Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
e9be9d5e MS |
2 | /* |
3 | * | |
4 | * Copyright (C) 2011 Novell Inc. | |
e9be9d5e MS |
5 | */ |
6 | ||
7 | #include <linux/fs.h> | |
8 | #include <linux/slab.h> | |
5b825c3a | 9 | #include <linux/cred.h> |
e9be9d5e | 10 | #include <linux/xattr.h> |
5201dc44 | 11 | #include <linux/posix_acl.h> |
5f8415d6 | 12 | #include <linux/ratelimit.h> |
10c5db28 | 13 | #include <linux/fiemap.h> |
e9be9d5e MS |
14 | #include "overlayfs.h" |
15 | ||
ba1e563c | 16 | |
e9be9d5e MS |
17 | int ovl_setattr(struct dentry *dentry, struct iattr *attr) |
18 | { | |
19 | int err; | |
997336f2 | 20 | bool full_copy_up = false; |
e9be9d5e | 21 | struct dentry *upperdentry; |
1175b6b8 | 22 | const struct cred *old_cred; |
e9be9d5e | 23 | |
31051c85 | 24 | err = setattr_prepare(dentry, attr); |
cf9a6784 MS |
25 | if (err) |
26 | return err; | |
27 | ||
e9be9d5e MS |
28 | err = ovl_want_write(dentry); |
29 | if (err) | |
30 | goto out; | |
31 | ||
5812160e MS |
32 | if (attr->ia_valid & ATTR_SIZE) { |
33 | struct inode *realinode = d_inode(ovl_dentry_real(dentry)); | |
34 | ||
35 | err = -ETXTBSY; | |
36 | if (atomic_read(&realinode->i_writecount) < 0) | |
37 | goto out_drop_write; | |
997336f2 VG |
38 | |
39 | /* Truncate should trigger data copy up as well */ | |
40 | full_copy_up = true; | |
5812160e MS |
41 | } |
42 | ||
997336f2 VG |
43 | if (!full_copy_up) |
44 | err = ovl_copy_up(dentry); | |
45 | else | |
46 | err = ovl_copy_up_with_data(dentry); | |
acff81ec | 47 | if (!err) { |
5812160e MS |
48 | struct inode *winode = NULL; |
49 | ||
acff81ec MS |
50 | upperdentry = ovl_dentry_upper(dentry); |
51 | ||
5812160e MS |
52 | if (attr->ia_valid & ATTR_SIZE) { |
53 | winode = d_inode(upperdentry); | |
54 | err = get_write_access(winode); | |
55 | if (err) | |
56 | goto out_drop_write; | |
57 | } | |
58 | ||
b99c2d91 MS |
59 | if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) |
60 | attr->ia_valid &= ~ATTR_MODE; | |
61 | ||
e67f0216 | 62 | /* |
15fd2ea9 VG |
63 | * We might have to translate ovl file into real file object |
64 | * once use cases emerge. For now, simply don't let underlying | |
65 | * filesystem rely on attr->ia_file | |
e67f0216 VG |
66 | */ |
67 | attr->ia_valid &= ~ATTR_FILE; | |
68 | ||
15fd2ea9 VG |
69 | /* |
70 | * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN | |
71 | * set. Overlayfs does not pass O_TRUNC flag to underlying | |
72 | * filesystem during open -> do not pass ATTR_OPEN. This | |
73 | * disables optimization in fuse which assumes open(O_TRUNC) | |
74 | * already set file size to 0. But we never passed O_TRUNC to | |
75 | * fuse. So by clearing ATTR_OPEN, fuse will be forced to send | |
76 | * setattr request to server. | |
77 | */ | |
78 | attr->ia_valid &= ~ATTR_OPEN; | |
79 | ||
5955102c | 80 | inode_lock(upperdentry->d_inode); |
1175b6b8 | 81 | old_cred = ovl_override_creds(dentry->d_sb); |
e9be9d5e | 82 | err = notify_change(upperdentry, attr, NULL); |
1175b6b8 | 83 | revert_creds(old_cred); |
b81de061 KK |
84 | if (!err) |
85 | ovl_copyattr(upperdentry->d_inode, dentry->d_inode); | |
5955102c | 86 | inode_unlock(upperdentry->d_inode); |
5812160e MS |
87 | |
88 | if (winode) | |
89 | put_write_access(winode); | |
e9be9d5e | 90 | } |
5812160e | 91 | out_drop_write: |
e9be9d5e MS |
92 | ovl_drop_write(dentry); |
93 | out: | |
94 | return err; | |
95 | } | |
96 | ||
07f1e596 | 97 | static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid) |
da309e8c | 98 | { |
0f831ec8 | 99 | bool samefs = ovl_same_fs(dentry->d_sb); |
e487d889 | 100 | unsigned int xinobits = ovl_xino_bits(dentry->d_sb); |
dfe51d47 | 101 | unsigned int xinoshift = 64 - xinobits; |
da309e8c AG |
102 | |
103 | if (samefs) { | |
104 | /* | |
105 | * When all layers are on the same fs, all real inode | |
106 | * number are unique, so we use the overlay st_dev, | |
107 | * which is friendly to du -x. | |
108 | */ | |
109 | stat->dev = dentry->d_sb->s_dev; | |
e487d889 AG |
110 | return 0; |
111 | } else if (xinobits) { | |
e487d889 AG |
112 | /* |
113 | * All inode numbers of underlying fs should not be using the | |
114 | * high xinobits, so we use high xinobits to partition the | |
115 | * overlay st_ino address space. The high bits holds the fsid | |
dfe51d47 AG |
116 | * (upper fsid is 0). The lowest xinobit is reserved for mapping |
117 | * the non-peresistent inode numbers range in case of overflow. | |
118 | * This way all overlay inode numbers are unique and use the | |
119 | * overlay st_dev. | |
e487d889 | 120 | */ |
926e94d7 | 121 | if (likely(!(stat->ino >> xinoshift))) { |
dfe51d47 | 122 | stat->ino |= ((u64)fsid) << (xinoshift + 1); |
e487d889 AG |
123 | stat->dev = dentry->d_sb->s_dev; |
124 | return 0; | |
926e94d7 AG |
125 | } else if (ovl_xino_warn(dentry->d_sb)) { |
126 | pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n", | |
127 | dentry, stat->ino, xinobits); | |
e487d889 AG |
128 | } |
129 | } | |
130 | ||
131 | /* The inode could not be mapped to a unified st_ino address space */ | |
132 | if (S_ISDIR(dentry->d_inode->i_mode)) { | |
da309e8c AG |
133 | /* |
134 | * Always use the overlay st_dev for directories, so 'find | |
135 | * -xdev' will scan the entire overlay mount and won't cross the | |
136 | * overlay mount boundaries. | |
137 | * | |
138 | * If not all layers are on the same fs the pair {real st_ino; | |
139 | * overlay st_dev} is not unique, so use the non persistent | |
140 | * overlay st_ino for directories. | |
141 | */ | |
142 | stat->dev = dentry->d_sb->s_dev; | |
143 | stat->ino = dentry->d_inode->i_ino; | |
b7bf9908 | 144 | } else { |
da309e8c AG |
145 | /* |
146 | * For non-samefs setup, if we cannot map all layers st_ino | |
147 | * to a unified address space, we need to make sure that st_dev | |
b7bf9908 AG |
148 | * is unique per underlying fs, so we use the unique anonymous |
149 | * bdev assigned to the underlying fs. | |
da309e8c | 150 | */ |
07f1e596 | 151 | stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev; |
da309e8c AG |
152 | } |
153 | ||
154 | return 0; | |
155 | } | |
156 | ||
5b712091 MS |
157 | int ovl_getattr(const struct path *path, struct kstat *stat, |
158 | u32 request_mask, unsigned int flags) | |
e9be9d5e | 159 | { |
a528d35e | 160 | struct dentry *dentry = path->dentry; |
72b608f0 | 161 | enum ovl_path_type type; |
e9be9d5e | 162 | struct path realpath; |
1175b6b8 | 163 | const struct cred *old_cred; |
5b712091 | 164 | bool is_dir = S_ISDIR(dentry->d_inode->i_mode); |
07f1e596 | 165 | int fsid = 0; |
1175b6b8 | 166 | int err; |
67d756c2 VG |
167 | bool metacopy_blocks = false; |
168 | ||
169 | metacopy_blocks = ovl_is_metacopy_dentry(dentry); | |
e9be9d5e | 170 | |
72b608f0 | 171 | type = ovl_path_real(dentry, &realpath); |
1175b6b8 | 172 | old_cred = ovl_override_creds(dentry->d_sb); |
a528d35e | 173 | err = vfs_getattr(&realpath, stat, request_mask, flags); |
72b608f0 AG |
174 | if (err) |
175 | goto out; | |
176 | ||
177 | /* | |
da309e8c AG |
178 | * For non-dir or same fs, we use st_ino of the copy up origin. |
179 | * This guaranties constant st_dev/st_ino across copy up. | |
e487d889 AG |
180 | * With xino feature and non-samefs, we use st_ino of the copy up |
181 | * origin masked with high bits that represent the layer id. | |
72b608f0 | 182 | * |
da309e8c | 183 | * If lower filesystem supports NFS file handles, this also guaranties |
72b608f0 AG |
184 | * persistent st_ino across mount cycle. |
185 | */ | |
0f831ec8 | 186 | if (!is_dir || ovl_same_dev(dentry->d_sb)) { |
da309e8c | 187 | if (!OVL_TYPE_UPPER(type)) { |
07f1e596 | 188 | fsid = ovl_layer_lower(dentry)->fsid; |
da309e8c | 189 | } else if (OVL_TYPE_ORIGIN(type)) { |
72b608f0 | 190 | struct kstat lowerstat; |
67d756c2 VG |
191 | u32 lowermask = STATX_INO | STATX_BLOCKS | |
192 | (!is_dir ? STATX_NLINK : 0); | |
72b608f0 AG |
193 | |
194 | ovl_path_lower(dentry, &realpath); | |
195 | err = vfs_getattr(&realpath, &lowerstat, | |
5b712091 | 196 | lowermask, flags); |
72b608f0 AG |
197 | if (err) |
198 | goto out; | |
199 | ||
72b608f0 | 200 | /* |
359f392c | 201 | * Lower hardlinks may be broken on copy up to different |
72b608f0 AG |
202 | * upper files, so we cannot use the lower origin st_ino |
203 | * for those different files, even for the same fs case. | |
86eaa130 AG |
204 | * |
205 | * Similarly, several redirected dirs can point to the | |
206 | * same dir on a lower layer. With the "verify_lower" | |
207 | * feature, we do not use the lower origin st_ino, if | |
208 | * we haven't verified that this redirect is unique. | |
209 | * | |
359f392c | 210 | * With inodes index enabled, it is safe to use st_ino |
86eaa130 AG |
211 | * of an indexed origin. The index validates that the |
212 | * upper hardlink is not broken and that a redirected | |
213 | * dir is the only redirect to that origin. | |
72b608f0 | 214 | */ |
86eaa130 AG |
215 | if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) || |
216 | (!ovl_verify_lower(dentry->d_sb) && | |
9f99e50d | 217 | (is_dir || lowerstat.nlink == 1))) { |
07f1e596 | 218 | fsid = ovl_layer_lower(dentry)->fsid; |
b7bf9908 | 219 | stat->ino = lowerstat.ino; |
9f99e50d | 220 | } |
67d756c2 VG |
221 | |
222 | /* | |
223 | * If we are querying a metacopy dentry and lower | |
224 | * dentry is data dentry, then use the blocks we | |
225 | * queried just now. We don't have to do additional | |
226 | * vfs_getattr(). If lower itself is metacopy, then | |
227 | * additional vfs_getattr() is unavoidable. | |
228 | */ | |
229 | if (metacopy_blocks && | |
230 | realpath.dentry == ovl_dentry_lowerdata(dentry)) { | |
231 | stat->blocks = lowerstat.blocks; | |
232 | metacopy_blocks = false; | |
233 | } | |
234 | } | |
235 | ||
236 | if (metacopy_blocks) { | |
237 | /* | |
238 | * If lower is not same as lowerdata or if there was | |
239 | * no origin on upper, we can end up here. | |
240 | */ | |
241 | struct kstat lowerdatastat; | |
242 | u32 lowermask = STATX_BLOCKS; | |
243 | ||
244 | ovl_path_lowerdata(dentry, &realpath); | |
245 | err = vfs_getattr(&realpath, &lowerdatastat, | |
246 | lowermask, flags); | |
247 | if (err) | |
248 | goto out; | |
249 | stat->blocks = lowerdatastat.blocks; | |
72b608f0 | 250 | } |
72b608f0 | 251 | } |
5b712091 | 252 | |
07f1e596 | 253 | err = ovl_map_dev_ino(dentry, stat, fsid); |
da309e8c AG |
254 | if (err) |
255 | goto out; | |
256 | ||
5b712091 MS |
257 | /* |
258 | * It's probably not worth it to count subdirs to get the | |
259 | * correct link count. nlink=1 seems to pacify 'find' and | |
260 | * other utilities. | |
261 | */ | |
262 | if (is_dir && OVL_TYPE_MERGE(type)) | |
263 | stat->nlink = 1; | |
264 | ||
5f8415d6 AG |
265 | /* |
266 | * Return the overlay inode nlinks for indexed upper inodes. | |
267 | * Overlay inode nlink counts the union of the upper hardlinks | |
268 | * and non-covered lower hardlinks. It does not include the upper | |
269 | * index hardlink. | |
270 | */ | |
271 | if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
272 | stat->nlink = dentry->d_inode->i_nlink; | |
273 | ||
72b608f0 | 274 | out: |
1175b6b8 | 275 | revert_creds(old_cred); |
72b608f0 | 276 | |
1175b6b8 | 277 | return err; |
e9be9d5e MS |
278 | } |
279 | ||
280 | int ovl_permission(struct inode *inode, int mask) | |
281 | { | |
09d8b586 MS |
282 | struct inode *upperinode = ovl_inode_upper(inode); |
283 | struct inode *realinode = upperinode ?: ovl_inode_lower(inode); | |
c0ca3d70 | 284 | const struct cred *old_cred; |
e9be9d5e MS |
285 | int err; |
286 | ||
e9be9d5e | 287 | /* Careful in RCU walk mode */ |
e9be9d5e MS |
288 | if (!realinode) { |
289 | WARN_ON(!(mask & MAY_NOT_BLOCK)); | |
a999d7e1 | 290 | return -ECHILD; |
e9be9d5e MS |
291 | } |
292 | ||
c0ca3d70 VG |
293 | /* |
294 | * Check overlay inode with the creds of task and underlying inode | |
295 | * with creds of mounter | |
296 | */ | |
297 | err = generic_permission(inode, mask); | |
298 | if (err) | |
299 | return err; | |
300 | ||
ec7ba118 MS |
301 | old_cred = ovl_override_creds(inode->i_sb); |
302 | if (!upperinode && | |
303 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 304 | mask &= ~(MAY_WRITE | MAY_APPEND); |
ec7ba118 | 305 | /* Make sure mounter can read file for copy up later */ |
500cac3c VG |
306 | mask |= MAY_READ; |
307 | } | |
9c630ebe | 308 | err = inode_permission(realinode, mask); |
c0ca3d70 VG |
309 | revert_creds(old_cred); |
310 | ||
311 | return err; | |
e9be9d5e MS |
312 | } |
313 | ||
6b255391 | 314 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
315 | struct inode *inode, |
316 | struct delayed_call *done) | |
e9be9d5e | 317 | { |
1175b6b8 VG |
318 | const struct cred *old_cred; |
319 | const char *p; | |
e9be9d5e | 320 | |
6b255391 AV |
321 | if (!dentry) |
322 | return ERR_PTR(-ECHILD); | |
323 | ||
1175b6b8 | 324 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 325 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
326 | revert_creds(old_cred); |
327 | return p; | |
e9be9d5e MS |
328 | } |
329 | ||
610afc0b | 330 | bool ovl_is_private_xattr(struct super_block *sb, const char *name) |
e9be9d5e | 331 | { |
2d2f2d73 MS |
332 | struct ovl_fs *ofs = sb->s_fs_info; |
333 | ||
334 | if (ofs->config.userxattr) | |
335 | return strncmp(name, OVL_XATTR_USER_PREFIX, | |
336 | sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0; | |
337 | else | |
338 | return strncmp(name, OVL_XATTR_TRUSTED_PREFIX, | |
339 | sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0; | |
e9be9d5e MS |
340 | } |
341 | ||
1d88f183 MS |
342 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
343 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
344 | { |
345 | int err; | |
1d88f183 MS |
346 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
347 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
1175b6b8 | 348 | const struct cred *old_cred; |
e9be9d5e MS |
349 | |
350 | err = ovl_want_write(dentry); | |
351 | if (err) | |
352 | goto out; | |
353 | ||
1d88f183 MS |
354 | if (!value && !upperdentry) { |
355 | err = vfs_getxattr(realdentry, name, NULL, 0); | |
0e585ccc AG |
356 | if (err < 0) |
357 | goto out_drop_write; | |
358 | } | |
359 | ||
1d88f183 MS |
360 | if (!upperdentry) { |
361 | err = ovl_copy_up(dentry); | |
362 | if (err) | |
363 | goto out_drop_write; | |
e9be9d5e | 364 | |
1d88f183 MS |
365 | realdentry = ovl_dentry_upper(dentry); |
366 | } | |
0e585ccc | 367 | |
1175b6b8 | 368 | old_cred = ovl_override_creds(dentry->d_sb); |
0e585ccc | 369 | if (value) |
1d88f183 | 370 | err = vfs_setxattr(realdentry, name, value, size, flags); |
0e585ccc AG |
371 | else { |
372 | WARN_ON(flags != XATTR_REPLACE); | |
1d88f183 | 373 | err = vfs_removexattr(realdentry, name); |
0e585ccc | 374 | } |
1175b6b8 | 375 | revert_creds(old_cred); |
e9be9d5e | 376 | |
d9854c87 MS |
377 | /* copy c/mtime */ |
378 | ovl_copyattr(d_inode(realdentry), inode); | |
379 | ||
e9be9d5e MS |
380 | out_drop_write: |
381 | ovl_drop_write(dentry); | |
382 | out: | |
383 | return err; | |
384 | } | |
385 | ||
1d88f183 | 386 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 387 | void *value, size_t size) |
e9be9d5e | 388 | { |
1175b6b8 VG |
389 | ssize_t res; |
390 | const struct cred *old_cred; | |
1d88f183 MS |
391 | struct dentry *realdentry = |
392 | ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry); | |
52148463 | 393 | |
1175b6b8 VG |
394 | old_cred = ovl_override_creds(dentry->d_sb); |
395 | res = vfs_getxattr(realdentry, name, value, size); | |
396 | revert_creds(old_cred); | |
397 | return res; | |
e9be9d5e MS |
398 | } |
399 | ||
610afc0b | 400 | static bool ovl_can_list(struct super_block *sb, const char *s) |
a082c6f6 | 401 | { |
8f6ee74c MS |
402 | /* Never list private (.overlay) */ |
403 | if (ovl_is_private_xattr(sb, s)) | |
404 | return false; | |
405 | ||
a082c6f6 MS |
406 | /* List all non-trusted xatts */ |
407 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) | |
408 | return true; | |
409 | ||
8f6ee74c MS |
410 | /* list other trusted for superuser only */ |
411 | return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN); | |
a082c6f6 MS |
412 | } |
413 | ||
e9be9d5e MS |
414 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
415 | { | |
b581755b | 416 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 417 | ssize_t res; |
7cb35119 MS |
418 | size_t len; |
419 | char *s; | |
1175b6b8 | 420 | const struct cred *old_cred; |
e9be9d5e | 421 | |
1175b6b8 | 422 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 423 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 424 | revert_creds(old_cred); |
e9be9d5e MS |
425 | if (res <= 0 || size == 0) |
426 | return res; | |
427 | ||
e9be9d5e | 428 | /* filter out private xattrs */ |
7cb35119 MS |
429 | for (s = list, len = res; len;) { |
430 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 431 | |
7cb35119 MS |
432 | /* underlying fs providing us with an broken xattr list? */ |
433 | if (WARN_ON(slen > len)) | |
434 | return -EIO; | |
e9be9d5e | 435 | |
7cb35119 | 436 | len -= slen; |
610afc0b | 437 | if (!ovl_can_list(dentry->d_sb, s)) { |
e9be9d5e | 438 | res -= slen; |
7cb35119 | 439 | memmove(s, s + slen, len); |
e9be9d5e | 440 | } else { |
7cb35119 | 441 | s += slen; |
e9be9d5e MS |
442 | } |
443 | } | |
444 | ||
445 | return res; | |
446 | } | |
447 | ||
39a25b2b VG |
448 | struct posix_acl *ovl_get_acl(struct inode *inode, int type) |
449 | { | |
09d8b586 | 450 | struct inode *realinode = ovl_inode_real(inode); |
1175b6b8 VG |
451 | const struct cred *old_cred; |
452 | struct posix_acl *acl; | |
39a25b2b | 453 | |
5201dc44 | 454 | if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode)) |
39a25b2b VG |
455 | return NULL; |
456 | ||
1175b6b8 | 457 | old_cred = ovl_override_creds(inode->i_sb); |
5201dc44 | 458 | acl = get_acl(realinode, type); |
1175b6b8 VG |
459 | revert_creds(old_cred); |
460 | ||
461 | return acl; | |
39a25b2b VG |
462 | } |
463 | ||
95582b00 | 464 | int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags) |
d719e8f2 | 465 | { |
8f35cf51 MS |
466 | if (flags & S_ATIME) { |
467 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
468 | struct path upperpath = { | |
08f4c7c8 | 469 | .mnt = ovl_upper_mnt(ofs), |
8f35cf51 MS |
470 | .dentry = ovl_upperdentry_dereference(OVL_I(inode)), |
471 | }; | |
472 | ||
473 | if (upperpath.dentry) { | |
474 | touch_atime(&upperpath); | |
475 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
476 | } | |
d719e8f2 | 477 | } |
d719e8f2 MS |
478 | return 0; |
479 | } | |
480 | ||
9e142c41 MS |
481 | static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
482 | u64 start, u64 len) | |
483 | { | |
484 | int err; | |
c11faf32 | 485 | struct inode *realinode = ovl_inode_realdata(inode); |
9e142c41 MS |
486 | const struct cred *old_cred; |
487 | ||
488 | if (!realinode->i_op->fiemap) | |
489 | return -EOPNOTSUPP; | |
490 | ||
491 | old_cred = ovl_override_creds(inode->i_sb); | |
492 | err = realinode->i_op->fiemap(realinode, fieinfo, start, len); | |
493 | revert_creds(old_cred); | |
494 | ||
495 | return err; | |
496 | } | |
497 | ||
e9be9d5e MS |
498 | static const struct inode_operations ovl_file_inode_operations = { |
499 | .setattr = ovl_setattr, | |
500 | .permission = ovl_permission, | |
501 | .getattr = ovl_getattr, | |
e9be9d5e | 502 | .listxattr = ovl_listxattr, |
39a25b2b | 503 | .get_acl = ovl_get_acl, |
d719e8f2 | 504 | .update_time = ovl_update_time, |
9e142c41 | 505 | .fiemap = ovl_fiemap, |
e9be9d5e MS |
506 | }; |
507 | ||
508 | static const struct inode_operations ovl_symlink_inode_operations = { | |
509 | .setattr = ovl_setattr, | |
6b255391 | 510 | .get_link = ovl_get_link, |
e9be9d5e | 511 | .getattr = ovl_getattr, |
e9be9d5e | 512 | .listxattr = ovl_listxattr, |
d719e8f2 | 513 | .update_time = ovl_update_time, |
e9be9d5e MS |
514 | }; |
515 | ||
9e142c41 MS |
516 | static const struct inode_operations ovl_special_inode_operations = { |
517 | .setattr = ovl_setattr, | |
518 | .permission = ovl_permission, | |
519 | .getattr = ovl_getattr, | |
520 | .listxattr = ovl_listxattr, | |
521 | .get_acl = ovl_get_acl, | |
522 | .update_time = ovl_update_time, | |
523 | }; | |
524 | ||
69383c59 | 525 | static const struct address_space_operations ovl_aops = { |
5b910bd6 AG |
526 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
527 | .direct_IO = noop_direct_IO, | |
528 | }; | |
529 | ||
b1eaa950 AG |
530 | /* |
531 | * It is possible to stack overlayfs instance on top of another | |
a5a84682 | 532 | * overlayfs instance as lower layer. We need to annotate the |
b1eaa950 AG |
533 | * stackable i_mutex locks according to stack level of the super |
534 | * block instance. An overlayfs instance can never be in stack | |
535 | * depth 0 (there is always a real fs below it). An overlayfs | |
536 | * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth]. | |
537 | * | |
538 | * For example, here is a snip from /proc/lockdep_chains after | |
539 | * dir_iterate of nested overlayfs: | |
540 | * | |
541 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
542 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
543 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
b1f9d385 AG |
544 | * |
545 | * Locking order w.r.t ovl_want_write() is important for nested overlayfs. | |
546 | * | |
547 | * This chain is valid: | |
548 | * - inode->i_rwsem (inode_lock[2]) | |
549 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
550 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
551 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
552 | * | |
553 | * And this chain is valid: | |
554 | * - inode->i_rwsem (inode_lock[2]) | |
555 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
556 | * - lowerinode->i_rwsem (inode_lock[1]) | |
557 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
558 | * | |
559 | * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is | |
560 | * held, because it is in reverse order of the non-nested case using the same | |
561 | * upper fs: | |
562 | * - inode->i_rwsem (inode_lock[1]) | |
563 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
564 | * - OVL_I(inode)->lock (ovl_inode_lock[1]) | |
b1eaa950 AG |
565 | */ |
566 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
567 | ||
568 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
569 | { | |
570 | #ifdef CONFIG_LOCKDEP | |
571 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
572 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 573 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
574 | |
575 | int depth = inode->i_sb->s_stack_depth - 1; | |
576 | ||
577 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
578 | depth = 0; | |
579 | ||
580 | if (S_ISDIR(inode->i_mode)) | |
581 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
582 | else | |
583 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
584 | |
585 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
586 | #endif |
587 | } | |
588 | ||
4d314f78 AG |
589 | static void ovl_next_ino(struct inode *inode) |
590 | { | |
591 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
592 | ||
593 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
594 | if (unlikely(!inode->i_ino)) | |
595 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
596 | } | |
597 | ||
62c832ed | 598 | static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid) |
e9be9d5e | 599 | { |
12574a9f | 600 | int xinobits = ovl_xino_bits(inode->i_sb); |
dfe51d47 | 601 | unsigned int xinoshift = 64 - xinobits; |
12574a9f | 602 | |
695b46e7 | 603 | /* |
6dde1e42 AG |
604 | * When d_ino is consistent with st_ino (samefs or i_ino has enough |
605 | * bits to encode layer), set the same value used for st_ino to i_ino, | |
606 | * so inode number exposed via /proc/locks and a like will be | |
607 | * consistent with d_ino and st_ino values. An i_ino value inconsistent | |
735c907d | 608 | * with d_ino also causes nfsd readdirplus to fail. |
695b46e7 | 609 | */ |
4d314f78 | 610 | inode->i_ino = ino; |
dfe51d47 AG |
611 | if (ovl_same_fs(inode->i_sb)) { |
612 | return; | |
613 | } else if (xinobits && likely(!(ino >> xinoshift))) { | |
614 | inode->i_ino |= (unsigned long)fsid << (xinoshift + 1); | |
615 | return; | |
616 | } | |
617 | ||
618 | /* | |
619 | * For directory inodes on non-samefs with xino disabled or xino | |
620 | * overflow, we allocate a non-persistent inode number, to be used for | |
621 | * resolving st_ino collisions in ovl_map_dev_ino(). | |
622 | * | |
623 | * To avoid ino collision with legitimate xino values from upper | |
624 | * layer (fsid 0), use the lowest xinobit to map the non | |
625 | * persistent inode numbers to the unified st_ino address space. | |
626 | */ | |
627 | if (S_ISDIR(inode->i_mode)) { | |
4d314f78 | 628 | ovl_next_ino(inode); |
dfe51d47 AG |
629 | if (xinobits) { |
630 | inode->i_ino &= ~0UL >> xinobits; | |
631 | inode->i_ino |= 1UL << xinoshift; | |
632 | } | |
12574a9f | 633 | } |
62c832ed AG |
634 | } |
635 | ||
636 | void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip, | |
637 | unsigned long ino, int fsid) | |
638 | { | |
639 | struct inode *realinode; | |
640 | ||
641 | if (oip->upperdentry) | |
642 | OVL_I(inode)->__upperdentry = oip->upperdentry; | |
643 | if (oip->lowerpath && oip->lowerpath->dentry) | |
644 | OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry)); | |
645 | if (oip->lowerdata) | |
646 | OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata)); | |
647 | ||
648 | realinode = ovl_inode_real(inode); | |
649 | ovl_copyattr(realinode, inode); | |
650 | ovl_copyflags(realinode, inode); | |
651 | ovl_map_ino(inode, ino, fsid); | |
652 | } | |
653 | ||
654 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
655 | { | |
e9be9d5e | 656 | inode->i_mode = mode; |
d719e8f2 | 657 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
658 | #ifdef CONFIG_FS_POSIX_ACL |
659 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
660 | #endif | |
e9be9d5e | 661 | |
b1eaa950 AG |
662 | ovl_lockdep_annotate_inode_mutex_key(inode); |
663 | ||
ca4c8a3a MS |
664 | switch (mode & S_IFMT) { |
665 | case S_IFREG: | |
666 | inode->i_op = &ovl_file_inode_operations; | |
d1d04ef8 | 667 | inode->i_fop = &ovl_file_operations; |
5b910bd6 | 668 | inode->i_mapping->a_ops = &ovl_aops; |
ca4c8a3a MS |
669 | break; |
670 | ||
e9be9d5e | 671 | case S_IFDIR: |
e9be9d5e MS |
672 | inode->i_op = &ovl_dir_inode_operations; |
673 | inode->i_fop = &ovl_dir_operations; | |
674 | break; | |
675 | ||
676 | case S_IFLNK: | |
677 | inode->i_op = &ovl_symlink_inode_operations; | |
678 | break; | |
679 | ||
51f7e52d | 680 | default: |
9e142c41 | 681 | inode->i_op = &ovl_special_inode_operations; |
ca4c8a3a | 682 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 683 | break; |
51f7e52d MS |
684 | } |
685 | } | |
e9be9d5e | 686 | |
5f8415d6 AG |
687 | /* |
688 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
689 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
690 | * upper inode, the following nlink modifying operations can happen: | |
691 | * | |
692 | * 1. Lower hardlink copy up | |
693 | * 2. Upper hardlink created, unlinked or renamed over | |
694 | * 3. Lower hardlink whiteout or renamed over | |
695 | * | |
696 | * For the first, copy up case, the union nlink does not change, whether the | |
697 | * operation succeeds or fails, but the upper inode nlink may change. | |
698 | * Therefore, before copy up, we store the union nlink value relative to the | |
2d2f2d73 | 699 | * lower inode nlink in the index inode xattr .overlay.nlink. |
5f8415d6 AG |
700 | * |
701 | * For the second, upper hardlink case, the union nlink should be incremented | |
702 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
703 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
704 | * value relative to the upper inode nlink in the index inode. | |
705 | * | |
706 | * For the last, lower cover up case, we simplify things by preceding the | |
707 | * whiteout or cover up with copy up. This makes sure that there is an index | |
708 | * upper inode where the nlink xattr can be stored before the copied up upper | |
709 | * entry is unlink. | |
710 | */ | |
711 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
712 | ||
713 | /* | |
714 | * On-disk format for indexed nlink: | |
715 | * | |
716 | * nlink relative to the upper inode - "U[+-]NUM" | |
717 | * nlink relative to the lower inode - "L[+-]NUM" | |
718 | */ | |
719 | ||
720 | static int ovl_set_nlink_common(struct dentry *dentry, | |
721 | struct dentry *realdentry, const char *format) | |
722 | { | |
723 | struct inode *inode = d_inode(dentry); | |
724 | struct inode *realinode = d_inode(realdentry); | |
725 | char buf[13]; | |
726 | int len; | |
727 | ||
728 | len = snprintf(buf, sizeof(buf), format, | |
729 | (int) (inode->i_nlink - realinode->i_nlink)); | |
730 | ||
6787341a MS |
731 | if (WARN_ON(len >= sizeof(buf))) |
732 | return -EIO; | |
733 | ||
610afc0b | 734 | return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry), |
26150ab5 | 735 | OVL_XATTR_NLINK, buf, len); |
5f8415d6 AG |
736 | } |
737 | ||
738 | int ovl_set_nlink_upper(struct dentry *dentry) | |
739 | { | |
740 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
741 | } | |
742 | ||
743 | int ovl_set_nlink_lower(struct dentry *dentry) | |
744 | { | |
745 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
746 | } | |
747 | ||
610afc0b | 748 | unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry, |
caf70cb2 AG |
749 | struct dentry *upperdentry, |
750 | unsigned int fallback) | |
5f8415d6 AG |
751 | { |
752 | int nlink_diff; | |
753 | int nlink; | |
754 | char buf[13]; | |
755 | int err; | |
756 | ||
757 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
758 | return fallback; | |
759 | ||
610afc0b | 760 | err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK, |
d5dc7486 | 761 | &buf, sizeof(buf) - 1); |
5f8415d6 AG |
762 | if (err < 0) |
763 | goto fail; | |
764 | ||
765 | buf[err] = '\0'; | |
766 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
767 | (buf[1] != '+' && buf[1] != '-')) | |
768 | goto fail; | |
769 | ||
770 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
771 | if (err < 0) | |
772 | goto fail; | |
773 | ||
774 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
775 | nlink += nlink_diff; | |
776 | ||
777 | if (nlink <= 0) | |
778 | goto fail; | |
779 | ||
780 | return nlink; | |
781 | ||
782 | fail: | |
1bd0a3ae | 783 | pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n", |
5f8415d6 AG |
784 | upperdentry, err); |
785 | return fallback; | |
786 | } | |
787 | ||
ca4c8a3a | 788 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
789 | { |
790 | struct inode *inode; | |
791 | ||
792 | inode = new_inode(sb); | |
793 | if (inode) | |
62c832ed | 794 | ovl_fill_inode(inode, mode, rdev); |
51f7e52d MS |
795 | |
796 | return inode; | |
797 | } | |
798 | ||
799 | static int ovl_inode_test(struct inode *inode, void *data) | |
800 | { | |
25b7713a | 801 | return inode->i_private == data; |
51f7e52d MS |
802 | } |
803 | ||
804 | static int ovl_inode_set(struct inode *inode, void *data) | |
805 | { | |
25b7713a | 806 | inode->i_private = data; |
51f7e52d MS |
807 | return 0; |
808 | } | |
809 | ||
b9ac5c27 | 810 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
4b91c30a | 811 | struct dentry *upperdentry, bool strict) |
b9ac5c27 | 812 | { |
4b91c30a AG |
813 | /* |
814 | * For directories, @strict verify from lookup path performs consistency | |
815 | * checks, so NULL lower/upper in dentry must match NULL lower/upper in | |
816 | * inode. Non @strict verify from NFS handle decode path passes NULL for | |
817 | * 'unknown' lower/upper. | |
818 | */ | |
819 | if (S_ISDIR(inode->i_mode) && strict) { | |
31747eda AG |
820 | /* Real lower dir moved to upper layer under us? */ |
821 | if (!lowerdentry && ovl_inode_lower(inode)) | |
822 | return false; | |
823 | ||
824 | /* Lookup of an uncovered redirect origin? */ | |
825 | if (!upperdentry && ovl_inode_upper(inode)) | |
826 | return false; | |
827 | } | |
828 | ||
939ae4ef AG |
829 | /* |
830 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
831 | * This happens when finding a copied up overlay inode for a renamed | |
832 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
833 | * by origin because lower fs does not support file handles. | |
834 | */ | |
835 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
836 | return false; |
837 | ||
838 | /* | |
839 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
840 | * This happens when finding a lower alias for a copied up hard link. | |
841 | */ | |
842 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
843 | return false; | |
844 | ||
845 | return true; | |
846 | } | |
847 | ||
4b91c30a AG |
848 | struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real, |
849 | bool is_upper) | |
9436a1a3 | 850 | { |
4b91c30a | 851 | struct inode *inode, *key = d_inode(real); |
9436a1a3 AG |
852 | |
853 | inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
854 | if (!inode) | |
855 | return NULL; | |
856 | ||
4b91c30a AG |
857 | if (!ovl_verify_inode(inode, is_upper ? NULL : real, |
858 | is_upper ? real : NULL, false)) { | |
9436a1a3 AG |
859 | iput(inode); |
860 | return ERR_PTR(-ESTALE); | |
861 | } | |
862 | ||
863 | return inode; | |
864 | } | |
865 | ||
146d62e5 AG |
866 | bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir) |
867 | { | |
868 | struct inode *key = d_inode(dir); | |
869 | struct inode *trap; | |
870 | bool res; | |
871 | ||
872 | trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
873 | if (!trap) | |
874 | return false; | |
875 | ||
876 | res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) && | |
877 | !ovl_inode_lower(trap); | |
878 | ||
879 | iput(trap); | |
880 | return res; | |
881 | } | |
882 | ||
883 | /* | |
884 | * Create an inode cache entry for layer root dir, that will intentionally | |
885 | * fail ovl_verify_inode(), so any lookup that will find some layer root | |
886 | * will fail. | |
887 | */ | |
888 | struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir) | |
889 | { | |
890 | struct inode *key = d_inode(dir); | |
891 | struct inode *trap; | |
892 | ||
893 | if (!d_is_dir(dir)) | |
894 | return ERR_PTR(-ENOTDIR); | |
895 | ||
896 | trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test, | |
897 | ovl_inode_set, key); | |
898 | if (!trap) | |
899 | return ERR_PTR(-ENOMEM); | |
900 | ||
901 | if (!(trap->i_state & I_NEW)) { | |
902 | /* Conflicting layer roots? */ | |
903 | iput(trap); | |
904 | return ERR_PTR(-ELOOP); | |
905 | } | |
906 | ||
907 | trap->i_mode = S_IFDIR; | |
908 | trap->i_flags = S_DEAD; | |
909 | unlock_new_inode(trap); | |
910 | ||
911 | return trap; | |
912 | } | |
913 | ||
764baba8 AG |
914 | /* |
915 | * Does overlay inode need to be hashed by lower inode? | |
916 | */ | |
917 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
74c6e384 | 918 | struct dentry *lower, bool index) |
764baba8 AG |
919 | { |
920 | struct ovl_fs *ofs = sb->s_fs_info; | |
921 | ||
922 | /* No, if pure upper */ | |
923 | if (!lower) | |
924 | return false; | |
925 | ||
926 | /* Yes, if already indexed */ | |
927 | if (index) | |
928 | return true; | |
929 | ||
930 | /* Yes, if won't be copied up */ | |
08f4c7c8 | 931 | if (!ovl_upper_mnt(ofs)) |
764baba8 AG |
932 | return true; |
933 | ||
934 | /* No, if lower hardlink is or will be broken on copy up */ | |
935 | if ((upper || !ovl_indexdir(sb)) && | |
936 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
937 | return false; | |
938 | ||
939 | /* No, if non-indexed upper with NFS export */ | |
940 | if (sb->s_export_op && upper) | |
941 | return false; | |
942 | ||
943 | /* Otherwise, hash by lower inode for fsnotify */ | |
944 | return true; | |
945 | } | |
946 | ||
01b39dcc AG |
947 | static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode, |
948 | struct inode *key) | |
949 | { | |
950 | return newinode ? inode_insert5(newinode, (unsigned long) key, | |
951 | ovl_inode_test, ovl_inode_set, key) : | |
952 | iget5_locked(sb, (unsigned long) key, | |
953 | ovl_inode_test, ovl_inode_set, key); | |
954 | } | |
955 | ||
ac6a52eb VG |
956 | struct inode *ovl_get_inode(struct super_block *sb, |
957 | struct ovl_inode_params *oip) | |
51f7e52d | 958 | { |
610afc0b | 959 | struct ovl_fs *ofs = OVL_FS(sb); |
ac6a52eb VG |
960 | struct dentry *upperdentry = oip->upperdentry; |
961 | struct ovl_path *lowerpath = oip->lowerpath; | |
09d8b586 | 962 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; |
51f7e52d | 963 | struct inode *inode; |
12574a9f | 964 | struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL; |
ac6a52eb VG |
965 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, |
966 | oip->index); | |
300b124f | 967 | int fsid = bylower ? lowerpath->layer->fsid : 0; |
28166ab3 | 968 | bool is_dir; |
695b46e7 | 969 | unsigned long ino = 0; |
acf3062a | 970 | int err = oip->newinode ? -EEXIST : -ENOMEM; |
6eaf0111 | 971 | |
09d8b586 MS |
972 | if (!realinode) |
973 | realinode = d_inode(lowerdentry); | |
974 | ||
6eaf0111 | 975 | /* |
764baba8 AG |
976 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
977 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 978 | */ |
31747eda | 979 | is_dir = S_ISDIR(realinode->i_mode); |
764baba8 AG |
980 | if (upperdentry || bylower) { |
981 | struct inode *key = d_inode(bylower ? lowerdentry : | |
982 | upperdentry); | |
31747eda | 983 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 984 | |
01b39dcc | 985 | inode = ovl_iget5(sb, oip->newinode, key); |
09d8b586 | 986 | if (!inode) |
027065b7 | 987 | goto out_err; |
09d8b586 | 988 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
989 | /* |
990 | * Verify that the underlying files stored in the inode | |
991 | * match those in the dentry. | |
992 | */ | |
4b91c30a AG |
993 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry, |
994 | true)) { | |
b9ac5c27 | 995 | iput(inode); |
027065b7 VG |
996 | err = -ESTALE; |
997 | goto out_err; | |
b9ac5c27 MS |
998 | } |
999 | ||
09d8b586 | 1000 | dput(upperdentry); |
9cec54c8 | 1001 | kfree(oip->redirect); |
09d8b586 MS |
1002 | goto out; |
1003 | } | |
e6d2ebdd | 1004 | |
31747eda AG |
1005 | /* Recalculate nlink for non-dir due to indexing */ |
1006 | if (!is_dir) | |
610afc0b MS |
1007 | nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry, |
1008 | nlink); | |
5f8415d6 | 1009 | set_nlink(inode, nlink); |
695b46e7 | 1010 | ino = key->i_ino; |
e6d2ebdd | 1011 | } else { |
764baba8 | 1012 | /* Lower hardlink that will be broken on copy up */ |
0aceb53e | 1013 | inode = new_inode(sb); |
027065b7 VG |
1014 | if (!inode) { |
1015 | err = -ENOMEM; | |
1016 | goto out_err; | |
1017 | } | |
300b124f AG |
1018 | ino = realinode->i_ino; |
1019 | fsid = lowerpath->layer->fsid; | |
e9be9d5e | 1020 | } |
62c832ed AG |
1021 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev); |
1022 | ovl_inode_init(inode, oip, ino, fsid); | |
13c72075 | 1023 | |
610afc0b | 1024 | if (upperdentry && ovl_is_impuredir(sb, upperdentry)) |
13c72075 MS |
1025 | ovl_set_flag(OVL_IMPURE, inode); |
1026 | ||
ac6a52eb | 1027 | if (oip->index) |
0471a9cd VG |
1028 | ovl_set_flag(OVL_INDEX, inode); |
1029 | ||
9cec54c8 VG |
1030 | OVL_I(inode)->redirect = oip->redirect; |
1031 | ||
a00c2d59 VG |
1032 | if (bylower) |
1033 | ovl_set_flag(OVL_CONST_INO, inode); | |
1034 | ||
b79e05aa | 1035 | /* Check for non-merge dir that may have whiteouts */ |
31747eda | 1036 | if (is_dir) { |
ac6a52eb | 1037 | if (((upperdentry && lowerdentry) || oip->numlower > 1) || |
610afc0b | 1038 | ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) { |
b79e05aa AG |
1039 | ovl_set_flag(OVL_WHITEOUTS, inode); |
1040 | } | |
1041 | } | |
1042 | ||
e6d2ebdd MS |
1043 | if (inode->i_state & I_NEW) |
1044 | unlock_new_inode(inode); | |
1045 | out: | |
e9be9d5e | 1046 | return inode; |
b9ac5c27 | 1047 | |
027065b7 | 1048 | out_err: |
1bd0a3ae | 1049 | pr_warn_ratelimited("failed to get inode (%i)\n", err); |
027065b7 | 1050 | inode = ERR_PTR(err); |
b9ac5c27 | 1051 | goto out; |
e9be9d5e | 1052 | } |