]>
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(); | |
6aa7de05 | 356 | while (READ_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); | |
954c736f AG |
471 | if (upperdentry && |
472 | (file_inode(file) == d_inode(upperdentry) || | |
473 | file_inode(file) == d_inode(dentry))) | |
7c6893e3 MS |
474 | return 0; |
475 | ||
476 | /* Lower layer: can't write to real file, sorry... */ | |
477 | return -EPERM; | |
478 | } | |
479 | ||
480 | /** | |
481 | * mnt_want_write_file - get write access to a file's mount | |
482 | * @file: the file who's mount on which to take a write | |
483 | * | |
484 | * This is like mnt_want_write, but it takes a file and can | |
485 | * do some optimisations if the file is open for write already | |
486 | * | |
487 | * Mostly called by filesystems from their ioctl operation before performing | |
488 | * modification. On overlayfs this needs to check if the file is on a read-only | |
489 | * lower layer and deny access in that case. | |
490 | */ | |
491 | int mnt_want_write_file(struct file *file) | |
492 | { | |
493 | int ret; | |
494 | ||
495 | ret = may_write_real(file); | |
496 | if (!ret) { | |
497 | sb_start_write(file_inode(file)->i_sb); | |
498 | ret = __mnt_want_write_file(file); | |
499 | if (ret) | |
500 | sb_end_write(file_inode(file)->i_sb); | |
501 | } | |
502 | return ret; | |
503 | } | |
96029c4e | 504 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
505 | ||
8366025e | 506 | /** |
eb04c282 | 507 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
508 | * @mnt: the mount on which to give up write access |
509 | * | |
510 | * Tells the low-level filesystem that we are done | |
511 | * performing writes to it. Must be matched with | |
eb04c282 | 512 | * __mnt_want_write() call above. |
8366025e | 513 | */ |
eb04c282 | 514 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 515 | { |
d3ef3d73 | 516 | preempt_disable(); |
83adc753 | 517 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 518 | preempt_enable(); |
8366025e | 519 | } |
c088e31d | 520 | EXPORT_SYMBOL_GPL(__mnt_drop_write); |
eb04c282 JK |
521 | |
522 | /** | |
523 | * mnt_drop_write - give up write access to a mount | |
524 | * @mnt: the mount on which to give up write access | |
525 | * | |
526 | * Tells the low-level filesystem that we are done performing writes to it and | |
527 | * also allows filesystem to be frozen again. Must be matched with | |
528 | * mnt_want_write() call above. | |
529 | */ | |
530 | void mnt_drop_write(struct vfsmount *mnt) | |
531 | { | |
532 | __mnt_drop_write(mnt); | |
533 | sb_end_write(mnt->mnt_sb); | |
534 | } | |
8366025e DH |
535 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
536 | ||
eb04c282 JK |
537 | void __mnt_drop_write_file(struct file *file) |
538 | { | |
539 | __mnt_drop_write(file->f_path.mnt); | |
540 | } | |
541 | ||
7c6893e3 | 542 | void mnt_drop_write_file_path(struct file *file) |
2a79f17e AV |
543 | { |
544 | mnt_drop_write(file->f_path.mnt); | |
545 | } | |
7c6893e3 MS |
546 | |
547 | void mnt_drop_write_file(struct file *file) | |
548 | { | |
549 | __mnt_drop_write(file->f_path.mnt); | |
550 | sb_end_write(file_inode(file)->i_sb); | |
551 | } | |
2a79f17e AV |
552 | EXPORT_SYMBOL(mnt_drop_write_file); |
553 | ||
83adc753 | 554 | static int mnt_make_readonly(struct mount *mnt) |
8366025e | 555 | { |
3d733633 DH |
556 | int ret = 0; |
557 | ||
719ea2fb | 558 | lock_mount_hash(); |
83adc753 | 559 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 560 | /* |
d3ef3d73 | 561 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
562 | * should be visible before we do. | |
3d733633 | 563 | */ |
d3ef3d73 | 564 | smp_mb(); |
565 | ||
3d733633 | 566 | /* |
d3ef3d73 | 567 | * With writers on hold, if this value is zero, then there are |
568 | * definitely no active writers (although held writers may subsequently | |
569 | * increment the count, they'll have to wait, and decrement it after | |
570 | * seeing MNT_READONLY). | |
571 | * | |
572 | * It is OK to have counter incremented on one CPU and decremented on | |
573 | * another: the sum will add up correctly. The danger would be when we | |
574 | * sum up each counter, if we read a counter before it is incremented, | |
575 | * but then read another CPU's count which it has been subsequently | |
576 | * decremented from -- we would see more decrements than we should. | |
577 | * MNT_WRITE_HOLD protects against this scenario, because | |
578 | * mnt_want_write first increments count, then smp_mb, then spins on | |
579 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
580 | * we're counting up here. | |
3d733633 | 581 | */ |
c6653a83 | 582 | if (mnt_get_writers(mnt) > 0) |
d3ef3d73 | 583 | ret = -EBUSY; |
584 | else | |
83adc753 | 585 | mnt->mnt.mnt_flags |= MNT_READONLY; |
d3ef3d73 | 586 | /* |
587 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
588 | * that become unheld will see MNT_READONLY. | |
589 | */ | |
590 | smp_wmb(); | |
83adc753 | 591 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
719ea2fb | 592 | unlock_mount_hash(); |
3d733633 | 593 | return ret; |
8366025e | 594 | } |
8366025e | 595 | |
83adc753 | 596 | static void __mnt_unmake_readonly(struct mount *mnt) |
2e4b7fcd | 597 | { |
719ea2fb | 598 | lock_mount_hash(); |
83adc753 | 599 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
719ea2fb | 600 | unlock_mount_hash(); |
2e4b7fcd DH |
601 | } |
602 | ||
4ed5e82f MS |
603 | int sb_prepare_remount_readonly(struct super_block *sb) |
604 | { | |
605 | struct mount *mnt; | |
606 | int err = 0; | |
607 | ||
8e8b8796 MS |
608 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
609 | if (atomic_long_read(&sb->s_remove_count)) | |
610 | return -EBUSY; | |
611 | ||
719ea2fb | 612 | lock_mount_hash(); |
4ed5e82f MS |
613 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
614 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
615 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; | |
616 | smp_mb(); | |
617 | if (mnt_get_writers(mnt) > 0) { | |
618 | err = -EBUSY; | |
619 | break; | |
620 | } | |
621 | } | |
622 | } | |
8e8b8796 MS |
623 | if (!err && atomic_long_read(&sb->s_remove_count)) |
624 | err = -EBUSY; | |
625 | ||
4ed5e82f MS |
626 | if (!err) { |
627 | sb->s_readonly_remount = 1; | |
628 | smp_wmb(); | |
629 | } | |
630 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
631 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
632 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
633 | } | |
719ea2fb | 634 | unlock_mount_hash(); |
4ed5e82f MS |
635 | |
636 | return err; | |
637 | } | |
638 | ||
b105e270 | 639 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 640 | { |
fcc139ae | 641 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 642 | #ifdef CONFIG_SMP |
68e8a9fe | 643 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 644 | #endif |
b105e270 | 645 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
646 | } |
647 | ||
8ffcb32e DH |
648 | static void delayed_free_vfsmnt(struct rcu_head *head) |
649 | { | |
650 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
651 | } | |
652 | ||
48a066e7 | 653 | /* call under rcu_read_lock */ |
294d71ff | 654 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
655 | { |
656 | struct mount *mnt; | |
657 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 658 | return 1; |
48a066e7 | 659 | if (bastard == NULL) |
294d71ff | 660 | return 0; |
48a066e7 AV |
661 | mnt = real_mount(bastard); |
662 | mnt_add_count(mnt, 1); | |
663 | if (likely(!read_seqretry(&mount_lock, seq))) | |
294d71ff | 664 | return 0; |
48a066e7 AV |
665 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
666 | mnt_add_count(mnt, -1); | |
294d71ff AV |
667 | return 1; |
668 | } | |
669 | return -1; | |
670 | } | |
671 | ||
672 | /* call under rcu_read_lock */ | |
673 | bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) | |
674 | { | |
675 | int res = __legitimize_mnt(bastard, seq); | |
676 | if (likely(!res)) | |
677 | return true; | |
678 | if (unlikely(res < 0)) { | |
679 | rcu_read_unlock(); | |
680 | mntput(bastard); | |
681 | rcu_read_lock(); | |
48a066e7 | 682 | } |
48a066e7 AV |
683 | return false; |
684 | } | |
685 | ||
1da177e4 | 686 | /* |
474279dc | 687 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 688 | * call under rcu_read_lock() |
1da177e4 | 689 | */ |
474279dc | 690 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 691 | { |
38129a13 | 692 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
693 | struct mount *p; |
694 | ||
38129a13 | 695 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
696 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
697 | return p; | |
698 | return NULL; | |
699 | } | |
700 | ||
a05964f3 | 701 | /* |
f015f126 DH |
702 | * lookup_mnt - Return the first child mount mounted at path |
703 | * | |
704 | * "First" means first mounted chronologically. If you create the | |
705 | * following mounts: | |
706 | * | |
707 | * mount /dev/sda1 /mnt | |
708 | * mount /dev/sda2 /mnt | |
709 | * mount /dev/sda3 /mnt | |
710 | * | |
711 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
712 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
713 | * /dev/sda2, then /dev/sda3, then NULL. | |
714 | * | |
715 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 716 | */ |
ca71cf71 | 717 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 718 | { |
c7105365 | 719 | struct mount *child_mnt; |
48a066e7 AV |
720 | struct vfsmount *m; |
721 | unsigned seq; | |
99b7db7b | 722 | |
48a066e7 AV |
723 | rcu_read_lock(); |
724 | do { | |
725 | seq = read_seqbegin(&mount_lock); | |
726 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
727 | m = child_mnt ? &child_mnt->mnt : NULL; | |
728 | } while (!legitimize_mnt(m, seq)); | |
729 | rcu_read_unlock(); | |
730 | return m; | |
a05964f3 RP |
731 | } |
732 | ||
7af1364f EB |
733 | /* |
734 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
735 | * current mount namespace. | |
736 | * | |
737 | * The common case is dentries are not mountpoints at all and that | |
738 | * test is handled inline. For the slow case when we are actually | |
739 | * dealing with a mountpoint of some kind, walk through all of the | |
740 | * mounts in the current mount namespace and test to see if the dentry | |
741 | * is a mountpoint. | |
742 | * | |
743 | * The mount_hashtable is not usable in the context because we | |
744 | * need to identify all mounts that may be in the current mount | |
745 | * namespace not just a mount that happens to have some specified | |
746 | * parent mount. | |
747 | */ | |
748 | bool __is_local_mountpoint(struct dentry *dentry) | |
749 | { | |
750 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
751 | struct mount *mnt; | |
752 | bool is_covered = false; | |
753 | ||
754 | if (!d_mountpoint(dentry)) | |
755 | goto out; | |
756 | ||
757 | down_read(&namespace_sem); | |
758 | list_for_each_entry(mnt, &ns->list, mnt_list) { | |
759 | is_covered = (mnt->mnt_mountpoint == dentry); | |
760 | if (is_covered) | |
761 | break; | |
762 | } | |
763 | up_read(&namespace_sem); | |
764 | out: | |
765 | return is_covered; | |
766 | } | |
767 | ||
e2dfa935 | 768 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 769 | { |
0818bf27 | 770 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
771 | struct mountpoint *mp; |
772 | ||
0818bf27 | 773 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 AV |
774 | if (mp->m_dentry == dentry) { |
775 | /* might be worth a WARN_ON() */ | |
776 | if (d_unlinked(dentry)) | |
777 | return ERR_PTR(-ENOENT); | |
778 | mp->m_count++; | |
779 | return mp; | |
780 | } | |
781 | } | |
e2dfa935 EB |
782 | return NULL; |
783 | } | |
784 | ||
3895dbf8 | 785 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 786 | { |
3895dbf8 | 787 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 788 | int ret; |
84d17192 | 789 | |
3895dbf8 EB |
790 | if (d_mountpoint(dentry)) { |
791 | mountpoint: | |
792 | read_seqlock_excl(&mount_lock); | |
793 | mp = lookup_mountpoint(dentry); | |
794 | read_sequnlock_excl(&mount_lock); | |
795 | if (mp) | |
796 | goto done; | |
797 | } | |
798 | ||
799 | if (!new) | |
800 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
801 | if (!new) | |
84d17192 AV |
802 | return ERR_PTR(-ENOMEM); |
803 | ||
3895dbf8 EB |
804 | |
805 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 806 | ret = d_set_mounted(dentry); |
eed81007 | 807 | |
3895dbf8 EB |
808 | /* Someone else set d_mounted? */ |
809 | if (ret == -EBUSY) | |
810 | goto mountpoint; | |
811 | ||
812 | /* The dentry is not available as a mountpoint? */ | |
813 | mp = ERR_PTR(ret); | |
814 | if (ret) | |
815 | goto done; | |
816 | ||
817 | /* Add the new mountpoint to the hash table */ | |
818 | read_seqlock_excl(&mount_lock); | |
819 | new->m_dentry = dentry; | |
820 | new->m_count = 1; | |
821 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
822 | INIT_HLIST_HEAD(&new->m_list); | |
823 | read_sequnlock_excl(&mount_lock); | |
824 | ||
825 | mp = new; | |
826 | new = NULL; | |
827 | done: | |
828 | kfree(new); | |
84d17192 AV |
829 | return mp; |
830 | } | |
831 | ||
832 | static void put_mountpoint(struct mountpoint *mp) | |
833 | { | |
834 | if (!--mp->m_count) { | |
835 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 836 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
837 | spin_lock(&dentry->d_lock); |
838 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
839 | spin_unlock(&dentry->d_lock); | |
0818bf27 | 840 | hlist_del(&mp->m_hash); |
84d17192 AV |
841 | kfree(mp); |
842 | } | |
843 | } | |
844 | ||
143c8c91 | 845 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 846 | { |
6b3286ed | 847 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
848 | } |
849 | ||
c088e31d SF |
850 | /* for aufs, CONFIG_AUFS_BR_FUSE */ |
851 | int is_current_mnt_ns(struct vfsmount *mnt) | |
852 | { | |
853 | return check_mnt(real_mount(mnt)); | |
854 | } | |
855 | EXPORT_SYMBOL_GPL(is_current_mnt_ns); | |
856 | ||
99b7db7b NP |
857 | /* |
858 | * vfsmount lock must be held for write | |
859 | */ | |
6b3286ed | 860 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
861 | { |
862 | if (ns) { | |
863 | ns->event = ++event; | |
864 | wake_up_interruptible(&ns->poll); | |
865 | } | |
866 | } | |
867 | ||
99b7db7b NP |
868 | /* |
869 | * vfsmount lock must be held for write | |
870 | */ | |
6b3286ed | 871 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
872 | { |
873 | if (ns && ns->event != event) { | |
874 | ns->event = event; | |
875 | wake_up_interruptible(&ns->poll); | |
876 | } | |
877 | } | |
878 | ||
99b7db7b NP |
879 | /* |
880 | * vfsmount lock must be held for write | |
881 | */ | |
7bdb11de | 882 | static void unhash_mnt(struct mount *mnt) |
419148da | 883 | { |
0714a533 | 884 | mnt->mnt_parent = mnt; |
a73324da | 885 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 886 | list_del_init(&mnt->mnt_child); |
38129a13 | 887 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 888 | hlist_del_init(&mnt->mnt_mp_list); |
84d17192 AV |
889 | put_mountpoint(mnt->mnt_mp); |
890 | mnt->mnt_mp = NULL; | |
1da177e4 LT |
891 | } |
892 | ||
7bdb11de EB |
893 | /* |
894 | * vfsmount lock must be held for write | |
895 | */ | |
896 | static void detach_mnt(struct mount *mnt, struct path *old_path) | |
897 | { | |
898 | old_path->dentry = mnt->mnt_mountpoint; | |
899 | old_path->mnt = &mnt->mnt_parent->mnt; | |
900 | unhash_mnt(mnt); | |
901 | } | |
902 | ||
6a46c573 EB |
903 | /* |
904 | * vfsmount lock must be held for write | |
905 | */ | |
906 | static void umount_mnt(struct mount *mnt) | |
907 | { | |
908 | /* old mountpoint will be dropped when we can do that */ | |
909 | mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; | |
910 | unhash_mnt(mnt); | |
911 | } | |
912 | ||
99b7db7b NP |
913 | /* |
914 | * vfsmount lock must be held for write | |
915 | */ | |
84d17192 AV |
916 | void mnt_set_mountpoint(struct mount *mnt, |
917 | struct mountpoint *mp, | |
44d964d6 | 918 | struct mount *child_mnt) |
b90fa9ae | 919 | { |
84d17192 | 920 | mp->m_count++; |
3a2393d7 | 921 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
84d17192 | 922 | child_mnt->mnt_mountpoint = dget(mp->m_dentry); |
3a2393d7 | 923 | child_mnt->mnt_parent = mnt; |
84d17192 | 924 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 925 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
926 | } |
927 | ||
1064f874 EB |
928 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
929 | { | |
930 | hlist_add_head_rcu(&mnt->mnt_hash, | |
931 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
932 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
933 | } | |
934 | ||
99b7db7b NP |
935 | /* |
936 | * vfsmount lock must be held for write | |
937 | */ | |
84d17192 AV |
938 | static void attach_mnt(struct mount *mnt, |
939 | struct mount *parent, | |
940 | struct mountpoint *mp) | |
1da177e4 | 941 | { |
84d17192 | 942 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 943 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
944 | } |
945 | ||
1064f874 | 946 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 947 | { |
1064f874 EB |
948 | struct mountpoint *old_mp = mnt->mnt_mp; |
949 | struct dentry *old_mountpoint = mnt->mnt_mountpoint; | |
950 | struct mount *old_parent = mnt->mnt_parent; | |
951 | ||
952 | list_del_init(&mnt->mnt_child); | |
953 | hlist_del_init(&mnt->mnt_mp_list); | |
954 | hlist_del_init_rcu(&mnt->mnt_hash); | |
955 | ||
956 | attach_mnt(mnt, parent, mp); | |
957 | ||
958 | put_mountpoint(old_mp); | |
959 | ||
960 | /* | |
961 | * Safely avoid even the suggestion this code might sleep or | |
962 | * lock the mount hash by taking advantage of the knowledge that | |
963 | * mnt_change_mountpoint will not release the final reference | |
964 | * to a mountpoint. | |
965 | * | |
966 | * During mounting, the mount passed in as the parent mount will | |
967 | * continue to use the old mountpoint and during unmounting, the | |
968 | * old mountpoint will continue to exist until namespace_unlock, | |
969 | * which happens well after mnt_change_mountpoint. | |
970 | */ | |
971 | spin_lock(&old_mountpoint->d_lock); | |
972 | old_mountpoint->d_lockref.count--; | |
973 | spin_unlock(&old_mountpoint->d_lock); | |
974 | ||
975 | mnt_add_count(old_parent, -1); | |
12a5b529 AV |
976 | } |
977 | ||
b90fa9ae | 978 | /* |
99b7db7b | 979 | * vfsmount lock must be held for write |
b90fa9ae | 980 | */ |
1064f874 | 981 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 982 | { |
0714a533 | 983 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 984 | struct mount *m; |
b90fa9ae | 985 | LIST_HEAD(head); |
143c8c91 | 986 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 987 | |
0714a533 | 988 | BUG_ON(parent == mnt); |
b90fa9ae | 989 | |
1a4eeaf2 | 990 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 991 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 992 | m->mnt_ns = n; |
f03c6599 | 993 | |
b90fa9ae RP |
994 | list_splice(&head, n->list.prev); |
995 | ||
d2921684 EB |
996 | n->mounts += n->pending_mounts; |
997 | n->pending_mounts = 0; | |
998 | ||
1064f874 | 999 | __attach_mnt(mnt, parent); |
6b3286ed | 1000 | touch_mnt_namespace(n); |
1da177e4 LT |
1001 | } |
1002 | ||
909b0a88 | 1003 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 1004 | { |
6b41d536 AV |
1005 | struct list_head *next = p->mnt_mounts.next; |
1006 | if (next == &p->mnt_mounts) { | |
1da177e4 | 1007 | while (1) { |
909b0a88 | 1008 | if (p == root) |
1da177e4 | 1009 | return NULL; |
6b41d536 AV |
1010 | next = p->mnt_child.next; |
1011 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 1012 | break; |
0714a533 | 1013 | p = p->mnt_parent; |
1da177e4 LT |
1014 | } |
1015 | } | |
6b41d536 | 1016 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
1017 | } |
1018 | ||
315fc83e | 1019 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 1020 | { |
6b41d536 AV |
1021 | struct list_head *prev = p->mnt_mounts.prev; |
1022 | while (prev != &p->mnt_mounts) { | |
1023 | p = list_entry(prev, struct mount, mnt_child); | |
1024 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
1025 | } |
1026 | return p; | |
1027 | } | |
1028 | ||
9d412a43 AV |
1029 | struct vfsmount * |
1030 | vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) | |
1031 | { | |
b105e270 | 1032 | struct mount *mnt; |
9d412a43 AV |
1033 | struct dentry *root; |
1034 | ||
1035 | if (!type) | |
1036 | return ERR_PTR(-ENODEV); | |
1037 | ||
1038 | mnt = alloc_vfsmnt(name); | |
1039 | if (!mnt) | |
1040 | return ERR_PTR(-ENOMEM); | |
1041 | ||
e462ec50 | 1042 | if (flags & SB_KERNMOUNT) |
b105e270 | 1043 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 AV |
1044 | |
1045 | root = mount_fs(type, flags, name, data); | |
1046 | if (IS_ERR(root)) { | |
8ffcb32e | 1047 | mnt_free_id(mnt); |
9d412a43 AV |
1048 | free_vfsmnt(mnt); |
1049 | return ERR_CAST(root); | |
1050 | } | |
1051 | ||
b105e270 AV |
1052 | mnt->mnt.mnt_root = root; |
1053 | mnt->mnt.mnt_sb = root->d_sb; | |
a73324da | 1054 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
0714a533 | 1055 | mnt->mnt_parent = mnt; |
719ea2fb | 1056 | lock_mount_hash(); |
39f7c4db | 1057 | list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); |
719ea2fb | 1058 | unlock_mount_hash(); |
b105e270 | 1059 | return &mnt->mnt; |
9d412a43 AV |
1060 | } |
1061 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1062 | ||
93faccbb EB |
1063 | struct vfsmount * |
1064 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1065 | const char *name, void *data) | |
1066 | { | |
1067 | /* Until it is worked out how to pass the user namespace | |
1068 | * through from the parent mount to the submount don't support | |
1069 | * unprivileged mounts with submounts. | |
1070 | */ | |
1071 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1072 | return ERR_PTR(-EPERM); | |
1073 | ||
e462ec50 | 1074 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1075 | } |
1076 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1077 | ||
87129cc0 | 1078 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1079 | int flag) |
1da177e4 | 1080 | { |
87129cc0 | 1081 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1082 | struct mount *mnt; |
1083 | int err; | |
1da177e4 | 1084 | |
be34d1a3 DH |
1085 | mnt = alloc_vfsmnt(old->mnt_devname); |
1086 | if (!mnt) | |
1087 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1088 | |
7a472ef4 | 1089 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1090 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1091 | else | |
1092 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1093 | |
be34d1a3 DH |
1094 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1095 | err = mnt_alloc_group_id(mnt); | |
1096 | if (err) | |
1097 | goto out_free; | |
1da177e4 | 1098 | } |
be34d1a3 | 1099 | |
5759c7e2 AV |
1100 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1101 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
132c94e3 | 1102 | /* Don't allow unprivileged users to change mount flags */ |
9566d674 EB |
1103 | if (flag & CL_UNPRIVILEGED) { |
1104 | mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; | |
1105 | ||
1106 | if (mnt->mnt.mnt_flags & MNT_READONLY) | |
1107 | mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; | |
1108 | ||
1109 | if (mnt->mnt.mnt_flags & MNT_NODEV) | |
1110 | mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; | |
1111 | ||
1112 | if (mnt->mnt.mnt_flags & MNT_NOSUID) | |
1113 | mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; | |
1114 | ||
1115 | if (mnt->mnt.mnt_flags & MNT_NOEXEC) | |
1116 | mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; | |
1117 | } | |
132c94e3 | 1118 | |
5ff9d8a6 | 1119 | /* Don't allow unprivileged users to reveal what is under a mount */ |
381cacb1 EB |
1120 | if ((flag & CL_UNPRIVILEGED) && |
1121 | (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) | |
5ff9d8a6 EB |
1122 | mnt->mnt.mnt_flags |= MNT_LOCKED; |
1123 | ||
be34d1a3 DH |
1124 | atomic_inc(&sb->s_active); |
1125 | mnt->mnt.mnt_sb = sb; | |
1126 | mnt->mnt.mnt_root = dget(root); | |
1127 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1128 | mnt->mnt_parent = mnt; | |
719ea2fb | 1129 | lock_mount_hash(); |
be34d1a3 | 1130 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1131 | unlock_mount_hash(); |
be34d1a3 | 1132 | |
7a472ef4 EB |
1133 | if ((flag & CL_SLAVE) || |
1134 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1135 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1136 | mnt->mnt_master = old; | |
1137 | CLEAR_MNT_SHARED(mnt); | |
1138 | } else if (!(flag & CL_PRIVATE)) { | |
1139 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1140 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1141 | if (IS_MNT_SLAVE(old)) | |
1142 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1143 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1144 | } else { |
1145 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1146 | } |
1147 | if (flag & CL_MAKE_SHARED) | |
1148 | set_mnt_shared(mnt); | |
1149 | ||
1150 | /* stick the duplicate mount on the same expiry list | |
1151 | * as the original if that was on one */ | |
1152 | if (flag & CL_EXPIRE) { | |
1153 | if (!list_empty(&old->mnt_expire)) | |
1154 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1155 | } | |
1156 | ||
cb338d06 | 1157 | return mnt; |
719f5d7f MS |
1158 | |
1159 | out_free: | |
8ffcb32e | 1160 | mnt_free_id(mnt); |
719f5d7f | 1161 | free_vfsmnt(mnt); |
be34d1a3 | 1162 | return ERR_PTR(err); |
1da177e4 LT |
1163 | } |
1164 | ||
9ea459e1 AV |
1165 | static void cleanup_mnt(struct mount *mnt) |
1166 | { | |
1167 | /* | |
1168 | * This probably indicates that somebody messed | |
1169 | * up a mnt_want/drop_write() pair. If this | |
1170 | * happens, the filesystem was probably unable | |
1171 | * to make r/w->r/o transitions. | |
1172 | */ | |
1173 | /* | |
1174 | * The locking used to deal with mnt_count decrement provides barriers, | |
1175 | * so mnt_get_writers() below is safe. | |
1176 | */ | |
1177 | WARN_ON(mnt_get_writers(mnt)); | |
1178 | if (unlikely(mnt->mnt_pins.first)) | |
1179 | mnt_pin_kill(mnt); | |
1180 | fsnotify_vfsmount_delete(&mnt->mnt); | |
1181 | dput(mnt->mnt.mnt_root); | |
1182 | deactivate_super(mnt->mnt.mnt_sb); | |
1183 | mnt_free_id(mnt); | |
1184 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1185 | } | |
1186 | ||
1187 | static void __cleanup_mnt(struct rcu_head *head) | |
1188 | { | |
1189 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1190 | } | |
1191 | ||
1192 | static LLIST_HEAD(delayed_mntput_list); | |
1193 | static void delayed_mntput(struct work_struct *unused) | |
1194 | { | |
1195 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1196 | struct mount *m, *t; |
9ea459e1 | 1197 | |
29785735 BP |
1198 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1199 | cleanup_mnt(m); | |
9ea459e1 AV |
1200 | } |
1201 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1202 | ||
900148dc | 1203 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1204 | { |
48a066e7 AV |
1205 | rcu_read_lock(); |
1206 | mnt_add_count(mnt, -1); | |
1207 | if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ | |
1208 | rcu_read_unlock(); | |
f03c6599 | 1209 | return; |
b3e19d92 | 1210 | } |
719ea2fb | 1211 | lock_mount_hash(); |
b3e19d92 | 1212 | if (mnt_get_count(mnt)) { |
48a066e7 | 1213 | rcu_read_unlock(); |
719ea2fb | 1214 | unlock_mount_hash(); |
99b7db7b NP |
1215 | return; |
1216 | } | |
48a066e7 AV |
1217 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1218 | rcu_read_unlock(); | |
1219 | unlock_mount_hash(); | |
1220 | return; | |
1221 | } | |
1222 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1223 | rcu_read_unlock(); | |
962830df | 1224 | |
39f7c4db | 1225 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1226 | |
1227 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1228 | struct mount *p, *tmp; | |
1229 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
1230 | umount_mnt(p); | |
1231 | } | |
1232 | } | |
719ea2fb | 1233 | unlock_mount_hash(); |
649a795a | 1234 | |
9ea459e1 AV |
1235 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1236 | struct task_struct *task = current; | |
1237 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1238 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
1239 | if (!task_work_add(task, &mnt->mnt_rcu, true)) | |
1240 | return; | |
1241 | } | |
1242 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1243 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1244 | return; | |
1245 | } | |
1246 | cleanup_mnt(mnt); | |
b3e19d92 | 1247 | } |
b3e19d92 NP |
1248 | |
1249 | void mntput(struct vfsmount *mnt) | |
1250 | { | |
1251 | if (mnt) { | |
863d684f | 1252 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1253 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1254 | if (unlikely(m->mnt_expiry_mark)) |
1255 | m->mnt_expiry_mark = 0; | |
1256 | mntput_no_expire(m); | |
b3e19d92 NP |
1257 | } |
1258 | } | |
1259 | EXPORT_SYMBOL(mntput); | |
1260 | ||
1261 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1262 | { | |
1263 | if (mnt) | |
83adc753 | 1264 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1265 | return mnt; |
1266 | } | |
1267 | EXPORT_SYMBOL(mntget); | |
1268 | ||
c6609c0a IK |
1269 | /* path_is_mountpoint() - Check if path is a mount in the current |
1270 | * namespace. | |
1271 | * | |
1272 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1273 | * not mounted in any namespace and that common case is handled inline. | |
1274 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1275 | * mounts using a given dentry in a different namespace. This function | |
1276 | * checks if the passed in path is a mountpoint rather than the dentry | |
1277 | * alone. | |
1278 | */ | |
1279 | bool path_is_mountpoint(const struct path *path) | |
1280 | { | |
1281 | unsigned seq; | |
1282 | bool res; | |
1283 | ||
1284 | if (!d_mountpoint(path->dentry)) | |
1285 | return false; | |
1286 | ||
1287 | rcu_read_lock(); | |
1288 | do { | |
1289 | seq = read_seqbegin(&mount_lock); | |
1290 | res = __path_is_mountpoint(path); | |
1291 | } while (read_seqretry(&mount_lock, seq)); | |
1292 | rcu_read_unlock(); | |
1293 | ||
1294 | return res; | |
1295 | } | |
1296 | EXPORT_SYMBOL(path_is_mountpoint); | |
1297 | ||
ca71cf71 | 1298 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1299 | { |
3064c356 AV |
1300 | struct mount *p; |
1301 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1302 | if (IS_ERR(p)) | |
1303 | return ERR_CAST(p); | |
1304 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1305 | return &p->mnt; | |
7b7b1ace | 1306 | } |
1da177e4 | 1307 | |
a1a2c409 | 1308 | #ifdef CONFIG_PROC_FS |
0226f492 | 1309 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1310 | static void *m_start(struct seq_file *m, loff_t *pos) |
1311 | { | |
ede1bf0d | 1312 | struct proc_mounts *p = m->private; |
1da177e4 | 1313 | |
390c6843 | 1314 | down_read(&namespace_sem); |
c7999c36 AV |
1315 | if (p->cached_event == p->ns->event) { |
1316 | void *v = p->cached_mount; | |
1317 | if (*pos == p->cached_index) | |
1318 | return v; | |
1319 | if (*pos == p->cached_index + 1) { | |
1320 | v = seq_list_next(v, &p->ns->list, &p->cached_index); | |
1321 | return p->cached_mount = v; | |
1322 | } | |
1323 | } | |
1324 | ||
1325 | p->cached_event = p->ns->event; | |
1326 | p->cached_mount = seq_list_start(&p->ns->list, *pos); | |
1327 | p->cached_index = *pos; | |
1328 | return p->cached_mount; | |
1da177e4 LT |
1329 | } |
1330 | ||
1331 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1332 | { | |
ede1bf0d | 1333 | struct proc_mounts *p = m->private; |
b0765fb8 | 1334 | |
c7999c36 AV |
1335 | p->cached_mount = seq_list_next(v, &p->ns->list, pos); |
1336 | p->cached_index = *pos; | |
1337 | return p->cached_mount; | |
1da177e4 LT |
1338 | } |
1339 | ||
1340 | static void m_stop(struct seq_file *m, void *v) | |
1341 | { | |
390c6843 | 1342 | up_read(&namespace_sem); |
1da177e4 LT |
1343 | } |
1344 | ||
0226f492 | 1345 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1346 | { |
ede1bf0d | 1347 | struct proc_mounts *p = m->private; |
1a4eeaf2 | 1348 | struct mount *r = list_entry(v, struct mount, mnt_list); |
0226f492 | 1349 | return p->show(m, &r->mnt); |
1da177e4 LT |
1350 | } |
1351 | ||
a1a2c409 | 1352 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1353 | .start = m_start, |
1354 | .next = m_next, | |
1355 | .stop = m_stop, | |
0226f492 | 1356 | .show = m_show, |
b4629fe2 | 1357 | }; |
a1a2c409 | 1358 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1359 | |
1da177e4 LT |
1360 | /** |
1361 | * may_umount_tree - check if a mount tree is busy | |
1362 | * @mnt: root of mount tree | |
1363 | * | |
1364 | * This is called to check if a tree of mounts has any | |
1365 | * open files, pwds, chroots or sub mounts that are | |
1366 | * busy. | |
1367 | */ | |
909b0a88 | 1368 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1369 | { |
909b0a88 | 1370 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1371 | int actual_refs = 0; |
1372 | int minimum_refs = 0; | |
315fc83e | 1373 | struct mount *p; |
909b0a88 | 1374 | BUG_ON(!m); |
1da177e4 | 1375 | |
b3e19d92 | 1376 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1377 | lock_mount_hash(); |
909b0a88 | 1378 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1379 | actual_refs += mnt_get_count(p); |
1da177e4 | 1380 | minimum_refs += 2; |
1da177e4 | 1381 | } |
719ea2fb | 1382 | unlock_mount_hash(); |
1da177e4 LT |
1383 | |
1384 | if (actual_refs > minimum_refs) | |
e3474a8e | 1385 | return 0; |
1da177e4 | 1386 | |
e3474a8e | 1387 | return 1; |
1da177e4 LT |
1388 | } |
1389 | ||
1390 | EXPORT_SYMBOL(may_umount_tree); | |
1391 | ||
1392 | /** | |
1393 | * may_umount - check if a mount point is busy | |
1394 | * @mnt: root of mount | |
1395 | * | |
1396 | * This is called to check if a mount point has any | |
1397 | * open files, pwds, chroots or sub mounts. If the | |
1398 | * mount has sub mounts this will return busy | |
1399 | * regardless of whether the sub mounts are busy. | |
1400 | * | |
1401 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1402 | * give false negatives. The main reason why it's here is that we need | |
1403 | * a non-destructive way to look for easily umountable filesystems. | |
1404 | */ | |
1405 | int may_umount(struct vfsmount *mnt) | |
1406 | { | |
e3474a8e | 1407 | int ret = 1; |
8ad08d8a | 1408 | down_read(&namespace_sem); |
719ea2fb | 1409 | lock_mount_hash(); |
1ab59738 | 1410 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1411 | ret = 0; |
719ea2fb | 1412 | unlock_mount_hash(); |
8ad08d8a | 1413 | up_read(&namespace_sem); |
a05964f3 | 1414 | return ret; |
1da177e4 LT |
1415 | } |
1416 | ||
1417 | EXPORT_SYMBOL(may_umount); | |
1418 | ||
38129a13 | 1419 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
e3197d83 | 1420 | |
97216be0 | 1421 | static void namespace_unlock(void) |
70fbcdf4 | 1422 | { |
a3b3c562 | 1423 | struct hlist_head head; |
97216be0 | 1424 | |
a3b3c562 | 1425 | hlist_move_list(&unmounted, &head); |
97216be0 | 1426 | |
97216be0 AV |
1427 | up_write(&namespace_sem); |
1428 | ||
a3b3c562 EB |
1429 | if (likely(hlist_empty(&head))) |
1430 | return; | |
1431 | ||
48a066e7 AV |
1432 | synchronize_rcu(); |
1433 | ||
87b95ce0 | 1434 | group_pin_kill(&head); |
70fbcdf4 RP |
1435 | } |
1436 | ||
97216be0 | 1437 | static inline void namespace_lock(void) |
e3197d83 | 1438 | { |
97216be0 | 1439 | down_write(&namespace_sem); |
e3197d83 AV |
1440 | } |
1441 | ||
e819f152 EB |
1442 | enum umount_tree_flags { |
1443 | UMOUNT_SYNC = 1, | |
1444 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1445 | UMOUNT_CONNECTED = 4, |
e819f152 | 1446 | }; |
f2d0a123 EB |
1447 | |
1448 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1449 | { | |
1450 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1451 | if (how & UMOUNT_SYNC) | |
1452 | return true; | |
1453 | ||
1454 | /* A mount without a parent has nothing to be connected to */ | |
1455 | if (!mnt_has_parent(mnt)) | |
1456 | return true; | |
1457 | ||
1458 | /* Because the reference counting rules change when mounts are | |
1459 | * unmounted and connected, umounted mounts may not be | |
1460 | * connected to mounted mounts. | |
1461 | */ | |
1462 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1463 | return true; | |
1464 | ||
1465 | /* Has it been requested that the mount remain connected? */ | |
1466 | if (how & UMOUNT_CONNECTED) | |
1467 | return false; | |
1468 | ||
1469 | /* Is the mount locked such that it needs to remain connected? */ | |
1470 | if (IS_MNT_LOCKED(mnt)) | |
1471 | return false; | |
1472 | ||
1473 | /* By default disconnect the mount */ | |
1474 | return true; | |
1475 | } | |
1476 | ||
99b7db7b | 1477 | /* |
48a066e7 | 1478 | * mount_lock must be held |
99b7db7b NP |
1479 | * namespace_sem must be held for write |
1480 | */ | |
e819f152 | 1481 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1482 | { |
c003b26f | 1483 | LIST_HEAD(tmp_list); |
315fc83e | 1484 | struct mount *p; |
1da177e4 | 1485 | |
5d88457e EB |
1486 | if (how & UMOUNT_PROPAGATE) |
1487 | propagate_mount_unlock(mnt); | |
1488 | ||
c003b26f | 1489 | /* Gather the mounts to umount */ |
590ce4bc EB |
1490 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1491 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1492 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1493 | } |
1da177e4 | 1494 | |
411a938b | 1495 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1496 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1497 | list_del_init(&p->mnt_child); |
c003b26f | 1498 | } |
88b368f2 | 1499 | |
c003b26f | 1500 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1501 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1502 | propagate_umount(&tmp_list); |
a05964f3 | 1503 | |
c003b26f | 1504 | while (!list_empty(&tmp_list)) { |
d2921684 | 1505 | struct mnt_namespace *ns; |
ce07d891 | 1506 | bool disconnect; |
c003b26f | 1507 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1508 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1509 | list_del_init(&p->mnt_list); |
d2921684 EB |
1510 | ns = p->mnt_ns; |
1511 | if (ns) { | |
1512 | ns->mounts--; | |
1513 | __touch_mnt_namespace(ns); | |
1514 | } | |
143c8c91 | 1515 | p->mnt_ns = NULL; |
e819f152 | 1516 | if (how & UMOUNT_SYNC) |
48a066e7 | 1517 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1518 | |
f2d0a123 | 1519 | disconnect = disconnect_mount(p, how); |
ce07d891 EB |
1520 | |
1521 | pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, | |
1522 | disconnect ? &unmounted : NULL); | |
676da58d | 1523 | if (mnt_has_parent(p)) { |
81b6b061 | 1524 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1525 | if (!disconnect) { |
1526 | /* Don't forget about p */ | |
1527 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1528 | } else { | |
1529 | umount_mnt(p); | |
1530 | } | |
7c4b93d8 | 1531 | } |
0f0afb1d | 1532 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
1533 | } |
1534 | } | |
1535 | ||
b54b9be7 | 1536 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1537 | |
1ab59738 | 1538 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1539 | { |
1ab59738 | 1540 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1541 | int retval; |
1542 | ||
1ab59738 | 1543 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1544 | if (retval) |
1545 | return retval; | |
1546 | ||
1547 | /* | |
1548 | * Allow userspace to request a mountpoint be expired rather than | |
1549 | * unmounting unconditionally. Unmount only happens if: | |
1550 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1551 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1552 | */ | |
1553 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1554 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1555 | flags & (MNT_FORCE | MNT_DETACH)) |
1556 | return -EINVAL; | |
1557 | ||
b3e19d92 NP |
1558 | /* |
1559 | * probably don't strictly need the lock here if we examined | |
1560 | * all race cases, but it's a slowpath. | |
1561 | */ | |
719ea2fb | 1562 | lock_mount_hash(); |
83adc753 | 1563 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1564 | unlock_mount_hash(); |
1da177e4 | 1565 | return -EBUSY; |
b3e19d92 | 1566 | } |
719ea2fb | 1567 | unlock_mount_hash(); |
1da177e4 | 1568 | |
863d684f | 1569 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1570 | return -EAGAIN; |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * If we may have to abort operations to get out of this | |
1575 | * mount, and they will themselves hold resources we must | |
1576 | * allow the fs to do things. In the Unix tradition of | |
1577 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1578 | * might fail to complete on the first run through as other tasks | |
1579 | * must return, and the like. Thats for the mount program to worry | |
1580 | * about for the moment. | |
1581 | */ | |
1582 | ||
42faad99 | 1583 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1584 | sb->s_op->umount_begin(sb); |
42faad99 | 1585 | } |
1da177e4 LT |
1586 | |
1587 | /* | |
1588 | * No sense to grab the lock for this test, but test itself looks | |
1589 | * somewhat bogus. Suggestions for better replacement? | |
1590 | * Ho-hum... In principle, we might treat that as umount + switch | |
1591 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1592 | * Actually it makes sense, especially if rootfs would contain a | |
1593 | * /reboot - static binary that would close all descriptors and | |
1594 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1595 | */ | |
1ab59738 | 1596 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1597 | /* |
1598 | * Special case for "unmounting" root ... | |
1599 | * we just try to remount it readonly. | |
1600 | */ | |
1db30ec0 | 1601 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1602 | return -EPERM; |
1da177e4 | 1603 | down_write(&sb->s_umount); |
bc98a42c | 1604 | if (!sb_rdonly(sb)) |
e462ec50 | 1605 | retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); |
1da177e4 LT |
1606 | up_write(&sb->s_umount); |
1607 | return retval; | |
1608 | } | |
1609 | ||
97216be0 | 1610 | namespace_lock(); |
719ea2fb | 1611 | lock_mount_hash(); |
5addc5dd | 1612 | event++; |
1da177e4 | 1613 | |
48a066e7 | 1614 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1615 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1616 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1617 | retval = 0; |
48a066e7 AV |
1618 | } else { |
1619 | shrink_submounts(mnt); | |
1620 | retval = -EBUSY; | |
1621 | if (!propagate_mount_busy(mnt, 2)) { | |
1622 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1623 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1624 | retval = 0; |
1625 | } | |
1da177e4 | 1626 | } |
719ea2fb | 1627 | unlock_mount_hash(); |
e3197d83 | 1628 | namespace_unlock(); |
1da177e4 LT |
1629 | return retval; |
1630 | } | |
1631 | ||
80b5dce8 EB |
1632 | /* |
1633 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1634 | * | |
1635 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1636 | * to an existing mountpoint, and wind up leaking the mount. | |
1637 | * detach_mounts allows lazily unmounting those mounts instead of | |
1638 | * leaking them. | |
1639 | * | |
1640 | * The caller may hold dentry->d_inode->i_mutex. | |
1641 | */ | |
1642 | void __detach_mounts(struct dentry *dentry) | |
1643 | { | |
1644 | struct mountpoint *mp; | |
1645 | struct mount *mnt; | |
1646 | ||
1647 | namespace_lock(); | |
3895dbf8 | 1648 | lock_mount_hash(); |
80b5dce8 | 1649 | mp = lookup_mountpoint(dentry); |
f53e5797 | 1650 | if (IS_ERR_OR_NULL(mp)) |
80b5dce8 EB |
1651 | goto out_unlock; |
1652 | ||
e06b933e | 1653 | event++; |
80b5dce8 EB |
1654 | while (!hlist_empty(&mp->m_list)) { |
1655 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1656 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 EB |
1657 | hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); |
1658 | umount_mnt(mnt); | |
ce07d891 | 1659 | } |
e0c9c0af | 1660 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1661 | } |
80b5dce8 EB |
1662 | put_mountpoint(mp); |
1663 | out_unlock: | |
3895dbf8 | 1664 | unlock_mount_hash(); |
80b5dce8 EB |
1665 | namespace_unlock(); |
1666 | } | |
1667 | ||
dd111b31 | 1668 | /* |
9b40bc90 AV |
1669 | * Is the caller allowed to modify his namespace? |
1670 | */ | |
1671 | static inline bool may_mount(void) | |
1672 | { | |
1673 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1674 | } | |
1675 | ||
9e8925b6 JL |
1676 | static inline bool may_mandlock(void) |
1677 | { | |
1678 | #ifndef CONFIG_MANDATORY_FILE_LOCKING | |
1679 | return false; | |
1680 | #endif | |
95ace754 | 1681 | return capable(CAP_SYS_ADMIN); |
9e8925b6 JL |
1682 | } |
1683 | ||
1da177e4 LT |
1684 | /* |
1685 | * Now umount can handle mount points as well as block devices. | |
1686 | * This is important for filesystems which use unnamed block devices. | |
1687 | * | |
1688 | * We now support a flag for forced unmount like the other 'big iron' | |
1689 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
1690 | */ | |
1691 | ||
bdc480e3 | 1692 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1da177e4 | 1693 | { |
2d8f3038 | 1694 | struct path path; |
900148dc | 1695 | struct mount *mnt; |
1da177e4 | 1696 | int retval; |
db1f05bb | 1697 | int lookup_flags = 0; |
1da177e4 | 1698 | |
db1f05bb MS |
1699 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) |
1700 | return -EINVAL; | |
1701 | ||
9b40bc90 AV |
1702 | if (!may_mount()) |
1703 | return -EPERM; | |
1704 | ||
db1f05bb MS |
1705 | if (!(flags & UMOUNT_NOFOLLOW)) |
1706 | lookup_flags |= LOOKUP_FOLLOW; | |
1707 | ||
197df04c | 1708 | retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); |
1da177e4 LT |
1709 | if (retval) |
1710 | goto out; | |
900148dc | 1711 | mnt = real_mount(path.mnt); |
1da177e4 | 1712 | retval = -EINVAL; |
2d8f3038 | 1713 | if (path.dentry != path.mnt->mnt_root) |
1da177e4 | 1714 | goto dput_and_out; |
143c8c91 | 1715 | if (!check_mnt(mnt)) |
1da177e4 | 1716 | goto dput_and_out; |
5ff9d8a6 EB |
1717 | if (mnt->mnt.mnt_flags & MNT_LOCKED) |
1718 | goto dput_and_out; | |
b2f5d4dc EB |
1719 | retval = -EPERM; |
1720 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) | |
1721 | goto dput_and_out; | |
1da177e4 | 1722 | |
900148dc | 1723 | retval = do_umount(mnt, flags); |
1da177e4 | 1724 | dput_and_out: |
429731b1 | 1725 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
2d8f3038 | 1726 | dput(path.dentry); |
900148dc | 1727 | mntput_no_expire(mnt); |
1da177e4 LT |
1728 | out: |
1729 | return retval; | |
1730 | } | |
1731 | ||
1732 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
1733 | ||
1734 | /* | |
b58fed8b | 1735 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1736 | */ |
bdc480e3 | 1737 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1738 | { |
b58fed8b | 1739 | return sys_umount(name, 0); |
1da177e4 LT |
1740 | } |
1741 | ||
1742 | #endif | |
1743 | ||
4ce5d2b1 | 1744 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1745 | { |
4ce5d2b1 | 1746 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1747 | return dentry->d_op == &ns_dentry_operations && |
1748 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1749 | } |
1750 | ||
58be2825 AV |
1751 | struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
1752 | { | |
1753 | return container_of(ns, struct mnt_namespace, ns); | |
1754 | } | |
1755 | ||
4ce5d2b1 EB |
1756 | static bool mnt_ns_loop(struct dentry *dentry) |
1757 | { | |
1758 | /* Could bind mounting the mount namespace inode cause a | |
1759 | * mount namespace loop? | |
1760 | */ | |
1761 | struct mnt_namespace *mnt_ns; | |
1762 | if (!is_mnt_ns_file(dentry)) | |
1763 | return false; | |
1764 | ||
f77c8014 | 1765 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1766 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1767 | } | |
1768 | ||
87129cc0 | 1769 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1770 | int flag) |
1da177e4 | 1771 | { |
84d17192 | 1772 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1773 | |
4ce5d2b1 EB |
1774 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1775 | return ERR_PTR(-EINVAL); | |
1776 | ||
1777 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1778 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1779 | |
36341f64 | 1780 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1781 | if (IS_ERR(q)) |
1782 | return q; | |
1783 | ||
a73324da | 1784 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1785 | |
1786 | p = mnt; | |
6b41d536 | 1787 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1788 | struct mount *s; |
7ec02ef1 | 1789 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1790 | continue; |
1791 | ||
909b0a88 | 1792 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1793 | if (!(flag & CL_COPY_UNBINDABLE) && |
1794 | IS_MNT_UNBINDABLE(s)) { | |
1795 | s = skip_mnt_tree(s); | |
1796 | continue; | |
1797 | } | |
1798 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1799 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1800 | s = skip_mnt_tree(s); |
1801 | continue; | |
1802 | } | |
0714a533 AV |
1803 | while (p != s->mnt_parent) { |
1804 | p = p->mnt_parent; | |
1805 | q = q->mnt_parent; | |
1da177e4 | 1806 | } |
87129cc0 | 1807 | p = s; |
84d17192 | 1808 | parent = q; |
87129cc0 | 1809 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1810 | if (IS_ERR(q)) |
1811 | goto out; | |
719ea2fb | 1812 | lock_mount_hash(); |
1a4eeaf2 | 1813 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1814 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1815 | unlock_mount_hash(); |
1da177e4 LT |
1816 | } |
1817 | } | |
1818 | return res; | |
be34d1a3 | 1819 | out: |
1da177e4 | 1820 | if (res) { |
719ea2fb | 1821 | lock_mount_hash(); |
e819f152 | 1822 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1823 | unlock_mount_hash(); |
1da177e4 | 1824 | } |
be34d1a3 | 1825 | return q; |
1da177e4 LT |
1826 | } |
1827 | ||
be34d1a3 DH |
1828 | /* Caller should check returned pointer for errors */ |
1829 | ||
ca71cf71 | 1830 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1831 | { |
cb338d06 | 1832 | struct mount *tree; |
97216be0 | 1833 | namespace_lock(); |
cd4a4017 EB |
1834 | if (!check_mnt(real_mount(path->mnt))) |
1835 | tree = ERR_PTR(-EINVAL); | |
1836 | else | |
1837 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1838 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1839 | namespace_unlock(); |
be34d1a3 | 1840 | if (IS_ERR(tree)) |
52e220d3 | 1841 | return ERR_CAST(tree); |
be34d1a3 | 1842 | return &tree->mnt; |
8aec0809 AV |
1843 | } |
1844 | ||
1845 | void drop_collected_mounts(struct vfsmount *mnt) | |
1846 | { | |
97216be0 | 1847 | namespace_lock(); |
719ea2fb | 1848 | lock_mount_hash(); |
e819f152 | 1849 | umount_tree(real_mount(mnt), UMOUNT_SYNC); |
719ea2fb | 1850 | unlock_mount_hash(); |
3ab6abee | 1851 | namespace_unlock(); |
8aec0809 AV |
1852 | } |
1853 | ||
c771d683 MS |
1854 | /** |
1855 | * clone_private_mount - create a private clone of a path | |
1856 | * | |
1857 | * This creates a new vfsmount, which will be the clone of @path. The new will | |
1858 | * not be attached anywhere in the namespace and will be private (i.e. changes | |
1859 | * to the originating mount won't be propagated into this). | |
1860 | * | |
1861 | * Release with mntput(). | |
1862 | */ | |
ca71cf71 | 1863 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
1864 | { |
1865 | struct mount *old_mnt = real_mount(path->mnt); | |
1866 | struct mount *new_mnt; | |
1867 | ||
1868 | if (IS_MNT_UNBINDABLE(old_mnt)) | |
1869 | return ERR_PTR(-EINVAL); | |
1870 | ||
c771d683 | 1871 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
c771d683 MS |
1872 | if (IS_ERR(new_mnt)) |
1873 | return ERR_CAST(new_mnt); | |
1874 | ||
1875 | return &new_mnt->mnt; | |
1876 | } | |
1877 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
1878 | ||
1f707137 AV |
1879 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
1880 | struct vfsmount *root) | |
1881 | { | |
1a4eeaf2 | 1882 | struct mount *mnt; |
1f707137 AV |
1883 | int res = f(root, arg); |
1884 | if (res) | |
1885 | return res; | |
1a4eeaf2 AV |
1886 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
1887 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
1888 | if (res) |
1889 | return res; | |
1890 | } | |
1891 | return 0; | |
1892 | } | |
c088e31d | 1893 | EXPORT_SYMBOL_GPL(iterate_mounts); |
1f707137 | 1894 | |
4b8b21f4 | 1895 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 1896 | { |
315fc83e | 1897 | struct mount *p; |
719f5d7f | 1898 | |
909b0a88 | 1899 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 1900 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 1901 | mnt_release_group_id(p); |
719f5d7f MS |
1902 | } |
1903 | } | |
1904 | ||
4b8b21f4 | 1905 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 1906 | { |
315fc83e | 1907 | struct mount *p; |
719f5d7f | 1908 | |
909b0a88 | 1909 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 1910 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 1911 | int err = mnt_alloc_group_id(p); |
719f5d7f | 1912 | if (err) { |
4b8b21f4 | 1913 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
1914 | return err; |
1915 | } | |
1916 | } | |
1917 | } | |
1918 | ||
1919 | return 0; | |
1920 | } | |
1921 | ||
d2921684 EB |
1922 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
1923 | { | |
1924 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
1925 | unsigned int mounts = 0, old, pending, sum; | |
1926 | struct mount *p; | |
1927 | ||
1928 | for (p = mnt; p; p = next_mnt(p, mnt)) | |
1929 | mounts++; | |
1930 | ||
1931 | old = ns->mounts; | |
1932 | pending = ns->pending_mounts; | |
1933 | sum = old + pending; | |
1934 | if ((old > sum) || | |
1935 | (pending > sum) || | |
1936 | (max < sum) || | |
1937 | (mounts > (max - sum))) | |
1938 | return -ENOSPC; | |
1939 | ||
1940 | ns->pending_mounts = pending + mounts; | |
1941 | return 0; | |
1942 | } | |
1943 | ||
b90fa9ae RP |
1944 | /* |
1945 | * @source_mnt : mount tree to be attached | |
21444403 RP |
1946 | * @nd : place the mount tree @source_mnt is attached |
1947 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
1948 | * store the parent mount and mountpoint dentry. | |
1949 | * (done when source_mnt is moved) | |
b90fa9ae RP |
1950 | * |
1951 | * NOTE: in the table below explains the semantics when a source mount | |
1952 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
1953 | * --------------------------------------------------------------------------- |
1954 | * | BIND MOUNT OPERATION | | |
1955 | * |************************************************************************** | |
1956 | * | source-->| shared | private | slave | unbindable | | |
1957 | * | dest | | | | | | |
1958 | * | | | | | | | | |
1959 | * | v | | | | | | |
1960 | * |************************************************************************** | |
1961 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
1962 | * | | | | | | | |
1963 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
1964 | * *************************************************************************** | |
b90fa9ae RP |
1965 | * A bind operation clones the source mount and mounts the clone on the |
1966 | * destination mount. | |
1967 | * | |
1968 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
1969 | * tree of the destination mount and the cloned mount is added to | |
1970 | * the peer group of the source mount. | |
1971 | * (+) the cloned mount is created under the destination mount and is marked | |
1972 | * as shared. The cloned mount is added to the peer group of the source | |
1973 | * mount. | |
5afe0022 RP |
1974 | * (+++) the mount is propagated to all the mounts in the propagation tree |
1975 | * of the destination mount and the cloned mount is made slave | |
1976 | * of the same master as that of the source mount. The cloned mount | |
1977 | * is marked as 'shared and slave'. | |
1978 | * (*) the cloned mount is made a slave of the same master as that of the | |
1979 | * source mount. | |
1980 | * | |
9676f0c6 RP |
1981 | * --------------------------------------------------------------------------- |
1982 | * | MOVE MOUNT OPERATION | | |
1983 | * |************************************************************************** | |
1984 | * | source-->| shared | private | slave | unbindable | | |
1985 | * | dest | | | | | | |
1986 | * | | | | | | | | |
1987 | * | v | | | | | | |
1988 | * |************************************************************************** | |
1989 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
1990 | * | | | | | | | |
1991 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
1992 | * *************************************************************************** | |
5afe0022 RP |
1993 | * |
1994 | * (+) the mount is moved to the destination. And is then propagated to | |
1995 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 1996 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
1997 | * (+++) the mount is moved to the destination and is then propagated to |
1998 | * all the mounts belonging to the destination mount's propagation tree. | |
1999 | * the mount is marked as 'shared and slave'. | |
2000 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2001 | * |
2002 | * if the source mount is a tree, the operations explained above is | |
2003 | * applied to each mount in the tree. | |
2004 | * Must be called without spinlocks held, since this function can sleep | |
2005 | * in allocations. | |
2006 | */ | |
0fb54e50 | 2007 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2008 | struct mount *dest_mnt, |
2009 | struct mountpoint *dest_mp, | |
2010 | struct path *parent_path) | |
b90fa9ae | 2011 | { |
38129a13 | 2012 | HLIST_HEAD(tree_list); |
d2921684 | 2013 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2014 | struct mountpoint *smp; |
315fc83e | 2015 | struct mount *child, *p; |
38129a13 | 2016 | struct hlist_node *n; |
719f5d7f | 2017 | int err; |
b90fa9ae | 2018 | |
1064f874 EB |
2019 | /* Preallocate a mountpoint in case the new mounts need |
2020 | * to be tucked under other mounts. | |
2021 | */ | |
2022 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2023 | if (IS_ERR(smp)) | |
2024 | return PTR_ERR(smp); | |
2025 | ||
d2921684 EB |
2026 | /* Is there space to add these mounts to the mount namespace? */ |
2027 | if (!parent_path) { | |
2028 | err = count_mounts(ns, source_mnt); | |
2029 | if (err) | |
2030 | goto out; | |
2031 | } | |
2032 | ||
fc7be130 | 2033 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2034 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2035 | if (err) |
2036 | goto out; | |
0b1b901b | 2037 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2038 | lock_mount_hash(); |
0b1b901b AV |
2039 | if (err) |
2040 | goto out_cleanup_ids; | |
909b0a88 | 2041 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2042 | set_mnt_shared(p); |
0b1b901b AV |
2043 | } else { |
2044 | lock_mount_hash(); | |
b90fa9ae | 2045 | } |
1a390689 | 2046 | if (parent_path) { |
0fb54e50 | 2047 | detach_mnt(source_mnt, parent_path); |
84d17192 | 2048 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2049 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2050 | } else { |
84d17192 | 2051 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2052 | commit_tree(source_mnt); |
21444403 | 2053 | } |
b90fa9ae | 2054 | |
38129a13 | 2055 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2056 | struct mount *q; |
38129a13 | 2057 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2058 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2059 | child->mnt_mountpoint); | |
2060 | if (q) | |
2061 | mnt_change_mountpoint(child, smp, q); | |
2062 | commit_tree(child); | |
b90fa9ae | 2063 | } |
1064f874 | 2064 | put_mountpoint(smp); |
719ea2fb | 2065 | unlock_mount_hash(); |
99b7db7b | 2066 | |
b90fa9ae | 2067 | return 0; |
719f5d7f MS |
2068 | |
2069 | out_cleanup_ids: | |
f2ebb3a9 AV |
2070 | while (!hlist_empty(&tree_list)) { |
2071 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2072 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2073 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2074 | } |
2075 | unlock_mount_hash(); | |
0b1b901b | 2076 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2077 | out: |
d2921684 | 2078 | ns->pending_mounts = 0; |
1064f874 EB |
2079 | |
2080 | read_seqlock_excl(&mount_lock); | |
2081 | put_mountpoint(smp); | |
2082 | read_sequnlock_excl(&mount_lock); | |
2083 | ||
719f5d7f | 2084 | return err; |
b90fa9ae RP |
2085 | } |
2086 | ||
84d17192 | 2087 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2088 | { |
2089 | struct vfsmount *mnt; | |
84d17192 | 2090 | struct dentry *dentry = path->dentry; |
b12cea91 | 2091 | retry: |
5955102c | 2092 | inode_lock(dentry->d_inode); |
84d17192 | 2093 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2094 | inode_unlock(dentry->d_inode); |
84d17192 | 2095 | return ERR_PTR(-ENOENT); |
b12cea91 | 2096 | } |
97216be0 | 2097 | namespace_lock(); |
b12cea91 | 2098 | mnt = lookup_mnt(path); |
84d17192 | 2099 | if (likely(!mnt)) { |
3895dbf8 | 2100 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2101 | if (IS_ERR(mp)) { |
97216be0 | 2102 | namespace_unlock(); |
5955102c | 2103 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2104 | return mp; |
2105 | } | |
2106 | return mp; | |
2107 | } | |
97216be0 | 2108 | namespace_unlock(); |
5955102c | 2109 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2110 | path_put(path); |
2111 | path->mnt = mnt; | |
84d17192 | 2112 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2113 | goto retry; |
2114 | } | |
2115 | ||
84d17192 | 2116 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2117 | { |
84d17192 | 2118 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2119 | |
2120 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2121 | put_mountpoint(where); |
3895dbf8 EB |
2122 | read_sequnlock_excl(&mount_lock); |
2123 | ||
328e6d90 | 2124 | namespace_unlock(); |
5955102c | 2125 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2126 | } |
2127 | ||
84d17192 | 2128 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2129 | { |
e462ec50 | 2130 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2131 | return -EINVAL; |
2132 | ||
e36cb0b8 DH |
2133 | if (d_is_dir(mp->m_dentry) != |
2134 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2135 | return -ENOTDIR; |
2136 | ||
84d17192 | 2137 | return attach_recursive_mnt(mnt, p, mp, NULL); |
1da177e4 LT |
2138 | } |
2139 | ||
7a2e8a8f VA |
2140 | /* |
2141 | * Sanity check the flags to change_mnt_propagation. | |
2142 | */ | |
2143 | ||
e462ec50 | 2144 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2145 | { |
e462ec50 | 2146 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2147 | |
2148 | /* Fail if any non-propagation flags are set */ | |
2149 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2150 | return 0; | |
2151 | /* Only one propagation flag should be set */ | |
2152 | if (!is_power_of_2(type)) | |
2153 | return 0; | |
2154 | return type; | |
2155 | } | |
2156 | ||
07b20889 RP |
2157 | /* |
2158 | * recursively change the type of the mountpoint. | |
2159 | */ | |
e462ec50 | 2160 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2161 | { |
315fc83e | 2162 | struct mount *m; |
4b8b21f4 | 2163 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2164 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2165 | int type; |
719f5d7f | 2166 | int err = 0; |
07b20889 | 2167 | |
2d92ab3c | 2168 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2169 | return -EINVAL; |
2170 | ||
e462ec50 | 2171 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2172 | if (!type) |
2173 | return -EINVAL; | |
2174 | ||
97216be0 | 2175 | namespace_lock(); |
719f5d7f MS |
2176 | if (type == MS_SHARED) { |
2177 | err = invent_group_ids(mnt, recurse); | |
2178 | if (err) | |
2179 | goto out_unlock; | |
2180 | } | |
2181 | ||
719ea2fb | 2182 | lock_mount_hash(); |
909b0a88 | 2183 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2184 | change_mnt_propagation(m, type); |
719ea2fb | 2185 | unlock_mount_hash(); |
719f5d7f MS |
2186 | |
2187 | out_unlock: | |
97216be0 | 2188 | namespace_unlock(); |
719f5d7f | 2189 | return err; |
07b20889 RP |
2190 | } |
2191 | ||
5ff9d8a6 EB |
2192 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2193 | { | |
2194 | struct mount *child; | |
2195 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2196 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2197 | continue; | |
2198 | ||
2199 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2200 | return true; | |
2201 | } | |
2202 | return false; | |
2203 | } | |
2204 | ||
1da177e4 LT |
2205 | /* |
2206 | * do loopback mount. | |
2207 | */ | |
808d4e3c | 2208 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2209 | int recurse) |
1da177e4 | 2210 | { |
2d92ab3c | 2211 | struct path old_path; |
84d17192 AV |
2212 | struct mount *mnt = NULL, *old, *parent; |
2213 | struct mountpoint *mp; | |
57eccb83 | 2214 | int err; |
1da177e4 LT |
2215 | if (!old_name || !*old_name) |
2216 | return -EINVAL; | |
815d405c | 2217 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2218 | if (err) |
2219 | return err; | |
2220 | ||
8823c079 | 2221 | err = -EINVAL; |
4ce5d2b1 | 2222 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2223 | goto out; |
8823c079 | 2224 | |
84d17192 AV |
2225 | mp = lock_mount(path); |
2226 | err = PTR_ERR(mp); | |
2227 | if (IS_ERR(mp)) | |
b12cea91 AV |
2228 | goto out; |
2229 | ||
87129cc0 | 2230 | old = real_mount(old_path.mnt); |
84d17192 | 2231 | parent = real_mount(path->mnt); |
87129cc0 | 2232 | |
1da177e4 | 2233 | err = -EINVAL; |
fc7be130 | 2234 | if (IS_MNT_UNBINDABLE(old)) |
b12cea91 | 2235 | goto out2; |
9676f0c6 | 2236 | |
e149ed2b AV |
2237 | if (!check_mnt(parent)) |
2238 | goto out2; | |
2239 | ||
2240 | if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) | |
b12cea91 | 2241 | goto out2; |
1da177e4 | 2242 | |
5ff9d8a6 EB |
2243 | if (!recurse && has_locked_children(old, old_path.dentry)) |
2244 | goto out2; | |
2245 | ||
ccd48bc7 | 2246 | if (recurse) |
4ce5d2b1 | 2247 | mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); |
ccd48bc7 | 2248 | else |
87129cc0 | 2249 | mnt = clone_mnt(old, old_path.dentry, 0); |
ccd48bc7 | 2250 | |
be34d1a3 DH |
2251 | if (IS_ERR(mnt)) { |
2252 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2253 | goto out2; |
be34d1a3 | 2254 | } |
ccd48bc7 | 2255 | |
5ff9d8a6 EB |
2256 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; |
2257 | ||
84d17192 | 2258 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2259 | if (err) { |
719ea2fb | 2260 | lock_mount_hash(); |
e819f152 | 2261 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2262 | unlock_mount_hash(); |
5b83d2c5 | 2263 | } |
b12cea91 | 2264 | out2: |
84d17192 | 2265 | unlock_mount(mp); |
ccd48bc7 | 2266 | out: |
2d92ab3c | 2267 | path_put(&old_path); |
1da177e4 LT |
2268 | return err; |
2269 | } | |
2270 | ||
2e4b7fcd DH |
2271 | static int change_mount_flags(struct vfsmount *mnt, int ms_flags) |
2272 | { | |
2273 | int error = 0; | |
2274 | int readonly_request = 0; | |
2275 | ||
2276 | if (ms_flags & MS_RDONLY) | |
2277 | readonly_request = 1; | |
2278 | if (readonly_request == __mnt_is_readonly(mnt)) | |
2279 | return 0; | |
2280 | ||
2281 | if (readonly_request) | |
83adc753 | 2282 | error = mnt_make_readonly(real_mount(mnt)); |
2e4b7fcd | 2283 | else |
83adc753 | 2284 | __mnt_unmake_readonly(real_mount(mnt)); |
2e4b7fcd DH |
2285 | return error; |
2286 | } | |
2287 | ||
1da177e4 LT |
2288 | /* |
2289 | * change filesystem flags. dir should be a physical root of filesystem. | |
2290 | * If you've mounted a non-root directory somewhere and want to do remount | |
2291 | * on it - tough luck. | |
2292 | */ | |
e462ec50 DH |
2293 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2294 | int mnt_flags, void *data) | |
1da177e4 LT |
2295 | { |
2296 | int err; | |
2d92ab3c | 2297 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2298 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 2299 | |
143c8c91 | 2300 | if (!check_mnt(mnt)) |
1da177e4 LT |
2301 | return -EINVAL; |
2302 | ||
2d92ab3c | 2303 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2304 | return -EINVAL; |
2305 | ||
07b64558 EB |
2306 | /* Don't allow changing of locked mnt flags. |
2307 | * | |
2308 | * No locks need to be held here while testing the various | |
2309 | * MNT_LOCK flags because those flags can never be cleared | |
2310 | * once they are set. | |
2311 | */ | |
2312 | if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && | |
2313 | !(mnt_flags & MNT_READONLY)) { | |
2314 | return -EPERM; | |
2315 | } | |
9566d674 EB |
2316 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && |
2317 | !(mnt_flags & MNT_NODEV)) { | |
67690f93 | 2318 | return -EPERM; |
9566d674 EB |
2319 | } |
2320 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && | |
2321 | !(mnt_flags & MNT_NOSUID)) { | |
2322 | return -EPERM; | |
2323 | } | |
2324 | if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && | |
2325 | !(mnt_flags & MNT_NOEXEC)) { | |
2326 | return -EPERM; | |
2327 | } | |
2328 | if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && | |
2329 | ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { | |
2330 | return -EPERM; | |
2331 | } | |
2332 | ||
ff36fe2c EP |
2333 | err = security_sb_remount(sb, data); |
2334 | if (err) | |
2335 | return err; | |
2336 | ||
1da177e4 | 2337 | down_write(&sb->s_umount); |
e462ec50 DH |
2338 | if (ms_flags & MS_BIND) |
2339 | err = change_mount_flags(path->mnt, ms_flags); | |
1db30ec0 | 2340 | else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
57eccb83 | 2341 | err = -EPERM; |
4aa98cf7 | 2342 | else |
e462ec50 | 2343 | err = do_remount_sb(sb, sb_flags, data, 0); |
7b43a79f | 2344 | if (!err) { |
719ea2fb | 2345 | lock_mount_hash(); |
a6138db8 | 2346 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
143c8c91 | 2347 | mnt->mnt.mnt_flags = mnt_flags; |
143c8c91 | 2348 | touch_mnt_namespace(mnt->mnt_ns); |
719ea2fb | 2349 | unlock_mount_hash(); |
0e55a7cc | 2350 | } |
6339dab8 | 2351 | up_write(&sb->s_umount); |
1da177e4 LT |
2352 | return err; |
2353 | } | |
2354 | ||
cbbe362c | 2355 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2356 | { |
315fc83e | 2357 | struct mount *p; |
909b0a88 | 2358 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2359 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2360 | return 1; |
2361 | } | |
2362 | return 0; | |
2363 | } | |
2364 | ||
808d4e3c | 2365 | static int do_move_mount(struct path *path, const char *old_name) |
1da177e4 | 2366 | { |
2d92ab3c | 2367 | struct path old_path, parent_path; |
676da58d | 2368 | struct mount *p; |
0fb54e50 | 2369 | struct mount *old; |
84d17192 | 2370 | struct mountpoint *mp; |
57eccb83 | 2371 | int err; |
1da177e4 LT |
2372 | if (!old_name || !*old_name) |
2373 | return -EINVAL; | |
2d92ab3c | 2374 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); |
1da177e4 LT |
2375 | if (err) |
2376 | return err; | |
2377 | ||
84d17192 AV |
2378 | mp = lock_mount(path); |
2379 | err = PTR_ERR(mp); | |
2380 | if (IS_ERR(mp)) | |
cc53ce53 DH |
2381 | goto out; |
2382 | ||
143c8c91 | 2383 | old = real_mount(old_path.mnt); |
fc7be130 | 2384 | p = real_mount(path->mnt); |
143c8c91 | 2385 | |
1da177e4 | 2386 | err = -EINVAL; |
fc7be130 | 2387 | if (!check_mnt(p) || !check_mnt(old)) |
1da177e4 LT |
2388 | goto out1; |
2389 | ||
5ff9d8a6 EB |
2390 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2391 | goto out1; | |
2392 | ||
1da177e4 | 2393 | err = -EINVAL; |
2d92ab3c | 2394 | if (old_path.dentry != old_path.mnt->mnt_root) |
21444403 | 2395 | goto out1; |
1da177e4 | 2396 | |
676da58d | 2397 | if (!mnt_has_parent(old)) |
21444403 | 2398 | goto out1; |
1da177e4 | 2399 | |
e36cb0b8 DH |
2400 | if (d_is_dir(path->dentry) != |
2401 | d_is_dir(old_path.dentry)) | |
21444403 RP |
2402 | goto out1; |
2403 | /* | |
2404 | * Don't move a mount residing in a shared parent. | |
2405 | */ | |
fc7be130 | 2406 | if (IS_MNT_SHARED(old->mnt_parent)) |
21444403 | 2407 | goto out1; |
9676f0c6 RP |
2408 | /* |
2409 | * Don't move a mount tree containing unbindable mounts to a destination | |
2410 | * mount which is shared. | |
2411 | */ | |
fc7be130 | 2412 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
9676f0c6 | 2413 | goto out1; |
1da177e4 | 2414 | err = -ELOOP; |
fc7be130 | 2415 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2416 | if (p == old) |
21444403 | 2417 | goto out1; |
1da177e4 | 2418 | |
84d17192 | 2419 | err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); |
4ac91378 | 2420 | if (err) |
21444403 | 2421 | goto out1; |
1da177e4 LT |
2422 | |
2423 | /* if the mount is moved, it should no longer be expire | |
2424 | * automatically */ | |
6776db3d | 2425 | list_del_init(&old->mnt_expire); |
1da177e4 | 2426 | out1: |
84d17192 | 2427 | unlock_mount(mp); |
1da177e4 | 2428 | out: |
1da177e4 | 2429 | if (!err) |
1a390689 | 2430 | path_put(&parent_path); |
2d92ab3c | 2431 | path_put(&old_path); |
1da177e4 LT |
2432 | return err; |
2433 | } | |
2434 | ||
9d412a43 AV |
2435 | static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) |
2436 | { | |
2437 | int err; | |
2438 | const char *subtype = strchr(fstype, '.'); | |
2439 | if (subtype) { | |
2440 | subtype++; | |
2441 | err = -EINVAL; | |
2442 | if (!subtype[0]) | |
2443 | goto err; | |
2444 | } else | |
2445 | subtype = ""; | |
2446 | ||
2447 | mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); | |
2448 | err = -ENOMEM; | |
2449 | if (!mnt->mnt_sb->s_subtype) | |
2450 | goto err; | |
2451 | return mnt; | |
2452 | ||
2453 | err: | |
2454 | mntput(mnt); | |
2455 | return ERR_PTR(err); | |
2456 | } | |
2457 | ||
9d412a43 AV |
2458 | /* |
2459 | * add a mount into a namespace's mount tree | |
2460 | */ | |
95bc5f25 | 2461 | static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) |
9d412a43 | 2462 | { |
84d17192 AV |
2463 | struct mountpoint *mp; |
2464 | struct mount *parent; | |
9d412a43 AV |
2465 | int err; |
2466 | ||
f2ebb3a9 | 2467 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2468 | |
84d17192 AV |
2469 | mp = lock_mount(path); |
2470 | if (IS_ERR(mp)) | |
2471 | return PTR_ERR(mp); | |
9d412a43 | 2472 | |
84d17192 | 2473 | parent = real_mount(path->mnt); |
9d412a43 | 2474 | err = -EINVAL; |
84d17192 | 2475 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2476 | /* that's acceptable only for automounts done in private ns */ |
2477 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
2478 | goto unlock; | |
2479 | /* ... and for those we'd better have mountpoint still alive */ | |
84d17192 | 2480 | if (!parent->mnt_ns) |
156cacb1 AV |
2481 | goto unlock; |
2482 | } | |
9d412a43 AV |
2483 | |
2484 | /* Refuse the same filesystem on the same mount point */ | |
2485 | err = -EBUSY; | |
95bc5f25 | 2486 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 AV |
2487 | path->mnt->mnt_root == path->dentry) |
2488 | goto unlock; | |
2489 | ||
2490 | err = -EINVAL; | |
e36cb0b8 | 2491 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
9d412a43 AV |
2492 | goto unlock; |
2493 | ||
95bc5f25 | 2494 | newmnt->mnt.mnt_flags = mnt_flags; |
84d17192 | 2495 | err = graft_tree(newmnt, parent, mp); |
9d412a43 AV |
2496 | |
2497 | unlock: | |
84d17192 | 2498 | unlock_mount(mp); |
9d412a43 AV |
2499 | return err; |
2500 | } | |
b1e75df4 | 2501 | |
8654df4e | 2502 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); |
1b852bce | 2503 | |
1da177e4 LT |
2504 | /* |
2505 | * create a new mount for userspace and request it to be added into the | |
2506 | * namespace's tree | |
2507 | */ | |
e462ec50 | 2508 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 2509 | int mnt_flags, const char *name, void *data) |
1da177e4 | 2510 | { |
0c55cfc4 | 2511 | struct file_system_type *type; |
1da177e4 | 2512 | struct vfsmount *mnt; |
15f9a3f3 | 2513 | int err; |
1da177e4 | 2514 | |
0c55cfc4 | 2515 | if (!fstype) |
1da177e4 LT |
2516 | return -EINVAL; |
2517 | ||
0c55cfc4 EB |
2518 | type = get_fs_type(fstype); |
2519 | if (!type) | |
2520 | return -ENODEV; | |
2521 | ||
e462ec50 | 2522 | mnt = vfs_kern_mount(type, sb_flags, name, data); |
0c55cfc4 EB |
2523 | if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && |
2524 | !mnt->mnt_sb->s_subtype) | |
2525 | mnt = fs_set_subtype(mnt, fstype); | |
2526 | ||
2527 | put_filesystem(type); | |
1da177e4 LT |
2528 | if (IS_ERR(mnt)) |
2529 | return PTR_ERR(mnt); | |
2530 | ||
8654df4e EB |
2531 | if (mount_too_revealing(mnt, &mnt_flags)) { |
2532 | mntput(mnt); | |
2533 | return -EPERM; | |
2534 | } | |
2535 | ||
95bc5f25 | 2536 | err = do_add_mount(real_mount(mnt), path, mnt_flags); |
15f9a3f3 AV |
2537 | if (err) |
2538 | mntput(mnt); | |
2539 | return err; | |
1da177e4 LT |
2540 | } |
2541 | ||
19a167af AV |
2542 | int finish_automount(struct vfsmount *m, struct path *path) |
2543 | { | |
6776db3d | 2544 | struct mount *mnt = real_mount(m); |
19a167af AV |
2545 | int err; |
2546 | /* The new mount record should have at least 2 refs to prevent it being | |
2547 | * expired before we get a chance to add it | |
2548 | */ | |
6776db3d | 2549 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
2550 | |
2551 | if (m->mnt_sb == path->mnt->mnt_sb && | |
2552 | m->mnt_root == path->dentry) { | |
b1e75df4 AV |
2553 | err = -ELOOP; |
2554 | goto fail; | |
19a167af AV |
2555 | } |
2556 | ||
95bc5f25 | 2557 | err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
b1e75df4 AV |
2558 | if (!err) |
2559 | return 0; | |
2560 | fail: | |
2561 | /* remove m from any expiration list it may be on */ | |
6776db3d | 2562 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 2563 | namespace_lock(); |
6776db3d | 2564 | list_del_init(&mnt->mnt_expire); |
97216be0 | 2565 | namespace_unlock(); |
19a167af | 2566 | } |
b1e75df4 AV |
2567 | mntput(m); |
2568 | mntput(m); | |
19a167af AV |
2569 | return err; |
2570 | } | |
2571 | ||
ea5b778a DH |
2572 | /** |
2573 | * mnt_set_expiry - Put a mount on an expiration list | |
2574 | * @mnt: The mount to list. | |
2575 | * @expiry_list: The list to add the mount to. | |
2576 | */ | |
2577 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
2578 | { | |
97216be0 | 2579 | namespace_lock(); |
ea5b778a | 2580 | |
6776db3d | 2581 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 2582 | |
97216be0 | 2583 | namespace_unlock(); |
ea5b778a DH |
2584 | } |
2585 | EXPORT_SYMBOL(mnt_set_expiry); | |
2586 | ||
1da177e4 LT |
2587 | /* |
2588 | * process a list of expirable mountpoints with the intent of discarding any | |
2589 | * mountpoints that aren't in use and haven't been touched since last we came | |
2590 | * here | |
2591 | */ | |
2592 | void mark_mounts_for_expiry(struct list_head *mounts) | |
2593 | { | |
761d5c38 | 2594 | struct mount *mnt, *next; |
1da177e4 LT |
2595 | LIST_HEAD(graveyard); |
2596 | ||
2597 | if (list_empty(mounts)) | |
2598 | return; | |
2599 | ||
97216be0 | 2600 | namespace_lock(); |
719ea2fb | 2601 | lock_mount_hash(); |
1da177e4 LT |
2602 | |
2603 | /* extract from the expiration list every vfsmount that matches the | |
2604 | * following criteria: | |
2605 | * - only referenced by its parent vfsmount | |
2606 | * - still marked for expiry (marked on the last call here; marks are | |
2607 | * cleared by mntput()) | |
2608 | */ | |
6776db3d | 2609 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 2610 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 2611 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 2612 | continue; |
6776db3d | 2613 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 2614 | } |
bcc5c7d2 | 2615 | while (!list_empty(&graveyard)) { |
6776db3d | 2616 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 2617 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 2618 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 2619 | } |
719ea2fb | 2620 | unlock_mount_hash(); |
3ab6abee | 2621 | namespace_unlock(); |
5528f911 TM |
2622 | } |
2623 | ||
2624 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
2625 | ||
2626 | /* | |
2627 | * Ripoff of 'select_parent()' | |
2628 | * | |
2629 | * search the list of submounts for a given mountpoint, and move any | |
2630 | * shrinkable submounts to the 'graveyard' list. | |
2631 | */ | |
692afc31 | 2632 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 2633 | { |
692afc31 | 2634 | struct mount *this_parent = parent; |
5528f911 TM |
2635 | struct list_head *next; |
2636 | int found = 0; | |
2637 | ||
2638 | repeat: | |
6b41d536 | 2639 | next = this_parent->mnt_mounts.next; |
5528f911 | 2640 | resume: |
6b41d536 | 2641 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 2642 | struct list_head *tmp = next; |
6b41d536 | 2643 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
2644 | |
2645 | next = tmp->next; | |
692afc31 | 2646 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 2647 | continue; |
5528f911 TM |
2648 | /* |
2649 | * Descend a level if the d_mounts list is non-empty. | |
2650 | */ | |
6b41d536 | 2651 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
2652 | this_parent = mnt; |
2653 | goto repeat; | |
2654 | } | |
1da177e4 | 2655 | |
1ab59738 | 2656 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 2657 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
2658 | found++; |
2659 | } | |
1da177e4 | 2660 | } |
5528f911 TM |
2661 | /* |
2662 | * All done at this level ... ascend and resume the search | |
2663 | */ | |
2664 | if (this_parent != parent) { | |
6b41d536 | 2665 | next = this_parent->mnt_child.next; |
0714a533 | 2666 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
2667 | goto resume; |
2668 | } | |
2669 | return found; | |
2670 | } | |
2671 | ||
2672 | /* | |
2673 | * process a list of expirable mountpoints with the intent of discarding any | |
2674 | * submounts of a specific parent mountpoint | |
99b7db7b | 2675 | * |
48a066e7 | 2676 | * mount_lock must be held for write |
5528f911 | 2677 | */ |
b54b9be7 | 2678 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
2679 | { |
2680 | LIST_HEAD(graveyard); | |
761d5c38 | 2681 | struct mount *m; |
5528f911 | 2682 | |
5528f911 | 2683 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 2684 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 2685 | while (!list_empty(&graveyard)) { |
761d5c38 | 2686 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 2687 | mnt_expire); |
143c8c91 | 2688 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 2689 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
2690 | } |
2691 | } | |
1da177e4 LT |
2692 | } |
2693 | ||
1da177e4 LT |
2694 | /* |
2695 | * Some copy_from_user() implementations do not return the exact number of | |
2696 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
2697 | * Note that this function differs from copy_from_user() in that it will oops | |
2698 | * on bad values of `to', rather than returning a short copy. | |
2699 | */ | |
b58fed8b RP |
2700 | static long exact_copy_from_user(void *to, const void __user * from, |
2701 | unsigned long n) | |
1da177e4 LT |
2702 | { |
2703 | char *t = to; | |
2704 | const char __user *f = from; | |
2705 | char c; | |
2706 | ||
2707 | if (!access_ok(VERIFY_READ, from, n)) | |
2708 | return n; | |
2709 | ||
2710 | while (n) { | |
2711 | if (__get_user(c, f)) { | |
2712 | memset(t, 0, n); | |
2713 | break; | |
2714 | } | |
2715 | *t++ = c; | |
2716 | f++; | |
2717 | n--; | |
2718 | } | |
2719 | return n; | |
2720 | } | |
2721 | ||
b40ef869 | 2722 | void *copy_mount_options(const void __user * data) |
1da177e4 LT |
2723 | { |
2724 | int i; | |
1da177e4 | 2725 | unsigned long size; |
b40ef869 | 2726 | char *copy; |
b58fed8b | 2727 | |
1da177e4 | 2728 | if (!data) |
b40ef869 | 2729 | return NULL; |
1da177e4 | 2730 | |
b40ef869 AV |
2731 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2732 | if (!copy) | |
2733 | return ERR_PTR(-ENOMEM); | |
1da177e4 LT |
2734 | |
2735 | /* We only care that *some* data at the address the user | |
2736 | * gave us is valid. Just in case, we'll zero | |
2737 | * the remainder of the page. | |
2738 | */ | |
2739 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
2740 | size = TASK_SIZE - (unsigned long)data; | |
2741 | if (size > PAGE_SIZE) | |
2742 | size = PAGE_SIZE; | |
2743 | ||
b40ef869 | 2744 | i = size - exact_copy_from_user(copy, data, size); |
1da177e4 | 2745 | if (!i) { |
b40ef869 AV |
2746 | kfree(copy); |
2747 | return ERR_PTR(-EFAULT); | |
1da177e4 LT |
2748 | } |
2749 | if (i != PAGE_SIZE) | |
b40ef869 AV |
2750 | memset(copy + i, 0, PAGE_SIZE - i); |
2751 | return copy; | |
1da177e4 LT |
2752 | } |
2753 | ||
b8850d1f | 2754 | char *copy_mount_string(const void __user *data) |
eca6f534 | 2755 | { |
b8850d1f | 2756 | return data ? strndup_user(data, PAGE_SIZE) : NULL; |
eca6f534 VN |
2757 | } |
2758 | ||
1da177e4 LT |
2759 | /* |
2760 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
2761 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
2762 | * | |
2763 | * data is a (void *) that can point to any structure up to | |
2764 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
2765 | * information (or be NULL). | |
2766 | * | |
2767 | * Pre-0.97 versions of mount() didn't have a flags word. | |
2768 | * When the flags word was introduced its top half was required | |
2769 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
2770 | * Therefore, if this magic number is present, it carries no information | |
2771 | * and must be discarded. | |
2772 | */ | |
5e6123f3 | 2773 | long do_mount(const char *dev_name, const char __user *dir_name, |
808d4e3c | 2774 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 2775 | { |
2d92ab3c | 2776 | struct path path; |
e462ec50 | 2777 | unsigned int mnt_flags = 0, sb_flags; |
1da177e4 | 2778 | int retval = 0; |
1da177e4 LT |
2779 | |
2780 | /* Discard magic */ | |
2781 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
2782 | flags &= ~MS_MGC_MSK; | |
2783 | ||
2784 | /* Basic sanity checks */ | |
1da177e4 LT |
2785 | if (data_page) |
2786 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
2787 | ||
e462ec50 DH |
2788 | if (flags & MS_NOUSER) |
2789 | return -EINVAL; | |
2790 | ||
a27ab9f2 | 2791 | /* ... and get the mountpoint */ |
5e6123f3 | 2792 | retval = user_path(dir_name, &path); |
a27ab9f2 TH |
2793 | if (retval) |
2794 | return retval; | |
2795 | ||
2796 | retval = security_sb_mount(dev_name, &path, | |
2797 | type_page, flags, data_page); | |
0d5cadb8 AV |
2798 | if (!retval && !may_mount()) |
2799 | retval = -EPERM; | |
e462ec50 | 2800 | if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) |
9e8925b6 | 2801 | retval = -EPERM; |
a27ab9f2 TH |
2802 | if (retval) |
2803 | goto dput_out; | |
2804 | ||
613cbe3d AK |
2805 | /* Default to relatime unless overriden */ |
2806 | if (!(flags & MS_NOATIME)) | |
2807 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 2808 | |
1da177e4 LT |
2809 | /* Separate the per-mountpoint flags */ |
2810 | if (flags & MS_NOSUID) | |
2811 | mnt_flags |= MNT_NOSUID; | |
2812 | if (flags & MS_NODEV) | |
2813 | mnt_flags |= MNT_NODEV; | |
2814 | if (flags & MS_NOEXEC) | |
2815 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
2816 | if (flags & MS_NOATIME) |
2817 | mnt_flags |= MNT_NOATIME; | |
2818 | if (flags & MS_NODIRATIME) | |
2819 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
2820 | if (flags & MS_STRICTATIME) |
2821 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
e462ec50 | 2822 | if (flags & SB_RDONLY) |
2e4b7fcd | 2823 | mnt_flags |= MNT_READONLY; |
fc33a7bb | 2824 | |
ffbc6f0e EB |
2825 | /* The default atime for remount is preservation */ |
2826 | if ((flags & MS_REMOUNT) && | |
2827 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
2828 | MS_STRICTATIME)) == 0)) { | |
2829 | mnt_flags &= ~MNT_ATIME_MASK; | |
2830 | mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; | |
2831 | } | |
2832 | ||
e462ec50 DH |
2833 | sb_flags = flags & (SB_RDONLY | |
2834 | SB_SYNCHRONOUS | | |
2835 | SB_MANDLOCK | | |
2836 | SB_DIRSYNC | | |
2837 | SB_SILENT | | |
917086ff | 2838 | SB_POSIXACL | |
d7ee9469 | 2839 | SB_LAZYTIME | |
917086ff | 2840 | SB_I_VERSION); |
1da177e4 | 2841 | |
1da177e4 | 2842 | if (flags & MS_REMOUNT) |
e462ec50 | 2843 | retval = do_remount(&path, flags, sb_flags, mnt_flags, |
1da177e4 LT |
2844 | data_page); |
2845 | else if (flags & MS_BIND) | |
2d92ab3c | 2846 | retval = do_loopback(&path, dev_name, flags & MS_REC); |
9676f0c6 | 2847 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
2d92ab3c | 2848 | retval = do_change_type(&path, flags); |
1da177e4 | 2849 | else if (flags & MS_MOVE) |
2d92ab3c | 2850 | retval = do_move_mount(&path, dev_name); |
1da177e4 | 2851 | else |
e462ec50 | 2852 | retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, |
1da177e4 LT |
2853 | dev_name, data_page); |
2854 | dput_out: | |
2d92ab3c | 2855 | path_put(&path); |
1da177e4 LT |
2856 | return retval; |
2857 | } | |
2858 | ||
537f7ccb EB |
2859 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
2860 | { | |
2861 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
2862 | } | |
2863 | ||
2864 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
2865 | { | |
2866 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
2867 | } | |
2868 | ||
771b1371 EB |
2869 | static void free_mnt_ns(struct mnt_namespace *ns) |
2870 | { | |
6344c433 | 2871 | ns_free_inum(&ns->ns); |
537f7ccb | 2872 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
2873 | put_user_ns(ns->user_ns); |
2874 | kfree(ns); | |
2875 | } | |
2876 | ||
8823c079 EB |
2877 | /* |
2878 | * Assign a sequence number so we can detect when we attempt to bind | |
2879 | * mount a reference to an older mount namespace into the current | |
2880 | * mount namespace, preventing reference counting loops. A 64bit | |
2881 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
2882 | * is effectively never, so we can ignore the possibility. | |
2883 | */ | |
2884 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
2885 | ||
771b1371 | 2886 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) |
cf8d2c11 TM |
2887 | { |
2888 | struct mnt_namespace *new_ns; | |
537f7ccb | 2889 | struct ucounts *ucounts; |
98f842e6 | 2890 | int ret; |
cf8d2c11 | 2891 | |
537f7ccb EB |
2892 | ucounts = inc_mnt_namespaces(user_ns); |
2893 | if (!ucounts) | |
df75e774 | 2894 | return ERR_PTR(-ENOSPC); |
537f7ccb | 2895 | |
cf8d2c11 | 2896 | new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); |
537f7ccb EB |
2897 | if (!new_ns) { |
2898 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 2899 | return ERR_PTR(-ENOMEM); |
537f7ccb | 2900 | } |
6344c433 | 2901 | ret = ns_alloc_inum(&new_ns->ns); |
98f842e6 EB |
2902 | if (ret) { |
2903 | kfree(new_ns); | |
537f7ccb | 2904 | dec_mnt_namespaces(ucounts); |
98f842e6 EB |
2905 | return ERR_PTR(ret); |
2906 | } | |
33c42940 | 2907 | new_ns->ns.ops = &mntns_operations; |
8823c079 | 2908 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); |
cf8d2c11 TM |
2909 | atomic_set(&new_ns->count, 1); |
2910 | new_ns->root = NULL; | |
2911 | INIT_LIST_HEAD(&new_ns->list); | |
2912 | init_waitqueue_head(&new_ns->poll); | |
2913 | new_ns->event = 0; | |
771b1371 | 2914 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 2915 | new_ns->ucounts = ucounts; |
d2921684 EB |
2916 | new_ns->mounts = 0; |
2917 | new_ns->pending_mounts = 0; | |
cf8d2c11 TM |
2918 | return new_ns; |
2919 | } | |
2920 | ||
0766f788 | 2921 | __latent_entropy |
9559f689 AV |
2922 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
2923 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 2924 | { |
6b3286ed | 2925 | struct mnt_namespace *new_ns; |
7f2da1e7 | 2926 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 2927 | struct mount *p, *q; |
9559f689 | 2928 | struct mount *old; |
cb338d06 | 2929 | struct mount *new; |
7a472ef4 | 2930 | int copy_flags; |
1da177e4 | 2931 | |
9559f689 AV |
2932 | BUG_ON(!ns); |
2933 | ||
2934 | if (likely(!(flags & CLONE_NEWNS))) { | |
2935 | get_mnt_ns(ns); | |
2936 | return ns; | |
2937 | } | |
2938 | ||
2939 | old = ns->root; | |
2940 | ||
771b1371 | 2941 | new_ns = alloc_mnt_ns(user_ns); |
cf8d2c11 TM |
2942 | if (IS_ERR(new_ns)) |
2943 | return new_ns; | |
1da177e4 | 2944 | |
97216be0 | 2945 | namespace_lock(); |
1da177e4 | 2946 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 2947 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 2948 | if (user_ns != ns->user_ns) |
132c94e3 | 2949 | copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; |
7a472ef4 | 2950 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 2951 | if (IS_ERR(new)) { |
328e6d90 | 2952 | namespace_unlock(); |
771b1371 | 2953 | free_mnt_ns(new_ns); |
be34d1a3 | 2954 | return ERR_CAST(new); |
1da177e4 | 2955 | } |
be08d6d2 | 2956 | new_ns->root = new; |
1a4eeaf2 | 2957 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
2958 | |
2959 | /* | |
2960 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
2961 | * as belonging to new namespace. We have already acquired a private | |
2962 | * fs_struct, so tsk->fs->lock is not needed. | |
2963 | */ | |
909b0a88 | 2964 | p = old; |
cb338d06 | 2965 | q = new; |
1da177e4 | 2966 | while (p) { |
143c8c91 | 2967 | q->mnt_ns = new_ns; |
d2921684 | 2968 | new_ns->mounts++; |
9559f689 AV |
2969 | if (new_fs) { |
2970 | if (&p->mnt == new_fs->root.mnt) { | |
2971 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 2972 | rootmnt = &p->mnt; |
1da177e4 | 2973 | } |
9559f689 AV |
2974 | if (&p->mnt == new_fs->pwd.mnt) { |
2975 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 2976 | pwdmnt = &p->mnt; |
1da177e4 | 2977 | } |
1da177e4 | 2978 | } |
909b0a88 AV |
2979 | p = next_mnt(p, old); |
2980 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
2981 | if (!q) |
2982 | break; | |
2983 | while (p->mnt.mnt_root != q->mnt.mnt_root) | |
2984 | p = next_mnt(p, old); | |
1da177e4 | 2985 | } |
328e6d90 | 2986 | namespace_unlock(); |
1da177e4 | 2987 | |
1da177e4 | 2988 | if (rootmnt) |
f03c6599 | 2989 | mntput(rootmnt); |
1da177e4 | 2990 | if (pwdmnt) |
f03c6599 | 2991 | mntput(pwdmnt); |
1da177e4 | 2992 | |
741a2951 | 2993 | return new_ns; |
1da177e4 LT |
2994 | } |
2995 | ||
cf8d2c11 TM |
2996 | /** |
2997 | * create_mnt_ns - creates a private namespace and adds a root filesystem | |
2998 | * @mnt: pointer to the new root filesystem mountpoint | |
2999 | */ | |
1a4eeaf2 | 3000 | static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) |
cf8d2c11 | 3001 | { |
771b1371 | 3002 | struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); |
cf8d2c11 | 3003 | if (!IS_ERR(new_ns)) { |
1a4eeaf2 AV |
3004 | struct mount *mnt = real_mount(m); |
3005 | mnt->mnt_ns = new_ns; | |
be08d6d2 | 3006 | new_ns->root = mnt; |
d2921684 | 3007 | new_ns->mounts++; |
b1983cd8 | 3008 | list_add(&mnt->mnt_list, &new_ns->list); |
c1334495 | 3009 | } else { |
1a4eeaf2 | 3010 | mntput(m); |
cf8d2c11 TM |
3011 | } |
3012 | return new_ns; | |
3013 | } | |
cf8d2c11 | 3014 | |
ea441d11 AV |
3015 | struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) |
3016 | { | |
3017 | struct mnt_namespace *ns; | |
d31da0f0 | 3018 | struct super_block *s; |
ea441d11 AV |
3019 | struct path path; |
3020 | int err; | |
3021 | ||
3022 | ns = create_mnt_ns(mnt); | |
3023 | if (IS_ERR(ns)) | |
3024 | return ERR_CAST(ns); | |
3025 | ||
3026 | err = vfs_path_lookup(mnt->mnt_root, mnt, | |
3027 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); | |
3028 | ||
3029 | put_mnt_ns(ns); | |
3030 | ||
3031 | if (err) | |
3032 | return ERR_PTR(err); | |
3033 | ||
3034 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3035 | s = path.mnt->mnt_sb; |
3036 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3037 | mntput(path.mnt); |
3038 | /* lock the sucker */ | |
d31da0f0 | 3039 | down_write(&s->s_umount); |
ea441d11 AV |
3040 | /* ... and return the root of (sub)tree on it */ |
3041 | return path.dentry; | |
3042 | } | |
3043 | EXPORT_SYMBOL(mount_subtree); | |
3044 | ||
bdc480e3 HC |
3045 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3046 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3047 | { |
eca6f534 VN |
3048 | int ret; |
3049 | char *kernel_type; | |
eca6f534 | 3050 | char *kernel_dev; |
b40ef869 | 3051 | void *options; |
1da177e4 | 3052 | |
b8850d1f TG |
3053 | kernel_type = copy_mount_string(type); |
3054 | ret = PTR_ERR(kernel_type); | |
3055 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3056 | goto out_type; |
1da177e4 | 3057 | |
b8850d1f TG |
3058 | kernel_dev = copy_mount_string(dev_name); |
3059 | ret = PTR_ERR(kernel_dev); | |
3060 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3061 | goto out_dev; |
1da177e4 | 3062 | |
b40ef869 AV |
3063 | options = copy_mount_options(data); |
3064 | ret = PTR_ERR(options); | |
3065 | if (IS_ERR(options)) | |
eca6f534 | 3066 | goto out_data; |
1da177e4 | 3067 | |
b40ef869 | 3068 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3069 | |
b40ef869 | 3070 | kfree(options); |
eca6f534 VN |
3071 | out_data: |
3072 | kfree(kernel_dev); | |
3073 | out_dev: | |
eca6f534 VN |
3074 | kfree(kernel_type); |
3075 | out_type: | |
3076 | return ret; | |
1da177e4 LT |
3077 | } |
3078 | ||
afac7cba AV |
3079 | /* |
3080 | * Return true if path is reachable from root | |
3081 | * | |
48a066e7 | 3082 | * namespace_sem or mount_lock is held |
afac7cba | 3083 | */ |
643822b4 | 3084 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3085 | const struct path *root) |
3086 | { | |
643822b4 | 3087 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3088 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3089 | mnt = mnt->mnt_parent; |
afac7cba | 3090 | } |
643822b4 | 3091 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3092 | } |
3093 | ||
640eb7e7 | 3094 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3095 | { |
25ab4c9b | 3096 | bool res; |
48a066e7 | 3097 | read_seqlock_excl(&mount_lock); |
643822b4 | 3098 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3099 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3100 | return res; |
3101 | } | |
3102 | EXPORT_SYMBOL(path_is_under); | |
3103 | ||
1da177e4 LT |
3104 | /* |
3105 | * pivot_root Semantics: | |
3106 | * Moves the root file system of the current process to the directory put_old, | |
3107 | * makes new_root as the new root file system of the current process, and sets | |
3108 | * root/cwd of all processes which had them on the current root to new_root. | |
3109 | * | |
3110 | * Restrictions: | |
3111 | * The new_root and put_old must be directories, and must not be on the | |
3112 | * same file system as the current process root. The put_old must be | |
3113 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3114 | * pointed to by put_old must yield the same directory as new_root. No other | |
3115 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3116 | * | |
4a0d11fa NB |
3117 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
3118 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
3119 | * in this situation. | |
3120 | * | |
1da177e4 LT |
3121 | * Notes: |
3122 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3123 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3124 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3125 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3126 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3127 | * first. | |
3128 | */ | |
3480b257 HC |
3129 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3130 | const char __user *, put_old) | |
1da177e4 | 3131 | { |
2d8f3038 | 3132 | struct path new, old, parent_path, root_parent, root; |
84d17192 AV |
3133 | struct mount *new_mnt, *root_mnt, *old_mnt; |
3134 | struct mountpoint *old_mp, *root_mp; | |
1da177e4 LT |
3135 | int error; |
3136 | ||
9b40bc90 | 3137 | if (!may_mount()) |
1da177e4 LT |
3138 | return -EPERM; |
3139 | ||
2d8f3038 | 3140 | error = user_path_dir(new_root, &new); |
1da177e4 LT |
3141 | if (error) |
3142 | goto out0; | |
1da177e4 | 3143 | |
2d8f3038 | 3144 | error = user_path_dir(put_old, &old); |
1da177e4 LT |
3145 | if (error) |
3146 | goto out1; | |
3147 | ||
2d8f3038 | 3148 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3149 | if (error) |
3150 | goto out2; | |
1da177e4 | 3151 | |
f7ad3c6b | 3152 | get_fs_root(current->fs, &root); |
84d17192 AV |
3153 | old_mp = lock_mount(&old); |
3154 | error = PTR_ERR(old_mp); | |
3155 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3156 | goto out3; |
3157 | ||
1da177e4 | 3158 | error = -EINVAL; |
419148da AV |
3159 | new_mnt = real_mount(new.mnt); |
3160 | root_mnt = real_mount(root.mnt); | |
84d17192 AV |
3161 | old_mnt = real_mount(old.mnt); |
3162 | if (IS_MNT_SHARED(old_mnt) || | |
fc7be130 AV |
3163 | IS_MNT_SHARED(new_mnt->mnt_parent) || |
3164 | IS_MNT_SHARED(root_mnt->mnt_parent)) | |
b12cea91 | 3165 | goto out4; |
143c8c91 | 3166 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3167 | goto out4; |
5ff9d8a6 EB |
3168 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3169 | goto out4; | |
1da177e4 | 3170 | error = -ENOENT; |
f3da392e | 3171 | if (d_unlinked(new.dentry)) |
b12cea91 | 3172 | goto out4; |
1da177e4 | 3173 | error = -EBUSY; |
84d17192 | 3174 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3175 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3176 | error = -EINVAL; |
8c3ee42e | 3177 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3178 | goto out4; /* not a mountpoint */ |
676da58d | 3179 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3180 | goto out4; /* not attached */ |
84d17192 | 3181 | root_mp = root_mnt->mnt_mp; |
2d8f3038 | 3182 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3183 | goto out4; /* not a mountpoint */ |
676da58d | 3184 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3185 | goto out4; /* not attached */ |
4ac91378 | 3186 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3187 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3188 | goto out4; |
0d082601 EB |
3189 | /* make certain new is below the root */ |
3190 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3191 | goto out4; | |
84d17192 | 3192 | root_mp->m_count++; /* pin it so it won't go away */ |
719ea2fb | 3193 | lock_mount_hash(); |
419148da AV |
3194 | detach_mnt(new_mnt, &parent_path); |
3195 | detach_mnt(root_mnt, &root_parent); | |
5ff9d8a6 EB |
3196 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3197 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3198 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3199 | } | |
4ac91378 | 3200 | /* mount old root on put_old */ |
84d17192 | 3201 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3202 | /* mount new_root on / */ |
84d17192 | 3203 | attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); |
6b3286ed | 3204 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3205 | /* A moved mount should not expire automatically */ |
3206 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3207 | put_mountpoint(root_mp); |
719ea2fb | 3208 | unlock_mount_hash(); |
2d8f3038 | 3209 | chroot_fs_refs(&root, &new); |
1da177e4 | 3210 | error = 0; |
b12cea91 | 3211 | out4: |
84d17192 | 3212 | unlock_mount(old_mp); |
b12cea91 AV |
3213 | if (!error) { |
3214 | path_put(&root_parent); | |
3215 | path_put(&parent_path); | |
3216 | } | |
3217 | out3: | |
8c3ee42e | 3218 | path_put(&root); |
b12cea91 | 3219 | out2: |
2d8f3038 | 3220 | path_put(&old); |
1da177e4 | 3221 | out1: |
2d8f3038 | 3222 | path_put(&new); |
1da177e4 | 3223 | out0: |
1da177e4 | 3224 | return error; |
1da177e4 LT |
3225 | } |
3226 | ||
3227 | static void __init init_mount_tree(void) | |
3228 | { | |
3229 | struct vfsmount *mnt; | |
6b3286ed | 3230 | struct mnt_namespace *ns; |
ac748a09 | 3231 | struct path root; |
0c55cfc4 | 3232 | struct file_system_type *type; |
1da177e4 | 3233 | |
0c55cfc4 EB |
3234 | type = get_fs_type("rootfs"); |
3235 | if (!type) | |
3236 | panic("Can't find rootfs type"); | |
3237 | mnt = vfs_kern_mount(type, 0, "rootfs", NULL); | |
3238 | put_filesystem(type); | |
1da177e4 LT |
3239 | if (IS_ERR(mnt)) |
3240 | panic("Can't create rootfs"); | |
b3e19d92 | 3241 | |
3b22edc5 TM |
3242 | ns = create_mnt_ns(mnt); |
3243 | if (IS_ERR(ns)) | |
1da177e4 | 3244 | panic("Can't allocate initial namespace"); |
6b3286ed KK |
3245 | |
3246 | init_task.nsproxy->mnt_ns = ns; | |
3247 | get_mnt_ns(ns); | |
3248 | ||
be08d6d2 AV |
3249 | root.mnt = mnt; |
3250 | root.dentry = mnt->mnt_root; | |
da362b09 | 3251 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
3252 | |
3253 | set_fs_pwd(current->fs, &root); | |
3254 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
3255 | } |
3256 | ||
74bf17cf | 3257 | void __init mnt_init(void) |
1da177e4 | 3258 | { |
15a67dd8 | 3259 | int err; |
1da177e4 | 3260 | |
7d6fec45 | 3261 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
20c2df83 | 3262 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); |
1da177e4 | 3263 | |
0818bf27 | 3264 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 3265 | sizeof(struct hlist_head), |
0818bf27 | 3266 | mhash_entries, 19, |
3d375d78 | 3267 | HASH_ZERO, |
0818bf27 AV |
3268 | &m_hash_shift, &m_hash_mask, 0, 0); |
3269 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
3270 | sizeof(struct hlist_head), | |
3271 | mphash_entries, 19, | |
3d375d78 | 3272 | HASH_ZERO, |
0818bf27 | 3273 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 3274 | |
84d17192 | 3275 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
3276 | panic("Failed to allocate mount hash table\n"); |
3277 | ||
4b93dc9b TH |
3278 | kernfs_init(); |
3279 | ||
15a67dd8 RD |
3280 | err = sysfs_init(); |
3281 | if (err) | |
3282 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 3283 | __func__, err); |
00d26666 GKH |
3284 | fs_kobj = kobject_create_and_add("fs", NULL); |
3285 | if (!fs_kobj) | |
8e24eea7 | 3286 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
1da177e4 LT |
3287 | init_rootfs(); |
3288 | init_mount_tree(); | |
3289 | } | |
3290 | ||
616511d0 | 3291 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 3292 | { |
d498b25a | 3293 | if (!atomic_dec_and_test(&ns->count)) |
616511d0 | 3294 | return; |
7b00ed6f | 3295 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 3296 | free_mnt_ns(ns); |
1da177e4 | 3297 | } |
9d412a43 AV |
3298 | |
3299 | struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) | |
3300 | { | |
423e0ab0 | 3301 | struct vfsmount *mnt; |
e462ec50 | 3302 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); |
423e0ab0 TC |
3303 | if (!IS_ERR(mnt)) { |
3304 | /* | |
3305 | * it is a longterm mount, don't release mnt until | |
3306 | * we unmount before file sys is unregistered | |
3307 | */ | |
f7a99c5b | 3308 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
3309 | } |
3310 | return mnt; | |
9d412a43 AV |
3311 | } |
3312 | EXPORT_SYMBOL_GPL(kern_mount_data); | |
423e0ab0 TC |
3313 | |
3314 | void kern_unmount(struct vfsmount *mnt) | |
3315 | { | |
3316 | /* release long term mount so mount point can be released */ | |
3317 | if (!IS_ERR_OR_NULL(mnt)) { | |
f7a99c5b | 3318 | real_mount(mnt)->mnt_ns = NULL; |
48a066e7 | 3319 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
3320 | mntput(mnt); |
3321 | } | |
3322 | } | |
3323 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 AV |
3324 | |
3325 | bool our_mnt(struct vfsmount *mnt) | |
3326 | { | |
143c8c91 | 3327 | return check_mnt(real_mount(mnt)); |
02125a82 | 3328 | } |
8823c079 | 3329 | |
3151527e EB |
3330 | bool current_chrooted(void) |
3331 | { | |
3332 | /* Does the current process have a non-standard root */ | |
3333 | struct path ns_root; | |
3334 | struct path fs_root; | |
3335 | bool chrooted; | |
3336 | ||
3337 | /* Find the namespace root */ | |
3338 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
3339 | ns_root.dentry = ns_root.mnt->mnt_root; | |
3340 | path_get(&ns_root); | |
3341 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
3342 | ; | |
3343 | ||
3344 | get_fs_root(current->fs, &fs_root); | |
3345 | ||
3346 | chrooted = !path_equal(&fs_root, &ns_root); | |
3347 | ||
3348 | path_put(&fs_root); | |
3349 | path_put(&ns_root); | |
3350 | ||
3351 | return chrooted; | |
3352 | } | |
3353 | ||
8654df4e EB |
3354 | static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, |
3355 | int *new_mnt_flags) | |
87a8ebd6 | 3356 | { |
8c6cf9cc | 3357 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 3358 | struct mount *mnt; |
e51db735 | 3359 | bool visible = false; |
87a8ebd6 | 3360 | |
44bb4385 | 3361 | down_read(&namespace_sem); |
87a8ebd6 | 3362 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 3363 | struct mount *child; |
77b1a97d EB |
3364 | int mnt_flags; |
3365 | ||
8654df4e | 3366 | if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) |
e51db735 EB |
3367 | continue; |
3368 | ||
7e96c1b0 EB |
3369 | /* This mount is not fully visible if it's root directory |
3370 | * is not the root directory of the filesystem. | |
3371 | */ | |
3372 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
3373 | continue; | |
3374 | ||
a1935c17 | 3375 | /* A local view of the mount flags */ |
77b1a97d | 3376 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 3377 | |
695e9df0 | 3378 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 3379 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
3380 | mnt_flags |= MNT_LOCK_READONLY; |
3381 | ||
8c6cf9cc EB |
3382 | /* Verify the mount flags are equal to or more permissive |
3383 | * than the proposed new mount. | |
3384 | */ | |
77b1a97d | 3385 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
3386 | !(new_flags & MNT_READONLY)) |
3387 | continue; | |
77b1a97d EB |
3388 | if ((mnt_flags & MNT_LOCK_ATIME) && |
3389 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
3390 | continue; |
3391 | ||
ceeb0e5d EB |
3392 | /* This mount is not fully visible if there are any |
3393 | * locked child mounts that cover anything except for | |
3394 | * empty directories. | |
e51db735 EB |
3395 | */ |
3396 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
3397 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 3398 | /* Only worry about locked mounts */ |
d71ed6c9 | 3399 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 3400 | continue; |
7236c85e EB |
3401 | /* Is the directory permanetly empty? */ |
3402 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 3403 | goto next; |
87a8ebd6 | 3404 | } |
8c6cf9cc | 3405 | /* Preserve the locked attributes */ |
77b1a97d | 3406 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 3407 | MNT_LOCK_ATIME); |
e51db735 EB |
3408 | visible = true; |
3409 | goto found; | |
3410 | next: ; | |
87a8ebd6 | 3411 | } |
e51db735 | 3412 | found: |
44bb4385 | 3413 | up_read(&namespace_sem); |
e51db735 | 3414 | return visible; |
87a8ebd6 EB |
3415 | } |
3416 | ||
8654df4e EB |
3417 | static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) |
3418 | { | |
a1935c17 | 3419 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
3420 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
3421 | unsigned long s_iflags; | |
3422 | ||
3423 | if (ns->user_ns == &init_user_ns) | |
3424 | return false; | |
3425 | ||
3426 | /* Can this filesystem be too revealing? */ | |
3427 | s_iflags = mnt->mnt_sb->s_iflags; | |
3428 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) | |
3429 | return false; | |
3430 | ||
a1935c17 EB |
3431 | if ((s_iflags & required_iflags) != required_iflags) { |
3432 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
3433 | required_iflags); | |
3434 | return true; | |
3435 | } | |
3436 | ||
8654df4e EB |
3437 | return !mnt_already_visible(ns, mnt, new_mnt_flags); |
3438 | } | |
3439 | ||
380cf5ba AL |
3440 | bool mnt_may_suid(struct vfsmount *mnt) |
3441 | { | |
3442 | /* | |
3443 | * Foreign mounts (accessed via fchdir or through /proc | |
3444 | * symlinks) are always treated as if they are nosuid. This | |
3445 | * prevents namespaces from trusting potentially unsafe | |
3446 | * suid/sgid bits, file caps, or security labels that originate | |
3447 | * in other namespaces. | |
3448 | */ | |
3449 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
3450 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
3451 | } | |
3452 | ||
64964528 | 3453 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 3454 | { |
58be2825 | 3455 | struct ns_common *ns = NULL; |
8823c079 EB |
3456 | struct nsproxy *nsproxy; |
3457 | ||
728dba3a EB |
3458 | task_lock(task); |
3459 | nsproxy = task->nsproxy; | |
8823c079 | 3460 | if (nsproxy) { |
58be2825 AV |
3461 | ns = &nsproxy->mnt_ns->ns; |
3462 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 3463 | } |
728dba3a | 3464 | task_unlock(task); |
8823c079 EB |
3465 | |
3466 | return ns; | |
3467 | } | |
3468 | ||
64964528 | 3469 | static void mntns_put(struct ns_common *ns) |
8823c079 | 3470 | { |
58be2825 | 3471 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
3472 | } |
3473 | ||
64964528 | 3474 | static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) |
8823c079 EB |
3475 | { |
3476 | struct fs_struct *fs = current->fs; | |
4f757f3c | 3477 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
8823c079 | 3478 | struct path root; |
4f757f3c | 3479 | int err; |
8823c079 | 3480 | |
0c55cfc4 | 3481 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
c7b96acf EB |
3482 | !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || |
3483 | !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) | |
ae11e0f1 | 3484 | return -EPERM; |
8823c079 EB |
3485 | |
3486 | if (fs->users != 1) | |
3487 | return -EINVAL; | |
3488 | ||
3489 | get_mnt_ns(mnt_ns); | |
4f757f3c | 3490 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
3491 | nsproxy->mnt_ns = mnt_ns; |
3492 | ||
3493 | /* Find the root */ | |
4f757f3c AV |
3494 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
3495 | "/", LOOKUP_DOWN, &root); | |
3496 | if (err) { | |
3497 | /* revert to old namespace */ | |
3498 | nsproxy->mnt_ns = old_mnt_ns; | |
3499 | put_mnt_ns(mnt_ns); | |
3500 | return err; | |
3501 | } | |
8823c079 | 3502 | |
4068367c AV |
3503 | put_mnt_ns(old_mnt_ns); |
3504 | ||
8823c079 EB |
3505 | /* Update the pwd and root */ |
3506 | set_fs_pwd(fs, &root); | |
3507 | set_fs_root(fs, &root); | |
3508 | ||
3509 | path_put(&root); | |
3510 | return 0; | |
3511 | } | |
3512 | ||
bcac25a5 AV |
3513 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
3514 | { | |
3515 | return to_mnt_ns(ns)->user_ns; | |
3516 | } | |
3517 | ||
8823c079 EB |
3518 | const struct proc_ns_operations mntns_operations = { |
3519 | .name = "mnt", | |
3520 | .type = CLONE_NEWNS, | |
3521 | .get = mntns_get, | |
3522 | .put = mntns_put, | |
3523 | .install = mntns_install, | |
bcac25a5 | 3524 | .owner = mntns_owner, |
8823c079 | 3525 | }; |