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> |
66dbfabf MS |
14 | #include <linux/fileattr.h> |
15 | #include <linux/security.h> | |
e9be9d5e MS |
16 | #include "overlayfs.h" |
17 | ||
ba1e563c | 18 | |
549c7297 CB |
19 | int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
20 | struct iattr *attr) | |
e9be9d5e MS |
21 | { |
22 | int err; | |
997336f2 | 23 | bool full_copy_up = false; |
e9be9d5e | 24 | struct dentry *upperdentry; |
1175b6b8 | 25 | const struct cred *old_cred; |
e9be9d5e | 26 | |
2f221d6f | 27 | err = setattr_prepare(&init_user_ns, dentry, attr); |
cf9a6784 MS |
28 | if (err) |
29 | return err; | |
30 | ||
e9be9d5e MS |
31 | err = ovl_want_write(dentry); |
32 | if (err) | |
33 | goto out; | |
34 | ||
5812160e | 35 | if (attr->ia_valid & ATTR_SIZE) { |
997336f2 VG |
36 | /* Truncate should trigger data copy up as well */ |
37 | full_copy_up = true; | |
5812160e MS |
38 | } |
39 | ||
997336f2 VG |
40 | if (!full_copy_up) |
41 | err = ovl_copy_up(dentry); | |
42 | else | |
43 | err = ovl_copy_up_with_data(dentry); | |
acff81ec | 44 | if (!err) { |
5812160e MS |
45 | struct inode *winode = NULL; |
46 | ||
acff81ec MS |
47 | upperdentry = ovl_dentry_upper(dentry); |
48 | ||
5812160e MS |
49 | if (attr->ia_valid & ATTR_SIZE) { |
50 | winode = d_inode(upperdentry); | |
51 | err = get_write_access(winode); | |
52 | if (err) | |
53 | goto out_drop_write; | |
54 | } | |
55 | ||
b99c2d91 MS |
56 | if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) |
57 | attr->ia_valid &= ~ATTR_MODE; | |
58 | ||
e67f0216 | 59 | /* |
15fd2ea9 VG |
60 | * We might have to translate ovl file into real file object |
61 | * once use cases emerge. For now, simply don't let underlying | |
62 | * filesystem rely on attr->ia_file | |
e67f0216 VG |
63 | */ |
64 | attr->ia_valid &= ~ATTR_FILE; | |
65 | ||
15fd2ea9 VG |
66 | /* |
67 | * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN | |
68 | * set. Overlayfs does not pass O_TRUNC flag to underlying | |
69 | * filesystem during open -> do not pass ATTR_OPEN. This | |
70 | * disables optimization in fuse which assumes open(O_TRUNC) | |
71 | * already set file size to 0. But we never passed O_TRUNC to | |
72 | * fuse. So by clearing ATTR_OPEN, fuse will be forced to send | |
73 | * setattr request to server. | |
74 | */ | |
75 | attr->ia_valid &= ~ATTR_OPEN; | |
76 | ||
5955102c | 77 | inode_lock(upperdentry->d_inode); |
1175b6b8 | 78 | old_cred = ovl_override_creds(dentry->d_sb); |
2f221d6f | 79 | err = notify_change(&init_user_ns, upperdentry, attr, NULL); |
1175b6b8 | 80 | revert_creds(old_cred); |
b81de061 KK |
81 | if (!err) |
82 | ovl_copyattr(upperdentry->d_inode, dentry->d_inode); | |
5955102c | 83 | inode_unlock(upperdentry->d_inode); |
5812160e MS |
84 | |
85 | if (winode) | |
86 | put_write_access(winode); | |
e9be9d5e | 87 | } |
5812160e | 88 | out_drop_write: |
e9be9d5e MS |
89 | ovl_drop_write(dentry); |
90 | out: | |
91 | return err; | |
92 | } | |
93 | ||
c68e7ec5 | 94 | static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid) |
da309e8c | 95 | { |
0f831ec8 | 96 | bool samefs = ovl_same_fs(dentry->d_sb); |
e487d889 | 97 | unsigned int xinobits = ovl_xino_bits(dentry->d_sb); |
dfe51d47 | 98 | unsigned int xinoshift = 64 - xinobits; |
da309e8c AG |
99 | |
100 | if (samefs) { | |
101 | /* | |
102 | * When all layers are on the same fs, all real inode | |
103 | * number are unique, so we use the overlay st_dev, | |
104 | * which is friendly to du -x. | |
105 | */ | |
106 | stat->dev = dentry->d_sb->s_dev; | |
c68e7ec5 | 107 | return; |
e487d889 | 108 | } else if (xinobits) { |
e487d889 AG |
109 | /* |
110 | * All inode numbers of underlying fs should not be using the | |
111 | * high xinobits, so we use high xinobits to partition the | |
112 | * overlay st_ino address space. The high bits holds the fsid | |
dfe51d47 | 113 | * (upper fsid is 0). The lowest xinobit is reserved for mapping |
f48bbfb2 | 114 | * the non-persistent inode numbers range in case of overflow. |
dfe51d47 AG |
115 | * This way all overlay inode numbers are unique and use the |
116 | * overlay st_dev. | |
e487d889 | 117 | */ |
926e94d7 | 118 | if (likely(!(stat->ino >> xinoshift))) { |
dfe51d47 | 119 | stat->ino |= ((u64)fsid) << (xinoshift + 1); |
e487d889 | 120 | stat->dev = dentry->d_sb->s_dev; |
c68e7ec5 | 121 | return; |
926e94d7 AG |
122 | } else if (ovl_xino_warn(dentry->d_sb)) { |
123 | pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n", | |
124 | dentry, stat->ino, xinobits); | |
e487d889 AG |
125 | } |
126 | } | |
127 | ||
128 | /* The inode could not be mapped to a unified st_ino address space */ | |
129 | if (S_ISDIR(dentry->d_inode->i_mode)) { | |
da309e8c AG |
130 | /* |
131 | * Always use the overlay st_dev for directories, so 'find | |
132 | * -xdev' will scan the entire overlay mount and won't cross the | |
133 | * overlay mount boundaries. | |
134 | * | |
135 | * If not all layers are on the same fs the pair {real st_ino; | |
136 | * overlay st_dev} is not unique, so use the non persistent | |
137 | * overlay st_ino for directories. | |
138 | */ | |
139 | stat->dev = dentry->d_sb->s_dev; | |
140 | stat->ino = dentry->d_inode->i_ino; | |
b7bf9908 | 141 | } else { |
da309e8c AG |
142 | /* |
143 | * For non-samefs setup, if we cannot map all layers st_ino | |
144 | * to a unified address space, we need to make sure that st_dev | |
b7bf9908 AG |
145 | * is unique per underlying fs, so we use the unique anonymous |
146 | * bdev assigned to the underlying fs. | |
da309e8c | 147 | */ |
07f1e596 | 148 | stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev; |
da309e8c | 149 | } |
da309e8c AG |
150 | } |
151 | ||
549c7297 CB |
152 | int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path, |
153 | struct kstat *stat, u32 request_mask, unsigned int flags) | |
e9be9d5e | 154 | { |
a528d35e | 155 | struct dentry *dentry = path->dentry; |
72b608f0 | 156 | enum ovl_path_type type; |
e9be9d5e | 157 | struct path realpath; |
1175b6b8 | 158 | const struct cred *old_cred; |
096a218a AG |
159 | struct inode *inode = d_inode(dentry); |
160 | bool is_dir = S_ISDIR(inode->i_mode); | |
07f1e596 | 161 | int fsid = 0; |
1175b6b8 | 162 | int err; |
67d756c2 VG |
163 | bool metacopy_blocks = false; |
164 | ||
165 | metacopy_blocks = ovl_is_metacopy_dentry(dentry); | |
e9be9d5e | 166 | |
72b608f0 | 167 | type = ovl_path_real(dentry, &realpath); |
1175b6b8 | 168 | old_cred = ovl_override_creds(dentry->d_sb); |
a528d35e | 169 | err = vfs_getattr(&realpath, stat, request_mask, flags); |
72b608f0 AG |
170 | if (err) |
171 | goto out; | |
172 | ||
096a218a AG |
173 | /* Report the effective immutable/append-only STATX flags */ |
174 | generic_fill_statx_attr(inode, stat); | |
175 | ||
72b608f0 | 176 | /* |
da309e8c AG |
177 | * For non-dir or same fs, we use st_ino of the copy up origin. |
178 | * This guaranties constant st_dev/st_ino across copy up. | |
e487d889 AG |
179 | * With xino feature and non-samefs, we use st_ino of the copy up |
180 | * origin masked with high bits that represent the layer id. | |
72b608f0 | 181 | * |
da309e8c | 182 | * If lower filesystem supports NFS file handles, this also guaranties |
72b608f0 AG |
183 | * persistent st_ino across mount cycle. |
184 | */ | |
0f831ec8 | 185 | if (!is_dir || ovl_same_dev(dentry->d_sb)) { |
da309e8c | 186 | if (!OVL_TYPE_UPPER(type)) { |
07f1e596 | 187 | fsid = ovl_layer_lower(dentry)->fsid; |
da309e8c | 188 | } else if (OVL_TYPE_ORIGIN(type)) { |
72b608f0 | 189 | struct kstat lowerstat; |
67d756c2 VG |
190 | u32 lowermask = STATX_INO | STATX_BLOCKS | |
191 | (!is_dir ? STATX_NLINK : 0); | |
72b608f0 AG |
192 | |
193 | ovl_path_lower(dentry, &realpath); | |
194 | err = vfs_getattr(&realpath, &lowerstat, | |
5b712091 | 195 | lowermask, flags); |
72b608f0 AG |
196 | if (err) |
197 | goto out; | |
198 | ||
72b608f0 | 199 | /* |
359f392c | 200 | * Lower hardlinks may be broken on copy up to different |
72b608f0 AG |
201 | * upper files, so we cannot use the lower origin st_ino |
202 | * for those different files, even for the same fs case. | |
86eaa130 AG |
203 | * |
204 | * Similarly, several redirected dirs can point to the | |
205 | * same dir on a lower layer. With the "verify_lower" | |
206 | * feature, we do not use the lower origin st_ino, if | |
207 | * we haven't verified that this redirect is unique. | |
208 | * | |
359f392c | 209 | * With inodes index enabled, it is safe to use st_ino |
86eaa130 AG |
210 | * of an indexed origin. The index validates that the |
211 | * upper hardlink is not broken and that a redirected | |
212 | * dir is the only redirect to that origin. | |
72b608f0 | 213 | */ |
86eaa130 AG |
214 | if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) || |
215 | (!ovl_verify_lower(dentry->d_sb) && | |
9f99e50d | 216 | (is_dir || lowerstat.nlink == 1))) { |
07f1e596 | 217 | fsid = ovl_layer_lower(dentry)->fsid; |
b7bf9908 | 218 | stat->ino = lowerstat.ino; |
9f99e50d | 219 | } |
67d756c2 VG |
220 | |
221 | /* | |
222 | * If we are querying a metacopy dentry and lower | |
223 | * dentry is data dentry, then use the blocks we | |
224 | * queried just now. We don't have to do additional | |
225 | * vfs_getattr(). If lower itself is metacopy, then | |
226 | * additional vfs_getattr() is unavoidable. | |
227 | */ | |
228 | if (metacopy_blocks && | |
229 | realpath.dentry == ovl_dentry_lowerdata(dentry)) { | |
230 | stat->blocks = lowerstat.blocks; | |
231 | metacopy_blocks = false; | |
232 | } | |
233 | } | |
234 | ||
235 | if (metacopy_blocks) { | |
236 | /* | |
237 | * If lower is not same as lowerdata or if there was | |
238 | * no origin on upper, we can end up here. | |
239 | */ | |
240 | struct kstat lowerdatastat; | |
241 | u32 lowermask = STATX_BLOCKS; | |
242 | ||
243 | ovl_path_lowerdata(dentry, &realpath); | |
244 | err = vfs_getattr(&realpath, &lowerdatastat, | |
245 | lowermask, flags); | |
246 | if (err) | |
247 | goto out; | |
248 | stat->blocks = lowerdatastat.blocks; | |
72b608f0 | 249 | } |
72b608f0 | 250 | } |
5b712091 | 251 | |
c68e7ec5 | 252 | ovl_map_dev_ino(dentry, stat, fsid); |
da309e8c | 253 | |
5b712091 MS |
254 | /* |
255 | * It's probably not worth it to count subdirs to get the | |
256 | * correct link count. nlink=1 seems to pacify 'find' and | |
257 | * other utilities. | |
258 | */ | |
259 | if (is_dir && OVL_TYPE_MERGE(type)) | |
260 | stat->nlink = 1; | |
261 | ||
5f8415d6 AG |
262 | /* |
263 | * Return the overlay inode nlinks for indexed upper inodes. | |
264 | * Overlay inode nlink counts the union of the upper hardlinks | |
265 | * and non-covered lower hardlinks. It does not include the upper | |
266 | * index hardlink. | |
267 | */ | |
268 | if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
269 | stat->nlink = dentry->d_inode->i_nlink; | |
270 | ||
72b608f0 | 271 | out: |
1175b6b8 | 272 | revert_creds(old_cred); |
72b608f0 | 273 | |
1175b6b8 | 274 | return err; |
e9be9d5e MS |
275 | } |
276 | ||
549c7297 CB |
277 | int ovl_permission(struct user_namespace *mnt_userns, |
278 | struct inode *inode, int mask) | |
e9be9d5e | 279 | { |
09d8b586 MS |
280 | struct inode *upperinode = ovl_inode_upper(inode); |
281 | struct inode *realinode = upperinode ?: ovl_inode_lower(inode); | |
c0ca3d70 | 282 | const struct cred *old_cred; |
e9be9d5e MS |
283 | int err; |
284 | ||
e9be9d5e | 285 | /* Careful in RCU walk mode */ |
e9be9d5e MS |
286 | if (!realinode) { |
287 | WARN_ON(!(mask & MAY_NOT_BLOCK)); | |
a999d7e1 | 288 | return -ECHILD; |
e9be9d5e MS |
289 | } |
290 | ||
c0ca3d70 VG |
291 | /* |
292 | * Check overlay inode with the creds of task and underlying inode | |
293 | * with creds of mounter | |
294 | */ | |
47291baa | 295 | err = generic_permission(&init_user_ns, inode, mask); |
c0ca3d70 VG |
296 | if (err) |
297 | return err; | |
298 | ||
ec7ba118 MS |
299 | old_cred = ovl_override_creds(inode->i_sb); |
300 | if (!upperinode && | |
301 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 302 | mask &= ~(MAY_WRITE | MAY_APPEND); |
ec7ba118 | 303 | /* Make sure mounter can read file for copy up later */ |
500cac3c VG |
304 | mask |= MAY_READ; |
305 | } | |
47291baa | 306 | err = inode_permission(&init_user_ns, realinode, mask); |
c0ca3d70 VG |
307 | revert_creds(old_cred); |
308 | ||
309 | return err; | |
e9be9d5e MS |
310 | } |
311 | ||
6b255391 | 312 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
313 | struct inode *inode, |
314 | struct delayed_call *done) | |
e9be9d5e | 315 | { |
1175b6b8 VG |
316 | const struct cred *old_cred; |
317 | const char *p; | |
e9be9d5e | 318 | |
6b255391 AV |
319 | if (!dentry) |
320 | return ERR_PTR(-ECHILD); | |
321 | ||
1175b6b8 | 322 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 323 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
324 | revert_creds(old_cred); |
325 | return p; | |
e9be9d5e MS |
326 | } |
327 | ||
610afc0b | 328 | bool ovl_is_private_xattr(struct super_block *sb, const char *name) |
e9be9d5e | 329 | { |
2d2f2d73 MS |
330 | struct ovl_fs *ofs = sb->s_fs_info; |
331 | ||
332 | if (ofs->config.userxattr) | |
333 | return strncmp(name, OVL_XATTR_USER_PREFIX, | |
334 | sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0; | |
335 | else | |
336 | return strncmp(name, OVL_XATTR_TRUSTED_PREFIX, | |
337 | sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0; | |
e9be9d5e MS |
338 | } |
339 | ||
1d88f183 MS |
340 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
341 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
342 | { |
343 | int err; | |
1d88f183 MS |
344 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
345 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
1175b6b8 | 346 | const struct cred *old_cred; |
e9be9d5e MS |
347 | |
348 | err = ovl_want_write(dentry); | |
349 | if (err) | |
350 | goto out; | |
351 | ||
1d88f183 | 352 | if (!value && !upperdentry) { |
554677b9 | 353 | old_cred = ovl_override_creds(dentry->d_sb); |
c7c7a1a1 | 354 | err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0); |
554677b9 | 355 | revert_creds(old_cred); |
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) |
c7c7a1a1 TA |
370 | err = vfs_setxattr(&init_user_ns, realdentry, name, value, size, |
371 | flags); | |
0e585ccc AG |
372 | else { |
373 | WARN_ON(flags != XATTR_REPLACE); | |
c7c7a1a1 | 374 | err = vfs_removexattr(&init_user_ns, realdentry, name); |
0e585ccc | 375 | } |
1175b6b8 | 376 | revert_creds(old_cred); |
e9be9d5e | 377 | |
d9854c87 MS |
378 | /* copy c/mtime */ |
379 | ovl_copyattr(d_inode(realdentry), inode); | |
380 | ||
e9be9d5e MS |
381 | out_drop_write: |
382 | ovl_drop_write(dentry); | |
383 | out: | |
384 | return err; | |
385 | } | |
386 | ||
1d88f183 | 387 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 388 | void *value, size_t size) |
e9be9d5e | 389 | { |
1175b6b8 VG |
390 | ssize_t res; |
391 | const struct cred *old_cred; | |
1d88f183 MS |
392 | struct dentry *realdentry = |
393 | ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry); | |
52148463 | 394 | |
1175b6b8 | 395 | old_cred = ovl_override_creds(dentry->d_sb); |
c7c7a1a1 | 396 | res = vfs_getxattr(&init_user_ns, realdentry, name, value, size); |
1175b6b8 VG |
397 | revert_creds(old_cred); |
398 | return res; | |
e9be9d5e MS |
399 | } |
400 | ||
610afc0b | 401 | static bool ovl_can_list(struct super_block *sb, const char *s) |
a082c6f6 | 402 | { |
8f6ee74c MS |
403 | /* Never list private (.overlay) */ |
404 | if (ovl_is_private_xattr(sb, s)) | |
405 | return false; | |
406 | ||
f48bbfb2 | 407 | /* List all non-trusted xattrs */ |
a082c6f6 MS |
408 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) |
409 | return true; | |
410 | ||
8f6ee74c MS |
411 | /* list other trusted for superuser only */ |
412 | return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN); | |
a082c6f6 MS |
413 | } |
414 | ||
e9be9d5e MS |
415 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
416 | { | |
b581755b | 417 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 418 | ssize_t res; |
7cb35119 MS |
419 | size_t len; |
420 | char *s; | |
1175b6b8 | 421 | const struct cred *old_cred; |
e9be9d5e | 422 | |
1175b6b8 | 423 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 424 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 425 | revert_creds(old_cred); |
e9be9d5e MS |
426 | if (res <= 0 || size == 0) |
427 | return res; | |
428 | ||
e9be9d5e | 429 | /* filter out private xattrs */ |
7cb35119 MS |
430 | for (s = list, len = res; len;) { |
431 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 432 | |
7cb35119 MS |
433 | /* underlying fs providing us with an broken xattr list? */ |
434 | if (WARN_ON(slen > len)) | |
435 | return -EIO; | |
e9be9d5e | 436 | |
7cb35119 | 437 | len -= slen; |
610afc0b | 438 | if (!ovl_can_list(dentry->d_sb, s)) { |
e9be9d5e | 439 | res -= slen; |
7cb35119 | 440 | memmove(s, s + slen, len); |
e9be9d5e | 441 | } else { |
7cb35119 | 442 | s += slen; |
e9be9d5e MS |
443 | } |
444 | } | |
445 | ||
446 | return res; | |
447 | } | |
448 | ||
39a25b2b VG |
449 | struct posix_acl *ovl_get_acl(struct inode *inode, int type) |
450 | { | |
09d8b586 | 451 | struct inode *realinode = ovl_inode_real(inode); |
1175b6b8 VG |
452 | const struct cred *old_cred; |
453 | struct posix_acl *acl; | |
39a25b2b | 454 | |
5201dc44 | 455 | if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode)) |
39a25b2b VG |
456 | return NULL; |
457 | ||
1175b6b8 | 458 | old_cred = ovl_override_creds(inode->i_sb); |
5201dc44 | 459 | acl = get_acl(realinode, type); |
1175b6b8 VG |
460 | revert_creds(old_cred); |
461 | ||
462 | return acl; | |
39a25b2b VG |
463 | } |
464 | ||
95582b00 | 465 | int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags) |
d719e8f2 | 466 | { |
8f35cf51 MS |
467 | if (flags & S_ATIME) { |
468 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
469 | struct path upperpath = { | |
08f4c7c8 | 470 | .mnt = ovl_upper_mnt(ofs), |
8f35cf51 MS |
471 | .dentry = ovl_upperdentry_dereference(OVL_I(inode)), |
472 | }; | |
473 | ||
474 | if (upperpath.dentry) { | |
475 | touch_atime(&upperpath); | |
476 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
477 | } | |
d719e8f2 | 478 | } |
d719e8f2 MS |
479 | return 0; |
480 | } | |
481 | ||
9e142c41 MS |
482 | static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
483 | u64 start, u64 len) | |
484 | { | |
485 | int err; | |
c11faf32 | 486 | struct inode *realinode = ovl_inode_realdata(inode); |
9e142c41 MS |
487 | const struct cred *old_cred; |
488 | ||
489 | if (!realinode->i_op->fiemap) | |
490 | return -EOPNOTSUPP; | |
491 | ||
492 | old_cred = ovl_override_creds(inode->i_sb); | |
493 | err = realinode->i_op->fiemap(realinode, fieinfo, start, len); | |
494 | revert_creds(old_cred); | |
495 | ||
496 | return err; | |
497 | } | |
498 | ||
66dbfabf MS |
499 | /* |
500 | * Work around the fact that security_file_ioctl() takes a file argument. | |
501 | * Introducing security_inode_fileattr_get/set() hooks would solve this issue | |
502 | * properly. | |
503 | */ | |
72db8211 | 504 | static int ovl_security_fileattr(struct path *realpath, struct fileattr *fa, |
66dbfabf MS |
505 | bool set) |
506 | { | |
66dbfabf MS |
507 | struct file *file; |
508 | unsigned int cmd; | |
509 | int err; | |
510 | ||
72db8211 | 511 | file = dentry_open(realpath, O_RDONLY, current_cred()); |
66dbfabf MS |
512 | if (IS_ERR(file)) |
513 | return PTR_ERR(file); | |
514 | ||
515 | if (set) | |
516 | cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS; | |
517 | else | |
518 | cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS; | |
519 | ||
520 | err = security_file_ioctl(file, cmd, 0); | |
521 | fput(file); | |
522 | ||
523 | return err; | |
524 | } | |
525 | ||
72db8211 AG |
526 | int ovl_real_fileattr_set(struct path *realpath, struct fileattr *fa) |
527 | { | |
528 | int err; | |
529 | ||
530 | err = ovl_security_fileattr(realpath, fa, true); | |
531 | if (err) | |
532 | return err; | |
533 | ||
534 | return vfs_fileattr_set(&init_user_ns, realpath->dentry, fa); | |
535 | } | |
536 | ||
66dbfabf MS |
537 | int ovl_fileattr_set(struct user_namespace *mnt_userns, |
538 | struct dentry *dentry, struct fileattr *fa) | |
539 | { | |
540 | struct inode *inode = d_inode(dentry); | |
72db8211 | 541 | struct path upperpath; |
66dbfabf | 542 | const struct cred *old_cred; |
096a218a | 543 | unsigned int flags; |
66dbfabf MS |
544 | int err; |
545 | ||
546 | err = ovl_want_write(dentry); | |
547 | if (err) | |
548 | goto out; | |
549 | ||
550 | err = ovl_copy_up(dentry); | |
551 | if (!err) { | |
72db8211 | 552 | ovl_path_real(dentry, &upperpath); |
66dbfabf MS |
553 | |
554 | old_cred = ovl_override_creds(inode->i_sb); | |
096a218a AG |
555 | /* |
556 | * Store immutable/append-only flags in xattr and clear them | |
557 | * in upper fileattr (in case they were set by older kernel) | |
558 | * so children of "ovl-immutable" directories lower aliases of | |
559 | * "ovl-immutable" hardlinks could be copied up. | |
560 | * Clear xattr when flags are cleared. | |
561 | */ | |
562 | err = ovl_set_protattr(inode, upperpath.dentry, fa); | |
563 | if (!err) | |
564 | err = ovl_real_fileattr_set(&upperpath, fa); | |
66dbfabf | 565 | revert_creds(old_cred); |
096a218a AG |
566 | |
567 | /* | |
568 | * Merge real inode flags with inode flags read from | |
569 | * overlay.protattr xattr | |
570 | */ | |
571 | flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK; | |
572 | ||
573 | BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK); | |
574 | flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK; | |
575 | inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK); | |
d8991e86 CX |
576 | |
577 | /* Update ctime */ | |
578 | ovl_copyattr(ovl_inode_real(inode), inode); | |
66dbfabf MS |
579 | } |
580 | ovl_drop_write(dentry); | |
581 | out: | |
582 | return err; | |
583 | } | |
584 | ||
096a218a AG |
585 | /* Convert inode protection flags to fileattr flags */ |
586 | static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa) | |
587 | { | |
588 | BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL); | |
589 | BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON); | |
590 | ||
591 | if (inode->i_flags & S_APPEND) { | |
592 | fa->flags |= FS_APPEND_FL; | |
593 | fa->fsx_xflags |= FS_XFLAG_APPEND; | |
594 | } | |
595 | if (inode->i_flags & S_IMMUTABLE) { | |
596 | fa->flags |= FS_IMMUTABLE_FL; | |
597 | fa->fsx_xflags |= FS_XFLAG_IMMUTABLE; | |
598 | } | |
599 | } | |
600 | ||
72db8211 AG |
601 | int ovl_real_fileattr_get(struct path *realpath, struct fileattr *fa) |
602 | { | |
603 | int err; | |
604 | ||
605 | err = ovl_security_fileattr(realpath, fa, false); | |
606 | if (err) | |
607 | return err; | |
608 | ||
609 | return vfs_fileattr_get(realpath->dentry, fa); | |
610 | } | |
611 | ||
66dbfabf MS |
612 | int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
613 | { | |
614 | struct inode *inode = d_inode(dentry); | |
72db8211 | 615 | struct path realpath; |
66dbfabf MS |
616 | const struct cred *old_cred; |
617 | int err; | |
618 | ||
72db8211 AG |
619 | ovl_path_real(dentry, &realpath); |
620 | ||
66dbfabf | 621 | old_cred = ovl_override_creds(inode->i_sb); |
72db8211 | 622 | err = ovl_real_fileattr_get(&realpath, fa); |
096a218a | 623 | ovl_fileattr_prot_flags(inode, fa); |
66dbfabf MS |
624 | revert_creds(old_cred); |
625 | ||
626 | return err; | |
627 | } | |
628 | ||
e9be9d5e MS |
629 | static const struct inode_operations ovl_file_inode_operations = { |
630 | .setattr = ovl_setattr, | |
631 | .permission = ovl_permission, | |
632 | .getattr = ovl_getattr, | |
e9be9d5e | 633 | .listxattr = ovl_listxattr, |
39a25b2b | 634 | .get_acl = ovl_get_acl, |
d719e8f2 | 635 | .update_time = ovl_update_time, |
9e142c41 | 636 | .fiemap = ovl_fiemap, |
66dbfabf MS |
637 | .fileattr_get = ovl_fileattr_get, |
638 | .fileattr_set = ovl_fileattr_set, | |
e9be9d5e MS |
639 | }; |
640 | ||
641 | static const struct inode_operations ovl_symlink_inode_operations = { | |
642 | .setattr = ovl_setattr, | |
6b255391 | 643 | .get_link = ovl_get_link, |
e9be9d5e | 644 | .getattr = ovl_getattr, |
e9be9d5e | 645 | .listxattr = ovl_listxattr, |
d719e8f2 | 646 | .update_time = ovl_update_time, |
e9be9d5e MS |
647 | }; |
648 | ||
9e142c41 MS |
649 | static const struct inode_operations ovl_special_inode_operations = { |
650 | .setattr = ovl_setattr, | |
651 | .permission = ovl_permission, | |
652 | .getattr = ovl_getattr, | |
653 | .listxattr = ovl_listxattr, | |
654 | .get_acl = ovl_get_acl, | |
655 | .update_time = ovl_update_time, | |
656 | }; | |
657 | ||
69383c59 | 658 | static const struct address_space_operations ovl_aops = { |
5b910bd6 AG |
659 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
660 | .direct_IO = noop_direct_IO, | |
661 | }; | |
662 | ||
b1eaa950 AG |
663 | /* |
664 | * It is possible to stack overlayfs instance on top of another | |
a5a84682 | 665 | * overlayfs instance as lower layer. We need to annotate the |
b1eaa950 AG |
666 | * stackable i_mutex locks according to stack level of the super |
667 | * block instance. An overlayfs instance can never be in stack | |
668 | * depth 0 (there is always a real fs below it). An overlayfs | |
f48bbfb2 | 669 | * inode lock will use the lockdep annotation ovl_i_mutex_key[depth]. |
b1eaa950 AG |
670 | * |
671 | * For example, here is a snip from /proc/lockdep_chains after | |
672 | * dir_iterate of nested overlayfs: | |
673 | * | |
674 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
675 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
676 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
b1f9d385 AG |
677 | * |
678 | * Locking order w.r.t ovl_want_write() is important for nested overlayfs. | |
679 | * | |
680 | * This chain is valid: | |
681 | * - inode->i_rwsem (inode_lock[2]) | |
682 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
683 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
684 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
685 | * | |
686 | * And this chain is valid: | |
687 | * - inode->i_rwsem (inode_lock[2]) | |
688 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
689 | * - lowerinode->i_rwsem (inode_lock[1]) | |
690 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
691 | * | |
692 | * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is | |
693 | * held, because it is in reverse order of the non-nested case using the same | |
694 | * upper fs: | |
695 | * - inode->i_rwsem (inode_lock[1]) | |
696 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
697 | * - OVL_I(inode)->lock (ovl_inode_lock[1]) | |
b1eaa950 AG |
698 | */ |
699 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
700 | ||
701 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
702 | { | |
703 | #ifdef CONFIG_LOCKDEP | |
704 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
705 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 706 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
707 | |
708 | int depth = inode->i_sb->s_stack_depth - 1; | |
709 | ||
710 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
711 | depth = 0; | |
712 | ||
713 | if (S_ISDIR(inode->i_mode)) | |
714 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
715 | else | |
716 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
717 | |
718 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
719 | #endif |
720 | } | |
721 | ||
4d314f78 AG |
722 | static void ovl_next_ino(struct inode *inode) |
723 | { | |
724 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
725 | ||
726 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
727 | if (unlikely(!inode->i_ino)) | |
728 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
729 | } | |
730 | ||
62c832ed | 731 | static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid) |
e9be9d5e | 732 | { |
12574a9f | 733 | int xinobits = ovl_xino_bits(inode->i_sb); |
dfe51d47 | 734 | unsigned int xinoshift = 64 - xinobits; |
12574a9f | 735 | |
695b46e7 | 736 | /* |
6dde1e42 AG |
737 | * When d_ino is consistent with st_ino (samefs or i_ino has enough |
738 | * bits to encode layer), set the same value used for st_ino to i_ino, | |
739 | * so inode number exposed via /proc/locks and a like will be | |
740 | * consistent with d_ino and st_ino values. An i_ino value inconsistent | |
735c907d | 741 | * with d_ino also causes nfsd readdirplus to fail. |
695b46e7 | 742 | */ |
4d314f78 | 743 | inode->i_ino = ino; |
dfe51d47 AG |
744 | if (ovl_same_fs(inode->i_sb)) { |
745 | return; | |
746 | } else if (xinobits && likely(!(ino >> xinoshift))) { | |
747 | inode->i_ino |= (unsigned long)fsid << (xinoshift + 1); | |
748 | return; | |
749 | } | |
750 | ||
751 | /* | |
752 | * For directory inodes on non-samefs with xino disabled or xino | |
753 | * overflow, we allocate a non-persistent inode number, to be used for | |
754 | * resolving st_ino collisions in ovl_map_dev_ino(). | |
755 | * | |
756 | * To avoid ino collision with legitimate xino values from upper | |
757 | * layer (fsid 0), use the lowest xinobit to map the non | |
758 | * persistent inode numbers to the unified st_ino address space. | |
759 | */ | |
760 | if (S_ISDIR(inode->i_mode)) { | |
4d314f78 | 761 | ovl_next_ino(inode); |
dfe51d47 AG |
762 | if (xinobits) { |
763 | inode->i_ino &= ~0UL >> xinobits; | |
764 | inode->i_ino |= 1UL << xinoshift; | |
765 | } | |
12574a9f | 766 | } |
62c832ed AG |
767 | } |
768 | ||
769 | void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip, | |
770 | unsigned long ino, int fsid) | |
771 | { | |
772 | struct inode *realinode; | |
773 | ||
774 | if (oip->upperdentry) | |
775 | OVL_I(inode)->__upperdentry = oip->upperdentry; | |
776 | if (oip->lowerpath && oip->lowerpath->dentry) | |
777 | OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry)); | |
778 | if (oip->lowerdata) | |
779 | OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata)); | |
780 | ||
781 | realinode = ovl_inode_real(inode); | |
782 | ovl_copyattr(realinode, inode); | |
783 | ovl_copyflags(realinode, inode); | |
784 | ovl_map_ino(inode, ino, fsid); | |
785 | } | |
786 | ||
787 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
788 | { | |
e9be9d5e | 789 | inode->i_mode = mode; |
d719e8f2 | 790 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
791 | #ifdef CONFIG_FS_POSIX_ACL |
792 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
793 | #endif | |
e9be9d5e | 794 | |
b1eaa950 AG |
795 | ovl_lockdep_annotate_inode_mutex_key(inode); |
796 | ||
ca4c8a3a MS |
797 | switch (mode & S_IFMT) { |
798 | case S_IFREG: | |
799 | inode->i_op = &ovl_file_inode_operations; | |
d1d04ef8 | 800 | inode->i_fop = &ovl_file_operations; |
5b910bd6 | 801 | inode->i_mapping->a_ops = &ovl_aops; |
ca4c8a3a MS |
802 | break; |
803 | ||
e9be9d5e | 804 | case S_IFDIR: |
e9be9d5e MS |
805 | inode->i_op = &ovl_dir_inode_operations; |
806 | inode->i_fop = &ovl_dir_operations; | |
807 | break; | |
808 | ||
809 | case S_IFLNK: | |
810 | inode->i_op = &ovl_symlink_inode_operations; | |
811 | break; | |
812 | ||
51f7e52d | 813 | default: |
9e142c41 | 814 | inode->i_op = &ovl_special_inode_operations; |
ca4c8a3a | 815 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 816 | break; |
51f7e52d MS |
817 | } |
818 | } | |
e9be9d5e | 819 | |
5f8415d6 AG |
820 | /* |
821 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
822 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
823 | * upper inode, the following nlink modifying operations can happen: | |
824 | * | |
825 | * 1. Lower hardlink copy up | |
826 | * 2. Upper hardlink created, unlinked or renamed over | |
827 | * 3. Lower hardlink whiteout or renamed over | |
828 | * | |
829 | * For the first, copy up case, the union nlink does not change, whether the | |
830 | * operation succeeds or fails, but the upper inode nlink may change. | |
831 | * Therefore, before copy up, we store the union nlink value relative to the | |
2d2f2d73 | 832 | * lower inode nlink in the index inode xattr .overlay.nlink. |
5f8415d6 AG |
833 | * |
834 | * For the second, upper hardlink case, the union nlink should be incremented | |
835 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
836 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
837 | * value relative to the upper inode nlink in the index inode. | |
838 | * | |
839 | * For the last, lower cover up case, we simplify things by preceding the | |
840 | * whiteout or cover up with copy up. This makes sure that there is an index | |
841 | * upper inode where the nlink xattr can be stored before the copied up upper | |
842 | * entry is unlink. | |
843 | */ | |
844 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
845 | ||
846 | /* | |
847 | * On-disk format for indexed nlink: | |
848 | * | |
849 | * nlink relative to the upper inode - "U[+-]NUM" | |
850 | * nlink relative to the lower inode - "L[+-]NUM" | |
851 | */ | |
852 | ||
853 | static int ovl_set_nlink_common(struct dentry *dentry, | |
854 | struct dentry *realdentry, const char *format) | |
855 | { | |
856 | struct inode *inode = d_inode(dentry); | |
857 | struct inode *realinode = d_inode(realdentry); | |
858 | char buf[13]; | |
859 | int len; | |
860 | ||
861 | len = snprintf(buf, sizeof(buf), format, | |
862 | (int) (inode->i_nlink - realinode->i_nlink)); | |
863 | ||
6787341a MS |
864 | if (WARN_ON(len >= sizeof(buf))) |
865 | return -EIO; | |
866 | ||
610afc0b | 867 | return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry), |
26150ab5 | 868 | OVL_XATTR_NLINK, buf, len); |
5f8415d6 AG |
869 | } |
870 | ||
871 | int ovl_set_nlink_upper(struct dentry *dentry) | |
872 | { | |
873 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
874 | } | |
875 | ||
876 | int ovl_set_nlink_lower(struct dentry *dentry) | |
877 | { | |
878 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
879 | } | |
880 | ||
610afc0b | 881 | unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry, |
caf70cb2 AG |
882 | struct dentry *upperdentry, |
883 | unsigned int fallback) | |
5f8415d6 AG |
884 | { |
885 | int nlink_diff; | |
886 | int nlink; | |
887 | char buf[13]; | |
888 | int err; | |
889 | ||
890 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
891 | return fallback; | |
892 | ||
610afc0b | 893 | err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK, |
d5dc7486 | 894 | &buf, sizeof(buf) - 1); |
5f8415d6 AG |
895 | if (err < 0) |
896 | goto fail; | |
897 | ||
898 | buf[err] = '\0'; | |
899 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
900 | (buf[1] != '+' && buf[1] != '-')) | |
901 | goto fail; | |
902 | ||
903 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
904 | if (err < 0) | |
905 | goto fail; | |
906 | ||
907 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
908 | nlink += nlink_diff; | |
909 | ||
910 | if (nlink <= 0) | |
911 | goto fail; | |
912 | ||
913 | return nlink; | |
914 | ||
915 | fail: | |
1bd0a3ae | 916 | pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n", |
5f8415d6 AG |
917 | upperdentry, err); |
918 | return fallback; | |
919 | } | |
920 | ||
ca4c8a3a | 921 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
922 | { |
923 | struct inode *inode; | |
924 | ||
925 | inode = new_inode(sb); | |
926 | if (inode) | |
62c832ed | 927 | ovl_fill_inode(inode, mode, rdev); |
51f7e52d MS |
928 | |
929 | return inode; | |
930 | } | |
931 | ||
932 | static int ovl_inode_test(struct inode *inode, void *data) | |
933 | { | |
25b7713a | 934 | return inode->i_private == data; |
51f7e52d MS |
935 | } |
936 | ||
937 | static int ovl_inode_set(struct inode *inode, void *data) | |
938 | { | |
25b7713a | 939 | inode->i_private = data; |
51f7e52d MS |
940 | return 0; |
941 | } | |
942 | ||
b9ac5c27 | 943 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
4b91c30a | 944 | struct dentry *upperdentry, bool strict) |
b9ac5c27 | 945 | { |
4b91c30a AG |
946 | /* |
947 | * For directories, @strict verify from lookup path performs consistency | |
948 | * checks, so NULL lower/upper in dentry must match NULL lower/upper in | |
949 | * inode. Non @strict verify from NFS handle decode path passes NULL for | |
950 | * 'unknown' lower/upper. | |
951 | */ | |
952 | if (S_ISDIR(inode->i_mode) && strict) { | |
31747eda AG |
953 | /* Real lower dir moved to upper layer under us? */ |
954 | if (!lowerdentry && ovl_inode_lower(inode)) | |
955 | return false; | |
956 | ||
957 | /* Lookup of an uncovered redirect origin? */ | |
958 | if (!upperdentry && ovl_inode_upper(inode)) | |
959 | return false; | |
960 | } | |
961 | ||
939ae4ef AG |
962 | /* |
963 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
964 | * This happens when finding a copied up overlay inode for a renamed | |
965 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
966 | * by origin because lower fs does not support file handles. | |
967 | */ | |
968 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
969 | return false; |
970 | ||
971 | /* | |
972 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
973 | * This happens when finding a lower alias for a copied up hard link. | |
974 | */ | |
975 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
976 | return false; | |
977 | ||
978 | return true; | |
979 | } | |
980 | ||
4b91c30a AG |
981 | struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real, |
982 | bool is_upper) | |
9436a1a3 | 983 | { |
4b91c30a | 984 | struct inode *inode, *key = d_inode(real); |
9436a1a3 AG |
985 | |
986 | inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
987 | if (!inode) | |
988 | return NULL; | |
989 | ||
4b91c30a AG |
990 | if (!ovl_verify_inode(inode, is_upper ? NULL : real, |
991 | is_upper ? real : NULL, false)) { | |
9436a1a3 AG |
992 | iput(inode); |
993 | return ERR_PTR(-ESTALE); | |
994 | } | |
995 | ||
996 | return inode; | |
997 | } | |
998 | ||
146d62e5 AG |
999 | bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir) |
1000 | { | |
1001 | struct inode *key = d_inode(dir); | |
1002 | struct inode *trap; | |
1003 | bool res; | |
1004 | ||
1005 | trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
1006 | if (!trap) | |
1007 | return false; | |
1008 | ||
1009 | res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) && | |
1010 | !ovl_inode_lower(trap); | |
1011 | ||
1012 | iput(trap); | |
1013 | return res; | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * Create an inode cache entry for layer root dir, that will intentionally | |
1018 | * fail ovl_verify_inode(), so any lookup that will find some layer root | |
1019 | * will fail. | |
1020 | */ | |
1021 | struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir) | |
1022 | { | |
1023 | struct inode *key = d_inode(dir); | |
1024 | struct inode *trap; | |
1025 | ||
1026 | if (!d_is_dir(dir)) | |
1027 | return ERR_PTR(-ENOTDIR); | |
1028 | ||
1029 | trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test, | |
1030 | ovl_inode_set, key); | |
1031 | if (!trap) | |
1032 | return ERR_PTR(-ENOMEM); | |
1033 | ||
1034 | if (!(trap->i_state & I_NEW)) { | |
1035 | /* Conflicting layer roots? */ | |
1036 | iput(trap); | |
1037 | return ERR_PTR(-ELOOP); | |
1038 | } | |
1039 | ||
1040 | trap->i_mode = S_IFDIR; | |
1041 | trap->i_flags = S_DEAD; | |
1042 | unlock_new_inode(trap); | |
1043 | ||
1044 | return trap; | |
1045 | } | |
1046 | ||
764baba8 AG |
1047 | /* |
1048 | * Does overlay inode need to be hashed by lower inode? | |
1049 | */ | |
1050 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
74c6e384 | 1051 | struct dentry *lower, bool index) |
764baba8 AG |
1052 | { |
1053 | struct ovl_fs *ofs = sb->s_fs_info; | |
1054 | ||
1055 | /* No, if pure upper */ | |
1056 | if (!lower) | |
1057 | return false; | |
1058 | ||
1059 | /* Yes, if already indexed */ | |
1060 | if (index) | |
1061 | return true; | |
1062 | ||
1063 | /* Yes, if won't be copied up */ | |
08f4c7c8 | 1064 | if (!ovl_upper_mnt(ofs)) |
764baba8 AG |
1065 | return true; |
1066 | ||
1067 | /* No, if lower hardlink is or will be broken on copy up */ | |
1068 | if ((upper || !ovl_indexdir(sb)) && | |
1069 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
1070 | return false; | |
1071 | ||
1072 | /* No, if non-indexed upper with NFS export */ | |
1073 | if (sb->s_export_op && upper) | |
1074 | return false; | |
1075 | ||
1076 | /* Otherwise, hash by lower inode for fsnotify */ | |
1077 | return true; | |
1078 | } | |
1079 | ||
01b39dcc AG |
1080 | static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode, |
1081 | struct inode *key) | |
1082 | { | |
1083 | return newinode ? inode_insert5(newinode, (unsigned long) key, | |
1084 | ovl_inode_test, ovl_inode_set, key) : | |
1085 | iget5_locked(sb, (unsigned long) key, | |
1086 | ovl_inode_test, ovl_inode_set, key); | |
1087 | } | |
1088 | ||
ac6a52eb VG |
1089 | struct inode *ovl_get_inode(struct super_block *sb, |
1090 | struct ovl_inode_params *oip) | |
51f7e52d | 1091 | { |
610afc0b | 1092 | struct ovl_fs *ofs = OVL_FS(sb); |
ac6a52eb VG |
1093 | struct dentry *upperdentry = oip->upperdentry; |
1094 | struct ovl_path *lowerpath = oip->lowerpath; | |
09d8b586 | 1095 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; |
51f7e52d | 1096 | struct inode *inode; |
12574a9f | 1097 | struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL; |
ac6a52eb VG |
1098 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, |
1099 | oip->index); | |
300b124f | 1100 | int fsid = bylower ? lowerpath->layer->fsid : 0; |
28166ab3 | 1101 | bool is_dir; |
695b46e7 | 1102 | unsigned long ino = 0; |
acf3062a | 1103 | int err = oip->newinode ? -EEXIST : -ENOMEM; |
6eaf0111 | 1104 | |
09d8b586 MS |
1105 | if (!realinode) |
1106 | realinode = d_inode(lowerdentry); | |
1107 | ||
6eaf0111 | 1108 | /* |
764baba8 AG |
1109 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
1110 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 1111 | */ |
31747eda | 1112 | is_dir = S_ISDIR(realinode->i_mode); |
764baba8 AG |
1113 | if (upperdentry || bylower) { |
1114 | struct inode *key = d_inode(bylower ? lowerdentry : | |
1115 | upperdentry); | |
31747eda | 1116 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 1117 | |
01b39dcc | 1118 | inode = ovl_iget5(sb, oip->newinode, key); |
09d8b586 | 1119 | if (!inode) |
027065b7 | 1120 | goto out_err; |
09d8b586 | 1121 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
1122 | /* |
1123 | * Verify that the underlying files stored in the inode | |
1124 | * match those in the dentry. | |
1125 | */ | |
4b91c30a AG |
1126 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry, |
1127 | true)) { | |
b9ac5c27 | 1128 | iput(inode); |
027065b7 VG |
1129 | err = -ESTALE; |
1130 | goto out_err; | |
b9ac5c27 MS |
1131 | } |
1132 | ||
09d8b586 | 1133 | dput(upperdentry); |
9cec54c8 | 1134 | kfree(oip->redirect); |
09d8b586 MS |
1135 | goto out; |
1136 | } | |
e6d2ebdd | 1137 | |
31747eda AG |
1138 | /* Recalculate nlink for non-dir due to indexing */ |
1139 | if (!is_dir) | |
610afc0b MS |
1140 | nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry, |
1141 | nlink); | |
5f8415d6 | 1142 | set_nlink(inode, nlink); |
695b46e7 | 1143 | ino = key->i_ino; |
e6d2ebdd | 1144 | } else { |
764baba8 | 1145 | /* Lower hardlink that will be broken on copy up */ |
0aceb53e | 1146 | inode = new_inode(sb); |
027065b7 VG |
1147 | if (!inode) { |
1148 | err = -ENOMEM; | |
1149 | goto out_err; | |
1150 | } | |
300b124f AG |
1151 | ino = realinode->i_ino; |
1152 | fsid = lowerpath->layer->fsid; | |
e9be9d5e | 1153 | } |
62c832ed AG |
1154 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev); |
1155 | ovl_inode_init(inode, oip, ino, fsid); | |
13c72075 | 1156 | |
610afc0b | 1157 | if (upperdentry && ovl_is_impuredir(sb, upperdentry)) |
13c72075 MS |
1158 | ovl_set_flag(OVL_IMPURE, inode); |
1159 | ||
ac6a52eb | 1160 | if (oip->index) |
0471a9cd VG |
1161 | ovl_set_flag(OVL_INDEX, inode); |
1162 | ||
9cec54c8 VG |
1163 | OVL_I(inode)->redirect = oip->redirect; |
1164 | ||
a00c2d59 VG |
1165 | if (bylower) |
1166 | ovl_set_flag(OVL_CONST_INO, inode); | |
1167 | ||
b79e05aa | 1168 | /* Check for non-merge dir that may have whiteouts */ |
31747eda | 1169 | if (is_dir) { |
ac6a52eb | 1170 | if (((upperdentry && lowerdentry) || oip->numlower > 1) || |
610afc0b | 1171 | ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) { |
b79e05aa AG |
1172 | ovl_set_flag(OVL_WHITEOUTS, inode); |
1173 | } | |
1174 | } | |
1175 | ||
096a218a AG |
1176 | /* Check for immutable/append-only inode flags in xattr */ |
1177 | if (upperdentry) | |
1178 | ovl_check_protattr(inode, upperdentry); | |
1179 | ||
e6d2ebdd MS |
1180 | if (inode->i_state & I_NEW) |
1181 | unlock_new_inode(inode); | |
1182 | out: | |
e9be9d5e | 1183 | return inode; |
b9ac5c27 | 1184 | |
027065b7 | 1185 | out_err: |
1bd0a3ae | 1186 | pr_warn_ratelimited("failed to get inode (%i)\n", err); |
027065b7 | 1187 | inode = ERR_PTR(err); |
b9ac5c27 | 1188 | goto out; |
e9be9d5e | 1189 | } |