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 */ |
4b7791b2 CB |
291 | ovl_i_path_real(inode, &realpath); |
292 | if (!realpath.dentry) { | |
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 | ||
4b7791b2 | 305 | realinode = d_inode(realpath.dentry); |
ec7ba118 MS |
306 | old_cred = ovl_override_creds(inode->i_sb); |
307 | if (!upperinode && | |
308 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 309 | mask &= ~(MAY_WRITE | MAY_APPEND); |
ec7ba118 | 310 | /* Make sure mounter can read file for copy up later */ |
500cac3c VG |
311 | mask |= MAY_READ; |
312 | } | |
4609e1f1 | 313 | err = inode_permission(mnt_idmap(realpath.mnt), realinode, mask); |
c0ca3d70 VG |
314 | revert_creds(old_cred); |
315 | ||
316 | return err; | |
e9be9d5e MS |
317 | } |
318 | ||
6b255391 | 319 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
320 | struct inode *inode, |
321 | struct delayed_call *done) | |
e9be9d5e | 322 | { |
1175b6b8 VG |
323 | const struct cred *old_cred; |
324 | const char *p; | |
e9be9d5e | 325 | |
6b255391 AV |
326 | if (!dentry) |
327 | return ERR_PTR(-ECHILD); | |
328 | ||
1175b6b8 | 329 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 330 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
331 | revert_creds(old_cred); |
332 | return p; | |
e9be9d5e MS |
333 | } |
334 | ||
610afc0b | 335 | bool ovl_is_private_xattr(struct super_block *sb, const char *name) |
e9be9d5e | 336 | { |
2d2f2d73 MS |
337 | struct ovl_fs *ofs = sb->s_fs_info; |
338 | ||
339 | if (ofs->config.userxattr) | |
340 | return strncmp(name, OVL_XATTR_USER_PREFIX, | |
341 | sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0; | |
342 | else | |
343 | return strncmp(name, OVL_XATTR_TRUSTED_PREFIX, | |
344 | sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0; | |
e9be9d5e MS |
345 | } |
346 | ||
1d88f183 MS |
347 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
348 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
349 | { |
350 | int err; | |
c914c0e2 | 351 | struct ovl_fs *ofs = OVL_FS(dentry->d_sb); |
1d88f183 MS |
352 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
353 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
8bc0095d | 354 | struct path realpath; |
1175b6b8 | 355 | const struct cred *old_cred; |
e9be9d5e MS |
356 | |
357 | err = ovl_want_write(dentry); | |
358 | if (err) | |
359 | goto out; | |
360 | ||
1d88f183 | 361 | if (!value && !upperdentry) { |
8bc0095d | 362 | ovl_path_lower(dentry, &realpath); |
554677b9 | 363 | old_cred = ovl_override_creds(dentry->d_sb); |
4609e1f1 | 364 | err = vfs_getxattr(mnt_idmap(realpath.mnt), realdentry, name, NULL, 0); |
554677b9 | 365 | revert_creds(old_cred); |
0e585ccc AG |
366 | if (err < 0) |
367 | goto out_drop_write; | |
368 | } | |
369 | ||
1d88f183 MS |
370 | if (!upperdentry) { |
371 | err = ovl_copy_up(dentry); | |
372 | if (err) | |
373 | goto out_drop_write; | |
e9be9d5e | 374 | |
1d88f183 MS |
375 | realdentry = ovl_dentry_upper(dentry); |
376 | } | |
0e585ccc | 377 | |
1175b6b8 | 378 | old_cred = ovl_override_creds(dentry->d_sb); |
c914c0e2 AG |
379 | if (value) { |
380 | err = ovl_do_setxattr(ofs, realdentry, name, value, size, | |
381 | flags); | |
382 | } else { | |
0e585ccc | 383 | WARN_ON(flags != XATTR_REPLACE); |
c914c0e2 | 384 | err = ovl_do_removexattr(ofs, realdentry, name); |
0e585ccc | 385 | } |
1175b6b8 | 386 | revert_creds(old_cred); |
e9be9d5e | 387 | |
d9854c87 | 388 | /* copy c/mtime */ |
2878dffc | 389 | ovl_copyattr(inode); |
d9854c87 | 390 | |
e9be9d5e MS |
391 | out_drop_write: |
392 | ovl_drop_write(dentry); | |
393 | out: | |
394 | return err; | |
395 | } | |
396 | ||
1d88f183 | 397 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 398 | void *value, size_t size) |
e9be9d5e | 399 | { |
1175b6b8 VG |
400 | ssize_t res; |
401 | const struct cred *old_cred; | |
8bc0095d | 402 | struct path realpath; |
52148463 | 403 | |
8bc0095d | 404 | ovl_i_path_real(inode, &realpath); |
1175b6b8 | 405 | old_cred = ovl_override_creds(dentry->d_sb); |
4609e1f1 | 406 | res = vfs_getxattr(mnt_idmap(realpath.mnt), realpath.dentry, name, value, size); |
1175b6b8 VG |
407 | revert_creds(old_cred); |
408 | return res; | |
e9be9d5e MS |
409 | } |
410 | ||
610afc0b | 411 | static bool ovl_can_list(struct super_block *sb, const char *s) |
a082c6f6 | 412 | { |
8f6ee74c MS |
413 | /* Never list private (.overlay) */ |
414 | if (ovl_is_private_xattr(sb, s)) | |
415 | return false; | |
416 | ||
f48bbfb2 | 417 | /* List all non-trusted xattrs */ |
a082c6f6 MS |
418 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) |
419 | return true; | |
420 | ||
8f6ee74c MS |
421 | /* list other trusted for superuser only */ |
422 | return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN); | |
a082c6f6 MS |
423 | } |
424 | ||
e9be9d5e MS |
425 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
426 | { | |
b581755b | 427 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 428 | ssize_t res; |
7cb35119 MS |
429 | size_t len; |
430 | char *s; | |
1175b6b8 | 431 | const struct cred *old_cred; |
e9be9d5e | 432 | |
1175b6b8 | 433 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 434 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 435 | revert_creds(old_cred); |
e9be9d5e MS |
436 | if (res <= 0 || size == 0) |
437 | return res; | |
438 | ||
e9be9d5e | 439 | /* filter out private xattrs */ |
7cb35119 MS |
440 | for (s = list, len = res; len;) { |
441 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 442 | |
7cb35119 MS |
443 | /* underlying fs providing us with an broken xattr list? */ |
444 | if (WARN_ON(slen > len)) | |
445 | return -EIO; | |
e9be9d5e | 446 | |
7cb35119 | 447 | len -= slen; |
610afc0b | 448 | if (!ovl_can_list(dentry->d_sb, s)) { |
e9be9d5e | 449 | res -= slen; |
7cb35119 | 450 | memmove(s, s + slen, len); |
e9be9d5e | 451 | } else { |
7cb35119 | 452 | s += slen; |
e9be9d5e MS |
453 | } |
454 | } | |
455 | ||
456 | return res; | |
457 | } | |
458 | ||
ded53656 | 459 | #ifdef CONFIG_FS_POSIX_ACL |
1aa5fef5 CB |
460 | /* |
461 | * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone | |
462 | * of the POSIX ACLs retrieved from the lower layer to this function to not | |
463 | * alter the POSIX ACLs for the underlying filesystem. | |
464 | */ | |
6c0a8bfb | 465 | static void ovl_idmap_posix_acl(const struct inode *realinode, |
abfcf55d | 466 | struct user_namespace *mnt_userns, |
1aa5fef5 CB |
467 | struct posix_acl *acl) |
468 | { | |
abfcf55d CB |
469 | struct user_namespace *fs_userns = i_user_ns(realinode); |
470 | ||
1aa5fef5 CB |
471 | for (unsigned int i = 0; i < acl->a_count; i++) { |
472 | vfsuid_t vfsuid; | |
473 | vfsgid_t vfsgid; | |
474 | ||
475 | struct posix_acl_entry *e = &acl->a_entries[i]; | |
476 | switch (e->e_tag) { | |
477 | case ACL_USER: | |
abfcf55d | 478 | vfsuid = make_vfsuid(mnt_userns, fs_userns, e->e_uid); |
1aa5fef5 CB |
479 | e->e_uid = vfsuid_into_kuid(vfsuid); |
480 | break; | |
481 | case ACL_GROUP: | |
abfcf55d | 482 | vfsgid = make_vfsgid(mnt_userns, fs_userns, e->e_gid); |
1aa5fef5 CB |
483 | e->e_gid = vfsgid_into_kgid(vfsgid); |
484 | break; | |
485 | } | |
486 | } | |
487 | } | |
488 | ||
6c0a8bfb CB |
489 | /* |
490 | * The @noperm argument is used to skip permission checking and is a temporary | |
491 | * measure. Quoting Miklos from an earlier discussion: | |
492 | * | |
493 | * > So there are two paths to getting an acl: | |
494 | * > 1) permission checking and 2) retrieving the value via getxattr(2). | |
495 | * > This is a similar situation as reading a symlink vs. following it. | |
496 | * > When following a symlink overlayfs always reads the link on the | |
497 | * > underlying fs just as if it was a readlink(2) call, calling | |
498 | * > security_inode_readlink() instead of security_inode_follow_link(). | |
499 | * > This is logical: we are reading the link from the underlying storage, | |
500 | * > and following it on overlayfs. | |
501 | * > | |
502 | * > Applying the same logic to acl: we do need to call the | |
503 | * > security_inode_getxattr() on the underlying fs, even if just want to | |
504 | * > check permissions on overlay. This is currently not done, which is an | |
505 | * > inconsistency. | |
506 | * > | |
507 | * > Maybe adding the check to ovl_get_acl() is the right way to go, but | |
508 | * > I'm a little afraid of a performance regression. Will look into that. | |
509 | * | |
510 | * Until we have made a decision allow this helper to take the @noperm | |
511 | * argument. We should hopefully be able to remove it soon. | |
512 | */ | |
31acceb9 CB |
513 | struct posix_acl *ovl_get_acl_path(const struct path *path, |
514 | const char *acl_name, bool noperm) | |
6c0a8bfb CB |
515 | { |
516 | struct posix_acl *real_acl, *clone; | |
517 | struct user_namespace *mnt_userns; | |
77435322 | 518 | struct mnt_idmap *idmap; |
6c0a8bfb CB |
519 | struct inode *realinode = d_inode(path->dentry); |
520 | ||
77435322 CB |
521 | idmap = mnt_idmap(path->mnt); |
522 | mnt_userns = mnt_idmap_owner(idmap); | |
6c0a8bfb CB |
523 | |
524 | if (noperm) | |
525 | real_acl = get_inode_acl(realinode, posix_acl_type(acl_name)); | |
526 | else | |
77435322 | 527 | real_acl = vfs_get_acl(idmap, path->dentry, acl_name); |
6c0a8bfb CB |
528 | if (IS_ERR_OR_NULL(real_acl)) |
529 | return real_acl; | |
530 | ||
531 | if (!is_idmapped_mnt(path->mnt)) | |
532 | return real_acl; | |
533 | ||
534 | /* | |
535 | * We cannot alter the ACLs returned from the relevant layer as that | |
536 | * would alter the cached values filesystem wide for the lower | |
537 | * filesystem. Instead we can clone the ACLs and then apply the | |
538 | * relevant idmapping of the layer. | |
539 | */ | |
540 | clone = posix_acl_clone(real_acl, GFP_KERNEL); | |
541 | posix_acl_release(real_acl); /* release original acl */ | |
542 | if (!clone) | |
543 | return ERR_PTR(-ENOMEM); | |
544 | ||
545 | ovl_idmap_posix_acl(realinode, mnt_userns, clone); | |
546 | return clone; | |
547 | } | |
548 | ||
1aa5fef5 CB |
549 | /* |
550 | * When the relevant layer is an idmapped mount we need to take the idmapping | |
551 | * of the layer into account and translate any ACL_{GROUP,USER} values | |
552 | * according to the idmapped mount. | |
553 | * | |
554 | * We cannot alter the ACLs returned from the relevant layer as that would | |
555 | * alter the cached values filesystem wide for the lower filesystem. Instead we | |
556 | * can clone the ACLs and then apply the relevant idmapping of the layer. | |
557 | * | |
558 | * This is obviously only relevant when idmapped layers are used. | |
559 | */ | |
77435322 | 560 | struct posix_acl *do_ovl_get_acl(struct mnt_idmap *idmap, |
6c0a8bfb CB |
561 | struct inode *inode, int type, |
562 | bool rcu, bool noperm) | |
39a25b2b | 563 | { |
09d8b586 | 564 | struct inode *realinode = ovl_inode_real(inode); |
6c0a8bfb | 565 | struct posix_acl *acl; |
1aa5fef5 | 566 | struct path realpath; |
39a25b2b | 567 | |
ded53656 | 568 | if (!IS_POSIXACL(realinode)) |
39a25b2b VG |
569 | return NULL; |
570 | ||
1aa5fef5 CB |
571 | /* Careful in RCU walk mode */ |
572 | ovl_i_path_real(inode, &realpath); | |
573 | if (!realpath.dentry) { | |
574 | WARN_ON(!rcu); | |
575 | return ERR_PTR(-ECHILD); | |
576 | } | |
332f606b | 577 | |
1aa5fef5 | 578 | if (rcu) { |
6c0a8bfb CB |
579 | /* |
580 | * If the layer is idmapped drop out of RCU path walk | |
581 | * so we can clone the ACLs. | |
582 | */ | |
583 | if (is_idmapped_mnt(realpath.mnt)) | |
584 | return ERR_PTR(-ECHILD); | |
585 | ||
1aa5fef5 CB |
586 | acl = get_cached_acl_rcu(realinode, type); |
587 | } else { | |
588 | const struct cred *old_cred; | |
589 | ||
590 | old_cred = ovl_override_creds(inode->i_sb); | |
6c0a8bfb | 591 | acl = ovl_get_acl_path(&realpath, posix_acl_xattr_name(type), noperm); |
1aa5fef5 CB |
592 | revert_creds(old_cred); |
593 | } | |
1175b6b8 | 594 | |
6c0a8bfb | 595 | return acl; |
39a25b2b | 596 | } |
0e641857 CB |
597 | |
598 | static int ovl_set_or_remove_acl(struct dentry *dentry, struct inode *inode, | |
599 | struct posix_acl *acl, int type) | |
600 | { | |
601 | int err; | |
602 | struct path realpath; | |
603 | const char *acl_name; | |
604 | const struct cred *old_cred; | |
605 | struct ovl_fs *ofs = OVL_FS(dentry->d_sb); | |
606 | struct dentry *upperdentry = ovl_dentry_upper(dentry); | |
607 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
608 | ||
609 | err = ovl_want_write(dentry); | |
610 | if (err) | |
611 | return err; | |
612 | ||
613 | /* | |
614 | * If ACL is to be removed from a lower file, check if it exists in | |
615 | * the first place before copying it up. | |
616 | */ | |
617 | acl_name = posix_acl_xattr_name(type); | |
618 | if (!acl && !upperdentry) { | |
619 | struct posix_acl *real_acl; | |
620 | ||
621 | ovl_path_lower(dentry, &realpath); | |
622 | old_cred = ovl_override_creds(dentry->d_sb); | |
77435322 | 623 | real_acl = vfs_get_acl(mnt_idmap(realpath.mnt), realdentry, |
0e641857 CB |
624 | acl_name); |
625 | revert_creds(old_cred); | |
0e641857 CB |
626 | if (IS_ERR(real_acl)) { |
627 | err = PTR_ERR(real_acl); | |
628 | goto out_drop_write; | |
629 | } | |
5b52aebe | 630 | posix_acl_release(real_acl); |
0e641857 CB |
631 | } |
632 | ||
633 | if (!upperdentry) { | |
634 | err = ovl_copy_up(dentry); | |
635 | if (err) | |
636 | goto out_drop_write; | |
637 | ||
638 | realdentry = ovl_dentry_upper(dentry); | |
639 | } | |
640 | ||
641 | old_cred = ovl_override_creds(dentry->d_sb); | |
642 | if (acl) | |
643 | err = ovl_do_set_acl(ofs, realdentry, acl_name, acl); | |
644 | else | |
645 | err = ovl_do_remove_acl(ofs, realdentry, acl_name); | |
646 | revert_creds(old_cred); | |
647 | ||
648 | /* copy c/mtime */ | |
649 | ovl_copyattr(inode); | |
650 | ||
651 | out_drop_write: | |
652 | ovl_drop_write(dentry); | |
653 | return err; | |
654 | } | |
655 | ||
13e83a49 | 656 | int ovl_set_acl(struct mnt_idmap *idmap, struct dentry *dentry, |
0e641857 CB |
657 | struct posix_acl *acl, int type) |
658 | { | |
659 | int err; | |
660 | struct inode *inode = d_inode(dentry); | |
661 | struct dentry *workdir = ovl_workdir(dentry); | |
662 | struct inode *realinode = ovl_inode_real(inode); | |
663 | ||
664 | if (!IS_POSIXACL(d_inode(workdir))) | |
665 | return -EOPNOTSUPP; | |
666 | if (!realinode->i_op->set_acl) | |
667 | return -EOPNOTSUPP; | |
668 | if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) | |
669 | return acl ? -EACCES : 0; | |
01beba79 | 670 | if (!inode_owner_or_capable(&nop_mnt_idmap, inode)) |
0e641857 CB |
671 | return -EPERM; |
672 | ||
673 | /* | |
674 | * Check if sgid bit needs to be cleared (actual setacl operation will | |
675 | * be done with mounter's capabilities and so that won't do it for us). | |
676 | */ | |
677 | if (unlikely(inode->i_mode & S_ISGID) && type == ACL_TYPE_ACCESS && | |
678 | !in_group_p(inode->i_gid) && | |
9452e93e | 679 | !capable_wrt_inode_uidgid(&nop_mnt_idmap, inode, CAP_FSETID)) { |
0e641857 CB |
680 | struct iattr iattr = { .ia_valid = ATTR_KILL_SGID }; |
681 | ||
c1632a0f | 682 | err = ovl_setattr(&nop_mnt_idmap, dentry, &iattr); |
0e641857 CB |
683 | if (err) |
684 | return err; | |
685 | } | |
686 | ||
687 | return ovl_set_or_remove_acl(dentry, inode, acl, type); | |
688 | } | |
ded53656 | 689 | #endif |
39a25b2b | 690 | |
95582b00 | 691 | int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags) |
d719e8f2 | 692 | { |
8f35cf51 MS |
693 | if (flags & S_ATIME) { |
694 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
695 | struct path upperpath = { | |
08f4c7c8 | 696 | .mnt = ovl_upper_mnt(ofs), |
8f35cf51 MS |
697 | .dentry = ovl_upperdentry_dereference(OVL_I(inode)), |
698 | }; | |
699 | ||
700 | if (upperpath.dentry) { | |
701 | touch_atime(&upperpath); | |
702 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
703 | } | |
d719e8f2 | 704 | } |
d719e8f2 MS |
705 | return 0; |
706 | } | |
707 | ||
9e142c41 MS |
708 | static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
709 | u64 start, u64 len) | |
710 | { | |
711 | int err; | |
c11faf32 | 712 | struct inode *realinode = ovl_inode_realdata(inode); |
9e142c41 MS |
713 | const struct cred *old_cred; |
714 | ||
715 | if (!realinode->i_op->fiemap) | |
716 | return -EOPNOTSUPP; | |
717 | ||
718 | old_cred = ovl_override_creds(inode->i_sb); | |
719 | err = realinode->i_op->fiemap(realinode, fieinfo, start, len); | |
720 | revert_creds(old_cred); | |
721 | ||
722 | return err; | |
723 | } | |
724 | ||
66dbfabf MS |
725 | /* |
726 | * Work around the fact that security_file_ioctl() takes a file argument. | |
727 | * Introducing security_inode_fileattr_get/set() hooks would solve this issue | |
728 | * properly. | |
729 | */ | |
2d343087 | 730 | static int ovl_security_fileattr(const struct path *realpath, struct fileattr *fa, |
66dbfabf MS |
731 | bool set) |
732 | { | |
66dbfabf MS |
733 | struct file *file; |
734 | unsigned int cmd; | |
735 | int err; | |
736 | ||
72db8211 | 737 | file = dentry_open(realpath, O_RDONLY, current_cred()); |
66dbfabf MS |
738 | if (IS_ERR(file)) |
739 | return PTR_ERR(file); | |
740 | ||
741 | if (set) | |
742 | cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS; | |
743 | else | |
744 | cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS; | |
745 | ||
746 | err = security_file_ioctl(file, cmd, 0); | |
747 | fput(file); | |
748 | ||
749 | return err; | |
750 | } | |
751 | ||
2d343087 | 752 | int ovl_real_fileattr_set(const struct path *realpath, struct fileattr *fa) |
72db8211 AG |
753 | { |
754 | int err; | |
755 | ||
756 | err = ovl_security_fileattr(realpath, fa, true); | |
757 | if (err) | |
758 | return err; | |
759 | ||
8782a9ae | 760 | return vfs_fileattr_set(mnt_idmap(realpath->mnt), realpath->dentry, fa); |
72db8211 AG |
761 | } |
762 | ||
8782a9ae | 763 | int ovl_fileattr_set(struct mnt_idmap *idmap, |
66dbfabf MS |
764 | struct dentry *dentry, struct fileattr *fa) |
765 | { | |
766 | struct inode *inode = d_inode(dentry); | |
72db8211 | 767 | struct path upperpath; |
66dbfabf | 768 | const struct cred *old_cred; |
096a218a | 769 | unsigned int flags; |
66dbfabf MS |
770 | int err; |
771 | ||
772 | err = ovl_want_write(dentry); | |
773 | if (err) | |
774 | goto out; | |
775 | ||
776 | err = ovl_copy_up(dentry); | |
777 | if (!err) { | |
72db8211 | 778 | ovl_path_real(dentry, &upperpath); |
66dbfabf MS |
779 | |
780 | old_cred = ovl_override_creds(inode->i_sb); | |
096a218a AG |
781 | /* |
782 | * Store immutable/append-only flags in xattr and clear them | |
783 | * in upper fileattr (in case they were set by older kernel) | |
784 | * so children of "ovl-immutable" directories lower aliases of | |
785 | * "ovl-immutable" hardlinks could be copied up. | |
786 | * Clear xattr when flags are cleared. | |
787 | */ | |
788 | err = ovl_set_protattr(inode, upperpath.dentry, fa); | |
789 | if (!err) | |
790 | err = ovl_real_fileattr_set(&upperpath, fa); | |
66dbfabf | 791 | revert_creds(old_cred); |
096a218a AG |
792 | |
793 | /* | |
794 | * Merge real inode flags with inode flags read from | |
795 | * overlay.protattr xattr | |
796 | */ | |
797 | flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK; | |
798 | ||
799 | BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK); | |
800 | flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK; | |
801 | inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK); | |
d8991e86 CX |
802 | |
803 | /* Update ctime */ | |
2878dffc | 804 | ovl_copyattr(inode); |
66dbfabf MS |
805 | } |
806 | ovl_drop_write(dentry); | |
807 | out: | |
808 | return err; | |
809 | } | |
810 | ||
096a218a AG |
811 | /* Convert inode protection flags to fileattr flags */ |
812 | static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa) | |
813 | { | |
814 | BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL); | |
815 | BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON); | |
816 | ||
817 | if (inode->i_flags & S_APPEND) { | |
818 | fa->flags |= FS_APPEND_FL; | |
819 | fa->fsx_xflags |= FS_XFLAG_APPEND; | |
820 | } | |
821 | if (inode->i_flags & S_IMMUTABLE) { | |
822 | fa->flags |= FS_IMMUTABLE_FL; | |
823 | fa->fsx_xflags |= FS_XFLAG_IMMUTABLE; | |
824 | } | |
825 | } | |
826 | ||
2d343087 | 827 | int ovl_real_fileattr_get(const struct path *realpath, struct fileattr *fa) |
72db8211 AG |
828 | { |
829 | int err; | |
830 | ||
831 | err = ovl_security_fileattr(realpath, fa, false); | |
832 | if (err) | |
833 | return err; | |
834 | ||
5b0a414d MS |
835 | err = vfs_fileattr_get(realpath->dentry, fa); |
836 | if (err == -ENOIOCTLCMD) | |
837 | err = -ENOTTY; | |
838 | return err; | |
72db8211 AG |
839 | } |
840 | ||
66dbfabf MS |
841 | int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa) |
842 | { | |
843 | struct inode *inode = d_inode(dentry); | |
72db8211 | 844 | struct path realpath; |
66dbfabf MS |
845 | const struct cred *old_cred; |
846 | int err; | |
847 | ||
72db8211 AG |
848 | ovl_path_real(dentry, &realpath); |
849 | ||
66dbfabf | 850 | old_cred = ovl_override_creds(inode->i_sb); |
72db8211 | 851 | err = ovl_real_fileattr_get(&realpath, fa); |
096a218a | 852 | ovl_fileattr_prot_flags(inode, fa); |
66dbfabf MS |
853 | revert_creds(old_cred); |
854 | ||
855 | return err; | |
856 | } | |
857 | ||
e9be9d5e MS |
858 | static const struct inode_operations ovl_file_inode_operations = { |
859 | .setattr = ovl_setattr, | |
860 | .permission = ovl_permission, | |
861 | .getattr = ovl_getattr, | |
e9be9d5e | 862 | .listxattr = ovl_listxattr, |
6c0a8bfb CB |
863 | .get_inode_acl = ovl_get_inode_acl, |
864 | .get_acl = ovl_get_acl, | |
0e641857 | 865 | .set_acl = ovl_set_acl, |
d719e8f2 | 866 | .update_time = ovl_update_time, |
9e142c41 | 867 | .fiemap = ovl_fiemap, |
66dbfabf MS |
868 | .fileattr_get = ovl_fileattr_get, |
869 | .fileattr_set = ovl_fileattr_set, | |
e9be9d5e MS |
870 | }; |
871 | ||
872 | static const struct inode_operations ovl_symlink_inode_operations = { | |
873 | .setattr = ovl_setattr, | |
6b255391 | 874 | .get_link = ovl_get_link, |
e9be9d5e | 875 | .getattr = ovl_getattr, |
e9be9d5e | 876 | .listxattr = ovl_listxattr, |
d719e8f2 | 877 | .update_time = ovl_update_time, |
e9be9d5e MS |
878 | }; |
879 | ||
9e142c41 MS |
880 | static const struct inode_operations ovl_special_inode_operations = { |
881 | .setattr = ovl_setattr, | |
882 | .permission = ovl_permission, | |
883 | .getattr = ovl_getattr, | |
884 | .listxattr = ovl_listxattr, | |
6c0a8bfb CB |
885 | .get_inode_acl = ovl_get_inode_acl, |
886 | .get_acl = ovl_get_acl, | |
0e641857 | 887 | .set_acl = ovl_set_acl, |
9e142c41 MS |
888 | .update_time = ovl_update_time, |
889 | }; | |
890 | ||
69383c59 | 891 | static const struct address_space_operations ovl_aops = { |
5b910bd6 AG |
892 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
893 | .direct_IO = noop_direct_IO, | |
894 | }; | |
895 | ||
b1eaa950 AG |
896 | /* |
897 | * It is possible to stack overlayfs instance on top of another | |
a5a84682 | 898 | * overlayfs instance as lower layer. We need to annotate the |
b1eaa950 AG |
899 | * stackable i_mutex locks according to stack level of the super |
900 | * block instance. An overlayfs instance can never be in stack | |
901 | * depth 0 (there is always a real fs below it). An overlayfs | |
f48bbfb2 | 902 | * inode lock will use the lockdep annotation ovl_i_mutex_key[depth]. |
b1eaa950 AG |
903 | * |
904 | * For example, here is a snip from /proc/lockdep_chains after | |
905 | * dir_iterate of nested overlayfs: | |
906 | * | |
907 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
908 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
909 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
b1f9d385 AG |
910 | * |
911 | * Locking order w.r.t ovl_want_write() is important for nested overlayfs. | |
912 | * | |
913 | * This chain is valid: | |
914 | * - inode->i_rwsem (inode_lock[2]) | |
915 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
916 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
917 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
918 | * | |
919 | * And this chain is valid: | |
920 | * - inode->i_rwsem (inode_lock[2]) | |
921 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
922 | * - lowerinode->i_rwsem (inode_lock[1]) | |
923 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
924 | * | |
925 | * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is | |
926 | * held, because it is in reverse order of the non-nested case using the same | |
927 | * upper fs: | |
928 | * - inode->i_rwsem (inode_lock[1]) | |
929 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
930 | * - OVL_I(inode)->lock (ovl_inode_lock[1]) | |
b1eaa950 AG |
931 | */ |
932 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
933 | ||
934 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
935 | { | |
936 | #ifdef CONFIG_LOCKDEP | |
937 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
938 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 939 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
940 | |
941 | int depth = inode->i_sb->s_stack_depth - 1; | |
942 | ||
943 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
944 | depth = 0; | |
945 | ||
946 | if (S_ISDIR(inode->i_mode)) | |
947 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
948 | else | |
949 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
950 | |
951 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
952 | #endif |
953 | } | |
954 | ||
4d314f78 AG |
955 | static void ovl_next_ino(struct inode *inode) |
956 | { | |
957 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
958 | ||
959 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
960 | if (unlikely(!inode->i_ino)) | |
961 | inode->i_ino = atomic_long_inc_return(&ofs->last_ino); | |
962 | } | |
963 | ||
62c832ed | 964 | static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid) |
e9be9d5e | 965 | { |
12574a9f | 966 | int xinobits = ovl_xino_bits(inode->i_sb); |
dfe51d47 | 967 | unsigned int xinoshift = 64 - xinobits; |
12574a9f | 968 | |
695b46e7 | 969 | /* |
6dde1e42 AG |
970 | * When d_ino is consistent with st_ino (samefs or i_ino has enough |
971 | * bits to encode layer), set the same value used for st_ino to i_ino, | |
972 | * so inode number exposed via /proc/locks and a like will be | |
973 | * consistent with d_ino and st_ino values. An i_ino value inconsistent | |
735c907d | 974 | * with d_ino also causes nfsd readdirplus to fail. |
695b46e7 | 975 | */ |
4d314f78 | 976 | inode->i_ino = ino; |
dfe51d47 AG |
977 | if (ovl_same_fs(inode->i_sb)) { |
978 | return; | |
979 | } else if (xinobits && likely(!(ino >> xinoshift))) { | |
980 | inode->i_ino |= (unsigned long)fsid << (xinoshift + 1); | |
981 | return; | |
982 | } | |
983 | ||
984 | /* | |
985 | * For directory inodes on non-samefs with xino disabled or xino | |
986 | * overflow, we allocate a non-persistent inode number, to be used for | |
987 | * resolving st_ino collisions in ovl_map_dev_ino(). | |
988 | * | |
989 | * To avoid ino collision with legitimate xino values from upper | |
990 | * layer (fsid 0), use the lowest xinobit to map the non | |
991 | * persistent inode numbers to the unified st_ino address space. | |
992 | */ | |
993 | if (S_ISDIR(inode->i_mode)) { | |
4d314f78 | 994 | ovl_next_ino(inode); |
dfe51d47 AG |
995 | if (xinobits) { |
996 | inode->i_ino &= ~0UL >> xinobits; | |
997 | inode->i_ino |= 1UL << xinoshift; | |
998 | } | |
12574a9f | 999 | } |
62c832ed AG |
1000 | } |
1001 | ||
1002 | void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip, | |
1003 | unsigned long ino, int fsid) | |
1004 | { | |
1005 | struct inode *realinode; | |
ffa5723c | 1006 | struct ovl_inode *oi = OVL_I(inode); |
62c832ed AG |
1007 | |
1008 | if (oip->upperdentry) | |
ffa5723c AG |
1009 | oi->__upperdentry = oip->upperdentry; |
1010 | if (oip->lowerpath && oip->lowerpath->dentry) { | |
1011 | oi->lowerpath.dentry = dget(oip->lowerpath->dentry); | |
1012 | oi->lowerpath.layer = oip->lowerpath->layer; | |
1013 | } | |
62c832ed | 1014 | if (oip->lowerdata) |
ffa5723c | 1015 | oi->lowerdata = igrab(d_inode(oip->lowerdata)); |
62c832ed AG |
1016 | |
1017 | realinode = ovl_inode_real(inode); | |
2878dffc | 1018 | ovl_copyattr(inode); |
62c832ed AG |
1019 | ovl_copyflags(realinode, inode); |
1020 | ovl_map_ino(inode, ino, fsid); | |
1021 | } | |
1022 | ||
1023 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1024 | { | |
e9be9d5e | 1025 | inode->i_mode = mode; |
d719e8f2 | 1026 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
1027 | #ifdef CONFIG_FS_POSIX_ACL |
1028 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
1029 | #endif | |
e9be9d5e | 1030 | |
b1eaa950 AG |
1031 | ovl_lockdep_annotate_inode_mutex_key(inode); |
1032 | ||
ca4c8a3a MS |
1033 | switch (mode & S_IFMT) { |
1034 | case S_IFREG: | |
1035 | inode->i_op = &ovl_file_inode_operations; | |
d1d04ef8 | 1036 | inode->i_fop = &ovl_file_operations; |
5b910bd6 | 1037 | inode->i_mapping->a_ops = &ovl_aops; |
ca4c8a3a MS |
1038 | break; |
1039 | ||
e9be9d5e | 1040 | case S_IFDIR: |
e9be9d5e MS |
1041 | inode->i_op = &ovl_dir_inode_operations; |
1042 | inode->i_fop = &ovl_dir_operations; | |
1043 | break; | |
1044 | ||
1045 | case S_IFLNK: | |
1046 | inode->i_op = &ovl_symlink_inode_operations; | |
1047 | break; | |
1048 | ||
51f7e52d | 1049 | default: |
9e142c41 | 1050 | inode->i_op = &ovl_special_inode_operations; |
ca4c8a3a | 1051 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 1052 | break; |
51f7e52d MS |
1053 | } |
1054 | } | |
e9be9d5e | 1055 | |
5f8415d6 AG |
1056 | /* |
1057 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
1058 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
1059 | * upper inode, the following nlink modifying operations can happen: | |
1060 | * | |
1061 | * 1. Lower hardlink copy up | |
1062 | * 2. Upper hardlink created, unlinked or renamed over | |
1063 | * 3. Lower hardlink whiteout or renamed over | |
1064 | * | |
1065 | * For the first, copy up case, the union nlink does not change, whether the | |
1066 | * operation succeeds or fails, but the upper inode nlink may change. | |
1067 | * Therefore, before copy up, we store the union nlink value relative to the | |
2d2f2d73 | 1068 | * lower inode nlink in the index inode xattr .overlay.nlink. |
5f8415d6 AG |
1069 | * |
1070 | * For the second, upper hardlink case, the union nlink should be incremented | |
1071 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
1072 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
1073 | * value relative to the upper inode nlink in the index inode. | |
1074 | * | |
1075 | * For the last, lower cover up case, we simplify things by preceding the | |
1076 | * whiteout or cover up with copy up. This makes sure that there is an index | |
1077 | * upper inode where the nlink xattr can be stored before the copied up upper | |
1078 | * entry is unlink. | |
1079 | */ | |
1080 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
1081 | ||
1082 | /* | |
1083 | * On-disk format for indexed nlink: | |
1084 | * | |
1085 | * nlink relative to the upper inode - "U[+-]NUM" | |
1086 | * nlink relative to the lower inode - "L[+-]NUM" | |
1087 | */ | |
1088 | ||
1089 | static int ovl_set_nlink_common(struct dentry *dentry, | |
1090 | struct dentry *realdentry, const char *format) | |
1091 | { | |
1092 | struct inode *inode = d_inode(dentry); | |
1093 | struct inode *realinode = d_inode(realdentry); | |
1094 | char buf[13]; | |
1095 | int len; | |
1096 | ||
1097 | len = snprintf(buf, sizeof(buf), format, | |
1098 | (int) (inode->i_nlink - realinode->i_nlink)); | |
1099 | ||
6787341a MS |
1100 | if (WARN_ON(len >= sizeof(buf))) |
1101 | return -EIO; | |
1102 | ||
c914c0e2 AG |
1103 | return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry), |
1104 | OVL_XATTR_NLINK, buf, len); | |
5f8415d6 AG |
1105 | } |
1106 | ||
1107 | int ovl_set_nlink_upper(struct dentry *dentry) | |
1108 | { | |
1109 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
1110 | } | |
1111 | ||
1112 | int ovl_set_nlink_lower(struct dentry *dentry) | |
1113 | { | |
1114 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
1115 | } | |
1116 | ||
610afc0b | 1117 | unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry, |
caf70cb2 AG |
1118 | struct dentry *upperdentry, |
1119 | unsigned int fallback) | |
5f8415d6 AG |
1120 | { |
1121 | int nlink_diff; | |
1122 | int nlink; | |
1123 | char buf[13]; | |
1124 | int err; | |
1125 | ||
1126 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
1127 | return fallback; | |
1128 | ||
dad7017a CB |
1129 | err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK, |
1130 | &buf, sizeof(buf) - 1); | |
5f8415d6 AG |
1131 | if (err < 0) |
1132 | goto fail; | |
1133 | ||
1134 | buf[err] = '\0'; | |
1135 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
1136 | (buf[1] != '+' && buf[1] != '-')) | |
1137 | goto fail; | |
1138 | ||
1139 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
1140 | if (err < 0) | |
1141 | goto fail; | |
1142 | ||
1143 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
1144 | nlink += nlink_diff; | |
1145 | ||
1146 | if (nlink <= 0) | |
1147 | goto fail; | |
1148 | ||
1149 | return nlink; | |
1150 | ||
1151 | fail: | |
1bd0a3ae | 1152 | pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n", |
5f8415d6 AG |
1153 | upperdentry, err); |
1154 | return fallback; | |
1155 | } | |
1156 | ||
ca4c8a3a | 1157 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
1158 | { |
1159 | struct inode *inode; | |
1160 | ||
1161 | inode = new_inode(sb); | |
1162 | if (inode) | |
62c832ed | 1163 | ovl_fill_inode(inode, mode, rdev); |
51f7e52d MS |
1164 | |
1165 | return inode; | |
1166 | } | |
1167 | ||
1168 | static int ovl_inode_test(struct inode *inode, void *data) | |
1169 | { | |
25b7713a | 1170 | return inode->i_private == data; |
51f7e52d MS |
1171 | } |
1172 | ||
1173 | static int ovl_inode_set(struct inode *inode, void *data) | |
1174 | { | |
25b7713a | 1175 | inode->i_private = data; |
51f7e52d MS |
1176 | return 0; |
1177 | } | |
1178 | ||
b9ac5c27 | 1179 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
4b91c30a | 1180 | struct dentry *upperdentry, bool strict) |
b9ac5c27 | 1181 | { |
4b91c30a AG |
1182 | /* |
1183 | * For directories, @strict verify from lookup path performs consistency | |
1184 | * checks, so NULL lower/upper in dentry must match NULL lower/upper in | |
1185 | * inode. Non @strict verify from NFS handle decode path passes NULL for | |
1186 | * 'unknown' lower/upper. | |
1187 | */ | |
1188 | if (S_ISDIR(inode->i_mode) && strict) { | |
31747eda AG |
1189 | /* Real lower dir moved to upper layer under us? */ |
1190 | if (!lowerdentry && ovl_inode_lower(inode)) | |
1191 | return false; | |
1192 | ||
1193 | /* Lookup of an uncovered redirect origin? */ | |
1194 | if (!upperdentry && ovl_inode_upper(inode)) | |
1195 | return false; | |
1196 | } | |
1197 | ||
939ae4ef AG |
1198 | /* |
1199 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
1200 | * This happens when finding a copied up overlay inode for a renamed | |
1201 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
1202 | * by origin because lower fs does not support file handles. | |
1203 | */ | |
1204 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
1205 | return false; |
1206 | ||
1207 | /* | |
1208 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
1209 | * This happens when finding a lower alias for a copied up hard link. | |
1210 | */ | |
1211 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
1212 | return false; | |
1213 | ||
1214 | return true; | |
1215 | } | |
1216 | ||
4b91c30a AG |
1217 | struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real, |
1218 | bool is_upper) | |
9436a1a3 | 1219 | { |
4b91c30a | 1220 | struct inode *inode, *key = d_inode(real); |
9436a1a3 AG |
1221 | |
1222 | inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
1223 | if (!inode) | |
1224 | return NULL; | |
1225 | ||
4b91c30a AG |
1226 | if (!ovl_verify_inode(inode, is_upper ? NULL : real, |
1227 | is_upper ? real : NULL, false)) { | |
9436a1a3 AG |
1228 | iput(inode); |
1229 | return ERR_PTR(-ESTALE); | |
1230 | } | |
1231 | ||
1232 | return inode; | |
1233 | } | |
1234 | ||
146d62e5 AG |
1235 | bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir) |
1236 | { | |
1237 | struct inode *key = d_inode(dir); | |
1238 | struct inode *trap; | |
1239 | bool res; | |
1240 | ||
1241 | trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
1242 | if (!trap) | |
1243 | return false; | |
1244 | ||
1245 | res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) && | |
1246 | !ovl_inode_lower(trap); | |
1247 | ||
1248 | iput(trap); | |
1249 | return res; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * Create an inode cache entry for layer root dir, that will intentionally | |
1254 | * fail ovl_verify_inode(), so any lookup that will find some layer root | |
1255 | * will fail. | |
1256 | */ | |
1257 | struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir) | |
1258 | { | |
1259 | struct inode *key = d_inode(dir); | |
1260 | struct inode *trap; | |
1261 | ||
1262 | if (!d_is_dir(dir)) | |
1263 | return ERR_PTR(-ENOTDIR); | |
1264 | ||
1265 | trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test, | |
1266 | ovl_inode_set, key); | |
1267 | if (!trap) | |
1268 | return ERR_PTR(-ENOMEM); | |
1269 | ||
1270 | if (!(trap->i_state & I_NEW)) { | |
1271 | /* Conflicting layer roots? */ | |
1272 | iput(trap); | |
1273 | return ERR_PTR(-ELOOP); | |
1274 | } | |
1275 | ||
1276 | trap->i_mode = S_IFDIR; | |
1277 | trap->i_flags = S_DEAD; | |
1278 | unlock_new_inode(trap); | |
1279 | ||
1280 | return trap; | |
1281 | } | |
1282 | ||
764baba8 AG |
1283 | /* |
1284 | * Does overlay inode need to be hashed by lower inode? | |
1285 | */ | |
1286 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
74c6e384 | 1287 | struct dentry *lower, bool index) |
764baba8 AG |
1288 | { |
1289 | struct ovl_fs *ofs = sb->s_fs_info; | |
1290 | ||
1291 | /* No, if pure upper */ | |
1292 | if (!lower) | |
1293 | return false; | |
1294 | ||
1295 | /* Yes, if already indexed */ | |
1296 | if (index) | |
1297 | return true; | |
1298 | ||
1299 | /* Yes, if won't be copied up */ | |
08f4c7c8 | 1300 | if (!ovl_upper_mnt(ofs)) |
764baba8 AG |
1301 | return true; |
1302 | ||
1303 | /* No, if lower hardlink is or will be broken on copy up */ | |
1304 | if ((upper || !ovl_indexdir(sb)) && | |
1305 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
1306 | return false; | |
1307 | ||
1308 | /* No, if non-indexed upper with NFS export */ | |
1309 | if (sb->s_export_op && upper) | |
1310 | return false; | |
1311 | ||
1312 | /* Otherwise, hash by lower inode for fsnotify */ | |
1313 | return true; | |
1314 | } | |
1315 | ||
01b39dcc AG |
1316 | static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode, |
1317 | struct inode *key) | |
1318 | { | |
1319 | return newinode ? inode_insert5(newinode, (unsigned long) key, | |
1320 | ovl_inode_test, ovl_inode_set, key) : | |
1321 | iget5_locked(sb, (unsigned long) key, | |
1322 | ovl_inode_test, ovl_inode_set, key); | |
1323 | } | |
1324 | ||
ac6a52eb VG |
1325 | struct inode *ovl_get_inode(struct super_block *sb, |
1326 | struct ovl_inode_params *oip) | |
51f7e52d | 1327 | { |
610afc0b | 1328 | struct ovl_fs *ofs = OVL_FS(sb); |
ac6a52eb VG |
1329 | struct dentry *upperdentry = oip->upperdentry; |
1330 | struct ovl_path *lowerpath = oip->lowerpath; | |
09d8b586 | 1331 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; |
51f7e52d | 1332 | struct inode *inode; |
12574a9f | 1333 | struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL; |
dad7017a CB |
1334 | struct path realpath = { |
1335 | .dentry = upperdentry ?: lowerdentry, | |
1336 | .mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt, | |
1337 | }; | |
ac6a52eb VG |
1338 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, |
1339 | oip->index); | |
300b124f | 1340 | int fsid = bylower ? lowerpath->layer->fsid : 0; |
28166ab3 | 1341 | bool is_dir; |
695b46e7 | 1342 | unsigned long ino = 0; |
acf3062a | 1343 | int err = oip->newinode ? -EEXIST : -ENOMEM; |
6eaf0111 | 1344 | |
09d8b586 MS |
1345 | if (!realinode) |
1346 | realinode = d_inode(lowerdentry); | |
1347 | ||
6eaf0111 | 1348 | /* |
764baba8 AG |
1349 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
1350 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 1351 | */ |
31747eda | 1352 | is_dir = S_ISDIR(realinode->i_mode); |
764baba8 AG |
1353 | if (upperdentry || bylower) { |
1354 | struct inode *key = d_inode(bylower ? lowerdentry : | |
1355 | upperdentry); | |
31747eda | 1356 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 1357 | |
01b39dcc | 1358 | inode = ovl_iget5(sb, oip->newinode, key); |
09d8b586 | 1359 | if (!inode) |
027065b7 | 1360 | goto out_err; |
09d8b586 | 1361 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
1362 | /* |
1363 | * Verify that the underlying files stored in the inode | |
1364 | * match those in the dentry. | |
1365 | */ | |
4b91c30a AG |
1366 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry, |
1367 | true)) { | |
b9ac5c27 | 1368 | iput(inode); |
027065b7 VG |
1369 | err = -ESTALE; |
1370 | goto out_err; | |
b9ac5c27 MS |
1371 | } |
1372 | ||
09d8b586 | 1373 | dput(upperdentry); |
9cec54c8 | 1374 | kfree(oip->redirect); |
09d8b586 MS |
1375 | goto out; |
1376 | } | |
e6d2ebdd | 1377 | |
31747eda AG |
1378 | /* Recalculate nlink for non-dir due to indexing */ |
1379 | if (!is_dir) | |
610afc0b MS |
1380 | nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry, |
1381 | nlink); | |
5f8415d6 | 1382 | set_nlink(inode, nlink); |
695b46e7 | 1383 | ino = key->i_ino; |
e6d2ebdd | 1384 | } else { |
764baba8 | 1385 | /* Lower hardlink that will be broken on copy up */ |
0aceb53e | 1386 | inode = new_inode(sb); |
027065b7 VG |
1387 | if (!inode) { |
1388 | err = -ENOMEM; | |
1389 | goto out_err; | |
1390 | } | |
300b124f AG |
1391 | ino = realinode->i_ino; |
1392 | fsid = lowerpath->layer->fsid; | |
e9be9d5e | 1393 | } |
62c832ed AG |
1394 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev); |
1395 | ovl_inode_init(inode, oip, ino, fsid); | |
13c72075 | 1396 | |
610afc0b | 1397 | if (upperdentry && ovl_is_impuredir(sb, upperdentry)) |
13c72075 MS |
1398 | ovl_set_flag(OVL_IMPURE, inode); |
1399 | ||
ac6a52eb | 1400 | if (oip->index) |
0471a9cd VG |
1401 | ovl_set_flag(OVL_INDEX, inode); |
1402 | ||
9cec54c8 VG |
1403 | OVL_I(inode)->redirect = oip->redirect; |
1404 | ||
a00c2d59 VG |
1405 | if (bylower) |
1406 | ovl_set_flag(OVL_CONST_INO, inode); | |
1407 | ||
b79e05aa | 1408 | /* Check for non-merge dir that may have whiteouts */ |
31747eda | 1409 | if (is_dir) { |
ac6a52eb | 1410 | if (((upperdentry && lowerdentry) || oip->numlower > 1) || |
dad7017a | 1411 | ovl_path_check_origin_xattr(ofs, &realpath)) { |
b79e05aa AG |
1412 | ovl_set_flag(OVL_WHITEOUTS, inode); |
1413 | } | |
1414 | } | |
1415 | ||
096a218a AG |
1416 | /* Check for immutable/append-only inode flags in xattr */ |
1417 | if (upperdentry) | |
1418 | ovl_check_protattr(inode, upperdentry); | |
1419 | ||
e6d2ebdd MS |
1420 | if (inode->i_state & I_NEW) |
1421 | unlock_new_inode(inode); | |
1422 | out: | |
e9be9d5e | 1423 | return inode; |
b9ac5c27 | 1424 | |
027065b7 | 1425 | out_err: |
1bd0a3ae | 1426 | pr_warn_ratelimited("failed to get inode (%i)\n", err); |
027065b7 | 1427 | inode = ERR_PTR(err); |
b9ac5c27 | 1428 | goto out; |
e9be9d5e | 1429 | } |