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> | |
332f606b | 16 | #include <linux/namei.h> |
6c0a8bfb CB |
17 | #include <linux/posix_acl.h> |
18 | #include <linux/posix_acl_xattr.h> | |
e9be9d5e MS |
19 | #include "overlayfs.h" |
20 | ||
ba1e563c | 21 | |
c1632a0f | 22 | int ovl_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 23 | struct iattr *attr) |
e9be9d5e MS |
24 | { |
25 | int err; | |
a15506ea | 26 | struct ovl_fs *ofs = OVL_FS(dentry->d_sb); |
997336f2 | 27 | bool full_copy_up = false; |
e9be9d5e | 28 | struct dentry *upperdentry; |
1175b6b8 | 29 | const struct cred *old_cred; |
e9be9d5e | 30 | |
c1632a0f | 31 | err = setattr_prepare(&nop_mnt_idmap, dentry, attr); |
cf9a6784 MS |
32 | if (err) |
33 | return err; | |
34 | ||
e9be9d5e MS |
35 | err = ovl_want_write(dentry); |
36 | if (err) | |
37 | goto out; | |
38 | ||
5812160e | 39 | if (attr->ia_valid & ATTR_SIZE) { |
997336f2 VG |
40 | /* Truncate should trigger data copy up as well */ |
41 | full_copy_up = true; | |
5812160e MS |
42 | } |
43 | ||
997336f2 VG |
44 | if (!full_copy_up) |
45 | err = ovl_copy_up(dentry); | |
46 | else | |
47 | err = ovl_copy_up_with_data(dentry); | |
acff81ec | 48 | if (!err) { |
5812160e MS |
49 | struct inode *winode = NULL; |
50 | ||
acff81ec MS |
51 | upperdentry = ovl_dentry_upper(dentry); |
52 | ||
5812160e MS |
53 | if (attr->ia_valid & ATTR_SIZE) { |
54 | winode = d_inode(upperdentry); | |
55 | err = get_write_access(winode); | |
56 | if (err) | |
57 | goto out_drop_write; | |
58 | } | |
59 | ||
b99c2d91 MS |
60 | if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) |
61 | attr->ia_valid &= ~ATTR_MODE; | |
62 | ||
e67f0216 | 63 | /* |
15fd2ea9 VG |
64 | * We might have to translate ovl file into real file object |
65 | * once use cases emerge. For now, simply don't let underlying | |
66 | * filesystem rely on attr->ia_file | |
e67f0216 VG |
67 | */ |
68 | attr->ia_valid &= ~ATTR_FILE; | |
69 | ||
15fd2ea9 VG |
70 | /* |
71 | * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN | |
72 | * set. Overlayfs does not pass O_TRUNC flag to underlying | |
73 | * filesystem during open -> do not pass ATTR_OPEN. This | |
74 | * disables optimization in fuse which assumes open(O_TRUNC) | |
75 | * already set file size to 0. But we never passed O_TRUNC to | |
76 | * fuse. So by clearing ATTR_OPEN, fuse will be forced to send | |
77 | * setattr request to server. | |
78 | */ | |
79 | attr->ia_valid &= ~ATTR_OPEN; | |
80 | ||
5955102c | 81 | inode_lock(upperdentry->d_inode); |
1175b6b8 | 82 | old_cred = ovl_override_creds(dentry->d_sb); |
a15506ea | 83 | err = ovl_do_notify_change(ofs, upperdentry, attr); |
1175b6b8 | 84 | revert_creds(old_cred); |
b81de061 | 85 | if (!err) |
2878dffc | 86 | ovl_copyattr(dentry->d_inode); |
5955102c | 87 | inode_unlock(upperdentry->d_inode); |
5812160e MS |
88 | |
89 | if (winode) | |
90 | put_write_access(winode); | |
e9be9d5e | 91 | } |
5812160e | 92 | out_drop_write: |
e9be9d5e MS |
93 | ovl_drop_write(dentry); |
94 | out: | |
95 | return err; | |
96 | } | |
97 | ||
c68e7ec5 | 98 | static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid) |
da309e8c | 99 | { |
0f831ec8 | 100 | bool samefs = ovl_same_fs(dentry->d_sb); |
e487d889 | 101 | unsigned int xinobits = ovl_xino_bits(dentry->d_sb); |
dfe51d47 | 102 | unsigned int xinoshift = 64 - xinobits; |
da309e8c AG |
103 | |
104 | if (samefs) { | |
105 | /* | |
106 | * When all layers are on the same fs, all real inode | |
107 | * number are unique, so we use the overlay st_dev, | |
108 | * which is friendly to du -x. | |
109 | */ | |
110 | stat->dev = dentry->d_sb->s_dev; | |
c68e7ec5 | 111 | return; |
e487d889 | 112 | } else if (xinobits) { |
e487d889 AG |
113 | /* |
114 | * All inode numbers of underlying fs should not be using the | |
115 | * high xinobits, so we use high xinobits to partition the | |
116 | * overlay st_ino address space. The high bits holds the fsid | |
dfe51d47 | 117 | * (upper fsid is 0). The lowest xinobit is reserved for mapping |
f48bbfb2 | 118 | * the non-persistent inode numbers range in case of overflow. |
dfe51d47 AG |
119 | * This way all overlay inode numbers are unique and use the |
120 | * overlay st_dev. | |
e487d889 | 121 | */ |
926e94d7 | 122 | if (likely(!(stat->ino >> xinoshift))) { |
dfe51d47 | 123 | stat->ino |= ((u64)fsid) << (xinoshift + 1); |
e487d889 | 124 | stat->dev = dentry->d_sb->s_dev; |
c68e7ec5 | 125 | return; |
926e94d7 AG |
126 | } else if (ovl_xino_warn(dentry->d_sb)) { |
127 | pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n", | |
128 | dentry, stat->ino, xinobits); | |
e487d889 AG |
129 | } |
130 | } | |
131 | ||
132 | /* The inode could not be mapped to a unified st_ino address space */ | |
133 | if (S_ISDIR(dentry->d_inode->i_mode)) { | |
da309e8c AG |
134 | /* |
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. | |
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. | |
142 | */ | |
143 | stat->dev = dentry->d_sb->s_dev; | |
144 | stat->ino = dentry->d_inode->i_ino; | |
b7bf9908 | 145 | } else { |
da309e8c AG |
146 | /* |
147 | * For non-samefs setup, if we cannot map all layers st_ino | |
148 | * to a unified address space, we need to make sure that st_dev | |
b7bf9908 AG |
149 | * is unique per underlying fs, so we use the unique anonymous |
150 | * bdev assigned to the underlying fs. | |
da309e8c | 151 | */ |
07f1e596 | 152 | stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev; |
da309e8c | 153 | } |
da309e8c AG |
154 | } |
155 | ||
b74d24f7 | 156 | int ovl_getattr(struct mnt_idmap *idmap, const struct path *path, |
549c7297 | 157 | struct kstat *stat, u32 request_mask, unsigned int flags) |
e9be9d5e | 158 | { |
a528d35e | 159 | struct dentry *dentry = path->dentry; |
72b608f0 | 160 | enum ovl_path_type type; |
e9be9d5e | 161 | struct path realpath; |
1175b6b8 | 162 | const struct cred *old_cred; |
096a218a AG |
163 | struct inode *inode = d_inode(dentry); |
164 | bool is_dir = S_ISDIR(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 | ||
096a218a AG |
177 | /* Report the effective immutable/append-only STATX flags */ |
178 | generic_fill_statx_attr(inode, stat); | |
179 | ||
72b608f0 | 180 | /* |
da309e8c AG |
181 | * For non-dir or same fs, we use st_ino of the copy up origin. |
182 | * This guaranties constant st_dev/st_ino across copy up. | |
e487d889 AG |
183 | * With xino feature and non-samefs, we use st_ino of the copy up |
184 | * origin masked with high bits that represent the layer id. | |
72b608f0 | 185 | * |
da309e8c | 186 | * If lower filesystem supports NFS file handles, this also guaranties |
72b608f0 AG |
187 | * persistent st_ino across mount cycle. |
188 | */ | |
0f831ec8 | 189 | if (!is_dir || ovl_same_dev(dentry->d_sb)) { |
da309e8c | 190 | if (!OVL_TYPE_UPPER(type)) { |
07f1e596 | 191 | fsid = ovl_layer_lower(dentry)->fsid; |
da309e8c | 192 | } else if (OVL_TYPE_ORIGIN(type)) { |
72b608f0 | 193 | struct kstat lowerstat; |
67d756c2 VG |
194 | u32 lowermask = STATX_INO | STATX_BLOCKS | |
195 | (!is_dir ? STATX_NLINK : 0); | |
72b608f0 AG |
196 | |
197 | ovl_path_lower(dentry, &realpath); | |
198 | err = vfs_getattr(&realpath, &lowerstat, | |
5b712091 | 199 | lowermask, flags); |
72b608f0 AG |
200 | if (err) |
201 | goto out; | |
202 | ||
72b608f0 | 203 | /* |
359f392c | 204 | * Lower hardlinks may be broken on copy up to different |
72b608f0 AG |
205 | * upper files, so we cannot use the lower origin st_ino |
206 | * for those different files, even for the same fs case. | |
86eaa130 AG |
207 | * |
208 | * Similarly, several redirected dirs can point to the | |
209 | * same dir on a lower layer. With the "verify_lower" | |
210 | * feature, we do not use the lower origin st_ino, if | |
211 | * we haven't verified that this redirect is unique. | |
212 | * | |
359f392c | 213 | * With inodes index enabled, it is safe to use st_ino |
86eaa130 AG |
214 | * of an indexed origin. The index validates that the |
215 | * upper hardlink is not broken and that a redirected | |
216 | * dir is the only redirect to that origin. | |
72b608f0 | 217 | */ |
86eaa130 AG |
218 | if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) || |
219 | (!ovl_verify_lower(dentry->d_sb) && | |
9f99e50d | 220 | (is_dir || lowerstat.nlink == 1))) { |
07f1e596 | 221 | fsid = ovl_layer_lower(dentry)->fsid; |
b7bf9908 | 222 | stat->ino = lowerstat.ino; |
9f99e50d | 223 | } |
67d756c2 VG |
224 | |
225 | /* | |
226 | * If we are querying a metacopy dentry and lower | |
227 | * dentry is data dentry, then use the blocks we | |
228 | * queried just now. We don't have to do additional | |
229 | * vfs_getattr(). If lower itself is metacopy, then | |
230 | * additional vfs_getattr() is unavoidable. | |
231 | */ | |
232 | if (metacopy_blocks && | |
233 | realpath.dentry == ovl_dentry_lowerdata(dentry)) { | |
234 | stat->blocks = lowerstat.blocks; | |
235 | metacopy_blocks = false; | |
236 | } | |
237 | } | |
238 | ||
239 | if (metacopy_blocks) { | |
240 | /* | |
241 | * If lower is not same as lowerdata or if there was | |
242 | * no origin on upper, we can end up here. | |
243 | */ | |
244 | struct kstat lowerdatastat; | |
245 | u32 lowermask = STATX_BLOCKS; | |
246 | ||
247 | ovl_path_lowerdata(dentry, &realpath); | |
248 | err = vfs_getattr(&realpath, &lowerdatastat, | |
249 | lowermask, flags); | |
250 | if (err) | |
251 | goto out; | |
252 | stat->blocks = lowerdatastat.blocks; | |
72b608f0 | 253 | } |
72b608f0 | 254 | } |
5b712091 | 255 | |
c68e7ec5 | 256 | ovl_map_dev_ino(dentry, stat, fsid); |
da309e8c | 257 | |
5b712091 MS |
258 | /* |
259 | * It's probably not worth it to count subdirs to get the | |
260 | * correct link count. nlink=1 seems to pacify 'find' and | |
261 | * other utilities. | |
262 | */ | |
263 | if (is_dir && OVL_TYPE_MERGE(type)) | |
264 | stat->nlink = 1; | |
265 | ||
5f8415d6 AG |
266 | /* |
267 | * Return the overlay inode nlinks for indexed upper inodes. | |
268 | * Overlay inode nlink counts the union of the upper hardlinks | |
269 | * and non-covered lower hardlinks. It does not include the upper | |
270 | * index hardlink. | |
271 | */ | |
272 | if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
273 | stat->nlink = dentry->d_inode->i_nlink; | |
274 | ||
72b608f0 | 275 | out: |
1175b6b8 | 276 | revert_creds(old_cred); |
72b608f0 | 277 | |
1175b6b8 | 278 | return err; |
e9be9d5e MS |
279 | } |
280 | ||
4609e1f1 | 281 | int ovl_permission(struct mnt_idmap *idmap, |
549c7297 | 282 | struct inode *inode, int mask) |
e9be9d5e | 283 | { |
09d8b586 | 284 | struct inode *upperinode = ovl_inode_upper(inode); |
4b7791b2 CB |
285 | struct inode *realinode; |
286 | struct path realpath; | |
c0ca3d70 | 287 | const struct cred *old_cred; |
e9be9d5e MS |
288 | int err; |
289 | ||
e9be9d5e | 290 | /* Careful in RCU walk mode */ |
1a73f5b8 ZC |
291 | realinode = ovl_i_path_real(inode, &realpath); |
292 | if (!realinode) { | |
e9be9d5e | 293 | WARN_ON(!(mask & MAY_NOT_BLOCK)); |
a999d7e1 | 294 | return -ECHILD; |
e9be9d5e MS |
295 | } |
296 | ||
c0ca3d70 VG |
297 | /* |
298 | * Check overlay inode with the creds of task and underlying inode | |
299 | * with creds of mounter | |
300 | */ | |
4609e1f1 | 301 | err = generic_permission(&nop_mnt_idmap, inode, mask); |
c0ca3d70 VG |
302 | if (err) |
303 | return err; | |
304 | ||
ec7ba118 MS |
305 | old_cred = ovl_override_creds(inode->i_sb); |
306 | if (!upperinode && | |
307 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 308 | mask &= ~(MAY_WRITE | MAY_APPEND); |
ec7ba118 | 309 | /* Make sure mounter can read file for copy up later */ |
500cac3c VG |
310 | mask |= MAY_READ; |
311 | } | |
4609e1f1 | 312 | err = inode_permission(mnt_idmap(realpath.mnt), realinode, mask); |
c0ca3d70 VG |
313 | revert_creds(old_cred); |
314 | ||
315 | return err; | |
e9be9d5e MS |
316 | } |
317 | ||
6b255391 | 318 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
319 | struct inode *inode, |
320 | struct delayed_call *done) | |
e9be9d5e | 321 | { |
1175b6b8 VG |
322 | const struct cred *old_cred; |
323 | const char *p; | |
e9be9d5e | 324 | |
6b255391 AV |
325 | if (!dentry) |
326 | return ERR_PTR(-ECHILD); | |
327 | ||
1175b6b8 | 328 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 329 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
330 | revert_creds(old_cred); |
331 | return p; | |
e9be9d5e MS |
332 | } |
333 | ||
610afc0b | 334 | bool ovl_is_private_xattr(struct super_block *sb, const char *name) |
e9be9d5e | 335 | { |
2d2f2d73 MS |
336 | struct ovl_fs *ofs = sb->s_fs_info; |
337 | ||
338 | if (ofs->config.userxattr) | |
339 | return strncmp(name, OVL_XATTR_USER_PREFIX, | |
340 | sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0; | |
341 | else | |
342 | return strncmp(name, OVL_XATTR_TRUSTED_PREFIX, | |
343 | sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0; | |
e9be9d5e MS |
344 | } |
345 | ||
1d88f183 MS |
346 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
347 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
348 | { |
349 | int err; | |
c914c0e2 | 350 | struct ovl_fs *ofs = OVL_FS(dentry->d_sb); |
1d88f183 MS |
351 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
352 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
8bc0095d | 353 | struct path realpath; |
1175b6b8 | 354 | const struct cred *old_cred; |
e9be9d5e MS |
355 | |
356 | err = ovl_want_write(dentry); | |
357 | if (err) | |
358 | goto out; | |
359 | ||
1d88f183 | 360 | if (!value && !upperdentry) { |
8bc0095d | 361 | ovl_path_lower(dentry, &realpath); |
554677b9 | 362 | old_cred = ovl_override_creds(dentry->d_sb); |
4609e1f1 | 363 | err = vfs_getxattr(mnt_idmap(realpath.mnt), realdentry, name, NULL, 0); |
554677b9 | 364 | revert_creds(old_cred); |
0e585ccc AG |
365 | if (err < 0) |
366 | goto out_drop_write; | |
367 | } | |
368 | ||
1d88f183 MS |
369 | if (!upperdentry) { |
370 | err = ovl_copy_up(dentry); | |
371 | if (err) | |
372 | goto out_drop_write; | |
e9be9d5e | 373 | |
1d88f183 MS |
374 | realdentry = ovl_dentry_upper(dentry); |
375 | } | |
0e585ccc | 376 | |
1175b6b8 | 377 | old_cred = ovl_override_creds(dentry->d_sb); |
c914c0e2 AG |
378 | if (value) { |
379 | err = ovl_do_setxattr(ofs, realdentry, name, value, size, | |
380 | flags); | |
381 | } else { | |
0e585ccc | 382 | WARN_ON(flags != XATTR_REPLACE); |
c914c0e2 | 383 | err = ovl_do_removexattr(ofs, realdentry, name); |
0e585ccc | 384 | } |
1175b6b8 | 385 | revert_creds(old_cred); |
e9be9d5e | 386 | |
d9854c87 | 387 | /* copy c/mtime */ |
2878dffc | 388 | ovl_copyattr(inode); |
d9854c87 | 389 | |
e9be9d5e MS |
390 | out_drop_write: |
391 | ovl_drop_write(dentry); | |
392 | out: | |
393 | return err; | |
394 | } | |
395 | ||
1d88f183 | 396 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 397 | void *value, size_t size) |
e9be9d5e | 398 | { |
1175b6b8 VG |
399 | ssize_t res; |
400 | const struct cred *old_cred; | |
8bc0095d | 401 | struct path realpath; |
52148463 | 402 | |
8bc0095d | 403 | ovl_i_path_real(inode, &realpath); |
1175b6b8 | 404 | old_cred = ovl_override_creds(dentry->d_sb); |
4609e1f1 | 405 | res = vfs_getxattr(mnt_idmap(realpath.mnt), realpath.dentry, name, value, size); |
1175b6b8 VG |
406 | revert_creds(old_cred); |
407 | return res; | |
e9be9d5e MS |
408 | } |
409 | ||
610afc0b | 410 | static bool ovl_can_list(struct super_block *sb, const char *s) |
a082c6f6 | 411 | { |
8f6ee74c MS |
412 | /* Never list private (.overlay) */ |
413 | if (ovl_is_private_xattr(sb, s)) | |
414 | return false; | |
415 | ||
f48bbfb2 | 416 | /* List all non-trusted xattrs */ |
a082c6f6 MS |
417 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) |
418 | return true; | |
419 | ||
8f6ee74c MS |
420 | /* list other trusted for superuser only */ |
421 | return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN); | |
a082c6f6 MS |
422 | } |
423 | ||
e9be9d5e MS |
424 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
425 | { | |
b581755b | 426 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 427 | ssize_t res; |
7cb35119 MS |
428 | size_t len; |
429 | char *s; | |
1175b6b8 | 430 | const struct cred *old_cred; |
e9be9d5e | 431 | |
1175b6b8 | 432 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 433 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 434 | revert_creds(old_cred); |
e9be9d5e MS |
435 | if (res <= 0 || size == 0) |
436 | return res; | |
437 | ||
e9be9d5e | 438 | /* filter out private xattrs */ |
7cb35119 MS |
439 | for (s = list, len = res; len;) { |
440 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 441 | |
7cb35119 MS |
442 | /* underlying fs providing us with an broken xattr list? */ |
443 | if (WARN_ON(slen > len)) | |
444 | return -EIO; | |
e9be9d5e | 445 | |
7cb35119 | 446 | len -= slen; |
610afc0b | 447 | if (!ovl_can_list(dentry->d_sb, s)) { |
e9be9d5e | 448 | res -= slen; |
7cb35119 | 449 | memmove(s, s + slen, len); |
e9be9d5e | 450 | } else { |
7cb35119 | 451 | s += slen; |
e9be9d5e MS |
452 | } |
453 | } | |
454 | ||
455 | return res; | |
456 | } | |
457 | ||
ded53656 | 458 | #ifdef CONFIG_FS_POSIX_ACL |
1aa5fef5 CB |
459 | /* |
460 | * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone | |
461 | * of the POSIX ACLs retrieved from the lower layer to this function to not | |
462 | * alter the POSIX ACLs for the underlying filesystem. | |
463 | */ | |
6c0a8bfb | 464 | static void ovl_idmap_posix_acl(const struct inode *realinode, |
4d7ca409 | 465 | struct mnt_idmap *idmap, |
1aa5fef5 CB |
466 | struct posix_acl *acl) |
467 | { | |
abfcf55d CB |
468 | struct user_namespace *fs_userns = i_user_ns(realinode); |
469 | ||
1aa5fef5 CB |
470 | for (unsigned int i = 0; i < acl->a_count; i++) { |
471 | vfsuid_t vfsuid; | |
472 | vfsgid_t vfsgid; | |
473 | ||
474 | struct posix_acl_entry *e = &acl->a_entries[i]; | |
475 | switch (e->e_tag) { | |
476 | case ACL_USER: | |
4d7ca409 | 477 | vfsuid = make_vfsuid(idmap, fs_userns, e->e_uid); |
1aa5fef5 CB |
478 | e->e_uid = vfsuid_into_kuid(vfsuid); |
479 | break; | |
480 | case ACL_GROUP: | |
4d7ca409 | 481 | vfsgid = make_vfsgid(idmap, fs_userns, e->e_gid); |
1aa5fef5 CB |
482 | e->e_gid = vfsgid_into_kgid(vfsgid); |
483 | break; | |
484 | } | |
485 | } | |
486 | } | |
487 | ||
6c0a8bfb CB |
488 | /* |
489 | * The @noperm argument is used to skip permission checking and is a temporary | |
490 | * measure. Quoting Miklos from an earlier discussion: | |
491 | * | |
492 | * > So there are two paths to getting an acl: | |
493 | * > 1) permission checking and 2) retrieving the value via getxattr(2). | |
494 | * > This is a similar situation as reading a symlink vs. following it. | |
495 | * > When following a symlink overlayfs always reads the link on the | |
496 | * > underlying fs just as if it was a readlink(2) call, calling | |
497 | * > security_inode_readlink() instead of security_inode_follow_link(). | |
498 | * > This is logical: we are reading the link from the underlying storage, | |
499 | * > and following it on overlayfs. | |
500 | * > | |
501 | * > Applying the same logic to acl: we do need to call the | |
502 | * > security_inode_getxattr() on the underlying fs, even if just want to | |
503 | * > check permissions on overlay. This is currently not done, which is an | |
504 | * > inconsistency. | |
505 | * > | |
506 | * > Maybe adding the check to ovl_get_acl() is the right way to go, but | |
507 | * > I'm a little afraid of a performance regression. Will look into that. | |
508 | * | |
509 | * Until we have made a decision allow this helper to take the @noperm | |
510 | * argument. We should hopefully be able to remove it soon. | |
511 | */ | |
31acceb9 CB |
512 | struct posix_acl *ovl_get_acl_path(const struct path *path, |
513 | const char *acl_name, bool noperm) | |
6c0a8bfb CB |
514 | { |
515 | struct posix_acl *real_acl, *clone; | |
77435322 | 516 | struct mnt_idmap *idmap; |
6c0a8bfb CB |
517 | struct inode *realinode = d_inode(path->dentry); |
518 | ||
77435322 | 519 | idmap = mnt_idmap(path->mnt); |
6c0a8bfb CB |
520 | |
521 | if (noperm) | |
522 | real_acl = get_inode_acl(realinode, posix_acl_type(acl_name)); | |
523 | else | |
77435322 | 524 | real_acl = vfs_get_acl(idmap, path->dentry, acl_name); |
6c0a8bfb CB |
525 | if (IS_ERR_OR_NULL(real_acl)) |
526 | return real_acl; | |
527 | ||
528 | if (!is_idmapped_mnt(path->mnt)) | |
529 | return real_acl; | |
530 | ||
531 | /* | |
532 | * We cannot alter the ACLs returned from the relevant layer as that | |
533 | * would alter the cached values filesystem wide for the lower | |
534 | * filesystem. Instead we can clone the ACLs and then apply the | |
535 | * relevant idmapping of the layer. | |
536 | */ | |
537 | clone = posix_acl_clone(real_acl, GFP_KERNEL); | |
538 | posix_acl_release(real_acl); /* release original acl */ | |
539 | if (!clone) | |
540 | return ERR_PTR(-ENOMEM); | |
541 | ||
4d7ca409 | 542 | ovl_idmap_posix_acl(realinode, idmap, clone); |
6c0a8bfb CB |
543 | return clone; |
544 | } | |
545 | ||
1aa5fef5 CB |
546 | /* |
547 | * When the relevant layer is an idmapped mount we need to take the idmapping | |
548 | * of the layer into account and translate any ACL_{GROUP,USER} values | |
549 | * according to the idmapped mount. | |
550 | * | |
551 | * We cannot alter the ACLs returned from the relevant layer as that would | |
552 | * alter the cached values filesystem wide for the lower filesystem. Instead we | |
553 | * can clone the ACLs and then apply the relevant idmapping of the layer. | |
554 | * | |
555 | * This is obviously only relevant when idmapped layers are used. | |
556 | */ | |
77435322 | 557 | struct posix_acl *do_ovl_get_acl(struct mnt_idmap *idmap, |
6c0a8bfb CB |
558 | struct inode *inode, int type, |
559 | bool rcu, bool noperm) | |
39a25b2b | 560 | { |
09d8b586 | 561 | struct inode *realinode = ovl_inode_real(inode); |
6c0a8bfb | 562 | struct posix_acl *acl; |
1aa5fef5 | 563 | struct path realpath; |
39a25b2b | 564 | |
ded53656 | 565 | if (!IS_POSIXACL(realinode)) |
39a25b2b VG |
566 | return NULL; |
567 | ||
1aa5fef5 CB |
568 | /* Careful in RCU walk mode */ |
569 | ovl_i_path_real(inode, &realpath); | |
570 | if (!realpath.dentry) { | |
571 | WARN_ON(!rcu); | |
572 | return ERR_PTR(-ECHILD); | |
573 | } | |
332f606b | 574 | |
1aa5fef5 | 575 | if (rcu) { |
6c0a8bfb CB |
576 | /* |
577 | * If the layer is idmapped drop out of RCU path walk | |
578 | * so we can clone the ACLs. | |
579 | */ | |
580 | if (is_idmapped_mnt(realpath.mnt)) | |
581 | return ERR_PTR(-ECHILD); | |
582 | ||
1aa5fef5 CB |
583 | acl = get_cached_acl_rcu(realinode, type); |
584 | } else { | |
585 | const struct cred *old_cred; | |
586 | ||
587 | old_cred = ovl_override_creds(inode->i_sb); | |
6c0a8bfb | 588 | acl = ovl_get_acl_path(&realpath, posix_acl_xattr_name(type), noperm); |
1aa5fef5 CB |
589 | revert_creds(old_cred); |
590 | } | |
1175b6b8 | 591 | |
6c0a8bfb | 592 | return acl; |
39a25b2b | 593 | } |
0e641857 CB |
594 | |
595 | static int ovl_set_or_remove_acl(struct dentry *dentry, struct inode *inode, | |
596 | struct posix_acl *acl, int type) | |
597 | { | |
598 | int err; | |
599 | struct path realpath; | |
600 | const char *acl_name; | |
601 | const struct cred *old_cred; | |
602 | struct ovl_fs *ofs = OVL_FS(dentry->d_sb); | |
603 | struct dentry *upperdentry = ovl_dentry_upper(dentry); | |
604 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
605 | ||
606 | err = ovl_want_write(dentry); | |
607 | if (err) | |
608 | return err; | |
609 | ||
610 | /* | |
611 | * If ACL is to be removed from a lower file, check if it exists in | |
612 | * the first place before copying it up. | |
613 | */ | |
614 | acl_name = posix_acl_xattr_name(type); | |
615 | if (!acl && !upperdentry) { | |
616 | struct posix_acl *real_acl; | |
617 | ||
618 | ovl_path_lower(dentry, &realpath); | |
619 | old_cred = ovl_override_creds(dentry->d_sb); | |
77435322 | 620 | real_acl = vfs_get_acl(mnt_idmap(realpath.mnt), realdentry, |
0e641857 CB |
621 | acl_name); |
622 | revert_creds(old_cred); | |
0e641857 CB |
623 | if (IS_ERR(real_acl)) { |
624 | err = PTR_ERR(real_acl); | |
625 | goto out_drop_write; | |
626 | } | |
5b52aebe | 627 | posix_acl_release(real_acl); |
0e641857 CB |
628 | } |
629 | ||
630 | if (!upperdentry) { | |
631 | err = ovl_copy_up(dentry); | |
632 | if (err) | |
633 | goto out_drop_write; | |
634 | ||
635 | realdentry = ovl_dentry_upper(dentry); | |
636 | } | |
637 | ||
638 | old_cred = ovl_override_creds(dentry->d_sb); | |
639 | if (acl) | |
640 | err = ovl_do_set_acl(ofs, realdentry, acl_name, acl); | |
641 | else | |
642 | err = ovl_do_remove_acl(ofs, realdentry, acl_name); | |
643 | revert_creds(old_cred); | |
644 | ||
645 | /* copy c/mtime */ | |
646 | ovl_copyattr(inode); | |
647 | ||
648 | out_drop_write: | |
649 | ovl_drop_write(dentry); | |
650 | return err; | |
651 | } | |
652 | ||
13e83a49 | 653 | int ovl_set_acl(struct mnt_idmap *idmap, struct dentry *dentry, |
0e641857 CB |
654 | struct posix_acl *acl, int type) |
655 | { | |
656 | int err; | |
657 | struct inode *inode = d_inode(dentry); | |
658 | struct dentry *workdir = ovl_workdir(dentry); | |
659 | struct inode *realinode = ovl_inode_real(inode); | |
660 | ||
661 | if (!IS_POSIXACL(d_inode(workdir))) | |
662 | return -EOPNOTSUPP; | |
663 | if (!realinode->i_op->set_acl) | |
664 | return -EOPNOTSUPP; | |
665 | if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) | |
666 | return acl ? -EACCES : 0; | |
01beba79 | 667 | if (!inode_owner_or_capable(&nop_mnt_idmap, inode)) |
0e641857 CB |
668 | return -EPERM; |
669 | ||
670 | /* | |
671 | * Check if sgid bit needs to be cleared (actual setacl operation will | |
672 | * be done with mounter's capabilities and so that won't do it for us). | |
673 | */ | |
674 | if (unlikely(inode->i_mode & S_ISGID) && type == ACL_TYPE_ACCESS && | |
675 | !in_group_p(inode->i_gid) && | |
9452e93e | 676 | !capable_wrt_inode_uidgid(&nop_mnt_idmap, inode, CAP_FSETID)) { |
0e641857 CB |
677 | struct iattr iattr = { .ia_valid = ATTR_KILL_SGID }; |
678 | ||
c1632a0f | 679 | err = ovl_setattr(&nop_mnt_idmap, dentry, &iattr); |
0e641857 CB |
680 | if (err) |
681 | return err; | |
682 | } | |
683 | ||
684 | return ovl_set_or_remove_acl(dentry, inode, acl, type); | |
685 | } | |
ded53656 | 686 | #endif |
39a25b2b | 687 | |
95582b00 | 688 | int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags) |
d719e8f2 | 689 | { |
8f35cf51 MS |
690 | if (flags & S_ATIME) { |
691 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
692 | struct path upperpath = { | |
08f4c7c8 | 693 | .mnt = ovl_upper_mnt(ofs), |
8f35cf51 MS |
694 | .dentry = ovl_upperdentry_dereference(OVL_I(inode)), |
695 | }; | |
696 | ||
697 | if (upperpath.dentry) { | |
698 | touch_atime(&upperpath); | |
699 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
700 | } | |
d719e8f2 | 701 | } |
d719e8f2 MS |
702 | return 0; |
703 | } | |
704 | ||
9e142c41 MS |
705 | static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
706 | u64 start, u64 len) | |
707 | { | |
708 | int err; | |
c11faf32 | 709 | struct inode *realinode = ovl_inode_realdata(inode); |
9e142c41 MS |
710 | const struct cred *old_cred; |
711 | ||
712 | if (!realinode->i_op->fiemap) | |
713 | return -EOPNOTSUPP; | |
714 | ||
715 | old_cred = ovl_override_creds(inode->i_sb); | |
716 | err = realinode->i_op->fiemap(realinode, fieinfo, start, len); | |
717 | revert_creds(old_cred); | |
718 | ||
719 | return err; | |
720 | } | |
721 | ||
66dbfabf MS |
722 | /* |
723 | * Work around the fact that security_file_ioctl() takes a file argument. | |
724 | * Introducing security_inode_fileattr_get/set() hooks would solve this issue | |
725 | * properly. | |
726 | */ | |
2d343087 | 727 | static int ovl_security_fileattr(const struct path *realpath, struct fileattr *fa, |
66dbfabf MS |
728 | bool set) |
729 | { | |
66dbfabf MS |
730 | struct file *file; |
731 | unsigned int cmd; | |
732 | int err; | |
733 | ||
72db8211 | 734 | file = dentry_open(realpath, O_RDONLY, current_cred()); |
66dbfabf MS |
735 | if (IS_ERR(file)) |
736 | return PTR_ERR(file); | |
737 | ||
738 | if (set) | |
739 | cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS; | |
740 | else | |
741 | cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS; | |
742 | ||
743 | err = security_file_ioctl(file, cmd, 0); | |
744 | fput(file); | |
745 | ||
746 | return err; | |
747 | } | |
748 | ||
2d343087 | 749 | int ovl_real_fileattr_set(const struct path *realpath, struct fileattr *fa) |
72db8211 AG |
750 | { |
751 | int err; | |
752 | ||
753 | err = ovl_security_fileattr(realpath, fa, true); | |
754 | if (err) | |
755 | return err; | |
756 | ||
8782a9ae | 757 | return vfs_fileattr_set(mnt_idmap(realpath->mnt), realpath->dentry, fa); |
72db8211 AG |
758 | } |
759 | ||
8782a9ae | 760 | int ovl_fileattr_set(struct mnt_idmap *idmap, |
66dbfabf MS |
761 | struct dentry *dentry, struct fileattr *fa) |
762 | { | |
763 | struct inode *inode = d_inode(dentry); | |
72db8211 | 764 | struct path upperpath; |
66dbfabf | 765 | const struct cred *old_cred; |
096a218a | 766 | unsigned int flags; |
66dbfabf MS |
767 | int err; |
768 | ||
769 | err = ovl_want_write(dentry); | |
770 | if (err) | |
771 | goto out; | |
772 | ||
773 | err = ovl_copy_up(dentry); | |
774 | if (!err) { | |
72db8211 | 775 | ovl_path_real(dentry, &upperpath); |
66dbfabf MS |
776 | |
777 | old_cred = ovl_override_creds(inode->i_sb); | |
096a218a AG |
778 | /* |
779 | * Store immutable/append-only flags in xattr and clear them | |
780 | * in upper fileattr (in case they were set by older kernel) | |
781 | * so children of "ovl-immutable" directories lower aliases of | |
782 | * "ovl-immutable" hardlinks could be copied up. | |
783 | * Clear xattr when flags are cleared. | |
784 | */ | |
785 | err = ovl_set_protattr(inode, upperpath.dentry, fa); | |
786 | if (!err) | |
787 | err = ovl_real_fileattr_set(&upperpath, fa); | |
66dbfabf | 788 | revert_creds(old_cred); |
096a218a AG |
789 | |
790 | /* | |
791 | * Merge real inode flags with inode flags read from | |
792 | * overlay.protattr xattr | |
793 | */ | |
794 | flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK; | |
795 | ||
796 | BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK); | |
797 | flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK; | |
798 | inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK); | |
d8991e86 CX |
799 | |
800 | /* Update ctime */ | |
2878dffc | 801 | ovl_copyattr(inode); |
66dbfabf MS |
802 | } |
803 | ovl_drop_write(dentry); | |
804 | out: | |
805 | return err; | |
806 | } | |
807 | ||
096a218a AG |
808 | /* Convert inode protection flags to fileattr flags */ |
809 | static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa) | |
810 | { | |
811 | BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL); | |
812 | BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON); | |
813 | ||
814 | if (inode->i_flags & S_APPEND) { | |
815 | fa->flags |= FS_APPEND_FL; | |
816 | fa->fsx_xflags |= FS_XFLAG_APPEND; | |
817 | } | |
818 | if (inode->i_flags & S_IMMUTABLE) { | |
819 | fa->flags |= FS_IMMUTABLE_FL; | |
820 | fa->fsx_xflags |= FS_XFLAG_IMMUTABLE; | |
821 | } | |
822 | } | |
823 | ||
2d343087 | 824 | int ovl_real_fileattr_get(const struct path *realpath, struct fileattr *fa) |
72db8211 AG |
825 | { |
826 | int err; | |
827 | ||
828 | err = ovl_security_fileattr(realpath, fa, false); | |
829 | if (err) | |
830 | return err; | |
831 | ||
5b0a414d MS |
832 | err = vfs_fileattr_get(realpath->dentry, fa); |
833 | if (err == -ENOIOCTLCMD) | |
834 | err = -ENOTTY; | |
835 | return err; | |
72db8211 AG |
836 | } |
837 | ||
66dbfabf MS |
838 | int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
839 | { | |
840 | struct inode *inode = d_inode(dentry); | |
72db8211 | 841 | struct path realpath; |
66dbfabf MS |
842 | const struct cred *old_cred; |
843 | int err; | |
844 | ||
72db8211 AG |
845 | ovl_path_real(dentry, &realpath); |
846 | ||
66dbfabf | 847 | old_cred = ovl_override_creds(inode->i_sb); |
72db8211 | 848 | err = ovl_real_fileattr_get(&realpath, fa); |
096a218a | 849 | ovl_fileattr_prot_flags(inode, fa); |
66dbfabf MS |
850 | revert_creds(old_cred); |
851 | ||
852 | return err; | |
853 | } | |
854 | ||
e9be9d5e MS |
855 | static const struct inode_operations ovl_file_inode_operations = { |
856 | .setattr = ovl_setattr, | |
857 | .permission = ovl_permission, | |
858 | .getattr = ovl_getattr, | |
e9be9d5e | 859 | .listxattr = ovl_listxattr, |
6c0a8bfb CB |
860 | .get_inode_acl = ovl_get_inode_acl, |
861 | .get_acl = ovl_get_acl, | |
0e641857 | 862 | .set_acl = ovl_set_acl, |
d719e8f2 | 863 | .update_time = ovl_update_time, |
9e142c41 | 864 | .fiemap = ovl_fiemap, |
66dbfabf MS |
865 | .fileattr_get = ovl_fileattr_get, |
866 | .fileattr_set = ovl_fileattr_set, | |
e9be9d5e MS |
867 | }; |
868 | ||
869 | static const struct inode_operations ovl_symlink_inode_operations = { | |
870 | .setattr = ovl_setattr, | |
6b255391 | 871 | .get_link = ovl_get_link, |
e9be9d5e | 872 | .getattr = ovl_getattr, |
e9be9d5e | 873 | .listxattr = ovl_listxattr, |
d719e8f2 | 874 | .update_time = ovl_update_time, |
e9be9d5e MS |
875 | }; |
876 | ||
9e142c41 MS |
877 | static const struct inode_operations ovl_special_inode_operations = { |
878 | .setattr = ovl_setattr, | |
879 | .permission = ovl_permission, | |
880 | .getattr = ovl_getattr, | |
881 | .listxattr = ovl_listxattr, | |
6c0a8bfb CB |
882 | .get_inode_acl = ovl_get_inode_acl, |
883 | .get_acl = ovl_get_acl, | |
0e641857 | 884 | .set_acl = ovl_set_acl, |
9e142c41 MS |
885 | .update_time = ovl_update_time, |
886 | }; | |
887 | ||
69383c59 | 888 | static const struct address_space_operations ovl_aops = { |
5b910bd6 AG |
889 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
890 | .direct_IO = noop_direct_IO, | |
891 | }; | |
892 | ||
b1eaa950 AG |
893 | /* |
894 | * It is possible to stack overlayfs instance on top of another | |
a5a84682 | 895 | * overlayfs instance as lower layer. We need to annotate the |
b1eaa950 AG |
896 | * stackable i_mutex locks according to stack level of the super |
897 | * block instance. An overlayfs instance can never be in stack | |
898 | * depth 0 (there is always a real fs below it). An overlayfs | |
f48bbfb2 | 899 | * inode lock will use the lockdep annotation ovl_i_mutex_key[depth]. |
b1eaa950 AG |
900 | * |
901 | * For example, here is a snip from /proc/lockdep_chains after | |
902 | * dir_iterate of nested overlayfs: | |
903 | * | |
904 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
905 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
906 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
b1f9d385 AG |
907 | * |
908 | * Locking order w.r.t ovl_want_write() is important for nested overlayfs. | |
909 | * | |
910 | * This chain is valid: | |
911 | * - inode->i_rwsem (inode_lock[2]) | |
912 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
913 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
914 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
915 | * | |
916 | * And this chain is valid: | |
917 | * - inode->i_rwsem (inode_lock[2]) | |
918 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
919 | * - lowerinode->i_rwsem (inode_lock[1]) | |
920 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
921 | * | |
922 | * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is | |
923 | * held, because it is in reverse order of the non-nested case using the same | |
924 | * upper fs: | |
925 | * - inode->i_rwsem (inode_lock[1]) | |
926 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
927 | * - OVL_I(inode)->lock (ovl_inode_lock[1]) | |
b1eaa950 AG |
928 | */ |
929 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
930 | ||
931 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
932 | { | |
933 | #ifdef CONFIG_LOCKDEP | |
934 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
935 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 936 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
937 | |
938 | int depth = inode->i_sb->s_stack_depth - 1; | |
939 | ||
940 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
941 | depth = 0; | |
942 | ||
943 | if (S_ISDIR(inode->i_mode)) | |
944 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
945 | else | |
946 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
947 | |
948 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
949 | #endif |
950 | } | |
951 | ||
4d314f78 AG |
952 | static void ovl_next_ino(struct inode *inode) |
953 | { | |
954 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
955 | ||
956 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
957 | if (unlikely(!inode->i_ino)) | |
958 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
959 | } | |
960 | ||
62c832ed | 961 | static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid) |
e9be9d5e | 962 | { |
12574a9f | 963 | int xinobits = ovl_xino_bits(inode->i_sb); |
dfe51d47 | 964 | unsigned int xinoshift = 64 - xinobits; |
12574a9f | 965 | |
695b46e7 | 966 | /* |
6dde1e42 AG |
967 | * When d_ino is consistent with st_ino (samefs or i_ino has enough |
968 | * bits to encode layer), set the same value used for st_ino to i_ino, | |
969 | * so inode number exposed via /proc/locks and a like will be | |
970 | * consistent with d_ino and st_ino values. An i_ino value inconsistent | |
735c907d | 971 | * with d_ino also causes nfsd readdirplus to fail. |
695b46e7 | 972 | */ |
4d314f78 | 973 | inode->i_ino = ino; |
dfe51d47 AG |
974 | if (ovl_same_fs(inode->i_sb)) { |
975 | return; | |
976 | } else if (xinobits && likely(!(ino >> xinoshift))) { | |
977 | inode->i_ino |= (unsigned long)fsid << (xinoshift + 1); | |
978 | return; | |
979 | } | |
980 | ||
981 | /* | |
982 | * For directory inodes on non-samefs with xino disabled or xino | |
983 | * overflow, we allocate a non-persistent inode number, to be used for | |
984 | * resolving st_ino collisions in ovl_map_dev_ino(). | |
985 | * | |
986 | * To avoid ino collision with legitimate xino values from upper | |
987 | * layer (fsid 0), use the lowest xinobit to map the non | |
988 | * persistent inode numbers to the unified st_ino address space. | |
989 | */ | |
990 | if (S_ISDIR(inode->i_mode)) { | |
4d314f78 | 991 | ovl_next_ino(inode); |
dfe51d47 AG |
992 | if (xinobits) { |
993 | inode->i_ino &= ~0UL >> xinobits; | |
994 | inode->i_ino |= 1UL << xinoshift; | |
995 | } | |
12574a9f | 996 | } |
62c832ed AG |
997 | } |
998 | ||
999 | void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip, | |
1000 | unsigned long ino, int fsid) | |
1001 | { | |
1002 | struct inode *realinode; | |
ffa5723c | 1003 | struct ovl_inode *oi = OVL_I(inode); |
62c832ed AG |
1004 | |
1005 | if (oip->upperdentry) | |
ffa5723c AG |
1006 | oi->__upperdentry = oip->upperdentry; |
1007 | if (oip->lowerpath && oip->lowerpath->dentry) { | |
1008 | oi->lowerpath.dentry = dget(oip->lowerpath->dentry); | |
1009 | oi->lowerpath.layer = oip->lowerpath->layer; | |
1010 | } | |
62c832ed | 1011 | if (oip->lowerdata) |
ffa5723c | 1012 | oi->lowerdata = igrab(d_inode(oip->lowerdata)); |
62c832ed AG |
1013 | |
1014 | realinode = ovl_inode_real(inode); | |
2878dffc | 1015 | ovl_copyattr(inode); |
62c832ed AG |
1016 | ovl_copyflags(realinode, inode); |
1017 | ovl_map_ino(inode, ino, fsid); | |
1018 | } | |
1019 | ||
1020 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1021 | { | |
e9be9d5e | 1022 | inode->i_mode = mode; |
d719e8f2 | 1023 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
1024 | #ifdef CONFIG_FS_POSIX_ACL |
1025 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
1026 | #endif | |
e9be9d5e | 1027 | |
b1eaa950 AG |
1028 | ovl_lockdep_annotate_inode_mutex_key(inode); |
1029 | ||
ca4c8a3a MS |
1030 | switch (mode & S_IFMT) { |
1031 | case S_IFREG: | |
1032 | inode->i_op = &ovl_file_inode_operations; | |
d1d04ef8 | 1033 | inode->i_fop = &ovl_file_operations; |
5b910bd6 | 1034 | inode->i_mapping->a_ops = &ovl_aops; |
ca4c8a3a MS |
1035 | break; |
1036 | ||
e9be9d5e | 1037 | case S_IFDIR: |
e9be9d5e MS |
1038 | inode->i_op = &ovl_dir_inode_operations; |
1039 | inode->i_fop = &ovl_dir_operations; | |
1040 | break; | |
1041 | ||
1042 | case S_IFLNK: | |
1043 | inode->i_op = &ovl_symlink_inode_operations; | |
1044 | break; | |
1045 | ||
51f7e52d | 1046 | default: |
9e142c41 | 1047 | inode->i_op = &ovl_special_inode_operations; |
ca4c8a3a | 1048 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 1049 | break; |
51f7e52d MS |
1050 | } |
1051 | } | |
e9be9d5e | 1052 | |
5f8415d6 AG |
1053 | /* |
1054 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
1055 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
1056 | * upper inode, the following nlink modifying operations can happen: | |
1057 | * | |
1058 | * 1. Lower hardlink copy up | |
1059 | * 2. Upper hardlink created, unlinked or renamed over | |
1060 | * 3. Lower hardlink whiteout or renamed over | |
1061 | * | |
1062 | * For the first, copy up case, the union nlink does not change, whether the | |
1063 | * operation succeeds or fails, but the upper inode nlink may change. | |
1064 | * Therefore, before copy up, we store the union nlink value relative to the | |
2d2f2d73 | 1065 | * lower inode nlink in the index inode xattr .overlay.nlink. |
5f8415d6 AG |
1066 | * |
1067 | * For the second, upper hardlink case, the union nlink should be incremented | |
1068 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
1069 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
1070 | * value relative to the upper inode nlink in the index inode. | |
1071 | * | |
1072 | * For the last, lower cover up case, we simplify things by preceding the | |
1073 | * whiteout or cover up with copy up. This makes sure that there is an index | |
1074 | * upper inode where the nlink xattr can be stored before the copied up upper | |
1075 | * entry is unlink. | |
1076 | */ | |
1077 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
1078 | ||
1079 | /* | |
1080 | * On-disk format for indexed nlink: | |
1081 | * | |
1082 | * nlink relative to the upper inode - "U[+-]NUM" | |
1083 | * nlink relative to the lower inode - "L[+-]NUM" | |
1084 | */ | |
1085 | ||
1086 | static int ovl_set_nlink_common(struct dentry *dentry, | |
1087 | struct dentry *realdentry, const char *format) | |
1088 | { | |
1089 | struct inode *inode = d_inode(dentry); | |
1090 | struct inode *realinode = d_inode(realdentry); | |
1091 | char buf[13]; | |
1092 | int len; | |
1093 | ||
1094 | len = snprintf(buf, sizeof(buf), format, | |
1095 | (int) (inode->i_nlink - realinode->i_nlink)); | |
1096 | ||
6787341a MS |
1097 | if (WARN_ON(len >= sizeof(buf))) |
1098 | return -EIO; | |
1099 | ||
c914c0e2 AG |
1100 | return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry), |
1101 | OVL_XATTR_NLINK, buf, len); | |
5f8415d6 AG |
1102 | } |
1103 | ||
1104 | int ovl_set_nlink_upper(struct dentry *dentry) | |
1105 | { | |
1106 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
1107 | } | |
1108 | ||
1109 | int ovl_set_nlink_lower(struct dentry *dentry) | |
1110 | { | |
1111 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
1112 | } | |
1113 | ||
610afc0b | 1114 | unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry, |
caf70cb2 AG |
1115 | struct dentry *upperdentry, |
1116 | unsigned int fallback) | |
5f8415d6 AG |
1117 | { |
1118 | int nlink_diff; | |
1119 | int nlink; | |
1120 | char buf[13]; | |
1121 | int err; | |
1122 | ||
1123 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
1124 | return fallback; | |
1125 | ||
dad7017a CB |
1126 | err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK, |
1127 | &buf, sizeof(buf) - 1); | |
5f8415d6 AG |
1128 | if (err < 0) |
1129 | goto fail; | |
1130 | ||
1131 | buf[err] = '\0'; | |
1132 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
1133 | (buf[1] != '+' && buf[1] != '-')) | |
1134 | goto fail; | |
1135 | ||
1136 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
1137 | if (err < 0) | |
1138 | goto fail; | |
1139 | ||
1140 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
1141 | nlink += nlink_diff; | |
1142 | ||
1143 | if (nlink <= 0) | |
1144 | goto fail; | |
1145 | ||
1146 | return nlink; | |
1147 | ||
1148 | fail: | |
1bd0a3ae | 1149 | pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n", |
5f8415d6 AG |
1150 | upperdentry, err); |
1151 | return fallback; | |
1152 | } | |
1153 | ||
ca4c8a3a | 1154 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
1155 | { |
1156 | struct inode *inode; | |
1157 | ||
1158 | inode = new_inode(sb); | |
1159 | if (inode) | |
62c832ed | 1160 | ovl_fill_inode(inode, mode, rdev); |
51f7e52d MS |
1161 | |
1162 | return inode; | |
1163 | } | |
1164 | ||
1165 | static int ovl_inode_test(struct inode *inode, void *data) | |
1166 | { | |
25b7713a | 1167 | return inode->i_private == data; |
51f7e52d MS |
1168 | } |
1169 | ||
1170 | static int ovl_inode_set(struct inode *inode, void *data) | |
1171 | { | |
25b7713a | 1172 | inode->i_private = data; |
51f7e52d MS |
1173 | return 0; |
1174 | } | |
1175 | ||
b9ac5c27 | 1176 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
4b91c30a | 1177 | struct dentry *upperdentry, bool strict) |
b9ac5c27 | 1178 | { |
4b91c30a AG |
1179 | /* |
1180 | * For directories, @strict verify from lookup path performs consistency | |
1181 | * checks, so NULL lower/upper in dentry must match NULL lower/upper in | |
1182 | * inode. Non @strict verify from NFS handle decode path passes NULL for | |
1183 | * 'unknown' lower/upper. | |
1184 | */ | |
1185 | if (S_ISDIR(inode->i_mode) && strict) { | |
31747eda AG |
1186 | /* Real lower dir moved to upper layer under us? */ |
1187 | if (!lowerdentry && ovl_inode_lower(inode)) | |
1188 | return false; | |
1189 | ||
1190 | /* Lookup of an uncovered redirect origin? */ | |
1191 | if (!upperdentry && ovl_inode_upper(inode)) | |
1192 | return false; | |
1193 | } | |
1194 | ||
939ae4ef AG |
1195 | /* |
1196 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
1197 | * This happens when finding a copied up overlay inode for a renamed | |
1198 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
1199 | * by origin because lower fs does not support file handles. | |
1200 | */ | |
1201 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
1202 | return false; |
1203 | ||
1204 | /* | |
1205 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
1206 | * This happens when finding a lower alias for a copied up hard link. | |
1207 | */ | |
1208 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
1209 | return false; | |
1210 | ||
1211 | return true; | |
1212 | } | |
1213 | ||
4b91c30a AG |
1214 | struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real, |
1215 | bool is_upper) | |
9436a1a3 | 1216 | { |
4b91c30a | 1217 | struct inode *inode, *key = d_inode(real); |
9436a1a3 AG |
1218 | |
1219 | inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
1220 | if (!inode) | |
1221 | return NULL; | |
1222 | ||
4b91c30a AG |
1223 | if (!ovl_verify_inode(inode, is_upper ? NULL : real, |
1224 | is_upper ? real : NULL, false)) { | |
9436a1a3 AG |
1225 | iput(inode); |
1226 | return ERR_PTR(-ESTALE); | |
1227 | } | |
1228 | ||
1229 | return inode; | |
1230 | } | |
1231 | ||
146d62e5 AG |
1232 | bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir) |
1233 | { | |
1234 | struct inode *key = d_inode(dir); | |
1235 | struct inode *trap; | |
1236 | bool res; | |
1237 | ||
1238 | trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
1239 | if (!trap) | |
1240 | return false; | |
1241 | ||
1242 | res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) && | |
1243 | !ovl_inode_lower(trap); | |
1244 | ||
1245 | iput(trap); | |
1246 | return res; | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * Create an inode cache entry for layer root dir, that will intentionally | |
1251 | * fail ovl_verify_inode(), so any lookup that will find some layer root | |
1252 | * will fail. | |
1253 | */ | |
1254 | struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir) | |
1255 | { | |
1256 | struct inode *key = d_inode(dir); | |
1257 | struct inode *trap; | |
1258 | ||
1259 | if (!d_is_dir(dir)) | |
1260 | return ERR_PTR(-ENOTDIR); | |
1261 | ||
1262 | trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test, | |
1263 | ovl_inode_set, key); | |
1264 | if (!trap) | |
1265 | return ERR_PTR(-ENOMEM); | |
1266 | ||
1267 | if (!(trap->i_state & I_NEW)) { | |
1268 | /* Conflicting layer roots? */ | |
1269 | iput(trap); | |
1270 | return ERR_PTR(-ELOOP); | |
1271 | } | |
1272 | ||
1273 | trap->i_mode = S_IFDIR; | |
1274 | trap->i_flags = S_DEAD; | |
1275 | unlock_new_inode(trap); | |
1276 | ||
1277 | return trap; | |
1278 | } | |
1279 | ||
764baba8 AG |
1280 | /* |
1281 | * Does overlay inode need to be hashed by lower inode? | |
1282 | */ | |
1283 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
74c6e384 | 1284 | struct dentry *lower, bool index) |
764baba8 AG |
1285 | { |
1286 | struct ovl_fs *ofs = sb->s_fs_info; | |
1287 | ||
1288 | /* No, if pure upper */ | |
1289 | if (!lower) | |
1290 | return false; | |
1291 | ||
1292 | /* Yes, if already indexed */ | |
1293 | if (index) | |
1294 | return true; | |
1295 | ||
1296 | /* Yes, if won't be copied up */ | |
08f4c7c8 | 1297 | if (!ovl_upper_mnt(ofs)) |
764baba8 AG |
1298 | return true; |
1299 | ||
1300 | /* No, if lower hardlink is or will be broken on copy up */ | |
1301 | if ((upper || !ovl_indexdir(sb)) && | |
1302 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
1303 | return false; | |
1304 | ||
1305 | /* No, if non-indexed upper with NFS export */ | |
1306 | if (sb->s_export_op && upper) | |
1307 | return false; | |
1308 | ||
1309 | /* Otherwise, hash by lower inode for fsnotify */ | |
1310 | return true; | |
1311 | } | |
1312 | ||
01b39dcc AG |
1313 | static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode, |
1314 | struct inode *key) | |
1315 | { | |
1316 | return newinode ? inode_insert5(newinode, (unsigned long) key, | |
1317 | ovl_inode_test, ovl_inode_set, key) : | |
1318 | iget5_locked(sb, (unsigned long) key, | |
1319 | ovl_inode_test, ovl_inode_set, key); | |
1320 | } | |
1321 | ||
ac6a52eb VG |
1322 | struct inode *ovl_get_inode(struct super_block *sb, |
1323 | struct ovl_inode_params *oip) | |
51f7e52d | 1324 | { |
610afc0b | 1325 | struct ovl_fs *ofs = OVL_FS(sb); |
ac6a52eb VG |
1326 | struct dentry *upperdentry = oip->upperdentry; |
1327 | struct ovl_path *lowerpath = oip->lowerpath; | |
09d8b586 | 1328 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; |
51f7e52d | 1329 | struct inode *inode; |
12574a9f | 1330 | struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL; |
dad7017a CB |
1331 | struct path realpath = { |
1332 | .dentry = upperdentry ?: lowerdentry, | |
1333 | .mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt, | |
1334 | }; | |
ac6a52eb VG |
1335 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, |
1336 | oip->index); | |
300b124f | 1337 | int fsid = bylower ? lowerpath->layer->fsid : 0; |
28166ab3 | 1338 | bool is_dir; |
695b46e7 | 1339 | unsigned long ino = 0; |
acf3062a | 1340 | int err = oip->newinode ? -EEXIST : -ENOMEM; |
6eaf0111 | 1341 | |
09d8b586 MS |
1342 | if (!realinode) |
1343 | realinode = d_inode(lowerdentry); | |
1344 | ||
6eaf0111 | 1345 | /* |
764baba8 AG |
1346 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
1347 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 1348 | */ |
31747eda | 1349 | is_dir = S_ISDIR(realinode->i_mode); |
764baba8 AG |
1350 | if (upperdentry || bylower) { |
1351 | struct inode *key = d_inode(bylower ? lowerdentry : | |
1352 | upperdentry); | |
31747eda | 1353 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 1354 | |
01b39dcc | 1355 | inode = ovl_iget5(sb, oip->newinode, key); |
09d8b586 | 1356 | if (!inode) |
027065b7 | 1357 | goto out_err; |
09d8b586 | 1358 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
1359 | /* |
1360 | * Verify that the underlying files stored in the inode | |
1361 | * match those in the dentry. | |
1362 | */ | |
4b91c30a AG |
1363 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry, |
1364 | true)) { | |
b9ac5c27 | 1365 | iput(inode); |
027065b7 VG |
1366 | err = -ESTALE; |
1367 | goto out_err; | |
b9ac5c27 MS |
1368 | } |
1369 | ||
09d8b586 | 1370 | dput(upperdentry); |
9cec54c8 | 1371 | kfree(oip->redirect); |
09d8b586 MS |
1372 | goto out; |
1373 | } | |
e6d2ebdd | 1374 | |
31747eda AG |
1375 | /* Recalculate nlink for non-dir due to indexing */ |
1376 | if (!is_dir) | |
610afc0b MS |
1377 | nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry, |
1378 | nlink); | |
5f8415d6 | 1379 | set_nlink(inode, nlink); |
695b46e7 | 1380 | ino = key->i_ino; |
e6d2ebdd | 1381 | } else { |
764baba8 | 1382 | /* Lower hardlink that will be broken on copy up */ |
0aceb53e | 1383 | inode = new_inode(sb); |
027065b7 VG |
1384 | if (!inode) { |
1385 | err = -ENOMEM; | |
1386 | goto out_err; | |
1387 | } | |
300b124f AG |
1388 | ino = realinode->i_ino; |
1389 | fsid = lowerpath->layer->fsid; | |
e9be9d5e | 1390 | } |
62c832ed AG |
1391 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev); |
1392 | ovl_inode_init(inode, oip, ino, fsid); | |
13c72075 | 1393 | |
610afc0b | 1394 | if (upperdentry && ovl_is_impuredir(sb, upperdentry)) |
13c72075 MS |
1395 | ovl_set_flag(OVL_IMPURE, inode); |
1396 | ||
ac6a52eb | 1397 | if (oip->index) |
0471a9cd VG |
1398 | ovl_set_flag(OVL_INDEX, inode); |
1399 | ||
9cec54c8 VG |
1400 | OVL_I(inode)->redirect = oip->redirect; |
1401 | ||
a00c2d59 VG |
1402 | if (bylower) |
1403 | ovl_set_flag(OVL_CONST_INO, inode); | |
1404 | ||
b79e05aa | 1405 | /* Check for non-merge dir that may have whiteouts */ |
31747eda | 1406 | if (is_dir) { |
ac6a52eb | 1407 | if (((upperdentry && lowerdentry) || oip->numlower > 1) || |
dad7017a | 1408 | ovl_path_check_origin_xattr(ofs, &realpath)) { |
b79e05aa AG |
1409 | ovl_set_flag(OVL_WHITEOUTS, inode); |
1410 | } | |
1411 | } | |
1412 | ||
096a218a AG |
1413 | /* Check for immutable/append-only inode flags in xattr */ |
1414 | if (upperdentry) | |
1415 | ovl_check_protattr(inode, upperdentry); | |
1416 | ||
e6d2ebdd MS |
1417 | if (inode->i_state & I_NEW) |
1418 | unlock_new_inode(inode); | |
1419 | out: | |
e9be9d5e | 1420 | return inode; |
b9ac5c27 | 1421 | |
027065b7 | 1422 | out_err: |
1bd0a3ae | 1423 | pr_warn_ratelimited("failed to get inode (%i)\n", err); |
027065b7 | 1424 | inode = ERR_PTR(err); |
b9ac5c27 | 1425 | goto out; |
e9be9d5e | 1426 | } |