]>
Commit | Line | Data |
---|---|---|
59bd9ded | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namespace.c | |
4 | * | |
5 | * (C) Copyright Al Viro 2000, 2001 | |
1da177e4 LT |
6 | * |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
a07b2000 | 23 | #include <linux/file.h> |
d10577a8 | 24 | #include <linux/uaccess.h> |
0bb80f24 | 25 | #include <linux/proc_ns.h> |
20b4fb48 | 26 | #include <linux/magic.h> |
57c8a661 | 27 | #include <linux/memblock.h> |
9caccd41 | 28 | #include <linux/proc_fs.h> |
9ea459e1 | 29 | #include <linux/task_work.h> |
9164bb4a | 30 | #include <linux/sched/task.h> |
e262e32d | 31 | #include <uapi/linux/mount.h> |
9bc61ab1 | 32 | #include <linux/fs_context.h> |
037f11b4 | 33 | #include <linux/shmem_fs.h> |
9164bb4a | 34 | |
07b20889 | 35 | #include "pnode.h" |
948730b0 | 36 | #include "internal.h" |
1da177e4 | 37 | |
d2921684 EB |
38 | /* Maximum number of mounts in a mount namespace */ |
39 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
40 | ||
0818bf27 AV |
41 | static unsigned int m_hash_mask __read_mostly; |
42 | static unsigned int m_hash_shift __read_mostly; | |
43 | static unsigned int mp_hash_mask __read_mostly; | |
44 | static unsigned int mp_hash_shift __read_mostly; | |
45 | ||
46 | static __initdata unsigned long mhash_entries; | |
47 | static int __init set_mhash_entries(char *str) | |
48 | { | |
49 | if (!str) | |
50 | return 0; | |
51 | mhash_entries = simple_strtoul(str, &str, 0); | |
52 | return 1; | |
53 | } | |
54 | __setup("mhash_entries=", set_mhash_entries); | |
55 | ||
56 | static __initdata unsigned long mphash_entries; | |
57 | static int __init set_mphash_entries(char *str) | |
58 | { | |
59 | if (!str) | |
60 | return 0; | |
61 | mphash_entries = simple_strtoul(str, &str, 0); | |
62 | return 1; | |
63 | } | |
64 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 65 | |
c7999c36 | 66 | static u64 event; |
73cd49ec | 67 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 68 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 69 | |
38129a13 | 70 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 71 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 72 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 73 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
74 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
75 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 76 | |
2a186721 CB |
77 | struct mount_kattr { |
78 | unsigned int attr_set; | |
79 | unsigned int attr_clr; | |
80 | unsigned int propagation; | |
81 | unsigned int lookup_flags; | |
82 | bool recurse; | |
9caccd41 | 83 | struct user_namespace *mnt_userns; |
2a186721 CB |
84 | }; |
85 | ||
f87fd4c2 | 86 | /* /sys/fs */ |
00d26666 GKH |
87 | struct kobject *fs_kobj; |
88 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 89 | |
99b7db7b NP |
90 | /* |
91 | * vfsmount lock may be taken for read to prevent changes to the | |
92 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
93 | * up the tree. | |
94 | * | |
95 | * It should be taken for write in all cases where the vfsmount | |
96 | * tree or hash is modified or when a vfsmount structure is modified. | |
97 | */ | |
48a066e7 | 98 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 99 | |
d033cb67 CB |
100 | static inline void lock_mount_hash(void) |
101 | { | |
102 | write_seqlock(&mount_lock); | |
103 | } | |
104 | ||
105 | static inline void unlock_mount_hash(void) | |
106 | { | |
107 | write_sequnlock(&mount_lock); | |
108 | } | |
109 | ||
38129a13 | 110 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 111 | { |
b58fed8b RP |
112 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
113 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
114 | tmp = tmp + (tmp >> m_hash_shift); |
115 | return &mount_hashtable[tmp & m_hash_mask]; | |
116 | } | |
117 | ||
118 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
119 | { | |
120 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
121 | tmp = tmp + (tmp >> mp_hash_shift); | |
122 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
123 | } |
124 | ||
b105e270 | 125 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 126 | { |
169b480e MW |
127 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
128 | ||
129 | if (res < 0) | |
130 | return res; | |
131 | mnt->mnt_id = res; | |
132 | return 0; | |
73cd49ec MS |
133 | } |
134 | ||
b105e270 | 135 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 136 | { |
169b480e | 137 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
138 | } |
139 | ||
719f5d7f MS |
140 | /* |
141 | * Allocate a new peer group ID | |
719f5d7f | 142 | */ |
4b8b21f4 | 143 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 144 | { |
169b480e | 145 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 146 | |
169b480e MW |
147 | if (res < 0) |
148 | return res; | |
149 | mnt->mnt_group_id = res; | |
150 | return 0; | |
719f5d7f MS |
151 | } |
152 | ||
153 | /* | |
154 | * Release a peer group ID | |
155 | */ | |
4b8b21f4 | 156 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 157 | { |
169b480e | 158 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 159 | mnt->mnt_group_id = 0; |
719f5d7f MS |
160 | } |
161 | ||
b3e19d92 NP |
162 | /* |
163 | * vfsmount lock must be held for read | |
164 | */ | |
83adc753 | 165 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
166 | { |
167 | #ifdef CONFIG_SMP | |
68e8a9fe | 168 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
169 | #else |
170 | preempt_disable(); | |
68e8a9fe | 171 | mnt->mnt_count += n; |
b3e19d92 NP |
172 | preempt_enable(); |
173 | #endif | |
174 | } | |
175 | ||
b3e19d92 NP |
176 | /* |
177 | * vfsmount lock must be held for write | |
178 | */ | |
edf7ddbf | 179 | int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
180 | { |
181 | #ifdef CONFIG_SMP | |
edf7ddbf | 182 | int count = 0; |
b3e19d92 NP |
183 | int cpu; |
184 | ||
185 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 186 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
187 | } |
188 | ||
189 | return count; | |
190 | #else | |
68e8a9fe | 191 | return mnt->mnt_count; |
b3e19d92 NP |
192 | #endif |
193 | } | |
194 | ||
b105e270 | 195 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 196 | { |
c63181e6 AV |
197 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
198 | if (mnt) { | |
73cd49ec MS |
199 | int err; |
200 | ||
c63181e6 | 201 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
202 | if (err) |
203 | goto out_free_cache; | |
204 | ||
205 | if (name) { | |
fcc139ae | 206 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 207 | if (!mnt->mnt_devname) |
88b38782 | 208 | goto out_free_id; |
73cd49ec MS |
209 | } |
210 | ||
b3e19d92 | 211 | #ifdef CONFIG_SMP |
c63181e6 AV |
212 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
213 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
214 | goto out_free_devname; |
215 | ||
c63181e6 | 216 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 217 | #else |
c63181e6 AV |
218 | mnt->mnt_count = 1; |
219 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
220 | #endif |
221 | ||
38129a13 | 222 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
223 | INIT_LIST_HEAD(&mnt->mnt_child); |
224 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
225 | INIT_LIST_HEAD(&mnt->mnt_list); | |
226 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
227 | INIT_LIST_HEAD(&mnt->mnt_share); | |
228 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
229 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 230 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 231 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
56cbb429 | 232 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
a6435940 | 233 | mnt->mnt.mnt_userns = &init_user_ns; |
1da177e4 | 234 | } |
c63181e6 | 235 | return mnt; |
88b38782 | 236 | |
d3ef3d73 NP |
237 | #ifdef CONFIG_SMP |
238 | out_free_devname: | |
fcc139ae | 239 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 240 | #endif |
88b38782 | 241 | out_free_id: |
c63181e6 | 242 | mnt_free_id(mnt); |
88b38782 | 243 | out_free_cache: |
c63181e6 | 244 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 245 | return NULL; |
1da177e4 LT |
246 | } |
247 | ||
3d733633 DH |
248 | /* |
249 | * Most r/o checks on a fs are for operations that take | |
250 | * discrete amounts of time, like a write() or unlink(). | |
251 | * We must keep track of when those operations start | |
252 | * (for permission checks) and when they end, so that | |
253 | * we can determine when writes are able to occur to | |
254 | * a filesystem. | |
255 | */ | |
256 | /* | |
257 | * __mnt_is_readonly: check whether a mount is read-only | |
258 | * @mnt: the mount to check for its write status | |
259 | * | |
260 | * This shouldn't be used directly ouside of the VFS. | |
261 | * It does not guarantee that the filesystem will stay | |
262 | * r/w, just that it is right *now*. This can not and | |
263 | * should not be used in place of IS_RDONLY(inode). | |
264 | * mnt_want/drop_write() will _keep_ the filesystem | |
265 | * r/w. | |
266 | */ | |
43f5e655 | 267 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 268 | { |
43f5e655 | 269 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
270 | } |
271 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
272 | ||
83adc753 | 273 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
274 | { |
275 | #ifdef CONFIG_SMP | |
68e8a9fe | 276 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 277 | #else |
68e8a9fe | 278 | mnt->mnt_writers++; |
d3ef3d73 NP |
279 | #endif |
280 | } | |
3d733633 | 281 | |
83adc753 | 282 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 283 | { |
d3ef3d73 | 284 | #ifdef CONFIG_SMP |
68e8a9fe | 285 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 286 | #else |
68e8a9fe | 287 | mnt->mnt_writers--; |
d3ef3d73 | 288 | #endif |
3d733633 | 289 | } |
3d733633 | 290 | |
83adc753 | 291 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 292 | { |
d3ef3d73 NP |
293 | #ifdef CONFIG_SMP |
294 | unsigned int count = 0; | |
3d733633 | 295 | int cpu; |
3d733633 DH |
296 | |
297 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 298 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 299 | } |
3d733633 | 300 | |
d3ef3d73 NP |
301 | return count; |
302 | #else | |
303 | return mnt->mnt_writers; | |
304 | #endif | |
3d733633 DH |
305 | } |
306 | ||
4ed5e82f MS |
307 | static int mnt_is_readonly(struct vfsmount *mnt) |
308 | { | |
309 | if (mnt->mnt_sb->s_readonly_remount) | |
310 | return 1; | |
311 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
312 | smp_rmb(); | |
313 | return __mnt_is_readonly(mnt); | |
314 | } | |
315 | ||
8366025e | 316 | /* |
eb04c282 JK |
317 | * Most r/o & frozen checks on a fs are for operations that take discrete |
318 | * amounts of time, like a write() or unlink(). We must keep track of when | |
319 | * those operations start (for permission checks) and when they end, so that we | |
320 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
321 | */ |
322 | /** | |
eb04c282 | 323 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 324 | * @m: the mount on which to take a write |
8366025e | 325 | * |
eb04c282 JK |
326 | * This tells the low-level filesystem that a write is about to be performed to |
327 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
328 | * returning success. This operation does not protect against filesystem being | |
329 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
330 | * called. This is effectively a refcount. | |
8366025e | 331 | */ |
eb04c282 | 332 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 333 | { |
83adc753 | 334 | struct mount *mnt = real_mount(m); |
3d733633 | 335 | int ret = 0; |
3d733633 | 336 | |
d3ef3d73 | 337 | preempt_disable(); |
c6653a83 | 338 | mnt_inc_writers(mnt); |
d3ef3d73 | 339 | /* |
c6653a83 | 340 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
341 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
342 | * incremented count after it has set MNT_WRITE_HOLD. | |
343 | */ | |
344 | smp_mb(); | |
6aa7de05 | 345 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 NP |
346 | cpu_relax(); |
347 | /* | |
348 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
349 | * be set to match its requirements. So we must not load that until | |
350 | * MNT_WRITE_HOLD is cleared. | |
351 | */ | |
352 | smp_rmb(); | |
4ed5e82f | 353 | if (mnt_is_readonly(m)) { |
c6653a83 | 354 | mnt_dec_writers(mnt); |
3d733633 | 355 | ret = -EROFS; |
3d733633 | 356 | } |
d3ef3d73 | 357 | preempt_enable(); |
eb04c282 JK |
358 | |
359 | return ret; | |
360 | } | |
361 | ||
362 | /** | |
363 | * mnt_want_write - get write access to a mount | |
364 | * @m: the mount on which to take a write | |
365 | * | |
366 | * This tells the low-level filesystem that a write is about to be performed to | |
367 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
368 | * is not frozen) before returning success. When the write operation is | |
369 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
370 | */ | |
371 | int mnt_want_write(struct vfsmount *m) | |
372 | { | |
373 | int ret; | |
374 | ||
375 | sb_start_write(m->mnt_sb); | |
376 | ret = __mnt_want_write(m); | |
377 | if (ret) | |
378 | sb_end_write(m->mnt_sb); | |
3d733633 | 379 | return ret; |
8366025e DH |
380 | } |
381 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
382 | ||
96029c4e | 383 | /** |
eb04c282 | 384 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
385 | * @file: the file who's mount on which to take a write |
386 | * | |
14e43bf4 EB |
387 | * This is like __mnt_want_write, but if the file is already open for writing it |
388 | * skips incrementing mnt_writers (since the open file already has a reference) | |
389 | * and instead only does the check for emergency r/o remounts. This must be | |
390 | * paired with __mnt_drop_write_file. | |
96029c4e | 391 | */ |
eb04c282 | 392 | int __mnt_want_write_file(struct file *file) |
96029c4e | 393 | { |
14e43bf4 EB |
394 | if (file->f_mode & FMODE_WRITER) { |
395 | /* | |
396 | * Superblock may have become readonly while there are still | |
397 | * writable fd's, e.g. due to a fs error with errors=remount-ro | |
398 | */ | |
399 | if (__mnt_is_readonly(file->f_path.mnt)) | |
400 | return -EROFS; | |
401 | return 0; | |
402 | } | |
403 | return __mnt_want_write(file->f_path.mnt); | |
96029c4e | 404 | } |
eb04c282 | 405 | |
7c6893e3 MS |
406 | /** |
407 | * mnt_want_write_file - get write access to a file's mount | |
408 | * @file: the file who's mount on which to take a write | |
409 | * | |
14e43bf4 EB |
410 | * This is like mnt_want_write, but if the file is already open for writing it |
411 | * skips incrementing mnt_writers (since the open file already has a reference) | |
412 | * and instead only does the freeze protection and the check for emergency r/o | |
413 | * remounts. This must be paired with mnt_drop_write_file. | |
7c6893e3 MS |
414 | */ |
415 | int mnt_want_write_file(struct file *file) | |
416 | { | |
417 | int ret; | |
418 | ||
a6795a58 | 419 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
420 | ret = __mnt_want_write_file(file); |
421 | if (ret) | |
a6795a58 | 422 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
423 | return ret; |
424 | } | |
96029c4e NP |
425 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
426 | ||
8366025e | 427 | /** |
eb04c282 | 428 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
429 | * @mnt: the mount on which to give up write access |
430 | * | |
431 | * Tells the low-level filesystem that we are done | |
432 | * performing writes to it. Must be matched with | |
eb04c282 | 433 | * __mnt_want_write() call above. |
8366025e | 434 | */ |
eb04c282 | 435 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 436 | { |
d3ef3d73 | 437 | preempt_disable(); |
83adc753 | 438 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 439 | preempt_enable(); |
8366025e | 440 | } |
eb04c282 JK |
441 | |
442 | /** | |
443 | * mnt_drop_write - give up write access to a mount | |
444 | * @mnt: the mount on which to give up write access | |
445 | * | |
446 | * Tells the low-level filesystem that we are done performing writes to it and | |
447 | * also allows filesystem to be frozen again. Must be matched with | |
448 | * mnt_want_write() call above. | |
449 | */ | |
450 | void mnt_drop_write(struct vfsmount *mnt) | |
451 | { | |
452 | __mnt_drop_write(mnt); | |
453 | sb_end_write(mnt->mnt_sb); | |
454 | } | |
8366025e DH |
455 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
456 | ||
eb04c282 JK |
457 | void __mnt_drop_write_file(struct file *file) |
458 | { | |
14e43bf4 EB |
459 | if (!(file->f_mode & FMODE_WRITER)) |
460 | __mnt_drop_write(file->f_path.mnt); | |
eb04c282 JK |
461 | } |
462 | ||
7c6893e3 MS |
463 | void mnt_drop_write_file(struct file *file) |
464 | { | |
a6795a58 | 465 | __mnt_drop_write_file(file); |
7c6893e3 MS |
466 | sb_end_write(file_inode(file)->i_sb); |
467 | } | |
2a79f17e AV |
468 | EXPORT_SYMBOL(mnt_drop_write_file); |
469 | ||
fbdc2f6c | 470 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 471 | { |
83adc753 | 472 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 473 | /* |
d3ef3d73 NP |
474 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
475 | * should be visible before we do. | |
3d733633 | 476 | */ |
d3ef3d73 NP |
477 | smp_mb(); |
478 | ||
3d733633 | 479 | /* |
d3ef3d73 NP |
480 | * With writers on hold, if this value is zero, then there are |
481 | * definitely no active writers (although held writers may subsequently | |
482 | * increment the count, they'll have to wait, and decrement it after | |
483 | * seeing MNT_READONLY). | |
484 | * | |
485 | * It is OK to have counter incremented on one CPU and decremented on | |
486 | * another: the sum will add up correctly. The danger would be when we | |
487 | * sum up each counter, if we read a counter before it is incremented, | |
488 | * but then read another CPU's count which it has been subsequently | |
489 | * decremented from -- we would see more decrements than we should. | |
490 | * MNT_WRITE_HOLD protects against this scenario, because | |
491 | * mnt_want_write first increments count, then smp_mb, then spins on | |
492 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
493 | * we're counting up here. | |
3d733633 | 494 | */ |
c6653a83 | 495 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
496 | return -EBUSY; |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
501 | static inline void mnt_unhold_writers(struct mount *mnt) | |
502 | { | |
d3ef3d73 NP |
503 | /* |
504 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
505 | * that become unheld will see MNT_READONLY. | |
506 | */ | |
507 | smp_wmb(); | |
83adc753 | 508 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
509 | } |
510 | ||
511 | static int mnt_make_readonly(struct mount *mnt) | |
512 | { | |
513 | int ret; | |
514 | ||
515 | ret = mnt_hold_writers(mnt); | |
516 | if (!ret) | |
517 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
518 | mnt_unhold_writers(mnt); | |
3d733633 | 519 | return ret; |
8366025e | 520 | } |
8366025e | 521 | |
4ed5e82f MS |
522 | int sb_prepare_remount_readonly(struct super_block *sb) |
523 | { | |
524 | struct mount *mnt; | |
525 | int err = 0; | |
526 | ||
8e8b8796 MS |
527 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
528 | if (atomic_long_read(&sb->s_remove_count)) | |
529 | return -EBUSY; | |
530 | ||
719ea2fb | 531 | lock_mount_hash(); |
4ed5e82f MS |
532 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
533 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
534 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
535 | smp_mb(); | |
536 | if (mnt_get_writers(mnt) > 0) { | |
537 | err = -EBUSY; | |
538 | break; | |
539 | } | |
540 | } | |
541 | } | |
8e8b8796 MS |
542 | if (!err && atomic_long_read(&sb->s_remove_count)) |
543 | err = -EBUSY; | |
544 | ||
4ed5e82f MS |
545 | if (!err) { |
546 | sb->s_readonly_remount = 1; | |
547 | smp_wmb(); | |
548 | } | |
549 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
550 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
551 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
552 | } | |
719ea2fb | 553 | unlock_mount_hash(); |
4ed5e82f MS |
554 | |
555 | return err; | |
556 | } | |
557 | ||
b105e270 | 558 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 559 | { |
a6435940 CB |
560 | struct user_namespace *mnt_userns; |
561 | ||
562 | mnt_userns = mnt_user_ns(&mnt->mnt); | |
563 | if (mnt_userns != &init_user_ns) | |
564 | put_user_ns(mnt_userns); | |
fcc139ae | 565 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 566 | #ifdef CONFIG_SMP |
68e8a9fe | 567 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 568 | #endif |
b105e270 | 569 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
570 | } |
571 | ||
8ffcb32e DH |
572 | static void delayed_free_vfsmnt(struct rcu_head *head) |
573 | { | |
574 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
575 | } | |
576 | ||
48a066e7 | 577 | /* call under rcu_read_lock */ |
294d71ff | 578 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
579 | { |
580 | struct mount *mnt; | |
581 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 582 | return 1; |
48a066e7 | 583 | if (bastard == NULL) |
294d71ff | 584 | return 0; |
48a066e7 AV |
585 | mnt = real_mount(bastard); |
586 | mnt_add_count(mnt, 1); | |
119e1ef8 | 587 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 588 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 589 | return 0; |
48a066e7 AV |
590 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
591 | mnt_add_count(mnt, -1); | |
294d71ff AV |
592 | return 1; |
593 | } | |
119e1ef8 AV |
594 | lock_mount_hash(); |
595 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
596 | mnt_add_count(mnt, -1); | |
597 | unlock_mount_hash(); | |
598 | return 1; | |
599 | } | |
600 | unlock_mount_hash(); | |
601 | /* caller will mntput() */ | |
294d71ff AV |
602 | return -1; |
603 | } | |
604 | ||
605 | /* call under rcu_read_lock */ | |
606 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
607 | { | |
608 | int res = __legitimize_mnt(bastard, seq); | |
609 | if (likely(!res)) | |
610 | return true; | |
611 | if (unlikely(res < 0)) { | |
612 | rcu_read_unlock(); | |
613 | mntput(bastard); | |
614 | rcu_read_lock(); | |
48a066e7 | 615 | } |
48a066e7 AV |
616 | return false; |
617 | } | |
618 | ||
1da177e4 | 619 | /* |
474279dc | 620 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 621 | * call under rcu_read_lock() |
1da177e4 | 622 | */ |
474279dc | 623 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 624 | { |
38129a13 | 625 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
626 | struct mount *p; |
627 | ||
38129a13 | 628 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
629 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
630 | return p; | |
631 | return NULL; | |
632 | } | |
633 | ||
a05964f3 | 634 | /* |
f015f126 DH |
635 | * lookup_mnt - Return the first child mount mounted at path |
636 | * | |
637 | * "First" means first mounted chronologically. If you create the | |
638 | * following mounts: | |
639 | * | |
640 | * mount /dev/sda1 /mnt | |
641 | * mount /dev/sda2 /mnt | |
642 | * mount /dev/sda3 /mnt | |
643 | * | |
644 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
645 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
646 | * /dev/sda2, then /dev/sda3, then NULL. | |
647 | * | |
648 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 649 | */ |
ca71cf71 | 650 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 651 | { |
c7105365 | 652 | struct mount *child_mnt; |
48a066e7 AV |
653 | struct vfsmount *m; |
654 | unsigned seq; | |
99b7db7b | 655 | |
48a066e7 AV |
656 | rcu_read_lock(); |
657 | do { | |
658 | seq = read_seqbegin(&mount_lock); | |
659 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
660 | m = child_mnt ? &child_mnt->mnt : NULL; | |
661 | } while (!legitimize_mnt(m, seq)); | |
662 | rcu_read_unlock(); | |
663 | return m; | |
a05964f3 RP |
664 | } |
665 | ||
9f6c61f9 MS |
666 | static inline void lock_ns_list(struct mnt_namespace *ns) |
667 | { | |
668 | spin_lock(&ns->ns_lock); | |
669 | } | |
670 | ||
671 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
672 | { | |
673 | spin_unlock(&ns->ns_lock); | |
674 | } | |
675 | ||
676 | static inline bool mnt_is_cursor(struct mount *mnt) | |
677 | { | |
678 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
679 | } | |
680 | ||
7af1364f EB |
681 | /* |
682 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
683 | * current mount namespace. | |
684 | * | |
685 | * The common case is dentries are not mountpoints at all and that | |
686 | * test is handled inline. For the slow case when we are actually | |
687 | * dealing with a mountpoint of some kind, walk through all of the | |
688 | * mounts in the current mount namespace and test to see if the dentry | |
689 | * is a mountpoint. | |
690 | * | |
691 | * The mount_hashtable is not usable in the context because we | |
692 | * need to identify all mounts that may be in the current mount | |
693 | * namespace not just a mount that happens to have some specified | |
694 | * parent mount. | |
695 | */ | |
696 | bool __is_local_mountpoint(struct dentry *dentry) | |
697 | { | |
698 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
699 | struct mount *mnt; | |
700 | bool is_covered = false; | |
701 | ||
7af1364f | 702 | down_read(&namespace_sem); |
9f6c61f9 | 703 | lock_ns_list(ns); |
7af1364f | 704 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
705 | if (mnt_is_cursor(mnt)) |
706 | continue; | |
7af1364f EB |
707 | is_covered = (mnt->mnt_mountpoint == dentry); |
708 | if (is_covered) | |
709 | break; | |
710 | } | |
9f6c61f9 | 711 | unlock_ns_list(ns); |
7af1364f | 712 | up_read(&namespace_sem); |
5ad05cc8 | 713 | |
7af1364f EB |
714 | return is_covered; |
715 | } | |
716 | ||
e2dfa935 | 717 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 718 | { |
0818bf27 | 719 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
720 | struct mountpoint *mp; |
721 | ||
0818bf27 | 722 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 723 | if (mp->m_dentry == dentry) { |
84d17192 AV |
724 | mp->m_count++; |
725 | return mp; | |
726 | } | |
727 | } | |
e2dfa935 EB |
728 | return NULL; |
729 | } | |
730 | ||
3895dbf8 | 731 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 732 | { |
3895dbf8 | 733 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 734 | int ret; |
84d17192 | 735 | |
3895dbf8 | 736 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
737 | /* might be worth a WARN_ON() */ |
738 | if (d_unlinked(dentry)) | |
739 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
740 | mountpoint: |
741 | read_seqlock_excl(&mount_lock); | |
742 | mp = lookup_mountpoint(dentry); | |
743 | read_sequnlock_excl(&mount_lock); | |
744 | if (mp) | |
745 | goto done; | |
746 | } | |
747 | ||
748 | if (!new) | |
749 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
750 | if (!new) | |
84d17192 AV |
751 | return ERR_PTR(-ENOMEM); |
752 | ||
3895dbf8 EB |
753 | |
754 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 755 | ret = d_set_mounted(dentry); |
eed81007 | 756 | |
3895dbf8 EB |
757 | /* Someone else set d_mounted? */ |
758 | if (ret == -EBUSY) | |
759 | goto mountpoint; | |
760 | ||
761 | /* The dentry is not available as a mountpoint? */ | |
762 | mp = ERR_PTR(ret); | |
763 | if (ret) | |
764 | goto done; | |
765 | ||
766 | /* Add the new mountpoint to the hash table */ | |
767 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 768 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
769 | new->m_count = 1; |
770 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
771 | INIT_HLIST_HEAD(&new->m_list); | |
772 | read_sequnlock_excl(&mount_lock); | |
773 | ||
774 | mp = new; | |
775 | new = NULL; | |
776 | done: | |
777 | kfree(new); | |
84d17192 AV |
778 | return mp; |
779 | } | |
780 | ||
4edbe133 AV |
781 | /* |
782 | * vfsmount lock must be held. Additionally, the caller is responsible | |
783 | * for serializing calls for given disposal list. | |
784 | */ | |
785 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
786 | { |
787 | if (!--mp->m_count) { | |
788 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 789 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
790 | spin_lock(&dentry->d_lock); |
791 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
792 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 793 | dput_to_list(dentry, list); |
0818bf27 | 794 | hlist_del(&mp->m_hash); |
84d17192 AV |
795 | kfree(mp); |
796 | } | |
797 | } | |
798 | ||
4edbe133 AV |
799 | /* called with namespace_lock and vfsmount lock */ |
800 | static void put_mountpoint(struct mountpoint *mp) | |
801 | { | |
802 | __put_mountpoint(mp, &ex_mountpoints); | |
803 | } | |
804 | ||
143c8c91 | 805 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 806 | { |
6b3286ed | 807 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
808 | } |
809 | ||
99b7db7b NP |
810 | /* |
811 | * vfsmount lock must be held for write | |
812 | */ | |
6b3286ed | 813 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
814 | { |
815 | if (ns) { | |
816 | ns->event = ++event; | |
817 | wake_up_interruptible(&ns->poll); | |
818 | } | |
819 | } | |
820 | ||
99b7db7b NP |
821 | /* |
822 | * vfsmount lock must be held for write | |
823 | */ | |
6b3286ed | 824 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
825 | { |
826 | if (ns && ns->event != event) { | |
827 | ns->event = event; | |
828 | wake_up_interruptible(&ns->poll); | |
829 | } | |
830 | } | |
831 | ||
99b7db7b NP |
832 | /* |
833 | * vfsmount lock must be held for write | |
834 | */ | |
e4e59906 | 835 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 836 | { |
e4e59906 | 837 | struct mountpoint *mp; |
0714a533 | 838 | mnt->mnt_parent = mnt; |
a73324da | 839 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 840 | list_del_init(&mnt->mnt_child); |
38129a13 | 841 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 842 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 843 | mp = mnt->mnt_mp; |
84d17192 | 844 | mnt->mnt_mp = NULL; |
e4e59906 | 845 | return mp; |
7bdb11de EB |
846 | } |
847 | ||
6a46c573 EB |
848 | /* |
849 | * vfsmount lock must be held for write | |
850 | */ | |
851 | static void umount_mnt(struct mount *mnt) | |
852 | { | |
e4e59906 | 853 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
854 | } |
855 | ||
99b7db7b NP |
856 | /* |
857 | * vfsmount lock must be held for write | |
858 | */ | |
84d17192 AV |
859 | void mnt_set_mountpoint(struct mount *mnt, |
860 | struct mountpoint *mp, | |
44d964d6 | 861 | struct mount *child_mnt) |
b90fa9ae | 862 | { |
84d17192 | 863 | mp->m_count++; |
3a2393d7 | 864 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 865 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 866 | child_mnt->mnt_parent = mnt; |
84d17192 | 867 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 868 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
869 | } |
870 | ||
1064f874 EB |
871 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
872 | { | |
873 | hlist_add_head_rcu(&mnt->mnt_hash, | |
874 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
875 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
876 | } | |
877 | ||
99b7db7b NP |
878 | /* |
879 | * vfsmount lock must be held for write | |
880 | */ | |
84d17192 AV |
881 | static void attach_mnt(struct mount *mnt, |
882 | struct mount *parent, | |
883 | struct mountpoint *mp) | |
1da177e4 | 884 | { |
84d17192 | 885 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 886 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
887 | } |
888 | ||
1064f874 | 889 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 890 | { |
1064f874 | 891 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
892 | struct mount *old_parent = mnt->mnt_parent; |
893 | ||
894 | list_del_init(&mnt->mnt_child); | |
895 | hlist_del_init(&mnt->mnt_mp_list); | |
896 | hlist_del_init_rcu(&mnt->mnt_hash); | |
897 | ||
898 | attach_mnt(mnt, parent, mp); | |
899 | ||
900 | put_mountpoint(old_mp); | |
1064f874 | 901 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
902 | } |
903 | ||
b90fa9ae | 904 | /* |
99b7db7b | 905 | * vfsmount lock must be held for write |
b90fa9ae | 906 | */ |
1064f874 | 907 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 908 | { |
0714a533 | 909 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 910 | struct mount *m; |
b90fa9ae | 911 | LIST_HEAD(head); |
143c8c91 | 912 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 913 | |
0714a533 | 914 | BUG_ON(parent == mnt); |
b90fa9ae | 915 | |
1a4eeaf2 | 916 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 917 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 918 | m->mnt_ns = n; |
f03c6599 | 919 | |
b90fa9ae RP |
920 | list_splice(&head, n->list.prev); |
921 | ||
d2921684 EB |
922 | n->mounts += n->pending_mounts; |
923 | n->pending_mounts = 0; | |
924 | ||
1064f874 | 925 | __attach_mnt(mnt, parent); |
6b3286ed | 926 | touch_mnt_namespace(n); |
1da177e4 LT |
927 | } |
928 | ||
909b0a88 | 929 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 930 | { |
6b41d536 AV |
931 | struct list_head *next = p->mnt_mounts.next; |
932 | if (next == &p->mnt_mounts) { | |
1da177e4 | 933 | while (1) { |
909b0a88 | 934 | if (p == root) |
1da177e4 | 935 | return NULL; |
6b41d536 AV |
936 | next = p->mnt_child.next; |
937 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 938 | break; |
0714a533 | 939 | p = p->mnt_parent; |
1da177e4 LT |
940 | } |
941 | } | |
6b41d536 | 942 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
943 | } |
944 | ||
315fc83e | 945 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 946 | { |
6b41d536 AV |
947 | struct list_head *prev = p->mnt_mounts.prev; |
948 | while (prev != &p->mnt_mounts) { | |
949 | p = list_entry(prev, struct mount, mnt_child); | |
950 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
951 | } |
952 | return p; | |
953 | } | |
954 | ||
8f291889 AV |
955 | /** |
956 | * vfs_create_mount - Create a mount for a configured superblock | |
957 | * @fc: The configuration context with the superblock attached | |
958 | * | |
959 | * Create a mount to an already configured superblock. If necessary, the | |
960 | * caller should invoke vfs_get_tree() before calling this. | |
961 | * | |
962 | * Note that this does not attach the mount to anything. | |
963 | */ | |
964 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 965 | { |
b105e270 | 966 | struct mount *mnt; |
9d412a43 | 967 | |
8f291889 AV |
968 | if (!fc->root) |
969 | return ERR_PTR(-EINVAL); | |
9d412a43 | 970 | |
8f291889 | 971 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
972 | if (!mnt) |
973 | return ERR_PTR(-ENOMEM); | |
974 | ||
8f291889 | 975 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 976 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 977 | |
8f291889 AV |
978 | atomic_inc(&fc->root->d_sb->s_active); |
979 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
980 | mnt->mnt.mnt_root = dget(fc->root); | |
981 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
982 | mnt->mnt_parent = mnt; | |
9d412a43 | 983 | |
719ea2fb | 984 | lock_mount_hash(); |
8f291889 | 985 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 986 | unlock_mount_hash(); |
b105e270 | 987 | return &mnt->mnt; |
9d412a43 | 988 | } |
8f291889 AV |
989 | EXPORT_SYMBOL(vfs_create_mount); |
990 | ||
991 | struct vfsmount *fc_mount(struct fs_context *fc) | |
992 | { | |
993 | int err = vfs_get_tree(fc); | |
994 | if (!err) { | |
995 | up_write(&fc->root->d_sb->s_umount); | |
996 | return vfs_create_mount(fc); | |
997 | } | |
998 | return ERR_PTR(err); | |
999 | } | |
1000 | EXPORT_SYMBOL(fc_mount); | |
1001 | ||
9bc61ab1 DH |
1002 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1003 | int flags, const char *name, | |
1004 | void *data) | |
9d412a43 | 1005 | { |
9bc61ab1 | 1006 | struct fs_context *fc; |
8f291889 | 1007 | struct vfsmount *mnt; |
9bc61ab1 | 1008 | int ret = 0; |
9d412a43 AV |
1009 | |
1010 | if (!type) | |
3e1aeb00 | 1011 | return ERR_PTR(-EINVAL); |
9d412a43 | 1012 | |
9bc61ab1 DH |
1013 | fc = fs_context_for_mount(type, flags); |
1014 | if (IS_ERR(fc)) | |
1015 | return ERR_CAST(fc); | |
1016 | ||
3e1aeb00 DH |
1017 | if (name) |
1018 | ret = vfs_parse_fs_string(fc, "source", | |
1019 | name, strlen(name)); | |
9bc61ab1 DH |
1020 | if (!ret) |
1021 | ret = parse_monolithic_mount_data(fc, data); | |
1022 | if (!ret) | |
8f291889 AV |
1023 | mnt = fc_mount(fc); |
1024 | else | |
1025 | mnt = ERR_PTR(ret); | |
9d412a43 | 1026 | |
9bc61ab1 | 1027 | put_fs_context(fc); |
8f291889 | 1028 | return mnt; |
9d412a43 AV |
1029 | } |
1030 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1031 | ||
93faccbb EB |
1032 | struct vfsmount * |
1033 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1034 | const char *name, void *data) | |
1035 | { | |
1036 | /* Until it is worked out how to pass the user namespace | |
1037 | * through from the parent mount to the submount don't support | |
1038 | * unprivileged mounts with submounts. | |
1039 | */ | |
1040 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1041 | return ERR_PTR(-EPERM); | |
1042 | ||
e462ec50 | 1043 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1044 | } |
1045 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1046 | ||
87129cc0 | 1047 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1048 | int flag) |
1da177e4 | 1049 | { |
87129cc0 | 1050 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1051 | struct mount *mnt; |
1052 | int err; | |
1da177e4 | 1053 | |
be34d1a3 DH |
1054 | mnt = alloc_vfsmnt(old->mnt_devname); |
1055 | if (!mnt) | |
1056 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1057 | |
7a472ef4 | 1058 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1059 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1060 | else | |
1061 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1062 | |
be34d1a3 DH |
1063 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1064 | err = mnt_alloc_group_id(mnt); | |
1065 | if (err) | |
1066 | goto out_free; | |
1da177e4 | 1067 | } |
be34d1a3 | 1068 | |
16a34adb AV |
1069 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1070 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1071 | |
be34d1a3 | 1072 | atomic_inc(&sb->s_active); |
a6435940 CB |
1073 | mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); |
1074 | if (mnt->mnt.mnt_userns != &init_user_ns) | |
1075 | mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); | |
be34d1a3 DH |
1076 | mnt->mnt.mnt_sb = sb; |
1077 | mnt->mnt.mnt_root = dget(root); | |
1078 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1079 | mnt->mnt_parent = mnt; | |
719ea2fb | 1080 | lock_mount_hash(); |
be34d1a3 | 1081 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1082 | unlock_mount_hash(); |
be34d1a3 | 1083 | |
7a472ef4 EB |
1084 | if ((flag & CL_SLAVE) || |
1085 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1086 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1087 | mnt->mnt_master = old; | |
1088 | CLEAR_MNT_SHARED(mnt); | |
1089 | } else if (!(flag & CL_PRIVATE)) { | |
1090 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1091 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1092 | if (IS_MNT_SLAVE(old)) | |
1093 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1094 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1095 | } else { |
1096 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1097 | } |
1098 | if (flag & CL_MAKE_SHARED) | |
1099 | set_mnt_shared(mnt); | |
1100 | ||
1101 | /* stick the duplicate mount on the same expiry list | |
1102 | * as the original if that was on one */ | |
1103 | if (flag & CL_EXPIRE) { | |
1104 | if (!list_empty(&old->mnt_expire)) | |
1105 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1106 | } | |
1107 | ||
cb338d06 | 1108 | return mnt; |
719f5d7f MS |
1109 | |
1110 | out_free: | |
8ffcb32e | 1111 | mnt_free_id(mnt); |
719f5d7f | 1112 | free_vfsmnt(mnt); |
be34d1a3 | 1113 | return ERR_PTR(err); |
1da177e4 LT |
1114 | } |
1115 | ||
9ea459e1 AV |
1116 | static void cleanup_mnt(struct mount *mnt) |
1117 | { | |
56cbb429 AV |
1118 | struct hlist_node *p; |
1119 | struct mount *m; | |
9ea459e1 | 1120 | /* |
56cbb429 AV |
1121 | * The warning here probably indicates that somebody messed |
1122 | * up a mnt_want/drop_write() pair. If this happens, the | |
1123 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1124 | * The locking used to deal with mnt_count decrement provides barriers, |
1125 | * so mnt_get_writers() below is safe. | |
1126 | */ | |
1127 | WARN_ON(mnt_get_writers(mnt)); | |
1128 | if (unlikely(mnt->mnt_pins.first)) | |
1129 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1130 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1131 | hlist_del(&m->mnt_umount); | |
1132 | mntput(&m->mnt); | |
1133 | } | |
9ea459e1 AV |
1134 | fsnotify_vfsmount_delete(&mnt->mnt); |
1135 | dput(mnt->mnt.mnt_root); | |
1136 | deactivate_super(mnt->mnt.mnt_sb); | |
1137 | mnt_free_id(mnt); | |
1138 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1139 | } | |
1140 | ||
1141 | static void __cleanup_mnt(struct rcu_head *head) | |
1142 | { | |
1143 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1144 | } | |
1145 | ||
1146 | static LLIST_HEAD(delayed_mntput_list); | |
1147 | static void delayed_mntput(struct work_struct *unused) | |
1148 | { | |
1149 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1150 | struct mount *m, *t; |
9ea459e1 | 1151 | |
29785735 BP |
1152 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1153 | cleanup_mnt(m); | |
9ea459e1 AV |
1154 | } |
1155 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1156 | ||
900148dc | 1157 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1158 | { |
4edbe133 | 1159 | LIST_HEAD(list); |
edf7ddbf | 1160 | int count; |
4edbe133 | 1161 | |
48a066e7 | 1162 | rcu_read_lock(); |
9ea0a46c AV |
1163 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1164 | /* | |
1165 | * Since we don't do lock_mount_hash() here, | |
1166 | * ->mnt_ns can change under us. However, if it's | |
1167 | * non-NULL, then there's a reference that won't | |
1168 | * be dropped until after an RCU delay done after | |
1169 | * turning ->mnt_ns NULL. So if we observe it | |
1170 | * non-NULL under rcu_read_lock(), the reference | |
1171 | * we are dropping is not the final one. | |
1172 | */ | |
1173 | mnt_add_count(mnt, -1); | |
48a066e7 | 1174 | rcu_read_unlock(); |
f03c6599 | 1175 | return; |
b3e19d92 | 1176 | } |
719ea2fb | 1177 | lock_mount_hash(); |
119e1ef8 AV |
1178 | /* |
1179 | * make sure that if __legitimize_mnt() has not seen us grab | |
1180 | * mount_lock, we'll see their refcount increment here. | |
1181 | */ | |
1182 | smp_mb(); | |
9ea0a46c | 1183 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1184 | count = mnt_get_count(mnt); |
1185 | if (count != 0) { | |
1186 | WARN_ON(count < 0); | |
48a066e7 | 1187 | rcu_read_unlock(); |
719ea2fb | 1188 | unlock_mount_hash(); |
99b7db7b NP |
1189 | return; |
1190 | } | |
48a066e7 AV |
1191 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1192 | rcu_read_unlock(); | |
1193 | unlock_mount_hash(); | |
1194 | return; | |
1195 | } | |
1196 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1197 | rcu_read_unlock(); | |
962830df | 1198 | |
39f7c4db | 1199 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1200 | |
1201 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1202 | struct mount *p, *tmp; | |
1203 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1204 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1205 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1206 | } |
1207 | } | |
719ea2fb | 1208 | unlock_mount_hash(); |
4edbe133 | 1209 | shrink_dentry_list(&list); |
649a795a | 1210 | |
9ea459e1 AV |
1211 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1212 | struct task_struct *task = current; | |
1213 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1214 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1215 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1216 | return; |
1217 | } | |
1218 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1219 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1220 | return; | |
1221 | } | |
1222 | cleanup_mnt(mnt); | |
b3e19d92 | 1223 | } |
b3e19d92 NP |
1224 | |
1225 | void mntput(struct vfsmount *mnt) | |
1226 | { | |
1227 | if (mnt) { | |
863d684f | 1228 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1229 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1230 | if (unlikely(m->mnt_expiry_mark)) |
1231 | m->mnt_expiry_mark = 0; | |
1232 | mntput_no_expire(m); | |
b3e19d92 NP |
1233 | } |
1234 | } | |
1235 | EXPORT_SYMBOL(mntput); | |
1236 | ||
1237 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1238 | { | |
1239 | if (mnt) | |
83adc753 | 1240 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1241 | return mnt; |
1242 | } | |
1243 | EXPORT_SYMBOL(mntget); | |
1244 | ||
1f287bc4 RD |
1245 | /** |
1246 | * path_is_mountpoint() - Check if path is a mount in the current namespace. | |
1247 | * @path: path to check | |
c6609c0a IK |
1248 | * |
1249 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1250 | * not mounted in any namespace and that common case is handled inline. | |
1251 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1252 | * mounts using a given dentry in a different namespace. This function | |
1253 | * checks if the passed in path is a mountpoint rather than the dentry | |
1254 | * alone. | |
1255 | */ | |
1256 | bool path_is_mountpoint(const struct path *path) | |
1257 | { | |
1258 | unsigned seq; | |
1259 | bool res; | |
1260 | ||
1261 | if (!d_mountpoint(path->dentry)) | |
1262 | return false; | |
1263 | ||
1264 | rcu_read_lock(); | |
1265 | do { | |
1266 | seq = read_seqbegin(&mount_lock); | |
1267 | res = __path_is_mountpoint(path); | |
1268 | } while (read_seqretry(&mount_lock, seq)); | |
1269 | rcu_read_unlock(); | |
1270 | ||
1271 | return res; | |
1272 | } | |
1273 | EXPORT_SYMBOL(path_is_mountpoint); | |
1274 | ||
ca71cf71 | 1275 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1276 | { |
3064c356 AV |
1277 | struct mount *p; |
1278 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1279 | if (IS_ERR(p)) | |
1280 | return ERR_CAST(p); | |
1281 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1282 | return &p->mnt; | |
7b7b1ace | 1283 | } |
1da177e4 | 1284 | |
a1a2c409 | 1285 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1286 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1287 | struct list_head *p) | |
1288 | { | |
1289 | struct mount *mnt, *ret = NULL; | |
1290 | ||
1291 | lock_ns_list(ns); | |
1292 | list_for_each_continue(p, &ns->list) { | |
1293 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1294 | if (!mnt_is_cursor(mnt)) { | |
1295 | ret = mnt; | |
1296 | break; | |
1297 | } | |
1298 | } | |
1299 | unlock_ns_list(ns); | |
1300 | ||
1301 | return ret; | |
1302 | } | |
1303 | ||
0226f492 | 1304 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1305 | static void *m_start(struct seq_file *m, loff_t *pos) |
1306 | { | |
ede1bf0d | 1307 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1308 | struct list_head *prev; |
1da177e4 | 1309 | |
390c6843 | 1310 | down_read(&namespace_sem); |
9f6c61f9 MS |
1311 | if (!*pos) { |
1312 | prev = &p->ns->list; | |
1313 | } else { | |
1314 | prev = &p->cursor.mnt_list; | |
1315 | ||
1316 | /* Read after we'd reached the end? */ | |
1317 | if (list_empty(prev)) | |
1318 | return NULL; | |
c7999c36 AV |
1319 | } |
1320 | ||
9f6c61f9 | 1321 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1322 | } |
1323 | ||
1324 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1325 | { | |
ede1bf0d | 1326 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1327 | struct mount *mnt = v; |
b0765fb8 | 1328 | |
9f6c61f9 MS |
1329 | ++*pos; |
1330 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | static void m_stop(struct seq_file *m, void *v) | |
1334 | { | |
9f6c61f9 MS |
1335 | struct proc_mounts *p = m->private; |
1336 | struct mount *mnt = v; | |
1337 | ||
1338 | lock_ns_list(p->ns); | |
1339 | if (mnt) | |
1340 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1341 | else | |
1342 | list_del_init(&p->cursor.mnt_list); | |
1343 | unlock_ns_list(p->ns); | |
390c6843 | 1344 | up_read(&namespace_sem); |
1da177e4 LT |
1345 | } |
1346 | ||
0226f492 | 1347 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1348 | { |
ede1bf0d | 1349 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1350 | struct mount *r = v; |
0226f492 | 1351 | return p->show(m, &r->mnt); |
1da177e4 LT |
1352 | } |
1353 | ||
a1a2c409 | 1354 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1355 | .start = m_start, |
1356 | .next = m_next, | |
1357 | .stop = m_stop, | |
0226f492 | 1358 | .show = m_show, |
b4629fe2 | 1359 | }; |
9f6c61f9 MS |
1360 | |
1361 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1362 | { | |
1363 | down_read(&namespace_sem); | |
1364 | lock_ns_list(ns); | |
1365 | list_del(&cursor->mnt_list); | |
1366 | unlock_ns_list(ns); | |
1367 | up_read(&namespace_sem); | |
1368 | } | |
a1a2c409 | 1369 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1370 | |
1da177e4 LT |
1371 | /** |
1372 | * may_umount_tree - check if a mount tree is busy | |
1f287bc4 | 1373 | * @m: root of mount tree |
1da177e4 LT |
1374 | * |
1375 | * This is called to check if a tree of mounts has any | |
1376 | * open files, pwds, chroots or sub mounts that are | |
1377 | * busy. | |
1378 | */ | |
909b0a88 | 1379 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1380 | { |
909b0a88 | 1381 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1382 | int actual_refs = 0; |
1383 | int minimum_refs = 0; | |
315fc83e | 1384 | struct mount *p; |
909b0a88 | 1385 | BUG_ON(!m); |
1da177e4 | 1386 | |
b3e19d92 | 1387 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1388 | lock_mount_hash(); |
909b0a88 | 1389 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1390 | actual_refs += mnt_get_count(p); |
1da177e4 | 1391 | minimum_refs += 2; |
1da177e4 | 1392 | } |
719ea2fb | 1393 | unlock_mount_hash(); |
1da177e4 LT |
1394 | |
1395 | if (actual_refs > minimum_refs) | |
e3474a8e | 1396 | return 0; |
1da177e4 | 1397 | |
e3474a8e | 1398 | return 1; |
1da177e4 LT |
1399 | } |
1400 | ||
1401 | EXPORT_SYMBOL(may_umount_tree); | |
1402 | ||
1403 | /** | |
1404 | * may_umount - check if a mount point is busy | |
1405 | * @mnt: root of mount | |
1406 | * | |
1407 | * This is called to check if a mount point has any | |
1408 | * open files, pwds, chroots or sub mounts. If the | |
1409 | * mount has sub mounts this will return busy | |
1410 | * regardless of whether the sub mounts are busy. | |
1411 | * | |
1412 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1413 | * give false negatives. The main reason why it's here is that we need | |
1414 | * a non-destructive way to look for easily umountable filesystems. | |
1415 | */ | |
1416 | int may_umount(struct vfsmount *mnt) | |
1417 | { | |
e3474a8e | 1418 | int ret = 1; |
8ad08d8a | 1419 | down_read(&namespace_sem); |
719ea2fb | 1420 | lock_mount_hash(); |
1ab59738 | 1421 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1422 | ret = 0; |
719ea2fb | 1423 | unlock_mount_hash(); |
8ad08d8a | 1424 | up_read(&namespace_sem); |
a05964f3 | 1425 | return ret; |
1da177e4 LT |
1426 | } |
1427 | ||
1428 | EXPORT_SYMBOL(may_umount); | |
1429 | ||
97216be0 | 1430 | static void namespace_unlock(void) |
70fbcdf4 | 1431 | { |
a3b3c562 | 1432 | struct hlist_head head; |
56cbb429 AV |
1433 | struct hlist_node *p; |
1434 | struct mount *m; | |
4edbe133 | 1435 | LIST_HEAD(list); |
97216be0 | 1436 | |
a3b3c562 | 1437 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1438 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1439 | |
97216be0 AV |
1440 | up_write(&namespace_sem); |
1441 | ||
4edbe133 AV |
1442 | shrink_dentry_list(&list); |
1443 | ||
a3b3c562 EB |
1444 | if (likely(hlist_empty(&head))) |
1445 | return; | |
1446 | ||
22cb7405 | 1447 | synchronize_rcu_expedited(); |
48a066e7 | 1448 | |
56cbb429 AV |
1449 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1450 | hlist_del(&m->mnt_umount); | |
1451 | mntput(&m->mnt); | |
1452 | } | |
70fbcdf4 RP |
1453 | } |
1454 | ||
97216be0 | 1455 | static inline void namespace_lock(void) |
e3197d83 | 1456 | { |
97216be0 | 1457 | down_write(&namespace_sem); |
e3197d83 AV |
1458 | } |
1459 | ||
e819f152 EB |
1460 | enum umount_tree_flags { |
1461 | UMOUNT_SYNC = 1, | |
1462 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1463 | UMOUNT_CONNECTED = 4, |
e819f152 | 1464 | }; |
f2d0a123 EB |
1465 | |
1466 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1467 | { | |
1468 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1469 | if (how & UMOUNT_SYNC) | |
1470 | return true; | |
1471 | ||
1472 | /* A mount without a parent has nothing to be connected to */ | |
1473 | if (!mnt_has_parent(mnt)) | |
1474 | return true; | |
1475 | ||
1476 | /* Because the reference counting rules change when mounts are | |
1477 | * unmounted and connected, umounted mounts may not be | |
1478 | * connected to mounted mounts. | |
1479 | */ | |
1480 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1481 | return true; | |
1482 | ||
1483 | /* Has it been requested that the mount remain connected? */ | |
1484 | if (how & UMOUNT_CONNECTED) | |
1485 | return false; | |
1486 | ||
1487 | /* Is the mount locked such that it needs to remain connected? */ | |
1488 | if (IS_MNT_LOCKED(mnt)) | |
1489 | return false; | |
1490 | ||
1491 | /* By default disconnect the mount */ | |
1492 | return true; | |
1493 | } | |
1494 | ||
99b7db7b | 1495 | /* |
48a066e7 | 1496 | * mount_lock must be held |
99b7db7b NP |
1497 | * namespace_sem must be held for write |
1498 | */ | |
e819f152 | 1499 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1500 | { |
c003b26f | 1501 | LIST_HEAD(tmp_list); |
315fc83e | 1502 | struct mount *p; |
1da177e4 | 1503 | |
5d88457e EB |
1504 | if (how & UMOUNT_PROPAGATE) |
1505 | propagate_mount_unlock(mnt); | |
1506 | ||
c003b26f | 1507 | /* Gather the mounts to umount */ |
590ce4bc EB |
1508 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1509 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1510 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1511 | } |
1da177e4 | 1512 | |
411a938b | 1513 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1514 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1515 | list_del_init(&p->mnt_child); |
c003b26f | 1516 | } |
88b368f2 | 1517 | |
c003b26f | 1518 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1519 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1520 | propagate_umount(&tmp_list); |
a05964f3 | 1521 | |
c003b26f | 1522 | while (!list_empty(&tmp_list)) { |
d2921684 | 1523 | struct mnt_namespace *ns; |
ce07d891 | 1524 | bool disconnect; |
c003b26f | 1525 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1526 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1527 | list_del_init(&p->mnt_list); |
d2921684 EB |
1528 | ns = p->mnt_ns; |
1529 | if (ns) { | |
1530 | ns->mounts--; | |
1531 | __touch_mnt_namespace(ns); | |
1532 | } | |
143c8c91 | 1533 | p->mnt_ns = NULL; |
e819f152 | 1534 | if (how & UMOUNT_SYNC) |
48a066e7 | 1535 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1536 | |
f2d0a123 | 1537 | disconnect = disconnect_mount(p, how); |
676da58d | 1538 | if (mnt_has_parent(p)) { |
81b6b061 | 1539 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1540 | if (!disconnect) { |
1541 | /* Don't forget about p */ | |
1542 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1543 | } else { | |
1544 | umount_mnt(p); | |
1545 | } | |
7c4b93d8 | 1546 | } |
0f0afb1d | 1547 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1548 | if (disconnect) |
1549 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1550 | } |
1551 | } | |
1552 | ||
b54b9be7 | 1553 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1554 | |
8d0347f6 DH |
1555 | static int do_umount_root(struct super_block *sb) |
1556 | { | |
1557 | int ret = 0; | |
1558 | ||
1559 | down_write(&sb->s_umount); | |
1560 | if (!sb_rdonly(sb)) { | |
1561 | struct fs_context *fc; | |
1562 | ||
1563 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1564 | SB_RDONLY); | |
1565 | if (IS_ERR(fc)) { | |
1566 | ret = PTR_ERR(fc); | |
1567 | } else { | |
1568 | ret = parse_monolithic_mount_data(fc, NULL); | |
1569 | if (!ret) | |
1570 | ret = reconfigure_super(fc); | |
1571 | put_fs_context(fc); | |
1572 | } | |
1573 | } | |
1574 | up_write(&sb->s_umount); | |
1575 | return ret; | |
1576 | } | |
1577 | ||
1ab59738 | 1578 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1579 | { |
1ab59738 | 1580 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1581 | int retval; |
1582 | ||
1ab59738 | 1583 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1584 | if (retval) |
1585 | return retval; | |
1586 | ||
1587 | /* | |
1588 | * Allow userspace to request a mountpoint be expired rather than | |
1589 | * unmounting unconditionally. Unmount only happens if: | |
1590 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1591 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1592 | */ | |
1593 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1594 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1595 | flags & (MNT_FORCE | MNT_DETACH)) |
1596 | return -EINVAL; | |
1597 | ||
b3e19d92 NP |
1598 | /* |
1599 | * probably don't strictly need the lock here if we examined | |
1600 | * all race cases, but it's a slowpath. | |
1601 | */ | |
719ea2fb | 1602 | lock_mount_hash(); |
83adc753 | 1603 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1604 | unlock_mount_hash(); |
1da177e4 | 1605 | return -EBUSY; |
b3e19d92 | 1606 | } |
719ea2fb | 1607 | unlock_mount_hash(); |
1da177e4 | 1608 | |
863d684f | 1609 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1610 | return -EAGAIN; |
1611 | } | |
1612 | ||
1613 | /* | |
1614 | * If we may have to abort operations to get out of this | |
1615 | * mount, and they will themselves hold resources we must | |
1616 | * allow the fs to do things. In the Unix tradition of | |
1617 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1618 | * might fail to complete on the first run through as other tasks | |
1619 | * must return, and the like. Thats for the mount program to worry | |
1620 | * about for the moment. | |
1621 | */ | |
1622 | ||
42faad99 | 1623 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1624 | sb->s_op->umount_begin(sb); |
42faad99 | 1625 | } |
1da177e4 LT |
1626 | |
1627 | /* | |
1628 | * No sense to grab the lock for this test, but test itself looks | |
1629 | * somewhat bogus. Suggestions for better replacement? | |
1630 | * Ho-hum... In principle, we might treat that as umount + switch | |
1631 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1632 | * Actually it makes sense, especially if rootfs would contain a | |
1633 | * /reboot - static binary that would close all descriptors and | |
1634 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1635 | */ | |
1ab59738 | 1636 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1637 | /* |
1638 | * Special case for "unmounting" root ... | |
1639 | * we just try to remount it readonly. | |
1640 | */ | |
bc6155d1 | 1641 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1642 | return -EPERM; |
8d0347f6 | 1643 | return do_umount_root(sb); |
1da177e4 LT |
1644 | } |
1645 | ||
97216be0 | 1646 | namespace_lock(); |
719ea2fb | 1647 | lock_mount_hash(); |
1da177e4 | 1648 | |
25d202ed EB |
1649 | /* Recheck MNT_LOCKED with the locks held */ |
1650 | retval = -EINVAL; | |
1651 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1652 | goto out; | |
1653 | ||
1654 | event++; | |
48a066e7 | 1655 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1656 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1657 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1658 | retval = 0; |
48a066e7 AV |
1659 | } else { |
1660 | shrink_submounts(mnt); | |
1661 | retval = -EBUSY; | |
1662 | if (!propagate_mount_busy(mnt, 2)) { | |
1663 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1664 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1665 | retval = 0; |
1666 | } | |
1da177e4 | 1667 | } |
25d202ed | 1668 | out: |
719ea2fb | 1669 | unlock_mount_hash(); |
e3197d83 | 1670 | namespace_unlock(); |
1da177e4 LT |
1671 | return retval; |
1672 | } | |
1673 | ||
80b5dce8 EB |
1674 | /* |
1675 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1676 | * | |
1677 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1678 | * to an existing mountpoint, and wind up leaking the mount. | |
1679 | * detach_mounts allows lazily unmounting those mounts instead of | |
1680 | * leaking them. | |
1681 | * | |
1682 | * The caller may hold dentry->d_inode->i_mutex. | |
1683 | */ | |
1684 | void __detach_mounts(struct dentry *dentry) | |
1685 | { | |
1686 | struct mountpoint *mp; | |
1687 | struct mount *mnt; | |
1688 | ||
1689 | namespace_lock(); | |
3895dbf8 | 1690 | lock_mount_hash(); |
80b5dce8 | 1691 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1692 | if (!mp) |
80b5dce8 EB |
1693 | goto out_unlock; |
1694 | ||
e06b933e | 1695 | event++; |
80b5dce8 EB |
1696 | while (!hlist_empty(&mp->m_list)) { |
1697 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1698 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1699 | umount_mnt(mnt); |
56cbb429 | 1700 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1701 | } |
e0c9c0af | 1702 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1703 | } |
80b5dce8 EB |
1704 | put_mountpoint(mp); |
1705 | out_unlock: | |
3895dbf8 | 1706 | unlock_mount_hash(); |
80b5dce8 EB |
1707 | namespace_unlock(); |
1708 | } | |
1709 | ||
dd111b31 | 1710 | /* |
9b40bc90 AV |
1711 | * Is the caller allowed to modify his namespace? |
1712 | */ | |
1713 | static inline bool may_mount(void) | |
1714 | { | |
1715 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1716 | } | |
1717 | ||
f7e33bdb | 1718 | static void warn_mandlock(void) |
9e8925b6 | 1719 | { |
f7e33bdb JL |
1720 | pr_warn_once("=======================================================\n" |
1721 | "WARNING: The mand mount option has been deprecated and\n" | |
1722 | " and is ignored by this kernel. Remove the mand\n" | |
1723 | " option from the mount to silence this warning.\n" | |
1724 | "=======================================================\n"); | |
9e8925b6 JL |
1725 | } |
1726 | ||
25ccd24f | 1727 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1728 | { |
25ccd24f | 1729 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1730 | |
9b40bc90 AV |
1731 | if (!may_mount()) |
1732 | return -EPERM; | |
41525f56 | 1733 | if (path->dentry != path->mnt->mnt_root) |
25ccd24f | 1734 | return -EINVAL; |
143c8c91 | 1735 | if (!check_mnt(mnt)) |
25ccd24f | 1736 | return -EINVAL; |
25d202ed | 1737 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1738 | return -EINVAL; |
b2f5d4dc | 1739 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1740 | return -EPERM; |
1741 | return 0; | |
1742 | } | |
1743 | ||
a0a6df9a | 1744 | // caller is responsible for flags being sane |
25ccd24f CH |
1745 | int path_umount(struct path *path, int flags) |
1746 | { | |
1747 | struct mount *mnt = real_mount(path->mnt); | |
1748 | int ret; | |
1749 | ||
1750 | ret = can_umount(path, flags); | |
1751 | if (!ret) | |
1752 | ret = do_umount(mnt, flags); | |
1da177e4 | 1753 | |
429731b1 | 1754 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1755 | dput(path->dentry); |
900148dc | 1756 | mntput_no_expire(mnt); |
25ccd24f | 1757 | return ret; |
1da177e4 LT |
1758 | } |
1759 | ||
09267def | 1760 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1761 | { |
1762 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1763 | struct path path; | |
1764 | int ret; | |
1765 | ||
a0a6df9a AV |
1766 | // basic validity checks done first |
1767 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1768 | return -EINVAL; | |
1769 | ||
41525f56 CH |
1770 | if (!(flags & UMOUNT_NOFOLLOW)) |
1771 | lookup_flags |= LOOKUP_FOLLOW; | |
1772 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1773 | if (ret) | |
1774 | return ret; | |
1775 | return path_umount(&path, flags); | |
1776 | } | |
1777 | ||
3a18ef5c DB |
1778 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1779 | { | |
1780 | return ksys_umount(name, flags); | |
1781 | } | |
1782 | ||
1da177e4 LT |
1783 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1784 | ||
1785 | /* | |
b58fed8b | 1786 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1787 | */ |
bdc480e3 | 1788 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1789 | { |
3a18ef5c | 1790 | return ksys_umount(name, 0); |
1da177e4 LT |
1791 | } |
1792 | ||
1793 | #endif | |
1794 | ||
4ce5d2b1 | 1795 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1796 | { |
4ce5d2b1 | 1797 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1798 | return dentry->d_op == &ns_dentry_operations && |
1799 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1800 | } |
1801 | ||
213921f9 | 1802 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1803 | { |
1804 | return container_of(ns, struct mnt_namespace, ns); | |
1805 | } | |
1806 | ||
303cc571 CB |
1807 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1808 | { | |
1809 | return &mnt->ns; | |
1810 | } | |
1811 | ||
4ce5d2b1 EB |
1812 | static bool mnt_ns_loop(struct dentry *dentry) |
1813 | { | |
1814 | /* Could bind mounting the mount namespace inode cause a | |
1815 | * mount namespace loop? | |
1816 | */ | |
1817 | struct mnt_namespace *mnt_ns; | |
1818 | if (!is_mnt_ns_file(dentry)) | |
1819 | return false; | |
1820 | ||
f77c8014 | 1821 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1822 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1823 | } | |
1824 | ||
87129cc0 | 1825 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1826 | int flag) |
1da177e4 | 1827 | { |
84d17192 | 1828 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1829 | |
4ce5d2b1 EB |
1830 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1831 | return ERR_PTR(-EINVAL); | |
1832 | ||
1833 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1834 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1835 | |
36341f64 | 1836 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1837 | if (IS_ERR(q)) |
1838 | return q; | |
1839 | ||
a73324da | 1840 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1841 | |
1842 | p = mnt; | |
6b41d536 | 1843 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1844 | struct mount *s; |
7ec02ef1 | 1845 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1846 | continue; |
1847 | ||
909b0a88 | 1848 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1849 | if (!(flag & CL_COPY_UNBINDABLE) && |
1850 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1851 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1852 | /* Both unbindable and locked. */ | |
1853 | q = ERR_PTR(-EPERM); | |
1854 | goto out; | |
1855 | } else { | |
1856 | s = skip_mnt_tree(s); | |
1857 | continue; | |
1858 | } | |
4ce5d2b1 EB |
1859 | } |
1860 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1861 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1862 | s = skip_mnt_tree(s); |
1863 | continue; | |
1864 | } | |
0714a533 AV |
1865 | while (p != s->mnt_parent) { |
1866 | p = p->mnt_parent; | |
1867 | q = q->mnt_parent; | |
1da177e4 | 1868 | } |
87129cc0 | 1869 | p = s; |
84d17192 | 1870 | parent = q; |
87129cc0 | 1871 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1872 | if (IS_ERR(q)) |
1873 | goto out; | |
719ea2fb | 1874 | lock_mount_hash(); |
1a4eeaf2 | 1875 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1876 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1877 | unlock_mount_hash(); |
1da177e4 LT |
1878 | } |
1879 | } | |
1880 | return res; | |
be34d1a3 | 1881 | out: |
1da177e4 | 1882 | if (res) { |
719ea2fb | 1883 | lock_mount_hash(); |
e819f152 | 1884 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1885 | unlock_mount_hash(); |
1da177e4 | 1886 | } |
be34d1a3 | 1887 | return q; |
1da177e4 LT |
1888 | } |
1889 | ||
be34d1a3 DH |
1890 | /* Caller should check returned pointer for errors */ |
1891 | ||
ca71cf71 | 1892 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1893 | { |
cb338d06 | 1894 | struct mount *tree; |
97216be0 | 1895 | namespace_lock(); |
cd4a4017 EB |
1896 | if (!check_mnt(real_mount(path->mnt))) |
1897 | tree = ERR_PTR(-EINVAL); | |
1898 | else | |
1899 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1900 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1901 | namespace_unlock(); |
be34d1a3 | 1902 | if (IS_ERR(tree)) |
52e220d3 | 1903 | return ERR_CAST(tree); |
be34d1a3 | 1904 | return &tree->mnt; |
8aec0809 AV |
1905 | } |
1906 | ||
a07b2000 AV |
1907 | static void free_mnt_ns(struct mnt_namespace *); |
1908 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1909 | ||
1910 | void dissolve_on_fput(struct vfsmount *mnt) | |
1911 | { | |
1912 | struct mnt_namespace *ns; | |
1913 | namespace_lock(); | |
1914 | lock_mount_hash(); | |
1915 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1916 | if (ns) { |
1917 | if (is_anon_ns(ns)) | |
1918 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1919 | else | |
1920 | ns = NULL; | |
1921 | } | |
a07b2000 AV |
1922 | unlock_mount_hash(); |
1923 | namespace_unlock(); | |
44dfd84a DH |
1924 | if (ns) |
1925 | free_mnt_ns(ns); | |
a07b2000 AV |
1926 | } |
1927 | ||
8aec0809 AV |
1928 | void drop_collected_mounts(struct vfsmount *mnt) |
1929 | { | |
97216be0 | 1930 | namespace_lock(); |
719ea2fb | 1931 | lock_mount_hash(); |
9c8e0a1b | 1932 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1933 | unlock_mount_hash(); |
3ab6abee | 1934 | namespace_unlock(); |
8aec0809 AV |
1935 | } |
1936 | ||
427215d8 MS |
1937 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1938 | { | |
1939 | struct mount *child; | |
1940 | ||
1941 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1942 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1943 | continue; | |
1944 | ||
1945 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1946 | return true; | |
1947 | } | |
1948 | return false; | |
1949 | } | |
1950 | ||
c771d683 MS |
1951 | /** |
1952 | * clone_private_mount - create a private clone of a path | |
1f287bc4 | 1953 | * @path: path to clone |
c771d683 | 1954 | * |
1f287bc4 RD |
1955 | * This creates a new vfsmount, which will be the clone of @path. The new mount |
1956 | * will not be attached anywhere in the namespace and will be private (i.e. | |
1957 | * changes to the originating mount won't be propagated into this). | |
c771d683 MS |
1958 | * |
1959 | * Release with mntput(). | |
1960 | */ | |
ca71cf71 | 1961 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1962 | { |
1963 | struct mount *old_mnt = real_mount(path->mnt); | |
1964 | struct mount *new_mnt; | |
1965 | ||
427215d8 | 1966 | down_read(&namespace_sem); |
c771d683 | 1967 | if (IS_MNT_UNBINDABLE(old_mnt)) |
427215d8 MS |
1968 | goto invalid; |
1969 | ||
1970 | if (!check_mnt(old_mnt)) | |
1971 | goto invalid; | |
1972 | ||
1973 | if (has_locked_children(old_mnt, path->dentry)) | |
1974 | goto invalid; | |
c771d683 | 1975 | |
c771d683 | 1976 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
427215d8 MS |
1977 | up_read(&namespace_sem); |
1978 | ||
c771d683 MS |
1979 | if (IS_ERR(new_mnt)) |
1980 | return ERR_CAST(new_mnt); | |
1981 | ||
df820f8d MS |
1982 | /* Longterm mount to be removed by kern_unmount*() */ |
1983 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
1984 | ||
c771d683 | 1985 | return &new_mnt->mnt; |
427215d8 MS |
1986 | |
1987 | invalid: | |
1988 | up_read(&namespace_sem); | |
1989 | return ERR_PTR(-EINVAL); | |
c771d683 MS |
1990 | } |
1991 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1992 | ||
1f707137 AV |
1993 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1994 | struct vfsmount *root) | |
1995 | { | |
1a4eeaf2 | 1996 | struct mount *mnt; |
1f707137 AV |
1997 | int res = f(root, arg); |
1998 | if (res) | |
1999 | return res; | |
1a4eeaf2 AV |
2000 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
2001 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
2002 | if (res) |
2003 | return res; | |
2004 | } | |
2005 | return 0; | |
2006 | } | |
2007 | ||
3bd045cc AV |
2008 | static void lock_mnt_tree(struct mount *mnt) |
2009 | { | |
2010 | struct mount *p; | |
2011 | ||
2012 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2013 | int flags = p->mnt.mnt_flags; | |
2014 | /* Don't allow unprivileged users to change mount flags */ | |
2015 | flags |= MNT_LOCK_ATIME; | |
2016 | ||
2017 | if (flags & MNT_READONLY) | |
2018 | flags |= MNT_LOCK_READONLY; | |
2019 | ||
2020 | if (flags & MNT_NODEV) | |
2021 | flags |= MNT_LOCK_NODEV; | |
2022 | ||
2023 | if (flags & MNT_NOSUID) | |
2024 | flags |= MNT_LOCK_NOSUID; | |
2025 | ||
2026 | if (flags & MNT_NOEXEC) | |
2027 | flags |= MNT_LOCK_NOEXEC; | |
2028 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2029 | if (list_empty(&p->mnt_expire)) | |
2030 | flags |= MNT_LOCKED; | |
2031 | p->mnt.mnt_flags = flags; | |
2032 | } | |
2033 | } | |
2034 | ||
4b8b21f4 | 2035 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2036 | { |
315fc83e | 2037 | struct mount *p; |
719f5d7f | 2038 | |
909b0a88 | 2039 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2040 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2041 | mnt_release_group_id(p); |
719f5d7f MS |
2042 | } |
2043 | } | |
2044 | ||
4b8b21f4 | 2045 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2046 | { |
315fc83e | 2047 | struct mount *p; |
719f5d7f | 2048 | |
909b0a88 | 2049 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2050 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2051 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2052 | if (err) { |
4b8b21f4 | 2053 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2054 | return err; |
2055 | } | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | return 0; | |
2060 | } | |
2061 | ||
d2921684 EB |
2062 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2063 | { | |
2064 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
2065 | unsigned int mounts = 0, old, pending, sum; | |
2066 | struct mount *p; | |
2067 | ||
2068 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
2069 | mounts++; | |
2070 | ||
2071 | old = ns->mounts; | |
2072 | pending = ns->pending_mounts; | |
2073 | sum = old + pending; | |
2074 | if ((old > sum) || | |
2075 | (pending > sum) || | |
2076 | (max < sum) || | |
2077 | (mounts > (max - sum))) | |
2078 | return -ENOSPC; | |
2079 | ||
2080 | ns->pending_mounts = pending + mounts; | |
2081 | return 0; | |
2082 | } | |
2083 | ||
b90fa9ae RP |
2084 | /* |
2085 | * @source_mnt : mount tree to be attached | |
21444403 RP |
2086 | * @nd : place the mount tree @source_mnt is attached |
2087 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
2088 | * store the parent mount and mountpoint dentry. | |
2089 | * (done when source_mnt is moved) | |
b90fa9ae RP |
2090 | * |
2091 | * NOTE: in the table below explains the semantics when a source mount | |
2092 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2093 | * --------------------------------------------------------------------------- |
2094 | * | BIND MOUNT OPERATION | | |
2095 | * |************************************************************************** | |
2096 | * | source-->| shared | private | slave | unbindable | | |
2097 | * | dest | | | | | | |
2098 | * | | | | | | | | |
2099 | * | v | | | | | | |
2100 | * |************************************************************************** | |
2101 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2102 | * | | | | | | | |
2103 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2104 | * *************************************************************************** | |
b90fa9ae RP |
2105 | * A bind operation clones the source mount and mounts the clone on the |
2106 | * destination mount. | |
2107 | * | |
2108 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2109 | * tree of the destination mount and the cloned mount is added to | |
2110 | * the peer group of the source mount. | |
2111 | * (+) the cloned mount is created under the destination mount and is marked | |
2112 | * as shared. The cloned mount is added to the peer group of the source | |
2113 | * mount. | |
5afe0022 RP |
2114 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2115 | * of the destination mount and the cloned mount is made slave | |
2116 | * of the same master as that of the source mount. The cloned mount | |
2117 | * is marked as 'shared and slave'. | |
2118 | * (*) the cloned mount is made a slave of the same master as that of the | |
2119 | * source mount. | |
2120 | * | |
9676f0c6 RP |
2121 | * --------------------------------------------------------------------------- |
2122 | * | MOVE MOUNT OPERATION | | |
2123 | * |************************************************************************** | |
2124 | * | source-->| shared | private | slave | unbindable | | |
2125 | * | dest | | | | | | |
2126 | * | | | | | | | | |
2127 | * | v | | | | | | |
2128 | * |************************************************************************** | |
2129 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2130 | * | | | | | | | |
2131 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2132 | * *************************************************************************** | |
5afe0022 RP |
2133 | * |
2134 | * (+) the mount is moved to the destination. And is then propagated to | |
2135 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2136 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2137 | * (+++) the mount is moved to the destination and is then propagated to |
2138 | * all the mounts belonging to the destination mount's propagation tree. | |
2139 | * the mount is marked as 'shared and slave'. | |
2140 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2141 | * |
2142 | * if the source mount is a tree, the operations explained above is | |
2143 | * applied to each mount in the tree. | |
2144 | * Must be called without spinlocks held, since this function can sleep | |
2145 | * in allocations. | |
2146 | */ | |
0fb54e50 | 2147 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2148 | struct mount *dest_mnt, |
2149 | struct mountpoint *dest_mp, | |
2763d119 | 2150 | bool moving) |
b90fa9ae | 2151 | { |
3bd045cc | 2152 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2153 | HLIST_HEAD(tree_list); |
d2921684 | 2154 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2155 | struct mountpoint *smp; |
315fc83e | 2156 | struct mount *child, *p; |
38129a13 | 2157 | struct hlist_node *n; |
719f5d7f | 2158 | int err; |
b90fa9ae | 2159 | |
1064f874 EB |
2160 | /* Preallocate a mountpoint in case the new mounts need |
2161 | * to be tucked under other mounts. | |
2162 | */ | |
2163 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2164 | if (IS_ERR(smp)) | |
2165 | return PTR_ERR(smp); | |
2166 | ||
d2921684 | 2167 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2168 | if (!moving) { |
d2921684 EB |
2169 | err = count_mounts(ns, source_mnt); |
2170 | if (err) | |
2171 | goto out; | |
2172 | } | |
2173 | ||
fc7be130 | 2174 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2175 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2176 | if (err) |
2177 | goto out; | |
0b1b901b | 2178 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2179 | lock_mount_hash(); |
0b1b901b AV |
2180 | if (err) |
2181 | goto out_cleanup_ids; | |
909b0a88 | 2182 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2183 | set_mnt_shared(p); |
0b1b901b AV |
2184 | } else { |
2185 | lock_mount_hash(); | |
b90fa9ae | 2186 | } |
2763d119 AV |
2187 | if (moving) { |
2188 | unhash_mnt(source_mnt); | |
84d17192 | 2189 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2190 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2191 | } else { |
44dfd84a DH |
2192 | if (source_mnt->mnt_ns) { |
2193 | /* move from anon - the caller will destroy */ | |
2194 | list_del_init(&source_mnt->mnt_ns->list); | |
2195 | } | |
84d17192 | 2196 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2197 | commit_tree(source_mnt); |
21444403 | 2198 | } |
b90fa9ae | 2199 | |
38129a13 | 2200 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2201 | struct mount *q; |
38129a13 | 2202 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2203 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2204 | child->mnt_mountpoint); | |
2205 | if (q) | |
2206 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2207 | /* Notice when we are propagating across user namespaces */ |
2208 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2209 | lock_mnt_tree(child); | |
d728cf79 | 2210 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2211 | commit_tree(child); |
b90fa9ae | 2212 | } |
1064f874 | 2213 | put_mountpoint(smp); |
719ea2fb | 2214 | unlock_mount_hash(); |
99b7db7b | 2215 | |
b90fa9ae | 2216 | return 0; |
719f5d7f MS |
2217 | |
2218 | out_cleanup_ids: | |
f2ebb3a9 AV |
2219 | while (!hlist_empty(&tree_list)) { |
2220 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2221 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2222 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2223 | } |
2224 | unlock_mount_hash(); | |
0b1b901b | 2225 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2226 | out: |
d2921684 | 2227 | ns->pending_mounts = 0; |
1064f874 EB |
2228 | |
2229 | read_seqlock_excl(&mount_lock); | |
2230 | put_mountpoint(smp); | |
2231 | read_sequnlock_excl(&mount_lock); | |
2232 | ||
719f5d7f | 2233 | return err; |
b90fa9ae RP |
2234 | } |
2235 | ||
84d17192 | 2236 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2237 | { |
2238 | struct vfsmount *mnt; | |
84d17192 | 2239 | struct dentry *dentry = path->dentry; |
b12cea91 | 2240 | retry: |
5955102c | 2241 | inode_lock(dentry->d_inode); |
84d17192 | 2242 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2243 | inode_unlock(dentry->d_inode); |
84d17192 | 2244 | return ERR_PTR(-ENOENT); |
b12cea91 | 2245 | } |
97216be0 | 2246 | namespace_lock(); |
b12cea91 | 2247 | mnt = lookup_mnt(path); |
84d17192 | 2248 | if (likely(!mnt)) { |
3895dbf8 | 2249 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2250 | if (IS_ERR(mp)) { |
97216be0 | 2251 | namespace_unlock(); |
5955102c | 2252 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2253 | return mp; |
2254 | } | |
2255 | return mp; | |
2256 | } | |
97216be0 | 2257 | namespace_unlock(); |
5955102c | 2258 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2259 | path_put(path); |
2260 | path->mnt = mnt; | |
84d17192 | 2261 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2262 | goto retry; |
2263 | } | |
2264 | ||
84d17192 | 2265 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2266 | { |
84d17192 | 2267 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2268 | |
2269 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2270 | put_mountpoint(where); |
3895dbf8 EB |
2271 | read_sequnlock_excl(&mount_lock); |
2272 | ||
328e6d90 | 2273 | namespace_unlock(); |
5955102c | 2274 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2275 | } |
2276 | ||
84d17192 | 2277 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2278 | { |
e462ec50 | 2279 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2280 | return -EINVAL; |
2281 | ||
e36cb0b8 DH |
2282 | if (d_is_dir(mp->m_dentry) != |
2283 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2284 | return -ENOTDIR; |
2285 | ||
2763d119 | 2286 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2287 | } |
2288 | ||
7a2e8a8f VA |
2289 | /* |
2290 | * Sanity check the flags to change_mnt_propagation. | |
2291 | */ | |
2292 | ||
e462ec50 | 2293 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2294 | { |
e462ec50 | 2295 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2296 | |
2297 | /* Fail if any non-propagation flags are set */ | |
2298 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2299 | return 0; | |
2300 | /* Only one propagation flag should be set */ | |
2301 | if (!is_power_of_2(type)) | |
2302 | return 0; | |
2303 | return type; | |
2304 | } | |
2305 | ||
07b20889 RP |
2306 | /* |
2307 | * recursively change the type of the mountpoint. | |
2308 | */ | |
e462ec50 | 2309 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2310 | { |
315fc83e | 2311 | struct mount *m; |
4b8b21f4 | 2312 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2313 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2314 | int type; |
719f5d7f | 2315 | int err = 0; |
07b20889 | 2316 | |
2d92ab3c | 2317 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2318 | return -EINVAL; |
2319 | ||
e462ec50 | 2320 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2321 | if (!type) |
2322 | return -EINVAL; | |
2323 | ||
97216be0 | 2324 | namespace_lock(); |
719f5d7f MS |
2325 | if (type == MS_SHARED) { |
2326 | err = invent_group_ids(mnt, recurse); | |
2327 | if (err) | |
2328 | goto out_unlock; | |
2329 | } | |
2330 | ||
719ea2fb | 2331 | lock_mount_hash(); |
909b0a88 | 2332 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2333 | change_mnt_propagation(m, type); |
719ea2fb | 2334 | unlock_mount_hash(); |
719f5d7f MS |
2335 | |
2336 | out_unlock: | |
97216be0 | 2337 | namespace_unlock(); |
719f5d7f | 2338 | return err; |
07b20889 RP |
2339 | } |
2340 | ||
a07b2000 AV |
2341 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2342 | { | |
2343 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2344 | ||
2345 | if (IS_MNT_UNBINDABLE(old)) | |
2346 | return mnt; | |
2347 | ||
2348 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2349 | return mnt; | |
2350 | ||
2351 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2352 | return mnt; | |
2353 | ||
2354 | if (recurse) | |
2355 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2356 | else | |
2357 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2358 | ||
2359 | if (!IS_ERR(mnt)) | |
2360 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2361 | ||
2362 | return mnt; | |
2363 | } | |
2364 | ||
1da177e4 LT |
2365 | /* |
2366 | * do loopback mount. | |
2367 | */ | |
808d4e3c | 2368 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2369 | int recurse) |
1da177e4 | 2370 | { |
2d92ab3c | 2371 | struct path old_path; |
a07b2000 | 2372 | struct mount *mnt = NULL, *parent; |
84d17192 | 2373 | struct mountpoint *mp; |
57eccb83 | 2374 | int err; |
1da177e4 LT |
2375 | if (!old_name || !*old_name) |
2376 | return -EINVAL; | |
815d405c | 2377 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2378 | if (err) |
2379 | return err; | |
2380 | ||
8823c079 | 2381 | err = -EINVAL; |
4ce5d2b1 | 2382 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2383 | goto out; |
8823c079 | 2384 | |
84d17192 | 2385 | mp = lock_mount(path); |
a07b2000 AV |
2386 | if (IS_ERR(mp)) { |
2387 | err = PTR_ERR(mp); | |
b12cea91 | 2388 | goto out; |
a07b2000 | 2389 | } |
b12cea91 | 2390 | |
84d17192 | 2391 | parent = real_mount(path->mnt); |
e149ed2b AV |
2392 | if (!check_mnt(parent)) |
2393 | goto out2; | |
2394 | ||
a07b2000 | 2395 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2396 | if (IS_ERR(mnt)) { |
2397 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2398 | goto out2; |
be34d1a3 | 2399 | } |
ccd48bc7 | 2400 | |
84d17192 | 2401 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2402 | if (err) { |
719ea2fb | 2403 | lock_mount_hash(); |
e819f152 | 2404 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2405 | unlock_mount_hash(); |
5b83d2c5 | 2406 | } |
b12cea91 | 2407 | out2: |
84d17192 | 2408 | unlock_mount(mp); |
ccd48bc7 | 2409 | out: |
2d92ab3c | 2410 | path_put(&old_path); |
1da177e4 LT |
2411 | return err; |
2412 | } | |
2413 | ||
a07b2000 AV |
2414 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2415 | { | |
2416 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2417 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2418 | struct mount *mnt, *p; | |
2419 | struct file *file; | |
2420 | ||
2421 | if (IS_ERR(ns)) | |
2422 | return ERR_CAST(ns); | |
2423 | ||
2424 | namespace_lock(); | |
2425 | mnt = __do_loopback(path, recursive); | |
2426 | if (IS_ERR(mnt)) { | |
2427 | namespace_unlock(); | |
2428 | free_mnt_ns(ns); | |
2429 | return ERR_CAST(mnt); | |
2430 | } | |
2431 | ||
2432 | lock_mount_hash(); | |
2433 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2434 | p->mnt_ns = ns; | |
2435 | ns->mounts++; | |
2436 | } | |
2437 | ns->root = mnt; | |
2438 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2439 | mntget(&mnt->mnt); | |
2440 | unlock_mount_hash(); | |
2441 | namespace_unlock(); | |
2442 | ||
2443 | mntput(path->mnt); | |
2444 | path->mnt = &mnt->mnt; | |
2445 | file = dentry_open(path, O_PATH, current_cred()); | |
2446 | if (IS_ERR(file)) | |
2447 | dissolve_on_fput(path->mnt); | |
2448 | else | |
2449 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2450 | return file; | |
2451 | } | |
2452 | ||
2658ce09 | 2453 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2454 | { |
2455 | struct file *file; | |
2456 | struct path path; | |
2457 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2458 | bool detached = flags & OPEN_TREE_CLONE; | |
2459 | int error; | |
2460 | int fd; | |
2461 | ||
2462 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2463 | ||
2464 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2465 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2466 | OPEN_TREE_CLOEXEC)) | |
2467 | return -EINVAL; | |
2468 | ||
2469 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2470 | return -EINVAL; | |
2471 | ||
2472 | if (flags & AT_NO_AUTOMOUNT) | |
2473 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2474 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2475 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2476 | if (flags & AT_EMPTY_PATH) | |
2477 | lookup_flags |= LOOKUP_EMPTY; | |
2478 | ||
2479 | if (detached && !may_mount()) | |
2480 | return -EPERM; | |
2481 | ||
2482 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2483 | if (fd < 0) | |
2484 | return fd; | |
2485 | ||
2486 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2487 | if (unlikely(error)) { | |
2488 | file = ERR_PTR(error); | |
2489 | } else { | |
2490 | if (detached) | |
2491 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2492 | else | |
2493 | file = dentry_open(&path, O_PATH, current_cred()); | |
2494 | path_put(&path); | |
2495 | } | |
2496 | if (IS_ERR(file)) { | |
2497 | put_unused_fd(fd); | |
2498 | return PTR_ERR(file); | |
2499 | } | |
2500 | fd_install(fd, file); | |
2501 | return fd; | |
2502 | } | |
2503 | ||
43f5e655 DH |
2504 | /* |
2505 | * Don't allow locked mount flags to be cleared. | |
2506 | * | |
2507 | * No locks need to be held here while testing the various MNT_LOCK | |
2508 | * flags because those flags can never be cleared once they are set. | |
2509 | */ | |
2510 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2511 | { |
43f5e655 DH |
2512 | unsigned int fl = mnt->mnt.mnt_flags; |
2513 | ||
2514 | if ((fl & MNT_LOCK_READONLY) && | |
2515 | !(mnt_flags & MNT_READONLY)) | |
2516 | return false; | |
2517 | ||
2518 | if ((fl & MNT_LOCK_NODEV) && | |
2519 | !(mnt_flags & MNT_NODEV)) | |
2520 | return false; | |
2521 | ||
2522 | if ((fl & MNT_LOCK_NOSUID) && | |
2523 | !(mnt_flags & MNT_NOSUID)) | |
2524 | return false; | |
2525 | ||
2526 | if ((fl & MNT_LOCK_NOEXEC) && | |
2527 | !(mnt_flags & MNT_NOEXEC)) | |
2528 | return false; | |
2529 | ||
2530 | if ((fl & MNT_LOCK_ATIME) && | |
2531 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2532 | return false; | |
2e4b7fcd | 2533 | |
43f5e655 DH |
2534 | return true; |
2535 | } | |
2536 | ||
2537 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2538 | { |
43f5e655 | 2539 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2540 | |
43f5e655 | 2541 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2542 | return 0; |
2543 | ||
2544 | if (readonly_request) | |
43f5e655 DH |
2545 | return mnt_make_readonly(mnt); |
2546 | ||
68847c94 CB |
2547 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2548 | return 0; | |
43f5e655 DH |
2549 | } |
2550 | ||
43f5e655 DH |
2551 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2552 | { | |
43f5e655 DH |
2553 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2554 | mnt->mnt.mnt_flags = mnt_flags; | |
2555 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2556 | } |
2557 | ||
f8b92ba6 DD |
2558 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2559 | { | |
2560 | struct super_block *sb = mnt->mnt_sb; | |
2561 | ||
2562 | if (!__mnt_is_readonly(mnt) && | |
2563 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { | |
2564 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2565 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
2566 | struct tm tm; | |
2567 | ||
2568 | time64_to_tm(sb->s_time_max, 0, &tm); | |
2569 | ||
0ecee669 EB |
2570 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
2571 | sb->s_type->name, | |
2572 | is_mounted(mnt) ? "remounted" : "mounted", | |
2573 | mntpath, | |
f8b92ba6 DD |
2574 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
2575 | ||
2576 | free_page((unsigned long)buf); | |
2577 | } | |
2578 | } | |
2579 | ||
43f5e655 DH |
2580 | /* |
2581 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2582 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2583 | * to mount(2). | |
2584 | */ | |
2585 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2586 | { | |
2587 | struct super_block *sb = path->mnt->mnt_sb; | |
2588 | struct mount *mnt = real_mount(path->mnt); | |
2589 | int ret; | |
2590 | ||
2591 | if (!check_mnt(mnt)) | |
2592 | return -EINVAL; | |
2593 | ||
2594 | if (path->dentry != mnt->mnt.mnt_root) | |
2595 | return -EINVAL; | |
2596 | ||
2597 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2598 | return -EPERM; | |
2599 | ||
e58ace1a CB |
2600 | /* |
2601 | * We're only checking whether the superblock is read-only not | |
2602 | * changing it, so only take down_read(&sb->s_umount). | |
2603 | */ | |
2604 | down_read(&sb->s_umount); | |
68847c94 | 2605 | lock_mount_hash(); |
43f5e655 DH |
2606 | ret = change_mount_ro_state(mnt, mnt_flags); |
2607 | if (ret == 0) | |
2608 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2609 | unlock_mount_hash(); |
e58ace1a | 2610 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2611 | |
2612 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2613 | ||
43f5e655 | 2614 | return ret; |
2e4b7fcd DH |
2615 | } |
2616 | ||
1da177e4 LT |
2617 | /* |
2618 | * change filesystem flags. dir should be a physical root of filesystem. | |
2619 | * If you've mounted a non-root directory somewhere and want to do remount | |
2620 | * on it - tough luck. | |
2621 | */ | |
e462ec50 DH |
2622 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2623 | int mnt_flags, void *data) | |
1da177e4 LT |
2624 | { |
2625 | int err; | |
2d92ab3c | 2626 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2627 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2628 | struct fs_context *fc; |
1da177e4 | 2629 | |
143c8c91 | 2630 | if (!check_mnt(mnt)) |
1da177e4 LT |
2631 | return -EINVAL; |
2632 | ||
2d92ab3c | 2633 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2634 | return -EINVAL; |
2635 | ||
43f5e655 | 2636 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2637 | return -EPERM; |
9566d674 | 2638 | |
8d0347f6 DH |
2639 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2640 | if (IS_ERR(fc)) | |
2641 | return PTR_ERR(fc); | |
ff36fe2c | 2642 | |
b330966f | 2643 | fc->oldapi = true; |
8d0347f6 DH |
2644 | err = parse_monolithic_mount_data(fc, data); |
2645 | if (!err) { | |
2646 | down_write(&sb->s_umount); | |
2647 | err = -EPERM; | |
2648 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2649 | err = reconfigure_super(fc); | |
68847c94 CB |
2650 | if (!err) { |
2651 | lock_mount_hash(); | |
8d0347f6 | 2652 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2653 | unlock_mount_hash(); |
2654 | } | |
8d0347f6 DH |
2655 | } |
2656 | up_write(&sb->s_umount); | |
0e55a7cc | 2657 | } |
f8b92ba6 DD |
2658 | |
2659 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2660 | ||
8d0347f6 | 2661 | put_fs_context(fc); |
1da177e4 LT |
2662 | return err; |
2663 | } | |
2664 | ||
cbbe362c | 2665 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2666 | { |
315fc83e | 2667 | struct mount *p; |
909b0a88 | 2668 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2669 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2670 | return 1; |
2671 | } | |
2672 | return 0; | |
2673 | } | |
2674 | ||
44dfd84a DH |
2675 | /* |
2676 | * Check that there aren't references to earlier/same mount namespaces in the | |
2677 | * specified subtree. Such references can act as pins for mount namespaces | |
2678 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2679 | * cycles to be made. | |
2680 | */ | |
2681 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2682 | { | |
2683 | struct mount *p; | |
2684 | bool ret = false; | |
2685 | ||
2686 | lock_mount_hash(); | |
2687 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2688 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2689 | goto out; | |
2690 | ||
2691 | ret = true; | |
2692 | out: | |
2693 | unlock_mount_hash(); | |
2694 | return ret; | |
2695 | } | |
2696 | ||
2db154b3 | 2697 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2698 | { |
44dfd84a | 2699 | struct mnt_namespace *ns; |
676da58d | 2700 | struct mount *p; |
0fb54e50 | 2701 | struct mount *old; |
2763d119 AV |
2702 | struct mount *parent; |
2703 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2704 | int err; |
44dfd84a | 2705 | bool attached; |
1da177e4 | 2706 | |
2db154b3 | 2707 | mp = lock_mount(new_path); |
84d17192 | 2708 | if (IS_ERR(mp)) |
2db154b3 | 2709 | return PTR_ERR(mp); |
cc53ce53 | 2710 | |
2db154b3 DH |
2711 | old = real_mount(old_path->mnt); |
2712 | p = real_mount(new_path->mnt); | |
2763d119 | 2713 | parent = old->mnt_parent; |
44dfd84a | 2714 | attached = mnt_has_parent(old); |
2763d119 | 2715 | old_mp = old->mnt_mp; |
44dfd84a | 2716 | ns = old->mnt_ns; |
143c8c91 | 2717 | |
1da177e4 | 2718 | err = -EINVAL; |
44dfd84a DH |
2719 | /* The mountpoint must be in our namespace. */ |
2720 | if (!check_mnt(p)) | |
2db154b3 | 2721 | goto out; |
1da177e4 | 2722 | |
570d7a98 EB |
2723 | /* The thing moved must be mounted... */ |
2724 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2725 | goto out; |
2726 | ||
570d7a98 EB |
2727 | /* ... and either ours or the root of anon namespace */ |
2728 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2729 | goto out; |
5ff9d8a6 | 2730 | |
2db154b3 DH |
2731 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2732 | goto out; | |
1da177e4 | 2733 | |
2db154b3 DH |
2734 | if (old_path->dentry != old_path->mnt->mnt_root) |
2735 | goto out; | |
1da177e4 | 2736 | |
2db154b3 DH |
2737 | if (d_is_dir(new_path->dentry) != |
2738 | d_is_dir(old_path->dentry)) | |
2739 | goto out; | |
21444403 RP |
2740 | /* |
2741 | * Don't move a mount residing in a shared parent. | |
2742 | */ | |
2763d119 | 2743 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2744 | goto out; |
9676f0c6 RP |
2745 | /* |
2746 | * Don't move a mount tree containing unbindable mounts to a destination | |
2747 | * mount which is shared. | |
2748 | */ | |
fc7be130 | 2749 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2750 | goto out; |
1da177e4 | 2751 | err = -ELOOP; |
44dfd84a DH |
2752 | if (!check_for_nsfs_mounts(old)) |
2753 | goto out; | |
fc7be130 | 2754 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2755 | if (p == old) |
2db154b3 | 2756 | goto out; |
1da177e4 | 2757 | |
2db154b3 | 2758 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2759 | attached); |
4ac91378 | 2760 | if (err) |
2db154b3 | 2761 | goto out; |
1da177e4 LT |
2762 | |
2763 | /* if the mount is moved, it should no longer be expire | |
2764 | * automatically */ | |
6776db3d | 2765 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2766 | if (attached) |
2767 | put_mountpoint(old_mp); | |
1da177e4 | 2768 | out: |
2db154b3 | 2769 | unlock_mount(mp); |
44dfd84a | 2770 | if (!err) { |
2763d119 AV |
2771 | if (attached) |
2772 | mntput_no_expire(parent); | |
2773 | else | |
44dfd84a DH |
2774 | free_mnt_ns(ns); |
2775 | } | |
2db154b3 DH |
2776 | return err; |
2777 | } | |
2778 | ||
2779 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2780 | { | |
2781 | struct path old_path; | |
2782 | int err; | |
2783 | ||
2784 | if (!old_name || !*old_name) | |
2785 | return -EINVAL; | |
2786 | ||
2787 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2788 | if (err) | |
2789 | return err; | |
2790 | ||
2791 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2792 | path_put(&old_path); |
1da177e4 LT |
2793 | return err; |
2794 | } | |
2795 | ||
9d412a43 AV |
2796 | /* |
2797 | * add a mount into a namespace's mount tree | |
2798 | */ | |
8f11538e AV |
2799 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
2800 | struct path *path, int mnt_flags) | |
9d412a43 | 2801 | { |
8f11538e | 2802 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 2803 | |
f2ebb3a9 | 2804 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2805 | |
84d17192 | 2806 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2807 | /* that's acceptable only for automounts done in private ns */ |
2808 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 2809 | return -EINVAL; |
156cacb1 | 2810 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 2811 | if (!parent->mnt_ns) |
8f11538e | 2812 | return -EINVAL; |
156cacb1 | 2813 | } |
9d412a43 AV |
2814 | |
2815 | /* Refuse the same filesystem on the same mount point */ | |
95bc5f25 | 2816 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 | 2817 | path->mnt->mnt_root == path->dentry) |
8f11538e | 2818 | return -EBUSY; |
9d412a43 | 2819 | |
e36cb0b8 | 2820 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 2821 | return -EINVAL; |
9d412a43 | 2822 | |
95bc5f25 | 2823 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 2824 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 2825 | } |
b1e75df4 | 2826 | |
132e4608 DH |
2827 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2828 | ||
2829 | /* | |
2830 | * Create a new mount using a superblock configuration and request it | |
2831 | * be added to the namespace tree. | |
2832 | */ | |
2833 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2834 | unsigned int mnt_flags) | |
2835 | { | |
2836 | struct vfsmount *mnt; | |
8f11538e | 2837 | struct mountpoint *mp; |
132e4608 DH |
2838 | struct super_block *sb = fc->root->d_sb; |
2839 | int error; | |
2840 | ||
c9ce29ed AV |
2841 | error = security_sb_kern_mount(sb); |
2842 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2843 | error = -EPERM; | |
2844 | ||
2845 | if (unlikely(error)) { | |
2846 | fc_drop_locked(fc); | |
2847 | return error; | |
132e4608 DH |
2848 | } |
2849 | ||
2850 | up_write(&sb->s_umount); | |
2851 | ||
2852 | mnt = vfs_create_mount(fc); | |
2853 | if (IS_ERR(mnt)) | |
2854 | return PTR_ERR(mnt); | |
2855 | ||
f8b92ba6 DD |
2856 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
2857 | ||
8f11538e AV |
2858 | mp = lock_mount(mountpoint); |
2859 | if (IS_ERR(mp)) { | |
2860 | mntput(mnt); | |
2861 | return PTR_ERR(mp); | |
2862 | } | |
2863 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
2864 | unlock_mount(mp); | |
0ecee669 EB |
2865 | if (error < 0) |
2866 | mntput(mnt); | |
132e4608 DH |
2867 | return error; |
2868 | } | |
1b852bce | 2869 | |
1da177e4 LT |
2870 | /* |
2871 | * create a new mount for userspace and request it to be added into the | |
2872 | * namespace's tree | |
2873 | */ | |
e462ec50 | 2874 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2875 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2876 | { |
0c55cfc4 | 2877 | struct file_system_type *type; |
a0c9a8b8 AV |
2878 | struct fs_context *fc; |
2879 | const char *subtype = NULL; | |
2880 | int err = 0; | |
1da177e4 | 2881 | |
0c55cfc4 | 2882 | if (!fstype) |
1da177e4 LT |
2883 | return -EINVAL; |
2884 | ||
0c55cfc4 EB |
2885 | type = get_fs_type(fstype); |
2886 | if (!type) | |
2887 | return -ENODEV; | |
2888 | ||
a0c9a8b8 AV |
2889 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2890 | subtype = strchr(fstype, '.'); | |
2891 | if (subtype) { | |
2892 | subtype++; | |
2893 | if (!*subtype) { | |
2894 | put_filesystem(type); | |
2895 | return -EINVAL; | |
2896 | } | |
a0c9a8b8 AV |
2897 | } |
2898 | } | |
0c55cfc4 | 2899 | |
a0c9a8b8 | 2900 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2901 | put_filesystem(type); |
a0c9a8b8 AV |
2902 | if (IS_ERR(fc)) |
2903 | return PTR_ERR(fc); | |
2904 | ||
3e1aeb00 DH |
2905 | if (subtype) |
2906 | err = vfs_parse_fs_string(fc, "subtype", | |
2907 | subtype, strlen(subtype)); | |
2908 | if (!err && name) | |
2909 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2910 | if (!err) |
2911 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
2912 | if (!err && !mount_capable(fc)) |
2913 | err = -EPERM; | |
a0c9a8b8 AV |
2914 | if (!err) |
2915 | err = vfs_get_tree(fc); | |
132e4608 DH |
2916 | if (!err) |
2917 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2918 | |
a0c9a8b8 | 2919 | put_fs_context(fc); |
15f9a3f3 | 2920 | return err; |
1da177e4 LT |
2921 | } |
2922 | ||
19a167af AV |
2923 | int finish_automount(struct vfsmount *m, struct path *path) |
2924 | { | |
26df6034 | 2925 | struct dentry *dentry = path->dentry; |
8f11538e | 2926 | struct mountpoint *mp; |
25e195aa | 2927 | struct mount *mnt; |
19a167af | 2928 | int err; |
25e195aa AV |
2929 | |
2930 | if (!m) | |
2931 | return 0; | |
2932 | if (IS_ERR(m)) | |
2933 | return PTR_ERR(m); | |
2934 | ||
2935 | mnt = real_mount(m); | |
19a167af AV |
2936 | /* The new mount record should have at least 2 refs to prevent it being |
2937 | * expired before we get a chance to add it | |
2938 | */ | |
6776db3d | 2939 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2940 | |
2941 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 2942 | m->mnt_root == dentry) { |
b1e75df4 | 2943 | err = -ELOOP; |
26df6034 | 2944 | goto discard; |
19a167af AV |
2945 | } |
2946 | ||
26df6034 AV |
2947 | /* |
2948 | * we don't want to use lock_mount() - in this case finding something | |
2949 | * that overmounts our mountpoint to be means "quitely drop what we've | |
2950 | * got", not "try to mount it on top". | |
2951 | */ | |
2952 | inode_lock(dentry->d_inode); | |
2953 | namespace_lock(); | |
2954 | if (unlikely(cant_mount(dentry))) { | |
2955 | err = -ENOENT; | |
2956 | goto discard_locked; | |
2957 | } | |
2958 | rcu_read_lock(); | |
2959 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { | |
2960 | rcu_read_unlock(); | |
2961 | err = 0; | |
2962 | goto discard_locked; | |
2963 | } | |
2964 | rcu_read_unlock(); | |
2965 | mp = get_mountpoint(dentry); | |
8f11538e AV |
2966 | if (IS_ERR(mp)) { |
2967 | err = PTR_ERR(mp); | |
26df6034 | 2968 | goto discard_locked; |
8f11538e | 2969 | } |
26df6034 | 2970 | |
8f11538e AV |
2971 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
2972 | unlock_mount(mp); | |
26df6034 AV |
2973 | if (unlikely(err)) |
2974 | goto discard; | |
2975 | mntput(m); | |
2976 | return 0; | |
2977 | ||
2978 | discard_locked: | |
2979 | namespace_unlock(); | |
2980 | inode_unlock(dentry->d_inode); | |
2981 | discard: | |
b1e75df4 | 2982 | /* remove m from any expiration list it may be on */ |
6776db3d | 2983 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2984 | namespace_lock(); |
6776db3d | 2985 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2986 | namespace_unlock(); |
19a167af | 2987 | } |
b1e75df4 AV |
2988 | mntput(m); |
2989 | mntput(m); | |
19a167af AV |
2990 | return err; |
2991 | } | |
2992 | ||
ea5b778a DH |
2993 | /** |
2994 | * mnt_set_expiry - Put a mount on an expiration list | |
2995 | * @mnt: The mount to list. | |
2996 | * @expiry_list: The list to add the mount to. | |
2997 | */ | |
2998 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2999 | { | |
97216be0 | 3000 | namespace_lock(); |
ea5b778a | 3001 | |
6776db3d | 3002 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 3003 | |
97216be0 | 3004 | namespace_unlock(); |
ea5b778a DH |
3005 | } |
3006 | EXPORT_SYMBOL(mnt_set_expiry); | |
3007 | ||
1da177e4 LT |
3008 | /* |
3009 | * process a list of expirable mountpoints with the intent of discarding any | |
3010 | * mountpoints that aren't in use and haven't been touched since last we came | |
3011 | * here | |
3012 | */ | |
3013 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3014 | { | |
761d5c38 | 3015 | struct mount *mnt, *next; |
1da177e4 LT |
3016 | LIST_HEAD(graveyard); |
3017 | ||
3018 | if (list_empty(mounts)) | |
3019 | return; | |
3020 | ||
97216be0 | 3021 | namespace_lock(); |
719ea2fb | 3022 | lock_mount_hash(); |
1da177e4 LT |
3023 | |
3024 | /* extract from the expiration list every vfsmount that matches the | |
3025 | * following criteria: | |
3026 | * - only referenced by its parent vfsmount | |
3027 | * - still marked for expiry (marked on the last call here; marks are | |
3028 | * cleared by mntput()) | |
3029 | */ | |
6776db3d | 3030 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3031 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3032 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3033 | continue; |
6776db3d | 3034 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3035 | } |
bcc5c7d2 | 3036 | while (!list_empty(&graveyard)) { |
6776db3d | 3037 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3038 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3039 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3040 | } |
719ea2fb | 3041 | unlock_mount_hash(); |
3ab6abee | 3042 | namespace_unlock(); |
5528f911 TM |
3043 | } |
3044 | ||
3045 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3046 | ||
3047 | /* | |
3048 | * Ripoff of 'select_parent()' | |
3049 | * | |
3050 | * search the list of submounts for a given mountpoint, and move any | |
3051 | * shrinkable submounts to the 'graveyard' list. | |
3052 | */ | |
692afc31 | 3053 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3054 | { |
692afc31 | 3055 | struct mount *this_parent = parent; |
5528f911 TM |
3056 | struct list_head *next; |
3057 | int found = 0; | |
3058 | ||
3059 | repeat: | |
6b41d536 | 3060 | next = this_parent->mnt_mounts.next; |
5528f911 | 3061 | resume: |
6b41d536 | 3062 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3063 | struct list_head *tmp = next; |
6b41d536 | 3064 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3065 | |
3066 | next = tmp->next; | |
692afc31 | 3067 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3068 | continue; |
5528f911 TM |
3069 | /* |
3070 | * Descend a level if the d_mounts list is non-empty. | |
3071 | */ | |
6b41d536 | 3072 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3073 | this_parent = mnt; |
3074 | goto repeat; | |
3075 | } | |
1da177e4 | 3076 | |
1ab59738 | 3077 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3078 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3079 | found++; |
3080 | } | |
1da177e4 | 3081 | } |
5528f911 TM |
3082 | /* |
3083 | * All done at this level ... ascend and resume the search | |
3084 | */ | |
3085 | if (this_parent != parent) { | |
6b41d536 | 3086 | next = this_parent->mnt_child.next; |
0714a533 | 3087 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3088 | goto resume; |
3089 | } | |
3090 | return found; | |
3091 | } | |
3092 | ||
3093 | /* | |
3094 | * process a list of expirable mountpoints with the intent of discarding any | |
3095 | * submounts of a specific parent mountpoint | |
99b7db7b | 3096 | * |
48a066e7 | 3097 | * mount_lock must be held for write |
5528f911 | 3098 | */ |
b54b9be7 | 3099 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3100 | { |
3101 | LIST_HEAD(graveyard); | |
761d5c38 | 3102 | struct mount *m; |
5528f911 | 3103 | |
5528f911 | 3104 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3105 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3106 | while (!list_empty(&graveyard)) { |
761d5c38 | 3107 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3108 | mnt_expire); |
143c8c91 | 3109 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3110 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3111 | } |
3112 | } | |
1da177e4 LT |
3113 | } |
3114 | ||
028abd92 | 3115 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3116 | { |
b40ef869 | 3117 | char *copy; |
d563d678 | 3118 | unsigned left, offset; |
b58fed8b | 3119 | |
1da177e4 | 3120 | if (!data) |
b40ef869 | 3121 | return NULL; |
1da177e4 | 3122 | |
b40ef869 AV |
3123 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3124 | if (!copy) | |
3125 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3126 | |
d563d678 | 3127 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3128 | |
d563d678 CM |
3129 | /* |
3130 | * Not all architectures have an exact copy_from_user(). Resort to | |
3131 | * byte at a time. | |
3132 | */ | |
3133 | offset = PAGE_SIZE - left; | |
3134 | while (left) { | |
3135 | char c; | |
3136 | if (get_user(c, (const char __user *)data + offset)) | |
3137 | break; | |
3138 | copy[offset] = c; | |
3139 | left--; | |
3140 | offset++; | |
3141 | } | |
3142 | ||
3143 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3144 | kfree(copy); |
3145 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3146 | } |
d563d678 | 3147 | |
b40ef869 | 3148 | return copy; |
1da177e4 LT |
3149 | } |
3150 | ||
028abd92 | 3151 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3152 | { |
fbdb4401 | 3153 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3154 | } |
3155 | ||
1da177e4 LT |
3156 | /* |
3157 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3158 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3159 | * | |
3160 | * data is a (void *) that can point to any structure up to | |
3161 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3162 | * information (or be NULL). | |
3163 | * | |
3164 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3165 | * When the flags word was introduced its top half was required | |
3166 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3167 | * Therefore, if this magic number is present, it carries no information | |
3168 | * and must be discarded. | |
3169 | */ | |
c60166f0 | 3170 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3171 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3172 | { |
e462ec50 | 3173 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3174 | int ret; |
1da177e4 LT |
3175 | |
3176 | /* Discard magic */ | |
3177 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3178 | flags &= ~MS_MGC_MSK; | |
3179 | ||
3180 | /* Basic sanity checks */ | |
1da177e4 LT |
3181 | if (data_page) |
3182 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3183 | ||
e462ec50 DH |
3184 | if (flags & MS_NOUSER) |
3185 | return -EINVAL; | |
3186 | ||
a1e6aaa3 CH |
3187 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3188 | if (ret) | |
3189 | return ret; | |
3190 | if (!may_mount()) | |
3191 | return -EPERM; | |
f7e33bdb JL |
3192 | if (flags & SB_MANDLOCK) |
3193 | warn_mandlock(); | |
a27ab9f2 | 3194 | |
613cbe3d AK |
3195 | /* Default to relatime unless overriden */ |
3196 | if (!(flags & MS_NOATIME)) | |
3197 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3198 | |
1da177e4 LT |
3199 | /* Separate the per-mountpoint flags */ |
3200 | if (flags & MS_NOSUID) | |
3201 | mnt_flags |= MNT_NOSUID; | |
3202 | if (flags & MS_NODEV) | |
3203 | mnt_flags |= MNT_NODEV; | |
3204 | if (flags & MS_NOEXEC) | |
3205 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3206 | if (flags & MS_NOATIME) |
3207 | mnt_flags |= MNT_NOATIME; | |
3208 | if (flags & MS_NODIRATIME) | |
3209 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3210 | if (flags & MS_STRICTATIME) |
3211 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3212 | if (flags & MS_RDONLY) |
2e4b7fcd | 3213 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3214 | if (flags & MS_NOSYMFOLLOW) |
3215 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3216 | |
ffbc6f0e EB |
3217 | /* The default atime for remount is preservation */ |
3218 | if ((flags & MS_REMOUNT) && | |
3219 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3220 | MS_STRICTATIME)) == 0)) { | |
3221 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3222 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3223 | } |
3224 | ||
e462ec50 DH |
3225 | sb_flags = flags & (SB_RDONLY | |
3226 | SB_SYNCHRONOUS | | |
3227 | SB_MANDLOCK | | |
3228 | SB_DIRSYNC | | |
3229 | SB_SILENT | | |
917086ff | 3230 | SB_POSIXACL | |
d7ee9469 | 3231 | SB_LAZYTIME | |
917086ff | 3232 | SB_I_VERSION); |
1da177e4 | 3233 | |
43f5e655 | 3234 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3235 | return do_reconfigure_mnt(path, mnt_flags); |
3236 | if (flags & MS_REMOUNT) | |
3237 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3238 | if (flags & MS_BIND) | |
3239 | return do_loopback(path, dev_name, flags & MS_REC); | |
3240 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3241 | return do_change_type(path, flags); | |
3242 | if (flags & MS_MOVE) | |
3243 | return do_move_mount_old(path, dev_name); | |
3244 | ||
3245 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3246 | data_page); | |
3247 | } | |
3248 | ||
3249 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3250 | const char *type_page, unsigned long flags, void *data_page) | |
3251 | { | |
3252 | struct path path; | |
3253 | int ret; | |
3254 | ||
3255 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3256 | if (ret) | |
3257 | return ret; | |
3258 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3259 | path_put(&path); |
a1e6aaa3 | 3260 | return ret; |
1da177e4 LT |
3261 | } |
3262 | ||
537f7ccb EB |
3263 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3264 | { | |
3265 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3266 | } | |
3267 | ||
3268 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3269 | { | |
3270 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3271 | } | |
3272 | ||
771b1371 EB |
3273 | static void free_mnt_ns(struct mnt_namespace *ns) |
3274 | { | |
74e83122 AV |
3275 | if (!is_anon_ns(ns)) |
3276 | ns_free_inum(&ns->ns); | |
537f7ccb | 3277 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3278 | put_user_ns(ns->user_ns); |
3279 | kfree(ns); | |
3280 | } | |
3281 | ||
8823c079 EB |
3282 | /* |
3283 | * Assign a sequence number so we can detect when we attempt to bind | |
3284 | * mount a reference to an older mount namespace into the current | |
3285 | * mount namespace, preventing reference counting loops. A 64bit | |
3286 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3287 | * is effectively never, so we can ignore the possibility. | |
3288 | */ | |
3289 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3290 | ||
74e83122 | 3291 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3292 | { |
3293 | struct mnt_namespace *new_ns; | |
537f7ccb | 3294 | struct ucounts *ucounts; |
98f842e6 | 3295 | int ret; |
cf8d2c11 | 3296 | |
537f7ccb EB |
3297 | ucounts = inc_mnt_namespaces(user_ns); |
3298 | if (!ucounts) | |
df75e774 | 3299 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3300 | |
74e83122 | 3301 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
3302 | if (!new_ns) { |
3303 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3304 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3305 | } |
74e83122 AV |
3306 | if (!anon) { |
3307 | ret = ns_alloc_inum(&new_ns->ns); | |
3308 | if (ret) { | |
3309 | kfree(new_ns); | |
3310 | dec_mnt_namespaces(ucounts); | |
3311 | return ERR_PTR(ret); | |
3312 | } | |
98f842e6 | 3313 | } |
33c42940 | 3314 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3315 | if (!anon) |
3316 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3317 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3318 | INIT_LIST_HEAD(&new_ns->list); |
3319 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3320 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3321 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3322 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3323 | return new_ns; |
3324 | } | |
3325 | ||
0766f788 | 3326 | __latent_entropy |
9559f689 AV |
3327 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3328 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3329 | { |
6b3286ed | 3330 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3331 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3332 | struct mount *p, *q; |
9559f689 | 3333 | struct mount *old; |
cb338d06 | 3334 | struct mount *new; |
7a472ef4 | 3335 | int copy_flags; |
1da177e4 | 3336 | |
9559f689 AV |
3337 | BUG_ON(!ns); |
3338 | ||
3339 | if (likely(!(flags & CLONE_NEWNS))) { | |
3340 | get_mnt_ns(ns); | |
3341 | return ns; | |
3342 | } | |
3343 | ||
3344 | old = ns->root; | |
3345 | ||
74e83122 | 3346 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3347 | if (IS_ERR(new_ns)) |
3348 | return new_ns; | |
1da177e4 | 3349 | |
97216be0 | 3350 | namespace_lock(); |
1da177e4 | 3351 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3352 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3353 | if (user_ns != ns->user_ns) |
3bd045cc | 3354 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3355 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3356 | if (IS_ERR(new)) { |
328e6d90 | 3357 | namespace_unlock(); |
771b1371 | 3358 | free_mnt_ns(new_ns); |
be34d1a3 | 3359 | return ERR_CAST(new); |
1da177e4 | 3360 | } |
3bd045cc AV |
3361 | if (user_ns != ns->user_ns) { |
3362 | lock_mount_hash(); | |
3363 | lock_mnt_tree(new); | |
3364 | unlock_mount_hash(); | |
3365 | } | |
be08d6d2 | 3366 | new_ns->root = new; |
1a4eeaf2 | 3367 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3368 | |
3369 | /* | |
3370 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3371 | * as belonging to new namespace. We have already acquired a private | |
3372 | * fs_struct, so tsk->fs->lock is not needed. | |
3373 | */ | |
909b0a88 | 3374 | p = old; |
cb338d06 | 3375 | q = new; |
1da177e4 | 3376 | while (p) { |
143c8c91 | 3377 | q->mnt_ns = new_ns; |
d2921684 | 3378 | new_ns->mounts++; |
9559f689 AV |
3379 | if (new_fs) { |
3380 | if (&p->mnt == new_fs->root.mnt) { | |
3381 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3382 | rootmnt = &p->mnt; |
1da177e4 | 3383 | } |
9559f689 AV |
3384 | if (&p->mnt == new_fs->pwd.mnt) { |
3385 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3386 | pwdmnt = &p->mnt; |
1da177e4 | 3387 | } |
1da177e4 | 3388 | } |
909b0a88 AV |
3389 | p = next_mnt(p, old); |
3390 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3391 | if (!q) |
3392 | break; | |
3393 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3394 | p = next_mnt(p, old); | |
1da177e4 | 3395 | } |
328e6d90 | 3396 | namespace_unlock(); |
1da177e4 | 3397 | |
1da177e4 | 3398 | if (rootmnt) |
f03c6599 | 3399 | mntput(rootmnt); |
1da177e4 | 3400 | if (pwdmnt) |
f03c6599 | 3401 | mntput(pwdmnt); |
1da177e4 | 3402 | |
741a2951 | 3403 | return new_ns; |
1da177e4 LT |
3404 | } |
3405 | ||
74e83122 | 3406 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3407 | { |
74e83122 | 3408 | struct mount *mnt = real_mount(m); |
ea441d11 | 3409 | struct mnt_namespace *ns; |
d31da0f0 | 3410 | struct super_block *s; |
ea441d11 AV |
3411 | struct path path; |
3412 | int err; | |
3413 | ||
74e83122 AV |
3414 | ns = alloc_mnt_ns(&init_user_ns, true); |
3415 | if (IS_ERR(ns)) { | |
3416 | mntput(m); | |
ea441d11 | 3417 | return ERR_CAST(ns); |
74e83122 AV |
3418 | } |
3419 | mnt->mnt_ns = ns; | |
3420 | ns->root = mnt; | |
3421 | ns->mounts++; | |
3422 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3423 | |
74e83122 | 3424 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3425 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3426 | ||
3427 | put_mnt_ns(ns); | |
3428 | ||
3429 | if (err) | |
3430 | return ERR_PTR(err); | |
3431 | ||
3432 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3433 | s = path.mnt->mnt_sb; |
3434 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3435 | mntput(path.mnt); |
3436 | /* lock the sucker */ | |
d31da0f0 | 3437 | down_write(&s->s_umount); |
ea441d11 AV |
3438 | /* ... and return the root of (sub)tree on it */ |
3439 | return path.dentry; | |
3440 | } | |
3441 | EXPORT_SYMBOL(mount_subtree); | |
3442 | ||
cccaa5e3 DB |
3443 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3444 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3445 | { |
eca6f534 VN |
3446 | int ret; |
3447 | char *kernel_type; | |
eca6f534 | 3448 | char *kernel_dev; |
b40ef869 | 3449 | void *options; |
1da177e4 | 3450 | |
b8850d1f TG |
3451 | kernel_type = copy_mount_string(type); |
3452 | ret = PTR_ERR(kernel_type); | |
3453 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3454 | goto out_type; |
1da177e4 | 3455 | |
b8850d1f TG |
3456 | kernel_dev = copy_mount_string(dev_name); |
3457 | ret = PTR_ERR(kernel_dev); | |
3458 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3459 | goto out_dev; |
1da177e4 | 3460 | |
b40ef869 AV |
3461 | options = copy_mount_options(data); |
3462 | ret = PTR_ERR(options); | |
3463 | if (IS_ERR(options)) | |
eca6f534 | 3464 | goto out_data; |
1da177e4 | 3465 | |
b40ef869 | 3466 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3467 | |
b40ef869 | 3468 | kfree(options); |
eca6f534 VN |
3469 | out_data: |
3470 | kfree(kernel_dev); | |
3471 | out_dev: | |
eca6f534 VN |
3472 | kfree(kernel_type); |
3473 | out_type: | |
3474 | return ret; | |
1da177e4 LT |
3475 | } |
3476 | ||
dd8b477f CB |
3477 | #define FSMOUNT_VALID_FLAGS \ |
3478 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3479 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ | |
3480 | MOUNT_ATTR_NOSYMFOLLOW) | |
5b490500 | 3481 | |
9caccd41 | 3482 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
2a186721 CB |
3483 | |
3484 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3485 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3486 | ||
5b490500 CB |
3487 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3488 | { | |
3489 | unsigned int mnt_flags = 0; | |
3490 | ||
3491 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3492 | mnt_flags |= MNT_READONLY; | |
3493 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3494 | mnt_flags |= MNT_NOSUID; | |
3495 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3496 | mnt_flags |= MNT_NODEV; | |
3497 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3498 | mnt_flags |= MNT_NOEXEC; | |
3499 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3500 | mnt_flags |= MNT_NODIRATIME; | |
dd8b477f CB |
3501 | if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) |
3502 | mnt_flags |= MNT_NOSYMFOLLOW; | |
5b490500 CB |
3503 | |
3504 | return mnt_flags; | |
3505 | } | |
3506 | ||
2db154b3 | 3507 | /* |
93766fbd DH |
3508 | * Create a kernel mount representation for a new, prepared superblock |
3509 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3510 | */ | |
3511 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3512 | unsigned int, attr_flags) | |
3513 | { | |
3514 | struct mnt_namespace *ns; | |
3515 | struct fs_context *fc; | |
3516 | struct file *file; | |
3517 | struct path newmount; | |
3518 | struct mount *mnt; | |
3519 | struct fd f; | |
3520 | unsigned int mnt_flags = 0; | |
3521 | long ret; | |
3522 | ||
3523 | if (!may_mount()) | |
3524 | return -EPERM; | |
3525 | ||
3526 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3527 | return -EINVAL; | |
3528 | ||
5b490500 | 3529 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3530 | return -EINVAL; |
3531 | ||
5b490500 | 3532 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3533 | |
3534 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3535 | case MOUNT_ATTR_STRICTATIME: | |
3536 | break; | |
3537 | case MOUNT_ATTR_NOATIME: | |
3538 | mnt_flags |= MNT_NOATIME; | |
3539 | break; | |
3540 | case MOUNT_ATTR_RELATIME: | |
3541 | mnt_flags |= MNT_RELATIME; | |
3542 | break; | |
3543 | default: | |
3544 | return -EINVAL; | |
3545 | } | |
3546 | ||
3547 | f = fdget(fs_fd); | |
3548 | if (!f.file) | |
3549 | return -EBADF; | |
3550 | ||
3551 | ret = -EINVAL; | |
3552 | if (f.file->f_op != &fscontext_fops) | |
3553 | goto err_fsfd; | |
3554 | ||
3555 | fc = f.file->private_data; | |
3556 | ||
3557 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3558 | if (ret < 0) | |
3559 | goto err_fsfd; | |
3560 | ||
3561 | /* There must be a valid superblock or we can't mount it */ | |
3562 | ret = -EINVAL; | |
3563 | if (!fc->root) | |
3564 | goto err_unlock; | |
3565 | ||
3566 | ret = -EPERM; | |
3567 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3568 | pr_warn("VFS: Mount too revealing\n"); | |
3569 | goto err_unlock; | |
3570 | } | |
3571 | ||
3572 | ret = -EBUSY; | |
3573 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3574 | goto err_unlock; | |
3575 | ||
f7e33bdb JL |
3576 | if (fc->sb_flags & SB_MANDLOCK) |
3577 | warn_mandlock(); | |
93766fbd DH |
3578 | |
3579 | newmount.mnt = vfs_create_mount(fc); | |
3580 | if (IS_ERR(newmount.mnt)) { | |
3581 | ret = PTR_ERR(newmount.mnt); | |
3582 | goto err_unlock; | |
3583 | } | |
3584 | newmount.dentry = dget(fc->root); | |
3585 | newmount.mnt->mnt_flags = mnt_flags; | |
3586 | ||
3587 | /* We've done the mount bit - now move the file context into more or | |
3588 | * less the same state as if we'd done an fspick(). We don't want to | |
3589 | * do any memory allocation or anything like that at this point as we | |
3590 | * don't want to have to handle any errors incurred. | |
3591 | */ | |
3592 | vfs_clean_context(fc); | |
3593 | ||
3594 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3595 | if (IS_ERR(ns)) { | |
3596 | ret = PTR_ERR(ns); | |
3597 | goto err_path; | |
3598 | } | |
3599 | mnt = real_mount(newmount.mnt); | |
3600 | mnt->mnt_ns = ns; | |
3601 | ns->root = mnt; | |
3602 | ns->mounts = 1; | |
3603 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3604 | mntget(newmount.mnt); |
93766fbd DH |
3605 | |
3606 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3607 | * it, not just simply put it. | |
3608 | */ | |
3609 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3610 | if (IS_ERR(file)) { | |
3611 | dissolve_on_fput(newmount.mnt); | |
3612 | ret = PTR_ERR(file); | |
3613 | goto err_path; | |
3614 | } | |
3615 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3616 | ||
3617 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3618 | if (ret >= 0) | |
3619 | fd_install(ret, file); | |
3620 | else | |
3621 | fput(file); | |
3622 | ||
3623 | err_path: | |
3624 | path_put(&newmount); | |
3625 | err_unlock: | |
3626 | mutex_unlock(&fc->uapi_mutex); | |
3627 | err_fsfd: | |
3628 | fdput(f); | |
3629 | return ret; | |
3630 | } | |
3631 | ||
3632 | /* | |
3633 | * Move a mount from one place to another. In combination with | |
3634 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3635 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3636 | * a mount subtree. | |
2db154b3 DH |
3637 | * |
3638 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3639 | */ | |
3640 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
3641 | int, from_dfd, const char __user *, from_pathname, |
3642 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
3643 | unsigned int, flags) |
3644 | { | |
3645 | struct path from_path, to_path; | |
3646 | unsigned int lflags; | |
3647 | int ret = 0; | |
3648 | ||
3649 | if (!may_mount()) | |
3650 | return -EPERM; | |
3651 | ||
3652 | if (flags & ~MOVE_MOUNT__MASK) | |
3653 | return -EINVAL; | |
3654 | ||
3655 | /* If someone gives a pathname, they aren't permitted to move | |
3656 | * from an fd that requires unmount as we can't get at the flag | |
3657 | * to clear it afterwards. | |
3658 | */ | |
3659 | lflags = 0; | |
3660 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3661 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3662 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3663 | ||
3664 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3665 | if (ret < 0) | |
3666 | return ret; | |
3667 | ||
3668 | lflags = 0; | |
3669 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3670 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3671 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3672 | ||
3673 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3674 | if (ret < 0) | |
3675 | goto out_from; | |
3676 | ||
3677 | ret = security_move_mount(&from_path, &to_path); | |
3678 | if (ret < 0) | |
3679 | goto out_to; | |
3680 | ||
3681 | ret = do_move_mount(&from_path, &to_path); | |
3682 | ||
3683 | out_to: | |
3684 | path_put(&to_path); | |
3685 | out_from: | |
3686 | path_put(&from_path); | |
3687 | return ret; | |
3688 | } | |
3689 | ||
afac7cba AV |
3690 | /* |
3691 | * Return true if path is reachable from root | |
3692 | * | |
48a066e7 | 3693 | * namespace_sem or mount_lock is held |
afac7cba | 3694 | */ |
643822b4 | 3695 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3696 | const struct path *root) |
3697 | { | |
643822b4 | 3698 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3699 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3700 | mnt = mnt->mnt_parent; |
afac7cba | 3701 | } |
643822b4 | 3702 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3703 | } |
3704 | ||
640eb7e7 | 3705 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3706 | { |
25ab4c9b | 3707 | bool res; |
48a066e7 | 3708 | read_seqlock_excl(&mount_lock); |
643822b4 | 3709 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3710 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3711 | return res; |
3712 | } | |
3713 | EXPORT_SYMBOL(path_is_under); | |
3714 | ||
1da177e4 LT |
3715 | /* |
3716 | * pivot_root Semantics: | |
3717 | * Moves the root file system of the current process to the directory put_old, | |
3718 | * makes new_root as the new root file system of the current process, and sets | |
3719 | * root/cwd of all processes which had them on the current root to new_root. | |
3720 | * | |
3721 | * Restrictions: | |
3722 | * The new_root and put_old must be directories, and must not be on the | |
3723 | * same file system as the current process root. The put_old must be | |
3724 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3725 | * pointed to by put_old must yield the same directory as new_root. No other | |
3726 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3727 | * | |
4a0d11fa | 3728 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 3729 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
3730 | * in this situation. |
3731 | * | |
1da177e4 LT |
3732 | * Notes: |
3733 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3734 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3735 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3736 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3737 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3738 | * first. | |
3739 | */ | |
3480b257 HC |
3740 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3741 | const char __user *, put_old) | |
1da177e4 | 3742 | { |
2763d119 AV |
3743 | struct path new, old, root; |
3744 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3745 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3746 | int error; |
3747 | ||
9b40bc90 | 3748 | if (!may_mount()) |
1da177e4 LT |
3749 | return -EPERM; |
3750 | ||
ce6595a2 AV |
3751 | error = user_path_at(AT_FDCWD, new_root, |
3752 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
3753 | if (error) |
3754 | goto out0; | |
1da177e4 | 3755 | |
ce6595a2 AV |
3756 | error = user_path_at(AT_FDCWD, put_old, |
3757 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
3758 | if (error) |
3759 | goto out1; | |
3760 | ||
2d8f3038 | 3761 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3762 | if (error) |
3763 | goto out2; | |
1da177e4 | 3764 | |
f7ad3c6b | 3765 | get_fs_root(current->fs, &root); |
84d17192 AV |
3766 | old_mp = lock_mount(&old); |
3767 | error = PTR_ERR(old_mp); | |
3768 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3769 | goto out3; |
3770 | ||
1da177e4 | 3771 | error = -EINVAL; |
419148da AV |
3772 | new_mnt = real_mount(new.mnt); |
3773 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3774 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3775 | ex_parent = new_mnt->mnt_parent; |
3776 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3777 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3778 | IS_MNT_SHARED(ex_parent) || |
3779 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3780 | goto out4; |
143c8c91 | 3781 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3782 | goto out4; |
5ff9d8a6 EB |
3783 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3784 | goto out4; | |
1da177e4 | 3785 | error = -ENOENT; |
f3da392e | 3786 | if (d_unlinked(new.dentry)) |
b12cea91 | 3787 | goto out4; |
1da177e4 | 3788 | error = -EBUSY; |
84d17192 | 3789 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3790 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3791 | error = -EINVAL; |
8c3ee42e | 3792 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3793 | goto out4; /* not a mountpoint */ |
676da58d | 3794 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3795 | goto out4; /* not attached */ |
2d8f3038 | 3796 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3797 | goto out4; /* not a mountpoint */ |
676da58d | 3798 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3799 | goto out4; /* not attached */ |
4ac91378 | 3800 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3801 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3802 | goto out4; |
0d082601 EB |
3803 | /* make certain new is below the root */ |
3804 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3805 | goto out4; | |
719ea2fb | 3806 | lock_mount_hash(); |
2763d119 AV |
3807 | umount_mnt(new_mnt); |
3808 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3809 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3810 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3811 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3812 | } | |
4ac91378 | 3813 | /* mount old root on put_old */ |
84d17192 | 3814 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3815 | /* mount new_root on / */ |
2763d119 AV |
3816 | attach_mnt(new_mnt, root_parent, root_mp); |
3817 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3818 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3819 | /* A moved mount should not expire automatically */ |
3820 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3821 | put_mountpoint(root_mp); |
719ea2fb | 3822 | unlock_mount_hash(); |
2d8f3038 | 3823 | chroot_fs_refs(&root, &new); |
1da177e4 | 3824 | error = 0; |
b12cea91 | 3825 | out4: |
84d17192 | 3826 | unlock_mount(old_mp); |
2763d119 AV |
3827 | if (!error) |
3828 | mntput_no_expire(ex_parent); | |
b12cea91 | 3829 | out3: |
8c3ee42e | 3830 | path_put(&root); |
b12cea91 | 3831 | out2: |
2d8f3038 | 3832 | path_put(&old); |
1da177e4 | 3833 | out1: |
2d8f3038 | 3834 | path_put(&new); |
1da177e4 | 3835 | out0: |
1da177e4 | 3836 | return error; |
1da177e4 LT |
3837 | } |
3838 | ||
2a186721 CB |
3839 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
3840 | { | |
3841 | unsigned int flags = mnt->mnt.mnt_flags; | |
3842 | ||
3843 | /* flags to clear */ | |
3844 | flags &= ~kattr->attr_clr; | |
3845 | /* flags to raise */ | |
3846 | flags |= kattr->attr_set; | |
3847 | ||
3848 | return flags; | |
3849 | } | |
3850 | ||
9caccd41 CB |
3851 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3852 | { | |
3853 | struct vfsmount *m = &mnt->mnt; | |
3854 | ||
3855 | if (!kattr->mnt_userns) | |
3856 | return 0; | |
3857 | ||
3858 | /* | |
3859 | * Once a mount has been idmapped we don't allow it to change its | |
3860 | * mapping. It makes things simpler and callers can just create | |
3861 | * another bind-mount they can idmap if they want to. | |
3862 | */ | |
3863 | if (mnt_user_ns(m) != &init_user_ns) | |
3864 | return -EPERM; | |
3865 | ||
3866 | /* The underlying filesystem doesn't support idmapped mounts yet. */ | |
3867 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) | |
3868 | return -EINVAL; | |
3869 | ||
2ca4dcc4 CB |
3870 | /* Don't yet support filesystem mountable in user namespaces. */ |
3871 | if (m->mnt_sb->s_user_ns != &init_user_ns) | |
3872 | return -EINVAL; | |
3873 | ||
9caccd41 | 3874 | /* We're not controlling the superblock. */ |
2ca4dcc4 | 3875 | if (!capable(CAP_SYS_ADMIN)) |
9caccd41 CB |
3876 | return -EPERM; |
3877 | ||
3878 | /* Mount has already been visible in the filesystem hierarchy. */ | |
3879 | if (!is_anon_ns(mnt->mnt_ns)) | |
3880 | return -EINVAL; | |
3881 | ||
3882 | return 0; | |
3883 | } | |
3884 | ||
2a186721 CB |
3885 | static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, |
3886 | struct mount *mnt, int *err) | |
3887 | { | |
3888 | struct mount *m = mnt, *last = NULL; | |
3889 | ||
3890 | if (!is_mounted(&m->mnt)) { | |
3891 | *err = -EINVAL; | |
3892 | goto out; | |
3893 | } | |
3894 | ||
3895 | if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { | |
3896 | *err = -EINVAL; | |
3897 | goto out; | |
3898 | } | |
3899 | ||
3900 | do { | |
3901 | unsigned int flags; | |
3902 | ||
3903 | flags = recalc_flags(kattr, m); | |
3904 | if (!can_change_locked_flags(m, flags)) { | |
3905 | *err = -EPERM; | |
3906 | goto out; | |
3907 | } | |
3908 | ||
9caccd41 CB |
3909 | *err = can_idmap_mount(kattr, m); |
3910 | if (*err) | |
3911 | goto out; | |
3912 | ||
2a186721 CB |
3913 | last = m; |
3914 | ||
3915 | if ((kattr->attr_set & MNT_READONLY) && | |
3916 | !(m->mnt.mnt_flags & MNT_READONLY)) { | |
3917 | *err = mnt_hold_writers(m); | |
3918 | if (*err) | |
3919 | goto out; | |
3920 | } | |
3921 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3922 | ||
3923 | out: | |
3924 | return last; | |
3925 | } | |
3926 | ||
9caccd41 CB |
3927 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3928 | { | |
3929 | struct user_namespace *mnt_userns; | |
3930 | ||
3931 | if (!kattr->mnt_userns) | |
3932 | return; | |
3933 | ||
3934 | mnt_userns = get_user_ns(kattr->mnt_userns); | |
3935 | /* Pairs with smp_load_acquire() in mnt_user_ns(). */ | |
3936 | smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); | |
3937 | } | |
3938 | ||
2a186721 CB |
3939 | static void mount_setattr_commit(struct mount_kattr *kattr, |
3940 | struct mount *mnt, struct mount *last, | |
3941 | int err) | |
3942 | { | |
3943 | struct mount *m = mnt; | |
3944 | ||
3945 | do { | |
3946 | if (!err) { | |
3947 | unsigned int flags; | |
3948 | ||
9caccd41 | 3949 | do_idmap_mount(kattr, m); |
2a186721 CB |
3950 | flags = recalc_flags(kattr, m); |
3951 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
3952 | } | |
3953 | ||
3954 | /* | |
3955 | * We either set MNT_READONLY above so make it visible | |
3956 | * before ~MNT_WRITE_HOLD or we failed to recursively | |
3957 | * apply mount options. | |
3958 | */ | |
3959 | if ((kattr->attr_set & MNT_READONLY) && | |
3960 | (m->mnt.mnt_flags & MNT_WRITE_HOLD)) | |
3961 | mnt_unhold_writers(m); | |
3962 | ||
3963 | if (!err && kattr->propagation) | |
3964 | change_mnt_propagation(m, kattr->propagation); | |
3965 | ||
3966 | /* | |
3967 | * On failure, only cleanup until we found the first mount | |
3968 | * we failed to handle. | |
3969 | */ | |
3970 | if (err && m == last) | |
3971 | break; | |
3972 | } while (kattr->recurse && (m = next_mnt(m, mnt))); | |
3973 | ||
3974 | if (!err) | |
3975 | touch_mnt_namespace(mnt->mnt_ns); | |
3976 | } | |
3977 | ||
3978 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
3979 | { | |
3980 | struct mount *mnt = real_mount(path->mnt), *last = NULL; | |
3981 | int err = 0; | |
3982 | ||
3983 | if (path->dentry != mnt->mnt.mnt_root) | |
3984 | return -EINVAL; | |
3985 | ||
3986 | if (kattr->propagation) { | |
3987 | /* | |
3988 | * Only take namespace_lock() if we're actually changing | |
3989 | * propagation. | |
3990 | */ | |
3991 | namespace_lock(); | |
3992 | if (kattr->propagation == MS_SHARED) { | |
3993 | err = invent_group_ids(mnt, kattr->recurse); | |
3994 | if (err) { | |
3995 | namespace_unlock(); | |
3996 | return err; | |
3997 | } | |
3998 | } | |
3999 | } | |
4000 | ||
4001 | lock_mount_hash(); | |
4002 | ||
4003 | /* | |
4004 | * Get the mount tree in a shape where we can change mount | |
4005 | * properties without failure. | |
4006 | */ | |
4007 | last = mount_setattr_prepare(kattr, mnt, &err); | |
4008 | if (last) /* Commit all changes or revert to the old state. */ | |
4009 | mount_setattr_commit(kattr, mnt, last, err); | |
4010 | ||
4011 | unlock_mount_hash(); | |
4012 | ||
4013 | if (kattr->propagation) { | |
4014 | namespace_unlock(); | |
4015 | if (err) | |
4016 | cleanup_group_ids(mnt, NULL); | |
4017 | } | |
4018 | ||
4019 | return err; | |
4020 | } | |
4021 | ||
9caccd41 CB |
4022 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
4023 | struct mount_kattr *kattr, unsigned int flags) | |
4024 | { | |
4025 | int err = 0; | |
4026 | struct ns_common *ns; | |
4027 | struct user_namespace *mnt_userns; | |
4028 | struct file *file; | |
4029 | ||
4030 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) | |
4031 | return 0; | |
4032 | ||
4033 | /* | |
4034 | * We currently do not support clearing an idmapped mount. If this ever | |
4035 | * is a use-case we can revisit this but for now let's keep it simple | |
4036 | * and not allow it. | |
4037 | */ | |
4038 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) | |
4039 | return -EINVAL; | |
4040 | ||
4041 | if (attr->userns_fd > INT_MAX) | |
4042 | return -EINVAL; | |
4043 | ||
4044 | file = fget(attr->userns_fd); | |
4045 | if (!file) | |
4046 | return -EBADF; | |
4047 | ||
4048 | if (!proc_ns_file(file)) { | |
4049 | err = -EINVAL; | |
4050 | goto out_fput; | |
4051 | } | |
4052 | ||
4053 | ns = get_proc_ns(file_inode(file)); | |
4054 | if (ns->ops->type != CLONE_NEWUSER) { | |
4055 | err = -EINVAL; | |
4056 | goto out_fput; | |
4057 | } | |
4058 | ||
4059 | /* | |
4060 | * The init_user_ns is used to indicate that a vfsmount is not idmapped. | |
4061 | * This is simpler than just having to treat NULL as unmapped. Users | |
4062 | * wanting to idmap a mount to init_user_ns can just use a namespace | |
4063 | * with an identity mapping. | |
4064 | */ | |
4065 | mnt_userns = container_of(ns, struct user_namespace, ns); | |
4066 | if (mnt_userns == &init_user_ns) { | |
4067 | err = -EPERM; | |
4068 | goto out_fput; | |
4069 | } | |
4070 | kattr->mnt_userns = get_user_ns(mnt_userns); | |
4071 | ||
4072 | out_fput: | |
4073 | fput(file); | |
4074 | return err; | |
4075 | } | |
4076 | ||
4077 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, | |
2a186721 CB |
4078 | struct mount_kattr *kattr, unsigned int flags) |
4079 | { | |
4080 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
4081 | ||
4082 | if (flags & AT_NO_AUTOMOUNT) | |
4083 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
4084 | if (flags & AT_SYMLINK_NOFOLLOW) | |
4085 | lookup_flags &= ~LOOKUP_FOLLOW; | |
4086 | if (flags & AT_EMPTY_PATH) | |
4087 | lookup_flags |= LOOKUP_EMPTY; | |
4088 | ||
4089 | *kattr = (struct mount_kattr) { | |
4090 | .lookup_flags = lookup_flags, | |
4091 | .recurse = !!(flags & AT_RECURSIVE), | |
4092 | }; | |
4093 | ||
4094 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
4095 | return -EINVAL; | |
4096 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
4097 | return -EINVAL; | |
4098 | kattr->propagation = attr->propagation; | |
4099 | ||
4100 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
4101 | return -EINVAL; | |
4102 | ||
2a186721 CB |
4103 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
4104 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
4105 | ||
4106 | /* | |
4107 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4108 | * users wanting to transition to a different atime setting cannot | |
4109 | * simply specify the atime setting in @attr_set, but must also | |
4110 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4111 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4112 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4113 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4114 | */ | |
4115 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4116 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4117 | return -EINVAL; | |
4118 | ||
4119 | /* | |
4120 | * Clear all previous time settings as they are mutually | |
4121 | * exclusive. | |
4122 | */ | |
4123 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4124 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4125 | case MOUNT_ATTR_RELATIME: | |
4126 | kattr->attr_set |= MNT_RELATIME; | |
4127 | break; | |
4128 | case MOUNT_ATTR_NOATIME: | |
4129 | kattr->attr_set |= MNT_NOATIME; | |
4130 | break; | |
4131 | case MOUNT_ATTR_STRICTATIME: | |
4132 | break; | |
4133 | default: | |
4134 | return -EINVAL; | |
4135 | } | |
4136 | } else { | |
4137 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4138 | return -EINVAL; | |
4139 | } | |
4140 | ||
9caccd41 CB |
4141 | return build_mount_idmapped(attr, usize, kattr, flags); |
4142 | } | |
4143 | ||
4144 | static void finish_mount_kattr(struct mount_kattr *kattr) | |
4145 | { | |
4146 | put_user_ns(kattr->mnt_userns); | |
4147 | kattr->mnt_userns = NULL; | |
2a186721 CB |
4148 | } |
4149 | ||
4150 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4151 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4152 | size_t, usize) | |
4153 | { | |
4154 | int err; | |
4155 | struct path target; | |
4156 | struct mount_attr attr; | |
4157 | struct mount_kattr kattr; | |
4158 | ||
4159 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4160 | ||
4161 | if (flags & ~(AT_EMPTY_PATH | | |
4162 | AT_RECURSIVE | | |
4163 | AT_SYMLINK_NOFOLLOW | | |
4164 | AT_NO_AUTOMOUNT)) | |
4165 | return -EINVAL; | |
4166 | ||
4167 | if (unlikely(usize > PAGE_SIZE)) | |
4168 | return -E2BIG; | |
4169 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4170 | return -EINVAL; | |
4171 | ||
4172 | if (!may_mount()) | |
4173 | return -EPERM; | |
4174 | ||
4175 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4176 | if (err) | |
4177 | return err; | |
4178 | ||
4179 | /* Don't bother walking through the mounts if this is a nop. */ | |
4180 | if (attr.attr_set == 0 && | |
4181 | attr.attr_clr == 0 && | |
4182 | attr.propagation == 0) | |
4183 | return 0; | |
4184 | ||
9caccd41 | 4185 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
2a186721 CB |
4186 | if (err) |
4187 | return err; | |
4188 | ||
4189 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
4190 | if (err) | |
4191 | return err; | |
4192 | ||
4193 | err = do_mount_setattr(&target, &kattr); | |
9caccd41 | 4194 | finish_mount_kattr(&kattr); |
2a186721 CB |
4195 | path_put(&target); |
4196 | return err; | |
4197 | } | |
4198 | ||
1da177e4 LT |
4199 | static void __init init_mount_tree(void) |
4200 | { | |
4201 | struct vfsmount *mnt; | |
74e83122 | 4202 | struct mount *m; |
6b3286ed | 4203 | struct mnt_namespace *ns; |
ac748a09 | 4204 | struct path root; |
1da177e4 | 4205 | |
fd3e007f | 4206 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4207 | if (IS_ERR(mnt)) |
4208 | panic("Can't create rootfs"); | |
b3e19d92 | 4209 | |
74e83122 | 4210 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4211 | if (IS_ERR(ns)) |
1da177e4 | 4212 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4213 | m = real_mount(mnt); |
4214 | m->mnt_ns = ns; | |
4215 | ns->root = m; | |
4216 | ns->mounts = 1; | |
4217 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4218 | init_task.nsproxy->mnt_ns = ns; |
4219 | get_mnt_ns(ns); | |
4220 | ||
be08d6d2 AV |
4221 | root.mnt = mnt; |
4222 | root.dentry = mnt->mnt_root; | |
da362b09 | 4223 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4224 | |
4225 | set_fs_pwd(current->fs, &root); | |
4226 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4227 | } |
4228 | ||
74bf17cf | 4229 | void __init mnt_init(void) |
1da177e4 | 4230 | { |
15a67dd8 | 4231 | int err; |
1da177e4 | 4232 | |
7d6fec45 | 4233 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 4234 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 4235 | |
0818bf27 | 4236 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4237 | sizeof(struct hlist_head), |
0818bf27 | 4238 | mhash_entries, 19, |
3d375d78 | 4239 | HASH_ZERO, |
0818bf27 AV |
4240 | &m_hash_shift, &m_hash_mask, 0, 0); |
4241 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4242 | sizeof(struct hlist_head), | |
4243 | mphash_entries, 19, | |
3d375d78 | 4244 | HASH_ZERO, |
0818bf27 | 4245 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4246 | |
84d17192 | 4247 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4248 | panic("Failed to allocate mount hash table\n"); |
4249 | ||
4b93dc9b TH |
4250 | kernfs_init(); |
4251 | ||
15a67dd8 RD |
4252 | err = sysfs_init(); |
4253 | if (err) | |
4254 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4255 | __func__, err); |
00d26666 GKH |
4256 | fs_kobj = kobject_create_and_add("fs", NULL); |
4257 | if (!fs_kobj) | |
8e24eea7 | 4258 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4259 | shmem_init(); |
1da177e4 LT |
4260 | init_rootfs(); |
4261 | init_mount_tree(); | |
4262 | } | |
4263 | ||
616511d0 | 4264 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4265 | { |
1a7b8969 | 4266 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4267 | return; |
7b00ed6f | 4268 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4269 | free_mnt_ns(ns); |
1da177e4 | 4270 | } |
9d412a43 | 4271 | |
d911b458 | 4272 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4273 | { |
423e0ab0 | 4274 | struct vfsmount *mnt; |
d911b458 | 4275 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4276 | if (!IS_ERR(mnt)) { |
4277 | /* | |
4278 | * it is a longterm mount, don't release mnt until | |
4279 | * we unmount before file sys is unregistered | |
4280 | */ | |
f7a99c5b | 4281 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4282 | } |
4283 | return mnt; | |
9d412a43 | 4284 | } |
d911b458 | 4285 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4286 | |
4287 | void kern_unmount(struct vfsmount *mnt) | |
4288 | { | |
4289 | /* release long term mount so mount point can be released */ | |
4290 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 4291 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 4292 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4293 | mntput(mnt); |
4294 | } | |
4295 | } | |
4296 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4297 | |
df820f8d MS |
4298 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4299 | { | |
4300 | unsigned int i; | |
4301 | ||
4302 | for (i = 0; i < num; i++) | |
4303 | if (mnt[i]) | |
4304 | real_mount(mnt[i])->mnt_ns = NULL; | |
4305 | synchronize_rcu_expedited(); | |
4306 | for (i = 0; i < num; i++) | |
4307 | mntput(mnt[i]); | |
4308 | } | |
4309 | EXPORT_SYMBOL(kern_unmount_array); | |
4310 | ||
02125a82 AV |
4311 | bool our_mnt(struct vfsmount *mnt) |
4312 | { | |
143c8c91 | 4313 | return check_mnt(real_mount(mnt)); |
02125a82 | 4314 | } |
8823c079 | 4315 | |
3151527e EB |
4316 | bool current_chrooted(void) |
4317 | { | |
4318 | /* Does the current process have a non-standard root */ | |
4319 | struct path ns_root; | |
4320 | struct path fs_root; | |
4321 | bool chrooted; | |
4322 | ||
4323 | /* Find the namespace root */ | |
4324 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4325 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4326 | path_get(&ns_root); | |
4327 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4328 | ; | |
4329 | ||
4330 | get_fs_root(current->fs, &fs_root); | |
4331 | ||
4332 | chrooted = !path_equal(&fs_root, &ns_root); | |
4333 | ||
4334 | path_put(&fs_root); | |
4335 | path_put(&ns_root); | |
4336 | ||
4337 | return chrooted; | |
4338 | } | |
4339 | ||
132e4608 DH |
4340 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4341 | const struct super_block *sb, | |
8654df4e | 4342 | int *new_mnt_flags) |
87a8ebd6 | 4343 | { |
8c6cf9cc | 4344 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4345 | struct mount *mnt; |
e51db735 | 4346 | bool visible = false; |
87a8ebd6 | 4347 | |
44bb4385 | 4348 | down_read(&namespace_sem); |
9f6c61f9 | 4349 | lock_ns_list(ns); |
87a8ebd6 | 4350 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4351 | struct mount *child; |
77b1a97d EB |
4352 | int mnt_flags; |
4353 | ||
9f6c61f9 MS |
4354 | if (mnt_is_cursor(mnt)) |
4355 | continue; | |
4356 | ||
132e4608 | 4357 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4358 | continue; |
4359 | ||
7e96c1b0 EB |
4360 | /* This mount is not fully visible if it's root directory |
4361 | * is not the root directory of the filesystem. | |
4362 | */ | |
4363 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4364 | continue; | |
4365 | ||
a1935c17 | 4366 | /* A local view of the mount flags */ |
77b1a97d | 4367 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4368 | |
695e9df0 | 4369 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4370 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4371 | mnt_flags |= MNT_LOCK_READONLY; |
4372 | ||
8c6cf9cc EB |
4373 | /* Verify the mount flags are equal to or more permissive |
4374 | * than the proposed new mount. | |
4375 | */ | |
77b1a97d | 4376 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4377 | !(new_flags & MNT_READONLY)) |
4378 | continue; | |
77b1a97d EB |
4379 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4380 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4381 | continue; |
4382 | ||
ceeb0e5d EB |
4383 | /* This mount is not fully visible if there are any |
4384 | * locked child mounts that cover anything except for | |
4385 | * empty directories. | |
e51db735 EB |
4386 | */ |
4387 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4388 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4389 | /* Only worry about locked mounts */ |
d71ed6c9 | 4390 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4391 | continue; |
7236c85e EB |
4392 | /* Is the directory permanetly empty? */ |
4393 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4394 | goto next; |
87a8ebd6 | 4395 | } |
8c6cf9cc | 4396 | /* Preserve the locked attributes */ |
77b1a97d | 4397 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4398 | MNT_LOCK_ATIME); |
e51db735 EB |
4399 | visible = true; |
4400 | goto found; | |
4401 | next: ; | |
87a8ebd6 | 4402 | } |
e51db735 | 4403 | found: |
9f6c61f9 | 4404 | unlock_ns_list(ns); |
44bb4385 | 4405 | up_read(&namespace_sem); |
e51db735 | 4406 | return visible; |
87a8ebd6 EB |
4407 | } |
4408 | ||
132e4608 | 4409 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4410 | { |
a1935c17 | 4411 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4412 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4413 | unsigned long s_iflags; | |
4414 | ||
4415 | if (ns->user_ns == &init_user_ns) | |
4416 | return false; | |
4417 | ||
4418 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4419 | s_iflags = sb->s_iflags; |
8654df4e EB |
4420 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4421 | return false; | |
4422 | ||
a1935c17 EB |
4423 | if ((s_iflags & required_iflags) != required_iflags) { |
4424 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4425 | required_iflags); | |
4426 | return true; | |
4427 | } | |
4428 | ||
132e4608 | 4429 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4430 | } |
4431 | ||
380cf5ba AL |
4432 | bool mnt_may_suid(struct vfsmount *mnt) |
4433 | { | |
4434 | /* | |
4435 | * Foreign mounts (accessed via fchdir or through /proc | |
4436 | * symlinks) are always treated as if they are nosuid. This | |
4437 | * prevents namespaces from trusting potentially unsafe | |
4438 | * suid/sgid bits, file caps, or security labels that originate | |
4439 | * in other namespaces. | |
4440 | */ | |
4441 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4442 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4443 | } | |
4444 | ||
64964528 | 4445 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4446 | { |
58be2825 | 4447 | struct ns_common *ns = NULL; |
8823c079 EB |
4448 | struct nsproxy *nsproxy; |
4449 | ||
728dba3a EB |
4450 | task_lock(task); |
4451 | nsproxy = task->nsproxy; | |
8823c079 | 4452 | if (nsproxy) { |
58be2825 AV |
4453 | ns = &nsproxy->mnt_ns->ns; |
4454 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4455 | } |
728dba3a | 4456 | task_unlock(task); |
8823c079 EB |
4457 | |
4458 | return ns; | |
4459 | } | |
4460 | ||
64964528 | 4461 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4462 | { |
58be2825 | 4463 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4464 | } |
4465 | ||
f2a8d52e | 4466 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4467 | { |
f2a8d52e CB |
4468 | struct nsproxy *nsproxy = nsset->nsproxy; |
4469 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4470 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4471 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4472 | struct path root; |
4f757f3c | 4473 | int err; |
8823c079 | 4474 | |
0c55cfc4 | 4475 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4476 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4477 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4478 | return -EPERM; |
8823c079 | 4479 | |
74e83122 AV |
4480 | if (is_anon_ns(mnt_ns)) |
4481 | return -EINVAL; | |
4482 | ||
8823c079 EB |
4483 | if (fs->users != 1) |
4484 | return -EINVAL; | |
4485 | ||
4486 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4487 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4488 | nsproxy->mnt_ns = mnt_ns; |
4489 | ||
4490 | /* Find the root */ | |
4f757f3c AV |
4491 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4492 | "/", LOOKUP_DOWN, &root); | |
4493 | if (err) { | |
4494 | /* revert to old namespace */ | |
4495 | nsproxy->mnt_ns = old_mnt_ns; | |
4496 | put_mnt_ns(mnt_ns); | |
4497 | return err; | |
4498 | } | |
8823c079 | 4499 | |
4068367c AV |
4500 | put_mnt_ns(old_mnt_ns); |
4501 | ||
8823c079 EB |
4502 | /* Update the pwd and root */ |
4503 | set_fs_pwd(fs, &root); | |
4504 | set_fs_root(fs, &root); | |
4505 | ||
4506 | path_put(&root); | |
4507 | return 0; | |
4508 | } | |
4509 | ||
bcac25a5 AV |
4510 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4511 | { | |
4512 | return to_mnt_ns(ns)->user_ns; | |
4513 | } | |
4514 | ||
8823c079 EB |
4515 | const struct proc_ns_operations mntns_operations = { |
4516 | .name = "mnt", | |
4517 | .type = CLONE_NEWNS, | |
4518 | .get = mntns_get, | |
4519 | .put = mntns_put, | |
4520 | .install = mntns_install, | |
bcac25a5 | 4521 | .owner = mntns_owner, |
8823c079 | 4522 | }; |