]>
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> |
9caccd41 | 28 | #include <linux/proc_fs.h> |
9ea459e1 | 29 | #include <linux/task_work.h> |
9164bb4a | 30 | #include <linux/sched/task.h> |
e262e32d | 31 | #include <uapi/linux/mount.h> |
9bc61ab1 | 32 | #include <linux/fs_context.h> |
037f11b4 | 33 | #include <linux/shmem_fs.h> |
9164bb4a | 34 | |
07b20889 | 35 | #include "pnode.h" |
948730b0 | 36 | #include "internal.h" |
1da177e4 | 37 | |
d2921684 EB |
38 | /* Maximum number of mounts in a mount namespace */ |
39 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
40 | ||
0818bf27 AV |
41 | static unsigned int m_hash_mask __read_mostly; |
42 | static unsigned int m_hash_shift __read_mostly; | |
43 | static unsigned int mp_hash_mask __read_mostly; | |
44 | static unsigned int mp_hash_shift __read_mostly; | |
45 | ||
46 | static __initdata unsigned long mhash_entries; | |
47 | static int __init set_mhash_entries(char *str) | |
48 | { | |
49 | if (!str) | |
50 | return 0; | |
51 | mhash_entries = simple_strtoul(str, &str, 0); | |
52 | return 1; | |
53 | } | |
54 | __setup("mhash_entries=", set_mhash_entries); | |
55 | ||
56 | static __initdata unsigned long mphash_entries; | |
57 | static int __init set_mphash_entries(char *str) | |
58 | { | |
59 | if (!str) | |
60 | return 0; | |
61 | mphash_entries = simple_strtoul(str, &str, 0); | |
62 | return 1; | |
63 | } | |
64 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 65 | |
c7999c36 | 66 | static u64 event; |
73cd49ec | 67 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 68 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 69 | |
38129a13 | 70 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 71 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 72 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 73 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
74 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
75 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 76 | |
2a186721 CB |
77 | struct mount_kattr { |
78 | unsigned int attr_set; | |
79 | unsigned int attr_clr; | |
80 | unsigned int propagation; | |
81 | unsigned int lookup_flags; | |
82 | bool recurse; | |
9caccd41 | 83 | struct user_namespace *mnt_userns; |
2a186721 CB |
84 | }; |
85 | ||
f87fd4c2 | 86 | /* /sys/fs */ |
00d26666 GKH |
87 | struct kobject *fs_kobj; |
88 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 89 | |
99b7db7b NP |
90 | /* |
91 | * vfsmount lock may be taken for read to prevent changes to the | |
92 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
93 | * up the tree. | |
94 | * | |
95 | * It should be taken for write in all cases where the vfsmount | |
96 | * tree or hash is modified or when a vfsmount structure is modified. | |
97 | */ | |
48a066e7 | 98 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 99 | |
d033cb67 CB |
100 | static inline void lock_mount_hash(void) |
101 | { | |
102 | write_seqlock(&mount_lock); | |
103 | } | |
104 | ||
105 | static inline void unlock_mount_hash(void) | |
106 | { | |
107 | write_sequnlock(&mount_lock); | |
108 | } | |
109 | ||
38129a13 | 110 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 111 | { |
b58fed8b RP |
112 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
113 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
114 | tmp = tmp + (tmp >> m_hash_shift); |
115 | return &mount_hashtable[tmp & m_hash_mask]; | |
116 | } | |
117 | ||
118 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
119 | { | |
120 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
121 | tmp = tmp + (tmp >> mp_hash_shift); | |
122 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
123 | } |
124 | ||
b105e270 | 125 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 126 | { |
169b480e MW |
127 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
128 | ||
129 | if (res < 0) | |
130 | return res; | |
131 | mnt->mnt_id = res; | |
132 | return 0; | |
73cd49ec MS |
133 | } |
134 | ||
b105e270 | 135 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 136 | { |
169b480e | 137 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
138 | } |
139 | ||
719f5d7f MS |
140 | /* |
141 | * Allocate a new peer group ID | |
719f5d7f | 142 | */ |
4b8b21f4 | 143 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 144 | { |
169b480e | 145 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 146 | |
169b480e MW |
147 | if (res < 0) |
148 | return res; | |
149 | mnt->mnt_group_id = res; | |
150 | return 0; | |
719f5d7f MS |
151 | } |
152 | ||
153 | /* | |
154 | * Release a peer group ID | |
155 | */ | |
4b8b21f4 | 156 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 157 | { |
169b480e | 158 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 159 | mnt->mnt_group_id = 0; |
719f5d7f MS |
160 | } |
161 | ||
b3e19d92 NP |
162 | /* |
163 | * vfsmount lock must be held for read | |
164 | */ | |
83adc753 | 165 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
166 | { |
167 | #ifdef CONFIG_SMP | |
68e8a9fe | 168 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
169 | #else |
170 | preempt_disable(); | |
68e8a9fe | 171 | mnt->mnt_count += n; |
b3e19d92 NP |
172 | preempt_enable(); |
173 | #endif | |
174 | } | |
175 | ||
b3e19d92 NP |
176 | /* |
177 | * vfsmount lock must be held for write | |
178 | */ | |
edf7ddbf | 179 | int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
180 | { |
181 | #ifdef CONFIG_SMP | |
edf7ddbf | 182 | int count = 0; |
b3e19d92 NP |
183 | int cpu; |
184 | ||
185 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 186 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
187 | } |
188 | ||
189 | return count; | |
190 | #else | |
68e8a9fe | 191 | return mnt->mnt_count; |
b3e19d92 NP |
192 | #endif |
193 | } | |
194 | ||
b105e270 | 195 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 196 | { |
c63181e6 AV |
197 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
198 | if (mnt) { | |
73cd49ec MS |
199 | int err; |
200 | ||
c63181e6 | 201 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
202 | if (err) |
203 | goto out_free_cache; | |
204 | ||
205 | if (name) { | |
fcc139ae | 206 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 207 | if (!mnt->mnt_devname) |
88b38782 | 208 | goto out_free_id; |
73cd49ec MS |
209 | } |
210 | ||
b3e19d92 | 211 | #ifdef CONFIG_SMP |
c63181e6 AV |
212 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
213 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
214 | goto out_free_devname; |
215 | ||
c63181e6 | 216 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 217 | #else |
c63181e6 AV |
218 | mnt->mnt_count = 1; |
219 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
220 | #endif |
221 | ||
38129a13 | 222 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
223 | INIT_LIST_HEAD(&mnt->mnt_child); |
224 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
225 | INIT_LIST_HEAD(&mnt->mnt_list); | |
226 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
227 | INIT_LIST_HEAD(&mnt->mnt_share); | |
228 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
229 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 230 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 231 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
56cbb429 | 232 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
a6435940 | 233 | mnt->mnt.mnt_userns = &init_user_ns; |
1da177e4 | 234 | } |
c63181e6 | 235 | return mnt; |
88b38782 | 236 | |
d3ef3d73 NP |
237 | #ifdef CONFIG_SMP |
238 | out_free_devname: | |
fcc139ae | 239 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 240 | #endif |
88b38782 | 241 | out_free_id: |
c63181e6 | 242 | mnt_free_id(mnt); |
88b38782 | 243 | out_free_cache: |
c63181e6 | 244 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 245 | return NULL; |
1da177e4 LT |
246 | } |
247 | ||
3d733633 DH |
248 | /* |
249 | * Most r/o checks on a fs are for operations that take | |
250 | * discrete amounts of time, like a write() or unlink(). | |
251 | * We must keep track of when those operations start | |
252 | * (for permission checks) and when they end, so that | |
253 | * we can determine when writes are able to occur to | |
254 | * a filesystem. | |
255 | */ | |
256 | /* | |
257 | * __mnt_is_readonly: check whether a mount is read-only | |
258 | * @mnt: the mount to check for its write status | |
259 | * | |
260 | * This shouldn't be used directly ouside of the VFS. | |
261 | * It does not guarantee that the filesystem will stay | |
262 | * r/w, just that it is right *now*. This can not and | |
263 | * should not be used in place of IS_RDONLY(inode). | |
264 | * mnt_want/drop_write() will _keep_ the filesystem | |
265 | * r/w. | |
266 | */ | |
43f5e655 | 267 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 268 | { |
43f5e655 | 269 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
270 | } |
271 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
272 | ||
83adc753 | 273 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
274 | { |
275 | #ifdef CONFIG_SMP | |
68e8a9fe | 276 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 277 | #else |
68e8a9fe | 278 | mnt->mnt_writers++; |
d3ef3d73 NP |
279 | #endif |
280 | } | |
3d733633 | 281 | |
83adc753 | 282 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 283 | { |
d3ef3d73 | 284 | #ifdef CONFIG_SMP |
68e8a9fe | 285 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 286 | #else |
68e8a9fe | 287 | mnt->mnt_writers--; |
d3ef3d73 | 288 | #endif |
3d733633 | 289 | } |
3d733633 | 290 | |
83adc753 | 291 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 292 | { |
d3ef3d73 NP |
293 | #ifdef CONFIG_SMP |
294 | unsigned int count = 0; | |
3d733633 | 295 | int cpu; |
3d733633 DH |
296 | |
297 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 298 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 299 | } |
3d733633 | 300 | |
d3ef3d73 NP |
301 | return count; |
302 | #else | |
303 | return mnt->mnt_writers; | |
304 | #endif | |
3d733633 DH |
305 | } |
306 | ||
4ed5e82f MS |
307 | static int mnt_is_readonly(struct vfsmount *mnt) |
308 | { | |
309 | if (mnt->mnt_sb->s_readonly_remount) | |
310 | return 1; | |
311 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
312 | smp_rmb(); | |
313 | return __mnt_is_readonly(mnt); | |
314 | } | |
315 | ||
8366025e | 316 | /* |
eb04c282 JK |
317 | * Most r/o & frozen checks on a fs are for operations that take discrete |
318 | * amounts of time, like a write() or unlink(). We must keep track of when | |
319 | * those operations start (for permission checks) and when they end, so that we | |
320 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
321 | */ |
322 | /** | |
eb04c282 | 323 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 324 | * @m: the mount on which to take a write |
8366025e | 325 | * |
eb04c282 JK |
326 | * This tells the low-level filesystem that a write is about to be performed to |
327 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
328 | * returning success. This operation does not protect against filesystem being | |
329 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
330 | * called. This is effectively a refcount. | |
8366025e | 331 | */ |
eb04c282 | 332 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 333 | { |
83adc753 | 334 | struct mount *mnt = real_mount(m); |
3d733633 | 335 | int ret = 0; |
3d733633 | 336 | |
d3ef3d73 | 337 | preempt_disable(); |
c6653a83 | 338 | mnt_inc_writers(mnt); |
d3ef3d73 | 339 | /* |
c6653a83 | 340 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
341 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
342 | * incremented count after it has set MNT_WRITE_HOLD. | |
343 | */ | |
344 | smp_mb(); | |
6aa7de05 | 345 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 NP |
346 | cpu_relax(); |
347 | /* | |
348 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
349 | * be set to match its requirements. So we must not load that until | |
350 | * MNT_WRITE_HOLD is cleared. | |
351 | */ | |
352 | smp_rmb(); | |
4ed5e82f | 353 | if (mnt_is_readonly(m)) { |
c6653a83 | 354 | mnt_dec_writers(mnt); |
3d733633 | 355 | ret = -EROFS; |
3d733633 | 356 | } |
d3ef3d73 | 357 | preempt_enable(); |
eb04c282 JK |
358 | |
359 | return ret; | |
360 | } | |
361 | ||
362 | /** | |
363 | * mnt_want_write - get write access to a mount | |
364 | * @m: the mount on which to take a write | |
365 | * | |
366 | * This tells the low-level filesystem that a write is about to be performed to | |
367 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
368 | * is not frozen) before returning success. When the write operation is | |
369 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
370 | */ | |
371 | int mnt_want_write(struct vfsmount *m) | |
372 | { | |
373 | int ret; | |
374 | ||
375 | sb_start_write(m->mnt_sb); | |
376 | ret = __mnt_want_write(m); | |
377 | if (ret) | |
378 | sb_end_write(m->mnt_sb); | |
3d733633 | 379 | return ret; |
8366025e DH |
380 | } |
381 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
382 | ||
96029c4e | 383 | /** |
eb04c282 | 384 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
385 | * @file: the file who's mount on which to take a write |
386 | * | |
14e43bf4 EB |
387 | * This is like __mnt_want_write, but if the file is already open for writing it |
388 | * skips incrementing mnt_writers (since the open file already has a reference) | |
389 | * and instead only does the check for emergency r/o remounts. This must be | |
390 | * paired with __mnt_drop_write_file. | |
96029c4e | 391 | */ |
eb04c282 | 392 | int __mnt_want_write_file(struct file *file) |
96029c4e | 393 | { |
14e43bf4 EB |
394 | if (file->f_mode & FMODE_WRITER) { |
395 | /* | |
396 | * Superblock may have become readonly while there are still | |
397 | * writable fd's, e.g. due to a fs error with errors=remount-ro | |
398 | */ | |
399 | if (__mnt_is_readonly(file->f_path.mnt)) | |
400 | return -EROFS; | |
401 | return 0; | |
402 | } | |
403 | return __mnt_want_write(file->f_path.mnt); | |
96029c4e | 404 | } |
eb04c282 | 405 | |
7c6893e3 MS |
406 | /** |
407 | * mnt_want_write_file - get write access to a file's mount | |
408 | * @file: the file who's mount on which to take a write | |
409 | * | |
14e43bf4 EB |
410 | * This is like mnt_want_write, but if the file is already open for writing it |
411 | * skips incrementing mnt_writers (since the open file already has a reference) | |
412 | * and instead only does the freeze protection and the check for emergency r/o | |
413 | * remounts. This must be paired with mnt_drop_write_file. | |
7c6893e3 MS |
414 | */ |
415 | int mnt_want_write_file(struct file *file) | |
416 | { | |
417 | int ret; | |
418 | ||
a6795a58 | 419 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
420 | ret = __mnt_want_write_file(file); |
421 | if (ret) | |
a6795a58 | 422 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
423 | return ret; |
424 | } | |
96029c4e NP |
425 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
426 | ||
8366025e | 427 | /** |
eb04c282 | 428 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
429 | * @mnt: the mount on which to give up write access |
430 | * | |
431 | * Tells the low-level filesystem that we are done | |
432 | * performing writes to it. Must be matched with | |
eb04c282 | 433 | * __mnt_want_write() call above. |
8366025e | 434 | */ |
eb04c282 | 435 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 436 | { |
d3ef3d73 | 437 | preempt_disable(); |
83adc753 | 438 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 439 | preempt_enable(); |
8366025e | 440 | } |
eb04c282 JK |
441 | |
442 | /** | |
443 | * mnt_drop_write - give up write access to a mount | |
444 | * @mnt: the mount on which to give up write access | |
445 | * | |
446 | * Tells the low-level filesystem that we are done performing writes to it and | |
447 | * also allows filesystem to be frozen again. Must be matched with | |
448 | * mnt_want_write() call above. | |
449 | */ | |
450 | void mnt_drop_write(struct vfsmount *mnt) | |
451 | { | |
452 | __mnt_drop_write(mnt); | |
453 | sb_end_write(mnt->mnt_sb); | |
454 | } | |
8366025e DH |
455 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
456 | ||
eb04c282 JK |
457 | void __mnt_drop_write_file(struct file *file) |
458 | { | |
14e43bf4 EB |
459 | if (!(file->f_mode & FMODE_WRITER)) |
460 | __mnt_drop_write(file->f_path.mnt); | |
eb04c282 JK |
461 | } |
462 | ||
7c6893e3 MS |
463 | void mnt_drop_write_file(struct file *file) |
464 | { | |
a6795a58 | 465 | __mnt_drop_write_file(file); |
7c6893e3 MS |
466 | sb_end_write(file_inode(file)->i_sb); |
467 | } | |
2a79f17e AV |
468 | EXPORT_SYMBOL(mnt_drop_write_file); |
469 | ||
fbdc2f6c | 470 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 471 | { |
83adc753 | 472 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 473 | /* |
d3ef3d73 NP |
474 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
475 | * should be visible before we do. | |
3d733633 | 476 | */ |
d3ef3d73 NP |
477 | smp_mb(); |
478 | ||
3d733633 | 479 | /* |
d3ef3d73 NP |
480 | * With writers on hold, if this value is zero, then there are |
481 | * definitely no active writers (although held writers may subsequently | |
482 | * increment the count, they'll have to wait, and decrement it after | |
483 | * seeing MNT_READONLY). | |
484 | * | |
485 | * It is OK to have counter incremented on one CPU and decremented on | |
486 | * another: the sum will add up correctly. The danger would be when we | |
487 | * sum up each counter, if we read a counter before it is incremented, | |
488 | * but then read another CPU's count which it has been subsequently | |
489 | * decremented from -- we would see more decrements than we should. | |
490 | * MNT_WRITE_HOLD protects against this scenario, because | |
491 | * mnt_want_write first increments count, then smp_mb, then spins on | |
492 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
493 | * we're counting up here. | |
3d733633 | 494 | */ |
c6653a83 | 495 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
496 | return -EBUSY; |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
501 | static inline void mnt_unhold_writers(struct mount *mnt) | |
502 | { | |
d3ef3d73 NP |
503 | /* |
504 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
505 | * that become unheld will see MNT_READONLY. | |
506 | */ | |
507 | smp_wmb(); | |
83adc753 | 508 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
509 | } |
510 | ||
511 | static int mnt_make_readonly(struct mount *mnt) | |
512 | { | |
513 | int ret; | |
514 | ||
515 | ret = mnt_hold_writers(mnt); | |
516 | if (!ret) | |
517 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
518 | mnt_unhold_writers(mnt); | |
3d733633 | 519 | return ret; |
8366025e | 520 | } |
8366025e | 521 | |
4ed5e82f MS |
522 | int sb_prepare_remount_readonly(struct super_block *sb) |
523 | { | |
524 | struct mount *mnt; | |
525 | int err = 0; | |
526 | ||
8e8b8796 MS |
527 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
528 | if (atomic_long_read(&sb->s_remove_count)) | |
529 | return -EBUSY; | |
530 | ||
719ea2fb | 531 | lock_mount_hash(); |
4ed5e82f MS |
532 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
533 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
534 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
535 | smp_mb(); | |
536 | if (mnt_get_writers(mnt) > 0) { | |
537 | err = -EBUSY; | |
538 | break; | |
539 | } | |
540 | } | |
541 | } | |
8e8b8796 MS |
542 | if (!err && atomic_long_read(&sb->s_remove_count)) |
543 | err = -EBUSY; | |
544 | ||
4ed5e82f MS |
545 | if (!err) { |
546 | sb->s_readonly_remount = 1; | |
547 | smp_wmb(); | |
548 | } | |
549 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
550 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
551 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
552 | } | |
719ea2fb | 553 | unlock_mount_hash(); |
4ed5e82f MS |
554 | |
555 | return err; | |
556 | } | |
557 | ||
b105e270 | 558 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 559 | { |
a6435940 CB |
560 | struct user_namespace *mnt_userns; |
561 | ||
562 | mnt_userns = mnt_user_ns(&mnt->mnt); | |
563 | if (mnt_userns != &init_user_ns) | |
564 | put_user_ns(mnt_userns); | |
fcc139ae | 565 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 566 | #ifdef CONFIG_SMP |
68e8a9fe | 567 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 568 | #endif |
b105e270 | 569 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
570 | } |
571 | ||
8ffcb32e DH |
572 | static void delayed_free_vfsmnt(struct rcu_head *head) |
573 | { | |
574 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
575 | } | |
576 | ||
48a066e7 | 577 | /* call under rcu_read_lock */ |
294d71ff | 578 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
579 | { |
580 | struct mount *mnt; | |
581 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 582 | return 1; |
48a066e7 | 583 | if (bastard == NULL) |
294d71ff | 584 | return 0; |
48a066e7 AV |
585 | mnt = real_mount(bastard); |
586 | mnt_add_count(mnt, 1); | |
119e1ef8 | 587 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 588 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 589 | return 0; |
48a066e7 AV |
590 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
591 | mnt_add_count(mnt, -1); | |
294d71ff AV |
592 | return 1; |
593 | } | |
119e1ef8 AV |
594 | lock_mount_hash(); |
595 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
596 | mnt_add_count(mnt, -1); | |
597 | unlock_mount_hash(); | |
598 | return 1; | |
599 | } | |
600 | unlock_mount_hash(); | |
601 | /* caller will mntput() */ | |
294d71ff AV |
602 | return -1; |
603 | } | |
604 | ||
605 | /* call under rcu_read_lock */ | |
606 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
607 | { | |
608 | int res = __legitimize_mnt(bastard, seq); | |
609 | if (likely(!res)) | |
610 | return true; | |
611 | if (unlikely(res < 0)) { | |
612 | rcu_read_unlock(); | |
613 | mntput(bastard); | |
614 | rcu_read_lock(); | |
48a066e7 | 615 | } |
48a066e7 AV |
616 | return false; |
617 | } | |
618 | ||
1da177e4 | 619 | /* |
474279dc | 620 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 621 | * call under rcu_read_lock() |
1da177e4 | 622 | */ |
474279dc | 623 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 624 | { |
38129a13 | 625 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
626 | struct mount *p; |
627 | ||
38129a13 | 628 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
629 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
630 | return p; | |
631 | return NULL; | |
632 | } | |
633 | ||
a05964f3 | 634 | /* |
f015f126 DH |
635 | * lookup_mnt - Return the first child mount mounted at path |
636 | * | |
637 | * "First" means first mounted chronologically. If you create the | |
638 | * following mounts: | |
639 | * | |
640 | * mount /dev/sda1 /mnt | |
641 | * mount /dev/sda2 /mnt | |
642 | * mount /dev/sda3 /mnt | |
643 | * | |
644 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
645 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
646 | * /dev/sda2, then /dev/sda3, then NULL. | |
647 | * | |
648 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 649 | */ |
ca71cf71 | 650 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 651 | { |
c7105365 | 652 | struct mount *child_mnt; |
48a066e7 AV |
653 | struct vfsmount *m; |
654 | unsigned seq; | |
99b7db7b | 655 | |
48a066e7 AV |
656 | rcu_read_lock(); |
657 | do { | |
658 | seq = read_seqbegin(&mount_lock); | |
659 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
660 | m = child_mnt ? &child_mnt->mnt : NULL; | |
661 | } while (!legitimize_mnt(m, seq)); | |
662 | rcu_read_unlock(); | |
663 | return m; | |
a05964f3 RP |
664 | } |
665 | ||
9f6c61f9 MS |
666 | static inline void lock_ns_list(struct mnt_namespace *ns) |
667 | { | |
668 | spin_lock(&ns->ns_lock); | |
669 | } | |
670 | ||
671 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
672 | { | |
673 | spin_unlock(&ns->ns_lock); | |
674 | } | |
675 | ||
676 | static inline bool mnt_is_cursor(struct mount *mnt) | |
677 | { | |
678 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
679 | } | |
680 | ||
7af1364f EB |
681 | /* |
682 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
683 | * current mount namespace. | |
684 | * | |
685 | * The common case is dentries are not mountpoints at all and that | |
686 | * test is handled inline. For the slow case when we are actually | |
687 | * dealing with a mountpoint of some kind, walk through all of the | |
688 | * mounts in the current mount namespace and test to see if the dentry | |
689 | * is a mountpoint. | |
690 | * | |
691 | * The mount_hashtable is not usable in the context because we | |
692 | * need to identify all mounts that may be in the current mount | |
693 | * namespace not just a mount that happens to have some specified | |
694 | * parent mount. | |
695 | */ | |
696 | bool __is_local_mountpoint(struct dentry *dentry) | |
697 | { | |
698 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
699 | struct mount *mnt; | |
700 | bool is_covered = false; | |
701 | ||
7af1364f | 702 | down_read(&namespace_sem); |
9f6c61f9 | 703 | lock_ns_list(ns); |
7af1364f | 704 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
705 | if (mnt_is_cursor(mnt)) |
706 | continue; | |
7af1364f EB |
707 | is_covered = (mnt->mnt_mountpoint == dentry); |
708 | if (is_covered) | |
709 | break; | |
710 | } | |
9f6c61f9 | 711 | unlock_ns_list(ns); |
7af1364f | 712 | up_read(&namespace_sem); |
5ad05cc8 | 713 | |
7af1364f EB |
714 | return is_covered; |
715 | } | |
716 | ||
e2dfa935 | 717 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 718 | { |
0818bf27 | 719 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
720 | struct mountpoint *mp; |
721 | ||
0818bf27 | 722 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 723 | if (mp->m_dentry == dentry) { |
84d17192 AV |
724 | mp->m_count++; |
725 | return mp; | |
726 | } | |
727 | } | |
e2dfa935 EB |
728 | return NULL; |
729 | } | |
730 | ||
3895dbf8 | 731 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 732 | { |
3895dbf8 | 733 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 734 | int ret; |
84d17192 | 735 | |
3895dbf8 | 736 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
737 | /* might be worth a WARN_ON() */ |
738 | if (d_unlinked(dentry)) | |
739 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
740 | mountpoint: |
741 | read_seqlock_excl(&mount_lock); | |
742 | mp = lookup_mountpoint(dentry); | |
743 | read_sequnlock_excl(&mount_lock); | |
744 | if (mp) | |
745 | goto done; | |
746 | } | |
747 | ||
748 | if (!new) | |
749 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
750 | if (!new) | |
84d17192 AV |
751 | return ERR_PTR(-ENOMEM); |
752 | ||
3895dbf8 EB |
753 | |
754 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 755 | ret = d_set_mounted(dentry); |
eed81007 | 756 | |
3895dbf8 EB |
757 | /* Someone else set d_mounted? */ |
758 | if (ret == -EBUSY) | |
759 | goto mountpoint; | |
760 | ||
761 | /* The dentry is not available as a mountpoint? */ | |
762 | mp = ERR_PTR(ret); | |
763 | if (ret) | |
764 | goto done; | |
765 | ||
766 | /* Add the new mountpoint to the hash table */ | |
767 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 768 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
769 | new->m_count = 1; |
770 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
771 | INIT_HLIST_HEAD(&new->m_list); | |
772 | read_sequnlock_excl(&mount_lock); | |
773 | ||
774 | mp = new; | |
775 | new = NULL; | |
776 | done: | |
777 | kfree(new); | |
84d17192 AV |
778 | return mp; |
779 | } | |
780 | ||
4edbe133 AV |
781 | /* |
782 | * vfsmount lock must be held. Additionally, the caller is responsible | |
783 | * for serializing calls for given disposal list. | |
784 | */ | |
785 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
786 | { |
787 | if (!--mp->m_count) { | |
788 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 789 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
790 | spin_lock(&dentry->d_lock); |
791 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
792 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 793 | dput_to_list(dentry, list); |
0818bf27 | 794 | hlist_del(&mp->m_hash); |
84d17192 AV |
795 | kfree(mp); |
796 | } | |
797 | } | |
798 | ||
4edbe133 AV |
799 | /* called with namespace_lock and vfsmount lock */ |
800 | static void put_mountpoint(struct mountpoint *mp) | |
801 | { | |
802 | __put_mountpoint(mp, &ex_mountpoints); | |
803 | } | |
804 | ||
143c8c91 | 805 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 806 | { |
6b3286ed | 807 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
808 | } |
809 | ||
99b7db7b NP |
810 | /* |
811 | * vfsmount lock must be held for write | |
812 | */ | |
6b3286ed | 813 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
814 | { |
815 | if (ns) { | |
816 | ns->event = ++event; | |
817 | wake_up_interruptible(&ns->poll); | |
818 | } | |
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 && ns->event != event) { | |
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 | */ | |
e4e59906 | 835 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 836 | { |
e4e59906 | 837 | struct mountpoint *mp; |
0714a533 | 838 | mnt->mnt_parent = mnt; |
a73324da | 839 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 840 | list_del_init(&mnt->mnt_child); |
38129a13 | 841 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 842 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 843 | mp = mnt->mnt_mp; |
84d17192 | 844 | mnt->mnt_mp = NULL; |
e4e59906 | 845 | return mp; |
7bdb11de EB |
846 | } |
847 | ||
6a46c573 EB |
848 | /* |
849 | * vfsmount lock must be held for write | |
850 | */ | |
851 | static void umount_mnt(struct mount *mnt) | |
852 | { | |
e4e59906 | 853 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
854 | } |
855 | ||
99b7db7b NP |
856 | /* |
857 | * vfsmount lock must be held for write | |
858 | */ | |
84d17192 AV |
859 | void mnt_set_mountpoint(struct mount *mnt, |
860 | struct mountpoint *mp, | |
44d964d6 | 861 | struct mount *child_mnt) |
b90fa9ae | 862 | { |
84d17192 | 863 | mp->m_count++; |
3a2393d7 | 864 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 865 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 866 | child_mnt->mnt_parent = mnt; |
84d17192 | 867 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 868 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
869 | } |
870 | ||
1064f874 EB |
871 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
872 | { | |
873 | hlist_add_head_rcu(&mnt->mnt_hash, | |
874 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
875 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
876 | } | |
877 | ||
99b7db7b NP |
878 | /* |
879 | * vfsmount lock must be held for write | |
880 | */ | |
84d17192 AV |
881 | static void attach_mnt(struct mount *mnt, |
882 | struct mount *parent, | |
883 | struct mountpoint *mp) | |
1da177e4 | 884 | { |
84d17192 | 885 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 886 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
887 | } |
888 | ||
1064f874 | 889 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 890 | { |
1064f874 | 891 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
892 | struct mount *old_parent = mnt->mnt_parent; |
893 | ||
894 | list_del_init(&mnt->mnt_child); | |
895 | hlist_del_init(&mnt->mnt_mp_list); | |
896 | hlist_del_init_rcu(&mnt->mnt_hash); | |
897 | ||
898 | attach_mnt(mnt, parent, mp); | |
899 | ||
900 | put_mountpoint(old_mp); | |
1064f874 | 901 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
902 | } |
903 | ||
b90fa9ae | 904 | /* |
99b7db7b | 905 | * vfsmount lock must be held for write |
b90fa9ae | 906 | */ |
1064f874 | 907 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 908 | { |
0714a533 | 909 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 910 | struct mount *m; |
b90fa9ae | 911 | LIST_HEAD(head); |
143c8c91 | 912 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 913 | |
0714a533 | 914 | BUG_ON(parent == mnt); |
b90fa9ae | 915 | |
1a4eeaf2 | 916 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 917 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 918 | m->mnt_ns = n; |
f03c6599 | 919 | |
b90fa9ae RP |
920 | list_splice(&head, n->list.prev); |
921 | ||
d2921684 EB |
922 | n->mounts += n->pending_mounts; |
923 | n->pending_mounts = 0; | |
924 | ||
1064f874 | 925 | __attach_mnt(mnt, parent); |
6b3286ed | 926 | touch_mnt_namespace(n); |
1da177e4 LT |
927 | } |
928 | ||
909b0a88 | 929 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 930 | { |
6b41d536 AV |
931 | struct list_head *next = p->mnt_mounts.next; |
932 | if (next == &p->mnt_mounts) { | |
1da177e4 | 933 | while (1) { |
909b0a88 | 934 | if (p == root) |
1da177e4 | 935 | return NULL; |
6b41d536 AV |
936 | next = p->mnt_child.next; |
937 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 938 | break; |
0714a533 | 939 | p = p->mnt_parent; |
1da177e4 LT |
940 | } |
941 | } | |
6b41d536 | 942 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
943 | } |
944 | ||
315fc83e | 945 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 946 | { |
6b41d536 AV |
947 | struct list_head *prev = p->mnt_mounts.prev; |
948 | while (prev != &p->mnt_mounts) { | |
949 | p = list_entry(prev, struct mount, mnt_child); | |
950 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
951 | } |
952 | return p; | |
953 | } | |
954 | ||
8f291889 AV |
955 | /** |
956 | * vfs_create_mount - Create a mount for a configured superblock | |
957 | * @fc: The configuration context with the superblock attached | |
958 | * | |
959 | * Create a mount to an already configured superblock. If necessary, the | |
960 | * caller should invoke vfs_get_tree() before calling this. | |
961 | * | |
962 | * Note that this does not attach the mount to anything. | |
963 | */ | |
964 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 965 | { |
b105e270 | 966 | struct mount *mnt; |
9d412a43 | 967 | |
8f291889 AV |
968 | if (!fc->root) |
969 | return ERR_PTR(-EINVAL); | |
9d412a43 | 970 | |
8f291889 | 971 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
972 | if (!mnt) |
973 | return ERR_PTR(-ENOMEM); | |
974 | ||
8f291889 | 975 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 976 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 977 | |
8f291889 AV |
978 | atomic_inc(&fc->root->d_sb->s_active); |
979 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
980 | mnt->mnt.mnt_root = dget(fc->root); | |
981 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
982 | mnt->mnt_parent = mnt; | |
9d412a43 | 983 | |
719ea2fb | 984 | lock_mount_hash(); |
8f291889 | 985 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 986 | unlock_mount_hash(); |
b105e270 | 987 | return &mnt->mnt; |
9d412a43 | 988 | } |
8f291889 AV |
989 | EXPORT_SYMBOL(vfs_create_mount); |
990 | ||
991 | struct vfsmount *fc_mount(struct fs_context *fc) | |
992 | { | |
993 | int err = vfs_get_tree(fc); | |
994 | if (!err) { | |
995 | up_write(&fc->root->d_sb->s_umount); | |
996 | return vfs_create_mount(fc); | |
997 | } | |
998 | return ERR_PTR(err); | |
999 | } | |
1000 | EXPORT_SYMBOL(fc_mount); | |
1001 | ||
9bc61ab1 DH |
1002 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1003 | int flags, const char *name, | |
1004 | void *data) | |
9d412a43 | 1005 | { |
9bc61ab1 | 1006 | struct fs_context *fc; |
8f291889 | 1007 | struct vfsmount *mnt; |
9bc61ab1 | 1008 | int ret = 0; |
9d412a43 AV |
1009 | |
1010 | if (!type) | |
3e1aeb00 | 1011 | return ERR_PTR(-EINVAL); |
9d412a43 | 1012 | |
9bc61ab1 DH |
1013 | fc = fs_context_for_mount(type, flags); |
1014 | if (IS_ERR(fc)) | |
1015 | return ERR_CAST(fc); | |
1016 | ||
3e1aeb00 DH |
1017 | if (name) |
1018 | ret = vfs_parse_fs_string(fc, "source", | |
1019 | name, strlen(name)); | |
9bc61ab1 DH |
1020 | if (!ret) |
1021 | ret = parse_monolithic_mount_data(fc, data); | |
1022 | if (!ret) | |
8f291889 AV |
1023 | mnt = fc_mount(fc); |
1024 | else | |
1025 | mnt = ERR_PTR(ret); | |
9d412a43 | 1026 | |
9bc61ab1 | 1027 | put_fs_context(fc); |
8f291889 | 1028 | return mnt; |
9d412a43 AV |
1029 | } |
1030 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1031 | ||
93faccbb EB |
1032 | struct vfsmount * |
1033 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1034 | const char *name, void *data) | |
1035 | { | |
1036 | /* Until it is worked out how to pass the user namespace | |
1037 | * through from the parent mount to the submount don't support | |
1038 | * unprivileged mounts with submounts. | |
1039 | */ | |
1040 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1041 | return ERR_PTR(-EPERM); | |
1042 | ||
e462ec50 | 1043 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1044 | } |
1045 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1046 | ||
87129cc0 | 1047 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1048 | int flag) |
1da177e4 | 1049 | { |
87129cc0 | 1050 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1051 | struct mount *mnt; |
1052 | int err; | |
1da177e4 | 1053 | |
be34d1a3 DH |
1054 | mnt = alloc_vfsmnt(old->mnt_devname); |
1055 | if (!mnt) | |
1056 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1057 | |
7a472ef4 | 1058 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1059 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1060 | else | |
1061 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1062 | |
be34d1a3 DH |
1063 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1064 | err = mnt_alloc_group_id(mnt); | |
1065 | if (err) | |
1066 | goto out_free; | |
1da177e4 | 1067 | } |
be34d1a3 | 1068 | |
16a34adb AV |
1069 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1070 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1071 | |
be34d1a3 | 1072 | atomic_inc(&sb->s_active); |
a6435940 CB |
1073 | mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); |
1074 | if (mnt->mnt.mnt_userns != &init_user_ns) | |
1075 | mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); | |
be34d1a3 DH |
1076 | mnt->mnt.mnt_sb = sb; |
1077 | mnt->mnt.mnt_root = dget(root); | |
1078 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1079 | mnt->mnt_parent = mnt; | |
719ea2fb | 1080 | lock_mount_hash(); |
be34d1a3 | 1081 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1082 | unlock_mount_hash(); |
be34d1a3 | 1083 | |
7a472ef4 EB |
1084 | if ((flag & CL_SLAVE) || |
1085 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1086 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1087 | mnt->mnt_master = old; | |
1088 | CLEAR_MNT_SHARED(mnt); | |
1089 | } else if (!(flag & CL_PRIVATE)) { | |
1090 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1091 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1092 | if (IS_MNT_SLAVE(old)) | |
1093 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1094 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1095 | } else { |
1096 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1097 | } |
1098 | if (flag & CL_MAKE_SHARED) | |
1099 | set_mnt_shared(mnt); | |
1100 | ||
1101 | /* stick the duplicate mount on the same expiry list | |
1102 | * as the original if that was on one */ | |
1103 | if (flag & CL_EXPIRE) { | |
1104 | if (!list_empty(&old->mnt_expire)) | |
1105 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1106 | } | |
1107 | ||
cb338d06 | 1108 | return mnt; |
719f5d7f MS |
1109 | |
1110 | out_free: | |
8ffcb32e | 1111 | mnt_free_id(mnt); |
719f5d7f | 1112 | free_vfsmnt(mnt); |
be34d1a3 | 1113 | return ERR_PTR(err); |
1da177e4 LT |
1114 | } |
1115 | ||
9ea459e1 AV |
1116 | static void cleanup_mnt(struct mount *mnt) |
1117 | { | |
56cbb429 AV |
1118 | struct hlist_node *p; |
1119 | struct mount *m; | |
9ea459e1 | 1120 | /* |
56cbb429 AV |
1121 | * The warning here probably indicates that somebody messed |
1122 | * up a mnt_want/drop_write() pair. If this happens, the | |
1123 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1124 | * The locking used to deal with mnt_count decrement provides barriers, |
1125 | * so mnt_get_writers() below is safe. | |
1126 | */ | |
1127 | WARN_ON(mnt_get_writers(mnt)); | |
1128 | if (unlikely(mnt->mnt_pins.first)) | |
1129 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1130 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1131 | hlist_del(&m->mnt_umount); | |
1132 | mntput(&m->mnt); | |
1133 | } | |
9ea459e1 AV |
1134 | fsnotify_vfsmount_delete(&mnt->mnt); |
1135 | dput(mnt->mnt.mnt_root); | |
1136 | deactivate_super(mnt->mnt.mnt_sb); | |
1137 | mnt_free_id(mnt); | |
1138 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1139 | } | |
1140 | ||
1141 | static void __cleanup_mnt(struct rcu_head *head) | |
1142 | { | |
1143 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1144 | } | |
1145 | ||
1146 | static LLIST_HEAD(delayed_mntput_list); | |
1147 | static void delayed_mntput(struct work_struct *unused) | |
1148 | { | |
1149 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1150 | struct mount *m, *t; |
9ea459e1 | 1151 | |
29785735 BP |
1152 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1153 | cleanup_mnt(m); | |
9ea459e1 AV |
1154 | } |
1155 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1156 | ||
900148dc | 1157 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1158 | { |
4edbe133 | 1159 | LIST_HEAD(list); |
edf7ddbf | 1160 | int count; |
4edbe133 | 1161 | |
48a066e7 | 1162 | rcu_read_lock(); |
9ea0a46c AV |
1163 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1164 | /* | |
1165 | * Since we don't do lock_mount_hash() here, | |
1166 | * ->mnt_ns can change under us. However, if it's | |
1167 | * non-NULL, then there's a reference that won't | |
1168 | * be dropped until after an RCU delay done after | |
1169 | * turning ->mnt_ns NULL. So if we observe it | |
1170 | * non-NULL under rcu_read_lock(), the reference | |
1171 | * we are dropping is not the final one. | |
1172 | */ | |
1173 | mnt_add_count(mnt, -1); | |
48a066e7 | 1174 | rcu_read_unlock(); |
f03c6599 | 1175 | return; |
b3e19d92 | 1176 | } |
719ea2fb | 1177 | lock_mount_hash(); |
119e1ef8 AV |
1178 | /* |
1179 | * make sure that if __legitimize_mnt() has not seen us grab | |
1180 | * mount_lock, we'll see their refcount increment here. | |
1181 | */ | |
1182 | smp_mb(); | |
9ea0a46c | 1183 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1184 | count = mnt_get_count(mnt); |
1185 | if (count != 0) { | |
1186 | WARN_ON(count < 0); | |
48a066e7 | 1187 | rcu_read_unlock(); |
719ea2fb | 1188 | unlock_mount_hash(); |
99b7db7b NP |
1189 | return; |
1190 | } | |
48a066e7 AV |
1191 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1192 | rcu_read_unlock(); | |
1193 | unlock_mount_hash(); | |
1194 | return; | |
1195 | } | |
1196 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1197 | rcu_read_unlock(); | |
962830df | 1198 | |
39f7c4db | 1199 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1200 | |
1201 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1202 | struct mount *p, *tmp; | |
1203 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1204 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1205 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1206 | } |
1207 | } | |
719ea2fb | 1208 | unlock_mount_hash(); |
4edbe133 | 1209 | shrink_dentry_list(&list); |
649a795a | 1210 | |
9ea459e1 AV |
1211 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1212 | struct task_struct *task = current; | |
1213 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1214 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1215 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1216 | return; |
1217 | } | |
1218 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1219 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1220 | return; | |
1221 | } | |
1222 | cleanup_mnt(mnt); | |
b3e19d92 | 1223 | } |
b3e19d92 NP |
1224 | |
1225 | void mntput(struct vfsmount *mnt) | |
1226 | { | |
1227 | if (mnt) { | |
863d684f | 1228 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1229 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1230 | if (unlikely(m->mnt_expiry_mark)) |
1231 | m->mnt_expiry_mark = 0; | |
1232 | mntput_no_expire(m); | |
b3e19d92 NP |
1233 | } |
1234 | } | |
1235 | EXPORT_SYMBOL(mntput); | |
1236 | ||
1237 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1238 | { | |
1239 | if (mnt) | |
83adc753 | 1240 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1241 | return mnt; |
1242 | } | |
1243 | EXPORT_SYMBOL(mntget); | |
1244 | ||
c6609c0a IK |
1245 | /* path_is_mountpoint() - Check if path is a mount in the current |
1246 | * namespace. | |
1247 | * | |
1248 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1249 | * not mounted in any namespace and that common case is handled inline. | |
1250 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1251 | * mounts using a given dentry in a different namespace. This function | |
1252 | * checks if the passed in path is a mountpoint rather than the dentry | |
1253 | * alone. | |
1254 | */ | |
1255 | bool path_is_mountpoint(const struct path *path) | |
1256 | { | |
1257 | unsigned seq; | |
1258 | bool res; | |
1259 | ||
1260 | if (!d_mountpoint(path->dentry)) | |
1261 | return false; | |
1262 | ||
1263 | rcu_read_lock(); | |
1264 | do { | |
1265 | seq = read_seqbegin(&mount_lock); | |
1266 | res = __path_is_mountpoint(path); | |
1267 | } while (read_seqretry(&mount_lock, seq)); | |
1268 | rcu_read_unlock(); | |
1269 | ||
1270 | return res; | |
1271 | } | |
1272 | EXPORT_SYMBOL(path_is_mountpoint); | |
1273 | ||
ca71cf71 | 1274 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1275 | { |
3064c356 AV |
1276 | struct mount *p; |
1277 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1278 | if (IS_ERR(p)) | |
1279 | return ERR_CAST(p); | |
1280 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1281 | return &p->mnt; | |
7b7b1ace | 1282 | } |
1da177e4 | 1283 | |
a1a2c409 | 1284 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1285 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1286 | struct list_head *p) | |
1287 | { | |
1288 | struct mount *mnt, *ret = NULL; | |
1289 | ||
1290 | lock_ns_list(ns); | |
1291 | list_for_each_continue(p, &ns->list) { | |
1292 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1293 | if (!mnt_is_cursor(mnt)) { | |
1294 | ret = mnt; | |
1295 | break; | |
1296 | } | |
1297 | } | |
1298 | unlock_ns_list(ns); | |
1299 | ||
1300 | return ret; | |
1301 | } | |
1302 | ||
0226f492 | 1303 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1304 | static void *m_start(struct seq_file *m, loff_t *pos) |
1305 | { | |
ede1bf0d | 1306 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1307 | struct list_head *prev; |
1da177e4 | 1308 | |
390c6843 | 1309 | down_read(&namespace_sem); |
9f6c61f9 MS |
1310 | if (!*pos) { |
1311 | prev = &p->ns->list; | |
1312 | } else { | |
1313 | prev = &p->cursor.mnt_list; | |
1314 | ||
1315 | /* Read after we'd reached the end? */ | |
1316 | if (list_empty(prev)) | |
1317 | return NULL; | |
c7999c36 AV |
1318 | } |
1319 | ||
9f6c61f9 | 1320 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1321 | } |
1322 | ||
1323 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1324 | { | |
ede1bf0d | 1325 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1326 | struct mount *mnt = v; |
b0765fb8 | 1327 | |
9f6c61f9 MS |
1328 | ++*pos; |
1329 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1330 | } |
1331 | ||
1332 | static void m_stop(struct seq_file *m, void *v) | |
1333 | { | |
9f6c61f9 MS |
1334 | struct proc_mounts *p = m->private; |
1335 | struct mount *mnt = v; | |
1336 | ||
1337 | lock_ns_list(p->ns); | |
1338 | if (mnt) | |
1339 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1340 | else | |
1341 | list_del_init(&p->cursor.mnt_list); | |
1342 | unlock_ns_list(p->ns); | |
390c6843 | 1343 | up_read(&namespace_sem); |
1da177e4 LT |
1344 | } |
1345 | ||
0226f492 | 1346 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1347 | { |
ede1bf0d | 1348 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1349 | struct mount *r = v; |
0226f492 | 1350 | return p->show(m, &r->mnt); |
1da177e4 LT |
1351 | } |
1352 | ||
a1a2c409 | 1353 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1354 | .start = m_start, |
1355 | .next = m_next, | |
1356 | .stop = m_stop, | |
0226f492 | 1357 | .show = m_show, |
b4629fe2 | 1358 | }; |
9f6c61f9 MS |
1359 | |
1360 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1361 | { | |
1362 | down_read(&namespace_sem); | |
1363 | lock_ns_list(ns); | |
1364 | list_del(&cursor->mnt_list); | |
1365 | unlock_ns_list(ns); | |
1366 | up_read(&namespace_sem); | |
1367 | } | |
a1a2c409 | 1368 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1369 | |
1da177e4 LT |
1370 | /** |
1371 | * may_umount_tree - check if a mount tree is busy | |
1372 | * @mnt: root of mount tree | |
1373 | * | |
1374 | * This is called to check if a tree of mounts has any | |
1375 | * open files, pwds, chroots or sub mounts that are | |
1376 | * busy. | |
1377 | */ | |
909b0a88 | 1378 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1379 | { |
909b0a88 | 1380 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1381 | int actual_refs = 0; |
1382 | int minimum_refs = 0; | |
315fc83e | 1383 | struct mount *p; |
909b0a88 | 1384 | BUG_ON(!m); |
1da177e4 | 1385 | |
b3e19d92 | 1386 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1387 | lock_mount_hash(); |
909b0a88 | 1388 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1389 | actual_refs += mnt_get_count(p); |
1da177e4 | 1390 | minimum_refs += 2; |
1da177e4 | 1391 | } |
719ea2fb | 1392 | unlock_mount_hash(); |
1da177e4 LT |
1393 | |
1394 | if (actual_refs > minimum_refs) | |
e3474a8e | 1395 | return 0; |
1da177e4 | 1396 | |
e3474a8e | 1397 | return 1; |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | EXPORT_SYMBOL(may_umount_tree); | |
1401 | ||
1402 | /** | |
1403 | * may_umount - check if a mount point is busy | |
1404 | * @mnt: root of mount | |
1405 | * | |
1406 | * This is called to check if a mount point has any | |
1407 | * open files, pwds, chroots or sub mounts. If the | |
1408 | * mount has sub mounts this will return busy | |
1409 | * regardless of whether the sub mounts are busy. | |
1410 | * | |
1411 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1412 | * give false negatives. The main reason why it's here is that we need | |
1413 | * a non-destructive way to look for easily umountable filesystems. | |
1414 | */ | |
1415 | int may_umount(struct vfsmount *mnt) | |
1416 | { | |
e3474a8e | 1417 | int ret = 1; |
8ad08d8a | 1418 | down_read(&namespace_sem); |
719ea2fb | 1419 | lock_mount_hash(); |
1ab59738 | 1420 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1421 | ret = 0; |
719ea2fb | 1422 | unlock_mount_hash(); |
8ad08d8a | 1423 | up_read(&namespace_sem); |
a05964f3 | 1424 | return ret; |
1da177e4 LT |
1425 | } |
1426 | ||
1427 | EXPORT_SYMBOL(may_umount); | |
1428 | ||
97216be0 | 1429 | static void namespace_unlock(void) |
70fbcdf4 | 1430 | { |
a3b3c562 | 1431 | struct hlist_head head; |
56cbb429 AV |
1432 | struct hlist_node *p; |
1433 | struct mount *m; | |
4edbe133 | 1434 | LIST_HEAD(list); |
97216be0 | 1435 | |
a3b3c562 | 1436 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1437 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1438 | |
97216be0 AV |
1439 | up_write(&namespace_sem); |
1440 | ||
4edbe133 AV |
1441 | shrink_dentry_list(&list); |
1442 | ||
a3b3c562 EB |
1443 | if (likely(hlist_empty(&head))) |
1444 | return; | |
1445 | ||
22cb7405 | 1446 | synchronize_rcu_expedited(); |
48a066e7 | 1447 | |
56cbb429 AV |
1448 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1449 | hlist_del(&m->mnt_umount); | |
1450 | mntput(&m->mnt); | |
1451 | } | |
70fbcdf4 RP |
1452 | } |
1453 | ||
97216be0 | 1454 | static inline void namespace_lock(void) |
e3197d83 | 1455 | { |
97216be0 | 1456 | down_write(&namespace_sem); |
e3197d83 AV |
1457 | } |
1458 | ||
e819f152 EB |
1459 | enum umount_tree_flags { |
1460 | UMOUNT_SYNC = 1, | |
1461 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1462 | UMOUNT_CONNECTED = 4, |
e819f152 | 1463 | }; |
f2d0a123 EB |
1464 | |
1465 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1466 | { | |
1467 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1468 | if (how & UMOUNT_SYNC) | |
1469 | return true; | |
1470 | ||
1471 | /* A mount without a parent has nothing to be connected to */ | |
1472 | if (!mnt_has_parent(mnt)) | |
1473 | return true; | |
1474 | ||
1475 | /* Because the reference counting rules change when mounts are | |
1476 | * unmounted and connected, umounted mounts may not be | |
1477 | * connected to mounted mounts. | |
1478 | */ | |
1479 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1480 | return true; | |
1481 | ||
1482 | /* Has it been requested that the mount remain connected? */ | |
1483 | if (how & UMOUNT_CONNECTED) | |
1484 | return false; | |
1485 | ||
1486 | /* Is the mount locked such that it needs to remain connected? */ | |
1487 | if (IS_MNT_LOCKED(mnt)) | |
1488 | return false; | |
1489 | ||
1490 | /* By default disconnect the mount */ | |
1491 | return true; | |
1492 | } | |
1493 | ||
99b7db7b | 1494 | /* |
48a066e7 | 1495 | * mount_lock must be held |
99b7db7b NP |
1496 | * namespace_sem must be held for write |
1497 | */ | |
e819f152 | 1498 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1499 | { |
c003b26f | 1500 | LIST_HEAD(tmp_list); |
315fc83e | 1501 | struct mount *p; |
1da177e4 | 1502 | |
5d88457e EB |
1503 | if (how & UMOUNT_PROPAGATE) |
1504 | propagate_mount_unlock(mnt); | |
1505 | ||
c003b26f | 1506 | /* Gather the mounts to umount */ |
590ce4bc EB |
1507 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1508 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1509 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1510 | } |
1da177e4 | 1511 | |
411a938b | 1512 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1513 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1514 | list_del_init(&p->mnt_child); |
c003b26f | 1515 | } |
88b368f2 | 1516 | |
c003b26f | 1517 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1518 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1519 | propagate_umount(&tmp_list); |
a05964f3 | 1520 | |
c003b26f | 1521 | while (!list_empty(&tmp_list)) { |
d2921684 | 1522 | struct mnt_namespace *ns; |
ce07d891 | 1523 | bool disconnect; |
c003b26f | 1524 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1525 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1526 | list_del_init(&p->mnt_list); |
d2921684 EB |
1527 | ns = p->mnt_ns; |
1528 | if (ns) { | |
1529 | ns->mounts--; | |
1530 | __touch_mnt_namespace(ns); | |
1531 | } | |
143c8c91 | 1532 | p->mnt_ns = NULL; |
e819f152 | 1533 | if (how & UMOUNT_SYNC) |
48a066e7 | 1534 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1535 | |
f2d0a123 | 1536 | disconnect = disconnect_mount(p, how); |
676da58d | 1537 | if (mnt_has_parent(p)) { |
81b6b061 | 1538 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1539 | if (!disconnect) { |
1540 | /* Don't forget about p */ | |
1541 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1542 | } else { | |
1543 | umount_mnt(p); | |
1544 | } | |
7c4b93d8 | 1545 | } |
0f0afb1d | 1546 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1547 | if (disconnect) |
1548 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1549 | } |
1550 | } | |
1551 | ||
b54b9be7 | 1552 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1553 | |
8d0347f6 DH |
1554 | static int do_umount_root(struct super_block *sb) |
1555 | { | |
1556 | int ret = 0; | |
1557 | ||
1558 | down_write(&sb->s_umount); | |
1559 | if (!sb_rdonly(sb)) { | |
1560 | struct fs_context *fc; | |
1561 | ||
1562 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1563 | SB_RDONLY); | |
1564 | if (IS_ERR(fc)) { | |
1565 | ret = PTR_ERR(fc); | |
1566 | } else { | |
1567 | ret = parse_monolithic_mount_data(fc, NULL); | |
1568 | if (!ret) | |
1569 | ret = reconfigure_super(fc); | |
1570 | put_fs_context(fc); | |
1571 | } | |
1572 | } | |
1573 | up_write(&sb->s_umount); | |
1574 | return ret; | |
1575 | } | |
1576 | ||
1ab59738 | 1577 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1578 | { |
1ab59738 | 1579 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1580 | int retval; |
1581 | ||
1ab59738 | 1582 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1583 | if (retval) |
1584 | return retval; | |
1585 | ||
1586 | /* | |
1587 | * Allow userspace to request a mountpoint be expired rather than | |
1588 | * unmounting unconditionally. Unmount only happens if: | |
1589 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1590 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1591 | */ | |
1592 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1593 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1594 | flags & (MNT_FORCE | MNT_DETACH)) |
1595 | return -EINVAL; | |
1596 | ||
b3e19d92 NP |
1597 | /* |
1598 | * probably don't strictly need the lock here if we examined | |
1599 | * all race cases, but it's a slowpath. | |
1600 | */ | |
719ea2fb | 1601 | lock_mount_hash(); |
83adc753 | 1602 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1603 | unlock_mount_hash(); |
1da177e4 | 1604 | return -EBUSY; |
b3e19d92 | 1605 | } |
719ea2fb | 1606 | unlock_mount_hash(); |
1da177e4 | 1607 | |
863d684f | 1608 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1609 | return -EAGAIN; |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * If we may have to abort operations to get out of this | |
1614 | * mount, and they will themselves hold resources we must | |
1615 | * allow the fs to do things. In the Unix tradition of | |
1616 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1617 | * might fail to complete on the first run through as other tasks | |
1618 | * must return, and the like. Thats for the mount program to worry | |
1619 | * about for the moment. | |
1620 | */ | |
1621 | ||
42faad99 | 1622 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1623 | sb->s_op->umount_begin(sb); |
42faad99 | 1624 | } |
1da177e4 LT |
1625 | |
1626 | /* | |
1627 | * No sense to grab the lock for this test, but test itself looks | |
1628 | * somewhat bogus. Suggestions for better replacement? | |
1629 | * Ho-hum... In principle, we might treat that as umount + switch | |
1630 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1631 | * Actually it makes sense, especially if rootfs would contain a | |
1632 | * /reboot - static binary that would close all descriptors and | |
1633 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1634 | */ | |
1ab59738 | 1635 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1636 | /* |
1637 | * Special case for "unmounting" root ... | |
1638 | * we just try to remount it readonly. | |
1639 | */ | |
bc6155d1 | 1640 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1641 | return -EPERM; |
8d0347f6 | 1642 | return do_umount_root(sb); |
1da177e4 LT |
1643 | } |
1644 | ||
97216be0 | 1645 | namespace_lock(); |
719ea2fb | 1646 | lock_mount_hash(); |
1da177e4 | 1647 | |
25d202ed EB |
1648 | /* Recheck MNT_LOCKED with the locks held */ |
1649 | retval = -EINVAL; | |
1650 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1651 | goto out; | |
1652 | ||
1653 | event++; | |
48a066e7 | 1654 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1655 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1656 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1657 | retval = 0; |
48a066e7 AV |
1658 | } else { |
1659 | shrink_submounts(mnt); | |
1660 | retval = -EBUSY; | |
1661 | if (!propagate_mount_busy(mnt, 2)) { | |
1662 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1663 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1664 | retval = 0; |
1665 | } | |
1da177e4 | 1666 | } |
25d202ed | 1667 | out: |
719ea2fb | 1668 | unlock_mount_hash(); |
e3197d83 | 1669 | namespace_unlock(); |
1da177e4 LT |
1670 | return retval; |
1671 | } | |
1672 | ||
80b5dce8 EB |
1673 | /* |
1674 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1675 | * | |
1676 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1677 | * to an existing mountpoint, and wind up leaking the mount. | |
1678 | * detach_mounts allows lazily unmounting those mounts instead of | |
1679 | * leaking them. | |
1680 | * | |
1681 | * The caller may hold dentry->d_inode->i_mutex. | |
1682 | */ | |
1683 | void __detach_mounts(struct dentry *dentry) | |
1684 | { | |
1685 | struct mountpoint *mp; | |
1686 | struct mount *mnt; | |
1687 | ||
1688 | namespace_lock(); | |
3895dbf8 | 1689 | lock_mount_hash(); |
80b5dce8 | 1690 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1691 | if (!mp) |
80b5dce8 EB |
1692 | goto out_unlock; |
1693 | ||
e06b933e | 1694 | event++; |
80b5dce8 EB |
1695 | while (!hlist_empty(&mp->m_list)) { |
1696 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1697 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1698 | umount_mnt(mnt); |
56cbb429 | 1699 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1700 | } |
e0c9c0af | 1701 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1702 | } |
80b5dce8 EB |
1703 | put_mountpoint(mp); |
1704 | out_unlock: | |
3895dbf8 | 1705 | unlock_mount_hash(); |
80b5dce8 EB |
1706 | namespace_unlock(); |
1707 | } | |
1708 | ||
dd111b31 | 1709 | /* |
9b40bc90 AV |
1710 | * Is the caller allowed to modify his namespace? |
1711 | */ | |
1712 | static inline bool may_mount(void) | |
1713 | { | |
1714 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1715 | } | |
1716 | ||
df2474a2 | 1717 | #ifdef CONFIG_MANDATORY_FILE_LOCKING |
9e8925b6 JL |
1718 | static inline bool may_mandlock(void) |
1719 | { | |
95ace754 | 1720 | return capable(CAP_SYS_ADMIN); |
9e8925b6 | 1721 | } |
df2474a2 JL |
1722 | #else |
1723 | static inline bool may_mandlock(void) | |
1724 | { | |
1725 | pr_warn("VFS: \"mand\" mount option not supported"); | |
1726 | return false; | |
1727 | } | |
1728 | #endif | |
9e8925b6 | 1729 | |
25ccd24f | 1730 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1731 | { |
25ccd24f | 1732 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1733 | |
9b40bc90 AV |
1734 | if (!may_mount()) |
1735 | return -EPERM; | |
41525f56 | 1736 | if (path->dentry != path->mnt->mnt_root) |
25ccd24f | 1737 | return -EINVAL; |
143c8c91 | 1738 | if (!check_mnt(mnt)) |
25ccd24f | 1739 | return -EINVAL; |
25d202ed | 1740 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1741 | return -EINVAL; |
b2f5d4dc | 1742 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1743 | return -EPERM; |
1744 | return 0; | |
1745 | } | |
1746 | ||
a0a6df9a | 1747 | // caller is responsible for flags being sane |
25ccd24f CH |
1748 | int path_umount(struct path *path, int flags) |
1749 | { | |
1750 | struct mount *mnt = real_mount(path->mnt); | |
1751 | int ret; | |
1752 | ||
1753 | ret = can_umount(path, flags); | |
1754 | if (!ret) | |
1755 | ret = do_umount(mnt, flags); | |
1da177e4 | 1756 | |
429731b1 | 1757 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1758 | dput(path->dentry); |
900148dc | 1759 | mntput_no_expire(mnt); |
25ccd24f | 1760 | return ret; |
1da177e4 LT |
1761 | } |
1762 | ||
09267def | 1763 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1764 | { |
1765 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1766 | struct path path; | |
1767 | int ret; | |
1768 | ||
a0a6df9a AV |
1769 | // basic validity checks done first |
1770 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1771 | return -EINVAL; | |
1772 | ||
41525f56 CH |
1773 | if (!(flags & UMOUNT_NOFOLLOW)) |
1774 | lookup_flags |= LOOKUP_FOLLOW; | |
1775 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1776 | if (ret) | |
1777 | return ret; | |
1778 | return path_umount(&path, flags); | |
1779 | } | |
1780 | ||
3a18ef5c DB |
1781 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1782 | { | |
1783 | return ksys_umount(name, flags); | |
1784 | } | |
1785 | ||
1da177e4 LT |
1786 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1787 | ||
1788 | /* | |
b58fed8b | 1789 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1790 | */ |
bdc480e3 | 1791 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1792 | { |
3a18ef5c | 1793 | return ksys_umount(name, 0); |
1da177e4 LT |
1794 | } |
1795 | ||
1796 | #endif | |
1797 | ||
4ce5d2b1 | 1798 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1799 | { |
4ce5d2b1 | 1800 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1801 | return dentry->d_op == &ns_dentry_operations && |
1802 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1803 | } |
1804 | ||
213921f9 | 1805 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1806 | { |
1807 | return container_of(ns, struct mnt_namespace, ns); | |
1808 | } | |
1809 | ||
303cc571 CB |
1810 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1811 | { | |
1812 | return &mnt->ns; | |
1813 | } | |
1814 | ||
4ce5d2b1 EB |
1815 | static bool mnt_ns_loop(struct dentry *dentry) |
1816 | { | |
1817 | /* Could bind mounting the mount namespace inode cause a | |
1818 | * mount namespace loop? | |
1819 | */ | |
1820 | struct mnt_namespace *mnt_ns; | |
1821 | if (!is_mnt_ns_file(dentry)) | |
1822 | return false; | |
1823 | ||
f77c8014 | 1824 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1825 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1826 | } | |
1827 | ||
87129cc0 | 1828 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1829 | int flag) |
1da177e4 | 1830 | { |
84d17192 | 1831 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1832 | |
4ce5d2b1 EB |
1833 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1834 | return ERR_PTR(-EINVAL); | |
1835 | ||
1836 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1837 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1838 | |
36341f64 | 1839 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1840 | if (IS_ERR(q)) |
1841 | return q; | |
1842 | ||
a73324da | 1843 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1844 | |
1845 | p = mnt; | |
6b41d536 | 1846 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1847 | struct mount *s; |
7ec02ef1 | 1848 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1849 | continue; |
1850 | ||
909b0a88 | 1851 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1852 | if (!(flag & CL_COPY_UNBINDABLE) && |
1853 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1854 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1855 | /* Both unbindable and locked. */ | |
1856 | q = ERR_PTR(-EPERM); | |
1857 | goto out; | |
1858 | } else { | |
1859 | s = skip_mnt_tree(s); | |
1860 | continue; | |
1861 | } | |
4ce5d2b1 EB |
1862 | } |
1863 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1864 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1865 | s = skip_mnt_tree(s); |
1866 | continue; | |
1867 | } | |
0714a533 AV |
1868 | while (p != s->mnt_parent) { |
1869 | p = p->mnt_parent; | |
1870 | q = q->mnt_parent; | |
1da177e4 | 1871 | } |
87129cc0 | 1872 | p = s; |
84d17192 | 1873 | parent = q; |
87129cc0 | 1874 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1875 | if (IS_ERR(q)) |
1876 | goto out; | |
719ea2fb | 1877 | lock_mount_hash(); |
1a4eeaf2 | 1878 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1879 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1880 | unlock_mount_hash(); |
1da177e4 LT |
1881 | } |
1882 | } | |
1883 | return res; | |
be34d1a3 | 1884 | out: |
1da177e4 | 1885 | if (res) { |
719ea2fb | 1886 | lock_mount_hash(); |
e819f152 | 1887 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1888 | unlock_mount_hash(); |
1da177e4 | 1889 | } |
be34d1a3 | 1890 | return q; |
1da177e4 LT |
1891 | } |
1892 | ||
be34d1a3 DH |
1893 | /* Caller should check returned pointer for errors */ |
1894 | ||
ca71cf71 | 1895 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1896 | { |
cb338d06 | 1897 | struct mount *tree; |
97216be0 | 1898 | namespace_lock(); |
cd4a4017 EB |
1899 | if (!check_mnt(real_mount(path->mnt))) |
1900 | tree = ERR_PTR(-EINVAL); | |
1901 | else | |
1902 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1903 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1904 | namespace_unlock(); |
be34d1a3 | 1905 | if (IS_ERR(tree)) |
52e220d3 | 1906 | return ERR_CAST(tree); |
be34d1a3 | 1907 | return &tree->mnt; |
8aec0809 AV |
1908 | } |
1909 | ||
a07b2000 AV |
1910 | static void free_mnt_ns(struct mnt_namespace *); |
1911 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1912 | ||
1913 | void dissolve_on_fput(struct vfsmount *mnt) | |
1914 | { | |
1915 | struct mnt_namespace *ns; | |
1916 | namespace_lock(); | |
1917 | lock_mount_hash(); | |
1918 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1919 | if (ns) { |
1920 | if (is_anon_ns(ns)) | |
1921 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1922 | else | |
1923 | ns = NULL; | |
1924 | } | |
a07b2000 AV |
1925 | unlock_mount_hash(); |
1926 | namespace_unlock(); | |
44dfd84a DH |
1927 | if (ns) |
1928 | free_mnt_ns(ns); | |
a07b2000 AV |
1929 | } |
1930 | ||
8aec0809 AV |
1931 | void drop_collected_mounts(struct vfsmount *mnt) |
1932 | { | |
97216be0 | 1933 | namespace_lock(); |
719ea2fb | 1934 | lock_mount_hash(); |
9c8e0a1b | 1935 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1936 | unlock_mount_hash(); |
3ab6abee | 1937 | namespace_unlock(); |
8aec0809 AV |
1938 | } |
1939 | ||
c771d683 MS |
1940 | /** |
1941 | * clone_private_mount - create a private clone of a path | |
1942 | * | |
1943 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1944 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1945 | * to the originating mount won't be propagated into this). | |
1946 | * | |
1947 | * Release with mntput(). | |
1948 | */ | |
ca71cf71 | 1949 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1950 | { |
1951 | struct mount *old_mnt = real_mount(path->mnt); | |
1952 | struct mount *new_mnt; | |
1953 | ||
1954 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1955 | return ERR_PTR(-EINVAL); | |
1956 | ||
c771d683 | 1957 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1958 | if (IS_ERR(new_mnt)) |
1959 | return ERR_CAST(new_mnt); | |
1960 | ||
df820f8d MS |
1961 | /* Longterm mount to be removed by kern_unmount*() */ |
1962 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
1963 | ||
c771d683 MS |
1964 | return &new_mnt->mnt; |
1965 | } | |
1966 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1967 | ||
1f707137 AV |
1968 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1969 | struct vfsmount *root) | |
1970 | { | |
1a4eeaf2 | 1971 | struct mount *mnt; |
1f707137 AV |
1972 | int res = f(root, arg); |
1973 | if (res) | |
1974 | return res; | |
1a4eeaf2 AV |
1975 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1976 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1977 | if (res) |
1978 | return res; | |
1979 | } | |
1980 | return 0; | |
1981 | } | |
1982 | ||
3bd045cc AV |
1983 | static void lock_mnt_tree(struct mount *mnt) |
1984 | { | |
1985 | struct mount *p; | |
1986 | ||
1987 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1988 | int flags = p->mnt.mnt_flags; | |
1989 | /* Don't allow unprivileged users to change mount flags */ | |
1990 | flags |= MNT_LOCK_ATIME; | |
1991 | ||
1992 | if (flags & MNT_READONLY) | |
1993 | flags |= MNT_LOCK_READONLY; | |
1994 | ||
1995 | if (flags & MNT_NODEV) | |
1996 | flags |= MNT_LOCK_NODEV; | |
1997 | ||
1998 | if (flags & MNT_NOSUID) | |
1999 | flags |= MNT_LOCK_NOSUID; | |
2000 | ||
2001 | if (flags & MNT_NOEXEC) | |
2002 | flags |= MNT_LOCK_NOEXEC; | |
2003 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2004 | if (list_empty(&p->mnt_expire)) | |
2005 | flags |= MNT_LOCKED; | |
2006 | p->mnt.mnt_flags = flags; | |
2007 | } | |
2008 | } | |
2009 | ||
4b8b21f4 | 2010 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2011 | { |
315fc83e | 2012 | struct mount *p; |
719f5d7f | 2013 | |
909b0a88 | 2014 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2015 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2016 | mnt_release_group_id(p); |
719f5d7f MS |
2017 | } |
2018 | } | |
2019 | ||
4b8b21f4 | 2020 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2021 | { |
315fc83e | 2022 | struct mount *p; |
719f5d7f | 2023 | |
909b0a88 | 2024 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2025 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2026 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2027 | if (err) { |
4b8b21f4 | 2028 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2029 | return err; |
2030 | } | |
2031 | } | |
2032 | } | |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | ||
d2921684 EB |
2037 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2038 | { | |
2039 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
2040 | unsigned int mounts = 0, old, pending, sum; | |
2041 | struct mount *p; | |
2042 | ||
2043 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
2044 | mounts++; | |
2045 | ||
2046 | old = ns->mounts; | |
2047 | pending = ns->pending_mounts; | |
2048 | sum = old + pending; | |
2049 | if ((old > sum) || | |
2050 | (pending > sum) || | |
2051 | (max < sum) || | |
2052 | (mounts > (max - sum))) | |
2053 | return -ENOSPC; | |
2054 | ||
2055 | ns->pending_mounts = pending + mounts; | |
2056 | return 0; | |
2057 | } | |
2058 | ||
b90fa9ae RP |
2059 | /* |
2060 | * @source_mnt : mount tree to be attached | |
21444403 RP |
2061 | * @nd : place the mount tree @source_mnt is attached |
2062 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
2063 | * store the parent mount and mountpoint dentry. | |
2064 | * (done when source_mnt is moved) | |
b90fa9ae RP |
2065 | * |
2066 | * NOTE: in the table below explains the semantics when a source mount | |
2067 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2068 | * --------------------------------------------------------------------------- |
2069 | * | BIND MOUNT OPERATION | | |
2070 | * |************************************************************************** | |
2071 | * | source-->| shared | private | slave | unbindable | | |
2072 | * | dest | | | | | | |
2073 | * | | | | | | | | |
2074 | * | v | | | | | | |
2075 | * |************************************************************************** | |
2076 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2077 | * | | | | | | | |
2078 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2079 | * *************************************************************************** | |
b90fa9ae RP |
2080 | * A bind operation clones the source mount and mounts the clone on the |
2081 | * destination mount. | |
2082 | * | |
2083 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2084 | * tree of the destination mount and the cloned mount is added to | |
2085 | * the peer group of the source mount. | |
2086 | * (+) the cloned mount is created under the destination mount and is marked | |
2087 | * as shared. The cloned mount is added to the peer group of the source | |
2088 | * mount. | |
5afe0022 RP |
2089 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2090 | * of the destination mount and the cloned mount is made slave | |
2091 | * of the same master as that of the source mount. The cloned mount | |
2092 | * is marked as 'shared and slave'. | |
2093 | * (*) the cloned mount is made a slave of the same master as that of the | |
2094 | * source mount. | |
2095 | * | |
9676f0c6 RP |
2096 | * --------------------------------------------------------------------------- |
2097 | * | MOVE MOUNT OPERATION | | |
2098 | * |************************************************************************** | |
2099 | * | source-->| shared | private | slave | unbindable | | |
2100 | * | dest | | | | | | |
2101 | * | | | | | | | | |
2102 | * | v | | | | | | |
2103 | * |************************************************************************** | |
2104 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2105 | * | | | | | | | |
2106 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2107 | * *************************************************************************** | |
5afe0022 RP |
2108 | * |
2109 | * (+) the mount is moved to the destination. And is then propagated to | |
2110 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2111 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2112 | * (+++) the mount is moved to the destination and is then propagated to |
2113 | * all the mounts belonging to the destination mount's propagation tree. | |
2114 | * the mount is marked as 'shared and slave'. | |
2115 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2116 | * |
2117 | * if the source mount is a tree, the operations explained above is | |
2118 | * applied to each mount in the tree. | |
2119 | * Must be called without spinlocks held, since this function can sleep | |
2120 | * in allocations. | |
2121 | */ | |
0fb54e50 | 2122 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2123 | struct mount *dest_mnt, |
2124 | struct mountpoint *dest_mp, | |
2763d119 | 2125 | bool moving) |
b90fa9ae | 2126 | { |
3bd045cc | 2127 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2128 | HLIST_HEAD(tree_list); |
d2921684 | 2129 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2130 | struct mountpoint *smp; |
315fc83e | 2131 | struct mount *child, *p; |
38129a13 | 2132 | struct hlist_node *n; |
719f5d7f | 2133 | int err; |
b90fa9ae | 2134 | |
1064f874 EB |
2135 | /* Preallocate a mountpoint in case the new mounts need |
2136 | * to be tucked under other mounts. | |
2137 | */ | |
2138 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2139 | if (IS_ERR(smp)) | |
2140 | return PTR_ERR(smp); | |
2141 | ||
d2921684 | 2142 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2143 | if (!moving) { |
d2921684 EB |
2144 | err = count_mounts(ns, source_mnt); |
2145 | if (err) | |
2146 | goto out; | |
2147 | } | |
2148 | ||
fc7be130 | 2149 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2150 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2151 | if (err) |
2152 | goto out; | |
0b1b901b | 2153 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2154 | lock_mount_hash(); |
0b1b901b AV |
2155 | if (err) |
2156 | goto out_cleanup_ids; | |
909b0a88 | 2157 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2158 | set_mnt_shared(p); |
0b1b901b AV |
2159 | } else { |
2160 | lock_mount_hash(); | |
b90fa9ae | 2161 | } |
2763d119 AV |
2162 | if (moving) { |
2163 | unhash_mnt(source_mnt); | |
84d17192 | 2164 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2165 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2166 | } else { |
44dfd84a DH |
2167 | if (source_mnt->mnt_ns) { |
2168 | /* move from anon - the caller will destroy */ | |
2169 | list_del_init(&source_mnt->mnt_ns->list); | |
2170 | } | |
84d17192 | 2171 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2172 | commit_tree(source_mnt); |
21444403 | 2173 | } |
b90fa9ae | 2174 | |
38129a13 | 2175 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2176 | struct mount *q; |
38129a13 | 2177 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2178 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2179 | child->mnt_mountpoint); | |
2180 | if (q) | |
2181 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2182 | /* Notice when we are propagating across user namespaces */ |
2183 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2184 | lock_mnt_tree(child); | |
d728cf79 | 2185 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2186 | commit_tree(child); |
b90fa9ae | 2187 | } |
1064f874 | 2188 | put_mountpoint(smp); |
719ea2fb | 2189 | unlock_mount_hash(); |
99b7db7b | 2190 | |
b90fa9ae | 2191 | return 0; |
719f5d7f MS |
2192 | |
2193 | out_cleanup_ids: | |
f2ebb3a9 AV |
2194 | while (!hlist_empty(&tree_list)) { |
2195 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2196 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2197 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2198 | } |
2199 | unlock_mount_hash(); | |
0b1b901b | 2200 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2201 | out: |
d2921684 | 2202 | ns->pending_mounts = 0; |
1064f874 EB |
2203 | |
2204 | read_seqlock_excl(&mount_lock); | |
2205 | put_mountpoint(smp); | |
2206 | read_sequnlock_excl(&mount_lock); | |
2207 | ||
719f5d7f | 2208 | return err; |
b90fa9ae RP |
2209 | } |
2210 | ||
84d17192 | 2211 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2212 | { |
2213 | struct vfsmount *mnt; | |
84d17192 | 2214 | struct dentry *dentry = path->dentry; |
b12cea91 | 2215 | retry: |
5955102c | 2216 | inode_lock(dentry->d_inode); |
84d17192 | 2217 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2218 | inode_unlock(dentry->d_inode); |
84d17192 | 2219 | return ERR_PTR(-ENOENT); |
b12cea91 | 2220 | } |
97216be0 | 2221 | namespace_lock(); |
b12cea91 | 2222 | mnt = lookup_mnt(path); |
84d17192 | 2223 | if (likely(!mnt)) { |
3895dbf8 | 2224 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2225 | if (IS_ERR(mp)) { |
97216be0 | 2226 | namespace_unlock(); |
5955102c | 2227 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2228 | return mp; |
2229 | } | |
2230 | return mp; | |
2231 | } | |
97216be0 | 2232 | namespace_unlock(); |
5955102c | 2233 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2234 | path_put(path); |
2235 | path->mnt = mnt; | |
84d17192 | 2236 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2237 | goto retry; |
2238 | } | |
2239 | ||
84d17192 | 2240 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2241 | { |
84d17192 | 2242 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2243 | |
2244 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2245 | put_mountpoint(where); |
3895dbf8 EB |
2246 | read_sequnlock_excl(&mount_lock); |
2247 | ||
328e6d90 | 2248 | namespace_unlock(); |
5955102c | 2249 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2250 | } |
2251 | ||
84d17192 | 2252 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2253 | { |
e462ec50 | 2254 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2255 | return -EINVAL; |
2256 | ||
e36cb0b8 DH |
2257 | if (d_is_dir(mp->m_dentry) != |
2258 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2259 | return -ENOTDIR; |
2260 | ||
2763d119 | 2261 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2262 | } |
2263 | ||
7a2e8a8f VA |
2264 | /* |
2265 | * Sanity check the flags to change_mnt_propagation. | |
2266 | */ | |
2267 | ||
e462ec50 | 2268 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2269 | { |
e462ec50 | 2270 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2271 | |
2272 | /* Fail if any non-propagation flags are set */ | |
2273 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2274 | return 0; | |
2275 | /* Only one propagation flag should be set */ | |
2276 | if (!is_power_of_2(type)) | |
2277 | return 0; | |
2278 | return type; | |
2279 | } | |
2280 | ||
07b20889 RP |
2281 | /* |
2282 | * recursively change the type of the mountpoint. | |
2283 | */ | |
e462ec50 | 2284 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2285 | { |
315fc83e | 2286 | struct mount *m; |
4b8b21f4 | 2287 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2288 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2289 | int type; |
719f5d7f | 2290 | int err = 0; |
07b20889 | 2291 | |
2d92ab3c | 2292 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2293 | return -EINVAL; |
2294 | ||
e462ec50 | 2295 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2296 | if (!type) |
2297 | return -EINVAL; | |
2298 | ||
97216be0 | 2299 | namespace_lock(); |
719f5d7f MS |
2300 | if (type == MS_SHARED) { |
2301 | err = invent_group_ids(mnt, recurse); | |
2302 | if (err) | |
2303 | goto out_unlock; | |
2304 | } | |
2305 | ||
719ea2fb | 2306 | lock_mount_hash(); |
909b0a88 | 2307 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2308 | change_mnt_propagation(m, type); |
719ea2fb | 2309 | unlock_mount_hash(); |
719f5d7f MS |
2310 | |
2311 | out_unlock: | |
97216be0 | 2312 | namespace_unlock(); |
719f5d7f | 2313 | return err; |
07b20889 RP |
2314 | } |
2315 | ||
5ff9d8a6 EB |
2316 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2317 | { | |
2318 | struct mount *child; | |
2319 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2320 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2321 | continue; | |
2322 | ||
2323 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2324 | return true; | |
2325 | } | |
2326 | return false; | |
2327 | } | |
2328 | ||
a07b2000 AV |
2329 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2330 | { | |
2331 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2332 | ||
2333 | if (IS_MNT_UNBINDABLE(old)) | |
2334 | return mnt; | |
2335 | ||
2336 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2337 | return mnt; | |
2338 | ||
2339 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2340 | return mnt; | |
2341 | ||
2342 | if (recurse) | |
2343 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2344 | else | |
2345 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2346 | ||
2347 | if (!IS_ERR(mnt)) | |
2348 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2349 | ||
2350 | return mnt; | |
2351 | } | |
2352 | ||
1da177e4 LT |
2353 | /* |
2354 | * do loopback mount. | |
2355 | */ | |
808d4e3c | 2356 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2357 | int recurse) |
1da177e4 | 2358 | { |
2d92ab3c | 2359 | struct path old_path; |
a07b2000 | 2360 | struct mount *mnt = NULL, *parent; |
84d17192 | 2361 | struct mountpoint *mp; |
57eccb83 | 2362 | int err; |
1da177e4 LT |
2363 | if (!old_name || !*old_name) |
2364 | return -EINVAL; | |
815d405c | 2365 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2366 | if (err) |
2367 | return err; | |
2368 | ||
8823c079 | 2369 | err = -EINVAL; |
4ce5d2b1 | 2370 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2371 | goto out; |
8823c079 | 2372 | |
84d17192 | 2373 | mp = lock_mount(path); |
a07b2000 AV |
2374 | if (IS_ERR(mp)) { |
2375 | err = PTR_ERR(mp); | |
b12cea91 | 2376 | goto out; |
a07b2000 | 2377 | } |
b12cea91 | 2378 | |
84d17192 | 2379 | parent = real_mount(path->mnt); |
e149ed2b AV |
2380 | if (!check_mnt(parent)) |
2381 | goto out2; | |
2382 | ||
a07b2000 | 2383 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2384 | if (IS_ERR(mnt)) { |
2385 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2386 | goto out2; |
be34d1a3 | 2387 | } |
ccd48bc7 | 2388 | |
84d17192 | 2389 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2390 | if (err) { |
719ea2fb | 2391 | lock_mount_hash(); |
e819f152 | 2392 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2393 | unlock_mount_hash(); |
5b83d2c5 | 2394 | } |
b12cea91 | 2395 | out2: |
84d17192 | 2396 | unlock_mount(mp); |
ccd48bc7 | 2397 | out: |
2d92ab3c | 2398 | path_put(&old_path); |
1da177e4 LT |
2399 | return err; |
2400 | } | |
2401 | ||
a07b2000 AV |
2402 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2403 | { | |
2404 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2405 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2406 | struct mount *mnt, *p; | |
2407 | struct file *file; | |
2408 | ||
2409 | if (IS_ERR(ns)) | |
2410 | return ERR_CAST(ns); | |
2411 | ||
2412 | namespace_lock(); | |
2413 | mnt = __do_loopback(path, recursive); | |
2414 | if (IS_ERR(mnt)) { | |
2415 | namespace_unlock(); | |
2416 | free_mnt_ns(ns); | |
2417 | return ERR_CAST(mnt); | |
2418 | } | |
2419 | ||
2420 | lock_mount_hash(); | |
2421 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2422 | p->mnt_ns = ns; | |
2423 | ns->mounts++; | |
2424 | } | |
2425 | ns->root = mnt; | |
2426 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2427 | mntget(&mnt->mnt); | |
2428 | unlock_mount_hash(); | |
2429 | namespace_unlock(); | |
2430 | ||
2431 | mntput(path->mnt); | |
2432 | path->mnt = &mnt->mnt; | |
2433 | file = dentry_open(path, O_PATH, current_cred()); | |
2434 | if (IS_ERR(file)) | |
2435 | dissolve_on_fput(path->mnt); | |
2436 | else | |
2437 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2438 | return file; | |
2439 | } | |
2440 | ||
2658ce09 | 2441 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2442 | { |
2443 | struct file *file; | |
2444 | struct path path; | |
2445 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2446 | bool detached = flags & OPEN_TREE_CLONE; | |
2447 | int error; | |
2448 | int fd; | |
2449 | ||
2450 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2451 | ||
2452 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2453 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2454 | OPEN_TREE_CLOEXEC)) | |
2455 | return -EINVAL; | |
2456 | ||
2457 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2458 | return -EINVAL; | |
2459 | ||
2460 | if (flags & AT_NO_AUTOMOUNT) | |
2461 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2462 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2463 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2464 | if (flags & AT_EMPTY_PATH) | |
2465 | lookup_flags |= LOOKUP_EMPTY; | |
2466 | ||
2467 | if (detached && !may_mount()) | |
2468 | return -EPERM; | |
2469 | ||
2470 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2471 | if (fd < 0) | |
2472 | return fd; | |
2473 | ||
2474 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2475 | if (unlikely(error)) { | |
2476 | file = ERR_PTR(error); | |
2477 | } else { | |
2478 | if (detached) | |
2479 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2480 | else | |
2481 | file = dentry_open(&path, O_PATH, current_cred()); | |
2482 | path_put(&path); | |
2483 | } | |
2484 | if (IS_ERR(file)) { | |
2485 | put_unused_fd(fd); | |
2486 | return PTR_ERR(file); | |
2487 | } | |
2488 | fd_install(fd, file); | |
2489 | return fd; | |
2490 | } | |
2491 | ||
43f5e655 DH |
2492 | /* |
2493 | * Don't allow locked mount flags to be cleared. | |
2494 | * | |
2495 | * No locks need to be held here while testing the various MNT_LOCK | |
2496 | * flags because those flags can never be cleared once they are set. | |
2497 | */ | |
2498 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2499 | { |
43f5e655 DH |
2500 | unsigned int fl = mnt->mnt.mnt_flags; |
2501 | ||
2502 | if ((fl & MNT_LOCK_READONLY) && | |
2503 | !(mnt_flags & MNT_READONLY)) | |
2504 | return false; | |
2505 | ||
2506 | if ((fl & MNT_LOCK_NODEV) && | |
2507 | !(mnt_flags & MNT_NODEV)) | |
2508 | return false; | |
2509 | ||
2510 | if ((fl & MNT_LOCK_NOSUID) && | |
2511 | !(mnt_flags & MNT_NOSUID)) | |
2512 | return false; | |
2513 | ||
2514 | if ((fl & MNT_LOCK_NOEXEC) && | |
2515 | !(mnt_flags & MNT_NOEXEC)) | |
2516 | return false; | |
2517 | ||
2518 | if ((fl & MNT_LOCK_ATIME) && | |
2519 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2520 | return false; | |
2e4b7fcd | 2521 | |
43f5e655 DH |
2522 | return true; |
2523 | } | |
2524 | ||
2525 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2526 | { |
43f5e655 | 2527 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2528 | |
43f5e655 | 2529 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2530 | return 0; |
2531 | ||
2532 | if (readonly_request) | |
43f5e655 DH |
2533 | return mnt_make_readonly(mnt); |
2534 | ||
68847c94 CB |
2535 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2536 | return 0; | |
43f5e655 DH |
2537 | } |
2538 | ||
43f5e655 DH |
2539 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2540 | { | |
43f5e655 DH |
2541 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2542 | mnt->mnt.mnt_flags = mnt_flags; | |
2543 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2544 | } |
2545 | ||
f8b92ba6 DD |
2546 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2547 | { | |
2548 | struct super_block *sb = mnt->mnt_sb; | |
2549 | ||
2550 | if (!__mnt_is_readonly(mnt) && | |
2551 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { | |
2552 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2553 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
2554 | struct tm tm; | |
2555 | ||
2556 | time64_to_tm(sb->s_time_max, 0, &tm); | |
2557 | ||
0ecee669 EB |
2558 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
2559 | sb->s_type->name, | |
2560 | is_mounted(mnt) ? "remounted" : "mounted", | |
2561 | mntpath, | |
f8b92ba6 DD |
2562 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
2563 | ||
2564 | free_page((unsigned long)buf); | |
2565 | } | |
2566 | } | |
2567 | ||
43f5e655 DH |
2568 | /* |
2569 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2570 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2571 | * to mount(2). | |
2572 | */ | |
2573 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2574 | { | |
2575 | struct super_block *sb = path->mnt->mnt_sb; | |
2576 | struct mount *mnt = real_mount(path->mnt); | |
2577 | int ret; | |
2578 | ||
2579 | if (!check_mnt(mnt)) | |
2580 | return -EINVAL; | |
2581 | ||
2582 | if (path->dentry != mnt->mnt.mnt_root) | |
2583 | return -EINVAL; | |
2584 | ||
2585 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2586 | return -EPERM; | |
2587 | ||
e58ace1a CB |
2588 | /* |
2589 | * We're only checking whether the superblock is read-only not | |
2590 | * changing it, so only take down_read(&sb->s_umount). | |
2591 | */ | |
2592 | down_read(&sb->s_umount); | |
68847c94 | 2593 | lock_mount_hash(); |
43f5e655 DH |
2594 | ret = change_mount_ro_state(mnt, mnt_flags); |
2595 | if (ret == 0) | |
2596 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2597 | unlock_mount_hash(); |
e58ace1a | 2598 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2599 | |
2600 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2601 | ||
43f5e655 | 2602 | return ret; |
2e4b7fcd DH |
2603 | } |
2604 | ||
1da177e4 LT |
2605 | /* |
2606 | * change filesystem flags. dir should be a physical root of filesystem. | |
2607 | * If you've mounted a non-root directory somewhere and want to do remount | |
2608 | * on it - tough luck. | |
2609 | */ | |
e462ec50 DH |
2610 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2611 | int mnt_flags, void *data) | |
1da177e4 LT |
2612 | { |
2613 | int err; | |
2d92ab3c | 2614 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2615 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2616 | struct fs_context *fc; |
1da177e4 | 2617 | |
143c8c91 | 2618 | if (!check_mnt(mnt)) |
1da177e4 LT |
2619 | return -EINVAL; |
2620 | ||
2d92ab3c | 2621 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2622 | return -EINVAL; |
2623 | ||
43f5e655 | 2624 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2625 | return -EPERM; |
9566d674 | 2626 | |
8d0347f6 DH |
2627 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2628 | if (IS_ERR(fc)) | |
2629 | return PTR_ERR(fc); | |
ff36fe2c | 2630 | |
b330966f | 2631 | fc->oldapi = true; |
8d0347f6 DH |
2632 | err = parse_monolithic_mount_data(fc, data); |
2633 | if (!err) { | |
2634 | down_write(&sb->s_umount); | |
2635 | err = -EPERM; | |
2636 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2637 | err = reconfigure_super(fc); | |
68847c94 CB |
2638 | if (!err) { |
2639 | lock_mount_hash(); | |
8d0347f6 | 2640 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2641 | unlock_mount_hash(); |
2642 | } | |
8d0347f6 DH |
2643 | } |
2644 | up_write(&sb->s_umount); | |
0e55a7cc | 2645 | } |
f8b92ba6 DD |
2646 | |
2647 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2648 | ||
8d0347f6 | 2649 | put_fs_context(fc); |
1da177e4 LT |
2650 | return err; |
2651 | } | |
2652 | ||
cbbe362c | 2653 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2654 | { |
315fc83e | 2655 | struct mount *p; |
909b0a88 | 2656 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2657 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2658 | return 1; |
2659 | } | |
2660 | return 0; | |
2661 | } | |
2662 | ||
44dfd84a DH |
2663 | /* |
2664 | * Check that there aren't references to earlier/same mount namespaces in the | |
2665 | * specified subtree. Such references can act as pins for mount namespaces | |
2666 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2667 | * cycles to be made. | |
2668 | */ | |
2669 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2670 | { | |
2671 | struct mount *p; | |
2672 | bool ret = false; | |
2673 | ||
2674 | lock_mount_hash(); | |
2675 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2676 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2677 | goto out; | |
2678 | ||
2679 | ret = true; | |
2680 | out: | |
2681 | unlock_mount_hash(); | |
2682 | return ret; | |
2683 | } | |
2684 | ||
2db154b3 | 2685 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2686 | { |
44dfd84a | 2687 | struct mnt_namespace *ns; |
676da58d | 2688 | struct mount *p; |
0fb54e50 | 2689 | struct mount *old; |
2763d119 AV |
2690 | struct mount *parent; |
2691 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2692 | int err; |
44dfd84a | 2693 | bool attached; |
1da177e4 | 2694 | |
2db154b3 | 2695 | mp = lock_mount(new_path); |
84d17192 | 2696 | if (IS_ERR(mp)) |
2db154b3 | 2697 | return PTR_ERR(mp); |
cc53ce53 | 2698 | |
2db154b3 DH |
2699 | old = real_mount(old_path->mnt); |
2700 | p = real_mount(new_path->mnt); | |
2763d119 | 2701 | parent = old->mnt_parent; |
44dfd84a | 2702 | attached = mnt_has_parent(old); |
2763d119 | 2703 | old_mp = old->mnt_mp; |
44dfd84a | 2704 | ns = old->mnt_ns; |
143c8c91 | 2705 | |
1da177e4 | 2706 | err = -EINVAL; |
44dfd84a DH |
2707 | /* The mountpoint must be in our namespace. */ |
2708 | if (!check_mnt(p)) | |
2db154b3 | 2709 | goto out; |
1da177e4 | 2710 | |
570d7a98 EB |
2711 | /* The thing moved must be mounted... */ |
2712 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2713 | goto out; |
2714 | ||
570d7a98 EB |
2715 | /* ... and either ours or the root of anon namespace */ |
2716 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2717 | goto out; |
5ff9d8a6 | 2718 | |
2db154b3 DH |
2719 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2720 | goto out; | |
1da177e4 | 2721 | |
2db154b3 DH |
2722 | if (old_path->dentry != old_path->mnt->mnt_root) |
2723 | goto out; | |
1da177e4 | 2724 | |
2db154b3 DH |
2725 | if (d_is_dir(new_path->dentry) != |
2726 | d_is_dir(old_path->dentry)) | |
2727 | goto out; | |
21444403 RP |
2728 | /* |
2729 | * Don't move a mount residing in a shared parent. | |
2730 | */ | |
2763d119 | 2731 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2732 | goto out; |
9676f0c6 RP |
2733 | /* |
2734 | * Don't move a mount tree containing unbindable mounts to a destination | |
2735 | * mount which is shared. | |
2736 | */ | |
fc7be130 | 2737 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2738 | goto out; |
1da177e4 | 2739 | err = -ELOOP; |
44dfd84a DH |
2740 | if (!check_for_nsfs_mounts(old)) |
2741 | goto out; | |
fc7be130 | 2742 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2743 | if (p == old) |
2db154b3 | 2744 | goto out; |
1da177e4 | 2745 | |
2db154b3 | 2746 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2747 | attached); |
4ac91378 | 2748 | if (err) |
2db154b3 | 2749 | goto out; |
1da177e4 LT |
2750 | |
2751 | /* if the mount is moved, it should no longer be expire | |
2752 | * automatically */ | |
6776db3d | 2753 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2754 | if (attached) |
2755 | put_mountpoint(old_mp); | |
1da177e4 | 2756 | out: |
2db154b3 | 2757 | unlock_mount(mp); |
44dfd84a | 2758 | if (!err) { |
2763d119 AV |
2759 | if (attached) |
2760 | mntput_no_expire(parent); | |
2761 | else | |
44dfd84a DH |
2762 | free_mnt_ns(ns); |
2763 | } | |
2db154b3 DH |
2764 | return err; |
2765 | } | |
2766 | ||
2767 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2768 | { | |
2769 | struct path old_path; | |
2770 | int err; | |
2771 | ||
2772 | if (!old_name || !*old_name) | |
2773 | return -EINVAL; | |
2774 | ||
2775 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2776 | if (err) | |
2777 | return err; | |
2778 | ||
2779 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2780 | path_put(&old_path); |
1da177e4 LT |
2781 | return err; |
2782 | } | |
2783 | ||
9d412a43 AV |
2784 | /* |
2785 | * add a mount into a namespace's mount tree | |
2786 | */ | |
8f11538e AV |
2787 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
2788 | struct path *path, int mnt_flags) | |
9d412a43 | 2789 | { |
8f11538e | 2790 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 2791 | |
f2ebb3a9 | 2792 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2793 | |
84d17192 | 2794 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2795 | /* that's acceptable only for automounts done in private ns */ |
2796 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 2797 | return -EINVAL; |
156cacb1 | 2798 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 2799 | if (!parent->mnt_ns) |
8f11538e | 2800 | return -EINVAL; |
156cacb1 | 2801 | } |
9d412a43 AV |
2802 | |
2803 | /* Refuse the same filesystem on the same mount point */ | |
95bc5f25 | 2804 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 | 2805 | path->mnt->mnt_root == path->dentry) |
8f11538e | 2806 | return -EBUSY; |
9d412a43 | 2807 | |
e36cb0b8 | 2808 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 2809 | return -EINVAL; |
9d412a43 | 2810 | |
95bc5f25 | 2811 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 2812 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 2813 | } |
b1e75df4 | 2814 | |
132e4608 DH |
2815 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2816 | ||
2817 | /* | |
2818 | * Create a new mount using a superblock configuration and request it | |
2819 | * be added to the namespace tree. | |
2820 | */ | |
2821 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2822 | unsigned int mnt_flags) | |
2823 | { | |
2824 | struct vfsmount *mnt; | |
8f11538e | 2825 | struct mountpoint *mp; |
132e4608 DH |
2826 | struct super_block *sb = fc->root->d_sb; |
2827 | int error; | |
2828 | ||
c9ce29ed AV |
2829 | error = security_sb_kern_mount(sb); |
2830 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2831 | error = -EPERM; | |
2832 | ||
2833 | if (unlikely(error)) { | |
2834 | fc_drop_locked(fc); | |
2835 | return error; | |
132e4608 DH |
2836 | } |
2837 | ||
2838 | up_write(&sb->s_umount); | |
2839 | ||
2840 | mnt = vfs_create_mount(fc); | |
2841 | if (IS_ERR(mnt)) | |
2842 | return PTR_ERR(mnt); | |
2843 | ||
f8b92ba6 DD |
2844 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
2845 | ||
8f11538e AV |
2846 | mp = lock_mount(mountpoint); |
2847 | if (IS_ERR(mp)) { | |
2848 | mntput(mnt); | |
2849 | return PTR_ERR(mp); | |
2850 | } | |
2851 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
2852 | unlock_mount(mp); | |
0ecee669 EB |
2853 | if (error < 0) |
2854 | mntput(mnt); | |
132e4608 DH |
2855 | return error; |
2856 | } | |
1b852bce | 2857 | |
1da177e4 LT |
2858 | /* |
2859 | * create a new mount for userspace and request it to be added into the | |
2860 | * namespace's tree | |
2861 | */ | |
e462ec50 | 2862 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2863 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2864 | { |
0c55cfc4 | 2865 | struct file_system_type *type; |
a0c9a8b8 AV |
2866 | struct fs_context *fc; |
2867 | const char *subtype = NULL; | |
2868 | int err = 0; | |
1da177e4 | 2869 | |
0c55cfc4 | 2870 | if (!fstype) |
1da177e4 LT |
2871 | return -EINVAL; |
2872 | ||
0c55cfc4 EB |
2873 | type = get_fs_type(fstype); |
2874 | if (!type) | |
2875 | return -ENODEV; | |
2876 | ||
a0c9a8b8 AV |
2877 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2878 | subtype = strchr(fstype, '.'); | |
2879 | if (subtype) { | |
2880 | subtype++; | |
2881 | if (!*subtype) { | |
2882 | put_filesystem(type); | |
2883 | return -EINVAL; | |
2884 | } | |
a0c9a8b8 AV |
2885 | } |
2886 | } | |
0c55cfc4 | 2887 | |
a0c9a8b8 | 2888 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2889 | put_filesystem(type); |
a0c9a8b8 AV |
2890 | if (IS_ERR(fc)) |
2891 | return PTR_ERR(fc); | |
2892 | ||
3e1aeb00 DH |
2893 | if (subtype) |
2894 | err = vfs_parse_fs_string(fc, "subtype", | |
2895 | subtype, strlen(subtype)); | |
2896 | if (!err && name) | |
2897 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2898 | if (!err) |
2899 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
2900 | if (!err && !mount_capable(fc)) |
2901 | err = -EPERM; | |
a0c9a8b8 AV |
2902 | if (!err) |
2903 | err = vfs_get_tree(fc); | |
132e4608 DH |
2904 | if (!err) |
2905 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2906 | |
a0c9a8b8 | 2907 | put_fs_context(fc); |
15f9a3f3 | 2908 | return err; |
1da177e4 LT |
2909 | } |
2910 | ||
19a167af AV |
2911 | int finish_automount(struct vfsmount *m, struct path *path) |
2912 | { | |
26df6034 | 2913 | struct dentry *dentry = path->dentry; |
8f11538e | 2914 | struct mountpoint *mp; |
25e195aa | 2915 | struct mount *mnt; |
19a167af | 2916 | int err; |
25e195aa AV |
2917 | |
2918 | if (!m) | |
2919 | return 0; | |
2920 | if (IS_ERR(m)) | |
2921 | return PTR_ERR(m); | |
2922 | ||
2923 | mnt = real_mount(m); | |
19a167af AV |
2924 | /* The new mount record should have at least 2 refs to prevent it being |
2925 | * expired before we get a chance to add it | |
2926 | */ | |
6776db3d | 2927 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2928 | |
2929 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 2930 | m->mnt_root == dentry) { |
b1e75df4 | 2931 | err = -ELOOP; |
26df6034 | 2932 | goto discard; |
19a167af AV |
2933 | } |
2934 | ||
26df6034 AV |
2935 | /* |
2936 | * we don't want to use lock_mount() - in this case finding something | |
2937 | * that overmounts our mountpoint to be means "quitely drop what we've | |
2938 | * got", not "try to mount it on top". | |
2939 | */ | |
2940 | inode_lock(dentry->d_inode); | |
2941 | namespace_lock(); | |
2942 | if (unlikely(cant_mount(dentry))) { | |
2943 | err = -ENOENT; | |
2944 | goto discard_locked; | |
2945 | } | |
2946 | rcu_read_lock(); | |
2947 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { | |
2948 | rcu_read_unlock(); | |
2949 | err = 0; | |
2950 | goto discard_locked; | |
2951 | } | |
2952 | rcu_read_unlock(); | |
2953 | mp = get_mountpoint(dentry); | |
8f11538e AV |
2954 | if (IS_ERR(mp)) { |
2955 | err = PTR_ERR(mp); | |
26df6034 | 2956 | goto discard_locked; |
8f11538e | 2957 | } |
26df6034 | 2958 | |
8f11538e AV |
2959 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
2960 | unlock_mount(mp); | |
26df6034 AV |
2961 | if (unlikely(err)) |
2962 | goto discard; | |
2963 | mntput(m); | |
2964 | return 0; | |
2965 | ||
2966 | discard_locked: | |
2967 | namespace_unlock(); | |
2968 | inode_unlock(dentry->d_inode); | |
2969 | discard: | |
b1e75df4 | 2970 | /* remove m from any expiration list it may be on */ |
6776db3d | 2971 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2972 | namespace_lock(); |
6776db3d | 2973 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2974 | namespace_unlock(); |
19a167af | 2975 | } |
b1e75df4 AV |
2976 | mntput(m); |
2977 | mntput(m); | |
19a167af AV |
2978 | return err; |
2979 | } | |
2980 | ||
ea5b778a DH |
2981 | /** |
2982 | * mnt_set_expiry - Put a mount on an expiration list | |
2983 | * @mnt: The mount to list. | |
2984 | * @expiry_list: The list to add the mount to. | |
2985 | */ | |
2986 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2987 | { | |
97216be0 | 2988 | namespace_lock(); |
ea5b778a | 2989 | |
6776db3d | 2990 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2991 | |
97216be0 | 2992 | namespace_unlock(); |
ea5b778a DH |
2993 | } |
2994 | EXPORT_SYMBOL(mnt_set_expiry); | |
2995 | ||
1da177e4 LT |
2996 | /* |
2997 | * process a list of expirable mountpoints with the intent of discarding any | |
2998 | * mountpoints that aren't in use and haven't been touched since last we came | |
2999 | * here | |
3000 | */ | |
3001 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3002 | { | |
761d5c38 | 3003 | struct mount *mnt, *next; |
1da177e4 LT |
3004 | LIST_HEAD(graveyard); |
3005 | ||
3006 | if (list_empty(mounts)) | |
3007 | return; | |
3008 | ||
97216be0 | 3009 | namespace_lock(); |
719ea2fb | 3010 | lock_mount_hash(); |
1da177e4 LT |
3011 | |
3012 | /* extract from the expiration list every vfsmount that matches the | |
3013 | * following criteria: | |
3014 | * - only referenced by its parent vfsmount | |
3015 | * - still marked for expiry (marked on the last call here; marks are | |
3016 | * cleared by mntput()) | |
3017 | */ | |
6776db3d | 3018 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3019 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3020 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3021 | continue; |
6776db3d | 3022 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3023 | } |
bcc5c7d2 | 3024 | while (!list_empty(&graveyard)) { |
6776db3d | 3025 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3026 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3027 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3028 | } |
719ea2fb | 3029 | unlock_mount_hash(); |
3ab6abee | 3030 | namespace_unlock(); |
5528f911 TM |
3031 | } |
3032 | ||
3033 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3034 | ||
3035 | /* | |
3036 | * Ripoff of 'select_parent()' | |
3037 | * | |
3038 | * search the list of submounts for a given mountpoint, and move any | |
3039 | * shrinkable submounts to the 'graveyard' list. | |
3040 | */ | |
692afc31 | 3041 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3042 | { |
692afc31 | 3043 | struct mount *this_parent = parent; |
5528f911 TM |
3044 | struct list_head *next; |
3045 | int found = 0; | |
3046 | ||
3047 | repeat: | |
6b41d536 | 3048 | next = this_parent->mnt_mounts.next; |
5528f911 | 3049 | resume: |
6b41d536 | 3050 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3051 | struct list_head *tmp = next; |
6b41d536 | 3052 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3053 | |
3054 | next = tmp->next; | |
692afc31 | 3055 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3056 | continue; |
5528f911 TM |
3057 | /* |
3058 | * Descend a level if the d_mounts list is non-empty. | |
3059 | */ | |
6b41d536 | 3060 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3061 | this_parent = mnt; |
3062 | goto repeat; | |
3063 | } | |
1da177e4 | 3064 | |
1ab59738 | 3065 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3066 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3067 | found++; |
3068 | } | |
1da177e4 | 3069 | } |
5528f911 TM |
3070 | /* |
3071 | * All done at this level ... ascend and resume the search | |
3072 | */ | |
3073 | if (this_parent != parent) { | |
6b41d536 | 3074 | next = this_parent->mnt_child.next; |
0714a533 | 3075 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3076 | goto resume; |
3077 | } | |
3078 | return found; | |
3079 | } | |
3080 | ||
3081 | /* | |
3082 | * process a list of expirable mountpoints with the intent of discarding any | |
3083 | * submounts of a specific parent mountpoint | |
99b7db7b | 3084 | * |
48a066e7 | 3085 | * mount_lock must be held for write |
5528f911 | 3086 | */ |
b54b9be7 | 3087 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3088 | { |
3089 | LIST_HEAD(graveyard); | |
761d5c38 | 3090 | struct mount *m; |
5528f911 | 3091 | |
5528f911 | 3092 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3093 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3094 | while (!list_empty(&graveyard)) { |
761d5c38 | 3095 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3096 | mnt_expire); |
143c8c91 | 3097 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3098 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3099 | } |
3100 | } | |
1da177e4 LT |
3101 | } |
3102 | ||
028abd92 | 3103 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3104 | { |
b40ef869 | 3105 | char *copy; |
d563d678 | 3106 | unsigned left, offset; |
b58fed8b | 3107 | |
1da177e4 | 3108 | if (!data) |
b40ef869 | 3109 | return NULL; |
1da177e4 | 3110 | |
b40ef869 AV |
3111 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3112 | if (!copy) | |
3113 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3114 | |
d563d678 | 3115 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3116 | |
d563d678 CM |
3117 | /* |
3118 | * Not all architectures have an exact copy_from_user(). Resort to | |
3119 | * byte at a time. | |
3120 | */ | |
3121 | offset = PAGE_SIZE - left; | |
3122 | while (left) { | |
3123 | char c; | |
3124 | if (get_user(c, (const char __user *)data + offset)) | |
3125 | break; | |
3126 | copy[offset] = c; | |
3127 | left--; | |
3128 | offset++; | |
3129 | } | |
3130 | ||
3131 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3132 | kfree(copy); |
3133 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3134 | } |
d563d678 | 3135 | |
b40ef869 | 3136 | return copy; |
1da177e4 LT |
3137 | } |
3138 | ||
028abd92 | 3139 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3140 | { |
fbdb4401 | 3141 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3142 | } |
3143 | ||
1da177e4 LT |
3144 | /* |
3145 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3146 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3147 | * | |
3148 | * data is a (void *) that can point to any structure up to | |
3149 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3150 | * information (or be NULL). | |
3151 | * | |
3152 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3153 | * When the flags word was introduced its top half was required | |
3154 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3155 | * Therefore, if this magic number is present, it carries no information | |
3156 | * and must be discarded. | |
3157 | */ | |
c60166f0 | 3158 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3159 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3160 | { |
e462ec50 | 3161 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3162 | int ret; |
1da177e4 LT |
3163 | |
3164 | /* Discard magic */ | |
3165 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3166 | flags &= ~MS_MGC_MSK; | |
3167 | ||
3168 | /* Basic sanity checks */ | |
1da177e4 LT |
3169 | if (data_page) |
3170 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3171 | ||
e462ec50 DH |
3172 | if (flags & MS_NOUSER) |
3173 | return -EINVAL; | |
3174 | ||
a1e6aaa3 CH |
3175 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3176 | if (ret) | |
3177 | return ret; | |
3178 | if (!may_mount()) | |
3179 | return -EPERM; | |
3180 | if ((flags & SB_MANDLOCK) && !may_mandlock()) | |
3181 | return -EPERM; | |
a27ab9f2 | 3182 | |
613cbe3d AK |
3183 | /* Default to relatime unless overriden */ |
3184 | if (!(flags & MS_NOATIME)) | |
3185 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3186 | |
1da177e4 LT |
3187 | /* Separate the per-mountpoint flags */ |
3188 | if (flags & MS_NOSUID) | |
3189 | mnt_flags |= MNT_NOSUID; | |
3190 | if (flags & MS_NODEV) | |
3191 | mnt_flags |= MNT_NODEV; | |
3192 | if (flags & MS_NOEXEC) | |
3193 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3194 | if (flags & MS_NOATIME) |
3195 | mnt_flags |= MNT_NOATIME; | |
3196 | if (flags & MS_NODIRATIME) | |
3197 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3198 | if (flags & MS_STRICTATIME) |
3199 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3200 | if (flags & MS_RDONLY) |
2e4b7fcd | 3201 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3202 | if (flags & MS_NOSYMFOLLOW) |
3203 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3204 | |
ffbc6f0e EB |
3205 | /* The default atime for remount is preservation */ |
3206 | if ((flags & MS_REMOUNT) && | |
3207 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3208 | MS_STRICTATIME)) == 0)) { | |
3209 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3210 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3211 | } |
3212 | ||
e462ec50 DH |
3213 | sb_flags = flags & (SB_RDONLY | |
3214 | SB_SYNCHRONOUS | | |
3215 | SB_MANDLOCK | | |
3216 | SB_DIRSYNC | | |
3217 | SB_SILENT | | |
917086ff | 3218 | SB_POSIXACL | |
d7ee9469 | 3219 | SB_LAZYTIME | |
917086ff | 3220 | SB_I_VERSION); |
1da177e4 | 3221 | |
43f5e655 | 3222 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3223 | return do_reconfigure_mnt(path, mnt_flags); |
3224 | if (flags & MS_REMOUNT) | |
3225 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3226 | if (flags & MS_BIND) | |
3227 | return do_loopback(path, dev_name, flags & MS_REC); | |
3228 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3229 | return do_change_type(path, flags); | |
3230 | if (flags & MS_MOVE) | |
3231 | return do_move_mount_old(path, dev_name); | |
3232 | ||
3233 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3234 | data_page); | |
3235 | } | |
3236 | ||
3237 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3238 | const char *type_page, unsigned long flags, void *data_page) | |
3239 | { | |
3240 | struct path path; | |
3241 | int ret; | |
3242 | ||
3243 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3244 | if (ret) | |
3245 | return ret; | |
3246 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3247 | path_put(&path); |
a1e6aaa3 | 3248 | return ret; |
1da177e4 LT |
3249 | } |
3250 | ||
537f7ccb EB |
3251 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3252 | { | |
3253 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3254 | } | |
3255 | ||
3256 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3257 | { | |
3258 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3259 | } | |
3260 | ||
771b1371 EB |
3261 | static void free_mnt_ns(struct mnt_namespace *ns) |
3262 | { | |
74e83122 AV |
3263 | if (!is_anon_ns(ns)) |
3264 | ns_free_inum(&ns->ns); | |
537f7ccb | 3265 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3266 | put_user_ns(ns->user_ns); |
3267 | kfree(ns); | |
3268 | } | |
3269 | ||
8823c079 EB |
3270 | /* |
3271 | * Assign a sequence number so we can detect when we attempt to bind | |
3272 | * mount a reference to an older mount namespace into the current | |
3273 | * mount namespace, preventing reference counting loops. A 64bit | |
3274 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3275 | * is effectively never, so we can ignore the possibility. | |
3276 | */ | |
3277 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3278 | ||
74e83122 | 3279 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3280 | { |
3281 | struct mnt_namespace *new_ns; | |
537f7ccb | 3282 | struct ucounts *ucounts; |
98f842e6 | 3283 | int ret; |
cf8d2c11 | 3284 | |
537f7ccb EB |
3285 | ucounts = inc_mnt_namespaces(user_ns); |
3286 | if (!ucounts) | |
df75e774 | 3287 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3288 | |
74e83122 | 3289 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
3290 | if (!new_ns) { |
3291 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3292 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3293 | } |
74e83122 AV |
3294 | if (!anon) { |
3295 | ret = ns_alloc_inum(&new_ns->ns); | |
3296 | if (ret) { | |
3297 | kfree(new_ns); | |
3298 | dec_mnt_namespaces(ucounts); | |
3299 | return ERR_PTR(ret); | |
3300 | } | |
98f842e6 | 3301 | } |
33c42940 | 3302 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3303 | if (!anon) |
3304 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3305 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3306 | INIT_LIST_HEAD(&new_ns->list); |
3307 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3308 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3309 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3310 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3311 | return new_ns; |
3312 | } | |
3313 | ||
0766f788 | 3314 | __latent_entropy |
9559f689 AV |
3315 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3316 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3317 | { |
6b3286ed | 3318 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3319 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3320 | struct mount *p, *q; |
9559f689 | 3321 | struct mount *old; |
cb338d06 | 3322 | struct mount *new; |
7a472ef4 | 3323 | int copy_flags; |
1da177e4 | 3324 | |
9559f689 AV |
3325 | BUG_ON(!ns); |
3326 | ||
3327 | if (likely(!(flags & CLONE_NEWNS))) { | |
3328 | get_mnt_ns(ns); | |
3329 | return ns; | |
3330 | } | |
3331 | ||
3332 | old = ns->root; | |
3333 | ||
74e83122 | 3334 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3335 | if (IS_ERR(new_ns)) |
3336 | return new_ns; | |
1da177e4 | 3337 | |
97216be0 | 3338 | namespace_lock(); |
1da177e4 | 3339 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3340 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3341 | if (user_ns != ns->user_ns) |
3bd045cc | 3342 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3343 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3344 | if (IS_ERR(new)) { |
328e6d90 | 3345 | namespace_unlock(); |
771b1371 | 3346 | free_mnt_ns(new_ns); |
be34d1a3 | 3347 | return ERR_CAST(new); |
1da177e4 | 3348 | } |
3bd045cc AV |
3349 | if (user_ns != ns->user_ns) { |
3350 | lock_mount_hash(); | |
3351 | lock_mnt_tree(new); | |
3352 | unlock_mount_hash(); | |
3353 | } | |
be08d6d2 | 3354 | new_ns->root = new; |
1a4eeaf2 | 3355 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3356 | |
3357 | /* | |
3358 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3359 | * as belonging to new namespace. We have already acquired a private | |
3360 | * fs_struct, so tsk->fs->lock is not needed. | |
3361 | */ | |
909b0a88 | 3362 | p = old; |
cb338d06 | 3363 | q = new; |
1da177e4 | 3364 | while (p) { |
143c8c91 | 3365 | q->mnt_ns = new_ns; |
d2921684 | 3366 | new_ns->mounts++; |
9559f689 AV |
3367 | if (new_fs) { |
3368 | if (&p->mnt == new_fs->root.mnt) { | |
3369 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3370 | rootmnt = &p->mnt; |
1da177e4 | 3371 | } |
9559f689 AV |
3372 | if (&p->mnt == new_fs->pwd.mnt) { |
3373 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3374 | pwdmnt = &p->mnt; |
1da177e4 | 3375 | } |
1da177e4 | 3376 | } |
909b0a88 AV |
3377 | p = next_mnt(p, old); |
3378 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3379 | if (!q) |
3380 | break; | |
3381 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3382 | p = next_mnt(p, old); | |
1da177e4 | 3383 | } |
328e6d90 | 3384 | namespace_unlock(); |
1da177e4 | 3385 | |
1da177e4 | 3386 | if (rootmnt) |
f03c6599 | 3387 | mntput(rootmnt); |
1da177e4 | 3388 | if (pwdmnt) |
f03c6599 | 3389 | mntput(pwdmnt); |
1da177e4 | 3390 | |
741a2951 | 3391 | return new_ns; |
1da177e4 LT |
3392 | } |
3393 | ||
74e83122 | 3394 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3395 | { |
74e83122 | 3396 | struct mount *mnt = real_mount(m); |
ea441d11 | 3397 | struct mnt_namespace *ns; |
d31da0f0 | 3398 | struct super_block *s; |
ea441d11 AV |
3399 | struct path path; |
3400 | int err; | |
3401 | ||
74e83122 AV |
3402 | ns = alloc_mnt_ns(&init_user_ns, true); |
3403 | if (IS_ERR(ns)) { | |
3404 | mntput(m); | |
ea441d11 | 3405 | return ERR_CAST(ns); |
74e83122 AV |
3406 | } |
3407 | mnt->mnt_ns = ns; | |
3408 | ns->root = mnt; | |
3409 | ns->mounts++; | |
3410 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3411 | |
74e83122 | 3412 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3413 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3414 | ||
3415 | put_mnt_ns(ns); | |
3416 | ||
3417 | if (err) | |
3418 | return ERR_PTR(err); | |
3419 | ||
3420 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3421 | s = path.mnt->mnt_sb; |
3422 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3423 | mntput(path.mnt); |
3424 | /* lock the sucker */ | |
d31da0f0 | 3425 | down_write(&s->s_umount); |
ea441d11 AV |
3426 | /* ... and return the root of (sub)tree on it */ |
3427 | return path.dentry; | |
3428 | } | |
3429 | EXPORT_SYMBOL(mount_subtree); | |
3430 | ||
cccaa5e3 DB |
3431 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3432 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3433 | { |
eca6f534 VN |
3434 | int ret; |
3435 | char *kernel_type; | |
eca6f534 | 3436 | char *kernel_dev; |
b40ef869 | 3437 | void *options; |
1da177e4 | 3438 | |
b8850d1f TG |
3439 | kernel_type = copy_mount_string(type); |
3440 | ret = PTR_ERR(kernel_type); | |
3441 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3442 | goto out_type; |
1da177e4 | 3443 | |
b8850d1f TG |
3444 | kernel_dev = copy_mount_string(dev_name); |
3445 | ret = PTR_ERR(kernel_dev); | |
3446 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3447 | goto out_dev; |
1da177e4 | 3448 | |
b40ef869 AV |
3449 | options = copy_mount_options(data); |
3450 | ret = PTR_ERR(options); | |
3451 | if (IS_ERR(options)) | |
eca6f534 | 3452 | goto out_data; |
1da177e4 | 3453 | |
b40ef869 | 3454 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3455 | |
b40ef869 | 3456 | kfree(options); |
eca6f534 VN |
3457 | out_data: |
3458 | kfree(kernel_dev); | |
3459 | out_dev: | |
eca6f534 VN |
3460 | kfree(kernel_type); |
3461 | out_type: | |
3462 | return ret; | |
1da177e4 LT |
3463 | } |
3464 | ||
5b490500 CB |
3465 | #define FSMOUNT_VALID_FLAGS \ |
3466 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3467 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME) | |
3468 | ||
9caccd41 | 3469 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
2a186721 CB |
3470 | |
3471 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3472 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3473 | ||
5b490500 CB |
3474 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3475 | { | |
3476 | unsigned int mnt_flags = 0; | |
3477 | ||
3478 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3479 | mnt_flags |= MNT_READONLY; | |
3480 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3481 | mnt_flags |= MNT_NOSUID; | |
3482 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3483 | mnt_flags |= MNT_NODEV; | |
3484 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3485 | mnt_flags |= MNT_NOEXEC; | |
3486 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3487 | mnt_flags |= MNT_NODIRATIME; | |
3488 | ||
3489 | return mnt_flags; | |
3490 | } | |
3491 | ||
2db154b3 | 3492 | /* |
93766fbd DH |
3493 | * Create a kernel mount representation for a new, prepared superblock |
3494 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3495 | */ | |
3496 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3497 | unsigned int, attr_flags) | |
3498 | { | |
3499 | struct mnt_namespace *ns; | |
3500 | struct fs_context *fc; | |
3501 | struct file *file; | |
3502 | struct path newmount; | |
3503 | struct mount *mnt; | |
3504 | struct fd f; | |
3505 | unsigned int mnt_flags = 0; | |
3506 | long ret; | |
3507 | ||
3508 | if (!may_mount()) | |
3509 | return -EPERM; | |
3510 | ||
3511 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3512 | return -EINVAL; | |
3513 | ||
5b490500 | 3514 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3515 | return -EINVAL; |
3516 | ||
5b490500 | 3517 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3518 | |
3519 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3520 | case MOUNT_ATTR_STRICTATIME: | |
3521 | break; | |
3522 | case MOUNT_ATTR_NOATIME: | |
3523 | mnt_flags |= MNT_NOATIME; | |
3524 | break; | |
3525 | case MOUNT_ATTR_RELATIME: | |
3526 | mnt_flags |= MNT_RELATIME; | |
3527 | break; | |
3528 | default: | |
3529 | return -EINVAL; | |
3530 | } | |
3531 | ||
3532 | f = fdget(fs_fd); | |
3533 | if (!f.file) | |
3534 | return -EBADF; | |
3535 | ||
3536 | ret = -EINVAL; | |
3537 | if (f.file->f_op != &fscontext_fops) | |
3538 | goto err_fsfd; | |
3539 | ||
3540 | fc = f.file->private_data; | |
3541 | ||
3542 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3543 | if (ret < 0) | |
3544 | goto err_fsfd; | |
3545 | ||
3546 | /* There must be a valid superblock or we can't mount it */ | |
3547 | ret = -EINVAL; | |
3548 | if (!fc->root) | |
3549 | goto err_unlock; | |
3550 | ||
3551 | ret = -EPERM; | |
3552 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3553 | pr_warn("VFS: Mount too revealing\n"); | |
3554 | goto err_unlock; | |
3555 | } | |
3556 | ||
3557 | ret = -EBUSY; | |
3558 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3559 | goto err_unlock; | |
3560 | ||
3561 | ret = -EPERM; | |
3562 | if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) | |
3563 | goto err_unlock; | |
3564 | ||
3565 | newmount.mnt = vfs_create_mount(fc); | |
3566 | if (IS_ERR(newmount.mnt)) { | |
3567 | ret = PTR_ERR(newmount.mnt); | |
3568 | goto err_unlock; | |
3569 | } | |
3570 | newmount.dentry = dget(fc->root); | |
3571 | newmount.mnt->mnt_flags = mnt_flags; | |
3572 | ||
3573 | /* We've done the mount bit - now move the file context into more or | |
3574 | * less the same state as if we'd done an fspick(). We don't want to | |
3575 | * do any memory allocation or anything like that at this point as we | |
3576 | * don't want to have to handle any errors incurred. | |
3577 | */ | |
3578 | vfs_clean_context(fc); | |
3579 | ||
3580 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3581 | if (IS_ERR(ns)) { | |
3582 | ret = PTR_ERR(ns); | |
3583 | goto err_path; | |
3584 | } | |
3585 | mnt = real_mount(newmount.mnt); | |
3586 | mnt->mnt_ns = ns; | |
3587 | ns->root = mnt; | |
3588 | ns->mounts = 1; | |
3589 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3590 | mntget(newmount.mnt); |
93766fbd DH |
3591 | |
3592 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3593 | * it, not just simply put it. | |
3594 | */ | |
3595 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3596 | if (IS_ERR(file)) { | |
3597 | dissolve_on_fput(newmount.mnt); | |
3598 | ret = PTR_ERR(file); | |
3599 | goto err_path; | |
3600 | } | |
3601 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3602 | ||
3603 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3604 | if (ret >= 0) | |
3605 | fd_install(ret, file); | |
3606 | else | |
3607 | fput(file); | |
3608 | ||
3609 | err_path: | |
3610 | path_put(&newmount); | |
3611 | err_unlock: | |
3612 | mutex_unlock(&fc->uapi_mutex); | |
3613 | err_fsfd: | |
3614 | fdput(f); | |
3615 | return ret; | |
3616 | } | |
3617 | ||
3618 | /* | |
3619 | * Move a mount from one place to another. In combination with | |
3620 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3621 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3622 | * a mount subtree. | |
2db154b3 DH |
3623 | * |
3624 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3625 | */ | |
3626 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
3627 | int, from_dfd, const char __user *, from_pathname, |
3628 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
3629 | unsigned int, flags) |
3630 | { | |
3631 | struct path from_path, to_path; | |
3632 | unsigned int lflags; | |
3633 | int ret = 0; | |
3634 | ||
3635 | if (!may_mount()) | |
3636 | return -EPERM; | |
3637 | ||
3638 | if (flags & ~MOVE_MOUNT__MASK) | |
3639 | return -EINVAL; | |
3640 | ||
3641 | /* If someone gives a pathname, they aren't permitted to move | |
3642 | * from an fd that requires unmount as we can't get at the flag | |
3643 | * to clear it afterwards. | |
3644 | */ | |
3645 | lflags = 0; | |
3646 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3647 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3648 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3649 | ||
3650 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3651 | if (ret < 0) | |
3652 | return ret; | |
3653 | ||
3654 | lflags = 0; | |
3655 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3656 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3657 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3658 | ||
3659 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3660 | if (ret < 0) | |
3661 | goto out_from; | |
3662 | ||
3663 | ret = security_move_mount(&from_path, &to_path); | |
3664 | if (ret < 0) | |
3665 | goto out_to; | |
3666 | ||
3667 | ret = do_move_mount(&from_path, &to_path); | |
3668 | ||
3669 | out_to: | |
3670 | path_put(&to_path); | |
3671 | out_from: | |
3672 | path_put(&from_path); | |
3673 | return ret; | |
3674 | } | |
3675 | ||
afac7cba AV |
3676 | /* |
3677 | * Return true if path is reachable from root | |
3678 | * | |
48a066e7 | 3679 | * namespace_sem or mount_lock is held |
afac7cba | 3680 | */ |
643822b4 | 3681 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3682 | const struct path *root) |
3683 | { | |
643822b4 | 3684 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3685 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3686 | mnt = mnt->mnt_parent; |
afac7cba | 3687 | } |
643822b4 | 3688 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3689 | } |
3690 | ||
640eb7e7 | 3691 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3692 | { |
25ab4c9b | 3693 | bool res; |
48a066e7 | 3694 | read_seqlock_excl(&mount_lock); |
643822b4 | 3695 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3696 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3697 | return res; |
3698 | } | |
3699 | EXPORT_SYMBOL(path_is_under); | |
3700 | ||
1da177e4 LT |
3701 | /* |
3702 | * pivot_root Semantics: | |
3703 | * Moves the root file system of the current process to the directory put_old, | |
3704 | * makes new_root as the new root file system of the current process, and sets | |
3705 | * root/cwd of all processes which had them on the current root to new_root. | |
3706 | * | |
3707 | * Restrictions: | |
3708 | * The new_root and put_old must be directories, and must not be on the | |
3709 | * same file system as the current process root. The put_old must be | |
3710 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3711 | * pointed to by put_old must yield the same directory as new_root. No other | |
3712 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3713 | * | |
4a0d11fa | 3714 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 3715 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
3716 | * in this situation. |
3717 | * | |
1da177e4 LT |
3718 | * Notes: |
3719 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3720 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3721 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3722 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3723 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3724 | * first. | |
3725 | */ | |
3480b257 HC |
3726 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3727 | const char __user *, put_old) | |
1da177e4 | 3728 | { |
2763d119 AV |
3729 | struct path new, old, root; |
3730 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3731 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3732 | int error; |
3733 | ||
9b40bc90 | 3734 | if (!may_mount()) |
1da177e4 LT |
3735 | return -EPERM; |
3736 | ||
ce6595a2 AV |
3737 | error = user_path_at(AT_FDCWD, new_root, |
3738 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
3739 | if (error) |
3740 | goto out0; | |
1da177e4 | 3741 | |
ce6595a2 AV |
3742 | error = user_path_at(AT_FDCWD, put_old, |
3743 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
3744 | if (error) |
3745 | goto out1; | |
3746 | ||
2d8f3038 | 3747 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3748 | if (error) |
3749 | goto out2; | |
1da177e4 | 3750 | |
f7ad3c6b | 3751 | get_fs_root(current->fs, &root); |
84d17192 AV |
3752 | old_mp = lock_mount(&old); |
3753 | error = PTR_ERR(old_mp); | |
3754 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3755 | goto out3; |
3756 | ||
1da177e4 | 3757 | error = -EINVAL; |
419148da AV |
3758 | new_mnt = real_mount(new.mnt); |
3759 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3760 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3761 | ex_parent = new_mnt->mnt_parent; |
3762 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3763 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3764 | IS_MNT_SHARED(ex_parent) || |
3765 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3766 | goto out4; |
143c8c91 | 3767 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3768 | goto out4; |
5ff9d8a6 EB |
3769 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3770 | goto out4; | |
1da177e4 | 3771 | error = -ENOENT; |
f3da392e | 3772 | if (d_unlinked(new.dentry)) |
b12cea91 | 3773 | goto out4; |
1da177e4 | 3774 | error = -EBUSY; |
84d17192 | 3775 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3776 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3777 | error = -EINVAL; |
8c3ee42e | 3778 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3779 | goto out4; /* not a mountpoint */ |
676da58d | 3780 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3781 | goto out4; /* not attached */ |
2d8f3038 | 3782 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3783 | goto out4; /* not a mountpoint */ |
676da58d | 3784 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3785 | goto out4; /* not attached */ |
4ac91378 | 3786 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3787 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3788 | goto out4; |
0d082601 EB |
3789 | /* make certain new is below the root */ |
3790 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3791 | goto out4; | |
719ea2fb | 3792 | lock_mount_hash(); |
2763d119 AV |
3793 | umount_mnt(new_mnt); |
3794 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3795 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3796 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3797 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3798 | } | |
4ac91378 | 3799 | /* mount old root on put_old */ |
84d17192 | 3800 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3801 | /* mount new_root on / */ |
2763d119 AV |
3802 | attach_mnt(new_mnt, root_parent, root_mp); |
3803 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3804 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3805 | /* A moved mount should not expire automatically */ |
3806 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3807 | put_mountpoint(root_mp); |
719ea2fb | 3808 | unlock_mount_hash(); |
2d8f3038 | 3809 | chroot_fs_refs(&root, &new); |
1da177e4 | 3810 | error = 0; |
b12cea91 | 3811 | out4: |
84d17192 | 3812 | unlock_mount(old_mp); |
2763d119 AV |
3813 | if (!error) |
3814 | mntput_no_expire(ex_parent); | |
b12cea91 | 3815 | out3: |
8c3ee42e | 3816 | path_put(&root); |
b12cea91 | 3817 | out2: |
2d8f3038 | 3818 | path_put(&old); |
1da177e4 | 3819 | out1: |
2d8f3038 | 3820 | path_put(&new); |
1da177e4 | 3821 | out0: |
1da177e4 | 3822 | return error; |
1da177e4 LT |
3823 | } |
3824 | ||
2a186721 CB |
3825 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
3826 | { | |
3827 | unsigned int flags = mnt->mnt.mnt_flags; | |
3828 | ||
3829 | /* flags to clear */ | |
3830 | flags &= ~kattr->attr_clr; | |
3831 | /* flags to raise */ | |
3832 | flags |= kattr->attr_set; | |
3833 | ||
3834 | return flags; | |
3835 | } | |
3836 | ||
9caccd41 CB |
3837 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3838 | { | |
3839 | struct vfsmount *m = &mnt->mnt; | |
3840 | ||
3841 | if (!kattr->mnt_userns) | |
3842 | return 0; | |
3843 | ||
3844 | /* | |
3845 | * Once a mount has been idmapped we don't allow it to change its | |
3846 | * mapping. It makes things simpler and callers can just create | |
3847 | * another bind-mount they can idmap if they want to. | |
3848 | */ | |
3849 | if (mnt_user_ns(m) != &init_user_ns) | |
3850 | return -EPERM; | |
3851 | ||
3852 | /* The underlying filesystem doesn't support idmapped mounts yet. */ | |
3853 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) | |
3854 | return -EINVAL; | |
3855 | ||
3856 | /* We're not controlling the superblock. */ | |
3857 | if (!ns_capable(m->mnt_sb->s_user_ns, CAP_SYS_ADMIN)) | |
3858 | return -EPERM; | |
3859 | ||
3860 | /* Mount has already been visible in the filesystem hierarchy. */ | |
3861 | if (!is_anon_ns(mnt->mnt_ns)) | |
3862 | return -EINVAL; | |
3863 | ||
3864 | return 0; | |
3865 | } | |
3866 | ||
2a186721 CB |
3867 | static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, |
3868 | struct mount *mnt, int *err) | |
3869 | { | |
3870 | struct mount *m = mnt, *last = NULL; | |
3871 | ||
3872 | if (!is_mounted(&m->mnt)) { | |
3873 | *err = -EINVAL; | |
3874 | goto out; | |
3875 | } | |
3876 | ||
3877 | if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { | |
3878 | *err = -EINVAL; | |
3879 | goto out; | |
3880 | } | |
3881 | ||
3882 | do { | |
3883 | unsigned int flags; | |
3884 | ||
3885 | flags = recalc_flags(kattr, m); | |
3886 | if (!can_change_locked_flags(m, flags)) { | |
3887 | *err = -EPERM; | |
3888 | goto out; | |
3889 | } | |
3890 | ||
9caccd41 CB |
3891 | *err = can_idmap_mount(kattr, m); |
3892 | if (*err) | |
3893 | goto out; | |
3894 | ||
2a186721 CB |
3895 | last = m; |
3896 | ||
3897 | if ((kattr->attr_set & MNT_READONLY) && | |
3898 | !(m->mnt.mnt_flags & MNT_READONLY)) { | |
3899 | *err = mnt_hold_writers(m); | |
3900 | if (*err) | |
3901 | goto out; | |
3902 | } | |
3903 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3904 | ||
3905 | out: | |
3906 | return last; | |
3907 | } | |
3908 | ||
9caccd41 CB |
3909 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3910 | { | |
3911 | struct user_namespace *mnt_userns; | |
3912 | ||
3913 | if (!kattr->mnt_userns) | |
3914 | return; | |
3915 | ||
3916 | mnt_userns = get_user_ns(kattr->mnt_userns); | |
3917 | /* Pairs with smp_load_acquire() in mnt_user_ns(). */ | |
3918 | smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); | |
3919 | } | |
3920 | ||
2a186721 CB |
3921 | static void mount_setattr_commit(struct mount_kattr *kattr, |
3922 | struct mount *mnt, struct mount *last, | |
3923 | int err) | |
3924 | { | |
3925 | struct mount *m = mnt; | |
3926 | ||
3927 | do { | |
3928 | if (!err) { | |
3929 | unsigned int flags; | |
3930 | ||
9caccd41 | 3931 | do_idmap_mount(kattr, m); |
2a186721 CB |
3932 | flags = recalc_flags(kattr, m); |
3933 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
3934 | } | |
3935 | ||
3936 | /* | |
3937 | * We either set MNT_READONLY above so make it visible | |
3938 | * before ~MNT_WRITE_HOLD or we failed to recursively | |
3939 | * apply mount options. | |
3940 | */ | |
3941 | if ((kattr->attr_set & MNT_READONLY) && | |
3942 | (m->mnt.mnt_flags & MNT_WRITE_HOLD)) | |
3943 | mnt_unhold_writers(m); | |
3944 | ||
3945 | if (!err && kattr->propagation) | |
3946 | change_mnt_propagation(m, kattr->propagation); | |
3947 | ||
3948 | /* | |
3949 | * On failure, only cleanup until we found the first mount | |
3950 | * we failed to handle. | |
3951 | */ | |
3952 | if (err && m == last) | |
3953 | break; | |
3954 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3955 | ||
3956 | if (!err) | |
3957 | touch_mnt_namespace(mnt->mnt_ns); | |
3958 | } | |
3959 | ||
3960 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
3961 | { | |
3962 | struct mount *mnt = real_mount(path->mnt), *last = NULL; | |
3963 | int err = 0; | |
3964 | ||
3965 | if (path->dentry != mnt->mnt.mnt_root) | |
3966 | return -EINVAL; | |
3967 | ||
3968 | if (kattr->propagation) { | |
3969 | /* | |
3970 | * Only take namespace_lock() if we're actually changing | |
3971 | * propagation. | |
3972 | */ | |
3973 | namespace_lock(); | |
3974 | if (kattr->propagation == MS_SHARED) { | |
3975 | err = invent_group_ids(mnt, kattr->recurse); | |
3976 | if (err) { | |
3977 | namespace_unlock(); | |
3978 | return err; | |
3979 | } | |
3980 | } | |
3981 | } | |
3982 | ||
3983 | lock_mount_hash(); | |
3984 | ||
3985 | /* | |
3986 | * Get the mount tree in a shape where we can change mount | |
3987 | * properties without failure. | |
3988 | */ | |
3989 | last = mount_setattr_prepare(kattr, mnt, &err); | |
3990 | if (last) /* Commit all changes or revert to the old state. */ | |
3991 | mount_setattr_commit(kattr, mnt, last, err); | |
3992 | ||
3993 | unlock_mount_hash(); | |
3994 | ||
3995 | if (kattr->propagation) { | |
3996 | namespace_unlock(); | |
3997 | if (err) | |
3998 | cleanup_group_ids(mnt, NULL); | |
3999 | } | |
4000 | ||
4001 | return err; | |
4002 | } | |
4003 | ||
9caccd41 CB |
4004 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
4005 | struct mount_kattr *kattr, unsigned int flags) | |
4006 | { | |
4007 | int err = 0; | |
4008 | struct ns_common *ns; | |
4009 | struct user_namespace *mnt_userns; | |
4010 | struct file *file; | |
4011 | ||
4012 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) | |
4013 | return 0; | |
4014 | ||
4015 | /* | |
4016 | * We currently do not support clearing an idmapped mount. If this ever | |
4017 | * is a use-case we can revisit this but for now let's keep it simple | |
4018 | * and not allow it. | |
4019 | */ | |
4020 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) | |
4021 | return -EINVAL; | |
4022 | ||
4023 | if (attr->userns_fd > INT_MAX) | |
4024 | return -EINVAL; | |
4025 | ||
4026 | file = fget(attr->userns_fd); | |
4027 | if (!file) | |
4028 | return -EBADF; | |
4029 | ||
4030 | if (!proc_ns_file(file)) { | |
4031 | err = -EINVAL; | |
4032 | goto out_fput; | |
4033 | } | |
4034 | ||
4035 | ns = get_proc_ns(file_inode(file)); | |
4036 | if (ns->ops->type != CLONE_NEWUSER) { | |
4037 | err = -EINVAL; | |
4038 | goto out_fput; | |
4039 | } | |
4040 | ||
4041 | /* | |
4042 | * The init_user_ns is used to indicate that a vfsmount is not idmapped. | |
4043 | * This is simpler than just having to treat NULL as unmapped. Users | |
4044 | * wanting to idmap a mount to init_user_ns can just use a namespace | |
4045 | * with an identity mapping. | |
4046 | */ | |
4047 | mnt_userns = container_of(ns, struct user_namespace, ns); | |
4048 | if (mnt_userns == &init_user_ns) { | |
4049 | err = -EPERM; | |
4050 | goto out_fput; | |
4051 | } | |
4052 | kattr->mnt_userns = get_user_ns(mnt_userns); | |
4053 | ||
4054 | out_fput: | |
4055 | fput(file); | |
4056 | return err; | |
4057 | } | |
4058 | ||
4059 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, | |
2a186721 CB |
4060 | struct mount_kattr *kattr, unsigned int flags) |
4061 | { | |
4062 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
4063 | ||
4064 | if (flags & AT_NO_AUTOMOUNT) | |
4065 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
4066 | if (flags & AT_SYMLINK_NOFOLLOW) | |
4067 | lookup_flags &= ~LOOKUP_FOLLOW; | |
4068 | if (flags & AT_EMPTY_PATH) | |
4069 | lookup_flags |= LOOKUP_EMPTY; | |
4070 | ||
4071 | *kattr = (struct mount_kattr) { | |
4072 | .lookup_flags = lookup_flags, | |
4073 | .recurse = !!(flags & AT_RECURSIVE), | |
4074 | }; | |
4075 | ||
4076 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
4077 | return -EINVAL; | |
4078 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
4079 | return -EINVAL; | |
4080 | kattr->propagation = attr->propagation; | |
4081 | ||
4082 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
4083 | return -EINVAL; | |
4084 | ||
2a186721 CB |
4085 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
4086 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
4087 | ||
4088 | /* | |
4089 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4090 | * users wanting to transition to a different atime setting cannot | |
4091 | * simply specify the atime setting in @attr_set, but must also | |
4092 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4093 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4094 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4095 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4096 | */ | |
4097 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4098 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4099 | return -EINVAL; | |
4100 | ||
4101 | /* | |
4102 | * Clear all previous time settings as they are mutually | |
4103 | * exclusive. | |
4104 | */ | |
4105 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4106 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4107 | case MOUNT_ATTR_RELATIME: | |
4108 | kattr->attr_set |= MNT_RELATIME; | |
4109 | break; | |
4110 | case MOUNT_ATTR_NOATIME: | |
4111 | kattr->attr_set |= MNT_NOATIME; | |
4112 | break; | |
4113 | case MOUNT_ATTR_STRICTATIME: | |
4114 | break; | |
4115 | default: | |
4116 | return -EINVAL; | |
4117 | } | |
4118 | } else { | |
4119 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4120 | return -EINVAL; | |
4121 | } | |
4122 | ||
9caccd41 CB |
4123 | return build_mount_idmapped(attr, usize, kattr, flags); |
4124 | } | |
4125 | ||
4126 | static void finish_mount_kattr(struct mount_kattr *kattr) | |
4127 | { | |
4128 | put_user_ns(kattr->mnt_userns); | |
4129 | kattr->mnt_userns = NULL; | |
2a186721 CB |
4130 | } |
4131 | ||
4132 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4133 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4134 | size_t, usize) | |
4135 | { | |
4136 | int err; | |
4137 | struct path target; | |
4138 | struct mount_attr attr; | |
4139 | struct mount_kattr kattr; | |
4140 | ||
4141 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4142 | ||
4143 | if (flags & ~(AT_EMPTY_PATH | | |
4144 | AT_RECURSIVE | | |
4145 | AT_SYMLINK_NOFOLLOW | | |
4146 | AT_NO_AUTOMOUNT)) | |
4147 | return -EINVAL; | |
4148 | ||
4149 | if (unlikely(usize > PAGE_SIZE)) | |
4150 | return -E2BIG; | |
4151 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4152 | return -EINVAL; | |
4153 | ||
4154 | if (!may_mount()) | |
4155 | return -EPERM; | |
4156 | ||
4157 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4158 | if (err) | |
4159 | return err; | |
4160 | ||
4161 | /* Don't bother walking through the mounts if this is a nop. */ | |
4162 | if (attr.attr_set == 0 && | |
4163 | attr.attr_clr == 0 && | |
4164 | attr.propagation == 0) | |
4165 | return 0; | |
4166 | ||
9caccd41 | 4167 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
2a186721 CB |
4168 | if (err) |
4169 | return err; | |
4170 | ||
4171 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
4172 | if (err) | |
4173 | return err; | |
4174 | ||
4175 | err = do_mount_setattr(&target, &kattr); | |
9caccd41 | 4176 | finish_mount_kattr(&kattr); |
2a186721 CB |
4177 | path_put(&target); |
4178 | return err; | |
4179 | } | |
4180 | ||
1da177e4 LT |
4181 | static void __init init_mount_tree(void) |
4182 | { | |
4183 | struct vfsmount *mnt; | |
74e83122 | 4184 | struct mount *m; |
6b3286ed | 4185 | struct mnt_namespace *ns; |
ac748a09 | 4186 | struct path root; |
1da177e4 | 4187 | |
fd3e007f | 4188 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4189 | if (IS_ERR(mnt)) |
4190 | panic("Can't create rootfs"); | |
b3e19d92 | 4191 | |
74e83122 | 4192 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4193 | if (IS_ERR(ns)) |
1da177e4 | 4194 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4195 | m = real_mount(mnt); |
4196 | m->mnt_ns = ns; | |
4197 | ns->root = m; | |
4198 | ns->mounts = 1; | |
4199 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4200 | init_task.nsproxy->mnt_ns = ns; |
4201 | get_mnt_ns(ns); | |
4202 | ||
be08d6d2 AV |
4203 | root.mnt = mnt; |
4204 | root.dentry = mnt->mnt_root; | |
da362b09 | 4205 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4206 | |
4207 | set_fs_pwd(current->fs, &root); | |
4208 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4209 | } |
4210 | ||
74bf17cf | 4211 | void __init mnt_init(void) |
1da177e4 | 4212 | { |
15a67dd8 | 4213 | int err; |
1da177e4 | 4214 | |
7d6fec45 | 4215 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 4216 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 4217 | |
0818bf27 | 4218 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4219 | sizeof(struct hlist_head), |
0818bf27 | 4220 | mhash_entries, 19, |
3d375d78 | 4221 | HASH_ZERO, |
0818bf27 AV |
4222 | &m_hash_shift, &m_hash_mask, 0, 0); |
4223 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4224 | sizeof(struct hlist_head), | |
4225 | mphash_entries, 19, | |
3d375d78 | 4226 | HASH_ZERO, |
0818bf27 | 4227 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4228 | |
84d17192 | 4229 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4230 | panic("Failed to allocate mount hash table\n"); |
4231 | ||
4b93dc9b TH |
4232 | kernfs_init(); |
4233 | ||
15a67dd8 RD |
4234 | err = sysfs_init(); |
4235 | if (err) | |
4236 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4237 | __func__, err); |
00d26666 GKH |
4238 | fs_kobj = kobject_create_and_add("fs", NULL); |
4239 | if (!fs_kobj) | |
8e24eea7 | 4240 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4241 | shmem_init(); |
1da177e4 LT |
4242 | init_rootfs(); |
4243 | init_mount_tree(); | |
4244 | } | |
4245 | ||
616511d0 | 4246 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4247 | { |
1a7b8969 | 4248 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4249 | return; |
7b00ed6f | 4250 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4251 | free_mnt_ns(ns); |
1da177e4 | 4252 | } |
9d412a43 | 4253 | |
d911b458 | 4254 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4255 | { |
423e0ab0 | 4256 | struct vfsmount *mnt; |
d911b458 | 4257 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4258 | if (!IS_ERR(mnt)) { |
4259 | /* | |
4260 | * it is a longterm mount, don't release mnt until | |
4261 | * we unmount before file sys is unregistered | |
4262 | */ | |
f7a99c5b | 4263 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4264 | } |
4265 | return mnt; | |
9d412a43 | 4266 | } |
d911b458 | 4267 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4268 | |
4269 | void kern_unmount(struct vfsmount *mnt) | |
4270 | { | |
4271 | /* release long term mount so mount point can be released */ | |
4272 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 4273 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 4274 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4275 | mntput(mnt); |
4276 | } | |
4277 | } | |
4278 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4279 | |
df820f8d MS |
4280 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4281 | { | |
4282 | unsigned int i; | |
4283 | ||
4284 | for (i = 0; i < num; i++) | |
4285 | if (mnt[i]) | |
4286 | real_mount(mnt[i])->mnt_ns = NULL; | |
4287 | synchronize_rcu_expedited(); | |
4288 | for (i = 0; i < num; i++) | |
4289 | mntput(mnt[i]); | |
4290 | } | |
4291 | EXPORT_SYMBOL(kern_unmount_array); | |
4292 | ||
02125a82 AV |
4293 | bool our_mnt(struct vfsmount *mnt) |
4294 | { | |
143c8c91 | 4295 | return check_mnt(real_mount(mnt)); |
02125a82 | 4296 | } |
8823c079 | 4297 | |
3151527e EB |
4298 | bool current_chrooted(void) |
4299 | { | |
4300 | /* Does the current process have a non-standard root */ | |
4301 | struct path ns_root; | |
4302 | struct path fs_root; | |
4303 | bool chrooted; | |
4304 | ||
4305 | /* Find the namespace root */ | |
4306 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4307 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4308 | path_get(&ns_root); | |
4309 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4310 | ; | |
4311 | ||
4312 | get_fs_root(current->fs, &fs_root); | |
4313 | ||
4314 | chrooted = !path_equal(&fs_root, &ns_root); | |
4315 | ||
4316 | path_put(&fs_root); | |
4317 | path_put(&ns_root); | |
4318 | ||
4319 | return chrooted; | |
4320 | } | |
4321 | ||
132e4608 DH |
4322 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4323 | const struct super_block *sb, | |
8654df4e | 4324 | int *new_mnt_flags) |
87a8ebd6 | 4325 | { |
8c6cf9cc | 4326 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4327 | struct mount *mnt; |
e51db735 | 4328 | bool visible = false; |
87a8ebd6 | 4329 | |
44bb4385 | 4330 | down_read(&namespace_sem); |
9f6c61f9 | 4331 | lock_ns_list(ns); |
87a8ebd6 | 4332 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4333 | struct mount *child; |
77b1a97d EB |
4334 | int mnt_flags; |
4335 | ||
9f6c61f9 MS |
4336 | if (mnt_is_cursor(mnt)) |
4337 | continue; | |
4338 | ||
132e4608 | 4339 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4340 | continue; |
4341 | ||
7e96c1b0 EB |
4342 | /* This mount is not fully visible if it's root directory |
4343 | * is not the root directory of the filesystem. | |
4344 | */ | |
4345 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4346 | continue; | |
4347 | ||
a1935c17 | 4348 | /* A local view of the mount flags */ |
77b1a97d | 4349 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4350 | |
695e9df0 | 4351 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4352 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4353 | mnt_flags |= MNT_LOCK_READONLY; |
4354 | ||
8c6cf9cc EB |
4355 | /* Verify the mount flags are equal to or more permissive |
4356 | * than the proposed new mount. | |
4357 | */ | |
77b1a97d | 4358 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4359 | !(new_flags & MNT_READONLY)) |
4360 | continue; | |
77b1a97d EB |
4361 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4362 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4363 | continue; |
4364 | ||
ceeb0e5d EB |
4365 | /* This mount is not fully visible if there are any |
4366 | * locked child mounts that cover anything except for | |
4367 | * empty directories. | |
e51db735 EB |
4368 | */ |
4369 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4370 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4371 | /* Only worry about locked mounts */ |
d71ed6c9 | 4372 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4373 | continue; |
7236c85e EB |
4374 | /* Is the directory permanetly empty? */ |
4375 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4376 | goto next; |
87a8ebd6 | 4377 | } |
8c6cf9cc | 4378 | /* Preserve the locked attributes */ |
77b1a97d | 4379 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4380 | MNT_LOCK_ATIME); |
e51db735 EB |
4381 | visible = true; |
4382 | goto found; | |
4383 | next: ; | |
87a8ebd6 | 4384 | } |
e51db735 | 4385 | found: |
9f6c61f9 | 4386 | unlock_ns_list(ns); |
44bb4385 | 4387 | up_read(&namespace_sem); |
e51db735 | 4388 | return visible; |
87a8ebd6 EB |
4389 | } |
4390 | ||
132e4608 | 4391 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4392 | { |
a1935c17 | 4393 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4394 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4395 | unsigned long s_iflags; | |
4396 | ||
4397 | if (ns->user_ns == &init_user_ns) | |
4398 | return false; | |
4399 | ||
4400 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4401 | s_iflags = sb->s_iflags; |
8654df4e EB |
4402 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4403 | return false; | |
4404 | ||
a1935c17 EB |
4405 | if ((s_iflags & required_iflags) != required_iflags) { |
4406 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4407 | required_iflags); | |
4408 | return true; | |
4409 | } | |
4410 | ||
132e4608 | 4411 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4412 | } |
4413 | ||
380cf5ba AL |
4414 | bool mnt_may_suid(struct vfsmount *mnt) |
4415 | { | |
4416 | /* | |
4417 | * Foreign mounts (accessed via fchdir or through /proc | |
4418 | * symlinks) are always treated as if they are nosuid. This | |
4419 | * prevents namespaces from trusting potentially unsafe | |
4420 | * suid/sgid bits, file caps, or security labels that originate | |
4421 | * in other namespaces. | |
4422 | */ | |
4423 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4424 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4425 | } | |
4426 | ||
64964528 | 4427 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4428 | { |
58be2825 | 4429 | struct ns_common *ns = NULL; |
8823c079 EB |
4430 | struct nsproxy *nsproxy; |
4431 | ||
728dba3a EB |
4432 | task_lock(task); |
4433 | nsproxy = task->nsproxy; | |
8823c079 | 4434 | if (nsproxy) { |
58be2825 AV |
4435 | ns = &nsproxy->mnt_ns->ns; |
4436 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4437 | } |
728dba3a | 4438 | task_unlock(task); |
8823c079 EB |
4439 | |
4440 | return ns; | |
4441 | } | |
4442 | ||
64964528 | 4443 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4444 | { |
58be2825 | 4445 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4446 | } |
4447 | ||
f2a8d52e | 4448 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4449 | { |
f2a8d52e CB |
4450 | struct nsproxy *nsproxy = nsset->nsproxy; |
4451 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4452 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4453 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4454 | struct path root; |
4f757f3c | 4455 | int err; |
8823c079 | 4456 | |
0c55cfc4 | 4457 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4458 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4459 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4460 | return -EPERM; |
8823c079 | 4461 | |
74e83122 AV |
4462 | if (is_anon_ns(mnt_ns)) |
4463 | return -EINVAL; | |
4464 | ||
8823c079 EB |
4465 | if (fs->users != 1) |
4466 | return -EINVAL; | |
4467 | ||
4468 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4469 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4470 | nsproxy->mnt_ns = mnt_ns; |
4471 | ||
4472 | /* Find the root */ | |
4f757f3c AV |
4473 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4474 | "/", LOOKUP_DOWN, &root); | |
4475 | if (err) { | |
4476 | /* revert to old namespace */ | |
4477 | nsproxy->mnt_ns = old_mnt_ns; | |
4478 | put_mnt_ns(mnt_ns); | |
4479 | return err; | |
4480 | } | |
8823c079 | 4481 | |
4068367c AV |
4482 | put_mnt_ns(old_mnt_ns); |
4483 | ||
8823c079 EB |
4484 | /* Update the pwd and root */ |
4485 | set_fs_pwd(fs, &root); | |
4486 | set_fs_root(fs, &root); | |
4487 | ||
4488 | path_put(&root); | |
4489 | return 0; | |
4490 | } | |
4491 | ||
bcac25a5 AV |
4492 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4493 | { | |
4494 | return to_mnt_ns(ns)->user_ns; | |
4495 | } | |
4496 | ||
8823c079 EB |
4497 | const struct proc_ns_operations mntns_operations = { |
4498 | .name = "mnt", | |
4499 | .type = CLONE_NEWNS, | |
4500 | .get = mntns_get, | |
4501 | .put = mntns_put, | |
4502 | .install = mntns_install, | |
bcac25a5 | 4503 | .owner = mntns_owner, |
8823c079 | 4504 | }; |