]>
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 */ | |
a07b2000 | 23 | #include <linux/file.h> |
d10577a8 | 24 | #include <linux/uaccess.h> |
0bb80f24 | 25 | #include <linux/proc_ns.h> |
20b4fb48 | 26 | #include <linux/magic.h> |
57c8a661 | 27 | #include <linux/memblock.h> |
9ea459e1 | 28 | #include <linux/task_work.h> |
9164bb4a | 29 | #include <linux/sched/task.h> |
e262e32d | 30 | #include <uapi/linux/mount.h> |
9bc61ab1 | 31 | #include <linux/fs_context.h> |
9164bb4a | 32 | |
07b20889 | 33 | #include "pnode.h" |
948730b0 | 34 | #include "internal.h" |
1da177e4 | 35 | |
d2921684 EB |
36 | /* Maximum number of mounts in a mount namespace */ |
37 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
38 | ||
0818bf27 AV |
39 | static unsigned int m_hash_mask __read_mostly; |
40 | static unsigned int m_hash_shift __read_mostly; | |
41 | static unsigned int mp_hash_mask __read_mostly; | |
42 | static unsigned int mp_hash_shift __read_mostly; | |
43 | ||
44 | static __initdata unsigned long mhash_entries; | |
45 | static int __init set_mhash_entries(char *str) | |
46 | { | |
47 | if (!str) | |
48 | return 0; | |
49 | mhash_entries = simple_strtoul(str, &str, 0); | |
50 | return 1; | |
51 | } | |
52 | __setup("mhash_entries=", set_mhash_entries); | |
53 | ||
54 | static __initdata unsigned long mphash_entries; | |
55 | static int __init set_mphash_entries(char *str) | |
56 | { | |
57 | if (!str) | |
58 | return 0; | |
59 | mphash_entries = simple_strtoul(str, &str, 0); | |
60 | return 1; | |
61 | } | |
62 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 63 | |
c7999c36 | 64 | static u64 event; |
73cd49ec | 65 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 66 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 67 | |
38129a13 | 68 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 69 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 70 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 71 | static DECLARE_RWSEM(namespace_sem); |
1da177e4 | 72 | |
f87fd4c2 | 73 | /* /sys/fs */ |
00d26666 GKH |
74 | struct kobject *fs_kobj; |
75 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 76 | |
99b7db7b NP |
77 | /* |
78 | * vfsmount lock may be taken for read to prevent changes to the | |
79 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
80 | * up the tree. | |
81 | * | |
82 | * It should be taken for write in all cases where the vfsmount | |
83 | * tree or hash is modified or when a vfsmount structure is modified. | |
84 | */ | |
48a066e7 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 86 | |
38129a13 | 87 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 88 | { |
b58fed8b RP |
89 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
90 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
91 | tmp = tmp + (tmp >> m_hash_shift); |
92 | return &mount_hashtable[tmp & m_hash_mask]; | |
93 | } | |
94 | ||
95 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
96 | { | |
97 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
98 | tmp = tmp + (tmp >> mp_hash_shift); | |
99 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
100 | } |
101 | ||
b105e270 | 102 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 103 | { |
169b480e MW |
104 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
105 | ||
106 | if (res < 0) | |
107 | return res; | |
108 | mnt->mnt_id = res; | |
109 | return 0; | |
73cd49ec MS |
110 | } |
111 | ||
b105e270 | 112 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 113 | { |
169b480e | 114 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
115 | } |
116 | ||
719f5d7f MS |
117 | /* |
118 | * Allocate a new peer group ID | |
719f5d7f | 119 | */ |
4b8b21f4 | 120 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 121 | { |
169b480e | 122 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 123 | |
169b480e MW |
124 | if (res < 0) |
125 | return res; | |
126 | mnt->mnt_group_id = res; | |
127 | return 0; | |
719f5d7f MS |
128 | } |
129 | ||
130 | /* | |
131 | * Release a peer group ID | |
132 | */ | |
4b8b21f4 | 133 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 134 | { |
169b480e | 135 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 136 | mnt->mnt_group_id = 0; |
719f5d7f MS |
137 | } |
138 | ||
b3e19d92 NP |
139 | /* |
140 | * vfsmount lock must be held for read | |
141 | */ | |
83adc753 | 142 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
143 | { |
144 | #ifdef CONFIG_SMP | |
68e8a9fe | 145 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
146 | #else |
147 | preempt_disable(); | |
68e8a9fe | 148 | mnt->mnt_count += n; |
b3e19d92 NP |
149 | preempt_enable(); |
150 | #endif | |
151 | } | |
152 | ||
b3e19d92 NP |
153 | /* |
154 | * vfsmount lock must be held for write | |
155 | */ | |
83adc753 | 156 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
157 | { |
158 | #ifdef CONFIG_SMP | |
f03c6599 | 159 | unsigned int count = 0; |
b3e19d92 NP |
160 | int cpu; |
161 | ||
162 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 163 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
164 | } |
165 | ||
166 | return count; | |
167 | #else | |
68e8a9fe | 168 | return mnt->mnt_count; |
b3e19d92 NP |
169 | #endif |
170 | } | |
171 | ||
87b95ce0 AV |
172 | static void drop_mountpoint(struct fs_pin *p) |
173 | { | |
174 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
175 | dput(m->mnt_ex_mountpoint); | |
176 | pin_remove(p); | |
177 | mntput(&m->mnt); | |
178 | } | |
179 | ||
b105e270 | 180 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 181 | { |
c63181e6 AV |
182 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
183 | if (mnt) { | |
73cd49ec MS |
184 | int err; |
185 | ||
c63181e6 | 186 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
187 | if (err) |
188 | goto out_free_cache; | |
189 | ||
190 | if (name) { | |
fcc139ae | 191 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 192 | if (!mnt->mnt_devname) |
88b38782 | 193 | goto out_free_id; |
73cd49ec MS |
194 | } |
195 | ||
b3e19d92 | 196 | #ifdef CONFIG_SMP |
c63181e6 AV |
197 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
198 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
199 | goto out_free_devname; |
200 | ||
c63181e6 | 201 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 202 | #else |
c63181e6 AV |
203 | mnt->mnt_count = 1; |
204 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
205 | #endif |
206 | ||
38129a13 | 207 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
208 | INIT_LIST_HEAD(&mnt->mnt_child); |
209 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
210 | INIT_LIST_HEAD(&mnt->mnt_list); | |
211 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
212 | INIT_LIST_HEAD(&mnt->mnt_share); | |
213 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
214 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 215 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 216 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
87b95ce0 | 217 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 218 | } |
c63181e6 | 219 | return mnt; |
88b38782 | 220 | |
d3ef3d73 | 221 | #ifdef CONFIG_SMP |
222 | out_free_devname: | |
fcc139ae | 223 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 224 | #endif |
88b38782 | 225 | out_free_id: |
c63181e6 | 226 | mnt_free_id(mnt); |
88b38782 | 227 | out_free_cache: |
c63181e6 | 228 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 229 | return NULL; |
1da177e4 LT |
230 | } |
231 | ||
3d733633 DH |
232 | /* |
233 | * Most r/o checks on a fs are for operations that take | |
234 | * discrete amounts of time, like a write() or unlink(). | |
235 | * We must keep track of when those operations start | |
236 | * (for permission checks) and when they end, so that | |
237 | * we can determine when writes are able to occur to | |
238 | * a filesystem. | |
239 | */ | |
240 | /* | |
241 | * __mnt_is_readonly: check whether a mount is read-only | |
242 | * @mnt: the mount to check for its write status | |
243 | * | |
244 | * This shouldn't be used directly ouside of the VFS. | |
245 | * It does not guarantee that the filesystem will stay | |
246 | * r/w, just that it is right *now*. This can not and | |
247 | * should not be used in place of IS_RDONLY(inode). | |
248 | * mnt_want/drop_write() will _keep_ the filesystem | |
249 | * r/w. | |
250 | */ | |
43f5e655 | 251 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 252 | { |
43f5e655 | 253 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
254 | } |
255 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
256 | ||
83adc753 | 257 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 | 258 | { |
259 | #ifdef CONFIG_SMP | |
68e8a9fe | 260 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 261 | #else |
68e8a9fe | 262 | mnt->mnt_writers++; |
d3ef3d73 | 263 | #endif |
264 | } | |
3d733633 | 265 | |
83adc753 | 266 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 267 | { |
d3ef3d73 | 268 | #ifdef CONFIG_SMP |
68e8a9fe | 269 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 270 | #else |
68e8a9fe | 271 | mnt->mnt_writers--; |
d3ef3d73 | 272 | #endif |
3d733633 | 273 | } |
3d733633 | 274 | |
83adc753 | 275 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 276 | { |
d3ef3d73 | 277 | #ifdef CONFIG_SMP |
278 | unsigned int count = 0; | |
3d733633 | 279 | int cpu; |
3d733633 DH |
280 | |
281 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 282 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 283 | } |
3d733633 | 284 | |
d3ef3d73 | 285 | return count; |
286 | #else | |
287 | return mnt->mnt_writers; | |
288 | #endif | |
3d733633 DH |
289 | } |
290 | ||
4ed5e82f MS |
291 | static int mnt_is_readonly(struct vfsmount *mnt) |
292 | { | |
293 | if (mnt->mnt_sb->s_readonly_remount) | |
294 | return 1; | |
295 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
296 | smp_rmb(); | |
297 | return __mnt_is_readonly(mnt); | |
298 | } | |
299 | ||
8366025e | 300 | /* |
eb04c282 JK |
301 | * Most r/o & frozen checks on a fs are for operations that take discrete |
302 | * amounts of time, like a write() or unlink(). We must keep track of when | |
303 | * those operations start (for permission checks) and when they end, so that we | |
304 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
305 | */ |
306 | /** | |
eb04c282 | 307 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 308 | * @m: the mount on which to take a write |
8366025e | 309 | * |
eb04c282 JK |
310 | * This tells the low-level filesystem that a write is about to be performed to |
311 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
312 | * returning success. This operation does not protect against filesystem being | |
313 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
314 | * called. This is effectively a refcount. | |
8366025e | 315 | */ |
eb04c282 | 316 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 317 | { |
83adc753 | 318 | struct mount *mnt = real_mount(m); |
3d733633 | 319 | int ret = 0; |
3d733633 | 320 | |
d3ef3d73 | 321 | preempt_disable(); |
c6653a83 | 322 | mnt_inc_writers(mnt); |
d3ef3d73 | 323 | /* |
c6653a83 | 324 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 325 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
326 | * incremented count after it has set MNT_WRITE_HOLD. | |
327 | */ | |
328 | smp_mb(); | |
6aa7de05 | 329 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 330 | cpu_relax(); |
331 | /* | |
332 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
333 | * be set to match its requirements. So we must not load that until | |
334 | * MNT_WRITE_HOLD is cleared. | |
335 | */ | |
336 | smp_rmb(); | |
4ed5e82f | 337 | if (mnt_is_readonly(m)) { |
c6653a83 | 338 | mnt_dec_writers(mnt); |
3d733633 | 339 | ret = -EROFS; |
3d733633 | 340 | } |
d3ef3d73 | 341 | preempt_enable(); |
eb04c282 JK |
342 | |
343 | return ret; | |
344 | } | |
345 | ||
346 | /** | |
347 | * mnt_want_write - get write access to a mount | |
348 | * @m: the mount on which to take a write | |
349 | * | |
350 | * This tells the low-level filesystem that a write is about to be performed to | |
351 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
352 | * is not frozen) before returning success. When the write operation is | |
353 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
354 | */ | |
355 | int mnt_want_write(struct vfsmount *m) | |
356 | { | |
357 | int ret; | |
358 | ||
359 | sb_start_write(m->mnt_sb); | |
360 | ret = __mnt_want_write(m); | |
361 | if (ret) | |
362 | sb_end_write(m->mnt_sb); | |
3d733633 | 363 | return ret; |
8366025e DH |
364 | } |
365 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
366 | ||
96029c4e | 367 | /** |
368 | * mnt_clone_write - get write access to a mount | |
369 | * @mnt: the mount on which to take a write | |
370 | * | |
371 | * This is effectively like mnt_want_write, except | |
372 | * it must only be used to take an extra write reference | |
373 | * on a mountpoint that we already know has a write reference | |
374 | * on it. This allows some optimisation. | |
375 | * | |
376 | * After finished, mnt_drop_write must be called as usual to | |
377 | * drop the reference. | |
378 | */ | |
379 | int mnt_clone_write(struct vfsmount *mnt) | |
380 | { | |
381 | /* superblock may be r/o */ | |
382 | if (__mnt_is_readonly(mnt)) | |
383 | return -EROFS; | |
384 | preempt_disable(); | |
83adc753 | 385 | mnt_inc_writers(real_mount(mnt)); |
96029c4e | 386 | preempt_enable(); |
387 | return 0; | |
388 | } | |
389 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
390 | ||
391 | /** | |
eb04c282 | 392 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e | 393 | * @file: the file who's mount on which to take a write |
394 | * | |
eb04c282 | 395 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e | 396 | * do some optimisations if the file is open for write already |
397 | */ | |
eb04c282 | 398 | int __mnt_want_write_file(struct file *file) |
96029c4e | 399 | { |
83f936c7 | 400 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 401 | return __mnt_want_write(file->f_path.mnt); |
96029c4e | 402 | else |
403 | return mnt_clone_write(file->f_path.mnt); | |
404 | } | |
eb04c282 | 405 | |
7c6893e3 MS |
406 | /** |
407 | * mnt_want_write_file - get write access to a file's mount | |
408 | * @file: the file who's mount on which to take a write | |
409 | * | |
410 | * This is like mnt_want_write, but it takes a file and can | |
411 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
412 | */ |
413 | int mnt_want_write_file(struct file *file) | |
414 | { | |
415 | int ret; | |
416 | ||
a6795a58 | 417 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
418 | ret = __mnt_want_write_file(file); |
419 | if (ret) | |
a6795a58 | 420 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
421 | return ret; |
422 | } | |
96029c4e | 423 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
424 | ||
8366025e | 425 | /** |
eb04c282 | 426 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
427 | * @mnt: the mount on which to give up write access |
428 | * | |
429 | * Tells the low-level filesystem that we are done | |
430 | * performing writes to it. Must be matched with | |
eb04c282 | 431 | * __mnt_want_write() call above. |
8366025e | 432 | */ |
eb04c282 | 433 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 434 | { |
d3ef3d73 | 435 | preempt_disable(); |
83adc753 | 436 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 437 | preempt_enable(); |
8366025e | 438 | } |
eb04c282 JK |
439 | |
440 | /** | |
441 | * mnt_drop_write - give up write access to a mount | |
442 | * @mnt: the mount on which to give up write access | |
443 | * | |
444 | * Tells the low-level filesystem that we are done performing writes to it and | |
445 | * also allows filesystem to be frozen again. Must be matched with | |
446 | * mnt_want_write() call above. | |
447 | */ | |
448 | void mnt_drop_write(struct vfsmount *mnt) | |
449 | { | |
450 | __mnt_drop_write(mnt); | |
451 | sb_end_write(mnt->mnt_sb); | |
452 | } | |
8366025e DH |
453 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
454 | ||
eb04c282 JK |
455 | void __mnt_drop_write_file(struct file *file) |
456 | { | |
457 | __mnt_drop_write(file->f_path.mnt); | |
458 | } | |
459 | ||
7c6893e3 MS |
460 | void mnt_drop_write_file(struct file *file) |
461 | { | |
a6795a58 | 462 | __mnt_drop_write_file(file); |
7c6893e3 MS |
463 | sb_end_write(file_inode(file)->i_sb); |
464 | } | |
2a79f17e AV |
465 | EXPORT_SYMBOL(mnt_drop_write_file); |
466 | ||
83adc753 | 467 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 468 | { |
3d733633 DH |
469 | int ret = 0; |
470 | ||
719ea2fb | 471 | lock_mount_hash(); |
83adc753 | 472 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 473 | /* |
d3ef3d73 | 474 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
475 | * should be visible before we do. | |
3d733633 | 476 | */ |
d3ef3d73 | 477 | smp_mb(); |
478 | ||
3d733633 | 479 | /* |
d3ef3d73 | 480 | * With writers on hold, if this value is zero, then there are |
481 | * definitely no active writers (although held writers may subsequently | |
482 | * increment the count, they'll have to wait, and decrement it after | |
483 | * seeing MNT_READONLY). | |
484 | * | |
485 | * It is OK to have counter incremented on one CPU and decremented on | |
486 | * another: the sum will add up correctly. The danger would be when we | |
487 | * sum up each counter, if we read a counter before it is incremented, | |
488 | * but then read another CPU's count which it has been subsequently | |
489 | * decremented from -- we would see more decrements than we should. | |
490 | * MNT_WRITE_HOLD protects against this scenario, because | |
491 | * mnt_want_write first increments count, then smp_mb, then spins on | |
492 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
493 | * we're counting up here. | |
3d733633 | 494 | */ |
c6653a83 | 495 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 496 | ret = -EBUSY; |
497 | else | |
83adc753 | 498 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 499 | /* |
500 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
501 | * that become unheld will see MNT_READONLY. | |
502 | */ | |
503 | smp_wmb(); | |
83adc753 | 504 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 505 | unlock_mount_hash(); |
3d733633 | 506 | return ret; |
8366025e | 507 | } |
8366025e | 508 | |
43f5e655 | 509 | static int __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 510 | { |
719ea2fb | 511 | lock_mount_hash(); |
83adc753 | 512 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 513 | unlock_mount_hash(); |
43f5e655 | 514 | return 0; |
2e4b7fcd DH |
515 | } |
516 | ||
4ed5e82f MS |
517 | int sb_prepare_remount_readonly(struct super_block *sb) |
518 | { | |
519 | struct mount *mnt; | |
520 | int err = 0; | |
521 | ||
8e8b8796 MS |
522 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
523 | if (atomic_long_read(&sb->s_remove_count)) | |
524 | return -EBUSY; | |
525 | ||
719ea2fb | 526 | lock_mount_hash(); |
4ed5e82f MS |
527 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
528 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
529 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
530 | smp_mb(); | |
531 | if (mnt_get_writers(mnt) > 0) { | |
532 | err = -EBUSY; | |
533 | break; | |
534 | } | |
535 | } | |
536 | } | |
8e8b8796 MS |
537 | if (!err && atomic_long_read(&sb->s_remove_count)) |
538 | err = -EBUSY; | |
539 | ||
4ed5e82f MS |
540 | if (!err) { |
541 | sb->s_readonly_remount = 1; | |
542 | smp_wmb(); | |
543 | } | |
544 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
545 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
546 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
547 | } | |
719ea2fb | 548 | unlock_mount_hash(); |
4ed5e82f MS |
549 | |
550 | return err; | |
551 | } | |
552 | ||
b105e270 | 553 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 554 | { |
fcc139ae | 555 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 556 | #ifdef CONFIG_SMP |
68e8a9fe | 557 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 558 | #endif |
b105e270 | 559 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
560 | } |
561 | ||
8ffcb32e DH |
562 | static void delayed_free_vfsmnt(struct rcu_head *head) |
563 | { | |
564 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
565 | } | |
566 | ||
48a066e7 | 567 | /* call under rcu_read_lock */ |
294d71ff | 568 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
569 | { |
570 | struct mount *mnt; | |
571 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 572 | return 1; |
48a066e7 | 573 | if (bastard == NULL) |
294d71ff | 574 | return 0; |
48a066e7 AV |
575 | mnt = real_mount(bastard); |
576 | mnt_add_count(mnt, 1); | |
119e1ef8 | 577 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 578 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 579 | return 0; |
48a066e7 AV |
580 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
581 | mnt_add_count(mnt, -1); | |
294d71ff AV |
582 | return 1; |
583 | } | |
119e1ef8 AV |
584 | lock_mount_hash(); |
585 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
586 | mnt_add_count(mnt, -1); | |
587 | unlock_mount_hash(); | |
588 | return 1; | |
589 | } | |
590 | unlock_mount_hash(); | |
591 | /* caller will mntput() */ | |
294d71ff AV |
592 | return -1; |
593 | } | |
594 | ||
595 | /* call under rcu_read_lock */ | |
596 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
597 | { | |
598 | int res = __legitimize_mnt(bastard, seq); | |
599 | if (likely(!res)) | |
600 | return true; | |
601 | if (unlikely(res < 0)) { | |
602 | rcu_read_unlock(); | |
603 | mntput(bastard); | |
604 | rcu_read_lock(); | |
48a066e7 | 605 | } |
48a066e7 AV |
606 | return false; |
607 | } | |
608 | ||
1da177e4 | 609 | /* |
474279dc | 610 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 611 | * call under rcu_read_lock() |
1da177e4 | 612 | */ |
474279dc | 613 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 614 | { |
38129a13 | 615 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
616 | struct mount *p; |
617 | ||
38129a13 | 618 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
619 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
620 | return p; | |
621 | return NULL; | |
622 | } | |
623 | ||
a05964f3 | 624 | /* |
f015f126 DH |
625 | * lookup_mnt - Return the first child mount mounted at path |
626 | * | |
627 | * "First" means first mounted chronologically. If you create the | |
628 | * following mounts: | |
629 | * | |
630 | * mount /dev/sda1 /mnt | |
631 | * mount /dev/sda2 /mnt | |
632 | * mount /dev/sda3 /mnt | |
633 | * | |
634 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
635 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
636 | * /dev/sda2, then /dev/sda3, then NULL. | |
637 | * | |
638 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 639 | */ |
ca71cf71 | 640 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 641 | { |
c7105365 | 642 | struct mount *child_mnt; |
48a066e7 AV |
643 | struct vfsmount *m; |
644 | unsigned seq; | |
99b7db7b | 645 | |
48a066e7 AV |
646 | rcu_read_lock(); |
647 | do { | |
648 | seq = read_seqbegin(&mount_lock); | |
649 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
650 | m = child_mnt ? &child_mnt->mnt : NULL; | |
651 | } while (!legitimize_mnt(m, seq)); | |
652 | rcu_read_unlock(); | |
653 | return m; | |
a05964f3 RP |
654 | } |
655 | ||
7af1364f EB |
656 | /* |
657 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
658 | * current mount namespace. | |
659 | * | |
660 | * The common case is dentries are not mountpoints at all and that | |
661 | * test is handled inline. For the slow case when we are actually | |
662 | * dealing with a mountpoint of some kind, walk through all of the | |
663 | * mounts in the current mount namespace and test to see if the dentry | |
664 | * is a mountpoint. | |
665 | * | |
666 | * The mount_hashtable is not usable in the context because we | |
667 | * need to identify all mounts that may be in the current mount | |
668 | * namespace not just a mount that happens to have some specified | |
669 | * parent mount. | |
670 | */ | |
671 | bool __is_local_mountpoint(struct dentry *dentry) | |
672 | { | |
673 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
674 | struct mount *mnt; | |
675 | bool is_covered = false; | |
676 | ||
677 | if (!d_mountpoint(dentry)) | |
678 | goto out; | |
679 | ||
680 | down_read(&namespace_sem); | |
681 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
682 | is_covered = (mnt->mnt_mountpoint == dentry); | |
683 | if (is_covered) | |
684 | break; | |
685 | } | |
686 | up_read(&namespace_sem); | |
687 | out: | |
688 | return is_covered; | |
689 | } | |
690 | ||
e2dfa935 | 691 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 692 | { |
0818bf27 | 693 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
694 | struct mountpoint *mp; |
695 | ||
0818bf27 | 696 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 697 | if (mp->m_dentry == dentry) { |
84d17192 AV |
698 | mp->m_count++; |
699 | return mp; | |
700 | } | |
701 | } | |
e2dfa935 EB |
702 | return NULL; |
703 | } | |
704 | ||
3895dbf8 | 705 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 706 | { |
3895dbf8 | 707 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 708 | int ret; |
84d17192 | 709 | |
3895dbf8 | 710 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
711 | /* might be worth a WARN_ON() */ |
712 | if (d_unlinked(dentry)) | |
713 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
714 | mountpoint: |
715 | read_seqlock_excl(&mount_lock); | |
716 | mp = lookup_mountpoint(dentry); | |
717 | read_sequnlock_excl(&mount_lock); | |
718 | if (mp) | |
719 | goto done; | |
720 | } | |
721 | ||
722 | if (!new) | |
723 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
724 | if (!new) | |
84d17192 AV |
725 | return ERR_PTR(-ENOMEM); |
726 | ||
3895dbf8 EB |
727 | |
728 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 729 | ret = d_set_mounted(dentry); |
eed81007 | 730 | |
3895dbf8 EB |
731 | /* Someone else set d_mounted? */ |
732 | if (ret == -EBUSY) | |
733 | goto mountpoint; | |
734 | ||
735 | /* The dentry is not available as a mountpoint? */ | |
736 | mp = ERR_PTR(ret); | |
737 | if (ret) | |
738 | goto done; | |
739 | ||
740 | /* Add the new mountpoint to the hash table */ | |
741 | read_seqlock_excl(&mount_lock); | |
742 | new->m_dentry = dentry; | |
743 | new->m_count = 1; | |
744 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
745 | INIT_HLIST_HEAD(&new->m_list); | |
746 | read_sequnlock_excl(&mount_lock); | |
747 | ||
748 | mp = new; | |
749 | new = NULL; | |
750 | done: | |
751 | kfree(new); | |
84d17192 AV |
752 | return mp; |
753 | } | |
754 | ||
755 | static void put_mountpoint(struct mountpoint *mp) | |
756 | { | |
757 | if (!--mp->m_count) { | |
758 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 759 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
760 | spin_lock(&dentry->d_lock); |
761 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
762 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 763 | hlist_del(&mp->m_hash); |
84d17192 AV |
764 | kfree(mp); |
765 | } | |
766 | } | |
767 | ||
143c8c91 | 768 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 769 | { |
6b3286ed | 770 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
771 | } |
772 | ||
99b7db7b NP |
773 | /* |
774 | * vfsmount lock must be held for write | |
775 | */ | |
6b3286ed | 776 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
777 | { |
778 | if (ns) { | |
779 | ns->event = ++event; | |
780 | wake_up_interruptible(&ns->poll); | |
781 | } | |
782 | } | |
783 | ||
99b7db7b NP |
784 | /* |
785 | * vfsmount lock must be held for write | |
786 | */ | |
6b3286ed | 787 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
788 | { |
789 | if (ns && ns->event != event) { | |
790 | ns->event = event; | |
791 | wake_up_interruptible(&ns->poll); | |
792 | } | |
793 | } | |
794 | ||
99b7db7b NP |
795 | /* |
796 | * vfsmount lock must be held for write | |
797 | */ | |
7bdb11de | 798 | static void unhash_mnt(struct mount *mnt) |
419148da | 799 | { |
0714a533 | 800 | mnt->mnt_parent = mnt; |
a73324da | 801 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 802 | list_del_init(&mnt->mnt_child); |
38129a13 | 803 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 804 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
805 | put_mountpoint(mnt->mnt_mp); |
806 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
807 | } |
808 | ||
7bdb11de EB |
809 | /* |
810 | * vfsmount lock must be held for write | |
811 | */ | |
812 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
813 | { | |
814 | old_path->dentry = mnt->mnt_mountpoint; | |
815 | old_path->mnt = &mnt->mnt_parent->mnt; | |
816 | unhash_mnt(mnt); | |
817 | } | |
818 | ||
6a46c573 EB |
819 | /* |
820 | * vfsmount lock must be held for write | |
821 | */ | |
822 | static void umount_mnt(struct mount *mnt) | |
823 | { | |
824 | /* old mountpoint will be dropped when we can do that */ | |
825 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
826 | unhash_mnt(mnt); | |
827 | } | |
828 | ||
99b7db7b NP |
829 | /* |
830 | * vfsmount lock must be held for write | |
831 | */ | |
84d17192 AV |
832 | void mnt_set_mountpoint(struct mount *mnt, |
833 | struct mountpoint *mp, | |
44d964d6 | 834 | struct mount *child_mnt) |
b90fa9ae | 835 | { |
84d17192 | 836 | mp->m_count++; |
3a2393d7 | 837 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 838 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 839 | child_mnt->mnt_parent = mnt; |
84d17192 | 840 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 841 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
842 | } |
843 | ||
1064f874 EB |
844 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
845 | { | |
846 | hlist_add_head_rcu(&mnt->mnt_hash, | |
847 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
848 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
849 | } | |
850 | ||
99b7db7b NP |
851 | /* |
852 | * vfsmount lock must be held for write | |
853 | */ | |
84d17192 AV |
854 | static void attach_mnt(struct mount *mnt, |
855 | struct mount *parent, | |
856 | struct mountpoint *mp) | |
1da177e4 | 857 | { |
84d17192 | 858 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 859 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
860 | } |
861 | ||
1064f874 | 862 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 863 | { |
1064f874 EB |
864 | struct mountpoint *old_mp = mnt->mnt_mp; |
865 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
866 | struct mount *old_parent = mnt->mnt_parent; | |
867 | ||
868 | list_del_init(&mnt->mnt_child); | |
869 | hlist_del_init(&mnt->mnt_mp_list); | |
870 | hlist_del_init_rcu(&mnt->mnt_hash); | |
871 | ||
872 | attach_mnt(mnt, parent, mp); | |
873 | ||
874 | put_mountpoint(old_mp); | |
875 | ||
876 | /* | |
877 | * Safely avoid even the suggestion this code might sleep or | |
878 | * lock the mount hash by taking advantage of the knowledge that | |
879 | * mnt_change_mountpoint will not release the final reference | |
880 | * to a mountpoint. | |
881 | * | |
882 | * During mounting, the mount passed in as the parent mount will | |
883 | * continue to use the old mountpoint and during unmounting, the | |
884 | * old mountpoint will continue to exist until namespace_unlock, | |
885 | * which happens well after mnt_change_mountpoint. | |
886 | */ | |
887 | spin_lock(&old_mountpoint->d_lock); | |
888 | old_mountpoint->d_lockref.count--; | |
889 | spin_unlock(&old_mountpoint->d_lock); | |
890 | ||
891 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
892 | } |
893 | ||
b90fa9ae | 894 | /* |
99b7db7b | 895 | * vfsmount lock must be held for write |
b90fa9ae | 896 | */ |
1064f874 | 897 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 898 | { |
0714a533 | 899 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 900 | struct mount *m; |
b90fa9ae | 901 | LIST_HEAD(head); |
143c8c91 | 902 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 903 | |
0714a533 | 904 | BUG_ON(parent == mnt); |
b90fa9ae | 905 | |
1a4eeaf2 | 906 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 907 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 908 | m->mnt_ns = n; |
f03c6599 | 909 | |
b90fa9ae RP |
910 | list_splice(&head, n->list.prev); |
911 | ||
d2921684 EB |
912 | n->mounts += n->pending_mounts; |
913 | n->pending_mounts = 0; | |
914 | ||
1064f874 | 915 | __attach_mnt(mnt, parent); |
6b3286ed | 916 | touch_mnt_namespace(n); |
1da177e4 LT |
917 | } |
918 | ||
909b0a88 | 919 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 920 | { |
6b41d536 AV |
921 | struct list_head *next = p->mnt_mounts.next; |
922 | if (next == &p->mnt_mounts) { | |
1da177e4 | 923 | while (1) { |
909b0a88 | 924 | if (p == root) |
1da177e4 | 925 | return NULL; |
6b41d536 AV |
926 | next = p->mnt_child.next; |
927 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 928 | break; |
0714a533 | 929 | p = p->mnt_parent; |
1da177e4 LT |
930 | } |
931 | } | |
6b41d536 | 932 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
933 | } |
934 | ||
315fc83e | 935 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 936 | { |
6b41d536 AV |
937 | struct list_head *prev = p->mnt_mounts.prev; |
938 | while (prev != &p->mnt_mounts) { | |
939 | p = list_entry(prev, struct mount, mnt_child); | |
940 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
941 | } |
942 | return p; | |
943 | } | |
944 | ||
8f291889 AV |
945 | /** |
946 | * vfs_create_mount - Create a mount for a configured superblock | |
947 | * @fc: The configuration context with the superblock attached | |
948 | * | |
949 | * Create a mount to an already configured superblock. If necessary, the | |
950 | * caller should invoke vfs_get_tree() before calling this. | |
951 | * | |
952 | * Note that this does not attach the mount to anything. | |
953 | */ | |
954 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 955 | { |
b105e270 | 956 | struct mount *mnt; |
9d412a43 | 957 | |
8f291889 AV |
958 | if (!fc->root) |
959 | return ERR_PTR(-EINVAL); | |
9d412a43 | 960 | |
8f291889 | 961 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
962 | if (!mnt) |
963 | return ERR_PTR(-ENOMEM); | |
964 | ||
8f291889 | 965 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 966 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 967 | |
8f291889 AV |
968 | atomic_inc(&fc->root->d_sb->s_active); |
969 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
970 | mnt->mnt.mnt_root = dget(fc->root); | |
971 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
972 | mnt->mnt_parent = mnt; | |
9d412a43 | 973 | |
719ea2fb | 974 | lock_mount_hash(); |
8f291889 | 975 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 976 | unlock_mount_hash(); |
b105e270 | 977 | return &mnt->mnt; |
9d412a43 | 978 | } |
8f291889 AV |
979 | EXPORT_SYMBOL(vfs_create_mount); |
980 | ||
981 | struct vfsmount *fc_mount(struct fs_context *fc) | |
982 | { | |
983 | int err = vfs_get_tree(fc); | |
984 | if (!err) { | |
985 | up_write(&fc->root->d_sb->s_umount); | |
986 | return vfs_create_mount(fc); | |
987 | } | |
988 | return ERR_PTR(err); | |
989 | } | |
990 | EXPORT_SYMBOL(fc_mount); | |
991 | ||
9bc61ab1 DH |
992 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
993 | int flags, const char *name, | |
994 | void *data) | |
9d412a43 | 995 | { |
9bc61ab1 | 996 | struct fs_context *fc; |
8f291889 | 997 | struct vfsmount *mnt; |
9bc61ab1 | 998 | int ret = 0; |
9d412a43 AV |
999 | |
1000 | if (!type) | |
3e1aeb00 | 1001 | return ERR_PTR(-EINVAL); |
9d412a43 | 1002 | |
9bc61ab1 DH |
1003 | fc = fs_context_for_mount(type, flags); |
1004 | if (IS_ERR(fc)) | |
1005 | return ERR_CAST(fc); | |
1006 | ||
3e1aeb00 DH |
1007 | if (name) |
1008 | ret = vfs_parse_fs_string(fc, "source", | |
1009 | name, strlen(name)); | |
9bc61ab1 DH |
1010 | if (!ret) |
1011 | ret = parse_monolithic_mount_data(fc, data); | |
1012 | if (!ret) | |
8f291889 AV |
1013 | mnt = fc_mount(fc); |
1014 | else | |
1015 | mnt = ERR_PTR(ret); | |
9d412a43 | 1016 | |
9bc61ab1 | 1017 | put_fs_context(fc); |
8f291889 | 1018 | return mnt; |
9d412a43 AV |
1019 | } |
1020 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1021 | ||
93faccbb EB |
1022 | struct vfsmount * |
1023 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1024 | const char *name, void *data) | |
1025 | { | |
1026 | /* Until it is worked out how to pass the user namespace | |
1027 | * through from the parent mount to the submount don't support | |
1028 | * unprivileged mounts with submounts. | |
1029 | */ | |
1030 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1031 | return ERR_PTR(-EPERM); | |
1032 | ||
e462ec50 | 1033 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1034 | } |
1035 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1036 | ||
87129cc0 | 1037 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1038 | int flag) |
1da177e4 | 1039 | { |
87129cc0 | 1040 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1041 | struct mount *mnt; |
1042 | int err; | |
1da177e4 | 1043 | |
be34d1a3 DH |
1044 | mnt = alloc_vfsmnt(old->mnt_devname); |
1045 | if (!mnt) | |
1046 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1047 | |
7a472ef4 | 1048 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1049 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1050 | else | |
1051 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1052 | |
be34d1a3 DH |
1053 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1054 | err = mnt_alloc_group_id(mnt); | |
1055 | if (err) | |
1056 | goto out_free; | |
1da177e4 | 1057 | } |
be34d1a3 | 1058 | |
16a34adb AV |
1059 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1060 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1061 | |
be34d1a3 DH |
1062 | atomic_inc(&sb->s_active); |
1063 | mnt->mnt.mnt_sb = sb; | |
1064 | mnt->mnt.mnt_root = dget(root); | |
1065 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1066 | mnt->mnt_parent = mnt; | |
719ea2fb | 1067 | lock_mount_hash(); |
be34d1a3 | 1068 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1069 | unlock_mount_hash(); |
be34d1a3 | 1070 | |
7a472ef4 EB |
1071 | if ((flag & CL_SLAVE) || |
1072 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1073 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1074 | mnt->mnt_master = old; | |
1075 | CLEAR_MNT_SHARED(mnt); | |
1076 | } else if (!(flag & CL_PRIVATE)) { | |
1077 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1078 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1079 | if (IS_MNT_SLAVE(old)) | |
1080 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1081 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1082 | } else { |
1083 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1084 | } |
1085 | if (flag & CL_MAKE_SHARED) | |
1086 | set_mnt_shared(mnt); | |
1087 | ||
1088 | /* stick the duplicate mount on the same expiry list | |
1089 | * as the original if that was on one */ | |
1090 | if (flag & CL_EXPIRE) { | |
1091 | if (!list_empty(&old->mnt_expire)) | |
1092 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1093 | } | |
1094 | ||
cb338d06 | 1095 | return mnt; |
719f5d7f MS |
1096 | |
1097 | out_free: | |
8ffcb32e | 1098 | mnt_free_id(mnt); |
719f5d7f | 1099 | free_vfsmnt(mnt); |
be34d1a3 | 1100 | return ERR_PTR(err); |
1da177e4 LT |
1101 | } |
1102 | ||
9ea459e1 AV |
1103 | static void cleanup_mnt(struct mount *mnt) |
1104 | { | |
1105 | /* | |
1106 | * This probably indicates that somebody messed | |
1107 | * up a mnt_want/drop_write() pair. If this | |
1108 | * happens, the filesystem was probably unable | |
1109 | * to make r/w->r/o transitions. | |
1110 | */ | |
1111 | /* | |
1112 | * The locking used to deal with mnt_count decrement provides barriers, | |
1113 | * so mnt_get_writers() below is safe. | |
1114 | */ | |
1115 | WARN_ON(mnt_get_writers(mnt)); | |
1116 | if (unlikely(mnt->mnt_pins.first)) | |
1117 | mnt_pin_kill(mnt); | |
1118 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1119 | dput(mnt->mnt.mnt_root); | |
1120 | deactivate_super(mnt->mnt.mnt_sb); | |
1121 | mnt_free_id(mnt); | |
1122 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1123 | } | |
1124 | ||
1125 | static void __cleanup_mnt(struct rcu_head *head) | |
1126 | { | |
1127 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1128 | } | |
1129 | ||
1130 | static LLIST_HEAD(delayed_mntput_list); | |
1131 | static void delayed_mntput(struct work_struct *unused) | |
1132 | { | |
1133 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1134 | struct mount *m, *t; |
9ea459e1 | 1135 | |
29785735 BP |
1136 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1137 | cleanup_mnt(m); | |
9ea459e1 AV |
1138 | } |
1139 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1140 | ||
900148dc | 1141 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1142 | { |
48a066e7 | 1143 | rcu_read_lock(); |
9ea0a46c AV |
1144 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1145 | /* | |
1146 | * Since we don't do lock_mount_hash() here, | |
1147 | * ->mnt_ns can change under us. However, if it's | |
1148 | * non-NULL, then there's a reference that won't | |
1149 | * be dropped until after an RCU delay done after | |
1150 | * turning ->mnt_ns NULL. So if we observe it | |
1151 | * non-NULL under rcu_read_lock(), the reference | |
1152 | * we are dropping is not the final one. | |
1153 | */ | |
1154 | mnt_add_count(mnt, -1); | |
48a066e7 | 1155 | rcu_read_unlock(); |
f03c6599 | 1156 | return; |
b3e19d92 | 1157 | } |
719ea2fb | 1158 | lock_mount_hash(); |
119e1ef8 AV |
1159 | /* |
1160 | * make sure that if __legitimize_mnt() has not seen us grab | |
1161 | * mount_lock, we'll see their refcount increment here. | |
1162 | */ | |
1163 | smp_mb(); | |
9ea0a46c | 1164 | mnt_add_count(mnt, -1); |
b3e19d92 | 1165 | if (mnt_get_count(mnt)) { |
48a066e7 | 1166 | rcu_read_unlock(); |
719ea2fb | 1167 | unlock_mount_hash(); |
99b7db7b NP |
1168 | return; |
1169 | } | |
48a066e7 AV |
1170 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1171 | rcu_read_unlock(); | |
1172 | unlock_mount_hash(); | |
1173 | return; | |
1174 | } | |
1175 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1176 | rcu_read_unlock(); | |
962830df | 1177 | |
39f7c4db | 1178 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1179 | |
1180 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1181 | struct mount *p, *tmp; | |
1182 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1183 | umount_mnt(p); | |
1184 | } | |
1185 | } | |
719ea2fb | 1186 | unlock_mount_hash(); |
649a795a | 1187 | |
9ea459e1 AV |
1188 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1189 | struct task_struct *task = current; | |
1190 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1191 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1192 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1193 | return; | |
1194 | } | |
1195 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1196 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1197 | return; | |
1198 | } | |
1199 | cleanup_mnt(mnt); | |
b3e19d92 | 1200 | } |
b3e19d92 NP |
1201 | |
1202 | void mntput(struct vfsmount *mnt) | |
1203 | { | |
1204 | if (mnt) { | |
863d684f | 1205 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1206 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1207 | if (unlikely(m->mnt_expiry_mark)) |
1208 | m->mnt_expiry_mark = 0; | |
1209 | mntput_no_expire(m); | |
b3e19d92 NP |
1210 | } |
1211 | } | |
1212 | EXPORT_SYMBOL(mntput); | |
1213 | ||
1214 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1215 | { | |
1216 | if (mnt) | |
83adc753 | 1217 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1218 | return mnt; |
1219 | } | |
1220 | EXPORT_SYMBOL(mntget); | |
1221 | ||
c6609c0a IK |
1222 | /* path_is_mountpoint() - Check if path is a mount in the current |
1223 | * namespace. | |
1224 | * | |
1225 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1226 | * not mounted in any namespace and that common case is handled inline. | |
1227 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1228 | * mounts using a given dentry in a different namespace. This function | |
1229 | * checks if the passed in path is a mountpoint rather than the dentry | |
1230 | * alone. | |
1231 | */ | |
1232 | bool path_is_mountpoint(const struct path *path) | |
1233 | { | |
1234 | unsigned seq; | |
1235 | bool res; | |
1236 | ||
1237 | if (!d_mountpoint(path->dentry)) | |
1238 | return false; | |
1239 | ||
1240 | rcu_read_lock(); | |
1241 | do { | |
1242 | seq = read_seqbegin(&mount_lock); | |
1243 | res = __path_is_mountpoint(path); | |
1244 | } while (read_seqretry(&mount_lock, seq)); | |
1245 | rcu_read_unlock(); | |
1246 | ||
1247 | return res; | |
1248 | } | |
1249 | EXPORT_SYMBOL(path_is_mountpoint); | |
1250 | ||
ca71cf71 | 1251 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1252 | { |
3064c356 AV |
1253 | struct mount *p; |
1254 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1255 | if (IS_ERR(p)) | |
1256 | return ERR_CAST(p); | |
1257 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1258 | return &p->mnt; | |
7b7b1ace | 1259 | } |
1da177e4 | 1260 | |
a1a2c409 | 1261 | #ifdef CONFIG_PROC_FS |
0226f492 | 1262 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1263 | static void *m_start(struct seq_file *m, loff_t *pos) |
1264 | { | |
ede1bf0d | 1265 | struct proc_mounts *p = m->private; |
1da177e4 | 1266 | |
390c6843 | 1267 | down_read(&namespace_sem); |
c7999c36 AV |
1268 | if (p->cached_event == p->ns->event) { |
1269 | void *v = p->cached_mount; | |
1270 | if (*pos == p->cached_index) | |
1271 | return v; | |
1272 | if (*pos == p->cached_index + 1) { | |
1273 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1274 | return p->cached_mount = v; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | p->cached_event = p->ns->event; | |
1279 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1280 | p->cached_index = *pos; | |
1281 | return p->cached_mount; | |
1da177e4 LT |
1282 | } |
1283 | ||
1284 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1285 | { | |
ede1bf0d | 1286 | struct proc_mounts *p = m->private; |
b0765fb8 | 1287 | |
c7999c36 AV |
1288 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1289 | p->cached_index = *pos; | |
1290 | return p->cached_mount; | |
1da177e4 LT |
1291 | } |
1292 | ||
1293 | static void m_stop(struct seq_file *m, void *v) | |
1294 | { | |
390c6843 | 1295 | up_read(&namespace_sem); |
1da177e4 LT |
1296 | } |
1297 | ||
0226f492 | 1298 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1299 | { |
ede1bf0d | 1300 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1301 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1302 | return p->show(m, &r->mnt); |
1da177e4 LT |
1303 | } |
1304 | ||
a1a2c409 | 1305 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1306 | .start = m_start, |
1307 | .next = m_next, | |
1308 | .stop = m_stop, | |
0226f492 | 1309 | .show = m_show, |
b4629fe2 | 1310 | }; |
a1a2c409 | 1311 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1312 | |
1da177e4 LT |
1313 | /** |
1314 | * may_umount_tree - check if a mount tree is busy | |
1315 | * @mnt: root of mount tree | |
1316 | * | |
1317 | * This is called to check if a tree of mounts has any | |
1318 | * open files, pwds, chroots or sub mounts that are | |
1319 | * busy. | |
1320 | */ | |
909b0a88 | 1321 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1322 | { |
909b0a88 | 1323 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1324 | int actual_refs = 0; |
1325 | int minimum_refs = 0; | |
315fc83e | 1326 | struct mount *p; |
909b0a88 | 1327 | BUG_ON(!m); |
1da177e4 | 1328 | |
b3e19d92 | 1329 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1330 | lock_mount_hash(); |
909b0a88 | 1331 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1332 | actual_refs += mnt_get_count(p); |
1da177e4 | 1333 | minimum_refs += 2; |
1da177e4 | 1334 | } |
719ea2fb | 1335 | unlock_mount_hash(); |
1da177e4 LT |
1336 | |
1337 | if (actual_refs > minimum_refs) | |
e3474a8e | 1338 | return 0; |
1da177e4 | 1339 | |
e3474a8e | 1340 | return 1; |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | EXPORT_SYMBOL(may_umount_tree); | |
1344 | ||
1345 | /** | |
1346 | * may_umount - check if a mount point is busy | |
1347 | * @mnt: root of mount | |
1348 | * | |
1349 | * This is called to check if a mount point has any | |
1350 | * open files, pwds, chroots or sub mounts. If the | |
1351 | * mount has sub mounts this will return busy | |
1352 | * regardless of whether the sub mounts are busy. | |
1353 | * | |
1354 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1355 | * give false negatives. The main reason why it's here is that we need | |
1356 | * a non-destructive way to look for easily umountable filesystems. | |
1357 | */ | |
1358 | int may_umount(struct vfsmount *mnt) | |
1359 | { | |
e3474a8e | 1360 | int ret = 1; |
8ad08d8a | 1361 | down_read(&namespace_sem); |
719ea2fb | 1362 | lock_mount_hash(); |
1ab59738 | 1363 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1364 | ret = 0; |
719ea2fb | 1365 | unlock_mount_hash(); |
8ad08d8a | 1366 | up_read(&namespace_sem); |
a05964f3 | 1367 | return ret; |
1da177e4 LT |
1368 | } |
1369 | ||
1370 | EXPORT_SYMBOL(may_umount); | |
1371 | ||
38129a13 | 1372 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1373 | |
97216be0 | 1374 | static void namespace_unlock(void) |
70fbcdf4 | 1375 | { |
a3b3c562 | 1376 | struct hlist_head head; |
97216be0 | 1377 | |
a3b3c562 | 1378 | hlist_move_list(&unmounted, &head); |
97216be0 | 1379 | |
97216be0 AV |
1380 | up_write(&namespace_sem); |
1381 | ||
a3b3c562 EB |
1382 | if (likely(hlist_empty(&head))) |
1383 | return; | |
1384 | ||
22cb7405 | 1385 | synchronize_rcu_expedited(); |
48a066e7 | 1386 | |
87b95ce0 | 1387 | group_pin_kill(&head); |
70fbcdf4 RP |
1388 | } |
1389 | ||
97216be0 | 1390 | static inline void namespace_lock(void) |
e3197d83 | 1391 | { |
97216be0 | 1392 | down_write(&namespace_sem); |
e3197d83 AV |
1393 | } |
1394 | ||
e819f152 EB |
1395 | enum umount_tree_flags { |
1396 | UMOUNT_SYNC = 1, | |
1397 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1398 | UMOUNT_CONNECTED = 4, |
e819f152 | 1399 | }; |
f2d0a123 EB |
1400 | |
1401 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1402 | { | |
1403 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1404 | if (how & UMOUNT_SYNC) | |
1405 | return true; | |
1406 | ||
1407 | /* A mount without a parent has nothing to be connected to */ | |
1408 | if (!mnt_has_parent(mnt)) | |
1409 | return true; | |
1410 | ||
1411 | /* Because the reference counting rules change when mounts are | |
1412 | * unmounted and connected, umounted mounts may not be | |
1413 | * connected to mounted mounts. | |
1414 | */ | |
1415 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1416 | return true; | |
1417 | ||
1418 | /* Has it been requested that the mount remain connected? */ | |
1419 | if (how & UMOUNT_CONNECTED) | |
1420 | return false; | |
1421 | ||
1422 | /* Is the mount locked such that it needs to remain connected? */ | |
1423 | if (IS_MNT_LOCKED(mnt)) | |
1424 | return false; | |
1425 | ||
1426 | /* By default disconnect the mount */ | |
1427 | return true; | |
1428 | } | |
1429 | ||
99b7db7b | 1430 | /* |
48a066e7 | 1431 | * mount_lock must be held |
99b7db7b NP |
1432 | * namespace_sem must be held for write |
1433 | */ | |
e819f152 | 1434 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1435 | { |
c003b26f | 1436 | LIST_HEAD(tmp_list); |
315fc83e | 1437 | struct mount *p; |
1da177e4 | 1438 | |
5d88457e EB |
1439 | if (how & UMOUNT_PROPAGATE) |
1440 | propagate_mount_unlock(mnt); | |
1441 | ||
c003b26f | 1442 | /* Gather the mounts to umount */ |
590ce4bc EB |
1443 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1444 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1445 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1446 | } |
1da177e4 | 1447 | |
411a938b | 1448 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1449 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1450 | list_del_init(&p->mnt_child); |
c003b26f | 1451 | } |
88b368f2 | 1452 | |
c003b26f | 1453 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1454 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1455 | propagate_umount(&tmp_list); |
a05964f3 | 1456 | |
c003b26f | 1457 | while (!list_empty(&tmp_list)) { |
d2921684 | 1458 | struct mnt_namespace *ns; |
ce07d891 | 1459 | bool disconnect; |
c003b26f | 1460 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1461 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1462 | list_del_init(&p->mnt_list); |
d2921684 EB |
1463 | ns = p->mnt_ns; |
1464 | if (ns) { | |
1465 | ns->mounts--; | |
1466 | __touch_mnt_namespace(ns); | |
1467 | } | |
143c8c91 | 1468 | p->mnt_ns = NULL; |
e819f152 | 1469 | if (how & UMOUNT_SYNC) |
48a066e7 | 1470 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1471 | |
f2d0a123 | 1472 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1473 | |
1474 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1475 | disconnect ? &unmounted : NULL); | |
676da58d | 1476 | if (mnt_has_parent(p)) { |
81b6b061 | 1477 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1478 | if (!disconnect) { |
1479 | /* Don't forget about p */ | |
1480 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1481 | } else { | |
1482 | umount_mnt(p); | |
1483 | } | |
7c4b93d8 | 1484 | } |
0f0afb1d | 1485 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1486 | } |
1487 | } | |
1488 | ||
b54b9be7 | 1489 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1490 | |
8d0347f6 DH |
1491 | static int do_umount_root(struct super_block *sb) |
1492 | { | |
1493 | int ret = 0; | |
1494 | ||
1495 | down_write(&sb->s_umount); | |
1496 | if (!sb_rdonly(sb)) { | |
1497 | struct fs_context *fc; | |
1498 | ||
1499 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1500 | SB_RDONLY); | |
1501 | if (IS_ERR(fc)) { | |
1502 | ret = PTR_ERR(fc); | |
1503 | } else { | |
1504 | ret = parse_monolithic_mount_data(fc, NULL); | |
1505 | if (!ret) | |
1506 | ret = reconfigure_super(fc); | |
1507 | put_fs_context(fc); | |
1508 | } | |
1509 | } | |
1510 | up_write(&sb->s_umount); | |
1511 | return ret; | |
1512 | } | |
1513 | ||
1ab59738 | 1514 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1515 | { |
1ab59738 | 1516 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1517 | int retval; |
1518 | ||
1ab59738 | 1519 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1520 | if (retval) |
1521 | return retval; | |
1522 | ||
1523 | /* | |
1524 | * Allow userspace to request a mountpoint be expired rather than | |
1525 | * unmounting unconditionally. Unmount only happens if: | |
1526 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1527 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1528 | */ | |
1529 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1530 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1531 | flags & (MNT_FORCE | MNT_DETACH)) |
1532 | return -EINVAL; | |
1533 | ||
b3e19d92 NP |
1534 | /* |
1535 | * probably don't strictly need the lock here if we examined | |
1536 | * all race cases, but it's a slowpath. | |
1537 | */ | |
719ea2fb | 1538 | lock_mount_hash(); |
83adc753 | 1539 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1540 | unlock_mount_hash(); |
1da177e4 | 1541 | return -EBUSY; |
b3e19d92 | 1542 | } |
719ea2fb | 1543 | unlock_mount_hash(); |
1da177e4 | 1544 | |
863d684f | 1545 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1546 | return -EAGAIN; |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * If we may have to abort operations to get out of this | |
1551 | * mount, and they will themselves hold resources we must | |
1552 | * allow the fs to do things. In the Unix tradition of | |
1553 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1554 | * might fail to complete on the first run through as other tasks | |
1555 | * must return, and the like. Thats for the mount program to worry | |
1556 | * about for the moment. | |
1557 | */ | |
1558 | ||
42faad99 | 1559 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1560 | sb->s_op->umount_begin(sb); |
42faad99 | 1561 | } |
1da177e4 LT |
1562 | |
1563 | /* | |
1564 | * No sense to grab the lock for this test, but test itself looks | |
1565 | * somewhat bogus. Suggestions for better replacement? | |
1566 | * Ho-hum... In principle, we might treat that as umount + switch | |
1567 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1568 | * Actually it makes sense, especially if rootfs would contain a | |
1569 | * /reboot - static binary that would close all descriptors and | |
1570 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1571 | */ | |
1ab59738 | 1572 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1573 | /* |
1574 | * Special case for "unmounting" root ... | |
1575 | * we just try to remount it readonly. | |
1576 | */ | |
bc6155d1 | 1577 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1578 | return -EPERM; |
8d0347f6 | 1579 | return do_umount_root(sb); |
1da177e4 LT |
1580 | } |
1581 | ||
97216be0 | 1582 | namespace_lock(); |
719ea2fb | 1583 | lock_mount_hash(); |
1da177e4 | 1584 | |
25d202ed EB |
1585 | /* Recheck MNT_LOCKED with the locks held */ |
1586 | retval = -EINVAL; | |
1587 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1588 | goto out; | |
1589 | ||
1590 | event++; | |
48a066e7 | 1591 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1592 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1593 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1594 | retval = 0; |
48a066e7 AV |
1595 | } else { |
1596 | shrink_submounts(mnt); | |
1597 | retval = -EBUSY; | |
1598 | if (!propagate_mount_busy(mnt, 2)) { | |
1599 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1600 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1601 | retval = 0; |
1602 | } | |
1da177e4 | 1603 | } |
25d202ed | 1604 | out: |
719ea2fb | 1605 | unlock_mount_hash(); |
e3197d83 | 1606 | namespace_unlock(); |
1da177e4 LT |
1607 | return retval; |
1608 | } | |
1609 | ||
80b5dce8 EB |
1610 | /* |
1611 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1612 | * | |
1613 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1614 | * to an existing mountpoint, and wind up leaking the mount. | |
1615 | * detach_mounts allows lazily unmounting those mounts instead of | |
1616 | * leaking them. | |
1617 | * | |
1618 | * The caller may hold dentry->d_inode->i_mutex. | |
1619 | */ | |
1620 | void __detach_mounts(struct dentry *dentry) | |
1621 | { | |
1622 | struct mountpoint *mp; | |
1623 | struct mount *mnt; | |
1624 | ||
1625 | namespace_lock(); | |
3895dbf8 | 1626 | lock_mount_hash(); |
80b5dce8 | 1627 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1628 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1629 | goto out_unlock; |
1630 | ||
e06b933e | 1631 | event++; |
80b5dce8 EB |
1632 | while (!hlist_empty(&mp->m_list)) { |
1633 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1634 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1635 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1636 | umount_mnt(mnt); | |
ce07d891 | 1637 | } |
e0c9c0af | 1638 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1639 | } |
80b5dce8 EB |
1640 | put_mountpoint(mp); |
1641 | out_unlock: | |
3895dbf8 | 1642 | unlock_mount_hash(); |
80b5dce8 EB |
1643 | namespace_unlock(); |
1644 | } | |
1645 | ||
dd111b31 | 1646 | /* |
9b40bc90 AV |
1647 | * Is the caller allowed to modify his namespace? |
1648 | */ | |
1649 | static inline bool may_mount(void) | |
1650 | { | |
1651 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1652 | } | |
1653 | ||
9e8925b6 JL |
1654 | static inline bool may_mandlock(void) |
1655 | { | |
1656 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1657 | return false; | |
1658 | #endif | |
95ace754 | 1659 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1660 | } |
1661 | ||
1da177e4 LT |
1662 | /* |
1663 | * Now umount can handle mount points as well as block devices. | |
1664 | * This is important for filesystems which use unnamed block devices. | |
1665 | * | |
1666 | * We now support a flag for forced unmount like the other 'big iron' | |
1667 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1668 | */ | |
1669 | ||
3a18ef5c | 1670 | int ksys_umount(char __user *name, int flags) |
1da177e4 | 1671 | { |
2d8f3038 | 1672 | struct path path; |
900148dc | 1673 | struct mount *mnt; |
1da177e4 | 1674 | int retval; |
db1f05bb | 1675 | int lookup_flags = 0; |
1da177e4 | 1676 | |
db1f05bb MS |
1677 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1678 | return -EINVAL; | |
1679 | ||
9b40bc90 AV |
1680 | if (!may_mount()) |
1681 | return -EPERM; | |
1682 | ||
db1f05bb MS |
1683 | if (!(flags & UMOUNT_NOFOLLOW)) |
1684 | lookup_flags |= LOOKUP_FOLLOW; | |
1685 | ||
57d46577 RGB |
1686 | lookup_flags |= LOOKUP_NO_EVAL; |
1687 | ||
197df04c | 1688 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1689 | if (retval) |
1690 | goto out; | |
900148dc | 1691 | mnt = real_mount(path.mnt); |
1da177e4 | 1692 | retval = -EINVAL; |
2d8f3038 | 1693 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1694 | goto dput_and_out; |
143c8c91 | 1695 | if (!check_mnt(mnt)) |
1da177e4 | 1696 | goto dput_and_out; |
25d202ed | 1697 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
5ff9d8a6 | 1698 | goto dput_and_out; |
b2f5d4dc EB |
1699 | retval = -EPERM; |
1700 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1701 | goto dput_and_out; | |
1da177e4 | 1702 | |
900148dc | 1703 | retval = do_umount(mnt, flags); |
1da177e4 | 1704 | dput_and_out: |
429731b1 | 1705 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1706 | dput(path.dentry); |
900148dc | 1707 | mntput_no_expire(mnt); |
1da177e4 LT |
1708 | out: |
1709 | return retval; | |
1710 | } | |
1711 | ||
3a18ef5c DB |
1712 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1713 | { | |
1714 | return ksys_umount(name, flags); | |
1715 | } | |
1716 | ||
1da177e4 LT |
1717 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1718 | ||
1719 | /* | |
b58fed8b | 1720 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1721 | */ |
bdc480e3 | 1722 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1723 | { |
3a18ef5c | 1724 | return ksys_umount(name, 0); |
1da177e4 LT |
1725 | } |
1726 | ||
1727 | #endif | |
1728 | ||
4ce5d2b1 | 1729 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1730 | { |
4ce5d2b1 | 1731 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1732 | return dentry->d_op == &ns_dentry_operations && |
1733 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1734 | } |
1735 | ||
58be2825 AV |
1736 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1737 | { | |
1738 | return container_of(ns, struct mnt_namespace, ns); | |
1739 | } | |
1740 | ||
4ce5d2b1 EB |
1741 | static bool mnt_ns_loop(struct dentry *dentry) |
1742 | { | |
1743 | /* Could bind mounting the mount namespace inode cause a | |
1744 | * mount namespace loop? | |
1745 | */ | |
1746 | struct mnt_namespace *mnt_ns; | |
1747 | if (!is_mnt_ns_file(dentry)) | |
1748 | return false; | |
1749 | ||
f77c8014 | 1750 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1751 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1752 | } | |
1753 | ||
87129cc0 | 1754 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1755 | int flag) |
1da177e4 | 1756 | { |
84d17192 | 1757 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1758 | |
4ce5d2b1 EB |
1759 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1760 | return ERR_PTR(-EINVAL); | |
1761 | ||
1762 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1763 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1764 | |
36341f64 | 1765 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1766 | if (IS_ERR(q)) |
1767 | return q; | |
1768 | ||
a73324da | 1769 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1770 | |
1771 | p = mnt; | |
6b41d536 | 1772 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1773 | struct mount *s; |
7ec02ef1 | 1774 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1775 | continue; |
1776 | ||
909b0a88 | 1777 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1778 | if (!(flag & CL_COPY_UNBINDABLE) && |
1779 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1780 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1781 | /* Both unbindable and locked. */ | |
1782 | q = ERR_PTR(-EPERM); | |
1783 | goto out; | |
1784 | } else { | |
1785 | s = skip_mnt_tree(s); | |
1786 | continue; | |
1787 | } | |
4ce5d2b1 EB |
1788 | } |
1789 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1790 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1791 | s = skip_mnt_tree(s); |
1792 | continue; | |
1793 | } | |
0714a533 AV |
1794 | while (p != s->mnt_parent) { |
1795 | p = p->mnt_parent; | |
1796 | q = q->mnt_parent; | |
1da177e4 | 1797 | } |
87129cc0 | 1798 | p = s; |
84d17192 | 1799 | parent = q; |
87129cc0 | 1800 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1801 | if (IS_ERR(q)) |
1802 | goto out; | |
719ea2fb | 1803 | lock_mount_hash(); |
1a4eeaf2 | 1804 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1805 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1806 | unlock_mount_hash(); |
1da177e4 LT |
1807 | } |
1808 | } | |
1809 | return res; | |
be34d1a3 | 1810 | out: |
1da177e4 | 1811 | if (res) { |
719ea2fb | 1812 | lock_mount_hash(); |
e819f152 | 1813 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1814 | unlock_mount_hash(); |
1da177e4 | 1815 | } |
be34d1a3 | 1816 | return q; |
1da177e4 LT |
1817 | } |
1818 | ||
be34d1a3 DH |
1819 | /* Caller should check returned pointer for errors */ |
1820 | ||
ca71cf71 | 1821 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1822 | { |
cb338d06 | 1823 | struct mount *tree; |
97216be0 | 1824 | namespace_lock(); |
cd4a4017 EB |
1825 | if (!check_mnt(real_mount(path->mnt))) |
1826 | tree = ERR_PTR(-EINVAL); | |
1827 | else | |
1828 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1829 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1830 | namespace_unlock(); |
be34d1a3 | 1831 | if (IS_ERR(tree)) |
52e220d3 | 1832 | return ERR_CAST(tree); |
be34d1a3 | 1833 | return &tree->mnt; |
8aec0809 AV |
1834 | } |
1835 | ||
a07b2000 AV |
1836 | static void free_mnt_ns(struct mnt_namespace *); |
1837 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1838 | ||
1839 | void dissolve_on_fput(struct vfsmount *mnt) | |
1840 | { | |
1841 | struct mnt_namespace *ns; | |
1842 | namespace_lock(); | |
1843 | lock_mount_hash(); | |
1844 | ns = real_mount(mnt)->mnt_ns; | |
1845 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1846 | unlock_mount_hash(); | |
1847 | namespace_unlock(); | |
1848 | free_mnt_ns(ns); | |
1849 | } | |
1850 | ||
8aec0809 AV |
1851 | void drop_collected_mounts(struct vfsmount *mnt) |
1852 | { | |
97216be0 | 1853 | namespace_lock(); |
719ea2fb | 1854 | lock_mount_hash(); |
9c8e0a1b | 1855 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1856 | unlock_mount_hash(); |
3ab6abee | 1857 | namespace_unlock(); |
8aec0809 AV |
1858 | } |
1859 | ||
c771d683 MS |
1860 | /** |
1861 | * clone_private_mount - create a private clone of a path | |
1862 | * | |
1863 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1864 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1865 | * to the originating mount won't be propagated into this). | |
1866 | * | |
1867 | * Release with mntput(). | |
1868 | */ | |
ca71cf71 | 1869 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1870 | { |
1871 | struct mount *old_mnt = real_mount(path->mnt); | |
1872 | struct mount *new_mnt; | |
1873 | ||
1874 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1875 | return ERR_PTR(-EINVAL); | |
1876 | ||
c771d683 | 1877 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1878 | if (IS_ERR(new_mnt)) |
1879 | return ERR_CAST(new_mnt); | |
1880 | ||
1881 | return &new_mnt->mnt; | |
1882 | } | |
1883 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1884 | ||
1f707137 AV |
1885 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1886 | struct vfsmount *root) | |
1887 | { | |
1a4eeaf2 | 1888 | struct mount *mnt; |
1f707137 AV |
1889 | int res = f(root, arg); |
1890 | if (res) | |
1891 | return res; | |
1a4eeaf2 AV |
1892 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1893 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1894 | if (res) |
1895 | return res; | |
1896 | } | |
1897 | return 0; | |
1898 | } | |
1899 | ||
3bd045cc AV |
1900 | static void lock_mnt_tree(struct mount *mnt) |
1901 | { | |
1902 | struct mount *p; | |
1903 | ||
1904 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
1905 | int flags = p->mnt.mnt_flags; | |
1906 | /* Don't allow unprivileged users to change mount flags */ | |
1907 | flags |= MNT_LOCK_ATIME; | |
1908 | ||
1909 | if (flags & MNT_READONLY) | |
1910 | flags |= MNT_LOCK_READONLY; | |
1911 | ||
1912 | if (flags & MNT_NODEV) | |
1913 | flags |= MNT_LOCK_NODEV; | |
1914 | ||
1915 | if (flags & MNT_NOSUID) | |
1916 | flags |= MNT_LOCK_NOSUID; | |
1917 | ||
1918 | if (flags & MNT_NOEXEC) | |
1919 | flags |= MNT_LOCK_NOEXEC; | |
1920 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
1921 | if (list_empty(&p->mnt_expire)) | |
1922 | flags |= MNT_LOCKED; | |
1923 | p->mnt.mnt_flags = flags; | |
1924 | } | |
1925 | } | |
1926 | ||
4b8b21f4 | 1927 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1928 | { |
315fc83e | 1929 | struct mount *p; |
719f5d7f | 1930 | |
909b0a88 | 1931 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1932 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1933 | mnt_release_group_id(p); |
719f5d7f MS |
1934 | } |
1935 | } | |
1936 | ||
4b8b21f4 | 1937 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1938 | { |
315fc83e | 1939 | struct mount *p; |
719f5d7f | 1940 | |
909b0a88 | 1941 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1942 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1943 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1944 | if (err) { |
4b8b21f4 | 1945 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1946 | return err; |
1947 | } | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | return 0; | |
1952 | } | |
1953 | ||
d2921684 EB |
1954 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1955 | { | |
1956 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1957 | unsigned int mounts = 0, old, pending, sum; | |
1958 | struct mount *p; | |
1959 | ||
1960 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1961 | mounts++; | |
1962 | ||
1963 | old = ns->mounts; | |
1964 | pending = ns->pending_mounts; | |
1965 | sum = old + pending; | |
1966 | if ((old > sum) || | |
1967 | (pending > sum) || | |
1968 | (max < sum) || | |
1969 | (mounts > (max - sum))) | |
1970 | return -ENOSPC; | |
1971 | ||
1972 | ns->pending_mounts = pending + mounts; | |
1973 | return 0; | |
1974 | } | |
1975 | ||
b90fa9ae RP |
1976 | /* |
1977 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1978 | * @nd : place the mount tree @source_mnt is attached |
1979 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1980 | * store the parent mount and mountpoint dentry. | |
1981 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1982 | * |
1983 | * NOTE: in the table below explains the semantics when a source mount | |
1984 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1985 | * --------------------------------------------------------------------------- |
1986 | * | BIND MOUNT OPERATION | | |
1987 | * |************************************************************************** | |
1988 | * | source-->| shared | private | slave | unbindable | | |
1989 | * | dest | | | | | | |
1990 | * | | | | | | | | |
1991 | * | v | | | | | | |
1992 | * |************************************************************************** | |
1993 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1994 | * | | | | | | | |
1995 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1996 | * *************************************************************************** | |
b90fa9ae RP |
1997 | * A bind operation clones the source mount and mounts the clone on the |
1998 | * destination mount. | |
1999 | * | |
2000 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2001 | * tree of the destination mount and the cloned mount is added to | |
2002 | * the peer group of the source mount. | |
2003 | * (+) the cloned mount is created under the destination mount and is marked | |
2004 | * as shared. The cloned mount is added to the peer group of the source | |
2005 | * mount. | |
5afe0022 RP |
2006 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2007 | * of the destination mount and the cloned mount is made slave | |
2008 | * of the same master as that of the source mount. The cloned mount | |
2009 | * is marked as 'shared and slave'. | |
2010 | * (*) the cloned mount is made a slave of the same master as that of the | |
2011 | * source mount. | |
2012 | * | |
9676f0c6 RP |
2013 | * --------------------------------------------------------------------------- |
2014 | * | MOVE MOUNT OPERATION | | |
2015 | * |************************************************************************** | |
2016 | * | source-->| shared | private | slave | unbindable | | |
2017 | * | dest | | | | | | |
2018 | * | | | | | | | | |
2019 | * | v | | | | | | |
2020 | * |************************************************************************** | |
2021 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2022 | * | | | | | | | |
2023 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2024 | * *************************************************************************** | |
5afe0022 RP |
2025 | * |
2026 | * (+) the mount is moved to the destination. And is then propagated to | |
2027 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2028 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2029 | * (+++) the mount is moved to the destination and is then propagated to |
2030 | * all the mounts belonging to the destination mount's propagation tree. | |
2031 | * the mount is marked as 'shared and slave'. | |
2032 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2033 | * |
2034 | * if the source mount is a tree, the operations explained above is | |
2035 | * applied to each mount in the tree. | |
2036 | * Must be called without spinlocks held, since this function can sleep | |
2037 | * in allocations. | |
2038 | */ | |
0fb54e50 | 2039 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2040 | struct mount *dest_mnt, |
2041 | struct mountpoint *dest_mp, | |
2042 | struct path *parent_path) | |
b90fa9ae | 2043 | { |
3bd045cc | 2044 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2045 | HLIST_HEAD(tree_list); |
d2921684 | 2046 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2047 | struct mountpoint *smp; |
315fc83e | 2048 | struct mount *child, *p; |
38129a13 | 2049 | struct hlist_node *n; |
719f5d7f | 2050 | int err; |
b90fa9ae | 2051 | |
1064f874 EB |
2052 | /* Preallocate a mountpoint in case the new mounts need |
2053 | * to be tucked under other mounts. | |
2054 | */ | |
2055 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2056 | if (IS_ERR(smp)) | |
2057 | return PTR_ERR(smp); | |
2058 | ||
d2921684 EB |
2059 | /* Is there space to add these mounts to the mount namespace? */ |
2060 | if (!parent_path) { | |
2061 | err = count_mounts(ns, source_mnt); | |
2062 | if (err) | |
2063 | goto out; | |
2064 | } | |
2065 | ||
fc7be130 | 2066 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2067 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2068 | if (err) |
2069 | goto out; | |
0b1b901b | 2070 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2071 | lock_mount_hash(); |
0b1b901b AV |
2072 | if (err) |
2073 | goto out_cleanup_ids; | |
909b0a88 | 2074 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2075 | set_mnt_shared(p); |
0b1b901b AV |
2076 | } else { |
2077 | lock_mount_hash(); | |
b90fa9ae | 2078 | } |
1a390689 | 2079 | if (parent_path) { |
0fb54e50 | 2080 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2081 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2082 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2083 | } else { |
84d17192 | 2084 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2085 | commit_tree(source_mnt); |
21444403 | 2086 | } |
b90fa9ae | 2087 | |
38129a13 | 2088 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2089 | struct mount *q; |
38129a13 | 2090 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2091 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2092 | child->mnt_mountpoint); | |
2093 | if (q) | |
2094 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2095 | /* Notice when we are propagating across user namespaces */ |
2096 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2097 | lock_mnt_tree(child); | |
1064f874 | 2098 | commit_tree(child); |
b90fa9ae | 2099 | } |
1064f874 | 2100 | put_mountpoint(smp); |
719ea2fb | 2101 | unlock_mount_hash(); |
99b7db7b | 2102 | |
b90fa9ae | 2103 | return 0; |
719f5d7f MS |
2104 | |
2105 | out_cleanup_ids: | |
f2ebb3a9 AV |
2106 | while (!hlist_empty(&tree_list)) { |
2107 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2108 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2109 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2110 | } |
2111 | unlock_mount_hash(); | |
0b1b901b | 2112 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2113 | out: |
d2921684 | 2114 | ns->pending_mounts = 0; |
1064f874 EB |
2115 | |
2116 | read_seqlock_excl(&mount_lock); | |
2117 | put_mountpoint(smp); | |
2118 | read_sequnlock_excl(&mount_lock); | |
2119 | ||
719f5d7f | 2120 | return err; |
b90fa9ae RP |
2121 | } |
2122 | ||
84d17192 | 2123 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2124 | { |
2125 | struct vfsmount *mnt; | |
84d17192 | 2126 | struct dentry *dentry = path->dentry; |
b12cea91 | 2127 | retry: |
5955102c | 2128 | inode_lock(dentry->d_inode); |
84d17192 | 2129 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2130 | inode_unlock(dentry->d_inode); |
84d17192 | 2131 | return ERR_PTR(-ENOENT); |
b12cea91 | 2132 | } |
97216be0 | 2133 | namespace_lock(); |
b12cea91 | 2134 | mnt = lookup_mnt(path); |
84d17192 | 2135 | if (likely(!mnt)) { |
3895dbf8 | 2136 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2137 | if (IS_ERR(mp)) { |
97216be0 | 2138 | namespace_unlock(); |
5955102c | 2139 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2140 | return mp; |
2141 | } | |
2142 | return mp; | |
2143 | } | |
97216be0 | 2144 | namespace_unlock(); |
5955102c | 2145 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2146 | path_put(path); |
2147 | path->mnt = mnt; | |
84d17192 | 2148 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2149 | goto retry; |
2150 | } | |
2151 | ||
84d17192 | 2152 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2153 | { |
84d17192 | 2154 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2155 | |
2156 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2157 | put_mountpoint(where); |
3895dbf8 EB |
2158 | read_sequnlock_excl(&mount_lock); |
2159 | ||
328e6d90 | 2160 | namespace_unlock(); |
5955102c | 2161 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2162 | } |
2163 | ||
84d17192 | 2164 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2165 | { |
e462ec50 | 2166 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2167 | return -EINVAL; |
2168 | ||
e36cb0b8 DH |
2169 | if (d_is_dir(mp->m_dentry) != |
2170 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2171 | return -ENOTDIR; |
2172 | ||
84d17192 | 2173 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2174 | } |
2175 | ||
7a2e8a8f VA |
2176 | /* |
2177 | * Sanity check the flags to change_mnt_propagation. | |
2178 | */ | |
2179 | ||
e462ec50 | 2180 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2181 | { |
e462ec50 | 2182 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2183 | |
2184 | /* Fail if any non-propagation flags are set */ | |
2185 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2186 | return 0; | |
2187 | /* Only one propagation flag should be set */ | |
2188 | if (!is_power_of_2(type)) | |
2189 | return 0; | |
2190 | return type; | |
2191 | } | |
2192 | ||
07b20889 RP |
2193 | /* |
2194 | * recursively change the type of the mountpoint. | |
2195 | */ | |
e462ec50 | 2196 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2197 | { |
315fc83e | 2198 | struct mount *m; |
4b8b21f4 | 2199 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2200 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2201 | int type; |
719f5d7f | 2202 | int err = 0; |
07b20889 | 2203 | |
2d92ab3c | 2204 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2205 | return -EINVAL; |
2206 | ||
e462ec50 | 2207 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2208 | if (!type) |
2209 | return -EINVAL; | |
2210 | ||
97216be0 | 2211 | namespace_lock(); |
719f5d7f MS |
2212 | if (type == MS_SHARED) { |
2213 | err = invent_group_ids(mnt, recurse); | |
2214 | if (err) | |
2215 | goto out_unlock; | |
2216 | } | |
2217 | ||
719ea2fb | 2218 | lock_mount_hash(); |
909b0a88 | 2219 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2220 | change_mnt_propagation(m, type); |
719ea2fb | 2221 | unlock_mount_hash(); |
719f5d7f MS |
2222 | |
2223 | out_unlock: | |
97216be0 | 2224 | namespace_unlock(); |
719f5d7f | 2225 | return err; |
07b20889 RP |
2226 | } |
2227 | ||
5ff9d8a6 EB |
2228 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2229 | { | |
2230 | struct mount *child; | |
2231 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2232 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2233 | continue; | |
2234 | ||
2235 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2236 | return true; | |
2237 | } | |
2238 | return false; | |
2239 | } | |
2240 | ||
a07b2000 AV |
2241 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2242 | { | |
2243 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2244 | ||
2245 | if (IS_MNT_UNBINDABLE(old)) | |
2246 | return mnt; | |
2247 | ||
2248 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2249 | return mnt; | |
2250 | ||
2251 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2252 | return mnt; | |
2253 | ||
2254 | if (recurse) | |
2255 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2256 | else | |
2257 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2258 | ||
2259 | if (!IS_ERR(mnt)) | |
2260 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2261 | ||
2262 | return mnt; | |
2263 | } | |
2264 | ||
1da177e4 LT |
2265 | /* |
2266 | * do loopback mount. | |
2267 | */ | |
808d4e3c | 2268 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2269 | int recurse) |
1da177e4 | 2270 | { |
2d92ab3c | 2271 | struct path old_path; |
a07b2000 | 2272 | struct mount *mnt = NULL, *parent; |
84d17192 | 2273 | struct mountpoint *mp; |
57eccb83 | 2274 | int err; |
1da177e4 LT |
2275 | if (!old_name || !*old_name) |
2276 | return -EINVAL; | |
815d405c | 2277 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2278 | if (err) |
2279 | return err; | |
2280 | ||
8823c079 | 2281 | err = -EINVAL; |
4ce5d2b1 | 2282 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2283 | goto out; |
8823c079 | 2284 | |
84d17192 | 2285 | mp = lock_mount(path); |
a07b2000 AV |
2286 | if (IS_ERR(mp)) { |
2287 | err = PTR_ERR(mp); | |
b12cea91 | 2288 | goto out; |
a07b2000 | 2289 | } |
b12cea91 | 2290 | |
84d17192 | 2291 | parent = real_mount(path->mnt); |
e149ed2b AV |
2292 | if (!check_mnt(parent)) |
2293 | goto out2; | |
2294 | ||
a07b2000 | 2295 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2296 | if (IS_ERR(mnt)) { |
2297 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2298 | goto out2; |
be34d1a3 | 2299 | } |
ccd48bc7 | 2300 | |
84d17192 | 2301 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2302 | if (err) { |
719ea2fb | 2303 | lock_mount_hash(); |
e819f152 | 2304 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2305 | unlock_mount_hash(); |
5b83d2c5 | 2306 | } |
b12cea91 | 2307 | out2: |
84d17192 | 2308 | unlock_mount(mp); |
ccd48bc7 | 2309 | out: |
2d92ab3c | 2310 | path_put(&old_path); |
1da177e4 LT |
2311 | return err; |
2312 | } | |
2313 | ||
a07b2000 AV |
2314 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2315 | { | |
2316 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2317 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2318 | struct mount *mnt, *p; | |
2319 | struct file *file; | |
2320 | ||
2321 | if (IS_ERR(ns)) | |
2322 | return ERR_CAST(ns); | |
2323 | ||
2324 | namespace_lock(); | |
2325 | mnt = __do_loopback(path, recursive); | |
2326 | if (IS_ERR(mnt)) { | |
2327 | namespace_unlock(); | |
2328 | free_mnt_ns(ns); | |
2329 | return ERR_CAST(mnt); | |
2330 | } | |
2331 | ||
2332 | lock_mount_hash(); | |
2333 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2334 | p->mnt_ns = ns; | |
2335 | ns->mounts++; | |
2336 | } | |
2337 | ns->root = mnt; | |
2338 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2339 | mntget(&mnt->mnt); | |
2340 | unlock_mount_hash(); | |
2341 | namespace_unlock(); | |
2342 | ||
2343 | mntput(path->mnt); | |
2344 | path->mnt = &mnt->mnt; | |
2345 | file = dentry_open(path, O_PATH, current_cred()); | |
2346 | if (IS_ERR(file)) | |
2347 | dissolve_on_fput(path->mnt); | |
2348 | else | |
2349 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2350 | return file; | |
2351 | } | |
2352 | ||
2353 | SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags) | |
2354 | { | |
2355 | struct file *file; | |
2356 | struct path path; | |
2357 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2358 | bool detached = flags & OPEN_TREE_CLONE; | |
2359 | int error; | |
2360 | int fd; | |
2361 | ||
2362 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2363 | ||
2364 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2365 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2366 | OPEN_TREE_CLOEXEC)) | |
2367 | return -EINVAL; | |
2368 | ||
2369 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2370 | return -EINVAL; | |
2371 | ||
2372 | if (flags & AT_NO_AUTOMOUNT) | |
2373 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2374 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2375 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2376 | if (flags & AT_EMPTY_PATH) | |
2377 | lookup_flags |= LOOKUP_EMPTY; | |
2378 | ||
2379 | if (detached && !may_mount()) | |
2380 | return -EPERM; | |
2381 | ||
2382 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2383 | if (fd < 0) | |
2384 | return fd; | |
2385 | ||
2386 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2387 | if (unlikely(error)) { | |
2388 | file = ERR_PTR(error); | |
2389 | } else { | |
2390 | if (detached) | |
2391 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2392 | else | |
2393 | file = dentry_open(&path, O_PATH, current_cred()); | |
2394 | path_put(&path); | |
2395 | } | |
2396 | if (IS_ERR(file)) { | |
2397 | put_unused_fd(fd); | |
2398 | return PTR_ERR(file); | |
2399 | } | |
2400 | fd_install(fd, file); | |
2401 | return fd; | |
2402 | } | |
2403 | ||
43f5e655 DH |
2404 | /* |
2405 | * Don't allow locked mount flags to be cleared. | |
2406 | * | |
2407 | * No locks need to be held here while testing the various MNT_LOCK | |
2408 | * flags because those flags can never be cleared once they are set. | |
2409 | */ | |
2410 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2411 | { |
43f5e655 DH |
2412 | unsigned int fl = mnt->mnt.mnt_flags; |
2413 | ||
2414 | if ((fl & MNT_LOCK_READONLY) && | |
2415 | !(mnt_flags & MNT_READONLY)) | |
2416 | return false; | |
2417 | ||
2418 | if ((fl & MNT_LOCK_NODEV) && | |
2419 | !(mnt_flags & MNT_NODEV)) | |
2420 | return false; | |
2421 | ||
2422 | if ((fl & MNT_LOCK_NOSUID) && | |
2423 | !(mnt_flags & MNT_NOSUID)) | |
2424 | return false; | |
2425 | ||
2426 | if ((fl & MNT_LOCK_NOEXEC) && | |
2427 | !(mnt_flags & MNT_NOEXEC)) | |
2428 | return false; | |
2429 | ||
2430 | if ((fl & MNT_LOCK_ATIME) && | |
2431 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2432 | return false; | |
2e4b7fcd | 2433 | |
43f5e655 DH |
2434 | return true; |
2435 | } | |
2436 | ||
2437 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2438 | { |
43f5e655 | 2439 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2440 | |
43f5e655 | 2441 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2442 | return 0; |
2443 | ||
2444 | if (readonly_request) | |
43f5e655 DH |
2445 | return mnt_make_readonly(mnt); |
2446 | ||
2447 | return __mnt_unmake_readonly(mnt); | |
2448 | } | |
2449 | ||
2450 | /* | |
2451 | * Update the user-settable attributes on a mount. The caller must hold | |
2452 | * sb->s_umount for writing. | |
2453 | */ | |
2454 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) | |
2455 | { | |
2456 | lock_mount_hash(); | |
2457 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; | |
2458 | mnt->mnt.mnt_flags = mnt_flags; | |
2459 | touch_mnt_namespace(mnt->mnt_ns); | |
2460 | unlock_mount_hash(); | |
2461 | } | |
2462 | ||
2463 | /* | |
2464 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2465 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2466 | * to mount(2). | |
2467 | */ | |
2468 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2469 | { | |
2470 | struct super_block *sb = path->mnt->mnt_sb; | |
2471 | struct mount *mnt = real_mount(path->mnt); | |
2472 | int ret; | |
2473 | ||
2474 | if (!check_mnt(mnt)) | |
2475 | return -EINVAL; | |
2476 | ||
2477 | if (path->dentry != mnt->mnt.mnt_root) | |
2478 | return -EINVAL; | |
2479 | ||
2480 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2481 | return -EPERM; | |
2482 | ||
2483 | down_write(&sb->s_umount); | |
2484 | ret = change_mount_ro_state(mnt, mnt_flags); | |
2485 | if (ret == 0) | |
2486 | set_mount_attributes(mnt, mnt_flags); | |
2487 | up_write(&sb->s_umount); | |
2488 | return ret; | |
2e4b7fcd DH |
2489 | } |
2490 | ||
1da177e4 LT |
2491 | /* |
2492 | * change filesystem flags. dir should be a physical root of filesystem. | |
2493 | * If you've mounted a non-root directory somewhere and want to do remount | |
2494 | * on it - tough luck. | |
2495 | */ | |
e462ec50 DH |
2496 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2497 | int mnt_flags, void *data) | |
1da177e4 LT |
2498 | { |
2499 | int err; | |
2d92ab3c | 2500 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2501 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2502 | struct fs_context *fc; |
1da177e4 | 2503 | |
143c8c91 | 2504 | if (!check_mnt(mnt)) |
1da177e4 LT |
2505 | return -EINVAL; |
2506 | ||
2d92ab3c | 2507 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2508 | return -EINVAL; |
2509 | ||
43f5e655 | 2510 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2511 | return -EPERM; |
9566d674 | 2512 | |
8d0347f6 DH |
2513 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2514 | if (IS_ERR(fc)) | |
2515 | return PTR_ERR(fc); | |
ff36fe2c | 2516 | |
8d0347f6 DH |
2517 | err = parse_monolithic_mount_data(fc, data); |
2518 | if (!err) { | |
2519 | down_write(&sb->s_umount); | |
2520 | err = -EPERM; | |
2521 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2522 | err = reconfigure_super(fc); | |
2523 | if (!err) | |
2524 | set_mount_attributes(mnt, mnt_flags); | |
2525 | } | |
2526 | up_write(&sb->s_umount); | |
0e55a7cc | 2527 | } |
8d0347f6 | 2528 | put_fs_context(fc); |
1da177e4 LT |
2529 | return err; |
2530 | } | |
2531 | ||
cbbe362c | 2532 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2533 | { |
315fc83e | 2534 | struct mount *p; |
909b0a88 | 2535 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2536 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2537 | return 1; |
2538 | } | |
2539 | return 0; | |
2540 | } | |
2541 | ||
2db154b3 | 2542 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2543 | { |
2db154b3 | 2544 | struct path parent_path = {.mnt = NULL, .dentry = NULL}; |
676da58d | 2545 | struct mount *p; |
0fb54e50 | 2546 | struct mount *old; |
84d17192 | 2547 | struct mountpoint *mp; |
57eccb83 | 2548 | int err; |
1da177e4 | 2549 | |
2db154b3 | 2550 | mp = lock_mount(new_path); |
84d17192 | 2551 | if (IS_ERR(mp)) |
2db154b3 | 2552 | return PTR_ERR(mp); |
cc53ce53 | 2553 | |
2db154b3 DH |
2554 | old = real_mount(old_path->mnt); |
2555 | p = real_mount(new_path->mnt); | |
143c8c91 | 2556 | |
1da177e4 | 2557 | err = -EINVAL; |
fc7be130 | 2558 | if (!check_mnt(p) || !check_mnt(old)) |
2db154b3 | 2559 | goto out; |
1da177e4 | 2560 | |
2db154b3 DH |
2561 | if (!mnt_has_parent(old)) |
2562 | goto out; | |
5ff9d8a6 | 2563 | |
2db154b3 DH |
2564 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2565 | goto out; | |
1da177e4 | 2566 | |
2db154b3 DH |
2567 | if (old_path->dentry != old_path->mnt->mnt_root) |
2568 | goto out; | |
1da177e4 | 2569 | |
2db154b3 DH |
2570 | if (d_is_dir(new_path->dentry) != |
2571 | d_is_dir(old_path->dentry)) | |
2572 | goto out; | |
21444403 RP |
2573 | /* |
2574 | * Don't move a mount residing in a shared parent. | |
2575 | */ | |
fc7be130 | 2576 | if (IS_MNT_SHARED(old->mnt_parent)) |
2db154b3 | 2577 | goto out; |
9676f0c6 RP |
2578 | /* |
2579 | * Don't move a mount tree containing unbindable mounts to a destination | |
2580 | * mount which is shared. | |
2581 | */ | |
fc7be130 | 2582 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2583 | goto out; |
1da177e4 | 2584 | err = -ELOOP; |
fc7be130 | 2585 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2586 | if (p == old) |
2db154b3 | 2587 | goto out; |
1da177e4 | 2588 | |
2db154b3 DH |
2589 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2590 | &parent_path); | |
4ac91378 | 2591 | if (err) |
2db154b3 | 2592 | goto out; |
1da177e4 LT |
2593 | |
2594 | /* if the mount is moved, it should no longer be expire | |
2595 | * automatically */ | |
6776db3d | 2596 | list_del_init(&old->mnt_expire); |
1da177e4 | 2597 | out: |
2db154b3 | 2598 | unlock_mount(mp); |
1da177e4 | 2599 | if (!err) |
1a390689 | 2600 | path_put(&parent_path); |
2db154b3 DH |
2601 | return err; |
2602 | } | |
2603 | ||
2604 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2605 | { | |
2606 | struct path old_path; | |
2607 | int err; | |
2608 | ||
2609 | if (!old_name || !*old_name) | |
2610 | return -EINVAL; | |
2611 | ||
2612 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2613 | if (err) | |
2614 | return err; | |
2615 | ||
2616 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2617 | path_put(&old_path); |
1da177e4 LT |
2618 | return err; |
2619 | } | |
2620 | ||
9d412a43 AV |
2621 | /* |
2622 | * add a mount into a namespace's mount tree | |
2623 | */ | |
95bc5f25 | 2624 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2625 | { |
84d17192 AV |
2626 | struct mountpoint *mp; |
2627 | struct mount *parent; | |
9d412a43 AV |
2628 | int err; |
2629 | ||
f2ebb3a9 | 2630 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2631 | |
84d17192 AV |
2632 | mp = lock_mount(path); |
2633 | if (IS_ERR(mp)) | |
2634 | return PTR_ERR(mp); | |
9d412a43 | 2635 | |
84d17192 | 2636 | parent = real_mount(path->mnt); |
9d412a43 | 2637 | err = -EINVAL; |
84d17192 | 2638 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2639 | /* that's acceptable only for automounts done in private ns */ |
2640 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2641 | goto unlock; | |
2642 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2643 | if (!parent->mnt_ns) |
156cacb1 AV |
2644 | goto unlock; |
2645 | } | |
9d412a43 AV |
2646 | |
2647 | /* Refuse the same filesystem on the same mount point */ | |
2648 | err = -EBUSY; | |
95bc5f25 | 2649 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2650 | path->mnt->mnt_root == path->dentry) |
2651 | goto unlock; | |
2652 | ||
2653 | err = -EINVAL; | |
e36cb0b8 | 2654 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2655 | goto unlock; |
2656 | ||
95bc5f25 | 2657 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2658 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2659 | |
2660 | unlock: | |
84d17192 | 2661 | unlock_mount(mp); |
9d412a43 AV |
2662 | return err; |
2663 | } | |
b1e75df4 | 2664 | |
132e4608 DH |
2665 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2666 | ||
2667 | /* | |
2668 | * Create a new mount using a superblock configuration and request it | |
2669 | * be added to the namespace tree. | |
2670 | */ | |
2671 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2672 | unsigned int mnt_flags) | |
2673 | { | |
2674 | struct vfsmount *mnt; | |
2675 | struct super_block *sb = fc->root->d_sb; | |
2676 | int error; | |
2677 | ||
c9ce29ed AV |
2678 | error = security_sb_kern_mount(sb); |
2679 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2680 | error = -EPERM; | |
2681 | ||
2682 | if (unlikely(error)) { | |
2683 | fc_drop_locked(fc); | |
2684 | return error; | |
132e4608 DH |
2685 | } |
2686 | ||
2687 | up_write(&sb->s_umount); | |
2688 | ||
2689 | mnt = vfs_create_mount(fc); | |
2690 | if (IS_ERR(mnt)) | |
2691 | return PTR_ERR(mnt); | |
2692 | ||
2693 | error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags); | |
2694 | if (error < 0) | |
2695 | mntput(mnt); | |
2696 | return error; | |
2697 | } | |
1b852bce | 2698 | |
1da177e4 LT |
2699 | /* |
2700 | * create a new mount for userspace and request it to be added into the | |
2701 | * namespace's tree | |
2702 | */ | |
e462ec50 | 2703 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2704 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2705 | { |
0c55cfc4 | 2706 | struct file_system_type *type; |
a0c9a8b8 AV |
2707 | struct fs_context *fc; |
2708 | const char *subtype = NULL; | |
2709 | int err = 0; | |
1da177e4 | 2710 | |
0c55cfc4 | 2711 | if (!fstype) |
1da177e4 LT |
2712 | return -EINVAL; |
2713 | ||
0c55cfc4 EB |
2714 | type = get_fs_type(fstype); |
2715 | if (!type) | |
2716 | return -ENODEV; | |
2717 | ||
a0c9a8b8 AV |
2718 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
2719 | subtype = strchr(fstype, '.'); | |
2720 | if (subtype) { | |
2721 | subtype++; | |
2722 | if (!*subtype) { | |
2723 | put_filesystem(type); | |
2724 | return -EINVAL; | |
2725 | } | |
2726 | } else { | |
2727 | subtype = ""; | |
2728 | } | |
2729 | } | |
0c55cfc4 | 2730 | |
a0c9a8b8 | 2731 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 2732 | put_filesystem(type); |
a0c9a8b8 AV |
2733 | if (IS_ERR(fc)) |
2734 | return PTR_ERR(fc); | |
2735 | ||
3e1aeb00 DH |
2736 | if (subtype) |
2737 | err = vfs_parse_fs_string(fc, "subtype", | |
2738 | subtype, strlen(subtype)); | |
2739 | if (!err && name) | |
2740 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
2741 | if (!err) |
2742 | err = parse_monolithic_mount_data(fc, data); | |
2743 | if (!err) | |
2744 | err = vfs_get_tree(fc); | |
132e4608 DH |
2745 | if (!err) |
2746 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 2747 | |
a0c9a8b8 | 2748 | put_fs_context(fc); |
15f9a3f3 | 2749 | return err; |
1da177e4 LT |
2750 | } |
2751 | ||
19a167af AV |
2752 | int finish_automount(struct vfsmount *m, struct path *path) |
2753 | { | |
6776db3d | 2754 | struct mount *mnt = real_mount(m); |
19a167af AV |
2755 | int err; |
2756 | /* The new mount record should have at least 2 refs to prevent it being | |
2757 | * expired before we get a chance to add it | |
2758 | */ | |
6776db3d | 2759 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2760 | |
2761 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2762 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2763 | err = -ELOOP; |
2764 | goto fail; | |
19a167af AV |
2765 | } |
2766 | ||
95bc5f25 | 2767 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2768 | if (!err) |
2769 | return 0; | |
2770 | fail: | |
2771 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2772 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2773 | namespace_lock(); |
6776db3d | 2774 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2775 | namespace_unlock(); |
19a167af | 2776 | } |
b1e75df4 AV |
2777 | mntput(m); |
2778 | mntput(m); | |
19a167af AV |
2779 | return err; |
2780 | } | |
2781 | ||
ea5b778a DH |
2782 | /** |
2783 | * mnt_set_expiry - Put a mount on an expiration list | |
2784 | * @mnt: The mount to list. | |
2785 | * @expiry_list: The list to add the mount to. | |
2786 | */ | |
2787 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2788 | { | |
97216be0 | 2789 | namespace_lock(); |
ea5b778a | 2790 | |
6776db3d | 2791 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2792 | |
97216be0 | 2793 | namespace_unlock(); |
ea5b778a DH |
2794 | } |
2795 | EXPORT_SYMBOL(mnt_set_expiry); | |
2796 | ||
1da177e4 LT |
2797 | /* |
2798 | * process a list of expirable mountpoints with the intent of discarding any | |
2799 | * mountpoints that aren't in use and haven't been touched since last we came | |
2800 | * here | |
2801 | */ | |
2802 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2803 | { | |
761d5c38 | 2804 | struct mount *mnt, *next; |
1da177e4 LT |
2805 | LIST_HEAD(graveyard); |
2806 | ||
2807 | if (list_empty(mounts)) | |
2808 | return; | |
2809 | ||
97216be0 | 2810 | namespace_lock(); |
719ea2fb | 2811 | lock_mount_hash(); |
1da177e4 LT |
2812 | |
2813 | /* extract from the expiration list every vfsmount that matches the | |
2814 | * following criteria: | |
2815 | * - only referenced by its parent vfsmount | |
2816 | * - still marked for expiry (marked on the last call here; marks are | |
2817 | * cleared by mntput()) | |
2818 | */ | |
6776db3d | 2819 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2820 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2821 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2822 | continue; |
6776db3d | 2823 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2824 | } |
bcc5c7d2 | 2825 | while (!list_empty(&graveyard)) { |
6776db3d | 2826 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2827 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2828 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2829 | } |
719ea2fb | 2830 | unlock_mount_hash(); |
3ab6abee | 2831 | namespace_unlock(); |
5528f911 TM |
2832 | } |
2833 | ||
2834 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2835 | ||
2836 | /* | |
2837 | * Ripoff of 'select_parent()' | |
2838 | * | |
2839 | * search the list of submounts for a given mountpoint, and move any | |
2840 | * shrinkable submounts to the 'graveyard' list. | |
2841 | */ | |
692afc31 | 2842 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2843 | { |
692afc31 | 2844 | struct mount *this_parent = parent; |
5528f911 TM |
2845 | struct list_head *next; |
2846 | int found = 0; | |
2847 | ||
2848 | repeat: | |
6b41d536 | 2849 | next = this_parent->mnt_mounts.next; |
5528f911 | 2850 | resume: |
6b41d536 | 2851 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2852 | struct list_head *tmp = next; |
6b41d536 | 2853 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2854 | |
2855 | next = tmp->next; | |
692afc31 | 2856 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2857 | continue; |
5528f911 TM |
2858 | /* |
2859 | * Descend a level if the d_mounts list is non-empty. | |
2860 | */ | |
6b41d536 | 2861 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2862 | this_parent = mnt; |
2863 | goto repeat; | |
2864 | } | |
1da177e4 | 2865 | |
1ab59738 | 2866 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2867 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2868 | found++; |
2869 | } | |
1da177e4 | 2870 | } |
5528f911 TM |
2871 | /* |
2872 | * All done at this level ... ascend and resume the search | |
2873 | */ | |
2874 | if (this_parent != parent) { | |
6b41d536 | 2875 | next = this_parent->mnt_child.next; |
0714a533 | 2876 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2877 | goto resume; |
2878 | } | |
2879 | return found; | |
2880 | } | |
2881 | ||
2882 | /* | |
2883 | * process a list of expirable mountpoints with the intent of discarding any | |
2884 | * submounts of a specific parent mountpoint | |
99b7db7b | 2885 | * |
48a066e7 | 2886 | * mount_lock must be held for write |
5528f911 | 2887 | */ |
b54b9be7 | 2888 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2889 | { |
2890 | LIST_HEAD(graveyard); | |
761d5c38 | 2891 | struct mount *m; |
5528f911 | 2892 | |
5528f911 | 2893 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2894 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2895 | while (!list_empty(&graveyard)) { |
761d5c38 | 2896 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2897 | mnt_expire); |
143c8c91 | 2898 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2899 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2900 | } |
2901 | } | |
1da177e4 LT |
2902 | } |
2903 | ||
1da177e4 LT |
2904 | /* |
2905 | * Some copy_from_user() implementations do not return the exact number of | |
2906 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2907 | * Note that this function differs from copy_from_user() in that it will oops | |
2908 | * on bad values of `to', rather than returning a short copy. | |
2909 | */ | |
b58fed8b RP |
2910 | static long exact_copy_from_user(void *to, const void __user * from, |
2911 | unsigned long n) | |
1da177e4 LT |
2912 | { |
2913 | char *t = to; | |
2914 | const char __user *f = from; | |
2915 | char c; | |
2916 | ||
96d4f267 | 2917 | if (!access_ok(from, n)) |
1da177e4 LT |
2918 | return n; |
2919 | ||
2920 | while (n) { | |
2921 | if (__get_user(c, f)) { | |
2922 | memset(t, 0, n); | |
2923 | break; | |
2924 | } | |
2925 | *t++ = c; | |
2926 | f++; | |
2927 | n--; | |
2928 | } | |
2929 | return n; | |
2930 | } | |
2931 | ||
b40ef869 | 2932 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2933 | { |
2934 | int i; | |
1da177e4 | 2935 | unsigned long size; |
b40ef869 | 2936 | char *copy; |
b58fed8b | 2937 | |
1da177e4 | 2938 | if (!data) |
b40ef869 | 2939 | return NULL; |
1da177e4 | 2940 | |
b40ef869 AV |
2941 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2942 | if (!copy) | |
2943 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2944 | |
2945 | /* We only care that *some* data at the address the user | |
2946 | * gave us is valid. Just in case, we'll zero | |
2947 | * the remainder of the page. | |
2948 | */ | |
2949 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2950 | size = TASK_SIZE - (unsigned long)data; | |
2951 | if (size > PAGE_SIZE) | |
2952 | size = PAGE_SIZE; | |
2953 | ||
b40ef869 | 2954 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2955 | if (!i) { |
b40ef869 AV |
2956 | kfree(copy); |
2957 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2958 | } |
2959 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2960 | memset(copy + i, 0, PAGE_SIZE - i); |
2961 | return copy; | |
1da177e4 LT |
2962 | } |
2963 | ||
b8850d1f | 2964 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2965 | { |
fbdb4401 | 2966 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
2967 | } |
2968 | ||
1da177e4 LT |
2969 | /* |
2970 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2971 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2972 | * | |
2973 | * data is a (void *) that can point to any structure up to | |
2974 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2975 | * information (or be NULL). | |
2976 | * | |
2977 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2978 | * When the flags word was introduced its top half was required | |
2979 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2980 | * Therefore, if this magic number is present, it carries no information | |
2981 | * and must be discarded. | |
2982 | */ | |
5e6123f3 | 2983 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2984 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2985 | { |
2d92ab3c | 2986 | struct path path; |
e462ec50 | 2987 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2988 | int retval = 0; |
1da177e4 LT |
2989 | |
2990 | /* Discard magic */ | |
2991 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2992 | flags &= ~MS_MGC_MSK; | |
2993 | ||
2994 | /* Basic sanity checks */ | |
1da177e4 LT |
2995 | if (data_page) |
2996 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2997 | ||
e462ec50 DH |
2998 | if (flags & MS_NOUSER) |
2999 | return -EINVAL; | |
3000 | ||
a27ab9f2 | 3001 | /* ... and get the mountpoint */ |
5e6123f3 | 3002 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
3003 | if (retval) |
3004 | return retval; | |
3005 | ||
3006 | retval = security_sb_mount(dev_name, &path, | |
3007 | type_page, flags, data_page); | |
0d5cadb8 AV |
3008 | if (!retval && !may_mount()) |
3009 | retval = -EPERM; | |
e462ec50 | 3010 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 3011 | retval = -EPERM; |
a27ab9f2 TH |
3012 | if (retval) |
3013 | goto dput_out; | |
3014 | ||
613cbe3d AK |
3015 | /* Default to relatime unless overriden */ |
3016 | if (!(flags & MS_NOATIME)) | |
3017 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3018 | |
1da177e4 LT |
3019 | /* Separate the per-mountpoint flags */ |
3020 | if (flags & MS_NOSUID) | |
3021 | mnt_flags |= MNT_NOSUID; | |
3022 | if (flags & MS_NODEV) | |
3023 | mnt_flags |= MNT_NODEV; | |
3024 | if (flags & MS_NOEXEC) | |
3025 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3026 | if (flags & MS_NOATIME) |
3027 | mnt_flags |= MNT_NOATIME; | |
3028 | if (flags & MS_NODIRATIME) | |
3029 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3030 | if (flags & MS_STRICTATIME) |
3031 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3032 | if (flags & MS_RDONLY) |
2e4b7fcd | 3033 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 3034 | |
ffbc6f0e EB |
3035 | /* The default atime for remount is preservation */ |
3036 | if ((flags & MS_REMOUNT) && | |
3037 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3038 | MS_STRICTATIME)) == 0)) { | |
3039 | mnt_flags &= ~MNT_ATIME_MASK; | |
3040 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
3041 | } | |
3042 | ||
e462ec50 DH |
3043 | sb_flags = flags & (SB_RDONLY | |
3044 | SB_SYNCHRONOUS | | |
3045 | SB_MANDLOCK | | |
3046 | SB_DIRSYNC | | |
3047 | SB_SILENT | | |
917086ff | 3048 | SB_POSIXACL | |
d7ee9469 | 3049 | SB_LAZYTIME | |
917086ff | 3050 | SB_I_VERSION); |
1da177e4 | 3051 | |
43f5e655 DH |
3052 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
3053 | retval = do_reconfigure_mnt(&path, mnt_flags); | |
3054 | else if (flags & MS_REMOUNT) | |
e462ec50 | 3055 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
3056 | data_page); |
3057 | else if (flags & MS_BIND) | |
2d92ab3c | 3058 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 3059 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 3060 | retval = do_change_type(&path, flags); |
1da177e4 | 3061 | else if (flags & MS_MOVE) |
2db154b3 | 3062 | retval = do_move_mount_old(&path, dev_name); |
1da177e4 | 3063 | else |
e462ec50 | 3064 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
3065 | dev_name, data_page); |
3066 | dput_out: | |
2d92ab3c | 3067 | path_put(&path); |
1da177e4 LT |
3068 | return retval; |
3069 | } | |
3070 | ||
537f7ccb EB |
3071 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3072 | { | |
3073 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3074 | } | |
3075 | ||
3076 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3077 | { | |
3078 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3079 | } | |
3080 | ||
771b1371 EB |
3081 | static void free_mnt_ns(struct mnt_namespace *ns) |
3082 | { | |
74e83122 AV |
3083 | if (!is_anon_ns(ns)) |
3084 | ns_free_inum(&ns->ns); | |
537f7ccb | 3085 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3086 | put_user_ns(ns->user_ns); |
3087 | kfree(ns); | |
3088 | } | |
3089 | ||
8823c079 EB |
3090 | /* |
3091 | * Assign a sequence number so we can detect when we attempt to bind | |
3092 | * mount a reference to an older mount namespace into the current | |
3093 | * mount namespace, preventing reference counting loops. A 64bit | |
3094 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3095 | * is effectively never, so we can ignore the possibility. | |
3096 | */ | |
3097 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3098 | ||
74e83122 | 3099 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3100 | { |
3101 | struct mnt_namespace *new_ns; | |
537f7ccb | 3102 | struct ucounts *ucounts; |
98f842e6 | 3103 | int ret; |
cf8d2c11 | 3104 | |
537f7ccb EB |
3105 | ucounts = inc_mnt_namespaces(user_ns); |
3106 | if (!ucounts) | |
df75e774 | 3107 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3108 | |
74e83122 | 3109 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
3110 | if (!new_ns) { |
3111 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3112 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3113 | } |
74e83122 AV |
3114 | if (!anon) { |
3115 | ret = ns_alloc_inum(&new_ns->ns); | |
3116 | if (ret) { | |
3117 | kfree(new_ns); | |
3118 | dec_mnt_namespaces(ucounts); | |
3119 | return ERR_PTR(ret); | |
3120 | } | |
98f842e6 | 3121 | } |
33c42940 | 3122 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3123 | if (!anon) |
3124 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
cf8d2c11 | 3125 | atomic_set(&new_ns->count, 1); |
cf8d2c11 TM |
3126 | INIT_LIST_HEAD(&new_ns->list); |
3127 | init_waitqueue_head(&new_ns->poll); | |
771b1371 | 3128 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3129 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3130 | return new_ns; |
3131 | } | |
3132 | ||
0766f788 | 3133 | __latent_entropy |
9559f689 AV |
3134 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3135 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3136 | { |
6b3286ed | 3137 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3138 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3139 | struct mount *p, *q; |
9559f689 | 3140 | struct mount *old; |
cb338d06 | 3141 | struct mount *new; |
7a472ef4 | 3142 | int copy_flags; |
1da177e4 | 3143 | |
9559f689 AV |
3144 | BUG_ON(!ns); |
3145 | ||
3146 | if (likely(!(flags & CLONE_NEWNS))) { | |
3147 | get_mnt_ns(ns); | |
3148 | return ns; | |
3149 | } | |
3150 | ||
3151 | old = ns->root; | |
3152 | ||
74e83122 | 3153 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3154 | if (IS_ERR(new_ns)) |
3155 | return new_ns; | |
1da177e4 | 3156 | |
97216be0 | 3157 | namespace_lock(); |
1da177e4 | 3158 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3159 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3160 | if (user_ns != ns->user_ns) |
3bd045cc | 3161 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3162 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3163 | if (IS_ERR(new)) { |
328e6d90 | 3164 | namespace_unlock(); |
771b1371 | 3165 | free_mnt_ns(new_ns); |
be34d1a3 | 3166 | return ERR_CAST(new); |
1da177e4 | 3167 | } |
3bd045cc AV |
3168 | if (user_ns != ns->user_ns) { |
3169 | lock_mount_hash(); | |
3170 | lock_mnt_tree(new); | |
3171 | unlock_mount_hash(); | |
3172 | } | |
be08d6d2 | 3173 | new_ns->root = new; |
1a4eeaf2 | 3174 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3175 | |
3176 | /* | |
3177 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3178 | * as belonging to new namespace. We have already acquired a private | |
3179 | * fs_struct, so tsk->fs->lock is not needed. | |
3180 | */ | |
909b0a88 | 3181 | p = old; |
cb338d06 | 3182 | q = new; |
1da177e4 | 3183 | while (p) { |
143c8c91 | 3184 | q->mnt_ns = new_ns; |
d2921684 | 3185 | new_ns->mounts++; |
9559f689 AV |
3186 | if (new_fs) { |
3187 | if (&p->mnt == new_fs->root.mnt) { | |
3188 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3189 | rootmnt = &p->mnt; |
1da177e4 | 3190 | } |
9559f689 AV |
3191 | if (&p->mnt == new_fs->pwd.mnt) { |
3192 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3193 | pwdmnt = &p->mnt; |
1da177e4 | 3194 | } |
1da177e4 | 3195 | } |
909b0a88 AV |
3196 | p = next_mnt(p, old); |
3197 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3198 | if (!q) |
3199 | break; | |
3200 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3201 | p = next_mnt(p, old); | |
1da177e4 | 3202 | } |
328e6d90 | 3203 | namespace_unlock(); |
1da177e4 | 3204 | |
1da177e4 | 3205 | if (rootmnt) |
f03c6599 | 3206 | mntput(rootmnt); |
1da177e4 | 3207 | if (pwdmnt) |
f03c6599 | 3208 | mntput(pwdmnt); |
1da177e4 | 3209 | |
741a2951 | 3210 | return new_ns; |
1da177e4 LT |
3211 | } |
3212 | ||
74e83122 | 3213 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3214 | { |
74e83122 | 3215 | struct mount *mnt = real_mount(m); |
ea441d11 | 3216 | struct mnt_namespace *ns; |
d31da0f0 | 3217 | struct super_block *s; |
ea441d11 AV |
3218 | struct path path; |
3219 | int err; | |
3220 | ||
74e83122 AV |
3221 | ns = alloc_mnt_ns(&init_user_ns, true); |
3222 | if (IS_ERR(ns)) { | |
3223 | mntput(m); | |
ea441d11 | 3224 | return ERR_CAST(ns); |
74e83122 AV |
3225 | } |
3226 | mnt->mnt_ns = ns; | |
3227 | ns->root = mnt; | |
3228 | ns->mounts++; | |
3229 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3230 | |
74e83122 | 3231 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3232 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3233 | ||
3234 | put_mnt_ns(ns); | |
3235 | ||
3236 | if (err) | |
3237 | return ERR_PTR(err); | |
3238 | ||
3239 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3240 | s = path.mnt->mnt_sb; |
3241 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3242 | mntput(path.mnt); |
3243 | /* lock the sucker */ | |
d31da0f0 | 3244 | down_write(&s->s_umount); |
ea441d11 AV |
3245 | /* ... and return the root of (sub)tree on it */ |
3246 | return path.dentry; | |
3247 | } | |
3248 | EXPORT_SYMBOL(mount_subtree); | |
3249 | ||
312db1aa DB |
3250 | int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, |
3251 | unsigned long flags, void __user *data) | |
1da177e4 | 3252 | { |
eca6f534 VN |
3253 | int ret; |
3254 | char *kernel_type; | |
eca6f534 | 3255 | char *kernel_dev; |
b40ef869 | 3256 | void *options; |
1da177e4 | 3257 | |
b8850d1f TG |
3258 | kernel_type = copy_mount_string(type); |
3259 | ret = PTR_ERR(kernel_type); | |
3260 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3261 | goto out_type; |
1da177e4 | 3262 | |
b8850d1f TG |
3263 | kernel_dev = copy_mount_string(dev_name); |
3264 | ret = PTR_ERR(kernel_dev); | |
3265 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3266 | goto out_dev; |
1da177e4 | 3267 | |
b40ef869 AV |
3268 | options = copy_mount_options(data); |
3269 | ret = PTR_ERR(options); | |
3270 | if (IS_ERR(options)) | |
eca6f534 | 3271 | goto out_data; |
1da177e4 | 3272 | |
b40ef869 | 3273 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3274 | |
b40ef869 | 3275 | kfree(options); |
eca6f534 VN |
3276 | out_data: |
3277 | kfree(kernel_dev); | |
3278 | out_dev: | |
eca6f534 VN |
3279 | kfree(kernel_type); |
3280 | out_type: | |
3281 | return ret; | |
1da177e4 LT |
3282 | } |
3283 | ||
312db1aa DB |
3284 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3285 | char __user *, type, unsigned long, flags, void __user *, data) | |
3286 | { | |
3287 | return ksys_mount(dev_name, dir_name, type, flags, data); | |
3288 | } | |
3289 | ||
2db154b3 DH |
3290 | /* |
3291 | * Move a mount from one place to another. | |
3292 | * | |
3293 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3294 | */ | |
3295 | SYSCALL_DEFINE5(move_mount, | |
3296 | int, from_dfd, const char *, from_pathname, | |
3297 | int, to_dfd, const char *, to_pathname, | |
3298 | unsigned int, flags) | |
3299 | { | |
3300 | struct path from_path, to_path; | |
3301 | unsigned int lflags; | |
3302 | int ret = 0; | |
3303 | ||
3304 | if (!may_mount()) | |
3305 | return -EPERM; | |
3306 | ||
3307 | if (flags & ~MOVE_MOUNT__MASK) | |
3308 | return -EINVAL; | |
3309 | ||
3310 | /* If someone gives a pathname, they aren't permitted to move | |
3311 | * from an fd that requires unmount as we can't get at the flag | |
3312 | * to clear it afterwards. | |
3313 | */ | |
3314 | lflags = 0; | |
3315 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3316 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3317 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3318 | ||
3319 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3320 | if (ret < 0) | |
3321 | return ret; | |
3322 | ||
3323 | lflags = 0; | |
3324 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3325 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3326 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3327 | ||
3328 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3329 | if (ret < 0) | |
3330 | goto out_from; | |
3331 | ||
3332 | ret = security_move_mount(&from_path, &to_path); | |
3333 | if (ret < 0) | |
3334 | goto out_to; | |
3335 | ||
3336 | ret = do_move_mount(&from_path, &to_path); | |
3337 | ||
3338 | out_to: | |
3339 | path_put(&to_path); | |
3340 | out_from: | |
3341 | path_put(&from_path); | |
3342 | return ret; | |
3343 | } | |
3344 | ||
afac7cba AV |
3345 | /* |
3346 | * Return true if path is reachable from root | |
3347 | * | |
48a066e7 | 3348 | * namespace_sem or mount_lock is held |
afac7cba | 3349 | */ |
643822b4 | 3350 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3351 | const struct path *root) |
3352 | { | |
643822b4 | 3353 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3354 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3355 | mnt = mnt->mnt_parent; |
afac7cba | 3356 | } |
643822b4 | 3357 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3358 | } |
3359 | ||
640eb7e7 | 3360 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3361 | { |
25ab4c9b | 3362 | bool res; |
48a066e7 | 3363 | read_seqlock_excl(&mount_lock); |
643822b4 | 3364 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3365 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3366 | return res; |
3367 | } | |
3368 | EXPORT_SYMBOL(path_is_under); | |
3369 | ||
1da177e4 LT |
3370 | /* |
3371 | * pivot_root Semantics: | |
3372 | * Moves the root file system of the current process to the directory put_old, | |
3373 | * makes new_root as the new root file system of the current process, and sets | |
3374 | * root/cwd of all processes which had them on the current root to new_root. | |
3375 | * | |
3376 | * Restrictions: | |
3377 | * The new_root and put_old must be directories, and must not be on the | |
3378 | * same file system as the current process root. The put_old must be | |
3379 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3380 | * pointed to by put_old must yield the same directory as new_root. No other | |
3381 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3382 | * | |
4a0d11fa NB |
3383 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3384 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3385 | * in this situation. | |
3386 | * | |
1da177e4 LT |
3387 | * Notes: |
3388 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3389 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3390 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3391 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3392 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3393 | * first. | |
3394 | */ | |
3480b257 HC |
3395 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3396 | const char __user *, put_old) | |
1da177e4 | 3397 | { |
2d8f3038 | 3398 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3399 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3400 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3401 | int error; |
3402 | ||
9b40bc90 | 3403 | if (!may_mount()) |
1da177e4 LT |
3404 | return -EPERM; |
3405 | ||
2d8f3038 | 3406 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3407 | if (error) |
3408 | goto out0; | |
1da177e4 | 3409 | |
2d8f3038 | 3410 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3411 | if (error) |
3412 | goto out1; | |
3413 | ||
2d8f3038 | 3414 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3415 | if (error) |
3416 | goto out2; | |
1da177e4 | 3417 | |
f7ad3c6b | 3418 | get_fs_root(current->fs, &root); |
84d17192 AV |
3419 | old_mp = lock_mount(&old); |
3420 | error = PTR_ERR(old_mp); | |
3421 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3422 | goto out3; |
3423 | ||
1da177e4 | 3424 | error = -EINVAL; |
419148da AV |
3425 | new_mnt = real_mount(new.mnt); |
3426 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3427 | old_mnt = real_mount(old.mnt); |
3428 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3429 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3430 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3431 | goto out4; |
143c8c91 | 3432 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3433 | goto out4; |
5ff9d8a6 EB |
3434 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3435 | goto out4; | |
1da177e4 | 3436 | error = -ENOENT; |
f3da392e | 3437 | if (d_unlinked(new.dentry)) |
b12cea91 | 3438 | goto out4; |
1da177e4 | 3439 | error = -EBUSY; |
84d17192 | 3440 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3441 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3442 | error = -EINVAL; |
8c3ee42e | 3443 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3444 | goto out4; /* not a mountpoint */ |
676da58d | 3445 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3446 | goto out4; /* not attached */ |
84d17192 | 3447 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3448 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3449 | goto out4; /* not a mountpoint */ |
676da58d | 3450 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3451 | goto out4; /* not attached */ |
4ac91378 | 3452 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3453 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3454 | goto out4; |
0d082601 EB |
3455 | /* make certain new is below the root */ |
3456 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3457 | goto out4; | |
84d17192 | 3458 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3459 | lock_mount_hash(); |
419148da AV |
3460 | detach_mnt(new_mnt, &parent_path); |
3461 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3462 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3463 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3464 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3465 | } | |
4ac91378 | 3466 | /* mount old root on put_old */ |
84d17192 | 3467 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3468 | /* mount new_root on / */ |
84d17192 | 3469 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3470 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3471 | /* A moved mount should not expire automatically */ |
3472 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3473 | put_mountpoint(root_mp); |
719ea2fb | 3474 | unlock_mount_hash(); |
2d8f3038 | 3475 | chroot_fs_refs(&root, &new); |
1da177e4 | 3476 | error = 0; |
b12cea91 | 3477 | out4: |
84d17192 | 3478 | unlock_mount(old_mp); |
b12cea91 AV |
3479 | if (!error) { |
3480 | path_put(&root_parent); | |
3481 | path_put(&parent_path); | |
3482 | } | |
3483 | out3: | |
8c3ee42e | 3484 | path_put(&root); |
b12cea91 | 3485 | out2: |
2d8f3038 | 3486 | path_put(&old); |
1da177e4 | 3487 | out1: |
2d8f3038 | 3488 | path_put(&new); |
1da177e4 | 3489 | out0: |
1da177e4 | 3490 | return error; |
1da177e4 LT |
3491 | } |
3492 | ||
3493 | static void __init init_mount_tree(void) | |
3494 | { | |
3495 | struct vfsmount *mnt; | |
74e83122 | 3496 | struct mount *m; |
6b3286ed | 3497 | struct mnt_namespace *ns; |
ac748a09 | 3498 | struct path root; |
0c55cfc4 | 3499 | struct file_system_type *type; |
1da177e4 | 3500 | |
0c55cfc4 EB |
3501 | type = get_fs_type("rootfs"); |
3502 | if (!type) | |
3503 | panic("Can't find rootfs type"); | |
3504 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3505 | put_filesystem(type); | |
1da177e4 LT |
3506 | if (IS_ERR(mnt)) |
3507 | panic("Can't create rootfs"); | |
b3e19d92 | 3508 | |
74e83122 | 3509 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 3510 | if (IS_ERR(ns)) |
1da177e4 | 3511 | panic("Can't allocate initial namespace"); |
74e83122 AV |
3512 | m = real_mount(mnt); |
3513 | m->mnt_ns = ns; | |
3514 | ns->root = m; | |
3515 | ns->mounts = 1; | |
3516 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
3517 | init_task.nsproxy->mnt_ns = ns; |
3518 | get_mnt_ns(ns); | |
3519 | ||
be08d6d2 AV |
3520 | root.mnt = mnt; |
3521 | root.dentry = mnt->mnt_root; | |
da362b09 | 3522 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3523 | |
3524 | set_fs_pwd(current->fs, &root); | |
3525 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3526 | } |
3527 | ||
74bf17cf | 3528 | void __init mnt_init(void) |
1da177e4 | 3529 | { |
15a67dd8 | 3530 | int err; |
1da177e4 | 3531 | |
7d6fec45 | 3532 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3533 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3534 | |
0818bf27 | 3535 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3536 | sizeof(struct hlist_head), |
0818bf27 | 3537 | mhash_entries, 19, |
3d375d78 | 3538 | HASH_ZERO, |
0818bf27 AV |
3539 | &m_hash_shift, &m_hash_mask, 0, 0); |
3540 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3541 | sizeof(struct hlist_head), | |
3542 | mphash_entries, 19, | |
3d375d78 | 3543 | HASH_ZERO, |
0818bf27 | 3544 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3545 | |
84d17192 | 3546 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3547 | panic("Failed to allocate mount hash table\n"); |
3548 | ||
4b93dc9b TH |
3549 | kernfs_init(); |
3550 | ||
15a67dd8 RD |
3551 | err = sysfs_init(); |
3552 | if (err) | |
3553 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3554 | __func__, err); |
00d26666 GKH |
3555 | fs_kobj = kobject_create_and_add("fs", NULL); |
3556 | if (!fs_kobj) | |
8e24eea7 | 3557 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3558 | init_rootfs(); |
3559 | init_mount_tree(); | |
3560 | } | |
3561 | ||
616511d0 | 3562 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3563 | { |
d498b25a | 3564 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3565 | return; |
7b00ed6f | 3566 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3567 | free_mnt_ns(ns); |
1da177e4 | 3568 | } |
9d412a43 | 3569 | |
d911b458 | 3570 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 3571 | { |
423e0ab0 | 3572 | struct vfsmount *mnt; |
d911b458 | 3573 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
3574 | if (!IS_ERR(mnt)) { |
3575 | /* | |
3576 | * it is a longterm mount, don't release mnt until | |
3577 | * we unmount before file sys is unregistered | |
3578 | */ | |
f7a99c5b | 3579 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3580 | } |
3581 | return mnt; | |
9d412a43 | 3582 | } |
d911b458 | 3583 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
3584 | |
3585 | void kern_unmount(struct vfsmount *mnt) | |
3586 | { | |
3587 | /* release long term mount so mount point can be released */ | |
3588 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3589 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3590 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3591 | mntput(mnt); |
3592 | } | |
3593 | } | |
3594 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3595 | |
3596 | bool our_mnt(struct vfsmount *mnt) | |
3597 | { | |
143c8c91 | 3598 | return check_mnt(real_mount(mnt)); |
02125a82 | 3599 | } |
8823c079 | 3600 | |
3151527e EB |
3601 | bool current_chrooted(void) |
3602 | { | |
3603 | /* Does the current process have a non-standard root */ | |
3604 | struct path ns_root; | |
3605 | struct path fs_root; | |
3606 | bool chrooted; | |
3607 | ||
3608 | /* Find the namespace root */ | |
3609 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3610 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3611 | path_get(&ns_root); | |
3612 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3613 | ; | |
3614 | ||
3615 | get_fs_root(current->fs, &fs_root); | |
3616 | ||
3617 | chrooted = !path_equal(&fs_root, &ns_root); | |
3618 | ||
3619 | path_put(&fs_root); | |
3620 | path_put(&ns_root); | |
3621 | ||
3622 | return chrooted; | |
3623 | } | |
3624 | ||
132e4608 DH |
3625 | static bool mnt_already_visible(struct mnt_namespace *ns, |
3626 | const struct super_block *sb, | |
8654df4e | 3627 | int *new_mnt_flags) |
87a8ebd6 | 3628 | { |
8c6cf9cc | 3629 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3630 | struct mount *mnt; |
e51db735 | 3631 | bool visible = false; |
87a8ebd6 | 3632 | |
44bb4385 | 3633 | down_read(&namespace_sem); |
87a8ebd6 | 3634 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3635 | struct mount *child; |
77b1a97d EB |
3636 | int mnt_flags; |
3637 | ||
132e4608 | 3638 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
3639 | continue; |
3640 | ||
7e96c1b0 EB |
3641 | /* This mount is not fully visible if it's root directory |
3642 | * is not the root directory of the filesystem. | |
3643 | */ | |
3644 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3645 | continue; | |
3646 | ||
a1935c17 | 3647 | /* A local view of the mount flags */ |
77b1a97d | 3648 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3649 | |
695e9df0 | 3650 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3651 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3652 | mnt_flags |= MNT_LOCK_READONLY; |
3653 | ||
8c6cf9cc EB |
3654 | /* Verify the mount flags are equal to or more permissive |
3655 | * than the proposed new mount. | |
3656 | */ | |
77b1a97d | 3657 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3658 | !(new_flags & MNT_READONLY)) |
3659 | continue; | |
77b1a97d EB |
3660 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3661 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3662 | continue; |
3663 | ||
ceeb0e5d EB |
3664 | /* This mount is not fully visible if there are any |
3665 | * locked child mounts that cover anything except for | |
3666 | * empty directories. | |
e51db735 EB |
3667 | */ |
3668 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3669 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3670 | /* Only worry about locked mounts */ |
d71ed6c9 | 3671 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3672 | continue; |
7236c85e EB |
3673 | /* Is the directory permanetly empty? */ |
3674 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3675 | goto next; |
87a8ebd6 | 3676 | } |
8c6cf9cc | 3677 | /* Preserve the locked attributes */ |
77b1a97d | 3678 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3679 | MNT_LOCK_ATIME); |
e51db735 EB |
3680 | visible = true; |
3681 | goto found; | |
3682 | next: ; | |
87a8ebd6 | 3683 | } |
e51db735 | 3684 | found: |
44bb4385 | 3685 | up_read(&namespace_sem); |
e51db735 | 3686 | return visible; |
87a8ebd6 EB |
3687 | } |
3688 | ||
132e4608 | 3689 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 3690 | { |
a1935c17 | 3691 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3692 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3693 | unsigned long s_iflags; | |
3694 | ||
3695 | if (ns->user_ns == &init_user_ns) | |
3696 | return false; | |
3697 | ||
3698 | /* Can this filesystem be too revealing? */ | |
132e4608 | 3699 | s_iflags = sb->s_iflags; |
8654df4e EB |
3700 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
3701 | return false; | |
3702 | ||
a1935c17 EB |
3703 | if ((s_iflags & required_iflags) != required_iflags) { |
3704 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3705 | required_iflags); | |
3706 | return true; | |
3707 | } | |
3708 | ||
132e4608 | 3709 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
3710 | } |
3711 | ||
380cf5ba AL |
3712 | bool mnt_may_suid(struct vfsmount *mnt) |
3713 | { | |
3714 | /* | |
3715 | * Foreign mounts (accessed via fchdir or through /proc | |
3716 | * symlinks) are always treated as if they are nosuid. This | |
3717 | * prevents namespaces from trusting potentially unsafe | |
3718 | * suid/sgid bits, file caps, or security labels that originate | |
3719 | * in other namespaces. | |
3720 | */ | |
3721 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3722 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3723 | } | |
3724 | ||
64964528 | 3725 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3726 | { |
58be2825 | 3727 | struct ns_common *ns = NULL; |
8823c079 EB |
3728 | struct nsproxy *nsproxy; |
3729 | ||
728dba3a EB |
3730 | task_lock(task); |
3731 | nsproxy = task->nsproxy; | |
8823c079 | 3732 | if (nsproxy) { |
58be2825 AV |
3733 | ns = &nsproxy->mnt_ns->ns; |
3734 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3735 | } |
728dba3a | 3736 | task_unlock(task); |
8823c079 EB |
3737 | |
3738 | return ns; | |
3739 | } | |
3740 | ||
64964528 | 3741 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3742 | { |
58be2825 | 3743 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3744 | } |
3745 | ||
64964528 | 3746 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3747 | { |
3748 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3749 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3750 | struct path root; |
4f757f3c | 3751 | int err; |
8823c079 | 3752 | |
0c55cfc4 | 3753 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3754 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3755 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3756 | return -EPERM; |
8823c079 | 3757 | |
74e83122 AV |
3758 | if (is_anon_ns(mnt_ns)) |
3759 | return -EINVAL; | |
3760 | ||
8823c079 EB |
3761 | if (fs->users != 1) |
3762 | return -EINVAL; | |
3763 | ||
3764 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3765 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3766 | nsproxy->mnt_ns = mnt_ns; |
3767 | ||
3768 | /* Find the root */ | |
4f757f3c AV |
3769 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3770 | "/", LOOKUP_DOWN, &root); | |
3771 | if (err) { | |
3772 | /* revert to old namespace */ | |
3773 | nsproxy->mnt_ns = old_mnt_ns; | |
3774 | put_mnt_ns(mnt_ns); | |
3775 | return err; | |
3776 | } | |
8823c079 | 3777 | |
4068367c AV |
3778 | put_mnt_ns(old_mnt_ns); |
3779 | ||
8823c079 EB |
3780 | /* Update the pwd and root */ |
3781 | set_fs_pwd(fs, &root); | |
3782 | set_fs_root(fs, &root); | |
3783 | ||
3784 | path_put(&root); | |
3785 | return 0; | |
3786 | } | |
3787 | ||
bcac25a5 AV |
3788 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3789 | { | |
3790 | return to_mnt_ns(ns)->user_ns; | |
3791 | } | |
3792 | ||
8823c079 EB |
3793 | const struct proc_ns_operations mntns_operations = { |
3794 | .name = "mnt", | |
3795 | .type = CLONE_NEWNS, | |
3796 | .get = mntns_get, | |
3797 | .put = mntns_put, | |
3798 | .install = mntns_install, | |
bcac25a5 | 3799 | .owner = mntns_owner, |
8823c079 | 3800 | }; |