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> |
bd303368 | 34 | #include <linux/mnt_idmapping.h> |
9164bb4a | 35 | |
07b20889 | 36 | #include "pnode.h" |
948730b0 | 37 | #include "internal.h" |
1da177e4 | 38 | |
d2921684 | 39 | /* Maximum number of mounts in a mount namespace */ |
ab171b95 | 40 | static unsigned int sysctl_mount_max __read_mostly = 100000; |
d2921684 | 41 | |
0818bf27 AV |
42 | static unsigned int m_hash_mask __read_mostly; |
43 | static unsigned int m_hash_shift __read_mostly; | |
44 | static unsigned int mp_hash_mask __read_mostly; | |
45 | static unsigned int mp_hash_shift __read_mostly; | |
46 | ||
47 | static __initdata unsigned long mhash_entries; | |
48 | static int __init set_mhash_entries(char *str) | |
49 | { | |
50 | if (!str) | |
51 | return 0; | |
52 | mhash_entries = simple_strtoul(str, &str, 0); | |
53 | return 1; | |
54 | } | |
55 | __setup("mhash_entries=", set_mhash_entries); | |
56 | ||
57 | static __initdata unsigned long mphash_entries; | |
58 | static int __init set_mphash_entries(char *str) | |
59 | { | |
60 | if (!str) | |
61 | return 0; | |
62 | mphash_entries = simple_strtoul(str, &str, 0); | |
63 | return 1; | |
64 | } | |
65 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 66 | |
c7999c36 | 67 | static u64 event; |
73cd49ec | 68 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 69 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 70 | |
38129a13 | 71 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 72 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 73 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 74 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
75 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
76 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 77 | |
2a186721 CB |
78 | struct mount_kattr { |
79 | unsigned int attr_set; | |
80 | unsigned int attr_clr; | |
81 | unsigned int propagation; | |
82 | unsigned int lookup_flags; | |
83 | bool recurse; | |
9caccd41 | 84 | struct user_namespace *mnt_userns; |
2a186721 CB |
85 | }; |
86 | ||
f87fd4c2 | 87 | /* /sys/fs */ |
00d26666 GKH |
88 | struct kobject *fs_kobj; |
89 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 90 | |
99b7db7b NP |
91 | /* |
92 | * vfsmount lock may be taken for read to prevent changes to the | |
93 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
94 | * up the tree. | |
95 | * | |
96 | * It should be taken for write in all cases where the vfsmount | |
97 | * tree or hash is modified or when a vfsmount structure is modified. | |
98 | */ | |
48a066e7 | 99 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 100 | |
d033cb67 CB |
101 | static inline void lock_mount_hash(void) |
102 | { | |
103 | write_seqlock(&mount_lock); | |
104 | } | |
105 | ||
106 | static inline void unlock_mount_hash(void) | |
107 | { | |
108 | write_sequnlock(&mount_lock); | |
109 | } | |
110 | ||
38129a13 | 111 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 112 | { |
b58fed8b RP |
113 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
114 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
115 | tmp = tmp + (tmp >> m_hash_shift); |
116 | return &mount_hashtable[tmp & m_hash_mask]; | |
117 | } | |
118 | ||
119 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
120 | { | |
121 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
122 | tmp = tmp + (tmp >> mp_hash_shift); | |
123 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
124 | } |
125 | ||
b105e270 | 126 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 127 | { |
169b480e MW |
128 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
129 | ||
130 | if (res < 0) | |
131 | return res; | |
132 | mnt->mnt_id = res; | |
133 | return 0; | |
73cd49ec MS |
134 | } |
135 | ||
b105e270 | 136 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 137 | { |
169b480e | 138 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
139 | } |
140 | ||
719f5d7f MS |
141 | /* |
142 | * Allocate a new peer group ID | |
719f5d7f | 143 | */ |
4b8b21f4 | 144 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 145 | { |
169b480e | 146 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 147 | |
169b480e MW |
148 | if (res < 0) |
149 | return res; | |
150 | mnt->mnt_group_id = res; | |
151 | return 0; | |
719f5d7f MS |
152 | } |
153 | ||
154 | /* | |
155 | * Release a peer group ID | |
156 | */ | |
4b8b21f4 | 157 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 158 | { |
169b480e | 159 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 160 | mnt->mnt_group_id = 0; |
719f5d7f MS |
161 | } |
162 | ||
b3e19d92 NP |
163 | /* |
164 | * vfsmount lock must be held for read | |
165 | */ | |
83adc753 | 166 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
167 | { |
168 | #ifdef CONFIG_SMP | |
68e8a9fe | 169 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
170 | #else |
171 | preempt_disable(); | |
68e8a9fe | 172 | mnt->mnt_count += n; |
b3e19d92 NP |
173 | preempt_enable(); |
174 | #endif | |
175 | } | |
176 | ||
b3e19d92 NP |
177 | /* |
178 | * vfsmount lock must be held for write | |
179 | */ | |
edf7ddbf | 180 | int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
181 | { |
182 | #ifdef CONFIG_SMP | |
edf7ddbf | 183 | int count = 0; |
b3e19d92 NP |
184 | int cpu; |
185 | ||
186 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 187 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
188 | } |
189 | ||
190 | return count; | |
191 | #else | |
68e8a9fe | 192 | return mnt->mnt_count; |
b3e19d92 NP |
193 | #endif |
194 | } | |
195 | ||
b105e270 | 196 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 197 | { |
c63181e6 AV |
198 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
199 | if (mnt) { | |
73cd49ec MS |
200 | int err; |
201 | ||
c63181e6 | 202 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
203 | if (err) |
204 | goto out_free_cache; | |
205 | ||
206 | if (name) { | |
79f6540b VA |
207 | mnt->mnt_devname = kstrdup_const(name, |
208 | GFP_KERNEL_ACCOUNT); | |
c63181e6 | 209 | if (!mnt->mnt_devname) |
88b38782 | 210 | goto out_free_id; |
73cd49ec MS |
211 | } |
212 | ||
b3e19d92 | 213 | #ifdef CONFIG_SMP |
c63181e6 AV |
214 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
215 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
216 | goto out_free_devname; |
217 | ||
c63181e6 | 218 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 219 | #else |
c63181e6 AV |
220 | mnt->mnt_count = 1; |
221 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
222 | #endif |
223 | ||
38129a13 | 224 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
225 | INIT_LIST_HEAD(&mnt->mnt_child); |
226 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
227 | INIT_LIST_HEAD(&mnt->mnt_list); | |
228 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
229 | INIT_LIST_HEAD(&mnt->mnt_share); | |
230 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
231 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 232 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 233 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
56cbb429 | 234 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
a6435940 | 235 | mnt->mnt.mnt_userns = &init_user_ns; |
1da177e4 | 236 | } |
c63181e6 | 237 | return mnt; |
88b38782 | 238 | |
d3ef3d73 | 239 | #ifdef CONFIG_SMP |
240 | out_free_devname: | |
fcc139ae | 241 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 242 | #endif |
88b38782 | 243 | out_free_id: |
c63181e6 | 244 | mnt_free_id(mnt); |
88b38782 | 245 | out_free_cache: |
c63181e6 | 246 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 247 | return NULL; |
1da177e4 LT |
248 | } |
249 | ||
3d733633 DH |
250 | /* |
251 | * Most r/o checks on a fs are for operations that take | |
252 | * discrete amounts of time, like a write() or unlink(). | |
253 | * We must keep track of when those operations start | |
254 | * (for permission checks) and when they end, so that | |
255 | * we can determine when writes are able to occur to | |
256 | * a filesystem. | |
257 | */ | |
258 | /* | |
259 | * __mnt_is_readonly: check whether a mount is read-only | |
260 | * @mnt: the mount to check for its write status | |
261 | * | |
262 | * This shouldn't be used directly ouside of the VFS. | |
263 | * It does not guarantee that the filesystem will stay | |
264 | * r/w, just that it is right *now*. This can not and | |
265 | * should not be used in place of IS_RDONLY(inode). | |
266 | * mnt_want/drop_write() will _keep_ the filesystem | |
267 | * r/w. | |
268 | */ | |
43f5e655 | 269 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 270 | { |
43f5e655 | 271 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
272 | } |
273 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
274 | ||
83adc753 | 275 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 276 | { |
277 | #ifdef CONFIG_SMP | |
68e8a9fe | 278 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 279 | #else |
68e8a9fe | 280 | mnt->mnt_writers++; |
d3ef3d73 | 281 | #endif |
282 | } | |
3d733633 | 283 | |
83adc753 | 284 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 285 | { |
d3ef3d73 | 286 | #ifdef CONFIG_SMP |
68e8a9fe | 287 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 288 | #else |
68e8a9fe | 289 | mnt->mnt_writers--; |
d3ef3d73 | 290 | #endif |
3d733633 | 291 | } |
3d733633 | 292 | |
83adc753 | 293 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 294 | { |
d3ef3d73 | 295 | #ifdef CONFIG_SMP |
296 | unsigned int count = 0; | |
3d733633 | 297 | int cpu; |
3d733633 DH |
298 | |
299 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 300 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 301 | } |
3d733633 | 302 | |
d3ef3d73 | 303 | return count; |
304 | #else | |
305 | return mnt->mnt_writers; | |
306 | #endif | |
3d733633 DH |
307 | } |
308 | ||
4ed5e82f MS |
309 | static int mnt_is_readonly(struct vfsmount *mnt) |
310 | { | |
311 | if (mnt->mnt_sb->s_readonly_remount) | |
312 | return 1; | |
313 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
314 | smp_rmb(); | |
315 | return __mnt_is_readonly(mnt); | |
316 | } | |
317 | ||
8366025e | 318 | /* |
eb04c282 JK |
319 | * Most r/o & frozen checks on a fs are for operations that take discrete |
320 | * amounts of time, like a write() or unlink(). We must keep track of when | |
321 | * those operations start (for permission checks) and when they end, so that we | |
322 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
323 | */ |
324 | /** | |
eb04c282 | 325 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 326 | * @m: the mount on which to take a write |
8366025e | 327 | * |
eb04c282 JK |
328 | * This tells the low-level filesystem that a write is about to be performed to |
329 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
330 | * returning success. This operation does not protect against filesystem being | |
331 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
332 | * called. This is effectively a refcount. | |
8366025e | 333 | */ |
eb04c282 | 334 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 335 | { |
83adc753 | 336 | struct mount *mnt = real_mount(m); |
3d733633 | 337 | int ret = 0; |
3d733633 | 338 | |
d3ef3d73 | 339 | preempt_disable(); |
c6653a83 | 340 | mnt_inc_writers(mnt); |
d3ef3d73 | 341 | /* |
c6653a83 | 342 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 343 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
344 | * incremented count after it has set MNT_WRITE_HOLD. | |
345 | */ | |
346 | smp_mb(); | |
6aa7de05 | 347 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 348 | cpu_relax(); |
349 | /* | |
350 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
351 | * be set to match its requirements. So we must not load that until | |
352 | * MNT_WRITE_HOLD is cleared. | |
353 | */ | |
354 | smp_rmb(); | |
4ed5e82f | 355 | if (mnt_is_readonly(m)) { |
c6653a83 | 356 | mnt_dec_writers(mnt); |
3d733633 | 357 | ret = -EROFS; |
3d733633 | 358 | } |
d3ef3d73 | 359 | preempt_enable(); |
eb04c282 JK |
360 | |
361 | return ret; | |
362 | } | |
363 | ||
364 | /** | |
365 | * mnt_want_write - get write access to a mount | |
366 | * @m: the mount on which to take a write | |
367 | * | |
368 | * This tells the low-level filesystem that a write is about to be performed to | |
369 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
370 | * is not frozen) before returning success. When the write operation is | |
371 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
372 | */ | |
373 | int mnt_want_write(struct vfsmount *m) | |
374 | { | |
375 | int ret; | |
376 | ||
377 | sb_start_write(m->mnt_sb); | |
378 | ret = __mnt_want_write(m); | |
379 | if (ret) | |
380 | sb_end_write(m->mnt_sb); | |
3d733633 | 381 | return ret; |
8366025e DH |
382 | } |
383 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
384 | ||
96029c4e | 385 | /** |
eb04c282 | 386 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 387 | * @file: the file who's mount on which to take a write |
388 | * | |
14e43bf4 EB |
389 | * This is like __mnt_want_write, but if the file is already open for writing it |
390 | * skips incrementing mnt_writers (since the open file already has a reference) | |
391 | * and instead only does the check for emergency r/o remounts. This must be | |
392 | * paired with __mnt_drop_write_file. | |
96029c4e | 393 | */ |
eb04c282 | 394 | int __mnt_want_write_file(struct file *file) |
96029c4e | 395 | { |
14e43bf4 EB |
396 | if (file->f_mode & FMODE_WRITER) { |
397 | /* | |
398 | * Superblock may have become readonly while there are still | |
399 | * writable fd's, e.g. due to a fs error with errors=remount-ro | |
400 | */ | |
401 | if (__mnt_is_readonly(file->f_path.mnt)) | |
402 | return -EROFS; | |
403 | return 0; | |
404 | } | |
405 | return __mnt_want_write(file->f_path.mnt); | |
96029c4e | 406 | } |
eb04c282 | 407 | |
7c6893e3 MS |
408 | /** |
409 | * mnt_want_write_file - get write access to a file's mount | |
410 | * @file: the file who's mount on which to take a write | |
411 | * | |
14e43bf4 EB |
412 | * This is like mnt_want_write, but if the file is already open for writing it |
413 | * skips incrementing mnt_writers (since the open file already has a reference) | |
414 | * and instead only does the freeze protection and the check for emergency r/o | |
415 | * remounts. This must be paired with mnt_drop_write_file. | |
7c6893e3 MS |
416 | */ |
417 | int mnt_want_write_file(struct file *file) | |
418 | { | |
419 | int ret; | |
420 | ||
a6795a58 | 421 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
422 | ret = __mnt_want_write_file(file); |
423 | if (ret) | |
a6795a58 | 424 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
425 | return ret; |
426 | } | |
96029c4e | 427 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
428 | ||
8366025e | 429 | /** |
eb04c282 | 430 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
431 | * @mnt: the mount on which to give up write access |
432 | * | |
433 | * Tells the low-level filesystem that we are done | |
434 | * performing writes to it. Must be matched with | |
eb04c282 | 435 | * __mnt_want_write() call above. |
8366025e | 436 | */ |
eb04c282 | 437 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 438 | { |
d3ef3d73 | 439 | preempt_disable(); |
83adc753 | 440 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 441 | preempt_enable(); |
8366025e | 442 | } |
eb04c282 JK |
443 | |
444 | /** | |
445 | * mnt_drop_write - give up write access to a mount | |
446 | * @mnt: the mount on which to give up write access | |
447 | * | |
448 | * Tells the low-level filesystem that we are done performing writes to it and | |
449 | * also allows filesystem to be frozen again. Must be matched with | |
450 | * mnt_want_write() call above. | |
451 | */ | |
452 | void mnt_drop_write(struct vfsmount *mnt) | |
453 | { | |
454 | __mnt_drop_write(mnt); | |
455 | sb_end_write(mnt->mnt_sb); | |
456 | } | |
8366025e DH |
457 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
458 | ||
eb04c282 JK |
459 | void __mnt_drop_write_file(struct file *file) |
460 | { | |
14e43bf4 EB |
461 | if (!(file->f_mode & FMODE_WRITER)) |
462 | __mnt_drop_write(file->f_path.mnt); | |
eb04c282 JK |
463 | } |
464 | ||
7c6893e3 MS |
465 | void mnt_drop_write_file(struct file *file) |
466 | { | |
a6795a58 | 467 | __mnt_drop_write_file(file); |
7c6893e3 MS |
468 | sb_end_write(file_inode(file)->i_sb); |
469 | } | |
2a79f17e AV |
470 | EXPORT_SYMBOL(mnt_drop_write_file); |
471 | ||
fbdc2f6c | 472 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 473 | { |
83adc753 | 474 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 475 | /* |
d3ef3d73 | 476 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
477 | * should be visible before we do. | |
3d733633 | 478 | */ |
d3ef3d73 | 479 | smp_mb(); |
480 | ||
3d733633 | 481 | /* |
d3ef3d73 | 482 | * With writers on hold, if this value is zero, then there are |
483 | * definitely no active writers (although held writers may subsequently | |
484 | * increment the count, they'll have to wait, and decrement it after | |
485 | * seeing MNT_READONLY). | |
486 | * | |
487 | * It is OK to have counter incremented on one CPU and decremented on | |
488 | * another: the sum will add up correctly. The danger would be when we | |
489 | * sum up each counter, if we read a counter before it is incremented, | |
490 | * but then read another CPU's count which it has been subsequently | |
491 | * decremented from -- we would see more decrements than we should. | |
492 | * MNT_WRITE_HOLD protects against this scenario, because | |
493 | * mnt_want_write first increments count, then smp_mb, then spins on | |
494 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
495 | * we're counting up here. | |
3d733633 | 496 | */ |
c6653a83 | 497 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
498 | return -EBUSY; |
499 | ||
500 | return 0; | |
501 | } | |
502 | ||
503 | static inline void mnt_unhold_writers(struct mount *mnt) | |
504 | { | |
d3ef3d73 | 505 | /* |
506 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
507 | * that become unheld will see MNT_READONLY. | |
508 | */ | |
509 | smp_wmb(); | |
83adc753 | 510 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
511 | } |
512 | ||
513 | static int mnt_make_readonly(struct mount *mnt) | |
514 | { | |
515 | int ret; | |
516 | ||
517 | ret = mnt_hold_writers(mnt); | |
518 | if (!ret) | |
519 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
520 | mnt_unhold_writers(mnt); | |
3d733633 | 521 | return ret; |
8366025e | 522 | } |
8366025e | 523 | |
4ed5e82f MS |
524 | int sb_prepare_remount_readonly(struct super_block *sb) |
525 | { | |
526 | struct mount *mnt; | |
527 | int err = 0; | |
528 | ||
8e8b8796 MS |
529 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
530 | if (atomic_long_read(&sb->s_remove_count)) | |
531 | return -EBUSY; | |
532 | ||
719ea2fb | 533 | lock_mount_hash(); |
4ed5e82f MS |
534 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
535 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
536 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
537 | smp_mb(); | |
538 | if (mnt_get_writers(mnt) > 0) { | |
539 | err = -EBUSY; | |
540 | break; | |
541 | } | |
542 | } | |
543 | } | |
8e8b8796 MS |
544 | if (!err && atomic_long_read(&sb->s_remove_count)) |
545 | err = -EBUSY; | |
546 | ||
4ed5e82f MS |
547 | if (!err) { |
548 | sb->s_readonly_remount = 1; | |
549 | smp_wmb(); | |
550 | } | |
551 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
552 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
553 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
554 | } | |
719ea2fb | 555 | unlock_mount_hash(); |
4ed5e82f MS |
556 | |
557 | return err; | |
558 | } | |
559 | ||
b105e270 | 560 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 561 | { |
a6435940 CB |
562 | struct user_namespace *mnt_userns; |
563 | ||
564 | mnt_userns = mnt_user_ns(&mnt->mnt); | |
bd303368 | 565 | if (!initial_idmapping(mnt_userns)) |
a6435940 | 566 | put_user_ns(mnt_userns); |
fcc139ae | 567 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 568 | #ifdef CONFIG_SMP |
68e8a9fe | 569 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 570 | #endif |
b105e270 | 571 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
572 | } |
573 | ||
8ffcb32e DH |
574 | static void delayed_free_vfsmnt(struct rcu_head *head) |
575 | { | |
576 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
577 | } | |
578 | ||
48a066e7 | 579 | /* call under rcu_read_lock */ |
294d71ff | 580 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
581 | { |
582 | struct mount *mnt; | |
583 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 584 | return 1; |
48a066e7 | 585 | if (bastard == NULL) |
294d71ff | 586 | return 0; |
48a066e7 AV |
587 | mnt = real_mount(bastard); |
588 | mnt_add_count(mnt, 1); | |
119e1ef8 | 589 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 590 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 591 | return 0; |
48a066e7 AV |
592 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
593 | mnt_add_count(mnt, -1); | |
294d71ff AV |
594 | return 1; |
595 | } | |
119e1ef8 AV |
596 | lock_mount_hash(); |
597 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
598 | mnt_add_count(mnt, -1); | |
599 | unlock_mount_hash(); | |
600 | return 1; | |
601 | } | |
602 | unlock_mount_hash(); | |
603 | /* caller will mntput() */ | |
294d71ff AV |
604 | return -1; |
605 | } | |
606 | ||
607 | /* call under rcu_read_lock */ | |
608 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
609 | { | |
610 | int res = __legitimize_mnt(bastard, seq); | |
611 | if (likely(!res)) | |
612 | return true; | |
613 | if (unlikely(res < 0)) { | |
614 | rcu_read_unlock(); | |
615 | mntput(bastard); | |
616 | rcu_read_lock(); | |
48a066e7 | 617 | } |
48a066e7 AV |
618 | return false; |
619 | } | |
620 | ||
1da177e4 | 621 | /* |
474279dc | 622 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 623 | * call under rcu_read_lock() |
1da177e4 | 624 | */ |
474279dc | 625 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 626 | { |
38129a13 | 627 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
628 | struct mount *p; |
629 | ||
38129a13 | 630 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
631 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
632 | return p; | |
633 | return NULL; | |
634 | } | |
635 | ||
a05964f3 | 636 | /* |
f015f126 DH |
637 | * lookup_mnt - Return the first child mount mounted at path |
638 | * | |
639 | * "First" means first mounted chronologically. If you create the | |
640 | * following mounts: | |
641 | * | |
642 | * mount /dev/sda1 /mnt | |
643 | * mount /dev/sda2 /mnt | |
644 | * mount /dev/sda3 /mnt | |
645 | * | |
646 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
647 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
648 | * /dev/sda2, then /dev/sda3, then NULL. | |
649 | * | |
650 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 651 | */ |
ca71cf71 | 652 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 653 | { |
c7105365 | 654 | struct mount *child_mnt; |
48a066e7 AV |
655 | struct vfsmount *m; |
656 | unsigned seq; | |
99b7db7b | 657 | |
48a066e7 AV |
658 | rcu_read_lock(); |
659 | do { | |
660 | seq = read_seqbegin(&mount_lock); | |
661 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
662 | m = child_mnt ? &child_mnt->mnt : NULL; | |
663 | } while (!legitimize_mnt(m, seq)); | |
664 | rcu_read_unlock(); | |
665 | return m; | |
a05964f3 RP |
666 | } |
667 | ||
9f6c61f9 MS |
668 | static inline void lock_ns_list(struct mnt_namespace *ns) |
669 | { | |
670 | spin_lock(&ns->ns_lock); | |
671 | } | |
672 | ||
673 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
674 | { | |
675 | spin_unlock(&ns->ns_lock); | |
676 | } | |
677 | ||
678 | static inline bool mnt_is_cursor(struct mount *mnt) | |
679 | { | |
680 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
681 | } | |
682 | ||
7af1364f EB |
683 | /* |
684 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
685 | * current mount namespace. | |
686 | * | |
687 | * The common case is dentries are not mountpoints at all and that | |
688 | * test is handled inline. For the slow case when we are actually | |
689 | * dealing with a mountpoint of some kind, walk through all of the | |
690 | * mounts in the current mount namespace and test to see if the dentry | |
691 | * is a mountpoint. | |
692 | * | |
693 | * The mount_hashtable is not usable in the context because we | |
694 | * need to identify all mounts that may be in the current mount | |
695 | * namespace not just a mount that happens to have some specified | |
696 | * parent mount. | |
697 | */ | |
698 | bool __is_local_mountpoint(struct dentry *dentry) | |
699 | { | |
700 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
701 | struct mount *mnt; | |
702 | bool is_covered = false; | |
703 | ||
7af1364f | 704 | down_read(&namespace_sem); |
9f6c61f9 | 705 | lock_ns_list(ns); |
7af1364f | 706 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
707 | if (mnt_is_cursor(mnt)) |
708 | continue; | |
7af1364f EB |
709 | is_covered = (mnt->mnt_mountpoint == dentry); |
710 | if (is_covered) | |
711 | break; | |
712 | } | |
9f6c61f9 | 713 | unlock_ns_list(ns); |
7af1364f | 714 | up_read(&namespace_sem); |
5ad05cc8 | 715 | |
7af1364f EB |
716 | return is_covered; |
717 | } | |
718 | ||
e2dfa935 | 719 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 720 | { |
0818bf27 | 721 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
722 | struct mountpoint *mp; |
723 | ||
0818bf27 | 724 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 725 | if (mp->m_dentry == dentry) { |
84d17192 AV |
726 | mp->m_count++; |
727 | return mp; | |
728 | } | |
729 | } | |
e2dfa935 EB |
730 | return NULL; |
731 | } | |
732 | ||
3895dbf8 | 733 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 734 | { |
3895dbf8 | 735 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 736 | int ret; |
84d17192 | 737 | |
3895dbf8 | 738 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
739 | /* might be worth a WARN_ON() */ |
740 | if (d_unlinked(dentry)) | |
741 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
742 | mountpoint: |
743 | read_seqlock_excl(&mount_lock); | |
744 | mp = lookup_mountpoint(dentry); | |
745 | read_sequnlock_excl(&mount_lock); | |
746 | if (mp) | |
747 | goto done; | |
748 | } | |
749 | ||
750 | if (!new) | |
751 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
752 | if (!new) | |
84d17192 AV |
753 | return ERR_PTR(-ENOMEM); |
754 | ||
3895dbf8 EB |
755 | |
756 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 757 | ret = d_set_mounted(dentry); |
eed81007 | 758 | |
3895dbf8 EB |
759 | /* Someone else set d_mounted? */ |
760 | if (ret == -EBUSY) | |
761 | goto mountpoint; | |
762 | ||
763 | /* The dentry is not available as a mountpoint? */ | |
764 | mp = ERR_PTR(ret); | |
765 | if (ret) | |
766 | goto done; | |
767 | ||
768 | /* Add the new mountpoint to the hash table */ | |
769 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 770 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
771 | new->m_count = 1; |
772 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
773 | INIT_HLIST_HEAD(&new->m_list); | |
774 | read_sequnlock_excl(&mount_lock); | |
775 | ||
776 | mp = new; | |
777 | new = NULL; | |
778 | done: | |
779 | kfree(new); | |
84d17192 AV |
780 | return mp; |
781 | } | |
782 | ||
4edbe133 AV |
783 | /* |
784 | * vfsmount lock must be held. Additionally, the caller is responsible | |
785 | * for serializing calls for given disposal list. | |
786 | */ | |
787 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
788 | { |
789 | if (!--mp->m_count) { | |
790 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 791 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
792 | spin_lock(&dentry->d_lock); |
793 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
794 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 795 | dput_to_list(dentry, list); |
0818bf27 | 796 | hlist_del(&mp->m_hash); |
84d17192 AV |
797 | kfree(mp); |
798 | } | |
799 | } | |
800 | ||
4edbe133 AV |
801 | /* called with namespace_lock and vfsmount lock */ |
802 | static void put_mountpoint(struct mountpoint *mp) | |
803 | { | |
804 | __put_mountpoint(mp, &ex_mountpoints); | |
805 | } | |
806 | ||
143c8c91 | 807 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 808 | { |
6b3286ed | 809 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
810 | } |
811 | ||
99b7db7b NP |
812 | /* |
813 | * vfsmount lock must be held for write | |
814 | */ | |
6b3286ed | 815 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
816 | { |
817 | if (ns) { | |
818 | ns->event = ++event; | |
819 | wake_up_interruptible(&ns->poll); | |
820 | } | |
821 | } | |
822 | ||
99b7db7b NP |
823 | /* |
824 | * vfsmount lock must be held for write | |
825 | */ | |
6b3286ed | 826 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
827 | { |
828 | if (ns && ns->event != event) { | |
829 | ns->event = event; | |
830 | wake_up_interruptible(&ns->poll); | |
831 | } | |
832 | } | |
833 | ||
99b7db7b NP |
834 | /* |
835 | * vfsmount lock must be held for write | |
836 | */ | |
e4e59906 | 837 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 838 | { |
e4e59906 | 839 | struct mountpoint *mp; |
0714a533 | 840 | mnt->mnt_parent = mnt; |
a73324da | 841 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 842 | list_del_init(&mnt->mnt_child); |
38129a13 | 843 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 844 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 845 | mp = mnt->mnt_mp; |
84d17192 | 846 | mnt->mnt_mp = NULL; |
e4e59906 | 847 | return mp; |
7bdb11de EB |
848 | } |
849 | ||
6a46c573 EB |
850 | /* |
851 | * vfsmount lock must be held for write | |
852 | */ | |
853 | static void umount_mnt(struct mount *mnt) | |
854 | { | |
e4e59906 | 855 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
856 | } |
857 | ||
99b7db7b NP |
858 | /* |
859 | * vfsmount lock must be held for write | |
860 | */ | |
84d17192 AV |
861 | void mnt_set_mountpoint(struct mount *mnt, |
862 | struct mountpoint *mp, | |
44d964d6 | 863 | struct mount *child_mnt) |
b90fa9ae | 864 | { |
84d17192 | 865 | mp->m_count++; |
3a2393d7 | 866 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 867 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 868 | child_mnt->mnt_parent = mnt; |
84d17192 | 869 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 870 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
871 | } |
872 | ||
1064f874 EB |
873 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
874 | { | |
875 | hlist_add_head_rcu(&mnt->mnt_hash, | |
876 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
877 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
878 | } | |
879 | ||
99b7db7b NP |
880 | /* |
881 | * vfsmount lock must be held for write | |
882 | */ | |
84d17192 AV |
883 | static void attach_mnt(struct mount *mnt, |
884 | struct mount *parent, | |
885 | struct mountpoint *mp) | |
1da177e4 | 886 | { |
84d17192 | 887 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 888 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
889 | } |
890 | ||
1064f874 | 891 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 892 | { |
1064f874 | 893 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
894 | struct mount *old_parent = mnt->mnt_parent; |
895 | ||
896 | list_del_init(&mnt->mnt_child); | |
897 | hlist_del_init(&mnt->mnt_mp_list); | |
898 | hlist_del_init_rcu(&mnt->mnt_hash); | |
899 | ||
900 | attach_mnt(mnt, parent, mp); | |
901 | ||
902 | put_mountpoint(old_mp); | |
1064f874 | 903 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
904 | } |
905 | ||
b90fa9ae | 906 | /* |
99b7db7b | 907 | * vfsmount lock must be held for write |
b90fa9ae | 908 | */ |
1064f874 | 909 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 910 | { |
0714a533 | 911 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 912 | struct mount *m; |
b90fa9ae | 913 | LIST_HEAD(head); |
143c8c91 | 914 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 915 | |
0714a533 | 916 | BUG_ON(parent == mnt); |
b90fa9ae | 917 | |
1a4eeaf2 | 918 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 919 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 920 | m->mnt_ns = n; |
f03c6599 | 921 | |
b90fa9ae RP |
922 | list_splice(&head, n->list.prev); |
923 | ||
d2921684 EB |
924 | n->mounts += n->pending_mounts; |
925 | n->pending_mounts = 0; | |
926 | ||
1064f874 | 927 | __attach_mnt(mnt, parent); |
6b3286ed | 928 | touch_mnt_namespace(n); |
1da177e4 LT |
929 | } |
930 | ||
909b0a88 | 931 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 932 | { |
6b41d536 AV |
933 | struct list_head *next = p->mnt_mounts.next; |
934 | if (next == &p->mnt_mounts) { | |
1da177e4 | 935 | while (1) { |
909b0a88 | 936 | if (p == root) |
1da177e4 | 937 | return NULL; |
6b41d536 AV |
938 | next = p->mnt_child.next; |
939 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 940 | break; |
0714a533 | 941 | p = p->mnt_parent; |
1da177e4 LT |
942 | } |
943 | } | |
6b41d536 | 944 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
945 | } |
946 | ||
315fc83e | 947 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 948 | { |
6b41d536 AV |
949 | struct list_head *prev = p->mnt_mounts.prev; |
950 | while (prev != &p->mnt_mounts) { | |
951 | p = list_entry(prev, struct mount, mnt_child); | |
952 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
953 | } |
954 | return p; | |
955 | } | |
956 | ||
8f291889 AV |
957 | /** |
958 | * vfs_create_mount - Create a mount for a configured superblock | |
959 | * @fc: The configuration context with the superblock attached | |
960 | * | |
961 | * Create a mount to an already configured superblock. If necessary, the | |
962 | * caller should invoke vfs_get_tree() before calling this. | |
963 | * | |
964 | * Note that this does not attach the mount to anything. | |
965 | */ | |
966 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 967 | { |
b105e270 | 968 | struct mount *mnt; |
bd303368 | 969 | struct user_namespace *fs_userns; |
9d412a43 | 970 | |
8f291889 AV |
971 | if (!fc->root) |
972 | return ERR_PTR(-EINVAL); | |
9d412a43 | 973 | |
8f291889 | 974 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
975 | if (!mnt) |
976 | return ERR_PTR(-ENOMEM); | |
977 | ||
8f291889 | 978 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 979 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 980 | |
8f291889 AV |
981 | atomic_inc(&fc->root->d_sb->s_active); |
982 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
983 | mnt->mnt.mnt_root = dget(fc->root); | |
984 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
985 | mnt->mnt_parent = mnt; | |
9d412a43 | 986 | |
bd303368 CB |
987 | fs_userns = mnt->mnt.mnt_sb->s_user_ns; |
988 | if (!initial_idmapping(fs_userns)) | |
989 | mnt->mnt.mnt_userns = get_user_ns(fs_userns); | |
990 | ||
719ea2fb | 991 | lock_mount_hash(); |
8f291889 | 992 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 993 | unlock_mount_hash(); |
b105e270 | 994 | return &mnt->mnt; |
9d412a43 | 995 | } |
8f291889 AV |
996 | EXPORT_SYMBOL(vfs_create_mount); |
997 | ||
998 | struct vfsmount *fc_mount(struct fs_context *fc) | |
999 | { | |
1000 | int err = vfs_get_tree(fc); | |
1001 | if (!err) { | |
1002 | up_write(&fc->root->d_sb->s_umount); | |
1003 | return vfs_create_mount(fc); | |
1004 | } | |
1005 | return ERR_PTR(err); | |
1006 | } | |
1007 | EXPORT_SYMBOL(fc_mount); | |
1008 | ||
9bc61ab1 DH |
1009 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1010 | int flags, const char *name, | |
1011 | void *data) | |
9d412a43 | 1012 | { |
9bc61ab1 | 1013 | struct fs_context *fc; |
8f291889 | 1014 | struct vfsmount *mnt; |
9bc61ab1 | 1015 | int ret = 0; |
9d412a43 AV |
1016 | |
1017 | if (!type) | |
3e1aeb00 | 1018 | return ERR_PTR(-EINVAL); |
9d412a43 | 1019 | |
9bc61ab1 DH |
1020 | fc = fs_context_for_mount(type, flags); |
1021 | if (IS_ERR(fc)) | |
1022 | return ERR_CAST(fc); | |
1023 | ||
3e1aeb00 DH |
1024 | if (name) |
1025 | ret = vfs_parse_fs_string(fc, "source", | |
1026 | name, strlen(name)); | |
9bc61ab1 DH |
1027 | if (!ret) |
1028 | ret = parse_monolithic_mount_data(fc, data); | |
1029 | if (!ret) | |
8f291889 AV |
1030 | mnt = fc_mount(fc); |
1031 | else | |
1032 | mnt = ERR_PTR(ret); | |
9d412a43 | 1033 | |
9bc61ab1 | 1034 | put_fs_context(fc); |
8f291889 | 1035 | return mnt; |
9d412a43 AV |
1036 | } |
1037 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1038 | ||
93faccbb EB |
1039 | struct vfsmount * |
1040 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1041 | const char *name, void *data) | |
1042 | { | |
1043 | /* Until it is worked out how to pass the user namespace | |
1044 | * through from the parent mount to the submount don't support | |
1045 | * unprivileged mounts with submounts. | |
1046 | */ | |
1047 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1048 | return ERR_PTR(-EPERM); | |
1049 | ||
e462ec50 | 1050 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1051 | } |
1052 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1053 | ||
87129cc0 | 1054 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1055 | int flag) |
1da177e4 | 1056 | { |
87129cc0 | 1057 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1058 | struct mount *mnt; |
1059 | int err; | |
1da177e4 | 1060 | |
be34d1a3 DH |
1061 | mnt = alloc_vfsmnt(old->mnt_devname); |
1062 | if (!mnt) | |
1063 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1064 | |
7a472ef4 | 1065 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1066 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1067 | else | |
1068 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1069 | |
be34d1a3 DH |
1070 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1071 | err = mnt_alloc_group_id(mnt); | |
1072 | if (err) | |
1073 | goto out_free; | |
1da177e4 | 1074 | } |
be34d1a3 | 1075 | |
16a34adb AV |
1076 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1077 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1078 | |
be34d1a3 | 1079 | atomic_inc(&sb->s_active); |
a6435940 | 1080 | mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); |
bd303368 | 1081 | if (!initial_idmapping(mnt->mnt.mnt_userns)) |
a6435940 | 1082 | mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); |
be34d1a3 DH |
1083 | mnt->mnt.mnt_sb = sb; |
1084 | mnt->mnt.mnt_root = dget(root); | |
1085 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1086 | mnt->mnt_parent = mnt; | |
719ea2fb | 1087 | lock_mount_hash(); |
be34d1a3 | 1088 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1089 | unlock_mount_hash(); |
be34d1a3 | 1090 | |
7a472ef4 EB |
1091 | if ((flag & CL_SLAVE) || |
1092 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1093 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1094 | mnt->mnt_master = old; | |
1095 | CLEAR_MNT_SHARED(mnt); | |
1096 | } else if (!(flag & CL_PRIVATE)) { | |
1097 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1098 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1099 | if (IS_MNT_SLAVE(old)) | |
1100 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1101 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1102 | } else { |
1103 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1104 | } |
1105 | if (flag & CL_MAKE_SHARED) | |
1106 | set_mnt_shared(mnt); | |
1107 | ||
1108 | /* stick the duplicate mount on the same expiry list | |
1109 | * as the original if that was on one */ | |
1110 | if (flag & CL_EXPIRE) { | |
1111 | if (!list_empty(&old->mnt_expire)) | |
1112 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1113 | } | |
1114 | ||
cb338d06 | 1115 | return mnt; |
719f5d7f MS |
1116 | |
1117 | out_free: | |
8ffcb32e | 1118 | mnt_free_id(mnt); |
719f5d7f | 1119 | free_vfsmnt(mnt); |
be34d1a3 | 1120 | return ERR_PTR(err); |
1da177e4 LT |
1121 | } |
1122 | ||
9ea459e1 AV |
1123 | static void cleanup_mnt(struct mount *mnt) |
1124 | { | |
56cbb429 AV |
1125 | struct hlist_node *p; |
1126 | struct mount *m; | |
9ea459e1 | 1127 | /* |
56cbb429 AV |
1128 | * The warning here probably indicates that somebody messed |
1129 | * up a mnt_want/drop_write() pair. If this happens, the | |
1130 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1131 | * The locking used to deal with mnt_count decrement provides barriers, |
1132 | * so mnt_get_writers() below is safe. | |
1133 | */ | |
1134 | WARN_ON(mnt_get_writers(mnt)); | |
1135 | if (unlikely(mnt->mnt_pins.first)) | |
1136 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1137 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1138 | hlist_del(&m->mnt_umount); | |
1139 | mntput(&m->mnt); | |
1140 | } | |
9ea459e1 AV |
1141 | fsnotify_vfsmount_delete(&mnt->mnt); |
1142 | dput(mnt->mnt.mnt_root); | |
1143 | deactivate_super(mnt->mnt.mnt_sb); | |
1144 | mnt_free_id(mnt); | |
1145 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1146 | } | |
1147 | ||
1148 | static void __cleanup_mnt(struct rcu_head *head) | |
1149 | { | |
1150 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1151 | } | |
1152 | ||
1153 | static LLIST_HEAD(delayed_mntput_list); | |
1154 | static void delayed_mntput(struct work_struct *unused) | |
1155 | { | |
1156 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1157 | struct mount *m, *t; |
9ea459e1 | 1158 | |
29785735 BP |
1159 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1160 | cleanup_mnt(m); | |
9ea459e1 AV |
1161 | } |
1162 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1163 | ||
900148dc | 1164 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1165 | { |
4edbe133 | 1166 | LIST_HEAD(list); |
edf7ddbf | 1167 | int count; |
4edbe133 | 1168 | |
48a066e7 | 1169 | rcu_read_lock(); |
9ea0a46c AV |
1170 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1171 | /* | |
1172 | * Since we don't do lock_mount_hash() here, | |
1173 | * ->mnt_ns can change under us. However, if it's | |
1174 | * non-NULL, then there's a reference that won't | |
1175 | * be dropped until after an RCU delay done after | |
1176 | * turning ->mnt_ns NULL. So if we observe it | |
1177 | * non-NULL under rcu_read_lock(), the reference | |
1178 | * we are dropping is not the final one. | |
1179 | */ | |
1180 | mnt_add_count(mnt, -1); | |
48a066e7 | 1181 | rcu_read_unlock(); |
f03c6599 | 1182 | return; |
b3e19d92 | 1183 | } |
719ea2fb | 1184 | lock_mount_hash(); |
119e1ef8 AV |
1185 | /* |
1186 | * make sure that if __legitimize_mnt() has not seen us grab | |
1187 | * mount_lock, we'll see their refcount increment here. | |
1188 | */ | |
1189 | smp_mb(); | |
9ea0a46c | 1190 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1191 | count = mnt_get_count(mnt); |
1192 | if (count != 0) { | |
1193 | WARN_ON(count < 0); | |
48a066e7 | 1194 | rcu_read_unlock(); |
719ea2fb | 1195 | unlock_mount_hash(); |
99b7db7b NP |
1196 | return; |
1197 | } | |
48a066e7 AV |
1198 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1199 | rcu_read_unlock(); | |
1200 | unlock_mount_hash(); | |
1201 | return; | |
1202 | } | |
1203 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1204 | rcu_read_unlock(); | |
962830df | 1205 | |
39f7c4db | 1206 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1207 | |
1208 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1209 | struct mount *p, *tmp; | |
1210 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1211 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1212 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1213 | } |
1214 | } | |
719ea2fb | 1215 | unlock_mount_hash(); |
4edbe133 | 1216 | shrink_dentry_list(&list); |
649a795a | 1217 | |
9ea459e1 AV |
1218 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1219 | struct task_struct *task = current; | |
1220 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1221 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1222 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1223 | return; |
1224 | } | |
1225 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1226 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1227 | return; | |
1228 | } | |
1229 | cleanup_mnt(mnt); | |
b3e19d92 | 1230 | } |
b3e19d92 NP |
1231 | |
1232 | void mntput(struct vfsmount *mnt) | |
1233 | { | |
1234 | if (mnt) { | |
863d684f | 1235 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1236 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1237 | if (unlikely(m->mnt_expiry_mark)) |
1238 | m->mnt_expiry_mark = 0; | |
1239 | mntput_no_expire(m); | |
b3e19d92 NP |
1240 | } |
1241 | } | |
1242 | EXPORT_SYMBOL(mntput); | |
1243 | ||
1244 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1245 | { | |
1246 | if (mnt) | |
83adc753 | 1247 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1248 | return mnt; |
1249 | } | |
1250 | EXPORT_SYMBOL(mntget); | |
1251 | ||
1f287bc4 RD |
1252 | /** |
1253 | * path_is_mountpoint() - Check if path is a mount in the current namespace. | |
1254 | * @path: path to check | |
c6609c0a IK |
1255 | * |
1256 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1257 | * not mounted in any namespace and that common case is handled inline. | |
1258 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1259 | * mounts using a given dentry in a different namespace. This function | |
1260 | * checks if the passed in path is a mountpoint rather than the dentry | |
1261 | * alone. | |
1262 | */ | |
1263 | bool path_is_mountpoint(const struct path *path) | |
1264 | { | |
1265 | unsigned seq; | |
1266 | bool res; | |
1267 | ||
1268 | if (!d_mountpoint(path->dentry)) | |
1269 | return false; | |
1270 | ||
1271 | rcu_read_lock(); | |
1272 | do { | |
1273 | seq = read_seqbegin(&mount_lock); | |
1274 | res = __path_is_mountpoint(path); | |
1275 | } while (read_seqretry(&mount_lock, seq)); | |
1276 | rcu_read_unlock(); | |
1277 | ||
1278 | return res; | |
1279 | } | |
1280 | EXPORT_SYMBOL(path_is_mountpoint); | |
1281 | ||
ca71cf71 | 1282 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1283 | { |
3064c356 AV |
1284 | struct mount *p; |
1285 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1286 | if (IS_ERR(p)) | |
1287 | return ERR_CAST(p); | |
1288 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1289 | return &p->mnt; | |
7b7b1ace | 1290 | } |
1da177e4 | 1291 | |
a1a2c409 | 1292 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1293 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1294 | struct list_head *p) | |
1295 | { | |
1296 | struct mount *mnt, *ret = NULL; | |
1297 | ||
1298 | lock_ns_list(ns); | |
1299 | list_for_each_continue(p, &ns->list) { | |
1300 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1301 | if (!mnt_is_cursor(mnt)) { | |
1302 | ret = mnt; | |
1303 | break; | |
1304 | } | |
1305 | } | |
1306 | unlock_ns_list(ns); | |
1307 | ||
1308 | return ret; | |
1309 | } | |
1310 | ||
0226f492 | 1311 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1312 | static void *m_start(struct seq_file *m, loff_t *pos) |
1313 | { | |
ede1bf0d | 1314 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1315 | struct list_head *prev; |
1da177e4 | 1316 | |
390c6843 | 1317 | down_read(&namespace_sem); |
9f6c61f9 MS |
1318 | if (!*pos) { |
1319 | prev = &p->ns->list; | |
1320 | } else { | |
1321 | prev = &p->cursor.mnt_list; | |
1322 | ||
1323 | /* Read after we'd reached the end? */ | |
1324 | if (list_empty(prev)) | |
1325 | return NULL; | |
c7999c36 AV |
1326 | } |
1327 | ||
9f6c61f9 | 1328 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1329 | } |
1330 | ||
1331 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1332 | { | |
ede1bf0d | 1333 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1334 | struct mount *mnt = v; |
b0765fb8 | 1335 | |
9f6c61f9 MS |
1336 | ++*pos; |
1337 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1338 | } |
1339 | ||
1340 | static void m_stop(struct seq_file *m, void *v) | |
1341 | { | |
9f6c61f9 MS |
1342 | struct proc_mounts *p = m->private; |
1343 | struct mount *mnt = v; | |
1344 | ||
1345 | lock_ns_list(p->ns); | |
1346 | if (mnt) | |
1347 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1348 | else | |
1349 | list_del_init(&p->cursor.mnt_list); | |
1350 | unlock_ns_list(p->ns); | |
390c6843 | 1351 | up_read(&namespace_sem); |
1da177e4 LT |
1352 | } |
1353 | ||
0226f492 | 1354 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1355 | { |
ede1bf0d | 1356 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1357 | struct mount *r = v; |
0226f492 | 1358 | return p->show(m, &r->mnt); |
1da177e4 LT |
1359 | } |
1360 | ||
a1a2c409 | 1361 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1362 | .start = m_start, |
1363 | .next = m_next, | |
1364 | .stop = m_stop, | |
0226f492 | 1365 | .show = m_show, |
b4629fe2 | 1366 | }; |
9f6c61f9 MS |
1367 | |
1368 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1369 | { | |
1370 | down_read(&namespace_sem); | |
1371 | lock_ns_list(ns); | |
1372 | list_del(&cursor->mnt_list); | |
1373 | unlock_ns_list(ns); | |
1374 | up_read(&namespace_sem); | |
1375 | } | |
a1a2c409 | 1376 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1377 | |
1da177e4 LT |
1378 | /** |
1379 | * may_umount_tree - check if a mount tree is busy | |
1f287bc4 | 1380 | * @m: root of mount tree |
1da177e4 LT |
1381 | * |
1382 | * This is called to check if a tree of mounts has any | |
1383 | * open files, pwds, chroots or sub mounts that are | |
1384 | * busy. | |
1385 | */ | |
909b0a88 | 1386 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1387 | { |
909b0a88 | 1388 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1389 | int actual_refs = 0; |
1390 | int minimum_refs = 0; | |
315fc83e | 1391 | struct mount *p; |
909b0a88 | 1392 | BUG_ON(!m); |
1da177e4 | 1393 | |
b3e19d92 | 1394 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1395 | lock_mount_hash(); |
909b0a88 | 1396 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1397 | actual_refs += mnt_get_count(p); |
1da177e4 | 1398 | minimum_refs += 2; |
1da177e4 | 1399 | } |
719ea2fb | 1400 | unlock_mount_hash(); |
1da177e4 LT |
1401 | |
1402 | if (actual_refs > minimum_refs) | |
e3474a8e | 1403 | return 0; |
1da177e4 | 1404 | |
e3474a8e | 1405 | return 1; |
1da177e4 LT |
1406 | } |
1407 | ||
1408 | EXPORT_SYMBOL(may_umount_tree); | |
1409 | ||
1410 | /** | |
1411 | * may_umount - check if a mount point is busy | |
1412 | * @mnt: root of mount | |
1413 | * | |
1414 | * This is called to check if a mount point has any | |
1415 | * open files, pwds, chroots or sub mounts. If the | |
1416 | * mount has sub mounts this will return busy | |
1417 | * regardless of whether the sub mounts are busy. | |
1418 | * | |
1419 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1420 | * give false negatives. The main reason why it's here is that we need | |
1421 | * a non-destructive way to look for easily umountable filesystems. | |
1422 | */ | |
1423 | int may_umount(struct vfsmount *mnt) | |
1424 | { | |
e3474a8e | 1425 | int ret = 1; |
8ad08d8a | 1426 | down_read(&namespace_sem); |
719ea2fb | 1427 | lock_mount_hash(); |
1ab59738 | 1428 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1429 | ret = 0; |
719ea2fb | 1430 | unlock_mount_hash(); |
8ad08d8a | 1431 | up_read(&namespace_sem); |
a05964f3 | 1432 | return ret; |
1da177e4 LT |
1433 | } |
1434 | ||
1435 | EXPORT_SYMBOL(may_umount); | |
1436 | ||
97216be0 | 1437 | static void namespace_unlock(void) |
70fbcdf4 | 1438 | { |
a3b3c562 | 1439 | struct hlist_head head; |
56cbb429 AV |
1440 | struct hlist_node *p; |
1441 | struct mount *m; | |
4edbe133 | 1442 | LIST_HEAD(list); |
97216be0 | 1443 | |
a3b3c562 | 1444 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1445 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1446 | |
97216be0 AV |
1447 | up_write(&namespace_sem); |
1448 | ||
4edbe133 AV |
1449 | shrink_dentry_list(&list); |
1450 | ||
a3b3c562 EB |
1451 | if (likely(hlist_empty(&head))) |
1452 | return; | |
1453 | ||
22cb7405 | 1454 | synchronize_rcu_expedited(); |
48a066e7 | 1455 | |
56cbb429 AV |
1456 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1457 | hlist_del(&m->mnt_umount); | |
1458 | mntput(&m->mnt); | |
1459 | } | |
70fbcdf4 RP |
1460 | } |
1461 | ||
97216be0 | 1462 | static inline void namespace_lock(void) |
e3197d83 | 1463 | { |
97216be0 | 1464 | down_write(&namespace_sem); |
e3197d83 AV |
1465 | } |
1466 | ||
e819f152 EB |
1467 | enum umount_tree_flags { |
1468 | UMOUNT_SYNC = 1, | |
1469 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1470 | UMOUNT_CONNECTED = 4, |
e819f152 | 1471 | }; |
f2d0a123 EB |
1472 | |
1473 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1474 | { | |
1475 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1476 | if (how & UMOUNT_SYNC) | |
1477 | return true; | |
1478 | ||
1479 | /* A mount without a parent has nothing to be connected to */ | |
1480 | if (!mnt_has_parent(mnt)) | |
1481 | return true; | |
1482 | ||
1483 | /* Because the reference counting rules change when mounts are | |
1484 | * unmounted and connected, umounted mounts may not be | |
1485 | * connected to mounted mounts. | |
1486 | */ | |
1487 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1488 | return true; | |
1489 | ||
1490 | /* Has it been requested that the mount remain connected? */ | |
1491 | if (how & UMOUNT_CONNECTED) | |
1492 | return false; | |
1493 | ||
1494 | /* Is the mount locked such that it needs to remain connected? */ | |
1495 | if (IS_MNT_LOCKED(mnt)) | |
1496 | return false; | |
1497 | ||
1498 | /* By default disconnect the mount */ | |
1499 | return true; | |
1500 | } | |
1501 | ||
99b7db7b | 1502 | /* |
48a066e7 | 1503 | * mount_lock must be held |
99b7db7b NP |
1504 | * namespace_sem must be held for write |
1505 | */ | |
e819f152 | 1506 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1507 | { |
c003b26f | 1508 | LIST_HEAD(tmp_list); |
315fc83e | 1509 | struct mount *p; |
1da177e4 | 1510 | |
5d88457e EB |
1511 | if (how & UMOUNT_PROPAGATE) |
1512 | propagate_mount_unlock(mnt); | |
1513 | ||
c003b26f | 1514 | /* Gather the mounts to umount */ |
590ce4bc EB |
1515 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1516 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1517 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1518 | } |
1da177e4 | 1519 | |
411a938b | 1520 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1521 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1522 | list_del_init(&p->mnt_child); |
c003b26f | 1523 | } |
88b368f2 | 1524 | |
c003b26f | 1525 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1526 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1527 | propagate_umount(&tmp_list); |
a05964f3 | 1528 | |
c003b26f | 1529 | while (!list_empty(&tmp_list)) { |
d2921684 | 1530 | struct mnt_namespace *ns; |
ce07d891 | 1531 | bool disconnect; |
c003b26f | 1532 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1533 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1534 | list_del_init(&p->mnt_list); |
d2921684 EB |
1535 | ns = p->mnt_ns; |
1536 | if (ns) { | |
1537 | ns->mounts--; | |
1538 | __touch_mnt_namespace(ns); | |
1539 | } | |
143c8c91 | 1540 | p->mnt_ns = NULL; |
e819f152 | 1541 | if (how & UMOUNT_SYNC) |
48a066e7 | 1542 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1543 | |
f2d0a123 | 1544 | disconnect = disconnect_mount(p, how); |
676da58d | 1545 | if (mnt_has_parent(p)) { |
81b6b061 | 1546 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1547 | if (!disconnect) { |
1548 | /* Don't forget about p */ | |
1549 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1550 | } else { | |
1551 | umount_mnt(p); | |
1552 | } | |
7c4b93d8 | 1553 | } |
0f0afb1d | 1554 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1555 | if (disconnect) |
1556 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1557 | } |
1558 | } | |
1559 | ||
b54b9be7 | 1560 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1561 | |
8d0347f6 DH |
1562 | static int do_umount_root(struct super_block *sb) |
1563 | { | |
1564 | int ret = 0; | |
1565 | ||
1566 | down_write(&sb->s_umount); | |
1567 | if (!sb_rdonly(sb)) { | |
1568 | struct fs_context *fc; | |
1569 | ||
1570 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1571 | SB_RDONLY); | |
1572 | if (IS_ERR(fc)) { | |
1573 | ret = PTR_ERR(fc); | |
1574 | } else { | |
1575 | ret = parse_monolithic_mount_data(fc, NULL); | |
1576 | if (!ret) | |
1577 | ret = reconfigure_super(fc); | |
1578 | put_fs_context(fc); | |
1579 | } | |
1580 | } | |
1581 | up_write(&sb->s_umount); | |
1582 | return ret; | |
1583 | } | |
1584 | ||
1ab59738 | 1585 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1586 | { |
1ab59738 | 1587 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1588 | int retval; |
1589 | ||
1ab59738 | 1590 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1591 | if (retval) |
1592 | return retval; | |
1593 | ||
1594 | /* | |
1595 | * Allow userspace to request a mountpoint be expired rather than | |
1596 | * unmounting unconditionally. Unmount only happens if: | |
1597 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1598 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1599 | */ | |
1600 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1601 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1602 | flags & (MNT_FORCE | MNT_DETACH)) |
1603 | return -EINVAL; | |
1604 | ||
b3e19d92 NP |
1605 | /* |
1606 | * probably don't strictly need the lock here if we examined | |
1607 | * all race cases, but it's a slowpath. | |
1608 | */ | |
719ea2fb | 1609 | lock_mount_hash(); |
83adc753 | 1610 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1611 | unlock_mount_hash(); |
1da177e4 | 1612 | return -EBUSY; |
b3e19d92 | 1613 | } |
719ea2fb | 1614 | unlock_mount_hash(); |
1da177e4 | 1615 | |
863d684f | 1616 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1617 | return -EAGAIN; |
1618 | } | |
1619 | ||
1620 | /* | |
1621 | * If we may have to abort operations to get out of this | |
1622 | * mount, and they will themselves hold resources we must | |
1623 | * allow the fs to do things. In the Unix tradition of | |
1624 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1625 | * might fail to complete on the first run through as other tasks | |
1626 | * must return, and the like. Thats for the mount program to worry | |
1627 | * about for the moment. | |
1628 | */ | |
1629 | ||
42faad99 | 1630 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1631 | sb->s_op->umount_begin(sb); |
42faad99 | 1632 | } |
1da177e4 LT |
1633 | |
1634 | /* | |
1635 | * No sense to grab the lock for this test, but test itself looks | |
1636 | * somewhat bogus. Suggestions for better replacement? | |
1637 | * Ho-hum... In principle, we might treat that as umount + switch | |
1638 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1639 | * Actually it makes sense, especially if rootfs would contain a | |
1640 | * /reboot - static binary that would close all descriptors and | |
1641 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1642 | */ | |
1ab59738 | 1643 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1644 | /* |
1645 | * Special case for "unmounting" root ... | |
1646 | * we just try to remount it readonly. | |
1647 | */ | |
bc6155d1 | 1648 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1649 | return -EPERM; |
8d0347f6 | 1650 | return do_umount_root(sb); |
1da177e4 LT |
1651 | } |
1652 | ||
97216be0 | 1653 | namespace_lock(); |
719ea2fb | 1654 | lock_mount_hash(); |
1da177e4 | 1655 | |
25d202ed EB |
1656 | /* Recheck MNT_LOCKED with the locks held */ |
1657 | retval = -EINVAL; | |
1658 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1659 | goto out; | |
1660 | ||
1661 | event++; | |
48a066e7 | 1662 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1663 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1664 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1665 | retval = 0; |
48a066e7 AV |
1666 | } else { |
1667 | shrink_submounts(mnt); | |
1668 | retval = -EBUSY; | |
1669 | if (!propagate_mount_busy(mnt, 2)) { | |
1670 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1671 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1672 | retval = 0; |
1673 | } | |
1da177e4 | 1674 | } |
25d202ed | 1675 | out: |
719ea2fb | 1676 | unlock_mount_hash(); |
e3197d83 | 1677 | namespace_unlock(); |
1da177e4 LT |
1678 | return retval; |
1679 | } | |
1680 | ||
80b5dce8 EB |
1681 | /* |
1682 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1683 | * | |
1684 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1685 | * to an existing mountpoint, and wind up leaking the mount. | |
1686 | * detach_mounts allows lazily unmounting those mounts instead of | |
1687 | * leaking them. | |
1688 | * | |
1689 | * The caller may hold dentry->d_inode->i_mutex. | |
1690 | */ | |
1691 | void __detach_mounts(struct dentry *dentry) | |
1692 | { | |
1693 | struct mountpoint *mp; | |
1694 | struct mount *mnt; | |
1695 | ||
1696 | namespace_lock(); | |
3895dbf8 | 1697 | lock_mount_hash(); |
80b5dce8 | 1698 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1699 | if (!mp) |
80b5dce8 EB |
1700 | goto out_unlock; |
1701 | ||
e06b933e | 1702 | event++; |
80b5dce8 EB |
1703 | while (!hlist_empty(&mp->m_list)) { |
1704 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1705 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1706 | umount_mnt(mnt); |
56cbb429 | 1707 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1708 | } |
e0c9c0af | 1709 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1710 | } |
80b5dce8 EB |
1711 | put_mountpoint(mp); |
1712 | out_unlock: | |
3895dbf8 | 1713 | unlock_mount_hash(); |
80b5dce8 EB |
1714 | namespace_unlock(); |
1715 | } | |
1716 | ||
dd111b31 | 1717 | /* |
9b40bc90 AV |
1718 | * Is the caller allowed to modify his namespace? |
1719 | */ | |
1720 | static inline bool may_mount(void) | |
1721 | { | |
1722 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1723 | } | |
1724 | ||
f7e33bdb | 1725 | static void warn_mandlock(void) |
9e8925b6 | 1726 | { |
f7e33bdb JL |
1727 | pr_warn_once("=======================================================\n" |
1728 | "WARNING: The mand mount option has been deprecated and\n" | |
1729 | " and is ignored by this kernel. Remove the mand\n" | |
1730 | " option from the mount to silence this warning.\n" | |
1731 | "=======================================================\n"); | |
9e8925b6 JL |
1732 | } |
1733 | ||
25ccd24f | 1734 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1735 | { |
25ccd24f | 1736 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1737 | |
9b40bc90 AV |
1738 | if (!may_mount()) |
1739 | return -EPERM; | |
41525f56 | 1740 | if (path->dentry != path->mnt->mnt_root) |
25ccd24f | 1741 | return -EINVAL; |
143c8c91 | 1742 | if (!check_mnt(mnt)) |
25ccd24f | 1743 | return -EINVAL; |
25d202ed | 1744 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1745 | return -EINVAL; |
b2f5d4dc | 1746 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1747 | return -EPERM; |
1748 | return 0; | |
1749 | } | |
1750 | ||
a0a6df9a | 1751 | // caller is responsible for flags being sane |
25ccd24f CH |
1752 | int path_umount(struct path *path, int flags) |
1753 | { | |
1754 | struct mount *mnt = real_mount(path->mnt); | |
1755 | int ret; | |
1756 | ||
1757 | ret = can_umount(path, flags); | |
1758 | if (!ret) | |
1759 | ret = do_umount(mnt, flags); | |
1da177e4 | 1760 | |
429731b1 | 1761 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1762 | dput(path->dentry); |
900148dc | 1763 | mntput_no_expire(mnt); |
25ccd24f | 1764 | return ret; |
1da177e4 LT |
1765 | } |
1766 | ||
09267def | 1767 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1768 | { |
1769 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1770 | struct path path; | |
1771 | int ret; | |
1772 | ||
a0a6df9a AV |
1773 | // basic validity checks done first |
1774 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1775 | return -EINVAL; | |
1776 | ||
41525f56 CH |
1777 | if (!(flags & UMOUNT_NOFOLLOW)) |
1778 | lookup_flags |= LOOKUP_FOLLOW; | |
1779 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1780 | if (ret) | |
1781 | return ret; | |
1782 | return path_umount(&path, flags); | |
1783 | } | |
1784 | ||
3a18ef5c DB |
1785 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1786 | { | |
1787 | return ksys_umount(name, flags); | |
1788 | } | |
1789 | ||
1da177e4 LT |
1790 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1791 | ||
1792 | /* | |
b58fed8b | 1793 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1794 | */ |
bdc480e3 | 1795 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1796 | { |
3a18ef5c | 1797 | return ksys_umount(name, 0); |
1da177e4 LT |
1798 | } |
1799 | ||
1800 | #endif | |
1801 | ||
4ce5d2b1 | 1802 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1803 | { |
4ce5d2b1 | 1804 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1805 | return dentry->d_op == &ns_dentry_operations && |
1806 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1807 | } |
1808 | ||
213921f9 | 1809 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1810 | { |
1811 | return container_of(ns, struct mnt_namespace, ns); | |
1812 | } | |
1813 | ||
303cc571 CB |
1814 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1815 | { | |
1816 | return &mnt->ns; | |
1817 | } | |
1818 | ||
4ce5d2b1 EB |
1819 | static bool mnt_ns_loop(struct dentry *dentry) |
1820 | { | |
1821 | /* Could bind mounting the mount namespace inode cause a | |
1822 | * mount namespace loop? | |
1823 | */ | |
1824 | struct mnt_namespace *mnt_ns; | |
1825 | if (!is_mnt_ns_file(dentry)) | |
1826 | return false; | |
1827 | ||
f77c8014 | 1828 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1829 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1830 | } | |
1831 | ||
87129cc0 | 1832 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1833 | int flag) |
1da177e4 | 1834 | { |
84d17192 | 1835 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1836 | |
4ce5d2b1 EB |
1837 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1838 | return ERR_PTR(-EINVAL); | |
1839 | ||
1840 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1841 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1842 | |
36341f64 | 1843 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1844 | if (IS_ERR(q)) |
1845 | return q; | |
1846 | ||
a73324da | 1847 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1848 | |
1849 | p = mnt; | |
6b41d536 | 1850 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1851 | struct mount *s; |
7ec02ef1 | 1852 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1853 | continue; |
1854 | ||
909b0a88 | 1855 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1856 | if (!(flag & CL_COPY_UNBINDABLE) && |
1857 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1858 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1859 | /* Both unbindable and locked. */ | |
1860 | q = ERR_PTR(-EPERM); | |
1861 | goto out; | |
1862 | } else { | |
1863 | s = skip_mnt_tree(s); | |
1864 | continue; | |
1865 | } | |
4ce5d2b1 EB |
1866 | } |
1867 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1868 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1869 | s = skip_mnt_tree(s); |
1870 | continue; | |
1871 | } | |
0714a533 AV |
1872 | while (p != s->mnt_parent) { |
1873 | p = p->mnt_parent; | |
1874 | q = q->mnt_parent; | |
1da177e4 | 1875 | } |
87129cc0 | 1876 | p = s; |
84d17192 | 1877 | parent = q; |
87129cc0 | 1878 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1879 | if (IS_ERR(q)) |
1880 | goto out; | |
719ea2fb | 1881 | lock_mount_hash(); |
1a4eeaf2 | 1882 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1883 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1884 | unlock_mount_hash(); |
1da177e4 LT |
1885 | } |
1886 | } | |
1887 | return res; | |
be34d1a3 | 1888 | out: |
1da177e4 | 1889 | if (res) { |
719ea2fb | 1890 | lock_mount_hash(); |
e819f152 | 1891 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1892 | unlock_mount_hash(); |
1da177e4 | 1893 | } |
be34d1a3 | 1894 | return q; |
1da177e4 LT |
1895 | } |
1896 | ||
be34d1a3 DH |
1897 | /* Caller should check returned pointer for errors */ |
1898 | ||
ca71cf71 | 1899 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1900 | { |
cb338d06 | 1901 | struct mount *tree; |
97216be0 | 1902 | namespace_lock(); |
cd4a4017 EB |
1903 | if (!check_mnt(real_mount(path->mnt))) |
1904 | tree = ERR_PTR(-EINVAL); | |
1905 | else | |
1906 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1907 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1908 | namespace_unlock(); |
be34d1a3 | 1909 | if (IS_ERR(tree)) |
52e220d3 | 1910 | return ERR_CAST(tree); |
be34d1a3 | 1911 | return &tree->mnt; |
8aec0809 AV |
1912 | } |
1913 | ||
a07b2000 AV |
1914 | static void free_mnt_ns(struct mnt_namespace *); |
1915 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1916 | ||
1917 | void dissolve_on_fput(struct vfsmount *mnt) | |
1918 | { | |
1919 | struct mnt_namespace *ns; | |
1920 | namespace_lock(); | |
1921 | lock_mount_hash(); | |
1922 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1923 | if (ns) { |
1924 | if (is_anon_ns(ns)) | |
1925 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1926 | else | |
1927 | ns = NULL; | |
1928 | } | |
a07b2000 AV |
1929 | unlock_mount_hash(); |
1930 | namespace_unlock(); | |
44dfd84a DH |
1931 | if (ns) |
1932 | free_mnt_ns(ns); | |
a07b2000 AV |
1933 | } |
1934 | ||
8aec0809 AV |
1935 | void drop_collected_mounts(struct vfsmount *mnt) |
1936 | { | |
97216be0 | 1937 | namespace_lock(); |
719ea2fb | 1938 | lock_mount_hash(); |
9c8e0a1b | 1939 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1940 | unlock_mount_hash(); |
3ab6abee | 1941 | namespace_unlock(); |
8aec0809 AV |
1942 | } |
1943 | ||
427215d8 MS |
1944 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1945 | { | |
1946 | struct mount *child; | |
1947 | ||
1948 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1949 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1950 | continue; | |
1951 | ||
1952 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1953 | return true; | |
1954 | } | |
1955 | return false; | |
1956 | } | |
1957 | ||
c771d683 MS |
1958 | /** |
1959 | * clone_private_mount - create a private clone of a path | |
1f287bc4 | 1960 | * @path: path to clone |
c771d683 | 1961 | * |
1f287bc4 RD |
1962 | * This creates a new vfsmount, which will be the clone of @path. The new mount |
1963 | * will not be attached anywhere in the namespace and will be private (i.e. | |
1964 | * changes to the originating mount won't be propagated into this). | |
c771d683 MS |
1965 | * |
1966 | * Release with mntput(). | |
1967 | */ | |
ca71cf71 | 1968 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1969 | { |
1970 | struct mount *old_mnt = real_mount(path->mnt); | |
1971 | struct mount *new_mnt; | |
1972 | ||
427215d8 | 1973 | down_read(&namespace_sem); |
c771d683 | 1974 | if (IS_MNT_UNBINDABLE(old_mnt)) |
427215d8 MS |
1975 | goto invalid; |
1976 | ||
1977 | if (!check_mnt(old_mnt)) | |
1978 | goto invalid; | |
1979 | ||
1980 | if (has_locked_children(old_mnt, path->dentry)) | |
1981 | goto invalid; | |
c771d683 | 1982 | |
c771d683 | 1983 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
427215d8 MS |
1984 | up_read(&namespace_sem); |
1985 | ||
c771d683 MS |
1986 | if (IS_ERR(new_mnt)) |
1987 | return ERR_CAST(new_mnt); | |
1988 | ||
df820f8d MS |
1989 | /* Longterm mount to be removed by kern_unmount*() */ |
1990 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
1991 | ||
c771d683 | 1992 | return &new_mnt->mnt; |
427215d8 MS |
1993 | |
1994 | invalid: | |
1995 | up_read(&namespace_sem); | |
1996 | return ERR_PTR(-EINVAL); | |
c771d683 MS |
1997 | } |
1998 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1999 | ||
1f707137 AV |
2000 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
2001 | struct vfsmount *root) | |
2002 | { | |
1a4eeaf2 | 2003 | struct mount *mnt; |
1f707137 AV |
2004 | int res = f(root, arg); |
2005 | if (res) | |
2006 | return res; | |
1a4eeaf2 AV |
2007 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
2008 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
2009 | if (res) |
2010 | return res; | |
2011 | } | |
2012 | return 0; | |
2013 | } | |
2014 | ||
3bd045cc AV |
2015 | static void lock_mnt_tree(struct mount *mnt) |
2016 | { | |
2017 | struct mount *p; | |
2018 | ||
2019 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2020 | int flags = p->mnt.mnt_flags; | |
2021 | /* Don't allow unprivileged users to change mount flags */ | |
2022 | flags |= MNT_LOCK_ATIME; | |
2023 | ||
2024 | if (flags & MNT_READONLY) | |
2025 | flags |= MNT_LOCK_READONLY; | |
2026 | ||
2027 | if (flags & MNT_NODEV) | |
2028 | flags |= MNT_LOCK_NODEV; | |
2029 | ||
2030 | if (flags & MNT_NOSUID) | |
2031 | flags |= MNT_LOCK_NOSUID; | |
2032 | ||
2033 | if (flags & MNT_NOEXEC) | |
2034 | flags |= MNT_LOCK_NOEXEC; | |
2035 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2036 | if (list_empty(&p->mnt_expire)) | |
2037 | flags |= MNT_LOCKED; | |
2038 | p->mnt.mnt_flags = flags; | |
2039 | } | |
2040 | } | |
2041 | ||
4b8b21f4 | 2042 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2043 | { |
315fc83e | 2044 | struct mount *p; |
719f5d7f | 2045 | |
909b0a88 | 2046 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2047 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2048 | mnt_release_group_id(p); |
719f5d7f MS |
2049 | } |
2050 | } | |
2051 | ||
4b8b21f4 | 2052 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2053 | { |
315fc83e | 2054 | struct mount *p; |
719f5d7f | 2055 | |
909b0a88 | 2056 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2057 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2058 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2059 | if (err) { |
4b8b21f4 | 2060 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2061 | return err; |
2062 | } | |
2063 | } | |
2064 | } | |
2065 | ||
2066 | return 0; | |
2067 | } | |
2068 | ||
d2921684 EB |
2069 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2070 | { | |
2071 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
2072 | unsigned int mounts = 0, old, pending, sum; | |
2073 | struct mount *p; | |
2074 | ||
2075 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
2076 | mounts++; | |
2077 | ||
2078 | old = ns->mounts; | |
2079 | pending = ns->pending_mounts; | |
2080 | sum = old + pending; | |
2081 | if ((old > sum) || | |
2082 | (pending > sum) || | |
2083 | (max < sum) || | |
2084 | (mounts > (max - sum))) | |
2085 | return -ENOSPC; | |
2086 | ||
2087 | ns->pending_mounts = pending + mounts; | |
2088 | return 0; | |
2089 | } | |
2090 | ||
b90fa9ae RP |
2091 | /* |
2092 | * @source_mnt : mount tree to be attached | |
21444403 RP |
2093 | * @nd : place the mount tree @source_mnt is attached |
2094 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
2095 | * store the parent mount and mountpoint dentry. | |
2096 | * (done when source_mnt is moved) | |
b90fa9ae RP |
2097 | * |
2098 | * NOTE: in the table below explains the semantics when a source mount | |
2099 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2100 | * --------------------------------------------------------------------------- |
2101 | * | BIND MOUNT OPERATION | | |
2102 | * |************************************************************************** | |
2103 | * | source-->| shared | private | slave | unbindable | | |
2104 | * | dest | | | | | | |
2105 | * | | | | | | | | |
2106 | * | v | | | | | | |
2107 | * |************************************************************************** | |
2108 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2109 | * | | | | | | | |
2110 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2111 | * *************************************************************************** | |
b90fa9ae RP |
2112 | * A bind operation clones the source mount and mounts the clone on the |
2113 | * destination mount. | |
2114 | * | |
2115 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2116 | * tree of the destination mount and the cloned mount is added to | |
2117 | * the peer group of the source mount. | |
2118 | * (+) the cloned mount is created under the destination mount and is marked | |
2119 | * as shared. The cloned mount is added to the peer group of the source | |
2120 | * mount. | |
5afe0022 RP |
2121 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2122 | * of the destination mount and the cloned mount is made slave | |
2123 | * of the same master as that of the source mount. The cloned mount | |
2124 | * is marked as 'shared and slave'. | |
2125 | * (*) the cloned mount is made a slave of the same master as that of the | |
2126 | * source mount. | |
2127 | * | |
9676f0c6 RP |
2128 | * --------------------------------------------------------------------------- |
2129 | * | MOVE MOUNT OPERATION | | |
2130 | * |************************************************************************** | |
2131 | * | source-->| shared | private | slave | unbindable | | |
2132 | * | dest | | | | | | |
2133 | * | | | | | | | | |
2134 | * | v | | | | | | |
2135 | * |************************************************************************** | |
2136 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2137 | * | | | | | | | |
2138 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2139 | * *************************************************************************** | |
5afe0022 RP |
2140 | * |
2141 | * (+) the mount is moved to the destination. And is then propagated to | |
2142 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2143 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2144 | * (+++) the mount is moved to the destination and is then propagated to |
2145 | * all the mounts belonging to the destination mount's propagation tree. | |
2146 | * the mount is marked as 'shared and slave'. | |
2147 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2148 | * |
2149 | * if the source mount is a tree, the operations explained above is | |
2150 | * applied to each mount in the tree. | |
2151 | * Must be called without spinlocks held, since this function can sleep | |
2152 | * in allocations. | |
2153 | */ | |
0fb54e50 | 2154 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2155 | struct mount *dest_mnt, |
2156 | struct mountpoint *dest_mp, | |
2763d119 | 2157 | bool moving) |
b90fa9ae | 2158 | { |
3bd045cc | 2159 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2160 | HLIST_HEAD(tree_list); |
d2921684 | 2161 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2162 | struct mountpoint *smp; |
315fc83e | 2163 | struct mount *child, *p; |
38129a13 | 2164 | struct hlist_node *n; |
719f5d7f | 2165 | int err; |
b90fa9ae | 2166 | |
1064f874 EB |
2167 | /* Preallocate a mountpoint in case the new mounts need |
2168 | * to be tucked under other mounts. | |
2169 | */ | |
2170 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2171 | if (IS_ERR(smp)) | |
2172 | return PTR_ERR(smp); | |
2173 | ||
d2921684 | 2174 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2175 | if (!moving) { |
d2921684 EB |
2176 | err = count_mounts(ns, source_mnt); |
2177 | if (err) | |
2178 | goto out; | |
2179 | } | |
2180 | ||
fc7be130 | 2181 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2182 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2183 | if (err) |
2184 | goto out; | |
0b1b901b | 2185 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2186 | lock_mount_hash(); |
0b1b901b AV |
2187 | if (err) |
2188 | goto out_cleanup_ids; | |
909b0a88 | 2189 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2190 | set_mnt_shared(p); |
0b1b901b AV |
2191 | } else { |
2192 | lock_mount_hash(); | |
b90fa9ae | 2193 | } |
2763d119 AV |
2194 | if (moving) { |
2195 | unhash_mnt(source_mnt); | |
84d17192 | 2196 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2197 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2198 | } else { |
44dfd84a DH |
2199 | if (source_mnt->mnt_ns) { |
2200 | /* move from anon - the caller will destroy */ | |
2201 | list_del_init(&source_mnt->mnt_ns->list); | |
2202 | } | |
84d17192 | 2203 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2204 | commit_tree(source_mnt); |
21444403 | 2205 | } |
b90fa9ae | 2206 | |
38129a13 | 2207 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2208 | struct mount *q; |
38129a13 | 2209 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2210 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2211 | child->mnt_mountpoint); | |
2212 | if (q) | |
2213 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2214 | /* Notice when we are propagating across user namespaces */ |
2215 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2216 | lock_mnt_tree(child); | |
d728cf79 | 2217 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2218 | commit_tree(child); |
b90fa9ae | 2219 | } |
1064f874 | 2220 | put_mountpoint(smp); |
719ea2fb | 2221 | unlock_mount_hash(); |
99b7db7b | 2222 | |
b90fa9ae | 2223 | return 0; |
719f5d7f MS |
2224 | |
2225 | out_cleanup_ids: | |
f2ebb3a9 AV |
2226 | while (!hlist_empty(&tree_list)) { |
2227 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2228 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2229 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2230 | } |
2231 | unlock_mount_hash(); | |
0b1b901b | 2232 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2233 | out: |
d2921684 | 2234 | ns->pending_mounts = 0; |
1064f874 EB |
2235 | |
2236 | read_seqlock_excl(&mount_lock); | |
2237 | put_mountpoint(smp); | |
2238 | read_sequnlock_excl(&mount_lock); | |
2239 | ||
719f5d7f | 2240 | return err; |
b90fa9ae RP |
2241 | } |
2242 | ||
84d17192 | 2243 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2244 | { |
2245 | struct vfsmount *mnt; | |
84d17192 | 2246 | struct dentry *dentry = path->dentry; |
b12cea91 | 2247 | retry: |
5955102c | 2248 | inode_lock(dentry->d_inode); |
84d17192 | 2249 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2250 | inode_unlock(dentry->d_inode); |
84d17192 | 2251 | return ERR_PTR(-ENOENT); |
b12cea91 | 2252 | } |
97216be0 | 2253 | namespace_lock(); |
b12cea91 | 2254 | mnt = lookup_mnt(path); |
84d17192 | 2255 | if (likely(!mnt)) { |
3895dbf8 | 2256 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2257 | if (IS_ERR(mp)) { |
97216be0 | 2258 | namespace_unlock(); |
5955102c | 2259 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2260 | return mp; |
2261 | } | |
2262 | return mp; | |
2263 | } | |
97216be0 | 2264 | namespace_unlock(); |
5955102c | 2265 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2266 | path_put(path); |
2267 | path->mnt = mnt; | |
84d17192 | 2268 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2269 | goto retry; |
2270 | } | |
2271 | ||
84d17192 | 2272 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2273 | { |
84d17192 | 2274 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2275 | |
2276 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2277 | put_mountpoint(where); |
3895dbf8 EB |
2278 | read_sequnlock_excl(&mount_lock); |
2279 | ||
328e6d90 | 2280 | namespace_unlock(); |
5955102c | 2281 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2282 | } |
2283 | ||
84d17192 | 2284 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2285 | { |
e462ec50 | 2286 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2287 | return -EINVAL; |
2288 | ||
e36cb0b8 DH |
2289 | if (d_is_dir(mp->m_dentry) != |
2290 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2291 | return -ENOTDIR; |
2292 | ||
2763d119 | 2293 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2294 | } |
2295 | ||
7a2e8a8f VA |
2296 | /* |
2297 | * Sanity check the flags to change_mnt_propagation. | |
2298 | */ | |
2299 | ||
e462ec50 | 2300 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2301 | { |
e462ec50 | 2302 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2303 | |
2304 | /* Fail if any non-propagation flags are set */ | |
2305 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2306 | return 0; | |
2307 | /* Only one propagation flag should be set */ | |
2308 | if (!is_power_of_2(type)) | |
2309 | return 0; | |
2310 | return type; | |
2311 | } | |
2312 | ||
07b20889 RP |
2313 | /* |
2314 | * recursively change the type of the mountpoint. | |
2315 | */ | |
e462ec50 | 2316 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2317 | { |
315fc83e | 2318 | struct mount *m; |
4b8b21f4 | 2319 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2320 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2321 | int type; |
719f5d7f | 2322 | int err = 0; |
07b20889 | 2323 | |
2d92ab3c | 2324 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2325 | return -EINVAL; |
2326 | ||
e462ec50 | 2327 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2328 | if (!type) |
2329 | return -EINVAL; | |
2330 | ||
97216be0 | 2331 | namespace_lock(); |
719f5d7f MS |
2332 | if (type == MS_SHARED) { |
2333 | err = invent_group_ids(mnt, recurse); | |
2334 | if (err) | |
2335 | goto out_unlock; | |
2336 | } | |
2337 | ||
719ea2fb | 2338 | lock_mount_hash(); |
909b0a88 | 2339 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2340 | change_mnt_propagation(m, type); |
719ea2fb | 2341 | unlock_mount_hash(); |
719f5d7f MS |
2342 | |
2343 | out_unlock: | |
97216be0 | 2344 | namespace_unlock(); |
719f5d7f | 2345 | return err; |
07b20889 RP |
2346 | } |
2347 | ||
a07b2000 AV |
2348 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2349 | { | |
2350 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2351 | ||
2352 | if (IS_MNT_UNBINDABLE(old)) | |
2353 | return mnt; | |
2354 | ||
2355 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2356 | return mnt; | |
2357 | ||
2358 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2359 | return mnt; | |
2360 | ||
2361 | if (recurse) | |
2362 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2363 | else | |
2364 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2365 | ||
2366 | if (!IS_ERR(mnt)) | |
2367 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2368 | ||
2369 | return mnt; | |
2370 | } | |
2371 | ||
1da177e4 LT |
2372 | /* |
2373 | * do loopback mount. | |
2374 | */ | |
808d4e3c | 2375 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2376 | int recurse) |
1da177e4 | 2377 | { |
2d92ab3c | 2378 | struct path old_path; |
a07b2000 | 2379 | struct mount *mnt = NULL, *parent; |
84d17192 | 2380 | struct mountpoint *mp; |
57eccb83 | 2381 | int err; |
1da177e4 LT |
2382 | if (!old_name || !*old_name) |
2383 | return -EINVAL; | |
815d405c | 2384 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2385 | if (err) |
2386 | return err; | |
2387 | ||
8823c079 | 2388 | err = -EINVAL; |
4ce5d2b1 | 2389 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2390 | goto out; |
8823c079 | 2391 | |
84d17192 | 2392 | mp = lock_mount(path); |
a07b2000 AV |
2393 | if (IS_ERR(mp)) { |
2394 | err = PTR_ERR(mp); | |
b12cea91 | 2395 | goto out; |
a07b2000 | 2396 | } |
b12cea91 | 2397 | |
84d17192 | 2398 | parent = real_mount(path->mnt); |
e149ed2b AV |
2399 | if (!check_mnt(parent)) |
2400 | goto out2; | |
2401 | ||
a07b2000 | 2402 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2403 | if (IS_ERR(mnt)) { |
2404 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2405 | goto out2; |
be34d1a3 | 2406 | } |
ccd48bc7 | 2407 | |
84d17192 | 2408 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2409 | if (err) { |
719ea2fb | 2410 | lock_mount_hash(); |
e819f152 | 2411 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2412 | unlock_mount_hash(); |
5b83d2c5 | 2413 | } |
b12cea91 | 2414 | out2: |
84d17192 | 2415 | unlock_mount(mp); |
ccd48bc7 | 2416 | out: |
2d92ab3c | 2417 | path_put(&old_path); |
1da177e4 LT |
2418 | return err; |
2419 | } | |
2420 | ||
a07b2000 AV |
2421 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2422 | { | |
2423 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2424 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2425 | struct mount *mnt, *p; | |
2426 | struct file *file; | |
2427 | ||
2428 | if (IS_ERR(ns)) | |
2429 | return ERR_CAST(ns); | |
2430 | ||
2431 | namespace_lock(); | |
2432 | mnt = __do_loopback(path, recursive); | |
2433 | if (IS_ERR(mnt)) { | |
2434 | namespace_unlock(); | |
2435 | free_mnt_ns(ns); | |
2436 | return ERR_CAST(mnt); | |
2437 | } | |
2438 | ||
2439 | lock_mount_hash(); | |
2440 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2441 | p->mnt_ns = ns; | |
2442 | ns->mounts++; | |
2443 | } | |
2444 | ns->root = mnt; | |
2445 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2446 | mntget(&mnt->mnt); | |
2447 | unlock_mount_hash(); | |
2448 | namespace_unlock(); | |
2449 | ||
2450 | mntput(path->mnt); | |
2451 | path->mnt = &mnt->mnt; | |
2452 | file = dentry_open(path, O_PATH, current_cred()); | |
2453 | if (IS_ERR(file)) | |
2454 | dissolve_on_fput(path->mnt); | |
2455 | else | |
2456 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2457 | return file; | |
2458 | } | |
2459 | ||
2658ce09 | 2460 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2461 | { |
2462 | struct file *file; | |
2463 | struct path path; | |
2464 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2465 | bool detached = flags & OPEN_TREE_CLONE; | |
2466 | int error; | |
2467 | int fd; | |
2468 | ||
2469 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2470 | ||
2471 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2472 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2473 | OPEN_TREE_CLOEXEC)) | |
2474 | return -EINVAL; | |
2475 | ||
2476 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2477 | return -EINVAL; | |
2478 | ||
2479 | if (flags & AT_NO_AUTOMOUNT) | |
2480 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2481 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2482 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2483 | if (flags & AT_EMPTY_PATH) | |
2484 | lookup_flags |= LOOKUP_EMPTY; | |
2485 | ||
2486 | if (detached && !may_mount()) | |
2487 | return -EPERM; | |
2488 | ||
2489 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2490 | if (fd < 0) | |
2491 | return fd; | |
2492 | ||
2493 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2494 | if (unlikely(error)) { | |
2495 | file = ERR_PTR(error); | |
2496 | } else { | |
2497 | if (detached) | |
2498 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2499 | else | |
2500 | file = dentry_open(&path, O_PATH, current_cred()); | |
2501 | path_put(&path); | |
2502 | } | |
2503 | if (IS_ERR(file)) { | |
2504 | put_unused_fd(fd); | |
2505 | return PTR_ERR(file); | |
2506 | } | |
2507 | fd_install(fd, file); | |
2508 | return fd; | |
2509 | } | |
2510 | ||
43f5e655 DH |
2511 | /* |
2512 | * Don't allow locked mount flags to be cleared. | |
2513 | * | |
2514 | * No locks need to be held here while testing the various MNT_LOCK | |
2515 | * flags because those flags can never be cleared once they are set. | |
2516 | */ | |
2517 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2518 | { |
43f5e655 DH |
2519 | unsigned int fl = mnt->mnt.mnt_flags; |
2520 | ||
2521 | if ((fl & MNT_LOCK_READONLY) && | |
2522 | !(mnt_flags & MNT_READONLY)) | |
2523 | return false; | |
2524 | ||
2525 | if ((fl & MNT_LOCK_NODEV) && | |
2526 | !(mnt_flags & MNT_NODEV)) | |
2527 | return false; | |
2528 | ||
2529 | if ((fl & MNT_LOCK_NOSUID) && | |
2530 | !(mnt_flags & MNT_NOSUID)) | |
2531 | return false; | |
2532 | ||
2533 | if ((fl & MNT_LOCK_NOEXEC) && | |
2534 | !(mnt_flags & MNT_NOEXEC)) | |
2535 | return false; | |
2536 | ||
2537 | if ((fl & MNT_LOCK_ATIME) && | |
2538 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2539 | return false; | |
2e4b7fcd | 2540 | |
43f5e655 DH |
2541 | return true; |
2542 | } | |
2543 | ||
2544 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2545 | { |
43f5e655 | 2546 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2547 | |
43f5e655 | 2548 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2549 | return 0; |
2550 | ||
2551 | if (readonly_request) | |
43f5e655 DH |
2552 | return mnt_make_readonly(mnt); |
2553 | ||
68847c94 CB |
2554 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2555 | return 0; | |
43f5e655 DH |
2556 | } |
2557 | ||
43f5e655 DH |
2558 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2559 | { | |
43f5e655 DH |
2560 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2561 | mnt->mnt.mnt_flags = mnt_flags; | |
2562 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2563 | } |
2564 | ||
f8b92ba6 DD |
2565 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2566 | { | |
2567 | struct super_block *sb = mnt->mnt_sb; | |
2568 | ||
2569 | if (!__mnt_is_readonly(mnt) && | |
2570 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { | |
2571 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2572 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
2573 | struct tm tm; | |
2574 | ||
2575 | time64_to_tm(sb->s_time_max, 0, &tm); | |
2576 | ||
0ecee669 EB |
2577 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
2578 | sb->s_type->name, | |
2579 | is_mounted(mnt) ? "remounted" : "mounted", | |
2580 | mntpath, | |
f8b92ba6 DD |
2581 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
2582 | ||
2583 | free_page((unsigned long)buf); | |
2584 | } | |
2585 | } | |
2586 | ||
43f5e655 DH |
2587 | /* |
2588 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2589 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2590 | * to mount(2). | |
2591 | */ | |
2592 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2593 | { | |
2594 | struct super_block *sb = path->mnt->mnt_sb; | |
2595 | struct mount *mnt = real_mount(path->mnt); | |
2596 | int ret; | |
2597 | ||
2598 | if (!check_mnt(mnt)) | |
2599 | return -EINVAL; | |
2600 | ||
2601 | if (path->dentry != mnt->mnt.mnt_root) | |
2602 | return -EINVAL; | |
2603 | ||
2604 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2605 | return -EPERM; | |
2606 | ||
e58ace1a CB |
2607 | /* |
2608 | * We're only checking whether the superblock is read-only not | |
2609 | * changing it, so only take down_read(&sb->s_umount). | |
2610 | */ | |
2611 | down_read(&sb->s_umount); | |
68847c94 | 2612 | lock_mount_hash(); |
43f5e655 DH |
2613 | ret = change_mount_ro_state(mnt, mnt_flags); |
2614 | if (ret == 0) | |
2615 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2616 | unlock_mount_hash(); |
e58ace1a | 2617 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2618 | |
2619 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2620 | ||
43f5e655 | 2621 | return ret; |
2e4b7fcd DH |
2622 | } |
2623 | ||
1da177e4 LT |
2624 | /* |
2625 | * change filesystem flags. dir should be a physical root of filesystem. | |
2626 | * If you've mounted a non-root directory somewhere and want to do remount | |
2627 | * on it - tough luck. | |
2628 | */ | |
e462ec50 DH |
2629 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2630 | int mnt_flags, void *data) | |
1da177e4 LT |
2631 | { |
2632 | int err; | |
2d92ab3c | 2633 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2634 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2635 | struct fs_context *fc; |
1da177e4 | 2636 | |
143c8c91 | 2637 | if (!check_mnt(mnt)) |
1da177e4 LT |
2638 | return -EINVAL; |
2639 | ||
2d92ab3c | 2640 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2641 | return -EINVAL; |
2642 | ||
43f5e655 | 2643 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2644 | return -EPERM; |
9566d674 | 2645 | |
8d0347f6 DH |
2646 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2647 | if (IS_ERR(fc)) | |
2648 | return PTR_ERR(fc); | |
ff36fe2c | 2649 | |
b330966f | 2650 | fc->oldapi = true; |
8d0347f6 DH |
2651 | err = parse_monolithic_mount_data(fc, data); |
2652 | if (!err) { | |
2653 | down_write(&sb->s_umount); | |
2654 | err = -EPERM; | |
2655 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2656 | err = reconfigure_super(fc); | |
68847c94 CB |
2657 | if (!err) { |
2658 | lock_mount_hash(); | |
8d0347f6 | 2659 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2660 | unlock_mount_hash(); |
2661 | } | |
8d0347f6 DH |
2662 | } |
2663 | up_write(&sb->s_umount); | |
0e55a7cc | 2664 | } |
f8b92ba6 DD |
2665 | |
2666 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2667 | ||
8d0347f6 | 2668 | put_fs_context(fc); |
1da177e4 LT |
2669 | return err; |
2670 | } | |
2671 | ||
cbbe362c | 2672 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2673 | { |
315fc83e | 2674 | struct mount *p; |
909b0a88 | 2675 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2676 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2677 | return 1; |
2678 | } | |
2679 | return 0; | |
2680 | } | |
2681 | ||
44dfd84a DH |
2682 | /* |
2683 | * Check that there aren't references to earlier/same mount namespaces in the | |
2684 | * specified subtree. Such references can act as pins for mount namespaces | |
2685 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2686 | * cycles to be made. | |
2687 | */ | |
2688 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2689 | { | |
2690 | struct mount *p; | |
2691 | bool ret = false; | |
2692 | ||
2693 | lock_mount_hash(); | |
2694 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2695 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2696 | goto out; | |
2697 | ||
2698 | ret = true; | |
2699 | out: | |
2700 | unlock_mount_hash(); | |
2701 | return ret; | |
2702 | } | |
2703 | ||
9ffb14ef PT |
2704 | static int do_set_group(struct path *from_path, struct path *to_path) |
2705 | { | |
2706 | struct mount *from, *to; | |
2707 | int err; | |
2708 | ||
2709 | from = real_mount(from_path->mnt); | |
2710 | to = real_mount(to_path->mnt); | |
2711 | ||
2712 | namespace_lock(); | |
2713 | ||
2714 | err = -EINVAL; | |
2715 | /* To and From must be mounted */ | |
2716 | if (!is_mounted(&from->mnt)) | |
2717 | goto out; | |
2718 | if (!is_mounted(&to->mnt)) | |
2719 | goto out; | |
2720 | ||
2721 | err = -EPERM; | |
2722 | /* We should be allowed to modify mount namespaces of both mounts */ | |
2723 | if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2724 | goto out; | |
2725 | if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2726 | goto out; | |
2727 | ||
2728 | err = -EINVAL; | |
2729 | /* To and From paths should be mount roots */ | |
2730 | if (from_path->dentry != from_path->mnt->mnt_root) | |
2731 | goto out; | |
2732 | if (to_path->dentry != to_path->mnt->mnt_root) | |
2733 | goto out; | |
2734 | ||
2735 | /* Setting sharing groups is only allowed across same superblock */ | |
2736 | if (from->mnt.mnt_sb != to->mnt.mnt_sb) | |
2737 | goto out; | |
2738 | ||
2739 | /* From mount root should be wider than To mount root */ | |
2740 | if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) | |
2741 | goto out; | |
2742 | ||
2743 | /* From mount should not have locked children in place of To's root */ | |
2744 | if (has_locked_children(from, to->mnt.mnt_root)) | |
2745 | goto out; | |
2746 | ||
2747 | /* Setting sharing groups is only allowed on private mounts */ | |
2748 | if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) | |
2749 | goto out; | |
2750 | ||
2751 | /* From should not be private */ | |
2752 | if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) | |
2753 | goto out; | |
2754 | ||
2755 | if (IS_MNT_SLAVE(from)) { | |
2756 | struct mount *m = from->mnt_master; | |
2757 | ||
2758 | list_add(&to->mnt_slave, &m->mnt_slave_list); | |
2759 | to->mnt_master = m; | |
2760 | } | |
2761 | ||
2762 | if (IS_MNT_SHARED(from)) { | |
2763 | to->mnt_group_id = from->mnt_group_id; | |
2764 | list_add(&to->mnt_share, &from->mnt_share); | |
2765 | lock_mount_hash(); | |
2766 | set_mnt_shared(to); | |
2767 | unlock_mount_hash(); | |
2768 | } | |
2769 | ||
2770 | err = 0; | |
2771 | out: | |
2772 | namespace_unlock(); | |
2773 | return err; | |
2774 | } | |
2775 | ||
2db154b3 | 2776 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2777 | { |
44dfd84a | 2778 | struct mnt_namespace *ns; |
676da58d | 2779 | struct mount *p; |
0fb54e50 | 2780 | struct mount *old; |
2763d119 AV |
2781 | struct mount *parent; |
2782 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2783 | int err; |
44dfd84a | 2784 | bool attached; |
1da177e4 | 2785 | |
2db154b3 | 2786 | mp = lock_mount(new_path); |
84d17192 | 2787 | if (IS_ERR(mp)) |
2db154b3 | 2788 | return PTR_ERR(mp); |
cc53ce53 | 2789 | |
2db154b3 DH |
2790 | old = real_mount(old_path->mnt); |
2791 | p = real_mount(new_path->mnt); | |
2763d119 | 2792 | parent = old->mnt_parent; |
44dfd84a | 2793 | attached = mnt_has_parent(old); |
2763d119 | 2794 | old_mp = old->mnt_mp; |
44dfd84a | 2795 | ns = old->mnt_ns; |
143c8c91 | 2796 | |
1da177e4 | 2797 | err = -EINVAL; |
44dfd84a DH |
2798 | /* The mountpoint must be in our namespace. */ |
2799 | if (!check_mnt(p)) | |
2db154b3 | 2800 | goto out; |
1da177e4 | 2801 | |
570d7a98 EB |
2802 | /* The thing moved must be mounted... */ |
2803 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2804 | goto out; |
2805 | ||
570d7a98 EB |
2806 | /* ... and either ours or the root of anon namespace */ |
2807 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2808 | goto out; |
5ff9d8a6 | 2809 | |
2db154b3 DH |
2810 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2811 | goto out; | |
1da177e4 | 2812 | |
2db154b3 DH |
2813 | if (old_path->dentry != old_path->mnt->mnt_root) |
2814 | goto out; | |
1da177e4 | 2815 | |
2db154b3 DH |
2816 | if (d_is_dir(new_path->dentry) != |
2817 | d_is_dir(old_path->dentry)) | |
2818 | goto out; | |
21444403 RP |
2819 | /* |
2820 | * Don't move a mount residing in a shared parent. | |
2821 | */ | |
2763d119 | 2822 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2823 | goto out; |
9676f0c6 RP |
2824 | /* |
2825 | * Don't move a mount tree containing unbindable mounts to a destination | |
2826 | * mount which is shared. | |
2827 | */ | |
fc7be130 | 2828 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2829 | goto out; |
1da177e4 | 2830 | err = -ELOOP; |
44dfd84a DH |
2831 | if (!check_for_nsfs_mounts(old)) |
2832 | goto out; | |
fc7be130 | 2833 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2834 | if (p == old) |
2db154b3 | 2835 | goto out; |
1da177e4 | 2836 | |
2db154b3 | 2837 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2838 | attached); |
4ac91378 | 2839 | if (err) |
2db154b3 | 2840 | goto out; |
1da177e4 LT |
2841 | |
2842 | /* if the mount is moved, it should no longer be expire | |
2843 | * automatically */ | |
6776db3d | 2844 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2845 | if (attached) |
2846 | put_mountpoint(old_mp); | |
1da177e4 | 2847 | out: |
2db154b3 | 2848 | unlock_mount(mp); |
44dfd84a | 2849 | if (!err) { |
2763d119 AV |
2850 | if (attached) |
2851 | mntput_no_expire(parent); | |
2852 | else | |
44dfd84a DH |
2853 | free_mnt_ns(ns); |
2854 | } | |
2db154b3 DH |
2855 | return err; |
2856 | } | |
2857 | ||
2858 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2859 | { | |
2860 | struct path old_path; | |
2861 | int err; | |
2862 | ||
2863 | if (!old_name || !*old_name) | |
2864 | return -EINVAL; | |
2865 | ||
2866 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2867 | if (err) | |
2868 | return err; | |
2869 | ||
2870 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2871 | path_put(&old_path); |
1da177e4 LT |
2872 | return err; |
2873 | } | |
2874 | ||
9d412a43 AV |
2875 | /* |
2876 | * add a mount into a namespace's mount tree | |
2877 | */ | |
8f11538e AV |
2878 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
2879 | struct path *path, int mnt_flags) | |
9d412a43 | 2880 | { |
8f11538e | 2881 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 2882 | |
f2ebb3a9 | 2883 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2884 | |
84d17192 | 2885 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2886 | /* that's acceptable only for automounts done in private ns */ |
2887 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 2888 | return -EINVAL; |
156cacb1 | 2889 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 2890 | if (!parent->mnt_ns) |
8f11538e | 2891 | return -EINVAL; |
156cacb1 | 2892 | } |
9d412a43 AV |
2893 | |
2894 | /* Refuse the same filesystem on the same mount point */ | |
95bc5f25 | 2895 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 | 2896 | path->mnt->mnt_root == path->dentry) |
8f11538e | 2897 | return -EBUSY; |
9d412a43 | 2898 | |
e36cb0b8 | 2899 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 2900 | return -EINVAL; |
9d412a43 | 2901 | |
95bc5f25 | 2902 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 2903 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 2904 | } |
b1e75df4 | 2905 | |
132e4608 DH |
2906 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2907 | ||
2908 | /* | |
2909 | * Create a new mount using a superblock configuration and request it | |
2910 | * be added to the namespace tree. | |
2911 | */ | |
2912 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2913 | unsigned int mnt_flags) | |
2914 | { | |
2915 | struct vfsmount *mnt; | |
8f11538e | 2916 | struct mountpoint *mp; |
132e4608 DH |
2917 | struct super_block *sb = fc->root->d_sb; |
2918 | int error; | |
2919 | ||
c9ce29ed AV |
2920 | error = security_sb_kern_mount(sb); |
2921 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2922 | error = -EPERM; | |
2923 | ||
2924 | if (unlikely(error)) { | |
2925 | fc_drop_locked(fc); | |
2926 | return error; | |
132e4608 DH |
2927 | } |
2928 | ||
2929 | up_write(&sb->s_umount); | |
2930 | ||
2931 | mnt = vfs_create_mount(fc); | |
2932 | if (IS_ERR(mnt)) | |
2933 | return PTR_ERR(mnt); | |
2934 | ||
f8b92ba6 DD |
2935 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
2936 | ||
8f11538e AV |
2937 | mp = lock_mount(mountpoint); |
2938 | if (IS_ERR(mp)) { | |
2939 | mntput(mnt); | |
2940 | return PTR_ERR(mp); | |
2941 | } | |
2942 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
2943 | unlock_mount(mp); | |
0ecee669 EB |
2944 | if (error < 0) |
2945 | mntput(mnt); | |
132e4608 DH |
2946 | return error; |
2947 | } | |
1b852bce | 2948 | |
1da177e4 LT |
2949 | /* |
2950 | * create a new mount for userspace and request it to be added into the | |
2951 | * namespace's tree | |
2952 | */ | |
e462ec50 | 2953 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2954 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2955 | { |
0c55cfc4 | 2956 | struct file_system_type *type; |
a0c9a8b8 AV |
2957 | struct fs_context *fc; |
2958 | const char *subtype = NULL; | |
2959 | int err = 0; | |
1da177e4 | 2960 | |
0c55cfc4 | 2961 | if (!fstype) |
1da177e4 LT |
2962 | return -EINVAL; |
2963 | ||
0c55cfc4 EB |
2964 | type = get_fs_type(fstype); |
2965 | if (!type) | |
2966 | return -ENODEV; | |
2967 | ||
a0c9a8b8 AV |
2968 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2969 | subtype = strchr(fstype, '.'); | |
2970 | if (subtype) { | |
2971 | subtype++; | |
2972 | if (!*subtype) { | |
2973 | put_filesystem(type); | |
2974 | return -EINVAL; | |
2975 | } | |
a0c9a8b8 AV |
2976 | } |
2977 | } | |
0c55cfc4 | 2978 | |
a0c9a8b8 | 2979 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2980 | put_filesystem(type); |
a0c9a8b8 AV |
2981 | if (IS_ERR(fc)) |
2982 | return PTR_ERR(fc); | |
2983 | ||
3e1aeb00 DH |
2984 | if (subtype) |
2985 | err = vfs_parse_fs_string(fc, "subtype", | |
2986 | subtype, strlen(subtype)); | |
2987 | if (!err && name) | |
2988 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2989 | if (!err) |
2990 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
2991 | if (!err && !mount_capable(fc)) |
2992 | err = -EPERM; | |
a0c9a8b8 AV |
2993 | if (!err) |
2994 | err = vfs_get_tree(fc); | |
132e4608 DH |
2995 | if (!err) |
2996 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2997 | |
a0c9a8b8 | 2998 | put_fs_context(fc); |
15f9a3f3 | 2999 | return err; |
1da177e4 LT |
3000 | } |
3001 | ||
19a167af AV |
3002 | int finish_automount(struct vfsmount *m, struct path *path) |
3003 | { | |
26df6034 | 3004 | struct dentry *dentry = path->dentry; |
8f11538e | 3005 | struct mountpoint *mp; |
25e195aa | 3006 | struct mount *mnt; |
19a167af | 3007 | int err; |
25e195aa AV |
3008 | |
3009 | if (!m) | |
3010 | return 0; | |
3011 | if (IS_ERR(m)) | |
3012 | return PTR_ERR(m); | |
3013 | ||
3014 | mnt = real_mount(m); | |
19a167af AV |
3015 | /* The new mount record should have at least 2 refs to prevent it being |
3016 | * expired before we get a chance to add it | |
3017 | */ | |
6776db3d | 3018 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
3019 | |
3020 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 3021 | m->mnt_root == dentry) { |
b1e75df4 | 3022 | err = -ELOOP; |
26df6034 | 3023 | goto discard; |
19a167af AV |
3024 | } |
3025 | ||
26df6034 AV |
3026 | /* |
3027 | * we don't want to use lock_mount() - in this case finding something | |
3028 | * that overmounts our mountpoint to be means "quitely drop what we've | |
3029 | * got", not "try to mount it on top". | |
3030 | */ | |
3031 | inode_lock(dentry->d_inode); | |
3032 | namespace_lock(); | |
3033 | if (unlikely(cant_mount(dentry))) { | |
3034 | err = -ENOENT; | |
3035 | goto discard_locked; | |
3036 | } | |
3037 | rcu_read_lock(); | |
3038 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { | |
3039 | rcu_read_unlock(); | |
3040 | err = 0; | |
3041 | goto discard_locked; | |
3042 | } | |
3043 | rcu_read_unlock(); | |
3044 | mp = get_mountpoint(dentry); | |
8f11538e AV |
3045 | if (IS_ERR(mp)) { |
3046 | err = PTR_ERR(mp); | |
26df6034 | 3047 | goto discard_locked; |
8f11538e | 3048 | } |
26df6034 | 3049 | |
8f11538e AV |
3050 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
3051 | unlock_mount(mp); | |
26df6034 AV |
3052 | if (unlikely(err)) |
3053 | goto discard; | |
3054 | mntput(m); | |
3055 | return 0; | |
3056 | ||
3057 | discard_locked: | |
3058 | namespace_unlock(); | |
3059 | inode_unlock(dentry->d_inode); | |
3060 | discard: | |
b1e75df4 | 3061 | /* remove m from any expiration list it may be on */ |
6776db3d | 3062 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 3063 | namespace_lock(); |
6776db3d | 3064 | list_del_init(&mnt->mnt_expire); |
97216be0 | 3065 | namespace_unlock(); |
19a167af | 3066 | } |
b1e75df4 AV |
3067 | mntput(m); |
3068 | mntput(m); | |
19a167af AV |
3069 | return err; |
3070 | } | |
3071 | ||
ea5b778a DH |
3072 | /** |
3073 | * mnt_set_expiry - Put a mount on an expiration list | |
3074 | * @mnt: The mount to list. | |
3075 | * @expiry_list: The list to add the mount to. | |
3076 | */ | |
3077 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
3078 | { | |
97216be0 | 3079 | namespace_lock(); |
ea5b778a | 3080 | |
6776db3d | 3081 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 3082 | |
97216be0 | 3083 | namespace_unlock(); |
ea5b778a DH |
3084 | } |
3085 | EXPORT_SYMBOL(mnt_set_expiry); | |
3086 | ||
1da177e4 LT |
3087 | /* |
3088 | * process a list of expirable mountpoints with the intent of discarding any | |
3089 | * mountpoints that aren't in use and haven't been touched since last we came | |
3090 | * here | |
3091 | */ | |
3092 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3093 | { | |
761d5c38 | 3094 | struct mount *mnt, *next; |
1da177e4 LT |
3095 | LIST_HEAD(graveyard); |
3096 | ||
3097 | if (list_empty(mounts)) | |
3098 | return; | |
3099 | ||
97216be0 | 3100 | namespace_lock(); |
719ea2fb | 3101 | lock_mount_hash(); |
1da177e4 LT |
3102 | |
3103 | /* extract from the expiration list every vfsmount that matches the | |
3104 | * following criteria: | |
3105 | * - only referenced by its parent vfsmount | |
3106 | * - still marked for expiry (marked on the last call here; marks are | |
3107 | * cleared by mntput()) | |
3108 | */ | |
6776db3d | 3109 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3110 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3111 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3112 | continue; |
6776db3d | 3113 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3114 | } |
bcc5c7d2 | 3115 | while (!list_empty(&graveyard)) { |
6776db3d | 3116 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3117 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3118 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3119 | } |
719ea2fb | 3120 | unlock_mount_hash(); |
3ab6abee | 3121 | namespace_unlock(); |
5528f911 TM |
3122 | } |
3123 | ||
3124 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3125 | ||
3126 | /* | |
3127 | * Ripoff of 'select_parent()' | |
3128 | * | |
3129 | * search the list of submounts for a given mountpoint, and move any | |
3130 | * shrinkable submounts to the 'graveyard' list. | |
3131 | */ | |
692afc31 | 3132 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3133 | { |
692afc31 | 3134 | struct mount *this_parent = parent; |
5528f911 TM |
3135 | struct list_head *next; |
3136 | int found = 0; | |
3137 | ||
3138 | repeat: | |
6b41d536 | 3139 | next = this_parent->mnt_mounts.next; |
5528f911 | 3140 | resume: |
6b41d536 | 3141 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3142 | struct list_head *tmp = next; |
6b41d536 | 3143 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3144 | |
3145 | next = tmp->next; | |
692afc31 | 3146 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3147 | continue; |
5528f911 TM |
3148 | /* |
3149 | * Descend a level if the d_mounts list is non-empty. | |
3150 | */ | |
6b41d536 | 3151 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3152 | this_parent = mnt; |
3153 | goto repeat; | |
3154 | } | |
1da177e4 | 3155 | |
1ab59738 | 3156 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3157 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3158 | found++; |
3159 | } | |
1da177e4 | 3160 | } |
5528f911 TM |
3161 | /* |
3162 | * All done at this level ... ascend and resume the search | |
3163 | */ | |
3164 | if (this_parent != parent) { | |
6b41d536 | 3165 | next = this_parent->mnt_child.next; |
0714a533 | 3166 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3167 | goto resume; |
3168 | } | |
3169 | return found; | |
3170 | } | |
3171 | ||
3172 | /* | |
3173 | * process a list of expirable mountpoints with the intent of discarding any | |
3174 | * submounts of a specific parent mountpoint | |
99b7db7b | 3175 | * |
48a066e7 | 3176 | * mount_lock must be held for write |
5528f911 | 3177 | */ |
b54b9be7 | 3178 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3179 | { |
3180 | LIST_HEAD(graveyard); | |
761d5c38 | 3181 | struct mount *m; |
5528f911 | 3182 | |
5528f911 | 3183 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3184 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3185 | while (!list_empty(&graveyard)) { |
761d5c38 | 3186 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3187 | mnt_expire); |
143c8c91 | 3188 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3189 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3190 | } |
3191 | } | |
1da177e4 LT |
3192 | } |
3193 | ||
028abd92 | 3194 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3195 | { |
b40ef869 | 3196 | char *copy; |
d563d678 | 3197 | unsigned left, offset; |
b58fed8b | 3198 | |
1da177e4 | 3199 | if (!data) |
b40ef869 | 3200 | return NULL; |
1da177e4 | 3201 | |
b40ef869 AV |
3202 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3203 | if (!copy) | |
3204 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3205 | |
d563d678 | 3206 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3207 | |
d563d678 CM |
3208 | /* |
3209 | * Not all architectures have an exact copy_from_user(). Resort to | |
3210 | * byte at a time. | |
3211 | */ | |
3212 | offset = PAGE_SIZE - left; | |
3213 | while (left) { | |
3214 | char c; | |
3215 | if (get_user(c, (const char __user *)data + offset)) | |
3216 | break; | |
3217 | copy[offset] = c; | |
3218 | left--; | |
3219 | offset++; | |
3220 | } | |
3221 | ||
3222 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3223 | kfree(copy); |
3224 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3225 | } |
d563d678 | 3226 | |
b40ef869 | 3227 | return copy; |
1da177e4 LT |
3228 | } |
3229 | ||
028abd92 | 3230 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3231 | { |
fbdb4401 | 3232 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3233 | } |
3234 | ||
1da177e4 LT |
3235 | /* |
3236 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3237 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3238 | * | |
3239 | * data is a (void *) that can point to any structure up to | |
3240 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3241 | * information (or be NULL). | |
3242 | * | |
3243 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3244 | * When the flags word was introduced its top half was required | |
3245 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3246 | * Therefore, if this magic number is present, it carries no information | |
3247 | * and must be discarded. | |
3248 | */ | |
c60166f0 | 3249 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3250 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3251 | { |
e462ec50 | 3252 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3253 | int ret; |
1da177e4 LT |
3254 | |
3255 | /* Discard magic */ | |
3256 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3257 | flags &= ~MS_MGC_MSK; | |
3258 | ||
3259 | /* Basic sanity checks */ | |
1da177e4 LT |
3260 | if (data_page) |
3261 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3262 | ||
e462ec50 DH |
3263 | if (flags & MS_NOUSER) |
3264 | return -EINVAL; | |
3265 | ||
a1e6aaa3 CH |
3266 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3267 | if (ret) | |
3268 | return ret; | |
3269 | if (!may_mount()) | |
3270 | return -EPERM; | |
f7e33bdb JL |
3271 | if (flags & SB_MANDLOCK) |
3272 | warn_mandlock(); | |
a27ab9f2 | 3273 | |
613cbe3d AK |
3274 | /* Default to relatime unless overriden */ |
3275 | if (!(flags & MS_NOATIME)) | |
3276 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3277 | |
1da177e4 LT |
3278 | /* Separate the per-mountpoint flags */ |
3279 | if (flags & MS_NOSUID) | |
3280 | mnt_flags |= MNT_NOSUID; | |
3281 | if (flags & MS_NODEV) | |
3282 | mnt_flags |= MNT_NODEV; | |
3283 | if (flags & MS_NOEXEC) | |
3284 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3285 | if (flags & MS_NOATIME) |
3286 | mnt_flags |= MNT_NOATIME; | |
3287 | if (flags & MS_NODIRATIME) | |
3288 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3289 | if (flags & MS_STRICTATIME) |
3290 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3291 | if (flags & MS_RDONLY) |
2e4b7fcd | 3292 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3293 | if (flags & MS_NOSYMFOLLOW) |
3294 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3295 | |
ffbc6f0e EB |
3296 | /* The default atime for remount is preservation */ |
3297 | if ((flags & MS_REMOUNT) && | |
3298 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3299 | MS_STRICTATIME)) == 0)) { | |
3300 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3301 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3302 | } |
3303 | ||
e462ec50 DH |
3304 | sb_flags = flags & (SB_RDONLY | |
3305 | SB_SYNCHRONOUS | | |
3306 | SB_MANDLOCK | | |
3307 | SB_DIRSYNC | | |
3308 | SB_SILENT | | |
917086ff | 3309 | SB_POSIXACL | |
d7ee9469 | 3310 | SB_LAZYTIME | |
917086ff | 3311 | SB_I_VERSION); |
1da177e4 | 3312 | |
43f5e655 | 3313 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3314 | return do_reconfigure_mnt(path, mnt_flags); |
3315 | if (flags & MS_REMOUNT) | |
3316 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3317 | if (flags & MS_BIND) | |
3318 | return do_loopback(path, dev_name, flags & MS_REC); | |
3319 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3320 | return do_change_type(path, flags); | |
3321 | if (flags & MS_MOVE) | |
3322 | return do_move_mount_old(path, dev_name); | |
3323 | ||
3324 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3325 | data_page); | |
3326 | } | |
3327 | ||
3328 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3329 | const char *type_page, unsigned long flags, void *data_page) | |
3330 | { | |
3331 | struct path path; | |
3332 | int ret; | |
3333 | ||
3334 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3335 | if (ret) | |
3336 | return ret; | |
3337 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3338 | path_put(&path); |
a1e6aaa3 | 3339 | return ret; |
1da177e4 LT |
3340 | } |
3341 | ||
537f7ccb EB |
3342 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3343 | { | |
3344 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3345 | } | |
3346 | ||
3347 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3348 | { | |
3349 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3350 | } | |
3351 | ||
771b1371 EB |
3352 | static void free_mnt_ns(struct mnt_namespace *ns) |
3353 | { | |
74e83122 AV |
3354 | if (!is_anon_ns(ns)) |
3355 | ns_free_inum(&ns->ns); | |
537f7ccb | 3356 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3357 | put_user_ns(ns->user_ns); |
3358 | kfree(ns); | |
3359 | } | |
3360 | ||
8823c079 EB |
3361 | /* |
3362 | * Assign a sequence number so we can detect when we attempt to bind | |
3363 | * mount a reference to an older mount namespace into the current | |
3364 | * mount namespace, preventing reference counting loops. A 64bit | |
3365 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3366 | * is effectively never, so we can ignore the possibility. | |
3367 | */ | |
3368 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3369 | ||
74e83122 | 3370 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3371 | { |
3372 | struct mnt_namespace *new_ns; | |
537f7ccb | 3373 | struct ucounts *ucounts; |
98f842e6 | 3374 | int ret; |
cf8d2c11 | 3375 | |
537f7ccb EB |
3376 | ucounts = inc_mnt_namespaces(user_ns); |
3377 | if (!ucounts) | |
df75e774 | 3378 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3379 | |
30acd0bd | 3380 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); |
537f7ccb EB |
3381 | if (!new_ns) { |
3382 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3383 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3384 | } |
74e83122 AV |
3385 | if (!anon) { |
3386 | ret = ns_alloc_inum(&new_ns->ns); | |
3387 | if (ret) { | |
3388 | kfree(new_ns); | |
3389 | dec_mnt_namespaces(ucounts); | |
3390 | return ERR_PTR(ret); | |
3391 | } | |
98f842e6 | 3392 | } |
33c42940 | 3393 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3394 | if (!anon) |
3395 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3396 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3397 | INIT_LIST_HEAD(&new_ns->list); |
3398 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3399 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3400 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3401 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3402 | return new_ns; |
3403 | } | |
3404 | ||
0766f788 | 3405 | __latent_entropy |
9559f689 AV |
3406 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3407 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3408 | { |
6b3286ed | 3409 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3410 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3411 | struct mount *p, *q; |
9559f689 | 3412 | struct mount *old; |
cb338d06 | 3413 | struct mount *new; |
7a472ef4 | 3414 | int copy_flags; |
1da177e4 | 3415 | |
9559f689 AV |
3416 | BUG_ON(!ns); |
3417 | ||
3418 | if (likely(!(flags & CLONE_NEWNS))) { | |
3419 | get_mnt_ns(ns); | |
3420 | return ns; | |
3421 | } | |
3422 | ||
3423 | old = ns->root; | |
3424 | ||
74e83122 | 3425 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3426 | if (IS_ERR(new_ns)) |
3427 | return new_ns; | |
1da177e4 | 3428 | |
97216be0 | 3429 | namespace_lock(); |
1da177e4 | 3430 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3431 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3432 | if (user_ns != ns->user_ns) |
3bd045cc | 3433 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3434 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3435 | if (IS_ERR(new)) { |
328e6d90 | 3436 | namespace_unlock(); |
771b1371 | 3437 | free_mnt_ns(new_ns); |
be34d1a3 | 3438 | return ERR_CAST(new); |
1da177e4 | 3439 | } |
3bd045cc AV |
3440 | if (user_ns != ns->user_ns) { |
3441 | lock_mount_hash(); | |
3442 | lock_mnt_tree(new); | |
3443 | unlock_mount_hash(); | |
3444 | } | |
be08d6d2 | 3445 | new_ns->root = new; |
1a4eeaf2 | 3446 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3447 | |
3448 | /* | |
3449 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3450 | * as belonging to new namespace. We have already acquired a private | |
3451 | * fs_struct, so tsk->fs->lock is not needed. | |
3452 | */ | |
909b0a88 | 3453 | p = old; |
cb338d06 | 3454 | q = new; |
1da177e4 | 3455 | while (p) { |
143c8c91 | 3456 | q->mnt_ns = new_ns; |
d2921684 | 3457 | new_ns->mounts++; |
9559f689 AV |
3458 | if (new_fs) { |
3459 | if (&p->mnt == new_fs->root.mnt) { | |
3460 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3461 | rootmnt = &p->mnt; |
1da177e4 | 3462 | } |
9559f689 AV |
3463 | if (&p->mnt == new_fs->pwd.mnt) { |
3464 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3465 | pwdmnt = &p->mnt; |
1da177e4 | 3466 | } |
1da177e4 | 3467 | } |
909b0a88 AV |
3468 | p = next_mnt(p, old); |
3469 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3470 | if (!q) |
3471 | break; | |
3472 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3473 | p = next_mnt(p, old); | |
1da177e4 | 3474 | } |
328e6d90 | 3475 | namespace_unlock(); |
1da177e4 | 3476 | |
1da177e4 | 3477 | if (rootmnt) |
f03c6599 | 3478 | mntput(rootmnt); |
1da177e4 | 3479 | if (pwdmnt) |
f03c6599 | 3480 | mntput(pwdmnt); |
1da177e4 | 3481 | |
741a2951 | 3482 | return new_ns; |
1da177e4 LT |
3483 | } |
3484 | ||
74e83122 | 3485 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3486 | { |
74e83122 | 3487 | struct mount *mnt = real_mount(m); |
ea441d11 | 3488 | struct mnt_namespace *ns; |
d31da0f0 | 3489 | struct super_block *s; |
ea441d11 AV |
3490 | struct path path; |
3491 | int err; | |
3492 | ||
74e83122 AV |
3493 | ns = alloc_mnt_ns(&init_user_ns, true); |
3494 | if (IS_ERR(ns)) { | |
3495 | mntput(m); | |
ea441d11 | 3496 | return ERR_CAST(ns); |
74e83122 AV |
3497 | } |
3498 | mnt->mnt_ns = ns; | |
3499 | ns->root = mnt; | |
3500 | ns->mounts++; | |
3501 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3502 | |
74e83122 | 3503 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3504 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3505 | ||
3506 | put_mnt_ns(ns); | |
3507 | ||
3508 | if (err) | |
3509 | return ERR_PTR(err); | |
3510 | ||
3511 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3512 | s = path.mnt->mnt_sb; |
3513 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3514 | mntput(path.mnt); |
3515 | /* lock the sucker */ | |
d31da0f0 | 3516 | down_write(&s->s_umount); |
ea441d11 AV |
3517 | /* ... and return the root of (sub)tree on it */ |
3518 | return path.dentry; | |
3519 | } | |
3520 | EXPORT_SYMBOL(mount_subtree); | |
3521 | ||
cccaa5e3 DB |
3522 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3523 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3524 | { |
eca6f534 VN |
3525 | int ret; |
3526 | char *kernel_type; | |
eca6f534 | 3527 | char *kernel_dev; |
b40ef869 | 3528 | void *options; |
1da177e4 | 3529 | |
b8850d1f TG |
3530 | kernel_type = copy_mount_string(type); |
3531 | ret = PTR_ERR(kernel_type); | |
3532 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3533 | goto out_type; |
1da177e4 | 3534 | |
b8850d1f TG |
3535 | kernel_dev = copy_mount_string(dev_name); |
3536 | ret = PTR_ERR(kernel_dev); | |
3537 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3538 | goto out_dev; |
1da177e4 | 3539 | |
b40ef869 AV |
3540 | options = copy_mount_options(data); |
3541 | ret = PTR_ERR(options); | |
3542 | if (IS_ERR(options)) | |
eca6f534 | 3543 | goto out_data; |
1da177e4 | 3544 | |
b40ef869 | 3545 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3546 | |
b40ef869 | 3547 | kfree(options); |
eca6f534 VN |
3548 | out_data: |
3549 | kfree(kernel_dev); | |
3550 | out_dev: | |
eca6f534 VN |
3551 | kfree(kernel_type); |
3552 | out_type: | |
3553 | return ret; | |
1da177e4 LT |
3554 | } |
3555 | ||
dd8b477f CB |
3556 | #define FSMOUNT_VALID_FLAGS \ |
3557 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3558 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ | |
3559 | MOUNT_ATTR_NOSYMFOLLOW) | |
5b490500 | 3560 | |
9caccd41 | 3561 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
2a186721 CB |
3562 | |
3563 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3564 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3565 | ||
5b490500 CB |
3566 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3567 | { | |
3568 | unsigned int mnt_flags = 0; | |
3569 | ||
3570 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3571 | mnt_flags |= MNT_READONLY; | |
3572 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3573 | mnt_flags |= MNT_NOSUID; | |
3574 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3575 | mnt_flags |= MNT_NODEV; | |
3576 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3577 | mnt_flags |= MNT_NOEXEC; | |
3578 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3579 | mnt_flags |= MNT_NODIRATIME; | |
dd8b477f CB |
3580 | if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) |
3581 | mnt_flags |= MNT_NOSYMFOLLOW; | |
5b490500 CB |
3582 | |
3583 | return mnt_flags; | |
3584 | } | |
3585 | ||
2db154b3 | 3586 | /* |
93766fbd DH |
3587 | * Create a kernel mount representation for a new, prepared superblock |
3588 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3589 | */ | |
3590 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3591 | unsigned int, attr_flags) | |
3592 | { | |
3593 | struct mnt_namespace *ns; | |
3594 | struct fs_context *fc; | |
3595 | struct file *file; | |
3596 | struct path newmount; | |
3597 | struct mount *mnt; | |
3598 | struct fd f; | |
3599 | unsigned int mnt_flags = 0; | |
3600 | long ret; | |
3601 | ||
3602 | if (!may_mount()) | |
3603 | return -EPERM; | |
3604 | ||
3605 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3606 | return -EINVAL; | |
3607 | ||
5b490500 | 3608 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3609 | return -EINVAL; |
3610 | ||
5b490500 | 3611 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3612 | |
3613 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3614 | case MOUNT_ATTR_STRICTATIME: | |
3615 | break; | |
3616 | case MOUNT_ATTR_NOATIME: | |
3617 | mnt_flags |= MNT_NOATIME; | |
3618 | break; | |
3619 | case MOUNT_ATTR_RELATIME: | |
3620 | mnt_flags |= MNT_RELATIME; | |
3621 | break; | |
3622 | default: | |
3623 | return -EINVAL; | |
3624 | } | |
3625 | ||
3626 | f = fdget(fs_fd); | |
3627 | if (!f.file) | |
3628 | return -EBADF; | |
3629 | ||
3630 | ret = -EINVAL; | |
3631 | if (f.file->f_op != &fscontext_fops) | |
3632 | goto err_fsfd; | |
3633 | ||
3634 | fc = f.file->private_data; | |
3635 | ||
3636 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3637 | if (ret < 0) | |
3638 | goto err_fsfd; | |
3639 | ||
3640 | /* There must be a valid superblock or we can't mount it */ | |
3641 | ret = -EINVAL; | |
3642 | if (!fc->root) | |
3643 | goto err_unlock; | |
3644 | ||
3645 | ret = -EPERM; | |
3646 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3647 | pr_warn("VFS: Mount too revealing\n"); | |
3648 | goto err_unlock; | |
3649 | } | |
3650 | ||
3651 | ret = -EBUSY; | |
3652 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3653 | goto err_unlock; | |
3654 | ||
f7e33bdb JL |
3655 | if (fc->sb_flags & SB_MANDLOCK) |
3656 | warn_mandlock(); | |
93766fbd DH |
3657 | |
3658 | newmount.mnt = vfs_create_mount(fc); | |
3659 | if (IS_ERR(newmount.mnt)) { | |
3660 | ret = PTR_ERR(newmount.mnt); | |
3661 | goto err_unlock; | |
3662 | } | |
3663 | newmount.dentry = dget(fc->root); | |
3664 | newmount.mnt->mnt_flags = mnt_flags; | |
3665 | ||
3666 | /* We've done the mount bit - now move the file context into more or | |
3667 | * less the same state as if we'd done an fspick(). We don't want to | |
3668 | * do any memory allocation or anything like that at this point as we | |
3669 | * don't want to have to handle any errors incurred. | |
3670 | */ | |
3671 | vfs_clean_context(fc); | |
3672 | ||
3673 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3674 | if (IS_ERR(ns)) { | |
3675 | ret = PTR_ERR(ns); | |
3676 | goto err_path; | |
3677 | } | |
3678 | mnt = real_mount(newmount.mnt); | |
3679 | mnt->mnt_ns = ns; | |
3680 | ns->root = mnt; | |
3681 | ns->mounts = 1; | |
3682 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3683 | mntget(newmount.mnt); |
93766fbd DH |
3684 | |
3685 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3686 | * it, not just simply put it. | |
3687 | */ | |
3688 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3689 | if (IS_ERR(file)) { | |
3690 | dissolve_on_fput(newmount.mnt); | |
3691 | ret = PTR_ERR(file); | |
3692 | goto err_path; | |
3693 | } | |
3694 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3695 | ||
3696 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3697 | if (ret >= 0) | |
3698 | fd_install(ret, file); | |
3699 | else | |
3700 | fput(file); | |
3701 | ||
3702 | err_path: | |
3703 | path_put(&newmount); | |
3704 | err_unlock: | |
3705 | mutex_unlock(&fc->uapi_mutex); | |
3706 | err_fsfd: | |
3707 | fdput(f); | |
3708 | return ret; | |
3709 | } | |
3710 | ||
3711 | /* | |
3712 | * Move a mount from one place to another. In combination with | |
3713 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3714 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3715 | * a mount subtree. | |
2db154b3 DH |
3716 | * |
3717 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3718 | */ | |
3719 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
3720 | int, from_dfd, const char __user *, from_pathname, |
3721 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
3722 | unsigned int, flags) |
3723 | { | |
3724 | struct path from_path, to_path; | |
3725 | unsigned int lflags; | |
3726 | int ret = 0; | |
3727 | ||
3728 | if (!may_mount()) | |
3729 | return -EPERM; | |
3730 | ||
3731 | if (flags & ~MOVE_MOUNT__MASK) | |
3732 | return -EINVAL; | |
3733 | ||
3734 | /* If someone gives a pathname, they aren't permitted to move | |
3735 | * from an fd that requires unmount as we can't get at the flag | |
3736 | * to clear it afterwards. | |
3737 | */ | |
3738 | lflags = 0; | |
3739 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3740 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3741 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3742 | ||
3743 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3744 | if (ret < 0) | |
3745 | return ret; | |
3746 | ||
3747 | lflags = 0; | |
3748 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3749 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3750 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3751 | ||
3752 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3753 | if (ret < 0) | |
3754 | goto out_from; | |
3755 | ||
3756 | ret = security_move_mount(&from_path, &to_path); | |
3757 | if (ret < 0) | |
3758 | goto out_to; | |
3759 | ||
9ffb14ef PT |
3760 | if (flags & MOVE_MOUNT_SET_GROUP) |
3761 | ret = do_set_group(&from_path, &to_path); | |
3762 | else | |
3763 | ret = do_move_mount(&from_path, &to_path); | |
2db154b3 DH |
3764 | |
3765 | out_to: | |
3766 | path_put(&to_path); | |
3767 | out_from: | |
3768 | path_put(&from_path); | |
3769 | return ret; | |
3770 | } | |
3771 | ||
afac7cba AV |
3772 | /* |
3773 | * Return true if path is reachable from root | |
3774 | * | |
48a066e7 | 3775 | * namespace_sem or mount_lock is held |
afac7cba | 3776 | */ |
643822b4 | 3777 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3778 | const struct path *root) |
3779 | { | |
643822b4 | 3780 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3781 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3782 | mnt = mnt->mnt_parent; |
afac7cba | 3783 | } |
643822b4 | 3784 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3785 | } |
3786 | ||
640eb7e7 | 3787 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3788 | { |
25ab4c9b | 3789 | bool res; |
48a066e7 | 3790 | read_seqlock_excl(&mount_lock); |
643822b4 | 3791 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3792 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3793 | return res; |
3794 | } | |
3795 | EXPORT_SYMBOL(path_is_under); | |
3796 | ||
1da177e4 LT |
3797 | /* |
3798 | * pivot_root Semantics: | |
3799 | * Moves the root file system of the current process to the directory put_old, | |
3800 | * makes new_root as the new root file system of the current process, and sets | |
3801 | * root/cwd of all processes which had them on the current root to new_root. | |
3802 | * | |
3803 | * Restrictions: | |
3804 | * The new_root and put_old must be directories, and must not be on the | |
3805 | * same file system as the current process root. The put_old must be | |
3806 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3807 | * pointed to by put_old must yield the same directory as new_root. No other | |
3808 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3809 | * | |
4a0d11fa | 3810 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 3811 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
3812 | * in this situation. |
3813 | * | |
1da177e4 LT |
3814 | * Notes: |
3815 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3816 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3817 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3818 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3819 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3820 | * first. | |
3821 | */ | |
3480b257 HC |
3822 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3823 | const char __user *, put_old) | |
1da177e4 | 3824 | { |
2763d119 AV |
3825 | struct path new, old, root; |
3826 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3827 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3828 | int error; |
3829 | ||
9b40bc90 | 3830 | if (!may_mount()) |
1da177e4 LT |
3831 | return -EPERM; |
3832 | ||
ce6595a2 AV |
3833 | error = user_path_at(AT_FDCWD, new_root, |
3834 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
3835 | if (error) |
3836 | goto out0; | |
1da177e4 | 3837 | |
ce6595a2 AV |
3838 | error = user_path_at(AT_FDCWD, put_old, |
3839 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
3840 | if (error) |
3841 | goto out1; | |
3842 | ||
2d8f3038 | 3843 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3844 | if (error) |
3845 | goto out2; | |
1da177e4 | 3846 | |
f7ad3c6b | 3847 | get_fs_root(current->fs, &root); |
84d17192 AV |
3848 | old_mp = lock_mount(&old); |
3849 | error = PTR_ERR(old_mp); | |
3850 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3851 | goto out3; |
3852 | ||
1da177e4 | 3853 | error = -EINVAL; |
419148da AV |
3854 | new_mnt = real_mount(new.mnt); |
3855 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3856 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3857 | ex_parent = new_mnt->mnt_parent; |
3858 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3859 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3860 | IS_MNT_SHARED(ex_parent) || |
3861 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3862 | goto out4; |
143c8c91 | 3863 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3864 | goto out4; |
5ff9d8a6 EB |
3865 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3866 | goto out4; | |
1da177e4 | 3867 | error = -ENOENT; |
f3da392e | 3868 | if (d_unlinked(new.dentry)) |
b12cea91 | 3869 | goto out4; |
1da177e4 | 3870 | error = -EBUSY; |
84d17192 | 3871 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3872 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3873 | error = -EINVAL; |
8c3ee42e | 3874 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3875 | goto out4; /* not a mountpoint */ |
676da58d | 3876 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3877 | goto out4; /* not attached */ |
2d8f3038 | 3878 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3879 | goto out4; /* not a mountpoint */ |
676da58d | 3880 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3881 | goto out4; /* not attached */ |
4ac91378 | 3882 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3883 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3884 | goto out4; |
0d082601 EB |
3885 | /* make certain new is below the root */ |
3886 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3887 | goto out4; | |
719ea2fb | 3888 | lock_mount_hash(); |
2763d119 AV |
3889 | umount_mnt(new_mnt); |
3890 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3891 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3892 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3893 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3894 | } | |
4ac91378 | 3895 | /* mount old root on put_old */ |
84d17192 | 3896 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3897 | /* mount new_root on / */ |
2763d119 AV |
3898 | attach_mnt(new_mnt, root_parent, root_mp); |
3899 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3900 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3901 | /* A moved mount should not expire automatically */ |
3902 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3903 | put_mountpoint(root_mp); |
719ea2fb | 3904 | unlock_mount_hash(); |
2d8f3038 | 3905 | chroot_fs_refs(&root, &new); |
1da177e4 | 3906 | error = 0; |
b12cea91 | 3907 | out4: |
84d17192 | 3908 | unlock_mount(old_mp); |
2763d119 AV |
3909 | if (!error) |
3910 | mntput_no_expire(ex_parent); | |
b12cea91 | 3911 | out3: |
8c3ee42e | 3912 | path_put(&root); |
b12cea91 | 3913 | out2: |
2d8f3038 | 3914 | path_put(&old); |
1da177e4 | 3915 | out1: |
2d8f3038 | 3916 | path_put(&new); |
1da177e4 | 3917 | out0: |
1da177e4 | 3918 | return error; |
1da177e4 LT |
3919 | } |
3920 | ||
2a186721 CB |
3921 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
3922 | { | |
3923 | unsigned int flags = mnt->mnt.mnt_flags; | |
3924 | ||
3925 | /* flags to clear */ | |
3926 | flags &= ~kattr->attr_clr; | |
3927 | /* flags to raise */ | |
3928 | flags |= kattr->attr_set; | |
3929 | ||
3930 | return flags; | |
3931 | } | |
3932 | ||
9caccd41 CB |
3933 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3934 | { | |
3935 | struct vfsmount *m = &mnt->mnt; | |
bd303368 | 3936 | struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; |
9caccd41 CB |
3937 | |
3938 | if (!kattr->mnt_userns) | |
3939 | return 0; | |
3940 | ||
bd303368 CB |
3941 | /* |
3942 | * Creating an idmapped mount with the filesystem wide idmapping | |
3943 | * doesn't make sense so block that. We don't allow mushy semantics. | |
3944 | */ | |
3945 | if (kattr->mnt_userns == fs_userns) | |
3946 | return -EINVAL; | |
3947 | ||
9caccd41 CB |
3948 | /* |
3949 | * Once a mount has been idmapped we don't allow it to change its | |
3950 | * mapping. It makes things simpler and callers can just create | |
3951 | * another bind-mount they can idmap if they want to. | |
3952 | */ | |
bb49e9e7 | 3953 | if (is_idmapped_mnt(m)) |
9caccd41 CB |
3954 | return -EPERM; |
3955 | ||
3956 | /* The underlying filesystem doesn't support idmapped mounts yet. */ | |
3957 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) | |
3958 | return -EINVAL; | |
3959 | ||
3960 | /* We're not controlling the superblock. */ | |
bd303368 | 3961 | if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) |
9caccd41 CB |
3962 | return -EPERM; |
3963 | ||
3964 | /* Mount has already been visible in the filesystem hierarchy. */ | |
3965 | if (!is_anon_ns(mnt->mnt_ns)) | |
3966 | return -EINVAL; | |
3967 | ||
3968 | return 0; | |
3969 | } | |
3970 | ||
2a186721 CB |
3971 | static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, |
3972 | struct mount *mnt, int *err) | |
3973 | { | |
3974 | struct mount *m = mnt, *last = NULL; | |
3975 | ||
3976 | if (!is_mounted(&m->mnt)) { | |
3977 | *err = -EINVAL; | |
3978 | goto out; | |
3979 | } | |
3980 | ||
3981 | if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { | |
3982 | *err = -EINVAL; | |
3983 | goto out; | |
3984 | } | |
3985 | ||
3986 | do { | |
3987 | unsigned int flags; | |
3988 | ||
3989 | flags = recalc_flags(kattr, m); | |
3990 | if (!can_change_locked_flags(m, flags)) { | |
3991 | *err = -EPERM; | |
3992 | goto out; | |
3993 | } | |
3994 | ||
9caccd41 CB |
3995 | *err = can_idmap_mount(kattr, m); |
3996 | if (*err) | |
3997 | goto out; | |
3998 | ||
2a186721 CB |
3999 | last = m; |
4000 | ||
4001 | if ((kattr->attr_set & MNT_READONLY) && | |
4002 | !(m->mnt.mnt_flags & MNT_READONLY)) { | |
4003 | *err = mnt_hold_writers(m); | |
4004 | if (*err) | |
4005 | goto out; | |
4006 | } | |
4007 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
4008 | ||
4009 | out: | |
4010 | return last; | |
4011 | } | |
4012 | ||
9caccd41 CB |
4013 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
4014 | { | |
bd303368 | 4015 | struct user_namespace *mnt_userns, *old_mnt_userns; |
9caccd41 CB |
4016 | |
4017 | if (!kattr->mnt_userns) | |
4018 | return; | |
4019 | ||
bd303368 CB |
4020 | /* |
4021 | * We're the only ones able to change the mount's idmapping. So | |
4022 | * mnt->mnt.mnt_userns is stable and we can retrieve it directly. | |
4023 | */ | |
4024 | old_mnt_userns = mnt->mnt.mnt_userns; | |
4025 | ||
9caccd41 CB |
4026 | mnt_userns = get_user_ns(kattr->mnt_userns); |
4027 | /* Pairs with smp_load_acquire() in mnt_user_ns(). */ | |
4028 | smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); | |
bd303368 CB |
4029 | |
4030 | /* | |
4031 | * If this is an idmapped filesystem drop the reference we've taken | |
4032 | * in vfs_create_mount() before. | |
4033 | */ | |
4034 | if (!initial_idmapping(old_mnt_userns)) | |
4035 | put_user_ns(old_mnt_userns); | |
9caccd41 CB |
4036 | } |
4037 | ||
2a186721 CB |
4038 | static void mount_setattr_commit(struct mount_kattr *kattr, |
4039 | struct mount *mnt, struct mount *last, | |
4040 | int err) | |
4041 | { | |
4042 | struct mount *m = mnt; | |
4043 | ||
4044 | do { | |
4045 | if (!err) { | |
4046 | unsigned int flags; | |
4047 | ||
9caccd41 | 4048 | do_idmap_mount(kattr, m); |
2a186721 CB |
4049 | flags = recalc_flags(kattr, m); |
4050 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
4051 | } | |
4052 | ||
4053 | /* | |
4054 | * We either set MNT_READONLY above so make it visible | |
4055 | * before ~MNT_WRITE_HOLD or we failed to recursively | |
4056 | * apply mount options. | |
4057 | */ | |
4058 | if ((kattr->attr_set & MNT_READONLY) && | |
4059 | (m->mnt.mnt_flags & MNT_WRITE_HOLD)) | |
4060 | mnt_unhold_writers(m); | |
4061 | ||
4062 | if (!err && kattr->propagation) | |
4063 | change_mnt_propagation(m, kattr->propagation); | |
4064 | ||
4065 | /* | |
4066 | * On failure, only cleanup until we found the first mount | |
4067 | * we failed to handle. | |
4068 | */ | |
4069 | if (err && m == last) | |
4070 | break; | |
4071 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
4072 | ||
4073 | if (!err) | |
4074 | touch_mnt_namespace(mnt->mnt_ns); | |
4075 | } | |
4076 | ||
4077 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
4078 | { | |
4079 | struct mount *mnt = real_mount(path->mnt), *last = NULL; | |
4080 | int err = 0; | |
4081 | ||
4082 | if (path->dentry != mnt->mnt.mnt_root) | |
4083 | return -EINVAL; | |
4084 | ||
4085 | if (kattr->propagation) { | |
4086 | /* | |
4087 | * Only take namespace_lock() if we're actually changing | |
4088 | * propagation. | |
4089 | */ | |
4090 | namespace_lock(); | |
4091 | if (kattr->propagation == MS_SHARED) { | |
4092 | err = invent_group_ids(mnt, kattr->recurse); | |
4093 | if (err) { | |
4094 | namespace_unlock(); | |
4095 | return err; | |
4096 | } | |
4097 | } | |
4098 | } | |
4099 | ||
4100 | lock_mount_hash(); | |
4101 | ||
4102 | /* | |
4103 | * Get the mount tree in a shape where we can change mount | |
4104 | * properties without failure. | |
4105 | */ | |
4106 | last = mount_setattr_prepare(kattr, mnt, &err); | |
4107 | if (last) /* Commit all changes or revert to the old state. */ | |
4108 | mount_setattr_commit(kattr, mnt, last, err); | |
4109 | ||
4110 | unlock_mount_hash(); | |
4111 | ||
4112 | if (kattr->propagation) { | |
4113 | namespace_unlock(); | |
4114 | if (err) | |
4115 | cleanup_group_ids(mnt, NULL); | |
4116 | } | |
4117 | ||
4118 | return err; | |
4119 | } | |
4120 | ||
9caccd41 CB |
4121 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
4122 | struct mount_kattr *kattr, unsigned int flags) | |
4123 | { | |
4124 | int err = 0; | |
4125 | struct ns_common *ns; | |
4126 | struct user_namespace *mnt_userns; | |
4127 | struct file *file; | |
4128 | ||
4129 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) | |
4130 | return 0; | |
4131 | ||
4132 | /* | |
4133 | * We currently do not support clearing an idmapped mount. If this ever | |
4134 | * is a use-case we can revisit this but for now let's keep it simple | |
4135 | * and not allow it. | |
4136 | */ | |
4137 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) | |
4138 | return -EINVAL; | |
4139 | ||
4140 | if (attr->userns_fd > INT_MAX) | |
4141 | return -EINVAL; | |
4142 | ||
4143 | file = fget(attr->userns_fd); | |
4144 | if (!file) | |
4145 | return -EBADF; | |
4146 | ||
4147 | if (!proc_ns_file(file)) { | |
4148 | err = -EINVAL; | |
4149 | goto out_fput; | |
4150 | } | |
4151 | ||
4152 | ns = get_proc_ns(file_inode(file)); | |
4153 | if (ns->ops->type != CLONE_NEWUSER) { | |
4154 | err = -EINVAL; | |
4155 | goto out_fput; | |
4156 | } | |
4157 | ||
4158 | /* | |
bd303368 CB |
4159 | * The initial idmapping cannot be used to create an idmapped |
4160 | * mount. We use the initial idmapping as an indicator of a mount | |
4161 | * that is not idmapped. It can simply be passed into helpers that | |
4162 | * are aware of idmapped mounts as a convenient shortcut. A user | |
4163 | * can just create a dedicated identity mapping to achieve the same | |
4164 | * result. | |
9caccd41 CB |
4165 | */ |
4166 | mnt_userns = container_of(ns, struct user_namespace, ns); | |
bd303368 | 4167 | if (initial_idmapping(mnt_userns)) { |
9caccd41 CB |
4168 | err = -EPERM; |
4169 | goto out_fput; | |
4170 | } | |
4171 | kattr->mnt_userns = get_user_ns(mnt_userns); | |
4172 | ||
4173 | out_fput: | |
4174 | fput(file); | |
4175 | return err; | |
4176 | } | |
4177 | ||
4178 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, | |
2a186721 CB |
4179 | struct mount_kattr *kattr, unsigned int flags) |
4180 | { | |
4181 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
4182 | ||
4183 | if (flags & AT_NO_AUTOMOUNT) | |
4184 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
4185 | if (flags & AT_SYMLINK_NOFOLLOW) | |
4186 | lookup_flags &= ~LOOKUP_FOLLOW; | |
4187 | if (flags & AT_EMPTY_PATH) | |
4188 | lookup_flags |= LOOKUP_EMPTY; | |
4189 | ||
4190 | *kattr = (struct mount_kattr) { | |
4191 | .lookup_flags = lookup_flags, | |
4192 | .recurse = !!(flags & AT_RECURSIVE), | |
4193 | }; | |
4194 | ||
4195 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
4196 | return -EINVAL; | |
4197 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
4198 | return -EINVAL; | |
4199 | kattr->propagation = attr->propagation; | |
4200 | ||
4201 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
4202 | return -EINVAL; | |
4203 | ||
2a186721 CB |
4204 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
4205 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
4206 | ||
4207 | /* | |
4208 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4209 | * users wanting to transition to a different atime setting cannot | |
4210 | * simply specify the atime setting in @attr_set, but must also | |
4211 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4212 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4213 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4214 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4215 | */ | |
4216 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4217 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4218 | return -EINVAL; | |
4219 | ||
4220 | /* | |
4221 | * Clear all previous time settings as they are mutually | |
4222 | * exclusive. | |
4223 | */ | |
4224 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4225 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4226 | case MOUNT_ATTR_RELATIME: | |
4227 | kattr->attr_set |= MNT_RELATIME; | |
4228 | break; | |
4229 | case MOUNT_ATTR_NOATIME: | |
4230 | kattr->attr_set |= MNT_NOATIME; | |
4231 | break; | |
4232 | case MOUNT_ATTR_STRICTATIME: | |
4233 | break; | |
4234 | default: | |
4235 | return -EINVAL; | |
4236 | } | |
4237 | } else { | |
4238 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4239 | return -EINVAL; | |
4240 | } | |
4241 | ||
9caccd41 CB |
4242 | return build_mount_idmapped(attr, usize, kattr, flags); |
4243 | } | |
4244 | ||
4245 | static void finish_mount_kattr(struct mount_kattr *kattr) | |
4246 | { | |
4247 | put_user_ns(kattr->mnt_userns); | |
4248 | kattr->mnt_userns = NULL; | |
2a186721 CB |
4249 | } |
4250 | ||
4251 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4252 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4253 | size_t, usize) | |
4254 | { | |
4255 | int err; | |
4256 | struct path target; | |
4257 | struct mount_attr attr; | |
4258 | struct mount_kattr kattr; | |
4259 | ||
4260 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4261 | ||
4262 | if (flags & ~(AT_EMPTY_PATH | | |
4263 | AT_RECURSIVE | | |
4264 | AT_SYMLINK_NOFOLLOW | | |
4265 | AT_NO_AUTOMOUNT)) | |
4266 | return -EINVAL; | |
4267 | ||
4268 | if (unlikely(usize > PAGE_SIZE)) | |
4269 | return -E2BIG; | |
4270 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4271 | return -EINVAL; | |
4272 | ||
4273 | if (!may_mount()) | |
4274 | return -EPERM; | |
4275 | ||
4276 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4277 | if (err) | |
4278 | return err; | |
4279 | ||
4280 | /* Don't bother walking through the mounts if this is a nop. */ | |
4281 | if (attr.attr_set == 0 && | |
4282 | attr.attr_clr == 0 && | |
4283 | attr.propagation == 0) | |
4284 | return 0; | |
4285 | ||
9caccd41 | 4286 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
2a186721 CB |
4287 | if (err) |
4288 | return err; | |
4289 | ||
4290 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
012e3322 CB |
4291 | if (!err) { |
4292 | err = do_mount_setattr(&target, &kattr); | |
4293 | path_put(&target); | |
4294 | } | |
9caccd41 | 4295 | finish_mount_kattr(&kattr); |
2a186721 CB |
4296 | return err; |
4297 | } | |
4298 | ||
1da177e4 LT |
4299 | static void __init init_mount_tree(void) |
4300 | { | |
4301 | struct vfsmount *mnt; | |
74e83122 | 4302 | struct mount *m; |
6b3286ed | 4303 | struct mnt_namespace *ns; |
ac748a09 | 4304 | struct path root; |
1da177e4 | 4305 | |
fd3e007f | 4306 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4307 | if (IS_ERR(mnt)) |
4308 | panic("Can't create rootfs"); | |
b3e19d92 | 4309 | |
74e83122 | 4310 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4311 | if (IS_ERR(ns)) |
1da177e4 | 4312 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4313 | m = real_mount(mnt); |
4314 | m->mnt_ns = ns; | |
4315 | ns->root = m; | |
4316 | ns->mounts = 1; | |
4317 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4318 | init_task.nsproxy->mnt_ns = ns; |
4319 | get_mnt_ns(ns); | |
4320 | ||
be08d6d2 AV |
4321 | root.mnt = mnt; |
4322 | root.dentry = mnt->mnt_root; | |
da362b09 | 4323 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4324 | |
4325 | set_fs_pwd(current->fs, &root); | |
4326 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4327 | } |
4328 | ||
74bf17cf | 4329 | void __init mnt_init(void) |
1da177e4 | 4330 | { |
15a67dd8 | 4331 | int err; |
1da177e4 | 4332 | |
7d6fec45 | 4333 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
79f6540b | 4334 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 4335 | |
0818bf27 | 4336 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4337 | sizeof(struct hlist_head), |
0818bf27 | 4338 | mhash_entries, 19, |
3d375d78 | 4339 | HASH_ZERO, |
0818bf27 AV |
4340 | &m_hash_shift, &m_hash_mask, 0, 0); |
4341 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4342 | sizeof(struct hlist_head), | |
4343 | mphash_entries, 19, | |
3d375d78 | 4344 | HASH_ZERO, |
0818bf27 | 4345 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4346 | |
84d17192 | 4347 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4348 | panic("Failed to allocate mount hash table\n"); |
4349 | ||
4b93dc9b TH |
4350 | kernfs_init(); |
4351 | ||
15a67dd8 RD |
4352 | err = sysfs_init(); |
4353 | if (err) | |
4354 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4355 | __func__, err); |
00d26666 GKH |
4356 | fs_kobj = kobject_create_and_add("fs", NULL); |
4357 | if (!fs_kobj) | |
8e24eea7 | 4358 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4359 | shmem_init(); |
1da177e4 LT |
4360 | init_rootfs(); |
4361 | init_mount_tree(); | |
4362 | } | |
4363 | ||
616511d0 | 4364 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4365 | { |
1a7b8969 | 4366 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4367 | return; |
7b00ed6f | 4368 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4369 | free_mnt_ns(ns); |
1da177e4 | 4370 | } |
9d412a43 | 4371 | |
d911b458 | 4372 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4373 | { |
423e0ab0 | 4374 | struct vfsmount *mnt; |
d911b458 | 4375 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4376 | if (!IS_ERR(mnt)) { |
4377 | /* | |
4378 | * it is a longterm mount, don't release mnt until | |
4379 | * we unmount before file sys is unregistered | |
4380 | */ | |
f7a99c5b | 4381 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4382 | } |
4383 | return mnt; | |
9d412a43 | 4384 | } |
d911b458 | 4385 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4386 | |
4387 | void kern_unmount(struct vfsmount *mnt) | |
4388 | { | |
4389 | /* release long term mount so mount point can be released */ | |
4390 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 4391 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 4392 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4393 | mntput(mnt); |
4394 | } | |
4395 | } | |
4396 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4397 | |
df820f8d MS |
4398 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4399 | { | |
4400 | unsigned int i; | |
4401 | ||
4402 | for (i = 0; i < num; i++) | |
4403 | if (mnt[i]) | |
4404 | real_mount(mnt[i])->mnt_ns = NULL; | |
4405 | synchronize_rcu_expedited(); | |
4406 | for (i = 0; i < num; i++) | |
4407 | mntput(mnt[i]); | |
4408 | } | |
4409 | EXPORT_SYMBOL(kern_unmount_array); | |
4410 | ||
02125a82 AV |
4411 | bool our_mnt(struct vfsmount *mnt) |
4412 | { | |
143c8c91 | 4413 | return check_mnt(real_mount(mnt)); |
02125a82 | 4414 | } |
8823c079 | 4415 | |
3151527e EB |
4416 | bool current_chrooted(void) |
4417 | { | |
4418 | /* Does the current process have a non-standard root */ | |
4419 | struct path ns_root; | |
4420 | struct path fs_root; | |
4421 | bool chrooted; | |
4422 | ||
4423 | /* Find the namespace root */ | |
4424 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4425 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4426 | path_get(&ns_root); | |
4427 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4428 | ; | |
4429 | ||
4430 | get_fs_root(current->fs, &fs_root); | |
4431 | ||
4432 | chrooted = !path_equal(&fs_root, &ns_root); | |
4433 | ||
4434 | path_put(&fs_root); | |
4435 | path_put(&ns_root); | |
4436 | ||
4437 | return chrooted; | |
4438 | } | |
4439 | ||
132e4608 DH |
4440 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4441 | const struct super_block *sb, | |
8654df4e | 4442 | int *new_mnt_flags) |
87a8ebd6 | 4443 | { |
8c6cf9cc | 4444 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4445 | struct mount *mnt; |
e51db735 | 4446 | bool visible = false; |
87a8ebd6 | 4447 | |
44bb4385 | 4448 | down_read(&namespace_sem); |
9f6c61f9 | 4449 | lock_ns_list(ns); |
87a8ebd6 | 4450 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4451 | struct mount *child; |
77b1a97d EB |
4452 | int mnt_flags; |
4453 | ||
9f6c61f9 MS |
4454 | if (mnt_is_cursor(mnt)) |
4455 | continue; | |
4456 | ||
132e4608 | 4457 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4458 | continue; |
4459 | ||
7e96c1b0 EB |
4460 | /* This mount is not fully visible if it's root directory |
4461 | * is not the root directory of the filesystem. | |
4462 | */ | |
4463 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4464 | continue; | |
4465 | ||
a1935c17 | 4466 | /* A local view of the mount flags */ |
77b1a97d | 4467 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4468 | |
695e9df0 | 4469 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4470 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4471 | mnt_flags |= MNT_LOCK_READONLY; |
4472 | ||
8c6cf9cc EB |
4473 | /* Verify the mount flags are equal to or more permissive |
4474 | * than the proposed new mount. | |
4475 | */ | |
77b1a97d | 4476 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4477 | !(new_flags & MNT_READONLY)) |
4478 | continue; | |
77b1a97d EB |
4479 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4480 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4481 | continue; |
4482 | ||
ceeb0e5d EB |
4483 | /* This mount is not fully visible if there are any |
4484 | * locked child mounts that cover anything except for | |
4485 | * empty directories. | |
e51db735 EB |
4486 | */ |
4487 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4488 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4489 | /* Only worry about locked mounts */ |
d71ed6c9 | 4490 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4491 | continue; |
7236c85e EB |
4492 | /* Is the directory permanetly empty? */ |
4493 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4494 | goto next; |
87a8ebd6 | 4495 | } |
8c6cf9cc | 4496 | /* Preserve the locked attributes */ |
77b1a97d | 4497 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4498 | MNT_LOCK_ATIME); |
e51db735 EB |
4499 | visible = true; |
4500 | goto found; | |
4501 | next: ; | |
87a8ebd6 | 4502 | } |
e51db735 | 4503 | found: |
9f6c61f9 | 4504 | unlock_ns_list(ns); |
44bb4385 | 4505 | up_read(&namespace_sem); |
e51db735 | 4506 | return visible; |
87a8ebd6 EB |
4507 | } |
4508 | ||
132e4608 | 4509 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4510 | { |
a1935c17 | 4511 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4512 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4513 | unsigned long s_iflags; | |
4514 | ||
4515 | if (ns->user_ns == &init_user_ns) | |
4516 | return false; | |
4517 | ||
4518 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4519 | s_iflags = sb->s_iflags; |
8654df4e EB |
4520 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4521 | return false; | |
4522 | ||
a1935c17 EB |
4523 | if ((s_iflags & required_iflags) != required_iflags) { |
4524 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4525 | required_iflags); | |
4526 | return true; | |
4527 | } | |
4528 | ||
132e4608 | 4529 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4530 | } |
4531 | ||
380cf5ba AL |
4532 | bool mnt_may_suid(struct vfsmount *mnt) |
4533 | { | |
4534 | /* | |
4535 | * Foreign mounts (accessed via fchdir or through /proc | |
4536 | * symlinks) are always treated as if they are nosuid. This | |
4537 | * prevents namespaces from trusting potentially unsafe | |
4538 | * suid/sgid bits, file caps, or security labels that originate | |
4539 | * in other namespaces. | |
4540 | */ | |
4541 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4542 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4543 | } | |
4544 | ||
64964528 | 4545 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4546 | { |
58be2825 | 4547 | struct ns_common *ns = NULL; |
8823c079 EB |
4548 | struct nsproxy *nsproxy; |
4549 | ||
728dba3a EB |
4550 | task_lock(task); |
4551 | nsproxy = task->nsproxy; | |
8823c079 | 4552 | if (nsproxy) { |
58be2825 AV |
4553 | ns = &nsproxy->mnt_ns->ns; |
4554 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4555 | } |
728dba3a | 4556 | task_unlock(task); |
8823c079 EB |
4557 | |
4558 | return ns; | |
4559 | } | |
4560 | ||
64964528 | 4561 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4562 | { |
58be2825 | 4563 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4564 | } |
4565 | ||
f2a8d52e | 4566 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4567 | { |
f2a8d52e CB |
4568 | struct nsproxy *nsproxy = nsset->nsproxy; |
4569 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4570 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4571 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4572 | struct path root; |
4f757f3c | 4573 | int err; |
8823c079 | 4574 | |
0c55cfc4 | 4575 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4576 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4577 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4578 | return -EPERM; |
8823c079 | 4579 | |
74e83122 AV |
4580 | if (is_anon_ns(mnt_ns)) |
4581 | return -EINVAL; | |
4582 | ||
8823c079 EB |
4583 | if (fs->users != 1) |
4584 | return -EINVAL; | |
4585 | ||
4586 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4587 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4588 | nsproxy->mnt_ns = mnt_ns; |
4589 | ||
4590 | /* Find the root */ | |
4f757f3c AV |
4591 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4592 | "/", LOOKUP_DOWN, &root); | |
4593 | if (err) { | |
4594 | /* revert to old namespace */ | |
4595 | nsproxy->mnt_ns = old_mnt_ns; | |
4596 | put_mnt_ns(mnt_ns); | |
4597 | return err; | |
4598 | } | |
8823c079 | 4599 | |
4068367c AV |
4600 | put_mnt_ns(old_mnt_ns); |
4601 | ||
8823c079 EB |
4602 | /* Update the pwd and root */ |
4603 | set_fs_pwd(fs, &root); | |
4604 | set_fs_root(fs, &root); | |
4605 | ||
4606 | path_put(&root); | |
4607 | return 0; | |
4608 | } | |
4609 | ||
bcac25a5 AV |
4610 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4611 | { | |
4612 | return to_mnt_ns(ns)->user_ns; | |
4613 | } | |
4614 | ||
8823c079 EB |
4615 | const struct proc_ns_operations mntns_operations = { |
4616 | .name = "mnt", | |
4617 | .type = CLONE_NEWNS, | |
4618 | .get = mntns_get, | |
4619 | .put = mntns_put, | |
4620 | .install = mntns_install, | |
bcac25a5 | 4621 | .owner = mntns_owner, |
8823c079 | 4622 | }; |
ab171b95 LC |
4623 | |
4624 | #ifdef CONFIG_SYSCTL | |
4625 | static struct ctl_table fs_namespace_sysctls[] = { | |
4626 | { | |
4627 | .procname = "mount-max", | |
4628 | .data = &sysctl_mount_max, | |
4629 | .maxlen = sizeof(unsigned int), | |
4630 | .mode = 0644, | |
4631 | .proc_handler = proc_dointvec_minmax, | |
4632 | .extra1 = SYSCTL_ONE, | |
4633 | }, | |
4634 | { } | |
4635 | }; | |
4636 | ||
4637 | static int __init init_fs_namespace_sysctls(void) | |
4638 | { | |
4639 | register_sysctl_init("fs", fs_namespace_sysctls); | |
4640 | return 0; | |
4641 | } | |
4642 | fs_initcall(init_fs_namespace_sysctls); | |
4643 | ||
4644 | #endif /* CONFIG_SYSCTL */ |