Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
57c8a661 | 26 | #include <linux/memblock.h> |
9ea459e1 | 27 | #include <linux/task_work.h> |
9164bb4a | 28 | #include <linux/sched/task.h> |
e262e32d | 29 | #include <uapi/linux/mount.h> |
9164bb4a | 30 | |
07b20889 | 31 | #include "pnode.h" |
948730b0 | 32 | #include "internal.h" |
1da177e4 | 33 | |
d2921684 EB |
34 | /* Maximum number of mounts in a mount namespace */ |
35 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
36 | ||
0818bf27 AV |
37 | static unsigned int m_hash_mask __read_mostly; |
38 | static unsigned int m_hash_shift __read_mostly; | |
39 | static unsigned int mp_hash_mask __read_mostly; | |
40 | static unsigned int mp_hash_shift __read_mostly; | |
41 | ||
42 | static __initdata unsigned long mhash_entries; | |
43 | static int __init set_mhash_entries(char *str) | |
44 | { | |
45 | if (!str) | |
46 | return 0; | |
47 | mhash_entries = simple_strtoul(str, &str, 0); | |
48 | return 1; | |
49 | } | |
50 | __setup("mhash_entries=", set_mhash_entries); | |
51 | ||
52 | static __initdata unsigned long mphash_entries; | |
53 | static int __init set_mphash_entries(char *str) | |
54 | { | |
55 | if (!str) | |
56 | return 0; | |
57 | mphash_entries = simple_strtoul(str, &str, 0); | |
58 | return 1; | |
59 | } | |
60 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 61 | |
c7999c36 | 62 | static u64 event; |
73cd49ec | 63 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 64 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 65 | |
38129a13 | 66 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 67 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 68 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 69 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 70 | |
f87fd4c2 | 71 | /* /sys/fs */ |
00d26666 GKH |
72 | struct kobject *fs_kobj; |
73 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 74 | |
99b7db7b NP |
75 | /* |
76 | * vfsmount lock may be taken for read to prevent changes to the | |
77 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
78 | * up the tree. | |
79 | * | |
80 | * It should be taken for write in all cases where the vfsmount | |
81 | * tree or hash is modified or when a vfsmount structure is modified. | |
82 | */ | |
48a066e7 | 83 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 84 | |
38129a13 | 85 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 86 | { |
b58fed8b RP |
87 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
88 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
89 | tmp = tmp + (tmp >> m_hash_shift); |
90 | return &mount_hashtable[tmp & m_hash_mask]; | |
91 | } | |
92 | ||
93 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
94 | { | |
95 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
96 | tmp = tmp + (tmp >> mp_hash_shift); | |
97 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
98 | } |
99 | ||
b105e270 | 100 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 101 | { |
169b480e MW |
102 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
103 | ||
104 | if (res < 0) | |
105 | return res; | |
106 | mnt->mnt_id = res; | |
107 | return 0; | |
73cd49ec MS |
108 | } |
109 | ||
b105e270 | 110 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 111 | { |
169b480e | 112 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
113 | } |
114 | ||
719f5d7f MS |
115 | /* |
116 | * Allocate a new peer group ID | |
719f5d7f | 117 | */ |
4b8b21f4 | 118 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 119 | { |
169b480e | 120 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 121 | |
169b480e MW |
122 | if (res < 0) |
123 | return res; | |
124 | mnt->mnt_group_id = res; | |
125 | return 0; | |
719f5d7f MS |
126 | } |
127 | ||
128 | /* | |
129 | * Release a peer group ID | |
130 | */ | |
4b8b21f4 | 131 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 132 | { |
169b480e | 133 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 134 | mnt->mnt_group_id = 0; |
719f5d7f MS |
135 | } |
136 | ||
b3e19d92 NP |
137 | /* |
138 | * vfsmount lock must be held for read | |
139 | */ | |
83adc753 | 140 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
141 | { |
142 | #ifdef CONFIG_SMP | |
68e8a9fe | 143 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
144 | #else |
145 | preempt_disable(); | |
68e8a9fe | 146 | mnt->mnt_count += n; |
b3e19d92 NP |
147 | preempt_enable(); |
148 | #endif | |
149 | } | |
150 | ||
b3e19d92 NP |
151 | /* |
152 | * vfsmount lock must be held for write | |
153 | */ | |
83adc753 | 154 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
155 | { |
156 | #ifdef CONFIG_SMP | |
f03c6599 | 157 | unsigned int count = 0; |
b3e19d92 NP |
158 | int cpu; |
159 | ||
160 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 161 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
162 | } |
163 | ||
164 | return count; | |
165 | #else | |
68e8a9fe | 166 | return mnt->mnt_count; |
b3e19d92 NP |
167 | #endif |
168 | } | |
169 | ||
87b95ce0 AV |
170 | static void drop_mountpoint(struct fs_pin *p) |
171 | { | |
172 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
173 | dput(m->mnt_ex_mountpoint); | |
174 | pin_remove(p); | |
175 | mntput(&m->mnt); | |
176 | } | |
177 | ||
b105e270 | 178 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 179 | { |
c63181e6 AV |
180 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
181 | if (mnt) { | |
73cd49ec MS |
182 | int err; |
183 | ||
c63181e6 | 184 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
185 | if (err) |
186 | goto out_free_cache; | |
187 | ||
188 | if (name) { | |
fcc139ae | 189 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 190 | if (!mnt->mnt_devname) |
88b38782 | 191 | goto out_free_id; |
73cd49ec MS |
192 | } |
193 | ||
b3e19d92 | 194 | #ifdef CONFIG_SMP |
c63181e6 AV |
195 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
196 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
197 | goto out_free_devname; |
198 | ||
c63181e6 | 199 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 200 | #else |
c63181e6 AV |
201 | mnt->mnt_count = 1; |
202 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
203 | #endif |
204 | ||
38129a13 | 205 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
206 | INIT_LIST_HEAD(&mnt->mnt_child); |
207 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
208 | INIT_LIST_HEAD(&mnt->mnt_list); | |
209 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
210 | INIT_LIST_HEAD(&mnt->mnt_share); | |
211 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
212 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 213 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 214 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
87b95ce0 | 215 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 216 | } |
c63181e6 | 217 | return mnt; |
88b38782 | 218 | |
d3ef3d73 | 219 | #ifdef CONFIG_SMP |
220 | out_free_devname: | |
fcc139ae | 221 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 222 | #endif |
88b38782 | 223 | out_free_id: |
c63181e6 | 224 | mnt_free_id(mnt); |
88b38782 | 225 | out_free_cache: |
c63181e6 | 226 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 227 | return NULL; |
1da177e4 LT |
228 | } |
229 | ||
3d733633 DH |
230 | /* |
231 | * Most r/o checks on a fs are for operations that take | |
232 | * discrete amounts of time, like a write() or unlink(). | |
233 | * We must keep track of when those operations start | |
234 | * (for permission checks) and when they end, so that | |
235 | * we can determine when writes are able to occur to | |
236 | * a filesystem. | |
237 | */ | |
238 | /* | |
239 | * __mnt_is_readonly: check whether a mount is read-only | |
240 | * @mnt: the mount to check for its write status | |
241 | * | |
242 | * This shouldn't be used directly ouside of the VFS. | |
243 | * It does not guarantee that the filesystem will stay | |
244 | * r/w, just that it is right *now*. This can not and | |
245 | * should not be used in place of IS_RDONLY(inode). | |
246 | * mnt_want/drop_write() will _keep_ the filesystem | |
247 | * r/w. | |
248 | */ | |
43f5e655 | 249 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 250 | { |
43f5e655 | 251 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
252 | } |
253 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
254 | ||
83adc753 | 255 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 256 | { |
257 | #ifdef CONFIG_SMP | |
68e8a9fe | 258 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 259 | #else |
68e8a9fe | 260 | mnt->mnt_writers++; |
d3ef3d73 | 261 | #endif |
262 | } | |
3d733633 | 263 | |
83adc753 | 264 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 265 | { |
d3ef3d73 | 266 | #ifdef CONFIG_SMP |
68e8a9fe | 267 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 268 | #else |
68e8a9fe | 269 | mnt->mnt_writers--; |
d3ef3d73 | 270 | #endif |
3d733633 | 271 | } |
3d733633 | 272 | |
83adc753 | 273 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 274 | { |
d3ef3d73 | 275 | #ifdef CONFIG_SMP |
276 | unsigned int count = 0; | |
3d733633 | 277 | int cpu; |
3d733633 DH |
278 | |
279 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 280 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 281 | } |
3d733633 | 282 | |
d3ef3d73 | 283 | return count; |
284 | #else | |
285 | return mnt->mnt_writers; | |
286 | #endif | |
3d733633 DH |
287 | } |
288 | ||
4ed5e82f MS |
289 | static int mnt_is_readonly(struct vfsmount *mnt) |
290 | { | |
291 | if (mnt->mnt_sb->s_readonly_remount) | |
292 | return 1; | |
293 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
294 | smp_rmb(); | |
295 | return __mnt_is_readonly(mnt); | |
296 | } | |
297 | ||
8366025e | 298 | /* |
eb04c282 JK |
299 | * Most r/o & frozen checks on a fs are for operations that take discrete |
300 | * amounts of time, like a write() or unlink(). We must keep track of when | |
301 | * those operations start (for permission checks) and when they end, so that we | |
302 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
303 | */ |
304 | /** | |
eb04c282 | 305 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 306 | * @m: the mount on which to take a write |
8366025e | 307 | * |
eb04c282 JK |
308 | * This tells the low-level filesystem that a write is about to be performed to |
309 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
310 | * returning success. This operation does not protect against filesystem being | |
311 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
312 | * called. This is effectively a refcount. | |
8366025e | 313 | */ |
eb04c282 | 314 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 315 | { |
83adc753 | 316 | struct mount *mnt = real_mount(m); |
3d733633 | 317 | int ret = 0; |
3d733633 | 318 | |
d3ef3d73 | 319 | preempt_disable(); |
c6653a83 | 320 | mnt_inc_writers(mnt); |
d3ef3d73 | 321 | /* |
c6653a83 | 322 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 323 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
324 | * incremented count after it has set MNT_WRITE_HOLD. | |
325 | */ | |
326 | smp_mb(); | |
6aa7de05 | 327 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 328 | cpu_relax(); |
329 | /* | |
330 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
331 | * be set to match its requirements. So we must not load that until | |
332 | * MNT_WRITE_HOLD is cleared. | |
333 | */ | |
334 | smp_rmb(); | |
4ed5e82f | 335 | if (mnt_is_readonly(m)) { |
c6653a83 | 336 | mnt_dec_writers(mnt); |
3d733633 | 337 | ret = -EROFS; |
3d733633 | 338 | } |
d3ef3d73 | 339 | preempt_enable(); |
eb04c282 JK |
340 | |
341 | return ret; | |
342 | } | |
343 | ||
344 | /** | |
345 | * mnt_want_write - get write access to a mount | |
346 | * @m: the mount on which to take a write | |
347 | * | |
348 | * This tells the low-level filesystem that a write is about to be performed to | |
349 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
350 | * is not frozen) before returning success. When the write operation is | |
351 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
352 | */ | |
353 | int mnt_want_write(struct vfsmount *m) | |
354 | { | |
355 | int ret; | |
356 | ||
357 | sb_start_write(m->mnt_sb); | |
358 | ret = __mnt_want_write(m); | |
359 | if (ret) | |
360 | sb_end_write(m->mnt_sb); | |
3d733633 | 361 | return ret; |
8366025e DH |
362 | } |
363 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
364 | ||
96029c4e | 365 | /** |
366 | * mnt_clone_write - get write access to a mount | |
367 | * @mnt: the mount on which to take a write | |
368 | * | |
369 | * This is effectively like mnt_want_write, except | |
370 | * it must only be used to take an extra write reference | |
371 | * on a mountpoint that we already know has a write reference | |
372 | * on it. This allows some optimisation. | |
373 | * | |
374 | * After finished, mnt_drop_write must be called as usual to | |
375 | * drop the reference. | |
376 | */ | |
377 | int mnt_clone_write(struct vfsmount *mnt) | |
378 | { | |
379 | /* superblock may be r/o */ | |
380 | if (__mnt_is_readonly(mnt)) | |
381 | return -EROFS; | |
382 | preempt_disable(); | |
83adc753 | 383 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 384 | preempt_enable(); |
385 | return 0; | |
386 | } | |
387 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
388 | ||
389 | /** | |
eb04c282 | 390 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 391 | * @file: the file who's mount on which to take a write |
392 | * | |
eb04c282 | 393 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 394 | * do some optimisations if the file is open for write already |
395 | */ | |
eb04c282 | 396 | int __mnt_want_write_file(struct file *file) |
96029c4e | 397 | { |
83f936c7 | 398 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 399 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 400 | else |
401 | return mnt_clone_write(file->f_path.mnt); | |
402 | } | |
eb04c282 | 403 | |
7c6893e3 MS |
404 | /** |
405 | * mnt_want_write_file - get write access to a file's mount | |
406 | * @file: the file who's mount on which to take a write | |
407 | * | |
408 | * This is like mnt_want_write, but it takes a file and can | |
409 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
410 | */ |
411 | int mnt_want_write_file(struct file *file) | |
412 | { | |
413 | int ret; | |
414 | ||
a6795a58 | 415 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
416 | ret = __mnt_want_write_file(file); |
417 | if (ret) | |
a6795a58 | 418 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
419 | return ret; |
420 | } | |
96029c4e | 421 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
422 | ||
8366025e | 423 | /** |
eb04c282 | 424 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
425 | * @mnt: the mount on which to give up write access |
426 | * | |
427 | * Tells the low-level filesystem that we are done | |
428 | * performing writes to it. Must be matched with | |
eb04c282 | 429 | * __mnt_want_write() call above. |
8366025e | 430 | */ |
eb04c282 | 431 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 432 | { |
d3ef3d73 | 433 | preempt_disable(); |
83adc753 | 434 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 435 | preempt_enable(); |
8366025e | 436 | } |
eb04c282 JK |
437 | |
438 | /** | |
439 | * mnt_drop_write - give up write access to a mount | |
440 | * @mnt: the mount on which to give up write access | |
441 | * | |
442 | * Tells the low-level filesystem that we are done performing writes to it and | |
443 | * also allows filesystem to be frozen again. Must be matched with | |
444 | * mnt_want_write() call above. | |
445 | */ | |
446 | void mnt_drop_write(struct vfsmount *mnt) | |
447 | { | |
448 | __mnt_drop_write(mnt); | |
449 | sb_end_write(mnt->mnt_sb); | |
450 | } | |
8366025e DH |
451 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
452 | ||
eb04c282 JK |
453 | void __mnt_drop_write_file(struct file *file) |
454 | { | |
455 | __mnt_drop_write(file->f_path.mnt); | |
456 | } | |
457 | ||
7c6893e3 MS |
458 | void mnt_drop_write_file(struct file *file) |
459 | { | |
a6795a58 | 460 | __mnt_drop_write_file(file); |
7c6893e3 MS |
461 | sb_end_write(file_inode(file)->i_sb); |
462 | } | |
2a79f17e AV |
463 | EXPORT_SYMBOL(mnt_drop_write_file); |
464 | ||
83adc753 | 465 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 466 | { |
3d733633 DH |
467 | int ret = 0; |
468 | ||
719ea2fb | 469 | lock_mount_hash(); |
83adc753 | 470 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 471 | /* |
d3ef3d73 | 472 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
473 | * should be visible before we do. | |
3d733633 | 474 | */ |
d3ef3d73 | 475 | smp_mb(); |
476 | ||
3d733633 | 477 | /* |
d3ef3d73 | 478 | * With writers on hold, if this value is zero, then there are |
479 | * definitely no active writers (although held writers may subsequently | |
480 | * increment the count, they'll have to wait, and decrement it after | |
481 | * seeing MNT_READONLY). | |
482 | * | |
483 | * It is OK to have counter incremented on one CPU and decremented on | |
484 | * another: the sum will add up correctly. The danger would be when we | |
485 | * sum up each counter, if we read a counter before it is incremented, | |
486 | * but then read another CPU's count which it has been subsequently | |
487 | * decremented from -- we would see more decrements than we should. | |
488 | * MNT_WRITE_HOLD protects against this scenario, because | |
489 | * mnt_want_write first increments count, then smp_mb, then spins on | |
490 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
491 | * we're counting up here. | |
3d733633 | 492 | */ |
c6653a83 | 493 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 494 | ret = -EBUSY; |
495 | else | |
83adc753 | 496 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 497 | /* |
498 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
499 | * that become unheld will see MNT_READONLY. | |
500 | */ | |
501 | smp_wmb(); | |
83adc753 | 502 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 503 | unlock_mount_hash(); |
3d733633 | 504 | return ret; |
8366025e | 505 | } |
8366025e | 506 | |
43f5e655 | 507 | static int __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 508 | { |
719ea2fb | 509 | lock_mount_hash(); |
83adc753 | 510 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 511 | unlock_mount_hash(); |
43f5e655 | 512 | return 0; |
2e4b7fcd DH |
513 | } |
514 | ||
4ed5e82f MS |
515 | int sb_prepare_remount_readonly(struct super_block *sb) |
516 | { | |
517 | struct mount *mnt; | |
518 | int err = 0; | |
519 | ||
8e8b8796 MS |
520 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
521 | if (atomic_long_read(&sb->s_remove_count)) | |
522 | return -EBUSY; | |
523 | ||
719ea2fb | 524 | lock_mount_hash(); |
4ed5e82f MS |
525 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
526 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
527 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
528 | smp_mb(); | |
529 | if (mnt_get_writers(mnt) > 0) { | |
530 | err = -EBUSY; | |
531 | break; | |
532 | } | |
533 | } | |
534 | } | |
8e8b8796 MS |
535 | if (!err && atomic_long_read(&sb->s_remove_count)) |
536 | err = -EBUSY; | |
537 | ||
4ed5e82f MS |
538 | if (!err) { |
539 | sb->s_readonly_remount = 1; | |
540 | smp_wmb(); | |
541 | } | |
542 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
543 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
544 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
545 | } | |
719ea2fb | 546 | unlock_mount_hash(); |
4ed5e82f MS |
547 | |
548 | return err; | |
549 | } | |
550 | ||
b105e270 | 551 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 552 | { |
fcc139ae | 553 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 554 | #ifdef CONFIG_SMP |
68e8a9fe | 555 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 556 | #endif |
b105e270 | 557 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
558 | } |
559 | ||
8ffcb32e DH |
560 | static void delayed_free_vfsmnt(struct rcu_head *head) |
561 | { | |
562 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
563 | } | |
564 | ||
48a066e7 | 565 | /* call under rcu_read_lock */ |
294d71ff | 566 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
567 | { |
568 | struct mount *mnt; | |
569 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 570 | return 1; |
48a066e7 | 571 | if (bastard == NULL) |
294d71ff | 572 | return 0; |
48a066e7 AV |
573 | mnt = real_mount(bastard); |
574 | mnt_add_count(mnt, 1); | |
119e1ef8 | 575 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 576 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 577 | return 0; |
48a066e7 AV |
578 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
579 | mnt_add_count(mnt, -1); | |
294d71ff AV |
580 | return 1; |
581 | } | |
119e1ef8 AV |
582 | lock_mount_hash(); |
583 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
584 | mnt_add_count(mnt, -1); | |
585 | unlock_mount_hash(); | |
586 | return 1; | |
587 | } | |
588 | unlock_mount_hash(); | |
589 | /* caller will mntput() */ | |
294d71ff AV |
590 | return -1; |
591 | } | |
592 | ||
593 | /* call under rcu_read_lock */ | |
594 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
595 | { | |
596 | int res = __legitimize_mnt(bastard, seq); | |
597 | if (likely(!res)) | |
598 | return true; | |
599 | if (unlikely(res < 0)) { | |
600 | rcu_read_unlock(); | |
601 | mntput(bastard); | |
602 | rcu_read_lock(); | |
48a066e7 | 603 | } |
48a066e7 AV |
604 | return false; |
605 | } | |
606 | ||
1da177e4 | 607 | /* |
474279dc | 608 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 609 | * call under rcu_read_lock() |
1da177e4 | 610 | */ |
474279dc | 611 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 612 | { |
38129a13 | 613 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
614 | struct mount *p; |
615 | ||
38129a13 | 616 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
617 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
618 | return p; | |
619 | return NULL; | |
620 | } | |
621 | ||
a05964f3 | 622 | /* |
f015f126 DH |
623 | * lookup_mnt - Return the first child mount mounted at path |
624 | * | |
625 | * "First" means first mounted chronologically. If you create the | |
626 | * following mounts: | |
627 | * | |
628 | * mount /dev/sda1 /mnt | |
629 | * mount /dev/sda2 /mnt | |
630 | * mount /dev/sda3 /mnt | |
631 | * | |
632 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
633 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
634 | * /dev/sda2, then /dev/sda3, then NULL. | |
635 | * | |
636 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 637 | */ |
ca71cf71 | 638 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 639 | { |
c7105365 | 640 | struct mount *child_mnt; |
48a066e7 AV |
641 | struct vfsmount *m; |
642 | unsigned seq; | |
99b7db7b | 643 | |
48a066e7 AV |
644 | rcu_read_lock(); |
645 | do { | |
646 | seq = read_seqbegin(&mount_lock); | |
647 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
648 | m = child_mnt ? &child_mnt->mnt : NULL; | |
649 | } while (!legitimize_mnt(m, seq)); | |
650 | rcu_read_unlock(); | |
651 | return m; | |
a05964f3 RP |
652 | } |
653 | ||
7af1364f EB |
654 | /* |
655 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
656 | * current mount namespace. | |
657 | * | |
658 | * The common case is dentries are not mountpoints at all and that | |
659 | * test is handled inline. For the slow case when we are actually | |
660 | * dealing with a mountpoint of some kind, walk through all of the | |
661 | * mounts in the current mount namespace and test to see if the dentry | |
662 | * is a mountpoint. | |
663 | * | |
664 | * The mount_hashtable is not usable in the context because we | |
665 | * need to identify all mounts that may be in the current mount | |
666 | * namespace not just a mount that happens to have some specified | |
667 | * parent mount. | |
668 | */ | |
669 | bool __is_local_mountpoint(struct dentry *dentry) | |
670 | { | |
671 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
672 | struct mount *mnt; | |
673 | bool is_covered = false; | |
674 | ||
675 | if (!d_mountpoint(dentry)) | |
676 | goto out; | |
677 | ||
678 | down_read(&namespace_sem); | |
679 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
680 | is_covered = (mnt->mnt_mountpoint == dentry); | |
681 | if (is_covered) | |
682 | break; | |
683 | } | |
684 | up_read(&namespace_sem); | |
685 | out: | |
686 | return is_covered; | |
687 | } | |
688 | ||
e2dfa935 | 689 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 690 | { |
0818bf27 | 691 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
692 | struct mountpoint *mp; |
693 | ||
0818bf27 | 694 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 695 | if (mp->m_dentry == dentry) { |
84d17192 AV |
696 | mp->m_count++; |
697 | return mp; | |
698 | } | |
699 | } | |
e2dfa935 EB |
700 | return NULL; |
701 | } | |
702 | ||
3895dbf8 | 703 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 704 | { |
3895dbf8 | 705 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 706 | int ret; |
84d17192 | 707 | |
3895dbf8 | 708 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
709 | /* might be worth a WARN_ON() */ |
710 | if (d_unlinked(dentry)) | |
711 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
712 | mountpoint: |
713 | read_seqlock_excl(&mount_lock); | |
714 | mp = lookup_mountpoint(dentry); | |
715 | read_sequnlock_excl(&mount_lock); | |
716 | if (mp) | |
717 | goto done; | |
718 | } | |
719 | ||
720 | if (!new) | |
721 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
722 | if (!new) | |
84d17192 AV |
723 | return ERR_PTR(-ENOMEM); |
724 | ||
3895dbf8 EB |
725 | |
726 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 727 | ret = d_set_mounted(dentry); |
eed81007 | 728 | |
3895dbf8 EB |
729 | /* Someone else set d_mounted? */ |
730 | if (ret == -EBUSY) | |
731 | goto mountpoint; | |
732 | ||
733 | /* The dentry is not available as a mountpoint? */ | |
734 | mp = ERR_PTR(ret); | |
735 | if (ret) | |
736 | goto done; | |
737 | ||
738 | /* Add the new mountpoint to the hash table */ | |
739 | read_seqlock_excl(&mount_lock); | |
740 | new->m_dentry = dentry; | |
741 | new->m_count = 1; | |
742 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
743 | INIT_HLIST_HEAD(&new->m_list); | |
744 | read_sequnlock_excl(&mount_lock); | |
745 | ||
746 | mp = new; | |
747 | new = NULL; | |
748 | done: | |
749 | kfree(new); | |
84d17192 AV |
750 | return mp; |
751 | } | |
752 | ||
753 | static void put_mountpoint(struct mountpoint *mp) | |
754 | { | |
755 | if (!--mp->m_count) { | |
756 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 757 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
758 | spin_lock(&dentry->d_lock); |
759 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
760 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 761 | hlist_del(&mp->m_hash); |
84d17192 AV |
762 | kfree(mp); |
763 | } | |
764 | } | |
765 | ||
143c8c91 | 766 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 767 | { |
6b3286ed | 768 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
769 | } |
770 | ||
99b7db7b NP |
771 | /* |
772 | * vfsmount lock must be held for write | |
773 | */ | |
6b3286ed | 774 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
775 | { |
776 | if (ns) { | |
777 | ns->event = ++event; | |
778 | wake_up_interruptible(&ns->poll); | |
779 | } | |
780 | } | |
781 | ||
99b7db7b NP |
782 | /* |
783 | * vfsmount lock must be held for write | |
784 | */ | |
6b3286ed | 785 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
786 | { |
787 | if (ns && ns->event != event) { | |
788 | ns->event = event; | |
789 | wake_up_interruptible(&ns->poll); | |
790 | } | |
791 | } | |
792 | ||
99b7db7b NP |
793 | /* |
794 | * vfsmount lock must be held for write | |
795 | */ | |
7bdb11de | 796 | static void unhash_mnt(struct mount *mnt) |
419148da | 797 | { |
0714a533 | 798 | mnt->mnt_parent = mnt; |
a73324da | 799 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 800 | list_del_init(&mnt->mnt_child); |
38129a13 | 801 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 802 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
803 | put_mountpoint(mnt->mnt_mp); |
804 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
805 | } |
806 | ||
7bdb11de EB |
807 | /* |
808 | * vfsmount lock must be held for write | |
809 | */ | |
810 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
811 | { | |
812 | old_path->dentry = mnt->mnt_mountpoint; | |
813 | old_path->mnt = &mnt->mnt_parent->mnt; | |
814 | unhash_mnt(mnt); | |
815 | } | |
816 | ||
6a46c573 EB |
817 | /* |
818 | * vfsmount lock must be held for write | |
819 | */ | |
820 | static void umount_mnt(struct mount *mnt) | |
821 | { | |
822 | /* old mountpoint will be dropped when we can do that */ | |
823 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
824 | unhash_mnt(mnt); | |
825 | } | |
826 | ||
99b7db7b NP |
827 | /* |
828 | * vfsmount lock must be held for write | |
829 | */ | |
84d17192 AV |
830 | void mnt_set_mountpoint(struct mount *mnt, |
831 | struct mountpoint *mp, | |
44d964d6 | 832 | struct mount *child_mnt) |
b90fa9ae | 833 | { |
84d17192 | 834 | mp->m_count++; |
3a2393d7 | 835 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 836 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 837 | child_mnt->mnt_parent = mnt; |
84d17192 | 838 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 839 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
840 | } |
841 | ||
1064f874 EB |
842 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
843 | { | |
844 | hlist_add_head_rcu(&mnt->mnt_hash, | |
845 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
846 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
847 | } | |
848 | ||
99b7db7b NP |
849 | /* |
850 | * vfsmount lock must be held for write | |
851 | */ | |
84d17192 AV |
852 | static void attach_mnt(struct mount *mnt, |
853 | struct mount *parent, | |
854 | struct mountpoint *mp) | |
1da177e4 | 855 | { |
84d17192 | 856 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 857 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
858 | } |
859 | ||
1064f874 | 860 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 861 | { |
1064f874 EB |
862 | struct mountpoint *old_mp = mnt->mnt_mp; |
863 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
864 | struct mount *old_parent = mnt->mnt_parent; | |
865 | ||
866 | list_del_init(&mnt->mnt_child); | |
867 | hlist_del_init(&mnt->mnt_mp_list); | |
868 | hlist_del_init_rcu(&mnt->mnt_hash); | |
869 | ||
870 | attach_mnt(mnt, parent, mp); | |
871 | ||
872 | put_mountpoint(old_mp); | |
873 | ||
874 | /* | |
875 | * Safely avoid even the suggestion this code might sleep or | |
876 | * lock the mount hash by taking advantage of the knowledge that | |
877 | * mnt_change_mountpoint will not release the final reference | |
878 | * to a mountpoint. | |
879 | * | |
880 | * During mounting, the mount passed in as the parent mount will | |
881 | * continue to use the old mountpoint and during unmounting, the | |
882 | * old mountpoint will continue to exist until namespace_unlock, | |
883 | * which happens well after mnt_change_mountpoint. | |
884 | */ | |
885 | spin_lock(&old_mountpoint->d_lock); | |
886 | old_mountpoint->d_lockref.count--; | |
887 | spin_unlock(&old_mountpoint->d_lock); | |
888 | ||
889 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
890 | } |
891 | ||
b90fa9ae | 892 | /* |
99b7db7b | 893 | * vfsmount lock must be held for write |
b90fa9ae | 894 | */ |
1064f874 | 895 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 896 | { |
0714a533 | 897 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 898 | struct mount *m; |
b90fa9ae | 899 | LIST_HEAD(head); |
143c8c91 | 900 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 901 | |
0714a533 | 902 | BUG_ON(parent == mnt); |
b90fa9ae | 903 | |
1a4eeaf2 | 904 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 905 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 906 | m->mnt_ns = n; |
f03c6599 | 907 | |
b90fa9ae RP |
908 | list_splice(&head, n->list.prev); |
909 | ||
d2921684 EB |
910 | n->mounts += n->pending_mounts; |
911 | n->pending_mounts = 0; | |
912 | ||
1064f874 | 913 | __attach_mnt(mnt, parent); |
6b3286ed | 914 | touch_mnt_namespace(n); |
1da177e4 LT |
915 | } |
916 | ||
909b0a88 | 917 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 918 | { |
6b41d536 AV |
919 | struct list_head *next = p->mnt_mounts.next; |
920 | if (next == &p->mnt_mounts) { | |
1da177e4 | 921 | while (1) { |
909b0a88 | 922 | if (p == root) |
1da177e4 | 923 | return NULL; |
6b41d536 AV |
924 | next = p->mnt_child.next; |
925 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 926 | break; |
0714a533 | 927 | p = p->mnt_parent; |
1da177e4 LT |
928 | } |
929 | } | |
6b41d536 | 930 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
931 | } |
932 | ||
315fc83e | 933 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 934 | { |
6b41d536 AV |
935 | struct list_head *prev = p->mnt_mounts.prev; |
936 | while (prev != &p->mnt_mounts) { | |
937 | p = list_entry(prev, struct mount, mnt_child); | |
938 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
939 | } |
940 | return p; | |
941 | } | |
942 | ||
9d412a43 AV |
943 | struct vfsmount * |
944 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
945 | { | |
b105e270 | 946 | struct mount *mnt; |
9d412a43 AV |
947 | struct dentry *root; |
948 | ||
949 | if (!type) | |
950 | return ERR_PTR(-ENODEV); | |
951 | ||
952 | mnt = alloc_vfsmnt(name); | |
953 | if (!mnt) | |
954 | return ERR_PTR(-ENOMEM); | |
955 | ||
e462ec50 | 956 | if (flags & SB_KERNMOUNT) |
b105e270 | 957 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
958 | |
959 | root = mount_fs(type, flags, name, data); | |
960 | if (IS_ERR(root)) { | |
8ffcb32e | 961 | mnt_free_id(mnt); |
9d412a43 AV |
962 | free_vfsmnt(mnt); |
963 | return ERR_CAST(root); | |
964 | } | |
965 | ||
b105e270 AV |
966 | mnt->mnt.mnt_root = root; |
967 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 968 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 969 | mnt->mnt_parent = mnt; |
719ea2fb | 970 | lock_mount_hash(); |
39f7c4db | 971 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 972 | unlock_mount_hash(); |
b105e270 | 973 | return &mnt->mnt; |
9d412a43 AV |
974 | } |
975 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
976 | ||
93faccbb EB |
977 | struct vfsmount * |
978 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
979 | const char *name, void *data) | |
980 | { | |
981 | /* Until it is worked out how to pass the user namespace | |
982 | * through from the parent mount to the submount don't support | |
983 | * unprivileged mounts with submounts. | |
984 | */ | |
985 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
986 | return ERR_PTR(-EPERM); | |
987 | ||
e462ec50 | 988 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
989 | } |
990 | EXPORT_SYMBOL_GPL(vfs_submount); | |
991 | ||
87129cc0 | 992 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 993 | int flag) |
1da177e4 | 994 | { |
87129cc0 | 995 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
996 | struct mount *mnt; |
997 | int err; | |
1da177e4 | 998 | |
be34d1a3 DH |
999 | mnt = alloc_vfsmnt(old->mnt_devname); |
1000 | if (!mnt) | |
1001 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1002 | |
7a472ef4 | 1003 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1004 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1005 | else | |
1006 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1007 | |
be34d1a3 DH |
1008 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1009 | err = mnt_alloc_group_id(mnt); | |
1010 | if (err) | |
1011 | goto out_free; | |
1da177e4 | 1012 | } |
be34d1a3 | 1013 | |
16a34adb AV |
1014 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1015 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
132c94e3 | 1016 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1017 | if (flag & CL_UNPRIVILEGED) { |
1018 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1019 | ||
1020 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1021 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1022 | ||
1023 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1024 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1025 | ||
1026 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1027 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1028 | ||
1029 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1030 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1031 | } | |
132c94e3 | 1032 | |
5ff9d8a6 | 1033 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1034 | if ((flag & CL_UNPRIVILEGED) && |
1035 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1036 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1037 | ||
be34d1a3 DH |
1038 | atomic_inc(&sb->s_active); |
1039 | mnt->mnt.mnt_sb = sb; | |
1040 | mnt->mnt.mnt_root = dget(root); | |
1041 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1042 | mnt->mnt_parent = mnt; | |
719ea2fb | 1043 | lock_mount_hash(); |
be34d1a3 | 1044 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1045 | unlock_mount_hash(); |
be34d1a3 | 1046 | |
7a472ef4 EB |
1047 | if ((flag & CL_SLAVE) || |
1048 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1049 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1050 | mnt->mnt_master = old; | |
1051 | CLEAR_MNT_SHARED(mnt); | |
1052 | } else if (!(flag & CL_PRIVATE)) { | |
1053 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1054 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1055 | if (IS_MNT_SLAVE(old)) | |
1056 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1057 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1058 | } else { |
1059 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1060 | } |
1061 | if (flag & CL_MAKE_SHARED) | |
1062 | set_mnt_shared(mnt); | |
1063 | ||
1064 | /* stick the duplicate mount on the same expiry list | |
1065 | * as the original if that was on one */ | |
1066 | if (flag & CL_EXPIRE) { | |
1067 | if (!list_empty(&old->mnt_expire)) | |
1068 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1069 | } | |
1070 | ||
cb338d06 | 1071 | return mnt; |
719f5d7f MS |
1072 | |
1073 | out_free: | |
8ffcb32e | 1074 | mnt_free_id(mnt); |
719f5d7f | 1075 | free_vfsmnt(mnt); |
be34d1a3 | 1076 | return ERR_PTR(err); |
1da177e4 LT |
1077 | } |
1078 | ||
9ea459e1 AV |
1079 | static void cleanup_mnt(struct mount *mnt) |
1080 | { | |
1081 | /* | |
1082 | * This probably indicates that somebody messed | |
1083 | * up a mnt_want/drop_write() pair. If this | |
1084 | * happens, the filesystem was probably unable | |
1085 | * to make r/w->r/o transitions. | |
1086 | */ | |
1087 | /* | |
1088 | * The locking used to deal with mnt_count decrement provides barriers, | |
1089 | * so mnt_get_writers() below is safe. | |
1090 | */ | |
1091 | WARN_ON(mnt_get_writers(mnt)); | |
1092 | if (unlikely(mnt->mnt_pins.first)) | |
1093 | mnt_pin_kill(mnt); | |
1094 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1095 | dput(mnt->mnt.mnt_root); | |
1096 | deactivate_super(mnt->mnt.mnt_sb); | |
1097 | mnt_free_id(mnt); | |
1098 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1099 | } | |
1100 | ||
1101 | static void __cleanup_mnt(struct rcu_head *head) | |
1102 | { | |
1103 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1104 | } | |
1105 | ||
1106 | static LLIST_HEAD(delayed_mntput_list); | |
1107 | static void delayed_mntput(struct work_struct *unused) | |
1108 | { | |
1109 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1110 | struct mount *m, *t; |
9ea459e1 | 1111 | |
29785735 BP |
1112 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1113 | cleanup_mnt(m); | |
9ea459e1 AV |
1114 | } |
1115 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1116 | ||
900148dc | 1117 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1118 | { |
48a066e7 | 1119 | rcu_read_lock(); |
9ea0a46c AV |
1120 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1121 | /* | |
1122 | * Since we don't do lock_mount_hash() here, | |
1123 | * ->mnt_ns can change under us. However, if it's | |
1124 | * non-NULL, then there's a reference that won't | |
1125 | * be dropped until after an RCU delay done after | |
1126 | * turning ->mnt_ns NULL. So if we observe it | |
1127 | * non-NULL under rcu_read_lock(), the reference | |
1128 | * we are dropping is not the final one. | |
1129 | */ | |
1130 | mnt_add_count(mnt, -1); | |
48a066e7 | 1131 | rcu_read_unlock(); |
f03c6599 | 1132 | return; |
b3e19d92 | 1133 | } |
719ea2fb | 1134 | lock_mount_hash(); |
119e1ef8 AV |
1135 | /* |
1136 | * make sure that if __legitimize_mnt() has not seen us grab | |
1137 | * mount_lock, we'll see their refcount increment here. | |
1138 | */ | |
1139 | smp_mb(); | |
9ea0a46c | 1140 | mnt_add_count(mnt, -1); |
b3e19d92 | 1141 | if (mnt_get_count(mnt)) { |
48a066e7 | 1142 | rcu_read_unlock(); |
719ea2fb | 1143 | unlock_mount_hash(); |
99b7db7b NP |
1144 | return; |
1145 | } | |
48a066e7 AV |
1146 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1147 | rcu_read_unlock(); | |
1148 | unlock_mount_hash(); | |
1149 | return; | |
1150 | } | |
1151 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1152 | rcu_read_unlock(); | |
962830df | 1153 | |
39f7c4db | 1154 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1155 | |
1156 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1157 | struct mount *p, *tmp; | |
1158 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1159 | umount_mnt(p); | |
1160 | } | |
1161 | } | |
719ea2fb | 1162 | unlock_mount_hash(); |
649a795a | 1163 | |
9ea459e1 AV |
1164 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1165 | struct task_struct *task = current; | |
1166 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1167 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1168 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1169 | return; | |
1170 | } | |
1171 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1172 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1173 | return; | |
1174 | } | |
1175 | cleanup_mnt(mnt); | |
b3e19d92 | 1176 | } |
b3e19d92 NP |
1177 | |
1178 | void mntput(struct vfsmount *mnt) | |
1179 | { | |
1180 | if (mnt) { | |
863d684f | 1181 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1182 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1183 | if (unlikely(m->mnt_expiry_mark)) |
1184 | m->mnt_expiry_mark = 0; | |
1185 | mntput_no_expire(m); | |
b3e19d92 NP |
1186 | } |
1187 | } | |
1188 | EXPORT_SYMBOL(mntput); | |
1189 | ||
1190 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1191 | { | |
1192 | if (mnt) | |
83adc753 | 1193 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1194 | return mnt; |
1195 | } | |
1196 | EXPORT_SYMBOL(mntget); | |
1197 | ||
c6609c0a IK |
1198 | /* path_is_mountpoint() - Check if path is a mount in the current |
1199 | * namespace. | |
1200 | * | |
1201 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1202 | * not mounted in any namespace and that common case is handled inline. | |
1203 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1204 | * mounts using a given dentry in a different namespace. This function | |
1205 | * checks if the passed in path is a mountpoint rather than the dentry | |
1206 | * alone. | |
1207 | */ | |
1208 | bool path_is_mountpoint(const struct path *path) | |
1209 | { | |
1210 | unsigned seq; | |
1211 | bool res; | |
1212 | ||
1213 | if (!d_mountpoint(path->dentry)) | |
1214 | return false; | |
1215 | ||
1216 | rcu_read_lock(); | |
1217 | do { | |
1218 | seq = read_seqbegin(&mount_lock); | |
1219 | res = __path_is_mountpoint(path); | |
1220 | } while (read_seqretry(&mount_lock, seq)); | |
1221 | rcu_read_unlock(); | |
1222 | ||
1223 | return res; | |
1224 | } | |
1225 | EXPORT_SYMBOL(path_is_mountpoint); | |
1226 | ||
ca71cf71 | 1227 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1228 | { |
3064c356 AV |
1229 | struct mount *p; |
1230 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1231 | if (IS_ERR(p)) | |
1232 | return ERR_CAST(p); | |
1233 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1234 | return &p->mnt; | |
7b7b1ace | 1235 | } |
1da177e4 | 1236 | |
a1a2c409 | 1237 | #ifdef CONFIG_PROC_FS |
0226f492 | 1238 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1239 | static void *m_start(struct seq_file *m, loff_t *pos) |
1240 | { | |
ede1bf0d | 1241 | struct proc_mounts *p = m->private; |
1da177e4 | 1242 | |
390c6843 | 1243 | down_read(&namespace_sem); |
c7999c36 AV |
1244 | if (p->cached_event == p->ns->event) { |
1245 | void *v = p->cached_mount; | |
1246 | if (*pos == p->cached_index) | |
1247 | return v; | |
1248 | if (*pos == p->cached_index + 1) { | |
1249 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1250 | return p->cached_mount = v; | |
1251 | } | |
1252 | } | |
1253 | ||
1254 | p->cached_event = p->ns->event; | |
1255 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1256 | p->cached_index = *pos; | |
1257 | return p->cached_mount; | |
1da177e4 LT |
1258 | } |
1259 | ||
1260 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1261 | { | |
ede1bf0d | 1262 | struct proc_mounts *p = m->private; |
b0765fb8 | 1263 | |
c7999c36 AV |
1264 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1265 | p->cached_index = *pos; | |
1266 | return p->cached_mount; | |
1da177e4 LT |
1267 | } |
1268 | ||
1269 | static void m_stop(struct seq_file *m, void *v) | |
1270 | { | |
390c6843 | 1271 | up_read(&namespace_sem); |
1da177e4 LT |
1272 | } |
1273 | ||
0226f492 | 1274 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1275 | { |
ede1bf0d | 1276 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1277 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1278 | return p->show(m, &r->mnt); |
1da177e4 LT |
1279 | } |
1280 | ||
a1a2c409 | 1281 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1282 | .start = m_start, |
1283 | .next = m_next, | |
1284 | .stop = m_stop, | |
0226f492 | 1285 | .show = m_show, |
b4629fe2 | 1286 | }; |
a1a2c409 | 1287 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1288 | |
1da177e4 LT |
1289 | /** |
1290 | * may_umount_tree - check if a mount tree is busy | |
1291 | * @mnt: root of mount tree | |
1292 | * | |
1293 | * This is called to check if a tree of mounts has any | |
1294 | * open files, pwds, chroots or sub mounts that are | |
1295 | * busy. | |
1296 | */ | |
909b0a88 | 1297 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1298 | { |
909b0a88 | 1299 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1300 | int actual_refs = 0; |
1301 | int minimum_refs = 0; | |
315fc83e | 1302 | struct mount *p; |
909b0a88 | 1303 | BUG_ON(!m); |
1da177e4 | 1304 | |
b3e19d92 | 1305 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1306 | lock_mount_hash(); |
909b0a88 | 1307 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1308 | actual_refs += mnt_get_count(p); |
1da177e4 | 1309 | minimum_refs += 2; |
1da177e4 | 1310 | } |
719ea2fb | 1311 | unlock_mount_hash(); |
1da177e4 LT |
1312 | |
1313 | if (actual_refs > minimum_refs) | |
e3474a8e | 1314 | return 0; |
1da177e4 | 1315 | |
e3474a8e | 1316 | return 1; |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | EXPORT_SYMBOL(may_umount_tree); | |
1320 | ||
1321 | /** | |
1322 | * may_umount - check if a mount point is busy | |
1323 | * @mnt: root of mount | |
1324 | * | |
1325 | * This is called to check if a mount point has any | |
1326 | * open files, pwds, chroots or sub mounts. If the | |
1327 | * mount has sub mounts this will return busy | |
1328 | * regardless of whether the sub mounts are busy. | |
1329 | * | |
1330 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1331 | * give false negatives. The main reason why it's here is that we need | |
1332 | * a non-destructive way to look for easily umountable filesystems. | |
1333 | */ | |
1334 | int may_umount(struct vfsmount *mnt) | |
1335 | { | |
e3474a8e | 1336 | int ret = 1; |
8ad08d8a | 1337 | down_read(&namespace_sem); |
719ea2fb | 1338 | lock_mount_hash(); |
1ab59738 | 1339 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1340 | ret = 0; |
719ea2fb | 1341 | unlock_mount_hash(); |
8ad08d8a | 1342 | up_read(&namespace_sem); |
a05964f3 | 1343 | return ret; |
1da177e4 LT |
1344 | } |
1345 | ||
1346 | EXPORT_SYMBOL(may_umount); | |
1347 | ||
38129a13 | 1348 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1349 | |
97216be0 | 1350 | static void namespace_unlock(void) |
70fbcdf4 | 1351 | { |
a3b3c562 | 1352 | struct hlist_head head; |
97216be0 | 1353 | |
a3b3c562 | 1354 | hlist_move_list(&unmounted, &head); |
97216be0 | 1355 | |
97216be0 AV |
1356 | up_write(&namespace_sem); |
1357 | ||
a3b3c562 EB |
1358 | if (likely(hlist_empty(&head))) |
1359 | return; | |
1360 | ||
22cb7405 | 1361 | synchronize_rcu_expedited(); |
48a066e7 | 1362 | |
87b95ce0 | 1363 | group_pin_kill(&head); |
70fbcdf4 RP |
1364 | } |
1365 | ||
97216be0 | 1366 | static inline void namespace_lock(void) |
e3197d83 | 1367 | { |
97216be0 | 1368 | down_write(&namespace_sem); |
e3197d83 AV |
1369 | } |
1370 | ||
e819f152 EB |
1371 | enum umount_tree_flags { |
1372 | UMOUNT_SYNC = 1, | |
1373 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1374 | UMOUNT_CONNECTED = 4, |
e819f152 | 1375 | }; |
f2d0a123 EB |
1376 | |
1377 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1378 | { | |
1379 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1380 | if (how & UMOUNT_SYNC) | |
1381 | return true; | |
1382 | ||
1383 | /* A mount without a parent has nothing to be connected to */ | |
1384 | if (!mnt_has_parent(mnt)) | |
1385 | return true; | |
1386 | ||
1387 | /* Because the reference counting rules change when mounts are | |
1388 | * unmounted and connected, umounted mounts may not be | |
1389 | * connected to mounted mounts. | |
1390 | */ | |
1391 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1392 | return true; | |
1393 | ||
1394 | /* Has it been requested that the mount remain connected? */ | |
1395 | if (how & UMOUNT_CONNECTED) | |
1396 | return false; | |
1397 | ||
1398 | /* Is the mount locked such that it needs to remain connected? */ | |
1399 | if (IS_MNT_LOCKED(mnt)) | |
1400 | return false; | |
1401 | ||
1402 | /* By default disconnect the mount */ | |
1403 | return true; | |
1404 | } | |
1405 | ||
99b7db7b | 1406 | /* |
48a066e7 | 1407 | * mount_lock must be held |
99b7db7b NP |
1408 | * namespace_sem must be held for write |
1409 | */ | |
e819f152 | 1410 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1411 | { |
c003b26f | 1412 | LIST_HEAD(tmp_list); |
315fc83e | 1413 | struct mount *p; |
1da177e4 | 1414 | |
5d88457e EB |
1415 | if (how & UMOUNT_PROPAGATE) |
1416 | propagate_mount_unlock(mnt); | |
1417 | ||
c003b26f | 1418 | /* Gather the mounts to umount */ |
590ce4bc EB |
1419 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1420 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1421 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1422 | } |
1da177e4 | 1423 | |
411a938b | 1424 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1425 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1426 | list_del_init(&p->mnt_child); |
c003b26f | 1427 | } |
88b368f2 | 1428 | |
c003b26f | 1429 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1430 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1431 | propagate_umount(&tmp_list); |
a05964f3 | 1432 | |
c003b26f | 1433 | while (!list_empty(&tmp_list)) { |
d2921684 | 1434 | struct mnt_namespace *ns; |
ce07d891 | 1435 | bool disconnect; |
c003b26f | 1436 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1437 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1438 | list_del_init(&p->mnt_list); |
d2921684 EB |
1439 | ns = p->mnt_ns; |
1440 | if (ns) { | |
1441 | ns->mounts--; | |
1442 | __touch_mnt_namespace(ns); | |
1443 | } | |
143c8c91 | 1444 | p->mnt_ns = NULL; |
e819f152 | 1445 | if (how & UMOUNT_SYNC) |
48a066e7 | 1446 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1447 | |
f2d0a123 | 1448 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1449 | |
1450 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1451 | disconnect ? &unmounted : NULL); | |
676da58d | 1452 | if (mnt_has_parent(p)) { |
81b6b061 | 1453 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1454 | if (!disconnect) { |
1455 | /* Don't forget about p */ | |
1456 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1457 | } else { | |
1458 | umount_mnt(p); | |
1459 | } | |
7c4b93d8 | 1460 | } |
0f0afb1d | 1461 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1462 | } |
1463 | } | |
1464 | ||
b54b9be7 | 1465 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1466 | |
1ab59738 | 1467 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1468 | { |
1ab59738 | 1469 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1470 | int retval; |
1471 | ||
1ab59738 | 1472 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1473 | if (retval) |
1474 | return retval; | |
1475 | ||
1476 | /* | |
1477 | * Allow userspace to request a mountpoint be expired rather than | |
1478 | * unmounting unconditionally. Unmount only happens if: | |
1479 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1480 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1481 | */ | |
1482 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1483 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1484 | flags & (MNT_FORCE | MNT_DETACH)) |
1485 | return -EINVAL; | |
1486 | ||
b3e19d92 NP |
1487 | /* |
1488 | * probably don't strictly need the lock here if we examined | |
1489 | * all race cases, but it's a slowpath. | |
1490 | */ | |
719ea2fb | 1491 | lock_mount_hash(); |
83adc753 | 1492 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1493 | unlock_mount_hash(); |
1da177e4 | 1494 | return -EBUSY; |
b3e19d92 | 1495 | } |
719ea2fb | 1496 | unlock_mount_hash(); |
1da177e4 | 1497 | |
863d684f | 1498 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1499 | return -EAGAIN; |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * If we may have to abort operations to get out of this | |
1504 | * mount, and they will themselves hold resources we must | |
1505 | * allow the fs to do things. In the Unix tradition of | |
1506 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1507 | * might fail to complete on the first run through as other tasks | |
1508 | * must return, and the like. Thats for the mount program to worry | |
1509 | * about for the moment. | |
1510 | */ | |
1511 | ||
42faad99 | 1512 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1513 | sb->s_op->umount_begin(sb); |
42faad99 | 1514 | } |
1da177e4 LT |
1515 | |
1516 | /* | |
1517 | * No sense to grab the lock for this test, but test itself looks | |
1518 | * somewhat bogus. Suggestions for better replacement? | |
1519 | * Ho-hum... In principle, we might treat that as umount + switch | |
1520 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1521 | * Actually it makes sense, especially if rootfs would contain a | |
1522 | * /reboot - static binary that would close all descriptors and | |
1523 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1524 | */ | |
1ab59738 | 1525 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1526 | /* |
1527 | * Special case for "unmounting" root ... | |
1528 | * we just try to remount it readonly. | |
1529 | */ | |
bc6155d1 | 1530 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1531 | return -EPERM; |
1da177e4 | 1532 | down_write(&sb->s_umount); |
bc98a42c | 1533 | if (!sb_rdonly(sb)) |
e462ec50 | 1534 | retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); |
1da177e4 LT |
1535 | up_write(&sb->s_umount); |
1536 | return retval; | |
1537 | } | |
1538 | ||
97216be0 | 1539 | namespace_lock(); |
719ea2fb | 1540 | lock_mount_hash(); |
1da177e4 | 1541 | |
25d202ed EB |
1542 | /* Recheck MNT_LOCKED with the locks held */ |
1543 | retval = -EINVAL; | |
1544 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1545 | goto out; | |
1546 | ||
1547 | event++; | |
48a066e7 | 1548 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1549 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1550 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1551 | retval = 0; |
48a066e7 AV |
1552 | } else { |
1553 | shrink_submounts(mnt); | |
1554 | retval = -EBUSY; | |
1555 | if (!propagate_mount_busy(mnt, 2)) { | |
1556 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1557 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1558 | retval = 0; |
1559 | } | |
1da177e4 | 1560 | } |
25d202ed | 1561 | out: |
719ea2fb | 1562 | unlock_mount_hash(); |
e3197d83 | 1563 | namespace_unlock(); |
1da177e4 LT |
1564 | return retval; |
1565 | } | |
1566 | ||
80b5dce8 EB |
1567 | /* |
1568 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1569 | * | |
1570 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1571 | * to an existing mountpoint, and wind up leaking the mount. | |
1572 | * detach_mounts allows lazily unmounting those mounts instead of | |
1573 | * leaking them. | |
1574 | * | |
1575 | * The caller may hold dentry->d_inode->i_mutex. | |
1576 | */ | |
1577 | void __detach_mounts(struct dentry *dentry) | |
1578 | { | |
1579 | struct mountpoint *mp; | |
1580 | struct mount *mnt; | |
1581 | ||
1582 | namespace_lock(); | |
3895dbf8 | 1583 | lock_mount_hash(); |
80b5dce8 | 1584 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1585 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1586 | goto out_unlock; |
1587 | ||
e06b933e | 1588 | event++; |
80b5dce8 EB |
1589 | while (!hlist_empty(&mp->m_list)) { |
1590 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1591 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1592 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1593 | umount_mnt(mnt); | |
ce07d891 | 1594 | } |
e0c9c0af | 1595 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1596 | } |
80b5dce8 EB |
1597 | put_mountpoint(mp); |
1598 | out_unlock: | |
3895dbf8 | 1599 | unlock_mount_hash(); |
80b5dce8 EB |
1600 | namespace_unlock(); |
1601 | } | |
1602 | ||
dd111b31 | 1603 | /* |
9b40bc90 AV |
1604 | * Is the caller allowed to modify his namespace? |
1605 | */ | |
1606 | static inline bool may_mount(void) | |
1607 | { | |
1608 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1609 | } | |
1610 | ||
9e8925b6 JL |
1611 | static inline bool may_mandlock(void) |
1612 | { | |
1613 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1614 | return false; | |
1615 | #endif | |
95ace754 | 1616 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1617 | } |
1618 | ||
1da177e4 LT |
1619 | /* |
1620 | * Now umount can handle mount points as well as block devices. | |
1621 | * This is important for filesystems which use unnamed block devices. | |
1622 | * | |
1623 | * We now support a flag for forced unmount like the other 'big iron' | |
1624 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1625 | */ | |
1626 | ||
3a18ef5c | 1627 | int ksys_umount(char __user *name, int flags) |
1da177e4 | 1628 | { |
2d8f3038 | 1629 | struct path path; |
900148dc | 1630 | struct mount *mnt; |
1da177e4 | 1631 | int retval; |
db1f05bb | 1632 | int lookup_flags = 0; |
1da177e4 | 1633 | |
db1f05bb MS |
1634 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1635 | return -EINVAL; | |
1636 | ||
9b40bc90 AV |
1637 | if (!may_mount()) |
1638 | return -EPERM; | |
1639 | ||
db1f05bb MS |
1640 | if (!(flags & UMOUNT_NOFOLLOW)) |
1641 | lookup_flags |= LOOKUP_FOLLOW; | |
1642 | ||
197df04c | 1643 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1644 | if (retval) |
1645 | goto out; | |
900148dc | 1646 | mnt = real_mount(path.mnt); |
1da177e4 | 1647 | retval = -EINVAL; |
2d8f3038 | 1648 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1649 | goto dput_and_out; |
143c8c91 | 1650 | if (!check_mnt(mnt)) |
1da177e4 | 1651 | goto dput_and_out; |
25d202ed | 1652 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
5ff9d8a6 | 1653 | goto dput_and_out; |
b2f5d4dc EB |
1654 | retval = -EPERM; |
1655 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1656 | goto dput_and_out; | |
1da177e4 | 1657 | |
900148dc | 1658 | retval = do_umount(mnt, flags); |
1da177e4 | 1659 | dput_and_out: |
429731b1 | 1660 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1661 | dput(path.dentry); |
900148dc | 1662 | mntput_no_expire(mnt); |
1da177e4 LT |
1663 | out: |
1664 | return retval; | |
1665 | } | |
1666 | ||
3a18ef5c DB |
1667 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1668 | { | |
1669 | return ksys_umount(name, flags); | |
1670 | } | |
1671 | ||
1da177e4 LT |
1672 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1673 | ||
1674 | /* | |
b58fed8b | 1675 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1676 | */ |
bdc480e3 | 1677 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1678 | { |
3a18ef5c | 1679 | return ksys_umount(name, 0); |
1da177e4 LT |
1680 | } |
1681 | ||
1682 | #endif | |
1683 | ||
4ce5d2b1 | 1684 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1685 | { |
4ce5d2b1 | 1686 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1687 | return dentry->d_op == &ns_dentry_operations && |
1688 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1689 | } |
1690 | ||
58be2825 AV |
1691 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1692 | { | |
1693 | return container_of(ns, struct mnt_namespace, ns); | |
1694 | } | |
1695 | ||
4ce5d2b1 EB |
1696 | static bool mnt_ns_loop(struct dentry *dentry) |
1697 | { | |
1698 | /* Could bind mounting the mount namespace inode cause a | |
1699 | * mount namespace loop? | |
1700 | */ | |
1701 | struct mnt_namespace *mnt_ns; | |
1702 | if (!is_mnt_ns_file(dentry)) | |
1703 | return false; | |
1704 | ||
f77c8014 | 1705 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1706 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1707 | } | |
1708 | ||
87129cc0 | 1709 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1710 | int flag) |
1da177e4 | 1711 | { |
84d17192 | 1712 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1713 | |
4ce5d2b1 EB |
1714 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1715 | return ERR_PTR(-EINVAL); | |
1716 | ||
1717 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1718 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1719 | |
36341f64 | 1720 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1721 | if (IS_ERR(q)) |
1722 | return q; | |
1723 | ||
a73324da | 1724 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1725 | |
1726 | p = mnt; | |
6b41d536 | 1727 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1728 | struct mount *s; |
7ec02ef1 | 1729 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1730 | continue; |
1731 | ||
909b0a88 | 1732 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1733 | if (!(flag & CL_COPY_UNBINDABLE) && |
1734 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1735 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1736 | /* Both unbindable and locked. */ | |
1737 | q = ERR_PTR(-EPERM); | |
1738 | goto out; | |
1739 | } else { | |
1740 | s = skip_mnt_tree(s); | |
1741 | continue; | |
1742 | } | |
4ce5d2b1 EB |
1743 | } |
1744 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1745 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1746 | s = skip_mnt_tree(s); |
1747 | continue; | |
1748 | } | |
0714a533 AV |
1749 | while (p != s->mnt_parent) { |
1750 | p = p->mnt_parent; | |
1751 | q = q->mnt_parent; | |
1da177e4 | 1752 | } |
87129cc0 | 1753 | p = s; |
84d17192 | 1754 | parent = q; |
87129cc0 | 1755 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1756 | if (IS_ERR(q)) |
1757 | goto out; | |
719ea2fb | 1758 | lock_mount_hash(); |
1a4eeaf2 | 1759 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1760 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1761 | unlock_mount_hash(); |
1da177e4 LT |
1762 | } |
1763 | } | |
1764 | return res; | |
be34d1a3 | 1765 | out: |
1da177e4 | 1766 | if (res) { |
719ea2fb | 1767 | lock_mount_hash(); |
e819f152 | 1768 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1769 | unlock_mount_hash(); |
1da177e4 | 1770 | } |
be34d1a3 | 1771 | return q; |
1da177e4 LT |
1772 | } |
1773 | ||
be34d1a3 DH |
1774 | /* Caller should check returned pointer for errors */ |
1775 | ||
ca71cf71 | 1776 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1777 | { |
cb338d06 | 1778 | struct mount *tree; |
97216be0 | 1779 | namespace_lock(); |
cd4a4017 EB |
1780 | if (!check_mnt(real_mount(path->mnt))) |
1781 | tree = ERR_PTR(-EINVAL); | |
1782 | else | |
1783 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1784 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1785 | namespace_unlock(); |
be34d1a3 | 1786 | if (IS_ERR(tree)) |
52e220d3 | 1787 | return ERR_CAST(tree); |
be34d1a3 | 1788 | return &tree->mnt; |
8aec0809 AV |
1789 | } |
1790 | ||
1791 | void drop_collected_mounts(struct vfsmount *mnt) | |
1792 | { | |
97216be0 | 1793 | namespace_lock(); |
719ea2fb | 1794 | lock_mount_hash(); |
9c8e0a1b | 1795 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1796 | unlock_mount_hash(); |
3ab6abee | 1797 | namespace_unlock(); |
8aec0809 AV |
1798 | } |
1799 | ||
c771d683 MS |
1800 | /** |
1801 | * clone_private_mount - create a private clone of a path | |
1802 | * | |
1803 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1804 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1805 | * to the originating mount won't be propagated into this). | |
1806 | * | |
1807 | * Release with mntput(). | |
1808 | */ | |
ca71cf71 | 1809 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1810 | { |
1811 | struct mount *old_mnt = real_mount(path->mnt); | |
1812 | struct mount *new_mnt; | |
1813 | ||
1814 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1815 | return ERR_PTR(-EINVAL); | |
1816 | ||
c771d683 | 1817 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1818 | if (IS_ERR(new_mnt)) |
1819 | return ERR_CAST(new_mnt); | |
1820 | ||
1821 | return &new_mnt->mnt; | |
1822 | } | |
1823 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1824 | ||
1f707137 AV |
1825 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1826 | struct vfsmount *root) | |
1827 | { | |
1a4eeaf2 | 1828 | struct mount *mnt; |
1f707137 AV |
1829 | int res = f(root, arg); |
1830 | if (res) | |
1831 | return res; | |
1a4eeaf2 AV |
1832 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1833 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1834 | if (res) |
1835 | return res; | |
1836 | } | |
1837 | return 0; | |
1838 | } | |
1839 | ||
4b8b21f4 | 1840 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1841 | { |
315fc83e | 1842 | struct mount *p; |
719f5d7f | 1843 | |
909b0a88 | 1844 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1845 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1846 | mnt_release_group_id(p); |
719f5d7f MS |
1847 | } |
1848 | } | |
1849 | ||
4b8b21f4 | 1850 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1851 | { |
315fc83e | 1852 | struct mount *p; |
719f5d7f | 1853 | |
909b0a88 | 1854 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1855 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1856 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1857 | if (err) { |
4b8b21f4 | 1858 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1859 | return err; |
1860 | } | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | return 0; | |
1865 | } | |
1866 | ||
d2921684 EB |
1867 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1868 | { | |
1869 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1870 | unsigned int mounts = 0, old, pending, sum; | |
1871 | struct mount *p; | |
1872 | ||
1873 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1874 | mounts++; | |
1875 | ||
1876 | old = ns->mounts; | |
1877 | pending = ns->pending_mounts; | |
1878 | sum = old + pending; | |
1879 | if ((old > sum) || | |
1880 | (pending > sum) || | |
1881 | (max < sum) || | |
1882 | (mounts > (max - sum))) | |
1883 | return -ENOSPC; | |
1884 | ||
1885 | ns->pending_mounts = pending + mounts; | |
1886 | return 0; | |
1887 | } | |
1888 | ||
b90fa9ae RP |
1889 | /* |
1890 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1891 | * @nd : place the mount tree @source_mnt is attached |
1892 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1893 | * store the parent mount and mountpoint dentry. | |
1894 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1895 | * |
1896 | * NOTE: in the table below explains the semantics when a source mount | |
1897 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1898 | * --------------------------------------------------------------------------- |
1899 | * | BIND MOUNT OPERATION | | |
1900 | * |************************************************************************** | |
1901 | * | source-->| shared | private | slave | unbindable | | |
1902 | * | dest | | | | | | |
1903 | * | | | | | | | | |
1904 | * | v | | | | | | |
1905 | * |************************************************************************** | |
1906 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1907 | * | | | | | | | |
1908 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1909 | * *************************************************************************** | |
b90fa9ae RP |
1910 | * A bind operation clones the source mount and mounts the clone on the |
1911 | * destination mount. | |
1912 | * | |
1913 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1914 | * tree of the destination mount and the cloned mount is added to | |
1915 | * the peer group of the source mount. | |
1916 | * (+) the cloned mount is created under the destination mount and is marked | |
1917 | * as shared. The cloned mount is added to the peer group of the source | |
1918 | * mount. | |
5afe0022 RP |
1919 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1920 | * of the destination mount and the cloned mount is made slave | |
1921 | * of the same master as that of the source mount. The cloned mount | |
1922 | * is marked as 'shared and slave'. | |
1923 | * (*) the cloned mount is made a slave of the same master as that of the | |
1924 | * source mount. | |
1925 | * | |
9676f0c6 RP |
1926 | * --------------------------------------------------------------------------- |
1927 | * | MOVE MOUNT OPERATION | | |
1928 | * |************************************************************************** | |
1929 | * | source-->| shared | private | slave | unbindable | | |
1930 | * | dest | | | | | | |
1931 | * | | | | | | | | |
1932 | * | v | | | | | | |
1933 | * |************************************************************************** | |
1934 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1935 | * | | | | | | | |
1936 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1937 | * *************************************************************************** | |
5afe0022 RP |
1938 | * |
1939 | * (+) the mount is moved to the destination. And is then propagated to | |
1940 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1941 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1942 | * (+++) the mount is moved to the destination and is then propagated to |
1943 | * all the mounts belonging to the destination mount's propagation tree. | |
1944 | * the mount is marked as 'shared and slave'. | |
1945 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1946 | * |
1947 | * if the source mount is a tree, the operations explained above is | |
1948 | * applied to each mount in the tree. | |
1949 | * Must be called without spinlocks held, since this function can sleep | |
1950 | * in allocations. | |
1951 | */ | |
0fb54e50 | 1952 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1953 | struct mount *dest_mnt, |
1954 | struct mountpoint *dest_mp, | |
1955 | struct path *parent_path) | |
b90fa9ae | 1956 | { |
38129a13 | 1957 | HLIST_HEAD(tree_list); |
d2921684 | 1958 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 1959 | struct mountpoint *smp; |
315fc83e | 1960 | struct mount *child, *p; |
38129a13 | 1961 | struct hlist_node *n; |
719f5d7f | 1962 | int err; |
b90fa9ae | 1963 | |
1064f874 EB |
1964 | /* Preallocate a mountpoint in case the new mounts need |
1965 | * to be tucked under other mounts. | |
1966 | */ | |
1967 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
1968 | if (IS_ERR(smp)) | |
1969 | return PTR_ERR(smp); | |
1970 | ||
d2921684 EB |
1971 | /* Is there space to add these mounts to the mount namespace? */ |
1972 | if (!parent_path) { | |
1973 | err = count_mounts(ns, source_mnt); | |
1974 | if (err) | |
1975 | goto out; | |
1976 | } | |
1977 | ||
fc7be130 | 1978 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 1979 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
1980 | if (err) |
1981 | goto out; | |
0b1b901b | 1982 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 1983 | lock_mount_hash(); |
0b1b901b AV |
1984 | if (err) |
1985 | goto out_cleanup_ids; | |
909b0a88 | 1986 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 1987 | set_mnt_shared(p); |
0b1b901b AV |
1988 | } else { |
1989 | lock_mount_hash(); | |
b90fa9ae | 1990 | } |
1a390689 | 1991 | if (parent_path) { |
0fb54e50 | 1992 | detach_mnt(source_mnt, parent_path); |
84d17192 | 1993 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 1994 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 1995 | } else { |
84d17192 | 1996 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 1997 | commit_tree(source_mnt); |
21444403 | 1998 | } |
b90fa9ae | 1999 | |
38129a13 | 2000 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2001 | struct mount *q; |
38129a13 | 2002 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2003 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2004 | child->mnt_mountpoint); | |
2005 | if (q) | |
2006 | mnt_change_mountpoint(child, smp, q); | |
2007 | commit_tree(child); | |
b90fa9ae | 2008 | } |
1064f874 | 2009 | put_mountpoint(smp); |
719ea2fb | 2010 | unlock_mount_hash(); |
99b7db7b | 2011 | |
b90fa9ae | 2012 | return 0; |
719f5d7f MS |
2013 | |
2014 | out_cleanup_ids: | |
f2ebb3a9 AV |
2015 | while (!hlist_empty(&tree_list)) { |
2016 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2017 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2018 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2019 | } |
2020 | unlock_mount_hash(); | |
0b1b901b | 2021 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2022 | out: |
d2921684 | 2023 | ns->pending_mounts = 0; |
1064f874 EB |
2024 | |
2025 | read_seqlock_excl(&mount_lock); | |
2026 | put_mountpoint(smp); | |
2027 | read_sequnlock_excl(&mount_lock); | |
2028 | ||
719f5d7f | 2029 | return err; |
b90fa9ae RP |
2030 | } |
2031 | ||
84d17192 | 2032 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2033 | { |
2034 | struct vfsmount *mnt; | |
84d17192 | 2035 | struct dentry *dentry = path->dentry; |
b12cea91 | 2036 | retry: |
5955102c | 2037 | inode_lock(dentry->d_inode); |
84d17192 | 2038 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2039 | inode_unlock(dentry->d_inode); |
84d17192 | 2040 | return ERR_PTR(-ENOENT); |
b12cea91 | 2041 | } |
97216be0 | 2042 | namespace_lock(); |
b12cea91 | 2043 | mnt = lookup_mnt(path); |
84d17192 | 2044 | if (likely(!mnt)) { |
3895dbf8 | 2045 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2046 | if (IS_ERR(mp)) { |
97216be0 | 2047 | namespace_unlock(); |
5955102c | 2048 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2049 | return mp; |
2050 | } | |
2051 | return mp; | |
2052 | } | |
97216be0 | 2053 | namespace_unlock(); |
5955102c | 2054 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2055 | path_put(path); |
2056 | path->mnt = mnt; | |
84d17192 | 2057 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2058 | goto retry; |
2059 | } | |
2060 | ||
84d17192 | 2061 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2062 | { |
84d17192 | 2063 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2064 | |
2065 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2066 | put_mountpoint(where); |
3895dbf8 EB |
2067 | read_sequnlock_excl(&mount_lock); |
2068 | ||
328e6d90 | 2069 | namespace_unlock(); |
5955102c | 2070 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2071 | } |
2072 | ||
84d17192 | 2073 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2074 | { |
e462ec50 | 2075 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2076 | return -EINVAL; |
2077 | ||
e36cb0b8 DH |
2078 | if (d_is_dir(mp->m_dentry) != |
2079 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2080 | return -ENOTDIR; |
2081 | ||
84d17192 | 2082 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2083 | } |
2084 | ||
7a2e8a8f VA |
2085 | /* |
2086 | * Sanity check the flags to change_mnt_propagation. | |
2087 | */ | |
2088 | ||
e462ec50 | 2089 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2090 | { |
e462ec50 | 2091 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2092 | |
2093 | /* Fail if any non-propagation flags are set */ | |
2094 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2095 | return 0; | |
2096 | /* Only one propagation flag should be set */ | |
2097 | if (!is_power_of_2(type)) | |
2098 | return 0; | |
2099 | return type; | |
2100 | } | |
2101 | ||
07b20889 RP |
2102 | /* |
2103 | * recursively change the type of the mountpoint. | |
2104 | */ | |
e462ec50 | 2105 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2106 | { |
315fc83e | 2107 | struct mount *m; |
4b8b21f4 | 2108 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2109 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2110 | int type; |
719f5d7f | 2111 | int err = 0; |
07b20889 | 2112 | |
2d92ab3c | 2113 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2114 | return -EINVAL; |
2115 | ||
e462ec50 | 2116 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2117 | if (!type) |
2118 | return -EINVAL; | |
2119 | ||
97216be0 | 2120 | namespace_lock(); |
719f5d7f MS |
2121 | if (type == MS_SHARED) { |
2122 | err = invent_group_ids(mnt, recurse); | |
2123 | if (err) | |
2124 | goto out_unlock; | |
2125 | } | |
2126 | ||
719ea2fb | 2127 | lock_mount_hash(); |
909b0a88 | 2128 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2129 | change_mnt_propagation(m, type); |
719ea2fb | 2130 | unlock_mount_hash(); |
719f5d7f MS |
2131 | |
2132 | out_unlock: | |
97216be0 | 2133 | namespace_unlock(); |
719f5d7f | 2134 | return err; |
07b20889 RP |
2135 | } |
2136 | ||
5ff9d8a6 EB |
2137 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2138 | { | |
2139 | struct mount *child; | |
2140 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2141 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2142 | continue; | |
2143 | ||
2144 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2145 | return true; | |
2146 | } | |
2147 | return false; | |
2148 | } | |
2149 | ||
1da177e4 LT |
2150 | /* |
2151 | * do loopback mount. | |
2152 | */ | |
808d4e3c | 2153 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2154 | int recurse) |
1da177e4 | 2155 | { |
2d92ab3c | 2156 | struct path old_path; |
84d17192 AV |
2157 | struct mount *mnt = NULL, *old, *parent; |
2158 | struct mountpoint *mp; | |
57eccb83 | 2159 | int err; |
1da177e4 LT |
2160 | if (!old_name || !*old_name) |
2161 | return -EINVAL; | |
815d405c | 2162 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2163 | if (err) |
2164 | return err; | |
2165 | ||
8823c079 | 2166 | err = -EINVAL; |
4ce5d2b1 | 2167 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2168 | goto out; |
8823c079 | 2169 | |
84d17192 AV |
2170 | mp = lock_mount(path); |
2171 | err = PTR_ERR(mp); | |
2172 | if (IS_ERR(mp)) | |
b12cea91 AV |
2173 | goto out; |
2174 | ||
87129cc0 | 2175 | old = real_mount(old_path.mnt); |
84d17192 | 2176 | parent = real_mount(path->mnt); |
87129cc0 | 2177 | |
1da177e4 | 2178 | err = -EINVAL; |
fc7be130 | 2179 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2180 | goto out2; |
9676f0c6 | 2181 | |
e149ed2b AV |
2182 | if (!check_mnt(parent)) |
2183 | goto out2; | |
2184 | ||
2185 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2186 | goto out2; |
1da177e4 | 2187 | |
5ff9d8a6 EB |
2188 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2189 | goto out2; | |
2190 | ||
ccd48bc7 | 2191 | if (recurse) |
4ce5d2b1 | 2192 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2193 | else |
87129cc0 | 2194 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2195 | |
be34d1a3 DH |
2196 | if (IS_ERR(mnt)) { |
2197 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2198 | goto out2; |
be34d1a3 | 2199 | } |
ccd48bc7 | 2200 | |
5ff9d8a6 EB |
2201 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2202 | ||
84d17192 | 2203 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2204 | if (err) { |
719ea2fb | 2205 | lock_mount_hash(); |
e819f152 | 2206 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2207 | unlock_mount_hash(); |
5b83d2c5 | 2208 | } |
b12cea91 | 2209 | out2: |
84d17192 | 2210 | unlock_mount(mp); |
ccd48bc7 | 2211 | out: |
2d92ab3c | 2212 | path_put(&old_path); |
1da177e4 LT |
2213 | return err; |
2214 | } | |
2215 | ||
43f5e655 DH |
2216 | /* |
2217 | * Don't allow locked mount flags to be cleared. | |
2218 | * | |
2219 | * No locks need to be held here while testing the various MNT_LOCK | |
2220 | * flags because those flags can never be cleared once they are set. | |
2221 | */ | |
2222 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2223 | { |
43f5e655 DH |
2224 | unsigned int fl = mnt->mnt.mnt_flags; |
2225 | ||
2226 | if ((fl & MNT_LOCK_READONLY) && | |
2227 | !(mnt_flags & MNT_READONLY)) | |
2228 | return false; | |
2229 | ||
2230 | if ((fl & MNT_LOCK_NODEV) && | |
2231 | !(mnt_flags & MNT_NODEV)) | |
2232 | return false; | |
2233 | ||
2234 | if ((fl & MNT_LOCK_NOSUID) && | |
2235 | !(mnt_flags & MNT_NOSUID)) | |
2236 | return false; | |
2237 | ||
2238 | if ((fl & MNT_LOCK_NOEXEC) && | |
2239 | !(mnt_flags & MNT_NOEXEC)) | |
2240 | return false; | |
2241 | ||
2242 | if ((fl & MNT_LOCK_ATIME) && | |
2243 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2244 | return false; | |
2e4b7fcd | 2245 | |
43f5e655 DH |
2246 | return true; |
2247 | } | |
2248 | ||
2249 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2250 | { |
43f5e655 | 2251 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2252 | |
43f5e655 | 2253 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2254 | return 0; |
2255 | ||
2256 | if (readonly_request) | |
43f5e655 DH |
2257 | return mnt_make_readonly(mnt); |
2258 | ||
2259 | return __mnt_unmake_readonly(mnt); | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * Update the user-settable attributes on a mount. The caller must hold | |
2264 | * sb->s_umount for writing. | |
2265 | */ | |
2266 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) | |
2267 | { | |
2268 | lock_mount_hash(); | |
2269 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; | |
2270 | mnt->mnt.mnt_flags = mnt_flags; | |
2271 | touch_mnt_namespace(mnt->mnt_ns); | |
2272 | unlock_mount_hash(); | |
2273 | } | |
2274 | ||
2275 | /* | |
2276 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2277 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2278 | * to mount(2). | |
2279 | */ | |
2280 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2281 | { | |
2282 | struct super_block *sb = path->mnt->mnt_sb; | |
2283 | struct mount *mnt = real_mount(path->mnt); | |
2284 | int ret; | |
2285 | ||
2286 | if (!check_mnt(mnt)) | |
2287 | return -EINVAL; | |
2288 | ||
2289 | if (path->dentry != mnt->mnt.mnt_root) | |
2290 | return -EINVAL; | |
2291 | ||
2292 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2293 | return -EPERM; | |
2294 | ||
2295 | down_write(&sb->s_umount); | |
2296 | ret = change_mount_ro_state(mnt, mnt_flags); | |
2297 | if (ret == 0) | |
2298 | set_mount_attributes(mnt, mnt_flags); | |
2299 | up_write(&sb->s_umount); | |
2300 | return ret; | |
2e4b7fcd DH |
2301 | } |
2302 | ||
1da177e4 LT |
2303 | /* |
2304 | * change filesystem flags. dir should be a physical root of filesystem. | |
2305 | * If you've mounted a non-root directory somewhere and want to do remount | |
2306 | * on it - tough luck. | |
2307 | */ | |
e462ec50 DH |
2308 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2309 | int mnt_flags, void *data) | |
1da177e4 LT |
2310 | { |
2311 | int err; | |
2d92ab3c | 2312 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2313 | struct mount *mnt = real_mount(path->mnt); |
204cc0cc | 2314 | void *sec_opts = NULL; |
1da177e4 | 2315 | |
143c8c91 | 2316 | if (!check_mnt(mnt)) |
1da177e4 LT |
2317 | return -EINVAL; |
2318 | ||
2d92ab3c | 2319 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2320 | return -EINVAL; |
2321 | ||
43f5e655 | 2322 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2323 | return -EPERM; |
9566d674 | 2324 | |
c039bc3c | 2325 | if (data && !(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)) { |
204cc0cc | 2326 | err = security_sb_eat_lsm_opts(data, &sec_opts); |
c039bc3c AV |
2327 | if (err) |
2328 | return err; | |
2329 | } | |
204cc0cc AV |
2330 | err = security_sb_remount(sb, sec_opts); |
2331 | security_free_mnt_opts(&sec_opts); | |
ff36fe2c EP |
2332 | if (err) |
2333 | return err; | |
2334 | ||
1da177e4 | 2335 | down_write(&sb->s_umount); |
43f5e655 DH |
2336 | err = -EPERM; |
2337 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
e462ec50 | 2338 | err = do_remount_sb(sb, sb_flags, data, 0); |
43f5e655 DH |
2339 | if (!err) |
2340 | set_mount_attributes(mnt, mnt_flags); | |
0e55a7cc | 2341 | } |
6339dab8 | 2342 | up_write(&sb->s_umount); |
1da177e4 LT |
2343 | return err; |
2344 | } | |
2345 | ||
cbbe362c | 2346 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2347 | { |
315fc83e | 2348 | struct mount *p; |
909b0a88 | 2349 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2350 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2351 | return 1; |
2352 | } | |
2353 | return 0; | |
2354 | } | |
2355 | ||
808d4e3c | 2356 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2357 | { |
2d92ab3c | 2358 | struct path old_path, parent_path; |
676da58d | 2359 | struct mount *p; |
0fb54e50 | 2360 | struct mount *old; |
84d17192 | 2361 | struct mountpoint *mp; |
57eccb83 | 2362 | int err; |
1da177e4 LT |
2363 | if (!old_name || !*old_name) |
2364 | return -EINVAL; | |
2d92ab3c | 2365 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2366 | if (err) |
2367 | return err; | |
2368 | ||
84d17192 AV |
2369 | mp = lock_mount(path); |
2370 | err = PTR_ERR(mp); | |
2371 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2372 | goto out; |
2373 | ||
143c8c91 | 2374 | old = real_mount(old_path.mnt); |
fc7be130 | 2375 | p = real_mount(path->mnt); |
143c8c91 | 2376 | |
1da177e4 | 2377 | err = -EINVAL; |
fc7be130 | 2378 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2379 | goto out1; |
2380 | ||
5ff9d8a6 EB |
2381 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2382 | goto out1; | |
2383 | ||
1da177e4 | 2384 | err = -EINVAL; |
2d92ab3c | 2385 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2386 | goto out1; |
1da177e4 | 2387 | |
676da58d | 2388 | if (!mnt_has_parent(old)) |
21444403 | 2389 | goto out1; |
1da177e4 | 2390 | |
e36cb0b8 DH |
2391 | if (d_is_dir(path->dentry) != |
2392 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2393 | goto out1; |
2394 | /* | |
2395 | * Don't move a mount residing in a shared parent. | |
2396 | */ | |
fc7be130 | 2397 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2398 | goto out1; |
9676f0c6 RP |
2399 | /* |
2400 | * Don't move a mount tree containing unbindable mounts to a destination | |
2401 | * mount which is shared. | |
2402 | */ | |
fc7be130 | 2403 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2404 | goto out1; |
1da177e4 | 2405 | err = -ELOOP; |
fc7be130 | 2406 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2407 | if (p == old) |
21444403 | 2408 | goto out1; |
1da177e4 | 2409 | |
84d17192 | 2410 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2411 | if (err) |
21444403 | 2412 | goto out1; |
1da177e4 LT |
2413 | |
2414 | /* if the mount is moved, it should no longer be expire | |
2415 | * automatically */ | |
6776db3d | 2416 | list_del_init(&old->mnt_expire); |
1da177e4 | 2417 | out1: |
84d17192 | 2418 | unlock_mount(mp); |
1da177e4 | 2419 | out: |
1da177e4 | 2420 | if (!err) |
1a390689 | 2421 | path_put(&parent_path); |
2d92ab3c | 2422 | path_put(&old_path); |
1da177e4 LT |
2423 | return err; |
2424 | } | |
2425 | ||
9d412a43 AV |
2426 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2427 | { | |
2428 | int err; | |
2429 | const char *subtype = strchr(fstype, '.'); | |
2430 | if (subtype) { | |
2431 | subtype++; | |
2432 | err = -EINVAL; | |
2433 | if (!subtype[0]) | |
2434 | goto err; | |
2435 | } else | |
2436 | subtype = ""; | |
2437 | ||
2438 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2439 | err = -ENOMEM; | |
2440 | if (!mnt->mnt_sb->s_subtype) | |
2441 | goto err; | |
2442 | return mnt; | |
2443 | ||
2444 | err: | |
2445 | mntput(mnt); | |
2446 | return ERR_PTR(err); | |
2447 | } | |
2448 | ||
9d412a43 AV |
2449 | /* |
2450 | * add a mount into a namespace's mount tree | |
2451 | */ | |
95bc5f25 | 2452 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2453 | { |
84d17192 AV |
2454 | struct mountpoint *mp; |
2455 | struct mount *parent; | |
9d412a43 AV |
2456 | int err; |
2457 | ||
f2ebb3a9 | 2458 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2459 | |
84d17192 AV |
2460 | mp = lock_mount(path); |
2461 | if (IS_ERR(mp)) | |
2462 | return PTR_ERR(mp); | |
9d412a43 | 2463 | |
84d17192 | 2464 | parent = real_mount(path->mnt); |
9d412a43 | 2465 | err = -EINVAL; |
84d17192 | 2466 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2467 | /* that's acceptable only for automounts done in private ns */ |
2468 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2469 | goto unlock; | |
2470 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2471 | if (!parent->mnt_ns) |
156cacb1 AV |
2472 | goto unlock; |
2473 | } | |
9d412a43 AV |
2474 | |
2475 | /* Refuse the same filesystem on the same mount point */ | |
2476 | err = -EBUSY; | |
95bc5f25 | 2477 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2478 | path->mnt->mnt_root == path->dentry) |
2479 | goto unlock; | |
2480 | ||
2481 | err = -EINVAL; | |
e36cb0b8 | 2482 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2483 | goto unlock; |
2484 | ||
95bc5f25 | 2485 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2486 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2487 | |
2488 | unlock: | |
84d17192 | 2489 | unlock_mount(mp); |
9d412a43 AV |
2490 | return err; |
2491 | } | |
b1e75df4 | 2492 | |
8654df4e | 2493 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2494 | |
1da177e4 LT |
2495 | /* |
2496 | * create a new mount for userspace and request it to be added into the | |
2497 | * namespace's tree | |
2498 | */ | |
e462ec50 | 2499 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2500 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2501 | { |
0c55cfc4 | 2502 | struct file_system_type *type; |
1da177e4 | 2503 | struct vfsmount *mnt; |
15f9a3f3 | 2504 | int err; |
1da177e4 | 2505 | |
0c55cfc4 | 2506 | if (!fstype) |
1da177e4 LT |
2507 | return -EINVAL; |
2508 | ||
0c55cfc4 EB |
2509 | type = get_fs_type(fstype); |
2510 | if (!type) | |
2511 | return -ENODEV; | |
2512 | ||
e462ec50 | 2513 | mnt = vfs_kern_mount(type, sb_flags, name, data); |
0c55cfc4 EB |
2514 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
2515 | !mnt->mnt_sb->s_subtype) | |
2516 | mnt = fs_set_subtype(mnt, fstype); | |
2517 | ||
2518 | put_filesystem(type); | |
1da177e4 LT |
2519 | if (IS_ERR(mnt)) |
2520 | return PTR_ERR(mnt); | |
2521 | ||
8654df4e EB |
2522 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2523 | mntput(mnt); | |
2524 | return -EPERM; | |
2525 | } | |
2526 | ||
95bc5f25 | 2527 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2528 | if (err) |
2529 | mntput(mnt); | |
2530 | return err; | |
1da177e4 LT |
2531 | } |
2532 | ||
19a167af AV |
2533 | int finish_automount(struct vfsmount *m, struct path *path) |
2534 | { | |
6776db3d | 2535 | struct mount *mnt = real_mount(m); |
19a167af AV |
2536 | int err; |
2537 | /* The new mount record should have at least 2 refs to prevent it being | |
2538 | * expired before we get a chance to add it | |
2539 | */ | |
6776db3d | 2540 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2541 | |
2542 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2543 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2544 | err = -ELOOP; |
2545 | goto fail; | |
19a167af AV |
2546 | } |
2547 | ||
95bc5f25 | 2548 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2549 | if (!err) |
2550 | return 0; | |
2551 | fail: | |
2552 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2553 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2554 | namespace_lock(); |
6776db3d | 2555 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2556 | namespace_unlock(); |
19a167af | 2557 | } |
b1e75df4 AV |
2558 | mntput(m); |
2559 | mntput(m); | |
19a167af AV |
2560 | return err; |
2561 | } | |
2562 | ||
ea5b778a DH |
2563 | /** |
2564 | * mnt_set_expiry - Put a mount on an expiration list | |
2565 | * @mnt: The mount to list. | |
2566 | * @expiry_list: The list to add the mount to. | |
2567 | */ | |
2568 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2569 | { | |
97216be0 | 2570 | namespace_lock(); |
ea5b778a | 2571 | |
6776db3d | 2572 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2573 | |
97216be0 | 2574 | namespace_unlock(); |
ea5b778a DH |
2575 | } |
2576 | EXPORT_SYMBOL(mnt_set_expiry); | |
2577 | ||
1da177e4 LT |
2578 | /* |
2579 | * process a list of expirable mountpoints with the intent of discarding any | |
2580 | * mountpoints that aren't in use and haven't been touched since last we came | |
2581 | * here | |
2582 | */ | |
2583 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2584 | { | |
761d5c38 | 2585 | struct mount *mnt, *next; |
1da177e4 LT |
2586 | LIST_HEAD(graveyard); |
2587 | ||
2588 | if (list_empty(mounts)) | |
2589 | return; | |
2590 | ||
97216be0 | 2591 | namespace_lock(); |
719ea2fb | 2592 | lock_mount_hash(); |
1da177e4 LT |
2593 | |
2594 | /* extract from the expiration list every vfsmount that matches the | |
2595 | * following criteria: | |
2596 | * - only referenced by its parent vfsmount | |
2597 | * - still marked for expiry (marked on the last call here; marks are | |
2598 | * cleared by mntput()) | |
2599 | */ | |
6776db3d | 2600 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2601 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2602 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2603 | continue; |
6776db3d | 2604 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2605 | } |
bcc5c7d2 | 2606 | while (!list_empty(&graveyard)) { |
6776db3d | 2607 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2608 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2609 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2610 | } |
719ea2fb | 2611 | unlock_mount_hash(); |
3ab6abee | 2612 | namespace_unlock(); |
5528f911 TM |
2613 | } |
2614 | ||
2615 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2616 | ||
2617 | /* | |
2618 | * Ripoff of 'select_parent()' | |
2619 | * | |
2620 | * search the list of submounts for a given mountpoint, and move any | |
2621 | * shrinkable submounts to the 'graveyard' list. | |
2622 | */ | |
692afc31 | 2623 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2624 | { |
692afc31 | 2625 | struct mount *this_parent = parent; |
5528f911 TM |
2626 | struct list_head *next; |
2627 | int found = 0; | |
2628 | ||
2629 | repeat: | |
6b41d536 | 2630 | next = this_parent->mnt_mounts.next; |
5528f911 | 2631 | resume: |
6b41d536 | 2632 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2633 | struct list_head *tmp = next; |
6b41d536 | 2634 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2635 | |
2636 | next = tmp->next; | |
692afc31 | 2637 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2638 | continue; |
5528f911 TM |
2639 | /* |
2640 | * Descend a level if the d_mounts list is non-empty. | |
2641 | */ | |
6b41d536 | 2642 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2643 | this_parent = mnt; |
2644 | goto repeat; | |
2645 | } | |
1da177e4 | 2646 | |
1ab59738 | 2647 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2648 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2649 | found++; |
2650 | } | |
1da177e4 | 2651 | } |
5528f911 TM |
2652 | /* |
2653 | * All done at this level ... ascend and resume the search | |
2654 | */ | |
2655 | if (this_parent != parent) { | |
6b41d536 | 2656 | next = this_parent->mnt_child.next; |
0714a533 | 2657 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2658 | goto resume; |
2659 | } | |
2660 | return found; | |
2661 | } | |
2662 | ||
2663 | /* | |
2664 | * process a list of expirable mountpoints with the intent of discarding any | |
2665 | * submounts of a specific parent mountpoint | |
99b7db7b | 2666 | * |
48a066e7 | 2667 | * mount_lock must be held for write |
5528f911 | 2668 | */ |
b54b9be7 | 2669 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2670 | { |
2671 | LIST_HEAD(graveyard); | |
761d5c38 | 2672 | struct mount *m; |
5528f911 | 2673 | |
5528f911 | 2674 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2675 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2676 | while (!list_empty(&graveyard)) { |
761d5c38 | 2677 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2678 | mnt_expire); |
143c8c91 | 2679 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2680 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2681 | } |
2682 | } | |
1da177e4 LT |
2683 | } |
2684 | ||
1da177e4 LT |
2685 | /* |
2686 | * Some copy_from_user() implementations do not return the exact number of | |
2687 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2688 | * Note that this function differs from copy_from_user() in that it will oops | |
2689 | * on bad values of `to', rather than returning a short copy. | |
2690 | */ | |
b58fed8b RP |
2691 | static long exact_copy_from_user(void *to, const void __user * from, |
2692 | unsigned long n) | |
1da177e4 LT |
2693 | { |
2694 | char *t = to; | |
2695 | const char __user *f = from; | |
2696 | char c; | |
2697 | ||
96d4f267 | 2698 | if (!access_ok(from, n)) |
1da177e4 LT |
2699 | return n; |
2700 | ||
9da3f2b7 | 2701 | current->kernel_uaccess_faults_ok++; |
1da177e4 LT |
2702 | while (n) { |
2703 | if (__get_user(c, f)) { | |
2704 | memset(t, 0, n); | |
2705 | break; | |
2706 | } | |
2707 | *t++ = c; | |
2708 | f++; | |
2709 | n--; | |
2710 | } | |
9da3f2b7 | 2711 | current->kernel_uaccess_faults_ok--; |
1da177e4 LT |
2712 | return n; |
2713 | } | |
2714 | ||
b40ef869 | 2715 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2716 | { |
2717 | int i; | |
1da177e4 | 2718 | unsigned long size; |
b40ef869 | 2719 | char *copy; |
b58fed8b | 2720 | |
1da177e4 | 2721 | if (!data) |
b40ef869 | 2722 | return NULL; |
1da177e4 | 2723 | |
b40ef869 AV |
2724 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2725 | if (!copy) | |
2726 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2727 | |
2728 | /* We only care that *some* data at the address the user | |
2729 | * gave us is valid. Just in case, we'll zero | |
2730 | * the remainder of the page. | |
2731 | */ | |
2732 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2733 | size = TASK_SIZE - (unsigned long)data; | |
2734 | if (size > PAGE_SIZE) | |
2735 | size = PAGE_SIZE; | |
2736 | ||
b40ef869 | 2737 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2738 | if (!i) { |
b40ef869 AV |
2739 | kfree(copy); |
2740 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2741 | } |
2742 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2743 | memset(copy + i, 0, PAGE_SIZE - i); |
2744 | return copy; | |
1da177e4 LT |
2745 | } |
2746 | ||
b8850d1f | 2747 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2748 | { |
b8850d1f | 2749 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2750 | } |
2751 | ||
1da177e4 LT |
2752 | /* |
2753 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2754 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2755 | * | |
2756 | * data is a (void *) that can point to any structure up to | |
2757 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2758 | * information (or be NULL). | |
2759 | * | |
2760 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2761 | * When the flags word was introduced its top half was required | |
2762 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2763 | * Therefore, if this magic number is present, it carries no information | |
2764 | * and must be discarded. | |
2765 | */ | |
5e6123f3 | 2766 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2767 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2768 | { |
2d92ab3c | 2769 | struct path path; |
e462ec50 | 2770 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2771 | int retval = 0; |
1da177e4 LT |
2772 | |
2773 | /* Discard magic */ | |
2774 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2775 | flags &= ~MS_MGC_MSK; | |
2776 | ||
2777 | /* Basic sanity checks */ | |
1da177e4 LT |
2778 | if (data_page) |
2779 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2780 | ||
e462ec50 DH |
2781 | if (flags & MS_NOUSER) |
2782 | return -EINVAL; | |
2783 | ||
a27ab9f2 | 2784 | /* ... and get the mountpoint */ |
5e6123f3 | 2785 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2786 | if (retval) |
2787 | return retval; | |
2788 | ||
2789 | retval = security_sb_mount(dev_name, &path, | |
2790 | type_page, flags, data_page); | |
0d5cadb8 AV |
2791 | if (!retval && !may_mount()) |
2792 | retval = -EPERM; | |
e462ec50 | 2793 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 2794 | retval = -EPERM; |
a27ab9f2 TH |
2795 | if (retval) |
2796 | goto dput_out; | |
2797 | ||
613cbe3d AK |
2798 | /* Default to relatime unless overriden */ |
2799 | if (!(flags & MS_NOATIME)) | |
2800 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2801 | |
1da177e4 LT |
2802 | /* Separate the per-mountpoint flags */ |
2803 | if (flags & MS_NOSUID) | |
2804 | mnt_flags |= MNT_NOSUID; | |
2805 | if (flags & MS_NODEV) | |
2806 | mnt_flags |= MNT_NODEV; | |
2807 | if (flags & MS_NOEXEC) | |
2808 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2809 | if (flags & MS_NOATIME) |
2810 | mnt_flags |= MNT_NOATIME; | |
2811 | if (flags & MS_NODIRATIME) | |
2812 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2813 | if (flags & MS_STRICTATIME) |
2814 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 2815 | if (flags & MS_RDONLY) |
2e4b7fcd | 2816 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 2817 | |
ffbc6f0e EB |
2818 | /* The default atime for remount is preservation */ |
2819 | if ((flags & MS_REMOUNT) && | |
2820 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2821 | MS_STRICTATIME)) == 0)) { | |
2822 | mnt_flags &= ~MNT_ATIME_MASK; | |
2823 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2824 | } | |
2825 | ||
e462ec50 DH |
2826 | sb_flags = flags & (SB_RDONLY | |
2827 | SB_SYNCHRONOUS | | |
2828 | SB_MANDLOCK | | |
2829 | SB_DIRSYNC | | |
2830 | SB_SILENT | | |
917086ff | 2831 | SB_POSIXACL | |
d7ee9469 | 2832 | SB_LAZYTIME | |
917086ff | 2833 | SB_I_VERSION); |
1da177e4 | 2834 | |
43f5e655 DH |
2835 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
2836 | retval = do_reconfigure_mnt(&path, mnt_flags); | |
2837 | else if (flags & MS_REMOUNT) | |
e462ec50 | 2838 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
2839 | data_page); |
2840 | else if (flags & MS_BIND) | |
2d92ab3c | 2841 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2842 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2843 | retval = do_change_type(&path, flags); |
1da177e4 | 2844 | else if (flags & MS_MOVE) |
2d92ab3c | 2845 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2846 | else |
e462ec50 | 2847 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
2848 | dev_name, data_page); |
2849 | dput_out: | |
2d92ab3c | 2850 | path_put(&path); |
1da177e4 LT |
2851 | return retval; |
2852 | } | |
2853 | ||
537f7ccb EB |
2854 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2855 | { | |
2856 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2857 | } | |
2858 | ||
2859 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2860 | { | |
2861 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2862 | } | |
2863 | ||
771b1371 EB |
2864 | static void free_mnt_ns(struct mnt_namespace *ns) |
2865 | { | |
6344c433 | 2866 | ns_free_inum(&ns->ns); |
537f7ccb | 2867 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2868 | put_user_ns(ns->user_ns); |
2869 | kfree(ns); | |
2870 | } | |
2871 | ||
8823c079 EB |
2872 | /* |
2873 | * Assign a sequence number so we can detect when we attempt to bind | |
2874 | * mount a reference to an older mount namespace into the current | |
2875 | * mount namespace, preventing reference counting loops. A 64bit | |
2876 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2877 | * is effectively never, so we can ignore the possibility. | |
2878 | */ | |
2879 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2880 | ||
771b1371 | 2881 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2882 | { |
2883 | struct mnt_namespace *new_ns; | |
537f7ccb | 2884 | struct ucounts *ucounts; |
98f842e6 | 2885 | int ret; |
cf8d2c11 | 2886 | |
537f7ccb EB |
2887 | ucounts = inc_mnt_namespaces(user_ns); |
2888 | if (!ucounts) | |
df75e774 | 2889 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2890 | |
cf8d2c11 | 2891 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2892 | if (!new_ns) { |
2893 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2894 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2895 | } |
6344c433 | 2896 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2897 | if (ret) { |
2898 | kfree(new_ns); | |
537f7ccb | 2899 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2900 | return ERR_PTR(ret); |
2901 | } | |
33c42940 | 2902 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2903 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2904 | atomic_set(&new_ns->count, 1); |
2905 | new_ns->root = NULL; | |
2906 | INIT_LIST_HEAD(&new_ns->list); | |
2907 | init_waitqueue_head(&new_ns->poll); | |
2908 | new_ns->event = 0; | |
771b1371 | 2909 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2910 | new_ns->ucounts = ucounts; |
d2921684 EB |
2911 | new_ns->mounts = 0; |
2912 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2913 | return new_ns; |
2914 | } | |
2915 | ||
0766f788 | 2916 | __latent_entropy |
9559f689 AV |
2917 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2918 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2919 | { |
6b3286ed | 2920 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2921 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2922 | struct mount *p, *q; |
9559f689 | 2923 | struct mount *old; |
cb338d06 | 2924 | struct mount *new; |
7a472ef4 | 2925 | int copy_flags; |
1da177e4 | 2926 | |
9559f689 AV |
2927 | BUG_ON(!ns); |
2928 | ||
2929 | if (likely(!(flags & CLONE_NEWNS))) { | |
2930 | get_mnt_ns(ns); | |
2931 | return ns; | |
2932 | } | |
2933 | ||
2934 | old = ns->root; | |
2935 | ||
771b1371 | 2936 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2937 | if (IS_ERR(new_ns)) |
2938 | return new_ns; | |
1da177e4 | 2939 | |
97216be0 | 2940 | namespace_lock(); |
1da177e4 | 2941 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2942 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2943 | if (user_ns != ns->user_ns) |
132c94e3 | 2944 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2945 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2946 | if (IS_ERR(new)) { |
328e6d90 | 2947 | namespace_unlock(); |
771b1371 | 2948 | free_mnt_ns(new_ns); |
be34d1a3 | 2949 | return ERR_CAST(new); |
1da177e4 | 2950 | } |
be08d6d2 | 2951 | new_ns->root = new; |
1a4eeaf2 | 2952 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2953 | |
2954 | /* | |
2955 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2956 | * as belonging to new namespace. We have already acquired a private | |
2957 | * fs_struct, so tsk->fs->lock is not needed. | |
2958 | */ | |
909b0a88 | 2959 | p = old; |
cb338d06 | 2960 | q = new; |
1da177e4 | 2961 | while (p) { |
143c8c91 | 2962 | q->mnt_ns = new_ns; |
d2921684 | 2963 | new_ns->mounts++; |
9559f689 AV |
2964 | if (new_fs) { |
2965 | if (&p->mnt == new_fs->root.mnt) { | |
2966 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2967 | rootmnt = &p->mnt; |
1da177e4 | 2968 | } |
9559f689 AV |
2969 | if (&p->mnt == new_fs->pwd.mnt) { |
2970 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2971 | pwdmnt = &p->mnt; |
1da177e4 | 2972 | } |
1da177e4 | 2973 | } |
909b0a88 AV |
2974 | p = next_mnt(p, old); |
2975 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2976 | if (!q) |
2977 | break; | |
2978 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2979 | p = next_mnt(p, old); | |
1da177e4 | 2980 | } |
328e6d90 | 2981 | namespace_unlock(); |
1da177e4 | 2982 | |
1da177e4 | 2983 | if (rootmnt) |
f03c6599 | 2984 | mntput(rootmnt); |
1da177e4 | 2985 | if (pwdmnt) |
f03c6599 | 2986 | mntput(pwdmnt); |
1da177e4 | 2987 | |
741a2951 | 2988 | return new_ns; |
1da177e4 LT |
2989 | } |
2990 | ||
cf8d2c11 TM |
2991 | /** |
2992 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2993 | * @mnt: pointer to the new root filesystem mountpoint | |
2994 | */ | |
1a4eeaf2 | 2995 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2996 | { |
771b1371 | 2997 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2998 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2999 | struct mount *mnt = real_mount(m); |
3000 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 3001 | new_ns->root = mnt; |
d2921684 | 3002 | new_ns->mounts++; |
b1983cd8 | 3003 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 3004 | } else { |
1a4eeaf2 | 3005 | mntput(m); |
cf8d2c11 TM |
3006 | } |
3007 | return new_ns; | |
3008 | } | |
cf8d2c11 | 3009 | |
ea441d11 AV |
3010 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3011 | { | |
3012 | struct mnt_namespace *ns; | |
d31da0f0 | 3013 | struct super_block *s; |
ea441d11 AV |
3014 | struct path path; |
3015 | int err; | |
3016 | ||
3017 | ns = create_mnt_ns(mnt); | |
3018 | if (IS_ERR(ns)) | |
3019 | return ERR_CAST(ns); | |
3020 | ||
3021 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3022 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3023 | ||
3024 | put_mnt_ns(ns); | |
3025 | ||
3026 | if (err) | |
3027 | return ERR_PTR(err); | |
3028 | ||
3029 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3030 | s = path.mnt->mnt_sb; |
3031 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3032 | mntput(path.mnt); |
3033 | /* lock the sucker */ | |
d31da0f0 | 3034 | down_write(&s->s_umount); |
ea441d11 AV |
3035 | /* ... and return the root of (sub)tree on it */ |
3036 | return path.dentry; | |
3037 | } | |
3038 | EXPORT_SYMBOL(mount_subtree); | |
3039 | ||
312db1aa DB |
3040 | int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, |
3041 | unsigned long flags, void __user *data) | |
1da177e4 | 3042 | { |
eca6f534 VN |
3043 | int ret; |
3044 | char *kernel_type; | |
eca6f534 | 3045 | char *kernel_dev; |
b40ef869 | 3046 | void *options; |
1da177e4 | 3047 | |
b8850d1f TG |
3048 | kernel_type = copy_mount_string(type); |
3049 | ret = PTR_ERR(kernel_type); | |
3050 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3051 | goto out_type; |
1da177e4 | 3052 | |
b8850d1f TG |
3053 | kernel_dev = copy_mount_string(dev_name); |
3054 | ret = PTR_ERR(kernel_dev); | |
3055 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3056 | goto out_dev; |
1da177e4 | 3057 | |
b40ef869 AV |
3058 | options = copy_mount_options(data); |
3059 | ret = PTR_ERR(options); | |
3060 | if (IS_ERR(options)) | |
eca6f534 | 3061 | goto out_data; |
1da177e4 | 3062 | |
b40ef869 | 3063 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3064 | |
b40ef869 | 3065 | kfree(options); |
eca6f534 VN |
3066 | out_data: |
3067 | kfree(kernel_dev); | |
3068 | out_dev: | |
eca6f534 VN |
3069 | kfree(kernel_type); |
3070 | out_type: | |
3071 | return ret; | |
1da177e4 LT |
3072 | } |
3073 | ||
312db1aa DB |
3074 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3075 | char __user *, type, unsigned long, flags, void __user *, data) | |
3076 | { | |
3077 | return ksys_mount(dev_name, dir_name, type, flags, data); | |
3078 | } | |
3079 | ||
afac7cba AV |
3080 | /* |
3081 | * Return true if path is reachable from root | |
3082 | * | |
48a066e7 | 3083 | * namespace_sem or mount_lock is held |
afac7cba | 3084 | */ |
643822b4 | 3085 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3086 | const struct path *root) |
3087 | { | |
643822b4 | 3088 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3089 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3090 | mnt = mnt->mnt_parent; |
afac7cba | 3091 | } |
643822b4 | 3092 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3093 | } |
3094 | ||
640eb7e7 | 3095 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3096 | { |
25ab4c9b | 3097 | bool res; |
48a066e7 | 3098 | read_seqlock_excl(&mount_lock); |
643822b4 | 3099 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3100 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3101 | return res; |
3102 | } | |
3103 | EXPORT_SYMBOL(path_is_under); | |
3104 | ||
1da177e4 LT |
3105 | /* |
3106 | * pivot_root Semantics: | |
3107 | * Moves the root file system of the current process to the directory put_old, | |
3108 | * makes new_root as the new root file system of the current process, and sets | |
3109 | * root/cwd of all processes which had them on the current root to new_root. | |
3110 | * | |
3111 | * Restrictions: | |
3112 | * The new_root and put_old must be directories, and must not be on the | |
3113 | * same file system as the current process root. The put_old must be | |
3114 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3115 | * pointed to by put_old must yield the same directory as new_root. No other | |
3116 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3117 | * | |
4a0d11fa NB |
3118 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3119 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3120 | * in this situation. | |
3121 | * | |
1da177e4 LT |
3122 | * Notes: |
3123 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3124 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3125 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3126 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3127 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3128 | * first. | |
3129 | */ | |
3480b257 HC |
3130 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3131 | const char __user *, put_old) | |
1da177e4 | 3132 | { |
2d8f3038 | 3133 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3134 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3135 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3136 | int error; |
3137 | ||
9b40bc90 | 3138 | if (!may_mount()) |
1da177e4 LT |
3139 | return -EPERM; |
3140 | ||
2d8f3038 | 3141 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3142 | if (error) |
3143 | goto out0; | |
1da177e4 | 3144 | |
2d8f3038 | 3145 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3146 | if (error) |
3147 | goto out1; | |
3148 | ||
2d8f3038 | 3149 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3150 | if (error) |
3151 | goto out2; | |
1da177e4 | 3152 | |
f7ad3c6b | 3153 | get_fs_root(current->fs, &root); |
84d17192 AV |
3154 | old_mp = lock_mount(&old); |
3155 | error = PTR_ERR(old_mp); | |
3156 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3157 | goto out3; |
3158 | ||
1da177e4 | 3159 | error = -EINVAL; |
419148da AV |
3160 | new_mnt = real_mount(new.mnt); |
3161 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3162 | old_mnt = real_mount(old.mnt); |
3163 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3164 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3165 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3166 | goto out4; |
143c8c91 | 3167 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3168 | goto out4; |
5ff9d8a6 EB |
3169 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3170 | goto out4; | |
1da177e4 | 3171 | error = -ENOENT; |
f3da392e | 3172 | if (d_unlinked(new.dentry)) |
b12cea91 | 3173 | goto out4; |
1da177e4 | 3174 | error = -EBUSY; |
84d17192 | 3175 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3176 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3177 | error = -EINVAL; |
8c3ee42e | 3178 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3179 | goto out4; /* not a mountpoint */ |
676da58d | 3180 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3181 | goto out4; /* not attached */ |
84d17192 | 3182 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3183 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3184 | goto out4; /* not a mountpoint */ |
676da58d | 3185 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3186 | goto out4; /* not attached */ |
4ac91378 | 3187 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3188 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3189 | goto out4; |
0d082601 EB |
3190 | /* make certain new is below the root */ |
3191 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3192 | goto out4; | |
84d17192 | 3193 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3194 | lock_mount_hash(); |
419148da AV |
3195 | detach_mnt(new_mnt, &parent_path); |
3196 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3197 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3198 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3199 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3200 | } | |
4ac91378 | 3201 | /* mount old root on put_old */ |
84d17192 | 3202 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3203 | /* mount new_root on / */ |
84d17192 | 3204 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3205 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3206 | /* A moved mount should not expire automatically */ |
3207 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3208 | put_mountpoint(root_mp); |
719ea2fb | 3209 | unlock_mount_hash(); |
2d8f3038 | 3210 | chroot_fs_refs(&root, &new); |
1da177e4 | 3211 | error = 0; |
b12cea91 | 3212 | out4: |
84d17192 | 3213 | unlock_mount(old_mp); |
b12cea91 AV |
3214 | if (!error) { |
3215 | path_put(&root_parent); | |
3216 | path_put(&parent_path); | |
3217 | } | |
3218 | out3: | |
8c3ee42e | 3219 | path_put(&root); |
b12cea91 | 3220 | out2: |
2d8f3038 | 3221 | path_put(&old); |
1da177e4 | 3222 | out1: |
2d8f3038 | 3223 | path_put(&new); |
1da177e4 | 3224 | out0: |
1da177e4 | 3225 | return error; |
1da177e4 LT |
3226 | } |
3227 | ||
3228 | static void __init init_mount_tree(void) | |
3229 | { | |
3230 | struct vfsmount *mnt; | |
6b3286ed | 3231 | struct mnt_namespace *ns; |
ac748a09 | 3232 | struct path root; |
0c55cfc4 | 3233 | struct file_system_type *type; |
1da177e4 | 3234 | |
0c55cfc4 EB |
3235 | type = get_fs_type("rootfs"); |
3236 | if (!type) | |
3237 | panic("Can't find rootfs type"); | |
3238 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3239 | put_filesystem(type); | |
1da177e4 LT |
3240 | if (IS_ERR(mnt)) |
3241 | panic("Can't create rootfs"); | |
b3e19d92 | 3242 | |
3b22edc5 TM |
3243 | ns = create_mnt_ns(mnt); |
3244 | if (IS_ERR(ns)) | |
1da177e4 | 3245 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3246 | |
3247 | init_task.nsproxy->mnt_ns = ns; | |
3248 | get_mnt_ns(ns); | |
3249 | ||
be08d6d2 AV |
3250 | root.mnt = mnt; |
3251 | root.dentry = mnt->mnt_root; | |
da362b09 | 3252 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3253 | |
3254 | set_fs_pwd(current->fs, &root); | |
3255 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3256 | } |
3257 | ||
74bf17cf | 3258 | void __init mnt_init(void) |
1da177e4 | 3259 | { |
15a67dd8 | 3260 | int err; |
1da177e4 | 3261 | |
7d6fec45 | 3262 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3263 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3264 | |
0818bf27 | 3265 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3266 | sizeof(struct hlist_head), |
0818bf27 | 3267 | mhash_entries, 19, |
3d375d78 | 3268 | HASH_ZERO, |
0818bf27 AV |
3269 | &m_hash_shift, &m_hash_mask, 0, 0); |
3270 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3271 | sizeof(struct hlist_head), | |
3272 | mphash_entries, 19, | |
3d375d78 | 3273 | HASH_ZERO, |
0818bf27 | 3274 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3275 | |
84d17192 | 3276 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3277 | panic("Failed to allocate mount hash table\n"); |
3278 | ||
4b93dc9b TH |
3279 | kernfs_init(); |
3280 | ||
15a67dd8 RD |
3281 | err = sysfs_init(); |
3282 | if (err) | |
3283 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3284 | __func__, err); |
00d26666 GKH |
3285 | fs_kobj = kobject_create_and_add("fs", NULL); |
3286 | if (!fs_kobj) | |
8e24eea7 | 3287 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3288 | init_rootfs(); |
3289 | init_mount_tree(); | |
3290 | } | |
3291 | ||
616511d0 | 3292 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3293 | { |
d498b25a | 3294 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3295 | return; |
7b00ed6f | 3296 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3297 | free_mnt_ns(ns); |
1da177e4 | 3298 | } |
9d412a43 AV |
3299 | |
3300 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3301 | { | |
423e0ab0 | 3302 | struct vfsmount *mnt; |
e462ec50 | 3303 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); |
423e0ab0 TC |
3304 | if (!IS_ERR(mnt)) { |
3305 | /* | |
3306 | * it is a longterm mount, don't release mnt until | |
3307 | * we unmount before file sys is unregistered | |
3308 | */ | |
f7a99c5b | 3309 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3310 | } |
3311 | return mnt; | |
9d412a43 AV |
3312 | } |
3313 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3314 | |
3315 | void kern_unmount(struct vfsmount *mnt) | |
3316 | { | |
3317 | /* release long term mount so mount point can be released */ | |
3318 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3319 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3320 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3321 | mntput(mnt); |
3322 | } | |
3323 | } | |
3324 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3325 | |
3326 | bool our_mnt(struct vfsmount *mnt) | |
3327 | { | |
143c8c91 | 3328 | return check_mnt(real_mount(mnt)); |
02125a82 | 3329 | } |
8823c079 | 3330 | |
3151527e EB |
3331 | bool current_chrooted(void) |
3332 | { | |
3333 | /* Does the current process have a non-standard root */ | |
3334 | struct path ns_root; | |
3335 | struct path fs_root; | |
3336 | bool chrooted; | |
3337 | ||
3338 | /* Find the namespace root */ | |
3339 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3340 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3341 | path_get(&ns_root); | |
3342 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3343 | ; | |
3344 | ||
3345 | get_fs_root(current->fs, &fs_root); | |
3346 | ||
3347 | chrooted = !path_equal(&fs_root, &ns_root); | |
3348 | ||
3349 | path_put(&fs_root); | |
3350 | path_put(&ns_root); | |
3351 | ||
3352 | return chrooted; | |
3353 | } | |
3354 | ||
8654df4e EB |
3355 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3356 | int *new_mnt_flags) | |
87a8ebd6 | 3357 | { |
8c6cf9cc | 3358 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3359 | struct mount *mnt; |
e51db735 | 3360 | bool visible = false; |
87a8ebd6 | 3361 | |
44bb4385 | 3362 | down_read(&namespace_sem); |
87a8ebd6 | 3363 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3364 | struct mount *child; |
77b1a97d EB |
3365 | int mnt_flags; |
3366 | ||
8654df4e | 3367 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3368 | continue; |
3369 | ||
7e96c1b0 EB |
3370 | /* This mount is not fully visible if it's root directory |
3371 | * is not the root directory of the filesystem. | |
3372 | */ | |
3373 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3374 | continue; | |
3375 | ||
a1935c17 | 3376 | /* A local view of the mount flags */ |
77b1a97d | 3377 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3378 | |
695e9df0 | 3379 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3380 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3381 | mnt_flags |= MNT_LOCK_READONLY; |
3382 | ||
8c6cf9cc EB |
3383 | /* Verify the mount flags are equal to or more permissive |
3384 | * than the proposed new mount. | |
3385 | */ | |
77b1a97d | 3386 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3387 | !(new_flags & MNT_READONLY)) |
3388 | continue; | |
77b1a97d EB |
3389 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3390 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3391 | continue; |
3392 | ||
ceeb0e5d EB |
3393 | /* This mount is not fully visible if there are any |
3394 | * locked child mounts that cover anything except for | |
3395 | * empty directories. | |
e51db735 EB |
3396 | */ |
3397 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3398 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3399 | /* Only worry about locked mounts */ |
d71ed6c9 | 3400 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3401 | continue; |
7236c85e EB |
3402 | /* Is the directory permanetly empty? */ |
3403 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3404 | goto next; |
87a8ebd6 | 3405 | } |
8c6cf9cc | 3406 | /* Preserve the locked attributes */ |
77b1a97d | 3407 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3408 | MNT_LOCK_ATIME); |
e51db735 EB |
3409 | visible = true; |
3410 | goto found; | |
3411 | next: ; | |
87a8ebd6 | 3412 | } |
e51db735 | 3413 | found: |
44bb4385 | 3414 | up_read(&namespace_sem); |
e51db735 | 3415 | return visible; |
87a8ebd6 EB |
3416 | } |
3417 | ||
8654df4e EB |
3418 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3419 | { | |
a1935c17 | 3420 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3421 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3422 | unsigned long s_iflags; | |
3423 | ||
3424 | if (ns->user_ns == &init_user_ns) | |
3425 | return false; | |
3426 | ||
3427 | /* Can this filesystem be too revealing? */ | |
3428 | s_iflags = mnt->mnt_sb->s_iflags; | |
3429 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3430 | return false; | |
3431 | ||
a1935c17 EB |
3432 | if ((s_iflags & required_iflags) != required_iflags) { |
3433 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3434 | required_iflags); | |
3435 | return true; | |
3436 | } | |
3437 | ||
8654df4e EB |
3438 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3439 | } | |
3440 | ||
380cf5ba AL |
3441 | bool mnt_may_suid(struct vfsmount *mnt) |
3442 | { | |
3443 | /* | |
3444 | * Foreign mounts (accessed via fchdir or through /proc | |
3445 | * symlinks) are always treated as if they are nosuid. This | |
3446 | * prevents namespaces from trusting potentially unsafe | |
3447 | * suid/sgid bits, file caps, or security labels that originate | |
3448 | * in other namespaces. | |
3449 | */ | |
3450 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3451 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3452 | } | |
3453 | ||
64964528 | 3454 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3455 | { |
58be2825 | 3456 | struct ns_common *ns = NULL; |
8823c079 EB |
3457 | struct nsproxy *nsproxy; |
3458 | ||
728dba3a EB |
3459 | task_lock(task); |
3460 | nsproxy = task->nsproxy; | |
8823c079 | 3461 | if (nsproxy) { |
58be2825 AV |
3462 | ns = &nsproxy->mnt_ns->ns; |
3463 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3464 | } |
728dba3a | 3465 | task_unlock(task); |
8823c079 EB |
3466 | |
3467 | return ns; | |
3468 | } | |
3469 | ||
64964528 | 3470 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3471 | { |
58be2825 | 3472 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3473 | } |
3474 | ||
64964528 | 3475 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3476 | { |
3477 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3478 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3479 | struct path root; |
4f757f3c | 3480 | int err; |
8823c079 | 3481 | |
0c55cfc4 | 3482 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3483 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3484 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3485 | return -EPERM; |
8823c079 EB |
3486 | |
3487 | if (fs->users != 1) | |
3488 | return -EINVAL; | |
3489 | ||
3490 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3491 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3492 | nsproxy->mnt_ns = mnt_ns; |
3493 | ||
3494 | /* Find the root */ | |
4f757f3c AV |
3495 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3496 | "/", LOOKUP_DOWN, &root); | |
3497 | if (err) { | |
3498 | /* revert to old namespace */ | |
3499 | nsproxy->mnt_ns = old_mnt_ns; | |
3500 | put_mnt_ns(mnt_ns); | |
3501 | return err; | |
3502 | } | |
8823c079 | 3503 | |
4068367c AV |
3504 | put_mnt_ns(old_mnt_ns); |
3505 | ||
8823c079 EB |
3506 | /* Update the pwd and root */ |
3507 | set_fs_pwd(fs, &root); | |
3508 | set_fs_root(fs, &root); | |
3509 | ||
3510 | path_put(&root); | |
3511 | return 0; | |
3512 | } | |
3513 | ||
bcac25a5 AV |
3514 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3515 | { | |
3516 | return to_mnt_ns(ns)->user_ns; | |
3517 | } | |
3518 | ||
8823c079 EB |
3519 | const struct proc_ns_operations mntns_operations = { |
3520 | .name = "mnt", | |
3521 | .type = CLONE_NEWNS, | |
3522 | .get = mntns_get, | |
3523 | .put = mntns_put, | |
3524 | .install = mntns_install, | |
bcac25a5 | 3525 | .owner = mntns_owner, |
8823c079 | 3526 | }; |