blob: ab467ee5834115b4bea458ebcd5607047d542d95 [file] [log] [blame]
Thomas Gleixner59bd9de2019-05-28 10:10:12 -07001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
Al Virod10577a2011-12-07 13:06:11 -050012#include <linux/export.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080013#include <linux/capability.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080014#include <linux/mnt_namespace.h>
Eric W. Biederman771b1372012-07-26 21:08:32 -070015#include <linux/user_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/namei.h>
17#include <linux/security.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +010018#include <linux/cred.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010019#include <linux/idr.h>
Rob Landley57f150a2013-09-11 14:26:10 -070020#include <linux/init.h> /* init_rootfs */
Al Virod10577a2011-12-07 13:06:11 -050021#include <linux/fs_struct.h> /* get_fs_root et.al. */
22#include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
Al Viroa07b2002018-11-05 17:40:30 +000023#include <linux/file.h>
Al Virod10577a2011-12-07 13:06:11 -050024#include <linux/uaccess.h>
David Howells0bb80f22013-04-12 01:50:06 +010025#include <linux/proc_ns.h>
Linus Torvalds20b4fb42013-05-01 17:51:54 -070026#include <linux/magic.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070027#include <linux/memblock.h>
Christian Brauner9caccd42021-01-21 14:19:54 +010028#include <linux/proc_fs.h>
Al Viro9ea459e12014-08-08 13:08:20 -040029#include <linux/task_work.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010030#include <linux/sched/task.h>
David Howellse262e32d2018-11-01 23:07:23 +000031#include <uapi/linux/mount.h>
David Howells9bc61ab2018-11-04 03:19:03 -050032#include <linux/fs_context.h>
Al Viro037f11b2019-06-01 18:09:44 -040033#include <linux/shmem_fs.h>
Christian Braunerbd303362021-12-03 12:17:07 +010034#include <linux/mnt_idmapping.h>
Ingo Molnar9164bb42017-02-04 01:20:53 +010035
Ram Pai07b20882005-11-07 17:19:07 -050036#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070037#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Eric W. Biedermand2921682016-09-28 00:27:17 -050039/* Maximum number of mounts in a mount namespace */
Luis Chamberlainab171b92022-01-21 22:13:27 -080040static unsigned int sysctl_mount_max __read_mostly = 100000;
Eric W. Biedermand2921682016-09-28 00:27:17 -050041
Al Viro0818bf22014-02-28 13:46:44 -050042static unsigned int m_hash_mask __read_mostly;
43static unsigned int m_hash_shift __read_mostly;
44static unsigned int mp_hash_mask __read_mostly;
45static unsigned int mp_hash_shift __read_mostly;
46
47static __initdata unsigned long mhash_entries;
48static int __init set_mhash_entries(char *str)
49{
50 if (!str)
51 return 0;
52 mhash_entries = simple_strtoul(str, &str, 0);
53 return 1;
54}
55__setup("mhash_entries=", set_mhash_entries);
56
57static __initdata unsigned long mphash_entries;
58static int __init set_mphash_entries(char *str)
59{
60 if (!str)
61 return 0;
62 mphash_entries = simple_strtoul(str, &str, 0);
63 return 1;
64}
65__setup("mphash_entries=", set_mphash_entries);
Eric Dumazet13f14b42008-02-06 01:37:57 -080066
Al Viroc7999c32014-02-27 14:40:10 -050067static u64 event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010068static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010069static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Al Viro38129a12014-03-20 21:10:51 -040071static struct hlist_head *mount_hashtable __read_mostly;
Al Viro0818bf22014-02-28 13:46:44 -050072static struct hlist_head *mountpoint_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080073static struct kmem_cache *mnt_cache __read_mostly;
Al Viro59aa0da2013-09-16 21:34:53 -040074static DECLARE_RWSEM(namespace_sem);
Al Viro4edbe132019-06-30 10:39:08 -040075static HLIST_HEAD(unmounted); /* protected by namespace_sem */
76static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Christian Brauner256c8ae2022-10-26 12:51:27 +020078struct mnt_idmap {
79 struct user_namespace *owner;
80 refcount_t count;
81};
82
83/*
84 * Carries the initial idmapping of 0:0:4294967295 which is an identity
85 * mapping. This means that {g,u}id 0 is mapped to {g,u}id 0, {g,u}id 1 is
86 * mapped to {g,u}id 1, [...], {g,u}id 1000 to {g,u}id 1000, [...].
87 */
88struct mnt_idmap nop_mnt_idmap = {
89 .owner = &init_user_ns,
90 .count = REFCOUNT_INIT(1),
91};
92EXPORT_SYMBOL_GPL(nop_mnt_idmap);
93
Christian Brauner2a186722021-01-21 14:19:53 +010094struct mount_kattr {
95 unsigned int attr_set;
96 unsigned int attr_clr;
97 unsigned int propagation;
98 unsigned int lookup_flags;
99 bool recurse;
Christian Brauner9caccd42021-01-21 14:19:54 +0100100 struct user_namespace *mnt_userns;
Christian Brauner256c8ae2022-10-26 12:51:27 +0200101 struct mnt_idmap *mnt_idmap;
Christian Brauner2a186722021-01-21 14:19:53 +0100102};
103
Miklos Szeredif87fd4c2006-01-16 22:14:23 -0800104/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -0600105struct kobject *fs_kobj;
106EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -0800107
Nick Piggin99b7db72010-08-18 04:37:39 +1000108/*
109 * vfsmount lock may be taken for read to prevent changes to the
110 * vfsmount hash, ie. during mountpoint lookups or walking back
111 * up the tree.
112 *
113 * It should be taken for write in all cases where the vfsmount
114 * tree or hash is modified or when a vfsmount structure is modified.
115 */
Al Viro48a066e2013-09-29 22:06:07 -0400116__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
Nick Piggin99b7db72010-08-18 04:37:39 +1000117
Christian Braunerd033cb62021-01-21 14:19:49 +0100118static inline void lock_mount_hash(void)
119{
120 write_seqlock(&mount_lock);
121}
122
123static inline void unlock_mount_hash(void)
124{
125 write_sequnlock(&mount_lock);
126}
127
Al Viro38129a12014-03-20 21:10:51 -0400128static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
Ram Paib58fed82005-11-07 17:16:09 -0500130 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
131 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Al Viro0818bf22014-02-28 13:46:44 -0500132 tmp = tmp + (tmp >> m_hash_shift);
133 return &mount_hashtable[tmp & m_hash_mask];
134}
135
136static inline struct hlist_head *mp_hash(struct dentry *dentry)
137{
138 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
139 tmp = tmp + (tmp >> mp_hash_shift);
140 return &mountpoint_hashtable[tmp & mp_hash_mask];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Al Virob105e272011-11-24 20:38:33 -0500143static int mnt_alloc_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100144{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400145 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100146
Matthew Wilcox169b4802018-06-11 12:31:36 -0400147 if (res < 0)
148 return res;
149 mnt->mnt_id = res;
150 return 0;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100151}
152
Al Virob105e272011-11-24 20:38:33 -0500153static void mnt_free_id(struct mount *mnt)
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100154{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400155 ida_free(&mnt_id_ida, mnt->mnt_id);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100156}
157
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100158/*
159 * Allocate a new peer group ID
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100160 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500161static int mnt_alloc_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100162{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400163 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
Al Virof21f6222009-06-24 03:12:00 -0400164
Matthew Wilcox169b4802018-06-11 12:31:36 -0400165 if (res < 0)
166 return res;
167 mnt->mnt_group_id = res;
168 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100169}
170
171/*
172 * Release a peer group ID
173 */
Al Viro4b8b21f2011-11-24 19:54:23 -0500174void mnt_release_group_id(struct mount *mnt)
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100175{
Matthew Wilcox169b4802018-06-11 12:31:36 -0400176 ida_free(&mnt_group_ida, mnt->mnt_group_id);
Al Viro15169fe2011-11-25 00:50:41 -0500177 mnt->mnt_group_id = 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100178}
179
Nick Pigginb3e19d92011-01-07 17:50:11 +1100180/*
181 * vfsmount lock must be held for read
182 */
Al Viro83adc752011-11-24 22:37:54 -0500183static inline void mnt_add_count(struct mount *mnt, int n)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100184{
185#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500186 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100187#else
188 preempt_disable();
Al Viro68e8a9f2011-11-24 22:53:09 -0500189 mnt->mnt_count += n;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100190 preempt_enable();
191#endif
192}
193
Nick Pigginb3e19d92011-01-07 17:50:11 +1100194/*
195 * vfsmount lock must be held for write
196 */
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700197int mnt_get_count(struct mount *mnt)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100198{
199#ifdef CONFIG_SMP
Eric Biggersedf7ddb2020-10-31 21:40:21 -0700200 int count = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100201 int cpu;
202
203 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500204 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100205 }
206
207 return count;
208#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500209 return mnt->mnt_count;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100210#endif
211}
212
Christian Brauner256c8ae2022-10-26 12:51:27 +0200213/**
214 * mnt_idmap_owner - retrieve owner of the mount's idmapping
215 * @idmap: mount idmapping
216 *
217 * This helper will go away once the conversion to use struct mnt_idmap
218 * everywhere has finished at which point the helper will be unexported.
219 *
220 * Only code that needs to perform permission checks based on the owner of the
221 * idmapping will get access to it. All other code will solely rely on
222 * idmappings. This will get us type safety so it's impossible to conflate
223 * filesystems idmappings with mount idmappings.
224 *
225 * Return: The owner of the idmapping.
226 */
227struct user_namespace *mnt_idmap_owner(const struct mnt_idmap *idmap)
228{
229 return idmap->owner;
230}
231EXPORT_SYMBOL_GPL(mnt_idmap_owner);
232
233/**
234 * mnt_user_ns - retrieve owner of an idmapped mount
235 * @mnt: the relevant vfsmount
236 *
237 * This helper will go away once the conversion to use struct mnt_idmap
238 * everywhere has finished at which point the helper will be unexported.
239 *
240 * Only code that needs to perform permission checks based on the owner of the
241 * idmapping will get access to it. All other code will solely rely on
242 * idmappings. This will get us type safety so it's impossible to conflate
243 * filesystems idmappings with mount idmappings.
244 *
245 * Return: The owner of the idmapped.
246 */
247struct user_namespace *mnt_user_ns(const struct vfsmount *mnt)
248{
249 struct mnt_idmap *idmap = mnt_idmap(mnt);
250
251 /* Return the actual owner of the filesystem instead of the nop. */
252 if (idmap == &nop_mnt_idmap &&
253 !initial_idmapping(mnt->mnt_sb->s_user_ns))
254 return mnt->mnt_sb->s_user_ns;
255 return mnt_idmap_owner(idmap);
256}
257EXPORT_SYMBOL_GPL(mnt_user_ns);
258
259/**
260 * alloc_mnt_idmap - allocate a new idmapping for the mount
261 * @mnt_userns: owning userns of the idmapping
262 *
263 * Allocate a new struct mnt_idmap which carries the idmapping of the mount.
264 *
265 * Return: On success a new idmap, on error an error pointer is returned.
266 */
267static struct mnt_idmap *alloc_mnt_idmap(struct user_namespace *mnt_userns)
268{
269 struct mnt_idmap *idmap;
270
271 idmap = kzalloc(sizeof(struct mnt_idmap), GFP_KERNEL_ACCOUNT);
272 if (!idmap)
273 return ERR_PTR(-ENOMEM);
274
275 idmap->owner = get_user_ns(mnt_userns);
276 refcount_set(&idmap->count, 1);
277 return idmap;
278}
279
280/**
281 * mnt_idmap_get - get a reference to an idmapping
282 * @idmap: the idmap to bump the reference on
283 *
284 * If @idmap is not the @nop_mnt_idmap bump the reference count.
285 *
286 * Return: @idmap with reference count bumped if @not_mnt_idmap isn't passed.
287 */
288static inline struct mnt_idmap *mnt_idmap_get(struct mnt_idmap *idmap)
289{
290 if (idmap != &nop_mnt_idmap)
291 refcount_inc(&idmap->count);
292
293 return idmap;
294}
295
296/**
297 * mnt_idmap_put - put a reference to an idmapping
298 * @idmap: the idmap to put the reference on
299 *
300 * If this is a non-initial idmapping, put the reference count when a mount is
301 * released and free it if we're the last user.
302 */
303static inline void mnt_idmap_put(struct mnt_idmap *idmap)
304{
305 if (idmap != &nop_mnt_idmap && refcount_dec_and_test(&idmap->count)) {
306 put_user_ns(idmap->owner);
307 kfree(idmap);
308 }
309}
310
Al Virob105e272011-11-24 20:38:33 -0500311static struct mount *alloc_vfsmnt(const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Al Viroc63181e2011-11-25 02:35:16 -0500313 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
314 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100315 int err;
316
Al Viroc63181e2011-11-25 02:35:16 -0500317 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800318 if (err)
319 goto out_free_cache;
320
321 if (name) {
Vasily Averin79f65402021-09-02 14:55:10 -0700322 mnt->mnt_devname = kstrdup_const(name,
323 GFP_KERNEL_ACCOUNT);
Al Viroc63181e2011-11-25 02:35:16 -0500324 if (!mnt->mnt_devname)
Li Zefan88b38782008-07-21 18:06:36 +0800325 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100326 }
327
Nick Pigginb3e19d92011-01-07 17:50:11 +1100328#ifdef CONFIG_SMP
Al Viroc63181e2011-11-25 02:35:16 -0500329 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
330 if (!mnt->mnt_pcp)
Nick Pigginb3e19d92011-01-07 17:50:11 +1100331 goto out_free_devname;
332
Al Viroc63181e2011-11-25 02:35:16 -0500333 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +1100334#else
Al Viroc63181e2011-11-25 02:35:16 -0500335 mnt->mnt_count = 1;
336 mnt->mnt_writers = 0;
Nick Pigginb3e19d92011-01-07 17:50:11 +1100337#endif
338
Al Viro38129a12014-03-20 21:10:51 -0400339 INIT_HLIST_NODE(&mnt->mnt_hash);
Al Viroc63181e2011-11-25 02:35:16 -0500340 INIT_LIST_HEAD(&mnt->mnt_child);
341 INIT_LIST_HEAD(&mnt->mnt_mounts);
342 INIT_LIST_HEAD(&mnt->mnt_list);
343 INIT_LIST_HEAD(&mnt->mnt_expire);
344 INIT_LIST_HEAD(&mnt->mnt_share);
345 INIT_LIST_HEAD(&mnt->mnt_slave_list);
346 INIT_LIST_HEAD(&mnt->mnt_slave);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700347 INIT_HLIST_NODE(&mnt->mnt_mp_list);
Eric W. Biederman99b19d12016-10-24 16:16:13 -0500348 INIT_LIST_HEAD(&mnt->mnt_umounting);
Al Viro56cbb4292019-07-04 16:57:51 -0400349 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
Christian Brauner256c8ae2022-10-26 12:51:27 +0200350 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 }
Al Viroc63181e2011-11-25 02:35:16 -0500352 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800353
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000354#ifdef CONFIG_SMP
355out_free_devname:
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800356 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000357#endif
Li Zefan88b38782008-07-21 18:06:36 +0800358out_free_id:
Al Viroc63181e2011-11-25 02:35:16 -0500359 mnt_free_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800360out_free_cache:
Al Viroc63181e2011-11-25 02:35:16 -0500361 kmem_cache_free(mnt_cache, mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800362 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Dave Hansen83660252008-02-15 14:37:30 -0800365/*
366 * Most r/o checks on a fs are for operations that take
367 * discrete amounts of time, like a write() or unlink().
368 * We must keep track of when those operations start
369 * (for permission checks) and when they end, so that
370 * we can determine when writes are able to occur to
371 * a filesystem.
372 */
Dave Hansen3d733632008-02-15 14:37:59 -0800373/*
374 * __mnt_is_readonly: check whether a mount is read-only
375 * @mnt: the mount to check for its write status
376 *
377 * This shouldn't be used directly ouside of the VFS.
378 * It does not guarantee that the filesystem will stay
379 * r/w, just that it is right *now*. This can not and
380 * should not be used in place of IS_RDONLY(inode).
381 * mnt_want/drop_write() will _keep_ the filesystem
382 * r/w.
383 */
David Howells43f5e652018-11-01 23:07:25 +0000384bool __mnt_is_readonly(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800385{
David Howells43f5e652018-11-01 23:07:25 +0000386 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800387}
388EXPORT_SYMBOL_GPL(__mnt_is_readonly);
389
Al Viro83adc752011-11-24 22:37:54 -0500390static inline void mnt_inc_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800391{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000392#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500393 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000394#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500395 mnt->mnt_writers++;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000396#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800397}
Dave Hansen3d733632008-02-15 14:37:59 -0800398
Al Viro83adc752011-11-24 22:37:54 -0500399static inline void mnt_dec_writers(struct mount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800400{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000401#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500402 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000403#else
Al Viro68e8a9f2011-11-24 22:53:09 -0500404 mnt->mnt_writers--;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000405#endif
406}
407
Al Viro83adc752011-11-24 22:37:54 -0500408static unsigned int mnt_get_writers(struct mount *mnt)
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000409{
410#ifdef CONFIG_SMP
411 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800412 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800413
414 for_each_possible_cpu(cpu) {
Al Viro68e8a9f2011-11-24 22:53:09 -0500415 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
Dave Hansen3d733632008-02-15 14:37:59 -0800416 }
Dave Hansen3d733632008-02-15 14:37:59 -0800417
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000418 return count;
419#else
420 return mnt->mnt_writers;
421#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800422}
423
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100424static int mnt_is_readonly(struct vfsmount *mnt)
425{
426 if (mnt->mnt_sb->s_readonly_remount)
427 return 1;
428 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
429 smp_rmb();
430 return __mnt_is_readonly(mnt);
431}
432
Dave Hansen3d733632008-02-15 14:37:59 -0800433/*
Jan Karaeb04c282012-06-12 16:20:35 +0200434 * Most r/o & frozen checks on a fs are for operations that take discrete
435 * amounts of time, like a write() or unlink(). We must keep track of when
436 * those operations start (for permission checks) and when they end, so that we
437 * can determine when writes are able to occur to a filesystem.
Dave Hansen3d733632008-02-15 14:37:59 -0800438 */
Dave Hansen83660252008-02-15 14:37:30 -0800439/**
Jan Karaeb04c282012-06-12 16:20:35 +0200440 * __mnt_want_write - get write access to a mount without freeze protection
Al Viro83adc752011-11-24 22:37:54 -0500441 * @m: the mount on which to take a write
Dave Hansen83660252008-02-15 14:37:30 -0800442 *
Jan Karaeb04c282012-06-12 16:20:35 +0200443 * This tells the low-level filesystem that a write is about to be performed to
444 * it, and makes sure that writes are allowed (mnt it read-write) before
445 * returning success. This operation does not protect against filesystem being
446 * frozen. When the write operation is finished, __mnt_drop_write() must be
447 * called. This is effectively a refcount.
Dave Hansen83660252008-02-15 14:37:30 -0800448 */
Jan Karaeb04c282012-06-12 16:20:35 +0200449int __mnt_want_write(struct vfsmount *m)
Dave Hansen83660252008-02-15 14:37:30 -0800450{
Al Viro83adc752011-11-24 22:37:54 -0500451 struct mount *mnt = real_mount(m);
Dave Hansen3d733632008-02-15 14:37:59 -0800452 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800453
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000454 preempt_disable();
Nick Pigginc6653a82011-01-07 17:50:10 +1100455 mnt_inc_writers(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000456 /*
Nick Pigginc6653a82011-01-07 17:50:10 +1100457 * The store to mnt_inc_writers must be visible before we pass
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000458 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
459 * incremented count after it has set MNT_WRITE_HOLD.
460 */
461 smp_mb();
Sebastian Andrzej Siewior0f8821d2021-11-25 13:07:11 +0100462 might_lock(&mount_lock.lock);
463 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
464 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
465 cpu_relax();
466 } else {
467 /*
468 * This prevents priority inversion, if the task
469 * setting MNT_WRITE_HOLD got preempted on a remote
470 * CPU, and it prevents life lock if the task setting
471 * MNT_WRITE_HOLD has a lower priority and is bound to
472 * the same CPU as the task that is spinning here.
473 */
474 preempt_enable();
475 lock_mount_hash();
476 unlock_mount_hash();
477 preempt_disable();
478 }
479 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000480 /*
481 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
482 * be set to match its requirements. So we must not load that until
483 * MNT_WRITE_HOLD is cleared.
484 */
485 smp_rmb();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100486 if (mnt_is_readonly(m)) {
Nick Pigginc6653a82011-01-07 17:50:10 +1100487 mnt_dec_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800488 ret = -EROFS;
Dave Hansen3d733632008-02-15 14:37:59 -0800489 }
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000490 preempt_enable();
Jan Karaeb04c282012-06-12 16:20:35 +0200491
492 return ret;
493}
494
495/**
496 * mnt_want_write - get write access to a mount
497 * @m: the mount on which to take a write
498 *
499 * This tells the low-level filesystem that a write is about to be performed to
500 * it, and makes sure that writes are allowed (mount is read-write, filesystem
501 * is not frozen) before returning success. When the write operation is
502 * finished, mnt_drop_write() must be called. This is effectively a refcount.
503 */
504int mnt_want_write(struct vfsmount *m)
505{
506 int ret;
507
508 sb_start_write(m->mnt_sb);
509 ret = __mnt_want_write(m);
510 if (ret)
511 sb_end_write(m->mnt_sb);
Dave Hansen3d733632008-02-15 14:37:59 -0800512 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800513}
514EXPORT_SYMBOL_GPL(mnt_want_write);
515
516/**
Jan Karaeb04c282012-06-12 16:20:35 +0200517 * __mnt_want_write_file - get write access to a file's mount
518 * @file: the file who's mount on which to take a write
519 *
Eric Biggers14e43bf42020-09-22 09:44:18 -0700520 * This is like __mnt_want_write, but if the file is already open for writing it
521 * skips incrementing mnt_writers (since the open file already has a reference)
522 * and instead only does the check for emergency r/o remounts. This must be
523 * paired with __mnt_drop_write_file.
Jan Karaeb04c282012-06-12 16:20:35 +0200524 */
525int __mnt_want_write_file(struct file *file)
526{
Eric Biggers14e43bf42020-09-22 09:44:18 -0700527 if (file->f_mode & FMODE_WRITER) {
528 /*
529 * Superblock may have become readonly while there are still
530 * writable fd's, e.g. due to a fs error with errors=remount-ro
531 */
532 if (__mnt_is_readonly(file->f_path.mnt))
533 return -EROFS;
534 return 0;
535 }
536 return __mnt_want_write(file->f_path.mnt);
Jan Karaeb04c282012-06-12 16:20:35 +0200537}
538
539/**
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200540 * mnt_want_write_file - get write access to a file's mount
541 * @file: the file who's mount on which to take a write
542 *
Eric Biggers14e43bf42020-09-22 09:44:18 -0700543 * This is like mnt_want_write, but if the file is already open for writing it
544 * skips incrementing mnt_writers (since the open file already has a reference)
545 * and instead only does the freeze protection and the check for emergency r/o
546 * remounts. This must be paired with mnt_drop_write_file.
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200547 */
548int mnt_want_write_file(struct file *file)
549{
550 int ret;
551
Miklos Szeredia6795a52018-07-18 15:44:43 +0200552 sb_start_write(file_inode(file)->i_sb);
Jan Karaeb04c282012-06-12 16:20:35 +0200553 ret = __mnt_want_write_file(file);
554 if (ret)
Miklos Szeredia6795a52018-07-18 15:44:43 +0200555 sb_end_write(file_inode(file)->i_sb);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200556 return ret;
557}
npiggin@suse.de96029c42009-04-26 20:25:55 +1000558EXPORT_SYMBOL_GPL(mnt_want_write_file);
559
560/**
Jan Karaeb04c282012-06-12 16:20:35 +0200561 * __mnt_drop_write - give up write access to a mount
Dave Hansen83660252008-02-15 14:37:30 -0800562 * @mnt: the mount on which to give up write access
563 *
564 * Tells the low-level filesystem that we are done
565 * performing writes to it. Must be matched with
Jan Karaeb04c282012-06-12 16:20:35 +0200566 * __mnt_want_write() call above.
Dave Hansen83660252008-02-15 14:37:30 -0800567 */
Jan Karaeb04c282012-06-12 16:20:35 +0200568void __mnt_drop_write(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800569{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000570 preempt_disable();
Al Viro83adc752011-11-24 22:37:54 -0500571 mnt_dec_writers(real_mount(mnt));
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000572 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800573}
Jan Karaeb04c282012-06-12 16:20:35 +0200574
575/**
576 * mnt_drop_write - give up write access to a mount
577 * @mnt: the mount on which to give up write access
578 *
579 * Tells the low-level filesystem that we are done performing writes to it and
580 * also allows filesystem to be frozen again. Must be matched with
581 * mnt_want_write() call above.
582 */
583void mnt_drop_write(struct vfsmount *mnt)
584{
585 __mnt_drop_write(mnt);
586 sb_end_write(mnt->mnt_sb);
587}
Dave Hansen83660252008-02-15 14:37:30 -0800588EXPORT_SYMBOL_GPL(mnt_drop_write);
589
Jan Karaeb04c282012-06-12 16:20:35 +0200590void __mnt_drop_write_file(struct file *file)
591{
Eric Biggers14e43bf42020-09-22 09:44:18 -0700592 if (!(file->f_mode & FMODE_WRITER))
593 __mnt_drop_write(file->f_path.mnt);
Jan Karaeb04c282012-06-12 16:20:35 +0200594}
595
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200596void mnt_drop_write_file(struct file *file)
597{
Miklos Szeredia6795a52018-07-18 15:44:43 +0200598 __mnt_drop_write_file(file);
Miklos Szeredi7c6893e2017-09-05 12:53:12 +0200599 sb_end_write(file_inode(file)->i_sb);
600}
Al Viro2a79f172011-12-09 08:06:57 -0500601EXPORT_SYMBOL(mnt_drop_write_file);
602
Christian Brauner538f4f022022-02-03 14:14:07 +0100603/**
604 * mnt_hold_writers - prevent write access to the given mount
605 * @mnt: mnt to prevent write access to
606 *
607 * Prevents write access to @mnt if there are no active writers for @mnt.
608 * This function needs to be called and return successfully before changing
609 * properties of @mnt that need to remain stable for callers with write access
610 * to @mnt.
611 *
612 * After this functions has been called successfully callers must pair it with
613 * a call to mnt_unhold_writers() in order to stop preventing write access to
614 * @mnt.
615 *
616 * Context: This function expects lock_mount_hash() to be held serializing
617 * setting MNT_WRITE_HOLD.
618 * Return: On success 0 is returned.
619 * On error, -EBUSY is returned.
620 */
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100621static inline int mnt_hold_writers(struct mount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800622{
Al Viro83adc752011-11-24 22:37:54 -0500623 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000624 /*
625 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
626 * should be visible before we do.
627 */
628 smp_mb();
629
630 /*
631 * With writers on hold, if this value is zero, then there are
632 * definitely no active writers (although held writers may subsequently
633 * increment the count, they'll have to wait, and decrement it after
634 * seeing MNT_READONLY).
635 *
636 * It is OK to have counter incremented on one CPU and decremented on
637 * another: the sum will add up correctly. The danger would be when we
638 * sum up each counter, if we read a counter before it is incremented,
639 * but then read another CPU's count which it has been subsequently
640 * decremented from -- we would see more decrements than we should.
641 * MNT_WRITE_HOLD protects against this scenario, because
642 * mnt_want_write first increments count, then smp_mb, then spins on
643 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
644 * we're counting up here.
645 */
Nick Pigginc6653a82011-01-07 17:50:10 +1100646 if (mnt_get_writers(mnt) > 0)
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100647 return -EBUSY;
648
649 return 0;
650}
651
Christian Brauner538f4f022022-02-03 14:14:07 +0100652/**
653 * mnt_unhold_writers - stop preventing write access to the given mount
654 * @mnt: mnt to stop preventing write access to
655 *
656 * Stop preventing write access to @mnt allowing callers to gain write access
657 * to @mnt again.
658 *
659 * This function can only be called after a successful call to
660 * mnt_hold_writers().
661 *
662 * Context: This function expects lock_mount_hash() to be held.
663 */
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100664static inline void mnt_unhold_writers(struct mount *mnt)
665{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000666 /*
667 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
668 * that become unheld will see MNT_READONLY.
669 */
670 smp_wmb();
Al Viro83adc752011-11-24 22:37:54 -0500671 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
Christian Braunerfbdc2f62021-01-21 14:19:51 +0100672}
673
674static int mnt_make_readonly(struct mount *mnt)
675{
676 int ret;
677
678 ret = mnt_hold_writers(mnt);
679 if (!ret)
680 mnt->mnt.mnt_flags |= MNT_READONLY;
681 mnt_unhold_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800682 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800683}
Dave Hansen83660252008-02-15 14:37:30 -0800684
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100685int sb_prepare_remount_readonly(struct super_block *sb)
686{
687 struct mount *mnt;
688 int err = 0;
689
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100690 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
691 if (atomic_long_read(&sb->s_remove_count))
692 return -EBUSY;
693
Al Viro719ea2f2013-09-29 11:24:49 -0400694 lock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100695 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
696 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
Christian Braunerad1844a2022-02-03 14:14:10 +0100697 err = mnt_hold_writers(mnt);
698 if (err)
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100699 break;
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100700 }
701 }
Miklos Szeredi8e8b8792011-11-21 12:11:33 +0100702 if (!err && atomic_long_read(&sb->s_remove_count))
703 err = -EBUSY;
704
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100705 if (!err) {
706 sb->s_readonly_remount = 1;
707 smp_wmb();
708 }
709 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
710 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
711 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
712 }
Al Viro719ea2f2013-09-29 11:24:49 -0400713 unlock_mount_hash();
Miklos Szeredi4ed5e82f2011-11-21 12:11:31 +0100714
715 return err;
716}
717
Al Virob105e272011-11-24 20:38:33 -0500718static void free_vfsmnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Christian Brauner256c8ae2022-10-26 12:51:27 +0200720 mnt_idmap_put(mnt_idmap(&mnt->mnt));
Andrzej Hajdafcc139a2015-02-13 14:36:41 -0800721 kfree_const(mnt->mnt_devname);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000722#ifdef CONFIG_SMP
Al Viro68e8a9f2011-11-24 22:53:09 -0500723 free_percpu(mnt->mnt_pcp);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000724#endif
Al Virob105e272011-11-24 20:38:33 -0500725 kmem_cache_free(mnt_cache, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
David Howells8ffcb322014-01-24 12:17:54 +0000728static void delayed_free_vfsmnt(struct rcu_head *head)
729{
730 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
731}
732
Al Viro48a066e2013-09-29 22:06:07 -0400733/* call under rcu_read_lock */
Al Viro294d71f2015-05-08 11:43:53 -0400734int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro48a066e2013-09-29 22:06:07 -0400735{
736 struct mount *mnt;
737 if (read_seqretry(&mount_lock, seq))
Al Viro294d71f2015-05-08 11:43:53 -0400738 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400739 if (bastard == NULL)
Al Viro294d71f2015-05-08 11:43:53 -0400740 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400741 mnt = real_mount(bastard);
742 mnt_add_count(mnt, 1);
Al Viro119e1ef2018-08-09 17:51:32 -0400743 smp_mb(); // see mntput_no_expire()
Al Viro48a066e2013-09-29 22:06:07 -0400744 if (likely(!read_seqretry(&mount_lock, seq)))
Al Viro294d71f2015-05-08 11:43:53 -0400745 return 0;
Al Viro48a066e2013-09-29 22:06:07 -0400746 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
747 mnt_add_count(mnt, -1);
Al Viro294d71f2015-05-08 11:43:53 -0400748 return 1;
Al Viro48a066e2013-09-29 22:06:07 -0400749 }
Al Viro119e1ef2018-08-09 17:51:32 -0400750 lock_mount_hash();
751 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
752 mnt_add_count(mnt, -1);
753 unlock_mount_hash();
754 return 1;
755 }
756 unlock_mount_hash();
757 /* caller will mntput() */
Al Viro294d71f2015-05-08 11:43:53 -0400758 return -1;
759}
760
761/* call under rcu_read_lock */
Al Viro7e4745a2022-07-05 12:22:46 -0400762static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
Al Viro294d71f2015-05-08 11:43:53 -0400763{
764 int res = __legitimize_mnt(bastard, seq);
765 if (likely(!res))
766 return true;
767 if (unlikely(res < 0)) {
768 rcu_read_unlock();
769 mntput(bastard);
770 rcu_read_lock();
771 }
Al Viro48a066e2013-09-29 22:06:07 -0400772 return false;
773}
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775/*
Al Viro474279d2013-10-01 16:11:26 -0400776 * find the first mount at @dentry on vfsmount @mnt.
Al Viro48a066e2013-09-29 22:06:07 -0400777 * call under rcu_read_lock()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Al Viro474279d2013-10-01 16:11:26 -0400779struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
Al Viro38129a12014-03-20 21:10:51 -0400781 struct hlist_head *head = m_hash(mnt, dentry);
Al Viro474279d2013-10-01 16:11:26 -0400782 struct mount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Al Viro38129a12014-03-20 21:10:51 -0400784 hlist_for_each_entry_rcu(p, head, mnt_hash)
Al Viro474279d2013-10-01 16:11:26 -0400785 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
786 return p;
787 return NULL;
788}
789
790/*
David Howellsf015f1262012-06-25 12:55:28 +0100791 * lookup_mnt - Return the first child mount mounted at path
792 *
793 * "First" means first mounted chronologically. If you create the
794 * following mounts:
795 *
796 * mount /dev/sda1 /mnt
797 * mount /dev/sda2 /mnt
798 * mount /dev/sda3 /mnt
799 *
800 * Then lookup_mnt() on the base /mnt dentry in the root mount will
801 * return successively the root dentry and vfsmount of /dev/sda1, then
802 * /dev/sda2, then /dev/sda3, then NULL.
803 *
804 * lookup_mnt takes a reference to the found vfsmount.
Ram Paia05964f2005-11-07 17:20:17 -0500805 */
Al Viroca71cf72016-11-20 19:45:28 -0500806struct vfsmount *lookup_mnt(const struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500807{
Al Viroc7105362011-11-24 18:22:03 -0500808 struct mount *child_mnt;
Al Viro48a066e2013-09-29 22:06:07 -0400809 struct vfsmount *m;
810 unsigned seq;
Nick Piggin99b7db72010-08-18 04:37:39 +1000811
Al Viro48a066e2013-09-29 22:06:07 -0400812 rcu_read_lock();
813 do {
814 seq = read_seqbegin(&mount_lock);
815 child_mnt = __lookup_mnt(path->mnt, path->dentry);
816 m = child_mnt ? &child_mnt->mnt : NULL;
817 } while (!legitimize_mnt(m, seq));
818 rcu_read_unlock();
819 return m;
Ram Paia05964f2005-11-07 17:20:17 -0500820}
821
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200822static inline void lock_ns_list(struct mnt_namespace *ns)
823{
824 spin_lock(&ns->ns_lock);
825}
826
827static inline void unlock_ns_list(struct mnt_namespace *ns)
828{
829 spin_unlock(&ns->ns_lock);
830}
831
832static inline bool mnt_is_cursor(struct mount *mnt)
833{
834 return mnt->mnt.mnt_flags & MNT_CURSOR;
835}
836
Eric W. Biederman7af13642013-10-04 19:15:13 -0700837/*
838 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
839 * current mount namespace.
840 *
841 * The common case is dentries are not mountpoints at all and that
842 * test is handled inline. For the slow case when we are actually
843 * dealing with a mountpoint of some kind, walk through all of the
844 * mounts in the current mount namespace and test to see if the dentry
845 * is a mountpoint.
846 *
847 * The mount_hashtable is not usable in the context because we
848 * need to identify all mounts that may be in the current mount
849 * namespace not just a mount that happens to have some specified
850 * parent mount.
851 */
852bool __is_local_mountpoint(struct dentry *dentry)
853{
854 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
855 struct mount *mnt;
856 bool is_covered = false;
857
Eric W. Biederman7af13642013-10-04 19:15:13 -0700858 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200859 lock_ns_list(ns);
Eric W. Biederman7af13642013-10-04 19:15:13 -0700860 list_for_each_entry(mnt, &ns->list, mnt_list) {
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200861 if (mnt_is_cursor(mnt))
862 continue;
Eric W. Biederman7af13642013-10-04 19:15:13 -0700863 is_covered = (mnt->mnt_mountpoint == dentry);
864 if (is_covered)
865 break;
866 }
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +0200867 unlock_ns_list(ns);
Eric W. Biederman7af13642013-10-04 19:15:13 -0700868 up_read(&namespace_sem);
Nikolay Borisov5ad05cc2020-03-04 18:12:45 +0200869
Eric W. Biederman7af13642013-10-04 19:15:13 -0700870 return is_covered;
871}
872
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800873static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
Al Viro84d171922013-03-15 10:53:28 -0400874{
Al Viro0818bf22014-02-28 13:46:44 -0500875 struct hlist_head *chain = mp_hash(dentry);
Al Viro84d171922013-03-15 10:53:28 -0400876 struct mountpoint *mp;
877
Al Viro0818bf22014-02-28 13:46:44 -0500878 hlist_for_each_entry(mp, chain, m_hash) {
Al Viro84d171922013-03-15 10:53:28 -0400879 if (mp->m_dentry == dentry) {
Al Viro84d171922013-03-15 10:53:28 -0400880 mp->m_count++;
881 return mp;
882 }
883 }
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800884 return NULL;
885}
886
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300887static struct mountpoint *get_mountpoint(struct dentry *dentry)
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800888{
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300889 struct mountpoint *mp, *new = NULL;
Eric W. Biedermane2dfa932014-02-24 17:32:34 -0800890 int ret;
Al Viro84d171922013-03-15 10:53:28 -0400891
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300892 if (d_mountpoint(dentry)) {
Benjamin Coddington1e9c75f2018-10-03 10:18:33 -0400893 /* might be worth a WARN_ON() */
894 if (d_unlinked(dentry))
895 return ERR_PTR(-ENOENT);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300896mountpoint:
897 read_seqlock_excl(&mount_lock);
898 mp = lookup_mountpoint(dentry);
899 read_sequnlock_excl(&mount_lock);
900 if (mp)
901 goto done;
Al Viro84d171922013-03-15 10:53:28 -0400902 }
Miklos Szeredieed81002013-09-05 14:39:11 +0200903
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300904 if (!new)
905 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
906 if (!new)
907 return ERR_PTR(-ENOMEM);
908
909
910 /* Exactly one processes may set d_mounted */
911 ret = d_set_mounted(dentry);
912
913 /* Someone else set d_mounted? */
914 if (ret == -EBUSY)
915 goto mountpoint;
916
917 /* The dentry is not available as a mountpoint? */
918 mp = ERR_PTR(ret);
919 if (ret)
920 goto done;
921
922 /* Add the new mountpoint to the hash table */
923 read_seqlock_excl(&mount_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400924 new->m_dentry = dget(dentry);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +1300925 new->m_count = 1;
926 hlist_add_head(&new->m_hash, mp_hash(dentry));
927 INIT_HLIST_HEAD(&new->m_list);
928 read_sequnlock_excl(&mount_lock);
929
930 mp = new;
931 new = NULL;
932done:
933 kfree(new);
Al Viro84d171922013-03-15 10:53:28 -0400934 return mp;
935}
936
Al Viro4edbe132019-06-30 10:39:08 -0400937/*
938 * vfsmount lock must be held. Additionally, the caller is responsible
939 * for serializing calls for given disposal list.
940 */
941static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
Al Viro84d171922013-03-15 10:53:28 -0400942{
943 if (!--mp->m_count) {
944 struct dentry *dentry = mp->m_dentry;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700945 BUG_ON(!hlist_empty(&mp->m_list));
Al Viro84d171922013-03-15 10:53:28 -0400946 spin_lock(&dentry->d_lock);
947 dentry->d_flags &= ~DCACHE_MOUNTED;
948 spin_unlock(&dentry->d_lock);
Al Viro4edbe132019-06-30 10:39:08 -0400949 dput_to_list(dentry, list);
Al Viro0818bf22014-02-28 13:46:44 -0500950 hlist_del(&mp->m_hash);
Al Viro84d171922013-03-15 10:53:28 -0400951 kfree(mp);
952 }
953}
954
Al Viro4edbe132019-06-30 10:39:08 -0400955/* called with namespace_lock and vfsmount lock */
956static void put_mountpoint(struct mountpoint *mp)
957{
958 __put_mountpoint(mp, &ex_mountpoints);
959}
960
Al Viro143c8c92011-11-25 00:46:35 -0500961static inline int check_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800963 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964}
965
Nick Piggin99b7db72010-08-18 04:37:39 +1000966/*
967 * vfsmount lock must be held for write
968 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800969static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500970{
971 if (ns) {
972 ns->event = ++event;
973 wake_up_interruptible(&ns->poll);
974 }
975}
976
Nick Piggin99b7db72010-08-18 04:37:39 +1000977/*
978 * vfsmount lock must be held for write
979 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800980static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500981{
982 if (ns && ns->event != event) {
983 ns->event = event;
984 wake_up_interruptible(&ns->poll);
985 }
986}
987
Nick Piggin99b7db72010-08-18 04:37:39 +1000988/*
989 * vfsmount lock must be held for write
990 */
Al Viroe4e59902019-06-29 12:58:42 -0400991static struct mountpoint *unhash_mnt(struct mount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
Al Viroe4e59902019-06-29 12:58:42 -0400993 struct mountpoint *mp;
Al Viro0714a532011-11-24 22:19:58 -0500994 mnt->mnt_parent = mnt;
Al Viroa73324d2011-11-24 22:25:07 -0500995 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
Al Viro6b41d532011-11-24 23:24:33 -0500996 list_del_init(&mnt->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400997 hlist_del_init_rcu(&mnt->mnt_hash);
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -0700998 hlist_del_init(&mnt->mnt_mp_list);
Al Viroe4e59902019-06-29 12:58:42 -0400999 mp = mnt->mnt_mp;
Al Viro84d171922013-03-15 10:53:28 -04001000 mnt->mnt_mp = NULL;
Al Viroe4e59902019-06-29 12:58:42 -04001001 return mp;
Eric W. Biederman7bdb11de2014-12-29 13:03:41 -06001002}
1003
1004/*
1005 * vfsmount lock must be held for write
1006 */
Eric W. Biederman6a46c572015-01-15 22:58:33 -06001007static void umount_mnt(struct mount *mnt)
1008{
Al Viroe4e59902019-06-29 12:58:42 -04001009 put_mountpoint(unhash_mnt(mnt));
Eric W. Biederman6a46c572015-01-15 22:58:33 -06001010}
1011
1012/*
1013 * vfsmount lock must be held for write
1014 */
Al Viro84d171922013-03-15 10:53:28 -04001015void mnt_set_mountpoint(struct mount *mnt,
1016 struct mountpoint *mp,
Al Viro44d964d62011-11-24 21:28:22 -05001017 struct mount *child_mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -05001018{
Al Viro84d171922013-03-15 10:53:28 -04001019 mp->m_count++;
Al Viro3a2393d2011-11-25 03:19:09 -05001020 mnt_add_count(mnt, 1); /* essentially, that's mntget */
Al Viro4edbe132019-06-30 10:39:08 -04001021 child_mnt->mnt_mountpoint = mp->m_dentry;
Al Viro3a2393d2011-11-25 03:19:09 -05001022 child_mnt->mnt_parent = mnt;
Al Viro84d171922013-03-15 10:53:28 -04001023 child_mnt->mnt_mp = mp;
Eric W. Biederman0a5eb7c2013-09-22 19:37:01 -07001024 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
Ram Paib90fa9a2005-11-07 17:19:50 -05001025}
1026
Eric W. Biederman1064f872017-01-20 18:28:35 +13001027static void __attach_mnt(struct mount *mnt, struct mount *parent)
1028{
1029 hlist_add_head_rcu(&mnt->mnt_hash,
1030 m_hash(&parent->mnt, mnt->mnt_mountpoint));
1031 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
1032}
1033
Nick Piggin99b7db72010-08-18 04:37:39 +10001034/*
1035 * vfsmount lock must be held for write
1036 */
Al Viro84d171922013-03-15 10:53:28 -04001037static void attach_mnt(struct mount *mnt,
1038 struct mount *parent,
1039 struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040{
Al Viro84d171922013-03-15 10:53:28 -04001041 mnt_set_mountpoint(parent, mp, mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001042 __attach_mnt(mnt, parent);
Ram Paib90fa9a2005-11-07 17:19:50 -05001043}
1044
Eric W. Biederman1064f872017-01-20 18:28:35 +13001045void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
Al Viro12a5b522014-08-10 03:44:55 -04001046{
Eric W. Biederman1064f872017-01-20 18:28:35 +13001047 struct mountpoint *old_mp = mnt->mnt_mp;
Eric W. Biederman1064f872017-01-20 18:28:35 +13001048 struct mount *old_parent = mnt->mnt_parent;
1049
1050 list_del_init(&mnt->mnt_child);
1051 hlist_del_init(&mnt->mnt_mp_list);
1052 hlist_del_init_rcu(&mnt->mnt_hash);
1053
1054 attach_mnt(mnt, parent, mp);
1055
1056 put_mountpoint(old_mp);
Eric W. Biederman1064f872017-01-20 18:28:35 +13001057 mnt_add_count(old_parent, -1);
Al Viro12a5b522014-08-10 03:44:55 -04001058}
1059
Ram Paib90fa9a2005-11-07 17:19:50 -05001060/*
Nick Piggin99b7db72010-08-18 04:37:39 +10001061 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -05001062 */
Eric W. Biederman1064f872017-01-20 18:28:35 +13001063static void commit_tree(struct mount *mnt)
Ram Paib90fa9a2005-11-07 17:19:50 -05001064{
Al Viro0714a532011-11-24 22:19:58 -05001065 struct mount *parent = mnt->mnt_parent;
Al Viro83adc752011-11-24 22:37:54 -05001066 struct mount *m;
Ram Paib90fa9a2005-11-07 17:19:50 -05001067 LIST_HEAD(head);
Al Viro143c8c92011-11-25 00:46:35 -05001068 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -05001069
Al Viro0714a532011-11-24 22:19:58 -05001070 BUG_ON(parent == mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -05001071
Al Viro1a4eeaf2011-11-25 02:19:55 -05001072 list_add_tail(&head, &mnt->mnt_list);
Al Virof7a99c52012-06-09 00:59:08 -04001073 list_for_each_entry(m, &head, mnt_list)
Al Viro143c8c92011-11-25 00:46:35 -05001074 m->mnt_ns = n;
Al Virof03c6592011-01-14 22:30:21 -05001075
Ram Paib90fa9a2005-11-07 17:19:50 -05001076 list_splice(&head, n->list.prev);
1077
Eric W. Biedermand2921682016-09-28 00:27:17 -05001078 n->mounts += n->pending_mounts;
1079 n->pending_mounts = 0;
1080
Eric W. Biederman1064f872017-01-20 18:28:35 +13001081 __attach_mnt(mnt, parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001082 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083}
1084
Al Viro909b0a82011-11-25 03:06:56 -05001085static struct mount *next_mnt(struct mount *p, struct mount *root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
Al Viro6b41d532011-11-24 23:24:33 -05001087 struct list_head *next = p->mnt_mounts.next;
1088 if (next == &p->mnt_mounts) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 while (1) {
Al Viro909b0a82011-11-25 03:06:56 -05001090 if (p == root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 return NULL;
Al Viro6b41d532011-11-24 23:24:33 -05001092 next = p->mnt_child.next;
1093 if (next != &p->mnt_parent->mnt_mounts)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 break;
Al Viro0714a532011-11-24 22:19:58 -05001095 p = p->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 }
1097 }
Al Viro6b41d532011-11-24 23:24:33 -05001098 return list_entry(next, struct mount, mnt_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099}
1100
Al Viro315fc832011-11-24 18:57:30 -05001101static struct mount *skip_mnt_tree(struct mount *p)
Ram Pai9676f0c2005-11-07 17:21:20 -05001102{
Al Viro6b41d532011-11-24 23:24:33 -05001103 struct list_head *prev = p->mnt_mounts.prev;
1104 while (prev != &p->mnt_mounts) {
1105 p = list_entry(prev, struct mount, mnt_child);
1106 prev = p->mnt_mounts.prev;
Ram Pai9676f0c2005-11-07 17:21:20 -05001107 }
1108 return p;
1109}
1110
Al Viro8f2918892018-11-04 06:48:34 -05001111/**
1112 * vfs_create_mount - Create a mount for a configured superblock
1113 * @fc: The configuration context with the superblock attached
1114 *
1115 * Create a mount to an already configured superblock. If necessary, the
1116 * caller should invoke vfs_get_tree() before calling this.
1117 *
1118 * Note that this does not attach the mount to anything.
1119 */
1120struct vfsmount *vfs_create_mount(struct fs_context *fc)
Al Viro9d412a42011-03-17 22:08:28 -04001121{
Al Virob105e272011-11-24 20:38:33 -05001122 struct mount *mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001123
Al Viro8f2918892018-11-04 06:48:34 -05001124 if (!fc->root)
1125 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -04001126
Al Viro8f2918892018-11-04 06:48:34 -05001127 mnt = alloc_vfsmnt(fc->source ?: "none");
Al Viro9d412a42011-03-17 22:08:28 -04001128 if (!mnt)
1129 return ERR_PTR(-ENOMEM);
1130
Al Viro8f2918892018-11-04 06:48:34 -05001131 if (fc->sb_flags & SB_KERNMOUNT)
Al Virob105e272011-11-24 20:38:33 -05001132 mnt->mnt.mnt_flags = MNT_INTERNAL;
Al Viro9d412a42011-03-17 22:08:28 -04001133
Al Viro8f2918892018-11-04 06:48:34 -05001134 atomic_inc(&fc->root->d_sb->s_active);
1135 mnt->mnt.mnt_sb = fc->root->d_sb;
1136 mnt->mnt.mnt_root = dget(fc->root);
1137 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1138 mnt->mnt_parent = mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001139
Al Viro719ea2f2013-09-29 11:24:49 -04001140 lock_mount_hash();
Al Viro8f2918892018-11-04 06:48:34 -05001141 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001142 unlock_mount_hash();
Al Virob105e272011-11-24 20:38:33 -05001143 return &mnt->mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001144}
Al Viro8f2918892018-11-04 06:48:34 -05001145EXPORT_SYMBOL(vfs_create_mount);
1146
1147struct vfsmount *fc_mount(struct fs_context *fc)
1148{
1149 int err = vfs_get_tree(fc);
1150 if (!err) {
1151 up_write(&fc->root->d_sb->s_umount);
1152 return vfs_create_mount(fc);
1153 }
1154 return ERR_PTR(err);
1155}
1156EXPORT_SYMBOL(fc_mount);
1157
David Howells9bc61ab2018-11-04 03:19:03 -05001158struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1159 int flags, const char *name,
1160 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
David Howells9bc61ab2018-11-04 03:19:03 -05001162 struct fs_context *fc;
Al Viro8f2918892018-11-04 06:48:34 -05001163 struct vfsmount *mnt;
David Howells9bc61ab2018-11-04 03:19:03 -05001164 int ret = 0;
Al Viro9d412a42011-03-17 22:08:28 -04001165
1166 if (!type)
David Howells3e1aeb02018-11-01 23:07:25 +00001167 return ERR_PTR(-EINVAL);
Al Viro9d412a42011-03-17 22:08:28 -04001168
David Howells9bc61ab2018-11-04 03:19:03 -05001169 fc = fs_context_for_mount(type, flags);
1170 if (IS_ERR(fc))
1171 return ERR_CAST(fc);
1172
David Howells3e1aeb02018-11-01 23:07:25 +00001173 if (name)
1174 ret = vfs_parse_fs_string(fc, "source",
1175 name, strlen(name));
David Howells9bc61ab2018-11-04 03:19:03 -05001176 if (!ret)
1177 ret = parse_monolithic_mount_data(fc, data);
1178 if (!ret)
Al Viro8f2918892018-11-04 06:48:34 -05001179 mnt = fc_mount(fc);
1180 else
1181 mnt = ERR_PTR(ret);
Al Viro9d412a42011-03-17 22:08:28 -04001182
David Howells9bc61ab2018-11-04 03:19:03 -05001183 put_fs_context(fc);
Al Viro8f2918892018-11-04 06:48:34 -05001184 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04001185}
1186EXPORT_SYMBOL_GPL(vfs_kern_mount);
1187
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001188struct vfsmount *
1189vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1190 const char *name, void *data)
1191{
1192 /* Until it is worked out how to pass the user namespace
1193 * through from the parent mount to the submount don't support
1194 * unprivileged mounts with submounts.
1195 */
1196 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1197 return ERR_PTR(-EPERM);
1198
David Howellse462ec52017-07-17 08:45:35 +01001199 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
Eric W. Biederman93faccbb2017-02-01 06:06:16 +13001200}
1201EXPORT_SYMBOL_GPL(vfs_submount);
1202
Al Viro87129cc2011-11-24 21:24:27 -05001203static struct mount *clone_mnt(struct mount *old, struct dentry *root,
Ram Pai36341f62005-11-07 17:17:22 -05001204 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Al Viro87129cc2011-11-24 21:24:27 -05001206 struct super_block *sb = old->mnt.mnt_sb;
David Howellsbe34d1a2012-06-25 12:55:18 +01001207 struct mount *mnt;
1208 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
David Howellsbe34d1a2012-06-25 12:55:18 +01001210 mnt = alloc_vfsmnt(old->mnt_devname);
1211 if (!mnt)
1212 return ERR_PTR(-ENOMEM);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001213
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001214 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
David Howellsbe34d1a2012-06-25 12:55:18 +01001215 mnt->mnt_group_id = 0; /* not a peer of original */
1216 else
1217 mnt->mnt_group_id = old->mnt_group_id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001218
David Howellsbe34d1a2012-06-25 12:55:18 +01001219 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1220 err = mnt_alloc_group_id(mnt);
1221 if (err)
1222 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
David Howellsbe34d1a2012-06-25 12:55:18 +01001224
Al Viro16a34ad2018-04-19 22:03:08 -04001225 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1226 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07001227
David Howellsbe34d1a2012-06-25 12:55:18 +01001228 atomic_inc(&sb->s_active);
Christian Brauner256c8ae2022-10-26 12:51:27 +02001229 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1230
David Howellsbe34d1a2012-06-25 12:55:18 +01001231 mnt->mnt.mnt_sb = sb;
1232 mnt->mnt.mnt_root = dget(root);
1233 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1234 mnt->mnt_parent = mnt;
Al Viro719ea2f2013-09-29 11:24:49 -04001235 lock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001236 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
Al Viro719ea2f2013-09-29 11:24:49 -04001237 unlock_mount_hash();
David Howellsbe34d1a2012-06-25 12:55:18 +01001238
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07001239 if ((flag & CL_SLAVE) ||
1240 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
David Howellsbe34d1a2012-06-25 12:55:18 +01001241 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1242 mnt->mnt_master = old;
1243 CLEAR_MNT_SHARED(mnt);
1244 } else if (!(flag & CL_PRIVATE)) {
1245 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1246 list_add(&mnt->mnt_share, &old->mnt_share);
1247 if (IS_MNT_SLAVE(old))
1248 list_add(&mnt->mnt_slave, &old->mnt_slave);
1249 mnt->mnt_master = old->mnt_master;
Al Viro5235d442016-11-20 19:33:09 -05001250 } else {
1251 CLEAR_MNT_SHARED(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001252 }
1253 if (flag & CL_MAKE_SHARED)
1254 set_mnt_shared(mnt);
1255
1256 /* stick the duplicate mount on the same expiry list
1257 * as the original if that was on one */
1258 if (flag & CL_EXPIRE) {
1259 if (!list_empty(&old->mnt_expire))
1260 list_add(&mnt->mnt_expire, &old->mnt_expire);
1261 }
1262
Al Virocb338d02011-11-24 20:55:08 -05001263 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001264
1265 out_free:
David Howells8ffcb322014-01-24 12:17:54 +00001266 mnt_free_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001267 free_vfsmnt(mnt);
David Howellsbe34d1a2012-06-25 12:55:18 +01001268 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269}
1270
Al Viro9ea459e12014-08-08 13:08:20 -04001271static void cleanup_mnt(struct mount *mnt)
1272{
Al Viro56cbb4292019-07-04 16:57:51 -04001273 struct hlist_node *p;
1274 struct mount *m;
Al Viro9ea459e12014-08-08 13:08:20 -04001275 /*
Al Viro56cbb4292019-07-04 16:57:51 -04001276 * The warning here probably indicates that somebody messed
1277 * up a mnt_want/drop_write() pair. If this happens, the
1278 * filesystem was probably unable to make r/w->r/o transitions.
Al Viro9ea459e12014-08-08 13:08:20 -04001279 * The locking used to deal with mnt_count decrement provides barriers,
1280 * so mnt_get_writers() below is safe.
1281 */
1282 WARN_ON(mnt_get_writers(mnt));
1283 if (unlikely(mnt->mnt_pins.first))
1284 mnt_pin_kill(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001285 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1286 hlist_del(&m->mnt_umount);
1287 mntput(&m->mnt);
1288 }
Al Viro9ea459e12014-08-08 13:08:20 -04001289 fsnotify_vfsmount_delete(&mnt->mnt);
1290 dput(mnt->mnt.mnt_root);
1291 deactivate_super(mnt->mnt.mnt_sb);
1292 mnt_free_id(mnt);
1293 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1294}
1295
1296static void __cleanup_mnt(struct rcu_head *head)
1297{
1298 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1299}
1300
1301static LLIST_HEAD(delayed_mntput_list);
1302static void delayed_mntput(struct work_struct *unused)
1303{
1304 struct llist_node *node = llist_del_all(&delayed_mntput_list);
Byungchul Park29785732017-08-07 17:44:45 +09001305 struct mount *m, *t;
Al Viro9ea459e12014-08-08 13:08:20 -04001306
Byungchul Park29785732017-08-07 17:44:45 +09001307 llist_for_each_entry_safe(m, t, node, mnt_llist)
1308 cleanup_mnt(m);
Al Viro9ea459e12014-08-08 13:08:20 -04001309}
1310static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1311
Al Viro900148d2011-11-25 00:33:11 -05001312static void mntput_no_expire(struct mount *mnt)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001313{
Al Viro4edbe132019-06-30 10:39:08 -04001314 LIST_HEAD(list);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001315 int count;
Al Viro4edbe132019-06-30 10:39:08 -04001316
Al Viro48a066e2013-09-29 22:06:07 -04001317 rcu_read_lock();
Al Viro9ea0a462018-08-09 17:21:17 -04001318 if (likely(READ_ONCE(mnt->mnt_ns))) {
1319 /*
1320 * Since we don't do lock_mount_hash() here,
1321 * ->mnt_ns can change under us. However, if it's
1322 * non-NULL, then there's a reference that won't
1323 * be dropped until after an RCU delay done after
1324 * turning ->mnt_ns NULL. So if we observe it
1325 * non-NULL under rcu_read_lock(), the reference
1326 * we are dropping is not the final one.
1327 */
1328 mnt_add_count(mnt, -1);
Al Viro48a066e2013-09-29 22:06:07 -04001329 rcu_read_unlock();
Al Virof03c6592011-01-14 22:30:21 -05001330 return;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001331 }
Al Viro719ea2f2013-09-29 11:24:49 -04001332 lock_mount_hash();
Al Viro119e1ef2018-08-09 17:51:32 -04001333 /*
1334 * make sure that if __legitimize_mnt() has not seen us grab
1335 * mount_lock, we'll see their refcount increment here.
1336 */
1337 smp_mb();
Al Viro9ea0a462018-08-09 17:21:17 -04001338 mnt_add_count(mnt, -1);
Eric Biggersedf7ddb2020-10-31 21:40:21 -07001339 count = mnt_get_count(mnt);
1340 if (count != 0) {
1341 WARN_ON(count < 0);
Al Viro48a066e2013-09-29 22:06:07 -04001342 rcu_read_unlock();
Al Viro719ea2f2013-09-29 11:24:49 -04001343 unlock_mount_hash();
Nick Pigginb3e19d92011-01-07 17:50:11 +11001344 return;
1345 }
Al Viro48a066e2013-09-29 22:06:07 -04001346 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1347 rcu_read_unlock();
1348 unlock_mount_hash();
1349 return;
1350 }
1351 mnt->mnt.mnt_flags |= MNT_DOOMED;
1352 rcu_read_unlock();
Andi Kleen962830d2012-05-08 13:32:02 +09301353
Miklos Szeredi39f7c4d2011-11-21 12:11:30 +01001354 list_del(&mnt->mnt_instance);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001355
1356 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1357 struct mount *p, *tmp;
1358 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
Al Viro4edbe132019-06-30 10:39:08 -04001359 __put_mountpoint(unhash_mnt(p), &list);
Al Viro56cbb4292019-07-04 16:57:51 -04001360 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001361 }
1362 }
Al Viro719ea2f2013-09-29 11:24:49 -04001363 unlock_mount_hash();
Al Viro4edbe132019-06-30 10:39:08 -04001364 shrink_dentry_list(&list);
Al Viro649a7952013-09-28 12:41:25 -04001365
Al Viro9ea459e12014-08-08 13:08:20 -04001366 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1367 struct task_struct *task = current;
1368 if (likely(!(task->flags & PF_KTHREAD))) {
1369 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
Jens Axboe91989c72020-10-16 09:02:26 -06001370 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
Al Viro9ea459e12014-08-08 13:08:20 -04001371 return;
1372 }
1373 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1374 schedule_delayed_work(&delayed_mntput_work, 1);
1375 return;
1376 }
1377 cleanup_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -05001378}
Nick Pigginb3e19d92011-01-07 17:50:11 +11001379
1380void mntput(struct vfsmount *mnt)
1381{
1382 if (mnt) {
Al Viro863d6842011-11-25 00:57:42 -05001383 struct mount *m = real_mount(mnt);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001384 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
Al Viro863d6842011-11-25 00:57:42 -05001385 if (unlikely(m->mnt_expiry_mark))
1386 m->mnt_expiry_mark = 0;
1387 mntput_no_expire(m);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001388 }
1389}
1390EXPORT_SYMBOL(mntput);
1391
1392struct vfsmount *mntget(struct vfsmount *mnt)
1393{
1394 if (mnt)
Al Viro83adc752011-11-24 22:37:54 -05001395 mnt_add_count(real_mount(mnt), 1);
Nick Pigginb3e19d92011-01-07 17:50:11 +11001396 return mnt;
1397}
1398EXPORT_SYMBOL(mntget);
1399
Randy Dunlap1f287bc2021-03-17 19:52:25 -07001400/**
1401 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1402 * @path: path to check
Ian Kentc6609c02016-11-24 08:03:41 +11001403 *
1404 * d_mountpoint() can only be used reliably to establish if a dentry is
1405 * not mounted in any namespace and that common case is handled inline.
1406 * d_mountpoint() isn't aware of the possibility there may be multiple
1407 * mounts using a given dentry in a different namespace. This function
1408 * checks if the passed in path is a mountpoint rather than the dentry
1409 * alone.
1410 */
1411bool path_is_mountpoint(const struct path *path)
1412{
1413 unsigned seq;
1414 bool res;
1415
1416 if (!d_mountpoint(path->dentry))
1417 return false;
1418
1419 rcu_read_lock();
1420 do {
1421 seq = read_seqbegin(&mount_lock);
1422 res = __path_is_mountpoint(path);
1423 } while (read_seqretry(&mount_lock, seq));
1424 rcu_read_unlock();
1425
1426 return res;
1427}
1428EXPORT_SYMBOL(path_is_mountpoint);
1429
Al Viroca71cf72016-11-20 19:45:28 -05001430struct vfsmount *mnt_clone_internal(const struct path *path)
Al Viro7b7b1ac2005-11-07 17:13:39 -05001431{
Al Viro3064c352014-08-07 09:12:31 -04001432 struct mount *p;
1433 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1434 if (IS_ERR(p))
1435 return ERR_CAST(p);
1436 p->mnt.mnt_flags |= MNT_INTERNAL;
1437 return &p->mnt;
Al Viro7b7b1ac2005-11-07 17:13:39 -05001438}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001440#ifdef CONFIG_PROC_FS
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001441static struct mount *mnt_list_next(struct mnt_namespace *ns,
1442 struct list_head *p)
1443{
1444 struct mount *mnt, *ret = NULL;
1445
1446 lock_ns_list(ns);
1447 list_for_each_continue(p, &ns->list) {
1448 mnt = list_entry(p, typeof(*mnt), mnt_list);
1449 if (!mnt_is_cursor(mnt)) {
1450 ret = mnt;
1451 break;
1452 }
1453 }
1454 unlock_ns_list(ns);
1455
1456 return ret;
1457}
1458
Al Viro0226f492011-12-06 12:21:54 -05001459/* iterator; we want it to have access to namespace_sem, thus here... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460static void *m_start(struct seq_file *m, loff_t *pos)
1461{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001462 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001463 struct list_head *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Ram Pai390c6842005-11-07 17:17:51 -05001465 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001466 if (!*pos) {
1467 prev = &p->ns->list;
1468 } else {
1469 prev = &p->cursor.mnt_list;
1470
1471 /* Read after we'd reached the end? */
1472 if (list_empty(prev))
1473 return NULL;
Al Viroc7999c32014-02-27 14:40:10 -05001474 }
1475
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001476 return mnt_list_next(p->ns, prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477}
1478
1479static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1480{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001481 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001482 struct mount *mnt = v;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -07001483
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001484 ++*pos;
1485 return mnt_list_next(p->ns, &mnt->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486}
1487
1488static void m_stop(struct seq_file *m, void *v)
1489{
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001490 struct proc_mounts *p = m->private;
1491 struct mount *mnt = v;
1492
1493 lock_ns_list(p->ns);
1494 if (mnt)
1495 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1496 else
1497 list_del_init(&p->cursor.mnt_list);
1498 unlock_ns_list(p->ns);
Ram Pai390c6842005-11-07 17:17:51 -05001499 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
Al Viro0226f492011-12-06 12:21:54 -05001502static int m_show(struct seq_file *m, void *v)
Al Viro9f5596a2010-02-05 00:40:25 -05001503{
Yann Droneaudede1bf02015-06-30 14:57:30 -07001504 struct proc_mounts *p = m->private;
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001505 struct mount *r = v;
Al Viro0226f492011-12-06 12:21:54 -05001506 return p->show(m, &r->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001509const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 .start = m_start,
1511 .next = m_next,
1512 .stop = m_stop,
Al Viro0226f492011-12-06 12:21:54 -05001513 .show = m_show,
Chuck Leverb4629fe2006-03-20 13:44:12 -05001514};
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02001515
1516void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1517{
1518 down_read(&namespace_sem);
1519 lock_ns_list(ns);
1520 list_del(&cursor->mnt_list);
1521 unlock_ns_list(ns);
1522 up_read(&namespace_sem);
1523}
Miklos Szeredia1a2c402008-03-27 13:06:24 +01001524#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -05001525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526/**
1527 * may_umount_tree - check if a mount tree is busy
Randy Dunlap1f287bc2021-03-17 19:52:25 -07001528 * @m: root of mount tree
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 *
1530 * This is called to check if a tree of mounts has any
1531 * open files, pwds, chroots or sub mounts that are
1532 * busy.
1533 */
Al Viro909b0a82011-11-25 03:06:56 -05001534int may_umount_tree(struct vfsmount *m)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
Al Viro909b0a82011-11-25 03:06:56 -05001536 struct mount *mnt = real_mount(m);
Ram Pai36341f62005-11-07 17:17:22 -05001537 int actual_refs = 0;
1538 int minimum_refs = 0;
Al Viro315fc832011-11-24 18:57:30 -05001539 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05001540 BUG_ON(!m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Nick Pigginb3e19d92011-01-07 17:50:11 +11001542 /* write lock needed for mnt_get_count */
Al Viro719ea2f2013-09-29 11:24:49 -04001543 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05001544 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viro83adc752011-11-24 22:37:54 -05001545 actual_refs += mnt_get_count(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
Al Viro719ea2f2013-09-29 11:24:49 -04001548 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549
1550 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -08001551 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Ian Kente3474a82006-03-27 01:14:51 -08001553 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554}
1555
1556EXPORT_SYMBOL(may_umount_tree);
1557
1558/**
1559 * may_umount - check if a mount point is busy
1560 * @mnt: root of mount
1561 *
1562 * This is called to check if a mount point has any
1563 * open files, pwds, chroots or sub mounts. If the
1564 * mount has sub mounts this will return busy
1565 * regardless of whether the sub mounts are busy.
1566 *
1567 * Doesn't take quota and stuff into account. IOW, in some cases it will
1568 * give false negatives. The main reason why it's here is that we need
1569 * a non-destructive way to look for easily umountable filesystems.
1570 */
1571int may_umount(struct vfsmount *mnt)
1572{
Ian Kente3474a82006-03-27 01:14:51 -08001573 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001574 down_read(&namespace_sem);
Al Viro719ea2f2013-09-29 11:24:49 -04001575 lock_mount_hash();
Al Viro1ab59732011-11-24 21:35:16 -05001576 if (propagate_mount_busy(real_mount(mnt), 2))
Ian Kente3474a82006-03-27 01:14:51 -08001577 ret = 0;
Al Viro719ea2f2013-09-29 11:24:49 -04001578 unlock_mount_hash();
Al Viro8ad08d82010-01-16 12:56:08 -05001579 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001580 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581}
1582
1583EXPORT_SYMBOL(may_umount);
1584
Al Viro97216be2013-03-16 15:12:40 -04001585static void namespace_unlock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001587 struct hlist_head head;
Al Viro56cbb4292019-07-04 16:57:51 -04001588 struct hlist_node *p;
1589 struct mount *m;
Al Viro4edbe132019-06-30 10:39:08 -04001590 LIST_HEAD(list);
Al Viro97216be2013-03-16 15:12:40 -04001591
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001592 hlist_move_list(&unmounted, &head);
Al Viro4edbe132019-06-30 10:39:08 -04001593 list_splice_init(&ex_mountpoints, &list);
Al Viro97216be2013-03-16 15:12:40 -04001594
Al Viro97216be2013-03-16 15:12:40 -04001595 up_write(&namespace_sem);
1596
Al Viro4edbe132019-06-30 10:39:08 -04001597 shrink_dentry_list(&list);
1598
Eric W. Biedermana3b3c562015-04-02 20:33:53 -05001599 if (likely(hlist_empty(&head)))
1600 return;
1601
NeilBrown22cb7402018-11-30 10:33:18 +11001602 synchronize_rcu_expedited();
Al Viro48a066e2013-09-29 22:06:07 -04001603
Al Viro56cbb4292019-07-04 16:57:51 -04001604 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1605 hlist_del(&m->mnt_umount);
1606 mntput(&m->mnt);
1607 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001608}
1609
Al Viro97216be2013-03-16 15:12:40 -04001610static inline void namespace_lock(void)
Al Viroe3197d82013-03-16 14:35:16 -04001611{
Al Viro97216be2013-03-16 15:12:40 -04001612 down_write(&namespace_sem);
Al Viroe3197d82013-03-16 14:35:16 -04001613}
1614
Eric W. Biedermane819f152014-12-24 07:20:01 -06001615enum umount_tree_flags {
1616 UMOUNT_SYNC = 1,
1617 UMOUNT_PROPAGATE = 2,
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001618 UMOUNT_CONNECTED = 4,
Eric W. Biedermane819f152014-12-24 07:20:01 -06001619};
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001620
1621static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1622{
1623 /* Leaving mounts connected is only valid for lazy umounts */
1624 if (how & UMOUNT_SYNC)
1625 return true;
1626
1627 /* A mount without a parent has nothing to be connected to */
1628 if (!mnt_has_parent(mnt))
1629 return true;
1630
1631 /* Because the reference counting rules change when mounts are
1632 * unmounted and connected, umounted mounts may not be
1633 * connected to mounted mounts.
1634 */
1635 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1636 return true;
1637
1638 /* Has it been requested that the mount remain connected? */
1639 if (how & UMOUNT_CONNECTED)
1640 return false;
1641
1642 /* Is the mount locked such that it needs to remain connected? */
1643 if (IS_MNT_LOCKED(mnt))
1644 return false;
1645
1646 /* By default disconnect the mount */
1647 return true;
1648}
1649
Nick Piggin99b7db72010-08-18 04:37:39 +10001650/*
Al Viro48a066e2013-09-29 22:06:07 -04001651 * mount_lock must be held
Nick Piggin99b7db72010-08-18 04:37:39 +10001652 * namespace_sem must be held for write
1653 */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001654static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001655{
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001656 LIST_HEAD(tmp_list);
Al Viro315fc832011-11-24 18:57:30 -05001657 struct mount *p;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001658
Eric W. Biederman5d884572015-01-03 05:39:35 -06001659 if (how & UMOUNT_PROPAGATE)
1660 propagate_mount_unlock(mnt);
1661
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001662 /* Gather the mounts to umount */
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001663 for (p = mnt; p; p = next_mnt(p, mnt)) {
1664 p->mnt.mnt_flags |= MNT_UMOUNT;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001665 list_move(&p->mnt_list, &tmp_list);
Eric W. Biederman590ce4bc2014-12-22 18:30:08 -06001666 }
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001667
Eric W. Biederman411a9382014-12-22 19:12:07 -06001668 /* Hide the mounts from mnt_mounts */
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001669 list_for_each_entry(p, &tmp_list, mnt_list) {
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001670 list_del_init(&p->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -04001671 }
Ram Pai70fbcdf2005-11-07 17:17:04 -05001672
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001673 /* Add propogated mounts to the tmp_list */
Eric W. Biedermane819f152014-12-24 07:20:01 -06001674 if (how & UMOUNT_PROPAGATE)
Al Viro7b8a53f2011-01-15 20:08:44 -05001675 propagate_umount(&tmp_list);
Ram Paia05964f2005-11-07 17:20:17 -05001676
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001677 while (!list_empty(&tmp_list)) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05001678 struct mnt_namespace *ns;
Eric W. Biedermance07d892014-12-23 21:37:03 -06001679 bool disconnect;
Eric W. Biedermanc003b262014-12-18 13:10:48 -06001680 p = list_first_entry(&tmp_list, struct mount, mnt_list);
Al Viro6776db3d2011-11-25 00:22:05 -05001681 list_del_init(&p->mnt_expire);
Al Viro1a4eeaf2011-11-25 02:19:55 -05001682 list_del_init(&p->mnt_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05001683 ns = p->mnt_ns;
1684 if (ns) {
1685 ns->mounts--;
1686 __touch_mnt_namespace(ns);
1687 }
Al Viro143c8c92011-11-25 00:46:35 -05001688 p->mnt_ns = NULL;
Eric W. Biedermane819f152014-12-24 07:20:01 -06001689 if (how & UMOUNT_SYNC)
Al Viro48a066e2013-09-29 22:06:07 -04001690 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
Al Viro87b95ce2015-01-10 19:01:08 -05001691
Eric W. Biedermanf2d0a122015-07-17 14:15:30 -05001692 disconnect = disconnect_mount(p, how);
Al Viro676da582011-11-24 21:47:05 -05001693 if (mnt_has_parent(p)) {
Al Viro81b6b062014-08-30 18:32:05 -04001694 mnt_add_count(p->mnt_parent, -1);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001695 if (!disconnect) {
1696 /* Don't forget about p */
1697 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1698 } else {
1699 umount_mnt(p);
1700 }
Al Viro7c4b93d2008-03-21 23:59:49 -04001701 }
Al Viro0f0afb12011-11-24 20:43:10 -05001702 change_mnt_propagation(p, MS_PRIVATE);
Al Viro19a1c402019-07-24 12:45:46 -04001703 if (disconnect)
1704 hlist_add_head(&p->mnt_umount, &unmounted);
Al Viro38129a12014-03-20 21:10:51 -04001705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706}
1707
Al Virob54b9be2013-03-16 14:39:34 -04001708static void shrink_submounts(struct mount *mnt);
Al Viroc35038b2008-03-22 00:46:23 -04001709
David Howells8d0347f2018-11-04 09:28:36 -05001710static int do_umount_root(struct super_block *sb)
1711{
1712 int ret = 0;
1713
1714 down_write(&sb->s_umount);
1715 if (!sb_rdonly(sb)) {
1716 struct fs_context *fc;
1717
1718 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1719 SB_RDONLY);
1720 if (IS_ERR(fc)) {
1721 ret = PTR_ERR(fc);
1722 } else {
1723 ret = parse_monolithic_mount_data(fc, NULL);
1724 if (!ret)
1725 ret = reconfigure_super(fc);
1726 put_fs_context(fc);
1727 }
1728 }
1729 up_write(&sb->s_umount);
1730 return ret;
1731}
1732
Al Viro1ab59732011-11-24 21:35:16 -05001733static int do_umount(struct mount *mnt, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
Al Viro1ab59732011-11-24 21:35:16 -05001735 struct super_block *sb = mnt->mnt.mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 int retval;
1737
Al Viro1ab59732011-11-24 21:35:16 -05001738 retval = security_sb_umount(&mnt->mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 if (retval)
1740 return retval;
1741
1742 /*
1743 * Allow userspace to request a mountpoint be expired rather than
1744 * unmounting unconditionally. Unmount only happens if:
1745 * (1) the mark is already set (the mark is cleared by mntput())
1746 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1747 */
1748 if (flags & MNT_EXPIRE) {
Al Viro1ab59732011-11-24 21:35:16 -05001749 if (&mnt->mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 flags & (MNT_FORCE | MNT_DETACH))
1751 return -EINVAL;
1752
Nick Pigginb3e19d92011-01-07 17:50:11 +11001753 /*
1754 * probably don't strictly need the lock here if we examined
1755 * all race cases, but it's a slowpath.
1756 */
Al Viro719ea2f2013-09-29 11:24:49 -04001757 lock_mount_hash();
Al Viro83adc752011-11-24 22:37:54 -05001758 if (mnt_get_count(mnt) != 2) {
Al Viro719ea2f2013-09-29 11:24:49 -04001759 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 return -EBUSY;
Nick Pigginb3e19d92011-01-07 17:50:11 +11001761 }
Al Viro719ea2f2013-09-29 11:24:49 -04001762 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Al Viro863d6842011-11-25 00:57:42 -05001764 if (!xchg(&mnt->mnt_expiry_mark, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 return -EAGAIN;
1766 }
1767
1768 /*
1769 * If we may have to abort operations to get out of this
1770 * mount, and they will themselves hold resources we must
1771 * allow the fs to do things. In the Unix tradition of
1772 * 'Gee thats tricky lets do it in userspace' the umount_begin
1773 * might fail to complete on the first run through as other tasks
1774 * must return, and the like. Thats for the mount program to worry
1775 * about for the moment.
1776 */
1777
Al Viro42faad92008-04-24 07:21:56 -04001778 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001779 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
1782 /*
1783 * No sense to grab the lock for this test, but test itself looks
1784 * somewhat bogus. Suggestions for better replacement?
1785 * Ho-hum... In principle, we might treat that as umount + switch
1786 * to rootfs. GC would eventually take care of the old vfsmount.
1787 * Actually it makes sense, especially if rootfs would contain a
1788 * /reboot - static binary that would close all descriptors and
1789 * call reboot(9). Then init(8) could umount root and exec /reboot.
1790 */
Al Viro1ab59732011-11-24 21:35:16 -05001791 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 /*
1793 * Special case for "unmounting" root ...
1794 * we just try to remount it readonly.
1795 */
Eric W. Biedermanbc6155d2017-09-18 17:58:08 -05001796 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
Andy Lutomirskia1480dc2014-10-08 12:32:47 -07001797 return -EPERM;
David Howells8d0347f2018-11-04 09:28:36 -05001798 return do_umount_root(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 }
1800
Al Viro97216be2013-03-16 15:12:40 -04001801 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04001802 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001804 /* Recheck MNT_LOCKED with the locks held */
1805 retval = -EINVAL;
1806 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1807 goto out;
1808
1809 event++;
Al Viro48a066e2013-09-29 22:06:07 -04001810 if (flags & MNT_DETACH) {
Al Viro1a4eeaf2011-11-25 02:19:55 -05001811 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001812 umount_tree(mnt, UMOUNT_PROPAGATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 retval = 0;
Al Viro48a066e2013-09-29 22:06:07 -04001814 } else {
1815 shrink_submounts(mnt);
1816 retval = -EBUSY;
1817 if (!propagate_mount_busy(mnt, 2)) {
1818 if (!list_empty(&mnt->mnt_list))
Eric W. Biedermane819f152014-12-24 07:20:01 -06001819 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Viro48a066e2013-09-29 22:06:07 -04001820 retval = 0;
1821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 }
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001823out:
Al Viro719ea2f2013-09-29 11:24:49 -04001824 unlock_mount_hash();
Al Viroe3197d82013-03-16 14:35:16 -04001825 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 return retval;
1827}
1828
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001829/*
1830 * __detach_mounts - lazily unmount all mounts on the specified dentry
1831 *
1832 * During unlink, rmdir, and d_drop it is possible to loose the path
1833 * to an existing mountpoint, and wind up leaking the mount.
1834 * detach_mounts allows lazily unmounting those mounts instead of
1835 * leaking them.
1836 *
1837 * The caller may hold dentry->d_inode->i_mutex.
1838 */
1839void __detach_mounts(struct dentry *dentry)
1840{
1841 struct mountpoint *mp;
1842 struct mount *mnt;
1843
1844 namespace_lock();
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001845 lock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001846 mp = lookup_mountpoint(dentry);
Al Viroadc9b5c2019-06-29 12:06:51 -04001847 if (!mp)
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001848 goto out_unlock;
1849
Andrey Ulanove06b9332016-04-15 14:24:41 -07001850 event++;
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001851 while (!hlist_empty(&mp->m_list)) {
1852 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001853 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
Eric W. Biedermanfe78fcc82015-07-17 14:54:27 -05001854 umount_mnt(mnt);
Al Viro56cbb4292019-07-04 16:57:51 -04001855 hlist_add_head(&mnt->mnt_umount, &unmounted);
Eric W. Biedermance07d892014-12-23 21:37:03 -06001856 }
Eric W. Biedermane0c9c0a2015-04-01 18:30:06 -05001857 else umount_tree(mnt, UMOUNT_CONNECTED);
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001858 }
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001859 put_mountpoint(mp);
1860out_unlock:
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001861 unlock_mount_hash();
Eric W. Biederman80b5dce2013-10-03 01:31:18 -07001862 namespace_unlock();
1863}
1864
David Howellsdd111b32017-07-04 17:25:09 +01001865/*
Al Viro9b40bc92013-02-22 22:45:42 -05001866 * Is the caller allowed to modify his namespace?
1867 */
Al Viroa5f85d72022-03-01 00:05:29 -05001868bool may_mount(void)
Al Viro9b40bc92013-02-22 22:45:42 -05001869{
1870 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1871}
1872
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04001873static void warn_mandlock(void)
Jeff Layton9e8925b2015-11-16 09:49:34 -05001874{
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04001875 pr_warn_once("=======================================================\n"
1876 "WARNING: The mand mount option has been deprecated and\n"
1877 " and is ignored by this kernel. Remove the mand\n"
1878 " option from the mount to silence this warning.\n"
1879 "=======================================================\n");
Jeff Layton9e8925b2015-11-16 09:49:34 -05001880}
1881
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001882static int can_umount(const struct path *path, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001884 struct mount *mnt = real_mount(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Al Viro9b40bc92013-02-22 22:45:42 -05001886 if (!may_mount())
1887 return -EPERM;
Christoph Hellwig41525f52020-07-21 10:54:34 +02001888 if (path->dentry != path->mnt->mnt_root)
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001889 return -EINVAL;
Al Viro143c8c92011-11-25 00:46:35 -05001890 if (!check_mnt(mnt))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001891 return -EINVAL;
Eric W. Biederman25d202e2018-10-22 10:21:38 -05001892 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001893 return -EINVAL;
Eric W. Biedermanb2f5d4d2014-10-04 14:44:03 -07001894 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001895 return -EPERM;
1896 return 0;
1897}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Al Viroa0a6df92021-01-04 15:25:34 -05001899// caller is responsible for flags being sane
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001900int path_umount(struct path *path, int flags)
1901{
1902 struct mount *mnt = real_mount(path->mnt);
1903 int ret;
1904
1905 ret = can_umount(path, flags);
1906 if (!ret)
1907 ret = do_umount(mnt, flags);
1908
Jan Blunck429731b2008-02-14 19:34:31 -08001909 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Christoph Hellwig41525f52020-07-21 10:54:34 +02001910 dput(path->dentry);
Al Viro900148d2011-11-25 00:33:11 -05001911 mntput_no_expire(mnt);
Christoph Hellwig25ccd24f2020-08-06 16:07:10 +02001912 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913}
1914
Christoph Hellwig09267de2020-07-23 08:23:08 +02001915static int ksys_umount(char __user *name, int flags)
Christoph Hellwig41525f52020-07-21 10:54:34 +02001916{
1917 int lookup_flags = LOOKUP_MOUNTPOINT;
1918 struct path path;
1919 int ret;
1920
Al Viroa0a6df92021-01-04 15:25:34 -05001921 // basic validity checks done first
1922 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1923 return -EINVAL;
1924
Christoph Hellwig41525f52020-07-21 10:54:34 +02001925 if (!(flags & UMOUNT_NOFOLLOW))
1926 lookup_flags |= LOOKUP_FOLLOW;
1927 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1928 if (ret)
1929 return ret;
1930 return path_umount(&path, flags);
1931}
1932
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001933SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1934{
1935 return ksys_umount(name, flags);
1936}
1937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1939
1940/*
Ram Paib58fed82005-11-07 17:16:09 -05001941 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001943SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944{
Dominik Brodowski3a18ef52018-03-11 11:34:40 +01001945 return ksys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
1947
1948#endif
1949
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001950static bool is_mnt_ns_file(struct dentry *dentry)
Eric W. Biederman8823c072010-03-07 18:49:36 -08001951{
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001952 /* Is this a proxy for a mount namespace? */
Al Viroe149ed22014-11-01 10:57:28 -04001953 return dentry->d_op == &ns_dentry_operations &&
1954 dentry->d_fsdata == &mntns_operations;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001955}
1956
Eric Biggers213921f2020-01-04 12:59:55 -08001957static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
Al Viro58be28252014-11-01 00:00:23 -04001958{
1959 return container_of(ns, struct mnt_namespace, ns);
1960}
1961
Christian Brauner303cc572020-05-05 16:04:31 +02001962struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1963{
1964 return &mnt->ns;
1965}
1966
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001967static bool mnt_ns_loop(struct dentry *dentry)
1968{
1969 /* Could bind mounting the mount namespace inode cause a
1970 * mount namespace loop?
1971 */
1972 struct mnt_namespace *mnt_ns;
1973 if (!is_mnt_ns_file(dentry))
1974 return false;
1975
Al Virof77c8012014-11-01 03:13:17 -04001976 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
Eric W. Biederman8823c072010-03-07 18:49:36 -08001977 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1978}
1979
Al Viro87129cc2011-11-24 21:24:27 -05001980struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001981 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
Al Viro84d171922013-03-15 10:53:28 -04001983 struct mount *res, *p, *q, *r, *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07001985 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1986 return ERR_PTR(-EINVAL);
1987
1988 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
David Howellsbe34d1a2012-06-25 12:55:18 +01001989 return ERR_PTR(-EINVAL);
Ram Pai9676f0c2005-11-07 17:21:20 -05001990
Ram Pai36341f62005-11-07 17:17:22 -05001991 res = q = clone_mnt(mnt, dentry, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01001992 if (IS_ERR(q))
1993 return q;
1994
Al Viroa73324d2011-11-24 22:25:07 -05001995 q->mnt_mountpoint = mnt->mnt_mountpoint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997 p = mnt;
Al Viro6b41d532011-11-24 23:24:33 -05001998 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Al Viro315fc832011-11-24 18:57:30 -05001999 struct mount *s;
Jan Blunck7ec02ef2008-04-29 00:59:40 -07002000 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 continue;
2002
Al Viro909b0a82011-11-25 03:06:56 -05002003 for (s = r; s; s = next_mnt(s, r)) {
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002004 if (!(flag & CL_COPY_UNBINDABLE) &&
2005 IS_MNT_UNBINDABLE(s)) {
Eric W. Biedermandf7342b2018-10-25 09:04:18 -05002006 if (s->mnt.mnt_flags & MNT_LOCKED) {
2007 /* Both unbindable and locked. */
2008 q = ERR_PTR(-EPERM);
2009 goto out;
2010 } else {
2011 s = skip_mnt_tree(s);
2012 continue;
2013 }
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002014 }
2015 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2016 is_mnt_ns_file(s->mnt.mnt_root)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05002017 s = skip_mnt_tree(s);
2018 continue;
2019 }
Al Viro0714a532011-11-24 22:19:58 -05002020 while (p != s->mnt_parent) {
2021 p = p->mnt_parent;
2022 q = q->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
Al Viro87129cc2011-11-24 21:24:27 -05002024 p = s;
Al Viro84d171922013-03-15 10:53:28 -04002025 parent = q;
Al Viro87129cc2011-11-24 21:24:27 -05002026 q = clone_mnt(p, p->mnt.mnt_root, flag);
David Howellsbe34d1a2012-06-25 12:55:18 +01002027 if (IS_ERR(q))
2028 goto out;
Al Viro719ea2f2013-09-29 11:24:49 -04002029 lock_mount_hash();
Al Viro1a4eeaf2011-11-25 02:19:55 -05002030 list_add_tail(&q->mnt_list, &res->mnt_list);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002031 attach_mnt(q, parent, p->mnt_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04002032 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 }
2034 }
2035 return res;
David Howellsbe34d1a2012-06-25 12:55:18 +01002036out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (res) {
Al Viro719ea2f2013-09-29 11:24:49 -04002038 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002039 umount_tree(res, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002040 unlock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
David Howellsbe34d1a2012-06-25 12:55:18 +01002042 return q;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043}
2044
David Howellsbe34d1a2012-06-25 12:55:18 +01002045/* Caller should check returned pointer for errors */
2046
Al Viroca71cf72016-11-20 19:45:28 -05002047struct vfsmount *collect_mounts(const struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04002048{
Al Virocb338d02011-11-24 20:55:08 -05002049 struct mount *tree;
Al Viro97216be2013-03-16 15:12:40 -04002050 namespace_lock();
Eric W. Biedermancd4a4012015-01-07 14:28:26 -06002051 if (!check_mnt(real_mount(path->mnt)))
2052 tree = ERR_PTR(-EINVAL);
2053 else
2054 tree = copy_tree(real_mount(path->mnt), path->dentry,
2055 CL_COPY_ALL | CL_PRIVATE);
Al Viro328e6d92013-03-16 14:49:45 -04002056 namespace_unlock();
David Howellsbe34d1a2012-06-25 12:55:18 +01002057 if (IS_ERR(tree))
Dan Carpenter52e220d2013-08-14 12:44:39 +03002058 return ERR_CAST(tree);
David Howellsbe34d1a2012-06-25 12:55:18 +01002059 return &tree->mnt;
Al Viro8aec0802007-06-07 12:20:32 -04002060}
2061
Al Viroa07b2002018-11-05 17:40:30 +00002062static void free_mnt_ns(struct mnt_namespace *);
2063static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2064
2065void dissolve_on_fput(struct vfsmount *mnt)
2066{
2067 struct mnt_namespace *ns;
2068 namespace_lock();
2069 lock_mount_hash();
2070 ns = real_mount(mnt)->mnt_ns;
David Howells44dfd842018-11-05 17:40:31 +00002071 if (ns) {
2072 if (is_anon_ns(ns))
2073 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2074 else
2075 ns = NULL;
2076 }
Al Viroa07b2002018-11-05 17:40:30 +00002077 unlock_mount_hash();
2078 namespace_unlock();
David Howells44dfd842018-11-05 17:40:31 +00002079 if (ns)
2080 free_mnt_ns(ns);
Al Viroa07b2002018-11-05 17:40:30 +00002081}
2082
Al Viro8aec0802007-06-07 12:20:32 -04002083void drop_collected_mounts(struct vfsmount *mnt)
2084{
Al Viro97216be2013-03-16 15:12:40 -04002085 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04002086 lock_mount_hash();
Eric W. Biederman9c8e0a12018-10-25 12:05:11 -05002087 umount_tree(real_mount(mnt), 0);
Al Viro719ea2f2013-09-29 11:24:49 -04002088 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04002089 namespace_unlock();
Al Viro8aec0802007-06-07 12:20:32 -04002090}
2091
Miklos Szeredi427215d2021-08-09 10:19:47 +02002092static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2093{
2094 struct mount *child;
2095
2096 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2097 if (!is_subdir(child->mnt_mountpoint, dentry))
2098 continue;
2099
2100 if (child->mnt.mnt_flags & MNT_LOCKED)
2101 return true;
2102 }
2103 return false;
2104}
2105
Miklos Szeredic771d682014-10-24 00:14:36 +02002106/**
2107 * clone_private_mount - create a private clone of a path
Randy Dunlap1f287bc2021-03-17 19:52:25 -07002108 * @path: path to clone
Miklos Szeredic771d682014-10-24 00:14:36 +02002109 *
Randy Dunlap1f287bc2021-03-17 19:52:25 -07002110 * This creates a new vfsmount, which will be the clone of @path. The new mount
2111 * will not be attached anywhere in the namespace and will be private (i.e.
2112 * changes to the originating mount won't be propagated into this).
Miklos Szeredic771d682014-10-24 00:14:36 +02002113 *
2114 * Release with mntput().
2115 */
Al Viroca71cf72016-11-20 19:45:28 -05002116struct vfsmount *clone_private_mount(const struct path *path)
Miklos Szeredic771d682014-10-24 00:14:36 +02002117{
2118 struct mount *old_mnt = real_mount(path->mnt);
2119 struct mount *new_mnt;
2120
Miklos Szeredi427215d2021-08-09 10:19:47 +02002121 down_read(&namespace_sem);
Miklos Szeredic771d682014-10-24 00:14:36 +02002122 if (IS_MNT_UNBINDABLE(old_mnt))
Miklos Szeredi427215d2021-08-09 10:19:47 +02002123 goto invalid;
2124
2125 if (!check_mnt(old_mnt))
2126 goto invalid;
2127
2128 if (has_locked_children(old_mnt, path->dentry))
2129 goto invalid;
Miklos Szeredic771d682014-10-24 00:14:36 +02002130
Miklos Szeredic771d682014-10-24 00:14:36 +02002131 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
Miklos Szeredi427215d2021-08-09 10:19:47 +02002132 up_read(&namespace_sem);
2133
Miklos Szeredic771d682014-10-24 00:14:36 +02002134 if (IS_ERR(new_mnt))
2135 return ERR_CAST(new_mnt);
2136
Miklos Szeredidf820f82020-06-04 10:48:19 +02002137 /* Longterm mount to be removed by kern_unmount*() */
2138 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2139
Miklos Szeredic771d682014-10-24 00:14:36 +02002140 return &new_mnt->mnt;
Miklos Szeredi427215d2021-08-09 10:19:47 +02002141
2142invalid:
2143 up_read(&namespace_sem);
2144 return ERR_PTR(-EINVAL);
Miklos Szeredic771d682014-10-24 00:14:36 +02002145}
2146EXPORT_SYMBOL_GPL(clone_private_mount);
2147
Al Viro1f707132010-01-30 22:51:25 -05002148int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2149 struct vfsmount *root)
2150{
Al Viro1a4eeaf2011-11-25 02:19:55 -05002151 struct mount *mnt;
Al Viro1f707132010-01-30 22:51:25 -05002152 int res = f(root, arg);
2153 if (res)
2154 return res;
Al Viro1a4eeaf2011-11-25 02:19:55 -05002155 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2156 res = f(&mnt->mnt, arg);
Al Viro1f707132010-01-30 22:51:25 -05002157 if (res)
2158 return res;
2159 }
2160 return 0;
2161}
2162
Al Viro3bd045c2019-01-30 13:15:45 -05002163static void lock_mnt_tree(struct mount *mnt)
2164{
2165 struct mount *p;
2166
2167 for (p = mnt; p; p = next_mnt(p, mnt)) {
2168 int flags = p->mnt.mnt_flags;
2169 /* Don't allow unprivileged users to change mount flags */
2170 flags |= MNT_LOCK_ATIME;
2171
2172 if (flags & MNT_READONLY)
2173 flags |= MNT_LOCK_READONLY;
2174
2175 if (flags & MNT_NODEV)
2176 flags |= MNT_LOCK_NODEV;
2177
2178 if (flags & MNT_NOSUID)
2179 flags |= MNT_LOCK_NOSUID;
2180
2181 if (flags & MNT_NOEXEC)
2182 flags |= MNT_LOCK_NOEXEC;
2183 /* Don't allow unprivileged users to reveal what is under a mount */
2184 if (list_empty(&p->mnt_expire))
2185 flags |= MNT_LOCKED;
2186 p->mnt.mnt_flags = flags;
2187 }
2188}
2189
Al Viro4b8b21f2011-11-24 19:54:23 -05002190static void cleanup_group_ids(struct mount *mnt, struct mount *end)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002191{
Al Viro315fc832011-11-24 18:57:30 -05002192 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002193
Al Viro909b0a82011-11-25 03:06:56 -05002194 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002195 if (p->mnt_group_id && !IS_MNT_SHARED(p))
Al Viro4b8b21f2011-11-24 19:54:23 -05002196 mnt_release_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002197 }
2198}
2199
Al Viro4b8b21f2011-11-24 19:54:23 -05002200static int invent_group_ids(struct mount *mnt, bool recurse)
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002201{
Al Viro315fc832011-11-24 18:57:30 -05002202 struct mount *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002203
Al Viro909b0a82011-11-25 03:06:56 -05002204 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
Al Virofc7be132011-11-25 01:05:37 -05002205 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002206 int err = mnt_alloc_group_id(p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002207 if (err) {
Al Viro4b8b21f2011-11-24 19:54:23 -05002208 cleanup_group_ids(mnt, p);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002209 return err;
2210 }
2211 }
2212 }
2213
2214 return 0;
2215}
2216
Eric W. Biedermand2921682016-09-28 00:27:17 -05002217int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2218{
2219 unsigned int max = READ_ONCE(sysctl_mount_max);
Al Viro124f75f2022-02-13 22:42:30 -05002220 unsigned int mounts = 0;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002221 struct mount *p;
2222
Al Viro124f75f2022-02-13 22:42:30 -05002223 if (ns->mounts >= max)
2224 return -ENOSPC;
2225 max -= ns->mounts;
2226 if (ns->pending_mounts >= max)
2227 return -ENOSPC;
2228 max -= ns->pending_mounts;
2229
Eric W. Biedermand2921682016-09-28 00:27:17 -05002230 for (p = mnt; p; p = next_mnt(p, mnt))
2231 mounts++;
2232
Al Viro124f75f2022-02-13 22:42:30 -05002233 if (mounts > max)
Eric W. Biedermand2921682016-09-28 00:27:17 -05002234 return -ENOSPC;
2235
Al Viro124f75f2022-02-13 22:42:30 -05002236 ns->pending_mounts += mounts;
Eric W. Biedermand2921682016-09-28 00:27:17 -05002237 return 0;
2238}
2239
Ram Paib90fa9a2005-11-07 17:19:50 -05002240/*
2241 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05002242 * @nd : place the mount tree @source_mnt is attached
2243 * @parent_nd : if non-null, detach the source_mnt from its parent and
2244 * store the parent mount and mountpoint dentry.
2245 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05002246 *
2247 * NOTE: in the table below explains the semantics when a source mount
2248 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05002249 * ---------------------------------------------------------------------------
2250 * | BIND MOUNT OPERATION |
2251 * |**************************************************************************
2252 * | source-->| shared | private | slave | unbindable |
2253 * | dest | | | | |
2254 * | | | | | | |
2255 * | v | | | | |
2256 * |**************************************************************************
2257 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2258 * | | | | | |
2259 * |non-shared| shared (+) | private | slave (*) | invalid |
2260 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05002261 * A bind operation clones the source mount and mounts the clone on the
2262 * destination mount.
2263 *
2264 * (++) the cloned mount is propagated to all the mounts in the propagation
2265 * tree of the destination mount and the cloned mount is added to
2266 * the peer group of the source mount.
2267 * (+) the cloned mount is created under the destination mount and is marked
2268 * as shared. The cloned mount is added to the peer group of the source
2269 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05002270 * (+++) the mount is propagated to all the mounts in the propagation tree
2271 * of the destination mount and the cloned mount is made slave
2272 * of the same master as that of the source mount. The cloned mount
2273 * is marked as 'shared and slave'.
2274 * (*) the cloned mount is made a slave of the same master as that of the
2275 * source mount.
2276 *
Ram Pai9676f0c2005-11-07 17:21:20 -05002277 * ---------------------------------------------------------------------------
2278 * | MOVE MOUNT OPERATION |
2279 * |**************************************************************************
2280 * | source-->| shared | private | slave | unbindable |
2281 * | dest | | | | |
2282 * | | | | | | |
2283 * | v | | | | |
2284 * |**************************************************************************
2285 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2286 * | | | | | |
2287 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2288 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05002289 *
2290 * (+) the mount is moved to the destination. And is then propagated to
2291 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05002292 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05002293 * (+++) the mount is moved to the destination and is then propagated to
2294 * all the mounts belonging to the destination mount's propagation tree.
2295 * the mount is marked as 'shared and slave'.
2296 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05002297 *
2298 * if the source mount is a tree, the operations explained above is
2299 * applied to each mount in the tree.
2300 * Must be called without spinlocks held, since this function can sleep
2301 * in allocations.
2302 */
Al Viro0fb54e52011-11-24 19:59:16 -05002303static int attach_recursive_mnt(struct mount *source_mnt,
Al Viro84d171922013-03-15 10:53:28 -04002304 struct mount *dest_mnt,
2305 struct mountpoint *dest_mp,
Al Viro2763d112019-06-30 19:18:53 -04002306 bool moving)
Ram Paib90fa9a2005-11-07 17:19:50 -05002307{
Al Viro3bd045c2019-01-30 13:15:45 -05002308 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
Al Viro38129a12014-03-20 21:10:51 -04002309 HLIST_HEAD(tree_list);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002310 struct mnt_namespace *ns = dest_mnt->mnt_ns;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002311 struct mountpoint *smp;
Al Viro315fc832011-11-24 18:57:30 -05002312 struct mount *child, *p;
Al Viro38129a12014-03-20 21:10:51 -04002313 struct hlist_node *n;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002314 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002315
Eric W. Biederman1064f872017-01-20 18:28:35 +13002316 /* Preallocate a mountpoint in case the new mounts need
2317 * to be tucked under other mounts.
2318 */
2319 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2320 if (IS_ERR(smp))
2321 return PTR_ERR(smp);
2322
Eric W. Biedermand2921682016-09-28 00:27:17 -05002323 /* Is there space to add these mounts to the mount namespace? */
Al Viro2763d112019-06-30 19:18:53 -04002324 if (!moving) {
Eric W. Biedermand2921682016-09-28 00:27:17 -05002325 err = count_mounts(ns, source_mnt);
2326 if (err)
2327 goto out;
2328 }
2329
Al Virofc7be132011-11-25 01:05:37 -05002330 if (IS_MNT_SHARED(dest_mnt)) {
Al Viro0fb54e52011-11-24 19:59:16 -05002331 err = invent_group_ids(source_mnt, true);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002332 if (err)
2333 goto out;
Al Viro0b1b9012014-03-21 10:14:08 -04002334 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
Al Virof2ebb3a2014-02-27 09:35:45 -05002335 lock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002336 if (err)
2337 goto out_cleanup_ids;
Al Viro909b0a82011-11-25 03:06:56 -05002338 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
Al Viro0f0afb12011-11-24 20:43:10 -05002339 set_mnt_shared(p);
Al Viro0b1b9012014-03-21 10:14:08 -04002340 } else {
2341 lock_mount_hash();
Ram Paib90fa9a2005-11-07 17:19:50 -05002342 }
Al Viro2763d112019-06-30 19:18:53 -04002343 if (moving) {
2344 unhash_mnt(source_mnt);
Al Viro84d171922013-03-15 10:53:28 -04002345 attach_mnt(source_mnt, dest_mnt, dest_mp);
Al Viro143c8c92011-11-25 00:46:35 -05002346 touch_mnt_namespace(source_mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05002347 } else {
David Howells44dfd842018-11-05 17:40:31 +00002348 if (source_mnt->mnt_ns) {
2349 /* move from anon - the caller will destroy */
2350 list_del_init(&source_mnt->mnt_ns->list);
2351 }
Al Viro84d171922013-03-15 10:53:28 -04002352 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002353 commit_tree(source_mnt);
Ram Pai21444402005-11-07 17:20:03 -05002354 }
Ram Paib90fa9a2005-11-07 17:19:50 -05002355
Al Viro38129a12014-03-20 21:10:51 -04002356 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
Al Viro1d6a32a2014-03-20 20:34:43 -04002357 struct mount *q;
Al Viro38129a12014-03-20 21:10:51 -04002358 hlist_del_init(&child->mnt_hash);
Eric W. Biederman1064f872017-01-20 18:28:35 +13002359 q = __lookup_mnt(&child->mnt_parent->mnt,
2360 child->mnt_mountpoint);
2361 if (q)
2362 mnt_change_mountpoint(child, smp, q);
Al Viro3bd045c2019-01-30 13:15:45 -05002363 /* Notice when we are propagating across user namespaces */
2364 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2365 lock_mnt_tree(child);
Christian Braunerd728cf72019-06-17 23:22:14 +02002366 child->mnt.mnt_flags &= ~MNT_LOCKED;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002367 commit_tree(child);
Ram Paib90fa9a2005-11-07 17:19:50 -05002368 }
Eric W. Biederman1064f872017-01-20 18:28:35 +13002369 put_mountpoint(smp);
Al Viro719ea2f2013-09-29 11:24:49 -04002370 unlock_mount_hash();
Nick Piggin99b7db72010-08-18 04:37:39 +10002371
Ram Paib90fa9a2005-11-07 17:19:50 -05002372 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002373
2374 out_cleanup_ids:
Al Virof2ebb3a2014-02-27 09:35:45 -05002375 while (!hlist_empty(&tree_list)) {
2376 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
Eric W. Biedermand2921682016-09-28 00:27:17 -05002377 child->mnt_parent->mnt_ns->pending_mounts = 0;
Eric W. Biedermane819f152014-12-24 07:20:01 -06002378 umount_tree(child, UMOUNT_SYNC);
Al Virof2ebb3a2014-02-27 09:35:45 -05002379 }
2380 unlock_mount_hash();
Al Viro0b1b9012014-03-21 10:14:08 -04002381 cleanup_group_ids(source_mnt, NULL);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002382 out:
Eric W. Biedermand2921682016-09-28 00:27:17 -05002383 ns->pending_mounts = 0;
Eric W. Biederman1064f872017-01-20 18:28:35 +13002384
2385 read_seqlock_excl(&mount_lock);
2386 put_mountpoint(smp);
2387 read_sequnlock_excl(&mount_lock);
2388
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002389 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05002390}
2391
Al Viro84d171922013-03-15 10:53:28 -04002392static struct mountpoint *lock_mount(struct path *path)
Al Virob12cea92011-03-18 08:55:38 -04002393{
2394 struct vfsmount *mnt;
Al Viro84d171922013-03-15 10:53:28 -04002395 struct dentry *dentry = path->dentry;
Al Virob12cea92011-03-18 08:55:38 -04002396retry:
Al Viro59551022016-01-22 15:40:57 -05002397 inode_lock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002398 if (unlikely(cant_mount(dentry))) {
Al Viro59551022016-01-22 15:40:57 -05002399 inode_unlock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002400 return ERR_PTR(-ENOENT);
Al Virob12cea92011-03-18 08:55:38 -04002401 }
Al Viro97216be2013-03-16 15:12:40 -04002402 namespace_lock();
Al Virob12cea92011-03-18 08:55:38 -04002403 mnt = lookup_mnt(path);
Al Viro84d171922013-03-15 10:53:28 -04002404 if (likely(!mnt)) {
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002405 struct mountpoint *mp = get_mountpoint(dentry);
Al Viro84d171922013-03-15 10:53:28 -04002406 if (IS_ERR(mp)) {
Al Viro97216be2013-03-16 15:12:40 -04002407 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002408 inode_unlock(dentry->d_inode);
Al Viro84d171922013-03-15 10:53:28 -04002409 return mp;
2410 }
2411 return mp;
2412 }
Al Viro97216be2013-03-16 15:12:40 -04002413 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002414 inode_unlock(path->dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002415 path_put(path);
2416 path->mnt = mnt;
Al Viro84d171922013-03-15 10:53:28 -04002417 dentry = path->dentry = dget(mnt->mnt_root);
Al Virob12cea92011-03-18 08:55:38 -04002418 goto retry;
2419}
2420
Al Viro84d171922013-03-15 10:53:28 -04002421static void unlock_mount(struct mountpoint *where)
Al Virob12cea92011-03-18 08:55:38 -04002422{
Al Viro84d171922013-03-15 10:53:28 -04002423 struct dentry *dentry = where->m_dentry;
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002424
2425 read_seqlock_excl(&mount_lock);
Al Viro84d171922013-03-15 10:53:28 -04002426 put_mountpoint(where);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13002427 read_sequnlock_excl(&mount_lock);
2428
Al Viro328e6d92013-03-16 14:49:45 -04002429 namespace_unlock();
Al Viro59551022016-01-22 15:40:57 -05002430 inode_unlock(dentry->d_inode);
Al Virob12cea92011-03-18 08:55:38 -04002431}
2432
Al Viro84d171922013-03-15 10:53:28 -04002433static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
David Howellse462ec52017-07-17 08:45:35 +01002435 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 return -EINVAL;
2437
David Howellse36cb0b2015-01-29 12:02:35 +00002438 if (d_is_dir(mp->m_dentry) !=
2439 d_is_dir(mnt->mnt.mnt_root))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 return -ENOTDIR;
2441
Al Viro2763d112019-06-30 19:18:53 -04002442 return attach_recursive_mnt(mnt, p, mp, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443}
2444
2445/*
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002446 * Sanity check the flags to change_mnt_propagation.
2447 */
2448
David Howellse462ec52017-07-17 08:45:35 +01002449static int flags_to_propagation_type(int ms_flags)
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002450{
David Howellse462ec52017-07-17 08:45:35 +01002451 int type = ms_flags & ~(MS_REC | MS_SILENT);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002452
2453 /* Fail if any non-propagation flags are set */
2454 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2455 return 0;
2456 /* Only one propagation flag should be set */
2457 if (!is_power_of_2(type))
2458 return 0;
2459 return type;
2460}
2461
2462/*
Ram Pai07b20882005-11-07 17:19:07 -05002463 * recursively change the type of the mountpoint.
2464 */
David Howellse462ec52017-07-17 08:45:35 +01002465static int do_change_type(struct path *path, int ms_flags)
Ram Pai07b20882005-11-07 17:19:07 -05002466{
Al Viro315fc832011-11-24 18:57:30 -05002467 struct mount *m;
Al Viro4b8b21f2011-11-24 19:54:23 -05002468 struct mount *mnt = real_mount(path->mnt);
David Howellse462ec52017-07-17 08:45:35 +01002469 int recurse = ms_flags & MS_REC;
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002470 int type;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002471 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05002472
Al Viro2d92ab32008-08-02 00:51:11 -04002473 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05002474 return -EINVAL;
2475
David Howellse462ec52017-07-17 08:45:35 +01002476 type = flags_to_propagation_type(ms_flags);
Valerie Aurora7a2e8a82010-08-26 11:07:22 -07002477 if (!type)
2478 return -EINVAL;
2479
Al Viro97216be2013-03-16 15:12:40 -04002480 namespace_lock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002481 if (type == MS_SHARED) {
2482 err = invent_group_ids(mnt, recurse);
2483 if (err)
2484 goto out_unlock;
2485 }
2486
Al Viro719ea2f2013-09-29 11:24:49 -04002487 lock_mount_hash();
Al Viro909b0a82011-11-25 03:06:56 -05002488 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
Al Viro0f0afb12011-11-24 20:43:10 -05002489 change_mnt_propagation(m, type);
Al Viro719ea2f2013-09-29 11:24:49 -04002490 unlock_mount_hash();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002491
2492 out_unlock:
Al Viro97216be2013-03-16 15:12:40 -04002493 namespace_unlock();
Miklos Szeredi719f5d72008-03-27 13:06:23 +01002494 return err;
Ram Pai07b20882005-11-07 17:19:07 -05002495}
2496
Al Viroa07b2002018-11-05 17:40:30 +00002497static struct mount *__do_loopback(struct path *old_path, int recurse)
2498{
2499 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2500
2501 if (IS_MNT_UNBINDABLE(old))
2502 return mnt;
2503
2504 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2505 return mnt;
2506
2507 if (!recurse && has_locked_children(old, old_path->dentry))
2508 return mnt;
2509
2510 if (recurse)
2511 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2512 else
2513 mnt = clone_mnt(old, old_path->dentry, 0);
2514
2515 if (!IS_ERR(mnt))
2516 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2517
2518 return mnt;
2519}
2520
Ram Pai07b20882005-11-07 17:19:07 -05002521/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 * do loopback mount.
2523 */
Al Viro808d4e32012-10-11 11:42:01 -04002524static int do_loopback(struct path *path, const char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08002525 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526{
Al Viro2d92ab32008-08-02 00:51:11 -04002527 struct path old_path;
Al Viroa07b2002018-11-05 17:40:30 +00002528 struct mount *mnt = NULL, *parent;
Al Viro84d171922013-03-15 10:53:28 -04002529 struct mountpoint *mp;
Al Viro57eccb82013-02-22 22:49:10 -05002530 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 if (!old_name || !*old_name)
2532 return -EINVAL;
Trond Myklebust815d4052011-09-26 20:36:09 -04002533 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 if (err)
2535 return err;
2536
Eric W. Biederman8823c072010-03-07 18:49:36 -08002537 err = -EINVAL;
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07002538 if (mnt_ns_loop(old_path.dentry))
David Howellsdd111b32017-07-04 17:25:09 +01002539 goto out;
Eric W. Biederman8823c072010-03-07 18:49:36 -08002540
Al Viro84d171922013-03-15 10:53:28 -04002541 mp = lock_mount(path);
Al Viroa07b2002018-11-05 17:40:30 +00002542 if (IS_ERR(mp)) {
2543 err = PTR_ERR(mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08002544 goto out;
Al Viroa07b2002018-11-05 17:40:30 +00002545 }
Ram Pai9676f0c2005-11-07 17:21:20 -05002546
Al Viro84d171922013-03-15 10:53:28 -04002547 parent = real_mount(path->mnt);
Al Viroe149ed22014-11-01 10:57:28 -04002548 if (!check_mnt(parent))
2549 goto out2;
2550
Al Viroa07b2002018-11-05 17:40:30 +00002551 mnt = __do_loopback(&old_path, recurse);
David Howellsbe34d1a2012-06-25 12:55:18 +01002552 if (IS_ERR(mnt)) {
2553 err = PTR_ERR(mnt);
Andrey Vagine9c5d8a2013-04-09 17:33:29 +04002554 goto out2;
David Howellsbe34d1a2012-06-25 12:55:18 +01002555 }
Al Viroccd48bc2005-11-07 17:15:04 -05002556
Al Viro84d171922013-03-15 10:53:28 -04002557 err = graft_tree(mnt, parent, mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002558 if (err) {
Al Viro719ea2f2013-09-29 11:24:49 -04002559 lock_mount_hash();
Eric W. Biedermane819f152014-12-24 07:20:01 -06002560 umount_tree(mnt, UMOUNT_SYNC);
Al Viro719ea2f2013-09-29 11:24:49 -04002561 unlock_mount_hash();
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05002562 }
Al Virob12cea92011-03-18 08:55:38 -04002563out2:
Al Viro84d171922013-03-15 10:53:28 -04002564 unlock_mount(mp);
Al Viroccd48bc2005-11-07 17:15:04 -05002565out:
Al Viro2d92ab32008-08-02 00:51:11 -04002566 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 return err;
2568}
2569
Al Viroa07b2002018-11-05 17:40:30 +00002570static struct file *open_detached_copy(struct path *path, bool recursive)
2571{
2572 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2573 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2574 struct mount *mnt, *p;
2575 struct file *file;
2576
2577 if (IS_ERR(ns))
2578 return ERR_CAST(ns);
2579
2580 namespace_lock();
2581 mnt = __do_loopback(path, recursive);
2582 if (IS_ERR(mnt)) {
2583 namespace_unlock();
2584 free_mnt_ns(ns);
2585 return ERR_CAST(mnt);
2586 }
2587
2588 lock_mount_hash();
2589 for (p = mnt; p; p = next_mnt(p, mnt)) {
2590 p->mnt_ns = ns;
2591 ns->mounts++;
2592 }
2593 ns->root = mnt;
2594 list_add_tail(&ns->list, &mnt->mnt_list);
2595 mntget(&mnt->mnt);
2596 unlock_mount_hash();
2597 namespace_unlock();
2598
2599 mntput(path->mnt);
2600 path->mnt = &mnt->mnt;
2601 file = dentry_open(path, O_PATH, current_cred());
2602 if (IS_ERR(file))
2603 dissolve_on_fput(path->mnt);
2604 else
2605 file->f_mode |= FMODE_NEED_UNMOUNT;
2606 return file;
2607}
2608
Ben Dooks2658ce02019-10-15 11:35:02 +01002609SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
Al Viroa07b2002018-11-05 17:40:30 +00002610{
2611 struct file *file;
2612 struct path path;
2613 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2614 bool detached = flags & OPEN_TREE_CLONE;
2615 int error;
2616 int fd;
2617
2618 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2619
2620 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2621 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2622 OPEN_TREE_CLOEXEC))
2623 return -EINVAL;
2624
2625 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2626 return -EINVAL;
2627
2628 if (flags & AT_NO_AUTOMOUNT)
2629 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2630 if (flags & AT_SYMLINK_NOFOLLOW)
2631 lookup_flags &= ~LOOKUP_FOLLOW;
2632 if (flags & AT_EMPTY_PATH)
2633 lookup_flags |= LOOKUP_EMPTY;
2634
2635 if (detached && !may_mount())
2636 return -EPERM;
2637
2638 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2639 if (fd < 0)
2640 return fd;
2641
2642 error = user_path_at(dfd, filename, lookup_flags, &path);
2643 if (unlikely(error)) {
2644 file = ERR_PTR(error);
2645 } else {
2646 if (detached)
2647 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2648 else
2649 file = dentry_open(&path, O_PATH, current_cred());
2650 path_put(&path);
2651 }
2652 if (IS_ERR(file)) {
2653 put_unused_fd(fd);
2654 return PTR_ERR(file);
2655 }
2656 fd_install(fd, file);
2657 return fd;
2658}
2659
David Howells43f5e652018-11-01 23:07:25 +00002660/*
2661 * Don't allow locked mount flags to be cleared.
2662 *
2663 * No locks need to be held here while testing the various MNT_LOCK
2664 * flags because those flags can never be cleared once they are set.
2665 */
2666static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002667{
David Howells43f5e652018-11-01 23:07:25 +00002668 unsigned int fl = mnt->mnt.mnt_flags;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002669
David Howells43f5e652018-11-01 23:07:25 +00002670 if ((fl & MNT_LOCK_READONLY) &&
2671 !(mnt_flags & MNT_READONLY))
2672 return false;
2673
2674 if ((fl & MNT_LOCK_NODEV) &&
2675 !(mnt_flags & MNT_NODEV))
2676 return false;
2677
2678 if ((fl & MNT_LOCK_NOSUID) &&
2679 !(mnt_flags & MNT_NOSUID))
2680 return false;
2681
2682 if ((fl & MNT_LOCK_NOEXEC) &&
2683 !(mnt_flags & MNT_NOEXEC))
2684 return false;
2685
2686 if ((fl & MNT_LOCK_ATIME) &&
2687 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2688 return false;
2689
2690 return true;
2691}
2692
2693static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2694{
2695 bool readonly_request = (mnt_flags & MNT_READONLY);
2696
2697 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002698 return 0;
2699
2700 if (readonly_request)
David Howells43f5e652018-11-01 23:07:25 +00002701 return mnt_make_readonly(mnt);
2702
Christian Brauner68847c92021-01-21 14:19:48 +01002703 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2704 return 0;
David Howells43f5e652018-11-01 23:07:25 +00002705}
2706
David Howells43f5e652018-11-01 23:07:25 +00002707static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2708{
David Howells43f5e652018-11-01 23:07:25 +00002709 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2710 mnt->mnt.mnt_flags = mnt_flags;
2711 touch_mnt_namespace(mnt->mnt_ns);
David Howells43f5e652018-11-01 23:07:25 +00002712}
2713
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002714static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2715{
2716 struct super_block *sb = mnt->mnt_sb;
2717
2718 if (!__mnt_is_readonly(mnt) &&
Anthony Iliopoulosa128b052022-03-22 14:39:22 -07002719 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002720 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2721 char *buf = (char *)__get_free_page(GFP_KERNEL);
2722 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2723 struct tm tm;
2724
2725 time64_to_tm(sb->s_time_max, 0, &tm);
2726
Eric Biggers0ecee662019-10-16 19:48:14 -07002727 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2728 sb->s_type->name,
2729 is_mounted(mnt) ? "remounted" : "mounted",
2730 mntpath,
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002731 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2732
2733 free_page((unsigned long)buf);
Anthony Iliopoulosa128b052022-03-22 14:39:22 -07002734 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002735 }
2736}
2737
David Howells43f5e652018-11-01 23:07:25 +00002738/*
2739 * Handle reconfiguration of the mountpoint only without alteration of the
2740 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2741 * to mount(2).
2742 */
2743static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2744{
2745 struct super_block *sb = path->mnt->mnt_sb;
2746 struct mount *mnt = real_mount(path->mnt);
2747 int ret;
2748
2749 if (!check_mnt(mnt))
2750 return -EINVAL;
2751
2752 if (path->dentry != mnt->mnt.mnt_root)
2753 return -EINVAL;
2754
2755 if (!can_change_locked_flags(mnt, mnt_flags))
2756 return -EPERM;
2757
Christian Braunere58ace12021-01-21 14:19:50 +01002758 /*
2759 * We're only checking whether the superblock is read-only not
2760 * changing it, so only take down_read(&sb->s_umount).
2761 */
2762 down_read(&sb->s_umount);
Christian Brauner68847c92021-01-21 14:19:48 +01002763 lock_mount_hash();
David Howells43f5e652018-11-01 23:07:25 +00002764 ret = change_mount_ro_state(mnt, mnt_flags);
2765 if (ret == 0)
2766 set_mount_attributes(mnt, mnt_flags);
Christian Brauner68847c92021-01-21 14:19:48 +01002767 unlock_mount_hash();
Christian Braunere58ace12021-01-21 14:19:50 +01002768 up_read(&sb->s_umount);
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002769
2770 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2771
David Howells43f5e652018-11-01 23:07:25 +00002772 return ret;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002773}
2774
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775/*
2776 * change filesystem flags. dir should be a physical root of filesystem.
2777 * If you've mounted a non-root directory somewhere and want to do remount
2778 * on it - tough luck.
2779 */
David Howellse462ec52017-07-17 08:45:35 +01002780static int do_remount(struct path *path, int ms_flags, int sb_flags,
2781 int mnt_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782{
2783 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04002784 struct super_block *sb = path->mnt->mnt_sb;
Al Viro143c8c92011-11-25 00:46:35 -05002785 struct mount *mnt = real_mount(path->mnt);
David Howells8d0347f2018-11-04 09:28:36 -05002786 struct fs_context *fc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
Al Viro143c8c92011-11-25 00:46:35 -05002788 if (!check_mnt(mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 return -EINVAL;
2790
Al Viro2d92ab32008-08-02 00:51:11 -04002791 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792 return -EINVAL;
2793
David Howells43f5e652018-11-01 23:07:25 +00002794 if (!can_change_locked_flags(mnt, mnt_flags))
Eric W. Biederman07b64552014-07-28 17:10:56 -07002795 return -EPERM;
Eric W. Biederman9566d672014-07-28 17:26:07 -07002796
David Howells8d0347f2018-11-04 09:28:36 -05002797 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2798 if (IS_ERR(fc))
2799 return PTR_ERR(fc);
Eric Parisff36fe22011-03-03 16:09:14 -05002800
Miklos Szeredib3309662020-07-14 14:45:41 +02002801 fc->oldapi = true;
David Howells8d0347f2018-11-04 09:28:36 -05002802 err = parse_monolithic_mount_data(fc, data);
2803 if (!err) {
2804 down_write(&sb->s_umount);
2805 err = -EPERM;
2806 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2807 err = reconfigure_super(fc);
Christian Brauner68847c92021-01-21 14:19:48 +01002808 if (!err) {
2809 lock_mount_hash();
David Howells8d0347f2018-11-04 09:28:36 -05002810 set_mount_attributes(mnt, mnt_flags);
Christian Brauner68847c92021-01-21 14:19:48 +01002811 unlock_mount_hash();
2812 }
David Howells8d0347f2018-11-04 09:28:36 -05002813 }
2814 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07002815 }
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07002816
2817 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2818
David Howells8d0347f2018-11-04 09:28:36 -05002819 put_fs_context(fc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 return err;
2821}
2822
Al Virocbbe3622011-11-24 20:01:19 -05002823static inline int tree_contains_unbindable(struct mount *mnt)
Ram Pai9676f0c2005-11-07 17:21:20 -05002824{
Al Viro315fc832011-11-24 18:57:30 -05002825 struct mount *p;
Al Viro909b0a82011-11-25 03:06:56 -05002826 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Virofc7be132011-11-25 01:05:37 -05002827 if (IS_MNT_UNBINDABLE(p))
Ram Pai9676f0c2005-11-07 17:21:20 -05002828 return 1;
2829 }
2830 return 0;
2831}
2832
David Howells44dfd842018-11-05 17:40:31 +00002833/*
2834 * Check that there aren't references to earlier/same mount namespaces in the
2835 * specified subtree. Such references can act as pins for mount namespaces
2836 * that aren't checked by the mount-cycle checking code, thereby allowing
2837 * cycles to be made.
2838 */
2839static bool check_for_nsfs_mounts(struct mount *subtree)
2840{
2841 struct mount *p;
2842 bool ret = false;
2843
2844 lock_mount_hash();
2845 for (p = subtree; p; p = next_mnt(p, subtree))
2846 if (mnt_ns_loop(p->mnt.mnt_root))
2847 goto out;
2848
2849 ret = true;
2850out:
2851 unlock_mount_hash();
2852 return ret;
2853}
2854
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03002855static int do_set_group(struct path *from_path, struct path *to_path)
2856{
2857 struct mount *from, *to;
2858 int err;
2859
2860 from = real_mount(from_path->mnt);
2861 to = real_mount(to_path->mnt);
2862
2863 namespace_lock();
2864
2865 err = -EINVAL;
2866 /* To and From must be mounted */
2867 if (!is_mounted(&from->mnt))
2868 goto out;
2869 if (!is_mounted(&to->mnt))
2870 goto out;
2871
2872 err = -EPERM;
2873 /* We should be allowed to modify mount namespaces of both mounts */
2874 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2875 goto out;
2876 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2877 goto out;
2878
2879 err = -EINVAL;
2880 /* To and From paths should be mount roots */
2881 if (from_path->dentry != from_path->mnt->mnt_root)
2882 goto out;
2883 if (to_path->dentry != to_path->mnt->mnt_root)
2884 goto out;
2885
2886 /* Setting sharing groups is only allowed across same superblock */
2887 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2888 goto out;
2889
2890 /* From mount root should be wider than To mount root */
2891 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2892 goto out;
2893
2894 /* From mount should not have locked children in place of To's root */
2895 if (has_locked_children(from, to->mnt.mnt_root))
2896 goto out;
2897
2898 /* Setting sharing groups is only allowed on private mounts */
2899 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2900 goto out;
2901
2902 /* From should not be private */
2903 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2904 goto out;
2905
2906 if (IS_MNT_SLAVE(from)) {
2907 struct mount *m = from->mnt_master;
2908
2909 list_add(&to->mnt_slave, &m->mnt_slave_list);
2910 to->mnt_master = m;
2911 }
2912
2913 if (IS_MNT_SHARED(from)) {
2914 to->mnt_group_id = from->mnt_group_id;
2915 list_add(&to->mnt_share, &from->mnt_share);
2916 lock_mount_hash();
2917 set_mnt_shared(to);
2918 unlock_mount_hash();
2919 }
2920
2921 err = 0;
2922out:
2923 namespace_unlock();
2924 return err;
2925}
2926
David Howells2db154b2018-11-05 17:40:30 +00002927static int do_move_mount(struct path *old_path, struct path *new_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928{
David Howells44dfd842018-11-05 17:40:31 +00002929 struct mnt_namespace *ns;
Al Viro676da582011-11-24 21:47:05 -05002930 struct mount *p;
Al Viro0fb54e52011-11-24 19:59:16 -05002931 struct mount *old;
Al Viro2763d112019-06-30 19:18:53 -04002932 struct mount *parent;
2933 struct mountpoint *mp, *old_mp;
Al Viro57eccb82013-02-22 22:49:10 -05002934 int err;
David Howells44dfd842018-11-05 17:40:31 +00002935 bool attached;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
David Howells2db154b2018-11-05 17:40:30 +00002937 mp = lock_mount(new_path);
Al Viro84d171922013-03-15 10:53:28 -04002938 if (IS_ERR(mp))
David Howells2db154b2018-11-05 17:40:30 +00002939 return PTR_ERR(mp);
David Howellscc53ce52011-01-14 18:45:26 +00002940
David Howells2db154b2018-11-05 17:40:30 +00002941 old = real_mount(old_path->mnt);
2942 p = real_mount(new_path->mnt);
Al Viro2763d112019-06-30 19:18:53 -04002943 parent = old->mnt_parent;
David Howells44dfd842018-11-05 17:40:31 +00002944 attached = mnt_has_parent(old);
Al Viro2763d112019-06-30 19:18:53 -04002945 old_mp = old->mnt_mp;
David Howells44dfd842018-11-05 17:40:31 +00002946 ns = old->mnt_ns;
Al Viro143c8c92011-11-25 00:46:35 -05002947
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 err = -EINVAL;
David Howells44dfd842018-11-05 17:40:31 +00002949 /* The mountpoint must be in our namespace. */
2950 if (!check_mnt(p))
David Howells2db154b2018-11-05 17:40:30 +00002951 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952
Eric Biggers570d7a92019-06-29 13:27:44 -07002953 /* The thing moved must be mounted... */
2954 if (!is_mounted(&old->mnt))
David Howells44dfd842018-11-05 17:40:31 +00002955 goto out;
2956
Eric Biggers570d7a92019-06-29 13:27:44 -07002957 /* ... and either ours or the root of anon namespace */
2958 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
David Howells2db154b2018-11-05 17:40:30 +00002959 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
David Howells2db154b2018-11-05 17:40:30 +00002961 if (old->mnt.mnt_flags & MNT_LOCKED)
2962 goto out;
2963
2964 if (old_path->dentry != old_path->mnt->mnt_root)
2965 goto out;
2966
2967 if (d_is_dir(new_path->dentry) !=
2968 d_is_dir(old_path->dentry))
2969 goto out;
Ram Pai21444402005-11-07 17:20:03 -05002970 /*
2971 * Don't move a mount residing in a shared parent.
2972 */
Al Viro2763d112019-06-30 19:18:53 -04002973 if (attached && IS_MNT_SHARED(parent))
David Howells2db154b2018-11-05 17:40:30 +00002974 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05002975 /*
2976 * Don't move a mount tree containing unbindable mounts to a destination
2977 * mount which is shared.
2978 */
Al Virofc7be132011-11-25 01:05:37 -05002979 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
David Howells2db154b2018-11-05 17:40:30 +00002980 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 err = -ELOOP;
David Howells44dfd842018-11-05 17:40:31 +00002982 if (!check_for_nsfs_mounts(old))
2983 goto out;
Al Virofc7be132011-11-25 01:05:37 -05002984 for (; mnt_has_parent(p); p = p->mnt_parent)
Al Viro676da582011-11-24 21:47:05 -05002985 if (p == old)
David Howells2db154b2018-11-05 17:40:30 +00002986 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
David Howells2db154b2018-11-05 17:40:30 +00002988 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
Al Viro2763d112019-06-30 19:18:53 -04002989 attached);
Jan Blunck4ac91372008-02-14 19:34:32 -08002990 if (err)
David Howells2db154b2018-11-05 17:40:30 +00002991 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993 /* if the mount is moved, it should no longer be expire
2994 * automatically */
Al Viro6776db3d2011-11-25 00:22:05 -05002995 list_del_init(&old->mnt_expire);
Al Viro2763d112019-06-30 19:18:53 -04002996 if (attached)
2997 put_mountpoint(old_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998out:
David Howells2db154b2018-11-05 17:40:30 +00002999 unlock_mount(mp);
David Howells44dfd842018-11-05 17:40:31 +00003000 if (!err) {
Al Viro2763d112019-06-30 19:18:53 -04003001 if (attached)
3002 mntput_no_expire(parent);
3003 else
David Howells44dfd842018-11-05 17:40:31 +00003004 free_mnt_ns(ns);
3005 }
David Howells2db154b2018-11-05 17:40:30 +00003006 return err;
3007}
3008
3009static int do_move_mount_old(struct path *path, const char *old_name)
3010{
3011 struct path old_path;
3012 int err;
3013
3014 if (!old_name || !*old_name)
3015 return -EINVAL;
3016
3017 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3018 if (err)
3019 return err;
3020
3021 err = do_move_mount(&old_path, path);
Al Viro2d92ab32008-08-02 00:51:11 -04003022 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 return err;
3024}
3025
Al Viro9d412a42011-03-17 22:08:28 -04003026/*
3027 * add a mount into a namespace's mount tree
3028 */
Al Viro8f115382020-01-11 10:14:09 -05003029static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
Al Viro1e2d84642021-06-18 20:27:57 -04003030 const struct path *path, int mnt_flags)
Al Viro9d412a42011-03-17 22:08:28 -04003031{
Al Viro8f115382020-01-11 10:14:09 -05003032 struct mount *parent = real_mount(path->mnt);
Al Viro9d412a42011-03-17 22:08:28 -04003033
Al Virof2ebb3a2014-02-27 09:35:45 -05003034 mnt_flags &= ~MNT_INTERNAL_FLAGS;
Al Viro9d412a42011-03-17 22:08:28 -04003035
Al Viro84d171922013-03-15 10:53:28 -04003036 if (unlikely(!check_mnt(parent))) {
Al Viro156cacb2012-09-21 08:19:02 -04003037 /* that's acceptable only for automounts done in private ns */
3038 if (!(mnt_flags & MNT_SHRINKABLE))
Al Viro8f115382020-01-11 10:14:09 -05003039 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04003040 /* ... and for those we'd better have mountpoint still alive */
Al Viro84d171922013-03-15 10:53:28 -04003041 if (!parent->mnt_ns)
Al Viro8f115382020-01-11 10:14:09 -05003042 return -EINVAL;
Al Viro156cacb2012-09-21 08:19:02 -04003043 }
Al Viro9d412a42011-03-17 22:08:28 -04003044
3045 /* Refuse the same filesystem on the same mount point */
Al Viro95bc5f22011-11-25 00:30:56 -05003046 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
Al Viro9d412a42011-03-17 22:08:28 -04003047 path->mnt->mnt_root == path->dentry)
Al Viro8f115382020-01-11 10:14:09 -05003048 return -EBUSY;
Al Viro9d412a42011-03-17 22:08:28 -04003049
David Howellse36cb0b2015-01-29 12:02:35 +00003050 if (d_is_symlink(newmnt->mnt.mnt_root))
Al Viro8f115382020-01-11 10:14:09 -05003051 return -EINVAL;
Al Viro9d412a42011-03-17 22:08:28 -04003052
Al Viro95bc5f22011-11-25 00:30:56 -05003053 newmnt->mnt.mnt_flags = mnt_flags;
Al Viro8f115382020-01-11 10:14:09 -05003054 return graft_tree(newmnt, parent, mp);
Al Viro9d412a42011-03-17 22:08:28 -04003055}
Al Virob1e75df2011-01-17 01:47:59 -05003056
David Howells132e4602018-11-04 07:43:08 -05003057static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3058
3059/*
3060 * Create a new mount using a superblock configuration and request it
3061 * be added to the namespace tree.
3062 */
3063static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3064 unsigned int mnt_flags)
3065{
3066 struct vfsmount *mnt;
Al Viro8f115382020-01-11 10:14:09 -05003067 struct mountpoint *mp;
David Howells132e4602018-11-04 07:43:08 -05003068 struct super_block *sb = fc->root->d_sb;
3069 int error;
3070
Al Viroc9ce29e2018-12-20 15:04:50 -05003071 error = security_sb_kern_mount(sb);
3072 if (!error && mount_too_revealing(sb, &mnt_flags))
3073 error = -EPERM;
3074
3075 if (unlikely(error)) {
3076 fc_drop_locked(fc);
3077 return error;
David Howells132e4602018-11-04 07:43:08 -05003078 }
3079
3080 up_write(&sb->s_umount);
3081
3082 mnt = vfs_create_mount(fc);
3083 if (IS_ERR(mnt))
3084 return PTR_ERR(mnt);
3085
Deepa Dinamanif8b92ba2019-04-15 14:17:12 -07003086 mnt_warn_timestamp_expiry(mountpoint, mnt);
3087
Al Viro8f115382020-01-11 10:14:09 -05003088 mp = lock_mount(mountpoint);
3089 if (IS_ERR(mp)) {
3090 mntput(mnt);
3091 return PTR_ERR(mp);
3092 }
3093 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3094 unlock_mount(mp);
Eric Biggers0ecee662019-10-16 19:48:14 -07003095 if (error < 0)
3096 mntput(mnt);
David Howells132e4602018-11-04 07:43:08 -05003097 return error;
3098}
Eric W. Biederman1b852bc2015-05-08 23:22:29 -05003099
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100/*
3101 * create a new mount for userspace and request it to be added into the
3102 * namespace's tree
3103 */
David Howellse462ec52017-07-17 08:45:35 +01003104static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
Al Viro808d4e32012-10-11 11:42:01 -04003105 int mnt_flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003107 struct file_system_type *type;
Al Viroa0c9a8b2018-11-04 07:18:51 -05003108 struct fs_context *fc;
3109 const char *subtype = NULL;
3110 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003112 if (!fstype)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 return -EINVAL;
3114
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003115 type = get_fs_type(fstype);
3116 if (!type)
3117 return -ENODEV;
3118
Al Viroa0c9a8b2018-11-04 07:18:51 -05003119 if (type->fs_flags & FS_HAS_SUBTYPE) {
3120 subtype = strchr(fstype, '.');
3121 if (subtype) {
3122 subtype++;
3123 if (!*subtype) {
3124 put_filesystem(type);
3125 return -EINVAL;
3126 }
Al Viroa0c9a8b2018-11-04 07:18:51 -05003127 }
Eric W. Biederman8654df42016-06-09 16:06:06 -05003128 }
3129
Al Viroa0c9a8b2018-11-04 07:18:51 -05003130 fc = fs_context_for_mount(type, sb_flags);
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07003131 put_filesystem(type);
Al Viroa0c9a8b2018-11-04 07:18:51 -05003132 if (IS_ERR(fc))
3133 return PTR_ERR(fc);
3134
David Howells3e1aeb02018-11-01 23:07:25 +00003135 if (subtype)
3136 err = vfs_parse_fs_string(fc, "subtype",
3137 subtype, strlen(subtype));
3138 if (!err && name)
3139 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
Al Viroa0c9a8b2018-11-04 07:18:51 -05003140 if (!err)
3141 err = parse_monolithic_mount_data(fc, data);
Al Viroc3aabf02019-05-13 12:57:22 -04003142 if (!err && !mount_capable(fc))
3143 err = -EPERM;
Al Viroa0c9a8b2018-11-04 07:18:51 -05003144 if (!err)
3145 err = vfs_get_tree(fc);
David Howells132e4602018-11-04 07:43:08 -05003146 if (!err)
3147 err = do_new_mount_fc(fc, path, mnt_flags);
Al Viroa0c9a8b2018-11-04 07:18:51 -05003148
Al Viroa0c9a8b2018-11-04 07:18:51 -05003149 put_fs_context(fc);
Al Viro15f9a3f2011-01-17 01:41:58 -05003150 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151}
3152
Al Viro1e2d84642021-06-18 20:27:57 -04003153int finish_automount(struct vfsmount *m, const struct path *path)
Al Viro19a167a2011-01-17 01:35:23 -05003154{
Al Viro26df6032020-01-11 10:44:29 -05003155 struct dentry *dentry = path->dentry;
Al Viro8f115382020-01-11 10:14:09 -05003156 struct mountpoint *mp;
Al Viro25e195a2020-01-11 11:27:46 -05003157 struct mount *mnt;
Al Viro19a167a2011-01-17 01:35:23 -05003158 int err;
Al Viro25e195a2020-01-11 11:27:46 -05003159
3160 if (!m)
3161 return 0;
3162 if (IS_ERR(m))
3163 return PTR_ERR(m);
3164
3165 mnt = real_mount(m);
Al Viro19a167a2011-01-17 01:35:23 -05003166 /* The new mount record should have at least 2 refs to prevent it being
3167 * expired before we get a chance to add it
3168 */
Al Viro6776db3d2011-11-25 00:22:05 -05003169 BUG_ON(mnt_get_count(mnt) < 2);
Al Viro19a167a2011-01-17 01:35:23 -05003170
3171 if (m->mnt_sb == path->mnt->mnt_sb &&
Al Viro26df6032020-01-11 10:44:29 -05003172 m->mnt_root == dentry) {
Al Virob1e75df2011-01-17 01:47:59 -05003173 err = -ELOOP;
Al Viro26df6032020-01-11 10:44:29 -05003174 goto discard;
Al Viro19a167a2011-01-17 01:35:23 -05003175 }
3176
Al Viro26df6032020-01-11 10:44:29 -05003177 /*
3178 * we don't want to use lock_mount() - in this case finding something
3179 * that overmounts our mountpoint to be means "quitely drop what we've
3180 * got", not "try to mount it on top".
3181 */
3182 inode_lock(dentry->d_inode);
3183 namespace_lock();
3184 if (unlikely(cant_mount(dentry))) {
3185 err = -ENOENT;
3186 goto discard_locked;
3187 }
3188 rcu_read_lock();
3189 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3190 rcu_read_unlock();
3191 err = 0;
3192 goto discard_locked;
3193 }
3194 rcu_read_unlock();
3195 mp = get_mountpoint(dentry);
Al Viro8f115382020-01-11 10:14:09 -05003196 if (IS_ERR(mp)) {
3197 err = PTR_ERR(mp);
Al Viro26df6032020-01-11 10:44:29 -05003198 goto discard_locked;
Al Viro8f115382020-01-11 10:14:09 -05003199 }
Al Viro26df6032020-01-11 10:44:29 -05003200
Al Viro8f115382020-01-11 10:14:09 -05003201 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3202 unlock_mount(mp);
Al Viro26df6032020-01-11 10:44:29 -05003203 if (unlikely(err))
3204 goto discard;
3205 mntput(m);
3206 return 0;
3207
3208discard_locked:
3209 namespace_unlock();
3210 inode_unlock(dentry->d_inode);
3211discard:
Al Virob1e75df2011-01-17 01:47:59 -05003212 /* remove m from any expiration list it may be on */
Al Viro6776db3d2011-11-25 00:22:05 -05003213 if (!list_empty(&mnt->mnt_expire)) {
Al Viro97216be2013-03-16 15:12:40 -04003214 namespace_lock();
Al Viro6776db3d2011-11-25 00:22:05 -05003215 list_del_init(&mnt->mnt_expire);
Al Viro97216be2013-03-16 15:12:40 -04003216 namespace_unlock();
Al Viro19a167a2011-01-17 01:35:23 -05003217 }
Al Virob1e75df2011-01-17 01:47:59 -05003218 mntput(m);
3219 mntput(m);
Al Viro19a167a2011-01-17 01:35:23 -05003220 return err;
3221}
3222
David Howellsea5b7782011-01-14 19:10:03 +00003223/**
3224 * mnt_set_expiry - Put a mount on an expiration list
3225 * @mnt: The mount to list.
3226 * @expiry_list: The list to add the mount to.
3227 */
3228void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3229{
Al Viro97216be2013-03-16 15:12:40 -04003230 namespace_lock();
David Howellsea5b7782011-01-14 19:10:03 +00003231
Al Viro6776db3d2011-11-25 00:22:05 -05003232 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
David Howellsea5b7782011-01-14 19:10:03 +00003233
Al Viro97216be2013-03-16 15:12:40 -04003234 namespace_unlock();
David Howellsea5b7782011-01-14 19:10:03 +00003235}
3236EXPORT_SYMBOL(mnt_set_expiry);
3237
3238/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 * process a list of expirable mountpoints with the intent of discarding any
3240 * mountpoints that aren't in use and haven't been touched since last we came
3241 * here
3242 */
3243void mark_mounts_for_expiry(struct list_head *mounts)
3244{
Al Viro761d5c32011-11-24 21:07:43 -05003245 struct mount *mnt, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 LIST_HEAD(graveyard);
3247
3248 if (list_empty(mounts))
3249 return;
3250
Al Viro97216be2013-03-16 15:12:40 -04003251 namespace_lock();
Al Viro719ea2f2013-09-29 11:24:49 -04003252 lock_mount_hash();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253
3254 /* extract from the expiration list every vfsmount that matches the
3255 * following criteria:
3256 * - only referenced by its parent vfsmount
3257 * - still marked for expiry (marked on the last call here; marks are
3258 * cleared by mntput())
3259 */
Al Viro6776db3d2011-11-25 00:22:05 -05003260 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Al Viro863d6842011-11-25 00:57:42 -05003261 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Viro1ab59732011-11-24 21:35:16 -05003262 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 continue;
Al Viro6776db3d2011-11-25 00:22:05 -05003264 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 }
Al Virobcc5c7d2008-03-22 00:21:53 -04003266 while (!list_empty(&graveyard)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003267 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003268 touch_mnt_namespace(mnt->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003269 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003270 }
Al Viro719ea2f2013-09-29 11:24:49 -04003271 unlock_mount_hash();
Al Viro3ab6abe2013-03-16 14:42:19 -04003272 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273}
3274
3275EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3276
3277/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04003278 * Ripoff of 'select_parent()'
3279 *
3280 * search the list of submounts for a given mountpoint, and move any
3281 * shrinkable submounts to the 'graveyard' list.
3282 */
Al Viro692afc32011-11-24 21:15:14 -05003283static int select_submounts(struct mount *parent, struct list_head *graveyard)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003284{
Al Viro692afc32011-11-24 21:15:14 -05003285 struct mount *this_parent = parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003286 struct list_head *next;
3287 int found = 0;
3288
3289repeat:
Al Viro6b41d532011-11-24 23:24:33 -05003290 next = this_parent->mnt_mounts.next;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003291resume:
Al Viro6b41d532011-11-24 23:24:33 -05003292 while (next != &this_parent->mnt_mounts) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003293 struct list_head *tmp = next;
Al Viro6b41d532011-11-24 23:24:33 -05003294 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003295
3296 next = tmp->next;
Al Viro692afc32011-11-24 21:15:14 -05003297 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
Trond Myklebust5528f9112006-06-09 09:34:17 -04003298 continue;
3299 /*
3300 * Descend a level if the d_mounts list is non-empty.
3301 */
Al Viro6b41d532011-11-24 23:24:33 -05003302 if (!list_empty(&mnt->mnt_mounts)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04003303 this_parent = mnt;
3304 goto repeat;
3305 }
3306
Al Viro1ab59732011-11-24 21:35:16 -05003307 if (!propagate_mount_busy(mnt, 1)) {
Al Viro6776db3d2011-11-25 00:22:05 -05003308 list_move_tail(&mnt->mnt_expire, graveyard);
Trond Myklebust5528f9112006-06-09 09:34:17 -04003309 found++;
3310 }
3311 }
3312 /*
3313 * All done at this level ... ascend and resume the search
3314 */
3315 if (this_parent != parent) {
Al Viro6b41d532011-11-24 23:24:33 -05003316 next = this_parent->mnt_child.next;
Al Viro0714a532011-11-24 22:19:58 -05003317 this_parent = this_parent->mnt_parent;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003318 goto resume;
3319 }
3320 return found;
3321}
3322
3323/*
3324 * process a list of expirable mountpoints with the intent of discarding any
3325 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10003326 *
Al Viro48a066e2013-09-29 22:06:07 -04003327 * mount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04003328 */
Al Virob54b9be2013-03-16 14:39:34 -04003329static void shrink_submounts(struct mount *mnt)
Trond Myklebust5528f9112006-06-09 09:34:17 -04003330{
3331 LIST_HEAD(graveyard);
Al Viro761d5c32011-11-24 21:07:43 -05003332 struct mount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04003333
Trond Myklebust5528f9112006-06-09 09:34:17 -04003334 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04003335 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04003336 while (!list_empty(&graveyard)) {
Al Viro761d5c32011-11-24 21:07:43 -05003337 m = list_first_entry(&graveyard, struct mount,
Al Viro6776db3d2011-11-25 00:22:05 -05003338 mnt_expire);
Al Viro143c8c92011-11-25 00:46:35 -05003339 touch_mnt_namespace(m->mnt_ns);
Eric W. Biedermane819f152014-12-24 07:20:01 -06003340 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
Al Virobcc5c7d2008-03-22 00:21:53 -04003341 }
3342 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04003343}
3344
Christoph Hellwig028abd92020-09-17 10:22:34 +02003345static void *copy_mount_options(const void __user * data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346{
Al Virob40ef862015-12-14 18:44:44 -05003347 char *copy;
Catalin Marinasd563d672020-07-01 17:46:06 +01003348 unsigned left, offset;
Ram Paib58fed82005-11-07 17:16:09 -05003349
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 if (!data)
Al Virob40ef862015-12-14 18:44:44 -05003351 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352
Al Virob40ef862015-12-14 18:44:44 -05003353 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3354 if (!copy)
3355 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
Catalin Marinasd563d672020-07-01 17:46:06 +01003357 left = copy_from_user(copy, data, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358
Catalin Marinasd563d672020-07-01 17:46:06 +01003359 /*
3360 * Not all architectures have an exact copy_from_user(). Resort to
3361 * byte at a time.
3362 */
3363 offset = PAGE_SIZE - left;
3364 while (left) {
3365 char c;
3366 if (get_user(c, (const char __user *)data + offset))
3367 break;
3368 copy[offset] = c;
3369 left--;
3370 offset++;
3371 }
3372
3373 if (left == PAGE_SIZE) {
Al Virob40ef862015-12-14 18:44:44 -05003374 kfree(copy);
3375 return ERR_PTR(-EFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 }
Catalin Marinasd563d672020-07-01 17:46:06 +01003377
Al Virob40ef862015-12-14 18:44:44 -05003378 return copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379}
3380
Christoph Hellwig028abd92020-09-17 10:22:34 +02003381static char *copy_mount_string(const void __user *data)
Vegard Nossumeca6f532009-09-18 13:05:45 -07003382{
Chandan Rajendrafbdb4402019-01-22 12:21:52 +05303383 return data ? strndup_user(data, PATH_MAX) : NULL;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003384}
3385
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386/*
3387 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3388 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3389 *
3390 * data is a (void *) that can point to any structure up to
3391 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3392 * information (or be NULL).
3393 *
3394 * Pre-0.97 versions of mount() didn't have a flags word.
3395 * When the flags word was introduced its top half was required
3396 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3397 * Therefore, if this magic number is present, it carries no information
3398 * and must be discarded.
3399 */
Christoph Hellwigc60166f2020-07-21 11:12:08 +02003400int path_mount(const char *dev_name, struct path *path,
Al Viro808d4e32012-10-11 11:42:01 -04003401 const char *type_page, unsigned long flags, void *data_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402{
David Howellse462ec52017-07-17 08:45:35 +01003403 unsigned int mnt_flags = 0, sb_flags;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003404 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405
3406 /* Discard magic */
3407 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3408 flags &= ~MS_MGC_MSK;
3409
3410 /* Basic sanity checks */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 if (data_page)
3412 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3413
David Howellse462ec52017-07-17 08:45:35 +01003414 if (flags & MS_NOUSER)
3415 return -EINVAL;
3416
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003417 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3418 if (ret)
3419 return ret;
3420 if (!may_mount())
3421 return -EPERM;
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04003422 if (flags & SB_MANDLOCK)
3423 warn_mandlock();
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09003424
Andi Kleen613cbe32009-04-19 18:40:43 +02003425 /* Default to relatime unless overriden */
3426 if (!(flags & MS_NOATIME))
3427 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00003428
Linus Torvalds1da177e2005-04-16 15:20:36 -07003429 /* Separate the per-mountpoint flags */
3430 if (flags & MS_NOSUID)
3431 mnt_flags |= MNT_NOSUID;
3432 if (flags & MS_NODEV)
3433 mnt_flags |= MNT_NODEV;
3434 if (flags & MS_NOEXEC)
3435 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003436 if (flags & MS_NOATIME)
3437 mnt_flags |= MNT_NOATIME;
3438 if (flags & MS_NODIRATIME)
3439 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00003440 if (flags & MS_STRICTATIME)
3441 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
David Howellsa9e5b732018-04-20 13:35:02 +01003442 if (flags & MS_RDONLY)
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08003443 mnt_flags |= MNT_READONLY;
Mattias Nisslerdab741e02020-08-27 11:09:46 -06003444 if (flags & MS_NOSYMFOLLOW)
3445 mnt_flags |= MNT_NOSYMFOLLOW;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08003446
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003447 /* The default atime for remount is preservation */
3448 if ((flags & MS_REMOUNT) &&
3449 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3450 MS_STRICTATIME)) == 0)) {
3451 mnt_flags &= ~MNT_ATIME_MASK;
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003452 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
Eric W. Biedermanffbc6f02014-07-28 17:36:04 -07003453 }
3454
David Howellse462ec52017-07-17 08:45:35 +01003455 sb_flags = flags & (SB_RDONLY |
3456 SB_SYNCHRONOUS |
3457 SB_MANDLOCK |
3458 SB_DIRSYNC |
3459 SB_SILENT |
Mimi Zohar917086f2017-10-08 00:28:21 -04003460 SB_POSIXACL |
Markus Trippelsdorfd7ee9462017-10-11 07:01:31 +02003461 SB_LAZYTIME |
Mimi Zohar917086f2017-10-08 00:28:21 -04003462 SB_I_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463
David Howells43f5e652018-11-01 23:07:25 +00003464 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003465 return do_reconfigure_mnt(path, mnt_flags);
3466 if (flags & MS_REMOUNT)
3467 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3468 if (flags & MS_BIND)
3469 return do_loopback(path, dev_name, flags & MS_REC);
3470 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3471 return do_change_type(path, flags);
3472 if (flags & MS_MOVE)
3473 return do_move_mount_old(path, dev_name);
3474
3475 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3476 data_page);
3477}
3478
3479long do_mount(const char *dev_name, const char __user *dir_name,
3480 const char *type_page, unsigned long flags, void *data_page)
3481{
3482 struct path path;
3483 int ret;
3484
3485 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3486 if (ret)
3487 return ret;
3488 ret = path_mount(dev_name, &path, type_page, flags, data_page);
Al Viro2d92ab32008-08-02 00:51:11 -04003489 path_put(&path);
Christoph Hellwiga1e6aaa2020-06-04 09:41:08 +02003490 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491}
3492
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003493static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3494{
3495 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3496}
3497
3498static void dec_mnt_namespaces(struct ucounts *ucounts)
3499{
3500 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3501}
3502
Eric W. Biederman771b1372012-07-26 21:08:32 -07003503static void free_mnt_ns(struct mnt_namespace *ns)
3504{
Al Viro74e83122019-01-30 13:30:21 -05003505 if (!is_anon_ns(ns))
3506 ns_free_inum(&ns->ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003507 dec_mnt_namespaces(ns->ucounts);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003508 put_user_ns(ns->user_ns);
3509 kfree(ns);
3510}
3511
Eric W. Biederman8823c072010-03-07 18:49:36 -08003512/*
3513 * Assign a sequence number so we can detect when we attempt to bind
3514 * mount a reference to an older mount namespace into the current
3515 * mount namespace, preventing reference counting loops. A 64bit
3516 * number incrementing at 10Ghz will take 12,427 years to wrap which
3517 * is effectively never, so we can ignore the possibility.
3518 */
3519static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3520
Al Viro74e83122019-01-30 13:30:21 -05003521static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003522{
3523 struct mnt_namespace *new_ns;
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003524 struct ucounts *ucounts;
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003525 int ret;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003526
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003527 ucounts = inc_mnt_namespaces(user_ns);
3528 if (!ucounts)
Eric W. Biedermandf75e772016-09-22 13:08:36 -05003529 return ERR_PTR(-ENOSPC);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003530
Vasily Averin30acd0b2021-09-02 14:55:27 -07003531 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003532 if (!new_ns) {
3533 dec_mnt_namespaces(ucounts);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003534 return ERR_PTR(-ENOMEM);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003535 }
Al Viro74e83122019-01-30 13:30:21 -05003536 if (!anon) {
3537 ret = ns_alloc_inum(&new_ns->ns);
3538 if (ret) {
3539 kfree(new_ns);
3540 dec_mnt_namespaces(ucounts);
3541 return ERR_PTR(ret);
3542 }
Eric W. Biederman98f842e2011-06-15 10:21:48 -07003543 }
Al Viro33c42942014-11-01 02:32:53 -04003544 new_ns->ns.ops = &mntns_operations;
Al Viro74e83122019-01-30 13:30:21 -05003545 if (!anon)
3546 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03003547 refcount_set(&new_ns->ns.count, 1);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003548 INIT_LIST_HEAD(&new_ns->list);
3549 init_waitqueue_head(&new_ns->poll);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02003550 spin_lock_init(&new_ns->ns_lock);
Eric W. Biederman771b1372012-07-26 21:08:32 -07003551 new_ns->user_ns = get_user_ns(user_ns);
Eric W. Biederman537f7cc2016-08-08 14:37:37 -05003552 new_ns->ucounts = ucounts;
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003553 return new_ns;
3554}
3555
Emese Revfy0766f782016-06-20 20:42:34 +02003556__latent_entropy
Al Viro9559f682013-09-28 20:47:57 -04003557struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3558 struct user_namespace *user_ns, struct fs_struct *new_fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08003560 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04003561 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Al Viro315fc832011-11-24 18:57:30 -05003562 struct mount *p, *q;
Al Viro9559f682013-09-28 20:47:57 -04003563 struct mount *old;
Al Virocb338d02011-11-24 20:55:08 -05003564 struct mount *new;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003565 int copy_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566
Al Viro9559f682013-09-28 20:47:57 -04003567 BUG_ON(!ns);
3568
3569 if (likely(!(flags & CLONE_NEWNS))) {
3570 get_mnt_ns(ns);
3571 return ns;
3572 }
3573
3574 old = ns->root;
3575
Al Viro74e83122019-01-30 13:30:21 -05003576 new_ns = alloc_mnt_ns(user_ns, false);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003577 if (IS_ERR(new_ns))
3578 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579
Al Viro97216be2013-03-16 15:12:40 -04003580 namespace_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 /* First pass: copy the tree topology */
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003582 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
Al Viro9559f682013-09-28 20:47:57 -04003583 if (user_ns != ns->user_ns)
Al Viro3bd045c2019-01-30 13:15:45 -05003584 copy_flags |= CL_SHARED_TO_SLAVE;
Eric W. Biederman7a472ef2012-07-31 13:13:04 -07003585 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
David Howellsbe34d1a2012-06-25 12:55:18 +01003586 if (IS_ERR(new)) {
Al Viro328e6d92013-03-16 14:49:45 -04003587 namespace_unlock();
Eric W. Biederman771b1372012-07-26 21:08:32 -07003588 free_mnt_ns(new_ns);
David Howellsbe34d1a2012-06-25 12:55:18 +01003589 return ERR_CAST(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 }
Al Viro3bd045c2019-01-30 13:15:45 -05003591 if (user_ns != ns->user_ns) {
3592 lock_mount_hash();
3593 lock_mnt_tree(new);
3594 unlock_mount_hash();
3595 }
Al Virobe08d6d2011-12-06 13:32:36 -05003596 new_ns->root = new;
Al Viro1a4eeaf2011-11-25 02:19:55 -05003597 list_add_tail(&new_ns->list, &new->mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598
3599 /*
3600 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3601 * as belonging to new namespace. We have already acquired a private
3602 * fs_struct, so tsk->fs->lock is not needed.
3603 */
Al Viro909b0a82011-11-25 03:06:56 -05003604 p = old;
Al Virocb338d02011-11-24 20:55:08 -05003605 q = new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 while (p) {
Al Viro143c8c92011-11-25 00:46:35 -05003607 q->mnt_ns = new_ns;
Eric W. Biedermand2921682016-09-28 00:27:17 -05003608 new_ns->mounts++;
Al Viro9559f682013-09-28 20:47:57 -04003609 if (new_fs) {
3610 if (&p->mnt == new_fs->root.mnt) {
3611 new_fs->root.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003612 rootmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613 }
Al Viro9559f682013-09-28 20:47:57 -04003614 if (&p->mnt == new_fs->pwd.mnt) {
3615 new_fs->pwd.mnt = mntget(&q->mnt);
Al Viro315fc832011-11-24 18:57:30 -05003616 pwdmnt = &p->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 }
Al Viro909b0a82011-11-25 03:06:56 -05003619 p = next_mnt(p, old);
3620 q = next_mnt(q, new);
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003621 if (!q)
3622 break;
Al Viro61d8e422022-11-24 22:55:57 -05003623 // an mntns binding we'd skipped?
Eric W. Biederman4ce5d2b2013-03-30 01:35:18 -07003624 while (p->mnt.mnt_root != q->mnt.mnt_root)
Al Viro61d8e422022-11-24 22:55:57 -05003625 p = next_mnt(skip_mnt_tree(p), old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 }
Al Viro328e6d92013-03-16 14:49:45 -04003627 namespace_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 if (rootmnt)
Al Virof03c6592011-01-14 22:30:21 -05003630 mntput(rootmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 if (pwdmnt)
Al Virof03c6592011-01-14 22:30:21 -05003632 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633
JANAK DESAI741a2952006-02-07 12:59:00 -08003634 return new_ns;
3635}
3636
Al Viro74e83122019-01-30 13:30:21 -05003637struct dentry *mount_subtree(struct vfsmount *m, const char *name)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04003638{
Al Viro74e83122019-01-30 13:30:21 -05003639 struct mount *mnt = real_mount(m);
Al Viroea441d12011-11-16 21:43:59 -05003640 struct mnt_namespace *ns;
Al Virod31da0f2011-11-22 12:31:21 -05003641 struct super_block *s;
Al Viroea441d12011-11-16 21:43:59 -05003642 struct path path;
3643 int err;
3644
Al Viro74e83122019-01-30 13:30:21 -05003645 ns = alloc_mnt_ns(&init_user_ns, true);
3646 if (IS_ERR(ns)) {
3647 mntput(m);
Al Viroea441d12011-11-16 21:43:59 -05003648 return ERR_CAST(ns);
Al Viro74e83122019-01-30 13:30:21 -05003649 }
3650 mnt->mnt_ns = ns;
3651 ns->root = mnt;
3652 ns->mounts++;
3653 list_add(&mnt->mnt_list, &ns->list);
Al Viroea441d12011-11-16 21:43:59 -05003654
Al Viro74e83122019-01-30 13:30:21 -05003655 err = vfs_path_lookup(m->mnt_root, m,
Al Viroea441d12011-11-16 21:43:59 -05003656 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3657
3658 put_mnt_ns(ns);
3659
3660 if (err)
3661 return ERR_PTR(err);
3662
3663 /* trade a vfsmount reference for active sb one */
Al Virod31da0f2011-11-22 12:31:21 -05003664 s = path.mnt->mnt_sb;
3665 atomic_inc(&s->s_active);
Al Viroea441d12011-11-16 21:43:59 -05003666 mntput(path.mnt);
3667 /* lock the sucker */
Al Virod31da0f2011-11-22 12:31:21 -05003668 down_write(&s->s_umount);
Al Viroea441d12011-11-16 21:43:59 -05003669 /* ... and return the root of (sub)tree on it */
3670 return path.dentry;
3671}
3672EXPORT_SYMBOL(mount_subtree);
3673
Dominik Brodowskicccaa5e2018-10-23 22:41:09 +02003674SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3675 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676{
Vegard Nossumeca6f532009-09-18 13:05:45 -07003677 int ret;
3678 char *kernel_type;
Vegard Nossumeca6f532009-09-18 13:05:45 -07003679 char *kernel_dev;
Al Virob40ef862015-12-14 18:44:44 -05003680 void *options;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681
Tim Gardnerb8850d12014-08-28 11:26:03 -06003682 kernel_type = copy_mount_string(type);
3683 ret = PTR_ERR(kernel_type);
3684 if (IS_ERR(kernel_type))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003685 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686
Tim Gardnerb8850d12014-08-28 11:26:03 -06003687 kernel_dev = copy_mount_string(dev_name);
3688 ret = PTR_ERR(kernel_dev);
3689 if (IS_ERR(kernel_dev))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003690 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
Al Virob40ef862015-12-14 18:44:44 -05003692 options = copy_mount_options(data);
3693 ret = PTR_ERR(options);
3694 if (IS_ERR(options))
Vegard Nossumeca6f532009-09-18 13:05:45 -07003695 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
Al Virob40ef862015-12-14 18:44:44 -05003697 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003698
Al Virob40ef862015-12-14 18:44:44 -05003699 kfree(options);
Vegard Nossumeca6f532009-09-18 13:05:45 -07003700out_data:
3701 kfree(kernel_dev);
3702out_dev:
Vegard Nossumeca6f532009-09-18 13:05:45 -07003703 kfree(kernel_type);
3704out_type:
3705 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706}
3707
Christian Braunerdd8b4772021-06-01 11:33:59 +02003708#define FSMOUNT_VALID_FLAGS \
3709 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3710 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3711 MOUNT_ATTR_NOSYMFOLLOW)
Christian Brauner5b490502021-01-21 14:19:52 +01003712
Christian Brauner9caccd42021-01-21 14:19:54 +01003713#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
Christian Brauner2a186722021-01-21 14:19:53 +01003714
3715#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3716 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3717
Christian Brauner5b490502021-01-21 14:19:52 +01003718static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3719{
3720 unsigned int mnt_flags = 0;
3721
3722 if (attr_flags & MOUNT_ATTR_RDONLY)
3723 mnt_flags |= MNT_READONLY;
3724 if (attr_flags & MOUNT_ATTR_NOSUID)
3725 mnt_flags |= MNT_NOSUID;
3726 if (attr_flags & MOUNT_ATTR_NODEV)
3727 mnt_flags |= MNT_NODEV;
3728 if (attr_flags & MOUNT_ATTR_NOEXEC)
3729 mnt_flags |= MNT_NOEXEC;
3730 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3731 mnt_flags |= MNT_NODIRATIME;
Christian Braunerdd8b4772021-06-01 11:33:59 +02003732 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3733 mnt_flags |= MNT_NOSYMFOLLOW;
Christian Brauner5b490502021-01-21 14:19:52 +01003734
3735 return mnt_flags;
3736}
3737
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738/*
David Howells93766fb2018-11-01 23:36:14 +00003739 * Create a kernel mount representation for a new, prepared superblock
3740 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3741 */
3742SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3743 unsigned int, attr_flags)
3744{
3745 struct mnt_namespace *ns;
3746 struct fs_context *fc;
3747 struct file *file;
3748 struct path newmount;
3749 struct mount *mnt;
3750 struct fd f;
3751 unsigned int mnt_flags = 0;
3752 long ret;
3753
3754 if (!may_mount())
3755 return -EPERM;
3756
3757 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3758 return -EINVAL;
3759
Christian Brauner5b490502021-01-21 14:19:52 +01003760 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
David Howells93766fb2018-11-01 23:36:14 +00003761 return -EINVAL;
3762
Christian Brauner5b490502021-01-21 14:19:52 +01003763 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
David Howells93766fb2018-11-01 23:36:14 +00003764
3765 switch (attr_flags & MOUNT_ATTR__ATIME) {
3766 case MOUNT_ATTR_STRICTATIME:
3767 break;
3768 case MOUNT_ATTR_NOATIME:
3769 mnt_flags |= MNT_NOATIME;
3770 break;
3771 case MOUNT_ATTR_RELATIME:
3772 mnt_flags |= MNT_RELATIME;
3773 break;
3774 default:
3775 return -EINVAL;
3776 }
3777
3778 f = fdget(fs_fd);
3779 if (!f.file)
3780 return -EBADF;
3781
3782 ret = -EINVAL;
3783 if (f.file->f_op != &fscontext_fops)
3784 goto err_fsfd;
3785
3786 fc = f.file->private_data;
3787
3788 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3789 if (ret < 0)
3790 goto err_fsfd;
3791
3792 /* There must be a valid superblock or we can't mount it */
3793 ret = -EINVAL;
3794 if (!fc->root)
3795 goto err_unlock;
3796
3797 ret = -EPERM;
3798 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3799 pr_warn("VFS: Mount too revealing\n");
3800 goto err_unlock;
3801 }
3802
3803 ret = -EBUSY;
3804 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3805 goto err_unlock;
3806
Jeff Laytonf7e33bd2021-08-19 14:56:38 -04003807 if (fc->sb_flags & SB_MANDLOCK)
3808 warn_mandlock();
David Howells93766fb2018-11-01 23:36:14 +00003809
3810 newmount.mnt = vfs_create_mount(fc);
3811 if (IS_ERR(newmount.mnt)) {
3812 ret = PTR_ERR(newmount.mnt);
3813 goto err_unlock;
3814 }
3815 newmount.dentry = dget(fc->root);
3816 newmount.mnt->mnt_flags = mnt_flags;
3817
3818 /* We've done the mount bit - now move the file context into more or
3819 * less the same state as if we'd done an fspick(). We don't want to
3820 * do any memory allocation or anything like that at this point as we
3821 * don't want to have to handle any errors incurred.
3822 */
3823 vfs_clean_context(fc);
3824
3825 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3826 if (IS_ERR(ns)) {
3827 ret = PTR_ERR(ns);
3828 goto err_path;
3829 }
3830 mnt = real_mount(newmount.mnt);
3831 mnt->mnt_ns = ns;
3832 ns->root = mnt;
3833 ns->mounts = 1;
3834 list_add(&mnt->mnt_list, &ns->list);
Eric Biggers1b0b9cc2019-06-12 11:43:13 -07003835 mntget(newmount.mnt);
David Howells93766fb2018-11-01 23:36:14 +00003836
3837 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3838 * it, not just simply put it.
3839 */
3840 file = dentry_open(&newmount, O_PATH, fc->cred);
3841 if (IS_ERR(file)) {
3842 dissolve_on_fput(newmount.mnt);
3843 ret = PTR_ERR(file);
3844 goto err_path;
3845 }
3846 file->f_mode |= FMODE_NEED_UNMOUNT;
3847
3848 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3849 if (ret >= 0)
3850 fd_install(ret, file);
3851 else
3852 fput(file);
3853
3854err_path:
3855 path_put(&newmount);
3856err_unlock:
3857 mutex_unlock(&fc->uapi_mutex);
3858err_fsfd:
3859 fdput(f);
3860 return ret;
3861}
3862
3863/*
3864 * Move a mount from one place to another. In combination with
3865 * fsopen()/fsmount() this is used to install a new mount and in combination
3866 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3867 * a mount subtree.
David Howells2db154b2018-11-05 17:40:30 +00003868 *
3869 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3870 */
3871SYSCALL_DEFINE5(move_mount,
Ben Dooks2658ce02019-10-15 11:35:02 +01003872 int, from_dfd, const char __user *, from_pathname,
3873 int, to_dfd, const char __user *, to_pathname,
David Howells2db154b2018-11-05 17:40:30 +00003874 unsigned int, flags)
3875{
3876 struct path from_path, to_path;
3877 unsigned int lflags;
3878 int ret = 0;
3879
3880 if (!may_mount())
3881 return -EPERM;
3882
3883 if (flags & ~MOVE_MOUNT__MASK)
3884 return -EINVAL;
3885
3886 /* If someone gives a pathname, they aren't permitted to move
3887 * from an fd that requires unmount as we can't get at the flag
3888 * to clear it afterwards.
3889 */
3890 lflags = 0;
3891 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3892 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3893 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3894
3895 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3896 if (ret < 0)
3897 return ret;
3898
3899 lflags = 0;
3900 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3901 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3902 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3903
3904 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3905 if (ret < 0)
3906 goto out_from;
3907
3908 ret = security_move_mount(&from_path, &to_path);
3909 if (ret < 0)
3910 goto out_to;
3911
Pavel Tikhomirov9ffb14e2021-07-15 13:07:13 +03003912 if (flags & MOVE_MOUNT_SET_GROUP)
3913 ret = do_set_group(&from_path, &to_path);
3914 else
3915 ret = do_move_mount(&from_path, &to_path);
David Howells2db154b2018-11-05 17:40:30 +00003916
3917out_to:
3918 path_put(&to_path);
3919out_from:
3920 path_put(&from_path);
3921 return ret;
3922}
3923
3924/*
Al Viroafac7cb2011-11-23 19:34:49 -05003925 * Return true if path is reachable from root
3926 *
Al Viro48a066e2013-09-29 22:06:07 -04003927 * namespace_sem or mount_lock is held
Al Viroafac7cb2011-11-23 19:34:49 -05003928 */
Al Viro643822b2011-11-24 22:00:28 -05003929bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
Al Viroafac7cb2011-11-23 19:34:49 -05003930 const struct path *root)
3931{
Al Viro643822b2011-11-24 22:00:28 -05003932 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -05003933 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05003934 mnt = mnt->mnt_parent;
Al Viroafac7cb2011-11-23 19:34:49 -05003935 }
Al Viro643822b2011-11-24 22:00:28 -05003936 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
Al Viroafac7cb2011-11-23 19:34:49 -05003937}
3938
Mickaël Salaün640eb7e2016-11-14 22:14:35 +01003939bool path_is_under(const struct path *path1, const struct path *path2)
Al Viroafac7cb2011-11-23 19:34:49 -05003940{
Yaowei Bai25ab4c92015-11-17 14:40:10 +08003941 bool res;
Al Viro48a066e2013-09-29 22:06:07 -04003942 read_seqlock_excl(&mount_lock);
Al Viro643822b2011-11-24 22:00:28 -05003943 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
Al Viro48a066e2013-09-29 22:06:07 -04003944 read_sequnlock_excl(&mount_lock);
Al Viroafac7cb2011-11-23 19:34:49 -05003945 return res;
3946}
3947EXPORT_SYMBOL(path_is_under);
3948
3949/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 * pivot_root Semantics:
3951 * Moves the root file system of the current process to the directory put_old,
3952 * makes new_root as the new root file system of the current process, and sets
3953 * root/cwd of all processes which had them on the current root to new_root.
3954 *
3955 * Restrictions:
3956 * The new_root and put_old must be directories, and must not be on the
3957 * same file system as the current process root. The put_old must be
3958 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3959 * pointed to by put_old must yield the same directory as new_root. No other
3960 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3961 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08003962 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
Mauro Carvalho Chehab0c1bc6b2020-04-14 18:48:37 +02003963 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
Neil Brown4a0d11f2006-01-08 01:03:18 -08003964 * in this situation.
3965 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 * Notes:
3967 * - we don't move root/cwd if they are not at the root (reason: if something
3968 * cared enough to change them, it's probably wrong to force them elsewhere)
3969 * - it's okay to pick a root that isn't the root of a file system, e.g.
3970 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3971 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3972 * first.
3973 */
Heiko Carstens3480b252009-01-14 14:14:16 +01003974SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3975 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976{
Al Viro2763d112019-06-30 19:18:53 -04003977 struct path new, old, root;
3978 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
Al Viro84d171922013-03-15 10:53:28 -04003979 struct mountpoint *old_mp, *root_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 int error;
3981
Al Viro9b40bc92013-02-22 22:45:42 -05003982 if (!may_mount())
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 return -EPERM;
3984
Al Viroce6595a2019-07-14 16:42:44 -04003985 error = user_path_at(AT_FDCWD, new_root,
3986 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 if (error)
3988 goto out0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989
Al Viroce6595a2019-07-14 16:42:44 -04003990 error = user_path_at(AT_FDCWD, put_old,
3991 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 if (error)
3993 goto out1;
3994
Al Viro2d8f3032008-07-22 09:59:21 -04003995 error = security_sb_pivotroot(&old, &new);
Al Virob12cea92011-03-18 08:55:38 -04003996 if (error)
3997 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02003999 get_fs_root(current->fs, &root);
Al Viro84d171922013-03-15 10:53:28 -04004000 old_mp = lock_mount(&old);
4001 error = PTR_ERR(old_mp);
4002 if (IS_ERR(old_mp))
Al Virob12cea92011-03-18 08:55:38 -04004003 goto out3;
4004
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 error = -EINVAL;
Al Viro419148d2011-11-24 19:41:16 -05004006 new_mnt = real_mount(new.mnt);
4007 root_mnt = real_mount(root.mnt);
Al Viro84d171922013-03-15 10:53:28 -04004008 old_mnt = real_mount(old.mnt);
Al Viro2763d112019-06-30 19:18:53 -04004009 ex_parent = new_mnt->mnt_parent;
4010 root_parent = root_mnt->mnt_parent;
Al Viro84d171922013-03-15 10:53:28 -04004011 if (IS_MNT_SHARED(old_mnt) ||
Al Viro2763d112019-06-30 19:18:53 -04004012 IS_MNT_SHARED(ex_parent) ||
4013 IS_MNT_SHARED(root_parent))
Al Virob12cea92011-03-18 08:55:38 -04004014 goto out4;
Al Viro143c8c92011-11-25 00:46:35 -05004015 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004016 goto out4;
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07004017 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4018 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04004020 if (d_unlinked(new.dentry))
Al Virob12cea92011-03-18 08:55:38 -04004021 goto out4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022 error = -EBUSY;
Al Viro84d171922013-03-15 10:53:28 -04004023 if (new_mnt == root_mnt || old_mnt == root_mnt)
Al Virob12cea92011-03-18 08:55:38 -04004024 goto out4; /* loop, on the same file system */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004025 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04004026 if (root.mnt->mnt_root != root.dentry)
Al Virob12cea92011-03-18 08:55:38 -04004027 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05004028 if (!mnt_has_parent(root_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004029 goto out4; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04004030 if (new.mnt->mnt_root != new.dentry)
Al Virob12cea92011-03-18 08:55:38 -04004031 goto out4; /* not a mountpoint */
Al Viro676da582011-11-24 21:47:05 -05004032 if (!mnt_has_parent(new_mnt))
Al Virob12cea92011-03-18 08:55:38 -04004033 goto out4; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08004034 /* make sure we can reach put_old from new_root */
Al Viro84d171922013-03-15 10:53:28 -04004035 if (!is_path_reachable(old_mnt, old.dentry, &new))
Al Virob12cea92011-03-18 08:55:38 -04004036 goto out4;
Eric W. Biederman0d082602014-10-08 10:42:27 -07004037 /* make certain new is below the root */
4038 if (!is_path_reachable(new_mnt, new.dentry, &root))
4039 goto out4;
Al Viro719ea2f2013-09-29 11:24:49 -04004040 lock_mount_hash();
Al Viro2763d112019-06-30 19:18:53 -04004041 umount_mnt(new_mnt);
4042 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
Eric W. Biederman5ff9d8a2013-03-29 21:04:39 -07004043 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4044 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4045 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4046 }
Jan Blunck4ac91372008-02-14 19:34:32 -08004047 /* mount old root on put_old */
Al Viro84d171922013-03-15 10:53:28 -04004048 attach_mnt(root_mnt, old_mnt, old_mp);
Jan Blunck4ac91372008-02-14 19:34:32 -08004049 /* mount new_root on / */
Al Viro2763d112019-06-30 19:18:53 -04004050 attach_mnt(new_mnt, root_parent, root_mp);
4051 mnt_add_count(root_parent, -1);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08004052 touch_mnt_namespace(current->nsproxy->mnt_ns);
Eric W. Biederman4fed6552014-10-08 10:42:57 -07004053 /* A moved mount should not expire automatically */
4054 list_del_init(&new_mnt->mnt_expire);
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13004055 put_mountpoint(root_mp);
Al Viro719ea2f2013-09-29 11:24:49 -04004056 unlock_mount_hash();
Al Viro2d8f3032008-07-22 09:59:21 -04004057 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058 error = 0;
Al Virob12cea92011-03-18 08:55:38 -04004059out4:
Al Viro84d171922013-03-15 10:53:28 -04004060 unlock_mount(old_mp);
Al Viro2763d112019-06-30 19:18:53 -04004061 if (!error)
4062 mntput_no_expire(ex_parent);
Al Virob12cea92011-03-18 08:55:38 -04004063out3:
Al Viro8c3ee422008-03-22 18:00:39 -04004064 path_put(&root);
Al Virob12cea92011-03-18 08:55:38 -04004065out2:
Al Viro2d8f3032008-07-22 09:59:21 -04004066 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067out1:
Al Viro2d8f3032008-07-22 09:59:21 -04004068 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071}
4072
Christian Brauner2a186722021-01-21 14:19:53 +01004073static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4074{
4075 unsigned int flags = mnt->mnt.mnt_flags;
4076
4077 /* flags to clear */
4078 flags &= ~kattr->attr_clr;
4079 /* flags to raise */
4080 flags |= kattr->attr_set;
4081
4082 return flags;
4083}
4084
Christian Brauner9caccd42021-01-21 14:19:54 +01004085static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4086{
4087 struct vfsmount *m = &mnt->mnt;
Christian Braunerbd303362021-12-03 12:17:07 +01004088 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
Christian Brauner9caccd42021-01-21 14:19:54 +01004089
Christian Brauner256c8ae2022-10-26 12:51:27 +02004090 if (!kattr->mnt_idmap)
Christian Brauner9caccd42021-01-21 14:19:54 +01004091 return 0;
4092
4093 /*
Christian Braunerbd303362021-12-03 12:17:07 +01004094 * Creating an idmapped mount with the filesystem wide idmapping
4095 * doesn't make sense so block that. We don't allow mushy semantics.
4096 */
Christian Brauner256c8ae2022-10-26 12:51:27 +02004097 if (mnt_idmap_owner(kattr->mnt_idmap) == fs_userns)
Christian Braunerbd303362021-12-03 12:17:07 +01004098 return -EINVAL;
4099
4100 /*
Christian Brauner9caccd42021-01-21 14:19:54 +01004101 * Once a mount has been idmapped we don't allow it to change its
4102 * mapping. It makes things simpler and callers can just create
4103 * another bind-mount they can idmap if they want to.
4104 */
Christian Braunerbb49e9e2021-12-03 12:16:58 +01004105 if (is_idmapped_mnt(m))
Christian Brauner9caccd42021-01-21 14:19:54 +01004106 return -EPERM;
4107
4108 /* The underlying filesystem doesn't support idmapped mounts yet. */
4109 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4110 return -EINVAL;
4111
4112 /* We're not controlling the superblock. */
Christian Braunerbd303362021-12-03 12:17:07 +01004113 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
Christian Brauner9caccd42021-01-21 14:19:54 +01004114 return -EPERM;
4115
4116 /* Mount has already been visible in the filesystem hierarchy. */
4117 if (!is_anon_ns(mnt->mnt_ns))
4118 return -EINVAL;
4119
4120 return 0;
4121}
4122
Christian Braunera26f7882022-02-03 14:14:08 +01004123/**
4124 * mnt_allow_writers() - check whether the attribute change allows writers
4125 * @kattr: the new mount attributes
4126 * @mnt: the mount to which @kattr will be applied
4127 *
4128 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4129 *
4130 * Return: true if writers need to be held, false if not
4131 */
4132static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4133 const struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004134{
Christian Braunere1bbcd22022-05-10 11:58:40 +02004135 return (!(kattr->attr_set & MNT_READONLY) ||
4136 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
Christian Brauner256c8ae2022-10-26 12:51:27 +02004137 !kattr->mnt_idmap;
Christian Braunera26f7882022-02-03 14:14:08 +01004138}
Christian Brauner2a186722021-01-21 14:19:53 +01004139
Christian Brauner87bb5b62022-02-03 14:14:11 +01004140static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004141{
Al Viroe2570392022-02-28 23:04:20 -05004142 struct mount *m;
4143 int err;
Christian Brauner2a186722021-01-21 14:19:53 +01004144
Al Viroe2570392022-02-28 23:04:20 -05004145 for (m = mnt; m; m = next_mnt(m, mnt)) {
4146 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4147 err = -EPERM;
4148 break;
Christian Brauner2a186722021-01-21 14:19:53 +01004149 }
4150
Christian Brauner87bb5b62022-02-03 14:14:11 +01004151 err = can_idmap_mount(kattr, m);
4152 if (err)
Al Viroe2570392022-02-28 23:04:20 -05004153 break;
Christian Brauner9caccd42021-01-21 14:19:54 +01004154
Al Viroe2570392022-02-28 23:04:20 -05004155 if (!mnt_allow_writers(kattr, m)) {
4156 err = mnt_hold_writers(m);
4157 if (err)
4158 break;
Christian Brauner2a186722021-01-21 14:19:53 +01004159 }
Christian Brauner2a186722021-01-21 14:19:53 +01004160
Al Viroe2570392022-02-28 23:04:20 -05004161 if (!kattr->recurse)
4162 return 0;
4163 }
Christian Brauner2a186722021-01-21 14:19:53 +01004164
Al Viroe2570392022-02-28 23:04:20 -05004165 if (err) {
4166 struct mount *p;
4167
Christian Brauner0014eda2022-04-20 15:19:25 +02004168 /*
4169 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4170 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4171 * mounts and needs to take care to include the first mount.
4172 */
4173 for (p = mnt; p; p = next_mnt(p, mnt)) {
Al Viroe2570392022-02-28 23:04:20 -05004174 /* If we had to hold writers unblock them. */
4175 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4176 mnt_unhold_writers(p);
Christian Brauner0014eda2022-04-20 15:19:25 +02004177
4178 /*
4179 * We're done once the first mount we changed got
4180 * MNT_WRITE_HOLD unset.
4181 */
4182 if (p == m)
4183 break;
Al Viroe2570392022-02-28 23:04:20 -05004184 }
4185 }
4186 return err;
Christian Brauner2a186722021-01-21 14:19:53 +01004187}
4188
Christian Brauner9caccd42021-01-21 14:19:54 +01004189static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4190{
Christian Brauner256c8ae2022-10-26 12:51:27 +02004191 if (!kattr->mnt_idmap)
Christian Brauner9caccd42021-01-21 14:19:54 +01004192 return;
4193
Christian Braunerbd303362021-12-03 12:17:07 +01004194 /*
Christian Brauner256c8ae2022-10-26 12:51:27 +02004195 * Pairs with smp_load_acquire() in mnt_idmap().
4196 *
4197 * Since we only allow a mount to change the idmapping once and
4198 * verified this in can_idmap_mount() we know that the mount has
4199 * @nop_mnt_idmap attached to it. So there's no need to drop any
4200 * references.
Christian Braunerbd303362021-12-03 12:17:07 +01004201 */
Christian Brauner256c8ae2022-10-26 12:51:27 +02004202 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
Christian Brauner9caccd42021-01-21 14:19:54 +01004203}
4204
Al Viroe2570392022-02-28 23:04:20 -05004205static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
Christian Brauner2a186722021-01-21 14:19:53 +01004206{
Al Viroe2570392022-02-28 23:04:20 -05004207 struct mount *m;
Christian Brauner2a186722021-01-21 14:19:53 +01004208
Al Viroe2570392022-02-28 23:04:20 -05004209 for (m = mnt; m; m = next_mnt(m, mnt)) {
4210 unsigned int flags;
Christian Brauner2a186722021-01-21 14:19:53 +01004211
Al Viroe2570392022-02-28 23:04:20 -05004212 do_idmap_mount(kattr, m);
4213 flags = recalc_flags(kattr, m);
4214 WRITE_ONCE(m->mnt.mnt_flags, flags);
Christian Brauner2a186722021-01-21 14:19:53 +01004215
Christian Brauner03b6abe2022-02-03 14:14:09 +01004216 /* If we had to hold writers unblock them. */
4217 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
Christian Brauner2a186722021-01-21 14:19:53 +01004218 mnt_unhold_writers(m);
4219
Al Viroe2570392022-02-28 23:04:20 -05004220 if (kattr->propagation)
Christian Brauner2a186722021-01-21 14:19:53 +01004221 change_mnt_propagation(m, kattr->propagation);
Al Viroe2570392022-02-28 23:04:20 -05004222 if (!kattr->recurse)
Christian Brauner2a186722021-01-21 14:19:53 +01004223 break;
Al Viroe2570392022-02-28 23:04:20 -05004224 }
4225 touch_mnt_namespace(mnt->mnt_ns);
Christian Brauner2a186722021-01-21 14:19:53 +01004226}
4227
4228static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4229{
Christian Brauner87bb5b62022-02-03 14:14:11 +01004230 struct mount *mnt = real_mount(path->mnt);
Christian Brauner2a186722021-01-21 14:19:53 +01004231 int err = 0;
4232
4233 if (path->dentry != mnt->mnt.mnt_root)
4234 return -EINVAL;
4235
Christian Brauner256c8ae2022-10-26 12:51:27 +02004236 if (kattr->mnt_userns) {
4237 struct mnt_idmap *mnt_idmap;
4238
4239 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4240 if (IS_ERR(mnt_idmap))
4241 return PTR_ERR(mnt_idmap);
4242 kattr->mnt_idmap = mnt_idmap;
4243 }
4244
Christian Brauner2a186722021-01-21 14:19:53 +01004245 if (kattr->propagation) {
4246 /*
4247 * Only take namespace_lock() if we're actually changing
4248 * propagation.
4249 */
4250 namespace_lock();
4251 if (kattr->propagation == MS_SHARED) {
4252 err = invent_group_ids(mnt, kattr->recurse);
4253 if (err) {
4254 namespace_unlock();
4255 return err;
4256 }
4257 }
4258 }
4259
Christian Brauner87bb5b62022-02-03 14:14:11 +01004260 err = -EINVAL;
Christian Brauner2a186722021-01-21 14:19:53 +01004261 lock_mount_hash();
4262
Christian Brauner87bb5b62022-02-03 14:14:11 +01004263 /* Ensure that this isn't anything purely vfs internal. */
4264 if (!is_mounted(&mnt->mnt))
4265 goto out;
Christian Brauner2a186722021-01-21 14:19:53 +01004266
Christian Brauner87bb5b62022-02-03 14:14:11 +01004267 /*
4268 * If this is an attached mount make sure it's located in the callers
4269 * mount namespace. If it's not don't let the caller interact with it.
4270 * If this is a detached mount make sure it has an anonymous mount
4271 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4272 */
4273 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4274 goto out;
4275
4276 /*
4277 * First, we get the mount tree in a shape where we can change mount
4278 * properties without failure. If we succeeded to do so we commit all
4279 * changes and if we failed we clean up.
4280 */
4281 err = mount_setattr_prepare(kattr, mnt);
Al Viroe2570392022-02-28 23:04:20 -05004282 if (!err)
4283 mount_setattr_commit(kattr, mnt);
Christian Brauner87bb5b62022-02-03 14:14:11 +01004284
4285out:
Christian Brauner2a186722021-01-21 14:19:53 +01004286 unlock_mount_hash();
4287
4288 if (kattr->propagation) {
4289 namespace_unlock();
4290 if (err)
4291 cleanup_group_ids(mnt, NULL);
4292 }
4293
4294 return err;
4295}
4296
Christian Brauner9caccd42021-01-21 14:19:54 +01004297static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4298 struct mount_kattr *kattr, unsigned int flags)
4299{
4300 int err = 0;
4301 struct ns_common *ns;
4302 struct user_namespace *mnt_userns;
4303 struct file *file;
4304
4305 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4306 return 0;
4307
4308 /*
4309 * We currently do not support clearing an idmapped mount. If this ever
4310 * is a use-case we can revisit this but for now let's keep it simple
4311 * and not allow it.
4312 */
4313 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4314 return -EINVAL;
4315
4316 if (attr->userns_fd > INT_MAX)
4317 return -EINVAL;
4318
4319 file = fget(attr->userns_fd);
4320 if (!file)
4321 return -EBADF;
4322
4323 if (!proc_ns_file(file)) {
4324 err = -EINVAL;
4325 goto out_fput;
4326 }
4327
4328 ns = get_proc_ns(file_inode(file));
4329 if (ns->ops->type != CLONE_NEWUSER) {
4330 err = -EINVAL;
4331 goto out_fput;
4332 }
4333
4334 /*
Christian Braunerbd303362021-12-03 12:17:07 +01004335 * The initial idmapping cannot be used to create an idmapped
4336 * mount. We use the initial idmapping as an indicator of a mount
4337 * that is not idmapped. It can simply be passed into helpers that
4338 * are aware of idmapped mounts as a convenient shortcut. A user
4339 * can just create a dedicated identity mapping to achieve the same
4340 * result.
Christian Brauner9caccd42021-01-21 14:19:54 +01004341 */
4342 mnt_userns = container_of(ns, struct user_namespace, ns);
Christian Braunerbd303362021-12-03 12:17:07 +01004343 if (initial_idmapping(mnt_userns)) {
Christian Brauner9caccd42021-01-21 14:19:54 +01004344 err = -EPERM;
4345 goto out_fput;
4346 }
Seth Forsheebf1ac162022-08-16 11:47:52 -05004347
4348 /* We're not controlling the target namespace. */
4349 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4350 err = -EPERM;
4351 goto out_fput;
4352 }
4353
Christian Brauner9caccd42021-01-21 14:19:54 +01004354 kattr->mnt_userns = get_user_ns(mnt_userns);
4355
4356out_fput:
4357 fput(file);
4358 return err;
4359}
4360
4361static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
Christian Brauner2a186722021-01-21 14:19:53 +01004362 struct mount_kattr *kattr, unsigned int flags)
4363{
4364 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4365
4366 if (flags & AT_NO_AUTOMOUNT)
4367 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4368 if (flags & AT_SYMLINK_NOFOLLOW)
4369 lookup_flags &= ~LOOKUP_FOLLOW;
4370 if (flags & AT_EMPTY_PATH)
4371 lookup_flags |= LOOKUP_EMPTY;
4372
4373 *kattr = (struct mount_kattr) {
4374 .lookup_flags = lookup_flags,
4375 .recurse = !!(flags & AT_RECURSIVE),
4376 };
4377
4378 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4379 return -EINVAL;
4380 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4381 return -EINVAL;
4382 kattr->propagation = attr->propagation;
4383
4384 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4385 return -EINVAL;
4386
Christian Brauner2a186722021-01-21 14:19:53 +01004387 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4388 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4389
4390 /*
4391 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4392 * users wanting to transition to a different atime setting cannot
4393 * simply specify the atime setting in @attr_set, but must also
4394 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4395 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4396 * @attr_clr and that @attr_set can't have any atime bits set if
4397 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4398 */
4399 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4400 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4401 return -EINVAL;
4402
4403 /*
4404 * Clear all previous time settings as they are mutually
4405 * exclusive.
4406 */
4407 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4408 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4409 case MOUNT_ATTR_RELATIME:
4410 kattr->attr_set |= MNT_RELATIME;
4411 break;
4412 case MOUNT_ATTR_NOATIME:
4413 kattr->attr_set |= MNT_NOATIME;
4414 break;
4415 case MOUNT_ATTR_STRICTATIME:
4416 break;
4417 default:
4418 return -EINVAL;
4419 }
4420 } else {
4421 if (attr->attr_set & MOUNT_ATTR__ATIME)
4422 return -EINVAL;
4423 }
4424
Christian Brauner9caccd42021-01-21 14:19:54 +01004425 return build_mount_idmapped(attr, usize, kattr, flags);
4426}
4427
4428static void finish_mount_kattr(struct mount_kattr *kattr)
4429{
4430 put_user_ns(kattr->mnt_userns);
4431 kattr->mnt_userns = NULL;
Christian Brauner256c8ae2022-10-26 12:51:27 +02004432
4433 if (kattr->mnt_idmap)
4434 mnt_idmap_put(kattr->mnt_idmap);
Christian Brauner2a186722021-01-21 14:19:53 +01004435}
4436
4437SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4438 unsigned int, flags, struct mount_attr __user *, uattr,
4439 size_t, usize)
4440{
4441 int err;
4442 struct path target;
4443 struct mount_attr attr;
4444 struct mount_kattr kattr;
4445
4446 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4447
4448 if (flags & ~(AT_EMPTY_PATH |
4449 AT_RECURSIVE |
4450 AT_SYMLINK_NOFOLLOW |
4451 AT_NO_AUTOMOUNT))
4452 return -EINVAL;
4453
4454 if (unlikely(usize > PAGE_SIZE))
4455 return -E2BIG;
4456 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4457 return -EINVAL;
4458
4459 if (!may_mount())
4460 return -EPERM;
4461
4462 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4463 if (err)
4464 return err;
4465
4466 /* Don't bother walking through the mounts if this is a nop. */
4467 if (attr.attr_set == 0 &&
4468 attr.attr_clr == 0 &&
4469 attr.propagation == 0)
4470 return 0;
4471
Christian Brauner9caccd42021-01-21 14:19:54 +01004472 err = build_mount_kattr(&attr, usize, &kattr, flags);
Christian Brauner2a186722021-01-21 14:19:53 +01004473 if (err)
4474 return err;
4475
4476 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
Christian Brauner012e3322021-12-30 20:23:09 +01004477 if (!err) {
4478 err = do_mount_setattr(&target, &kattr);
4479 path_put(&target);
4480 }
Christian Brauner9caccd42021-01-21 14:19:54 +01004481 finish_mount_kattr(&kattr);
Christian Brauner2a186722021-01-21 14:19:53 +01004482 return err;
4483}
4484
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485static void __init init_mount_tree(void)
4486{
4487 struct vfsmount *mnt;
Al Viro74e83122019-01-30 13:30:21 -05004488 struct mount *m;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08004489 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08004490 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491
Al Virofd3e0072019-05-30 17:48:35 -04004492 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004493 if (IS_ERR(mnt))
4494 panic("Can't create rootfs");
Nick Pigginb3e19d92011-01-07 17:50:11 +11004495
Al Viro74e83122019-01-30 13:30:21 -05004496 ns = alloc_mnt_ns(&init_user_ns, false);
Trond Myklebust3b22edc2009-06-23 17:29:49 -04004497 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004498 panic("Can't allocate initial namespace");
Al Viro74e83122019-01-30 13:30:21 -05004499 m = real_mount(mnt);
4500 m->mnt_ns = ns;
4501 ns->root = m;
4502 ns->mounts = 1;
4503 list_add(&m->mnt_list, &ns->list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08004504 init_task.nsproxy->mnt_ns = ns;
4505 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004506
Al Virobe08d6d2011-12-06 13:32:36 -05004507 root.mnt = mnt;
4508 root.dentry = mnt->mnt_root;
Eric W. Biedermanda362b02014-10-07 12:19:53 -07004509 mnt->mnt_flags |= MNT_LOCKED;
Jan Blunckac748a02008-02-14 19:34:39 -08004510
4511 set_fs_pwd(current->fs, &root);
4512 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004513}
4514
Denis Cheng74bf17c2007-10-16 23:26:30 -07004515void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004516{
Randy Dunlap15a67dd2006-09-29 01:58:57 -07004517 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004518
Al Viro7d6fec42011-11-23 12:14:10 -05004519 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
Vasily Averin79f65402021-09-02 14:55:10 -07004520 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004521
Al Viro0818bf22014-02-28 13:46:44 -05004522 mount_hashtable = alloc_large_system_hash("Mount-cache",
Al Viro38129a12014-03-20 21:10:51 -04004523 sizeof(struct hlist_head),
Al Viro0818bf22014-02-28 13:46:44 -05004524 mhash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07004525 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05004526 &m_hash_shift, &m_hash_mask, 0, 0);
4527 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4528 sizeof(struct hlist_head),
4529 mphash_entries, 19,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07004530 HASH_ZERO,
Al Viro0818bf22014-02-28 13:46:44 -05004531 &mp_hash_shift, &mp_hash_mask, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004532
Al Viro84d171922013-03-15 10:53:28 -04004533 if (!mount_hashtable || !mountpoint_hashtable)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004534 panic("Failed to allocate mount hash table\n");
4535
Tejun Heo4b93dc92013-11-28 14:54:43 -05004536 kernfs_init();
4537
Randy Dunlap15a67dd2006-09-29 01:58:57 -07004538 err = sysfs_init();
4539 if (err)
4540 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07004541 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06004542 fs_kobj = kobject_create_and_add("fs", NULL);
4543 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07004544 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Al Viro037f11b2019-06-01 18:09:44 -04004545 shmem_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004546 init_rootfs();
4547 init_mount_tree();
4548}
4549
Trond Myklebust616511d2009-06-22 15:09:13 -04004550void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004551{
Kirill Tkhai1a7b89692020-08-03 13:16:42 +03004552 if (!refcount_dec_and_test(&ns->ns.count))
Trond Myklebust616511d2009-06-22 15:09:13 -04004553 return;
Al Viro7b00ed62013-09-16 21:19:20 -04004554 drop_collected_mounts(&ns->root->mnt);
Eric W. Biederman771b1372012-07-26 21:08:32 -07004555 free_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004556}
Al Viro9d412a42011-03-17 22:08:28 -04004557
David Howellsd911b452018-11-01 23:07:26 +00004558struct vfsmount *kern_mount(struct file_system_type *type)
Al Viro9d412a42011-03-17 22:08:28 -04004559{
Tim Chen423e0ab2011-07-19 09:32:38 -07004560 struct vfsmount *mnt;
David Howellsd911b452018-11-01 23:07:26 +00004561 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
Tim Chen423e0ab2011-07-19 09:32:38 -07004562 if (!IS_ERR(mnt)) {
4563 /*
4564 * it is a longterm mount, don't release mnt until
4565 * we unmount before file sys is unregistered
4566 */
Al Virof7a99c52012-06-09 00:59:08 -04004567 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
Tim Chen423e0ab2011-07-19 09:32:38 -07004568 }
4569 return mnt;
Al Viro9d412a42011-03-17 22:08:28 -04004570}
David Howellsd911b452018-11-01 23:07:26 +00004571EXPORT_SYMBOL_GPL(kern_mount);
Tim Chen423e0ab2011-07-19 09:32:38 -07004572
4573void kern_unmount(struct vfsmount *mnt)
4574{
4575 /* release long term mount so mount point can be released */
4576 if (!IS_ERR_OR_NULL(mnt)) {
Al Virof7a99c52012-06-09 00:59:08 -04004577 real_mount(mnt)->mnt_ns = NULL;
Al Viro48a066e2013-09-29 22:06:07 -04004578 synchronize_rcu(); /* yecchhh... */
Tim Chen423e0ab2011-07-19 09:32:38 -07004579 mntput(mnt);
4580 }
4581}
4582EXPORT_SYMBOL(kern_unmount);
Al Viro02125a82011-12-05 08:43:34 -05004583
Miklos Szeredidf820f82020-06-04 10:48:19 +02004584void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4585{
4586 unsigned int i;
4587
4588 for (i = 0; i < num; i++)
4589 if (mnt[i])
4590 real_mount(mnt[i])->mnt_ns = NULL;
4591 synchronize_rcu_expedited();
4592 for (i = 0; i < num; i++)
4593 mntput(mnt[i]);
4594}
4595EXPORT_SYMBOL(kern_unmount_array);
4596
Al Viro02125a82011-12-05 08:43:34 -05004597bool our_mnt(struct vfsmount *mnt)
4598{
Al Viro143c8c92011-11-25 00:46:35 -05004599 return check_mnt(real_mount(mnt));
Al Viro02125a82011-12-05 08:43:34 -05004600}
Eric W. Biederman8823c072010-03-07 18:49:36 -08004601
Eric W. Biederman31515272013-03-15 01:45:51 -07004602bool current_chrooted(void)
4603{
4604 /* Does the current process have a non-standard root */
4605 struct path ns_root;
4606 struct path fs_root;
4607 bool chrooted;
4608
4609 /* Find the namespace root */
4610 ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4611 ns_root.dentry = ns_root.mnt->mnt_root;
4612 path_get(&ns_root);
4613 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4614 ;
4615
4616 get_fs_root(current->fs, &fs_root);
4617
4618 chrooted = !path_equal(&fs_root, &ns_root);
4619
4620 path_put(&fs_root);
4621 path_put(&ns_root);
4622
4623 return chrooted;
4624}
4625
David Howells132e4602018-11-04 07:43:08 -05004626static bool mnt_already_visible(struct mnt_namespace *ns,
4627 const struct super_block *sb,
Eric W. Biederman8654df42016-06-09 16:06:06 -05004628 int *new_mnt_flags)
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004629{
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05004630 int new_flags = *new_mnt_flags;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004631 struct mount *mnt;
Eric W. Biedermane51db732013-03-30 19:57:41 -07004632 bool visible = false;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004633
Al Viro44bb4382013-09-16 21:37:36 -04004634 down_read(&namespace_sem);
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02004635 lock_ns_list(ns);
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004636 list_for_each_entry(mnt, &ns->list, mnt_list) {
Eric W. Biedermane51db732013-03-30 19:57:41 -07004637 struct mount *child;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004638 int mnt_flags;
4639
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02004640 if (mnt_is_cursor(mnt))
4641 continue;
4642
David Howells132e4602018-11-04 07:43:08 -05004643 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
Eric W. Biedermane51db732013-03-30 19:57:41 -07004644 continue;
4645
Eric W. Biederman7e96c1b2015-05-08 16:36:50 -05004646 /* This mount is not fully visible if it's root directory
4647 * is not the root directory of the filesystem.
4648 */
4649 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4650 continue;
4651
Eric W. Biedermana1935c12016-06-15 06:59:49 -05004652 /* A local view of the mount flags */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004653 mnt_flags = mnt->mnt.mnt_flags;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004654
Eric W. Biederman695e9df2016-06-10 12:21:40 -05004655 /* Don't miss readonly hidden in the superblock flags */
David Howellsbc98a422017-07-17 08:45:34 +01004656 if (sb_rdonly(mnt->mnt.mnt_sb))
Eric W. Biederman695e9df2016-06-10 12:21:40 -05004657 mnt_flags |= MNT_LOCK_READONLY;
4658
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05004659 /* Verify the mount flags are equal to or more permissive
4660 * than the proposed new mount.
4661 */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004662 if ((mnt_flags & MNT_LOCK_READONLY) &&
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05004663 !(new_flags & MNT_READONLY))
4664 continue;
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004665 if ((mnt_flags & MNT_LOCK_ATIME) &&
4666 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05004667 continue;
4668
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06004669 /* This mount is not fully visible if there are any
4670 * locked child mounts that cover anything except for
4671 * empty directories.
Eric W. Biedermane51db732013-03-30 19:57:41 -07004672 */
4673 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4674 struct inode *inode = child->mnt_mountpoint->d_inode;
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06004675 /* Only worry about locked mounts */
Eric W. Biedermand71ed6c2016-05-27 14:50:05 -05004676 if (!(child->mnt.mnt_flags & MNT_LOCKED))
Eric W. Biedermanceeb0e52015-01-07 08:10:09 -06004677 continue;
Eric W. Biederman7236c852015-05-13 20:51:09 -05004678 /* Is the directory permanetly empty? */
4679 if (!is_empty_dir_inode(inode))
Eric W. Biedermane51db732013-03-30 19:57:41 -07004680 goto next;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004681 }
Eric W. Biederman8c6cf9c2015-05-08 23:49:47 -05004682 /* Preserve the locked attributes */
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004683 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
Eric W. Biederman77b1a972015-06-04 09:43:11 -05004684 MNT_LOCK_ATIME);
Eric W. Biedermane51db732013-03-30 19:57:41 -07004685 visible = true;
4686 goto found;
4687 next: ;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004688 }
Eric W. Biedermane51db732013-03-30 19:57:41 -07004689found:
Miklos Szeredi9f6c61f2020-05-14 16:44:24 +02004690 unlock_ns_list(ns);
Al Viro44bb4382013-09-16 21:37:36 -04004691 up_read(&namespace_sem);
Eric W. Biedermane51db732013-03-30 19:57:41 -07004692 return visible;
Eric W. Biederman87a8ebd2013-03-24 14:28:27 -07004693}
4694
David Howells132e4602018-11-04 07:43:08 -05004695static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
Eric W. Biederman8654df42016-06-09 16:06:06 -05004696{
Eric W. Biedermana1935c12016-06-15 06:59:49 -05004697 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
Eric W. Biederman8654df42016-06-09 16:06:06 -05004698 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4699 unsigned long s_iflags;
4700
4701 if (ns->user_ns == &init_user_ns)
4702 return false;
4703
4704 /* Can this filesystem be too revealing? */
David Howells132e4602018-11-04 07:43:08 -05004705 s_iflags = sb->s_iflags;
Eric W. Biederman8654df42016-06-09 16:06:06 -05004706 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4707 return false;
4708
Eric W. Biedermana1935c12016-06-15 06:59:49 -05004709 if ((s_iflags & required_iflags) != required_iflags) {
4710 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4711 required_iflags);
4712 return true;
4713 }
4714
David Howells132e4602018-11-04 07:43:08 -05004715 return !mnt_already_visible(ns, sb, new_mnt_flags);
Eric W. Biederman8654df42016-06-09 16:06:06 -05004716}
4717
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05004718bool mnt_may_suid(struct vfsmount *mnt)
4719{
4720 /*
4721 * Foreign mounts (accessed via fchdir or through /proc
4722 * symlinks) are always treated as if they are nosuid. This
4723 * prevents namespaces from trusting potentially unsafe
4724 * suid/sgid bits, file caps, or security labels that originate
4725 * in other namespaces.
4726 */
4727 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4728 current_in_userns(mnt->mnt_sb->s_user_ns);
4729}
4730
Al Viro64964522014-11-01 00:37:32 -04004731static struct ns_common *mntns_get(struct task_struct *task)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004732{
Al Viro58be28252014-11-01 00:00:23 -04004733 struct ns_common *ns = NULL;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004734 struct nsproxy *nsproxy;
4735
Eric W. Biederman728dba32014-02-03 19:13:49 -08004736 task_lock(task);
4737 nsproxy = task->nsproxy;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004738 if (nsproxy) {
Al Viro58be28252014-11-01 00:00:23 -04004739 ns = &nsproxy->mnt_ns->ns;
4740 get_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08004741 }
Eric W. Biederman728dba32014-02-03 19:13:49 -08004742 task_unlock(task);
Eric W. Biederman8823c072010-03-07 18:49:36 -08004743
4744 return ns;
4745}
4746
Al Viro64964522014-11-01 00:37:32 -04004747static void mntns_put(struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004748{
Al Viro58be28252014-11-01 00:00:23 -04004749 put_mnt_ns(to_mnt_ns(ns));
Eric W. Biederman8823c072010-03-07 18:49:36 -08004750}
4751
Christian Braunerf2a8d522020-05-05 16:04:30 +02004752static int mntns_install(struct nsset *nsset, struct ns_common *ns)
Eric W. Biederman8823c072010-03-07 18:49:36 -08004753{
Christian Braunerf2a8d522020-05-05 16:04:30 +02004754 struct nsproxy *nsproxy = nsset->nsproxy;
4755 struct fs_struct *fs = nsset->fs;
Al Viro4f757f32017-04-15 17:31:22 -04004756 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
Christian Braunerf2a8d522020-05-05 16:04:30 +02004757 struct user_namespace *user_ns = nsset->cred->user_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004758 struct path root;
Al Viro4f757f32017-04-15 17:31:22 -04004759 int err;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004760
Eric W. Biederman0c55cfc2012-07-26 21:42:03 -07004761 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
Christian Braunerf2a8d522020-05-05 16:04:30 +02004762 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4763 !ns_capable(user_ns, CAP_SYS_ADMIN))
Zhao Hongjiangae11e0f2012-09-13 16:38:03 +08004764 return -EPERM;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004765
Al Viro74e83122019-01-30 13:30:21 -05004766 if (is_anon_ns(mnt_ns))
4767 return -EINVAL;
4768
Eric W. Biederman8823c072010-03-07 18:49:36 -08004769 if (fs->users != 1)
4770 return -EINVAL;
4771
4772 get_mnt_ns(mnt_ns);
Al Viro4f757f32017-04-15 17:31:22 -04004773 old_mnt_ns = nsproxy->mnt_ns;
Eric W. Biederman8823c072010-03-07 18:49:36 -08004774 nsproxy->mnt_ns = mnt_ns;
4775
4776 /* Find the root */
Al Viro4f757f32017-04-15 17:31:22 -04004777 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4778 "/", LOOKUP_DOWN, &root);
4779 if (err) {
4780 /* revert to old namespace */
4781 nsproxy->mnt_ns = old_mnt_ns;
4782 put_mnt_ns(mnt_ns);
4783 return err;
4784 }
Eric W. Biederman8823c072010-03-07 18:49:36 -08004785
Andrei Vagin40683672017-06-08 17:32:29 -07004786 put_mnt_ns(old_mnt_ns);
4787
Eric W. Biederman8823c072010-03-07 18:49:36 -08004788 /* Update the pwd and root */
4789 set_fs_pwd(fs, &root);
4790 set_fs_root(fs, &root);
4791
4792 path_put(&root);
4793 return 0;
4794}
4795
Andrey Vaginbcac25a2016-09-06 00:47:13 -07004796static struct user_namespace *mntns_owner(struct ns_common *ns)
4797{
4798 return to_mnt_ns(ns)->user_ns;
4799}
4800
Eric W. Biederman8823c072010-03-07 18:49:36 -08004801const struct proc_ns_operations mntns_operations = {
4802 .name = "mnt",
4803 .type = CLONE_NEWNS,
4804 .get = mntns_get,
4805 .put = mntns_put,
4806 .install = mntns_install,
Andrey Vaginbcac25a2016-09-06 00:47:13 -07004807 .owner = mntns_owner,
Eric W. Biederman8823c072010-03-07 18:49:36 -08004808};
Luis Chamberlainab171b92022-01-21 22:13:27 -08004809
4810#ifdef CONFIG_SYSCTL
4811static struct ctl_table fs_namespace_sysctls[] = {
4812 {
4813 .procname = "mount-max",
4814 .data = &sysctl_mount_max,
4815 .maxlen = sizeof(unsigned int),
4816 .mode = 0644,
4817 .proc_handler = proc_dointvec_minmax,
4818 .extra1 = SYSCTL_ONE,
4819 },
4820 { }
4821};
4822
4823static int __init init_fs_namespace_sysctls(void)
4824{
4825 register_sysctl_init("fs", fs_namespace_sysctls);
4826 return 0;
4827}
4828fs_initcall(init_fs_namespace_sysctls);
4829
4830#endif /* CONFIG_SYSCTL */