]>
Commit | Line | Data |
---|---|---|
e9be9d5e MS |
1 | /* |
2 | * | |
3 | * Copyright (C) 2011 Novell Inc. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms of the GNU General Public License version 2 as published by | |
7 | * the Free Software Foundation. | |
8 | */ | |
9 | ||
10 | #include <linux/fs.h> | |
11 | #include <linux/slab.h> | |
5b825c3a | 12 | #include <linux/cred.h> |
e9be9d5e | 13 | #include <linux/xattr.h> |
5201dc44 | 14 | #include <linux/posix_acl.h> |
5f8415d6 | 15 | #include <linux/ratelimit.h> |
e9be9d5e MS |
16 | #include "overlayfs.h" |
17 | ||
ba1e563c CR |
18 | |
19 | static dev_t ovl_get_pseudo_dev(struct dentry *dentry) | |
20 | { | |
21 | struct ovl_entry *oe = dentry->d_fsdata; | |
22 | ||
23 | return oe->lowerstack[0].layer->pseudo_dev; | |
24 | } | |
25 | ||
e9be9d5e MS |
26 | int ovl_setattr(struct dentry *dentry, struct iattr *attr) |
27 | { | |
28 | int err; | |
29 | struct dentry *upperdentry; | |
1175b6b8 | 30 | const struct cred *old_cred; |
e9be9d5e | 31 | |
cf9a6784 MS |
32 | /* |
33 | * Check for permissions before trying to copy-up. This is redundant | |
34 | * since it will be rechecked later by ->setattr() on upper dentry. But | |
35 | * without this, copy-up can be triggered by just about anybody. | |
36 | * | |
37 | * We don't initialize inode->size, which just means that | |
38 | * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not | |
39 | * check for a swapfile (which this won't be anyway). | |
40 | */ | |
31051c85 | 41 | err = setattr_prepare(dentry, attr); |
cf9a6784 MS |
42 | if (err) |
43 | return err; | |
44 | ||
e9be9d5e MS |
45 | err = ovl_want_write(dentry); |
46 | if (err) | |
47 | goto out; | |
48 | ||
acff81ec MS |
49 | err = ovl_copy_up(dentry); |
50 | if (!err) { | |
51 | upperdentry = ovl_dentry_upper(dentry); | |
52 | ||
b99c2d91 MS |
53 | if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) |
54 | attr->ia_valid &= ~ATTR_MODE; | |
55 | ||
5955102c | 56 | inode_lock(upperdentry->d_inode); |
1175b6b8 | 57 | old_cred = ovl_override_creds(dentry->d_sb); |
e9be9d5e | 58 | err = notify_change(upperdentry, attr, NULL); |
1175b6b8 | 59 | revert_creds(old_cred); |
b81de061 KK |
60 | if (!err) |
61 | ovl_copyattr(upperdentry->d_inode, dentry->d_inode); | |
5955102c | 62 | inode_unlock(upperdentry->d_inode); |
e9be9d5e MS |
63 | } |
64 | ovl_drop_write(dentry); | |
65 | out: | |
66 | return err; | |
67 | } | |
68 | ||
5b712091 MS |
69 | int ovl_getattr(const struct path *path, struct kstat *stat, |
70 | u32 request_mask, unsigned int flags) | |
e9be9d5e | 71 | { |
a528d35e | 72 | struct dentry *dentry = path->dentry; |
72b608f0 | 73 | enum ovl_path_type type; |
e9be9d5e | 74 | struct path realpath; |
1175b6b8 | 75 | const struct cred *old_cred; |
5b712091 | 76 | bool is_dir = S_ISDIR(dentry->d_inode->i_mode); |
a0c5ad30 | 77 | bool samefs = ovl_same_sb(dentry->d_sb); |
1175b6b8 | 78 | int err; |
e9be9d5e | 79 | |
72b608f0 | 80 | type = ovl_path_real(dentry, &realpath); |
1175b6b8 | 81 | old_cred = ovl_override_creds(dentry->d_sb); |
a528d35e | 82 | err = vfs_getattr(&realpath, stat, request_mask, flags); |
72b608f0 AG |
83 | if (err) |
84 | goto out; | |
85 | ||
86 | /* | |
a0c5ad30 AG |
87 | * For non-dir or same fs, we use st_ino of the copy up origin, if we |
88 | * know it. This guaranties constant st_dev/st_ino across copy up. | |
72b608f0 AG |
89 | * |
90 | * If filesystem supports NFS export ops, this also guaranties | |
91 | * persistent st_ino across mount cycle. | |
92 | */ | |
a0c5ad30 | 93 | if (!is_dir || samefs) { |
72b608f0 AG |
94 | if (OVL_TYPE_ORIGIN(type)) { |
95 | struct kstat lowerstat; | |
5b712091 | 96 | u32 lowermask = STATX_INO | (!is_dir ? STATX_NLINK : 0); |
72b608f0 AG |
97 | |
98 | ovl_path_lower(dentry, &realpath); | |
99 | err = vfs_getattr(&realpath, &lowerstat, | |
5b712091 | 100 | lowermask, flags); |
72b608f0 AG |
101 | if (err) |
102 | goto out; | |
103 | ||
72b608f0 | 104 | /* |
359f392c | 105 | * Lower hardlinks may be broken on copy up to different |
72b608f0 AG |
106 | * upper files, so we cannot use the lower origin st_ino |
107 | * for those different files, even for the same fs case. | |
359f392c AG |
108 | * With inodes index enabled, it is safe to use st_ino |
109 | * of an indexed hardlinked origin. The index validates | |
110 | * that the upper hardlink is not broken. | |
72b608f0 | 111 | */ |
359f392c | 112 | if (is_dir || lowerstat.nlink == 1 || |
a39a8c30 | 113 | ovl_test_flag(OVL_INDEX, d_inode(dentry))) { |
72b608f0 | 114 | stat->ino = lowerstat.ino; |
a0c5ad30 | 115 | stat->dev = ovl_get_pseudo_dev(dentry); |
a39a8c30 | 116 | } |
72b608f0 | 117 | } |
a0c5ad30 AG |
118 | if (samefs) { |
119 | /* | |
120 | * When all layers are on the same fs, all real inode | |
121 | * number are unique, so we use the overlay st_dev, | |
122 | * which is friendly to du -x. | |
123 | */ | |
124 | stat->dev = dentry->d_sb->s_dev; | |
125 | } else if (!OVL_TYPE_UPPER(type)) { | |
126 | /* | |
127 | * For non-samefs setup, to make sure that st_dev/st_ino | |
128 | * pair is unique across the system, we use a unique | |
129 | * anonymous st_dev for lower layer inode. | |
130 | */ | |
131 | stat->dev = ovl_get_pseudo_dev(dentry); | |
132 | } | |
133 | } else { | |
5b712091 | 134 | /* |
5b712091 MS |
135 | * Always use the overlay st_dev for directories, so 'find |
136 | * -xdev' will scan the entire overlay mount and won't cross the | |
137 | * overlay mount boundaries. | |
a0c5ad30 AG |
138 | * |
139 | * If not all layers are on the same fs the pair {real st_ino; | |
140 | * overlay st_dev} is not unique, so use the non persistent | |
141 | * overlay st_ino for directories. | |
5b712091 MS |
142 | */ |
143 | stat->dev = dentry->d_sb->s_dev; | |
144 | stat->ino = dentry->d_inode->i_ino; | |
72b608f0 | 145 | } |
5b712091 MS |
146 | |
147 | /* | |
148 | * It's probably not worth it to count subdirs to get the | |
149 | * correct link count. nlink=1 seems to pacify 'find' and | |
150 | * other utilities. | |
151 | */ | |
152 | if (is_dir && OVL_TYPE_MERGE(type)) | |
153 | stat->nlink = 1; | |
154 | ||
5f8415d6 AG |
155 | /* |
156 | * Return the overlay inode nlinks for indexed upper inodes. | |
157 | * Overlay inode nlink counts the union of the upper hardlinks | |
158 | * and non-covered lower hardlinks. It does not include the upper | |
159 | * index hardlink. | |
160 | */ | |
161 | if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
162 | stat->nlink = dentry->d_inode->i_nlink; | |
163 | ||
72b608f0 | 164 | out: |
1175b6b8 | 165 | revert_creds(old_cred); |
72b608f0 | 166 | |
1175b6b8 | 167 | return err; |
e9be9d5e MS |
168 | } |
169 | ||
170 | int ovl_permission(struct inode *inode, int mask) | |
171 | { | |
09d8b586 MS |
172 | struct inode *upperinode = ovl_inode_upper(inode); |
173 | struct inode *realinode = upperinode ?: ovl_inode_lower(inode); | |
e9be9d5e MS |
174 | int err; |
175 | ||
e9be9d5e | 176 | /* Careful in RCU walk mode */ |
e9be9d5e MS |
177 | if (!realinode) { |
178 | WARN_ON(!(mask & MAY_NOT_BLOCK)); | |
a999d7e1 | 179 | return -ECHILD; |
e9be9d5e MS |
180 | } |
181 | ||
c0ca3d70 VG |
182 | /* |
183 | * Check overlay inode with the creds of task and underlying inode | |
184 | * with creds of mounter | |
185 | */ | |
186 | err = generic_permission(inode, mask); | |
187 | if (err) | |
188 | return err; | |
189 | ||
09d8b586 MS |
190 | if (!upperinode && |
191 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 192 | mask &= ~(MAY_WRITE | MAY_APPEND); |
500cac3c VG |
193 | /* Make sure mounter can read file for copy up later */ |
194 | mask |= MAY_READ; | |
195 | } | |
a0b09f9b | 196 | err = ovl_creator_permission(inode->i_sb, realinode, mask); |
c0ca3d70 VG |
197 | |
198 | return err; | |
e9be9d5e MS |
199 | } |
200 | ||
6b255391 | 201 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
202 | struct inode *inode, |
203 | struct delayed_call *done) | |
e9be9d5e | 204 | { |
1175b6b8 VG |
205 | const struct cred *old_cred; |
206 | const char *p; | |
e9be9d5e | 207 | |
6b255391 AV |
208 | if (!dentry) |
209 | return ERR_PTR(-ECHILD); | |
210 | ||
1175b6b8 | 211 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 212 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
213 | revert_creds(old_cred); |
214 | return p; | |
e9be9d5e MS |
215 | } |
216 | ||
0956254a | 217 | bool ovl_is_private_xattr(const char *name) |
e9be9d5e | 218 | { |
fe2b7595 AG |
219 | return strncmp(name, OVL_XATTR_PREFIX, |
220 | sizeof(OVL_XATTR_PREFIX) - 1) == 0; | |
e9be9d5e MS |
221 | } |
222 | ||
1d88f183 MS |
223 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
224 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
225 | { |
226 | int err; | |
1d88f183 MS |
227 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
228 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
1175b6b8 | 229 | const struct cred *old_cred; |
e9be9d5e MS |
230 | |
231 | err = ovl_want_write(dentry); | |
232 | if (err) | |
233 | goto out; | |
234 | ||
1d88f183 MS |
235 | if (!value && !upperdentry) { |
236 | err = vfs_getxattr(realdentry, name, NULL, 0); | |
0e585ccc AG |
237 | if (err < 0) |
238 | goto out_drop_write; | |
239 | } | |
240 | ||
1d88f183 MS |
241 | if (!upperdentry) { |
242 | err = ovl_copy_up(dentry); | |
243 | if (err) | |
244 | goto out_drop_write; | |
e9be9d5e | 245 | |
1d88f183 MS |
246 | realdentry = ovl_dentry_upper(dentry); |
247 | } | |
0e585ccc | 248 | |
1175b6b8 | 249 | old_cred = ovl_override_creds(dentry->d_sb); |
0e585ccc | 250 | if (value) |
1d88f183 | 251 | err = vfs_setxattr(realdentry, name, value, size, flags); |
0e585ccc AG |
252 | else { |
253 | WARN_ON(flags != XATTR_REPLACE); | |
1d88f183 | 254 | err = vfs_removexattr(realdentry, name); |
0e585ccc | 255 | } |
1175b6b8 | 256 | revert_creds(old_cred); |
e9be9d5e MS |
257 | |
258 | out_drop_write: | |
259 | ovl_drop_write(dentry); | |
260 | out: | |
261 | return err; | |
262 | } | |
263 | ||
1d88f183 | 264 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 265 | void *value, size_t size) |
e9be9d5e | 266 | { |
1175b6b8 VG |
267 | ssize_t res; |
268 | const struct cred *old_cred; | |
1d88f183 MS |
269 | struct dentry *realdentry = |
270 | ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry); | |
52148463 | 271 | |
1175b6b8 VG |
272 | old_cred = ovl_override_creds(dentry->d_sb); |
273 | res = vfs_getxattr(realdentry, name, value, size); | |
274 | revert_creds(old_cred); | |
275 | return res; | |
e9be9d5e MS |
276 | } |
277 | ||
a082c6f6 MS |
278 | static bool ovl_can_list(const char *s) |
279 | { | |
280 | /* List all non-trusted xatts */ | |
281 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) | |
282 | return true; | |
283 | ||
284 | /* Never list trusted.overlay, list other trusted for superuser only */ | |
285 | return !ovl_is_private_xattr(s) && capable(CAP_SYS_ADMIN); | |
286 | } | |
287 | ||
e9be9d5e MS |
288 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
289 | { | |
b581755b | 290 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 291 | ssize_t res; |
7cb35119 MS |
292 | size_t len; |
293 | char *s; | |
1175b6b8 | 294 | const struct cred *old_cred; |
e9be9d5e | 295 | |
1175b6b8 | 296 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 297 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 298 | revert_creds(old_cred); |
e9be9d5e MS |
299 | if (res <= 0 || size == 0) |
300 | return res; | |
301 | ||
e9be9d5e | 302 | /* filter out private xattrs */ |
7cb35119 MS |
303 | for (s = list, len = res; len;) { |
304 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 305 | |
7cb35119 MS |
306 | /* underlying fs providing us with an broken xattr list? */ |
307 | if (WARN_ON(slen > len)) | |
308 | return -EIO; | |
e9be9d5e | 309 | |
7cb35119 | 310 | len -= slen; |
a082c6f6 | 311 | if (!ovl_can_list(s)) { |
e9be9d5e | 312 | res -= slen; |
7cb35119 | 313 | memmove(s, s + slen, len); |
e9be9d5e | 314 | } else { |
7cb35119 | 315 | s += slen; |
e9be9d5e MS |
316 | } |
317 | } | |
318 | ||
319 | return res; | |
320 | } | |
321 | ||
39a25b2b VG |
322 | struct posix_acl *ovl_get_acl(struct inode *inode, int type) |
323 | { | |
09d8b586 | 324 | struct inode *realinode = ovl_inode_real(inode); |
1175b6b8 VG |
325 | const struct cred *old_cred; |
326 | struct posix_acl *acl; | |
39a25b2b | 327 | |
5201dc44 | 328 | if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode)) |
39a25b2b VG |
329 | return NULL; |
330 | ||
1175b6b8 | 331 | old_cred = ovl_override_creds(inode->i_sb); |
5201dc44 | 332 | acl = get_acl(realinode, type); |
1175b6b8 VG |
333 | revert_creds(old_cred); |
334 | ||
335 | return acl; | |
39a25b2b VG |
336 | } |
337 | ||
59be0971 | 338 | static bool ovl_open_need_copy_up(struct dentry *dentry, int flags) |
e9be9d5e | 339 | { |
59be0971 AG |
340 | if (ovl_dentry_upper(dentry) && |
341 | ovl_dentry_has_upper_alias(dentry)) | |
e9be9d5e MS |
342 | return false; |
343 | ||
59be0971 | 344 | if (special_file(d_inode(dentry)->i_mode)) |
e9be9d5e MS |
345 | return false; |
346 | ||
347 | if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC)) | |
348 | return false; | |
349 | ||
350 | return true; | |
351 | } | |
352 | ||
2d902671 | 353 | int ovl_open_maybe_copy_up(struct dentry *dentry, unsigned int file_flags) |
e9be9d5e | 354 | { |
2d902671 | 355 | int err = 0; |
e9be9d5e | 356 | |
59be0971 | 357 | if (ovl_open_need_copy_up(dentry, file_flags)) { |
e9be9d5e | 358 | err = ovl_want_write(dentry); |
2d902671 | 359 | if (!err) { |
9aba6521 | 360 | err = ovl_copy_up_flags(dentry, file_flags); |
2d902671 MS |
361 | ovl_drop_write(dentry); |
362 | } | |
e9be9d5e MS |
363 | } |
364 | ||
2d902671 | 365 | return err; |
e9be9d5e MS |
366 | } |
367 | ||
d719e8f2 MS |
368 | int ovl_update_time(struct inode *inode, struct timespec *ts, int flags) |
369 | { | |
370 | struct dentry *alias; | |
371 | struct path upperpath; | |
372 | ||
373 | if (!(flags & S_ATIME)) | |
374 | return 0; | |
375 | ||
376 | alias = d_find_any_alias(inode); | |
377 | if (!alias) | |
378 | return 0; | |
379 | ||
380 | ovl_path_upper(alias, &upperpath); | |
381 | if (upperpath.dentry) { | |
382 | touch_atime(&upperpath); | |
383 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
384 | } | |
385 | ||
386 | dput(alias); | |
387 | ||
388 | return 0; | |
389 | } | |
390 | ||
e9be9d5e MS |
391 | static const struct inode_operations ovl_file_inode_operations = { |
392 | .setattr = ovl_setattr, | |
393 | .permission = ovl_permission, | |
394 | .getattr = ovl_getattr, | |
e9be9d5e | 395 | .listxattr = ovl_listxattr, |
39a25b2b | 396 | .get_acl = ovl_get_acl, |
d719e8f2 | 397 | .update_time = ovl_update_time, |
e9be9d5e MS |
398 | }; |
399 | ||
400 | static const struct inode_operations ovl_symlink_inode_operations = { | |
401 | .setattr = ovl_setattr, | |
6b255391 | 402 | .get_link = ovl_get_link, |
e9be9d5e | 403 | .getattr = ovl_getattr, |
e9be9d5e | 404 | .listxattr = ovl_listxattr, |
d719e8f2 | 405 | .update_time = ovl_update_time, |
e9be9d5e MS |
406 | }; |
407 | ||
b1eaa950 AG |
408 | /* |
409 | * It is possible to stack overlayfs instance on top of another | |
410 | * overlayfs instance as lower layer. We need to annonate the | |
411 | * stackable i_mutex locks according to stack level of the super | |
412 | * block instance. An overlayfs instance can never be in stack | |
413 | * depth 0 (there is always a real fs below it). An overlayfs | |
414 | * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth]. | |
415 | * | |
416 | * For example, here is a snip from /proc/lockdep_chains after | |
417 | * dir_iterate of nested overlayfs: | |
418 | * | |
419 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
420 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
421 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
422 | */ | |
423 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
424 | ||
425 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
426 | { | |
427 | #ifdef CONFIG_LOCKDEP | |
428 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
429 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 430 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
431 | |
432 | int depth = inode->i_sb->s_stack_depth - 1; | |
433 | ||
434 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
435 | depth = 0; | |
436 | ||
437 | if (S_ISDIR(inode->i_mode)) | |
438 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
439 | else | |
440 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
441 | |
442 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
443 | #endif |
444 | } | |
445 | ||
ca4c8a3a | 446 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev) |
e9be9d5e | 447 | { |
e9be9d5e MS |
448 | inode->i_ino = get_next_ino(); |
449 | inode->i_mode = mode; | |
d719e8f2 | 450 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
451 | #ifdef CONFIG_FS_POSIX_ACL |
452 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
453 | #endif | |
e9be9d5e | 454 | |
b1eaa950 AG |
455 | ovl_lockdep_annotate_inode_mutex_key(inode); |
456 | ||
ca4c8a3a MS |
457 | switch (mode & S_IFMT) { |
458 | case S_IFREG: | |
459 | inode->i_op = &ovl_file_inode_operations; | |
460 | break; | |
461 | ||
e9be9d5e | 462 | case S_IFDIR: |
e9be9d5e MS |
463 | inode->i_op = &ovl_dir_inode_operations; |
464 | inode->i_fop = &ovl_dir_operations; | |
465 | break; | |
466 | ||
467 | case S_IFLNK: | |
468 | inode->i_op = &ovl_symlink_inode_operations; | |
469 | break; | |
470 | ||
51f7e52d | 471 | default: |
e9be9d5e | 472 | inode->i_op = &ovl_file_inode_operations; |
ca4c8a3a | 473 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 474 | break; |
51f7e52d MS |
475 | } |
476 | } | |
e9be9d5e | 477 | |
5f8415d6 AG |
478 | /* |
479 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
480 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
481 | * upper inode, the following nlink modifying operations can happen: | |
482 | * | |
483 | * 1. Lower hardlink copy up | |
484 | * 2. Upper hardlink created, unlinked or renamed over | |
485 | * 3. Lower hardlink whiteout or renamed over | |
486 | * | |
487 | * For the first, copy up case, the union nlink does not change, whether the | |
488 | * operation succeeds or fails, but the upper inode nlink may change. | |
489 | * Therefore, before copy up, we store the union nlink value relative to the | |
490 | * lower inode nlink in the index inode xattr trusted.overlay.nlink. | |
491 | * | |
492 | * For the second, upper hardlink case, the union nlink should be incremented | |
493 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
494 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
495 | * value relative to the upper inode nlink in the index inode. | |
496 | * | |
497 | * For the last, lower cover up case, we simplify things by preceding the | |
498 | * whiteout or cover up with copy up. This makes sure that there is an index | |
499 | * upper inode where the nlink xattr can be stored before the copied up upper | |
500 | * entry is unlink. | |
501 | */ | |
502 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
503 | ||
504 | /* | |
505 | * On-disk format for indexed nlink: | |
506 | * | |
507 | * nlink relative to the upper inode - "U[+-]NUM" | |
508 | * nlink relative to the lower inode - "L[+-]NUM" | |
509 | */ | |
510 | ||
511 | static int ovl_set_nlink_common(struct dentry *dentry, | |
512 | struct dentry *realdentry, const char *format) | |
513 | { | |
514 | struct inode *inode = d_inode(dentry); | |
515 | struct inode *realinode = d_inode(realdentry); | |
516 | char buf[13]; | |
517 | int len; | |
518 | ||
519 | len = snprintf(buf, sizeof(buf), format, | |
520 | (int) (inode->i_nlink - realinode->i_nlink)); | |
521 | ||
6787341a MS |
522 | if (WARN_ON(len >= sizeof(buf))) |
523 | return -EIO; | |
524 | ||
5f8415d6 AG |
525 | return ovl_do_setxattr(ovl_dentry_upper(dentry), |
526 | OVL_XATTR_NLINK, buf, len, 0); | |
527 | } | |
528 | ||
529 | int ovl_set_nlink_upper(struct dentry *dentry) | |
530 | { | |
531 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
532 | } | |
533 | ||
534 | int ovl_set_nlink_lower(struct dentry *dentry) | |
535 | { | |
536 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
537 | } | |
538 | ||
caf70cb2 AG |
539 | unsigned int ovl_get_nlink(struct dentry *lowerdentry, |
540 | struct dentry *upperdentry, | |
541 | unsigned int fallback) | |
5f8415d6 AG |
542 | { |
543 | int nlink_diff; | |
544 | int nlink; | |
545 | char buf[13]; | |
546 | int err; | |
547 | ||
548 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
549 | return fallback; | |
550 | ||
551 | err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1); | |
552 | if (err < 0) | |
553 | goto fail; | |
554 | ||
555 | buf[err] = '\0'; | |
556 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
557 | (buf[1] != '+' && buf[1] != '-')) | |
558 | goto fail; | |
559 | ||
560 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
561 | if (err < 0) | |
562 | goto fail; | |
563 | ||
564 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
565 | nlink += nlink_diff; | |
566 | ||
567 | if (nlink <= 0) | |
568 | goto fail; | |
569 | ||
570 | return nlink; | |
571 | ||
572 | fail: | |
573 | pr_warn_ratelimited("overlayfs: failed to get index nlink (%pd2, err=%i)\n", | |
574 | upperdentry, err); | |
575 | return fallback; | |
576 | } | |
577 | ||
ca4c8a3a | 578 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
579 | { |
580 | struct inode *inode; | |
581 | ||
582 | inode = new_inode(sb); | |
583 | if (inode) | |
ca4c8a3a | 584 | ovl_fill_inode(inode, mode, rdev); |
51f7e52d MS |
585 | |
586 | return inode; | |
587 | } | |
588 | ||
589 | static int ovl_inode_test(struct inode *inode, void *data) | |
590 | { | |
25b7713a | 591 | return inode->i_private == data; |
51f7e52d MS |
592 | } |
593 | ||
594 | static int ovl_inode_set(struct inode *inode, void *data) | |
595 | { | |
25b7713a | 596 | inode->i_private = data; |
51f7e52d MS |
597 | return 0; |
598 | } | |
599 | ||
b9ac5c27 MS |
600 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
601 | struct dentry *upperdentry) | |
602 | { | |
d2c9584e AG |
603 | if (S_ISDIR(inode->i_mode)) { |
604 | /* Real lower dir moved to upper layer under us? */ | |
605 | if (!lowerdentry && ovl_inode_lower(inode)) | |
606 | return false; | |
607 | ||
608 | /* Lookup of an uncovered redirect origin? */ | |
609 | if (!upperdentry && ovl_inode_upper(inode)) | |
610 | return false; | |
611 | } | |
612 | ||
939ae4ef AG |
613 | /* |
614 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
615 | * This happens when finding a copied up overlay inode for a renamed | |
616 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
617 | * by origin because lower fs does not support file handles. | |
618 | */ | |
619 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
620 | return false; |
621 | ||
622 | /* | |
623 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
624 | * This happens when finding a lower alias for a copied up hard link. | |
625 | */ | |
626 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
627 | return false; | |
628 | ||
629 | return true; | |
630 | } | |
631 | ||
cd3d03d0 AG |
632 | /* |
633 | * Does overlay inode need to be hashed by lower inode? | |
634 | */ | |
635 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
636 | struct dentry *lower, struct dentry *index) | |
637 | { | |
638 | struct ovl_fs *ofs = sb->s_fs_info; | |
639 | ||
640 | /* No, if pure upper */ | |
641 | if (!lower) | |
642 | return false; | |
643 | ||
644 | /* Yes, if already indexed */ | |
645 | if (index) | |
646 | return true; | |
647 | ||
648 | /* Yes, if won't be copied up */ | |
649 | if (!ofs->upper_mnt) | |
650 | return true; | |
651 | ||
652 | /* No, if lower hardlink is or will be broken on copy up */ | |
653 | if ((upper || !ovl_indexdir(sb)) && | |
654 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
655 | return false; | |
656 | ||
657 | /* No, if non-indexed upper with NFS export */ | |
658 | if (sb->s_export_op && upper) | |
659 | return false; | |
660 | ||
661 | /* Otherwise, hash by lower inode for fsnotify */ | |
662 | return true; | |
663 | } | |
664 | ||
6eaf0111 AG |
665 | struct inode *ovl_get_inode(struct dentry *dentry, struct dentry *upperdentry, |
666 | struct dentry *index) | |
51f7e52d | 667 | { |
cd3d03d0 | 668 | struct super_block *sb = dentry->d_sb; |
09d8b586 MS |
669 | struct dentry *lowerdentry = ovl_dentry_lower(dentry); |
670 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; | |
51f7e52d | 671 | struct inode *inode; |
cd3d03d0 | 672 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, index); |
d2c9584e | 673 | bool is_dir; |
6eaf0111 | 674 | |
09d8b586 MS |
675 | if (!realinode) |
676 | realinode = d_inode(lowerdentry); | |
677 | ||
6eaf0111 | 678 | /* |
cd3d03d0 AG |
679 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
680 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 681 | */ |
d2c9584e | 682 | is_dir = S_ISDIR(realinode->i_mode); |
cd3d03d0 AG |
683 | if (upperdentry || bylower) { |
684 | struct inode *key = d_inode(bylower ? lowerdentry : | |
685 | upperdentry); | |
d2c9584e | 686 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 687 | |
cd3d03d0 | 688 | inode = iget5_locked(sb, (unsigned long) key, |
b9ac5c27 | 689 | ovl_inode_test, ovl_inode_set, key); |
09d8b586 | 690 | if (!inode) |
b9ac5c27 | 691 | goto out_nomem; |
09d8b586 | 692 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
693 | /* |
694 | * Verify that the underlying files stored in the inode | |
695 | * match those in the dentry. | |
696 | */ | |
697 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry)) { | |
698 | iput(inode); | |
699 | inode = ERR_PTR(-ESTALE); | |
700 | goto out; | |
701 | } | |
702 | ||
09d8b586 MS |
703 | dput(upperdentry); |
704 | goto out; | |
705 | } | |
e6d2ebdd | 706 | |
d2c9584e AG |
707 | /* Recalculate nlink for non-dir due to indexing */ |
708 | if (!is_dir) | |
709 | nlink = ovl_get_nlink(lowerdentry, upperdentry, nlink); | |
5f8415d6 | 710 | set_nlink(inode, nlink); |
e6d2ebdd | 711 | } else { |
cd3d03d0 AG |
712 | /* Lower hardlink that will be broken on copy up */ |
713 | inode = new_inode(sb); | |
e6d2ebdd | 714 | if (!inode) |
b9ac5c27 | 715 | goto out_nomem; |
e9be9d5e | 716 | } |
e6d2ebdd | 717 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev); |
09d8b586 | 718 | ovl_inode_init(inode, upperdentry, lowerdentry); |
13c72075 MS |
719 | |
720 | if (upperdentry && ovl_is_impuredir(upperdentry)) | |
721 | ovl_set_flag(OVL_IMPURE, inode); | |
722 | ||
b79e05aa | 723 | /* Check for non-merge dir that may have whiteouts */ |
d2c9584e | 724 | if (is_dir) { |
b79e05aa AG |
725 | struct ovl_entry *oe = dentry->d_fsdata; |
726 | ||
727 | if (((upperdentry && lowerdentry) || oe->numlower > 1) || | |
728 | ovl_check_origin_xattr(upperdentry ?: lowerdentry)) { | |
729 | ovl_set_flag(OVL_WHITEOUTS, inode); | |
730 | } | |
731 | } | |
732 | ||
e6d2ebdd MS |
733 | if (inode->i_state & I_NEW) |
734 | unlock_new_inode(inode); | |
735 | out: | |
e9be9d5e | 736 | return inode; |
b9ac5c27 MS |
737 | |
738 | out_nomem: | |
739 | inode = ERR_PTR(-ENOMEM); | |
740 | goto out; | |
e9be9d5e | 741 | } |