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 | |
68279f9c AD |
42 | static unsigned int m_hash_mask __ro_after_init; |
43 | static unsigned int m_hash_shift __ro_after_init; | |
44 | static unsigned int mp_hash_mask __ro_after_init; | |
45 | static unsigned int mp_hash_shift __ro_after_init; | |
0818bf27 AV |
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 | |
68279f9c AD |
71 | static struct hlist_head *mount_hashtable __ro_after_init; |
72 | static struct hlist_head *mountpoint_hashtable __ro_after_init; | |
73 | static struct kmem_cache *mnt_cache __ro_after_init; | |
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 */ |
68279f9c | 89 | struct kobject *fs_kobj __ro_after_init; |
00d26666 | 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 | 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 | 277 | { |
278 | #ifdef CONFIG_SMP | |
68e8a9fe | 279 | this_cpu_inc(mnt->mnt_pcp->mnt_writers); |
d3ef3d73 | 280 | #else |
68e8a9fe | 281 | mnt->mnt_writers++; |
d3ef3d73 | 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 | 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 | 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 | { | |
d7439fb1 | 312 | if (READ_ONCE(mnt->mnt_sb->s_readonly_remount)) |
4ed5e82f | 313 | return 1; |
d7439fb1 JK |
314 | /* |
315 | * The barrier pairs with the barrier in sb_start_ro_state_change() | |
316 | * making sure if we don't see s_readonly_remount set yet, we also will | |
317 | * not see any superblock / mount flag changes done by remount. | |
318 | * It also pairs with the barrier in sb_end_ro_state_change() | |
319 | * assuring that if we see s_readonly_remount already cleared, we will | |
320 | * see the values of superblock / mount flags updated by remount. | |
321 | */ | |
4ed5e82f MS |
322 | smp_rmb(); |
323 | return __mnt_is_readonly(mnt); | |
324 | } | |
325 | ||
8366025e | 326 | /* |
eb04c282 JK |
327 | * Most r/o & frozen checks on a fs are for operations that take discrete |
328 | * amounts of time, like a write() or unlink(). We must keep track of when | |
329 | * those operations start (for permission checks) and when they end, so that we | |
330 | * can determine when writes are able to occur to a filesystem. | |
8366025e DH |
331 | */ |
332 | /** | |
3e15dcf7 | 333 | * mnt_get_write_access - get write access to a mount without freeze protection |
83adc753 | 334 | * @m: the mount on which to take a write |
8366025e | 335 | * |
eb04c282 JK |
336 | * This tells the low-level filesystem that a write is about to be performed to |
337 | * it, and makes sure that writes are allowed (mnt it read-write) before | |
338 | * returning success. This operation does not protect against filesystem being | |
3e15dcf7 | 339 | * frozen. When the write operation is finished, mnt_put_write_access() must be |
eb04c282 | 340 | * called. This is effectively a refcount. |
8366025e | 341 | */ |
3e15dcf7 | 342 | int mnt_get_write_access(struct vfsmount *m) |
8366025e | 343 | { |
83adc753 | 344 | struct mount *mnt = real_mount(m); |
3d733633 | 345 | int ret = 0; |
3d733633 | 346 | |
d3ef3d73 | 347 | preempt_disable(); |
c6653a83 | 348 | mnt_inc_writers(mnt); |
d3ef3d73 | 349 | /* |
c6653a83 | 350 | * The store to mnt_inc_writers must be visible before we pass |
d3ef3d73 | 351 | * MNT_WRITE_HOLD loop below, so that the slowpath can see our |
352 | * incremented count after it has set MNT_WRITE_HOLD. | |
353 | */ | |
354 | smp_mb(); | |
0f8821da SAS |
355 | might_lock(&mount_lock.lock); |
356 | while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) { | |
357 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { | |
358 | cpu_relax(); | |
359 | } else { | |
360 | /* | |
361 | * This prevents priority inversion, if the task | |
362 | * setting MNT_WRITE_HOLD got preempted on a remote | |
363 | * CPU, and it prevents life lock if the task setting | |
364 | * MNT_WRITE_HOLD has a lower priority and is bound to | |
365 | * the same CPU as the task that is spinning here. | |
366 | */ | |
367 | preempt_enable(); | |
368 | lock_mount_hash(); | |
369 | unlock_mount_hash(); | |
370 | preempt_disable(); | |
371 | } | |
372 | } | |
d3ef3d73 | 373 | /* |
d7439fb1 JK |
374 | * The barrier pairs with the barrier sb_start_ro_state_change() making |
375 | * sure that if we see MNT_WRITE_HOLD cleared, we will also see | |
376 | * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in | |
377 | * mnt_is_readonly() and bail in case we are racing with remount | |
378 | * read-only. | |
d3ef3d73 | 379 | */ |
380 | smp_rmb(); | |
4ed5e82f | 381 | if (mnt_is_readonly(m)) { |
c6653a83 | 382 | mnt_dec_writers(mnt); |
3d733633 | 383 | ret = -EROFS; |
3d733633 | 384 | } |
d3ef3d73 | 385 | preempt_enable(); |
eb04c282 JK |
386 | |
387 | return ret; | |
388 | } | |
ddf9e2ff | 389 | EXPORT_SYMBOL_GPL(mnt_get_write_access); |
eb04c282 JK |
390 | |
391 | /** | |
392 | * mnt_want_write - get write access to a mount | |
393 | * @m: the mount on which to take a write | |
394 | * | |
395 | * This tells the low-level filesystem that a write is about to be performed to | |
396 | * it, and makes sure that writes are allowed (mount is read-write, filesystem | |
397 | * is not frozen) before returning success. When the write operation is | |
398 | * finished, mnt_drop_write() must be called. This is effectively a refcount. | |
399 | */ | |
400 | int mnt_want_write(struct vfsmount *m) | |
401 | { | |
402 | int ret; | |
403 | ||
404 | sb_start_write(m->mnt_sb); | |
3e15dcf7 | 405 | ret = mnt_get_write_access(m); |
eb04c282 JK |
406 | if (ret) |
407 | sb_end_write(m->mnt_sb); | |
3d733633 | 408 | return ret; |
8366025e DH |
409 | } |
410 | EXPORT_SYMBOL_GPL(mnt_want_write); | |
411 | ||
96029c4e | 412 | /** |
3e15dcf7 | 413 | * mnt_get_write_access_file - get write access to a file's mount |
96029c4e | 414 | * @file: the file who's mount on which to take a write |
415 | * | |
3e15dcf7 | 416 | * This is like mnt_get_write_access, but if @file is already open for write it |
14e43bf4 EB |
417 | * skips incrementing mnt_writers (since the open file already has a reference) |
418 | * and instead only does the check for emergency r/o remounts. This must be | |
3e15dcf7 | 419 | * paired with mnt_put_write_access_file. |
96029c4e | 420 | */ |
3e15dcf7 | 421 | int mnt_get_write_access_file(struct file *file) |
96029c4e | 422 | { |
14e43bf4 EB |
423 | if (file->f_mode & FMODE_WRITER) { |
424 | /* | |
425 | * Superblock may have become readonly while there are still | |
426 | * writable fd's, e.g. due to a fs error with errors=remount-ro | |
427 | */ | |
428 | if (__mnt_is_readonly(file->f_path.mnt)) | |
429 | return -EROFS; | |
430 | return 0; | |
431 | } | |
3e15dcf7 | 432 | return mnt_get_write_access(file->f_path.mnt); |
96029c4e | 433 | } |
eb04c282 | 434 | |
7c6893e3 MS |
435 | /** |
436 | * mnt_want_write_file - get write access to a file's mount | |
437 | * @file: the file who's mount on which to take a write | |
438 | * | |
14e43bf4 EB |
439 | * This is like mnt_want_write, but if the file is already open for writing it |
440 | * skips incrementing mnt_writers (since the open file already has a reference) | |
441 | * and instead only does the freeze protection and the check for emergency r/o | |
442 | * remounts. This must be paired with mnt_drop_write_file. | |
7c6893e3 MS |
443 | */ |
444 | int mnt_want_write_file(struct file *file) | |
445 | { | |
446 | int ret; | |
447 | ||
a6795a58 | 448 | sb_start_write(file_inode(file)->i_sb); |
3e15dcf7 | 449 | ret = mnt_get_write_access_file(file); |
eb04c282 | 450 | if (ret) |
a6795a58 | 451 | sb_end_write(file_inode(file)->i_sb); |
7c6893e3 MS |
452 | return ret; |
453 | } | |
96029c4e | 454 | EXPORT_SYMBOL_GPL(mnt_want_write_file); |
455 | ||
8366025e | 456 | /** |
3e15dcf7 | 457 | * mnt_put_write_access - give up write access to a mount |
8366025e DH |
458 | * @mnt: the mount on which to give up write access |
459 | * | |
460 | * Tells the low-level filesystem that we are done | |
461 | * performing writes to it. Must be matched with | |
3e15dcf7 | 462 | * mnt_get_write_access() call above. |
8366025e | 463 | */ |
3e15dcf7 | 464 | void mnt_put_write_access(struct vfsmount *mnt) |
8366025e | 465 | { |
d3ef3d73 | 466 | preempt_disable(); |
83adc753 | 467 | mnt_dec_writers(real_mount(mnt)); |
d3ef3d73 | 468 | preempt_enable(); |
8366025e | 469 | } |
ddf9e2ff | 470 | EXPORT_SYMBOL_GPL(mnt_put_write_access); |
eb04c282 JK |
471 | |
472 | /** | |
473 | * mnt_drop_write - give up write access to a mount | |
474 | * @mnt: the mount on which to give up write access | |
475 | * | |
476 | * Tells the low-level filesystem that we are done performing writes to it and | |
477 | * also allows filesystem to be frozen again. Must be matched with | |
478 | * mnt_want_write() call above. | |
479 | */ | |
480 | void mnt_drop_write(struct vfsmount *mnt) | |
481 | { | |
3e15dcf7 | 482 | mnt_put_write_access(mnt); |
eb04c282 JK |
483 | sb_end_write(mnt->mnt_sb); |
484 | } | |
8366025e DH |
485 | EXPORT_SYMBOL_GPL(mnt_drop_write); |
486 | ||
3e15dcf7 | 487 | void mnt_put_write_access_file(struct file *file) |
eb04c282 | 488 | { |
14e43bf4 | 489 | if (!(file->f_mode & FMODE_WRITER)) |
3e15dcf7 | 490 | mnt_put_write_access(file->f_path.mnt); |
eb04c282 JK |
491 | } |
492 | ||
7c6893e3 MS |
493 | void mnt_drop_write_file(struct file *file) |
494 | { | |
3e15dcf7 | 495 | mnt_put_write_access_file(file); |
7c6893e3 MS |
496 | sb_end_write(file_inode(file)->i_sb); |
497 | } | |
2a79f17e AV |
498 | EXPORT_SYMBOL(mnt_drop_write_file); |
499 | ||
538f4f02 CB |
500 | /** |
501 | * mnt_hold_writers - prevent write access to the given mount | |
502 | * @mnt: mnt to prevent write access to | |
503 | * | |
504 | * Prevents write access to @mnt if there are no active writers for @mnt. | |
505 | * This function needs to be called and return successfully before changing | |
506 | * properties of @mnt that need to remain stable for callers with write access | |
507 | * to @mnt. | |
508 | * | |
509 | * After this functions has been called successfully callers must pair it with | |
510 | * a call to mnt_unhold_writers() in order to stop preventing write access to | |
511 | * @mnt. | |
512 | * | |
513 | * Context: This function expects lock_mount_hash() to be held serializing | |
514 | * setting MNT_WRITE_HOLD. | |
515 | * Return: On success 0 is returned. | |
516 | * On error, -EBUSY is returned. | |
517 | */ | |
fbdc2f6c | 518 | static inline int mnt_hold_writers(struct mount *mnt) |
8366025e | 519 | { |
83adc753 | 520 | mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; |
3d733633 | 521 | /* |
d3ef3d73 | 522 | * After storing MNT_WRITE_HOLD, we'll read the counters. This store |
523 | * should be visible before we do. | |
3d733633 | 524 | */ |
d3ef3d73 | 525 | smp_mb(); |
526 | ||
3d733633 | 527 | /* |
d3ef3d73 | 528 | * With writers on hold, if this value is zero, then there are |
529 | * definitely no active writers (although held writers may subsequently | |
530 | * increment the count, they'll have to wait, and decrement it after | |
531 | * seeing MNT_READONLY). | |
532 | * | |
533 | * It is OK to have counter incremented on one CPU and decremented on | |
534 | * another: the sum will add up correctly. The danger would be when we | |
535 | * sum up each counter, if we read a counter before it is incremented, | |
536 | * but then read another CPU's count which it has been subsequently | |
537 | * decremented from -- we would see more decrements than we should. | |
538 | * MNT_WRITE_HOLD protects against this scenario, because | |
539 | * mnt_want_write first increments count, then smp_mb, then spins on | |
540 | * MNT_WRITE_HOLD, so it can't be decremented by another CPU while | |
541 | * we're counting up here. | |
3d733633 | 542 | */ |
c6653a83 | 543 | if (mnt_get_writers(mnt) > 0) |
fbdc2f6c CB |
544 | return -EBUSY; |
545 | ||
546 | return 0; | |
547 | } | |
548 | ||
538f4f02 CB |
549 | /** |
550 | * mnt_unhold_writers - stop preventing write access to the given mount | |
551 | * @mnt: mnt to stop preventing write access to | |
552 | * | |
553 | * Stop preventing write access to @mnt allowing callers to gain write access | |
554 | * to @mnt again. | |
555 | * | |
556 | * This function can only be called after a successful call to | |
557 | * mnt_hold_writers(). | |
558 | * | |
559 | * Context: This function expects lock_mount_hash() to be held. | |
560 | */ | |
fbdc2f6c CB |
561 | static inline void mnt_unhold_writers(struct mount *mnt) |
562 | { | |
d3ef3d73 | 563 | /* |
564 | * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers | |
565 | * that become unheld will see MNT_READONLY. | |
566 | */ | |
567 | smp_wmb(); | |
83adc753 | 568 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; |
fbdc2f6c CB |
569 | } |
570 | ||
571 | static int mnt_make_readonly(struct mount *mnt) | |
572 | { | |
573 | int ret; | |
574 | ||
575 | ret = mnt_hold_writers(mnt); | |
576 | if (!ret) | |
577 | mnt->mnt.mnt_flags |= MNT_READONLY; | |
578 | mnt_unhold_writers(mnt); | |
3d733633 | 579 | return ret; |
8366025e | 580 | } |
8366025e | 581 | |
4ed5e82f MS |
582 | int sb_prepare_remount_readonly(struct super_block *sb) |
583 | { | |
584 | struct mount *mnt; | |
585 | int err = 0; | |
586 | ||
8e8b8796 MS |
587 | /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ |
588 | if (atomic_long_read(&sb->s_remove_count)) | |
589 | return -EBUSY; | |
590 | ||
719ea2fb | 591 | lock_mount_hash(); |
4ed5e82f MS |
592 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
593 | if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { | |
ad1844a0 CB |
594 | err = mnt_hold_writers(mnt); |
595 | if (err) | |
4ed5e82f | 596 | break; |
4ed5e82f MS |
597 | } |
598 | } | |
8e8b8796 MS |
599 | if (!err && atomic_long_read(&sb->s_remove_count)) |
600 | err = -EBUSY; | |
601 | ||
d7439fb1 JK |
602 | if (!err) |
603 | sb_start_ro_state_change(sb); | |
4ed5e82f MS |
604 | list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { |
605 | if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) | |
606 | mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; | |
607 | } | |
719ea2fb | 608 | unlock_mount_hash(); |
4ed5e82f MS |
609 | |
610 | return err; | |
611 | } | |
612 | ||
b105e270 | 613 | static void free_vfsmnt(struct mount *mnt) |
1da177e4 | 614 | { |
256c8aed | 615 | mnt_idmap_put(mnt_idmap(&mnt->mnt)); |
fcc139ae | 616 | kfree_const(mnt->mnt_devname); |
d3ef3d73 | 617 | #ifdef CONFIG_SMP |
68e8a9fe | 618 | free_percpu(mnt->mnt_pcp); |
d3ef3d73 | 619 | #endif |
b105e270 | 620 | kmem_cache_free(mnt_cache, mnt); |
1da177e4 LT |
621 | } |
622 | ||
8ffcb32e DH |
623 | static void delayed_free_vfsmnt(struct rcu_head *head) |
624 | { | |
625 | free_vfsmnt(container_of(head, struct mount, mnt_rcu)); | |
626 | } | |
627 | ||
48a066e7 | 628 | /* call under rcu_read_lock */ |
294d71ff | 629 | int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
48a066e7 AV |
630 | { |
631 | struct mount *mnt; | |
632 | if (read_seqretry(&mount_lock, seq)) | |
294d71ff | 633 | return 1; |
48a066e7 | 634 | if (bastard == NULL) |
294d71ff | 635 | return 0; |
48a066e7 AV |
636 | mnt = real_mount(bastard); |
637 | mnt_add_count(mnt, 1); | |
119e1ef8 | 638 | smp_mb(); // see mntput_no_expire() |
48a066e7 | 639 | if (likely(!read_seqretry(&mount_lock, seq))) |
294d71ff | 640 | return 0; |
48a066e7 AV |
641 | if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { |
642 | mnt_add_count(mnt, -1); | |
294d71ff AV |
643 | return 1; |
644 | } | |
119e1ef8 AV |
645 | lock_mount_hash(); |
646 | if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { | |
647 | mnt_add_count(mnt, -1); | |
648 | unlock_mount_hash(); | |
649 | return 1; | |
650 | } | |
651 | unlock_mount_hash(); | |
652 | /* caller will mntput() */ | |
294d71ff AV |
653 | return -1; |
654 | } | |
655 | ||
656 | /* call under rcu_read_lock */ | |
7e4745a0 | 657 | static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) |
294d71ff AV |
658 | { |
659 | int res = __legitimize_mnt(bastard, seq); | |
660 | if (likely(!res)) | |
661 | return true; | |
662 | if (unlikely(res < 0)) { | |
663 | rcu_read_unlock(); | |
664 | mntput(bastard); | |
665 | rcu_read_lock(); | |
48a066e7 | 666 | } |
48a066e7 AV |
667 | return false; |
668 | } | |
669 | ||
104026c2 CB |
670 | /** |
671 | * __lookup_mnt - find first child mount | |
672 | * @mnt: parent mount | |
673 | * @dentry: mountpoint | |
674 | * | |
675 | * If @mnt has a child mount @c mounted @dentry find and return it. | |
676 | * | |
677 | * Note that the child mount @c need not be unique. There are cases | |
678 | * where shadow mounts are created. For example, during mount | |
679 | * propagation when a source mount @mnt whose root got overmounted by a | |
680 | * mount @o after path lookup but before @namespace_sem could be | |
681 | * acquired gets copied and propagated. So @mnt gets copied including | |
682 | * @o. When @mnt is propagated to a destination mount @d that already | |
683 | * has another mount @n mounted at the same mountpoint then the source | |
684 | * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on | |
685 | * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt | |
686 | * on @dentry. | |
687 | * | |
688 | * Return: The first child of @mnt mounted @dentry or NULL. | |
1da177e4 | 689 | */ |
474279dc | 690 | struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 691 | { |
38129a13 | 692 | struct hlist_head *head = m_hash(mnt, dentry); |
474279dc AV |
693 | struct mount *p; |
694 | ||
38129a13 | 695 | hlist_for_each_entry_rcu(p, head, mnt_hash) |
474279dc AV |
696 | if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) |
697 | return p; | |
698 | return NULL; | |
699 | } | |
700 | ||
a05964f3 | 701 | /* |
f015f126 DH |
702 | * lookup_mnt - Return the first child mount mounted at path |
703 | * | |
704 | * "First" means first mounted chronologically. If you create the | |
705 | * following mounts: | |
706 | * | |
707 | * mount /dev/sda1 /mnt | |
708 | * mount /dev/sda2 /mnt | |
709 | * mount /dev/sda3 /mnt | |
710 | * | |
711 | * Then lookup_mnt() on the base /mnt dentry in the root mount will | |
712 | * return successively the root dentry and vfsmount of /dev/sda1, then | |
713 | * /dev/sda2, then /dev/sda3, then NULL. | |
714 | * | |
715 | * lookup_mnt takes a reference to the found vfsmount. | |
a05964f3 | 716 | */ |
ca71cf71 | 717 | struct vfsmount *lookup_mnt(const struct path *path) |
a05964f3 | 718 | { |
c7105365 | 719 | struct mount *child_mnt; |
48a066e7 AV |
720 | struct vfsmount *m; |
721 | unsigned seq; | |
99b7db7b | 722 | |
48a066e7 AV |
723 | rcu_read_lock(); |
724 | do { | |
725 | seq = read_seqbegin(&mount_lock); | |
726 | child_mnt = __lookup_mnt(path->mnt, path->dentry); | |
727 | m = child_mnt ? &child_mnt->mnt : NULL; | |
728 | } while (!legitimize_mnt(m, seq)); | |
729 | rcu_read_unlock(); | |
730 | return m; | |
a05964f3 RP |
731 | } |
732 | ||
9f6c61f9 MS |
733 | static inline void lock_ns_list(struct mnt_namespace *ns) |
734 | { | |
735 | spin_lock(&ns->ns_lock); | |
736 | } | |
737 | ||
738 | static inline void unlock_ns_list(struct mnt_namespace *ns) | |
739 | { | |
740 | spin_unlock(&ns->ns_lock); | |
741 | } | |
742 | ||
743 | static inline bool mnt_is_cursor(struct mount *mnt) | |
744 | { | |
745 | return mnt->mnt.mnt_flags & MNT_CURSOR; | |
746 | } | |
747 | ||
7af1364f EB |
748 | /* |
749 | * __is_local_mountpoint - Test to see if dentry is a mountpoint in the | |
750 | * current mount namespace. | |
751 | * | |
752 | * The common case is dentries are not mountpoints at all and that | |
753 | * test is handled inline. For the slow case when we are actually | |
754 | * dealing with a mountpoint of some kind, walk through all of the | |
755 | * mounts in the current mount namespace and test to see if the dentry | |
756 | * is a mountpoint. | |
757 | * | |
758 | * The mount_hashtable is not usable in the context because we | |
759 | * need to identify all mounts that may be in the current mount | |
760 | * namespace not just a mount that happens to have some specified | |
761 | * parent mount. | |
762 | */ | |
763 | bool __is_local_mountpoint(struct dentry *dentry) | |
764 | { | |
765 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; | |
766 | struct mount *mnt; | |
767 | bool is_covered = false; | |
768 | ||
7af1364f | 769 | down_read(&namespace_sem); |
9f6c61f9 | 770 | lock_ns_list(ns); |
7af1364f | 771 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
9f6c61f9 MS |
772 | if (mnt_is_cursor(mnt)) |
773 | continue; | |
7af1364f EB |
774 | is_covered = (mnt->mnt_mountpoint == dentry); |
775 | if (is_covered) | |
776 | break; | |
777 | } | |
9f6c61f9 | 778 | unlock_ns_list(ns); |
7af1364f | 779 | up_read(&namespace_sem); |
5ad05cc8 | 780 | |
7af1364f EB |
781 | return is_covered; |
782 | } | |
783 | ||
e2dfa935 | 784 | static struct mountpoint *lookup_mountpoint(struct dentry *dentry) |
84d17192 | 785 | { |
0818bf27 | 786 | struct hlist_head *chain = mp_hash(dentry); |
84d17192 AV |
787 | struct mountpoint *mp; |
788 | ||
0818bf27 | 789 | hlist_for_each_entry(mp, chain, m_hash) { |
84d17192 | 790 | if (mp->m_dentry == dentry) { |
84d17192 AV |
791 | mp->m_count++; |
792 | return mp; | |
793 | } | |
794 | } | |
e2dfa935 EB |
795 | return NULL; |
796 | } | |
797 | ||
3895dbf8 | 798 | static struct mountpoint *get_mountpoint(struct dentry *dentry) |
e2dfa935 | 799 | { |
3895dbf8 | 800 | struct mountpoint *mp, *new = NULL; |
e2dfa935 | 801 | int ret; |
84d17192 | 802 | |
3895dbf8 | 803 | if (d_mountpoint(dentry)) { |
1e9c75fb BC |
804 | /* might be worth a WARN_ON() */ |
805 | if (d_unlinked(dentry)) | |
806 | return ERR_PTR(-ENOENT); | |
3895dbf8 EB |
807 | mountpoint: |
808 | read_seqlock_excl(&mount_lock); | |
809 | mp = lookup_mountpoint(dentry); | |
810 | read_sequnlock_excl(&mount_lock); | |
811 | if (mp) | |
812 | goto done; | |
813 | } | |
814 | ||
815 | if (!new) | |
816 | new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); | |
817 | if (!new) | |
84d17192 AV |
818 | return ERR_PTR(-ENOMEM); |
819 | ||
3895dbf8 EB |
820 | |
821 | /* Exactly one processes may set d_mounted */ | |
eed81007 | 822 | ret = d_set_mounted(dentry); |
eed81007 | 823 | |
3895dbf8 EB |
824 | /* Someone else set d_mounted? */ |
825 | if (ret == -EBUSY) | |
826 | goto mountpoint; | |
827 | ||
828 | /* The dentry is not available as a mountpoint? */ | |
829 | mp = ERR_PTR(ret); | |
830 | if (ret) | |
831 | goto done; | |
832 | ||
833 | /* Add the new mountpoint to the hash table */ | |
834 | read_seqlock_excl(&mount_lock); | |
4edbe133 | 835 | new->m_dentry = dget(dentry); |
3895dbf8 EB |
836 | new->m_count = 1; |
837 | hlist_add_head(&new->m_hash, mp_hash(dentry)); | |
838 | INIT_HLIST_HEAD(&new->m_list); | |
839 | read_sequnlock_excl(&mount_lock); | |
840 | ||
841 | mp = new; | |
842 | new = NULL; | |
843 | done: | |
844 | kfree(new); | |
84d17192 AV |
845 | return mp; |
846 | } | |
847 | ||
4edbe133 AV |
848 | /* |
849 | * vfsmount lock must be held. Additionally, the caller is responsible | |
850 | * for serializing calls for given disposal list. | |
851 | */ | |
852 | static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) | |
84d17192 AV |
853 | { |
854 | if (!--mp->m_count) { | |
855 | struct dentry *dentry = mp->m_dentry; | |
0a5eb7c8 | 856 | BUG_ON(!hlist_empty(&mp->m_list)); |
84d17192 AV |
857 | spin_lock(&dentry->d_lock); |
858 | dentry->d_flags &= ~DCACHE_MOUNTED; | |
859 | spin_unlock(&dentry->d_lock); | |
4edbe133 | 860 | dput_to_list(dentry, list); |
0818bf27 | 861 | hlist_del(&mp->m_hash); |
84d17192 AV |
862 | kfree(mp); |
863 | } | |
864 | } | |
865 | ||
4edbe133 AV |
866 | /* called with namespace_lock and vfsmount lock */ |
867 | static void put_mountpoint(struct mountpoint *mp) | |
868 | { | |
869 | __put_mountpoint(mp, &ex_mountpoints); | |
870 | } | |
871 | ||
143c8c91 | 872 | static inline int check_mnt(struct mount *mnt) |
1da177e4 | 873 | { |
6b3286ed | 874 | return mnt->mnt_ns == current->nsproxy->mnt_ns; |
1da177e4 LT |
875 | } |
876 | ||
99b7db7b NP |
877 | /* |
878 | * vfsmount lock must be held for write | |
879 | */ | |
6b3286ed | 880 | static void touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
881 | { |
882 | if (ns) { | |
883 | ns->event = ++event; | |
884 | wake_up_interruptible(&ns->poll); | |
885 | } | |
886 | } | |
887 | ||
99b7db7b NP |
888 | /* |
889 | * vfsmount lock must be held for write | |
890 | */ | |
6b3286ed | 891 | static void __touch_mnt_namespace(struct mnt_namespace *ns) |
5addc5dd AV |
892 | { |
893 | if (ns && ns->event != event) { | |
894 | ns->event = event; | |
895 | wake_up_interruptible(&ns->poll); | |
896 | } | |
897 | } | |
898 | ||
99b7db7b NP |
899 | /* |
900 | * vfsmount lock must be held for write | |
901 | */ | |
e4e59906 | 902 | static struct mountpoint *unhash_mnt(struct mount *mnt) |
419148da | 903 | { |
e4e59906 | 904 | struct mountpoint *mp; |
0714a533 | 905 | mnt->mnt_parent = mnt; |
a73324da | 906 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; |
6b41d536 | 907 | list_del_init(&mnt->mnt_child); |
38129a13 | 908 | hlist_del_init_rcu(&mnt->mnt_hash); |
0a5eb7c8 | 909 | hlist_del_init(&mnt->mnt_mp_list); |
e4e59906 | 910 | mp = mnt->mnt_mp; |
84d17192 | 911 | mnt->mnt_mp = NULL; |
e4e59906 | 912 | return mp; |
7bdb11de EB |
913 | } |
914 | ||
6a46c573 EB |
915 | /* |
916 | * vfsmount lock must be held for write | |
917 | */ | |
918 | static void umount_mnt(struct mount *mnt) | |
919 | { | |
e4e59906 | 920 | put_mountpoint(unhash_mnt(mnt)); |
6a46c573 EB |
921 | } |
922 | ||
99b7db7b NP |
923 | /* |
924 | * vfsmount lock must be held for write | |
925 | */ | |
84d17192 AV |
926 | void mnt_set_mountpoint(struct mount *mnt, |
927 | struct mountpoint *mp, | |
44d964d6 | 928 | struct mount *child_mnt) |
b90fa9ae | 929 | { |
84d17192 | 930 | mp->m_count++; |
3a2393d7 | 931 | mnt_add_count(mnt, 1); /* essentially, that's mntget */ |
4edbe133 | 932 | child_mnt->mnt_mountpoint = mp->m_dentry; |
3a2393d7 | 933 | child_mnt->mnt_parent = mnt; |
84d17192 | 934 | child_mnt->mnt_mp = mp; |
0a5eb7c8 | 935 | hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); |
b90fa9ae RP |
936 | } |
937 | ||
6ac39281 CB |
938 | /** |
939 | * mnt_set_mountpoint_beneath - mount a mount beneath another one | |
940 | * | |
941 | * @new_parent: the source mount | |
942 | * @top_mnt: the mount beneath which @new_parent is mounted | |
943 | * @new_mp: the new mountpoint of @top_mnt on @new_parent | |
944 | * | |
945 | * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and | |
946 | * parent @top_mnt->mnt_parent and mount it on top of @new_parent at | |
947 | * @new_mp. And mount @new_parent on the old parent and old | |
948 | * mountpoint of @top_mnt. | |
949 | * | |
950 | * Context: This function expects namespace_lock() and lock_mount_hash() | |
951 | * to have been acquired in that order. | |
952 | */ | |
953 | static void mnt_set_mountpoint_beneath(struct mount *new_parent, | |
954 | struct mount *top_mnt, | |
955 | struct mountpoint *new_mp) | |
956 | { | |
957 | struct mount *old_top_parent = top_mnt->mnt_parent; | |
958 | struct mountpoint *old_top_mp = top_mnt->mnt_mp; | |
959 | ||
960 | mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent); | |
961 | mnt_change_mountpoint(new_parent, new_mp, top_mnt); | |
962 | } | |
963 | ||
964 | ||
1064f874 EB |
965 | static void __attach_mnt(struct mount *mnt, struct mount *parent) |
966 | { | |
967 | hlist_add_head_rcu(&mnt->mnt_hash, | |
968 | m_hash(&parent->mnt, mnt->mnt_mountpoint)); | |
969 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
970 | } | |
971 | ||
6ac39281 CB |
972 | /** |
973 | * attach_mnt - mount a mount, attach to @mount_hashtable and parent's | |
974 | * list of child mounts | |
975 | * @parent: the parent | |
976 | * @mnt: the new mount | |
977 | * @mp: the new mountpoint | |
978 | * @beneath: whether to mount @mnt beneath or on top of @parent | |
979 | * | |
980 | * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt | |
981 | * to @parent's child mount list and to @mount_hashtable. | |
982 | * | |
983 | * If @beneath is true, remove @mnt from its current parent and | |
984 | * mountpoint and mount it on @mp on @parent, and mount @parent on the | |
985 | * old parent and old mountpoint of @mnt. Finally, attach @parent to | |
986 | * @mnt_hashtable and @parent->mnt_parent->mnt_mounts. | |
987 | * | |
988 | * Note, when __attach_mnt() is called @mnt->mnt_parent already points | |
989 | * to the correct parent. | |
990 | * | |
991 | * Context: This function expects namespace_lock() and lock_mount_hash() | |
992 | * to have been acquired in that order. | |
99b7db7b | 993 | */ |
6ac39281 CB |
994 | static void attach_mnt(struct mount *mnt, struct mount *parent, |
995 | struct mountpoint *mp, bool beneath) | |
1da177e4 | 996 | { |
6ac39281 CB |
997 | if (beneath) |
998 | mnt_set_mountpoint_beneath(mnt, parent, mp); | |
999 | else | |
1000 | mnt_set_mountpoint(parent, mp, mnt); | |
1001 | /* | |
1002 | * Note, @mnt->mnt_parent has to be used. If @mnt was mounted | |
1003 | * beneath @parent then @mnt will need to be attached to | |
1004 | * @parent's old parent, not @parent. IOW, @mnt->mnt_parent | |
1005 | * isn't the same mount as @parent. | |
1006 | */ | |
1007 | __attach_mnt(mnt, mnt->mnt_parent); | |
b90fa9ae RP |
1008 | } |
1009 | ||
1064f874 | 1010 | void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) |
12a5b529 | 1011 | { |
1064f874 | 1012 | struct mountpoint *old_mp = mnt->mnt_mp; |
1064f874 EB |
1013 | struct mount *old_parent = mnt->mnt_parent; |
1014 | ||
1015 | list_del_init(&mnt->mnt_child); | |
1016 | hlist_del_init(&mnt->mnt_mp_list); | |
1017 | hlist_del_init_rcu(&mnt->mnt_hash); | |
1018 | ||
6ac39281 | 1019 | attach_mnt(mnt, parent, mp, false); |
1064f874 EB |
1020 | |
1021 | put_mountpoint(old_mp); | |
1064f874 | 1022 | mnt_add_count(old_parent, -1); |
12a5b529 AV |
1023 | } |
1024 | ||
b90fa9ae | 1025 | /* |
99b7db7b | 1026 | * vfsmount lock must be held for write |
b90fa9ae | 1027 | */ |
1064f874 | 1028 | static void commit_tree(struct mount *mnt) |
b90fa9ae | 1029 | { |
0714a533 | 1030 | struct mount *parent = mnt->mnt_parent; |
83adc753 | 1031 | struct mount *m; |
b90fa9ae | 1032 | LIST_HEAD(head); |
143c8c91 | 1033 | struct mnt_namespace *n = parent->mnt_ns; |
b90fa9ae | 1034 | |
0714a533 | 1035 | BUG_ON(parent == mnt); |
b90fa9ae | 1036 | |
1a4eeaf2 | 1037 | list_add_tail(&head, &mnt->mnt_list); |
f7a99c5b | 1038 | list_for_each_entry(m, &head, mnt_list) |
143c8c91 | 1039 | m->mnt_ns = n; |
f03c6599 | 1040 | |
b90fa9ae RP |
1041 | list_splice(&head, n->list.prev); |
1042 | ||
d2921684 EB |
1043 | n->mounts += n->pending_mounts; |
1044 | n->pending_mounts = 0; | |
1045 | ||
1064f874 | 1046 | __attach_mnt(mnt, parent); |
6b3286ed | 1047 | touch_mnt_namespace(n); |
1da177e4 LT |
1048 | } |
1049 | ||
909b0a88 | 1050 | static struct mount *next_mnt(struct mount *p, struct mount *root) |
1da177e4 | 1051 | { |
6b41d536 AV |
1052 | struct list_head *next = p->mnt_mounts.next; |
1053 | if (next == &p->mnt_mounts) { | |
1da177e4 | 1054 | while (1) { |
909b0a88 | 1055 | if (p == root) |
1da177e4 | 1056 | return NULL; |
6b41d536 AV |
1057 | next = p->mnt_child.next; |
1058 | if (next != &p->mnt_parent->mnt_mounts) | |
1da177e4 | 1059 | break; |
0714a533 | 1060 | p = p->mnt_parent; |
1da177e4 LT |
1061 | } |
1062 | } | |
6b41d536 | 1063 | return list_entry(next, struct mount, mnt_child); |
1da177e4 LT |
1064 | } |
1065 | ||
315fc83e | 1066 | static struct mount *skip_mnt_tree(struct mount *p) |
9676f0c6 | 1067 | { |
6b41d536 AV |
1068 | struct list_head *prev = p->mnt_mounts.prev; |
1069 | while (prev != &p->mnt_mounts) { | |
1070 | p = list_entry(prev, struct mount, mnt_child); | |
1071 | prev = p->mnt_mounts.prev; | |
9676f0c6 RP |
1072 | } |
1073 | return p; | |
1074 | } | |
1075 | ||
8f291889 AV |
1076 | /** |
1077 | * vfs_create_mount - Create a mount for a configured superblock | |
1078 | * @fc: The configuration context with the superblock attached | |
1079 | * | |
1080 | * Create a mount to an already configured superblock. If necessary, the | |
1081 | * caller should invoke vfs_get_tree() before calling this. | |
1082 | * | |
1083 | * Note that this does not attach the mount to anything. | |
1084 | */ | |
1085 | struct vfsmount *vfs_create_mount(struct fs_context *fc) | |
9d412a43 | 1086 | { |
b105e270 | 1087 | struct mount *mnt; |
9d412a43 | 1088 | |
8f291889 AV |
1089 | if (!fc->root) |
1090 | return ERR_PTR(-EINVAL); | |
9d412a43 | 1091 | |
8f291889 | 1092 | mnt = alloc_vfsmnt(fc->source ?: "none"); |
9d412a43 AV |
1093 | if (!mnt) |
1094 | return ERR_PTR(-ENOMEM); | |
1095 | ||
8f291889 | 1096 | if (fc->sb_flags & SB_KERNMOUNT) |
b105e270 | 1097 | mnt->mnt.mnt_flags = MNT_INTERNAL; |
9d412a43 | 1098 | |
8f291889 AV |
1099 | atomic_inc(&fc->root->d_sb->s_active); |
1100 | mnt->mnt.mnt_sb = fc->root->d_sb; | |
1101 | mnt->mnt.mnt_root = dget(fc->root); | |
1102 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1103 | mnt->mnt_parent = mnt; | |
9d412a43 | 1104 | |
719ea2fb | 1105 | lock_mount_hash(); |
8f291889 | 1106 | list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); |
719ea2fb | 1107 | unlock_mount_hash(); |
b105e270 | 1108 | return &mnt->mnt; |
9d412a43 | 1109 | } |
8f291889 AV |
1110 | EXPORT_SYMBOL(vfs_create_mount); |
1111 | ||
1112 | struct vfsmount *fc_mount(struct fs_context *fc) | |
1113 | { | |
1114 | int err = vfs_get_tree(fc); | |
1115 | if (!err) { | |
1116 | up_write(&fc->root->d_sb->s_umount); | |
1117 | return vfs_create_mount(fc); | |
1118 | } | |
1119 | return ERR_PTR(err); | |
1120 | } | |
1121 | EXPORT_SYMBOL(fc_mount); | |
1122 | ||
9bc61ab1 DH |
1123 | struct vfsmount *vfs_kern_mount(struct file_system_type *type, |
1124 | int flags, const char *name, | |
1125 | void *data) | |
9d412a43 | 1126 | { |
9bc61ab1 | 1127 | struct fs_context *fc; |
8f291889 | 1128 | struct vfsmount *mnt; |
9bc61ab1 | 1129 | int ret = 0; |
9d412a43 AV |
1130 | |
1131 | if (!type) | |
3e1aeb00 | 1132 | return ERR_PTR(-EINVAL); |
9d412a43 | 1133 | |
9bc61ab1 DH |
1134 | fc = fs_context_for_mount(type, flags); |
1135 | if (IS_ERR(fc)) | |
1136 | return ERR_CAST(fc); | |
1137 | ||
3e1aeb00 DH |
1138 | if (name) |
1139 | ret = vfs_parse_fs_string(fc, "source", | |
1140 | name, strlen(name)); | |
9bc61ab1 DH |
1141 | if (!ret) |
1142 | ret = parse_monolithic_mount_data(fc, data); | |
1143 | if (!ret) | |
8f291889 AV |
1144 | mnt = fc_mount(fc); |
1145 | else | |
1146 | mnt = ERR_PTR(ret); | |
9d412a43 | 1147 | |
9bc61ab1 | 1148 | put_fs_context(fc); |
8f291889 | 1149 | return mnt; |
9d412a43 AV |
1150 | } |
1151 | EXPORT_SYMBOL_GPL(vfs_kern_mount); | |
1152 | ||
93faccbb EB |
1153 | struct vfsmount * |
1154 | vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, | |
1155 | const char *name, void *data) | |
1156 | { | |
1157 | /* Until it is worked out how to pass the user namespace | |
1158 | * through from the parent mount to the submount don't support | |
1159 | * unprivileged mounts with submounts. | |
1160 | */ | |
1161 | if (mountpoint->d_sb->s_user_ns != &init_user_ns) | |
1162 | return ERR_PTR(-EPERM); | |
1163 | ||
e462ec50 | 1164 | return vfs_kern_mount(type, SB_SUBMOUNT, name, data); |
93faccbb EB |
1165 | } |
1166 | EXPORT_SYMBOL_GPL(vfs_submount); | |
1167 | ||
87129cc0 | 1168 | static struct mount *clone_mnt(struct mount *old, struct dentry *root, |
36341f64 | 1169 | int flag) |
1da177e4 | 1170 | { |
87129cc0 | 1171 | struct super_block *sb = old->mnt.mnt_sb; |
be34d1a3 DH |
1172 | struct mount *mnt; |
1173 | int err; | |
1da177e4 | 1174 | |
be34d1a3 DH |
1175 | mnt = alloc_vfsmnt(old->mnt_devname); |
1176 | if (!mnt) | |
1177 | return ERR_PTR(-ENOMEM); | |
719f5d7f | 1178 | |
7a472ef4 | 1179 | if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) |
be34d1a3 DH |
1180 | mnt->mnt_group_id = 0; /* not a peer of original */ |
1181 | else | |
1182 | mnt->mnt_group_id = old->mnt_group_id; | |
b90fa9ae | 1183 | |
be34d1a3 DH |
1184 | if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { |
1185 | err = mnt_alloc_group_id(mnt); | |
1186 | if (err) | |
1187 | goto out_free; | |
1da177e4 | 1188 | } |
be34d1a3 | 1189 | |
16a34adb AV |
1190 | mnt->mnt.mnt_flags = old->mnt.mnt_flags; |
1191 | mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); | |
5ff9d8a6 | 1192 | |
be34d1a3 | 1193 | atomic_inc(&sb->s_active); |
256c8aed CB |
1194 | mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt)); |
1195 | ||
be34d1a3 DH |
1196 | mnt->mnt.mnt_sb = sb; |
1197 | mnt->mnt.mnt_root = dget(root); | |
1198 | mnt->mnt_mountpoint = mnt->mnt.mnt_root; | |
1199 | mnt->mnt_parent = mnt; | |
719ea2fb | 1200 | lock_mount_hash(); |
be34d1a3 | 1201 | list_add_tail(&mnt->mnt_instance, &sb->s_mounts); |
719ea2fb | 1202 | unlock_mount_hash(); |
be34d1a3 | 1203 | |
7a472ef4 EB |
1204 | if ((flag & CL_SLAVE) || |
1205 | ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { | |
be34d1a3 DH |
1206 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); |
1207 | mnt->mnt_master = old; | |
1208 | CLEAR_MNT_SHARED(mnt); | |
1209 | } else if (!(flag & CL_PRIVATE)) { | |
1210 | if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) | |
1211 | list_add(&mnt->mnt_share, &old->mnt_share); | |
1212 | if (IS_MNT_SLAVE(old)) | |
1213 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
1214 | mnt->mnt_master = old->mnt_master; | |
5235d448 AV |
1215 | } else { |
1216 | CLEAR_MNT_SHARED(mnt); | |
be34d1a3 DH |
1217 | } |
1218 | if (flag & CL_MAKE_SHARED) | |
1219 | set_mnt_shared(mnt); | |
1220 | ||
1221 | /* stick the duplicate mount on the same expiry list | |
1222 | * as the original if that was on one */ | |
1223 | if (flag & CL_EXPIRE) { | |
1224 | if (!list_empty(&old->mnt_expire)) | |
1225 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
1226 | } | |
1227 | ||
cb338d06 | 1228 | return mnt; |
719f5d7f MS |
1229 | |
1230 | out_free: | |
8ffcb32e | 1231 | mnt_free_id(mnt); |
719f5d7f | 1232 | free_vfsmnt(mnt); |
be34d1a3 | 1233 | return ERR_PTR(err); |
1da177e4 LT |
1234 | } |
1235 | ||
9ea459e1 AV |
1236 | static void cleanup_mnt(struct mount *mnt) |
1237 | { | |
56cbb429 AV |
1238 | struct hlist_node *p; |
1239 | struct mount *m; | |
9ea459e1 | 1240 | /* |
56cbb429 AV |
1241 | * The warning here probably indicates that somebody messed |
1242 | * up a mnt_want/drop_write() pair. If this happens, the | |
1243 | * filesystem was probably unable to make r/w->r/o transitions. | |
9ea459e1 AV |
1244 | * The locking used to deal with mnt_count decrement provides barriers, |
1245 | * so mnt_get_writers() below is safe. | |
1246 | */ | |
1247 | WARN_ON(mnt_get_writers(mnt)); | |
1248 | if (unlikely(mnt->mnt_pins.first)) | |
1249 | mnt_pin_kill(mnt); | |
56cbb429 AV |
1250 | hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { |
1251 | hlist_del(&m->mnt_umount); | |
1252 | mntput(&m->mnt); | |
1253 | } | |
9ea459e1 AV |
1254 | fsnotify_vfsmount_delete(&mnt->mnt); |
1255 | dput(mnt->mnt.mnt_root); | |
1256 | deactivate_super(mnt->mnt.mnt_sb); | |
1257 | mnt_free_id(mnt); | |
1258 | call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); | |
1259 | } | |
1260 | ||
1261 | static void __cleanup_mnt(struct rcu_head *head) | |
1262 | { | |
1263 | cleanup_mnt(container_of(head, struct mount, mnt_rcu)); | |
1264 | } | |
1265 | ||
1266 | static LLIST_HEAD(delayed_mntput_list); | |
1267 | static void delayed_mntput(struct work_struct *unused) | |
1268 | { | |
1269 | struct llist_node *node = llist_del_all(&delayed_mntput_list); | |
29785735 | 1270 | struct mount *m, *t; |
9ea459e1 | 1271 | |
29785735 BP |
1272 | llist_for_each_entry_safe(m, t, node, mnt_llist) |
1273 | cleanup_mnt(m); | |
9ea459e1 AV |
1274 | } |
1275 | static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); | |
1276 | ||
900148dc | 1277 | static void mntput_no_expire(struct mount *mnt) |
b3e19d92 | 1278 | { |
4edbe133 | 1279 | LIST_HEAD(list); |
edf7ddbf | 1280 | int count; |
4edbe133 | 1281 | |
48a066e7 | 1282 | rcu_read_lock(); |
9ea0a46c AV |
1283 | if (likely(READ_ONCE(mnt->mnt_ns))) { |
1284 | /* | |
1285 | * Since we don't do lock_mount_hash() here, | |
1286 | * ->mnt_ns can change under us. However, if it's | |
1287 | * non-NULL, then there's a reference that won't | |
1288 | * be dropped until after an RCU delay done after | |
1289 | * turning ->mnt_ns NULL. So if we observe it | |
1290 | * non-NULL under rcu_read_lock(), the reference | |
1291 | * we are dropping is not the final one. | |
1292 | */ | |
1293 | mnt_add_count(mnt, -1); | |
48a066e7 | 1294 | rcu_read_unlock(); |
f03c6599 | 1295 | return; |
b3e19d92 | 1296 | } |
719ea2fb | 1297 | lock_mount_hash(); |
119e1ef8 AV |
1298 | /* |
1299 | * make sure that if __legitimize_mnt() has not seen us grab | |
1300 | * mount_lock, we'll see their refcount increment here. | |
1301 | */ | |
1302 | smp_mb(); | |
9ea0a46c | 1303 | mnt_add_count(mnt, -1); |
edf7ddbf EB |
1304 | count = mnt_get_count(mnt); |
1305 | if (count != 0) { | |
1306 | WARN_ON(count < 0); | |
48a066e7 | 1307 | rcu_read_unlock(); |
719ea2fb | 1308 | unlock_mount_hash(); |
99b7db7b NP |
1309 | return; |
1310 | } | |
48a066e7 AV |
1311 | if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { |
1312 | rcu_read_unlock(); | |
1313 | unlock_mount_hash(); | |
1314 | return; | |
1315 | } | |
1316 | mnt->mnt.mnt_flags |= MNT_DOOMED; | |
1317 | rcu_read_unlock(); | |
962830df | 1318 | |
39f7c4db | 1319 | list_del(&mnt->mnt_instance); |
ce07d891 EB |
1320 | |
1321 | if (unlikely(!list_empty(&mnt->mnt_mounts))) { | |
1322 | struct mount *p, *tmp; | |
1323 | list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { | |
4edbe133 | 1324 | __put_mountpoint(unhash_mnt(p), &list); |
56cbb429 | 1325 | hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); |
ce07d891 EB |
1326 | } |
1327 | } | |
719ea2fb | 1328 | unlock_mount_hash(); |
4edbe133 | 1329 | shrink_dentry_list(&list); |
649a795a | 1330 | |
9ea459e1 AV |
1331 | if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { |
1332 | struct task_struct *task = current; | |
1333 | if (likely(!(task->flags & PF_KTHREAD))) { | |
1334 | init_task_work(&mnt->mnt_rcu, __cleanup_mnt); | |
91989c70 | 1335 | if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) |
9ea459e1 AV |
1336 | return; |
1337 | } | |
1338 | if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) | |
1339 | schedule_delayed_work(&delayed_mntput_work, 1); | |
1340 | return; | |
1341 | } | |
1342 | cleanup_mnt(mnt); | |
b3e19d92 | 1343 | } |
b3e19d92 NP |
1344 | |
1345 | void mntput(struct vfsmount *mnt) | |
1346 | { | |
1347 | if (mnt) { | |
863d684f | 1348 | struct mount *m = real_mount(mnt); |
6c4d1c99 | 1349 | /* avoid cacheline pingpong */ |
863d684f | 1350 | if (unlikely(m->mnt_expiry_mark)) |
6c4d1c99 | 1351 | WRITE_ONCE(m->mnt_expiry_mark, 0); |
863d684f | 1352 | mntput_no_expire(m); |
b3e19d92 NP |
1353 | } |
1354 | } | |
1355 | EXPORT_SYMBOL(mntput); | |
1356 | ||
1357 | struct vfsmount *mntget(struct vfsmount *mnt) | |
1358 | { | |
1359 | if (mnt) | |
83adc753 | 1360 | mnt_add_count(real_mount(mnt), 1); |
b3e19d92 NP |
1361 | return mnt; |
1362 | } | |
1363 | EXPORT_SYMBOL(mntget); | |
1364 | ||
da27f796 RR |
1365 | /* |
1366 | * Make a mount point inaccessible to new lookups. | |
1367 | * Because there may still be current users, the caller MUST WAIT | |
1368 | * for an RCU grace period before destroying the mount point. | |
1369 | */ | |
1370 | void mnt_make_shortterm(struct vfsmount *mnt) | |
1371 | { | |
1372 | if (mnt) | |
1373 | real_mount(mnt)->mnt_ns = NULL; | |
1374 | } | |
1375 | ||
1f287bc4 RD |
1376 | /** |
1377 | * path_is_mountpoint() - Check if path is a mount in the current namespace. | |
1378 | * @path: path to check | |
c6609c0a IK |
1379 | * |
1380 | * d_mountpoint() can only be used reliably to establish if a dentry is | |
1381 | * not mounted in any namespace and that common case is handled inline. | |
1382 | * d_mountpoint() isn't aware of the possibility there may be multiple | |
1383 | * mounts using a given dentry in a different namespace. This function | |
1384 | * checks if the passed in path is a mountpoint rather than the dentry | |
1385 | * alone. | |
1386 | */ | |
1387 | bool path_is_mountpoint(const struct path *path) | |
1388 | { | |
1389 | unsigned seq; | |
1390 | bool res; | |
1391 | ||
1392 | if (!d_mountpoint(path->dentry)) | |
1393 | return false; | |
1394 | ||
1395 | rcu_read_lock(); | |
1396 | do { | |
1397 | seq = read_seqbegin(&mount_lock); | |
1398 | res = __path_is_mountpoint(path); | |
1399 | } while (read_seqretry(&mount_lock, seq)); | |
1400 | rcu_read_unlock(); | |
1401 | ||
1402 | return res; | |
1403 | } | |
1404 | EXPORT_SYMBOL(path_is_mountpoint); | |
1405 | ||
ca71cf71 | 1406 | struct vfsmount *mnt_clone_internal(const struct path *path) |
7b7b1ace | 1407 | { |
3064c356 AV |
1408 | struct mount *p; |
1409 | p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); | |
1410 | if (IS_ERR(p)) | |
1411 | return ERR_CAST(p); | |
1412 | p->mnt.mnt_flags |= MNT_INTERNAL; | |
1413 | return &p->mnt; | |
7b7b1ace | 1414 | } |
1da177e4 | 1415 | |
a1a2c409 | 1416 | #ifdef CONFIG_PROC_FS |
9f6c61f9 MS |
1417 | static struct mount *mnt_list_next(struct mnt_namespace *ns, |
1418 | struct list_head *p) | |
1419 | { | |
1420 | struct mount *mnt, *ret = NULL; | |
1421 | ||
1422 | lock_ns_list(ns); | |
1423 | list_for_each_continue(p, &ns->list) { | |
1424 | mnt = list_entry(p, typeof(*mnt), mnt_list); | |
1425 | if (!mnt_is_cursor(mnt)) { | |
1426 | ret = mnt; | |
1427 | break; | |
1428 | } | |
1429 | } | |
1430 | unlock_ns_list(ns); | |
1431 | ||
1432 | return ret; | |
1433 | } | |
1434 | ||
0226f492 | 1435 | /* iterator; we want it to have access to namespace_sem, thus here... */ |
1da177e4 LT |
1436 | static void *m_start(struct seq_file *m, loff_t *pos) |
1437 | { | |
ede1bf0d | 1438 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1439 | struct list_head *prev; |
1da177e4 | 1440 | |
390c6843 | 1441 | down_read(&namespace_sem); |
9f6c61f9 MS |
1442 | if (!*pos) { |
1443 | prev = &p->ns->list; | |
1444 | } else { | |
1445 | prev = &p->cursor.mnt_list; | |
1446 | ||
1447 | /* Read after we'd reached the end? */ | |
1448 | if (list_empty(prev)) | |
1449 | return NULL; | |
c7999c36 AV |
1450 | } |
1451 | ||
9f6c61f9 | 1452 | return mnt_list_next(p->ns, prev); |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
1456 | { | |
ede1bf0d | 1457 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1458 | struct mount *mnt = v; |
b0765fb8 | 1459 | |
9f6c61f9 MS |
1460 | ++*pos; |
1461 | return mnt_list_next(p->ns, &mnt->mnt_list); | |
1da177e4 LT |
1462 | } |
1463 | ||
1464 | static void m_stop(struct seq_file *m, void *v) | |
1465 | { | |
9f6c61f9 MS |
1466 | struct proc_mounts *p = m->private; |
1467 | struct mount *mnt = v; | |
1468 | ||
1469 | lock_ns_list(p->ns); | |
1470 | if (mnt) | |
1471 | list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); | |
1472 | else | |
1473 | list_del_init(&p->cursor.mnt_list); | |
1474 | unlock_ns_list(p->ns); | |
390c6843 | 1475 | up_read(&namespace_sem); |
1da177e4 LT |
1476 | } |
1477 | ||
0226f492 | 1478 | static int m_show(struct seq_file *m, void *v) |
2d4d4864 | 1479 | { |
ede1bf0d | 1480 | struct proc_mounts *p = m->private; |
9f6c61f9 | 1481 | struct mount *r = v; |
0226f492 | 1482 | return p->show(m, &r->mnt); |
1da177e4 LT |
1483 | } |
1484 | ||
a1a2c409 | 1485 | const struct seq_operations mounts_op = { |
1da177e4 LT |
1486 | .start = m_start, |
1487 | .next = m_next, | |
1488 | .stop = m_stop, | |
0226f492 | 1489 | .show = m_show, |
b4629fe2 | 1490 | }; |
9f6c61f9 MS |
1491 | |
1492 | void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) | |
1493 | { | |
1494 | down_read(&namespace_sem); | |
1495 | lock_ns_list(ns); | |
1496 | list_del(&cursor->mnt_list); | |
1497 | unlock_ns_list(ns); | |
1498 | up_read(&namespace_sem); | |
1499 | } | |
a1a2c409 | 1500 | #endif /* CONFIG_PROC_FS */ |
b4629fe2 | 1501 | |
1da177e4 LT |
1502 | /** |
1503 | * may_umount_tree - check if a mount tree is busy | |
1f287bc4 | 1504 | * @m: root of mount tree |
1da177e4 LT |
1505 | * |
1506 | * This is called to check if a tree of mounts has any | |
1507 | * open files, pwds, chroots or sub mounts that are | |
1508 | * busy. | |
1509 | */ | |
909b0a88 | 1510 | int may_umount_tree(struct vfsmount *m) |
1da177e4 | 1511 | { |
909b0a88 | 1512 | struct mount *mnt = real_mount(m); |
36341f64 RP |
1513 | int actual_refs = 0; |
1514 | int minimum_refs = 0; | |
315fc83e | 1515 | struct mount *p; |
909b0a88 | 1516 | BUG_ON(!m); |
1da177e4 | 1517 | |
b3e19d92 | 1518 | /* write lock needed for mnt_get_count */ |
719ea2fb | 1519 | lock_mount_hash(); |
909b0a88 | 1520 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
83adc753 | 1521 | actual_refs += mnt_get_count(p); |
1da177e4 | 1522 | minimum_refs += 2; |
1da177e4 | 1523 | } |
719ea2fb | 1524 | unlock_mount_hash(); |
1da177e4 LT |
1525 | |
1526 | if (actual_refs > minimum_refs) | |
e3474a8e | 1527 | return 0; |
1da177e4 | 1528 | |
e3474a8e | 1529 | return 1; |
1da177e4 LT |
1530 | } |
1531 | ||
1532 | EXPORT_SYMBOL(may_umount_tree); | |
1533 | ||
1534 | /** | |
1535 | * may_umount - check if a mount point is busy | |
1536 | * @mnt: root of mount | |
1537 | * | |
1538 | * This is called to check if a mount point has any | |
1539 | * open files, pwds, chroots or sub mounts. If the | |
1540 | * mount has sub mounts this will return busy | |
1541 | * regardless of whether the sub mounts are busy. | |
1542 | * | |
1543 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
1544 | * give false negatives. The main reason why it's here is that we need | |
1545 | * a non-destructive way to look for easily umountable filesystems. | |
1546 | */ | |
1547 | int may_umount(struct vfsmount *mnt) | |
1548 | { | |
e3474a8e | 1549 | int ret = 1; |
8ad08d8a | 1550 | down_read(&namespace_sem); |
719ea2fb | 1551 | lock_mount_hash(); |
1ab59738 | 1552 | if (propagate_mount_busy(real_mount(mnt), 2)) |
e3474a8e | 1553 | ret = 0; |
719ea2fb | 1554 | unlock_mount_hash(); |
8ad08d8a | 1555 | up_read(&namespace_sem); |
a05964f3 | 1556 | return ret; |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | EXPORT_SYMBOL(may_umount); | |
1560 | ||
97216be0 | 1561 | static void namespace_unlock(void) |
70fbcdf4 | 1562 | { |
a3b3c562 | 1563 | struct hlist_head head; |
56cbb429 AV |
1564 | struct hlist_node *p; |
1565 | struct mount *m; | |
4edbe133 | 1566 | LIST_HEAD(list); |
97216be0 | 1567 | |
a3b3c562 | 1568 | hlist_move_list(&unmounted, &head); |
4edbe133 | 1569 | list_splice_init(&ex_mountpoints, &list); |
97216be0 | 1570 | |
97216be0 AV |
1571 | up_write(&namespace_sem); |
1572 | ||
4edbe133 AV |
1573 | shrink_dentry_list(&list); |
1574 | ||
a3b3c562 EB |
1575 | if (likely(hlist_empty(&head))) |
1576 | return; | |
1577 | ||
22cb7405 | 1578 | synchronize_rcu_expedited(); |
48a066e7 | 1579 | |
56cbb429 AV |
1580 | hlist_for_each_entry_safe(m, p, &head, mnt_umount) { |
1581 | hlist_del(&m->mnt_umount); | |
1582 | mntput(&m->mnt); | |
1583 | } | |
70fbcdf4 RP |
1584 | } |
1585 | ||
97216be0 | 1586 | static inline void namespace_lock(void) |
e3197d83 | 1587 | { |
97216be0 | 1588 | down_write(&namespace_sem); |
e3197d83 AV |
1589 | } |
1590 | ||
e819f152 EB |
1591 | enum umount_tree_flags { |
1592 | UMOUNT_SYNC = 1, | |
1593 | UMOUNT_PROPAGATE = 2, | |
e0c9c0af | 1594 | UMOUNT_CONNECTED = 4, |
e819f152 | 1595 | }; |
f2d0a123 EB |
1596 | |
1597 | static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) | |
1598 | { | |
1599 | /* Leaving mounts connected is only valid for lazy umounts */ | |
1600 | if (how & UMOUNT_SYNC) | |
1601 | return true; | |
1602 | ||
1603 | /* A mount without a parent has nothing to be connected to */ | |
1604 | if (!mnt_has_parent(mnt)) | |
1605 | return true; | |
1606 | ||
1607 | /* Because the reference counting rules change when mounts are | |
1608 | * unmounted and connected, umounted mounts may not be | |
1609 | * connected to mounted mounts. | |
1610 | */ | |
1611 | if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) | |
1612 | return true; | |
1613 | ||
1614 | /* Has it been requested that the mount remain connected? */ | |
1615 | if (how & UMOUNT_CONNECTED) | |
1616 | return false; | |
1617 | ||
1618 | /* Is the mount locked such that it needs to remain connected? */ | |
1619 | if (IS_MNT_LOCKED(mnt)) | |
1620 | return false; | |
1621 | ||
1622 | /* By default disconnect the mount */ | |
1623 | return true; | |
1624 | } | |
1625 | ||
99b7db7b | 1626 | /* |
48a066e7 | 1627 | * mount_lock must be held |
99b7db7b NP |
1628 | * namespace_sem must be held for write |
1629 | */ | |
e819f152 | 1630 | static void umount_tree(struct mount *mnt, enum umount_tree_flags how) |
1da177e4 | 1631 | { |
c003b26f | 1632 | LIST_HEAD(tmp_list); |
315fc83e | 1633 | struct mount *p; |
1da177e4 | 1634 | |
5d88457e EB |
1635 | if (how & UMOUNT_PROPAGATE) |
1636 | propagate_mount_unlock(mnt); | |
1637 | ||
c003b26f | 1638 | /* Gather the mounts to umount */ |
590ce4bc EB |
1639 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1640 | p->mnt.mnt_flags |= MNT_UMOUNT; | |
c003b26f | 1641 | list_move(&p->mnt_list, &tmp_list); |
590ce4bc | 1642 | } |
1da177e4 | 1643 | |
411a938b | 1644 | /* Hide the mounts from mnt_mounts */ |
c003b26f | 1645 | list_for_each_entry(p, &tmp_list, mnt_list) { |
88b368f2 | 1646 | list_del_init(&p->mnt_child); |
c003b26f | 1647 | } |
88b368f2 | 1648 | |
c003b26f | 1649 | /* Add propogated mounts to the tmp_list */ |
e819f152 | 1650 | if (how & UMOUNT_PROPAGATE) |
7b8a53fd | 1651 | propagate_umount(&tmp_list); |
a05964f3 | 1652 | |
c003b26f | 1653 | while (!list_empty(&tmp_list)) { |
d2921684 | 1654 | struct mnt_namespace *ns; |
ce07d891 | 1655 | bool disconnect; |
c003b26f | 1656 | p = list_first_entry(&tmp_list, struct mount, mnt_list); |
6776db3d | 1657 | list_del_init(&p->mnt_expire); |
1a4eeaf2 | 1658 | list_del_init(&p->mnt_list); |
d2921684 EB |
1659 | ns = p->mnt_ns; |
1660 | if (ns) { | |
1661 | ns->mounts--; | |
1662 | __touch_mnt_namespace(ns); | |
1663 | } | |
143c8c91 | 1664 | p->mnt_ns = NULL; |
e819f152 | 1665 | if (how & UMOUNT_SYNC) |
48a066e7 | 1666 | p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; |
87b95ce0 | 1667 | |
f2d0a123 | 1668 | disconnect = disconnect_mount(p, how); |
676da58d | 1669 | if (mnt_has_parent(p)) { |
81b6b061 | 1670 | mnt_add_count(p->mnt_parent, -1); |
ce07d891 EB |
1671 | if (!disconnect) { |
1672 | /* Don't forget about p */ | |
1673 | list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); | |
1674 | } else { | |
1675 | umount_mnt(p); | |
1676 | } | |
7c4b93d8 | 1677 | } |
0f0afb1d | 1678 | change_mnt_propagation(p, MS_PRIVATE); |
19a1c409 AV |
1679 | if (disconnect) |
1680 | hlist_add_head(&p->mnt_umount, &unmounted); | |
1da177e4 LT |
1681 | } |
1682 | } | |
1683 | ||
b54b9be7 | 1684 | static void shrink_submounts(struct mount *mnt); |
c35038be | 1685 | |
8d0347f6 DH |
1686 | static int do_umount_root(struct super_block *sb) |
1687 | { | |
1688 | int ret = 0; | |
1689 | ||
1690 | down_write(&sb->s_umount); | |
1691 | if (!sb_rdonly(sb)) { | |
1692 | struct fs_context *fc; | |
1693 | ||
1694 | fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, | |
1695 | SB_RDONLY); | |
1696 | if (IS_ERR(fc)) { | |
1697 | ret = PTR_ERR(fc); | |
1698 | } else { | |
1699 | ret = parse_monolithic_mount_data(fc, NULL); | |
1700 | if (!ret) | |
1701 | ret = reconfigure_super(fc); | |
1702 | put_fs_context(fc); | |
1703 | } | |
1704 | } | |
1705 | up_write(&sb->s_umount); | |
1706 | return ret; | |
1707 | } | |
1708 | ||
1ab59738 | 1709 | static int do_umount(struct mount *mnt, int flags) |
1da177e4 | 1710 | { |
1ab59738 | 1711 | struct super_block *sb = mnt->mnt.mnt_sb; |
1da177e4 LT |
1712 | int retval; |
1713 | ||
1ab59738 | 1714 | retval = security_sb_umount(&mnt->mnt, flags); |
1da177e4 LT |
1715 | if (retval) |
1716 | return retval; | |
1717 | ||
1718 | /* | |
1719 | * Allow userspace to request a mountpoint be expired rather than | |
1720 | * unmounting unconditionally. Unmount only happens if: | |
1721 | * (1) the mark is already set (the mark is cleared by mntput()) | |
1722 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
1723 | */ | |
1724 | if (flags & MNT_EXPIRE) { | |
1ab59738 | 1725 | if (&mnt->mnt == current->fs->root.mnt || |
1da177e4 LT |
1726 | flags & (MNT_FORCE | MNT_DETACH)) |
1727 | return -EINVAL; | |
1728 | ||
b3e19d92 NP |
1729 | /* |
1730 | * probably don't strictly need the lock here if we examined | |
1731 | * all race cases, but it's a slowpath. | |
1732 | */ | |
719ea2fb | 1733 | lock_mount_hash(); |
83adc753 | 1734 | if (mnt_get_count(mnt) != 2) { |
719ea2fb | 1735 | unlock_mount_hash(); |
1da177e4 | 1736 | return -EBUSY; |
b3e19d92 | 1737 | } |
719ea2fb | 1738 | unlock_mount_hash(); |
1da177e4 | 1739 | |
863d684f | 1740 | if (!xchg(&mnt->mnt_expiry_mark, 1)) |
1da177e4 LT |
1741 | return -EAGAIN; |
1742 | } | |
1743 | ||
1744 | /* | |
1745 | * If we may have to abort operations to get out of this | |
1746 | * mount, and they will themselves hold resources we must | |
1747 | * allow the fs to do things. In the Unix tradition of | |
1748 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
1749 | * might fail to complete on the first run through as other tasks | |
1750 | * must return, and the like. Thats for the mount program to worry | |
1751 | * about for the moment. | |
1752 | */ | |
1753 | ||
42faad99 | 1754 | if (flags & MNT_FORCE && sb->s_op->umount_begin) { |
42faad99 | 1755 | sb->s_op->umount_begin(sb); |
42faad99 | 1756 | } |
1da177e4 LT |
1757 | |
1758 | /* | |
1759 | * No sense to grab the lock for this test, but test itself looks | |
1760 | * somewhat bogus. Suggestions for better replacement? | |
1761 | * Ho-hum... In principle, we might treat that as umount + switch | |
1762 | * to rootfs. GC would eventually take care of the old vfsmount. | |
1763 | * Actually it makes sense, especially if rootfs would contain a | |
1764 | * /reboot - static binary that would close all descriptors and | |
1765 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
1766 | */ | |
1ab59738 | 1767 | if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { |
1da177e4 LT |
1768 | /* |
1769 | * Special case for "unmounting" root ... | |
1770 | * we just try to remount it readonly. | |
1771 | */ | |
bc6155d1 | 1772 | if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) |
a1480dcc | 1773 | return -EPERM; |
8d0347f6 | 1774 | return do_umount_root(sb); |
1da177e4 LT |
1775 | } |
1776 | ||
97216be0 | 1777 | namespace_lock(); |
719ea2fb | 1778 | lock_mount_hash(); |
1da177e4 | 1779 | |
25d202ed EB |
1780 | /* Recheck MNT_LOCKED with the locks held */ |
1781 | retval = -EINVAL; | |
1782 | if (mnt->mnt.mnt_flags & MNT_LOCKED) | |
1783 | goto out; | |
1784 | ||
1785 | event++; | |
48a066e7 | 1786 | if (flags & MNT_DETACH) { |
1a4eeaf2 | 1787 | if (!list_empty(&mnt->mnt_list)) |
e819f152 | 1788 | umount_tree(mnt, UMOUNT_PROPAGATE); |
1da177e4 | 1789 | retval = 0; |
48a066e7 AV |
1790 | } else { |
1791 | shrink_submounts(mnt); | |
1792 | retval = -EBUSY; | |
1793 | if (!propagate_mount_busy(mnt, 2)) { | |
1794 | if (!list_empty(&mnt->mnt_list)) | |
e819f152 | 1795 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
48a066e7 AV |
1796 | retval = 0; |
1797 | } | |
1da177e4 | 1798 | } |
25d202ed | 1799 | out: |
719ea2fb | 1800 | unlock_mount_hash(); |
e3197d83 | 1801 | namespace_unlock(); |
1da177e4 LT |
1802 | return retval; |
1803 | } | |
1804 | ||
80b5dce8 EB |
1805 | /* |
1806 | * __detach_mounts - lazily unmount all mounts on the specified dentry | |
1807 | * | |
1808 | * During unlink, rmdir, and d_drop it is possible to loose the path | |
1809 | * to an existing mountpoint, and wind up leaking the mount. | |
1810 | * detach_mounts allows lazily unmounting those mounts instead of | |
1811 | * leaking them. | |
1812 | * | |
1813 | * The caller may hold dentry->d_inode->i_mutex. | |
1814 | */ | |
1815 | void __detach_mounts(struct dentry *dentry) | |
1816 | { | |
1817 | struct mountpoint *mp; | |
1818 | struct mount *mnt; | |
1819 | ||
1820 | namespace_lock(); | |
3895dbf8 | 1821 | lock_mount_hash(); |
80b5dce8 | 1822 | mp = lookup_mountpoint(dentry); |
adc9b5c0 | 1823 | if (!mp) |
80b5dce8 EB |
1824 | goto out_unlock; |
1825 | ||
e06b933e | 1826 | event++; |
80b5dce8 EB |
1827 | while (!hlist_empty(&mp->m_list)) { |
1828 | mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); | |
ce07d891 | 1829 | if (mnt->mnt.mnt_flags & MNT_UMOUNT) { |
fe78fcc8 | 1830 | umount_mnt(mnt); |
56cbb429 | 1831 | hlist_add_head(&mnt->mnt_umount, &unmounted); |
ce07d891 | 1832 | } |
e0c9c0af | 1833 | else umount_tree(mnt, UMOUNT_CONNECTED); |
80b5dce8 | 1834 | } |
80b5dce8 EB |
1835 | put_mountpoint(mp); |
1836 | out_unlock: | |
3895dbf8 | 1837 | unlock_mount_hash(); |
80b5dce8 EB |
1838 | namespace_unlock(); |
1839 | } | |
1840 | ||
dd111b31 | 1841 | /* |
9b40bc90 AV |
1842 | * Is the caller allowed to modify his namespace? |
1843 | */ | |
a5f85d78 | 1844 | bool may_mount(void) |
9b40bc90 AV |
1845 | { |
1846 | return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); | |
1847 | } | |
1848 | ||
78aa08a8 CB |
1849 | /** |
1850 | * path_mounted - check whether path is mounted | |
1851 | * @path: path to check | |
1852 | * | |
1853 | * Determine whether @path refers to the root of a mount. | |
1854 | * | |
1855 | * Return: true if @path is the root of a mount, false if not. | |
1856 | */ | |
1857 | static inline bool path_mounted(const struct path *path) | |
1858 | { | |
1859 | return path->mnt->mnt_root == path->dentry; | |
1860 | } | |
1861 | ||
f7e33bdb | 1862 | static void warn_mandlock(void) |
9e8925b6 | 1863 | { |
f7e33bdb JL |
1864 | pr_warn_once("=======================================================\n" |
1865 | "WARNING: The mand mount option has been deprecated and\n" | |
1866 | " and is ignored by this kernel. Remove the mand\n" | |
1867 | " option from the mount to silence this warning.\n" | |
1868 | "=======================================================\n"); | |
9e8925b6 JL |
1869 | } |
1870 | ||
25ccd24f | 1871 | static int can_umount(const struct path *path, int flags) |
1da177e4 | 1872 | { |
25ccd24f | 1873 | struct mount *mnt = real_mount(path->mnt); |
1da177e4 | 1874 | |
9b40bc90 AV |
1875 | if (!may_mount()) |
1876 | return -EPERM; | |
78aa08a8 | 1877 | if (!path_mounted(path)) |
25ccd24f | 1878 | return -EINVAL; |
143c8c91 | 1879 | if (!check_mnt(mnt)) |
25ccd24f | 1880 | return -EINVAL; |
25d202ed | 1881 | if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ |
25ccd24f | 1882 | return -EINVAL; |
b2f5d4dc | 1883 | if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) |
25ccd24f CH |
1884 | return -EPERM; |
1885 | return 0; | |
1886 | } | |
1887 | ||
a0a6df9a | 1888 | // caller is responsible for flags being sane |
25ccd24f CH |
1889 | int path_umount(struct path *path, int flags) |
1890 | { | |
1891 | struct mount *mnt = real_mount(path->mnt); | |
1892 | int ret; | |
1893 | ||
1894 | ret = can_umount(path, flags); | |
1895 | if (!ret) | |
1896 | ret = do_umount(mnt, flags); | |
1da177e4 | 1897 | |
429731b1 | 1898 | /* we mustn't call path_put() as that would clear mnt_expiry_mark */ |
41525f56 | 1899 | dput(path->dentry); |
900148dc | 1900 | mntput_no_expire(mnt); |
25ccd24f | 1901 | return ret; |
1da177e4 LT |
1902 | } |
1903 | ||
09267def | 1904 | static int ksys_umount(char __user *name, int flags) |
41525f56 CH |
1905 | { |
1906 | int lookup_flags = LOOKUP_MOUNTPOINT; | |
1907 | struct path path; | |
1908 | int ret; | |
1909 | ||
a0a6df9a AV |
1910 | // basic validity checks done first |
1911 | if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) | |
1912 | return -EINVAL; | |
1913 | ||
41525f56 CH |
1914 | if (!(flags & UMOUNT_NOFOLLOW)) |
1915 | lookup_flags |= LOOKUP_FOLLOW; | |
1916 | ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); | |
1917 | if (ret) | |
1918 | return ret; | |
1919 | return path_umount(&path, flags); | |
1920 | } | |
1921 | ||
3a18ef5c DB |
1922 | SYSCALL_DEFINE2(umount, char __user *, name, int, flags) |
1923 | { | |
1924 | return ksys_umount(name, flags); | |
1925 | } | |
1926 | ||
1da177e4 LT |
1927 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT |
1928 | ||
1929 | /* | |
b58fed8b | 1930 | * The 2.0 compatible umount. No flags. |
1da177e4 | 1931 | */ |
bdc480e3 | 1932 | SYSCALL_DEFINE1(oldumount, char __user *, name) |
1da177e4 | 1933 | { |
3a18ef5c | 1934 | return ksys_umount(name, 0); |
1da177e4 LT |
1935 | } |
1936 | ||
1937 | #endif | |
1938 | ||
4ce5d2b1 | 1939 | static bool is_mnt_ns_file(struct dentry *dentry) |
8823c079 | 1940 | { |
4ce5d2b1 | 1941 | /* Is this a proxy for a mount namespace? */ |
e149ed2b AV |
1942 | return dentry->d_op == &ns_dentry_operations && |
1943 | dentry->d_fsdata == &mntns_operations; | |
4ce5d2b1 EB |
1944 | } |
1945 | ||
213921f9 | 1946 | static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) |
58be2825 AV |
1947 | { |
1948 | return container_of(ns, struct mnt_namespace, ns); | |
1949 | } | |
1950 | ||
303cc571 CB |
1951 | struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) |
1952 | { | |
1953 | return &mnt->ns; | |
1954 | } | |
1955 | ||
4ce5d2b1 EB |
1956 | static bool mnt_ns_loop(struct dentry *dentry) |
1957 | { | |
1958 | /* Could bind mounting the mount namespace inode cause a | |
1959 | * mount namespace loop? | |
1960 | */ | |
1961 | struct mnt_namespace *mnt_ns; | |
1962 | if (!is_mnt_ns_file(dentry)) | |
1963 | return false; | |
1964 | ||
f77c8014 | 1965 | mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); |
8823c079 EB |
1966 | return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; |
1967 | } | |
1968 | ||
87129cc0 | 1969 | struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, |
36341f64 | 1970 | int flag) |
1da177e4 | 1971 | { |
84d17192 | 1972 | struct mount *res, *p, *q, *r, *parent; |
1da177e4 | 1973 | |
4ce5d2b1 EB |
1974 | if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) |
1975 | return ERR_PTR(-EINVAL); | |
1976 | ||
1977 | if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) | |
be34d1a3 | 1978 | return ERR_PTR(-EINVAL); |
9676f0c6 | 1979 | |
36341f64 | 1980 | res = q = clone_mnt(mnt, dentry, flag); |
be34d1a3 DH |
1981 | if (IS_ERR(q)) |
1982 | return q; | |
1983 | ||
a73324da | 1984 | q->mnt_mountpoint = mnt->mnt_mountpoint; |
1da177e4 LT |
1985 | |
1986 | p = mnt; | |
6b41d536 | 1987 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
315fc83e | 1988 | struct mount *s; |
7ec02ef1 | 1989 | if (!is_subdir(r->mnt_mountpoint, dentry)) |
1da177e4 LT |
1990 | continue; |
1991 | ||
909b0a88 | 1992 | for (s = r; s; s = next_mnt(s, r)) { |
4ce5d2b1 EB |
1993 | if (!(flag & CL_COPY_UNBINDABLE) && |
1994 | IS_MNT_UNBINDABLE(s)) { | |
df7342b2 EB |
1995 | if (s->mnt.mnt_flags & MNT_LOCKED) { |
1996 | /* Both unbindable and locked. */ | |
1997 | q = ERR_PTR(-EPERM); | |
1998 | goto out; | |
1999 | } else { | |
2000 | s = skip_mnt_tree(s); | |
2001 | continue; | |
2002 | } | |
4ce5d2b1 EB |
2003 | } |
2004 | if (!(flag & CL_COPY_MNT_NS_FILE) && | |
2005 | is_mnt_ns_file(s->mnt.mnt_root)) { | |
9676f0c6 RP |
2006 | s = skip_mnt_tree(s); |
2007 | continue; | |
2008 | } | |
0714a533 AV |
2009 | while (p != s->mnt_parent) { |
2010 | p = p->mnt_parent; | |
2011 | q = q->mnt_parent; | |
1da177e4 | 2012 | } |
87129cc0 | 2013 | p = s; |
84d17192 | 2014 | parent = q; |
87129cc0 | 2015 | q = clone_mnt(p, p->mnt.mnt_root, flag); |
be34d1a3 DH |
2016 | if (IS_ERR(q)) |
2017 | goto out; | |
719ea2fb | 2018 | lock_mount_hash(); |
1a4eeaf2 | 2019 | list_add_tail(&q->mnt_list, &res->mnt_list); |
6ac39281 | 2020 | attach_mnt(q, parent, p->mnt_mp, false); |
719ea2fb | 2021 | unlock_mount_hash(); |
1da177e4 LT |
2022 | } |
2023 | } | |
2024 | return res; | |
be34d1a3 | 2025 | out: |
1da177e4 | 2026 | if (res) { |
719ea2fb | 2027 | lock_mount_hash(); |
e819f152 | 2028 | umount_tree(res, UMOUNT_SYNC); |
719ea2fb | 2029 | unlock_mount_hash(); |
1da177e4 | 2030 | } |
be34d1a3 | 2031 | return q; |
1da177e4 LT |
2032 | } |
2033 | ||
be34d1a3 DH |
2034 | /* Caller should check returned pointer for errors */ |
2035 | ||
ca71cf71 | 2036 | struct vfsmount *collect_mounts(const struct path *path) |
8aec0809 | 2037 | { |
cb338d06 | 2038 | struct mount *tree; |
97216be0 | 2039 | namespace_lock(); |
cd4a4017 EB |
2040 | if (!check_mnt(real_mount(path->mnt))) |
2041 | tree = ERR_PTR(-EINVAL); | |
2042 | else | |
2043 | tree = copy_tree(real_mount(path->mnt), path->dentry, | |
2044 | CL_COPY_ALL | CL_PRIVATE); | |
328e6d90 | 2045 | namespace_unlock(); |
be34d1a3 | 2046 | if (IS_ERR(tree)) |
52e220d3 | 2047 | return ERR_CAST(tree); |
be34d1a3 | 2048 | return &tree->mnt; |
8aec0809 AV |
2049 | } |
2050 | ||
a07b2000 AV |
2051 | static void free_mnt_ns(struct mnt_namespace *); |
2052 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); | |
2053 | ||
2054 | void dissolve_on_fput(struct vfsmount *mnt) | |
2055 | { | |
2056 | struct mnt_namespace *ns; | |
2057 | namespace_lock(); | |
2058 | lock_mount_hash(); | |
2059 | ns = real_mount(mnt)->mnt_ns; | |
44dfd84a DH |
2060 | if (ns) { |
2061 | if (is_anon_ns(ns)) | |
2062 | umount_tree(real_mount(mnt), UMOUNT_CONNECTED); | |
2063 | else | |
2064 | ns = NULL; | |
2065 | } | |
a07b2000 AV |
2066 | unlock_mount_hash(); |
2067 | namespace_unlock(); | |
44dfd84a DH |
2068 | if (ns) |
2069 | free_mnt_ns(ns); | |
a07b2000 AV |
2070 | } |
2071 | ||
8aec0809 AV |
2072 | void drop_collected_mounts(struct vfsmount *mnt) |
2073 | { | |
97216be0 | 2074 | namespace_lock(); |
719ea2fb | 2075 | lock_mount_hash(); |
9c8e0a1b | 2076 | umount_tree(real_mount(mnt), 0); |
719ea2fb | 2077 | unlock_mount_hash(); |
3ab6abee | 2078 | namespace_unlock(); |
8aec0809 AV |
2079 | } |
2080 | ||
427215d8 MS |
2081 | static bool has_locked_children(struct mount *mnt, struct dentry *dentry) |
2082 | { | |
2083 | struct mount *child; | |
2084 | ||
2085 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
2086 | if (!is_subdir(child->mnt_mountpoint, dentry)) | |
2087 | continue; | |
2088 | ||
2089 | if (child->mnt.mnt_flags & MNT_LOCKED) | |
2090 | return true; | |
2091 | } | |
2092 | return false; | |
2093 | } | |
2094 | ||
c771d683 MS |
2095 | /** |
2096 | * clone_private_mount - create a private clone of a path | |
1f287bc4 | 2097 | * @path: path to clone |
c771d683 | 2098 | * |
1f287bc4 RD |
2099 | * This creates a new vfsmount, which will be the clone of @path. The new mount |
2100 | * will not be attached anywhere in the namespace and will be private (i.e. | |
2101 | * changes to the originating mount won't be propagated into this). | |
c771d683 MS |
2102 | * |
2103 | * Release with mntput(). | |
2104 | */ | |
ca71cf71 | 2105 | struct vfsmount *clone_private_mount(const struct path *path) |
c771d683 MS |
2106 | { |
2107 | struct mount *old_mnt = real_mount(path->mnt); | |
2108 | struct mount *new_mnt; | |
2109 | ||
427215d8 | 2110 | down_read(&namespace_sem); |
c771d683 | 2111 | if (IS_MNT_UNBINDABLE(old_mnt)) |
427215d8 MS |
2112 | goto invalid; |
2113 | ||
2114 | if (!check_mnt(old_mnt)) | |
2115 | goto invalid; | |
2116 | ||
2117 | if (has_locked_children(old_mnt, path->dentry)) | |
2118 | goto invalid; | |
c771d683 | 2119 | |
c771d683 | 2120 | new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); |
427215d8 MS |
2121 | up_read(&namespace_sem); |
2122 | ||
c771d683 MS |
2123 | if (IS_ERR(new_mnt)) |
2124 | return ERR_CAST(new_mnt); | |
2125 | ||
df820f8d MS |
2126 | /* Longterm mount to be removed by kern_unmount*() */ |
2127 | new_mnt->mnt_ns = MNT_NS_INTERNAL; | |
2128 | ||
c771d683 | 2129 | return &new_mnt->mnt; |
427215d8 MS |
2130 | |
2131 | invalid: | |
2132 | up_read(&namespace_sem); | |
2133 | return ERR_PTR(-EINVAL); | |
c771d683 MS |
2134 | } |
2135 | EXPORT_SYMBOL_GPL(clone_private_mount); | |
2136 | ||
1f707137 AV |
2137 | int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, |
2138 | struct vfsmount *root) | |
2139 | { | |
1a4eeaf2 | 2140 | struct mount *mnt; |
1f707137 AV |
2141 | int res = f(root, arg); |
2142 | if (res) | |
2143 | return res; | |
1a4eeaf2 AV |
2144 | list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { |
2145 | res = f(&mnt->mnt, arg); | |
1f707137 AV |
2146 | if (res) |
2147 | return res; | |
2148 | } | |
2149 | return 0; | |
2150 | } | |
2151 | ||
3bd045cc AV |
2152 | static void lock_mnt_tree(struct mount *mnt) |
2153 | { | |
2154 | struct mount *p; | |
2155 | ||
2156 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2157 | int flags = p->mnt.mnt_flags; | |
2158 | /* Don't allow unprivileged users to change mount flags */ | |
2159 | flags |= MNT_LOCK_ATIME; | |
2160 | ||
2161 | if (flags & MNT_READONLY) | |
2162 | flags |= MNT_LOCK_READONLY; | |
2163 | ||
2164 | if (flags & MNT_NODEV) | |
2165 | flags |= MNT_LOCK_NODEV; | |
2166 | ||
2167 | if (flags & MNT_NOSUID) | |
2168 | flags |= MNT_LOCK_NOSUID; | |
2169 | ||
2170 | if (flags & MNT_NOEXEC) | |
2171 | flags |= MNT_LOCK_NOEXEC; | |
2172 | /* Don't allow unprivileged users to reveal what is under a mount */ | |
2173 | if (list_empty(&p->mnt_expire)) | |
2174 | flags |= MNT_LOCKED; | |
2175 | p->mnt.mnt_flags = flags; | |
2176 | } | |
2177 | } | |
2178 | ||
4b8b21f4 | 2179 | static void cleanup_group_ids(struct mount *mnt, struct mount *end) |
719f5d7f | 2180 | { |
315fc83e | 2181 | struct mount *p; |
719f5d7f | 2182 | |
909b0a88 | 2183 | for (p = mnt; p != end; p = next_mnt(p, mnt)) { |
fc7be130 | 2184 | if (p->mnt_group_id && !IS_MNT_SHARED(p)) |
4b8b21f4 | 2185 | mnt_release_group_id(p); |
719f5d7f MS |
2186 | } |
2187 | } | |
2188 | ||
4b8b21f4 | 2189 | static int invent_group_ids(struct mount *mnt, bool recurse) |
719f5d7f | 2190 | { |
315fc83e | 2191 | struct mount *p; |
719f5d7f | 2192 | |
909b0a88 | 2193 | for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { |
fc7be130 | 2194 | if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { |
4b8b21f4 | 2195 | int err = mnt_alloc_group_id(p); |
719f5d7f | 2196 | if (err) { |
4b8b21f4 | 2197 | cleanup_group_ids(mnt, p); |
719f5d7f MS |
2198 | return err; |
2199 | } | |
2200 | } | |
2201 | } | |
2202 | ||
2203 | return 0; | |
2204 | } | |
2205 | ||
d2921684 EB |
2206 | int count_mounts(struct mnt_namespace *ns, struct mount *mnt) |
2207 | { | |
2208 | unsigned int max = READ_ONCE(sysctl_mount_max); | |
124f75f8 | 2209 | unsigned int mounts = 0; |
d2921684 EB |
2210 | struct mount *p; |
2211 | ||
124f75f8 AV |
2212 | if (ns->mounts >= max) |
2213 | return -ENOSPC; | |
2214 | max -= ns->mounts; | |
2215 | if (ns->pending_mounts >= max) | |
2216 | return -ENOSPC; | |
2217 | max -= ns->pending_mounts; | |
2218 | ||
d2921684 EB |
2219 | for (p = mnt; p; p = next_mnt(p, mnt)) |
2220 | mounts++; | |
2221 | ||
124f75f8 | 2222 | if (mounts > max) |
d2921684 EB |
2223 | return -ENOSPC; |
2224 | ||
124f75f8 | 2225 | ns->pending_mounts += mounts; |
d2921684 EB |
2226 | return 0; |
2227 | } | |
2228 | ||
6ac39281 CB |
2229 | enum mnt_tree_flags_t { |
2230 | MNT_TREE_MOVE = BIT(0), | |
2231 | MNT_TREE_BENEATH = BIT(1), | |
2232 | }; | |
2233 | ||
2234 | /** | |
2235 | * attach_recursive_mnt - attach a source mount tree | |
2236 | * @source_mnt: mount tree to be attached | |
2237 | * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath | |
2238 | * @dest_mp: the mountpoint @source_mnt will be mounted at | |
2239 | * @flags: modify how @source_mnt is supposed to be attached | |
b90fa9ae RP |
2240 | * |
2241 | * NOTE: in the table below explains the semantics when a source mount | |
2242 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
2243 | * --------------------------------------------------------------------------- |
2244 | * | BIND MOUNT OPERATION | | |
2245 | * |************************************************************************** | |
2246 | * | source-->| shared | private | slave | unbindable | | |
2247 | * | dest | | | | | | |
2248 | * | | | | | | | | |
2249 | * | v | | | | | | |
2250 | * |************************************************************************** | |
2251 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
2252 | * | | | | | | | |
2253 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
2254 | * *************************************************************************** | |
b90fa9ae RP |
2255 | * A bind operation clones the source mount and mounts the clone on the |
2256 | * destination mount. | |
2257 | * | |
2258 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
2259 | * tree of the destination mount and the cloned mount is added to | |
2260 | * the peer group of the source mount. | |
2261 | * (+) the cloned mount is created under the destination mount and is marked | |
2262 | * as shared. The cloned mount is added to the peer group of the source | |
2263 | * mount. | |
5afe0022 RP |
2264 | * (+++) the mount is propagated to all the mounts in the propagation tree |
2265 | * of the destination mount and the cloned mount is made slave | |
2266 | * of the same master as that of the source mount. The cloned mount | |
2267 | * is marked as 'shared and slave'. | |
2268 | * (*) the cloned mount is made a slave of the same master as that of the | |
2269 | * source mount. | |
2270 | * | |
9676f0c6 RP |
2271 | * --------------------------------------------------------------------------- |
2272 | * | MOVE MOUNT OPERATION | | |
2273 | * |************************************************************************** | |
2274 | * | source-->| shared | private | slave | unbindable | | |
2275 | * | dest | | | | | | |
2276 | * | | | | | | | | |
2277 | * | v | | | | | | |
2278 | * |************************************************************************** | |
2279 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
2280 | * | | | | | | | |
2281 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
2282 | * *************************************************************************** | |
5afe0022 RP |
2283 | * |
2284 | * (+) the mount is moved to the destination. And is then propagated to | |
2285 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 2286 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
2287 | * (+++) the mount is moved to the destination and is then propagated to |
2288 | * all the mounts belonging to the destination mount's propagation tree. | |
2289 | * the mount is marked as 'shared and slave'. | |
2290 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
2291 | * |
2292 | * if the source mount is a tree, the operations explained above is | |
2293 | * applied to each mount in the tree. | |
2294 | * Must be called without spinlocks held, since this function can sleep | |
2295 | * in allocations. | |
6ac39281 CB |
2296 | * |
2297 | * Context: The function expects namespace_lock() to be held. | |
2298 | * Return: If @source_mnt was successfully attached 0 is returned. | |
2299 | * Otherwise a negative error code is returned. | |
b90fa9ae | 2300 | */ |
0fb54e50 | 2301 | static int attach_recursive_mnt(struct mount *source_mnt, |
6ac39281 CB |
2302 | struct mount *top_mnt, |
2303 | struct mountpoint *dest_mp, | |
2304 | enum mnt_tree_flags_t flags) | |
b90fa9ae | 2305 | { |
3bd045cc | 2306 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; |
38129a13 | 2307 | HLIST_HEAD(tree_list); |
6ac39281 | 2308 | struct mnt_namespace *ns = top_mnt->mnt_ns; |
1064f874 | 2309 | struct mountpoint *smp; |
6ac39281 | 2310 | struct mount *child, *dest_mnt, *p; |
38129a13 | 2311 | struct hlist_node *n; |
6ac39281 CB |
2312 | int err = 0; |
2313 | bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH; | |
b90fa9ae | 2314 | |
6ac39281 CB |
2315 | /* |
2316 | * Preallocate a mountpoint in case the new mounts need to be | |
2317 | * mounted beneath mounts on the same mountpoint. | |
1064f874 EB |
2318 | */ |
2319 | smp = get_mountpoint(source_mnt->mnt.mnt_root); | |
2320 | if (IS_ERR(smp)) | |
2321 | return PTR_ERR(smp); | |
2322 | ||
d2921684 | 2323 | /* Is there space to add these mounts to the mount namespace? */ |
2763d119 | 2324 | if (!moving) { |
d2921684 EB |
2325 | err = count_mounts(ns, source_mnt); |
2326 | if (err) | |
2327 | goto out; | |
2328 | } | |
2329 | ||
6ac39281 CB |
2330 | if (beneath) |
2331 | dest_mnt = top_mnt->mnt_parent; | |
2332 | else | |
2333 | dest_mnt = top_mnt; | |
2334 | ||
fc7be130 | 2335 | if (IS_MNT_SHARED(dest_mnt)) { |
0fb54e50 | 2336 | err = invent_group_ids(source_mnt, true); |
719f5d7f MS |
2337 | if (err) |
2338 | goto out; | |
0b1b901b | 2339 | err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); |
6ac39281 CB |
2340 | } |
2341 | lock_mount_hash(); | |
2342 | if (err) | |
2343 | goto out_cleanup_ids; | |
2344 | ||
2345 | if (IS_MNT_SHARED(dest_mnt)) { | |
909b0a88 | 2346 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) |
0f0afb1d | 2347 | set_mnt_shared(p); |
b90fa9ae | 2348 | } |
6ac39281 | 2349 | |
2763d119 | 2350 | if (moving) { |
6ac39281 CB |
2351 | if (beneath) |
2352 | dest_mp = smp; | |
2763d119 | 2353 | unhash_mnt(source_mnt); |
6ac39281 | 2354 | attach_mnt(source_mnt, top_mnt, dest_mp, beneath); |
143c8c91 | 2355 | touch_mnt_namespace(source_mnt->mnt_ns); |
21444403 | 2356 | } else { |
44dfd84a DH |
2357 | if (source_mnt->mnt_ns) { |
2358 | /* move from anon - the caller will destroy */ | |
2359 | list_del_init(&source_mnt->mnt_ns->list); | |
2360 | } | |
6ac39281 CB |
2361 | if (beneath) |
2362 | mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp); | |
2363 | else | |
2364 | mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); | |
1064f874 | 2365 | commit_tree(source_mnt); |
21444403 | 2366 | } |
b90fa9ae | 2367 | |
38129a13 | 2368 | hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { |
1d6a32ac | 2369 | struct mount *q; |
38129a13 | 2370 | hlist_del_init(&child->mnt_hash); |
1064f874 EB |
2371 | q = __lookup_mnt(&child->mnt_parent->mnt, |
2372 | child->mnt_mountpoint); | |
2373 | if (q) | |
2374 | mnt_change_mountpoint(child, smp, q); | |
3bd045cc AV |
2375 | /* Notice when we are propagating across user namespaces */ |
2376 | if (child->mnt_parent->mnt_ns->user_ns != user_ns) | |
2377 | lock_mnt_tree(child); | |
d728cf79 | 2378 | child->mnt.mnt_flags &= ~MNT_LOCKED; |
1064f874 | 2379 | commit_tree(child); |
b90fa9ae | 2380 | } |
1064f874 | 2381 | put_mountpoint(smp); |
719ea2fb | 2382 | unlock_mount_hash(); |
99b7db7b | 2383 | |
b90fa9ae | 2384 | return 0; |
719f5d7f MS |
2385 | |
2386 | out_cleanup_ids: | |
f2ebb3a9 AV |
2387 | while (!hlist_empty(&tree_list)) { |
2388 | child = hlist_entry(tree_list.first, struct mount, mnt_hash); | |
d2921684 | 2389 | child->mnt_parent->mnt_ns->pending_mounts = 0; |
e819f152 | 2390 | umount_tree(child, UMOUNT_SYNC); |
f2ebb3a9 AV |
2391 | } |
2392 | unlock_mount_hash(); | |
0b1b901b | 2393 | cleanup_group_ids(source_mnt, NULL); |
719f5d7f | 2394 | out: |
d2921684 | 2395 | ns->pending_mounts = 0; |
1064f874 EB |
2396 | |
2397 | read_seqlock_excl(&mount_lock); | |
2398 | put_mountpoint(smp); | |
2399 | read_sequnlock_excl(&mount_lock); | |
2400 | ||
719f5d7f | 2401 | return err; |
b90fa9ae RP |
2402 | } |
2403 | ||
6ac39281 CB |
2404 | /** |
2405 | * do_lock_mount - lock mount and mountpoint | |
2406 | * @path: target path | |
2407 | * @beneath: whether the intention is to mount beneath @path | |
2408 | * | |
2409 | * Follow the mount stack on @path until the top mount @mnt is found. If | |
2410 | * the initial @path->{mnt,dentry} is a mountpoint lookup the first | |
2411 | * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root} | |
2412 | * until nothing is stacked on top of it anymore. | |
2413 | * | |
2414 | * Acquire the inode_lock() on the top mount's ->mnt_root to protect | |
2415 | * against concurrent removal of the new mountpoint from another mount | |
2416 | * namespace. | |
2417 | * | |
2418 | * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint | |
2419 | * @mp on @mnt->mnt_parent must be acquired. This protects against a | |
2420 | * concurrent unlink of @mp->mnt_dentry from another mount namespace | |
2421 | * where @mnt doesn't have a child mount mounted @mp. A concurrent | |
2422 | * removal of @mnt->mnt_root doesn't matter as nothing will be mounted | |
2423 | * on top of it for @beneath. | |
2424 | * | |
2425 | * In addition, @beneath needs to make sure that @mnt hasn't been | |
2426 | * unmounted or moved from its current mountpoint in between dropping | |
2427 | * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt | |
2428 | * being unmounted would be detected later by e.g., calling | |
2429 | * check_mnt(mnt) in the function it's called from. For the @beneath | |
2430 | * case however, it's useful to detect it directly in do_lock_mount(). | |
2431 | * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points | |
2432 | * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will | |
2433 | * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL. | |
2434 | * | |
2435 | * Return: Either the target mountpoint on the top mount or the top | |
2436 | * mount's mountpoint. | |
2437 | */ | |
2438 | static struct mountpoint *do_lock_mount(struct path *path, bool beneath) | |
b12cea91 | 2439 | { |
6ac39281 | 2440 | struct vfsmount *mnt = path->mnt; |
64f44b27 | 2441 | struct dentry *dentry; |
6ac39281 | 2442 | struct mountpoint *mp = ERR_PTR(-ENOENT); |
64f44b27 CB |
2443 | |
2444 | for (;;) { | |
6ac39281 CB |
2445 | struct mount *m; |
2446 | ||
2447 | if (beneath) { | |
2448 | m = real_mount(mnt); | |
2449 | read_seqlock_excl(&mount_lock); | |
2450 | dentry = dget(m->mnt_mountpoint); | |
2451 | read_sequnlock_excl(&mount_lock); | |
2452 | } else { | |
2453 | dentry = path->dentry; | |
2454 | } | |
2455 | ||
64f44b27 CB |
2456 | inode_lock(dentry->d_inode); |
2457 | if (unlikely(cant_mount(dentry))) { | |
5955102c | 2458 | inode_unlock(dentry->d_inode); |
6ac39281 | 2459 | goto out; |
84d17192 | 2460 | } |
64f44b27 CB |
2461 | |
2462 | namespace_lock(); | |
2463 | ||
6ac39281 | 2464 | if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) { |
97216be0 | 2465 | namespace_unlock(); |
5955102c | 2466 | inode_unlock(dentry->d_inode); |
6ac39281 | 2467 | goto out; |
84d17192 | 2468 | } |
6ac39281 | 2469 | |
64f44b27 CB |
2470 | mnt = lookup_mnt(path); |
2471 | if (likely(!mnt)) | |
2472 | break; | |
2473 | ||
2474 | namespace_unlock(); | |
2475 | inode_unlock(dentry->d_inode); | |
6ac39281 CB |
2476 | if (beneath) |
2477 | dput(dentry); | |
64f44b27 CB |
2478 | path_put(path); |
2479 | path->mnt = mnt; | |
2480 | path->dentry = dget(mnt->mnt_root); | |
84d17192 | 2481 | } |
64f44b27 CB |
2482 | |
2483 | mp = get_mountpoint(dentry); | |
2484 | if (IS_ERR(mp)) { | |
2485 | namespace_unlock(); | |
2486 | inode_unlock(dentry->d_inode); | |
2487 | } | |
2488 | ||
6ac39281 CB |
2489 | out: |
2490 | if (beneath) | |
2491 | dput(dentry); | |
2492 | ||
64f44b27 | 2493 | return mp; |
b12cea91 AV |
2494 | } |
2495 | ||
6ac39281 CB |
2496 | static inline struct mountpoint *lock_mount(struct path *path) |
2497 | { | |
2498 | return do_lock_mount(path, false); | |
b12cea91 AV |
2499 | } |
2500 | ||
84d17192 | 2501 | static void unlock_mount(struct mountpoint *where) |
b12cea91 | 2502 | { |
84d17192 | 2503 | struct dentry *dentry = where->m_dentry; |
3895dbf8 EB |
2504 | |
2505 | read_seqlock_excl(&mount_lock); | |
84d17192 | 2506 | put_mountpoint(where); |
3895dbf8 EB |
2507 | read_sequnlock_excl(&mount_lock); |
2508 | ||
328e6d90 | 2509 | namespace_unlock(); |
5955102c | 2510 | inode_unlock(dentry->d_inode); |
b12cea91 AV |
2511 | } |
2512 | ||
84d17192 | 2513 | static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) |
1da177e4 | 2514 | { |
e462ec50 | 2515 | if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) |
1da177e4 LT |
2516 | return -EINVAL; |
2517 | ||
e36cb0b8 DH |
2518 | if (d_is_dir(mp->m_dentry) != |
2519 | d_is_dir(mnt->mnt.mnt_root)) | |
1da177e4 LT |
2520 | return -ENOTDIR; |
2521 | ||
6ac39281 | 2522 | return attach_recursive_mnt(mnt, p, mp, 0); |
1da177e4 LT |
2523 | } |
2524 | ||
7a2e8a8f VA |
2525 | /* |
2526 | * Sanity check the flags to change_mnt_propagation. | |
2527 | */ | |
2528 | ||
e462ec50 | 2529 | static int flags_to_propagation_type(int ms_flags) |
7a2e8a8f | 2530 | { |
e462ec50 | 2531 | int type = ms_flags & ~(MS_REC | MS_SILENT); |
7a2e8a8f VA |
2532 | |
2533 | /* Fail if any non-propagation flags are set */ | |
2534 | if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
2535 | return 0; | |
2536 | /* Only one propagation flag should be set */ | |
2537 | if (!is_power_of_2(type)) | |
2538 | return 0; | |
2539 | return type; | |
2540 | } | |
2541 | ||
07b20889 RP |
2542 | /* |
2543 | * recursively change the type of the mountpoint. | |
2544 | */ | |
e462ec50 | 2545 | static int do_change_type(struct path *path, int ms_flags) |
07b20889 | 2546 | { |
315fc83e | 2547 | struct mount *m; |
4b8b21f4 | 2548 | struct mount *mnt = real_mount(path->mnt); |
e462ec50 | 2549 | int recurse = ms_flags & MS_REC; |
7a2e8a8f | 2550 | int type; |
719f5d7f | 2551 | int err = 0; |
07b20889 | 2552 | |
78aa08a8 | 2553 | if (!path_mounted(path)) |
07b20889 RP |
2554 | return -EINVAL; |
2555 | ||
e462ec50 | 2556 | type = flags_to_propagation_type(ms_flags); |
7a2e8a8f VA |
2557 | if (!type) |
2558 | return -EINVAL; | |
2559 | ||
97216be0 | 2560 | namespace_lock(); |
719f5d7f MS |
2561 | if (type == MS_SHARED) { |
2562 | err = invent_group_ids(mnt, recurse); | |
2563 | if (err) | |
2564 | goto out_unlock; | |
2565 | } | |
2566 | ||
719ea2fb | 2567 | lock_mount_hash(); |
909b0a88 | 2568 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) |
0f0afb1d | 2569 | change_mnt_propagation(m, type); |
719ea2fb | 2570 | unlock_mount_hash(); |
719f5d7f MS |
2571 | |
2572 | out_unlock: | |
97216be0 | 2573 | namespace_unlock(); |
719f5d7f | 2574 | return err; |
07b20889 RP |
2575 | } |
2576 | ||
a07b2000 AV |
2577 | static struct mount *__do_loopback(struct path *old_path, int recurse) |
2578 | { | |
2579 | struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); | |
2580 | ||
2581 | if (IS_MNT_UNBINDABLE(old)) | |
2582 | return mnt; | |
2583 | ||
2584 | if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) | |
2585 | return mnt; | |
2586 | ||
2587 | if (!recurse && has_locked_children(old, old_path->dentry)) | |
2588 | return mnt; | |
2589 | ||
2590 | if (recurse) | |
2591 | mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); | |
2592 | else | |
2593 | mnt = clone_mnt(old, old_path->dentry, 0); | |
2594 | ||
2595 | if (!IS_ERR(mnt)) | |
2596 | mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
2597 | ||
2598 | return mnt; | |
2599 | } | |
2600 | ||
1da177e4 LT |
2601 | /* |
2602 | * do loopback mount. | |
2603 | */ | |
808d4e3c | 2604 | static int do_loopback(struct path *path, const char *old_name, |
2dafe1c4 | 2605 | int recurse) |
1da177e4 | 2606 | { |
2d92ab3c | 2607 | struct path old_path; |
a07b2000 | 2608 | struct mount *mnt = NULL, *parent; |
84d17192 | 2609 | struct mountpoint *mp; |
57eccb83 | 2610 | int err; |
1da177e4 LT |
2611 | if (!old_name || !*old_name) |
2612 | return -EINVAL; | |
815d405c | 2613 | err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); |
1da177e4 LT |
2614 | if (err) |
2615 | return err; | |
2616 | ||
8823c079 | 2617 | err = -EINVAL; |
4ce5d2b1 | 2618 | if (mnt_ns_loop(old_path.dentry)) |
dd111b31 | 2619 | goto out; |
8823c079 | 2620 | |
84d17192 | 2621 | mp = lock_mount(path); |
a07b2000 AV |
2622 | if (IS_ERR(mp)) { |
2623 | err = PTR_ERR(mp); | |
b12cea91 | 2624 | goto out; |
a07b2000 | 2625 | } |
b12cea91 | 2626 | |
84d17192 | 2627 | parent = real_mount(path->mnt); |
e149ed2b AV |
2628 | if (!check_mnt(parent)) |
2629 | goto out2; | |
2630 | ||
a07b2000 | 2631 | mnt = __do_loopback(&old_path, recurse); |
be34d1a3 DH |
2632 | if (IS_ERR(mnt)) { |
2633 | err = PTR_ERR(mnt); | |
e9c5d8a5 | 2634 | goto out2; |
be34d1a3 | 2635 | } |
ccd48bc7 | 2636 | |
84d17192 | 2637 | err = graft_tree(mnt, parent, mp); |
ccd48bc7 | 2638 | if (err) { |
719ea2fb | 2639 | lock_mount_hash(); |
e819f152 | 2640 | umount_tree(mnt, UMOUNT_SYNC); |
719ea2fb | 2641 | unlock_mount_hash(); |
5b83d2c5 | 2642 | } |
b12cea91 | 2643 | out2: |
84d17192 | 2644 | unlock_mount(mp); |
ccd48bc7 | 2645 | out: |
2d92ab3c | 2646 | path_put(&old_path); |
1da177e4 LT |
2647 | return err; |
2648 | } | |
2649 | ||
a07b2000 AV |
2650 | static struct file *open_detached_copy(struct path *path, bool recursive) |
2651 | { | |
2652 | struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; | |
2653 | struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); | |
2654 | struct mount *mnt, *p; | |
2655 | struct file *file; | |
2656 | ||
2657 | if (IS_ERR(ns)) | |
2658 | return ERR_CAST(ns); | |
2659 | ||
2660 | namespace_lock(); | |
2661 | mnt = __do_loopback(path, recursive); | |
2662 | if (IS_ERR(mnt)) { | |
2663 | namespace_unlock(); | |
2664 | free_mnt_ns(ns); | |
2665 | return ERR_CAST(mnt); | |
2666 | } | |
2667 | ||
2668 | lock_mount_hash(); | |
2669 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
2670 | p->mnt_ns = ns; | |
2671 | ns->mounts++; | |
2672 | } | |
2673 | ns->root = mnt; | |
2674 | list_add_tail(&ns->list, &mnt->mnt_list); | |
2675 | mntget(&mnt->mnt); | |
2676 | unlock_mount_hash(); | |
2677 | namespace_unlock(); | |
2678 | ||
2679 | mntput(path->mnt); | |
2680 | path->mnt = &mnt->mnt; | |
2681 | file = dentry_open(path, O_PATH, current_cred()); | |
2682 | if (IS_ERR(file)) | |
2683 | dissolve_on_fput(path->mnt); | |
2684 | else | |
2685 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
2686 | return file; | |
2687 | } | |
2688 | ||
2658ce09 | 2689 | SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) |
a07b2000 AV |
2690 | { |
2691 | struct file *file; | |
2692 | struct path path; | |
2693 | int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
2694 | bool detached = flags & OPEN_TREE_CLONE; | |
2695 | int error; | |
2696 | int fd; | |
2697 | ||
2698 | BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); | |
2699 | ||
2700 | if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | | |
2701 | AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | | |
2702 | OPEN_TREE_CLOEXEC)) | |
2703 | return -EINVAL; | |
2704 | ||
2705 | if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) | |
2706 | return -EINVAL; | |
2707 | ||
2708 | if (flags & AT_NO_AUTOMOUNT) | |
2709 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
2710 | if (flags & AT_SYMLINK_NOFOLLOW) | |
2711 | lookup_flags &= ~LOOKUP_FOLLOW; | |
2712 | if (flags & AT_EMPTY_PATH) | |
2713 | lookup_flags |= LOOKUP_EMPTY; | |
2714 | ||
2715 | if (detached && !may_mount()) | |
2716 | return -EPERM; | |
2717 | ||
2718 | fd = get_unused_fd_flags(flags & O_CLOEXEC); | |
2719 | if (fd < 0) | |
2720 | return fd; | |
2721 | ||
2722 | error = user_path_at(dfd, filename, lookup_flags, &path); | |
2723 | if (unlikely(error)) { | |
2724 | file = ERR_PTR(error); | |
2725 | } else { | |
2726 | if (detached) | |
2727 | file = open_detached_copy(&path, flags & AT_RECURSIVE); | |
2728 | else | |
2729 | file = dentry_open(&path, O_PATH, current_cred()); | |
2730 | path_put(&path); | |
2731 | } | |
2732 | if (IS_ERR(file)) { | |
2733 | put_unused_fd(fd); | |
2734 | return PTR_ERR(file); | |
2735 | } | |
2736 | fd_install(fd, file); | |
2737 | return fd; | |
2738 | } | |
2739 | ||
43f5e655 DH |
2740 | /* |
2741 | * Don't allow locked mount flags to be cleared. | |
2742 | * | |
2743 | * No locks need to be held here while testing the various MNT_LOCK | |
2744 | * flags because those flags can never be cleared once they are set. | |
2745 | */ | |
2746 | static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2747 | { |
43f5e655 DH |
2748 | unsigned int fl = mnt->mnt.mnt_flags; |
2749 | ||
2750 | if ((fl & MNT_LOCK_READONLY) && | |
2751 | !(mnt_flags & MNT_READONLY)) | |
2752 | return false; | |
2753 | ||
2754 | if ((fl & MNT_LOCK_NODEV) && | |
2755 | !(mnt_flags & MNT_NODEV)) | |
2756 | return false; | |
2757 | ||
2758 | if ((fl & MNT_LOCK_NOSUID) && | |
2759 | !(mnt_flags & MNT_NOSUID)) | |
2760 | return false; | |
2761 | ||
2762 | if ((fl & MNT_LOCK_NOEXEC) && | |
2763 | !(mnt_flags & MNT_NOEXEC)) | |
2764 | return false; | |
2765 | ||
2766 | if ((fl & MNT_LOCK_ATIME) && | |
2767 | ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) | |
2768 | return false; | |
2e4b7fcd | 2769 | |
43f5e655 DH |
2770 | return true; |
2771 | } | |
2772 | ||
2773 | static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) | |
2e4b7fcd | 2774 | { |
43f5e655 | 2775 | bool readonly_request = (mnt_flags & MNT_READONLY); |
2e4b7fcd | 2776 | |
43f5e655 | 2777 | if (readonly_request == __mnt_is_readonly(&mnt->mnt)) |
2e4b7fcd DH |
2778 | return 0; |
2779 | ||
2780 | if (readonly_request) | |
43f5e655 DH |
2781 | return mnt_make_readonly(mnt); |
2782 | ||
68847c94 CB |
2783 | mnt->mnt.mnt_flags &= ~MNT_READONLY; |
2784 | return 0; | |
43f5e655 DH |
2785 | } |
2786 | ||
43f5e655 DH |
2787 | static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) |
2788 | { | |
43f5e655 DH |
2789 | mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; |
2790 | mnt->mnt.mnt_flags = mnt_flags; | |
2791 | touch_mnt_namespace(mnt->mnt_ns); | |
43f5e655 DH |
2792 | } |
2793 | ||
f8b92ba6 DD |
2794 | static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) |
2795 | { | |
2796 | struct super_block *sb = mnt->mnt_sb; | |
2797 | ||
2798 | if (!__mnt_is_readonly(mnt) && | |
a128b054 | 2799 | (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && |
f8b92ba6 DD |
2800 | (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { |
2801 | char *buf = (char *)__get_free_page(GFP_KERNEL); | |
2802 | char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); | |
f8b92ba6 | 2803 | |
74e60b8b | 2804 | pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n", |
0ecee669 EB |
2805 | sb->s_type->name, |
2806 | is_mounted(mnt) ? "remounted" : "mounted", | |
74e60b8b AS |
2807 | mntpath, &sb->s_time_max, |
2808 | (unsigned long long)sb->s_time_max); | |
f8b92ba6 DD |
2809 | |
2810 | free_page((unsigned long)buf); | |
a128b054 | 2811 | sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; |
f8b92ba6 DD |
2812 | } |
2813 | } | |
2814 | ||
43f5e655 DH |
2815 | /* |
2816 | * Handle reconfiguration of the mountpoint only without alteration of the | |
2817 | * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND | |
2818 | * to mount(2). | |
2819 | */ | |
2820 | static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) | |
2821 | { | |
2822 | struct super_block *sb = path->mnt->mnt_sb; | |
2823 | struct mount *mnt = real_mount(path->mnt); | |
2824 | int ret; | |
2825 | ||
2826 | if (!check_mnt(mnt)) | |
2827 | return -EINVAL; | |
2828 | ||
78aa08a8 | 2829 | if (!path_mounted(path)) |
43f5e655 DH |
2830 | return -EINVAL; |
2831 | ||
2832 | if (!can_change_locked_flags(mnt, mnt_flags)) | |
2833 | return -EPERM; | |
2834 | ||
e58ace1a CB |
2835 | /* |
2836 | * We're only checking whether the superblock is read-only not | |
2837 | * changing it, so only take down_read(&sb->s_umount). | |
2838 | */ | |
2839 | down_read(&sb->s_umount); | |
68847c94 | 2840 | lock_mount_hash(); |
43f5e655 DH |
2841 | ret = change_mount_ro_state(mnt, mnt_flags); |
2842 | if (ret == 0) | |
2843 | set_mount_attributes(mnt, mnt_flags); | |
68847c94 | 2844 | unlock_mount_hash(); |
e58ace1a | 2845 | up_read(&sb->s_umount); |
f8b92ba6 DD |
2846 | |
2847 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2848 | ||
43f5e655 | 2849 | return ret; |
2e4b7fcd DH |
2850 | } |
2851 | ||
1da177e4 LT |
2852 | /* |
2853 | * change filesystem flags. dir should be a physical root of filesystem. | |
2854 | * If you've mounted a non-root directory somewhere and want to do remount | |
2855 | * on it - tough luck. | |
2856 | */ | |
e462ec50 DH |
2857 | static int do_remount(struct path *path, int ms_flags, int sb_flags, |
2858 | int mnt_flags, void *data) | |
1da177e4 LT |
2859 | { |
2860 | int err; | |
2d92ab3c | 2861 | struct super_block *sb = path->mnt->mnt_sb; |
143c8c91 | 2862 | struct mount *mnt = real_mount(path->mnt); |
8d0347f6 | 2863 | struct fs_context *fc; |
1da177e4 | 2864 | |
143c8c91 | 2865 | if (!check_mnt(mnt)) |
1da177e4 LT |
2866 | return -EINVAL; |
2867 | ||
78aa08a8 | 2868 | if (!path_mounted(path)) |
1da177e4 LT |
2869 | return -EINVAL; |
2870 | ||
43f5e655 | 2871 | if (!can_change_locked_flags(mnt, mnt_flags)) |
9566d674 | 2872 | return -EPERM; |
9566d674 | 2873 | |
8d0347f6 DH |
2874 | fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); |
2875 | if (IS_ERR(fc)) | |
2876 | return PTR_ERR(fc); | |
ff36fe2c | 2877 | |
b330966f | 2878 | fc->oldapi = true; |
8d0347f6 DH |
2879 | err = parse_monolithic_mount_data(fc, data); |
2880 | if (!err) { | |
2881 | down_write(&sb->s_umount); | |
2882 | err = -EPERM; | |
2883 | if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { | |
2884 | err = reconfigure_super(fc); | |
68847c94 CB |
2885 | if (!err) { |
2886 | lock_mount_hash(); | |
8d0347f6 | 2887 | set_mount_attributes(mnt, mnt_flags); |
68847c94 CB |
2888 | unlock_mount_hash(); |
2889 | } | |
8d0347f6 DH |
2890 | } |
2891 | up_write(&sb->s_umount); | |
0e55a7cc | 2892 | } |
f8b92ba6 DD |
2893 | |
2894 | mnt_warn_timestamp_expiry(path, &mnt->mnt); | |
2895 | ||
8d0347f6 | 2896 | put_fs_context(fc); |
1da177e4 LT |
2897 | return err; |
2898 | } | |
2899 | ||
cbbe362c | 2900 | static inline int tree_contains_unbindable(struct mount *mnt) |
9676f0c6 | 2901 | { |
315fc83e | 2902 | struct mount *p; |
909b0a88 | 2903 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
fc7be130 | 2904 | if (IS_MNT_UNBINDABLE(p)) |
9676f0c6 RP |
2905 | return 1; |
2906 | } | |
2907 | return 0; | |
2908 | } | |
2909 | ||
44dfd84a DH |
2910 | /* |
2911 | * Check that there aren't references to earlier/same mount namespaces in the | |
2912 | * specified subtree. Such references can act as pins for mount namespaces | |
2913 | * that aren't checked by the mount-cycle checking code, thereby allowing | |
2914 | * cycles to be made. | |
2915 | */ | |
2916 | static bool check_for_nsfs_mounts(struct mount *subtree) | |
2917 | { | |
2918 | struct mount *p; | |
2919 | bool ret = false; | |
2920 | ||
2921 | lock_mount_hash(); | |
2922 | for (p = subtree; p; p = next_mnt(p, subtree)) | |
2923 | if (mnt_ns_loop(p->mnt.mnt_root)) | |
2924 | goto out; | |
2925 | ||
2926 | ret = true; | |
2927 | out: | |
2928 | unlock_mount_hash(); | |
2929 | return ret; | |
2930 | } | |
2931 | ||
9ffb14ef PT |
2932 | static int do_set_group(struct path *from_path, struct path *to_path) |
2933 | { | |
2934 | struct mount *from, *to; | |
2935 | int err; | |
2936 | ||
2937 | from = real_mount(from_path->mnt); | |
2938 | to = real_mount(to_path->mnt); | |
2939 | ||
2940 | namespace_lock(); | |
2941 | ||
2942 | err = -EINVAL; | |
2943 | /* To and From must be mounted */ | |
2944 | if (!is_mounted(&from->mnt)) | |
2945 | goto out; | |
2946 | if (!is_mounted(&to->mnt)) | |
2947 | goto out; | |
2948 | ||
2949 | err = -EPERM; | |
2950 | /* We should be allowed to modify mount namespaces of both mounts */ | |
2951 | if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2952 | goto out; | |
2953 | if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) | |
2954 | goto out; | |
2955 | ||
2956 | err = -EINVAL; | |
2957 | /* To and From paths should be mount roots */ | |
78aa08a8 | 2958 | if (!path_mounted(from_path)) |
9ffb14ef | 2959 | goto out; |
78aa08a8 | 2960 | if (!path_mounted(to_path)) |
9ffb14ef PT |
2961 | goto out; |
2962 | ||
2963 | /* Setting sharing groups is only allowed across same superblock */ | |
2964 | if (from->mnt.mnt_sb != to->mnt.mnt_sb) | |
2965 | goto out; | |
2966 | ||
2967 | /* From mount root should be wider than To mount root */ | |
2968 | if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) | |
2969 | goto out; | |
2970 | ||
2971 | /* From mount should not have locked children in place of To's root */ | |
2972 | if (has_locked_children(from, to->mnt.mnt_root)) | |
2973 | goto out; | |
2974 | ||
2975 | /* Setting sharing groups is only allowed on private mounts */ | |
2976 | if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) | |
2977 | goto out; | |
2978 | ||
2979 | /* From should not be private */ | |
2980 | if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) | |
2981 | goto out; | |
2982 | ||
2983 | if (IS_MNT_SLAVE(from)) { | |
2984 | struct mount *m = from->mnt_master; | |
2985 | ||
2986 | list_add(&to->mnt_slave, &m->mnt_slave_list); | |
2987 | to->mnt_master = m; | |
2988 | } | |
2989 | ||
2990 | if (IS_MNT_SHARED(from)) { | |
2991 | to->mnt_group_id = from->mnt_group_id; | |
2992 | list_add(&to->mnt_share, &from->mnt_share); | |
2993 | lock_mount_hash(); | |
2994 | set_mnt_shared(to); | |
2995 | unlock_mount_hash(); | |
2996 | } | |
2997 | ||
2998 | err = 0; | |
2999 | out: | |
3000 | namespace_unlock(); | |
3001 | return err; | |
3002 | } | |
3003 | ||
6ac39281 CB |
3004 | /** |
3005 | * path_overmounted - check if path is overmounted | |
3006 | * @path: path to check | |
3007 | * | |
3008 | * Check if path is overmounted, i.e., if there's a mount on top of | |
3009 | * @path->mnt with @path->dentry as mountpoint. | |
3010 | * | |
3011 | * Context: This function expects namespace_lock() to be held. | |
3012 | * Return: If path is overmounted true is returned, false if not. | |
3013 | */ | |
3014 | static inline bool path_overmounted(const struct path *path) | |
3015 | { | |
3016 | rcu_read_lock(); | |
3017 | if (unlikely(__lookup_mnt(path->mnt, path->dentry))) { | |
3018 | rcu_read_unlock(); | |
3019 | return true; | |
3020 | } | |
3021 | rcu_read_unlock(); | |
3022 | return false; | |
3023 | } | |
3024 | ||
3025 | /** | |
3026 | * can_move_mount_beneath - check that we can mount beneath the top mount | |
3027 | * @from: mount to mount beneath | |
3028 | * @to: mount under which to mount | |
3029 | * | |
3030 | * - Make sure that @to->dentry is actually the root of a mount under | |
3031 | * which we can mount another mount. | |
3032 | * - Make sure that nothing can be mounted beneath the caller's current | |
3033 | * root or the rootfs of the namespace. | |
3034 | * - Make sure that the caller can unmount the topmost mount ensuring | |
3035 | * that the caller could reveal the underlying mountpoint. | |
3036 | * - Ensure that nothing has been mounted on top of @from before we | |
3037 | * grabbed @namespace_sem to avoid creating pointless shadow mounts. | |
3038 | * - Prevent mounting beneath a mount if the propagation relationship | |
3039 | * between the source mount, parent mount, and top mount would lead to | |
3040 | * nonsensical mount trees. | |
3041 | * | |
3042 | * Context: This function expects namespace_lock() to be held. | |
3043 | * Return: On success 0, and on error a negative error code is returned. | |
3044 | */ | |
3045 | static int can_move_mount_beneath(const struct path *from, | |
3046 | const struct path *to, | |
3047 | const struct mountpoint *mp) | |
3048 | { | |
3049 | struct mount *mnt_from = real_mount(from->mnt), | |
3050 | *mnt_to = real_mount(to->mnt), | |
3051 | *parent_mnt_to = mnt_to->mnt_parent; | |
3052 | ||
3053 | if (!mnt_has_parent(mnt_to)) | |
3054 | return -EINVAL; | |
3055 | ||
3056 | if (!path_mounted(to)) | |
3057 | return -EINVAL; | |
3058 | ||
3059 | if (IS_MNT_LOCKED(mnt_to)) | |
3060 | return -EINVAL; | |
3061 | ||
3062 | /* Avoid creating shadow mounts during mount propagation. */ | |
3063 | if (path_overmounted(from)) | |
3064 | return -EINVAL; | |
3065 | ||
3066 | /* | |
3067 | * Mounting beneath the rootfs only makes sense when the | |
3068 | * semantics of pivot_root(".", ".") are used. | |
3069 | */ | |
3070 | if (&mnt_to->mnt == current->fs->root.mnt) | |
3071 | return -EINVAL; | |
3072 | if (parent_mnt_to == current->nsproxy->mnt_ns->root) | |
3073 | return -EINVAL; | |
3074 | ||
3075 | for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent) | |
3076 | if (p == mnt_to) | |
3077 | return -EINVAL; | |
3078 | ||
3079 | /* | |
3080 | * If the parent mount propagates to the child mount this would | |
3081 | * mean mounting @mnt_from on @mnt_to->mnt_parent and then | |
3082 | * propagating a copy @c of @mnt_from on top of @mnt_to. This | |
3083 | * defeats the whole purpose of mounting beneath another mount. | |
3084 | */ | |
3085 | if (propagation_would_overmount(parent_mnt_to, mnt_to, mp)) | |
3086 | return -EINVAL; | |
3087 | ||
3088 | /* | |
3089 | * If @mnt_to->mnt_parent propagates to @mnt_from this would | |
3090 | * mean propagating a copy @c of @mnt_from on top of @mnt_from. | |
3091 | * Afterwards @mnt_from would be mounted on top of | |
3092 | * @mnt_to->mnt_parent and @mnt_to would be unmounted from | |
3093 | * @mnt->mnt_parent and remounted on @mnt_from. But since @c is | |
3094 | * already mounted on @mnt_from, @mnt_to would ultimately be | |
3095 | * remounted on top of @c. Afterwards, @mnt_from would be | |
3096 | * covered by a copy @c of @mnt_from and @c would be covered by | |
3097 | * @mnt_from itself. This defeats the whole purpose of mounting | |
3098 | * @mnt_from beneath @mnt_to. | |
3099 | */ | |
3100 | if (propagation_would_overmount(parent_mnt_to, mnt_from, mp)) | |
3101 | return -EINVAL; | |
3102 | ||
3103 | return 0; | |
3104 | } | |
3105 | ||
3106 | static int do_move_mount(struct path *old_path, struct path *new_path, | |
3107 | bool beneath) | |
1da177e4 | 3108 | { |
44dfd84a | 3109 | struct mnt_namespace *ns; |
676da58d | 3110 | struct mount *p; |
0fb54e50 | 3111 | struct mount *old; |
2763d119 AV |
3112 | struct mount *parent; |
3113 | struct mountpoint *mp, *old_mp; | |
57eccb83 | 3114 | int err; |
44dfd84a | 3115 | bool attached; |
6ac39281 | 3116 | enum mnt_tree_flags_t flags = 0; |
1da177e4 | 3117 | |
6ac39281 | 3118 | mp = do_lock_mount(new_path, beneath); |
84d17192 | 3119 | if (IS_ERR(mp)) |
2db154b3 | 3120 | return PTR_ERR(mp); |
cc53ce53 | 3121 | |
2db154b3 DH |
3122 | old = real_mount(old_path->mnt); |
3123 | p = real_mount(new_path->mnt); | |
2763d119 | 3124 | parent = old->mnt_parent; |
44dfd84a | 3125 | attached = mnt_has_parent(old); |
6ac39281 CB |
3126 | if (attached) |
3127 | flags |= MNT_TREE_MOVE; | |
2763d119 | 3128 | old_mp = old->mnt_mp; |
44dfd84a | 3129 | ns = old->mnt_ns; |
143c8c91 | 3130 | |
1da177e4 | 3131 | err = -EINVAL; |
44dfd84a DH |
3132 | /* The mountpoint must be in our namespace. */ |
3133 | if (!check_mnt(p)) | |
2db154b3 | 3134 | goto out; |
1da177e4 | 3135 | |
570d7a98 EB |
3136 | /* The thing moved must be mounted... */ |
3137 | if (!is_mounted(&old->mnt)) | |
44dfd84a DH |
3138 | goto out; |
3139 | ||
570d7a98 EB |
3140 | /* ... and either ours or the root of anon namespace */ |
3141 | if (!(attached ? check_mnt(old) : is_anon_ns(ns))) | |
2db154b3 | 3142 | goto out; |
5ff9d8a6 | 3143 | |
2db154b3 DH |
3144 | if (old->mnt.mnt_flags & MNT_LOCKED) |
3145 | goto out; | |
1da177e4 | 3146 | |
78aa08a8 | 3147 | if (!path_mounted(old_path)) |
2db154b3 | 3148 | goto out; |
1da177e4 | 3149 | |
2db154b3 DH |
3150 | if (d_is_dir(new_path->dentry) != |
3151 | d_is_dir(old_path->dentry)) | |
3152 | goto out; | |
21444403 RP |
3153 | /* |
3154 | * Don't move a mount residing in a shared parent. | |
3155 | */ | |
2763d119 | 3156 | if (attached && IS_MNT_SHARED(parent)) |
2db154b3 | 3157 | goto out; |
6ac39281 CB |
3158 | |
3159 | if (beneath) { | |
3160 | err = can_move_mount_beneath(old_path, new_path, mp); | |
3161 | if (err) | |
3162 | goto out; | |
3163 | ||
3164 | err = -EINVAL; | |
3165 | p = p->mnt_parent; | |
3166 | flags |= MNT_TREE_BENEATH; | |
3167 | } | |
3168 | ||
9676f0c6 RP |
3169 | /* |
3170 | * Don't move a mount tree containing unbindable mounts to a destination | |
3171 | * mount which is shared. | |
3172 | */ | |
fc7be130 | 3173 | if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) |
2db154b3 | 3174 | goto out; |
1da177e4 | 3175 | err = -ELOOP; |
44dfd84a DH |
3176 | if (!check_for_nsfs_mounts(old)) |
3177 | goto out; | |
fc7be130 | 3178 | for (; mnt_has_parent(p); p = p->mnt_parent) |
676da58d | 3179 | if (p == old) |
2db154b3 | 3180 | goto out; |
1da177e4 | 3181 | |
6ac39281 | 3182 | err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags); |
4ac91378 | 3183 | if (err) |
2db154b3 | 3184 | goto out; |
1da177e4 LT |
3185 | |
3186 | /* if the mount is moved, it should no longer be expire | |
3187 | * automatically */ | |
6776db3d | 3188 | list_del_init(&old->mnt_expire); |
2763d119 AV |
3189 | if (attached) |
3190 | put_mountpoint(old_mp); | |
1da177e4 | 3191 | out: |
2db154b3 | 3192 | unlock_mount(mp); |
44dfd84a | 3193 | if (!err) { |
2763d119 AV |
3194 | if (attached) |
3195 | mntput_no_expire(parent); | |
3196 | else | |
44dfd84a DH |
3197 | free_mnt_ns(ns); |
3198 | } | |
2db154b3 DH |
3199 | return err; |
3200 | } | |
3201 | ||
3202 | static int do_move_mount_old(struct path *path, const char *old_name) | |
3203 | { | |
3204 | struct path old_path; | |
3205 | int err; | |
3206 | ||
3207 | if (!old_name || !*old_name) | |
3208 | return -EINVAL; | |
3209 | ||
3210 | err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); | |
3211 | if (err) | |
3212 | return err; | |
3213 | ||
6ac39281 | 3214 | err = do_move_mount(&old_path, path, false); |
2d92ab3c | 3215 | path_put(&old_path); |
1da177e4 LT |
3216 | return err; |
3217 | } | |
3218 | ||
9d412a43 AV |
3219 | /* |
3220 | * add a mount into a namespace's mount tree | |
3221 | */ | |
8f11538e | 3222 | static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, |
1e2d8464 | 3223 | const struct path *path, int mnt_flags) |
9d412a43 | 3224 | { |
8f11538e | 3225 | struct mount *parent = real_mount(path->mnt); |
9d412a43 | 3226 | |
f2ebb3a9 | 3227 | mnt_flags &= ~MNT_INTERNAL_FLAGS; |
9d412a43 | 3228 | |
84d17192 | 3229 | if (unlikely(!check_mnt(parent))) { |
156cacb1 AV |
3230 | /* that's acceptable only for automounts done in private ns */ |
3231 | if (!(mnt_flags & MNT_SHRINKABLE)) | |
8f11538e | 3232 | return -EINVAL; |
156cacb1 | 3233 | /* ... and for those we'd better have mountpoint still alive */ |
84d17192 | 3234 | if (!parent->mnt_ns) |
8f11538e | 3235 | return -EINVAL; |
156cacb1 | 3236 | } |
9d412a43 AV |
3237 | |
3238 | /* Refuse the same filesystem on the same mount point */ | |
78aa08a8 | 3239 | if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path)) |
8f11538e | 3240 | return -EBUSY; |
9d412a43 | 3241 | |
e36cb0b8 | 3242 | if (d_is_symlink(newmnt->mnt.mnt_root)) |
8f11538e | 3243 | return -EINVAL; |
9d412a43 | 3244 | |
95bc5f25 | 3245 | newmnt->mnt.mnt_flags = mnt_flags; |
8f11538e | 3246 | return graft_tree(newmnt, parent, mp); |
9d412a43 | 3247 | } |
b1e75df4 | 3248 | |
132e4608 DH |
3249 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); |
3250 | ||
3251 | /* | |
3252 | * Create a new mount using a superblock configuration and request it | |
3253 | * be added to the namespace tree. | |
3254 | */ | |
3255 | static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, | |
3256 | unsigned int mnt_flags) | |
3257 | { | |
3258 | struct vfsmount *mnt; | |
8f11538e | 3259 | struct mountpoint *mp; |
132e4608 DH |
3260 | struct super_block *sb = fc->root->d_sb; |
3261 | int error; | |
3262 | ||
c9ce29ed AV |
3263 | error = security_sb_kern_mount(sb); |
3264 | if (!error && mount_too_revealing(sb, &mnt_flags)) | |
3265 | error = -EPERM; | |
3266 | ||
3267 | if (unlikely(error)) { | |
3268 | fc_drop_locked(fc); | |
3269 | return error; | |
132e4608 DH |
3270 | } |
3271 | ||
3272 | up_write(&sb->s_umount); | |
3273 | ||
3274 | mnt = vfs_create_mount(fc); | |
3275 | if (IS_ERR(mnt)) | |
3276 | return PTR_ERR(mnt); | |
3277 | ||
f8b92ba6 DD |
3278 | mnt_warn_timestamp_expiry(mountpoint, mnt); |
3279 | ||
8f11538e AV |
3280 | mp = lock_mount(mountpoint); |
3281 | if (IS_ERR(mp)) { | |
3282 | mntput(mnt); | |
3283 | return PTR_ERR(mp); | |
3284 | } | |
3285 | error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); | |
3286 | unlock_mount(mp); | |
0ecee669 EB |
3287 | if (error < 0) |
3288 | mntput(mnt); | |
132e4608 DH |
3289 | return error; |
3290 | } | |
1b852bce | 3291 | |
1da177e4 LT |
3292 | /* |
3293 | * create a new mount for userspace and request it to be added into the | |
3294 | * namespace's tree | |
3295 | */ | |
e462ec50 | 3296 | static int do_new_mount(struct path *path, const char *fstype, int sb_flags, |
808d4e3c | 3297 | int mnt_flags, const char *name, void *data) |
1da177e4 | 3298 | { |
0c55cfc4 | 3299 | struct file_system_type *type; |
a0c9a8b8 AV |
3300 | struct fs_context *fc; |
3301 | const char *subtype = NULL; | |
3302 | int err = 0; | |
1da177e4 | 3303 | |
0c55cfc4 | 3304 | if (!fstype) |
1da177e4 LT |
3305 | return -EINVAL; |
3306 | ||
0c55cfc4 EB |
3307 | type = get_fs_type(fstype); |
3308 | if (!type) | |
3309 | return -ENODEV; | |
3310 | ||
a0c9a8b8 AV |
3311 | if (type->fs_flags & FS_HAS_SUBTYPE) { |
3312 | subtype = strchr(fstype, '.'); | |
3313 | if (subtype) { | |
3314 | subtype++; | |
3315 | if (!*subtype) { | |
3316 | put_filesystem(type); | |
3317 | return -EINVAL; | |
3318 | } | |
a0c9a8b8 AV |
3319 | } |
3320 | } | |
0c55cfc4 | 3321 | |
a0c9a8b8 | 3322 | fc = fs_context_for_mount(type, sb_flags); |
0c55cfc4 | 3323 | put_filesystem(type); |
a0c9a8b8 AV |
3324 | if (IS_ERR(fc)) |
3325 | return PTR_ERR(fc); | |
3326 | ||
3e1aeb00 DH |
3327 | if (subtype) |
3328 | err = vfs_parse_fs_string(fc, "subtype", | |
3329 | subtype, strlen(subtype)); | |
3330 | if (!err && name) | |
3331 | err = vfs_parse_fs_string(fc, "source", name, strlen(name)); | |
a0c9a8b8 AV |
3332 | if (!err) |
3333 | err = parse_monolithic_mount_data(fc, data); | |
c3aabf07 AV |
3334 | if (!err && !mount_capable(fc)) |
3335 | err = -EPERM; | |
a0c9a8b8 AV |
3336 | if (!err) |
3337 | err = vfs_get_tree(fc); | |
132e4608 DH |
3338 | if (!err) |
3339 | err = do_new_mount_fc(fc, path, mnt_flags); | |
8654df4e | 3340 | |
a0c9a8b8 | 3341 | put_fs_context(fc); |
15f9a3f3 | 3342 | return err; |
1da177e4 LT |
3343 | } |
3344 | ||
1e2d8464 | 3345 | int finish_automount(struct vfsmount *m, const struct path *path) |
19a167af | 3346 | { |
26df6034 | 3347 | struct dentry *dentry = path->dentry; |
8f11538e | 3348 | struct mountpoint *mp; |
25e195aa | 3349 | struct mount *mnt; |
19a167af | 3350 | int err; |
25e195aa AV |
3351 | |
3352 | if (!m) | |
3353 | return 0; | |
3354 | if (IS_ERR(m)) | |
3355 | return PTR_ERR(m); | |
3356 | ||
3357 | mnt = real_mount(m); | |
19a167af AV |
3358 | /* The new mount record should have at least 2 refs to prevent it being |
3359 | * expired before we get a chance to add it | |
3360 | */ | |
6776db3d | 3361 | BUG_ON(mnt_get_count(mnt) < 2); |
19a167af AV |
3362 | |
3363 | if (m->mnt_sb == path->mnt->mnt_sb && | |
26df6034 | 3364 | m->mnt_root == dentry) { |
b1e75df4 | 3365 | err = -ELOOP; |
26df6034 | 3366 | goto discard; |
19a167af AV |
3367 | } |
3368 | ||
26df6034 AV |
3369 | /* |
3370 | * we don't want to use lock_mount() - in this case finding something | |
3371 | * that overmounts our mountpoint to be means "quitely drop what we've | |
3372 | * got", not "try to mount it on top". | |
3373 | */ | |
3374 | inode_lock(dentry->d_inode); | |
3375 | namespace_lock(); | |
3376 | if (unlikely(cant_mount(dentry))) { | |
3377 | err = -ENOENT; | |
3378 | goto discard_locked; | |
3379 | } | |
6ac39281 | 3380 | if (path_overmounted(path)) { |
26df6034 AV |
3381 | err = 0; |
3382 | goto discard_locked; | |
3383 | } | |
26df6034 | 3384 | mp = get_mountpoint(dentry); |
8f11538e AV |
3385 | if (IS_ERR(mp)) { |
3386 | err = PTR_ERR(mp); | |
26df6034 | 3387 | goto discard_locked; |
8f11538e | 3388 | } |
26df6034 | 3389 | |
8f11538e AV |
3390 | err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); |
3391 | unlock_mount(mp); | |
26df6034 AV |
3392 | if (unlikely(err)) |
3393 | goto discard; | |
3394 | mntput(m); | |
3395 | return 0; | |
3396 | ||
3397 | discard_locked: | |
3398 | namespace_unlock(); | |
3399 | inode_unlock(dentry->d_inode); | |
3400 | discard: | |
b1e75df4 | 3401 | /* remove m from any expiration list it may be on */ |
6776db3d | 3402 | if (!list_empty(&mnt->mnt_expire)) { |
97216be0 | 3403 | namespace_lock(); |
6776db3d | 3404 | list_del_init(&mnt->mnt_expire); |
97216be0 | 3405 | namespace_unlock(); |
19a167af | 3406 | } |
b1e75df4 AV |
3407 | mntput(m); |
3408 | mntput(m); | |
19a167af AV |
3409 | return err; |
3410 | } | |
3411 | ||
ea5b778a DH |
3412 | /** |
3413 | * mnt_set_expiry - Put a mount on an expiration list | |
3414 | * @mnt: The mount to list. | |
3415 | * @expiry_list: The list to add the mount to. | |
3416 | */ | |
3417 | void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) | |
3418 | { | |
97216be0 | 3419 | namespace_lock(); |
ea5b778a | 3420 | |
6776db3d | 3421 | list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); |
ea5b778a | 3422 | |
97216be0 | 3423 | namespace_unlock(); |
ea5b778a DH |
3424 | } |
3425 | EXPORT_SYMBOL(mnt_set_expiry); | |
3426 | ||
1da177e4 LT |
3427 | /* |
3428 | * process a list of expirable mountpoints with the intent of discarding any | |
3429 | * mountpoints that aren't in use and haven't been touched since last we came | |
3430 | * here | |
3431 | */ | |
3432 | void mark_mounts_for_expiry(struct list_head *mounts) | |
3433 | { | |
761d5c38 | 3434 | struct mount *mnt, *next; |
1da177e4 LT |
3435 | LIST_HEAD(graveyard); |
3436 | ||
3437 | if (list_empty(mounts)) | |
3438 | return; | |
3439 | ||
97216be0 | 3440 | namespace_lock(); |
719ea2fb | 3441 | lock_mount_hash(); |
1da177e4 LT |
3442 | |
3443 | /* extract from the expiration list every vfsmount that matches the | |
3444 | * following criteria: | |
3445 | * - only referenced by its parent vfsmount | |
3446 | * - still marked for expiry (marked on the last call here; marks are | |
3447 | * cleared by mntput()) | |
3448 | */ | |
6776db3d | 3449 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
863d684f | 3450 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1ab59738 | 3451 | propagate_mount_busy(mnt, 1)) |
1da177e4 | 3452 | continue; |
6776db3d | 3453 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 | 3454 | } |
bcc5c7d2 | 3455 | while (!list_empty(&graveyard)) { |
6776db3d | 3456 | mnt = list_first_entry(&graveyard, struct mount, mnt_expire); |
143c8c91 | 3457 | touch_mnt_namespace(mnt->mnt_ns); |
e819f152 | 3458 | umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 | 3459 | } |
719ea2fb | 3460 | unlock_mount_hash(); |
3ab6abee | 3461 | namespace_unlock(); |
5528f911 TM |
3462 | } |
3463 | ||
3464 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
3465 | ||
3466 | /* | |
3467 | * Ripoff of 'select_parent()' | |
3468 | * | |
3469 | * search the list of submounts for a given mountpoint, and move any | |
3470 | * shrinkable submounts to the 'graveyard' list. | |
3471 | */ | |
692afc31 | 3472 | static int select_submounts(struct mount *parent, struct list_head *graveyard) |
5528f911 | 3473 | { |
692afc31 | 3474 | struct mount *this_parent = parent; |
5528f911 TM |
3475 | struct list_head *next; |
3476 | int found = 0; | |
3477 | ||
3478 | repeat: | |
6b41d536 | 3479 | next = this_parent->mnt_mounts.next; |
5528f911 | 3480 | resume: |
6b41d536 | 3481 | while (next != &this_parent->mnt_mounts) { |
5528f911 | 3482 | struct list_head *tmp = next; |
6b41d536 | 3483 | struct mount *mnt = list_entry(tmp, struct mount, mnt_child); |
5528f911 TM |
3484 | |
3485 | next = tmp->next; | |
692afc31 | 3486 | if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) |
1da177e4 | 3487 | continue; |
5528f911 TM |
3488 | /* |
3489 | * Descend a level if the d_mounts list is non-empty. | |
3490 | */ | |
6b41d536 | 3491 | if (!list_empty(&mnt->mnt_mounts)) { |
5528f911 TM |
3492 | this_parent = mnt; |
3493 | goto repeat; | |
3494 | } | |
1da177e4 | 3495 | |
1ab59738 | 3496 | if (!propagate_mount_busy(mnt, 1)) { |
6776db3d | 3497 | list_move_tail(&mnt->mnt_expire, graveyard); |
5528f911 TM |
3498 | found++; |
3499 | } | |
1da177e4 | 3500 | } |
5528f911 TM |
3501 | /* |
3502 | * All done at this level ... ascend and resume the search | |
3503 | */ | |
3504 | if (this_parent != parent) { | |
6b41d536 | 3505 | next = this_parent->mnt_child.next; |
0714a533 | 3506 | this_parent = this_parent->mnt_parent; |
5528f911 TM |
3507 | goto resume; |
3508 | } | |
3509 | return found; | |
3510 | } | |
3511 | ||
3512 | /* | |
3513 | * process a list of expirable mountpoints with the intent of discarding any | |
3514 | * submounts of a specific parent mountpoint | |
99b7db7b | 3515 | * |
48a066e7 | 3516 | * mount_lock must be held for write |
5528f911 | 3517 | */ |
b54b9be7 | 3518 | static void shrink_submounts(struct mount *mnt) |
5528f911 TM |
3519 | { |
3520 | LIST_HEAD(graveyard); | |
761d5c38 | 3521 | struct mount *m; |
5528f911 | 3522 | |
5528f911 | 3523 | /* extract submounts of 'mountpoint' from the expiration list */ |
c35038be | 3524 | while (select_submounts(mnt, &graveyard)) { |
bcc5c7d2 | 3525 | while (!list_empty(&graveyard)) { |
761d5c38 | 3526 | m = list_first_entry(&graveyard, struct mount, |
6776db3d | 3527 | mnt_expire); |
143c8c91 | 3528 | touch_mnt_namespace(m->mnt_ns); |
e819f152 | 3529 | umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); |
bcc5c7d2 AV |
3530 | } |
3531 | } | |
1da177e4 LT |
3532 | } |
3533 | ||
028abd92 | 3534 | static void *copy_mount_options(const void __user * data) |
1da177e4 | 3535 | { |
b40ef869 | 3536 | char *copy; |
d563d678 | 3537 | unsigned left, offset; |
b58fed8b | 3538 | |
1da177e4 | 3539 | if (!data) |
b40ef869 | 3540 | return NULL; |
1da177e4 | 3541 | |
b40ef869 AV |
3542 | copy = kmalloc(PAGE_SIZE, GFP_KERNEL); |
3543 | if (!copy) | |
3544 | return ERR_PTR(-ENOMEM); | |
1da177e4 | 3545 | |
d563d678 | 3546 | left = copy_from_user(copy, data, PAGE_SIZE); |
1da177e4 | 3547 | |
d563d678 CM |
3548 | /* |
3549 | * Not all architectures have an exact copy_from_user(). Resort to | |
3550 | * byte at a time. | |
3551 | */ | |
3552 | offset = PAGE_SIZE - left; | |
3553 | while (left) { | |
3554 | char c; | |
3555 | if (get_user(c, (const char __user *)data + offset)) | |
3556 | break; | |
3557 | copy[offset] = c; | |
3558 | left--; | |
3559 | offset++; | |
3560 | } | |
3561 | ||
3562 | if (left == PAGE_SIZE) { | |
b40ef869 AV |
3563 | kfree(copy); |
3564 | return ERR_PTR(-EFAULT); | |
1da177e4 | 3565 | } |
d563d678 | 3566 | |
b40ef869 | 3567 | return copy; |
1da177e4 LT |
3568 | } |
3569 | ||
028abd92 | 3570 | static char *copy_mount_string(const void __user *data) |
eca6f534 | 3571 | { |
fbdb4401 | 3572 | return data ? strndup_user(data, PATH_MAX) : NULL; |
eca6f534 VN |
3573 | } |
3574 | ||
1da177e4 LT |
3575 | /* |
3576 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
3577 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
3578 | * | |
3579 | * data is a (void *) that can point to any structure up to | |
3580 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
3581 | * information (or be NULL). | |
3582 | * | |
3583 | * Pre-0.97 versions of mount() didn't have a flags word. | |
3584 | * When the flags word was introduced its top half was required | |
3585 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
3586 | * Therefore, if this magic number is present, it carries no information | |
3587 | * and must be discarded. | |
3588 | */ | |
c60166f0 | 3589 | int path_mount(const char *dev_name, struct path *path, |
808d4e3c | 3590 | const char *type_page, unsigned long flags, void *data_page) |
1da177e4 | 3591 | { |
e462ec50 | 3592 | unsigned int mnt_flags = 0, sb_flags; |
a1e6aaa3 | 3593 | int ret; |
1da177e4 LT |
3594 | |
3595 | /* Discard magic */ | |
3596 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
3597 | flags &= ~MS_MGC_MSK; | |
3598 | ||
3599 | /* Basic sanity checks */ | |
1da177e4 LT |
3600 | if (data_page) |
3601 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
3602 | ||
e462ec50 DH |
3603 | if (flags & MS_NOUSER) |
3604 | return -EINVAL; | |
3605 | ||
a1e6aaa3 CH |
3606 | ret = security_sb_mount(dev_name, path, type_page, flags, data_page); |
3607 | if (ret) | |
3608 | return ret; | |
3609 | if (!may_mount()) | |
3610 | return -EPERM; | |
f7e33bdb JL |
3611 | if (flags & SB_MANDLOCK) |
3612 | warn_mandlock(); | |
a27ab9f2 | 3613 | |
613cbe3d AK |
3614 | /* Default to relatime unless overriden */ |
3615 | if (!(flags & MS_NOATIME)) | |
3616 | mnt_flags |= MNT_RELATIME; | |
0a1c01c9 | 3617 | |
1da177e4 LT |
3618 | /* Separate the per-mountpoint flags */ |
3619 | if (flags & MS_NOSUID) | |
3620 | mnt_flags |= MNT_NOSUID; | |
3621 | if (flags & MS_NODEV) | |
3622 | mnt_flags |= MNT_NODEV; | |
3623 | if (flags & MS_NOEXEC) | |
3624 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
3625 | if (flags & MS_NOATIME) |
3626 | mnt_flags |= MNT_NOATIME; | |
3627 | if (flags & MS_NODIRATIME) | |
3628 | mnt_flags |= MNT_NODIRATIME; | |
d0adde57 MG |
3629 | if (flags & MS_STRICTATIME) |
3630 | mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); | |
a9e5b732 | 3631 | if (flags & MS_RDONLY) |
2e4b7fcd | 3632 | mnt_flags |= MNT_READONLY; |
dab741e0 MN |
3633 | if (flags & MS_NOSYMFOLLOW) |
3634 | mnt_flags |= MNT_NOSYMFOLLOW; | |
fc33a7bb | 3635 | |
ffbc6f0e EB |
3636 | /* The default atime for remount is preservation */ |
3637 | if ((flags & MS_REMOUNT) && | |
3638 | ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | | |
3639 | MS_STRICTATIME)) == 0)) { | |
3640 | mnt_flags &= ~MNT_ATIME_MASK; | |
a1e6aaa3 | 3641 | mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; |
ffbc6f0e EB |
3642 | } |
3643 | ||
e462ec50 DH |
3644 | sb_flags = flags & (SB_RDONLY | |
3645 | SB_SYNCHRONOUS | | |
3646 | SB_MANDLOCK | | |
3647 | SB_DIRSYNC | | |
3648 | SB_SILENT | | |
917086ff | 3649 | SB_POSIXACL | |
d7ee9469 | 3650 | SB_LAZYTIME | |
917086ff | 3651 | SB_I_VERSION); |
1da177e4 | 3652 | |
43f5e655 | 3653 | if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) |
a1e6aaa3 CH |
3654 | return do_reconfigure_mnt(path, mnt_flags); |
3655 | if (flags & MS_REMOUNT) | |
3656 | return do_remount(path, flags, sb_flags, mnt_flags, data_page); | |
3657 | if (flags & MS_BIND) | |
3658 | return do_loopback(path, dev_name, flags & MS_REC); | |
3659 | if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) | |
3660 | return do_change_type(path, flags); | |
3661 | if (flags & MS_MOVE) | |
3662 | return do_move_mount_old(path, dev_name); | |
3663 | ||
3664 | return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, | |
3665 | data_page); | |
3666 | } | |
3667 | ||
3668 | long do_mount(const char *dev_name, const char __user *dir_name, | |
3669 | const char *type_page, unsigned long flags, void *data_page) | |
3670 | { | |
3671 | struct path path; | |
3672 | int ret; | |
3673 | ||
3674 | ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); | |
3675 | if (ret) | |
3676 | return ret; | |
3677 | ret = path_mount(dev_name, &path, type_page, flags, data_page); | |
2d92ab3c | 3678 | path_put(&path); |
a1e6aaa3 | 3679 | return ret; |
1da177e4 LT |
3680 | } |
3681 | ||
537f7ccb EB |
3682 | static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) |
3683 | { | |
3684 | return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); | |
3685 | } | |
3686 | ||
3687 | static void dec_mnt_namespaces(struct ucounts *ucounts) | |
3688 | { | |
3689 | dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); | |
3690 | } | |
3691 | ||
771b1371 EB |
3692 | static void free_mnt_ns(struct mnt_namespace *ns) |
3693 | { | |
74e83122 AV |
3694 | if (!is_anon_ns(ns)) |
3695 | ns_free_inum(&ns->ns); | |
537f7ccb | 3696 | dec_mnt_namespaces(ns->ucounts); |
771b1371 EB |
3697 | put_user_ns(ns->user_ns); |
3698 | kfree(ns); | |
3699 | } | |
3700 | ||
8823c079 EB |
3701 | /* |
3702 | * Assign a sequence number so we can detect when we attempt to bind | |
3703 | * mount a reference to an older mount namespace into the current | |
3704 | * mount namespace, preventing reference counting loops. A 64bit | |
3705 | * number incrementing at 10Ghz will take 12,427 years to wrap which | |
3706 | * is effectively never, so we can ignore the possibility. | |
3707 | */ | |
3708 | static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); | |
3709 | ||
74e83122 | 3710 | static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) |
cf8d2c11 TM |
3711 | { |
3712 | struct mnt_namespace *new_ns; | |
537f7ccb | 3713 | struct ucounts *ucounts; |
98f842e6 | 3714 | int ret; |
cf8d2c11 | 3715 | |
537f7ccb EB |
3716 | ucounts = inc_mnt_namespaces(user_ns); |
3717 | if (!ucounts) | |
df75e774 | 3718 | return ERR_PTR(-ENOSPC); |
537f7ccb | 3719 | |
30acd0bd | 3720 | new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); |
537f7ccb EB |
3721 | if (!new_ns) { |
3722 | dec_mnt_namespaces(ucounts); | |
cf8d2c11 | 3723 | return ERR_PTR(-ENOMEM); |
537f7ccb | 3724 | } |
74e83122 AV |
3725 | if (!anon) { |
3726 | ret = ns_alloc_inum(&new_ns->ns); | |
3727 | if (ret) { | |
3728 | kfree(new_ns); | |
3729 | dec_mnt_namespaces(ucounts); | |
3730 | return ERR_PTR(ret); | |
3731 | } | |
98f842e6 | 3732 | } |
33c42940 | 3733 | new_ns->ns.ops = &mntns_operations; |
74e83122 AV |
3734 | if (!anon) |
3735 | new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); | |
1a7b8969 | 3736 | refcount_set(&new_ns->ns.count, 1); |
cf8d2c11 TM |
3737 | INIT_LIST_HEAD(&new_ns->list); |
3738 | init_waitqueue_head(&new_ns->poll); | |
9f6c61f9 | 3739 | spin_lock_init(&new_ns->ns_lock); |
771b1371 | 3740 | new_ns->user_ns = get_user_ns(user_ns); |
537f7ccb | 3741 | new_ns->ucounts = ucounts; |
cf8d2c11 TM |
3742 | return new_ns; |
3743 | } | |
3744 | ||
0766f788 | 3745 | __latent_entropy |
9559f689 AV |
3746 | struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, |
3747 | struct user_namespace *user_ns, struct fs_struct *new_fs) | |
1da177e4 | 3748 | { |
6b3286ed | 3749 | struct mnt_namespace *new_ns; |
7f2da1e7 | 3750 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; |
315fc83e | 3751 | struct mount *p, *q; |
9559f689 | 3752 | struct mount *old; |
cb338d06 | 3753 | struct mount *new; |
7a472ef4 | 3754 | int copy_flags; |
1da177e4 | 3755 | |
9559f689 AV |
3756 | BUG_ON(!ns); |
3757 | ||
3758 | if (likely(!(flags & CLONE_NEWNS))) { | |
3759 | get_mnt_ns(ns); | |
3760 | return ns; | |
3761 | } | |
3762 | ||
3763 | old = ns->root; | |
3764 | ||
74e83122 | 3765 | new_ns = alloc_mnt_ns(user_ns, false); |
cf8d2c11 TM |
3766 | if (IS_ERR(new_ns)) |
3767 | return new_ns; | |
1da177e4 | 3768 | |
97216be0 | 3769 | namespace_lock(); |
1da177e4 | 3770 | /* First pass: copy the tree topology */ |
4ce5d2b1 | 3771 | copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; |
9559f689 | 3772 | if (user_ns != ns->user_ns) |
3bd045cc | 3773 | copy_flags |= CL_SHARED_TO_SLAVE; |
7a472ef4 | 3774 | new = copy_tree(old, old->mnt.mnt_root, copy_flags); |
be34d1a3 | 3775 | if (IS_ERR(new)) { |
328e6d90 | 3776 | namespace_unlock(); |
771b1371 | 3777 | free_mnt_ns(new_ns); |
be34d1a3 | 3778 | return ERR_CAST(new); |
1da177e4 | 3779 | } |
3bd045cc AV |
3780 | if (user_ns != ns->user_ns) { |
3781 | lock_mount_hash(); | |
3782 | lock_mnt_tree(new); | |
3783 | unlock_mount_hash(); | |
3784 | } | |
be08d6d2 | 3785 | new_ns->root = new; |
1a4eeaf2 | 3786 | list_add_tail(&new_ns->list, &new->mnt_list); |
1da177e4 LT |
3787 | |
3788 | /* | |
3789 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
3790 | * as belonging to new namespace. We have already acquired a private | |
3791 | * fs_struct, so tsk->fs->lock is not needed. | |
3792 | */ | |
909b0a88 | 3793 | p = old; |
cb338d06 | 3794 | q = new; |
1da177e4 | 3795 | while (p) { |
143c8c91 | 3796 | q->mnt_ns = new_ns; |
d2921684 | 3797 | new_ns->mounts++; |
9559f689 AV |
3798 | if (new_fs) { |
3799 | if (&p->mnt == new_fs->root.mnt) { | |
3800 | new_fs->root.mnt = mntget(&q->mnt); | |
315fc83e | 3801 | rootmnt = &p->mnt; |
1da177e4 | 3802 | } |
9559f689 AV |
3803 | if (&p->mnt == new_fs->pwd.mnt) { |
3804 | new_fs->pwd.mnt = mntget(&q->mnt); | |
315fc83e | 3805 | pwdmnt = &p->mnt; |
1da177e4 | 3806 | } |
1da177e4 | 3807 | } |
909b0a88 AV |
3808 | p = next_mnt(p, old); |
3809 | q = next_mnt(q, new); | |
4ce5d2b1 EB |
3810 | if (!q) |
3811 | break; | |
61d8e426 | 3812 | // an mntns binding we'd skipped? |
4ce5d2b1 | 3813 | while (p->mnt.mnt_root != q->mnt.mnt_root) |
61d8e426 | 3814 | p = next_mnt(skip_mnt_tree(p), old); |
1da177e4 | 3815 | } |
328e6d90 | 3816 | namespace_unlock(); |
1da177e4 | 3817 | |
1da177e4 | 3818 | if (rootmnt) |
f03c6599 | 3819 | mntput(rootmnt); |
1da177e4 | 3820 | if (pwdmnt) |
f03c6599 | 3821 | mntput(pwdmnt); |
1da177e4 | 3822 | |
741a2951 | 3823 | return new_ns; |
1da177e4 LT |
3824 | } |
3825 | ||
74e83122 | 3826 | struct dentry *mount_subtree(struct vfsmount *m, const char *name) |
ea441d11 | 3827 | { |
74e83122 | 3828 | struct mount *mnt = real_mount(m); |
ea441d11 | 3829 | struct mnt_namespace *ns; |
d31da0f0 | 3830 | struct super_block *s; |
ea441d11 AV |
3831 | struct path path; |
3832 | int err; | |
3833 | ||
74e83122 AV |
3834 | ns = alloc_mnt_ns(&init_user_ns, true); |
3835 | if (IS_ERR(ns)) { | |
3836 | mntput(m); | |
ea441d11 | 3837 | return ERR_CAST(ns); |
74e83122 AV |
3838 | } |
3839 | mnt->mnt_ns = ns; | |
3840 | ns->root = mnt; | |
3841 | ns->mounts++; | |
3842 | list_add(&mnt->mnt_list, &ns->list); | |
ea441d11 | 3843 | |
74e83122 | 3844 | err = vfs_path_lookup(m->mnt_root, m, |
ea441d11 AV |
3845 | name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); |
3846 | ||
3847 | put_mnt_ns(ns); | |
3848 | ||
3849 | if (err) | |
3850 | return ERR_PTR(err); | |
3851 | ||
3852 | /* trade a vfsmount reference for active sb one */ | |
d31da0f0 AV |
3853 | s = path.mnt->mnt_sb; |
3854 | atomic_inc(&s->s_active); | |
ea441d11 AV |
3855 | mntput(path.mnt); |
3856 | /* lock the sucker */ | |
d31da0f0 | 3857 | down_write(&s->s_umount); |
ea441d11 AV |
3858 | /* ... and return the root of (sub)tree on it */ |
3859 | return path.dentry; | |
3860 | } | |
3861 | EXPORT_SYMBOL(mount_subtree); | |
3862 | ||
cccaa5e3 DB |
3863 | SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, |
3864 | char __user *, type, unsigned long, flags, void __user *, data) | |
1da177e4 | 3865 | { |
eca6f534 VN |
3866 | int ret; |
3867 | char *kernel_type; | |
eca6f534 | 3868 | char *kernel_dev; |
b40ef869 | 3869 | void *options; |
1da177e4 | 3870 | |
b8850d1f TG |
3871 | kernel_type = copy_mount_string(type); |
3872 | ret = PTR_ERR(kernel_type); | |
3873 | if (IS_ERR(kernel_type)) | |
eca6f534 | 3874 | goto out_type; |
1da177e4 | 3875 | |
b8850d1f TG |
3876 | kernel_dev = copy_mount_string(dev_name); |
3877 | ret = PTR_ERR(kernel_dev); | |
3878 | if (IS_ERR(kernel_dev)) | |
eca6f534 | 3879 | goto out_dev; |
1da177e4 | 3880 | |
b40ef869 AV |
3881 | options = copy_mount_options(data); |
3882 | ret = PTR_ERR(options); | |
3883 | if (IS_ERR(options)) | |
eca6f534 | 3884 | goto out_data; |
1da177e4 | 3885 | |
b40ef869 | 3886 | ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); |
1da177e4 | 3887 | |
b40ef869 | 3888 | kfree(options); |
eca6f534 VN |
3889 | out_data: |
3890 | kfree(kernel_dev); | |
3891 | out_dev: | |
eca6f534 VN |
3892 | kfree(kernel_type); |
3893 | out_type: | |
3894 | return ret; | |
1da177e4 LT |
3895 | } |
3896 | ||
dd8b477f CB |
3897 | #define FSMOUNT_VALID_FLAGS \ |
3898 | (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ | |
3899 | MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ | |
3900 | MOUNT_ATTR_NOSYMFOLLOW) | |
5b490500 | 3901 | |
9caccd41 | 3902 | #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) |
2a186721 CB |
3903 | |
3904 | #define MOUNT_SETATTR_PROPAGATION_FLAGS \ | |
3905 | (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) | |
3906 | ||
5b490500 CB |
3907 | static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) |
3908 | { | |
3909 | unsigned int mnt_flags = 0; | |
3910 | ||
3911 | if (attr_flags & MOUNT_ATTR_RDONLY) | |
3912 | mnt_flags |= MNT_READONLY; | |
3913 | if (attr_flags & MOUNT_ATTR_NOSUID) | |
3914 | mnt_flags |= MNT_NOSUID; | |
3915 | if (attr_flags & MOUNT_ATTR_NODEV) | |
3916 | mnt_flags |= MNT_NODEV; | |
3917 | if (attr_flags & MOUNT_ATTR_NOEXEC) | |
3918 | mnt_flags |= MNT_NOEXEC; | |
3919 | if (attr_flags & MOUNT_ATTR_NODIRATIME) | |
3920 | mnt_flags |= MNT_NODIRATIME; | |
dd8b477f CB |
3921 | if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) |
3922 | mnt_flags |= MNT_NOSYMFOLLOW; | |
5b490500 CB |
3923 | |
3924 | return mnt_flags; | |
3925 | } | |
3926 | ||
2db154b3 | 3927 | /* |
93766fbd DH |
3928 | * Create a kernel mount representation for a new, prepared superblock |
3929 | * (specified by fs_fd) and attach to an open_tree-like file descriptor. | |
3930 | */ | |
3931 | SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, | |
3932 | unsigned int, attr_flags) | |
3933 | { | |
3934 | struct mnt_namespace *ns; | |
3935 | struct fs_context *fc; | |
3936 | struct file *file; | |
3937 | struct path newmount; | |
3938 | struct mount *mnt; | |
3939 | struct fd f; | |
3940 | unsigned int mnt_flags = 0; | |
3941 | long ret; | |
3942 | ||
3943 | if (!may_mount()) | |
3944 | return -EPERM; | |
3945 | ||
3946 | if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) | |
3947 | return -EINVAL; | |
3948 | ||
5b490500 | 3949 | if (attr_flags & ~FSMOUNT_VALID_FLAGS) |
93766fbd DH |
3950 | return -EINVAL; |
3951 | ||
5b490500 | 3952 | mnt_flags = attr_flags_to_mnt_flags(attr_flags); |
93766fbd DH |
3953 | |
3954 | switch (attr_flags & MOUNT_ATTR__ATIME) { | |
3955 | case MOUNT_ATTR_STRICTATIME: | |
3956 | break; | |
3957 | case MOUNT_ATTR_NOATIME: | |
3958 | mnt_flags |= MNT_NOATIME; | |
3959 | break; | |
3960 | case MOUNT_ATTR_RELATIME: | |
3961 | mnt_flags |= MNT_RELATIME; | |
3962 | break; | |
3963 | default: | |
3964 | return -EINVAL; | |
3965 | } | |
3966 | ||
3967 | f = fdget(fs_fd); | |
3968 | if (!f.file) | |
3969 | return -EBADF; | |
3970 | ||
3971 | ret = -EINVAL; | |
3972 | if (f.file->f_op != &fscontext_fops) | |
3973 | goto err_fsfd; | |
3974 | ||
3975 | fc = f.file->private_data; | |
3976 | ||
3977 | ret = mutex_lock_interruptible(&fc->uapi_mutex); | |
3978 | if (ret < 0) | |
3979 | goto err_fsfd; | |
3980 | ||
3981 | /* There must be a valid superblock or we can't mount it */ | |
3982 | ret = -EINVAL; | |
3983 | if (!fc->root) | |
3984 | goto err_unlock; | |
3985 | ||
3986 | ret = -EPERM; | |
3987 | if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { | |
3988 | pr_warn("VFS: Mount too revealing\n"); | |
3989 | goto err_unlock; | |
3990 | } | |
3991 | ||
3992 | ret = -EBUSY; | |
3993 | if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) | |
3994 | goto err_unlock; | |
3995 | ||
f7e33bdb JL |
3996 | if (fc->sb_flags & SB_MANDLOCK) |
3997 | warn_mandlock(); | |
93766fbd DH |
3998 | |
3999 | newmount.mnt = vfs_create_mount(fc); | |
4000 | if (IS_ERR(newmount.mnt)) { | |
4001 | ret = PTR_ERR(newmount.mnt); | |
4002 | goto err_unlock; | |
4003 | } | |
4004 | newmount.dentry = dget(fc->root); | |
4005 | newmount.mnt->mnt_flags = mnt_flags; | |
4006 | ||
4007 | /* We've done the mount bit - now move the file context into more or | |
4008 | * less the same state as if we'd done an fspick(). We don't want to | |
4009 | * do any memory allocation or anything like that at this point as we | |
4010 | * don't want to have to handle any errors incurred. | |
4011 | */ | |
4012 | vfs_clean_context(fc); | |
4013 | ||
4014 | ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); | |
4015 | if (IS_ERR(ns)) { | |
4016 | ret = PTR_ERR(ns); | |
4017 | goto err_path; | |
4018 | } | |
4019 | mnt = real_mount(newmount.mnt); | |
4020 | mnt->mnt_ns = ns; | |
4021 | ns->root = mnt; | |
4022 | ns->mounts = 1; | |
4023 | list_add(&mnt->mnt_list, &ns->list); | |
1b0b9cc8 | 4024 | mntget(newmount.mnt); |
93766fbd DH |
4025 | |
4026 | /* Attach to an apparent O_PATH fd with a note that we need to unmount | |
4027 | * it, not just simply put it. | |
4028 | */ | |
4029 | file = dentry_open(&newmount, O_PATH, fc->cred); | |
4030 | if (IS_ERR(file)) { | |
4031 | dissolve_on_fput(newmount.mnt); | |
4032 | ret = PTR_ERR(file); | |
4033 | goto err_path; | |
4034 | } | |
4035 | file->f_mode |= FMODE_NEED_UNMOUNT; | |
4036 | ||
4037 | ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); | |
4038 | if (ret >= 0) | |
4039 | fd_install(ret, file); | |
4040 | else | |
4041 | fput(file); | |
4042 | ||
4043 | err_path: | |
4044 | path_put(&newmount); | |
4045 | err_unlock: | |
4046 | mutex_unlock(&fc->uapi_mutex); | |
4047 | err_fsfd: | |
4048 | fdput(f); | |
4049 | return ret; | |
4050 | } | |
4051 | ||
4052 | /* | |
4053 | * Move a mount from one place to another. In combination with | |
4054 | * fsopen()/fsmount() this is used to install a new mount and in combination | |
4055 | * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy | |
4056 | * a mount subtree. | |
2db154b3 DH |
4057 | * |
4058 | * Note the flags value is a combination of MOVE_MOUNT_* flags. | |
4059 | */ | |
4060 | SYSCALL_DEFINE5(move_mount, | |
2658ce09 BD |
4061 | int, from_dfd, const char __user *, from_pathname, |
4062 | int, to_dfd, const char __user *, to_pathname, | |
2db154b3 DH |
4063 | unsigned int, flags) |
4064 | { | |
4065 | struct path from_path, to_path; | |
4066 | unsigned int lflags; | |
4067 | int ret = 0; | |
4068 | ||
4069 | if (!may_mount()) | |
4070 | return -EPERM; | |
4071 | ||
4072 | if (flags & ~MOVE_MOUNT__MASK) | |
4073 | return -EINVAL; | |
4074 | ||
6ac39281 CB |
4075 | if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) == |
4076 | (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) | |
4077 | return -EINVAL; | |
4078 | ||
2db154b3 DH |
4079 | /* If someone gives a pathname, they aren't permitted to move |
4080 | * from an fd that requires unmount as we can't get at the flag | |
4081 | * to clear it afterwards. | |
4082 | */ | |
4083 | lflags = 0; | |
4084 | if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
4085 | if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
4086 | if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
4087 | ||
4088 | ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); | |
4089 | if (ret < 0) | |
4090 | return ret; | |
4091 | ||
4092 | lflags = 0; | |
4093 | if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; | |
4094 | if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; | |
4095 | if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; | |
4096 | ||
4097 | ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); | |
4098 | if (ret < 0) | |
4099 | goto out_from; | |
4100 | ||
4101 | ret = security_move_mount(&from_path, &to_path); | |
4102 | if (ret < 0) | |
4103 | goto out_to; | |
4104 | ||
9ffb14ef PT |
4105 | if (flags & MOVE_MOUNT_SET_GROUP) |
4106 | ret = do_set_group(&from_path, &to_path); | |
4107 | else | |
6ac39281 CB |
4108 | ret = do_move_mount(&from_path, &to_path, |
4109 | (flags & MOVE_MOUNT_BENEATH)); | |
2db154b3 DH |
4110 | |
4111 | out_to: | |
4112 | path_put(&to_path); | |
4113 | out_from: | |
4114 | path_put(&from_path); | |
4115 | return ret; | |
4116 | } | |
4117 | ||
afac7cba AV |
4118 | /* |
4119 | * Return true if path is reachable from root | |
4120 | * | |
48a066e7 | 4121 | * namespace_sem or mount_lock is held |
afac7cba | 4122 | */ |
643822b4 | 4123 | bool is_path_reachable(struct mount *mnt, struct dentry *dentry, |
afac7cba AV |
4124 | const struct path *root) |
4125 | { | |
643822b4 | 4126 | while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { |
a73324da | 4127 | dentry = mnt->mnt_mountpoint; |
0714a533 | 4128 | mnt = mnt->mnt_parent; |
afac7cba | 4129 | } |
643822b4 | 4130 | return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); |
afac7cba AV |
4131 | } |
4132 | ||
640eb7e7 | 4133 | bool path_is_under(const struct path *path1, const struct path *path2) |
afac7cba | 4134 | { |
25ab4c9b | 4135 | bool res; |
48a066e7 | 4136 | read_seqlock_excl(&mount_lock); |
643822b4 | 4137 | res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); |
48a066e7 | 4138 | read_sequnlock_excl(&mount_lock); |
afac7cba AV |
4139 | return res; |
4140 | } | |
4141 | EXPORT_SYMBOL(path_is_under); | |
4142 | ||
1da177e4 LT |
4143 | /* |
4144 | * pivot_root Semantics: | |
4145 | * Moves the root file system of the current process to the directory put_old, | |
4146 | * makes new_root as the new root file system of the current process, and sets | |
4147 | * root/cwd of all processes which had them on the current root to new_root. | |
4148 | * | |
4149 | * Restrictions: | |
4150 | * The new_root and put_old must be directories, and must not be on the | |
4151 | * same file system as the current process root. The put_old must be | |
4152 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
4153 | * pointed to by put_old must yield the same directory as new_root. No other | |
4154 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
4155 | * | |
4a0d11fa | 4156 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
0c1bc6b8 | 4157 | * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives |
4a0d11fa NB |
4158 | * in this situation. |
4159 | * | |
1da177e4 LT |
4160 | * Notes: |
4161 | * - we don't move root/cwd if they are not at the root (reason: if something | |
4162 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
4163 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
4164 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
4165 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
4166 | * first. | |
4167 | */ | |
3480b257 HC |
4168 | SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, |
4169 | const char __user *, put_old) | |
1da177e4 | 4170 | { |
2763d119 AV |
4171 | struct path new, old, root; |
4172 | struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; | |
84d17192 | 4173 | struct mountpoint *old_mp, *root_mp; |
1da177e4 LT |
4174 | int error; |
4175 | ||
9b40bc90 | 4176 | if (!may_mount()) |
1da177e4 LT |
4177 | return -EPERM; |
4178 | ||
ce6595a2 AV |
4179 | error = user_path_at(AT_FDCWD, new_root, |
4180 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); | |
1da177e4 LT |
4181 | if (error) |
4182 | goto out0; | |
1da177e4 | 4183 | |
ce6595a2 AV |
4184 | error = user_path_at(AT_FDCWD, put_old, |
4185 | LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); | |
1da177e4 LT |
4186 | if (error) |
4187 | goto out1; | |
4188 | ||
2d8f3038 | 4189 | error = security_sb_pivotroot(&old, &new); |
b12cea91 AV |
4190 | if (error) |
4191 | goto out2; | |
1da177e4 | 4192 | |
f7ad3c6b | 4193 | get_fs_root(current->fs, &root); |
84d17192 AV |
4194 | old_mp = lock_mount(&old); |
4195 | error = PTR_ERR(old_mp); | |
4196 | if (IS_ERR(old_mp)) | |
b12cea91 AV |
4197 | goto out3; |
4198 | ||
1da177e4 | 4199 | error = -EINVAL; |
419148da AV |
4200 | new_mnt = real_mount(new.mnt); |
4201 | root_mnt = real_mount(root.mnt); | |
84d17192 | 4202 | old_mnt = real_mount(old.mnt); |
2763d119 AV |
4203 | ex_parent = new_mnt->mnt_parent; |
4204 | root_parent = root_mnt->mnt_parent; | |
84d17192 | 4205 | if (IS_MNT_SHARED(old_mnt) || |
2763d119 AV |
4206 | IS_MNT_SHARED(ex_parent) || |
4207 | IS_MNT_SHARED(root_parent)) | |
b12cea91 | 4208 | goto out4; |
143c8c91 | 4209 | if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) |
b12cea91 | 4210 | goto out4; |
5ff9d8a6 EB |
4211 | if (new_mnt->mnt.mnt_flags & MNT_LOCKED) |
4212 | goto out4; | |
1da177e4 | 4213 | error = -ENOENT; |
f3da392e | 4214 | if (d_unlinked(new.dentry)) |
b12cea91 | 4215 | goto out4; |
1da177e4 | 4216 | error = -EBUSY; |
84d17192 | 4217 | if (new_mnt == root_mnt || old_mnt == root_mnt) |
b12cea91 | 4218 | goto out4; /* loop, on the same file system */ |
1da177e4 | 4219 | error = -EINVAL; |
78aa08a8 | 4220 | if (!path_mounted(&root)) |
b12cea91 | 4221 | goto out4; /* not a mountpoint */ |
676da58d | 4222 | if (!mnt_has_parent(root_mnt)) |
b12cea91 | 4223 | goto out4; /* not attached */ |
78aa08a8 | 4224 | if (!path_mounted(&new)) |
b12cea91 | 4225 | goto out4; /* not a mountpoint */ |
676da58d | 4226 | if (!mnt_has_parent(new_mnt)) |
b12cea91 | 4227 | goto out4; /* not attached */ |
4ac91378 | 4228 | /* make sure we can reach put_old from new_root */ |
84d17192 | 4229 | if (!is_path_reachable(old_mnt, old.dentry, &new)) |
b12cea91 | 4230 | goto out4; |
0d082601 EB |
4231 | /* make certain new is below the root */ |
4232 | if (!is_path_reachable(new_mnt, new.dentry, &root)) | |
4233 | goto out4; | |
719ea2fb | 4234 | lock_mount_hash(); |
2763d119 AV |
4235 | umount_mnt(new_mnt); |
4236 | root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ | |
5ff9d8a6 EB |
4237 | if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { |
4238 | new_mnt->mnt.mnt_flags |= MNT_LOCKED; | |
4239 | root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; | |
4240 | } | |
4ac91378 | 4241 | /* mount old root on put_old */ |
6ac39281 | 4242 | attach_mnt(root_mnt, old_mnt, old_mp, false); |
4ac91378 | 4243 | /* mount new_root on / */ |
6ac39281 | 4244 | attach_mnt(new_mnt, root_parent, root_mp, false); |
2763d119 | 4245 | mnt_add_count(root_parent, -1); |
6b3286ed | 4246 | touch_mnt_namespace(current->nsproxy->mnt_ns); |
4fed655c EB |
4247 | /* A moved mount should not expire automatically */ |
4248 | list_del_init(&new_mnt->mnt_expire); | |
3895dbf8 | 4249 | put_mountpoint(root_mp); |
719ea2fb | 4250 | unlock_mount_hash(); |
2d8f3038 | 4251 | chroot_fs_refs(&root, &new); |
1da177e4 | 4252 | error = 0; |
b12cea91 | 4253 | out4: |
84d17192 | 4254 | unlock_mount(old_mp); |
2763d119 AV |
4255 | if (!error) |
4256 | mntput_no_expire(ex_parent); | |
b12cea91 | 4257 | out3: |
8c3ee42e | 4258 | path_put(&root); |
b12cea91 | 4259 | out2: |
2d8f3038 | 4260 | path_put(&old); |
1da177e4 | 4261 | out1: |
2d8f3038 | 4262 | path_put(&new); |
1da177e4 | 4263 | out0: |
1da177e4 | 4264 | return error; |
1da177e4 LT |
4265 | } |
4266 | ||
2a186721 CB |
4267 | static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) |
4268 | { | |
4269 | unsigned int flags = mnt->mnt.mnt_flags; | |
4270 | ||
4271 | /* flags to clear */ | |
4272 | flags &= ~kattr->attr_clr; | |
4273 | /* flags to raise */ | |
4274 | flags |= kattr->attr_set; | |
4275 | ||
4276 | return flags; | |
4277 | } | |
4278 | ||
9caccd41 CB |
4279 | static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
4280 | { | |
4281 | struct vfsmount *m = &mnt->mnt; | |
bd303368 | 4282 | struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; |
9caccd41 | 4283 | |
256c8aed | 4284 | if (!kattr->mnt_idmap) |
9caccd41 CB |
4285 | return 0; |
4286 | ||
bd303368 CB |
4287 | /* |
4288 | * Creating an idmapped mount with the filesystem wide idmapping | |
4289 | * doesn't make sense so block that. We don't allow mushy semantics. | |
4290 | */ | |
3707d84c | 4291 | if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb)) |
bd303368 CB |
4292 | return -EINVAL; |
4293 | ||
9caccd41 CB |
4294 | /* |
4295 | * Once a mount has been idmapped we don't allow it to change its | |
4296 | * mapping. It makes things simpler and callers can just create | |
4297 | * another bind-mount they can idmap if they want to. | |
4298 | */ | |
bb49e9e7 | 4299 | if (is_idmapped_mnt(m)) |
9caccd41 CB |
4300 | return -EPERM; |
4301 | ||
4302 | /* The underlying filesystem doesn't support idmapped mounts yet. */ | |
4303 | if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) | |
4304 | return -EINVAL; | |
4305 | ||
4306 | /* We're not controlling the superblock. */ | |
bd303368 | 4307 | if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) |
9caccd41 CB |
4308 | return -EPERM; |
4309 | ||
4310 | /* Mount has already been visible in the filesystem hierarchy. */ | |
4311 | if (!is_anon_ns(mnt->mnt_ns)) | |
4312 | return -EINVAL; | |
4313 | ||
4314 | return 0; | |
4315 | } | |
4316 | ||
a26f788b CB |
4317 | /** |
4318 | * mnt_allow_writers() - check whether the attribute change allows writers | |
4319 | * @kattr: the new mount attributes | |
4320 | * @mnt: the mount to which @kattr will be applied | |
4321 | * | |
4322 | * Check whether thew new mount attributes in @kattr allow concurrent writers. | |
4323 | * | |
4324 | * Return: true if writers need to be held, false if not | |
4325 | */ | |
4326 | static inline bool mnt_allow_writers(const struct mount_kattr *kattr, | |
4327 | const struct mount *mnt) | |
2a186721 | 4328 | { |
e1bbcd27 CB |
4329 | return (!(kattr->attr_set & MNT_READONLY) || |
4330 | (mnt->mnt.mnt_flags & MNT_READONLY)) && | |
256c8aed | 4331 | !kattr->mnt_idmap; |
a26f788b | 4332 | } |
2a186721 | 4333 | |
87bb5b60 | 4334 | static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) |
2a186721 | 4335 | { |
e257039f AV |
4336 | struct mount *m; |
4337 | int err; | |
2a186721 | 4338 | |
e257039f AV |
4339 | for (m = mnt; m; m = next_mnt(m, mnt)) { |
4340 | if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { | |
4341 | err = -EPERM; | |
4342 | break; | |
4343 | } | |
2a186721 | 4344 | |
87bb5b60 CB |
4345 | err = can_idmap_mount(kattr, m); |
4346 | if (err) | |
e257039f | 4347 | break; |
2a186721 | 4348 | |
e257039f AV |
4349 | if (!mnt_allow_writers(kattr, m)) { |
4350 | err = mnt_hold_writers(m); | |
4351 | if (err) | |
4352 | break; | |
2a186721 CB |
4353 | } |
4354 | ||
e257039f AV |
4355 | if (!kattr->recurse) |
4356 | return 0; | |
4357 | } | |
9caccd41 | 4358 | |
e257039f AV |
4359 | if (err) { |
4360 | struct mount *p; | |
2a186721 | 4361 | |
0014edae CB |
4362 | /* |
4363 | * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will | |
4364 | * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all | |
4365 | * mounts and needs to take care to include the first mount. | |
4366 | */ | |
4367 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
e257039f AV |
4368 | /* If we had to hold writers unblock them. */ |
4369 | if (p->mnt.mnt_flags & MNT_WRITE_HOLD) | |
4370 | mnt_unhold_writers(p); | |
0014edae CB |
4371 | |
4372 | /* | |
4373 | * We're done once the first mount we changed got | |
4374 | * MNT_WRITE_HOLD unset. | |
4375 | */ | |
4376 | if (p == m) | |
4377 | break; | |
2a186721 | 4378 | } |
e257039f AV |
4379 | } |
4380 | return err; | |
2a186721 CB |
4381 | } |
4382 | ||
9caccd41 CB |
4383 | static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) |
4384 | { | |
256c8aed | 4385 | if (!kattr->mnt_idmap) |
9caccd41 CB |
4386 | return; |
4387 | ||
bd303368 | 4388 | /* |
256c8aed CB |
4389 | * Pairs with smp_load_acquire() in mnt_idmap(). |
4390 | * | |
4391 | * Since we only allow a mount to change the idmapping once and | |
4392 | * verified this in can_idmap_mount() we know that the mount has | |
4393 | * @nop_mnt_idmap attached to it. So there's no need to drop any | |
4394 | * references. | |
bd303368 | 4395 | */ |
256c8aed | 4396 | smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap)); |
9caccd41 CB |
4397 | } |
4398 | ||
e257039f | 4399 | static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) |
2a186721 | 4400 | { |
e257039f | 4401 | struct mount *m; |
2a186721 | 4402 | |
e257039f AV |
4403 | for (m = mnt; m; m = next_mnt(m, mnt)) { |
4404 | unsigned int flags; | |
2a186721 | 4405 | |
e257039f AV |
4406 | do_idmap_mount(kattr, m); |
4407 | flags = recalc_flags(kattr, m); | |
4408 | WRITE_ONCE(m->mnt.mnt_flags, flags); | |
2a186721 | 4409 | |
03b6abee CB |
4410 | /* If we had to hold writers unblock them. */ |
4411 | if (m->mnt.mnt_flags & MNT_WRITE_HOLD) | |
2a186721 CB |
4412 | mnt_unhold_writers(m); |
4413 | ||
e257039f | 4414 | if (kattr->propagation) |
2a186721 | 4415 | change_mnt_propagation(m, kattr->propagation); |
e257039f | 4416 | if (!kattr->recurse) |
2a186721 | 4417 | break; |
e257039f AV |
4418 | } |
4419 | touch_mnt_namespace(mnt->mnt_ns); | |
2a186721 CB |
4420 | } |
4421 | ||
4422 | static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) | |
4423 | { | |
87bb5b60 | 4424 | struct mount *mnt = real_mount(path->mnt); |
2a186721 CB |
4425 | int err = 0; |
4426 | ||
78aa08a8 | 4427 | if (!path_mounted(path)) |
2a186721 CB |
4428 | return -EINVAL; |
4429 | ||
256c8aed CB |
4430 | if (kattr->mnt_userns) { |
4431 | struct mnt_idmap *mnt_idmap; | |
4432 | ||
4433 | mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns); | |
4434 | if (IS_ERR(mnt_idmap)) | |
4435 | return PTR_ERR(mnt_idmap); | |
4436 | kattr->mnt_idmap = mnt_idmap; | |
4437 | } | |
4438 | ||
2a186721 CB |
4439 | if (kattr->propagation) { |
4440 | /* | |
4441 | * Only take namespace_lock() if we're actually changing | |
4442 | * propagation. | |
4443 | */ | |
4444 | namespace_lock(); | |
4445 | if (kattr->propagation == MS_SHARED) { | |
4446 | err = invent_group_ids(mnt, kattr->recurse); | |
4447 | if (err) { | |
4448 | namespace_unlock(); | |
4449 | return err; | |
4450 | } | |
4451 | } | |
4452 | } | |
4453 | ||
87bb5b60 | 4454 | err = -EINVAL; |
2a186721 CB |
4455 | lock_mount_hash(); |
4456 | ||
87bb5b60 CB |
4457 | /* Ensure that this isn't anything purely vfs internal. */ |
4458 | if (!is_mounted(&mnt->mnt)) | |
4459 | goto out; | |
4460 | ||
2a186721 | 4461 | /* |
87bb5b60 CB |
4462 | * If this is an attached mount make sure it's located in the callers |
4463 | * mount namespace. If it's not don't let the caller interact with it. | |
4464 | * If this is a detached mount make sure it has an anonymous mount | |
4465 | * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. | |
2a186721 | 4466 | */ |
87bb5b60 CB |
4467 | if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) |
4468 | goto out; | |
2a186721 | 4469 | |
87bb5b60 CB |
4470 | /* |
4471 | * First, we get the mount tree in a shape where we can change mount | |
4472 | * properties without failure. If we succeeded to do so we commit all | |
4473 | * changes and if we failed we clean up. | |
4474 | */ | |
4475 | err = mount_setattr_prepare(kattr, mnt); | |
e257039f AV |
4476 | if (!err) |
4477 | mount_setattr_commit(kattr, mnt); | |
2a186721 | 4478 | |
87bb5b60 | 4479 | out: |
2a186721 CB |
4480 | unlock_mount_hash(); |
4481 | ||
4482 | if (kattr->propagation) { | |
2a186721 CB |
4483 | if (err) |
4484 | cleanup_group_ids(mnt, NULL); | |
cb2239c1 | 4485 | namespace_unlock(); |
2a186721 CB |
4486 | } |
4487 | ||
4488 | return err; | |
4489 | } | |
4490 | ||
9caccd41 CB |
4491 | static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, |
4492 | struct mount_kattr *kattr, unsigned int flags) | |
4493 | { | |
4494 | int err = 0; | |
4495 | struct ns_common *ns; | |
4496 | struct user_namespace *mnt_userns; | |
96e85e95 | 4497 | struct fd f; |
9caccd41 CB |
4498 | |
4499 | if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) | |
4500 | return 0; | |
4501 | ||
4502 | /* | |
4503 | * We currently do not support clearing an idmapped mount. If this ever | |
4504 | * is a use-case we can revisit this but for now let's keep it simple | |
4505 | * and not allow it. | |
4506 | */ | |
4507 | if (attr->attr_clr & MOUNT_ATTR_IDMAP) | |
4508 | return -EINVAL; | |
4509 | ||
4510 | if (attr->userns_fd > INT_MAX) | |
4511 | return -EINVAL; | |
4512 | ||
96e85e95 AV |
4513 | f = fdget(attr->userns_fd); |
4514 | if (!f.file) | |
9caccd41 CB |
4515 | return -EBADF; |
4516 | ||
96e85e95 | 4517 | if (!proc_ns_file(f.file)) { |
9caccd41 CB |
4518 | err = -EINVAL; |
4519 | goto out_fput; | |
4520 | } | |
4521 | ||
96e85e95 | 4522 | ns = get_proc_ns(file_inode(f.file)); |
9caccd41 CB |
4523 | if (ns->ops->type != CLONE_NEWUSER) { |
4524 | err = -EINVAL; | |
4525 | goto out_fput; | |
4526 | } | |
4527 | ||
4528 | /* | |
bd303368 CB |
4529 | * The initial idmapping cannot be used to create an idmapped |
4530 | * mount. We use the initial idmapping as an indicator of a mount | |
4531 | * that is not idmapped. It can simply be passed into helpers that | |
4532 | * are aware of idmapped mounts as a convenient shortcut. A user | |
4533 | * can just create a dedicated identity mapping to achieve the same | |
4534 | * result. | |
9caccd41 CB |
4535 | */ |
4536 | mnt_userns = container_of(ns, struct user_namespace, ns); | |
3707d84c | 4537 | if (mnt_userns == &init_user_ns) { |
9caccd41 CB |
4538 | err = -EPERM; |
4539 | goto out_fput; | |
4540 | } | |
bf1ac16e SF |
4541 | |
4542 | /* We're not controlling the target namespace. */ | |
4543 | if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { | |
4544 | err = -EPERM; | |
4545 | goto out_fput; | |
4546 | } | |
4547 | ||
9caccd41 CB |
4548 | kattr->mnt_userns = get_user_ns(mnt_userns); |
4549 | ||
4550 | out_fput: | |
96e85e95 | 4551 | fdput(f); |
9caccd41 CB |
4552 | return err; |
4553 | } | |
4554 | ||
4555 | static int build_mount_kattr(const struct mount_attr *attr, size_t usize, | |
2a186721 CB |
4556 | struct mount_kattr *kattr, unsigned int flags) |
4557 | { | |
4558 | unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; | |
4559 | ||
4560 | if (flags & AT_NO_AUTOMOUNT) | |
4561 | lookup_flags &= ~LOOKUP_AUTOMOUNT; | |
4562 | if (flags & AT_SYMLINK_NOFOLLOW) | |
4563 | lookup_flags &= ~LOOKUP_FOLLOW; | |
4564 | if (flags & AT_EMPTY_PATH) | |
4565 | lookup_flags |= LOOKUP_EMPTY; | |
4566 | ||
4567 | *kattr = (struct mount_kattr) { | |
4568 | .lookup_flags = lookup_flags, | |
4569 | .recurse = !!(flags & AT_RECURSIVE), | |
4570 | }; | |
4571 | ||
4572 | if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) | |
4573 | return -EINVAL; | |
4574 | if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) | |
4575 | return -EINVAL; | |
4576 | kattr->propagation = attr->propagation; | |
4577 | ||
4578 | if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) | |
4579 | return -EINVAL; | |
4580 | ||
2a186721 CB |
4581 | kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); |
4582 | kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); | |
4583 | ||
4584 | /* | |
4585 | * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, | |
4586 | * users wanting to transition to a different atime setting cannot | |
4587 | * simply specify the atime setting in @attr_set, but must also | |
4588 | * specify MOUNT_ATTR__ATIME in the @attr_clr field. | |
4589 | * So ensure that MOUNT_ATTR__ATIME can't be partially set in | |
4590 | * @attr_clr and that @attr_set can't have any atime bits set if | |
4591 | * MOUNT_ATTR__ATIME isn't set in @attr_clr. | |
4592 | */ | |
4593 | if (attr->attr_clr & MOUNT_ATTR__ATIME) { | |
4594 | if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) | |
4595 | return -EINVAL; | |
4596 | ||
4597 | /* | |
4598 | * Clear all previous time settings as they are mutually | |
4599 | * exclusive. | |
4600 | */ | |
4601 | kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; | |
4602 | switch (attr->attr_set & MOUNT_ATTR__ATIME) { | |
4603 | case MOUNT_ATTR_RELATIME: | |
4604 | kattr->attr_set |= MNT_RELATIME; | |
4605 | break; | |
4606 | case MOUNT_ATTR_NOATIME: | |
4607 | kattr->attr_set |= MNT_NOATIME; | |
4608 | break; | |
4609 | case MOUNT_ATTR_STRICTATIME: | |
4610 | break; | |
4611 | default: | |
4612 | return -EINVAL; | |
4613 | } | |
4614 | } else { | |
4615 | if (attr->attr_set & MOUNT_ATTR__ATIME) | |
4616 | return -EINVAL; | |
4617 | } | |
4618 | ||
9caccd41 CB |
4619 | return build_mount_idmapped(attr, usize, kattr, flags); |
4620 | } | |
4621 | ||
4622 | static void finish_mount_kattr(struct mount_kattr *kattr) | |
4623 | { | |
4624 | put_user_ns(kattr->mnt_userns); | |
4625 | kattr->mnt_userns = NULL; | |
256c8aed CB |
4626 | |
4627 | if (kattr->mnt_idmap) | |
4628 | mnt_idmap_put(kattr->mnt_idmap); | |
2a186721 CB |
4629 | } |
4630 | ||
4631 | SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, | |
4632 | unsigned int, flags, struct mount_attr __user *, uattr, | |
4633 | size_t, usize) | |
4634 | { | |
4635 | int err; | |
4636 | struct path target; | |
4637 | struct mount_attr attr; | |
4638 | struct mount_kattr kattr; | |
4639 | ||
4640 | BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); | |
4641 | ||
4642 | if (flags & ~(AT_EMPTY_PATH | | |
4643 | AT_RECURSIVE | | |
4644 | AT_SYMLINK_NOFOLLOW | | |
4645 | AT_NO_AUTOMOUNT)) | |
4646 | return -EINVAL; | |
4647 | ||
4648 | if (unlikely(usize > PAGE_SIZE)) | |
4649 | return -E2BIG; | |
4650 | if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) | |
4651 | return -EINVAL; | |
4652 | ||
4653 | if (!may_mount()) | |
4654 | return -EPERM; | |
4655 | ||
4656 | err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); | |
4657 | if (err) | |
4658 | return err; | |
4659 | ||
4660 | /* Don't bother walking through the mounts if this is a nop. */ | |
4661 | if (attr.attr_set == 0 && | |
4662 | attr.attr_clr == 0 && | |
4663 | attr.propagation == 0) | |
4664 | return 0; | |
4665 | ||
9caccd41 | 4666 | err = build_mount_kattr(&attr, usize, &kattr, flags); |
2a186721 CB |
4667 | if (err) |
4668 | return err; | |
4669 | ||
4670 | err = user_path_at(dfd, path, kattr.lookup_flags, &target); | |
012e3322 CB |
4671 | if (!err) { |
4672 | err = do_mount_setattr(&target, &kattr); | |
4673 | path_put(&target); | |
4674 | } | |
9caccd41 | 4675 | finish_mount_kattr(&kattr); |
2a186721 CB |
4676 | return err; |
4677 | } | |
4678 | ||
1da177e4 LT |
4679 | static void __init init_mount_tree(void) |
4680 | { | |
4681 | struct vfsmount *mnt; | |
74e83122 | 4682 | struct mount *m; |
6b3286ed | 4683 | struct mnt_namespace *ns; |
ac748a09 | 4684 | struct path root; |
1da177e4 | 4685 | |
fd3e007f | 4686 | mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); |
1da177e4 LT |
4687 | if (IS_ERR(mnt)) |
4688 | panic("Can't create rootfs"); | |
b3e19d92 | 4689 | |
74e83122 | 4690 | ns = alloc_mnt_ns(&init_user_ns, false); |
3b22edc5 | 4691 | if (IS_ERR(ns)) |
1da177e4 | 4692 | panic("Can't allocate initial namespace"); |
74e83122 AV |
4693 | m = real_mount(mnt); |
4694 | m->mnt_ns = ns; | |
4695 | ns->root = m; | |
4696 | ns->mounts = 1; | |
4697 | list_add(&m->mnt_list, &ns->list); | |
6b3286ed KK |
4698 | init_task.nsproxy->mnt_ns = ns; |
4699 | get_mnt_ns(ns); | |
4700 | ||
be08d6d2 AV |
4701 | root.mnt = mnt; |
4702 | root.dentry = mnt->mnt_root; | |
da362b09 | 4703 | mnt->mnt_flags |= MNT_LOCKED; |
ac748a09 JB |
4704 | |
4705 | set_fs_pwd(current->fs, &root); | |
4706 | set_fs_root(current->fs, &root); | |
1da177e4 LT |
4707 | } |
4708 | ||
74bf17cf | 4709 | void __init mnt_init(void) |
1da177e4 | 4710 | { |
15a67dd8 | 4711 | int err; |
1da177e4 | 4712 | |
7d6fec45 | 4713 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), |
79f6540b | 4714 | 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); |
1da177e4 | 4715 | |
0818bf27 | 4716 | mount_hashtable = alloc_large_system_hash("Mount-cache", |
38129a13 | 4717 | sizeof(struct hlist_head), |
0818bf27 | 4718 | mhash_entries, 19, |
3d375d78 | 4719 | HASH_ZERO, |
0818bf27 AV |
4720 | &m_hash_shift, &m_hash_mask, 0, 0); |
4721 | mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", | |
4722 | sizeof(struct hlist_head), | |
4723 | mphash_entries, 19, | |
3d375d78 | 4724 | HASH_ZERO, |
0818bf27 | 4725 | &mp_hash_shift, &mp_hash_mask, 0, 0); |
1da177e4 | 4726 | |
84d17192 | 4727 | if (!mount_hashtable || !mountpoint_hashtable) |
1da177e4 LT |
4728 | panic("Failed to allocate mount hash table\n"); |
4729 | ||
4b93dc9b TH |
4730 | kernfs_init(); |
4731 | ||
15a67dd8 RD |
4732 | err = sysfs_init(); |
4733 | if (err) | |
4734 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
8e24eea7 | 4735 | __func__, err); |
00d26666 GKH |
4736 | fs_kobj = kobject_create_and_add("fs", NULL); |
4737 | if (!fs_kobj) | |
8e24eea7 | 4738 | printk(KERN_WARNING "%s: kobj create error\n", __func__); |
037f11b4 | 4739 | shmem_init(); |
1da177e4 LT |
4740 | init_rootfs(); |
4741 | init_mount_tree(); | |
4742 | } | |
4743 | ||
616511d0 | 4744 | void put_mnt_ns(struct mnt_namespace *ns) |
1da177e4 | 4745 | { |
1a7b8969 | 4746 | if (!refcount_dec_and_test(&ns->ns.count)) |
616511d0 | 4747 | return; |
7b00ed6f | 4748 | drop_collected_mounts(&ns->root->mnt); |
771b1371 | 4749 | free_mnt_ns(ns); |
1da177e4 | 4750 | } |
9d412a43 | 4751 | |
d911b458 | 4752 | struct vfsmount *kern_mount(struct file_system_type *type) |
9d412a43 | 4753 | { |
423e0ab0 | 4754 | struct vfsmount *mnt; |
d911b458 | 4755 | mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); |
423e0ab0 TC |
4756 | if (!IS_ERR(mnt)) { |
4757 | /* | |
4758 | * it is a longterm mount, don't release mnt until | |
4759 | * we unmount before file sys is unregistered | |
4760 | */ | |
f7a99c5b | 4761 | real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; |
423e0ab0 TC |
4762 | } |
4763 | return mnt; | |
9d412a43 | 4764 | } |
d911b458 | 4765 | EXPORT_SYMBOL_GPL(kern_mount); |
423e0ab0 TC |
4766 | |
4767 | void kern_unmount(struct vfsmount *mnt) | |
4768 | { | |
4769 | /* release long term mount so mount point can be released */ | |
da27f796 RR |
4770 | if (!IS_ERR(mnt)) { |
4771 | mnt_make_shortterm(mnt); | |
48a066e7 | 4772 | synchronize_rcu(); /* yecchhh... */ |
423e0ab0 TC |
4773 | mntput(mnt); |
4774 | } | |
4775 | } | |
4776 | EXPORT_SYMBOL(kern_unmount); | |
02125a82 | 4777 | |
df820f8d MS |
4778 | void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) |
4779 | { | |
4780 | unsigned int i; | |
4781 | ||
4782 | for (i = 0; i < num; i++) | |
da27f796 | 4783 | mnt_make_shortterm(mnt[i]); |
df820f8d MS |
4784 | synchronize_rcu_expedited(); |
4785 | for (i = 0; i < num; i++) | |
4786 | mntput(mnt[i]); | |
4787 | } | |
4788 | EXPORT_SYMBOL(kern_unmount_array); | |
4789 | ||
02125a82 AV |
4790 | bool our_mnt(struct vfsmount *mnt) |
4791 | { | |
143c8c91 | 4792 | return check_mnt(real_mount(mnt)); |
02125a82 | 4793 | } |
8823c079 | 4794 | |
3151527e EB |
4795 | bool current_chrooted(void) |
4796 | { | |
4797 | /* Does the current process have a non-standard root */ | |
4798 | struct path ns_root; | |
4799 | struct path fs_root; | |
4800 | bool chrooted; | |
4801 | ||
4802 | /* Find the namespace root */ | |
4803 | ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; | |
4804 | ns_root.dentry = ns_root.mnt->mnt_root; | |
4805 | path_get(&ns_root); | |
4806 | while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) | |
4807 | ; | |
4808 | ||
4809 | get_fs_root(current->fs, &fs_root); | |
4810 | ||
4811 | chrooted = !path_equal(&fs_root, &ns_root); | |
4812 | ||
4813 | path_put(&fs_root); | |
4814 | path_put(&ns_root); | |
4815 | ||
4816 | return chrooted; | |
4817 | } | |
4818 | ||
132e4608 DH |
4819 | static bool mnt_already_visible(struct mnt_namespace *ns, |
4820 | const struct super_block *sb, | |
8654df4e | 4821 | int *new_mnt_flags) |
87a8ebd6 | 4822 | { |
8c6cf9cc | 4823 | int new_flags = *new_mnt_flags; |
87a8ebd6 | 4824 | struct mount *mnt; |
e51db735 | 4825 | bool visible = false; |
87a8ebd6 | 4826 | |
44bb4385 | 4827 | down_read(&namespace_sem); |
9f6c61f9 | 4828 | lock_ns_list(ns); |
87a8ebd6 | 4829 | list_for_each_entry(mnt, &ns->list, mnt_list) { |
e51db735 | 4830 | struct mount *child; |
77b1a97d EB |
4831 | int mnt_flags; |
4832 | ||
9f6c61f9 MS |
4833 | if (mnt_is_cursor(mnt)) |
4834 | continue; | |
4835 | ||
132e4608 | 4836 | if (mnt->mnt.mnt_sb->s_type != sb->s_type) |
e51db735 EB |
4837 | continue; |
4838 | ||
7e96c1b0 EB |
4839 | /* This mount is not fully visible if it's root directory |
4840 | * is not the root directory of the filesystem. | |
4841 | */ | |
4842 | if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) | |
4843 | continue; | |
4844 | ||
a1935c17 | 4845 | /* A local view of the mount flags */ |
77b1a97d | 4846 | mnt_flags = mnt->mnt.mnt_flags; |
77b1a97d | 4847 | |
695e9df0 | 4848 | /* Don't miss readonly hidden in the superblock flags */ |
bc98a42c | 4849 | if (sb_rdonly(mnt->mnt.mnt_sb)) |
695e9df0 EB |
4850 | mnt_flags |= MNT_LOCK_READONLY; |
4851 | ||
8c6cf9cc EB |
4852 | /* Verify the mount flags are equal to or more permissive |
4853 | * than the proposed new mount. | |
4854 | */ | |
77b1a97d | 4855 | if ((mnt_flags & MNT_LOCK_READONLY) && |
8c6cf9cc EB |
4856 | !(new_flags & MNT_READONLY)) |
4857 | continue; | |
77b1a97d EB |
4858 | if ((mnt_flags & MNT_LOCK_ATIME) && |
4859 | ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) | |
8c6cf9cc EB |
4860 | continue; |
4861 | ||
ceeb0e5d EB |
4862 | /* This mount is not fully visible if there are any |
4863 | * locked child mounts that cover anything except for | |
4864 | * empty directories. | |
e51db735 EB |
4865 | */ |
4866 | list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { | |
4867 | struct inode *inode = child->mnt_mountpoint->d_inode; | |
ceeb0e5d | 4868 | /* Only worry about locked mounts */ |
d71ed6c9 | 4869 | if (!(child->mnt.mnt_flags & MNT_LOCKED)) |
ceeb0e5d | 4870 | continue; |
7236c85e EB |
4871 | /* Is the directory permanetly empty? */ |
4872 | if (!is_empty_dir_inode(inode)) | |
e51db735 | 4873 | goto next; |
87a8ebd6 | 4874 | } |
8c6cf9cc | 4875 | /* Preserve the locked attributes */ |
77b1a97d | 4876 | *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ |
77b1a97d | 4877 | MNT_LOCK_ATIME); |
e51db735 EB |
4878 | visible = true; |
4879 | goto found; | |
4880 | next: ; | |
87a8ebd6 | 4881 | } |
e51db735 | 4882 | found: |
9f6c61f9 | 4883 | unlock_ns_list(ns); |
44bb4385 | 4884 | up_read(&namespace_sem); |
e51db735 | 4885 | return visible; |
87a8ebd6 EB |
4886 | } |
4887 | ||
132e4608 | 4888 | static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) |
8654df4e | 4889 | { |
a1935c17 | 4890 | const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; |
8654df4e EB |
4891 | struct mnt_namespace *ns = current->nsproxy->mnt_ns; |
4892 | unsigned long s_iflags; | |
4893 | ||
4894 | if (ns->user_ns == &init_user_ns) | |
4895 | return false; | |
4896 | ||
4897 | /* Can this filesystem be too revealing? */ | |
132e4608 | 4898 | s_iflags = sb->s_iflags; |
8654df4e EB |
4899 | if (!(s_iflags & SB_I_USERNS_VISIBLE)) |
4900 | return false; | |
4901 | ||
a1935c17 EB |
4902 | if ((s_iflags & required_iflags) != required_iflags) { |
4903 | WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", | |
4904 | required_iflags); | |
4905 | return true; | |
4906 | } | |
4907 | ||
132e4608 | 4908 | return !mnt_already_visible(ns, sb, new_mnt_flags); |
8654df4e EB |
4909 | } |
4910 | ||
380cf5ba AL |
4911 | bool mnt_may_suid(struct vfsmount *mnt) |
4912 | { | |
4913 | /* | |
4914 | * Foreign mounts (accessed via fchdir or through /proc | |
4915 | * symlinks) are always treated as if they are nosuid. This | |
4916 | * prevents namespaces from trusting potentially unsafe | |
4917 | * suid/sgid bits, file caps, or security labels that originate | |
4918 | * in other namespaces. | |
4919 | */ | |
4920 | return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && | |
4921 | current_in_userns(mnt->mnt_sb->s_user_ns); | |
4922 | } | |
4923 | ||
64964528 | 4924 | static struct ns_common *mntns_get(struct task_struct *task) |
8823c079 | 4925 | { |
58be2825 | 4926 | struct ns_common *ns = NULL; |
8823c079 EB |
4927 | struct nsproxy *nsproxy; |
4928 | ||
728dba3a EB |
4929 | task_lock(task); |
4930 | nsproxy = task->nsproxy; | |
8823c079 | 4931 | if (nsproxy) { |
58be2825 AV |
4932 | ns = &nsproxy->mnt_ns->ns; |
4933 | get_mnt_ns(to_mnt_ns(ns)); | |
8823c079 | 4934 | } |
728dba3a | 4935 | task_unlock(task); |
8823c079 EB |
4936 | |
4937 | return ns; | |
4938 | } | |
4939 | ||
64964528 | 4940 | static void mntns_put(struct ns_common *ns) |
8823c079 | 4941 | { |
58be2825 | 4942 | put_mnt_ns(to_mnt_ns(ns)); |
8823c079 EB |
4943 | } |
4944 | ||
f2a8d52e | 4945 | static int mntns_install(struct nsset *nsset, struct ns_common *ns) |
8823c079 | 4946 | { |
f2a8d52e CB |
4947 | struct nsproxy *nsproxy = nsset->nsproxy; |
4948 | struct fs_struct *fs = nsset->fs; | |
4f757f3c | 4949 | struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; |
f2a8d52e | 4950 | struct user_namespace *user_ns = nsset->cred->user_ns; |
8823c079 | 4951 | struct path root; |
4f757f3c | 4952 | int err; |
8823c079 | 4953 | |
0c55cfc4 | 4954 | if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || |
f2a8d52e CB |
4955 | !ns_capable(user_ns, CAP_SYS_CHROOT) || |
4956 | !ns_capable(user_ns, CAP_SYS_ADMIN)) | |
ae11e0f1 | 4957 | return -EPERM; |
8823c079 | 4958 | |
74e83122 AV |
4959 | if (is_anon_ns(mnt_ns)) |
4960 | return -EINVAL; | |
4961 | ||
8823c079 EB |
4962 | if (fs->users != 1) |
4963 | return -EINVAL; | |
4964 | ||
4965 | get_mnt_ns(mnt_ns); | |
4f757f3c | 4966 | old_mnt_ns = nsproxy->mnt_ns; |
8823c079 EB |
4967 | nsproxy->mnt_ns = mnt_ns; |
4968 | ||
4969 | /* Find the root */ | |
4f757f3c AV |
4970 | err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, |
4971 | "/", LOOKUP_DOWN, &root); | |
4972 | if (err) { | |
4973 | /* revert to old namespace */ | |
4974 | nsproxy->mnt_ns = old_mnt_ns; | |
4975 | put_mnt_ns(mnt_ns); | |
4976 | return err; | |
4977 | } | |
8823c079 | 4978 | |
4068367c AV |
4979 | put_mnt_ns(old_mnt_ns); |
4980 | ||
8823c079 EB |
4981 | /* Update the pwd and root */ |
4982 | set_fs_pwd(fs, &root); | |
4983 | set_fs_root(fs, &root); | |
4984 | ||
4985 | path_put(&root); | |
4986 | return 0; | |
4987 | } | |
4988 | ||
bcac25a5 AV |
4989 | static struct user_namespace *mntns_owner(struct ns_common *ns) |
4990 | { | |
4991 | return to_mnt_ns(ns)->user_ns; | |
4992 | } | |
4993 | ||
8823c079 EB |
4994 | const struct proc_ns_operations mntns_operations = { |
4995 | .name = "mnt", | |
4996 | .type = CLONE_NEWNS, | |
4997 | .get = mntns_get, | |
4998 | .put = mntns_put, | |
4999 | .install = mntns_install, | |
bcac25a5 | 5000 | .owner = mntns_owner, |
8823c079 | 5001 | }; |
ab171b95 LC |
5002 | |
5003 | #ifdef CONFIG_SYSCTL | |
5004 | static struct ctl_table fs_namespace_sysctls[] = { | |
5005 | { | |
5006 | .procname = "mount-max", | |
5007 | .data = &sysctl_mount_max, | |
5008 | .maxlen = sizeof(unsigned int), | |
5009 | .mode = 0644, | |
5010 | .proc_handler = proc_dointvec_minmax, | |
5011 | .extra1 = SYSCTL_ONE, | |
5012 | }, | |
5013 | { } | |
5014 | }; | |
5015 | ||
5016 | static int __init init_fs_namespace_sysctls(void) | |
5017 | { | |
5018 | register_sysctl_init("fs", fs_namespace_sysctls); | |
5019 | return 0; | |
5020 | } | |
5021 | fs_initcall(init_fs_namespace_sysctls); | |
5022 | ||
5023 | #endif /* CONFIG_SYSCTL */ |