]>
Commit | Line | Data |
---|---|---|
e9be9d5e MS |
1 | /* |
2 | * | |
3 | * Copyright (C) 2011 Novell Inc. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms of the GNU General Public License version 2 as published by | |
7 | * the Free Software Foundation. | |
8 | */ | |
9 | ||
5b825c3a | 10 | #include <uapi/linux/magic.h> |
e9be9d5e MS |
11 | #include <linux/fs.h> |
12 | #include <linux/namei.h> | |
13 | #include <linux/xattr.h> | |
e9be9d5e | 14 | #include <linux/mount.h> |
e9be9d5e MS |
15 | #include <linux/parser.h> |
16 | #include <linux/module.h> | |
cc259639 | 17 | #include <linux/statfs.h> |
f45827e8 | 18 | #include <linux/seq_file.h> |
d837a49b | 19 | #include <linux/posix_acl_xattr.h> |
e9be9d5e | 20 | #include "overlayfs.h" |
bbb1e54d | 21 | #include "ovl_entry.h" |
e9be9d5e MS |
22 | |
23 | MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>"); | |
24 | MODULE_DESCRIPTION("Overlay filesystem"); | |
25 | MODULE_LICENSE("GPL"); | |
26 | ||
e9be9d5e MS |
27 | |
28 | struct ovl_dir_cache; | |
29 | ||
a78d9f0d MS |
30 | #define OVL_MAX_STACK 500 |
31 | ||
688ea0e5 MS |
32 | static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR); |
33 | module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644); | |
34 | MODULE_PARM_DESC(ovl_redirect_dir_def, | |
35 | "Default to on or off for the redirect_dir feature"); | |
e9be9d5e | 36 | |
02bcd157 AG |
37 | static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX); |
38 | module_param_named(index, ovl_index_def, bool, 0644); | |
39 | MODULE_PARM_DESC(ovl_index_def, | |
40 | "Default to on or off for the inodes index feature"); | |
41 | ||
e9be9d5e MS |
42 | static void ovl_dentry_release(struct dentry *dentry) |
43 | { | |
44 | struct ovl_entry *oe = dentry->d_fsdata; | |
45 | ||
46 | if (oe) { | |
dd662667 MS |
47 | unsigned int i; |
48 | ||
dd662667 MS |
49 | for (i = 0; i < oe->numlower; i++) |
50 | dput(oe->lowerstack[i].dentry); | |
e9be9d5e MS |
51 | kfree_rcu(oe, rcu); |
52 | } | |
53 | } | |
54 | ||
b0990fbb AG |
55 | static int ovl_check_append_only(struct inode *inode, int flag) |
56 | { | |
57 | /* | |
58 | * This test was moot in vfs may_open() because overlay inode does | |
59 | * not have the S_APPEND flag, so re-check on real upper inode | |
60 | */ | |
61 | if (IS_APPEND(inode)) { | |
62 | if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) | |
63 | return -EPERM; | |
64 | if (flag & O_TRUNC) | |
65 | return -EPERM; | |
66 | } | |
67 | ||
68 | return 0; | |
69 | } | |
70 | ||
2d902671 MS |
71 | static struct dentry *ovl_d_real(struct dentry *dentry, |
72 | const struct inode *inode, | |
73 | unsigned int open_flags) | |
d101a125 MS |
74 | { |
75 | struct dentry *real; | |
b0990fbb | 76 | int err; |
d101a125 | 77 | |
ca4c8a3a | 78 | if (!d_is_reg(dentry)) { |
d101a125 MS |
79 | if (!inode || inode == d_inode(dentry)) |
80 | return dentry; | |
81 | goto bug; | |
82 | } | |
83 | ||
2d902671 MS |
84 | if (d_is_negative(dentry)) |
85 | return dentry; | |
86 | ||
87 | if (open_flags) { | |
b0990fbb | 88 | err = ovl_open_maybe_copy_up(dentry, open_flags); |
2d902671 MS |
89 | if (err) |
90 | return ERR_PTR(err); | |
91 | } | |
92 | ||
d101a125 | 93 | real = ovl_dentry_upper(dentry); |
b0990fbb AG |
94 | if (real && (!inode || inode == d_inode(real))) { |
95 | if (!inode) { | |
96 | err = ovl_check_append_only(d_inode(real), open_flags); | |
97 | if (err) | |
98 | return ERR_PTR(err); | |
99 | } | |
d101a125 | 100 | return real; |
b0990fbb | 101 | } |
d101a125 MS |
102 | |
103 | real = ovl_dentry_lower(dentry); | |
104 | if (!real) | |
105 | goto bug; | |
106 | ||
c4fcfc16 MS |
107 | /* Handle recursion */ |
108 | real = d_real(real, inode, open_flags); | |
109 | ||
d101a125 MS |
110 | if (!inode || inode == d_inode(real)) |
111 | return real; | |
d101a125 | 112 | bug: |
656189d2 | 113 | WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry, |
d101a125 MS |
114 | inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0); |
115 | return dentry; | |
116 | } | |
117 | ||
7c03b5d4 MS |
118 | static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags) |
119 | { | |
120 | struct ovl_entry *oe = dentry->d_fsdata; | |
121 | unsigned int i; | |
122 | int ret = 1; | |
123 | ||
124 | for (i = 0; i < oe->numlower; i++) { | |
125 | struct dentry *d = oe->lowerstack[i].dentry; | |
126 | ||
127 | if (d->d_flags & DCACHE_OP_REVALIDATE) { | |
128 | ret = d->d_op->d_revalidate(d, flags); | |
129 | if (ret < 0) | |
130 | return ret; | |
131 | if (!ret) { | |
132 | if (!(flags & LOOKUP_RCU)) | |
133 | d_invalidate(d); | |
134 | return -ESTALE; | |
135 | } | |
136 | } | |
137 | } | |
138 | return 1; | |
139 | } | |
140 | ||
141 | static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags) | |
142 | { | |
143 | struct ovl_entry *oe = dentry->d_fsdata; | |
144 | unsigned int i; | |
145 | int ret = 1; | |
146 | ||
147 | for (i = 0; i < oe->numlower; i++) { | |
148 | struct dentry *d = oe->lowerstack[i].dentry; | |
149 | ||
150 | if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) { | |
151 | ret = d->d_op->d_weak_revalidate(d, flags); | |
152 | if (ret <= 0) | |
153 | break; | |
154 | } | |
155 | } | |
156 | return ret; | |
157 | } | |
158 | ||
e9be9d5e MS |
159 | static const struct dentry_operations ovl_dentry_operations = { |
160 | .d_release = ovl_dentry_release, | |
d101a125 | 161 | .d_real = ovl_d_real, |
e9be9d5e MS |
162 | }; |
163 | ||
7c03b5d4 MS |
164 | static const struct dentry_operations ovl_reval_dentry_operations = { |
165 | .d_release = ovl_dentry_release, | |
d101a125 | 166 | .d_real = ovl_d_real, |
7c03b5d4 MS |
167 | .d_revalidate = ovl_dentry_revalidate, |
168 | .d_weak_revalidate = ovl_dentry_weak_revalidate, | |
169 | }; | |
170 | ||
13cf199d AG |
171 | static struct kmem_cache *ovl_inode_cachep; |
172 | ||
173 | static struct inode *ovl_alloc_inode(struct super_block *sb) | |
174 | { | |
175 | struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL); | |
176 | ||
5c2f6e8a HN |
177 | if (!oi) |
178 | return NULL; | |
179 | ||
04a01ac7 | 180 | oi->cache = NULL; |
cf31c463 | 181 | oi->redirect = NULL; |
04a01ac7 | 182 | oi->version = 0; |
13c72075 | 183 | oi->flags = 0; |
09d8b586 | 184 | oi->__upperdentry = NULL; |
25b7713a | 185 | oi->lower = NULL; |
a015dafc | 186 | mutex_init(&oi->lock); |
25b7713a | 187 | |
13cf199d AG |
188 | return &oi->vfs_inode; |
189 | } | |
190 | ||
191 | static void ovl_i_callback(struct rcu_head *head) | |
192 | { | |
193 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
194 | ||
195 | kmem_cache_free(ovl_inode_cachep, OVL_I(inode)); | |
196 | } | |
197 | ||
198 | static void ovl_destroy_inode(struct inode *inode) | |
199 | { | |
09d8b586 MS |
200 | struct ovl_inode *oi = OVL_I(inode); |
201 | ||
202 | dput(oi->__upperdentry); | |
cf31c463 | 203 | kfree(oi->redirect); |
a015dafc | 204 | mutex_destroy(&oi->lock); |
09d8b586 | 205 | |
13cf199d AG |
206 | call_rcu(&inode->i_rcu, ovl_i_callback); |
207 | } | |
208 | ||
e9be9d5e MS |
209 | static void ovl_put_super(struct super_block *sb) |
210 | { | |
211 | struct ovl_fs *ufs = sb->s_fs_info; | |
dd662667 | 212 | unsigned i; |
e9be9d5e | 213 | |
02bcd157 | 214 | dput(ufs->indexdir); |
e9be9d5e | 215 | dput(ufs->workdir); |
cd499a84 AG |
216 | if (ufs->workdir_locked) |
217 | ovl_inuse_unlock(ufs->workbasedir); | |
2cac0c00 | 218 | dput(ufs->workbasedir); |
cd499a84 | 219 | if (ufs->upper_mnt && ufs->upperdir_locked) |
2cac0c00 | 220 | ovl_inuse_unlock(ufs->upper_mnt->mnt_root); |
e9be9d5e | 221 | mntput(ufs->upper_mnt); |
dd662667 MS |
222 | for (i = 0; i < ufs->numlower; i++) |
223 | mntput(ufs->lower_mnt[i]); | |
5ffdbe8b | 224 | kfree(ufs->lower_mnt); |
e9be9d5e | 225 | |
f45827e8 EZ |
226 | kfree(ufs->config.lowerdir); |
227 | kfree(ufs->config.upperdir); | |
228 | kfree(ufs->config.workdir); | |
3fe6e52f | 229 | put_cred(ufs->creator_cred); |
e9be9d5e MS |
230 | kfree(ufs); |
231 | } | |
232 | ||
e593b2bf AG |
233 | static int ovl_sync_fs(struct super_block *sb, int wait) |
234 | { | |
235 | struct ovl_fs *ufs = sb->s_fs_info; | |
236 | struct super_block *upper_sb; | |
237 | int ret; | |
238 | ||
239 | if (!ufs->upper_mnt) | |
240 | return 0; | |
241 | upper_sb = ufs->upper_mnt->mnt_sb; | |
242 | if (!upper_sb->s_op->sync_fs) | |
243 | return 0; | |
244 | ||
245 | /* real inodes have already been synced by sync_filesystem(ovl_sb) */ | |
246 | down_read(&upper_sb->s_umount); | |
247 | ret = upper_sb->s_op->sync_fs(upper_sb, wait); | |
248 | up_read(&upper_sb->s_umount); | |
249 | return ret; | |
250 | } | |
251 | ||
cc259639 AW |
252 | /** |
253 | * ovl_statfs | |
254 | * @sb: The overlayfs super block | |
255 | * @buf: The struct kstatfs to fill in with stats | |
256 | * | |
257 | * Get the filesystem statistics. As writes always target the upper layer | |
4ebc5818 | 258 | * filesystem pass the statfs to the upper filesystem (if it exists) |
cc259639 AW |
259 | */ |
260 | static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf) | |
261 | { | |
262 | struct ovl_fs *ofs = dentry->d_sb->s_fs_info; | |
263 | struct dentry *root_dentry = dentry->d_sb->s_root; | |
264 | struct path path; | |
265 | int err; | |
266 | ||
4ebc5818 | 267 | ovl_path_real(root_dentry, &path); |
cc259639 AW |
268 | |
269 | err = vfs_statfs(&path, buf); | |
270 | if (!err) { | |
6b2d5fe4 | 271 | buf->f_namelen = ofs->namelen; |
cc259639 AW |
272 | buf->f_type = OVERLAYFS_SUPER_MAGIC; |
273 | } | |
274 | ||
275 | return err; | |
276 | } | |
277 | ||
02bcd157 AG |
278 | /* Will this overlay be forced to mount/remount ro? */ |
279 | static bool ovl_force_readonly(struct ovl_fs *ufs) | |
280 | { | |
281 | return (!ufs->upper_mnt || !ufs->workdir); | |
282 | } | |
283 | ||
f45827e8 EZ |
284 | /** |
285 | * ovl_show_options | |
286 | * | |
287 | * Prints the mount options for a given superblock. | |
288 | * Returns zero; does not fail. | |
289 | */ | |
290 | static int ovl_show_options(struct seq_file *m, struct dentry *dentry) | |
291 | { | |
292 | struct super_block *sb = dentry->d_sb; | |
293 | struct ovl_fs *ufs = sb->s_fs_info; | |
294 | ||
a068acf2 | 295 | seq_show_option(m, "lowerdir", ufs->config.lowerdir); |
53a08cb9 | 296 | if (ufs->config.upperdir) { |
a068acf2 KC |
297 | seq_show_option(m, "upperdir", ufs->config.upperdir); |
298 | seq_show_option(m, "workdir", ufs->config.workdir); | |
53a08cb9 | 299 | } |
8d3095f4 MS |
300 | if (ufs->config.default_permissions) |
301 | seq_puts(m, ",default_permissions"); | |
c5bef3a7 AG |
302 | if (ufs->config.redirect_dir != ovl_redirect_dir_def) |
303 | seq_printf(m, ",redirect_dir=%s", | |
304 | ufs->config.redirect_dir ? "on" : "off"); | |
02bcd157 AG |
305 | if (ufs->config.index != ovl_index_def) |
306 | seq_printf(m, ",index=%s", | |
307 | ufs->config.index ? "on" : "off"); | |
f45827e8 EZ |
308 | return 0; |
309 | } | |
310 | ||
3cdf6fe9 SL |
311 | static int ovl_remount(struct super_block *sb, int *flags, char *data) |
312 | { | |
313 | struct ovl_fs *ufs = sb->s_fs_info; | |
314 | ||
02bcd157 | 315 | if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs)) |
3cdf6fe9 SL |
316 | return -EROFS; |
317 | ||
318 | return 0; | |
319 | } | |
320 | ||
e9be9d5e | 321 | static const struct super_operations ovl_super_operations = { |
13cf199d AG |
322 | .alloc_inode = ovl_alloc_inode, |
323 | .destroy_inode = ovl_destroy_inode, | |
324 | .drop_inode = generic_delete_inode, | |
e9be9d5e | 325 | .put_super = ovl_put_super, |
e593b2bf | 326 | .sync_fs = ovl_sync_fs, |
cc259639 | 327 | .statfs = ovl_statfs, |
f45827e8 | 328 | .show_options = ovl_show_options, |
3cdf6fe9 | 329 | .remount_fs = ovl_remount, |
e9be9d5e MS |
330 | }; |
331 | ||
332 | enum { | |
333 | OPT_LOWERDIR, | |
334 | OPT_UPPERDIR, | |
335 | OPT_WORKDIR, | |
8d3095f4 | 336 | OPT_DEFAULT_PERMISSIONS, |
a6c60655 MS |
337 | OPT_REDIRECT_DIR_ON, |
338 | OPT_REDIRECT_DIR_OFF, | |
02bcd157 AG |
339 | OPT_INDEX_ON, |
340 | OPT_INDEX_OFF, | |
e9be9d5e MS |
341 | OPT_ERR, |
342 | }; | |
343 | ||
344 | static const match_table_t ovl_tokens = { | |
345 | {OPT_LOWERDIR, "lowerdir=%s"}, | |
346 | {OPT_UPPERDIR, "upperdir=%s"}, | |
347 | {OPT_WORKDIR, "workdir=%s"}, | |
8d3095f4 | 348 | {OPT_DEFAULT_PERMISSIONS, "default_permissions"}, |
a6c60655 MS |
349 | {OPT_REDIRECT_DIR_ON, "redirect_dir=on"}, |
350 | {OPT_REDIRECT_DIR_OFF, "redirect_dir=off"}, | |
02bcd157 AG |
351 | {OPT_INDEX_ON, "index=on"}, |
352 | {OPT_INDEX_OFF, "index=off"}, | |
e9be9d5e MS |
353 | {OPT_ERR, NULL} |
354 | }; | |
355 | ||
91c77947 MS |
356 | static char *ovl_next_opt(char **s) |
357 | { | |
358 | char *sbegin = *s; | |
359 | char *p; | |
360 | ||
361 | if (sbegin == NULL) | |
362 | return NULL; | |
363 | ||
364 | for (p = sbegin; *p; p++) { | |
365 | if (*p == '\\') { | |
366 | p++; | |
367 | if (!*p) | |
368 | break; | |
369 | } else if (*p == ',') { | |
370 | *p = '\0'; | |
371 | *s = p + 1; | |
372 | return sbegin; | |
373 | } | |
374 | } | |
375 | *s = NULL; | |
376 | return sbegin; | |
377 | } | |
378 | ||
e9be9d5e MS |
379 | static int ovl_parse_opt(char *opt, struct ovl_config *config) |
380 | { | |
381 | char *p; | |
382 | ||
91c77947 | 383 | while ((p = ovl_next_opt(&opt)) != NULL) { |
e9be9d5e MS |
384 | int token; |
385 | substring_t args[MAX_OPT_ARGS]; | |
386 | ||
387 | if (!*p) | |
388 | continue; | |
389 | ||
390 | token = match_token(p, ovl_tokens, args); | |
391 | switch (token) { | |
392 | case OPT_UPPERDIR: | |
393 | kfree(config->upperdir); | |
394 | config->upperdir = match_strdup(&args[0]); | |
395 | if (!config->upperdir) | |
396 | return -ENOMEM; | |
397 | break; | |
398 | ||
399 | case OPT_LOWERDIR: | |
400 | kfree(config->lowerdir); | |
401 | config->lowerdir = match_strdup(&args[0]); | |
402 | if (!config->lowerdir) | |
403 | return -ENOMEM; | |
404 | break; | |
405 | ||
406 | case OPT_WORKDIR: | |
407 | kfree(config->workdir); | |
408 | config->workdir = match_strdup(&args[0]); | |
409 | if (!config->workdir) | |
410 | return -ENOMEM; | |
411 | break; | |
412 | ||
8d3095f4 MS |
413 | case OPT_DEFAULT_PERMISSIONS: |
414 | config->default_permissions = true; | |
415 | break; | |
416 | ||
a6c60655 MS |
417 | case OPT_REDIRECT_DIR_ON: |
418 | config->redirect_dir = true; | |
419 | break; | |
420 | ||
421 | case OPT_REDIRECT_DIR_OFF: | |
422 | config->redirect_dir = false; | |
423 | break; | |
424 | ||
02bcd157 AG |
425 | case OPT_INDEX_ON: |
426 | config->index = true; | |
427 | break; | |
428 | ||
429 | case OPT_INDEX_OFF: | |
430 | config->index = false; | |
431 | break; | |
432 | ||
e9be9d5e | 433 | default: |
bead55ef | 434 | pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p); |
e9be9d5e MS |
435 | return -EINVAL; |
436 | } | |
437 | } | |
71cbad7e | 438 | |
439 | /* Workdir is useless in non-upper mount */ | |
440 | if (!config->upperdir && config->workdir) { | |
441 | pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n", | |
442 | config->workdir); | |
443 | kfree(config->workdir); | |
444 | config->workdir = NULL; | |
445 | } | |
446 | ||
e9be9d5e MS |
447 | return 0; |
448 | } | |
449 | ||
450 | #define OVL_WORKDIR_NAME "work" | |
02bcd157 | 451 | #define OVL_INDEXDIR_NAME "index" |
e9be9d5e | 452 | |
6b8aa129 AG |
453 | static struct dentry *ovl_workdir_create(struct super_block *sb, |
454 | struct ovl_fs *ufs, | |
455 | struct dentry *dentry, | |
456 | const char *name, bool persist) | |
e9be9d5e MS |
457 | { |
458 | struct inode *dir = dentry->d_inode; | |
6b8aa129 | 459 | struct vfsmount *mnt = ufs->upper_mnt; |
e9be9d5e MS |
460 | struct dentry *work; |
461 | int err; | |
462 | bool retried = false; | |
6b8aa129 | 463 | bool locked = false; |
e9be9d5e MS |
464 | |
465 | err = mnt_want_write(mnt); | |
466 | if (err) | |
6b8aa129 | 467 | goto out_err; |
e9be9d5e | 468 | |
5955102c | 469 | inode_lock_nested(dir, I_MUTEX_PARENT); |
6b8aa129 AG |
470 | locked = true; |
471 | ||
e9be9d5e | 472 | retry: |
6b8aa129 | 473 | work = lookup_one_len(name, dentry, strlen(name)); |
e9be9d5e MS |
474 | |
475 | if (!IS_ERR(work)) { | |
c11b9fdd MS |
476 | struct iattr attr = { |
477 | .ia_valid = ATTR_MODE, | |
32a3d848 | 478 | .ia_mode = S_IFDIR | 0, |
c11b9fdd | 479 | }; |
e9be9d5e MS |
480 | |
481 | if (work->d_inode) { | |
482 | err = -EEXIST; | |
483 | if (retried) | |
484 | goto out_dput; | |
485 | ||
6b8aa129 AG |
486 | if (persist) |
487 | goto out_unlock; | |
488 | ||
e9be9d5e | 489 | retried = true; |
eea2fb48 | 490 | ovl_workdir_cleanup(dir, mnt, work, 0); |
e9be9d5e MS |
491 | dput(work); |
492 | goto retry; | |
493 | } | |
494 | ||
32a3d848 AV |
495 | err = ovl_create_real(dir, work, |
496 | &(struct cattr){.mode = S_IFDIR | 0}, | |
497 | NULL, true); | |
e9be9d5e MS |
498 | if (err) |
499 | goto out_dput; | |
c11b9fdd | 500 | |
cb348edb MS |
501 | /* |
502 | * Try to remove POSIX ACL xattrs from workdir. We are good if: | |
503 | * | |
504 | * a) success (there was a POSIX ACL xattr and was removed) | |
505 | * b) -ENODATA (there was no POSIX ACL xattr) | |
506 | * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported) | |
507 | * | |
508 | * There are various other error values that could effectively | |
509 | * mean that the xattr doesn't exist (e.g. -ERANGE is returned | |
510 | * if the xattr name is too long), but the set of filesystems | |
511 | * allowed as upper are limited to "normal" ones, where checking | |
512 | * for the above two errors is sufficient. | |
513 | */ | |
c11b9fdd | 514 | err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT); |
e1ff3dd1 | 515 | if (err && err != -ENODATA && err != -EOPNOTSUPP) |
c11b9fdd MS |
516 | goto out_dput; |
517 | ||
518 | err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS); | |
e1ff3dd1 | 519 | if (err && err != -ENODATA && err != -EOPNOTSUPP) |
c11b9fdd MS |
520 | goto out_dput; |
521 | ||
522 | /* Clear any inherited mode bits */ | |
523 | inode_lock(work->d_inode); | |
524 | err = notify_change(work, &attr, NULL); | |
525 | inode_unlock(work->d_inode); | |
526 | if (err) | |
527 | goto out_dput; | |
6b8aa129 AG |
528 | } else { |
529 | err = PTR_ERR(work); | |
530 | goto out_err; | |
e9be9d5e MS |
531 | } |
532 | out_unlock: | |
e9be9d5e | 533 | mnt_drop_write(mnt); |
6b8aa129 AG |
534 | if (locked) |
535 | inode_unlock(dir); | |
e9be9d5e MS |
536 | |
537 | return work; | |
538 | ||
539 | out_dput: | |
540 | dput(work); | |
6b8aa129 AG |
541 | out_err: |
542 | pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n", | |
543 | ufs->config.workdir, name, -err); | |
544 | sb->s_flags |= MS_RDONLY; | |
545 | work = NULL; | |
e9be9d5e MS |
546 | goto out_unlock; |
547 | } | |
548 | ||
91c77947 MS |
549 | static void ovl_unescape(char *s) |
550 | { | |
551 | char *d = s; | |
552 | ||
553 | for (;; s++, d++) { | |
554 | if (*s == '\\') | |
555 | s++; | |
556 | *d = *s; | |
557 | if (!*s) | |
558 | break; | |
559 | } | |
560 | } | |
561 | ||
ab508822 MS |
562 | static int ovl_mount_dir_noesc(const char *name, struct path *path) |
563 | { | |
a78d9f0d | 564 | int err = -EINVAL; |
ab508822 | 565 | |
a78d9f0d MS |
566 | if (!*name) { |
567 | pr_err("overlayfs: empty lowerdir\n"); | |
568 | goto out; | |
569 | } | |
ab508822 MS |
570 | err = kern_path(name, LOOKUP_FOLLOW, path); |
571 | if (err) { | |
572 | pr_err("overlayfs: failed to resolve '%s': %i\n", name, err); | |
573 | goto out; | |
574 | } | |
575 | err = -EINVAL; | |
7c03b5d4 | 576 | if (ovl_dentry_weird(path->dentry)) { |
ab508822 MS |
577 | pr_err("overlayfs: filesystem on '%s' not supported\n", name); |
578 | goto out_put; | |
579 | } | |
2b8c30e9 | 580 | if (!d_is_dir(path->dentry)) { |
ab508822 MS |
581 | pr_err("overlayfs: '%s' not a directory\n", name); |
582 | goto out_put; | |
583 | } | |
584 | return 0; | |
585 | ||
586 | out_put: | |
587 | path_put(path); | |
588 | out: | |
589 | return err; | |
590 | } | |
591 | ||
592 | static int ovl_mount_dir(const char *name, struct path *path) | |
593 | { | |
594 | int err = -ENOMEM; | |
595 | char *tmp = kstrdup(name, GFP_KERNEL); | |
596 | ||
597 | if (tmp) { | |
598 | ovl_unescape(tmp); | |
599 | err = ovl_mount_dir_noesc(tmp, path); | |
7c03b5d4 MS |
600 | |
601 | if (!err) | |
602 | if (ovl_dentry_remote(path->dentry)) { | |
603 | pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n", | |
604 | tmp); | |
605 | path_put(path); | |
606 | err = -EINVAL; | |
607 | } | |
ab508822 MS |
608 | kfree(tmp); |
609 | } | |
610 | return err; | |
611 | } | |
612 | ||
6b2d5fe4 MS |
613 | static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs, |
614 | const char *name) | |
ab508822 | 615 | { |
ab508822 | 616 | struct kstatfs statfs; |
6b2d5fe4 MS |
617 | int err = vfs_statfs(path, &statfs); |
618 | ||
619 | if (err) | |
620 | pr_err("overlayfs: statfs failed on '%s'\n", name); | |
621 | else | |
622 | ofs->namelen = max(ofs->namelen, statfs.f_namelen); | |
623 | ||
624 | return err; | |
625 | } | |
626 | ||
627 | static int ovl_lower_dir(const char *name, struct path *path, | |
628 | struct ovl_fs *ofs, int *stack_depth, bool *remote) | |
629 | { | |
630 | int err; | |
ab508822 | 631 | |
a78d9f0d | 632 | err = ovl_mount_dir_noesc(name, path); |
ab508822 MS |
633 | if (err) |
634 | goto out; | |
635 | ||
6b2d5fe4 MS |
636 | err = ovl_check_namelen(path, ofs, name); |
637 | if (err) | |
ab508822 | 638 | goto out_put; |
6b2d5fe4 | 639 | |
ab508822 MS |
640 | *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth); |
641 | ||
7c03b5d4 MS |
642 | if (ovl_dentry_remote(path->dentry)) |
643 | *remote = true; | |
644 | ||
02bcd157 AG |
645 | /* |
646 | * The inodes index feature needs to encode and decode file | |
647 | * handles, so it requires that all layers support them. | |
648 | */ | |
649 | if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) { | |
650 | ofs->config.index = false; | |
651 | pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name); | |
652 | } | |
653 | ||
ab508822 MS |
654 | return 0; |
655 | ||
656 | out_put: | |
657 | path_put(path); | |
658 | out: | |
659 | return err; | |
660 | } | |
661 | ||
e9be9d5e MS |
662 | /* Workdir should not be subdir of upperdir and vice versa */ |
663 | static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir) | |
664 | { | |
665 | bool ok = false; | |
666 | ||
667 | if (workdir != upperdir) { | |
668 | ok = (lock_rename(workdir, upperdir) == NULL); | |
669 | unlock_rename(workdir, upperdir); | |
670 | } | |
671 | return ok; | |
672 | } | |
673 | ||
a78d9f0d MS |
674 | static unsigned int ovl_split_lowerdirs(char *str) |
675 | { | |
676 | unsigned int ctr = 1; | |
677 | char *s, *d; | |
678 | ||
679 | for (s = d = str;; s++, d++) { | |
680 | if (*s == '\\') { | |
681 | s++; | |
682 | } else if (*s == ':') { | |
683 | *d = '\0'; | |
684 | ctr++; | |
685 | continue; | |
686 | } | |
687 | *d = *s; | |
688 | if (!*s) | |
689 | break; | |
690 | } | |
691 | return ctr; | |
692 | } | |
693 | ||
0eb45fc3 AG |
694 | static int __maybe_unused |
695 | ovl_posix_acl_xattr_get(const struct xattr_handler *handler, | |
696 | struct dentry *dentry, struct inode *inode, | |
697 | const char *name, void *buffer, size_t size) | |
698 | { | |
1d88f183 | 699 | return ovl_xattr_get(dentry, inode, handler->name, buffer, size); |
0eb45fc3 AG |
700 | } |
701 | ||
0c97be22 AG |
702 | static int __maybe_unused |
703 | ovl_posix_acl_xattr_set(const struct xattr_handler *handler, | |
704 | struct dentry *dentry, struct inode *inode, | |
705 | const char *name, const void *value, | |
706 | size_t size, int flags) | |
d837a49b MS |
707 | { |
708 | struct dentry *workdir = ovl_workdir(dentry); | |
09d8b586 | 709 | struct inode *realinode = ovl_inode_real(inode); |
d837a49b MS |
710 | struct posix_acl *acl = NULL; |
711 | int err; | |
712 | ||
713 | /* Check that everything is OK before copy-up */ | |
714 | if (value) { | |
715 | acl = posix_acl_from_xattr(&init_user_ns, value, size); | |
716 | if (IS_ERR(acl)) | |
717 | return PTR_ERR(acl); | |
718 | } | |
719 | err = -EOPNOTSUPP; | |
720 | if (!IS_POSIXACL(d_inode(workdir))) | |
721 | goto out_acl_release; | |
722 | if (!realinode->i_op->set_acl) | |
723 | goto out_acl_release; | |
724 | if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) { | |
725 | err = acl ? -EACCES : 0; | |
726 | goto out_acl_release; | |
727 | } | |
728 | err = -EPERM; | |
729 | if (!inode_owner_or_capable(inode)) | |
730 | goto out_acl_release; | |
731 | ||
732 | posix_acl_release(acl); | |
733 | ||
fd3220d3 MS |
734 | /* |
735 | * Check if sgid bit needs to be cleared (actual setacl operation will | |
736 | * be done with mounter's capabilities and so that won't do it for us). | |
737 | */ | |
738 | if (unlikely(inode->i_mode & S_ISGID) && | |
739 | handler->flags == ACL_TYPE_ACCESS && | |
740 | !in_group_p(inode->i_gid) && | |
741 | !capable_wrt_inode_uidgid(inode, CAP_FSETID)) { | |
742 | struct iattr iattr = { .ia_valid = ATTR_KILL_SGID }; | |
743 | ||
744 | err = ovl_setattr(dentry, &iattr); | |
745 | if (err) | |
746 | return err; | |
747 | } | |
748 | ||
1d88f183 | 749 | err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags); |
ce31513a | 750 | if (!err) |
09d8b586 | 751 | ovl_copyattr(ovl_inode_real(inode), inode); |
ce31513a MS |
752 | |
753 | return err; | |
d837a49b MS |
754 | |
755 | out_acl_release: | |
756 | posix_acl_release(acl); | |
757 | return err; | |
758 | } | |
759 | ||
0eb45fc3 AG |
760 | static int ovl_own_xattr_get(const struct xattr_handler *handler, |
761 | struct dentry *dentry, struct inode *inode, | |
762 | const char *name, void *buffer, size_t size) | |
763 | { | |
48fab5d7 | 764 | return -EOPNOTSUPP; |
0eb45fc3 AG |
765 | } |
766 | ||
d837a49b MS |
767 | static int ovl_own_xattr_set(const struct xattr_handler *handler, |
768 | struct dentry *dentry, struct inode *inode, | |
769 | const char *name, const void *value, | |
770 | size_t size, int flags) | |
771 | { | |
48fab5d7 | 772 | return -EOPNOTSUPP; |
d837a49b MS |
773 | } |
774 | ||
0eb45fc3 AG |
775 | static int ovl_other_xattr_get(const struct xattr_handler *handler, |
776 | struct dentry *dentry, struct inode *inode, | |
777 | const char *name, void *buffer, size_t size) | |
778 | { | |
1d88f183 | 779 | return ovl_xattr_get(dentry, inode, name, buffer, size); |
0eb45fc3 AG |
780 | } |
781 | ||
0e585ccc AG |
782 | static int ovl_other_xattr_set(const struct xattr_handler *handler, |
783 | struct dentry *dentry, struct inode *inode, | |
784 | const char *name, const void *value, | |
785 | size_t size, int flags) | |
786 | { | |
1d88f183 | 787 | return ovl_xattr_set(dentry, inode, name, value, size, flags); |
0e585ccc AG |
788 | } |
789 | ||
0c97be22 AG |
790 | static const struct xattr_handler __maybe_unused |
791 | ovl_posix_acl_access_xattr_handler = { | |
d837a49b MS |
792 | .name = XATTR_NAME_POSIX_ACL_ACCESS, |
793 | .flags = ACL_TYPE_ACCESS, | |
0eb45fc3 | 794 | .get = ovl_posix_acl_xattr_get, |
d837a49b MS |
795 | .set = ovl_posix_acl_xattr_set, |
796 | }; | |
797 | ||
0c97be22 AG |
798 | static const struct xattr_handler __maybe_unused |
799 | ovl_posix_acl_default_xattr_handler = { | |
d837a49b MS |
800 | .name = XATTR_NAME_POSIX_ACL_DEFAULT, |
801 | .flags = ACL_TYPE_DEFAULT, | |
0eb45fc3 | 802 | .get = ovl_posix_acl_xattr_get, |
d837a49b MS |
803 | .set = ovl_posix_acl_xattr_set, |
804 | }; | |
805 | ||
806 | static const struct xattr_handler ovl_own_xattr_handler = { | |
807 | .prefix = OVL_XATTR_PREFIX, | |
0eb45fc3 | 808 | .get = ovl_own_xattr_get, |
d837a49b MS |
809 | .set = ovl_own_xattr_set, |
810 | }; | |
811 | ||
812 | static const struct xattr_handler ovl_other_xattr_handler = { | |
813 | .prefix = "", /* catch all */ | |
0eb45fc3 | 814 | .get = ovl_other_xattr_get, |
d837a49b MS |
815 | .set = ovl_other_xattr_set, |
816 | }; | |
817 | ||
818 | static const struct xattr_handler *ovl_xattr_handlers[] = { | |
0c97be22 | 819 | #ifdef CONFIG_FS_POSIX_ACL |
d837a49b MS |
820 | &ovl_posix_acl_access_xattr_handler, |
821 | &ovl_posix_acl_default_xattr_handler, | |
0c97be22 | 822 | #endif |
d837a49b MS |
823 | &ovl_own_xattr_handler, |
824 | &ovl_other_xattr_handler, | |
825 | NULL | |
826 | }; | |
827 | ||
e9be9d5e MS |
828 | static int ovl_fill_super(struct super_block *sb, void *data, int silent) |
829 | { | |
33006cdf KC |
830 | struct path upperpath = { }; |
831 | struct path workpath = { }; | |
e9be9d5e MS |
832 | struct dentry *root_dentry; |
833 | struct ovl_entry *oe; | |
834 | struct ovl_fs *ufs; | |
a78d9f0d MS |
835 | struct path *stack = NULL; |
836 | char *lowertmp; | |
837 | char *lower; | |
838 | unsigned int numlower; | |
839 | unsigned int stacklen = 0; | |
dd662667 | 840 | unsigned int i; |
7c03b5d4 | 841 | bool remote = false; |
51f8f3c4 | 842 | struct cred *cred; |
e9be9d5e MS |
843 | int err; |
844 | ||
f45827e8 EZ |
845 | err = -ENOMEM; |
846 | ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL); | |
847 | if (!ufs) | |
e9be9d5e MS |
848 | goto out; |
849 | ||
688ea0e5 | 850 | ufs->config.redirect_dir = ovl_redirect_dir_def; |
02bcd157 | 851 | ufs->config.index = ovl_index_def; |
f45827e8 EZ |
852 | err = ovl_parse_opt((char *) data, &ufs->config); |
853 | if (err) | |
854 | goto out_free_config; | |
855 | ||
e9be9d5e | 856 | err = -EINVAL; |
53a08cb9 | 857 | if (!ufs->config.lowerdir) { |
07f2af7b KK |
858 | if (!silent) |
859 | pr_err("overlayfs: missing 'lowerdir'\n"); | |
e9be9d5e MS |
860 | goto out_free_config; |
861 | } | |
862 | ||
53a08cb9 | 863 | sb->s_stack_depth = 0; |
cf9a6784 | 864 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
53a08cb9 | 865 | if (ufs->config.upperdir) { |
53a08cb9 MS |
866 | if (!ufs->config.workdir) { |
867 | pr_err("overlayfs: missing 'workdir'\n"); | |
868 | goto out_free_config; | |
869 | } | |
e9be9d5e | 870 | |
53a08cb9 MS |
871 | err = ovl_mount_dir(ufs->config.upperdir, &upperpath); |
872 | if (err) | |
873 | goto out_free_config; | |
e9be9d5e | 874 | |
71cbad7e | 875 | /* Upper fs should not be r/o */ |
876 | if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) { | |
877 | pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n"); | |
878 | err = -EINVAL; | |
879 | goto out_put_upperpath; | |
880 | } | |
881 | ||
6b2d5fe4 MS |
882 | err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir); |
883 | if (err) | |
884 | goto out_put_upperpath; | |
885 | ||
2cac0c00 | 886 | err = -EBUSY; |
cd499a84 AG |
887 | if (ovl_inuse_trylock(upperpath.dentry)) { |
888 | ufs->upperdir_locked = true; | |
889 | } else if (ufs->config.index) { | |
890 | pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n"); | |
2cac0c00 | 891 | goto out_put_upperpath; |
cd499a84 AG |
892 | } else { |
893 | pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n"); | |
2cac0c00 AG |
894 | } |
895 | ||
53a08cb9 MS |
896 | err = ovl_mount_dir(ufs->config.workdir, &workpath); |
897 | if (err) | |
2cac0c00 | 898 | goto out_unlock_upperdentry; |
53a08cb9 | 899 | |
2f83fd8c | 900 | err = -EINVAL; |
53a08cb9 MS |
901 | if (upperpath.mnt != workpath.mnt) { |
902 | pr_err("overlayfs: workdir and upperdir must reside under the same mount\n"); | |
903 | goto out_put_workpath; | |
904 | } | |
905 | if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) { | |
906 | pr_err("overlayfs: workdir and upperdir must be separate subtrees\n"); | |
907 | goto out_put_workpath; | |
908 | } | |
2cac0c00 AG |
909 | |
910 | err = -EBUSY; | |
cd499a84 AG |
911 | if (ovl_inuse_trylock(workpath.dentry)) { |
912 | ufs->workdir_locked = true; | |
913 | } else if (ufs->config.index) { | |
914 | pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n"); | |
2cac0c00 | 915 | goto out_put_workpath; |
cd499a84 AG |
916 | } else { |
917 | pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n"); | |
2cac0c00 AG |
918 | } |
919 | ||
920 | ufs->workbasedir = workpath.dentry; | |
53a08cb9 | 921 | sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth; |
cc259639 | 922 | } |
a78d9f0d MS |
923 | err = -ENOMEM; |
924 | lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL); | |
925 | if (!lowertmp) | |
2cac0c00 | 926 | goto out_unlock_workdentry; |
69c433ed | 927 | |
a78d9f0d MS |
928 | err = -EINVAL; |
929 | stacklen = ovl_split_lowerdirs(lowertmp); | |
6be4506e | 930 | if (stacklen > OVL_MAX_STACK) { |
fd36570a | 931 | pr_err("overlayfs: too many lower directories, limit is %d\n", |
6be4506e | 932 | OVL_MAX_STACK); |
a78d9f0d | 933 | goto out_free_lowertmp; |
6be4506e | 934 | } else if (!ufs->config.upperdir && stacklen == 1) { |
935 | pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n"); | |
936 | goto out_free_lowertmp; | |
937 | } | |
a78d9f0d | 938 | |
313684c4 | 939 | err = -ENOMEM; |
a78d9f0d MS |
940 | stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL); |
941 | if (!stack) | |
942 | goto out_free_lowertmp; | |
943 | ||
313684c4 | 944 | err = -EINVAL; |
a78d9f0d MS |
945 | lower = lowertmp; |
946 | for (numlower = 0; numlower < stacklen; numlower++) { | |
6b2d5fe4 MS |
947 | err = ovl_lower_dir(lower, &stack[numlower], ufs, |
948 | &sb->s_stack_depth, &remote); | |
a78d9f0d MS |
949 | if (err) |
950 | goto out_put_lowerpath; | |
951 | ||
952 | lower = strchr(lower, '\0') + 1; | |
953 | } | |
954 | ||
69c433ed | 955 | err = -EINVAL; |
ab508822 | 956 | sb->s_stack_depth++; |
69c433ed MS |
957 | if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) { |
958 | pr_err("overlayfs: maximum fs stacking depth exceeded\n"); | |
3b7a9a24 | 959 | goto out_put_lowerpath; |
69c433ed MS |
960 | } |
961 | ||
53a08cb9 MS |
962 | if (ufs->config.upperdir) { |
963 | ufs->upper_mnt = clone_private_mount(&upperpath); | |
964 | err = PTR_ERR(ufs->upper_mnt); | |
965 | if (IS_ERR(ufs->upper_mnt)) { | |
966 | pr_err("overlayfs: failed to clone upperpath\n"); | |
967 | goto out_put_lowerpath; | |
968 | } | |
2cac0c00 | 969 | |
d719e8f2 MS |
970 | /* Don't inherit atime flags */ |
971 | ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME); | |
972 | ||
973 | sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran; | |
3b7a9a24 | 974 | |
0dc04886 SF |
975 | if (ufs->upper_mnt->mnt_flags & MNT_NOSUID) |
976 | sb->s_iflags |= SB_I_NOSUID; | |
977 | ||
6b8aa129 AG |
978 | ufs->workdir = ovl_workdir_create(sb, ufs, workpath.dentry, |
979 | OVL_WORKDIR_NAME, false); | |
45aebeaf VG |
980 | /* |
981 | * Upper should support d_type, else whiteouts are visible. | |
982 | * Given workdir and upper are on same fs, we can do | |
21765194 VG |
983 | * iterate_dir() on workdir. This check requires successful |
984 | * creation of workdir in previous step. | |
45aebeaf | 985 | */ |
21765194 | 986 | if (ufs->workdir) { |
e7f52429 AG |
987 | struct dentry *temp; |
988 | ||
21765194 VG |
989 | err = ovl_check_d_type_supported(&workpath); |
990 | if (err < 0) | |
991 | goto out_put_workdir; | |
45aebeaf | 992 | |
e7c0b599 VG |
993 | /* |
994 | * We allowed this configuration and don't want to | |
995 | * break users over kernel upgrade. So warn instead | |
996 | * of erroring out. | |
997 | */ | |
998 | if (!err) | |
999 | pr_warn("overlayfs: upper fs needs to support d_type.\n"); | |
e7f52429 AG |
1000 | |
1001 | /* Check if upper/work fs supports O_TMPFILE */ | |
1002 | temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0); | |
1003 | ufs->tmpfile = !IS_ERR(temp); | |
1004 | if (ufs->tmpfile) | |
1005 | dput(temp); | |
1006 | else | |
1007 | pr_warn("overlayfs: upper fs does not support tmpfile.\n"); | |
82b749b2 AG |
1008 | |
1009 | /* | |
1010 | * Check if upper/work fs supports trusted.overlay.* | |
1011 | * xattr | |
1012 | */ | |
1013 | err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE, | |
1014 | "0", 1, 0); | |
1015 | if (err) { | |
1016 | ufs->noxattr = true; | |
1017 | pr_warn("overlayfs: upper fs does not support xattr.\n"); | |
1018 | } else { | |
1019 | vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE); | |
1020 | } | |
02bcd157 AG |
1021 | |
1022 | /* Check if upper/work fs supports file handles */ | |
1023 | if (ufs->config.index && | |
1024 | !ovl_can_decode_fh(ufs->workdir->d_sb)) { | |
1025 | ufs->config.index = false; | |
1026 | pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n"); | |
1027 | } | |
45aebeaf | 1028 | } |
e9be9d5e MS |
1029 | } |
1030 | ||
2f83fd8c | 1031 | err = -ENOMEM; |
a78d9f0d | 1032 | ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL); |
dd662667 | 1033 | if (ufs->lower_mnt == NULL) |
3b7a9a24 | 1034 | goto out_put_workdir; |
a78d9f0d MS |
1035 | for (i = 0; i < numlower; i++) { |
1036 | struct vfsmount *mnt = clone_private_mount(&stack[i]); | |
dd662667 | 1037 | |
2f83fd8c | 1038 | err = PTR_ERR(mnt); |
a78d9f0d MS |
1039 | if (IS_ERR(mnt)) { |
1040 | pr_err("overlayfs: failed to clone lowerpath\n"); | |
1041 | goto out_put_lower_mnt; | |
1042 | } | |
1043 | /* | |
1044 | * Make lower_mnt R/O. That way fchmod/fchown on lower file | |
1045 | * will fail instead of modifying lower fs. | |
1046 | */ | |
d719e8f2 | 1047 | mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME; |
dd662667 | 1048 | |
0dc04886 SF |
1049 | if (mnt->mnt_flags & MNT_NOSUID) |
1050 | sb->s_iflags |= SB_I_NOSUID; | |
1051 | ||
a78d9f0d MS |
1052 | ufs->lower_mnt[ufs->numlower] = mnt; |
1053 | ufs->numlower++; | |
7bcd74b9 AG |
1054 | |
1055 | /* Check if all lower layers are on same sb */ | |
1056 | if (i == 0) | |
1057 | ufs->same_sb = mnt->mnt_sb; | |
1058 | else if (ufs->same_sb != mnt->mnt_sb) | |
1059 | ufs->same_sb = NULL; | |
a78d9f0d | 1060 | } |
e9be9d5e | 1061 | |
71cbad7e | 1062 | /* If the upper fs is nonexistent, we mark overlayfs r/o too */ |
1063 | if (!ufs->upper_mnt) | |
e9be9d5e | 1064 | sb->s_flags |= MS_RDONLY; |
7bcd74b9 AG |
1065 | else if (ufs->upper_mnt->mnt_sb != ufs->same_sb) |
1066 | ufs->same_sb = NULL; | |
e9be9d5e | 1067 | |
02bcd157 | 1068 | if (!(ovl_force_readonly(ufs)) && ufs->config.index) { |
8b88a2e6 AG |
1069 | /* Verify lower root is upper root origin */ |
1070 | err = ovl_verify_origin(upperpath.dentry, ufs->lower_mnt[0], | |
54fb347e | 1071 | stack[0].dentry, false, true); |
8b88a2e6 AG |
1072 | if (err) { |
1073 | pr_err("overlayfs: failed to verify upper root origin\n"); | |
1074 | goto out_put_lower_mnt; | |
1075 | } | |
1076 | ||
02bcd157 AG |
1077 | ufs->indexdir = ovl_workdir_create(sb, ufs, workpath.dentry, |
1078 | OVL_INDEXDIR_NAME, true); | |
54fb347e AG |
1079 | if (ufs->indexdir) { |
1080 | /* Verify upper root is index dir origin */ | |
1081 | err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt, | |
1082 | upperpath.dentry, true, true); | |
1083 | if (err) | |
1084 | pr_err("overlayfs: failed to verify index dir origin\n"); | |
415543d5 | 1085 | |
caf70cb2 | 1086 | /* Cleanup bad/stale/orphan index entries */ |
415543d5 AG |
1087 | if (!err) |
1088 | err = ovl_indexdir_cleanup(ufs->indexdir, | |
1089 | ufs->upper_mnt, | |
1090 | stack, numlower); | |
54fb347e AG |
1091 | } |
1092 | if (err || !ufs->indexdir) | |
02bcd157 | 1093 | pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n"); |
54fb347e AG |
1094 | if (err) |
1095 | goto out_put_indexdir; | |
02bcd157 AG |
1096 | } |
1097 | ||
1098 | /* Show index=off/on in /proc/mounts for any of the reasons above */ | |
1099 | if (!ufs->indexdir) | |
1100 | ufs->config.index = false; | |
1101 | ||
7c03b5d4 MS |
1102 | if (remote) |
1103 | sb->s_d_op = &ovl_reval_dentry_operations; | |
1104 | else | |
1105 | sb->s_d_op = &ovl_dentry_operations; | |
e9be9d5e | 1106 | |
8fc646b4 | 1107 | err = -ENOMEM; |
51f8f3c4 KK |
1108 | ufs->creator_cred = cred = prepare_creds(); |
1109 | if (!cred) | |
02bcd157 | 1110 | goto out_put_indexdir; |
3fe6e52f | 1111 | |
51f8f3c4 KK |
1112 | /* Never override disk quota limits or use reserved space */ |
1113 | cap_lower(cred->cap_effective, CAP_SYS_RESOURCE); | |
1114 | ||
e9be9d5e | 1115 | err = -ENOMEM; |
a78d9f0d | 1116 | oe = ovl_alloc_entry(numlower); |
3b7a9a24 | 1117 | if (!oe) |
3fe6e52f | 1118 | goto out_put_cred; |
e9be9d5e | 1119 | |
655042cc VG |
1120 | sb->s_magic = OVERLAYFS_SUPER_MAGIC; |
1121 | sb->s_op = &ovl_super_operations; | |
1122 | sb->s_xattr = ovl_xattr_handlers; | |
1123 | sb->s_fs_info = ufs; | |
1124 | sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK; | |
1125 | ||
ca4c8a3a | 1126 | root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0)); |
e9be9d5e | 1127 | if (!root_dentry) |
3b7a9a24 | 1128 | goto out_free_oe; |
e9be9d5e MS |
1129 | |
1130 | mntput(upperpath.mnt); | |
a78d9f0d MS |
1131 | for (i = 0; i < numlower; i++) |
1132 | mntput(stack[i].mnt); | |
2cac0c00 | 1133 | mntput(workpath.mnt); |
a78d9f0d | 1134 | kfree(lowertmp); |
e9be9d5e | 1135 | |
f3a15685 | 1136 | if (upperpath.dentry) { |
55acc661 | 1137 | oe->has_upper = true; |
13c72075 MS |
1138 | if (ovl_is_impuredir(upperpath.dentry)) |
1139 | ovl_set_flag(OVL_IMPURE, d_inode(root_dentry)); | |
f3a15685 | 1140 | } |
a78d9f0d MS |
1141 | for (i = 0; i < numlower; i++) { |
1142 | oe->lowerstack[i].dentry = stack[i].dentry; | |
1143 | oe->lowerstack[i].mnt = ufs->lower_mnt[i]; | |
1144 | } | |
0f95502a | 1145 | kfree(stack); |
e9be9d5e MS |
1146 | |
1147 | root_dentry->d_fsdata = oe; | |
1148 | ||
09d8b586 MS |
1149 | ovl_inode_init(d_inode(root_dentry), upperpath.dentry, |
1150 | ovl_dentry_lower(root_dentry)); | |
ed06e069 | 1151 | |
e9be9d5e | 1152 | sb->s_root = root_dentry; |
e9be9d5e MS |
1153 | |
1154 | return 0; | |
1155 | ||
3b7a9a24 MS |
1156 | out_free_oe: |
1157 | kfree(oe); | |
3fe6e52f AM |
1158 | out_put_cred: |
1159 | put_cred(ufs->creator_cred); | |
02bcd157 AG |
1160 | out_put_indexdir: |
1161 | dput(ufs->indexdir); | |
e9be9d5e | 1162 | out_put_lower_mnt: |
dd662667 MS |
1163 | for (i = 0; i < ufs->numlower; i++) |
1164 | mntput(ufs->lower_mnt[i]); | |
1165 | kfree(ufs->lower_mnt); | |
3b7a9a24 MS |
1166 | out_put_workdir: |
1167 | dput(ufs->workdir); | |
e9be9d5e | 1168 | mntput(ufs->upper_mnt); |
e9be9d5e | 1169 | out_put_lowerpath: |
a78d9f0d MS |
1170 | for (i = 0; i < numlower; i++) |
1171 | path_put(&stack[i]); | |
1172 | kfree(stack); | |
1173 | out_free_lowertmp: | |
1174 | kfree(lowertmp); | |
2cac0c00 | 1175 | out_unlock_workdentry: |
cd499a84 AG |
1176 | if (ufs->workdir_locked) |
1177 | ovl_inuse_unlock(workpath.dentry); | |
3b7a9a24 MS |
1178 | out_put_workpath: |
1179 | path_put(&workpath); | |
2cac0c00 | 1180 | out_unlock_upperdentry: |
cd499a84 AG |
1181 | if (ufs->upperdir_locked) |
1182 | ovl_inuse_unlock(upperpath.dentry); | |
e9be9d5e MS |
1183 | out_put_upperpath: |
1184 | path_put(&upperpath); | |
e9be9d5e | 1185 | out_free_config: |
f45827e8 EZ |
1186 | kfree(ufs->config.lowerdir); |
1187 | kfree(ufs->config.upperdir); | |
1188 | kfree(ufs->config.workdir); | |
1189 | kfree(ufs); | |
e9be9d5e MS |
1190 | out: |
1191 | return err; | |
1192 | } | |
1193 | ||
1194 | static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags, | |
1195 | const char *dev_name, void *raw_data) | |
1196 | { | |
1197 | return mount_nodev(fs_type, flags, raw_data, ovl_fill_super); | |
1198 | } | |
1199 | ||
1200 | static struct file_system_type ovl_fs_type = { | |
1201 | .owner = THIS_MODULE, | |
ef94b186 | 1202 | .name = "overlay", |
e9be9d5e MS |
1203 | .mount = ovl_mount, |
1204 | .kill_sb = kill_anon_super, | |
0a414bdc | 1205 | .fs_flags = FS_USERNS_MOUNT, |
e9be9d5e | 1206 | }; |
ef94b186 | 1207 | MODULE_ALIAS_FS("overlay"); |
e9be9d5e | 1208 | |
13cf199d AG |
1209 | static void ovl_inode_init_once(void *foo) |
1210 | { | |
1211 | struct ovl_inode *oi = foo; | |
1212 | ||
1213 | inode_init_once(&oi->vfs_inode); | |
1214 | } | |
1215 | ||
e9be9d5e MS |
1216 | static int __init ovl_init(void) |
1217 | { | |
13cf199d AG |
1218 | int err; |
1219 | ||
1220 | ovl_inode_cachep = kmem_cache_create("ovl_inode", | |
1221 | sizeof(struct ovl_inode), 0, | |
1222 | (SLAB_RECLAIM_ACCOUNT| | |
1223 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), | |
1224 | ovl_inode_init_once); | |
1225 | if (ovl_inode_cachep == NULL) | |
1226 | return -ENOMEM; | |
1227 | ||
1228 | err = register_filesystem(&ovl_fs_type); | |
1229 | if (err) | |
1230 | kmem_cache_destroy(ovl_inode_cachep); | |
1231 | ||
1232 | return err; | |
e9be9d5e MS |
1233 | } |
1234 | ||
1235 | static void __exit ovl_exit(void) | |
1236 | { | |
1237 | unregister_filesystem(&ovl_fs_type); | |
13cf199d AG |
1238 | |
1239 | /* | |
1240 | * Make sure all delayed rcu free inodes are flushed before we | |
1241 | * destroy cache. | |
1242 | */ | |
1243 | rcu_barrier(); | |
1244 | kmem_cache_destroy(ovl_inode_cachep); | |
1245 | ||
e9be9d5e MS |
1246 | } |
1247 | ||
1248 | module_init(ovl_init); | |
1249 | module_exit(ovl_exit); |