]>
Commit | Line | Data |
---|---|---|
59bd9ded | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/namespace.c | |
4 | * | |
5 | * (C) Copyright Al Viro 2000, 2001 | |
1da177e4 LT |
6 | * |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 | 11 | #include <linux/syscalls.h> |
d10577a8 | 12 | #include <linux/export.h> |
16f7e0fe | 13 | #include <linux/capability.h> |
6b3286ed | 14 | #include <linux/mnt_namespace.h> |
771b1371 | 15 | #include <linux/user_namespace.h> |
1da177e4 LT |
16 | #include <linux/namei.h> |
17 | #include <linux/security.h> | |
5b825c3a | 18 | #include <linux/cred.h> |
73cd49ec | 19 | #include <linux/idr.h> |
57f150a5 | 20 | #include <linux/init.h> /* init_rootfs */ |
d10577a8 AV |
21 | #include <linux/fs_struct.h> /* get_fs_root et.al. */ |
22 | #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ | |
a07b2000 | 23 | #include <linux/file.h> |
d10577a8 | 24 | #include <linux/uaccess.h> |
0bb80f24 | 25 | #include <linux/proc_ns.h> |
20b4fb48 | 26 | #include <linux/magic.h> |
57c8a661 | 27 | #include <linux/memblock.h> |
9caccd41 | 28 | #include <linux/proc_fs.h> |
9ea459e1 | 29 | #include <linux/task_work.h> |
9164bb4a | 30 | #include <linux/sched/task.h> |
e262e32d | 31 | #include <uapi/linux/mount.h> |
9bc61ab1 | 32 | #include <linux/fs_context.h> |
037f11b4 | 33 | #include <linux/shmem_fs.h> |
bd303368 | 34 | #include <linux/mnt_idmapping.h> |
9164bb4a | 35 | |
07b20889 | 36 | #include "pnode.h" |
948730b0 | 37 | #include "internal.h" |
1da177e4 | 38 | |
d2921684 | 39 | /* Maximum number of mounts in a mount namespace */ |
ab171b95 | 40 | static unsigned int sysctl_mount_max __read_mostly = 100000; |
d2921684 | 41 | |
0818bf27 AV |
42 | static unsigned int m_hash_mask __read_mostly; |
43 | static unsigned int m_hash_shift __read_mostly; | |
44 | static unsigned int mp_hash_mask __read_mostly; | |
45 | static unsigned int mp_hash_shift __read_mostly; | |
46 | ||
47 | static __initdata unsigned long mhash_entries; | |
48 | static int __init set_mhash_entries(char *str) | |
49 | { | |
50 | if (!str) | |
51 | return 0; | |
52 | mhash_entries = simple_strtoul(str, &str, 0); | |
53 | return 1; | |
54 | } | |
55 | __setup("mhash_entries=", set_mhash_entries); | |
56 | ||
57 | static __initdata unsigned long mphash_entries; | |
58 | static int __init set_mphash_entries(char *str) | |
59 | { | |
60 | if (!str) | |
61 | return 0; | |
62 | mphash_entries = simple_strtoul(str, &str, 0); | |
63 | return 1; | |
64 | } | |
65 | __setup("mphash_entries=", set_mphash_entries); | |
13f14b4d | 66 | |
c7999c36 | 67 | static u64 event; |
73cd49ec | 68 | static DEFINE_IDA(mnt_id_ida); |
719f5d7f | 69 | static DEFINE_IDA(mnt_group_ida); |
1da177e4 | 70 | |
38129a13 | 71 | static struct hlist_head *mount_hashtable __read_mostly; |
0818bf27 | 72 | static struct hlist_head *mountpoint_hashtable __read_mostly; |
e18b890b | 73 | static struct kmem_cache *mnt_cache __read_mostly; |
59aa0da8 | 74 | static DECLARE_RWSEM(namespace_sem); |
4edbe133 AV |
75 | static HLIST_HEAD(unmounted); /* protected by namespace_sem */ |
76 | static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ | |
1da177e4 | 77 | |
2a186721 CB |
78 | struct mount_kattr { |
79 | unsigned int attr_set; | |
80 | unsigned int attr_clr; | |
81 | unsigned int propagation; | |
82 | unsigned int lookup_flags; | |
83 | bool recurse; | |
9caccd41 | 84 | struct user_namespace *mnt_userns; |
256c8aed | 85 | struct mnt_idmap *mnt_idmap; |
2a186721 CB |
86 | }; |
87 | ||
f87fd4c2 | 88 | /* /sys/fs */ |
00d26666 GKH |
89 | struct kobject *fs_kobj; |
90 | EXPORT_SYMBOL_GPL(fs_kobj); | |
f87fd4c2 | 91 | |
99b7db7b NP |
92 | /* |
93 | * vfsmount lock may be taken for read to prevent changes to the | |
94 | * vfsmount hash, ie. during mountpoint lookups or walking back | |
95 | * up the tree. | |
96 | * | |
97 | * It should be taken for write in all cases where the vfsmount | |
98 | * tree or hash is modified or when a vfsmount structure is modified. | |
99 | */ | |
48a066e7 | 100 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); |
99b7db7b | 101 | |
d033cb67 CB |
102 | static inline void lock_mount_hash(void) |
103 | { | |
104 | write_seqlock(&mount_lock); | |
105 | } | |
106 | ||
107 | static inline void unlock_mount_hash(void) | |
108 | { | |
109 | write_sequnlock(&mount_lock); | |
110 | } | |
111 | ||
38129a13 | 112 | static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 113 | { |
b58fed8b RP |
114 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
115 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
0818bf27 AV |
116 | tmp = tmp + (tmp >> m_hash_shift); |
117 | return &mount_hashtable[tmp & m_hash_mask]; | |
118 | } | |
119 | ||
120 | static inline struct hlist_head *mp_hash(struct dentry *dentry) | |
121 | { | |
122 | unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); | |
123 | tmp = tmp + (tmp >> mp_hash_shift); | |
124 | return &mountpoint_hashtable[tmp & mp_hash_mask]; | |
1da177e4 LT |
125 | } |
126 | ||
b105e270 | 127 | static int mnt_alloc_id(struct mount *mnt) |
73cd49ec | 128 | { |
169b480e MW |
129 | int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); |
130 | ||
131 | if (res < 0) | |
132 | return res; | |
133 | mnt->mnt_id = res; | |
134 | return 0; | |
73cd49ec MS |
135 | } |
136 | ||
b105e270 | 137 | static void mnt_free_id(struct mount *mnt) |
73cd49ec | 138 | { |
169b480e | 139 | ida_free(&mnt_id_ida, mnt->mnt_id); |
73cd49ec MS |
140 | } |
141 | ||
719f5d7f MS |
142 | /* |
143 | * Allocate a new peer group ID | |
719f5d7f | 144 | */ |
4b8b21f4 | 145 | static int mnt_alloc_group_id(struct mount *mnt) |
719f5d7f | 146 | { |
169b480e | 147 | int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); |
f21f6220 | 148 | |
169b480e MW |
149 | if (res < 0) |
150 | return res; | |
151 | mnt->mnt_group_id = res; | |
152 | return 0; | |
719f5d7f MS |
153 | } |
154 | ||
155 | /* | |
156 | * Release a peer group ID | |
157 | */ | |
4b8b21f4 | 158 | void mnt_release_group_id(struct mount *mnt) |
719f5d7f | 159 | { |
169b480e | 160 | ida_free(&mnt_group_ida, mnt->mnt_group_id); |
15169fe7 | 161 | mnt->mnt_group_id = 0; |
719f5d7f MS |
162 | } |
163 | ||
b3e19d92 NP |
164 | /* |
165 | * vfsmount lock must be held for read | |
166 | */ | |
83adc753 | 167 | static inline void mnt_add_count(struct mount *mnt, int n) |
b3e19d92 NP |
168 | { |
169 | #ifdef CONFIG_SMP | |
68e8a9fe | 170 | this_cpu_add(mnt->mnt_pcp->mnt_count, n); |
b3e19d92 NP |
171 | #else |
172 | preempt_disable(); | |
68e8a9fe | 173 | mnt->mnt_count += n; |
b3e19d92 NP |
174 | preempt_enable(); |
175 | #endif | |
176 | } | |
177 | ||
b3e19d92 NP |
178 | /* |
179 | * vfsmount lock must be held for write | |
180 | */ | |
edf7ddbf | 181 | int mnt_get_count(struct mount *mnt) |
b3e19d92 NP |
182 | { |
183 | #ifdef CONFIG_SMP | |
edf7ddbf | 184 | int count = 0; |
b3e19d92 NP |
185 | int cpu; |
186 | ||
187 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 188 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; |
b3e19d92 NP |
189 | } |
190 | ||
191 | return count; | |
192 | #else | |
68e8a9fe | 193 | return mnt->mnt_count; |
b3e19d92 NP |
194 | #endif |
195 | } | |
196 | ||
b105e270 | 197 | static struct mount *alloc_vfsmnt(const char *name) |
1da177e4 | 198 | { |
c63181e6 AV |
199 | struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); |
200 | if (mnt) { | |
73cd49ec MS |
201 | int err; |
202 | ||
c63181e6 | 203 | err = mnt_alloc_id(mnt); |
88b38782 LZ |
204 | if (err) |
205 | goto out_free_cache; | |
206 | ||
207 | if (name) { | |
79f6540b VA |
208 | mnt->mnt_devname = kstrdup_const(name, |
209 | GFP_KERNEL_ACCOUNT); | |
c63181e6 | 210 | if (!mnt->mnt_devname) |
88b38782 | 211 | goto out_free_id; |
73cd49ec MS |
212 | } |
213 | ||
b3e19d92 | 214 | #ifdef CONFIG_SMP |
c63181e6 AV |
215 | mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); |
216 | if (!mnt->mnt_pcp) | |
b3e19d92 NP |
217 | goto out_free_devname; |
218 | ||
c63181e6 | 219 | this_cpu_add(mnt->mnt_pcp->mnt_count, 1); |
b3e19d92 | 220 | #else |
c63181e6 AV |
221 | mnt->mnt_count = 1; |
222 | mnt->mnt_writers = 0; | |
b3e19d92 NP |
223 | #endif |
224 | ||
38129a13 | 225 | INIT_HLIST_NODE(&mnt->mnt_hash); |
c63181e6 AV |
226 | INIT_LIST_HEAD(&mnt->mnt_child); |
227 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
228 | INIT_LIST_HEAD(&mnt->mnt_list); | |
229 | INIT_LIST_HEAD(&mnt->mnt_expire); | |
230 | INIT_LIST_HEAD(&mnt->mnt_share); | |
231 | INIT_LIST_HEAD(&mnt->mnt_slave_list); | |
232 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
0a5eb7c8 | 233 | INIT_HLIST_NODE(&mnt->mnt_mp_list); |
99b19d16 | 234 | INIT_LIST_HEAD(&mnt->mnt_umounting); |
56cbb429 | 235 | INIT_HLIST_HEAD(&mnt->mnt_stuck_children); |
256c8aed | 236 | mnt->mnt.mnt_idmap = &nop_mnt_idmap; |
1da177e4 | 237 | } |
c63181e6 | 238 | return mnt; |
88b38782 | 239 | |
d3ef3d73 NP |
240 | #ifdef CONFIG_SMP |
241 | out_free_devname: | |
fcc139ae | 242 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 243 | #endif |
88b38782 | 244 | out_free_id: |
c63181e6 | 245 | mnt_free_id(mnt); |
88b38782 | 246 | out_free_cache: |
c63181e6 | 247 | kmem_cache_free(mnt_cache, mnt); |
88b38782 | 248 | return NULL; |
1da177e4 LT |
249 | } |
250 | ||
3d733633 DH |
251 | /* |
252 | * Most r/o checks on a fs are for operations that take | |
253 | * discrete amounts of time, like a write() or unlink(). | |
254 | * We must keep track of when those operations start | |
255 | * (for permission checks) and when they end, so that | |
256 | * we can determine when writes are able to occur to | |
257 | * a filesystem. | |
258 | */ | |
259 | /* | |
260 | * __mnt_is_readonly: check whether a mount is read-only | |
261 | * @mnt: the mount to check for its write status | |
262 | * | |
263 | * This shouldn't be used directly ouside of the VFS. | |
264 | * It does not guarantee that the filesystem will stay | |
265 | * r/w, just that it is right *now*. This can not and | |
266 | * should not be used in place of IS_RDONLY(inode). | |
267 | * mnt_want/drop_write() will _keep_ the filesystem | |
268 | * r/w. | |
269 | */ | |
43f5e655 | 270 | bool __mnt_is_readonly(struct vfsmount *mnt) |
3d733633 | 271 | { |
43f5e655 | 272 | return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); |
3d733633 DH |
273 | } |
274 | EXPORT_SYMBOL_GPL(__mnt_is_readonly); | |
275 | ||
83adc753 | 276 | static inline void mnt_inc_writers(struct mount *mnt) |
d3ef3d73 NP |
277 | { |
278 | #ifdef CONFIG_SMP | |
68e8a9fe | 279 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 280 | #else |
68e8a9fe | 281 | mnt->mnt_writers++; |
d3ef3d73 NP |
282 | #endif |
283 | } | |
3d733633 | 284 | |
83adc753 | 285 | static inline void mnt_dec_writers(struct mount *mnt) |
3d733633 | 286 | { |
d3ef3d73 | 287 | #ifdef CONFIG_SMP |
68e8a9fe | 288 | this_cpu_dec(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 289 | #else |
68e8a9fe | 290 | mnt->mnt_writers--; |
d3ef3d73 | 291 | #endif |
3d733633 | 292 | } |
3d733633 | 293 | |
83adc753 | 294 | static unsigned int mnt_get_writers(struct mount *mnt) |
3d733633 | 295 | { |
d3ef3d73 NP |
296 | #ifdef CONFIG_SMP |
297 | unsigned int count = 0; | |
3d733633 | 298 | int cpu; |
3d733633 DH |
299 | |
300 | for_each_possible_cpu(cpu) { | |
68e8a9fe | 301 | count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; |
3d733633 | 302 | } |
3d733633 | 303 | |
d3ef3d73 NP |
304 | return count; |
305 | #else | |
306 | return mnt->mnt_writers; | |
307 | #endif | |
3d733633 DH |
308 | } |
309 | ||
4ed5e82f MS |
310 | static int mnt_is_readonly(struct vfsmount *mnt) |
311 | { | |
312 | if (mnt->mnt_sb->s_readonly_remount) | |
313 | return 1; | |
314 | /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ | |
315 | smp_rmb(); | |
316 | return __mnt_is_readonly(mnt); | |
317 | } | |
318 | ||
8366025e | 319 | /* |
eb04c282 JK |
320 | * Most r/o & frozen checks on a fs are for operations that take discrete |
321 | * amounts of time, like a write() or unlink(). We must keep track of when | |
322 | * those operations start (for permission checks) and when they end, so that we | |
323 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
324 | */ |
325 | /** | |
eb04c282 | 326 | * __mnt_want_write - get write access to a mount without freeze protection |
83adc753 | 327 | * @m: the mount on which to take a write |
8366025e | 328 | * |
eb04c282 JK |
329 | * This tells the low-level filesystem that a write is about to be performed to |
330 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
331 | * returning success. This operation does not protect against filesystem being | |
332 | * frozen. When the write operation is finished, __mnt_drop_write() must be | |
333 | * called. This is effectively a refcount. | |
8366025e | 334 | */ |
eb04c282 | 335 | int __mnt_want_write(struct vfsmount *m) |
8366025e | 336 | { |
83adc753 | 337 | struct mount *mnt = real_mount(m); |
3d733633 | 338 | int ret = 0; |
3d733633 | 339 | |
d3ef3d73 | 340 | preempt_disable(); |
c6653a83 | 341 | mnt_inc_writers(mnt); |
d3ef3d73 | 342 | /* |
c6653a83 | 343 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 NP |
344 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
345 | * incremented count after it has set MNT_WRITE_HOLD. | |
346 | */ | |
347 | smp_mb(); | |
0f8821da SAS |
348 | might_lock(&mount_lock.lock); |
349 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) { | |
350 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
351 | cpu_relax(); | |
352 | } else { | |
353 | /* | |
354 | * This prevents priority inversion, if the task | |
355 | * setting MNT_WRITE_HOLD got preempted on a remote | |
356 | * CPU, and it prevents life lock if the task setting | |
357 | * MNT_WRITE_HOLD has a lower priority and is bound to | |
358 | * the same CPU as the task that is spinning here. | |
359 | */ | |
360 | preempt_enable(); | |
361 | lock_mount_hash(); | |
362 | unlock_mount_hash(); | |
363 | preempt_disable(); | |
364 | } | |
365 | } | |
d3ef3d73 NP |
366 | /* |
367 | * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will | |
368 | * be set to match its requirements. So we must not load that until | |
369 | * MNT_WRITE_HOLD is cleared. | |
370 | */ | |
371 | smp_rmb(); | |
4ed5e82f | 372 | if (mnt_is_readonly(m)) { |
c6653a83 | 373 | mnt_dec_writers(mnt); |
3d733633 | 374 | ret = -EROFS; |
3d733633 | 375 | } |
d3ef3d73 | 376 | preempt_enable(); |
eb04c282 JK |
377 | |
378 | return ret; | |
379 | } | |
380 | ||
381 | /** | |
382 | * mnt_want_write - get write access to a mount | |
383 | * @m: the mount on which to take a write | |
384 | * | |
385 | * This tells the low-level filesystem that a write is about to be performed to | |
386 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
387 | * is not frozen) before returning success. When the write operation is | |
388 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
389 | */ | |
390 | int mnt_want_write(struct vfsmount *m) | |
391 | { | |
392 | int ret; | |
393 | ||
394 | sb_start_write(m->mnt_sb); | |
395 | ret = __mnt_want_write(m); | |
396 | if (ret) | |
397 | sb_end_write(m->mnt_sb); | |
3d733633 | 398 | return ret; |
8366025e DH |
399 | } |
400 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
401 | ||
96029c4e | 402 | /** |
eb04c282 | 403 | * __mnt_want_write_file - get write access to a file's mount |
96029c4e NP |
404 | * @file: the file who's mount on which to take a write |
405 | * | |
14e43bf4 EB |
406 | * This is like __mnt_want_write, but if the file is already open for writing it |
407 | * skips incrementing mnt_writers (since the open file already has a reference) | |
408 | * and instead only does the check for emergency r/o remounts. This must be | |
409 | * paired with __mnt_drop_write_file. | |
96029c4e | 410 | */ |
eb04c282 | 411 | int __mnt_want_write_file(struct file *file) |
96029c4e | 412 | { |
14e43bf4 EB |
413 | if (file->f_mode & FMODE_WRITER) { |
414 | /* | |
415 | * Superblock may have become readonly while there are still | |
416 | * writable fd's, e.g. due to a fs error with errors=remount-ro | |
417 | */ | |
418 | if (__mnt_is_readonly(file->f_path.mnt)) | |
419 | return -EROFS; | |
420 | return 0; | |
421 | } | |
422 | return __mnt_want_write(file->f_path.mnt); | |
96029c4e | 423 | } |
eb04c282 | 424 | |
7c6893e3 MS |
425 | /** |
426 | * mnt_want_write_file - get write access to a file's mount | |
427 | * @file: the file who's mount on which to take a write | |
428 | * | |
14e43bf4 EB |
429 | * This is like mnt_want_write, but if the file is already open for writing it |
430 | * skips incrementing mnt_writers (since the open file already has a reference) | |
431 | * and instead only does the freeze protection and the check for emergency r/o | |
432 | * remounts. This must be paired with mnt_drop_write_file. | |
7c6893e3 MS |
433 | */ |
434 | int mnt_want_write_file(struct file *file) | |
435 | { | |
436 | int ret; | |
437 | ||
a6795a58 | 438 | sb_start_write(file_inode(file)->i_sb); |
eb04c282 JK |
439 | ret = __mnt_want_write_file(file); |
440 | if (ret) | |
a6795a58 | 441 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
442 | return ret; |
443 | } | |
96029c4e NP |
444 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
445 | ||
8366025e | 446 | /** |
eb04c282 | 447 | * __mnt_drop_write - give up write access to a mount |
8366025e DH |
448 | * @mnt: the mount on which to give up write access |
449 | * | |
450 | * Tells the low-level filesystem that we are done | |
451 | * performing writes to it. Must be matched with | |
eb04c282 | 452 | * __mnt_want_write() call above. |
8366025e | 453 | */ |
eb04c282 | 454 | void __mnt_drop_write(struct vfsmount *mnt) |
8366025e | 455 | { |
d3ef3d73 | 456 | preempt_disable(); |
83adc753 | 457 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 458 | preempt_enable(); |
8366025e | 459 | } |
eb04c282 JK |
460 | |
461 | /** | |
462 | * mnt_drop_write - give up write access to a mount | |
463 | * @mnt: the mount on which to give up write access | |
464 | * | |
465 | * Tells the low-level filesystem that we are done performing writes to it and | |
466 | * also allows filesystem to be frozen again. Must be matched with | |
467 | * mnt_want_write() call above. | |
468 | */ | |
469 | void mnt_drop_write(struct vfsmount *mnt) | |
470 | { | |
471 | __mnt_drop_write(mnt); | |
472 | sb_end_write(mnt->mnt_sb); | |
473 | } | |
8366025e DH |
474 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
475 | ||
eb04c282 JK |
476 | void __mnt_drop_write_file(struct file *file) |
477 | { | |
14e43bf4 EB |
478 | if (!(file->f_mode & FMODE_WRITER)) |
479 | __mnt_drop_write(file->f_path.mnt); | |
eb04c282 JK |
480 | } |
481 | ||
7c6893e3 MS |
482 | void mnt_drop_write_file(struct file *file) |
483 | { | |
a6795a58 | 484 | __mnt_drop_write_file(file); |
7c6893e3 MS |
485 | sb_end_write(file_inode(file)->i_sb); |
486 | } | |
2a79f17e AV |
487 | EXPORT_SYMBOL(mnt_drop_write_file); |
488 | ||
538f4f02 CB |
489 | /** |
490 | * mnt_hold_writers - prevent write access to the given mount | |
491 | * @mnt: mnt to prevent write access to | |
492 | * | |
493 | * Prevents write access to @mnt if there are no active writers for @mnt. | |
494 | * This function needs to be called and return successfully before changing | |
495 | * properties of @mnt that need to remain stable for callers with write access | |
496 | * to @mnt. | |
497 | * | |
498 | * After this functions has been called successfully callers must pair it with | |
499 | * a call to mnt_unhold_writers() in order to stop preventing write access to | |
500 | * @mnt. | |
501 | * | |
502 | * Context: This function expects lock_mount_hash() to be held serializing | |
503 | * setting MNT_WRITE_HOLD. | |
504 | * Return: On success 0 is returned. | |
505 | * On error, -EBUSY is returned. | |
506 | */ | |
fbdc2f6c | 507 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 508 | { |
83adc753 | 509 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 510 | /* |
d3ef3d73 NP |
511 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
512 | * should be visible before we do. | |
3d733633 | 513 | */ |
d3ef3d73 NP |
514 | smp_mb(); |
515 | ||
3d733633 | 516 | /* |
d3ef3d73 NP |
517 | * With writers on hold, if this value is zero, then there are |
518 | * definitely no active writers (although held writers may subsequently | |
519 | * increment the count, they'll have to wait, and decrement it after | |
520 | * seeing MNT_READONLY). | |
521 | * | |
522 | * It is OK to have counter incremented on one CPU and decremented on | |
523 | * another: the sum will add up correctly. The danger would be when we | |
524 | * sum up each counter, if we read a counter before it is incremented, | |
525 | * but then read another CPU's count which it has been subsequently | |
526 | * decremented from -- we would see more decrements than we should. | |
527 | * MNT_WRITE_HOLD protects against this scenario, because | |
528 | * mnt_want_write first increments count, then smp_mb, then spins on | |
529 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
530 | * we're counting up here. | |
3d733633 | 531 | */ |
c6653a83 | 532 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
533 | return -EBUSY; |
534 | ||
535 | return 0; | |
536 | } | |
537 | ||
538f4f02 CB |
538 | /** |
539 | * mnt_unhold_writers - stop preventing write access to the given mount | |
540 | * @mnt: mnt to stop preventing write access to | |
541 | * | |
542 | * Stop preventing write access to @mnt allowing callers to gain write access | |
543 | * to @mnt again. | |
544 | * | |
545 | * This function can only be called after a successful call to | |
546 | * mnt_hold_writers(). | |
547 | * | |
548 | * Context: This function expects lock_mount_hash() to be held. | |
549 | */ | |
fbdc2f6c CB |
550 | static inline void mnt_unhold_writers(struct mount *mnt) |
551 | { | |
d3ef3d73 NP |
552 | /* |
553 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
554 | * that become unheld will see MNT_READONLY. | |
555 | */ | |
556 | smp_wmb(); | |
83adc753 | 557 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
558 | } |
559 | ||
560 | static int mnt_make_readonly(struct mount *mnt) | |
561 | { | |
562 | int ret; | |
563 | ||
564 | ret = mnt_hold_writers(mnt); | |
565 | if (!ret) | |
566 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
567 | mnt_unhold_writers(mnt); | |
3d733633 | 568 | return ret; |
8366025e | 569 | } |
8366025e | 570 | |
4ed5e82f MS |
571 | int sb_prepare_remount_readonly(struct super_block *sb) |
572 | { | |
573 | struct mount *mnt; | |
574 | int err = 0; | |
575 | ||
8e8b8796 MS |
576 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
577 | if (atomic_long_read(&sb->s_remove_count)) | |
578 | return -EBUSY; | |
579 | ||
719ea2fb | 580 | lock_mount_hash(); |
4ed5e82f MS |
581 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
582 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
ad1844a0 CB |
583 | err = mnt_hold_writers(mnt); |
584 | if (err) | |
4ed5e82f | 585 | break; |
4ed5e82f MS |
586 | } |
587 | } | |
8e8b8796 MS |
588 | if (!err && atomic_long_read(&sb->s_remove_count)) |
589 | err = -EBUSY; | |
590 | ||
4ed5e82f MS |
591 | if (!err) { |
592 | sb->s_readonly_remount = 1; | |
593 | smp_wmb(); | |
594 | } | |
595 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { | |
596 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
597 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
598 | } | |
719ea2fb | 599 | unlock_mount_hash(); |
4ed5e82f MS |
600 | |
601 | return err; | |
602 | } | |
603 | ||
b105e270 | 604 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 605 | { |
256c8aed | 606 | mnt_idmap_put(mnt_idmap(&mnt->mnt)); |
fcc139ae | 607 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 608 | #ifdef CONFIG_SMP |
68e8a9fe | 609 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 610 | #endif |
b105e270 | 611 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
612 | } |
613 | ||
8ffcb32e DH |
614 | static void delayed_free_vfsmnt(struct rcu_head *head) |
615 | { | |
616 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
617 | } | |
618 | ||
48a066e7 | 619 | /* call under rcu_read_lock */ |
294d71ff | 620 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
621 | { |
622 | struct mount *mnt; | |
623 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 624 | return 1; |
48a066e7 | 625 | if (bastard == NULL) |
294d71ff | 626 | return 0; |
48a066e7 AV |
627 | mnt = real_mount(bastard); |
628 | mnt_add_count(mnt, 1); | |
119e1ef8 | 629 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 630 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 631 | return 0; |
48a066e7 AV |
632 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
633 | mnt_add_count(mnt, -1); | |
294d71ff AV |
634 | return 1; |
635 | } | |
119e1ef8 AV |
636 | lock_mount_hash(); |
637 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
638 | mnt_add_count(mnt, -1); | |
639 | unlock_mount_hash(); | |
640 | return 1; | |
641 | } | |
642 | unlock_mount_hash(); | |
643 | /* caller will mntput() */ | |
294d71ff AV |
644 | return -1; |
645 | } | |
646 | ||
647 | /* call under rcu_read_lock */ | |
7e4745a0 | 648 | static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
294d71ff AV |
649 | { |
650 | int res = __legitimize_mnt(bastard, seq); | |
651 | if (likely(!res)) | |
652 | return true; | |
653 | if (unlikely(res < 0)) { | |
654 | rcu_read_unlock(); | |
655 | mntput(bastard); | |
656 | rcu_read_lock(); | |
48a066e7 | 657 | } |
48a066e7 AV |
658 | return false; |
659 | } | |
660 | ||
1da177e4 | 661 | /* |
474279dc | 662 | * find the first mount at @dentry on vfsmount @mnt. |
48a066e7 | 663 | * call under rcu_read_lock() |
1da177e4 | 664 | */ |
474279dc | 665 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 666 | { |
38129a13 | 667 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
668 | struct mount *p; |
669 | ||
38129a13 | 670 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
671 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
672 | return p; | |
673 | return NULL; | |
674 | } | |
675 | ||
a05964f3 | 676 | /* |
f015f126 DH |
677 | * lookup_mnt - Return the first child mount mounted at path |
678 | * | |
679 | * "First" means first mounted chronologically. If you create the | |
680 | * following mounts: | |
681 | * | |
682 | * mount /dev/sda1 /mnt | |
683 | * mount /dev/sda2 /mnt | |
684 | * mount /dev/sda3 /mnt | |
685 | * | |
686 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
687 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
688 | * /dev/sda2, then /dev/sda3, then NULL. | |
689 | * | |
690 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 691 | */ |
ca71cf71 | 692 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 693 | { |
c7105365 | 694 | struct mount *child_mnt; |
48a066e7 AV |
695 | struct vfsmount *m; |
696 | unsigned seq; | |
99b7db7b | 697 | |
48a066e7 AV |
698 | rcu_read_lock(); |
699 | do { | |
700 | seq = read_seqbegin(&mount_lock); | |
701 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
702 | m = child_mnt ? &child_mnt->mnt : NULL; | |
703 | } while (!legitimize_mnt(m, seq)); | |
704 | rcu_read_unlock(); | |
705 | return m; | |
a05964f3 RP |
706 | } |
707 | ||
9f6c61f9 MS |
708 | static inline void lock_ns_list(struct mnt_namespace *ns) |
709 | { | |
710 | spin_lock(&ns->ns_lock); | |
711 | } | |
712 | ||
713 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
714 | { | |
715 | spin_unlock(&ns->ns_lock); | |
716 | } | |
717 | ||
718 | static inline bool mnt_is_cursor(struct mount *mnt) | |
719 | { | |
720 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
721 | } | |
722 | ||
7af1364f EB |
723 | /* |
724 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
725 | * current mount namespace. | |
726 | * | |
727 | * The common case is dentries are not mountpoints at all and that | |
728 | * test is handled inline. For the slow case when we are actually | |
729 | * dealing with a mountpoint of some kind, walk through all of the | |
730 | * mounts in the current mount namespace and test to see if the dentry | |
731 | * is a mountpoint. | |
732 | * | |
733 | * The mount_hashtable is not usable in the context because we | |
734 | * need to identify all mounts that may be in the current mount | |
735 | * namespace not just a mount that happens to have some specified | |
736 | * parent mount. | |
737 | */ | |
738 | bool __is_local_mountpoint(struct dentry *dentry) | |
739 | { | |
740 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
741 | struct mount *mnt; | |
742 | bool is_covered = false; | |
743 | ||
7af1364f | 744 | down_read(&namespace_sem); |
9f6c61f9 | 745 | lock_ns_list(ns); |
7af1364f | 746 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
747 | if (mnt_is_cursor(mnt)) |
748 | continue; | |
7af1364f EB |
749 | is_covered = (mnt->mnt_mountpoint == dentry); |
750 | if (is_covered) | |
751 | break; | |
752 | } | |
9f6c61f9 | 753 | unlock_ns_list(ns); |
7af1364f | 754 | up_read(&namespace_sem); |
5ad05cc8 | 755 | |
7af1364f EB |
756 | return is_covered; |
757 | } | |
758 | ||
e2dfa935 | 759 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 760 | { |
0818bf27 | 761 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
762 | struct mountpoint *mp; |
763 | ||
0818bf27 | 764 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 765 | if (mp->m_dentry == dentry) { |
84d17192 AV |
766 | mp->m_count++; |
767 | return mp; | |
768 | } | |
769 | } | |
e2dfa935 EB |
770 | return NULL; |
771 | } | |
772 | ||
3895dbf8 | 773 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 774 | { |
3895dbf8 | 775 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 776 | int ret; |
84d17192 | 777 | |
3895dbf8 | 778 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
779 | /* might be worth a WARN_ON() */ |
780 | if (d_unlinked(dentry)) | |
781 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
782 | mountpoint: |
783 | read_seqlock_excl(&mount_lock); | |
784 | mp = lookup_mountpoint(dentry); | |
785 | read_sequnlock_excl(&mount_lock); | |
786 | if (mp) | |
787 | goto done; | |
788 | } | |
789 | ||
790 | if (!new) | |
791 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
792 | if (!new) | |
84d17192 AV |
793 | return ERR_PTR(-ENOMEM); |
794 | ||
3895dbf8 EB |
795 | |
796 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 797 | ret = d_set_mounted(dentry); |
eed81007 | 798 | |
3895dbf8 EB |
799 | /* Someone else set d_mounted? */ |
800 | if (ret == -EBUSY) | |
801 | goto mountpoint; | |
802 | ||
803 | /* The dentry is not available as a mountpoint? */ | |
804 | mp = ERR_PTR(ret); | |
805 | if (ret) | |
806 | goto done; | |
807 | ||
808 | /* Add the new mountpoint to the hash table */ | |
809 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 810 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
811 | new->m_count = 1; |
812 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
813 | INIT_HLIST_HEAD(&new->m_list); | |
814 | read_sequnlock_excl(&mount_lock); | |
815 | ||
816 | mp = new; | |
817 | new = NULL; | |
818 | done: | |
819 | kfree(new); | |
84d17192 AV |
820 | return mp; |
821 | } | |
822 | ||
4edbe133 AV |
823 | /* |
824 | * vfsmount lock must be held. Additionally, the caller is responsible | |
825 | * for serializing calls for given disposal list. | |
826 | */ | |
827 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
828 | { |
829 | if (!--mp->m_count) { | |
830 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 831 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
832 | spin_lock(&dentry->d_lock); |
833 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
834 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 835 | dput_to_list(dentry, list); |
0818bf27 | 836 | hlist_del(&mp->m_hash); |
84d17192 AV |
837 | kfree(mp); |
838 | } | |
839 | } | |
840 | ||
4edbe133 AV |
841 | /* called with namespace_lock and vfsmount lock */ |
842 | static void put_mountpoint(struct mountpoint *mp) | |
843 | { | |
844 | __put_mountpoint(mp, &ex_mountpoints); | |
845 | } | |
846 | ||
143c8c91 | 847 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 848 | { |
6b3286ed | 849 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
850 | } |
851 | ||
99b7db7b NP |
852 | /* |
853 | * vfsmount lock must be held for write | |
854 | */ | |
6b3286ed | 855 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
856 | { |
857 | if (ns) { | |
858 | ns->event = ++event; | |
859 | wake_up_interruptible(&ns->poll); | |
860 | } | |
861 | } | |
862 | ||
99b7db7b NP |
863 | /* |
864 | * vfsmount lock must be held for write | |
865 | */ | |
6b3286ed | 866 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
867 | { |
868 | if (ns && ns->event != event) { | |
869 | ns->event = event; | |
870 | wake_up_interruptible(&ns->poll); | |
871 | } | |
872 | } | |
873 | ||
99b7db7b NP |
874 | /* |
875 | * vfsmount lock must be held for write | |
876 | */ | |
e4e59906 | 877 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 878 | { |
e4e59906 | 879 | struct mountpoint *mp; |
0714a533 | 880 | mnt->mnt_parent = mnt; |
a73324da | 881 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 882 | list_del_init(&mnt->mnt_child); |
38129a13 | 883 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 884 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 885 | mp = mnt->mnt_mp; |
84d17192 | 886 | mnt->mnt_mp = NULL; |
e4e59906 | 887 | return mp; |
7bdb11de EB |
888 | } |
889 | ||
6a46c573 EB |
890 | /* |
891 | * vfsmount lock must be held for write | |
892 | */ | |
893 | static void umount_mnt(struct mount *mnt) | |
894 | { | |
e4e59906 | 895 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
896 | } |
897 | ||
99b7db7b NP |
898 | /* |
899 | * vfsmount lock must be held for write | |
900 | */ | |
84d17192 AV |
901 | void mnt_set_mountpoint(struct mount *mnt, |
902 | struct mountpoint *mp, | |
44d964d6 | 903 | struct mount *child_mnt) |
b90fa9ae | 904 | { |
84d17192 | 905 | mp->m_count++; |
3a2393d7 | 906 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 907 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 908 | child_mnt->mnt_parent = mnt; |
84d17192 | 909 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 910 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
911 | } |
912 | ||
1064f874 EB |
913 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
914 | { | |
915 | hlist_add_head_rcu(&mnt->mnt_hash, | |
916 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
917 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
918 | } | |
919 | ||
99b7db7b NP |
920 | /* |
921 | * vfsmount lock must be held for write | |
922 | */ | |
84d17192 AV |
923 | static void attach_mnt(struct mount *mnt, |
924 | struct mount *parent, | |
925 | struct mountpoint *mp) | |
1da177e4 | 926 | { |
84d17192 | 927 | mnt_set_mountpoint(parent, mp, mnt); |
1064f874 | 928 | __attach_mnt(mnt, parent); |
b90fa9ae RP |
929 | } |
930 | ||
1064f874 | 931 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 932 | { |
1064f874 | 933 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
934 | struct mount *old_parent = mnt->mnt_parent; |
935 | ||
936 | list_del_init(&mnt->mnt_child); | |
937 | hlist_del_init(&mnt->mnt_mp_list); | |
938 | hlist_del_init_rcu(&mnt->mnt_hash); | |
939 | ||
940 | attach_mnt(mnt, parent, mp); | |
941 | ||
942 | put_mountpoint(old_mp); | |
1064f874 | 943 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
944 | } |
945 | ||
b90fa9ae | 946 | /* |
99b7db7b | 947 | * vfsmount lock must be held for write |
b90fa9ae | 948 | */ |
1064f874 | 949 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 950 | { |
0714a533 | 951 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 952 | struct mount *m; |
b90fa9ae | 953 | LIST_HEAD(head); |
143c8c91 | 954 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 955 | |
0714a533 | 956 | BUG_ON(parent == mnt); |
b90fa9ae | 957 | |
1a4eeaf2 | 958 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 959 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 960 | m->mnt_ns = n; |
f03c6599 | 961 | |
b90fa9ae RP |
962 | list_splice(&head, n->list.prev); |
963 | ||
d2921684 EB |
964 | n->mounts += n->pending_mounts; |
965 | n->pending_mounts = 0; | |
966 | ||
1064f874 | 967 | __attach_mnt(mnt, parent); |
6b3286ed | 968 | touch_mnt_namespace(n); |
1da177e4 LT |
969 | } |
970 | ||
909b0a88 | 971 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 972 | { |
6b41d536 AV |
973 | struct list_head *next = p->mnt_mounts.next; |
974 | if (next == &p->mnt_mounts) { | |
1da177e4 | 975 | while (1) { |
909b0a88 | 976 | if (p == root) |
1da177e4 | 977 | return NULL; |
6b41d536 AV |
978 | next = p->mnt_child.next; |
979 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 980 | break; |
0714a533 | 981 | p = p->mnt_parent; |
1da177e4 LT |
982 | } |
983 | } | |
6b41d536 | 984 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
985 | } |
986 | ||
315fc83e | 987 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 988 | { |
6b41d536 AV |
989 | struct list_head *prev = p->mnt_mounts.prev; |
990 | while (prev != &p->mnt_mounts) { | |
991 | p = list_entry(prev, struct mount, mnt_child); | |
992 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
993 | } |
994 | return p; | |
995 | } | |
996 | ||
8f291889 AV |
997 | /** |
998 | * vfs_create_mount - Create a mount for a configured superblock | |
999 | * @fc: The configuration context with the superblock attached | |
1000 | * | |
1001 | * Create a mount to an already configured superblock. If necessary, the | |
1002 | * caller should invoke vfs_get_tree() before calling this. | |
1003 | * | |
1004 | * Note that this does not attach the mount to anything. | |
1005 | */ | |
1006 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 1007 | { |
b105e270 | 1008 | struct mount *mnt; |
9d412a43 | 1009 | |
8f291889 AV |
1010 | if (!fc->root) |
1011 | return ERR_PTR(-EINVAL); | |
9d412a43 | 1012 | |
8f291889 | 1013 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
1014 | if (!mnt) |
1015 | return ERR_PTR(-ENOMEM); | |
1016 | ||
8f291889 | 1017 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 1018 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 1019 | |
8f291889 AV |
1020 | atomic_inc(&fc->root->d_sb->s_active); |
1021 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
1022 | mnt->mnt.mnt_root = dget(fc->root); | |
1023 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1024 | mnt->mnt_parent = mnt; | |
9d412a43 | 1025 | |
719ea2fb | 1026 | lock_mount_hash(); |
8f291889 | 1027 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 1028 | unlock_mount_hash(); |
b105e270 | 1029 | return &mnt->mnt; |
9d412a43 | 1030 | } |
8f291889 AV |
1031 | EXPORT_SYMBOL(vfs_create_mount); |
1032 | ||
1033 | struct vfsmount *fc_mount(struct fs_context *fc) | |
1034 | { | |
1035 | int err = vfs_get_tree(fc); | |
1036 | if (!err) { | |
1037 | up_write(&fc->root->d_sb->s_umount); | |
1038 | return vfs_create_mount(fc); | |
1039 | } | |
1040 | return ERR_PTR(err); | |
1041 | } | |
1042 | EXPORT_SYMBOL(fc_mount); | |
1043 | ||
9bc61ab1 DH |
1044 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1045 | int flags, const char *name, | |
1046 | void *data) | |
9d412a43 | 1047 | { |
9bc61ab1 | 1048 | struct fs_context *fc; |
8f291889 | 1049 | struct vfsmount *mnt; |
9bc61ab1 | 1050 | int ret = 0; |
9d412a43 AV |
1051 | |
1052 | if (!type) | |
3e1aeb00 | 1053 | return ERR_PTR(-EINVAL); |
9d412a43 | 1054 | |
9bc61ab1 DH |
1055 | fc = fs_context_for_mount(type, flags); |
1056 | if (IS_ERR(fc)) | |
1057 | return ERR_CAST(fc); | |
1058 | ||
3e1aeb00 DH |
1059 | if (name) |
1060 | ret = vfs_parse_fs_string(fc, "source", | |
1061 | name, strlen(name)); | |
9bc61ab1 DH |
1062 | if (!ret) |
1063 | ret = parse_monolithic_mount_data(fc, data); | |
1064 | if (!ret) | |
8f291889 AV |
1065 | mnt = fc_mount(fc); |
1066 | else | |
1067 | mnt = ERR_PTR(ret); | |
9d412a43 | 1068 | |
9bc61ab1 | 1069 | put_fs_context(fc); |
8f291889 | 1070 | return mnt; |
9d412a43 AV |
1071 | } |
1072 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1073 | ||
93faccbb EB |
1074 | struct vfsmount * |
1075 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1076 | const char *name, void *data) | |
1077 | { | |
1078 | /* Until it is worked out how to pass the user namespace | |
1079 | * through from the parent mount to the submount don't support | |
1080 | * unprivileged mounts with submounts. | |
1081 | */ | |
1082 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1083 | return ERR_PTR(-EPERM); | |
1084 | ||
e462ec50 | 1085 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1086 | } |
1087 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1088 | ||
87129cc0 | 1089 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1090 | int flag) |
1da177e4 | 1091 | { |
87129cc0 | 1092 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1093 | struct mount *mnt; |
1094 | int err; | |
1da177e4 | 1095 | |
be34d1a3 DH |
1096 | mnt = alloc_vfsmnt(old->mnt_devname); |
1097 | if (!mnt) | |
1098 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1099 | |
7a472ef4 | 1100 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1101 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1102 | else | |
1103 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1104 | |
be34d1a3 DH |
1105 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1106 | err = mnt_alloc_group_id(mnt); | |
1107 | if (err) | |
1108 | goto out_free; | |
1da177e4 | 1109 | } |
be34d1a3 | 1110 | |
16a34adb AV |
1111 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1112 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1113 | |
be34d1a3 | 1114 | atomic_inc(&sb->s_active); |
256c8aed CB |
1115 | mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt)); |
1116 | ||
be34d1a3 DH |
1117 | mnt->mnt.mnt_sb = sb; |
1118 | mnt->mnt.mnt_root = dget(root); | |
1119 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1120 | mnt->mnt_parent = mnt; | |
719ea2fb | 1121 | lock_mount_hash(); |
be34d1a3 | 1122 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1123 | unlock_mount_hash(); |
be34d1a3 | 1124 | |
7a472ef4 EB |
1125 | if ((flag & CL_SLAVE) || |
1126 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1127 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1128 | mnt->mnt_master = old; | |
1129 | CLEAR_MNT_SHARED(mnt); | |
1130 | } else if (!(flag & CL_PRIVATE)) { | |
1131 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1132 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1133 | if (IS_MNT_SLAVE(old)) | |
1134 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1135 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1136 | } else { |
1137 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1138 | } |
1139 | if (flag & CL_MAKE_SHARED) | |
1140 | set_mnt_shared(mnt); | |
1141 | ||
1142 | /* stick the duplicate mount on the same expiry list | |
1143 | * as the original if that was on one */ | |
1144 | if (flag & CL_EXPIRE) { | |
1145 | if (!list_empty(&old->mnt_expire)) | |
1146 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1147 | } | |
1148 | ||
cb338d06 | 1149 | return mnt; |
719f5d7f MS |
1150 | |
1151 | out_free: | |
8ffcb32e | 1152 | mnt_free_id(mnt); |
719f5d7f | 1153 | free_vfsmnt(mnt); |
be34d1a3 | 1154 | return ERR_PTR(err); |
1da177e4 LT |
1155 | } |
1156 | ||
9ea459e1 AV |
1157 | static void cleanup_mnt(struct mount *mnt) |
1158 | { | |
56cbb429 AV |
1159 | struct hlist_node *p; |
1160 | struct mount *m; | |
9ea459e1 | 1161 | /* |
56cbb429 AV |
1162 | * The warning here probably indicates that somebody messed |
1163 | * up a mnt_want/drop_write() pair. If this happens, the | |
1164 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1165 | * The locking used to deal with mnt_count decrement provides barriers, |
1166 | * so mnt_get_writers() below is safe. | |
1167 | */ | |
1168 | WARN_ON(mnt_get_writers(mnt)); | |
1169 | if (unlikely(mnt->mnt_pins.first)) | |
1170 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1171 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1172 | hlist_del(&m->mnt_umount); | |
1173 | mntput(&m->mnt); | |
1174 | } | |
9ea459e1 AV |
1175 | fsnotify_vfsmount_delete(&mnt->mnt); |
1176 | dput(mnt->mnt.mnt_root); | |
1177 | deactivate_super(mnt->mnt.mnt_sb); | |
1178 | mnt_free_id(mnt); | |
1179 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1180 | } | |
1181 | ||
1182 | static void __cleanup_mnt(struct rcu_head *head) | |
1183 | { | |
1184 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1185 | } | |
1186 | ||
1187 | static LLIST_HEAD(delayed_mntput_list); | |
1188 | static void delayed_mntput(struct work_struct *unused) | |
1189 | { | |
1190 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1191 | struct mount *m, *t; |
9ea459e1 | 1192 | |
29785735 BP |
1193 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1194 | cleanup_mnt(m); | |
9ea459e1 AV |
1195 | } |
1196 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1197 | ||
900148dc | 1198 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1199 | { |
4edbe133 | 1200 | LIST_HEAD(list); |
edf7ddbf | 1201 | int count; |
4edbe133 | 1202 | |
48a066e7 | 1203 | rcu_read_lock(); |
9ea0a46c AV |
1204 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1205 | /* | |
1206 | * Since we don't do lock_mount_hash() here, | |
1207 | * ->mnt_ns can change under us. However, if it's | |
1208 | * non-NULL, then there's a reference that won't | |
1209 | * be dropped until after an RCU delay done after | |
1210 | * turning ->mnt_ns NULL. So if we observe it | |
1211 | * non-NULL under rcu_read_lock(), the reference | |
1212 | * we are dropping is not the final one. | |
1213 | */ | |
1214 | mnt_add_count(mnt, -1); | |
48a066e7 | 1215 | rcu_read_unlock(); |
f03c6599 | 1216 | return; |
b3e19d92 | 1217 | } |
719ea2fb | 1218 | lock_mount_hash(); |
119e1ef8 AV |
1219 | /* |
1220 | * make sure that if __legitimize_mnt() has not seen us grab | |
1221 | * mount_lock, we'll see their refcount increment here. | |
1222 | */ | |
1223 | smp_mb(); | |
9ea0a46c | 1224 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1225 | count = mnt_get_count(mnt); |
1226 | if (count != 0) { | |
1227 | WARN_ON(count < 0); | |
48a066e7 | 1228 | rcu_read_unlock(); |
719ea2fb | 1229 | unlock_mount_hash(); |
99b7db7b NP |
1230 | return; |
1231 | } | |
48a066e7 AV |
1232 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1233 | rcu_read_unlock(); | |
1234 | unlock_mount_hash(); | |
1235 | return; | |
1236 | } | |
1237 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1238 | rcu_read_unlock(); | |
962830df | 1239 | |
39f7c4db | 1240 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1241 | |
1242 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1243 | struct mount *p, *tmp; | |
1244 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1245 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1246 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1247 | } |
1248 | } | |
719ea2fb | 1249 | unlock_mount_hash(); |
4edbe133 | 1250 | shrink_dentry_list(&list); |
649a795a | 1251 | |
9ea459e1 AV |
1252 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1253 | struct task_struct *task = current; | |
1254 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1255 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1256 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1257 | return; |
1258 | } | |
1259 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1260 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1261 | return; | |
1262 | } | |
1263 | cleanup_mnt(mnt); | |
b3e19d92 | 1264 | } |
b3e19d92 NP |
1265 | |
1266 | void mntput(struct vfsmount *mnt) | |
1267 | { | |
1268 | if (mnt) { | |
863d684f | 1269 | struct mount *m = real_mount(mnt); |
b3e19d92 | 1270 | /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ |
863d684f AV |
1271 | if (unlikely(m->mnt_expiry_mark)) |
1272 | m->mnt_expiry_mark = 0; | |
1273 | mntput_no_expire(m); | |
b3e19d92 NP |
1274 | } |
1275 | } | |
1276 | EXPORT_SYMBOL(mntput); | |
1277 | ||
1278 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1279 | { | |
1280 | if (mnt) | |
83adc753 | 1281 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1282 | return mnt; |
1283 | } | |
1284 | EXPORT_SYMBOL(mntget); | |
1285 | ||
da27f796 RR |
1286 | /* |
1287 | * Make a mount point inaccessible to new lookups. | |
1288 | * Because there may still be current users, the caller MUST WAIT | |
1289 | * for an RCU grace period before destroying the mount point. | |
1290 | */ | |
1291 | void mnt_make_shortterm(struct vfsmount *mnt) | |
1292 | { | |
1293 | if (mnt) | |
1294 | real_mount(mnt)->mnt_ns = NULL; | |
1295 | } | |
1296 | ||
1f287bc4 RD |
1297 | /** |
1298 | * path_is_mountpoint() - Check if path is a mount in the current namespace. | |
1299 | * @path: path to check | |
c6609c0a IK |
1300 | * |
1301 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1302 | * not mounted in any namespace and that common case is handled inline. | |
1303 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1304 | * mounts using a given dentry in a different namespace. This function | |
1305 | * checks if the passed in path is a mountpoint rather than the dentry | |
1306 | * alone. | |
1307 | */ | |
1308 | bool path_is_mountpoint(const struct path *path) | |
1309 | { | |
1310 | unsigned seq; | |
1311 | bool res; | |
1312 | ||
1313 | if (!d_mountpoint(path->dentry)) | |
1314 | return false; | |
1315 | ||
1316 | rcu_read_lock(); | |
1317 | do { | |
1318 | seq = read_seqbegin(&mount_lock); | |
1319 | res = __path_is_mountpoint(path); | |
1320 | } while (read_seqretry(&mount_lock, seq)); | |
1321 | rcu_read_unlock(); | |
1322 | ||
1323 | return res; | |
1324 | } | |
1325 | EXPORT_SYMBOL(path_is_mountpoint); | |
1326 | ||
ca71cf71 | 1327 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1328 | { |
3064c356 AV |
1329 | struct mount *p; |
1330 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1331 | if (IS_ERR(p)) | |
1332 | return ERR_CAST(p); | |
1333 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1334 | return &p->mnt; | |
7b7b1ace | 1335 | } |
1da177e4 | 1336 | |
a1a2c409 | 1337 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1338 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1339 | struct list_head *p) | |
1340 | { | |
1341 | struct mount *mnt, *ret = NULL; | |
1342 | ||
1343 | lock_ns_list(ns); | |
1344 | list_for_each_continue(p, &ns->list) { | |
1345 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1346 | if (!mnt_is_cursor(mnt)) { | |
1347 | ret = mnt; | |
1348 | break; | |
1349 | } | |
1350 | } | |
1351 | unlock_ns_list(ns); | |
1352 | ||
1353 | return ret; | |
1354 | } | |
1355 | ||
0226f492 | 1356 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1357 | static void *m_start(struct seq_file *m, loff_t *pos) |
1358 | { | |
ede1bf0d | 1359 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1360 | struct list_head *prev; |
1da177e4 | 1361 | |
390c6843 | 1362 | down_read(&namespace_sem); |
9f6c61f9 MS |
1363 | if (!*pos) { |
1364 | prev = &p->ns->list; | |
1365 | } else { | |
1366 | prev = &p->cursor.mnt_list; | |
1367 | ||
1368 | /* Read after we'd reached the end? */ | |
1369 | if (list_empty(prev)) | |
1370 | return NULL; | |
c7999c36 AV |
1371 | } |
1372 | ||
9f6c61f9 | 1373 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1374 | } |
1375 | ||
1376 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1377 | { | |
ede1bf0d | 1378 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1379 | struct mount *mnt = v; |
b0765fb8 | 1380 | |
9f6c61f9 MS |
1381 | ++*pos; |
1382 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1383 | } |
1384 | ||
1385 | static void m_stop(struct seq_file *m, void *v) | |
1386 | { | |
9f6c61f9 MS |
1387 | struct proc_mounts *p = m->private; |
1388 | struct mount *mnt = v; | |
1389 | ||
1390 | lock_ns_list(p->ns); | |
1391 | if (mnt) | |
1392 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1393 | else | |
1394 | list_del_init(&p->cursor.mnt_list); | |
1395 | unlock_ns_list(p->ns); | |
390c6843 | 1396 | up_read(&namespace_sem); |
1da177e4 LT |
1397 | } |
1398 | ||
0226f492 | 1399 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1400 | { |
ede1bf0d | 1401 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1402 | struct mount *r = v; |
0226f492 | 1403 | return p->show(m, &r->mnt); |
1da177e4 LT |
1404 | } |
1405 | ||
a1a2c409 | 1406 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1407 | .start = m_start, |
1408 | .next = m_next, | |
1409 | .stop = m_stop, | |
0226f492 | 1410 | .show = m_show, |
b4629fe2 | 1411 | }; |
9f6c61f9 MS |
1412 | |
1413 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1414 | { | |
1415 | down_read(&namespace_sem); | |
1416 | lock_ns_list(ns); | |
1417 | list_del(&cursor->mnt_list); | |
1418 | unlock_ns_list(ns); | |
1419 | up_read(&namespace_sem); | |
1420 | } | |
a1a2c409 | 1421 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1422 | |
1da177e4 LT |
1423 | /** |
1424 | * may_umount_tree - check if a mount tree is busy | |
1f287bc4 | 1425 | * @m: root of mount tree |
1da177e4 LT |
1426 | * |
1427 | * This is called to check if a tree of mounts has any | |
1428 | * open files, pwds, chroots or sub mounts that are | |
1429 | * busy. | |
1430 | */ | |
909b0a88 | 1431 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1432 | { |
909b0a88 | 1433 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1434 | int actual_refs = 0; |
1435 | int minimum_refs = 0; | |
315fc83e | 1436 | struct mount *p; |
909b0a88 | 1437 | BUG_ON(!m); |
1da177e4 | 1438 | |
b3e19d92 | 1439 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1440 | lock_mount_hash(); |
909b0a88 | 1441 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1442 | actual_refs += mnt_get_count(p); |
1da177e4 | 1443 | minimum_refs += 2; |
1da177e4 | 1444 | } |
719ea2fb | 1445 | unlock_mount_hash(); |
1da177e4 LT |
1446 | |
1447 | if (actual_refs > minimum_refs) | |
e3474a8e | 1448 | return 0; |
1da177e4 | 1449 | |
e3474a8e | 1450 | return 1; |
1da177e4 LT |
1451 | } |
1452 | ||
1453 | EXPORT_SYMBOL(may_umount_tree); | |
1454 | ||
1455 | /** | |
1456 | * may_umount - check if a mount point is busy | |
1457 | * @mnt: root of mount | |
1458 | * | |
1459 | * This is called to check if a mount point has any | |
1460 | * open files, pwds, chroots or sub mounts. If the | |
1461 | * mount has sub mounts this will return busy | |
1462 | * regardless of whether the sub mounts are busy. | |
1463 | * | |
1464 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1465 | * give false negatives. The main reason why it's here is that we need | |
1466 | * a non-destructive way to look for easily umountable filesystems. | |
1467 | */ | |
1468 | int may_umount(struct vfsmount *mnt) | |
1469 | { | |
e3474a8e | 1470 | int ret = 1; |
8ad08d8a | 1471 | down_read(&namespace_sem); |
719ea2fb | 1472 | lock_mount_hash(); |
1ab59738 | 1473 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1474 | ret = 0; |
719ea2fb | 1475 | unlock_mount_hash(); |
8ad08d8a | 1476 | up_read(&namespace_sem); |
a05964f3 | 1477 | return ret; |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | EXPORT_SYMBOL(may_umount); | |
1481 | ||
97216be0 | 1482 | static void namespace_unlock(void) |
70fbcdf4 | 1483 | { |
a3b3c562 | 1484 | struct hlist_head head; |
56cbb429 AV |
1485 | struct hlist_node *p; |
1486 | struct mount *m; | |
4edbe133 | 1487 | LIST_HEAD(list); |
97216be0 | 1488 | |
a3b3c562 | 1489 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1490 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1491 | |
97216be0 AV |
1492 | up_write(&namespace_sem); |
1493 | ||
4edbe133 AV |
1494 | shrink_dentry_list(&list); |
1495 | ||
a3b3c562 EB |
1496 | if (likely(hlist_empty(&head))) |
1497 | return; | |
1498 | ||
22cb7405 | 1499 | synchronize_rcu_expedited(); |
48a066e7 | 1500 | |
56cbb429 AV |
1501 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1502 | hlist_del(&m->mnt_umount); | |
1503 | mntput(&m->mnt); | |
1504 | } | |
70fbcdf4 RP |
1505 | } |
1506 | ||
97216be0 | 1507 | static inline void namespace_lock(void) |
e3197d83 | 1508 | { |
97216be0 | 1509 | down_write(&namespace_sem); |
e3197d83 AV |
1510 | } |
1511 | ||
e819f152 EB |
1512 | enum umount_tree_flags { |
1513 | UMOUNT_SYNC = 1, | |
1514 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1515 | UMOUNT_CONNECTED = 4, |
e819f152 | 1516 | }; |
f2d0a123 EB |
1517 | |
1518 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1519 | { | |
1520 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1521 | if (how & UMOUNT_SYNC) | |
1522 | return true; | |
1523 | ||
1524 | /* A mount without a parent has nothing to be connected to */ | |
1525 | if (!mnt_has_parent(mnt)) | |
1526 | return true; | |
1527 | ||
1528 | /* Because the reference counting rules change when mounts are | |
1529 | * unmounted and connected, umounted mounts may not be | |
1530 | * connected to mounted mounts. | |
1531 | */ | |
1532 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1533 | return true; | |
1534 | ||
1535 | /* Has it been requested that the mount remain connected? */ | |
1536 | if (how & UMOUNT_CONNECTED) | |
1537 | return false; | |
1538 | ||
1539 | /* Is the mount locked such that it needs to remain connected? */ | |
1540 | if (IS_MNT_LOCKED(mnt)) | |
1541 | return false; | |
1542 | ||
1543 | /* By default disconnect the mount */ | |
1544 | return true; | |
1545 | } | |
1546 | ||
99b7db7b | 1547 | /* |
48a066e7 | 1548 | * mount_lock must be held |
99b7db7b NP |
1549 | * namespace_sem must be held for write |
1550 | */ | |
e819f152 | 1551 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1552 | { |
c003b26f | 1553 | LIST_HEAD(tmp_list); |
315fc83e | 1554 | struct mount *p; |
1da177e4 | 1555 | |
5d88457e EB |
1556 | if (how & UMOUNT_PROPAGATE) |
1557 | propagate_mount_unlock(mnt); | |
1558 | ||
c003b26f | 1559 | /* Gather the mounts to umount */ |
590ce4bc EB |
1560 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1561 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1562 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1563 | } |
1da177e4 | 1564 | |
411a938b | 1565 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1566 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1567 | list_del_init(&p->mnt_child); |
c003b26f | 1568 | } |
88b368f2 | 1569 | |
c003b26f | 1570 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1571 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1572 | propagate_umount(&tmp_list); |
a05964f3 | 1573 | |
c003b26f | 1574 | while (!list_empty(&tmp_list)) { |
d2921684 | 1575 | struct mnt_namespace *ns; |
ce07d891 | 1576 | bool disconnect; |
c003b26f | 1577 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1578 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1579 | list_del_init(&p->mnt_list); |
d2921684 EB |
1580 | ns = p->mnt_ns; |
1581 | if (ns) { | |
1582 | ns->mounts--; | |
1583 | __touch_mnt_namespace(ns); | |
1584 | } | |
143c8c91 | 1585 | p->mnt_ns = NULL; |
e819f152 | 1586 | if (how & UMOUNT_SYNC) |
48a066e7 | 1587 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1588 | |
f2d0a123 | 1589 | disconnect = disconnect_mount(p, how); |
676da58d | 1590 | if (mnt_has_parent(p)) { |
81b6b061 | 1591 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1592 | if (!disconnect) { |
1593 | /* Don't forget about p */ | |
1594 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1595 | } else { | |
1596 | umount_mnt(p); | |
1597 | } | |
7c4b93d8 | 1598 | } |
0f0afb1d | 1599 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1600 | if (disconnect) |
1601 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1602 | } |
1603 | } | |
1604 | ||
b54b9be7 | 1605 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1606 | |
8d0347f6 DH |
1607 | static int do_umount_root(struct super_block *sb) |
1608 | { | |
1609 | int ret = 0; | |
1610 | ||
1611 | down_write(&sb->s_umount); | |
1612 | if (!sb_rdonly(sb)) { | |
1613 | struct fs_context *fc; | |
1614 | ||
1615 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1616 | SB_RDONLY); | |
1617 | if (IS_ERR(fc)) { | |
1618 | ret = PTR_ERR(fc); | |
1619 | } else { | |
1620 | ret = parse_monolithic_mount_data(fc, NULL); | |
1621 | if (!ret) | |
1622 | ret = reconfigure_super(fc); | |
1623 | put_fs_context(fc); | |
1624 | } | |
1625 | } | |
1626 | up_write(&sb->s_umount); | |
1627 | return ret; | |
1628 | } | |
1629 | ||
1ab59738 | 1630 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1631 | { |
1ab59738 | 1632 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1633 | int retval; |
1634 | ||
1ab59738 | 1635 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1636 | if (retval) |
1637 | return retval; | |
1638 | ||
1639 | /* | |
1640 | * Allow userspace to request a mountpoint be expired rather than | |
1641 | * unmounting unconditionally. Unmount only happens if: | |
1642 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1643 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1644 | */ | |
1645 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1646 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1647 | flags & (MNT_FORCE | MNT_DETACH)) |
1648 | return -EINVAL; | |
1649 | ||
b3e19d92 NP |
1650 | /* |
1651 | * probably don't strictly need the lock here if we examined | |
1652 | * all race cases, but it's a slowpath. | |
1653 | */ | |
719ea2fb | 1654 | lock_mount_hash(); |
83adc753 | 1655 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1656 | unlock_mount_hash(); |
1da177e4 | 1657 | return -EBUSY; |
b3e19d92 | 1658 | } |
719ea2fb | 1659 | unlock_mount_hash(); |
1da177e4 | 1660 | |
863d684f | 1661 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1662 | return -EAGAIN; |
1663 | } | |
1664 | ||
1665 | /* | |
1666 | * If we may have to abort operations to get out of this | |
1667 | * mount, and they will themselves hold resources we must | |
1668 | * allow the fs to do things. In the Unix tradition of | |
1669 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1670 | * might fail to complete on the first run through as other tasks | |
1671 | * must return, and the like. Thats for the mount program to worry | |
1672 | * about for the moment. | |
1673 | */ | |
1674 | ||
42faad99 | 1675 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1676 | sb->s_op->umount_begin(sb); |
42faad99 | 1677 | } |
1da177e4 LT |
1678 | |
1679 | /* | |
1680 | * No sense to grab the lock for this test, but test itself looks | |
1681 | * somewhat bogus. Suggestions for better replacement? | |
1682 | * Ho-hum... In principle, we might treat that as umount + switch | |
1683 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1684 | * Actually it makes sense, especially if rootfs would contain a | |
1685 | * /reboot - static binary that would close all descriptors and | |
1686 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1687 | */ | |
1ab59738 | 1688 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1689 | /* |
1690 | * Special case for "unmounting" root ... | |
1691 | * we just try to remount it readonly. | |
1692 | */ | |
bc6155d1 | 1693 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1694 | return -EPERM; |
8d0347f6 | 1695 | return do_umount_root(sb); |
1da177e4 LT |
1696 | } |
1697 | ||
97216be0 | 1698 | namespace_lock(); |
719ea2fb | 1699 | lock_mount_hash(); |
1da177e4 | 1700 | |
25d202ed EB |
1701 | /* Recheck MNT_LOCKED with the locks held */ |
1702 | retval = -EINVAL; | |
1703 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1704 | goto out; | |
1705 | ||
1706 | event++; | |
48a066e7 | 1707 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1708 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1709 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1710 | retval = 0; |
48a066e7 AV |
1711 | } else { |
1712 | shrink_submounts(mnt); | |
1713 | retval = -EBUSY; | |
1714 | if (!propagate_mount_busy(mnt, 2)) { | |
1715 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1716 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1717 | retval = 0; |
1718 | } | |
1da177e4 | 1719 | } |
25d202ed | 1720 | out: |
719ea2fb | 1721 | unlock_mount_hash(); |
e3197d83 | 1722 | namespace_unlock(); |
1da177e4 LT |
1723 | return retval; |
1724 | } | |
1725 | ||
80b5dce8 EB |
1726 | /* |
1727 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1728 | * | |
1729 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1730 | * to an existing mountpoint, and wind up leaking the mount. | |
1731 | * detach_mounts allows lazily unmounting those mounts instead of | |
1732 | * leaking them. | |
1733 | * | |
1734 | * The caller may hold dentry->d_inode->i_mutex. | |
1735 | */ | |
1736 | void __detach_mounts(struct dentry *dentry) | |
1737 | { | |
1738 | struct mountpoint *mp; | |
1739 | struct mount *mnt; | |
1740 | ||
1741 | namespace_lock(); | |
3895dbf8 | 1742 | lock_mount_hash(); |
80b5dce8 | 1743 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1744 | if (!mp) |
80b5dce8 EB |
1745 | goto out_unlock; |
1746 | ||
e06b933e | 1747 | event++; |
80b5dce8 EB |
1748 | while (!hlist_empty(&mp->m_list)) { |
1749 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1750 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1751 | umount_mnt(mnt); |
56cbb429 | 1752 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1753 | } |
e0c9c0af | 1754 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1755 | } |
80b5dce8 EB |
1756 | put_mountpoint(mp); |
1757 | out_unlock: | |
3895dbf8 | 1758 | unlock_mount_hash(); |
80b5dce8 EB |
1759 | namespace_unlock(); |
1760 | } | |
1761 | ||
dd111b31 | 1762 | /* |
9b40bc90 AV |
1763 | * Is the caller allowed to modify his namespace? |
1764 | */ | |
a5f85d78 | 1765 | bool may_mount(void) |
9b40bc90 AV |
1766 | { |
1767 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1768 | } | |
1769 | ||
f7e33bdb | 1770 | static void warn_mandlock(void) |
9e8925b6 | 1771 | { |
f7e33bdb JL |
1772 | pr_warn_once("=======================================================\n" |
1773 | "WARNING: The mand mount option has been deprecated and\n" | |
1774 | " and is ignored by this kernel. Remove the mand\n" | |
1775 | " option from the mount to silence this warning.\n" | |
1776 | "=======================================================\n"); | |
9e8925b6 JL |
1777 | } |
1778 | ||
25ccd24f | 1779 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1780 | { |
25ccd24f | 1781 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1782 | |
9b40bc90 AV |
1783 | if (!may_mount()) |
1784 | return -EPERM; | |
41525f56 | 1785 | if (path->dentry != path->mnt->mnt_root) |
25ccd24f | 1786 | return -EINVAL; |
143c8c91 | 1787 | if (!check_mnt(mnt)) |
25ccd24f | 1788 | return -EINVAL; |
25d202ed | 1789 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1790 | return -EINVAL; |
b2f5d4dc | 1791 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1792 | return -EPERM; |
1793 | return 0; | |
1794 | } | |
1795 | ||
a0a6df9a | 1796 | // caller is responsible for flags being sane |
25ccd24f CH |
1797 | int path_umount(struct path *path, int flags) |
1798 | { | |
1799 | struct mount *mnt = real_mount(path->mnt); | |
1800 | int ret; | |
1801 | ||
1802 | ret = can_umount(path, flags); | |
1803 | if (!ret) | |
1804 | ret = do_umount(mnt, flags); | |
1da177e4 | 1805 | |
429731b1 | 1806 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1807 | dput(path->dentry); |
900148dc | 1808 | mntput_no_expire(mnt); |
25ccd24f | 1809 | return ret; |
1da177e4 LT |
1810 | } |
1811 | ||
09267def | 1812 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1813 | { |
1814 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1815 | struct path path; | |
1816 | int ret; | |
1817 | ||
a0a6df9a AV |
1818 | // basic validity checks done first |
1819 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1820 | return -EINVAL; | |
1821 | ||
41525f56 CH |
1822 | if (!(flags & UMOUNT_NOFOLLOW)) |
1823 | lookup_flags |= LOOKUP_FOLLOW; | |
1824 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1825 | if (ret) | |
1826 | return ret; | |
1827 | return path_umount(&path, flags); | |
1828 | } | |
1829 | ||
3a18ef5c DB |
1830 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1831 | { | |
1832 | return ksys_umount(name, flags); | |
1833 | } | |
1834 | ||
1da177e4 LT |
1835 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1836 | ||
1837 | /* | |
b58fed8b | 1838 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1839 | */ |
bdc480e3 | 1840 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1841 | { |
3a18ef5c | 1842 | return ksys_umount(name, 0); |
1da177e4 LT |
1843 | } |
1844 | ||
1845 | #endif | |
1846 | ||
4ce5d2b1 | 1847 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1848 | { |
4ce5d2b1 | 1849 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1850 | return dentry->d_op == &ns_dentry_operations && |
1851 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1852 | } |
1853 | ||
213921f9 | 1854 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1855 | { |
1856 | return container_of(ns, struct mnt_namespace, ns); | |
1857 | } | |
1858 | ||
303cc571 CB |
1859 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1860 | { | |
1861 | return &mnt->ns; | |
1862 | } | |
1863 | ||
4ce5d2b1 EB |
1864 | static bool mnt_ns_loop(struct dentry *dentry) |
1865 | { | |
1866 | /* Could bind mounting the mount namespace inode cause a | |
1867 | * mount namespace loop? | |
1868 | */ | |
1869 | struct mnt_namespace *mnt_ns; | |
1870 | if (!is_mnt_ns_file(dentry)) | |
1871 | return false; | |
1872 | ||
f77c8014 | 1873 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1874 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1875 | } | |
1876 | ||
87129cc0 | 1877 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1878 | int flag) |
1da177e4 | 1879 | { |
84d17192 | 1880 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1881 | |
4ce5d2b1 EB |
1882 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1883 | return ERR_PTR(-EINVAL); | |
1884 | ||
1885 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1886 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1887 | |
36341f64 | 1888 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1889 | if (IS_ERR(q)) |
1890 | return q; | |
1891 | ||
a73324da | 1892 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1893 | |
1894 | p = mnt; | |
6b41d536 | 1895 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1896 | struct mount *s; |
7ec02ef1 | 1897 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1898 | continue; |
1899 | ||
909b0a88 | 1900 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1901 | if (!(flag & CL_COPY_UNBINDABLE) && |
1902 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1903 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1904 | /* Both unbindable and locked. */ | |
1905 | q = ERR_PTR(-EPERM); | |
1906 | goto out; | |
1907 | } else { | |
1908 | s = skip_mnt_tree(s); | |
1909 | continue; | |
1910 | } | |
4ce5d2b1 EB |
1911 | } |
1912 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
1913 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
1914 | s = skip_mnt_tree(s); |
1915 | continue; | |
1916 | } | |
0714a533 AV |
1917 | while (p != s->mnt_parent) { |
1918 | p = p->mnt_parent; | |
1919 | q = q->mnt_parent; | |
1da177e4 | 1920 | } |
87129cc0 | 1921 | p = s; |
84d17192 | 1922 | parent = q; |
87129cc0 | 1923 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
1924 | if (IS_ERR(q)) |
1925 | goto out; | |
719ea2fb | 1926 | lock_mount_hash(); |
1a4eeaf2 | 1927 | list_add_tail(&q->mnt_list, &res->mnt_list); |
1064f874 | 1928 | attach_mnt(q, parent, p->mnt_mp); |
719ea2fb | 1929 | unlock_mount_hash(); |
1da177e4 LT |
1930 | } |
1931 | } | |
1932 | return res; | |
be34d1a3 | 1933 | out: |
1da177e4 | 1934 | if (res) { |
719ea2fb | 1935 | lock_mount_hash(); |
e819f152 | 1936 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 1937 | unlock_mount_hash(); |
1da177e4 | 1938 | } |
be34d1a3 | 1939 | return q; |
1da177e4 LT |
1940 | } |
1941 | ||
be34d1a3 DH |
1942 | /* Caller should check returned pointer for errors */ |
1943 | ||
ca71cf71 | 1944 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 1945 | { |
cb338d06 | 1946 | struct mount *tree; |
97216be0 | 1947 | namespace_lock(); |
cd4a4017 EB |
1948 | if (!check_mnt(real_mount(path->mnt))) |
1949 | tree = ERR_PTR(-EINVAL); | |
1950 | else | |
1951 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
1952 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 1953 | namespace_unlock(); |
be34d1a3 | 1954 | if (IS_ERR(tree)) |
52e220d3 | 1955 | return ERR_CAST(tree); |
be34d1a3 | 1956 | return &tree->mnt; |
8aec0809 AV |
1957 | } |
1958 | ||
a07b2000 AV |
1959 | static void free_mnt_ns(struct mnt_namespace *); |
1960 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
1961 | ||
1962 | void dissolve_on_fput(struct vfsmount *mnt) | |
1963 | { | |
1964 | struct mnt_namespace *ns; | |
1965 | namespace_lock(); | |
1966 | lock_mount_hash(); | |
1967 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
1968 | if (ns) { |
1969 | if (is_anon_ns(ns)) | |
1970 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
1971 | else | |
1972 | ns = NULL; | |
1973 | } | |
a07b2000 AV |
1974 | unlock_mount_hash(); |
1975 | namespace_unlock(); | |
44dfd84a DH |
1976 | if (ns) |
1977 | free_mnt_ns(ns); | |
a07b2000 AV |
1978 | } |
1979 | ||
8aec0809 AV |
1980 | void drop_collected_mounts(struct vfsmount *mnt) |
1981 | { | |
97216be0 | 1982 | namespace_lock(); |
719ea2fb | 1983 | lock_mount_hash(); |
9c8e0a1b | 1984 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 1985 | unlock_mount_hash(); |
3ab6abee | 1986 | namespace_unlock(); |
8aec0809 AV |
1987 | } |
1988 | ||
427215d8 MS |
1989 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
1990 | { | |
1991 | struct mount *child; | |
1992 | ||
1993 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
1994 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
1995 | continue; | |
1996 | ||
1997 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
1998 | return true; | |
1999 | } | |
2000 | return false; | |
2001 | } | |
2002 | ||
c771d683 MS |
2003 | /** |
2004 | * clone_private_mount - create a private clone of a path | |
1f287bc4 | 2005 | * @path: path to clone |
c771d683 | 2006 | * |
1f287bc4 RD |
2007 | * This creates a new vfsmount, which will be the clone of @path. The new mount |
2008 | * will not be attached anywhere in the namespace and will be private (i.e. | |
2009 | * changes to the originating mount won't be propagated into this). | |
c771d683 MS |
2010 | * |
2011 | * Release with mntput(). | |
2012 | */ | |
ca71cf71 | 2013 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
2014 | { |
2015 | struct mount *old_mnt = real_mount(path->mnt); | |
2016 | struct mount *new_mnt; | |
2017 | ||
427215d8 | 2018 | down_read(&namespace_sem); |
c771d683 | 2019 | if (IS_MNT_UNBINDABLE(old_mnt)) |
427215d8 MS |
2020 | goto invalid; |
2021 | ||
2022 | if (!check_mnt(old_mnt)) | |
2023 | goto invalid; | |
2024 | ||
2025 | if (has_locked_children(old_mnt, path->dentry)) | |
2026 | goto invalid; | |
c771d683 | 2027 | |
c771d683 | 2028 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
427215d8 MS |
2029 | up_read(&namespace_sem); |
2030 | ||
c771d683 MS |
2031 | if (IS_ERR(new_mnt)) |
2032 | return ERR_CAST(new_mnt); | |
2033 | ||
df820f8d MS |
2034 | /* Longterm mount to be removed by kern_unmount*() */ |
2035 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
2036 | ||
c771d683 | 2037 | return &new_mnt->mnt; |
427215d8 MS |
2038 | |
2039 | invalid: | |
2040 | up_read(&namespace_sem); | |
2041 | return ERR_PTR(-EINVAL); | |
c771d683 MS |
2042 | } |
2043 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
2044 | ||
1f707137 AV |
2045 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
2046 | struct vfsmount *root) | |
2047 | { | |
1a4eeaf2 | 2048 | struct mount *mnt; |
1f707137 AV |
2049 | int res = f(root, arg); |
2050 | if (res) | |
2051 | return res; | |
1a4eeaf2 AV |
2052 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
2053 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
2054 | if (res) |
2055 | return res; | |
2056 | } | |
2057 | return 0; | |
2058 | } | |
2059 | ||
3bd045cc AV |
2060 | static void lock_mnt_tree(struct mount *mnt) |
2061 | { | |
2062 | struct mount *p; | |
2063 | ||
2064 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2065 | int flags = p->mnt.mnt_flags; | |
2066 | /* Don't allow unprivileged users to change mount flags */ | |
2067 | flags |= MNT_LOCK_ATIME; | |
2068 | ||
2069 | if (flags & MNT_READONLY) | |
2070 | flags |= MNT_LOCK_READONLY; | |
2071 | ||
2072 | if (flags & MNT_NODEV) | |
2073 | flags |= MNT_LOCK_NODEV; | |
2074 | ||
2075 | if (flags & MNT_NOSUID) | |
2076 | flags |= MNT_LOCK_NOSUID; | |
2077 | ||
2078 | if (flags & MNT_NOEXEC) | |
2079 | flags |= MNT_LOCK_NOEXEC; | |
2080 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2081 | if (list_empty(&p->mnt_expire)) | |
2082 | flags |= MNT_LOCKED; | |
2083 | p->mnt.mnt_flags = flags; | |
2084 | } | |
2085 | } | |
2086 | ||
4b8b21f4 | 2087 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2088 | { |
315fc83e | 2089 | struct mount *p; |
719f5d7f | 2090 | |
909b0a88 | 2091 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2092 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2093 | mnt_release_group_id(p); |
719f5d7f MS |
2094 | } |
2095 | } | |
2096 | ||
4b8b21f4 | 2097 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2098 | { |
315fc83e | 2099 | struct mount *p; |
719f5d7f | 2100 | |
909b0a88 | 2101 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2102 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2103 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2104 | if (err) { |
4b8b21f4 | 2105 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2106 | return err; |
2107 | } | |
2108 | } | |
2109 | } | |
2110 | ||
2111 | return 0; | |
2112 | } | |
2113 | ||
d2921684 EB |
2114 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2115 | { | |
2116 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
124f75f8 | 2117 | unsigned int mounts = 0; |
d2921684 EB |
2118 | struct mount *p; |
2119 | ||
124f75f8 AV |
2120 | if (ns->mounts >= max) |
2121 | return -ENOSPC; | |
2122 | max -= ns->mounts; | |
2123 | if (ns->pending_mounts >= max) | |
2124 | return -ENOSPC; | |
2125 | max -= ns->pending_mounts; | |
2126 | ||
d2921684 EB |
2127 | for (p = mnt; p; p = next_mnt(p, mnt)) |
2128 | mounts++; | |
2129 | ||
124f75f8 | 2130 | if (mounts > max) |
d2921684 EB |
2131 | return -ENOSPC; |
2132 | ||
124f75f8 | 2133 | ns->pending_mounts += mounts; |
d2921684 EB |
2134 | return 0; |
2135 | } | |
2136 | ||
b90fa9ae RP |
2137 | /* |
2138 | * @source_mnt : mount tree to be attached | |
21444403 RP |
2139 | * @nd : place the mount tree @source_mnt is attached |
2140 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
2141 | * store the parent mount and mountpoint dentry. | |
2142 | * (done when source_mnt is moved) | |
b90fa9ae RP |
2143 | * |
2144 | * NOTE: in the table below explains the semantics when a source mount | |
2145 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2146 | * --------------------------------------------------------------------------- |
2147 | * | BIND MOUNT OPERATION | | |
2148 | * |************************************************************************** | |
2149 | * | source-->| shared | private | slave | unbindable | | |
2150 | * | dest | | | | | | |
2151 | * | | | | | | | | |
2152 | * | v | | | | | | |
2153 | * |************************************************************************** | |
2154 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2155 | * | | | | | | | |
2156 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2157 | * *************************************************************************** | |
b90fa9ae RP |
2158 | * A bind operation clones the source mount and mounts the clone on the |
2159 | * destination mount. | |
2160 | * | |
2161 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2162 | * tree of the destination mount and the cloned mount is added to | |
2163 | * the peer group of the source mount. | |
2164 | * (+) the cloned mount is created under the destination mount and is marked | |
2165 | * as shared. The cloned mount is added to the peer group of the source | |
2166 | * mount. | |
5afe0022 RP |
2167 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2168 | * of the destination mount and the cloned mount is made slave | |
2169 | * of the same master as that of the source mount. The cloned mount | |
2170 | * is marked as 'shared and slave'. | |
2171 | * (*) the cloned mount is made a slave of the same master as that of the | |
2172 | * source mount. | |
2173 | * | |
9676f0c6 RP |
2174 | * --------------------------------------------------------------------------- |
2175 | * | MOVE MOUNT OPERATION | | |
2176 | * |************************************************************************** | |
2177 | * | source-->| shared | private | slave | unbindable | | |
2178 | * | dest | | | | | | |
2179 | * | | | | | | | | |
2180 | * | v | | | | | | |
2181 | * |************************************************************************** | |
2182 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2183 | * | | | | | | | |
2184 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2185 | * *************************************************************************** | |
5afe0022 RP |
2186 | * |
2187 | * (+) the mount is moved to the destination. And is then propagated to | |
2188 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2189 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2190 | * (+++) the mount is moved to the destination and is then propagated to |
2191 | * all the mounts belonging to the destination mount's propagation tree. | |
2192 | * the mount is marked as 'shared and slave'. | |
2193 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2194 | * |
2195 | * if the source mount is a tree, the operations explained above is | |
2196 | * applied to each mount in the tree. | |
2197 | * Must be called without spinlocks held, since this function can sleep | |
2198 | * in allocations. | |
2199 | */ | |
0fb54e50 | 2200 | static int attach_recursive_mnt(struct mount *source_mnt, |
84d17192 AV |
2201 | struct mount *dest_mnt, |
2202 | struct mountpoint *dest_mp, | |
2763d119 | 2203 | bool moving) |
b90fa9ae | 2204 | { |
3bd045cc | 2205 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2206 | HLIST_HEAD(tree_list); |
d2921684 | 2207 | struct mnt_namespace *ns = dest_mnt->mnt_ns; |
1064f874 | 2208 | struct mountpoint *smp; |
315fc83e | 2209 | struct mount *child, *p; |
38129a13 | 2210 | struct hlist_node *n; |
719f5d7f | 2211 | int err; |
b90fa9ae | 2212 | |
1064f874 EB |
2213 | /* Preallocate a mountpoint in case the new mounts need |
2214 | * to be tucked under other mounts. | |
2215 | */ | |
2216 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2217 | if (IS_ERR(smp)) | |
2218 | return PTR_ERR(smp); | |
2219 | ||
d2921684 | 2220 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2221 | if (!moving) { |
d2921684 EB |
2222 | err = count_mounts(ns, source_mnt); |
2223 | if (err) | |
2224 | goto out; | |
2225 | } | |
2226 | ||
fc7be130 | 2227 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2228 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2229 | if (err) |
2230 | goto out; | |
0b1b901b | 2231 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
f2ebb3a9 | 2232 | lock_mount_hash(); |
0b1b901b AV |
2233 | if (err) |
2234 | goto out_cleanup_ids; | |
909b0a88 | 2235 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2236 | set_mnt_shared(p); |
0b1b901b AV |
2237 | } else { |
2238 | lock_mount_hash(); | |
b90fa9ae | 2239 | } |
2763d119 AV |
2240 | if (moving) { |
2241 | unhash_mnt(source_mnt); | |
84d17192 | 2242 | attach_mnt(source_mnt, dest_mnt, dest_mp); |
143c8c91 | 2243 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2244 | } else { |
44dfd84a DH |
2245 | if (source_mnt->mnt_ns) { |
2246 | /* move from anon - the caller will destroy */ | |
2247 | list_del_init(&source_mnt->mnt_ns->list); | |
2248 | } | |
84d17192 | 2249 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); |
1064f874 | 2250 | commit_tree(source_mnt); |
21444403 | 2251 | } |
b90fa9ae | 2252 | |
38129a13 | 2253 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2254 | struct mount *q; |
38129a13 | 2255 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2256 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2257 | child->mnt_mountpoint); | |
2258 | if (q) | |
2259 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2260 | /* Notice when we are propagating across user namespaces */ |
2261 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2262 | lock_mnt_tree(child); | |
d728cf79 | 2263 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2264 | commit_tree(child); |
b90fa9ae | 2265 | } |
1064f874 | 2266 | put_mountpoint(smp); |
719ea2fb | 2267 | unlock_mount_hash(); |
99b7db7b | 2268 | |
b90fa9ae | 2269 | return 0; |
719f5d7f MS |
2270 | |
2271 | out_cleanup_ids: | |
f2ebb3a9 AV |
2272 | while (!hlist_empty(&tree_list)) { |
2273 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2274 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2275 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2276 | } |
2277 | unlock_mount_hash(); | |
0b1b901b | 2278 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2279 | out: |
d2921684 | 2280 | ns->pending_mounts = 0; |
1064f874 EB |
2281 | |
2282 | read_seqlock_excl(&mount_lock); | |
2283 | put_mountpoint(smp); | |
2284 | read_sequnlock_excl(&mount_lock); | |
2285 | ||
719f5d7f | 2286 | return err; |
b90fa9ae RP |
2287 | } |
2288 | ||
84d17192 | 2289 | static struct mountpoint *lock_mount(struct path *path) |
b12cea91 AV |
2290 | { |
2291 | struct vfsmount *mnt; | |
84d17192 | 2292 | struct dentry *dentry = path->dentry; |
b12cea91 | 2293 | retry: |
5955102c | 2294 | inode_lock(dentry->d_inode); |
84d17192 | 2295 | if (unlikely(cant_mount(dentry))) { |
5955102c | 2296 | inode_unlock(dentry->d_inode); |
84d17192 | 2297 | return ERR_PTR(-ENOENT); |
b12cea91 | 2298 | } |
97216be0 | 2299 | namespace_lock(); |
b12cea91 | 2300 | mnt = lookup_mnt(path); |
84d17192 | 2301 | if (likely(!mnt)) { |
3895dbf8 | 2302 | struct mountpoint *mp = get_mountpoint(dentry); |
84d17192 | 2303 | if (IS_ERR(mp)) { |
97216be0 | 2304 | namespace_unlock(); |
5955102c | 2305 | inode_unlock(dentry->d_inode); |
84d17192 AV |
2306 | return mp; |
2307 | } | |
2308 | return mp; | |
2309 | } | |
97216be0 | 2310 | namespace_unlock(); |
5955102c | 2311 | inode_unlock(path->dentry->d_inode); |
b12cea91 AV |
2312 | path_put(path); |
2313 | path->mnt = mnt; | |
84d17192 | 2314 | dentry = path->dentry = dget(mnt->mnt_root); |
b12cea91 AV |
2315 | goto retry; |
2316 | } | |
2317 | ||
84d17192 | 2318 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2319 | { |
84d17192 | 2320 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2321 | |
2322 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2323 | put_mountpoint(where); |
3895dbf8 EB |
2324 | read_sequnlock_excl(&mount_lock); |
2325 | ||
328e6d90 | 2326 | namespace_unlock(); |
5955102c | 2327 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2328 | } |
2329 | ||
84d17192 | 2330 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2331 | { |
e462ec50 | 2332 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2333 | return -EINVAL; |
2334 | ||
e36cb0b8 DH |
2335 | if (d_is_dir(mp->m_dentry) != |
2336 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2337 | return -ENOTDIR; |
2338 | ||
2763d119 | 2339 | return attach_recursive_mnt(mnt, p, mp, false); |
1da177e4 LT |
2340 | } |
2341 | ||
7a2e8a8f VA |
2342 | /* |
2343 | * Sanity check the flags to change_mnt_propagation. | |
2344 | */ | |
2345 | ||
e462ec50 | 2346 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2347 | { |
e462ec50 | 2348 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2349 | |
2350 | /* Fail if any non-propagation flags are set */ | |
2351 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2352 | return 0; | |
2353 | /* Only one propagation flag should be set */ | |
2354 | if (!is_power_of_2(type)) | |
2355 | return 0; | |
2356 | return type; | |
2357 | } | |
2358 | ||
07b20889 RP |
2359 | /* |
2360 | * recursively change the type of the mountpoint. | |
2361 | */ | |
e462ec50 | 2362 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2363 | { |
315fc83e | 2364 | struct mount *m; |
4b8b21f4 | 2365 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2366 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2367 | int type; |
719f5d7f | 2368 | int err = 0; |
07b20889 | 2369 | |
2d92ab3c | 2370 | if (path->dentry != path->mnt->mnt_root) |
07b20889 RP |
2371 | return -EINVAL; |
2372 | ||
e462ec50 | 2373 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2374 | if (!type) |
2375 | return -EINVAL; | |
2376 | ||
97216be0 | 2377 | namespace_lock(); |
719f5d7f MS |
2378 | if (type == MS_SHARED) { |
2379 | err = invent_group_ids(mnt, recurse); | |
2380 | if (err) | |
2381 | goto out_unlock; | |
2382 | } | |
2383 | ||
719ea2fb | 2384 | lock_mount_hash(); |
909b0a88 | 2385 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2386 | change_mnt_propagation(m, type); |
719ea2fb | 2387 | unlock_mount_hash(); |
719f5d7f MS |
2388 | |
2389 | out_unlock: | |
97216be0 | 2390 | namespace_unlock(); |
719f5d7f | 2391 | return err; |
07b20889 RP |
2392 | } |
2393 | ||
a07b2000 AV |
2394 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2395 | { | |
2396 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2397 | ||
2398 | if (IS_MNT_UNBINDABLE(old)) | |
2399 | return mnt; | |
2400 | ||
2401 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2402 | return mnt; | |
2403 | ||
2404 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2405 | return mnt; | |
2406 | ||
2407 | if (recurse) | |
2408 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2409 | else | |
2410 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2411 | ||
2412 | if (!IS_ERR(mnt)) | |
2413 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2414 | ||
2415 | return mnt; | |
2416 | } | |
2417 | ||
1da177e4 LT |
2418 | /* |
2419 | * do loopback mount. | |
2420 | */ | |
808d4e3c | 2421 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2422 | int recurse) |
1da177e4 | 2423 | { |
2d92ab3c | 2424 | struct path old_path; |
a07b2000 | 2425 | struct mount *mnt = NULL, *parent; |
84d17192 | 2426 | struct mountpoint *mp; |
57eccb83 | 2427 | int err; |
1da177e4 LT |
2428 | if (!old_name || !*old_name) |
2429 | return -EINVAL; | |
815d405c | 2430 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2431 | if (err) |
2432 | return err; | |
2433 | ||
8823c079 | 2434 | err = -EINVAL; |
4ce5d2b1 | 2435 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2436 | goto out; |
8823c079 | 2437 | |
84d17192 | 2438 | mp = lock_mount(path); |
a07b2000 AV |
2439 | if (IS_ERR(mp)) { |
2440 | err = PTR_ERR(mp); | |
b12cea91 | 2441 | goto out; |
a07b2000 | 2442 | } |
b12cea91 | 2443 | |
84d17192 | 2444 | parent = real_mount(path->mnt); |
e149ed2b AV |
2445 | if (!check_mnt(parent)) |
2446 | goto out2; | |
2447 | ||
a07b2000 | 2448 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2449 | if (IS_ERR(mnt)) { |
2450 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2451 | goto out2; |
be34d1a3 | 2452 | } |
ccd48bc7 | 2453 | |
84d17192 | 2454 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2455 | if (err) { |
719ea2fb | 2456 | lock_mount_hash(); |
e819f152 | 2457 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2458 | unlock_mount_hash(); |
5b83d2c5 | 2459 | } |
b12cea91 | 2460 | out2: |
84d17192 | 2461 | unlock_mount(mp); |
ccd48bc7 | 2462 | out: |
2d92ab3c | 2463 | path_put(&old_path); |
1da177e4 LT |
2464 | return err; |
2465 | } | |
2466 | ||
a07b2000 AV |
2467 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2468 | { | |
2469 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2470 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2471 | struct mount *mnt, *p; | |
2472 | struct file *file; | |
2473 | ||
2474 | if (IS_ERR(ns)) | |
2475 | return ERR_CAST(ns); | |
2476 | ||
2477 | namespace_lock(); | |
2478 | mnt = __do_loopback(path, recursive); | |
2479 | if (IS_ERR(mnt)) { | |
2480 | namespace_unlock(); | |
2481 | free_mnt_ns(ns); | |
2482 | return ERR_CAST(mnt); | |
2483 | } | |
2484 | ||
2485 | lock_mount_hash(); | |
2486 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2487 | p->mnt_ns = ns; | |
2488 | ns->mounts++; | |
2489 | } | |
2490 | ns->root = mnt; | |
2491 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2492 | mntget(&mnt->mnt); | |
2493 | unlock_mount_hash(); | |
2494 | namespace_unlock(); | |
2495 | ||
2496 | mntput(path->mnt); | |
2497 | path->mnt = &mnt->mnt; | |
2498 | file = dentry_open(path, O_PATH, current_cred()); | |
2499 | if (IS_ERR(file)) | |
2500 | dissolve_on_fput(path->mnt); | |
2501 | else | |
2502 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2503 | return file; | |
2504 | } | |
2505 | ||
2658ce09 | 2506 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2507 | { |
2508 | struct file *file; | |
2509 | struct path path; | |
2510 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2511 | bool detached = flags & OPEN_TREE_CLONE; | |
2512 | int error; | |
2513 | int fd; | |
2514 | ||
2515 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2516 | ||
2517 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2518 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2519 | OPEN_TREE_CLOEXEC)) | |
2520 | return -EINVAL; | |
2521 | ||
2522 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2523 | return -EINVAL; | |
2524 | ||
2525 | if (flags & AT_NO_AUTOMOUNT) | |
2526 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2527 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2528 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2529 | if (flags & AT_EMPTY_PATH) | |
2530 | lookup_flags |= LOOKUP_EMPTY; | |
2531 | ||
2532 | if (detached && !may_mount()) | |
2533 | return -EPERM; | |
2534 | ||
2535 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2536 | if (fd < 0) | |
2537 | return fd; | |
2538 | ||
2539 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2540 | if (unlikely(error)) { | |
2541 | file = ERR_PTR(error); | |
2542 | } else { | |
2543 | if (detached) | |
2544 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2545 | else | |
2546 | file = dentry_open(&path, O_PATH, current_cred()); | |
2547 | path_put(&path); | |
2548 | } | |
2549 | if (IS_ERR(file)) { | |
2550 | put_unused_fd(fd); | |
2551 | return PTR_ERR(file); | |
2552 | } | |
2553 | fd_install(fd, file); | |
2554 | return fd; | |
2555 | } | |
2556 | ||
43f5e655 DH |
2557 | /* |
2558 | * Don't allow locked mount flags to be cleared. | |
2559 | * | |
2560 | * No locks need to be held here while testing the various MNT_LOCK | |
2561 | * flags because those flags can never be cleared once they are set. | |
2562 | */ | |
2563 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2564 | { |
43f5e655 DH |
2565 | unsigned int fl = mnt->mnt.mnt_flags; |
2566 | ||
2567 | if ((fl & MNT_LOCK_READONLY) && | |
2568 | !(mnt_flags & MNT_READONLY)) | |
2569 | return false; | |
2570 | ||
2571 | if ((fl & MNT_LOCK_NODEV) && | |
2572 | !(mnt_flags & MNT_NODEV)) | |
2573 | return false; | |
2574 | ||
2575 | if ((fl & MNT_LOCK_NOSUID) && | |
2576 | !(mnt_flags & MNT_NOSUID)) | |
2577 | return false; | |
2578 | ||
2579 | if ((fl & MNT_LOCK_NOEXEC) && | |
2580 | !(mnt_flags & MNT_NOEXEC)) | |
2581 | return false; | |
2582 | ||
2583 | if ((fl & MNT_LOCK_ATIME) && | |
2584 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2585 | return false; | |
2e4b7fcd | 2586 | |
43f5e655 DH |
2587 | return true; |
2588 | } | |
2589 | ||
2590 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2591 | { |
43f5e655 | 2592 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2593 | |
43f5e655 | 2594 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2595 | return 0; |
2596 | ||
2597 | if (readonly_request) | |
43f5e655 DH |
2598 | return mnt_make_readonly(mnt); |
2599 | ||
68847c94 CB |
2600 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2601 | return 0; | |
43f5e655 DH |
2602 | } |
2603 | ||
43f5e655 DH |
2604 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2605 | { | |
43f5e655 DH |
2606 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2607 | mnt->mnt.mnt_flags = mnt_flags; | |
2608 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2609 | } |
2610 | ||
f8b92ba6 DD |
2611 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2612 | { | |
2613 | struct super_block *sb = mnt->mnt_sb; | |
2614 | ||
2615 | if (!__mnt_is_readonly(mnt) && | |
a128b054 | 2616 | (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && |
f8b92ba6 DD |
2617 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { |
2618 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2619 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
2620 | struct tm tm; | |
2621 | ||
2622 | time64_to_tm(sb->s_time_max, 0, &tm); | |
2623 | ||
0ecee669 EB |
2624 | pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", |
2625 | sb->s_type->name, | |
2626 | is_mounted(mnt) ? "remounted" : "mounted", | |
2627 | mntpath, | |
f8b92ba6 DD |
2628 | tm.tm_year+1900, (unsigned long long)sb->s_time_max); |
2629 | ||
2630 | free_page((unsigned long)buf); | |
a128b054 | 2631 | sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; |
f8b92ba6 DD |
2632 | } |
2633 | } | |
2634 | ||
43f5e655 DH |
2635 | /* |
2636 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2637 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2638 | * to mount(2). | |
2639 | */ | |
2640 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2641 | { | |
2642 | struct super_block *sb = path->mnt->mnt_sb; | |
2643 | struct mount *mnt = real_mount(path->mnt); | |
2644 | int ret; | |
2645 | ||
2646 | if (!check_mnt(mnt)) | |
2647 | return -EINVAL; | |
2648 | ||
2649 | if (path->dentry != mnt->mnt.mnt_root) | |
2650 | return -EINVAL; | |
2651 | ||
2652 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2653 | return -EPERM; | |
2654 | ||
e58ace1a CB |
2655 | /* |
2656 | * We're only checking whether the superblock is read-only not | |
2657 | * changing it, so only take down_read(&sb->s_umount). | |
2658 | */ | |
2659 | down_read(&sb->s_umount); | |
68847c94 | 2660 | lock_mount_hash(); |
43f5e655 DH |
2661 | ret = change_mount_ro_state(mnt, mnt_flags); |
2662 | if (ret == 0) | |
2663 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2664 | unlock_mount_hash(); |
e58ace1a | 2665 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2666 | |
2667 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2668 | ||
43f5e655 | 2669 | return ret; |
2e4b7fcd DH |
2670 | } |
2671 | ||
1da177e4 LT |
2672 | /* |
2673 | * change filesystem flags. dir should be a physical root of filesystem. | |
2674 | * If you've mounted a non-root directory somewhere and want to do remount | |
2675 | * on it - tough luck. | |
2676 | */ | |
e462ec50 DH |
2677 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2678 | int mnt_flags, void *data) | |
1da177e4 LT |
2679 | { |
2680 | int err; | |
2d92ab3c | 2681 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2682 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2683 | struct fs_context *fc; |
1da177e4 | 2684 | |
143c8c91 | 2685 | if (!check_mnt(mnt)) |
1da177e4 LT |
2686 | return -EINVAL; |
2687 | ||
2d92ab3c | 2688 | if (path->dentry != path->mnt->mnt_root) |
1da177e4 LT |
2689 | return -EINVAL; |
2690 | ||
43f5e655 | 2691 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2692 | return -EPERM; |
9566d674 | 2693 | |
8d0347f6 DH |
2694 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2695 | if (IS_ERR(fc)) | |
2696 | return PTR_ERR(fc); | |
ff36fe2c | 2697 | |
b330966f | 2698 | fc->oldapi = true; |
8d0347f6 DH |
2699 | err = parse_monolithic_mount_data(fc, data); |
2700 | if (!err) { | |
2701 | down_write(&sb->s_umount); | |
2702 | err = -EPERM; | |
2703 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2704 | err = reconfigure_super(fc); | |
68847c94 CB |
2705 | if (!err) { |
2706 | lock_mount_hash(); | |
8d0347f6 | 2707 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2708 | unlock_mount_hash(); |
2709 | } | |
8d0347f6 DH |
2710 | } |
2711 | up_write(&sb->s_umount); | |
0e55a7cc | 2712 | } |
f8b92ba6 DD |
2713 | |
2714 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2715 | ||
8d0347f6 | 2716 | put_fs_context(fc); |
1da177e4 LT |
2717 | return err; |
2718 | } | |
2719 | ||
cbbe362c | 2720 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2721 | { |
315fc83e | 2722 | struct mount *p; |
909b0a88 | 2723 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2724 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2725 | return 1; |
2726 | } | |
2727 | return 0; | |
2728 | } | |
2729 | ||
44dfd84a DH |
2730 | /* |
2731 | * Check that there aren't references to earlier/same mount namespaces in the | |
2732 | * specified subtree. Such references can act as pins for mount namespaces | |
2733 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2734 | * cycles to be made. | |
2735 | */ | |
2736 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2737 | { | |
2738 | struct mount *p; | |
2739 | bool ret = false; | |
2740 | ||
2741 | lock_mount_hash(); | |
2742 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2743 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2744 | goto out; | |
2745 | ||
2746 | ret = true; | |
2747 | out: | |
2748 | unlock_mount_hash(); | |
2749 | return ret; | |
2750 | } | |
2751 | ||
9ffb14ef PT |
2752 | static int do_set_group(struct path *from_path, struct path *to_path) |
2753 | { | |
2754 | struct mount *from, *to; | |
2755 | int err; | |
2756 | ||
2757 | from = real_mount(from_path->mnt); | |
2758 | to = real_mount(to_path->mnt); | |
2759 | ||
2760 | namespace_lock(); | |
2761 | ||
2762 | err = -EINVAL; | |
2763 | /* To and From must be mounted */ | |
2764 | if (!is_mounted(&from->mnt)) | |
2765 | goto out; | |
2766 | if (!is_mounted(&to->mnt)) | |
2767 | goto out; | |
2768 | ||
2769 | err = -EPERM; | |
2770 | /* We should be allowed to modify mount namespaces of both mounts */ | |
2771 | if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2772 | goto out; | |
2773 | if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2774 | goto out; | |
2775 | ||
2776 | err = -EINVAL; | |
2777 | /* To and From paths should be mount roots */ | |
2778 | if (from_path->dentry != from_path->mnt->mnt_root) | |
2779 | goto out; | |
2780 | if (to_path->dentry != to_path->mnt->mnt_root) | |
2781 | goto out; | |
2782 | ||
2783 | /* Setting sharing groups is only allowed across same superblock */ | |
2784 | if (from->mnt.mnt_sb != to->mnt.mnt_sb) | |
2785 | goto out; | |
2786 | ||
2787 | /* From mount root should be wider than To mount root */ | |
2788 | if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) | |
2789 | goto out; | |
2790 | ||
2791 | /* From mount should not have locked children in place of To's root */ | |
2792 | if (has_locked_children(from, to->mnt.mnt_root)) | |
2793 | goto out; | |
2794 | ||
2795 | /* Setting sharing groups is only allowed on private mounts */ | |
2796 | if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) | |
2797 | goto out; | |
2798 | ||
2799 | /* From should not be private */ | |
2800 | if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) | |
2801 | goto out; | |
2802 | ||
2803 | if (IS_MNT_SLAVE(from)) { | |
2804 | struct mount *m = from->mnt_master; | |
2805 | ||
2806 | list_add(&to->mnt_slave, &m->mnt_slave_list); | |
2807 | to->mnt_master = m; | |
2808 | } | |
2809 | ||
2810 | if (IS_MNT_SHARED(from)) { | |
2811 | to->mnt_group_id = from->mnt_group_id; | |
2812 | list_add(&to->mnt_share, &from->mnt_share); | |
2813 | lock_mount_hash(); | |
2814 | set_mnt_shared(to); | |
2815 | unlock_mount_hash(); | |
2816 | } | |
2817 | ||
2818 | err = 0; | |
2819 | out: | |
2820 | namespace_unlock(); | |
2821 | return err; | |
2822 | } | |
2823 | ||
2db154b3 | 2824 | static int do_move_mount(struct path *old_path, struct path *new_path) |
1da177e4 | 2825 | { |
44dfd84a | 2826 | struct mnt_namespace *ns; |
676da58d | 2827 | struct mount *p; |
0fb54e50 | 2828 | struct mount *old; |
2763d119 AV |
2829 | struct mount *parent; |
2830 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 2831 | int err; |
44dfd84a | 2832 | bool attached; |
1da177e4 | 2833 | |
2db154b3 | 2834 | mp = lock_mount(new_path); |
84d17192 | 2835 | if (IS_ERR(mp)) |
2db154b3 | 2836 | return PTR_ERR(mp); |
cc53ce53 | 2837 | |
2db154b3 DH |
2838 | old = real_mount(old_path->mnt); |
2839 | p = real_mount(new_path->mnt); | |
2763d119 | 2840 | parent = old->mnt_parent; |
44dfd84a | 2841 | attached = mnt_has_parent(old); |
2763d119 | 2842 | old_mp = old->mnt_mp; |
44dfd84a | 2843 | ns = old->mnt_ns; |
143c8c91 | 2844 | |
1da177e4 | 2845 | err = -EINVAL; |
44dfd84a DH |
2846 | /* The mountpoint must be in our namespace. */ |
2847 | if (!check_mnt(p)) | |
2db154b3 | 2848 | goto out; |
1da177e4 | 2849 | |
570d7a98 EB |
2850 | /* The thing moved must be mounted... */ |
2851 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
2852 | goto out; |
2853 | ||
570d7a98 EB |
2854 | /* ... and either ours or the root of anon namespace */ |
2855 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 2856 | goto out; |
5ff9d8a6 | 2857 | |
2db154b3 DH |
2858 | if (old->mnt.mnt_flags & MNT_LOCKED) |
2859 | goto out; | |
1da177e4 | 2860 | |
2db154b3 DH |
2861 | if (old_path->dentry != old_path->mnt->mnt_root) |
2862 | goto out; | |
1da177e4 | 2863 | |
2db154b3 DH |
2864 | if (d_is_dir(new_path->dentry) != |
2865 | d_is_dir(old_path->dentry)) | |
2866 | goto out; | |
21444403 RP |
2867 | /* |
2868 | * Don't move a mount residing in a shared parent. | |
2869 | */ | |
2763d119 | 2870 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 2871 | goto out; |
9676f0c6 RP |
2872 | /* |
2873 | * Don't move a mount tree containing unbindable mounts to a destination | |
2874 | * mount which is shared. | |
2875 | */ | |
fc7be130 | 2876 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 2877 | goto out; |
1da177e4 | 2878 | err = -ELOOP; |
44dfd84a DH |
2879 | if (!check_for_nsfs_mounts(old)) |
2880 | goto out; | |
fc7be130 | 2881 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 2882 | if (p == old) |
2db154b3 | 2883 | goto out; |
1da177e4 | 2884 | |
2db154b3 | 2885 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, |
2763d119 | 2886 | attached); |
4ac91378 | 2887 | if (err) |
2db154b3 | 2888 | goto out; |
1da177e4 LT |
2889 | |
2890 | /* if the mount is moved, it should no longer be expire | |
2891 | * automatically */ | |
6776db3d | 2892 | list_del_init(&old->mnt_expire); |
2763d119 AV |
2893 | if (attached) |
2894 | put_mountpoint(old_mp); | |
1da177e4 | 2895 | out: |
2db154b3 | 2896 | unlock_mount(mp); |
44dfd84a | 2897 | if (!err) { |
2763d119 AV |
2898 | if (attached) |
2899 | mntput_no_expire(parent); | |
2900 | else | |
44dfd84a DH |
2901 | free_mnt_ns(ns); |
2902 | } | |
2db154b3 DH |
2903 | return err; |
2904 | } | |
2905 | ||
2906 | static int do_move_mount_old(struct path *path, const char *old_name) | |
2907 | { | |
2908 | struct path old_path; | |
2909 | int err; | |
2910 | ||
2911 | if (!old_name || !*old_name) | |
2912 | return -EINVAL; | |
2913 | ||
2914 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
2915 | if (err) | |
2916 | return err; | |
2917 | ||
2918 | err = do_move_mount(&old_path, path); | |
2d92ab3c | 2919 | path_put(&old_path); |
1da177e4 LT |
2920 | return err; |
2921 | } | |
2922 | ||
9d412a43 AV |
2923 | /* |
2924 | * add a mount into a namespace's mount tree | |
2925 | */ | |
8f11538e | 2926 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
1e2d8464 | 2927 | const struct path *path, int mnt_flags) |
9d412a43 | 2928 | { |
8f11538e | 2929 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 2930 | |
f2ebb3a9 | 2931 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 2932 | |
84d17192 | 2933 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
2934 | /* that's acceptable only for automounts done in private ns */ |
2935 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 2936 | return -EINVAL; |
156cacb1 | 2937 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 2938 | if (!parent->mnt_ns) |
8f11538e | 2939 | return -EINVAL; |
156cacb1 | 2940 | } |
9d412a43 AV |
2941 | |
2942 | /* Refuse the same filesystem on the same mount point */ | |
95bc5f25 | 2943 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && |
9d412a43 | 2944 | path->mnt->mnt_root == path->dentry) |
8f11538e | 2945 | return -EBUSY; |
9d412a43 | 2946 | |
e36cb0b8 | 2947 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 2948 | return -EINVAL; |
9d412a43 | 2949 | |
95bc5f25 | 2950 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 2951 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 2952 | } |
b1e75df4 | 2953 | |
132e4608 DH |
2954 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
2955 | ||
2956 | /* | |
2957 | * Create a new mount using a superblock configuration and request it | |
2958 | * be added to the namespace tree. | |
2959 | */ | |
2960 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
2961 | unsigned int mnt_flags) | |
2962 | { | |
2963 | struct vfsmount *mnt; | |
8f11538e | 2964 | struct mountpoint *mp; |
132e4608 DH |
2965 | struct super_block *sb = fc->root->d_sb; |
2966 | int error; | |
2967 | ||
c9ce29ed AV |
2968 | error = security_sb_kern_mount(sb); |
2969 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
2970 | error = -EPERM; | |
2971 | ||
2972 | if (unlikely(error)) { | |
2973 | fc_drop_locked(fc); | |
2974 | return error; | |
132e4608 DH |
2975 | } |
2976 | ||
2977 | up_write(&sb->s_umount); | |
2978 | ||
2979 | mnt = vfs_create_mount(fc); | |
2980 | if (IS_ERR(mnt)) | |
2981 | return PTR_ERR(mnt); | |
2982 | ||
f8b92ba6 DD |
2983 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
2984 | ||
8f11538e AV |
2985 | mp = lock_mount(mountpoint); |
2986 | if (IS_ERR(mp)) { | |
2987 | mntput(mnt); | |
2988 | return PTR_ERR(mp); | |
2989 | } | |
2990 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
2991 | unlock_mount(mp); | |
0ecee669 EB |
2992 | if (error < 0) |
2993 | mntput(mnt); | |
132e4608 DH |
2994 | return error; |
2995 | } | |
1b852bce | 2996 | |
1da177e4 LT |
2997 | /* |
2998 | * create a new mount for userspace and request it to be added into the | |
2999 | * namespace's tree | |
3000 | */ | |
e462ec50 | 3001 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 3002 | int mnt_flags, const char *name, void *data) |
1da177e4 | 3003 | { |
0c55cfc4 | 3004 | struct file_system_type *type; |
a0c9a8b8 AV |
3005 | struct fs_context *fc; |
3006 | const char *subtype = NULL; | |
3007 | int err = 0; | |
1da177e4 | 3008 | |
0c55cfc4 | 3009 | if (!fstype) |
1da177e4 LT |
3010 | return -EINVAL; |
3011 | ||
0c55cfc4 EB |
3012 | type = get_fs_type(fstype); |
3013 | if (!type) | |
3014 | return -ENODEV; | |
3015 | ||
a0c9a8b8 AV |
3016 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
3017 | subtype = strchr(fstype, '.'); | |
3018 | if (subtype) { | |
3019 | subtype++; | |
3020 | if (!*subtype) { | |
3021 | put_filesystem(type); | |
3022 | return -EINVAL; | |
3023 | } | |
a0c9a8b8 AV |
3024 | } |
3025 | } | |
0c55cfc4 | 3026 | |
a0c9a8b8 | 3027 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 3028 | put_filesystem(type); |
a0c9a8b8 AV |
3029 | if (IS_ERR(fc)) |
3030 | return PTR_ERR(fc); | |
3031 | ||
3e1aeb00 DH |
3032 | if (subtype) |
3033 | err = vfs_parse_fs_string(fc, "subtype", | |
3034 | subtype, strlen(subtype)); | |
3035 | if (!err && name) | |
3036 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
3037 | if (!err) |
3038 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
3039 | if (!err && !mount_capable(fc)) |
3040 | err = -EPERM; | |
a0c9a8b8 AV |
3041 | if (!err) |
3042 | err = vfs_get_tree(fc); | |
132e4608 DH |
3043 | if (!err) |
3044 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 3045 | |
a0c9a8b8 | 3046 | put_fs_context(fc); |
15f9a3f3 | 3047 | return err; |
1da177e4 LT |
3048 | } |
3049 | ||
1e2d8464 | 3050 | int finish_automount(struct vfsmount *m, const struct path *path) |
19a167af | 3051 | { |
26df6034 | 3052 | struct dentry *dentry = path->dentry; |
8f11538e | 3053 | struct mountpoint *mp; |
25e195aa | 3054 | struct mount *mnt; |
19a167af | 3055 | int err; |
25e195aa AV |
3056 | |
3057 | if (!m) | |
3058 | return 0; | |
3059 | if (IS_ERR(m)) | |
3060 | return PTR_ERR(m); | |
3061 | ||
3062 | mnt = real_mount(m); | |
19a167af AV |
3063 | /* The new mount record should have at least 2 refs to prevent it being |
3064 | * expired before we get a chance to add it | |
3065 | */ | |
6776db3d | 3066 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
3067 | |
3068 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 3069 | m->mnt_root == dentry) { |
b1e75df4 | 3070 | err = -ELOOP; |
26df6034 | 3071 | goto discard; |
19a167af AV |
3072 | } |
3073 | ||
26df6034 AV |
3074 | /* |
3075 | * we don't want to use lock_mount() - in this case finding something | |
3076 | * that overmounts our mountpoint to be means "quitely drop what we've | |
3077 | * got", not "try to mount it on top". | |
3078 | */ | |
3079 | inode_lock(dentry->d_inode); | |
3080 | namespace_lock(); | |
3081 | if (unlikely(cant_mount(dentry))) { | |
3082 | err = -ENOENT; | |
3083 | goto discard_locked; | |
3084 | } | |
3085 | rcu_read_lock(); | |
3086 | if (unlikely(__lookup_mnt(path->mnt, dentry))) { | |
3087 | rcu_read_unlock(); | |
3088 | err = 0; | |
3089 | goto discard_locked; | |
3090 | } | |
3091 | rcu_read_unlock(); | |
3092 | mp = get_mountpoint(dentry); | |
8f11538e AV |
3093 | if (IS_ERR(mp)) { |
3094 | err = PTR_ERR(mp); | |
26df6034 | 3095 | goto discard_locked; |
8f11538e | 3096 | } |
26df6034 | 3097 | |
8f11538e AV |
3098 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
3099 | unlock_mount(mp); | |
26df6034 AV |
3100 | if (unlikely(err)) |
3101 | goto discard; | |
3102 | mntput(m); | |
3103 | return 0; | |
3104 | ||
3105 | discard_locked: | |
3106 | namespace_unlock(); | |
3107 | inode_unlock(dentry->d_inode); | |
3108 | discard: | |
b1e75df4 | 3109 | /* remove m from any expiration list it may be on */ |
6776db3d | 3110 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 3111 | namespace_lock(); |
6776db3d | 3112 | list_del_init(&mnt->mnt_expire); |
97216be0 | 3113 | namespace_unlock(); |
19a167af | 3114 | } |
b1e75df4 AV |
3115 | mntput(m); |
3116 | mntput(m); | |
19a167af AV |
3117 | return err; |
3118 | } | |
3119 | ||
ea5b778a DH |
3120 | /** |
3121 | * mnt_set_expiry - Put a mount on an expiration list | |
3122 | * @mnt: The mount to list. | |
3123 | * @expiry_list: The list to add the mount to. | |
3124 | */ | |
3125 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
3126 | { | |
97216be0 | 3127 | namespace_lock(); |
ea5b778a | 3128 | |
6776db3d | 3129 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 3130 | |
97216be0 | 3131 | namespace_unlock(); |
ea5b778a DH |
3132 | } |
3133 | EXPORT_SYMBOL(mnt_set_expiry); | |
3134 | ||
1da177e4 LT |
3135 | /* |
3136 | * process a list of expirable mountpoints with the intent of discarding any | |
3137 | * mountpoints that aren't in use and haven't been touched since last we came | |
3138 | * here | |
3139 | */ | |
3140 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3141 | { | |
761d5c38 | 3142 | struct mount *mnt, *next; |
1da177e4 LT |
3143 | LIST_HEAD(graveyard); |
3144 | ||
3145 | if (list_empty(mounts)) | |
3146 | return; | |
3147 | ||
97216be0 | 3148 | namespace_lock(); |
719ea2fb | 3149 | lock_mount_hash(); |
1da177e4 LT |
3150 | |
3151 | /* extract from the expiration list every vfsmount that matches the | |
3152 | * following criteria: | |
3153 | * - only referenced by its parent vfsmount | |
3154 | * - still marked for expiry (marked on the last call here; marks are | |
3155 | * cleared by mntput()) | |
3156 | */ | |
6776db3d | 3157 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3158 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3159 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3160 | continue; |
6776db3d | 3161 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3162 | } |
bcc5c7d2 | 3163 | while (!list_empty(&graveyard)) { |
6776db3d | 3164 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3165 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3166 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3167 | } |
719ea2fb | 3168 | unlock_mount_hash(); |
3ab6abee | 3169 | namespace_unlock(); |
5528f911 TM |
3170 | } |
3171 | ||
3172 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3173 | ||
3174 | /* | |
3175 | * Ripoff of 'select_parent()' | |
3176 | * | |
3177 | * search the list of submounts for a given mountpoint, and move any | |
3178 | * shrinkable submounts to the 'graveyard' list. | |
3179 | */ | |
692afc31 | 3180 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3181 | { |
692afc31 | 3182 | struct mount *this_parent = parent; |
5528f911 TM |
3183 | struct list_head *next; |
3184 | int found = 0; | |
3185 | ||
3186 | repeat: | |
6b41d536 | 3187 | next = this_parent->mnt_mounts.next; |
5528f911 | 3188 | resume: |
6b41d536 | 3189 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3190 | struct list_head *tmp = next; |
6b41d536 | 3191 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3192 | |
3193 | next = tmp->next; | |
692afc31 | 3194 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3195 | continue; |
5528f911 TM |
3196 | /* |
3197 | * Descend a level if the d_mounts list is non-empty. | |
3198 | */ | |
6b41d536 | 3199 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3200 | this_parent = mnt; |
3201 | goto repeat; | |
3202 | } | |
1da177e4 | 3203 | |
1ab59738 | 3204 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3205 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3206 | found++; |
3207 | } | |
1da177e4 | 3208 | } |
5528f911 TM |
3209 | /* |
3210 | * All done at this level ... ascend and resume the search | |
3211 | */ | |
3212 | if (this_parent != parent) { | |
6b41d536 | 3213 | next = this_parent->mnt_child.next; |
0714a533 | 3214 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3215 | goto resume; |
3216 | } | |
3217 | return found; | |
3218 | } | |
3219 | ||
3220 | /* | |
3221 | * process a list of expirable mountpoints with the intent of discarding any | |
3222 | * submounts of a specific parent mountpoint | |
99b7db7b | 3223 | * |
48a066e7 | 3224 | * mount_lock must be held for write |
5528f911 | 3225 | */ |
b54b9be7 | 3226 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3227 | { |
3228 | LIST_HEAD(graveyard); | |
761d5c38 | 3229 | struct mount *m; |
5528f911 | 3230 | |
5528f911 | 3231 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3232 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3233 | while (!list_empty(&graveyard)) { |
761d5c38 | 3234 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3235 | mnt_expire); |
143c8c91 | 3236 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3237 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3238 | } |
3239 | } | |
1da177e4 LT |
3240 | } |
3241 | ||
028abd92 | 3242 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3243 | { |
b40ef869 | 3244 | char *copy; |
d563d678 | 3245 | unsigned left, offset; |
b58fed8b | 3246 | |
1da177e4 | 3247 | if (!data) |
b40ef869 | 3248 | return NULL; |
1da177e4 | 3249 | |
b40ef869 AV |
3250 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3251 | if (!copy) | |
3252 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3253 | |
d563d678 | 3254 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3255 | |
d563d678 CM |
3256 | /* |
3257 | * Not all architectures have an exact copy_from_user(). Resort to | |
3258 | * byte at a time. | |
3259 | */ | |
3260 | offset = PAGE_SIZE - left; | |
3261 | while (left) { | |
3262 | char c; | |
3263 | if (get_user(c, (const char __user *)data + offset)) | |
3264 | break; | |
3265 | copy[offset] = c; | |
3266 | left--; | |
3267 | offset++; | |
3268 | } | |
3269 | ||
3270 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3271 | kfree(copy); |
3272 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3273 | } |
d563d678 | 3274 | |
b40ef869 | 3275 | return copy; |
1da177e4 LT |
3276 | } |
3277 | ||
028abd92 | 3278 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3279 | { |
fbdb4401 | 3280 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3281 | } |
3282 | ||
1da177e4 LT |
3283 | /* |
3284 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3285 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3286 | * | |
3287 | * data is a (void *) that can point to any structure up to | |
3288 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3289 | * information (or be NULL). | |
3290 | * | |
3291 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3292 | * When the flags word was introduced its top half was required | |
3293 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3294 | * Therefore, if this magic number is present, it carries no information | |
3295 | * and must be discarded. | |
3296 | */ | |
c60166f0 | 3297 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3298 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3299 | { |
e462ec50 | 3300 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3301 | int ret; |
1da177e4 LT |
3302 | |
3303 | /* Discard magic */ | |
3304 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3305 | flags &= ~MS_MGC_MSK; | |
3306 | ||
3307 | /* Basic sanity checks */ | |
1da177e4 LT |
3308 | if (data_page) |
3309 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3310 | ||
e462ec50 DH |
3311 | if (flags & MS_NOUSER) |
3312 | return -EINVAL; | |
3313 | ||
a1e6aaa3 CH |
3314 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3315 | if (ret) | |
3316 | return ret; | |
3317 | if (!may_mount()) | |
3318 | return -EPERM; | |
f7e33bdb JL |
3319 | if (flags & SB_MANDLOCK) |
3320 | warn_mandlock(); | |
a27ab9f2 | 3321 | |
613cbe3d AK |
3322 | /* Default to relatime unless overriden */ |
3323 | if (!(flags & MS_NOATIME)) | |
3324 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3325 | |
1da177e4 LT |
3326 | /* Separate the per-mountpoint flags */ |
3327 | if (flags & MS_NOSUID) | |
3328 | mnt_flags |= MNT_NOSUID; | |
3329 | if (flags & MS_NODEV) | |
3330 | mnt_flags |= MNT_NODEV; | |
3331 | if (flags & MS_NOEXEC) | |
3332 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3333 | if (flags & MS_NOATIME) |
3334 | mnt_flags |= MNT_NOATIME; | |
3335 | if (flags & MS_NODIRATIME) | |
3336 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3337 | if (flags & MS_STRICTATIME) |
3338 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3339 | if (flags & MS_RDONLY) |
2e4b7fcd | 3340 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3341 | if (flags & MS_NOSYMFOLLOW) |
3342 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3343 | |
ffbc6f0e EB |
3344 | /* The default atime for remount is preservation */ |
3345 | if ((flags & MS_REMOUNT) && | |
3346 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3347 | MS_STRICTATIME)) == 0)) { | |
3348 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3349 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3350 | } |
3351 | ||
e462ec50 DH |
3352 | sb_flags = flags & (SB_RDONLY | |
3353 | SB_SYNCHRONOUS | | |
3354 | SB_MANDLOCK | | |
3355 | SB_DIRSYNC | | |
3356 | SB_SILENT | | |
917086ff | 3357 | SB_POSIXACL | |
d7ee9469 | 3358 | SB_LAZYTIME | |
917086ff | 3359 | SB_I_VERSION); |
1da177e4 | 3360 | |
43f5e655 | 3361 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3362 | return do_reconfigure_mnt(path, mnt_flags); |
3363 | if (flags & MS_REMOUNT) | |
3364 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3365 | if (flags & MS_BIND) | |
3366 | return do_loopback(path, dev_name, flags & MS_REC); | |
3367 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3368 | return do_change_type(path, flags); | |
3369 | if (flags & MS_MOVE) | |
3370 | return do_move_mount_old(path, dev_name); | |
3371 | ||
3372 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3373 | data_page); | |
3374 | } | |
3375 | ||
3376 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3377 | const char *type_page, unsigned long flags, void *data_page) | |
3378 | { | |
3379 | struct path path; | |
3380 | int ret; | |
3381 | ||
3382 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3383 | if (ret) | |
3384 | return ret; | |
3385 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3386 | path_put(&path); |
a1e6aaa3 | 3387 | return ret; |
1da177e4 LT |
3388 | } |
3389 | ||
537f7ccb EB |
3390 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3391 | { | |
3392 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3393 | } | |
3394 | ||
3395 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3396 | { | |
3397 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3398 | } | |
3399 | ||
771b1371 EB |
3400 | static void free_mnt_ns(struct mnt_namespace *ns) |
3401 | { | |
74e83122 AV |
3402 | if (!is_anon_ns(ns)) |
3403 | ns_free_inum(&ns->ns); | |
537f7ccb | 3404 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3405 | put_user_ns(ns->user_ns); |
3406 | kfree(ns); | |
3407 | } | |
3408 | ||
8823c079 EB |
3409 | /* |
3410 | * Assign a sequence number so we can detect when we attempt to bind | |
3411 | * mount a reference to an older mount namespace into the current | |
3412 | * mount namespace, preventing reference counting loops. A 64bit | |
3413 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3414 | * is effectively never, so we can ignore the possibility. | |
3415 | */ | |
3416 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3417 | ||
74e83122 | 3418 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3419 | { |
3420 | struct mnt_namespace *new_ns; | |
537f7ccb | 3421 | struct ucounts *ucounts; |
98f842e6 | 3422 | int ret; |
cf8d2c11 | 3423 | |
537f7ccb EB |
3424 | ucounts = inc_mnt_namespaces(user_ns); |
3425 | if (!ucounts) | |
df75e774 | 3426 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3427 | |
30acd0bd | 3428 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); |
537f7ccb EB |
3429 | if (!new_ns) { |
3430 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3431 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3432 | } |
74e83122 AV |
3433 | if (!anon) { |
3434 | ret = ns_alloc_inum(&new_ns->ns); | |
3435 | if (ret) { | |
3436 | kfree(new_ns); | |
3437 | dec_mnt_namespaces(ucounts); | |
3438 | return ERR_PTR(ret); | |
3439 | } | |
98f842e6 | 3440 | } |
33c42940 | 3441 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3442 | if (!anon) |
3443 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3444 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3445 | INIT_LIST_HEAD(&new_ns->list); |
3446 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3447 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3448 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3449 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3450 | return new_ns; |
3451 | } | |
3452 | ||
0766f788 | 3453 | __latent_entropy |
9559f689 AV |
3454 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3455 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3456 | { |
6b3286ed | 3457 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3458 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3459 | struct mount *p, *q; |
9559f689 | 3460 | struct mount *old; |
cb338d06 | 3461 | struct mount *new; |
7a472ef4 | 3462 | int copy_flags; |
1da177e4 | 3463 | |
9559f689 AV |
3464 | BUG_ON(!ns); |
3465 | ||
3466 | if (likely(!(flags & CLONE_NEWNS))) { | |
3467 | get_mnt_ns(ns); | |
3468 | return ns; | |
3469 | } | |
3470 | ||
3471 | old = ns->root; | |
3472 | ||
74e83122 | 3473 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3474 | if (IS_ERR(new_ns)) |
3475 | return new_ns; | |
1da177e4 | 3476 | |
97216be0 | 3477 | namespace_lock(); |
1da177e4 | 3478 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3479 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3480 | if (user_ns != ns->user_ns) |
3bd045cc | 3481 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3482 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3483 | if (IS_ERR(new)) { |
328e6d90 | 3484 | namespace_unlock(); |
771b1371 | 3485 | free_mnt_ns(new_ns); |
be34d1a3 | 3486 | return ERR_CAST(new); |
1da177e4 | 3487 | } |
3bd045cc AV |
3488 | if (user_ns != ns->user_ns) { |
3489 | lock_mount_hash(); | |
3490 | lock_mnt_tree(new); | |
3491 | unlock_mount_hash(); | |
3492 | } | |
be08d6d2 | 3493 | new_ns->root = new; |
1a4eeaf2 | 3494 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3495 | |
3496 | /* | |
3497 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3498 | * as belonging to new namespace. We have already acquired a private | |
3499 | * fs_struct, so tsk->fs->lock is not needed. | |
3500 | */ | |
909b0a88 | 3501 | p = old; |
cb338d06 | 3502 | q = new; |
1da177e4 | 3503 | while (p) { |
143c8c91 | 3504 | q->mnt_ns = new_ns; |
d2921684 | 3505 | new_ns->mounts++; |
9559f689 AV |
3506 | if (new_fs) { |
3507 | if (&p->mnt == new_fs->root.mnt) { | |
3508 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3509 | rootmnt = &p->mnt; |
1da177e4 | 3510 | } |
9559f689 AV |
3511 | if (&p->mnt == new_fs->pwd.mnt) { |
3512 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3513 | pwdmnt = &p->mnt; |
1da177e4 | 3514 | } |
1da177e4 | 3515 | } |
909b0a88 AV |
3516 | p = next_mnt(p, old); |
3517 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3518 | if (!q) |
3519 | break; | |
61d8e426 | 3520 | // an mntns binding we'd skipped? |
4ce5d2b1 | 3521 | while (p->mnt.mnt_root != q->mnt.mnt_root) |
61d8e426 | 3522 | p = next_mnt(skip_mnt_tree(p), old); |
1da177e4 | 3523 | } |
328e6d90 | 3524 | namespace_unlock(); |
1da177e4 | 3525 | |
1da177e4 | 3526 | if (rootmnt) |
f03c6599 | 3527 | mntput(rootmnt); |
1da177e4 | 3528 | if (pwdmnt) |
f03c6599 | 3529 | mntput(pwdmnt); |
1da177e4 | 3530 | |
741a2951 | 3531 | return new_ns; |
1da177e4 LT |
3532 | } |
3533 | ||
74e83122 | 3534 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3535 | { |
74e83122 | 3536 | struct mount *mnt = real_mount(m); |
ea441d11 | 3537 | struct mnt_namespace *ns; |
d31da0f0 | 3538 | struct super_block *s; |
ea441d11 AV |
3539 | struct path path; |
3540 | int err; | |
3541 | ||
74e83122 AV |
3542 | ns = alloc_mnt_ns(&init_user_ns, true); |
3543 | if (IS_ERR(ns)) { | |
3544 | mntput(m); | |
ea441d11 | 3545 | return ERR_CAST(ns); |
74e83122 AV |
3546 | } |
3547 | mnt->mnt_ns = ns; | |
3548 | ns->root = mnt; | |
3549 | ns->mounts++; | |
3550 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3551 | |
74e83122 | 3552 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3553 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3554 | ||
3555 | put_mnt_ns(ns); | |
3556 | ||
3557 | if (err) | |
3558 | return ERR_PTR(err); | |
3559 | ||
3560 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3561 | s = path.mnt->mnt_sb; |
3562 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3563 | mntput(path.mnt); |
3564 | /* lock the sucker */ | |
d31da0f0 | 3565 | down_write(&s->s_umount); |
ea441d11 AV |
3566 | /* ... and return the root of (sub)tree on it */ |
3567 | return path.dentry; | |
3568 | } | |
3569 | EXPORT_SYMBOL(mount_subtree); | |
3570 | ||
cccaa5e3 DB |
3571 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3572 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3573 | { |
eca6f534 VN |
3574 | int ret; |
3575 | char *kernel_type; | |
eca6f534 | 3576 | char *kernel_dev; |
b40ef869 | 3577 | void *options; |
1da177e4 | 3578 | |
b8850d1f TG |
3579 | kernel_type = copy_mount_string(type); |
3580 | ret = PTR_ERR(kernel_type); | |
3581 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3582 | goto out_type; |
1da177e4 | 3583 | |
b8850d1f TG |
3584 | kernel_dev = copy_mount_string(dev_name); |
3585 | ret = PTR_ERR(kernel_dev); | |
3586 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3587 | goto out_dev; |
1da177e4 | 3588 | |
b40ef869 AV |
3589 | options = copy_mount_options(data); |
3590 | ret = PTR_ERR(options); | |
3591 | if (IS_ERR(options)) | |
eca6f534 | 3592 | goto out_data; |
1da177e4 | 3593 | |
b40ef869 | 3594 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3595 | |
b40ef869 | 3596 | kfree(options); |
eca6f534 VN |
3597 | out_data: |
3598 | kfree(kernel_dev); | |
3599 | out_dev: | |
eca6f534 VN |
3600 | kfree(kernel_type); |
3601 | out_type: | |
3602 | return ret; | |
1da177e4 LT |
3603 | } |
3604 | ||
dd8b477f CB |
3605 | #define FSMOUNT_VALID_FLAGS \ |
3606 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3607 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ | |
3608 | MOUNT_ATTR_NOSYMFOLLOW) | |
5b490500 | 3609 | |
9caccd41 | 3610 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
2a186721 CB |
3611 | |
3612 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3613 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3614 | ||
5b490500 CB |
3615 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3616 | { | |
3617 | unsigned int mnt_flags = 0; | |
3618 | ||
3619 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3620 | mnt_flags |= MNT_READONLY; | |
3621 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3622 | mnt_flags |= MNT_NOSUID; | |
3623 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3624 | mnt_flags |= MNT_NODEV; | |
3625 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3626 | mnt_flags |= MNT_NOEXEC; | |
3627 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3628 | mnt_flags |= MNT_NODIRATIME; | |
dd8b477f CB |
3629 | if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) |
3630 | mnt_flags |= MNT_NOSYMFOLLOW; | |
5b490500 CB |
3631 | |
3632 | return mnt_flags; | |
3633 | } | |
3634 | ||
2db154b3 | 3635 | /* |
93766fbd DH |
3636 | * Create a kernel mount representation for a new, prepared superblock |
3637 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3638 | */ | |
3639 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3640 | unsigned int, attr_flags) | |
3641 | { | |
3642 | struct mnt_namespace *ns; | |
3643 | struct fs_context *fc; | |
3644 | struct file *file; | |
3645 | struct path newmount; | |
3646 | struct mount *mnt; | |
3647 | struct fd f; | |
3648 | unsigned int mnt_flags = 0; | |
3649 | long ret; | |
3650 | ||
3651 | if (!may_mount()) | |
3652 | return -EPERM; | |
3653 | ||
3654 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3655 | return -EINVAL; | |
3656 | ||
5b490500 | 3657 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3658 | return -EINVAL; |
3659 | ||
5b490500 | 3660 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3661 | |
3662 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3663 | case MOUNT_ATTR_STRICTATIME: | |
3664 | break; | |
3665 | case MOUNT_ATTR_NOATIME: | |
3666 | mnt_flags |= MNT_NOATIME; | |
3667 | break; | |
3668 | case MOUNT_ATTR_RELATIME: | |
3669 | mnt_flags |= MNT_RELATIME; | |
3670 | break; | |
3671 | default: | |
3672 | return -EINVAL; | |
3673 | } | |
3674 | ||
3675 | f = fdget(fs_fd); | |
3676 | if (!f.file) | |
3677 | return -EBADF; | |
3678 | ||
3679 | ret = -EINVAL; | |
3680 | if (f.file->f_op != &fscontext_fops) | |
3681 | goto err_fsfd; | |
3682 | ||
3683 | fc = f.file->private_data; | |
3684 | ||
3685 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3686 | if (ret < 0) | |
3687 | goto err_fsfd; | |
3688 | ||
3689 | /* There must be a valid superblock or we can't mount it */ | |
3690 | ret = -EINVAL; | |
3691 | if (!fc->root) | |
3692 | goto err_unlock; | |
3693 | ||
3694 | ret = -EPERM; | |
3695 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3696 | pr_warn("VFS: Mount too revealing\n"); | |
3697 | goto err_unlock; | |
3698 | } | |
3699 | ||
3700 | ret = -EBUSY; | |
3701 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3702 | goto err_unlock; | |
3703 | ||
f7e33bdb JL |
3704 | if (fc->sb_flags & SB_MANDLOCK) |
3705 | warn_mandlock(); | |
93766fbd DH |
3706 | |
3707 | newmount.mnt = vfs_create_mount(fc); | |
3708 | if (IS_ERR(newmount.mnt)) { | |
3709 | ret = PTR_ERR(newmount.mnt); | |
3710 | goto err_unlock; | |
3711 | } | |
3712 | newmount.dentry = dget(fc->root); | |
3713 | newmount.mnt->mnt_flags = mnt_flags; | |
3714 | ||
3715 | /* We've done the mount bit - now move the file context into more or | |
3716 | * less the same state as if we'd done an fspick(). We don't want to | |
3717 | * do any memory allocation or anything like that at this point as we | |
3718 | * don't want to have to handle any errors incurred. | |
3719 | */ | |
3720 | vfs_clean_context(fc); | |
3721 | ||
3722 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
3723 | if (IS_ERR(ns)) { | |
3724 | ret = PTR_ERR(ns); | |
3725 | goto err_path; | |
3726 | } | |
3727 | mnt = real_mount(newmount.mnt); | |
3728 | mnt->mnt_ns = ns; | |
3729 | ns->root = mnt; | |
3730 | ns->mounts = 1; | |
3731 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 3732 | mntget(newmount.mnt); |
93766fbd DH |
3733 | |
3734 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
3735 | * it, not just simply put it. | |
3736 | */ | |
3737 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
3738 | if (IS_ERR(file)) { | |
3739 | dissolve_on_fput(newmount.mnt); | |
3740 | ret = PTR_ERR(file); | |
3741 | goto err_path; | |
3742 | } | |
3743 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
3744 | ||
3745 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
3746 | if (ret >= 0) | |
3747 | fd_install(ret, file); | |
3748 | else | |
3749 | fput(file); | |
3750 | ||
3751 | err_path: | |
3752 | path_put(&newmount); | |
3753 | err_unlock: | |
3754 | mutex_unlock(&fc->uapi_mutex); | |
3755 | err_fsfd: | |
3756 | fdput(f); | |
3757 | return ret; | |
3758 | } | |
3759 | ||
3760 | /* | |
3761 | * Move a mount from one place to another. In combination with | |
3762 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
3763 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
3764 | * a mount subtree. | |
2db154b3 DH |
3765 | * |
3766 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
3767 | */ | |
3768 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
3769 | int, from_dfd, const char __user *, from_pathname, |
3770 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
3771 | unsigned int, flags) |
3772 | { | |
3773 | struct path from_path, to_path; | |
3774 | unsigned int lflags; | |
3775 | int ret = 0; | |
3776 | ||
3777 | if (!may_mount()) | |
3778 | return -EPERM; | |
3779 | ||
3780 | if (flags & ~MOVE_MOUNT__MASK) | |
3781 | return -EINVAL; | |
3782 | ||
3783 | /* If someone gives a pathname, they aren't permitted to move | |
3784 | * from an fd that requires unmount as we can't get at the flag | |
3785 | * to clear it afterwards. | |
3786 | */ | |
3787 | lflags = 0; | |
3788 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3789 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3790 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3791 | ||
3792 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
3793 | if (ret < 0) | |
3794 | return ret; | |
3795 | ||
3796 | lflags = 0; | |
3797 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
3798 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
3799 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
3800 | ||
3801 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
3802 | if (ret < 0) | |
3803 | goto out_from; | |
3804 | ||
3805 | ret = security_move_mount(&from_path, &to_path); | |
3806 | if (ret < 0) | |
3807 | goto out_to; | |
3808 | ||
9ffb14ef PT |
3809 | if (flags & MOVE_MOUNT_SET_GROUP) |
3810 | ret = do_set_group(&from_path, &to_path); | |
3811 | else | |
3812 | ret = do_move_mount(&from_path, &to_path); | |
2db154b3 DH |
3813 | |
3814 | out_to: | |
3815 | path_put(&to_path); | |
3816 | out_from: | |
3817 | path_put(&from_path); | |
3818 | return ret; | |
3819 | } | |
3820 | ||
afac7cba AV |
3821 | /* |
3822 | * Return true if path is reachable from root | |
3823 | * | |
48a066e7 | 3824 | * namespace_sem or mount_lock is held |
afac7cba | 3825 | */ |
643822b4 | 3826 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
3827 | const struct path *root) |
3828 | { | |
643822b4 | 3829 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 3830 | dentry = mnt->mnt_mountpoint; |
0714a533 | 3831 | mnt = mnt->mnt_parent; |
afac7cba | 3832 | } |
643822b4 | 3833 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
3834 | } |
3835 | ||
640eb7e7 | 3836 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 3837 | { |
25ab4c9b | 3838 | bool res; |
48a066e7 | 3839 | read_seqlock_excl(&mount_lock); |
643822b4 | 3840 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 3841 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
3842 | return res; |
3843 | } | |
3844 | EXPORT_SYMBOL(path_is_under); | |
3845 | ||
1da177e4 LT |
3846 | /* |
3847 | * pivot_root Semantics: | |
3848 | * Moves the root file system of the current process to the directory put_old, | |
3849 | * makes new_root as the new root file system of the current process, and sets | |
3850 | * root/cwd of all processes which had them on the current root to new_root. | |
3851 | * | |
3852 | * Restrictions: | |
3853 | * The new_root and put_old must be directories, and must not be on the | |
3854 | * same file system as the current process root. The put_old must be | |
3855 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
3856 | * pointed to by put_old must yield the same directory as new_root. No other | |
3857 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
3858 | * | |
4a0d11fa | 3859 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 3860 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
3861 | * in this situation. |
3862 | * | |
1da177e4 LT |
3863 | * Notes: |
3864 | * - we don't move root/cwd if they are not at the root (reason: if something | |
3865 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
3866 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
3867 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
3868 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
3869 | * first. | |
3870 | */ | |
3480b257 HC |
3871 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
3872 | const char __user *, put_old) | |
1da177e4 | 3873 | { |
2763d119 AV |
3874 | struct path new, old, root; |
3875 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 3876 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
3877 | int error; |
3878 | ||
9b40bc90 | 3879 | if (!may_mount()) |
1da177e4 LT |
3880 | return -EPERM; |
3881 | ||
ce6595a2 AV |
3882 | error = user_path_at(AT_FDCWD, new_root, |
3883 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
3884 | if (error) |
3885 | goto out0; | |
1da177e4 | 3886 | |
ce6595a2 AV |
3887 | error = user_path_at(AT_FDCWD, put_old, |
3888 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
3889 | if (error) |
3890 | goto out1; | |
3891 | ||
2d8f3038 | 3892 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
3893 | if (error) |
3894 | goto out2; | |
1da177e4 | 3895 | |
f7ad3c6b | 3896 | get_fs_root(current->fs, &root); |
84d17192 AV |
3897 | old_mp = lock_mount(&old); |
3898 | error = PTR_ERR(old_mp); | |
3899 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
3900 | goto out3; |
3901 | ||
1da177e4 | 3902 | error = -EINVAL; |
419148da AV |
3903 | new_mnt = real_mount(new.mnt); |
3904 | root_mnt = real_mount(root.mnt); | |
84d17192 | 3905 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
3906 | ex_parent = new_mnt->mnt_parent; |
3907 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 3908 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
3909 | IS_MNT_SHARED(ex_parent) || |
3910 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 3911 | goto out4; |
143c8c91 | 3912 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 3913 | goto out4; |
5ff9d8a6 EB |
3914 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
3915 | goto out4; | |
1da177e4 | 3916 | error = -ENOENT; |
f3da392e | 3917 | if (d_unlinked(new.dentry)) |
b12cea91 | 3918 | goto out4; |
1da177e4 | 3919 | error = -EBUSY; |
84d17192 | 3920 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 3921 | goto out4; /* loop, on the same file system */ |
1da177e4 | 3922 | error = -EINVAL; |
8c3ee42e | 3923 | if (root.mnt->mnt_root != root.dentry) |
b12cea91 | 3924 | goto out4; /* not a mountpoint */ |
676da58d | 3925 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 3926 | goto out4; /* not attached */ |
2d8f3038 | 3927 | if (new.mnt->mnt_root != new.dentry) |
b12cea91 | 3928 | goto out4; /* not a mountpoint */ |
676da58d | 3929 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 3930 | goto out4; /* not attached */ |
4ac91378 | 3931 | /* make sure we can reach put_old from new_root */ |
84d17192 | 3932 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 3933 | goto out4; |
0d082601 EB |
3934 | /* make certain new is below the root */ |
3935 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
3936 | goto out4; | |
719ea2fb | 3937 | lock_mount_hash(); |
2763d119 AV |
3938 | umount_mnt(new_mnt); |
3939 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
3940 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
3941 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
3942 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
3943 | } | |
4ac91378 | 3944 | /* mount old root on put_old */ |
84d17192 | 3945 | attach_mnt(root_mnt, old_mnt, old_mp); |
4ac91378 | 3946 | /* mount new_root on / */ |
2763d119 AV |
3947 | attach_mnt(new_mnt, root_parent, root_mp); |
3948 | mnt_add_count(root_parent, -1); | |
6b3286ed | 3949 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
3950 | /* A moved mount should not expire automatically */ |
3951 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 3952 | put_mountpoint(root_mp); |
719ea2fb | 3953 | unlock_mount_hash(); |
2d8f3038 | 3954 | chroot_fs_refs(&root, &new); |
1da177e4 | 3955 | error = 0; |
b12cea91 | 3956 | out4: |
84d17192 | 3957 | unlock_mount(old_mp); |
2763d119 AV |
3958 | if (!error) |
3959 | mntput_no_expire(ex_parent); | |
b12cea91 | 3960 | out3: |
8c3ee42e | 3961 | path_put(&root); |
b12cea91 | 3962 | out2: |
2d8f3038 | 3963 | path_put(&old); |
1da177e4 | 3964 | out1: |
2d8f3038 | 3965 | path_put(&new); |
1da177e4 | 3966 | out0: |
1da177e4 | 3967 | return error; |
1da177e4 LT |
3968 | } |
3969 | ||
2a186721 CB |
3970 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
3971 | { | |
3972 | unsigned int flags = mnt->mnt.mnt_flags; | |
3973 | ||
3974 | /* flags to clear */ | |
3975 | flags &= ~kattr->attr_clr; | |
3976 | /* flags to raise */ | |
3977 | flags |= kattr->attr_set; | |
3978 | ||
3979 | return flags; | |
3980 | } | |
3981 | ||
9caccd41 CB |
3982 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
3983 | { | |
3984 | struct vfsmount *m = &mnt->mnt; | |
bd303368 | 3985 | struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; |
9caccd41 | 3986 | |
256c8aed | 3987 | if (!kattr->mnt_idmap) |
9caccd41 CB |
3988 | return 0; |
3989 | ||
bd303368 CB |
3990 | /* |
3991 | * Creating an idmapped mount with the filesystem wide idmapping | |
3992 | * doesn't make sense so block that. We don't allow mushy semantics. | |
3993 | */ | |
3707d84c | 3994 | if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb)) |
bd303368 CB |
3995 | return -EINVAL; |
3996 | ||
9caccd41 CB |
3997 | /* |
3998 | * Once a mount has been idmapped we don't allow it to change its | |
3999 | * mapping. It makes things simpler and callers can just create | |
4000 | * another bind-mount they can idmap if they want to. | |
4001 | */ | |
bb49e9e7 | 4002 | if (is_idmapped_mnt(m)) |
9caccd41 CB |
4003 | return -EPERM; |
4004 | ||
4005 | /* The underlying filesystem doesn't support idmapped mounts yet. */ | |
4006 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) | |
4007 | return -EINVAL; | |
4008 | ||
4009 | /* We're not controlling the superblock. */ | |
bd303368 | 4010 | if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) |
9caccd41 CB |
4011 | return -EPERM; |
4012 | ||
4013 | /* Mount has already been visible in the filesystem hierarchy. */ | |
4014 | if (!is_anon_ns(mnt->mnt_ns)) | |
4015 | return -EINVAL; | |
4016 | ||
4017 | return 0; | |
4018 | } | |
4019 | ||
a26f788b CB |
4020 | /** |
4021 | * mnt_allow_writers() - check whether the attribute change allows writers | |
4022 | * @kattr: the new mount attributes | |
4023 | * @mnt: the mount to which @kattr will be applied | |
4024 | * | |
4025 | * Check whether thew new mount attributes in @kattr allow concurrent writers. | |
4026 | * | |
4027 | * Return: true if writers need to be held, false if not | |
4028 | */ | |
4029 | static inline bool mnt_allow_writers(const struct mount_kattr *kattr, | |
4030 | const struct mount *mnt) | |
2a186721 | 4031 | { |
e1bbcd27 CB |
4032 | return (!(kattr->attr_set & MNT_READONLY) || |
4033 | (mnt->mnt.mnt_flags & MNT_READONLY)) && | |
256c8aed | 4034 | !kattr->mnt_idmap; |
a26f788b | 4035 | } |
2a186721 | 4036 | |
87bb5b60 | 4037 | static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) |
2a186721 | 4038 | { |
e257039f AV |
4039 | struct mount *m; |
4040 | int err; | |
2a186721 | 4041 | |
e257039f AV |
4042 | for (m = mnt; m; m = next_mnt(m, mnt)) { |
4043 | if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { | |
4044 | err = -EPERM; | |
4045 | break; | |
4046 | } | |
2a186721 | 4047 | |
87bb5b60 CB |
4048 | err = can_idmap_mount(kattr, m); |
4049 | if (err) | |
e257039f | 4050 | break; |
2a186721 | 4051 | |
e257039f AV |
4052 | if (!mnt_allow_writers(kattr, m)) { |
4053 | err = mnt_hold_writers(m); | |
4054 | if (err) | |
4055 | break; | |
2a186721 CB |
4056 | } |
4057 | ||
e257039f AV |
4058 | if (!kattr->recurse) |
4059 | return 0; | |
4060 | } | |
9caccd41 | 4061 | |
e257039f AV |
4062 | if (err) { |
4063 | struct mount *p; | |
2a186721 | 4064 | |
0014edae CB |
4065 | /* |
4066 | * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will | |
4067 | * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all | |
4068 | * mounts and needs to take care to include the first mount. | |
4069 | */ | |
4070 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
e257039f AV |
4071 | /* If we had to hold writers unblock them. */ |
4072 | if (p->mnt.mnt_flags & MNT_WRITE_HOLD) | |
4073 | mnt_unhold_writers(p); | |
0014edae CB |
4074 | |
4075 | /* | |
4076 | * We're done once the first mount we changed got | |
4077 | * MNT_WRITE_HOLD unset. | |
4078 | */ | |
4079 | if (p == m) | |
4080 | break; | |
2a186721 | 4081 | } |
e257039f AV |
4082 | } |
4083 | return err; | |
2a186721 CB |
4084 | } |
4085 | ||
9caccd41 CB |
4086 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
4087 | { | |
256c8aed | 4088 | if (!kattr->mnt_idmap) |
9caccd41 CB |
4089 | return; |
4090 | ||
bd303368 | 4091 | /* |
256c8aed CB |
4092 | * Pairs with smp_load_acquire() in mnt_idmap(). |
4093 | * | |
4094 | * Since we only allow a mount to change the idmapping once and | |
4095 | * verified this in can_idmap_mount() we know that the mount has | |
4096 | * @nop_mnt_idmap attached to it. So there's no need to drop any | |
4097 | * references. | |
bd303368 | 4098 | */ |
256c8aed | 4099 | smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap)); |
9caccd41 CB |
4100 | } |
4101 | ||
e257039f | 4102 | static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) |
2a186721 | 4103 | { |
e257039f | 4104 | struct mount *m; |
2a186721 | 4105 | |
e257039f AV |
4106 | for (m = mnt; m; m = next_mnt(m, mnt)) { |
4107 | unsigned int flags; | |
2a186721 | 4108 | |
e257039f AV |
4109 | do_idmap_mount(kattr, m); |
4110 | flags = recalc_flags(kattr, m); | |
4111 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
2a186721 | 4112 | |
03b6abee CB |
4113 | /* If we had to hold writers unblock them. */ |
4114 | if (m->mnt.mnt_flags & MNT_WRITE_HOLD) | |
2a186721 CB |
4115 | mnt_unhold_writers(m); |
4116 | ||
e257039f | 4117 | if (kattr->propagation) |
2a186721 | 4118 | change_mnt_propagation(m, kattr->propagation); |
e257039f | 4119 | if (!kattr->recurse) |
2a186721 | 4120 | break; |
e257039f AV |
4121 | } |
4122 | touch_mnt_namespace(mnt->mnt_ns); | |
2a186721 CB |
4123 | } |
4124 | ||
4125 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
4126 | { | |
87bb5b60 | 4127 | struct mount *mnt = real_mount(path->mnt); |
2a186721 CB |
4128 | int err = 0; |
4129 | ||
4130 | if (path->dentry != mnt->mnt.mnt_root) | |
4131 | return -EINVAL; | |
4132 | ||
256c8aed CB |
4133 | if (kattr->mnt_userns) { |
4134 | struct mnt_idmap *mnt_idmap; | |
4135 | ||
4136 | mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns); | |
4137 | if (IS_ERR(mnt_idmap)) | |
4138 | return PTR_ERR(mnt_idmap); | |
4139 | kattr->mnt_idmap = mnt_idmap; | |
4140 | } | |
4141 | ||
2a186721 CB |
4142 | if (kattr->propagation) { |
4143 | /* | |
4144 | * Only take namespace_lock() if we're actually changing | |
4145 | * propagation. | |
4146 | */ | |
4147 | namespace_lock(); | |
4148 | if (kattr->propagation == MS_SHARED) { | |
4149 | err = invent_group_ids(mnt, kattr->recurse); | |
4150 | if (err) { | |
4151 | namespace_unlock(); | |
4152 | return err; | |
4153 | } | |
4154 | } | |
4155 | } | |
4156 | ||
87bb5b60 | 4157 | err = -EINVAL; |
2a186721 CB |
4158 | lock_mount_hash(); |
4159 | ||
87bb5b60 CB |
4160 | /* Ensure that this isn't anything purely vfs internal. */ |
4161 | if (!is_mounted(&mnt->mnt)) | |
4162 | goto out; | |
4163 | ||
2a186721 | 4164 | /* |
87bb5b60 CB |
4165 | * If this is an attached mount make sure it's located in the callers |
4166 | * mount namespace. If it's not don't let the caller interact with it. | |
4167 | * If this is a detached mount make sure it has an anonymous mount | |
4168 | * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. | |
2a186721 | 4169 | */ |
87bb5b60 CB |
4170 | if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) |
4171 | goto out; | |
2a186721 | 4172 | |
87bb5b60 CB |
4173 | /* |
4174 | * First, we get the mount tree in a shape where we can change mount | |
4175 | * properties without failure. If we succeeded to do so we commit all | |
4176 | * changes and if we failed we clean up. | |
4177 | */ | |
4178 | err = mount_setattr_prepare(kattr, mnt); | |
e257039f AV |
4179 | if (!err) |
4180 | mount_setattr_commit(kattr, mnt); | |
2a186721 | 4181 | |
87bb5b60 | 4182 | out: |
2a186721 CB |
4183 | unlock_mount_hash(); |
4184 | ||
4185 | if (kattr->propagation) { | |
4186 | namespace_unlock(); | |
4187 | if (err) | |
4188 | cleanup_group_ids(mnt, NULL); | |
4189 | } | |
4190 | ||
4191 | return err; | |
4192 | } | |
4193 | ||
9caccd41 CB |
4194 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
4195 | struct mount_kattr *kattr, unsigned int flags) | |
4196 | { | |
4197 | int err = 0; | |
4198 | struct ns_common *ns; | |
4199 | struct user_namespace *mnt_userns; | |
4200 | struct file *file; | |
4201 | ||
4202 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) | |
4203 | return 0; | |
4204 | ||
4205 | /* | |
4206 | * We currently do not support clearing an idmapped mount. If this ever | |
4207 | * is a use-case we can revisit this but for now let's keep it simple | |
4208 | * and not allow it. | |
4209 | */ | |
4210 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) | |
4211 | return -EINVAL; | |
4212 | ||
4213 | if (attr->userns_fd > INT_MAX) | |
4214 | return -EINVAL; | |
4215 | ||
4216 | file = fget(attr->userns_fd); | |
4217 | if (!file) | |
4218 | return -EBADF; | |
4219 | ||
4220 | if (!proc_ns_file(file)) { | |
4221 | err = -EINVAL; | |
4222 | goto out_fput; | |
4223 | } | |
4224 | ||
4225 | ns = get_proc_ns(file_inode(file)); | |
4226 | if (ns->ops->type != CLONE_NEWUSER) { | |
4227 | err = -EINVAL; | |
4228 | goto out_fput; | |
4229 | } | |
4230 | ||
4231 | /* | |
bd303368 CB |
4232 | * The initial idmapping cannot be used to create an idmapped |
4233 | * mount. We use the initial idmapping as an indicator of a mount | |
4234 | * that is not idmapped. It can simply be passed into helpers that | |
4235 | * are aware of idmapped mounts as a convenient shortcut. A user | |
4236 | * can just create a dedicated identity mapping to achieve the same | |
4237 | * result. | |
9caccd41 CB |
4238 | */ |
4239 | mnt_userns = container_of(ns, struct user_namespace, ns); | |
3707d84c | 4240 | if (mnt_userns == &init_user_ns) { |
9caccd41 CB |
4241 | err = -EPERM; |
4242 | goto out_fput; | |
4243 | } | |
bf1ac16e SF |
4244 | |
4245 | /* We're not controlling the target namespace. */ | |
4246 | if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { | |
4247 | err = -EPERM; | |
4248 | goto out_fput; | |
4249 | } | |
4250 | ||
9caccd41 CB |
4251 | kattr->mnt_userns = get_user_ns(mnt_userns); |
4252 | ||
4253 | out_fput: | |
4254 | fput(file); | |
4255 | return err; | |
4256 | } | |
4257 | ||
4258 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, | |
2a186721 CB |
4259 | struct mount_kattr *kattr, unsigned int flags) |
4260 | { | |
4261 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
4262 | ||
4263 | if (flags & AT_NO_AUTOMOUNT) | |
4264 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
4265 | if (flags & AT_SYMLINK_NOFOLLOW) | |
4266 | lookup_flags &= ~LOOKUP_FOLLOW; | |
4267 | if (flags & AT_EMPTY_PATH) | |
4268 | lookup_flags |= LOOKUP_EMPTY; | |
4269 | ||
4270 | *kattr = (struct mount_kattr) { | |
4271 | .lookup_flags = lookup_flags, | |
4272 | .recurse = !!(flags & AT_RECURSIVE), | |
4273 | }; | |
4274 | ||
4275 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
4276 | return -EINVAL; | |
4277 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
4278 | return -EINVAL; | |
4279 | kattr->propagation = attr->propagation; | |
4280 | ||
4281 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
4282 | return -EINVAL; | |
4283 | ||
2a186721 CB |
4284 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
4285 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
4286 | ||
4287 | /* | |
4288 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4289 | * users wanting to transition to a different atime setting cannot | |
4290 | * simply specify the atime setting in @attr_set, but must also | |
4291 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4292 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4293 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4294 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4295 | */ | |
4296 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4297 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4298 | return -EINVAL; | |
4299 | ||
4300 | /* | |
4301 | * Clear all previous time settings as they are mutually | |
4302 | * exclusive. | |
4303 | */ | |
4304 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4305 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4306 | case MOUNT_ATTR_RELATIME: | |
4307 | kattr->attr_set |= MNT_RELATIME; | |
4308 | break; | |
4309 | case MOUNT_ATTR_NOATIME: | |
4310 | kattr->attr_set |= MNT_NOATIME; | |
4311 | break; | |
4312 | case MOUNT_ATTR_STRICTATIME: | |
4313 | break; | |
4314 | default: | |
4315 | return -EINVAL; | |
4316 | } | |
4317 | } else { | |
4318 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4319 | return -EINVAL; | |
4320 | } | |
4321 | ||
9caccd41 CB |
4322 | return build_mount_idmapped(attr, usize, kattr, flags); |
4323 | } | |
4324 | ||
4325 | static void finish_mount_kattr(struct mount_kattr *kattr) | |
4326 | { | |
4327 | put_user_ns(kattr->mnt_userns); | |
4328 | kattr->mnt_userns = NULL; | |
256c8aed CB |
4329 | |
4330 | if (kattr->mnt_idmap) | |
4331 | mnt_idmap_put(kattr->mnt_idmap); | |
2a186721 CB |
4332 | } |
4333 | ||
4334 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4335 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4336 | size_t, usize) | |
4337 | { | |
4338 | int err; | |
4339 | struct path target; | |
4340 | struct mount_attr attr; | |
4341 | struct mount_kattr kattr; | |
4342 | ||
4343 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4344 | ||
4345 | if (flags & ~(AT_EMPTY_PATH | | |
4346 | AT_RECURSIVE | | |
4347 | AT_SYMLINK_NOFOLLOW | | |
4348 | AT_NO_AUTOMOUNT)) | |
4349 | return -EINVAL; | |
4350 | ||
4351 | if (unlikely(usize > PAGE_SIZE)) | |
4352 | return -E2BIG; | |
4353 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4354 | return -EINVAL; | |
4355 | ||
4356 | if (!may_mount()) | |
4357 | return -EPERM; | |
4358 | ||
4359 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4360 | if (err) | |
4361 | return err; | |
4362 | ||
4363 | /* Don't bother walking through the mounts if this is a nop. */ | |
4364 | if (attr.attr_set == 0 && | |
4365 | attr.attr_clr == 0 && | |
4366 | attr.propagation == 0) | |
4367 | return 0; | |
4368 | ||
9caccd41 | 4369 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
2a186721 CB |
4370 | if (err) |
4371 | return err; | |
4372 | ||
4373 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
012e3322 CB |
4374 | if (!err) { |
4375 | err = do_mount_setattr(&target, &kattr); | |
4376 | path_put(&target); | |
4377 | } | |
9caccd41 | 4378 | finish_mount_kattr(&kattr); |
2a186721 CB |
4379 | return err; |
4380 | } | |
4381 | ||
1da177e4 LT |
4382 | static void __init init_mount_tree(void) |
4383 | { | |
4384 | struct vfsmount *mnt; | |
74e83122 | 4385 | struct mount *m; |
6b3286ed | 4386 | struct mnt_namespace *ns; |
ac748a09 | 4387 | struct path root; |
1da177e4 | 4388 | |
fd3e007f | 4389 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4390 | if (IS_ERR(mnt)) |
4391 | panic("Can't create rootfs"); | |
b3e19d92 | 4392 | |
74e83122 | 4393 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4394 | if (IS_ERR(ns)) |
1da177e4 | 4395 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4396 | m = real_mount(mnt); |
4397 | m->mnt_ns = ns; | |
4398 | ns->root = m; | |
4399 | ns->mounts = 1; | |
4400 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4401 | init_task.nsproxy->mnt_ns = ns; |
4402 | get_mnt_ns(ns); | |
4403 | ||
be08d6d2 AV |
4404 | root.mnt = mnt; |
4405 | root.dentry = mnt->mnt_root; | |
da362b09 | 4406 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4407 | |
4408 | set_fs_pwd(current->fs, &root); | |
4409 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4410 | } |
4411 | ||
74bf17cf | 4412 | void __init mnt_init(void) |
1da177e4 | 4413 | { |
15a67dd8 | 4414 | int err; |
1da177e4 | 4415 | |
7d6fec45 | 4416 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
79f6540b | 4417 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 4418 | |
0818bf27 | 4419 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4420 | sizeof(struct hlist_head), |
0818bf27 | 4421 | mhash_entries, 19, |
3d375d78 | 4422 | HASH_ZERO, |
0818bf27 AV |
4423 | &m_hash_shift, &m_hash_mask, 0, 0); |
4424 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4425 | sizeof(struct hlist_head), | |
4426 | mphash_entries, 19, | |
3d375d78 | 4427 | HASH_ZERO, |
0818bf27 | 4428 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4429 | |
84d17192 | 4430 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4431 | panic("Failed to allocate mount hash table\n"); |
4432 | ||
4b93dc9b TH |
4433 | kernfs_init(); |
4434 | ||
15a67dd8 RD |
4435 | err = sysfs_init(); |
4436 | if (err) | |
4437 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4438 | __func__, err); |
00d26666 GKH |
4439 | fs_kobj = kobject_create_and_add("fs", NULL); |
4440 | if (!fs_kobj) | |
8e24eea7 | 4441 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4442 | shmem_init(); |
1da177e4 LT |
4443 | init_rootfs(); |
4444 | init_mount_tree(); | |
4445 | } | |
4446 | ||
616511d0 | 4447 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4448 | { |
1a7b8969 | 4449 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4450 | return; |
7b00ed6f | 4451 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4452 | free_mnt_ns(ns); |
1da177e4 | 4453 | } |
9d412a43 | 4454 | |
d911b458 | 4455 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4456 | { |
423e0ab0 | 4457 | struct vfsmount *mnt; |
d911b458 | 4458 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4459 | if (!IS_ERR(mnt)) { |
4460 | /* | |
4461 | * it is a longterm mount, don't release mnt until | |
4462 | * we unmount before file sys is unregistered | |
4463 | */ | |
f7a99c5b | 4464 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4465 | } |
4466 | return mnt; | |
9d412a43 | 4467 | } |
d911b458 | 4468 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4469 | |
4470 | void kern_unmount(struct vfsmount *mnt) | |
4471 | { | |
4472 | /* release long term mount so mount point can be released */ | |
da27f796 RR |
4473 | if (!IS_ERR(mnt)) { |
4474 | mnt_make_shortterm(mnt); | |
48a066e7 | 4475 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4476 | mntput(mnt); |
4477 | } | |
4478 | } | |
4479 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4480 | |
df820f8d MS |
4481 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4482 | { | |
4483 | unsigned int i; | |
4484 | ||
4485 | for (i = 0; i < num; i++) | |
da27f796 | 4486 | mnt_make_shortterm(mnt[i]); |
df820f8d MS |
4487 | synchronize_rcu_expedited(); |
4488 | for (i = 0; i < num; i++) | |
4489 | mntput(mnt[i]); | |
4490 | } | |
4491 | EXPORT_SYMBOL(kern_unmount_array); | |
4492 | ||
02125a82 AV |
4493 | bool our_mnt(struct vfsmount *mnt) |
4494 | { | |
143c8c91 | 4495 | return check_mnt(real_mount(mnt)); |
02125a82 | 4496 | } |
8823c079 | 4497 | |
3151527e EB |
4498 | bool current_chrooted(void) |
4499 | { | |
4500 | /* Does the current process have a non-standard root */ | |
4501 | struct path ns_root; | |
4502 | struct path fs_root; | |
4503 | bool chrooted; | |
4504 | ||
4505 | /* Find the namespace root */ | |
4506 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4507 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4508 | path_get(&ns_root); | |
4509 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4510 | ; | |
4511 | ||
4512 | get_fs_root(current->fs, &fs_root); | |
4513 | ||
4514 | chrooted = !path_equal(&fs_root, &ns_root); | |
4515 | ||
4516 | path_put(&fs_root); | |
4517 | path_put(&ns_root); | |
4518 | ||
4519 | return chrooted; | |
4520 | } | |
4521 | ||
132e4608 DH |
4522 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4523 | const struct super_block *sb, | |
8654df4e | 4524 | int *new_mnt_flags) |
87a8ebd6 | 4525 | { |
8c6cf9cc | 4526 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4527 | struct mount *mnt; |
e51db735 | 4528 | bool visible = false; |
87a8ebd6 | 4529 | |
44bb4385 | 4530 | down_read(&namespace_sem); |
9f6c61f9 | 4531 | lock_ns_list(ns); |
87a8ebd6 | 4532 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4533 | struct mount *child; |
77b1a97d EB |
4534 | int mnt_flags; |
4535 | ||
9f6c61f9 MS |
4536 | if (mnt_is_cursor(mnt)) |
4537 | continue; | |
4538 | ||
132e4608 | 4539 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4540 | continue; |
4541 | ||
7e96c1b0 EB |
4542 | /* This mount is not fully visible if it's root directory |
4543 | * is not the root directory of the filesystem. | |
4544 | */ | |
4545 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4546 | continue; | |
4547 | ||
a1935c17 | 4548 | /* A local view of the mount flags */ |
77b1a97d | 4549 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4550 | |
695e9df0 | 4551 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4552 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4553 | mnt_flags |= MNT_LOCK_READONLY; |
4554 | ||
8c6cf9cc EB |
4555 | /* Verify the mount flags are equal to or more permissive |
4556 | * than the proposed new mount. | |
4557 | */ | |
77b1a97d | 4558 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4559 | !(new_flags & MNT_READONLY)) |
4560 | continue; | |
77b1a97d EB |
4561 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4562 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4563 | continue; |
4564 | ||
ceeb0e5d EB |
4565 | /* This mount is not fully visible if there are any |
4566 | * locked child mounts that cover anything except for | |
4567 | * empty directories. | |
e51db735 EB |
4568 | */ |
4569 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4570 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4571 | /* Only worry about locked mounts */ |
d71ed6c9 | 4572 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4573 | continue; |
7236c85e EB |
4574 | /* Is the directory permanetly empty? */ |
4575 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4576 | goto next; |
87a8ebd6 | 4577 | } |
8c6cf9cc | 4578 | /* Preserve the locked attributes */ |
77b1a97d | 4579 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4580 | MNT_LOCK_ATIME); |
e51db735 EB |
4581 | visible = true; |
4582 | goto found; | |
4583 | next: ; | |
87a8ebd6 | 4584 | } |
e51db735 | 4585 | found: |
9f6c61f9 | 4586 | unlock_ns_list(ns); |
44bb4385 | 4587 | up_read(&namespace_sem); |
e51db735 | 4588 | return visible; |
87a8ebd6 EB |
4589 | } |
4590 | ||
132e4608 | 4591 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4592 | { |
a1935c17 | 4593 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4594 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4595 | unsigned long s_iflags; | |
4596 | ||
4597 | if (ns->user_ns == &init_user_ns) | |
4598 | return false; | |
4599 | ||
4600 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4601 | s_iflags = sb->s_iflags; |
8654df4e EB |
4602 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4603 | return false; | |
4604 | ||
a1935c17 EB |
4605 | if ((s_iflags & required_iflags) != required_iflags) { |
4606 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4607 | required_iflags); | |
4608 | return true; | |
4609 | } | |
4610 | ||
132e4608 | 4611 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4612 | } |
4613 | ||
380cf5ba AL |
4614 | bool mnt_may_suid(struct vfsmount *mnt) |
4615 | { | |
4616 | /* | |
4617 | * Foreign mounts (accessed via fchdir or through /proc | |
4618 | * symlinks) are always treated as if they are nosuid. This | |
4619 | * prevents namespaces from trusting potentially unsafe | |
4620 | * suid/sgid bits, file caps, or security labels that originate | |
4621 | * in other namespaces. | |
4622 | */ | |
4623 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4624 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4625 | } | |
4626 | ||
64964528 | 4627 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4628 | { |
58be2825 | 4629 | struct ns_common *ns = NULL; |
8823c079 EB |
4630 | struct nsproxy *nsproxy; |
4631 | ||
728dba3a EB |
4632 | task_lock(task); |
4633 | nsproxy = task->nsproxy; | |
8823c079 | 4634 | if (nsproxy) { |
58be2825 AV |
4635 | ns = &nsproxy->mnt_ns->ns; |
4636 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4637 | } |
728dba3a | 4638 | task_unlock(task); |
8823c079 EB |
4639 | |
4640 | return ns; | |
4641 | } | |
4642 | ||
64964528 | 4643 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4644 | { |
58be2825 | 4645 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4646 | } |
4647 | ||
f2a8d52e | 4648 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4649 | { |
f2a8d52e CB |
4650 | struct nsproxy *nsproxy = nsset->nsproxy; |
4651 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4652 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4653 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4654 | struct path root; |
4f757f3c | 4655 | int err; |
8823c079 | 4656 | |
0c55cfc4 | 4657 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4658 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4659 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4660 | return -EPERM; |
8823c079 | 4661 | |
74e83122 AV |
4662 | if (is_anon_ns(mnt_ns)) |
4663 | return -EINVAL; | |
4664 | ||
8823c079 EB |
4665 | if (fs->users != 1) |
4666 | return -EINVAL; | |
4667 | ||
4668 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4669 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4670 | nsproxy->mnt_ns = mnt_ns; |
4671 | ||
4672 | /* Find the root */ | |
4f757f3c AV |
4673 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4674 | "/", LOOKUP_DOWN, &root); | |
4675 | if (err) { | |
4676 | /* revert to old namespace */ | |
4677 | nsproxy->mnt_ns = old_mnt_ns; | |
4678 | put_mnt_ns(mnt_ns); | |
4679 | return err; | |
4680 | } | |
8823c079 | 4681 | |
4068367c AV |
4682 | put_mnt_ns(old_mnt_ns); |
4683 | ||
8823c079 EB |
4684 | /* Update the pwd and root */ |
4685 | set_fs_pwd(fs, &root); | |
4686 | set_fs_root(fs, &root); | |
4687 | ||
4688 | path_put(&root); | |
4689 | return 0; | |
4690 | } | |
4691 | ||
bcac25a5 AV |
4692 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4693 | { | |
4694 | return to_mnt_ns(ns)->user_ns; | |
4695 | } | |
4696 | ||
8823c079 EB |
4697 | const struct proc_ns_operations mntns_operations = { |
4698 | .name = "mnt", | |
4699 | .type = CLONE_NEWNS, | |
4700 | .get = mntns_get, | |
4701 | .put = mntns_put, | |
4702 | .install = mntns_install, | |
bcac25a5 | 4703 | .owner = mntns_owner, |
8823c079 | 4704 | }; |
ab171b95 LC |
4705 | |
4706 | #ifdef CONFIG_SYSCTL | |
4707 | static struct ctl_table fs_namespace_sysctls[] = { | |
4708 | { | |
4709 | .procname = "mount-max", | |
4710 | .data = &sysctl_mount_max, | |
4711 | .maxlen = sizeof(unsigned int), | |
4712 | .mode = 0644, | |
4713 | .proc_handler = proc_dointvec_minmax, | |
4714 | .extra1 = SYSCTL_ONE, | |
4715 | }, | |
4716 | { } | |
4717 | }; | |
4718 | ||
4719 | static int __init init_fs_namespace_sysctls(void) | |
4720 | { | |
4721 | register_sysctl_init("fs", fs_namespace_sysctls); | |
4722 | return 0; | |
4723 | } | |
4724 | fs_initcall(init_fs_namespace_sysctls); | |
4725 | ||
4726 | #endif /* CONFIG_SYSCTL */ |