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 | 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 | 285 | { |
286 | #ifdef CONFIG_SMP | |
68e8a9fe | 287 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 288 | #else |
68e8a9fe | 289 | mnt->mnt_writers++; |
d3ef3d73 | 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 | 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 | 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 | 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(); | |
1e75529e | 356 | while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) |
d3ef3d73 | 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 | 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 | 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 | 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 | 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 | 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); | |
471 | if (upperdentry && file_inode(file) == d_inode(upperdentry)) | |
472 | return 0; | |
473 | ||
474 | /* Lower layer: can't write to real file, sorry... */ | |
475 | return -EPERM; | |
476 | } | |
477 | ||
478 | /** | |
479 | * mnt_want_write_file - get write access to a file's mount | |
480 | * @file: the file who's mount on which to take a write | |
481 | * | |
482 | * This is like mnt_want_write, but it takes a file and can | |
483 | * do some optimisations if the file is open for write already | |
484 | * | |
485 | * Mostly called by filesystems from their ioctl operation before performing | |
486 | * modification. On overlayfs this needs to check if the file is on a read-only | |
487 | * lower layer and deny access in that case. | |
488 | */ | |
489 | int mnt_want_write_file(struct file *file) | |
490 | { | |
491 | int ret; | |
492 | ||
493 | ret = may_write_real(file); | |
494 | if (!ret) { | |
495 | sb_start_write(file_inode(file)->i_sb); | |
496 | ret = __mnt_want_write_file(file); | |
497 | if (ret) | |
498 | sb_end_write(file_inode(file)->i_sb); | |
499 | } | |
500 | return ret; | |
501 | } | |
96029c4e | 502 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
503 | ||
8366025e | 504 | /** |
eb04c282 | 505 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
506 | * @mnt: the mount on which to give up write access |
507 | * | |
508 | * Tells the low-level filesystem that we are done | |
509 | * performing writes to it. Must be matched with | |
eb04c282 | 510 | * __mnt_want_write() call above. |
8366025e | 511 | */ |
eb04c282 | 512 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 513 | { |
d3ef3d73 | 514 | preempt_disable(); |
83adc753 | 515 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 516 | preempt_enable(); |
8366025e | 517 | } |
eb04c282 JK |
518 | |
519 | /** | |
520 | * mnt_drop_write - give up write access to a mount | |
521 | * @mnt: the mount on which to give up write access | |
522 | * | |
523 | * Tells the low-level filesystem that we are done performing writes to it and | |
524 | * also allows filesystem to be frozen again. Must be matched with | |
525 | * mnt_want_write() call above. | |
526 | */ | |
527 | void mnt_drop_write(struct vfsmount *mnt) | |
528 | { | |
529 | __mnt_drop_write(mnt); | |
530 | sb_end_write(mnt->mnt_sb); | |
531 | } | |
8366025e DH |
532 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
533 | ||
eb04c282 JK |
534 | void __mnt_drop_write_file(struct file *file) |
535 | { | |
536 | __mnt_drop_write(file->f_path.mnt); | |
537 | } | |
538 | ||
7c6893e3 | 539 | void mnt_drop_write_file_path(struct file *file) |
2a79f17e AV |
540 | { |
541 | mnt_drop_write(file->f_path.mnt); | |
542 | } | |
7c6893e3 MS |
543 | |
544 | void mnt_drop_write_file(struct file *file) | |
545 | { | |
546 | __mnt_drop_write(file->f_path.mnt); | |
547 | sb_end_write(file_inode(file)->i_sb); | |
548 | } | |
2a79f17e AV |
549 | EXPORT_SYMBOL(mnt_drop_write_file); |
550 | ||
83adc753 | 551 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 552 | { |
3d733633 DH |
553 | int ret = 0; |
554 | ||
719ea2fb | 555 | lock_mount_hash(); |
83adc753 | 556 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 557 | /* |
d3ef3d73 | 558 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
559 | * should be visible before we do. | |
3d733633 | 560 | */ |
d3ef3d73 | 561 | smp_mb(); |
562 | ||
3d733633 | 563 | /* |
d3ef3d73 | 564 | * With writers on hold, if this value is zero, then there are |
565 | * definitely no active writers (although held writers may subsequently | |
566 | * increment the count, they'll have to wait, and decrement it after | |
567 | * seeing MNT_READONLY). | |
568 | * | |
569 | * It is OK to have counter incremented on one CPU and decremented on | |
570 | * another: the sum will add up correctly. The danger would be when we | |
571 | * sum up each counter, if we read a counter before it is incremented, | |
572 | * but then read another CPU's count which it has been subsequently | |
573 | * decremented from -- we would see more decrements than we should. | |
574 | * MNT_WRITE_HOLD protects against this scenario, because | |
575 | * mnt_want_write first increments count, then smp_mb, then spins on | |
576 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
577 | * we're counting up here. | |
3d733633 | 578 | */ |
c6653a83 | 579 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 580 | ret = -EBUSY; |
581 | else | |
83adc753 | 582 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 583 | /* |
584 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
585 | * that become unheld will see MNT_READONLY. | |
586 | */ | |
587 | smp_wmb(); | |
83adc753 | 588 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 589 | unlock_mount_hash(); |
3d733633 | 590 | return ret; |
8366025e | 591 | } |
8366025e | 592 | |
83adc753 | 593 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 594 | { |
719ea2fb | 595 | lock_mount_hash(); |
83adc753 | 596 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 597 | unlock_mount_hash(); |
2e4b7fcd DH |
598 | } |
599 | ||
4ed5e82f MS |
600 | int sb_prepare_remount_readonly(struct super_block *sb) |
601 | { | |
602 | struct mount *mnt; | |
603 | int err = 0; | |
604 | ||
8e8b8796 MS |
605 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
606 | if (atomic_long_read(&sb->s_remove_count)) | |
607 | return -EBUSY; | |
608 | ||
719ea2fb | 609 | lock_mount_hash(); |
4ed5e82f MS |
610 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
611 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
612 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
613 | smp_mb(); | |
614 | if (mnt_get_writers(mnt) > 0) { | |
615 | err = -EBUSY; | |
616 | break; | |
617 | } | |
618 | } | |
619 | } | |
8e8b8796 MS |
620 | if (!err && atomic_long_read(&sb->s_remove_count)) |
621 | err = -EBUSY; | |
622 | ||
4ed5e82f MS |
623 | if (!err) { |
624 | sb->s_readonly_remount = 1; | |
625 | smp_wmb(); | |
626 | } | |
627 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
628 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
629 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
630 | } | |
719ea2fb | 631 | unlock_mount_hash(); |
4ed5e82f MS |
632 | |
633 | return err; | |
634 | } | |
635 | ||
b105e270 | 636 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 637 | { |
fcc139ae | 638 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 639 | #ifdef CONFIG_SMP |
68e8a9fe | 640 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 641 | #endif |
b105e270 | 642 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
643 | } |
644 | ||
8ffcb32e DH |
645 | static void delayed_free_vfsmnt(struct rcu_head *head) |
646 | { | |
647 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
648 | } | |
649 | ||
48a066e7 | 650 | /* call under rcu_read_lock */ |
294d71ff | 651 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
652 | { |
653 | struct mount *mnt; | |
654 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 655 | return 1; |
48a066e7 | 656 | if (bastard == NULL) |
294d71ff | 657 | return 0; |
48a066e7 AV |
658 | mnt = real_mount(bastard); |
659 | mnt_add_count(mnt, 1); | |
660 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 661 | return 0; |
48a066e7 AV |
662 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
663 | mnt_add_count(mnt, -1); | |
294d71ff AV |
664 | return 1; |
665 | } | |
666 | return -1; | |
667 | } | |
668 | ||
669 | /* call under rcu_read_lock */ | |
670 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
671 | { | |
672 | int res = __legitimize_mnt(bastard, seq); | |
673 | if (likely(!res)) | |
674 | return true; | |
675 | if (unlikely(res < 0)) { | |
676 | rcu_read_unlock(); | |
677 | mntput(bastard); | |
678 | rcu_read_lock(); | |
48a066e7 | 679 | } |
48a066e7 AV |
680 | return false; |
681 | } | |
682 | ||
1da177e4 | 683 | /* |
474279dc | 684 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 685 | * call under rcu_read_lock() |
1da177e4 | 686 | */ |
474279dc | 687 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 688 | { |
38129a13 | 689 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
690 | struct mount *p; |
691 | ||
38129a13 | 692 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
693 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
694 | return p; | |
695 | return NULL; | |
696 | } | |
697 | ||
a05964f3 | 698 | /* |
f015f126 DH |
699 | * lookup_mnt - Return the first child mount mounted at path |
700 | * | |
701 | * "First" means first mounted chronologically. If you create the | |
702 | * following mounts: | |
703 | * | |
704 | * mount /dev/sda1 /mnt | |
705 | * mount /dev/sda2 /mnt | |
706 | * mount /dev/sda3 /mnt | |
707 | * | |
708 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
709 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
710 | * /dev/sda2, then /dev/sda3, then NULL. | |
711 | * | |
712 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 713 | */ |
ca71cf71 | 714 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 715 | { |
c7105365 | 716 | struct mount *child_mnt; |
48a066e7 AV |
717 | struct vfsmount *m; |
718 | unsigned seq; | |
99b7db7b | 719 | |
48a066e7 AV |
720 | rcu_read_lock(); |
721 | do { | |
722 | seq = read_seqbegin(&mount_lock); | |
723 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
724 | m = child_mnt ? &child_mnt->mnt : NULL; | |
725 | } while (!legitimize_mnt(m, seq)); | |
726 | rcu_read_unlock(); | |
727 | return m; | |
a05964f3 RP |
728 | } |
729 | ||
7af1364f EB |
730 | /* |
731 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
732 | * current mount namespace. | |
733 | * | |
734 | * The common case is dentries are not mountpoints at all and that | |
735 | * test is handled inline. For the slow case when we are actually | |
736 | * dealing with a mountpoint of some kind, walk through all of the | |
737 | * mounts in the current mount namespace and test to see if the dentry | |
738 | * is a mountpoint. | |
739 | * | |
740 | * The mount_hashtable is not usable in the context because we | |
741 | * need to identify all mounts that may be in the current mount | |
742 | * namespace not just a mount that happens to have some specified | |
743 | * parent mount. | |
744 | */ | |
745 | bool __is_local_mountpoint(struct dentry *dentry) | |
746 | { | |
747 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
748 | struct mount *mnt; | |
749 | bool is_covered = false; | |
750 | ||
751 | if (!d_mountpoint(dentry)) | |
752 | goto out; | |
753 | ||
754 | down_read(&namespace_sem); | |
755 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
756 | is_covered = (mnt->mnt_mountpoint == dentry); | |
757 | if (is_covered) | |
758 | break; | |
759 | } | |
760 | up_read(&namespace_sem); | |
761 | out: | |
762 | return is_covered; | |
763 | } | |
764 | ||
e2dfa935 | 765 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 766 | { |
0818bf27 | 767 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
768 | struct mountpoint *mp; |
769 | ||
0818bf27 | 770 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
771 | if (mp->m_dentry == dentry) { |
772 | /* might be worth a WARN_ON() */ | |
773 | if (d_unlinked(dentry)) | |
774 | return ERR_PTR(-ENOENT); | |
775 | mp->m_count++; | |
776 | return mp; | |
777 | } | |
778 | } | |
e2dfa935 EB |
779 | return NULL; |
780 | } | |
781 | ||
3895dbf8 | 782 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 783 | { |
3895dbf8 | 784 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 785 | int ret; |
84d17192 | 786 | |
3895dbf8 EB |
787 | if (d_mountpoint(dentry)) { |
788 | mountpoint: | |
789 | read_seqlock_excl(&mount_lock); | |
790 | mp = lookup_mountpoint(dentry); | |
791 | read_sequnlock_excl(&mount_lock); | |
792 | if (mp) | |
793 | goto done; | |
794 | } | |
795 | ||
796 | if (!new) | |
797 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
798 | if (!new) | |
84d17192 AV |
799 | return ERR_PTR(-ENOMEM); |
800 | ||
3895dbf8 EB |
801 | |
802 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 803 | ret = d_set_mounted(dentry); |
eed81007 | 804 | |
3895dbf8 EB |
805 | /* Someone else set d_mounted? */ |
806 | if (ret == -EBUSY) | |
807 | goto mountpoint; | |
808 | ||
809 | /* The dentry is not available as a mountpoint? */ | |
810 | mp = ERR_PTR(ret); | |
811 | if (ret) | |
812 | goto done; | |
813 | ||
814 | /* Add the new mountpoint to the hash table */ | |
815 | read_seqlock_excl(&mount_lock); | |
816 | new->m_dentry = dentry; | |
817 | new->m_count = 1; | |
818 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
819 | INIT_HLIST_HEAD(&new->m_list); | |
820 | read_sequnlock_excl(&mount_lock); | |
821 | ||
822 | mp = new; | |
823 | new = NULL; | |
824 | done: | |
825 | kfree(new); | |
84d17192 AV |
826 | return mp; |
827 | } | |
828 | ||
829 | static void put_mountpoint(struct mountpoint *mp) | |
830 | { | |
831 | if (!--mp->m_count) { | |
832 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 833 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
834 | spin_lock(&dentry->d_lock); |
835 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
836 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 837 | hlist_del(&mp->m_hash); |
84d17192 AV |
838 | kfree(mp); |
839 | } | |
840 | } | |
841 | ||
143c8c91 | 842 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 843 | { |
6b3286ed | 844 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
845 | } |
846 | ||
99b7db7b NP |
847 | /* |
848 | * vfsmount lock must be held for write | |
849 | */ | |
6b3286ed | 850 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
851 | { |
852 | if (ns) { | |
853 | ns->event = ++event; | |
854 | wake_up_interruptible(&ns->poll); | |
855 | } | |
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 && ns->event != event) { | |
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 | */ | |
7bdb11de | 872 | static void unhash_mnt(struct mount *mnt) |
419148da | 873 | { |
0714a533 | 874 | mnt->mnt_parent = mnt; |
a73324da | 875 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 876 | list_del_init(&mnt->mnt_child); |
38129a13 | 877 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 878 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
879 | put_mountpoint(mnt->mnt_mp); |
880 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
881 | } |
882 | ||
7bdb11de EB |
883 | /* |
884 | * vfsmount lock must be held for write | |
885 | */ | |
886 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
887 | { | |
888 | old_path->dentry = mnt->mnt_mountpoint; | |
889 | old_path->mnt = &mnt->mnt_parent->mnt; | |
890 | unhash_mnt(mnt); | |
891 | } | |
892 | ||
6a46c573 EB |
893 | /* |
894 | * vfsmount lock must be held for write | |
895 | */ | |
896 | static void umount_mnt(struct mount *mnt) | |
897 | { | |
898 | /* old mountpoint will be dropped when we can do that */ | |
899 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
900 | unhash_mnt(mnt); | |
901 | } | |
902 | ||
99b7db7b NP |
903 | /* |
904 | * vfsmount lock must be held for write | |
905 | */ | |
84d17192 AV |
906 | void mnt_set_mountpoint(struct mount *mnt, |
907 | struct mountpoint *mp, | |
44d964d6 | 908 | struct mount *child_mnt) |
b90fa9ae | 909 | { |
84d17192 | 910 | mp->m_count++; |
3a2393d7 | 911 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 912 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 913 | child_mnt->mnt_parent = mnt; |
84d17192 | 914 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 915 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
916 | } |
917 | ||
1064f874 EB |
918 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
919 | { | |
920 | hlist_add_head_rcu(&mnt->mnt_hash, | |
921 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
922 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
923 | } | |
924 | ||
99b7db7b NP |
925 | /* |
926 | * vfsmount lock must be held for write | |
927 | */ | |
84d17192 AV |
928 | static void attach_mnt(struct mount *mnt, |
929 | struct mount *parent, | |
930 | struct mountpoint *mp) | |
1da177e4 | 931 | { |
84d17192 | 932 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 933 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
934 | } |
935 | ||
1064f874 | 936 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 937 | { |
1064f874 EB |
938 | struct mountpoint *old_mp = mnt->mnt_mp; |
939 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
940 | struct mount *old_parent = mnt->mnt_parent; | |
941 | ||
942 | list_del_init(&mnt->mnt_child); | |
943 | hlist_del_init(&mnt->mnt_mp_list); | |
944 | hlist_del_init_rcu(&mnt->mnt_hash); | |
945 | ||
946 | attach_mnt(mnt, parent, mp); | |
947 | ||
948 | put_mountpoint(old_mp); | |
949 | ||
950 | /* | |
951 | * Safely avoid even the suggestion this code might sleep or | |
952 | * lock the mount hash by taking advantage of the knowledge that | |
953 | * mnt_change_mountpoint will not release the final reference | |
954 | * to a mountpoint. | |
955 | * | |
956 | * During mounting, the mount passed in as the parent mount will | |
957 | * continue to use the old mountpoint and during unmounting, the | |
958 | * old mountpoint will continue to exist until namespace_unlock, | |
959 | * which happens well after mnt_change_mountpoint. | |
960 | */ | |
961 | spin_lock(&old_mountpoint->d_lock); | |
962 | old_mountpoint->d_lockref.count--; | |
963 | spin_unlock(&old_mountpoint->d_lock); | |
964 | ||
965 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
966 | } |
967 | ||
b90fa9ae | 968 | /* |
99b7db7b | 969 | * vfsmount lock must be held for write |
b90fa9ae | 970 | */ |
1064f874 | 971 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 972 | { |
0714a533 | 973 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 974 | struct mount *m; |
b90fa9ae | 975 | LIST_HEAD(head); |
143c8c91 | 976 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 977 | |
0714a533 | 978 | BUG_ON(parent == mnt); |
b90fa9ae | 979 | |
1a4eeaf2 | 980 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 981 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 982 | m->mnt_ns = n; |
f03c6599 | 983 | |
b90fa9ae RP |
984 | list_splice(&head, n->list.prev); |
985 | ||
d2921684 EB |
986 | n->mounts += n->pending_mounts; |
987 | n->pending_mounts = 0; | |
988 | ||
1064f874 | 989 | __attach_mnt(mnt, parent); |
6b3286ed | 990 | touch_mnt_namespace(n); |
1da177e4 LT |
991 | } |
992 | ||
909b0a88 | 993 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 994 | { |
6b41d536 AV |
995 | struct list_head *next = p->mnt_mounts.next; |
996 | if (next == &p->mnt_mounts) { | |
1da177e4 | 997 | while (1) { |
909b0a88 | 998 | if (p == root) |
1da177e4 | 999 | return NULL; |
6b41d536 AV |
1000 | next = p->mnt_child.next; |
1001 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 1002 | break; |
0714a533 | 1003 | p = p->mnt_parent; |
1da177e4 LT |
1004 | } |
1005 | } | |
6b41d536 | 1006 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
1007 | } |
1008 | ||
315fc83e | 1009 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 1010 | { |
6b41d536 AV |
1011 | struct list_head *prev = p->mnt_mounts.prev; |
1012 | while (prev != &p->mnt_mounts) { | |
1013 | p = list_entry(prev, struct mount, mnt_child); | |
1014 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
1015 | } |
1016 | return p; | |
1017 | } | |
1018 | ||
9d412a43 AV |
1019 | struct vfsmount * |
1020 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
1021 | { | |
b105e270 | 1022 | struct mount *mnt; |
9d412a43 AV |
1023 | struct dentry *root; |
1024 | ||
1025 | if (!type) | |
1026 | return ERR_PTR(-ENODEV); | |
1027 | ||
1028 | mnt = alloc_vfsmnt(name); | |
1029 | if (!mnt) | |
1030 | return ERR_PTR(-ENOMEM); | |
1031 | ||
e462ec50 | 1032 | if (flags & SB_KERNMOUNT) |
b105e270 | 1033 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
1034 | |
1035 | root = mount_fs(type, flags, name, data); | |
1036 | if (IS_ERR(root)) { | |
8ffcb32e | 1037 | mnt_free_id(mnt); |
9d412a43 AV |
1038 | free_vfsmnt(mnt); |
1039 | return ERR_CAST(root); | |
1040 | } | |
1041 | ||
b105e270 AV |
1042 | mnt->mnt.mnt_root = root; |
1043 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 1044 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1045 | mnt->mnt_parent = mnt; |
719ea2fb | 1046 | lock_mount_hash(); |
39f7c4db | 1047 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 1048 | unlock_mount_hash(); |
b105e270 | 1049 | return &mnt->mnt; |
9d412a43 AV |
1050 | } |
1051 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1052 | ||
93faccbb EB |
1053 | struct vfsmount * |
1054 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1055 | const char *name, void *data) | |
1056 | { | |
1057 | /* Until it is worked out how to pass the user namespace | |
1058 | * through from the parent mount to the submount don't support | |
1059 | * unprivileged mounts with submounts. | |
1060 | */ | |
1061 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1062 | return ERR_PTR(-EPERM); | |
1063 | ||
e462ec50 | 1064 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1065 | } |
1066 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1067 | ||
87129cc0 | 1068 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1069 | int flag) |
1da177e4 | 1070 | { |
87129cc0 | 1071 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1072 | struct mount *mnt; |
1073 | int err; | |
1da177e4 | 1074 | |
be34d1a3 DH |
1075 | mnt = alloc_vfsmnt(old->mnt_devname); |
1076 | if (!mnt) | |
1077 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1078 | |
7a472ef4 | 1079 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1080 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1081 | else | |
1082 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1083 | |
be34d1a3 DH |
1084 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1085 | err = mnt_alloc_group_id(mnt); | |
1086 | if (err) | |
1087 | goto out_free; | |
1da177e4 | 1088 | } |
be34d1a3 | 1089 | |
f2ebb3a9 | 1090 | mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); |
132c94e3 | 1091 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1092 | if (flag & CL_UNPRIVILEGED) { |
1093 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1094 | ||
1095 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1096 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1097 | ||
1098 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1099 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1100 | ||
1101 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1102 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1103 | ||
1104 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1105 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1106 | } | |
132c94e3 | 1107 | |
5ff9d8a6 | 1108 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1109 | if ((flag & CL_UNPRIVILEGED) && |
1110 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1111 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1112 | ||
be34d1a3 DH |
1113 | atomic_inc(&sb->s_active); |
1114 | mnt->mnt.mnt_sb = sb; | |
1115 | mnt->mnt.mnt_root = dget(root); | |
1116 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1117 | mnt->mnt_parent = mnt; | |
719ea2fb | 1118 | lock_mount_hash(); |
be34d1a3 | 1119 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1120 | unlock_mount_hash(); |
be34d1a3 | 1121 | |
7a472ef4 EB |
1122 | if ((flag & CL_SLAVE) || |
1123 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1124 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1125 | mnt->mnt_master = old; | |
1126 | CLEAR_MNT_SHARED(mnt); | |
1127 | } else if (!(flag & CL_PRIVATE)) { | |
1128 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1129 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1130 | if (IS_MNT_SLAVE(old)) | |
1131 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1132 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1133 | } else { |
1134 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1135 | } |
1136 | if (flag & CL_MAKE_SHARED) | |
1137 | set_mnt_shared(mnt); | |
1138 | ||
1139 | /* stick the duplicate mount on the same expiry list | |
1140 | * as the original if that was on one */ | |
1141 | if (flag & CL_EXPIRE) { | |
1142 | if (!list_empty(&old->mnt_expire)) | |
1143 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1144 | } | |
1145 | ||
cb338d06 | 1146 | return mnt; |
719f5d7f MS |
1147 | |
1148 | out_free: | |
8ffcb32e | 1149 | mnt_free_id(mnt); |
719f5d7f | 1150 | free_vfsmnt(mnt); |
be34d1a3 | 1151 | return ERR_PTR(err); |
1da177e4 LT |
1152 | } |
1153 | ||
9ea459e1 AV |
1154 | static void cleanup_mnt(struct mount *mnt) |
1155 | { | |
1156 | /* | |
1157 | * This probably indicates that somebody messed | |
1158 | * up a mnt_want/drop_write() pair. If this | |
1159 | * happens, the filesystem was probably unable | |
1160 | * to make r/w->r/o transitions. | |
1161 | */ | |
1162 | /* | |
1163 | * The locking used to deal with mnt_count decrement provides barriers, | |
1164 | * so mnt_get_writers() below is safe. | |
1165 | */ | |
1166 | WARN_ON(mnt_get_writers(mnt)); | |
1167 | if (unlikely(mnt->mnt_pins.first)) | |
1168 | mnt_pin_kill(mnt); | |
1169 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1170 | dput(mnt->mnt.mnt_root); | |
1171 | deactivate_super(mnt->mnt.mnt_sb); | |
1172 | mnt_free_id(mnt); | |
1173 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1174 | } | |
1175 | ||
1176 | static void __cleanup_mnt(struct rcu_head *head) | |
1177 | { | |
1178 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1179 | } | |
1180 | ||
1181 | static LLIST_HEAD(delayed_mntput_list); | |
1182 | static void delayed_mntput(struct work_struct *unused) | |
1183 | { | |
1184 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1185 | struct mount *m, *t; |
9ea459e1 | 1186 | |
29785735 BP |
1187 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1188 | cleanup_mnt(m); | |
9ea459e1 AV |
1189 | } |
1190 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1191 | ||
900148dc | 1192 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1193 | { |
48a066e7 AV |
1194 | rcu_read_lock(); |
1195 | mnt_add_count(mnt, -1); | |
1196 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1197 | rcu_read_unlock(); | |
f03c6599 | 1198 | return; |
b3e19d92 | 1199 | } |
719ea2fb | 1200 | lock_mount_hash(); |
b3e19d92 | 1201 | if (mnt_get_count(mnt)) { |
48a066e7 | 1202 | rcu_read_unlock(); |
719ea2fb | 1203 | unlock_mount_hash(); |
99b7db7b NP |
1204 | return; |
1205 | } | |
48a066e7 AV |
1206 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1207 | rcu_read_unlock(); | |
1208 | unlock_mount_hash(); | |
1209 | return; | |
1210 | } | |
1211 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1212 | rcu_read_unlock(); | |
962830df | 1213 | |
39f7c4db | 1214 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1215 | |
1216 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1217 | struct mount *p, *tmp; | |
1218 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1219 | umount_mnt(p); | |
1220 | } | |
1221 | } | |
719ea2fb | 1222 | unlock_mount_hash(); |
649a795a | 1223 | |
9ea459e1 AV |
1224 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1225 | struct task_struct *task = current; | |
1226 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1227 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1228 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1229 | return; | |
1230 | } | |
1231 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1232 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1233 | return; | |
1234 | } | |
1235 | cleanup_mnt(mnt); | |
b3e19d92 | 1236 | } |
b3e19d92 NP |
1237 | |
1238 | void mntput(struct vfsmount *mnt) | |
1239 | { | |
1240 | if (mnt) { | |
863d684f | 1241 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1242 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1243 | if (unlikely(m->mnt_expiry_mark)) |
1244 | m->mnt_expiry_mark = 0; | |
1245 | mntput_no_expire(m); | |
b3e19d92 NP |
1246 | } |
1247 | } | |
1248 | EXPORT_SYMBOL(mntput); | |
1249 | ||
1250 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1251 | { | |
1252 | if (mnt) | |
83adc753 | 1253 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1254 | return mnt; |
1255 | } | |
1256 | EXPORT_SYMBOL(mntget); | |
1257 | ||
c6609c0a IK |
1258 | /* path_is_mountpoint() - Check if path is a mount in the current |
1259 | * namespace. | |
1260 | * | |
1261 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1262 | * not mounted in any namespace and that common case is handled inline. | |
1263 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1264 | * mounts using a given dentry in a different namespace. This function | |
1265 | * checks if the passed in path is a mountpoint rather than the dentry | |
1266 | * alone. | |
1267 | */ | |
1268 | bool path_is_mountpoint(const struct path *path) | |
1269 | { | |
1270 | unsigned seq; | |
1271 | bool res; | |
1272 | ||
1273 | if (!d_mountpoint(path->dentry)) | |
1274 | return false; | |
1275 | ||
1276 | rcu_read_lock(); | |
1277 | do { | |
1278 | seq = read_seqbegin(&mount_lock); | |
1279 | res = __path_is_mountpoint(path); | |
1280 | } while (read_seqretry(&mount_lock, seq)); | |
1281 | rcu_read_unlock(); | |
1282 | ||
1283 | return res; | |
1284 | } | |
1285 | EXPORT_SYMBOL(path_is_mountpoint); | |
1286 | ||
ca71cf71 | 1287 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1288 | { |
3064c356 AV |
1289 | struct mount *p; |
1290 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1291 | if (IS_ERR(p)) | |
1292 | return ERR_CAST(p); | |
1293 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1294 | return &p->mnt; | |
7b7b1ace | 1295 | } |
1da177e4 | 1296 | |
a1a2c409 | 1297 | #ifdef CONFIG_PROC_FS |
0226f492 | 1298 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1299 | static void *m_start(struct seq_file *m, loff_t *pos) |
1300 | { | |
ede1bf0d | 1301 | struct proc_mounts *p = m->private; |
1da177e4 | 1302 | |
390c6843 | 1303 | down_read(&namespace_sem); |
c7999c36 AV |
1304 | if (p->cached_event == p->ns->event) { |
1305 | void *v = p->cached_mount; | |
1306 | if (*pos == p->cached_index) | |
1307 | return v; | |
1308 | if (*pos == p->cached_index + 1) { | |
1309 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1310 | return p->cached_mount = v; | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | p->cached_event = p->ns->event; | |
1315 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1316 | p->cached_index = *pos; | |
1317 | return p->cached_mount; | |
1da177e4 LT |
1318 | } |
1319 | ||
1320 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1321 | { | |
ede1bf0d | 1322 | struct proc_mounts *p = m->private; |
b0765fb8 | 1323 | |
c7999c36 AV |
1324 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1325 | p->cached_index = *pos; | |
1326 | return p->cached_mount; | |
1da177e4 LT |
1327 | } |
1328 | ||
1329 | static void m_stop(struct seq_file *m, void *v) | |
1330 | { | |
390c6843 | 1331 | up_read(&namespace_sem); |
1da177e4 LT |
1332 | } |
1333 | ||
0226f492 | 1334 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1335 | { |
ede1bf0d | 1336 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1337 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1338 | return p->show(m, &r->mnt); |
1da177e4 LT |
1339 | } |
1340 | ||
a1a2c409 | 1341 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1342 | .start = m_start, |
1343 | .next = m_next, | |
1344 | .stop = m_stop, | |
0226f492 | 1345 | .show = m_show, |
b4629fe2 | 1346 | }; |
a1a2c409 | 1347 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1348 | |
1da177e4 LT |
1349 | /** |
1350 | * may_umount_tree - check if a mount tree is busy | |
1351 | * @mnt: root of mount tree | |
1352 | * | |
1353 | * This is called to check if a tree of mounts has any | |
1354 | * open files, pwds, chroots or sub mounts that are | |
1355 | * busy. | |
1356 | */ | |
909b0a88 | 1357 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1358 | { |
909b0a88 | 1359 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1360 | int actual_refs = 0; |
1361 | int minimum_refs = 0; | |
315fc83e | 1362 | struct mount *p; |
909b0a88 | 1363 | BUG_ON(!m); |
1da177e4 | 1364 | |
b3e19d92 | 1365 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1366 | lock_mount_hash(); |
909b0a88 | 1367 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1368 | actual_refs += mnt_get_count(p); |
1da177e4 | 1369 | minimum_refs += 2; |
1da177e4 | 1370 | } |
719ea2fb | 1371 | unlock_mount_hash(); |
1da177e4 LT |
1372 | |
1373 | if (actual_refs > minimum_refs) | |
e3474a8e | 1374 | return 0; |
1da177e4 | 1375 | |
e3474a8e | 1376 | return 1; |
1da177e4 LT |
1377 | } |
1378 | ||
1379 | EXPORT_SYMBOL(may_umount_tree); | |
1380 | ||
1381 | /** | |
1382 | * may_umount - check if a mount point is busy | |
1383 | * @mnt: root of mount | |
1384 | * | |
1385 | * This is called to check if a mount point has any | |
1386 | * open files, pwds, chroots or sub mounts. If the | |
1387 | * mount has sub mounts this will return busy | |
1388 | * regardless of whether the sub mounts are busy. | |
1389 | * | |
1390 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1391 | * give false negatives. The main reason why it's here is that we need | |
1392 | * a non-destructive way to look for easily umountable filesystems. | |
1393 | */ | |
1394 | int may_umount(struct vfsmount *mnt) | |
1395 | { | |
e3474a8e | 1396 | int ret = 1; |
8ad08d8a | 1397 | down_read(&namespace_sem); |
719ea2fb | 1398 | lock_mount_hash(); |
1ab59738 | 1399 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1400 | ret = 0; |
719ea2fb | 1401 | unlock_mount_hash(); |
8ad08d8a | 1402 | up_read(&namespace_sem); |
a05964f3 | 1403 | return ret; |
1da177e4 LT |
1404 | } |
1405 | ||
1406 | EXPORT_SYMBOL(may_umount); | |
1407 | ||
38129a13 | 1408 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1409 | |
97216be0 | 1410 | static void namespace_unlock(void) |
70fbcdf4 | 1411 | { |
a3b3c562 | 1412 | struct hlist_head head; |
97216be0 | 1413 | |
a3b3c562 | 1414 | hlist_move_list(&unmounted, &head); |
97216be0 | 1415 | |
97216be0 AV |
1416 | up_write(&namespace_sem); |
1417 | ||
a3b3c562 EB |
1418 | if (likely(hlist_empty(&head))) |
1419 | return; | |
1420 | ||
48a066e7 AV |
1421 | synchronize_rcu(); |
1422 | ||
87b95ce0 | 1423 | group_pin_kill(&head); |
70fbcdf4 RP |
1424 | } |
1425 | ||
97216be0 | 1426 | static inline void namespace_lock(void) |
e3197d83 | 1427 | { |
97216be0 | 1428 | down_write(&namespace_sem); |
e3197d83 AV |
1429 | } |
1430 | ||
e819f152 EB |
1431 | enum umount_tree_flags { |
1432 | UMOUNT_SYNC = 1, | |
1433 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1434 | UMOUNT_CONNECTED = 4, |
e819f152 | 1435 | }; |
f2d0a123 EB |
1436 | |
1437 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1438 | { | |
1439 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1440 | if (how & UMOUNT_SYNC) | |
1441 | return true; | |
1442 | ||
1443 | /* A mount without a parent has nothing to be connected to */ | |
1444 | if (!mnt_has_parent(mnt)) | |
1445 | return true; | |
1446 | ||
1447 | /* Because the reference counting rules change when mounts are | |
1448 | * unmounted and connected, umounted mounts may not be | |
1449 | * connected to mounted mounts. | |
1450 | */ | |
1451 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1452 | return true; | |
1453 | ||
1454 | /* Has it been requested that the mount remain connected? */ | |
1455 | if (how & UMOUNT_CONNECTED) | |
1456 | return false; | |
1457 | ||
1458 | /* Is the mount locked such that it needs to remain connected? */ | |
1459 | if (IS_MNT_LOCKED(mnt)) | |
1460 | return false; | |
1461 | ||
1462 | /* By default disconnect the mount */ | |
1463 | return true; | |
1464 | } | |
1465 | ||
99b7db7b | 1466 | /* |
48a066e7 | 1467 | * mount_lock must be held |
99b7db7b NP |
1468 | * namespace_sem must be held for write |
1469 | */ | |
e819f152 | 1470 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1471 | { |
c003b26f | 1472 | LIST_HEAD(tmp_list); |
315fc83e | 1473 | struct mount *p; |
1da177e4 | 1474 | |
5d88457e EB |
1475 | if (how & UMOUNT_PROPAGATE) |
1476 | propagate_mount_unlock(mnt); | |
1477 | ||
c003b26f | 1478 | /* Gather the mounts to umount */ |
590ce4bc EB |
1479 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1480 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1481 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1482 | } |
1da177e4 | 1483 | |
411a938b | 1484 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1485 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1486 | list_del_init(&p->mnt_child); |
c003b26f | 1487 | } |
88b368f2 | 1488 | |
c003b26f | 1489 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1490 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1491 | propagate_umount(&tmp_list); |
a05964f3 | 1492 | |
c003b26f | 1493 | while (!list_empty(&tmp_list)) { |
d2921684 | 1494 | struct mnt_namespace *ns; |
ce07d891 | 1495 | bool disconnect; |
c003b26f | 1496 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1497 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1498 | list_del_init(&p->mnt_list); |
d2921684 EB |
1499 | ns = p->mnt_ns; |
1500 | if (ns) { | |
1501 | ns->mounts--; | |
1502 | __touch_mnt_namespace(ns); | |
1503 | } | |
143c8c91 | 1504 | p->mnt_ns = NULL; |
e819f152 | 1505 | if (how & UMOUNT_SYNC) |
48a066e7 | 1506 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1507 | |
f2d0a123 | 1508 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1509 | |
1510 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1511 | disconnect ? &unmounted : NULL); | |
676da58d | 1512 | if (mnt_has_parent(p)) { |
81b6b061 | 1513 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1514 | if (!disconnect) { |
1515 | /* Don't forget about p */ | |
1516 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1517 | } else { | |
1518 | umount_mnt(p); | |
1519 | } | |
7c4b93d8 | 1520 | } |
0f0afb1d | 1521 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1522 | } |
1523 | } | |
1524 | ||
b54b9be7 | 1525 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1526 | |
1ab59738 | 1527 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1528 | { |
1ab59738 | 1529 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1530 | int retval; |
1531 | ||
1ab59738 | 1532 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1533 | if (retval) |
1534 | return retval; | |
1535 | ||
1536 | /* | |
1537 | * Allow userspace to request a mountpoint be expired rather than | |
1538 | * unmounting unconditionally. Unmount only happens if: | |
1539 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1540 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1541 | */ | |
1542 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1543 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1544 | flags & (MNT_FORCE | MNT_DETACH)) |
1545 | return -EINVAL; | |
1546 | ||
b3e19d92 NP |
1547 | /* |
1548 | * probably don't strictly need the lock here if we examined | |
1549 | * all race cases, but it's a slowpath. | |
1550 | */ | |
719ea2fb | 1551 | lock_mount_hash(); |
83adc753 | 1552 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1553 | unlock_mount_hash(); |
1da177e4 | 1554 | return -EBUSY; |
b3e19d92 | 1555 | } |
719ea2fb | 1556 | unlock_mount_hash(); |
1da177e4 | 1557 | |
863d684f | 1558 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1559 | return -EAGAIN; |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * If we may have to abort operations to get out of this | |
1564 | * mount, and they will themselves hold resources we must | |
1565 | * allow the fs to do things. In the Unix tradition of | |
1566 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1567 | * might fail to complete on the first run through as other tasks | |
1568 | * must return, and the like. Thats for the mount program to worry | |
1569 | * about for the moment. | |
1570 | */ | |
1571 | ||
42faad99 | 1572 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1573 | sb->s_op->umount_begin(sb); |
42faad99 | 1574 | } |
1da177e4 LT |
1575 | |
1576 | /* | |
1577 | * No sense to grab the lock for this test, but test itself looks | |
1578 | * somewhat bogus. Suggestions for better replacement? | |
1579 | * Ho-hum... In principle, we might treat that as umount + switch | |
1580 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1581 | * Actually it makes sense, especially if rootfs would contain a | |
1582 | * /reboot - static binary that would close all descriptors and | |
1583 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1584 | */ | |
1ab59738 | 1585 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1586 | /* |
1587 | * Special case for "unmounting" root ... | |
1588 | * we just try to remount it readonly. | |
1589 | */ | |
a1480dcc AL |
1590 | if (!capable(CAP_SYS_ADMIN)) |
1591 | return -EPERM; | |
1da177e4 | 1592 | down_write(&sb->s_umount); |
bc98a42c | 1593 | if (!sb_rdonly(sb)) |
e462ec50 | 1594 | retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); |
1da177e4 LT |
1595 | up_write(&sb->s_umount); |
1596 | return retval; | |
1597 | } | |
1598 | ||
97216be0 | 1599 | namespace_lock(); |
719ea2fb | 1600 | lock_mount_hash(); |
5addc5dd | 1601 | event++; |
1da177e4 | 1602 | |
48a066e7 | 1603 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1604 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1605 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1606 | retval = 0; |
48a066e7 AV |
1607 | } else { |
1608 | shrink_submounts(mnt); | |
1609 | retval = -EBUSY; | |
1610 | if (!propagate_mount_busy(mnt, 2)) { | |
1611 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1612 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1613 | retval = 0; |
1614 | } | |
1da177e4 | 1615 | } |
719ea2fb | 1616 | unlock_mount_hash(); |
e3197d83 | 1617 | namespace_unlock(); |
1da177e4 LT |
1618 | return retval; |
1619 | } | |
1620 | ||
80b5dce8 EB |
1621 | /* |
1622 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1623 | * | |
1624 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1625 | * to an existing mountpoint, and wind up leaking the mount. | |
1626 | * detach_mounts allows lazily unmounting those mounts instead of | |
1627 | * leaking them. | |
1628 | * | |
1629 | * The caller may hold dentry->d_inode->i_mutex. | |
1630 | */ | |
1631 | void __detach_mounts(struct dentry *dentry) | |
1632 | { | |
1633 | struct mountpoint *mp; | |
1634 | struct mount *mnt; | |
1635 | ||
1636 | namespace_lock(); | |
3895dbf8 | 1637 | lock_mount_hash(); |
80b5dce8 | 1638 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1639 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1640 | goto out_unlock; |
1641 | ||
e06b933e | 1642 | event++; |
80b5dce8 EB |
1643 | while (!hlist_empty(&mp->m_list)) { |
1644 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1645 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1646 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1647 | umount_mnt(mnt); | |
ce07d891 | 1648 | } |
e0c9c0af | 1649 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1650 | } |
80b5dce8 EB |
1651 | put_mountpoint(mp); |
1652 | out_unlock: | |
3895dbf8 | 1653 | unlock_mount_hash(); |
80b5dce8 EB |
1654 | namespace_unlock(); |
1655 | } | |
1656 | ||
dd111b31 | 1657 | /* |
9b40bc90 AV |
1658 | * Is the caller allowed to modify his namespace? |
1659 | */ | |
1660 | static inline bool may_mount(void) | |
1661 | { | |
1662 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1663 | } | |
1664 | ||
9e8925b6 JL |
1665 | static inline bool may_mandlock(void) |
1666 | { | |
1667 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1668 | return false; | |
1669 | #endif | |
95ace754 | 1670 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1671 | } |
1672 | ||
1da177e4 LT |
1673 | /* |
1674 | * Now umount can handle mount points as well as block devices. | |
1675 | * This is important for filesystems which use unnamed block devices. | |
1676 | * | |
1677 | * We now support a flag for forced unmount like the other 'big iron' | |
1678 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1679 | */ | |
1680 | ||
bdc480e3 | 1681 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1682 | { |
2d8f3038 | 1683 | struct path path; |
900148dc | 1684 | struct mount *mnt; |
1da177e4 | 1685 | int retval; |
db1f05bb | 1686 | int lookup_flags = 0; |
1da177e4 | 1687 | |
db1f05bb MS |
1688 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1689 | return -EINVAL; | |
1690 | ||
9b40bc90 AV |
1691 | if (!may_mount()) |
1692 | return -EPERM; | |
1693 | ||
db1f05bb MS |
1694 | if (!(flags & UMOUNT_NOFOLLOW)) |
1695 | lookup_flags |= LOOKUP_FOLLOW; | |
1696 | ||
197df04c | 1697 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1698 | if (retval) |
1699 | goto out; | |
900148dc | 1700 | mnt = real_mount(path.mnt); |
1da177e4 | 1701 | retval = -EINVAL; |
2d8f3038 | 1702 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1703 | goto dput_and_out; |
143c8c91 | 1704 | if (!check_mnt(mnt)) |
1da177e4 | 1705 | goto dput_and_out; |
5ff9d8a6 EB |
1706 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1707 | goto dput_and_out; | |
b2f5d4dc EB |
1708 | retval = -EPERM; |
1709 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1710 | goto dput_and_out; | |
1da177e4 | 1711 | |
900148dc | 1712 | retval = do_umount(mnt, flags); |
1da177e4 | 1713 | dput_and_out: |
429731b1 | 1714 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1715 | dput(path.dentry); |
900148dc | 1716 | mntput_no_expire(mnt); |
1da177e4 LT |
1717 | out: |
1718 | return retval; | |
1719 | } | |
1720 | ||
1721 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1722 | ||
1723 | /* | |
b58fed8b | 1724 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1725 | */ |
bdc480e3 | 1726 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1727 | { |
b58fed8b | 1728 | return sys_umount(name, 0); |
1da177e4 LT |
1729 | } |
1730 | ||
1731 | #endif | |
1732 | ||
4ce5d2b1 | 1733 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1734 | { |
4ce5d2b1 | 1735 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1736 | return dentry->d_op == &ns_dentry_operations && |
1737 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1738 | } |
1739 | ||
58be2825 AV |
1740 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1741 | { | |
1742 | return container_of(ns, struct mnt_namespace, ns); | |
1743 | } | |
1744 | ||
4ce5d2b1 EB |
1745 | static bool mnt_ns_loop(struct dentry *dentry) |
1746 | { | |
1747 | /* Could bind mounting the mount namespace inode cause a | |
1748 | * mount namespace loop? | |
1749 | */ | |
1750 | struct mnt_namespace *mnt_ns; | |
1751 | if (!is_mnt_ns_file(dentry)) | |
1752 | return false; | |
1753 | ||
f77c8014 | 1754 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1755 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1756 | } | |
1757 | ||
87129cc0 | 1758 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1759 | int flag) |
1da177e4 | 1760 | { |
84d17192 | 1761 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1762 | |
4ce5d2b1 EB |
1763 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1764 | return ERR_PTR(-EINVAL); | |
1765 | ||
1766 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1767 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1768 | |
36341f64 | 1769 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1770 | if (IS_ERR(q)) |
1771 | return q; | |
1772 | ||
a73324da | 1773 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1774 | |
1775 | p = mnt; | |
6b41d536 | 1776 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1777 | struct mount *s; |
7ec02ef1 | 1778 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1779 | continue; |
1780 | ||
909b0a88 | 1781 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1782 | if (!(flag & CL_COPY_UNBINDABLE) && |
1783 | IS_MNT_UNBINDABLE(s)) { | |
1784 | s = skip_mnt_tree(s); | |
1785 | continue; | |
1786 | } | |
1787 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1788 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1789 | s = skip_mnt_tree(s); |
1790 | continue; | |
1791 | } | |
0714a533 AV |
1792 | while (p != s->mnt_parent) { |
1793 | p = p->mnt_parent; | |
1794 | q = q->mnt_parent; | |
1da177e4 | 1795 | } |
87129cc0 | 1796 | p = s; |
84d17192 | 1797 | parent = q; |
87129cc0 | 1798 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1799 | if (IS_ERR(q)) |
1800 | goto out; | |
719ea2fb | 1801 | lock_mount_hash(); |
1a4eeaf2 | 1802 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1803 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1804 | unlock_mount_hash(); |
1da177e4 LT |
1805 | } |
1806 | } | |
1807 | return res; | |
be34d1a3 | 1808 | out: |
1da177e4 | 1809 | if (res) { |
719ea2fb | 1810 | lock_mount_hash(); |
e819f152 | 1811 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1812 | unlock_mount_hash(); |
1da177e4 | 1813 | } |
be34d1a3 | 1814 | return q; |
1da177e4 LT |
1815 | } |
1816 | ||
be34d1a3 DH |
1817 | /* Caller should check returned pointer for errors */ |
1818 | ||
ca71cf71 | 1819 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1820 | { |
cb338d06 | 1821 | struct mount *tree; |
97216be0 | 1822 | namespace_lock(); |
cd4a4017 EB |
1823 | if (!check_mnt(real_mount(path->mnt))) |
1824 | tree = ERR_PTR(-EINVAL); | |
1825 | else | |
1826 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1827 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1828 | namespace_unlock(); |
be34d1a3 | 1829 | if (IS_ERR(tree)) |
52e220d3 | 1830 | return ERR_CAST(tree); |
be34d1a3 | 1831 | return &tree->mnt; |
8aec0809 AV |
1832 | } |
1833 | ||
1834 | void drop_collected_mounts(struct vfsmount *mnt) | |
1835 | { | |
97216be0 | 1836 | namespace_lock(); |
719ea2fb | 1837 | lock_mount_hash(); |
e819f152 | 1838 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1839 | unlock_mount_hash(); |
3ab6abee | 1840 | namespace_unlock(); |
8aec0809 AV |
1841 | } |
1842 | ||
c771d683 MS |
1843 | /** |
1844 | * clone_private_mount - create a private clone of a path | |
1845 | * | |
1846 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1847 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1848 | * to the originating mount won't be propagated into this). | |
1849 | * | |
1850 | * Release with mntput(). | |
1851 | */ | |
ca71cf71 | 1852 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1853 | { |
1854 | struct mount *old_mnt = real_mount(path->mnt); | |
1855 | struct mount *new_mnt; | |
1856 | ||
1857 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1858 | return ERR_PTR(-EINVAL); | |
1859 | ||
c771d683 | 1860 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1861 | if (IS_ERR(new_mnt)) |
1862 | return ERR_CAST(new_mnt); | |
1863 | ||
1864 | return &new_mnt->mnt; | |
1865 | } | |
1866 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1867 | ||
1f707137 AV |
1868 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1869 | struct vfsmount *root) | |
1870 | { | |
1a4eeaf2 | 1871 | struct mount *mnt; |
1f707137 AV |
1872 | int res = f(root, arg); |
1873 | if (res) | |
1874 | return res; | |
1a4eeaf2 AV |
1875 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1876 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1877 | if (res) |
1878 | return res; | |
1879 | } | |
1880 | return 0; | |
1881 | } | |
1882 | ||
4b8b21f4 | 1883 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1884 | { |
315fc83e | 1885 | struct mount *p; |
719f5d7f | 1886 | |
909b0a88 | 1887 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1888 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1889 | mnt_release_group_id(p); |
719f5d7f MS |
1890 | } |
1891 | } | |
1892 | ||
4b8b21f4 | 1893 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1894 | { |
315fc83e | 1895 | struct mount *p; |
719f5d7f | 1896 | |
909b0a88 | 1897 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1898 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1899 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1900 | if (err) { |
4b8b21f4 | 1901 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1902 | return err; |
1903 | } | |
1904 | } | |
1905 | } | |
1906 | ||
1907 | return 0; | |
1908 | } | |
1909 | ||
d2921684 EB |
1910 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1911 | { | |
1912 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1913 | unsigned int mounts = 0, old, pending, sum; | |
1914 | struct mount *p; | |
1915 | ||
1916 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1917 | mounts++; | |
1918 | ||
1919 | old = ns->mounts; | |
1920 | pending = ns->pending_mounts; | |
1921 | sum = old + pending; | |
1922 | if ((old > sum) || | |
1923 | (pending > sum) || | |
1924 | (max < sum) || | |
1925 | (mounts > (max - sum))) | |
1926 | return -ENOSPC; | |
1927 | ||
1928 | ns->pending_mounts = pending + mounts; | |
1929 | return 0; | |
1930 | } | |
1931 | ||
b90fa9ae RP |
1932 | /* |
1933 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1934 | * @nd : place the mount tree @source_mnt is attached |
1935 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1936 | * store the parent mount and mountpoint dentry. | |
1937 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1938 | * |
1939 | * NOTE: in the table below explains the semantics when a source mount | |
1940 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1941 | * --------------------------------------------------------------------------- |
1942 | * | BIND MOUNT OPERATION | | |
1943 | * |************************************************************************** | |
1944 | * | source-->| shared | private | slave | unbindable | | |
1945 | * | dest | | | | | | |
1946 | * | | | | | | | | |
1947 | * | v | | | | | | |
1948 | * |************************************************************************** | |
1949 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1950 | * | | | | | | | |
1951 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1952 | * *************************************************************************** | |
b90fa9ae RP |
1953 | * A bind operation clones the source mount and mounts the clone on the |
1954 | * destination mount. | |
1955 | * | |
1956 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1957 | * tree of the destination mount and the cloned mount is added to | |
1958 | * the peer group of the source mount. | |
1959 | * (+) the cloned mount is created under the destination mount and is marked | |
1960 | * as shared. The cloned mount is added to the peer group of the source | |
1961 | * mount. | |
5afe0022 RP |
1962 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1963 | * of the destination mount and the cloned mount is made slave | |
1964 | * of the same master as that of the source mount. The cloned mount | |
1965 | * is marked as 'shared and slave'. | |
1966 | * (*) the cloned mount is made a slave of the same master as that of the | |
1967 | * source mount. | |
1968 | * | |
9676f0c6 RP |
1969 | * --------------------------------------------------------------------------- |
1970 | * | MOVE MOUNT OPERATION | | |
1971 | * |************************************************************************** | |
1972 | * | source-->| shared | private | slave | unbindable | | |
1973 | * | dest | | | | | | |
1974 | * | | | | | | | | |
1975 | * | v | | | | | | |
1976 | * |************************************************************************** | |
1977 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1978 | * | | | | | | | |
1979 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1980 | * *************************************************************************** | |
5afe0022 RP |
1981 | * |
1982 | * (+) the mount is moved to the destination. And is then propagated to | |
1983 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1984 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1985 | * (+++) the mount is moved to the destination and is then propagated to |
1986 | * all the mounts belonging to the destination mount's propagation tree. | |
1987 | * the mount is marked as 'shared and slave'. | |
1988 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
1989 | * |
1990 | * if the source mount is a tree, the operations explained above is | |
1991 | * applied to each mount in the tree. | |
1992 | * Must be called without spinlocks held, since this function can sleep | |
1993 | * in allocations. | |
1994 | */ | |
0fb54e50 | 1995 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
1996 | struct mount *dest_mnt, |
1997 | struct mountpoint *dest_mp, | |
1998 | struct path *parent_path) | |
b90fa9ae | 1999 | { |
38129a13 | 2000 | HLIST_HEAD(tree_list); |
d2921684 | 2001 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2002 | struct mountpoint *smp; |
315fc83e | 2003 | struct mount *child, *p; |
38129a13 | 2004 | struct hlist_node *n; |
719f5d7f | 2005 | int err; |
b90fa9ae | 2006 | |
1064f874 EB |
2007 | /* Preallocate a mountpoint in case the new mounts need |
2008 | * to be tucked under other mounts. | |
2009 | */ | |
2010 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2011 | if (IS_ERR(smp)) | |
2012 | return PTR_ERR(smp); | |
2013 | ||
d2921684 EB |
2014 | /* Is there space to add these mounts to the mount namespace? */ |
2015 | if (!parent_path) { | |
2016 | err = count_mounts(ns, source_mnt); | |
2017 | if (err) | |
2018 | goto out; | |
2019 | } | |
2020 | ||
fc7be130 | 2021 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2022 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2023 | if (err) |
2024 | goto out; | |
0b1b901b | 2025 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2026 | lock_mount_hash(); |
0b1b901b AV |
2027 | if (err) |
2028 | goto out_cleanup_ids; | |
909b0a88 | 2029 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2030 | set_mnt_shared(p); |
0b1b901b AV |
2031 | } else { |
2032 | lock_mount_hash(); | |
b90fa9ae | 2033 | } |
1a390689 | 2034 | if (parent_path) { |
0fb54e50 | 2035 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2036 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2037 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2038 | } else { |
84d17192 | 2039 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2040 | commit_tree(source_mnt); |
21444403 | 2041 | } |
b90fa9ae | 2042 | |
38129a13 | 2043 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2044 | struct mount *q; |
38129a13 | 2045 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2046 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2047 | child->mnt_mountpoint); | |
2048 | if (q) | |
2049 | mnt_change_mountpoint(child, smp, q); | |
2050 | commit_tree(child); | |
b90fa9ae | 2051 | } |
1064f874 | 2052 | put_mountpoint(smp); |
719ea2fb | 2053 | unlock_mount_hash(); |
99b7db7b | 2054 | |
b90fa9ae | 2055 | return 0; |
719f5d7f MS |
2056 | |
2057 | out_cleanup_ids: | |
f2ebb3a9 AV |
2058 | while (!hlist_empty(&tree_list)) { |
2059 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2060 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2061 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2062 | } |
2063 | unlock_mount_hash(); | |
0b1b901b | 2064 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2065 | out: |
d2921684 | 2066 | ns->pending_mounts = 0; |
1064f874 EB |
2067 | |
2068 | read_seqlock_excl(&mount_lock); | |
2069 | put_mountpoint(smp); | |
2070 | read_sequnlock_excl(&mount_lock); | |
2071 | ||
719f5d7f | 2072 | return err; |
b90fa9ae RP |
2073 | } |
2074 | ||
84d17192 | 2075 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2076 | { |
2077 | struct vfsmount *mnt; | |
84d17192 | 2078 | struct dentry *dentry = path->dentry; |
b12cea91 | 2079 | retry: |
5955102c | 2080 | inode_lock(dentry->d_inode); |
84d17192 | 2081 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2082 | inode_unlock(dentry->d_inode); |
84d17192 | 2083 | return ERR_PTR(-ENOENT); |
b12cea91 | 2084 | } |
97216be0 | 2085 | namespace_lock(); |
b12cea91 | 2086 | mnt = lookup_mnt(path); |
84d17192 | 2087 | if (likely(!mnt)) { |
3895dbf8 | 2088 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2089 | if (IS_ERR(mp)) { |
97216be0 | 2090 | namespace_unlock(); |
5955102c | 2091 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2092 | return mp; |
2093 | } | |
2094 | return mp; | |
2095 | } | |
97216be0 | 2096 | namespace_unlock(); |
5955102c | 2097 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2098 | path_put(path); |
2099 | path->mnt = mnt; | |
84d17192 | 2100 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2101 | goto retry; |
2102 | } | |
2103 | ||
84d17192 | 2104 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2105 | { |
84d17192 | 2106 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2107 | |
2108 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2109 | put_mountpoint(where); |
3895dbf8 EB |
2110 | read_sequnlock_excl(&mount_lock); |
2111 | ||
328e6d90 | 2112 | namespace_unlock(); |
5955102c | 2113 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2114 | } |
2115 | ||
84d17192 | 2116 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2117 | { |
e462ec50 | 2118 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2119 | return -EINVAL; |
2120 | ||
e36cb0b8 DH |
2121 | if (d_is_dir(mp->m_dentry) != |
2122 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2123 | return -ENOTDIR; |
2124 | ||
84d17192 | 2125 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2126 | } |
2127 | ||
7a2e8a8f VA |
2128 | /* |
2129 | * Sanity check the flags to change_mnt_propagation. | |
2130 | */ | |
2131 | ||
e462ec50 | 2132 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2133 | { |
e462ec50 | 2134 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2135 | |
2136 | /* Fail if any non-propagation flags are set */ | |
2137 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2138 | return 0; | |
2139 | /* Only one propagation flag should be set */ | |
2140 | if (!is_power_of_2(type)) | |
2141 | return 0; | |
2142 | return type; | |
2143 | } | |
2144 | ||
07b20889 RP |
2145 | /* |
2146 | * recursively change the type of the mountpoint. | |
2147 | */ | |
e462ec50 | 2148 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2149 | { |
315fc83e | 2150 | struct mount *m; |
4b8b21f4 | 2151 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2152 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2153 | int type; |
719f5d7f | 2154 | int err = 0; |
07b20889 | 2155 | |
2d92ab3c | 2156 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2157 | return -EINVAL; |
2158 | ||
e462ec50 | 2159 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2160 | if (!type) |
2161 | return -EINVAL; | |
2162 | ||
97216be0 | 2163 | namespace_lock(); |
719f5d7f MS |
2164 | if (type == MS_SHARED) { |
2165 | err = invent_group_ids(mnt, recurse); | |
2166 | if (err) | |
2167 | goto out_unlock; | |
2168 | } | |
2169 | ||
719ea2fb | 2170 | lock_mount_hash(); |
909b0a88 | 2171 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2172 | change_mnt_propagation(m, type); |
719ea2fb | 2173 | unlock_mount_hash(); |
719f5d7f MS |
2174 | |
2175 | out_unlock: | |
97216be0 | 2176 | namespace_unlock(); |
719f5d7f | 2177 | return err; |
07b20889 RP |
2178 | } |
2179 | ||
5ff9d8a6 EB |
2180 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2181 | { | |
2182 | struct mount *child; | |
2183 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2184 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2185 | continue; | |
2186 | ||
2187 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2188 | return true; | |
2189 | } | |
2190 | return false; | |
2191 | } | |
2192 | ||
1da177e4 LT |
2193 | /* |
2194 | * do loopback mount. | |
2195 | */ | |
808d4e3c | 2196 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2197 | int recurse) |
1da177e4 | 2198 | { |
2d92ab3c | 2199 | struct path old_path; |
84d17192 AV |
2200 | struct mount *mnt = NULL, *old, *parent; |
2201 | struct mountpoint *mp; | |
57eccb83 | 2202 | int err; |
1da177e4 LT |
2203 | if (!old_name || !*old_name) |
2204 | return -EINVAL; | |
815d405c | 2205 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2206 | if (err) |
2207 | return err; | |
2208 | ||
8823c079 | 2209 | err = -EINVAL; |
4ce5d2b1 | 2210 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2211 | goto out; |
8823c079 | 2212 | |
84d17192 AV |
2213 | mp = lock_mount(path); |
2214 | err = PTR_ERR(mp); | |
2215 | if (IS_ERR(mp)) | |
b12cea91 AV |
2216 | goto out; |
2217 | ||
87129cc0 | 2218 | old = real_mount(old_path.mnt); |
84d17192 | 2219 | parent = real_mount(path->mnt); |
87129cc0 | 2220 | |
1da177e4 | 2221 | err = -EINVAL; |
fc7be130 | 2222 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2223 | goto out2; |
9676f0c6 | 2224 | |
e149ed2b AV |
2225 | if (!check_mnt(parent)) |
2226 | goto out2; | |
2227 | ||
2228 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2229 | goto out2; |
1da177e4 | 2230 | |
5ff9d8a6 EB |
2231 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2232 | goto out2; | |
2233 | ||
ccd48bc7 | 2234 | if (recurse) |
4ce5d2b1 | 2235 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2236 | else |
87129cc0 | 2237 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2238 | |
be34d1a3 DH |
2239 | if (IS_ERR(mnt)) { |
2240 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2241 | goto out2; |
be34d1a3 | 2242 | } |
ccd48bc7 | 2243 | |
5ff9d8a6 EB |
2244 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2245 | ||
84d17192 | 2246 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2247 | if (err) { |
719ea2fb | 2248 | lock_mount_hash(); |
e819f152 | 2249 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2250 | unlock_mount_hash(); |
5b83d2c5 | 2251 | } |
b12cea91 | 2252 | out2: |
84d17192 | 2253 | unlock_mount(mp); |
ccd48bc7 | 2254 | out: |
2d92ab3c | 2255 | path_put(&old_path); |
1da177e4 LT |
2256 | return err; |
2257 | } | |
2258 | ||
2e4b7fcd DH |
2259 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2260 | { | |
2261 | int error = 0; | |
2262 | int readonly_request = 0; | |
2263 | ||
2264 | if (ms_flags & MS_RDONLY) | |
2265 | readonly_request = 1; | |
2266 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2267 | return 0; | |
2268 | ||
2269 | if (readonly_request) | |
83adc753 | 2270 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2271 | else |
83adc753 | 2272 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2273 | return error; |
2274 | } | |
2275 | ||
1da177e4 LT |
2276 | /* |
2277 | * change filesystem flags. dir should be a physical root of filesystem. | |
2278 | * If you've mounted a non-root directory somewhere and want to do remount | |
2279 | * on it - tough luck. | |
2280 | */ | |
e462ec50 DH |
2281 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2282 | int mnt_flags, void *data) | |
1da177e4 LT |
2283 | { |
2284 | int err; | |
2d92ab3c | 2285 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2286 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2287 | |
143c8c91 | 2288 | if (!check_mnt(mnt)) |
1da177e4 LT |
2289 | return -EINVAL; |
2290 | ||
2d92ab3c | 2291 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2292 | return -EINVAL; |
2293 | ||
07b64558 EB |
2294 | /* Don't allow changing of locked mnt flags. |
2295 | * | |
2296 | * No locks need to be held here while testing the various | |
2297 | * MNT_LOCK flags because those flags can never be cleared | |
2298 | * once they are set. | |
2299 | */ | |
2300 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2301 | !(mnt_flags & MNT_READONLY)) { | |
2302 | return -EPERM; | |
2303 | } | |
9566d674 EB |
2304 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2305 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2306 | return -EPERM; |
9566d674 EB |
2307 | } |
2308 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2309 | !(mnt_flags & MNT_NOSUID)) { | |
2310 | return -EPERM; | |
2311 | } | |
2312 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2313 | !(mnt_flags & MNT_NOEXEC)) { | |
2314 | return -EPERM; | |
2315 | } | |
2316 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2317 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2318 | return -EPERM; | |
2319 | } | |
2320 | ||
ff36fe2c EP |
2321 | err = security_sb_remount(sb, data); |
2322 | if (err) | |
2323 | return err; | |
2324 | ||
1da177e4 | 2325 | down_write(&sb->s_umount); |
e462ec50 DH |
2326 | if (ms_flags & MS_BIND) |
2327 | err = change_mount_flags(path->mnt, ms_flags); | |
57eccb83 AV |
2328 | else if (!capable(CAP_SYS_ADMIN)) |
2329 | err = -EPERM; | |
4aa98cf7 | 2330 | else |
e462ec50 | 2331 | err = do_remount_sb(sb, sb_flags, data, 0); |
7b43a79f | 2332 | if (!err) { |
719ea2fb | 2333 | lock_mount_hash(); |
a6138db8 | 2334 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2335 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2336 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2337 | unlock_mount_hash(); |
0e55a7cc | 2338 | } |
6339dab8 | 2339 | up_write(&sb->s_umount); |
1da177e4 LT |
2340 | return err; |
2341 | } | |
2342 | ||
cbbe362c | 2343 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2344 | { |
315fc83e | 2345 | struct mount *p; |
909b0a88 | 2346 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2347 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2348 | return 1; |
2349 | } | |
2350 | return 0; | |
2351 | } | |
2352 | ||
808d4e3c | 2353 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2354 | { |
2d92ab3c | 2355 | struct path old_path, parent_path; |
676da58d | 2356 | struct mount *p; |
0fb54e50 | 2357 | struct mount *old; |
84d17192 | 2358 | struct mountpoint *mp; |
57eccb83 | 2359 | int err; |
1da177e4 LT |
2360 | if (!old_name || !*old_name) |
2361 | return -EINVAL; | |
2d92ab3c | 2362 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2363 | if (err) |
2364 | return err; | |
2365 | ||
84d17192 AV |
2366 | mp = lock_mount(path); |
2367 | err = PTR_ERR(mp); | |
2368 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2369 | goto out; |
2370 | ||
143c8c91 | 2371 | old = real_mount(old_path.mnt); |
fc7be130 | 2372 | p = real_mount(path->mnt); |
143c8c91 | 2373 | |
1da177e4 | 2374 | err = -EINVAL; |
fc7be130 | 2375 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2376 | goto out1; |
2377 | ||
5ff9d8a6 EB |
2378 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2379 | goto out1; | |
2380 | ||
1da177e4 | 2381 | err = -EINVAL; |
2d92ab3c | 2382 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2383 | goto out1; |
1da177e4 | 2384 | |
676da58d | 2385 | if (!mnt_has_parent(old)) |
21444403 | 2386 | goto out1; |
1da177e4 | 2387 | |
e36cb0b8 DH |
2388 | if (d_is_dir(path->dentry) != |
2389 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2390 | goto out1; |
2391 | /* | |
2392 | * Don't move a mount residing in a shared parent. | |
2393 | */ | |
fc7be130 | 2394 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2395 | goto out1; |
9676f0c6 RP |
2396 | /* |
2397 | * Don't move a mount tree containing unbindable mounts to a destination | |
2398 | * mount which is shared. | |
2399 | */ | |
fc7be130 | 2400 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2401 | goto out1; |
1da177e4 | 2402 | err = -ELOOP; |
fc7be130 | 2403 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2404 | if (p == old) |
21444403 | 2405 | goto out1; |
1da177e4 | 2406 | |
84d17192 | 2407 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2408 | if (err) |
21444403 | 2409 | goto out1; |
1da177e4 LT |
2410 | |
2411 | /* if the mount is moved, it should no longer be expire | |
2412 | * automatically */ | |
6776db3d | 2413 | list_del_init(&old->mnt_expire); |
1da177e4 | 2414 | out1: |
84d17192 | 2415 | unlock_mount(mp); |
1da177e4 | 2416 | out: |
1da177e4 | 2417 | if (!err) |
1a390689 | 2418 | path_put(&parent_path); |
2d92ab3c | 2419 | path_put(&old_path); |
1da177e4 LT |
2420 | return err; |
2421 | } | |
2422 | ||
9d412a43 AV |
2423 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2424 | { | |
2425 | int err; | |
2426 | const char *subtype = strchr(fstype, '.'); | |
2427 | if (subtype) { | |
2428 | subtype++; | |
2429 | err = -EINVAL; | |
2430 | if (!subtype[0]) | |
2431 | goto err; | |
2432 | } else | |
2433 | subtype = ""; | |
2434 | ||
2435 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2436 | err = -ENOMEM; | |
2437 | if (!mnt->mnt_sb->s_subtype) | |
2438 | goto err; | |
2439 | return mnt; | |
2440 | ||
2441 | err: | |
2442 | mntput(mnt); | |
2443 | return ERR_PTR(err); | |
2444 | } | |
2445 | ||
9d412a43 AV |
2446 | /* |
2447 | * add a mount into a namespace's mount tree | |
2448 | */ | |
95bc5f25 | 2449 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2450 | { |
84d17192 AV |
2451 | struct mountpoint *mp; |
2452 | struct mount *parent; | |
9d412a43 AV |
2453 | int err; |
2454 | ||
f2ebb3a9 | 2455 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2456 | |
84d17192 AV |
2457 | mp = lock_mount(path); |
2458 | if (IS_ERR(mp)) | |
2459 | return PTR_ERR(mp); | |
9d412a43 | 2460 | |
84d17192 | 2461 | parent = real_mount(path->mnt); |
9d412a43 | 2462 | err = -EINVAL; |
84d17192 | 2463 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2464 | /* that's acceptable only for automounts done in private ns */ |
2465 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2466 | goto unlock; | |
2467 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2468 | if (!parent->mnt_ns) |
156cacb1 AV |
2469 | goto unlock; |
2470 | } | |
9d412a43 AV |
2471 | |
2472 | /* Refuse the same filesystem on the same mount point */ | |
2473 | err = -EBUSY; | |
95bc5f25 | 2474 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2475 | path->mnt->mnt_root == path->dentry) |
2476 | goto unlock; | |
2477 | ||
2478 | err = -EINVAL; | |
e36cb0b8 | 2479 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2480 | goto unlock; |
2481 | ||
95bc5f25 | 2482 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2483 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2484 | |
2485 | unlock: | |
84d17192 | 2486 | unlock_mount(mp); |
9d412a43 AV |
2487 | return err; |
2488 | } | |
b1e75df4 | 2489 | |
8654df4e | 2490 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2491 | |
1da177e4 LT |
2492 | /* |
2493 | * create a new mount for userspace and request it to be added into the | |
2494 | * namespace's tree | |
2495 | */ | |
e462ec50 | 2496 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2497 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2498 | { |
0c55cfc4 | 2499 | struct file_system_type *type; |
1da177e4 | 2500 | struct vfsmount *mnt; |
15f9a3f3 | 2501 | int err; |
1da177e4 | 2502 | |
0c55cfc4 | 2503 | if (!fstype) |
1da177e4 LT |
2504 | return -EINVAL; |
2505 | ||
0c55cfc4 EB |
2506 | type = get_fs_type(fstype); |
2507 | if (!type) | |
2508 | return -ENODEV; | |
2509 | ||
e462ec50 | 2510 | mnt = vfs_kern_mount(type, sb_flags, name, data); |
0c55cfc4 EB |
2511 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
2512 | !mnt->mnt_sb->s_subtype) | |
2513 | mnt = fs_set_subtype(mnt, fstype); | |
2514 | ||
2515 | put_filesystem(type); | |
1da177e4 LT |
2516 | if (IS_ERR(mnt)) |
2517 | return PTR_ERR(mnt); | |
2518 | ||
8654df4e EB |
2519 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2520 | mntput(mnt); | |
2521 | return -EPERM; | |
2522 | } | |
2523 | ||
95bc5f25 | 2524 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2525 | if (err) |
2526 | mntput(mnt); | |
2527 | return err; | |
1da177e4 LT |
2528 | } |
2529 | ||
19a167af AV |
2530 | int finish_automount(struct vfsmount *m, struct path *path) |
2531 | { | |
6776db3d | 2532 | struct mount *mnt = real_mount(m); |
19a167af AV |
2533 | int err; |
2534 | /* The new mount record should have at least 2 refs to prevent it being | |
2535 | * expired before we get a chance to add it | |
2536 | */ | |
6776db3d | 2537 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2538 | |
2539 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2540 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2541 | err = -ELOOP; |
2542 | goto fail; | |
19a167af AV |
2543 | } |
2544 | ||
95bc5f25 | 2545 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2546 | if (!err) |
2547 | return 0; | |
2548 | fail: | |
2549 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2550 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2551 | namespace_lock(); |
6776db3d | 2552 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2553 | namespace_unlock(); |
19a167af | 2554 | } |
b1e75df4 AV |
2555 | mntput(m); |
2556 | mntput(m); | |
19a167af AV |
2557 | return err; |
2558 | } | |
2559 | ||
ea5b778a DH |
2560 | /** |
2561 | * mnt_set_expiry - Put a mount on an expiration list | |
2562 | * @mnt: The mount to list. | |
2563 | * @expiry_list: The list to add the mount to. | |
2564 | */ | |
2565 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2566 | { | |
97216be0 | 2567 | namespace_lock(); |
ea5b778a | 2568 | |
6776db3d | 2569 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2570 | |
97216be0 | 2571 | namespace_unlock(); |
ea5b778a DH |
2572 | } |
2573 | EXPORT_SYMBOL(mnt_set_expiry); | |
2574 | ||
1da177e4 LT |
2575 | /* |
2576 | * process a list of expirable mountpoints with the intent of discarding any | |
2577 | * mountpoints that aren't in use and haven't been touched since last we came | |
2578 | * here | |
2579 | */ | |
2580 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2581 | { | |
761d5c38 | 2582 | struct mount *mnt, *next; |
1da177e4 LT |
2583 | LIST_HEAD(graveyard); |
2584 | ||
2585 | if (list_empty(mounts)) | |
2586 | return; | |
2587 | ||
97216be0 | 2588 | namespace_lock(); |
719ea2fb | 2589 | lock_mount_hash(); |
1da177e4 LT |
2590 | |
2591 | /* extract from the expiration list every vfsmount that matches the | |
2592 | * following criteria: | |
2593 | * - only referenced by its parent vfsmount | |
2594 | * - still marked for expiry (marked on the last call here; marks are | |
2595 | * cleared by mntput()) | |
2596 | */ | |
6776db3d | 2597 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2598 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2599 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2600 | continue; |
6776db3d | 2601 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2602 | } |
bcc5c7d2 | 2603 | while (!list_empty(&graveyard)) { |
6776db3d | 2604 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2605 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2606 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2607 | } |
719ea2fb | 2608 | unlock_mount_hash(); |
3ab6abee | 2609 | namespace_unlock(); |
5528f911 TM |
2610 | } |
2611 | ||
2612 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2613 | ||
2614 | /* | |
2615 | * Ripoff of 'select_parent()' | |
2616 | * | |
2617 | * search the list of submounts for a given mountpoint, and move any | |
2618 | * shrinkable submounts to the 'graveyard' list. | |
2619 | */ | |
692afc31 | 2620 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2621 | { |
692afc31 | 2622 | struct mount *this_parent = parent; |
5528f911 TM |
2623 | struct list_head *next; |
2624 | int found = 0; | |
2625 | ||
2626 | repeat: | |
6b41d536 | 2627 | next = this_parent->mnt_mounts.next; |
5528f911 | 2628 | resume: |
6b41d536 | 2629 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2630 | struct list_head *tmp = next; |
6b41d536 | 2631 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2632 | |
2633 | next = tmp->next; | |
692afc31 | 2634 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2635 | continue; |
5528f911 TM |
2636 | /* |
2637 | * Descend a level if the d_mounts list is non-empty. | |
2638 | */ | |
6b41d536 | 2639 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2640 | this_parent = mnt; |
2641 | goto repeat; | |
2642 | } | |
1da177e4 | 2643 | |
1ab59738 | 2644 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2645 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2646 | found++; |
2647 | } | |
1da177e4 | 2648 | } |
5528f911 TM |
2649 | /* |
2650 | * All done at this level ... ascend and resume the search | |
2651 | */ | |
2652 | if (this_parent != parent) { | |
6b41d536 | 2653 | next = this_parent->mnt_child.next; |
0714a533 | 2654 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2655 | goto resume; |
2656 | } | |
2657 | return found; | |
2658 | } | |
2659 | ||
2660 | /* | |
2661 | * process a list of expirable mountpoints with the intent of discarding any | |
2662 | * submounts of a specific parent mountpoint | |
99b7db7b | 2663 | * |
48a066e7 | 2664 | * mount_lock must be held for write |
5528f911 | 2665 | */ |
b54b9be7 | 2666 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2667 | { |
2668 | LIST_HEAD(graveyard); | |
761d5c38 | 2669 | struct mount *m; |
5528f911 | 2670 | |
5528f911 | 2671 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2672 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2673 | while (!list_empty(&graveyard)) { |
761d5c38 | 2674 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2675 | mnt_expire); |
143c8c91 | 2676 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2677 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2678 | } |
2679 | } | |
1da177e4 LT |
2680 | } |
2681 | ||
1da177e4 LT |
2682 | /* |
2683 | * Some copy_from_user() implementations do not return the exact number of | |
2684 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2685 | * Note that this function differs from copy_from_user() in that it will oops | |
2686 | * on bad values of `to', rather than returning a short copy. | |
2687 | */ | |
b58fed8b RP |
2688 | static long exact_copy_from_user(void *to, const void __user * from, |
2689 | unsigned long n) | |
1da177e4 LT |
2690 | { |
2691 | char *t = to; | |
2692 | const char __user *f = from; | |
2693 | char c; | |
2694 | ||
2695 | if (!access_ok(VERIFY_READ, from, n)) | |
2696 | return n; | |
2697 | ||
2698 | while (n) { | |
2699 | if (__get_user(c, f)) { | |
2700 | memset(t, 0, n); | |
2701 | break; | |
2702 | } | |
2703 | *t++ = c; | |
2704 | f++; | |
2705 | n--; | |
2706 | } | |
2707 | return n; | |
2708 | } | |
2709 | ||
b40ef869 | 2710 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2711 | { |
2712 | int i; | |
1da177e4 | 2713 | unsigned long size; |
b40ef869 | 2714 | char *copy; |
b58fed8b | 2715 | |
1da177e4 | 2716 | if (!data) |
b40ef869 | 2717 | return NULL; |
1da177e4 | 2718 | |
b40ef869 AV |
2719 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2720 | if (!copy) | |
2721 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2722 | |
2723 | /* We only care that *some* data at the address the user | |
2724 | * gave us is valid. Just in case, we'll zero | |
2725 | * the remainder of the page. | |
2726 | */ | |
2727 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2728 | size = TASK_SIZE - (unsigned long)data; | |
2729 | if (size > PAGE_SIZE) | |
2730 | size = PAGE_SIZE; | |
2731 | ||
b40ef869 | 2732 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2733 | if (!i) { |
b40ef869 AV |
2734 | kfree(copy); |
2735 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2736 | } |
2737 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2738 | memset(copy + i, 0, PAGE_SIZE - i); |
2739 | return copy; | |
1da177e4 LT |
2740 | } |
2741 | ||
b8850d1f | 2742 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2743 | { |
b8850d1f | 2744 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2745 | } |
2746 | ||
1da177e4 LT |
2747 | /* |
2748 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2749 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2750 | * | |
2751 | * data is a (void *) that can point to any structure up to | |
2752 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2753 | * information (or be NULL). | |
2754 | * | |
2755 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2756 | * When the flags word was introduced its top half was required | |
2757 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2758 | * Therefore, if this magic number is present, it carries no information | |
2759 | * and must be discarded. | |
2760 | */ | |
5e6123f3 | 2761 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2762 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2763 | { |
2d92ab3c | 2764 | struct path path; |
e462ec50 | 2765 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2766 | int retval = 0; |
1da177e4 LT |
2767 | |
2768 | /* Discard magic */ | |
2769 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2770 | flags &= ~MS_MGC_MSK; | |
2771 | ||
2772 | /* Basic sanity checks */ | |
1da177e4 LT |
2773 | if (data_page) |
2774 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2775 | ||
e462ec50 DH |
2776 | if (flags & MS_NOUSER) |
2777 | return -EINVAL; | |
2778 | ||
a27ab9f2 | 2779 | /* ... and get the mountpoint */ |
5e6123f3 | 2780 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2781 | if (retval) |
2782 | return retval; | |
2783 | ||
2784 | retval = security_sb_mount(dev_name, &path, | |
2785 | type_page, flags, data_page); | |
0d5cadb8 AV |
2786 | if (!retval && !may_mount()) |
2787 | retval = -EPERM; | |
e462ec50 | 2788 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 2789 | retval = -EPERM; |
a27ab9f2 TH |
2790 | if (retval) |
2791 | goto dput_out; | |
2792 | ||
613cbe3d AK |
2793 | /* Default to relatime unless overriden */ |
2794 | if (!(flags & MS_NOATIME)) | |
2795 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2796 | |
1da177e4 LT |
2797 | /* Separate the per-mountpoint flags */ |
2798 | if (flags & MS_NOSUID) | |
2799 | mnt_flags |= MNT_NOSUID; | |
2800 | if (flags & MS_NODEV) | |
2801 | mnt_flags |= MNT_NODEV; | |
2802 | if (flags & MS_NOEXEC) | |
2803 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2804 | if (flags & MS_NOATIME) |
2805 | mnt_flags |= MNT_NOATIME; | |
2806 | if (flags & MS_NODIRATIME) | |
2807 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2808 | if (flags & MS_STRICTATIME) |
2809 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
e462ec50 | 2810 | if (flags & SB_RDONLY) |
2e4b7fcd | 2811 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 2812 | |
ffbc6f0e EB |
2813 | /* The default atime for remount is preservation */ |
2814 | if ((flags & MS_REMOUNT) && | |
2815 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2816 | MS_STRICTATIME)) == 0)) { | |
2817 | mnt_flags &= ~MNT_ATIME_MASK; | |
2818 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2819 | } | |
2820 | ||
e462ec50 DH |
2821 | sb_flags = flags & (SB_RDONLY | |
2822 | SB_SYNCHRONOUS | | |
2823 | SB_MANDLOCK | | |
2824 | SB_DIRSYNC | | |
2825 | SB_SILENT | | |
2826 | SB_POSIXACL); | |
1da177e4 | 2827 | |
1da177e4 | 2828 | if (flags & MS_REMOUNT) |
e462ec50 | 2829 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
2830 | data_page); |
2831 | else if (flags & MS_BIND) | |
2d92ab3c | 2832 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2833 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2834 | retval = do_change_type(&path, flags); |
1da177e4 | 2835 | else if (flags & MS_MOVE) |
2d92ab3c | 2836 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2837 | else |
e462ec50 | 2838 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
2839 | dev_name, data_page); |
2840 | dput_out: | |
2d92ab3c | 2841 | path_put(&path); |
1da177e4 LT |
2842 | return retval; |
2843 | } | |
2844 | ||
537f7ccb EB |
2845 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2846 | { | |
2847 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2848 | } | |
2849 | ||
2850 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2851 | { | |
2852 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2853 | } | |
2854 | ||
771b1371 EB |
2855 | static void free_mnt_ns(struct mnt_namespace *ns) |
2856 | { | |
6344c433 | 2857 | ns_free_inum(&ns->ns); |
537f7ccb | 2858 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2859 | put_user_ns(ns->user_ns); |
2860 | kfree(ns); | |
2861 | } | |
2862 | ||
8823c079 EB |
2863 | /* |
2864 | * Assign a sequence number so we can detect when we attempt to bind | |
2865 | * mount a reference to an older mount namespace into the current | |
2866 | * mount namespace, preventing reference counting loops. A 64bit | |
2867 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2868 | * is effectively never, so we can ignore the possibility. | |
2869 | */ | |
2870 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2871 | ||
771b1371 | 2872 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2873 | { |
2874 | struct mnt_namespace *new_ns; | |
537f7ccb | 2875 | struct ucounts *ucounts; |
98f842e6 | 2876 | int ret; |
cf8d2c11 | 2877 | |
537f7ccb EB |
2878 | ucounts = inc_mnt_namespaces(user_ns); |
2879 | if (!ucounts) | |
df75e774 | 2880 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2881 | |
cf8d2c11 | 2882 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2883 | if (!new_ns) { |
2884 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2885 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2886 | } |
6344c433 | 2887 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2888 | if (ret) { |
2889 | kfree(new_ns); | |
537f7ccb | 2890 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2891 | return ERR_PTR(ret); |
2892 | } | |
33c42940 | 2893 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2894 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2895 | atomic_set(&new_ns->count, 1); |
2896 | new_ns->root = NULL; | |
2897 | INIT_LIST_HEAD(&new_ns->list); | |
2898 | init_waitqueue_head(&new_ns->poll); | |
2899 | new_ns->event = 0; | |
771b1371 | 2900 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2901 | new_ns->ucounts = ucounts; |
d2921684 EB |
2902 | new_ns->mounts = 0; |
2903 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2904 | return new_ns; |
2905 | } | |
2906 | ||
0766f788 | 2907 | __latent_entropy |
9559f689 AV |
2908 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2909 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2910 | { |
6b3286ed | 2911 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2912 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2913 | struct mount *p, *q; |
9559f689 | 2914 | struct mount *old; |
cb338d06 | 2915 | struct mount *new; |
7a472ef4 | 2916 | int copy_flags; |
1da177e4 | 2917 | |
9559f689 AV |
2918 | BUG_ON(!ns); |
2919 | ||
2920 | if (likely(!(flags & CLONE_NEWNS))) { | |
2921 | get_mnt_ns(ns); | |
2922 | return ns; | |
2923 | } | |
2924 | ||
2925 | old = ns->root; | |
2926 | ||
771b1371 | 2927 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2928 | if (IS_ERR(new_ns)) |
2929 | return new_ns; | |
1da177e4 | 2930 | |
97216be0 | 2931 | namespace_lock(); |
1da177e4 | 2932 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2933 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2934 | if (user_ns != ns->user_ns) |
132c94e3 | 2935 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2936 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2937 | if (IS_ERR(new)) { |
328e6d90 | 2938 | namespace_unlock(); |
771b1371 | 2939 | free_mnt_ns(new_ns); |
be34d1a3 | 2940 | return ERR_CAST(new); |
1da177e4 | 2941 | } |
be08d6d2 | 2942 | new_ns->root = new; |
1a4eeaf2 | 2943 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2944 | |
2945 | /* | |
2946 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2947 | * as belonging to new namespace. We have already acquired a private | |
2948 | * fs_struct, so tsk->fs->lock is not needed. | |
2949 | */ | |
909b0a88 | 2950 | p = old; |
cb338d06 | 2951 | q = new; |
1da177e4 | 2952 | while (p) { |
143c8c91 | 2953 | q->mnt_ns = new_ns; |
d2921684 | 2954 | new_ns->mounts++; |
9559f689 AV |
2955 | if (new_fs) { |
2956 | if (&p->mnt == new_fs->root.mnt) { | |
2957 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2958 | rootmnt = &p->mnt; |
1da177e4 | 2959 | } |
9559f689 AV |
2960 | if (&p->mnt == new_fs->pwd.mnt) { |
2961 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2962 | pwdmnt = &p->mnt; |
1da177e4 | 2963 | } |
1da177e4 | 2964 | } |
909b0a88 AV |
2965 | p = next_mnt(p, old); |
2966 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2967 | if (!q) |
2968 | break; | |
2969 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2970 | p = next_mnt(p, old); | |
1da177e4 | 2971 | } |
328e6d90 | 2972 | namespace_unlock(); |
1da177e4 | 2973 | |
1da177e4 | 2974 | if (rootmnt) |
f03c6599 | 2975 | mntput(rootmnt); |
1da177e4 | 2976 | if (pwdmnt) |
f03c6599 | 2977 | mntput(pwdmnt); |
1da177e4 | 2978 | |
741a2951 | 2979 | return new_ns; |
1da177e4 LT |
2980 | } |
2981 | ||
cf8d2c11 TM |
2982 | /** |
2983 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2984 | * @mnt: pointer to the new root filesystem mountpoint | |
2985 | */ | |
1a4eeaf2 | 2986 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 2987 | { |
771b1371 | 2988 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 2989 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
2990 | struct mount *mnt = real_mount(m); |
2991 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 2992 | new_ns->root = mnt; |
d2921684 | 2993 | new_ns->mounts++; |
b1983cd8 | 2994 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 2995 | } else { |
1a4eeaf2 | 2996 | mntput(m); |
cf8d2c11 TM |
2997 | } |
2998 | return new_ns; | |
2999 | } | |
cf8d2c11 | 3000 | |
ea441d11 AV |
3001 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3002 | { | |
3003 | struct mnt_namespace *ns; | |
d31da0f0 | 3004 | struct super_block *s; |
ea441d11 AV |
3005 | struct path path; |
3006 | int err; | |
3007 | ||
3008 | ns = create_mnt_ns(mnt); | |
3009 | if (IS_ERR(ns)) | |
3010 | return ERR_CAST(ns); | |
3011 | ||
3012 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3013 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3014 | ||
3015 | put_mnt_ns(ns); | |
3016 | ||
3017 | if (err) | |
3018 | return ERR_PTR(err); | |
3019 | ||
3020 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3021 | s = path.mnt->mnt_sb; |
3022 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3023 | mntput(path.mnt); |
3024 | /* lock the sucker */ | |
d31da0f0 | 3025 | down_write(&s->s_umount); |
ea441d11 AV |
3026 | /* ... and return the root of (sub)tree on it */ |
3027 | return path.dentry; | |
3028 | } | |
3029 | EXPORT_SYMBOL(mount_subtree); | |
3030 | ||
bdc480e3 HC |
3031 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3032 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3033 | { |
eca6f534 VN |
3034 | int ret; |
3035 | char *kernel_type; | |
eca6f534 | 3036 | char *kernel_dev; |
b40ef869 | 3037 | void *options; |
1da177e4 | 3038 | |
b8850d1f TG |
3039 | kernel_type = copy_mount_string(type); |
3040 | ret = PTR_ERR(kernel_type); | |
3041 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3042 | goto out_type; |
1da177e4 | 3043 | |
b8850d1f TG |
3044 | kernel_dev = copy_mount_string(dev_name); |
3045 | ret = PTR_ERR(kernel_dev); | |
3046 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3047 | goto out_dev; |
1da177e4 | 3048 | |
b40ef869 AV |
3049 | options = copy_mount_options(data); |
3050 | ret = PTR_ERR(options); | |
3051 | if (IS_ERR(options)) | |
eca6f534 | 3052 | goto out_data; |
1da177e4 | 3053 | |
b40ef869 | 3054 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3055 | |
b40ef869 | 3056 | kfree(options); |
eca6f534 VN |
3057 | out_data: |
3058 | kfree(kernel_dev); | |
3059 | out_dev: | |
eca6f534 VN |
3060 | kfree(kernel_type); |
3061 | out_type: | |
3062 | return ret; | |
1da177e4 LT |
3063 | } |
3064 | ||
afac7cba AV |
3065 | /* |
3066 | * Return true if path is reachable from root | |
3067 | * | |
48a066e7 | 3068 | * namespace_sem or mount_lock is held |
afac7cba | 3069 | */ |
643822b4 | 3070 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3071 | const struct path *root) |
3072 | { | |
643822b4 | 3073 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3074 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3075 | mnt = mnt->mnt_parent; |
afac7cba | 3076 | } |
643822b4 | 3077 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3078 | } |
3079 | ||
640eb7e7 | 3080 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3081 | { |
25ab4c9b | 3082 | bool res; |
48a066e7 | 3083 | read_seqlock_excl(&mount_lock); |
643822b4 | 3084 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3085 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3086 | return res; |
3087 | } | |
3088 | EXPORT_SYMBOL(path_is_under); | |
3089 | ||
1da177e4 LT |
3090 | /* |
3091 | * pivot_root Semantics: | |
3092 | * Moves the root file system of the current process to the directory put_old, | |
3093 | * makes new_root as the new root file system of the current process, and sets | |
3094 | * root/cwd of all processes which had them on the current root to new_root. | |
3095 | * | |
3096 | * Restrictions: | |
3097 | * The new_root and put_old must be directories, and must not be on the | |
3098 | * same file system as the current process root. The put_old must be | |
3099 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3100 | * pointed to by put_old must yield the same directory as new_root. No other | |
3101 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3102 | * | |
4a0d11fa NB |
3103 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3104 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3105 | * in this situation. | |
3106 | * | |
1da177e4 LT |
3107 | * Notes: |
3108 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3109 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3110 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3111 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3112 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3113 | * first. | |
3114 | */ | |
3480b257 HC |
3115 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3116 | const char __user *, put_old) | |
1da177e4 | 3117 | { |
2d8f3038 | 3118 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3119 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3120 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3121 | int error; |
3122 | ||
9b40bc90 | 3123 | if (!may_mount()) |
1da177e4 LT |
3124 | return -EPERM; |
3125 | ||
2d8f3038 | 3126 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3127 | if (error) |
3128 | goto out0; | |
1da177e4 | 3129 | |
2d8f3038 | 3130 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3131 | if (error) |
3132 | goto out1; | |
3133 | ||
2d8f3038 | 3134 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3135 | if (error) |
3136 | goto out2; | |
1da177e4 | 3137 | |
f7ad3c6b | 3138 | get_fs_root(current->fs, &root); |
84d17192 AV |
3139 | old_mp = lock_mount(&old); |
3140 | error = PTR_ERR(old_mp); | |
3141 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3142 | goto out3; |
3143 | ||
1da177e4 | 3144 | error = -EINVAL; |
419148da AV |
3145 | new_mnt = real_mount(new.mnt); |
3146 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3147 | old_mnt = real_mount(old.mnt); |
3148 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3149 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3150 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3151 | goto out4; |
143c8c91 | 3152 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3153 | goto out4; |
5ff9d8a6 EB |
3154 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3155 | goto out4; | |
1da177e4 | 3156 | error = -ENOENT; |
f3da392e | 3157 | if (d_unlinked(new.dentry)) |
b12cea91 | 3158 | goto out4; |
1da177e4 | 3159 | error = -EBUSY; |
84d17192 | 3160 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3161 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3162 | error = -EINVAL; |
8c3ee42e | 3163 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3164 | goto out4; /* not a mountpoint */ |
676da58d | 3165 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3166 | goto out4; /* not attached */ |
84d17192 | 3167 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3168 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3169 | goto out4; /* not a mountpoint */ |
676da58d | 3170 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3171 | goto out4; /* not attached */ |
4ac91378 | 3172 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3173 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3174 | goto out4; |
0d082601 EB |
3175 | /* make certain new is below the root */ |
3176 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3177 | goto out4; | |
84d17192 | 3178 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3179 | lock_mount_hash(); |
419148da AV |
3180 | detach_mnt(new_mnt, &parent_path); |
3181 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3182 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3183 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3184 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3185 | } | |
4ac91378 | 3186 | /* mount old root on put_old */ |
84d17192 | 3187 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3188 | /* mount new_root on / */ |
84d17192 | 3189 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3190 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3191 | /* A moved mount should not expire automatically */ |
3192 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3193 | put_mountpoint(root_mp); |
719ea2fb | 3194 | unlock_mount_hash(); |
2d8f3038 | 3195 | chroot_fs_refs(&root, &new); |
1da177e4 | 3196 | error = 0; |
b12cea91 | 3197 | out4: |
84d17192 | 3198 | unlock_mount(old_mp); |
b12cea91 AV |
3199 | if (!error) { |
3200 | path_put(&root_parent); | |
3201 | path_put(&parent_path); | |
3202 | } | |
3203 | out3: | |
8c3ee42e | 3204 | path_put(&root); |
b12cea91 | 3205 | out2: |
2d8f3038 | 3206 | path_put(&old); |
1da177e4 | 3207 | out1: |
2d8f3038 | 3208 | path_put(&new); |
1da177e4 | 3209 | out0: |
1da177e4 | 3210 | return error; |
1da177e4 LT |
3211 | } |
3212 | ||
3213 | static void __init init_mount_tree(void) | |
3214 | { | |
3215 | struct vfsmount *mnt; | |
6b3286ed | 3216 | struct mnt_namespace *ns; |
ac748a09 | 3217 | struct path root; |
0c55cfc4 | 3218 | struct file_system_type *type; |
1da177e4 | 3219 | |
0c55cfc4 EB |
3220 | type = get_fs_type("rootfs"); |
3221 | if (!type) | |
3222 | panic("Can't find rootfs type"); | |
3223 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3224 | put_filesystem(type); | |
1da177e4 LT |
3225 | if (IS_ERR(mnt)) |
3226 | panic("Can't create rootfs"); | |
b3e19d92 | 3227 | |
3b22edc5 TM |
3228 | ns = create_mnt_ns(mnt); |
3229 | if (IS_ERR(ns)) | |
1da177e4 | 3230 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3231 | |
3232 | init_task.nsproxy->mnt_ns = ns; | |
3233 | get_mnt_ns(ns); | |
3234 | ||
be08d6d2 AV |
3235 | root.mnt = mnt; |
3236 | root.dentry = mnt->mnt_root; | |
da362b09 | 3237 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3238 | |
3239 | set_fs_pwd(current->fs, &root); | |
3240 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3241 | } |
3242 | ||
74bf17cf | 3243 | void __init mnt_init(void) |
1da177e4 | 3244 | { |
15a67dd8 | 3245 | int err; |
1da177e4 | 3246 | |
7d6fec45 | 3247 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3248 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3249 | |
0818bf27 | 3250 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3251 | sizeof(struct hlist_head), |
0818bf27 | 3252 | mhash_entries, 19, |
3d375d78 | 3253 | HASH_ZERO, |
0818bf27 AV |
3254 | &m_hash_shift, &m_hash_mask, 0, 0); |
3255 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3256 | sizeof(struct hlist_head), | |
3257 | mphash_entries, 19, | |
3d375d78 | 3258 | HASH_ZERO, |
0818bf27 | 3259 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3260 | |
84d17192 | 3261 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3262 | panic("Failed to allocate mount hash table\n"); |
3263 | ||
4b93dc9b TH |
3264 | kernfs_init(); |
3265 | ||
15a67dd8 RD |
3266 | err = sysfs_init(); |
3267 | if (err) | |
3268 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3269 | __func__, err); |
00d26666 GKH |
3270 | fs_kobj = kobject_create_and_add("fs", NULL); |
3271 | if (!fs_kobj) | |
8e24eea7 | 3272 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3273 | init_rootfs(); |
3274 | init_mount_tree(); | |
3275 | } | |
3276 | ||
616511d0 | 3277 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3278 | { |
d498b25a | 3279 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3280 | return; |
7b00ed6f | 3281 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3282 | free_mnt_ns(ns); |
1da177e4 | 3283 | } |
9d412a43 AV |
3284 | |
3285 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3286 | { | |
423e0ab0 | 3287 | struct vfsmount *mnt; |
e462ec50 | 3288 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); |
423e0ab0 TC |
3289 | if (!IS_ERR(mnt)) { |
3290 | /* | |
3291 | * it is a longterm mount, don't release mnt until | |
3292 | * we unmount before file sys is unregistered | |
3293 | */ | |
f7a99c5b | 3294 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3295 | } |
3296 | return mnt; | |
9d412a43 AV |
3297 | } |
3298 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3299 | |
3300 | void kern_unmount(struct vfsmount *mnt) | |
3301 | { | |
3302 | /* release long term mount so mount point can be released */ | |
3303 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3304 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3305 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3306 | mntput(mnt); |
3307 | } | |
3308 | } | |
3309 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3310 | |
3311 | bool our_mnt(struct vfsmount *mnt) | |
3312 | { | |
143c8c91 | 3313 | return check_mnt(real_mount(mnt)); |
02125a82 | 3314 | } |
8823c079 | 3315 | |
3151527e EB |
3316 | bool current_chrooted(void) |
3317 | { | |
3318 | /* Does the current process have a non-standard root */ | |
3319 | struct path ns_root; | |
3320 | struct path fs_root; | |
3321 | bool chrooted; | |
3322 | ||
3323 | /* Find the namespace root */ | |
3324 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3325 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3326 | path_get(&ns_root); | |
3327 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3328 | ; | |
3329 | ||
3330 | get_fs_root(current->fs, &fs_root); | |
3331 | ||
3332 | chrooted = !path_equal(&fs_root, &ns_root); | |
3333 | ||
3334 | path_put(&fs_root); | |
3335 | path_put(&ns_root); | |
3336 | ||
3337 | return chrooted; | |
3338 | } | |
3339 | ||
8654df4e EB |
3340 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3341 | int *new_mnt_flags) | |
87a8ebd6 | 3342 | { |
8c6cf9cc | 3343 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3344 | struct mount *mnt; |
e51db735 | 3345 | bool visible = false; |
87a8ebd6 | 3346 | |
44bb4385 | 3347 | down_read(&namespace_sem); |
87a8ebd6 | 3348 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3349 | struct mount *child; |
77b1a97d EB |
3350 | int mnt_flags; |
3351 | ||
8654df4e | 3352 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3353 | continue; |
3354 | ||
7e96c1b0 EB |
3355 | /* This mount is not fully visible if it's root directory |
3356 | * is not the root directory of the filesystem. | |
3357 | */ | |
3358 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3359 | continue; | |
3360 | ||
a1935c17 | 3361 | /* A local view of the mount flags */ |
77b1a97d | 3362 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3363 | |
695e9df0 | 3364 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3365 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3366 | mnt_flags |= MNT_LOCK_READONLY; |
3367 | ||
8c6cf9cc EB |
3368 | /* Verify the mount flags are equal to or more permissive |
3369 | * than the proposed new mount. | |
3370 | */ | |
77b1a97d | 3371 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3372 | !(new_flags & MNT_READONLY)) |
3373 | continue; | |
77b1a97d EB |
3374 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3375 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3376 | continue; |
3377 | ||
ceeb0e5d EB |
3378 | /* This mount is not fully visible if there are any |
3379 | * locked child mounts that cover anything except for | |
3380 | * empty directories. | |
e51db735 EB |
3381 | */ |
3382 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3383 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3384 | /* Only worry about locked mounts */ |
d71ed6c9 | 3385 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3386 | continue; |
7236c85e EB |
3387 | /* Is the directory permanetly empty? */ |
3388 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3389 | goto next; |
87a8ebd6 | 3390 | } |
8c6cf9cc | 3391 | /* Preserve the locked attributes */ |
77b1a97d | 3392 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3393 | MNT_LOCK_ATIME); |
e51db735 EB |
3394 | visible = true; |
3395 | goto found; | |
3396 | next: ; | |
87a8ebd6 | 3397 | } |
e51db735 | 3398 | found: |
44bb4385 | 3399 | up_read(&namespace_sem); |
e51db735 | 3400 | return visible; |
87a8ebd6 EB |
3401 | } |
3402 | ||
8654df4e EB |
3403 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3404 | { | |
a1935c17 | 3405 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3406 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3407 | unsigned long s_iflags; | |
3408 | ||
3409 | if (ns->user_ns == &init_user_ns) | |
3410 | return false; | |
3411 | ||
3412 | /* Can this filesystem be too revealing? */ | |
3413 | s_iflags = mnt->mnt_sb->s_iflags; | |
3414 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3415 | return false; | |
3416 | ||
a1935c17 EB |
3417 | if ((s_iflags & required_iflags) != required_iflags) { |
3418 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3419 | required_iflags); | |
3420 | return true; | |
3421 | } | |
3422 | ||
8654df4e EB |
3423 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3424 | } | |
3425 | ||
380cf5ba AL |
3426 | bool mnt_may_suid(struct vfsmount *mnt) |
3427 | { | |
3428 | /* | |
3429 | * Foreign mounts (accessed via fchdir or through /proc | |
3430 | * symlinks) are always treated as if they are nosuid. This | |
3431 | * prevents namespaces from trusting potentially unsafe | |
3432 | * suid/sgid bits, file caps, or security labels that originate | |
3433 | * in other namespaces. | |
3434 | */ | |
3435 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3436 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3437 | } | |
3438 | ||
64964528 | 3439 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3440 | { |
58be2825 | 3441 | struct ns_common *ns = NULL; |
8823c079 EB |
3442 | struct nsproxy *nsproxy; |
3443 | ||
728dba3a EB |
3444 | task_lock(task); |
3445 | nsproxy = task->nsproxy; | |
8823c079 | 3446 | if (nsproxy) { |
58be2825 AV |
3447 | ns = &nsproxy->mnt_ns->ns; |
3448 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3449 | } |
728dba3a | 3450 | task_unlock(task); |
8823c079 EB |
3451 | |
3452 | return ns; | |
3453 | } | |
3454 | ||
64964528 | 3455 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3456 | { |
58be2825 | 3457 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3458 | } |
3459 | ||
64964528 | 3460 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3461 | { |
3462 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3463 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3464 | struct path root; |
4f757f3c | 3465 | int err; |
8823c079 | 3466 | |
0c55cfc4 | 3467 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3468 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3469 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3470 | return -EPERM; |
8823c079 EB |
3471 | |
3472 | if (fs->users != 1) | |
3473 | return -EINVAL; | |
3474 | ||
3475 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3476 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3477 | nsproxy->mnt_ns = mnt_ns; |
3478 | ||
3479 | /* Find the root */ | |
4f757f3c AV |
3480 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3481 | "/", LOOKUP_DOWN, &root); | |
3482 | if (err) { | |
3483 | /* revert to old namespace */ | |
3484 | nsproxy->mnt_ns = old_mnt_ns; | |
3485 | put_mnt_ns(mnt_ns); | |
3486 | return err; | |
3487 | } | |
8823c079 | 3488 | |
4068367c AV |
3489 | put_mnt_ns(old_mnt_ns); |
3490 | ||
8823c079 EB |
3491 | /* Update the pwd and root */ |
3492 | set_fs_pwd(fs, &root); | |
3493 | set_fs_root(fs, &root); | |
3494 | ||
3495 | path_put(&root); | |
3496 | return 0; | |
3497 | } | |
3498 | ||
bcac25a5 AV |
3499 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3500 | { | |
3501 | return to_mnt_ns(ns)->user_ns; | |
3502 | } | |
3503 | ||
8823c079 EB |
3504 | const struct proc_ns_operations mntns_operations = { |
3505 | .name = "mnt", | |
3506 | .type = CLONE_NEWNS, | |
3507 | .get = mntns_get, | |
3508 | .put = mntns_put, | |
3509 | .install = mntns_install, | |
bcac25a5 | 3510 | .owner = mntns_owner, |
8823c079 | 3511 | }; |