]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
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 */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
57c8a661 | 26 | #include <linux/memblock.h> |
9ea459e1 | 27 | #include <linux/task_work.h> |
9164bb4a | 28 | #include <linux/sched/task.h> |
e262e32d | 29 | #include <uapi/linux/mount.h> |
9bc61ab1 | 30 | #include <linux/fs_context.h> |
9164bb4a | 31 | |
07b20889 | 32 | #include "pnode.h" |
948730b0 | 33 | #include "internal.h" |
1da177e4 | 34 | |
d2921684 EB |
35 | /* Maximum number of mounts in a mount namespace */ |
36 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
37 | ||
0818bf27 AV |
38 | static unsigned int m_hash_mask __read_mostly; |
39 | static unsigned int m_hash_shift __read_mostly; | |
40 | static unsigned int mp_hash_mask __read_mostly; | |
41 | static unsigned int mp_hash_shift __read_mostly; | |
42 | ||
43 | static __initdata unsigned long mhash_entries; | |
44 | static int __init set_mhash_entries(char *str) | |
45 | { | |
46 | if (!str) | |
47 | return 0; | |
48 | mhash_entries = simple_strtoul(str, &str, 0); | |
49 | return 1; | |
50 | } | |
51 | __setup("mhash_entries=", set_mhash_entries); | |
52 | ||
53 | static __initdata unsigned long mphash_entries; | |
54 | static int __init set_mphash_entries(char *str) | |
55 | { | |
56 | if (!str) | |
57 | return 0; | |
58 | mphash_entries = simple_strtoul(str, &str, 0); | |
59 | return 1; | |
60 | } | |
61 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 62 | |
c7999c36 | 63 | static u64 event; |
73cd49ec | 64 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 65 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 66 | |
38129a13 | 67 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 68 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 69 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 70 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 71 | |
f87fd4c2 | 72 | /* /sys/fs */ |
00d26666 GKH |
73 | struct kobject *fs_kobj; |
74 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 75 | |
99b7db7b NP |
76 | /* |
77 | * vfsmount lock may be taken for read to prevent changes to the | |
78 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
79 | * up the tree. | |
80 | * | |
81 | * It should be taken for write in all cases where the vfsmount | |
82 | * tree or hash is modified or when a vfsmount structure is modified. | |
83 | */ | |
48a066e7 | 84 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 85 | |
38129a13 | 86 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 87 | { |
b58fed8b RP |
88 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
89 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
90 | tmp = tmp + (tmp >> m_hash_shift); |
91 | return &mount_hashtable[tmp & m_hash_mask]; | |
92 | } | |
93 | ||
94 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
95 | { | |
96 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
97 | tmp = tmp + (tmp >> mp_hash_shift); | |
98 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
99 | } |
100 | ||
b105e270 | 101 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 102 | { |
169b480e MW |
103 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
104 | ||
105 | if (res < 0) | |
106 | return res; | |
107 | mnt->mnt_id = res; | |
108 | return 0; | |
73cd49ec MS |
109 | } |
110 | ||
b105e270 | 111 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 112 | { |
169b480e | 113 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
114 | } |
115 | ||
719f5d7f MS |
116 | /* |
117 | * Allocate a new peer group ID | |
719f5d7f | 118 | */ |
4b8b21f4 | 119 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 120 | { |
169b480e | 121 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 122 | |
169b480e MW |
123 | if (res < 0) |
124 | return res; | |
125 | mnt->mnt_group_id = res; | |
126 | return 0; | |
719f5d7f MS |
127 | } |
128 | ||
129 | /* | |
130 | * Release a peer group ID | |
131 | */ | |
4b8b21f4 | 132 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 133 | { |
169b480e | 134 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 135 | mnt->mnt_group_id = 0; |
719f5d7f MS |
136 | } |
137 | ||
b3e19d92 NP |
138 | /* |
139 | * vfsmount lock must be held for read | |
140 | */ | |
83adc753 | 141 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
142 | { |
143 | #ifdef CONFIG_SMP | |
68e8a9fe | 144 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
145 | #else |
146 | preempt_disable(); | |
68e8a9fe | 147 | mnt->mnt_count += n; |
b3e19d92 NP |
148 | preempt_enable(); |
149 | #endif | |
150 | } | |
151 | ||
b3e19d92 NP |
152 | /* |
153 | * vfsmount lock must be held for write | |
154 | */ | |
83adc753 | 155 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
156 | { |
157 | #ifdef CONFIG_SMP | |
f03c6599 | 158 | unsigned int count = 0; |
b3e19d92 NP |
159 | int cpu; |
160 | ||
161 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 162 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
163 | } |
164 | ||
165 | return count; | |
166 | #else | |
68e8a9fe | 167 | return mnt->mnt_count; |
b3e19d92 NP |
168 | #endif |
169 | } | |
170 | ||
87b95ce0 AV |
171 | static void drop_mountpoint(struct fs_pin *p) |
172 | { | |
173 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
174 | dput(m->mnt_ex_mountpoint); | |
175 | pin_remove(p); | |
176 | mntput(&m->mnt); | |
177 | } | |
178 | ||
b105e270 | 179 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 180 | { |
c63181e6 AV |
181 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
182 | if (mnt) { | |
73cd49ec MS |
183 | int err; |
184 | ||
c63181e6 | 185 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
186 | if (err) |
187 | goto out_free_cache; | |
188 | ||
189 | if (name) { | |
fcc139ae | 190 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 191 | if (!mnt->mnt_devname) |
88b38782 | 192 | goto out_free_id; |
73cd49ec MS |
193 | } |
194 | ||
b3e19d92 | 195 | #ifdef CONFIG_SMP |
c63181e6 AV |
196 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
197 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
198 | goto out_free_devname; |
199 | ||
c63181e6 | 200 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 201 | #else |
c63181e6 AV |
202 | mnt->mnt_count = 1; |
203 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
204 | #endif |
205 | ||
38129a13 | 206 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
207 | INIT_LIST_HEAD(&mnt->mnt_child); |
208 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
209 | INIT_LIST_HEAD(&mnt->mnt_list); | |
210 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
211 | INIT_LIST_HEAD(&mnt->mnt_share); | |
212 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
213 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 214 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 215 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
87b95ce0 | 216 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 217 | } |
c63181e6 | 218 | return mnt; |
88b38782 | 219 | |
d3ef3d73 | 220 | #ifdef CONFIG_SMP |
221 | out_free_devname: | |
fcc139ae | 222 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 223 | #endif |
88b38782 | 224 | out_free_id: |
c63181e6 | 225 | mnt_free_id(mnt); |
88b38782 | 226 | out_free_cache: |
c63181e6 | 227 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 228 | return NULL; |
1da177e4 LT |
229 | } |
230 | ||
3d733633 DH |
231 | /* |
232 | * Most r/o checks on a fs are for operations that take | |
233 | * discrete amounts of time, like a write() or unlink(). | |
234 | * We must keep track of when those operations start | |
235 | * (for permission checks) and when they end, so that | |
236 | * we can determine when writes are able to occur to | |
237 | * a filesystem. | |
238 | */ | |
239 | /* | |
240 | * __mnt_is_readonly: check whether a mount is read-only | |
241 | * @mnt: the mount to check for its write status | |
242 | * | |
243 | * This shouldn't be used directly ouside of the VFS. | |
244 | * It does not guarantee that the filesystem will stay | |
245 | * r/w, just that it is right *now*. This can not and | |
246 | * should not be used in place of IS_RDONLY(inode). | |
247 | * mnt_want/drop_write() will _keep_ the filesystem | |
248 | * r/w. | |
249 | */ | |
43f5e655 | 250 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 251 | { |
43f5e655 | 252 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
253 | } |
254 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
255 | ||
83adc753 | 256 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 257 | { |
258 | #ifdef CONFIG_SMP | |
68e8a9fe | 259 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 260 | #else |
68e8a9fe | 261 | mnt->mnt_writers++; |
d3ef3d73 | 262 | #endif |
263 | } | |
3d733633 | 264 | |
83adc753 | 265 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 266 | { |
d3ef3d73 | 267 | #ifdef CONFIG_SMP |
68e8a9fe | 268 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 269 | #else |
68e8a9fe | 270 | mnt->mnt_writers--; |
d3ef3d73 | 271 | #endif |
3d733633 | 272 | } |
3d733633 | 273 | |
83adc753 | 274 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 275 | { |
d3ef3d73 | 276 | #ifdef CONFIG_SMP |
277 | unsigned int count = 0; | |
3d733633 | 278 | int cpu; |
3d733633 DH |
279 | |
280 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 281 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 282 | } |
3d733633 | 283 | |
d3ef3d73 | 284 | return count; |
285 | #else | |
286 | return mnt->mnt_writers; | |
287 | #endif | |
3d733633 DH |
288 | } |
289 | ||
4ed5e82f MS |
290 | static int mnt_is_readonly(struct vfsmount *mnt) |
291 | { | |
292 | if (mnt->mnt_sb->s_readonly_remount) | |
293 | return 1; | |
294 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
295 | smp_rmb(); | |
296 | return __mnt_is_readonly(mnt); | |
297 | } | |
298 | ||
8366025e | 299 | /* |
eb04c282 JK |
300 | * Most r/o & frozen checks on a fs are for operations that take discrete |
301 | * amounts of time, like a write() or unlink(). We must keep track of when | |
302 | * those operations start (for permission checks) and when they end, so that we | |
303 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
304 | */ |
305 | /** | |
eb04c282 | 306 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 307 | * @m: the mount on which to take a write |
8366025e | 308 | * |
eb04c282 JK |
309 | * This tells the low-level filesystem that a write is about to be performed to |
310 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
311 | * returning success. This operation does not protect against filesystem being | |
312 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
313 | * called. This is effectively a refcount. | |
8366025e | 314 | */ |
eb04c282 | 315 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 316 | { |
83adc753 | 317 | struct mount *mnt = real_mount(m); |
3d733633 | 318 | int ret = 0; |
3d733633 | 319 | |
d3ef3d73 | 320 | preempt_disable(); |
c6653a83 | 321 | mnt_inc_writers(mnt); |
d3ef3d73 | 322 | /* |
c6653a83 | 323 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 324 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
325 | * incremented count after it has set MNT_WRITE_HOLD. | |
326 | */ | |
327 | smp_mb(); | |
6aa7de05 | 328 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 329 | cpu_relax(); |
330 | /* | |
331 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
332 | * be set to match its requirements. So we must not load that until | |
333 | * MNT_WRITE_HOLD is cleared. | |
334 | */ | |
335 | smp_rmb(); | |
4ed5e82f | 336 | if (mnt_is_readonly(m)) { |
c6653a83 | 337 | mnt_dec_writers(mnt); |
3d733633 | 338 | ret = -EROFS; |
3d733633 | 339 | } |
d3ef3d73 | 340 | preempt_enable(); |
eb04c282 JK |
341 | |
342 | return ret; | |
343 | } | |
344 | ||
345 | /** | |
346 | * mnt_want_write - get write access to a mount | |
347 | * @m: the mount on which to take a write | |
348 | * | |
349 | * This tells the low-level filesystem that a write is about to be performed to | |
350 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
351 | * is not frozen) before returning success. When the write operation is | |
352 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
353 | */ | |
354 | int mnt_want_write(struct vfsmount *m) | |
355 | { | |
356 | int ret; | |
357 | ||
358 | sb_start_write(m->mnt_sb); | |
359 | ret = __mnt_want_write(m); | |
360 | if (ret) | |
361 | sb_end_write(m->mnt_sb); | |
3d733633 | 362 | return ret; |
8366025e DH |
363 | } |
364 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
365 | ||
96029c4e | 366 | /** |
367 | * mnt_clone_write - get write access to a mount | |
368 | * @mnt: the mount on which to take a write | |
369 | * | |
370 | * This is effectively like mnt_want_write, except | |
371 | * it must only be used to take an extra write reference | |
372 | * on a mountpoint that we already know has a write reference | |
373 | * on it. This allows some optimisation. | |
374 | * | |
375 | * After finished, mnt_drop_write must be called as usual to | |
376 | * drop the reference. | |
377 | */ | |
378 | int mnt_clone_write(struct vfsmount *mnt) | |
379 | { | |
380 | /* superblock may be r/o */ | |
381 | if (__mnt_is_readonly(mnt)) | |
382 | return -EROFS; | |
383 | preempt_disable(); | |
83adc753 | 384 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 385 | preempt_enable(); |
386 | return 0; | |
387 | } | |
388 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
389 | ||
390 | /** | |
eb04c282 | 391 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 392 | * @file: the file who's mount on which to take a write |
393 | * | |
eb04c282 | 394 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 395 | * do some optimisations if the file is open for write already |
396 | */ | |
eb04c282 | 397 | int __mnt_want_write_file(struct file *file) |
96029c4e | 398 | { |
83f936c7 | 399 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 400 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 401 | else |
402 | return mnt_clone_write(file->f_path.mnt); | |
403 | } | |
eb04c282 | 404 | |
7c6893e3 MS |
405 | /** |
406 | * mnt_want_write_file - get write access to a file's mount | |
407 | * @file: the file who's mount on which to take a write | |
408 | * | |
409 | * This is like mnt_want_write, but it takes a file and can | |
410 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
411 | */ |
412 | int mnt_want_write_file(struct file *file) | |
413 | { | |
414 | int ret; | |
415 | ||
a6795a58 | 416 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
417 | ret = __mnt_want_write_file(file); |
418 | if (ret) | |
a6795a58 | 419 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
420 | return ret; |
421 | } | |
96029c4e | 422 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
423 | ||
8366025e | 424 | /** |
eb04c282 | 425 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
426 | * @mnt: the mount on which to give up write access |
427 | * | |
428 | * Tells the low-level filesystem that we are done | |
429 | * performing writes to it. Must be matched with | |
eb04c282 | 430 | * __mnt_want_write() call above. |
8366025e | 431 | */ |
eb04c282 | 432 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 433 | { |
d3ef3d73 | 434 | preempt_disable(); |
83adc753 | 435 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 436 | preempt_enable(); |
8366025e | 437 | } |
eb04c282 JK |
438 | |
439 | /** | |
440 | * mnt_drop_write - give up write access to a mount | |
441 | * @mnt: the mount on which to give up write access | |
442 | * | |
443 | * Tells the low-level filesystem that we are done performing writes to it and | |
444 | * also allows filesystem to be frozen again. Must be matched with | |
445 | * mnt_want_write() call above. | |
446 | */ | |
447 | void mnt_drop_write(struct vfsmount *mnt) | |
448 | { | |
449 | __mnt_drop_write(mnt); | |
450 | sb_end_write(mnt->mnt_sb); | |
451 | } | |
8366025e DH |
452 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
453 | ||
eb04c282 JK |
454 | void __mnt_drop_write_file(struct file *file) |
455 | { | |
456 | __mnt_drop_write(file->f_path.mnt); | |
457 | } | |
458 | ||
7c6893e3 MS |
459 | void mnt_drop_write_file(struct file *file) |
460 | { | |
a6795a58 | 461 | __mnt_drop_write_file(file); |
7c6893e3 MS |
462 | sb_end_write(file_inode(file)->i_sb); |
463 | } | |
2a79f17e AV |
464 | EXPORT_SYMBOL(mnt_drop_write_file); |
465 | ||
83adc753 | 466 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 467 | { |
3d733633 DH |
468 | int ret = 0; |
469 | ||
719ea2fb | 470 | lock_mount_hash(); |
83adc753 | 471 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 472 | /* |
d3ef3d73 | 473 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
474 | * should be visible before we do. | |
3d733633 | 475 | */ |
d3ef3d73 | 476 | smp_mb(); |
477 | ||
3d733633 | 478 | /* |
d3ef3d73 | 479 | * With writers on hold, if this value is zero, then there are |
480 | * definitely no active writers (although held writers may subsequently | |
481 | * increment the count, they'll have to wait, and decrement it after | |
482 | * seeing MNT_READONLY). | |
483 | * | |
484 | * It is OK to have counter incremented on one CPU and decremented on | |
485 | * another: the sum will add up correctly. The danger would be when we | |
486 | * sum up each counter, if we read a counter before it is incremented, | |
487 | * but then read another CPU's count which it has been subsequently | |
488 | * decremented from -- we would see more decrements than we should. | |
489 | * MNT_WRITE_HOLD protects against this scenario, because | |
490 | * mnt_want_write first increments count, then smp_mb, then spins on | |
491 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
492 | * we're counting up here. | |
3d733633 | 493 | */ |
c6653a83 | 494 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 495 | ret = -EBUSY; |
496 | else | |
83adc753 | 497 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 498 | /* |
499 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
500 | * that become unheld will see MNT_READONLY. | |
501 | */ | |
502 | smp_wmb(); | |
83adc753 | 503 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 504 | unlock_mount_hash(); |
3d733633 | 505 | return ret; |
8366025e | 506 | } |
8366025e | 507 | |
43f5e655 | 508 | static int __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 509 | { |
719ea2fb | 510 | lock_mount_hash(); |
83adc753 | 511 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 512 | unlock_mount_hash(); |
43f5e655 | 513 | return 0; |
2e4b7fcd DH |
514 | } |
515 | ||
4ed5e82f MS |
516 | int sb_prepare_remount_readonly(struct super_block *sb) |
517 | { | |
518 | struct mount *mnt; | |
519 | int err = 0; | |
520 | ||
8e8b8796 MS |
521 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
522 | if (atomic_long_read(&sb->s_remove_count)) | |
523 | return -EBUSY; | |
524 | ||
719ea2fb | 525 | lock_mount_hash(); |
4ed5e82f MS |
526 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
527 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
528 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
529 | smp_mb(); | |
530 | if (mnt_get_writers(mnt) > 0) { | |
531 | err = -EBUSY; | |
532 | break; | |
533 | } | |
534 | } | |
535 | } | |
8e8b8796 MS |
536 | if (!err && atomic_long_read(&sb->s_remove_count)) |
537 | err = -EBUSY; | |
538 | ||
4ed5e82f MS |
539 | if (!err) { |
540 | sb->s_readonly_remount = 1; | |
541 | smp_wmb(); | |
542 | } | |
543 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
544 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
545 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
546 | } | |
719ea2fb | 547 | unlock_mount_hash(); |
4ed5e82f MS |
548 | |
549 | return err; | |
550 | } | |
551 | ||
b105e270 | 552 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 553 | { |
fcc139ae | 554 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 555 | #ifdef CONFIG_SMP |
68e8a9fe | 556 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 557 | #endif |
b105e270 | 558 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
559 | } |
560 | ||
8ffcb32e DH |
561 | static void delayed_free_vfsmnt(struct rcu_head *head) |
562 | { | |
563 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
564 | } | |
565 | ||
48a066e7 | 566 | /* call under rcu_read_lock */ |
294d71ff | 567 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
568 | { |
569 | struct mount *mnt; | |
570 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 571 | return 1; |
48a066e7 | 572 | if (bastard == NULL) |
294d71ff | 573 | return 0; |
48a066e7 AV |
574 | mnt = real_mount(bastard); |
575 | mnt_add_count(mnt, 1); | |
119e1ef8 | 576 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 577 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 578 | return 0; |
48a066e7 AV |
579 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
580 | mnt_add_count(mnt, -1); | |
294d71ff AV |
581 | return 1; |
582 | } | |
119e1ef8 AV |
583 | lock_mount_hash(); |
584 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
585 | mnt_add_count(mnt, -1); | |
586 | unlock_mount_hash(); | |
587 | return 1; | |
588 | } | |
589 | unlock_mount_hash(); | |
590 | /* caller will mntput() */ | |
294d71ff AV |
591 | return -1; |
592 | } | |
593 | ||
594 | /* call under rcu_read_lock */ | |
595 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
596 | { | |
597 | int res = __legitimize_mnt(bastard, seq); | |
598 | if (likely(!res)) | |
599 | return true; | |
600 | if (unlikely(res < 0)) { | |
601 | rcu_read_unlock(); | |
602 | mntput(bastard); | |
603 | rcu_read_lock(); | |
48a066e7 | 604 | } |
48a066e7 AV |
605 | return false; |
606 | } | |
607 | ||
1da177e4 | 608 | /* |
474279dc | 609 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 610 | * call under rcu_read_lock() |
1da177e4 | 611 | */ |
474279dc | 612 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 613 | { |
38129a13 | 614 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
615 | struct mount *p; |
616 | ||
38129a13 | 617 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
618 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
619 | return p; | |
620 | return NULL; | |
621 | } | |
622 | ||
a05964f3 | 623 | /* |
f015f126 DH |
624 | * lookup_mnt - Return the first child mount mounted at path |
625 | * | |
626 | * "First" means first mounted chronologically. If you create the | |
627 | * following mounts: | |
628 | * | |
629 | * mount /dev/sda1 /mnt | |
630 | * mount /dev/sda2 /mnt | |
631 | * mount /dev/sda3 /mnt | |
632 | * | |
633 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
634 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
635 | * /dev/sda2, then /dev/sda3, then NULL. | |
636 | * | |
637 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 638 | */ |
ca71cf71 | 639 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 640 | { |
c7105365 | 641 | struct mount *child_mnt; |
48a066e7 AV |
642 | struct vfsmount *m; |
643 | unsigned seq; | |
99b7db7b | 644 | |
48a066e7 AV |
645 | rcu_read_lock(); |
646 | do { | |
647 | seq = read_seqbegin(&mount_lock); | |
648 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
649 | m = child_mnt ? &child_mnt->mnt : NULL; | |
650 | } while (!legitimize_mnt(m, seq)); | |
651 | rcu_read_unlock(); | |
652 | return m; | |
a05964f3 RP |
653 | } |
654 | ||
7af1364f EB |
655 | /* |
656 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
657 | * current mount namespace. | |
658 | * | |
659 | * The common case is dentries are not mountpoints at all and that | |
660 | * test is handled inline. For the slow case when we are actually | |
661 | * dealing with a mountpoint of some kind, walk through all of the | |
662 | * mounts in the current mount namespace and test to see if the dentry | |
663 | * is a mountpoint. | |
664 | * | |
665 | * The mount_hashtable is not usable in the context because we | |
666 | * need to identify all mounts that may be in the current mount | |
667 | * namespace not just a mount that happens to have some specified | |
668 | * parent mount. | |
669 | */ | |
670 | bool __is_local_mountpoint(struct dentry *dentry) | |
671 | { | |
672 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
673 | struct mount *mnt; | |
674 | bool is_covered = false; | |
675 | ||
676 | if (!d_mountpoint(dentry)) | |
677 | goto out; | |
678 | ||
679 | down_read(&namespace_sem); | |
680 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
681 | is_covered = (mnt->mnt_mountpoint == dentry); | |
682 | if (is_covered) | |
683 | break; | |
684 | } | |
685 | up_read(&namespace_sem); | |
686 | out: | |
687 | return is_covered; | |
688 | } | |
689 | ||
e2dfa935 | 690 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 691 | { |
0818bf27 | 692 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
693 | struct mountpoint *mp; |
694 | ||
0818bf27 | 695 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 696 | if (mp->m_dentry == dentry) { |
84d17192 AV |
697 | mp->m_count++; |
698 | return mp; | |
699 | } | |
700 | } | |
e2dfa935 EB |
701 | return NULL; |
702 | } | |
703 | ||
3895dbf8 | 704 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 705 | { |
3895dbf8 | 706 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 707 | int ret; |
84d17192 | 708 | |
3895dbf8 | 709 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
710 | /* might be worth a WARN_ON() */ |
711 | if (d_unlinked(dentry)) | |
712 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
713 | mountpoint: |
714 | read_seqlock_excl(&mount_lock); | |
715 | mp = lookup_mountpoint(dentry); | |
716 | read_sequnlock_excl(&mount_lock); | |
717 | if (mp) | |
718 | goto done; | |
719 | } | |
720 | ||
721 | if (!new) | |
722 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
723 | if (!new) | |
84d17192 AV |
724 | return ERR_PTR(-ENOMEM); |
725 | ||
3895dbf8 EB |
726 | |
727 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 728 | ret = d_set_mounted(dentry); |
eed81007 | 729 | |
3895dbf8 EB |
730 | /* Someone else set d_mounted? */ |
731 | if (ret == -EBUSY) | |
732 | goto mountpoint; | |
733 | ||
734 | /* The dentry is not available as a mountpoint? */ | |
735 | mp = ERR_PTR(ret); | |
736 | if (ret) | |
737 | goto done; | |
738 | ||
739 | /* Add the new mountpoint to the hash table */ | |
740 | read_seqlock_excl(&mount_lock); | |
741 | new->m_dentry = dentry; | |
742 | new->m_count = 1; | |
743 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
744 | INIT_HLIST_HEAD(&new->m_list); | |
745 | read_sequnlock_excl(&mount_lock); | |
746 | ||
747 | mp = new; | |
748 | new = NULL; | |
749 | done: | |
750 | kfree(new); | |
84d17192 AV |
751 | return mp; |
752 | } | |
753 | ||
754 | static void put_mountpoint(struct mountpoint *mp) | |
755 | { | |
756 | if (!--mp->m_count) { | |
757 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 758 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
759 | spin_lock(&dentry->d_lock); |
760 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
761 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 762 | hlist_del(&mp->m_hash); |
84d17192 AV |
763 | kfree(mp); |
764 | } | |
765 | } | |
766 | ||
143c8c91 | 767 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 768 | { |
6b3286ed | 769 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
770 | } |
771 | ||
99b7db7b NP |
772 | /* |
773 | * vfsmount lock must be held for write | |
774 | */ | |
6b3286ed | 775 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
776 | { |
777 | if (ns) { | |
778 | ns->event = ++event; | |
779 | wake_up_interruptible(&ns->poll); | |
780 | } | |
781 | } | |
782 | ||
99b7db7b NP |
783 | /* |
784 | * vfsmount lock must be held for write | |
785 | */ | |
6b3286ed | 786 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
787 | { |
788 | if (ns && ns->event != event) { | |
789 | ns->event = event; | |
790 | wake_up_interruptible(&ns->poll); | |
791 | } | |
792 | } | |
793 | ||
99b7db7b NP |
794 | /* |
795 | * vfsmount lock must be held for write | |
796 | */ | |
7bdb11de | 797 | static void unhash_mnt(struct mount *mnt) |
419148da | 798 | { |
0714a533 | 799 | mnt->mnt_parent = mnt; |
a73324da | 800 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 801 | list_del_init(&mnt->mnt_child); |
38129a13 | 802 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 803 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
804 | put_mountpoint(mnt->mnt_mp); |
805 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
806 | } |
807 | ||
7bdb11de EB |
808 | /* |
809 | * vfsmount lock must be held for write | |
810 | */ | |
811 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
812 | { | |
813 | old_path->dentry = mnt->mnt_mountpoint; | |
814 | old_path->mnt = &mnt->mnt_parent->mnt; | |
815 | unhash_mnt(mnt); | |
816 | } | |
817 | ||
6a46c573 EB |
818 | /* |
819 | * vfsmount lock must be held for write | |
820 | */ | |
821 | static void umount_mnt(struct mount *mnt) | |
822 | { | |
823 | /* old mountpoint will be dropped when we can do that */ | |
824 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
825 | unhash_mnt(mnt); | |
826 | } | |
827 | ||
99b7db7b NP |
828 | /* |
829 | * vfsmount lock must be held for write | |
830 | */ | |
84d17192 AV |
831 | void mnt_set_mountpoint(struct mount *mnt, |
832 | struct mountpoint *mp, | |
44d964d6 | 833 | struct mount *child_mnt) |
b90fa9ae | 834 | { |
84d17192 | 835 | mp->m_count++; |
3a2393d7 | 836 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 837 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 838 | child_mnt->mnt_parent = mnt; |
84d17192 | 839 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 840 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
841 | } |
842 | ||
1064f874 EB |
843 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
844 | { | |
845 | hlist_add_head_rcu(&mnt->mnt_hash, | |
846 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
847 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
848 | } | |
849 | ||
99b7db7b NP |
850 | /* |
851 | * vfsmount lock must be held for write | |
852 | */ | |
84d17192 AV |
853 | static void attach_mnt(struct mount *mnt, |
854 | struct mount *parent, | |
855 | struct mountpoint *mp) | |
1da177e4 | 856 | { |
84d17192 | 857 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 858 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
859 | } |
860 | ||
1064f874 | 861 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 862 | { |
1064f874 EB |
863 | struct mountpoint *old_mp = mnt->mnt_mp; |
864 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
865 | struct mount *old_parent = mnt->mnt_parent; | |
866 | ||
867 | list_del_init(&mnt->mnt_child); | |
868 | hlist_del_init(&mnt->mnt_mp_list); | |
869 | hlist_del_init_rcu(&mnt->mnt_hash); | |
870 | ||
871 | attach_mnt(mnt, parent, mp); | |
872 | ||
873 | put_mountpoint(old_mp); | |
874 | ||
875 | /* | |
876 | * Safely avoid even the suggestion this code might sleep or | |
877 | * lock the mount hash by taking advantage of the knowledge that | |
878 | * mnt_change_mountpoint will not release the final reference | |
879 | * to a mountpoint. | |
880 | * | |
881 | * During mounting, the mount passed in as the parent mount will | |
882 | * continue to use the old mountpoint and during unmounting, the | |
883 | * old mountpoint will continue to exist until namespace_unlock, | |
884 | * which happens well after mnt_change_mountpoint. | |
885 | */ | |
886 | spin_lock(&old_mountpoint->d_lock); | |
887 | old_mountpoint->d_lockref.count--; | |
888 | spin_unlock(&old_mountpoint->d_lock); | |
889 | ||
890 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
891 | } |
892 | ||
b90fa9ae | 893 | /* |
99b7db7b | 894 | * vfsmount lock must be held for write |
b90fa9ae | 895 | */ |
1064f874 | 896 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 897 | { |
0714a533 | 898 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 899 | struct mount *m; |
b90fa9ae | 900 | LIST_HEAD(head); |
143c8c91 | 901 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 902 | |
0714a533 | 903 | BUG_ON(parent == mnt); |
b90fa9ae | 904 | |
1a4eeaf2 | 905 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 906 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 907 | m->mnt_ns = n; |
f03c6599 | 908 | |
b90fa9ae RP |
909 | list_splice(&head, n->list.prev); |
910 | ||
d2921684 EB |
911 | n->mounts += n->pending_mounts; |
912 | n->pending_mounts = 0; | |
913 | ||
1064f874 | 914 | __attach_mnt(mnt, parent); |
6b3286ed | 915 | touch_mnt_namespace(n); |
1da177e4 LT |
916 | } |
917 | ||
909b0a88 | 918 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 919 | { |
6b41d536 AV |
920 | struct list_head *next = p->mnt_mounts.next; |
921 | if (next == &p->mnt_mounts) { | |
1da177e4 | 922 | while (1) { |
909b0a88 | 923 | if (p == root) |
1da177e4 | 924 | return NULL; |
6b41d536 AV |
925 | next = p->mnt_child.next; |
926 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 927 | break; |
0714a533 | 928 | p = p->mnt_parent; |
1da177e4 LT |
929 | } |
930 | } | |
6b41d536 | 931 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
932 | } |
933 | ||
315fc83e | 934 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 935 | { |
6b41d536 AV |
936 | struct list_head *prev = p->mnt_mounts.prev; |
937 | while (prev != &p->mnt_mounts) { | |
938 | p = list_entry(prev, struct mount, mnt_child); | |
939 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
940 | } |
941 | return p; | |
942 | } | |
943 | ||
9bc61ab1 DH |
944 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
945 | int flags, const char *name, | |
946 | void *data) | |
9d412a43 | 947 | { |
9bc61ab1 | 948 | struct fs_context *fc; |
b105e270 | 949 | struct mount *mnt; |
9bc61ab1 | 950 | int ret = 0; |
9d412a43 AV |
951 | |
952 | if (!type) | |
953 | return ERR_PTR(-ENODEV); | |
954 | ||
9bc61ab1 DH |
955 | fc = fs_context_for_mount(type, flags); |
956 | if (IS_ERR(fc)) | |
957 | return ERR_CAST(fc); | |
958 | ||
959 | if (name) { | |
960 | fc->source = kstrdup(name, GFP_KERNEL); | |
961 | if (!fc->source) | |
962 | ret = -ENOMEM; | |
963 | } | |
964 | if (!ret) | |
965 | ret = parse_monolithic_mount_data(fc, data); | |
966 | if (!ret) | |
967 | ret = vfs_get_tree(fc); | |
968 | if (ret) { | |
969 | put_fs_context(fc); | |
970 | return ERR_PTR(ret); | |
971 | } | |
972 | up_write(&fc->root->d_sb->s_umount); | |
9d412a43 | 973 | mnt = alloc_vfsmnt(name); |
9bc61ab1 DH |
974 | if (!mnt) { |
975 | put_fs_context(fc); | |
9d412a43 | 976 | return ERR_PTR(-ENOMEM); |
9bc61ab1 | 977 | } |
9d412a43 | 978 | |
e462ec50 | 979 | if (flags & SB_KERNMOUNT) |
b105e270 | 980 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 981 | |
9bc61ab1 DH |
982 | atomic_inc(&fc->root->d_sb->s_active); |
983 | mnt->mnt.mnt_root = dget(fc->root); | |
984 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
a73324da | 985 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 986 | mnt->mnt_parent = mnt; |
719ea2fb | 987 | lock_mount_hash(); |
9bc61ab1 | 988 | list_add_tail(&mnt->mnt_instance, &fc->root->d_sb->s_mounts); |
719ea2fb | 989 | unlock_mount_hash(); |
9bc61ab1 | 990 | put_fs_context(fc); |
b105e270 | 991 | return &mnt->mnt; |
9d412a43 AV |
992 | } |
993 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
994 | ||
93faccbb EB |
995 | struct vfsmount * |
996 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
997 | const char *name, void *data) | |
998 | { | |
999 | /* Until it is worked out how to pass the user namespace | |
1000 | * through from the parent mount to the submount don't support | |
1001 | * unprivileged mounts with submounts. | |
1002 | */ | |
1003 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1004 | return ERR_PTR(-EPERM); | |
1005 | ||
e462ec50 | 1006 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1007 | } |
1008 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1009 | ||
87129cc0 | 1010 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1011 | int flag) |
1da177e4 | 1012 | { |
87129cc0 | 1013 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1014 | struct mount *mnt; |
1015 | int err; | |
1da177e4 | 1016 | |
be34d1a3 DH |
1017 | mnt = alloc_vfsmnt(old->mnt_devname); |
1018 | if (!mnt) | |
1019 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1020 | |
7a472ef4 | 1021 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1022 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1023 | else | |
1024 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1025 | |
be34d1a3 DH |
1026 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1027 | err = mnt_alloc_group_id(mnt); | |
1028 | if (err) | |
1029 | goto out_free; | |
1da177e4 | 1030 | } |
be34d1a3 | 1031 | |
16a34adb AV |
1032 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1033 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1034 | |
be34d1a3 DH |
1035 | atomic_inc(&sb->s_active); |
1036 | mnt->mnt.mnt_sb = sb; | |
1037 | mnt->mnt.mnt_root = dget(root); | |
1038 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1039 | mnt->mnt_parent = mnt; | |
719ea2fb | 1040 | lock_mount_hash(); |
be34d1a3 | 1041 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1042 | unlock_mount_hash(); |
be34d1a3 | 1043 | |
7a472ef4 EB |
1044 | if ((flag & CL_SLAVE) || |
1045 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1046 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1047 | mnt->mnt_master = old; | |
1048 | CLEAR_MNT_SHARED(mnt); | |
1049 | } else if (!(flag & CL_PRIVATE)) { | |
1050 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1051 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1052 | if (IS_MNT_SLAVE(old)) | |
1053 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1054 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1055 | } else { |
1056 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1057 | } |
1058 | if (flag & CL_MAKE_SHARED) | |
1059 | set_mnt_shared(mnt); | |
1060 | ||
1061 | /* stick the duplicate mount on the same expiry list | |
1062 | * as the original if that was on one */ | |
1063 | if (flag & CL_EXPIRE) { | |
1064 | if (!list_empty(&old->mnt_expire)) | |
1065 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1066 | } | |
1067 | ||
cb338d06 | 1068 | return mnt; |
719f5d7f MS |
1069 | |
1070 | out_free: | |
8ffcb32e | 1071 | mnt_free_id(mnt); |
719f5d7f | 1072 | free_vfsmnt(mnt); |
be34d1a3 | 1073 | return ERR_PTR(err); |
1da177e4 LT |
1074 | } |
1075 | ||
9ea459e1 AV |
1076 | static void cleanup_mnt(struct mount *mnt) |
1077 | { | |
1078 | /* | |
1079 | * This probably indicates that somebody messed | |
1080 | * up a mnt_want/drop_write() pair. If this | |
1081 | * happens, the filesystem was probably unable | |
1082 | * to make r/w->r/o transitions. | |
1083 | */ | |
1084 | /* | |
1085 | * The locking used to deal with mnt_count decrement provides barriers, | |
1086 | * so mnt_get_writers() below is safe. | |
1087 | */ | |
1088 | WARN_ON(mnt_get_writers(mnt)); | |
1089 | if (unlikely(mnt->mnt_pins.first)) | |
1090 | mnt_pin_kill(mnt); | |
1091 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1092 | dput(mnt->mnt.mnt_root); | |
1093 | deactivate_super(mnt->mnt.mnt_sb); | |
1094 | mnt_free_id(mnt); | |
1095 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1096 | } | |
1097 | ||
1098 | static void __cleanup_mnt(struct rcu_head *head) | |
1099 | { | |
1100 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1101 | } | |
1102 | ||
1103 | static LLIST_HEAD(delayed_mntput_list); | |
1104 | static void delayed_mntput(struct work_struct *unused) | |
1105 | { | |
1106 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1107 | struct mount *m, *t; |
9ea459e1 | 1108 | |
29785735 BP |
1109 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1110 | cleanup_mnt(m); | |
9ea459e1 AV |
1111 | } |
1112 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1113 | ||
900148dc | 1114 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1115 | { |
48a066e7 | 1116 | rcu_read_lock(); |
9ea0a46c AV |
1117 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1118 | /* | |
1119 | * Since we don't do lock_mount_hash() here, | |
1120 | * ->mnt_ns can change under us. However, if it's | |
1121 | * non-NULL, then there's a reference that won't | |
1122 | * be dropped until after an RCU delay done after | |
1123 | * turning ->mnt_ns NULL. So if we observe it | |
1124 | * non-NULL under rcu_read_lock(), the reference | |
1125 | * we are dropping is not the final one. | |
1126 | */ | |
1127 | mnt_add_count(mnt, -1); | |
48a066e7 | 1128 | rcu_read_unlock(); |
f03c6599 | 1129 | return; |
b3e19d92 | 1130 | } |
719ea2fb | 1131 | lock_mount_hash(); |
119e1ef8 AV |
1132 | /* |
1133 | * make sure that if __legitimize_mnt() has not seen us grab | |
1134 | * mount_lock, we'll see their refcount increment here. | |
1135 | */ | |
1136 | smp_mb(); | |
9ea0a46c | 1137 | mnt_add_count(mnt, -1); |
b3e19d92 | 1138 | if (mnt_get_count(mnt)) { |
48a066e7 | 1139 | rcu_read_unlock(); |
719ea2fb | 1140 | unlock_mount_hash(); |
99b7db7b NP |
1141 | return; |
1142 | } | |
48a066e7 AV |
1143 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1144 | rcu_read_unlock(); | |
1145 | unlock_mount_hash(); | |
1146 | return; | |
1147 | } | |
1148 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1149 | rcu_read_unlock(); | |
962830df | 1150 | |
39f7c4db | 1151 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1152 | |
1153 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1154 | struct mount *p, *tmp; | |
1155 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1156 | umount_mnt(p); | |
1157 | } | |
1158 | } | |
719ea2fb | 1159 | unlock_mount_hash(); |
649a795a | 1160 | |
9ea459e1 AV |
1161 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1162 | struct task_struct *task = current; | |
1163 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1164 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1165 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1166 | return; | |
1167 | } | |
1168 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1169 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1170 | return; | |
1171 | } | |
1172 | cleanup_mnt(mnt); | |
b3e19d92 | 1173 | } |
b3e19d92 NP |
1174 | |
1175 | void mntput(struct vfsmount *mnt) | |
1176 | { | |
1177 | if (mnt) { | |
863d684f | 1178 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1179 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1180 | if (unlikely(m->mnt_expiry_mark)) |
1181 | m->mnt_expiry_mark = 0; | |
1182 | mntput_no_expire(m); | |
b3e19d92 NP |
1183 | } |
1184 | } | |
1185 | EXPORT_SYMBOL(mntput); | |
1186 | ||
1187 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1188 | { | |
1189 | if (mnt) | |
83adc753 | 1190 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1191 | return mnt; |
1192 | } | |
1193 | EXPORT_SYMBOL(mntget); | |
1194 | ||
c6609c0a IK |
1195 | /* path_is_mountpoint() - Check if path is a mount in the current |
1196 | * namespace. | |
1197 | * | |
1198 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1199 | * not mounted in any namespace and that common case is handled inline. | |
1200 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1201 | * mounts using a given dentry in a different namespace. This function | |
1202 | * checks if the passed in path is a mountpoint rather than the dentry | |
1203 | * alone. | |
1204 | */ | |
1205 | bool path_is_mountpoint(const struct path *path) | |
1206 | { | |
1207 | unsigned seq; | |
1208 | bool res; | |
1209 | ||
1210 | if (!d_mountpoint(path->dentry)) | |
1211 | return false; | |
1212 | ||
1213 | rcu_read_lock(); | |
1214 | do { | |
1215 | seq = read_seqbegin(&mount_lock); | |
1216 | res = __path_is_mountpoint(path); | |
1217 | } while (read_seqretry(&mount_lock, seq)); | |
1218 | rcu_read_unlock(); | |
1219 | ||
1220 | return res; | |
1221 | } | |
1222 | EXPORT_SYMBOL(path_is_mountpoint); | |
1223 | ||
ca71cf71 | 1224 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1225 | { |
3064c356 AV |
1226 | struct mount *p; |
1227 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1228 | if (IS_ERR(p)) | |
1229 | return ERR_CAST(p); | |
1230 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1231 | return &p->mnt; | |
7b7b1ace | 1232 | } |
1da177e4 | 1233 | |
a1a2c409 | 1234 | #ifdef CONFIG_PROC_FS |
0226f492 | 1235 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1236 | static void *m_start(struct seq_file *m, loff_t *pos) |
1237 | { | |
ede1bf0d | 1238 | struct proc_mounts *p = m->private; |
1da177e4 | 1239 | |
390c6843 | 1240 | down_read(&namespace_sem); |
c7999c36 AV |
1241 | if (p->cached_event == p->ns->event) { |
1242 | void *v = p->cached_mount; | |
1243 | if (*pos == p->cached_index) | |
1244 | return v; | |
1245 | if (*pos == p->cached_index + 1) { | |
1246 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1247 | return p->cached_mount = v; | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | p->cached_event = p->ns->event; | |
1252 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1253 | p->cached_index = *pos; | |
1254 | return p->cached_mount; | |
1da177e4 LT |
1255 | } |
1256 | ||
1257 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1258 | { | |
ede1bf0d | 1259 | struct proc_mounts *p = m->private; |
b0765fb8 | 1260 | |
c7999c36 AV |
1261 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1262 | p->cached_index = *pos; | |
1263 | return p->cached_mount; | |
1da177e4 LT |
1264 | } |
1265 | ||
1266 | static void m_stop(struct seq_file *m, void *v) | |
1267 | { | |
390c6843 | 1268 | up_read(&namespace_sem); |
1da177e4 LT |
1269 | } |
1270 | ||
0226f492 | 1271 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1272 | { |
ede1bf0d | 1273 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1274 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1275 | return p->show(m, &r->mnt); |
1da177e4 LT |
1276 | } |
1277 | ||
a1a2c409 | 1278 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1279 | .start = m_start, |
1280 | .next = m_next, | |
1281 | .stop = m_stop, | |
0226f492 | 1282 | .show = m_show, |
b4629fe2 | 1283 | }; |
a1a2c409 | 1284 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1285 | |
1da177e4 LT |
1286 | /** |
1287 | * may_umount_tree - check if a mount tree is busy | |
1288 | * @mnt: root of mount tree | |
1289 | * | |
1290 | * This is called to check if a tree of mounts has any | |
1291 | * open files, pwds, chroots or sub mounts that are | |
1292 | * busy. | |
1293 | */ | |
909b0a88 | 1294 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1295 | { |
909b0a88 | 1296 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1297 | int actual_refs = 0; |
1298 | int minimum_refs = 0; | |
315fc83e | 1299 | struct mount *p; |
909b0a88 | 1300 | BUG_ON(!m); |
1da177e4 | 1301 | |
b3e19d92 | 1302 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1303 | lock_mount_hash(); |
909b0a88 | 1304 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1305 | actual_refs += mnt_get_count(p); |
1da177e4 | 1306 | minimum_refs += 2; |
1da177e4 | 1307 | } |
719ea2fb | 1308 | unlock_mount_hash(); |
1da177e4 LT |
1309 | |
1310 | if (actual_refs > minimum_refs) | |
e3474a8e | 1311 | return 0; |
1da177e4 | 1312 | |
e3474a8e | 1313 | return 1; |
1da177e4 LT |
1314 | } |
1315 | ||
1316 | EXPORT_SYMBOL(may_umount_tree); | |
1317 | ||
1318 | /** | |
1319 | * may_umount - check if a mount point is busy | |
1320 | * @mnt: root of mount | |
1321 | * | |
1322 | * This is called to check if a mount point has any | |
1323 | * open files, pwds, chroots or sub mounts. If the | |
1324 | * mount has sub mounts this will return busy | |
1325 | * regardless of whether the sub mounts are busy. | |
1326 | * | |
1327 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1328 | * give false negatives. The main reason why it's here is that we need | |
1329 | * a non-destructive way to look for easily umountable filesystems. | |
1330 | */ | |
1331 | int may_umount(struct vfsmount *mnt) | |
1332 | { | |
e3474a8e | 1333 | int ret = 1; |
8ad08d8a | 1334 | down_read(&namespace_sem); |
719ea2fb | 1335 | lock_mount_hash(); |
1ab59738 | 1336 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1337 | ret = 0; |
719ea2fb | 1338 | unlock_mount_hash(); |
8ad08d8a | 1339 | up_read(&namespace_sem); |
a05964f3 | 1340 | return ret; |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | EXPORT_SYMBOL(may_umount); | |
1344 | ||
38129a13 | 1345 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1346 | |
97216be0 | 1347 | static void namespace_unlock(void) |
70fbcdf4 | 1348 | { |
a3b3c562 | 1349 | struct hlist_head head; |
97216be0 | 1350 | |
a3b3c562 | 1351 | hlist_move_list(&unmounted, &head); |
97216be0 | 1352 | |
97216be0 AV |
1353 | up_write(&namespace_sem); |
1354 | ||
a3b3c562 EB |
1355 | if (likely(hlist_empty(&head))) |
1356 | return; | |
1357 | ||
22cb7405 | 1358 | synchronize_rcu_expedited(); |
48a066e7 | 1359 | |
87b95ce0 | 1360 | group_pin_kill(&head); |
70fbcdf4 RP |
1361 | } |
1362 | ||
97216be0 | 1363 | static inline void namespace_lock(void) |
e3197d83 | 1364 | { |
97216be0 | 1365 | down_write(&namespace_sem); |
e3197d83 AV |
1366 | } |
1367 | ||
e819f152 EB |
1368 | enum umount_tree_flags { |
1369 | UMOUNT_SYNC = 1, | |
1370 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1371 | UMOUNT_CONNECTED = 4, |
e819f152 | 1372 | }; |
f2d0a123 EB |
1373 | |
1374 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1375 | { | |
1376 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1377 | if (how & UMOUNT_SYNC) | |
1378 | return true; | |
1379 | ||
1380 | /* A mount without a parent has nothing to be connected to */ | |
1381 | if (!mnt_has_parent(mnt)) | |
1382 | return true; | |
1383 | ||
1384 | /* Because the reference counting rules change when mounts are | |
1385 | * unmounted and connected, umounted mounts may not be | |
1386 | * connected to mounted mounts. | |
1387 | */ | |
1388 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1389 | return true; | |
1390 | ||
1391 | /* Has it been requested that the mount remain connected? */ | |
1392 | if (how & UMOUNT_CONNECTED) | |
1393 | return false; | |
1394 | ||
1395 | /* Is the mount locked such that it needs to remain connected? */ | |
1396 | if (IS_MNT_LOCKED(mnt)) | |
1397 | return false; | |
1398 | ||
1399 | /* By default disconnect the mount */ | |
1400 | return true; | |
1401 | } | |
1402 | ||
99b7db7b | 1403 | /* |
48a066e7 | 1404 | * mount_lock must be held |
99b7db7b NP |
1405 | * namespace_sem must be held for write |
1406 | */ | |
e819f152 | 1407 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1408 | { |
c003b26f | 1409 | LIST_HEAD(tmp_list); |
315fc83e | 1410 | struct mount *p; |
1da177e4 | 1411 | |
5d88457e EB |
1412 | if (how & UMOUNT_PROPAGATE) |
1413 | propagate_mount_unlock(mnt); | |
1414 | ||
c003b26f | 1415 | /* Gather the mounts to umount */ |
590ce4bc EB |
1416 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1417 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1418 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1419 | } |
1da177e4 | 1420 | |
411a938b | 1421 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1422 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1423 | list_del_init(&p->mnt_child); |
c003b26f | 1424 | } |
88b368f2 | 1425 | |
c003b26f | 1426 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1427 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1428 | propagate_umount(&tmp_list); |
a05964f3 | 1429 | |
c003b26f | 1430 | while (!list_empty(&tmp_list)) { |
d2921684 | 1431 | struct mnt_namespace *ns; |
ce07d891 | 1432 | bool disconnect; |
c003b26f | 1433 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1434 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1435 | list_del_init(&p->mnt_list); |
d2921684 EB |
1436 | ns = p->mnt_ns; |
1437 | if (ns) { | |
1438 | ns->mounts--; | |
1439 | __touch_mnt_namespace(ns); | |
1440 | } | |
143c8c91 | 1441 | p->mnt_ns = NULL; |
e819f152 | 1442 | if (how & UMOUNT_SYNC) |
48a066e7 | 1443 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1444 | |
f2d0a123 | 1445 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1446 | |
1447 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1448 | disconnect ? &unmounted : NULL); | |
676da58d | 1449 | if (mnt_has_parent(p)) { |
81b6b061 | 1450 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1451 | if (!disconnect) { |
1452 | /* Don't forget about p */ | |
1453 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1454 | } else { | |
1455 | umount_mnt(p); | |
1456 | } | |
7c4b93d8 | 1457 | } |
0f0afb1d | 1458 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1459 | } |
1460 | } | |
1461 | ||
b54b9be7 | 1462 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1463 | |
1ab59738 | 1464 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1465 | { |
1ab59738 | 1466 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1467 | int retval; |
1468 | ||
1ab59738 | 1469 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1470 | if (retval) |
1471 | return retval; | |
1472 | ||
1473 | /* | |
1474 | * Allow userspace to request a mountpoint be expired rather than | |
1475 | * unmounting unconditionally. Unmount only happens if: | |
1476 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1477 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1478 | */ | |
1479 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1480 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1481 | flags & (MNT_FORCE | MNT_DETACH)) |
1482 | return -EINVAL; | |
1483 | ||
b3e19d92 NP |
1484 | /* |
1485 | * probably don't strictly need the lock here if we examined | |
1486 | * all race cases, but it's a slowpath. | |
1487 | */ | |
719ea2fb | 1488 | lock_mount_hash(); |
83adc753 | 1489 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1490 | unlock_mount_hash(); |
1da177e4 | 1491 | return -EBUSY; |
b3e19d92 | 1492 | } |
719ea2fb | 1493 | unlock_mount_hash(); |
1da177e4 | 1494 | |
863d684f | 1495 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1496 | return -EAGAIN; |
1497 | } | |
1498 | ||
1499 | /* | |
1500 | * If we may have to abort operations to get out of this | |
1501 | * mount, and they will themselves hold resources we must | |
1502 | * allow the fs to do things. In the Unix tradition of | |
1503 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1504 | * might fail to complete on the first run through as other tasks | |
1505 | * must return, and the like. Thats for the mount program to worry | |
1506 | * about for the moment. | |
1507 | */ | |
1508 | ||
42faad99 | 1509 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1510 | sb->s_op->umount_begin(sb); |
42faad99 | 1511 | } |
1da177e4 LT |
1512 | |
1513 | /* | |
1514 | * No sense to grab the lock for this test, but test itself looks | |
1515 | * somewhat bogus. Suggestions for better replacement? | |
1516 | * Ho-hum... In principle, we might treat that as umount + switch | |
1517 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1518 | * Actually it makes sense, especially if rootfs would contain a | |
1519 | * /reboot - static binary that would close all descriptors and | |
1520 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1521 | */ | |
1ab59738 | 1522 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1523 | /* |
1524 | * Special case for "unmounting" root ... | |
1525 | * we just try to remount it readonly. | |
1526 | */ | |
bc6155d1 | 1527 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1528 | return -EPERM; |
1da177e4 | 1529 | down_write(&sb->s_umount); |
bc98a42c | 1530 | if (!sb_rdonly(sb)) |
e462ec50 | 1531 | retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); |
1da177e4 LT |
1532 | up_write(&sb->s_umount); |
1533 | return retval; | |
1534 | } | |
1535 | ||
97216be0 | 1536 | namespace_lock(); |
719ea2fb | 1537 | lock_mount_hash(); |
1da177e4 | 1538 | |
25d202ed EB |
1539 | /* Recheck MNT_LOCKED with the locks held */ |
1540 | retval = -EINVAL; | |
1541 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1542 | goto out; | |
1543 | ||
1544 | event++; | |
48a066e7 | 1545 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1546 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1547 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1548 | retval = 0; |
48a066e7 AV |
1549 | } else { |
1550 | shrink_submounts(mnt); | |
1551 | retval = -EBUSY; | |
1552 | if (!propagate_mount_busy(mnt, 2)) { | |
1553 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1554 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1555 | retval = 0; |
1556 | } | |
1da177e4 | 1557 | } |
25d202ed | 1558 | out: |
719ea2fb | 1559 | unlock_mount_hash(); |
e3197d83 | 1560 | namespace_unlock(); |
1da177e4 LT |
1561 | return retval; |
1562 | } | |
1563 | ||
80b5dce8 EB |
1564 | /* |
1565 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1566 | * | |
1567 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1568 | * to an existing mountpoint, and wind up leaking the mount. | |
1569 | * detach_mounts allows lazily unmounting those mounts instead of | |
1570 | * leaking them. | |
1571 | * | |
1572 | * The caller may hold dentry->d_inode->i_mutex. | |
1573 | */ | |
1574 | void __detach_mounts(struct dentry *dentry) | |
1575 | { | |
1576 | struct mountpoint *mp; | |
1577 | struct mount *mnt; | |
1578 | ||
1579 | namespace_lock(); | |
3895dbf8 | 1580 | lock_mount_hash(); |
80b5dce8 | 1581 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1582 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1583 | goto out_unlock; |
1584 | ||
e06b933e | 1585 | event++; |
80b5dce8 EB |
1586 | while (!hlist_empty(&mp->m_list)) { |
1587 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1588 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1589 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1590 | umount_mnt(mnt); | |
ce07d891 | 1591 | } |
e0c9c0af | 1592 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1593 | } |
80b5dce8 EB |
1594 | put_mountpoint(mp); |
1595 | out_unlock: | |
3895dbf8 | 1596 | unlock_mount_hash(); |
80b5dce8 EB |
1597 | namespace_unlock(); |
1598 | } | |
1599 | ||
dd111b31 | 1600 | /* |
9b40bc90 AV |
1601 | * Is the caller allowed to modify his namespace? |
1602 | */ | |
1603 | static inline bool may_mount(void) | |
1604 | { | |
1605 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1606 | } | |
1607 | ||
9e8925b6 JL |
1608 | static inline bool may_mandlock(void) |
1609 | { | |
1610 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1611 | return false; | |
1612 | #endif | |
95ace754 | 1613 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1614 | } |
1615 | ||
1da177e4 LT |
1616 | /* |
1617 | * Now umount can handle mount points as well as block devices. | |
1618 | * This is important for filesystems which use unnamed block devices. | |
1619 | * | |
1620 | * We now support a flag for forced unmount like the other 'big iron' | |
1621 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1622 | */ | |
1623 | ||
3a18ef5c | 1624 | int ksys_umount(char __user *name, int flags) |
1da177e4 | 1625 | { |
2d8f3038 | 1626 | struct path path; |
900148dc | 1627 | struct mount *mnt; |
1da177e4 | 1628 | int retval; |
db1f05bb | 1629 | int lookup_flags = 0; |
1da177e4 | 1630 | |
db1f05bb MS |
1631 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1632 | return -EINVAL; | |
1633 | ||
9b40bc90 AV |
1634 | if (!may_mount()) |
1635 | return -EPERM; | |
1636 | ||
db1f05bb MS |
1637 | if (!(flags & UMOUNT_NOFOLLOW)) |
1638 | lookup_flags |= LOOKUP_FOLLOW; | |
1639 | ||
197df04c | 1640 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1641 | if (retval) |
1642 | goto out; | |
900148dc | 1643 | mnt = real_mount(path.mnt); |
1da177e4 | 1644 | retval = -EINVAL; |
2d8f3038 | 1645 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1646 | goto dput_and_out; |
143c8c91 | 1647 | if (!check_mnt(mnt)) |
1da177e4 | 1648 | goto dput_and_out; |
25d202ed | 1649 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
5ff9d8a6 | 1650 | goto dput_and_out; |
b2f5d4dc EB |
1651 | retval = -EPERM; |
1652 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1653 | goto dput_and_out; | |
1da177e4 | 1654 | |
900148dc | 1655 | retval = do_umount(mnt, flags); |
1da177e4 | 1656 | dput_and_out: |
429731b1 | 1657 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1658 | dput(path.dentry); |
900148dc | 1659 | mntput_no_expire(mnt); |
1da177e4 LT |
1660 | out: |
1661 | return retval; | |
1662 | } | |
1663 | ||
3a18ef5c DB |
1664 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1665 | { | |
1666 | return ksys_umount(name, flags); | |
1667 | } | |
1668 | ||
1da177e4 LT |
1669 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1670 | ||
1671 | /* | |
b58fed8b | 1672 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1673 | */ |
bdc480e3 | 1674 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1675 | { |
3a18ef5c | 1676 | return ksys_umount(name, 0); |
1da177e4 LT |
1677 | } |
1678 | ||
1679 | #endif | |
1680 | ||
4ce5d2b1 | 1681 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1682 | { |
4ce5d2b1 | 1683 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1684 | return dentry->d_op == &ns_dentry_operations && |
1685 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1686 | } |
1687 | ||
58be2825 AV |
1688 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1689 | { | |
1690 | return container_of(ns, struct mnt_namespace, ns); | |
1691 | } | |
1692 | ||
4ce5d2b1 EB |
1693 | static bool mnt_ns_loop(struct dentry *dentry) |
1694 | { | |
1695 | /* Could bind mounting the mount namespace inode cause a | |
1696 | * mount namespace loop? | |
1697 | */ | |
1698 | struct mnt_namespace *mnt_ns; | |
1699 | if (!is_mnt_ns_file(dentry)) | |
1700 | return false; | |
1701 | ||
f77c8014 | 1702 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1703 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1704 | } | |
1705 | ||
87129cc0 | 1706 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1707 | int flag) |
1da177e4 | 1708 | { |
84d17192 | 1709 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1710 | |
4ce5d2b1 EB |
1711 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1712 | return ERR_PTR(-EINVAL); | |
1713 | ||
1714 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1715 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1716 | |
36341f64 | 1717 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1718 | if (IS_ERR(q)) |
1719 | return q; | |
1720 | ||
a73324da | 1721 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1722 | |
1723 | p = mnt; | |
6b41d536 | 1724 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1725 | struct mount *s; |
7ec02ef1 | 1726 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1727 | continue; |
1728 | ||
909b0a88 | 1729 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1730 | if (!(flag & CL_COPY_UNBINDABLE) && |
1731 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1732 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1733 | /* Both unbindable and locked. */ | |
1734 | q = ERR_PTR(-EPERM); | |
1735 | goto out; | |
1736 | } else { | |
1737 | s = skip_mnt_tree(s); | |
1738 | continue; | |
1739 | } | |
4ce5d2b1 EB |
1740 | } |
1741 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1742 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1743 | s = skip_mnt_tree(s); |
1744 | continue; | |
1745 | } | |
0714a533 AV |
1746 | while (p != s->mnt_parent) { |
1747 | p = p->mnt_parent; | |
1748 | q = q->mnt_parent; | |
1da177e4 | 1749 | } |
87129cc0 | 1750 | p = s; |
84d17192 | 1751 | parent = q; |
87129cc0 | 1752 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1753 | if (IS_ERR(q)) |
1754 | goto out; | |
719ea2fb | 1755 | lock_mount_hash(); |
1a4eeaf2 | 1756 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1757 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1758 | unlock_mount_hash(); |
1da177e4 LT |
1759 | } |
1760 | } | |
1761 | return res; | |
be34d1a3 | 1762 | out: |
1da177e4 | 1763 | if (res) { |
719ea2fb | 1764 | lock_mount_hash(); |
e819f152 | 1765 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1766 | unlock_mount_hash(); |
1da177e4 | 1767 | } |
be34d1a3 | 1768 | return q; |
1da177e4 LT |
1769 | } |
1770 | ||
be34d1a3 DH |
1771 | /* Caller should check returned pointer for errors */ |
1772 | ||
ca71cf71 | 1773 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1774 | { |
cb338d06 | 1775 | struct mount *tree; |
97216be0 | 1776 | namespace_lock(); |
cd4a4017 EB |
1777 | if (!check_mnt(real_mount(path->mnt))) |
1778 | tree = ERR_PTR(-EINVAL); | |
1779 | else | |
1780 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1781 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1782 | namespace_unlock(); |
be34d1a3 | 1783 | if (IS_ERR(tree)) |
52e220d3 | 1784 | return ERR_CAST(tree); |
be34d1a3 | 1785 | return &tree->mnt; |
8aec0809 AV |
1786 | } |
1787 | ||
1788 | void drop_collected_mounts(struct vfsmount *mnt) | |
1789 | { | |
97216be0 | 1790 | namespace_lock(); |
719ea2fb | 1791 | lock_mount_hash(); |
9c8e0a1b | 1792 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1793 | unlock_mount_hash(); |
3ab6abee | 1794 | namespace_unlock(); |
8aec0809 AV |
1795 | } |
1796 | ||
c771d683 MS |
1797 | /** |
1798 | * clone_private_mount - create a private clone of a path | |
1799 | * | |
1800 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1801 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1802 | * to the originating mount won't be propagated into this). | |
1803 | * | |
1804 | * Release with mntput(). | |
1805 | */ | |
ca71cf71 | 1806 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1807 | { |
1808 | struct mount *old_mnt = real_mount(path->mnt); | |
1809 | struct mount *new_mnt; | |
1810 | ||
1811 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1812 | return ERR_PTR(-EINVAL); | |
1813 | ||
c771d683 | 1814 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1815 | if (IS_ERR(new_mnt)) |
1816 | return ERR_CAST(new_mnt); | |
1817 | ||
1818 | return &new_mnt->mnt; | |
1819 | } | |
1820 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1821 | ||
1f707137 AV |
1822 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1823 | struct vfsmount *root) | |
1824 | { | |
1a4eeaf2 | 1825 | struct mount *mnt; |
1f707137 AV |
1826 | int res = f(root, arg); |
1827 | if (res) | |
1828 | return res; | |
1a4eeaf2 AV |
1829 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1830 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1831 | if (res) |
1832 | return res; | |
1833 | } | |
1834 | return 0; | |
1835 | } | |
1836 | ||
3bd045cc AV |
1837 | static void lock_mnt_tree(struct mount *mnt) |
1838 | { | |
1839 | struct mount *p; | |
1840 | ||
1841 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1842 | int flags = p->mnt.mnt_flags; | |
1843 | /* Don't allow unprivileged users to change mount flags */ | |
1844 | flags |= MNT_LOCK_ATIME; | |
1845 | ||
1846 | if (flags & MNT_READONLY) | |
1847 | flags |= MNT_LOCK_READONLY; | |
1848 | ||
1849 | if (flags & MNT_NODEV) | |
1850 | flags |= MNT_LOCK_NODEV; | |
1851 | ||
1852 | if (flags & MNT_NOSUID) | |
1853 | flags |= MNT_LOCK_NOSUID; | |
1854 | ||
1855 | if (flags & MNT_NOEXEC) | |
1856 | flags |= MNT_LOCK_NOEXEC; | |
1857 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
1858 | if (list_empty(&p->mnt_expire)) | |
1859 | flags |= MNT_LOCKED; | |
1860 | p->mnt.mnt_flags = flags; | |
1861 | } | |
1862 | } | |
1863 | ||
4b8b21f4 | 1864 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1865 | { |
315fc83e | 1866 | struct mount *p; |
719f5d7f | 1867 | |
909b0a88 | 1868 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1869 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1870 | mnt_release_group_id(p); |
719f5d7f MS |
1871 | } |
1872 | } | |
1873 | ||
4b8b21f4 | 1874 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1875 | { |
315fc83e | 1876 | struct mount *p; |
719f5d7f | 1877 | |
909b0a88 | 1878 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1879 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1880 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1881 | if (err) { |
4b8b21f4 | 1882 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1883 | return err; |
1884 | } | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | return 0; | |
1889 | } | |
1890 | ||
d2921684 EB |
1891 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1892 | { | |
1893 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1894 | unsigned int mounts = 0, old, pending, sum; | |
1895 | struct mount *p; | |
1896 | ||
1897 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1898 | mounts++; | |
1899 | ||
1900 | old = ns->mounts; | |
1901 | pending = ns->pending_mounts; | |
1902 | sum = old + pending; | |
1903 | if ((old > sum) || | |
1904 | (pending > sum) || | |
1905 | (max < sum) || | |
1906 | (mounts > (max - sum))) | |
1907 | return -ENOSPC; | |
1908 | ||
1909 | ns->pending_mounts = pending + mounts; | |
1910 | return 0; | |
1911 | } | |
1912 | ||
b90fa9ae RP |
1913 | /* |
1914 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1915 | * @nd : place the mount tree @source_mnt is attached |
1916 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1917 | * store the parent mount and mountpoint dentry. | |
1918 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1919 | * |
1920 | * NOTE: in the table below explains the semantics when a source mount | |
1921 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1922 | * --------------------------------------------------------------------------- |
1923 | * | BIND MOUNT OPERATION | | |
1924 | * |************************************************************************** | |
1925 | * | source-->| shared | private | slave | unbindable | | |
1926 | * | dest | | | | | | |
1927 | * | | | | | | | | |
1928 | * | v | | | | | | |
1929 | * |************************************************************************** | |
1930 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1931 | * | | | | | | | |
1932 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1933 | * *************************************************************************** | |
b90fa9ae RP |
1934 | * A bind operation clones the source mount and mounts the clone on the |
1935 | * destination mount. | |
1936 | * | |
1937 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1938 | * tree of the destination mount and the cloned mount is added to | |
1939 | * the peer group of the source mount. | |
1940 | * (+) the cloned mount is created under the destination mount and is marked | |
1941 | * as shared. The cloned mount is added to the peer group of the source | |
1942 | * mount. | |
5afe0022 RP |
1943 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1944 | * of the destination mount and the cloned mount is made slave | |
1945 | * of the same master as that of the source mount. The cloned mount | |
1946 | * is marked as 'shared and slave'. | |
1947 | * (*) the cloned mount is made a slave of the same master as that of the | |
1948 | * source mount. | |
1949 | * | |
9676f0c6 RP |
1950 | * --------------------------------------------------------------------------- |
1951 | * | MOVE MOUNT OPERATION | | |
1952 | * |************************************************************************** | |
1953 | * | source-->| shared | private | slave | unbindable | | |
1954 | * | dest | | | | | | |
1955 | * | | | | | | | | |
1956 | * | v | | | | | | |
1957 | * |************************************************************************** | |
1958 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1959 | * | | | | | | | |
1960 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1961 | * *************************************************************************** | |
5afe0022 RP |
1962 | * |
1963 | * (+) the mount is moved to the destination. And is then propagated to | |
1964 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1965 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1966 | * (+++) the mount is moved to the destination and is then propagated to |
1967 | * all the mounts belonging to the destination mount's propagation tree. | |
1968 | * the mount is marked as 'shared and slave'. | |
1969 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1970 | * |
1971 | * if the source mount is a tree, the operations explained above is | |
1972 | * applied to each mount in the tree. | |
1973 | * Must be called without spinlocks held, since this function can sleep | |
1974 | * in allocations. | |
1975 | */ | |
0fb54e50 | 1976 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1977 | struct mount *dest_mnt, |
1978 | struct mountpoint *dest_mp, | |
1979 | struct path *parent_path) | |
b90fa9ae | 1980 | { |
3bd045cc | 1981 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 1982 | HLIST_HEAD(tree_list); |
d2921684 | 1983 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 1984 | struct mountpoint *smp; |
315fc83e | 1985 | struct mount *child, *p; |
38129a13 | 1986 | struct hlist_node *n; |
719f5d7f | 1987 | int err; |
b90fa9ae | 1988 | |
1064f874 EB |
1989 | /* Preallocate a mountpoint in case the new mounts need |
1990 | * to be tucked under other mounts. | |
1991 | */ | |
1992 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
1993 | if (IS_ERR(smp)) | |
1994 | return PTR_ERR(smp); | |
1995 | ||
d2921684 EB |
1996 | /* Is there space to add these mounts to the mount namespace? */ |
1997 | if (!parent_path) { | |
1998 | err = count_mounts(ns, source_mnt); | |
1999 | if (err) | |
2000 | goto out; | |
2001 | } | |
2002 | ||
fc7be130 | 2003 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2004 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2005 | if (err) |
2006 | goto out; | |
0b1b901b | 2007 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2008 | lock_mount_hash(); |
0b1b901b AV |
2009 | if (err) |
2010 | goto out_cleanup_ids; | |
909b0a88 | 2011 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2012 | set_mnt_shared(p); |
0b1b901b AV |
2013 | } else { |
2014 | lock_mount_hash(); | |
b90fa9ae | 2015 | } |
1a390689 | 2016 | if (parent_path) { |
0fb54e50 | 2017 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2018 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2019 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2020 | } else { |
84d17192 | 2021 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2022 | commit_tree(source_mnt); |
21444403 | 2023 | } |
b90fa9ae | 2024 | |
38129a13 | 2025 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2026 | struct mount *q; |
38129a13 | 2027 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2028 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2029 | child->mnt_mountpoint); | |
2030 | if (q) | |
2031 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2032 | /* Notice when we are propagating across user namespaces */ |
2033 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2034 | lock_mnt_tree(child); | |
1064f874 | 2035 | commit_tree(child); |
b90fa9ae | 2036 | } |
1064f874 | 2037 | put_mountpoint(smp); |
719ea2fb | 2038 | unlock_mount_hash(); |
99b7db7b | 2039 | |
b90fa9ae | 2040 | return 0; |
719f5d7f MS |
2041 | |
2042 | out_cleanup_ids: | |
f2ebb3a9 AV |
2043 | while (!hlist_empty(&tree_list)) { |
2044 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2045 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2046 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2047 | } |
2048 | unlock_mount_hash(); | |
0b1b901b | 2049 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2050 | out: |
d2921684 | 2051 | ns->pending_mounts = 0; |
1064f874 EB |
2052 | |
2053 | read_seqlock_excl(&mount_lock); | |
2054 | put_mountpoint(smp); | |
2055 | read_sequnlock_excl(&mount_lock); | |
2056 | ||
719f5d7f | 2057 | return err; |
b90fa9ae RP |
2058 | } |
2059 | ||
84d17192 | 2060 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2061 | { |
2062 | struct vfsmount *mnt; | |
84d17192 | 2063 | struct dentry *dentry = path->dentry; |
b12cea91 | 2064 | retry: |
5955102c | 2065 | inode_lock(dentry->d_inode); |
84d17192 | 2066 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2067 | inode_unlock(dentry->d_inode); |
84d17192 | 2068 | return ERR_PTR(-ENOENT); |
b12cea91 | 2069 | } |
97216be0 | 2070 | namespace_lock(); |
b12cea91 | 2071 | mnt = lookup_mnt(path); |
84d17192 | 2072 | if (likely(!mnt)) { |
3895dbf8 | 2073 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2074 | if (IS_ERR(mp)) { |
97216be0 | 2075 | namespace_unlock(); |
5955102c | 2076 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2077 | return mp; |
2078 | } | |
2079 | return mp; | |
2080 | } | |
97216be0 | 2081 | namespace_unlock(); |
5955102c | 2082 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2083 | path_put(path); |
2084 | path->mnt = mnt; | |
84d17192 | 2085 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2086 | goto retry; |
2087 | } | |
2088 | ||
84d17192 | 2089 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2090 | { |
84d17192 | 2091 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2092 | |
2093 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2094 | put_mountpoint(where); |
3895dbf8 EB |
2095 | read_sequnlock_excl(&mount_lock); |
2096 | ||
328e6d90 | 2097 | namespace_unlock(); |
5955102c | 2098 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2099 | } |
2100 | ||
84d17192 | 2101 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2102 | { |
e462ec50 | 2103 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2104 | return -EINVAL; |
2105 | ||
e36cb0b8 DH |
2106 | if (d_is_dir(mp->m_dentry) != |
2107 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2108 | return -ENOTDIR; |
2109 | ||
84d17192 | 2110 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2111 | } |
2112 | ||
7a2e8a8f VA |
2113 | /* |
2114 | * Sanity check the flags to change_mnt_propagation. | |
2115 | */ | |
2116 | ||
e462ec50 | 2117 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2118 | { |
e462ec50 | 2119 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2120 | |
2121 | /* Fail if any non-propagation flags are set */ | |
2122 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2123 | return 0; | |
2124 | /* Only one propagation flag should be set */ | |
2125 | if (!is_power_of_2(type)) | |
2126 | return 0; | |
2127 | return type; | |
2128 | } | |
2129 | ||
07b20889 RP |
2130 | /* |
2131 | * recursively change the type of the mountpoint. | |
2132 | */ | |
e462ec50 | 2133 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2134 | { |
315fc83e | 2135 | struct mount *m; |
4b8b21f4 | 2136 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2137 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2138 | int type; |
719f5d7f | 2139 | int err = 0; |
07b20889 | 2140 | |
2d92ab3c | 2141 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2142 | return -EINVAL; |
2143 | ||
e462ec50 | 2144 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2145 | if (!type) |
2146 | return -EINVAL; | |
2147 | ||
97216be0 | 2148 | namespace_lock(); |
719f5d7f MS |
2149 | if (type == MS_SHARED) { |
2150 | err = invent_group_ids(mnt, recurse); | |
2151 | if (err) | |
2152 | goto out_unlock; | |
2153 | } | |
2154 | ||
719ea2fb | 2155 | lock_mount_hash(); |
909b0a88 | 2156 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2157 | change_mnt_propagation(m, type); |
719ea2fb | 2158 | unlock_mount_hash(); |
719f5d7f MS |
2159 | |
2160 | out_unlock: | |
97216be0 | 2161 | namespace_unlock(); |
719f5d7f | 2162 | return err; |
07b20889 RP |
2163 | } |
2164 | ||
5ff9d8a6 EB |
2165 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2166 | { | |
2167 | struct mount *child; | |
2168 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2169 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2170 | continue; | |
2171 | ||
2172 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2173 | return true; | |
2174 | } | |
2175 | return false; | |
2176 | } | |
2177 | ||
1da177e4 LT |
2178 | /* |
2179 | * do loopback mount. | |
2180 | */ | |
808d4e3c | 2181 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2182 | int recurse) |
1da177e4 | 2183 | { |
2d92ab3c | 2184 | struct path old_path; |
84d17192 AV |
2185 | struct mount *mnt = NULL, *old, *parent; |
2186 | struct mountpoint *mp; | |
57eccb83 | 2187 | int err; |
1da177e4 LT |
2188 | if (!old_name || !*old_name) |
2189 | return -EINVAL; | |
815d405c | 2190 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2191 | if (err) |
2192 | return err; | |
2193 | ||
8823c079 | 2194 | err = -EINVAL; |
4ce5d2b1 | 2195 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2196 | goto out; |
8823c079 | 2197 | |
84d17192 AV |
2198 | mp = lock_mount(path); |
2199 | err = PTR_ERR(mp); | |
2200 | if (IS_ERR(mp)) | |
b12cea91 AV |
2201 | goto out; |
2202 | ||
87129cc0 | 2203 | old = real_mount(old_path.mnt); |
84d17192 | 2204 | parent = real_mount(path->mnt); |
87129cc0 | 2205 | |
1da177e4 | 2206 | err = -EINVAL; |
fc7be130 | 2207 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2208 | goto out2; |
9676f0c6 | 2209 | |
e149ed2b AV |
2210 | if (!check_mnt(parent)) |
2211 | goto out2; | |
2212 | ||
2213 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2214 | goto out2; |
1da177e4 | 2215 | |
5ff9d8a6 EB |
2216 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2217 | goto out2; | |
2218 | ||
ccd48bc7 | 2219 | if (recurse) |
4ce5d2b1 | 2220 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2221 | else |
87129cc0 | 2222 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2223 | |
be34d1a3 DH |
2224 | if (IS_ERR(mnt)) { |
2225 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2226 | goto out2; |
be34d1a3 | 2227 | } |
ccd48bc7 | 2228 | |
5ff9d8a6 EB |
2229 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2230 | ||
84d17192 | 2231 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2232 | if (err) { |
719ea2fb | 2233 | lock_mount_hash(); |
e819f152 | 2234 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2235 | unlock_mount_hash(); |
5b83d2c5 | 2236 | } |
b12cea91 | 2237 | out2: |
84d17192 | 2238 | unlock_mount(mp); |
ccd48bc7 | 2239 | out: |
2d92ab3c | 2240 | path_put(&old_path); |
1da177e4 LT |
2241 | return err; |
2242 | } | |
2243 | ||
43f5e655 DH |
2244 | /* |
2245 | * Don't allow locked mount flags to be cleared. | |
2246 | * | |
2247 | * No locks need to be held here while testing the various MNT_LOCK | |
2248 | * flags because those flags can never be cleared once they are set. | |
2249 | */ | |
2250 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2251 | { |
43f5e655 DH |
2252 | unsigned int fl = mnt->mnt.mnt_flags; |
2253 | ||
2254 | if ((fl & MNT_LOCK_READONLY) && | |
2255 | !(mnt_flags & MNT_READONLY)) | |
2256 | return false; | |
2257 | ||
2258 | if ((fl & MNT_LOCK_NODEV) && | |
2259 | !(mnt_flags & MNT_NODEV)) | |
2260 | return false; | |
2261 | ||
2262 | if ((fl & MNT_LOCK_NOSUID) && | |
2263 | !(mnt_flags & MNT_NOSUID)) | |
2264 | return false; | |
2265 | ||
2266 | if ((fl & MNT_LOCK_NOEXEC) && | |
2267 | !(mnt_flags & MNT_NOEXEC)) | |
2268 | return false; | |
2269 | ||
2270 | if ((fl & MNT_LOCK_ATIME) && | |
2271 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2272 | return false; | |
2e4b7fcd | 2273 | |
43f5e655 DH |
2274 | return true; |
2275 | } | |
2276 | ||
2277 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2278 | { |
43f5e655 | 2279 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2280 | |
43f5e655 | 2281 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2282 | return 0; |
2283 | ||
2284 | if (readonly_request) | |
43f5e655 DH |
2285 | return mnt_make_readonly(mnt); |
2286 | ||
2287 | return __mnt_unmake_readonly(mnt); | |
2288 | } | |
2289 | ||
2290 | /* | |
2291 | * Update the user-settable attributes on a mount. The caller must hold | |
2292 | * sb->s_umount for writing. | |
2293 | */ | |
2294 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) | |
2295 | { | |
2296 | lock_mount_hash(); | |
2297 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; | |
2298 | mnt->mnt.mnt_flags = mnt_flags; | |
2299 | touch_mnt_namespace(mnt->mnt_ns); | |
2300 | unlock_mount_hash(); | |
2301 | } | |
2302 | ||
2303 | /* | |
2304 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2305 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2306 | * to mount(2). | |
2307 | */ | |
2308 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2309 | { | |
2310 | struct super_block *sb = path->mnt->mnt_sb; | |
2311 | struct mount *mnt = real_mount(path->mnt); | |
2312 | int ret; | |
2313 | ||
2314 | if (!check_mnt(mnt)) | |
2315 | return -EINVAL; | |
2316 | ||
2317 | if (path->dentry != mnt->mnt.mnt_root) | |
2318 | return -EINVAL; | |
2319 | ||
2320 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2321 | return -EPERM; | |
2322 | ||
2323 | down_write(&sb->s_umount); | |
2324 | ret = change_mount_ro_state(mnt, mnt_flags); | |
2325 | if (ret == 0) | |
2326 | set_mount_attributes(mnt, mnt_flags); | |
2327 | up_write(&sb->s_umount); | |
2328 | return ret; | |
2e4b7fcd DH |
2329 | } |
2330 | ||
1da177e4 LT |
2331 | /* |
2332 | * change filesystem flags. dir should be a physical root of filesystem. | |
2333 | * If you've mounted a non-root directory somewhere and want to do remount | |
2334 | * on it - tough luck. | |
2335 | */ | |
e462ec50 DH |
2336 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2337 | int mnt_flags, void *data) | |
1da177e4 LT |
2338 | { |
2339 | int err; | |
2d92ab3c | 2340 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2341 | struct mount *mnt = real_mount(path->mnt); |
204cc0cc | 2342 | void *sec_opts = NULL; |
1da177e4 | 2343 | |
143c8c91 | 2344 | if (!check_mnt(mnt)) |
1da177e4 LT |
2345 | return -EINVAL; |
2346 | ||
2d92ab3c | 2347 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2348 | return -EINVAL; |
2349 | ||
43f5e655 | 2350 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2351 | return -EPERM; |
9566d674 | 2352 | |
c039bc3c | 2353 | if (data && !(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)) { |
204cc0cc | 2354 | err = security_sb_eat_lsm_opts(data, &sec_opts); |
c039bc3c AV |
2355 | if (err) |
2356 | return err; | |
2357 | } | |
204cc0cc AV |
2358 | err = security_sb_remount(sb, sec_opts); |
2359 | security_free_mnt_opts(&sec_opts); | |
ff36fe2c EP |
2360 | if (err) |
2361 | return err; | |
2362 | ||
1da177e4 | 2363 | down_write(&sb->s_umount); |
43f5e655 DH |
2364 | err = -EPERM; |
2365 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
e462ec50 | 2366 | err = do_remount_sb(sb, sb_flags, data, 0); |
43f5e655 DH |
2367 | if (!err) |
2368 | set_mount_attributes(mnt, mnt_flags); | |
0e55a7cc | 2369 | } |
6339dab8 | 2370 | up_write(&sb->s_umount); |
1da177e4 LT |
2371 | return err; |
2372 | } | |
2373 | ||
cbbe362c | 2374 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2375 | { |
315fc83e | 2376 | struct mount *p; |
909b0a88 | 2377 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2378 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2379 | return 1; |
2380 | } | |
2381 | return 0; | |
2382 | } | |
2383 | ||
808d4e3c | 2384 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2385 | { |
2d92ab3c | 2386 | struct path old_path, parent_path; |
676da58d | 2387 | struct mount *p; |
0fb54e50 | 2388 | struct mount *old; |
84d17192 | 2389 | struct mountpoint *mp; |
57eccb83 | 2390 | int err; |
1da177e4 LT |
2391 | if (!old_name || !*old_name) |
2392 | return -EINVAL; | |
2d92ab3c | 2393 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2394 | if (err) |
2395 | return err; | |
2396 | ||
84d17192 AV |
2397 | mp = lock_mount(path); |
2398 | err = PTR_ERR(mp); | |
2399 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2400 | goto out; |
2401 | ||
143c8c91 | 2402 | old = real_mount(old_path.mnt); |
fc7be130 | 2403 | p = real_mount(path->mnt); |
143c8c91 | 2404 | |
1da177e4 | 2405 | err = -EINVAL; |
fc7be130 | 2406 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2407 | goto out1; |
2408 | ||
5ff9d8a6 EB |
2409 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2410 | goto out1; | |
2411 | ||
1da177e4 | 2412 | err = -EINVAL; |
2d92ab3c | 2413 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2414 | goto out1; |
1da177e4 | 2415 | |
676da58d | 2416 | if (!mnt_has_parent(old)) |
21444403 | 2417 | goto out1; |
1da177e4 | 2418 | |
e36cb0b8 DH |
2419 | if (d_is_dir(path->dentry) != |
2420 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2421 | goto out1; |
2422 | /* | |
2423 | * Don't move a mount residing in a shared parent. | |
2424 | */ | |
fc7be130 | 2425 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2426 | goto out1; |
9676f0c6 RP |
2427 | /* |
2428 | * Don't move a mount tree containing unbindable mounts to a destination | |
2429 | * mount which is shared. | |
2430 | */ | |
fc7be130 | 2431 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2432 | goto out1; |
1da177e4 | 2433 | err = -ELOOP; |
fc7be130 | 2434 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2435 | if (p == old) |
21444403 | 2436 | goto out1; |
1da177e4 | 2437 | |
84d17192 | 2438 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2439 | if (err) |
21444403 | 2440 | goto out1; |
1da177e4 LT |
2441 | |
2442 | /* if the mount is moved, it should no longer be expire | |
2443 | * automatically */ | |
6776db3d | 2444 | list_del_init(&old->mnt_expire); |
1da177e4 | 2445 | out1: |
84d17192 | 2446 | unlock_mount(mp); |
1da177e4 | 2447 | out: |
1da177e4 | 2448 | if (!err) |
1a390689 | 2449 | path_put(&parent_path); |
2d92ab3c | 2450 | path_put(&old_path); |
1da177e4 LT |
2451 | return err; |
2452 | } | |
2453 | ||
9d412a43 AV |
2454 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2455 | { | |
2456 | int err; | |
2457 | const char *subtype = strchr(fstype, '.'); | |
2458 | if (subtype) { | |
2459 | subtype++; | |
2460 | err = -EINVAL; | |
2461 | if (!subtype[0]) | |
2462 | goto err; | |
2463 | } else | |
2464 | subtype = ""; | |
2465 | ||
2466 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2467 | err = -ENOMEM; | |
2468 | if (!mnt->mnt_sb->s_subtype) | |
2469 | goto err; | |
2470 | return mnt; | |
2471 | ||
2472 | err: | |
2473 | mntput(mnt); | |
2474 | return ERR_PTR(err); | |
2475 | } | |
2476 | ||
9d412a43 AV |
2477 | /* |
2478 | * add a mount into a namespace's mount tree | |
2479 | */ | |
95bc5f25 | 2480 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2481 | { |
84d17192 AV |
2482 | struct mountpoint *mp; |
2483 | struct mount *parent; | |
9d412a43 AV |
2484 | int err; |
2485 | ||
f2ebb3a9 | 2486 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2487 | |
84d17192 AV |
2488 | mp = lock_mount(path); |
2489 | if (IS_ERR(mp)) | |
2490 | return PTR_ERR(mp); | |
9d412a43 | 2491 | |
84d17192 | 2492 | parent = real_mount(path->mnt); |
9d412a43 | 2493 | err = -EINVAL; |
84d17192 | 2494 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2495 | /* that's acceptable only for automounts done in private ns */ |
2496 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2497 | goto unlock; | |
2498 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2499 | if (!parent->mnt_ns) |
156cacb1 AV |
2500 | goto unlock; |
2501 | } | |
9d412a43 AV |
2502 | |
2503 | /* Refuse the same filesystem on the same mount point */ | |
2504 | err = -EBUSY; | |
95bc5f25 | 2505 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2506 | path->mnt->mnt_root == path->dentry) |
2507 | goto unlock; | |
2508 | ||
2509 | err = -EINVAL; | |
e36cb0b8 | 2510 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2511 | goto unlock; |
2512 | ||
95bc5f25 | 2513 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2514 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2515 | |
2516 | unlock: | |
84d17192 | 2517 | unlock_mount(mp); |
9d412a43 AV |
2518 | return err; |
2519 | } | |
b1e75df4 | 2520 | |
8654df4e | 2521 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2522 | |
1da177e4 LT |
2523 | /* |
2524 | * create a new mount for userspace and request it to be added into the | |
2525 | * namespace's tree | |
2526 | */ | |
e462ec50 | 2527 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2528 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2529 | { |
0c55cfc4 | 2530 | struct file_system_type *type; |
1da177e4 | 2531 | struct vfsmount *mnt; |
15f9a3f3 | 2532 | int err; |
1da177e4 | 2533 | |
0c55cfc4 | 2534 | if (!fstype) |
1da177e4 LT |
2535 | return -EINVAL; |
2536 | ||
0c55cfc4 EB |
2537 | type = get_fs_type(fstype); |
2538 | if (!type) | |
2539 | return -ENODEV; | |
2540 | ||
e462ec50 | 2541 | mnt = vfs_kern_mount(type, sb_flags, name, data); |
0c55cfc4 EB |
2542 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
2543 | !mnt->mnt_sb->s_subtype) | |
2544 | mnt = fs_set_subtype(mnt, fstype); | |
2545 | ||
2546 | put_filesystem(type); | |
1da177e4 LT |
2547 | if (IS_ERR(mnt)) |
2548 | return PTR_ERR(mnt); | |
2549 | ||
8654df4e EB |
2550 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2551 | mntput(mnt); | |
2552 | return -EPERM; | |
2553 | } | |
2554 | ||
95bc5f25 | 2555 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2556 | if (err) |
2557 | mntput(mnt); | |
2558 | return err; | |
1da177e4 LT |
2559 | } |
2560 | ||
19a167af AV |
2561 | int finish_automount(struct vfsmount *m, struct path *path) |
2562 | { | |
6776db3d | 2563 | struct mount *mnt = real_mount(m); |
19a167af AV |
2564 | int err; |
2565 | /* The new mount record should have at least 2 refs to prevent it being | |
2566 | * expired before we get a chance to add it | |
2567 | */ | |
6776db3d | 2568 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2569 | |
2570 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2571 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2572 | err = -ELOOP; |
2573 | goto fail; | |
19a167af AV |
2574 | } |
2575 | ||
95bc5f25 | 2576 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2577 | if (!err) |
2578 | return 0; | |
2579 | fail: | |
2580 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2581 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2582 | namespace_lock(); |
6776db3d | 2583 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2584 | namespace_unlock(); |
19a167af | 2585 | } |
b1e75df4 AV |
2586 | mntput(m); |
2587 | mntput(m); | |
19a167af AV |
2588 | return err; |
2589 | } | |
2590 | ||
ea5b778a DH |
2591 | /** |
2592 | * mnt_set_expiry - Put a mount on an expiration list | |
2593 | * @mnt: The mount to list. | |
2594 | * @expiry_list: The list to add the mount to. | |
2595 | */ | |
2596 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2597 | { | |
97216be0 | 2598 | namespace_lock(); |
ea5b778a | 2599 | |
6776db3d | 2600 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2601 | |
97216be0 | 2602 | namespace_unlock(); |
ea5b778a DH |
2603 | } |
2604 | EXPORT_SYMBOL(mnt_set_expiry); | |
2605 | ||
1da177e4 LT |
2606 | /* |
2607 | * process a list of expirable mountpoints with the intent of discarding any | |
2608 | * mountpoints that aren't in use and haven't been touched since last we came | |
2609 | * here | |
2610 | */ | |
2611 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2612 | { | |
761d5c38 | 2613 | struct mount *mnt, *next; |
1da177e4 LT |
2614 | LIST_HEAD(graveyard); |
2615 | ||
2616 | if (list_empty(mounts)) | |
2617 | return; | |
2618 | ||
97216be0 | 2619 | namespace_lock(); |
719ea2fb | 2620 | lock_mount_hash(); |
1da177e4 LT |
2621 | |
2622 | /* extract from the expiration list every vfsmount that matches the | |
2623 | * following criteria: | |
2624 | * - only referenced by its parent vfsmount | |
2625 | * - still marked for expiry (marked on the last call here; marks are | |
2626 | * cleared by mntput()) | |
2627 | */ | |
6776db3d | 2628 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2629 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2630 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2631 | continue; |
6776db3d | 2632 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2633 | } |
bcc5c7d2 | 2634 | while (!list_empty(&graveyard)) { |
6776db3d | 2635 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2636 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2637 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2638 | } |
719ea2fb | 2639 | unlock_mount_hash(); |
3ab6abee | 2640 | namespace_unlock(); |
5528f911 TM |
2641 | } |
2642 | ||
2643 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2644 | ||
2645 | /* | |
2646 | * Ripoff of 'select_parent()' | |
2647 | * | |
2648 | * search the list of submounts for a given mountpoint, and move any | |
2649 | * shrinkable submounts to the 'graveyard' list. | |
2650 | */ | |
692afc31 | 2651 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2652 | { |
692afc31 | 2653 | struct mount *this_parent = parent; |
5528f911 TM |
2654 | struct list_head *next; |
2655 | int found = 0; | |
2656 | ||
2657 | repeat: | |
6b41d536 | 2658 | next = this_parent->mnt_mounts.next; |
5528f911 | 2659 | resume: |
6b41d536 | 2660 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2661 | struct list_head *tmp = next; |
6b41d536 | 2662 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2663 | |
2664 | next = tmp->next; | |
692afc31 | 2665 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2666 | continue; |
5528f911 TM |
2667 | /* |
2668 | * Descend a level if the d_mounts list is non-empty. | |
2669 | */ | |
6b41d536 | 2670 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2671 | this_parent = mnt; |
2672 | goto repeat; | |
2673 | } | |
1da177e4 | 2674 | |
1ab59738 | 2675 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2676 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2677 | found++; |
2678 | } | |
1da177e4 | 2679 | } |
5528f911 TM |
2680 | /* |
2681 | * All done at this level ... ascend and resume the search | |
2682 | */ | |
2683 | if (this_parent != parent) { | |
6b41d536 | 2684 | next = this_parent->mnt_child.next; |
0714a533 | 2685 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2686 | goto resume; |
2687 | } | |
2688 | return found; | |
2689 | } | |
2690 | ||
2691 | /* | |
2692 | * process a list of expirable mountpoints with the intent of discarding any | |
2693 | * submounts of a specific parent mountpoint | |
99b7db7b | 2694 | * |
48a066e7 | 2695 | * mount_lock must be held for write |
5528f911 | 2696 | */ |
b54b9be7 | 2697 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2698 | { |
2699 | LIST_HEAD(graveyard); | |
761d5c38 | 2700 | struct mount *m; |
5528f911 | 2701 | |
5528f911 | 2702 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2703 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2704 | while (!list_empty(&graveyard)) { |
761d5c38 | 2705 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2706 | mnt_expire); |
143c8c91 | 2707 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2708 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2709 | } |
2710 | } | |
1da177e4 LT |
2711 | } |
2712 | ||
1da177e4 LT |
2713 | /* |
2714 | * Some copy_from_user() implementations do not return the exact number of | |
2715 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2716 | * Note that this function differs from copy_from_user() in that it will oops | |
2717 | * on bad values of `to', rather than returning a short copy. | |
2718 | */ | |
b58fed8b RP |
2719 | static long exact_copy_from_user(void *to, const void __user * from, |
2720 | unsigned long n) | |
1da177e4 LT |
2721 | { |
2722 | char *t = to; | |
2723 | const char __user *f = from; | |
2724 | char c; | |
2725 | ||
96d4f267 | 2726 | if (!access_ok(from, n)) |
1da177e4 LT |
2727 | return n; |
2728 | ||
9da3f2b7 | 2729 | current->kernel_uaccess_faults_ok++; |
1da177e4 LT |
2730 | while (n) { |
2731 | if (__get_user(c, f)) { | |
2732 | memset(t, 0, n); | |
2733 | break; | |
2734 | } | |
2735 | *t++ = c; | |
2736 | f++; | |
2737 | n--; | |
2738 | } | |
9da3f2b7 | 2739 | current->kernel_uaccess_faults_ok--; |
1da177e4 LT |
2740 | return n; |
2741 | } | |
2742 | ||
b40ef869 | 2743 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2744 | { |
2745 | int i; | |
1da177e4 | 2746 | unsigned long size; |
b40ef869 | 2747 | char *copy; |
b58fed8b | 2748 | |
1da177e4 | 2749 | if (!data) |
b40ef869 | 2750 | return NULL; |
1da177e4 | 2751 | |
b40ef869 AV |
2752 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2753 | if (!copy) | |
2754 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2755 | |
2756 | /* We only care that *some* data at the address the user | |
2757 | * gave us is valid. Just in case, we'll zero | |
2758 | * the remainder of the page. | |
2759 | */ | |
2760 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2761 | size = TASK_SIZE - (unsigned long)data; | |
2762 | if (size > PAGE_SIZE) | |
2763 | size = PAGE_SIZE; | |
2764 | ||
b40ef869 | 2765 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2766 | if (!i) { |
b40ef869 AV |
2767 | kfree(copy); |
2768 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2769 | } |
2770 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2771 | memset(copy + i, 0, PAGE_SIZE - i); |
2772 | return copy; | |
1da177e4 LT |
2773 | } |
2774 | ||
b8850d1f | 2775 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2776 | { |
b8850d1f | 2777 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2778 | } |
2779 | ||
1da177e4 LT |
2780 | /* |
2781 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2782 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2783 | * | |
2784 | * data is a (void *) that can point to any structure up to | |
2785 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2786 | * information (or be NULL). | |
2787 | * | |
2788 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2789 | * When the flags word was introduced its top half was required | |
2790 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2791 | * Therefore, if this magic number is present, it carries no information | |
2792 | * and must be discarded. | |
2793 | */ | |
5e6123f3 | 2794 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2795 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2796 | { |
2d92ab3c | 2797 | struct path path; |
e462ec50 | 2798 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2799 | int retval = 0; |
1da177e4 LT |
2800 | |
2801 | /* Discard magic */ | |
2802 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2803 | flags &= ~MS_MGC_MSK; | |
2804 | ||
2805 | /* Basic sanity checks */ | |
1da177e4 LT |
2806 | if (data_page) |
2807 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2808 | ||
e462ec50 DH |
2809 | if (flags & MS_NOUSER) |
2810 | return -EINVAL; | |
2811 | ||
a27ab9f2 | 2812 | /* ... and get the mountpoint */ |
5e6123f3 | 2813 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2814 | if (retval) |
2815 | return retval; | |
2816 | ||
2817 | retval = security_sb_mount(dev_name, &path, | |
2818 | type_page, flags, data_page); | |
0d5cadb8 AV |
2819 | if (!retval && !may_mount()) |
2820 | retval = -EPERM; | |
e462ec50 | 2821 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 2822 | retval = -EPERM; |
a27ab9f2 TH |
2823 | if (retval) |
2824 | goto dput_out; | |
2825 | ||
613cbe3d AK |
2826 | /* Default to relatime unless overriden */ |
2827 | if (!(flags & MS_NOATIME)) | |
2828 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2829 | |
1da177e4 LT |
2830 | /* Separate the per-mountpoint flags */ |
2831 | if (flags & MS_NOSUID) | |
2832 | mnt_flags |= MNT_NOSUID; | |
2833 | if (flags & MS_NODEV) | |
2834 | mnt_flags |= MNT_NODEV; | |
2835 | if (flags & MS_NOEXEC) | |
2836 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2837 | if (flags & MS_NOATIME) |
2838 | mnt_flags |= MNT_NOATIME; | |
2839 | if (flags & MS_NODIRATIME) | |
2840 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2841 | if (flags & MS_STRICTATIME) |
2842 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 2843 | if (flags & MS_RDONLY) |
2e4b7fcd | 2844 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 2845 | |
ffbc6f0e EB |
2846 | /* The default atime for remount is preservation */ |
2847 | if ((flags & MS_REMOUNT) && | |
2848 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2849 | MS_STRICTATIME)) == 0)) { | |
2850 | mnt_flags &= ~MNT_ATIME_MASK; | |
2851 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2852 | } | |
2853 | ||
e462ec50 DH |
2854 | sb_flags = flags & (SB_RDONLY | |
2855 | SB_SYNCHRONOUS | | |
2856 | SB_MANDLOCK | | |
2857 | SB_DIRSYNC | | |
2858 | SB_SILENT | | |
917086ff | 2859 | SB_POSIXACL | |
d7ee9469 | 2860 | SB_LAZYTIME | |
917086ff | 2861 | SB_I_VERSION); |
1da177e4 | 2862 | |
43f5e655 DH |
2863 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
2864 | retval = do_reconfigure_mnt(&path, mnt_flags); | |
2865 | else if (flags & MS_REMOUNT) | |
e462ec50 | 2866 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
2867 | data_page); |
2868 | else if (flags & MS_BIND) | |
2d92ab3c | 2869 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2870 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2871 | retval = do_change_type(&path, flags); |
1da177e4 | 2872 | else if (flags & MS_MOVE) |
2d92ab3c | 2873 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2874 | else |
e462ec50 | 2875 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
2876 | dev_name, data_page); |
2877 | dput_out: | |
2d92ab3c | 2878 | path_put(&path); |
1da177e4 LT |
2879 | return retval; |
2880 | } | |
2881 | ||
537f7ccb EB |
2882 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2883 | { | |
2884 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2885 | } | |
2886 | ||
2887 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2888 | { | |
2889 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2890 | } | |
2891 | ||
771b1371 EB |
2892 | static void free_mnt_ns(struct mnt_namespace *ns) |
2893 | { | |
74e83122 AV |
2894 | if (!is_anon_ns(ns)) |
2895 | ns_free_inum(&ns->ns); | |
537f7ccb | 2896 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2897 | put_user_ns(ns->user_ns); |
2898 | kfree(ns); | |
2899 | } | |
2900 | ||
8823c079 EB |
2901 | /* |
2902 | * Assign a sequence number so we can detect when we attempt to bind | |
2903 | * mount a reference to an older mount namespace into the current | |
2904 | * mount namespace, preventing reference counting loops. A 64bit | |
2905 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2906 | * is effectively never, so we can ignore the possibility. | |
2907 | */ | |
2908 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2909 | ||
74e83122 | 2910 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
2911 | { |
2912 | struct mnt_namespace *new_ns; | |
537f7ccb | 2913 | struct ucounts *ucounts; |
98f842e6 | 2914 | int ret; |
cf8d2c11 | 2915 | |
537f7ccb EB |
2916 | ucounts = inc_mnt_namespaces(user_ns); |
2917 | if (!ucounts) | |
df75e774 | 2918 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2919 | |
74e83122 | 2920 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2921 | if (!new_ns) { |
2922 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2923 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2924 | } |
74e83122 AV |
2925 | if (!anon) { |
2926 | ret = ns_alloc_inum(&new_ns->ns); | |
2927 | if (ret) { | |
2928 | kfree(new_ns); | |
2929 | dec_mnt_namespaces(ucounts); | |
2930 | return ERR_PTR(ret); | |
2931 | } | |
98f842e6 | 2932 | } |
33c42940 | 2933 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
2934 | if (!anon) |
2935 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
cf8d2c11 | 2936 | atomic_set(&new_ns->count, 1); |
cf8d2c11 TM |
2937 | INIT_LIST_HEAD(&new_ns->list); |
2938 | init_waitqueue_head(&new_ns->poll); | |
771b1371 | 2939 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2940 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
2941 | return new_ns; |
2942 | } | |
2943 | ||
0766f788 | 2944 | __latent_entropy |
9559f689 AV |
2945 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2946 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2947 | { |
6b3286ed | 2948 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2949 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2950 | struct mount *p, *q; |
9559f689 | 2951 | struct mount *old; |
cb338d06 | 2952 | struct mount *new; |
7a472ef4 | 2953 | int copy_flags; |
1da177e4 | 2954 | |
9559f689 AV |
2955 | BUG_ON(!ns); |
2956 | ||
2957 | if (likely(!(flags & CLONE_NEWNS))) { | |
2958 | get_mnt_ns(ns); | |
2959 | return ns; | |
2960 | } | |
2961 | ||
2962 | old = ns->root; | |
2963 | ||
74e83122 | 2964 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
2965 | if (IS_ERR(new_ns)) |
2966 | return new_ns; | |
1da177e4 | 2967 | |
97216be0 | 2968 | namespace_lock(); |
1da177e4 | 2969 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2970 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2971 | if (user_ns != ns->user_ns) |
3bd045cc | 2972 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 2973 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2974 | if (IS_ERR(new)) { |
328e6d90 | 2975 | namespace_unlock(); |
771b1371 | 2976 | free_mnt_ns(new_ns); |
be34d1a3 | 2977 | return ERR_CAST(new); |
1da177e4 | 2978 | } |
3bd045cc AV |
2979 | if (user_ns != ns->user_ns) { |
2980 | lock_mount_hash(); | |
2981 | lock_mnt_tree(new); | |
2982 | unlock_mount_hash(); | |
2983 | } | |
be08d6d2 | 2984 | new_ns->root = new; |
1a4eeaf2 | 2985 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2986 | |
2987 | /* | |
2988 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2989 | * as belonging to new namespace. We have already acquired a private | |
2990 | * fs_struct, so tsk->fs->lock is not needed. | |
2991 | */ | |
909b0a88 | 2992 | p = old; |
cb338d06 | 2993 | q = new; |
1da177e4 | 2994 | while (p) { |
143c8c91 | 2995 | q->mnt_ns = new_ns; |
d2921684 | 2996 | new_ns->mounts++; |
9559f689 AV |
2997 | if (new_fs) { |
2998 | if (&p->mnt == new_fs->root.mnt) { | |
2999 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3000 | rootmnt = &p->mnt; |
1da177e4 | 3001 | } |
9559f689 AV |
3002 | if (&p->mnt == new_fs->pwd.mnt) { |
3003 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3004 | pwdmnt = &p->mnt; |
1da177e4 | 3005 | } |
1da177e4 | 3006 | } |
909b0a88 AV |
3007 | p = next_mnt(p, old); |
3008 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3009 | if (!q) |
3010 | break; | |
3011 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3012 | p = next_mnt(p, old); | |
1da177e4 | 3013 | } |
328e6d90 | 3014 | namespace_unlock(); |
1da177e4 | 3015 | |
1da177e4 | 3016 | if (rootmnt) |
f03c6599 | 3017 | mntput(rootmnt); |
1da177e4 | 3018 | if (pwdmnt) |
f03c6599 | 3019 | mntput(pwdmnt); |
1da177e4 | 3020 | |
741a2951 | 3021 | return new_ns; |
1da177e4 LT |
3022 | } |
3023 | ||
74e83122 | 3024 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3025 | { |
74e83122 | 3026 | struct mount *mnt = real_mount(m); |
ea441d11 | 3027 | struct mnt_namespace *ns; |
d31da0f0 | 3028 | struct super_block *s; |
ea441d11 AV |
3029 | struct path path; |
3030 | int err; | |
3031 | ||
74e83122 AV |
3032 | ns = alloc_mnt_ns(&init_user_ns, true); |
3033 | if (IS_ERR(ns)) { | |
3034 | mntput(m); | |
ea441d11 | 3035 | return ERR_CAST(ns); |
74e83122 AV |
3036 | } |
3037 | mnt->mnt_ns = ns; | |
3038 | ns->root = mnt; | |
3039 | ns->mounts++; | |
3040 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3041 | |
74e83122 | 3042 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3043 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3044 | ||
3045 | put_mnt_ns(ns); | |
3046 | ||
3047 | if (err) | |
3048 | return ERR_PTR(err); | |
3049 | ||
3050 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3051 | s = path.mnt->mnt_sb; |
3052 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3053 | mntput(path.mnt); |
3054 | /* lock the sucker */ | |
d31da0f0 | 3055 | down_write(&s->s_umount); |
ea441d11 AV |
3056 | /* ... and return the root of (sub)tree on it */ |
3057 | return path.dentry; | |
3058 | } | |
3059 | EXPORT_SYMBOL(mount_subtree); | |
3060 | ||
312db1aa DB |
3061 | int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, |
3062 | unsigned long flags, void __user *data) | |
1da177e4 | 3063 | { |
eca6f534 VN |
3064 | int ret; |
3065 | char *kernel_type; | |
eca6f534 | 3066 | char *kernel_dev; |
b40ef869 | 3067 | void *options; |
1da177e4 | 3068 | |
b8850d1f TG |
3069 | kernel_type = copy_mount_string(type); |
3070 | ret = PTR_ERR(kernel_type); | |
3071 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3072 | goto out_type; |
1da177e4 | 3073 | |
b8850d1f TG |
3074 | kernel_dev = copy_mount_string(dev_name); |
3075 | ret = PTR_ERR(kernel_dev); | |
3076 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3077 | goto out_dev; |
1da177e4 | 3078 | |
b40ef869 AV |
3079 | options = copy_mount_options(data); |
3080 | ret = PTR_ERR(options); | |
3081 | if (IS_ERR(options)) | |
eca6f534 | 3082 | goto out_data; |
1da177e4 | 3083 | |
b40ef869 | 3084 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3085 | |
b40ef869 | 3086 | kfree(options); |
eca6f534 VN |
3087 | out_data: |
3088 | kfree(kernel_dev); | |
3089 | out_dev: | |
eca6f534 VN |
3090 | kfree(kernel_type); |
3091 | out_type: | |
3092 | return ret; | |
1da177e4 LT |
3093 | } |
3094 | ||
312db1aa DB |
3095 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3096 | char __user *, type, unsigned long, flags, void __user *, data) | |
3097 | { | |
3098 | return ksys_mount(dev_name, dir_name, type, flags, data); | |
3099 | } | |
3100 | ||
afac7cba AV |
3101 | /* |
3102 | * Return true if path is reachable from root | |
3103 | * | |
48a066e7 | 3104 | * namespace_sem or mount_lock is held |
afac7cba | 3105 | */ |
643822b4 | 3106 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3107 | const struct path *root) |
3108 | { | |
643822b4 | 3109 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3110 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3111 | mnt = mnt->mnt_parent; |
afac7cba | 3112 | } |
643822b4 | 3113 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3114 | } |
3115 | ||
640eb7e7 | 3116 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3117 | { |
25ab4c9b | 3118 | bool res; |
48a066e7 | 3119 | read_seqlock_excl(&mount_lock); |
643822b4 | 3120 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3121 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3122 | return res; |
3123 | } | |
3124 | EXPORT_SYMBOL(path_is_under); | |
3125 | ||
1da177e4 LT |
3126 | /* |
3127 | * pivot_root Semantics: | |
3128 | * Moves the root file system of the current process to the directory put_old, | |
3129 | * makes new_root as the new root file system of the current process, and sets | |
3130 | * root/cwd of all processes which had them on the current root to new_root. | |
3131 | * | |
3132 | * Restrictions: | |
3133 | * The new_root and put_old must be directories, and must not be on the | |
3134 | * same file system as the current process root. The put_old must be | |
3135 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3136 | * pointed to by put_old must yield the same directory as new_root. No other | |
3137 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3138 | * | |
4a0d11fa NB |
3139 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3140 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3141 | * in this situation. | |
3142 | * | |
1da177e4 LT |
3143 | * Notes: |
3144 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3145 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3146 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3147 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3148 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3149 | * first. | |
3150 | */ | |
3480b257 HC |
3151 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3152 | const char __user *, put_old) | |
1da177e4 | 3153 | { |
2d8f3038 | 3154 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3155 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3156 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3157 | int error; |
3158 | ||
9b40bc90 | 3159 | if (!may_mount()) |
1da177e4 LT |
3160 | return -EPERM; |
3161 | ||
2d8f3038 | 3162 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3163 | if (error) |
3164 | goto out0; | |
1da177e4 | 3165 | |
2d8f3038 | 3166 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3167 | if (error) |
3168 | goto out1; | |
3169 | ||
2d8f3038 | 3170 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3171 | if (error) |
3172 | goto out2; | |
1da177e4 | 3173 | |
f7ad3c6b | 3174 | get_fs_root(current->fs, &root); |
84d17192 AV |
3175 | old_mp = lock_mount(&old); |
3176 | error = PTR_ERR(old_mp); | |
3177 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3178 | goto out3; |
3179 | ||
1da177e4 | 3180 | error = -EINVAL; |
419148da AV |
3181 | new_mnt = real_mount(new.mnt); |
3182 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3183 | old_mnt = real_mount(old.mnt); |
3184 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3185 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3186 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3187 | goto out4; |
143c8c91 | 3188 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3189 | goto out4; |
5ff9d8a6 EB |
3190 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3191 | goto out4; | |
1da177e4 | 3192 | error = -ENOENT; |
f3da392e | 3193 | if (d_unlinked(new.dentry)) |
b12cea91 | 3194 | goto out4; |
1da177e4 | 3195 | error = -EBUSY; |
84d17192 | 3196 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3197 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3198 | error = -EINVAL; |
8c3ee42e | 3199 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3200 | goto out4; /* not a mountpoint */ |
676da58d | 3201 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3202 | goto out4; /* not attached */ |
84d17192 | 3203 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3204 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3205 | goto out4; /* not a mountpoint */ |
676da58d | 3206 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3207 | goto out4; /* not attached */ |
4ac91378 | 3208 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3209 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3210 | goto out4; |
0d082601 EB |
3211 | /* make certain new is below the root */ |
3212 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3213 | goto out4; | |
84d17192 | 3214 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3215 | lock_mount_hash(); |
419148da AV |
3216 | detach_mnt(new_mnt, &parent_path); |
3217 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3218 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3219 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3220 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3221 | } | |
4ac91378 | 3222 | /* mount old root on put_old */ |
84d17192 | 3223 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3224 | /* mount new_root on / */ |
84d17192 | 3225 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3226 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3227 | /* A moved mount should not expire automatically */ |
3228 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3229 | put_mountpoint(root_mp); |
719ea2fb | 3230 | unlock_mount_hash(); |
2d8f3038 | 3231 | chroot_fs_refs(&root, &new); |
1da177e4 | 3232 | error = 0; |
b12cea91 | 3233 | out4: |
84d17192 | 3234 | unlock_mount(old_mp); |
b12cea91 AV |
3235 | if (!error) { |
3236 | path_put(&root_parent); | |
3237 | path_put(&parent_path); | |
3238 | } | |
3239 | out3: | |
8c3ee42e | 3240 | path_put(&root); |
b12cea91 | 3241 | out2: |
2d8f3038 | 3242 | path_put(&old); |
1da177e4 | 3243 | out1: |
2d8f3038 | 3244 | path_put(&new); |
1da177e4 | 3245 | out0: |
1da177e4 | 3246 | return error; |
1da177e4 LT |
3247 | } |
3248 | ||
3249 | static void __init init_mount_tree(void) | |
3250 | { | |
3251 | struct vfsmount *mnt; | |
74e83122 | 3252 | struct mount *m; |
6b3286ed | 3253 | struct mnt_namespace *ns; |
ac748a09 | 3254 | struct path root; |
0c55cfc4 | 3255 | struct file_system_type *type; |
1da177e4 | 3256 | |
0c55cfc4 EB |
3257 | type = get_fs_type("rootfs"); |
3258 | if (!type) | |
3259 | panic("Can't find rootfs type"); | |
3260 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3261 | put_filesystem(type); | |
1da177e4 LT |
3262 | if (IS_ERR(mnt)) |
3263 | panic("Can't create rootfs"); | |
b3e19d92 | 3264 | |
74e83122 | 3265 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 3266 | if (IS_ERR(ns)) |
1da177e4 | 3267 | panic("Can't allocate initial namespace"); |
74e83122 AV |
3268 | m = real_mount(mnt); |
3269 | m->mnt_ns = ns; | |
3270 | ns->root = m; | |
3271 | ns->mounts = 1; | |
3272 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
3273 | init_task.nsproxy->mnt_ns = ns; |
3274 | get_mnt_ns(ns); | |
3275 | ||
be08d6d2 AV |
3276 | root.mnt = mnt; |
3277 | root.dentry = mnt->mnt_root; | |
da362b09 | 3278 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3279 | |
3280 | set_fs_pwd(current->fs, &root); | |
3281 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3282 | } |
3283 | ||
74bf17cf | 3284 | void __init mnt_init(void) |
1da177e4 | 3285 | { |
15a67dd8 | 3286 | int err; |
1da177e4 | 3287 | |
7d6fec45 | 3288 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3289 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3290 | |
0818bf27 | 3291 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3292 | sizeof(struct hlist_head), |
0818bf27 | 3293 | mhash_entries, 19, |
3d375d78 | 3294 | HASH_ZERO, |
0818bf27 AV |
3295 | &m_hash_shift, &m_hash_mask, 0, 0); |
3296 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3297 | sizeof(struct hlist_head), | |
3298 | mphash_entries, 19, | |
3d375d78 | 3299 | HASH_ZERO, |
0818bf27 | 3300 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3301 | |
84d17192 | 3302 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3303 | panic("Failed to allocate mount hash table\n"); |
3304 | ||
4b93dc9b TH |
3305 | kernfs_init(); |
3306 | ||
15a67dd8 RD |
3307 | err = sysfs_init(); |
3308 | if (err) | |
3309 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3310 | __func__, err); |
00d26666 GKH |
3311 | fs_kobj = kobject_create_and_add("fs", NULL); |
3312 | if (!fs_kobj) | |
8e24eea7 | 3313 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3314 | init_rootfs(); |
3315 | init_mount_tree(); | |
3316 | } | |
3317 | ||
616511d0 | 3318 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3319 | { |
d498b25a | 3320 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3321 | return; |
7b00ed6f | 3322 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3323 | free_mnt_ns(ns); |
1da177e4 | 3324 | } |
9d412a43 AV |
3325 | |
3326 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3327 | { | |
423e0ab0 | 3328 | struct vfsmount *mnt; |
e462ec50 | 3329 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); |
423e0ab0 TC |
3330 | if (!IS_ERR(mnt)) { |
3331 | /* | |
3332 | * it is a longterm mount, don't release mnt until | |
3333 | * we unmount before file sys is unregistered | |
3334 | */ | |
f7a99c5b | 3335 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3336 | } |
3337 | return mnt; | |
9d412a43 AV |
3338 | } |
3339 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3340 | |
3341 | void kern_unmount(struct vfsmount *mnt) | |
3342 | { | |
3343 | /* release long term mount so mount point can be released */ | |
3344 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3345 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3346 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3347 | mntput(mnt); |
3348 | } | |
3349 | } | |
3350 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3351 | |
3352 | bool our_mnt(struct vfsmount *mnt) | |
3353 | { | |
143c8c91 | 3354 | return check_mnt(real_mount(mnt)); |
02125a82 | 3355 | } |
8823c079 | 3356 | |
3151527e EB |
3357 | bool current_chrooted(void) |
3358 | { | |
3359 | /* Does the current process have a non-standard root */ | |
3360 | struct path ns_root; | |
3361 | struct path fs_root; | |
3362 | bool chrooted; | |
3363 | ||
3364 | /* Find the namespace root */ | |
3365 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3366 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3367 | path_get(&ns_root); | |
3368 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3369 | ; | |
3370 | ||
3371 | get_fs_root(current->fs, &fs_root); | |
3372 | ||
3373 | chrooted = !path_equal(&fs_root, &ns_root); | |
3374 | ||
3375 | path_put(&fs_root); | |
3376 | path_put(&ns_root); | |
3377 | ||
3378 | return chrooted; | |
3379 | } | |
3380 | ||
8654df4e EB |
3381 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3382 | int *new_mnt_flags) | |
87a8ebd6 | 3383 | { |
8c6cf9cc | 3384 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3385 | struct mount *mnt; |
e51db735 | 3386 | bool visible = false; |
87a8ebd6 | 3387 | |
44bb4385 | 3388 | down_read(&namespace_sem); |
87a8ebd6 | 3389 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3390 | struct mount *child; |
77b1a97d EB |
3391 | int mnt_flags; |
3392 | ||
8654df4e | 3393 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3394 | continue; |
3395 | ||
7e96c1b0 EB |
3396 | /* This mount is not fully visible if it's root directory |
3397 | * is not the root directory of the filesystem. | |
3398 | */ | |
3399 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3400 | continue; | |
3401 | ||
a1935c17 | 3402 | /* A local view of the mount flags */ |
77b1a97d | 3403 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3404 | |
695e9df0 | 3405 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3406 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3407 | mnt_flags |= MNT_LOCK_READONLY; |
3408 | ||
8c6cf9cc EB |
3409 | /* Verify the mount flags are equal to or more permissive |
3410 | * than the proposed new mount. | |
3411 | */ | |
77b1a97d | 3412 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3413 | !(new_flags & MNT_READONLY)) |
3414 | continue; | |
77b1a97d EB |
3415 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3416 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3417 | continue; |
3418 | ||
ceeb0e5d EB |
3419 | /* This mount is not fully visible if there are any |
3420 | * locked child mounts that cover anything except for | |
3421 | * empty directories. | |
e51db735 EB |
3422 | */ |
3423 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3424 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3425 | /* Only worry about locked mounts */ |
d71ed6c9 | 3426 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3427 | continue; |
7236c85e EB |
3428 | /* Is the directory permanetly empty? */ |
3429 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3430 | goto next; |
87a8ebd6 | 3431 | } |
8c6cf9cc | 3432 | /* Preserve the locked attributes */ |
77b1a97d | 3433 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3434 | MNT_LOCK_ATIME); |
e51db735 EB |
3435 | visible = true; |
3436 | goto found; | |
3437 | next: ; | |
87a8ebd6 | 3438 | } |
e51db735 | 3439 | found: |
44bb4385 | 3440 | up_read(&namespace_sem); |
e51db735 | 3441 | return visible; |
87a8ebd6 EB |
3442 | } |
3443 | ||
8654df4e EB |
3444 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3445 | { | |
a1935c17 | 3446 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3447 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3448 | unsigned long s_iflags; | |
3449 | ||
3450 | if (ns->user_ns == &init_user_ns) | |
3451 | return false; | |
3452 | ||
3453 | /* Can this filesystem be too revealing? */ | |
3454 | s_iflags = mnt->mnt_sb->s_iflags; | |
3455 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3456 | return false; | |
3457 | ||
a1935c17 EB |
3458 | if ((s_iflags & required_iflags) != required_iflags) { |
3459 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3460 | required_iflags); | |
3461 | return true; | |
3462 | } | |
3463 | ||
8654df4e EB |
3464 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3465 | } | |
3466 | ||
380cf5ba AL |
3467 | bool mnt_may_suid(struct vfsmount *mnt) |
3468 | { | |
3469 | /* | |
3470 | * Foreign mounts (accessed via fchdir or through /proc | |
3471 | * symlinks) are always treated as if they are nosuid. This | |
3472 | * prevents namespaces from trusting potentially unsafe | |
3473 | * suid/sgid bits, file caps, or security labels that originate | |
3474 | * in other namespaces. | |
3475 | */ | |
3476 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3477 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3478 | } | |
3479 | ||
64964528 | 3480 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3481 | { |
58be2825 | 3482 | struct ns_common *ns = NULL; |
8823c079 EB |
3483 | struct nsproxy *nsproxy; |
3484 | ||
728dba3a EB |
3485 | task_lock(task); |
3486 | nsproxy = task->nsproxy; | |
8823c079 | 3487 | if (nsproxy) { |
58be2825 AV |
3488 | ns = &nsproxy->mnt_ns->ns; |
3489 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3490 | } |
728dba3a | 3491 | task_unlock(task); |
8823c079 EB |
3492 | |
3493 | return ns; | |
3494 | } | |
3495 | ||
64964528 | 3496 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3497 | { |
58be2825 | 3498 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3499 | } |
3500 | ||
64964528 | 3501 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3502 | { |
3503 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3504 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3505 | struct path root; |
4f757f3c | 3506 | int err; |
8823c079 | 3507 | |
0c55cfc4 | 3508 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3509 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3510 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3511 | return -EPERM; |
8823c079 | 3512 | |
74e83122 AV |
3513 | if (is_anon_ns(mnt_ns)) |
3514 | return -EINVAL; | |
3515 | ||
8823c079 EB |
3516 | if (fs->users != 1) |
3517 | return -EINVAL; | |
3518 | ||
3519 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3520 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3521 | nsproxy->mnt_ns = mnt_ns; |
3522 | ||
3523 | /* Find the root */ | |
4f757f3c AV |
3524 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3525 | "/", LOOKUP_DOWN, &root); | |
3526 | if (err) { | |
3527 | /* revert to old namespace */ | |
3528 | nsproxy->mnt_ns = old_mnt_ns; | |
3529 | put_mnt_ns(mnt_ns); | |
3530 | return err; | |
3531 | } | |
8823c079 | 3532 | |
4068367c AV |
3533 | put_mnt_ns(old_mnt_ns); |
3534 | ||
8823c079 EB |
3535 | /* Update the pwd and root */ |
3536 | set_fs_pwd(fs, &root); | |
3537 | set_fs_root(fs, &root); | |
3538 | ||
3539 | path_put(&root); | |
3540 | return 0; | |
3541 | } | |
3542 | ||
bcac25a5 AV |
3543 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3544 | { | |
3545 | return to_mnt_ns(ns)->user_ns; | |
3546 | } | |
3547 | ||
8823c079 EB |
3548 | const struct proc_ns_operations mntns_operations = { |
3549 | .name = "mnt", | |
3550 | .type = CLONE_NEWNS, | |
3551 | .get = mntns_get, | |
3552 | .put = mntns_put, | |
3553 | .install = mntns_install, | |
bcac25a5 | 3554 | .owner = mntns_owner, |
8823c079 | 3555 | }; |