]>
Commit | Line | Data |
---|---|---|
59bd9ded | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namespace.c | |
4 | * | |
5 | * (C) Copyright Al Viro 2000, 2001 | |
1da177e4 LT |
6 | * |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
a07b2000 | 23 | #include <linux/file.h> |
d10577a8 | 24 | #include <linux/uaccess.h> |
0bb80f24 | 25 | #include <linux/proc_ns.h> |
20b4fb48 | 26 | #include <linux/magic.h> |
57c8a661 | 27 | #include <linux/memblock.h> |
9ea459e1 | 28 | #include <linux/task_work.h> |
9164bb4a | 29 | #include <linux/sched/task.h> |
e262e32d | 30 | #include <uapi/linux/mount.h> |
9bc61ab1 | 31 | #include <linux/fs_context.h> |
037f11b4 | 32 | #include <linux/shmem_fs.h> |
9164bb4a | 33 | |
07b20889 | 34 | #include "pnode.h" |
948730b0 | 35 | #include "internal.h" |
1da177e4 | 36 | |
d2921684 EB |
37 | /* Maximum number of mounts in a mount namespace */ |
38 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
39 | ||
0818bf27 AV |
40 | static unsigned int m_hash_mask __read_mostly; |
41 | static unsigned int m_hash_shift __read_mostly; | |
42 | static unsigned int mp_hash_mask __read_mostly; | |
43 | static unsigned int mp_hash_shift __read_mostly; | |
44 | ||
45 | static __initdata unsigned long mhash_entries; | |
46 | static int __init set_mhash_entries(char *str) | |
47 | { | |
48 | if (!str) | |
49 | return 0; | |
50 | mhash_entries = simple_strtoul(str, &str, 0); | |
51 | return 1; | |
52 | } | |
53 | __setup("mhash_entries=", set_mhash_entries); | |
54 | ||
55 | static __initdata unsigned long mphash_entries; | |
56 | static int __init set_mphash_entries(char *str) | |
57 | { | |
58 | if (!str) | |
59 | return 0; | |
60 | mphash_entries = simple_strtoul(str, &str, 0); | |
61 | return 1; | |
62 | } | |
63 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 64 | |
c7999c36 | 65 | static u64 event; |
73cd49ec | 66 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 67 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 68 | |
38129a13 | 69 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 70 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 71 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 72 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
73 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
74 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 75 | |
2a186721 CB |
76 | struct mount_kattr { |
77 | unsigned int attr_set; | |
78 | unsigned int attr_clr; | |
79 | unsigned int propagation; | |
80 | unsigned int lookup_flags; | |
81 | bool recurse; | |
82 | }; | |
83 | ||
f87fd4c2 | 84 | /* /sys/fs */ |
00d26666 GKH |
85 | struct kobject *fs_kobj; |
86 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 87 | |
99b7db7b NP |
88 | /* |
89 | * vfsmount lock may be taken for read to prevent changes to the | |
90 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
91 | * up the tree. | |
92 | * | |
93 | * It should be taken for write in all cases where the vfsmount | |
94 | * tree or hash is modified or when a vfsmount structure is modified. | |
95 | */ | |
48a066e7 | 96 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 97 | |
d033cb67 CB |
98 | static inline void lock_mount_hash(void) |
99 | { | |
100 | write_seqlock(&mount_lock); | |
101 | } | |
102 | ||
103 | static inline void unlock_mount_hash(void) | |
104 | { | |
105 | write_sequnlock(&mount_lock); | |
106 | } | |
107 | ||
38129a13 | 108 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 109 | { |
b58fed8b RP |
110 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
111 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
112 | tmp = tmp + (tmp >> m_hash_shift); |
113 | return &mount_hashtable[tmp & m_hash_mask]; | |
114 | } | |
115 | ||
116 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
117 | { | |
118 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
119 | tmp = tmp + (tmp >> mp_hash_shift); | |
120 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
121 | } |
122 | ||
b105e270 | 123 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 124 | { |
169b480e MW |
125 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
126 | ||
127 | if (res < 0) | |
128 | return res; | |
129 | mnt->mnt_id = res; | |
130 | return 0; | |
73cd49ec MS |
131 | } |
132 | ||
b105e270 | 133 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 134 | { |
169b480e | 135 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
136 | } |
137 | ||
719f5d7f MS |
138 | /* |
139 | * Allocate a new peer group ID | |
719f5d7f | 140 | */ |
4b8b21f4 | 141 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 142 | { |
169b480e | 143 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 144 | |
169b480e MW |
145 | if (res < 0) |
146 | return res; | |
147 | mnt->mnt_group_id = res; | |
148 | return 0; | |
719f5d7f MS |
149 | } |
150 | ||
151 | /* | |
152 | * Release a peer group ID | |
153 | */ | |
4b8b21f4 | 154 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 155 | { |
169b480e | 156 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 157 | mnt->mnt_group_id = 0; |
719f5d7f MS |
158 | } |
159 | ||
b3e19d92 NP |
160 | /* |
161 | * vfsmount lock must be held for read | |
162 | */ | |
83adc753 | 163 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
164 | { |
165 | #ifdef CONFIG_SMP | |
68e8a9fe | 166 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
167 | #else |
168 | preempt_disable(); | |
68e8a9fe | 169 | mnt->mnt_count += n; |
b3e19d92 NP |
170 | preempt_enable(); |
171 | #endif | |
172 | } | |
173 | ||
b3e19d92 NP |
174 | /* |
175 | * vfsmount lock must be held for write | |
176 | */ | |
edf7ddbf | 177 | int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
178 | { |
179 | #ifdef CONFIG_SMP | |
edf7ddbf | 180 | int count = 0; |
b3e19d92 NP |
181 | int cpu; |
182 | ||
183 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 184 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
185 | } |
186 | ||
187 | return count; | |
188 | #else | |
68e8a9fe | 189 | return mnt->mnt_count; |
b3e19d92 NP |
190 | #endif |
191 | } | |
192 | ||
b105e270 | 193 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 194 | { |
c63181e6 AV |
195 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
196 | if (mnt) { | |
73cd49ec MS |
197 | int err; |
198 | ||
c63181e6 | 199 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
200 | if (err) |
201 | goto out_free_cache; | |
202 | ||
203 | if (name) { | |
fcc139ae | 204 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 205 | if (!mnt->mnt_devname) |
88b38782 | 206 | goto out_free_id; |
73cd49ec MS |
207 | } |
208 | ||
b3e19d92 | 209 | #ifdef CONFIG_SMP |
c63181e6 AV |
210 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
211 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
212 | goto out_free_devname; |
213 | ||
c63181e6 | 214 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 215 | #else |
c63181e6 AV |
216 | mnt->mnt_count = 1; |
217 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
218 | #endif |
219 | ||
38129a13 | 220 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
221 | INIT_LIST_HEAD(&mnt->mnt_child); |
222 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
223 | INIT_LIST_HEAD(&mnt->mnt_list); | |
224 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
225 | INIT_LIST_HEAD(&mnt->mnt_share); | |
226 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
227 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 228 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 229 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
56cbb429 | 230 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
a6435940 | 231 | mnt->mnt.mnt_userns = &init_user_ns; |
1da177e4 | 232 | } |
c63181e6 | 233 | return mnt; |
88b38782 | 234 | |
d3ef3d73 NP |
235 | #ifdef CONFIG_SMP |
236 | out_free_devname: | |
fcc139ae | 237 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 238 | #endif |
88b38782 | 239 | out_free_id: |
c63181e6 | 240 | mnt_free_id(mnt); |
88b38782 | 241 | out_free_cache: |
c63181e6 | 242 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 243 | return NULL; |
1da177e4 LT |
244 | } |
245 | ||
3d733633 DH |
246 | /* |
247 | * Most r/o checks on a fs are for operations that take | |
248 | * discrete amounts of time, like a write() or unlink(). | |
249 | * We must keep track of when those operations start | |
250 | * (for permission checks) and when they end, so that | |
251 | * we can determine when writes are able to occur to | |
252 | * a filesystem. | |
253 | */ | |
254 | /* | |
255 | * __mnt_is_readonly: check whether a mount is read-only | |
256 | * @mnt: the mount to check for its write status | |
257 | * | |
258 | * This shouldn't be used directly ouside of the VFS. | |
259 | * It does not guarantee that the filesystem will stay | |
260 | * r/w, just that it is right *now*. This can not and | |
261 | * should not be used in place of IS_RDONLY(inode). | |
262 | * mnt_want/drop_write() will _keep_ the filesystem | |
263 | * r/w. | |
264 | */ | |
43f5e655 | 265 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 266 | { |
43f5e655 | 267 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
268 | } |
269 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
270 | ||
83adc753 | 271 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
272 | { |
273 | #ifdef CONFIG_SMP | |
68e8a9fe | 274 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 275 | #else |
68e8a9fe | 276 | mnt->mnt_writers++; |
d3ef3d73 NP |
277 | #endif |
278 | } | |
3d733633 | 279 | |
83adc753 | 280 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 281 | { |
d3ef3d73 | 282 | #ifdef CONFIG_SMP |
68e8a9fe | 283 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 284 | #else |
68e8a9fe | 285 | mnt->mnt_writers--; |
d3ef3d73 | 286 | #endif |
3d733633 | 287 | } |
3d733633 | 288 | |
83adc753 | 289 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 290 | { |
d3ef3d73 NP |
291 | #ifdef CONFIG_SMP |
292 | unsigned int count = 0; | |
3d733633 | 293 | int cpu; |
3d733633 DH |
294 | |
295 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 296 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 297 | } |
3d733633 | 298 | |
d3ef3d73 NP |
299 | return count; |
300 | #else | |
301 | return mnt->mnt_writers; | |
302 | #endif | |
3d733633 DH |
303 | } |
304 | ||
4ed5e82f MS |
305 | static int mnt_is_readonly(struct vfsmount *mnt) |
306 | { | |
307 | if (mnt->mnt_sb->s_readonly_remount) | |
308 | return 1; | |
309 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
310 | smp_rmb(); | |
311 | return __mnt_is_readonly(mnt); | |
312 | } | |
313 | ||
8366025e | 314 | /* |
eb04c282 JK |
315 | * Most r/o & frozen checks on a fs are for operations that take discrete |
316 | * amounts of time, like a write() or unlink(). We must keep track of when | |
317 | * those operations start (for permission checks) and when they end, so that we | |
318 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
319 | */ |
320 | /** | |
eb04c282 | 321 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 322 | * @m: the mount on which to take a write |
8366025e | 323 | * |
eb04c282 JK |
324 | * This tells the low-level filesystem that a write is about to be performed to |
325 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
326 | * returning success. This operation does not protect against filesystem being | |
327 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
328 | * called. This is effectively a refcount. | |
8366025e | 329 | */ |
eb04c282 | 330 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 331 | { |
83adc753 | 332 | struct mount *mnt = real_mount(m); |
3d733633 | 333 | int ret = 0; |
3d733633 | 334 | |
d3ef3d73 | 335 | preempt_disable(); |
c6653a83 | 336 | mnt_inc_writers(mnt); |
d3ef3d73 | 337 | /* |
c6653a83 | 338 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
339 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
340 | * incremented count after it has set MNT_WRITE_HOLD. | |
341 | */ | |
342 | smp_mb(); | |
6aa7de05 | 343 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 NP |
344 | cpu_relax(); |
345 | /* | |
346 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
347 | * be set to match its requirements. So we must not load that until | |
348 | * MNT_WRITE_HOLD is cleared. | |
349 | */ | |
350 | smp_rmb(); | |
4ed5e82f | 351 | if (mnt_is_readonly(m)) { |
c6653a83 | 352 | mnt_dec_writers(mnt); |
3d733633 | 353 | ret = -EROFS; |
3d733633 | 354 | } |
d3ef3d73 | 355 | preempt_enable(); |
eb04c282 JK |
356 | |
357 | return ret; | |
358 | } | |
359 | ||
360 | /** | |
361 | * mnt_want_write - get write access to a mount | |
362 | * @m: the mount on which to take a write | |
363 | * | |
364 | * This tells the low-level filesystem that a write is about to be performed to | |
365 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
366 | * is not frozen) before returning success. When the write operation is | |
367 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
368 | */ | |
369 | int mnt_want_write(struct vfsmount *m) | |
370 | { | |
371 | int ret; | |
372 | ||
373 | sb_start_write(m->mnt_sb); | |
374 | ret = __mnt_want_write(m); | |
375 | if (ret) | |
376 | sb_end_write(m->mnt_sb); | |
3d733633 | 377 | return ret; |
8366025e DH |
378 | } |
379 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
380 | ||
96029c4e NP |
381 | /** |
382 | * mnt_clone_write - get write access to a mount | |
383 | * @mnt: the mount on which to take a write | |
384 | * | |
385 | * This is effectively like mnt_want_write, except | |
386 | * it must only be used to take an extra write reference | |
387 | * on a mountpoint that we already know has a write reference | |
388 | * on it. This allows some optimisation. | |
389 | * | |
390 | * After finished, mnt_drop_write must be called as usual to | |
391 | * drop the reference. | |
392 | */ | |
393 | int mnt_clone_write(struct vfsmount *mnt) | |
394 | { | |
395 | /* superblock may be r/o */ | |
396 | if (__mnt_is_readonly(mnt)) | |
397 | return -EROFS; | |
398 | preempt_disable(); | |
83adc753 | 399 | mnt_inc_writers(real_mount(mnt)); |
96029c4e NP |
400 | preempt_enable(); |
401 | return 0; | |
402 | } | |
403 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
404 | ||
405 | /** | |
eb04c282 | 406 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
407 | * @file: the file who's mount on which to take a write |
408 | * | |
eb04c282 | 409 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e NP |
410 | * do some optimisations if the file is open for write already |
411 | */ | |
eb04c282 | 412 | int __mnt_want_write_file(struct file *file) |
96029c4e | 413 | { |
83f936c7 | 414 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 415 | return __mnt_want_write(file->f_path.mnt); |
96029c4e NP |
416 | else |
417 | return mnt_clone_write(file->f_path.mnt); | |
418 | } | |
eb04c282 | 419 | |
7c6893e3 MS |
420 | /** |
421 | * mnt_want_write_file - get write access to a file's mount | |
422 | * @file: the file who's mount on which to take a write | |
423 | * | |
424 | * This is like mnt_want_write, but it takes a file and can | |
425 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
426 | */ |
427 | int mnt_want_write_file(struct file *file) | |
428 | { | |
429 | int ret; | |
430 | ||
a6795a58 | 431 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
432 | ret = __mnt_want_write_file(file); |
433 | if (ret) | |
a6795a58 | 434 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
435 | return ret; |
436 | } | |
96029c4e NP |
437 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
438 | ||
8366025e | 439 | /** |
eb04c282 | 440 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
441 | * @mnt: the mount on which to give up write access |
442 | * | |
443 | * Tells the low-level filesystem that we are done | |
444 | * performing writes to it. Must be matched with | |
eb04c282 | 445 | * __mnt_want_write() call above. |
8366025e | 446 | */ |
eb04c282 | 447 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 448 | { |
d3ef3d73 | 449 | preempt_disable(); |
83adc753 | 450 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 451 | preempt_enable(); |
8366025e | 452 | } |
eb04c282 JK |
453 | |
454 | /** | |
455 | * mnt_drop_write - give up write access to a mount | |
456 | * @mnt: the mount on which to give up write access | |
457 | * | |
458 | * Tells the low-level filesystem that we are done performing writes to it and | |
459 | * also allows filesystem to be frozen again. Must be matched with | |
460 | * mnt_want_write() call above. | |
461 | */ | |
462 | void mnt_drop_write(struct vfsmount *mnt) | |
463 | { | |
464 | __mnt_drop_write(mnt); | |
465 | sb_end_write(mnt->mnt_sb); | |
466 | } | |
8366025e DH |
467 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
468 | ||
eb04c282 JK |
469 | void __mnt_drop_write_file(struct file *file) |
470 | { | |
471 | __mnt_drop_write(file->f_path.mnt); | |
472 | } | |
473 | ||
7c6893e3 MS |
474 | void mnt_drop_write_file(struct file *file) |
475 | { | |
a6795a58 | 476 | __mnt_drop_write_file(file); |
7c6893e3 MS |
477 | sb_end_write(file_inode(file)->i_sb); |
478 | } | |
2a79f17e AV |
479 | EXPORT_SYMBOL(mnt_drop_write_file); |
480 | ||
fbdc2f6c | 481 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 482 | { |
83adc753 | 483 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 484 | /* |
d3ef3d73 NP |
485 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
486 | * should be visible before we do. | |
3d733633 | 487 | */ |
d3ef3d73 NP |
488 | smp_mb(); |
489 | ||
3d733633 | 490 | /* |
d3ef3d73 NP |
491 | * With writers on hold, if this value is zero, then there are |
492 | * definitely no active writers (although held writers may subsequently | |
493 | * increment the count, they'll have to wait, and decrement it after | |
494 | * seeing MNT_READONLY). | |
495 | * | |
496 | * It is OK to have counter incremented on one CPU and decremented on | |
497 | * another: the sum will add up correctly. The danger would be when we | |
498 | * sum up each counter, if we read a counter before it is incremented, | |
499 | * but then read another CPU's count which it has been subsequently | |
500 | * decremented from -- we would see more decrements than we should. | |
501 | * MNT_WRITE_HOLD protects against this scenario, because | |
502 | * mnt_want_write first increments count, then smp_mb, then spins on | |
503 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
504 | * we're counting up here. | |
3d733633 | 505 | */ |
c6653a83 | 506 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
507 | return -EBUSY; |
508 | ||
509 | return 0; | |
510 | } | |
511 | ||
512 | static inline void mnt_unhold_writers(struct mount *mnt) | |
513 | { | |
d3ef3d73 NP |
514 | /* |
515 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
516 | * that become unheld will see MNT_READONLY. | |
517 | */ | |
518 | smp_wmb(); | |
83adc753 | 519 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
520 | } |
521 | ||
522 | static int mnt_make_readonly(struct mount *mnt) | |
523 | { | |
524 | int ret; | |
525 | ||
526 | ret = mnt_hold_writers(mnt); | |
527 | if (!ret) | |
528 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
529 | mnt_unhold_writers(mnt); | |
3d733633 | 530 | return ret; |
8366025e | 531 | } |
8366025e | 532 | |
4ed5e82f MS |
533 | int sb_prepare_remount_readonly(struct super_block *sb) |
534 | { | |
535 | struct mount *mnt; | |
536 | int err = 0; | |
537 | ||
8e8b8796 MS |
538 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
539 | if (atomic_long_read(&sb->s_remove_count)) | |
540 | return -EBUSY; | |
541 | ||
719ea2fb | 542 | lock_mount_hash(); |
4ed5e82f MS |
543 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
544 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
545 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
546 | smp_mb(); | |
547 | if (mnt_get_writers(mnt) > 0) { | |
548 | err = -EBUSY; | |
549 | break; | |
550 | } | |
551 | } | |
552 | } | |
8e8b8796 MS |
553 | if (!err && atomic_long_read(&sb->s_remove_count)) |
554 | err = -EBUSY; | |
555 | ||
4ed5e82f MS |
556 | if (!err) { |
557 | sb->s_readonly_remount = 1; | |
558 | smp_wmb(); | |
559 | } | |
560 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
561 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
562 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
563 | } | |
719ea2fb | 564 | unlock_mount_hash(); |
4ed5e82f MS |
565 | |
566 | return err; | |
567 | } | |
568 | ||
b105e270 | 569 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 570 | { |
a6435940 CB |
571 | struct user_namespace *mnt_userns; |
572 | ||
573 | mnt_userns = mnt_user_ns(&mnt->mnt); | |
574 | if (mnt_userns != &init_user_ns) | |
575 | put_user_ns(mnt_userns); | |
fcc139ae | 576 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 577 | #ifdef CONFIG_SMP |
68e8a9fe | 578 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 579 | #endif |
b105e270 | 580 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
581 | } |
582 | ||
8ffcb32e DH |
583 | static void delayed_free_vfsmnt(struct rcu_head *head) |
584 | { | |
585 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
586 | } | |
587 | ||
48a066e7 | 588 | /* call under rcu_read_lock */ |
294d71ff | 589 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
590 | { |
591 | struct mount *mnt; | |
592 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 593 | return 1; |
48a066e7 | 594 | if (bastard == NULL) |
294d71ff | 595 | return 0; |
48a066e7 AV |
596 | mnt = real_mount(bastard); |
597 | mnt_add_count(mnt, 1); | |
119e1ef8 | 598 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 599 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 600 | return 0; |
48a066e7 AV |
601 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
602 | mnt_add_count(mnt, -1); | |
294d71ff AV |
603 | return 1; |
604 | } | |
119e1ef8 AV |
605 | lock_mount_hash(); |
606 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
607 | mnt_add_count(mnt, -1); | |
608 | unlock_mount_hash(); | |
609 | return 1; | |
610 | } | |
611 | unlock_mount_hash(); | |
612 | /* caller will mntput() */ | |
294d71ff AV |
613 | return -1; |
614 | } | |
615 | ||
616 | /* call under rcu_read_lock */ | |
617 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
618 | { | |
619 | int res = __legitimize_mnt(bastard, seq); | |
620 | if (likely(!res)) | |
621 | return true; | |
622 | if (unlikely(res < 0)) { | |
623 | rcu_read_unlock(); | |
624 | mntput(bastard); | |
625 | rcu_read_lock(); | |
48a066e7 | 626 | } |
48a066e7 AV |
627 | return false; |
628 | } | |
629 | ||
1da177e4 | 630 | /* |
474279dc | 631 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 632 | * call under rcu_read_lock() |
1da177e4 | 633 | */ |
474279dc | 634 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 635 | { |
38129a13 | 636 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
637 | struct mount *p; |
638 | ||
38129a13 | 639 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
640 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
641 | return p; | |
642 | return NULL; | |
643 | } | |
644 | ||
a05964f3 | 645 | /* |
f015f126 DH |
646 | * lookup_mnt - Return the first child mount mounted at path |
647 | * | |
648 | * "First" means first mounted chronologically. If you create the | |
649 | * following mounts: | |
650 | * | |
651 | * mount /dev/sda1 /mnt | |
652 | * mount /dev/sda2 /mnt | |
653 | * mount /dev/sda3 /mnt | |
654 | * | |
655 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
656 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
657 | * /dev/sda2, then /dev/sda3, then NULL. | |
658 | * | |
659 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 660 | */ |
ca71cf71 | 661 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 662 | { |
c7105365 | 663 | struct mount *child_mnt; |
48a066e7 AV |
664 | struct vfsmount *m; |
665 | unsigned seq; | |
99b7db7b | 666 | |
48a066e7 AV |
667 | rcu_read_lock(); |
668 | do { | |
669 | seq = read_seqbegin(&mount_lock); | |
670 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
671 | m = child_mnt ? &child_mnt->mnt : NULL; | |
672 | } while (!legitimize_mnt(m, seq)); | |
673 | rcu_read_unlock(); | |
674 | return m; | |
a05964f3 RP |
675 | } |
676 | ||
9f6c61f9 MS |
677 | static inline void lock_ns_list(struct mnt_namespace *ns) |
678 | { | |
679 | spin_lock(&ns->ns_lock); | |
680 | } | |
681 | ||
682 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
683 | { | |
684 | spin_unlock(&ns->ns_lock); | |
685 | } | |
686 | ||
687 | static inline bool mnt_is_cursor(struct mount *mnt) | |
688 | { | |
689 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
690 | } | |
691 | ||
7af1364f EB |
692 | /* |
693 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
694 | * current mount namespace. | |
695 | * | |
696 | * The common case is dentries are not mountpoints at all and that | |
697 | * test is handled inline. For the slow case when we are actually | |
698 | * dealing with a mountpoint of some kind, walk through all of the | |
699 | * mounts in the current mount namespace and test to see if the dentry | |
700 | * is a mountpoint. | |
701 | * | |
702 | * The mount_hashtable is not usable in the context because we | |
703 | * need to identify all mounts that may be in the current mount | |
704 | * namespace not just a mount that happens to have some specified | |
705 | * parent mount. | |
706 | */ | |
707 | bool __is_local_mountpoint(struct dentry *dentry) | |
708 | { | |
709 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
710 | struct mount *mnt; | |
711 | bool is_covered = false; | |
712 | ||
7af1364f | 713 | down_read(&namespace_sem); |
9f6c61f9 | 714 | lock_ns_list(ns); |
7af1364f | 715 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
716 | if (mnt_is_cursor(mnt)) |
717 | continue; | |
7af1364f EB |
718 | is_covered = (mnt->mnt_mountpoint == dentry); |
719 | if (is_covered) | |
720 | break; | |
721 | } | |
9f6c61f9 | 722 | unlock_ns_list(ns); |
7af1364f | 723 | up_read(&namespace_sem); |
5ad05cc8 | 724 | |
7af1364f EB |
725 | return is_covered; |
726 | } | |
727 | ||
e2dfa935 | 728 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 729 | { |
0818bf27 | 730 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
731 | struct mountpoint *mp; |
732 | ||
0818bf27 | 733 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 734 | if (mp->m_dentry == dentry) { |
84d17192 AV |
735 | mp->m_count++; |
736 | return mp; | |
737 | } | |
738 | } | |
e2dfa935 EB |
739 | return NULL; |
740 | } | |
741 | ||
3895dbf8 | 742 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 743 | { |
3895dbf8 | 744 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 745 | int ret; |
84d17192 | 746 | |
3895dbf8 | 747 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
748 | /* might be worth a WARN_ON() */ |
749 | if (d_unlinked(dentry)) | |
750 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
751 | mountpoint: |
752 | read_seqlock_excl(&mount_lock); | |
753 | mp = lookup_mountpoint(dentry); | |
754 | read_sequnlock_excl(&mount_lock); | |
755 | if (mp) | |
756 | goto done; | |
757 | } | |
758 | ||
759 | if (!new) | |
760 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
761 | if (!new) | |
84d17192 AV |
762 | return ERR_PTR(-ENOMEM); |
763 | ||
3895dbf8 EB |
764 | |
765 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 766 | ret = d_set_mounted(dentry); |
eed81007 | 767 | |
3895dbf8 EB |
768 | /* Someone else set d_mounted? */ |
769 | if (ret == -EBUSY) | |
770 | goto mountpoint; | |
771 | ||
772 | /* The dentry is not available as a mountpoint? */ | |
773 | mp = ERR_PTR(ret); | |
774 | if (ret) | |
775 | goto done; | |
776 | ||
777 | /* Add the new mountpoint to the hash table */ | |
778 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 779 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
780 | new->m_count = 1; |
781 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
782 | INIT_HLIST_HEAD(&new->m_list); | |
783 | read_sequnlock_excl(&mount_lock); | |
784 | ||
785 | mp = new; | |
786 | new = NULL; | |
787 | done: | |
788 | kfree(new); | |
84d17192 AV |
789 | return mp; |
790 | } | |
791 | ||
4edbe133 AV |
792 | /* |
793 | * vfsmount lock must be held. Additionally, the caller is responsible | |
794 | * for serializing calls for given disposal list. | |
795 | */ | |
796 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
797 | { |
798 | if (!--mp->m_count) { | |
799 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 800 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
801 | spin_lock(&dentry->d_lock); |
802 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
803 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 804 | dput_to_list(dentry, list); |
0818bf27 | 805 | hlist_del(&mp->m_hash); |
84d17192 AV |
806 | kfree(mp); |
807 | } | |
808 | } | |
809 | ||
4edbe133 AV |
810 | /* called with namespace_lock and vfsmount lock */ |
811 | static void put_mountpoint(struct mountpoint *mp) | |
812 | { | |
813 | __put_mountpoint(mp, &ex_mountpoints); | |
814 | } | |
815 | ||
143c8c91 | 816 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 817 | { |
6b3286ed | 818 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
819 | } |
820 | ||
99b7db7b NP |
821 | /* |
822 | * vfsmount lock must be held for write | |
823 | */ | |
6b3286ed | 824 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
825 | { |
826 | if (ns) { | |
827 | ns->event = ++event; | |
828 | wake_up_interruptible(&ns->poll); | |
829 | } | |
830 | } | |
831 | ||
99b7db7b NP |
832 | /* |
833 | * vfsmount lock must be held for write | |
834 | */ | |
6b3286ed | 835 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
836 | { |
837 | if (ns && ns->event != event) { | |
838 | ns->event = event; | |
839 | wake_up_interruptible(&ns->poll); | |
840 | } | |
841 | } | |
842 | ||
99b7db7b NP |
843 | /* |
844 | * vfsmount lock must be held for write | |
845 | */ | |
e4e59906 | 846 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 847 | { |
e4e59906 | 848 | struct mountpoint *mp; |
0714a533 | 849 | mnt->mnt_parent = mnt; |
a73324da | 850 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 851 | list_del_init(&mnt->mnt_child); |
38129a13 | 852 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 853 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 854 | mp = mnt->mnt_mp; |
84d17192 | 855 | mnt->mnt_mp = NULL; |
e4e59906 | 856 | return mp; |
7bdb11de EB |
857 | } |
858 | ||
6a46c573 EB |
859 | /* |
860 | * vfsmount lock must be held for write | |
861 | */ | |
862 | static void umount_mnt(struct mount *mnt) | |
863 | { | |
e4e59906 | 864 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
865 | } |
866 | ||
99b7db7b NP |
867 | /* |
868 | * vfsmount lock must be held for write | |
869 | */ | |
84d17192 AV |
870 | void mnt_set_mountpoint(struct mount *mnt, |
871 | struct mountpoint *mp, | |
44d964d6 | 872 | struct mount *child_mnt) |
b90fa9ae | 873 | { |
84d17192 | 874 | mp->m_count++; |
3a2393d7 | 875 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 876 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 877 | child_mnt->mnt_parent = mnt; |
84d17192 | 878 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 879 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
880 | } |
881 | ||
1064f874 EB |
882 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
883 | { | |
884 | hlist_add_head_rcu(&mnt->mnt_hash, | |
885 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
886 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
887 | } | |
888 | ||
99b7db7b NP |
889 | /* |
890 | * vfsmount lock must be held for write | |
891 | */ | |
84d17192 AV |
892 | static void attach_mnt(struct mount *mnt, |
893 | struct mount *parent, | |
894 | struct mountpoint *mp) | |
1da177e4 | 895 | { |
84d17192 | 896 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 897 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
898 | } |
899 | ||
1064f874 | 900 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 901 | { |
1064f874 | 902 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
903 | struct mount *old_parent = mnt->mnt_parent; |
904 | ||
905 | list_del_init(&mnt->mnt_child); | |
906 | hlist_del_init(&mnt->mnt_mp_list); | |
907 | hlist_del_init_rcu(&mnt->mnt_hash); | |
908 | ||
909 | attach_mnt(mnt, parent, mp); | |
910 | ||
911 | put_mountpoint(old_mp); | |
1064f874 | 912 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
913 | } |
914 | ||
b90fa9ae | 915 | /* |
99b7db7b | 916 | * vfsmount lock must be held for write |
b90fa9ae | 917 | */ |
1064f874 | 918 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 919 | { |
0714a533 | 920 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 921 | struct mount *m; |
b90fa9ae | 922 | LIST_HEAD(head); |
143c8c91 | 923 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 924 | |
0714a533 | 925 | BUG_ON(parent == mnt); |
b90fa9ae | 926 | |
1a4eeaf2 | 927 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 928 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 929 | m->mnt_ns = n; |
f03c6599 | 930 | |
b90fa9ae RP |
931 | list_splice(&head, n->list.prev); |
932 | ||
d2921684 EB |
933 | n->mounts += n->pending_mounts; |
934 | n->pending_mounts = 0; | |
935 | ||
1064f874 | 936 | __attach_mnt(mnt, parent); |
6b3286ed | 937 | touch_mnt_namespace(n); |
1da177e4 LT |
938 | } |
939 | ||
909b0a88 | 940 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 941 | { |
6b41d536 AV |
942 | struct list_head *next = p->mnt_mounts.next; |
943 | if (next == &p->mnt_mounts) { | |
1da177e4 | 944 | while (1) { |
909b0a88 | 945 | if (p == root) |
1da177e4 | 946 | return NULL; |
6b41d536 AV |
947 | next = p->mnt_child.next; |
948 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 949 | break; |
0714a533 | 950 | p = p->mnt_parent; |
1da177e4 LT |
951 | } |
952 | } | |
6b41d536 | 953 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
954 | } |
955 | ||
315fc83e | 956 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 957 | { |
6b41d536 AV |
958 | struct list_head *prev = p->mnt_mounts.prev; |
959 | while (prev != &p->mnt_mounts) { | |
960 | p = list_entry(prev, struct mount, mnt_child); | |
961 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
962 | } |
963 | return p; | |
964 | } | |
965 | ||
8f291889 AV |
966 | /** |
967 | * vfs_create_mount - Create a mount for a configured superblock | |
968 | * @fc: The configuration context with the superblock attached | |
969 | * | |
970 | * Create a mount to an already configured superblock. If necessary, the | |
971 | * caller should invoke vfs_get_tree() before calling this. | |
972 | * | |
973 | * Note that this does not attach the mount to anything. | |
974 | */ | |
975 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 976 | { |
b105e270 | 977 | struct mount *mnt; |
9d412a43 | 978 | |
8f291889 AV |
979 | if (!fc->root) |
980 | return ERR_PTR(-EINVAL); | |
9d412a43 | 981 | |
8f291889 | 982 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
983 | if (!mnt) |
984 | return ERR_PTR(-ENOMEM); | |
985 | ||
8f291889 | 986 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 987 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 988 | |
8f291889 AV |
989 | atomic_inc(&fc->root->d_sb->s_active); |
990 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
991 | mnt->mnt.mnt_root = dget(fc->root); | |
992 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
993 | mnt->mnt_parent = mnt; | |
9d412a43 | 994 | |
719ea2fb | 995 | lock_mount_hash(); |
8f291889 | 996 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 997 | unlock_mount_hash(); |
b105e270 | 998 | return &mnt->mnt; |
9d412a43 | 999 | } |
8f291889 AV |
1000 | EXPORT_SYMBOL(vfs_create_mount); |
1001 | ||
1002 | struct vfsmount *fc_mount(struct fs_context *fc) | |
1003 | { | |
1004 | int err = vfs_get_tree(fc); | |
1005 | if (!err) { | |
1006 | up_write(&fc->root->d_sb->s_umount); | |
1007 | return vfs_create_mount(fc); | |
1008 | } | |
1009 | return ERR_PTR(err); | |
1010 | } | |
1011 | EXPORT_SYMBOL(fc_mount); | |
1012 | ||
9bc61ab1 DH |
1013 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1014 | int flags, const char *name, | |
1015 | void *data) | |
9d412a43 | 1016 | { |
9bc61ab1 | 1017 | struct fs_context *fc; |
8f291889 | 1018 | struct vfsmount *mnt; |
9bc61ab1 | 1019 | int ret = 0; |
9d412a43 AV |
1020 | |
1021 | if (!type) | |
3e1aeb00 | 1022 | return ERR_PTR(-EINVAL); |
9d412a43 | 1023 | |
9bc61ab1 DH |
1024 | fc = fs_context_for_mount(type, flags); |
1025 | if (IS_ERR(fc)) | |
1026 | return ERR_CAST(fc); | |
1027 | ||
3e1aeb00 DH |
1028 | if (name) |
1029 | ret = vfs_parse_fs_string(fc, "source", | |
1030 | name, strlen(name)); | |
9bc61ab1 DH |
1031 | if (!ret) |
1032 | ret = parse_monolithic_mount_data(fc, data); | |
1033 | if (!ret) | |
8f291889 AV |
1034 | mnt = fc_mount(fc); |
1035 | else | |
1036 | mnt = ERR_PTR(ret); | |
9d412a43 | 1037 | |
9bc61ab1 | 1038 | put_fs_context(fc); |
8f291889 | 1039 | return mnt; |
9d412a43 AV |
1040 | } |
1041 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1042 | ||
93faccbb EB |
1043 | struct vfsmount * |
1044 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1045 | const char *name, void *data) | |
1046 | { | |
1047 | /* Until it is worked out how to pass the user namespace | |
1048 | * through from the parent mount to the submount don't support | |
1049 | * unprivileged mounts with submounts. | |
1050 | */ | |
1051 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1052 | return ERR_PTR(-EPERM); | |
1053 | ||
e462ec50 | 1054 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1055 | } |
1056 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1057 | ||
87129cc0 | 1058 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1059 | int flag) |
1da177e4 | 1060 | { |
87129cc0 | 1061 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1062 | struct mount *mnt; |
1063 | int err; | |
1da177e4 | 1064 | |
be34d1a3 DH |
1065 | mnt = alloc_vfsmnt(old->mnt_devname); |
1066 | if (!mnt) | |
1067 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1068 | |
7a472ef4 | 1069 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1070 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1071 | else | |
1072 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1073 | |
be34d1a3 DH |
1074 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1075 | err = mnt_alloc_group_id(mnt); | |
1076 | if (err) | |
1077 | goto out_free; | |
1da177e4 | 1078 | } |
be34d1a3 | 1079 | |
16a34adb AV |
1080 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1081 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1082 | |
be34d1a3 | 1083 | atomic_inc(&sb->s_active); |
a6435940 CB |
1084 | mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); |
1085 | if (mnt->mnt.mnt_userns != &init_user_ns) | |
1086 | mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); | |
be34d1a3 DH |
1087 | mnt->mnt.mnt_sb = sb; |
1088 | mnt->mnt.mnt_root = dget(root); | |
1089 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1090 | mnt->mnt_parent = mnt; | |
719ea2fb | 1091 | lock_mount_hash(); |
be34d1a3 | 1092 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1093 | unlock_mount_hash(); |
be34d1a3 | 1094 | |
7a472ef4 EB |
1095 | if ((flag & CL_SLAVE) || |
1096 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1097 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1098 | mnt->mnt_master = old; | |
1099 | CLEAR_MNT_SHARED(mnt); | |
1100 | } else if (!(flag & CL_PRIVATE)) { | |
1101 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1102 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1103 | if (IS_MNT_SLAVE(old)) | |
1104 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1105 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1106 | } else { |
1107 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1108 | } |
1109 | if (flag & CL_MAKE_SHARED) | |
1110 | set_mnt_shared(mnt); | |
1111 | ||
1112 | /* stick the duplicate mount on the same expiry list | |
1113 | * as the original if that was on one */ | |
1114 | if (flag & CL_EXPIRE) { | |
1115 | if (!list_empty(&old->mnt_expire)) | |
1116 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1117 | } | |
1118 | ||
cb338d06 | 1119 | return mnt; |
719f5d7f MS |
1120 | |
1121 | out_free: | |
8ffcb32e | 1122 | mnt_free_id(mnt); |
719f5d7f | 1123 | free_vfsmnt(mnt); |
be34d1a3 | 1124 | return ERR_PTR(err); |
1da177e4 LT |
1125 | } |
1126 | ||
9ea459e1 AV |
1127 | static void cleanup_mnt(struct mount *mnt) |
1128 | { | |
56cbb429 AV |
1129 | struct hlist_node *p; |
1130 | struct mount *m; | |
9ea459e1 | 1131 | /* |
56cbb429 AV |
1132 | * The warning here probably indicates that somebody messed |
1133 | * up a mnt_want/drop_write() pair. If this happens, the | |
1134 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1135 | * The locking used to deal with mnt_count decrement provides barriers, |
1136 | * so mnt_get_writers() below is safe. | |
1137 | */ | |
1138 | WARN_ON(mnt_get_writers(mnt)); | |
1139 | if (unlikely(mnt->mnt_pins.first)) | |
1140 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1141 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1142 | hlist_del(&m->mnt_umount); | |
1143 | mntput(&m->mnt); | |
1144 | } | |
9ea459e1 AV |
1145 | fsnotify_vfsmount_delete(&mnt->mnt); |
1146 | dput(mnt->mnt.mnt_root); | |
1147 | deactivate_super(mnt->mnt.mnt_sb); | |
1148 | mnt_free_id(mnt); | |
1149 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1150 | } | |
1151 | ||
1152 | static void __cleanup_mnt(struct rcu_head *head) | |
1153 | { | |
1154 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1155 | } | |
1156 | ||
1157 | static LLIST_HEAD(delayed_mntput_list); | |
1158 | static void delayed_mntput(struct work_struct *unused) | |
1159 | { | |
1160 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1161 | struct mount *m, *t; |
9ea459e1 | 1162 | |
29785735 BP |
1163 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1164 | cleanup_mnt(m); | |
9ea459e1 AV |
1165 | } |
1166 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1167 | ||
900148dc | 1168 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1169 | { |
4edbe133 | 1170 | LIST_HEAD(list); |
edf7ddbf | 1171 | int count; |
4edbe133 | 1172 | |
48a066e7 | 1173 | rcu_read_lock(); |
9ea0a46c AV |
1174 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1175 | /* | |
1176 | * Since we don't do lock_mount_hash() here, | |
1177 | * ->mnt_ns can change under us. However, if it's | |
1178 | * non-NULL, then there's a reference that won't | |
1179 | * be dropped until after an RCU delay done after | |
1180 | * turning ->mnt_ns NULL. So if we observe it | |
1181 | * non-NULL under rcu_read_lock(), the reference | |
1182 | * we are dropping is not the final one. | |
1183 | */ | |
1184 | mnt_add_count(mnt, -1); | |
48a066e7 | 1185 | rcu_read_unlock(); |
f03c6599 | 1186 | return; |
b3e19d92 | 1187 | } |
719ea2fb | 1188 | lock_mount_hash(); |
119e1ef8 AV |
1189 | /* |
1190 | * make sure that if __legitimize_mnt() has not seen us grab | |
1191 | * mount_lock, we'll see their refcount increment here. | |
1192 | */ | |
1193 | smp_mb(); | |
9ea0a46c | 1194 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1195 | count = mnt_get_count(mnt); |
1196 | if (count != 0) { | |
1197 | WARN_ON(count < 0); | |
48a066e7 | 1198 | rcu_read_unlock(); |
719ea2fb | 1199 | unlock_mount_hash(); |
99b7db7b NP |
1200 | return; |
1201 | } | |
48a066e7 AV |
1202 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1203 | rcu_read_unlock(); | |
1204 | unlock_mount_hash(); | |
1205 | return; | |
1206 | } | |
1207 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1208 | rcu_read_unlock(); | |
962830df | 1209 | |
39f7c4db | 1210 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1211 | |
1212 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1213 | struct mount *p, *tmp; | |
1214 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1215 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1216 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1217 | } |
1218 | } | |
719ea2fb | 1219 | unlock_mount_hash(); |
4edbe133 | 1220 | shrink_dentry_list(&list); |
649a795a | 1221 | |
9ea459e1 AV |
1222 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1223 | struct task_struct *task = current; | |
1224 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1225 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1226 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1227 | return; |
1228 | } | |
1229 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1230 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1231 | return; | |
1232 | } | |
1233 | cleanup_mnt(mnt); | |
b3e19d92 | 1234 | } |
b3e19d92 NP |
1235 | |
1236 | void mntput(struct vfsmount *mnt) | |
1237 | { | |
1238 | if (mnt) { | |
863d684f | 1239 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1240 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1241 | if (unlikely(m->mnt_expiry_mark)) |
1242 | m->mnt_expiry_mark = 0; | |
1243 | mntput_no_expire(m); | |
b3e19d92 NP |
1244 | } |
1245 | } | |
1246 | EXPORT_SYMBOL(mntput); | |
1247 | ||
1248 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1249 | { | |
1250 | if (mnt) | |
83adc753 | 1251 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1252 | return mnt; |
1253 | } | |
1254 | EXPORT_SYMBOL(mntget); | |
1255 | ||
c6609c0a IK |
1256 | /* path_is_mountpoint() - Check if path is a mount in the current |
1257 | * namespace. | |
1258 | * | |
1259 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1260 | * not mounted in any namespace and that common case is handled inline. | |
1261 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1262 | * mounts using a given dentry in a different namespace. This function | |
1263 | * checks if the passed in path is a mountpoint rather than the dentry | |
1264 | * alone. | |
1265 | */ | |
1266 | bool path_is_mountpoint(const struct path *path) | |
1267 | { | |
1268 | unsigned seq; | |
1269 | bool res; | |
1270 | ||
1271 | if (!d_mountpoint(path->dentry)) | |
1272 | return false; | |
1273 | ||
1274 | rcu_read_lock(); | |
1275 | do { | |
1276 | seq = read_seqbegin(&mount_lock); | |
1277 | res = __path_is_mountpoint(path); | |
1278 | } while (read_seqretry(&mount_lock, seq)); | |
1279 | rcu_read_unlock(); | |
1280 | ||
1281 | return res; | |
1282 | } | |
1283 | EXPORT_SYMBOL(path_is_mountpoint); | |
1284 | ||
ca71cf71 | 1285 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1286 | { |
3064c356 AV |
1287 | struct mount *p; |
1288 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1289 | if (IS_ERR(p)) | |
1290 | return ERR_CAST(p); | |
1291 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1292 | return &p->mnt; | |
7b7b1ace | 1293 | } |
1da177e4 | 1294 | |
a1a2c409 | 1295 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1296 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1297 | struct list_head *p) | |
1298 | { | |
1299 | struct mount *mnt, *ret = NULL; | |
1300 | ||
1301 | lock_ns_list(ns); | |
1302 | list_for_each_continue(p, &ns->list) { | |
1303 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1304 | if (!mnt_is_cursor(mnt)) { | |
1305 | ret = mnt; | |
1306 | break; | |
1307 | } | |
1308 | } | |
1309 | unlock_ns_list(ns); | |
1310 | ||
1311 | return ret; | |
1312 | } | |
1313 | ||
0226f492 | 1314 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1315 | static void *m_start(struct seq_file *m, loff_t *pos) |
1316 | { | |
ede1bf0d | 1317 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1318 | struct list_head *prev; |
1da177e4 | 1319 | |
390c6843 | 1320 | down_read(&namespace_sem); |
9f6c61f9 MS |
1321 | if (!*pos) { |
1322 | prev = &p->ns->list; | |
1323 | } else { | |
1324 | prev = &p->cursor.mnt_list; | |
1325 | ||
1326 | /* Read after we'd reached the end? */ | |
1327 | if (list_empty(prev)) | |
1328 | return NULL; | |
c7999c36 AV |
1329 | } |
1330 | ||
9f6c61f9 | 1331 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1332 | } |
1333 | ||
1334 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1335 | { | |
ede1bf0d | 1336 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1337 | struct mount *mnt = v; |
b0765fb8 | 1338 | |
9f6c61f9 MS |
1339 | ++*pos; |
1340 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | static void m_stop(struct seq_file *m, void *v) | |
1344 | { | |
9f6c61f9 MS |
1345 | struct proc_mounts *p = m->private; |
1346 | struct mount *mnt = v; | |
1347 | ||
1348 | lock_ns_list(p->ns); | |
1349 | if (mnt) | |
1350 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1351 | else | |
1352 | list_del_init(&p->cursor.mnt_list); | |
1353 | unlock_ns_list(p->ns); | |
390c6843 | 1354 | up_read(&namespace_sem); |
1da177e4 LT |
1355 | } |
1356 | ||
0226f492 | 1357 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1358 | { |
ede1bf0d | 1359 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1360 | struct mount *r = v; |
0226f492 | 1361 | return p->show(m, &r->mnt); |
1da177e4 LT |
1362 | } |
1363 | ||
a1a2c409 | 1364 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1365 | .start = m_start, |
1366 | .next = m_next, | |
1367 | .stop = m_stop, | |
0226f492 | 1368 | .show = m_show, |
b4629fe2 | 1369 | }; |
9f6c61f9 MS |
1370 | |
1371 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1372 | { | |
1373 | down_read(&namespace_sem); | |
1374 | lock_ns_list(ns); | |
1375 | list_del(&cursor->mnt_list); | |
1376 | unlock_ns_list(ns); | |
1377 | up_read(&namespace_sem); | |
1378 | } | |
a1a2c409 | 1379 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1380 | |
1da177e4 LT |
1381 | /** |
1382 | * may_umount_tree - check if a mount tree is busy | |
1383 | * @mnt: root of mount tree | |
1384 | * | |
1385 | * This is called to check if a tree of mounts has any | |
1386 | * open files, pwds, chroots or sub mounts that are | |
1387 | * busy. | |
1388 | */ | |
909b0a88 | 1389 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1390 | { |
909b0a88 | 1391 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1392 | int actual_refs = 0; |
1393 | int minimum_refs = 0; | |
315fc83e | 1394 | struct mount *p; |
909b0a88 | 1395 | BUG_ON(!m); |
1da177e4 | 1396 | |
b3e19d92 | 1397 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1398 | lock_mount_hash(); |
909b0a88 | 1399 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1400 | actual_refs += mnt_get_count(p); |
1da177e4 | 1401 | minimum_refs += 2; |
1da177e4 | 1402 | } |
719ea2fb | 1403 | unlock_mount_hash(); |
1da177e4 LT |
1404 | |
1405 | if (actual_refs > minimum_refs) | |
e3474a8e | 1406 | return 0; |
1da177e4 | 1407 | |
e3474a8e | 1408 | return 1; |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | EXPORT_SYMBOL(may_umount_tree); | |
1412 | ||
1413 | /** | |
1414 | * may_umount - check if a mount point is busy | |
1415 | * @mnt: root of mount | |
1416 | * | |
1417 | * This is called to check if a mount point has any | |
1418 | * open files, pwds, chroots or sub mounts. If the | |
1419 | * mount has sub mounts this will return busy | |
1420 | * regardless of whether the sub mounts are busy. | |
1421 | * | |
1422 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1423 | * give false negatives. The main reason why it's here is that we need | |
1424 | * a non-destructive way to look for easily umountable filesystems. | |
1425 | */ | |
1426 | int may_umount(struct vfsmount *mnt) | |
1427 | { | |
e3474a8e | 1428 | int ret = 1; |
8ad08d8a | 1429 | down_read(&namespace_sem); |
719ea2fb | 1430 | lock_mount_hash(); |
1ab59738 | 1431 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1432 | ret = 0; |
719ea2fb | 1433 | unlock_mount_hash(); |
8ad08d8a | 1434 | up_read(&namespace_sem); |
a05964f3 | 1435 | return ret; |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | EXPORT_SYMBOL(may_umount); | |
1439 | ||
97216be0 | 1440 | static void namespace_unlock(void) |
70fbcdf4 | 1441 | { |
a3b3c562 | 1442 | struct hlist_head head; |
56cbb429 AV |
1443 | struct hlist_node *p; |
1444 | struct mount *m; | |
4edbe133 | 1445 | LIST_HEAD(list); |
97216be0 | 1446 | |
a3b3c562 | 1447 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1448 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1449 | |
97216be0 AV |
1450 | up_write(&namespace_sem); |
1451 | ||
4edbe133 AV |
1452 | shrink_dentry_list(&list); |
1453 | ||
a3b3c562 EB |
1454 | if (likely(hlist_empty(&head))) |
1455 | return; | |
1456 | ||
22cb7405 | 1457 | synchronize_rcu_expedited(); |
48a066e7 | 1458 | |
56cbb429 AV |
1459 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1460 | hlist_del(&m->mnt_umount); | |
1461 | mntput(&m->mnt); | |
1462 | } | |
70fbcdf4 RP |
1463 | } |
1464 | ||
97216be0 | 1465 | static inline void namespace_lock(void) |
e3197d83 | 1466 | { |
97216be0 | 1467 | down_write(&namespace_sem); |
e3197d83 AV |
1468 | } |
1469 | ||
e819f152 EB |
1470 | enum umount_tree_flags { |
1471 | UMOUNT_SYNC = 1, | |
1472 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1473 | UMOUNT_CONNECTED = 4, |
e819f152 | 1474 | }; |
f2d0a123 EB |
1475 | |
1476 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1477 | { | |
1478 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1479 | if (how & UMOUNT_SYNC) | |
1480 | return true; | |
1481 | ||
1482 | /* A mount without a parent has nothing to be connected to */ | |
1483 | if (!mnt_has_parent(mnt)) | |
1484 | return true; | |
1485 | ||
1486 | /* Because the reference counting rules change when mounts are | |
1487 | * unmounted and connected, umounted mounts may not be | |
1488 | * connected to mounted mounts. | |
1489 | */ | |
1490 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1491 | return true; | |
1492 | ||
1493 | /* Has it been requested that the mount remain connected? */ | |
1494 | if (how & UMOUNT_CONNECTED) | |
1495 | return false; | |
1496 | ||
1497 | /* Is the mount locked such that it needs to remain connected? */ | |
1498 | if (IS_MNT_LOCKED(mnt)) | |
1499 | return false; | |
1500 | ||
1501 | /* By default disconnect the mount */ | |
1502 | return true; | |
1503 | } | |
1504 | ||
99b7db7b | 1505 | /* |
48a066e7 | 1506 | * mount_lock must be held |
99b7db7b NP |
1507 | * namespace_sem must be held for write |
1508 | */ | |
e819f152 | 1509 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1510 | { |
c003b26f | 1511 | LIST_HEAD(tmp_list); |
315fc83e | 1512 | struct mount *p; |
1da177e4 | 1513 | |
5d88457e EB |
1514 | if (how & UMOUNT_PROPAGATE) |
1515 | propagate_mount_unlock(mnt); | |
1516 | ||
c003b26f | 1517 | /* Gather the mounts to umount */ |
590ce4bc EB |
1518 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1519 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1520 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1521 | } |
1da177e4 | 1522 | |
411a938b | 1523 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1524 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1525 | list_del_init(&p->mnt_child); |
c003b26f | 1526 | } |
88b368f2 | 1527 | |
c003b26f | 1528 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1529 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1530 | propagate_umount(&tmp_list); |
a05964f3 | 1531 | |
c003b26f | 1532 | while (!list_empty(&tmp_list)) { |
d2921684 | 1533 | struct mnt_namespace *ns; |
ce07d891 | 1534 | bool disconnect; |
c003b26f | 1535 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1536 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1537 | list_del_init(&p->mnt_list); |
d2921684 EB |
1538 | ns = p->mnt_ns; |
1539 | if (ns) { | |
1540 | ns->mounts--; | |
1541 | __touch_mnt_namespace(ns); | |
1542 | } | |
143c8c91 | 1543 | p->mnt_ns = NULL; |
e819f152 | 1544 | if (how & UMOUNT_SYNC) |
48a066e7 | 1545 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1546 | |
f2d0a123 | 1547 | disconnect = disconnect_mount(p, how); |
676da58d | 1548 | if (mnt_has_parent(p)) { |
81b6b061 | 1549 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1550 | if (!disconnect) { |
1551 | /* Don't forget about p */ | |
1552 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1553 | } else { | |
1554 | umount_mnt(p); | |
1555 | } | |
7c4b93d8 | 1556 | } |
0f0afb1d | 1557 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1558 | if (disconnect) |
1559 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1560 | } |
1561 | } | |
1562 | ||
b54b9be7 | 1563 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1564 | |
8d0347f6 DH |
1565 | static int do_umount_root(struct super_block *sb) |
1566 | { | |
1567 | int ret = 0; | |
1568 | ||
1569 | down_write(&sb->s_umount); | |
1570 | if (!sb_rdonly(sb)) { | |
1571 | struct fs_context *fc; | |
1572 | ||
1573 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1574 | SB_RDONLY); | |
1575 | if (IS_ERR(fc)) { | |
1576 | ret = PTR_ERR(fc); | |
1577 | } else { | |
1578 | ret = parse_monolithic_mount_data(fc, NULL); | |
1579 | if (!ret) | |
1580 | ret = reconfigure_super(fc); | |
1581 | put_fs_context(fc); | |
1582 | } | |
1583 | } | |
1584 | up_write(&sb->s_umount); | |
1585 | return ret; | |
1586 | } | |
1587 | ||
1ab59738 | 1588 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1589 | { |
1ab59738 | 1590 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1591 | int retval; |
1592 | ||
1ab59738 | 1593 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1594 | if (retval) |
1595 | return retval; | |
1596 | ||
1597 | /* | |
1598 | * Allow userspace to request a mountpoint be expired rather than | |
1599 | * unmounting unconditionally. Unmount only happens if: | |
1600 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1601 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1602 | */ | |
1603 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1604 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1605 | flags & (MNT_FORCE | MNT_DETACH)) |
1606 | return -EINVAL; | |
1607 | ||
b3e19d92 NP |
1608 | /* |
1609 | * probably don't strictly need the lock here if we examined | |
1610 | * all race cases, but it's a slowpath. | |
1611 | */ | |
719ea2fb | 1612 | lock_mount_hash(); |
83adc753 | 1613 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1614 | unlock_mount_hash(); |
1da177e4 | 1615 | return -EBUSY; |
b3e19d92 | 1616 | } |
719ea2fb | 1617 | unlock_mount_hash(); |
1da177e4 | 1618 | |
863d684f | 1619 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1620 | return -EAGAIN; |
1621 | } | |
1622 | ||
1623 | /* | |
1624 | * If we may have to abort operations to get out of this | |
1625 | * mount, and they will themselves hold resources we must | |
1626 | * allow the fs to do things. In the Unix tradition of | |
1627 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1628 | * might fail to complete on the first run through as other tasks | |
1629 | * must return, and the like. Thats for the mount program to worry | |
1630 | * about for the moment. | |
1631 | */ | |
1632 | ||
42faad99 | 1633 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1634 | sb->s_op->umount_begin(sb); |
42faad99 | 1635 | } |
1da177e4 LT |
1636 | |
1637 | /* | |
1638 | * No sense to grab the lock for this test, but test itself looks | |
1639 | * somewhat bogus. Suggestions for better replacement? | |
1640 | * Ho-hum... In principle, we might treat that as umount + switch | |
1641 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1642 | * Actually it makes sense, especially if rootfs would contain a | |
1643 | * /reboot - static binary that would close all descriptors and | |
1644 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1645 | */ | |
1ab59738 | 1646 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1647 | /* |
1648 | * Special case for "unmounting" root ... | |
1649 | * we just try to remount it readonly. | |
1650 | */ | |
bc6155d1 | 1651 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1652 | return -EPERM; |
8d0347f6 | 1653 | return do_umount_root(sb); |
1da177e4 LT |
1654 | } |
1655 | ||
97216be0 | 1656 | namespace_lock(); |
719ea2fb | 1657 | lock_mount_hash(); |
1da177e4 | 1658 | |
25d202ed EB |
1659 | /* Recheck MNT_LOCKED with the locks held */ |
1660 | retval = -EINVAL; | |
1661 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1662 | goto out; | |
1663 | ||
1664 | event++; | |
48a066e7 | 1665 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1666 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1667 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1668 | retval = 0; |
48a066e7 AV |
1669 | } else { |
1670 | shrink_submounts(mnt); | |
1671 | retval = -EBUSY; | |
1672 | if (!propagate_mount_busy(mnt, 2)) { | |
1673 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1674 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1675 | retval = 0; |
1676 | } | |
1da177e4 | 1677 | } |
25d202ed | 1678 | out: |
719ea2fb | 1679 | unlock_mount_hash(); |
e3197d83 | 1680 | namespace_unlock(); |
1da177e4 LT |
1681 | return retval; |
1682 | } | |
1683 | ||
80b5dce8 EB |
1684 | /* |
1685 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1686 | * | |
1687 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1688 | * to an existing mountpoint, and wind up leaking the mount. | |
1689 | * detach_mounts allows lazily unmounting those mounts instead of | |
1690 | * leaking them. | |
1691 | * | |
1692 | * The caller may hold dentry->d_inode->i_mutex. | |
1693 | */ | |
1694 | void __detach_mounts(struct dentry *dentry) | |
1695 | { | |
1696 | struct mountpoint *mp; | |
1697 | struct mount *mnt; | |
1698 | ||
1699 | namespace_lock(); | |
3895dbf8 | 1700 | lock_mount_hash(); |
80b5dce8 | 1701 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1702 | if (!mp) |
80b5dce8 EB |
1703 | goto out_unlock; |
1704 | ||
e06b933e | 1705 | event++; |
80b5dce8 EB |
1706 | while (!hlist_empty(&mp->m_list)) { |
1707 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1708 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1709 | umount_mnt(mnt); |
56cbb429 | 1710 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1711 | } |
e0c9c0af | 1712 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1713 | } |
80b5dce8 EB |
1714 | put_mountpoint(mp); |
1715 | out_unlock: | |
3895dbf8 | 1716 | unlock_mount_hash(); |
80b5dce8 EB |
1717 | namespace_unlock(); |
1718 | } | |
1719 | ||
dd111b31 | 1720 | /* |
9b40bc90 AV |
1721 | * Is the caller allowed to modify his namespace? |
1722 | */ | |
1723 | static inline bool may_mount(void) | |
1724 | { | |
1725 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1726 | } | |
1727 | ||
df2474a2 | 1728 | #ifdef CONFIG_MANDATORY_FILE_LOCKING |
9e8925b6 JL |
1729 | static inline bool may_mandlock(void) |
1730 | { | |
95ace754 | 1731 | return capable(CAP_SYS_ADMIN); |
9e8925b6 | 1732 | } |
df2474a2 JL |
1733 | #else |
1734 | static inline bool may_mandlock(void) | |
1735 | { | |
1736 | pr_warn("VFS: \"mand\" mount option not supported"); | |
1737 | return false; | |
1738 | } | |
1739 | #endif | |
9e8925b6 | 1740 | |
25ccd24f | 1741 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1742 | { |
25ccd24f | 1743 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1744 | |
9b40bc90 AV |
1745 | if (!may_mount()) |
1746 | return -EPERM; | |
41525f56 | 1747 | if (path->dentry != path->mnt->mnt_root) |
25ccd24f | 1748 | return -EINVAL; |
143c8c91 | 1749 | if (!check_mnt(mnt)) |
25ccd24f | 1750 | return -EINVAL; |
25d202ed | 1751 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1752 | return -EINVAL; |
b2f5d4dc | 1753 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1754 | return -EPERM; |
1755 | return 0; | |
1756 | } | |
1757 | ||
a0a6df9a | 1758 | // caller is responsible for flags being sane |
25ccd24f CH |
1759 | int path_umount(struct path *path, int flags) |
1760 | { | |
1761 | struct mount *mnt = real_mount(path->mnt); | |
1762 | int ret; | |
1763 | ||
1764 | ret = can_umount(path, flags); | |
1765 | if (!ret) | |
1766 | ret = do_umount(mnt, flags); | |
1da177e4 | 1767 | |
429731b1 | 1768 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1769 | dput(path->dentry); |
900148dc | 1770 | mntput_no_expire(mnt); |
25ccd24f | 1771 | return ret; |
1da177e4 LT |
1772 | } |
1773 | ||
09267def | 1774 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1775 | { |
1776 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1777 | struct path path; | |
1778 | int ret; | |
1779 | ||
a0a6df9a AV |
1780 | // basic validity checks done first |
1781 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1782 | return -EINVAL; | |
1783 | ||
41525f56 CH |
1784 | if (!(flags & UMOUNT_NOFOLLOW)) |
1785 | lookup_flags |= LOOKUP_FOLLOW; | |
1786 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1787 | if (ret) | |
1788 | return ret; | |
1789 | return path_umount(&path, flags); | |
1790 | } | |
1791 | ||
3a18ef5c DB |
1792 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1793 | { | |
1794 | return ksys_umount(name, flags); | |
1795 | } | |
1796 | ||
1da177e4 LT |
1797 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1798 | ||
1799 | /* | |
b58fed8b | 1800 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1801 | */ |
bdc480e3 | 1802 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1803 | { |
3a18ef5c | 1804 | return ksys_umount(name, 0); |
1da177e4 LT |
1805 | } |
1806 | ||
1807 | #endif | |
1808 | ||
4ce5d2b1 | 1809 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1810 | { |
4ce5d2b1 | 1811 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1812 | return dentry->d_op == &ns_dentry_operations && |
1813 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1814 | } |
1815 | ||
213921f9 | 1816 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1817 | { |
1818 | return container_of(ns, struct mnt_namespace, ns); | |
1819 | } | |
1820 | ||
303cc571 CB |
1821 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1822 | { | |
1823 | return &mnt->ns; | |
1824 | } | |
1825 | ||
4ce5d2b1 EB |
1826 | static bool mnt_ns_loop(struct dentry *dentry) |
1827 | { | |
1828 | /* Could bind mounting the mount namespace inode cause a | |
1829 | * mount namespace loop? | |
1830 | */ | |
1831 | struct mnt_namespace *mnt_ns; | |
1832 | if (!is_mnt_ns_file(dentry)) | |
1833 | return false; | |
1834 | ||
f77c8014 | 1835 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1836 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1837 | } | |
1838 | ||
87129cc0 | 1839 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1840 | int flag) |
1da177e4 | 1841 | { |
84d17192 | 1842 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1843 | |
4ce5d2b1 EB |
1844 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1845 | return ERR_PTR(-EINVAL); | |
1846 | ||
1847 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1848 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1849 | |
36341f64 | 1850 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1851 | if (IS_ERR(q)) |
1852 | return q; | |
1853 | ||
a73324da | 1854 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1855 | |
1856 | p = mnt; | |
6b41d536 | 1857 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1858 | struct mount *s; |
7ec02ef1 | 1859 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1860 | continue; |
1861 | ||
909b0a88 | 1862 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1863 | if (!(flag & CL_COPY_UNBINDABLE) && |
1864 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1865 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1866 | /* Both unbindable and locked. */ | |
1867 | q = ERR_PTR(-EPERM); | |
1868 | goto out; | |
1869 | } else { | |
1870 | s = skip_mnt_tree(s); | |
1871 | continue; | |
1872 | } | |
4ce5d2b1 EB |
1873 | } |
1874 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1875 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1876 | s = skip_mnt_tree(s); |
1877 | continue; | |
1878 | } | |
0714a533 AV |
1879 | while (p != s->mnt_parent) { |
1880 | p = p->mnt_parent; | |
1881 | q = q->mnt_parent; | |
1da177e4 | 1882 | } |
87129cc0 | 1883 | p = s; |
84d17192 | 1884 | parent = q; |
87129cc0 | 1885 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1886 | if (IS_ERR(q)) |
1887 | goto out; | |
719ea2fb | 1888 | lock_mount_hash(); |
1a4eeaf2 | 1889 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1890 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1891 | unlock_mount_hash(); |
1da177e4 LT |
1892 | } |
1893 | } | |
1894 | return res; | |
be34d1a3 | 1895 | out: |
1da177e4 | 1896 | if (res) { |
719ea2fb | 1897 | lock_mount_hash(); |
e819f152 | 1898 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1899 | unlock_mount_hash(); |
1da177e4 | 1900 | } |
be34d1a3 | 1901 | return q; |
1da177e4 LT |
1902 | } |
1903 | ||
be34d1a3 DH |
1904 | /* Caller should check returned pointer for errors */ |
1905 | ||
ca71cf71 | 1906 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1907 | { |
cb338d06 | 1908 | struct mount *tree; |
97216be0 | 1909 | namespace_lock(); |
cd4a4017 EB |
1910 | if (!check_mnt(real_mount(path->mnt))) |
1911 | tree = ERR_PTR(-EINVAL); | |
1912 | else | |
1913 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1914 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1915 | namespace_unlock(); |
be34d1a3 | 1916 | if (IS_ERR(tree)) |
52e220d3 | 1917 | return ERR_CAST(tree); |
be34d1a3 | 1918 | return &tree->mnt; |
8aec0809 AV |
1919 | } |
1920 | ||
a07b2000 AV |
1921 | static void free_mnt_ns(struct mnt_namespace *); |
1922 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1923 | ||
1924 | void dissolve_on_fput(struct vfsmount *mnt) | |
1925 | { | |
1926 | struct mnt_namespace *ns; | |
1927 | namespace_lock(); | |
1928 | lock_mount_hash(); | |
1929 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1930 | if (ns) { |
1931 | if (is_anon_ns(ns)) | |
1932 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1933 | else | |
1934 | ns = NULL; | |
1935 | } | |
a07b2000 AV |
1936 | unlock_mount_hash(); |
1937 | namespace_unlock(); | |
44dfd84a DH |
1938 | if (ns) |
1939 | free_mnt_ns(ns); | |
a07b2000 AV |
1940 | } |
1941 | ||
8aec0809 AV |
1942 | void drop_collected_mounts(struct vfsmount *mnt) |
1943 | { | |
97216be0 | 1944 | namespace_lock(); |
719ea2fb | 1945 | lock_mount_hash(); |
9c8e0a1b | 1946 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1947 | unlock_mount_hash(); |
3ab6abee | 1948 | namespace_unlock(); |
8aec0809 AV |
1949 | } |
1950 | ||
c771d683 MS |
1951 | /** |
1952 | * clone_private_mount - create a private clone of a path | |
1953 | * | |
1954 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1955 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1956 | * to the originating mount won't be propagated into this). | |
1957 | * | |
1958 | * Release with mntput(). | |
1959 | */ | |
ca71cf71 | 1960 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1961 | { |
1962 | struct mount *old_mnt = real_mount(path->mnt); | |
1963 | struct mount *new_mnt; | |
1964 | ||
1965 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1966 | return ERR_PTR(-EINVAL); | |
1967 | ||
c771d683 | 1968 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1969 | if (IS_ERR(new_mnt)) |
1970 | return ERR_CAST(new_mnt); | |
1971 | ||
df820f8d MS |
1972 | /* Longterm mount to be removed by kern_unmount*() */ |
1973 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
1974 | ||
c771d683 MS |
1975 | return &new_mnt->mnt; |
1976 | } | |
1977 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1978 | ||
1f707137 AV |
1979 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1980 | struct vfsmount *root) | |
1981 | { | |
1a4eeaf2 | 1982 | struct mount *mnt; |
1f707137 AV |
1983 | int res = f(root, arg); |
1984 | if (res) | |
1985 | return res; | |
1a4eeaf2 AV |
1986 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1987 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1988 | if (res) |
1989 | return res; | |
1990 | } | |
1991 | return 0; | |
1992 | } | |
1993 | ||
3bd045cc AV |
1994 | static void lock_mnt_tree(struct mount *mnt) |
1995 | { | |
1996 | struct mount *p; | |
1997 | ||
1998 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1999 | int flags = p->mnt.mnt_flags; | |
2000 | /* Don't allow unprivileged users to change mount flags */ | |
2001 | flags |= MNT_LOCK_ATIME; | |
2002 | ||
2003 | if (flags & MNT_READONLY) | |
2004 | flags |= MNT_LOCK_READONLY; | |
2005 | ||
2006 | if (flags & MNT_NODEV) | |
2007 | flags |= MNT_LOCK_NODEV; | |
2008 | ||
2009 | if (flags & MNT_NOSUID) | |
2010 | flags |= MNT_LOCK_NOSUID; | |
2011 | ||
2012 | if (flags & MNT_NOEXEC) | |
2013 | flags |= MNT_LOCK_NOEXEC; | |
2014 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2015 | if (list_empty(&p->mnt_expire)) | |
2016 | flags |= MNT_LOCKED; | |
2017 | p->mnt.mnt_flags = flags; | |
2018 | } | |
2019 | } | |
2020 | ||
4b8b21f4 | 2021 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2022 | { |
315fc83e | 2023 | struct mount *p; |
719f5d7f | 2024 | |
909b0a88 | 2025 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2026 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2027 | mnt_release_group_id(p); |
719f5d7f MS |
2028 | } |
2029 | } | |
2030 | ||
4b8b21f4 | 2031 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2032 | { |
315fc83e | 2033 | struct mount *p; |
719f5d7f | 2034 | |
909b0a88 | 2035 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2036 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2037 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2038 | if (err) { |
4b8b21f4 | 2039 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2040 | return err; |
2041 | } | |
2042 | } | |
2043 | } | |
2044 | ||
2045 | return 0; | |
2046 | } | |
2047 | ||
d2921684 EB |
2048 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2049 | { | |
2050 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
2051 | unsigned int mounts = 0, old, pending, sum; | |
2052 | struct mount *p; | |
2053 | ||
2054 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
2055 | mounts++; | |
2056 | ||
2057 | old = ns->mounts; | |
2058 | pending = ns->pending_mounts; | |
2059 | sum = old + pending; | |
2060 | if ((old > sum) || | |
2061 | (pending > sum) || | |
2062 | (max < sum) || | |
2063 | (mounts > (max - sum))) | |
2064 | return -ENOSPC; | |
2065 | ||
2066 | ns->pending_mounts = pending + mounts; | |
2067 | return 0; | |
2068 | } | |
2069 | ||
b90fa9ae RP |
2070 | /* |
2071 | * @source_mnt : mount tree to be attached | |
21444403 RP |
2072 | * @nd : place the mount tree @source_mnt is attached |
2073 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
2074 | * store the parent mount and mountpoint dentry. | |
2075 | * (done when source_mnt is moved) | |
b90fa9ae RP |
2076 | * |
2077 | * NOTE: in the table below explains the semantics when a source mount | |
2078 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2079 | * --------------------------------------------------------------------------- |
2080 | * | BIND MOUNT OPERATION | | |
2081 | * |************************************************************************** | |
2082 | * | source-->| shared | private | slave | unbindable | | |
2083 | * | dest | | | | | | |
2084 | * | | | | | | | | |
2085 | * | v | | | | | | |
2086 | * |************************************************************************** | |
2087 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2088 | * | | | | | | | |
2089 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2090 | * *************************************************************************** | |
b90fa9ae RP |
2091 | * A bind operation clones the source mount and mounts the clone on the |
2092 | * destination mount. | |
2093 | * | |
2094 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2095 | * tree of the destination mount and the cloned mount is added to | |
2096 | * the peer group of the source mount. | |
2097 | * (+) the cloned mount is created under the destination mount and is marked | |
2098 | * as shared. The cloned mount is added to the peer group of the source | |
2099 | * mount. | |
5afe0022 RP |
2100 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2101 | * of the destination mount and the cloned mount is made slave | |
2102 | * of the same master as that of the source mount. The cloned mount | |
2103 | * is marked as 'shared and slave'. | |
2104 | * (*) the cloned mount is made a slave of the same master as that of the | |
2105 | * source mount. | |
2106 | * | |
9676f0c6 RP |
2107 | * --------------------------------------------------------------------------- |
2108 | * | MOVE MOUNT OPERATION | | |
2109 | * |************************************************************************** | |
2110 | * | source-->| shared | private | slave | unbindable | | |
2111 | * | dest | | | | | | |
2112 | * | | | | | | | | |
2113 | * | v | | | | | | |
2114 | * |************************************************************************** | |
2115 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2116 | * | | | | | | | |
2117 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2118 | * *************************************************************************** | |
5afe0022 RP |
2119 | * |
2120 | * (+) the mount is moved to the destination. And is then propagated to | |
2121 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2122 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2123 | * (+++) the mount is moved to the destination and is then propagated to |
2124 | * all the mounts belonging to the destination mount's propagation tree. | |
2125 | * the mount is marked as 'shared and slave'. | |
2126 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2127 | * |
2128 | * if the source mount is a tree, the operations explained above is | |
2129 | * applied to each mount in the tree. | |
2130 | * Must be called without spinlocks held, since this function can sleep | |
2131 | * in allocations. | |
2132 | */ | |
0fb54e50 | 2133 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2134 | struct mount *dest_mnt, |
2135 | struct mountpoint *dest_mp, | |
2763d119 | 2136 | bool moving) |
b90fa9ae | 2137 | { |
3bd045cc | 2138 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2139 | HLIST_HEAD(tree_list); |
d2921684 | 2140 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2141 | struct mountpoint *smp; |
315fc83e | 2142 | struct mount *child, *p; |
38129a13 | 2143 | struct hlist_node *n; |
719f5d7f | 2144 | int err; |
b90fa9ae | 2145 | |
1064f874 EB |
2146 | /* Preallocate a mountpoint in case the new mounts need |
2147 | * to be tucked under other mounts. | |
2148 | */ | |
2149 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2150 | if (IS_ERR(smp)) | |
2151 | return PTR_ERR(smp); | |
2152 | ||
d2921684 | 2153 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2154 | if (!moving) { |
d2921684 EB |
2155 | err = count_mounts(ns, source_mnt); |
2156 | if (err) | |
2157 | goto out; | |
2158 | } | |
2159 | ||
fc7be130 | 2160 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2161 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2162 | if (err) |
2163 | goto out; | |
0b1b901b | 2164 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2165 | lock_mount_hash(); |
0b1b901b AV |
2166 | if (err) |
2167 | goto out_cleanup_ids; | |
909b0a88 | 2168 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2169 | set_mnt_shared(p); |
0b1b901b AV |
2170 | } else { |
2171 | lock_mount_hash(); | |
b90fa9ae | 2172 | } |
2763d119 AV |
2173 | if (moving) { |
2174 | unhash_mnt(source_mnt); | |
84d17192 | 2175 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2176 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2177 | } else { |
44dfd84a DH |
2178 | if (source_mnt->mnt_ns) { |
2179 | /* move from anon - the caller will destroy */ | |
2180 | list_del_init(&source_mnt->mnt_ns->list); | |
2181 | } | |
84d17192 | 2182 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2183 | commit_tree(source_mnt); |
21444403 | 2184 | } |
b90fa9ae | 2185 | |
38129a13 | 2186 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2187 | struct mount *q; |
38129a13 | 2188 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2189 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2190 | child->mnt_mountpoint); | |
2191 | if (q) | |
2192 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2193 | /* Notice when we are propagating across user namespaces */ |
2194 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2195 | lock_mnt_tree(child); | |
d728cf79 | 2196 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2197 | commit_tree(child); |
b90fa9ae | 2198 | } |
1064f874 | 2199 | put_mountpoint(smp); |
719ea2fb | 2200 | unlock_mount_hash(); |
99b7db7b | 2201 | |
b90fa9ae | 2202 | return 0; |
719f5d7f MS |
2203 | |
2204 | out_cleanup_ids: | |
f2ebb3a9 AV |
2205 | while (!hlist_empty(&tree_list)) { |
2206 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2207 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2208 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2209 | } |
2210 | unlock_mount_hash(); | |
0b1b901b | 2211 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2212 | out: |
d2921684 | 2213 | ns->pending_mounts = 0; |
1064f874 EB |
2214 | |
2215 | read_seqlock_excl(&mount_lock); | |
2216 | put_mountpoint(smp); | |
2217 | read_sequnlock_excl(&mount_lock); | |
2218 | ||
719f5d7f | 2219 | return err; |
b90fa9ae RP |
2220 | } |
2221 | ||
84d17192 | 2222 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2223 | { |
2224 | struct vfsmount *mnt; | |
84d17192 | 2225 | struct dentry *dentry = path->dentry; |
b12cea91 | 2226 | retry: |
5955102c | 2227 | inode_lock(dentry->d_inode); |
84d17192 | 2228 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2229 | inode_unlock(dentry->d_inode); |
84d17192 | 2230 | return ERR_PTR(-ENOENT); |
b12cea91 | 2231 | } |
97216be0 | 2232 | namespace_lock(); |
b12cea91 | 2233 | mnt = lookup_mnt(path); |
84d17192 | 2234 | if (likely(!mnt)) { |
3895dbf8 | 2235 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2236 | if (IS_ERR(mp)) { |
97216be0 | 2237 | namespace_unlock(); |
5955102c | 2238 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2239 | return mp; |
2240 | } | |
2241 | return mp; | |
2242 | } | |
97216be0 | 2243 | namespace_unlock(); |
5955102c | 2244 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2245 | path_put(path); |
2246 | path->mnt = mnt; | |
84d17192 | 2247 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2248 | goto retry; |
2249 | } | |
2250 | ||
84d17192 | 2251 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2252 | { |
84d17192 | 2253 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2254 | |
2255 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2256 | put_mountpoint(where); |
3895dbf8 EB |
2257 | read_sequnlock_excl(&mount_lock); |
2258 | ||
328e6d90 | 2259 | namespace_unlock(); |
5955102c | 2260 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2261 | } |
2262 | ||
84d17192 | 2263 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2264 | { |
e462ec50 | 2265 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2266 | return -EINVAL; |
2267 | ||
e36cb0b8 DH |
2268 | if (d_is_dir(mp->m_dentry) != |
2269 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2270 | return -ENOTDIR; |
2271 | ||
2763d119 | 2272 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2273 | } |
2274 | ||
7a2e8a8f VA |
2275 | /* |
2276 | * Sanity check the flags to change_mnt_propagation. | |
2277 | */ | |
2278 | ||
e462ec50 | 2279 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2280 | { |
e462ec50 | 2281 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2282 | |
2283 | /* Fail if any non-propagation flags are set */ | |
2284 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2285 | return 0; | |
2286 | /* Only one propagation flag should be set */ | |
2287 | if (!is_power_of_2(type)) | |
2288 | return 0; | |
2289 | return type; | |
2290 | } | |
2291 | ||
07b20889 RP |
2292 | /* |
2293 | * recursively change the type of the mountpoint. | |
2294 | */ | |
e462ec50 | 2295 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2296 | { |
315fc83e | 2297 | struct mount *m; |
4b8b21f4 | 2298 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2299 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2300 | int type; |
719f5d7f | 2301 | int err = 0; |
07b20889 | 2302 | |
2d92ab3c | 2303 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2304 | return -EINVAL; |
2305 | ||
e462ec50 | 2306 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2307 | if (!type) |
2308 | return -EINVAL; | |
2309 | ||
97216be0 | 2310 | namespace_lock(); |
719f5d7f MS |
2311 | if (type == MS_SHARED) { |
2312 | err = invent_group_ids(mnt, recurse); | |
2313 | if (err) | |
2314 | goto out_unlock; | |
2315 | } | |
2316 | ||
719ea2fb | 2317 | lock_mount_hash(); |
909b0a88 | 2318 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2319 | change_mnt_propagation(m, type); |
719ea2fb | 2320 | unlock_mount_hash(); |
719f5d7f MS |
2321 | |
2322 | out_unlock: | |
97216be0 | 2323 | namespace_unlock(); |
719f5d7f | 2324 | return err; |
07b20889 RP |
2325 | } |
2326 | ||
5ff9d8a6 EB |
2327 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2328 | { | |
2329 | struct mount *child; | |
2330 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2331 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2332 | continue; | |
2333 | ||
2334 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2335 | return true; | |
2336 | } | |
2337 | return false; | |
2338 | } | |
2339 | ||
a07b2000 AV |
2340 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2341 | { | |
2342 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2343 | ||
2344 | if (IS_MNT_UNBINDABLE(old)) | |
2345 | return mnt; | |
2346 | ||
2347 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2348 | return mnt; | |
2349 | ||
2350 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2351 | return mnt; | |
2352 | ||
2353 | if (recurse) | |
2354 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2355 | else | |
2356 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2357 | ||
2358 | if (!IS_ERR(mnt)) | |
2359 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2360 | ||
2361 | return mnt; | |
2362 | } | |
2363 | ||
1da177e4 LT |
2364 | /* |
2365 | * do loopback mount. | |
2366 | */ | |
808d4e3c | 2367 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2368 | int recurse) |
1da177e4 | 2369 | { |
2d92ab3c | 2370 | struct path old_path; |
a07b2000 | 2371 | struct mount *mnt = NULL, *parent; |
84d17192 | 2372 | struct mountpoint *mp; |
57eccb83 | 2373 | int err; |
1da177e4 LT |
2374 | if (!old_name || !*old_name) |
2375 | return -EINVAL; | |
815d405c | 2376 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2377 | if (err) |
2378 | return err; | |
2379 | ||
8823c079 | 2380 | err = -EINVAL; |
4ce5d2b1 | 2381 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2382 | goto out; |
8823c079 | 2383 | |
84d17192 | 2384 | mp = lock_mount(path); |
a07b2000 AV |
2385 | if (IS_ERR(mp)) { |
2386 | err = PTR_ERR(mp); | |
b12cea91 | 2387 | goto out; |
a07b2000 | 2388 | } |
b12cea91 | 2389 | |
84d17192 | 2390 | parent = real_mount(path->mnt); |
e149ed2b AV |
2391 | if (!check_mnt(parent)) |
2392 | goto out2; | |
2393 | ||
a07b2000 | 2394 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2395 | if (IS_ERR(mnt)) { |
2396 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2397 | goto out2; |
be34d1a3 | 2398 | } |
ccd48bc7 | 2399 | |
84d17192 | 2400 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2401 | if (err) { |
719ea2fb | 2402 | lock_mount_hash(); |
e819f152 | 2403 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2404 | unlock_mount_hash(); |
5b83d2c5 | 2405 | } |
b12cea91 | 2406 | out2: |
84d17192 | 2407 | unlock_mount(mp); |
ccd48bc7 | 2408 | out: |
2d92ab3c | 2409 | path_put(&old_path); |
1da177e4 LT |
2410 | return err; |
2411 | } | |
2412 | ||
a07b2000 AV |
2413 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2414 | { | |
2415 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2416 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2417 | struct mount *mnt, *p; | |
2418 | struct file *file; | |
2419 | ||
2420 | if (IS_ERR(ns)) | |
2421 | return ERR_CAST(ns); | |
2422 | ||
2423 | namespace_lock(); | |
2424 | mnt = __do_loopback(path, recursive); | |
2425 | if (IS_ERR(mnt)) { | |
2426 | namespace_unlock(); | |
2427 | free_mnt_ns(ns); | |
2428 | return ERR_CAST(mnt); | |
2429 | } | |
2430 | ||
2431 | lock_mount_hash(); | |
2432 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2433 | p->mnt_ns = ns; | |
2434 | ns->mounts++; | |
2435 | } | |
2436 | ns->root = mnt; | |
2437 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2438 | mntget(&mnt->mnt); | |
2439 | unlock_mount_hash(); | |
2440 | namespace_unlock(); | |
2441 | ||
2442 | mntput(path->mnt); | |
2443 | path->mnt = &mnt->mnt; | |
2444 | file = dentry_open(path, O_PATH, current_cred()); | |
2445 | if (IS_ERR(file)) | |
2446 | dissolve_on_fput(path->mnt); | |
2447 | else | |
2448 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2449 | return file; | |
2450 | } | |
2451 | ||
2658ce09 | 2452 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2453 | { |
2454 | struct file *file; | |
2455 | struct path path; | |
2456 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2457 | bool detached = flags & OPEN_TREE_CLONE; | |
2458 | int error; | |
2459 | int fd; | |
2460 | ||
2461 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2462 | ||
2463 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2464 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2465 | OPEN_TREE_CLOEXEC)) | |
2466 | return -EINVAL; | |
2467 | ||
2468 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2469 | return -EINVAL; | |
2470 | ||
2471 | if (flags & AT_NO_AUTOMOUNT) | |
2472 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2473 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2474 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2475 | if (flags & AT_EMPTY_PATH) | |
2476 | lookup_flags |= LOOKUP_EMPTY; | |
2477 | ||
2478 | if (detached && !may_mount()) | |
2479 | return -EPERM; | |
2480 | ||
2481 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2482 | if (fd < 0) | |
2483 | return fd; | |
2484 | ||
2485 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2486 | if (unlikely(error)) { | |
2487 | file = ERR_PTR(error); | |
2488 | } else { | |
2489 | if (detached) | |
2490 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2491 | else | |
2492 | file = dentry_open(&path, O_PATH, current_cred()); | |
2493 | path_put(&path); | |
2494 | } | |
2495 | if (IS_ERR(file)) { | |
2496 | put_unused_fd(fd); | |
2497 | return PTR_ERR(file); | |
2498 | } | |
2499 | fd_install(fd, file); | |
2500 | return fd; | |
2501 | } | |
2502 | ||
43f5e655 DH |
2503 | /* |
2504 | * Don't allow locked mount flags to be cleared. | |
2505 | * | |
2506 | * No locks need to be held here while testing the various MNT_LOCK | |
2507 | * flags because those flags can never be cleared once they are set. | |
2508 | */ | |
2509 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2510 | { |
43f5e655 DH |
2511 | unsigned int fl = mnt->mnt.mnt_flags; |
2512 | ||
2513 | if ((fl & MNT_LOCK_READONLY) && | |
2514 | !(mnt_flags & MNT_READONLY)) | |
2515 | return false; | |
2516 | ||
2517 | if ((fl & MNT_LOCK_NODEV) && | |
2518 | !(mnt_flags & MNT_NODEV)) | |
2519 | return false; | |
2520 | ||
2521 | if ((fl & MNT_LOCK_NOSUID) && | |
2522 | !(mnt_flags & MNT_NOSUID)) | |
2523 | return false; | |
2524 | ||
2525 | if ((fl & MNT_LOCK_NOEXEC) && | |
2526 | !(mnt_flags & MNT_NOEXEC)) | |
2527 | return false; | |
2528 | ||
2529 | if ((fl & MNT_LOCK_ATIME) && | |
2530 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2531 | return false; | |
2e4b7fcd | 2532 | |
43f5e655 DH |
2533 | return true; |
2534 | } | |
2535 | ||
2536 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2537 | { |
43f5e655 | 2538 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2539 | |
43f5e655 | 2540 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2541 | return 0; |
2542 | ||
2543 | if (readonly_request) | |
43f5e655 DH |
2544 | return mnt_make_readonly(mnt); |
2545 | ||
68847c94 CB |
2546 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2547 | return 0; | |
43f5e655 DH |
2548 | } |
2549 | ||
43f5e655 DH |
2550 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2551 | { | |
43f5e655 DH |
2552 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2553 | mnt->mnt.mnt_flags = mnt_flags; | |
2554 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2555 | } |
2556 | ||
f8b92ba6 DD |
2557 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2558 | { | |
2559 | struct super_block *sb = mnt->mnt_sb; | |
2560 | ||
2561 | if (!__mnt_is_readonly(mnt) && | |
2562 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { | |
2563 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2564 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
2565 | struct tm tm; | |
2566 | ||
2567 | time64_to_tm(sb->s_time_max, 0, &tm); | |
2568 | ||
0ecee669 EB |
2569 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
2570 | sb->s_type->name, | |
2571 | is_mounted(mnt) ? "remounted" : "mounted", | |
2572 | mntpath, | |
f8b92ba6 DD |
2573 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
2574 | ||
2575 | free_page((unsigned long)buf); | |
2576 | } | |
2577 | } | |
2578 | ||
43f5e655 DH |
2579 | /* |
2580 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2581 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2582 | * to mount(2). | |
2583 | */ | |
2584 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2585 | { | |
2586 | struct super_block *sb = path->mnt->mnt_sb; | |
2587 | struct mount *mnt = real_mount(path->mnt); | |
2588 | int ret; | |
2589 | ||
2590 | if (!check_mnt(mnt)) | |
2591 | return -EINVAL; | |
2592 | ||
2593 | if (path->dentry != mnt->mnt.mnt_root) | |
2594 | return -EINVAL; | |
2595 | ||
2596 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2597 | return -EPERM; | |
2598 | ||
e58ace1a CB |
2599 | /* |
2600 | * We're only checking whether the superblock is read-only not | |
2601 | * changing it, so only take down_read(&sb->s_umount). | |
2602 | */ | |
2603 | down_read(&sb->s_umount); | |
68847c94 | 2604 | lock_mount_hash(); |
43f5e655 DH |
2605 | ret = change_mount_ro_state(mnt, mnt_flags); |
2606 | if (ret == 0) | |
2607 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2608 | unlock_mount_hash(); |
e58ace1a | 2609 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2610 | |
2611 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2612 | ||
43f5e655 | 2613 | return ret; |
2e4b7fcd DH |
2614 | } |
2615 | ||
1da177e4 LT |
2616 | /* |
2617 | * change filesystem flags. dir should be a physical root of filesystem. | |
2618 | * If you've mounted a non-root directory somewhere and want to do remount | |
2619 | * on it - tough luck. | |
2620 | */ | |
e462ec50 DH |
2621 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2622 | int mnt_flags, void *data) | |
1da177e4 LT |
2623 | { |
2624 | int err; | |
2d92ab3c | 2625 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2626 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2627 | struct fs_context *fc; |
1da177e4 | 2628 | |
143c8c91 | 2629 | if (!check_mnt(mnt)) |
1da177e4 LT |
2630 | return -EINVAL; |
2631 | ||
2d92ab3c | 2632 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2633 | return -EINVAL; |
2634 | ||
43f5e655 | 2635 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2636 | return -EPERM; |
9566d674 | 2637 | |
8d0347f6 DH |
2638 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2639 | if (IS_ERR(fc)) | |
2640 | return PTR_ERR(fc); | |
ff36fe2c | 2641 | |
b330966f | 2642 | fc->oldapi = true; |
8d0347f6 DH |
2643 | err = parse_monolithic_mount_data(fc, data); |
2644 | if (!err) { | |
2645 | down_write(&sb->s_umount); | |
2646 | err = -EPERM; | |
2647 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2648 | err = reconfigure_super(fc); | |
68847c94 CB |
2649 | if (!err) { |
2650 | lock_mount_hash(); | |
8d0347f6 | 2651 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2652 | unlock_mount_hash(); |
2653 | } | |
8d0347f6 DH |
2654 | } |
2655 | up_write(&sb->s_umount); | |
0e55a7cc | 2656 | } |
f8b92ba6 DD |
2657 | |
2658 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2659 | ||
8d0347f6 | 2660 | put_fs_context(fc); |
1da177e4 LT |
2661 | return err; |
2662 | } | |
2663 | ||
cbbe362c | 2664 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2665 | { |
315fc83e | 2666 | struct mount *p; |
909b0a88 | 2667 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2668 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2669 | return 1; |
2670 | } | |
2671 | return 0; | |
2672 | } | |
2673 | ||
44dfd84a DH |
2674 | /* |
2675 | * Check that there aren't references to earlier/same mount namespaces in the | |
2676 | * specified subtree. Such references can act as pins for mount namespaces | |
2677 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2678 | * cycles to be made. | |
2679 | */ | |
2680 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2681 | { | |
2682 | struct mount *p; | |
2683 | bool ret = false; | |
2684 | ||
2685 | lock_mount_hash(); | |
2686 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2687 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2688 | goto out; | |
2689 | ||
2690 | ret = true; | |
2691 | out: | |
2692 | unlock_mount_hash(); | |
2693 | return ret; | |
2694 | } | |
2695 | ||
2db154b3 | 2696 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2697 | { |
44dfd84a | 2698 | struct mnt_namespace *ns; |
676da58d | 2699 | struct mount *p; |
0fb54e50 | 2700 | struct mount *old; |
2763d119 AV |
2701 | struct mount *parent; |
2702 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2703 | int err; |
44dfd84a | 2704 | bool attached; |
1da177e4 | 2705 | |
2db154b3 | 2706 | mp = lock_mount(new_path); |
84d17192 | 2707 | if (IS_ERR(mp)) |
2db154b3 | 2708 | return PTR_ERR(mp); |
cc53ce53 | 2709 | |
2db154b3 DH |
2710 | old = real_mount(old_path->mnt); |
2711 | p = real_mount(new_path->mnt); | |
2763d119 | 2712 | parent = old->mnt_parent; |
44dfd84a | 2713 | attached = mnt_has_parent(old); |
2763d119 | 2714 | old_mp = old->mnt_mp; |
44dfd84a | 2715 | ns = old->mnt_ns; |
143c8c91 | 2716 | |
1da177e4 | 2717 | err = -EINVAL; |
44dfd84a DH |
2718 | /* The mountpoint must be in our namespace. */ |
2719 | if (!check_mnt(p)) | |
2db154b3 | 2720 | goto out; |
1da177e4 | 2721 | |
570d7a98 EB |
2722 | /* The thing moved must be mounted... */ |
2723 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2724 | goto out; |
2725 | ||
570d7a98 EB |
2726 | /* ... and either ours or the root of anon namespace */ |
2727 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2728 | goto out; |
5ff9d8a6 | 2729 | |
2db154b3 DH |
2730 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2731 | goto out; | |
1da177e4 | 2732 | |
2db154b3 DH |
2733 | if (old_path->dentry != old_path->mnt->mnt_root) |
2734 | goto out; | |
1da177e4 | 2735 | |
2db154b3 DH |
2736 | if (d_is_dir(new_path->dentry) != |
2737 | d_is_dir(old_path->dentry)) | |
2738 | goto out; | |
21444403 RP |
2739 | /* |
2740 | * Don't move a mount residing in a shared parent. | |
2741 | */ | |
2763d119 | 2742 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2743 | goto out; |
9676f0c6 RP |
2744 | /* |
2745 | * Don't move a mount tree containing unbindable mounts to a destination | |
2746 | * mount which is shared. | |
2747 | */ | |
fc7be130 | 2748 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2749 | goto out; |
1da177e4 | 2750 | err = -ELOOP; |
44dfd84a DH |
2751 | if (!check_for_nsfs_mounts(old)) |
2752 | goto out; | |
fc7be130 | 2753 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2754 | if (p == old) |
2db154b3 | 2755 | goto out; |
1da177e4 | 2756 | |
2db154b3 | 2757 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2758 | attached); |
4ac91378 | 2759 | if (err) |
2db154b3 | 2760 | goto out; |
1da177e4 LT |
2761 | |
2762 | /* if the mount is moved, it should no longer be expire | |
2763 | * automatically */ | |
6776db3d | 2764 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2765 | if (attached) |
2766 | put_mountpoint(old_mp); | |
1da177e4 | 2767 | out: |
2db154b3 | 2768 | unlock_mount(mp); |
44dfd84a | 2769 | if (!err) { |
2763d119 AV |
2770 | if (attached) |
2771 | mntput_no_expire(parent); | |
2772 | else | |
44dfd84a DH |
2773 | free_mnt_ns(ns); |
2774 | } | |
2db154b3 DH |
2775 | return err; |
2776 | } | |
2777 | ||
2778 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2779 | { | |
2780 | struct path old_path; | |
2781 | int err; | |
2782 | ||
2783 | if (!old_name || !*old_name) | |
2784 | return -EINVAL; | |
2785 | ||
2786 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2787 | if (err) | |
2788 | return err; | |
2789 | ||
2790 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2791 | path_put(&old_path); |
1da177e4 LT |
2792 | return err; |
2793 | } | |
2794 | ||
9d412a43 AV |
2795 | /* |
2796 | * add a mount into a namespace's mount tree | |
2797 | */ | |
8f11538e AV |
2798 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
2799 | struct path *path, int mnt_flags) | |
9d412a43 | 2800 | { |
8f11538e | 2801 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 2802 | |
f2ebb3a9 | 2803 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2804 | |
84d17192 | 2805 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2806 | /* that's acceptable only for automounts done in private ns */ |
2807 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 2808 | return -EINVAL; |
156cacb1 | 2809 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 2810 | if (!parent->mnt_ns) |
8f11538e | 2811 | return -EINVAL; |
156cacb1 | 2812 | } |
9d412a43 AV |
2813 | |
2814 | /* Refuse the same filesystem on the same mount point */ | |
95bc5f25 | 2815 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 | 2816 | path->mnt->mnt_root == path->dentry) |
8f11538e | 2817 | return -EBUSY; |
9d412a43 | 2818 | |
e36cb0b8 | 2819 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 2820 | return -EINVAL; |
9d412a43 | 2821 | |
95bc5f25 | 2822 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 2823 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 2824 | } |
b1e75df4 | 2825 | |
132e4608 DH |
2826 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2827 | ||
2828 | /* | |
2829 | * Create a new mount using a superblock configuration and request it | |
2830 | * be added to the namespace tree. | |
2831 | */ | |
2832 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2833 | unsigned int mnt_flags) | |
2834 | { | |
2835 | struct vfsmount *mnt; | |
8f11538e | 2836 | struct mountpoint *mp; |
132e4608 DH |
2837 | struct super_block *sb = fc->root->d_sb; |
2838 | int error; | |
2839 | ||
c9ce29ed AV |
2840 | error = security_sb_kern_mount(sb); |
2841 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2842 | error = -EPERM; | |
2843 | ||
2844 | if (unlikely(error)) { | |
2845 | fc_drop_locked(fc); | |
2846 | return error; | |
132e4608 DH |
2847 | } |
2848 | ||
2849 | up_write(&sb->s_umount); | |
2850 | ||
2851 | mnt = vfs_create_mount(fc); | |
2852 | if (IS_ERR(mnt)) | |
2853 | return PTR_ERR(mnt); | |
2854 | ||
f8b92ba6 DD |
2855 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
2856 | ||
8f11538e AV |
2857 | mp = lock_mount(mountpoint); |
2858 | if (IS_ERR(mp)) { | |
2859 | mntput(mnt); | |
2860 | return PTR_ERR(mp); | |
2861 | } | |
2862 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
2863 | unlock_mount(mp); | |
0ecee669 EB |
2864 | if (error < 0) |
2865 | mntput(mnt); | |
132e4608 DH |
2866 | return error; |
2867 | } | |
1b852bce | 2868 | |
1da177e4 LT |
2869 | /* |
2870 | * create a new mount for userspace and request it to be added into the | |
2871 | * namespace's tree | |
2872 | */ | |
e462ec50 | 2873 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2874 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2875 | { |
0c55cfc4 | 2876 | struct file_system_type *type; |
a0c9a8b8 AV |
2877 | struct fs_context *fc; |
2878 | const char *subtype = NULL; | |
2879 | int err = 0; | |
1da177e4 | 2880 | |
0c55cfc4 | 2881 | if (!fstype) |
1da177e4 LT |
2882 | return -EINVAL; |
2883 | ||
0c55cfc4 EB |
2884 | type = get_fs_type(fstype); |
2885 | if (!type) | |
2886 | return -ENODEV; | |
2887 | ||
a0c9a8b8 AV |
2888 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2889 | subtype = strchr(fstype, '.'); | |
2890 | if (subtype) { | |
2891 | subtype++; | |
2892 | if (!*subtype) { | |
2893 | put_filesystem(type); | |
2894 | return -EINVAL; | |
2895 | } | |
a0c9a8b8 AV |
2896 | } |
2897 | } | |
0c55cfc4 | 2898 | |
a0c9a8b8 | 2899 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2900 | put_filesystem(type); |
a0c9a8b8 AV |
2901 | if (IS_ERR(fc)) |
2902 | return PTR_ERR(fc); | |
2903 | ||
3e1aeb00 DH |
2904 | if (subtype) |
2905 | err = vfs_parse_fs_string(fc, "subtype", | |
2906 | subtype, strlen(subtype)); | |
2907 | if (!err && name) | |
2908 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2909 | if (!err) |
2910 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
2911 | if (!err && !mount_capable(fc)) |
2912 | err = -EPERM; | |
a0c9a8b8 AV |
2913 | if (!err) |
2914 | err = vfs_get_tree(fc); | |
132e4608 DH |
2915 | if (!err) |
2916 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2917 | |
a0c9a8b8 | 2918 | put_fs_context(fc); |
15f9a3f3 | 2919 | return err; |
1da177e4 LT |
2920 | } |
2921 | ||
19a167af AV |
2922 | int finish_automount(struct vfsmount *m, struct path *path) |
2923 | { | |
26df6034 | 2924 | struct dentry *dentry = path->dentry; |
8f11538e | 2925 | struct mountpoint *mp; |
25e195aa | 2926 | struct mount *mnt; |
19a167af | 2927 | int err; |
25e195aa AV |
2928 | |
2929 | if (!m) | |
2930 | return 0; | |
2931 | if (IS_ERR(m)) | |
2932 | return PTR_ERR(m); | |
2933 | ||
2934 | mnt = real_mount(m); | |
19a167af AV |
2935 | /* The new mount record should have at least 2 refs to prevent it being |
2936 | * expired before we get a chance to add it | |
2937 | */ | |
6776db3d | 2938 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2939 | |
2940 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 2941 | m->mnt_root == dentry) { |
b1e75df4 | 2942 | err = -ELOOP; |
26df6034 | 2943 | goto discard; |
19a167af AV |
2944 | } |
2945 | ||
26df6034 AV |
2946 | /* |
2947 | * we don't want to use lock_mount() - in this case finding something | |
2948 | * that overmounts our mountpoint to be means "quitely drop what we've | |
2949 | * got", not "try to mount it on top". | |
2950 | */ | |
2951 | inode_lock(dentry->d_inode); | |
2952 | namespace_lock(); | |
2953 | if (unlikely(cant_mount(dentry))) { | |
2954 | err = -ENOENT; | |
2955 | goto discard_locked; | |
2956 | } | |
2957 | rcu_read_lock(); | |
2958 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { | |
2959 | rcu_read_unlock(); | |
2960 | err = 0; | |
2961 | goto discard_locked; | |
2962 | } | |
2963 | rcu_read_unlock(); | |
2964 | mp = get_mountpoint(dentry); | |
8f11538e AV |
2965 | if (IS_ERR(mp)) { |
2966 | err = PTR_ERR(mp); | |
26df6034 | 2967 | goto discard_locked; |
8f11538e | 2968 | } |
26df6034 | 2969 | |
8f11538e AV |
2970 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
2971 | unlock_mount(mp); | |
26df6034 AV |
2972 | if (unlikely(err)) |
2973 | goto discard; | |
2974 | mntput(m); | |
2975 | return 0; | |
2976 | ||
2977 | discard_locked: | |
2978 | namespace_unlock(); | |
2979 | inode_unlock(dentry->d_inode); | |
2980 | discard: | |
b1e75df4 | 2981 | /* remove m from any expiration list it may be on */ |
6776db3d | 2982 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2983 | namespace_lock(); |
6776db3d | 2984 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2985 | namespace_unlock(); |
19a167af | 2986 | } |
b1e75df4 AV |
2987 | mntput(m); |
2988 | mntput(m); | |
19a167af AV |
2989 | return err; |
2990 | } | |
2991 | ||
ea5b778a DH |
2992 | /** |
2993 | * mnt_set_expiry - Put a mount on an expiration list | |
2994 | * @mnt: The mount to list. | |
2995 | * @expiry_list: The list to add the mount to. | |
2996 | */ | |
2997 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2998 | { | |
97216be0 | 2999 | namespace_lock(); |
ea5b778a | 3000 | |
6776db3d | 3001 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 3002 | |
97216be0 | 3003 | namespace_unlock(); |
ea5b778a DH |
3004 | } |
3005 | EXPORT_SYMBOL(mnt_set_expiry); | |
3006 | ||
1da177e4 LT |
3007 | /* |
3008 | * process a list of expirable mountpoints with the intent of discarding any | |
3009 | * mountpoints that aren't in use and haven't been touched since last we came | |
3010 | * here | |
3011 | */ | |
3012 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3013 | { | |
761d5c38 | 3014 | struct mount *mnt, *next; |
1da177e4 LT |
3015 | LIST_HEAD(graveyard); |
3016 | ||
3017 | if (list_empty(mounts)) | |
3018 | return; | |
3019 | ||
97216be0 | 3020 | namespace_lock(); |
719ea2fb | 3021 | lock_mount_hash(); |
1da177e4 LT |
3022 | |
3023 | /* extract from the expiration list every vfsmount that matches the | |
3024 | * following criteria: | |
3025 | * - only referenced by its parent vfsmount | |
3026 | * - still marked for expiry (marked on the last call here; marks are | |
3027 | * cleared by mntput()) | |
3028 | */ | |
6776db3d | 3029 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3030 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3031 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3032 | continue; |
6776db3d | 3033 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3034 | } |
bcc5c7d2 | 3035 | while (!list_empty(&graveyard)) { |
6776db3d | 3036 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3037 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3038 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3039 | } |
719ea2fb | 3040 | unlock_mount_hash(); |
3ab6abee | 3041 | namespace_unlock(); |
5528f911 TM |
3042 | } |
3043 | ||
3044 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3045 | ||
3046 | /* | |
3047 | * Ripoff of 'select_parent()' | |
3048 | * | |
3049 | * search the list of submounts for a given mountpoint, and move any | |
3050 | * shrinkable submounts to the 'graveyard' list. | |
3051 | */ | |
692afc31 | 3052 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3053 | { |
692afc31 | 3054 | struct mount *this_parent = parent; |
5528f911 TM |
3055 | struct list_head *next; |
3056 | int found = 0; | |
3057 | ||
3058 | repeat: | |
6b41d536 | 3059 | next = this_parent->mnt_mounts.next; |
5528f911 | 3060 | resume: |
6b41d536 | 3061 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3062 | struct list_head *tmp = next; |
6b41d536 | 3063 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3064 | |
3065 | next = tmp->next; | |
692afc31 | 3066 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3067 | continue; |
5528f911 TM |
3068 | /* |
3069 | * Descend a level if the d_mounts list is non-empty. | |
3070 | */ | |
6b41d536 | 3071 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3072 | this_parent = mnt; |
3073 | goto repeat; | |
3074 | } | |
1da177e4 | 3075 | |
1ab59738 | 3076 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3077 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3078 | found++; |
3079 | } | |
1da177e4 | 3080 | } |
5528f911 TM |
3081 | /* |
3082 | * All done at this level ... ascend and resume the search | |
3083 | */ | |
3084 | if (this_parent != parent) { | |
6b41d536 | 3085 | next = this_parent->mnt_child.next; |
0714a533 | 3086 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3087 | goto resume; |
3088 | } | |
3089 | return found; | |
3090 | } | |
3091 | ||
3092 | /* | |
3093 | * process a list of expirable mountpoints with the intent of discarding any | |
3094 | * submounts of a specific parent mountpoint | |
99b7db7b | 3095 | * |
48a066e7 | 3096 | * mount_lock must be held for write |
5528f911 | 3097 | */ |
b54b9be7 | 3098 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3099 | { |
3100 | LIST_HEAD(graveyard); | |
761d5c38 | 3101 | struct mount *m; |
5528f911 | 3102 | |
5528f911 | 3103 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3104 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3105 | while (!list_empty(&graveyard)) { |
761d5c38 | 3106 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3107 | mnt_expire); |
143c8c91 | 3108 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3109 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3110 | } |
3111 | } | |
1da177e4 LT |
3112 | } |
3113 | ||
028abd92 | 3114 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3115 | { |
b40ef869 | 3116 | char *copy; |
d563d678 | 3117 | unsigned left, offset; |
b58fed8b | 3118 | |
1da177e4 | 3119 | if (!data) |
b40ef869 | 3120 | return NULL; |
1da177e4 | 3121 | |
b40ef869 AV |
3122 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3123 | if (!copy) | |
3124 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3125 | |
d563d678 | 3126 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3127 | |
d563d678 CM |
3128 | /* |
3129 | * Not all architectures have an exact copy_from_user(). Resort to | |
3130 | * byte at a time. | |
3131 | */ | |
3132 | offset = PAGE_SIZE - left; | |
3133 | while (left) { | |
3134 | char c; | |
3135 | if (get_user(c, (const char __user *)data + offset)) | |
3136 | break; | |
3137 | copy[offset] = c; | |
3138 | left--; | |
3139 | offset++; | |
3140 | } | |
3141 | ||
3142 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3143 | kfree(copy); |
3144 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3145 | } |
d563d678 | 3146 | |
b40ef869 | 3147 | return copy; |
1da177e4 LT |
3148 | } |
3149 | ||
028abd92 | 3150 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3151 | { |
fbdb4401 | 3152 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3153 | } |
3154 | ||
1da177e4 LT |
3155 | /* |
3156 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3157 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3158 | * | |
3159 | * data is a (void *) that can point to any structure up to | |
3160 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3161 | * information (or be NULL). | |
3162 | * | |
3163 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3164 | * When the flags word was introduced its top half was required | |
3165 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3166 | * Therefore, if this magic number is present, it carries no information | |
3167 | * and must be discarded. | |
3168 | */ | |
c60166f0 | 3169 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3170 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3171 | { |
e462ec50 | 3172 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3173 | int ret; |
1da177e4 LT |
3174 | |
3175 | /* Discard magic */ | |
3176 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3177 | flags &= ~MS_MGC_MSK; | |
3178 | ||
3179 | /* Basic sanity checks */ | |
1da177e4 LT |
3180 | if (data_page) |
3181 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3182 | ||
e462ec50 DH |
3183 | if (flags & MS_NOUSER) |
3184 | return -EINVAL; | |
3185 | ||
a1e6aaa3 CH |
3186 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3187 | if (ret) | |
3188 | return ret; | |
3189 | if (!may_mount()) | |
3190 | return -EPERM; | |
3191 | if ((flags & SB_MANDLOCK) && !may_mandlock()) | |
3192 | return -EPERM; | |
a27ab9f2 | 3193 | |
613cbe3d AK |
3194 | /* Default to relatime unless overriden */ |
3195 | if (!(flags & MS_NOATIME)) | |
3196 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3197 | |
1da177e4 LT |
3198 | /* Separate the per-mountpoint flags */ |
3199 | if (flags & MS_NOSUID) | |
3200 | mnt_flags |= MNT_NOSUID; | |
3201 | if (flags & MS_NODEV) | |
3202 | mnt_flags |= MNT_NODEV; | |
3203 | if (flags & MS_NOEXEC) | |
3204 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3205 | if (flags & MS_NOATIME) |
3206 | mnt_flags |= MNT_NOATIME; | |
3207 | if (flags & MS_NODIRATIME) | |
3208 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3209 | if (flags & MS_STRICTATIME) |
3210 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3211 | if (flags & MS_RDONLY) |
2e4b7fcd | 3212 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3213 | if (flags & MS_NOSYMFOLLOW) |
3214 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3215 | |
ffbc6f0e EB |
3216 | /* The default atime for remount is preservation */ |
3217 | if ((flags & MS_REMOUNT) && | |
3218 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3219 | MS_STRICTATIME)) == 0)) { | |
3220 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3221 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3222 | } |
3223 | ||
e462ec50 DH |
3224 | sb_flags = flags & (SB_RDONLY | |
3225 | SB_SYNCHRONOUS | | |
3226 | SB_MANDLOCK | | |
3227 | SB_DIRSYNC | | |
3228 | SB_SILENT | | |
917086ff | 3229 | SB_POSIXACL | |
d7ee9469 | 3230 | SB_LAZYTIME | |
917086ff | 3231 | SB_I_VERSION); |
1da177e4 | 3232 | |
43f5e655 | 3233 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3234 | return do_reconfigure_mnt(path, mnt_flags); |
3235 | if (flags & MS_REMOUNT) | |
3236 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3237 | if (flags & MS_BIND) | |
3238 | return do_loopback(path, dev_name, flags & MS_REC); | |
3239 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3240 | return do_change_type(path, flags); | |
3241 | if (flags & MS_MOVE) | |
3242 | return do_move_mount_old(path, dev_name); | |
3243 | ||
3244 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3245 | data_page); | |
3246 | } | |
3247 | ||
3248 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3249 | const char *type_page, unsigned long flags, void *data_page) | |
3250 | { | |
3251 | struct path path; | |
3252 | int ret; | |
3253 | ||
3254 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3255 | if (ret) | |
3256 | return ret; | |
3257 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3258 | path_put(&path); |
a1e6aaa3 | 3259 | return ret; |
1da177e4 LT |
3260 | } |
3261 | ||
537f7ccb EB |
3262 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3263 | { | |
3264 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3265 | } | |
3266 | ||
3267 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3268 | { | |
3269 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3270 | } | |
3271 | ||
771b1371 EB |
3272 | static void free_mnt_ns(struct mnt_namespace *ns) |
3273 | { | |
74e83122 AV |
3274 | if (!is_anon_ns(ns)) |
3275 | ns_free_inum(&ns->ns); | |
537f7ccb | 3276 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3277 | put_user_ns(ns->user_ns); |
3278 | kfree(ns); | |
3279 | } | |
3280 | ||
8823c079 EB |
3281 | /* |
3282 | * Assign a sequence number so we can detect when we attempt to bind | |
3283 | * mount a reference to an older mount namespace into the current | |
3284 | * mount namespace, preventing reference counting loops. A 64bit | |
3285 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3286 | * is effectively never, so we can ignore the possibility. | |
3287 | */ | |
3288 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3289 | ||
74e83122 | 3290 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3291 | { |
3292 | struct mnt_namespace *new_ns; | |
537f7ccb | 3293 | struct ucounts *ucounts; |
98f842e6 | 3294 | int ret; |
cf8d2c11 | 3295 | |
537f7ccb EB |
3296 | ucounts = inc_mnt_namespaces(user_ns); |
3297 | if (!ucounts) | |
df75e774 | 3298 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3299 | |
74e83122 | 3300 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
3301 | if (!new_ns) { |
3302 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3303 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3304 | } |
74e83122 AV |
3305 | if (!anon) { |
3306 | ret = ns_alloc_inum(&new_ns->ns); | |
3307 | if (ret) { | |
3308 | kfree(new_ns); | |
3309 | dec_mnt_namespaces(ucounts); | |
3310 | return ERR_PTR(ret); | |
3311 | } | |
98f842e6 | 3312 | } |
33c42940 | 3313 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3314 | if (!anon) |
3315 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3316 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3317 | INIT_LIST_HEAD(&new_ns->list); |
3318 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3319 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3320 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3321 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3322 | return new_ns; |
3323 | } | |
3324 | ||
0766f788 | 3325 | __latent_entropy |
9559f689 AV |
3326 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3327 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3328 | { |
6b3286ed | 3329 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3330 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3331 | struct mount *p, *q; |
9559f689 | 3332 | struct mount *old; |
cb338d06 | 3333 | struct mount *new; |
7a472ef4 | 3334 | int copy_flags; |
1da177e4 | 3335 | |
9559f689 AV |
3336 | BUG_ON(!ns); |
3337 | ||
3338 | if (likely(!(flags & CLONE_NEWNS))) { | |
3339 | get_mnt_ns(ns); | |
3340 | return ns; | |
3341 | } | |
3342 | ||
3343 | old = ns->root; | |
3344 | ||
74e83122 | 3345 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3346 | if (IS_ERR(new_ns)) |
3347 | return new_ns; | |
1da177e4 | 3348 | |
97216be0 | 3349 | namespace_lock(); |
1da177e4 | 3350 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3351 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3352 | if (user_ns != ns->user_ns) |
3bd045cc | 3353 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3354 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3355 | if (IS_ERR(new)) { |
328e6d90 | 3356 | namespace_unlock(); |
771b1371 | 3357 | free_mnt_ns(new_ns); |
be34d1a3 | 3358 | return ERR_CAST(new); |
1da177e4 | 3359 | } |
3bd045cc AV |
3360 | if (user_ns != ns->user_ns) { |
3361 | lock_mount_hash(); | |
3362 | lock_mnt_tree(new); | |
3363 | unlock_mount_hash(); | |
3364 | } | |
be08d6d2 | 3365 | new_ns->root = new; |
1a4eeaf2 | 3366 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3367 | |
3368 | /* | |
3369 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3370 | * as belonging to new namespace. We have already acquired a private | |
3371 | * fs_struct, so tsk->fs->lock is not needed. | |
3372 | */ | |
909b0a88 | 3373 | p = old; |
cb338d06 | 3374 | q = new; |
1da177e4 | 3375 | while (p) { |
143c8c91 | 3376 | q->mnt_ns = new_ns; |
d2921684 | 3377 | new_ns->mounts++; |
9559f689 AV |
3378 | if (new_fs) { |
3379 | if (&p->mnt == new_fs->root.mnt) { | |
3380 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3381 | rootmnt = &p->mnt; |
1da177e4 | 3382 | } |
9559f689 AV |
3383 | if (&p->mnt == new_fs->pwd.mnt) { |
3384 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3385 | pwdmnt = &p->mnt; |
1da177e4 | 3386 | } |
1da177e4 | 3387 | } |
909b0a88 AV |
3388 | p = next_mnt(p, old); |
3389 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3390 | if (!q) |
3391 | break; | |
3392 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3393 | p = next_mnt(p, old); | |
1da177e4 | 3394 | } |
328e6d90 | 3395 | namespace_unlock(); |
1da177e4 | 3396 | |
1da177e4 | 3397 | if (rootmnt) |
f03c6599 | 3398 | mntput(rootmnt); |
1da177e4 | 3399 | if (pwdmnt) |
f03c6599 | 3400 | mntput(pwdmnt); |
1da177e4 | 3401 | |
741a2951 | 3402 | return new_ns; |
1da177e4 LT |
3403 | } |
3404 | ||
74e83122 | 3405 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3406 | { |
74e83122 | 3407 | struct mount *mnt = real_mount(m); |
ea441d11 | 3408 | struct mnt_namespace *ns; |
d31da0f0 | 3409 | struct super_block *s; |
ea441d11 AV |
3410 | struct path path; |
3411 | int err; | |
3412 | ||
74e83122 AV |
3413 | ns = alloc_mnt_ns(&init_user_ns, true); |
3414 | if (IS_ERR(ns)) { | |
3415 | mntput(m); | |
ea441d11 | 3416 | return ERR_CAST(ns); |
74e83122 AV |
3417 | } |
3418 | mnt->mnt_ns = ns; | |
3419 | ns->root = mnt; | |
3420 | ns->mounts++; | |
3421 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3422 | |
74e83122 | 3423 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3424 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3425 | ||
3426 | put_mnt_ns(ns); | |
3427 | ||
3428 | if (err) | |
3429 | return ERR_PTR(err); | |
3430 | ||
3431 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3432 | s = path.mnt->mnt_sb; |
3433 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3434 | mntput(path.mnt); |
3435 | /* lock the sucker */ | |
d31da0f0 | 3436 | down_write(&s->s_umount); |
ea441d11 AV |
3437 | /* ... and return the root of (sub)tree on it */ |
3438 | return path.dentry; | |
3439 | } | |
3440 | EXPORT_SYMBOL(mount_subtree); | |
3441 | ||
cccaa5e3 DB |
3442 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3443 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3444 | { |
eca6f534 VN |
3445 | int ret; |
3446 | char *kernel_type; | |
eca6f534 | 3447 | char *kernel_dev; |
b40ef869 | 3448 | void *options; |
1da177e4 | 3449 | |
b8850d1f TG |
3450 | kernel_type = copy_mount_string(type); |
3451 | ret = PTR_ERR(kernel_type); | |
3452 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3453 | goto out_type; |
1da177e4 | 3454 | |
b8850d1f TG |
3455 | kernel_dev = copy_mount_string(dev_name); |
3456 | ret = PTR_ERR(kernel_dev); | |
3457 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3458 | goto out_dev; |
1da177e4 | 3459 | |
b40ef869 AV |
3460 | options = copy_mount_options(data); |
3461 | ret = PTR_ERR(options); | |
3462 | if (IS_ERR(options)) | |
eca6f534 | 3463 | goto out_data; |
1da177e4 | 3464 | |
b40ef869 | 3465 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3466 | |
b40ef869 | 3467 | kfree(options); |
eca6f534 VN |
3468 | out_data: |
3469 | kfree(kernel_dev); | |
3470 | out_dev: | |
eca6f534 VN |
3471 | kfree(kernel_type); |
3472 | out_type: | |
3473 | return ret; | |
1da177e4 LT |
3474 | } |
3475 | ||
5b490500 CB |
3476 | #define FSMOUNT_VALID_FLAGS \ |
3477 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3478 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME) | |
3479 | ||
2a186721 CB |
3480 | #define MOUNT_SETATTR_VALID_FLAGS FSMOUNT_VALID_FLAGS |
3481 | ||
3482 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3483 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3484 | ||
5b490500 CB |
3485 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3486 | { | |
3487 | unsigned int mnt_flags = 0; | |
3488 | ||
3489 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3490 | mnt_flags |= MNT_READONLY; | |
3491 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3492 | mnt_flags |= MNT_NOSUID; | |
3493 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3494 | mnt_flags |= MNT_NODEV; | |
3495 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3496 | mnt_flags |= MNT_NOEXEC; | |
3497 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3498 | mnt_flags |= MNT_NODIRATIME; | |
3499 | ||
3500 | return mnt_flags; | |
3501 | } | |
3502 | ||
2db154b3 | 3503 | /* |
93766fbd DH |
3504 | * Create a kernel mount representation for a new, prepared superblock |
3505 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3506 | */ | |
3507 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3508 | unsigned int, attr_flags) | |
3509 | { | |
3510 | struct mnt_namespace *ns; | |
3511 | struct fs_context *fc; | |
3512 | struct file *file; | |
3513 | struct path newmount; | |
3514 | struct mount *mnt; | |
3515 | struct fd f; | |
3516 | unsigned int mnt_flags = 0; | |
3517 | long ret; | |
3518 | ||
3519 | if (!may_mount()) | |
3520 | return -EPERM; | |
3521 | ||
3522 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3523 | return -EINVAL; | |
3524 | ||
5b490500 | 3525 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3526 | return -EINVAL; |
3527 | ||
5b490500 | 3528 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3529 | |
3530 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3531 | case MOUNT_ATTR_STRICTATIME: | |
3532 | break; | |
3533 | case MOUNT_ATTR_NOATIME: | |
3534 | mnt_flags |= MNT_NOATIME; | |
3535 | break; | |
3536 | case MOUNT_ATTR_RELATIME: | |
3537 | mnt_flags |= MNT_RELATIME; | |
3538 | break; | |
3539 | default: | |
3540 | return -EINVAL; | |
3541 | } | |
3542 | ||
3543 | f = fdget(fs_fd); | |
3544 | if (!f.file) | |
3545 | return -EBADF; | |
3546 | ||
3547 | ret = -EINVAL; | |
3548 | if (f.file->f_op != &fscontext_fops) | |
3549 | goto err_fsfd; | |
3550 | ||
3551 | fc = f.file->private_data; | |
3552 | ||
3553 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3554 | if (ret < 0) | |
3555 | goto err_fsfd; | |
3556 | ||
3557 | /* There must be a valid superblock or we can't mount it */ | |
3558 | ret = -EINVAL; | |
3559 | if (!fc->root) | |
3560 | goto err_unlock; | |
3561 | ||
3562 | ret = -EPERM; | |
3563 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3564 | pr_warn("VFS: Mount too revealing\n"); | |
3565 | goto err_unlock; | |
3566 | } | |
3567 | ||
3568 | ret = -EBUSY; | |
3569 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3570 | goto err_unlock; | |
3571 | ||
3572 | ret = -EPERM; | |
3573 | if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) | |
3574 | goto err_unlock; | |
3575 | ||
3576 | newmount.mnt = vfs_create_mount(fc); | |
3577 | if (IS_ERR(newmount.mnt)) { | |
3578 | ret = PTR_ERR(newmount.mnt); | |
3579 | goto err_unlock; | |
3580 | } | |
3581 | newmount.dentry = dget(fc->root); | |
3582 | newmount.mnt->mnt_flags = mnt_flags; | |
3583 | ||
3584 | /* We've done the mount bit - now move the file context into more or | |
3585 | * less the same state as if we'd done an fspick(). We don't want to | |
3586 | * do any memory allocation or anything like that at this point as we | |
3587 | * don't want to have to handle any errors incurred. | |
3588 | */ | |
3589 | vfs_clean_context(fc); | |
3590 | ||
3591 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3592 | if (IS_ERR(ns)) { | |
3593 | ret = PTR_ERR(ns); | |
3594 | goto err_path; | |
3595 | } | |
3596 | mnt = real_mount(newmount.mnt); | |
3597 | mnt->mnt_ns = ns; | |
3598 | ns->root = mnt; | |
3599 | ns->mounts = 1; | |
3600 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3601 | mntget(newmount.mnt); |
93766fbd DH |
3602 | |
3603 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3604 | * it, not just simply put it. | |
3605 | */ | |
3606 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3607 | if (IS_ERR(file)) { | |
3608 | dissolve_on_fput(newmount.mnt); | |
3609 | ret = PTR_ERR(file); | |
3610 | goto err_path; | |
3611 | } | |
3612 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3613 | ||
3614 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3615 | if (ret >= 0) | |
3616 | fd_install(ret, file); | |
3617 | else | |
3618 | fput(file); | |
3619 | ||
3620 | err_path: | |
3621 | path_put(&newmount); | |
3622 | err_unlock: | |
3623 | mutex_unlock(&fc->uapi_mutex); | |
3624 | err_fsfd: | |
3625 | fdput(f); | |
3626 | return ret; | |
3627 | } | |
3628 | ||
3629 | /* | |
3630 | * Move a mount from one place to another. In combination with | |
3631 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3632 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3633 | * a mount subtree. | |
2db154b3 DH |
3634 | * |
3635 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3636 | */ | |
3637 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
3638 | int, from_dfd, const char __user *, from_pathname, |
3639 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
3640 | unsigned int, flags) |
3641 | { | |
3642 | struct path from_path, to_path; | |
3643 | unsigned int lflags; | |
3644 | int ret = 0; | |
3645 | ||
3646 | if (!may_mount()) | |
3647 | return -EPERM; | |
3648 | ||
3649 | if (flags & ~MOVE_MOUNT__MASK) | |
3650 | return -EINVAL; | |
3651 | ||
3652 | /* If someone gives a pathname, they aren't permitted to move | |
3653 | * from an fd that requires unmount as we can't get at the flag | |
3654 | * to clear it afterwards. | |
3655 | */ | |
3656 | lflags = 0; | |
3657 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3658 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3659 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3660 | ||
3661 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3662 | if (ret < 0) | |
3663 | return ret; | |
3664 | ||
3665 | lflags = 0; | |
3666 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3667 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3668 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3669 | ||
3670 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3671 | if (ret < 0) | |
3672 | goto out_from; | |
3673 | ||
3674 | ret = security_move_mount(&from_path, &to_path); | |
3675 | if (ret < 0) | |
3676 | goto out_to; | |
3677 | ||
3678 | ret = do_move_mount(&from_path, &to_path); | |
3679 | ||
3680 | out_to: | |
3681 | path_put(&to_path); | |
3682 | out_from: | |
3683 | path_put(&from_path); | |
3684 | return ret; | |
3685 | } | |
3686 | ||
afac7cba AV |
3687 | /* |
3688 | * Return true if path is reachable from root | |
3689 | * | |
48a066e7 | 3690 | * namespace_sem or mount_lock is held |
afac7cba | 3691 | */ |
643822b4 | 3692 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3693 | const struct path *root) |
3694 | { | |
643822b4 | 3695 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3696 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3697 | mnt = mnt->mnt_parent; |
afac7cba | 3698 | } |
643822b4 | 3699 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3700 | } |
3701 | ||
640eb7e7 | 3702 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3703 | { |
25ab4c9b | 3704 | bool res; |
48a066e7 | 3705 | read_seqlock_excl(&mount_lock); |
643822b4 | 3706 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3707 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3708 | return res; |
3709 | } | |
3710 | EXPORT_SYMBOL(path_is_under); | |
3711 | ||
1da177e4 LT |
3712 | /* |
3713 | * pivot_root Semantics: | |
3714 | * Moves the root file system of the current process to the directory put_old, | |
3715 | * makes new_root as the new root file system of the current process, and sets | |
3716 | * root/cwd of all processes which had them on the current root to new_root. | |
3717 | * | |
3718 | * Restrictions: | |
3719 | * The new_root and put_old must be directories, and must not be on the | |
3720 | * same file system as the current process root. The put_old must be | |
3721 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3722 | * pointed to by put_old must yield the same directory as new_root. No other | |
3723 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3724 | * | |
4a0d11fa | 3725 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 3726 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
3727 | * in this situation. |
3728 | * | |
1da177e4 LT |
3729 | * Notes: |
3730 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3731 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3732 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3733 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3734 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3735 | * first. | |
3736 | */ | |
3480b257 HC |
3737 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3738 | const char __user *, put_old) | |
1da177e4 | 3739 | { |
2763d119 AV |
3740 | struct path new, old, root; |
3741 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3742 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3743 | int error; |
3744 | ||
9b40bc90 | 3745 | if (!may_mount()) |
1da177e4 LT |
3746 | return -EPERM; |
3747 | ||
ce6595a2 AV |
3748 | error = user_path_at(AT_FDCWD, new_root, |
3749 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
3750 | if (error) |
3751 | goto out0; | |
1da177e4 | 3752 | |
ce6595a2 AV |
3753 | error = user_path_at(AT_FDCWD, put_old, |
3754 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
3755 | if (error) |
3756 | goto out1; | |
3757 | ||
2d8f3038 | 3758 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3759 | if (error) |
3760 | goto out2; | |
1da177e4 | 3761 | |
f7ad3c6b | 3762 | get_fs_root(current->fs, &root); |
84d17192 AV |
3763 | old_mp = lock_mount(&old); |
3764 | error = PTR_ERR(old_mp); | |
3765 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3766 | goto out3; |
3767 | ||
1da177e4 | 3768 | error = -EINVAL; |
419148da AV |
3769 | new_mnt = real_mount(new.mnt); |
3770 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3771 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3772 | ex_parent = new_mnt->mnt_parent; |
3773 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3774 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3775 | IS_MNT_SHARED(ex_parent) || |
3776 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3777 | goto out4; |
143c8c91 | 3778 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3779 | goto out4; |
5ff9d8a6 EB |
3780 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3781 | goto out4; | |
1da177e4 | 3782 | error = -ENOENT; |
f3da392e | 3783 | if (d_unlinked(new.dentry)) |
b12cea91 | 3784 | goto out4; |
1da177e4 | 3785 | error = -EBUSY; |
84d17192 | 3786 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3787 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3788 | error = -EINVAL; |
8c3ee42e | 3789 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3790 | goto out4; /* not a mountpoint */ |
676da58d | 3791 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3792 | goto out4; /* not attached */ |
2d8f3038 | 3793 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3794 | goto out4; /* not a mountpoint */ |
676da58d | 3795 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3796 | goto out4; /* not attached */ |
4ac91378 | 3797 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3798 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3799 | goto out4; |
0d082601 EB |
3800 | /* make certain new is below the root */ |
3801 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3802 | goto out4; | |
719ea2fb | 3803 | lock_mount_hash(); |
2763d119 AV |
3804 | umount_mnt(new_mnt); |
3805 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3806 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3807 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3808 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3809 | } | |
4ac91378 | 3810 | /* mount old root on put_old */ |
84d17192 | 3811 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3812 | /* mount new_root on / */ |
2763d119 AV |
3813 | attach_mnt(new_mnt, root_parent, root_mp); |
3814 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3815 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3816 | /* A moved mount should not expire automatically */ |
3817 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3818 | put_mountpoint(root_mp); |
719ea2fb | 3819 | unlock_mount_hash(); |
2d8f3038 | 3820 | chroot_fs_refs(&root, &new); |
1da177e4 | 3821 | error = 0; |
b12cea91 | 3822 | out4: |
84d17192 | 3823 | unlock_mount(old_mp); |
2763d119 AV |
3824 | if (!error) |
3825 | mntput_no_expire(ex_parent); | |
b12cea91 | 3826 | out3: |
8c3ee42e | 3827 | path_put(&root); |
b12cea91 | 3828 | out2: |
2d8f3038 | 3829 | path_put(&old); |
1da177e4 | 3830 | out1: |
2d8f3038 | 3831 | path_put(&new); |
1da177e4 | 3832 | out0: |
1da177e4 | 3833 | return error; |
1da177e4 LT |
3834 | } |
3835 | ||
2a186721 CB |
3836 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
3837 | { | |
3838 | unsigned int flags = mnt->mnt.mnt_flags; | |
3839 | ||
3840 | /* flags to clear */ | |
3841 | flags &= ~kattr->attr_clr; | |
3842 | /* flags to raise */ | |
3843 | flags |= kattr->attr_set; | |
3844 | ||
3845 | return flags; | |
3846 | } | |
3847 | ||
3848 | static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, | |
3849 | struct mount *mnt, int *err) | |
3850 | { | |
3851 | struct mount *m = mnt, *last = NULL; | |
3852 | ||
3853 | if (!is_mounted(&m->mnt)) { | |
3854 | *err = -EINVAL; | |
3855 | goto out; | |
3856 | } | |
3857 | ||
3858 | if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { | |
3859 | *err = -EINVAL; | |
3860 | goto out; | |
3861 | } | |
3862 | ||
3863 | do { | |
3864 | unsigned int flags; | |
3865 | ||
3866 | flags = recalc_flags(kattr, m); | |
3867 | if (!can_change_locked_flags(m, flags)) { | |
3868 | *err = -EPERM; | |
3869 | goto out; | |
3870 | } | |
3871 | ||
3872 | last = m; | |
3873 | ||
3874 | if ((kattr->attr_set & MNT_READONLY) && | |
3875 | !(m->mnt.mnt_flags & MNT_READONLY)) { | |
3876 | *err = mnt_hold_writers(m); | |
3877 | if (*err) | |
3878 | goto out; | |
3879 | } | |
3880 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3881 | ||
3882 | out: | |
3883 | return last; | |
3884 | } | |
3885 | ||
3886 | static void mount_setattr_commit(struct mount_kattr *kattr, | |
3887 | struct mount *mnt, struct mount *last, | |
3888 | int err) | |
3889 | { | |
3890 | struct mount *m = mnt; | |
3891 | ||
3892 | do { | |
3893 | if (!err) { | |
3894 | unsigned int flags; | |
3895 | ||
3896 | flags = recalc_flags(kattr, m); | |
3897 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
3898 | } | |
3899 | ||
3900 | /* | |
3901 | * We either set MNT_READONLY above so make it visible | |
3902 | * before ~MNT_WRITE_HOLD or we failed to recursively | |
3903 | * apply mount options. | |
3904 | */ | |
3905 | if ((kattr->attr_set & MNT_READONLY) && | |
3906 | (m->mnt.mnt_flags & MNT_WRITE_HOLD)) | |
3907 | mnt_unhold_writers(m); | |
3908 | ||
3909 | if (!err && kattr->propagation) | |
3910 | change_mnt_propagation(m, kattr->propagation); | |
3911 | ||
3912 | /* | |
3913 | * On failure, only cleanup until we found the first mount | |
3914 | * we failed to handle. | |
3915 | */ | |
3916 | if (err && m == last) | |
3917 | break; | |
3918 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3919 | ||
3920 | if (!err) | |
3921 | touch_mnt_namespace(mnt->mnt_ns); | |
3922 | } | |
3923 | ||
3924 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
3925 | { | |
3926 | struct mount *mnt = real_mount(path->mnt), *last = NULL; | |
3927 | int err = 0; | |
3928 | ||
3929 | if (path->dentry != mnt->mnt.mnt_root) | |
3930 | return -EINVAL; | |
3931 | ||
3932 | if (kattr->propagation) { | |
3933 | /* | |
3934 | * Only take namespace_lock() if we're actually changing | |
3935 | * propagation. | |
3936 | */ | |
3937 | namespace_lock(); | |
3938 | if (kattr->propagation == MS_SHARED) { | |
3939 | err = invent_group_ids(mnt, kattr->recurse); | |
3940 | if (err) { | |
3941 | namespace_unlock(); | |
3942 | return err; | |
3943 | } | |
3944 | } | |
3945 | } | |
3946 | ||
3947 | lock_mount_hash(); | |
3948 | ||
3949 | /* | |
3950 | * Get the mount tree in a shape where we can change mount | |
3951 | * properties without failure. | |
3952 | */ | |
3953 | last = mount_setattr_prepare(kattr, mnt, &err); | |
3954 | if (last) /* Commit all changes or revert to the old state. */ | |
3955 | mount_setattr_commit(kattr, mnt, last, err); | |
3956 | ||
3957 | unlock_mount_hash(); | |
3958 | ||
3959 | if (kattr->propagation) { | |
3960 | namespace_unlock(); | |
3961 | if (err) | |
3962 | cleanup_group_ids(mnt, NULL); | |
3963 | } | |
3964 | ||
3965 | return err; | |
3966 | } | |
3967 | ||
3968 | static int build_mount_kattr(const struct mount_attr *attr, | |
3969 | struct mount_kattr *kattr, unsigned int flags) | |
3970 | { | |
3971 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
3972 | ||
3973 | if (flags & AT_NO_AUTOMOUNT) | |
3974 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
3975 | if (flags & AT_SYMLINK_NOFOLLOW) | |
3976 | lookup_flags &= ~LOOKUP_FOLLOW; | |
3977 | if (flags & AT_EMPTY_PATH) | |
3978 | lookup_flags |= LOOKUP_EMPTY; | |
3979 | ||
3980 | *kattr = (struct mount_kattr) { | |
3981 | .lookup_flags = lookup_flags, | |
3982 | .recurse = !!(flags & AT_RECURSIVE), | |
3983 | }; | |
3984 | ||
3985 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
3986 | return -EINVAL; | |
3987 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
3988 | return -EINVAL; | |
3989 | kattr->propagation = attr->propagation; | |
3990 | ||
3991 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
3992 | return -EINVAL; | |
3993 | ||
3994 | if (attr->userns_fd) | |
3995 | return -EINVAL; | |
3996 | ||
3997 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); | |
3998 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
3999 | ||
4000 | /* | |
4001 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4002 | * users wanting to transition to a different atime setting cannot | |
4003 | * simply specify the atime setting in @attr_set, but must also | |
4004 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4005 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4006 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4007 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4008 | */ | |
4009 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4010 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4011 | return -EINVAL; | |
4012 | ||
4013 | /* | |
4014 | * Clear all previous time settings as they are mutually | |
4015 | * exclusive. | |
4016 | */ | |
4017 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4018 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4019 | case MOUNT_ATTR_RELATIME: | |
4020 | kattr->attr_set |= MNT_RELATIME; | |
4021 | break; | |
4022 | case MOUNT_ATTR_NOATIME: | |
4023 | kattr->attr_set |= MNT_NOATIME; | |
4024 | break; | |
4025 | case MOUNT_ATTR_STRICTATIME: | |
4026 | break; | |
4027 | default: | |
4028 | return -EINVAL; | |
4029 | } | |
4030 | } else { | |
4031 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4032 | return -EINVAL; | |
4033 | } | |
4034 | ||
4035 | return 0; | |
4036 | } | |
4037 | ||
4038 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4039 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4040 | size_t, usize) | |
4041 | { | |
4042 | int err; | |
4043 | struct path target; | |
4044 | struct mount_attr attr; | |
4045 | struct mount_kattr kattr; | |
4046 | ||
4047 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4048 | ||
4049 | if (flags & ~(AT_EMPTY_PATH | | |
4050 | AT_RECURSIVE | | |
4051 | AT_SYMLINK_NOFOLLOW | | |
4052 | AT_NO_AUTOMOUNT)) | |
4053 | return -EINVAL; | |
4054 | ||
4055 | if (unlikely(usize > PAGE_SIZE)) | |
4056 | return -E2BIG; | |
4057 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4058 | return -EINVAL; | |
4059 | ||
4060 | if (!may_mount()) | |
4061 | return -EPERM; | |
4062 | ||
4063 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4064 | if (err) | |
4065 | return err; | |
4066 | ||
4067 | /* Don't bother walking through the mounts if this is a nop. */ | |
4068 | if (attr.attr_set == 0 && | |
4069 | attr.attr_clr == 0 && | |
4070 | attr.propagation == 0) | |
4071 | return 0; | |
4072 | ||
4073 | err = build_mount_kattr(&attr, &kattr, flags); | |
4074 | if (err) | |
4075 | return err; | |
4076 | ||
4077 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
4078 | if (err) | |
4079 | return err; | |
4080 | ||
4081 | err = do_mount_setattr(&target, &kattr); | |
4082 | path_put(&target); | |
4083 | return err; | |
4084 | } | |
4085 | ||
1da177e4 LT |
4086 | static void __init init_mount_tree(void) |
4087 | { | |
4088 | struct vfsmount *mnt; | |
74e83122 | 4089 | struct mount *m; |
6b3286ed | 4090 | struct mnt_namespace *ns; |
ac748a09 | 4091 | struct path root; |
1da177e4 | 4092 | |
fd3e007f | 4093 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4094 | if (IS_ERR(mnt)) |
4095 | panic("Can't create rootfs"); | |
b3e19d92 | 4096 | |
74e83122 | 4097 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4098 | if (IS_ERR(ns)) |
1da177e4 | 4099 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4100 | m = real_mount(mnt); |
4101 | m->mnt_ns = ns; | |
4102 | ns->root = m; | |
4103 | ns->mounts = 1; | |
4104 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4105 | init_task.nsproxy->mnt_ns = ns; |
4106 | get_mnt_ns(ns); | |
4107 | ||
be08d6d2 AV |
4108 | root.mnt = mnt; |
4109 | root.dentry = mnt->mnt_root; | |
da362b09 | 4110 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4111 | |
4112 | set_fs_pwd(current->fs, &root); | |
4113 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4114 | } |
4115 | ||
74bf17cf | 4116 | void __init mnt_init(void) |
1da177e4 | 4117 | { |
15a67dd8 | 4118 | int err; |
1da177e4 | 4119 | |
7d6fec45 | 4120 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 4121 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 4122 | |
0818bf27 | 4123 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4124 | sizeof(struct hlist_head), |
0818bf27 | 4125 | mhash_entries, 19, |
3d375d78 | 4126 | HASH_ZERO, |
0818bf27 AV |
4127 | &m_hash_shift, &m_hash_mask, 0, 0); |
4128 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4129 | sizeof(struct hlist_head), | |
4130 | mphash_entries, 19, | |
3d375d78 | 4131 | HASH_ZERO, |
0818bf27 | 4132 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4133 | |
84d17192 | 4134 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4135 | panic("Failed to allocate mount hash table\n"); |
4136 | ||
4b93dc9b TH |
4137 | kernfs_init(); |
4138 | ||
15a67dd8 RD |
4139 | err = sysfs_init(); |
4140 | if (err) | |
4141 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4142 | __func__, err); |
00d26666 GKH |
4143 | fs_kobj = kobject_create_and_add("fs", NULL); |
4144 | if (!fs_kobj) | |
8e24eea7 | 4145 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4146 | shmem_init(); |
1da177e4 LT |
4147 | init_rootfs(); |
4148 | init_mount_tree(); | |
4149 | } | |
4150 | ||
616511d0 | 4151 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4152 | { |
1a7b8969 | 4153 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4154 | return; |
7b00ed6f | 4155 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4156 | free_mnt_ns(ns); |
1da177e4 | 4157 | } |
9d412a43 | 4158 | |
d911b458 | 4159 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4160 | { |
423e0ab0 | 4161 | struct vfsmount *mnt; |
d911b458 | 4162 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4163 | if (!IS_ERR(mnt)) { |
4164 | /* | |
4165 | * it is a longterm mount, don't release mnt until | |
4166 | * we unmount before file sys is unregistered | |
4167 | */ | |
f7a99c5b | 4168 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4169 | } |
4170 | return mnt; | |
9d412a43 | 4171 | } |
d911b458 | 4172 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4173 | |
4174 | void kern_unmount(struct vfsmount *mnt) | |
4175 | { | |
4176 | /* release long term mount so mount point can be released */ | |
4177 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 4178 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 4179 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4180 | mntput(mnt); |
4181 | } | |
4182 | } | |
4183 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4184 | |
df820f8d MS |
4185 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4186 | { | |
4187 | unsigned int i; | |
4188 | ||
4189 | for (i = 0; i < num; i++) | |
4190 | if (mnt[i]) | |
4191 | real_mount(mnt[i])->mnt_ns = NULL; | |
4192 | synchronize_rcu_expedited(); | |
4193 | for (i = 0; i < num; i++) | |
4194 | mntput(mnt[i]); | |
4195 | } | |
4196 | EXPORT_SYMBOL(kern_unmount_array); | |
4197 | ||
02125a82 AV |
4198 | bool our_mnt(struct vfsmount *mnt) |
4199 | { | |
143c8c91 | 4200 | return check_mnt(real_mount(mnt)); |
02125a82 | 4201 | } |
8823c079 | 4202 | |
3151527e EB |
4203 | bool current_chrooted(void) |
4204 | { | |
4205 | /* Does the current process have a non-standard root */ | |
4206 | struct path ns_root; | |
4207 | struct path fs_root; | |
4208 | bool chrooted; | |
4209 | ||
4210 | /* Find the namespace root */ | |
4211 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4212 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4213 | path_get(&ns_root); | |
4214 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4215 | ; | |
4216 | ||
4217 | get_fs_root(current->fs, &fs_root); | |
4218 | ||
4219 | chrooted = !path_equal(&fs_root, &ns_root); | |
4220 | ||
4221 | path_put(&fs_root); | |
4222 | path_put(&ns_root); | |
4223 | ||
4224 | return chrooted; | |
4225 | } | |
4226 | ||
132e4608 DH |
4227 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4228 | const struct super_block *sb, | |
8654df4e | 4229 | int *new_mnt_flags) |
87a8ebd6 | 4230 | { |
8c6cf9cc | 4231 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4232 | struct mount *mnt; |
e51db735 | 4233 | bool visible = false; |
87a8ebd6 | 4234 | |
44bb4385 | 4235 | down_read(&namespace_sem); |
9f6c61f9 | 4236 | lock_ns_list(ns); |
87a8ebd6 | 4237 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4238 | struct mount *child; |
77b1a97d EB |
4239 | int mnt_flags; |
4240 | ||
9f6c61f9 MS |
4241 | if (mnt_is_cursor(mnt)) |
4242 | continue; | |
4243 | ||
132e4608 | 4244 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4245 | continue; |
4246 | ||
7e96c1b0 EB |
4247 | /* This mount is not fully visible if it's root directory |
4248 | * is not the root directory of the filesystem. | |
4249 | */ | |
4250 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4251 | continue; | |
4252 | ||
a1935c17 | 4253 | /* A local view of the mount flags */ |
77b1a97d | 4254 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4255 | |
695e9df0 | 4256 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4257 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4258 | mnt_flags |= MNT_LOCK_READONLY; |
4259 | ||
8c6cf9cc EB |
4260 | /* Verify the mount flags are equal to or more permissive |
4261 | * than the proposed new mount. | |
4262 | */ | |
77b1a97d | 4263 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4264 | !(new_flags & MNT_READONLY)) |
4265 | continue; | |
77b1a97d EB |
4266 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4267 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4268 | continue; |
4269 | ||
ceeb0e5d EB |
4270 | /* This mount is not fully visible if there are any |
4271 | * locked child mounts that cover anything except for | |
4272 | * empty directories. | |
e51db735 EB |
4273 | */ |
4274 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4275 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4276 | /* Only worry about locked mounts */ |
d71ed6c9 | 4277 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4278 | continue; |
7236c85e EB |
4279 | /* Is the directory permanetly empty? */ |
4280 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4281 | goto next; |
87a8ebd6 | 4282 | } |
8c6cf9cc | 4283 | /* Preserve the locked attributes */ |
77b1a97d | 4284 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4285 | MNT_LOCK_ATIME); |
e51db735 EB |
4286 | visible = true; |
4287 | goto found; | |
4288 | next: ; | |
87a8ebd6 | 4289 | } |
e51db735 | 4290 | found: |
9f6c61f9 | 4291 | unlock_ns_list(ns); |
44bb4385 | 4292 | up_read(&namespace_sem); |
e51db735 | 4293 | return visible; |
87a8ebd6 EB |
4294 | } |
4295 | ||
132e4608 | 4296 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4297 | { |
a1935c17 | 4298 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4299 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4300 | unsigned long s_iflags; | |
4301 | ||
4302 | if (ns->user_ns == &init_user_ns) | |
4303 | return false; | |
4304 | ||
4305 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4306 | s_iflags = sb->s_iflags; |
8654df4e EB |
4307 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4308 | return false; | |
4309 | ||
a1935c17 EB |
4310 | if ((s_iflags & required_iflags) != required_iflags) { |
4311 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4312 | required_iflags); | |
4313 | return true; | |
4314 | } | |
4315 | ||
132e4608 | 4316 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4317 | } |
4318 | ||
380cf5ba AL |
4319 | bool mnt_may_suid(struct vfsmount *mnt) |
4320 | { | |
4321 | /* | |
4322 | * Foreign mounts (accessed via fchdir or through /proc | |
4323 | * symlinks) are always treated as if they are nosuid. This | |
4324 | * prevents namespaces from trusting potentially unsafe | |
4325 | * suid/sgid bits, file caps, or security labels that originate | |
4326 | * in other namespaces. | |
4327 | */ | |
4328 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4329 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4330 | } | |
4331 | ||
64964528 | 4332 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4333 | { |
58be2825 | 4334 | struct ns_common *ns = NULL; |
8823c079 EB |
4335 | struct nsproxy *nsproxy; |
4336 | ||
728dba3a EB |
4337 | task_lock(task); |
4338 | nsproxy = task->nsproxy; | |
8823c079 | 4339 | if (nsproxy) { |
58be2825 AV |
4340 | ns = &nsproxy->mnt_ns->ns; |
4341 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4342 | } |
728dba3a | 4343 | task_unlock(task); |
8823c079 EB |
4344 | |
4345 | return ns; | |
4346 | } | |
4347 | ||
64964528 | 4348 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4349 | { |
58be2825 | 4350 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4351 | } |
4352 | ||
f2a8d52e | 4353 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4354 | { |
f2a8d52e CB |
4355 | struct nsproxy *nsproxy = nsset->nsproxy; |
4356 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4357 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4358 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4359 | struct path root; |
4f757f3c | 4360 | int err; |
8823c079 | 4361 | |
0c55cfc4 | 4362 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4363 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4364 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4365 | return -EPERM; |
8823c079 | 4366 | |
74e83122 AV |
4367 | if (is_anon_ns(mnt_ns)) |
4368 | return -EINVAL; | |
4369 | ||
8823c079 EB |
4370 | if (fs->users != 1) |
4371 | return -EINVAL; | |
4372 | ||
4373 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4374 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4375 | nsproxy->mnt_ns = mnt_ns; |
4376 | ||
4377 | /* Find the root */ | |
4f757f3c AV |
4378 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4379 | "/", LOOKUP_DOWN, &root); | |
4380 | if (err) { | |
4381 | /* revert to old namespace */ | |
4382 | nsproxy->mnt_ns = old_mnt_ns; | |
4383 | put_mnt_ns(mnt_ns); | |
4384 | return err; | |
4385 | } | |
8823c079 | 4386 | |
4068367c AV |
4387 | put_mnt_ns(old_mnt_ns); |
4388 | ||
8823c079 EB |
4389 | /* Update the pwd and root */ |
4390 | set_fs_pwd(fs, &root); | |
4391 | set_fs_root(fs, &root); | |
4392 | ||
4393 | path_put(&root); | |
4394 | return 0; | |
4395 | } | |
4396 | ||
bcac25a5 AV |
4397 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4398 | { | |
4399 | return to_mnt_ns(ns)->user_ns; | |
4400 | } | |
4401 | ||
8823c079 EB |
4402 | const struct proc_ns_operations mntns_operations = { |
4403 | .name = "mnt", | |
4404 | .type = CLONE_NEWNS, | |
4405 | .get = mntns_get, | |
4406 | .put = mntns_put, | |
4407 | .install = mntns_install, | |
bcac25a5 | 4408 | .owner = mntns_owner, |
8823c079 | 4409 | }; |