Commit | Line | Data |
---|---|---|
59bd9ded | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namespace.c | |
4 | * | |
5 | * (C) Copyright Al Viro 2000, 2001 | |
1da177e4 LT |
6 | * |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
a07b2000 | 23 | #include <linux/file.h> |
d10577a8 | 24 | #include <linux/uaccess.h> |
0bb80f24 | 25 | #include <linux/proc_ns.h> |
20b4fb48 | 26 | #include <linux/magic.h> |
57c8a661 | 27 | #include <linux/memblock.h> |
9ea459e1 | 28 | #include <linux/task_work.h> |
9164bb4a | 29 | #include <linux/sched/task.h> |
e262e32d | 30 | #include <uapi/linux/mount.h> |
9bc61ab1 | 31 | #include <linux/fs_context.h> |
9164bb4a | 32 | |
07b20889 | 33 | #include "pnode.h" |
948730b0 | 34 | #include "internal.h" |
1da177e4 | 35 | |
d2921684 EB |
36 | /* Maximum number of mounts in a mount namespace */ |
37 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
38 | ||
0818bf27 AV |
39 | static unsigned int m_hash_mask __read_mostly; |
40 | static unsigned int m_hash_shift __read_mostly; | |
41 | static unsigned int mp_hash_mask __read_mostly; | |
42 | static unsigned int mp_hash_shift __read_mostly; | |
43 | ||
44 | static __initdata unsigned long mhash_entries; | |
45 | static int __init set_mhash_entries(char *str) | |
46 | { | |
47 | if (!str) | |
48 | return 0; | |
49 | mhash_entries = simple_strtoul(str, &str, 0); | |
50 | return 1; | |
51 | } | |
52 | __setup("mhash_entries=", set_mhash_entries); | |
53 | ||
54 | static __initdata unsigned long mphash_entries; | |
55 | static int __init set_mphash_entries(char *str) | |
56 | { | |
57 | if (!str) | |
58 | return 0; | |
59 | mphash_entries = simple_strtoul(str, &str, 0); | |
60 | return 1; | |
61 | } | |
62 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 63 | |
c7999c36 | 64 | static u64 event; |
73cd49ec | 65 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 66 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 67 | |
38129a13 | 68 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 69 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 70 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 71 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
72 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
73 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 74 | |
f87fd4c2 | 75 | /* /sys/fs */ |
00d26666 GKH |
76 | struct kobject *fs_kobj; |
77 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 78 | |
99b7db7b NP |
79 | /* |
80 | * vfsmount lock may be taken for read to prevent changes to the | |
81 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
82 | * up the tree. | |
83 | * | |
84 | * It should be taken for write in all cases where the vfsmount | |
85 | * tree or hash is modified or when a vfsmount structure is modified. | |
86 | */ | |
48a066e7 | 87 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 88 | |
38129a13 | 89 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 90 | { |
b58fed8b RP |
91 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
92 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
93 | tmp = tmp + (tmp >> m_hash_shift); |
94 | return &mount_hashtable[tmp & m_hash_mask]; | |
95 | } | |
96 | ||
97 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
98 | { | |
99 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
100 | tmp = tmp + (tmp >> mp_hash_shift); | |
101 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
102 | } |
103 | ||
b105e270 | 104 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 105 | { |
169b480e MW |
106 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
107 | ||
108 | if (res < 0) | |
109 | return res; | |
110 | mnt->mnt_id = res; | |
111 | return 0; | |
73cd49ec MS |
112 | } |
113 | ||
b105e270 | 114 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 115 | { |
169b480e | 116 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
117 | } |
118 | ||
719f5d7f MS |
119 | /* |
120 | * Allocate a new peer group ID | |
719f5d7f | 121 | */ |
4b8b21f4 | 122 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 123 | { |
169b480e | 124 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 125 | |
169b480e MW |
126 | if (res < 0) |
127 | return res; | |
128 | mnt->mnt_group_id = res; | |
129 | return 0; | |
719f5d7f MS |
130 | } |
131 | ||
132 | /* | |
133 | * Release a peer group ID | |
134 | */ | |
4b8b21f4 | 135 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 136 | { |
169b480e | 137 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 138 | mnt->mnt_group_id = 0; |
719f5d7f MS |
139 | } |
140 | ||
b3e19d92 NP |
141 | /* |
142 | * vfsmount lock must be held for read | |
143 | */ | |
83adc753 | 144 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
145 | { |
146 | #ifdef CONFIG_SMP | |
68e8a9fe | 147 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
148 | #else |
149 | preempt_disable(); | |
68e8a9fe | 150 | mnt->mnt_count += n; |
b3e19d92 NP |
151 | preempt_enable(); |
152 | #endif | |
153 | } | |
154 | ||
b3e19d92 NP |
155 | /* |
156 | * vfsmount lock must be held for write | |
157 | */ | |
83adc753 | 158 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
159 | { |
160 | #ifdef CONFIG_SMP | |
f03c6599 | 161 | unsigned int count = 0; |
b3e19d92 NP |
162 | int cpu; |
163 | ||
164 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 165 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
166 | } |
167 | ||
168 | return count; | |
169 | #else | |
68e8a9fe | 170 | return mnt->mnt_count; |
b3e19d92 NP |
171 | #endif |
172 | } | |
173 | ||
87b95ce0 AV |
174 | static void drop_mountpoint(struct fs_pin *p) |
175 | { | |
176 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
87b95ce0 AV |
177 | pin_remove(p); |
178 | mntput(&m->mnt); | |
179 | } | |
180 | ||
b105e270 | 181 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 182 | { |
c63181e6 AV |
183 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
184 | if (mnt) { | |
73cd49ec MS |
185 | int err; |
186 | ||
c63181e6 | 187 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
188 | if (err) |
189 | goto out_free_cache; | |
190 | ||
191 | if (name) { | |
fcc139ae | 192 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 193 | if (!mnt->mnt_devname) |
88b38782 | 194 | goto out_free_id; |
73cd49ec MS |
195 | } |
196 | ||
b3e19d92 | 197 | #ifdef CONFIG_SMP |
c63181e6 AV |
198 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
199 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
200 | goto out_free_devname; |
201 | ||
c63181e6 | 202 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 203 | #else |
c63181e6 AV |
204 | mnt->mnt_count = 1; |
205 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
206 | #endif |
207 | ||
38129a13 | 208 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
209 | INIT_LIST_HEAD(&mnt->mnt_child); |
210 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
211 | INIT_LIST_HEAD(&mnt->mnt_list); | |
212 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
213 | INIT_LIST_HEAD(&mnt->mnt_share); | |
214 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
215 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 216 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 217 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
87b95ce0 | 218 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 219 | } |
c63181e6 | 220 | return mnt; |
88b38782 | 221 | |
d3ef3d73 | 222 | #ifdef CONFIG_SMP |
223 | out_free_devname: | |
fcc139ae | 224 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 225 | #endif |
88b38782 | 226 | out_free_id: |
c63181e6 | 227 | mnt_free_id(mnt); |
88b38782 | 228 | out_free_cache: |
c63181e6 | 229 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 230 | return NULL; |
1da177e4 LT |
231 | } |
232 | ||
3d733633 DH |
233 | /* |
234 | * Most r/o checks on a fs are for operations that take | |
235 | * discrete amounts of time, like a write() or unlink(). | |
236 | * We must keep track of when those operations start | |
237 | * (for permission checks) and when they end, so that | |
238 | * we can determine when writes are able to occur to | |
239 | * a filesystem. | |
240 | */ | |
241 | /* | |
242 | * __mnt_is_readonly: check whether a mount is read-only | |
243 | * @mnt: the mount to check for its write status | |
244 | * | |
245 | * This shouldn't be used directly ouside of the VFS. | |
246 | * It does not guarantee that the filesystem will stay | |
247 | * r/w, just that it is right *now*. This can not and | |
248 | * should not be used in place of IS_RDONLY(inode). | |
249 | * mnt_want/drop_write() will _keep_ the filesystem | |
250 | * r/w. | |
251 | */ | |
43f5e655 | 252 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 253 | { |
43f5e655 | 254 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
255 | } |
256 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
257 | ||
83adc753 | 258 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 259 | { |
260 | #ifdef CONFIG_SMP | |
68e8a9fe | 261 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 262 | #else |
68e8a9fe | 263 | mnt->mnt_writers++; |
d3ef3d73 | 264 | #endif |
265 | } | |
3d733633 | 266 | |
83adc753 | 267 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 268 | { |
d3ef3d73 | 269 | #ifdef CONFIG_SMP |
68e8a9fe | 270 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 271 | #else |
68e8a9fe | 272 | mnt->mnt_writers--; |
d3ef3d73 | 273 | #endif |
3d733633 | 274 | } |
3d733633 | 275 | |
83adc753 | 276 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 277 | { |
d3ef3d73 | 278 | #ifdef CONFIG_SMP |
279 | unsigned int count = 0; | |
3d733633 | 280 | int cpu; |
3d733633 DH |
281 | |
282 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 283 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 284 | } |
3d733633 | 285 | |
d3ef3d73 | 286 | return count; |
287 | #else | |
288 | return mnt->mnt_writers; | |
289 | #endif | |
3d733633 DH |
290 | } |
291 | ||
4ed5e82f MS |
292 | static int mnt_is_readonly(struct vfsmount *mnt) |
293 | { | |
294 | if (mnt->mnt_sb->s_readonly_remount) | |
295 | return 1; | |
296 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
297 | smp_rmb(); | |
298 | return __mnt_is_readonly(mnt); | |
299 | } | |
300 | ||
8366025e | 301 | /* |
eb04c282 JK |
302 | * Most r/o & frozen checks on a fs are for operations that take discrete |
303 | * amounts of time, like a write() or unlink(). We must keep track of when | |
304 | * those operations start (for permission checks) and when they end, so that we | |
305 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
306 | */ |
307 | /** | |
eb04c282 | 308 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 309 | * @m: the mount on which to take a write |
8366025e | 310 | * |
eb04c282 JK |
311 | * This tells the low-level filesystem that a write is about to be performed to |
312 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
313 | * returning success. This operation does not protect against filesystem being | |
314 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
315 | * called. This is effectively a refcount. | |
8366025e | 316 | */ |
eb04c282 | 317 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 318 | { |
83adc753 | 319 | struct mount *mnt = real_mount(m); |
3d733633 | 320 | int ret = 0; |
3d733633 | 321 | |
d3ef3d73 | 322 | preempt_disable(); |
c6653a83 | 323 | mnt_inc_writers(mnt); |
d3ef3d73 | 324 | /* |
c6653a83 | 325 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 326 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
327 | * incremented count after it has set MNT_WRITE_HOLD. | |
328 | */ | |
329 | smp_mb(); | |
6aa7de05 | 330 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 331 | cpu_relax(); |
332 | /* | |
333 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
334 | * be set to match its requirements. So we must not load that until | |
335 | * MNT_WRITE_HOLD is cleared. | |
336 | */ | |
337 | smp_rmb(); | |
4ed5e82f | 338 | if (mnt_is_readonly(m)) { |
c6653a83 | 339 | mnt_dec_writers(mnt); |
3d733633 | 340 | ret = -EROFS; |
3d733633 | 341 | } |
d3ef3d73 | 342 | preempt_enable(); |
eb04c282 JK |
343 | |
344 | return ret; | |
345 | } | |
346 | ||
347 | /** | |
348 | * mnt_want_write - get write access to a mount | |
349 | * @m: the mount on which to take a write | |
350 | * | |
351 | * This tells the low-level filesystem that a write is about to be performed to | |
352 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
353 | * is not frozen) before returning success. When the write operation is | |
354 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
355 | */ | |
356 | int mnt_want_write(struct vfsmount *m) | |
357 | { | |
358 | int ret; | |
359 | ||
360 | sb_start_write(m->mnt_sb); | |
361 | ret = __mnt_want_write(m); | |
362 | if (ret) | |
363 | sb_end_write(m->mnt_sb); | |
3d733633 | 364 | return ret; |
8366025e DH |
365 | } |
366 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
367 | ||
96029c4e | 368 | /** |
369 | * mnt_clone_write - get write access to a mount | |
370 | * @mnt: the mount on which to take a write | |
371 | * | |
372 | * This is effectively like mnt_want_write, except | |
373 | * it must only be used to take an extra write reference | |
374 | * on a mountpoint that we already know has a write reference | |
375 | * on it. This allows some optimisation. | |
376 | * | |
377 | * After finished, mnt_drop_write must be called as usual to | |
378 | * drop the reference. | |
379 | */ | |
380 | int mnt_clone_write(struct vfsmount *mnt) | |
381 | { | |
382 | /* superblock may be r/o */ | |
383 | if (__mnt_is_readonly(mnt)) | |
384 | return -EROFS; | |
385 | preempt_disable(); | |
83adc753 | 386 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 387 | preempt_enable(); |
388 | return 0; | |
389 | } | |
390 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
391 | ||
392 | /** | |
eb04c282 | 393 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 394 | * @file: the file who's mount on which to take a write |
395 | * | |
eb04c282 | 396 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 397 | * do some optimisations if the file is open for write already |
398 | */ | |
eb04c282 | 399 | int __mnt_want_write_file(struct file *file) |
96029c4e | 400 | { |
83f936c7 | 401 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 402 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 403 | else |
404 | return mnt_clone_write(file->f_path.mnt); | |
405 | } | |
eb04c282 | 406 | |
7c6893e3 MS |
407 | /** |
408 | * mnt_want_write_file - get write access to a file's mount | |
409 | * @file: the file who's mount on which to take a write | |
410 | * | |
411 | * This is like mnt_want_write, but it takes a file and can | |
412 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
413 | */ |
414 | int mnt_want_write_file(struct file *file) | |
415 | { | |
416 | int ret; | |
417 | ||
a6795a58 | 418 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
419 | ret = __mnt_want_write_file(file); |
420 | if (ret) | |
a6795a58 | 421 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
422 | return ret; |
423 | } | |
96029c4e | 424 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
425 | ||
8366025e | 426 | /** |
eb04c282 | 427 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
428 | * @mnt: the mount on which to give up write access |
429 | * | |
430 | * Tells the low-level filesystem that we are done | |
431 | * performing writes to it. Must be matched with | |
eb04c282 | 432 | * __mnt_want_write() call above. |
8366025e | 433 | */ |
eb04c282 | 434 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 435 | { |
d3ef3d73 | 436 | preempt_disable(); |
83adc753 | 437 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 438 | preempt_enable(); |
8366025e | 439 | } |
eb04c282 JK |
440 | |
441 | /** | |
442 | * mnt_drop_write - give up write access to a mount | |
443 | * @mnt: the mount on which to give up write access | |
444 | * | |
445 | * Tells the low-level filesystem that we are done performing writes to it and | |
446 | * also allows filesystem to be frozen again. Must be matched with | |
447 | * mnt_want_write() call above. | |
448 | */ | |
449 | void mnt_drop_write(struct vfsmount *mnt) | |
450 | { | |
451 | __mnt_drop_write(mnt); | |
452 | sb_end_write(mnt->mnt_sb); | |
453 | } | |
8366025e DH |
454 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
455 | ||
eb04c282 JK |
456 | void __mnt_drop_write_file(struct file *file) |
457 | { | |
458 | __mnt_drop_write(file->f_path.mnt); | |
459 | } | |
460 | ||
7c6893e3 MS |
461 | void mnt_drop_write_file(struct file *file) |
462 | { | |
a6795a58 | 463 | __mnt_drop_write_file(file); |
7c6893e3 MS |
464 | sb_end_write(file_inode(file)->i_sb); |
465 | } | |
2a79f17e AV |
466 | EXPORT_SYMBOL(mnt_drop_write_file); |
467 | ||
83adc753 | 468 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 469 | { |
3d733633 DH |
470 | int ret = 0; |
471 | ||
719ea2fb | 472 | lock_mount_hash(); |
83adc753 | 473 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 474 | /* |
d3ef3d73 | 475 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
476 | * should be visible before we do. | |
3d733633 | 477 | */ |
d3ef3d73 | 478 | smp_mb(); |
479 | ||
3d733633 | 480 | /* |
d3ef3d73 | 481 | * With writers on hold, if this value is zero, then there are |
482 | * definitely no active writers (although held writers may subsequently | |
483 | * increment the count, they'll have to wait, and decrement it after | |
484 | * seeing MNT_READONLY). | |
485 | * | |
486 | * It is OK to have counter incremented on one CPU and decremented on | |
487 | * another: the sum will add up correctly. The danger would be when we | |
488 | * sum up each counter, if we read a counter before it is incremented, | |
489 | * but then read another CPU's count which it has been subsequently | |
490 | * decremented from -- we would see more decrements than we should. | |
491 | * MNT_WRITE_HOLD protects against this scenario, because | |
492 | * mnt_want_write first increments count, then smp_mb, then spins on | |
493 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
494 | * we're counting up here. | |
3d733633 | 495 | */ |
c6653a83 | 496 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 497 | ret = -EBUSY; |
498 | else | |
83adc753 | 499 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 500 | /* |
501 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
502 | * that become unheld will see MNT_READONLY. | |
503 | */ | |
504 | smp_wmb(); | |
83adc753 | 505 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 506 | unlock_mount_hash(); |
3d733633 | 507 | return ret; |
8366025e | 508 | } |
8366025e | 509 | |
43f5e655 | 510 | static int __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 511 | { |
719ea2fb | 512 | lock_mount_hash(); |
83adc753 | 513 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 514 | unlock_mount_hash(); |
43f5e655 | 515 | return 0; |
2e4b7fcd DH |
516 | } |
517 | ||
4ed5e82f MS |
518 | int sb_prepare_remount_readonly(struct super_block *sb) |
519 | { | |
520 | struct mount *mnt; | |
521 | int err = 0; | |
522 | ||
8e8b8796 MS |
523 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
524 | if (atomic_long_read(&sb->s_remove_count)) | |
525 | return -EBUSY; | |
526 | ||
719ea2fb | 527 | lock_mount_hash(); |
4ed5e82f MS |
528 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
529 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
530 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
531 | smp_mb(); | |
532 | if (mnt_get_writers(mnt) > 0) { | |
533 | err = -EBUSY; | |
534 | break; | |
535 | } | |
536 | } | |
537 | } | |
8e8b8796 MS |
538 | if (!err && atomic_long_read(&sb->s_remove_count)) |
539 | err = -EBUSY; | |
540 | ||
4ed5e82f MS |
541 | if (!err) { |
542 | sb->s_readonly_remount = 1; | |
543 | smp_wmb(); | |
544 | } | |
545 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
546 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
547 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
548 | } | |
719ea2fb | 549 | unlock_mount_hash(); |
4ed5e82f MS |
550 | |
551 | return err; | |
552 | } | |
553 | ||
b105e270 | 554 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 555 | { |
fcc139ae | 556 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 557 | #ifdef CONFIG_SMP |
68e8a9fe | 558 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 559 | #endif |
b105e270 | 560 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
561 | } |
562 | ||
8ffcb32e DH |
563 | static void delayed_free_vfsmnt(struct rcu_head *head) |
564 | { | |
565 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
566 | } | |
567 | ||
48a066e7 | 568 | /* call under rcu_read_lock */ |
294d71ff | 569 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
570 | { |
571 | struct mount *mnt; | |
572 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 573 | return 1; |
48a066e7 | 574 | if (bastard == NULL) |
294d71ff | 575 | return 0; |
48a066e7 AV |
576 | mnt = real_mount(bastard); |
577 | mnt_add_count(mnt, 1); | |
119e1ef8 | 578 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 579 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 580 | return 0; |
48a066e7 AV |
581 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
582 | mnt_add_count(mnt, -1); | |
294d71ff AV |
583 | return 1; |
584 | } | |
119e1ef8 AV |
585 | lock_mount_hash(); |
586 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
587 | mnt_add_count(mnt, -1); | |
588 | unlock_mount_hash(); | |
589 | return 1; | |
590 | } | |
591 | unlock_mount_hash(); | |
592 | /* caller will mntput() */ | |
294d71ff AV |
593 | return -1; |
594 | } | |
595 | ||
596 | /* call under rcu_read_lock */ | |
597 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
598 | { | |
599 | int res = __legitimize_mnt(bastard, seq); | |
600 | if (likely(!res)) | |
601 | return true; | |
602 | if (unlikely(res < 0)) { | |
603 | rcu_read_unlock(); | |
604 | mntput(bastard); | |
605 | rcu_read_lock(); | |
48a066e7 | 606 | } |
48a066e7 AV |
607 | return false; |
608 | } | |
609 | ||
1da177e4 | 610 | /* |
474279dc | 611 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 612 | * call under rcu_read_lock() |
1da177e4 | 613 | */ |
474279dc | 614 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 615 | { |
38129a13 | 616 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
617 | struct mount *p; |
618 | ||
38129a13 | 619 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
620 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
621 | return p; | |
622 | return NULL; | |
623 | } | |
624 | ||
a05964f3 | 625 | /* |
f015f126 DH |
626 | * lookup_mnt - Return the first child mount mounted at path |
627 | * | |
628 | * "First" means first mounted chronologically. If you create the | |
629 | * following mounts: | |
630 | * | |
631 | * mount /dev/sda1 /mnt | |
632 | * mount /dev/sda2 /mnt | |
633 | * mount /dev/sda3 /mnt | |
634 | * | |
635 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
636 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
637 | * /dev/sda2, then /dev/sda3, then NULL. | |
638 | * | |
639 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 640 | */ |
ca71cf71 | 641 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 642 | { |
c7105365 | 643 | struct mount *child_mnt; |
48a066e7 AV |
644 | struct vfsmount *m; |
645 | unsigned seq; | |
99b7db7b | 646 | |
48a066e7 AV |
647 | rcu_read_lock(); |
648 | do { | |
649 | seq = read_seqbegin(&mount_lock); | |
650 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
651 | m = child_mnt ? &child_mnt->mnt : NULL; | |
652 | } while (!legitimize_mnt(m, seq)); | |
653 | rcu_read_unlock(); | |
654 | return m; | |
a05964f3 RP |
655 | } |
656 | ||
7af1364f EB |
657 | /* |
658 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
659 | * current mount namespace. | |
660 | * | |
661 | * The common case is dentries are not mountpoints at all and that | |
662 | * test is handled inline. For the slow case when we are actually | |
663 | * dealing with a mountpoint of some kind, walk through all of the | |
664 | * mounts in the current mount namespace and test to see if the dentry | |
665 | * is a mountpoint. | |
666 | * | |
667 | * The mount_hashtable is not usable in the context because we | |
668 | * need to identify all mounts that may be in the current mount | |
669 | * namespace not just a mount that happens to have some specified | |
670 | * parent mount. | |
671 | */ | |
672 | bool __is_local_mountpoint(struct dentry *dentry) | |
673 | { | |
674 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
675 | struct mount *mnt; | |
676 | bool is_covered = false; | |
677 | ||
678 | if (!d_mountpoint(dentry)) | |
679 | goto out; | |
680 | ||
681 | down_read(&namespace_sem); | |
682 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
683 | is_covered = (mnt->mnt_mountpoint == dentry); | |
684 | if (is_covered) | |
685 | break; | |
686 | } | |
687 | up_read(&namespace_sem); | |
688 | out: | |
689 | return is_covered; | |
690 | } | |
691 | ||
e2dfa935 | 692 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 693 | { |
0818bf27 | 694 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
695 | struct mountpoint *mp; |
696 | ||
0818bf27 | 697 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 698 | if (mp->m_dentry == dentry) { |
84d17192 AV |
699 | mp->m_count++; |
700 | return mp; | |
701 | } | |
702 | } | |
e2dfa935 EB |
703 | return NULL; |
704 | } | |
705 | ||
3895dbf8 | 706 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 707 | { |
3895dbf8 | 708 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 709 | int ret; |
84d17192 | 710 | |
3895dbf8 | 711 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
712 | /* might be worth a WARN_ON() */ |
713 | if (d_unlinked(dentry)) | |
714 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
715 | mountpoint: |
716 | read_seqlock_excl(&mount_lock); | |
717 | mp = lookup_mountpoint(dentry); | |
718 | read_sequnlock_excl(&mount_lock); | |
719 | if (mp) | |
720 | goto done; | |
721 | } | |
722 | ||
723 | if (!new) | |
724 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
725 | if (!new) | |
84d17192 AV |
726 | return ERR_PTR(-ENOMEM); |
727 | ||
3895dbf8 EB |
728 | |
729 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 730 | ret = d_set_mounted(dentry); |
eed81007 | 731 | |
3895dbf8 EB |
732 | /* Someone else set d_mounted? */ |
733 | if (ret == -EBUSY) | |
734 | goto mountpoint; | |
735 | ||
736 | /* The dentry is not available as a mountpoint? */ | |
737 | mp = ERR_PTR(ret); | |
738 | if (ret) | |
739 | goto done; | |
740 | ||
741 | /* Add the new mountpoint to the hash table */ | |
742 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 743 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
744 | new->m_count = 1; |
745 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
746 | INIT_HLIST_HEAD(&new->m_list); | |
747 | read_sequnlock_excl(&mount_lock); | |
748 | ||
749 | mp = new; | |
750 | new = NULL; | |
751 | done: | |
752 | kfree(new); | |
84d17192 AV |
753 | return mp; |
754 | } | |
755 | ||
4edbe133 AV |
756 | /* |
757 | * vfsmount lock must be held. Additionally, the caller is responsible | |
758 | * for serializing calls for given disposal list. | |
759 | */ | |
760 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
761 | { |
762 | if (!--mp->m_count) { | |
763 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 764 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
765 | spin_lock(&dentry->d_lock); |
766 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
767 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 768 | dput_to_list(dentry, list); |
0818bf27 | 769 | hlist_del(&mp->m_hash); |
84d17192 AV |
770 | kfree(mp); |
771 | } | |
772 | } | |
773 | ||
4edbe133 AV |
774 | /* called with namespace_lock and vfsmount lock */ |
775 | static void put_mountpoint(struct mountpoint *mp) | |
776 | { | |
777 | __put_mountpoint(mp, &ex_mountpoints); | |
778 | } | |
779 | ||
143c8c91 | 780 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 781 | { |
6b3286ed | 782 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
783 | } |
784 | ||
99b7db7b NP |
785 | /* |
786 | * vfsmount lock must be held for write | |
787 | */ | |
6b3286ed | 788 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
789 | { |
790 | if (ns) { | |
791 | ns->event = ++event; | |
792 | wake_up_interruptible(&ns->poll); | |
793 | } | |
794 | } | |
795 | ||
99b7db7b NP |
796 | /* |
797 | * vfsmount lock must be held for write | |
798 | */ | |
6b3286ed | 799 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
800 | { |
801 | if (ns && ns->event != event) { | |
802 | ns->event = event; | |
803 | wake_up_interruptible(&ns->poll); | |
804 | } | |
805 | } | |
806 | ||
99b7db7b NP |
807 | /* |
808 | * vfsmount lock must be held for write | |
809 | */ | |
e4e59906 | 810 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 811 | { |
e4e59906 | 812 | struct mountpoint *mp; |
0714a533 | 813 | mnt->mnt_parent = mnt; |
a73324da | 814 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 815 | list_del_init(&mnt->mnt_child); |
38129a13 | 816 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 817 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 818 | mp = mnt->mnt_mp; |
84d17192 | 819 | mnt->mnt_mp = NULL; |
e4e59906 | 820 | return mp; |
1da177e4 LT |
821 | } |
822 | ||
6a46c573 EB |
823 | /* |
824 | * vfsmount lock must be held for write | |
825 | */ | |
826 | static void umount_mnt(struct mount *mnt) | |
827 | { | |
e4e59906 | 828 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
829 | } |
830 | ||
99b7db7b NP |
831 | /* |
832 | * vfsmount lock must be held for write | |
833 | */ | |
84d17192 AV |
834 | void mnt_set_mountpoint(struct mount *mnt, |
835 | struct mountpoint *mp, | |
44d964d6 | 836 | struct mount *child_mnt) |
b90fa9ae | 837 | { |
84d17192 | 838 | mp->m_count++; |
3a2393d7 | 839 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 840 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 841 | child_mnt->mnt_parent = mnt; |
84d17192 | 842 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 843 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
844 | } |
845 | ||
1064f874 EB |
846 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
847 | { | |
848 | hlist_add_head_rcu(&mnt->mnt_hash, | |
849 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
850 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
851 | } | |
852 | ||
99b7db7b NP |
853 | /* |
854 | * vfsmount lock must be held for write | |
855 | */ | |
84d17192 AV |
856 | static void attach_mnt(struct mount *mnt, |
857 | struct mount *parent, | |
858 | struct mountpoint *mp) | |
1da177e4 | 859 | { |
84d17192 | 860 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 861 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
862 | } |
863 | ||
1064f874 | 864 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 865 | { |
1064f874 | 866 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
867 | struct mount *old_parent = mnt->mnt_parent; |
868 | ||
869 | list_del_init(&mnt->mnt_child); | |
870 | hlist_del_init(&mnt->mnt_mp_list); | |
871 | hlist_del_init_rcu(&mnt->mnt_hash); | |
872 | ||
873 | attach_mnt(mnt, parent, mp); | |
874 | ||
875 | put_mountpoint(old_mp); | |
1064f874 | 876 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
877 | } |
878 | ||
b90fa9ae | 879 | /* |
99b7db7b | 880 | * vfsmount lock must be held for write |
b90fa9ae | 881 | */ |
1064f874 | 882 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 883 | { |
0714a533 | 884 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 885 | struct mount *m; |
b90fa9ae | 886 | LIST_HEAD(head); |
143c8c91 | 887 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 888 | |
0714a533 | 889 | BUG_ON(parent == mnt); |
b90fa9ae | 890 | |
1a4eeaf2 | 891 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 892 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 893 | m->mnt_ns = n; |
f03c6599 | 894 | |
b90fa9ae RP |
895 | list_splice(&head, n->list.prev); |
896 | ||
d2921684 EB |
897 | n->mounts += n->pending_mounts; |
898 | n->pending_mounts = 0; | |
899 | ||
1064f874 | 900 | __attach_mnt(mnt, parent); |
6b3286ed | 901 | touch_mnt_namespace(n); |
1da177e4 LT |
902 | } |
903 | ||
909b0a88 | 904 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 905 | { |
6b41d536 AV |
906 | struct list_head *next = p->mnt_mounts.next; |
907 | if (next == &p->mnt_mounts) { | |
1da177e4 | 908 | while (1) { |
909b0a88 | 909 | if (p == root) |
1da177e4 | 910 | return NULL; |
6b41d536 AV |
911 | next = p->mnt_child.next; |
912 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 913 | break; |
0714a533 | 914 | p = p->mnt_parent; |
1da177e4 LT |
915 | } |
916 | } | |
6b41d536 | 917 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
918 | } |
919 | ||
315fc83e | 920 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 921 | { |
6b41d536 AV |
922 | struct list_head *prev = p->mnt_mounts.prev; |
923 | while (prev != &p->mnt_mounts) { | |
924 | p = list_entry(prev, struct mount, mnt_child); | |
925 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
926 | } |
927 | return p; | |
928 | } | |
929 | ||
8f291889 AV |
930 | /** |
931 | * vfs_create_mount - Create a mount for a configured superblock | |
932 | * @fc: The configuration context with the superblock attached | |
933 | * | |
934 | * Create a mount to an already configured superblock. If necessary, the | |
935 | * caller should invoke vfs_get_tree() before calling this. | |
936 | * | |
937 | * Note that this does not attach the mount to anything. | |
938 | */ | |
939 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 940 | { |
b105e270 | 941 | struct mount *mnt; |
9d412a43 | 942 | |
8f291889 AV |
943 | if (!fc->root) |
944 | return ERR_PTR(-EINVAL); | |
9d412a43 | 945 | |
8f291889 | 946 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
947 | if (!mnt) |
948 | return ERR_PTR(-ENOMEM); | |
949 | ||
8f291889 | 950 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 951 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 952 | |
8f291889 AV |
953 | atomic_inc(&fc->root->d_sb->s_active); |
954 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
955 | mnt->mnt.mnt_root = dget(fc->root); | |
956 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
957 | mnt->mnt_parent = mnt; | |
9d412a43 | 958 | |
719ea2fb | 959 | lock_mount_hash(); |
8f291889 | 960 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 961 | unlock_mount_hash(); |
b105e270 | 962 | return &mnt->mnt; |
9d412a43 | 963 | } |
8f291889 AV |
964 | EXPORT_SYMBOL(vfs_create_mount); |
965 | ||
966 | struct vfsmount *fc_mount(struct fs_context *fc) | |
967 | { | |
968 | int err = vfs_get_tree(fc); | |
969 | if (!err) { | |
970 | up_write(&fc->root->d_sb->s_umount); | |
971 | return vfs_create_mount(fc); | |
972 | } | |
973 | return ERR_PTR(err); | |
974 | } | |
975 | EXPORT_SYMBOL(fc_mount); | |
976 | ||
9bc61ab1 DH |
977 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
978 | int flags, const char *name, | |
979 | void *data) | |
9d412a43 | 980 | { |
9bc61ab1 | 981 | struct fs_context *fc; |
8f291889 | 982 | struct vfsmount *mnt; |
9bc61ab1 | 983 | int ret = 0; |
9d412a43 AV |
984 | |
985 | if (!type) | |
3e1aeb00 | 986 | return ERR_PTR(-EINVAL); |
9d412a43 | 987 | |
9bc61ab1 DH |
988 | fc = fs_context_for_mount(type, flags); |
989 | if (IS_ERR(fc)) | |
990 | return ERR_CAST(fc); | |
991 | ||
3e1aeb00 DH |
992 | if (name) |
993 | ret = vfs_parse_fs_string(fc, "source", | |
994 | name, strlen(name)); | |
9bc61ab1 DH |
995 | if (!ret) |
996 | ret = parse_monolithic_mount_data(fc, data); | |
997 | if (!ret) | |
8f291889 AV |
998 | mnt = fc_mount(fc); |
999 | else | |
1000 | mnt = ERR_PTR(ret); | |
9d412a43 | 1001 | |
9bc61ab1 | 1002 | put_fs_context(fc); |
8f291889 | 1003 | return mnt; |
9d412a43 AV |
1004 | } |
1005 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1006 | ||
93faccbb EB |
1007 | struct vfsmount * |
1008 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1009 | const char *name, void *data) | |
1010 | { | |
1011 | /* Until it is worked out how to pass the user namespace | |
1012 | * through from the parent mount to the submount don't support | |
1013 | * unprivileged mounts with submounts. | |
1014 | */ | |
1015 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1016 | return ERR_PTR(-EPERM); | |
1017 | ||
e462ec50 | 1018 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1019 | } |
1020 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1021 | ||
87129cc0 | 1022 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1023 | int flag) |
1da177e4 | 1024 | { |
87129cc0 | 1025 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1026 | struct mount *mnt; |
1027 | int err; | |
1da177e4 | 1028 | |
be34d1a3 DH |
1029 | mnt = alloc_vfsmnt(old->mnt_devname); |
1030 | if (!mnt) | |
1031 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1032 | |
7a472ef4 | 1033 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1034 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1035 | else | |
1036 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1037 | |
be34d1a3 DH |
1038 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1039 | err = mnt_alloc_group_id(mnt); | |
1040 | if (err) | |
1041 | goto out_free; | |
1da177e4 | 1042 | } |
be34d1a3 | 1043 | |
16a34adb AV |
1044 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1045 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1046 | |
be34d1a3 DH |
1047 | atomic_inc(&sb->s_active); |
1048 | mnt->mnt.mnt_sb = sb; | |
1049 | mnt->mnt.mnt_root = dget(root); | |
1050 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1051 | mnt->mnt_parent = mnt; | |
719ea2fb | 1052 | lock_mount_hash(); |
be34d1a3 | 1053 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1054 | unlock_mount_hash(); |
be34d1a3 | 1055 | |
7a472ef4 EB |
1056 | if ((flag & CL_SLAVE) || |
1057 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1058 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1059 | mnt->mnt_master = old; | |
1060 | CLEAR_MNT_SHARED(mnt); | |
1061 | } else if (!(flag & CL_PRIVATE)) { | |
1062 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1063 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1064 | if (IS_MNT_SLAVE(old)) | |
1065 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1066 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1067 | } else { |
1068 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1069 | } |
1070 | if (flag & CL_MAKE_SHARED) | |
1071 | set_mnt_shared(mnt); | |
1072 | ||
1073 | /* stick the duplicate mount on the same expiry list | |
1074 | * as the original if that was on one */ | |
1075 | if (flag & CL_EXPIRE) { | |
1076 | if (!list_empty(&old->mnt_expire)) | |
1077 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1078 | } | |
1079 | ||
cb338d06 | 1080 | return mnt; |
719f5d7f MS |
1081 | |
1082 | out_free: | |
8ffcb32e | 1083 | mnt_free_id(mnt); |
719f5d7f | 1084 | free_vfsmnt(mnt); |
be34d1a3 | 1085 | return ERR_PTR(err); |
1da177e4 LT |
1086 | } |
1087 | ||
9ea459e1 AV |
1088 | static void cleanup_mnt(struct mount *mnt) |
1089 | { | |
1090 | /* | |
1091 | * This probably indicates that somebody messed | |
1092 | * up a mnt_want/drop_write() pair. If this | |
1093 | * happens, the filesystem was probably unable | |
1094 | * to make r/w->r/o transitions. | |
1095 | */ | |
1096 | /* | |
1097 | * The locking used to deal with mnt_count decrement provides barriers, | |
1098 | * so mnt_get_writers() below is safe. | |
1099 | */ | |
1100 | WARN_ON(mnt_get_writers(mnt)); | |
1101 | if (unlikely(mnt->mnt_pins.first)) | |
1102 | mnt_pin_kill(mnt); | |
1103 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1104 | dput(mnt->mnt.mnt_root); | |
1105 | deactivate_super(mnt->mnt.mnt_sb); | |
1106 | mnt_free_id(mnt); | |
1107 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1108 | } | |
1109 | ||
1110 | static void __cleanup_mnt(struct rcu_head *head) | |
1111 | { | |
1112 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1113 | } | |
1114 | ||
1115 | static LLIST_HEAD(delayed_mntput_list); | |
1116 | static void delayed_mntput(struct work_struct *unused) | |
1117 | { | |
1118 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1119 | struct mount *m, *t; |
9ea459e1 | 1120 | |
29785735 BP |
1121 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1122 | cleanup_mnt(m); | |
9ea459e1 AV |
1123 | } |
1124 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1125 | ||
900148dc | 1126 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1127 | { |
4edbe133 AV |
1128 | LIST_HEAD(list); |
1129 | ||
48a066e7 | 1130 | rcu_read_lock(); |
9ea0a46c AV |
1131 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1132 | /* | |
1133 | * Since we don't do lock_mount_hash() here, | |
1134 | * ->mnt_ns can change under us. However, if it's | |
1135 | * non-NULL, then there's a reference that won't | |
1136 | * be dropped until after an RCU delay done after | |
1137 | * turning ->mnt_ns NULL. So if we observe it | |
1138 | * non-NULL under rcu_read_lock(), the reference | |
1139 | * we are dropping is not the final one. | |
1140 | */ | |
1141 | mnt_add_count(mnt, -1); | |
48a066e7 | 1142 | rcu_read_unlock(); |
f03c6599 | 1143 | return; |
b3e19d92 | 1144 | } |
719ea2fb | 1145 | lock_mount_hash(); |
119e1ef8 AV |
1146 | /* |
1147 | * make sure that if __legitimize_mnt() has not seen us grab | |
1148 | * mount_lock, we'll see their refcount increment here. | |
1149 | */ | |
1150 | smp_mb(); | |
9ea0a46c | 1151 | mnt_add_count(mnt, -1); |
b3e19d92 | 1152 | if (mnt_get_count(mnt)) { |
48a066e7 | 1153 | rcu_read_unlock(); |
719ea2fb | 1154 | unlock_mount_hash(); |
99b7db7b NP |
1155 | return; |
1156 | } | |
48a066e7 AV |
1157 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1158 | rcu_read_unlock(); | |
1159 | unlock_mount_hash(); | |
1160 | return; | |
1161 | } | |
1162 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1163 | rcu_read_unlock(); | |
962830df | 1164 | |
39f7c4db | 1165 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1166 | |
1167 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1168 | struct mount *p, *tmp; | |
1169 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1170 | __put_mountpoint(unhash_mnt(p), &list); |
ce07d891 EB |
1171 | } |
1172 | } | |
719ea2fb | 1173 | unlock_mount_hash(); |
4edbe133 | 1174 | shrink_dentry_list(&list); |
649a795a | 1175 | |
9ea459e1 AV |
1176 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1177 | struct task_struct *task = current; | |
1178 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1179 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1180 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1181 | return; | |
1182 | } | |
1183 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1184 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1185 | return; | |
1186 | } | |
1187 | cleanup_mnt(mnt); | |
b3e19d92 | 1188 | } |
b3e19d92 NP |
1189 | |
1190 | void mntput(struct vfsmount *mnt) | |
1191 | { | |
1192 | if (mnt) { | |
863d684f | 1193 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1194 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1195 | if (unlikely(m->mnt_expiry_mark)) |
1196 | m->mnt_expiry_mark = 0; | |
1197 | mntput_no_expire(m); | |
b3e19d92 NP |
1198 | } |
1199 | } | |
1200 | EXPORT_SYMBOL(mntput); | |
1201 | ||
1202 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1203 | { | |
1204 | if (mnt) | |
83adc753 | 1205 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1206 | return mnt; |
1207 | } | |
1208 | EXPORT_SYMBOL(mntget); | |
1209 | ||
c6609c0a IK |
1210 | /* path_is_mountpoint() - Check if path is a mount in the current |
1211 | * namespace. | |
1212 | * | |
1213 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1214 | * not mounted in any namespace and that common case is handled inline. | |
1215 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1216 | * mounts using a given dentry in a different namespace. This function | |
1217 | * checks if the passed in path is a mountpoint rather than the dentry | |
1218 | * alone. | |
1219 | */ | |
1220 | bool path_is_mountpoint(const struct path *path) | |
1221 | { | |
1222 | unsigned seq; | |
1223 | bool res; | |
1224 | ||
1225 | if (!d_mountpoint(path->dentry)) | |
1226 | return false; | |
1227 | ||
1228 | rcu_read_lock(); | |
1229 | do { | |
1230 | seq = read_seqbegin(&mount_lock); | |
1231 | res = __path_is_mountpoint(path); | |
1232 | } while (read_seqretry(&mount_lock, seq)); | |
1233 | rcu_read_unlock(); | |
1234 | ||
1235 | return res; | |
1236 | } | |
1237 | EXPORT_SYMBOL(path_is_mountpoint); | |
1238 | ||
ca71cf71 | 1239 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1240 | { |
3064c356 AV |
1241 | struct mount *p; |
1242 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1243 | if (IS_ERR(p)) | |
1244 | return ERR_CAST(p); | |
1245 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1246 | return &p->mnt; | |
7b7b1ace | 1247 | } |
1da177e4 | 1248 | |
a1a2c409 | 1249 | #ifdef CONFIG_PROC_FS |
0226f492 | 1250 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1251 | static void *m_start(struct seq_file *m, loff_t *pos) |
1252 | { | |
ede1bf0d | 1253 | struct proc_mounts *p = m->private; |
1da177e4 | 1254 | |
390c6843 | 1255 | down_read(&namespace_sem); |
c7999c36 AV |
1256 | if (p->cached_event == p->ns->event) { |
1257 | void *v = p->cached_mount; | |
1258 | if (*pos == p->cached_index) | |
1259 | return v; | |
1260 | if (*pos == p->cached_index + 1) { | |
1261 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1262 | return p->cached_mount = v; | |
1263 | } | |
1264 | } | |
1265 | ||
1266 | p->cached_event = p->ns->event; | |
1267 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1268 | p->cached_index = *pos; | |
1269 | return p->cached_mount; | |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1273 | { | |
ede1bf0d | 1274 | struct proc_mounts *p = m->private; |
b0765fb8 | 1275 | |
c7999c36 AV |
1276 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1277 | p->cached_index = *pos; | |
1278 | return p->cached_mount; | |
1da177e4 LT |
1279 | } |
1280 | ||
1281 | static void m_stop(struct seq_file *m, void *v) | |
1282 | { | |
390c6843 | 1283 | up_read(&namespace_sem); |
1da177e4 LT |
1284 | } |
1285 | ||
0226f492 | 1286 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1287 | { |
ede1bf0d | 1288 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1289 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1290 | return p->show(m, &r->mnt); |
1da177e4 LT |
1291 | } |
1292 | ||
a1a2c409 | 1293 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1294 | .start = m_start, |
1295 | .next = m_next, | |
1296 | .stop = m_stop, | |
0226f492 | 1297 | .show = m_show, |
b4629fe2 | 1298 | }; |
a1a2c409 | 1299 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1300 | |
1da177e4 LT |
1301 | /** |
1302 | * may_umount_tree - check if a mount tree is busy | |
1303 | * @mnt: root of mount tree | |
1304 | * | |
1305 | * This is called to check if a tree of mounts has any | |
1306 | * open files, pwds, chroots or sub mounts that are | |
1307 | * busy. | |
1308 | */ | |
909b0a88 | 1309 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1310 | { |
909b0a88 | 1311 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1312 | int actual_refs = 0; |
1313 | int minimum_refs = 0; | |
315fc83e | 1314 | struct mount *p; |
909b0a88 | 1315 | BUG_ON(!m); |
1da177e4 | 1316 | |
b3e19d92 | 1317 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1318 | lock_mount_hash(); |
909b0a88 | 1319 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1320 | actual_refs += mnt_get_count(p); |
1da177e4 | 1321 | minimum_refs += 2; |
1da177e4 | 1322 | } |
719ea2fb | 1323 | unlock_mount_hash(); |
1da177e4 LT |
1324 | |
1325 | if (actual_refs > minimum_refs) | |
e3474a8e | 1326 | return 0; |
1da177e4 | 1327 | |
e3474a8e | 1328 | return 1; |
1da177e4 LT |
1329 | } |
1330 | ||
1331 | EXPORT_SYMBOL(may_umount_tree); | |
1332 | ||
1333 | /** | |
1334 | * may_umount - check if a mount point is busy | |
1335 | * @mnt: root of mount | |
1336 | * | |
1337 | * This is called to check if a mount point has any | |
1338 | * open files, pwds, chroots or sub mounts. If the | |
1339 | * mount has sub mounts this will return busy | |
1340 | * regardless of whether the sub mounts are busy. | |
1341 | * | |
1342 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1343 | * give false negatives. The main reason why it's here is that we need | |
1344 | * a non-destructive way to look for easily umountable filesystems. | |
1345 | */ | |
1346 | int may_umount(struct vfsmount *mnt) | |
1347 | { | |
e3474a8e | 1348 | int ret = 1; |
8ad08d8a | 1349 | down_read(&namespace_sem); |
719ea2fb | 1350 | lock_mount_hash(); |
1ab59738 | 1351 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1352 | ret = 0; |
719ea2fb | 1353 | unlock_mount_hash(); |
8ad08d8a | 1354 | up_read(&namespace_sem); |
a05964f3 | 1355 | return ret; |
1da177e4 LT |
1356 | } |
1357 | ||
1358 | EXPORT_SYMBOL(may_umount); | |
1359 | ||
97216be0 | 1360 | static void namespace_unlock(void) |
70fbcdf4 | 1361 | { |
a3b3c562 | 1362 | struct hlist_head head; |
4edbe133 | 1363 | LIST_HEAD(list); |
97216be0 | 1364 | |
a3b3c562 | 1365 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1366 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1367 | |
97216be0 AV |
1368 | up_write(&namespace_sem); |
1369 | ||
4edbe133 AV |
1370 | shrink_dentry_list(&list); |
1371 | ||
a3b3c562 EB |
1372 | if (likely(hlist_empty(&head))) |
1373 | return; | |
1374 | ||
22cb7405 | 1375 | synchronize_rcu_expedited(); |
48a066e7 | 1376 | |
87b95ce0 | 1377 | group_pin_kill(&head); |
70fbcdf4 RP |
1378 | } |
1379 | ||
97216be0 | 1380 | static inline void namespace_lock(void) |
e3197d83 | 1381 | { |
97216be0 | 1382 | down_write(&namespace_sem); |
e3197d83 AV |
1383 | } |
1384 | ||
e819f152 EB |
1385 | enum umount_tree_flags { |
1386 | UMOUNT_SYNC = 1, | |
1387 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1388 | UMOUNT_CONNECTED = 4, |
e819f152 | 1389 | }; |
f2d0a123 EB |
1390 | |
1391 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1392 | { | |
1393 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1394 | if (how & UMOUNT_SYNC) | |
1395 | return true; | |
1396 | ||
1397 | /* A mount without a parent has nothing to be connected to */ | |
1398 | if (!mnt_has_parent(mnt)) | |
1399 | return true; | |
1400 | ||
1401 | /* Because the reference counting rules change when mounts are | |
1402 | * unmounted and connected, umounted mounts may not be | |
1403 | * connected to mounted mounts. | |
1404 | */ | |
1405 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1406 | return true; | |
1407 | ||
1408 | /* Has it been requested that the mount remain connected? */ | |
1409 | if (how & UMOUNT_CONNECTED) | |
1410 | return false; | |
1411 | ||
1412 | /* Is the mount locked such that it needs to remain connected? */ | |
1413 | if (IS_MNT_LOCKED(mnt)) | |
1414 | return false; | |
1415 | ||
1416 | /* By default disconnect the mount */ | |
1417 | return true; | |
1418 | } | |
1419 | ||
99b7db7b | 1420 | /* |
48a066e7 | 1421 | * mount_lock must be held |
99b7db7b NP |
1422 | * namespace_sem must be held for write |
1423 | */ | |
e819f152 | 1424 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1425 | { |
c003b26f | 1426 | LIST_HEAD(tmp_list); |
315fc83e | 1427 | struct mount *p; |
1da177e4 | 1428 | |
5d88457e EB |
1429 | if (how & UMOUNT_PROPAGATE) |
1430 | propagate_mount_unlock(mnt); | |
1431 | ||
c003b26f | 1432 | /* Gather the mounts to umount */ |
590ce4bc EB |
1433 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1434 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1435 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1436 | } |
1da177e4 | 1437 | |
411a938b | 1438 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1439 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1440 | list_del_init(&p->mnt_child); |
c003b26f | 1441 | } |
88b368f2 | 1442 | |
c003b26f | 1443 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1444 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1445 | propagate_umount(&tmp_list); |
a05964f3 | 1446 | |
c003b26f | 1447 | while (!list_empty(&tmp_list)) { |
d2921684 | 1448 | struct mnt_namespace *ns; |
ce07d891 | 1449 | bool disconnect; |
c003b26f | 1450 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1451 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1452 | list_del_init(&p->mnt_list); |
d2921684 EB |
1453 | ns = p->mnt_ns; |
1454 | if (ns) { | |
1455 | ns->mounts--; | |
1456 | __touch_mnt_namespace(ns); | |
1457 | } | |
143c8c91 | 1458 | p->mnt_ns = NULL; |
e819f152 | 1459 | if (how & UMOUNT_SYNC) |
48a066e7 | 1460 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1461 | |
f2d0a123 | 1462 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1463 | |
1464 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1465 | disconnect ? &unmounted : NULL); | |
676da58d | 1466 | if (mnt_has_parent(p)) { |
81b6b061 | 1467 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1468 | if (!disconnect) { |
1469 | /* Don't forget about p */ | |
1470 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1471 | } else { | |
1472 | umount_mnt(p); | |
1473 | } | |
7c4b93d8 | 1474 | } |
0f0afb1d | 1475 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1476 | } |
1477 | } | |
1478 | ||
b54b9be7 | 1479 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1480 | |
8d0347f6 DH |
1481 | static int do_umount_root(struct super_block *sb) |
1482 | { | |
1483 | int ret = 0; | |
1484 | ||
1485 | down_write(&sb->s_umount); | |
1486 | if (!sb_rdonly(sb)) { | |
1487 | struct fs_context *fc; | |
1488 | ||
1489 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1490 | SB_RDONLY); | |
1491 | if (IS_ERR(fc)) { | |
1492 | ret = PTR_ERR(fc); | |
1493 | } else { | |
1494 | ret = parse_monolithic_mount_data(fc, NULL); | |
1495 | if (!ret) | |
1496 | ret = reconfigure_super(fc); | |
1497 | put_fs_context(fc); | |
1498 | } | |
1499 | } | |
1500 | up_write(&sb->s_umount); | |
1501 | return ret; | |
1502 | } | |
1503 | ||
1ab59738 | 1504 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1505 | { |
1ab59738 | 1506 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1507 | int retval; |
1508 | ||
1ab59738 | 1509 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1510 | if (retval) |
1511 | return retval; | |
1512 | ||
1513 | /* | |
1514 | * Allow userspace to request a mountpoint be expired rather than | |
1515 | * unmounting unconditionally. Unmount only happens if: | |
1516 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1517 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1518 | */ | |
1519 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1520 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1521 | flags & (MNT_FORCE | MNT_DETACH)) |
1522 | return -EINVAL; | |
1523 | ||
b3e19d92 NP |
1524 | /* |
1525 | * probably don't strictly need the lock here if we examined | |
1526 | * all race cases, but it's a slowpath. | |
1527 | */ | |
719ea2fb | 1528 | lock_mount_hash(); |
83adc753 | 1529 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1530 | unlock_mount_hash(); |
1da177e4 | 1531 | return -EBUSY; |
b3e19d92 | 1532 | } |
719ea2fb | 1533 | unlock_mount_hash(); |
1da177e4 | 1534 | |
863d684f | 1535 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1536 | return -EAGAIN; |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * If we may have to abort operations to get out of this | |
1541 | * mount, and they will themselves hold resources we must | |
1542 | * allow the fs to do things. In the Unix tradition of | |
1543 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1544 | * might fail to complete on the first run through as other tasks | |
1545 | * must return, and the like. Thats for the mount program to worry | |
1546 | * about for the moment. | |
1547 | */ | |
1548 | ||
42faad99 | 1549 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1550 | sb->s_op->umount_begin(sb); |
42faad99 | 1551 | } |
1da177e4 LT |
1552 | |
1553 | /* | |
1554 | * No sense to grab the lock for this test, but test itself looks | |
1555 | * somewhat bogus. Suggestions for better replacement? | |
1556 | * Ho-hum... In principle, we might treat that as umount + switch | |
1557 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1558 | * Actually it makes sense, especially if rootfs would contain a | |
1559 | * /reboot - static binary that would close all descriptors and | |
1560 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1561 | */ | |
1ab59738 | 1562 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1563 | /* |
1564 | * Special case for "unmounting" root ... | |
1565 | * we just try to remount it readonly. | |
1566 | */ | |
bc6155d1 | 1567 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1568 | return -EPERM; |
8d0347f6 | 1569 | return do_umount_root(sb); |
1da177e4 LT |
1570 | } |
1571 | ||
97216be0 | 1572 | namespace_lock(); |
719ea2fb | 1573 | lock_mount_hash(); |
1da177e4 | 1574 | |
25d202ed EB |
1575 | /* Recheck MNT_LOCKED with the locks held */ |
1576 | retval = -EINVAL; | |
1577 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1578 | goto out; | |
1579 | ||
1580 | event++; | |
48a066e7 | 1581 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1582 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1583 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1584 | retval = 0; |
48a066e7 AV |
1585 | } else { |
1586 | shrink_submounts(mnt); | |
1587 | retval = -EBUSY; | |
1588 | if (!propagate_mount_busy(mnt, 2)) { | |
1589 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1590 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1591 | retval = 0; |
1592 | } | |
1da177e4 | 1593 | } |
25d202ed | 1594 | out: |
719ea2fb | 1595 | unlock_mount_hash(); |
e3197d83 | 1596 | namespace_unlock(); |
1da177e4 LT |
1597 | return retval; |
1598 | } | |
1599 | ||
80b5dce8 EB |
1600 | /* |
1601 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1602 | * | |
1603 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1604 | * to an existing mountpoint, and wind up leaking the mount. | |
1605 | * detach_mounts allows lazily unmounting those mounts instead of | |
1606 | * leaking them. | |
1607 | * | |
1608 | * The caller may hold dentry->d_inode->i_mutex. | |
1609 | */ | |
1610 | void __detach_mounts(struct dentry *dentry) | |
1611 | { | |
1612 | struct mountpoint *mp; | |
1613 | struct mount *mnt; | |
1614 | ||
1615 | namespace_lock(); | |
3895dbf8 | 1616 | lock_mount_hash(); |
80b5dce8 | 1617 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1618 | if (!mp) |
80b5dce8 EB |
1619 | goto out_unlock; |
1620 | ||
e06b933e | 1621 | event++; |
80b5dce8 EB |
1622 | while (!hlist_empty(&mp->m_list)) { |
1623 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1624 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1625 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1626 | umount_mnt(mnt); | |
ce07d891 | 1627 | } |
e0c9c0af | 1628 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1629 | } |
80b5dce8 EB |
1630 | put_mountpoint(mp); |
1631 | out_unlock: | |
3895dbf8 | 1632 | unlock_mount_hash(); |
80b5dce8 EB |
1633 | namespace_unlock(); |
1634 | } | |
1635 | ||
dd111b31 | 1636 | /* |
9b40bc90 AV |
1637 | * Is the caller allowed to modify his namespace? |
1638 | */ | |
1639 | static inline bool may_mount(void) | |
1640 | { | |
1641 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1642 | } | |
1643 | ||
9e8925b6 JL |
1644 | static inline bool may_mandlock(void) |
1645 | { | |
1646 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1647 | return false; | |
1648 | #endif | |
95ace754 | 1649 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1650 | } |
1651 | ||
1da177e4 LT |
1652 | /* |
1653 | * Now umount can handle mount points as well as block devices. | |
1654 | * This is important for filesystems which use unnamed block devices. | |
1655 | * | |
1656 | * We now support a flag for forced unmount like the other 'big iron' | |
1657 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1658 | */ | |
1659 | ||
3a18ef5c | 1660 | int ksys_umount(char __user *name, int flags) |
1da177e4 | 1661 | { |
2d8f3038 | 1662 | struct path path; |
900148dc | 1663 | struct mount *mnt; |
1da177e4 | 1664 | int retval; |
db1f05bb | 1665 | int lookup_flags = 0; |
1da177e4 | 1666 | |
db1f05bb MS |
1667 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1668 | return -EINVAL; | |
1669 | ||
9b40bc90 AV |
1670 | if (!may_mount()) |
1671 | return -EPERM; | |
1672 | ||
db1f05bb MS |
1673 | if (!(flags & UMOUNT_NOFOLLOW)) |
1674 | lookup_flags |= LOOKUP_FOLLOW; | |
1675 | ||
57d46577 RGB |
1676 | lookup_flags |= LOOKUP_NO_EVAL; |
1677 | ||
197df04c | 1678 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1679 | if (retval) |
1680 | goto out; | |
900148dc | 1681 | mnt = real_mount(path.mnt); |
1da177e4 | 1682 | retval = -EINVAL; |
2d8f3038 | 1683 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1684 | goto dput_and_out; |
143c8c91 | 1685 | if (!check_mnt(mnt)) |
1da177e4 | 1686 | goto dput_and_out; |
25d202ed | 1687 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
5ff9d8a6 | 1688 | goto dput_and_out; |
b2f5d4dc EB |
1689 | retval = -EPERM; |
1690 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1691 | goto dput_and_out; | |
1da177e4 | 1692 | |
900148dc | 1693 | retval = do_umount(mnt, flags); |
1da177e4 | 1694 | dput_and_out: |
429731b1 | 1695 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1696 | dput(path.dentry); |
900148dc | 1697 | mntput_no_expire(mnt); |
1da177e4 LT |
1698 | out: |
1699 | return retval; | |
1700 | } | |
1701 | ||
3a18ef5c DB |
1702 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1703 | { | |
1704 | return ksys_umount(name, flags); | |
1705 | } | |
1706 | ||
1da177e4 LT |
1707 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1708 | ||
1709 | /* | |
b58fed8b | 1710 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1711 | */ |
bdc480e3 | 1712 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1713 | { |
3a18ef5c | 1714 | return ksys_umount(name, 0); |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | #endif | |
1718 | ||
4ce5d2b1 | 1719 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1720 | { |
4ce5d2b1 | 1721 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1722 | return dentry->d_op == &ns_dentry_operations && |
1723 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1724 | } |
1725 | ||
58be2825 AV |
1726 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1727 | { | |
1728 | return container_of(ns, struct mnt_namespace, ns); | |
1729 | } | |
1730 | ||
4ce5d2b1 EB |
1731 | static bool mnt_ns_loop(struct dentry *dentry) |
1732 | { | |
1733 | /* Could bind mounting the mount namespace inode cause a | |
1734 | * mount namespace loop? | |
1735 | */ | |
1736 | struct mnt_namespace *mnt_ns; | |
1737 | if (!is_mnt_ns_file(dentry)) | |
1738 | return false; | |
1739 | ||
f77c8014 | 1740 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1741 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1742 | } | |
1743 | ||
87129cc0 | 1744 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1745 | int flag) |
1da177e4 | 1746 | { |
84d17192 | 1747 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1748 | |
4ce5d2b1 EB |
1749 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1750 | return ERR_PTR(-EINVAL); | |
1751 | ||
1752 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1753 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1754 | |
36341f64 | 1755 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1756 | if (IS_ERR(q)) |
1757 | return q; | |
1758 | ||
a73324da | 1759 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1760 | |
1761 | p = mnt; | |
6b41d536 | 1762 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1763 | struct mount *s; |
7ec02ef1 | 1764 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1765 | continue; |
1766 | ||
909b0a88 | 1767 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1768 | if (!(flag & CL_COPY_UNBINDABLE) && |
1769 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1770 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1771 | /* Both unbindable and locked. */ | |
1772 | q = ERR_PTR(-EPERM); | |
1773 | goto out; | |
1774 | } else { | |
1775 | s = skip_mnt_tree(s); | |
1776 | continue; | |
1777 | } | |
4ce5d2b1 EB |
1778 | } |
1779 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1780 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1781 | s = skip_mnt_tree(s); |
1782 | continue; | |
1783 | } | |
0714a533 AV |
1784 | while (p != s->mnt_parent) { |
1785 | p = p->mnt_parent; | |
1786 | q = q->mnt_parent; | |
1da177e4 | 1787 | } |
87129cc0 | 1788 | p = s; |
84d17192 | 1789 | parent = q; |
87129cc0 | 1790 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1791 | if (IS_ERR(q)) |
1792 | goto out; | |
719ea2fb | 1793 | lock_mount_hash(); |
1a4eeaf2 | 1794 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1795 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1796 | unlock_mount_hash(); |
1da177e4 LT |
1797 | } |
1798 | } | |
1799 | return res; | |
be34d1a3 | 1800 | out: |
1da177e4 | 1801 | if (res) { |
719ea2fb | 1802 | lock_mount_hash(); |
e819f152 | 1803 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1804 | unlock_mount_hash(); |
1da177e4 | 1805 | } |
be34d1a3 | 1806 | return q; |
1da177e4 LT |
1807 | } |
1808 | ||
be34d1a3 DH |
1809 | /* Caller should check returned pointer for errors */ |
1810 | ||
ca71cf71 | 1811 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1812 | { |
cb338d06 | 1813 | struct mount *tree; |
97216be0 | 1814 | namespace_lock(); |
cd4a4017 EB |
1815 | if (!check_mnt(real_mount(path->mnt))) |
1816 | tree = ERR_PTR(-EINVAL); | |
1817 | else | |
1818 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1819 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1820 | namespace_unlock(); |
be34d1a3 | 1821 | if (IS_ERR(tree)) |
52e220d3 | 1822 | return ERR_CAST(tree); |
be34d1a3 | 1823 | return &tree->mnt; |
8aec0809 AV |
1824 | } |
1825 | ||
a07b2000 AV |
1826 | static void free_mnt_ns(struct mnt_namespace *); |
1827 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1828 | ||
1829 | void dissolve_on_fput(struct vfsmount *mnt) | |
1830 | { | |
1831 | struct mnt_namespace *ns; | |
1832 | namespace_lock(); | |
1833 | lock_mount_hash(); | |
1834 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1835 | if (ns) { |
1836 | if (is_anon_ns(ns)) | |
1837 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1838 | else | |
1839 | ns = NULL; | |
1840 | } | |
a07b2000 AV |
1841 | unlock_mount_hash(); |
1842 | namespace_unlock(); | |
44dfd84a DH |
1843 | if (ns) |
1844 | free_mnt_ns(ns); | |
a07b2000 AV |
1845 | } |
1846 | ||
8aec0809 AV |
1847 | void drop_collected_mounts(struct vfsmount *mnt) |
1848 | { | |
97216be0 | 1849 | namespace_lock(); |
719ea2fb | 1850 | lock_mount_hash(); |
9c8e0a1b | 1851 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1852 | unlock_mount_hash(); |
3ab6abee | 1853 | namespace_unlock(); |
8aec0809 AV |
1854 | } |
1855 | ||
c771d683 MS |
1856 | /** |
1857 | * clone_private_mount - create a private clone of a path | |
1858 | * | |
1859 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1860 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1861 | * to the originating mount won't be propagated into this). | |
1862 | * | |
1863 | * Release with mntput(). | |
1864 | */ | |
ca71cf71 | 1865 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1866 | { |
1867 | struct mount *old_mnt = real_mount(path->mnt); | |
1868 | struct mount *new_mnt; | |
1869 | ||
1870 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1871 | return ERR_PTR(-EINVAL); | |
1872 | ||
c771d683 | 1873 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1874 | if (IS_ERR(new_mnt)) |
1875 | return ERR_CAST(new_mnt); | |
1876 | ||
1877 | return &new_mnt->mnt; | |
1878 | } | |
1879 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1880 | ||
1f707137 AV |
1881 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1882 | struct vfsmount *root) | |
1883 | { | |
1a4eeaf2 | 1884 | struct mount *mnt; |
1f707137 AV |
1885 | int res = f(root, arg); |
1886 | if (res) | |
1887 | return res; | |
1a4eeaf2 AV |
1888 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1889 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1890 | if (res) |
1891 | return res; | |
1892 | } | |
1893 | return 0; | |
1894 | } | |
1895 | ||
3bd045cc AV |
1896 | static void lock_mnt_tree(struct mount *mnt) |
1897 | { | |
1898 | struct mount *p; | |
1899 | ||
1900 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1901 | int flags = p->mnt.mnt_flags; | |
1902 | /* Don't allow unprivileged users to change mount flags */ | |
1903 | flags |= MNT_LOCK_ATIME; | |
1904 | ||
1905 | if (flags & MNT_READONLY) | |
1906 | flags |= MNT_LOCK_READONLY; | |
1907 | ||
1908 | if (flags & MNT_NODEV) | |
1909 | flags |= MNT_LOCK_NODEV; | |
1910 | ||
1911 | if (flags & MNT_NOSUID) | |
1912 | flags |= MNT_LOCK_NOSUID; | |
1913 | ||
1914 | if (flags & MNT_NOEXEC) | |
1915 | flags |= MNT_LOCK_NOEXEC; | |
1916 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
1917 | if (list_empty(&p->mnt_expire)) | |
1918 | flags |= MNT_LOCKED; | |
1919 | p->mnt.mnt_flags = flags; | |
1920 | } | |
1921 | } | |
1922 | ||
4b8b21f4 | 1923 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1924 | { |
315fc83e | 1925 | struct mount *p; |
719f5d7f | 1926 | |
909b0a88 | 1927 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1928 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1929 | mnt_release_group_id(p); |
719f5d7f MS |
1930 | } |
1931 | } | |
1932 | ||
4b8b21f4 | 1933 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1934 | { |
315fc83e | 1935 | struct mount *p; |
719f5d7f | 1936 | |
909b0a88 | 1937 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1938 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1939 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1940 | if (err) { |
4b8b21f4 | 1941 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1942 | return err; |
1943 | } | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | return 0; | |
1948 | } | |
1949 | ||
d2921684 EB |
1950 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1951 | { | |
1952 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1953 | unsigned int mounts = 0, old, pending, sum; | |
1954 | struct mount *p; | |
1955 | ||
1956 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1957 | mounts++; | |
1958 | ||
1959 | old = ns->mounts; | |
1960 | pending = ns->pending_mounts; | |
1961 | sum = old + pending; | |
1962 | if ((old > sum) || | |
1963 | (pending > sum) || | |
1964 | (max < sum) || | |
1965 | (mounts > (max - sum))) | |
1966 | return -ENOSPC; | |
1967 | ||
1968 | ns->pending_mounts = pending + mounts; | |
1969 | return 0; | |
1970 | } | |
1971 | ||
b90fa9ae RP |
1972 | /* |
1973 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1974 | * @nd : place the mount tree @source_mnt is attached |
1975 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1976 | * store the parent mount and mountpoint dentry. | |
1977 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1978 | * |
1979 | * NOTE: in the table below explains the semantics when a source mount | |
1980 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1981 | * --------------------------------------------------------------------------- |
1982 | * | BIND MOUNT OPERATION | | |
1983 | * |************************************************************************** | |
1984 | * | source-->| shared | private | slave | unbindable | | |
1985 | * | dest | | | | | | |
1986 | * | | | | | | | | |
1987 | * | v | | | | | | |
1988 | * |************************************************************************** | |
1989 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1990 | * | | | | | | | |
1991 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1992 | * *************************************************************************** | |
b90fa9ae RP |
1993 | * A bind operation clones the source mount and mounts the clone on the |
1994 | * destination mount. | |
1995 | * | |
1996 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1997 | * tree of the destination mount and the cloned mount is added to | |
1998 | * the peer group of the source mount. | |
1999 | * (+) the cloned mount is created under the destination mount and is marked | |
2000 | * as shared. The cloned mount is added to the peer group of the source | |
2001 | * mount. | |
5afe0022 RP |
2002 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2003 | * of the destination mount and the cloned mount is made slave | |
2004 | * of the same master as that of the source mount. The cloned mount | |
2005 | * is marked as 'shared and slave'. | |
2006 | * (*) the cloned mount is made a slave of the same master as that of the | |
2007 | * source mount. | |
2008 | * | |
9676f0c6 RP |
2009 | * --------------------------------------------------------------------------- |
2010 | * | MOVE MOUNT OPERATION | | |
2011 | * |************************************************************************** | |
2012 | * | source-->| shared | private | slave | unbindable | | |
2013 | * | dest | | | | | | |
2014 | * | | | | | | | | |
2015 | * | v | | | | | | |
2016 | * |************************************************************************** | |
2017 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2018 | * | | | | | | | |
2019 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2020 | * *************************************************************************** | |
5afe0022 RP |
2021 | * |
2022 | * (+) the mount is moved to the destination. And is then propagated to | |
2023 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2024 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2025 | * (+++) the mount is moved to the destination and is then propagated to |
2026 | * all the mounts belonging to the destination mount's propagation tree. | |
2027 | * the mount is marked as 'shared and slave'. | |
2028 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2029 | * |
2030 | * if the source mount is a tree, the operations explained above is | |
2031 | * applied to each mount in the tree. | |
2032 | * Must be called without spinlocks held, since this function can sleep | |
2033 | * in allocations. | |
2034 | */ | |
0fb54e50 | 2035 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2036 | struct mount *dest_mnt, |
2037 | struct mountpoint *dest_mp, | |
2763d119 | 2038 | bool moving) |
b90fa9ae | 2039 | { |
3bd045cc | 2040 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2041 | HLIST_HEAD(tree_list); |
d2921684 | 2042 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2043 | struct mountpoint *smp; |
315fc83e | 2044 | struct mount *child, *p; |
38129a13 | 2045 | struct hlist_node *n; |
719f5d7f | 2046 | int err; |
b90fa9ae | 2047 | |
1064f874 EB |
2048 | /* Preallocate a mountpoint in case the new mounts need |
2049 | * to be tucked under other mounts. | |
2050 | */ | |
2051 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2052 | if (IS_ERR(smp)) | |
2053 | return PTR_ERR(smp); | |
2054 | ||
d2921684 | 2055 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2056 | if (!moving) { |
d2921684 EB |
2057 | err = count_mounts(ns, source_mnt); |
2058 | if (err) | |
2059 | goto out; | |
2060 | } | |
2061 | ||
fc7be130 | 2062 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2063 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2064 | if (err) |
2065 | goto out; | |
0b1b901b | 2066 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2067 | lock_mount_hash(); |
0b1b901b AV |
2068 | if (err) |
2069 | goto out_cleanup_ids; | |
909b0a88 | 2070 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2071 | set_mnt_shared(p); |
0b1b901b AV |
2072 | } else { |
2073 | lock_mount_hash(); | |
b90fa9ae | 2074 | } |
2763d119 AV |
2075 | if (moving) { |
2076 | unhash_mnt(source_mnt); | |
84d17192 | 2077 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2078 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2079 | } else { |
44dfd84a DH |
2080 | if (source_mnt->mnt_ns) { |
2081 | /* move from anon - the caller will destroy */ | |
2082 | list_del_init(&source_mnt->mnt_ns->list); | |
2083 | } | |
84d17192 | 2084 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2085 | commit_tree(source_mnt); |
21444403 | 2086 | } |
b90fa9ae | 2087 | |
38129a13 | 2088 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2089 | struct mount *q; |
38129a13 | 2090 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2091 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2092 | child->mnt_mountpoint); | |
2093 | if (q) | |
2094 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2095 | /* Notice when we are propagating across user namespaces */ |
2096 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2097 | lock_mnt_tree(child); | |
d728cf79 | 2098 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2099 | commit_tree(child); |
b90fa9ae | 2100 | } |
1064f874 | 2101 | put_mountpoint(smp); |
719ea2fb | 2102 | unlock_mount_hash(); |
99b7db7b | 2103 | |
b90fa9ae | 2104 | return 0; |
719f5d7f MS |
2105 | |
2106 | out_cleanup_ids: | |
f2ebb3a9 AV |
2107 | while (!hlist_empty(&tree_list)) { |
2108 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2109 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2110 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2111 | } |
2112 | unlock_mount_hash(); | |
0b1b901b | 2113 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2114 | out: |
d2921684 | 2115 | ns->pending_mounts = 0; |
1064f874 EB |
2116 | |
2117 | read_seqlock_excl(&mount_lock); | |
2118 | put_mountpoint(smp); | |
2119 | read_sequnlock_excl(&mount_lock); | |
2120 | ||
719f5d7f | 2121 | return err; |
b90fa9ae RP |
2122 | } |
2123 | ||
84d17192 | 2124 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2125 | { |
2126 | struct vfsmount *mnt; | |
84d17192 | 2127 | struct dentry *dentry = path->dentry; |
b12cea91 | 2128 | retry: |
5955102c | 2129 | inode_lock(dentry->d_inode); |
84d17192 | 2130 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2131 | inode_unlock(dentry->d_inode); |
84d17192 | 2132 | return ERR_PTR(-ENOENT); |
b12cea91 | 2133 | } |
97216be0 | 2134 | namespace_lock(); |
b12cea91 | 2135 | mnt = lookup_mnt(path); |
84d17192 | 2136 | if (likely(!mnt)) { |
3895dbf8 | 2137 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2138 | if (IS_ERR(mp)) { |
97216be0 | 2139 | namespace_unlock(); |
5955102c | 2140 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2141 | return mp; |
2142 | } | |
2143 | return mp; | |
2144 | } | |
97216be0 | 2145 | namespace_unlock(); |
5955102c | 2146 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2147 | path_put(path); |
2148 | path->mnt = mnt; | |
84d17192 | 2149 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2150 | goto retry; |
2151 | } | |
2152 | ||
84d17192 | 2153 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2154 | { |
84d17192 | 2155 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2156 | |
2157 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2158 | put_mountpoint(where); |
3895dbf8 EB |
2159 | read_sequnlock_excl(&mount_lock); |
2160 | ||
328e6d90 | 2161 | namespace_unlock(); |
5955102c | 2162 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2163 | } |
2164 | ||
84d17192 | 2165 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2166 | { |
e462ec50 | 2167 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2168 | return -EINVAL; |
2169 | ||
e36cb0b8 DH |
2170 | if (d_is_dir(mp->m_dentry) != |
2171 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2172 | return -ENOTDIR; |
2173 | ||
2763d119 | 2174 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2175 | } |
2176 | ||
7a2e8a8f VA |
2177 | /* |
2178 | * Sanity check the flags to change_mnt_propagation. | |
2179 | */ | |
2180 | ||
e462ec50 | 2181 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2182 | { |
e462ec50 | 2183 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2184 | |
2185 | /* Fail if any non-propagation flags are set */ | |
2186 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2187 | return 0; | |
2188 | /* Only one propagation flag should be set */ | |
2189 | if (!is_power_of_2(type)) | |
2190 | return 0; | |
2191 | return type; | |
2192 | } | |
2193 | ||
07b20889 RP |
2194 | /* |
2195 | * recursively change the type of the mountpoint. | |
2196 | */ | |
e462ec50 | 2197 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2198 | { |
315fc83e | 2199 | struct mount *m; |
4b8b21f4 | 2200 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2201 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2202 | int type; |
719f5d7f | 2203 | int err = 0; |
07b20889 | 2204 | |
2d92ab3c | 2205 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2206 | return -EINVAL; |
2207 | ||
e462ec50 | 2208 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2209 | if (!type) |
2210 | return -EINVAL; | |
2211 | ||
97216be0 | 2212 | namespace_lock(); |
719f5d7f MS |
2213 | if (type == MS_SHARED) { |
2214 | err = invent_group_ids(mnt, recurse); | |
2215 | if (err) | |
2216 | goto out_unlock; | |
2217 | } | |
2218 | ||
719ea2fb | 2219 | lock_mount_hash(); |
909b0a88 | 2220 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2221 | change_mnt_propagation(m, type); |
719ea2fb | 2222 | unlock_mount_hash(); |
719f5d7f MS |
2223 | |
2224 | out_unlock: | |
97216be0 | 2225 | namespace_unlock(); |
719f5d7f | 2226 | return err; |
07b20889 RP |
2227 | } |
2228 | ||
5ff9d8a6 EB |
2229 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2230 | { | |
2231 | struct mount *child; | |
2232 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2233 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2234 | continue; | |
2235 | ||
2236 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2237 | return true; | |
2238 | } | |
2239 | return false; | |
2240 | } | |
2241 | ||
a07b2000 AV |
2242 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2243 | { | |
2244 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2245 | ||
2246 | if (IS_MNT_UNBINDABLE(old)) | |
2247 | return mnt; | |
2248 | ||
2249 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2250 | return mnt; | |
2251 | ||
2252 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2253 | return mnt; | |
2254 | ||
2255 | if (recurse) | |
2256 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2257 | else | |
2258 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2259 | ||
2260 | if (!IS_ERR(mnt)) | |
2261 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2262 | ||
2263 | return mnt; | |
2264 | } | |
2265 | ||
1da177e4 LT |
2266 | /* |
2267 | * do loopback mount. | |
2268 | */ | |
808d4e3c | 2269 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2270 | int recurse) |
1da177e4 | 2271 | { |
2d92ab3c | 2272 | struct path old_path; |
a07b2000 | 2273 | struct mount *mnt = NULL, *parent; |
84d17192 | 2274 | struct mountpoint *mp; |
57eccb83 | 2275 | int err; |
1da177e4 LT |
2276 | if (!old_name || !*old_name) |
2277 | return -EINVAL; | |
815d405c | 2278 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2279 | if (err) |
2280 | return err; | |
2281 | ||
8823c079 | 2282 | err = -EINVAL; |
4ce5d2b1 | 2283 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2284 | goto out; |
8823c079 | 2285 | |
84d17192 | 2286 | mp = lock_mount(path); |
a07b2000 AV |
2287 | if (IS_ERR(mp)) { |
2288 | err = PTR_ERR(mp); | |
b12cea91 | 2289 | goto out; |
a07b2000 | 2290 | } |
b12cea91 | 2291 | |
84d17192 | 2292 | parent = real_mount(path->mnt); |
e149ed2b AV |
2293 | if (!check_mnt(parent)) |
2294 | goto out2; | |
2295 | ||
a07b2000 | 2296 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2297 | if (IS_ERR(mnt)) { |
2298 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2299 | goto out2; |
be34d1a3 | 2300 | } |
ccd48bc7 | 2301 | |
84d17192 | 2302 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2303 | if (err) { |
719ea2fb | 2304 | lock_mount_hash(); |
e819f152 | 2305 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2306 | unlock_mount_hash(); |
5b83d2c5 | 2307 | } |
b12cea91 | 2308 | out2: |
84d17192 | 2309 | unlock_mount(mp); |
ccd48bc7 | 2310 | out: |
2d92ab3c | 2311 | path_put(&old_path); |
1da177e4 LT |
2312 | return err; |
2313 | } | |
2314 | ||
a07b2000 AV |
2315 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2316 | { | |
2317 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2318 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2319 | struct mount *mnt, *p; | |
2320 | struct file *file; | |
2321 | ||
2322 | if (IS_ERR(ns)) | |
2323 | return ERR_CAST(ns); | |
2324 | ||
2325 | namespace_lock(); | |
2326 | mnt = __do_loopback(path, recursive); | |
2327 | if (IS_ERR(mnt)) { | |
2328 | namespace_unlock(); | |
2329 | free_mnt_ns(ns); | |
2330 | return ERR_CAST(mnt); | |
2331 | } | |
2332 | ||
2333 | lock_mount_hash(); | |
2334 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2335 | p->mnt_ns = ns; | |
2336 | ns->mounts++; | |
2337 | } | |
2338 | ns->root = mnt; | |
2339 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2340 | mntget(&mnt->mnt); | |
2341 | unlock_mount_hash(); | |
2342 | namespace_unlock(); | |
2343 | ||
2344 | mntput(path->mnt); | |
2345 | path->mnt = &mnt->mnt; | |
2346 | file = dentry_open(path, O_PATH, current_cred()); | |
2347 | if (IS_ERR(file)) | |
2348 | dissolve_on_fput(path->mnt); | |
2349 | else | |
2350 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2351 | return file; | |
2352 | } | |
2353 | ||
2354 | SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags) | |
2355 | { | |
2356 | struct file *file; | |
2357 | struct path path; | |
2358 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2359 | bool detached = flags & OPEN_TREE_CLONE; | |
2360 | int error; | |
2361 | int fd; | |
2362 | ||
2363 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2364 | ||
2365 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2366 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2367 | OPEN_TREE_CLOEXEC)) | |
2368 | return -EINVAL; | |
2369 | ||
2370 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2371 | return -EINVAL; | |
2372 | ||
2373 | if (flags & AT_NO_AUTOMOUNT) | |
2374 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2375 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2376 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2377 | if (flags & AT_EMPTY_PATH) | |
2378 | lookup_flags |= LOOKUP_EMPTY; | |
2379 | ||
2380 | if (detached && !may_mount()) | |
2381 | return -EPERM; | |
2382 | ||
2383 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2384 | if (fd < 0) | |
2385 | return fd; | |
2386 | ||
2387 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2388 | if (unlikely(error)) { | |
2389 | file = ERR_PTR(error); | |
2390 | } else { | |
2391 | if (detached) | |
2392 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2393 | else | |
2394 | file = dentry_open(&path, O_PATH, current_cred()); | |
2395 | path_put(&path); | |
2396 | } | |
2397 | if (IS_ERR(file)) { | |
2398 | put_unused_fd(fd); | |
2399 | return PTR_ERR(file); | |
2400 | } | |
2401 | fd_install(fd, file); | |
2402 | return fd; | |
2403 | } | |
2404 | ||
43f5e655 DH |
2405 | /* |
2406 | * Don't allow locked mount flags to be cleared. | |
2407 | * | |
2408 | * No locks need to be held here while testing the various MNT_LOCK | |
2409 | * flags because those flags can never be cleared once they are set. | |
2410 | */ | |
2411 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2412 | { |
43f5e655 DH |
2413 | unsigned int fl = mnt->mnt.mnt_flags; |
2414 | ||
2415 | if ((fl & MNT_LOCK_READONLY) && | |
2416 | !(mnt_flags & MNT_READONLY)) | |
2417 | return false; | |
2418 | ||
2419 | if ((fl & MNT_LOCK_NODEV) && | |
2420 | !(mnt_flags & MNT_NODEV)) | |
2421 | return false; | |
2422 | ||
2423 | if ((fl & MNT_LOCK_NOSUID) && | |
2424 | !(mnt_flags & MNT_NOSUID)) | |
2425 | return false; | |
2426 | ||
2427 | if ((fl & MNT_LOCK_NOEXEC) && | |
2428 | !(mnt_flags & MNT_NOEXEC)) | |
2429 | return false; | |
2430 | ||
2431 | if ((fl & MNT_LOCK_ATIME) && | |
2432 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2433 | return false; | |
2e4b7fcd | 2434 | |
43f5e655 DH |
2435 | return true; |
2436 | } | |
2437 | ||
2438 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2439 | { |
43f5e655 | 2440 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2441 | |
43f5e655 | 2442 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2443 | return 0; |
2444 | ||
2445 | if (readonly_request) | |
43f5e655 DH |
2446 | return mnt_make_readonly(mnt); |
2447 | ||
2448 | return __mnt_unmake_readonly(mnt); | |
2449 | } | |
2450 | ||
2451 | /* | |
2452 | * Update the user-settable attributes on a mount. The caller must hold | |
2453 | * sb->s_umount for writing. | |
2454 | */ | |
2455 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) | |
2456 | { | |
2457 | lock_mount_hash(); | |
2458 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; | |
2459 | mnt->mnt.mnt_flags = mnt_flags; | |
2460 | touch_mnt_namespace(mnt->mnt_ns); | |
2461 | unlock_mount_hash(); | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2466 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2467 | * to mount(2). | |
2468 | */ | |
2469 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2470 | { | |
2471 | struct super_block *sb = path->mnt->mnt_sb; | |
2472 | struct mount *mnt = real_mount(path->mnt); | |
2473 | int ret; | |
2474 | ||
2475 | if (!check_mnt(mnt)) | |
2476 | return -EINVAL; | |
2477 | ||
2478 | if (path->dentry != mnt->mnt.mnt_root) | |
2479 | return -EINVAL; | |
2480 | ||
2481 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2482 | return -EPERM; | |
2483 | ||
2484 | down_write(&sb->s_umount); | |
2485 | ret = change_mount_ro_state(mnt, mnt_flags); | |
2486 | if (ret == 0) | |
2487 | set_mount_attributes(mnt, mnt_flags); | |
2488 | up_write(&sb->s_umount); | |
2489 | return ret; | |
2e4b7fcd DH |
2490 | } |
2491 | ||
1da177e4 LT |
2492 | /* |
2493 | * change filesystem flags. dir should be a physical root of filesystem. | |
2494 | * If you've mounted a non-root directory somewhere and want to do remount | |
2495 | * on it - tough luck. | |
2496 | */ | |
e462ec50 DH |
2497 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2498 | int mnt_flags, void *data) | |
1da177e4 LT |
2499 | { |
2500 | int err; | |
2d92ab3c | 2501 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2502 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2503 | struct fs_context *fc; |
1da177e4 | 2504 | |
143c8c91 | 2505 | if (!check_mnt(mnt)) |
1da177e4 LT |
2506 | return -EINVAL; |
2507 | ||
2d92ab3c | 2508 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2509 | return -EINVAL; |
2510 | ||
43f5e655 | 2511 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2512 | return -EPERM; |
9566d674 | 2513 | |
8d0347f6 DH |
2514 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2515 | if (IS_ERR(fc)) | |
2516 | return PTR_ERR(fc); | |
ff36fe2c | 2517 | |
8d0347f6 DH |
2518 | err = parse_monolithic_mount_data(fc, data); |
2519 | if (!err) { | |
2520 | down_write(&sb->s_umount); | |
2521 | err = -EPERM; | |
2522 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2523 | err = reconfigure_super(fc); | |
2524 | if (!err) | |
2525 | set_mount_attributes(mnt, mnt_flags); | |
2526 | } | |
2527 | up_write(&sb->s_umount); | |
0e55a7cc | 2528 | } |
8d0347f6 | 2529 | put_fs_context(fc); |
1da177e4 LT |
2530 | return err; |
2531 | } | |
2532 | ||
cbbe362c | 2533 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2534 | { |
315fc83e | 2535 | struct mount *p; |
909b0a88 | 2536 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2537 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2538 | return 1; |
2539 | } | |
2540 | return 0; | |
2541 | } | |
2542 | ||
44dfd84a DH |
2543 | /* |
2544 | * Check that there aren't references to earlier/same mount namespaces in the | |
2545 | * specified subtree. Such references can act as pins for mount namespaces | |
2546 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2547 | * cycles to be made. | |
2548 | */ | |
2549 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2550 | { | |
2551 | struct mount *p; | |
2552 | bool ret = false; | |
2553 | ||
2554 | lock_mount_hash(); | |
2555 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2556 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2557 | goto out; | |
2558 | ||
2559 | ret = true; | |
2560 | out: | |
2561 | unlock_mount_hash(); | |
2562 | return ret; | |
2563 | } | |
2564 | ||
2db154b3 | 2565 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2566 | { |
44dfd84a | 2567 | struct mnt_namespace *ns; |
676da58d | 2568 | struct mount *p; |
0fb54e50 | 2569 | struct mount *old; |
2763d119 AV |
2570 | struct mount *parent; |
2571 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2572 | int err; |
44dfd84a | 2573 | bool attached; |
1da177e4 | 2574 | |
2db154b3 | 2575 | mp = lock_mount(new_path); |
84d17192 | 2576 | if (IS_ERR(mp)) |
2db154b3 | 2577 | return PTR_ERR(mp); |
cc53ce53 | 2578 | |
2db154b3 DH |
2579 | old = real_mount(old_path->mnt); |
2580 | p = real_mount(new_path->mnt); | |
2763d119 | 2581 | parent = old->mnt_parent; |
44dfd84a | 2582 | attached = mnt_has_parent(old); |
2763d119 | 2583 | old_mp = old->mnt_mp; |
44dfd84a | 2584 | ns = old->mnt_ns; |
143c8c91 | 2585 | |
1da177e4 | 2586 | err = -EINVAL; |
44dfd84a DH |
2587 | /* The mountpoint must be in our namespace. */ |
2588 | if (!check_mnt(p)) | |
2db154b3 | 2589 | goto out; |
1da177e4 | 2590 | |
570d7a98 EB |
2591 | /* The thing moved must be mounted... */ |
2592 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2593 | goto out; |
2594 | ||
570d7a98 EB |
2595 | /* ... and either ours or the root of anon namespace */ |
2596 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2597 | goto out; |
5ff9d8a6 | 2598 | |
2db154b3 DH |
2599 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2600 | goto out; | |
1da177e4 | 2601 | |
2db154b3 DH |
2602 | if (old_path->dentry != old_path->mnt->mnt_root) |
2603 | goto out; | |
1da177e4 | 2604 | |
2db154b3 DH |
2605 | if (d_is_dir(new_path->dentry) != |
2606 | d_is_dir(old_path->dentry)) | |
2607 | goto out; | |
21444403 RP |
2608 | /* |
2609 | * Don't move a mount residing in a shared parent. | |
2610 | */ | |
2763d119 | 2611 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2612 | goto out; |
9676f0c6 RP |
2613 | /* |
2614 | * Don't move a mount tree containing unbindable mounts to a destination | |
2615 | * mount which is shared. | |
2616 | */ | |
fc7be130 | 2617 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2618 | goto out; |
1da177e4 | 2619 | err = -ELOOP; |
44dfd84a DH |
2620 | if (!check_for_nsfs_mounts(old)) |
2621 | goto out; | |
fc7be130 | 2622 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2623 | if (p == old) |
2db154b3 | 2624 | goto out; |
1da177e4 | 2625 | |
2db154b3 | 2626 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2627 | attached); |
4ac91378 | 2628 | if (err) |
2db154b3 | 2629 | goto out; |
1da177e4 LT |
2630 | |
2631 | /* if the mount is moved, it should no longer be expire | |
2632 | * automatically */ | |
6776db3d | 2633 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2634 | if (attached) |
2635 | put_mountpoint(old_mp); | |
1da177e4 | 2636 | out: |
2db154b3 | 2637 | unlock_mount(mp); |
44dfd84a | 2638 | if (!err) { |
2763d119 AV |
2639 | if (attached) |
2640 | mntput_no_expire(parent); | |
2641 | else | |
44dfd84a DH |
2642 | free_mnt_ns(ns); |
2643 | } | |
2db154b3 DH |
2644 | return err; |
2645 | } | |
2646 | ||
2647 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2648 | { | |
2649 | struct path old_path; | |
2650 | int err; | |
2651 | ||
2652 | if (!old_name || !*old_name) | |
2653 | return -EINVAL; | |
2654 | ||
2655 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2656 | if (err) | |
2657 | return err; | |
2658 | ||
2659 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2660 | path_put(&old_path); |
1da177e4 LT |
2661 | return err; |
2662 | } | |
2663 | ||
9d412a43 AV |
2664 | /* |
2665 | * add a mount into a namespace's mount tree | |
2666 | */ | |
95bc5f25 | 2667 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2668 | { |
84d17192 AV |
2669 | struct mountpoint *mp; |
2670 | struct mount *parent; | |
9d412a43 AV |
2671 | int err; |
2672 | ||
f2ebb3a9 | 2673 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2674 | |
84d17192 AV |
2675 | mp = lock_mount(path); |
2676 | if (IS_ERR(mp)) | |
2677 | return PTR_ERR(mp); | |
9d412a43 | 2678 | |
84d17192 | 2679 | parent = real_mount(path->mnt); |
9d412a43 | 2680 | err = -EINVAL; |
84d17192 | 2681 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2682 | /* that's acceptable only for automounts done in private ns */ |
2683 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2684 | goto unlock; | |
2685 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2686 | if (!parent->mnt_ns) |
156cacb1 AV |
2687 | goto unlock; |
2688 | } | |
9d412a43 AV |
2689 | |
2690 | /* Refuse the same filesystem on the same mount point */ | |
2691 | err = -EBUSY; | |
95bc5f25 | 2692 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2693 | path->mnt->mnt_root == path->dentry) |
2694 | goto unlock; | |
2695 | ||
2696 | err = -EINVAL; | |
e36cb0b8 | 2697 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2698 | goto unlock; |
2699 | ||
95bc5f25 | 2700 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2701 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2702 | |
2703 | unlock: | |
84d17192 | 2704 | unlock_mount(mp); |
9d412a43 AV |
2705 | return err; |
2706 | } | |
b1e75df4 | 2707 | |
132e4608 DH |
2708 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2709 | ||
2710 | /* | |
2711 | * Create a new mount using a superblock configuration and request it | |
2712 | * be added to the namespace tree. | |
2713 | */ | |
2714 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2715 | unsigned int mnt_flags) | |
2716 | { | |
2717 | struct vfsmount *mnt; | |
2718 | struct super_block *sb = fc->root->d_sb; | |
2719 | int error; | |
2720 | ||
c9ce29ed AV |
2721 | error = security_sb_kern_mount(sb); |
2722 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2723 | error = -EPERM; | |
2724 | ||
2725 | if (unlikely(error)) { | |
2726 | fc_drop_locked(fc); | |
2727 | return error; | |
132e4608 DH |
2728 | } |
2729 | ||
2730 | up_write(&sb->s_umount); | |
2731 | ||
2732 | mnt = vfs_create_mount(fc); | |
2733 | if (IS_ERR(mnt)) | |
2734 | return PTR_ERR(mnt); | |
2735 | ||
2736 | error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags); | |
2737 | if (error < 0) | |
2738 | mntput(mnt); | |
2739 | return error; | |
2740 | } | |
1b852bce | 2741 | |
1da177e4 LT |
2742 | /* |
2743 | * create a new mount for userspace and request it to be added into the | |
2744 | * namespace's tree | |
2745 | */ | |
e462ec50 | 2746 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2747 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2748 | { |
0c55cfc4 | 2749 | struct file_system_type *type; |
a0c9a8b8 AV |
2750 | struct fs_context *fc; |
2751 | const char *subtype = NULL; | |
2752 | int err = 0; | |
1da177e4 | 2753 | |
0c55cfc4 | 2754 | if (!fstype) |
1da177e4 LT |
2755 | return -EINVAL; |
2756 | ||
0c55cfc4 EB |
2757 | type = get_fs_type(fstype); |
2758 | if (!type) | |
2759 | return -ENODEV; | |
2760 | ||
a0c9a8b8 AV |
2761 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2762 | subtype = strchr(fstype, '.'); | |
2763 | if (subtype) { | |
2764 | subtype++; | |
2765 | if (!*subtype) { | |
2766 | put_filesystem(type); | |
2767 | return -EINVAL; | |
2768 | } | |
2769 | } else { | |
2770 | subtype = ""; | |
2771 | } | |
2772 | } | |
0c55cfc4 | 2773 | |
a0c9a8b8 | 2774 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2775 | put_filesystem(type); |
a0c9a8b8 AV |
2776 | if (IS_ERR(fc)) |
2777 | return PTR_ERR(fc); | |
2778 | ||
3e1aeb00 DH |
2779 | if (subtype) |
2780 | err = vfs_parse_fs_string(fc, "subtype", | |
2781 | subtype, strlen(subtype)); | |
2782 | if (!err && name) | |
2783 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2784 | if (!err) |
2785 | err = parse_monolithic_mount_data(fc, data); | |
2786 | if (!err) | |
2787 | err = vfs_get_tree(fc); | |
132e4608 DH |
2788 | if (!err) |
2789 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2790 | |
a0c9a8b8 | 2791 | put_fs_context(fc); |
15f9a3f3 | 2792 | return err; |
1da177e4 LT |
2793 | } |
2794 | ||
19a167af AV |
2795 | int finish_automount(struct vfsmount *m, struct path *path) |
2796 | { | |
6776db3d | 2797 | struct mount *mnt = real_mount(m); |
19a167af AV |
2798 | int err; |
2799 | /* The new mount record should have at least 2 refs to prevent it being | |
2800 | * expired before we get a chance to add it | |
2801 | */ | |
6776db3d | 2802 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2803 | |
2804 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2805 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2806 | err = -ELOOP; |
2807 | goto fail; | |
19a167af AV |
2808 | } |
2809 | ||
95bc5f25 | 2810 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2811 | if (!err) |
2812 | return 0; | |
2813 | fail: | |
2814 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2815 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2816 | namespace_lock(); |
6776db3d | 2817 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2818 | namespace_unlock(); |
19a167af | 2819 | } |
b1e75df4 AV |
2820 | mntput(m); |
2821 | mntput(m); | |
19a167af AV |
2822 | return err; |
2823 | } | |
2824 | ||
ea5b778a DH |
2825 | /** |
2826 | * mnt_set_expiry - Put a mount on an expiration list | |
2827 | * @mnt: The mount to list. | |
2828 | * @expiry_list: The list to add the mount to. | |
2829 | */ | |
2830 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2831 | { | |
97216be0 | 2832 | namespace_lock(); |
ea5b778a | 2833 | |
6776db3d | 2834 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2835 | |
97216be0 | 2836 | namespace_unlock(); |
ea5b778a DH |
2837 | } |
2838 | EXPORT_SYMBOL(mnt_set_expiry); | |
2839 | ||
1da177e4 LT |
2840 | /* |
2841 | * process a list of expirable mountpoints with the intent of discarding any | |
2842 | * mountpoints that aren't in use and haven't been touched since last we came | |
2843 | * here | |
2844 | */ | |
2845 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2846 | { | |
761d5c38 | 2847 | struct mount *mnt, *next; |
1da177e4 LT |
2848 | LIST_HEAD(graveyard); |
2849 | ||
2850 | if (list_empty(mounts)) | |
2851 | return; | |
2852 | ||
97216be0 | 2853 | namespace_lock(); |
719ea2fb | 2854 | lock_mount_hash(); |
1da177e4 LT |
2855 | |
2856 | /* extract from the expiration list every vfsmount that matches the | |
2857 | * following criteria: | |
2858 | * - only referenced by its parent vfsmount | |
2859 | * - still marked for expiry (marked on the last call here; marks are | |
2860 | * cleared by mntput()) | |
2861 | */ | |
6776db3d | 2862 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2863 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2864 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2865 | continue; |
6776db3d | 2866 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2867 | } |
bcc5c7d2 | 2868 | while (!list_empty(&graveyard)) { |
6776db3d | 2869 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2870 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2871 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2872 | } |
719ea2fb | 2873 | unlock_mount_hash(); |
3ab6abee | 2874 | namespace_unlock(); |
5528f911 TM |
2875 | } |
2876 | ||
2877 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2878 | ||
2879 | /* | |
2880 | * Ripoff of 'select_parent()' | |
2881 | * | |
2882 | * search the list of submounts for a given mountpoint, and move any | |
2883 | * shrinkable submounts to the 'graveyard' list. | |
2884 | */ | |
692afc31 | 2885 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2886 | { |
692afc31 | 2887 | struct mount *this_parent = parent; |
5528f911 TM |
2888 | struct list_head *next; |
2889 | int found = 0; | |
2890 | ||
2891 | repeat: | |
6b41d536 | 2892 | next = this_parent->mnt_mounts.next; |
5528f911 | 2893 | resume: |
6b41d536 | 2894 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2895 | struct list_head *tmp = next; |
6b41d536 | 2896 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2897 | |
2898 | next = tmp->next; | |
692afc31 | 2899 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2900 | continue; |
5528f911 TM |
2901 | /* |
2902 | * Descend a level if the d_mounts list is non-empty. | |
2903 | */ | |
6b41d536 | 2904 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2905 | this_parent = mnt; |
2906 | goto repeat; | |
2907 | } | |
1da177e4 | 2908 | |
1ab59738 | 2909 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2910 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2911 | found++; |
2912 | } | |
1da177e4 | 2913 | } |
5528f911 TM |
2914 | /* |
2915 | * All done at this level ... ascend and resume the search | |
2916 | */ | |
2917 | if (this_parent != parent) { | |
6b41d536 | 2918 | next = this_parent->mnt_child.next; |
0714a533 | 2919 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2920 | goto resume; |
2921 | } | |
2922 | return found; | |
2923 | } | |
2924 | ||
2925 | /* | |
2926 | * process a list of expirable mountpoints with the intent of discarding any | |
2927 | * submounts of a specific parent mountpoint | |
99b7db7b | 2928 | * |
48a066e7 | 2929 | * mount_lock must be held for write |
5528f911 | 2930 | */ |
b54b9be7 | 2931 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2932 | { |
2933 | LIST_HEAD(graveyard); | |
761d5c38 | 2934 | struct mount *m; |
5528f911 | 2935 | |
5528f911 | 2936 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2937 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2938 | while (!list_empty(&graveyard)) { |
761d5c38 | 2939 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2940 | mnt_expire); |
143c8c91 | 2941 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2942 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2943 | } |
2944 | } | |
1da177e4 LT |
2945 | } |
2946 | ||
1da177e4 LT |
2947 | /* |
2948 | * Some copy_from_user() implementations do not return the exact number of | |
2949 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2950 | * Note that this function differs from copy_from_user() in that it will oops | |
2951 | * on bad values of `to', rather than returning a short copy. | |
2952 | */ | |
b58fed8b RP |
2953 | static long exact_copy_from_user(void *to, const void __user * from, |
2954 | unsigned long n) | |
1da177e4 LT |
2955 | { |
2956 | char *t = to; | |
2957 | const char __user *f = from; | |
2958 | char c; | |
2959 | ||
96d4f267 | 2960 | if (!access_ok(from, n)) |
1da177e4 LT |
2961 | return n; |
2962 | ||
2963 | while (n) { | |
2964 | if (__get_user(c, f)) { | |
2965 | memset(t, 0, n); | |
2966 | break; | |
2967 | } | |
2968 | *t++ = c; | |
2969 | f++; | |
2970 | n--; | |
2971 | } | |
2972 | return n; | |
2973 | } | |
2974 | ||
b40ef869 | 2975 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2976 | { |
2977 | int i; | |
1da177e4 | 2978 | unsigned long size; |
b40ef869 | 2979 | char *copy; |
b58fed8b | 2980 | |
1da177e4 | 2981 | if (!data) |
b40ef869 | 2982 | return NULL; |
1da177e4 | 2983 | |
b40ef869 AV |
2984 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2985 | if (!copy) | |
2986 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2987 | |
2988 | /* We only care that *some* data at the address the user | |
2989 | * gave us is valid. Just in case, we'll zero | |
2990 | * the remainder of the page. | |
2991 | */ | |
2992 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2993 | size = TASK_SIZE - (unsigned long)data; | |
2994 | if (size > PAGE_SIZE) | |
2995 | size = PAGE_SIZE; | |
2996 | ||
b40ef869 | 2997 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2998 | if (!i) { |
b40ef869 AV |
2999 | kfree(copy); |
3000 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
3001 | } |
3002 | if (i != PAGE_SIZE) | |
b40ef869 AV |
3003 | memset(copy + i, 0, PAGE_SIZE - i); |
3004 | return copy; | |
1da177e4 LT |
3005 | } |
3006 | ||
b8850d1f | 3007 | char *copy_mount_string(const void __user *data) |
eca6f534 | 3008 | { |
fbdb4401 | 3009 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3010 | } |
3011 | ||
1da177e4 LT |
3012 | /* |
3013 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3014 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3015 | * | |
3016 | * data is a (void *) that can point to any structure up to | |
3017 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3018 | * information (or be NULL). | |
3019 | * | |
3020 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3021 | * When the flags word was introduced its top half was required | |
3022 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3023 | * Therefore, if this magic number is present, it carries no information | |
3024 | * and must be discarded. | |
3025 | */ | |
5e6123f3 | 3026 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 3027 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3028 | { |
2d92ab3c | 3029 | struct path path; |
e462ec50 | 3030 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 3031 | int retval = 0; |
1da177e4 LT |
3032 | |
3033 | /* Discard magic */ | |
3034 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3035 | flags &= ~MS_MGC_MSK; | |
3036 | ||
3037 | /* Basic sanity checks */ | |
1da177e4 LT |
3038 | if (data_page) |
3039 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3040 | ||
e462ec50 DH |
3041 | if (flags & MS_NOUSER) |
3042 | return -EINVAL; | |
3043 | ||
a27ab9f2 | 3044 | /* ... and get the mountpoint */ |
5e6123f3 | 3045 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
3046 | if (retval) |
3047 | return retval; | |
3048 | ||
3049 | retval = security_sb_mount(dev_name, &path, | |
3050 | type_page, flags, data_page); | |
0d5cadb8 AV |
3051 | if (!retval && !may_mount()) |
3052 | retval = -EPERM; | |
e462ec50 | 3053 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 3054 | retval = -EPERM; |
a27ab9f2 TH |
3055 | if (retval) |
3056 | goto dput_out; | |
3057 | ||
613cbe3d AK |
3058 | /* Default to relatime unless overriden */ |
3059 | if (!(flags & MS_NOATIME)) | |
3060 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3061 | |
1da177e4 LT |
3062 | /* Separate the per-mountpoint flags */ |
3063 | if (flags & MS_NOSUID) | |
3064 | mnt_flags |= MNT_NOSUID; | |
3065 | if (flags & MS_NODEV) | |
3066 | mnt_flags |= MNT_NODEV; | |
3067 | if (flags & MS_NOEXEC) | |
3068 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3069 | if (flags & MS_NOATIME) |
3070 | mnt_flags |= MNT_NOATIME; | |
3071 | if (flags & MS_NODIRATIME) | |
3072 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3073 | if (flags & MS_STRICTATIME) |
3074 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3075 | if (flags & MS_RDONLY) |
2e4b7fcd | 3076 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 3077 | |
ffbc6f0e EB |
3078 | /* The default atime for remount is preservation */ |
3079 | if ((flags & MS_REMOUNT) && | |
3080 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3081 | MS_STRICTATIME)) == 0)) { | |
3082 | mnt_flags &= ~MNT_ATIME_MASK; | |
3083 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
3084 | } | |
3085 | ||
e462ec50 DH |
3086 | sb_flags = flags & (SB_RDONLY | |
3087 | SB_SYNCHRONOUS | | |
3088 | SB_MANDLOCK | | |
3089 | SB_DIRSYNC | | |
3090 | SB_SILENT | | |
917086ff | 3091 | SB_POSIXACL | |
d7ee9469 | 3092 | SB_LAZYTIME | |
917086ff | 3093 | SB_I_VERSION); |
1da177e4 | 3094 | |
43f5e655 DH |
3095 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
3096 | retval = do_reconfigure_mnt(&path, mnt_flags); | |
3097 | else if (flags & MS_REMOUNT) | |
e462ec50 | 3098 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
3099 | data_page); |
3100 | else if (flags & MS_BIND) | |
2d92ab3c | 3101 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 3102 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 3103 | retval = do_change_type(&path, flags); |
1da177e4 | 3104 | else if (flags & MS_MOVE) |
2db154b3 | 3105 | retval = do_move_mount_old(&path, dev_name); |
1da177e4 | 3106 | else |
e462ec50 | 3107 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
3108 | dev_name, data_page); |
3109 | dput_out: | |
2d92ab3c | 3110 | path_put(&path); |
1da177e4 LT |
3111 | return retval; |
3112 | } | |
3113 | ||
537f7ccb EB |
3114 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3115 | { | |
3116 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3117 | } | |
3118 | ||
3119 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3120 | { | |
3121 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3122 | } | |
3123 | ||
771b1371 EB |
3124 | static void free_mnt_ns(struct mnt_namespace *ns) |
3125 | { | |
74e83122 AV |
3126 | if (!is_anon_ns(ns)) |
3127 | ns_free_inum(&ns->ns); | |
537f7ccb | 3128 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3129 | put_user_ns(ns->user_ns); |
3130 | kfree(ns); | |
3131 | } | |
3132 | ||
8823c079 EB |
3133 | /* |
3134 | * Assign a sequence number so we can detect when we attempt to bind | |
3135 | * mount a reference to an older mount namespace into the current | |
3136 | * mount namespace, preventing reference counting loops. A 64bit | |
3137 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3138 | * is effectively never, so we can ignore the possibility. | |
3139 | */ | |
3140 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3141 | ||
74e83122 | 3142 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3143 | { |
3144 | struct mnt_namespace *new_ns; | |
537f7ccb | 3145 | struct ucounts *ucounts; |
98f842e6 | 3146 | int ret; |
cf8d2c11 | 3147 | |
537f7ccb EB |
3148 | ucounts = inc_mnt_namespaces(user_ns); |
3149 | if (!ucounts) | |
df75e774 | 3150 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3151 | |
74e83122 | 3152 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
3153 | if (!new_ns) { |
3154 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3155 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3156 | } |
74e83122 AV |
3157 | if (!anon) { |
3158 | ret = ns_alloc_inum(&new_ns->ns); | |
3159 | if (ret) { | |
3160 | kfree(new_ns); | |
3161 | dec_mnt_namespaces(ucounts); | |
3162 | return ERR_PTR(ret); | |
3163 | } | |
98f842e6 | 3164 | } |
33c42940 | 3165 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3166 | if (!anon) |
3167 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
cf8d2c11 | 3168 | atomic_set(&new_ns->count, 1); |
cf8d2c11 TM |
3169 | INIT_LIST_HEAD(&new_ns->list); |
3170 | init_waitqueue_head(&new_ns->poll); | |
771b1371 | 3171 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3172 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3173 | return new_ns; |
3174 | } | |
3175 | ||
0766f788 | 3176 | __latent_entropy |
9559f689 AV |
3177 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3178 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3179 | { |
6b3286ed | 3180 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3181 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3182 | struct mount *p, *q; |
9559f689 | 3183 | struct mount *old; |
cb338d06 | 3184 | struct mount *new; |
7a472ef4 | 3185 | int copy_flags; |
1da177e4 | 3186 | |
9559f689 AV |
3187 | BUG_ON(!ns); |
3188 | ||
3189 | if (likely(!(flags & CLONE_NEWNS))) { | |
3190 | get_mnt_ns(ns); | |
3191 | return ns; | |
3192 | } | |
3193 | ||
3194 | old = ns->root; | |
3195 | ||
74e83122 | 3196 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3197 | if (IS_ERR(new_ns)) |
3198 | return new_ns; | |
1da177e4 | 3199 | |
97216be0 | 3200 | namespace_lock(); |
1da177e4 | 3201 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3202 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3203 | if (user_ns != ns->user_ns) |
3bd045cc | 3204 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3205 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3206 | if (IS_ERR(new)) { |
328e6d90 | 3207 | namespace_unlock(); |
771b1371 | 3208 | free_mnt_ns(new_ns); |
be34d1a3 | 3209 | return ERR_CAST(new); |
1da177e4 | 3210 | } |
3bd045cc AV |
3211 | if (user_ns != ns->user_ns) { |
3212 | lock_mount_hash(); | |
3213 | lock_mnt_tree(new); | |
3214 | unlock_mount_hash(); | |
3215 | } | |
be08d6d2 | 3216 | new_ns->root = new; |
1a4eeaf2 | 3217 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3218 | |
3219 | /* | |
3220 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3221 | * as belonging to new namespace. We have already acquired a private | |
3222 | * fs_struct, so tsk->fs->lock is not needed. | |
3223 | */ | |
909b0a88 | 3224 | p = old; |
cb338d06 | 3225 | q = new; |
1da177e4 | 3226 | while (p) { |
143c8c91 | 3227 | q->mnt_ns = new_ns; |
d2921684 | 3228 | new_ns->mounts++; |
9559f689 AV |
3229 | if (new_fs) { |
3230 | if (&p->mnt == new_fs->root.mnt) { | |
3231 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3232 | rootmnt = &p->mnt; |
1da177e4 | 3233 | } |
9559f689 AV |
3234 | if (&p->mnt == new_fs->pwd.mnt) { |
3235 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3236 | pwdmnt = &p->mnt; |
1da177e4 | 3237 | } |
1da177e4 | 3238 | } |
909b0a88 AV |
3239 | p = next_mnt(p, old); |
3240 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3241 | if (!q) |
3242 | break; | |
3243 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3244 | p = next_mnt(p, old); | |
1da177e4 | 3245 | } |
328e6d90 | 3246 | namespace_unlock(); |
1da177e4 | 3247 | |
1da177e4 | 3248 | if (rootmnt) |
f03c6599 | 3249 | mntput(rootmnt); |
1da177e4 | 3250 | if (pwdmnt) |
f03c6599 | 3251 | mntput(pwdmnt); |
1da177e4 | 3252 | |
741a2951 | 3253 | return new_ns; |
1da177e4 LT |
3254 | } |
3255 | ||
74e83122 | 3256 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3257 | { |
74e83122 | 3258 | struct mount *mnt = real_mount(m); |
ea441d11 | 3259 | struct mnt_namespace *ns; |
d31da0f0 | 3260 | struct super_block *s; |
ea441d11 AV |
3261 | struct path path; |
3262 | int err; | |
3263 | ||
74e83122 AV |
3264 | ns = alloc_mnt_ns(&init_user_ns, true); |
3265 | if (IS_ERR(ns)) { | |
3266 | mntput(m); | |
ea441d11 | 3267 | return ERR_CAST(ns); |
74e83122 AV |
3268 | } |
3269 | mnt->mnt_ns = ns; | |
3270 | ns->root = mnt; | |
3271 | ns->mounts++; | |
3272 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3273 | |
74e83122 | 3274 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3275 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3276 | ||
3277 | put_mnt_ns(ns); | |
3278 | ||
3279 | if (err) | |
3280 | return ERR_PTR(err); | |
3281 | ||
3282 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3283 | s = path.mnt->mnt_sb; |
3284 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3285 | mntput(path.mnt); |
3286 | /* lock the sucker */ | |
d31da0f0 | 3287 | down_write(&s->s_umount); |
ea441d11 AV |
3288 | /* ... and return the root of (sub)tree on it */ |
3289 | return path.dentry; | |
3290 | } | |
3291 | EXPORT_SYMBOL(mount_subtree); | |
3292 | ||
312db1aa DB |
3293 | int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, |
3294 | unsigned long flags, void __user *data) | |
1da177e4 | 3295 | { |
eca6f534 VN |
3296 | int ret; |
3297 | char *kernel_type; | |
eca6f534 | 3298 | char *kernel_dev; |
b40ef869 | 3299 | void *options; |
1da177e4 | 3300 | |
b8850d1f TG |
3301 | kernel_type = copy_mount_string(type); |
3302 | ret = PTR_ERR(kernel_type); | |
3303 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3304 | goto out_type; |
1da177e4 | 3305 | |
b8850d1f TG |
3306 | kernel_dev = copy_mount_string(dev_name); |
3307 | ret = PTR_ERR(kernel_dev); | |
3308 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3309 | goto out_dev; |
1da177e4 | 3310 | |
b40ef869 AV |
3311 | options = copy_mount_options(data); |
3312 | ret = PTR_ERR(options); | |
3313 | if (IS_ERR(options)) | |
eca6f534 | 3314 | goto out_data; |
1da177e4 | 3315 | |
b40ef869 | 3316 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3317 | |
b40ef869 | 3318 | kfree(options); |
eca6f534 VN |
3319 | out_data: |
3320 | kfree(kernel_dev); | |
3321 | out_dev: | |
eca6f534 VN |
3322 | kfree(kernel_type); |
3323 | out_type: | |
3324 | return ret; | |
1da177e4 LT |
3325 | } |
3326 | ||
312db1aa DB |
3327 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3328 | char __user *, type, unsigned long, flags, void __user *, data) | |
3329 | { | |
3330 | return ksys_mount(dev_name, dir_name, type, flags, data); | |
3331 | } | |
3332 | ||
2db154b3 | 3333 | /* |
93766fbd DH |
3334 | * Create a kernel mount representation for a new, prepared superblock |
3335 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3336 | */ | |
3337 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3338 | unsigned int, attr_flags) | |
3339 | { | |
3340 | struct mnt_namespace *ns; | |
3341 | struct fs_context *fc; | |
3342 | struct file *file; | |
3343 | struct path newmount; | |
3344 | struct mount *mnt; | |
3345 | struct fd f; | |
3346 | unsigned int mnt_flags = 0; | |
3347 | long ret; | |
3348 | ||
3349 | if (!may_mount()) | |
3350 | return -EPERM; | |
3351 | ||
3352 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3353 | return -EINVAL; | |
3354 | ||
3355 | if (attr_flags & ~(MOUNT_ATTR_RDONLY | | |
3356 | MOUNT_ATTR_NOSUID | | |
3357 | MOUNT_ATTR_NODEV | | |
3358 | MOUNT_ATTR_NOEXEC | | |
3359 | MOUNT_ATTR__ATIME | | |
3360 | MOUNT_ATTR_NODIRATIME)) | |
3361 | return -EINVAL; | |
3362 | ||
3363 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3364 | mnt_flags |= MNT_READONLY; | |
3365 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3366 | mnt_flags |= MNT_NOSUID; | |
3367 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3368 | mnt_flags |= MNT_NODEV; | |
3369 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3370 | mnt_flags |= MNT_NOEXEC; | |
3371 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3372 | mnt_flags |= MNT_NODIRATIME; | |
3373 | ||
3374 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3375 | case MOUNT_ATTR_STRICTATIME: | |
3376 | break; | |
3377 | case MOUNT_ATTR_NOATIME: | |
3378 | mnt_flags |= MNT_NOATIME; | |
3379 | break; | |
3380 | case MOUNT_ATTR_RELATIME: | |
3381 | mnt_flags |= MNT_RELATIME; | |
3382 | break; | |
3383 | default: | |
3384 | return -EINVAL; | |
3385 | } | |
3386 | ||
3387 | f = fdget(fs_fd); | |
3388 | if (!f.file) | |
3389 | return -EBADF; | |
3390 | ||
3391 | ret = -EINVAL; | |
3392 | if (f.file->f_op != &fscontext_fops) | |
3393 | goto err_fsfd; | |
3394 | ||
3395 | fc = f.file->private_data; | |
3396 | ||
3397 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3398 | if (ret < 0) | |
3399 | goto err_fsfd; | |
3400 | ||
3401 | /* There must be a valid superblock or we can't mount it */ | |
3402 | ret = -EINVAL; | |
3403 | if (!fc->root) | |
3404 | goto err_unlock; | |
3405 | ||
3406 | ret = -EPERM; | |
3407 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3408 | pr_warn("VFS: Mount too revealing\n"); | |
3409 | goto err_unlock; | |
3410 | } | |
3411 | ||
3412 | ret = -EBUSY; | |
3413 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3414 | goto err_unlock; | |
3415 | ||
3416 | ret = -EPERM; | |
3417 | if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) | |
3418 | goto err_unlock; | |
3419 | ||
3420 | newmount.mnt = vfs_create_mount(fc); | |
3421 | if (IS_ERR(newmount.mnt)) { | |
3422 | ret = PTR_ERR(newmount.mnt); | |
3423 | goto err_unlock; | |
3424 | } | |
3425 | newmount.dentry = dget(fc->root); | |
3426 | newmount.mnt->mnt_flags = mnt_flags; | |
3427 | ||
3428 | /* We've done the mount bit - now move the file context into more or | |
3429 | * less the same state as if we'd done an fspick(). We don't want to | |
3430 | * do any memory allocation or anything like that at this point as we | |
3431 | * don't want to have to handle any errors incurred. | |
3432 | */ | |
3433 | vfs_clean_context(fc); | |
3434 | ||
3435 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3436 | if (IS_ERR(ns)) { | |
3437 | ret = PTR_ERR(ns); | |
3438 | goto err_path; | |
3439 | } | |
3440 | mnt = real_mount(newmount.mnt); | |
3441 | mnt->mnt_ns = ns; | |
3442 | ns->root = mnt; | |
3443 | ns->mounts = 1; | |
3444 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3445 | mntget(newmount.mnt); |
93766fbd DH |
3446 | |
3447 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3448 | * it, not just simply put it. | |
3449 | */ | |
3450 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3451 | if (IS_ERR(file)) { | |
3452 | dissolve_on_fput(newmount.mnt); | |
3453 | ret = PTR_ERR(file); | |
3454 | goto err_path; | |
3455 | } | |
3456 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3457 | ||
3458 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3459 | if (ret >= 0) | |
3460 | fd_install(ret, file); | |
3461 | else | |
3462 | fput(file); | |
3463 | ||
3464 | err_path: | |
3465 | path_put(&newmount); | |
3466 | err_unlock: | |
3467 | mutex_unlock(&fc->uapi_mutex); | |
3468 | err_fsfd: | |
3469 | fdput(f); | |
3470 | return ret; | |
3471 | } | |
3472 | ||
3473 | /* | |
3474 | * Move a mount from one place to another. In combination with | |
3475 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3476 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3477 | * a mount subtree. | |
2db154b3 DH |
3478 | * |
3479 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3480 | */ | |
3481 | SYSCALL_DEFINE5(move_mount, | |
3482 | int, from_dfd, const char *, from_pathname, | |
3483 | int, to_dfd, const char *, to_pathname, | |
3484 | unsigned int, flags) | |
3485 | { | |
3486 | struct path from_path, to_path; | |
3487 | unsigned int lflags; | |
3488 | int ret = 0; | |
3489 | ||
3490 | if (!may_mount()) | |
3491 | return -EPERM; | |
3492 | ||
3493 | if (flags & ~MOVE_MOUNT__MASK) | |
3494 | return -EINVAL; | |
3495 | ||
3496 | /* If someone gives a pathname, they aren't permitted to move | |
3497 | * from an fd that requires unmount as we can't get at the flag | |
3498 | * to clear it afterwards. | |
3499 | */ | |
3500 | lflags = 0; | |
3501 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3502 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3503 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3504 | ||
3505 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3506 | if (ret < 0) | |
3507 | return ret; | |
3508 | ||
3509 | lflags = 0; | |
3510 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3511 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3512 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3513 | ||
3514 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3515 | if (ret < 0) | |
3516 | goto out_from; | |
3517 | ||
3518 | ret = security_move_mount(&from_path, &to_path); | |
3519 | if (ret < 0) | |
3520 | goto out_to; | |
3521 | ||
3522 | ret = do_move_mount(&from_path, &to_path); | |
3523 | ||
3524 | out_to: | |
3525 | path_put(&to_path); | |
3526 | out_from: | |
3527 | path_put(&from_path); | |
3528 | return ret; | |
3529 | } | |
3530 | ||
afac7cba AV |
3531 | /* |
3532 | * Return true if path is reachable from root | |
3533 | * | |
48a066e7 | 3534 | * namespace_sem or mount_lock is held |
afac7cba | 3535 | */ |
643822b4 | 3536 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3537 | const struct path *root) |
3538 | { | |
643822b4 | 3539 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3540 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3541 | mnt = mnt->mnt_parent; |
afac7cba | 3542 | } |
643822b4 | 3543 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3544 | } |
3545 | ||
640eb7e7 | 3546 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3547 | { |
25ab4c9b | 3548 | bool res; |
48a066e7 | 3549 | read_seqlock_excl(&mount_lock); |
643822b4 | 3550 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3551 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3552 | return res; |
3553 | } | |
3554 | EXPORT_SYMBOL(path_is_under); | |
3555 | ||
1da177e4 LT |
3556 | /* |
3557 | * pivot_root Semantics: | |
3558 | * Moves the root file system of the current process to the directory put_old, | |
3559 | * makes new_root as the new root file system of the current process, and sets | |
3560 | * root/cwd of all processes which had them on the current root to new_root. | |
3561 | * | |
3562 | * Restrictions: | |
3563 | * The new_root and put_old must be directories, and must not be on the | |
3564 | * same file system as the current process root. The put_old must be | |
3565 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3566 | * pointed to by put_old must yield the same directory as new_root. No other | |
3567 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3568 | * | |
4a0d11fa NB |
3569 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3570 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3571 | * in this situation. | |
3572 | * | |
1da177e4 LT |
3573 | * Notes: |
3574 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3575 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3576 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3577 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3578 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3579 | * first. | |
3580 | */ | |
3480b257 HC |
3581 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3582 | const char __user *, put_old) | |
1da177e4 | 3583 | { |
2763d119 AV |
3584 | struct path new, old, root; |
3585 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3586 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3587 | int error; |
3588 | ||
9b40bc90 | 3589 | if (!may_mount()) |
1da177e4 LT |
3590 | return -EPERM; |
3591 | ||
2d8f3038 | 3592 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3593 | if (error) |
3594 | goto out0; | |
1da177e4 | 3595 | |
2d8f3038 | 3596 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3597 | if (error) |
3598 | goto out1; | |
3599 | ||
2d8f3038 | 3600 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3601 | if (error) |
3602 | goto out2; | |
1da177e4 | 3603 | |
f7ad3c6b | 3604 | get_fs_root(current->fs, &root); |
84d17192 AV |
3605 | old_mp = lock_mount(&old); |
3606 | error = PTR_ERR(old_mp); | |
3607 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3608 | goto out3; |
3609 | ||
1da177e4 | 3610 | error = -EINVAL; |
419148da AV |
3611 | new_mnt = real_mount(new.mnt); |
3612 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3613 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3614 | ex_parent = new_mnt->mnt_parent; |
3615 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3616 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3617 | IS_MNT_SHARED(ex_parent) || |
3618 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3619 | goto out4; |
143c8c91 | 3620 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3621 | goto out4; |
5ff9d8a6 EB |
3622 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3623 | goto out4; | |
1da177e4 | 3624 | error = -ENOENT; |
f3da392e | 3625 | if (d_unlinked(new.dentry)) |
b12cea91 | 3626 | goto out4; |
1da177e4 | 3627 | error = -EBUSY; |
84d17192 | 3628 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3629 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3630 | error = -EINVAL; |
8c3ee42e | 3631 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3632 | goto out4; /* not a mountpoint */ |
676da58d | 3633 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3634 | goto out4; /* not attached */ |
2d8f3038 | 3635 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3636 | goto out4; /* not a mountpoint */ |
676da58d | 3637 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3638 | goto out4; /* not attached */ |
4ac91378 | 3639 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3640 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3641 | goto out4; |
0d082601 EB |
3642 | /* make certain new is below the root */ |
3643 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3644 | goto out4; | |
719ea2fb | 3645 | lock_mount_hash(); |
2763d119 AV |
3646 | umount_mnt(new_mnt); |
3647 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3648 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3649 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3650 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3651 | } | |
4ac91378 | 3652 | /* mount old root on put_old */ |
84d17192 | 3653 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3654 | /* mount new_root on / */ |
2763d119 AV |
3655 | attach_mnt(new_mnt, root_parent, root_mp); |
3656 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3657 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3658 | /* A moved mount should not expire automatically */ |
3659 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3660 | put_mountpoint(root_mp); |
719ea2fb | 3661 | unlock_mount_hash(); |
2d8f3038 | 3662 | chroot_fs_refs(&root, &new); |
1da177e4 | 3663 | error = 0; |
b12cea91 | 3664 | out4: |
84d17192 | 3665 | unlock_mount(old_mp); |
2763d119 AV |
3666 | if (!error) |
3667 | mntput_no_expire(ex_parent); | |
b12cea91 | 3668 | out3: |
8c3ee42e | 3669 | path_put(&root); |
b12cea91 | 3670 | out2: |
2d8f3038 | 3671 | path_put(&old); |
1da177e4 | 3672 | out1: |
2d8f3038 | 3673 | path_put(&new); |
1da177e4 | 3674 | out0: |
1da177e4 | 3675 | return error; |
1da177e4 LT |
3676 | } |
3677 | ||
3678 | static void __init init_mount_tree(void) | |
3679 | { | |
3680 | struct vfsmount *mnt; | |
74e83122 | 3681 | struct mount *m; |
6b3286ed | 3682 | struct mnt_namespace *ns; |
ac748a09 | 3683 | struct path root; |
0c55cfc4 | 3684 | struct file_system_type *type; |
1da177e4 | 3685 | |
0c55cfc4 EB |
3686 | type = get_fs_type("rootfs"); |
3687 | if (!type) | |
3688 | panic("Can't find rootfs type"); | |
3689 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3690 | put_filesystem(type); | |
1da177e4 LT |
3691 | if (IS_ERR(mnt)) |
3692 | panic("Can't create rootfs"); | |
b3e19d92 | 3693 | |
74e83122 | 3694 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 3695 | if (IS_ERR(ns)) |
1da177e4 | 3696 | panic("Can't allocate initial namespace"); |
74e83122 AV |
3697 | m = real_mount(mnt); |
3698 | m->mnt_ns = ns; | |
3699 | ns->root = m; | |
3700 | ns->mounts = 1; | |
3701 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
3702 | init_task.nsproxy->mnt_ns = ns; |
3703 | get_mnt_ns(ns); | |
3704 | ||
be08d6d2 AV |
3705 | root.mnt = mnt; |
3706 | root.dentry = mnt->mnt_root; | |
da362b09 | 3707 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3708 | |
3709 | set_fs_pwd(current->fs, &root); | |
3710 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3711 | } |
3712 | ||
74bf17cf | 3713 | void __init mnt_init(void) |
1da177e4 | 3714 | { |
15a67dd8 | 3715 | int err; |
1da177e4 | 3716 | |
7d6fec45 | 3717 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3718 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3719 | |
0818bf27 | 3720 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3721 | sizeof(struct hlist_head), |
0818bf27 | 3722 | mhash_entries, 19, |
3d375d78 | 3723 | HASH_ZERO, |
0818bf27 AV |
3724 | &m_hash_shift, &m_hash_mask, 0, 0); |
3725 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3726 | sizeof(struct hlist_head), | |
3727 | mphash_entries, 19, | |
3d375d78 | 3728 | HASH_ZERO, |
0818bf27 | 3729 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3730 | |
84d17192 | 3731 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3732 | panic("Failed to allocate mount hash table\n"); |
3733 | ||
4b93dc9b TH |
3734 | kernfs_init(); |
3735 | ||
15a67dd8 RD |
3736 | err = sysfs_init(); |
3737 | if (err) | |
3738 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3739 | __func__, err); |
00d26666 GKH |
3740 | fs_kobj = kobject_create_and_add("fs", NULL); |
3741 | if (!fs_kobj) | |
8e24eea7 | 3742 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3743 | init_rootfs(); |
3744 | init_mount_tree(); | |
3745 | } | |
3746 | ||
616511d0 | 3747 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3748 | { |
d498b25a | 3749 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3750 | return; |
7b00ed6f | 3751 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3752 | free_mnt_ns(ns); |
1da177e4 | 3753 | } |
9d412a43 | 3754 | |
d911b458 | 3755 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 3756 | { |
423e0ab0 | 3757 | struct vfsmount *mnt; |
d911b458 | 3758 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
3759 | if (!IS_ERR(mnt)) { |
3760 | /* | |
3761 | * it is a longterm mount, don't release mnt until | |
3762 | * we unmount before file sys is unregistered | |
3763 | */ | |
f7a99c5b | 3764 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3765 | } |
3766 | return mnt; | |
9d412a43 | 3767 | } |
d911b458 | 3768 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
3769 | |
3770 | void kern_unmount(struct vfsmount *mnt) | |
3771 | { | |
3772 | /* release long term mount so mount point can be released */ | |
3773 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3774 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3775 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3776 | mntput(mnt); |
3777 | } | |
3778 | } | |
3779 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3780 | |
3781 | bool our_mnt(struct vfsmount *mnt) | |
3782 | { | |
143c8c91 | 3783 | return check_mnt(real_mount(mnt)); |
02125a82 | 3784 | } |
8823c079 | 3785 | |
3151527e EB |
3786 | bool current_chrooted(void) |
3787 | { | |
3788 | /* Does the current process have a non-standard root */ | |
3789 | struct path ns_root; | |
3790 | struct path fs_root; | |
3791 | bool chrooted; | |
3792 | ||
3793 | /* Find the namespace root */ | |
3794 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3795 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3796 | path_get(&ns_root); | |
3797 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3798 | ; | |
3799 | ||
3800 | get_fs_root(current->fs, &fs_root); | |
3801 | ||
3802 | chrooted = !path_equal(&fs_root, &ns_root); | |
3803 | ||
3804 | path_put(&fs_root); | |
3805 | path_put(&ns_root); | |
3806 | ||
3807 | return chrooted; | |
3808 | } | |
3809 | ||
132e4608 DH |
3810 | static bool mnt_already_visible(struct mnt_namespace *ns, |
3811 | const struct super_block *sb, | |
8654df4e | 3812 | int *new_mnt_flags) |
87a8ebd6 | 3813 | { |
8c6cf9cc | 3814 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3815 | struct mount *mnt; |
e51db735 | 3816 | bool visible = false; |
87a8ebd6 | 3817 | |
44bb4385 | 3818 | down_read(&namespace_sem); |
87a8ebd6 | 3819 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3820 | struct mount *child; |
77b1a97d EB |
3821 | int mnt_flags; |
3822 | ||
132e4608 | 3823 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
3824 | continue; |
3825 | ||
7e96c1b0 EB |
3826 | /* This mount is not fully visible if it's root directory |
3827 | * is not the root directory of the filesystem. | |
3828 | */ | |
3829 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3830 | continue; | |
3831 | ||
a1935c17 | 3832 | /* A local view of the mount flags */ |
77b1a97d | 3833 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3834 | |
695e9df0 | 3835 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3836 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3837 | mnt_flags |= MNT_LOCK_READONLY; |
3838 | ||
8c6cf9cc EB |
3839 | /* Verify the mount flags are equal to or more permissive |
3840 | * than the proposed new mount. | |
3841 | */ | |
77b1a97d | 3842 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3843 | !(new_flags & MNT_READONLY)) |
3844 | continue; | |
77b1a97d EB |
3845 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3846 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3847 | continue; |
3848 | ||
ceeb0e5d EB |
3849 | /* This mount is not fully visible if there are any |
3850 | * locked child mounts that cover anything except for | |
3851 | * empty directories. | |
e51db735 EB |
3852 | */ |
3853 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3854 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3855 | /* Only worry about locked mounts */ |
d71ed6c9 | 3856 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3857 | continue; |
7236c85e EB |
3858 | /* Is the directory permanetly empty? */ |
3859 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3860 | goto next; |
87a8ebd6 | 3861 | } |
8c6cf9cc | 3862 | /* Preserve the locked attributes */ |
77b1a97d | 3863 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3864 | MNT_LOCK_ATIME); |
e51db735 EB |
3865 | visible = true; |
3866 | goto found; | |
3867 | next: ; | |
87a8ebd6 | 3868 | } |
e51db735 | 3869 | found: |
44bb4385 | 3870 | up_read(&namespace_sem); |
e51db735 | 3871 | return visible; |
87a8ebd6 EB |
3872 | } |
3873 | ||
132e4608 | 3874 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 3875 | { |
a1935c17 | 3876 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3877 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3878 | unsigned long s_iflags; | |
3879 | ||
3880 | if (ns->user_ns == &init_user_ns) | |
3881 | return false; | |
3882 | ||
3883 | /* Can this filesystem be too revealing? */ | |
132e4608 | 3884 | s_iflags = sb->s_iflags; |
8654df4e EB |
3885 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
3886 | return false; | |
3887 | ||
a1935c17 EB |
3888 | if ((s_iflags & required_iflags) != required_iflags) { |
3889 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3890 | required_iflags); | |
3891 | return true; | |
3892 | } | |
3893 | ||
132e4608 | 3894 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
3895 | } |
3896 | ||
380cf5ba AL |
3897 | bool mnt_may_suid(struct vfsmount *mnt) |
3898 | { | |
3899 | /* | |
3900 | * Foreign mounts (accessed via fchdir or through /proc | |
3901 | * symlinks) are always treated as if they are nosuid. This | |
3902 | * prevents namespaces from trusting potentially unsafe | |
3903 | * suid/sgid bits, file caps, or security labels that originate | |
3904 | * in other namespaces. | |
3905 | */ | |
3906 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3907 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3908 | } | |
3909 | ||
64964528 | 3910 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3911 | { |
58be2825 | 3912 | struct ns_common *ns = NULL; |
8823c079 EB |
3913 | struct nsproxy *nsproxy; |
3914 | ||
728dba3a EB |
3915 | task_lock(task); |
3916 | nsproxy = task->nsproxy; | |
8823c079 | 3917 | if (nsproxy) { |
58be2825 AV |
3918 | ns = &nsproxy->mnt_ns->ns; |
3919 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3920 | } |
728dba3a | 3921 | task_unlock(task); |
8823c079 EB |
3922 | |
3923 | return ns; | |
3924 | } | |
3925 | ||
64964528 | 3926 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3927 | { |
58be2825 | 3928 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3929 | } |
3930 | ||
64964528 | 3931 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3932 | { |
3933 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3934 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3935 | struct path root; |
4f757f3c | 3936 | int err; |
8823c079 | 3937 | |
0c55cfc4 | 3938 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3939 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3940 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3941 | return -EPERM; |
8823c079 | 3942 | |
74e83122 AV |
3943 | if (is_anon_ns(mnt_ns)) |
3944 | return -EINVAL; | |
3945 | ||
8823c079 EB |
3946 | if (fs->users != 1) |
3947 | return -EINVAL; | |
3948 | ||
3949 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3950 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3951 | nsproxy->mnt_ns = mnt_ns; |
3952 | ||
3953 | /* Find the root */ | |
4f757f3c AV |
3954 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3955 | "/", LOOKUP_DOWN, &root); | |
3956 | if (err) { | |
3957 | /* revert to old namespace */ | |
3958 | nsproxy->mnt_ns = old_mnt_ns; | |
3959 | put_mnt_ns(mnt_ns); | |
3960 | return err; | |
3961 | } | |
8823c079 | 3962 | |
4068367c AV |
3963 | put_mnt_ns(old_mnt_ns); |
3964 | ||
8823c079 EB |
3965 | /* Update the pwd and root */ |
3966 | set_fs_pwd(fs, &root); | |
3967 | set_fs_root(fs, &root); | |
3968 | ||
3969 | path_put(&root); | |
3970 | return 0; | |
3971 | } | |
3972 | ||
bcac25a5 AV |
3973 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3974 | { | |
3975 | return to_mnt_ns(ns)->user_ns; | |
3976 | } | |
3977 | ||
8823c079 EB |
3978 | const struct proc_ns_operations mntns_operations = { |
3979 | .name = "mnt", | |
3980 | .type = CLONE_NEWNS, | |
3981 | .get = mntns_get, | |
3982 | .put = mntns_put, | |
3983 | .install = mntns_install, | |
bcac25a5 | 3984 | .owner = mntns_owner, |
8823c079 | 3985 | }; |