~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/namespace.c

Version: ~ [ linux-5.19-rc8 ] ~ [ linux-5.18.14 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.57 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.133 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.207 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.253 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.289 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.324 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  *  linux/fs/namespace.c
  4  *
  5  * (C) Copyright Al Viro 2000, 2001
  6  *
  7  * Based on code from fs/super.c, copyright Linus Torvalds and others.
  8  * Heavily rewritten.
  9  */
 10 
 11 #include <linux/syscalls.h>
 12 #include <linux/export.h>
 13 #include <linux/capability.h>
 14 #include <linux/mnt_namespace.h>
 15 #include <linux/user_namespace.h>
 16 #include <linux/namei.h>
 17 #include <linux/security.h>
 18 #include <linux/cred.h>
 19 #include <linux/idr.h>
 20 #include <linux/init.h>         /* init_rootfs */
 21 #include <linux/fs_struct.h>    /* get_fs_root et.al. */
 22 #include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
 23 #include <linux/file.h>
 24 #include <linux/uaccess.h>
 25 #include <linux/proc_ns.h>
 26 #include <linux/magic.h>
 27 #include <linux/memblock.h>
 28 #include <linux/proc_fs.h>
 29 #include <linux/task_work.h>
 30 #include <linux/sched/task.h>
 31 #include <uapi/linux/mount.h>
 32 #include <linux/fs_context.h>
 33 #include <linux/shmem_fs.h>
 34 
 35 #include "pnode.h"
 36 #include "internal.h"
 37 
 38 /* Maximum number of mounts in a mount namespace */
 39 unsigned int sysctl_mount_max __read_mostly = 100000;
 40 
 41 static unsigned int m_hash_mask __read_mostly;
 42 static unsigned int m_hash_shift __read_mostly;
 43 static unsigned int mp_hash_mask __read_mostly;
 44 static unsigned int mp_hash_shift __read_mostly;
 45 
 46 static __initdata unsigned long mhash_entries;
 47 static int __init set_mhash_entries(char *str)
 48 {
 49         if (!str)
 50                 return 0;
 51         mhash_entries = simple_strtoul(str, &str, 0);
 52         return 1;
 53 }
 54 __setup("mhash_entries=", set_mhash_entries);
 55 
 56 static __initdata unsigned long mphash_entries;
 57 static int __init set_mphash_entries(char *str)
 58 {
 59         if (!str)
 60                 return 0;
 61         mphash_entries = simple_strtoul(str, &str, 0);
 62         return 1;
 63 }
 64 __setup("mphash_entries=", set_mphash_entries);
 65 
 66 static u64 event;
 67 static DEFINE_IDA(mnt_id_ida);
 68 static DEFINE_IDA(mnt_group_ida);
 69 
 70 static struct hlist_head *mount_hashtable __read_mostly;
 71 static struct hlist_head *mountpoint_hashtable __read_mostly;
 72 static struct kmem_cache *mnt_cache __read_mostly;
 73 static DECLARE_RWSEM(namespace_sem);
 74 static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
 75 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
 76 
 77 struct mount_kattr {
 78         unsigned int attr_set;
 79         unsigned int attr_clr;
 80         unsigned int propagation;
 81         unsigned int lookup_flags;
 82         bool recurse;
 83         struct user_namespace *mnt_userns;
 84 };
 85 
 86 /* /sys/fs */
 87 struct kobject *fs_kobj;
 88 EXPORT_SYMBOL_GPL(fs_kobj);
 89 
 90 /*
 91  * vfsmount lock may be taken for read to prevent changes to the
 92  * vfsmount hash, ie. during mountpoint lookups or walking back
 93  * up the tree.
 94  *
 95  * It should be taken for write in all cases where the vfsmount
 96  * tree or hash is modified or when a vfsmount structure is modified.
 97  */
 98 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
 99 
100 static inline void lock_mount_hash(void)
101 {
102         write_seqlock(&mount_lock);
103 }
104 
105 static inline void unlock_mount_hash(void)
106 {
107         write_sequnlock(&mount_lock);
108 }
109 
110 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
111 {
112         unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
113         tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
114         tmp = tmp + (tmp >> m_hash_shift);
115         return &mount_hashtable[tmp & m_hash_mask];
116 }
117 
118 static inline struct hlist_head *mp_hash(struct dentry *dentry)
119 {
120         unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
121         tmp = tmp + (tmp >> mp_hash_shift);
122         return &mountpoint_hashtable[tmp & mp_hash_mask];
123 }
124 
125 static int mnt_alloc_id(struct mount *mnt)
126 {
127         int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
128 
129         if (res < 0)
130                 return res;
131         mnt->mnt_id = res;
132         return 0;
133 }
134 
135 static void mnt_free_id(struct mount *mnt)
136 {
137         ida_free(&mnt_id_ida, mnt->mnt_id);
138 }
139 
140 /*
141  * Allocate a new peer group ID
142  */
143 static int mnt_alloc_group_id(struct mount *mnt)
144 {
145         int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
146 
147         if (res < 0)
148                 return res;
149         mnt->mnt_group_id = res;
150         return 0;
151 }
152 
153 /*
154  * Release a peer group ID
155  */
156 void mnt_release_group_id(struct mount *mnt)
157 {
158         ida_free(&mnt_group_ida, mnt->mnt_group_id);
159         mnt->mnt_group_id = 0;
160 }
161 
162 /*
163  * vfsmount lock must be held for read
164  */
165 static inline void mnt_add_count(struct mount *mnt, int n)
166 {
167 #ifdef CONFIG_SMP
168         this_cpu_add(mnt->mnt_pcp->mnt_count, n);
169 #else
170         preempt_disable();
171         mnt->mnt_count += n;
172         preempt_enable();
173 #endif
174 }
175 
176 /*
177  * vfsmount lock must be held for write
178  */
179 int mnt_get_count(struct mount *mnt)
180 {
181 #ifdef CONFIG_SMP
182         int count = 0;
183         int cpu;
184 
185         for_each_possible_cpu(cpu) {
186                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
187         }
188 
189         return count;
190 #else
191         return mnt->mnt_count;
192 #endif
193 }
194 
195 static struct mount *alloc_vfsmnt(const char *name)
196 {
197         struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
198         if (mnt) {
199                 int err;
200 
201                 err = mnt_alloc_id(mnt);
202                 if (err)
203                         goto out_free_cache;
204 
205                 if (name) {
206                         mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
207                         if (!mnt->mnt_devname)
208                                 goto out_free_id;
209                 }
210 
211 #ifdef CONFIG_SMP
212                 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
213                 if (!mnt->mnt_pcp)
214                         goto out_free_devname;
215 
216                 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
217 #else
218                 mnt->mnt_count = 1;
219                 mnt->mnt_writers = 0;
220 #endif
221 
222                 INIT_HLIST_NODE(&mnt->mnt_hash);
223                 INIT_LIST_HEAD(&mnt->mnt_child);
224                 INIT_LIST_HEAD(&mnt->mnt_mounts);
225                 INIT_LIST_HEAD(&mnt->mnt_list);
226                 INIT_LIST_HEAD(&mnt->mnt_expire);
227                 INIT_LIST_HEAD(&mnt->mnt_share);
228                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
229                 INIT_LIST_HEAD(&mnt->mnt_slave);
230                 INIT_HLIST_NODE(&mnt->mnt_mp_list);
231                 INIT_LIST_HEAD(&mnt->mnt_umounting);
232                 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
233                 mnt->mnt.mnt_userns = &init_user_ns;
234         }
235         return mnt;
236 
237 #ifdef CONFIG_SMP
238 out_free_devname:
239         kfree_const(mnt->mnt_devname);
240 #endif
241 out_free_id:
242         mnt_free_id(mnt);
243 out_free_cache:
244         kmem_cache_free(mnt_cache, mnt);
245         return NULL;
246 }
247 
248 /*
249  * Most r/o checks on a fs are for operations that take
250  * discrete amounts of time, like a write() or unlink().
251  * We must keep track of when those operations start
252  * (for permission checks) and when they end, so that
253  * we can determine when writes are able to occur to
254  * a filesystem.
255  */
256 /*
257  * __mnt_is_readonly: check whether a mount is read-only
258  * @mnt: the mount to check for its write status
259  *
260  * This shouldn't be used directly ouside of the VFS.
261  * It does not guarantee that the filesystem will stay
262  * r/w, just that it is right *now*.  This can not and
263  * should not be used in place of IS_RDONLY(inode).
264  * mnt_want/drop_write() will _keep_ the filesystem
265  * r/w.
266  */
267 bool __mnt_is_readonly(struct vfsmount *mnt)
268 {
269         return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
270 }
271 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
272 
273 static inline void mnt_inc_writers(struct mount *mnt)
274 {
275 #ifdef CONFIG_SMP
276         this_cpu_inc(mnt->mnt_pcp->mnt_writers);
277 #else
278         mnt->mnt_writers++;
279 #endif
280 }
281 
282 static inline void mnt_dec_writers(struct mount *mnt)
283 {
284 #ifdef CONFIG_SMP
285         this_cpu_dec(mnt->mnt_pcp->mnt_writers);
286 #else
287         mnt->mnt_writers--;
288 #endif
289 }
290 
291 static unsigned int mnt_get_writers(struct mount *mnt)
292 {
293 #ifdef CONFIG_SMP
294         unsigned int count = 0;
295         int cpu;
296 
297         for_each_possible_cpu(cpu) {
298                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
299         }
300 
301         return count;
302 #else
303         return mnt->mnt_writers;
304 #endif
305 }
306 
307 static int mnt_is_readonly(struct vfsmount *mnt)
308 {
309         if (mnt->mnt_sb->s_readonly_remount)
310                 return 1;
311         /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
312         smp_rmb();
313         return __mnt_is_readonly(mnt);
314 }
315 
316 /*
317  * Most r/o & frozen checks on a fs are for operations that take discrete
318  * amounts of time, like a write() or unlink().  We must keep track of when
319  * those operations start (for permission checks) and when they end, so that we
320  * can determine when writes are able to occur to a filesystem.
321  */
322 /**
323  * __mnt_want_write - get write access to a mount without freeze protection
324  * @m: the mount on which to take a write
325  *
326  * This tells the low-level filesystem that a write is about to be performed to
327  * it, and makes sure that writes are allowed (mnt it read-write) before
328  * returning success. This operation does not protect against filesystem being
329  * frozen. When the write operation is finished, __mnt_drop_write() must be
330  * called. This is effectively a refcount.
331  */
332 int __mnt_want_write(struct vfsmount *m)
333 {
334         struct mount *mnt = real_mount(m);
335         int ret = 0;
336 
337         preempt_disable();
338         mnt_inc_writers(mnt);
339         /*
340          * The store to mnt_inc_writers must be visible before we pass
341          * MNT_WRITE_HOLD loop below, so that the slowpath can see our
342          * incremented count after it has set MNT_WRITE_HOLD.
343          */
344         smp_mb();
345         while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
346                 cpu_relax();
347         /*
348          * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
349          * be set to match its requirements. So we must not load that until
350          * MNT_WRITE_HOLD is cleared.
351          */
352         smp_rmb();
353         if (mnt_is_readonly(m)) {
354                 mnt_dec_writers(mnt);
355                 ret = -EROFS;
356         }
357         preempt_enable();
358 
359         return ret;
360 }
361 
362 /**
363  * mnt_want_write - get write access to a mount
364  * @m: the mount on which to take a write
365  *
366  * This tells the low-level filesystem that a write is about to be performed to
367  * it, and makes sure that writes are allowed (mount is read-write, filesystem
368  * is not frozen) before returning success.  When the write operation is
369  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
370  */
371 int mnt_want_write(struct vfsmount *m)
372 {
373         int ret;
374 
375         sb_start_write(m->mnt_sb);
376         ret = __mnt_want_write(m);
377         if (ret)
378                 sb_end_write(m->mnt_sb);
379         return ret;
380 }
381 EXPORT_SYMBOL_GPL(mnt_want_write);
382 
383 /**
384  * __mnt_want_write_file - get write access to a file's mount
385  * @file: the file who's mount on which to take a write
386  *
387  * This is like __mnt_want_write, but if the file is already open for writing it
388  * skips incrementing mnt_writers (since the open file already has a reference)
389  * and instead only does the check for emergency r/o remounts.  This must be
390  * paired with __mnt_drop_write_file.
391  */
392 int __mnt_want_write_file(struct file *file)
393 {
394         if (file->f_mode & FMODE_WRITER) {
395                 /*
396                  * Superblock may have become readonly while there are still
397                  * writable fd's, e.g. due to a fs error with errors=remount-ro
398                  */
399                 if (__mnt_is_readonly(file->f_path.mnt))
400                         return -EROFS;
401                 return 0;
402         }
403         return __mnt_want_write(file->f_path.mnt);
404 }
405 
406 /**
407  * mnt_want_write_file - get write access to a file's mount
408  * @file: the file who's mount on which to take a write
409  *
410  * This is like mnt_want_write, but if the file is already open for writing it
411  * skips incrementing mnt_writers (since the open file already has a reference)
412  * and instead only does the freeze protection and the check for emergency r/o
413  * remounts.  This must be paired with mnt_drop_write_file.
414  */
415 int mnt_want_write_file(struct file *file)
416 {
417         int ret;
418 
419         sb_start_write(file_inode(file)->i_sb);
420         ret = __mnt_want_write_file(file);
421         if (ret)
422                 sb_end_write(file_inode(file)->i_sb);
423         return ret;
424 }
425 EXPORT_SYMBOL_GPL(mnt_want_write_file);
426 
427 /**
428  * __mnt_drop_write - give up write access to a mount
429  * @mnt: the mount on which to give up write access
430  *
431  * Tells the low-level filesystem that we are done
432  * performing writes to it.  Must be matched with
433  * __mnt_want_write() call above.
434  */
435 void __mnt_drop_write(struct vfsmount *mnt)
436 {
437         preempt_disable();
438         mnt_dec_writers(real_mount(mnt));
439         preempt_enable();
440 }
441 
442 /**
443  * mnt_drop_write - give up write access to a mount
444  * @mnt: the mount on which to give up write access
445  *
446  * Tells the low-level filesystem that we are done performing writes to it and
447  * also allows filesystem to be frozen again.  Must be matched with
448  * mnt_want_write() call above.
449  */
450 void mnt_drop_write(struct vfsmount *mnt)
451 {
452         __mnt_drop_write(mnt);
453         sb_end_write(mnt->mnt_sb);
454 }
455 EXPORT_SYMBOL_GPL(mnt_drop_write);
456 
457 void __mnt_drop_write_file(struct file *file)
458 {
459         if (!(file->f_mode & FMODE_WRITER))
460                 __mnt_drop_write(file->f_path.mnt);
461 }
462 
463 void mnt_drop_write_file(struct file *file)
464 {
465         __mnt_drop_write_file(file);
466         sb_end_write(file_inode(file)->i_sb);
467 }
468 EXPORT_SYMBOL(mnt_drop_write_file);
469 
470 static inline int mnt_hold_writers(struct mount *mnt)
471 {
472         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
473         /*
474          * After storing MNT_WRITE_HOLD, we'll read the counters. This store
475          * should be visible before we do.
476          */
477         smp_mb();
478 
479         /*
480          * With writers on hold, if this value is zero, then there are
481          * definitely no active writers (although held writers may subsequently
482          * increment the count, they'll have to wait, and decrement it after
483          * seeing MNT_READONLY).
484          *
485          * It is OK to have counter incremented on one CPU and decremented on
486          * another: the sum will add up correctly. The danger would be when we
487          * sum up each counter, if we read a counter before it is incremented,
488          * but then read another CPU's count which it has been subsequently
489          * decremented from -- we would see more decrements than we should.
490          * MNT_WRITE_HOLD protects against this scenario, because
491          * mnt_want_write first increments count, then smp_mb, then spins on
492          * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
493          * we're counting up here.
494          */
495         if (mnt_get_writers(mnt) > 0)
496                 return -EBUSY;
497 
498         return 0;
499 }
500 
501 static inline void mnt_unhold_writers(struct mount *mnt)
502 {
503         /*
504          * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
505          * that become unheld will see MNT_READONLY.
506          */
507         smp_wmb();
508         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
509 }
510 
511 static int mnt_make_readonly(struct mount *mnt)
512 {
513         int ret;
514 
515         ret = mnt_hold_writers(mnt);
516         if (!ret)
517                 mnt->mnt.mnt_flags |= MNT_READONLY;
518         mnt_unhold_writers(mnt);
519         return ret;
520 }
521 
522 int sb_prepare_remount_readonly(struct super_block *sb)
523 {
524         struct mount *mnt;
525         int err = 0;
526 
527         /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
528         if (atomic_long_read(&sb->s_remove_count))
529                 return -EBUSY;
530 
531         lock_mount_hash();
532         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
533                 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
534                         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
535                         smp_mb();
536                         if (mnt_get_writers(mnt) > 0) {
537                                 err = -EBUSY;
538                                 break;
539                         }
540                 }
541         }
542         if (!err && atomic_long_read(&sb->s_remove_count))
543                 err = -EBUSY;
544 
545         if (!err) {
546                 sb->s_readonly_remount = 1;
547                 smp_wmb();
548         }
549         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
550                 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
551                         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
552         }
553         unlock_mount_hash();
554 
555         return err;
556 }
557 
558 static void free_vfsmnt(struct mount *mnt)
559 {
560         struct user_namespace *mnt_userns;
561 
562         mnt_userns = mnt_user_ns(&mnt->mnt);
563         if (mnt_userns != &init_user_ns)
564                 put_user_ns(mnt_userns);
565         kfree_const(mnt->mnt_devname);
566 #ifdef CONFIG_SMP
567         free_percpu(mnt->mnt_pcp);
568 #endif
569         kmem_cache_free(mnt_cache, mnt);
570 }
571 
572 static void delayed_free_vfsmnt(struct rcu_head *head)
573 {
574         free_vfsmnt(container_of(head, struct mount, mnt_rcu));
575 }
576 
577 /* call under rcu_read_lock */
578 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
579 {
580         struct mount *mnt;
581         if (read_seqretry(&mount_lock, seq))
582                 return 1;
583         if (bastard == NULL)
584                 return 0;
585         mnt = real_mount(bastard);
586         mnt_add_count(mnt, 1);
587         smp_mb();                       // see mntput_no_expire()
588         if (likely(!read_seqretry(&mount_lock, seq)))
589                 return 0;
590         if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
591                 mnt_add_count(mnt, -1);
592                 return 1;
593         }
594         lock_mount_hash();
595         if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
596                 mnt_add_count(mnt, -1);
597                 unlock_mount_hash();
598                 return 1;
599         }
600         unlock_mount_hash();
601         /* caller will mntput() */
602         return -1;
603 }
604 
605 /* call under rcu_read_lock */
606 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
607 {
608         int res = __legitimize_mnt(bastard, seq);
609         if (likely(!res))
610                 return true;
611         if (unlikely(res < 0)) {
612                 rcu_read_unlock();
613                 mntput(bastard);
614                 rcu_read_lock();
615         }
616         return false;
617 }
618 
619 /*
620  * find the first mount at @dentry on vfsmount @mnt.
621  * call under rcu_read_lock()
622  */
623 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
624 {
625         struct hlist_head *head = m_hash(mnt, dentry);
626         struct mount *p;
627 
628         hlist_for_each_entry_rcu(p, head, mnt_hash)
629                 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
630                         return p;
631         return NULL;
632 }
633 
634 /*
635  * lookup_mnt - Return the first child mount mounted at path
636  *
637  * "First" means first mounted chronologically.  If you create the
638  * following mounts:
639  *
640  * mount /dev/sda1 /mnt
641  * mount /dev/sda2 /mnt
642  * mount /dev/sda3 /mnt
643  *
644  * Then lookup_mnt() on the base /mnt dentry in the root mount will
645  * return successively the root dentry and vfsmount of /dev/sda1, then
646  * /dev/sda2, then /dev/sda3, then NULL.
647  *
648  * lookup_mnt takes a reference to the found vfsmount.
649  */
650 struct vfsmount *lookup_mnt(const struct path *path)
651 {
652         struct mount *child_mnt;
653         struct vfsmount *m;
654         unsigned seq;
655 
656         rcu_read_lock();
657         do {
658                 seq = read_seqbegin(&mount_lock);
659                 child_mnt = __lookup_mnt(path->mnt, path->dentry);
660                 m = child_mnt ? &child_mnt->mnt : NULL;
661         } while (!legitimize_mnt(m, seq));
662         rcu_read_unlock();
663         return m;
664 }
665 
666 static inline void lock_ns_list(struct mnt_namespace *ns)
667 {
668         spin_lock(&ns->ns_lock);
669 }
670 
671 static inline void unlock_ns_list(struct mnt_namespace *ns)
672 {
673         spin_unlock(&ns->ns_lock);
674 }
675 
676 static inline bool mnt_is_cursor(struct mount *mnt)
677 {
678         return mnt->mnt.mnt_flags & MNT_CURSOR;
679 }
680 
681 /*
682  * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
683  *                         current mount namespace.
684  *
685  * The common case is dentries are not mountpoints at all and that
686  * test is handled inline.  For the slow case when we are actually
687  * dealing with a mountpoint of some kind, walk through all of the
688  * mounts in the current mount namespace and test to see if the dentry
689  * is a mountpoint.
690  *
691  * The mount_hashtable is not usable in the context because we
692  * need to identify all mounts that may be in the current mount
693  * namespace not just a mount that happens to have some specified
694  * parent mount.
695  */
696 bool __is_local_mountpoint(struct dentry *dentry)
697 {
698         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
699         struct mount *mnt;
700         bool is_covered = false;
701 
702         down_read(&namespace_sem);
703         lock_ns_list(ns);
704         list_for_each_entry(mnt, &ns->list, mnt_list) {
705                 if (mnt_is_cursor(mnt))
706                         continue;
707                 is_covered = (mnt->mnt_mountpoint == dentry);
708                 if (is_covered)
709                         break;
710         }
711         unlock_ns_list(ns);
712         up_read(&namespace_sem);
713 
714         return is_covered;
715 }
716 
717 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
718 {
719         struct hlist_head *chain = mp_hash(dentry);
720         struct mountpoint *mp;
721 
722         hlist_for_each_entry(mp, chain, m_hash) {
723                 if (mp->m_dentry == dentry) {
724                         mp->m_count++;
725                         return mp;
726                 }
727         }
728         return NULL;
729 }
730 
731 static struct mountpoint *get_mountpoint(struct dentry *dentry)
732 {
733         struct mountpoint *mp, *new = NULL;
734         int ret;
735 
736         if (d_mountpoint(dentry)) {
737                 /* might be worth a WARN_ON() */
738                 if (d_unlinked(dentry))
739                         return ERR_PTR(-ENOENT);
740 mountpoint:
741                 read_seqlock_excl(&mount_lock);
742                 mp = lookup_mountpoint(dentry);
743                 read_sequnlock_excl(&mount_lock);
744                 if (mp)
745                         goto done;
746         }
747 
748         if (!new)
749                 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
750         if (!new)
751                 return ERR_PTR(-ENOMEM);
752 
753 
754         /* Exactly one processes may set d_mounted */
755         ret = d_set_mounted(dentry);
756 
757         /* Someone else set d_mounted? */
758         if (ret == -EBUSY)
759                 goto mountpoint;
760 
761         /* The dentry is not available as a mountpoint? */
762         mp = ERR_PTR(ret);
763         if (ret)
764                 goto done;
765 
766         /* Add the new mountpoint to the hash table */
767         read_seqlock_excl(&mount_lock);
768         new->m_dentry = dget(dentry);
769         new->m_count = 1;
770         hlist_add_head(&new->m_hash, mp_hash(dentry));
771         INIT_HLIST_HEAD(&new->m_list);
772         read_sequnlock_excl(&mount_lock);
773 
774         mp = new;
775         new = NULL;
776 done:
777         kfree(new);
778         return mp;
779 }
780 
781 /*
782  * vfsmount lock must be held.  Additionally, the caller is responsible
783  * for serializing calls for given disposal list.
784  */
785 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
786 {
787         if (!--mp->m_count) {
788                 struct dentry *dentry = mp->m_dentry;
789                 BUG_ON(!hlist_empty(&mp->m_list));
790                 spin_lock(&dentry->d_lock);
791                 dentry->d_flags &= ~DCACHE_MOUNTED;
792                 spin_unlock(&dentry->d_lock);
793                 dput_to_list(dentry, list);
794                 hlist_del(&mp->m_hash);
795                 kfree(mp);
796         }
797 }
798 
799 /* called with namespace_lock and vfsmount lock */
800 static void put_mountpoint(struct mountpoint *mp)
801 {
802         __put_mountpoint(mp, &ex_mountpoints);
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 struct mountpoint *unhash_mnt(struct mount *mnt)
836 {
837         struct mountpoint *mp;
838         mnt->mnt_parent = mnt;
839         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
840         list_del_init(&mnt->mnt_child);
841         hlist_del_init_rcu(&mnt->mnt_hash);
842         hlist_del_init(&mnt->mnt_mp_list);
843         mp = mnt->mnt_mp;
844         mnt->mnt_mp = NULL;
845         return mp;
846 }
847 
848 /*
849  * vfsmount lock must be held for write
850  */
851 static void umount_mnt(struct mount *mnt)
852 {
853         put_mountpoint(unhash_mnt(mnt));
854 }
855 
856 /*
857  * vfsmount lock must be held for write
858  */
859 void mnt_set_mountpoint(struct mount *mnt,
860                         struct mountpoint *mp,
861                         struct mount *child_mnt)
862 {
863         mp->m_count++;
864         mnt_add_count(mnt, 1);  /* essentially, that's mntget */
865         child_mnt->mnt_mountpoint = mp->m_dentry;
866         child_mnt->mnt_parent = mnt;
867         child_mnt->mnt_mp = mp;
868         hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
869 }
870 
871 static void __attach_mnt(struct mount *mnt, struct mount *parent)
872 {
873         hlist_add_head_rcu(&mnt->mnt_hash,
874                            m_hash(&parent->mnt, mnt->mnt_mountpoint));
875         list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
876 }
877 
878 /*
879  * vfsmount lock must be held for write
880  */
881 static void attach_mnt(struct mount *mnt,
882                         struct mount *parent,
883                         struct mountpoint *mp)
884 {
885         mnt_set_mountpoint(parent, mp, mnt);
886         __attach_mnt(mnt, parent);
887 }
888 
889 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
890 {
891         struct mountpoint *old_mp = mnt->mnt_mp;
892         struct mount *old_parent = mnt->mnt_parent;
893 
894         list_del_init(&mnt->mnt_child);
895         hlist_del_init(&mnt->mnt_mp_list);
896         hlist_del_init_rcu(&mnt->mnt_hash);
897 
898         attach_mnt(mnt, parent, mp);
899 
900         put_mountpoint(old_mp);
901         mnt_add_count(old_parent, -1);
902 }
903 
904 /*
905  * vfsmount lock must be held for write
906  */
907 static void commit_tree(struct mount *mnt)
908 {
909         struct mount *parent = mnt->mnt_parent;
910         struct mount *m;
911         LIST_HEAD(head);
912         struct mnt_namespace *n = parent->mnt_ns;
913 
914         BUG_ON(parent == mnt);
915 
916         list_add_tail(&head, &mnt->mnt_list);
917         list_for_each_entry(m, &head, mnt_list)
918                 m->mnt_ns = n;
919 
920         list_splice(&head, n->list.prev);
921 
922         n->mounts += n->pending_mounts;
923         n->pending_mounts = 0;
924 
925         __attach_mnt(mnt, parent);
926         touch_mnt_namespace(n);
927 }
928 
929 static struct mount *next_mnt(struct mount *p, struct mount *root)
930 {
931         struct list_head *next = p->mnt_mounts.next;
932         if (next == &p->mnt_mounts) {
933                 while (1) {
934                         if (p == root)
935                                 return NULL;
936                         next = p->mnt_child.next;
937                         if (next != &p->mnt_parent->mnt_mounts)
938                                 break;
939                         p = p->mnt_parent;
940                 }
941         }
942         return list_entry(next, struct mount, mnt_child);
943 }
944 
945 static struct mount *skip_mnt_tree(struct mount *p)
946 {
947         struct list_head *prev = p->mnt_mounts.prev;
948         while (prev != &p->mnt_mounts) {
949                 p = list_entry(prev, struct mount, mnt_child);
950                 prev = p->mnt_mounts.prev;
951         }
952         return p;
953 }
954 
955 /**
956  * vfs_create_mount - Create a mount for a configured superblock
957  * @fc: The configuration context with the superblock attached
958  *
959  * Create a mount to an already configured superblock.  If necessary, the
960  * caller should invoke vfs_get_tree() before calling this.
961  *
962  * Note that this does not attach the mount to anything.
963  */
964 struct vfsmount *vfs_create_mount(struct fs_context *fc)
965 {
966         struct mount *mnt;
967 
968         if (!fc->root)
969                 return ERR_PTR(-EINVAL);
970 
971         mnt = alloc_vfsmnt(fc->source ?: "none");
972         if (!mnt)
973                 return ERR_PTR(-ENOMEM);
974 
975         if (fc->sb_flags & SB_KERNMOUNT)
976                 mnt->mnt.mnt_flags = MNT_INTERNAL;
977 
978         atomic_inc(&fc->root->d_sb->s_active);
979         mnt->mnt.mnt_sb         = fc->root->d_sb;
980         mnt->mnt.mnt_root       = dget(fc->root);
981         mnt->mnt_mountpoint     = mnt->mnt.mnt_root;
982         mnt->mnt_parent         = mnt;
983 
984         lock_mount_hash();
985         list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
986         unlock_mount_hash();
987         return &mnt->mnt;
988 }
989 EXPORT_SYMBOL(vfs_create_mount);
990 
991 struct vfsmount *fc_mount(struct fs_context *fc)
992 {
993         int err = vfs_get_tree(fc);
994         if (!err) {
995                 up_write(&fc->root->d_sb->s_umount);
996                 return vfs_create_mount(fc);
997         }
998         return ERR_PTR(err);
999 }
1000 EXPORT_SYMBOL(fc_mount);
1001 
1002 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1003                                 int flags, const char *name,
1004                                 void *data)
1005 {
1006         struct fs_context *fc;
1007         struct vfsmount *mnt;
1008         int ret = 0;
1009 
1010         if (!type)
1011                 return ERR_PTR(-EINVAL);
1012 
1013         fc = fs_context_for_mount(type, flags);
1014         if (IS_ERR(fc))
1015                 return ERR_CAST(fc);
1016 
1017         if (name)
1018                 ret = vfs_parse_fs_string(fc, "source",
1019                                           name, strlen(name));
1020         if (!ret)
1021                 ret = parse_monolithic_mount_data(fc, data);
1022         if (!ret)
1023                 mnt = fc_mount(fc);
1024         else
1025                 mnt = ERR_PTR(ret);
1026 
1027         put_fs_context(fc);
1028         return mnt;
1029 }
1030 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1031 
1032 struct vfsmount *
1033 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1034              const char *name, void *data)
1035 {
1036         /* Until it is worked out how to pass the user namespace
1037          * through from the parent mount to the submount don't support
1038          * unprivileged mounts with submounts.
1039          */
1040         if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1041                 return ERR_PTR(-EPERM);
1042 
1043         return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1044 }
1045 EXPORT_SYMBOL_GPL(vfs_submount);
1046 
1047 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1048                                         int flag)
1049 {
1050         struct super_block *sb = old->mnt.mnt_sb;
1051         struct mount *mnt;
1052         int err;
1053 
1054         mnt = alloc_vfsmnt(old->mnt_devname);
1055         if (!mnt)
1056                 return ERR_PTR(-ENOMEM);
1057 
1058         if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1059                 mnt->mnt_group_id = 0; /* not a peer of original */
1060         else
1061                 mnt->mnt_group_id = old->mnt_group_id;
1062 
1063         if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1064                 err = mnt_alloc_group_id(mnt);
1065                 if (err)
1066                         goto out_free;
1067         }
1068 
1069         mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1070         mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1071 
1072         atomic_inc(&sb->s_active);
1073         mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1074         if (mnt->mnt.mnt_userns != &init_user_ns)
1075                 mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1076         mnt->mnt.mnt_sb = sb;
1077         mnt->mnt.mnt_root = dget(root);
1078         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1079         mnt->mnt_parent = mnt;
1080         lock_mount_hash();
1081         list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1082         unlock_mount_hash();
1083 
1084         if ((flag & CL_SLAVE) ||
1085             ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1086                 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1087                 mnt->mnt_master = old;
1088                 CLEAR_MNT_SHARED(mnt);
1089         } else if (!(flag & CL_PRIVATE)) {
1090                 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1091                         list_add(&mnt->mnt_share, &old->mnt_share);
1092                 if (IS_MNT_SLAVE(old))
1093                         list_add(&mnt->mnt_slave, &old->mnt_slave);
1094                 mnt->mnt_master = old->mnt_master;
1095         } else {
1096                 CLEAR_MNT_SHARED(mnt);
1097         }
1098         if (flag & CL_MAKE_SHARED)
1099                 set_mnt_shared(mnt);
1100 
1101         /* stick the duplicate mount on the same expiry list
1102          * as the original if that was on one */
1103         if (flag & CL_EXPIRE) {
1104                 if (!list_empty(&old->mnt_expire))
1105                         list_add(&mnt->mnt_expire, &old->mnt_expire);
1106         }
1107 
1108         return mnt;
1109 
1110  out_free:
1111         mnt_free_id(mnt);
1112         free_vfsmnt(mnt);
1113         return ERR_PTR(err);
1114 }
1115 
1116 static void cleanup_mnt(struct mount *mnt)
1117 {
1118         struct hlist_node *p;
1119         struct mount *m;
1120         /*
1121          * The warning here probably indicates that somebody messed
1122          * up a mnt_want/drop_write() pair.  If this happens, the
1123          * filesystem was probably unable to make r/w->r/o transitions.
1124          * The locking used to deal with mnt_count decrement provides barriers,
1125          * so mnt_get_writers() below is safe.
1126          */
1127         WARN_ON(mnt_get_writers(mnt));
1128         if (unlikely(mnt->mnt_pins.first))
1129                 mnt_pin_kill(mnt);
1130         hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1131                 hlist_del(&m->mnt_umount);
1132                 mntput(&m->mnt);
1133         }
1134         fsnotify_vfsmount_delete(&mnt->mnt);
1135         dput(mnt->mnt.mnt_root);
1136         deactivate_super(mnt->mnt.mnt_sb);
1137         mnt_free_id(mnt);
1138         call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1139 }
1140 
1141 static void __cleanup_mnt(struct rcu_head *head)
1142 {
1143         cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1144 }
1145 
1146 static LLIST_HEAD(delayed_mntput_list);
1147 static void delayed_mntput(struct work_struct *unused)
1148 {
1149         struct llist_node *node = llist_del_all(&delayed_mntput_list);
1150         struct mount *m, *t;
1151 
1152         llist_for_each_entry_safe(m, t, node, mnt_llist)
1153                 cleanup_mnt(m);
1154 }
1155 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1156 
1157 static void mntput_no_expire(struct mount *mnt)
1158 {
1159         LIST_HEAD(list);
1160         int count;
1161 
1162         rcu_read_lock();
1163         if (likely(READ_ONCE(mnt->mnt_ns))) {
1164                 /*
1165                  * Since we don't do lock_mount_hash() here,
1166                  * ->mnt_ns can change under us.  However, if it's
1167                  * non-NULL, then there's a reference that won't
1168                  * be dropped until after an RCU delay done after
1169                  * turning ->mnt_ns NULL.  So if we observe it
1170                  * non-NULL under rcu_read_lock(), the reference
1171                  * we are dropping is not the final one.
1172                  */
1173                 mnt_add_count(mnt, -1);
1174                 rcu_read_unlock();
1175                 return;
1176         }
1177         lock_mount_hash();
1178         /*
1179          * make sure that if __legitimize_mnt() has not seen us grab
1180          * mount_lock, we'll see their refcount increment here.
1181          */
1182         smp_mb();
1183         mnt_add_count(mnt, -1);
1184         count = mnt_get_count(mnt);
1185         if (count != 0) {
1186                 WARN_ON(count < 0);
1187                 rcu_read_unlock();
1188                 unlock_mount_hash();
1189                 return;
1190         }
1191         if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1192                 rcu_read_unlock();
1193                 unlock_mount_hash();
1194                 return;
1195         }
1196         mnt->mnt.mnt_flags |= MNT_DOOMED;
1197         rcu_read_unlock();
1198 
1199         list_del(&mnt->mnt_instance);
1200 
1201         if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1202                 struct mount *p, *tmp;
1203                 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1204                         __put_mountpoint(unhash_mnt(p), &list);
1205                         hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1206                 }
1207         }
1208         unlock_mount_hash();
1209         shrink_dentry_list(&list);
1210 
1211         if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1212                 struct task_struct *task = current;
1213                 if (likely(!(task->flags & PF_KTHREAD))) {
1214                         init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1215                         if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1216                                 return;
1217                 }
1218                 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1219                         schedule_delayed_work(&delayed_mntput_work, 1);
1220                 return;
1221         }
1222         cleanup_mnt(mnt);
1223 }
1224 
1225 void mntput(struct vfsmount *mnt)
1226 {
1227         if (mnt) {
1228                 struct mount *m = real_mount(mnt);
1229                 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1230                 if (unlikely(m->mnt_expiry_mark))
1231                         m->mnt_expiry_mark = 0;
1232                 mntput_no_expire(m);
1233         }
1234 }
1235 EXPORT_SYMBOL(mntput);
1236 
1237 struct vfsmount *mntget(struct vfsmount *mnt)
1238 {
1239         if (mnt)
1240                 mnt_add_count(real_mount(mnt), 1);
1241         return mnt;
1242 }
1243 EXPORT_SYMBOL(mntget);
1244 
1245 /**
1246  * path_is_mountpoint() - Check if path is a mount in the current namespace.
1247  * @path: path to check
1248  *
1249  *  d_mountpoint() can only be used reliably to establish if a dentry is
1250  *  not mounted in any namespace and that common case is handled inline.
1251  *  d_mountpoint() isn't aware of the possibility there may be multiple
1252  *  mounts using a given dentry in a different namespace. This function
1253  *  checks if the passed in path is a mountpoint rather than the dentry
1254  *  alone.
1255  */
1256 bool path_is_mountpoint(const struct path *path)
1257 {
1258         unsigned seq;
1259         bool res;
1260 
1261         if (!d_mountpoint(path->dentry))
1262                 return false;
1263 
1264         rcu_read_lock();
1265         do {
1266                 seq = read_seqbegin(&mount_lock);
1267                 res = __path_is_mountpoint(path);
1268         } while (read_seqretry(&mount_lock, seq));
1269         rcu_read_unlock();
1270 
1271         return res;
1272 }
1273 EXPORT_SYMBOL(path_is_mountpoint);
1274 
1275 struct vfsmount *mnt_clone_internal(const struct path *path)
1276 {
1277         struct mount *p;
1278         p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1279         if (IS_ERR(p))
1280                 return ERR_CAST(p);
1281         p->mnt.mnt_flags |= MNT_INTERNAL;
1282         return &p->mnt;
1283 }
1284 
1285 #ifdef CONFIG_PROC_FS
1286 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1287                                    struct list_head *p)
1288 {
1289         struct mount *mnt, *ret = NULL;
1290 
1291         lock_ns_list(ns);
1292         list_for_each_continue(p, &ns->list) {
1293                 mnt = list_entry(p, typeof(*mnt), mnt_list);
1294                 if (!mnt_is_cursor(mnt)) {
1295                         ret = mnt;
1296                         break;
1297                 }
1298         }
1299         unlock_ns_list(ns);
1300 
1301         return ret;
1302 }
1303 
1304 /* iterator; we want it to have access to namespace_sem, thus here... */
1305 static void *m_start(struct seq_file *m, loff_t *pos)
1306 {
1307         struct proc_mounts *p = m->private;
1308         struct list_head *prev;
1309 
1310         down_read(&namespace_sem);
1311         if (!*pos) {
1312                 prev = &p->ns->list;
1313         } else {
1314                 prev = &p->cursor.mnt_list;
1315 
1316                 /* Read after we'd reached the end? */
1317                 if (list_empty(prev))
1318                         return NULL;
1319         }
1320 
1321         return mnt_list_next(p->ns, prev);
1322 }
1323 
1324 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1325 {
1326         struct proc_mounts *p = m->private;
1327         struct mount *mnt = v;
1328 
1329         ++*pos;
1330         return mnt_list_next(p->ns, &mnt->mnt_list);
1331 }
1332 
1333 static void m_stop(struct seq_file *m, void *v)
1334 {
1335         struct proc_mounts *p = m->private;
1336         struct mount *mnt = v;
1337 
1338         lock_ns_list(p->ns);
1339         if (mnt)
1340                 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1341         else
1342                 list_del_init(&p->cursor.mnt_list);
1343         unlock_ns_list(p->ns);
1344         up_read(&namespace_sem);
1345 }
1346 
1347 static int m_show(struct seq_file *m, void *v)
1348 {
1349         struct proc_mounts *p = m->private;
1350         struct mount *r = v;
1351         return p->show(m, &r->mnt);
1352 }
1353 
1354 const struct seq_operations mounts_op = {
1355         .start  = m_start,
1356         .next   = m_next,
1357         .stop   = m_stop,
1358         .show   = m_show,
1359 };
1360 
1361 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1362 {
1363         down_read(&namespace_sem);
1364         lock_ns_list(ns);
1365         list_del(&cursor->mnt_list);
1366         unlock_ns_list(ns);
1367         up_read(&namespace_sem);
1368 }
1369 #endif  /* CONFIG_PROC_FS */
1370 
1371 /**
1372  * may_umount_tree - check if a mount tree is busy
1373  * @m: root of mount tree
1374  *
1375  * This is called to check if a tree of mounts has any
1376  * open files, pwds, chroots or sub mounts that are
1377  * busy.
1378  */
1379 int may_umount_tree(struct vfsmount *m)
1380 {
1381         struct mount *mnt = real_mount(m);
1382         int actual_refs = 0;
1383         int minimum_refs = 0;
1384         struct mount *p;
1385         BUG_ON(!m);
1386 
1387         /* write lock needed for mnt_get_count */
1388         lock_mount_hash();
1389         for (p = mnt; p; p = next_mnt(p, mnt)) {
1390                 actual_refs += mnt_get_count(p);
1391                 minimum_refs += 2;
1392         }
1393         unlock_mount_hash();
1394 
1395         if (actual_refs > minimum_refs)
1396                 return 0;
1397 
1398         return 1;
1399 }
1400 
1401 EXPORT_SYMBOL(may_umount_tree);
1402 
1403 /**
1404  * may_umount - check if a mount point is busy
1405  * @mnt: root of mount
1406  *
1407  * This is called to check if a mount point has any
1408  * open files, pwds, chroots or sub mounts. If the
1409  * mount has sub mounts this will return busy
1410  * regardless of whether the sub mounts are busy.
1411  *
1412  * Doesn't take quota and stuff into account. IOW, in some cases it will
1413  * give false negatives. The main reason why it's here is that we need
1414  * a non-destructive way to look for easily umountable filesystems.
1415  */
1416 int may_umount(struct vfsmount *mnt)
1417 {
1418         int ret = 1;
1419         down_read(&namespace_sem);
1420         lock_mount_hash();
1421         if (propagate_mount_busy(real_mount(mnt), 2))
1422                 ret = 0;
1423         unlock_mount_hash();
1424         up_read(&namespace_sem);
1425         return ret;
1426 }
1427 
1428 EXPORT_SYMBOL(may_umount);
1429 
1430 static void namespace_unlock(void)
1431 {
1432         struct hlist_head head;
1433         struct hlist_node *p;
1434         struct mount *m;
1435         LIST_HEAD(list);
1436 
1437         hlist_move_list(&unmounted, &head);
1438         list_splice_init(&ex_mountpoints, &list);
1439 
1440         up_write(&namespace_sem);
1441 
1442         shrink_dentry_list(&list);
1443 
1444         if (likely(hlist_empty(&head)))
1445                 return;
1446 
1447         synchronize_rcu_expedited();
1448 
1449         hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1450                 hlist_del(&m->mnt_umount);
1451                 mntput(&m->mnt);
1452         }
1453 }
1454 
1455 static inline void namespace_lock(void)
1456 {
1457         down_write(&namespace_sem);
1458 }
1459 
1460 enum umount_tree_flags {
1461         UMOUNT_SYNC = 1,
1462         UMOUNT_PROPAGATE = 2,
1463         UMOUNT_CONNECTED = 4,
1464 };
1465 
1466 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1467 {
1468         /* Leaving mounts connected is only valid for lazy umounts */
1469         if (how & UMOUNT_SYNC)
1470                 return true;
1471 
1472         /* A mount without a parent has nothing to be connected to */
1473         if (!mnt_has_parent(mnt))
1474                 return true;
1475 
1476         /* Because the reference counting rules change when mounts are
1477          * unmounted and connected, umounted mounts may not be
1478          * connected to mounted mounts.
1479          */
1480         if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1481                 return true;
1482 
1483         /* Has it been requested that the mount remain connected? */
1484         if (how & UMOUNT_CONNECTED)
1485                 return false;
1486 
1487         /* Is the mount locked such that it needs to remain connected? */
1488         if (IS_MNT_LOCKED(mnt))
1489                 return false;
1490 
1491         /* By default disconnect the mount */
1492         return true;
1493 }
1494 
1495 /*
1496  * mount_lock must be held
1497  * namespace_sem must be held for write
1498  */
1499 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1500 {
1501         LIST_HEAD(tmp_list);
1502         struct mount *p;
1503 
1504         if (how & UMOUNT_PROPAGATE)
1505                 propagate_mount_unlock(mnt);
1506 
1507         /* Gather the mounts to umount */
1508         for (p = mnt; p; p = next_mnt(p, mnt)) {
1509                 p->mnt.mnt_flags |= MNT_UMOUNT;
1510                 list_move(&p->mnt_list, &tmp_list);
1511         }
1512 
1513         /* Hide the mounts from mnt_mounts */
1514         list_for_each_entry(p, &tmp_list, mnt_list) {
1515                 list_del_init(&p->mnt_child);
1516         }
1517 
1518         /* Add propogated mounts to the tmp_list */
1519         if (how & UMOUNT_PROPAGATE)
1520                 propagate_umount(&tmp_list);
1521 
1522         while (!list_empty(&tmp_list)) {
1523                 struct mnt_namespace *ns;
1524                 bool disconnect;
1525                 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1526                 list_del_init(&p->mnt_expire);
1527                 list_del_init(&p->mnt_list);
1528                 ns = p->mnt_ns;
1529                 if (ns) {
1530                         ns->mounts--;
1531                         __touch_mnt_namespace(ns);
1532                 }
1533                 p->mnt_ns = NULL;
1534                 if (how & UMOUNT_SYNC)
1535                         p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1536 
1537                 disconnect = disconnect_mount(p, how);
1538                 if (mnt_has_parent(p)) {
1539                         mnt_add_count(p->mnt_parent, -1);
1540                         if (!disconnect) {
1541                                 /* Don't forget about p */
1542                                 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1543                         } else {
1544                                 umount_mnt(p);
1545                         }
1546                 }
1547                 change_mnt_propagation(p, MS_PRIVATE);
1548                 if (disconnect)
1549                         hlist_add_head(&p->mnt_umount, &unmounted);
1550         }
1551 }
1552 
1553 static void shrink_submounts(struct mount *mnt);
1554 
1555 static int do_umount_root(struct super_block *sb)
1556 {
1557         int ret = 0;
1558 
1559         down_write(&sb->s_umount);
1560         if (!sb_rdonly(sb)) {
1561                 struct fs_context *fc;
1562 
1563                 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1564                                                 SB_RDONLY);
1565                 if (IS_ERR(fc)) {
1566                         ret = PTR_ERR(fc);
1567                 } else {
1568                         ret = parse_monolithic_mount_data(fc, NULL);
1569                         if (!ret)
1570                                 ret = reconfigure_super(fc);
1571                         put_fs_context(fc);
1572                 }
1573         }
1574         up_write(&sb->s_umount);
1575         return ret;
1576 }
1577 
1578 static int do_umount(struct mount *mnt, int flags)
1579 {
1580         struct super_block *sb = mnt->mnt.mnt_sb;
1581         int retval;
1582 
1583         retval = security_sb_umount(&mnt->mnt, flags);
1584         if (retval)
1585                 return retval;
1586 
1587         /*
1588          * Allow userspace to request a mountpoint be expired rather than
1589          * unmounting unconditionally. Unmount only happens if:
1590          *  (1) the mark is already set (the mark is cleared by mntput())
1591          *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1592          */
1593         if (flags & MNT_EXPIRE) {
1594                 if (&mnt->mnt == current->fs->root.mnt ||
1595                     flags & (MNT_FORCE | MNT_DETACH))
1596                         return -EINVAL;
1597 
1598                 /*
1599                  * probably don't strictly need the lock here if we examined
1600                  * all race cases, but it's a slowpath.
1601                  */
1602                 lock_mount_hash();
1603                 if (mnt_get_count(mnt) != 2) {
1604                         unlock_mount_hash();
1605                         return -EBUSY;
1606                 }
1607                 unlock_mount_hash();
1608 
1609                 if (!xchg(&mnt->mnt_expiry_mark, 1))
1610                         return -EAGAIN;
1611         }
1612 
1613         /*
1614          * If we may have to abort operations to get out of this
1615          * mount, and they will themselves hold resources we must
1616          * allow the fs to do things. In the Unix tradition of
1617          * 'Gee thats tricky lets do it in userspace' the umount_begin
1618          * might fail to complete on the first run through as other tasks
1619          * must return, and the like. Thats for the mount program to worry
1620          * about for the moment.
1621          */
1622 
1623         if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1624                 sb->s_op->umount_begin(sb);
1625         }
1626 
1627         /*
1628          * No sense to grab the lock for this test, but test itself looks
1629          * somewhat bogus. Suggestions for better replacement?
1630          * Ho-hum... In principle, we might treat that as umount + switch
1631          * to rootfs. GC would eventually take care of the old vfsmount.
1632          * Actually it makes sense, especially if rootfs would contain a
1633          * /reboot - static binary that would close all descriptors and
1634          * call reboot(9). Then init(8) could umount root and exec /reboot.
1635          */
1636         if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1637                 /*
1638                  * Special case for "unmounting" root ...
1639                  * we just try to remount it readonly.
1640                  */
1641                 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1642                         return -EPERM;
1643                 return do_umount_root(sb);
1644         }
1645 
1646         namespace_lock();
1647         lock_mount_hash();
1648 
1649         /* Recheck MNT_LOCKED with the locks held */
1650         retval = -EINVAL;
1651         if (mnt->mnt.mnt_flags & MNT_LOCKED)
1652                 goto out;
1653 
1654         event++;
1655         if (flags & MNT_DETACH) {
1656                 if (!list_empty(&mnt->mnt_list))
1657                         umount_tree(mnt, UMOUNT_PROPAGATE);
1658                 retval = 0;
1659         } else {
1660                 shrink_submounts(mnt);
1661                 retval = -EBUSY;
1662                 if (!propagate_mount_busy(mnt, 2)) {
1663                         if (!list_empty(&mnt->mnt_list))
1664                                 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1665                         retval = 0;
1666                 }
1667         }
1668 out:
1669         unlock_mount_hash();
1670         namespace_unlock();
1671         return retval;
1672 }
1673 
1674 /*
1675  * __detach_mounts - lazily unmount all mounts on the specified dentry
1676  *
1677  * During unlink, rmdir, and d_drop it is possible to loose the path
1678  * to an existing mountpoint, and wind up leaking the mount.
1679  * detach_mounts allows lazily unmounting those mounts instead of
1680  * leaking them.
1681  *
1682  * The caller may hold dentry->d_inode->i_mutex.
1683  */
1684 void __detach_mounts(struct dentry *dentry)
1685 {
1686         struct mountpoint *mp;
1687         struct mount *mnt;
1688 
1689         namespace_lock();
1690         lock_mount_hash();
1691         mp = lookup_mountpoint(dentry);
1692         if (!mp)
1693                 goto out_unlock;
1694 
1695         event++;
1696         while (!hlist_empty(&mp->m_list)) {
1697                 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1698                 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1699                         umount_mnt(mnt);
1700                         hlist_add_head(&mnt->mnt_umount, &unmounted);
1701                 }
1702                 else umount_tree(mnt, UMOUNT_CONNECTED);
1703         }
1704         put_mountpoint(mp);
1705 out_unlock:
1706         unlock_mount_hash();
1707         namespace_unlock();
1708 }
1709 
1710 /*
1711  * Is the caller allowed to modify his namespace?
1712  */
1713 static inline bool may_mount(void)
1714 {
1715         return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1716 }
1717 
1718 #ifdef  CONFIG_MANDATORY_FILE_LOCKING
1719 static bool may_mandlock(void)
1720 {
1721         pr_warn_once("======================================================\n"
1722                      "WARNING: the mand mount option is being deprecated and\n"
1723                      "         will be removed in v5.15!\n"
1724                      "======================================================\n");
1725         return capable(CAP_SYS_ADMIN);
1726 }
1727 #else
1728 static inline bool may_mandlock(void)
1729 {
1730         pr_warn("VFS: \"mand\" mount option not supported");
1731         return false;
1732 }
1733 #endif
1734 
1735 static int can_umount(const struct path *path, int flags)
1736 {
1737         struct mount *mnt = real_mount(path->mnt);
1738 
1739         if (!may_mount())
1740                 return -EPERM;
1741         if (path->dentry != path->mnt->mnt_root)
1742                 return -EINVAL;
1743         if (!check_mnt(mnt))
1744                 return -EINVAL;
1745         if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1746                 return -EINVAL;
1747         if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1748                 return -EPERM;
1749         return 0;
1750 }
1751 
1752 // caller is responsible for flags being sane
1753 int path_umount(struct path *path, int flags)
1754 {
1755         struct mount *mnt = real_mount(path->mnt);
1756         int ret;
1757 
1758         ret = can_umount(path, flags);
1759         if (!ret)
1760                 ret = do_umount(mnt, flags);
1761 
1762         /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1763         dput(path->dentry);
1764         mntput_no_expire(mnt);
1765         return ret;
1766 }
1767 
1768 static int ksys_umount(char __user *name, int flags)
1769 {
1770         int lookup_flags = LOOKUP_MOUNTPOINT;
1771         struct path path;
1772         int ret;
1773 
1774         // basic validity checks done first
1775         if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1776                 return -EINVAL;
1777 
1778         if (!(flags & UMOUNT_NOFOLLOW))
1779                 lookup_flags |= LOOKUP_FOLLOW;
1780         ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1781         if (ret)
1782                 return ret;
1783         return path_umount(&path, flags);
1784 }
1785 
1786 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1787 {
1788         return ksys_umount(name, flags);
1789 }
1790 
1791 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1792 
1793 /*
1794  *      The 2.0 compatible umount. No flags.
1795  */
1796 SYSCALL_DEFINE1(oldumount, char __user *, name)
1797 {
1798         return ksys_umount(name, 0);
1799 }
1800 
1801 #endif
1802 
1803 static bool is_mnt_ns_file(struct dentry *dentry)
1804 {
1805         /* Is this a proxy for a mount namespace? */
1806         return dentry->d_op == &ns_dentry_operations &&
1807                dentry->d_fsdata == &mntns_operations;
1808 }
1809 
1810 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1811 {
1812         return container_of(ns, struct mnt_namespace, ns);
1813 }
1814 
1815 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1816 {
1817         return &mnt->ns;
1818 }
1819 
1820 static bool mnt_ns_loop(struct dentry *dentry)
1821 {
1822         /* Could bind mounting the mount namespace inode cause a
1823          * mount namespace loop?
1824          */
1825         struct mnt_namespace *mnt_ns;
1826         if (!is_mnt_ns_file(dentry))
1827                 return false;
1828 
1829         mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1830         return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1831 }
1832 
1833 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1834                                         int flag)
1835 {
1836         struct mount *res, *p, *q, *r, *parent;
1837 
1838         if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1839                 return ERR_PTR(-EINVAL);
1840 
1841         if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1842                 return ERR_PTR(-EINVAL);
1843 
1844         res = q = clone_mnt(mnt, dentry, flag);
1845         if (IS_ERR(q))
1846                 return q;
1847 
1848         q->mnt_mountpoint = mnt->mnt_mountpoint;
1849 
1850         p = mnt;
1851         list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1852                 struct mount *s;
1853                 if (!is_subdir(r->mnt_mountpoint, dentry))
1854                         continue;
1855 
1856                 for (s = r; s; s = next_mnt(s, r)) {
1857                         if (!(flag & CL_COPY_UNBINDABLE) &&
1858                             IS_MNT_UNBINDABLE(s)) {
1859                                 if (s->mnt.mnt_flags & MNT_LOCKED) {
1860                                         /* Both unbindable and locked. */
1861                                         q = ERR_PTR(-EPERM);
1862                                         goto out;
1863                                 } else {
1864                                         s = skip_mnt_tree(s);
1865                                         continue;
1866                                 }
1867                         }
1868                         if (!(flag & CL_COPY_MNT_NS_FILE) &&
1869                             is_mnt_ns_file(s->mnt.mnt_root)) {
1870                                 s = skip_mnt_tree(s);
1871                                 continue;
1872                         }
1873                         while (p != s->mnt_parent) {
1874                                 p = p->mnt_parent;
1875                                 q = q->mnt_parent;
1876                         }
1877                         p = s;
1878                         parent = q;
1879                         q = clone_mnt(p, p->mnt.mnt_root, flag);
1880                         if (IS_ERR(q))
1881                                 goto out;
1882                         lock_mount_hash();
1883                         list_add_tail(&q->mnt_list, &res->mnt_list);
1884                         attach_mnt(q, parent, p->mnt_mp);
1885                         unlock_mount_hash();
1886                 }
1887         }
1888         return res;
1889 out:
1890         if (res) {
1891                 lock_mount_hash();
1892                 umount_tree(res, UMOUNT_SYNC);
1893                 unlock_mount_hash();
1894         }
1895         return q;
1896 }
1897 
1898 /* Caller should check returned pointer for errors */
1899 
1900 struct vfsmount *collect_mounts(const struct path *path)
1901 {
1902         struct mount *tree;
1903         namespace_lock();
1904         if (!check_mnt(real_mount(path->mnt)))
1905                 tree = ERR_PTR(-EINVAL);
1906         else
1907                 tree = copy_tree(real_mount(path->mnt), path->dentry,
1908                                  CL_COPY_ALL | CL_PRIVATE);
1909         namespace_unlock();
1910         if (IS_ERR(tree))
1911                 return ERR_CAST(tree);
1912         return &tree->mnt;
1913 }
1914 
1915 static void free_mnt_ns(struct mnt_namespace *);
1916 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1917 
1918 void dissolve_on_fput(struct vfsmount *mnt)
1919 {
1920         struct mnt_namespace *ns;
1921         namespace_lock();
1922         lock_mount_hash();
1923         ns = real_mount(mnt)->mnt_ns;
1924         if (ns) {
1925                 if (is_anon_ns(ns))
1926                         umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1927                 else
1928                         ns = NULL;
1929         }
1930         unlock_mount_hash();
1931         namespace_unlock();
1932         if (ns)
1933                 free_mnt_ns(ns);
1934 }
1935 
1936 void drop_collected_mounts(struct vfsmount *mnt)
1937 {
1938         namespace_lock();
1939         lock_mount_hash();
1940         umount_tree(real_mount(mnt), 0);
1941         unlock_mount_hash();
1942         namespace_unlock();
1943 }
1944 
1945 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1946 {
1947         struct mount *child;
1948 
1949         list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1950                 if (!is_subdir(child->mnt_mountpoint, dentry))
1951                         continue;
1952 
1953                 if (child->mnt.mnt_flags & MNT_LOCKED)
1954                         return true;
1955         }
1956         return false;
1957 }
1958 
1959 /**
1960  * clone_private_mount - create a private clone of a path
1961  * @path: path to clone
1962  *
1963  * This creates a new vfsmount, which will be the clone of @path.  The new mount
1964  * will not be attached anywhere in the namespace and will be private (i.e.
1965  * changes to the originating mount won't be propagated into this).
1966  *
1967  * Release with mntput().
1968  */
1969 struct vfsmount *clone_private_mount(const struct path *path)
1970 {
1971         struct mount *old_mnt = real_mount(path->mnt);
1972         struct mount *new_mnt;
1973 
1974         down_read(&namespace_sem);
1975         if (IS_MNT_UNBINDABLE(old_mnt))
1976                 goto invalid;
1977 
1978         if (!check_mnt(old_mnt))
1979                 goto invalid;
1980 
1981         if (has_locked_children(old_mnt, path->dentry))
1982                 goto invalid;
1983 
1984         new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1985         up_read(&namespace_sem);
1986 
1987         if (IS_ERR(new_mnt))
1988                 return ERR_CAST(new_mnt);
1989 
1990         /* Longterm mount to be removed by kern_unmount*() */
1991         new_mnt->mnt_ns = MNT_NS_INTERNAL;
1992 
1993         return &new_mnt->mnt;
1994 
1995 invalid:
1996         up_read(&namespace_sem);
1997         return ERR_PTR(-EINVAL);
1998 }
1999 EXPORT_SYMBOL_GPL(clone_private_mount);
2000 
2001 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2002                    struct vfsmount *root)
2003 {
2004         struct mount *mnt;
2005         int res = f(root, arg);
2006         if (res)
2007                 return res;
2008         list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2009                 res = f(&mnt->mnt, arg);
2010                 if (res)
2011                         return res;
2012         }
2013         return 0;
2014 }
2015 
2016 static void lock_mnt_tree(struct mount *mnt)
2017 {
2018         struct mount *p;
2019 
2020         for (p = mnt; p; p = next_mnt(p, mnt)) {
2021                 int flags = p->mnt.mnt_flags;
2022                 /* Don't allow unprivileged users to change mount flags */
2023                 flags |= MNT_LOCK_ATIME;
2024 
2025                 if (flags & MNT_READONLY)
2026                         flags |= MNT_LOCK_READONLY;
2027 
2028                 if (flags & MNT_NODEV)
2029                         flags |= MNT_LOCK_NODEV;
2030 
2031                 if (flags & MNT_NOSUID)
2032                         flags |= MNT_LOCK_NOSUID;
2033 
2034                 if (flags & MNT_NOEXEC)
2035                         flags |= MNT_LOCK_NOEXEC;
2036                 /* Don't allow unprivileged users to reveal what is under a mount */
2037                 if (list_empty(&p->mnt_expire))
2038                         flags |= MNT_LOCKED;
2039                 p->mnt.mnt_flags = flags;
2040         }
2041 }
2042 
2043 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2044 {
2045         struct mount *p;
2046 
2047         for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2048                 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2049                         mnt_release_group_id(p);
2050         }
2051 }
2052 
2053 static int invent_group_ids(struct mount *mnt, bool recurse)
2054 {
2055         struct mount *p;
2056 
2057         for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2058                 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2059                         int err = mnt_alloc_group_id(p);
2060                         if (err) {
2061                                 cleanup_group_ids(mnt, p);
2062                                 return err;
2063                         }
2064                 }
2065         }
2066 
2067         return 0;
2068 }
2069 
2070 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2071 {
2072         unsigned int max = READ_ONCE(sysctl_mount_max);
2073         unsigned int mounts = 0, old, pending, sum;
2074         struct mount *p;
2075 
2076         for (p = mnt; p; p = next_mnt(p, mnt))
2077                 mounts++;
2078 
2079         old = ns->mounts;
2080         pending = ns->pending_mounts;
2081         sum = old + pending;
2082         if ((old > sum) ||
2083             (pending > sum) ||
2084             (max < sum) ||
2085             (mounts > (max - sum)))
2086                 return -ENOSPC;
2087 
2088         ns->pending_mounts = pending + mounts;
2089         return 0;
2090 }
2091 
2092 /*
2093  *  @source_mnt : mount tree to be attached
2094  *  @nd         : place the mount tree @source_mnt is attached
2095  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
2096  *                 store the parent mount and mountpoint dentry.
2097  *                 (done when source_mnt is moved)
2098  *
2099  *  NOTE: in the table below explains the semantics when a source mount
2100  *  of a given type is attached to a destination mount of a given type.
2101  * ---------------------------------------------------------------------------
2102  * |         BIND MOUNT OPERATION                                            |
2103  * |**************************************************************************
2104  * | source-->| shared        |       private  |       slave    | unbindable |
2105  * | dest     |               |                |                |            |
2106  * |   |      |               |                |                |            |
2107  * |   v      |               |                |                |            |
2108  * |**************************************************************************
2109  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
2110  * |          |               |                |                |            |
2111  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
2112  * ***************************************************************************
2113  * A bind operation clones the source mount and mounts the clone on the
2114  * destination mount.
2115  *
2116  * (++)  the cloned mount is propagated to all the mounts in the propagation
2117  *       tree of the destination mount and the cloned mount is added to
2118  *       the peer group of the source mount.
2119  * (+)   the cloned mount is created under the destination mount and is marked
2120  *       as shared. The cloned mount is added to the peer group of the source
2121  *       mount.
2122  * (+++) the mount is propagated to all the mounts in the propagation tree
2123  *       of the destination mount and the cloned mount is made slave
2124  *       of the same master as that of the source mount. The cloned mount
2125  *       is marked as 'shared and slave'.
2126  * (*)   the cloned mount is made a slave of the same master as that of the
2127  *       source mount.
2128  *
2129  * ---------------------------------------------------------------------------
2130  * |                    MOVE MOUNT OPERATION                                 |
2131  * |**************************************************************************
2132  * | source-->| shared        |       private  |       slave    | unbindable |
2133  * | dest     |               |                |                |            |
2134  * |   |      |               |                |                |            |
2135  * |   v      |               |                |                |            |
2136  * |**************************************************************************
2137  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
2138  * |          |               |                |                |            |
2139  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
2140  * ***************************************************************************
2141  *
2142  * (+)  the mount is moved to the destination. And is then propagated to
2143  *      all the mounts in the propagation tree of the destination mount.
2144  * (+*)  the mount is moved to the destination.
2145  * (+++)  the mount is moved to the destination and is then propagated to
2146  *      all the mounts belonging to the destination mount's propagation tree.
2147  *      the mount is marked as 'shared and slave'.
2148  * (*)  the mount continues to be a slave at the new location.
2149  *
2150  * if the source mount is a tree, the operations explained above is
2151  * applied to each mount in the tree.
2152  * Must be called without spinlocks held, since this function can sleep
2153  * in allocations.
2154  */
2155 static int attach_recursive_mnt(struct mount *source_mnt,
2156                         struct mount *dest_mnt,
2157                         struct mountpoint *dest_mp,
2158                         bool moving)
2159 {
2160         struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2161         HLIST_HEAD(tree_list);
2162         struct mnt_namespace *ns = dest_mnt->mnt_ns;
2163         struct mountpoint *smp;
2164         struct mount *child, *p;
2165         struct hlist_node *n;
2166         int err;
2167 
2168         /* Preallocate a mountpoint in case the new mounts need
2169          * to be tucked under other mounts.
2170          */
2171         smp = get_mountpoint(source_mnt->mnt.mnt_root);
2172         if (IS_ERR(smp))
2173                 return PTR_ERR(smp);
2174 
2175         /* Is there space to add these mounts to the mount namespace? */
2176         if (!moving) {
2177                 err = count_mounts(ns, source_mnt);
2178                 if (err)
2179                         goto out;
2180         }
2181 
2182         if (IS_MNT_SHARED(dest_mnt)) {
2183                 err = invent_group_ids(source_mnt, true);
2184                 if (err)
2185                         goto out;
2186                 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2187                 lock_mount_hash();
2188                 if (err)
2189                         goto out_cleanup_ids;
2190                 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2191                         set_mnt_shared(p);
2192         } else {
2193                 lock_mount_hash();
2194         }
2195         if (moving) {
2196                 unhash_mnt(source_mnt);
2197                 attach_mnt(source_mnt, dest_mnt, dest_mp);
2198                 touch_mnt_namespace(source_mnt->mnt_ns);
2199         } else {
2200                 if (source_mnt->mnt_ns) {
2201                         /* move from anon - the caller will destroy */
2202                         list_del_init(&source_mnt->mnt_ns->list);
2203                 }
2204                 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2205                 commit_tree(source_mnt);
2206         }
2207 
2208         hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2209                 struct mount *q;
2210                 hlist_del_init(&child->mnt_hash);
2211                 q = __lookup_mnt(&child->mnt_parent->mnt,
2212                                  child->mnt_mountpoint);
2213                 if (q)
2214                         mnt_change_mountpoint(child, smp, q);
2215                 /* Notice when we are propagating across user namespaces */
2216                 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2217                         lock_mnt_tree(child);
2218                 child->mnt.mnt_flags &= ~MNT_LOCKED;
2219                 commit_tree(child);
2220         }
2221         put_mountpoint(smp);
2222         unlock_mount_hash();
2223 
2224         return 0;
2225 
2226  out_cleanup_ids:
2227         while (!hlist_empty(&tree_list)) {
2228                 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2229                 child->mnt_parent->mnt_ns->pending_mounts = 0;
2230                 umount_tree(child, UMOUNT_SYNC);
2231         }
2232         unlock_mount_hash();
2233         cleanup_group_ids(source_mnt, NULL);
2234  out:
2235         ns->pending_mounts = 0;
2236 
2237         read_seqlock_excl(&mount_lock);
2238         put_mountpoint(smp);
2239         read_sequnlock_excl(&mount_lock);
2240 
2241         return err;
2242 }
2243 
2244 static struct mountpoint *lock_mount(struct path *path)
2245 {
2246         struct vfsmount *mnt;
2247         struct dentry *dentry = path->dentry;
2248 retry:
2249         inode_lock(dentry->d_inode);
2250         if (unlikely(cant_mount(dentry))) {
2251                 inode_unlock(dentry->d_inode);
2252                 return ERR_PTR(-ENOENT);
2253         }
2254         namespace_lock();
2255         mnt = lookup_mnt(path);
2256         if (likely(!mnt)) {
2257                 struct mountpoint *mp = get_mountpoint(dentry);
2258                 if (IS_ERR(mp)) {
2259                         namespace_unlock();
2260                         inode_unlock(dentry->d_inode);
2261                         return mp;
2262                 }
2263                 return mp;
2264         }
2265         namespace_unlock();
2266         inode_unlock(path->dentry->d_inode);
2267         path_put(path);
2268         path->mnt = mnt;
2269         dentry = path->dentry = dget(mnt->mnt_root);
2270         goto retry;
2271 }
2272 
2273 static void unlock_mount(struct mountpoint *where)
2274 {
2275         struct dentry *dentry = where->m_dentry;
2276 
2277         read_seqlock_excl(&mount_lock);
2278         put_mountpoint(where);
2279         read_sequnlock_excl(&mount_lock);
2280 
2281         namespace_unlock();
2282         inode_unlock(dentry->d_inode);
2283 }
2284 
2285 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2286 {
2287         if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2288                 return -EINVAL;
2289 
2290         if (d_is_dir(mp->m_dentry) !=
2291               d_is_dir(mnt->mnt.mnt_root))
2292                 return -ENOTDIR;
2293 
2294         return attach_recursive_mnt(mnt, p, mp, false);
2295 }
2296 
2297 /*
2298  * Sanity check the flags to change_mnt_propagation.
2299  */
2300 
2301 static int flags_to_propagation_type(int ms_flags)
2302 {
2303         int type = ms_flags & ~(MS_REC | MS_SILENT);
2304 
2305         /* Fail if any non-propagation flags are set */
2306         if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2307                 return 0;
2308         /* Only one propagation flag should be set */
2309         if (!is_power_of_2(type))
2310                 return 0;
2311         return type;
2312 }
2313 
2314 /*
2315  * recursively change the type of the mountpoint.
2316  */
2317 static int do_change_type(struct path *path, int ms_flags)
2318 {
2319         struct mount *m;
2320         struct mount *mnt = real_mount(path->mnt);
2321         int recurse = ms_flags & MS_REC;
2322         int type;
2323         int err = 0;
2324 
2325         if (path->dentry != path->mnt->mnt_root)
2326                 return -EINVAL;
2327 
2328         type = flags_to_propagation_type(ms_flags);
2329         if (!type)
2330                 return -EINVAL;
2331 
2332         namespace_lock();
2333         if (type == MS_SHARED) {
2334                 err = invent_group_ids(mnt, recurse);
2335                 if (err)
2336                         goto out_unlock;
2337         }
2338 
2339         lock_mount_hash();
2340         for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2341                 change_mnt_propagation(m, type);
2342         unlock_mount_hash();
2343 
2344  out_unlock:
2345         namespace_unlock();
2346         return err;
2347 }
2348 
2349 static struct mount *__do_loopback(struct path *old_path, int recurse)
2350 {
2351         struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2352 
2353         if (IS_MNT_UNBINDABLE(old))
2354                 return mnt;
2355 
2356         if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2357                 return mnt;
2358 
2359         if (!recurse && has_locked_children(old, old_path->dentry))
2360                 return mnt;
2361 
2362         if (recurse)
2363                 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2364         else
2365                 mnt = clone_mnt(old, old_path->dentry, 0);
2366 
2367         if (!IS_ERR(mnt))
2368                 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2369 
2370         return mnt;
2371 }
2372 
2373 /*
2374  * do loopback mount.
2375  */
2376 static int do_loopback(struct path *path, const char *old_name,
2377                                 int recurse)
2378 {
2379         struct path old_path;
2380         struct mount *mnt = NULL, *parent;
2381         struct mountpoint *mp;
2382         int err;
2383         if (!old_name || !*old_name)
2384                 return -EINVAL;
2385         err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2386         if (err)
2387                 return err;
2388 
2389         err = -EINVAL;
2390         if (mnt_ns_loop(old_path.dentry))
2391                 goto out;
2392 
2393         mp = lock_mount(path);
2394         if (IS_ERR(mp)) {
2395                 err = PTR_ERR(mp);
2396                 goto out;
2397         }
2398 
2399         parent = real_mount(path->mnt);
2400         if (!check_mnt(parent))
2401                 goto out2;
2402 
2403         mnt = __do_loopback(&old_path, recurse);
2404         if (IS_ERR(mnt)) {
2405                 err = PTR_ERR(mnt);
2406                 goto out2;
2407         }
2408 
2409         err = graft_tree(mnt, parent, mp);
2410         if (err) {
2411                 lock_mount_hash();
2412                 umount_tree(mnt, UMOUNT_SYNC);
2413                 unlock_mount_hash();
2414         }
2415 out2:
2416         unlock_mount(mp);
2417 out:
2418         path_put(&old_path);
2419         return err;
2420 }
2421 
2422 static struct file *open_detached_copy(struct path *path, bool recursive)
2423 {
2424         struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2425         struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2426         struct mount *mnt, *p;
2427         struct file *file;
2428 
2429         if (IS_ERR(ns))
2430                 return ERR_CAST(ns);
2431 
2432         namespace_lock();
2433         mnt = __do_loopback(path, recursive);
2434         if (IS_ERR(mnt)) {
2435                 namespace_unlock();
2436                 free_mnt_ns(ns);
2437                 return ERR_CAST(mnt);
2438         }
2439 
2440         lock_mount_hash();
2441         for (p = mnt; p; p = next_mnt(p, mnt)) {
2442                 p->mnt_ns = ns;
2443                 ns->mounts++;
2444         }
2445         ns->root = mnt;
2446         list_add_tail(&ns->list, &mnt->mnt_list);
2447         mntget(&mnt->mnt);
2448         unlock_mount_hash();
2449         namespace_unlock();
2450 
2451         mntput(path->mnt);
2452         path->mnt = &mnt->mnt;
2453         file = dentry_open(path, O_PATH, current_cred());
2454         if (IS_ERR(file))
2455                 dissolve_on_fput(path->mnt);
2456         else
2457                 file->f_mode |= FMODE_NEED_UNMOUNT;
2458         return file;
2459 }
2460 
2461 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2462 {
2463         struct file *file;
2464         struct path path;
2465         int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2466         bool detached = flags & OPEN_TREE_CLONE;
2467         int error;
2468         int fd;
2469 
2470         BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2471 
2472         if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2473                       AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2474                       OPEN_TREE_CLOEXEC))
2475                 return -EINVAL;
2476 
2477         if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2478                 return -EINVAL;
2479 
2480         if (flags & AT_NO_AUTOMOUNT)
2481                 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2482         if (flags & AT_SYMLINK_NOFOLLOW)
2483                 lookup_flags &= ~LOOKUP_FOLLOW;
2484         if (flags & AT_EMPTY_PATH)
2485                 lookup_flags |= LOOKUP_EMPTY;
2486 
2487         if (detached && !may_mount())
2488                 return -EPERM;
2489 
2490         fd = get_unused_fd_flags(flags & O_CLOEXEC);
2491         if (fd < 0)
2492                 return fd;
2493 
2494         error = user_path_at(dfd, filename, lookup_flags, &path);
2495         if (unlikely(error)) {
2496                 file = ERR_PTR(error);
2497         } else {
2498                 if (detached)
2499                         file = open_detached_copy(&path, flags & AT_RECURSIVE);
2500                 else
2501                         file = dentry_open(&path, O_PATH, current_cred());
2502                 path_put(&path);
2503         }
2504         if (IS_ERR(file)) {
2505                 put_unused_fd(fd);
2506                 return PTR_ERR(file);
2507         }
2508         fd_install(fd, file);
2509         return fd;
2510 }
2511 
2512 /*
2513  * Don't allow locked mount flags to be cleared.
2514  *
2515  * No locks need to be held here while testing the various MNT_LOCK
2516  * flags because those flags can never be cleared once they are set.
2517  */
2518 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2519 {
2520         unsigned int fl = mnt->mnt.mnt_flags;
2521 
2522         if ((fl & MNT_LOCK_READONLY) &&
2523             !(mnt_flags & MNT_READONLY))
2524                 return false;
2525 
2526         if ((fl & MNT_LOCK_NODEV) &&
2527             !(mnt_flags & MNT_NODEV))
2528                 return false;
2529 
2530         if ((fl & MNT_LOCK_NOSUID) &&
2531             !(mnt_flags & MNT_NOSUID))
2532                 return false;
2533 
2534         if ((fl & MNT_LOCK_NOEXEC) &&
2535             !(mnt_flags & MNT_NOEXEC))
2536                 return false;
2537 
2538         if ((fl & MNT_LOCK_ATIME) &&
2539             ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2540                 return false;
2541 
2542         return true;
2543 }
2544 
2545 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2546 {
2547         bool readonly_request = (mnt_flags & MNT_READONLY);
2548 
2549         if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2550                 return 0;
2551 
2552         if (readonly_request)
2553                 return mnt_make_readonly(mnt);
2554 
2555         mnt->mnt.mnt_flags &= ~MNT_READONLY;
2556         return 0;
2557 }
2558 
2559 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2560 {
2561         mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2562         mnt->mnt.mnt_flags = mnt_flags;
2563         touch_mnt_namespace(mnt->mnt_ns);
2564 }
2565 
2566 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2567 {
2568         struct super_block *sb = mnt->mnt_sb;
2569 
2570         if (!__mnt_is_readonly(mnt) &&
2571            (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2572                 char *buf = (char *)__get_free_page(GFP_KERNEL);
2573                 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2574                 struct tm tm;
2575 
2576                 time64_to_tm(sb->s_time_max, 0, &tm);
2577 
2578                 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2579                         sb->s_type->name,
2580                         is_mounted(mnt) ? "remounted" : "mounted",
2581                         mntpath,
2582                         tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2583 
2584                 free_page((unsigned long)buf);
2585         }
2586 }
2587 
2588 /*
2589  * Handle reconfiguration of the mountpoint only without alteration of the
2590  * superblock it refers to.  This is triggered by specifying MS_REMOUNT|MS_BIND
2591  * to mount(2).
2592  */
2593 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2594 {
2595         struct super_block *sb = path->mnt->mnt_sb;
2596         struct mount *mnt = real_mount(path->mnt);
2597         int ret;
2598 
2599         if (!check_mnt(mnt))
2600                 return -EINVAL;
2601 
2602         if (path->dentry != mnt->mnt.mnt_root)
2603                 return -EINVAL;
2604 
2605         if (!can_change_locked_flags(mnt, mnt_flags))
2606                 return -EPERM;
2607 
2608         /*
2609          * We're only checking whether the superblock is read-only not
2610          * changing it, so only take down_read(&sb->s_umount).
2611          */
2612         down_read(&sb->s_umount);
2613         lock_mount_hash();
2614         ret = change_mount_ro_state(mnt, mnt_flags);
2615         if (ret == 0)
2616                 set_mount_attributes(mnt, mnt_flags);
2617         unlock_mount_hash();
2618         up_read(&sb->s_umount);
2619 
2620         mnt_warn_timestamp_expiry(path, &mnt->mnt);
2621 
2622         return ret;
2623 }
2624 
2625 /*
2626  * change filesystem flags. dir should be a physical root of filesystem.
2627  * If you've mounted a non-root directory somewhere and want to do remount
2628  * on it - tough luck.
2629  */
2630 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2631                       int mnt_flags, void *data)
2632 {
2633         int err;
2634         struct super_block *sb = path->mnt->mnt_sb;
2635         struct mount *mnt = real_mount(path->mnt);
2636         struct fs_context *fc;
2637 
2638         if (!check_mnt(mnt))
2639                 return -EINVAL;
2640 
2641         if (path->dentry != path->mnt->mnt_root)
2642                 return -EINVAL;
2643 
2644         if (!can_change_locked_flags(mnt, mnt_flags))
2645                 return -EPERM;
2646 
2647         fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2648         if (IS_ERR(fc))
2649                 return PTR_ERR(fc);
2650 
2651         fc->oldapi = true;
2652         err = parse_monolithic_mount_data(fc, data);
2653         if (!err) {
2654                 down_write(&sb->s_umount);
2655                 err = -EPERM;
2656                 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2657                         err = reconfigure_super(fc);
2658                         if (!err) {
2659                                 lock_mount_hash();
2660                                 set_mount_attributes(mnt, mnt_flags);
2661                                 unlock_mount_hash();
2662                         }
2663                 }
2664                 up_write(&sb->s_umount);
2665         }
2666 
2667         mnt_warn_timestamp_expiry(path, &mnt->mnt);
2668 
2669         put_fs_context(fc);
2670         return err;
2671 }
2672 
2673 static inline int tree_contains_unbindable(struct mount *mnt)
2674 {
2675         struct mount *p;
2676         for (p = mnt; p; p = next_mnt(p, mnt)) {
2677                 if (IS_MNT_UNBINDABLE(p))
2678                         return 1;
2679         }
2680         return 0;
2681 }
2682 
2683 /*
2684  * Check that there aren't references to earlier/same mount namespaces in the
2685  * specified subtree.  Such references can act as pins for mount namespaces
2686  * that aren't checked by the mount-cycle checking code, thereby allowing
2687  * cycles to be made.
2688  */
2689 static bool check_for_nsfs_mounts(struct mount *subtree)
2690 {
2691         struct mount *p;
2692         bool ret = false;
2693 
2694         lock_mount_hash();
2695         for (p = subtree; p; p = next_mnt(p, subtree))
2696                 if (mnt_ns_loop(p->mnt.mnt_root))
2697                         goto out;
2698 
2699         ret = true;
2700 out:
2701         unlock_mount_hash();
2702         return ret;
2703 }
2704 
2705 static int do_move_mount(struct path *old_path, struct path *new_path)
2706 {
2707         struct mnt_namespace *ns;
2708         struct mount *p;
2709         struct mount *old;
2710         struct mount *parent;
2711         struct mountpoint *mp, *old_mp;
2712         int err;
2713         bool attached;
2714 
2715         mp = lock_mount(new_path);
2716         if (IS_ERR(mp))
2717                 return PTR_ERR(mp);
2718 
2719         old = real_mount(old_path->mnt);
2720         p = real_mount(new_path->mnt);
2721         parent = old->mnt_parent;
2722         attached = mnt_has_parent(old);
2723         old_mp = old->mnt_mp;
2724         ns = old->mnt_ns;
2725 
2726         err = -EINVAL;
2727         /* The mountpoint must be in our namespace. */
2728         if (!check_mnt(p))
2729                 goto out;
2730 
2731         /* The thing moved must be mounted... */
2732         if (!is_mounted(&old->mnt))
2733                 goto out;
2734 
2735         /* ... and either ours or the root of anon namespace */
2736         if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2737                 goto out;
2738 
2739         if (old->mnt.mnt_flags & MNT_LOCKED)
2740                 goto out;
2741 
2742         if (old_path->dentry != old_path->mnt->mnt_root)
2743                 goto out;
2744 
2745         if (d_is_dir(new_path->dentry) !=
2746             d_is_dir(old_path->dentry))
2747                 goto out;
2748         /*
2749          * Don't move a mount residing in a shared parent.
2750          */
2751         if (attached && IS_MNT_SHARED(parent))
2752                 goto out;
2753         /*
2754          * Don't move a mount tree containing unbindable mounts to a destination
2755          * mount which is shared.
2756          */
2757         if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2758                 goto out;
2759         err = -ELOOP;
2760         if (!check_for_nsfs_mounts(old))
2761                 goto out;
2762         for (; mnt_has_parent(p); p = p->mnt_parent)
2763                 if (p == old)
2764                         goto out;
2765 
2766         err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2767                                    attached);
2768         if (err)
2769                 goto out;
2770 
2771         /* if the mount is moved, it should no longer be expire
2772          * automatically */
2773         list_del_init(&old->mnt_expire);
2774         if (attached)
2775                 put_mountpoint(old_mp);
2776 out:
2777         unlock_mount(mp);
2778         if (!err) {
2779                 if (attached)
2780                         mntput_no_expire(parent);
2781                 else
2782                         free_mnt_ns(ns);
2783         }
2784         return err;
2785 }
2786 
2787 static int do_move_mount_old(struct path *path, const char *old_name)
2788 {
2789         struct path old_path;
2790         int err;
2791 
2792         if (!old_name || !*old_name)
2793                 return -EINVAL;
2794 
2795         err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2796         if (err)
2797                 return err;
2798 
2799         err = do_move_mount(&old_path, path);
2800         path_put(&old_path);
2801         return err;
2802 }
2803 
2804 /*
2805  * add a mount into a namespace's mount tree
2806  */
2807 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2808                         struct path *path, int mnt_flags)
2809 {
2810         struct mount *parent = real_mount(path->mnt);
2811 
2812         mnt_flags &= ~MNT_INTERNAL_FLAGS;
2813 
2814         if (unlikely(!check_mnt(parent))) {
2815                 /* that's acceptable only for automounts done in private ns */
2816                 if (!(mnt_flags & MNT_SHRINKABLE))
2817                         return -EINVAL;
2818                 /* ... and for those we'd better have mountpoint still alive */
2819                 if (!parent->mnt_ns)
2820                         return -EINVAL;
2821         }
2822 
2823         /* Refuse the same filesystem on the same mount point */
2824         if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2825             path->mnt->mnt_root == path->dentry)
2826                 return -EBUSY;
2827 
2828         if (d_is_symlink(newmnt->mnt.mnt_root))
2829                 return -EINVAL;
2830 
2831         newmnt->mnt.mnt_flags = mnt_flags;
2832         return graft_tree(newmnt, parent, mp);
2833 }
2834 
2835 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2836 
2837 /*
2838  * Create a new mount using a superblock configuration and request it
2839  * be added to the namespace tree.
2840  */
2841 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2842                            unsigned int mnt_flags)
2843 {
2844         struct vfsmount *mnt;
2845         struct mountpoint *mp;
2846         struct super_block *sb = fc->root->d_sb;
2847         int error;
2848 
2849         error = security_sb_kern_mount(sb);
2850         if (!error && mount_too_revealing(sb, &mnt_flags))
2851                 error = -EPERM;
2852 
2853         if (unlikely(error)) {
2854                 fc_drop_locked(fc);
2855                 return error;
2856         }
2857 
2858         up_write(&sb->s_umount);
2859 
2860         mnt = vfs_create_mount(fc);
2861         if (IS_ERR(mnt))
2862                 return PTR_ERR(mnt);
2863 
2864         mnt_warn_timestamp_expiry(mountpoint, mnt);
2865 
2866         mp = lock_mount(mountpoint);
2867         if (IS_ERR(mp)) {
2868                 mntput(mnt);
2869                 return PTR_ERR(mp);
2870         }
2871         error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2872         unlock_mount(mp);
2873         if (error < 0)
2874                 mntput(mnt);
2875         return error;
2876 }
2877 
2878 /*
2879  * create a new mount for userspace and request it to be added into the
2880  * namespace's tree
2881  */
2882 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2883                         int mnt_flags, const char *name, void *data)
2884 {
2885         struct file_system_type *type;
2886         struct fs_context *fc;
2887         const char *subtype = NULL;
2888         int err = 0;
2889 
2890         if (!fstype)
2891                 return -EINVAL;
2892 
2893         type = get_fs_type(fstype);
2894         if (!type)
2895                 return -ENODEV;
2896 
2897         if (type->fs_flags & FS_HAS_SUBTYPE) {
2898                 subtype = strchr(fstype, '.');
2899                 if (subtype) {
2900                         subtype++;
2901                         if (!*subtype) {
2902                                 put_filesystem(type);
2903                                 return -EINVAL;
2904                         }
2905                 }
2906         }
2907 
2908         fc = fs_context_for_mount(type, sb_flags);
2909         put_filesystem(type);
2910         if (IS_ERR(fc))
2911                 return PTR_ERR(fc);
2912 
2913         if (subtype)
2914                 err = vfs_parse_fs_string(fc, "subtype",
2915                                           subtype, strlen(subtype));
2916         if (!err && name)
2917                 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2918         if (!err)
2919                 err = parse_monolithic_mount_data(fc, data);
2920         if (!err && !mount_capable(fc))
2921                 err = -EPERM;
2922         if (!err)
2923                 err = vfs_get_tree(fc);
2924         if (!err)
2925                 err = do_new_mount_fc(fc, path, mnt_flags);
2926 
2927         put_fs_context(fc);
2928         return err;
2929 }
2930 
2931 int finish_automount(struct vfsmount *m, struct path *path)
2932 {
2933         struct dentry *dentry = path->dentry;
2934         struct mountpoint *mp;
2935         struct mount *mnt;
2936         int err;
2937 
2938         if (!m)
2939                 return 0;
2940         if (IS_ERR(m))
2941                 return PTR_ERR(m);
2942 
2943         mnt = real_mount(m);
2944         /* The new mount record should have at least 2 refs to prevent it being
2945          * expired before we get a chance to add it
2946          */
2947         BUG_ON(mnt_get_count(mnt) < 2);
2948 
2949         if (m->mnt_sb == path->mnt->mnt_sb &&
2950             m->mnt_root == dentry) {
2951                 err = -ELOOP;
2952                 goto discard;
2953         }
2954 
2955         /*
2956          * we don't want to use lock_mount() - in this case finding something
2957          * that overmounts our mountpoint to be means "quitely drop what we've
2958          * got", not "try to mount it on top".
2959          */
2960         inode_lock(dentry->d_inode);
2961         namespace_lock();
2962         if (unlikely(cant_mount(dentry))) {
2963                 err = -ENOENT;
2964                 goto discard_locked;
2965         }
2966         rcu_read_lock();
2967         if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2968                 rcu_read_unlock();
2969                 err = 0;
2970                 goto discard_locked;
2971         }
2972         rcu_read_unlock();
2973         mp = get_mountpoint(dentry);
2974         if (IS_ERR(mp)) {
2975                 err = PTR_ERR(mp);
2976                 goto discard_locked;
2977         }
2978 
2979         err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2980         unlock_mount(mp);
2981         if (unlikely(err))
2982                 goto discard;
2983         mntput(m);
2984         return 0;
2985 
2986 discard_locked:
2987         namespace_unlock();
2988         inode_unlock(dentry->d_inode);
2989 discard:
2990         /* remove m from any expiration list it may be on */
2991         if (!list_empty(&mnt->mnt_expire)) {
2992                 namespace_lock();
2993                 list_del_init(&mnt->mnt_expire);
2994                 namespace_unlock();
2995         }
2996         mntput(m);
2997         mntput(m);
2998         return err;
2999 }
3000 
3001 /**
3002  * mnt_set_expiry - Put a mount on an expiration list
3003  * @mnt: The mount to list.
3004  * @expiry_list: The list to add the mount to.
3005  */
3006 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3007 {
3008         namespace_lock();
3009 
3010         list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3011 
3012         namespace_unlock();
3013 }
3014 EXPORT_SYMBOL(mnt_set_expiry);
3015 
3016 /*
3017  * process a list of expirable mountpoints with the intent of discarding any
3018  * mountpoints that aren't in use and haven't been touched since last we came
3019  * here
3020  */
3021 void mark_mounts_for_expiry(struct list_head *mounts)
3022 {
3023         struct mount *mnt, *next;
3024         LIST_HEAD(graveyard);
3025 
3026         if (list_empty(mounts))
3027                 return;
3028 
3029         namespace_lock();
3030         lock_mount_hash();
3031 
3032         /* extract from the expiration list every vfsmount that matches the
3033          * following criteria:
3034          * - only referenced by its parent vfsmount
3035          * - still marked for expiry (marked on the last call here; marks are
3036          *   cleared by mntput())
3037          */
3038         list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3039                 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3040                         propagate_mount_busy(mnt, 1))
3041                         continue;
3042                 list_move(&mnt->mnt_expire, &graveyard);
3043         }
3044         while (!list_empty(&graveyard)) {
3045                 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3046                 touch_mnt_namespace(mnt->mnt_ns);
3047                 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3048         }
3049         unlock_mount_hash();
3050         namespace_unlock();
3051 }
3052 
3053 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3054 
3055 /*
3056  * Ripoff of 'select_parent()'
3057  *
3058  * search the list of submounts for a given mountpoint, and move any
3059  * shrinkable submounts to the 'graveyard' list.
3060  */
3061 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3062 {
3063         struct mount *this_parent = parent;
3064         struct list_head *next;
3065         int found = 0;
3066 
3067 repeat:
3068         next = this_parent->mnt_mounts.next;
3069 resume:
3070         while (next != &this_parent->mnt_mounts) {
3071                 struct list_head *tmp = next;
3072                 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3073 
3074                 next = tmp->next;
3075                 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3076                         continue;
3077                 /*
3078                  * Descend a level if the d_mounts list is non-empty.
3079                  */
3080                 if (!list_empty(&mnt->mnt_mounts)) {
3081                         this_parent = mnt;
3082                         goto repeat;
3083                 }
3084 
3085                 if (!propagate_mount_busy(mnt, 1)) {
3086                         list_move_tail(&mnt->mnt_expire, graveyard);
3087                         found++;
3088                 }
3089         }
3090         /*
3091          * All done at this level ... ascend and resume the search
3092          */
3093         if (this_parent != parent) {
3094                 next = this_parent->mnt_child.next;
3095                 this_parent = this_parent->mnt_parent;
3096                 goto resume;
3097         }
3098         return found;
3099 }
3100 
3101 /*
3102  * process a list of expirable mountpoints with the intent of discarding any
3103  * submounts of a specific parent mountpoint
3104  *
3105  * mount_lock must be held for write
3106  */
3107 static void shrink_submounts(struct mount *mnt)
3108 {
3109         LIST_HEAD(graveyard);
3110         struct mount *m;
3111 
3112         /* extract submounts of 'mountpoint' from the expiration list */
3113         while (select_submounts(mnt, &graveyard)) {
3114                 while (!list_empty(&graveyard)) {
3115                         m = list_first_entry(&graveyard, struct mount,
3116                                                 mnt_expire);
3117                         touch_mnt_namespace(m->mnt_ns);
3118                         umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3119                 }
3120         }
3121 }
3122 
3123 static void *copy_mount_options(const void __user * data)
3124 {
3125         char *copy;
3126         unsigned left, offset;
3127 
3128         if (!data)
3129                 return NULL;
3130 
3131         copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3132         if (!copy)
3133                 return ERR_PTR(-ENOMEM);
3134 
3135         left = copy_from_user(copy, data, PAGE_SIZE);
3136 
3137         /*
3138          * Not all architectures have an exact copy_from_user(). Resort to
3139          * byte at a time.
3140          */
3141         offset = PAGE_SIZE - left;
3142         while (left) {
3143                 char c;
3144                 if (get_user(c, (const char __user *)data + offset))
3145                         break;
3146                 copy[offset] = c;
3147                 left--;
3148                 offset++;
3149         }
3150 
3151         if (left == PAGE_SIZE) {
3152                 kfree(copy);
3153                 return ERR_PTR(-EFAULT);
3154         }
3155 
3156         return copy;
3157 }
3158 
3159 static char *copy_mount_string(const void __user *data)
3160 {
3161         return data ? strndup_user(data, PATH_MAX) : NULL;
3162 }
3163 
3164 /*
3165  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3166  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3167  *
3168  * data is a (void *) that can point to any structure up to
3169  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3170  * information (or be NULL).
3171  *
3172  * Pre-0.97 versions of mount() didn't have a flags word.
3173  * When the flags word was introduced its top half was required
3174  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3175  * Therefore, if this magic number is present, it carries no information
3176  * and must be discarded.
3177  */
3178 int path_mount(const char *dev_name, struct path *path,
3179                 const char *type_page, unsigned long flags, void *data_page)
3180 {
3181         unsigned int mnt_flags = 0, sb_flags;
3182         int ret;
3183 
3184         /* Discard magic */
3185         if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3186                 flags &= ~MS_MGC_MSK;
3187 
3188         /* Basic sanity checks */
3189         if (data_page)
3190                 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3191 
3192         if (flags & MS_NOUSER)
3193                 return -EINVAL;
3194 
3195         ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3196         if (ret)
3197                 return ret;
3198         if (!may_mount())
3199                 return -EPERM;
3200         if ((flags & SB_MANDLOCK) && !may_mandlock())
3201                 return -EPERM;
3202 
3203         /* Default to relatime unless overriden */
3204         if (!(flags & MS_NOATIME))
3205                 mnt_flags |= MNT_RELATIME;
3206 
3207         /* Separate the per-mountpoint flags */
3208         if (flags & MS_NOSUID)
3209                 mnt_flags |= MNT_NOSUID;
3210         if (flags & MS_NODEV)
3211                 mnt_flags |= MNT_NODEV;
3212         if (flags & MS_NOEXEC)
3213                 mnt_flags |= MNT_NOEXEC;
3214         if (flags & MS_NOATIME)
3215                 mnt_flags |= MNT_NOATIME;
3216         if (flags & MS_NODIRATIME)
3217                 mnt_flags |= MNT_NODIRATIME;
3218         if (flags & MS_STRICTATIME)
3219                 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3220         if (flags & MS_RDONLY)
3221                 mnt_flags |= MNT_READONLY;
3222         if (flags & MS_NOSYMFOLLOW)
3223                 mnt_flags |= MNT_NOSYMFOLLOW;
3224 
3225         /* The default atime for remount is preservation */
3226         if ((flags & MS_REMOUNT) &&
3227             ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3228                        MS_STRICTATIME)) == 0)) {
3229                 mnt_flags &= ~MNT_ATIME_MASK;
3230                 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3231         }
3232 
3233         sb_flags = flags & (SB_RDONLY |
3234                             SB_SYNCHRONOUS |
3235                             SB_MANDLOCK |
3236                             SB_DIRSYNC |
3237                             SB_SILENT |
3238                             SB_POSIXACL |
3239                             SB_LAZYTIME |
3240                             SB_I_VERSION);
3241 
3242         if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3243                 return do_reconfigure_mnt(path, mnt_flags);
3244         if (flags & MS_REMOUNT)
3245                 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3246         if (flags & MS_BIND)
3247                 return do_loopback(path, dev_name, flags & MS_REC);
3248         if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3249                 return do_change_type(path, flags);
3250         if (flags & MS_MOVE)
3251                 return do_move_mount_old(path, dev_name);
3252 
3253         return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3254                             data_page);
3255 }
3256 
3257 long do_mount(const char *dev_name, const char __user *dir_name,
3258                 const char *type_page, unsigned long flags, void *data_page)
3259 {
3260         struct path path;
3261         int ret;
3262 
3263         ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3264         if (ret)
3265                 return ret;
3266         ret = path_mount(dev_name, &path, type_page, flags, data_page);
3267         path_put(&path);
3268         return ret;
3269 }
3270 
3271 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3272 {
3273         return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3274 }
3275 
3276 static void dec_mnt_namespaces(struct ucounts *ucounts)
3277 {
3278         dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3279 }
3280 
3281 static void free_mnt_ns(struct mnt_namespace *ns)
3282 {
3283         if (!is_anon_ns(ns))
3284                 ns_free_inum(&ns->ns);
3285         dec_mnt_namespaces(ns->ucounts);
3286         put_user_ns(ns->user_ns);
3287         kfree(ns);
3288 }
3289 
3290 /*
3291  * Assign a sequence number so we can detect when we attempt to bind
3292  * mount a reference to an older mount namespace into the current
3293  * mount namespace, preventing reference counting loops.  A 64bit
3294  * number incrementing at 10Ghz will take 12,427 years to wrap which
3295  * is effectively never, so we can ignore the possibility.
3296  */
3297 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3298 
3299 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3300 {
3301         struct mnt_namespace *new_ns;
3302         struct ucounts *ucounts;
3303         int ret;
3304 
3305         ucounts = inc_mnt_namespaces(user_ns);
3306         if (!ucounts)
3307                 return ERR_PTR(-ENOSPC);
3308 
3309         new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3310         if (!new_ns) {
3311                 dec_mnt_namespaces(ucounts);
3312                 return ERR_PTR(-ENOMEM);
3313         }
3314         if (!anon) {
3315                 ret = ns_alloc_inum(&new_ns->ns);
3316                 if (ret) {
3317                         kfree(new_ns);
3318                         dec_mnt_namespaces(ucounts);
3319                         return ERR_PTR(ret);
3320                 }
3321         }
3322         new_ns->ns.ops = &mntns_operations;
3323         if (!anon)
3324                 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3325         refcount_set(&new_ns->ns.count, 1);
3326         INIT_LIST_HEAD(&new_ns->list);
3327         init_waitqueue_head(&new_ns->poll);
3328         spin_lock_init(&new_ns->ns_lock);
3329         new_ns->user_ns = get_user_ns(user_ns);
3330         new_ns->ucounts = ucounts;
3331         return new_ns;
3332 }
3333 
3334 __latent_entropy
3335 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3336                 struct user_namespace *user_ns, struct fs_struct *new_fs)
3337 {
3338         struct mnt_namespace *new_ns;
3339         struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3340         struct mount *p, *q;
3341         struct mount *old;
3342         struct mount *new;
3343         int copy_flags;
3344 
3345         BUG_ON(!ns);
3346 
3347         if (likely(!(flags & CLONE_NEWNS))) {
3348                 get_mnt_ns(ns);
3349                 return ns;
3350         }
3351 
3352         old = ns->root;
3353 
3354         new_ns = alloc_mnt_ns(user_ns, false);
3355         if (IS_ERR(new_ns))
3356                 return new_ns;
3357 
3358         namespace_lock();
3359         /* First pass: copy the tree topology */
3360         copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3361         if (user_ns != ns->user_ns)
3362                 copy_flags |= CL_SHARED_TO_SLAVE;
3363         new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3364         if (IS_ERR(new)) {
3365                 namespace_unlock();
3366                 free_mnt_ns(new_ns);
3367                 return ERR_CAST(new);
3368         }
3369         if (user_ns != ns->user_ns) {
3370                 lock_mount_hash();
3371                 lock_mnt_tree(new);
3372                 unlock_mount_hash();
3373         }
3374         new_ns->root = new;
3375         list_add_tail(&new_ns->list, &new->mnt_list);
3376 
3377         /*
3378          * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3379          * as belonging to new namespace.  We have already acquired a private
3380          * fs_struct, so tsk->fs->lock is not needed.
3381          */
3382         p = old;
3383         q = new;
3384         while (p) {
3385                 q->mnt_ns = new_ns;
3386                 new_ns->mounts++;
3387                 if (new_fs) {
3388                         if (&p->mnt == new_fs->root.mnt) {
3389                                 new_fs->root.mnt = mntget(&q->mnt);
3390                                 rootmnt = &p->mnt;
3391                         }
3392                         if (&p->mnt == new_fs->pwd.mnt) {
3393                                 new_fs->pwd.mnt = mntget(&q->mnt);
3394                                 pwdmnt = &p->mnt;
3395                         }
3396                 }
3397                 p = next_mnt(p, old);
3398                 q = next_mnt(q, new);
3399                 if (!q)
3400                         break;
3401                 while (p->mnt.mnt_root != q->mnt.mnt_root)
3402                         p = next_mnt(p, old);
3403         }
3404         namespace_unlock();
3405 
3406         if (rootmnt)
3407                 mntput(rootmnt);
3408         if (pwdmnt)
3409                 mntput(pwdmnt);
3410 
3411         return new_ns;
3412 }
3413 
3414 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3415 {
3416         struct mount *mnt = real_mount(m);
3417         struct mnt_namespace *ns;
3418         struct super_block *s;
3419         struct path path;
3420         int err;
3421 
3422         ns = alloc_mnt_ns(&init_user_ns, true);
3423         if (IS_ERR(ns)) {
3424                 mntput(m);
3425                 return ERR_CAST(ns);
3426         }
3427         mnt->mnt_ns = ns;
3428         ns->root = mnt;
3429         ns->mounts++;
3430         list_add(&mnt->mnt_list, &ns->list);
3431 
3432         err = vfs_path_lookup(m->mnt_root, m,
3433                         name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3434 
3435         put_mnt_ns(ns);
3436 
3437         if (err)
3438                 return ERR_PTR(err);
3439 
3440         /* trade a vfsmount reference for active sb one */
3441         s = path.mnt->mnt_sb;
3442         atomic_inc(&s->s_active);
3443         mntput(path.mnt);
3444         /* lock the sucker */
3445         down_write(&s->s_umount);
3446         /* ... and return the root of (sub)tree on it */
3447         return path.dentry;
3448 }
3449 EXPORT_SYMBOL(mount_subtree);
3450 
3451 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3452                 char __user *, type, unsigned long, flags, void __user *, data)
3453 {
3454         int ret;
3455         char *kernel_type;
3456         char *kernel_dev;
3457         void *options;
3458 
3459         kernel_type = copy_mount_string(type);
3460         ret = PTR_ERR(kernel_type);
3461         if (IS_ERR(kernel_type))
3462                 goto out_type;
3463 
3464         kernel_dev = copy_mount_string(dev_name);
3465         ret = PTR_ERR(kernel_dev);
3466         if (IS_ERR(kernel_dev))
3467                 goto out_dev;
3468 
3469         options = copy_mount_options(data);
3470         ret = PTR_ERR(options);
3471         if (IS_ERR(options))
3472                 goto out_data;
3473 
3474         ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3475 
3476         kfree(options);
3477 out_data:
3478         kfree(kernel_dev);
3479 out_dev:
3480         kfree(kernel_type);
3481 out_type:
3482         return ret;
3483 }
3484 
3485 #define FSMOUNT_VALID_FLAGS                                                    \
3486         (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV |            \
3487          MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME |       \
3488          MOUNT_ATTR_NOSYMFOLLOW)
3489 
3490 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3491 
3492 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3493         (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3494 
3495 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3496 {
3497         unsigned int mnt_flags = 0;
3498 
3499         if (attr_flags & MOUNT_ATTR_RDONLY)
3500                 mnt_flags |= MNT_READONLY;
3501         if (attr_flags & MOUNT_ATTR_NOSUID)
3502                 mnt_flags |= MNT_NOSUID;
3503         if (attr_flags & MOUNT_ATTR_NODEV)
3504                 mnt_flags |= MNT_NODEV;
3505         if (attr_flags & MOUNT_ATTR_NOEXEC)
3506                 mnt_flags |= MNT_NOEXEC;
3507         if (attr_flags & MOUNT_ATTR_NODIRATIME)
3508                 mnt_flags |= MNT_NODIRATIME;
3509         if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3510                 mnt_flags |= MNT_NOSYMFOLLOW;
3511 
3512         return mnt_flags;
3513 }
3514 
3515 /*
3516  * Create a kernel mount representation for a new, prepared superblock
3517  * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3518  */
3519 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3520                 unsigned int, attr_flags)
3521 {
3522         struct mnt_namespace *ns;
3523         struct fs_context *fc;
3524         struct file *file;
3525         struct path newmount;
3526         struct mount *mnt;
3527         struct fd f;
3528         unsigned int mnt_flags = 0;
3529         long ret;
3530 
3531         if (!may_mount())
3532                 return -EPERM;
3533 
3534         if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3535                 return -EINVAL;
3536 
3537         if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3538                 return -EINVAL;
3539 
3540         mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3541 
3542         switch (attr_flags & MOUNT_ATTR__ATIME) {
3543         case MOUNT_ATTR_STRICTATIME:
3544                 break;
3545         case MOUNT_ATTR_NOATIME:
3546                 mnt_flags |= MNT_NOATIME;
3547                 break;
3548         case MOUNT_ATTR_RELATIME:
3549                 mnt_flags |= MNT_RELATIME;
3550                 break;
3551         default:
3552                 return -EINVAL;
3553         }
3554 
3555         f = fdget(fs_fd);
3556         if (!f.file)
3557                 return -EBADF;
3558 
3559         ret = -EINVAL;
3560         if (f.file->f_op != &fscontext_fops)
3561                 goto err_fsfd;
3562 
3563         fc = f.file->private_data;
3564 
3565         ret = mutex_lock_interruptible(&fc->uapi_mutex);
3566         if (ret < 0)
3567                 goto err_fsfd;
3568 
3569         /* There must be a valid superblock or we can't mount it */
3570         ret = -EINVAL;
3571         if (!fc->root)
3572                 goto err_unlock;
3573 
3574         ret = -EPERM;
3575         if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3576                 pr_warn("VFS: Mount too revealing\n");
3577                 goto err_unlock;
3578         }
3579 
3580         ret = -EBUSY;
3581         if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3582                 goto err_unlock;
3583 
3584         ret = -EPERM;
3585         if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3586                 goto err_unlock;
3587 
3588         newmount.mnt = vfs_create_mount(fc);
3589         if (IS_ERR(newmount.mnt)) {
3590                 ret = PTR_ERR(newmount.mnt);
3591                 goto err_unlock;
3592         }
3593         newmount.dentry = dget(fc->root);
3594         newmount.mnt->mnt_flags = mnt_flags;
3595 
3596         /* We've done the mount bit - now move the file context into more or
3597          * less the same state as if we'd done an fspick().  We don't want to
3598          * do any memory allocation or anything like that at this point as we
3599          * don't want to have to handle any errors incurred.
3600          */
3601         vfs_clean_context(fc);
3602 
3603         ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3604         if (IS_ERR(ns)) {
3605                 ret = PTR_ERR(ns);
3606                 goto err_path;
3607         }
3608         mnt = real_mount(newmount.mnt);
3609         mnt->mnt_ns = ns;
3610         ns->root = mnt;
3611         ns->mounts = 1;
3612         list_add(&mnt->mnt_list, &ns->list);
3613         mntget(newmount.mnt);
3614 
3615         /* Attach to an apparent O_PATH fd with a note that we need to unmount
3616          * it, not just simply put it.
3617          */
3618         file = dentry_open(&newmount, O_PATH, fc->cred);
3619         if (IS_ERR(file)) {
3620                 dissolve_on_fput(newmount.mnt);
3621                 ret = PTR_ERR(file);
3622                 goto err_path;
3623         }
3624         file->f_mode |= FMODE_NEED_UNMOUNT;
3625 
3626         ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3627         if (ret >= 0)
3628                 fd_install(ret, file);
3629         else
3630                 fput(file);
3631 
3632 err_path:
3633         path_put(&newmount);
3634 err_unlock:
3635         mutex_unlock(&fc->uapi_mutex);
3636 err_fsfd:
3637         fdput(f);
3638         return ret;
3639 }
3640 
3641 /*
3642  * Move a mount from one place to another.  In combination with
3643  * fsopen()/fsmount() this is used to install a new mount and in combination
3644  * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3645  * a mount subtree.
3646  *
3647  * Note the flags value is a combination of MOVE_MOUNT_* flags.
3648  */
3649 SYSCALL_DEFINE5(move_mount,
3650                 int, from_dfd, const char __user *, from_pathname,
3651                 int, to_dfd, const char __user *, to_pathname,
3652                 unsigned int, flags)
3653 {
3654         struct path from_path, to_path;
3655         unsigned int lflags;
3656         int ret = 0;
3657 
3658         if (!may_mount())
3659                 return -EPERM;
3660 
3661         if (flags & ~MOVE_MOUNT__MASK)
3662                 return -EINVAL;
3663 
3664         /* If someone gives a pathname, they aren't permitted to move
3665          * from an fd that requires unmount as we can't get at the flag
3666          * to clear it afterwards.
3667          */
3668         lflags = 0;
3669         if (flags & MOVE_MOUNT_F_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3670         if (flags & MOVE_MOUNT_F_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3671         if (flags & MOVE_MOUNT_F_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3672 
3673         ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3674         if (ret < 0)
3675                 return ret;
3676 
3677         lflags = 0;
3678         if (flags & MOVE_MOUNT_T_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3679         if (flags & MOVE_MOUNT_T_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3680         if (flags & MOVE_MOUNT_T_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3681 
3682         ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3683         if (ret < 0)
3684                 goto out_from;
3685 
3686         ret = security_move_mount(&from_path, &to_path);
3687         if (ret < 0)
3688                 goto out_to;
3689 
3690         ret = do_move_mount(&from_path, &to_path);
3691 
3692 out_to:
3693         path_put(&to_path);
3694 out_from:
3695         path_put(&from_path);
3696         return ret;
3697 }
3698 
3699 /*
3700  * Return true if path is reachable from root
3701  *
3702  * namespace_sem or mount_lock is held
3703  */
3704 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3705                          const struct path *root)
3706 {
3707         while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3708                 dentry = mnt->mnt_mountpoint;
3709                 mnt = mnt->mnt_parent;
3710         }
3711         return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3712 }
3713 
3714 bool path_is_under(const struct path *path1, const struct path *path2)
3715 {
3716         bool res;
3717         read_seqlock_excl(&mount_lock);
3718         res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3719         read_sequnlock_excl(&mount_lock);
3720         return res;
3721 }
3722 EXPORT_SYMBOL(path_is_under);
3723 
3724 /*
3725  * pivot_root Semantics:
3726  * Moves the root file system of the current process to the directory put_old,
3727  * makes new_root as the new root file system of the current process, and sets
3728  * root/cwd of all processes which had them on the current root to new_root.
3729  *
3730  * Restrictions:
3731  * The new_root and put_old must be directories, and  must not be on the
3732  * same file  system as the current process root. The put_old  must  be
3733  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3734  * pointed to by put_old must yield the same directory as new_root. No other
3735  * file system may be mounted on put_old. After all, new_root is a mountpoint.
3736  *
3737  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3738  * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3739  * in this situation.
3740  *
3741  * Notes:
3742  *  - we don't move root/cwd if they are not at the root (reason: if something
3743  *    cared enough to change them, it's probably wrong to force them elsewhere)
3744  *  - it's okay to pick a root that isn't the root of a file system, e.g.
3745  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3746  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3747  *    first.
3748  */
3749 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3750                 const char __user *, put_old)
3751 {
3752         struct path new, old, root;
3753         struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3754         struct mountpoint *old_mp, *root_mp;
3755         int error;
3756 
3757         if (!may_mount())
3758                 return -EPERM;
3759 
3760         error = user_path_at(AT_FDCWD, new_root,
3761                              LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3762         if (error)
3763                 goto out0;
3764 
3765         error = user_path_at(AT_FDCWD, put_old,
3766                              LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3767         if (error)
3768                 goto out1;
3769 
3770         error = security_sb_pivotroot(&old, &new);
3771         if (error)
3772                 goto out2;
3773 
3774         get_fs_root(current->fs, &root);
3775         old_mp = lock_mount(&old);
3776         error = PTR_ERR(old_mp);
3777         if (IS_ERR(old_mp))
3778                 goto out3;
3779 
3780         error = -EINVAL;
3781         new_mnt = real_mount(new.mnt);
3782         root_mnt = real_mount(root.mnt);
3783         old_mnt = real_mount(old.mnt);
3784         ex_parent = new_mnt->mnt_parent;
3785         root_parent = root_mnt->mnt_parent;
3786         if (IS_MNT_SHARED(old_mnt) ||
3787                 IS_MNT_SHARED(ex_parent) ||
3788                 IS_MNT_SHARED(root_parent))
3789                 goto out4;
3790         if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3791                 goto out4;
3792         if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3793                 goto out4;
3794         error = -ENOENT;
3795         if (d_unlinked(new.dentry))
3796                 goto out4;
3797         error = -EBUSY;
3798         if (new_mnt == root_mnt || old_mnt == root_mnt)
3799                 goto out4; /* loop, on the same file system  */
3800         error = -EINVAL;
3801         if (root.mnt->mnt_root != root.dentry)
3802                 goto out4; /* not a mountpoint */
3803         if (!mnt_has_parent(root_mnt))
3804                 goto out4; /* not attached */
3805         if (new.mnt->mnt_root != new.dentry)
3806                 goto out4; /* not a mountpoint */
3807         if (!mnt_has_parent(new_mnt))
3808                 goto out4; /* not attached */
3809         /* make sure we can reach put_old from new_root */
3810         if (!is_path_reachable(old_mnt, old.dentry, &new))
3811                 goto out4;
3812         /* make certain new is below the root */
3813         if (!is_path_reachable(new_mnt, new.dentry, &root))
3814                 goto out4;
3815         lock_mount_hash();
3816         umount_mnt(new_mnt);
3817         root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
3818         if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3819                 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3820                 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3821         }
3822         /* mount old root on put_old */
3823         attach_mnt(root_mnt, old_mnt, old_mp);
3824         /* mount new_root on / */
3825         attach_mnt(new_mnt, root_parent, root_mp);
3826         mnt_add_count(root_parent, -1);
3827         touch_mnt_namespace(current->nsproxy->mnt_ns);
3828         /* A moved mount should not expire automatically */
3829         list_del_init(&new_mnt->mnt_expire);
3830         put_mountpoint(root_mp);
3831         unlock_mount_hash();
3832         chroot_fs_refs(&root, &new);
3833         error = 0;
3834 out4:
3835         unlock_mount(old_mp);
3836         if (!error)
3837                 mntput_no_expire(ex_parent);
3838 out3:
3839         path_put(&root);
3840 out2:
3841         path_put(&old);
3842 out1:
3843         path_put(&new);
3844 out0:
3845         return error;
3846 }
3847 
3848 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3849 {
3850         unsigned int flags = mnt->mnt.mnt_flags;
3851 
3852         /*  flags to clear */
3853         flags &= ~kattr->attr_clr;
3854         /* flags to raise */
3855         flags |= kattr->attr_set;
3856 
3857         return flags;
3858 }
3859 
3860 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3861 {
3862         struct vfsmount *m = &mnt->mnt;
3863 
3864         if (!kattr->mnt_userns)
3865                 return 0;
3866 
3867         /*
3868          * Once a mount has been idmapped we don't allow it to change its
3869          * mapping. It makes things simpler and callers can just create
3870          * another bind-mount they can idmap if they want to.
3871          */
3872         if (mnt_user_ns(m) != &init_user_ns)
3873                 return -EPERM;
3874 
3875         /* The underlying filesystem doesn't support idmapped mounts yet. */
3876         if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3877                 return -EINVAL;
3878 
3879         /* Don't yet support filesystem mountable in user namespaces. */
3880         if (m->mnt_sb->s_user_ns != &init_user_ns)
3881                 return -EINVAL;
3882 
3883         /* We're not controlling the superblock. */
3884         if (!capable(CAP_SYS_ADMIN))
3885                 return -EPERM;
3886 
3887         /* Mount has already been visible in the filesystem hierarchy. */
3888         if (!is_anon_ns(mnt->mnt_ns))
3889                 return -EINVAL;
3890 
3891         return 0;
3892 }
3893 
3894 static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3895                                            struct mount *mnt, int *err)
3896 {
3897         struct mount *m = mnt, *last = NULL;
3898 
3899         if (!is_mounted(&m->mnt)) {
3900                 *err = -EINVAL;
3901                 goto out;
3902         }
3903 
3904         if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3905                 *err = -EINVAL;
3906                 goto out;
3907         }
3908 
3909         do {
3910                 unsigned int flags;
3911 
3912                 flags = recalc_flags(kattr, m);
3913                 if (!can_change_locked_flags(m, flags)) {
3914                         *err = -EPERM;
3915                         goto out;
3916                 }
3917 
3918                 *err = can_idmap_mount(kattr, m);
3919                 if (*err)
3920                         goto out;
3921 
3922                 last = m;
3923 
3924                 if ((kattr->attr_set & MNT_READONLY) &&
3925                     !(m->mnt.mnt_flags & MNT_READONLY)) {
3926                         *err = mnt_hold_writers(m);
3927                         if (*err)
3928                                 goto out;
3929                 }
3930         } while (kattr->recurse && (m = next_mnt(m, mnt)));
3931 
3932 out:
3933         return last;
3934 }
3935 
3936 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3937 {
3938         struct user_namespace *mnt_userns;
3939 
3940         if (!kattr->mnt_userns)
3941                 return;
3942 
3943         mnt_userns = get_user_ns(kattr->mnt_userns);
3944         /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3945         smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3946 }
3947 
3948 static void mount_setattr_commit(struct mount_kattr *kattr,
3949                                  struct mount *mnt, struct mount *last,
3950                                  int err)
3951 {
3952         struct mount *m = mnt;
3953 
3954         do {
3955                 if (!err) {
3956                         unsigned int flags;
3957 
3958                         do_idmap_mount(kattr, m);
3959                         flags = recalc_flags(kattr, m);
3960                         WRITE_ONCE(m->mnt.mnt_flags, flags);
3961                 }
3962 
3963                 /*
3964                  * We either set MNT_READONLY above so make it visible
3965                  * before ~MNT_WRITE_HOLD or we failed to recursively
3966                  * apply mount options.
3967                  */
3968                 if ((kattr->attr_set & MNT_READONLY) &&
3969                     (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3970                         mnt_unhold_writers(m);
3971 
3972                 if (!err && kattr->propagation)
3973                         change_mnt_propagation(m, kattr->propagation);
3974 
3975                 /*
3976                  * On failure, only cleanup until we found the first mount
3977                  * we failed to handle.
3978                  */
3979                 if (err && m == last)
3980                         break;
3981         } while (kattr->recurse && (m = next_mnt(m, mnt)));
3982 
3983         if (!err)
3984                 touch_mnt_namespace(mnt->mnt_ns);
3985 }
3986 
3987 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3988 {
3989         struct mount *mnt = real_mount(path->mnt), *last = NULL;
3990         int err = 0;
3991 
3992         if (path->dentry != mnt->mnt.mnt_root)
3993                 return -EINVAL;
3994 
3995         if (kattr->propagation) {
3996                 /*
3997                  * Only take namespace_lock() if we're actually changing
3998                  * propagation.
3999                  */
4000                 namespace_lock();
4001                 if (kattr->propagation == MS_SHARED) {
4002                         err = invent_group_ids(mnt, kattr->recurse);
4003                         if (err) {
4004                                 namespace_unlock();
4005                                 return err;
4006                         }
4007                 }
4008         }
4009 
4010         lock_mount_hash();
4011 
4012         /*
4013          * Get the mount tree in a shape where we can change mount
4014          * properties without failure.
4015          */
4016         last = mount_setattr_prepare(kattr, mnt, &err);
4017         if (last) /* Commit all changes or revert to the old state. */
4018                 mount_setattr_commit(kattr, mnt, last, err);
4019 
4020         unlock_mount_hash();
4021 
4022         if (kattr->propagation) {
4023                 namespace_unlock();
4024                 if (err)
4025                         cleanup_group_ids(mnt, NULL);
4026         }
4027 
4028         return err;
4029 }
4030 
4031 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4032                                 struct mount_kattr *kattr, unsigned int flags)
4033 {
4034         int err = 0;
4035         struct ns_common *ns;
4036         struct user_namespace *mnt_userns;
4037         struct file *file;
4038 
4039         if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4040                 return 0;
4041 
4042         /*
4043          * We currently do not support clearing an idmapped mount. If this ever
4044          * is a use-case we can revisit this but for now let's keep it simple
4045          * and not allow it.
4046          */
4047         if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4048                 return -EINVAL;
4049 
4050         if (attr->userns_fd > INT_MAX)
4051                 return -EINVAL;
4052 
4053         file = fget(attr->userns_fd);
4054         if (!file)
4055                 return -EBADF;
4056 
4057         if (!proc_ns_file(file)) {
4058                 err = -EINVAL;
4059                 goto out_fput;
4060         }
4061 
4062         ns = get_proc_ns(file_inode(file));
4063         if (ns->ops->type != CLONE_NEWUSER) {
4064                 err = -EINVAL;
4065                 goto out_fput;
4066         }
4067 
4068         /*
4069          * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4070          * This is simpler than just having to treat NULL as unmapped. Users
4071          * wanting to idmap a mount to init_user_ns can just use a namespace
4072          * with an identity mapping.
4073          */
4074         mnt_userns = container_of(ns, struct user_namespace, ns);
4075         if (mnt_userns == &init_user_ns) {
4076                 err = -EPERM;
4077                 goto out_fput;
4078         }
4079         kattr->mnt_userns = get_user_ns(mnt_userns);
4080 
4081 out_fput:
4082         fput(file);
4083         return err;
4084 }
4085 
4086 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4087                              struct mount_kattr *kattr, unsigned int flags)
4088 {
4089         unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4090 
4091         if (flags & AT_NO_AUTOMOUNT)
4092                 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4093         if (flags & AT_SYMLINK_NOFOLLOW)
4094                 lookup_flags &= ~LOOKUP_FOLLOW;
4095         if (flags & AT_EMPTY_PATH)
4096                 lookup_flags |= LOOKUP_EMPTY;
4097 
4098         *kattr = (struct mount_kattr) {
4099                 .lookup_flags   = lookup_flags,
4100                 .recurse        = !!(flags & AT_RECURSIVE),
4101         };
4102 
4103         if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4104                 return -EINVAL;
4105         if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4106                 return -EINVAL;
4107         kattr->propagation = attr->propagation;
4108 
4109         if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4110                 return -EINVAL;
4111 
4112         kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4113         kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4114 
4115         /*
4116          * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4117          * users wanting to transition to a different atime setting cannot
4118          * simply specify the atime setting in @attr_set, but must also
4119          * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4120          * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4121          * @attr_clr and that @attr_set can't have any atime bits set if
4122          * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4123          */
4124         if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4125                 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4126                         return -EINVAL;
4127 
4128                 /*
4129                  * Clear all previous time settings as they are mutually
4130                  * exclusive.
4131                  */
4132                 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4133                 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4134                 case MOUNT_ATTR_RELATIME:
4135                         kattr->attr_set |= MNT_RELATIME;
4136                         break;
4137                 case MOUNT_ATTR_NOATIME:
4138                         kattr->attr_set |= MNT_NOATIME;
4139                         break;
4140                 case MOUNT_ATTR_STRICTATIME:
4141                         break;
4142                 default:
4143                         return -EINVAL;
4144                 }
4145         } else {
4146                 if (attr->attr_set & MOUNT_ATTR__ATIME)
4147                         return -EINVAL;
4148         }
4149 
4150         return build_mount_idmapped(attr, usize, kattr, flags);
4151 }
4152 
4153 static void finish_mount_kattr(struct mount_kattr *kattr)
4154 {
4155         put_user_ns(kattr->mnt_userns);
4156         kattr->mnt_userns = NULL;
4157 }
4158 
4159 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4160                 unsigned int, flags, struct mount_attr __user *, uattr,
4161                 size_t, usize)
4162 {
4163         int err;
4164         struct path target;
4165         struct mount_attr attr;
4166         struct mount_kattr kattr;
4167 
4168         BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4169 
4170         if (flags & ~(AT_EMPTY_PATH |
4171                       AT_RECURSIVE |
4172                       AT_SYMLINK_NOFOLLOW |
4173                       AT_NO_AUTOMOUNT))
4174                 return -EINVAL;
4175 
4176         if (unlikely(usize > PAGE_SIZE))
4177                 return -E2BIG;
4178         if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4179                 return -EINVAL;
4180 
4181         if (!may_mount())
4182                 return -EPERM;
4183 
4184         err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4185         if (err)
4186                 return err;
4187 
4188         /* Don't bother walking through the mounts if this is a nop. */
4189         if (attr.attr_set == 0 &&
4190             attr.attr_clr == 0 &&
4191             attr.propagation == 0)
4192                 return 0;
4193 
4194         err = build_mount_kattr(&attr, usize, &kattr, flags);
4195         if (err)
4196                 return err;
4197 
4198         err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4199         if (err)
4200                 return err;
4201 
4202         err = do_mount_setattr(&target, &kattr);
4203         finish_mount_kattr(&kattr);
4204         path_put(&target);
4205         return err;
4206 }
4207 
4208 static void __init init_mount_tree(void)
4209 {
4210         struct vfsmount *mnt;
4211         struct mount *m;
4212         struct mnt_namespace *ns;
4213         struct path root;
4214 
4215         mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4216         if (IS_ERR(mnt))
4217                 panic("Can't create rootfs");
4218 
4219         ns = alloc_mnt_ns(&init_user_ns, false);
4220         if (IS_ERR(ns))
4221                 panic("Can't allocate initial namespace");
4222         m = real_mount(mnt);
4223         m->mnt_ns = ns;
4224         ns->root = m;
4225         ns->mounts = 1;
4226         list_add(&m->mnt_list, &ns->list);
4227         init_task.nsproxy->mnt_ns = ns;
4228         get_mnt_ns(ns);
4229 
4230         root.mnt = mnt;
4231         root.dentry = mnt->mnt_root;
4232         mnt->mnt_flags |= MNT_LOCKED;
4233 
4234         set_fs_pwd(current->fs, &root);
4235         set_fs_root(current->fs, &root);
4236 }
4237 
4238 void __init mnt_init(void)
4239 {
4240         int err;
4241 
4242         mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4243                         0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4244 
4245         mount_hashtable = alloc_large_system_hash("Mount-cache",
4246                                 sizeof(struct hlist_head),
4247                                 mhash_entries, 19,
4248                                 HASH_ZERO,
4249                                 &m_hash_shift, &m_hash_mask, 0, 0);
4250         mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4251                                 sizeof(struct hlist_head),
4252                                 mphash_entries, 19,
4253                                 HASH_ZERO,
4254                                 &mp_hash_shift, &mp_hash_mask, 0, 0);
4255 
4256         if (!mount_hashtable || !mountpoint_hashtable)
4257                 panic("Failed to allocate mount hash table\n");
4258 
4259         kernfs_init();
4260 
4261         err = sysfs_init();
4262         if (err)
4263                 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4264                         __func__, err);
4265         fs_kobj = kobject_create_and_add("fs", NULL);
4266         if (!fs_kobj)
4267                 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4268         shmem_init();
4269         init_rootfs();
4270         init_mount_tree();
4271 }
4272 
4273 void put_mnt_ns(struct mnt_namespace *ns)
4274 {
4275         if (!refcount_dec_and_test(&ns->ns.count))
4276                 return;
4277         drop_collected_mounts(&ns->root->mnt);
4278         free_mnt_ns(ns);
4279 }
4280 
4281 struct vfsmount *kern_mount(struct file_system_type *type)
4282 {
4283         struct vfsmount *mnt;
4284         mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4285         if (!IS_ERR(mnt)) {
4286                 /*
4287                  * it is a longterm mount, don't release mnt until
4288                  * we unmount before file sys is unregistered
4289                 */
4290                 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4291         }
4292         return mnt;
4293 }
4294 EXPORT_SYMBOL_GPL(kern_mount);
4295 
4296 void kern_unmount(struct vfsmount *mnt)
4297 {
4298         /* release long term mount so mount point can be released */
4299         if (!IS_ERR_OR_NULL(mnt)) {
4300                 real_mount(mnt)->mnt_ns = NULL;
4301                 synchronize_rcu();      /* yecchhh... */
4302                 mntput(mnt);
4303         }
4304 }
4305 EXPORT_SYMBOL(kern_unmount);
4306 
4307 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4308 {
4309         unsigned int i;
4310 
4311         for (i = 0; i < num; i++)
4312                 if (mnt[i])
4313                         real_mount(mnt[i])->mnt_ns = NULL;
4314         synchronize_rcu_expedited();
4315         for (i = 0; i < num; i++)
4316                 mntput(mnt[i]);
4317 }
4318 EXPORT_SYMBOL(kern_unmount_array);
4319 
4320 bool our_mnt(struct vfsmount *mnt)
4321 {
4322         return check_mnt(real_mount(mnt));
4323 }
4324 
4325 bool current_chrooted(void)
4326 {
4327         /* Does the current process have a non-standard root */
4328         struct path ns_root;
4329         struct path fs_root;
4330         bool chrooted;
4331 
4332         /* Find the namespace root */
4333         ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4334         ns_root.dentry = ns_root.mnt->mnt_root;
4335         path_get(&ns_root);
4336         while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4337                 ;
4338 
4339         get_fs_root(current->fs, &fs_root);
4340 
4341         chrooted = !path_equal(&fs_root, &ns_root);
4342 
4343         path_put(&fs_root);
4344         path_put(&ns_root);
4345 
4346         return chrooted;
4347 }
4348 
4349 static bool mnt_already_visible(struct mnt_namespace *ns,
4350                                 const struct super_block *sb,
4351                                 int *new_mnt_flags)
4352 {
4353         int new_flags = *new_mnt_flags;
4354         struct mount *mnt;
4355         bool visible = false;
4356 
4357         down_read(&namespace_sem);
4358         lock_ns_list(ns);
4359         list_for_each_entry(mnt, &ns->list, mnt_list) {
4360                 struct mount *child;
4361                 int mnt_flags;
4362 
4363                 if (mnt_is_cursor(mnt))
4364                         continue;
4365 
4366                 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4367                         continue;
4368 
4369                 /* This mount is not fully visible if it's root directory
4370                  * is not the root directory of the filesystem.
4371                  */
4372                 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4373                         continue;
4374 
4375                 /* A local view of the mount flags */
4376                 mnt_flags = mnt->mnt.mnt_flags;
4377 
4378                 /* Don't miss readonly hidden in the superblock flags */
4379                 if (sb_rdonly(mnt->mnt.mnt_sb))
4380                         mnt_flags |= MNT_LOCK_READONLY;
4381 
4382                 /* Verify the mount flags are equal to or more permissive
4383                  * than the proposed new mount.
4384                  */
4385                 if ((mnt_flags & MNT_LOCK_READONLY) &&
4386                     !(new_flags & MNT_READONLY))
4387                         continue;
4388                 if ((mnt_flags & MNT_LOCK_ATIME) &&
4389                     ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4390                         continue;
4391 
4392                 /* This mount is not fully visible if there are any
4393                  * locked child mounts that cover anything except for
4394                  * empty directories.
4395                  */
4396                 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4397                         struct inode *inode = child->mnt_mountpoint->d_inode;
4398                         /* Only worry about locked mounts */
4399                         if (!(child->mnt.mnt_flags & MNT_LOCKED))
4400                                 continue;
4401                         /* Is the directory permanetly empty? */
4402                         if (!is_empty_dir_inode(inode))
4403                                 goto next;
4404                 }
4405                 /* Preserve the locked attributes */
4406                 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4407                                                MNT_LOCK_ATIME);
4408                 visible = true;
4409                 goto found;
4410         next:   ;
4411         }
4412 found:
4413         unlock_ns_list(ns);
4414         up_read(&namespace_sem);
4415         return visible;
4416 }
4417 
4418 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4419 {
4420         const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4421         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4422         unsigned long s_iflags;
4423 
4424         if (ns->user_ns == &init_user_ns)
4425                 return false;
4426 
4427         /* Can this filesystem be too revealing? */
4428         s_iflags = sb->s_iflags;
4429         if (!(s_iflags & SB_I_USERNS_VISIBLE))
4430                 return false;
4431 
4432         if ((s_iflags & required_iflags) != required_iflags) {
4433                 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4434                           required_iflags);
4435                 return true;
4436         }
4437 
4438         return !mnt_already_visible(ns, sb, new_mnt_flags);
4439 }
4440 
4441 bool mnt_may_suid(struct vfsmount *mnt)
4442 {
4443         /*
4444          * Foreign mounts (accessed via fchdir or through /proc
4445          * symlinks) are always treated as if they are nosuid.  This
4446          * prevents namespaces from trusting potentially unsafe
4447          * suid/sgid bits, file caps, or security labels that originate
4448          * in other namespaces.
4449          */
4450         return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4451                current_in_userns(mnt->mnt_sb->s_user_ns);
4452 }
4453 
4454 static struct ns_common *mntns_get(struct task_struct *task)
4455 {
4456         struct ns_common *ns = NULL;
4457         struct nsproxy *nsproxy;
4458 
4459         task_lock(task);
4460         nsproxy = task->nsproxy;
4461         if (nsproxy) {
4462                 ns = &nsproxy->mnt_ns->ns;
4463                 get_mnt_ns(to_mnt_ns(ns));
4464         }
4465         task_unlock(task);
4466 
4467         return ns;
4468 }
4469 
4470 static void mntns_put(struct ns_common *ns)
4471 {
4472         put_mnt_ns(to_mnt_ns(ns));
4473 }
4474 
4475 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4476 {
4477         struct nsproxy *nsproxy = nsset->nsproxy;
4478         struct fs_struct *fs = nsset->fs;
4479         struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4480         struct user_namespace *user_ns = nsset->cred->user_ns;
4481         struct path root;
4482         int err;
4483 
4484         if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4485             !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4486             !ns_capable(user_ns, CAP_SYS_ADMIN))
4487                 return -EPERM;
4488 
4489         if (is_anon_ns(mnt_ns))
4490                 return -EINVAL;
4491 
4492         if (fs->users != 1)
4493                 return -EINVAL;
4494 
4495         get_mnt_ns(mnt_ns);
4496         old_mnt_ns = nsproxy->mnt_ns;
4497         nsproxy->mnt_ns = mnt_ns;
4498 
4499         /* Find the root */
4500         err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4501                                 "/", LOOKUP_DOWN, &root);
4502         if (err) {
4503                 /* revert to old namespace */
4504                 nsproxy->mnt_ns = old_mnt_ns;
4505                 put_mnt_ns(mnt_ns);
4506                 return err;
4507         }
4508 
4509         put_mnt_ns(old_mnt_ns);
4510 
4511         /* Update the pwd and root */
4512         set_fs_pwd(fs, &root);
4513         set_fs_root(fs, &root);
4514 
4515         path_put(&root);
4516         return 0;
4517 }
4518 
4519 static struct user_namespace *mntns_owner(struct ns_common *ns)
4520 {
4521         return to_mnt_ns(ns)->user_ns;
4522 }
4523 
4524 const struct proc_ns_operations mntns_operations = {
4525         .name           = "mnt",
4526         .type           = CLONE_NEWNS,
4527         .get            = mntns_get,
4528         .put            = mntns_put,
4529         .install        = mntns_install,
4530         .owner          = mntns_owner,
4531 };
4532 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp