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> |
e9be9d5e MS |
13 | #include "overlayfs.h" |
14 | ||
ba1e563c | 15 | |
e9be9d5e MS |
16 | int ovl_setattr(struct dentry *dentry, struct iattr *attr) |
17 | { | |
18 | int err; | |
997336f2 | 19 | bool full_copy_up = false; |
e9be9d5e | 20 | struct dentry *upperdentry; |
1175b6b8 | 21 | const struct cred *old_cred; |
e9be9d5e | 22 | |
31051c85 | 23 | err = setattr_prepare(dentry, attr); |
cf9a6784 MS |
24 | if (err) |
25 | return err; | |
26 | ||
e9be9d5e MS |
27 | err = ovl_want_write(dentry); |
28 | if (err) | |
29 | goto out; | |
30 | ||
5812160e MS |
31 | if (attr->ia_valid & ATTR_SIZE) { |
32 | struct inode *realinode = d_inode(ovl_dentry_real(dentry)); | |
33 | ||
34 | err = -ETXTBSY; | |
35 | if (atomic_read(&realinode->i_writecount) < 0) | |
36 | goto out_drop_write; | |
997336f2 VG |
37 | |
38 | /* Truncate should trigger data copy up as well */ | |
39 | full_copy_up = true; | |
5812160e MS |
40 | } |
41 | ||
997336f2 VG |
42 | if (!full_copy_up) |
43 | err = ovl_copy_up(dentry); | |
44 | else | |
45 | err = ovl_copy_up_with_data(dentry); | |
acff81ec | 46 | if (!err) { |
5812160e MS |
47 | struct inode *winode = NULL; |
48 | ||
acff81ec MS |
49 | upperdentry = ovl_dentry_upper(dentry); |
50 | ||
5812160e MS |
51 | if (attr->ia_valid & ATTR_SIZE) { |
52 | winode = d_inode(upperdentry); | |
53 | err = get_write_access(winode); | |
54 | if (err) | |
55 | goto out_drop_write; | |
56 | } | |
57 | ||
b99c2d91 MS |
58 | if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID)) |
59 | attr->ia_valid &= ~ATTR_MODE; | |
60 | ||
5955102c | 61 | inode_lock(upperdentry->d_inode); |
1175b6b8 | 62 | old_cred = ovl_override_creds(dentry->d_sb); |
e9be9d5e | 63 | err = notify_change(upperdentry, attr, NULL); |
1175b6b8 | 64 | revert_creds(old_cred); |
b81de061 KK |
65 | if (!err) |
66 | ovl_copyattr(upperdentry->d_inode, dentry->d_inode); | |
5955102c | 67 | inode_unlock(upperdentry->d_inode); |
5812160e MS |
68 | |
69 | if (winode) | |
70 | put_write_access(winode); | |
e9be9d5e | 71 | } |
5812160e | 72 | out_drop_write: |
e9be9d5e MS |
73 | ovl_drop_write(dentry); |
74 | out: | |
75 | return err; | |
76 | } | |
77 | ||
07f1e596 | 78 | static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid) |
da309e8c | 79 | { |
0f831ec8 | 80 | bool samefs = ovl_same_fs(dentry->d_sb); |
e487d889 | 81 | unsigned int xinobits = ovl_xino_bits(dentry->d_sb); |
da309e8c AG |
82 | |
83 | if (samefs) { | |
84 | /* | |
85 | * When all layers are on the same fs, all real inode | |
86 | * number are unique, so we use the overlay st_dev, | |
87 | * which is friendly to du -x. | |
88 | */ | |
89 | stat->dev = dentry->d_sb->s_dev; | |
e487d889 AG |
90 | return 0; |
91 | } else if (xinobits) { | |
92 | unsigned int shift = 64 - xinobits; | |
93 | /* | |
94 | * All inode numbers of underlying fs should not be using the | |
95 | * high xinobits, so we use high xinobits to partition the | |
96 | * overlay st_ino address space. The high bits holds the fsid | |
97 | * (upper fsid is 0). This way overlay inode numbers are unique | |
98 | * and all inodes use overlay st_dev. Inode numbers are also | |
99 | * persistent for a given layer configuration. | |
100 | */ | |
101 | if (stat->ino >> shift) { | |
1bd0a3ae | 102 | pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n", |
e487d889 AG |
103 | dentry, stat->ino, xinobits); |
104 | } else { | |
07f1e596 | 105 | stat->ino |= ((u64)fsid) << shift; |
e487d889 AG |
106 | stat->dev = dentry->d_sb->s_dev; |
107 | return 0; | |
108 | } | |
109 | } | |
110 | ||
111 | /* The inode could not be mapped to a unified st_ino address space */ | |
112 | if (S_ISDIR(dentry->d_inode->i_mode)) { | |
da309e8c AG |
113 | /* |
114 | * Always use the overlay st_dev for directories, so 'find | |
115 | * -xdev' will scan the entire overlay mount and won't cross the | |
116 | * overlay mount boundaries. | |
117 | * | |
118 | * If not all layers are on the same fs the pair {real st_ino; | |
119 | * overlay st_dev} is not unique, so use the non persistent | |
120 | * overlay st_ino for directories. | |
121 | */ | |
122 | stat->dev = dentry->d_sb->s_dev; | |
123 | stat->ino = dentry->d_inode->i_ino; | |
b7bf9908 | 124 | } else { |
da309e8c AG |
125 | /* |
126 | * For non-samefs setup, if we cannot map all layers st_ino | |
127 | * to a unified address space, we need to make sure that st_dev | |
b7bf9908 AG |
128 | * is unique per underlying fs, so we use the unique anonymous |
129 | * bdev assigned to the underlying fs. | |
da309e8c | 130 | */ |
07f1e596 | 131 | stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev; |
da309e8c AG |
132 | } |
133 | ||
134 | return 0; | |
135 | } | |
136 | ||
5b712091 MS |
137 | int ovl_getattr(const struct path *path, struct kstat *stat, |
138 | u32 request_mask, unsigned int flags) | |
e9be9d5e | 139 | { |
a528d35e | 140 | struct dentry *dentry = path->dentry; |
72b608f0 | 141 | enum ovl_path_type type; |
e9be9d5e | 142 | struct path realpath; |
1175b6b8 | 143 | const struct cred *old_cred; |
5b712091 | 144 | bool is_dir = S_ISDIR(dentry->d_inode->i_mode); |
07f1e596 | 145 | int fsid = 0; |
1175b6b8 | 146 | int err; |
67d756c2 VG |
147 | bool metacopy_blocks = false; |
148 | ||
149 | metacopy_blocks = ovl_is_metacopy_dentry(dentry); | |
e9be9d5e | 150 | |
72b608f0 | 151 | type = ovl_path_real(dentry, &realpath); |
1175b6b8 | 152 | old_cred = ovl_override_creds(dentry->d_sb); |
a528d35e | 153 | err = vfs_getattr(&realpath, stat, request_mask, flags); |
72b608f0 AG |
154 | if (err) |
155 | goto out; | |
156 | ||
157 | /* | |
da309e8c AG |
158 | * For non-dir or same fs, we use st_ino of the copy up origin. |
159 | * This guaranties constant st_dev/st_ino across copy up. | |
e487d889 AG |
160 | * With xino feature and non-samefs, we use st_ino of the copy up |
161 | * origin masked with high bits that represent the layer id. | |
72b608f0 | 162 | * |
da309e8c | 163 | * If lower filesystem supports NFS file handles, this also guaranties |
72b608f0 AG |
164 | * persistent st_ino across mount cycle. |
165 | */ | |
0f831ec8 | 166 | if (!is_dir || ovl_same_dev(dentry->d_sb)) { |
da309e8c | 167 | if (!OVL_TYPE_UPPER(type)) { |
07f1e596 | 168 | fsid = ovl_layer_lower(dentry)->fsid; |
da309e8c | 169 | } else if (OVL_TYPE_ORIGIN(type)) { |
72b608f0 | 170 | struct kstat lowerstat; |
67d756c2 VG |
171 | u32 lowermask = STATX_INO | STATX_BLOCKS | |
172 | (!is_dir ? STATX_NLINK : 0); | |
72b608f0 AG |
173 | |
174 | ovl_path_lower(dentry, &realpath); | |
175 | err = vfs_getattr(&realpath, &lowerstat, | |
5b712091 | 176 | lowermask, flags); |
72b608f0 AG |
177 | if (err) |
178 | goto out; | |
179 | ||
72b608f0 | 180 | /* |
359f392c | 181 | * Lower hardlinks may be broken on copy up to different |
72b608f0 AG |
182 | * upper files, so we cannot use the lower origin st_ino |
183 | * for those different files, even for the same fs case. | |
86eaa130 AG |
184 | * |
185 | * Similarly, several redirected dirs can point to the | |
186 | * same dir on a lower layer. With the "verify_lower" | |
187 | * feature, we do not use the lower origin st_ino, if | |
188 | * we haven't verified that this redirect is unique. | |
189 | * | |
359f392c | 190 | * With inodes index enabled, it is safe to use st_ino |
86eaa130 AG |
191 | * of an indexed origin. The index validates that the |
192 | * upper hardlink is not broken and that a redirected | |
193 | * dir is the only redirect to that origin. | |
72b608f0 | 194 | */ |
86eaa130 AG |
195 | if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) || |
196 | (!ovl_verify_lower(dentry->d_sb) && | |
9f99e50d | 197 | (is_dir || lowerstat.nlink == 1))) { |
07f1e596 | 198 | fsid = ovl_layer_lower(dentry)->fsid; |
b7bf9908 | 199 | stat->ino = lowerstat.ino; |
9f99e50d | 200 | } |
67d756c2 VG |
201 | |
202 | /* | |
203 | * If we are querying a metacopy dentry and lower | |
204 | * dentry is data dentry, then use the blocks we | |
205 | * queried just now. We don't have to do additional | |
206 | * vfs_getattr(). If lower itself is metacopy, then | |
207 | * additional vfs_getattr() is unavoidable. | |
208 | */ | |
209 | if (metacopy_blocks && | |
210 | realpath.dentry == ovl_dentry_lowerdata(dentry)) { | |
211 | stat->blocks = lowerstat.blocks; | |
212 | metacopy_blocks = false; | |
213 | } | |
214 | } | |
215 | ||
216 | if (metacopy_blocks) { | |
217 | /* | |
218 | * If lower is not same as lowerdata or if there was | |
219 | * no origin on upper, we can end up here. | |
220 | */ | |
221 | struct kstat lowerdatastat; | |
222 | u32 lowermask = STATX_BLOCKS; | |
223 | ||
224 | ovl_path_lowerdata(dentry, &realpath); | |
225 | err = vfs_getattr(&realpath, &lowerdatastat, | |
226 | lowermask, flags); | |
227 | if (err) | |
228 | goto out; | |
229 | stat->blocks = lowerdatastat.blocks; | |
72b608f0 | 230 | } |
72b608f0 | 231 | } |
5b712091 | 232 | |
07f1e596 | 233 | err = ovl_map_dev_ino(dentry, stat, fsid); |
da309e8c AG |
234 | if (err) |
235 | goto out; | |
236 | ||
5b712091 MS |
237 | /* |
238 | * It's probably not worth it to count subdirs to get the | |
239 | * correct link count. nlink=1 seems to pacify 'find' and | |
240 | * other utilities. | |
241 | */ | |
242 | if (is_dir && OVL_TYPE_MERGE(type)) | |
243 | stat->nlink = 1; | |
244 | ||
5f8415d6 AG |
245 | /* |
246 | * Return the overlay inode nlinks for indexed upper inodes. | |
247 | * Overlay inode nlink counts the union of the upper hardlinks | |
248 | * and non-covered lower hardlinks. It does not include the upper | |
249 | * index hardlink. | |
250 | */ | |
251 | if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
252 | stat->nlink = dentry->d_inode->i_nlink; | |
253 | ||
72b608f0 | 254 | out: |
1175b6b8 | 255 | revert_creds(old_cred); |
72b608f0 | 256 | |
1175b6b8 | 257 | return err; |
e9be9d5e MS |
258 | } |
259 | ||
260 | int ovl_permission(struct inode *inode, int mask) | |
261 | { | |
09d8b586 MS |
262 | struct inode *upperinode = ovl_inode_upper(inode); |
263 | struct inode *realinode = upperinode ?: ovl_inode_lower(inode); | |
c0ca3d70 | 264 | const struct cred *old_cred; |
e9be9d5e MS |
265 | int err; |
266 | ||
e9be9d5e | 267 | /* Careful in RCU walk mode */ |
e9be9d5e MS |
268 | if (!realinode) { |
269 | WARN_ON(!(mask & MAY_NOT_BLOCK)); | |
a999d7e1 | 270 | return -ECHILD; |
e9be9d5e MS |
271 | } |
272 | ||
c0ca3d70 VG |
273 | /* |
274 | * Check overlay inode with the creds of task and underlying inode | |
275 | * with creds of mounter | |
276 | */ | |
277 | err = generic_permission(inode, mask); | |
278 | if (err) | |
279 | return err; | |
280 | ||
ec7ba118 MS |
281 | old_cred = ovl_override_creds(inode->i_sb); |
282 | if (!upperinode && | |
283 | !special_file(realinode->i_mode) && mask & MAY_WRITE) { | |
754f8cb7 | 284 | mask &= ~(MAY_WRITE | MAY_APPEND); |
ec7ba118 | 285 | /* Make sure mounter can read file for copy up later */ |
500cac3c VG |
286 | mask |= MAY_READ; |
287 | } | |
9c630ebe | 288 | err = inode_permission(realinode, mask); |
c0ca3d70 VG |
289 | revert_creds(old_cred); |
290 | ||
291 | return err; | |
e9be9d5e MS |
292 | } |
293 | ||
6b255391 | 294 | static const char *ovl_get_link(struct dentry *dentry, |
fceef393 AV |
295 | struct inode *inode, |
296 | struct delayed_call *done) | |
e9be9d5e | 297 | { |
1175b6b8 VG |
298 | const struct cred *old_cred; |
299 | const char *p; | |
e9be9d5e | 300 | |
6b255391 AV |
301 | if (!dentry) |
302 | return ERR_PTR(-ECHILD); | |
303 | ||
1175b6b8 | 304 | old_cred = ovl_override_creds(dentry->d_sb); |
7764235b | 305 | p = vfs_get_link(ovl_dentry_real(dentry), done); |
1175b6b8 VG |
306 | revert_creds(old_cred); |
307 | return p; | |
e9be9d5e MS |
308 | } |
309 | ||
0956254a | 310 | bool ovl_is_private_xattr(const char *name) |
e9be9d5e | 311 | { |
fe2b7595 AG |
312 | return strncmp(name, OVL_XATTR_PREFIX, |
313 | sizeof(OVL_XATTR_PREFIX) - 1) == 0; | |
e9be9d5e MS |
314 | } |
315 | ||
1d88f183 MS |
316 | int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name, |
317 | const void *value, size_t size, int flags) | |
e9be9d5e MS |
318 | { |
319 | int err; | |
1d88f183 MS |
320 | struct dentry *upperdentry = ovl_i_dentry_upper(inode); |
321 | struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry); | |
1175b6b8 | 322 | const struct cred *old_cred; |
e9be9d5e MS |
323 | |
324 | err = ovl_want_write(dentry); | |
325 | if (err) | |
326 | goto out; | |
327 | ||
1d88f183 MS |
328 | if (!value && !upperdentry) { |
329 | err = vfs_getxattr(realdentry, name, NULL, 0); | |
0e585ccc AG |
330 | if (err < 0) |
331 | goto out_drop_write; | |
332 | } | |
333 | ||
1d88f183 MS |
334 | if (!upperdentry) { |
335 | err = ovl_copy_up(dentry); | |
336 | if (err) | |
337 | goto out_drop_write; | |
e9be9d5e | 338 | |
1d88f183 MS |
339 | realdentry = ovl_dentry_upper(dentry); |
340 | } | |
0e585ccc | 341 | |
1175b6b8 | 342 | old_cred = ovl_override_creds(dentry->d_sb); |
0e585ccc | 343 | if (value) |
1d88f183 | 344 | err = vfs_setxattr(realdentry, name, value, size, flags); |
0e585ccc AG |
345 | else { |
346 | WARN_ON(flags != XATTR_REPLACE); | |
1d88f183 | 347 | err = vfs_removexattr(realdentry, name); |
0e585ccc | 348 | } |
1175b6b8 | 349 | revert_creds(old_cred); |
e9be9d5e | 350 | |
d9854c87 MS |
351 | /* copy c/mtime */ |
352 | ovl_copyattr(d_inode(realdentry), inode); | |
353 | ||
e9be9d5e MS |
354 | out_drop_write: |
355 | ovl_drop_write(dentry); | |
356 | out: | |
357 | return err; | |
358 | } | |
359 | ||
1d88f183 | 360 | int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name, |
0eb45fc3 | 361 | void *value, size_t size) |
e9be9d5e | 362 | { |
1175b6b8 VG |
363 | ssize_t res; |
364 | const struct cred *old_cred; | |
1d88f183 MS |
365 | struct dentry *realdentry = |
366 | ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry); | |
52148463 | 367 | |
1175b6b8 VG |
368 | old_cred = ovl_override_creds(dentry->d_sb); |
369 | res = vfs_getxattr(realdentry, name, value, size); | |
370 | revert_creds(old_cred); | |
371 | return res; | |
e9be9d5e MS |
372 | } |
373 | ||
a082c6f6 MS |
374 | static bool ovl_can_list(const char *s) |
375 | { | |
376 | /* List all non-trusted xatts */ | |
377 | if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0) | |
378 | return true; | |
379 | ||
380 | /* Never list trusted.overlay, list other trusted for superuser only */ | |
5c2e9f34 MS |
381 | return !ovl_is_private_xattr(s) && |
382 | ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN); | |
a082c6f6 MS |
383 | } |
384 | ||
e9be9d5e MS |
385 | ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size) |
386 | { | |
b581755b | 387 | struct dentry *realdentry = ovl_dentry_real(dentry); |
e9be9d5e | 388 | ssize_t res; |
7cb35119 MS |
389 | size_t len; |
390 | char *s; | |
1175b6b8 | 391 | const struct cred *old_cred; |
e9be9d5e | 392 | |
1175b6b8 | 393 | old_cred = ovl_override_creds(dentry->d_sb); |
b581755b | 394 | res = vfs_listxattr(realdentry, list, size); |
1175b6b8 | 395 | revert_creds(old_cred); |
e9be9d5e MS |
396 | if (res <= 0 || size == 0) |
397 | return res; | |
398 | ||
e9be9d5e | 399 | /* filter out private xattrs */ |
7cb35119 MS |
400 | for (s = list, len = res; len;) { |
401 | size_t slen = strnlen(s, len) + 1; | |
e9be9d5e | 402 | |
7cb35119 MS |
403 | /* underlying fs providing us with an broken xattr list? */ |
404 | if (WARN_ON(slen > len)) | |
405 | return -EIO; | |
e9be9d5e | 406 | |
7cb35119 | 407 | len -= slen; |
a082c6f6 | 408 | if (!ovl_can_list(s)) { |
e9be9d5e | 409 | res -= slen; |
7cb35119 | 410 | memmove(s, s + slen, len); |
e9be9d5e | 411 | } else { |
7cb35119 | 412 | s += slen; |
e9be9d5e MS |
413 | } |
414 | } | |
415 | ||
416 | return res; | |
417 | } | |
418 | ||
39a25b2b VG |
419 | struct posix_acl *ovl_get_acl(struct inode *inode, int type) |
420 | { | |
09d8b586 | 421 | struct inode *realinode = ovl_inode_real(inode); |
1175b6b8 VG |
422 | const struct cred *old_cred; |
423 | struct posix_acl *acl; | |
39a25b2b | 424 | |
5201dc44 | 425 | if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode)) |
39a25b2b VG |
426 | return NULL; |
427 | ||
1175b6b8 | 428 | old_cred = ovl_override_creds(inode->i_sb); |
5201dc44 | 429 | acl = get_acl(realinode, type); |
1175b6b8 VG |
430 | revert_creds(old_cred); |
431 | ||
432 | return acl; | |
39a25b2b VG |
433 | } |
434 | ||
95582b00 | 435 | int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags) |
d719e8f2 | 436 | { |
8f35cf51 MS |
437 | if (flags & S_ATIME) { |
438 | struct ovl_fs *ofs = inode->i_sb->s_fs_info; | |
439 | struct path upperpath = { | |
440 | .mnt = ofs->upper_mnt, | |
441 | .dentry = ovl_upperdentry_dereference(OVL_I(inode)), | |
442 | }; | |
443 | ||
444 | if (upperpath.dentry) { | |
445 | touch_atime(&upperpath); | |
446 | inode->i_atime = d_inode(upperpath.dentry)->i_atime; | |
447 | } | |
d719e8f2 | 448 | } |
d719e8f2 MS |
449 | return 0; |
450 | } | |
451 | ||
9e142c41 MS |
452 | static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
453 | u64 start, u64 len) | |
454 | { | |
455 | int err; | |
456 | struct inode *realinode = ovl_inode_real(inode); | |
457 | const struct cred *old_cred; | |
458 | ||
459 | if (!realinode->i_op->fiemap) | |
460 | return -EOPNOTSUPP; | |
461 | ||
462 | old_cred = ovl_override_creds(inode->i_sb); | |
80d34810 AG |
463 | |
464 | if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) | |
465 | filemap_write_and_wait(realinode->i_mapping); | |
466 | ||
9e142c41 MS |
467 | err = realinode->i_op->fiemap(realinode, fieinfo, start, len); |
468 | revert_creds(old_cred); | |
469 | ||
470 | return err; | |
471 | } | |
472 | ||
e9be9d5e MS |
473 | static const struct inode_operations ovl_file_inode_operations = { |
474 | .setattr = ovl_setattr, | |
475 | .permission = ovl_permission, | |
476 | .getattr = ovl_getattr, | |
e9be9d5e | 477 | .listxattr = ovl_listxattr, |
39a25b2b | 478 | .get_acl = ovl_get_acl, |
d719e8f2 | 479 | .update_time = ovl_update_time, |
9e142c41 | 480 | .fiemap = ovl_fiemap, |
e9be9d5e MS |
481 | }; |
482 | ||
483 | static const struct inode_operations ovl_symlink_inode_operations = { | |
484 | .setattr = ovl_setattr, | |
6b255391 | 485 | .get_link = ovl_get_link, |
e9be9d5e | 486 | .getattr = ovl_getattr, |
e9be9d5e | 487 | .listxattr = ovl_listxattr, |
d719e8f2 | 488 | .update_time = ovl_update_time, |
e9be9d5e MS |
489 | }; |
490 | ||
9e142c41 MS |
491 | static const struct inode_operations ovl_special_inode_operations = { |
492 | .setattr = ovl_setattr, | |
493 | .permission = ovl_permission, | |
494 | .getattr = ovl_getattr, | |
495 | .listxattr = ovl_listxattr, | |
496 | .get_acl = ovl_get_acl, | |
497 | .update_time = ovl_update_time, | |
498 | }; | |
499 | ||
69383c59 | 500 | static const struct address_space_operations ovl_aops = { |
5b910bd6 AG |
501 | /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */ |
502 | .direct_IO = noop_direct_IO, | |
503 | }; | |
504 | ||
b1eaa950 AG |
505 | /* |
506 | * It is possible to stack overlayfs instance on top of another | |
507 | * overlayfs instance as lower layer. We need to annonate the | |
508 | * stackable i_mutex locks according to stack level of the super | |
509 | * block instance. An overlayfs instance can never be in stack | |
510 | * depth 0 (there is always a real fs below it). An overlayfs | |
511 | * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth]. | |
512 | * | |
513 | * For example, here is a snip from /proc/lockdep_chains after | |
514 | * dir_iterate of nested overlayfs: | |
515 | * | |
516 | * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2) | |
517 | * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1) | |
518 | * [...] &type->i_mutex_dir_key (stack_depth=0) | |
b1f9d385 AG |
519 | * |
520 | * Locking order w.r.t ovl_want_write() is important for nested overlayfs. | |
521 | * | |
522 | * This chain is valid: | |
523 | * - inode->i_rwsem (inode_lock[2]) | |
524 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
525 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
526 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
527 | * | |
528 | * And this chain is valid: | |
529 | * - inode->i_rwsem (inode_lock[2]) | |
530 | * - OVL_I(inode)->lock (ovl_inode_lock[2]) | |
531 | * - lowerinode->i_rwsem (inode_lock[1]) | |
532 | * - OVL_I(lowerinode)->lock (ovl_inode_lock[1]) | |
533 | * | |
534 | * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is | |
535 | * held, because it is in reverse order of the non-nested case using the same | |
536 | * upper fs: | |
537 | * - inode->i_rwsem (inode_lock[1]) | |
538 | * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0]) | |
539 | * - OVL_I(inode)->lock (ovl_inode_lock[1]) | |
b1eaa950 AG |
540 | */ |
541 | #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH | |
542 | ||
543 | static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode) | |
544 | { | |
545 | #ifdef CONFIG_LOCKDEP | |
546 | static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING]; | |
547 | static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING]; | |
4eae06de | 548 | static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING]; |
b1eaa950 AG |
549 | |
550 | int depth = inode->i_sb->s_stack_depth - 1; | |
551 | ||
552 | if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING)) | |
553 | depth = 0; | |
554 | ||
555 | if (S_ISDIR(inode->i_mode)) | |
556 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]); | |
557 | else | |
558 | lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]); | |
4eae06de AG |
559 | |
560 | lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]); | |
b1eaa950 AG |
561 | #endif |
562 | } | |
563 | ||
695b46e7 | 564 | static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev, |
12574a9f | 565 | unsigned long ino, int fsid) |
e9be9d5e | 566 | { |
12574a9f AG |
567 | int xinobits = ovl_xino_bits(inode->i_sb); |
568 | ||
695b46e7 | 569 | /* |
6dde1e42 AG |
570 | * When d_ino is consistent with st_ino (samefs or i_ino has enough |
571 | * bits to encode layer), set the same value used for st_ino to i_ino, | |
572 | * so inode number exposed via /proc/locks and a like will be | |
573 | * consistent with d_ino and st_ino values. An i_ino value inconsistent | |
735c907d AG |
574 | * with d_ino also causes nfsd readdirplus to fail. |
575 | * | |
576 | * When called from ovl_create_object() => ovl_new_inode(), with | |
577 | * ino = 0, i_ino will be updated to consistent value later on in | |
578 | * ovl_get_inode() => ovl_fill_inode(). | |
695b46e7 | 579 | */ |
0f831ec8 | 580 | if (ovl_same_dev(inode->i_sb)) { |
695b46e7 | 581 | inode->i_ino = ino; |
12574a9f AG |
582 | if (xinobits && fsid && !(ino >> (64 - xinobits))) |
583 | inode->i_ino |= (unsigned long)fsid << (64 - xinobits); | |
584 | } else { | |
695b46e7 | 585 | inode->i_ino = get_next_ino(); |
12574a9f | 586 | } |
e9be9d5e | 587 | inode->i_mode = mode; |
d719e8f2 | 588 | inode->i_flags |= S_NOCMTIME; |
2a3a2a3f MS |
589 | #ifdef CONFIG_FS_POSIX_ACL |
590 | inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; | |
591 | #endif | |
e9be9d5e | 592 | |
b1eaa950 AG |
593 | ovl_lockdep_annotate_inode_mutex_key(inode); |
594 | ||
ca4c8a3a MS |
595 | switch (mode & S_IFMT) { |
596 | case S_IFREG: | |
597 | inode->i_op = &ovl_file_inode_operations; | |
d1d04ef8 | 598 | inode->i_fop = &ovl_file_operations; |
5b910bd6 | 599 | inode->i_mapping->a_ops = &ovl_aops; |
ca4c8a3a MS |
600 | break; |
601 | ||
e9be9d5e | 602 | case S_IFDIR: |
e9be9d5e MS |
603 | inode->i_op = &ovl_dir_inode_operations; |
604 | inode->i_fop = &ovl_dir_operations; | |
605 | break; | |
606 | ||
607 | case S_IFLNK: | |
608 | inode->i_op = &ovl_symlink_inode_operations; | |
609 | break; | |
610 | ||
51f7e52d | 611 | default: |
9e142c41 | 612 | inode->i_op = &ovl_special_inode_operations; |
ca4c8a3a | 613 | init_special_inode(inode, mode, rdev); |
e9be9d5e | 614 | break; |
51f7e52d MS |
615 | } |
616 | } | |
e9be9d5e | 617 | |
5f8415d6 AG |
618 | /* |
619 | * With inodes index enabled, an overlay inode nlink counts the union of upper | |
620 | * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure | |
621 | * upper inode, the following nlink modifying operations can happen: | |
622 | * | |
623 | * 1. Lower hardlink copy up | |
624 | * 2. Upper hardlink created, unlinked or renamed over | |
625 | * 3. Lower hardlink whiteout or renamed over | |
626 | * | |
627 | * For the first, copy up case, the union nlink does not change, whether the | |
628 | * operation succeeds or fails, but the upper inode nlink may change. | |
629 | * Therefore, before copy up, we store the union nlink value relative to the | |
630 | * lower inode nlink in the index inode xattr trusted.overlay.nlink. | |
631 | * | |
632 | * For the second, upper hardlink case, the union nlink should be incremented | |
633 | * or decremented IFF the operation succeeds, aligned with nlink change of the | |
634 | * upper inode. Therefore, before link/unlink/rename, we store the union nlink | |
635 | * value relative to the upper inode nlink in the index inode. | |
636 | * | |
637 | * For the last, lower cover up case, we simplify things by preceding the | |
638 | * whiteout or cover up with copy up. This makes sure that there is an index | |
639 | * upper inode where the nlink xattr can be stored before the copied up upper | |
640 | * entry is unlink. | |
641 | */ | |
642 | #define OVL_NLINK_ADD_UPPER (1 << 0) | |
643 | ||
644 | /* | |
645 | * On-disk format for indexed nlink: | |
646 | * | |
647 | * nlink relative to the upper inode - "U[+-]NUM" | |
648 | * nlink relative to the lower inode - "L[+-]NUM" | |
649 | */ | |
650 | ||
651 | static int ovl_set_nlink_common(struct dentry *dentry, | |
652 | struct dentry *realdentry, const char *format) | |
653 | { | |
654 | struct inode *inode = d_inode(dentry); | |
655 | struct inode *realinode = d_inode(realdentry); | |
656 | char buf[13]; | |
657 | int len; | |
658 | ||
659 | len = snprintf(buf, sizeof(buf), format, | |
660 | (int) (inode->i_nlink - realinode->i_nlink)); | |
661 | ||
6787341a MS |
662 | if (WARN_ON(len >= sizeof(buf))) |
663 | return -EIO; | |
664 | ||
5f8415d6 AG |
665 | return ovl_do_setxattr(ovl_dentry_upper(dentry), |
666 | OVL_XATTR_NLINK, buf, len, 0); | |
667 | } | |
668 | ||
669 | int ovl_set_nlink_upper(struct dentry *dentry) | |
670 | { | |
671 | return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i"); | |
672 | } | |
673 | ||
674 | int ovl_set_nlink_lower(struct dentry *dentry) | |
675 | { | |
676 | return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i"); | |
677 | } | |
678 | ||
caf70cb2 AG |
679 | unsigned int ovl_get_nlink(struct dentry *lowerdentry, |
680 | struct dentry *upperdentry, | |
681 | unsigned int fallback) | |
5f8415d6 AG |
682 | { |
683 | int nlink_diff; | |
684 | int nlink; | |
685 | char buf[13]; | |
686 | int err; | |
687 | ||
688 | if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1) | |
689 | return fallback; | |
690 | ||
691 | err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1); | |
692 | if (err < 0) | |
693 | goto fail; | |
694 | ||
695 | buf[err] = '\0'; | |
696 | if ((buf[0] != 'L' && buf[0] != 'U') || | |
697 | (buf[1] != '+' && buf[1] != '-')) | |
698 | goto fail; | |
699 | ||
700 | err = kstrtoint(buf + 1, 10, &nlink_diff); | |
701 | if (err < 0) | |
702 | goto fail; | |
703 | ||
704 | nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink; | |
705 | nlink += nlink_diff; | |
706 | ||
707 | if (nlink <= 0) | |
708 | goto fail; | |
709 | ||
710 | return nlink; | |
711 | ||
712 | fail: | |
1bd0a3ae | 713 | pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n", |
5f8415d6 AG |
714 | upperdentry, err); |
715 | return fallback; | |
716 | } | |
717 | ||
ca4c8a3a | 718 | struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev) |
51f7e52d MS |
719 | { |
720 | struct inode *inode; | |
721 | ||
722 | inode = new_inode(sb); | |
723 | if (inode) | |
12574a9f | 724 | ovl_fill_inode(inode, mode, rdev, 0, 0); |
51f7e52d MS |
725 | |
726 | return inode; | |
727 | } | |
728 | ||
729 | static int ovl_inode_test(struct inode *inode, void *data) | |
730 | { | |
25b7713a | 731 | return inode->i_private == data; |
51f7e52d MS |
732 | } |
733 | ||
734 | static int ovl_inode_set(struct inode *inode, void *data) | |
735 | { | |
25b7713a | 736 | inode->i_private = data; |
51f7e52d MS |
737 | return 0; |
738 | } | |
739 | ||
b9ac5c27 | 740 | static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry, |
4b91c30a | 741 | struct dentry *upperdentry, bool strict) |
b9ac5c27 | 742 | { |
4b91c30a AG |
743 | /* |
744 | * For directories, @strict verify from lookup path performs consistency | |
745 | * checks, so NULL lower/upper in dentry must match NULL lower/upper in | |
746 | * inode. Non @strict verify from NFS handle decode path passes NULL for | |
747 | * 'unknown' lower/upper. | |
748 | */ | |
749 | if (S_ISDIR(inode->i_mode) && strict) { | |
31747eda AG |
750 | /* Real lower dir moved to upper layer under us? */ |
751 | if (!lowerdentry && ovl_inode_lower(inode)) | |
752 | return false; | |
753 | ||
754 | /* Lookup of an uncovered redirect origin? */ | |
755 | if (!upperdentry && ovl_inode_upper(inode)) | |
756 | return false; | |
757 | } | |
758 | ||
939ae4ef AG |
759 | /* |
760 | * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL. | |
761 | * This happens when finding a copied up overlay inode for a renamed | |
762 | * or hardlinked overlay dentry and lower dentry cannot be followed | |
763 | * by origin because lower fs does not support file handles. | |
764 | */ | |
765 | if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry)) | |
b9ac5c27 MS |
766 | return false; |
767 | ||
768 | /* | |
769 | * Allow non-NULL __upperdentry in inode even if upperdentry is NULL. | |
770 | * This happens when finding a lower alias for a copied up hard link. | |
771 | */ | |
772 | if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry)) | |
773 | return false; | |
774 | ||
775 | return true; | |
776 | } | |
777 | ||
4b91c30a AG |
778 | struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real, |
779 | bool is_upper) | |
9436a1a3 | 780 | { |
4b91c30a | 781 | struct inode *inode, *key = d_inode(real); |
9436a1a3 AG |
782 | |
783 | inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
784 | if (!inode) | |
785 | return NULL; | |
786 | ||
4b91c30a AG |
787 | if (!ovl_verify_inode(inode, is_upper ? NULL : real, |
788 | is_upper ? real : NULL, false)) { | |
9436a1a3 AG |
789 | iput(inode); |
790 | return ERR_PTR(-ESTALE); | |
791 | } | |
792 | ||
793 | return inode; | |
794 | } | |
795 | ||
146d62e5 AG |
796 | bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir) |
797 | { | |
798 | struct inode *key = d_inode(dir); | |
799 | struct inode *trap; | |
800 | bool res; | |
801 | ||
802 | trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key); | |
803 | if (!trap) | |
804 | return false; | |
805 | ||
806 | res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) && | |
807 | !ovl_inode_lower(trap); | |
808 | ||
809 | iput(trap); | |
810 | return res; | |
811 | } | |
812 | ||
813 | /* | |
814 | * Create an inode cache entry for layer root dir, that will intentionally | |
815 | * fail ovl_verify_inode(), so any lookup that will find some layer root | |
816 | * will fail. | |
817 | */ | |
818 | struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir) | |
819 | { | |
820 | struct inode *key = d_inode(dir); | |
821 | struct inode *trap; | |
822 | ||
823 | if (!d_is_dir(dir)) | |
824 | return ERR_PTR(-ENOTDIR); | |
825 | ||
826 | trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test, | |
827 | ovl_inode_set, key); | |
828 | if (!trap) | |
829 | return ERR_PTR(-ENOMEM); | |
830 | ||
831 | if (!(trap->i_state & I_NEW)) { | |
832 | /* Conflicting layer roots? */ | |
833 | iput(trap); | |
834 | return ERR_PTR(-ELOOP); | |
835 | } | |
836 | ||
837 | trap->i_mode = S_IFDIR; | |
838 | trap->i_flags = S_DEAD; | |
839 | unlock_new_inode(trap); | |
840 | ||
841 | return trap; | |
842 | } | |
843 | ||
764baba8 AG |
844 | /* |
845 | * Does overlay inode need to be hashed by lower inode? | |
846 | */ | |
847 | static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper, | |
848 | struct dentry *lower, struct dentry *index) | |
849 | { | |
850 | struct ovl_fs *ofs = sb->s_fs_info; | |
851 | ||
852 | /* No, if pure upper */ | |
853 | if (!lower) | |
854 | return false; | |
855 | ||
856 | /* Yes, if already indexed */ | |
857 | if (index) | |
858 | return true; | |
859 | ||
860 | /* Yes, if won't be copied up */ | |
861 | if (!ofs->upper_mnt) | |
862 | return true; | |
863 | ||
864 | /* No, if lower hardlink is or will be broken on copy up */ | |
865 | if ((upper || !ovl_indexdir(sb)) && | |
866 | !d_is_dir(lower) && d_inode(lower)->i_nlink > 1) | |
867 | return false; | |
868 | ||
869 | /* No, if non-indexed upper with NFS export */ | |
870 | if (sb->s_export_op && upper) | |
871 | return false; | |
872 | ||
873 | /* Otherwise, hash by lower inode for fsnotify */ | |
874 | return true; | |
875 | } | |
876 | ||
01b39dcc AG |
877 | static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode, |
878 | struct inode *key) | |
879 | { | |
880 | return newinode ? inode_insert5(newinode, (unsigned long) key, | |
881 | ovl_inode_test, ovl_inode_set, key) : | |
882 | iget5_locked(sb, (unsigned long) key, | |
883 | ovl_inode_test, ovl_inode_set, key); | |
884 | } | |
885 | ||
ac6a52eb VG |
886 | struct inode *ovl_get_inode(struct super_block *sb, |
887 | struct ovl_inode_params *oip) | |
51f7e52d | 888 | { |
ac6a52eb VG |
889 | struct dentry *upperdentry = oip->upperdentry; |
890 | struct ovl_path *lowerpath = oip->lowerpath; | |
09d8b586 | 891 | struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL; |
51f7e52d | 892 | struct inode *inode; |
12574a9f | 893 | struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL; |
ac6a52eb VG |
894 | bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry, |
895 | oip->index); | |
300b124f | 896 | int fsid = bylower ? lowerpath->layer->fsid : 0; |
9d3dfea3 | 897 | bool is_dir, metacopy = false; |
695b46e7 | 898 | unsigned long ino = 0; |
acf3062a | 899 | int err = oip->newinode ? -EEXIST : -ENOMEM; |
6eaf0111 | 900 | |
09d8b586 MS |
901 | if (!realinode) |
902 | realinode = d_inode(lowerdentry); | |
903 | ||
6eaf0111 | 904 | /* |
764baba8 AG |
905 | * Copy up origin (lower) may exist for non-indexed upper, but we must |
906 | * not use lower as hash key if this is a broken hardlink. | |
6eaf0111 | 907 | */ |
31747eda | 908 | is_dir = S_ISDIR(realinode->i_mode); |
764baba8 AG |
909 | if (upperdentry || bylower) { |
910 | struct inode *key = d_inode(bylower ? lowerdentry : | |
911 | upperdentry); | |
31747eda | 912 | unsigned int nlink = is_dir ? 1 : realinode->i_nlink; |
b9ac5c27 | 913 | |
01b39dcc | 914 | inode = ovl_iget5(sb, oip->newinode, key); |
09d8b586 | 915 | if (!inode) |
027065b7 | 916 | goto out_err; |
09d8b586 | 917 | if (!(inode->i_state & I_NEW)) { |
b9ac5c27 MS |
918 | /* |
919 | * Verify that the underlying files stored in the inode | |
920 | * match those in the dentry. | |
921 | */ | |
4b91c30a AG |
922 | if (!ovl_verify_inode(inode, lowerdentry, upperdentry, |
923 | true)) { | |
b9ac5c27 | 924 | iput(inode); |
027065b7 VG |
925 | err = -ESTALE; |
926 | goto out_err; | |
b9ac5c27 MS |
927 | } |
928 | ||
09d8b586 | 929 | dput(upperdentry); |
9cec54c8 | 930 | kfree(oip->redirect); |
09d8b586 MS |
931 | goto out; |
932 | } | |
e6d2ebdd | 933 | |
31747eda AG |
934 | /* Recalculate nlink for non-dir due to indexing */ |
935 | if (!is_dir) | |
936 | nlink = ovl_get_nlink(lowerdentry, upperdentry, nlink); | |
5f8415d6 | 937 | set_nlink(inode, nlink); |
695b46e7 | 938 | ino = key->i_ino; |
e6d2ebdd | 939 | } else { |
764baba8 | 940 | /* Lower hardlink that will be broken on copy up */ |
0aceb53e | 941 | inode = new_inode(sb); |
027065b7 VG |
942 | if (!inode) { |
943 | err = -ENOMEM; | |
944 | goto out_err; | |
945 | } | |
300b124f AG |
946 | ino = realinode->i_ino; |
947 | fsid = lowerpath->layer->fsid; | |
e9be9d5e | 948 | } |
12574a9f | 949 | ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev, ino, fsid); |
2664bd08 | 950 | ovl_inode_init(inode, upperdentry, lowerdentry, oip->lowerdata); |
13c72075 MS |
951 | |
952 | if (upperdentry && ovl_is_impuredir(upperdentry)) | |
953 | ovl_set_flag(OVL_IMPURE, inode); | |
954 | ||
ac6a52eb | 955 | if (oip->index) |
0471a9cd VG |
956 | ovl_set_flag(OVL_INDEX, inode); |
957 | ||
9d3dfea3 VG |
958 | if (upperdentry) { |
959 | err = ovl_check_metacopy_xattr(upperdentry); | |
960 | if (err < 0) | |
961 | goto out_err; | |
962 | metacopy = err; | |
963 | if (!metacopy) | |
964 | ovl_set_flag(OVL_UPPERDATA, inode); | |
965 | } | |
966 | ||
9cec54c8 VG |
967 | OVL_I(inode)->redirect = oip->redirect; |
968 | ||
a00c2d59 VG |
969 | if (bylower) |
970 | ovl_set_flag(OVL_CONST_INO, inode); | |
971 | ||
b79e05aa | 972 | /* Check for non-merge dir that may have whiteouts */ |
31747eda | 973 | if (is_dir) { |
ac6a52eb | 974 | if (((upperdentry && lowerdentry) || oip->numlower > 1) || |
b79e05aa AG |
975 | ovl_check_origin_xattr(upperdentry ?: lowerdentry)) { |
976 | ovl_set_flag(OVL_WHITEOUTS, inode); | |
977 | } | |
978 | } | |
979 | ||
e6d2ebdd MS |
980 | if (inode->i_state & I_NEW) |
981 | unlock_new_inode(inode); | |
982 | out: | |
e9be9d5e | 983 | return inode; |
b9ac5c27 | 984 | |
027065b7 | 985 | out_err: |
1bd0a3ae | 986 | pr_warn_ratelimited("failed to get inode (%i)\n", err); |
027065b7 | 987 | inode = ERR_PTR(err); |
b9ac5c27 | 988 | goto out; |
e9be9d5e | 989 | } |