount_hash();
2036 
2037         return 0;
2038 
2039  out_cleanup_ids:
2040         while (!hlist_empty(&tree_list)) {
2041                 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2042                 child->mnt_parent->mnt_ns->pending_mounts = 0;
2043                 umount_tree(child, UMOUNT_SYNC);
2044         }
2045         unlock_mount_hash();
2046         cleanup_group_ids(source_mnt, NULL);
2047  out:
2048         ns->pending_mounts = 0;
2049         return err;
2050 }
2051 
2052 static struct mountpoint *lock_mount(struct path *path)
2053 {
2054         struct vfsmount *mnt;
2055         struct dentry *dentry = path->dentry;
2056 retry:
2057         inode_lock(dentry->d_inode);
2058         if (unlikely(cant_mount(dentry))) {
2059                 inode_unlock(dentry->d_inode);
2060                 return ERR_PTR(-ENOENT);
2061         }
2062         namespace_lock();
2063         mnt = lookup_mnt(path);
2064         if (likely(!mnt)) {
2065                 struct mountpoint *mp = get_mountpoint(dentry);
2066                 if (IS_ERR(mp)) {
2067                         namespace_unlock();
2068                         inode_unlock(dentry->d_inode);
2069                         return mp;
2070                 }
2071                 return mp;
2072         }
2073         namespace_unlock();
2074         inode_unlock(path->dentry->d_inode);
2075         path_put(path);
2076         path->mnt = mnt;
2077         dentry = path->dentry = dget(mnt->mnt_root);
2078         goto retry;
2079 }
2080 
2081 static void unlock_mount(struct mountpoint *where)
2082 {
2083         struct dentry *dentry = where->m_dentry;
2084 
2085         read_seqlock_excl(&mount_lock);
2086         put_mountpoint(where);
2087         read_sequnlock_excl(&mount_lock);
2088 
2089         namespace_unlock();
2090         inode_unlock(dentry->d_inode);
2091 }
2092 
2093 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2094 {
2095         if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2096                 return -EINVAL;
2097 
2098         if (d_is_dir(mp->m_dentry) !=
2099               d_is_dir(mnt->mnt.mnt_root))
2100                 return -ENOTDIR;
2101 
2102         return attach_recursive_mnt(mnt, p, mp, NULL);
2103 }
2104 
2105 /*
2106  * Sanity check the flags to change_mnt_propagation.
2107  */
2108 
2109 static int flags_to_propagation_type(int flags)
2110 {
2111         int type = flags & ~(MS_REC | MS_SILENT);
2112 
2113         /* Fail if any non-propagation flags are set */
2114         if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2115                 return 0;
2116         /* Only one propagation flag should be set */
2117         if (!is_power_of_2(type))
2118                 return 0;
2119         return type;
2120 }
2121 
2122 /*
2123  * recursively change the type of the mountpoint.
2124  */
2125 static int do_change_type(struct path *path, int flag)
2126 {
2127         struct mount *m;
2128         struct mount *mnt = real_mount(path->mnt);
2129         int recurse = flag & MS_REC;
2130         int type;
2131         int err = 0;
2132 
2133         if (path->dentry != path->mnt->mnt_root)
2134                 return -EINVAL;
2135 
2136         type = flags_to_propagation_type(flag);
2137         if (!type)
2138                 return -EINVAL;
2139 
2140         namespace_lock();
2141         if (type == MS_SHARED) {
2142                 err = invent_group_ids(mnt, recurse);
2143                 if (err)
2144                         goto out_unlock;
2145         }
2146 
2147         lock_mount_hash();
2148         for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2149                 change_mnt_propagation(m, type);
2150         unlock_mount_hash();
2151 
2152  out_unlock:
2153         namespace_unlock();
2154         return err;
2155 }
2156 
2157 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2158 {
2159         struct mount *child;
2160         list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2161                 if (!is_subdir(child->mnt_mountpoint, dentry))
2162                         continue;
2163 
2164                 if (child->mnt.mnt_flags & MNT_LOCKED)
2165                         return true;
2166         }
2167         return false;
2168 }
2169 
2170 /*
2171  * do loopback mount.
2172  */
2173 static int do_loopback(struct path *path, const char *old_name,
2174                                 int recurse)
2175 {
2176         struct path old_path;
2177         struct mount *mnt = NULL, *old, *parent;
2178         struct mountpoint *mp;
2179         int err;
2180         if (!old_name || !*old_name)
2181                 return -EINVAL;
2182         err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2183         if (err)
2184                 return err;
2185 
2186         err = -EINVAL;
2187         if (mnt_ns_loop(old_path.dentry))
2188                 goto out; 
2189 
2190         mp = lock_mount(path);
2191         err = PTR_ERR(mp);
2192         if (IS_ERR(mp))
2193                 goto out;
2194 
2195         old = real_mount(old_path.mnt);
2196         parent = real_mount(path->mnt);
2197 
2198         err = -EINVAL;
2199         if (IS_MNT_UNBINDABLE(old))
2200                 goto out2;
2201 
2202         if (!check_mnt(parent))
2203                 goto out2;
2204 
2205         if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2206                 goto out2;
2207 
2208         if (!recurse && has_locked_children(old, old_path.dentry))
2209                 goto out2;
2210 
2211         if (recurse)
2212                 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2213         else
2214                 mnt = clone_mnt(old, old_path.dentry, 0);
2215 
2216         if (IS_ERR(mnt)) {
2217                 err = PTR_ERR(mnt);
2218                 goto out2;
2219         }
2220 
2221         mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2222 
2223         err = graft_tree(mnt, parent, mp);
2224         if (err) {
2225                 lock_mount_hash();
2226                 umount_tree(mnt, UMOUNT_SYNC);
2227                 unlock_mount_hash();
2228         }
2229 out2:
2230         unlock_mount(mp);
2231 out:
2232         path_put(&old_path);
2233         return err;
2234 }
2235 
2236 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2237 {
2238         int error = 0;
2239         int readonly_request = 0;
2240 
2241         if (ms_flags & MS_RDONLY)
2242                 readonly_request = 1;
2243         if (readonly_request == __mnt_is_readonly(mnt))
2244                 return 0;
2245 
2246         if (readonly_request)
2247                 error = mnt_make_readonly(real_mount(mnt));
2248         else
2249                 __mnt_unmake_readonly(real_mount(mnt));
2250         return error;
2251 }
2252 
2253 /*
2254  * change filesystem flags. dir should be a physical root of filesystem.
2255  * If you've mounted a non-root directory somewhere and want to do remount
2256  * on it - tough luck.
2257  */
2258 static int do_remount(struct path *path, int flags, int mnt_flags,
2259                       void *data)
2260 {
2261         int err;
2262         struct super_block *sb = path->mnt->mnt_sb;
2263         struct mount *mnt = real_mount(path->mnt);
2264 
2265         if (!check_mnt(mnt))
2266                 return -EINVAL;
2267 
2268         if (path->dentry != path->mnt->mnt_root)
2269                 return -EINVAL;
2270 
2271         /* Don't allow changing of locked mnt flags.
2272          *
2273          * No locks need to be held here while testing the various
2274          * MNT_LOCK flags because those flags can never be cleared
2275          * once they are set.
2276          */
2277         if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2278             !(mnt_flags & MNT_READONLY)) {
2279                 return -EPERM;
2280         }
2281         if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2282             !(mnt_flags & MNT_NODEV)) {
2283                 return -EPERM;
2284         }
2285         if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2286             !(mnt_flags & MNT_NOSUID)) {
2287                 return -EPERM;
2288         }
2289         if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2290             !(mnt_flags & MNT_NOEXEC)) {
2291                 return -EPERM;
2292         }
2293         if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2294             ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2295                 return -EPERM;
2296         }
2297 
2298         err = security_sb_remount(sb, data);
2299         if (err)
2300                 return err;
2301 
2302         down_write(&sb->s_umount);
2303         if (flags & MS_BIND)
2304                 err = change_mount_flags(path->mnt, flags);
2305         else if (!capable(CAP_SYS_ADMIN))
2306                 err = -EPERM;
2307         else
2308                 err = do_remount_sb(sb, flags, data, 0);
2309         if (!err) {
2310                 lock_mount_hash();
2311                 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2312                 mnt->mnt.mnt_flags = mnt_flags;
2313                 touch_mnt_namespace(mnt->mnt_ns);
2314                 unlock_mount_hash();
2315         }
2316         up_write(&sb->s_umount);
2317         return err;
2318 }
2319 
2320 static inline int tree_contains_unbindable(struct mount *mnt)
2321 {
2322         struct mount *p;
2323         for (p = mnt; p; p = next_mnt(p, mnt)) {
2324                 if (IS_MNT_UNBINDABLE(p))
2325                         return 1;
2326         }
2327         return 0;
2328 }
2329 
2330 static int do_move_mount(struct path *path, const char *old_name)
2331 {
2332         struct path old_path, parent_path;
2333         struct mount *p;
2334         struct mount *old;
2335         struct mountpoint *mp;
2336         int err;
2337         if (!old_name || !*old_name)
2338                 return -EINVAL;
2339         err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2340         if (err)
2341                 return err;
2342 
2343         mp = lock_mount(path);
2344         err = PTR_ERR(mp);
2345         if (IS_ERR(mp))
2346                 goto out;
2347 
2348         old = real_mount(old_path.mnt);
2349         p = real_mount(path->mnt);
2350 
2351         err = -EINVAL;
2352         if (!check_mnt(p) || !check_mnt(old))
2353                 goto out1;
2354 
2355         if (old->mnt.mnt_flags & MNT_LOCKED)
2356                 goto out1;
2357 
2358         err = -EINVAL;
2359         if (old_path.dentry != old_path.mnt->mnt_root)
2360                 goto out1;
2361 
2362         if (!mnt_has_parent(old))
2363                 goto out1;
2364 
2365         if (d_is_dir(path->dentry) !=
2366               d_is_dir(old_path.dentry))
2367                 goto out1;
2368         /*
2369          * Don't move a mount residing in a shared parent.
2370          */
2371         if (IS_MNT_SHARED(old->mnt_parent))
2372                 goto out1;
2373         /*
2374          * Don't move a mount tree containing unbindable mounts to a destination
2375          * mount which is shared.
2376          */
2377         if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2378                 goto out1;
2379         err = -ELOOP;
2380         for (; mnt_has_parent(p); p = p->mnt_parent)
2381                 if (p == old)
2382                         goto out1;
2383 
2384         err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2385         if (err)
2386                 goto out1;
2387 
2388         /* if the mount is moved, it should no longer be expire
2389          * automatically */
2390         list_del_init(&old->mnt_expire);
2391 out1:
2392         unlock_mount(mp);
2393 out:
2394         if (!err)
2395                 path_put(&parent_path);
2396         path_put(&old_path);
2397         return err;
2398 }
2399 
2400 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2401 {
2402         int err;
2403         const char *subtype = strchr(fstype, '.');
2404         if (subtype) {
2405                 subtype++;
2406                 err = -EINVAL;
2407                 if (!subtype[0])
2408                         goto err;
2409         } else
2410                 subtype = "";
2411 
2412         mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2413         err = -ENOMEM;
2414         if (!mnt->mnt_sb->s_subtype)
2415                 goto err;
2416         return mnt;
2417 
2418  err:
2419         mntput(mnt);
2420         return ERR_PTR(err);
2421 }
2422 
2423 /*
2424  * add a mount into a namespace's mount tree
2425  */
2426 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2427 {
2428         struct mountpoint *mp;
2429         struct mount *parent;
2430         int err;
2431 
2432         mnt_flags &= ~MNT_INTERNAL_FLAGS;
2433 
2434         mp = lock_mount(path);
2435         if (IS_ERR(mp))
2436                 return PTR_ERR(mp);
2437 
2438         parent = real_mount(path->mnt);
2439         err = -EINVAL;
2440         if (unlikely(!check_mnt(parent))) {
2441                 /* that's acceptable only for automounts done in private ns */
2442                 if (!(mnt_flags & MNT_SHRINKABLE))
2443                         goto unlock;
2444                 /* ... and for those we'd better have mountpoint still alive */
2445                 if (!parent->mnt_ns)
2446                         goto unlock;
2447         }
2448 
2449         /* Refuse the same filesystem on the same mount point */
2450         err = -EBUSY;
2451         if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2452             path->mnt->mnt_root == path->dentry)
2453                 goto unlock;
2454 
2455         err = -EINVAL;
2456         if (d_is_symlink(newmnt->mnt.mnt_root))
2457                 goto unlock;
2458 
2459         newmnt->mnt.mnt_flags = mnt_flags;
2460         err = graft_tree(newmnt, parent, mp);
2461 
2462 unlock:
2463         unlock_mount(mp);
2464         return err;
2465 }
2466 
2467 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2468 
2469 /*
2470  * create a new mount for userspace and request it to be added into the
2471  * namespace's tree
2472  */
2473 static int do_new_mount(struct path *path, const char *fstype, int flags,
2474                         int mnt_flags, const char *name, void *data)
2475 {
2476         struct file_system_type *type;
2477         struct vfsmount *mnt;
2478         int err;
2479 
2480         if (!fstype)
2481                 return -EINVAL;
2482 
2483         type = get_fs_type(fstype);
2484         if (!type)
2485                 return -ENODEV;
2486 
2487         mnt = vfs_kern_mount(type, flags, name, data);
2488         if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2489             !mnt->mnt_sb->s_subtype)
2490                 mnt = fs_set_subtype(mnt, fstype);
2491 
2492         put_filesystem(type);
2493         if (IS_ERR(mnt))
2494                 return PTR_ERR(mnt);
2495 
2496         if (mount_too_revealing(mnt, &mnt_flags)) {
2497                 mntput(mnt);
2498                 return -EPERM;
2499         }
2500 
2501         err = do_add_mount(real_mount(mnt), path, mnt_flags);
2502         if (err)
2503                 mntput(mnt);
2504         return err;
2505 }
2506 
2507 int finish_automount(struct vfsmount *m, struct path *path)
2508 {
2509         struct mount *mnt = real_mount(m);
2510         int err;
2511         /* The new mount record should have at least 2 refs to prevent it being
2512          * expired before we get a chance to add it
2513          */
2514         BUG_ON(mnt_get_count(mnt) < 2);
2515 
2516         if (m->mnt_sb == path->mnt->mnt_sb &&
2517             m->mnt_root == path->dentry) {
2518                 err = -ELOOP;
2519                 goto fail;
2520         }
2521 
2522         err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2523         if (!err)
2524                 return 0;
2525 fail:
2526         /* remove m from any expiration list it may be on */
2527         if (!list_empty(&mnt->mnt_expire)) {
2528                 namespace_lock();
2529                 list_del_init(&mnt->mnt_expire);
2530                 namespace_unlock();
2531         }
2532         mntput(m);
2533         mntput(m);
2534         return err;
2535 }
2536 
2537 /**
2538  * mnt_set_expiry - Put a mount on an expiration list
2539  * @mnt: The mount to list.
2540  * @expiry_list: The list to add the mount to.
2541  */
2542 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2543 {
2544         namespace_lock();
2545 
2546         list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2547 
2548         namespace_unlock();
2549 }
2550 EXPORT_SYMBOL(mnt_set_expiry);
2551 
2552 /*
2553  * process a list of expirable mountpoints with the intent of discarding any
2554  * mountpoints that aren't in use and haven't been touched since last we came
2555  * here
2556  */
2557 void mark_mounts_for_expiry(struct list_head *mounts)
2558 {
2559         struct mount *mnt, *next;
2560         LIST_HEAD(graveyard);
2561 
2562         if (list_empty(mounts))
2563                 return;
2564 
2565         namespace_lock();
2566         lock_mount_hash();
2567 
2568         /* extract from the expiration list every vfsmount that matches the
2569          * following criteria:
2570          * - only referenced by its parent vfsmount
2571          * - still marked for expiry (marked on the last call here; marks are
2572          *   cleared by mntput())
2573          */
2574         list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2575                 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2576                         propagate_mount_busy(mnt, 1))
2577                         continue;
2578                 list_move(&mnt->mnt_expire, &graveyard);
2579         }
2580         while (!list_empty(&graveyard)) {
2581                 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2582                 touch_mnt_namespace(mnt->mnt_ns);
2583                 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2584         }
2585         unlock_mount_hash();
2586         namespace_unlock();
2587 }
2588 
2589 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2590 
2591 /*
2592  * Ripoff of 'select_parent()'
2593  *
2594  * search the list of submounts for a given mountpoint, and move any
2595  * shrinkable submounts to the 'graveyard' list.
2596  */
2597 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2598 {
2599         struct mount *this_parent = parent;
2600         struct list_head *next;
2601         int found = 0;
2602 
2603 repeat:
2604         next = this_parent->mnt_mounts.next;
2605 resume:
2606         while (next != &this_parent->mnt_mounts) {
2607                 struct list_head *tmp = next;
2608                 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2609 
2610                 next = tmp->next;
2611                 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2612                         continue;
2613                 /*
2614                  * Descend a level if the d_mounts list is non-empty.
2615                  */
2616                 if (!list_empty(&mnt->mnt_mounts)) {
2617                         this_parent = mnt;
2618                         goto repeat;
2619                 }
2620 
2621                 if (!propagate_mount_busy(mnt, 1)) {
2622                         list_move_tail(&mnt->mnt_expire, graveyard);
2623                         found++;
2624                 }
2625         }
2626         /*
2627          * All done at this level ... ascend and resume the search
2628          */
2629         if (this_parent != parent) {
2630                 next = this_parent->mnt_child.next;
2631                 this_parent = this_parent->mnt_parent;
2632                 goto resume;
2633         }
2634         return found;
2635 }
2636 
2637 /*
2638  * process a list of expirable mountpoints with the intent of discarding any
2639  * submounts of a specific parent mountpoint
2640  *
2641  * mount_lock must be held for write
2642  */
2643 static void shrink_submounts(struct mount *mnt)
2644 {
2645         LIST_HEAD(graveyard);
2646         struct mount *m;
2647 
2648         /* extract submounts of 'mountpoint' from the expiration list */
2649         while (select_submounts(mnt, &graveyard)) {
2650                 while (!list_empty(&graveyard)) {
2651                         m = list_first_entry(&graveyard, struct mount,
2652                                                 mnt_expire);
2653                         touch_mnt_namespace(m->mnt_ns);
2654                         umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2655                 }
2656         }
2657 }
2658 
2659 /*
2660  * Some copy_from_user() implementations do not return the exact number of
2661  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2662  * Note that this function differs from copy_from_user() in that it will oops
2663  * on bad values of `to', rather than returning a short copy.
2664  */
2665 static long exact_copy_from_user(void *to, const void __user * from,
2666                                  unsigned long n)
2667 {
2668         char *t = to;
2669         const char __user *f = from;
2670         char c;
2671 
2672         if (!access_ok(VERIFY_READ, from, n))
2673                 return n;
2674 
2675         while (n) {
2676                 if (__get_user(c, f)) {
2677                         memset(t, 0, n);
2678                         break;
2679                 }
2680                 *t++ = c;
2681                 f++;
2682                 n--;
2683         }
2684         return n;
2685 }
2686 
2687 void *copy_mount_options(const void __user * data)
2688 {
2689         int i;
2690         unsigned long size;
2691         char *copy;
2692 
2693         if (!data)
2694                 return NULL;
2695 
2696         copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2697         if (!copy)
2698                 return ERR_PTR(-ENOMEM);
2699 
2700         /* We only care that *some* data at the address the user
2701          * gave us is valid.  Just in case, we'll zero
2702          * the remainder of the page.
2703          */
2704         /* copy_from_user cannot cross TASK_SIZE ! */
2705         size = TASK_SIZE - (unsigned long)data;
2706         if (size > PAGE_SIZE)
2707                 size = PAGE_SIZE;
2708 
2709         i = size - exact_copy_from_user(copy, data, size);
2710         if (!i) {
2711                 kfree(copy);
2712                 return ERR_PTR(-EFAULT);
2713         }
2714         if (i != PAGE_SIZE)
2715                 memset(copy + i, 0, PAGE_SIZE - i);
2716         return copy;
2717 }
2718 
2719 char *copy_mount_string(const void __user *data)
2720 {
2721         return data ? strndup_user(data, PAGE_SIZE) : NULL;
2722 }
2723 
2724 /*
2725  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2726  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2727  *
2728  * data is a (void *) that can point to any structure up to
2729  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2730  * information (or be NULL).
2731  *
2732  * Pre-0.97 versions of mount() didn't have a flags word.
2733  * When the flags word was introduced its top half was required
2734  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2735  * Therefore, if this magic number is present, it carries no information
2736  * and must be discarded.
2737  */
2738 long do_mount(const char *dev_name, const char __user *dir_name,
2739                 const char *type_page, unsigned long flags, void *data_page)
2740 {
2741         struct path path;
2742         int retval = 0;
2743         int mnt_flags = 0;
2744 
2745         /* Discard magic */
2746         if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2747                 flags &= ~MS_MGC_MSK;
2748 
2749         /* Basic sanity checks */
2750         if (data_page)
2751                 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2752 
2753         /* ... and get the mountpoint */
2754         retval = user_path(dir_name, &path);
2755         if (retval)
2756                 return retval;
2757 
2758         retval = security_sb_mount(dev_name, &path,
2759                                    type_page, flags, data_page);
2760         if (!retval && !may_mount())
2761                 retval = -EPERM;
2762         if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2763                 retval = -EPERM;
2764         if (retval)
2765                 goto dput_out;
2766 
2767         /* Default to relatime unless overriden */
2768         if (!(flags & MS_NOATIME))
2769                 mnt_flags |= MNT_RELATIME;
2770 
2771         /* Separate the per-mountpoint flags */
2772         if (flags & MS_NOSUID)
2773                 mnt_flags |= MNT_NOSUID;
2774         if (flags & MS_NODEV)
2775                 mnt_flags |= MNT_NODEV;
2776         if (flags & MS_NOEXEC)
2777                 mnt_flags |= MNT_NOEXEC;
2778         if (flags & MS_NOATIME)
2779                 mnt_flags |= MNT_NOATIME;
2780         if (flags & MS_NODIRATIME)
2781                 mnt_flags |= MNT_NODIRATIME;
2782         if (flags & MS_STRICTATIME)
2783                 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2784         if (flags & MS_RDONLY)
2785                 mnt_flags |= MNT_READONLY;
2786 
2787         /* The default atime for remount is preservation */
2788         if ((flags & MS_REMOUNT) &&
2789             ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2790                        MS_STRICTATIME)) == 0)) {
2791                 mnt_flags &= ~MNT_ATIME_MASK;
2792                 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2793         }
2794 
2795         flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2796                    MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2797                    MS_STRICTATIME | MS_NOREMOTELOCK);
2798 
2799         if (flags & MS_REMOUNT)
2800                 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2801                                     data_page);
2802         else if (flags & MS_BIND)
2803                 retval = do_loopback(&path, dev_name, flags & MS_REC);
2804         else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2805                 retval = do_change_type(&path, flags);
2806         else if (flags & MS_MOVE)
2807                 retval = do_move_mount(&path, dev_name);
2808         else
2809                 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2810                                       dev_name, data_page);
2811 dput_out:
2812         path_put(&path);
2813         return retval;
2814 }
2815 
2816 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2817 {
2818         return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2819 }
2820 
2821 static void dec_mnt_namespaces(struct ucounts *ucounts)
2822 {
2823         dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2824 }
2825 
2826 static void free_mnt_ns(struct mnt_namespace *ns)
2827 {
2828         ns_free_inum(&ns->ns);
2829         dec_mnt_namespaces(ns->ucounts);
2830         put_user_ns(ns->user_ns);
2831         kfree(ns);
2832 }
2833 
2834 /*
2835  * Assign a sequence number so we can detect when we attempt to bind
2836  * mount a reference to an older mount namespace into the current
2837  * mount namespace, preventing reference counting loops.  A 64bit
2838  * number incrementing at 10Ghz will take 12,427 years to wrap which
2839  * is effectively never, so we can ignore the possibility.
2840  */
2841 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2842 
2843 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2844 {
2845         struct mnt_namespace *new_ns;
2846         struct ucounts *ucounts;
2847         int ret;
2848 
2849         ucounts = inc_mnt_namespaces(user_ns);
2850         if (!ucounts)
2851                 return ERR_PTR(-ENOSPC);
2852 
2853         new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2854         if (!new_ns) {
2855                 dec_mnt_namespaces(ucounts);
2856                 return ERR_PTR(-ENOMEM);
2857         }
2858         ret = ns_alloc_inum(&new_ns->ns);
2859         if (ret) {
2860                 kfree(new_ns);
2861                 dec_mnt_namespaces(ucounts);
2862                 return ERR_PTR(ret);
2863         }
2864         new_ns->ns.ops = &mntns_operations;
2865         new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2866         atomic_set(&new_ns->count, 1);
2867         new_ns->root = NULL;
2868         INIT_LIST_HEAD(&new_ns->list);
2869         init_waitqueue_head(&new_ns->poll);
2870         new_ns->event = 0;
2871         new_ns->user_ns = get_user_ns(user_ns);
2872         new_ns->ucounts = ucounts;
2873         new_ns->mounts = 0;
2874         new_ns->pending_mounts = 0;
2875         return new_ns;
2876 }
2877 
2878 __latent_entropy
2879 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2880                 struct user_namespace *user_ns, struct fs_struct *new_fs)
2881 {
2882         struct mnt_namespace *new_ns;
2883         struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2884         struct mount *p, *q;
2885         struct mount *old;
2886         struct mount *new;
2887         int copy_flags;
2888 
2889         BUG_ON(!ns);
2890 
2891         if (likely(!(flags & CLONE_NEWNS))) {
2892                 get_mnt_ns(ns);
2893                 return ns;
2894         }
2895 
2896         old = ns->root;
2897 
2898         new_ns = alloc_mnt_ns(user_ns);
2899         if (IS_ERR(new_ns))
2900                 return new_ns;
2901 
2902         namespace_lock();
2903         /* First pass: copy the tree topology */
2904         copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2905         if (user_ns != ns->user_ns)
2906                 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2907         new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2908         if (IS_ERR(new)) {
2909                 namespace_unlock();
2910                 free_mnt_ns(new_ns);
2911                 return ERR_CAST(new);
2912         }
2913         new_ns->root = new;
2914         list_add_tail(&new_ns->list, &new->mnt_list);
2915 
2916         /*
2917          * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2918          * as belonging to new namespace.  We have already acquired a private
2919          * fs_struct, so tsk->fs->lock is not needed.
2920          */
2921         p = old;
2922         q = new;
2923         while (p) {
2924                 q->mnt_ns = new_ns;
2925                 new_ns->mounts++;
2926                 if (new_fs) {
2927                         if (&p->mnt == new_fs->root.mnt) {
2928                                 new_fs->root.mnt = mntget(&q->mnt);
2929                                 rootmnt = &p->mnt;
2930                         }
2931                         if (&p->mnt == new_fs->pwd.mnt) {
2932                                 new_fs->pwd.mnt = mntget(&q->mnt);
2933                                 pwdmnt = &p->mnt;
2934                         }
2935                 }
2936                 p = next_mnt(p, old);
2937                 q = next_mnt(q, new);
2938                 if (!q)
2939                         break;
2940                 while (p->mnt.mnt_root != q->mnt.mnt_root)
2941                         p = next_mnt(p, old);
2942         }
2943         namespace_unlock();
2944 
2945         if (rootmnt)
2946                 mntput(rootmnt);
2947         if (pwdmnt)
2948                 mntput(pwdmnt);
2949 
2950         return new_ns;
2951 }
2952 
2953 /**
2954  * create_mnt_ns - creates a private namespace and adds a root filesystem
2955  * @mnt: pointer to the new root filesystem mountpoint
2956  */
2957 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2958 {
2959         struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2960         if (!IS_ERR(new_ns)) {
2961                 struct mount *mnt = real_mount(m);
2962                 mnt->mnt_ns = new_ns;
2963                 new_ns->root = mnt;
2964                 new_ns->mounts++;
2965                 list_add(&mnt->mnt_list, &new_ns->list);
2966         } else {
2967                 mntput(m);
2968         }
2969         return new_ns;
2970 }
2971 
2972 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2973 {
2974         struct mnt_namespace *ns;
2975         struct super_block *s;
2976         struct path path;
2977         int err;
2978 
2979         ns = create_mnt_ns(mnt);
2980         if (IS_ERR(ns))
2981                 return ERR_CAST(ns);
2982 
2983         err = vfs_path_lookup(mnt->mnt_root, mnt,
2984                         name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2985 
2986         put_mnt_ns(ns);
2987 
2988         if (err)
2989                 return ERR_PTR(err);
2990 
2991         /* trade a vfsmount reference for active sb one */
2992         s = path.mnt->mnt_sb;
2993         atomic_inc(&s->s_active);
2994         mntput(path.mnt);
2995         /* lock the sucker */
2996         down_write(&s->s_umount);
2997         /* ... and return the root of (sub)tree on it */
2998         return path.dentry;
2999 }
3000 EXPORT_SYMBOL(mount_subtree);
3001 
3002 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3003                 char __user *, type, unsigned long, flags, void __user *, data)
3004 {
3005         int ret;
3006         char *kernel_type;
3007         char *kernel_dev;
3008         void *options;
3009 
3010         kernel_type = copy_mount_string(type);
3011         ret = PTR_ERR(kernel_type);
3012         if (IS_ERR(kernel_type))
3013                 goto out_type;
3014 
3015         kernel_dev = copy_mount_string(dev_name);
3016         ret = PTR_ERR(kernel_dev);
3017         if (IS_ERR(kernel_dev))
3018                 goto out_dev;
3019 
3020         options = copy_mount_options(data);
3021         ret = PTR_ERR(options);
3022         if (IS_ERR(options))
3023                 goto out_data;
3024 
3025         ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3026 
3027         kfree(options);
3028 out_data:
3029         kfree(kernel_dev);
3030 out_dev:
3031         kfree(kernel_type);
3032 out_type:
3033         return ret;
3034 }
3035 
3036 /*
3037  * Return true if path is reachable from root
3038  *
3039  * namespace_sem or mount_lock is held
3040  */
3041 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3042                          const struct path *root)
3043 {
3044         while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3045                 dentry = mnt->mnt_mountpoint;
3046                 mnt = mnt->mnt_parent;
3047         }
3048         return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3049 }
3050 
3051 bool path_is_under(const struct path *path1, const struct path *path2)
3052 {
3053         bool res;
3054         read_seqlock_excl(&mount_lock);
3055         res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3056         read_sequnlock_excl(&mount_lock);
3057         return res;
3058 }
3059 EXPORT_SYMBOL(path_is_under);
3060 
3061 /*
3062  * pivot_root Semantics:
3063  * Moves the root file system of the current process to the directory put_old,
3064  * makes new_root as the new root file system of the current process, and sets
3065  * root/cwd of all processes which had them on the current root to new_root.
3066  *
3067  * Restrictions:
3068  * The new_root and put_old must be directories, and  must not be on the
3069  * same file  system as the current process root. The put_old  must  be
3070  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3071  * pointed to by put_old must yield the same directory as new_root. No other
3072  * file system may be mounted on put_old. After all, new_root is a mountpoint.
3073  *
3074  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3075  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3076  * in this situation.
3077  *
3078  * Notes:
3079  *  - we don't move root/cwd if they are not at the root (reason: if something
3080  *    cared enough to change them, it's probably wrong to force them elsewhere)
3081  *  - it's okay to pick a root that isn't the root of a file system, e.g.
3082  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3083  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3084  *    first.
3085  */
3086 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3087                 const char __user *, put_old)
3088 {
3089         struct path new, old, parent_path, root_parent, root;
3090         struct mount *new_mnt, *root_mnt, *old_mnt;
3091         struct mountpoint *old_mp, *root_mp;
3092         int error;
3093 
3094         if (!may_mount())
3095                 return -EPERM;
3096 
3097         error = user_path_dir(new_root, &new);
3098         if (error)
3099                 goto out0;
3100 
3101         error = user_path_dir(put_old, &old);
3102         if (error)
3103                 goto out1;
3104 
3105         error = security_sb_pivotroot(&old, &new);
3106         if (error)
3107                 goto out2;
3108 
3109         get_fs_root(current->fs, &root);
3110         old_mp = lock_mount(&old);
3111         error = PTR_ERR(old_mp);
3112         if (IS_ERR(old_mp))
3113                 goto out3;
3114 
3115         error = -EINVAL;
3116         new_mnt = real_mount(new.mnt);
3117         root_mnt = real_mount(root.mnt);
3118         old_mnt = real_mount(old.mnt);
3119         if (IS_MNT_SHARED(old_mnt) ||
3120                 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3121                 IS_MNT_SHARED(root_mnt->mnt_parent))
3122                 goto out4;
3123         if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3124                 goto out4;
3125         if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3126                 goto out4;
3127         error = -ENOENT;
3128         if (d_unlinked(new.dentry))
3129                 goto out4;
3130         error = -EBUSY;
3131         if (new_mnt == root_mnt || old_mnt == root_mnt)
3132                 goto out4; /* loop, on the same file system  */
3133         error = -EINVAL;
3134         if (root.mnt->mnt_root != root.dentry)
3135                 goto out4; /* not a mountpoint */
3136         if (!mnt_has_parent(root_mnt))
3137                 goto out4; /* not attached */
3138         root_mp = root_mnt->mnt_mp;
3139         if (new.mnt->mnt_root != new.dentry)
3140                 goto out4; /* not a mountpoint */
3141         if (!mnt_has_parent(new_mnt))
3142                 goto out4; /* not attached */
3143         /* make sure we can reach put_old from new_root */
3144         if (!is_path_reachable(old_mnt, old.dentry, &new))
3145                 goto out4;
3146         /* make certain new is below the root */
3147         if (!is_path_reachable(new_mnt, new.dentry, &root))
3148                 goto out4;
3149         root_mp->m_count++; /* pin it so it won't go away */
3150         lock_mount_hash();
3151         detach_mnt(new_mnt, &parent_path);
3152         detach_mnt(root_mnt, &root_parent);
3153         if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3154                 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3155                 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3156         }
3157         /* mount old root on put_old */
3158         attach_mnt(root_mnt, old_mnt, old_mp);
3159         /* mount new_root on / */
3160         attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3161         touch_mnt_namespace(current->nsproxy->mnt_ns);
3162         /* A moved mount should not expire automatically */
3163         list_del_init(&new_mnt->mnt_expire);
3164         put_mountpoint(root_mp);
3165         unlock_mount_hash();
3166         chroot_fs_refs(&root, &new);
3167         error = 0;
3168 out4:
3169         unlock_mount(old_mp);
3170         if (!error) {
3171                 path_put(&root_parent);
3172                 path_put(&parent_path);
3173         }
3174 out3:
3175         path_put(&root);
3176 out2:
3177         path_put(&old);
3178 out1:
3179         path_put(&new);
3180 out0:
3181         return error;
3182 }
3183 
3184 static void __init init_mount_tree(void)
3185 {
3186         struct vfsmount *mnt;
3187         struct mnt_namespace *ns;
3188         struct path root;
3189         struct file_system_type *type;
3190 
3191         type = get_fs_type("rootfs");
3192         if (!type)
3193                 panic("Can't find rootfs type");
3194         mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3195         put_filesystem(type);
3196         if (IS_ERR(mnt))
3197                 panic("Can't create rootfs");
3198 
3199         ns = create_mnt_ns(mnt);
3200         if (IS_ERR(ns))
3201                 panic("Can't allocate initial namespace");
3202 
3203         init_task.nsproxy->mnt_ns = ns;
3204         get_mnt_ns(ns);
3205 
3206         root.mnt = mnt;
3207         root.dentry = mnt->mnt_root;
3208         mnt->mnt_flags |= MNT_LOCKED;
3209 
3210         set_fs_pwd(current->fs, &root);
3211         set_fs_root(current->fs, &root);
3212 }
3213 
3214 void __init mnt_init(void)
3215 {
3216         unsigned u;
3217         int err;
3218 
3219         mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3220                         0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3221 
3222         mount_hashtable = alloc_large_system_hash("Mount-cache",
3223                                 sizeof(struct hlist_head),
3224                                 mhash_entries, 19,
3225                                 0,
3226                                 &m_hash_shift, &m_hash_mask, 0, 0);
3227         mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3228                                 sizeof(struct hlist_head),
3229                                 mphash_entries, 19,
3230                                 0,
3231                                 &mp_hash_shift, &mp_hash_mask, 0, 0);
3232 
3233         if (!mount_hashtable || !mountpoint_hashtable)
3234                 panic("Failed to allocate mount hash table\n");
3235 
3236         for (u = 0; u <= m_hash_mask; u++)
3237                 INIT_HLIST_HEAD(&mount_hashtable[u]);
3238         for (u = 0; u <= mp_hash_mask; u++)
3239                 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3240 
3241         kernfs_init();
3242 
3243         err = sysfs_init();
3244         if (err)
3245                 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3246                         __func__, err);
3247         fs_kobj = kobject_create_and_add("fs", NULL);
3248         if (!fs_kobj)
3249                 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3250         init_rootfs();
3251         init_mount_tree();
3252 }
3253 
3254 void put_mnt_ns(struct mnt_namespace *ns)
3255 {
3256         if (!atomic_dec_and_test(&ns->count))
3257                 return;
3258         drop_collected_mounts(&ns->root->mnt);
3259         free_mnt_ns(ns);
3260 }
3261 
3262 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3263 {
3264         struct vfsmount *mnt;
3265         mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3266         if (!IS_ERR(mnt)) {
3267                 /*
3268                  * it is a longterm mount, don't release mnt until
3269                  * we unmount before file sys is unregistered
3270                 */
3271                 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3272         }
3273         return mnt;
3274 }
3275 EXPORT_SYMBOL_GPL(kern_mount_data);
3276 
3277 void kern_unmount(struct vfsmount *mnt)
3278 {
3279         /* release long term mount so mount point can be released */
3280         if (!IS_ERR_OR_NULL(mnt)) {
3281                 real_mount(mnt)->mnt_ns = NULL;
3282                 synchronize_rcu();      /* yecchhh... */
3283                 mntput(mnt);
3284         }
3285 }
3286 EXPORT_SYMBOL(kern_unmount);
3287 
3288 bool our_mnt(struct vfsmount *mnt)
3289 {
3290         return check_mnt(real_mount(mnt));
3291 }
3292 
3293 bool current_chrooted(void)
3294 {
3295         /* Does the current process have a non-standard root */
3296         struct path ns_root;
3297         struct path fs_root;
3298         bool chrooted;
3299 
3300         /* Find the namespace root */
3301         ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3302         ns_root.dentry = ns_root.mnt->mnt_root;
3303         path_get(&ns_root);
3304         while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3305                 ;
3306 
3307         get_fs_root(current->fs, &fs_root);
3308 
3309         chrooted = !path_equal(&fs_root, &ns_root);
3310 
3311         path_put(&fs_root);
3312         path_put(&ns_root);
3313 
3314         return chrooted;
3315 }
3316 
3317 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3318                                 int *new_mnt_flags)
3319 {
3320         int new_flags = *new_mnt_flags;
3321         struct mount *mnt;
3322         bool visible = false;
3323 
3324         down_read(&namespace_sem);
3325         list_for_each_entry(mnt, &ns->list, mnt_list) {
3326                 struct mount *child;
3327                 int mnt_flags;
3328 
3329                 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3330                         continue;
3331 
3332                 /* This mount is not fully visible if it's root directory
3333                  * is not the root directory of the filesystem.
3334                  */
3335                 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3336                         continue;
3337 
3338                 /* A local view of the mount flags */
3339                 mnt_flags = mnt->mnt.mnt_flags;
3340 
3341                 /* Don't miss readonly hidden in the superblock flags */
3342                 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3343                         mnt_flags |= MNT_LOCK_READONLY;
3344 
3345                 /* Verify the mount flags are equal to or more permissive
3346                  * than the proposed new mount.
3347                  */
3348                 if ((mnt_flags & MNT_LOCK_READONLY) &&
3349                     !(new_flags & MNT_READONLY))
3350                         continue;
3351                 if ((mnt_flags & MNT_LOCK_ATIME) &&
3352                     ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3353                         continue;
3354 
3355                 /* This mount is not fully visible if there are any
3356                  * locked child mounts that cover anything except for
3357                  * empty directories.
3358                  */
3359                 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3360                         struct inode *inode = child->mnt_mountpoint->d_inode;
3361                         /* Only worry about locked mounts */
3362                         if (!(child->mnt.mnt_flags & MNT_LOCKED))
3363                                 continue;
3364                         /* Is the directory permanetly empty? */
3365                         if (!is_empty_dir_inode(inode))
3366                                 goto next;
3367                 }
3368                 /* Preserve the locked attributes */
3369                 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3370                                                MNT_LOCK_ATIME);
3371                 visible = true;
3372                 goto found;
3373         next:   ;
3374         }
3375 found:
3376         up_read(&namespace_sem);
3377         return visible;
3378 }
3379 
3380 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3381 {
3382         const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3383         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3384         unsigned long s_iflags;
3385 
3386         if (ns->user_ns == &init_user_ns)
3387                 return false;
3388 
3389         /* Can this filesystem be too revealing? */
3390         s_iflags = mnt->mnt_sb->s_iflags;
3391         if (!(s_iflags & SB_I_USERNS_VISIBLE))
3392                 return false;
3393 
3394         if ((s_iflags & required_iflags) != required_iflags) {
3395                 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3396                           required_iflags);
3397                 return true;
3398         }
3399 
3400         return !mnt_already_visible(ns, mnt, new_mnt_flags);
3401 }
3402 
3403 bool mnt_may_suid(struct vfsmount *mnt)
3404 {
3405         /*
3406          * Foreign mounts (accessed via fchdir or through /proc
3407          * symlinks) are always treated as if they are nosuid.  This
3408          * prevents namespaces from trusting potentially unsafe
3409          * suid/sgid bits, file caps, or security labels that originate
3410          * in other namespaces.
3411          */
3412         return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3413                current_in_userns(mnt->mnt_sb->s_user_ns);
3414 }
3415 
3416 static struct ns_common *mntns_get(struct task_struct *task)
3417 {
3418         struct ns_common *ns = NULL;
3419         struct nsproxy *nsproxy;
3420 
3421         task_lock(task);
3422         nsproxy = task->nsproxy;
3423         if (nsproxy) {
3424                 ns = &nsproxy->mnt_ns->ns;
3425                 get_mnt_ns(to_mnt_ns(ns));
3426         }
3427         task_unlock(task);
3428 
3429         return ns;
3430 }
3431 
3432 static void mntns_put(struct ns_common *ns)
3433 {
3434         put_mnt_ns(to_mnt_ns(ns));
3435 }
3436 
3437 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3438 {
3439         struct fs_struct *fs = current->fs;
3440         struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3441         struct path root;
3442 
3443         if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3444             !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3445             !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3446                 return -EPERM;
3447 
3448         if (fs->users != 1)
3449                 return -EINVAL;
3450 
3451         get_mnt_ns(mnt_ns);
3452         put_mnt_ns(nsproxy->mnt_ns);
3453         nsproxy->mnt_ns = mnt_ns;
3454 
3455         /* Find the root */
3456         root.mnt    = &mnt_ns->root->mnt;
3457         root.dentry = mnt_ns->root->mnt.mnt_root;
3458         path_get(&root);
3459         while(d_mountpoint(root.dentry) && follow_down_one(&root))
3460                 ;
3461 
3462         /* Update the pwd and root */
3463         set_fs_pwd(fs, &root);
3464         set_fs_root(fs, &root);
3465 
3466         path_put(&root);
3467         return 0;
3468 }
3469 
3470 static struct user_namespace *mntns_owner(struct ns_common *ns)
3471 {
3472         return to_mnt_ns(ns)->user_ns;
3473 }
3474 
3475 const struct proc_ns_operations mntns_operations = {
3476         .name           = "mnt",
3477         .type           = CLONE_NEWNS,
3478         .get            = mntns_get,
3479         .put            = mntns_put,
3480         .install        = mntns_install,
3481         .owner          = mntns_owner,
3482 };
3483 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us