blob: 95b2fff91f6730db8e2da9d9b3f2125c9f9ca098 [file] [log] [blame]
Thomas Gleixner59bd9de2019-05-28 10:10:12 -07001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +010018#include <linux/cred.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010019#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070020#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050021#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
Al Viroa07b2002018-11-05 17:40:30 +000023#include <linux/file.h>
Al Virod10577a2011-12-07 13:06:11 -050024#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010025#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070026#include <linux/magic.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070027#include <linux/memblock.h>
Christian Brauner9caccd42021-01-21 14:19:54 +010028#include <linux/proc_fs.h>
Al Viro9ea459e12014-08-08 13:08:20 -040029#include <linux/task_work.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010030#include <linux/sched/task.h>
David Howellse262e32d2018-11-01 23:07:23 +000031#include <uapi/linux/mount.h>
David Howells9bc61ab2018-11-04 03:19:03 -050032#include <linux/fs_context.h>
Al Viro037f11b2019-06-01 18:09:44 -040033#include <linux/shmem_fs.h>
Christian Braunerbd303362021-12-03 12:17:07 +010034#include <linux/mnt_idmapping.h>
Miklos Szeredib4c2bea2023-10-25 16:02:03 +020035#include <linux/nospec.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010036
Ram Pai07b20882005-11-07 17:19:07 -050037#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070038#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Eric W. Biedermand2921682016-09-28 00:27:17 -050040/* Maximum number of mounts in a mount namespace */
Luis Chamberlainab171b92022-01-21 22:13:27 -080041static unsigned int sysctl_mount_max __read_mostly = 100000;
Eric W. Biedermand2921682016-09-28 00:27:17 -050042
Alexey Dobriyan68279f92023-10-11 19:55:00 +030043static unsigned int m_hash_mask __ro_after_init;
44static unsigned int m_hash_shift __ro_after_init;
45static unsigned int mp_hash_mask __ro_after_init;
46static unsigned int mp_hash_shift __ro_after_init;
Al Viro0818bf22014-02-28 13:46:44 -050047
48static __initdata unsigned long mhash_entries;
49static int __init set_mhash_entries(char *str)
50{
51 if (!str)
52 return 0;
53 mhash_entries = simple_strtoul(str, &str, 0);
54 return 1;
55}
56__setup("mhash_entries=", set_mhash_entries);
57
58static __initdata unsigned long mphash_entries;
59static int __init set_mphash_entries(char *str)
60{
61 if (!str)
62 return 0;
63 mphash_entries = simple_strtoul(str, &str, 0);
64 return 1;
65}
66__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080067
Al Viroc7999c32014-02-27 14:40:10 -050068static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010069static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010070static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Miklos Szeredi98d2b432023-10-25 16:01:59 +020072/* Don't allow confusion with old 32bit mount ID */
73static atomic64_t mnt_id_ctr = ATOMIC64_INIT(1ULL << 32);
74
Alexey Dobriyan68279f92023-10-11 19:55:00 +030075static struct hlist_head *mount_hashtable __ro_after_init;
76static struct hlist_head *mountpoint_hashtable __ro_after_init;
77static struct kmem_cache *mnt_cache __ro_after_init;
Al Viro59aa0da2013-09-16 21:34:53 -040078static DECLARE_RWSEM(namespace_sem);
Al Viro4edbe132019-06-30 10:39:08 -040079static HLIST_HEAD(unmounted); /* protected by namespace_sem */
80static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Christian Brauner2a186722021-01-21 14:19:53 +010082struct mount_kattr {
83 unsigned int attr_set;
84 unsigned int attr_clr;
85 unsigned int propagation;
86 unsigned int lookup_flags;
87 bool recurse;
Christian Brauner9caccd42021-01-21 14:19:54 +010088 struct user_namespace *mnt_userns;
Christian Brauner256c8ae2022-10-26 12:51:27 +020089 struct mnt_idmap *mnt_idmap;
Christian Brauner2a186722021-01-21 14:19:53 +010090};
91
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080092/* /sys/fs */
Alexey Dobriyan68279f92023-10-11 19:55:00 +030093struct kobject *fs_kobj __ro_after_init;
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060094EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080095
Nick Piggin99b7db72010-08-18 04:37:39 +100096/*
97 * vfsmount lock may be taken for read to prevent changes to the
98 * vfsmount hash, ie. during mountpoint lookups or walking back
99 * up the tree.
100 *
101 * It should be taken for write in all cases where the vfsmount
102 * tree or hash is modified or when a vfsmount structure is modified.
103 */
Al Viro48a066e2013-09-29 22:06:07 -0400104__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +1000105
Christian Braunerd033cb62021-01-21 14:19:49 +0100106static inline void lock_mount_hash(void)
107{
108 write_seqlock(&mount_lock);
109}
110
111static inline void unlock_mount_hash(void)
112{
113 write_sequnlock(&mount_lock);
114}
115
Al Viro38129a12014-03-20 21:10:51 -0400116static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Ram Paib58fed82005-11-07 17:16:09 -0500118 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
119 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -0500120 tmp = tmp + (tmp >> m_hash_shift);
121 return &mount_hashtable[tmp & m_hash_mask];
122}
123
124static inline struct hlist_head *mp_hash(struct dentry *dentry)
125{
126 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
127 tmp = tmp + (tmp >> mp_hash_shift);
128 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129}
130
Al Virob105e272011-11-24 20:38:33 -0500131static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100132{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400133 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100134
Matthew Wilcox169b4802018-06-11 12:31:36 -0400135 if (res < 0)
136 return res;
137 mnt->mnt_id = res;
Miklos Szeredi98d2b432023-10-25 16:01:59 +0200138 mnt->mnt_id_unique = atomic64_inc_return(&mnt_id_ctr);
Matthew Wilcox169b4802018-06-11 12:31:36 -0400139 return 0;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100140}
141
Al Virob105e272011-11-24 20:38:33 -0500142static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100143{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400144 ida_free(&mnt_id_ida, mnt->mnt_id);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100145}
146
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100147/*
148 * Allocate a new peer group ID
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100149 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500150static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100151{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400152 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
Al Virof21f6222009-06-24 03:12:00 -0400153
Matthew Wilcox169b4802018-06-11 12:31:36 -0400154 if (res < 0)
155 return res;
156 mnt->mnt_group_id = res;
157 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158}
159
160/*
161 * Release a peer group ID
162 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500163void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100164{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400165 ida_free(&mnt_group_ida, mnt->mnt_group_id);
Al Viro15169fe2011-11-25 00:50:41 -0500166 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100167}
168
Nick Pigginb3e19d92011-01-07 17:50:11 +1100169/*
170 * vfsmount lock must be held for read
171 */
Al Viro83adc752011-11-24 22:37:54 -0500172static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100173{
174#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500175 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100176#else
177 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500178 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100179 preempt_enable();
180#endif
181}
182
Nick Pigginb3e19d92011-01-07 17:50:11 +1100183/*
184 * vfsmount lock must be held for write
185 */
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700186int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100187{
188#ifdef CONFIG_SMP
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700189 int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190 int cpu;
191
192 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500193 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100194 }
195
196 return count;
197#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500198 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100199#endif
200}
201
Al Virob105e272011-11-24 20:38:33 -0500202static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Al Viroc63181e2011-11-25 02:35:16 -0500204 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
205 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100206 int err;
207
Al Viroc63181e2011-11-25 02:35:16 -0500208 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800209 if (err)
210 goto out_free_cache;
211
212 if (name) {
Vasily Averin79f65402021-09-02 14:55:10 -0700213 mnt->mnt_devname = kstrdup_const(name,
214 GFP_KERNEL_ACCOUNT);
Al Viroc63181e2011-11-25 02:35:16 -0500215 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800216 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100217 }
218
Nick Pigginb3e19d92011-01-07 17:50:11 +1100219#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500220 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100222 goto out_free_devname;
223
Al Viroc63181e2011-11-25 02:35:16 -0500224 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100225#else
Al Viroc63181e2011-11-25 02:35:16 -0500226 mnt->mnt_count = 1;
227 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100228#endif
229
Al Viro38129a12014-03-20 21:10:51 -0400230 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500231 INIT_LIST_HEAD(&mnt->mnt_child);
232 INIT_LIST_HEAD(&mnt->mnt_mounts);
233 INIT_LIST_HEAD(&mnt->mnt_list);
234 INIT_LIST_HEAD(&mnt->mnt_expire);
235 INIT_LIST_HEAD(&mnt->mnt_share);
236 INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700238 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Eric W. Biederman99b19d12016-10-24 16:16:13 -0500239 INIT_LIST_HEAD(&mnt->mnt_umounting);
Al Viro56cbb4292019-07-04 16:57:51 -0400240 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
Christian Brauner256c8ae2022-10-26 12:51:27 +0200241 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
Al Viroc63181e2011-11-25 02:35:16 -0500243 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800244
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000245#ifdef CONFIG_SMP
246out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800247 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248#endif
Li Zefan88b38782008-07-21 18:06:36 +0800249out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500250 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800251out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500252 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800253 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254}
255
Dave Hansen83660252008-02-15 14:37:30 -0800256/*
257 * Most r/o checks on a fs are for operations that take
258 * discrete amounts of time, like a write() or unlink().
259 * We must keep track of when those operations start
260 * (for permission checks) and when they end, so that
261 * we can determine when writes are able to occur to
262 * a filesystem.
263 */
Dave Hansen3d733632008-02-15 14:37:59 -0800264/*
265 * __mnt_is_readonly: check whether a mount is read-only
266 * @mnt: the mount to check for its write status
267 *
268 * This shouldn't be used directly ouside of the VFS.
269 * It does not guarantee that the filesystem will stay
270 * r/w, just that it is right *now*. This can not and
271 * should not be used in place of IS_RDONLY(inode).
272 * mnt_want/drop_write() will _keep_ the filesystem
273 * r/w.
274 */
David Howells43f5e652018-11-01 23:07:25 +0000275bool __mnt_is_readonly(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800276{
David Howells43f5e652018-11-01 23:07:25 +0000277 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800278}
279EXPORT_SYMBOL_GPL(__mnt_is_readonly);
280
Al Viro83adc752011-11-24 22:37:54 -0500281static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800282{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000283#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500284 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000285#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500286 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000287#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800288}
Dave Hansen3d733632008-02-15 14:37:59 -0800289
Al Viro83adc752011-11-24 22:37:54 -0500290static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800291{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000292#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500293 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000294#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500295 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000296#endif
297}
298
Al Viro83adc752011-11-24 22:37:54 -0500299static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000300{
301#ifdef CONFIG_SMP
302 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800303 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800304
305 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500306 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800307 }
Dave Hansen3d733632008-02-15 14:37:59 -0800308
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000309 return count;
310#else
311 return mnt->mnt_writers;
312#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800313}
314
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100315static int mnt_is_readonly(struct vfsmount *mnt)
316{
Jan Karad7439fb2023-06-20 13:28:32 +0200317 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100318 return 1;
Jan Karad7439fb2023-06-20 13:28:32 +0200319 /*
320 * The barrier pairs with the barrier in sb_start_ro_state_change()
321 * making sure if we don't see s_readonly_remount set yet, we also will
322 * not see any superblock / mount flag changes done by remount.
323 * It also pairs with the barrier in sb_end_ro_state_change()
324 * assuring that if we see s_readonly_remount already cleared, we will
325 * see the values of superblock / mount flags updated by remount.
326 */
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100327 smp_rmb();
328 return __mnt_is_readonly(mnt);
329}
330
Dave Hansen3d733632008-02-15 14:37:59 -0800331/*
Jan Karaeb04c282012-06-12 16:20:35 +0200332 * Most r/o & frozen checks on a fs are for operations that take discrete
333 * amounts of time, like a write() or unlink(). We must keep track of when
334 * those operations start (for permission checks) and when they end, so that we
335 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800336 */
Dave Hansen83660252008-02-15 14:37:30 -0800337/**
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300338 * mnt_get_write_access - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500339 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800340 *
Jan Karaeb04c282012-06-12 16:20:35 +0200341 * This tells the low-level filesystem that a write is about to be performed to
342 * it, and makes sure that writes are allowed (mnt it read-write) before
343 * returning success. This operation does not protect against filesystem being
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300344 * frozen. When the write operation is finished, mnt_put_write_access() must be
Jan Karaeb04c282012-06-12 16:20:35 +0200345 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800346 */
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300347int mnt_get_write_access(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800348{
Al Viro83adc752011-11-24 22:37:54 -0500349 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800350 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800351
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000352 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100353 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000354 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100355 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000356 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
357 * incremented count after it has set MNT_WRITE_HOLD.
358 */
359 smp_mb();
Sebastian Andrzej Siewior0f8821d2021-11-25 13:07:11 +0100360 might_lock(&mount_lock.lock);
361 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
362 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
363 cpu_relax();
364 } else {
365 /*
366 * This prevents priority inversion, if the task
367 * setting MNT_WRITE_HOLD got preempted on a remote
368 * CPU, and it prevents life lock if the task setting
369 * MNT_WRITE_HOLD has a lower priority and is bound to
370 * the same CPU as the task that is spinning here.
371 */
372 preempt_enable();
373 lock_mount_hash();
374 unlock_mount_hash();
375 preempt_disable();
376 }
377 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000378 /*
Jan Karad7439fb2023-06-20 13:28:32 +0200379 * The barrier pairs with the barrier sb_start_ro_state_change() making
380 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
381 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
382 * mnt_is_readonly() and bail in case we are racing with remount
383 * read-only.
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000384 */
385 smp_rmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100386 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100387 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800388 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800389 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000390 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200391
392 return ret;
393}
Amir Goldsteinddf9e2f2023-09-08 16:29:00 +0300394EXPORT_SYMBOL_GPL(mnt_get_write_access);
Jan Karaeb04c282012-06-12 16:20:35 +0200395
396/**
397 * mnt_want_write - get write access to a mount
398 * @m: the mount on which to take a write
399 *
400 * This tells the low-level filesystem that a write is about to be performed to
401 * it, and makes sure that writes are allowed (mount is read-write, filesystem
402 * is not frozen) before returning success. When the write operation is
403 * finished, mnt_drop_write() must be called. This is effectively a refcount.
404 */
405int mnt_want_write(struct vfsmount *m)
406{
407 int ret;
408
409 sb_start_write(m->mnt_sb);
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300410 ret = mnt_get_write_access(m);
Jan Karaeb04c282012-06-12 16:20:35 +0200411 if (ret)
412 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800413 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800414}
415EXPORT_SYMBOL_GPL(mnt_want_write);
416
417/**
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300418 * mnt_get_write_access_file - get write access to a file's mount
Jan Karaeb04c282012-06-12 16:20:35 +0200419 * @file: the file who's mount on which to take a write
420 *
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300421 * This is like mnt_get_write_access, but if @file is already open for write it
Eric Biggers14e43bf42020-09-22 09:44:18 -0700422 * skips incrementing mnt_writers (since the open file already has a reference)
423 * and instead only does the check for emergency r/o remounts. This must be
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300424 * paired with mnt_put_write_access_file.
Jan Karaeb04c282012-06-12 16:20:35 +0200425 */
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300426int mnt_get_write_access_file(struct file *file)
Jan Karaeb04c282012-06-12 16:20:35 +0200427{
Eric Biggers14e43bf42020-09-22 09:44:18 -0700428 if (file->f_mode & FMODE_WRITER) {
429 /*
430 * Superblock may have become readonly while there are still
431 * writable fd's, e.g. due to a fs error with errors=remount-ro
432 */
433 if (__mnt_is_readonly(file->f_path.mnt))
434 return -EROFS;
435 return 0;
436 }
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300437 return mnt_get_write_access(file->f_path.mnt);
Jan Karaeb04c282012-06-12 16:20:35 +0200438}
439
440/**
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200441 * mnt_want_write_file - get write access to a file's mount
442 * @file: the file who's mount on which to take a write
443 *
Eric Biggers14e43bf42020-09-22 09:44:18 -0700444 * This is like mnt_want_write, but if the file is already open for writing it
445 * skips incrementing mnt_writers (since the open file already has a reference)
446 * and instead only does the freeze protection and the check for emergency r/o
447 * remounts. This must be paired with mnt_drop_write_file.
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200448 */
449int mnt_want_write_file(struct file *file)
450{
451 int ret;
452
Miklos Szeredia6795a52018-07-18 15:44:43 +0200453 sb_start_write(file_inode(file)->i_sb);
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300454 ret = mnt_get_write_access_file(file);
Jan Karaeb04c282012-06-12 16:20:35 +0200455 if (ret)
Miklos Szeredia6795a52018-07-18 15:44:43 +0200456 sb_end_write(file_inode(file)->i_sb);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200457 return ret;
458}
npiggin@suse.de96029c42009-04-26 20:25:55 +1000459EXPORT_SYMBOL_GPL(mnt_want_write_file);
460
461/**
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300462 * mnt_put_write_access - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800463 * @mnt: the mount on which to give up write access
464 *
465 * Tells the low-level filesystem that we are done
466 * performing writes to it. Must be matched with
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300467 * mnt_get_write_access() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800468 */
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300469void mnt_put_write_access(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800470{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000471 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500472 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000473 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800474}
Amir Goldsteinddf9e2f2023-09-08 16:29:00 +0300475EXPORT_SYMBOL_GPL(mnt_put_write_access);
Jan Karaeb04c282012-06-12 16:20:35 +0200476
477/**
478 * mnt_drop_write - give up write access to a mount
479 * @mnt: the mount on which to give up write access
480 *
481 * Tells the low-level filesystem that we are done performing writes to it and
482 * also allows filesystem to be frozen again. Must be matched with
483 * mnt_want_write() call above.
484 */
485void mnt_drop_write(struct vfsmount *mnt)
486{
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300487 mnt_put_write_access(mnt);
Jan Karaeb04c282012-06-12 16:20:35 +0200488 sb_end_write(mnt->mnt_sb);
489}
Dave Hansen83660252008-02-15 14:37:30 -0800490EXPORT_SYMBOL_GPL(mnt_drop_write);
491
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300492void mnt_put_write_access_file(struct file *file)
Jan Karaeb04c282012-06-12 16:20:35 +0200493{
Eric Biggers14e43bf42020-09-22 09:44:18 -0700494 if (!(file->f_mode & FMODE_WRITER))
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300495 mnt_put_write_access(file->f_path.mnt);
Jan Karaeb04c282012-06-12 16:20:35 +0200496}
497
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200498void mnt_drop_write_file(struct file *file)
499{
Amir Goldstein3e15dcf2023-09-08 16:28:59 +0300500 mnt_put_write_access_file(file);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200501 sb_end_write(file_inode(file)->i_sb);
502}
Al Viro2a79f172011-12-09 08:06:57 -0500503EXPORT_SYMBOL(mnt_drop_write_file);
504
Christian Brauner538f4f022022-02-03 14:14:07 +0100505/**
506 * mnt_hold_writers - prevent write access to the given mount
507 * @mnt: mnt to prevent write access to
508 *
509 * Prevents write access to @mnt if there are no active writers for @mnt.
510 * This function needs to be called and return successfully before changing
511 * properties of @mnt that need to remain stable for callers with write access
512 * to @mnt.
513 *
514 * After this functions has been called successfully callers must pair it with
515 * a call to mnt_unhold_writers() in order to stop preventing write access to
516 * @mnt.
517 *
518 * Context: This function expects lock_mount_hash() to be held serializing
519 * setting MNT_WRITE_HOLD.
520 * Return: On success 0 is returned.
521 * On error, -EBUSY is returned.
522 */
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100523static inline int mnt_hold_writers(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800524{
Al Viro83adc752011-11-24 22:37:54 -0500525 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000526 /*
527 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
528 * should be visible before we do.
529 */
530 smp_mb();
531
532 /*
533 * With writers on hold, if this value is zero, then there are
534 * definitely no active writers (although held writers may subsequently
535 * increment the count, they'll have to wait, and decrement it after
536 * seeing MNT_READONLY).
537 *
538 * It is OK to have counter incremented on one CPU and decremented on
539 * another: the sum will add up correctly. The danger would be when we
540 * sum up each counter, if we read a counter before it is incremented,
541 * but then read another CPU's count which it has been subsequently
542 * decremented from -- we would see more decrements than we should.
543 * MNT_WRITE_HOLD protects against this scenario, because
544 * mnt_want_write first increments count, then smp_mb, then spins on
545 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
546 * we're counting up here.
547 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100548 if (mnt_get_writers(mnt) > 0)
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100549 return -EBUSY;
550
551 return 0;
552}
553
Christian Brauner538f4f022022-02-03 14:14:07 +0100554/**
555 * mnt_unhold_writers - stop preventing write access to the given mount
556 * @mnt: mnt to stop preventing write access to
557 *
558 * Stop preventing write access to @mnt allowing callers to gain write access
559 * to @mnt again.
560 *
561 * This function can only be called after a successful call to
562 * mnt_hold_writers().
563 *
564 * Context: This function expects lock_mount_hash() to be held.
565 */
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100566static inline void mnt_unhold_writers(struct mount *mnt)
567{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000568 /*
569 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
570 * that become unheld will see MNT_READONLY.
571 */
572 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500573 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100574}
575
576static int mnt_make_readonly(struct mount *mnt)
577{
578 int ret;
579
580 ret = mnt_hold_writers(mnt);
581 if (!ret)
582 mnt->mnt.mnt_flags |= MNT_READONLY;
583 mnt_unhold_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800584 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800585}
Dave Hansen83660252008-02-15 14:37:30 -0800586
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100587int sb_prepare_remount_readonly(struct super_block *sb)
588{
589 struct mount *mnt;
590 int err = 0;
591
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100592 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
593 if (atomic_long_read(&sb->s_remove_count))
594 return -EBUSY;
595
Al Viro719ea2f2013-09-29 11:24:49 -0400596 lock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100597 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
598 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
Christian Braunerad1844a2022-02-03 14:14:10 +0100599 err = mnt_hold_writers(mnt);
600 if (err)
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100601 break;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100602 }
603 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100604 if (!err && atomic_long_read(&sb->s_remove_count))
605 err = -EBUSY;
606
Jan Karad7439fb2023-06-20 13:28:32 +0200607 if (!err)
608 sb_start_ro_state_change(sb);
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100609 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
610 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
611 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
612 }
Al Viro719ea2f2013-09-29 11:24:49 -0400613 unlock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100614
615 return err;
616}
617
Al Virob105e272011-11-24 20:38:33 -0500618static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Christian Brauner256c8ae2022-10-26 12:51:27 +0200620 mnt_idmap_put(mnt_idmap(&mnt->mnt));
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800621 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000622#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500623 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000624#endif
Al Virob105e272011-11-24 20:38:33 -0500625 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
David Howells8ffcb322014-01-24 12:17:54 +0000628static void delayed_free_vfsmnt(struct rcu_head *head)
629{
630 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
631}
632
Al Viro48a066e2013-09-29 22:06:07 -0400633/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400634int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400635{
636 struct mount *mnt;
637 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400638 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400639 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400640 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400641 mnt = real_mount(bastard);
642 mnt_add_count(mnt, 1);
Al Viro119e1ef2018-08-09 17:51:32 -0400643 smp_mb(); // see mntput_no_expire()
Al Viro48a066e2013-09-29 22:06:07 -0400644 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400645 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400646 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
647 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400648 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400649 }
Al Viro119e1ef2018-08-09 17:51:32 -0400650 lock_mount_hash();
651 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
652 mnt_add_count(mnt, -1);
653 unlock_mount_hash();
654 return 1;
655 }
656 unlock_mount_hash();
657 /* caller will mntput() */
Al Viro294d71f2015-05-08 11:43:53 -0400658 return -1;
659}
660
661/* call under rcu_read_lock */
Al Viro7e4745a2022-07-05 12:22:46 -0400662static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro294d71f2015-05-08 11:43:53 -0400663{
664 int res = __legitimize_mnt(bastard, seq);
665 if (likely(!res))
666 return true;
667 if (unlikely(res < 0)) {
668 rcu_read_unlock();
669 mntput(bastard);
670 rcu_read_lock();
671 }
Al Viro48a066e2013-09-29 22:06:07 -0400672 return false;
673}
674
Christian Brauner104026c2023-05-03 13:18:40 +0200675/**
676 * __lookup_mnt - find first child mount
677 * @mnt: parent mount
678 * @dentry: mountpoint
679 *
680 * If @mnt has a child mount @c mounted @dentry find and return it.
681 *
682 * Note that the child mount @c need not be unique. There are cases
683 * where shadow mounts are created. For example, during mount
684 * propagation when a source mount @mnt whose root got overmounted by a
685 * mount @o after path lookup but before @namespace_sem could be
686 * acquired gets copied and propagated. So @mnt gets copied including
687 * @o. When @mnt is propagated to a destination mount @d that already
688 * has another mount @n mounted at the same mountpoint then the source
689 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
690 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
691 * on @dentry.
692 *
693 * Return: The first child of @mnt mounted @dentry or NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 */
Al Viro474279d2013-10-01 16:11:26 -0400695struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Al Viro38129a12014-03-20 21:10:51 -0400697 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400698 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Al Viro38129a12014-03-20 21:10:51 -0400700 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400701 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
702 return p;
703 return NULL;
704}
705
706/*
David Howellsf015f1262012-06-25 12:55:28 +0100707 * lookup_mnt - Return the first child mount mounted at path
708 *
709 * "First" means first mounted chronologically. If you create the
710 * following mounts:
711 *
712 * mount /dev/sda1 /mnt
713 * mount /dev/sda2 /mnt
714 * mount /dev/sda3 /mnt
715 *
716 * Then lookup_mnt() on the base /mnt dentry in the root mount will
717 * return successively the root dentry and vfsmount of /dev/sda1, then
718 * /dev/sda2, then /dev/sda3, then NULL.
719 *
720 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500721 */
Al Viroca71cf72016-11-20 19:45:28 -0500722struct vfsmount *lookup_mnt(const struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500723{
Al Viroc7105362011-11-24 18:22:03 -0500724 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400725 struct vfsmount *m;
726 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000727
Al Viro48a066e2013-09-29 22:06:07 -0400728 rcu_read_lock();
729 do {
730 seq = read_seqbegin(&mount_lock);
731 child_mnt = __lookup_mnt(path->mnt, path->dentry);
732 m = child_mnt ? &child_mnt->mnt : NULL;
733 } while (!legitimize_mnt(m, seq));
734 rcu_read_unlock();
735 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500736}
737
Eric W. Biederman7af13642013-10-04 19:15:13 -0700738/*
739 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
740 * current mount namespace.
741 *
742 * The common case is dentries are not mountpoints at all and that
743 * test is handled inline. For the slow case when we are actually
744 * dealing with a mountpoint of some kind, walk through all of the
745 * mounts in the current mount namespace and test to see if the dentry
746 * is a mountpoint.
747 *
748 * The mount_hashtable is not usable in the context because we
749 * need to identify all mounts that may be in the current mount
750 * namespace not just a mount that happens to have some specified
751 * parent mount.
752 */
753bool __is_local_mountpoint(struct dentry *dentry)
754{
755 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +0200756 struct mount *mnt, *n;
Eric W. Biederman7af13642013-10-04 19:15:13 -0700757 bool is_covered = false;
758
Eric W. Biederman7af13642013-10-04 19:15:13 -0700759 down_read(&namespace_sem);
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +0200760 rbtree_postorder_for_each_entry_safe(mnt, n, &ns->mounts, mnt_node) {
Eric W. Biederman7af13642013-10-04 19:15:13 -0700761 is_covered = (mnt->mnt_mountpoint == dentry);
762 if (is_covered)
763 break;
764 }
765 up_read(&namespace_sem);
Nikolay Borisov5ad05cc2020-03-04 18:12:45 +0200766
Eric W. Biederman7af13642013-10-04 19:15:13 -0700767 return is_covered;
768}
769
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800770static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d171922013-03-15 10:53:28 -0400771{
Al Viro0818bf22014-02-28 13:46:44 -0500772 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d171922013-03-15 10:53:28 -0400773 struct mountpoint *mp;
774
Al Viro0818bf22014-02-28 13:46:44 -0500775 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d171922013-03-15 10:53:28 -0400776 if (mp->m_dentry == dentry) {
Al Viro84d171922013-03-15 10:53:28 -0400777 mp->m_count++;
778 return mp;
779 }
780 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800781 return NULL;
782}
783
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300784static struct mountpoint *get_mountpoint(struct dentry *dentry)
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800785{
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300786 struct mountpoint *mp, *new = NULL;
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800787 int ret;
Al Viro84d171922013-03-15 10:53:28 -0400788
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300789 if (d_mountpoint(dentry)) {
Benjamin Coddington1e9c75f2018-10-03 10:18:33 -0400790 /* might be worth a WARN_ON() */
791 if (d_unlinked(dentry))
792 return ERR_PTR(-ENOENT);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300793mountpoint:
794 read_seqlock_excl(&mount_lock);
795 mp = lookup_mountpoint(dentry);
796 read_sequnlock_excl(&mount_lock);
797 if (mp)
798 goto done;
Al Viro84d171922013-03-15 10:53:28 -0400799 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200800
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300801 if (!new)
802 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
803 if (!new)
804 return ERR_PTR(-ENOMEM);
805
806
807 /* Exactly one processes may set d_mounted */
808 ret = d_set_mounted(dentry);
809
810 /* Someone else set d_mounted? */
811 if (ret == -EBUSY)
812 goto mountpoint;
813
814 /* The dentry is not available as a mountpoint? */
815 mp = ERR_PTR(ret);
816 if (ret)
817 goto done;
818
819 /* Add the new mountpoint to the hash table */
820 read_seqlock_excl(&mount_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400821 new->m_dentry = dget(dentry);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300822 new->m_count = 1;
823 hlist_add_head(&new->m_hash, mp_hash(dentry));
824 INIT_HLIST_HEAD(&new->m_list);
825 read_sequnlock_excl(&mount_lock);
826
827 mp = new;
828 new = NULL;
829done:
830 kfree(new);
Al Viro84d171922013-03-15 10:53:28 -0400831 return mp;
832}
833
Al Viro4edbe132019-06-30 10:39:08 -0400834/*
835 * vfsmount lock must be held. Additionally, the caller is responsible
836 * for serializing calls for given disposal list.
837 */
838static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
Al Viro84d171922013-03-15 10:53:28 -0400839{
840 if (!--mp->m_count) {
841 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700842 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d171922013-03-15 10:53:28 -0400843 spin_lock(&dentry->d_lock);
844 dentry->d_flags &= ~DCACHE_MOUNTED;
845 spin_unlock(&dentry->d_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400846 dput_to_list(dentry, list);
Al Viro0818bf22014-02-28 13:46:44 -0500847 hlist_del(&mp->m_hash);
Al Viro84d171922013-03-15 10:53:28 -0400848 kfree(mp);
849 }
850}
851
Al Viro4edbe132019-06-30 10:39:08 -0400852/* called with namespace_lock and vfsmount lock */
853static void put_mountpoint(struct mountpoint *mp)
854{
855 __put_mountpoint(mp, &ex_mountpoints);
856}
857
Al Viro143c8c92011-11-25 00:46:35 -0500858static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800860 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861}
862
Nick Piggin99b7db72010-08-18 04:37:39 +1000863/*
864 * vfsmount lock must be held for write
865 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800866static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500867{
868 if (ns) {
869 ns->event = ++event;
870 wake_up_interruptible(&ns->poll);
871 }
872}
873
Nick Piggin99b7db72010-08-18 04:37:39 +1000874/*
875 * vfsmount lock must be held for write
876 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800877static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500878{
879 if (ns && ns->event != event) {
880 ns->event = event;
881 wake_up_interruptible(&ns->poll);
882 }
883}
884
Nick Piggin99b7db72010-08-18 04:37:39 +1000885/*
886 * vfsmount lock must be held for write
887 */
Al Viroe4e59902019-06-29 12:58:42 -0400888static struct mountpoint *unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Al Viroe4e59902019-06-29 12:58:42 -0400890 struct mountpoint *mp;
Al Viro0714a532011-11-24 22:19:58 -0500891 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500892 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500893 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400894 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700895 hlist_del_init(&mnt->mnt_mp_list);
Al Viroe4e59902019-06-29 12:58:42 -0400896 mp = mnt->mnt_mp;
Al Viro84d171922013-03-15 10:53:28 -0400897 mnt->mnt_mp = NULL;
Al Viroe4e59902019-06-29 12:58:42 -0400898 return mp;
Eric W. Biederman7bdb11de2014-12-29 13:03:41 -0600899}
900
901/*
902 * vfsmount lock must be held for write
903 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600904static void umount_mnt(struct mount *mnt)
905{
Al Viroe4e59902019-06-29 12:58:42 -0400906 put_mountpoint(unhash_mnt(mnt));
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600907}
908
909/*
910 * vfsmount lock must be held for write
911 */
Al Viro84d171922013-03-15 10:53:28 -0400912void mnt_set_mountpoint(struct mount *mnt,
913 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500914 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500915{
Al Viro84d171922013-03-15 10:53:28 -0400916 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500917 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro4edbe132019-06-30 10:39:08 -0400918 child_mnt->mnt_mountpoint = mp->m_dentry;
Al Viro3a2393d2011-11-25 03:19:09 -0500919 child_mnt->mnt_parent = mnt;
Al Viro84d171922013-03-15 10:53:28 -0400920 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700921 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500922}
923
Christian Brauner6ac39282023-05-03 13:18:42 +0200924/**
925 * mnt_set_mountpoint_beneath - mount a mount beneath another one
926 *
927 * @new_parent: the source mount
928 * @top_mnt: the mount beneath which @new_parent is mounted
929 * @new_mp: the new mountpoint of @top_mnt on @new_parent
930 *
931 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
932 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
933 * @new_mp. And mount @new_parent on the old parent and old
934 * mountpoint of @top_mnt.
935 *
936 * Context: This function expects namespace_lock() and lock_mount_hash()
937 * to have been acquired in that order.
938 */
939static void mnt_set_mountpoint_beneath(struct mount *new_parent,
940 struct mount *top_mnt,
941 struct mountpoint *new_mp)
942{
943 struct mount *old_top_parent = top_mnt->mnt_parent;
944 struct mountpoint *old_top_mp = top_mnt->mnt_mp;
945
946 mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
947 mnt_change_mountpoint(new_parent, new_mp, top_mnt);
948}
949
950
Eric W. Biederman1064f872017-01-20 18:28:35 +1300951static void __attach_mnt(struct mount *mnt, struct mount *parent)
952{
953 hlist_add_head_rcu(&mnt->mnt_hash,
954 m_hash(&parent->mnt, mnt->mnt_mountpoint));
955 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
956}
957
Christian Brauner6ac39282023-05-03 13:18:42 +0200958/**
959 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
960 * list of child mounts
961 * @parent: the parent
962 * @mnt: the new mount
963 * @mp: the new mountpoint
964 * @beneath: whether to mount @mnt beneath or on top of @parent
965 *
966 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
967 * to @parent's child mount list and to @mount_hashtable.
968 *
969 * If @beneath is true, remove @mnt from its current parent and
970 * mountpoint and mount it on @mp on @parent, and mount @parent on the
971 * old parent and old mountpoint of @mnt. Finally, attach @parent to
972 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
973 *
974 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
975 * to the correct parent.
976 *
977 * Context: This function expects namespace_lock() and lock_mount_hash()
978 * to have been acquired in that order.
Nick Piggin99b7db72010-08-18 04:37:39 +1000979 */
Christian Brauner6ac39282023-05-03 13:18:42 +0200980static void attach_mnt(struct mount *mnt, struct mount *parent,
981 struct mountpoint *mp, bool beneath)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
Christian Brauner6ac39282023-05-03 13:18:42 +0200983 if (beneath)
984 mnt_set_mountpoint_beneath(mnt, parent, mp);
985 else
986 mnt_set_mountpoint(parent, mp, mnt);
987 /*
988 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
989 * beneath @parent then @mnt will need to be attached to
990 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
991 * isn't the same mount as @parent.
992 */
993 __attach_mnt(mnt, mnt->mnt_parent);
Ram Paib90fa9a2005-11-07 17:19:50 -0500994}
995
Eric W. Biederman1064f872017-01-20 18:28:35 +1300996void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
Al Viro12a5b522014-08-10 03:44:55 -0400997{
Eric W. Biederman1064f872017-01-20 18:28:35 +1300998 struct mountpoint *old_mp = mnt->mnt_mp;
Eric W. Biederman1064f872017-01-20 18:28:35 +1300999 struct mount *old_parent = mnt->mnt_parent;
1000
1001 list_del_init(&mnt->mnt_child);
1002 hlist_del_init(&mnt->mnt_mp_list);
1003 hlist_del_init_rcu(&mnt->mnt_hash);
1004
Christian Brauner6ac39282023-05-03 13:18:42 +02001005 attach_mnt(mnt, parent, mp, false);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001006
1007 put_mountpoint(old_mp);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001008 mnt_add_count(old_parent, -1);
Al Viro12a5b522014-08-10 03:44:55 -04001009}
1010
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001011static inline struct mount *node_to_mount(struct rb_node *node)
1012{
Miklos Szeredib4c2bea2023-10-25 16:02:03 +02001013 return node ? rb_entry(node, struct mount, mnt_node) : NULL;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001014}
1015
1016static void mnt_add_to_ns(struct mnt_namespace *ns, struct mount *mnt)
1017{
1018 struct rb_node **link = &ns->mounts.rb_node;
1019 struct rb_node *parent = NULL;
1020
1021 WARN_ON(mnt->mnt.mnt_flags & MNT_ONRB);
1022 mnt->mnt_ns = ns;
1023 while (*link) {
1024 parent = *link;
1025 if (mnt->mnt_id_unique < node_to_mount(parent)->mnt_id_unique)
1026 link = &parent->rb_left;
1027 else
1028 link = &parent->rb_right;
1029 }
1030 rb_link_node(&mnt->mnt_node, parent, link);
1031 rb_insert_color(&mnt->mnt_node, &ns->mounts);
1032 mnt->mnt.mnt_flags |= MNT_ONRB;
1033}
1034
Ram Paib90fa9a2005-11-07 17:19:50 -05001035/*
Nick Piggin99b7db72010-08-18 04:37:39 +10001036 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -05001037 */
Eric W. Biederman1064f872017-01-20 18:28:35 +13001038static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -05001039{
Al Viro0714a532011-11-24 22:19:58 -05001040 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -05001041 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -05001042 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -05001043 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -05001044
Al Viro0714a532011-11-24 22:19:58 -05001045 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -05001046
Al Viro1a4eeaf2011-11-25 02:19:55 -05001047 list_add_tail(&head, &mnt->mnt_list);
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001048 while (!list_empty(&head)) {
1049 m = list_first_entry(&head, typeof(*m), mnt_list);
1050 list_del(&m->mnt_list);
Al Virof03c6592011-01-14 22:30:21 -05001051
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001052 mnt_add_to_ns(n, m);
1053 }
1054 n->nr_mounts += n->pending_mounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05001055 n->pending_mounts = 0;
1056
Eric W. Biederman1064f872017-01-20 18:28:35 +13001057 __attach_mnt(mnt, parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001058 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059}
1060
Al Viro909b0a82011-11-25 03:06:56 -05001061static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
Al Viro6b41d532011-11-24 23:24:33 -05001063 struct list_head *next = p->mnt_mounts.next;
1064 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -05001066 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001068 next = p->mnt_child.next;
1069 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 break;
Al Viro0714a532011-11-24 22:19:58 -05001071 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073 }
Al Viro6b41d532011-11-24 23:24:33 -05001074 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
Al Viro315fc832011-11-24 18:57:30 -05001077static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -05001078{
Al Viro6b41d532011-11-24 23:24:33 -05001079 struct list_head *prev = p->mnt_mounts.prev;
1080 while (prev != &p->mnt_mounts) {
1081 p = list_entry(prev, struct mount, mnt_child);
1082 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -05001083 }
1084 return p;
1085}
1086
Al Viro8f2918892018-11-04 06:48:34 -05001087/**
1088 * vfs_create_mount - Create a mount for a configured superblock
1089 * @fc: The configuration context with the superblock attached
1090 *
1091 * Create a mount to an already configured superblock. If necessary, the
1092 * caller should invoke vfs_get_tree() before calling this.
1093 *
1094 * Note that this does not attach the mount to anything.
1095 */
1096struct vfsmount *vfs_create_mount(struct fs_context *fc)
Al Viro9d412a42011-03-17 22:08:28 -04001097{
Al Virob105e272011-11-24 20:38:33 -05001098 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001099
Al Viro8f2918892018-11-04 06:48:34 -05001100 if (!fc->root)
1101 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -04001102
Al Viro8f2918892018-11-04 06:48:34 -05001103 mnt = alloc_vfsmnt(fc->source ?: "none");
Al Viro9d412a42011-03-17 22:08:28 -04001104 if (!mnt)
1105 return ERR_PTR(-ENOMEM);
1106
Al Viro8f2918892018-11-04 06:48:34 -05001107 if (fc->sb_flags & SB_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -05001108 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -04001109
Al Viro8f2918892018-11-04 06:48:34 -05001110 atomic_inc(&fc->root->d_sb->s_active);
1111 mnt->mnt.mnt_sb = fc->root->d_sb;
1112 mnt->mnt.mnt_root = dget(fc->root);
1113 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1114 mnt->mnt_parent = mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001115
Al Viro719ea2f2013-09-29 11:24:49 -04001116 lock_mount_hash();
Al Viro8f2918892018-11-04 06:48:34 -05001117 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001118 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -05001119 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001120}
Al Viro8f2918892018-11-04 06:48:34 -05001121EXPORT_SYMBOL(vfs_create_mount);
1122
1123struct vfsmount *fc_mount(struct fs_context *fc)
1124{
1125 int err = vfs_get_tree(fc);
1126 if (!err) {
1127 up_write(&fc->root->d_sb->s_umount);
1128 return vfs_create_mount(fc);
1129 }
1130 return ERR_PTR(err);
1131}
1132EXPORT_SYMBOL(fc_mount);
1133
David Howells9bc61ab2018-11-04 03:19:03 -05001134struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1135 int flags, const char *name,
1136 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
David Howells9bc61ab2018-11-04 03:19:03 -05001138 struct fs_context *fc;
Al Viro8f2918892018-11-04 06:48:34 -05001139 struct vfsmount *mnt;
David Howells9bc61ab2018-11-04 03:19:03 -05001140 int ret = 0;
Al Viro9d412a42011-03-17 22:08:28 -04001141
1142 if (!type)
David Howells3e1aeb02018-11-01 23:07:25 +00001143 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -04001144
David Howells9bc61ab2018-11-04 03:19:03 -05001145 fc = fs_context_for_mount(type, flags);
1146 if (IS_ERR(fc))
1147 return ERR_CAST(fc);
1148
David Howells3e1aeb02018-11-01 23:07:25 +00001149 if (name)
1150 ret = vfs_parse_fs_string(fc, "source",
1151 name, strlen(name));
David Howells9bc61ab2018-11-04 03:19:03 -05001152 if (!ret)
1153 ret = parse_monolithic_mount_data(fc, data);
1154 if (!ret)
Al Viro8f2918892018-11-04 06:48:34 -05001155 mnt = fc_mount(fc);
1156 else
1157 mnt = ERR_PTR(ret);
Al Viro9d412a42011-03-17 22:08:28 -04001158
David Howells9bc61ab2018-11-04 03:19:03 -05001159 put_fs_context(fc);
Al Viro8f2918892018-11-04 06:48:34 -05001160 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001161}
1162EXPORT_SYMBOL_GPL(vfs_kern_mount);
1163
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001164struct vfsmount *
1165vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1166 const char *name, void *data)
1167{
1168 /* Until it is worked out how to pass the user namespace
1169 * through from the parent mount to the submount don't support
1170 * unprivileged mounts with submounts.
1171 */
1172 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1173 return ERR_PTR(-EPERM);
1174
David Howellse462ec52017-07-17 08:45:35 +01001175 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001176}
1177EXPORT_SYMBOL_GPL(vfs_submount);
1178
Al Viro87129cc2011-11-24 21:24:27 -05001179static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -05001180 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
Al Viro87129cc2011-11-24 21:24:27 -05001182 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +01001183 struct mount *mnt;
1184 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
David Howellsbe34d1a2012-06-25 12:55:18 +01001186 mnt = alloc_vfsmnt(old->mnt_devname);
1187 if (!mnt)
1188 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001189
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001190 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +01001191 mnt->mnt_group_id = 0; /* not a peer of original */
1192 else
1193 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001194
David Howellsbe34d1a2012-06-25 12:55:18 +01001195 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1196 err = mnt_alloc_group_id(mnt);
1197 if (err)
1198 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001200
Al Viro16a34ad2018-04-19 22:03:08 -04001201 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001202 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL|MNT_ONRB);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001203
David Howellsbe34d1a2012-06-25 12:55:18 +01001204 atomic_inc(&sb->s_active);
Christian Brauner256c8ae2022-10-26 12:51:27 +02001205 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1206
David Howellsbe34d1a2012-06-25 12:55:18 +01001207 mnt->mnt.mnt_sb = sb;
1208 mnt->mnt.mnt_root = dget(root);
1209 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1210 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001211 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001212 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001213 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001214
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001215 if ((flag & CL_SLAVE) ||
1216 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001217 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1218 mnt->mnt_master = old;
1219 CLEAR_MNT_SHARED(mnt);
1220 } else if (!(flag & CL_PRIVATE)) {
1221 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1222 list_add(&mnt->mnt_share, &old->mnt_share);
1223 if (IS_MNT_SLAVE(old))
1224 list_add(&mnt->mnt_slave, &old->mnt_slave);
1225 mnt->mnt_master = old->mnt_master;
Al Viro5235d442016-11-20 19:33:09 -05001226 } else {
1227 CLEAR_MNT_SHARED(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001228 }
1229 if (flag & CL_MAKE_SHARED)
1230 set_mnt_shared(mnt);
1231
1232 /* stick the duplicate mount on the same expiry list
1233 * as the original if that was on one */
1234 if (flag & CL_EXPIRE) {
1235 if (!list_empty(&old->mnt_expire))
1236 list_add(&mnt->mnt_expire, &old->mnt_expire);
1237 }
1238
Al Virocb338d02011-11-24 20:55:08 -05001239 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001240
1241 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001242 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001243 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001244 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245}
1246
Al Viro9ea459e12014-08-08 13:08:20 -04001247static void cleanup_mnt(struct mount *mnt)
1248{
Al Viro56cbb4292019-07-04 16:57:51 -04001249 struct hlist_node *p;
1250 struct mount *m;
Al Viro9ea459e12014-08-08 13:08:20 -04001251 /*
Al Viro56cbb4292019-07-04 16:57:51 -04001252 * The warning here probably indicates that somebody messed
1253 * up a mnt_want/drop_write() pair. If this happens, the
1254 * filesystem was probably unable to make r/w->r/o transitions.
Al Viro9ea459e12014-08-08 13:08:20 -04001255 * The locking used to deal with mnt_count decrement provides barriers,
1256 * so mnt_get_writers() below is safe.
1257 */
1258 WARN_ON(mnt_get_writers(mnt));
1259 if (unlikely(mnt->mnt_pins.first))
1260 mnt_pin_kill(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001261 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1262 hlist_del(&m->mnt_umount);
1263 mntput(&m->mnt);
1264 }
Al Viro9ea459e12014-08-08 13:08:20 -04001265 fsnotify_vfsmount_delete(&mnt->mnt);
1266 dput(mnt->mnt.mnt_root);
1267 deactivate_super(mnt->mnt.mnt_sb);
1268 mnt_free_id(mnt);
1269 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1270}
1271
1272static void __cleanup_mnt(struct rcu_head *head)
1273{
1274 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1275}
1276
1277static LLIST_HEAD(delayed_mntput_list);
1278static void delayed_mntput(struct work_struct *unused)
1279{
1280 struct llist_node *node = llist_del_all(&delayed_mntput_list);
Byungchul Park29785732017-08-07 17:44:45 +09001281 struct mount *m, *t;
Al Viro9ea459e12014-08-08 13:08:20 -04001282
Byungchul Park29785732017-08-07 17:44:45 +09001283 llist_for_each_entry_safe(m, t, node, mnt_llist)
1284 cleanup_mnt(m);
Al Viro9ea459e12014-08-08 13:08:20 -04001285}
1286static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1287
Al Viro900148d2011-11-25 00:33:11 -05001288static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001289{
Al Viro4edbe132019-06-30 10:39:08 -04001290 LIST_HEAD(list);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001291 int count;
Al Viro4edbe132019-06-30 10:39:08 -04001292
Al Viro48a066e2013-09-29 22:06:07 -04001293 rcu_read_lock();
Al Viro9ea0a462018-08-09 17:21:17 -04001294 if (likely(READ_ONCE(mnt->mnt_ns))) {
1295 /*
1296 * Since we don't do lock_mount_hash() here,
1297 * ->mnt_ns can change under us. However, if it's
1298 * non-NULL, then there's a reference that won't
1299 * be dropped until after an RCU delay done after
1300 * turning ->mnt_ns NULL. So if we observe it
1301 * non-NULL under rcu_read_lock(), the reference
1302 * we are dropping is not the final one.
1303 */
1304 mnt_add_count(mnt, -1);
Al Viro48a066e2013-09-29 22:06:07 -04001305 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001306 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001307 }
Al Viro719ea2f2013-09-29 11:24:49 -04001308 lock_mount_hash();
Al Viro119e1ef2018-08-09 17:51:32 -04001309 /*
1310 * make sure that if __legitimize_mnt() has not seen us grab
1311 * mount_lock, we'll see their refcount increment here.
1312 */
1313 smp_mb();
Al Viro9ea0a462018-08-09 17:21:17 -04001314 mnt_add_count(mnt, -1);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001315 count = mnt_get_count(mnt);
1316 if (count != 0) {
1317 WARN_ON(count < 0);
Al Viro48a066e2013-09-29 22:06:07 -04001318 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001319 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001320 return;
1321 }
Al Viro48a066e2013-09-29 22:06:07 -04001322 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1323 rcu_read_unlock();
1324 unlock_mount_hash();
1325 return;
1326 }
1327 mnt->mnt.mnt_flags |= MNT_DOOMED;
1328 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301329
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001330 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001331
1332 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1333 struct mount *p, *tmp;
1334 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
Al Viro4edbe132019-06-30 10:39:08 -04001335 __put_mountpoint(unhash_mnt(p), &list);
Al Viro56cbb4292019-07-04 16:57:51 -04001336 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001337 }
1338 }
Al Viro719ea2f2013-09-29 11:24:49 -04001339 unlock_mount_hash();
Al Viro4edbe132019-06-30 10:39:08 -04001340 shrink_dentry_list(&list);
Al Viro649a7952013-09-28 12:41:25 -04001341
Al Viro9ea459e12014-08-08 13:08:20 -04001342 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1343 struct task_struct *task = current;
1344 if (likely(!(task->flags & PF_KTHREAD))) {
1345 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
Jens Axboe91989c72020-10-16 09:02:26 -06001346 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
Al Viro9ea459e12014-08-08 13:08:20 -04001347 return;
1348 }
1349 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1350 schedule_delayed_work(&delayed_mntput_work, 1);
1351 return;
1352 }
1353 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001354}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001355
1356void mntput(struct vfsmount *mnt)
1357{
1358 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001359 struct mount *m = real_mount(mnt);
Mateusz Guzik6c4d1c92023-10-04 13:19:16 +02001360 /* avoid cacheline pingpong */
Al Viro863d6842011-11-25 00:57:42 -05001361 if (unlikely(m->mnt_expiry_mark))
Mateusz Guzik6c4d1c92023-10-04 13:19:16 +02001362 WRITE_ONCE(m->mnt_expiry_mark, 0);
Al Viro863d6842011-11-25 00:57:42 -05001363 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001364 }
1365}
1366EXPORT_SYMBOL(mntput);
1367
1368struct vfsmount *mntget(struct vfsmount *mnt)
1369{
1370 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001371 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001372 return mnt;
1373}
1374EXPORT_SYMBOL(mntget);
1375
Rik van Rielda27f792023-01-27 13:46:51 -05001376/*
1377 * Make a mount point inaccessible to new lookups.
1378 * Because there may still be current users, the caller MUST WAIT
1379 * for an RCU grace period before destroying the mount point.
1380 */
1381void mnt_make_shortterm(struct vfsmount *mnt)
1382{
1383 if (mnt)
1384 real_mount(mnt)->mnt_ns = NULL;
1385}
1386
Randy Dunlap1f287bc2021-03-17 19:52:25 -07001387/**
1388 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1389 * @path: path to check
Ian Kentc6609c02016-11-24 08:03:41 +11001390 *
1391 * d_mountpoint() can only be used reliably to establish if a dentry is
1392 * not mounted in any namespace and that common case is handled inline.
1393 * d_mountpoint() isn't aware of the possibility there may be multiple
1394 * mounts using a given dentry in a different namespace. This function
1395 * checks if the passed in path is a mountpoint rather than the dentry
1396 * alone.
1397 */
1398bool path_is_mountpoint(const struct path *path)
1399{
1400 unsigned seq;
1401 bool res;
1402
1403 if (!d_mountpoint(path->dentry))
1404 return false;
1405
1406 rcu_read_lock();
1407 do {
1408 seq = read_seqbegin(&mount_lock);
1409 res = __path_is_mountpoint(path);
1410 } while (read_seqretry(&mount_lock, seq));
1411 rcu_read_unlock();
1412
1413 return res;
1414}
1415EXPORT_SYMBOL(path_is_mountpoint);
1416
Al Viroca71cf72016-11-20 19:45:28 -05001417struct vfsmount *mnt_clone_internal(const struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001418{
Al Viro3064c352014-08-07 09:12:31 -04001419 struct mount *p;
1420 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1421 if (IS_ERR(p))
1422 return ERR_CAST(p);
1423 p->mnt.mnt_flags |= MNT_INTERNAL;
1424 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001425}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001427/*
1428 * Returns the mount which either has the specified mnt_id, or has the next
1429 * smallest id afer the specified one.
1430 */
1431static struct mount *mnt_find_id_at(struct mnt_namespace *ns, u64 mnt_id)
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001432{
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001433 struct rb_node *node = ns->mounts.rb_node;
1434 struct mount *ret = NULL;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001435
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001436 while (node) {
1437 struct mount *m = node_to_mount(node);
1438
1439 if (mnt_id <= m->mnt_id_unique) {
1440 ret = node_to_mount(node);
1441 if (mnt_id == m->mnt_id_unique)
1442 break;
1443 node = node->rb_left;
1444 } else {
1445 node = node->rb_right;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001446 }
1447 }
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001448 return ret;
1449}
1450
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001451#ifdef CONFIG_PROC_FS
1452
Al Viro0226f492011-12-06 12:21:54 -05001453/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454static void *m_start(struct seq_file *m, loff_t *pos)
1455{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001456 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Ram Pai390c6842005-11-07 17:17:51 -05001458 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001459
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001460 return mnt_find_id_at(p->ns, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
1463static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1464{
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001465 struct mount *next = NULL, *mnt = v;
1466 struct rb_node *node = rb_next(&mnt->mnt_node);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001467
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001468 ++*pos;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001469 if (node) {
1470 next = node_to_mount(node);
1471 *pos = next->mnt_id_unique;
1472 }
1473 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
1476static void m_stop(struct seq_file *m, void *v)
1477{
Ram Pai390c6842005-11-07 17:17:51 -05001478 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479}
1480
Al Viro0226f492011-12-06 12:21:54 -05001481static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001482{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001483 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001484 struct mount *r = v;
Al Viro0226f492011-12-06 12:21:54 -05001485 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486}
1487
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001488const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 .start = m_start,
1490 .next = m_next,
1491 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001492 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001493};
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001494
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001495#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497/**
1498 * may_umount_tree - check if a mount tree is busy
Randy Dunlap1f287bc2021-03-17 19:52:25 -07001499 * @m: root of mount tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 *
1501 * This is called to check if a tree of mounts has any
1502 * open files, pwds, chroots or sub mounts that are
1503 * busy.
1504 */
Al Viro909b0a82011-11-25 03:06:56 -05001505int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
Al Viro909b0a82011-11-25 03:06:56 -05001507 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001508 int actual_refs = 0;
1509 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001510 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001511 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Nick Pigginb3e19d92011-01-07 17:50:11 +11001513 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001514 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001515 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001516 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 }
Al Viro719ea2f2013-09-29 11:24:49 -04001519 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001522 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Ian Kente3474a82006-03-27 01:14:51 -08001524 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
1527EXPORT_SYMBOL(may_umount_tree);
1528
1529/**
1530 * may_umount - check if a mount point is busy
1531 * @mnt: root of mount
1532 *
1533 * This is called to check if a mount point has any
1534 * open files, pwds, chroots or sub mounts. If the
1535 * mount has sub mounts this will return busy
1536 * regardless of whether the sub mounts are busy.
1537 *
1538 * Doesn't take quota and stuff into account. IOW, in some cases it will
1539 * give false negatives. The main reason why it's here is that we need
1540 * a non-destructive way to look for easily umountable filesystems.
1541 */
1542int may_umount(struct vfsmount *mnt)
1543{
Ian Kente3474a82006-03-27 01:14:51 -08001544 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001545 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001546 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001547 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001548 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001549 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001550 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001551 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552}
1553
1554EXPORT_SYMBOL(may_umount);
1555
Al Viro97216be2013-03-16 15:12:40 -04001556static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001558 struct hlist_head head;
Al Viro56cbb4292019-07-04 16:57:51 -04001559 struct hlist_node *p;
1560 struct mount *m;
Al Viro4edbe132019-06-30 10:39:08 -04001561 LIST_HEAD(list);
Al Viro97216be2013-03-16 15:12:40 -04001562
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001563 hlist_move_list(&unmounted, &head);
Al Viro4edbe132019-06-30 10:39:08 -04001564 list_splice_init(&ex_mountpoints, &list);
Al Viro97216be2013-03-16 15:12:40 -04001565
Al Viro97216be2013-03-16 15:12:40 -04001566 up_write(&namespace_sem);
1567
Al Viro4edbe132019-06-30 10:39:08 -04001568 shrink_dentry_list(&list);
1569
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001570 if (likely(hlist_empty(&head)))
1571 return;
1572
NeilBrown22cb7402018-11-30 10:33:18 +11001573 synchronize_rcu_expedited();
Al Viro48a066e2013-09-29 22:06:07 -04001574
Al Viro56cbb4292019-07-04 16:57:51 -04001575 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1576 hlist_del(&m->mnt_umount);
1577 mntput(&m->mnt);
1578 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001579}
1580
Al Viro97216be2013-03-16 15:12:40 -04001581static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001582{
Al Viro97216be2013-03-16 15:12:40 -04001583 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001584}
1585
Eric W. Biedermane819f152014-12-24 07:20:01 -06001586enum umount_tree_flags {
1587 UMOUNT_SYNC = 1,
1588 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001589 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001590};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001591
1592static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1593{
1594 /* Leaving mounts connected is only valid for lazy umounts */
1595 if (how & UMOUNT_SYNC)
1596 return true;
1597
1598 /* A mount without a parent has nothing to be connected to */
1599 if (!mnt_has_parent(mnt))
1600 return true;
1601
1602 /* Because the reference counting rules change when mounts are
1603 * unmounted and connected, umounted mounts may not be
1604 * connected to mounted mounts.
1605 */
1606 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1607 return true;
1608
1609 /* Has it been requested that the mount remain connected? */
1610 if (how & UMOUNT_CONNECTED)
1611 return false;
1612
1613 /* Is the mount locked such that it needs to remain connected? */
1614 if (IS_MNT_LOCKED(mnt))
1615 return false;
1616
1617 /* By default disconnect the mount */
1618 return true;
1619}
1620
Nick Piggin99b7db72010-08-18 04:37:39 +10001621/*
Al Viro48a066e2013-09-29 22:06:07 -04001622 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001623 * namespace_sem must be held for write
1624 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001625static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001626{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001627 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001628 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001629
Eric W. Biederman5d884572015-01-03 05:39:35 -06001630 if (how & UMOUNT_PROPAGATE)
1631 propagate_mount_unlock(mnt);
1632
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001633 /* Gather the mounts to umount */
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001634 for (p = mnt; p; p = next_mnt(p, mnt)) {
1635 p->mnt.mnt_flags |= MNT_UMOUNT;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001636 if (p->mnt.mnt_flags & MNT_ONRB)
1637 move_from_ns(p, &tmp_list);
1638 else
1639 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001640 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001641
Eric W. Biederman411a9382014-12-22 19:12:07 -06001642 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001643 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001644 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001645 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001646
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001647 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001648 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001649 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001650
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001651 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001652 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001653 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001654 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001655 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001656 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001657 ns = p->mnt_ns;
1658 if (ns) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001659 ns->nr_mounts--;
Eric W. Biedermand2921682016-09-28 00:27:17 -05001660 __touch_mnt_namespace(ns);
1661 }
Al Viro143c8c92011-11-25 00:46:35 -05001662 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001663 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001664 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001665
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001666 disconnect = disconnect_mount(p, how);
Al Viro676da582011-11-24 21:47:05 -05001667 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001668 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001669 if (!disconnect) {
1670 /* Don't forget about p */
1671 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1672 } else {
1673 umount_mnt(p);
1674 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001675 }
Al Viro0f0afb12011-11-24 20:43:10 -05001676 change_mnt_propagation(p, MS_PRIVATE);
Al Viro19a1c402019-07-24 12:45:46 -04001677 if (disconnect)
1678 hlist_add_head(&p->mnt_umount, &unmounted);
Al Viro38129a12014-03-20 21:10:51 -04001679 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680}
1681
Al Virob54b9be2013-03-16 14:39:34 -04001682static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001683
David Howells8d0347f2018-11-04 09:28:36 -05001684static int do_umount_root(struct super_block *sb)
1685{
1686 int ret = 0;
1687
1688 down_write(&sb->s_umount);
1689 if (!sb_rdonly(sb)) {
1690 struct fs_context *fc;
1691
1692 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1693 SB_RDONLY);
1694 if (IS_ERR(fc)) {
1695 ret = PTR_ERR(fc);
1696 } else {
1697 ret = parse_monolithic_mount_data(fc, NULL);
1698 if (!ret)
1699 ret = reconfigure_super(fc);
1700 put_fs_context(fc);
1701 }
1702 }
1703 up_write(&sb->s_umount);
1704 return ret;
1705}
1706
Al Viro1ab59732011-11-24 21:35:16 -05001707static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708{
Al Viro1ab59732011-11-24 21:35:16 -05001709 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 int retval;
1711
Al Viro1ab59732011-11-24 21:35:16 -05001712 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 if (retval)
1714 return retval;
1715
1716 /*
1717 * Allow userspace to request a mountpoint be expired rather than
1718 * unmounting unconditionally. Unmount only happens if:
1719 * (1) the mark is already set (the mark is cleared by mntput())
1720 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1721 */
1722 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001723 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 flags & (MNT_FORCE | MNT_DETACH))
1725 return -EINVAL;
1726
Nick Pigginb3e19d92011-01-07 17:50:11 +11001727 /*
1728 * probably don't strictly need the lock here if we examined
1729 * all race cases, but it's a slowpath.
1730 */
Al Viro719ea2f2013-09-29 11:24:49 -04001731 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001732 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001733 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001735 }
Al Viro719ea2f2013-09-29 11:24:49 -04001736 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Al Viro863d6842011-11-25 00:57:42 -05001738 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 return -EAGAIN;
1740 }
1741
1742 /*
1743 * If we may have to abort operations to get out of this
1744 * mount, and they will themselves hold resources we must
1745 * allow the fs to do things. In the Unix tradition of
1746 * 'Gee thats tricky lets do it in userspace' the umount_begin
1747 * might fail to complete on the first run through as other tasks
1748 * must return, and the like. Thats for the mount program to worry
1749 * about for the moment.
1750 */
1751
Al Viro42faad92008-04-24 07:21:56 -04001752 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001753 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 /*
1757 * No sense to grab the lock for this test, but test itself looks
1758 * somewhat bogus. Suggestions for better replacement?
1759 * Ho-hum... In principle, we might treat that as umount + switch
1760 * to rootfs. GC would eventually take care of the old vfsmount.
1761 * Actually it makes sense, especially if rootfs would contain a
1762 * /reboot - static binary that would close all descriptors and
1763 * call reboot(9). Then init(8) could umount root and exec /reboot.
1764 */
Al Viro1ab59732011-11-24 21:35:16 -05001765 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 /*
1767 * Special case for "unmounting" root ...
1768 * we just try to remount it readonly.
1769 */
Eric W. Biedermanbc6155d2017-09-18 17:58:08 -05001770 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001771 return -EPERM;
David Howells8d0347f2018-11-04 09:28:36 -05001772 return do_umount_root(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 }
1774
Al Viro97216be2013-03-16 15:12:40 -04001775 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001776 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001778 /* Recheck MNT_LOCKED with the locks held */
1779 retval = -EINVAL;
1780 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1781 goto out;
1782
1783 event++;
Al Viro48a066e2013-09-29 22:06:07 -04001784 if (flags & MNT_DETACH) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001785 if (mnt->mnt.mnt_flags & MNT_ONRB ||
1786 !list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001787 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001789 } else {
1790 shrink_submounts(mnt);
1791 retval = -EBUSY;
1792 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02001793 if (mnt->mnt.mnt_flags & MNT_ONRB ||
1794 !list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001795 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001796 retval = 0;
1797 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 }
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001799out:
Al Viro719ea2f2013-09-29 11:24:49 -04001800 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001801 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 return retval;
1803}
1804
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001805/*
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();
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001821 lock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001822 mp = lookup_mountpoint(dentry);
Al Viroadc9b5c2019-06-29 12:06:51 -04001823 if (!mp)
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001824 goto out_unlock;
1825
Andrey Ulanove06b9332016-04-15 14:24:41 -07001826 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001827 while (!hlist_empty(&mp->m_list)) {
1828 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001829 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc82015-07-17 14:54:27 -05001830 umount_mnt(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001831 hlist_add_head(&mnt->mnt_umount, &unmounted);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001832 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001833 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001834 }
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001835 put_mountpoint(mp);
1836out_unlock:
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001837 unlock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001838 namespace_unlock();
1839}
1840
David Howellsdd111b32017-07-04 17:25:09 +01001841/*
Al Viro9b40bc92013-02-22 22:45:42 -05001842 * Is the caller allowed to modify his namespace?
1843 */
Al Viroa5f85d72022-03-01 00:05:29 -05001844bool may_mount(void)
Al Viro9b40bc92013-02-22 22:45:42 -05001845{
1846 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1847}
1848
Christian Brauner78aa08a2023-05-03 13:18:39 +02001849/**
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
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04001862static void warn_mandlock(void)
Jeff Layton9e8925b2015-11-16 09:49:34 -05001863{
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04001864 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");
Jeff Layton9e8925b2015-11-16 09:49:34 -05001869}
1870
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001871static int can_umount(const struct path *path, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001873 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
Al Viro9b40bc92013-02-22 22:45:42 -05001875 if (!may_mount())
1876 return -EPERM;
Christian Brauner78aa08a2023-05-03 13:18:39 +02001877 if (!path_mounted(path))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001878 return -EINVAL;
Al Viro143c8c92011-11-25 00:46:35 -05001879 if (!check_mnt(mnt))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001880 return -EINVAL;
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001881 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001882 return -EINVAL;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001883 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001884 return -EPERM;
1885 return 0;
1886}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Al Viroa0a6df92021-01-04 15:25:34 -05001888// caller is responsible for flags being sane
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001889int path_umount(struct path *path, int flags)
1890{
1891 struct mount *mnt = real_mount(path->mnt);
1892 int ret;
1893
1894 ret = can_umount(path, flags);
1895 if (!ret)
1896 ret = do_umount(mnt, flags);
1897
Jan Blunck429731b2008-02-14 19:34:31 -08001898 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Christoph Hellwig41525f52020-07-21 10:54:34 +02001899 dput(path->dentry);
Al Viro900148d2011-11-25 00:33:11 -05001900 mntput_no_expire(mnt);
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001901 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902}
1903
Christoph Hellwig09267de2020-07-23 08:23:08 +02001904static int ksys_umount(char __user *name, int flags)
Christoph Hellwig41525f52020-07-21 10:54:34 +02001905{
1906 int lookup_flags = LOOKUP_MOUNTPOINT;
1907 struct path path;
1908 int ret;
1909
Al Viroa0a6df92021-01-04 15:25:34 -05001910 // basic validity checks done first
1911 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1912 return -EINVAL;
1913
Christoph Hellwig41525f52020-07-21 10:54:34 +02001914 if (!(flags & UMOUNT_NOFOLLOW))
1915 lookup_flags |= LOOKUP_FOLLOW;
1916 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1917 if (ret)
1918 return ret;
1919 return path_umount(&path, flags);
1920}
1921
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001922SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1923{
1924 return ksys_umount(name, flags);
1925}
1926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1928
1929/*
Ram Paib58fed82005-11-07 17:16:09 -05001930 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001932SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933{
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001934 return ksys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
1936
1937#endif
1938
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001939static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001940{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001941 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001942 return dentry->d_op == &ns_dentry_operations &&
1943 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001944}
1945
Eric Biggers213921f2020-01-04 12:59:55 -08001946static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
Al Viro58be28252014-11-01 00:00:23 -04001947{
1948 return container_of(ns, struct mnt_namespace, ns);
1949}
1950
Christian Brauner303cc572020-05-05 16:04:31 +02001951struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1952{
1953 return &mnt->ns;
1954}
1955
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001956static 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
Al Virof77c8012014-11-01 03:13:17 -04001965 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001966 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1967}
1968
Al Viro87129cc2011-11-24 21:24:27 -05001969struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001970 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971{
Al Viro84d171922013-03-15 10:53:28 -04001972 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001974 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1975 return ERR_PTR(-EINVAL);
1976
1977 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001978 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001979
Ram Pai36341f62005-11-07 17:17:22 -05001980 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001981 if (IS_ERR(q))
1982 return q;
1983
Al Viroa73324d2011-11-24 22:25:07 -05001984 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001987 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001988 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001989 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 continue;
1991
Al Viro909b0a82011-11-25 03:06:56 -05001992 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001993 if (!(flag & CL_COPY_UNBINDABLE) &&
1994 IS_MNT_UNBINDABLE(s)) {
Eric W. Biedermandf7342b2018-10-25 09:04:18 -05001995 if (s->mnt.mnt_flags & MNT_LOCKED) {
1996 /* Both unbindable and locked. */
1997 q = ERR_PTR(-EPERM);
1998 goto out;
1999 } else {
2000 s = skip_mnt_tree(s);
2001 continue;
2002 }
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002003 }
2004 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2005 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05002006 s = skip_mnt_tree(s);
2007 continue;
2008 }
Al Viro0714a532011-11-24 22:19:58 -05002009 while (p != s->mnt_parent) {
2010 p = p->mnt_parent;
2011 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 }
Al Viro87129cc2011-11-24 21:24:27 -05002013 p = s;
Al Viro84d171922013-03-15 10:53:28 -04002014 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05002015 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01002016 if (IS_ERR(q))
2017 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04002018 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05002019 list_add_tail(&q->mnt_list, &res->mnt_list);
Christian Brauner6ac39282023-05-03 13:18:42 +02002020 attach_mnt(q, parent, p->mnt_mp, false);
Al Viro719ea2f2013-09-29 11:24:49 -04002021 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 }
2023 }
2024 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01002025out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04002027 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002028 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002029 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
David Howellsbe34d1a2012-06-25 12:55:18 +01002031 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
David Howellsbe34d1a2012-06-25 12:55:18 +01002034/* Caller should check returned pointer for errors */
2035
Al Viroca71cf72016-11-20 19:45:28 -05002036struct vfsmount *collect_mounts(const struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04002037{
Al Virocb338d02011-11-24 20:55:08 -05002038 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04002039 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06002040 if (!check_mnt(real_mount(path->mnt)))
2041 tree = ERR_PTR(-EINVAL);
2042 else
2043 tree = copy_tree(real_mount(path->mnt), path->dentry,
2044 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04002045 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01002046 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03002047 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01002048 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04002049}
2050
Al Viroa07b2002018-11-05 17:40:30 +00002051static 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;
David Howells44dfd842018-11-05 17:40:31 +00002060 if (ns) {
2061 if (is_anon_ns(ns))
2062 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2063 else
2064 ns = NULL;
2065 }
Al Viroa07b2002018-11-05 17:40:30 +00002066 unlock_mount_hash();
2067 namespace_unlock();
David Howells44dfd842018-11-05 17:40:31 +00002068 if (ns)
2069 free_mnt_ns(ns);
Al Viroa07b2002018-11-05 17:40:30 +00002070}
2071
Al Viro8aec0802007-06-07 12:20:32 -04002072void drop_collected_mounts(struct vfsmount *mnt)
2073{
Al Viro97216be2013-03-16 15:12:40 -04002074 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002075 lock_mount_hash();
Eric W. Biederman9c8e0a12018-10-25 12:05:11 -05002076 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04002077 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002078 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04002079}
2080
Miklos Szeredi427215d2021-08-09 10:19:47 +02002081static 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
Miklos Szeredic771d682014-10-24 00:14:36 +02002095/**
2096 * clone_private_mount - create a private clone of a path
Randy Dunlap1f287bc2021-03-17 19:52:25 -07002097 * @path: path to clone
Miklos Szeredic771d682014-10-24 00:14:36 +02002098 *
Randy Dunlap1f287bc2021-03-17 19:52:25 -07002099 * 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).
Miklos Szeredic771d682014-10-24 00:14:36 +02002102 *
2103 * Release with mntput().
2104 */
Al Viroca71cf72016-11-20 19:45:28 -05002105struct vfsmount *clone_private_mount(const struct path *path)
Miklos Szeredic771d682014-10-24 00:14:36 +02002106{
2107 struct mount *old_mnt = real_mount(path->mnt);
2108 struct mount *new_mnt;
2109
Miklos Szeredi427215d2021-08-09 10:19:47 +02002110 down_read(&namespace_sem);
Miklos Szeredic771d682014-10-24 00:14:36 +02002111 if (IS_MNT_UNBINDABLE(old_mnt))
Miklos Szeredi427215d2021-08-09 10:19:47 +02002112 goto invalid;
2113
2114 if (!check_mnt(old_mnt))
2115 goto invalid;
2116
2117 if (has_locked_children(old_mnt, path->dentry))
2118 goto invalid;
Miklos Szeredic771d682014-10-24 00:14:36 +02002119
Miklos Szeredic771d682014-10-24 00:14:36 +02002120 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
Miklos Szeredi427215d2021-08-09 10:19:47 +02002121 up_read(&namespace_sem);
2122
Miklos Szeredic771d682014-10-24 00:14:36 +02002123 if (IS_ERR(new_mnt))
2124 return ERR_CAST(new_mnt);
2125
Miklos Szeredidf820f82020-06-04 10:48:19 +02002126 /* Longterm mount to be removed by kern_unmount*() */
2127 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2128
Miklos Szeredic771d682014-10-24 00:14:36 +02002129 return &new_mnt->mnt;
Miklos Szeredi427215d2021-08-09 10:19:47 +02002130
2131invalid:
2132 up_read(&namespace_sem);
2133 return ERR_PTR(-EINVAL);
Miklos Szeredic771d682014-10-24 00:14:36 +02002134}
2135EXPORT_SYMBOL_GPL(clone_private_mount);
2136
Al Viro1f707132010-01-30 22:51:25 -05002137int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2138 struct vfsmount *root)
2139{
Al Viro1a4eeaf2011-11-25 02:19:55 -05002140 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05002141 int res = f(root, arg);
2142 if (res)
2143 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002144 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2145 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05002146 if (res)
2147 return res;
2148 }
2149 return 0;
2150}
2151
Al Viro3bd045c2019-01-30 13:15:45 -05002152static void lock_mnt_tree(struct mount *mnt)
2153{
2154 struct mount *p;
2155
2156 for (p = mnt; p; p = next_mnt(p, 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(&p->mnt_expire))
2174 flags |= MNT_LOCKED;
2175 p->mnt.mnt_flags = flags;
2176 }
2177}
2178
Al Viro4b8b21f2011-11-24 19:54:23 -05002179static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002180{
Al Viro315fc832011-11-24 18:57:30 -05002181 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002182
Al Viro909b0a82011-11-25 03:06:56 -05002183 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002184 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05002185 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002186 }
2187}
2188
Al Viro4b8b21f2011-11-24 19:54:23 -05002189static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002190{
Al Viro315fc832011-11-24 18:57:30 -05002191 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002192
Al Viro909b0a82011-11-25 03:06:56 -05002193 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05002194 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002195 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002196 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002197 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002198 return err;
2199 }
2200 }
2201 }
2202
2203 return 0;
2204}
2205
Eric W. Biedermand2921682016-09-28 00:27:17 -05002206int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2207{
2208 unsigned int max = READ_ONCE(sysctl_mount_max);
Al Viro124f75f2022-02-13 22:42:30 -05002209 unsigned int mounts = 0;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002210 struct mount *p;
2211
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02002212 if (ns->nr_mounts >= max)
Al Viro124f75f2022-02-13 22:42:30 -05002213 return -ENOSPC;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02002214 max -= ns->nr_mounts;
Al Viro124f75f2022-02-13 22:42:30 -05002215 if (ns->pending_mounts >= max)
2216 return -ENOSPC;
2217 max -= ns->pending_mounts;
2218
Eric W. Biedermand2921682016-09-28 00:27:17 -05002219 for (p = mnt; p; p = next_mnt(p, mnt))
2220 mounts++;
2221
Al Viro124f75f2022-02-13 22:42:30 -05002222 if (mounts > max)
Eric W. Biedermand2921682016-09-28 00:27:17 -05002223 return -ENOSPC;
2224
Al Viro124f75f2022-02-13 22:42:30 -05002225 ns->pending_mounts += mounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002226 return 0;
2227}
2228
Christian Brauner6ac39282023-05-03 13:18:42 +02002229enum 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
Ram Paib90fa9a2005-11-07 17:19:50 -05002240 *
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.
Ram Pai9676f0c2005-11-07 17:21:20 -05002243 * ---------------------------------------------------------------------------
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 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05002255 * 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.
Ram Pai5afe0022005-11-07 17:21:01 -05002264 * (+++) 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 *
Ram Pai9676f0c2005-11-07 17:21:20 -05002271 * ---------------------------------------------------------------------------
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 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05002283 *
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.
Ram Pai21444402005-11-07 17:20:03 -05002286 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05002287 * (+++) 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.
Ram Paib90fa9a2005-11-07 17:19:50 -05002291 *
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.
Christian Brauner6ac39282023-05-03 13:18:42 +02002296 *
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.
Ram Paib90fa9a2005-11-07 17:19:50 -05002300 */
Al Viro0fb54e52011-11-24 19:59:16 -05002301static int attach_recursive_mnt(struct mount *source_mnt,
Christian Brauner6ac39282023-05-03 13:18:42 +02002302 struct mount *top_mnt,
2303 struct mountpoint *dest_mp,
2304 enum mnt_tree_flags_t flags)
Ram Paib90fa9a2005-11-07 17:19:50 -05002305{
Al Viro3bd045c2019-01-30 13:15:45 -05002306 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Al Viro38129a12014-03-20 21:10:51 -04002307 HLIST_HEAD(tree_list);
Christian Brauner6ac39282023-05-03 13:18:42 +02002308 struct mnt_namespace *ns = top_mnt->mnt_ns;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002309 struct mountpoint *smp;
Christian Brauner6ac39282023-05-03 13:18:42 +02002310 struct mount *child, *dest_mnt, *p;
Al Viro38129a12014-03-20 21:10:51 -04002311 struct hlist_node *n;
Christian Brauner6ac39282023-05-03 13:18:42 +02002312 int err = 0;
2313 bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
Ram Paib90fa9a2005-11-07 17:19:50 -05002314
Christian Brauner6ac39282023-05-03 13:18:42 +02002315 /*
2316 * Preallocate a mountpoint in case the new mounts need to be
2317 * mounted beneath mounts on the same mountpoint.
Eric W. Biederman1064f872017-01-20 18:28:35 +13002318 */
2319 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2320 if (IS_ERR(smp))
2321 return PTR_ERR(smp);
2322
Eric W. Biedermand2921682016-09-28 00:27:17 -05002323 /* Is there space to add these mounts to the mount namespace? */
Al Viro2763d112019-06-30 19:18:53 -04002324 if (!moving) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05002325 err = count_mounts(ns, source_mnt);
2326 if (err)
2327 goto out;
2328 }
2329
Christian Brauner6ac39282023-05-03 13:18:42 +02002330 if (beneath)
2331 dest_mnt = top_mnt->mnt_parent;
2332 else
2333 dest_mnt = top_mnt;
2334
Al Virofc7be132011-11-25 01:05:37 -05002335 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05002336 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002337 if (err)
2338 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04002339 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Christian Brauner6ac39282023-05-03 13:18:42 +02002340 }
2341 lock_mount_hash();
2342 if (err)
2343 goto out_cleanup_ids;
2344
2345 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro909b0a82011-11-25 03:06:56 -05002346 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05002347 set_mnt_shared(p);
Ram Paib90fa9a2005-11-07 17:19:50 -05002348 }
Christian Brauner6ac39282023-05-03 13:18:42 +02002349
Al Viro2763d112019-06-30 19:18:53 -04002350 if (moving) {
Christian Brauner6ac39282023-05-03 13:18:42 +02002351 if (beneath)
2352 dest_mp = smp;
Al Viro2763d112019-06-30 19:18:53 -04002353 unhash_mnt(source_mnt);
Christian Brauner6ac39282023-05-03 13:18:42 +02002354 attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
Al Viro143c8c92011-11-25 00:46:35 -05002355 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05002356 } else {
David Howells44dfd842018-11-05 17:40:31 +00002357 if (source_mnt->mnt_ns) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02002358 LIST_HEAD(head);
2359
David Howells44dfd842018-11-05 17:40:31 +00002360 /* move from anon - the caller will destroy */
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02002361 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2362 move_from_ns(p, &head);
2363 list_del_init(&head);
David Howells44dfd842018-11-05 17:40:31 +00002364 }
Christian Brauner6ac39282023-05-03 13:18:42 +02002365 if (beneath)
2366 mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
2367 else
2368 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002369 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05002370 }
Ram Paib90fa9a2005-11-07 17:19:50 -05002371
Al Viro38129a12014-03-20 21:10:51 -04002372 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04002373 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04002374 hlist_del_init(&child->mnt_hash);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002375 q = __lookup_mnt(&child->mnt_parent->mnt,
2376 child->mnt_mountpoint);
2377 if (q)
2378 mnt_change_mountpoint(child, smp, q);
Al Viro3bd045c2019-01-30 13:15:45 -05002379 /* Notice when we are propagating across user namespaces */
2380 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2381 lock_mnt_tree(child);
Christian Braunerd728cf72019-06-17 23:22:14 +02002382 child->mnt.mnt_flags &= ~MNT_LOCKED;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002383 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05002384 }
Eric W. Biederman1064f872017-01-20 18:28:35 +13002385 put_mountpoint(smp);
Al Viro719ea2f2013-09-29 11:24:49 -04002386 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10002387
Ram Paib90fa9a2005-11-07 17:19:50 -05002388 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002389
2390 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05002391 while (!hlist_empty(&tree_list)) {
2392 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002393 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06002394 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05002395 }
2396 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002397 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002398 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05002399 ns->pending_mounts = 0;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002400
2401 read_seqlock_excl(&mount_lock);
2402 put_mountpoint(smp);
2403 read_sequnlock_excl(&mount_lock);
2404
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002405 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002406}
2407
Christian Brauner6ac39282023-05-03 13:18:42 +02002408/**
2409 * do_lock_mount - lock mount and mountpoint
2410 * @path: target path
2411 * @beneath: whether the intention is to mount beneath @path
2412 *
2413 * Follow the mount stack on @path until the top mount @mnt is found. If
2414 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2415 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2416 * until nothing is stacked on top of it anymore.
2417 *
2418 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2419 * against concurrent removal of the new mountpoint from another mount
2420 * namespace.
2421 *
2422 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2423 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2424 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2425 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2426 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2427 * on top of it for @beneath.
2428 *
2429 * In addition, @beneath needs to make sure that @mnt hasn't been
2430 * unmounted or moved from its current mountpoint in between dropping
2431 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2432 * being unmounted would be detected later by e.g., calling
2433 * check_mnt(mnt) in the function it's called from. For the @beneath
2434 * case however, it's useful to detect it directly in do_lock_mount().
2435 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2436 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2437 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2438 *
2439 * Return: Either the target mountpoint on the top mount or the top
2440 * mount's mountpoint.
2441 */
2442static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
Al Virob12cea92011-03-18 08:55:38 -04002443{
Christian Brauner6ac39282023-05-03 13:18:42 +02002444 struct vfsmount *mnt = path->mnt;
Christian Brauner64f44b22023-05-03 13:18:41 +02002445 struct dentry *dentry;
Christian Brauner6ac39282023-05-03 13:18:42 +02002446 struct mountpoint *mp = ERR_PTR(-ENOENT);
Christian Brauner64f44b22023-05-03 13:18:41 +02002447
2448 for (;;) {
Christian Brauner6ac39282023-05-03 13:18:42 +02002449 struct mount *m;
2450
2451 if (beneath) {
2452 m = real_mount(mnt);
2453 read_seqlock_excl(&mount_lock);
2454 dentry = dget(m->mnt_mountpoint);
2455 read_sequnlock_excl(&mount_lock);
2456 } else {
2457 dentry = path->dentry;
2458 }
2459
Christian Brauner64f44b22023-05-03 13:18:41 +02002460 inode_lock(dentry->d_inode);
2461 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002462 inode_unlock(dentry->d_inode);
Christian Brauner6ac39282023-05-03 13:18:42 +02002463 goto out;
Al Viro84d171922013-03-15 10:53:28 -04002464 }
Christian Brauner64f44b22023-05-03 13:18:41 +02002465
2466 namespace_lock();
2467
Christian Brauner6ac39282023-05-03 13:18:42 +02002468 if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) {
Al Viro84d171922013-03-15 10:53:28 -04002469 namespace_unlock();
2470 inode_unlock(dentry->d_inode);
Christian Brauner6ac39282023-05-03 13:18:42 +02002471 goto out;
Al Viro84d171922013-03-15 10:53:28 -04002472 }
Christian Brauner6ac39282023-05-03 13:18:42 +02002473
Christian Brauner64f44b22023-05-03 13:18:41 +02002474 mnt = lookup_mnt(path);
2475 if (likely(!mnt))
2476 break;
2477
2478 namespace_unlock();
2479 inode_unlock(dentry->d_inode);
Christian Brauner6ac39282023-05-03 13:18:42 +02002480 if (beneath)
2481 dput(dentry);
Christian Brauner64f44b22023-05-03 13:18:41 +02002482 path_put(path);
2483 path->mnt = mnt;
2484 path->dentry = dget(mnt->mnt_root);
Al Viro84d171922013-03-15 10:53:28 -04002485 }
Christian Brauner64f44b22023-05-03 13:18:41 +02002486
2487 mp = get_mountpoint(dentry);
2488 if (IS_ERR(mp)) {
2489 namespace_unlock();
2490 inode_unlock(dentry->d_inode);
2491 }
2492
Christian Brauner6ac39282023-05-03 13:18:42 +02002493out:
2494 if (beneath)
2495 dput(dentry);
2496
Christian Brauner64f44b22023-05-03 13:18:41 +02002497 return mp;
Al Virob12cea92011-03-18 08:55:38 -04002498}
2499
Christian Brauner6ac39282023-05-03 13:18:42 +02002500static inline struct mountpoint *lock_mount(struct path *path)
2501{
2502 return do_lock_mount(path, false);
Al Virob12cea92011-03-18 08:55:38 -04002503}
2504
Al Viro84d171922013-03-15 10:53:28 -04002505static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002506{
Al Viro84d171922013-03-15 10:53:28 -04002507 struct dentry *dentry = where->m_dentry;
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002508
2509 read_seqlock_excl(&mount_lock);
Al Viro84d171922013-03-15 10:53:28 -04002510 put_mountpoint(where);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002511 read_sequnlock_excl(&mount_lock);
2512
Al Viro328e6d92013-03-16 14:49:45 -04002513 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002514 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002515}
2516
Al Viro84d171922013-03-15 10:53:28 -04002517static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518{
David Howellse462ec52017-07-17 08:45:35 +01002519 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 return -EINVAL;
2521
David Howellse36cb0b2015-01-29 12:02:35 +00002522 if (d_is_dir(mp->m_dentry) !=
2523 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 return -ENOTDIR;
2525
Christian Brauner6ac39282023-05-03 13:18:42 +02002526 return attach_recursive_mnt(mnt, p, mp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527}
2528
2529/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002530 * Sanity check the flags to change_mnt_propagation.
2531 */
2532
David Howellse462ec52017-07-17 08:45:35 +01002533static int flags_to_propagation_type(int ms_flags)
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002534{
David Howellse462ec52017-07-17 08:45:35 +01002535 int type = ms_flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002536
2537 /* Fail if any non-propagation flags are set */
2538 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2539 return 0;
2540 /* Only one propagation flag should be set */
2541 if (!is_power_of_2(type))
2542 return 0;
2543 return type;
2544}
2545
2546/*
Ram Pai07b20882005-11-07 17:19:07 -05002547 * recursively change the type of the mountpoint.
2548 */
David Howellse462ec52017-07-17 08:45:35 +01002549static int do_change_type(struct path *path, int ms_flags)
Ram Pai07b20882005-11-07 17:19:07 -05002550{
Al Viro315fc832011-11-24 18:57:30 -05002551 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002552 struct mount *mnt = real_mount(path->mnt);
David Howellse462ec52017-07-17 08:45:35 +01002553 int recurse = ms_flags & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002554 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002555 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002556
Christian Brauner78aa08a2023-05-03 13:18:39 +02002557 if (!path_mounted(path))
Ram Pai07b20882005-11-07 17:19:07 -05002558 return -EINVAL;
2559
David Howellse462ec52017-07-17 08:45:35 +01002560 type = flags_to_propagation_type(ms_flags);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002561 if (!type)
2562 return -EINVAL;
2563
Al Viro97216be2013-03-16 15:12:40 -04002564 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002565 if (type == MS_SHARED) {
2566 err = invent_group_ids(mnt, recurse);
2567 if (err)
2568 goto out_unlock;
2569 }
2570
Al Viro719ea2f2013-09-29 11:24:49 -04002571 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002572 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002573 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002574 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002575
2576 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002577 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002578 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002579}
2580
Al Viroa07b2002018-11-05 17:40:30 +00002581static struct mount *__do_loopback(struct path *old_path, int recurse)
2582{
2583 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2584
2585 if (IS_MNT_UNBINDABLE(old))
2586 return mnt;
2587
2588 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2589 return mnt;
2590
2591 if (!recurse && has_locked_children(old, old_path->dentry))
2592 return mnt;
2593
2594 if (recurse)
2595 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2596 else
2597 mnt = clone_mnt(old, old_path->dentry, 0);
2598
2599 if (!IS_ERR(mnt))
2600 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2601
2602 return mnt;
2603}
2604
Ram Pai07b20882005-11-07 17:19:07 -05002605/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 * do loopback mount.
2607 */
Al Viro808d4e32012-10-11 11:42:01 -04002608static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002609 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610{
Al Viro2d92ab32008-08-02 00:51:11 -04002611 struct path old_path;
Al Viroa07b2002018-11-05 17:40:30 +00002612 struct mount *mnt = NULL, *parent;
Al Viro84d171922013-03-15 10:53:28 -04002613 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002614 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 if (!old_name || !*old_name)
2616 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002617 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 if (err)
2619 return err;
2620
Eric W. Biederman8823c072010-03-07 18:49:36 -08002621 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002622 if (mnt_ns_loop(old_path.dentry))
David Howellsdd111b32017-07-04 17:25:09 +01002623 goto out;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002624
Al Viro84d171922013-03-15 10:53:28 -04002625 mp = lock_mount(path);
Al Viroa07b2002018-11-05 17:40:30 +00002626 if (IS_ERR(mp)) {
2627 err = PTR_ERR(mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002628 goto out;
Al Viroa07b2002018-11-05 17:40:30 +00002629 }
Ram Pai9676f0c2005-11-07 17:21:20 -05002630
Al Viro84d171922013-03-15 10:53:28 -04002631 parent = real_mount(path->mnt);
Al Viroe149ed22014-11-01 10:57:28 -04002632 if (!check_mnt(parent))
2633 goto out2;
2634
Al Viroa07b2002018-11-05 17:40:30 +00002635 mnt = __do_loopback(&old_path, recurse);
David Howellsbe34d1a2012-06-25 12:55:18 +01002636 if (IS_ERR(mnt)) {
2637 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002638 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002639 }
Al Viroccd48bc2005-11-07 17:15:04 -05002640
Al Viro84d171922013-03-15 10:53:28 -04002641 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002642 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002643 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002644 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002645 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002646 }
Al Virob12cea92011-03-18 08:55:38 -04002647out2:
Al Viro84d171922013-03-15 10:53:28 -04002648 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002649out:
Al Viro2d92ab32008-08-02 00:51:11 -04002650 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 return err;
2652}
2653
Al Viroa07b2002018-11-05 17:40:30 +00002654static struct file *open_detached_copy(struct path *path, bool recursive)
2655{
2656 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2657 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2658 struct mount *mnt, *p;
2659 struct file *file;
2660
2661 if (IS_ERR(ns))
2662 return ERR_CAST(ns);
2663
2664 namespace_lock();
2665 mnt = __do_loopback(path, recursive);
2666 if (IS_ERR(mnt)) {
2667 namespace_unlock();
2668 free_mnt_ns(ns);
2669 return ERR_CAST(mnt);
2670 }
2671
2672 lock_mount_hash();
2673 for (p = mnt; p; p = next_mnt(p, mnt)) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02002674 mnt_add_to_ns(ns, p);
2675 ns->nr_mounts++;
Al Viroa07b2002018-11-05 17:40:30 +00002676 }
2677 ns->root = mnt;
Al Viroa07b2002018-11-05 17:40:30 +00002678 mntget(&mnt->mnt);
2679 unlock_mount_hash();
2680 namespace_unlock();
2681
2682 mntput(path->mnt);
2683 path->mnt = &mnt->mnt;
2684 file = dentry_open(path, O_PATH, current_cred());
2685 if (IS_ERR(file))
2686 dissolve_on_fput(path->mnt);
2687 else
2688 file->f_mode |= FMODE_NEED_UNMOUNT;
2689 return file;
2690}
2691
Ben Dooks2658ce02019-10-15 11:35:02 +01002692SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
Al Viroa07b2002018-11-05 17:40:30 +00002693{
2694 struct file *file;
2695 struct path path;
2696 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2697 bool detached = flags & OPEN_TREE_CLONE;
2698 int error;
2699 int fd;
2700
2701 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2702
2703 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2704 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2705 OPEN_TREE_CLOEXEC))
2706 return -EINVAL;
2707
2708 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2709 return -EINVAL;
2710
2711 if (flags & AT_NO_AUTOMOUNT)
2712 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2713 if (flags & AT_SYMLINK_NOFOLLOW)
2714 lookup_flags &= ~LOOKUP_FOLLOW;
2715 if (flags & AT_EMPTY_PATH)
2716 lookup_flags |= LOOKUP_EMPTY;
2717
2718 if (detached && !may_mount())
2719 return -EPERM;
2720
2721 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2722 if (fd < 0)
2723 return fd;
2724
2725 error = user_path_at(dfd, filename, lookup_flags, &path);
2726 if (unlikely(error)) {
2727 file = ERR_PTR(error);
2728 } else {
2729 if (detached)
2730 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2731 else
2732 file = dentry_open(&path, O_PATH, current_cred());
2733 path_put(&path);
2734 }
2735 if (IS_ERR(file)) {
2736 put_unused_fd(fd);
2737 return PTR_ERR(file);
2738 }
2739 fd_install(fd, file);
2740 return fd;
2741}
2742
David Howells43f5e652018-11-01 23:07:25 +00002743/*
2744 * Don't allow locked mount flags to be cleared.
2745 *
2746 * No locks need to be held here while testing the various MNT_LOCK
2747 * flags because those flags can never be cleared once they are set.
2748 */
2749static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002750{
David Howells43f5e652018-11-01 23:07:25 +00002751 unsigned int fl = mnt->mnt.mnt_flags;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002752
David Howells43f5e652018-11-01 23:07:25 +00002753 if ((fl & MNT_LOCK_READONLY) &&
2754 !(mnt_flags & MNT_READONLY))
2755 return false;
2756
2757 if ((fl & MNT_LOCK_NODEV) &&
2758 !(mnt_flags & MNT_NODEV))
2759 return false;
2760
2761 if ((fl & MNT_LOCK_NOSUID) &&
2762 !(mnt_flags & MNT_NOSUID))
2763 return false;
2764
2765 if ((fl & MNT_LOCK_NOEXEC) &&
2766 !(mnt_flags & MNT_NOEXEC))
2767 return false;
2768
2769 if ((fl & MNT_LOCK_ATIME) &&
2770 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2771 return false;
2772
2773 return true;
2774}
2775
2776static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2777{
2778 bool readonly_request = (mnt_flags & MNT_READONLY);
2779
2780 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002781 return 0;
2782
2783 if (readonly_request)
David Howells43f5e652018-11-01 23:07:25 +00002784 return mnt_make_readonly(mnt);
2785
Christian Brauner68847c92021-01-21 14:19:48 +01002786 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2787 return 0;
David Howells43f5e652018-11-01 23:07:25 +00002788}
2789
David Howells43f5e652018-11-01 23:07:25 +00002790static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2791{
David Howells43f5e652018-11-01 23:07:25 +00002792 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2793 mnt->mnt.mnt_flags = mnt_flags;
2794 touch_mnt_namespace(mnt->mnt_ns);
David Howells43f5e652018-11-01 23:07:25 +00002795}
2796
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002797static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2798{
2799 struct super_block *sb = mnt->mnt_sb;
2800
2801 if (!__mnt_is_readonly(mnt) &&
Anthony Iliopoulosa128b052022-03-22 14:39:22 -07002802 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002803 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2804 char *buf = (char *)__get_free_page(GFP_KERNEL);
2805 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002806
Andy Shevchenko74e60b82023-03-14 17:09:06 +02002807 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
Eric Biggers0ecee662019-10-16 19:48:14 -07002808 sb->s_type->name,
2809 is_mounted(mnt) ? "remounted" : "mounted",
Andy Shevchenko74e60b82023-03-14 17:09:06 +02002810 mntpath, &sb->s_time_max,
2811 (unsigned long long)sb->s_time_max);
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002812
2813 free_page((unsigned long)buf);
Anthony Iliopoulosa128b052022-03-22 14:39:22 -07002814 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002815 }
2816}
2817
David Howells43f5e652018-11-01 23:07:25 +00002818/*
2819 * Handle reconfiguration of the mountpoint only without alteration of the
2820 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2821 * to mount(2).
2822 */
2823static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2824{
2825 struct super_block *sb = path->mnt->mnt_sb;
2826 struct mount *mnt = real_mount(path->mnt);
2827 int ret;
2828
2829 if (!check_mnt(mnt))
2830 return -EINVAL;
2831
Christian Brauner78aa08a2023-05-03 13:18:39 +02002832 if (!path_mounted(path))
David Howells43f5e652018-11-01 23:07:25 +00002833 return -EINVAL;
2834
2835 if (!can_change_locked_flags(mnt, mnt_flags))
2836 return -EPERM;
2837
Christian Braunere58ace12021-01-21 14:19:50 +01002838 /*
2839 * We're only checking whether the superblock is read-only not
2840 * changing it, so only take down_read(&sb->s_umount).
2841 */
2842 down_read(&sb->s_umount);
Christian Brauner68847c92021-01-21 14:19:48 +01002843 lock_mount_hash();
David Howells43f5e652018-11-01 23:07:25 +00002844 ret = change_mount_ro_state(mnt, mnt_flags);
2845 if (ret == 0)
2846 set_mount_attributes(mnt, mnt_flags);
Christian Brauner68847c92021-01-21 14:19:48 +01002847 unlock_mount_hash();
Christian Braunere58ace12021-01-21 14:19:50 +01002848 up_read(&sb->s_umount);
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002849
2850 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2851
David Howells43f5e652018-11-01 23:07:25 +00002852 return ret;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002853}
2854
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855/*
2856 * change filesystem flags. dir should be a physical root of filesystem.
2857 * If you've mounted a non-root directory somewhere and want to do remount
2858 * on it - tough luck.
2859 */
David Howellse462ec52017-07-17 08:45:35 +01002860static int do_remount(struct path *path, int ms_flags, int sb_flags,
2861 int mnt_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862{
2863 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002864 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002865 struct mount *mnt = real_mount(path->mnt);
David Howells8d0347f2018-11-04 09:28:36 -05002866 struct fs_context *fc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
Al Viro143c8c92011-11-25 00:46:35 -05002868 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869 return -EINVAL;
2870
Christian Brauner78aa08a2023-05-03 13:18:39 +02002871 if (!path_mounted(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 return -EINVAL;
2873
David Howells43f5e652018-11-01 23:07:25 +00002874 if (!can_change_locked_flags(mnt, mnt_flags))
Eric W. Biederman07b64552014-07-28 17:10:56 -07002875 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002876
David Howells8d0347f2018-11-04 09:28:36 -05002877 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2878 if (IS_ERR(fc))
2879 return PTR_ERR(fc);
Eric Parisff36fe22011-03-03 16:09:14 -05002880
Christian Braunerf67d9222023-11-22 12:17:37 -05002881 /*
2882 * Indicate to the filesystem that the remount request is coming
2883 * from the legacy mount system call.
2884 */
Miklos Szeredib3309662020-07-14 14:45:41 +02002885 fc->oldapi = true;
Christian Braunerf67d9222023-11-22 12:17:37 -05002886
David Howells8d0347f2018-11-04 09:28:36 -05002887 err = parse_monolithic_mount_data(fc, data);
2888 if (!err) {
2889 down_write(&sb->s_umount);
2890 err = -EPERM;
2891 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2892 err = reconfigure_super(fc);
Christian Brauner68847c92021-01-21 14:19:48 +01002893 if (!err) {
2894 lock_mount_hash();
David Howells8d0347f2018-11-04 09:28:36 -05002895 set_mount_attributes(mnt, mnt_flags);
Christian Brauner68847c92021-01-21 14:19:48 +01002896 unlock_mount_hash();
2897 }
David Howells8d0347f2018-11-04 09:28:36 -05002898 }
2899 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07002900 }
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002901
2902 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2903
David Howells8d0347f2018-11-04 09:28:36 -05002904 put_fs_context(fc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 return err;
2906}
2907
Al Virocbbe3622011-11-24 20:01:19 -05002908static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002909{
Al Viro315fc832011-11-24 18:57:30 -05002910 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002911 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002912 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002913 return 1;
2914 }
2915 return 0;
2916}
2917
David Howells44dfd842018-11-05 17:40:31 +00002918/*
2919 * Check that there aren't references to earlier/same mount namespaces in the
2920 * specified subtree. Such references can act as pins for mount namespaces
2921 * that aren't checked by the mount-cycle checking code, thereby allowing
2922 * cycles to be made.
2923 */
2924static bool check_for_nsfs_mounts(struct mount *subtree)
2925{
2926 struct mount *p;
2927 bool ret = false;
2928
2929 lock_mount_hash();
2930 for (p = subtree; p; p = next_mnt(p, subtree))
2931 if (mnt_ns_loop(p->mnt.mnt_root))
2932 goto out;
2933
2934 ret = true;
2935out:
2936 unlock_mount_hash();
2937 return ret;
2938}
2939
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03002940static int do_set_group(struct path *from_path, struct path *to_path)
2941{
2942 struct mount *from, *to;
2943 int err;
2944
2945 from = real_mount(from_path->mnt);
2946 to = real_mount(to_path->mnt);
2947
2948 namespace_lock();
2949
2950 err = -EINVAL;
2951 /* To and From must be mounted */
2952 if (!is_mounted(&from->mnt))
2953 goto out;
2954 if (!is_mounted(&to->mnt))
2955 goto out;
2956
2957 err = -EPERM;
2958 /* We should be allowed to modify mount namespaces of both mounts */
2959 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2960 goto out;
2961 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2962 goto out;
2963
2964 err = -EINVAL;
2965 /* To and From paths should be mount roots */
Christian Brauner78aa08a2023-05-03 13:18:39 +02002966 if (!path_mounted(from_path))
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03002967 goto out;
Christian Brauner78aa08a2023-05-03 13:18:39 +02002968 if (!path_mounted(to_path))
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03002969 goto out;
2970
2971 /* Setting sharing groups is only allowed across same superblock */
2972 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2973 goto out;
2974
2975 /* From mount root should be wider than To mount root */
2976 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2977 goto out;
2978
2979 /* From mount should not have locked children in place of To's root */
2980 if (has_locked_children(from, to->mnt.mnt_root))
2981 goto out;
2982
2983 /* Setting sharing groups is only allowed on private mounts */
2984 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2985 goto out;
2986
2987 /* From should not be private */
2988 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2989 goto out;
2990
2991 if (IS_MNT_SLAVE(from)) {
2992 struct mount *m = from->mnt_master;
2993
2994 list_add(&to->mnt_slave, &m->mnt_slave_list);
2995 to->mnt_master = m;
2996 }
2997
2998 if (IS_MNT_SHARED(from)) {
2999 to->mnt_group_id = from->mnt_group_id;
3000 list_add(&to->mnt_share, &from->mnt_share);
3001 lock_mount_hash();
3002 set_mnt_shared(to);
3003 unlock_mount_hash();
3004 }
3005
3006 err = 0;
3007out:
3008 namespace_unlock();
3009 return err;
3010}
3011
Christian Brauner6ac39282023-05-03 13:18:42 +02003012/**
3013 * path_overmounted - check if path is overmounted
3014 * @path: path to check
3015 *
3016 * Check if path is overmounted, i.e., if there's a mount on top of
3017 * @path->mnt with @path->dentry as mountpoint.
3018 *
3019 * Context: This function expects namespace_lock() to be held.
3020 * Return: If path is overmounted true is returned, false if not.
3021 */
3022static inline bool path_overmounted(const struct path *path)
3023{
3024 rcu_read_lock();
3025 if (unlikely(__lookup_mnt(path->mnt, path->dentry))) {
3026 rcu_read_unlock();
3027 return true;
3028 }
3029 rcu_read_unlock();
3030 return false;
3031}
3032
3033/**
3034 * can_move_mount_beneath - check that we can mount beneath the top mount
3035 * @from: mount to mount beneath
3036 * @to: mount under which to mount
Christian Braunerd7802b72023-11-21 00:15:55 +01003037 * @mp: mountpoint of @to
Christian Brauner6ac39282023-05-03 13:18:42 +02003038 *
3039 * - Make sure that @to->dentry is actually the root of a mount under
3040 * which we can mount another mount.
3041 * - Make sure that nothing can be mounted beneath the caller's current
3042 * root or the rootfs of the namespace.
3043 * - Make sure that the caller can unmount the topmost mount ensuring
3044 * that the caller could reveal the underlying mountpoint.
3045 * - Ensure that nothing has been mounted on top of @from before we
3046 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3047 * - Prevent mounting beneath a mount if the propagation relationship
3048 * between the source mount, parent mount, and top mount would lead to
3049 * nonsensical mount trees.
3050 *
3051 * Context: This function expects namespace_lock() to be held.
3052 * Return: On success 0, and on error a negative error code is returned.
3053 */
3054static int can_move_mount_beneath(const struct path *from,
3055 const struct path *to,
3056 const struct mountpoint *mp)
3057{
3058 struct mount *mnt_from = real_mount(from->mnt),
3059 *mnt_to = real_mount(to->mnt),
3060 *parent_mnt_to = mnt_to->mnt_parent;
3061
3062 if (!mnt_has_parent(mnt_to))
3063 return -EINVAL;
3064
3065 if (!path_mounted(to))
3066 return -EINVAL;
3067
3068 if (IS_MNT_LOCKED(mnt_to))
3069 return -EINVAL;
3070
3071 /* Avoid creating shadow mounts during mount propagation. */
3072 if (path_overmounted(from))
3073 return -EINVAL;
3074
3075 /*
3076 * Mounting beneath the rootfs only makes sense when the
3077 * semantics of pivot_root(".", ".") are used.
3078 */
3079 if (&mnt_to->mnt == current->fs->root.mnt)
3080 return -EINVAL;
3081 if (parent_mnt_to == current->nsproxy->mnt_ns->root)
3082 return -EINVAL;
3083
3084 for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
3085 if (p == mnt_to)
3086 return -EINVAL;
3087
3088 /*
3089 * If the parent mount propagates to the child mount this would
3090 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3091 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3092 * defeats the whole purpose of mounting beneath another mount.
3093 */
3094 if (propagation_would_overmount(parent_mnt_to, mnt_to, mp))
3095 return -EINVAL;
3096
3097 /*
3098 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3099 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3100 * Afterwards @mnt_from would be mounted on top of
3101 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3102 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3103 * already mounted on @mnt_from, @mnt_to would ultimately be
3104 * remounted on top of @c. Afterwards, @mnt_from would be
3105 * covered by a copy @c of @mnt_from and @c would be covered by
3106 * @mnt_from itself. This defeats the whole purpose of mounting
3107 * @mnt_from beneath @mnt_to.
3108 */
3109 if (propagation_would_overmount(parent_mnt_to, mnt_from, mp))
3110 return -EINVAL;
3111
3112 return 0;
3113}
3114
3115static int do_move_mount(struct path *old_path, struct path *new_path,
3116 bool beneath)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117{
David Howells44dfd842018-11-05 17:40:31 +00003118 struct mnt_namespace *ns;
Al Viro676da582011-11-24 21:47:05 -05003119 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05003120 struct mount *old;
Al Viro2763d112019-06-30 19:18:53 -04003121 struct mount *parent;
3122 struct mountpoint *mp, *old_mp;
Al Viro57eccb82013-02-22 22:49:10 -05003123 int err;
David Howells44dfd842018-11-05 17:40:31 +00003124 bool attached;
Christian Brauner6ac39282023-05-03 13:18:42 +02003125 enum mnt_tree_flags_t flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126
Christian Brauner6ac39282023-05-03 13:18:42 +02003127 mp = do_lock_mount(new_path, beneath);
Al Viro84d171922013-03-15 10:53:28 -04003128 if (IS_ERR(mp))
David Howells2db154b2018-11-05 17:40:30 +00003129 return PTR_ERR(mp);
David Howellscc53ce52011-01-14 18:45:26 +00003130
David Howells2db154b2018-11-05 17:40:30 +00003131 old = real_mount(old_path->mnt);
3132 p = real_mount(new_path->mnt);
Al Viro2763d112019-06-30 19:18:53 -04003133 parent = old->mnt_parent;
David Howells44dfd842018-11-05 17:40:31 +00003134 attached = mnt_has_parent(old);
Christian Brauner6ac39282023-05-03 13:18:42 +02003135 if (attached)
3136 flags |= MNT_TREE_MOVE;
Al Viro2763d112019-06-30 19:18:53 -04003137 old_mp = old->mnt_mp;
David Howells44dfd842018-11-05 17:40:31 +00003138 ns = old->mnt_ns;
Al Viro143c8c92011-11-25 00:46:35 -05003139
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 err = -EINVAL;
David Howells44dfd842018-11-05 17:40:31 +00003141 /* The mountpoint must be in our namespace. */
3142 if (!check_mnt(p))
David Howells2db154b2018-11-05 17:40:30 +00003143 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144
Eric Biggers570d7a92019-06-29 13:27:44 -07003145 /* The thing moved must be mounted... */
3146 if (!is_mounted(&old->mnt))
David Howells44dfd842018-11-05 17:40:31 +00003147 goto out;
3148
Eric Biggers570d7a92019-06-29 13:27:44 -07003149 /* ... and either ours or the root of anon namespace */
3150 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
David Howells2db154b2018-11-05 17:40:30 +00003151 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152
David Howells2db154b2018-11-05 17:40:30 +00003153 if (old->mnt.mnt_flags & MNT_LOCKED)
3154 goto out;
3155
Christian Brauner78aa08a2023-05-03 13:18:39 +02003156 if (!path_mounted(old_path))
David Howells2db154b2018-11-05 17:40:30 +00003157 goto out;
3158
3159 if (d_is_dir(new_path->dentry) !=
3160 d_is_dir(old_path->dentry))
3161 goto out;
Ram Pai21444402005-11-07 17:20:03 -05003162 /*
3163 * Don't move a mount residing in a shared parent.
3164 */
Al Viro2763d112019-06-30 19:18:53 -04003165 if (attached && IS_MNT_SHARED(parent))
David Howells2db154b2018-11-05 17:40:30 +00003166 goto out;
Christian Brauner6ac39282023-05-03 13:18:42 +02003167
3168 if (beneath) {
3169 err = can_move_mount_beneath(old_path, new_path, mp);
3170 if (err)
3171 goto out;
3172
3173 err = -EINVAL;
3174 p = p->mnt_parent;
3175 flags |= MNT_TREE_BENEATH;
3176 }
3177
Ram Pai9676f0c2005-11-07 17:21:20 -05003178 /*
3179 * Don't move a mount tree containing unbindable mounts to a destination
3180 * mount which is shared.
3181 */
Al Virofc7be132011-11-25 01:05:37 -05003182 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
David Howells2db154b2018-11-05 17:40:30 +00003183 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 err = -ELOOP;
David Howells44dfd842018-11-05 17:40:31 +00003185 if (!check_for_nsfs_mounts(old))
3186 goto out;
Al Virofc7be132011-11-25 01:05:37 -05003187 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05003188 if (p == old)
David Howells2db154b2018-11-05 17:40:30 +00003189 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190
Christian Brauner6ac39282023-05-03 13:18:42 +02003191 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
Jan Blunck4ac91372008-02-14 19:34:32 -08003192 if (err)
David Howells2db154b2018-11-05 17:40:30 +00003193 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194
3195 /* if the mount is moved, it should no longer be expire
3196 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05003197 list_del_init(&old->mnt_expire);
Al Viro2763d112019-06-30 19:18:53 -04003198 if (attached)
3199 put_mountpoint(old_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200out:
David Howells2db154b2018-11-05 17:40:30 +00003201 unlock_mount(mp);
David Howells44dfd842018-11-05 17:40:31 +00003202 if (!err) {
Al Viro2763d112019-06-30 19:18:53 -04003203 if (attached)
3204 mntput_no_expire(parent);
3205 else
David Howells44dfd842018-11-05 17:40:31 +00003206 free_mnt_ns(ns);
3207 }
David Howells2db154b2018-11-05 17:40:30 +00003208 return err;
3209}
3210
3211static int do_move_mount_old(struct path *path, const char *old_name)
3212{
3213 struct path old_path;
3214 int err;
3215
3216 if (!old_name || !*old_name)
3217 return -EINVAL;
3218
3219 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3220 if (err)
3221 return err;
3222
Christian Brauner6ac39282023-05-03 13:18:42 +02003223 err = do_move_mount(&old_path, path, false);
Al Viro2d92ab32008-08-02 00:51:11 -04003224 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 return err;
3226}
3227
Al Viro9d412a42011-03-17 22:08:28 -04003228/*
3229 * add a mount into a namespace's mount tree
3230 */
Al Viro8f115382020-01-11 10:14:09 -05003231static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
Al Viro1e2d84642021-06-18 20:27:57 -04003232 const struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04003233{
Al Viro8f115382020-01-11 10:14:09 -05003234 struct mount *parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04003235
Al Virof2ebb3a2014-02-27 09:35:45 -05003236 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04003237
Al Viro84d171922013-03-15 10:53:28 -04003238 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04003239 /* that's acceptable only for automounts done in private ns */
3240 if (!(mnt_flags & MNT_SHRINKABLE))
Al Viro8f115382020-01-11 10:14:09 -05003241 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04003242 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d171922013-03-15 10:53:28 -04003243 if (!parent->mnt_ns)
Al Viro8f115382020-01-11 10:14:09 -05003244 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04003245 }
Al Viro9d412a42011-03-17 22:08:28 -04003246
3247 /* Refuse the same filesystem on the same mount point */
Christian Brauner78aa08a2023-05-03 13:18:39 +02003248 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
Al Viro8f115382020-01-11 10:14:09 -05003249 return -EBUSY;
Al Viro9d412a42011-03-17 22:08:28 -04003250
David Howellse36cb0b2015-01-29 12:02:35 +00003251 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro8f115382020-01-11 10:14:09 -05003252 return -EINVAL;
Al Viro9d412a42011-03-17 22:08:28 -04003253
Al Viro95bc5f22011-11-25 00:30:56 -05003254 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro8f115382020-01-11 10:14:09 -05003255 return graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04003256}
Al Virob1e75df2011-01-17 01:47:59 -05003257
David Howells132e4602018-11-04 07:43:08 -05003258static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3259
3260/*
3261 * Create a new mount using a superblock configuration and request it
3262 * be added to the namespace tree.
3263 */
3264static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3265 unsigned int mnt_flags)
3266{
3267 struct vfsmount *mnt;
Al Viro8f115382020-01-11 10:14:09 -05003268 struct mountpoint *mp;
David Howells132e4602018-11-04 07:43:08 -05003269 struct super_block *sb = fc->root->d_sb;
3270 int error;
3271
Al Viroc9ce29e2018-12-20 15:04:50 -05003272 error = security_sb_kern_mount(sb);
3273 if (!error && mount_too_revealing(sb, &mnt_flags))
3274 error = -EPERM;
3275
3276 if (unlikely(error)) {
3277 fc_drop_locked(fc);
3278 return error;
David Howells132e4602018-11-04 07:43:08 -05003279 }
3280
3281 up_write(&sb->s_umount);
3282
3283 mnt = vfs_create_mount(fc);
3284 if (IS_ERR(mnt))
3285 return PTR_ERR(mnt);
3286
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07003287 mnt_warn_timestamp_expiry(mountpoint, mnt);
3288
Al Viro8f115382020-01-11 10:14:09 -05003289 mp = lock_mount(mountpoint);
3290 if (IS_ERR(mp)) {
3291 mntput(mnt);
3292 return PTR_ERR(mp);
3293 }
3294 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3295 unlock_mount(mp);
Eric Biggers0ecee662019-10-16 19:48:14 -07003296 if (error < 0)
3297 mntput(mnt);
David Howells132e4602018-11-04 07:43:08 -05003298 return error;
3299}
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05003300
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301/*
3302 * create a new mount for userspace and request it to be added into the
3303 * namespace's tree
3304 */
David Howellse462ec52017-07-17 08:45:35 +01003305static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
Al Viro808d4e32012-10-11 11:42:01 -04003306 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003308 struct file_system_type *type;
Al Viroa0c9a8b2018-11-04 07:18:51 -05003309 struct fs_context *fc;
3310 const char *subtype = NULL;
3311 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003313 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314 return -EINVAL;
3315
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003316 type = get_fs_type(fstype);
3317 if (!type)
3318 return -ENODEV;
3319
Al Viroa0c9a8b2018-11-04 07:18:51 -05003320 if (type->fs_flags & FS_HAS_SUBTYPE) {
3321 subtype = strchr(fstype, '.');
3322 if (subtype) {
3323 subtype++;
3324 if (!*subtype) {
3325 put_filesystem(type);
3326 return -EINVAL;
3327 }
Al Viroa0c9a8b2018-11-04 07:18:51 -05003328 }
Eric W. Biederman8654df42016-06-09 16:06:06 -05003329 }
3330
Al Viroa0c9a8b2018-11-04 07:18:51 -05003331 fc = fs_context_for_mount(type, sb_flags);
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003332 put_filesystem(type);
Al Viroa0c9a8b2018-11-04 07:18:51 -05003333 if (IS_ERR(fc))
3334 return PTR_ERR(fc);
3335
Christian Braunerf67d9222023-11-22 12:17:37 -05003336 /*
3337 * Indicate to the filesystem that the mount request is coming
3338 * from the legacy mount system call.
3339 */
3340 fc->oldapi = true;
3341
David Howells3e1aeb02018-11-01 23:07:25 +00003342 if (subtype)
3343 err = vfs_parse_fs_string(fc, "subtype",
3344 subtype, strlen(subtype));
3345 if (!err && name)
3346 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
Al Viroa0c9a8b2018-11-04 07:18:51 -05003347 if (!err)
3348 err = parse_monolithic_mount_data(fc, data);
Al Viroc3aabf02019-05-13 12:57:22 -04003349 if (!err && !mount_capable(fc))
3350 err = -EPERM;
Al Viroa0c9a8b2018-11-04 07:18:51 -05003351 if (!err)
3352 err = vfs_get_tree(fc);
David Howells132e4602018-11-04 07:43:08 -05003353 if (!err)
3354 err = do_new_mount_fc(fc, path, mnt_flags);
Al Viroa0c9a8b2018-11-04 07:18:51 -05003355
Al Viroa0c9a8b2018-11-04 07:18:51 -05003356 put_fs_context(fc);
Al Viro15f9a3f2011-01-17 01:41:58 -05003357 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358}
3359
Al Viro1e2d84642021-06-18 20:27:57 -04003360int finish_automount(struct vfsmount *m, const struct path *path)
Al Viro19a167a2011-01-17 01:35:23 -05003361{
Al Viro26df6032020-01-11 10:44:29 -05003362 struct dentry *dentry = path->dentry;
Al Viro8f115382020-01-11 10:14:09 -05003363 struct mountpoint *mp;
Al Viro25e195a2020-01-11 11:27:46 -05003364 struct mount *mnt;
Al Viro19a167a2011-01-17 01:35:23 -05003365 int err;
Al Viro25e195a2020-01-11 11:27:46 -05003366
3367 if (!m)
3368 return 0;
3369 if (IS_ERR(m))
3370 return PTR_ERR(m);
3371
3372 mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05003373 /* The new mount record should have at least 2 refs to prevent it being
3374 * expired before we get a chance to add it
3375 */
Al Viro6776db3d2011-11-25 00:22:05 -05003376 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05003377
3378 if (m->mnt_sb == path->mnt->mnt_sb &&
Al Viro26df6032020-01-11 10:44:29 -05003379 m->mnt_root == dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05003380 err = -ELOOP;
Al Viro26df6032020-01-11 10:44:29 -05003381 goto discard;
Al Viro19a167a2011-01-17 01:35:23 -05003382 }
3383
Al Viro26df6032020-01-11 10:44:29 -05003384 /*
3385 * we don't want to use lock_mount() - in this case finding something
3386 * that overmounts our mountpoint to be means "quitely drop what we've
3387 * got", not "try to mount it on top".
3388 */
3389 inode_lock(dentry->d_inode);
3390 namespace_lock();
3391 if (unlikely(cant_mount(dentry))) {
3392 err = -ENOENT;
3393 goto discard_locked;
3394 }
Christian Brauner6ac39282023-05-03 13:18:42 +02003395 if (path_overmounted(path)) {
Al Viro26df6032020-01-11 10:44:29 -05003396 err = 0;
3397 goto discard_locked;
3398 }
Al Viro26df6032020-01-11 10:44:29 -05003399 mp = get_mountpoint(dentry);
Al Viro8f115382020-01-11 10:14:09 -05003400 if (IS_ERR(mp)) {
3401 err = PTR_ERR(mp);
Al Viro26df6032020-01-11 10:44:29 -05003402 goto discard_locked;
Al Viro8f115382020-01-11 10:14:09 -05003403 }
Al Viro26df6032020-01-11 10:44:29 -05003404
Al Viro8f115382020-01-11 10:14:09 -05003405 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3406 unlock_mount(mp);
Al Viro26df6032020-01-11 10:44:29 -05003407 if (unlikely(err))
3408 goto discard;
3409 mntput(m);
3410 return 0;
3411
3412discard_locked:
3413 namespace_unlock();
3414 inode_unlock(dentry->d_inode);
3415discard:
Al Virob1e75df2011-01-17 01:47:59 -05003416 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05003417 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04003418 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05003419 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04003420 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05003421 }
Al Virob1e75df2011-01-17 01:47:59 -05003422 mntput(m);
3423 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05003424 return err;
3425}
3426
David Howellsea5b7782011-01-14 19:10:03 +00003427/**
3428 * mnt_set_expiry - Put a mount on an expiration list
3429 * @mnt: The mount to list.
3430 * @expiry_list: The list to add the mount to.
3431 */
3432void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3433{
Al Viro97216be2013-03-16 15:12:40 -04003434 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00003435
Al Viro6776db3d2011-11-25 00:22:05 -05003436 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00003437
Al Viro97216be2013-03-16 15:12:40 -04003438 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00003439}
3440EXPORT_SYMBOL(mnt_set_expiry);
3441
3442/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 * process a list of expirable mountpoints with the intent of discarding any
3444 * mountpoints that aren't in use and haven't been touched since last we came
3445 * here
3446 */
3447void mark_mounts_for_expiry(struct list_head *mounts)
3448{
Al Viro761d5c32011-11-24 21:07:43 -05003449 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 LIST_HEAD(graveyard);
3451
3452 if (list_empty(mounts))
3453 return;
3454
Al Viro97216be2013-03-16 15:12:40 -04003455 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04003456 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457
3458 /* extract from the expiration list every vfsmount that matches the
3459 * following criteria:
3460 * - only referenced by its parent vfsmount
3461 * - still marked for expiry (marked on the last call here; marks are
3462 * cleared by mntput())
3463 */
Al Viro6776db3d2011-11-25 00:22:05 -05003464 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05003465 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05003466 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05003468 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 }
Al Virobcc5c7d2008-03-22 00:21:53 -04003470 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003471 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003472 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003473 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003474 }
Al Viro719ea2f2013-09-29 11:24:49 -04003475 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04003476 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477}
3478
3479EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3480
3481/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04003482 * Ripoff of 'select_parent()'
3483 *
3484 * search the list of submounts for a given mountpoint, and move any
3485 * shrinkable submounts to the 'graveyard' list.
3486 */
Al Viro692afc32011-11-24 21:15:14 -05003487static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003488{
Al Viro692afc32011-11-24 21:15:14 -05003489 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003490 struct list_head *next;
3491 int found = 0;
3492
3493repeat:
Al Viro6b41d532011-11-24 23:24:33 -05003494 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003495resume:
Al Viro6b41d532011-11-24 23:24:33 -05003496 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003497 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05003498 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003499
3500 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05003501 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04003502 continue;
3503 /*
3504 * Descend a level if the d_mounts list is non-empty.
3505 */
Al Viro6b41d532011-11-24 23:24:33 -05003506 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003507 this_parent = mnt;
3508 goto repeat;
3509 }
3510
Al Viro1ab59732011-11-24 21:35:16 -05003511 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003512 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003513 found++;
3514 }
3515 }
3516 /*
3517 * All done at this level ... ascend and resume the search
3518 */
3519 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05003520 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05003521 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003522 goto resume;
3523 }
3524 return found;
3525}
3526
3527/*
3528 * process a list of expirable mountpoints with the intent of discarding any
3529 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10003530 *
Al Viro48a066e2013-09-29 22:06:07 -04003531 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04003532 */
Al Virob54b9be2013-03-16 14:39:34 -04003533static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003534{
3535 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05003536 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003537
Trond Myklebust5528f9112006-06-09 09:34:17 -04003538 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04003539 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04003540 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05003541 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05003542 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003543 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003544 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003545 }
3546 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04003547}
3548
Christoph Hellwig028abd92020-09-17 10:22:34 +02003549static void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550{
Al Virob40ef862015-12-14 18:44:44 -05003551 char *copy;
Catalin Marinasd563d672020-07-01 17:46:06 +01003552 unsigned left, offset;
Ram Paib58fed82005-11-07 17:16:09 -05003553
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05003555 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556
Al Virob40ef862015-12-14 18:44:44 -05003557 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3558 if (!copy)
3559 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Catalin Marinasd563d672020-07-01 17:46:06 +01003561 left = copy_from_user(copy, data, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562
Catalin Marinasd563d672020-07-01 17:46:06 +01003563 /*
3564 * Not all architectures have an exact copy_from_user(). Resort to
3565 * byte at a time.
3566 */
3567 offset = PAGE_SIZE - left;
3568 while (left) {
3569 char c;
3570 if (get_user(c, (const char __user *)data + offset))
3571 break;
3572 copy[offset] = c;
3573 left--;
3574 offset++;
3575 }
3576
3577 if (left == PAGE_SIZE) {
Al Virob40ef862015-12-14 18:44:44 -05003578 kfree(copy);
3579 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 }
Catalin Marinasd563d672020-07-01 17:46:06 +01003581
Al Virob40ef862015-12-14 18:44:44 -05003582 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583}
3584
Christoph Hellwig028abd92020-09-17 10:22:34 +02003585static char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07003586{
Chandan Rajendrafbdb4402019-01-22 12:21:52 +05303587 return data ? strndup_user(data, PATH_MAX) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003588}
3589
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590/*
3591 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3592 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3593 *
3594 * data is a (void *) that can point to any structure up to
3595 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3596 * information (or be NULL).
3597 *
3598 * Pre-0.97 versions of mount() didn't have a flags word.
3599 * When the flags word was introduced its top half was required
3600 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3601 * Therefore, if this magic number is present, it carries no information
3602 * and must be discarded.
3603 */
Christoph Hellwigc60166f2020-07-21 11:12:08 +02003604int path_mount(const char *dev_name, struct path *path,
Al Viro808d4e32012-10-11 11:42:01 -04003605 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606{
David Howellse462ec52017-07-17 08:45:35 +01003607 unsigned int mnt_flags = 0, sb_flags;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003608 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609
3610 /* Discard magic */
3611 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3612 flags &= ~MS_MGC_MSK;
3613
3614 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 if (data_page)
3616 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3617
David Howellse462ec52017-07-17 08:45:35 +01003618 if (flags & MS_NOUSER)
3619 return -EINVAL;
3620
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003621 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3622 if (ret)
3623 return ret;
3624 if (!may_mount())
3625 return -EPERM;
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04003626 if (flags & SB_MANDLOCK)
3627 warn_mandlock();
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09003628
Andi Kleen613cbe32009-04-19 18:40:43 +02003629 /* Default to relatime unless overriden */
3630 if (!(flags & MS_NOATIME))
3631 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00003632
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 /* Separate the per-mountpoint flags */
3634 if (flags & MS_NOSUID)
3635 mnt_flags |= MNT_NOSUID;
3636 if (flags & MS_NODEV)
3637 mnt_flags |= MNT_NODEV;
3638 if (flags & MS_NOEXEC)
3639 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003640 if (flags & MS_NOATIME)
3641 mnt_flags |= MNT_NOATIME;
3642 if (flags & MS_NODIRATIME)
3643 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00003644 if (flags & MS_STRICTATIME)
3645 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
David Howellsa9e5b732018-04-20 13:35:02 +01003646 if (flags & MS_RDONLY)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08003647 mnt_flags |= MNT_READONLY;
Mattias Nisslerdab741e02020-08-27 11:09:46 -06003648 if (flags & MS_NOSYMFOLLOW)
3649 mnt_flags |= MNT_NOSYMFOLLOW;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003650
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003651 /* The default atime for remount is preservation */
3652 if ((flags & MS_REMOUNT) &&
3653 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3654 MS_STRICTATIME)) == 0)) {
3655 mnt_flags &= ~MNT_ATIME_MASK;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003656 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003657 }
3658
David Howellse462ec52017-07-17 08:45:35 +01003659 sb_flags = flags & (SB_RDONLY |
3660 SB_SYNCHRONOUS |
3661 SB_MANDLOCK |
3662 SB_DIRSYNC |
3663 SB_SILENT |
Mimi Zohar917086f2017-10-08 00:28:21 -04003664 SB_POSIXACL |
Markus Trippelsdorfd7ee9462017-10-11 07:01:31 +02003665 SB_LAZYTIME |
Mimi Zohar917086f2017-10-08 00:28:21 -04003666 SB_I_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
David Howells43f5e652018-11-01 23:07:25 +00003668 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003669 return do_reconfigure_mnt(path, mnt_flags);
3670 if (flags & MS_REMOUNT)
3671 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3672 if (flags & MS_BIND)
3673 return do_loopback(path, dev_name, flags & MS_REC);
3674 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3675 return do_change_type(path, flags);
3676 if (flags & MS_MOVE)
3677 return do_move_mount_old(path, dev_name);
3678
3679 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3680 data_page);
3681}
3682
3683long do_mount(const char *dev_name, const char __user *dir_name,
3684 const char *type_page, unsigned long flags, void *data_page)
3685{
3686 struct path path;
3687 int ret;
3688
3689 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3690 if (ret)
3691 return ret;
3692 ret = path_mount(dev_name, &path, type_page, flags, data_page);
Al Viro2d92ab32008-08-02 00:51:11 -04003693 path_put(&path);
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003694 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695}
3696
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003697static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3698{
3699 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3700}
3701
3702static void dec_mnt_namespaces(struct ucounts *ucounts)
3703{
3704 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3705}
3706
Eric W. Biederman771b1372012-07-26 21:08:32 -07003707static void free_mnt_ns(struct mnt_namespace *ns)
3708{
Al Viro74e83122019-01-30 13:30:21 -05003709 if (!is_anon_ns(ns))
3710 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003711 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003712 put_user_ns(ns->user_ns);
3713 kfree(ns);
3714}
3715
Eric W. Biederman8823c072010-03-07 18:49:36 -08003716/*
3717 * Assign a sequence number so we can detect when we attempt to bind
3718 * mount a reference to an older mount namespace into the current
3719 * mount namespace, preventing reference counting loops. A 64bit
3720 * number incrementing at 10Ghz will take 12,427 years to wrap which
3721 * is effectively never, so we can ignore the possibility.
3722 */
3723static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3724
Al Viro74e83122019-01-30 13:30:21 -05003725static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003726{
3727 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003728 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003729 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003730
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003731 ucounts = inc_mnt_namespaces(user_ns);
3732 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05003733 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003734
Vasily Averin30acd0b2021-09-02 14:55:27 -07003735 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003736 if (!new_ns) {
3737 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003738 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003739 }
Al Viro74e83122019-01-30 13:30:21 -05003740 if (!anon) {
3741 ret = ns_alloc_inum(&new_ns->ns);
3742 if (ret) {
3743 kfree(new_ns);
3744 dec_mnt_namespaces(ucounts);
3745 return ERR_PTR(ret);
3746 }
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003747 }
Al Viro33c42942014-11-01 02:32:53 -04003748 new_ns->ns.ops = &mntns_operations;
Al Viro74e83122019-01-30 13:30:21 -05003749 if (!anon)
3750 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03003751 refcount_set(&new_ns->ns.count, 1);
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02003752 new_ns->mounts = RB_ROOT;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003753 init_waitqueue_head(&new_ns->poll);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003754 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003755 new_ns->ucounts = ucounts;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003756 return new_ns;
3757}
3758
Emese Revfy0766f782016-06-20 20:42:34 +02003759__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04003760struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3761 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003763 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04003764 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05003765 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04003766 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05003767 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003768 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769
Al Viro9559f682013-09-28 20:47:57 -04003770 BUG_ON(!ns);
3771
3772 if (likely(!(flags & CLONE_NEWNS))) {
3773 get_mnt_ns(ns);
3774 return ns;
3775 }
3776
3777 old = ns->root;
3778
Al Viro74e83122019-01-30 13:30:21 -05003779 new_ns = alloc_mnt_ns(user_ns, false);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003780 if (IS_ERR(new_ns))
3781 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003782
Al Viro97216be2013-03-16 15:12:40 -04003783 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003785 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04003786 if (user_ns != ns->user_ns)
Al Viro3bd045c2019-01-30 13:15:45 -05003787 copy_flags |= CL_SHARED_TO_SLAVE;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003788 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01003789 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04003790 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07003791 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01003792 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793 }
Al Viro3bd045c2019-01-30 13:15:45 -05003794 if (user_ns != ns->user_ns) {
3795 lock_mount_hash();
3796 lock_mnt_tree(new);
3797 unlock_mount_hash();
3798 }
Al Virobe08d6d2011-12-06 13:32:36 -05003799 new_ns->root = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800
3801 /*
3802 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3803 * as belonging to new namespace. We have already acquired a private
3804 * fs_struct, so tsk->fs->lock is not needed.
3805 */
Al Viro909b0a82011-11-25 03:06:56 -05003806 p = old;
Al Virocb338d02011-11-24 20:55:08 -05003807 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 while (p) {
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02003809 mnt_add_to_ns(new_ns, q);
3810 new_ns->nr_mounts++;
Al Viro9559f682013-09-28 20:47:57 -04003811 if (new_fs) {
3812 if (&p->mnt == new_fs->root.mnt) {
3813 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003814 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815 }
Al Viro9559f682013-09-28 20:47:57 -04003816 if (&p->mnt == new_fs->pwd.mnt) {
3817 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003818 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 }
Al Viro909b0a82011-11-25 03:06:56 -05003821 p = next_mnt(p, old);
3822 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003823 if (!q)
3824 break;
Al Viro61d8e422022-11-24 22:55:57 -05003825 // an mntns binding we'd skipped?
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003826 while (p->mnt.mnt_root != q->mnt.mnt_root)
Al Viro61d8e422022-11-24 22:55:57 -05003827 p = next_mnt(skip_mnt_tree(p), old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 }
Al Viro328e6d92013-03-16 14:49:45 -04003829 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05003832 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05003834 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835
JANAK DESAI741a2952006-02-07 12:59:00 -08003836 return new_ns;
3837}
3838
Al Viro74e83122019-01-30 13:30:21 -05003839struct dentry *mount_subtree(struct vfsmount *m, const char *name)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003840{
Al Viro74e83122019-01-30 13:30:21 -05003841 struct mount *mnt = real_mount(m);
Al Viroea441d12011-11-16 21:43:59 -05003842 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05003843 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05003844 struct path path;
3845 int err;
3846
Al Viro74e83122019-01-30 13:30:21 -05003847 ns = alloc_mnt_ns(&init_user_ns, true);
3848 if (IS_ERR(ns)) {
3849 mntput(m);
Al Viroea441d12011-11-16 21:43:59 -05003850 return ERR_CAST(ns);
Al Viro74e83122019-01-30 13:30:21 -05003851 }
Al Viro74e83122019-01-30 13:30:21 -05003852 ns->root = mnt;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02003853 ns->nr_mounts++;
3854 mnt_add_to_ns(ns, mnt);
Al Viroea441d12011-11-16 21:43:59 -05003855
Al Viro74e83122019-01-30 13:30:21 -05003856 err = vfs_path_lookup(m->mnt_root, m,
Al Viroea441d12011-11-16 21:43:59 -05003857 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3858
3859 put_mnt_ns(ns);
3860
3861 if (err)
3862 return ERR_PTR(err);
3863
3864 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05003865 s = path.mnt->mnt_sb;
3866 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05003867 mntput(path.mnt);
3868 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05003869 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05003870 /* ... and return the root of (sub)tree on it */
3871 return path.dentry;
3872}
3873EXPORT_SYMBOL(mount_subtree);
3874
Dominik Brodowskicccaa5e2018-10-23 22:41:09 +02003875SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3876 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003877{
Vegard Nossumeca6f532009-09-18 13:05:45 -07003878 int ret;
3879 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003880 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05003881 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882
Tim Gardnerb8850d12014-08-28 11:26:03 -06003883 kernel_type = copy_mount_string(type);
3884 ret = PTR_ERR(kernel_type);
3885 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003886 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887
Tim Gardnerb8850d12014-08-28 11:26:03 -06003888 kernel_dev = copy_mount_string(dev_name);
3889 ret = PTR_ERR(kernel_dev);
3890 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003891 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892
Al Virob40ef862015-12-14 18:44:44 -05003893 options = copy_mount_options(data);
3894 ret = PTR_ERR(options);
3895 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003896 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897
Al Virob40ef862015-12-14 18:44:44 -05003898 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003899
Al Virob40ef862015-12-14 18:44:44 -05003900 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003901out_data:
3902 kfree(kernel_dev);
3903out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07003904 kfree(kernel_type);
3905out_type:
3906 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907}
3908
Christian Braunerdd8b4772021-06-01 11:33:59 +02003909#define FSMOUNT_VALID_FLAGS \
3910 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3911 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3912 MOUNT_ATTR_NOSYMFOLLOW)
Christian Brauner5b490502021-01-21 14:19:52 +01003913
Christian Brauner9caccd42021-01-21 14:19:54 +01003914#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
Christian Brauner2a186722021-01-21 14:19:53 +01003915
3916#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3917 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3918
Christian Brauner5b490502021-01-21 14:19:52 +01003919static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3920{
3921 unsigned int mnt_flags = 0;
3922
3923 if (attr_flags & MOUNT_ATTR_RDONLY)
3924 mnt_flags |= MNT_READONLY;
3925 if (attr_flags & MOUNT_ATTR_NOSUID)
3926 mnt_flags |= MNT_NOSUID;
3927 if (attr_flags & MOUNT_ATTR_NODEV)
3928 mnt_flags |= MNT_NODEV;
3929 if (attr_flags & MOUNT_ATTR_NOEXEC)
3930 mnt_flags |= MNT_NOEXEC;
3931 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3932 mnt_flags |= MNT_NODIRATIME;
Christian Braunerdd8b4772021-06-01 11:33:59 +02003933 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3934 mnt_flags |= MNT_NOSYMFOLLOW;
Christian Brauner5b490502021-01-21 14:19:52 +01003935
3936 return mnt_flags;
3937}
3938
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939/*
David Howells93766fb2018-11-01 23:36:14 +00003940 * Create a kernel mount representation for a new, prepared superblock
3941 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3942 */
3943SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3944 unsigned int, attr_flags)
3945{
3946 struct mnt_namespace *ns;
3947 struct fs_context *fc;
3948 struct file *file;
3949 struct path newmount;
3950 struct mount *mnt;
3951 struct fd f;
3952 unsigned int mnt_flags = 0;
3953 long ret;
3954
3955 if (!may_mount())
3956 return -EPERM;
3957
3958 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3959 return -EINVAL;
3960
Christian Brauner5b490502021-01-21 14:19:52 +01003961 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
David Howells93766fb2018-11-01 23:36:14 +00003962 return -EINVAL;
3963
Christian Brauner5b490502021-01-21 14:19:52 +01003964 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
David Howells93766fb2018-11-01 23:36:14 +00003965
3966 switch (attr_flags & MOUNT_ATTR__ATIME) {
3967 case MOUNT_ATTR_STRICTATIME:
3968 break;
3969 case MOUNT_ATTR_NOATIME:
3970 mnt_flags |= MNT_NOATIME;
3971 break;
3972 case MOUNT_ATTR_RELATIME:
3973 mnt_flags |= MNT_RELATIME;
3974 break;
3975 default:
3976 return -EINVAL;
3977 }
3978
3979 f = fdget(fs_fd);
3980 if (!f.file)
3981 return -EBADF;
3982
3983 ret = -EINVAL;
3984 if (f.file->f_op != &fscontext_fops)
3985 goto err_fsfd;
3986
3987 fc = f.file->private_data;
3988
3989 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3990 if (ret < 0)
3991 goto err_fsfd;
3992
3993 /* There must be a valid superblock or we can't mount it */
3994 ret = -EINVAL;
3995 if (!fc->root)
3996 goto err_unlock;
3997
3998 ret = -EPERM;
3999 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
4000 pr_warn("VFS: Mount too revealing\n");
4001 goto err_unlock;
4002 }
4003
4004 ret = -EBUSY;
4005 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
4006 goto err_unlock;
4007
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04004008 if (fc->sb_flags & SB_MANDLOCK)
4009 warn_mandlock();
David Howells93766fb2018-11-01 23:36:14 +00004010
4011 newmount.mnt = vfs_create_mount(fc);
4012 if (IS_ERR(newmount.mnt)) {
4013 ret = PTR_ERR(newmount.mnt);
4014 goto err_unlock;
4015 }
4016 newmount.dentry = dget(fc->root);
4017 newmount.mnt->mnt_flags = mnt_flags;
4018
4019 /* We've done the mount bit - now move the file context into more or
4020 * less the same state as if we'd done an fspick(). We don't want to
4021 * do any memory allocation or anything like that at this point as we
4022 * don't want to have to handle any errors incurred.
4023 */
4024 vfs_clean_context(fc);
4025
4026 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
4027 if (IS_ERR(ns)) {
4028 ret = PTR_ERR(ns);
4029 goto err_path;
4030 }
4031 mnt = real_mount(newmount.mnt);
David Howells93766fb2018-11-01 23:36:14 +00004032 ns->root = mnt;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02004033 ns->nr_mounts = 1;
4034 mnt_add_to_ns(ns, mnt);
Eric Biggers1b0b9cc2019-06-12 11:43:13 -07004035 mntget(newmount.mnt);
David Howells93766fb2018-11-01 23:36:14 +00004036
4037 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4038 * it, not just simply put it.
4039 */
4040 file = dentry_open(&newmount, O_PATH, fc->cred);
4041 if (IS_ERR(file)) {
4042 dissolve_on_fput(newmount.mnt);
4043 ret = PTR_ERR(file);
4044 goto err_path;
4045 }
4046 file->f_mode |= FMODE_NEED_UNMOUNT;
4047
4048 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
4049 if (ret >= 0)
4050 fd_install(ret, file);
4051 else
4052 fput(file);
4053
4054err_path:
4055 path_put(&newmount);
4056err_unlock:
4057 mutex_unlock(&fc->uapi_mutex);
4058err_fsfd:
4059 fdput(f);
4060 return ret;
4061}
4062
4063/*
4064 * Move a mount from one place to another. In combination with
4065 * fsopen()/fsmount() this is used to install a new mount and in combination
4066 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4067 * a mount subtree.
David Howells2db154b2018-11-05 17:40:30 +00004068 *
4069 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4070 */
4071SYSCALL_DEFINE5(move_mount,
Ben Dooks2658ce02019-10-15 11:35:02 +01004072 int, from_dfd, const char __user *, from_pathname,
4073 int, to_dfd, const char __user *, to_pathname,
David Howells2db154b2018-11-05 17:40:30 +00004074 unsigned int, flags)
4075{
4076 struct path from_path, to_path;
4077 unsigned int lflags;
4078 int ret = 0;
4079
4080 if (!may_mount())
4081 return -EPERM;
4082
4083 if (flags & ~MOVE_MOUNT__MASK)
4084 return -EINVAL;
4085
Christian Brauner6ac39282023-05-03 13:18:42 +02004086 if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) ==
4087 (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP))
4088 return -EINVAL;
4089
David Howells2db154b2018-11-05 17:40:30 +00004090 /* If someone gives a pathname, they aren't permitted to move
4091 * from an fd that requires unmount as we can't get at the flag
4092 * to clear it afterwards.
4093 */
4094 lflags = 0;
4095 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4096 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4097 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4098
4099 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
4100 if (ret < 0)
4101 return ret;
4102
4103 lflags = 0;
4104 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4105 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4106 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4107
4108 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
4109 if (ret < 0)
4110 goto out_from;
4111
4112 ret = security_move_mount(&from_path, &to_path);
4113 if (ret < 0)
4114 goto out_to;
4115
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03004116 if (flags & MOVE_MOUNT_SET_GROUP)
4117 ret = do_set_group(&from_path, &to_path);
4118 else
Christian Brauner6ac39282023-05-03 13:18:42 +02004119 ret = do_move_mount(&from_path, &to_path,
4120 (flags & MOVE_MOUNT_BENEATH));
David Howells2db154b2018-11-05 17:40:30 +00004121
4122out_to:
4123 path_put(&to_path);
4124out_from:
4125 path_put(&from_path);
4126 return ret;
4127}
4128
4129/*
Al Viroafac7cb2011-11-23 19:34:49 -05004130 * Return true if path is reachable from root
4131 *
Al Viro48a066e2013-09-29 22:06:07 -04004132 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05004133 */
Al Viro643822b2011-11-24 22:00:28 -05004134bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05004135 const struct path *root)
4136{
Al Viro643822b2011-11-24 22:00:28 -05004137 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05004138 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05004139 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05004140 }
Al Viro643822b2011-11-24 22:00:28 -05004141 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05004142}
4143
Mickaël Salaün640eb7e2016-11-14 22:14:35 +01004144bool path_is_under(const struct path *path1, const struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05004145{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08004146 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04004147 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05004148 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04004149 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05004150 return res;
4151}
4152EXPORT_SYMBOL(path_is_under);
4153
4154/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 * pivot_root Semantics:
4156 * Moves the root file system of the current process to the directory put_old,
4157 * makes new_root as the new root file system of the current process, and sets
4158 * root/cwd of all processes which had them on the current root to new_root.
4159 *
4160 * Restrictions:
4161 * The new_root and put_old must be directories, and must not be on the
4162 * same file system as the current process root. The put_old must be
4163 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4164 * pointed to by put_old must yield the same directory as new_root. No other
4165 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4166 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08004167 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
Mauro Carvalho Chehab0c1bc6b2020-04-14 18:48:37 +02004168 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
Neil Brown4a0d11f2006-01-08 01:03:18 -08004169 * in this situation.
4170 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171 * Notes:
4172 * - we don't move root/cwd if they are not at the root (reason: if something
4173 * cared enough to change them, it's probably wrong to force them elsewhere)
4174 * - it's okay to pick a root that isn't the root of a file system, e.g.
4175 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4176 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4177 * first.
4178 */
Heiko Carstens3480b252009-01-14 14:14:16 +01004179SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
4180 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004181{
Al Viro2763d112019-06-30 19:18:53 -04004182 struct path new, old, root;
4183 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
Al Viro84d171922013-03-15 10:53:28 -04004184 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 int error;
4186
Al Viro9b40bc92013-02-22 22:45:42 -05004187 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 return -EPERM;
4189
Al Viroce6595a2019-07-14 16:42:44 -04004190 error = user_path_at(AT_FDCWD, new_root,
4191 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004192 if (error)
4193 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004194
Al Viroce6595a2019-07-14 16:42:44 -04004195 error = user_path_at(AT_FDCWD, put_old,
4196 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004197 if (error)
4198 goto out1;
4199
Al Viro2d8f3032008-07-22 09:59:21 -04004200 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04004201 if (error)
4202 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02004204 get_fs_root(current->fs, &root);
Al Viro84d171922013-03-15 10:53:28 -04004205 old_mp = lock_mount(&old);
4206 error = PTR_ERR(old_mp);
4207 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04004208 goto out3;
4209
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05004211 new_mnt = real_mount(new.mnt);
4212 root_mnt = real_mount(root.mnt);
Al Viro84d171922013-03-15 10:53:28 -04004213 old_mnt = real_mount(old.mnt);
Al Viro2763d112019-06-30 19:18:53 -04004214 ex_parent = new_mnt->mnt_parent;
4215 root_parent = root_mnt->mnt_parent;
Al Viro84d171922013-03-15 10:53:28 -04004216 if (IS_MNT_SHARED(old_mnt) ||
Al Viro2763d112019-06-30 19:18:53 -04004217 IS_MNT_SHARED(ex_parent) ||
4218 IS_MNT_SHARED(root_parent))
Al Virob12cea92011-03-18 08:55:38 -04004219 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05004220 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004221 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07004222 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4223 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004224 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04004225 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04004226 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004227 error = -EBUSY;
Al Viro84d171922013-03-15 10:53:28 -04004228 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04004229 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004230 error = -EINVAL;
Christian Brauner78aa08a2023-05-03 13:18:39 +02004231 if (!path_mounted(&root))
Al Virob12cea92011-03-18 08:55:38 -04004232 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05004233 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004234 goto out4; /* not attached */
Christian Brauner78aa08a2023-05-03 13:18:39 +02004235 if (!path_mounted(&new))
Al Virob12cea92011-03-18 08:55:38 -04004236 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05004237 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004238 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08004239 /* make sure we can reach put_old from new_root */
Al Viro84d171922013-03-15 10:53:28 -04004240 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04004241 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07004242 /* make certain new is below the root */
4243 if (!is_path_reachable(new_mnt, new.dentry, &root))
4244 goto out4;
Al Viro719ea2f2013-09-29 11:24:49 -04004245 lock_mount_hash();
Al Viro2763d112019-06-30 19:18:53 -04004246 umount_mnt(new_mnt);
4247 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07004248 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4249 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4250 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4251 }
Jan Blunck4ac91372008-02-14 19:34:32 -08004252 /* mount old root on put_old */
Christian Brauner6ac39282023-05-03 13:18:42 +02004253 attach_mnt(root_mnt, old_mnt, old_mp, false);
Jan Blunck4ac91372008-02-14 19:34:32 -08004254 /* mount new_root on / */
Christian Brauner6ac39282023-05-03 13:18:42 +02004255 attach_mnt(new_mnt, root_parent, root_mp, false);
Al Viro2763d112019-06-30 19:18:53 -04004256 mnt_add_count(root_parent, -1);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08004257 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07004258 /* A moved mount should not expire automatically */
4259 list_del_init(&new_mnt->mnt_expire);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13004260 put_mountpoint(root_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04004261 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04004262 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004263 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04004264out4:
Al Viro84d171922013-03-15 10:53:28 -04004265 unlock_mount(old_mp);
Al Viro2763d112019-06-30 19:18:53 -04004266 if (!error)
4267 mntput_no_expire(ex_parent);
Al Virob12cea92011-03-18 08:55:38 -04004268out3:
Al Viro8c3ee422008-03-22 18:00:39 -04004269 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04004270out2:
Al Viro2d8f3032008-07-22 09:59:21 -04004271 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004272out1:
Al Viro2d8f3032008-07-22 09:59:21 -04004273 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004274out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004275 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004276}
4277
Christian Brauner2a186722021-01-21 14:19:53 +01004278static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4279{
4280 unsigned int flags = mnt->mnt.mnt_flags;
4281
4282 /* flags to clear */
4283 flags &= ~kattr->attr_clr;
4284 /* flags to raise */
4285 flags |= kattr->attr_set;
4286
4287 return flags;
4288}
4289
Christian Brauner9caccd42021-01-21 14:19:54 +01004290static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4291{
4292 struct vfsmount *m = &mnt->mnt;
Christian Braunerbd303362021-12-03 12:17:07 +01004293 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
Christian Brauner9caccd42021-01-21 14:19:54 +01004294
Christian Brauner256c8ae2022-10-26 12:51:27 +02004295 if (!kattr->mnt_idmap)
Christian Brauner9caccd42021-01-21 14:19:54 +01004296 return 0;
4297
4298 /*
Christian Braunerbd303362021-12-03 12:17:07 +01004299 * Creating an idmapped mount with the filesystem wide idmapping
4300 * doesn't make sense so block that. We don't allow mushy semantics.
4301 */
Christian Braunere65a29f2023-11-22 13:44:37 +01004302 if (kattr->mnt_userns == m->mnt_sb->s_user_ns)
Christian Braunerbd303362021-12-03 12:17:07 +01004303 return -EINVAL;
4304
4305 /*
Christian Brauner9caccd42021-01-21 14:19:54 +01004306 * Once a mount has been idmapped we don't allow it to change its
4307 * mapping. It makes things simpler and callers can just create
4308 * another bind-mount they can idmap if they want to.
4309 */
Christian Braunerbb49e9e2021-12-03 12:16:58 +01004310 if (is_idmapped_mnt(m))
Christian Brauner9caccd42021-01-21 14:19:54 +01004311 return -EPERM;
4312
4313 /* The underlying filesystem doesn't support idmapped mounts yet. */
4314 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4315 return -EINVAL;
4316
4317 /* We're not controlling the superblock. */
Christian Braunerbd303362021-12-03 12:17:07 +01004318 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
Christian Brauner9caccd42021-01-21 14:19:54 +01004319 return -EPERM;
4320
4321 /* Mount has already been visible in the filesystem hierarchy. */
4322 if (!is_anon_ns(mnt->mnt_ns))
4323 return -EINVAL;
4324
4325 return 0;
4326}
4327
Christian Braunera26f7882022-02-03 14:14:08 +01004328/**
4329 * mnt_allow_writers() - check whether the attribute change allows writers
4330 * @kattr: the new mount attributes
4331 * @mnt: the mount to which @kattr will be applied
4332 *
4333 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4334 *
4335 * Return: true if writers need to be held, false if not
4336 */
4337static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4338 const struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004339{
Christian Braunere1bbcd22022-05-10 11:58:40 +02004340 return (!(kattr->attr_set & MNT_READONLY) ||
4341 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
Christian Brauner256c8ae2022-10-26 12:51:27 +02004342 !kattr->mnt_idmap;
Christian Braunera26f7882022-02-03 14:14:08 +01004343}
Christian Brauner2a186722021-01-21 14:19:53 +01004344
Christian Brauner87bb5b62022-02-03 14:14:11 +01004345static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004346{
Al Viroe2570392022-02-28 23:04:20 -05004347 struct mount *m;
4348 int err;
Christian Brauner2a186722021-01-21 14:19:53 +01004349
Al Viroe2570392022-02-28 23:04:20 -05004350 for (m = mnt; m; m = next_mnt(m, mnt)) {
4351 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4352 err = -EPERM;
4353 break;
Christian Brauner2a186722021-01-21 14:19:53 +01004354 }
4355
Christian Brauner87bb5b62022-02-03 14:14:11 +01004356 err = can_idmap_mount(kattr, m);
4357 if (err)
Al Viroe2570392022-02-28 23:04:20 -05004358 break;
Christian Brauner9caccd42021-01-21 14:19:54 +01004359
Al Viroe2570392022-02-28 23:04:20 -05004360 if (!mnt_allow_writers(kattr, m)) {
4361 err = mnt_hold_writers(m);
4362 if (err)
4363 break;
Christian Brauner2a186722021-01-21 14:19:53 +01004364 }
Christian Brauner2a186722021-01-21 14:19:53 +01004365
Al Viroe2570392022-02-28 23:04:20 -05004366 if (!kattr->recurse)
4367 return 0;
4368 }
Christian Brauner2a186722021-01-21 14:19:53 +01004369
Al Viroe2570392022-02-28 23:04:20 -05004370 if (err) {
4371 struct mount *p;
4372
Christian Brauner0014eda2022-04-20 15:19:25 +02004373 /*
4374 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4375 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4376 * mounts and needs to take care to include the first mount.
4377 */
4378 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viroe2570392022-02-28 23:04:20 -05004379 /* If we had to hold writers unblock them. */
4380 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4381 mnt_unhold_writers(p);
Christian Brauner0014eda2022-04-20 15:19:25 +02004382
4383 /*
4384 * We're done once the first mount we changed got
4385 * MNT_WRITE_HOLD unset.
4386 */
4387 if (p == m)
4388 break;
Al Viroe2570392022-02-28 23:04:20 -05004389 }
4390 }
4391 return err;
Christian Brauner2a186722021-01-21 14:19:53 +01004392}
4393
Christian Brauner9caccd42021-01-21 14:19:54 +01004394static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4395{
Christian Brauner256c8ae2022-10-26 12:51:27 +02004396 if (!kattr->mnt_idmap)
Christian Brauner9caccd42021-01-21 14:19:54 +01004397 return;
4398
Christian Braunerbd303362021-12-03 12:17:07 +01004399 /*
Christian Brauner256c8ae2022-10-26 12:51:27 +02004400 * Pairs with smp_load_acquire() in mnt_idmap().
4401 *
4402 * Since we only allow a mount to change the idmapping once and
4403 * verified this in can_idmap_mount() we know that the mount has
4404 * @nop_mnt_idmap attached to it. So there's no need to drop any
4405 * references.
Christian Braunerbd303362021-12-03 12:17:07 +01004406 */
Christian Brauner256c8ae2022-10-26 12:51:27 +02004407 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
Christian Brauner9caccd42021-01-21 14:19:54 +01004408}
4409
Al Viroe2570392022-02-28 23:04:20 -05004410static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004411{
Al Viroe2570392022-02-28 23:04:20 -05004412 struct mount *m;
Christian Brauner2a186722021-01-21 14:19:53 +01004413
Al Viroe2570392022-02-28 23:04:20 -05004414 for (m = mnt; m; m = next_mnt(m, mnt)) {
4415 unsigned int flags;
Christian Brauner2a186722021-01-21 14:19:53 +01004416
Al Viroe2570392022-02-28 23:04:20 -05004417 do_idmap_mount(kattr, m);
4418 flags = recalc_flags(kattr, m);
4419 WRITE_ONCE(m->mnt.mnt_flags, flags);
Christian Brauner2a186722021-01-21 14:19:53 +01004420
Christian Brauner03b6abe2022-02-03 14:14:09 +01004421 /* If we had to hold writers unblock them. */
4422 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
Christian Brauner2a186722021-01-21 14:19:53 +01004423 mnt_unhold_writers(m);
4424
Al Viroe2570392022-02-28 23:04:20 -05004425 if (kattr->propagation)
Christian Brauner2a186722021-01-21 14:19:53 +01004426 change_mnt_propagation(m, kattr->propagation);
Al Viroe2570392022-02-28 23:04:20 -05004427 if (!kattr->recurse)
Christian Brauner2a186722021-01-21 14:19:53 +01004428 break;
Al Viroe2570392022-02-28 23:04:20 -05004429 }
4430 touch_mnt_namespace(mnt->mnt_ns);
Christian Brauner2a186722021-01-21 14:19:53 +01004431}
4432
4433static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4434{
Christian Brauner87bb5b62022-02-03 14:14:11 +01004435 struct mount *mnt = real_mount(path->mnt);
Christian Brauner2a186722021-01-21 14:19:53 +01004436 int err = 0;
4437
Christian Brauner78aa08a2023-05-03 13:18:39 +02004438 if (!path_mounted(path))
Christian Brauner2a186722021-01-21 14:19:53 +01004439 return -EINVAL;
4440
Christian Brauner256c8ae2022-10-26 12:51:27 +02004441 if (kattr->mnt_userns) {
4442 struct mnt_idmap *mnt_idmap;
4443
4444 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4445 if (IS_ERR(mnt_idmap))
4446 return PTR_ERR(mnt_idmap);
4447 kattr->mnt_idmap = mnt_idmap;
4448 }
4449
Christian Brauner2a186722021-01-21 14:19:53 +01004450 if (kattr->propagation) {
4451 /*
4452 * Only take namespace_lock() if we're actually changing
4453 * propagation.
4454 */
4455 namespace_lock();
4456 if (kattr->propagation == MS_SHARED) {
4457 err = invent_group_ids(mnt, kattr->recurse);
4458 if (err) {
4459 namespace_unlock();
4460 return err;
4461 }
4462 }
4463 }
4464
Christian Brauner87bb5b62022-02-03 14:14:11 +01004465 err = -EINVAL;
Christian Brauner2a186722021-01-21 14:19:53 +01004466 lock_mount_hash();
4467
Christian Brauner87bb5b62022-02-03 14:14:11 +01004468 /* Ensure that this isn't anything purely vfs internal. */
4469 if (!is_mounted(&mnt->mnt))
4470 goto out;
Christian Brauner2a186722021-01-21 14:19:53 +01004471
Christian Brauner87bb5b62022-02-03 14:14:11 +01004472 /*
4473 * If this is an attached mount make sure it's located in the callers
4474 * mount namespace. If it's not don't let the caller interact with it.
4475 * If this is a detached mount make sure it has an anonymous mount
4476 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4477 */
4478 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4479 goto out;
4480
4481 /*
4482 * First, we get the mount tree in a shape where we can change mount
4483 * properties without failure. If we succeeded to do so we commit all
4484 * changes and if we failed we clean up.
4485 */
4486 err = mount_setattr_prepare(kattr, mnt);
Al Viroe2570392022-02-28 23:04:20 -05004487 if (!err)
4488 mount_setattr_commit(kattr, mnt);
Christian Brauner87bb5b62022-02-03 14:14:11 +01004489
4490out:
Christian Brauner2a186722021-01-21 14:19:53 +01004491 unlock_mount_hash();
4492
4493 if (kattr->propagation) {
Christian Brauner2a186722021-01-21 14:19:53 +01004494 if (err)
4495 cleanup_group_ids(mnt, NULL);
Christian Braunercb2239c2023-03-30 09:13:16 +02004496 namespace_unlock();
Christian Brauner2a186722021-01-21 14:19:53 +01004497 }
4498
4499 return err;
4500}
4501
Christian Brauner9caccd42021-01-21 14:19:54 +01004502static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4503 struct mount_kattr *kattr, unsigned int flags)
4504{
4505 int err = 0;
4506 struct ns_common *ns;
4507 struct user_namespace *mnt_userns;
Al Viro96e85e92022-06-06 00:30:50 -04004508 struct fd f;
Christian Brauner9caccd42021-01-21 14:19:54 +01004509
4510 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4511 return 0;
4512
4513 /*
4514 * We currently do not support clearing an idmapped mount. If this ever
4515 * is a use-case we can revisit this but for now let's keep it simple
4516 * and not allow it.
4517 */
4518 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4519 return -EINVAL;
4520
4521 if (attr->userns_fd > INT_MAX)
4522 return -EINVAL;
4523
Al Viro96e85e92022-06-06 00:30:50 -04004524 f = fdget(attr->userns_fd);
4525 if (!f.file)
Christian Brauner9caccd42021-01-21 14:19:54 +01004526 return -EBADF;
4527
Al Viro96e85e92022-06-06 00:30:50 -04004528 if (!proc_ns_file(f.file)) {
Christian Brauner9caccd42021-01-21 14:19:54 +01004529 err = -EINVAL;
4530 goto out_fput;
4531 }
4532
Al Viro96e85e92022-06-06 00:30:50 -04004533 ns = get_proc_ns(file_inode(f.file));
Christian Brauner9caccd42021-01-21 14:19:54 +01004534 if (ns->ops->type != CLONE_NEWUSER) {
4535 err = -EINVAL;
4536 goto out_fput;
4537 }
4538
4539 /*
Christian Braunerbd303362021-12-03 12:17:07 +01004540 * The initial idmapping cannot be used to create an idmapped
4541 * mount. We use the initial idmapping as an indicator of a mount
4542 * that is not idmapped. It can simply be passed into helpers that
4543 * are aware of idmapped mounts as a convenient shortcut. A user
4544 * can just create a dedicated identity mapping to achieve the same
4545 * result.
Christian Brauner9caccd42021-01-21 14:19:54 +01004546 */
4547 mnt_userns = container_of(ns, struct user_namespace, ns);
Christian Brauner3707d842023-01-13 12:49:33 +01004548 if (mnt_userns == &init_user_ns) {
Christian Brauner9caccd42021-01-21 14:19:54 +01004549 err = -EPERM;
4550 goto out_fput;
4551 }
Seth Forsheebf1ac162022-08-16 11:47:52 -05004552
4553 /* We're not controlling the target namespace. */
4554 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4555 err = -EPERM;
4556 goto out_fput;
4557 }
4558
Christian Brauner9caccd42021-01-21 14:19:54 +01004559 kattr->mnt_userns = get_user_ns(mnt_userns);
4560
4561out_fput:
Al Viro96e85e92022-06-06 00:30:50 -04004562 fdput(f);
Christian Brauner9caccd42021-01-21 14:19:54 +01004563 return err;
4564}
4565
4566static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
Christian Brauner2a186722021-01-21 14:19:53 +01004567 struct mount_kattr *kattr, unsigned int flags)
4568{
4569 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4570
4571 if (flags & AT_NO_AUTOMOUNT)
4572 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4573 if (flags & AT_SYMLINK_NOFOLLOW)
4574 lookup_flags &= ~LOOKUP_FOLLOW;
4575 if (flags & AT_EMPTY_PATH)
4576 lookup_flags |= LOOKUP_EMPTY;
4577
4578 *kattr = (struct mount_kattr) {
4579 .lookup_flags = lookup_flags,
4580 .recurse = !!(flags & AT_RECURSIVE),
4581 };
4582
4583 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4584 return -EINVAL;
4585 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4586 return -EINVAL;
4587 kattr->propagation = attr->propagation;
4588
4589 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4590 return -EINVAL;
4591
Christian Brauner2a186722021-01-21 14:19:53 +01004592 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4593 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4594
4595 /*
4596 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4597 * users wanting to transition to a different atime setting cannot
4598 * simply specify the atime setting in @attr_set, but must also
4599 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4600 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4601 * @attr_clr and that @attr_set can't have any atime bits set if
4602 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4603 */
4604 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4605 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4606 return -EINVAL;
4607
4608 /*
4609 * Clear all previous time settings as they are mutually
4610 * exclusive.
4611 */
4612 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4613 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4614 case MOUNT_ATTR_RELATIME:
4615 kattr->attr_set |= MNT_RELATIME;
4616 break;
4617 case MOUNT_ATTR_NOATIME:
4618 kattr->attr_set |= MNT_NOATIME;
4619 break;
4620 case MOUNT_ATTR_STRICTATIME:
4621 break;
4622 default:
4623 return -EINVAL;
4624 }
4625 } else {
4626 if (attr->attr_set & MOUNT_ATTR__ATIME)
4627 return -EINVAL;
4628 }
4629
Christian Brauner9caccd42021-01-21 14:19:54 +01004630 return build_mount_idmapped(attr, usize, kattr, flags);
4631}
4632
4633static void finish_mount_kattr(struct mount_kattr *kattr)
4634{
4635 put_user_ns(kattr->mnt_userns);
4636 kattr->mnt_userns = NULL;
Christian Brauner256c8ae2022-10-26 12:51:27 +02004637
4638 if (kattr->mnt_idmap)
4639 mnt_idmap_put(kattr->mnt_idmap);
Christian Brauner2a186722021-01-21 14:19:53 +01004640}
4641
4642SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4643 unsigned int, flags, struct mount_attr __user *, uattr,
4644 size_t, usize)
4645{
4646 int err;
4647 struct path target;
4648 struct mount_attr attr;
4649 struct mount_kattr kattr;
4650
4651 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4652
4653 if (flags & ~(AT_EMPTY_PATH |
4654 AT_RECURSIVE |
4655 AT_SYMLINK_NOFOLLOW |
4656 AT_NO_AUTOMOUNT))
4657 return -EINVAL;
4658
4659 if (unlikely(usize > PAGE_SIZE))
4660 return -E2BIG;
4661 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4662 return -EINVAL;
4663
4664 if (!may_mount())
4665 return -EPERM;
4666
4667 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4668 if (err)
4669 return err;
4670
4671 /* Don't bother walking through the mounts if this is a nop. */
4672 if (attr.attr_set == 0 &&
4673 attr.attr_clr == 0 &&
4674 attr.propagation == 0)
4675 return 0;
4676
Christian Brauner9caccd42021-01-21 14:19:54 +01004677 err = build_mount_kattr(&attr, usize, &kattr, flags);
Christian Brauner2a186722021-01-21 14:19:53 +01004678 if (err)
4679 return err;
4680
4681 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
Christian Brauner012e3322021-12-30 20:23:09 +01004682 if (!err) {
4683 err = do_mount_setattr(&target, &kattr);
4684 path_put(&target);
4685 }
Christian Brauner9caccd42021-01-21 14:19:54 +01004686 finish_mount_kattr(&kattr);
Christian Brauner2a186722021-01-21 14:19:53 +01004687 return err;
4688}
4689
Miklos Szeredi56c94c62023-10-25 16:02:01 +02004690int show_path(struct seq_file *m, struct dentry *root)
4691{
4692 if (root->d_sb->s_op->show_path)
4693 return root->d_sb->s_op->show_path(m, root);
4694
4695 seq_dentry(m, root, " \t\n\\");
4696 return 0;
4697}
4698
Miklos Szeredi46eae992023-10-25 16:02:02 +02004699static struct vfsmount *lookup_mnt_in_ns(u64 id, struct mnt_namespace *ns)
4700{
4701 struct mount *mnt = mnt_find_id_at(ns, id);
4702
4703 if (!mnt || mnt->mnt_id_unique != id)
4704 return NULL;
4705
4706 return &mnt->mnt;
4707}
4708
4709struct kstatmount {
Christian Brauner68385d72023-11-19 20:58:49 +01004710 struct statmount __user *buf;
4711 size_t bufsize;
4712 struct vfsmount *mnt;
4713 u64 mask;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004714 struct path root;
4715 struct statmount sm;
Christian Brauner68385d72023-11-19 20:58:49 +01004716 struct seq_file seq;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004717};
4718
Miklos Szeredi46eae992023-10-25 16:02:02 +02004719static u64 mnt_to_attr_flags(struct vfsmount *mnt)
4720{
4721 unsigned int mnt_flags = READ_ONCE(mnt->mnt_flags);
4722 u64 attr_flags = 0;
4723
4724 if (mnt_flags & MNT_READONLY)
4725 attr_flags |= MOUNT_ATTR_RDONLY;
4726 if (mnt_flags & MNT_NOSUID)
4727 attr_flags |= MOUNT_ATTR_NOSUID;
4728 if (mnt_flags & MNT_NODEV)
4729 attr_flags |= MOUNT_ATTR_NODEV;
4730 if (mnt_flags & MNT_NOEXEC)
4731 attr_flags |= MOUNT_ATTR_NOEXEC;
4732 if (mnt_flags & MNT_NODIRATIME)
4733 attr_flags |= MOUNT_ATTR_NODIRATIME;
4734 if (mnt_flags & MNT_NOSYMFOLLOW)
4735 attr_flags |= MOUNT_ATTR_NOSYMFOLLOW;
4736
4737 if (mnt_flags & MNT_NOATIME)
4738 attr_flags |= MOUNT_ATTR_NOATIME;
4739 else if (mnt_flags & MNT_RELATIME)
4740 attr_flags |= MOUNT_ATTR_RELATIME;
4741 else
4742 attr_flags |= MOUNT_ATTR_STRICTATIME;
4743
4744 if (is_idmapped_mnt(mnt))
4745 attr_flags |= MOUNT_ATTR_IDMAP;
4746
4747 return attr_flags;
4748}
4749
4750static u64 mnt_to_propagation_flags(struct mount *m)
4751{
4752 u64 propagation = 0;
4753
4754 if (IS_MNT_SHARED(m))
4755 propagation |= MS_SHARED;
4756 if (IS_MNT_SLAVE(m))
4757 propagation |= MS_SLAVE;
4758 if (IS_MNT_UNBINDABLE(m))
4759 propagation |= MS_UNBINDABLE;
4760 if (!propagation)
4761 propagation |= MS_PRIVATE;
4762
4763 return propagation;
4764}
4765
Christian Brauner6971beb2023-11-19 19:03:33 +01004766static void statmount_sb_basic(struct kstatmount *s)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004767{
4768 struct super_block *sb = s->mnt->mnt_sb;
4769
Christian Brauner6971beb2023-11-19 19:03:33 +01004770 s->sm.mask |= STATMOUNT_SB_BASIC;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004771 s->sm.sb_dev_major = MAJOR(sb->s_dev);
4772 s->sm.sb_dev_minor = MINOR(sb->s_dev);
4773 s->sm.sb_magic = sb->s_magic;
4774 s->sm.sb_flags = sb->s_flags & (SB_RDONLY|SB_SYNCHRONOUS|SB_DIRSYNC|SB_LAZYTIME);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004775}
4776
Christian Brauner6971beb2023-11-19 19:03:33 +01004777static void statmount_mnt_basic(struct kstatmount *s)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004778{
4779 struct mount *m = real_mount(s->mnt);
4780
Christian Brauner6971beb2023-11-19 19:03:33 +01004781 s->sm.mask |= STATMOUNT_MNT_BASIC;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004782 s->sm.mnt_id = m->mnt_id_unique;
4783 s->sm.mnt_parent_id = m->mnt_parent->mnt_id_unique;
4784 s->sm.mnt_id_old = m->mnt_id;
4785 s->sm.mnt_parent_id_old = m->mnt_parent->mnt_id;
4786 s->sm.mnt_attr = mnt_to_attr_flags(&m->mnt);
4787 s->sm.mnt_propagation = mnt_to_propagation_flags(m);
4788 s->sm.mnt_peer_group = IS_MNT_SHARED(m) ? m->mnt_group_id : 0;
4789 s->sm.mnt_master = IS_MNT_SLAVE(m) ? m->mnt_master->mnt_group_id : 0;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004790}
4791
Christian Brauner6971beb2023-11-19 19:03:33 +01004792static void statmount_propagate_from(struct kstatmount *s)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004793{
4794 struct mount *m = real_mount(s->mnt);
4795
Christian Brauner6971beb2023-11-19 19:03:33 +01004796 s->sm.mask |= STATMOUNT_PROPAGATE_FROM;
4797 if (IS_MNT_SLAVE(m))
4798 s->sm.propagate_from = get_dominating_id(m, &current->fs->root);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004799}
4800
Christian Brauner68385d72023-11-19 20:58:49 +01004801static int statmount_mnt_root(struct kstatmount *s, struct seq_file *seq)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004802{
Christian Brauner68385d72023-11-19 20:58:49 +01004803 int ret;
4804 size_t start = seq->count;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004805
Christian Brauner68385d72023-11-19 20:58:49 +01004806 ret = show_path(seq, s->mnt->mnt_root);
4807 if (ret)
4808 return ret;
4809
4810 if (unlikely(seq_has_overflowed(seq)))
4811 return -EAGAIN;
4812
4813 /*
4814 * Unescape the result. It would be better if supplied string was not
4815 * escaped in the first place, but that's a pretty invasive change.
4816 */
4817 seq->buf[seq->count] = '\0';
4818 seq->count = start;
4819 seq_commit(seq, string_unescape_inplace(seq->buf + start, UNESCAPE_OCTAL));
4820 return 0;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004821}
4822
Christian Brauner68385d72023-11-19 20:58:49 +01004823static int statmount_mnt_point(struct kstatmount *s, struct seq_file *seq)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004824{
4825 struct vfsmount *mnt = s->mnt;
4826 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Christian Brauner68385d72023-11-19 20:58:49 +01004827 int err;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004828
Christian Brauner68385d72023-11-19 20:58:49 +01004829 err = seq_path_root(seq, &mnt_path, &s->root, "");
Miklos Szeredi46eae992023-10-25 16:02:02 +02004830 return err == SEQ_SKIP ? 0 : err;
4831}
4832
Christian Brauner68385d72023-11-19 20:58:49 +01004833static int statmount_fs_type(struct kstatmount *s, struct seq_file *seq)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004834{
Miklos Szeredi46eae992023-10-25 16:02:02 +02004835 struct super_block *sb = s->mnt->mnt_sb;
4836
4837 seq_puts(seq, sb->s_type->name);
4838 return 0;
4839}
4840
Christian Brauner68385d72023-11-19 20:58:49 +01004841static int statmount_string(struct kstatmount *s, u64 flag)
4842{
4843 int ret;
4844 size_t kbufsize;
4845 struct seq_file *seq = &s->seq;
4846 struct statmount *sm = &s->sm;
4847
4848 switch (flag) {
4849 case STATMOUNT_FS_TYPE:
4850 sm->fs_type = seq->count;
4851 ret = statmount_fs_type(s, seq);
4852 break;
4853 case STATMOUNT_MNT_ROOT:
4854 sm->mnt_root = seq->count;
4855 ret = statmount_mnt_root(s, seq);
4856 break;
4857 case STATMOUNT_MNT_POINT:
4858 sm->mnt_point = seq->count;
4859 ret = statmount_mnt_point(s, seq);
4860 break;
4861 default:
4862 WARN_ON_ONCE(true);
4863 return -EINVAL;
4864 }
4865
4866 if (unlikely(check_add_overflow(sizeof(*sm), seq->count, &kbufsize)))
4867 return -EOVERFLOW;
4868 if (kbufsize >= s->bufsize)
4869 return -EOVERFLOW;
4870
4871 /* signal a retry */
4872 if (unlikely(seq_has_overflowed(seq)))
4873 return -EAGAIN;
4874
4875 if (ret)
4876 return ret;
4877
4878 seq->buf[seq->count++] = '\0';
4879 sm->mask |= flag;
4880 return 0;
4881}
4882
4883static int copy_statmount_to_user(struct kstatmount *s)
Miklos Szeredi46eae992023-10-25 16:02:02 +02004884{
4885 struct statmount *sm = &s->sm;
Christian Brauner68385d72023-11-19 20:58:49 +01004886 struct seq_file *seq = &s->seq;
4887 char __user *str = ((char __user *)s->buf) + sizeof(*sm);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004888 size_t copysize = min_t(size_t, s->bufsize, sizeof(*sm));
Christian Brauner68385d72023-11-19 20:58:49 +01004889
4890 if (seq->count && copy_to_user(str, seq->buf, seq->count))
4891 return -EFAULT;
4892
4893 /* Return the number of bytes copied to the buffer */
4894 sm->size = copysize + seq->count;
4895 if (copy_to_user(s->buf, sm, copysize))
4896 return -EFAULT;
4897
4898 return 0;
4899}
4900
4901static int do_statmount(struct kstatmount *s)
4902{
4903 struct mount *m = real_mount(s->mnt);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004904 int err;
4905
4906 /*
4907 * Don't trigger audit denials. We just want to determine what
4908 * mounts to show users.
4909 */
4910 if (!is_path_reachable(m, m->mnt.mnt_root, &s->root) &&
4911 !ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN))
4912 return -EPERM;
4913
4914 err = security_sb_statfs(s->mnt->mnt_root);
4915 if (err)
4916 return err;
4917
Christian Brauner6971beb2023-11-19 19:03:33 +01004918 if (s->mask & STATMOUNT_SB_BASIC)
4919 statmount_sb_basic(s);
4920
4921 if (s->mask & STATMOUNT_MNT_BASIC)
4922 statmount_mnt_basic(s);
4923
4924 if (s->mask & STATMOUNT_PROPAGATE_FROM)
4925 statmount_propagate_from(s);
4926
Christian Brauner68385d72023-11-19 20:58:49 +01004927 if (s->mask & STATMOUNT_FS_TYPE)
4928 err = statmount_string(s, STATMOUNT_FS_TYPE);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004929
Christian Brauner68385d72023-11-19 20:58:49 +01004930 if (!err && s->mask & STATMOUNT_MNT_ROOT)
4931 err = statmount_string(s, STATMOUNT_MNT_ROOT);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004932
Christian Brauner68385d72023-11-19 20:58:49 +01004933 if (!err && s->mask & STATMOUNT_MNT_POINT)
4934 err = statmount_string(s, STATMOUNT_MNT_POINT);
Miklos Szeredi46eae992023-10-25 16:02:02 +02004935
Christian Brauner68385d72023-11-19 20:58:49 +01004936 if (err)
4937 return err;
4938
4939 return 0;
4940}
4941
4942static inline bool retry_statmount(const long ret, size_t *seq_size)
4943{
4944 if (likely(ret != -EAGAIN))
4945 return false;
4946 if (unlikely(check_mul_overflow(*seq_size, 2, seq_size)))
4947 return false;
4948 if (unlikely(*seq_size > MAX_RW_COUNT))
4949 return false;
4950 return true;
4951}
4952
4953static int prepare_kstatmount(struct kstatmount *ks, struct mnt_id_req *kreq,
4954 struct statmount __user *buf, size_t bufsize,
4955 size_t seq_size)
4956{
4957 if (!access_ok(buf, bufsize))
Miklos Szeredi46eae992023-10-25 16:02:02 +02004958 return -EFAULT;
4959
Christian Brauner68385d72023-11-19 20:58:49 +01004960 memset(ks, 0, sizeof(*ks));
Miklos Szeredib4c2bea2023-10-25 16:02:03 +02004961 ks->mask = kreq->param;
Christian Brauner68385d72023-11-19 20:58:49 +01004962 ks->buf = buf;
4963 ks->bufsize = bufsize;
4964 ks->seq.size = seq_size;
4965 ks->seq.buf = kvmalloc(seq_size, GFP_KERNEL_ACCOUNT);
4966 if (!ks->seq.buf)
4967 return -ENOMEM;
Miklos Szeredi46eae992023-10-25 16:02:02 +02004968 return 0;
4969}
4970
Christian Brauner35e27a52023-11-29 12:27:15 +01004971static int copy_mnt_id_req(const struct mnt_id_req __user *req,
4972 struct mnt_id_req *kreq)
4973{
4974 int ret;
4975 size_t usize;
4976
4977 BUILD_BUG_ON(sizeof(struct mnt_id_req) != MNT_ID_REQ_SIZE_VER0);
4978
4979 ret = get_user(usize, &req->size);
4980 if (ret)
4981 return -EFAULT;
4982 if (unlikely(usize > PAGE_SIZE))
4983 return -E2BIG;
4984 if (unlikely(usize < MNT_ID_REQ_SIZE_VER0))
4985 return -EINVAL;
4986 memset(kreq, 0, sizeof(*kreq));
4987 ret = copy_struct_from_user(kreq, sizeof(*kreq), req, usize);
4988 if (ret)
4989 return ret;
4990 if (kreq->spare != 0)
4991 return -EINVAL;
4992 return 0;
4993}
4994
Miklos Szeredi46eae992023-10-25 16:02:02 +02004995SYSCALL_DEFINE4(statmount, const struct mnt_id_req __user *, req,
4996 struct statmount __user *, buf, size_t, bufsize,
4997 unsigned int, flags)
4998{
4999 struct vfsmount *mnt;
5000 struct mnt_id_req kreq;
Christian Brauner68385d72023-11-19 20:58:49 +01005001 struct kstatmount ks;
5002 /* We currently support retrieval of 3 strings. */
5003 size_t seq_size = 3 * PATH_MAX;
Miklos Szeredi46eae992023-10-25 16:02:02 +02005004 int ret;
5005
5006 if (flags)
5007 return -EINVAL;
5008
Christian Brauner35e27a52023-11-29 12:27:15 +01005009 ret = copy_mnt_id_req(req, &kreq);
5010 if (ret)
5011 return ret;
Miklos Szeredi46eae992023-10-25 16:02:02 +02005012
Christian Brauner68385d72023-11-19 20:58:49 +01005013retry:
5014 ret = prepare_kstatmount(&ks, &kreq, buf, bufsize, seq_size);
5015 if (ret)
5016 return ret;
5017
Miklos Szeredi46eae992023-10-25 16:02:02 +02005018 down_read(&namespace_sem);
5019 mnt = lookup_mnt_in_ns(kreq.mnt_id, current->nsproxy->mnt_ns);
Christian Brauner68385d72023-11-19 20:58:49 +01005020 if (!mnt) {
5021 up_read(&namespace_sem);
5022 kvfree(ks.seq.buf);
5023 return -ENOENT;
Miklos Szeredi46eae992023-10-25 16:02:02 +02005024 }
Christian Brauner68385d72023-11-19 20:58:49 +01005025
5026 ks.mnt = mnt;
5027 get_fs_root(current->fs, &ks.root);
5028 ret = do_statmount(&ks);
5029 path_put(&ks.root);
Miklos Szeredi46eae992023-10-25 16:02:02 +02005030 up_read(&namespace_sem);
5031
Christian Brauner68385d72023-11-19 20:58:49 +01005032 if (!ret)
5033 ret = copy_statmount_to_user(&ks);
5034 kvfree(ks.seq.buf);
5035 if (retry_statmount(ret, &seq_size))
5036 goto retry;
Miklos Szeredi46eae992023-10-25 16:02:02 +02005037 return ret;
5038}
5039
Miklos Szeredib4c2bea2023-10-25 16:02:03 +02005040static struct mount *listmnt_next(struct mount *curr)
5041{
5042 return node_to_mount(rb_next(&curr->mnt_node));
5043}
5044
5045static ssize_t do_listmount(struct mount *first, struct path *orig, u64 mnt_id,
5046 u64 __user *buf, size_t bufsize,
5047 const struct path *root)
5048{
5049 struct mount *r;
5050 ssize_t ctr;
5051 int err;
5052
5053 /*
5054 * Don't trigger audit denials. We just want to determine what
5055 * mounts to show users.
5056 */
5057 if (!is_path_reachable(real_mount(orig->mnt), orig->dentry, root) &&
5058 !ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN))
5059 return -EPERM;
5060
5061 err = security_sb_statfs(orig->dentry);
5062 if (err)
5063 return err;
5064
5065 for (ctr = 0, r = first; r && ctr < bufsize; r = listmnt_next(r)) {
5066 if (r->mnt_id_unique == mnt_id)
5067 continue;
5068 if (!is_path_reachable(r, r->mnt.mnt_root, orig))
5069 continue;
5070 ctr = array_index_nospec(ctr, bufsize);
5071 if (put_user(r->mnt_id_unique, buf + ctr))
5072 return -EFAULT;
5073 if (check_add_overflow(ctr, 1, &ctr))
5074 return -ERANGE;
5075 }
5076 return ctr;
5077}
5078
5079SYSCALL_DEFINE4(listmount, const struct mnt_id_req __user *, req,
5080 u64 __user *, buf, size_t, bufsize, unsigned int, flags)
5081{
5082 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
5083 struct mnt_id_req kreq;
5084 struct mount *first;
5085 struct path root, orig;
5086 u64 mnt_id, last_mnt_id;
5087 ssize_t ret;
5088
5089 if (flags)
5090 return -EINVAL;
5091
Christian Brauner35e27a52023-11-29 12:27:15 +01005092 ret = copy_mnt_id_req(req, &kreq);
5093 if (ret)
5094 return ret;
Miklos Szeredib4c2bea2023-10-25 16:02:03 +02005095 mnt_id = kreq.mnt_id;
5096 last_mnt_id = kreq.param;
5097
5098 down_read(&namespace_sem);
5099 get_fs_root(current->fs, &root);
5100 if (mnt_id == LSMT_ROOT) {
5101 orig = root;
5102 } else {
5103 ret = -ENOENT;
5104 orig.mnt = lookup_mnt_in_ns(mnt_id, ns);
5105 if (!orig.mnt)
5106 goto err;
5107 orig.dentry = orig.mnt->mnt_root;
5108 }
5109 if (!last_mnt_id)
5110 first = node_to_mount(rb_first(&ns->mounts));
5111 else
5112 first = mnt_find_id_at(ns, last_mnt_id + 1);
5113
5114 ret = do_listmount(first, &orig, mnt_id, buf, bufsize, &root);
5115err:
5116 path_put(&root);
5117 up_read(&namespace_sem);
5118 return ret;
5119}
5120
5121
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122static void __init init_mount_tree(void)
5123{
5124 struct vfsmount *mnt;
Al Viro74e83122019-01-30 13:30:21 -05005125 struct mount *m;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08005126 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08005127 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005128
Al Virofd3e0072019-05-30 17:48:35 -04005129 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005130 if (IS_ERR(mnt))
5131 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11005132
Al Viro74e83122019-01-30 13:30:21 -05005133 ns = alloc_mnt_ns(&init_user_ns, false);
Trond Myklebust3b22edc2009-06-23 17:29:49 -04005134 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 panic("Can't allocate initial namespace");
Al Viro74e83122019-01-30 13:30:21 -05005136 m = real_mount(mnt);
Al Viro74e83122019-01-30 13:30:21 -05005137 ns->root = m;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02005138 ns->nr_mounts = 1;
5139 mnt_add_to_ns(ns, m);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08005140 init_task.nsproxy->mnt_ns = ns;
5141 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005142
Al Virobe08d6d2011-12-06 13:32:36 -05005143 root.mnt = mnt;
5144 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07005145 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08005146
5147 set_fs_pwd(current->fs, &root);
5148 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005149}
5150
Denis Cheng74bf17c2007-10-16 23:26:30 -07005151void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005152{
Randy Dunlap15a67dd2006-09-29 01:58:57 -07005153 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005154
Al Viro7d6fec42011-11-23 12:14:10 -05005155 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Vasily Averin79f65402021-09-02 14:55:10 -07005156 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005157
Al Viro0818bf22014-02-28 13:46:44 -05005158 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04005159 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05005160 mhash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07005161 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05005162 &m_hash_shift, &m_hash_mask, 0, 0);
5163 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
5164 sizeof(struct hlist_head),
5165 mphash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07005166 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05005167 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005168
Al Viro84d171922013-03-15 10:53:28 -04005169 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005170 panic("Failed to allocate mount hash table\n");
5171
Tejun Heo4b93dc92013-11-28 14:54:43 -05005172 kernfs_init();
5173
Randy Dunlap15a67dd2006-09-29 01:58:57 -07005174 err = sysfs_init();
5175 if (err)
5176 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07005177 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06005178 fs_kobj = kobject_create_and_add("fs", NULL);
5179 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07005180 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Al Viro037f11b2019-06-01 18:09:44 -04005181 shmem_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005182 init_rootfs();
5183 init_mount_tree();
5184}
5185
Trond Myklebust616511d2009-06-22 15:09:13 -04005186void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005187{
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03005188 if (!refcount_dec_and_test(&ns->ns.count))
Trond Myklebust616511d2009-06-22 15:09:13 -04005189 return;
Al Viro7b00ed62013-09-16 21:19:20 -04005190 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07005191 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005192}
Al Viro9d412a42011-03-17 22:08:28 -04005193
David Howellsd911b452018-11-01 23:07:26 +00005194struct vfsmount *kern_mount(struct file_system_type *type)
Al Viro9d412a42011-03-17 22:08:28 -04005195{
Tim Chen423e0ab2011-07-19 09:32:38 -07005196 struct vfsmount *mnt;
David Howellsd911b452018-11-01 23:07:26 +00005197 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
Tim Chen423e0ab2011-07-19 09:32:38 -07005198 if (!IS_ERR(mnt)) {
5199 /*
5200 * it is a longterm mount, don't release mnt until
5201 * we unmount before file sys is unregistered
5202 */
Al Virof7a99c52012-06-09 00:59:08 -04005203 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07005204 }
5205 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04005206}
David Howellsd911b452018-11-01 23:07:26 +00005207EXPORT_SYMBOL_GPL(kern_mount);
Tim Chen423e0ab2011-07-19 09:32:38 -07005208
5209void kern_unmount(struct vfsmount *mnt)
5210{
5211 /* release long term mount so mount point can be released */
Rik van Rielda27f792023-01-27 13:46:51 -05005212 if (!IS_ERR(mnt)) {
5213 mnt_make_shortterm(mnt);
Al Viro48a066e2013-09-29 22:06:07 -04005214 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07005215 mntput(mnt);
5216 }
5217}
5218EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05005219
Miklos Szeredidf820f82020-06-04 10:48:19 +02005220void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
5221{
5222 unsigned int i;
5223
5224 for (i = 0; i < num; i++)
Rik van Rielda27f792023-01-27 13:46:51 -05005225 mnt_make_shortterm(mnt[i]);
Miklos Szeredidf820f82020-06-04 10:48:19 +02005226 synchronize_rcu_expedited();
5227 for (i = 0; i < num; i++)
5228 mntput(mnt[i]);
5229}
5230EXPORT_SYMBOL(kern_unmount_array);
5231
Al Viro02125a82011-12-05 08:43:34 -05005232bool our_mnt(struct vfsmount *mnt)
5233{
Al Viro143c8c92011-11-25 00:46:35 -05005234 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05005235}
Eric W. Biederman8823c072010-03-07 18:49:36 -08005236
Eric W. Biederman31515272013-03-15 01:45:51 -07005237bool current_chrooted(void)
5238{
5239 /* Does the current process have a non-standard root */
5240 struct path ns_root;
5241 struct path fs_root;
5242 bool chrooted;
5243
5244 /* Find the namespace root */
5245 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
5246 ns_root.dentry = ns_root.mnt->mnt_root;
5247 path_get(&ns_root);
5248 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
5249 ;
5250
5251 get_fs_root(current->fs, &fs_root);
5252
5253 chrooted = !path_equal(&fs_root, &ns_root);
5254
5255 path_put(&fs_root);
5256 path_put(&ns_root);
5257
5258 return chrooted;
5259}
5260
David Howells132e4602018-11-04 07:43:08 -05005261static bool mnt_already_visible(struct mnt_namespace *ns,
5262 const struct super_block *sb,
Eric W. Biederman8654df42016-06-09 16:06:06 -05005263 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07005264{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05005265 int new_flags = *new_mnt_flags;
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02005266 struct mount *mnt, *n;
Eric W. Biedermane51db732013-03-30 19:57:41 -07005267 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07005268
Al Viro44bb4382013-09-16 21:37:36 -04005269 down_read(&namespace_sem);
Miklos Szeredi2eea9ce2023-10-25 16:02:00 +02005270 rbtree_postorder_for_each_entry_safe(mnt, n, &ns->mounts, mnt_node) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07005271 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005272 int mnt_flags;
5273
David Howells132e4602018-11-04 07:43:08 -05005274 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07005275 continue;
5276
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05005277 /* This mount is not fully visible if it's root directory
5278 * is not the root directory of the filesystem.
5279 */
5280 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
5281 continue;
5282
Eric W. Biedermana1935c12016-06-15 06:59:49 -05005283 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005284 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005285
Eric W. Biederman695e9df2016-06-10 12:21:40 -05005286 /* Don't miss readonly hidden in the superblock flags */
David Howellsbc98a422017-07-17 08:45:34 +01005287 if (sb_rdonly(mnt->mnt.mnt_sb))
Eric W. Biederman695e9df2016-06-10 12:21:40 -05005288 mnt_flags |= MNT_LOCK_READONLY;
5289
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05005290 /* Verify the mount flags are equal to or more permissive
5291 * than the proposed new mount.
5292 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005293 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05005294 !(new_flags & MNT_READONLY))
5295 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005296 if ((mnt_flags & MNT_LOCK_ATIME) &&
5297 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05005298 continue;
5299
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06005300 /* This mount is not fully visible if there are any
5301 * locked child mounts that cover anything except for
5302 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07005303 */
5304 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
5305 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06005306 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05005307 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06005308 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05005309 /* Is the directory permanetly empty? */
5310 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07005311 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07005312 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05005313 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005314 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05005315 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07005316 visible = true;
5317 goto found;
5318 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07005319 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07005320found:
Al Viro44bb4382013-09-16 21:37:36 -04005321 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07005322 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07005323}
5324
David Howells132e4602018-11-04 07:43:08 -05005325static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
Eric W. Biederman8654df42016-06-09 16:06:06 -05005326{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05005327 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05005328 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
5329 unsigned long s_iflags;
5330
5331 if (ns->user_ns == &init_user_ns)
5332 return false;
5333
5334 /* Can this filesystem be too revealing? */
David Howells132e4602018-11-04 07:43:08 -05005335 s_iflags = sb->s_iflags;
Eric W. Biederman8654df42016-06-09 16:06:06 -05005336 if (!(s_iflags & SB_I_USERNS_VISIBLE))
5337 return false;
5338
Eric W. Biedermana1935c12016-06-15 06:59:49 -05005339 if ((s_iflags & required_iflags) != required_iflags) {
5340 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
5341 required_iflags);
5342 return true;
5343 }
5344
David Howells132e4602018-11-04 07:43:08 -05005345 return !mnt_already_visible(ns, sb, new_mnt_flags);
Eric W. Biederman8654df42016-06-09 16:06:06 -05005346}
5347
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05005348bool mnt_may_suid(struct vfsmount *mnt)
5349{
5350 /*
5351 * Foreign mounts (accessed via fchdir or through /proc
5352 * symlinks) are always treated as if they are nosuid. This
5353 * prevents namespaces from trusting potentially unsafe
5354 * suid/sgid bits, file caps, or security labels that originate
5355 * in other namespaces.
5356 */
5357 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
5358 current_in_userns(mnt->mnt_sb->s_user_ns);
5359}
5360
Al Viro64964522014-11-01 00:37:32 -04005361static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08005362{
Al Viro58be28252014-11-01 00:00:23 -04005363 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005364 struct nsproxy *nsproxy;
5365
Eric W. Biederman728dba32014-02-03 19:13:49 -08005366 task_lock(task);
5367 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005368 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04005369 ns = &nsproxy->mnt_ns->ns;
5370 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08005371 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08005372 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08005373
5374 return ns;
5375}
5376
Al Viro64964522014-11-01 00:37:32 -04005377static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08005378{
Al Viro58be28252014-11-01 00:00:23 -04005379 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08005380}
5381
Christian Braunerf2a8d522020-05-05 16:04:30 +02005382static int mntns_install(struct nsset *nsset, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08005383{
Christian Braunerf2a8d522020-05-05 16:04:30 +02005384 struct nsproxy *nsproxy = nsset->nsproxy;
5385 struct fs_struct *fs = nsset->fs;
Al Viro4f757f32017-04-15 17:31:22 -04005386 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
Christian Braunerf2a8d522020-05-05 16:04:30 +02005387 struct user_namespace *user_ns = nsset->cred->user_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005388 struct path root;
Al Viro4f757f32017-04-15 17:31:22 -04005389 int err;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005390
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07005391 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Christian Braunerf2a8d522020-05-05 16:04:30 +02005392 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
5393 !ns_capable(user_ns, CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08005394 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005395
Al Viro74e83122019-01-30 13:30:21 -05005396 if (is_anon_ns(mnt_ns))
5397 return -EINVAL;
5398
Eric W. Biederman8823c072010-03-07 18:49:36 -08005399 if (fs->users != 1)
5400 return -EINVAL;
5401
5402 get_mnt_ns(mnt_ns);
Al Viro4f757f32017-04-15 17:31:22 -04005403 old_mnt_ns = nsproxy->mnt_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08005404 nsproxy->mnt_ns = mnt_ns;
5405
5406 /* Find the root */
Al Viro4f757f32017-04-15 17:31:22 -04005407 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
5408 "/", LOOKUP_DOWN, &root);
5409 if (err) {
5410 /* revert to old namespace */
5411 nsproxy->mnt_ns = old_mnt_ns;
5412 put_mnt_ns(mnt_ns);
5413 return err;
5414 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08005415
Andrei Vagin40683672017-06-08 17:32:29 -07005416 put_mnt_ns(old_mnt_ns);
5417
Eric W. Biederman8823c072010-03-07 18:49:36 -08005418 /* Update the pwd and root */
5419 set_fs_pwd(fs, &root);
5420 set_fs_root(fs, &root);
5421
5422 path_put(&root);
5423 return 0;
5424}
5425
Andrey Vaginbcac25a2016-09-06 00:47:13 -07005426static struct user_namespace *mntns_owner(struct ns_common *ns)
5427{
5428 return to_mnt_ns(ns)->user_ns;
5429}
5430
Eric W. Biederman8823c072010-03-07 18:49:36 -08005431const struct proc_ns_operations mntns_operations = {
5432 .name = "mnt",
5433 .type = CLONE_NEWNS,
5434 .get = mntns_get,
5435 .put = mntns_put,
5436 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07005437 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08005438};
Luis Chamberlainab171b92022-01-21 22:13:27 -08005439
5440#ifdef CONFIG_SYSCTL
5441static struct ctl_table fs_namespace_sysctls[] = {
5442 {
5443 .procname = "mount-max",
5444 .data = &sysctl_mount_max,
5445 .maxlen = sizeof(unsigned int),
5446 .mode = 0644,
5447 .proc_handler = proc_dointvec_minmax,
5448 .extra1 = SYSCTL_ONE,
5449 },
5450 { }
5451};
5452
5453static int __init init_fs_namespace_sysctls(void)
5454{
5455 register_sysctl_init("fs", fs_namespace_sysctls);
5456 return 0;
5457}
5458fs_initcall(init_fs_namespace_sysctls);
5459
5460#endif /* CONFIG_SYSCTL */