blob: 9d33909d0f9e34d8a54c7d0d1d7b44652cc8b69e [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>
Al Viro9ea459e12014-08-08 13:08:20 -040028#include <linux/task_work.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010029#include <linux/sched/task.h>
David Howellse262e32d2018-11-01 23:07:23 +000030#include <uapi/linux/mount.h>
David Howells9bc61ab2018-11-04 03:19:03 -050031#include <linux/fs_context.h>
Al Viro037f11b2019-06-01 18:09:44 -040032#include <linux/shmem_fs.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010033
Ram Pai07b20882005-11-07 17:19:07 -050034#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Eric W. Biedermand2921682016-09-28 00:27:17 -050037/* Maximum number of mounts in a mount namespace */
38unsigned int sysctl_mount_max __read_mostly = 100000;
39
Al Viro0818bf22014-02-28 13:46:44 -050040static unsigned int m_hash_mask __read_mostly;
41static unsigned int m_hash_shift __read_mostly;
42static unsigned int mp_hash_mask __read_mostly;
43static unsigned int mp_hash_shift __read_mostly;
44
45static __initdata unsigned long mhash_entries;
46static int __init set_mhash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mhash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mhash_entries=", set_mhash_entries);
54
55static __initdata unsigned long mphash_entries;
56static int __init set_mphash_entries(char *str)
57{
58 if (!str)
59 return 0;
60 mphash_entries = simple_strtoul(str, &str, 0);
61 return 1;
62}
63__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080064
Al Viroc7999c32014-02-27 14:40:10 -050065static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010066static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010067static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Al Viro38129a12014-03-20 21:10:51 -040069static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050070static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080071static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040072static DECLARE_RWSEM(namespace_sem);
Al Viro4edbe132019-06-30 10:39:08 -040073static HLIST_HEAD(unmounted); /* protected by namespace_sem */
74static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080076/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060077struct kobject *fs_kobj;
78EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080079
Nick Piggin99b7db72010-08-18 04:37:39 +100080/*
81 * vfsmount lock may be taken for read to prevent changes to the
82 * vfsmount hash, ie. during mountpoint lookups or walking back
83 * up the tree.
84 *
85 * It should be taken for write in all cases where the vfsmount
86 * tree or hash is modified or when a vfsmount structure is modified.
87 */
Al Viro48a066e2013-09-29 22:06:07 -040088__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +100089
Al Viro38129a12014-03-20 21:10:51 -040090static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
Ram Paib58fed82005-11-07 17:16:09 -050092 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
93 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -050094 tmp = tmp + (tmp >> m_hash_shift);
95 return &mount_hashtable[tmp & m_hash_mask];
96}
97
98static inline struct hlist_head *mp_hash(struct dentry *dentry)
99{
100 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
101 tmp = tmp + (tmp >> mp_hash_shift);
102 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103}
104
Al Virob105e272011-11-24 20:38:33 -0500105static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100106{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400107 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100108
Matthew Wilcox169b4802018-06-11 12:31:36 -0400109 if (res < 0)
110 return res;
111 mnt->mnt_id = res;
112 return 0;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100113}
114
Al Virob105e272011-11-24 20:38:33 -0500115static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100116{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400117 ida_free(&mnt_id_ida, mnt->mnt_id);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100118}
119
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100120/*
121 * Allocate a new peer group ID
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100122 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500123static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100124{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400125 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
Al Virof21f6222009-06-24 03:12:00 -0400126
Matthew Wilcox169b4802018-06-11 12:31:36 -0400127 if (res < 0)
128 return res;
129 mnt->mnt_group_id = res;
130 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100131}
132
133/*
134 * Release a peer group ID
135 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500136void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100137{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400138 ida_free(&mnt_group_ida, mnt->mnt_group_id);
Al Viro15169fe2011-11-25 00:50:41 -0500139 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100140}
141
Nick Pigginb3e19d92011-01-07 17:50:11 +1100142/*
143 * vfsmount lock must be held for read
144 */
Al Viro83adc752011-11-24 22:37:54 -0500145static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100146{
147#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500148 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100149#else
150 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500151 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100152 preempt_enable();
153#endif
154}
155
Nick Pigginb3e19d92011-01-07 17:50:11 +1100156/*
157 * vfsmount lock must be held for write
158 */
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700159int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100160{
161#ifdef CONFIG_SMP
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700162 int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100163 int cpu;
164
165 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500166 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100167 }
168
169 return count;
170#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500171 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100172#endif
173}
174
Al Virob105e272011-11-24 20:38:33 -0500175static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
Al Viroc63181e2011-11-25 02:35:16 -0500177 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
178 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100179 int err;
180
Al Viroc63181e2011-11-25 02:35:16 -0500181 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800182 if (err)
183 goto out_free_cache;
184
185 if (name) {
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800186 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
Al Viroc63181e2011-11-25 02:35:16 -0500187 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800188 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100189 }
190
Nick Pigginb3e19d92011-01-07 17:50:11 +1100191#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500192 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100194 goto out_free_devname;
195
Al Viroc63181e2011-11-25 02:35:16 -0500196 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100197#else
Al Viroc63181e2011-11-25 02:35:16 -0500198 mnt->mnt_count = 1;
199 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100200#endif
201
Al Viro38129a12014-03-20 21:10:51 -0400202 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500203 INIT_LIST_HEAD(&mnt->mnt_child);
204 INIT_LIST_HEAD(&mnt->mnt_mounts);
205 INIT_LIST_HEAD(&mnt->mnt_list);
206 INIT_LIST_HEAD(&mnt->mnt_expire);
207 INIT_LIST_HEAD(&mnt->mnt_share);
208 INIT_LIST_HEAD(&mnt->mnt_slave_list);
209 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700210 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Eric W. Biederman99b19d12016-10-24 16:16:13 -0500211 INIT_LIST_HEAD(&mnt->mnt_umounting);
Al Viro56cbb4292019-07-04 16:57:51 -0400212 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 }
Al Viroc63181e2011-11-25 02:35:16 -0500214 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800215
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000216#ifdef CONFIG_SMP
217out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800218 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000219#endif
Li Zefan88b38782008-07-21 18:06:36 +0800220out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500221 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800222out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500223 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800224 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Dave Hansen83660252008-02-15 14:37:30 -0800227/*
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
233 * a filesystem.
234 */
Dave Hansen3d733632008-02-15 14:37:59 -0800235/*
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
238 *
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
244 * r/w.
245 */
David Howells43f5e652018-11-01 23:07:25 +0000246bool __mnt_is_readonly(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800247{
David Howells43f5e652018-11-01 23:07:25 +0000248 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800249}
250EXPORT_SYMBOL_GPL(__mnt_is_readonly);
251
Al Viro83adc752011-11-24 22:37:54 -0500252static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800253{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000254#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500255 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000256#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500257 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000258#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800259}
Dave Hansen3d733632008-02-15 14:37:59 -0800260
Al Viro83adc752011-11-24 22:37:54 -0500261static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800262{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000263#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500264 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000265#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500266 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000267#endif
268}
269
Al Viro83adc752011-11-24 22:37:54 -0500270static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000271{
272#ifdef CONFIG_SMP
273 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800274 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800275
276 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500277 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800278 }
Dave Hansen3d733632008-02-15 14:37:59 -0800279
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000280 return count;
281#else
282 return mnt->mnt_writers;
283#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800284}
285
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100286static int mnt_is_readonly(struct vfsmount *mnt)
287{
288 if (mnt->mnt_sb->s_readonly_remount)
289 return 1;
290 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
291 smp_rmb();
292 return __mnt_is_readonly(mnt);
293}
294
Dave Hansen3d733632008-02-15 14:37:59 -0800295/*
Jan Karaeb04c282012-06-12 16:20:35 +0200296 * Most r/o & frozen checks on a fs are for operations that take discrete
297 * amounts of time, like a write() or unlink(). We must keep track of when
298 * those operations start (for permission checks) and when they end, so that we
299 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800300 */
Dave Hansen83660252008-02-15 14:37:30 -0800301/**
Jan Karaeb04c282012-06-12 16:20:35 +0200302 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500303 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800304 *
Jan Karaeb04c282012-06-12 16:20:35 +0200305 * This tells the low-level filesystem that a write is about to be performed to
306 * it, and makes sure that writes are allowed (mnt it read-write) before
307 * returning success. This operation does not protect against filesystem being
308 * frozen. When the write operation is finished, __mnt_drop_write() must be
309 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800310 */
Jan Karaeb04c282012-06-12 16:20:35 +0200311int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800312{
Al Viro83adc752011-11-24 22:37:54 -0500313 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800314 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800315
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000316 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100317 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000318 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100319 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000320 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
321 * incremented count after it has set MNT_WRITE_HOLD.
322 */
323 smp_mb();
Mark Rutland6aa7de02017-10-23 14:07:29 -0700324 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000325 cpu_relax();
326 /*
327 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
328 * be set to match its requirements. So we must not load that until
329 * MNT_WRITE_HOLD is cleared.
330 */
331 smp_rmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100332 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100333 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800334 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800335 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000336 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200337
338 return ret;
339}
340
341/**
342 * mnt_want_write - get write access to a mount
343 * @m: the mount on which to take a write
344 *
345 * This tells the low-level filesystem that a write is about to be performed to
346 * it, and makes sure that writes are allowed (mount is read-write, filesystem
347 * is not frozen) before returning success. When the write operation is
348 * finished, mnt_drop_write() must be called. This is effectively a refcount.
349 */
350int mnt_want_write(struct vfsmount *m)
351{
352 int ret;
353
354 sb_start_write(m->mnt_sb);
355 ret = __mnt_want_write(m);
356 if (ret)
357 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800358 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800359}
360EXPORT_SYMBOL_GPL(mnt_want_write);
361
362/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000363 * mnt_clone_write - get write access to a mount
364 * @mnt: the mount on which to take a write
365 *
366 * This is effectively like mnt_want_write, except
367 * it must only be used to take an extra write reference
368 * on a mountpoint that we already know has a write reference
369 * on it. This allows some optimisation.
370 *
371 * After finished, mnt_drop_write must be called as usual to
372 * drop the reference.
373 */
374int mnt_clone_write(struct vfsmount *mnt)
375{
376 /* superblock may be r/o */
377 if (__mnt_is_readonly(mnt))
378 return -EROFS;
379 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500380 mnt_inc_writers(real_mount(mnt));
npiggin@suse.de96029c42009-04-26 20:25:55 +1000381 preempt_enable();
382 return 0;
383}
384EXPORT_SYMBOL_GPL(mnt_clone_write);
385
386/**
Jan Karaeb04c282012-06-12 16:20:35 +0200387 * __mnt_want_write_file - get write access to a file's mount
388 * @file: the file who's mount on which to take a write
389 *
390 * This is like __mnt_want_write, but it takes a file and can
391 * do some optimisations if the file is open for write already
392 */
393int __mnt_want_write_file(struct file *file)
394{
Al Viro83f936c2014-03-14 12:02:47 -0400395 if (!(file->f_mode & FMODE_WRITER))
Jan Karaeb04c282012-06-12 16:20:35 +0200396 return __mnt_want_write(file->f_path.mnt);
397 else
398 return mnt_clone_write(file->f_path.mnt);
399}
400
401/**
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200402 * mnt_want_write_file - get write access to a file's mount
403 * @file: the file who's mount on which to take a write
404 *
405 * This is like mnt_want_write, but it takes a file and can
406 * do some optimisations if the file is open for write already
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200407 */
408int mnt_want_write_file(struct file *file)
409{
410 int ret;
411
Miklos Szeredia6795a52018-07-18 15:44:43 +0200412 sb_start_write(file_inode(file)->i_sb);
Jan Karaeb04c282012-06-12 16:20:35 +0200413 ret = __mnt_want_write_file(file);
414 if (ret)
Miklos Szeredia6795a52018-07-18 15:44:43 +0200415 sb_end_write(file_inode(file)->i_sb);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200416 return ret;
417}
npiggin@suse.de96029c42009-04-26 20:25:55 +1000418EXPORT_SYMBOL_GPL(mnt_want_write_file);
419
420/**
Jan Karaeb04c282012-06-12 16:20:35 +0200421 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800422 * @mnt: the mount on which to give up write access
423 *
424 * Tells the low-level filesystem that we are done
425 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200426 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800427 */
Jan Karaeb04c282012-06-12 16:20:35 +0200428void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800429{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000430 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500431 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000432 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800433}
Jan Karaeb04c282012-06-12 16:20:35 +0200434
435/**
436 * mnt_drop_write - give up write access to a mount
437 * @mnt: the mount on which to give up write access
438 *
439 * Tells the low-level filesystem that we are done performing writes to it and
440 * also allows filesystem to be frozen again. Must be matched with
441 * mnt_want_write() call above.
442 */
443void mnt_drop_write(struct vfsmount *mnt)
444{
445 __mnt_drop_write(mnt);
446 sb_end_write(mnt->mnt_sb);
447}
Dave Hansen83660252008-02-15 14:37:30 -0800448EXPORT_SYMBOL_GPL(mnt_drop_write);
449
Jan Karaeb04c282012-06-12 16:20:35 +0200450void __mnt_drop_write_file(struct file *file)
451{
452 __mnt_drop_write(file->f_path.mnt);
453}
454
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200455void mnt_drop_write_file(struct file *file)
456{
Miklos Szeredia6795a52018-07-18 15:44:43 +0200457 __mnt_drop_write_file(file);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200458 sb_end_write(file_inode(file)->i_sb);
459}
Al Viro2a79f172011-12-09 08:06:57 -0500460EXPORT_SYMBOL(mnt_drop_write_file);
461
Al Viro83adc752011-11-24 22:37:54 -0500462static int mnt_make_readonly(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800463{
Dave Hansen3d733632008-02-15 14:37:59 -0800464 int ret = 0;
465
Al Viro719ea2f2013-09-29 11:24:49 -0400466 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500467 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000468 /*
469 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
470 * should be visible before we do.
471 */
472 smp_mb();
473
474 /*
475 * With writers on hold, if this value is zero, then there are
476 * definitely no active writers (although held writers may subsequently
477 * increment the count, they'll have to wait, and decrement it after
478 * seeing MNT_READONLY).
479 *
480 * It is OK to have counter incremented on one CPU and decremented on
481 * another: the sum will add up correctly. The danger would be when we
482 * sum up each counter, if we read a counter before it is incremented,
483 * but then read another CPU's count which it has been subsequently
484 * decremented from -- we would see more decrements than we should.
485 * MNT_WRITE_HOLD protects against this scenario, because
486 * mnt_want_write first increments count, then smp_mb, then spins on
487 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
488 * we're counting up here.
489 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100490 if (mnt_get_writers(mnt) > 0)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000491 ret = -EBUSY;
492 else
Al Viro83adc752011-11-24 22:37:54 -0500493 mnt->mnt.mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000494 /*
495 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
496 * that become unheld will see MNT_READONLY.
497 */
498 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500499 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Al Viro719ea2f2013-09-29 11:24:49 -0400500 unlock_mount_hash();
Dave Hansen3d733632008-02-15 14:37:59 -0800501 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800502}
Dave Hansen83660252008-02-15 14:37:30 -0800503
David Howells43f5e652018-11-01 23:07:25 +0000504static int __mnt_unmake_readonly(struct mount *mnt)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800505{
Al Viro719ea2f2013-09-29 11:24:49 -0400506 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -0500507 mnt->mnt.mnt_flags &= ~MNT_READONLY;
Al Viro719ea2f2013-09-29 11:24:49 -0400508 unlock_mount_hash();
David Howells43f5e652018-11-01 23:07:25 +0000509 return 0;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800510}
511
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100512int sb_prepare_remount_readonly(struct super_block *sb)
513{
514 struct mount *mnt;
515 int err = 0;
516
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100517 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
518 if (atomic_long_read(&sb->s_remove_count))
519 return -EBUSY;
520
Al Viro719ea2f2013-09-29 11:24:49 -0400521 lock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100522 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
523 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
524 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
525 smp_mb();
526 if (mnt_get_writers(mnt) > 0) {
527 err = -EBUSY;
528 break;
529 }
530 }
531 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100532 if (!err && atomic_long_read(&sb->s_remove_count))
533 err = -EBUSY;
534
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100535 if (!err) {
536 sb->s_readonly_remount = 1;
537 smp_wmb();
538 }
539 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
540 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
541 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
542 }
Al Viro719ea2f2013-09-29 11:24:49 -0400543 unlock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100544
545 return err;
546}
547
Al Virob105e272011-11-24 20:38:33 -0500548static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800550 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000551#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500552 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000553#endif
Al Virob105e272011-11-24 20:38:33 -0500554 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
556
David Howells8ffcb322014-01-24 12:17:54 +0000557static void delayed_free_vfsmnt(struct rcu_head *head)
558{
559 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
560}
561
Al Viro48a066e2013-09-29 22:06:07 -0400562/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400563int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400564{
565 struct mount *mnt;
566 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400567 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400568 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400569 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400570 mnt = real_mount(bastard);
571 mnt_add_count(mnt, 1);
Al Viro119e1ef2018-08-09 17:51:32 -0400572 smp_mb(); // see mntput_no_expire()
Al Viro48a066e2013-09-29 22:06:07 -0400573 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400574 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400575 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
576 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400577 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400578 }
Al Viro119e1ef2018-08-09 17:51:32 -0400579 lock_mount_hash();
580 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
581 mnt_add_count(mnt, -1);
582 unlock_mount_hash();
583 return 1;
584 }
585 unlock_mount_hash();
586 /* caller will mntput() */
Al Viro294d71f2015-05-08 11:43:53 -0400587 return -1;
588}
589
590/* call under rcu_read_lock */
591bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
592{
593 int res = __legitimize_mnt(bastard, seq);
594 if (likely(!res))
595 return true;
596 if (unlikely(res < 0)) {
597 rcu_read_unlock();
598 mntput(bastard);
599 rcu_read_lock();
600 }
Al Viro48a066e2013-09-29 22:06:07 -0400601 return false;
602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/*
Al Viro474279d2013-10-01 16:11:26 -0400605 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400606 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Al Viro474279d2013-10-01 16:11:26 -0400608struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Al Viro38129a12014-03-20 21:10:51 -0400610 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400611 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Al Viro38129a12014-03-20 21:10:51 -0400613 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
615 return p;
616 return NULL;
617}
618
619/*
David Howellsf015f1262012-06-25 12:55:28 +0100620 * lookup_mnt - Return the first child mount mounted at path
621 *
622 * "First" means first mounted chronologically. If you create the
623 * following mounts:
624 *
625 * mount /dev/sda1 /mnt
626 * mount /dev/sda2 /mnt
627 * mount /dev/sda3 /mnt
628 *
629 * Then lookup_mnt() on the base /mnt dentry in the root mount will
630 * return successively the root dentry and vfsmount of /dev/sda1, then
631 * /dev/sda2, then /dev/sda3, then NULL.
632 *
633 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500634 */
Al Viroca71cf72016-11-20 19:45:28 -0500635struct vfsmount *lookup_mnt(const struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500636{
Al Viroc7105362011-11-24 18:22:03 -0500637 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400638 struct vfsmount *m;
639 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000640
Al Viro48a066e2013-09-29 22:06:07 -0400641 rcu_read_lock();
642 do {
643 seq = read_seqbegin(&mount_lock);
644 child_mnt = __lookup_mnt(path->mnt, path->dentry);
645 m = child_mnt ? &child_mnt->mnt : NULL;
646 } while (!legitimize_mnt(m, seq));
647 rcu_read_unlock();
648 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500649}
650
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200651static inline void lock_ns_list(struct mnt_namespace *ns)
652{
653 spin_lock(&ns->ns_lock);
654}
655
656static inline void unlock_ns_list(struct mnt_namespace *ns)
657{
658 spin_unlock(&ns->ns_lock);
659}
660
661static inline bool mnt_is_cursor(struct mount *mnt)
662{
663 return mnt->mnt.mnt_flags & MNT_CURSOR;
664}
665
Eric W. Biederman7af13642013-10-04 19:15:13 -0700666/*
667 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
668 * current mount namespace.
669 *
670 * The common case is dentries are not mountpoints at all and that
671 * test is handled inline. For the slow case when we are actually
672 * dealing with a mountpoint of some kind, walk through all of the
673 * mounts in the current mount namespace and test to see if the dentry
674 * is a mountpoint.
675 *
676 * The mount_hashtable is not usable in the context because we
677 * need to identify all mounts that may be in the current mount
678 * namespace not just a mount that happens to have some specified
679 * parent mount.
680 */
681bool __is_local_mountpoint(struct dentry *dentry)
682{
683 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
684 struct mount *mnt;
685 bool is_covered = false;
686
Eric W. Biederman7af13642013-10-04 19:15:13 -0700687 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200688 lock_ns_list(ns);
Eric W. Biederman7af13642013-10-04 19:15:13 -0700689 list_for_each_entry(mnt, &ns->list, mnt_list) {
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200690 if (mnt_is_cursor(mnt))
691 continue;
Eric W. Biederman7af13642013-10-04 19:15:13 -0700692 is_covered = (mnt->mnt_mountpoint == dentry);
693 if (is_covered)
694 break;
695 }
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200696 unlock_ns_list(ns);
Eric W. Biederman7af13642013-10-04 19:15:13 -0700697 up_read(&namespace_sem);
Nikolay Borisov5ad05cc2020-03-04 18:12:45 +0200698
Eric W. Biederman7af13642013-10-04 19:15:13 -0700699 return is_covered;
700}
701
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800702static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d171922013-03-15 10:53:28 -0400703{
Al Viro0818bf22014-02-28 13:46:44 -0500704 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d171922013-03-15 10:53:28 -0400705 struct mountpoint *mp;
706
Al Viro0818bf22014-02-28 13:46:44 -0500707 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d171922013-03-15 10:53:28 -0400708 if (mp->m_dentry == dentry) {
Al Viro84d171922013-03-15 10:53:28 -0400709 mp->m_count++;
710 return mp;
711 }
712 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800713 return NULL;
714}
715
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300716static struct mountpoint *get_mountpoint(struct dentry *dentry)
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800717{
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300718 struct mountpoint *mp, *new = NULL;
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800719 int ret;
Al Viro84d171922013-03-15 10:53:28 -0400720
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300721 if (d_mountpoint(dentry)) {
Benjamin Coddington1e9c75f2018-10-03 10:18:33 -0400722 /* might be worth a WARN_ON() */
723 if (d_unlinked(dentry))
724 return ERR_PTR(-ENOENT);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300725mountpoint:
726 read_seqlock_excl(&mount_lock);
727 mp = lookup_mountpoint(dentry);
728 read_sequnlock_excl(&mount_lock);
729 if (mp)
730 goto done;
Al Viro84d171922013-03-15 10:53:28 -0400731 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200732
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300733 if (!new)
734 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
735 if (!new)
736 return ERR_PTR(-ENOMEM);
737
738
739 /* Exactly one processes may set d_mounted */
740 ret = d_set_mounted(dentry);
741
742 /* Someone else set d_mounted? */
743 if (ret == -EBUSY)
744 goto mountpoint;
745
746 /* The dentry is not available as a mountpoint? */
747 mp = ERR_PTR(ret);
748 if (ret)
749 goto done;
750
751 /* Add the new mountpoint to the hash table */
752 read_seqlock_excl(&mount_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400753 new->m_dentry = dget(dentry);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300754 new->m_count = 1;
755 hlist_add_head(&new->m_hash, mp_hash(dentry));
756 INIT_HLIST_HEAD(&new->m_list);
757 read_sequnlock_excl(&mount_lock);
758
759 mp = new;
760 new = NULL;
761done:
762 kfree(new);
Al Viro84d171922013-03-15 10:53:28 -0400763 return mp;
764}
765
Al Viro4edbe132019-06-30 10:39:08 -0400766/*
767 * vfsmount lock must be held. Additionally, the caller is responsible
768 * for serializing calls for given disposal list.
769 */
770static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
Al Viro84d171922013-03-15 10:53:28 -0400771{
772 if (!--mp->m_count) {
773 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700774 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d171922013-03-15 10:53:28 -0400775 spin_lock(&dentry->d_lock);
776 dentry->d_flags &= ~DCACHE_MOUNTED;
777 spin_unlock(&dentry->d_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400778 dput_to_list(dentry, list);
Al Viro0818bf22014-02-28 13:46:44 -0500779 hlist_del(&mp->m_hash);
Al Viro84d171922013-03-15 10:53:28 -0400780 kfree(mp);
781 }
782}
783
Al Viro4edbe132019-06-30 10:39:08 -0400784/* called with namespace_lock and vfsmount lock */
785static void put_mountpoint(struct mountpoint *mp)
786{
787 __put_mountpoint(mp, &ex_mountpoints);
788}
789
Al Viro143c8c92011-11-25 00:46:35 -0500790static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800792 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Nick Piggin99b7db72010-08-18 04:37:39 +1000795/*
796 * vfsmount lock must be held for write
797 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800798static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500799{
800 if (ns) {
801 ns->event = ++event;
802 wake_up_interruptible(&ns->poll);
803 }
804}
805
Nick Piggin99b7db72010-08-18 04:37:39 +1000806/*
807 * vfsmount lock must be held for write
808 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800809static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500810{
811 if (ns && ns->event != event) {
812 ns->event = event;
813 wake_up_interruptible(&ns->poll);
814 }
815}
816
Nick Piggin99b7db72010-08-18 04:37:39 +1000817/*
818 * vfsmount lock must be held for write
819 */
Al Viroe4e59902019-06-29 12:58:42 -0400820static struct mountpoint *unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
Al Viroe4e59902019-06-29 12:58:42 -0400822 struct mountpoint *mp;
Al Viro0714a532011-11-24 22:19:58 -0500823 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500824 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500825 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400826 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700827 hlist_del_init(&mnt->mnt_mp_list);
Al Viroe4e59902019-06-29 12:58:42 -0400828 mp = mnt->mnt_mp;
Al Viro84d171922013-03-15 10:53:28 -0400829 mnt->mnt_mp = NULL;
Al Viroe4e59902019-06-29 12:58:42 -0400830 return mp;
Eric W. Biederman7bdb11de2014-12-29 13:03:41 -0600831}
832
833/*
834 * vfsmount lock must be held for write
835 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600836static void umount_mnt(struct mount *mnt)
837{
Al Viroe4e59902019-06-29 12:58:42 -0400838 put_mountpoint(unhash_mnt(mnt));
Eric W. Biederman6a46c572015-01-15 22:58:33 -0600839}
840
841/*
842 * vfsmount lock must be held for write
843 */
Al Viro84d171922013-03-15 10:53:28 -0400844void mnt_set_mountpoint(struct mount *mnt,
845 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -0500846 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500847{
Al Viro84d171922013-03-15 10:53:28 -0400848 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -0500849 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro4edbe132019-06-30 10:39:08 -0400850 child_mnt->mnt_mountpoint = mp->m_dentry;
Al Viro3a2393d2011-11-25 03:19:09 -0500851 child_mnt->mnt_parent = mnt;
Al Viro84d171922013-03-15 10:53:28 -0400852 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700853 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500854}
855
Eric W. Biederman1064f872017-01-20 18:28:35 +1300856static void __attach_mnt(struct mount *mnt, struct mount *parent)
857{
858 hlist_add_head_rcu(&mnt->mnt_hash,
859 m_hash(&parent->mnt, mnt->mnt_mountpoint));
860 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
861}
862
Nick Piggin99b7db72010-08-18 04:37:39 +1000863/*
864 * vfsmount lock must be held for write
865 */
Al Viro84d171922013-03-15 10:53:28 -0400866static void attach_mnt(struct mount *mnt,
867 struct mount *parent,
868 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Al Viro84d171922013-03-15 10:53:28 -0400870 mnt_set_mountpoint(parent, mp, mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +1300871 __attach_mnt(mnt, parent);
Ram Paib90fa9a2005-11-07 17:19:50 -0500872}
873
Eric W. Biederman1064f872017-01-20 18:28:35 +1300874void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
Al Viro12a5b522014-08-10 03:44:55 -0400875{
Eric W. Biederman1064f872017-01-20 18:28:35 +1300876 struct mountpoint *old_mp = mnt->mnt_mp;
Eric W. Biederman1064f872017-01-20 18:28:35 +1300877 struct mount *old_parent = mnt->mnt_parent;
878
879 list_del_init(&mnt->mnt_child);
880 hlist_del_init(&mnt->mnt_mp_list);
881 hlist_del_init_rcu(&mnt->mnt_hash);
882
883 attach_mnt(mnt, parent, mp);
884
885 put_mountpoint(old_mp);
Eric W. Biederman1064f872017-01-20 18:28:35 +1300886 mnt_add_count(old_parent, -1);
Al Viro12a5b522014-08-10 03:44:55 -0400887}
888
Ram Paib90fa9a2005-11-07 17:19:50 -0500889/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000890 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500891 */
Eric W. Biederman1064f872017-01-20 18:28:35 +1300892static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -0500893{
Al Viro0714a532011-11-24 22:19:58 -0500894 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -0500895 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -0500896 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -0500897 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500898
Al Viro0714a532011-11-24 22:19:58 -0500899 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500900
Al Viro1a4eeaf2011-11-25 02:19:55 -0500901 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -0400902 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -0500903 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -0500904
Ram Paib90fa9a2005-11-07 17:19:50 -0500905 list_splice(&head, n->list.prev);
906
Eric W. Biedermand2921682016-09-28 00:27:17 -0500907 n->mounts += n->pending_mounts;
908 n->pending_mounts = 0;
909
Eric W. Biederman1064f872017-01-20 18:28:35 +1300910 __attach_mnt(mnt, parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800911 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912}
913
Al Viro909b0a82011-11-25 03:06:56 -0500914static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915{
Al Viro6b41d532011-11-24 23:24:33 -0500916 struct list_head *next = p->mnt_mounts.next;
917 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -0500919 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -0500921 next = p->mnt_child.next;
922 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 break;
Al Viro0714a532011-11-24 22:19:58 -0500924 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 }
926 }
Al Viro6b41d532011-11-24 23:24:33 -0500927 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Al Viro315fc832011-11-24 18:57:30 -0500930static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -0500931{
Al Viro6b41d532011-11-24 23:24:33 -0500932 struct list_head *prev = p->mnt_mounts.prev;
933 while (prev != &p->mnt_mounts) {
934 p = list_entry(prev, struct mount, mnt_child);
935 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -0500936 }
937 return p;
938}
939
Al Viro8f2918892018-11-04 06:48:34 -0500940/**
941 * vfs_create_mount - Create a mount for a configured superblock
942 * @fc: The configuration context with the superblock attached
943 *
944 * Create a mount to an already configured superblock. If necessary, the
945 * caller should invoke vfs_get_tree() before calling this.
946 *
947 * Note that this does not attach the mount to anything.
948 */
949struct vfsmount *vfs_create_mount(struct fs_context *fc)
Al Viro9d412a42011-03-17 22:08:28 -0400950{
Al Virob105e272011-11-24 20:38:33 -0500951 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400952
Al Viro8f2918892018-11-04 06:48:34 -0500953 if (!fc->root)
954 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -0400955
Al Viro8f2918892018-11-04 06:48:34 -0500956 mnt = alloc_vfsmnt(fc->source ?: "none");
Al Viro9d412a42011-03-17 22:08:28 -0400957 if (!mnt)
958 return ERR_PTR(-ENOMEM);
959
Al Viro8f2918892018-11-04 06:48:34 -0500960 if (fc->sb_flags & SB_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -0500961 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -0400962
Al Viro8f2918892018-11-04 06:48:34 -0500963 atomic_inc(&fc->root->d_sb->s_active);
964 mnt->mnt.mnt_sb = fc->root->d_sb;
965 mnt->mnt.mnt_root = dget(fc->root);
966 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
967 mnt->mnt_parent = mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400968
Al Viro719ea2f2013-09-29 11:24:49 -0400969 lock_mount_hash();
Al Viro8f2918892018-11-04 06:48:34 -0500970 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -0400971 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -0500972 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -0400973}
Al Viro8f2918892018-11-04 06:48:34 -0500974EXPORT_SYMBOL(vfs_create_mount);
975
976struct vfsmount *fc_mount(struct fs_context *fc)
977{
978 int err = vfs_get_tree(fc);
979 if (!err) {
980 up_write(&fc->root->d_sb->s_umount);
981 return vfs_create_mount(fc);
982 }
983 return ERR_PTR(err);
984}
985EXPORT_SYMBOL(fc_mount);
986
David Howells9bc61ab2018-11-04 03:19:03 -0500987struct vfsmount *vfs_kern_mount(struct file_system_type *type,
988 int flags, const char *name,
989 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990{
David Howells9bc61ab2018-11-04 03:19:03 -0500991 struct fs_context *fc;
Al Viro8f2918892018-11-04 06:48:34 -0500992 struct vfsmount *mnt;
David Howells9bc61ab2018-11-04 03:19:03 -0500993 int ret = 0;
Al Viro9d412a42011-03-17 22:08:28 -0400994
995 if (!type)
David Howells3e1aeb02018-11-01 23:07:25 +0000996 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -0400997
David Howells9bc61ab2018-11-04 03:19:03 -0500998 fc = fs_context_for_mount(type, flags);
999 if (IS_ERR(fc))
1000 return ERR_CAST(fc);
1001
David Howells3e1aeb02018-11-01 23:07:25 +00001002 if (name)
1003 ret = vfs_parse_fs_string(fc, "source",
1004 name, strlen(name));
David Howells9bc61ab2018-11-04 03:19:03 -05001005 if (!ret)
1006 ret = parse_monolithic_mount_data(fc, data);
1007 if (!ret)
Al Viro8f2918892018-11-04 06:48:34 -05001008 mnt = fc_mount(fc);
1009 else
1010 mnt = ERR_PTR(ret);
Al Viro9d412a42011-03-17 22:08:28 -04001011
David Howells9bc61ab2018-11-04 03:19:03 -05001012 put_fs_context(fc);
Al Viro8f2918892018-11-04 06:48:34 -05001013 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001014}
1015EXPORT_SYMBOL_GPL(vfs_kern_mount);
1016
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001017struct vfsmount *
1018vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1019 const char *name, void *data)
1020{
1021 /* Until it is worked out how to pass the user namespace
1022 * through from the parent mount to the submount don't support
1023 * unprivileged mounts with submounts.
1024 */
1025 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1026 return ERR_PTR(-EPERM);
1027
David Howellse462ec52017-07-17 08:45:35 +01001028 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001029}
1030EXPORT_SYMBOL_GPL(vfs_submount);
1031
Al Viro87129cc2011-11-24 21:24:27 -05001032static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -05001033 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Al Viro87129cc2011-11-24 21:24:27 -05001035 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +01001036 struct mount *mnt;
1037 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
David Howellsbe34d1a2012-06-25 12:55:18 +01001039 mnt = alloc_vfsmnt(old->mnt_devname);
1040 if (!mnt)
1041 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001042
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001043 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +01001044 mnt->mnt_group_id = 0; /* not a peer of original */
1045 else
1046 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001047
David Howellsbe34d1a2012-06-25 12:55:18 +01001048 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1049 err = mnt_alloc_group_id(mnt);
1050 if (err)
1051 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001053
Al Viro16a34ad2018-04-19 22:03:08 -04001054 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1055 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001056
David Howellsbe34d1a2012-06-25 12:55:18 +01001057 atomic_inc(&sb->s_active);
1058 mnt->mnt.mnt_sb = sb;
1059 mnt->mnt.mnt_root = dget(root);
1060 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1061 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001062 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001063 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001064 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001065
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001066 if ((flag & CL_SLAVE) ||
1067 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001068 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1069 mnt->mnt_master = old;
1070 CLEAR_MNT_SHARED(mnt);
1071 } else if (!(flag & CL_PRIVATE)) {
1072 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1073 list_add(&mnt->mnt_share, &old->mnt_share);
1074 if (IS_MNT_SLAVE(old))
1075 list_add(&mnt->mnt_slave, &old->mnt_slave);
1076 mnt->mnt_master = old->mnt_master;
Al Viro5235d442016-11-20 19:33:09 -05001077 } else {
1078 CLEAR_MNT_SHARED(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001079 }
1080 if (flag & CL_MAKE_SHARED)
1081 set_mnt_shared(mnt);
1082
1083 /* stick the duplicate mount on the same expiry list
1084 * as the original if that was on one */
1085 if (flag & CL_EXPIRE) {
1086 if (!list_empty(&old->mnt_expire))
1087 list_add(&mnt->mnt_expire, &old->mnt_expire);
1088 }
1089
Al Virocb338d02011-11-24 20:55:08 -05001090 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001091
1092 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001093 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001094 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001095 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
Al Viro9ea459e12014-08-08 13:08:20 -04001098static void cleanup_mnt(struct mount *mnt)
1099{
Al Viro56cbb4292019-07-04 16:57:51 -04001100 struct hlist_node *p;
1101 struct mount *m;
Al Viro9ea459e12014-08-08 13:08:20 -04001102 /*
Al Viro56cbb4292019-07-04 16:57:51 -04001103 * The warning here probably indicates that somebody messed
1104 * up a mnt_want/drop_write() pair. If this happens, the
1105 * filesystem was probably unable to make r/w->r/o transitions.
Al Viro9ea459e12014-08-08 13:08:20 -04001106 * The locking used to deal with mnt_count decrement provides barriers,
1107 * so mnt_get_writers() below is safe.
1108 */
1109 WARN_ON(mnt_get_writers(mnt));
1110 if (unlikely(mnt->mnt_pins.first))
1111 mnt_pin_kill(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001112 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1113 hlist_del(&m->mnt_umount);
1114 mntput(&m->mnt);
1115 }
Al Viro9ea459e12014-08-08 13:08:20 -04001116 fsnotify_vfsmount_delete(&mnt->mnt);
1117 dput(mnt->mnt.mnt_root);
1118 deactivate_super(mnt->mnt.mnt_sb);
1119 mnt_free_id(mnt);
1120 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1121}
1122
1123static void __cleanup_mnt(struct rcu_head *head)
1124{
1125 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1126}
1127
1128static LLIST_HEAD(delayed_mntput_list);
1129static void delayed_mntput(struct work_struct *unused)
1130{
1131 struct llist_node *node = llist_del_all(&delayed_mntput_list);
Byungchul Park29785732017-08-07 17:44:45 +09001132 struct mount *m, *t;
Al Viro9ea459e12014-08-08 13:08:20 -04001133
Byungchul Park29785732017-08-07 17:44:45 +09001134 llist_for_each_entry_safe(m, t, node, mnt_llist)
1135 cleanup_mnt(m);
Al Viro9ea459e12014-08-08 13:08:20 -04001136}
1137static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1138
Al Viro900148d2011-11-25 00:33:11 -05001139static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001140{
Al Viro4edbe132019-06-30 10:39:08 -04001141 LIST_HEAD(list);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001142 int count;
Al Viro4edbe132019-06-30 10:39:08 -04001143
Al Viro48a066e2013-09-29 22:06:07 -04001144 rcu_read_lock();
Al Viro9ea0a462018-08-09 17:21:17 -04001145 if (likely(READ_ONCE(mnt->mnt_ns))) {
1146 /*
1147 * Since we don't do lock_mount_hash() here,
1148 * ->mnt_ns can change under us. However, if it's
1149 * non-NULL, then there's a reference that won't
1150 * be dropped until after an RCU delay done after
1151 * turning ->mnt_ns NULL. So if we observe it
1152 * non-NULL under rcu_read_lock(), the reference
1153 * we are dropping is not the final one.
1154 */
1155 mnt_add_count(mnt, -1);
Al Viro48a066e2013-09-29 22:06:07 -04001156 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001157 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001158 }
Al Viro719ea2f2013-09-29 11:24:49 -04001159 lock_mount_hash();
Al Viro119e1ef2018-08-09 17:51:32 -04001160 /*
1161 * make sure that if __legitimize_mnt() has not seen us grab
1162 * mount_lock, we'll see their refcount increment here.
1163 */
1164 smp_mb();
Al Viro9ea0a462018-08-09 17:21:17 -04001165 mnt_add_count(mnt, -1);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001166 count = mnt_get_count(mnt);
1167 if (count != 0) {
1168 WARN_ON(count < 0);
Al Viro48a066e2013-09-29 22:06:07 -04001169 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001170 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001171 return;
1172 }
Al Viro48a066e2013-09-29 22:06:07 -04001173 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1174 rcu_read_unlock();
1175 unlock_mount_hash();
1176 return;
1177 }
1178 mnt->mnt.mnt_flags |= MNT_DOOMED;
1179 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301180
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001181 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001182
1183 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1184 struct mount *p, *tmp;
1185 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
Al Viro4edbe132019-06-30 10:39:08 -04001186 __put_mountpoint(unhash_mnt(p), &list);
Al Viro56cbb4292019-07-04 16:57:51 -04001187 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001188 }
1189 }
Al Viro719ea2f2013-09-29 11:24:49 -04001190 unlock_mount_hash();
Al Viro4edbe132019-06-30 10:39:08 -04001191 shrink_dentry_list(&list);
Al Viro649a7952013-09-28 12:41:25 -04001192
Al Viro9ea459e12014-08-08 13:08:20 -04001193 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1194 struct task_struct *task = current;
1195 if (likely(!(task->flags & PF_KTHREAD))) {
1196 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
Jens Axboe91989c72020-10-16 09:02:26 -06001197 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
Al Viro9ea459e12014-08-08 13:08:20 -04001198 return;
1199 }
1200 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1201 schedule_delayed_work(&delayed_mntput_work, 1);
1202 return;
1203 }
1204 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001205}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001206
1207void mntput(struct vfsmount *mnt)
1208{
1209 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001210 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001211 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001212 if (unlikely(m->mnt_expiry_mark))
1213 m->mnt_expiry_mark = 0;
1214 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001215 }
1216}
1217EXPORT_SYMBOL(mntput);
1218
1219struct vfsmount *mntget(struct vfsmount *mnt)
1220{
1221 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001222 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001223 return mnt;
1224}
1225EXPORT_SYMBOL(mntget);
1226
Ian Kentc6609c02016-11-24 08:03:41 +11001227/* path_is_mountpoint() - Check if path is a mount in the current
1228 * namespace.
1229 *
1230 * d_mountpoint() can only be used reliably to establish if a dentry is
1231 * not mounted in any namespace and that common case is handled inline.
1232 * d_mountpoint() isn't aware of the possibility there may be multiple
1233 * mounts using a given dentry in a different namespace. This function
1234 * checks if the passed in path is a mountpoint rather than the dentry
1235 * alone.
1236 */
1237bool path_is_mountpoint(const struct path *path)
1238{
1239 unsigned seq;
1240 bool res;
1241
1242 if (!d_mountpoint(path->dentry))
1243 return false;
1244
1245 rcu_read_lock();
1246 do {
1247 seq = read_seqbegin(&mount_lock);
1248 res = __path_is_mountpoint(path);
1249 } while (read_seqretry(&mount_lock, seq));
1250 rcu_read_unlock();
1251
1252 return res;
1253}
1254EXPORT_SYMBOL(path_is_mountpoint);
1255
Al Viroca71cf72016-11-20 19:45:28 -05001256struct vfsmount *mnt_clone_internal(const struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001257{
Al Viro3064c352014-08-07 09:12:31 -04001258 struct mount *p;
1259 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1260 if (IS_ERR(p))
1261 return ERR_CAST(p);
1262 p->mnt.mnt_flags |= MNT_INTERNAL;
1263 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001264}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001266#ifdef CONFIG_PROC_FS
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001267static struct mount *mnt_list_next(struct mnt_namespace *ns,
1268 struct list_head *p)
1269{
1270 struct mount *mnt, *ret = NULL;
1271
1272 lock_ns_list(ns);
1273 list_for_each_continue(p, &ns->list) {
1274 mnt = list_entry(p, typeof(*mnt), mnt_list);
1275 if (!mnt_is_cursor(mnt)) {
1276 ret = mnt;
1277 break;
1278 }
1279 }
1280 unlock_ns_list(ns);
1281
1282 return ret;
1283}
1284
Al Viro0226f492011-12-06 12:21:54 -05001285/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286static void *m_start(struct seq_file *m, loff_t *pos)
1287{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001288 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001289 struct list_head *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Ram Pai390c6842005-11-07 17:17:51 -05001291 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001292 if (!*pos) {
1293 prev = &p->ns->list;
1294 } else {
1295 prev = &p->cursor.mnt_list;
1296
1297 /* Read after we'd reached the end? */
1298 if (list_empty(prev))
1299 return NULL;
Al Viroc7999c32014-02-27 14:40:10 -05001300 }
1301
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001302 return mnt_list_next(p->ns, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303}
1304
1305static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1306{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001307 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001308 struct mount *mnt = v;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001309
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001310 ++*pos;
1311 return mnt_list_next(p->ns, &mnt->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312}
1313
1314static void m_stop(struct seq_file *m, void *v)
1315{
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001316 struct proc_mounts *p = m->private;
1317 struct mount *mnt = v;
1318
1319 lock_ns_list(p->ns);
1320 if (mnt)
1321 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1322 else
1323 list_del_init(&p->cursor.mnt_list);
1324 unlock_ns_list(p->ns);
Ram Pai390c6842005-11-07 17:17:51 -05001325 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326}
1327
Al Viro0226f492011-12-06 12:21:54 -05001328static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001329{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001330 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001331 struct mount *r = v;
Al Viro0226f492011-12-06 12:21:54 -05001332 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
1334
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001335const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 .start = m_start,
1337 .next = m_next,
1338 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001339 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001340};
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001341
1342void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1343{
1344 down_read(&namespace_sem);
1345 lock_ns_list(ns);
1346 list_del(&cursor->mnt_list);
1347 unlock_ns_list(ns);
1348 up_read(&namespace_sem);
1349}
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001350#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352/**
1353 * may_umount_tree - check if a mount tree is busy
1354 * @mnt: root of mount tree
1355 *
1356 * This is called to check if a tree of mounts has any
1357 * open files, pwds, chroots or sub mounts that are
1358 * busy.
1359 */
Al Viro909b0a82011-11-25 03:06:56 -05001360int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361{
Al Viro909b0a82011-11-25 03:06:56 -05001362 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001363 int actual_refs = 0;
1364 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001365 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001366 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Nick Pigginb3e19d92011-01-07 17:50:11 +11001368 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001369 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001370 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001371 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
Al Viro719ea2f2013-09-29 11:24:49 -04001374 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
1376 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001377 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Ian Kente3474a82006-03-27 01:14:51 -08001379 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
1382EXPORT_SYMBOL(may_umount_tree);
1383
1384/**
1385 * may_umount - check if a mount point is busy
1386 * @mnt: root of mount
1387 *
1388 * This is called to check if a mount point has any
1389 * open files, pwds, chroots or sub mounts. If the
1390 * mount has sub mounts this will return busy
1391 * regardless of whether the sub mounts are busy.
1392 *
1393 * Doesn't take quota and stuff into account. IOW, in some cases it will
1394 * give false negatives. The main reason why it's here is that we need
1395 * a non-destructive way to look for easily umountable filesystems.
1396 */
1397int may_umount(struct vfsmount *mnt)
1398{
Ian Kente3474a82006-03-27 01:14:51 -08001399 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001400 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001401 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001402 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001403 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001404 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001405 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001406 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
1409EXPORT_SYMBOL(may_umount);
1410
Al Viro97216be2013-03-16 15:12:40 -04001411static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001413 struct hlist_head head;
Al Viro56cbb4292019-07-04 16:57:51 -04001414 struct hlist_node *p;
1415 struct mount *m;
Al Viro4edbe132019-06-30 10:39:08 -04001416 LIST_HEAD(list);
Al Viro97216be2013-03-16 15:12:40 -04001417
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001418 hlist_move_list(&unmounted, &head);
Al Viro4edbe132019-06-30 10:39:08 -04001419 list_splice_init(&ex_mountpoints, &list);
Al Viro97216be2013-03-16 15:12:40 -04001420
Al Viro97216be2013-03-16 15:12:40 -04001421 up_write(&namespace_sem);
1422
Al Viro4edbe132019-06-30 10:39:08 -04001423 shrink_dentry_list(&list);
1424
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001425 if (likely(hlist_empty(&head)))
1426 return;
1427
NeilBrown22cb7402018-11-30 10:33:18 +11001428 synchronize_rcu_expedited();
Al Viro48a066e2013-09-29 22:06:07 -04001429
Al Viro56cbb4292019-07-04 16:57:51 -04001430 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1431 hlist_del(&m->mnt_umount);
1432 mntput(&m->mnt);
1433 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001434}
1435
Al Viro97216be2013-03-16 15:12:40 -04001436static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001437{
Al Viro97216be2013-03-16 15:12:40 -04001438 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001439}
1440
Eric W. Biedermane819f152014-12-24 07:20:01 -06001441enum umount_tree_flags {
1442 UMOUNT_SYNC = 1,
1443 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001444 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001445};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001446
1447static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1448{
1449 /* Leaving mounts connected is only valid for lazy umounts */
1450 if (how & UMOUNT_SYNC)
1451 return true;
1452
1453 /* A mount without a parent has nothing to be connected to */
1454 if (!mnt_has_parent(mnt))
1455 return true;
1456
1457 /* Because the reference counting rules change when mounts are
1458 * unmounted and connected, umounted mounts may not be
1459 * connected to mounted mounts.
1460 */
1461 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1462 return true;
1463
1464 /* Has it been requested that the mount remain connected? */
1465 if (how & UMOUNT_CONNECTED)
1466 return false;
1467
1468 /* Is the mount locked such that it needs to remain connected? */
1469 if (IS_MNT_LOCKED(mnt))
1470 return false;
1471
1472 /* By default disconnect the mount */
1473 return true;
1474}
1475
Nick Piggin99b7db72010-08-18 04:37:39 +10001476/*
Al Viro48a066e2013-09-29 22:06:07 -04001477 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001478 * namespace_sem must be held for write
1479 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001480static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001481{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001482 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001483 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001484
Eric W. Biederman5d884572015-01-03 05:39:35 -06001485 if (how & UMOUNT_PROPAGATE)
1486 propagate_mount_unlock(mnt);
1487
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001488 /* Gather the mounts to umount */
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001489 for (p = mnt; p; p = next_mnt(p, mnt)) {
1490 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001491 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001492 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001493
Eric W. Biederman411a9382014-12-22 19:12:07 -06001494 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001495 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001496 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001497 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001498
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001499 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001500 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001501 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001502
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001503 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001504 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001505 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001506 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001507 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001508 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001509 ns = p->mnt_ns;
1510 if (ns) {
1511 ns->mounts--;
1512 __touch_mnt_namespace(ns);
1513 }
Al Viro143c8c92011-11-25 00:46:35 -05001514 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001515 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001516 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001517
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001518 disconnect = disconnect_mount(p, how);
Al Viro676da582011-11-24 21:47:05 -05001519 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001520 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001521 if (!disconnect) {
1522 /* Don't forget about p */
1523 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1524 } else {
1525 umount_mnt(p);
1526 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001527 }
Al Viro0f0afb12011-11-24 20:43:10 -05001528 change_mnt_propagation(p, MS_PRIVATE);
Al Viro19a1c402019-07-24 12:45:46 -04001529 if (disconnect)
1530 hlist_add_head(&p->mnt_umount, &unmounted);
Al Viro38129a12014-03-20 21:10:51 -04001531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532}
1533
Al Virob54b9be2013-03-16 14:39:34 -04001534static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001535
David Howells8d0347f2018-11-04 09:28:36 -05001536static int do_umount_root(struct super_block *sb)
1537{
1538 int ret = 0;
1539
1540 down_write(&sb->s_umount);
1541 if (!sb_rdonly(sb)) {
1542 struct fs_context *fc;
1543
1544 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1545 SB_RDONLY);
1546 if (IS_ERR(fc)) {
1547 ret = PTR_ERR(fc);
1548 } else {
1549 ret = parse_monolithic_mount_data(fc, NULL);
1550 if (!ret)
1551 ret = reconfigure_super(fc);
1552 put_fs_context(fc);
1553 }
1554 }
1555 up_write(&sb->s_umount);
1556 return ret;
1557}
1558
Al Viro1ab59732011-11-24 21:35:16 -05001559static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560{
Al Viro1ab59732011-11-24 21:35:16 -05001561 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 int retval;
1563
Al Viro1ab59732011-11-24 21:35:16 -05001564 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 if (retval)
1566 return retval;
1567
1568 /*
1569 * Allow userspace to request a mountpoint be expired rather than
1570 * unmounting unconditionally. Unmount only happens if:
1571 * (1) the mark is already set (the mark is cleared by mntput())
1572 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1573 */
1574 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001575 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 flags & (MNT_FORCE | MNT_DETACH))
1577 return -EINVAL;
1578
Nick Pigginb3e19d92011-01-07 17:50:11 +11001579 /*
1580 * probably don't strictly need the lock here if we examined
1581 * all race cases, but it's a slowpath.
1582 */
Al Viro719ea2f2013-09-29 11:24:49 -04001583 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001584 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001585 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001587 }
Al Viro719ea2f2013-09-29 11:24:49 -04001588 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
Al Viro863d6842011-11-25 00:57:42 -05001590 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 return -EAGAIN;
1592 }
1593
1594 /*
1595 * If we may have to abort operations to get out of this
1596 * mount, and they will themselves hold resources we must
1597 * allow the fs to do things. In the Unix tradition of
1598 * 'Gee thats tricky lets do it in userspace' the umount_begin
1599 * might fail to complete on the first run through as other tasks
1600 * must return, and the like. Thats for the mount program to worry
1601 * about for the moment.
1602 */
1603
Al Viro42faad92008-04-24 07:21:56 -04001604 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001605 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001606 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 /*
1609 * No sense to grab the lock for this test, but test itself looks
1610 * somewhat bogus. Suggestions for better replacement?
1611 * Ho-hum... In principle, we might treat that as umount + switch
1612 * to rootfs. GC would eventually take care of the old vfsmount.
1613 * Actually it makes sense, especially if rootfs would contain a
1614 * /reboot - static binary that would close all descriptors and
1615 * call reboot(9). Then init(8) could umount root and exec /reboot.
1616 */
Al Viro1ab59732011-11-24 21:35:16 -05001617 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 /*
1619 * Special case for "unmounting" root ...
1620 * we just try to remount it readonly.
1621 */
Eric W. Biedermanbc6155d2017-09-18 17:58:08 -05001622 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001623 return -EPERM;
David Howells8d0347f2018-11-04 09:28:36 -05001624 return do_umount_root(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
1626
Al Viro97216be2013-03-16 15:12:40 -04001627 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001628 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001630 /* Recheck MNT_LOCKED with the locks held */
1631 retval = -EINVAL;
1632 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1633 goto out;
1634
1635 event++;
Al Viro48a066e2013-09-29 22:06:07 -04001636 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001637 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001638 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001640 } else {
1641 shrink_submounts(mnt);
1642 retval = -EBUSY;
1643 if (!propagate_mount_busy(mnt, 2)) {
1644 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001645 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001646 retval = 0;
1647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001649out:
Al Viro719ea2f2013-09-29 11:24:49 -04001650 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001651 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 return retval;
1653}
1654
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001655/*
1656 * __detach_mounts - lazily unmount all mounts on the specified dentry
1657 *
1658 * During unlink, rmdir, and d_drop it is possible to loose the path
1659 * to an existing mountpoint, and wind up leaking the mount.
1660 * detach_mounts allows lazily unmounting those mounts instead of
1661 * leaking them.
1662 *
1663 * The caller may hold dentry->d_inode->i_mutex.
1664 */
1665void __detach_mounts(struct dentry *dentry)
1666{
1667 struct mountpoint *mp;
1668 struct mount *mnt;
1669
1670 namespace_lock();
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001671 lock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001672 mp = lookup_mountpoint(dentry);
Al Viroadc9b5c2019-06-29 12:06:51 -04001673 if (!mp)
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001674 goto out_unlock;
1675
Andrey Ulanove06b9332016-04-15 14:24:41 -07001676 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001677 while (!hlist_empty(&mp->m_list)) {
1678 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001679 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc82015-07-17 14:54:27 -05001680 umount_mnt(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001681 hlist_add_head(&mnt->mnt_umount, &unmounted);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001682 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001683 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001684 }
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001685 put_mountpoint(mp);
1686out_unlock:
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001687 unlock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001688 namespace_unlock();
1689}
1690
David Howellsdd111b32017-07-04 17:25:09 +01001691/*
Al Viro9b40bc92013-02-22 22:45:42 -05001692 * Is the caller allowed to modify his namespace?
1693 */
1694static inline bool may_mount(void)
1695{
1696 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1697}
1698
Jeff Laytondf2474a2019-08-15 15:21:17 -04001699#ifdef CONFIG_MANDATORY_FILE_LOCKING
Jeff Layton9e8925b2015-11-16 09:49:34 -05001700static inline bool may_mandlock(void)
1701{
Eric W. Biederman95ace752015-11-11 17:22:33 -06001702 return capable(CAP_SYS_ADMIN);
Jeff Layton9e8925b2015-11-16 09:49:34 -05001703}
Jeff Laytondf2474a2019-08-15 15:21:17 -04001704#else
1705static inline bool may_mandlock(void)
1706{
1707 pr_warn("VFS: \"mand\" mount option not supported");
1708 return false;
1709}
1710#endif
Jeff Layton9e8925b2015-11-16 09:49:34 -05001711
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001712static int can_umount(const struct path *path, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001714 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
Al Viro9b40bc92013-02-22 22:45:42 -05001716 if (!may_mount())
1717 return -EPERM;
Christoph Hellwig41525f52020-07-21 10:54:34 +02001718 if (path->dentry != path->mnt->mnt_root)
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001719 return -EINVAL;
Al Viro143c8c92011-11-25 00:46:35 -05001720 if (!check_mnt(mnt))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001721 return -EINVAL;
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001722 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001723 return -EINVAL;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001724 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001725 return -EPERM;
1726 return 0;
1727}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Al Viroa0a6df92021-01-04 15:25:34 -05001729// caller is responsible for flags being sane
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001730int path_umount(struct path *path, int flags)
1731{
1732 struct mount *mnt = real_mount(path->mnt);
1733 int ret;
1734
1735 ret = can_umount(path, flags);
1736 if (!ret)
1737 ret = do_umount(mnt, flags);
1738
Jan Blunck429731b2008-02-14 19:34:31 -08001739 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Christoph Hellwig41525f52020-07-21 10:54:34 +02001740 dput(path->dentry);
Al Viro900148d2011-11-25 00:33:11 -05001741 mntput_no_expire(mnt);
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001742 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
Christoph Hellwig09267de2020-07-23 08:23:08 +02001745static int ksys_umount(char __user *name, int flags)
Christoph Hellwig41525f52020-07-21 10:54:34 +02001746{
1747 int lookup_flags = LOOKUP_MOUNTPOINT;
1748 struct path path;
1749 int ret;
1750
Al Viroa0a6df92021-01-04 15:25:34 -05001751 // basic validity checks done first
1752 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1753 return -EINVAL;
1754
Christoph Hellwig41525f52020-07-21 10:54:34 +02001755 if (!(flags & UMOUNT_NOFOLLOW))
1756 lookup_flags |= LOOKUP_FOLLOW;
1757 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1758 if (ret)
1759 return ret;
1760 return path_umount(&path, flags);
1761}
1762
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001763SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1764{
1765 return ksys_umount(name, flags);
1766}
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1769
1770/*
Ram Paib58fed82005-11-07 17:16:09 -05001771 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001773SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774{
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001775 return ksys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
1778#endif
1779
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001780static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001781{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001782 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001783 return dentry->d_op == &ns_dentry_operations &&
1784 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001785}
1786
Eric Biggers213921f2020-01-04 12:59:55 -08001787static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
Al Viro58be28252014-11-01 00:00:23 -04001788{
1789 return container_of(ns, struct mnt_namespace, ns);
1790}
1791
Christian Brauner303cc572020-05-05 16:04:31 +02001792struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1793{
1794 return &mnt->ns;
1795}
1796
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001797static bool mnt_ns_loop(struct dentry *dentry)
1798{
1799 /* Could bind mounting the mount namespace inode cause a
1800 * mount namespace loop?
1801 */
1802 struct mnt_namespace *mnt_ns;
1803 if (!is_mnt_ns_file(dentry))
1804 return false;
1805
Al Virof77c8012014-11-01 03:13:17 -04001806 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001807 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1808}
1809
Al Viro87129cc2011-11-24 21:24:27 -05001810struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001811 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
Al Viro84d171922013-03-15 10:53:28 -04001813 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001815 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1816 return ERR_PTR(-EINVAL);
1817
1818 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001819 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001820
Ram Pai36341f62005-11-07 17:17:22 -05001821 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001822 if (IS_ERR(q))
1823 return q;
1824
Al Viroa73324d2011-11-24 22:25:07 -05001825 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001828 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001829 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001830 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 continue;
1832
Al Viro909b0a82011-11-25 03:06:56 -05001833 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001834 if (!(flag & CL_COPY_UNBINDABLE) &&
1835 IS_MNT_UNBINDABLE(s)) {
Eric W. Biedermandf7342b2018-10-25 09:04:18 -05001836 if (s->mnt.mnt_flags & MNT_LOCKED) {
1837 /* Both unbindable and locked. */
1838 q = ERR_PTR(-EPERM);
1839 goto out;
1840 } else {
1841 s = skip_mnt_tree(s);
1842 continue;
1843 }
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001844 }
1845 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1846 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001847 s = skip_mnt_tree(s);
1848 continue;
1849 }
Al Viro0714a532011-11-24 22:19:58 -05001850 while (p != s->mnt_parent) {
1851 p = p->mnt_parent;
1852 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 }
Al Viro87129cc2011-11-24 21:24:27 -05001854 p = s;
Al Viro84d171922013-03-15 10:53:28 -04001855 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05001856 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001857 if (IS_ERR(q))
1858 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04001859 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05001860 list_add_tail(&q->mnt_list, &res->mnt_list);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001861 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04001862 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 }
1864 }
1865 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01001866out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04001868 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06001869 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04001870 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001872 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873}
1874
David Howellsbe34d1a2012-06-25 12:55:18 +01001875/* Caller should check returned pointer for errors */
1876
Al Viroca71cf72016-11-20 19:45:28 -05001877struct vfsmount *collect_mounts(const struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001878{
Al Virocb338d02011-11-24 20:55:08 -05001879 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04001880 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06001881 if (!check_mnt(real_mount(path->mnt)))
1882 tree = ERR_PTR(-EINVAL);
1883 else
1884 tree = copy_tree(real_mount(path->mnt), path->dentry,
1885 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04001886 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01001887 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03001888 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01001889 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04001890}
1891
Al Viroa07b2002018-11-05 17:40:30 +00001892static void free_mnt_ns(struct mnt_namespace *);
1893static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1894
1895void dissolve_on_fput(struct vfsmount *mnt)
1896{
1897 struct mnt_namespace *ns;
1898 namespace_lock();
1899 lock_mount_hash();
1900 ns = real_mount(mnt)->mnt_ns;
David Howells44dfd842018-11-05 17:40:31 +00001901 if (ns) {
1902 if (is_anon_ns(ns))
1903 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1904 else
1905 ns = NULL;
1906 }
Al Viroa07b2002018-11-05 17:40:30 +00001907 unlock_mount_hash();
1908 namespace_unlock();
David Howells44dfd842018-11-05 17:40:31 +00001909 if (ns)
1910 free_mnt_ns(ns);
Al Viroa07b2002018-11-05 17:40:30 +00001911}
1912
Al Viro8aec0802007-06-07 12:20:32 -04001913void drop_collected_mounts(struct vfsmount *mnt)
1914{
Al Viro97216be2013-03-16 15:12:40 -04001915 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001916 lock_mount_hash();
Eric W. Biederman9c8e0a12018-10-25 12:05:11 -05001917 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04001918 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04001919 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04001920}
1921
Miklos Szeredic771d682014-10-24 00:14:36 +02001922/**
1923 * clone_private_mount - create a private clone of a path
1924 *
1925 * This creates a new vfsmount, which will be the clone of @path. The new will
1926 * not be attached anywhere in the namespace and will be private (i.e. changes
1927 * to the originating mount won't be propagated into this).
1928 *
1929 * Release with mntput().
1930 */
Al Viroca71cf72016-11-20 19:45:28 -05001931struct vfsmount *clone_private_mount(const struct path *path)
Miklos Szeredic771d682014-10-24 00:14:36 +02001932{
1933 struct mount *old_mnt = real_mount(path->mnt);
1934 struct mount *new_mnt;
1935
1936 if (IS_MNT_UNBINDABLE(old_mnt))
1937 return ERR_PTR(-EINVAL);
1938
Miklos Szeredic771d682014-10-24 00:14:36 +02001939 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
Miklos Szeredic771d682014-10-24 00:14:36 +02001940 if (IS_ERR(new_mnt))
1941 return ERR_CAST(new_mnt);
1942
Miklos Szeredidf820f82020-06-04 10:48:19 +02001943 /* Longterm mount to be removed by kern_unmount*() */
1944 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1945
Miklos Szeredic771d682014-10-24 00:14:36 +02001946 return &new_mnt->mnt;
1947}
1948EXPORT_SYMBOL_GPL(clone_private_mount);
1949
Al Viro1f707132010-01-30 22:51:25 -05001950int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1951 struct vfsmount *root)
1952{
Al Viro1a4eeaf2011-11-25 02:19:55 -05001953 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05001954 int res = f(root, arg);
1955 if (res)
1956 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05001957 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1958 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05001959 if (res)
1960 return res;
1961 }
1962 return 0;
1963}
1964
Al Viro3bd045c2019-01-30 13:15:45 -05001965static void lock_mnt_tree(struct mount *mnt)
1966{
1967 struct mount *p;
1968
1969 for (p = mnt; p; p = next_mnt(p, mnt)) {
1970 int flags = p->mnt.mnt_flags;
1971 /* Don't allow unprivileged users to change mount flags */
1972 flags |= MNT_LOCK_ATIME;
1973
1974 if (flags & MNT_READONLY)
1975 flags |= MNT_LOCK_READONLY;
1976
1977 if (flags & MNT_NODEV)
1978 flags |= MNT_LOCK_NODEV;
1979
1980 if (flags & MNT_NOSUID)
1981 flags |= MNT_LOCK_NOSUID;
1982
1983 if (flags & MNT_NOEXEC)
1984 flags |= MNT_LOCK_NOEXEC;
1985 /* Don't allow unprivileged users to reveal what is under a mount */
1986 if (list_empty(&p->mnt_expire))
1987 flags |= MNT_LOCKED;
1988 p->mnt.mnt_flags = flags;
1989 }
1990}
1991
Al Viro4b8b21f2011-11-24 19:54:23 -05001992static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001993{
Al Viro315fc832011-11-24 18:57:30 -05001994 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001995
Al Viro909b0a82011-11-25 03:06:56 -05001996 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05001997 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05001998 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001999 }
2000}
2001
Al Viro4b8b21f2011-11-24 19:54:23 -05002002static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002003{
Al Viro315fc832011-11-24 18:57:30 -05002004 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002005
Al Viro909b0a82011-11-25 03:06:56 -05002006 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05002007 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002008 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002009 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002010 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002011 return err;
2012 }
2013 }
2014 }
2015
2016 return 0;
2017}
2018
Eric W. Biedermand2921682016-09-28 00:27:17 -05002019int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2020{
2021 unsigned int max = READ_ONCE(sysctl_mount_max);
2022 unsigned int mounts = 0, old, pending, sum;
2023 struct mount *p;
2024
2025 for (p = mnt; p; p = next_mnt(p, mnt))
2026 mounts++;
2027
2028 old = ns->mounts;
2029 pending = ns->pending_mounts;
2030 sum = old + pending;
2031 if ((old > sum) ||
2032 (pending > sum) ||
2033 (max < sum) ||
2034 (mounts > (max - sum)))
2035 return -ENOSPC;
2036
2037 ns->pending_mounts = pending + mounts;
2038 return 0;
2039}
2040
Ram Paib90fa9a2005-11-07 17:19:50 -05002041/*
2042 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05002043 * @nd : place the mount tree @source_mnt is attached
2044 * @parent_nd : if non-null, detach the source_mnt from its parent and
2045 * store the parent mount and mountpoint dentry.
2046 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05002047 *
2048 * NOTE: in the table below explains the semantics when a source mount
2049 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05002050 * ---------------------------------------------------------------------------
2051 * | BIND MOUNT OPERATION |
2052 * |**************************************************************************
2053 * | source-->| shared | private | slave | unbindable |
2054 * | dest | | | | |
2055 * | | | | | | |
2056 * | v | | | | |
2057 * |**************************************************************************
2058 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2059 * | | | | | |
2060 * |non-shared| shared (+) | private | slave (*) | invalid |
2061 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05002062 * A bind operation clones the source mount and mounts the clone on the
2063 * destination mount.
2064 *
2065 * (++) the cloned mount is propagated to all the mounts in the propagation
2066 * tree of the destination mount and the cloned mount is added to
2067 * the peer group of the source mount.
2068 * (+) the cloned mount is created under the destination mount and is marked
2069 * as shared. The cloned mount is added to the peer group of the source
2070 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05002071 * (+++) the mount is propagated to all the mounts in the propagation tree
2072 * of the destination mount and the cloned mount is made slave
2073 * of the same master as that of the source mount. The cloned mount
2074 * is marked as 'shared and slave'.
2075 * (*) the cloned mount is made a slave of the same master as that of the
2076 * source mount.
2077 *
Ram Pai9676f0c2005-11-07 17:21:20 -05002078 * ---------------------------------------------------------------------------
2079 * | MOVE MOUNT OPERATION |
2080 * |**************************************************************************
2081 * | source-->| shared | private | slave | unbindable |
2082 * | dest | | | | |
2083 * | | | | | | |
2084 * | v | | | | |
2085 * |**************************************************************************
2086 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2087 * | | | | | |
2088 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2089 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05002090 *
2091 * (+) the mount is moved to the destination. And is then propagated to
2092 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05002093 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05002094 * (+++) the mount is moved to the destination and is then propagated to
2095 * all the mounts belonging to the destination mount's propagation tree.
2096 * the mount is marked as 'shared and slave'.
2097 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05002098 *
2099 * if the source mount is a tree, the operations explained above is
2100 * applied to each mount in the tree.
2101 * Must be called without spinlocks held, since this function can sleep
2102 * in allocations.
2103 */
Al Viro0fb54e52011-11-24 19:59:16 -05002104static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d171922013-03-15 10:53:28 -04002105 struct mount *dest_mnt,
2106 struct mountpoint *dest_mp,
Al Viro2763d112019-06-30 19:18:53 -04002107 bool moving)
Ram Paib90fa9a2005-11-07 17:19:50 -05002108{
Al Viro3bd045c2019-01-30 13:15:45 -05002109 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Al Viro38129a12014-03-20 21:10:51 -04002110 HLIST_HEAD(tree_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002111 struct mnt_namespace *ns = dest_mnt->mnt_ns;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002112 struct mountpoint *smp;
Al Viro315fc832011-11-24 18:57:30 -05002113 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04002114 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002115 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002116
Eric W. Biederman1064f872017-01-20 18:28:35 +13002117 /* Preallocate a mountpoint in case the new mounts need
2118 * to be tucked under other mounts.
2119 */
2120 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2121 if (IS_ERR(smp))
2122 return PTR_ERR(smp);
2123
Eric W. Biedermand2921682016-09-28 00:27:17 -05002124 /* Is there space to add these mounts to the mount namespace? */
Al Viro2763d112019-06-30 19:18:53 -04002125 if (!moving) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05002126 err = count_mounts(ns, source_mnt);
2127 if (err)
2128 goto out;
2129 }
2130
Al Virofc7be132011-11-25 01:05:37 -05002131 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05002132 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002133 if (err)
2134 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04002135 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05002136 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002137 if (err)
2138 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05002139 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05002140 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04002141 } else {
2142 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05002143 }
Al Viro2763d112019-06-30 19:18:53 -04002144 if (moving) {
2145 unhash_mnt(source_mnt);
Al Viro84d171922013-03-15 10:53:28 -04002146 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05002147 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05002148 } else {
David Howells44dfd842018-11-05 17:40:31 +00002149 if (source_mnt->mnt_ns) {
2150 /* move from anon - the caller will destroy */
2151 list_del_init(&source_mnt->mnt_ns->list);
2152 }
Al Viro84d171922013-03-15 10:53:28 -04002153 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002154 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05002155 }
Ram Paib90fa9a2005-11-07 17:19:50 -05002156
Al Viro38129a12014-03-20 21:10:51 -04002157 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04002158 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04002159 hlist_del_init(&child->mnt_hash);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002160 q = __lookup_mnt(&child->mnt_parent->mnt,
2161 child->mnt_mountpoint);
2162 if (q)
2163 mnt_change_mountpoint(child, smp, q);
Al Viro3bd045c2019-01-30 13:15:45 -05002164 /* Notice when we are propagating across user namespaces */
2165 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2166 lock_mnt_tree(child);
Christian Braunerd728cf72019-06-17 23:22:14 +02002167 child->mnt.mnt_flags &= ~MNT_LOCKED;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002168 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05002169 }
Eric W. Biederman1064f872017-01-20 18:28:35 +13002170 put_mountpoint(smp);
Al Viro719ea2f2013-09-29 11:24:49 -04002171 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10002172
Ram Paib90fa9a2005-11-07 17:19:50 -05002173 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002174
2175 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05002176 while (!hlist_empty(&tree_list)) {
2177 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002178 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06002179 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05002180 }
2181 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002182 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002183 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05002184 ns->pending_mounts = 0;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002185
2186 read_seqlock_excl(&mount_lock);
2187 put_mountpoint(smp);
2188 read_sequnlock_excl(&mount_lock);
2189
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002190 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002191}
2192
Al Viro84d171922013-03-15 10:53:28 -04002193static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04002194{
2195 struct vfsmount *mnt;
Al Viro84d171922013-03-15 10:53:28 -04002196 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04002197retry:
Al Viro59551022016-01-22 15:40:57 -05002198 inode_lock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002199 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002200 inode_unlock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002201 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04002202 }
Al Viro97216be2013-03-16 15:12:40 -04002203 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04002204 mnt = lookup_mnt(path);
Al Viro84d171922013-03-15 10:53:28 -04002205 if (likely(!mnt)) {
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002206 struct mountpoint *mp = get_mountpoint(dentry);
Al Viro84d171922013-03-15 10:53:28 -04002207 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04002208 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002209 inode_unlock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002210 return mp;
2211 }
2212 return mp;
2213 }
Al Viro97216be2013-03-16 15:12:40 -04002214 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002215 inode_unlock(path->dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002216 path_put(path);
2217 path->mnt = mnt;
Al Viro84d171922013-03-15 10:53:28 -04002218 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04002219 goto retry;
2220}
2221
Al Viro84d171922013-03-15 10:53:28 -04002222static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002223{
Al Viro84d171922013-03-15 10:53:28 -04002224 struct dentry *dentry = where->m_dentry;
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002225
2226 read_seqlock_excl(&mount_lock);
Al Viro84d171922013-03-15 10:53:28 -04002227 put_mountpoint(where);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002228 read_sequnlock_excl(&mount_lock);
2229
Al Viro328e6d92013-03-16 14:49:45 -04002230 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002231 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002232}
2233
Al Viro84d171922013-03-15 10:53:28 -04002234static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235{
David Howellse462ec52017-07-17 08:45:35 +01002236 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return -EINVAL;
2238
David Howellse36cb0b2015-01-29 12:02:35 +00002239 if (d_is_dir(mp->m_dentry) !=
2240 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 return -ENOTDIR;
2242
Al Viro2763d112019-06-30 19:18:53 -04002243 return attach_recursive_mnt(mnt, p, mp, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244}
2245
2246/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002247 * Sanity check the flags to change_mnt_propagation.
2248 */
2249
David Howellse462ec52017-07-17 08:45:35 +01002250static int flags_to_propagation_type(int ms_flags)
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002251{
David Howellse462ec52017-07-17 08:45:35 +01002252 int type = ms_flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002253
2254 /* Fail if any non-propagation flags are set */
2255 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2256 return 0;
2257 /* Only one propagation flag should be set */
2258 if (!is_power_of_2(type))
2259 return 0;
2260 return type;
2261}
2262
2263/*
Ram Pai07b20882005-11-07 17:19:07 -05002264 * recursively change the type of the mountpoint.
2265 */
David Howellse462ec52017-07-17 08:45:35 +01002266static int do_change_type(struct path *path, int ms_flags)
Ram Pai07b20882005-11-07 17:19:07 -05002267{
Al Viro315fc832011-11-24 18:57:30 -05002268 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002269 struct mount *mnt = real_mount(path->mnt);
David Howellse462ec52017-07-17 08:45:35 +01002270 int recurse = ms_flags & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002271 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002272 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002273
Al Viro2d92ab32008-08-02 00:51:11 -04002274 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002275 return -EINVAL;
2276
David Howellse462ec52017-07-17 08:45:35 +01002277 type = flags_to_propagation_type(ms_flags);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002278 if (!type)
2279 return -EINVAL;
2280
Al Viro97216be2013-03-16 15:12:40 -04002281 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002282 if (type == MS_SHARED) {
2283 err = invent_group_ids(mnt, recurse);
2284 if (err)
2285 goto out_unlock;
2286 }
2287
Al Viro719ea2f2013-09-29 11:24:49 -04002288 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002289 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002290 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002291 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002292
2293 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002294 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002295 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002296}
2297
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07002298static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2299{
2300 struct mount *child;
2301 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2302 if (!is_subdir(child->mnt_mountpoint, dentry))
2303 continue;
2304
2305 if (child->mnt.mnt_flags & MNT_LOCKED)
2306 return true;
2307 }
2308 return false;
2309}
2310
Al Viroa07b2002018-11-05 17:40:30 +00002311static struct mount *__do_loopback(struct path *old_path, int recurse)
2312{
2313 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2314
2315 if (IS_MNT_UNBINDABLE(old))
2316 return mnt;
2317
2318 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2319 return mnt;
2320
2321 if (!recurse && has_locked_children(old, old_path->dentry))
2322 return mnt;
2323
2324 if (recurse)
2325 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2326 else
2327 mnt = clone_mnt(old, old_path->dentry, 0);
2328
2329 if (!IS_ERR(mnt))
2330 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2331
2332 return mnt;
2333}
2334
Ram Pai07b20882005-11-07 17:19:07 -05002335/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 * do loopback mount.
2337 */
Al Viro808d4e32012-10-11 11:42:01 -04002338static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002339 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Al Viro2d92ab32008-08-02 00:51:11 -04002341 struct path old_path;
Al Viroa07b2002018-11-05 17:40:30 +00002342 struct mount *mnt = NULL, *parent;
Al Viro84d171922013-03-15 10:53:28 -04002343 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002344 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (!old_name || !*old_name)
2346 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002347 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 if (err)
2349 return err;
2350
Eric W. Biederman8823c072010-03-07 18:49:36 -08002351 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002352 if (mnt_ns_loop(old_path.dentry))
David Howellsdd111b32017-07-04 17:25:09 +01002353 goto out;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002354
Al Viro84d171922013-03-15 10:53:28 -04002355 mp = lock_mount(path);
Al Viroa07b2002018-11-05 17:40:30 +00002356 if (IS_ERR(mp)) {
2357 err = PTR_ERR(mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002358 goto out;
Al Viroa07b2002018-11-05 17:40:30 +00002359 }
Ram Pai9676f0c2005-11-07 17:21:20 -05002360
Al Viro84d171922013-03-15 10:53:28 -04002361 parent = real_mount(path->mnt);
Al Viroe149ed22014-11-01 10:57:28 -04002362 if (!check_mnt(parent))
2363 goto out2;
2364
Al Viroa07b2002018-11-05 17:40:30 +00002365 mnt = __do_loopback(&old_path, recurse);
David Howellsbe34d1a2012-06-25 12:55:18 +01002366 if (IS_ERR(mnt)) {
2367 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002368 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002369 }
Al Viroccd48bc2005-11-07 17:15:04 -05002370
Al Viro84d171922013-03-15 10:53:28 -04002371 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002372 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002373 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002374 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002375 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002376 }
Al Virob12cea92011-03-18 08:55:38 -04002377out2:
Al Viro84d171922013-03-15 10:53:28 -04002378 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002379out:
Al Viro2d92ab32008-08-02 00:51:11 -04002380 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 return err;
2382}
2383
Al Viroa07b2002018-11-05 17:40:30 +00002384static struct file *open_detached_copy(struct path *path, bool recursive)
2385{
2386 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2387 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2388 struct mount *mnt, *p;
2389 struct file *file;
2390
2391 if (IS_ERR(ns))
2392 return ERR_CAST(ns);
2393
2394 namespace_lock();
2395 mnt = __do_loopback(path, recursive);
2396 if (IS_ERR(mnt)) {
2397 namespace_unlock();
2398 free_mnt_ns(ns);
2399 return ERR_CAST(mnt);
2400 }
2401
2402 lock_mount_hash();
2403 for (p = mnt; p; p = next_mnt(p, mnt)) {
2404 p->mnt_ns = ns;
2405 ns->mounts++;
2406 }
2407 ns->root = mnt;
2408 list_add_tail(&ns->list, &mnt->mnt_list);
2409 mntget(&mnt->mnt);
2410 unlock_mount_hash();
2411 namespace_unlock();
2412
2413 mntput(path->mnt);
2414 path->mnt = &mnt->mnt;
2415 file = dentry_open(path, O_PATH, current_cred());
2416 if (IS_ERR(file))
2417 dissolve_on_fput(path->mnt);
2418 else
2419 file->f_mode |= FMODE_NEED_UNMOUNT;
2420 return file;
2421}
2422
Ben Dooks2658ce02019-10-15 11:35:02 +01002423SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
Al Viroa07b2002018-11-05 17:40:30 +00002424{
2425 struct file *file;
2426 struct path path;
2427 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2428 bool detached = flags & OPEN_TREE_CLONE;
2429 int error;
2430 int fd;
2431
2432 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2433
2434 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2435 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2436 OPEN_TREE_CLOEXEC))
2437 return -EINVAL;
2438
2439 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2440 return -EINVAL;
2441
2442 if (flags & AT_NO_AUTOMOUNT)
2443 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2444 if (flags & AT_SYMLINK_NOFOLLOW)
2445 lookup_flags &= ~LOOKUP_FOLLOW;
2446 if (flags & AT_EMPTY_PATH)
2447 lookup_flags |= LOOKUP_EMPTY;
2448
2449 if (detached && !may_mount())
2450 return -EPERM;
2451
2452 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2453 if (fd < 0)
2454 return fd;
2455
2456 error = user_path_at(dfd, filename, lookup_flags, &path);
2457 if (unlikely(error)) {
2458 file = ERR_PTR(error);
2459 } else {
2460 if (detached)
2461 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2462 else
2463 file = dentry_open(&path, O_PATH, current_cred());
2464 path_put(&path);
2465 }
2466 if (IS_ERR(file)) {
2467 put_unused_fd(fd);
2468 return PTR_ERR(file);
2469 }
2470 fd_install(fd, file);
2471 return fd;
2472}
2473
David Howells43f5e652018-11-01 23:07:25 +00002474/*
2475 * Don't allow locked mount flags to be cleared.
2476 *
2477 * No locks need to be held here while testing the various MNT_LOCK
2478 * flags because those flags can never be cleared once they are set.
2479 */
2480static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002481{
David Howells43f5e652018-11-01 23:07:25 +00002482 unsigned int fl = mnt->mnt.mnt_flags;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002483
David Howells43f5e652018-11-01 23:07:25 +00002484 if ((fl & MNT_LOCK_READONLY) &&
2485 !(mnt_flags & MNT_READONLY))
2486 return false;
2487
2488 if ((fl & MNT_LOCK_NODEV) &&
2489 !(mnt_flags & MNT_NODEV))
2490 return false;
2491
2492 if ((fl & MNT_LOCK_NOSUID) &&
2493 !(mnt_flags & MNT_NOSUID))
2494 return false;
2495
2496 if ((fl & MNT_LOCK_NOEXEC) &&
2497 !(mnt_flags & MNT_NOEXEC))
2498 return false;
2499
2500 if ((fl & MNT_LOCK_ATIME) &&
2501 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2502 return false;
2503
2504 return true;
2505}
2506
2507static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2508{
2509 bool readonly_request = (mnt_flags & MNT_READONLY);
2510
2511 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002512 return 0;
2513
2514 if (readonly_request)
David Howells43f5e652018-11-01 23:07:25 +00002515 return mnt_make_readonly(mnt);
2516
2517 return __mnt_unmake_readonly(mnt);
2518}
2519
2520/*
2521 * Update the user-settable attributes on a mount. The caller must hold
2522 * sb->s_umount for writing.
2523 */
2524static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2525{
2526 lock_mount_hash();
2527 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2528 mnt->mnt.mnt_flags = mnt_flags;
2529 touch_mnt_namespace(mnt->mnt_ns);
2530 unlock_mount_hash();
2531}
2532
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002533static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2534{
2535 struct super_block *sb = mnt->mnt_sb;
2536
2537 if (!__mnt_is_readonly(mnt) &&
2538 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2539 char *buf = (char *)__get_free_page(GFP_KERNEL);
2540 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2541 struct tm tm;
2542
2543 time64_to_tm(sb->s_time_max, 0, &tm);
2544
Eric Biggers0ecee662019-10-16 19:48:14 -07002545 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2546 sb->s_type->name,
2547 is_mounted(mnt) ? "remounted" : "mounted",
2548 mntpath,
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002549 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2550
2551 free_page((unsigned long)buf);
2552 }
2553}
2554
David Howells43f5e652018-11-01 23:07:25 +00002555/*
2556 * Handle reconfiguration of the mountpoint only without alteration of the
2557 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2558 * to mount(2).
2559 */
2560static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2561{
2562 struct super_block *sb = path->mnt->mnt_sb;
2563 struct mount *mnt = real_mount(path->mnt);
2564 int ret;
2565
2566 if (!check_mnt(mnt))
2567 return -EINVAL;
2568
2569 if (path->dentry != mnt->mnt.mnt_root)
2570 return -EINVAL;
2571
2572 if (!can_change_locked_flags(mnt, mnt_flags))
2573 return -EPERM;
2574
2575 down_write(&sb->s_umount);
2576 ret = change_mount_ro_state(mnt, mnt_flags);
2577 if (ret == 0)
2578 set_mount_attributes(mnt, mnt_flags);
2579 up_write(&sb->s_umount);
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002580
2581 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2582
David Howells43f5e652018-11-01 23:07:25 +00002583 return ret;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002584}
2585
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586/*
2587 * change filesystem flags. dir should be a physical root of filesystem.
2588 * If you've mounted a non-root directory somewhere and want to do remount
2589 * on it - tough luck.
2590 */
David Howellse462ec52017-07-17 08:45:35 +01002591static int do_remount(struct path *path, int ms_flags, int sb_flags,
2592 int mnt_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593{
2594 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002595 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002596 struct mount *mnt = real_mount(path->mnt);
David Howells8d0347f2018-11-04 09:28:36 -05002597 struct fs_context *fc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Al Viro143c8c92011-11-25 00:46:35 -05002599 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 return -EINVAL;
2601
Al Viro2d92ab32008-08-02 00:51:11 -04002602 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 return -EINVAL;
2604
David Howells43f5e652018-11-01 23:07:25 +00002605 if (!can_change_locked_flags(mnt, mnt_flags))
Eric W. Biederman07b64552014-07-28 17:10:56 -07002606 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002607
David Howells8d0347f2018-11-04 09:28:36 -05002608 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2609 if (IS_ERR(fc))
2610 return PTR_ERR(fc);
Eric Parisff36fe22011-03-03 16:09:14 -05002611
Miklos Szeredib3309662020-07-14 14:45:41 +02002612 fc->oldapi = true;
David Howells8d0347f2018-11-04 09:28:36 -05002613 err = parse_monolithic_mount_data(fc, data);
2614 if (!err) {
2615 down_write(&sb->s_umount);
2616 err = -EPERM;
2617 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2618 err = reconfigure_super(fc);
2619 if (!err)
2620 set_mount_attributes(mnt, mnt_flags);
2621 }
2622 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07002623 }
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002624
2625 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2626
David Howells8d0347f2018-11-04 09:28:36 -05002627 put_fs_context(fc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 return err;
2629}
2630
Al Virocbbe3622011-11-24 20:01:19 -05002631static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002632{
Al Viro315fc832011-11-24 18:57:30 -05002633 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002634 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002635 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002636 return 1;
2637 }
2638 return 0;
2639}
2640
David Howells44dfd842018-11-05 17:40:31 +00002641/*
2642 * Check that there aren't references to earlier/same mount namespaces in the
2643 * specified subtree. Such references can act as pins for mount namespaces
2644 * that aren't checked by the mount-cycle checking code, thereby allowing
2645 * cycles to be made.
2646 */
2647static bool check_for_nsfs_mounts(struct mount *subtree)
2648{
2649 struct mount *p;
2650 bool ret = false;
2651
2652 lock_mount_hash();
2653 for (p = subtree; p; p = next_mnt(p, subtree))
2654 if (mnt_ns_loop(p->mnt.mnt_root))
2655 goto out;
2656
2657 ret = true;
2658out:
2659 unlock_mount_hash();
2660 return ret;
2661}
2662
David Howells2db154b2018-11-05 17:40:30 +00002663static int do_move_mount(struct path *old_path, struct path *new_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664{
David Howells44dfd842018-11-05 17:40:31 +00002665 struct mnt_namespace *ns;
Al Viro676da582011-11-24 21:47:05 -05002666 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002667 struct mount *old;
Al Viro2763d112019-06-30 19:18:53 -04002668 struct mount *parent;
2669 struct mountpoint *mp, *old_mp;
Al Viro57eccb82013-02-22 22:49:10 -05002670 int err;
David Howells44dfd842018-11-05 17:40:31 +00002671 bool attached;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
David Howells2db154b2018-11-05 17:40:30 +00002673 mp = lock_mount(new_path);
Al Viro84d171922013-03-15 10:53:28 -04002674 if (IS_ERR(mp))
David Howells2db154b2018-11-05 17:40:30 +00002675 return PTR_ERR(mp);
David Howellscc53ce52011-01-14 18:45:26 +00002676
David Howells2db154b2018-11-05 17:40:30 +00002677 old = real_mount(old_path->mnt);
2678 p = real_mount(new_path->mnt);
Al Viro2763d112019-06-30 19:18:53 -04002679 parent = old->mnt_parent;
David Howells44dfd842018-11-05 17:40:31 +00002680 attached = mnt_has_parent(old);
Al Viro2763d112019-06-30 19:18:53 -04002681 old_mp = old->mnt_mp;
David Howells44dfd842018-11-05 17:40:31 +00002682 ns = old->mnt_ns;
Al Viro143c8c92011-11-25 00:46:35 -05002683
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 err = -EINVAL;
David Howells44dfd842018-11-05 17:40:31 +00002685 /* The mountpoint must be in our namespace. */
2686 if (!check_mnt(p))
David Howells2db154b2018-11-05 17:40:30 +00002687 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688
Eric Biggers570d7a92019-06-29 13:27:44 -07002689 /* The thing moved must be mounted... */
2690 if (!is_mounted(&old->mnt))
David Howells44dfd842018-11-05 17:40:31 +00002691 goto out;
2692
Eric Biggers570d7a92019-06-29 13:27:44 -07002693 /* ... and either ours or the root of anon namespace */
2694 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
David Howells2db154b2018-11-05 17:40:30 +00002695 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696
David Howells2db154b2018-11-05 17:40:30 +00002697 if (old->mnt.mnt_flags & MNT_LOCKED)
2698 goto out;
2699
2700 if (old_path->dentry != old_path->mnt->mnt_root)
2701 goto out;
2702
2703 if (d_is_dir(new_path->dentry) !=
2704 d_is_dir(old_path->dentry))
2705 goto out;
Ram Pai21444402005-11-07 17:20:03 -05002706 /*
2707 * Don't move a mount residing in a shared parent.
2708 */
Al Viro2763d112019-06-30 19:18:53 -04002709 if (attached && IS_MNT_SHARED(parent))
David Howells2db154b2018-11-05 17:40:30 +00002710 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002711 /*
2712 * Don't move a mount tree containing unbindable mounts to a destination
2713 * mount which is shared.
2714 */
Al Virofc7be132011-11-25 01:05:37 -05002715 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
David Howells2db154b2018-11-05 17:40:30 +00002716 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 err = -ELOOP;
David Howells44dfd842018-11-05 17:40:31 +00002718 if (!check_for_nsfs_mounts(old))
2719 goto out;
Al Virofc7be132011-11-25 01:05:37 -05002720 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002721 if (p == old)
David Howells2db154b2018-11-05 17:40:30 +00002722 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723
David Howells2db154b2018-11-05 17:40:30 +00002724 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
Al Viro2763d112019-06-30 19:18:53 -04002725 attached);
Jan Blunck4ac91372008-02-14 19:34:32 -08002726 if (err)
David Howells2db154b2018-11-05 17:40:30 +00002727 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728
2729 /* if the mount is moved, it should no longer be expire
2730 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002731 list_del_init(&old->mnt_expire);
Al Viro2763d112019-06-30 19:18:53 -04002732 if (attached)
2733 put_mountpoint(old_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734out:
David Howells2db154b2018-11-05 17:40:30 +00002735 unlock_mount(mp);
David Howells44dfd842018-11-05 17:40:31 +00002736 if (!err) {
Al Viro2763d112019-06-30 19:18:53 -04002737 if (attached)
2738 mntput_no_expire(parent);
2739 else
David Howells44dfd842018-11-05 17:40:31 +00002740 free_mnt_ns(ns);
2741 }
David Howells2db154b2018-11-05 17:40:30 +00002742 return err;
2743}
2744
2745static int do_move_mount_old(struct path *path, const char *old_name)
2746{
2747 struct path old_path;
2748 int err;
2749
2750 if (!old_name || !*old_name)
2751 return -EINVAL;
2752
2753 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2754 if (err)
2755 return err;
2756
2757 err = do_move_mount(&old_path, path);
Al Viro2d92ab32008-08-02 00:51:11 -04002758 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 return err;
2760}
2761
Al Viro9d412a42011-03-17 22:08:28 -04002762/*
2763 * add a mount into a namespace's mount tree
2764 */
Al Viro8f115382020-01-11 10:14:09 -05002765static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2766 struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04002767{
Al Viro8f115382020-01-11 10:14:09 -05002768 struct mount *parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04002769
Al Virof2ebb3a2014-02-27 09:35:45 -05002770 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04002771
Al Viro84d171922013-03-15 10:53:28 -04002772 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04002773 /* that's acceptable only for automounts done in private ns */
2774 if (!(mnt_flags & MNT_SHRINKABLE))
Al Viro8f115382020-01-11 10:14:09 -05002775 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04002776 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d171922013-03-15 10:53:28 -04002777 if (!parent->mnt_ns)
Al Viro8f115382020-01-11 10:14:09 -05002778 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04002779 }
Al Viro9d412a42011-03-17 22:08:28 -04002780
2781 /* Refuse the same filesystem on the same mount point */
Al Viro95bc5f22011-11-25 00:30:56 -05002782 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04002783 path->mnt->mnt_root == path->dentry)
Al Viro8f115382020-01-11 10:14:09 -05002784 return -EBUSY;
Al Viro9d412a42011-03-17 22:08:28 -04002785
David Howellse36cb0b2015-01-29 12:02:35 +00002786 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro8f115382020-01-11 10:14:09 -05002787 return -EINVAL;
Al Viro9d412a42011-03-17 22:08:28 -04002788
Al Viro95bc5f22011-11-25 00:30:56 -05002789 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro8f115382020-01-11 10:14:09 -05002790 return graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04002791}
Al Virob1e75df2011-01-17 01:47:59 -05002792
David Howells132e4602018-11-04 07:43:08 -05002793static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2794
2795/*
2796 * Create a new mount using a superblock configuration and request it
2797 * be added to the namespace tree.
2798 */
2799static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2800 unsigned int mnt_flags)
2801{
2802 struct vfsmount *mnt;
Al Viro8f115382020-01-11 10:14:09 -05002803 struct mountpoint *mp;
David Howells132e4602018-11-04 07:43:08 -05002804 struct super_block *sb = fc->root->d_sb;
2805 int error;
2806
Al Viroc9ce29e2018-12-20 15:04:50 -05002807 error = security_sb_kern_mount(sb);
2808 if (!error && mount_too_revealing(sb, &mnt_flags))
2809 error = -EPERM;
2810
2811 if (unlikely(error)) {
2812 fc_drop_locked(fc);
2813 return error;
David Howells132e4602018-11-04 07:43:08 -05002814 }
2815
2816 up_write(&sb->s_umount);
2817
2818 mnt = vfs_create_mount(fc);
2819 if (IS_ERR(mnt))
2820 return PTR_ERR(mnt);
2821
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002822 mnt_warn_timestamp_expiry(mountpoint, mnt);
2823
Al Viro8f115382020-01-11 10:14:09 -05002824 mp = lock_mount(mountpoint);
2825 if (IS_ERR(mp)) {
2826 mntput(mnt);
2827 return PTR_ERR(mp);
2828 }
2829 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2830 unlock_mount(mp);
Eric Biggers0ecee662019-10-16 19:48:14 -07002831 if (error < 0)
2832 mntput(mnt);
David Howells132e4602018-11-04 07:43:08 -05002833 return error;
2834}
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05002835
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836/*
2837 * create a new mount for userspace and request it to be added into the
2838 * namespace's tree
2839 */
David Howellse462ec52017-07-17 08:45:35 +01002840static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
Al Viro808d4e32012-10-11 11:42:01 -04002841 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002843 struct file_system_type *type;
Al Viroa0c9a8b2018-11-04 07:18:51 -05002844 struct fs_context *fc;
2845 const char *subtype = NULL;
2846 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002848 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 return -EINVAL;
2850
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002851 type = get_fs_type(fstype);
2852 if (!type)
2853 return -ENODEV;
2854
Al Viroa0c9a8b2018-11-04 07:18:51 -05002855 if (type->fs_flags & FS_HAS_SUBTYPE) {
2856 subtype = strchr(fstype, '.');
2857 if (subtype) {
2858 subtype++;
2859 if (!*subtype) {
2860 put_filesystem(type);
2861 return -EINVAL;
2862 }
Al Viroa0c9a8b2018-11-04 07:18:51 -05002863 }
Eric W. Biederman8654df42016-06-09 16:06:06 -05002864 }
2865
Al Viroa0c9a8b2018-11-04 07:18:51 -05002866 fc = fs_context_for_mount(type, sb_flags);
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07002867 put_filesystem(type);
Al Viroa0c9a8b2018-11-04 07:18:51 -05002868 if (IS_ERR(fc))
2869 return PTR_ERR(fc);
2870
David Howells3e1aeb02018-11-01 23:07:25 +00002871 if (subtype)
2872 err = vfs_parse_fs_string(fc, "subtype",
2873 subtype, strlen(subtype));
2874 if (!err && name)
2875 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
Al Viroa0c9a8b2018-11-04 07:18:51 -05002876 if (!err)
2877 err = parse_monolithic_mount_data(fc, data);
Al Viroc3aabf02019-05-13 12:57:22 -04002878 if (!err && !mount_capable(fc))
2879 err = -EPERM;
Al Viroa0c9a8b2018-11-04 07:18:51 -05002880 if (!err)
2881 err = vfs_get_tree(fc);
David Howells132e4602018-11-04 07:43:08 -05002882 if (!err)
2883 err = do_new_mount_fc(fc, path, mnt_flags);
Al Viroa0c9a8b2018-11-04 07:18:51 -05002884
Al Viroa0c9a8b2018-11-04 07:18:51 -05002885 put_fs_context(fc);
Al Viro15f9a3f2011-01-17 01:41:58 -05002886 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887}
2888
Al Viro19a167a2011-01-17 01:35:23 -05002889int finish_automount(struct vfsmount *m, struct path *path)
2890{
Al Viro26df6032020-01-11 10:44:29 -05002891 struct dentry *dentry = path->dentry;
Al Viro8f115382020-01-11 10:14:09 -05002892 struct mountpoint *mp;
Al Viro25e195a2020-01-11 11:27:46 -05002893 struct mount *mnt;
Al Viro19a167a2011-01-17 01:35:23 -05002894 int err;
Al Viro25e195a2020-01-11 11:27:46 -05002895
2896 if (!m)
2897 return 0;
2898 if (IS_ERR(m))
2899 return PTR_ERR(m);
2900
2901 mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05002902 /* The new mount record should have at least 2 refs to prevent it being
2903 * expired before we get a chance to add it
2904 */
Al Viro6776db3d2011-11-25 00:22:05 -05002905 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05002906
2907 if (m->mnt_sb == path->mnt->mnt_sb &&
Al Viro26df6032020-01-11 10:44:29 -05002908 m->mnt_root == dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05002909 err = -ELOOP;
Al Viro26df6032020-01-11 10:44:29 -05002910 goto discard;
Al Viro19a167a2011-01-17 01:35:23 -05002911 }
2912
Al Viro26df6032020-01-11 10:44:29 -05002913 /*
2914 * we don't want to use lock_mount() - in this case finding something
2915 * that overmounts our mountpoint to be means "quitely drop what we've
2916 * got", not "try to mount it on top".
2917 */
2918 inode_lock(dentry->d_inode);
2919 namespace_lock();
2920 if (unlikely(cant_mount(dentry))) {
2921 err = -ENOENT;
2922 goto discard_locked;
2923 }
2924 rcu_read_lock();
2925 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2926 rcu_read_unlock();
2927 err = 0;
2928 goto discard_locked;
2929 }
2930 rcu_read_unlock();
2931 mp = get_mountpoint(dentry);
Al Viro8f115382020-01-11 10:14:09 -05002932 if (IS_ERR(mp)) {
2933 err = PTR_ERR(mp);
Al Viro26df6032020-01-11 10:44:29 -05002934 goto discard_locked;
Al Viro8f115382020-01-11 10:14:09 -05002935 }
Al Viro26df6032020-01-11 10:44:29 -05002936
Al Viro8f115382020-01-11 10:14:09 -05002937 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2938 unlock_mount(mp);
Al Viro26df6032020-01-11 10:44:29 -05002939 if (unlikely(err))
2940 goto discard;
2941 mntput(m);
2942 return 0;
2943
2944discard_locked:
2945 namespace_unlock();
2946 inode_unlock(dentry->d_inode);
2947discard:
Al Virob1e75df2011-01-17 01:47:59 -05002948 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05002949 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04002950 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05002951 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04002952 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05002953 }
Al Virob1e75df2011-01-17 01:47:59 -05002954 mntput(m);
2955 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05002956 return err;
2957}
2958
David Howellsea5b7782011-01-14 19:10:03 +00002959/**
2960 * mnt_set_expiry - Put a mount on an expiration list
2961 * @mnt: The mount to list.
2962 * @expiry_list: The list to add the mount to.
2963 */
2964void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2965{
Al Viro97216be2013-03-16 15:12:40 -04002966 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00002967
Al Viro6776db3d2011-11-25 00:22:05 -05002968 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00002969
Al Viro97216be2013-03-16 15:12:40 -04002970 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00002971}
2972EXPORT_SYMBOL(mnt_set_expiry);
2973
2974/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 * process a list of expirable mountpoints with the intent of discarding any
2976 * mountpoints that aren't in use and haven't been touched since last we came
2977 * here
2978 */
2979void mark_mounts_for_expiry(struct list_head *mounts)
2980{
Al Viro761d5c32011-11-24 21:07:43 -05002981 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 LIST_HEAD(graveyard);
2983
2984 if (list_empty(mounts))
2985 return;
2986
Al Viro97216be2013-03-16 15:12:40 -04002987 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002988 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
2990 /* extract from the expiration list every vfsmount that matches the
2991 * following criteria:
2992 * - only referenced by its parent vfsmount
2993 * - still marked for expiry (marked on the last call here; marks are
2994 * cleared by mntput())
2995 */
Al Viro6776db3d2011-11-25 00:22:05 -05002996 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05002997 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05002998 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05003000 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 }
Al Virobcc5c7d2008-03-22 00:21:53 -04003002 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003003 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003004 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003005 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003006 }
Al Viro719ea2f2013-09-29 11:24:49 -04003007 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04003008 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009}
3010
3011EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3012
3013/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04003014 * Ripoff of 'select_parent()'
3015 *
3016 * search the list of submounts for a given mountpoint, and move any
3017 * shrinkable submounts to the 'graveyard' list.
3018 */
Al Viro692afc32011-11-24 21:15:14 -05003019static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003020{
Al Viro692afc32011-11-24 21:15:14 -05003021 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003022 struct list_head *next;
3023 int found = 0;
3024
3025repeat:
Al Viro6b41d532011-11-24 23:24:33 -05003026 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003027resume:
Al Viro6b41d532011-11-24 23:24:33 -05003028 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003029 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05003030 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003031
3032 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05003033 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04003034 continue;
3035 /*
3036 * Descend a level if the d_mounts list is non-empty.
3037 */
Al Viro6b41d532011-11-24 23:24:33 -05003038 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003039 this_parent = mnt;
3040 goto repeat;
3041 }
3042
Al Viro1ab59732011-11-24 21:35:16 -05003043 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003044 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003045 found++;
3046 }
3047 }
3048 /*
3049 * All done at this level ... ascend and resume the search
3050 */
3051 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05003052 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05003053 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003054 goto resume;
3055 }
3056 return found;
3057}
3058
3059/*
3060 * process a list of expirable mountpoints with the intent of discarding any
3061 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10003062 *
Al Viro48a066e2013-09-29 22:06:07 -04003063 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04003064 */
Al Virob54b9be2013-03-16 14:39:34 -04003065static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003066{
3067 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05003068 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003069
Trond Myklebust5528f9112006-06-09 09:34:17 -04003070 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04003071 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04003072 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05003073 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05003074 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003075 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003076 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003077 }
3078 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04003079}
3080
Christoph Hellwig028abd92020-09-17 10:22:34 +02003081static void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082{
Al Virob40ef862015-12-14 18:44:44 -05003083 char *copy;
Catalin Marinasd563d672020-07-01 17:46:06 +01003084 unsigned left, offset;
Ram Paib58fed82005-11-07 17:16:09 -05003085
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05003087 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
Al Virob40ef862015-12-14 18:44:44 -05003089 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3090 if (!copy)
3091 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092
Catalin Marinasd563d672020-07-01 17:46:06 +01003093 left = copy_from_user(copy, data, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
Catalin Marinasd563d672020-07-01 17:46:06 +01003095 /*
3096 * Not all architectures have an exact copy_from_user(). Resort to
3097 * byte at a time.
3098 */
3099 offset = PAGE_SIZE - left;
3100 while (left) {
3101 char c;
3102 if (get_user(c, (const char __user *)data + offset))
3103 break;
3104 copy[offset] = c;
3105 left--;
3106 offset++;
3107 }
3108
3109 if (left == PAGE_SIZE) {
Al Virob40ef862015-12-14 18:44:44 -05003110 kfree(copy);
3111 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112 }
Catalin Marinasd563d672020-07-01 17:46:06 +01003113
Al Virob40ef862015-12-14 18:44:44 -05003114 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115}
3116
Christoph Hellwig028abd92020-09-17 10:22:34 +02003117static char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07003118{
Chandan Rajendrafbdb4402019-01-22 12:21:52 +05303119 return data ? strndup_user(data, PATH_MAX) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003120}
3121
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122/*
3123 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3124 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3125 *
3126 * data is a (void *) that can point to any structure up to
3127 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3128 * information (or be NULL).
3129 *
3130 * Pre-0.97 versions of mount() didn't have a flags word.
3131 * When the flags word was introduced its top half was required
3132 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3133 * Therefore, if this magic number is present, it carries no information
3134 * and must be discarded.
3135 */
Christoph Hellwigc60166f2020-07-21 11:12:08 +02003136int path_mount(const char *dev_name, struct path *path,
Al Viro808d4e32012-10-11 11:42:01 -04003137 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138{
David Howellse462ec52017-07-17 08:45:35 +01003139 unsigned int mnt_flags = 0, sb_flags;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003140 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141
3142 /* Discard magic */
3143 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3144 flags &= ~MS_MGC_MSK;
3145
3146 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 if (data_page)
3148 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3149
David Howellse462ec52017-07-17 08:45:35 +01003150 if (flags & MS_NOUSER)
3151 return -EINVAL;
3152
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003153 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3154 if (ret)
3155 return ret;
3156 if (!may_mount())
3157 return -EPERM;
3158 if ((flags & SB_MANDLOCK) && !may_mandlock())
3159 return -EPERM;
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09003160
Andi Kleen613cbe32009-04-19 18:40:43 +02003161 /* Default to relatime unless overriden */
3162 if (!(flags & MS_NOATIME))
3163 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00003164
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 /* Separate the per-mountpoint flags */
3166 if (flags & MS_NOSUID)
3167 mnt_flags |= MNT_NOSUID;
3168 if (flags & MS_NODEV)
3169 mnt_flags |= MNT_NODEV;
3170 if (flags & MS_NOEXEC)
3171 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003172 if (flags & MS_NOATIME)
3173 mnt_flags |= MNT_NOATIME;
3174 if (flags & MS_NODIRATIME)
3175 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00003176 if (flags & MS_STRICTATIME)
3177 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
David Howellsa9e5b732018-04-20 13:35:02 +01003178 if (flags & MS_RDONLY)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08003179 mnt_flags |= MNT_READONLY;
Mattias Nisslerdab741e02020-08-27 11:09:46 -06003180 if (flags & MS_NOSYMFOLLOW)
3181 mnt_flags |= MNT_NOSYMFOLLOW;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003182
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003183 /* The default atime for remount is preservation */
3184 if ((flags & MS_REMOUNT) &&
3185 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3186 MS_STRICTATIME)) == 0)) {
3187 mnt_flags &= ~MNT_ATIME_MASK;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003188 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003189 }
3190
David Howellse462ec52017-07-17 08:45:35 +01003191 sb_flags = flags & (SB_RDONLY |
3192 SB_SYNCHRONOUS |
3193 SB_MANDLOCK |
3194 SB_DIRSYNC |
3195 SB_SILENT |
Mimi Zohar917086f2017-10-08 00:28:21 -04003196 SB_POSIXACL |
Markus Trippelsdorfd7ee9462017-10-11 07:01:31 +02003197 SB_LAZYTIME |
Mimi Zohar917086f2017-10-08 00:28:21 -04003198 SB_I_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199
David Howells43f5e652018-11-01 23:07:25 +00003200 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003201 return do_reconfigure_mnt(path, mnt_flags);
3202 if (flags & MS_REMOUNT)
3203 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3204 if (flags & MS_BIND)
3205 return do_loopback(path, dev_name, flags & MS_REC);
3206 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3207 return do_change_type(path, flags);
3208 if (flags & MS_MOVE)
3209 return do_move_mount_old(path, dev_name);
3210
3211 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3212 data_page);
3213}
3214
3215long do_mount(const char *dev_name, const char __user *dir_name,
3216 const char *type_page, unsigned long flags, void *data_page)
3217{
3218 struct path path;
3219 int ret;
3220
3221 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3222 if (ret)
3223 return ret;
3224 ret = path_mount(dev_name, &path, type_page, flags, data_page);
Al Viro2d92ab32008-08-02 00:51:11 -04003225 path_put(&path);
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003226 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227}
3228
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003229static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3230{
3231 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3232}
3233
3234static void dec_mnt_namespaces(struct ucounts *ucounts)
3235{
3236 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3237}
3238
Eric W. Biederman771b1372012-07-26 21:08:32 -07003239static void free_mnt_ns(struct mnt_namespace *ns)
3240{
Al Viro74e83122019-01-30 13:30:21 -05003241 if (!is_anon_ns(ns))
3242 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003243 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003244 put_user_ns(ns->user_ns);
3245 kfree(ns);
3246}
3247
Eric W. Biederman8823c072010-03-07 18:49:36 -08003248/*
3249 * Assign a sequence number so we can detect when we attempt to bind
3250 * mount a reference to an older mount namespace into the current
3251 * mount namespace, preventing reference counting loops. A 64bit
3252 * number incrementing at 10Ghz will take 12,427 years to wrap which
3253 * is effectively never, so we can ignore the possibility.
3254 */
3255static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3256
Al Viro74e83122019-01-30 13:30:21 -05003257static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003258{
3259 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003260 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003261 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003262
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003263 ucounts = inc_mnt_namespaces(user_ns);
3264 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05003265 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003266
Al Viro74e83122019-01-30 13:30:21 -05003267 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003268 if (!new_ns) {
3269 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003270 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003271 }
Al Viro74e83122019-01-30 13:30:21 -05003272 if (!anon) {
3273 ret = ns_alloc_inum(&new_ns->ns);
3274 if (ret) {
3275 kfree(new_ns);
3276 dec_mnt_namespaces(ucounts);
3277 return ERR_PTR(ret);
3278 }
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003279 }
Al Viro33c42942014-11-01 02:32:53 -04003280 new_ns->ns.ops = &mntns_operations;
Al Viro74e83122019-01-30 13:30:21 -05003281 if (!anon)
3282 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03003283 refcount_set(&new_ns->ns.count, 1);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003284 INIT_LIST_HEAD(&new_ns->list);
3285 init_waitqueue_head(&new_ns->poll);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02003286 spin_lock_init(&new_ns->ns_lock);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003287 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003288 new_ns->ucounts = ucounts;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003289 return new_ns;
3290}
3291
Emese Revfy0766f782016-06-20 20:42:34 +02003292__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04003293struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3294 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003296 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04003297 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05003298 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04003299 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05003300 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003301 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302
Al Viro9559f682013-09-28 20:47:57 -04003303 BUG_ON(!ns);
3304
3305 if (likely(!(flags & CLONE_NEWNS))) {
3306 get_mnt_ns(ns);
3307 return ns;
3308 }
3309
3310 old = ns->root;
3311
Al Viro74e83122019-01-30 13:30:21 -05003312 new_ns = alloc_mnt_ns(user_ns, false);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003313 if (IS_ERR(new_ns))
3314 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315
Al Viro97216be2013-03-16 15:12:40 -04003316 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003318 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04003319 if (user_ns != ns->user_ns)
Al Viro3bd045c2019-01-30 13:15:45 -05003320 copy_flags |= CL_SHARED_TO_SLAVE;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003321 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01003322 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04003323 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07003324 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01003325 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 }
Al Viro3bd045c2019-01-30 13:15:45 -05003327 if (user_ns != ns->user_ns) {
3328 lock_mount_hash();
3329 lock_mnt_tree(new);
3330 unlock_mount_hash();
3331 }
Al Virobe08d6d2011-12-06 13:32:36 -05003332 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05003333 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
3335 /*
3336 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3337 * as belonging to new namespace. We have already acquired a private
3338 * fs_struct, so tsk->fs->lock is not needed.
3339 */
Al Viro909b0a82011-11-25 03:06:56 -05003340 p = old;
Al Virocb338d02011-11-24 20:55:08 -05003341 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05003343 q->mnt_ns = new_ns;
Eric W. Biedermand2921682016-09-28 00:27:17 -05003344 new_ns->mounts++;
Al Viro9559f682013-09-28 20:47:57 -04003345 if (new_fs) {
3346 if (&p->mnt == new_fs->root.mnt) {
3347 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003348 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 }
Al Viro9559f682013-09-28 20:47:57 -04003350 if (&p->mnt == new_fs->pwd.mnt) {
3351 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003352 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 }
Al Viro909b0a82011-11-25 03:06:56 -05003355 p = next_mnt(p, old);
3356 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003357 if (!q)
3358 break;
3359 while (p->mnt.mnt_root != q->mnt.mnt_root)
3360 p = next_mnt(p, old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 }
Al Viro328e6d92013-03-16 14:49:45 -04003362 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05003365 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05003367 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368
JANAK DESAI741a2952006-02-07 12:59:00 -08003369 return new_ns;
3370}
3371
Al Viro74e83122019-01-30 13:30:21 -05003372struct dentry *mount_subtree(struct vfsmount *m, const char *name)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003373{
Al Viro74e83122019-01-30 13:30:21 -05003374 struct mount *mnt = real_mount(m);
Al Viroea441d12011-11-16 21:43:59 -05003375 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05003376 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05003377 struct path path;
3378 int err;
3379
Al Viro74e83122019-01-30 13:30:21 -05003380 ns = alloc_mnt_ns(&init_user_ns, true);
3381 if (IS_ERR(ns)) {
3382 mntput(m);
Al Viroea441d12011-11-16 21:43:59 -05003383 return ERR_CAST(ns);
Al Viro74e83122019-01-30 13:30:21 -05003384 }
3385 mnt->mnt_ns = ns;
3386 ns->root = mnt;
3387 ns->mounts++;
3388 list_add(&mnt->mnt_list, &ns->list);
Al Viroea441d12011-11-16 21:43:59 -05003389
Al Viro74e83122019-01-30 13:30:21 -05003390 err = vfs_path_lookup(m->mnt_root, m,
Al Viroea441d12011-11-16 21:43:59 -05003391 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3392
3393 put_mnt_ns(ns);
3394
3395 if (err)
3396 return ERR_PTR(err);
3397
3398 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05003399 s = path.mnt->mnt_sb;
3400 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05003401 mntput(path.mnt);
3402 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05003403 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05003404 /* ... and return the root of (sub)tree on it */
3405 return path.dentry;
3406}
3407EXPORT_SYMBOL(mount_subtree);
3408
Dominik Brodowskicccaa5e2018-10-23 22:41:09 +02003409SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3410 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411{
Vegard Nossumeca6f532009-09-18 13:05:45 -07003412 int ret;
3413 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003414 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05003415 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416
Tim Gardnerb8850d12014-08-28 11:26:03 -06003417 kernel_type = copy_mount_string(type);
3418 ret = PTR_ERR(kernel_type);
3419 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003420 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421
Tim Gardnerb8850d12014-08-28 11:26:03 -06003422 kernel_dev = copy_mount_string(dev_name);
3423 ret = PTR_ERR(kernel_dev);
3424 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003425 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426
Al Virob40ef862015-12-14 18:44:44 -05003427 options = copy_mount_options(data);
3428 ret = PTR_ERR(options);
3429 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003430 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431
Al Virob40ef862015-12-14 18:44:44 -05003432 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003433
Al Virob40ef862015-12-14 18:44:44 -05003434 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003435out_data:
3436 kfree(kernel_dev);
3437out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07003438 kfree(kernel_type);
3439out_type:
3440 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441}
3442
3443/*
David Howells93766fb2018-11-01 23:36:14 +00003444 * Create a kernel mount representation for a new, prepared superblock
3445 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3446 */
3447SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3448 unsigned int, attr_flags)
3449{
3450 struct mnt_namespace *ns;
3451 struct fs_context *fc;
3452 struct file *file;
3453 struct path newmount;
3454 struct mount *mnt;
3455 struct fd f;
3456 unsigned int mnt_flags = 0;
3457 long ret;
3458
3459 if (!may_mount())
3460 return -EPERM;
3461
3462 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3463 return -EINVAL;
3464
3465 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3466 MOUNT_ATTR_NOSUID |
3467 MOUNT_ATTR_NODEV |
3468 MOUNT_ATTR_NOEXEC |
3469 MOUNT_ATTR__ATIME |
3470 MOUNT_ATTR_NODIRATIME))
3471 return -EINVAL;
3472
3473 if (attr_flags & MOUNT_ATTR_RDONLY)
3474 mnt_flags |= MNT_READONLY;
3475 if (attr_flags & MOUNT_ATTR_NOSUID)
3476 mnt_flags |= MNT_NOSUID;
3477 if (attr_flags & MOUNT_ATTR_NODEV)
3478 mnt_flags |= MNT_NODEV;
3479 if (attr_flags & MOUNT_ATTR_NOEXEC)
3480 mnt_flags |= MNT_NOEXEC;
3481 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3482 mnt_flags |= MNT_NODIRATIME;
3483
3484 switch (attr_flags & MOUNT_ATTR__ATIME) {
3485 case MOUNT_ATTR_STRICTATIME:
3486 break;
3487 case MOUNT_ATTR_NOATIME:
3488 mnt_flags |= MNT_NOATIME;
3489 break;
3490 case MOUNT_ATTR_RELATIME:
3491 mnt_flags |= MNT_RELATIME;
3492 break;
3493 default:
3494 return -EINVAL;
3495 }
3496
3497 f = fdget(fs_fd);
3498 if (!f.file)
3499 return -EBADF;
3500
3501 ret = -EINVAL;
3502 if (f.file->f_op != &fscontext_fops)
3503 goto err_fsfd;
3504
3505 fc = f.file->private_data;
3506
3507 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3508 if (ret < 0)
3509 goto err_fsfd;
3510
3511 /* There must be a valid superblock or we can't mount it */
3512 ret = -EINVAL;
3513 if (!fc->root)
3514 goto err_unlock;
3515
3516 ret = -EPERM;
3517 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3518 pr_warn("VFS: Mount too revealing\n");
3519 goto err_unlock;
3520 }
3521
3522 ret = -EBUSY;
3523 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3524 goto err_unlock;
3525
3526 ret = -EPERM;
3527 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3528 goto err_unlock;
3529
3530 newmount.mnt = vfs_create_mount(fc);
3531 if (IS_ERR(newmount.mnt)) {
3532 ret = PTR_ERR(newmount.mnt);
3533 goto err_unlock;
3534 }
3535 newmount.dentry = dget(fc->root);
3536 newmount.mnt->mnt_flags = mnt_flags;
3537
3538 /* We've done the mount bit - now move the file context into more or
3539 * less the same state as if we'd done an fspick(). We don't want to
3540 * do any memory allocation or anything like that at this point as we
3541 * don't want to have to handle any errors incurred.
3542 */
3543 vfs_clean_context(fc);
3544
3545 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3546 if (IS_ERR(ns)) {
3547 ret = PTR_ERR(ns);
3548 goto err_path;
3549 }
3550 mnt = real_mount(newmount.mnt);
3551 mnt->mnt_ns = ns;
3552 ns->root = mnt;
3553 ns->mounts = 1;
3554 list_add(&mnt->mnt_list, &ns->list);
Eric Biggers1b0b9cc2019-06-12 11:43:13 -07003555 mntget(newmount.mnt);
David Howells93766fb2018-11-01 23:36:14 +00003556
3557 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3558 * it, not just simply put it.
3559 */
3560 file = dentry_open(&newmount, O_PATH, fc->cred);
3561 if (IS_ERR(file)) {
3562 dissolve_on_fput(newmount.mnt);
3563 ret = PTR_ERR(file);
3564 goto err_path;
3565 }
3566 file->f_mode |= FMODE_NEED_UNMOUNT;
3567
3568 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3569 if (ret >= 0)
3570 fd_install(ret, file);
3571 else
3572 fput(file);
3573
3574err_path:
3575 path_put(&newmount);
3576err_unlock:
3577 mutex_unlock(&fc->uapi_mutex);
3578err_fsfd:
3579 fdput(f);
3580 return ret;
3581}
3582
3583/*
3584 * Move a mount from one place to another. In combination with
3585 * fsopen()/fsmount() this is used to install a new mount and in combination
3586 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3587 * a mount subtree.
David Howells2db154b2018-11-05 17:40:30 +00003588 *
3589 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3590 */
3591SYSCALL_DEFINE5(move_mount,
Ben Dooks2658ce02019-10-15 11:35:02 +01003592 int, from_dfd, const char __user *, from_pathname,
3593 int, to_dfd, const char __user *, to_pathname,
David Howells2db154b2018-11-05 17:40:30 +00003594 unsigned int, flags)
3595{
3596 struct path from_path, to_path;
3597 unsigned int lflags;
3598 int ret = 0;
3599
3600 if (!may_mount())
3601 return -EPERM;
3602
3603 if (flags & ~MOVE_MOUNT__MASK)
3604 return -EINVAL;
3605
3606 /* If someone gives a pathname, they aren't permitted to move
3607 * from an fd that requires unmount as we can't get at the flag
3608 * to clear it afterwards.
3609 */
3610 lflags = 0;
3611 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3612 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3613 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3614
3615 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3616 if (ret < 0)
3617 return ret;
3618
3619 lflags = 0;
3620 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3621 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3622 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3623
3624 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3625 if (ret < 0)
3626 goto out_from;
3627
3628 ret = security_move_mount(&from_path, &to_path);
3629 if (ret < 0)
3630 goto out_to;
3631
3632 ret = do_move_mount(&from_path, &to_path);
3633
3634out_to:
3635 path_put(&to_path);
3636out_from:
3637 path_put(&from_path);
3638 return ret;
3639}
3640
3641/*
Al Viroafac7cb2011-11-23 19:34:49 -05003642 * Return true if path is reachable from root
3643 *
Al Viro48a066e2013-09-29 22:06:07 -04003644 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05003645 */
Al Viro643822b2011-11-24 22:00:28 -05003646bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05003647 const struct path *root)
3648{
Al Viro643822b2011-11-24 22:00:28 -05003649 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05003650 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05003651 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05003652 }
Al Viro643822b2011-11-24 22:00:28 -05003653 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05003654}
3655
Mickaël Salaün640eb7e2016-11-14 22:14:35 +01003656bool path_is_under(const struct path *path1, const struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05003657{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003658 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04003659 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05003660 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04003661 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05003662 return res;
3663}
3664EXPORT_SYMBOL(path_is_under);
3665
3666/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667 * pivot_root Semantics:
3668 * Moves the root file system of the current process to the directory put_old,
3669 * makes new_root as the new root file system of the current process, and sets
3670 * root/cwd of all processes which had them on the current root to new_root.
3671 *
3672 * Restrictions:
3673 * The new_root and put_old must be directories, and must not be on the
3674 * same file system as the current process root. The put_old must be
3675 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3676 * pointed to by put_old must yield the same directory as new_root. No other
3677 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3678 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08003679 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
Mauro Carvalho Chehab0c1bc6b2020-04-14 18:48:37 +02003680 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
Neil Brown4a0d11f2006-01-08 01:03:18 -08003681 * in this situation.
3682 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 * Notes:
3684 * - we don't move root/cwd if they are not at the root (reason: if something
3685 * cared enough to change them, it's probably wrong to force them elsewhere)
3686 * - it's okay to pick a root that isn't the root of a file system, e.g.
3687 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3688 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3689 * first.
3690 */
Heiko Carstens3480b252009-01-14 14:14:16 +01003691SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3692 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693{
Al Viro2763d112019-06-30 19:18:53 -04003694 struct path new, old, root;
3695 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
Al Viro84d171922013-03-15 10:53:28 -04003696 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697 int error;
3698
Al Viro9b40bc92013-02-22 22:45:42 -05003699 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700 return -EPERM;
3701
Al Viroce6595a2019-07-14 16:42:44 -04003702 error = user_path_at(AT_FDCWD, new_root,
3703 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 if (error)
3705 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706
Al Viroce6595a2019-07-14 16:42:44 -04003707 error = user_path_at(AT_FDCWD, put_old,
3708 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 if (error)
3710 goto out1;
3711
Al Viro2d8f3032008-07-22 09:59:21 -04003712 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04003713 if (error)
3714 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003716 get_fs_root(current->fs, &root);
Al Viro84d171922013-03-15 10:53:28 -04003717 old_mp = lock_mount(&old);
3718 error = PTR_ERR(old_mp);
3719 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04003720 goto out3;
3721
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05003723 new_mnt = real_mount(new.mnt);
3724 root_mnt = real_mount(root.mnt);
Al Viro84d171922013-03-15 10:53:28 -04003725 old_mnt = real_mount(old.mnt);
Al Viro2763d112019-06-30 19:18:53 -04003726 ex_parent = new_mnt->mnt_parent;
3727 root_parent = root_mnt->mnt_parent;
Al Viro84d171922013-03-15 10:53:28 -04003728 if (IS_MNT_SHARED(old_mnt) ||
Al Viro2763d112019-06-30 19:18:53 -04003729 IS_MNT_SHARED(ex_parent) ||
3730 IS_MNT_SHARED(root_parent))
Al Virob12cea92011-03-18 08:55:38 -04003731 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05003732 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003733 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003734 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3735 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003737 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04003738 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 error = -EBUSY;
Al Viro84d171922013-03-15 10:53:28 -04003740 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04003741 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04003743 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003744 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003745 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003746 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04003747 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04003748 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05003749 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04003750 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08003751 /* make sure we can reach put_old from new_root */
Al Viro84d171922013-03-15 10:53:28 -04003752 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04003753 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07003754 /* make certain new is below the root */
3755 if (!is_path_reachable(new_mnt, new.dentry, &root))
3756 goto out4;
Al Viro719ea2f2013-09-29 11:24:49 -04003757 lock_mount_hash();
Al Viro2763d112019-06-30 19:18:53 -04003758 umount_mnt(new_mnt);
3759 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07003760 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3761 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3762 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3763 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003764 /* mount old root on put_old */
Al Viro84d171922013-03-15 10:53:28 -04003765 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08003766 /* mount new_root on / */
Al Viro2763d112019-06-30 19:18:53 -04003767 attach_mnt(new_mnt, root_parent, root_mp);
3768 mnt_add_count(root_parent, -1);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003769 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07003770 /* A moved mount should not expire automatically */
3771 list_del_init(&new_mnt->mnt_expire);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13003772 put_mountpoint(root_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04003773 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04003774 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04003776out4:
Al Viro84d171922013-03-15 10:53:28 -04003777 unlock_mount(old_mp);
Al Viro2763d112019-06-30 19:18:53 -04003778 if (!error)
3779 mntput_no_expire(ex_parent);
Al Virob12cea92011-03-18 08:55:38 -04003780out3:
Al Viro8c3ee422008-03-22 18:00:39 -04003781 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04003782out2:
Al Viro2d8f3032008-07-22 09:59:21 -04003783 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784out1:
Al Viro2d8f3032008-07-22 09:59:21 -04003785 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788}
3789
3790static void __init init_mount_tree(void)
3791{
3792 struct vfsmount *mnt;
Al Viro74e83122019-01-30 13:30:21 -05003793 struct mount *m;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003794 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08003795 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796
Al Virofd3e0072019-05-30 17:48:35 -04003797 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 if (IS_ERR(mnt))
3799 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11003800
Al Viro74e83122019-01-30 13:30:21 -05003801 ns = alloc_mnt_ns(&init_user_ns, false);
Trond Myklebust3b22edc2009-06-23 17:29:49 -04003802 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 panic("Can't allocate initial namespace");
Al Viro74e83122019-01-30 13:30:21 -05003804 m = real_mount(mnt);
3805 m->mnt_ns = ns;
3806 ns->root = m;
3807 ns->mounts = 1;
3808 list_add(&m->mnt_list, &ns->list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003809 init_task.nsproxy->mnt_ns = ns;
3810 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811
Al Virobe08d6d2011-12-06 13:32:36 -05003812 root.mnt = mnt;
3813 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07003814 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08003815
3816 set_fs_pwd(current->fs, &root);
3817 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818}
3819
Denis Cheng74bf17c2007-10-16 23:26:30 -07003820void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003821{
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003822 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
Al Viro7d6fec42011-11-23 12:14:10 -05003824 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Paul Mundt20c2df82007-07-20 10:11:58 +09003825 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003826
Al Viro0818bf22014-02-28 13:46:44 -05003827 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04003828 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05003829 mhash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003830 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05003831 &m_hash_shift, &m_hash_mask, 0, 0);
3832 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3833 sizeof(struct hlist_head),
3834 mphash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003835 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05003836 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837
Al Viro84d171922013-03-15 10:53:28 -04003838 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 panic("Failed to allocate mount hash table\n");
3840
Tejun Heo4b93dc92013-11-28 14:54:43 -05003841 kernfs_init();
3842
Randy Dunlap15a67dd2006-09-29 01:58:57 -07003843 err = sysfs_init();
3844 if (err)
3845 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003846 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06003847 fs_kobj = kobject_create_and_add("fs", NULL);
3848 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07003849 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Al Viro037f11b2019-06-01 18:09:44 -04003850 shmem_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851 init_rootfs();
3852 init_mount_tree();
3853}
3854
Trond Myklebust616511d2009-06-22 15:09:13 -04003855void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856{
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03003857 if (!refcount_dec_and_test(&ns->ns.count))
Trond Myklebust616511d2009-06-22 15:09:13 -04003858 return;
Al Viro7b00ed62013-09-16 21:19:20 -04003859 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003860 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861}
Al Viro9d412a42011-03-17 22:08:28 -04003862
David Howellsd911b452018-11-01 23:07:26 +00003863struct vfsmount *kern_mount(struct file_system_type *type)
Al Viro9d412a42011-03-17 22:08:28 -04003864{
Tim Chen423e0ab2011-07-19 09:32:38 -07003865 struct vfsmount *mnt;
David Howellsd911b452018-11-01 23:07:26 +00003866 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
Tim Chen423e0ab2011-07-19 09:32:38 -07003867 if (!IS_ERR(mnt)) {
3868 /*
3869 * it is a longterm mount, don't release mnt until
3870 * we unmount before file sys is unregistered
3871 */
Al Virof7a99c52012-06-09 00:59:08 -04003872 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07003873 }
3874 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04003875}
David Howellsd911b452018-11-01 23:07:26 +00003876EXPORT_SYMBOL_GPL(kern_mount);
Tim Chen423e0ab2011-07-19 09:32:38 -07003877
3878void kern_unmount(struct vfsmount *mnt)
3879{
3880 /* release long term mount so mount point can be released */
3881 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04003882 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04003883 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07003884 mntput(mnt);
3885 }
3886}
3887EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05003888
Miklos Szeredidf820f82020-06-04 10:48:19 +02003889void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
3890{
3891 unsigned int i;
3892
3893 for (i = 0; i < num; i++)
3894 if (mnt[i])
3895 real_mount(mnt[i])->mnt_ns = NULL;
3896 synchronize_rcu_expedited();
3897 for (i = 0; i < num; i++)
3898 mntput(mnt[i]);
3899}
3900EXPORT_SYMBOL(kern_unmount_array);
3901
Al Viro02125a82011-12-05 08:43:34 -05003902bool our_mnt(struct vfsmount *mnt)
3903{
Al Viro143c8c92011-11-25 00:46:35 -05003904 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05003905}
Eric W. Biederman8823c072010-03-07 18:49:36 -08003906
Eric W. Biederman31515272013-03-15 01:45:51 -07003907bool current_chrooted(void)
3908{
3909 /* Does the current process have a non-standard root */
3910 struct path ns_root;
3911 struct path fs_root;
3912 bool chrooted;
3913
3914 /* Find the namespace root */
3915 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3916 ns_root.dentry = ns_root.mnt->mnt_root;
3917 path_get(&ns_root);
3918 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3919 ;
3920
3921 get_fs_root(current->fs, &fs_root);
3922
3923 chrooted = !path_equal(&fs_root, &ns_root);
3924
3925 path_put(&fs_root);
3926 path_put(&ns_root);
3927
3928 return chrooted;
3929}
3930
David Howells132e4602018-11-04 07:43:08 -05003931static bool mnt_already_visible(struct mnt_namespace *ns,
3932 const struct super_block *sb,
Eric W. Biederman8654df42016-06-09 16:06:06 -05003933 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003934{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003935 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003936 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07003937 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003938
Al Viro44bb4382013-09-16 21:37:36 -04003939 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02003940 lock_ns_list(ns);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003941 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07003942 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003943 int mnt_flags;
3944
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02003945 if (mnt_is_cursor(mnt))
3946 continue;
3947
David Howells132e4602018-11-04 07:43:08 -05003948 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07003949 continue;
3950
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05003951 /* This mount is not fully visible if it's root directory
3952 * is not the root directory of the filesystem.
3953 */
3954 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3955 continue;
3956
Eric W. Biedermana1935c12016-06-15 06:59:49 -05003957 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003958 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003959
Eric W. Biederman695e9df2016-06-10 12:21:40 -05003960 /* Don't miss readonly hidden in the superblock flags */
David Howellsbc98a422017-07-17 08:45:34 +01003961 if (sb_rdonly(mnt->mnt.mnt_sb))
Eric W. Biederman695e9df2016-06-10 12:21:40 -05003962 mnt_flags |= MNT_LOCK_READONLY;
3963
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003964 /* Verify the mount flags are equal to or more permissive
3965 * than the proposed new mount.
3966 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003967 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003968 !(new_flags & MNT_READONLY))
3969 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003970 if ((mnt_flags & MNT_LOCK_ATIME) &&
3971 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003972 continue;
3973
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003974 /* This mount is not fully visible if there are any
3975 * locked child mounts that cover anything except for
3976 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07003977 */
3978 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3979 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003980 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05003981 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06003982 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05003983 /* Is the directory permanetly empty? */
3984 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07003985 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003986 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05003987 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003988 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05003989 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003990 visible = true;
3991 goto found;
3992 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003993 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07003994found:
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02003995 unlock_ns_list(ns);
Al Viro44bb4382013-09-16 21:37:36 -04003996 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07003997 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07003998}
3999
David Howells132e4602018-11-04 07:43:08 -05004000static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
Eric W. Biederman8654df42016-06-09 16:06:06 -05004001{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05004002 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05004003 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4004 unsigned long s_iflags;
4005
4006 if (ns->user_ns == &init_user_ns)
4007 return false;
4008
4009 /* Can this filesystem be too revealing? */
David Howells132e4602018-11-04 07:43:08 -05004010 s_iflags = sb->s_iflags;
Eric W. Biederman8654df42016-06-09 16:06:06 -05004011 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4012 return false;
4013
Eric W. Biedermana1935c12016-06-15 06:59:49 -05004014 if ((s_iflags & required_iflags) != required_iflags) {
4015 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4016 required_iflags);
4017 return true;
4018 }
4019
David Howells132e4602018-11-04 07:43:08 -05004020 return !mnt_already_visible(ns, sb, new_mnt_flags);
Eric W. Biederman8654df42016-06-09 16:06:06 -05004021}
4022
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05004023bool mnt_may_suid(struct vfsmount *mnt)
4024{
4025 /*
4026 * Foreign mounts (accessed via fchdir or through /proc
4027 * symlinks) are always treated as if they are nosuid. This
4028 * prevents namespaces from trusting potentially unsafe
4029 * suid/sgid bits, file caps, or security labels that originate
4030 * in other namespaces.
4031 */
4032 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4033 current_in_userns(mnt->mnt_sb->s_user_ns);
4034}
4035
Al Viro64964522014-11-01 00:37:32 -04004036static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004037{
Al Viro58be28252014-11-01 00:00:23 -04004038 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004039 struct nsproxy *nsproxy;
4040
Eric W. Biederman728dba32014-02-03 19:13:49 -08004041 task_lock(task);
4042 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004043 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04004044 ns = &nsproxy->mnt_ns->ns;
4045 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08004046 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08004047 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08004048
4049 return ns;
4050}
4051
Al Viro64964522014-11-01 00:37:32 -04004052static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004053{
Al Viro58be28252014-11-01 00:00:23 -04004054 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08004055}
4056
Christian Braunerf2a8d522020-05-05 16:04:30 +02004057static int mntns_install(struct nsset *nsset, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004058{
Christian Braunerf2a8d522020-05-05 16:04:30 +02004059 struct nsproxy *nsproxy = nsset->nsproxy;
4060 struct fs_struct *fs = nsset->fs;
Al Viro4f757f32017-04-15 17:31:22 -04004061 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
Christian Braunerf2a8d522020-05-05 16:04:30 +02004062 struct user_namespace *user_ns = nsset->cred->user_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004063 struct path root;
Al Viro4f757f32017-04-15 17:31:22 -04004064 int err;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004065
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07004066 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Christian Braunerf2a8d522020-05-05 16:04:30 +02004067 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4068 !ns_capable(user_ns, CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08004069 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004070
Al Viro74e83122019-01-30 13:30:21 -05004071 if (is_anon_ns(mnt_ns))
4072 return -EINVAL;
4073
Eric W. Biederman8823c072010-03-07 18:49:36 -08004074 if (fs->users != 1)
4075 return -EINVAL;
4076
4077 get_mnt_ns(mnt_ns);
Al Viro4f757f32017-04-15 17:31:22 -04004078 old_mnt_ns = nsproxy->mnt_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004079 nsproxy->mnt_ns = mnt_ns;
4080
4081 /* Find the root */
Al Viro4f757f32017-04-15 17:31:22 -04004082 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4083 "/", LOOKUP_DOWN, &root);
4084 if (err) {
4085 /* revert to old namespace */
4086 nsproxy->mnt_ns = old_mnt_ns;
4087 put_mnt_ns(mnt_ns);
4088 return err;
4089 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08004090
Andrei Vagin40683672017-06-08 17:32:29 -07004091 put_mnt_ns(old_mnt_ns);
4092
Eric W. Biederman8823c072010-03-07 18:49:36 -08004093 /* Update the pwd and root */
4094 set_fs_pwd(fs, &root);
4095 set_fs_root(fs, &root);
4096
4097 path_put(&root);
4098 return 0;
4099}
4100
Andrey Vaginbcac25a2016-09-06 00:47:13 -07004101static struct user_namespace *mntns_owner(struct ns_common *ns)
4102{
4103 return to_mnt_ns(ns)->user_ns;
4104}
4105
Eric W. Biederman8823c072010-03-07 18:49:36 -08004106const struct proc_ns_operations mntns_operations = {
4107 .name = "mnt",
4108 .type = CLONE_NEWNS,
4109 .get = mntns_get,
4110 .put = mntns_put,
4111 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07004112 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08004113};