]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
23 | #include <linux/uaccess.h> | |
0bb80f24 | 24 | #include <linux/proc_ns.h> |
20b4fb48 | 25 | #include <linux/magic.h> |
0818bf27 | 26 | #include <linux/bootmem.h> |
9ea459e1 | 27 | #include <linux/task_work.h> |
9164bb4a IM |
28 | #include <linux/sched/task.h> |
29 | ||
07b20889 | 30 | #include "pnode.h" |
948730b0 | 31 | #include "internal.h" |
1da177e4 | 32 | |
d2921684 EB |
33 | /* Maximum number of mounts in a mount namespace */ |
34 | unsigned int sysctl_mount_max __read_mostly = 100000; | |
35 | ||
0818bf27 AV |
36 | static unsigned int m_hash_mask __read_mostly; |
37 | static unsigned int m_hash_shift __read_mostly; | |
38 | static unsigned int mp_hash_mask __read_mostly; | |
39 | static unsigned int mp_hash_shift __read_mostly; | |
40 | ||
41 | static __initdata unsigned long mhash_entries; | |
42 | static int __init set_mhash_entries(char *str) | |
43 | { | |
44 | if (!str) | |
45 | return 0; | |
46 | mhash_entries = simple_strtoul(str, &str, 0); | |
47 | return 1; | |
48 | } | |
49 | __setup("mhash_entries=", set_mhash_entries); | |
50 | ||
51 | static __initdata unsigned long mphash_entries; | |
52 | static int __init set_mphash_entries(char *str) | |
53 | { | |
54 | if (!str) | |
55 | return 0; | |
56 | mphash_entries = simple_strtoul(str, &str, 0); | |
57 | return 1; | |
58 | } | |
59 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 60 | |
c7999c36 | 61 | static u64 event; |
73cd49ec | 62 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 63 | static DEFINE_IDA(mnt_group_ida); |
99b7db7b | 64 | static DEFINE_SPINLOCK(mnt_id_lock); |
f21f6220 AV |
65 | static int mnt_id_start = 0; |
66 | static int mnt_group_start = 1; | |
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 MS |
103 | { |
104 | int res; | |
105 | ||
106 | retry: | |
107 | ida_pre_get(&mnt_id_ida, GFP_KERNEL); | |
99b7db7b | 108 | spin_lock(&mnt_id_lock); |
15169fe7 | 109 | res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); |
f21f6220 | 110 | if (!res) |
15169fe7 | 111 | mnt_id_start = mnt->mnt_id + 1; |
99b7db7b | 112 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
113 | if (res == -EAGAIN) |
114 | goto retry; | |
115 | ||
116 | return res; | |
117 | } | |
118 | ||
b105e270 | 119 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 120 | { |
15169fe7 | 121 | int id = mnt->mnt_id; |
99b7db7b | 122 | spin_lock(&mnt_id_lock); |
f21f6220 AV |
123 | ida_remove(&mnt_id_ida, id); |
124 | if (mnt_id_start > id) | |
125 | mnt_id_start = id; | |
99b7db7b | 126 | spin_unlock(&mnt_id_lock); |
73cd49ec MS |
127 | } |
128 | ||
719f5d7f MS |
129 | /* |
130 | * Allocate a new peer group ID | |
131 | * | |
132 | * mnt_group_ida is protected by namespace_sem | |
133 | */ | |
4b8b21f4 | 134 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 135 | { |
f21f6220 AV |
136 | int res; |
137 | ||
719f5d7f MS |
138 | if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) |
139 | return -ENOMEM; | |
140 | ||
f21f6220 AV |
141 | res = ida_get_new_above(&mnt_group_ida, |
142 | mnt_group_start, | |
15169fe7 | 143 | &mnt->mnt_group_id); |
f21f6220 | 144 | if (!res) |
15169fe7 | 145 | mnt_group_start = mnt->mnt_group_id + 1; |
f21f6220 AV |
146 | |
147 | return res; | |
719f5d7f MS |
148 | } |
149 | ||
150 | /* | |
151 | * Release a peer group ID | |
152 | */ | |
4b8b21f4 | 153 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 154 | { |
15169fe7 | 155 | int id = mnt->mnt_group_id; |
f21f6220 AV |
156 | ida_remove(&mnt_group_ida, id); |
157 | if (mnt_group_start > id) | |
158 | mnt_group_start = id; | |
15169fe7 | 159 | mnt->mnt_group_id = 0; |
719f5d7f MS |
160 | } |
161 | ||
b3e19d92 NP |
162 | /* |
163 | * vfsmount lock must be held for read | |
164 | */ | |
83adc753 | 165 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
166 | { |
167 | #ifdef CONFIG_SMP | |
68e8a9fe | 168 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
169 | #else |
170 | preempt_disable(); | |
68e8a9fe | 171 | mnt->mnt_count += n; |
b3e19d92 NP |
172 | preempt_enable(); |
173 | #endif | |
174 | } | |
175 | ||
b3e19d92 NP |
176 | /* |
177 | * vfsmount lock must be held for write | |
178 | */ | |
83adc753 | 179 | unsigned int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
180 | { |
181 | #ifdef CONFIG_SMP | |
f03c6599 | 182 | unsigned int count = 0; |
b3e19d92 NP |
183 | int cpu; |
184 | ||
185 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 186 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
187 | } |
188 | ||
189 | return count; | |
190 | #else | |
68e8a9fe | 191 | return mnt->mnt_count; |
b3e19d92 NP |
192 | #endif |
193 | } | |
194 | ||
87b95ce0 AV |
195 | static void drop_mountpoint(struct fs_pin *p) |
196 | { | |
197 | struct mount *m = container_of(p, struct mount, mnt_umount); | |
198 | dput(m->mnt_ex_mountpoint); | |
199 | pin_remove(p); | |
200 | mntput(&m->mnt); | |
201 | } | |
202 | ||
b105e270 | 203 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 204 | { |
c63181e6 AV |
205 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
206 | if (mnt) { | |
73cd49ec MS |
207 | int err; |
208 | ||
c63181e6 | 209 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
210 | if (err) |
211 | goto out_free_cache; | |
212 | ||
213 | if (name) { | |
fcc139ae | 214 | mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); |
c63181e6 | 215 | if (!mnt->mnt_devname) |
88b38782 | 216 | goto out_free_id; |
73cd49ec MS |
217 | } |
218 | ||
b3e19d92 | 219 | #ifdef CONFIG_SMP |
c63181e6 AV |
220 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
221 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
222 | goto out_free_devname; |
223 | ||
c63181e6 | 224 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 225 | #else |
c63181e6 AV |
226 | mnt->mnt_count = 1; |
227 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
228 | #endif |
229 | ||
38129a13 | 230 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
231 | INIT_LIST_HEAD(&mnt->mnt_child); |
232 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
233 | INIT_LIST_HEAD(&mnt->mnt_list); | |
234 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
235 | INIT_LIST_HEAD(&mnt->mnt_share); | |
236 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
237 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 238 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 239 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
87b95ce0 | 240 | init_fs_pin(&mnt->mnt_umount, drop_mountpoint); |
1da177e4 | 241 | } |
c63181e6 | 242 | return mnt; |
88b38782 | 243 | |
d3ef3d73 NP |
244 | #ifdef CONFIG_SMP |
245 | out_free_devname: | |
fcc139ae | 246 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 247 | #endif |
88b38782 | 248 | out_free_id: |
c63181e6 | 249 | mnt_free_id(mnt); |
88b38782 | 250 | out_free_cache: |
c63181e6 | 251 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 252 | return NULL; |
1da177e4 LT |
253 | } |
254 | ||
3d733633 DH |
255 | /* |
256 | * Most r/o checks on a fs are for operations that take | |
257 | * discrete amounts of time, like a write() or unlink(). | |
258 | * We must keep track of when those operations start | |
259 | * (for permission checks) and when they end, so that | |
260 | * we can determine when writes are able to occur to | |
261 | * a filesystem. | |
262 | */ | |
263 | /* | |
264 | * __mnt_is_readonly: check whether a mount is read-only | |
265 | * @mnt: the mount to check for its write status | |
266 | * | |
267 | * This shouldn't be used directly ouside of the VFS. | |
268 | * It does not guarantee that the filesystem will stay | |
269 | * r/w, just that it is right *now*. This can not and | |
270 | * should not be used in place of IS_RDONLY(inode). | |
271 | * mnt_want/drop_write() will _keep_ the filesystem | |
272 | * r/w. | |
273 | */ | |
274 | int __mnt_is_readonly(struct vfsmount *mnt) | |
275 | { | |
2e4b7fcd DH |
276 | if (mnt->mnt_flags & MNT_READONLY) |
277 | return 1; | |
bc98a42c | 278 | if (sb_rdonly(mnt->mnt_sb)) |
2e4b7fcd DH |
279 | return 1; |
280 | return 0; | |
3d733633 DH |
281 | } |
282 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
283 | ||
83adc753 | 284 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
285 | { |
286 | #ifdef CONFIG_SMP | |
68e8a9fe | 287 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 288 | #else |
68e8a9fe | 289 | mnt->mnt_writers++; |
d3ef3d73 NP |
290 | #endif |
291 | } | |
3d733633 | 292 | |
83adc753 | 293 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 294 | { |
d3ef3d73 | 295 | #ifdef CONFIG_SMP |
68e8a9fe | 296 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 297 | #else |
68e8a9fe | 298 | mnt->mnt_writers--; |
d3ef3d73 | 299 | #endif |
3d733633 | 300 | } |
3d733633 | 301 | |
83adc753 | 302 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 303 | { |
d3ef3d73 NP |
304 | #ifdef CONFIG_SMP |
305 | unsigned int count = 0; | |
3d733633 | 306 | int cpu; |
3d733633 DH |
307 | |
308 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 309 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 310 | } |
3d733633 | 311 | |
d3ef3d73 NP |
312 | return count; |
313 | #else | |
314 | return mnt->mnt_writers; | |
315 | #endif | |
3d733633 DH |
316 | } |
317 | ||
4ed5e82f MS |
318 | static int mnt_is_readonly(struct vfsmount *mnt) |
319 | { | |
320 | if (mnt->mnt_sb->s_readonly_remount) | |
321 | return 1; | |
322 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
323 | smp_rmb(); | |
324 | return __mnt_is_readonly(mnt); | |
325 | } | |
326 | ||
8366025e | 327 | /* |
eb04c282 JK |
328 | * Most r/o & frozen checks on a fs are for operations that take discrete |
329 | * amounts of time, like a write() or unlink(). We must keep track of when | |
330 | * those operations start (for permission checks) and when they end, so that we | |
331 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
332 | */ |
333 | /** | |
eb04c282 | 334 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 335 | * @m: the mount on which to take a write |
8366025e | 336 | * |
eb04c282 JK |
337 | * This tells the low-level filesystem that a write is about to be performed to |
338 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
339 | * returning success. This operation does not protect against filesystem being | |
340 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
341 | * called. This is effectively a refcount. | |
8366025e | 342 | */ |
eb04c282 | 343 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 344 | { |
83adc753 | 345 | struct mount *mnt = real_mount(m); |
3d733633 | 346 | int ret = 0; |
3d733633 | 347 | |
d3ef3d73 | 348 | preempt_disable(); |
c6653a83 | 349 | mnt_inc_writers(mnt); |
d3ef3d73 | 350 | /* |
c6653a83 | 351 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
352 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
353 | * incremented count after it has set MNT_WRITE_HOLD. | |
354 | */ | |
355 | smp_mb(); | |
6aa7de05 | 356 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 NP |
357 | cpu_relax(); |
358 | /* | |
359 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
360 | * be set to match its requirements. So we must not load that until | |
361 | * MNT_WRITE_HOLD is cleared. | |
362 | */ | |
363 | smp_rmb(); | |
4ed5e82f | 364 | if (mnt_is_readonly(m)) { |
c6653a83 | 365 | mnt_dec_writers(mnt); |
3d733633 | 366 | ret = -EROFS; |
3d733633 | 367 | } |
d3ef3d73 | 368 | preempt_enable(); |
eb04c282 JK |
369 | |
370 | return ret; | |
371 | } | |
372 | ||
373 | /** | |
374 | * mnt_want_write - get write access to a mount | |
375 | * @m: the mount on which to take a write | |
376 | * | |
377 | * This tells the low-level filesystem that a write is about to be performed to | |
378 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
379 | * is not frozen) before returning success. When the write operation is | |
380 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
381 | */ | |
382 | int mnt_want_write(struct vfsmount *m) | |
383 | { | |
384 | int ret; | |
385 | ||
386 | sb_start_write(m->mnt_sb); | |
387 | ret = __mnt_want_write(m); | |
388 | if (ret) | |
389 | sb_end_write(m->mnt_sb); | |
3d733633 | 390 | return ret; |
8366025e DH |
391 | } |
392 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
393 | ||
96029c4e NP |
394 | /** |
395 | * mnt_clone_write - get write access to a mount | |
396 | * @mnt: the mount on which to take a write | |
397 | * | |
398 | * This is effectively like mnt_want_write, except | |
399 | * it must only be used to take an extra write reference | |
400 | * on a mountpoint that we already know has a write reference | |
401 | * on it. This allows some optimisation. | |
402 | * | |
403 | * After finished, mnt_drop_write must be called as usual to | |
404 | * drop the reference. | |
405 | */ | |
406 | int mnt_clone_write(struct vfsmount *mnt) | |
407 | { | |
408 | /* superblock may be r/o */ | |
409 | if (__mnt_is_readonly(mnt)) | |
410 | return -EROFS; | |
411 | preempt_disable(); | |
83adc753 | 412 | mnt_inc_writers(real_mount(mnt)); |
96029c4e NP |
413 | preempt_enable(); |
414 | return 0; | |
415 | } | |
416 | EXPORT_SYMBOL_GPL(mnt_clone_write); | |
417 | ||
418 | /** | |
eb04c282 | 419 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
420 | * @file: the file who's mount on which to take a write |
421 | * | |
eb04c282 | 422 | * This is like __mnt_want_write, but it takes a file and can |
96029c4e NP |
423 | * do some optimisations if the file is open for write already |
424 | */ | |
eb04c282 | 425 | int __mnt_want_write_file(struct file *file) |
96029c4e | 426 | { |
83f936c7 | 427 | if (!(file->f_mode & FMODE_WRITER)) |
eb04c282 | 428 | return __mnt_want_write(file->f_path.mnt); |
96029c4e NP |
429 | else |
430 | return mnt_clone_write(file->f_path.mnt); | |
431 | } | |
eb04c282 JK |
432 | |
433 | /** | |
7c6893e3 | 434 | * mnt_want_write_file_path - get write access to a file's mount |
eb04c282 JK |
435 | * @file: the file who's mount on which to take a write |
436 | * | |
437 | * This is like mnt_want_write, but it takes a file and can | |
438 | * do some optimisations if the file is open for write already | |
7c6893e3 MS |
439 | * |
440 | * Called by the vfs for cases when we have an open file at hand, but will do an | |
441 | * inode operation on it (important distinction for files opened on overlayfs, | |
442 | * since the file operations will come from the real underlying file, while | |
443 | * inode operations come from the overlay). | |
eb04c282 | 444 | */ |
7c6893e3 | 445 | int mnt_want_write_file_path(struct file *file) |
eb04c282 JK |
446 | { |
447 | int ret; | |
448 | ||
449 | sb_start_write(file->f_path.mnt->mnt_sb); | |
450 | ret = __mnt_want_write_file(file); | |
451 | if (ret) | |
452 | sb_end_write(file->f_path.mnt->mnt_sb); | |
453 | return ret; | |
454 | } | |
7c6893e3 MS |
455 | |
456 | static inline int may_write_real(struct file *file) | |
457 | { | |
458 | struct dentry *dentry = file->f_path.dentry; | |
459 | struct dentry *upperdentry; | |
460 | ||
461 | /* Writable file? */ | |
462 | if (file->f_mode & FMODE_WRITER) | |
463 | return 0; | |
464 | ||
465 | /* Not overlayfs? */ | |
466 | if (likely(!(dentry->d_flags & DCACHE_OP_REAL))) | |
467 | return 0; | |
468 | ||
469 | /* File refers to upper, writable layer? */ | |
470 | upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER); | |
954c736f AG |
471 | if (upperdentry && |
472 | (file_inode(file) == d_inode(upperdentry) || | |
473 | file_inode(file) == d_inode(dentry))) | |
7c6893e3 MS |
474 | return 0; |
475 | ||
476 | /* Lower layer: can't write to real file, sorry... */ | |
477 | return -EPERM; | |
478 | } | |
479 | ||
480 | /** | |
481 | * mnt_want_write_file - get write access to a file's mount | |
482 | * @file: the file who's mount on which to take a write | |
483 | * | |
484 | * This is like mnt_want_write, but it takes a file and can | |
485 | * do some optimisations if the file is open for write already | |
486 | * | |
487 | * Mostly called by filesystems from their ioctl operation before performing | |
488 | * modification. On overlayfs this needs to check if the file is on a read-only | |
489 | * lower layer and deny access in that case. | |
490 | */ | |
491 | int mnt_want_write_file(struct file *file) | |
492 | { | |
493 | int ret; | |
494 | ||
495 | ret = may_write_real(file); | |
496 | if (!ret) { | |
497 | sb_start_write(file_inode(file)->i_sb); | |
498 | ret = __mnt_want_write_file(file); | |
499 | if (ret) | |
500 | sb_end_write(file_inode(file)->i_sb); | |
501 | } | |
502 | return ret; | |
503 | } | |
96029c4e NP |
504 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
505 | ||
8366025e | 506 | /** |
eb04c282 | 507 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
508 | * @mnt: the mount on which to give up write access |
509 | * | |
510 | * Tells the low-level filesystem that we are done | |
511 | * performing writes to it. Must be matched with | |
eb04c282 | 512 | * __mnt_want_write() call above. |
8366025e | 513 | */ |
eb04c282 | 514 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 515 | { |
d3ef3d73 | 516 | preempt_disable(); |
83adc753 | 517 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 518 | preempt_enable(); |
8366025e | 519 | } |
eb04c282 JK |
520 | |
521 | /** | |
522 | * mnt_drop_write - give up write access to a mount | |
523 | * @mnt: the mount on which to give up write access | |
524 | * | |
525 | * Tells the low-level filesystem that we are done performing writes to it and | |
526 | * also allows filesystem to be frozen again. Must be matched with | |
527 | * mnt_want_write() call above. | |
528 | */ | |
529 | void mnt_drop_write(struct vfsmount *mnt) | |
530 | { | |
531 | __mnt_drop_write(mnt); | |
532 | sb_end_write(mnt->mnt_sb); | |
533 | } | |
8366025e DH |
534 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
535 | ||
eb04c282 JK |
536 | void __mnt_drop_write_file(struct file *file) |
537 | { | |
538 | __mnt_drop_write(file->f_path.mnt); | |
539 | } | |
540 | ||
7c6893e3 | 541 | void mnt_drop_write_file_path(struct file *file) |
2a79f17e AV |
542 | { |
543 | mnt_drop_write(file->f_path.mnt); | |
544 | } | |
7c6893e3 MS |
545 | |
546 | void mnt_drop_write_file(struct file *file) | |
547 | { | |
548 | __mnt_drop_write(file->f_path.mnt); | |
549 | sb_end_write(file_inode(file)->i_sb); | |
550 | } | |
2a79f17e AV |
551 | EXPORT_SYMBOL(mnt_drop_write_file); |
552 | ||
83adc753 | 553 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 554 | { |
3d733633 DH |
555 | int ret = 0; |
556 | ||
719ea2fb | 557 | lock_mount_hash(); |
83adc753 | 558 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 559 | /* |
d3ef3d73 NP |
560 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
561 | * should be visible before we do. | |
3d733633 | 562 | */ |
d3ef3d73 NP |
563 | smp_mb(); |
564 | ||
3d733633 | 565 | /* |
d3ef3d73 NP |
566 | * With writers on hold, if this value is zero, then there are |
567 | * definitely no active writers (although held writers may subsequently | |
568 | * increment the count, they'll have to wait, and decrement it after | |
569 | * seeing MNT_READONLY). | |
570 | * | |
571 | * It is OK to have counter incremented on one CPU and decremented on | |
572 | * another: the sum will add up correctly. The danger would be when we | |
573 | * sum up each counter, if we read a counter before it is incremented, | |
574 | * but then read another CPU's count which it has been subsequently | |
575 | * decremented from -- we would see more decrements than we should. | |
576 | * MNT_WRITE_HOLD protects against this scenario, because | |
577 | * mnt_want_write first increments count, then smp_mb, then spins on | |
578 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
579 | * we're counting up here. | |
3d733633 | 580 | */ |
c6653a83 | 581 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 NP |
582 | ret = -EBUSY; |
583 | else | |
83adc753 | 584 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 NP |
585 | /* |
586 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
587 | * that become unheld will see MNT_READONLY. | |
588 | */ | |
589 | smp_wmb(); | |
83adc753 | 590 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 591 | unlock_mount_hash(); |
3d733633 | 592 | return ret; |
8366025e | 593 | } |
8366025e | 594 | |
83adc753 | 595 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 596 | { |
719ea2fb | 597 | lock_mount_hash(); |
83adc753 | 598 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 599 | unlock_mount_hash(); |
2e4b7fcd DH |
600 | } |
601 | ||
4ed5e82f MS |
602 | int sb_prepare_remount_readonly(struct super_block *sb) |
603 | { | |
604 | struct mount *mnt; | |
605 | int err = 0; | |
606 | ||
8e8b8796 MS |
607 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
608 | if (atomic_long_read(&sb->s_remove_count)) | |
609 | return -EBUSY; | |
610 | ||
719ea2fb | 611 | lock_mount_hash(); |
4ed5e82f MS |
612 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
613 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
614 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
615 | smp_mb(); | |
616 | if (mnt_get_writers(mnt) > 0) { | |
617 | err = -EBUSY; | |
618 | break; | |
619 | } | |
620 | } | |
621 | } | |
8e8b8796 MS |
622 | if (!err && atomic_long_read(&sb->s_remove_count)) |
623 | err = -EBUSY; | |
624 | ||
4ed5e82f MS |
625 | if (!err) { |
626 | sb->s_readonly_remount = 1; | |
627 | smp_wmb(); | |
628 | } | |
629 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
630 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
631 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
632 | } | |
719ea2fb | 633 | unlock_mount_hash(); |
4ed5e82f MS |
634 | |
635 | return err; | |
636 | } | |
637 | ||
b105e270 | 638 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 639 | { |
fcc139ae | 640 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 641 | #ifdef CONFIG_SMP |
68e8a9fe | 642 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 643 | #endif |
b105e270 | 644 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
645 | } |
646 | ||
8ffcb32e DH |
647 | static void delayed_free_vfsmnt(struct rcu_head *head) |
648 | { | |
649 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
650 | } | |
651 | ||
48a066e7 | 652 | /* call under rcu_read_lock */ |
294d71ff | 653 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
654 | { |
655 | struct mount *mnt; | |
656 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 657 | return 1; |
48a066e7 | 658 | if (bastard == NULL) |
294d71ff | 659 | return 0; |
48a066e7 AV |
660 | mnt = real_mount(bastard); |
661 | mnt_add_count(mnt, 1); | |
119e1ef8 | 662 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 663 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 664 | return 0; |
48a066e7 AV |
665 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
666 | mnt_add_count(mnt, -1); | |
294d71ff AV |
667 | return 1; |
668 | } | |
119e1ef8 AV |
669 | lock_mount_hash(); |
670 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
671 | mnt_add_count(mnt, -1); | |
672 | unlock_mount_hash(); | |
673 | return 1; | |
674 | } | |
675 | unlock_mount_hash(); | |
676 | /* caller will mntput() */ | |
294d71ff AV |
677 | return -1; |
678 | } | |
679 | ||
680 | /* call under rcu_read_lock */ | |
681 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
682 | { | |
683 | int res = __legitimize_mnt(bastard, seq); | |
684 | if (likely(!res)) | |
685 | return true; | |
686 | if (unlikely(res < 0)) { | |
687 | rcu_read_unlock(); | |
688 | mntput(bastard); | |
689 | rcu_read_lock(); | |
48a066e7 | 690 | } |
48a066e7 AV |
691 | return false; |
692 | } | |
693 | ||
1da177e4 | 694 | /* |
474279dc | 695 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 696 | * call under rcu_read_lock() |
1da177e4 | 697 | */ |
474279dc | 698 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 699 | { |
38129a13 | 700 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
701 | struct mount *p; |
702 | ||
38129a13 | 703 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
704 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
705 | return p; | |
706 | return NULL; | |
707 | } | |
708 | ||
a05964f3 | 709 | /* |
f015f126 DH |
710 | * lookup_mnt - Return the first child mount mounted at path |
711 | * | |
712 | * "First" means first mounted chronologically. If you create the | |
713 | * following mounts: | |
714 | * | |
715 | * mount /dev/sda1 /mnt | |
716 | * mount /dev/sda2 /mnt | |
717 | * mount /dev/sda3 /mnt | |
718 | * | |
719 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
720 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
721 | * /dev/sda2, then /dev/sda3, then NULL. | |
722 | * | |
723 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 724 | */ |
ca71cf71 | 725 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 726 | { |
c7105365 | 727 | struct mount *child_mnt; |
48a066e7 AV |
728 | struct vfsmount *m; |
729 | unsigned seq; | |
99b7db7b | 730 | |
48a066e7 AV |
731 | rcu_read_lock(); |
732 | do { | |
733 | seq = read_seqbegin(&mount_lock); | |
734 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
735 | m = child_mnt ? &child_mnt->mnt : NULL; | |
736 | } while (!legitimize_mnt(m, seq)); | |
737 | rcu_read_unlock(); | |
738 | return m; | |
a05964f3 RP |
739 | } |
740 | ||
7af1364f EB |
741 | /* |
742 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
743 | * current mount namespace. | |
744 | * | |
745 | * The common case is dentries are not mountpoints at all and that | |
746 | * test is handled inline. For the slow case when we are actually | |
747 | * dealing with a mountpoint of some kind, walk through all of the | |
748 | * mounts in the current mount namespace and test to see if the dentry | |
749 | * is a mountpoint. | |
750 | * | |
751 | * The mount_hashtable is not usable in the context because we | |
752 | * need to identify all mounts that may be in the current mount | |
753 | * namespace not just a mount that happens to have some specified | |
754 | * parent mount. | |
755 | */ | |
756 | bool __is_local_mountpoint(struct dentry *dentry) | |
757 | { | |
758 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
759 | struct mount *mnt; | |
760 | bool is_covered = false; | |
761 | ||
762 | if (!d_mountpoint(dentry)) | |
763 | goto out; | |
764 | ||
765 | down_read(&namespace_sem); | |
766 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
767 | is_covered = (mnt->mnt_mountpoint == dentry); | |
768 | if (is_covered) | |
769 | break; | |
770 | } | |
771 | up_read(&namespace_sem); | |
772 | out: | |
773 | return is_covered; | |
774 | } | |
775 | ||
e2dfa935 | 776 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 777 | { |
0818bf27 | 778 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
779 | struct mountpoint *mp; |
780 | ||
0818bf27 | 781 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
782 | if (mp->m_dentry == dentry) { |
783 | /* might be worth a WARN_ON() */ | |
784 | if (d_unlinked(dentry)) | |
785 | return ERR_PTR(-ENOENT); | |
786 | mp->m_count++; | |
787 | return mp; | |
788 | } | |
789 | } | |
e2dfa935 EB |
790 | return NULL; |
791 | } | |
792 | ||
3895dbf8 | 793 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 794 | { |
3895dbf8 | 795 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 796 | int ret; |
84d17192 | 797 | |
3895dbf8 EB |
798 | if (d_mountpoint(dentry)) { |
799 | mountpoint: | |
800 | read_seqlock_excl(&mount_lock); | |
801 | mp = lookup_mountpoint(dentry); | |
802 | read_sequnlock_excl(&mount_lock); | |
803 | if (mp) | |
804 | goto done; | |
805 | } | |
806 | ||
807 | if (!new) | |
808 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
809 | if (!new) | |
84d17192 AV |
810 | return ERR_PTR(-ENOMEM); |
811 | ||
3895dbf8 EB |
812 | |
813 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 814 | ret = d_set_mounted(dentry); |
eed81007 | 815 | |
3895dbf8 EB |
816 | /* Someone else set d_mounted? */ |
817 | if (ret == -EBUSY) | |
818 | goto mountpoint; | |
819 | ||
820 | /* The dentry is not available as a mountpoint? */ | |
821 | mp = ERR_PTR(ret); | |
822 | if (ret) | |
823 | goto done; | |
824 | ||
825 | /* Add the new mountpoint to the hash table */ | |
826 | read_seqlock_excl(&mount_lock); | |
827 | new->m_dentry = dentry; | |
828 | new->m_count = 1; | |
829 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
830 | INIT_HLIST_HEAD(&new->m_list); | |
831 | read_sequnlock_excl(&mount_lock); | |
832 | ||
833 | mp = new; | |
834 | new = NULL; | |
835 | done: | |
836 | kfree(new); | |
84d17192 AV |
837 | return mp; |
838 | } | |
839 | ||
840 | static void put_mountpoint(struct mountpoint *mp) | |
841 | { | |
842 | if (!--mp->m_count) { | |
843 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 844 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
845 | spin_lock(&dentry->d_lock); |
846 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
847 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 848 | hlist_del(&mp->m_hash); |
84d17192 AV |
849 | kfree(mp); |
850 | } | |
851 | } | |
852 | ||
143c8c91 | 853 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 854 | { |
6b3286ed | 855 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
856 | } |
857 | ||
99b7db7b NP |
858 | /* |
859 | * vfsmount lock must be held for write | |
860 | */ | |
6b3286ed | 861 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
862 | { |
863 | if (ns) { | |
864 | ns->event = ++event; | |
865 | wake_up_interruptible(&ns->poll); | |
866 | } | |
867 | } | |
868 | ||
99b7db7b NP |
869 | /* |
870 | * vfsmount lock must be held for write | |
871 | */ | |
6b3286ed | 872 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
873 | { |
874 | if (ns && ns->event != event) { | |
875 | ns->event = event; | |
876 | wake_up_interruptible(&ns->poll); | |
877 | } | |
878 | } | |
879 | ||
99b7db7b NP |
880 | /* |
881 | * vfsmount lock must be held for write | |
882 | */ | |
7bdb11de | 883 | static void unhash_mnt(struct mount *mnt) |
419148da | 884 | { |
0714a533 | 885 | mnt->mnt_parent = mnt; |
a73324da | 886 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 887 | list_del_init(&mnt->mnt_child); |
38129a13 | 888 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 889 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
890 | put_mountpoint(mnt->mnt_mp); |
891 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
892 | } |
893 | ||
7bdb11de EB |
894 | /* |
895 | * vfsmount lock must be held for write | |
896 | */ | |
897 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
898 | { | |
899 | old_path->dentry = mnt->mnt_mountpoint; | |
900 | old_path->mnt = &mnt->mnt_parent->mnt; | |
901 | unhash_mnt(mnt); | |
902 | } | |
903 | ||
6a46c573 EB |
904 | /* |
905 | * vfsmount lock must be held for write | |
906 | */ | |
907 | static void umount_mnt(struct mount *mnt) | |
908 | { | |
909 | /* old mountpoint will be dropped when we can do that */ | |
910 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
911 | unhash_mnt(mnt); | |
912 | } | |
913 | ||
99b7db7b NP |
914 | /* |
915 | * vfsmount lock must be held for write | |
916 | */ | |
84d17192 AV |
917 | void mnt_set_mountpoint(struct mount *mnt, |
918 | struct mountpoint *mp, | |
44d964d6 | 919 | struct mount *child_mnt) |
b90fa9ae | 920 | { |
84d17192 | 921 | mp->m_count++; |
3a2393d7 | 922 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 923 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 924 | child_mnt->mnt_parent = mnt; |
84d17192 | 925 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 926 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
927 | } |
928 | ||
1064f874 EB |
929 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
930 | { | |
931 | hlist_add_head_rcu(&mnt->mnt_hash, | |
932 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
933 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
934 | } | |
935 | ||
99b7db7b NP |
936 | /* |
937 | * vfsmount lock must be held for write | |
938 | */ | |
84d17192 AV |
939 | static void attach_mnt(struct mount *mnt, |
940 | struct mount *parent, | |
941 | struct mountpoint *mp) | |
1da177e4 | 942 | { |
84d17192 | 943 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 944 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
945 | } |
946 | ||
1064f874 | 947 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 948 | { |
1064f874 EB |
949 | struct mountpoint *old_mp = mnt->mnt_mp; |
950 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
951 | struct mount *old_parent = mnt->mnt_parent; | |
952 | ||
953 | list_del_init(&mnt->mnt_child); | |
954 | hlist_del_init(&mnt->mnt_mp_list); | |
955 | hlist_del_init_rcu(&mnt->mnt_hash); | |
956 | ||
957 | attach_mnt(mnt, parent, mp); | |
958 | ||
959 | put_mountpoint(old_mp); | |
960 | ||
961 | /* | |
962 | * Safely avoid even the suggestion this code might sleep or | |
963 | * lock the mount hash by taking advantage of the knowledge that | |
964 | * mnt_change_mountpoint will not release the final reference | |
965 | * to a mountpoint. | |
966 | * | |
967 | * During mounting, the mount passed in as the parent mount will | |
968 | * continue to use the old mountpoint and during unmounting, the | |
969 | * old mountpoint will continue to exist until namespace_unlock, | |
970 | * which happens well after mnt_change_mountpoint. | |
971 | */ | |
972 | spin_lock(&old_mountpoint->d_lock); | |
973 | old_mountpoint->d_lockref.count--; | |
974 | spin_unlock(&old_mountpoint->d_lock); | |
975 | ||
976 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
977 | } |
978 | ||
b90fa9ae | 979 | /* |
99b7db7b | 980 | * vfsmount lock must be held for write |
b90fa9ae | 981 | */ |
1064f874 | 982 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 983 | { |
0714a533 | 984 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 985 | struct mount *m; |
b90fa9ae | 986 | LIST_HEAD(head); |
143c8c91 | 987 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 988 | |
0714a533 | 989 | BUG_ON(parent == mnt); |
b90fa9ae | 990 | |
1a4eeaf2 | 991 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 992 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 993 | m->mnt_ns = n; |
f03c6599 | 994 | |
b90fa9ae RP |
995 | list_splice(&head, n->list.prev); |
996 | ||
d2921684 EB |
997 | n->mounts += n->pending_mounts; |
998 | n->pending_mounts = 0; | |
999 | ||
1064f874 | 1000 | __attach_mnt(mnt, parent); |
6b3286ed | 1001 | touch_mnt_namespace(n); |
1da177e4 LT |
1002 | } |
1003 | ||
909b0a88 | 1004 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 1005 | { |
6b41d536 AV |
1006 | struct list_head *next = p->mnt_mounts.next; |
1007 | if (next == &p->mnt_mounts) { | |
1da177e4 | 1008 | while (1) { |
909b0a88 | 1009 | if (p == root) |
1da177e4 | 1010 | return NULL; |
6b41d536 AV |
1011 | next = p->mnt_child.next; |
1012 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 1013 | break; |
0714a533 | 1014 | p = p->mnt_parent; |
1da177e4 LT |
1015 | } |
1016 | } | |
6b41d536 | 1017 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
1018 | } |
1019 | ||
315fc83e | 1020 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 1021 | { |
6b41d536 AV |
1022 | struct list_head *prev = p->mnt_mounts.prev; |
1023 | while (prev != &p->mnt_mounts) { | |
1024 | p = list_entry(prev, struct mount, mnt_child); | |
1025 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
1026 | } |
1027 | return p; | |
1028 | } | |
1029 | ||
9d412a43 AV |
1030 | struct vfsmount * |
1031 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
1032 | { | |
b105e270 | 1033 | struct mount *mnt; |
9d412a43 AV |
1034 | struct dentry *root; |
1035 | ||
1036 | if (!type) | |
1037 | return ERR_PTR(-ENODEV); | |
1038 | ||
1039 | mnt = alloc_vfsmnt(name); | |
1040 | if (!mnt) | |
1041 | return ERR_PTR(-ENOMEM); | |
1042 | ||
e462ec50 | 1043 | if (flags & SB_KERNMOUNT) |
b105e270 | 1044 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
1045 | |
1046 | root = mount_fs(type, flags, name, data); | |
1047 | if (IS_ERR(root)) { | |
8ffcb32e | 1048 | mnt_free_id(mnt); |
9d412a43 AV |
1049 | free_vfsmnt(mnt); |
1050 | return ERR_CAST(root); | |
1051 | } | |
1052 | ||
b105e270 AV |
1053 | mnt->mnt.mnt_root = root; |
1054 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 1055 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1056 | mnt->mnt_parent = mnt; |
719ea2fb | 1057 | lock_mount_hash(); |
39f7c4db | 1058 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 1059 | unlock_mount_hash(); |
b105e270 | 1060 | return &mnt->mnt; |
9d412a43 AV |
1061 | } |
1062 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1063 | ||
93faccbb EB |
1064 | struct vfsmount * |
1065 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1066 | const char *name, void *data) | |
1067 | { | |
1068 | /* Until it is worked out how to pass the user namespace | |
1069 | * through from the parent mount to the submount don't support | |
1070 | * unprivileged mounts with submounts. | |
1071 | */ | |
1072 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1073 | return ERR_PTR(-EPERM); | |
1074 | ||
e462ec50 | 1075 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1076 | } |
1077 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1078 | ||
87129cc0 | 1079 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1080 | int flag) |
1da177e4 | 1081 | { |
87129cc0 | 1082 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1083 | struct mount *mnt; |
1084 | int err; | |
1da177e4 | 1085 | |
be34d1a3 DH |
1086 | mnt = alloc_vfsmnt(old->mnt_devname); |
1087 | if (!mnt) | |
1088 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1089 | |
7a472ef4 | 1090 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1091 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1092 | else | |
1093 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1094 | |
be34d1a3 DH |
1095 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1096 | err = mnt_alloc_group_id(mnt); | |
1097 | if (err) | |
1098 | goto out_free; | |
1da177e4 | 1099 | } |
be34d1a3 | 1100 | |
16a34adb AV |
1101 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1102 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
132c94e3 | 1103 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1104 | if (flag & CL_UNPRIVILEGED) { |
1105 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1106 | ||
1107 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1108 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1109 | ||
1110 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1111 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1112 | ||
1113 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1114 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1115 | ||
1116 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1117 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1118 | } | |
132c94e3 | 1119 | |
5ff9d8a6 | 1120 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1121 | if ((flag & CL_UNPRIVILEGED) && |
1122 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1123 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1124 | ||
be34d1a3 DH |
1125 | atomic_inc(&sb->s_active); |
1126 | mnt->mnt.mnt_sb = sb; | |
1127 | mnt->mnt.mnt_root = dget(root); | |
1128 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1129 | mnt->mnt_parent = mnt; | |
719ea2fb | 1130 | lock_mount_hash(); |
be34d1a3 | 1131 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1132 | unlock_mount_hash(); |
be34d1a3 | 1133 | |
7a472ef4 EB |
1134 | if ((flag & CL_SLAVE) || |
1135 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1136 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1137 | mnt->mnt_master = old; | |
1138 | CLEAR_MNT_SHARED(mnt); | |
1139 | } else if (!(flag & CL_PRIVATE)) { | |
1140 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1141 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1142 | if (IS_MNT_SLAVE(old)) | |
1143 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1144 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1145 | } else { |
1146 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1147 | } |
1148 | if (flag & CL_MAKE_SHARED) | |
1149 | set_mnt_shared(mnt); | |
1150 | ||
1151 | /* stick the duplicate mount on the same expiry list | |
1152 | * as the original if that was on one */ | |
1153 | if (flag & CL_EXPIRE) { | |
1154 | if (!list_empty(&old->mnt_expire)) | |
1155 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1156 | } | |
1157 | ||
cb338d06 | 1158 | return mnt; |
719f5d7f MS |
1159 | |
1160 | out_free: | |
8ffcb32e | 1161 | mnt_free_id(mnt); |
719f5d7f | 1162 | free_vfsmnt(mnt); |
be34d1a3 | 1163 | return ERR_PTR(err); |
1da177e4 LT |
1164 | } |
1165 | ||
9ea459e1 AV |
1166 | static void cleanup_mnt(struct mount *mnt) |
1167 | { | |
1168 | /* | |
1169 | * This probably indicates that somebody messed | |
1170 | * up a mnt_want/drop_write() pair. If this | |
1171 | * happens, the filesystem was probably unable | |
1172 | * to make r/w->r/o transitions. | |
1173 | */ | |
1174 | /* | |
1175 | * The locking used to deal with mnt_count decrement provides barriers, | |
1176 | * so mnt_get_writers() below is safe. | |
1177 | */ | |
1178 | WARN_ON(mnt_get_writers(mnt)); | |
1179 | if (unlikely(mnt->mnt_pins.first)) | |
1180 | mnt_pin_kill(mnt); | |
1181 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1182 | dput(mnt->mnt.mnt_root); | |
1183 | deactivate_super(mnt->mnt.mnt_sb); | |
1184 | mnt_free_id(mnt); | |
1185 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1186 | } | |
1187 | ||
1188 | static void __cleanup_mnt(struct rcu_head *head) | |
1189 | { | |
1190 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1191 | } | |
1192 | ||
1193 | static LLIST_HEAD(delayed_mntput_list); | |
1194 | static void delayed_mntput(struct work_struct *unused) | |
1195 | { | |
1196 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1197 | struct mount *m, *t; |
9ea459e1 | 1198 | |
29785735 BP |
1199 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1200 | cleanup_mnt(m); | |
9ea459e1 AV |
1201 | } |
1202 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1203 | ||
900148dc | 1204 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1205 | { |
48a066e7 | 1206 | rcu_read_lock(); |
9ea0a46c AV |
1207 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1208 | /* | |
1209 | * Since we don't do lock_mount_hash() here, | |
1210 | * ->mnt_ns can change under us. However, if it's | |
1211 | * non-NULL, then there's a reference that won't | |
1212 | * be dropped until after an RCU delay done after | |
1213 | * turning ->mnt_ns NULL. So if we observe it | |
1214 | * non-NULL under rcu_read_lock(), the reference | |
1215 | * we are dropping is not the final one. | |
1216 | */ | |
1217 | mnt_add_count(mnt, -1); | |
48a066e7 | 1218 | rcu_read_unlock(); |
f03c6599 | 1219 | return; |
b3e19d92 | 1220 | } |
719ea2fb | 1221 | lock_mount_hash(); |
119e1ef8 AV |
1222 | /* |
1223 | * make sure that if __legitimize_mnt() has not seen us grab | |
1224 | * mount_lock, we'll see their refcount increment here. | |
1225 | */ | |
1226 | smp_mb(); | |
9ea0a46c | 1227 | mnt_add_count(mnt, -1); |
b3e19d92 | 1228 | if (mnt_get_count(mnt)) { |
48a066e7 | 1229 | rcu_read_unlock(); |
719ea2fb | 1230 | unlock_mount_hash(); |
99b7db7b NP |
1231 | return; |
1232 | } | |
48a066e7 AV |
1233 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1234 | rcu_read_unlock(); | |
1235 | unlock_mount_hash(); | |
1236 | return; | |
1237 | } | |
1238 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1239 | rcu_read_unlock(); | |
962830df | 1240 | |
39f7c4db | 1241 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1242 | |
1243 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1244 | struct mount *p, *tmp; | |
1245 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1246 | umount_mnt(p); | |
1247 | } | |
1248 | } | |
719ea2fb | 1249 | unlock_mount_hash(); |
649a795a | 1250 | |
9ea459e1 AV |
1251 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1252 | struct task_struct *task = current; | |
1253 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1254 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1255 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1256 | return; | |
1257 | } | |
1258 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1259 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1260 | return; | |
1261 | } | |
1262 | cleanup_mnt(mnt); | |
b3e19d92 | 1263 | } |
b3e19d92 NP |
1264 | |
1265 | void mntput(struct vfsmount *mnt) | |
1266 | { | |
1267 | if (mnt) { | |
863d684f | 1268 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1269 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1270 | if (unlikely(m->mnt_expiry_mark)) |
1271 | m->mnt_expiry_mark = 0; | |
1272 | mntput_no_expire(m); | |
b3e19d92 NP |
1273 | } |
1274 | } | |
1275 | EXPORT_SYMBOL(mntput); | |
1276 | ||
1277 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1278 | { | |
1279 | if (mnt) | |
83adc753 | 1280 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1281 | return mnt; |
1282 | } | |
1283 | EXPORT_SYMBOL(mntget); | |
1284 | ||
c6609c0a IK |
1285 | /* path_is_mountpoint() - Check if path is a mount in the current |
1286 | * namespace. | |
1287 | * | |
1288 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1289 | * not mounted in any namespace and that common case is handled inline. | |
1290 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1291 | * mounts using a given dentry in a different namespace. This function | |
1292 | * checks if the passed in path is a mountpoint rather than the dentry | |
1293 | * alone. | |
1294 | */ | |
1295 | bool path_is_mountpoint(const struct path *path) | |
1296 | { | |
1297 | unsigned seq; | |
1298 | bool res; | |
1299 | ||
1300 | if (!d_mountpoint(path->dentry)) | |
1301 | return false; | |
1302 | ||
1303 | rcu_read_lock(); | |
1304 | do { | |
1305 | seq = read_seqbegin(&mount_lock); | |
1306 | res = __path_is_mountpoint(path); | |
1307 | } while (read_seqretry(&mount_lock, seq)); | |
1308 | rcu_read_unlock(); | |
1309 | ||
1310 | return res; | |
1311 | } | |
1312 | EXPORT_SYMBOL(path_is_mountpoint); | |
1313 | ||
ca71cf71 | 1314 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1315 | { |
3064c356 AV |
1316 | struct mount *p; |
1317 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1318 | if (IS_ERR(p)) | |
1319 | return ERR_CAST(p); | |
1320 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1321 | return &p->mnt; | |
7b7b1ace | 1322 | } |
1da177e4 | 1323 | |
a1a2c409 | 1324 | #ifdef CONFIG_PROC_FS |
0226f492 | 1325 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1326 | static void *m_start(struct seq_file *m, loff_t *pos) |
1327 | { | |
ede1bf0d | 1328 | struct proc_mounts *p = m->private; |
1da177e4 | 1329 | |
390c6843 | 1330 | down_read(&namespace_sem); |
c7999c36 AV |
1331 | if (p->cached_event == p->ns->event) { |
1332 | void *v = p->cached_mount; | |
1333 | if (*pos == p->cached_index) | |
1334 | return v; | |
1335 | if (*pos == p->cached_index + 1) { | |
1336 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1337 | return p->cached_mount = v; | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | p->cached_event = p->ns->event; | |
1342 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1343 | p->cached_index = *pos; | |
1344 | return p->cached_mount; | |
1da177e4 LT |
1345 | } |
1346 | ||
1347 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1348 | { | |
ede1bf0d | 1349 | struct proc_mounts *p = m->private; |
b0765fb8 | 1350 | |
c7999c36 AV |
1351 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1352 | p->cached_index = *pos; | |
1353 | return p->cached_mount; | |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | static void m_stop(struct seq_file *m, void *v) | |
1357 | { | |
390c6843 | 1358 | up_read(&namespace_sem); |
1da177e4 LT |
1359 | } |
1360 | ||
0226f492 | 1361 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1362 | { |
ede1bf0d | 1363 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1364 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1365 | return p->show(m, &r->mnt); |
1da177e4 LT |
1366 | } |
1367 | ||
a1a2c409 | 1368 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1369 | .start = m_start, |
1370 | .next = m_next, | |
1371 | .stop = m_stop, | |
0226f492 | 1372 | .show = m_show, |
b4629fe2 | 1373 | }; |
a1a2c409 | 1374 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1375 | |
1da177e4 LT |
1376 | /** |
1377 | * may_umount_tree - check if a mount tree is busy | |
1378 | * @mnt: root of mount tree | |
1379 | * | |
1380 | * This is called to check if a tree of mounts has any | |
1381 | * open files, pwds, chroots or sub mounts that are | |
1382 | * busy. | |
1383 | */ | |
909b0a88 | 1384 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1385 | { |
909b0a88 | 1386 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1387 | int actual_refs = 0; |
1388 | int minimum_refs = 0; | |
315fc83e | 1389 | struct mount *p; |
909b0a88 | 1390 | BUG_ON(!m); |
1da177e4 | 1391 | |
b3e19d92 | 1392 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1393 | lock_mount_hash(); |
909b0a88 | 1394 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1395 | actual_refs += mnt_get_count(p); |
1da177e4 | 1396 | minimum_refs += 2; |
1da177e4 | 1397 | } |
719ea2fb | 1398 | unlock_mount_hash(); |
1da177e4 LT |
1399 | |
1400 | if (actual_refs > minimum_refs) | |
e3474a8e | 1401 | return 0; |
1da177e4 | 1402 | |
e3474a8e | 1403 | return 1; |
1da177e4 LT |
1404 | } |
1405 | ||
1406 | EXPORT_SYMBOL(may_umount_tree); | |
1407 | ||
1408 | /** | |
1409 | * may_umount - check if a mount point is busy | |
1410 | * @mnt: root of mount | |
1411 | * | |
1412 | * This is called to check if a mount point has any | |
1413 | * open files, pwds, chroots or sub mounts. If the | |
1414 | * mount has sub mounts this will return busy | |
1415 | * regardless of whether the sub mounts are busy. | |
1416 | * | |
1417 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1418 | * give false negatives. The main reason why it's here is that we need | |
1419 | * a non-destructive way to look for easily umountable filesystems. | |
1420 | */ | |
1421 | int may_umount(struct vfsmount *mnt) | |
1422 | { | |
e3474a8e | 1423 | int ret = 1; |
8ad08d8a | 1424 | down_read(&namespace_sem); |
719ea2fb | 1425 | lock_mount_hash(); |
1ab59738 | 1426 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1427 | ret = 0; |
719ea2fb | 1428 | unlock_mount_hash(); |
8ad08d8a | 1429 | up_read(&namespace_sem); |
a05964f3 | 1430 | return ret; |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | EXPORT_SYMBOL(may_umount); | |
1434 | ||
38129a13 | 1435 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1436 | |
97216be0 | 1437 | static void namespace_unlock(void) |
70fbcdf4 | 1438 | { |
a3b3c562 | 1439 | struct hlist_head head; |
97216be0 | 1440 | |
a3b3c562 | 1441 | hlist_move_list(&unmounted, &head); |
97216be0 | 1442 | |
97216be0 AV |
1443 | up_write(&namespace_sem); |
1444 | ||
a3b3c562 EB |
1445 | if (likely(hlist_empty(&head))) |
1446 | return; | |
1447 | ||
48a066e7 AV |
1448 | synchronize_rcu(); |
1449 | ||
87b95ce0 | 1450 | group_pin_kill(&head); |
70fbcdf4 RP |
1451 | } |
1452 | ||
97216be0 | 1453 | static inline void namespace_lock(void) |
e3197d83 | 1454 | { |
97216be0 | 1455 | down_write(&namespace_sem); |
e3197d83 AV |
1456 | } |
1457 | ||
e819f152 EB |
1458 | enum umount_tree_flags { |
1459 | UMOUNT_SYNC = 1, | |
1460 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1461 | UMOUNT_CONNECTED = 4, |
e819f152 | 1462 | }; |
f2d0a123 EB |
1463 | |
1464 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1465 | { | |
1466 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1467 | if (how & UMOUNT_SYNC) | |
1468 | return true; | |
1469 | ||
1470 | /* A mount without a parent has nothing to be connected to */ | |
1471 | if (!mnt_has_parent(mnt)) | |
1472 | return true; | |
1473 | ||
1474 | /* Because the reference counting rules change when mounts are | |
1475 | * unmounted and connected, umounted mounts may not be | |
1476 | * connected to mounted mounts. | |
1477 | */ | |
1478 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1479 | return true; | |
1480 | ||
1481 | /* Has it been requested that the mount remain connected? */ | |
1482 | if (how & UMOUNT_CONNECTED) | |
1483 | return false; | |
1484 | ||
1485 | /* Is the mount locked such that it needs to remain connected? */ | |
1486 | if (IS_MNT_LOCKED(mnt)) | |
1487 | return false; | |
1488 | ||
1489 | /* By default disconnect the mount */ | |
1490 | return true; | |
1491 | } | |
1492 | ||
99b7db7b | 1493 | /* |
48a066e7 | 1494 | * mount_lock must be held |
99b7db7b NP |
1495 | * namespace_sem must be held for write |
1496 | */ | |
e819f152 | 1497 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1498 | { |
c003b26f | 1499 | LIST_HEAD(tmp_list); |
315fc83e | 1500 | struct mount *p; |
1da177e4 | 1501 | |
5d88457e EB |
1502 | if (how & UMOUNT_PROPAGATE) |
1503 | propagate_mount_unlock(mnt); | |
1504 | ||
c003b26f | 1505 | /* Gather the mounts to umount */ |
590ce4bc EB |
1506 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1507 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1508 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1509 | } |
1da177e4 | 1510 | |
411a938b | 1511 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1512 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1513 | list_del_init(&p->mnt_child); |
c003b26f | 1514 | } |
88b368f2 | 1515 | |
c003b26f | 1516 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1517 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1518 | propagate_umount(&tmp_list); |
a05964f3 | 1519 | |
c003b26f | 1520 | while (!list_empty(&tmp_list)) { |
d2921684 | 1521 | struct mnt_namespace *ns; |
ce07d891 | 1522 | bool disconnect; |
c003b26f | 1523 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1524 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1525 | list_del_init(&p->mnt_list); |
d2921684 EB |
1526 | ns = p->mnt_ns; |
1527 | if (ns) { | |
1528 | ns->mounts--; | |
1529 | __touch_mnt_namespace(ns); | |
1530 | } | |
143c8c91 | 1531 | p->mnt_ns = NULL; |
e819f152 | 1532 | if (how & UMOUNT_SYNC) |
48a066e7 | 1533 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1534 | |
f2d0a123 | 1535 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1536 | |
1537 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1538 | disconnect ? &unmounted : NULL); | |
676da58d | 1539 | if (mnt_has_parent(p)) { |
81b6b061 | 1540 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1541 | if (!disconnect) { |
1542 | /* Don't forget about p */ | |
1543 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1544 | } else { | |
1545 | umount_mnt(p); | |
1546 | } | |
7c4b93d8 | 1547 | } |
0f0afb1d | 1548 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1549 | } |
1550 | } | |
1551 | ||
b54b9be7 | 1552 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1553 | |
1ab59738 | 1554 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1555 | { |
1ab59738 | 1556 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1557 | int retval; |
1558 | ||
1ab59738 | 1559 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1560 | if (retval) |
1561 | return retval; | |
1562 | ||
1563 | /* | |
1564 | * Allow userspace to request a mountpoint be expired rather than | |
1565 | * unmounting unconditionally. Unmount only happens if: | |
1566 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1567 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1568 | */ | |
1569 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1570 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1571 | flags & (MNT_FORCE | MNT_DETACH)) |
1572 | return -EINVAL; | |
1573 | ||
b3e19d92 NP |
1574 | /* |
1575 | * probably don't strictly need the lock here if we examined | |
1576 | * all race cases, but it's a slowpath. | |
1577 | */ | |
719ea2fb | 1578 | lock_mount_hash(); |
83adc753 | 1579 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1580 | unlock_mount_hash(); |
1da177e4 | 1581 | return -EBUSY; |
b3e19d92 | 1582 | } |
719ea2fb | 1583 | unlock_mount_hash(); |
1da177e4 | 1584 | |
863d684f | 1585 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1586 | return -EAGAIN; |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * If we may have to abort operations to get out of this | |
1591 | * mount, and they will themselves hold resources we must | |
1592 | * allow the fs to do things. In the Unix tradition of | |
1593 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1594 | * might fail to complete on the first run through as other tasks | |
1595 | * must return, and the like. Thats for the mount program to worry | |
1596 | * about for the moment. | |
1597 | */ | |
1598 | ||
42faad99 | 1599 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1600 | sb->s_op->umount_begin(sb); |
42faad99 | 1601 | } |
1da177e4 LT |
1602 | |
1603 | /* | |
1604 | * No sense to grab the lock for this test, but test itself looks | |
1605 | * somewhat bogus. Suggestions for better replacement? | |
1606 | * Ho-hum... In principle, we might treat that as umount + switch | |
1607 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1608 | * Actually it makes sense, especially if rootfs would contain a | |
1609 | * /reboot - static binary that would close all descriptors and | |
1610 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1611 | */ | |
1ab59738 | 1612 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1613 | /* |
1614 | * Special case for "unmounting" root ... | |
1615 | * we just try to remount it readonly. | |
1616 | */ | |
bc6155d1 | 1617 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1618 | return -EPERM; |
1da177e4 | 1619 | down_write(&sb->s_umount); |
bc98a42c | 1620 | if (!sb_rdonly(sb)) |
e462ec50 | 1621 | retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); |
1da177e4 LT |
1622 | up_write(&sb->s_umount); |
1623 | return retval; | |
1624 | } | |
1625 | ||
97216be0 | 1626 | namespace_lock(); |
719ea2fb | 1627 | lock_mount_hash(); |
5addc5dd | 1628 | event++; |
1da177e4 | 1629 | |
48a066e7 | 1630 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1631 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1632 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1633 | retval = 0; |
48a066e7 AV |
1634 | } else { |
1635 | shrink_submounts(mnt); | |
1636 | retval = -EBUSY; | |
1637 | if (!propagate_mount_busy(mnt, 2)) { | |
1638 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1639 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1640 | retval = 0; |
1641 | } | |
1da177e4 | 1642 | } |
719ea2fb | 1643 | unlock_mount_hash(); |
e3197d83 | 1644 | namespace_unlock(); |
1da177e4 LT |
1645 | return retval; |
1646 | } | |
1647 | ||
80b5dce8 EB |
1648 | /* |
1649 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1650 | * | |
1651 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1652 | * to an existing mountpoint, and wind up leaking the mount. | |
1653 | * detach_mounts allows lazily unmounting those mounts instead of | |
1654 | * leaking them. | |
1655 | * | |
1656 | * The caller may hold dentry->d_inode->i_mutex. | |
1657 | */ | |
1658 | void __detach_mounts(struct dentry *dentry) | |
1659 | { | |
1660 | struct mountpoint *mp; | |
1661 | struct mount *mnt; | |
1662 | ||
1663 | namespace_lock(); | |
3895dbf8 | 1664 | lock_mount_hash(); |
80b5dce8 | 1665 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1666 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1667 | goto out_unlock; |
1668 | ||
e06b933e | 1669 | event++; |
80b5dce8 EB |
1670 | while (!hlist_empty(&mp->m_list)) { |
1671 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1672 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1673 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1674 | umount_mnt(mnt); | |
ce07d891 | 1675 | } |
e0c9c0af | 1676 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1677 | } |
80b5dce8 EB |
1678 | put_mountpoint(mp); |
1679 | out_unlock: | |
3895dbf8 | 1680 | unlock_mount_hash(); |
80b5dce8 EB |
1681 | namespace_unlock(); |
1682 | } | |
1683 | ||
dd111b31 | 1684 | /* |
9b40bc90 AV |
1685 | * Is the caller allowed to modify his namespace? |
1686 | */ | |
1687 | static inline bool may_mount(void) | |
1688 | { | |
1689 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1690 | } | |
1691 | ||
9e8925b6 JL |
1692 | static inline bool may_mandlock(void) |
1693 | { | |
1694 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1695 | return false; | |
1696 | #endif | |
95ace754 | 1697 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1698 | } |
1699 | ||
1da177e4 LT |
1700 | /* |
1701 | * Now umount can handle mount points as well as block devices. | |
1702 | * This is important for filesystems which use unnamed block devices. | |
1703 | * | |
1704 | * We now support a flag for forced unmount like the other 'big iron' | |
1705 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1706 | */ | |
1707 | ||
3a18ef5c | 1708 | int ksys_umount(char __user *name, int flags) |
1da177e4 | 1709 | { |
2d8f3038 | 1710 | struct path path; |
900148dc | 1711 | struct mount *mnt; |
1da177e4 | 1712 | int retval; |
db1f05bb | 1713 | int lookup_flags = 0; |
1da177e4 | 1714 | |
db1f05bb MS |
1715 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1716 | return -EINVAL; | |
1717 | ||
9b40bc90 AV |
1718 | if (!may_mount()) |
1719 | return -EPERM; | |
1720 | ||
db1f05bb MS |
1721 | if (!(flags & UMOUNT_NOFOLLOW)) |
1722 | lookup_flags |= LOOKUP_FOLLOW; | |
1723 | ||
197df04c | 1724 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1725 | if (retval) |
1726 | goto out; | |
900148dc | 1727 | mnt = real_mount(path.mnt); |
1da177e4 | 1728 | retval = -EINVAL; |
2d8f3038 | 1729 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1730 | goto dput_and_out; |
143c8c91 | 1731 | if (!check_mnt(mnt)) |
1da177e4 | 1732 | goto dput_and_out; |
5ff9d8a6 EB |
1733 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1734 | goto dput_and_out; | |
b2f5d4dc EB |
1735 | retval = -EPERM; |
1736 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1737 | goto dput_and_out; | |
1da177e4 | 1738 | |
900148dc | 1739 | retval = do_umount(mnt, flags); |
1da177e4 | 1740 | dput_and_out: |
429731b1 | 1741 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1742 | dput(path.dentry); |
900148dc | 1743 | mntput_no_expire(mnt); |
1da177e4 LT |
1744 | out: |
1745 | return retval; | |
1746 | } | |
1747 | ||
3a18ef5c DB |
1748 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1749 | { | |
1750 | return ksys_umount(name, flags); | |
1751 | } | |
1752 | ||
1da177e4 LT |
1753 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1754 | ||
1755 | /* | |
b58fed8b | 1756 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1757 | */ |
bdc480e3 | 1758 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1759 | { |
3a18ef5c | 1760 | return ksys_umount(name, 0); |
1da177e4 LT |
1761 | } |
1762 | ||
1763 | #endif | |
1764 | ||
4ce5d2b1 | 1765 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1766 | { |
4ce5d2b1 | 1767 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1768 | return dentry->d_op == &ns_dentry_operations && |
1769 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1770 | } |
1771 | ||
58be2825 AV |
1772 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1773 | { | |
1774 | return container_of(ns, struct mnt_namespace, ns); | |
1775 | } | |
1776 | ||
4ce5d2b1 EB |
1777 | static bool mnt_ns_loop(struct dentry *dentry) |
1778 | { | |
1779 | /* Could bind mounting the mount namespace inode cause a | |
1780 | * mount namespace loop? | |
1781 | */ | |
1782 | struct mnt_namespace *mnt_ns; | |
1783 | if (!is_mnt_ns_file(dentry)) | |
1784 | return false; | |
1785 | ||
f77c8014 | 1786 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1787 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1788 | } | |
1789 | ||
87129cc0 | 1790 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1791 | int flag) |
1da177e4 | 1792 | { |
84d17192 | 1793 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1794 | |
4ce5d2b1 EB |
1795 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1796 | return ERR_PTR(-EINVAL); | |
1797 | ||
1798 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1799 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1800 | |
36341f64 | 1801 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1802 | if (IS_ERR(q)) |
1803 | return q; | |
1804 | ||
a73324da | 1805 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1806 | |
1807 | p = mnt; | |
6b41d536 | 1808 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1809 | struct mount *s; |
7ec02ef1 | 1810 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1811 | continue; |
1812 | ||
909b0a88 | 1813 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1814 | if (!(flag & CL_COPY_UNBINDABLE) && |
1815 | IS_MNT_UNBINDABLE(s)) { | |
1816 | s = skip_mnt_tree(s); | |
1817 | continue; | |
1818 | } | |
1819 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1820 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1821 | s = skip_mnt_tree(s); |
1822 | continue; | |
1823 | } | |
0714a533 AV |
1824 | while (p != s->mnt_parent) { |
1825 | p = p->mnt_parent; | |
1826 | q = q->mnt_parent; | |
1da177e4 | 1827 | } |
87129cc0 | 1828 | p = s; |
84d17192 | 1829 | parent = q; |
87129cc0 | 1830 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1831 | if (IS_ERR(q)) |
1832 | goto out; | |
719ea2fb | 1833 | lock_mount_hash(); |
1a4eeaf2 | 1834 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1835 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1836 | unlock_mount_hash(); |
1da177e4 LT |
1837 | } |
1838 | } | |
1839 | return res; | |
be34d1a3 | 1840 | out: |
1da177e4 | 1841 | if (res) { |
719ea2fb | 1842 | lock_mount_hash(); |
e819f152 | 1843 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1844 | unlock_mount_hash(); |
1da177e4 | 1845 | } |
be34d1a3 | 1846 | return q; |
1da177e4 LT |
1847 | } |
1848 | ||
be34d1a3 DH |
1849 | /* Caller should check returned pointer for errors */ |
1850 | ||
ca71cf71 | 1851 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1852 | { |
cb338d06 | 1853 | struct mount *tree; |
97216be0 | 1854 | namespace_lock(); |
cd4a4017 EB |
1855 | if (!check_mnt(real_mount(path->mnt))) |
1856 | tree = ERR_PTR(-EINVAL); | |
1857 | else | |
1858 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1859 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1860 | namespace_unlock(); |
be34d1a3 | 1861 | if (IS_ERR(tree)) |
52e220d3 | 1862 | return ERR_CAST(tree); |
be34d1a3 | 1863 | return &tree->mnt; |
8aec0809 AV |
1864 | } |
1865 | ||
1866 | void drop_collected_mounts(struct vfsmount *mnt) | |
1867 | { | |
97216be0 | 1868 | namespace_lock(); |
719ea2fb | 1869 | lock_mount_hash(); |
e819f152 | 1870 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1871 | unlock_mount_hash(); |
3ab6abee | 1872 | namespace_unlock(); |
8aec0809 AV |
1873 | } |
1874 | ||
c771d683 MS |
1875 | /** |
1876 | * clone_private_mount - create a private clone of a path | |
1877 | * | |
1878 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1879 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1880 | * to the originating mount won't be propagated into this). | |
1881 | * | |
1882 | * Release with mntput(). | |
1883 | */ | |
ca71cf71 | 1884 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1885 | { |
1886 | struct mount *old_mnt = real_mount(path->mnt); | |
1887 | struct mount *new_mnt; | |
1888 | ||
1889 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1890 | return ERR_PTR(-EINVAL); | |
1891 | ||
c771d683 | 1892 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1893 | if (IS_ERR(new_mnt)) |
1894 | return ERR_CAST(new_mnt); | |
1895 | ||
1896 | return &new_mnt->mnt; | |
1897 | } | |
1898 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1899 | ||
1f707137 AV |
1900 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1901 | struct vfsmount *root) | |
1902 | { | |
1a4eeaf2 | 1903 | struct mount *mnt; |
1f707137 AV |
1904 | int res = f(root, arg); |
1905 | if (res) | |
1906 | return res; | |
1a4eeaf2 AV |
1907 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1908 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1909 | if (res) |
1910 | return res; | |
1911 | } | |
1912 | return 0; | |
1913 | } | |
1914 | ||
4b8b21f4 | 1915 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1916 | { |
315fc83e | 1917 | struct mount *p; |
719f5d7f | 1918 | |
909b0a88 | 1919 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1920 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1921 | mnt_release_group_id(p); |
719f5d7f MS |
1922 | } |
1923 | } | |
1924 | ||
4b8b21f4 | 1925 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1926 | { |
315fc83e | 1927 | struct mount *p; |
719f5d7f | 1928 | |
909b0a88 | 1929 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1930 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1931 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1932 | if (err) { |
4b8b21f4 | 1933 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1934 | return err; |
1935 | } | |
1936 | } | |
1937 | } | |
1938 | ||
1939 | return 0; | |
1940 | } | |
1941 | ||
d2921684 EB |
1942 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1943 | { | |
1944 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1945 | unsigned int mounts = 0, old, pending, sum; | |
1946 | struct mount *p; | |
1947 | ||
1948 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1949 | mounts++; | |
1950 | ||
1951 | old = ns->mounts; | |
1952 | pending = ns->pending_mounts; | |
1953 | sum = old + pending; | |
1954 | if ((old > sum) || | |
1955 | (pending > sum) || | |
1956 | (max < sum) || | |
1957 | (mounts > (max - sum))) | |
1958 | return -ENOSPC; | |
1959 | ||
1960 | ns->pending_mounts = pending + mounts; | |
1961 | return 0; | |
1962 | } | |
1963 | ||
b90fa9ae RP |
1964 | /* |
1965 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1966 | * @nd : place the mount tree @source_mnt is attached |
1967 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1968 | * store the parent mount and mountpoint dentry. | |
1969 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1970 | * |
1971 | * NOTE: in the table below explains the semantics when a source mount | |
1972 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1973 | * --------------------------------------------------------------------------- |
1974 | * | BIND MOUNT OPERATION | | |
1975 | * |************************************************************************** | |
1976 | * | source-->| shared | private | slave | unbindable | | |
1977 | * | dest | | | | | | |
1978 | * | | | | | | | | |
1979 | * | v | | | | | | |
1980 | * |************************************************************************** | |
1981 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1982 | * | | | | | | | |
1983 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1984 | * *************************************************************************** | |
b90fa9ae RP |
1985 | * A bind operation clones the source mount and mounts the clone on the |
1986 | * destination mount. | |
1987 | * | |
1988 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1989 | * tree of the destination mount and the cloned mount is added to | |
1990 | * the peer group of the source mount. | |
1991 | * (+) the cloned mount is created under the destination mount and is marked | |
1992 | * as shared. The cloned mount is added to the peer group of the source | |
1993 | * mount. | |
5afe0022 RP |
1994 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1995 | * of the destination mount and the cloned mount is made slave | |
1996 | * of the same master as that of the source mount. The cloned mount | |
1997 | * is marked as 'shared and slave'. | |
1998 | * (*) the cloned mount is made a slave of the same master as that of the | |
1999 | * source mount. | |
2000 | * | |
9676f0c6 RP |
2001 | * --------------------------------------------------------------------------- |
2002 | * | MOVE MOUNT OPERATION | | |
2003 | * |************************************************************************** | |
2004 | * | source-->| shared | private | slave | unbindable | | |
2005 | * | dest | | | | | | |
2006 | * | | | | | | | | |
2007 | * | v | | | | | | |
2008 | * |************************************************************************** | |
2009 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2010 | * | | | | | | | |
2011 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2012 | * *************************************************************************** | |
5afe0022 RP |
2013 | * |
2014 | * (+) the mount is moved to the destination. And is then propagated to | |
2015 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2016 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2017 | * (+++) the mount is moved to the destination and is then propagated to |
2018 | * all the mounts belonging to the destination mount's propagation tree. | |
2019 | * the mount is marked as 'shared and slave'. | |
2020 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2021 | * |
2022 | * if the source mount is a tree, the operations explained above is | |
2023 | * applied to each mount in the tree. | |
2024 | * Must be called without spinlocks held, since this function can sleep | |
2025 | * in allocations. | |
2026 | */ | |
0fb54e50 | 2027 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2028 | struct mount *dest_mnt, |
2029 | struct mountpoint *dest_mp, | |
2030 | struct path *parent_path) | |
b90fa9ae | 2031 | { |
38129a13 | 2032 | HLIST_HEAD(tree_list); |
d2921684 | 2033 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2034 | struct mountpoint *smp; |
315fc83e | 2035 | struct mount *child, *p; |
38129a13 | 2036 | struct hlist_node *n; |
719f5d7f | 2037 | int err; |
b90fa9ae | 2038 | |
1064f874 EB |
2039 | /* Preallocate a mountpoint in case the new mounts need |
2040 | * to be tucked under other mounts. | |
2041 | */ | |
2042 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2043 | if (IS_ERR(smp)) | |
2044 | return PTR_ERR(smp); | |
2045 | ||
d2921684 EB |
2046 | /* Is there space to add these mounts to the mount namespace? */ |
2047 | if (!parent_path) { | |
2048 | err = count_mounts(ns, source_mnt); | |
2049 | if (err) | |
2050 | goto out; | |
2051 | } | |
2052 | ||
fc7be130 | 2053 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2054 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2055 | if (err) |
2056 | goto out; | |
0b1b901b | 2057 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2058 | lock_mount_hash(); |
0b1b901b AV |
2059 | if (err) |
2060 | goto out_cleanup_ids; | |
909b0a88 | 2061 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2062 | set_mnt_shared(p); |
0b1b901b AV |
2063 | } else { |
2064 | lock_mount_hash(); | |
b90fa9ae | 2065 | } |
1a390689 | 2066 | if (parent_path) { |
0fb54e50 | 2067 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2068 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2069 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2070 | } else { |
84d17192 | 2071 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2072 | commit_tree(source_mnt); |
21444403 | 2073 | } |
b90fa9ae | 2074 | |
38129a13 | 2075 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2076 | struct mount *q; |
38129a13 | 2077 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2078 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2079 | child->mnt_mountpoint); | |
2080 | if (q) | |
2081 | mnt_change_mountpoint(child, smp, q); | |
2082 | commit_tree(child); | |
b90fa9ae | 2083 | } |
1064f874 | 2084 | put_mountpoint(smp); |
719ea2fb | 2085 | unlock_mount_hash(); |
99b7db7b | 2086 | |
b90fa9ae | 2087 | return 0; |
719f5d7f MS |
2088 | |
2089 | out_cleanup_ids: | |
f2ebb3a9 AV |
2090 | while (!hlist_empty(&tree_list)) { |
2091 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2092 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2093 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2094 | } |
2095 | unlock_mount_hash(); | |
0b1b901b | 2096 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2097 | out: |
d2921684 | 2098 | ns->pending_mounts = 0; |
1064f874 EB |
2099 | |
2100 | read_seqlock_excl(&mount_lock); | |
2101 | put_mountpoint(smp); | |
2102 | read_sequnlock_excl(&mount_lock); | |
2103 | ||
719f5d7f | 2104 | return err; |
b90fa9ae RP |
2105 | } |
2106 | ||
84d17192 | 2107 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2108 | { |
2109 | struct vfsmount *mnt; | |
84d17192 | 2110 | struct dentry *dentry = path->dentry; |
b12cea91 | 2111 | retry: |
5955102c | 2112 | inode_lock(dentry->d_inode); |
84d17192 | 2113 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2114 | inode_unlock(dentry->d_inode); |
84d17192 | 2115 | return ERR_PTR(-ENOENT); |
b12cea91 | 2116 | } |
97216be0 | 2117 | namespace_lock(); |
b12cea91 | 2118 | mnt = lookup_mnt(path); |
84d17192 | 2119 | if (likely(!mnt)) { |
3895dbf8 | 2120 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2121 | if (IS_ERR(mp)) { |
97216be0 | 2122 | namespace_unlock(); |
5955102c | 2123 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2124 | return mp; |
2125 | } | |
2126 | return mp; | |
2127 | } | |
97216be0 | 2128 | namespace_unlock(); |
5955102c | 2129 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2130 | path_put(path); |
2131 | path->mnt = mnt; | |
84d17192 | 2132 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2133 | goto retry; |
2134 | } | |
2135 | ||
84d17192 | 2136 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2137 | { |
84d17192 | 2138 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2139 | |
2140 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2141 | put_mountpoint(where); |
3895dbf8 EB |
2142 | read_sequnlock_excl(&mount_lock); |
2143 | ||
328e6d90 | 2144 | namespace_unlock(); |
5955102c | 2145 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2146 | } |
2147 | ||
84d17192 | 2148 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2149 | { |
e462ec50 | 2150 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2151 | return -EINVAL; |
2152 | ||
e36cb0b8 DH |
2153 | if (d_is_dir(mp->m_dentry) != |
2154 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2155 | return -ENOTDIR; |
2156 | ||
84d17192 | 2157 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2158 | } |
2159 | ||
7a2e8a8f VA |
2160 | /* |
2161 | * Sanity check the flags to change_mnt_propagation. | |
2162 | */ | |
2163 | ||
e462ec50 | 2164 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2165 | { |
e462ec50 | 2166 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2167 | |
2168 | /* Fail if any non-propagation flags are set */ | |
2169 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2170 | return 0; | |
2171 | /* Only one propagation flag should be set */ | |
2172 | if (!is_power_of_2(type)) | |
2173 | return 0; | |
2174 | return type; | |
2175 | } | |
2176 | ||
07b20889 RP |
2177 | /* |
2178 | * recursively change the type of the mountpoint. | |
2179 | */ | |
e462ec50 | 2180 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2181 | { |
315fc83e | 2182 | struct mount *m; |
4b8b21f4 | 2183 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2184 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2185 | int type; |
719f5d7f | 2186 | int err = 0; |
07b20889 | 2187 | |
2d92ab3c | 2188 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2189 | return -EINVAL; |
2190 | ||
e462ec50 | 2191 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2192 | if (!type) |
2193 | return -EINVAL; | |
2194 | ||
97216be0 | 2195 | namespace_lock(); |
719f5d7f MS |
2196 | if (type == MS_SHARED) { |
2197 | err = invent_group_ids(mnt, recurse); | |
2198 | if (err) | |
2199 | goto out_unlock; | |
2200 | } | |
2201 | ||
719ea2fb | 2202 | lock_mount_hash(); |
909b0a88 | 2203 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2204 | change_mnt_propagation(m, type); |
719ea2fb | 2205 | unlock_mount_hash(); |
719f5d7f MS |
2206 | |
2207 | out_unlock: | |
97216be0 | 2208 | namespace_unlock(); |
719f5d7f | 2209 | return err; |
07b20889 RP |
2210 | } |
2211 | ||
5ff9d8a6 EB |
2212 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2213 | { | |
2214 | struct mount *child; | |
2215 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2216 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2217 | continue; | |
2218 | ||
2219 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2220 | return true; | |
2221 | } | |
2222 | return false; | |
2223 | } | |
2224 | ||
1da177e4 LT |
2225 | /* |
2226 | * do loopback mount. | |
2227 | */ | |
808d4e3c | 2228 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2229 | int recurse) |
1da177e4 | 2230 | { |
2d92ab3c | 2231 | struct path old_path; |
84d17192 AV |
2232 | struct mount *mnt = NULL, *old, *parent; |
2233 | struct mountpoint *mp; | |
57eccb83 | 2234 | int err; |
1da177e4 LT |
2235 | if (!old_name || !*old_name) |
2236 | return -EINVAL; | |
815d405c | 2237 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2238 | if (err) |
2239 | return err; | |
2240 | ||
8823c079 | 2241 | err = -EINVAL; |
4ce5d2b1 | 2242 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2243 | goto out; |
8823c079 | 2244 | |
84d17192 AV |
2245 | mp = lock_mount(path); |
2246 | err = PTR_ERR(mp); | |
2247 | if (IS_ERR(mp)) | |
b12cea91 AV |
2248 | goto out; |
2249 | ||
87129cc0 | 2250 | old = real_mount(old_path.mnt); |
84d17192 | 2251 | parent = real_mount(path->mnt); |
87129cc0 | 2252 | |
1da177e4 | 2253 | err = -EINVAL; |
fc7be130 | 2254 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2255 | goto out2; |
9676f0c6 | 2256 | |
e149ed2b AV |
2257 | if (!check_mnt(parent)) |
2258 | goto out2; | |
2259 | ||
2260 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2261 | goto out2; |
1da177e4 | 2262 | |
5ff9d8a6 EB |
2263 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2264 | goto out2; | |
2265 | ||
ccd48bc7 | 2266 | if (recurse) |
4ce5d2b1 | 2267 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2268 | else |
87129cc0 | 2269 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2270 | |
be34d1a3 DH |
2271 | if (IS_ERR(mnt)) { |
2272 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2273 | goto out2; |
be34d1a3 | 2274 | } |
ccd48bc7 | 2275 | |
5ff9d8a6 EB |
2276 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2277 | ||
84d17192 | 2278 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2279 | if (err) { |
719ea2fb | 2280 | lock_mount_hash(); |
e819f152 | 2281 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2282 | unlock_mount_hash(); |
5b83d2c5 | 2283 | } |
b12cea91 | 2284 | out2: |
84d17192 | 2285 | unlock_mount(mp); |
ccd48bc7 | 2286 | out: |
2d92ab3c | 2287 | path_put(&old_path); |
1da177e4 LT |
2288 | return err; |
2289 | } | |
2290 | ||
2e4b7fcd DH |
2291 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2292 | { | |
2293 | int error = 0; | |
2294 | int readonly_request = 0; | |
2295 | ||
2296 | if (ms_flags & MS_RDONLY) | |
2297 | readonly_request = 1; | |
2298 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2299 | return 0; | |
2300 | ||
2301 | if (readonly_request) | |
83adc753 | 2302 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2303 | else |
83adc753 | 2304 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2305 | return error; |
2306 | } | |
2307 | ||
1da177e4 LT |
2308 | /* |
2309 | * change filesystem flags. dir should be a physical root of filesystem. | |
2310 | * If you've mounted a non-root directory somewhere and want to do remount | |
2311 | * on it - tough luck. | |
2312 | */ | |
e462ec50 DH |
2313 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2314 | int mnt_flags, void *data) | |
1da177e4 LT |
2315 | { |
2316 | int err; | |
2d92ab3c | 2317 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2318 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2319 | |
143c8c91 | 2320 | if (!check_mnt(mnt)) |
1da177e4 LT |
2321 | return -EINVAL; |
2322 | ||
2d92ab3c | 2323 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2324 | return -EINVAL; |
2325 | ||
07b64558 EB |
2326 | /* Don't allow changing of locked mnt flags. |
2327 | * | |
2328 | * No locks need to be held here while testing the various | |
2329 | * MNT_LOCK flags because those flags can never be cleared | |
2330 | * once they are set. | |
2331 | */ | |
2332 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2333 | !(mnt_flags & MNT_READONLY)) { | |
2334 | return -EPERM; | |
2335 | } | |
9566d674 EB |
2336 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2337 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2338 | return -EPERM; |
9566d674 EB |
2339 | } |
2340 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2341 | !(mnt_flags & MNT_NOSUID)) { | |
2342 | return -EPERM; | |
2343 | } | |
2344 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2345 | !(mnt_flags & MNT_NOEXEC)) { | |
2346 | return -EPERM; | |
2347 | } | |
2348 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2349 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2350 | return -EPERM; | |
2351 | } | |
2352 | ||
ff36fe2c EP |
2353 | err = security_sb_remount(sb, data); |
2354 | if (err) | |
2355 | return err; | |
2356 | ||
1da177e4 | 2357 | down_write(&sb->s_umount); |
e462ec50 DH |
2358 | if (ms_flags & MS_BIND) |
2359 | err = change_mount_flags(path->mnt, ms_flags); | |
bc6155d1 | 2360 | else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
57eccb83 | 2361 | err = -EPERM; |
4aa98cf7 | 2362 | else |
e462ec50 | 2363 | err = do_remount_sb(sb, sb_flags, data, 0); |
7b43a79f | 2364 | if (!err) { |
719ea2fb | 2365 | lock_mount_hash(); |
a6138db8 | 2366 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2367 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2368 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2369 | unlock_mount_hash(); |
0e55a7cc | 2370 | } |
6339dab8 | 2371 | up_write(&sb->s_umount); |
1da177e4 LT |
2372 | return err; |
2373 | } | |
2374 | ||
cbbe362c | 2375 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2376 | { |
315fc83e | 2377 | struct mount *p; |
909b0a88 | 2378 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2379 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2380 | return 1; |
2381 | } | |
2382 | return 0; | |
2383 | } | |
2384 | ||
808d4e3c | 2385 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2386 | { |
2d92ab3c | 2387 | struct path old_path, parent_path; |
676da58d | 2388 | struct mount *p; |
0fb54e50 | 2389 | struct mount *old; |
84d17192 | 2390 | struct mountpoint *mp; |
57eccb83 | 2391 | int err; |
1da177e4 LT |
2392 | if (!old_name || !*old_name) |
2393 | return -EINVAL; | |
2d92ab3c | 2394 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2395 | if (err) |
2396 | return err; | |
2397 | ||
84d17192 AV |
2398 | mp = lock_mount(path); |
2399 | err = PTR_ERR(mp); | |
2400 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2401 | goto out; |
2402 | ||
143c8c91 | 2403 | old = real_mount(old_path.mnt); |
fc7be130 | 2404 | p = real_mount(path->mnt); |
143c8c91 | 2405 | |
1da177e4 | 2406 | err = -EINVAL; |
fc7be130 | 2407 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2408 | goto out1; |
2409 | ||
5ff9d8a6 EB |
2410 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2411 | goto out1; | |
2412 | ||
1da177e4 | 2413 | err = -EINVAL; |
2d92ab3c | 2414 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2415 | goto out1; |
1da177e4 | 2416 | |
676da58d | 2417 | if (!mnt_has_parent(old)) |
21444403 | 2418 | goto out1; |
1da177e4 | 2419 | |
e36cb0b8 DH |
2420 | if (d_is_dir(path->dentry) != |
2421 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2422 | goto out1; |
2423 | /* | |
2424 | * Don't move a mount residing in a shared parent. | |
2425 | */ | |
fc7be130 | 2426 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2427 | goto out1; |
9676f0c6 RP |
2428 | /* |
2429 | * Don't move a mount tree containing unbindable mounts to a destination | |
2430 | * mount which is shared. | |
2431 | */ | |
fc7be130 | 2432 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2433 | goto out1; |
1da177e4 | 2434 | err = -ELOOP; |
fc7be130 | 2435 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2436 | if (p == old) |
21444403 | 2437 | goto out1; |
1da177e4 | 2438 | |
84d17192 | 2439 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2440 | if (err) |
21444403 | 2441 | goto out1; |
1da177e4 LT |
2442 | |
2443 | /* if the mount is moved, it should no longer be expire | |
2444 | * automatically */ | |
6776db3d | 2445 | list_del_init(&old->mnt_expire); |
1da177e4 | 2446 | out1: |
84d17192 | 2447 | unlock_mount(mp); |
1da177e4 | 2448 | out: |
1da177e4 | 2449 | if (!err) |
1a390689 | 2450 | path_put(&parent_path); |
2d92ab3c | 2451 | path_put(&old_path); |
1da177e4 LT |
2452 | return err; |
2453 | } | |
2454 | ||
9d412a43 AV |
2455 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2456 | { | |
2457 | int err; | |
2458 | const char *subtype = strchr(fstype, '.'); | |
2459 | if (subtype) { | |
2460 | subtype++; | |
2461 | err = -EINVAL; | |
2462 | if (!subtype[0]) | |
2463 | goto err; | |
2464 | } else | |
2465 | subtype = ""; | |
2466 | ||
2467 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2468 | err = -ENOMEM; | |
2469 | if (!mnt->mnt_sb->s_subtype) | |
2470 | goto err; | |
2471 | return mnt; | |
2472 | ||
2473 | err: | |
2474 | mntput(mnt); | |
2475 | return ERR_PTR(err); | |
2476 | } | |
2477 | ||
9d412a43 AV |
2478 | /* |
2479 | * add a mount into a namespace's mount tree | |
2480 | */ | |
95bc5f25 | 2481 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2482 | { |
84d17192 AV |
2483 | struct mountpoint *mp; |
2484 | struct mount *parent; | |
9d412a43 AV |
2485 | int err; |
2486 | ||
f2ebb3a9 | 2487 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2488 | |
84d17192 AV |
2489 | mp = lock_mount(path); |
2490 | if (IS_ERR(mp)) | |
2491 | return PTR_ERR(mp); | |
9d412a43 | 2492 | |
84d17192 | 2493 | parent = real_mount(path->mnt); |
9d412a43 | 2494 | err = -EINVAL; |
84d17192 | 2495 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2496 | /* that's acceptable only for automounts done in private ns */ |
2497 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2498 | goto unlock; | |
2499 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2500 | if (!parent->mnt_ns) |
156cacb1 AV |
2501 | goto unlock; |
2502 | } | |
9d412a43 AV |
2503 | |
2504 | /* Refuse the same filesystem on the same mount point */ | |
2505 | err = -EBUSY; | |
95bc5f25 | 2506 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2507 | path->mnt->mnt_root == path->dentry) |
2508 | goto unlock; | |
2509 | ||
2510 | err = -EINVAL; | |
e36cb0b8 | 2511 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2512 | goto unlock; |
2513 | ||
95bc5f25 | 2514 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2515 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2516 | |
2517 | unlock: | |
84d17192 | 2518 | unlock_mount(mp); |
9d412a43 AV |
2519 | return err; |
2520 | } | |
b1e75df4 | 2521 | |
8654df4e | 2522 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2523 | |
1da177e4 LT |
2524 | /* |
2525 | * create a new mount for userspace and request it to be added into the | |
2526 | * namespace's tree | |
2527 | */ | |
e462ec50 | 2528 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2529 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2530 | { |
0c55cfc4 | 2531 | struct file_system_type *type; |
1da177e4 | 2532 | struct vfsmount *mnt; |
15f9a3f3 | 2533 | int err; |
1da177e4 | 2534 | |
0c55cfc4 | 2535 | if (!fstype) |
1da177e4 LT |
2536 | return -EINVAL; |
2537 | ||
0c55cfc4 EB |
2538 | type = get_fs_type(fstype); |
2539 | if (!type) | |
2540 | return -ENODEV; | |
2541 | ||
e462ec50 | 2542 | mnt = vfs_kern_mount(type, sb_flags, name, data); |
0c55cfc4 EB |
2543 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
2544 | !mnt->mnt_sb->s_subtype) | |
2545 | mnt = fs_set_subtype(mnt, fstype); | |
2546 | ||
2547 | put_filesystem(type); | |
1da177e4 LT |
2548 | if (IS_ERR(mnt)) |
2549 | return PTR_ERR(mnt); | |
2550 | ||
8654df4e EB |
2551 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2552 | mntput(mnt); | |
2553 | return -EPERM; | |
2554 | } | |
2555 | ||
95bc5f25 | 2556 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2557 | if (err) |
2558 | mntput(mnt); | |
2559 | return err; | |
1da177e4 LT |
2560 | } |
2561 | ||
19a167af AV |
2562 | int finish_automount(struct vfsmount *m, struct path *path) |
2563 | { | |
6776db3d | 2564 | struct mount *mnt = real_mount(m); |
19a167af AV |
2565 | int err; |
2566 | /* The new mount record should have at least 2 refs to prevent it being | |
2567 | * expired before we get a chance to add it | |
2568 | */ | |
6776db3d | 2569 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2570 | |
2571 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2572 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2573 | err = -ELOOP; |
2574 | goto fail; | |
19a167af AV |
2575 | } |
2576 | ||
95bc5f25 | 2577 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2578 | if (!err) |
2579 | return 0; | |
2580 | fail: | |
2581 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2582 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2583 | namespace_lock(); |
6776db3d | 2584 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2585 | namespace_unlock(); |
19a167af | 2586 | } |
b1e75df4 AV |
2587 | mntput(m); |
2588 | mntput(m); | |
19a167af AV |
2589 | return err; |
2590 | } | |
2591 | ||
ea5b778a DH |
2592 | /** |
2593 | * mnt_set_expiry - Put a mount on an expiration list | |
2594 | * @mnt: The mount to list. | |
2595 | * @expiry_list: The list to add the mount to. | |
2596 | */ | |
2597 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2598 | { | |
97216be0 | 2599 | namespace_lock(); |
ea5b778a | 2600 | |
6776db3d | 2601 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2602 | |
97216be0 | 2603 | namespace_unlock(); |
ea5b778a DH |
2604 | } |
2605 | EXPORT_SYMBOL(mnt_set_expiry); | |
2606 | ||
1da177e4 LT |
2607 | /* |
2608 | * process a list of expirable mountpoints with the intent of discarding any | |
2609 | * mountpoints that aren't in use and haven't been touched since last we came | |
2610 | * here | |
2611 | */ | |
2612 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2613 | { | |
761d5c38 | 2614 | struct mount *mnt, *next; |
1da177e4 LT |
2615 | LIST_HEAD(graveyard); |
2616 | ||
2617 | if (list_empty(mounts)) | |
2618 | return; | |
2619 | ||
97216be0 | 2620 | namespace_lock(); |
719ea2fb | 2621 | lock_mount_hash(); |
1da177e4 LT |
2622 | |
2623 | /* extract from the expiration list every vfsmount that matches the | |
2624 | * following criteria: | |
2625 | * - only referenced by its parent vfsmount | |
2626 | * - still marked for expiry (marked on the last call here; marks are | |
2627 | * cleared by mntput()) | |
2628 | */ | |
6776db3d | 2629 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2630 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2631 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2632 | continue; |
6776db3d | 2633 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2634 | } |
bcc5c7d2 | 2635 | while (!list_empty(&graveyard)) { |
6776db3d | 2636 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2637 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2638 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2639 | } |
719ea2fb | 2640 | unlock_mount_hash(); |
3ab6abee | 2641 | namespace_unlock(); |
5528f911 TM |
2642 | } |
2643 | ||
2644 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2645 | ||
2646 | /* | |
2647 | * Ripoff of 'select_parent()' | |
2648 | * | |
2649 | * search the list of submounts for a given mountpoint, and move any | |
2650 | * shrinkable submounts to the 'graveyard' list. | |
2651 | */ | |
692afc31 | 2652 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2653 | { |
692afc31 | 2654 | struct mount *this_parent = parent; |
5528f911 TM |
2655 | struct list_head *next; |
2656 | int found = 0; | |
2657 | ||
2658 | repeat: | |
6b41d536 | 2659 | next = this_parent->mnt_mounts.next; |
5528f911 | 2660 | resume: |
6b41d536 | 2661 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2662 | struct list_head *tmp = next; |
6b41d536 | 2663 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2664 | |
2665 | next = tmp->next; | |
692afc31 | 2666 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2667 | continue; |
5528f911 TM |
2668 | /* |
2669 | * Descend a level if the d_mounts list is non-empty. | |
2670 | */ | |
6b41d536 | 2671 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2672 | this_parent = mnt; |
2673 | goto repeat; | |
2674 | } | |
1da177e4 | 2675 | |
1ab59738 | 2676 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2677 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2678 | found++; |
2679 | } | |
1da177e4 | 2680 | } |
5528f911 TM |
2681 | /* |
2682 | * All done at this level ... ascend and resume the search | |
2683 | */ | |
2684 | if (this_parent != parent) { | |
6b41d536 | 2685 | next = this_parent->mnt_child.next; |
0714a533 | 2686 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2687 | goto resume; |
2688 | } | |
2689 | return found; | |
2690 | } | |
2691 | ||
2692 | /* | |
2693 | * process a list of expirable mountpoints with the intent of discarding any | |
2694 | * submounts of a specific parent mountpoint | |
99b7db7b | 2695 | * |
48a066e7 | 2696 | * mount_lock must be held for write |
5528f911 | 2697 | */ |
b54b9be7 | 2698 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2699 | { |
2700 | LIST_HEAD(graveyard); | |
761d5c38 | 2701 | struct mount *m; |
5528f911 | 2702 | |
5528f911 | 2703 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2704 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2705 | while (!list_empty(&graveyard)) { |
761d5c38 | 2706 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2707 | mnt_expire); |
143c8c91 | 2708 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2709 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2710 | } |
2711 | } | |
1da177e4 LT |
2712 | } |
2713 | ||
1da177e4 LT |
2714 | /* |
2715 | * Some copy_from_user() implementations do not return the exact number of | |
2716 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2717 | * Note that this function differs from copy_from_user() in that it will oops | |
2718 | * on bad values of `to', rather than returning a short copy. | |
2719 | */ | |
b58fed8b RP |
2720 | static long exact_copy_from_user(void *to, const void __user * from, |
2721 | unsigned long n) | |
1da177e4 LT |
2722 | { |
2723 | char *t = to; | |
2724 | const char __user *f = from; | |
2725 | char c; | |
2726 | ||
2727 | if (!access_ok(VERIFY_READ, from, n)) | |
2728 | return n; | |
2729 | ||
2730 | while (n) { | |
2731 | if (__get_user(c, f)) { | |
2732 | memset(t, 0, n); | |
2733 | break; | |
2734 | } | |
2735 | *t++ = c; | |
2736 | f++; | |
2737 | n--; | |
2738 | } | |
2739 | return n; | |
2740 | } | |
2741 | ||
b40ef869 | 2742 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2743 | { |
2744 | int i; | |
1da177e4 | 2745 | unsigned long size; |
b40ef869 | 2746 | char *copy; |
b58fed8b | 2747 | |
1da177e4 | 2748 | if (!data) |
b40ef869 | 2749 | return NULL; |
1da177e4 | 2750 | |
b40ef869 AV |
2751 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2752 | if (!copy) | |
2753 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2754 | |
2755 | /* We only care that *some* data at the address the user | |
2756 | * gave us is valid. Just in case, we'll zero | |
2757 | * the remainder of the page. | |
2758 | */ | |
2759 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2760 | size = TASK_SIZE - (unsigned long)data; | |
2761 | if (size > PAGE_SIZE) | |
2762 | size = PAGE_SIZE; | |
2763 | ||
b40ef869 | 2764 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2765 | if (!i) { |
b40ef869 AV |
2766 | kfree(copy); |
2767 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2768 | } |
2769 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2770 | memset(copy + i, 0, PAGE_SIZE - i); |
2771 | return copy; | |
1da177e4 LT |
2772 | } |
2773 | ||
b8850d1f | 2774 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2775 | { |
b8850d1f | 2776 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2777 | } |
2778 | ||
1da177e4 LT |
2779 | /* |
2780 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2781 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2782 | * | |
2783 | * data is a (void *) that can point to any structure up to | |
2784 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2785 | * information (or be NULL). | |
2786 | * | |
2787 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2788 | * When the flags word was introduced its top half was required | |
2789 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2790 | * Therefore, if this magic number is present, it carries no information | |
2791 | * and must be discarded. | |
2792 | */ | |
5e6123f3 | 2793 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2794 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2795 | { |
2d92ab3c | 2796 | struct path path; |
e462ec50 | 2797 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2798 | int retval = 0; |
1da177e4 LT |
2799 | |
2800 | /* Discard magic */ | |
2801 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2802 | flags &= ~MS_MGC_MSK; | |
2803 | ||
2804 | /* Basic sanity checks */ | |
1da177e4 LT |
2805 | if (data_page) |
2806 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2807 | ||
e462ec50 DH |
2808 | if (flags & MS_NOUSER) |
2809 | return -EINVAL; | |
2810 | ||
a27ab9f2 | 2811 | /* ... and get the mountpoint */ |
5e6123f3 | 2812 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2813 | if (retval) |
2814 | return retval; | |
2815 | ||
2816 | retval = security_sb_mount(dev_name, &path, | |
2817 | type_page, flags, data_page); | |
0d5cadb8 AV |
2818 | if (!retval && !may_mount()) |
2819 | retval = -EPERM; | |
e462ec50 | 2820 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 2821 | retval = -EPERM; |
a27ab9f2 TH |
2822 | if (retval) |
2823 | goto dput_out; | |
2824 | ||
613cbe3d AK |
2825 | /* Default to relatime unless overriden */ |
2826 | if (!(flags & MS_NOATIME)) | |
2827 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2828 | |
1da177e4 LT |
2829 | /* Separate the per-mountpoint flags */ |
2830 | if (flags & MS_NOSUID) | |
2831 | mnt_flags |= MNT_NOSUID; | |
2832 | if (flags & MS_NODEV) | |
2833 | mnt_flags |= MNT_NODEV; | |
2834 | if (flags & MS_NOEXEC) | |
2835 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2836 | if (flags & MS_NOATIME) |
2837 | mnt_flags |= MNT_NOATIME; | |
2838 | if (flags & MS_NODIRATIME) | |
2839 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2840 | if (flags & MS_STRICTATIME) |
2841 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 2842 | if (flags & MS_RDONLY) |
2e4b7fcd | 2843 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 2844 | |
ffbc6f0e EB |
2845 | /* The default atime for remount is preservation */ |
2846 | if ((flags & MS_REMOUNT) && | |
2847 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2848 | MS_STRICTATIME)) == 0)) { | |
2849 | mnt_flags &= ~MNT_ATIME_MASK; | |
2850 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2851 | } | |
2852 | ||
e462ec50 DH |
2853 | sb_flags = flags & (SB_RDONLY | |
2854 | SB_SYNCHRONOUS | | |
2855 | SB_MANDLOCK | | |
2856 | SB_DIRSYNC | | |
2857 | SB_SILENT | | |
917086ff | 2858 | SB_POSIXACL | |
d7ee9469 | 2859 | SB_LAZYTIME | |
917086ff | 2860 | SB_I_VERSION); |
1da177e4 | 2861 | |
1da177e4 | 2862 | if (flags & MS_REMOUNT) |
e462ec50 | 2863 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
2864 | data_page); |
2865 | else if (flags & MS_BIND) | |
2d92ab3c | 2866 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2867 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2868 | retval = do_change_type(&path, flags); |
1da177e4 | 2869 | else if (flags & MS_MOVE) |
2d92ab3c | 2870 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2871 | else |
e462ec50 | 2872 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
2873 | dev_name, data_page); |
2874 | dput_out: | |
2d92ab3c | 2875 | path_put(&path); |
1da177e4 LT |
2876 | return retval; |
2877 | } | |
2878 | ||
537f7ccb EB |
2879 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2880 | { | |
2881 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2882 | } | |
2883 | ||
2884 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2885 | { | |
2886 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2887 | } | |
2888 | ||
771b1371 EB |
2889 | static void free_mnt_ns(struct mnt_namespace *ns) |
2890 | { | |
6344c433 | 2891 | ns_free_inum(&ns->ns); |
537f7ccb | 2892 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2893 | put_user_ns(ns->user_ns); |
2894 | kfree(ns); | |
2895 | } | |
2896 | ||
8823c079 EB |
2897 | /* |
2898 | * Assign a sequence number so we can detect when we attempt to bind | |
2899 | * mount a reference to an older mount namespace into the current | |
2900 | * mount namespace, preventing reference counting loops. A 64bit | |
2901 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2902 | * is effectively never, so we can ignore the possibility. | |
2903 | */ | |
2904 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2905 | ||
771b1371 | 2906 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2907 | { |
2908 | struct mnt_namespace *new_ns; | |
537f7ccb | 2909 | struct ucounts *ucounts; |
98f842e6 | 2910 | int ret; |
cf8d2c11 | 2911 | |
537f7ccb EB |
2912 | ucounts = inc_mnt_namespaces(user_ns); |
2913 | if (!ucounts) | |
df75e774 | 2914 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2915 | |
cf8d2c11 | 2916 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2917 | if (!new_ns) { |
2918 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2919 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2920 | } |
6344c433 | 2921 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2922 | if (ret) { |
2923 | kfree(new_ns); | |
537f7ccb | 2924 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2925 | return ERR_PTR(ret); |
2926 | } | |
33c42940 | 2927 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2928 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2929 | atomic_set(&new_ns->count, 1); |
2930 | new_ns->root = NULL; | |
2931 | INIT_LIST_HEAD(&new_ns->list); | |
2932 | init_waitqueue_head(&new_ns->poll); | |
2933 | new_ns->event = 0; | |
771b1371 | 2934 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2935 | new_ns->ucounts = ucounts; |
d2921684 EB |
2936 | new_ns->mounts = 0; |
2937 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2938 | return new_ns; |
2939 | } | |
2940 | ||
0766f788 | 2941 | __latent_entropy |
9559f689 AV |
2942 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2943 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2944 | { |
6b3286ed | 2945 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2946 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2947 | struct mount *p, *q; |
9559f689 | 2948 | struct mount *old; |
cb338d06 | 2949 | struct mount *new; |
7a472ef4 | 2950 | int copy_flags; |
1da177e4 | 2951 | |
9559f689 AV |
2952 | BUG_ON(!ns); |
2953 | ||
2954 | if (likely(!(flags & CLONE_NEWNS))) { | |
2955 | get_mnt_ns(ns); | |
2956 | return ns; | |
2957 | } | |
2958 | ||
2959 | old = ns->root; | |
2960 | ||
771b1371 | 2961 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2962 | if (IS_ERR(new_ns)) |
2963 | return new_ns; | |
1da177e4 | 2964 | |
97216be0 | 2965 | namespace_lock(); |
1da177e4 | 2966 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2967 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2968 | if (user_ns != ns->user_ns) |
132c94e3 | 2969 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2970 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2971 | if (IS_ERR(new)) { |
328e6d90 | 2972 | namespace_unlock(); |
771b1371 | 2973 | free_mnt_ns(new_ns); |
be34d1a3 | 2974 | return ERR_CAST(new); |
1da177e4 | 2975 | } |
be08d6d2 | 2976 | new_ns->root = new; |
1a4eeaf2 | 2977 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2978 | |
2979 | /* | |
2980 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2981 | * as belonging to new namespace. We have already acquired a private | |
2982 | * fs_struct, so tsk->fs->lock is not needed. | |
2983 | */ | |
909b0a88 | 2984 | p = old; |
cb338d06 | 2985 | q = new; |
1da177e4 | 2986 | while (p) { |
143c8c91 | 2987 | q->mnt_ns = new_ns; |
d2921684 | 2988 | new_ns->mounts++; |
9559f689 AV |
2989 | if (new_fs) { |
2990 | if (&p->mnt == new_fs->root.mnt) { | |
2991 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2992 | rootmnt = &p->mnt; |
1da177e4 | 2993 | } |
9559f689 AV |
2994 | if (&p->mnt == new_fs->pwd.mnt) { |
2995 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2996 | pwdmnt = &p->mnt; |
1da177e4 | 2997 | } |
1da177e4 | 2998 | } |
909b0a88 AV |
2999 | p = next_mnt(p, old); |
3000 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3001 | if (!q) |
3002 | break; | |
3003 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
3004 | p = next_mnt(p, old); | |
1da177e4 | 3005 | } |
328e6d90 | 3006 | namespace_unlock(); |
1da177e4 | 3007 | |
1da177e4 | 3008 | if (rootmnt) |
f03c6599 | 3009 | mntput(rootmnt); |
1da177e4 | 3010 | if (pwdmnt) |
f03c6599 | 3011 | mntput(pwdmnt); |
1da177e4 | 3012 | |
741a2951 | 3013 | return new_ns; |
1da177e4 LT |
3014 | } |
3015 | ||
cf8d2c11 TM |
3016 | /** |
3017 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
3018 | * @mnt: pointer to the new root filesystem mountpoint | |
3019 | */ | |
1a4eeaf2 | 3020 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 3021 | { |
771b1371 | 3022 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 3023 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
3024 | struct mount *mnt = real_mount(m); |
3025 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 3026 | new_ns->root = mnt; |
d2921684 | 3027 | new_ns->mounts++; |
b1983cd8 | 3028 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 3029 | } else { |
1a4eeaf2 | 3030 | mntput(m); |
cf8d2c11 TM |
3031 | } |
3032 | return new_ns; | |
3033 | } | |
cf8d2c11 | 3034 | |
ea441d11 AV |
3035 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3036 | { | |
3037 | struct mnt_namespace *ns; | |
d31da0f0 | 3038 | struct super_block *s; |
ea441d11 AV |
3039 | struct path path; |
3040 | int err; | |
3041 | ||
3042 | ns = create_mnt_ns(mnt); | |
3043 | if (IS_ERR(ns)) | |
3044 | return ERR_CAST(ns); | |
3045 | ||
3046 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3047 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3048 | ||
3049 | put_mnt_ns(ns); | |
3050 | ||
3051 | if (err) | |
3052 | return ERR_PTR(err); | |
3053 | ||
3054 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3055 | s = path.mnt->mnt_sb; |
3056 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3057 | mntput(path.mnt); |
3058 | /* lock the sucker */ | |
d31da0f0 | 3059 | down_write(&s->s_umount); |
ea441d11 AV |
3060 | /* ... and return the root of (sub)tree on it */ |
3061 | return path.dentry; | |
3062 | } | |
3063 | EXPORT_SYMBOL(mount_subtree); | |
3064 | ||
312db1aa DB |
3065 | int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, |
3066 | unsigned long flags, void __user *data) | |
1da177e4 | 3067 | { |
eca6f534 VN |
3068 | int ret; |
3069 | char *kernel_type; | |
eca6f534 | 3070 | char *kernel_dev; |
b40ef869 | 3071 | void *options; |
1da177e4 | 3072 | |
b8850d1f TG |
3073 | kernel_type = copy_mount_string(type); |
3074 | ret = PTR_ERR(kernel_type); | |
3075 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3076 | goto out_type; |
1da177e4 | 3077 | |
b8850d1f TG |
3078 | kernel_dev = copy_mount_string(dev_name); |
3079 | ret = PTR_ERR(kernel_dev); | |
3080 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3081 | goto out_dev; |
1da177e4 | 3082 | |
b40ef869 AV |
3083 | options = copy_mount_options(data); |
3084 | ret = PTR_ERR(options); | |
3085 | if (IS_ERR(options)) | |
eca6f534 | 3086 | goto out_data; |
1da177e4 | 3087 | |
b40ef869 | 3088 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3089 | |
b40ef869 | 3090 | kfree(options); |
eca6f534 VN |
3091 | out_data: |
3092 | kfree(kernel_dev); | |
3093 | out_dev: | |
eca6f534 VN |
3094 | kfree(kernel_type); |
3095 | out_type: | |
3096 | return ret; | |
1da177e4 LT |
3097 | } |
3098 | ||
312db1aa DB |
3099 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3100 | char __user *, type, unsigned long, flags, void __user *, data) | |
3101 | { | |
3102 | return ksys_mount(dev_name, dir_name, type, flags, data); | |
3103 | } | |
3104 | ||
afac7cba AV |
3105 | /* |
3106 | * Return true if path is reachable from root | |
3107 | * | |
48a066e7 | 3108 | * namespace_sem or mount_lock is held |
afac7cba | 3109 | */ |
643822b4 | 3110 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3111 | const struct path *root) |
3112 | { | |
643822b4 | 3113 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3114 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3115 | mnt = mnt->mnt_parent; |
afac7cba | 3116 | } |
643822b4 | 3117 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3118 | } |
3119 | ||
640eb7e7 | 3120 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3121 | { |
25ab4c9b | 3122 | bool res; |
48a066e7 | 3123 | read_seqlock_excl(&mount_lock); |
643822b4 | 3124 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3125 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3126 | return res; |
3127 | } | |
3128 | EXPORT_SYMBOL(path_is_under); | |
3129 | ||
1da177e4 LT |
3130 | /* |
3131 | * pivot_root Semantics: | |
3132 | * Moves the root file system of the current process to the directory put_old, | |
3133 | * makes new_root as the new root file system of the current process, and sets | |
3134 | * root/cwd of all processes which had them on the current root to new_root. | |
3135 | * | |
3136 | * Restrictions: | |
3137 | * The new_root and put_old must be directories, and must not be on the | |
3138 | * same file system as the current process root. The put_old must be | |
3139 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3140 | * pointed to by put_old must yield the same directory as new_root. No other | |
3141 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3142 | * | |
4a0d11fa NB |
3143 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3144 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3145 | * in this situation. | |
3146 | * | |
1da177e4 LT |
3147 | * Notes: |
3148 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3149 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3150 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3151 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3152 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3153 | * first. | |
3154 | */ | |
3480b257 HC |
3155 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3156 | const char __user *, put_old) | |
1da177e4 | 3157 | { |
2d8f3038 | 3158 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3159 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3160 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3161 | int error; |
3162 | ||
9b40bc90 | 3163 | if (!may_mount()) |
1da177e4 LT |
3164 | return -EPERM; |
3165 | ||
2d8f3038 | 3166 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3167 | if (error) |
3168 | goto out0; | |
1da177e4 | 3169 | |
2d8f3038 | 3170 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3171 | if (error) |
3172 | goto out1; | |
3173 | ||
2d8f3038 | 3174 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3175 | if (error) |
3176 | goto out2; | |
1da177e4 | 3177 | |
f7ad3c6b | 3178 | get_fs_root(current->fs, &root); |
84d17192 AV |
3179 | old_mp = lock_mount(&old); |
3180 | error = PTR_ERR(old_mp); | |
3181 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3182 | goto out3; |
3183 | ||
1da177e4 | 3184 | error = -EINVAL; |
419148da AV |
3185 | new_mnt = real_mount(new.mnt); |
3186 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3187 | old_mnt = real_mount(old.mnt); |
3188 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3189 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3190 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3191 | goto out4; |
143c8c91 | 3192 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3193 | goto out4; |
5ff9d8a6 EB |
3194 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3195 | goto out4; | |
1da177e4 | 3196 | error = -ENOENT; |
f3da392e | 3197 | if (d_unlinked(new.dentry)) |
b12cea91 | 3198 | goto out4; |
1da177e4 | 3199 | error = -EBUSY; |
84d17192 | 3200 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3201 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3202 | error = -EINVAL; |
8c3ee42e | 3203 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3204 | goto out4; /* not a mountpoint */ |
676da58d | 3205 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3206 | goto out4; /* not attached */ |
84d17192 | 3207 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3208 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3209 | goto out4; /* not a mountpoint */ |
676da58d | 3210 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3211 | goto out4; /* not attached */ |
4ac91378 | 3212 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3213 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3214 | goto out4; |
0d082601 EB |
3215 | /* make certain new is below the root */ |
3216 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3217 | goto out4; | |
84d17192 | 3218 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3219 | lock_mount_hash(); |
419148da AV |
3220 | detach_mnt(new_mnt, &parent_path); |
3221 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3222 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3223 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3224 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3225 | } | |
4ac91378 | 3226 | /* mount old root on put_old */ |
84d17192 | 3227 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3228 | /* mount new_root on / */ |
84d17192 | 3229 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3230 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3231 | /* A moved mount should not expire automatically */ |
3232 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3233 | put_mountpoint(root_mp); |
719ea2fb | 3234 | unlock_mount_hash(); |
2d8f3038 | 3235 | chroot_fs_refs(&root, &new); |
1da177e4 | 3236 | error = 0; |
b12cea91 | 3237 | out4: |
84d17192 | 3238 | unlock_mount(old_mp); |
b12cea91 AV |
3239 | if (!error) { |
3240 | path_put(&root_parent); | |
3241 | path_put(&parent_path); | |
3242 | } | |
3243 | out3: | |
8c3ee42e | 3244 | path_put(&root); |
b12cea91 | 3245 | out2: |
2d8f3038 | 3246 | path_put(&old); |
1da177e4 | 3247 | out1: |
2d8f3038 | 3248 | path_put(&new); |
1da177e4 | 3249 | out0: |
1da177e4 | 3250 | return error; |
1da177e4 LT |
3251 | } |
3252 | ||
3253 | static void __init init_mount_tree(void) | |
3254 | { | |
3255 | struct vfsmount *mnt; | |
6b3286ed | 3256 | struct mnt_namespace *ns; |
ac748a09 | 3257 | struct path root; |
0c55cfc4 | 3258 | struct file_system_type *type; |
1da177e4 | 3259 | |
0c55cfc4 EB |
3260 | type = get_fs_type("rootfs"); |
3261 | if (!type) | |
3262 | panic("Can't find rootfs type"); | |
3263 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3264 | put_filesystem(type); | |
1da177e4 LT |
3265 | if (IS_ERR(mnt)) |
3266 | panic("Can't create rootfs"); | |
b3e19d92 | 3267 | |
3b22edc5 TM |
3268 | ns = create_mnt_ns(mnt); |
3269 | if (IS_ERR(ns)) | |
1da177e4 | 3270 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3271 | |
3272 | init_task.nsproxy->mnt_ns = ns; | |
3273 | get_mnt_ns(ns); | |
3274 | ||
be08d6d2 AV |
3275 | root.mnt = mnt; |
3276 | root.dentry = mnt->mnt_root; | |
da362b09 | 3277 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3278 | |
3279 | set_fs_pwd(current->fs, &root); | |
3280 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3281 | } |
3282 | ||
74bf17cf | 3283 | void __init mnt_init(void) |
1da177e4 | 3284 | { |
15a67dd8 | 3285 | int err; |
1da177e4 | 3286 | |
7d6fec45 | 3287 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3288 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3289 | |
0818bf27 | 3290 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3291 | sizeof(struct hlist_head), |
0818bf27 | 3292 | mhash_entries, 19, |
3d375d78 | 3293 | HASH_ZERO, |
0818bf27 AV |
3294 | &m_hash_shift, &m_hash_mask, 0, 0); |
3295 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3296 | sizeof(struct hlist_head), | |
3297 | mphash_entries, 19, | |
3d375d78 | 3298 | HASH_ZERO, |
0818bf27 | 3299 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3300 | |
84d17192 | 3301 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3302 | panic("Failed to allocate mount hash table\n"); |
3303 | ||
4b93dc9b TH |
3304 | kernfs_init(); |
3305 | ||
15a67dd8 RD |
3306 | err = sysfs_init(); |
3307 | if (err) | |
3308 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3309 | __func__, err); |
00d26666 GKH |
3310 | fs_kobj = kobject_create_and_add("fs", NULL); |
3311 | if (!fs_kobj) | |
8e24eea7 | 3312 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3313 | init_rootfs(); |
3314 | init_mount_tree(); | |
3315 | } | |
3316 | ||
616511d0 | 3317 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3318 | { |
d498b25a | 3319 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3320 | return; |
7b00ed6f | 3321 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3322 | free_mnt_ns(ns); |
1da177e4 | 3323 | } |
9d412a43 AV |
3324 | |
3325 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3326 | { | |
423e0ab0 | 3327 | struct vfsmount *mnt; |
e462ec50 | 3328 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); |
423e0ab0 TC |
3329 | if (!IS_ERR(mnt)) { |
3330 | /* | |
3331 | * it is a longterm mount, don't release mnt until | |
3332 | * we unmount before file sys is unregistered | |
3333 | */ | |
f7a99c5b | 3334 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3335 | } |
3336 | return mnt; | |
9d412a43 AV |
3337 | } |
3338 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3339 | |
3340 | void kern_unmount(struct vfsmount *mnt) | |
3341 | { | |
3342 | /* release long term mount so mount point can be released */ | |
3343 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3344 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3345 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3346 | mntput(mnt); |
3347 | } | |
3348 | } | |
3349 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3350 | |
3351 | bool our_mnt(struct vfsmount *mnt) | |
3352 | { | |
143c8c91 | 3353 | return check_mnt(real_mount(mnt)); |
02125a82 | 3354 | } |
8823c079 | 3355 | |
3151527e EB |
3356 | bool current_chrooted(void) |
3357 | { | |
3358 | /* Does the current process have a non-standard root */ | |
3359 | struct path ns_root; | |
3360 | struct path fs_root; | |
3361 | bool chrooted; | |
3362 | ||
3363 | /* Find the namespace root */ | |
3364 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3365 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3366 | path_get(&ns_root); | |
3367 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3368 | ; | |
3369 | ||
3370 | get_fs_root(current->fs, &fs_root); | |
3371 | ||
3372 | chrooted = !path_equal(&fs_root, &ns_root); | |
3373 | ||
3374 | path_put(&fs_root); | |
3375 | path_put(&ns_root); | |
3376 | ||
3377 | return chrooted; | |
3378 | } | |
3379 | ||
8654df4e EB |
3380 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3381 | int *new_mnt_flags) | |
87a8ebd6 | 3382 | { |
8c6cf9cc | 3383 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3384 | struct mount *mnt; |
e51db735 | 3385 | bool visible = false; |
87a8ebd6 | 3386 | |
44bb4385 | 3387 | down_read(&namespace_sem); |
87a8ebd6 | 3388 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3389 | struct mount *child; |
77b1a97d EB |
3390 | int mnt_flags; |
3391 | ||
8654df4e | 3392 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3393 | continue; |
3394 | ||
7e96c1b0 EB |
3395 | /* This mount is not fully visible if it's root directory |
3396 | * is not the root directory of the filesystem. | |
3397 | */ | |
3398 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3399 | continue; | |
3400 | ||
a1935c17 | 3401 | /* A local view of the mount flags */ |
77b1a97d | 3402 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3403 | |
695e9df0 | 3404 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3405 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3406 | mnt_flags |= MNT_LOCK_READONLY; |
3407 | ||
8c6cf9cc EB |
3408 | /* Verify the mount flags are equal to or more permissive |
3409 | * than the proposed new mount. | |
3410 | */ | |
77b1a97d | 3411 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3412 | !(new_flags & MNT_READONLY)) |
3413 | continue; | |
77b1a97d EB |
3414 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3415 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3416 | continue; |
3417 | ||
ceeb0e5d EB |
3418 | /* This mount is not fully visible if there are any |
3419 | * locked child mounts that cover anything except for | |
3420 | * empty directories. | |
e51db735 EB |
3421 | */ |
3422 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3423 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3424 | /* Only worry about locked mounts */ |
d71ed6c9 | 3425 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3426 | continue; |
7236c85e EB |
3427 | /* Is the directory permanetly empty? */ |
3428 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3429 | goto next; |
87a8ebd6 | 3430 | } |
8c6cf9cc | 3431 | /* Preserve the locked attributes */ |
77b1a97d | 3432 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3433 | MNT_LOCK_ATIME); |
e51db735 EB |
3434 | visible = true; |
3435 | goto found; | |
3436 | next: ; | |
87a8ebd6 | 3437 | } |
e51db735 | 3438 | found: |
44bb4385 | 3439 | up_read(&namespace_sem); |
e51db735 | 3440 | return visible; |
87a8ebd6 EB |
3441 | } |
3442 | ||
8654df4e EB |
3443 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3444 | { | |
a1935c17 | 3445 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3446 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3447 | unsigned long s_iflags; | |
3448 | ||
3449 | if (ns->user_ns == &init_user_ns) | |
3450 | return false; | |
3451 | ||
3452 | /* Can this filesystem be too revealing? */ | |
3453 | s_iflags = mnt->mnt_sb->s_iflags; | |
3454 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3455 | return false; | |
3456 | ||
a1935c17 EB |
3457 | if ((s_iflags & required_iflags) != required_iflags) { |
3458 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3459 | required_iflags); | |
3460 | return true; | |
3461 | } | |
3462 | ||
8654df4e EB |
3463 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3464 | } | |
3465 | ||
380cf5ba AL |
3466 | bool mnt_may_suid(struct vfsmount *mnt) |
3467 | { | |
3468 | /* | |
3469 | * Foreign mounts (accessed via fchdir or through /proc | |
3470 | * symlinks) are always treated as if they are nosuid. This | |
3471 | * prevents namespaces from trusting potentially unsafe | |
3472 | * suid/sgid bits, file caps, or security labels that originate | |
3473 | * in other namespaces. | |
3474 | */ | |
3475 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3476 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3477 | } | |
3478 | ||
64964528 | 3479 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3480 | { |
58be2825 | 3481 | struct ns_common *ns = NULL; |
8823c079 EB |
3482 | struct nsproxy *nsproxy; |
3483 | ||
728dba3a EB |
3484 | task_lock(task); |
3485 | nsproxy = task->nsproxy; | |
8823c079 | 3486 | if (nsproxy) { |
58be2825 AV |
3487 | ns = &nsproxy->mnt_ns->ns; |
3488 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3489 | } |
728dba3a | 3490 | task_unlock(task); |
8823c079 EB |
3491 | |
3492 | return ns; | |
3493 | } | |
3494 | ||
64964528 | 3495 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3496 | { |
58be2825 | 3497 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3498 | } |
3499 | ||
64964528 | 3500 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3501 | { |
3502 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3503 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3504 | struct path root; |
4f757f3c | 3505 | int err; |
8823c079 | 3506 | |
0c55cfc4 | 3507 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3508 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3509 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3510 | return -EPERM; |
8823c079 EB |
3511 | |
3512 | if (fs->users != 1) | |
3513 | return -EINVAL; | |
3514 | ||
3515 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3516 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3517 | nsproxy->mnt_ns = mnt_ns; |
3518 | ||
3519 | /* Find the root */ | |
4f757f3c AV |
3520 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3521 | "/", LOOKUP_DOWN, &root); | |
3522 | if (err) { | |
3523 | /* revert to old namespace */ | |
3524 | nsproxy->mnt_ns = old_mnt_ns; | |
3525 | put_mnt_ns(mnt_ns); | |
3526 | return err; | |
3527 | } | |
8823c079 | 3528 | |
4068367c AV |
3529 | put_mnt_ns(old_mnt_ns); |
3530 | ||
8823c079 EB |
3531 | /* Update the pwd and root */ |
3532 | set_fs_pwd(fs, &root); | |
3533 | set_fs_root(fs, &root); | |
3534 | ||
3535 | path_put(&root); | |
3536 | return 0; | |
3537 | } | |
3538 | ||
bcac25a5 AV |
3539 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3540 | { | |
3541 | return to_mnt_ns(ns)->user_ns; | |
3542 | } | |
3543 | ||
8823c079 EB |
3544 | const struct proc_ns_operations mntns_operations = { |
3545 | .name = "mnt", | |
3546 | .type = CLONE_NEWNS, | |
3547 | .get = mntns_get, | |
3548 | .put = mntns_put, | |
3549 | .install = mntns_install, | |
bcac25a5 | 3550 | .owner = mntns_owner, |
8823c079 | 3551 | }; |