1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11#include <linux/syscalls.h>
12#include <linux/export.h>
13#include <linux/capability.h>
14#include <linux/mnt_namespace.h>
15#include <linux/user_namespace.h>
16#include <linux/namei.h>
17#include <linux/security.h>
18#include <linux/cred.h>
19#include <linux/idr.h>
20#include <linux/init.h> /* init_rootfs */
21#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23#include <linux/file.h>
24#include <linux/uaccess.h>
25#include <linux/proc_ns.h>
26#include <linux/magic.h>
27#include <linux/memblock.h>
28#include <linux/proc_fs.h>
29#include <linux/task_work.h>
30#include <linux/sched/task.h>
31#include <uapi/linux/mount.h>
32#include <linux/fs_context.h>
33#include <linux/shmem_fs.h>
34#include <linux/mnt_idmapping.h>
35
36#include "pnode.h"
37#include "internal.h"
38
39/* Maximum number of mounts in a mount namespace */
40static unsigned int sysctl_mount_max __read_mostly = 100000;
41
42static unsigned int m_hash_mask __ro_after_init;
43static unsigned int m_hash_shift __ro_after_init;
44static unsigned int mp_hash_mask __ro_after_init;
45static unsigned int mp_hash_shift __ro_after_init;
46
47static __initdata unsigned long mhash_entries;
48static int __init set_mhash_entries(char *str)
49{
50 if (!str)
51 return 0;
52 mhash_entries = simple_strtoul(str, &str, 0);
53 return 1;
54}
55__setup("mhash_entries=", set_mhash_entries);
56
57static __initdata unsigned long mphash_entries;
58static int __init set_mphash_entries(char *str)
59{
60 if (!str)
61 return 0;
62 mphash_entries = simple_strtoul(str, &str, 0);
63 return 1;
64}
65__setup("mphash_entries=", set_mphash_entries);
66
67static u64 event;
68static DEFINE_IDA(mnt_id_ida);
69static DEFINE_IDA(mnt_group_ida);
70
71static struct hlist_head *mount_hashtable __ro_after_init;
72static struct hlist_head *mountpoint_hashtable __ro_after_init;
73static struct kmem_cache *mnt_cache __ro_after_init;
74static DECLARE_RWSEM(namespace_sem);
75static HLIST_HEAD(unmounted); /* protected by namespace_sem */
76static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
77
78struct mount_kattr {
79 unsigned int attr_set;
80 unsigned int attr_clr;
81 unsigned int propagation;
82 unsigned int lookup_flags;
83 bool recurse;
84 struct user_namespace *mnt_userns;
85 struct mnt_idmap *mnt_idmap;
86};
87
88/* /sys/fs */
89struct kobject *fs_kobj __ro_after_init;
90EXPORT_SYMBOL_GPL(fs_kobj);
91
92/*
93 * vfsmount lock may be taken for read to prevent changes to the
94 * vfsmount hash, ie. during mountpoint lookups or walking back
95 * up the tree.
96 *
97 * It should be taken for write in all cases where the vfsmount
98 * tree or hash is modified or when a vfsmount structure is modified.
99 */
100__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
101
102static inline void lock_mount_hash(void)
103{
104 write_seqlock(sl: &mount_lock);
105}
106
107static inline void unlock_mount_hash(void)
108{
109 write_sequnlock(sl: &mount_lock);
110}
111
112static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
113{
114 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
115 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
116 tmp = tmp + (tmp >> m_hash_shift);
117 return &mount_hashtable[tmp & m_hash_mask];
118}
119
120static inline struct hlist_head *mp_hash(struct dentry *dentry)
121{
122 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
123 tmp = tmp + (tmp >> mp_hash_shift);
124 return &mountpoint_hashtable[tmp & mp_hash_mask];
125}
126
127static int mnt_alloc_id(struct mount *mnt)
128{
129 int res = ida_alloc(ida: &mnt_id_ida, GFP_KERNEL);
130
131 if (res < 0)
132 return res;
133 mnt->mnt_id = res;
134 return 0;
135}
136
137static void mnt_free_id(struct mount *mnt)
138{
139 ida_free(&mnt_id_ida, id: mnt->mnt_id);
140}
141
142/*
143 * Allocate a new peer group ID
144 */
145static int mnt_alloc_group_id(struct mount *mnt)
146{
147 int res = ida_alloc_min(ida: &mnt_group_ida, min: 1, GFP_KERNEL);
148
149 if (res < 0)
150 return res;
151 mnt->mnt_group_id = res;
152 return 0;
153}
154
155/*
156 * Release a peer group ID
157 */
158void mnt_release_group_id(struct mount *mnt)
159{
160 ida_free(&mnt_group_ida, id: mnt->mnt_group_id);
161 mnt->mnt_group_id = 0;
162}
163
164/*
165 * vfsmount lock must be held for read
166 */
167static inline void mnt_add_count(struct mount *mnt, int n)
168{
169#ifdef CONFIG_SMP
170 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
171#else
172 preempt_disable();
173 mnt->mnt_count += n;
174 preempt_enable();
175#endif
176}
177
178/*
179 * vfsmount lock must be held for write
180 */
181int mnt_get_count(struct mount *mnt)
182{
183#ifdef CONFIG_SMP
184 int count = 0;
185 int cpu;
186
187 for_each_possible_cpu(cpu) {
188 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
189 }
190
191 return count;
192#else
193 return mnt->mnt_count;
194#endif
195}
196
197static struct mount *alloc_vfsmnt(const char *name)
198{
199 struct mount *mnt = kmem_cache_zalloc(k: mnt_cache, GFP_KERNEL);
200 if (mnt) {
201 int err;
202
203 err = mnt_alloc_id(mnt);
204 if (err)
205 goto out_free_cache;
206
207 if (name) {
208 mnt->mnt_devname = kstrdup_const(s: name,
209 GFP_KERNEL_ACCOUNT);
210 if (!mnt->mnt_devname)
211 goto out_free_id;
212 }
213
214#ifdef CONFIG_SMP
215 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
216 if (!mnt->mnt_pcp)
217 goto out_free_devname;
218
219 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
220#else
221 mnt->mnt_count = 1;
222 mnt->mnt_writers = 0;
223#endif
224
225 INIT_HLIST_NODE(h: &mnt->mnt_hash);
226 INIT_LIST_HEAD(list: &mnt->mnt_child);
227 INIT_LIST_HEAD(list: &mnt->mnt_mounts);
228 INIT_LIST_HEAD(list: &mnt->mnt_list);
229 INIT_LIST_HEAD(list: &mnt->mnt_expire);
230 INIT_LIST_HEAD(list: &mnt->mnt_share);
231 INIT_LIST_HEAD(list: &mnt->mnt_slave_list);
232 INIT_LIST_HEAD(list: &mnt->mnt_slave);
233 INIT_HLIST_NODE(h: &mnt->mnt_mp_list);
234 INIT_LIST_HEAD(list: &mnt->mnt_umounting);
235 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
236 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
237 }
238 return mnt;
239
240#ifdef CONFIG_SMP
241out_free_devname:
242 kfree_const(x: mnt->mnt_devname);
243#endif
244out_free_id:
245 mnt_free_id(mnt);
246out_free_cache:
247 kmem_cache_free(s: mnt_cache, objp: mnt);
248 return NULL;
249}
250
251/*
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
257 * a filesystem.
258 */
259/*
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
262 *
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
268 * r/w.
269 */
270bool __mnt_is_readonly(struct vfsmount *mnt)
271{
272 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(sb: mnt->mnt_sb);
273}
274EXPORT_SYMBOL_GPL(__mnt_is_readonly);
275
276static inline void mnt_inc_writers(struct mount *mnt)
277{
278#ifdef CONFIG_SMP
279 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
280#else
281 mnt->mnt_writers++;
282#endif
283}
284
285static inline void mnt_dec_writers(struct mount *mnt)
286{
287#ifdef CONFIG_SMP
288 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
289#else
290 mnt->mnt_writers--;
291#endif
292}
293
294static unsigned int mnt_get_writers(struct mount *mnt)
295{
296#ifdef CONFIG_SMP
297 unsigned int count = 0;
298 int cpu;
299
300 for_each_possible_cpu(cpu) {
301 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
302 }
303
304 return count;
305#else
306 return mnt->mnt_writers;
307#endif
308}
309
310static int mnt_is_readonly(struct vfsmount *mnt)
311{
312 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
313 return 1;
314 /*
315 * The barrier pairs with the barrier in sb_start_ro_state_change()
316 * making sure if we don't see s_readonly_remount set yet, we also will
317 * not see any superblock / mount flag changes done by remount.
318 * It also pairs with the barrier in sb_end_ro_state_change()
319 * assuring that if we see s_readonly_remount already cleared, we will
320 * see the values of superblock / mount flags updated by remount.
321 */
322 smp_rmb();
323 return __mnt_is_readonly(mnt);
324}
325
326/*
327 * Most r/o & frozen checks on a fs are for operations that take discrete
328 * amounts of time, like a write() or unlink(). We must keep track of when
329 * those operations start (for permission checks) and when they end, so that we
330 * can determine when writes are able to occur to a filesystem.
331 */
332/**
333 * mnt_get_write_access - get write access to a mount without freeze protection
334 * @m: the mount on which to take a write
335 *
336 * This tells the low-level filesystem that a write is about to be performed to
337 * it, and makes sure that writes are allowed (mnt it read-write) before
338 * returning success. This operation does not protect against filesystem being
339 * frozen. When the write operation is finished, mnt_put_write_access() must be
340 * called. This is effectively a refcount.
341 */
342int mnt_get_write_access(struct vfsmount *m)
343{
344 struct mount *mnt = real_mount(mnt: m);
345 int ret = 0;
346
347 preempt_disable();
348 mnt_inc_writers(mnt);
349 /*
350 * The store to mnt_inc_writers must be visible before we pass
351 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 * incremented count after it has set MNT_WRITE_HOLD.
353 */
354 smp_mb();
355 might_lock(&mount_lock.lock);
356 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
357 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
358 cpu_relax();
359 } else {
360 /*
361 * This prevents priority inversion, if the task
362 * setting MNT_WRITE_HOLD got preempted on a remote
363 * CPU, and it prevents life lock if the task setting
364 * MNT_WRITE_HOLD has a lower priority and is bound to
365 * the same CPU as the task that is spinning here.
366 */
367 preempt_enable();
368 lock_mount_hash();
369 unlock_mount_hash();
370 preempt_disable();
371 }
372 }
373 /*
374 * The barrier pairs with the barrier sb_start_ro_state_change() making
375 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
376 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
377 * mnt_is_readonly() and bail in case we are racing with remount
378 * read-only.
379 */
380 smp_rmb();
381 if (mnt_is_readonly(mnt: m)) {
382 mnt_dec_writers(mnt);
383 ret = -EROFS;
384 }
385 preempt_enable();
386
387 return ret;
388}
389EXPORT_SYMBOL_GPL(mnt_get_write_access);
390
391/**
392 * mnt_want_write - get write access to a mount
393 * @m: the mount on which to take a write
394 *
395 * This tells the low-level filesystem that a write is about to be performed to
396 * it, and makes sure that writes are allowed (mount is read-write, filesystem
397 * is not frozen) before returning success. When the write operation is
398 * finished, mnt_drop_write() must be called. This is effectively a refcount.
399 */
400int mnt_want_write(struct vfsmount *m)
401{
402 int ret;
403
404 sb_start_write(sb: m->mnt_sb);
405 ret = mnt_get_write_access(m);
406 if (ret)
407 sb_end_write(sb: m->mnt_sb);
408 return ret;
409}
410EXPORT_SYMBOL_GPL(mnt_want_write);
411
412/**
413 * mnt_get_write_access_file - get write access to a file's mount
414 * @file: the file who's mount on which to take a write
415 *
416 * This is like mnt_get_write_access, but if @file is already open for write it
417 * skips incrementing mnt_writers (since the open file already has a reference)
418 * and instead only does the check for emergency r/o remounts. This must be
419 * paired with mnt_put_write_access_file.
420 */
421int mnt_get_write_access_file(struct file *file)
422{
423 if (file->f_mode & FMODE_WRITER) {
424 /*
425 * Superblock may have become readonly while there are still
426 * writable fd's, e.g. due to a fs error with errors=remount-ro
427 */
428 if (__mnt_is_readonly(file->f_path.mnt))
429 return -EROFS;
430 return 0;
431 }
432 return mnt_get_write_access(file->f_path.mnt);
433}
434
435/**
436 * mnt_want_write_file - get write access to a file's mount
437 * @file: the file who's mount on which to take a write
438 *
439 * This is like mnt_want_write, but if the file is already open for writing it
440 * skips incrementing mnt_writers (since the open file already has a reference)
441 * and instead only does the freeze protection and the check for emergency r/o
442 * remounts. This must be paired with mnt_drop_write_file.
443 */
444int mnt_want_write_file(struct file *file)
445{
446 int ret;
447
448 sb_start_write(sb: file_inode(f: file)->i_sb);
449 ret = mnt_get_write_access_file(file);
450 if (ret)
451 sb_end_write(sb: file_inode(f: file)->i_sb);
452 return ret;
453}
454EXPORT_SYMBOL_GPL(mnt_want_write_file);
455
456/**
457 * mnt_put_write_access - give up write access to a mount
458 * @mnt: the mount on which to give up write access
459 *
460 * Tells the low-level filesystem that we are done
461 * performing writes to it. Must be matched with
462 * mnt_get_write_access() call above.
463 */
464void mnt_put_write_access(struct vfsmount *mnt)
465{
466 preempt_disable();
467 mnt_dec_writers(mnt: real_mount(mnt));
468 preempt_enable();
469}
470EXPORT_SYMBOL_GPL(mnt_put_write_access);
471
472/**
473 * mnt_drop_write - give up write access to a mount
474 * @mnt: the mount on which to give up write access
475 *
476 * Tells the low-level filesystem that we are done performing writes to it and
477 * also allows filesystem to be frozen again. Must be matched with
478 * mnt_want_write() call above.
479 */
480void mnt_drop_write(struct vfsmount *mnt)
481{
482 mnt_put_write_access(mnt);
483 sb_end_write(sb: mnt->mnt_sb);
484}
485EXPORT_SYMBOL_GPL(mnt_drop_write);
486
487void mnt_put_write_access_file(struct file *file)
488{
489 if (!(file->f_mode & FMODE_WRITER))
490 mnt_put_write_access(file->f_path.mnt);
491}
492
493void mnt_drop_write_file(struct file *file)
494{
495 mnt_put_write_access_file(file);
496 sb_end_write(sb: file_inode(f: file)->i_sb);
497}
498EXPORT_SYMBOL(mnt_drop_write_file);
499
500/**
501 * mnt_hold_writers - prevent write access to the given mount
502 * @mnt: mnt to prevent write access to
503 *
504 * Prevents write access to @mnt if there are no active writers for @mnt.
505 * This function needs to be called and return successfully before changing
506 * properties of @mnt that need to remain stable for callers with write access
507 * to @mnt.
508 *
509 * After this functions has been called successfully callers must pair it with
510 * a call to mnt_unhold_writers() in order to stop preventing write access to
511 * @mnt.
512 *
513 * Context: This function expects lock_mount_hash() to be held serializing
514 * setting MNT_WRITE_HOLD.
515 * Return: On success 0 is returned.
516 * On error, -EBUSY is returned.
517 */
518static inline int mnt_hold_writers(struct mount *mnt)
519{
520 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
521 /*
522 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
523 * should be visible before we do.
524 */
525 smp_mb();
526
527 /*
528 * With writers on hold, if this value is zero, then there are
529 * definitely no active writers (although held writers may subsequently
530 * increment the count, they'll have to wait, and decrement it after
531 * seeing MNT_READONLY).
532 *
533 * It is OK to have counter incremented on one CPU and decremented on
534 * another: the sum will add up correctly. The danger would be when we
535 * sum up each counter, if we read a counter before it is incremented,
536 * but then read another CPU's count which it has been subsequently
537 * decremented from -- we would see more decrements than we should.
538 * MNT_WRITE_HOLD protects against this scenario, because
539 * mnt_want_write first increments count, then smp_mb, then spins on
540 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
541 * we're counting up here.
542 */
543 if (mnt_get_writers(mnt) > 0)
544 return -EBUSY;
545
546 return 0;
547}
548
549/**
550 * mnt_unhold_writers - stop preventing write access to the given mount
551 * @mnt: mnt to stop preventing write access to
552 *
553 * Stop preventing write access to @mnt allowing callers to gain write access
554 * to @mnt again.
555 *
556 * This function can only be called after a successful call to
557 * mnt_hold_writers().
558 *
559 * Context: This function expects lock_mount_hash() to be held.
560 */
561static inline void mnt_unhold_writers(struct mount *mnt)
562{
563 /*
564 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
565 * that become unheld will see MNT_READONLY.
566 */
567 smp_wmb();
568 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
569}
570
571static int mnt_make_readonly(struct mount *mnt)
572{
573 int ret;
574
575 ret = mnt_hold_writers(mnt);
576 if (!ret)
577 mnt->mnt.mnt_flags |= MNT_READONLY;
578 mnt_unhold_writers(mnt);
579 return ret;
580}
581
582int sb_prepare_remount_readonly(struct super_block *sb)
583{
584 struct mount *mnt;
585 int err = 0;
586
587 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
588 if (atomic_long_read(v: &sb->s_remove_count))
589 return -EBUSY;
590
591 lock_mount_hash();
592 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
593 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
594 err = mnt_hold_writers(mnt);
595 if (err)
596 break;
597 }
598 }
599 if (!err && atomic_long_read(v: &sb->s_remove_count))
600 err = -EBUSY;
601
602 if (!err)
603 sb_start_ro_state_change(sb);
604 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
605 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
606 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
607 }
608 unlock_mount_hash();
609
610 return err;
611}
612
613static void free_vfsmnt(struct mount *mnt)
614{
615 mnt_idmap_put(idmap: mnt_idmap(mnt: &mnt->mnt));
616 kfree_const(x: mnt->mnt_devname);
617#ifdef CONFIG_SMP
618 free_percpu(pdata: mnt->mnt_pcp);
619#endif
620 kmem_cache_free(s: mnt_cache, objp: mnt);
621}
622
623static void delayed_free_vfsmnt(struct rcu_head *head)
624{
625 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
626}
627
628/* call under rcu_read_lock */
629int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
630{
631 struct mount *mnt;
632 if (read_seqretry(sl: &mount_lock, start: seq))
633 return 1;
634 if (bastard == NULL)
635 return 0;
636 mnt = real_mount(mnt: bastard);
637 mnt_add_count(mnt, n: 1);
638 smp_mb(); // see mntput_no_expire()
639 if (likely(!read_seqretry(&mount_lock, seq)))
640 return 0;
641 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
642 mnt_add_count(mnt, n: -1);
643 return 1;
644 }
645 lock_mount_hash();
646 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
647 mnt_add_count(mnt, n: -1);
648 unlock_mount_hash();
649 return 1;
650 }
651 unlock_mount_hash();
652 /* caller will mntput() */
653 return -1;
654}
655
656/* call under rcu_read_lock */
657static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
658{
659 int res = __legitimize_mnt(bastard, seq);
660 if (likely(!res))
661 return true;
662 if (unlikely(res < 0)) {
663 rcu_read_unlock();
664 mntput(mnt: bastard);
665 rcu_read_lock();
666 }
667 return false;
668}
669
670/**
671 * __lookup_mnt - find first child mount
672 * @mnt: parent mount
673 * @dentry: mountpoint
674 *
675 * If @mnt has a child mount @c mounted @dentry find and return it.
676 *
677 * Note that the child mount @c need not be unique. There are cases
678 * where shadow mounts are created. For example, during mount
679 * propagation when a source mount @mnt whose root got overmounted by a
680 * mount @o after path lookup but before @namespace_sem could be
681 * acquired gets copied and propagated. So @mnt gets copied including
682 * @o. When @mnt is propagated to a destination mount @d that already
683 * has another mount @n mounted at the same mountpoint then the source
684 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
685 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
686 * on @dentry.
687 *
688 * Return: The first child of @mnt mounted @dentry or NULL.
689 */
690struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
691{
692 struct hlist_head *head = m_hash(mnt, dentry);
693 struct mount *p;
694
695 hlist_for_each_entry_rcu(p, head, mnt_hash)
696 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
697 return p;
698 return NULL;
699}
700
701/*
702 * lookup_mnt - Return the first child mount mounted at path
703 *
704 * "First" means first mounted chronologically. If you create the
705 * following mounts:
706 *
707 * mount /dev/sda1 /mnt
708 * mount /dev/sda2 /mnt
709 * mount /dev/sda3 /mnt
710 *
711 * Then lookup_mnt() on the base /mnt dentry in the root mount will
712 * return successively the root dentry and vfsmount of /dev/sda1, then
713 * /dev/sda2, then /dev/sda3, then NULL.
714 *
715 * lookup_mnt takes a reference to the found vfsmount.
716 */
717struct vfsmount *lookup_mnt(const struct path *path)
718{
719 struct mount *child_mnt;
720 struct vfsmount *m;
721 unsigned seq;
722
723 rcu_read_lock();
724 do {
725 seq = read_seqbegin(sl: &mount_lock);
726 child_mnt = __lookup_mnt(mnt: path->mnt, dentry: path->dentry);
727 m = child_mnt ? &child_mnt->mnt : NULL;
728 } while (!legitimize_mnt(bastard: m, seq));
729 rcu_read_unlock();
730 return m;
731}
732
733static inline void lock_ns_list(struct mnt_namespace *ns)
734{
735 spin_lock(lock: &ns->ns_lock);
736}
737
738static inline void unlock_ns_list(struct mnt_namespace *ns)
739{
740 spin_unlock(lock: &ns->ns_lock);
741}
742
743static inline bool mnt_is_cursor(struct mount *mnt)
744{
745 return mnt->mnt.mnt_flags & MNT_CURSOR;
746}
747
748/*
749 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
750 * current mount namespace.
751 *
752 * The common case is dentries are not mountpoints at all and that
753 * test is handled inline. For the slow case when we are actually
754 * dealing with a mountpoint of some kind, walk through all of the
755 * mounts in the current mount namespace and test to see if the dentry
756 * is a mountpoint.
757 *
758 * The mount_hashtable is not usable in the context because we
759 * need to identify all mounts that may be in the current mount
760 * namespace not just a mount that happens to have some specified
761 * parent mount.
762 */
763bool __is_local_mountpoint(struct dentry *dentry)
764{
765 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
766 struct mount *mnt;
767 bool is_covered = false;
768
769 down_read(sem: &namespace_sem);
770 lock_ns_list(ns);
771 list_for_each_entry(mnt, &ns->list, mnt_list) {
772 if (mnt_is_cursor(mnt))
773 continue;
774 is_covered = (mnt->mnt_mountpoint == dentry);
775 if (is_covered)
776 break;
777 }
778 unlock_ns_list(ns);
779 up_read(sem: &namespace_sem);
780
781 return is_covered;
782}
783
784static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
785{
786 struct hlist_head *chain = mp_hash(dentry);
787 struct mountpoint *mp;
788
789 hlist_for_each_entry(mp, chain, m_hash) {
790 if (mp->m_dentry == dentry) {
791 mp->m_count++;
792 return mp;
793 }
794 }
795 return NULL;
796}
797
798static struct mountpoint *get_mountpoint(struct dentry *dentry)
799{
800 struct mountpoint *mp, *new = NULL;
801 int ret;
802
803 if (d_mountpoint(dentry)) {
804 /* might be worth a WARN_ON() */
805 if (d_unlinked(dentry))
806 return ERR_PTR(error: -ENOENT);
807mountpoint:
808 read_seqlock_excl(sl: &mount_lock);
809 mp = lookup_mountpoint(dentry);
810 read_sequnlock_excl(sl: &mount_lock);
811 if (mp)
812 goto done;
813 }
814
815 if (!new)
816 new = kmalloc(size: sizeof(struct mountpoint), GFP_KERNEL);
817 if (!new)
818 return ERR_PTR(error: -ENOMEM);
819
820
821 /* Exactly one processes may set d_mounted */
822 ret = d_set_mounted(dentry);
823
824 /* Someone else set d_mounted? */
825 if (ret == -EBUSY)
826 goto mountpoint;
827
828 /* The dentry is not available as a mountpoint? */
829 mp = ERR_PTR(error: ret);
830 if (ret)
831 goto done;
832
833 /* Add the new mountpoint to the hash table */
834 read_seqlock_excl(sl: &mount_lock);
835 new->m_dentry = dget(dentry);
836 new->m_count = 1;
837 hlist_add_head(n: &new->m_hash, h: mp_hash(dentry));
838 INIT_HLIST_HEAD(&new->m_list);
839 read_sequnlock_excl(sl: &mount_lock);
840
841 mp = new;
842 new = NULL;
843done:
844 kfree(objp: new);
845 return mp;
846}
847
848/*
849 * vfsmount lock must be held. Additionally, the caller is responsible
850 * for serializing calls for given disposal list.
851 */
852static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
853{
854 if (!--mp->m_count) {
855 struct dentry *dentry = mp->m_dentry;
856 BUG_ON(!hlist_empty(&mp->m_list));
857 spin_lock(lock: &dentry->d_lock);
858 dentry->d_flags &= ~DCACHE_MOUNTED;
859 spin_unlock(lock: &dentry->d_lock);
860 dput_to_list(dentry, list);
861 hlist_del(n: &mp->m_hash);
862 kfree(objp: mp);
863 }
864}
865
866/* called with namespace_lock and vfsmount lock */
867static void put_mountpoint(struct mountpoint *mp)
868{
869 __put_mountpoint(mp, list: &ex_mountpoints);
870}
871
872static inline int check_mnt(struct mount *mnt)
873{
874 return mnt->mnt_ns == current->nsproxy->mnt_ns;
875}
876
877/*
878 * vfsmount lock must be held for write
879 */
880static void touch_mnt_namespace(struct mnt_namespace *ns)
881{
882 if (ns) {
883 ns->event = ++event;
884 wake_up_interruptible(&ns->poll);
885 }
886}
887
888/*
889 * vfsmount lock must be held for write
890 */
891static void __touch_mnt_namespace(struct mnt_namespace *ns)
892{
893 if (ns && ns->event != event) {
894 ns->event = event;
895 wake_up_interruptible(&ns->poll);
896 }
897}
898
899/*
900 * vfsmount lock must be held for write
901 */
902static struct mountpoint *unhash_mnt(struct mount *mnt)
903{
904 struct mountpoint *mp;
905 mnt->mnt_parent = mnt;
906 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
907 list_del_init(entry: &mnt->mnt_child);
908 hlist_del_init_rcu(n: &mnt->mnt_hash);
909 hlist_del_init(n: &mnt->mnt_mp_list);
910 mp = mnt->mnt_mp;
911 mnt->mnt_mp = NULL;
912 return mp;
913}
914
915/*
916 * vfsmount lock must be held for write
917 */
918static void umount_mnt(struct mount *mnt)
919{
920 put_mountpoint(mp: unhash_mnt(mnt));
921}
922
923/*
924 * vfsmount lock must be held for write
925 */
926void mnt_set_mountpoint(struct mount *mnt,
927 struct mountpoint *mp,
928 struct mount *child_mnt)
929{
930 mp->m_count++;
931 mnt_add_count(mnt, n: 1); /* essentially, that's mntget */
932 child_mnt->mnt_mountpoint = mp->m_dentry;
933 child_mnt->mnt_parent = mnt;
934 child_mnt->mnt_mp = mp;
935 hlist_add_head(n: &child_mnt->mnt_mp_list, h: &mp->m_list);
936}
937
938/**
939 * mnt_set_mountpoint_beneath - mount a mount beneath another one
940 *
941 * @new_parent: the source mount
942 * @top_mnt: the mount beneath which @new_parent is mounted
943 * @new_mp: the new mountpoint of @top_mnt on @new_parent
944 *
945 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
946 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
947 * @new_mp. And mount @new_parent on the old parent and old
948 * mountpoint of @top_mnt.
949 *
950 * Context: This function expects namespace_lock() and lock_mount_hash()
951 * to have been acquired in that order.
952 */
953static void mnt_set_mountpoint_beneath(struct mount *new_parent,
954 struct mount *top_mnt,
955 struct mountpoint *new_mp)
956{
957 struct mount *old_top_parent = top_mnt->mnt_parent;
958 struct mountpoint *old_top_mp = top_mnt->mnt_mp;
959
960 mnt_set_mountpoint(mnt: old_top_parent, mp: old_top_mp, child_mnt: new_parent);
961 mnt_change_mountpoint(parent: new_parent, mp: new_mp, mnt: top_mnt);
962}
963
964
965static void __attach_mnt(struct mount *mnt, struct mount *parent)
966{
967 hlist_add_head_rcu(n: &mnt->mnt_hash,
968 h: m_hash(mnt: &parent->mnt, dentry: mnt->mnt_mountpoint));
969 list_add_tail(new: &mnt->mnt_child, head: &parent->mnt_mounts);
970}
971
972/**
973 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
974 * list of child mounts
975 * @parent: the parent
976 * @mnt: the new mount
977 * @mp: the new mountpoint
978 * @beneath: whether to mount @mnt beneath or on top of @parent
979 *
980 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
981 * to @parent's child mount list and to @mount_hashtable.
982 *
983 * If @beneath is true, remove @mnt from its current parent and
984 * mountpoint and mount it on @mp on @parent, and mount @parent on the
985 * old parent and old mountpoint of @mnt. Finally, attach @parent to
986 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
987 *
988 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
989 * to the correct parent.
990 *
991 * Context: This function expects namespace_lock() and lock_mount_hash()
992 * to have been acquired in that order.
993 */
994static void attach_mnt(struct mount *mnt, struct mount *parent,
995 struct mountpoint *mp, bool beneath)
996{
997 if (beneath)
998 mnt_set_mountpoint_beneath(new_parent: mnt, top_mnt: parent, new_mp: mp);
999 else
1000 mnt_set_mountpoint(mnt: parent, mp, child_mnt: mnt);
1001 /*
1002 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
1003 * beneath @parent then @mnt will need to be attached to
1004 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
1005 * isn't the same mount as @parent.
1006 */
1007 __attach_mnt(mnt, parent: mnt->mnt_parent);
1008}
1009
1010void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
1011{
1012 struct mountpoint *old_mp = mnt->mnt_mp;
1013 struct mount *old_parent = mnt->mnt_parent;
1014
1015 list_del_init(entry: &mnt->mnt_child);
1016 hlist_del_init(n: &mnt->mnt_mp_list);
1017 hlist_del_init_rcu(n: &mnt->mnt_hash);
1018
1019 attach_mnt(mnt, parent, mp, beneath: false);
1020
1021 put_mountpoint(mp: old_mp);
1022 mnt_add_count(mnt: old_parent, n: -1);
1023}
1024
1025/*
1026 * vfsmount lock must be held for write
1027 */
1028static void commit_tree(struct mount *mnt)
1029{
1030 struct mount *parent = mnt->mnt_parent;
1031 struct mount *m;
1032 LIST_HEAD(head);
1033 struct mnt_namespace *n = parent->mnt_ns;
1034
1035 BUG_ON(parent == mnt);
1036
1037 list_add_tail(new: &head, head: &mnt->mnt_list);
1038 list_for_each_entry(m, &head, mnt_list)
1039 m->mnt_ns = n;
1040
1041 list_splice(list: &head, head: n->list.prev);
1042
1043 n->mounts += n->pending_mounts;
1044 n->pending_mounts = 0;
1045
1046 __attach_mnt(mnt, parent);
1047 touch_mnt_namespace(ns: n);
1048}
1049
1050static struct mount *next_mnt(struct mount *p, struct mount *root)
1051{
1052 struct list_head *next = p->mnt_mounts.next;
1053 if (next == &p->mnt_mounts) {
1054 while (1) {
1055 if (p == root)
1056 return NULL;
1057 next = p->mnt_child.next;
1058 if (next != &p->mnt_parent->mnt_mounts)
1059 break;
1060 p = p->mnt_parent;
1061 }
1062 }
1063 return list_entry(next, struct mount, mnt_child);
1064}
1065
1066static struct mount *skip_mnt_tree(struct mount *p)
1067{
1068 struct list_head *prev = p->mnt_mounts.prev;
1069 while (prev != &p->mnt_mounts) {
1070 p = list_entry(prev, struct mount, mnt_child);
1071 prev = p->mnt_mounts.prev;
1072 }
1073 return p;
1074}
1075
1076/**
1077 * vfs_create_mount - Create a mount for a configured superblock
1078 * @fc: The configuration context with the superblock attached
1079 *
1080 * Create a mount to an already configured superblock. If necessary, the
1081 * caller should invoke vfs_get_tree() before calling this.
1082 *
1083 * Note that this does not attach the mount to anything.
1084 */
1085struct vfsmount *vfs_create_mount(struct fs_context *fc)
1086{
1087 struct mount *mnt;
1088
1089 if (!fc->root)
1090 return ERR_PTR(error: -EINVAL);
1091
1092 mnt = alloc_vfsmnt(name: fc->source ?: "none");
1093 if (!mnt)
1094 return ERR_PTR(error: -ENOMEM);
1095
1096 if (fc->sb_flags & SB_KERNMOUNT)
1097 mnt->mnt.mnt_flags = MNT_INTERNAL;
1098
1099 atomic_inc(v: &fc->root->d_sb->s_active);
1100 mnt->mnt.mnt_sb = fc->root->d_sb;
1101 mnt->mnt.mnt_root = dget(dentry: fc->root);
1102 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1103 mnt->mnt_parent = mnt;
1104
1105 lock_mount_hash();
1106 list_add_tail(new: &mnt->mnt_instance, head: &mnt->mnt.mnt_sb->s_mounts);
1107 unlock_mount_hash();
1108 return &mnt->mnt;
1109}
1110EXPORT_SYMBOL(vfs_create_mount);
1111
1112struct vfsmount *fc_mount(struct fs_context *fc)
1113{
1114 int err = vfs_get_tree(fc);
1115 if (!err) {
1116 up_write(sem: &fc->root->d_sb->s_umount);
1117 return vfs_create_mount(fc);
1118 }
1119 return ERR_PTR(error: err);
1120}
1121EXPORT_SYMBOL(fc_mount);
1122
1123struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1124 int flags, const char *name,
1125 void *data)
1126{
1127 struct fs_context *fc;
1128 struct vfsmount *mnt;
1129 int ret = 0;
1130
1131 if (!type)
1132 return ERR_PTR(error: -EINVAL);
1133
1134 fc = fs_context_for_mount(fs_type: type, sb_flags: flags);
1135 if (IS_ERR(ptr: fc))
1136 return ERR_CAST(ptr: fc);
1137
1138 if (name)
1139 ret = vfs_parse_fs_string(fc, key: "source",
1140 value: name, strlen(name));
1141 if (!ret)
1142 ret = parse_monolithic_mount_data(fc, data);
1143 if (!ret)
1144 mnt = fc_mount(fc);
1145 else
1146 mnt = ERR_PTR(error: ret);
1147
1148 put_fs_context(fc);
1149 return mnt;
1150}
1151EXPORT_SYMBOL_GPL(vfs_kern_mount);
1152
1153struct vfsmount *
1154vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1155 const char *name, void *data)
1156{
1157 /* Until it is worked out how to pass the user namespace
1158 * through from the parent mount to the submount don't support
1159 * unprivileged mounts with submounts.
1160 */
1161 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1162 return ERR_PTR(error: -EPERM);
1163
1164 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1165}
1166EXPORT_SYMBOL_GPL(vfs_submount);
1167
1168static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1169 int flag)
1170{
1171 struct super_block *sb = old->mnt.mnt_sb;
1172 struct mount *mnt;
1173 int err;
1174
1175 mnt = alloc_vfsmnt(name: old->mnt_devname);
1176 if (!mnt)
1177 return ERR_PTR(error: -ENOMEM);
1178
1179 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1180 mnt->mnt_group_id = 0; /* not a peer of original */
1181 else
1182 mnt->mnt_group_id = old->mnt_group_id;
1183
1184 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1185 err = mnt_alloc_group_id(mnt);
1186 if (err)
1187 goto out_free;
1188 }
1189
1190 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1191 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1192
1193 atomic_inc(v: &sb->s_active);
1194 mnt->mnt.mnt_idmap = mnt_idmap_get(idmap: mnt_idmap(mnt: &old->mnt));
1195
1196 mnt->mnt.mnt_sb = sb;
1197 mnt->mnt.mnt_root = dget(dentry: root);
1198 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1199 mnt->mnt_parent = mnt;
1200 lock_mount_hash();
1201 list_add_tail(new: &mnt->mnt_instance, head: &sb->s_mounts);
1202 unlock_mount_hash();
1203
1204 if ((flag & CL_SLAVE) ||
1205 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1206 list_add(new: &mnt->mnt_slave, head: &old->mnt_slave_list);
1207 mnt->mnt_master = old;
1208 CLEAR_MNT_SHARED(mnt);
1209 } else if (!(flag & CL_PRIVATE)) {
1210 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1211 list_add(new: &mnt->mnt_share, head: &old->mnt_share);
1212 if (IS_MNT_SLAVE(old))
1213 list_add(new: &mnt->mnt_slave, head: &old->mnt_slave);
1214 mnt->mnt_master = old->mnt_master;
1215 } else {
1216 CLEAR_MNT_SHARED(mnt);
1217 }
1218 if (flag & CL_MAKE_SHARED)
1219 set_mnt_shared(mnt);
1220
1221 /* stick the duplicate mount on the same expiry list
1222 * as the original if that was on one */
1223 if (flag & CL_EXPIRE) {
1224 if (!list_empty(head: &old->mnt_expire))
1225 list_add(new: &mnt->mnt_expire, head: &old->mnt_expire);
1226 }
1227
1228 return mnt;
1229
1230 out_free:
1231 mnt_free_id(mnt);
1232 free_vfsmnt(mnt);
1233 return ERR_PTR(error: err);
1234}
1235
1236static void cleanup_mnt(struct mount *mnt)
1237{
1238 struct hlist_node *p;
1239 struct mount *m;
1240 /*
1241 * The warning here probably indicates that somebody messed
1242 * up a mnt_want/drop_write() pair. If this happens, the
1243 * filesystem was probably unable to make r/w->r/o transitions.
1244 * The locking used to deal with mnt_count decrement provides barriers,
1245 * so mnt_get_writers() below is safe.
1246 */
1247 WARN_ON(mnt_get_writers(mnt));
1248 if (unlikely(mnt->mnt_pins.first))
1249 mnt_pin_kill(m: mnt);
1250 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1251 hlist_del(n: &m->mnt_umount);
1252 mntput(mnt: &m->mnt);
1253 }
1254 fsnotify_vfsmount_delete(mnt: &mnt->mnt);
1255 dput(mnt->mnt.mnt_root);
1256 deactivate_super(sb: mnt->mnt.mnt_sb);
1257 mnt_free_id(mnt);
1258 call_rcu(head: &mnt->mnt_rcu, func: delayed_free_vfsmnt);
1259}
1260
1261static void __cleanup_mnt(struct rcu_head *head)
1262{
1263 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1264}
1265
1266static LLIST_HEAD(delayed_mntput_list);
1267static void delayed_mntput(struct work_struct *unused)
1268{
1269 struct llist_node *node = llist_del_all(head: &delayed_mntput_list);
1270 struct mount *m, *t;
1271
1272 llist_for_each_entry_safe(m, t, node, mnt_llist)
1273 cleanup_mnt(mnt: m);
1274}
1275static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1276
1277static void mntput_no_expire(struct mount *mnt)
1278{
1279 LIST_HEAD(list);
1280 int count;
1281
1282 rcu_read_lock();
1283 if (likely(READ_ONCE(mnt->mnt_ns))) {
1284 /*
1285 * Since we don't do lock_mount_hash() here,
1286 * ->mnt_ns can change under us. However, if it's
1287 * non-NULL, then there's a reference that won't
1288 * be dropped until after an RCU delay done after
1289 * turning ->mnt_ns NULL. So if we observe it
1290 * non-NULL under rcu_read_lock(), the reference
1291 * we are dropping is not the final one.
1292 */
1293 mnt_add_count(mnt, n: -1);
1294 rcu_read_unlock();
1295 return;
1296 }
1297 lock_mount_hash();
1298 /*
1299 * make sure that if __legitimize_mnt() has not seen us grab
1300 * mount_lock, we'll see their refcount increment here.
1301 */
1302 smp_mb();
1303 mnt_add_count(mnt, n: -1);
1304 count = mnt_get_count(mnt);
1305 if (count != 0) {
1306 WARN_ON(count < 0);
1307 rcu_read_unlock();
1308 unlock_mount_hash();
1309 return;
1310 }
1311 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1312 rcu_read_unlock();
1313 unlock_mount_hash();
1314 return;
1315 }
1316 mnt->mnt.mnt_flags |= MNT_DOOMED;
1317 rcu_read_unlock();
1318
1319 list_del(entry: &mnt->mnt_instance);
1320
1321 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1322 struct mount *p, *tmp;
1323 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1324 __put_mountpoint(mp: unhash_mnt(mnt: p), list: &list);
1325 hlist_add_head(n: &p->mnt_umount, h: &mnt->mnt_stuck_children);
1326 }
1327 }
1328 unlock_mount_hash();
1329 shrink_dentry_list(&list);
1330
1331 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1332 struct task_struct *task = current;
1333 if (likely(!(task->flags & PF_KTHREAD))) {
1334 init_task_work(twork: &mnt->mnt_rcu, func: __cleanup_mnt);
1335 if (!task_work_add(task, twork: &mnt->mnt_rcu, mode: TWA_RESUME))
1336 return;
1337 }
1338 if (llist_add(new: &mnt->mnt_llist, head: &delayed_mntput_list))
1339 schedule_delayed_work(dwork: &delayed_mntput_work, delay: 1);
1340 return;
1341 }
1342 cleanup_mnt(mnt);
1343}
1344
1345void mntput(struct vfsmount *mnt)
1346{
1347 if (mnt) {
1348 struct mount *m = real_mount(mnt);
1349 /* avoid cacheline pingpong */
1350 if (unlikely(m->mnt_expiry_mark))
1351 WRITE_ONCE(m->mnt_expiry_mark, 0);
1352 mntput_no_expire(mnt: m);
1353 }
1354}
1355EXPORT_SYMBOL(mntput);
1356
1357struct vfsmount *mntget(struct vfsmount *mnt)
1358{
1359 if (mnt)
1360 mnt_add_count(mnt: real_mount(mnt), n: 1);
1361 return mnt;
1362}
1363EXPORT_SYMBOL(mntget);
1364
1365/*
1366 * Make a mount point inaccessible to new lookups.
1367 * Because there may still be current users, the caller MUST WAIT
1368 * for an RCU grace period before destroying the mount point.
1369 */
1370void mnt_make_shortterm(struct vfsmount *mnt)
1371{
1372 if (mnt)
1373 real_mount(mnt)->mnt_ns = NULL;
1374}
1375
1376/**
1377 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1378 * @path: path to check
1379 *
1380 * d_mountpoint() can only be used reliably to establish if a dentry is
1381 * not mounted in any namespace and that common case is handled inline.
1382 * d_mountpoint() isn't aware of the possibility there may be multiple
1383 * mounts using a given dentry in a different namespace. This function
1384 * checks if the passed in path is a mountpoint rather than the dentry
1385 * alone.
1386 */
1387bool path_is_mountpoint(const struct path *path)
1388{
1389 unsigned seq;
1390 bool res;
1391
1392 if (!d_mountpoint(dentry: path->dentry))
1393 return false;
1394
1395 rcu_read_lock();
1396 do {
1397 seq = read_seqbegin(sl: &mount_lock);
1398 res = __path_is_mountpoint(path);
1399 } while (read_seqretry(sl: &mount_lock, start: seq));
1400 rcu_read_unlock();
1401
1402 return res;
1403}
1404EXPORT_SYMBOL(path_is_mountpoint);
1405
1406struct vfsmount *mnt_clone_internal(const struct path *path)
1407{
1408 struct mount *p;
1409 p = clone_mnt(old: real_mount(mnt: path->mnt), root: path->dentry, CL_PRIVATE);
1410 if (IS_ERR(ptr: p))
1411 return ERR_CAST(ptr: p);
1412 p->mnt.mnt_flags |= MNT_INTERNAL;
1413 return &p->mnt;
1414}
1415
1416#ifdef CONFIG_PROC_FS
1417static struct mount *mnt_list_next(struct mnt_namespace *ns,
1418 struct list_head *p)
1419{
1420 struct mount *mnt, *ret = NULL;
1421
1422 lock_ns_list(ns);
1423 list_for_each_continue(p, &ns->list) {
1424 mnt = list_entry(p, typeof(*mnt), mnt_list);
1425 if (!mnt_is_cursor(mnt)) {
1426 ret = mnt;
1427 break;
1428 }
1429 }
1430 unlock_ns_list(ns);
1431
1432 return ret;
1433}
1434
1435/* iterator; we want it to have access to namespace_sem, thus here... */
1436static void *m_start(struct seq_file *m, loff_t *pos)
1437{
1438 struct proc_mounts *p = m->private;
1439 struct list_head *prev;
1440
1441 down_read(sem: &namespace_sem);
1442 if (!*pos) {
1443 prev = &p->ns->list;
1444 } else {
1445 prev = &p->cursor.mnt_list;
1446
1447 /* Read after we'd reached the end? */
1448 if (list_empty(head: prev))
1449 return NULL;
1450 }
1451
1452 return mnt_list_next(ns: p->ns, p: prev);
1453}
1454
1455static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1456{
1457 struct proc_mounts *p = m->private;
1458 struct mount *mnt = v;
1459
1460 ++*pos;
1461 return mnt_list_next(ns: p->ns, p: &mnt->mnt_list);
1462}
1463
1464static void m_stop(struct seq_file *m, void *v)
1465{
1466 struct proc_mounts *p = m->private;
1467 struct mount *mnt = v;
1468
1469 lock_ns_list(ns: p->ns);
1470 if (mnt)
1471 list_move_tail(list: &p->cursor.mnt_list, head: &mnt->mnt_list);
1472 else
1473 list_del_init(entry: &p->cursor.mnt_list);
1474 unlock_ns_list(ns: p->ns);
1475 up_read(sem: &namespace_sem);
1476}
1477
1478static int m_show(struct seq_file *m, void *v)
1479{
1480 struct proc_mounts *p = m->private;
1481 struct mount *r = v;
1482 return p->show(m, &r->mnt);
1483}
1484
1485const struct seq_operations mounts_op = {
1486 .start = m_start,
1487 .next = m_next,
1488 .stop = m_stop,
1489 .show = m_show,
1490};
1491
1492void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1493{
1494 down_read(sem: &namespace_sem);
1495 lock_ns_list(ns);
1496 list_del(entry: &cursor->mnt_list);
1497 unlock_ns_list(ns);
1498 up_read(sem: &namespace_sem);
1499}
1500#endif /* CONFIG_PROC_FS */
1501
1502/**
1503 * may_umount_tree - check if a mount tree is busy
1504 * @m: root of mount tree
1505 *
1506 * This is called to check if a tree of mounts has any
1507 * open files, pwds, chroots or sub mounts that are
1508 * busy.
1509 */
1510int may_umount_tree(struct vfsmount *m)
1511{
1512 struct mount *mnt = real_mount(mnt: m);
1513 int actual_refs = 0;
1514 int minimum_refs = 0;
1515 struct mount *p;
1516 BUG_ON(!m);
1517
1518 /* write lock needed for mnt_get_count */
1519 lock_mount_hash();
1520 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
1521 actual_refs += mnt_get_count(mnt: p);
1522 minimum_refs += 2;
1523 }
1524 unlock_mount_hash();
1525
1526 if (actual_refs > minimum_refs)
1527 return 0;
1528
1529 return 1;
1530}
1531
1532EXPORT_SYMBOL(may_umount_tree);
1533
1534/**
1535 * may_umount - check if a mount point is busy
1536 * @mnt: root of mount
1537 *
1538 * This is called to check if a mount point has any
1539 * open files, pwds, chroots or sub mounts. If the
1540 * mount has sub mounts this will return busy
1541 * regardless of whether the sub mounts are busy.
1542 *
1543 * Doesn't take quota and stuff into account. IOW, in some cases it will
1544 * give false negatives. The main reason why it's here is that we need
1545 * a non-destructive way to look for easily umountable filesystems.
1546 */
1547int may_umount(struct vfsmount *mnt)
1548{
1549 int ret = 1;
1550 down_read(sem: &namespace_sem);
1551 lock_mount_hash();
1552 if (propagate_mount_busy(real_mount(mnt), 2))
1553 ret = 0;
1554 unlock_mount_hash();
1555 up_read(sem: &namespace_sem);
1556 return ret;
1557}
1558
1559EXPORT_SYMBOL(may_umount);
1560
1561static void namespace_unlock(void)
1562{
1563 struct hlist_head head;
1564 struct hlist_node *p;
1565 struct mount *m;
1566 LIST_HEAD(list);
1567
1568 hlist_move_list(old: &unmounted, new: &head);
1569 list_splice_init(list: &ex_mountpoints, head: &list);
1570
1571 up_write(sem: &namespace_sem);
1572
1573 shrink_dentry_list(&list);
1574
1575 if (likely(hlist_empty(&head)))
1576 return;
1577
1578 synchronize_rcu_expedited();
1579
1580 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1581 hlist_del(n: &m->mnt_umount);
1582 mntput(&m->mnt);
1583 }
1584}
1585
1586static inline void namespace_lock(void)
1587{
1588 down_write(sem: &namespace_sem);
1589}
1590
1591enum umount_tree_flags {
1592 UMOUNT_SYNC = 1,
1593 UMOUNT_PROPAGATE = 2,
1594 UMOUNT_CONNECTED = 4,
1595};
1596
1597static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1598{
1599 /* Leaving mounts connected is only valid for lazy umounts */
1600 if (how & UMOUNT_SYNC)
1601 return true;
1602
1603 /* A mount without a parent has nothing to be connected to */
1604 if (!mnt_has_parent(mnt))
1605 return true;
1606
1607 /* Because the reference counting rules change when mounts are
1608 * unmounted and connected, umounted mounts may not be
1609 * connected to mounted mounts.
1610 */
1611 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1612 return true;
1613
1614 /* Has it been requested that the mount remain connected? */
1615 if (how & UMOUNT_CONNECTED)
1616 return false;
1617
1618 /* Is the mount locked such that it needs to remain connected? */
1619 if (IS_MNT_LOCKED(mnt))
1620 return false;
1621
1622 /* By default disconnect the mount */
1623 return true;
1624}
1625
1626/*
1627 * mount_lock must be held
1628 * namespace_sem must be held for write
1629 */
1630static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1631{
1632 LIST_HEAD(tmp_list);
1633 struct mount *p;
1634
1635 if (how & UMOUNT_PROPAGATE)
1636 propagate_mount_unlock(mnt);
1637
1638 /* Gather the mounts to umount */
1639 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
1640 p->mnt.mnt_flags |= MNT_UMOUNT;
1641 list_move(list: &p->mnt_list, head: &tmp_list);
1642 }
1643
1644 /* Hide the mounts from mnt_mounts */
1645 list_for_each_entry(p, &tmp_list, mnt_list) {
1646 list_del_init(entry: &p->mnt_child);
1647 }
1648
1649 /* Add propogated mounts to the tmp_list */
1650 if (how & UMOUNT_PROPAGATE)
1651 propagate_umount(&tmp_list);
1652
1653 while (!list_empty(head: &tmp_list)) {
1654 struct mnt_namespace *ns;
1655 bool disconnect;
1656 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1657 list_del_init(entry: &p->mnt_expire);
1658 list_del_init(entry: &p->mnt_list);
1659 ns = p->mnt_ns;
1660 if (ns) {
1661 ns->mounts--;
1662 __touch_mnt_namespace(ns);
1663 }
1664 p->mnt_ns = NULL;
1665 if (how & UMOUNT_SYNC)
1666 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1667
1668 disconnect = disconnect_mount(mnt: p, how);
1669 if (mnt_has_parent(mnt: p)) {
1670 mnt_add_count(mnt: p->mnt_parent, n: -1);
1671 if (!disconnect) {
1672 /* Don't forget about p */
1673 list_add_tail(new: &p->mnt_child, head: &p->mnt_parent->mnt_mounts);
1674 } else {
1675 umount_mnt(mnt: p);
1676 }
1677 }
1678 change_mnt_propagation(p, MS_PRIVATE);
1679 if (disconnect)
1680 hlist_add_head(n: &p->mnt_umount, h: &unmounted);
1681 }
1682}
1683
1684static void shrink_submounts(struct mount *mnt);
1685
1686static int do_umount_root(struct super_block *sb)
1687{
1688 int ret = 0;
1689
1690 down_write(sem: &sb->s_umount);
1691 if (!sb_rdonly(sb)) {
1692 struct fs_context *fc;
1693
1694 fc = fs_context_for_reconfigure(dentry: sb->s_root, SB_RDONLY,
1695 SB_RDONLY);
1696 if (IS_ERR(ptr: fc)) {
1697 ret = PTR_ERR(ptr: fc);
1698 } else {
1699 ret = parse_monolithic_mount_data(fc, NULL);
1700 if (!ret)
1701 ret = reconfigure_super(fc);
1702 put_fs_context(fc);
1703 }
1704 }
1705 up_write(sem: &sb->s_umount);
1706 return ret;
1707}
1708
1709static int do_umount(struct mount *mnt, int flags)
1710{
1711 struct super_block *sb = mnt->mnt.mnt_sb;
1712 int retval;
1713
1714 retval = security_sb_umount(mnt: &mnt->mnt, flags);
1715 if (retval)
1716 return retval;
1717
1718 /*
1719 * Allow userspace to request a mountpoint be expired rather than
1720 * unmounting unconditionally. Unmount only happens if:
1721 * (1) the mark is already set (the mark is cleared by mntput())
1722 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1723 */
1724 if (flags & MNT_EXPIRE) {
1725 if (&mnt->mnt == current->fs->root.mnt ||
1726 flags & (MNT_FORCE | MNT_DETACH))
1727 return -EINVAL;
1728
1729 /*
1730 * probably don't strictly need the lock here if we examined
1731 * all race cases, but it's a slowpath.
1732 */
1733 lock_mount_hash();
1734 if (mnt_get_count(mnt) != 2) {
1735 unlock_mount_hash();
1736 return -EBUSY;
1737 }
1738 unlock_mount_hash();
1739
1740 if (!xchg(&mnt->mnt_expiry_mark, 1))
1741 return -EAGAIN;
1742 }
1743
1744 /*
1745 * If we may have to abort operations to get out of this
1746 * mount, and they will themselves hold resources we must
1747 * allow the fs to do things. In the Unix tradition of
1748 * 'Gee thats tricky lets do it in userspace' the umount_begin
1749 * might fail to complete on the first run through as other tasks
1750 * must return, and the like. Thats for the mount program to worry
1751 * about for the moment.
1752 */
1753
1754 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1755 sb->s_op->umount_begin(sb);
1756 }
1757
1758 /*
1759 * No sense to grab the lock for this test, but test itself looks
1760 * somewhat bogus. Suggestions for better replacement?
1761 * Ho-hum... In principle, we might treat that as umount + switch
1762 * to rootfs. GC would eventually take care of the old vfsmount.
1763 * Actually it makes sense, especially if rootfs would contain a
1764 * /reboot - static binary that would close all descriptors and
1765 * call reboot(9). Then init(8) could umount root and exec /reboot.
1766 */
1767 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1768 /*
1769 * Special case for "unmounting" root ...
1770 * we just try to remount it readonly.
1771 */
1772 if (!ns_capable(ns: sb->s_user_ns, CAP_SYS_ADMIN))
1773 return -EPERM;
1774 return do_umount_root(sb);
1775 }
1776
1777 namespace_lock();
1778 lock_mount_hash();
1779
1780 /* Recheck MNT_LOCKED with the locks held */
1781 retval = -EINVAL;
1782 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1783 goto out;
1784
1785 event++;
1786 if (flags & MNT_DETACH) {
1787 if (!list_empty(head: &mnt->mnt_list))
1788 umount_tree(mnt, how: UMOUNT_PROPAGATE);
1789 retval = 0;
1790 } else {
1791 shrink_submounts(mnt);
1792 retval = -EBUSY;
1793 if (!propagate_mount_busy(mnt, 2)) {
1794 if (!list_empty(head: &mnt->mnt_list))
1795 umount_tree(mnt, how: UMOUNT_PROPAGATE|UMOUNT_SYNC);
1796 retval = 0;
1797 }
1798 }
1799out:
1800 unlock_mount_hash();
1801 namespace_unlock();
1802 return retval;
1803}
1804
1805/*
1806 * __detach_mounts - lazily unmount all mounts on the specified dentry
1807 *
1808 * During unlink, rmdir, and d_drop it is possible to loose the path
1809 * to an existing mountpoint, and wind up leaking the mount.
1810 * detach_mounts allows lazily unmounting those mounts instead of
1811 * leaking them.
1812 *
1813 * The caller may hold dentry->d_inode->i_mutex.
1814 */
1815void __detach_mounts(struct dentry *dentry)
1816{
1817 struct mountpoint *mp;
1818 struct mount *mnt;
1819
1820 namespace_lock();
1821 lock_mount_hash();
1822 mp = lookup_mountpoint(dentry);
1823 if (!mp)
1824 goto out_unlock;
1825
1826 event++;
1827 while (!hlist_empty(h: &mp->m_list)) {
1828 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1829 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1830 umount_mnt(mnt);
1831 hlist_add_head(n: &mnt->mnt_umount, h: &unmounted);
1832 }
1833 else umount_tree(mnt, how: UMOUNT_CONNECTED);
1834 }
1835 put_mountpoint(mp);
1836out_unlock:
1837 unlock_mount_hash();
1838 namespace_unlock();
1839}
1840
1841/*
1842 * Is the caller allowed to modify his namespace?
1843 */
1844bool may_mount(void)
1845{
1846 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1847}
1848
1849/**
1850 * path_mounted - check whether path is mounted
1851 * @path: path to check
1852 *
1853 * Determine whether @path refers to the root of a mount.
1854 *
1855 * Return: true if @path is the root of a mount, false if not.
1856 */
1857static inline bool path_mounted(const struct path *path)
1858{
1859 return path->mnt->mnt_root == path->dentry;
1860}
1861
1862static void warn_mandlock(void)
1863{
1864 pr_warn_once("=======================================================\n"
1865 "WARNING: The mand mount option has been deprecated and\n"
1866 " and is ignored by this kernel. Remove the mand\n"
1867 " option from the mount to silence this warning.\n"
1868 "=======================================================\n");
1869}
1870
1871static int can_umount(const struct path *path, int flags)
1872{
1873 struct mount *mnt = real_mount(mnt: path->mnt);
1874
1875 if (!may_mount())
1876 return -EPERM;
1877 if (!path_mounted(path))
1878 return -EINVAL;
1879 if (!check_mnt(mnt))
1880 return -EINVAL;
1881 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1882 return -EINVAL;
1883 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1884 return -EPERM;
1885 return 0;
1886}
1887
1888// caller is responsible for flags being sane
1889int path_umount(struct path *path, int flags)
1890{
1891 struct mount *mnt = real_mount(mnt: path->mnt);
1892 int ret;
1893
1894 ret = can_umount(path, flags);
1895 if (!ret)
1896 ret = do_umount(mnt, flags);
1897
1898 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1899 dput(path->dentry);
1900 mntput_no_expire(mnt);
1901 return ret;
1902}
1903
1904static int ksys_umount(char __user *name, int flags)
1905{
1906 int lookup_flags = LOOKUP_MOUNTPOINT;
1907 struct path path;
1908 int ret;
1909
1910 // basic validity checks done first
1911 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1912 return -EINVAL;
1913
1914 if (!(flags & UMOUNT_NOFOLLOW))
1915 lookup_flags |= LOOKUP_FOLLOW;
1916 ret = user_path_at(AT_FDCWD, name, flags: lookup_flags, path: &path);
1917 if (ret)
1918 return ret;
1919 return path_umount(path: &path, flags);
1920}
1921
1922SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1923{
1924 return ksys_umount(name, flags);
1925}
1926
1927#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1928
1929/*
1930 * The 2.0 compatible umount. No flags.
1931 */
1932SYSCALL_DEFINE1(oldumount, char __user *, name)
1933{
1934 return ksys_umount(name, flags: 0);
1935}
1936
1937#endif
1938
1939static bool is_mnt_ns_file(struct dentry *dentry)
1940{
1941 /* Is this a proxy for a mount namespace? */
1942 return dentry->d_op == &ns_dentry_operations &&
1943 dentry->d_fsdata == &mntns_operations;
1944}
1945
1946static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1947{
1948 return container_of(ns, struct mnt_namespace, ns);
1949}
1950
1951struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1952{
1953 return &mnt->ns;
1954}
1955
1956static bool mnt_ns_loop(struct dentry *dentry)
1957{
1958 /* Could bind mounting the mount namespace inode cause a
1959 * mount namespace loop?
1960 */
1961 struct mnt_namespace *mnt_ns;
1962 if (!is_mnt_ns_file(dentry))
1963 return false;
1964
1965 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1966 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1967}
1968
1969struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1970 int flag)
1971{
1972 struct mount *res, *p, *q, *r, *parent;
1973
1974 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1975 return ERR_PTR(error: -EINVAL);
1976
1977 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1978 return ERR_PTR(error: -EINVAL);
1979
1980 res = q = clone_mnt(old: mnt, root: dentry, flag);
1981 if (IS_ERR(ptr: q))
1982 return q;
1983
1984 q->mnt_mountpoint = mnt->mnt_mountpoint;
1985
1986 p = mnt;
1987 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1988 struct mount *s;
1989 if (!is_subdir(r->mnt_mountpoint, dentry))
1990 continue;
1991
1992 for (s = r; s; s = next_mnt(p: s, root: r)) {
1993 if (!(flag & CL_COPY_UNBINDABLE) &&
1994 IS_MNT_UNBINDABLE(s)) {
1995 if (s->mnt.mnt_flags & MNT_LOCKED) {
1996 /* Both unbindable and locked. */
1997 q = ERR_PTR(error: -EPERM);
1998 goto out;
1999 } else {
2000 s = skip_mnt_tree(p: s);
2001 continue;
2002 }
2003 }
2004 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2005 is_mnt_ns_file(dentry: s->mnt.mnt_root)) {
2006 s = skip_mnt_tree(p: s);
2007 continue;
2008 }
2009 while (p != s->mnt_parent) {
2010 p = p->mnt_parent;
2011 q = q->mnt_parent;
2012 }
2013 p = s;
2014 parent = q;
2015 q = clone_mnt(old: p, root: p->mnt.mnt_root, flag);
2016 if (IS_ERR(ptr: q))
2017 goto out;
2018 lock_mount_hash();
2019 list_add_tail(new: &q->mnt_list, head: &res->mnt_list);
2020 attach_mnt(mnt: q, parent, mp: p->mnt_mp, beneath: false);
2021 unlock_mount_hash();
2022 }
2023 }
2024 return res;
2025out:
2026 if (res) {
2027 lock_mount_hash();
2028 umount_tree(mnt: res, how: UMOUNT_SYNC);
2029 unlock_mount_hash();
2030 }
2031 return q;
2032}
2033
2034/* Caller should check returned pointer for errors */
2035
2036struct vfsmount *collect_mounts(const struct path *path)
2037{
2038 struct mount *tree;
2039 namespace_lock();
2040 if (!check_mnt(mnt: real_mount(mnt: path->mnt)))
2041 tree = ERR_PTR(error: -EINVAL);
2042 else
2043 tree = copy_tree(mnt: real_mount(mnt: path->mnt), dentry: path->dentry,
2044 CL_COPY_ALL | CL_PRIVATE);
2045 namespace_unlock();
2046 if (IS_ERR(ptr: tree))
2047 return ERR_CAST(ptr: tree);
2048 return &tree->mnt;
2049}
2050
2051static void free_mnt_ns(struct mnt_namespace *);
2052static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2053
2054void dissolve_on_fput(struct vfsmount *mnt)
2055{
2056 struct mnt_namespace *ns;
2057 namespace_lock();
2058 lock_mount_hash();
2059 ns = real_mount(mnt)->mnt_ns;
2060 if (ns) {
2061 if (is_anon_ns(ns))
2062 umount_tree(mnt: real_mount(mnt), how: UMOUNT_CONNECTED);
2063 else
2064 ns = NULL;
2065 }
2066 unlock_mount_hash();
2067 namespace_unlock();
2068 if (ns)
2069 free_mnt_ns(ns);
2070}
2071
2072void drop_collected_mounts(struct vfsmount *mnt)
2073{
2074 namespace_lock();
2075 lock_mount_hash();
2076 umount_tree(mnt: real_mount(mnt), how: 0);
2077 unlock_mount_hash();
2078 namespace_unlock();
2079}
2080
2081static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2082{
2083 struct mount *child;
2084
2085 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2086 if (!is_subdir(child->mnt_mountpoint, dentry))
2087 continue;
2088
2089 if (child->mnt.mnt_flags & MNT_LOCKED)
2090 return true;
2091 }
2092 return false;
2093}
2094
2095/**
2096 * clone_private_mount - create a private clone of a path
2097 * @path: path to clone
2098 *
2099 * This creates a new vfsmount, which will be the clone of @path. The new mount
2100 * will not be attached anywhere in the namespace and will be private (i.e.
2101 * changes to the originating mount won't be propagated into this).
2102 *
2103 * Release with mntput().
2104 */
2105struct vfsmount *clone_private_mount(const struct path *path)
2106{
2107 struct mount *old_mnt = real_mount(mnt: path->mnt);
2108 struct mount *new_mnt;
2109
2110 down_read(sem: &namespace_sem);
2111 if (IS_MNT_UNBINDABLE(old_mnt))
2112 goto invalid;
2113
2114 if (!check_mnt(mnt: old_mnt))
2115 goto invalid;
2116
2117 if (has_locked_children(mnt: old_mnt, dentry: path->dentry))
2118 goto invalid;
2119
2120 new_mnt = clone_mnt(old: old_mnt, root: path->dentry, CL_PRIVATE);
2121 up_read(sem: &namespace_sem);
2122
2123 if (IS_ERR(ptr: new_mnt))
2124 return ERR_CAST(ptr: new_mnt);
2125
2126 /* Longterm mount to be removed by kern_unmount*() */
2127 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2128
2129 return &new_mnt->mnt;
2130
2131invalid:
2132 up_read(sem: &namespace_sem);
2133 return ERR_PTR(error: -EINVAL);
2134}
2135EXPORT_SYMBOL_GPL(clone_private_mount);
2136
2137int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2138 struct vfsmount *root)
2139{
2140 struct mount *mnt;
2141 int res = f(root, arg);
2142 if (res)
2143 return res;
2144 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2145 res = f(&mnt->mnt, arg);
2146 if (res)
2147 return res;
2148 }
2149 return 0;
2150}
2151
2152static void lock_mnt_tree(struct mount *mnt)
2153{
2154 struct mount *p;
2155
2156 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
2157 int flags = p->mnt.mnt_flags;
2158 /* Don't allow unprivileged users to change mount flags */
2159 flags |= MNT_LOCK_ATIME;
2160
2161 if (flags & MNT_READONLY)
2162 flags |= MNT_LOCK_READONLY;
2163
2164 if (flags & MNT_NODEV)
2165 flags |= MNT_LOCK_NODEV;
2166
2167 if (flags & MNT_NOSUID)
2168 flags |= MNT_LOCK_NOSUID;
2169
2170 if (flags & MNT_NOEXEC)
2171 flags |= MNT_LOCK_NOEXEC;
2172 /* Don't allow unprivileged users to reveal what is under a mount */
2173 if (list_empty(head: &p->mnt_expire))
2174 flags |= MNT_LOCKED;
2175 p->mnt.mnt_flags = flags;
2176 }
2177}
2178
2179static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2180{
2181 struct mount *p;
2182
2183 for (p = mnt; p != end; p = next_mnt(p, root: mnt)) {
2184 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2185 mnt_release_group_id(mnt: p);
2186 }
2187}
2188
2189static int invent_group_ids(struct mount *mnt, bool recurse)
2190{
2191 struct mount *p;
2192
2193 for (p = mnt; p; p = recurse ? next_mnt(p, root: mnt) : NULL) {
2194 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2195 int err = mnt_alloc_group_id(mnt: p);
2196 if (err) {
2197 cleanup_group_ids(mnt, end: p);
2198 return err;
2199 }
2200 }
2201 }
2202
2203 return 0;
2204}
2205
2206int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2207{
2208 unsigned int max = READ_ONCE(sysctl_mount_max);
2209 unsigned int mounts = 0;
2210 struct mount *p;
2211
2212 if (ns->mounts >= max)
2213 return -ENOSPC;
2214 max -= ns->mounts;
2215 if (ns->pending_mounts >= max)
2216 return -ENOSPC;
2217 max -= ns->pending_mounts;
2218
2219 for (p = mnt; p; p = next_mnt(p, root: mnt))
2220 mounts++;
2221
2222 if (mounts > max)
2223 return -ENOSPC;
2224
2225 ns->pending_mounts += mounts;
2226 return 0;
2227}
2228
2229enum mnt_tree_flags_t {
2230 MNT_TREE_MOVE = BIT(0),
2231 MNT_TREE_BENEATH = BIT(1),
2232};
2233
2234/**
2235 * attach_recursive_mnt - attach a source mount tree
2236 * @source_mnt: mount tree to be attached
2237 * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath
2238 * @dest_mp: the mountpoint @source_mnt will be mounted at
2239 * @flags: modify how @source_mnt is supposed to be attached
2240 *
2241 * NOTE: in the table below explains the semantics when a source mount
2242 * of a given type is attached to a destination mount of a given type.
2243 * ---------------------------------------------------------------------------
2244 * | BIND MOUNT OPERATION |
2245 * |**************************************************************************
2246 * | source-->| shared | private | slave | unbindable |
2247 * | dest | | | | |
2248 * | | | | | | |
2249 * | v | | | | |
2250 * |**************************************************************************
2251 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2252 * | | | | | |
2253 * |non-shared| shared (+) | private | slave (*) | invalid |
2254 * ***************************************************************************
2255 * A bind operation clones the source mount and mounts the clone on the
2256 * destination mount.
2257 *
2258 * (++) the cloned mount is propagated to all the mounts in the propagation
2259 * tree of the destination mount and the cloned mount is added to
2260 * the peer group of the source mount.
2261 * (+) the cloned mount is created under the destination mount and is marked
2262 * as shared. The cloned mount is added to the peer group of the source
2263 * mount.
2264 * (+++) the mount is propagated to all the mounts in the propagation tree
2265 * of the destination mount and the cloned mount is made slave
2266 * of the same master as that of the source mount. The cloned mount
2267 * is marked as 'shared and slave'.
2268 * (*) the cloned mount is made a slave of the same master as that of the
2269 * source mount.
2270 *
2271 * ---------------------------------------------------------------------------
2272 * | MOVE MOUNT OPERATION |
2273 * |**************************************************************************
2274 * | source-->| shared | private | slave | unbindable |
2275 * | dest | | | | |
2276 * | | | | | | |
2277 * | v | | | | |
2278 * |**************************************************************************
2279 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2280 * | | | | | |
2281 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2282 * ***************************************************************************
2283 *
2284 * (+) the mount is moved to the destination. And is then propagated to
2285 * all the mounts in the propagation tree of the destination mount.
2286 * (+*) the mount is moved to the destination.
2287 * (+++) the mount is moved to the destination and is then propagated to
2288 * all the mounts belonging to the destination mount's propagation tree.
2289 * the mount is marked as 'shared and slave'.
2290 * (*) the mount continues to be a slave at the new location.
2291 *
2292 * if the source mount is a tree, the operations explained above is
2293 * applied to each mount in the tree.
2294 * Must be called without spinlocks held, since this function can sleep
2295 * in allocations.
2296 *
2297 * Context: The function expects namespace_lock() to be held.
2298 * Return: If @source_mnt was successfully attached 0 is returned.
2299 * Otherwise a negative error code is returned.
2300 */
2301static int attach_recursive_mnt(struct mount *source_mnt,
2302 struct mount *top_mnt,
2303 struct mountpoint *dest_mp,
2304 enum mnt_tree_flags_t flags)
2305{
2306 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2307 HLIST_HEAD(tree_list);
2308 struct mnt_namespace *ns = top_mnt->mnt_ns;
2309 struct mountpoint *smp;
2310 struct mount *child, *dest_mnt, *p;
2311 struct hlist_node *n;
2312 int err = 0;
2313 bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
2314
2315 /*
2316 * Preallocate a mountpoint in case the new mounts need to be
2317 * mounted beneath mounts on the same mountpoint.
2318 */
2319 smp = get_mountpoint(dentry: source_mnt->mnt.mnt_root);
2320 if (IS_ERR(ptr: smp))
2321 return PTR_ERR(ptr: smp);
2322
2323 /* Is there space to add these mounts to the mount namespace? */
2324 if (!moving) {
2325 err = count_mounts(ns, mnt: source_mnt);
2326 if (err)
2327 goto out;
2328 }
2329
2330 if (beneath)
2331 dest_mnt = top_mnt->mnt_parent;
2332 else
2333 dest_mnt = top_mnt;
2334
2335 if (IS_MNT_SHARED(dest_mnt)) {
2336 err = invent_group_ids(mnt: source_mnt, recurse: true);
2337 if (err)
2338 goto out;
2339 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2340 }
2341 lock_mount_hash();
2342 if (err)
2343 goto out_cleanup_ids;
2344
2345 if (IS_MNT_SHARED(dest_mnt)) {
2346 for (p = source_mnt; p; p = next_mnt(p, root: source_mnt))
2347 set_mnt_shared(p);
2348 }
2349
2350 if (moving) {
2351 if (beneath)
2352 dest_mp = smp;
2353 unhash_mnt(mnt: source_mnt);
2354 attach_mnt(mnt: source_mnt, parent: top_mnt, mp: dest_mp, beneath);
2355 touch_mnt_namespace(ns: source_mnt->mnt_ns);
2356 } else {
2357 if (source_mnt->mnt_ns) {
2358 /* move from anon - the caller will destroy */
2359 list_del_init(entry: &source_mnt->mnt_ns->list);
2360 }
2361 if (beneath)
2362 mnt_set_mountpoint_beneath(new_parent: source_mnt, top_mnt, new_mp: smp);
2363 else
2364 mnt_set_mountpoint(mnt: dest_mnt, mp: dest_mp, child_mnt: source_mnt);
2365 commit_tree(mnt: source_mnt);
2366 }
2367
2368 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2369 struct mount *q;
2370 hlist_del_init(n: &child->mnt_hash);
2371 q = __lookup_mnt(mnt: &child->mnt_parent->mnt,
2372 dentry: child->mnt_mountpoint);
2373 if (q)
2374 mnt_change_mountpoint(parent: child, mp: smp, mnt: q);
2375 /* Notice when we are propagating across user namespaces */
2376 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2377 lock_mnt_tree(mnt: child);
2378 child->mnt.mnt_flags &= ~MNT_LOCKED;
2379 commit_tree(mnt: child);
2380 }
2381 put_mountpoint(mp: smp);
2382 unlock_mount_hash();
2383
2384 return 0;
2385
2386 out_cleanup_ids:
2387 while (!hlist_empty(h: &tree_list)) {
2388 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2389 child->mnt_parent->mnt_ns->pending_mounts = 0;
2390 umount_tree(mnt: child, how: UMOUNT_SYNC);
2391 }
2392 unlock_mount_hash();
2393 cleanup_group_ids(mnt: source_mnt, NULL);
2394 out:
2395 ns->pending_mounts = 0;
2396
2397 read_seqlock_excl(sl: &mount_lock);
2398 put_mountpoint(mp: smp);
2399 read_sequnlock_excl(sl: &mount_lock);
2400
2401 return err;
2402}
2403
2404/**
2405 * do_lock_mount - lock mount and mountpoint
2406 * @path: target path
2407 * @beneath: whether the intention is to mount beneath @path
2408 *
2409 * Follow the mount stack on @path until the top mount @mnt is found. If
2410 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2411 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2412 * until nothing is stacked on top of it anymore.
2413 *
2414 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2415 * against concurrent removal of the new mountpoint from another mount
2416 * namespace.
2417 *
2418 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2419 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2420 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2421 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2422 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2423 * on top of it for @beneath.
2424 *
2425 * In addition, @beneath needs to make sure that @mnt hasn't been
2426 * unmounted or moved from its current mountpoint in between dropping
2427 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2428 * being unmounted would be detected later by e.g., calling
2429 * check_mnt(mnt) in the function it's called from. For the @beneath
2430 * case however, it's useful to detect it directly in do_lock_mount().
2431 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2432 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2433 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2434 *
2435 * Return: Either the target mountpoint on the top mount or the top
2436 * mount's mountpoint.
2437 */
2438static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
2439{
2440 struct vfsmount *mnt = path->mnt;
2441 struct dentry *dentry;
2442 struct mountpoint *mp = ERR_PTR(error: -ENOENT);
2443
2444 for (;;) {
2445 struct mount *m;
2446
2447 if (beneath) {
2448 m = real_mount(mnt);
2449 read_seqlock_excl(sl: &mount_lock);
2450 dentry = dget(dentry: m->mnt_mountpoint);
2451 read_sequnlock_excl(sl: &mount_lock);
2452 } else {
2453 dentry = path->dentry;
2454 }
2455
2456 inode_lock(inode: dentry->d_inode);
2457 if (unlikely(cant_mount(dentry))) {
2458 inode_unlock(inode: dentry->d_inode);
2459 goto out;
2460 }
2461
2462 namespace_lock();
2463
2464 if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) {
2465 namespace_unlock();
2466 inode_unlock(inode: dentry->d_inode);
2467 goto out;
2468 }
2469
2470 mnt = lookup_mnt(path);
2471 if (likely(!mnt))
2472 break;
2473
2474 namespace_unlock();
2475 inode_unlock(inode: dentry->d_inode);
2476 if (beneath)
2477 dput(dentry);
2478 path_put(path);
2479 path->mnt = mnt;
2480 path->dentry = dget(dentry: mnt->mnt_root);
2481 }
2482
2483 mp = get_mountpoint(dentry);
2484 if (IS_ERR(ptr: mp)) {
2485 namespace_unlock();
2486 inode_unlock(inode: dentry->d_inode);
2487 }
2488
2489out:
2490 if (beneath)
2491 dput(dentry);
2492
2493 return mp;
2494}
2495
2496static inline struct mountpoint *lock_mount(struct path *path)
2497{
2498 return do_lock_mount(path, beneath: false);
2499}
2500
2501static void unlock_mount(struct mountpoint *where)
2502{
2503 struct dentry *dentry = where->m_dentry;
2504
2505 read_seqlock_excl(sl: &mount_lock);
2506 put_mountpoint(mp: where);
2507 read_sequnlock_excl(sl: &mount_lock);
2508
2509 namespace_unlock();
2510 inode_unlock(inode: dentry->d_inode);
2511}
2512
2513static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2514{
2515 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2516 return -EINVAL;
2517
2518 if (d_is_dir(dentry: mp->m_dentry) !=
2519 d_is_dir(dentry: mnt->mnt.mnt_root))
2520 return -ENOTDIR;
2521
2522 return attach_recursive_mnt(source_mnt: mnt, top_mnt: p, dest_mp: mp, flags: 0);
2523}
2524
2525/*
2526 * Sanity check the flags to change_mnt_propagation.
2527 */
2528
2529static int flags_to_propagation_type(int ms_flags)
2530{
2531 int type = ms_flags & ~(MS_REC | MS_SILENT);
2532
2533 /* Fail if any non-propagation flags are set */
2534 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2535 return 0;
2536 /* Only one propagation flag should be set */
2537 if (!is_power_of_2(n: type))
2538 return 0;
2539 return type;
2540}
2541
2542/*
2543 * recursively change the type of the mountpoint.
2544 */
2545static int do_change_type(struct path *path, int ms_flags)
2546{
2547 struct mount *m;
2548 struct mount *mnt = real_mount(mnt: path->mnt);
2549 int recurse = ms_flags & MS_REC;
2550 int type;
2551 int err = 0;
2552
2553 if (!path_mounted(path))
2554 return -EINVAL;
2555
2556 type = flags_to_propagation_type(ms_flags);
2557 if (!type)
2558 return -EINVAL;
2559
2560 namespace_lock();
2561 if (type == MS_SHARED) {
2562 err = invent_group_ids(mnt, recurse);
2563 if (err)
2564 goto out_unlock;
2565 }
2566
2567 lock_mount_hash();
2568 for (m = mnt; m; m = (recurse ? next_mnt(p: m, root: mnt) : NULL))
2569 change_mnt_propagation(m, type);
2570 unlock_mount_hash();
2571
2572 out_unlock:
2573 namespace_unlock();
2574 return err;
2575}
2576
2577static struct mount *__do_loopback(struct path *old_path, int recurse)
2578{
2579 struct mount *mnt = ERR_PTR(error: -EINVAL), *old = real_mount(mnt: old_path->mnt);
2580
2581 if (IS_MNT_UNBINDABLE(old))
2582 return mnt;
2583
2584 if (!check_mnt(mnt: old) && old_path->dentry->d_op != &ns_dentry_operations)
2585 return mnt;
2586
2587 if (!recurse && has_locked_children(mnt: old, dentry: old_path->dentry))
2588 return mnt;
2589
2590 if (recurse)
2591 mnt = copy_tree(mnt: old, dentry: old_path->dentry, CL_COPY_MNT_NS_FILE);
2592 else
2593 mnt = clone_mnt(old, root: old_path->dentry, flag: 0);
2594
2595 if (!IS_ERR(ptr: mnt))
2596 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2597
2598 return mnt;
2599}
2600
2601/*
2602 * do loopback mount.
2603 */
2604static int do_loopback(struct path *path, const char *old_name,
2605 int recurse)
2606{
2607 struct path old_path;
2608 struct mount *mnt = NULL, *parent;
2609 struct mountpoint *mp;
2610 int err;
2611 if (!old_name || !*old_name)
2612 return -EINVAL;
2613 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2614 if (err)
2615 return err;
2616
2617 err = -EINVAL;
2618 if (mnt_ns_loop(dentry: old_path.dentry))
2619 goto out;
2620
2621 mp = lock_mount(path);
2622 if (IS_ERR(ptr: mp)) {
2623 err = PTR_ERR(ptr: mp);
2624 goto out;
2625 }
2626
2627 parent = real_mount(mnt: path->mnt);
2628 if (!check_mnt(mnt: parent))
2629 goto out2;
2630
2631 mnt = __do_loopback(old_path: &old_path, recurse);
2632 if (IS_ERR(ptr: mnt)) {
2633 err = PTR_ERR(ptr: mnt);
2634 goto out2;
2635 }
2636
2637 err = graft_tree(mnt, p: parent, mp);
2638 if (err) {
2639 lock_mount_hash();
2640 umount_tree(mnt, how: UMOUNT_SYNC);
2641 unlock_mount_hash();
2642 }
2643out2:
2644 unlock_mount(where: mp);
2645out:
2646 path_put(&old_path);
2647 return err;
2648}
2649
2650static struct file *open_detached_copy(struct path *path, bool recursive)
2651{
2652 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2653 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2654 struct mount *mnt, *p;
2655 struct file *file;
2656
2657 if (IS_ERR(ptr: ns))
2658 return ERR_CAST(ptr: ns);
2659
2660 namespace_lock();
2661 mnt = __do_loopback(old_path: path, recurse: recursive);
2662 if (IS_ERR(ptr: mnt)) {
2663 namespace_unlock();
2664 free_mnt_ns(ns);
2665 return ERR_CAST(ptr: mnt);
2666 }
2667
2668 lock_mount_hash();
2669 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
2670 p->mnt_ns = ns;
2671 ns->mounts++;
2672 }
2673 ns->root = mnt;
2674 list_add_tail(new: &ns->list, head: &mnt->mnt_list);
2675 mntget(&mnt->mnt);
2676 unlock_mount_hash();
2677 namespace_unlock();
2678
2679 mntput(path->mnt);
2680 path->mnt = &mnt->mnt;
2681 file = dentry_open(path, O_PATH, current_cred());
2682 if (IS_ERR(ptr: file))
2683 dissolve_on_fput(mnt: path->mnt);
2684 else
2685 file->f_mode |= FMODE_NEED_UNMOUNT;
2686 return file;
2687}
2688
2689SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2690{
2691 struct file *file;
2692 struct path path;
2693 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2694 bool detached = flags & OPEN_TREE_CLONE;
2695 int error;
2696 int fd;
2697
2698 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2699
2700 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2701 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2702 OPEN_TREE_CLOEXEC))
2703 return -EINVAL;
2704
2705 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2706 return -EINVAL;
2707
2708 if (flags & AT_NO_AUTOMOUNT)
2709 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2710 if (flags & AT_SYMLINK_NOFOLLOW)
2711 lookup_flags &= ~LOOKUP_FOLLOW;
2712 if (flags & AT_EMPTY_PATH)
2713 lookup_flags |= LOOKUP_EMPTY;
2714
2715 if (detached && !may_mount())
2716 return -EPERM;
2717
2718 fd = get_unused_fd_flags(flags: flags & O_CLOEXEC);
2719 if (fd < 0)
2720 return fd;
2721
2722 error = user_path_at(dfd, name: filename, flags: lookup_flags, path: &path);
2723 if (unlikely(error)) {
2724 file = ERR_PTR(error);
2725 } else {
2726 if (detached)
2727 file = open_detached_copy(path: &path, recursive: flags & AT_RECURSIVE);
2728 else
2729 file = dentry_open(path: &path, O_PATH, current_cred());
2730 path_put(&path);
2731 }
2732 if (IS_ERR(ptr: file)) {
2733 put_unused_fd(fd);
2734 return PTR_ERR(ptr: file);
2735 }
2736 fd_install(fd, file);
2737 return fd;
2738}
2739
2740/*
2741 * Don't allow locked mount flags to be cleared.
2742 *
2743 * No locks need to be held here while testing the various MNT_LOCK
2744 * flags because those flags can never be cleared once they are set.
2745 */
2746static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2747{
2748 unsigned int fl = mnt->mnt.mnt_flags;
2749
2750 if ((fl & MNT_LOCK_READONLY) &&
2751 !(mnt_flags & MNT_READONLY))
2752 return false;
2753
2754 if ((fl & MNT_LOCK_NODEV) &&
2755 !(mnt_flags & MNT_NODEV))
2756 return false;
2757
2758 if ((fl & MNT_LOCK_NOSUID) &&
2759 !(mnt_flags & MNT_NOSUID))
2760 return false;
2761
2762 if ((fl & MNT_LOCK_NOEXEC) &&
2763 !(mnt_flags & MNT_NOEXEC))
2764 return false;
2765
2766 if ((fl & MNT_LOCK_ATIME) &&
2767 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2768 return false;
2769
2770 return true;
2771}
2772
2773static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2774{
2775 bool readonly_request = (mnt_flags & MNT_READONLY);
2776
2777 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2778 return 0;
2779
2780 if (readonly_request)
2781 return mnt_make_readonly(mnt);
2782
2783 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2784 return 0;
2785}
2786
2787static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2788{
2789 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2790 mnt->mnt.mnt_flags = mnt_flags;
2791 touch_mnt_namespace(ns: mnt->mnt_ns);
2792}
2793
2794static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2795{
2796 struct super_block *sb = mnt->mnt_sb;
2797
2798 if (!__mnt_is_readonly(mnt) &&
2799 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2800 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2801 char *buf = (char *)__get_free_page(GFP_KERNEL);
2802 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(error: -ENOMEM);
2803
2804 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2805 sb->s_type->name,
2806 is_mounted(mnt) ? "remounted" : "mounted",
2807 mntpath, &sb->s_time_max,
2808 (unsigned long long)sb->s_time_max);
2809
2810 free_page((unsigned long)buf);
2811 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2812 }
2813}
2814
2815/*
2816 * Handle reconfiguration of the mountpoint only without alteration of the
2817 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2818 * to mount(2).
2819 */
2820static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2821{
2822 struct super_block *sb = path->mnt->mnt_sb;
2823 struct mount *mnt = real_mount(mnt: path->mnt);
2824 int ret;
2825
2826 if (!check_mnt(mnt))
2827 return -EINVAL;
2828
2829 if (!path_mounted(path))
2830 return -EINVAL;
2831
2832 if (!can_change_locked_flags(mnt, mnt_flags))
2833 return -EPERM;
2834
2835 /*
2836 * We're only checking whether the superblock is read-only not
2837 * changing it, so only take down_read(&sb->s_umount).
2838 */
2839 down_read(sem: &sb->s_umount);
2840 lock_mount_hash();
2841 ret = change_mount_ro_state(mnt, mnt_flags);
2842 if (ret == 0)
2843 set_mount_attributes(mnt, mnt_flags);
2844 unlock_mount_hash();
2845 up_read(sem: &sb->s_umount);
2846
2847 mnt_warn_timestamp_expiry(mountpoint: path, mnt: &mnt->mnt);
2848
2849 return ret;
2850}
2851
2852/*
2853 * change filesystem flags. dir should be a physical root of filesystem.
2854 * If you've mounted a non-root directory somewhere and want to do remount
2855 * on it - tough luck.
2856 */
2857static int do_remount(struct path *path, int ms_flags, int sb_flags,
2858 int mnt_flags, void *data)
2859{
2860 int err;
2861 struct super_block *sb = path->mnt->mnt_sb;
2862 struct mount *mnt = real_mount(mnt: path->mnt);
2863 struct fs_context *fc;
2864
2865 if (!check_mnt(mnt))
2866 return -EINVAL;
2867
2868 if (!path_mounted(path))
2869 return -EINVAL;
2870
2871 if (!can_change_locked_flags(mnt, mnt_flags))
2872 return -EPERM;
2873
2874 fc = fs_context_for_reconfigure(dentry: path->dentry, sb_flags, MS_RMT_MASK);
2875 if (IS_ERR(ptr: fc))
2876 return PTR_ERR(ptr: fc);
2877
2878 fc->oldapi = true;
2879 err = parse_monolithic_mount_data(fc, data);
2880 if (!err) {
2881 down_write(sem: &sb->s_umount);
2882 err = -EPERM;
2883 if (ns_capable(ns: sb->s_user_ns, CAP_SYS_ADMIN)) {
2884 err = reconfigure_super(fc);
2885 if (!err) {
2886 lock_mount_hash();
2887 set_mount_attributes(mnt, mnt_flags);
2888 unlock_mount_hash();
2889 }
2890 }
2891 up_write(sem: &sb->s_umount);
2892 }
2893
2894 mnt_warn_timestamp_expiry(mountpoint: path, mnt: &mnt->mnt);
2895
2896 put_fs_context(fc);
2897 return err;
2898}
2899
2900static inline int tree_contains_unbindable(struct mount *mnt)
2901{
2902 struct mount *p;
2903 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
2904 if (IS_MNT_UNBINDABLE(p))
2905 return 1;
2906 }
2907 return 0;
2908}
2909
2910/*
2911 * Check that there aren't references to earlier/same mount namespaces in the
2912 * specified subtree. Such references can act as pins for mount namespaces
2913 * that aren't checked by the mount-cycle checking code, thereby allowing
2914 * cycles to be made.
2915 */
2916static bool check_for_nsfs_mounts(struct mount *subtree)
2917{
2918 struct mount *p;
2919 bool ret = false;
2920
2921 lock_mount_hash();
2922 for (p = subtree; p; p = next_mnt(p, root: subtree))
2923 if (mnt_ns_loop(dentry: p->mnt.mnt_root))
2924 goto out;
2925
2926 ret = true;
2927out:
2928 unlock_mount_hash();
2929 return ret;
2930}
2931
2932static int do_set_group(struct path *from_path, struct path *to_path)
2933{
2934 struct mount *from, *to;
2935 int err;
2936
2937 from = real_mount(mnt: from_path->mnt);
2938 to = real_mount(mnt: to_path->mnt);
2939
2940 namespace_lock();
2941
2942 err = -EINVAL;
2943 /* To and From must be mounted */
2944 if (!is_mounted(mnt: &from->mnt))
2945 goto out;
2946 if (!is_mounted(mnt: &to->mnt))
2947 goto out;
2948
2949 err = -EPERM;
2950 /* We should be allowed to modify mount namespaces of both mounts */
2951 if (!ns_capable(ns: from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2952 goto out;
2953 if (!ns_capable(ns: to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2954 goto out;
2955
2956 err = -EINVAL;
2957 /* To and From paths should be mount roots */
2958 if (!path_mounted(path: from_path))
2959 goto out;
2960 if (!path_mounted(path: to_path))
2961 goto out;
2962
2963 /* Setting sharing groups is only allowed across same superblock */
2964 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2965 goto out;
2966
2967 /* From mount root should be wider than To mount root */
2968 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2969 goto out;
2970
2971 /* From mount should not have locked children in place of To's root */
2972 if (has_locked_children(mnt: from, dentry: to->mnt.mnt_root))
2973 goto out;
2974
2975 /* Setting sharing groups is only allowed on private mounts */
2976 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2977 goto out;
2978
2979 /* From should not be private */
2980 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2981 goto out;
2982
2983 if (IS_MNT_SLAVE(from)) {
2984 struct mount *m = from->mnt_master;
2985
2986 list_add(new: &to->mnt_slave, head: &m->mnt_slave_list);
2987 to->mnt_master = m;
2988 }
2989
2990 if (IS_MNT_SHARED(from)) {
2991 to->mnt_group_id = from->mnt_group_id;
2992 list_add(new: &to->mnt_share, head: &from->mnt_share);
2993 lock_mount_hash();
2994 set_mnt_shared(to);
2995 unlock_mount_hash();
2996 }
2997
2998 err = 0;
2999out:
3000 namespace_unlock();
3001 return err;
3002}
3003
3004/**
3005 * path_overmounted - check if path is overmounted
3006 * @path: path to check
3007 *
3008 * Check if path is overmounted, i.e., if there's a mount on top of
3009 * @path->mnt with @path->dentry as mountpoint.
3010 *
3011 * Context: This function expects namespace_lock() to be held.
3012 * Return: If path is overmounted true is returned, false if not.
3013 */
3014static inline bool path_overmounted(const struct path *path)
3015{
3016 rcu_read_lock();
3017 if (unlikely(__lookup_mnt(path->mnt, path->dentry))) {
3018 rcu_read_unlock();
3019 return true;
3020 }
3021 rcu_read_unlock();
3022 return false;
3023}
3024
3025/**
3026 * can_move_mount_beneath - check that we can mount beneath the top mount
3027 * @from: mount to mount beneath
3028 * @to: mount under which to mount
3029 *
3030 * - Make sure that @to->dentry is actually the root of a mount under
3031 * which we can mount another mount.
3032 * - Make sure that nothing can be mounted beneath the caller's current
3033 * root or the rootfs of the namespace.
3034 * - Make sure that the caller can unmount the topmost mount ensuring
3035 * that the caller could reveal the underlying mountpoint.
3036 * - Ensure that nothing has been mounted on top of @from before we
3037 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3038 * - Prevent mounting beneath a mount if the propagation relationship
3039 * between the source mount, parent mount, and top mount would lead to
3040 * nonsensical mount trees.
3041 *
3042 * Context: This function expects namespace_lock() to be held.
3043 * Return: On success 0, and on error a negative error code is returned.
3044 */
3045static int can_move_mount_beneath(const struct path *from,
3046 const struct path *to,
3047 const struct mountpoint *mp)
3048{
3049 struct mount *mnt_from = real_mount(mnt: from->mnt),
3050 *mnt_to = real_mount(mnt: to->mnt),
3051 *parent_mnt_to = mnt_to->mnt_parent;
3052
3053 if (!mnt_has_parent(mnt: mnt_to))
3054 return -EINVAL;
3055
3056 if (!path_mounted(path: to))
3057 return -EINVAL;
3058
3059 if (IS_MNT_LOCKED(mnt_to))
3060 return -EINVAL;
3061
3062 /* Avoid creating shadow mounts during mount propagation. */
3063 if (path_overmounted(path: from))
3064 return -EINVAL;
3065
3066 /*
3067 * Mounting beneath the rootfs only makes sense when the
3068 * semantics of pivot_root(".", ".") are used.
3069 */
3070 if (&mnt_to->mnt == current->fs->root.mnt)
3071 return -EINVAL;
3072 if (parent_mnt_to == current->nsproxy->mnt_ns->root)
3073 return -EINVAL;
3074
3075 for (struct mount *p = mnt_from; mnt_has_parent(mnt: p); p = p->mnt_parent)
3076 if (p == mnt_to)
3077 return -EINVAL;
3078
3079 /*
3080 * If the parent mount propagates to the child mount this would
3081 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3082 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3083 * defeats the whole purpose of mounting beneath another mount.
3084 */
3085 if (propagation_would_overmount(from: parent_mnt_to, to: mnt_to, mp))
3086 return -EINVAL;
3087
3088 /*
3089 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3090 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3091 * Afterwards @mnt_from would be mounted on top of
3092 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3093 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3094 * already mounted on @mnt_from, @mnt_to would ultimately be
3095 * remounted on top of @c. Afterwards, @mnt_from would be
3096 * covered by a copy @c of @mnt_from and @c would be covered by
3097 * @mnt_from itself. This defeats the whole purpose of mounting
3098 * @mnt_from beneath @mnt_to.
3099 */
3100 if (propagation_would_overmount(from: parent_mnt_to, to: mnt_from, mp))
3101 return -EINVAL;
3102
3103 return 0;
3104}
3105
3106static int do_move_mount(struct path *old_path, struct path *new_path,
3107 bool beneath)
3108{
3109 struct mnt_namespace *ns;
3110 struct mount *p;
3111 struct mount *old;
3112 struct mount *parent;
3113 struct mountpoint *mp, *old_mp;
3114 int err;
3115 bool attached;
3116 enum mnt_tree_flags_t flags = 0;
3117
3118 mp = do_lock_mount(path: new_path, beneath);
3119 if (IS_ERR(ptr: mp))
3120 return PTR_ERR(ptr: mp);
3121
3122 old = real_mount(mnt: old_path->mnt);
3123 p = real_mount(mnt: new_path->mnt);
3124 parent = old->mnt_parent;
3125 attached = mnt_has_parent(mnt: old);
3126 if (attached)
3127 flags |= MNT_TREE_MOVE;
3128 old_mp = old->mnt_mp;
3129 ns = old->mnt_ns;
3130
3131 err = -EINVAL;
3132 /* The mountpoint must be in our namespace. */
3133 if (!check_mnt(mnt: p))
3134 goto out;
3135
3136 /* The thing moved must be mounted... */
3137 if (!is_mounted(mnt: &old->mnt))
3138 goto out;
3139
3140 /* ... and either ours or the root of anon namespace */
3141 if (!(attached ? check_mnt(mnt: old) : is_anon_ns(ns)))
3142 goto out;
3143
3144 if (old->mnt.mnt_flags & MNT_LOCKED)
3145 goto out;
3146
3147 if (!path_mounted(path: old_path))
3148 goto out;
3149
3150 if (d_is_dir(dentry: new_path->dentry) !=
3151 d_is_dir(dentry: old_path->dentry))
3152 goto out;
3153 /*
3154 * Don't move a mount residing in a shared parent.
3155 */
3156 if (attached && IS_MNT_SHARED(parent))
3157 goto out;
3158
3159 if (beneath) {
3160 err = can_move_mount_beneath(from: old_path, to: new_path, mp);
3161 if (err)
3162 goto out;
3163
3164 err = -EINVAL;
3165 p = p->mnt_parent;
3166 flags |= MNT_TREE_BENEATH;
3167 }
3168
3169 /*
3170 * Don't move a mount tree containing unbindable mounts to a destination
3171 * mount which is shared.
3172 */
3173 if (IS_MNT_SHARED(p) && tree_contains_unbindable(mnt: old))
3174 goto out;
3175 err = -ELOOP;
3176 if (!check_for_nsfs_mounts(subtree: old))
3177 goto out;
3178 for (; mnt_has_parent(mnt: p); p = p->mnt_parent)
3179 if (p == old)
3180 goto out;
3181
3182 err = attach_recursive_mnt(source_mnt: old, top_mnt: real_mount(mnt: new_path->mnt), dest_mp: mp, flags);
3183 if (err)
3184 goto out;
3185
3186 /* if the mount is moved, it should no longer be expire
3187 * automatically */
3188 list_del_init(entry: &old->mnt_expire);
3189 if (attached)
3190 put_mountpoint(mp: old_mp);
3191out:
3192 unlock_mount(where: mp);
3193 if (!err) {
3194 if (attached)
3195 mntput_no_expire(mnt: parent);
3196 else
3197 free_mnt_ns(ns);
3198 }
3199 return err;
3200}
3201
3202static int do_move_mount_old(struct path *path, const char *old_name)
3203{
3204 struct path old_path;
3205 int err;
3206
3207 if (!old_name || !*old_name)
3208 return -EINVAL;
3209
3210 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3211 if (err)
3212 return err;
3213
3214 err = do_move_mount(old_path: &old_path, new_path: path, beneath: false);
3215 path_put(&old_path);
3216 return err;
3217}
3218
3219/*
3220 * add a mount into a namespace's mount tree
3221 */
3222static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
3223 const struct path *path, int mnt_flags)
3224{
3225 struct mount *parent = real_mount(mnt: path->mnt);
3226
3227 mnt_flags &= ~MNT_INTERNAL_FLAGS;
3228
3229 if (unlikely(!check_mnt(parent))) {
3230 /* that's acceptable only for automounts done in private ns */
3231 if (!(mnt_flags & MNT_SHRINKABLE))
3232 return -EINVAL;
3233 /* ... and for those we'd better have mountpoint still alive */
3234 if (!parent->mnt_ns)
3235 return -EINVAL;
3236 }
3237
3238 /* Refuse the same filesystem on the same mount point */
3239 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
3240 return -EBUSY;
3241
3242 if (d_is_symlink(dentry: newmnt->mnt.mnt_root))
3243 return -EINVAL;
3244
3245 newmnt->mnt.mnt_flags = mnt_flags;
3246 return graft_tree(mnt: newmnt, p: parent, mp);
3247}
3248
3249static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3250
3251/*
3252 * Create a new mount using a superblock configuration and request it
3253 * be added to the namespace tree.
3254 */
3255static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3256 unsigned int mnt_flags)
3257{
3258 struct vfsmount *mnt;
3259 struct mountpoint *mp;
3260 struct super_block *sb = fc->root->d_sb;
3261 int error;
3262
3263 error = security_sb_kern_mount(sb);
3264 if (!error && mount_too_revealing(sb, new_mnt_flags: &mnt_flags))
3265 error = -EPERM;
3266
3267 if (unlikely(error)) {
3268 fc_drop_locked(fc);
3269 return error;
3270 }
3271
3272 up_write(sem: &sb->s_umount);
3273
3274 mnt = vfs_create_mount(fc);
3275 if (IS_ERR(ptr: mnt))
3276 return PTR_ERR(ptr: mnt);
3277
3278 mnt_warn_timestamp_expiry(mountpoint, mnt);
3279
3280 mp = lock_mount(path: mountpoint);
3281 if (IS_ERR(ptr: mp)) {
3282 mntput(mnt);
3283 return PTR_ERR(ptr: mp);
3284 }
3285 error = do_add_mount(newmnt: real_mount(mnt), mp, path: mountpoint, mnt_flags);
3286 unlock_mount(where: mp);
3287 if (error < 0)
3288 mntput(mnt);
3289 return error;
3290}
3291
3292/*
3293 * create a new mount for userspace and request it to be added into the
3294 * namespace's tree
3295 */
3296static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3297 int mnt_flags, const char *name, void *data)
3298{
3299 struct file_system_type *type;
3300 struct fs_context *fc;
3301 const char *subtype = NULL;
3302 int err = 0;
3303
3304 if (!fstype)
3305 return -EINVAL;
3306
3307 type = get_fs_type(name: fstype);
3308 if (!type)
3309 return -ENODEV;
3310
3311 if (type->fs_flags & FS_HAS_SUBTYPE) {
3312 subtype = strchr(fstype, '.');
3313 if (subtype) {
3314 subtype++;
3315 if (!*subtype) {
3316 put_filesystem(fs: type);
3317 return -EINVAL;
3318 }
3319 }
3320 }
3321
3322 fc = fs_context_for_mount(fs_type: type, sb_flags);
3323 put_filesystem(fs: type);
3324 if (IS_ERR(ptr: fc))
3325 return PTR_ERR(ptr: fc);
3326
3327 if (subtype)
3328 err = vfs_parse_fs_string(fc, key: "subtype",
3329 value: subtype, strlen(subtype));
3330 if (!err && name)
3331 err = vfs_parse_fs_string(fc, key: "source", value: name, strlen(name));
3332 if (!err)
3333 err = parse_monolithic_mount_data(fc, data);
3334 if (!err && !mount_capable(fc))
3335 err = -EPERM;
3336 if (!err)
3337 err = vfs_get_tree(fc);
3338 if (!err)
3339 err = do_new_mount_fc(fc, mountpoint: path, mnt_flags);
3340
3341 put_fs_context(fc);
3342 return err;
3343}
3344
3345int finish_automount(struct vfsmount *m, const struct path *path)
3346{
3347 struct dentry *dentry = path->dentry;
3348 struct mountpoint *mp;
3349 struct mount *mnt;
3350 int err;
3351
3352 if (!m)
3353 return 0;
3354 if (IS_ERR(ptr: m))
3355 return PTR_ERR(ptr: m);
3356
3357 mnt = real_mount(mnt: m);
3358 /* The new mount record should have at least 2 refs to prevent it being
3359 * expired before we get a chance to add it
3360 */
3361 BUG_ON(mnt_get_count(mnt) < 2);
3362
3363 if (m->mnt_sb == path->mnt->mnt_sb &&
3364 m->mnt_root == dentry) {
3365 err = -ELOOP;
3366 goto discard;
3367 }
3368
3369 /*
3370 * we don't want to use lock_mount() - in this case finding something
3371 * that overmounts our mountpoint to be means "quitely drop what we've
3372 * got", not "try to mount it on top".
3373 */
3374 inode_lock(inode: dentry->d_inode);
3375 namespace_lock();
3376 if (unlikely(cant_mount(dentry))) {
3377 err = -ENOENT;
3378 goto discard_locked;
3379 }
3380 if (path_overmounted(path)) {
3381 err = 0;
3382 goto discard_locked;
3383 }
3384 mp = get_mountpoint(dentry);
3385 if (IS_ERR(ptr: mp)) {
3386 err = PTR_ERR(ptr: mp);
3387 goto discard_locked;
3388 }
3389
3390 err = do_add_mount(newmnt: mnt, mp, path, mnt_flags: path->mnt->mnt_flags | MNT_SHRINKABLE);
3391 unlock_mount(where: mp);
3392 if (unlikely(err))
3393 goto discard;
3394 mntput(m);
3395 return 0;
3396
3397discard_locked:
3398 namespace_unlock();
3399 inode_unlock(inode: dentry->d_inode);
3400discard:
3401 /* remove m from any expiration list it may be on */
3402 if (!list_empty(head: &mnt->mnt_expire)) {
3403 namespace_lock();
3404 list_del_init(entry: &mnt->mnt_expire);
3405 namespace_unlock();
3406 }
3407 mntput(m);
3408 mntput(m);
3409 return err;
3410}
3411
3412/**
3413 * mnt_set_expiry - Put a mount on an expiration list
3414 * @mnt: The mount to list.
3415 * @expiry_list: The list to add the mount to.
3416 */
3417void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3418{
3419 namespace_lock();
3420
3421 list_add_tail(new: &real_mount(mnt)->mnt_expire, head: expiry_list);
3422
3423 namespace_unlock();
3424}
3425EXPORT_SYMBOL(mnt_set_expiry);
3426
3427/*
3428 * process a list of expirable mountpoints with the intent of discarding any
3429 * mountpoints that aren't in use and haven't been touched since last we came
3430 * here
3431 */
3432void mark_mounts_for_expiry(struct list_head *mounts)
3433{
3434 struct mount *mnt, *next;
3435 LIST_HEAD(graveyard);
3436
3437 if (list_empty(head: mounts))
3438 return;
3439
3440 namespace_lock();
3441 lock_mount_hash();
3442
3443 /* extract from the expiration list every vfsmount that matches the
3444 * following criteria:
3445 * - only referenced by its parent vfsmount
3446 * - still marked for expiry (marked on the last call here; marks are
3447 * cleared by mntput())
3448 */
3449 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3450 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3451 propagate_mount_busy(mnt, 1))
3452 continue;
3453 list_move(list: &mnt->mnt_expire, head: &graveyard);
3454 }
3455 while (!list_empty(head: &graveyard)) {
3456 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3457 touch_mnt_namespace(ns: mnt->mnt_ns);
3458 umount_tree(mnt, how: UMOUNT_PROPAGATE|UMOUNT_SYNC);
3459 }
3460 unlock_mount_hash();
3461 namespace_unlock();
3462}
3463
3464EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3465
3466/*
3467 * Ripoff of 'select_parent()'
3468 *
3469 * search the list of submounts for a given mountpoint, and move any
3470 * shrinkable submounts to the 'graveyard' list.
3471 */
3472static int select_submounts(struct mount *parent, struct list_head *graveyard)
3473{
3474 struct mount *this_parent = parent;
3475 struct list_head *next;
3476 int found = 0;
3477
3478repeat:
3479 next = this_parent->mnt_mounts.next;
3480resume:
3481 while (next != &this_parent->mnt_mounts) {
3482 struct list_head *tmp = next;
3483 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3484
3485 next = tmp->next;
3486 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3487 continue;
3488 /*
3489 * Descend a level if the d_mounts list is non-empty.
3490 */
3491 if (!list_empty(head: &mnt->mnt_mounts)) {
3492 this_parent = mnt;
3493 goto repeat;
3494 }
3495
3496 if (!propagate_mount_busy(mnt, 1)) {
3497 list_move_tail(list: &mnt->mnt_expire, head: graveyard);
3498 found++;
3499 }
3500 }
3501 /*
3502 * All done at this level ... ascend and resume the search
3503 */
3504 if (this_parent != parent) {
3505 next = this_parent->mnt_child.next;
3506 this_parent = this_parent->mnt_parent;
3507 goto resume;
3508 }
3509 return found;
3510}
3511
3512/*
3513 * process a list of expirable mountpoints with the intent of discarding any
3514 * submounts of a specific parent mountpoint
3515 *
3516 * mount_lock must be held for write
3517 */
3518static void shrink_submounts(struct mount *mnt)
3519{
3520 LIST_HEAD(graveyard);
3521 struct mount *m;
3522
3523 /* extract submounts of 'mountpoint' from the expiration list */
3524 while (select_submounts(parent: mnt, graveyard: &graveyard)) {
3525 while (!list_empty(head: &graveyard)) {
3526 m = list_first_entry(&graveyard, struct mount,
3527 mnt_expire);
3528 touch_mnt_namespace(ns: m->mnt_ns);
3529 umount_tree(mnt: m, how: UMOUNT_PROPAGATE|UMOUNT_SYNC);
3530 }
3531 }
3532}
3533
3534static void *copy_mount_options(const void __user * data)
3535{
3536 char *copy;
3537 unsigned left, offset;
3538
3539 if (!data)
3540 return NULL;
3541
3542 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3543 if (!copy)
3544 return ERR_PTR(error: -ENOMEM);
3545
3546 left = copy_from_user(to: copy, from: data, PAGE_SIZE);
3547
3548 /*
3549 * Not all architectures have an exact copy_from_user(). Resort to
3550 * byte at a time.
3551 */
3552 offset = PAGE_SIZE - left;
3553 while (left) {
3554 char c;
3555 if (get_user(c, (const char __user *)data + offset))
3556 break;
3557 copy[offset] = c;
3558 left--;
3559 offset++;
3560 }
3561
3562 if (left == PAGE_SIZE) {
3563 kfree(objp: copy);
3564 return ERR_PTR(error: -EFAULT);
3565 }
3566
3567 return copy;
3568}
3569
3570static char *copy_mount_string(const void __user *data)
3571{
3572 return data ? strndup_user(data, PATH_MAX) : NULL;
3573}
3574
3575/*
3576 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3577 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3578 *
3579 * data is a (void *) that can point to any structure up to
3580 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3581 * information (or be NULL).
3582 *
3583 * Pre-0.97 versions of mount() didn't have a flags word.
3584 * When the flags word was introduced its top half was required
3585 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3586 * Therefore, if this magic number is present, it carries no information
3587 * and must be discarded.
3588 */
3589int path_mount(const char *dev_name, struct path *path,
3590 const char *type_page, unsigned long flags, void *data_page)
3591{
3592 unsigned int mnt_flags = 0, sb_flags;
3593 int ret;
3594
3595 /* Discard magic */
3596 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3597 flags &= ~MS_MGC_MSK;
3598
3599 /* Basic sanity checks */
3600 if (data_page)
3601 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3602
3603 if (flags & MS_NOUSER)
3604 return -EINVAL;
3605
3606 ret = security_sb_mount(dev_name, path, type: type_page, flags, data: data_page);
3607 if (ret)
3608 return ret;
3609 if (!may_mount())
3610 return -EPERM;
3611 if (flags & SB_MANDLOCK)
3612 warn_mandlock();
3613
3614 /* Default to relatime unless overriden */
3615 if (!(flags & MS_NOATIME))
3616 mnt_flags |= MNT_RELATIME;
3617
3618 /* Separate the per-mountpoint flags */
3619 if (flags & MS_NOSUID)
3620 mnt_flags |= MNT_NOSUID;
3621 if (flags & MS_NODEV)
3622 mnt_flags |= MNT_NODEV;
3623 if (flags & MS_NOEXEC)
3624 mnt_flags |= MNT_NOEXEC;
3625 if (flags & MS_NOATIME)
3626 mnt_flags |= MNT_NOATIME;
3627 if (flags & MS_NODIRATIME)
3628 mnt_flags |= MNT_NODIRATIME;
3629 if (flags & MS_STRICTATIME)
3630 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3631 if (flags & MS_RDONLY)
3632 mnt_flags |= MNT_READONLY;
3633 if (flags & MS_NOSYMFOLLOW)
3634 mnt_flags |= MNT_NOSYMFOLLOW;
3635
3636 /* The default atime for remount is preservation */
3637 if ((flags & MS_REMOUNT) &&
3638 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3639 MS_STRICTATIME)) == 0)) {
3640 mnt_flags &= ~MNT_ATIME_MASK;
3641 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3642 }
3643
3644 sb_flags = flags & (SB_RDONLY |
3645 SB_SYNCHRONOUS |
3646 SB_MANDLOCK |
3647 SB_DIRSYNC |
3648 SB_SILENT |
3649 SB_POSIXACL |
3650 SB_LAZYTIME |
3651 SB_I_VERSION);
3652
3653 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3654 return do_reconfigure_mnt(path, mnt_flags);
3655 if (flags & MS_REMOUNT)
3656 return do_remount(path, ms_flags: flags, sb_flags, mnt_flags, data: data_page);
3657 if (flags & MS_BIND)
3658 return do_loopback(path, old_name: dev_name, recurse: flags & MS_REC);
3659 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3660 return do_change_type(path, ms_flags: flags);
3661 if (flags & MS_MOVE)
3662 return do_move_mount_old(path, old_name: dev_name);
3663
3664 return do_new_mount(path, fstype: type_page, sb_flags, mnt_flags, name: dev_name,
3665 data: data_page);
3666}
3667
3668long do_mount(const char *dev_name, const char __user *dir_name,
3669 const char *type_page, unsigned long flags, void *data_page)
3670{
3671 struct path path;
3672 int ret;
3673
3674 ret = user_path_at(AT_FDCWD, name: dir_name, LOOKUP_FOLLOW, path: &path);
3675 if (ret)
3676 return ret;
3677 ret = path_mount(dev_name, path: &path, type_page, flags, data_page);
3678 path_put(&path);
3679 return ret;
3680}
3681
3682static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3683{
3684 return inc_ucount(ns, current_euid(), type: UCOUNT_MNT_NAMESPACES);
3685}
3686
3687static void dec_mnt_namespaces(struct ucounts *ucounts)
3688{
3689 dec_ucount(ucounts, type: UCOUNT_MNT_NAMESPACES);
3690}
3691
3692static void free_mnt_ns(struct mnt_namespace *ns)
3693{
3694 if (!is_anon_ns(ns))
3695 ns_free_inum(&ns->ns);
3696 dec_mnt_namespaces(ucounts: ns->ucounts);
3697 put_user_ns(ns: ns->user_ns);
3698 kfree(objp: ns);
3699}
3700
3701/*
3702 * Assign a sequence number so we can detect when we attempt to bind
3703 * mount a reference to an older mount namespace into the current
3704 * mount namespace, preventing reference counting loops. A 64bit
3705 * number incrementing at 10Ghz will take 12,427 years to wrap which
3706 * is effectively never, so we can ignore the possibility.
3707 */
3708static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3709
3710static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3711{
3712 struct mnt_namespace *new_ns;
3713 struct ucounts *ucounts;
3714 int ret;
3715
3716 ucounts = inc_mnt_namespaces(ns: user_ns);
3717 if (!ucounts)
3718 return ERR_PTR(error: -ENOSPC);
3719
3720 new_ns = kzalloc(size: sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3721 if (!new_ns) {
3722 dec_mnt_namespaces(ucounts);
3723 return ERR_PTR(error: -ENOMEM);
3724 }
3725 if (!anon) {
3726 ret = ns_alloc_inum(ns: &new_ns->ns);
3727 if (ret) {
3728 kfree(objp: new_ns);
3729 dec_mnt_namespaces(ucounts);
3730 return ERR_PTR(error: ret);
3731 }
3732 }
3733 new_ns->ns.ops = &mntns_operations;
3734 if (!anon)
3735 new_ns->seq = atomic64_add_return(i: 1, v: &mnt_ns_seq);
3736 refcount_set(r: &new_ns->ns.count, n: 1);
3737 INIT_LIST_HEAD(list: &new_ns->list);
3738 init_waitqueue_head(&new_ns->poll);
3739 spin_lock_init(&new_ns->ns_lock);
3740 new_ns->user_ns = get_user_ns(ns: user_ns);
3741 new_ns->ucounts = ucounts;
3742 return new_ns;
3743}
3744
3745__latent_entropy
3746struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3747 struct user_namespace *user_ns, struct fs_struct *new_fs)
3748{
3749 struct mnt_namespace *new_ns;
3750 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3751 struct mount *p, *q;
3752 struct mount *old;
3753 struct mount *new;
3754 int copy_flags;
3755
3756 BUG_ON(!ns);
3757
3758 if (likely(!(flags & CLONE_NEWNS))) {
3759 get_mnt_ns(ns);
3760 return ns;
3761 }
3762
3763 old = ns->root;
3764
3765 new_ns = alloc_mnt_ns(user_ns, anon: false);
3766 if (IS_ERR(ptr: new_ns))
3767 return new_ns;
3768
3769 namespace_lock();
3770 /* First pass: copy the tree topology */
3771 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3772 if (user_ns != ns->user_ns)
3773 copy_flags |= CL_SHARED_TO_SLAVE;
3774 new = copy_tree(mnt: old, dentry: old->mnt.mnt_root, flag: copy_flags);
3775 if (IS_ERR(ptr: new)) {
3776 namespace_unlock();
3777 free_mnt_ns(ns: new_ns);
3778 return ERR_CAST(ptr: new);
3779 }
3780 if (user_ns != ns->user_ns) {
3781 lock_mount_hash();
3782 lock_mnt_tree(mnt: new);
3783 unlock_mount_hash();
3784 }
3785 new_ns->root = new;
3786 list_add_tail(new: &new_ns->list, head: &new->mnt_list);
3787
3788 /*
3789 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3790 * as belonging to new namespace. We have already acquired a private
3791 * fs_struct, so tsk->fs->lock is not needed.
3792 */
3793 p = old;
3794 q = new;
3795 while (p) {
3796 q->mnt_ns = new_ns;
3797 new_ns->mounts++;
3798 if (new_fs) {
3799 if (&p->mnt == new_fs->root.mnt) {
3800 new_fs->root.mnt = mntget(&q->mnt);
3801 rootmnt = &p->mnt;
3802 }
3803 if (&p->mnt == new_fs->pwd.mnt) {
3804 new_fs->pwd.mnt = mntget(&q->mnt);
3805 pwdmnt = &p->mnt;
3806 }
3807 }
3808 p = next_mnt(p, root: old);
3809 q = next_mnt(p: q, root: new);
3810 if (!q)
3811 break;
3812 // an mntns binding we'd skipped?
3813 while (p->mnt.mnt_root != q->mnt.mnt_root)
3814 p = next_mnt(p: skip_mnt_tree(p), root: old);
3815 }
3816 namespace_unlock();
3817
3818 if (rootmnt)
3819 mntput(rootmnt);
3820 if (pwdmnt)
3821 mntput(pwdmnt);
3822
3823 return new_ns;
3824}
3825
3826struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3827{
3828 struct mount *mnt = real_mount(mnt: m);
3829 struct mnt_namespace *ns;
3830 struct super_block *s;
3831 struct path path;
3832 int err;
3833
3834 ns = alloc_mnt_ns(user_ns: &init_user_ns, anon: true);
3835 if (IS_ERR(ptr: ns)) {
3836 mntput(m);
3837 return ERR_CAST(ptr: ns);
3838 }
3839 mnt->mnt_ns = ns;
3840 ns->root = mnt;
3841 ns->mounts++;
3842 list_add(new: &mnt->mnt_list, head: &ns->list);
3843
3844 err = vfs_path_lookup(m->mnt_root, m,
3845 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3846
3847 put_mnt_ns(ns);
3848
3849 if (err)
3850 return ERR_PTR(error: err);
3851
3852 /* trade a vfsmount reference for active sb one */
3853 s = path.mnt->mnt_sb;
3854 atomic_inc(v: &s->s_active);
3855 mntput(path.mnt);
3856 /* lock the sucker */
3857 down_write(sem: &s->s_umount);
3858 /* ... and return the root of (sub)tree on it */
3859 return path.dentry;
3860}
3861EXPORT_SYMBOL(mount_subtree);
3862
3863SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3864 char __user *, type, unsigned long, flags, void __user *, data)
3865{
3866 int ret;
3867 char *kernel_type;
3868 char *kernel_dev;
3869 void *options;
3870
3871 kernel_type = copy_mount_string(data: type);
3872 ret = PTR_ERR(ptr: kernel_type);
3873 if (IS_ERR(ptr: kernel_type))
3874 goto out_type;
3875
3876 kernel_dev = copy_mount_string(data: dev_name);
3877 ret = PTR_ERR(ptr: kernel_dev);
3878 if (IS_ERR(ptr: kernel_dev))
3879 goto out_dev;
3880
3881 options = copy_mount_options(data);
3882 ret = PTR_ERR(ptr: options);
3883 if (IS_ERR(ptr: options))
3884 goto out_data;
3885
3886 ret = do_mount(dev_name: kernel_dev, dir_name, type_page: kernel_type, flags, data_page: options);
3887
3888 kfree(objp: options);
3889out_data:
3890 kfree(objp: kernel_dev);
3891out_dev:
3892 kfree(objp: kernel_type);
3893out_type:
3894 return ret;
3895}
3896
3897#define FSMOUNT_VALID_FLAGS \
3898 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3899 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3900 MOUNT_ATTR_NOSYMFOLLOW)
3901
3902#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3903
3904#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3905 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3906
3907static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3908{
3909 unsigned int mnt_flags = 0;
3910
3911 if (attr_flags & MOUNT_ATTR_RDONLY)
3912 mnt_flags |= MNT_READONLY;
3913 if (attr_flags & MOUNT_ATTR_NOSUID)
3914 mnt_flags |= MNT_NOSUID;
3915 if (attr_flags & MOUNT_ATTR_NODEV)
3916 mnt_flags |= MNT_NODEV;
3917 if (attr_flags & MOUNT_ATTR_NOEXEC)
3918 mnt_flags |= MNT_NOEXEC;
3919 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3920 mnt_flags |= MNT_NODIRATIME;
3921 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3922 mnt_flags |= MNT_NOSYMFOLLOW;
3923
3924 return mnt_flags;
3925}
3926
3927/*
3928 * Create a kernel mount representation for a new, prepared superblock
3929 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3930 */
3931SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3932 unsigned int, attr_flags)
3933{
3934 struct mnt_namespace *ns;
3935 struct fs_context *fc;
3936 struct file *file;
3937 struct path newmount;
3938 struct mount *mnt;
3939 struct fd f;
3940 unsigned int mnt_flags = 0;
3941 long ret;
3942
3943 if (!may_mount())
3944 return -EPERM;
3945
3946 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3947 return -EINVAL;
3948
3949 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3950 return -EINVAL;
3951
3952 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3953
3954 switch (attr_flags & MOUNT_ATTR__ATIME) {
3955 case MOUNT_ATTR_STRICTATIME:
3956 break;
3957 case MOUNT_ATTR_NOATIME:
3958 mnt_flags |= MNT_NOATIME;
3959 break;
3960 case MOUNT_ATTR_RELATIME:
3961 mnt_flags |= MNT_RELATIME;
3962 break;
3963 default:
3964 return -EINVAL;
3965 }
3966
3967 f = fdget(fd: fs_fd);
3968 if (!f.file)
3969 return -EBADF;
3970
3971 ret = -EINVAL;
3972 if (f.file->f_op != &fscontext_fops)
3973 goto err_fsfd;
3974
3975 fc = f.file->private_data;
3976
3977 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3978 if (ret < 0)
3979 goto err_fsfd;
3980
3981 /* There must be a valid superblock or we can't mount it */
3982 ret = -EINVAL;
3983 if (!fc->root)
3984 goto err_unlock;
3985
3986 ret = -EPERM;
3987 if (mount_too_revealing(sb: fc->root->d_sb, new_mnt_flags: &mnt_flags)) {
3988 pr_warn("VFS: Mount too revealing\n");
3989 goto err_unlock;
3990 }
3991
3992 ret = -EBUSY;
3993 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3994 goto err_unlock;
3995
3996 if (fc->sb_flags & SB_MANDLOCK)
3997 warn_mandlock();
3998
3999 newmount.mnt = vfs_create_mount(fc);
4000 if (IS_ERR(ptr: newmount.mnt)) {
4001 ret = PTR_ERR(ptr: newmount.mnt);
4002 goto err_unlock;
4003 }
4004 newmount.dentry = dget(dentry: fc->root);
4005 newmount.mnt->mnt_flags = mnt_flags;
4006
4007 /* We've done the mount bit - now move the file context into more or
4008 * less the same state as if we'd done an fspick(). We don't want to
4009 * do any memory allocation or anything like that at this point as we
4010 * don't want to have to handle any errors incurred.
4011 */
4012 vfs_clean_context(fc);
4013
4014 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, anon: true);
4015 if (IS_ERR(ptr: ns)) {
4016 ret = PTR_ERR(ptr: ns);
4017 goto err_path;
4018 }
4019 mnt = real_mount(mnt: newmount.mnt);
4020 mnt->mnt_ns = ns;
4021 ns->root = mnt;
4022 ns->mounts = 1;
4023 list_add(new: &mnt->mnt_list, head: &ns->list);
4024 mntget(newmount.mnt);
4025
4026 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4027 * it, not just simply put it.
4028 */
4029 file = dentry_open(path: &newmount, O_PATH, creds: fc->cred);
4030 if (IS_ERR(ptr: file)) {
4031 dissolve_on_fput(mnt: newmount.mnt);
4032 ret = PTR_ERR(ptr: file);
4033 goto err_path;
4034 }
4035 file->f_mode |= FMODE_NEED_UNMOUNT;
4036
4037 ret = get_unused_fd_flags(flags: (flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
4038 if (ret >= 0)
4039 fd_install(fd: ret, file);
4040 else
4041 fput(file);
4042
4043err_path:
4044 path_put(&newmount);
4045err_unlock:
4046 mutex_unlock(lock: &fc->uapi_mutex);
4047err_fsfd:
4048 fdput(fd: f);
4049 return ret;
4050}
4051
4052/*
4053 * Move a mount from one place to another. In combination with
4054 * fsopen()/fsmount() this is used to install a new mount and in combination
4055 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4056 * a mount subtree.
4057 *
4058 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4059 */
4060SYSCALL_DEFINE5(move_mount,
4061 int, from_dfd, const char __user *, from_pathname,
4062 int, to_dfd, const char __user *, to_pathname,
4063 unsigned int, flags)
4064{
4065 struct path from_path, to_path;
4066 unsigned int lflags;
4067 int ret = 0;
4068
4069 if (!may_mount())
4070 return -EPERM;
4071
4072 if (flags & ~MOVE_MOUNT__MASK)
4073 return -EINVAL;
4074
4075 if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) ==
4076 (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP))
4077 return -EINVAL;
4078
4079 /* If someone gives a pathname, they aren't permitted to move
4080 * from an fd that requires unmount as we can't get at the flag
4081 * to clear it afterwards.
4082 */
4083 lflags = 0;
4084 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4085 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4086 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4087
4088 ret = user_path_at(dfd: from_dfd, name: from_pathname, flags: lflags, path: &from_path);
4089 if (ret < 0)
4090 return ret;
4091
4092 lflags = 0;
4093 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4094 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4095 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4096
4097 ret = user_path_at(dfd: to_dfd, name: to_pathname, flags: lflags, path: &to_path);
4098 if (ret < 0)
4099 goto out_from;
4100
4101 ret = security_move_mount(from_path: &from_path, to_path: &to_path);
4102 if (ret < 0)
4103 goto out_to;
4104
4105 if (flags & MOVE_MOUNT_SET_GROUP)
4106 ret = do_set_group(from_path: &from_path, to_path: &to_path);
4107 else
4108 ret = do_move_mount(old_path: &from_path, new_path: &to_path,
4109 beneath: (flags & MOVE_MOUNT_BENEATH));
4110
4111out_to:
4112 path_put(&to_path);
4113out_from:
4114 path_put(&from_path);
4115 return ret;
4116}
4117
4118/*
4119 * Return true if path is reachable from root
4120 *
4121 * namespace_sem or mount_lock is held
4122 */
4123bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
4124 const struct path *root)
4125{
4126 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
4127 dentry = mnt->mnt_mountpoint;
4128 mnt = mnt->mnt_parent;
4129 }
4130 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
4131}
4132
4133bool path_is_under(const struct path *path1, const struct path *path2)
4134{
4135 bool res;
4136 read_seqlock_excl(sl: &mount_lock);
4137 res = is_path_reachable(mnt: real_mount(mnt: path1->mnt), dentry: path1->dentry, root: path2);
4138 read_sequnlock_excl(sl: &mount_lock);
4139 return res;
4140}
4141EXPORT_SYMBOL(path_is_under);
4142
4143/*
4144 * pivot_root Semantics:
4145 * Moves the root file system of the current process to the directory put_old,
4146 * makes new_root as the new root file system of the current process, and sets
4147 * root/cwd of all processes which had them on the current root to new_root.
4148 *
4149 * Restrictions:
4150 * The new_root and put_old must be directories, and must not be on the
4151 * same file system as the current process root. The put_old must be
4152 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4153 * pointed to by put_old must yield the same directory as new_root. No other
4154 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4155 *
4156 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
4157 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4158 * in this situation.
4159 *
4160 * Notes:
4161 * - we don't move root/cwd if they are not at the root (reason: if something
4162 * cared enough to change them, it's probably wrong to force them elsewhere)
4163 * - it's okay to pick a root that isn't the root of a file system, e.g.
4164 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4165 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4166 * first.
4167 */
4168SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
4169 const char __user *, put_old)
4170{
4171 struct path new, old, root;
4172 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
4173 struct mountpoint *old_mp, *root_mp;
4174 int error;
4175
4176 if (!may_mount())
4177 return -EPERM;
4178
4179 error = user_path_at(AT_FDCWD, name: new_root,
4180 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, path: &new);
4181 if (error)
4182 goto out0;
4183
4184 error = user_path_at(AT_FDCWD, name: put_old,
4185 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, path: &old);
4186 if (error)
4187 goto out1;
4188
4189 error = security_sb_pivotroot(old_path: &old, new_path: &new);
4190 if (error)
4191 goto out2;
4192
4193 get_fs_root(current->fs, root: &root);
4194 old_mp = lock_mount(path: &old);
4195 error = PTR_ERR(ptr: old_mp);
4196 if (IS_ERR(ptr: old_mp))
4197 goto out3;
4198
4199 error = -EINVAL;
4200 new_mnt = real_mount(mnt: new.mnt);
4201 root_mnt = real_mount(mnt: root.mnt);
4202 old_mnt = real_mount(mnt: old.mnt);
4203 ex_parent = new_mnt->mnt_parent;
4204 root_parent = root_mnt->mnt_parent;
4205 if (IS_MNT_SHARED(old_mnt) ||
4206 IS_MNT_SHARED(ex_parent) ||
4207 IS_MNT_SHARED(root_parent))
4208 goto out4;
4209 if (!check_mnt(mnt: root_mnt) || !check_mnt(mnt: new_mnt))
4210 goto out4;
4211 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4212 goto out4;
4213 error = -ENOENT;
4214 if (d_unlinked(dentry: new.dentry))
4215 goto out4;
4216 error = -EBUSY;
4217 if (new_mnt == root_mnt || old_mnt == root_mnt)
4218 goto out4; /* loop, on the same file system */
4219 error = -EINVAL;
4220 if (!path_mounted(path: &root))
4221 goto out4; /* not a mountpoint */
4222 if (!mnt_has_parent(mnt: root_mnt))
4223 goto out4; /* not attached */
4224 if (!path_mounted(path: &new))
4225 goto out4; /* not a mountpoint */
4226 if (!mnt_has_parent(mnt: new_mnt))
4227 goto out4; /* not attached */
4228 /* make sure we can reach put_old from new_root */
4229 if (!is_path_reachable(mnt: old_mnt, dentry: old.dentry, root: &new))
4230 goto out4;
4231 /* make certain new is below the root */
4232 if (!is_path_reachable(mnt: new_mnt, dentry: new.dentry, root: &root))
4233 goto out4;
4234 lock_mount_hash();
4235 umount_mnt(mnt: new_mnt);
4236 root_mp = unhash_mnt(mnt: root_mnt); /* we'll need its mountpoint */
4237 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4238 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4239 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4240 }
4241 /* mount old root on put_old */
4242 attach_mnt(mnt: root_mnt, parent: old_mnt, mp: old_mp, beneath: false);
4243 /* mount new_root on / */
4244 attach_mnt(mnt: new_mnt, parent: root_parent, mp: root_mp, beneath: false);
4245 mnt_add_count(mnt: root_parent, n: -1);
4246 touch_mnt_namespace(current->nsproxy->mnt_ns);
4247 /* A moved mount should not expire automatically */
4248 list_del_init(entry: &new_mnt->mnt_expire);
4249 put_mountpoint(mp: root_mp);
4250 unlock_mount_hash();
4251 chroot_fs_refs(&root, &new);
4252 error = 0;
4253out4:
4254 unlock_mount(where: old_mp);
4255 if (!error)
4256 mntput_no_expire(mnt: ex_parent);
4257out3:
4258 path_put(&root);
4259out2:
4260 path_put(&old);
4261out1:
4262 path_put(&new);
4263out0:
4264 return error;
4265}
4266
4267static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4268{
4269 unsigned int flags = mnt->mnt.mnt_flags;
4270
4271 /* flags to clear */
4272 flags &= ~kattr->attr_clr;
4273 /* flags to raise */
4274 flags |= kattr->attr_set;
4275
4276 return flags;
4277}
4278
4279static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4280{
4281 struct vfsmount *m = &mnt->mnt;
4282 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4283
4284 if (!kattr->mnt_idmap)
4285 return 0;
4286
4287 /*
4288 * Creating an idmapped mount with the filesystem wide idmapping
4289 * doesn't make sense so block that. We don't allow mushy semantics.
4290 */
4291 if (!check_fsmapping(idmap: kattr->mnt_idmap, sb: m->mnt_sb))
4292 return -EINVAL;
4293
4294 /*
4295 * Once a mount has been idmapped we don't allow it to change its
4296 * mapping. It makes things simpler and callers can just create
4297 * another bind-mount they can idmap if they want to.
4298 */
4299 if (is_idmapped_mnt(mnt: m))
4300 return -EPERM;
4301
4302 /* The underlying filesystem doesn't support idmapped mounts yet. */
4303 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4304 return -EINVAL;
4305
4306 /* We're not controlling the superblock. */
4307 if (!ns_capable(ns: fs_userns, CAP_SYS_ADMIN))
4308 return -EPERM;
4309
4310 /* Mount has already been visible in the filesystem hierarchy. */
4311 if (!is_anon_ns(ns: mnt->mnt_ns))
4312 return -EINVAL;
4313
4314 return 0;
4315}
4316
4317/**
4318 * mnt_allow_writers() - check whether the attribute change allows writers
4319 * @kattr: the new mount attributes
4320 * @mnt: the mount to which @kattr will be applied
4321 *
4322 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4323 *
4324 * Return: true if writers need to be held, false if not
4325 */
4326static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4327 const struct mount *mnt)
4328{
4329 return (!(kattr->attr_set & MNT_READONLY) ||
4330 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4331 !kattr->mnt_idmap;
4332}
4333
4334static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4335{
4336 struct mount *m;
4337 int err;
4338
4339 for (m = mnt; m; m = next_mnt(p: m, root: mnt)) {
4340 if (!can_change_locked_flags(mnt: m, mnt_flags: recalc_flags(kattr, mnt: m))) {
4341 err = -EPERM;
4342 break;
4343 }
4344
4345 err = can_idmap_mount(kattr, mnt: m);
4346 if (err)
4347 break;
4348
4349 if (!mnt_allow_writers(kattr, mnt: m)) {
4350 err = mnt_hold_writers(mnt: m);
4351 if (err)
4352 break;
4353 }
4354
4355 if (!kattr->recurse)
4356 return 0;
4357 }
4358
4359 if (err) {
4360 struct mount *p;
4361
4362 /*
4363 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4364 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4365 * mounts and needs to take care to include the first mount.
4366 */
4367 for (p = mnt; p; p = next_mnt(p, root: mnt)) {
4368 /* If we had to hold writers unblock them. */
4369 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4370 mnt_unhold_writers(mnt: p);
4371
4372 /*
4373 * We're done once the first mount we changed got
4374 * MNT_WRITE_HOLD unset.
4375 */
4376 if (p == m)
4377 break;
4378 }
4379 }
4380 return err;
4381}
4382
4383static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4384{
4385 if (!kattr->mnt_idmap)
4386 return;
4387
4388 /*
4389 * Pairs with smp_load_acquire() in mnt_idmap().
4390 *
4391 * Since we only allow a mount to change the idmapping once and
4392 * verified this in can_idmap_mount() we know that the mount has
4393 * @nop_mnt_idmap attached to it. So there's no need to drop any
4394 * references.
4395 */
4396 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4397}
4398
4399static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4400{
4401 struct mount *m;
4402
4403 for (m = mnt; m; m = next_mnt(p: m, root: mnt)) {
4404 unsigned int flags;
4405
4406 do_idmap_mount(kattr, mnt: m);
4407 flags = recalc_flags(kattr, mnt: m);
4408 WRITE_ONCE(m->mnt.mnt_flags, flags);
4409
4410 /* If we had to hold writers unblock them. */
4411 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4412 mnt_unhold_writers(mnt: m);
4413
4414 if (kattr->propagation)
4415 change_mnt_propagation(m, kattr->propagation);
4416 if (!kattr->recurse)
4417 break;
4418 }
4419 touch_mnt_namespace(ns: mnt->mnt_ns);
4420}
4421
4422static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4423{
4424 struct mount *mnt = real_mount(mnt: path->mnt);
4425 int err = 0;
4426
4427 if (!path_mounted(path))
4428 return -EINVAL;
4429
4430 if (kattr->mnt_userns) {
4431 struct mnt_idmap *mnt_idmap;
4432
4433 mnt_idmap = alloc_mnt_idmap(mnt_userns: kattr->mnt_userns);
4434 if (IS_ERR(ptr: mnt_idmap))
4435 return PTR_ERR(ptr: mnt_idmap);
4436 kattr->mnt_idmap = mnt_idmap;
4437 }
4438
4439 if (kattr->propagation) {
4440 /*
4441 * Only take namespace_lock() if we're actually changing
4442 * propagation.
4443 */
4444 namespace_lock();
4445 if (kattr->propagation == MS_SHARED) {
4446 err = invent_group_ids(mnt, recurse: kattr->recurse);
4447 if (err) {
4448 namespace_unlock();
4449 return err;
4450 }
4451 }
4452 }
4453
4454 err = -EINVAL;
4455 lock_mount_hash();
4456
4457 /* Ensure that this isn't anything purely vfs internal. */
4458 if (!is_mounted(mnt: &mnt->mnt))
4459 goto out;
4460
4461 /*
4462 * If this is an attached mount make sure it's located in the callers
4463 * mount namespace. If it's not don't let the caller interact with it.
4464 * If this is a detached mount make sure it has an anonymous mount
4465 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4466 */
4467 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(ns: mnt->mnt_ns)))
4468 goto out;
4469
4470 /*
4471 * First, we get the mount tree in a shape where we can change mount
4472 * properties without failure. If we succeeded to do so we commit all
4473 * changes and if we failed we clean up.
4474 */
4475 err = mount_setattr_prepare(kattr, mnt);
4476 if (!err)
4477 mount_setattr_commit(kattr, mnt);
4478
4479out:
4480 unlock_mount_hash();
4481
4482 if (kattr->propagation) {
4483 if (err)
4484 cleanup_group_ids(mnt, NULL);
4485 namespace_unlock();
4486 }
4487
4488 return err;
4489}
4490
4491static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4492 struct mount_kattr *kattr, unsigned int flags)
4493{
4494 int err = 0;
4495 struct ns_common *ns;
4496 struct user_namespace *mnt_userns;
4497 struct fd f;
4498
4499 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4500 return 0;
4501
4502 /*
4503 * We currently do not support clearing an idmapped mount. If this ever
4504 * is a use-case we can revisit this but for now let's keep it simple
4505 * and not allow it.
4506 */
4507 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4508 return -EINVAL;
4509
4510 if (attr->userns_fd > INT_MAX)
4511 return -EINVAL;
4512
4513 f = fdget(fd: attr->userns_fd);
4514 if (!f.file)
4515 return -EBADF;
4516
4517 if (!proc_ns_file(file: f.file)) {
4518 err = -EINVAL;
4519 goto out_fput;
4520 }
4521
4522 ns = get_proc_ns(file_inode(f.file));
4523 if (ns->ops->type != CLONE_NEWUSER) {
4524 err = -EINVAL;
4525 goto out_fput;
4526 }
4527
4528 /*
4529 * The initial idmapping cannot be used to create an idmapped
4530 * mount. We use the initial idmapping as an indicator of a mount
4531 * that is not idmapped. It can simply be passed into helpers that
4532 * are aware of idmapped mounts as a convenient shortcut. A user
4533 * can just create a dedicated identity mapping to achieve the same
4534 * result.
4535 */
4536 mnt_userns = container_of(ns, struct user_namespace, ns);
4537 if (mnt_userns == &init_user_ns) {
4538 err = -EPERM;
4539 goto out_fput;
4540 }
4541
4542 /* We're not controlling the target namespace. */
4543 if (!ns_capable(ns: mnt_userns, CAP_SYS_ADMIN)) {
4544 err = -EPERM;
4545 goto out_fput;
4546 }
4547
4548 kattr->mnt_userns = get_user_ns(ns: mnt_userns);
4549
4550out_fput:
4551 fdput(fd: f);
4552 return err;
4553}
4554
4555static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4556 struct mount_kattr *kattr, unsigned int flags)
4557{
4558 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4559
4560 if (flags & AT_NO_AUTOMOUNT)
4561 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4562 if (flags & AT_SYMLINK_NOFOLLOW)
4563 lookup_flags &= ~LOOKUP_FOLLOW;
4564 if (flags & AT_EMPTY_PATH)
4565 lookup_flags |= LOOKUP_EMPTY;
4566
4567 *kattr = (struct mount_kattr) {
4568 .lookup_flags = lookup_flags,
4569 .recurse = !!(flags & AT_RECURSIVE),
4570 };
4571
4572 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4573 return -EINVAL;
4574 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4575 return -EINVAL;
4576 kattr->propagation = attr->propagation;
4577
4578 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4579 return -EINVAL;
4580
4581 kattr->attr_set = attr_flags_to_mnt_flags(attr_flags: attr->attr_set);
4582 kattr->attr_clr = attr_flags_to_mnt_flags(attr_flags: attr->attr_clr);
4583
4584 /*
4585 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4586 * users wanting to transition to a different atime setting cannot
4587 * simply specify the atime setting in @attr_set, but must also
4588 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4589 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4590 * @attr_clr and that @attr_set can't have any atime bits set if
4591 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4592 */
4593 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4594 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4595 return -EINVAL;
4596
4597 /*
4598 * Clear all previous time settings as they are mutually
4599 * exclusive.
4600 */
4601 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4602 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4603 case MOUNT_ATTR_RELATIME:
4604 kattr->attr_set |= MNT_RELATIME;
4605 break;
4606 case MOUNT_ATTR_NOATIME:
4607 kattr->attr_set |= MNT_NOATIME;
4608 break;
4609 case MOUNT_ATTR_STRICTATIME:
4610 break;
4611 default:
4612 return -EINVAL;
4613 }
4614 } else {
4615 if (attr->attr_set & MOUNT_ATTR__ATIME)
4616 return -EINVAL;
4617 }
4618
4619 return build_mount_idmapped(attr, usize, kattr, flags);
4620}
4621
4622static void finish_mount_kattr(struct mount_kattr *kattr)
4623{
4624 put_user_ns(ns: kattr->mnt_userns);
4625 kattr->mnt_userns = NULL;
4626
4627 if (kattr->mnt_idmap)
4628 mnt_idmap_put(idmap: kattr->mnt_idmap);
4629}
4630
4631SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4632 unsigned int, flags, struct mount_attr __user *, uattr,
4633 size_t, usize)
4634{
4635 int err;
4636 struct path target;
4637 struct mount_attr attr;
4638 struct mount_kattr kattr;
4639
4640 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4641
4642 if (flags & ~(AT_EMPTY_PATH |
4643 AT_RECURSIVE |
4644 AT_SYMLINK_NOFOLLOW |
4645 AT_NO_AUTOMOUNT))
4646 return -EINVAL;
4647
4648 if (unlikely(usize > PAGE_SIZE))
4649 return -E2BIG;
4650 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4651 return -EINVAL;
4652
4653 if (!may_mount())
4654 return -EPERM;
4655
4656 err = copy_struct_from_user(dst: &attr, ksize: sizeof(attr), src: uattr, usize);
4657 if (err)
4658 return err;
4659
4660 /* Don't bother walking through the mounts if this is a nop. */
4661 if (attr.attr_set == 0 &&
4662 attr.attr_clr == 0 &&
4663 attr.propagation == 0)
4664 return 0;
4665
4666 err = build_mount_kattr(attr: &attr, usize, kattr: &kattr, flags);
4667 if (err)
4668 return err;
4669
4670 err = user_path_at(dfd, name: path, flags: kattr.lookup_flags, path: &target);
4671 if (!err) {
4672 err = do_mount_setattr(path: &target, kattr: &kattr);
4673 path_put(&target);
4674 }
4675 finish_mount_kattr(kattr: &kattr);
4676 return err;
4677}
4678
4679static void __init init_mount_tree(void)
4680{
4681 struct vfsmount *mnt;
4682 struct mount *m;
4683 struct mnt_namespace *ns;
4684 struct path root;
4685
4686 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4687 if (IS_ERR(ptr: mnt))
4688 panic(fmt: "Can't create rootfs");
4689
4690 ns = alloc_mnt_ns(user_ns: &init_user_ns, anon: false);
4691 if (IS_ERR(ptr: ns))
4692 panic(fmt: "Can't allocate initial namespace");
4693 m = real_mount(mnt);
4694 m->mnt_ns = ns;
4695 ns->root = m;
4696 ns->mounts = 1;
4697 list_add(new: &m->mnt_list, head: &ns->list);
4698 init_task.nsproxy->mnt_ns = ns;
4699 get_mnt_ns(ns);
4700
4701 root.mnt = mnt;
4702 root.dentry = mnt->mnt_root;
4703 mnt->mnt_flags |= MNT_LOCKED;
4704
4705 set_fs_pwd(current->fs, &root);
4706 set_fs_root(current->fs, &root);
4707}
4708
4709void __init mnt_init(void)
4710{
4711 int err;
4712
4713 mnt_cache = kmem_cache_create(name: "mnt_cache", size: sizeof(struct mount),
4714 align: 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4715
4716 mount_hashtable = alloc_large_system_hash(tablename: "Mount-cache",
4717 bucketsize: sizeof(struct hlist_head),
4718 numentries: mhash_entries, scale: 19,
4719 HASH_ZERO,
4720 hash_shift: &m_hash_shift, hash_mask: &m_hash_mask, low_limit: 0, high_limit: 0);
4721 mountpoint_hashtable = alloc_large_system_hash(tablename: "Mountpoint-cache",
4722 bucketsize: sizeof(struct hlist_head),
4723 numentries: mphash_entries, scale: 19,
4724 HASH_ZERO,
4725 hash_shift: &mp_hash_shift, hash_mask: &mp_hash_mask, low_limit: 0, high_limit: 0);
4726
4727 if (!mount_hashtable || !mountpoint_hashtable)
4728 panic(fmt: "Failed to allocate mount hash table\n");
4729
4730 kernfs_init();
4731
4732 err = sysfs_init();
4733 if (err)
4734 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4735 __func__, err);
4736 fs_kobj = kobject_create_and_add(name: "fs", NULL);
4737 if (!fs_kobj)
4738 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4739 shmem_init();
4740 init_rootfs();
4741 init_mount_tree();
4742}
4743
4744void put_mnt_ns(struct mnt_namespace *ns)
4745{
4746 if (!refcount_dec_and_test(r: &ns->ns.count))
4747 return;
4748 drop_collected_mounts(mnt: &ns->root->mnt);
4749 free_mnt_ns(ns);
4750}
4751
4752struct vfsmount *kern_mount(struct file_system_type *type)
4753{
4754 struct vfsmount *mnt;
4755 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4756 if (!IS_ERR(ptr: mnt)) {
4757 /*
4758 * it is a longterm mount, don't release mnt until
4759 * we unmount before file sys is unregistered
4760 */
4761 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4762 }
4763 return mnt;
4764}
4765EXPORT_SYMBOL_GPL(kern_mount);
4766
4767void kern_unmount(struct vfsmount *mnt)
4768{
4769 /* release long term mount so mount point can be released */
4770 if (!IS_ERR(ptr: mnt)) {
4771 mnt_make_shortterm(mnt);
4772 synchronize_rcu(); /* yecchhh... */
4773 mntput(mnt);
4774 }
4775}
4776EXPORT_SYMBOL(kern_unmount);
4777
4778void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4779{
4780 unsigned int i;
4781
4782 for (i = 0; i < num; i++)
4783 mnt_make_shortterm(mnt: mnt[i]);
4784 synchronize_rcu_expedited();
4785 for (i = 0; i < num; i++)
4786 mntput(mnt[i]);
4787}
4788EXPORT_SYMBOL(kern_unmount_array);
4789
4790bool our_mnt(struct vfsmount *mnt)
4791{
4792 return check_mnt(mnt: real_mount(mnt));
4793}
4794
4795bool current_chrooted(void)
4796{
4797 /* Does the current process have a non-standard root */
4798 struct path ns_root;
4799 struct path fs_root;
4800 bool chrooted;
4801
4802 /* Find the namespace root */
4803 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4804 ns_root.dentry = ns_root.mnt->mnt_root;
4805 path_get(&ns_root);
4806 while (d_mountpoint(dentry: ns_root.dentry) && follow_down_one(&ns_root))
4807 ;
4808
4809 get_fs_root(current->fs, root: &fs_root);
4810
4811 chrooted = !path_equal(path1: &fs_root, path2: &ns_root);
4812
4813 path_put(&fs_root);
4814 path_put(&ns_root);
4815
4816 return chrooted;
4817}
4818
4819static bool mnt_already_visible(struct mnt_namespace *ns,
4820 const struct super_block *sb,
4821 int *new_mnt_flags)
4822{
4823 int new_flags = *new_mnt_flags;
4824 struct mount *mnt;
4825 bool visible = false;
4826
4827 down_read(sem: &namespace_sem);
4828 lock_ns_list(ns);
4829 list_for_each_entry(mnt, &ns->list, mnt_list) {
4830 struct mount *child;
4831 int mnt_flags;
4832
4833 if (mnt_is_cursor(mnt))
4834 continue;
4835
4836 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4837 continue;
4838
4839 /* This mount is not fully visible if it's root directory
4840 * is not the root directory of the filesystem.
4841 */
4842 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4843 continue;
4844
4845 /* A local view of the mount flags */
4846 mnt_flags = mnt->mnt.mnt_flags;
4847
4848 /* Don't miss readonly hidden in the superblock flags */
4849 if (sb_rdonly(sb: mnt->mnt.mnt_sb))
4850 mnt_flags |= MNT_LOCK_READONLY;
4851
4852 /* Verify the mount flags are equal to or more permissive
4853 * than the proposed new mount.
4854 */
4855 if ((mnt_flags & MNT_LOCK_READONLY) &&
4856 !(new_flags & MNT_READONLY))
4857 continue;
4858 if ((mnt_flags & MNT_LOCK_ATIME) &&
4859 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4860 continue;
4861
4862 /* This mount is not fully visible if there are any
4863 * locked child mounts that cover anything except for
4864 * empty directories.
4865 */
4866 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4867 struct inode *inode = child->mnt_mountpoint->d_inode;
4868 /* Only worry about locked mounts */
4869 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4870 continue;
4871 /* Is the directory permanetly empty? */
4872 if (!is_empty_dir_inode(inode))
4873 goto next;
4874 }
4875 /* Preserve the locked attributes */
4876 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4877 MNT_LOCK_ATIME);
4878 visible = true;
4879 goto found;
4880 next: ;
4881 }
4882found:
4883 unlock_ns_list(ns);
4884 up_read(sem: &namespace_sem);
4885 return visible;
4886}
4887
4888static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4889{
4890 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4891 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4892 unsigned long s_iflags;
4893
4894 if (ns->user_ns == &init_user_ns)
4895 return false;
4896
4897 /* Can this filesystem be too revealing? */
4898 s_iflags = sb->s_iflags;
4899 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4900 return false;
4901
4902 if ((s_iflags & required_iflags) != required_iflags) {
4903 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4904 required_iflags);
4905 return true;
4906 }
4907
4908 return !mnt_already_visible(ns, sb, new_mnt_flags);
4909}
4910
4911bool mnt_may_suid(struct vfsmount *mnt)
4912{
4913 /*
4914 * Foreign mounts (accessed via fchdir or through /proc
4915 * symlinks) are always treated as if they are nosuid. This
4916 * prevents namespaces from trusting potentially unsafe
4917 * suid/sgid bits, file caps, or security labels that originate
4918 * in other namespaces.
4919 */
4920 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(mnt: real_mount(mnt)) &&
4921 current_in_userns(target_ns: mnt->mnt_sb->s_user_ns);
4922}
4923
4924static struct ns_common *mntns_get(struct task_struct *task)
4925{
4926 struct ns_common *ns = NULL;
4927 struct nsproxy *nsproxy;
4928
4929 task_lock(p: task);
4930 nsproxy = task->nsproxy;
4931 if (nsproxy) {
4932 ns = &nsproxy->mnt_ns->ns;
4933 get_mnt_ns(ns: to_mnt_ns(ns));
4934 }
4935 task_unlock(p: task);
4936
4937 return ns;
4938}
4939
4940static void mntns_put(struct ns_common *ns)
4941{
4942 put_mnt_ns(ns: to_mnt_ns(ns));
4943}
4944
4945static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4946{
4947 struct nsproxy *nsproxy = nsset->nsproxy;
4948 struct fs_struct *fs = nsset->fs;
4949 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4950 struct user_namespace *user_ns = nsset->cred->user_ns;
4951 struct path root;
4952 int err;
4953
4954 if (!ns_capable(ns: mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4955 !ns_capable(ns: user_ns, CAP_SYS_CHROOT) ||
4956 !ns_capable(ns: user_ns, CAP_SYS_ADMIN))
4957 return -EPERM;
4958
4959 if (is_anon_ns(ns: mnt_ns))
4960 return -EINVAL;
4961
4962 if (fs->users != 1)
4963 return -EINVAL;
4964
4965 get_mnt_ns(ns: mnt_ns);
4966 old_mnt_ns = nsproxy->mnt_ns;
4967 nsproxy->mnt_ns = mnt_ns;
4968
4969 /* Find the root */
4970 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4971 "/", LOOKUP_DOWN, &root);
4972 if (err) {
4973 /* revert to old namespace */
4974 nsproxy->mnt_ns = old_mnt_ns;
4975 put_mnt_ns(ns: mnt_ns);
4976 return err;
4977 }
4978
4979 put_mnt_ns(ns: old_mnt_ns);
4980
4981 /* Update the pwd and root */
4982 set_fs_pwd(fs, &root);
4983 set_fs_root(fs, &root);
4984
4985 path_put(&root);
4986 return 0;
4987}
4988
4989static struct user_namespace *mntns_owner(struct ns_common *ns)
4990{
4991 return to_mnt_ns(ns)->user_ns;
4992}
4993
4994const struct proc_ns_operations mntns_operations = {
4995 .name = "mnt",
4996 .type = CLONE_NEWNS,
4997 .get = mntns_get,
4998 .put = mntns_put,
4999 .install = mntns_install,
5000 .owner = mntns_owner,
5001};
5002
5003#ifdef CONFIG_SYSCTL
5004static struct ctl_table fs_namespace_sysctls[] = {
5005 {
5006 .procname = "mount-max",
5007 .data = &sysctl_mount_max,
5008 .maxlen = sizeof(unsigned int),
5009 .mode = 0644,
5010 .proc_handler = proc_dointvec_minmax,
5011 .extra1 = SYSCTL_ONE,
5012 },
5013 { }
5014};
5015
5016static int __init init_fs_namespace_sysctls(void)
5017{
5018 register_sysctl_init("fs", fs_namespace_sysctls);
5019 return 0;
5020}
5021fs_initcall(init_fs_namespace_sysctls);
5022
5023#endif /* CONFIG_SYSCTL */
5024

source code of linux/fs/namespace.c