]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - fs/overlayfs/super.c
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
UBUNTU: Ubuntu-5.3.0-29.31
[mirror_ubuntu-eoan-kernel.git] / fs / overlayfs / super.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 *
4 * Copyright (C) 2011 Novell Inc.
5 */
6
7 #include <uapi/linux/magic.h>
8 #include <linux/fs.h>
9 #include <linux/namei.h>
10 #include <linux/xattr.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/module.h>
14 #include <linux/statfs.h>
15 #include <linux/seq_file.h>
16 #include <linux/posix_acl_xattr.h>
17 #include <linux/exportfs.h>
18 #include "overlayfs.h"
19
20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
21 MODULE_DESCRIPTION("Overlay filesystem");
22 MODULE_LICENSE("GPL");
23
24
25 struct ovl_dir_cache;
26
27 #define OVL_MAX_STACK 500
28
29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
31 MODULE_PARM_DESC(redirect_dir,
32 "Default to on or off for the redirect_dir feature");
33
34 static bool ovl_redirect_always_follow =
35 IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
37 bool, 0644);
38 MODULE_PARM_DESC(redirect_always_follow,
39 "Follow redirects even if redirect_dir feature is turned off");
40
41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
42 module_param_named(index, ovl_index_def, bool, 0644);
43 MODULE_PARM_DESC(index,
44 "Default to on or off for the inodes index feature");
45
46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
48 MODULE_PARM_DESC(nfs_export,
49 "Default to on or off for the NFS export feature");
50
51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
53 MODULE_PARM_DESC(xino_auto,
54 "Auto enable xino feature");
55
56 static void ovl_entry_stack_free(struct ovl_entry *oe)
57 {
58 unsigned int i;
59
60 for (i = 0; i < oe->numlower; i++)
61 dput(oe->lowerstack[i].dentry);
62 }
63
64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
66 MODULE_PARM_DESC(metacopy,
67 "Default to on or off for the metadata only copy up feature");
68
69 static void ovl_dentry_release(struct dentry *dentry)
70 {
71 struct ovl_entry *oe = dentry->d_fsdata;
72
73 if (oe) {
74 ovl_entry_stack_free(oe);
75 kfree_rcu(oe, rcu);
76 }
77 }
78
79 static struct dentry *ovl_d_real(struct dentry *dentry,
80 const struct inode *inode)
81 {
82 struct dentry *real;
83
84 /* It's an overlay file */
85 if (inode && d_inode(dentry) == inode)
86 return dentry;
87
88 if (!d_is_reg(dentry)) {
89 if (!inode || inode == d_inode(dentry))
90 return dentry;
91 goto bug;
92 }
93
94 real = ovl_dentry_upper(dentry);
95 if (real && (inode == d_inode(real)))
96 return real;
97
98 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
99 return real;
100
101 real = ovl_dentry_lowerdata(dentry);
102 if (!real)
103 goto bug;
104
105 /* Handle recursion */
106 real = d_real(real, inode);
107
108 if (!inode || inode == d_inode(real))
109 return real;
110 bug:
111 WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
112 inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
113 return dentry;
114 }
115
116 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
117 {
118 struct ovl_entry *oe = dentry->d_fsdata;
119 unsigned int i;
120 int ret = 1;
121
122 for (i = 0; i < oe->numlower; i++) {
123 struct dentry *d = oe->lowerstack[i].dentry;
124
125 if (d->d_flags & DCACHE_OP_REVALIDATE) {
126 ret = d->d_op->d_revalidate(d, flags);
127 if (ret < 0)
128 return ret;
129 if (!ret) {
130 if (!(flags & LOOKUP_RCU))
131 d_invalidate(d);
132 return -ESTALE;
133 }
134 }
135 }
136 return 1;
137 }
138
139 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
140 {
141 struct ovl_entry *oe = dentry->d_fsdata;
142 unsigned int i;
143 int ret = 1;
144
145 for (i = 0; i < oe->numlower; i++) {
146 struct dentry *d = oe->lowerstack[i].dentry;
147
148 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
149 ret = d->d_op->d_weak_revalidate(d, flags);
150 if (ret <= 0)
151 break;
152 }
153 }
154 return ret;
155 }
156
157 static const struct dentry_operations ovl_dentry_operations = {
158 .d_release = ovl_dentry_release,
159 .d_real = ovl_d_real,
160 };
161
162 static const struct dentry_operations ovl_reval_dentry_operations = {
163 .d_release = ovl_dentry_release,
164 .d_real = ovl_d_real,
165 .d_revalidate = ovl_dentry_revalidate,
166 .d_weak_revalidate = ovl_dentry_weak_revalidate,
167 };
168
169 static struct kmem_cache *ovl_inode_cachep;
170
171 static struct inode *ovl_alloc_inode(struct super_block *sb)
172 {
173 struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
174
175 if (!oi)
176 return NULL;
177
178 oi->cache = NULL;
179 oi->redirect = NULL;
180 oi->version = 0;
181 oi->flags = 0;
182 oi->__upperdentry = NULL;
183 oi->lower = NULL;
184 oi->lowerdata = NULL;
185 mutex_init(&oi->lock);
186
187 return &oi->vfs_inode;
188 }
189
190 static void ovl_free_inode(struct inode *inode)
191 {
192 struct ovl_inode *oi = OVL_I(inode);
193
194 kfree(oi->redirect);
195 mutex_destroy(&oi->lock);
196 kmem_cache_free(ovl_inode_cachep, oi);
197 }
198
199 static void ovl_destroy_inode(struct inode *inode)
200 {
201 struct ovl_inode *oi = OVL_I(inode);
202
203 dput(oi->__upperdentry);
204 iput(oi->lower);
205 if (S_ISDIR(inode->i_mode))
206 ovl_dir_cache_free(inode);
207 else
208 iput(oi->lowerdata);
209 }
210
211 static void ovl_free_fs(struct ovl_fs *ofs)
212 {
213 unsigned i;
214
215 iput(ofs->workbasedir_trap);
216 iput(ofs->indexdir_trap);
217 iput(ofs->workdir_trap);
218 iput(ofs->upperdir_trap);
219 dput(ofs->indexdir);
220 dput(ofs->workdir);
221 if (ofs->workdir_locked)
222 ovl_inuse_unlock(ofs->workbasedir);
223 dput(ofs->workbasedir);
224 if (ofs->upperdir_locked)
225 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
226 mntput(ofs->upper_mnt);
227 for (i = 0; i < ofs->numlower; i++) {
228 iput(ofs->lower_layers[i].trap);
229 mntput(ofs->lower_layers[i].mnt);
230 }
231 for (i = 0; i < ofs->numlowerfs; i++)
232 free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
233 kfree(ofs->lower_layers);
234 kfree(ofs->lower_fs);
235
236 kfree(ofs->config.lowerdir);
237 kfree(ofs->config.upperdir);
238 kfree(ofs->config.workdir);
239 kfree(ofs->config.redirect_mode);
240 if (ofs->creator_cred)
241 put_cred(ofs->creator_cred);
242 kfree(ofs);
243 }
244
245 static void ovl_put_super(struct super_block *sb)
246 {
247 struct ovl_fs *ofs = sb->s_fs_info;
248
249 ovl_free_fs(ofs);
250 }
251
252 /* Sync real dirty inodes in upper filesystem (if it exists) */
253 static int ovl_sync_fs(struct super_block *sb, int wait)
254 {
255 struct ovl_fs *ofs = sb->s_fs_info;
256 struct super_block *upper_sb;
257 int ret;
258
259 if (!ofs->upper_mnt)
260 return 0;
261
262 /*
263 * If this is a sync(2) call or an emergency sync, all the super blocks
264 * will be iterated, including upper_sb, so no need to do anything.
265 *
266 * If this is a syncfs(2) call, then we do need to call
267 * sync_filesystem() on upper_sb, but enough if we do it when being
268 * called with wait == 1.
269 */
270 if (!wait)
271 return 0;
272
273 upper_sb = ofs->upper_mnt->mnt_sb;
274
275 down_read(&upper_sb->s_umount);
276 ret = sync_filesystem(upper_sb);
277 up_read(&upper_sb->s_umount);
278
279 return ret;
280 }
281
282 /**
283 * ovl_statfs
284 * @sb: The overlayfs super block
285 * @buf: The struct kstatfs to fill in with stats
286 *
287 * Get the filesystem statistics. As writes always target the upper layer
288 * filesystem pass the statfs to the upper filesystem (if it exists)
289 */
290 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
291 {
292 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
293 struct dentry *root_dentry = dentry->d_sb->s_root;
294 struct path path;
295 int err;
296
297 ovl_path_real(root_dentry, &path);
298
299 err = vfs_statfs(&path, buf);
300 if (!err) {
301 buf->f_namelen = ofs->namelen;
302 buf->f_type = OVERLAYFS_SUPER_MAGIC;
303 }
304
305 return err;
306 }
307
308 /* Will this overlay be forced to mount/remount ro? */
309 static bool ovl_force_readonly(struct ovl_fs *ofs)
310 {
311 return (!ofs->upper_mnt || !ofs->workdir);
312 }
313
314 static const char *ovl_redirect_mode_def(void)
315 {
316 return ovl_redirect_dir_def ? "on" : "off";
317 }
318
319 enum {
320 OVL_XINO_OFF,
321 OVL_XINO_AUTO,
322 OVL_XINO_ON,
323 };
324
325 static const char * const ovl_xino_str[] = {
326 "off",
327 "auto",
328 "on",
329 };
330
331 static inline int ovl_xino_def(void)
332 {
333 return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
334 }
335
336 /**
337 * ovl_show_options
338 *
339 * Prints the mount options for a given superblock.
340 * Returns zero; does not fail.
341 */
342 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
343 {
344 struct super_block *sb = dentry->d_sb;
345 struct ovl_fs *ofs = sb->s_fs_info;
346
347 seq_show_option(m, "lowerdir", ofs->config.lowerdir);
348 if (ofs->config.upperdir) {
349 seq_show_option(m, "upperdir", ofs->config.upperdir);
350 seq_show_option(m, "workdir", ofs->config.workdir);
351 }
352 if (ofs->config.default_permissions)
353 seq_puts(m, ",default_permissions");
354 if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
355 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
356 if (ofs->config.index != ovl_index_def)
357 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
358 if (ofs->config.nfs_export != ovl_nfs_export_def)
359 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
360 "on" : "off");
361 if (ofs->config.xino != ovl_xino_def())
362 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
363 if (ofs->config.metacopy != ovl_metacopy_def)
364 seq_printf(m, ",metacopy=%s",
365 ofs->config.metacopy ? "on" : "off");
366 return 0;
367 }
368
369 static int ovl_remount(struct super_block *sb, int *flags, char *data)
370 {
371 struct ovl_fs *ofs = sb->s_fs_info;
372
373 if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
374 return -EROFS;
375
376 return 0;
377 }
378
379 static const struct super_operations ovl_super_operations = {
380 .alloc_inode = ovl_alloc_inode,
381 .free_inode = ovl_free_inode,
382 .destroy_inode = ovl_destroy_inode,
383 .drop_inode = generic_delete_inode,
384 .put_super = ovl_put_super,
385 .sync_fs = ovl_sync_fs,
386 .statfs = ovl_statfs,
387 .show_options = ovl_show_options,
388 .remount_fs = ovl_remount,
389 };
390
391 enum {
392 OPT_LOWERDIR,
393 OPT_UPPERDIR,
394 OPT_WORKDIR,
395 OPT_DEFAULT_PERMISSIONS,
396 OPT_REDIRECT_DIR,
397 OPT_INDEX_ON,
398 OPT_INDEX_OFF,
399 OPT_NFS_EXPORT_ON,
400 OPT_NFS_EXPORT_OFF,
401 OPT_XINO_ON,
402 OPT_XINO_OFF,
403 OPT_XINO_AUTO,
404 OPT_METACOPY_ON,
405 OPT_METACOPY_OFF,
406 OPT_ERR,
407 };
408
409 static const match_table_t ovl_tokens = {
410 {OPT_LOWERDIR, "lowerdir=%s"},
411 {OPT_UPPERDIR, "upperdir=%s"},
412 {OPT_WORKDIR, "workdir=%s"},
413 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
414 {OPT_REDIRECT_DIR, "redirect_dir=%s"},
415 {OPT_INDEX_ON, "index=on"},
416 {OPT_INDEX_OFF, "index=off"},
417 {OPT_NFS_EXPORT_ON, "nfs_export=on"},
418 {OPT_NFS_EXPORT_OFF, "nfs_export=off"},
419 {OPT_XINO_ON, "xino=on"},
420 {OPT_XINO_OFF, "xino=off"},
421 {OPT_XINO_AUTO, "xino=auto"},
422 {OPT_METACOPY_ON, "metacopy=on"},
423 {OPT_METACOPY_OFF, "metacopy=off"},
424 {OPT_ERR, NULL}
425 };
426
427 static char *ovl_next_opt(char **s)
428 {
429 char *sbegin = *s;
430 char *p;
431
432 if (sbegin == NULL)
433 return NULL;
434
435 for (p = sbegin; *p; p++) {
436 if (*p == '\\') {
437 p++;
438 if (!*p)
439 break;
440 } else if (*p == ',') {
441 *p = '\0';
442 *s = p + 1;
443 return sbegin;
444 }
445 }
446 *s = NULL;
447 return sbegin;
448 }
449
450 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
451 {
452 if (strcmp(mode, "on") == 0) {
453 config->redirect_dir = true;
454 /*
455 * Does not make sense to have redirect creation without
456 * redirect following.
457 */
458 config->redirect_follow = true;
459 } else if (strcmp(mode, "follow") == 0) {
460 config->redirect_follow = true;
461 } else if (strcmp(mode, "off") == 0) {
462 if (ovl_redirect_always_follow)
463 config->redirect_follow = true;
464 } else if (strcmp(mode, "nofollow") != 0) {
465 pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
466 mode);
467 return -EINVAL;
468 }
469
470 return 0;
471 }
472
473 static int ovl_parse_opt(char *opt, struct ovl_config *config)
474 {
475 char *p;
476 int err;
477 bool metacopy_opt = false, redirect_opt = false;
478
479 config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
480 if (!config->redirect_mode)
481 return -ENOMEM;
482
483 while ((p = ovl_next_opt(&opt)) != NULL) {
484 int token;
485 substring_t args[MAX_OPT_ARGS];
486
487 if (!*p)
488 continue;
489
490 token = match_token(p, ovl_tokens, args);
491 switch (token) {
492 case OPT_UPPERDIR:
493 kfree(config->upperdir);
494 config->upperdir = match_strdup(&args[0]);
495 if (!config->upperdir)
496 return -ENOMEM;
497 break;
498
499 case OPT_LOWERDIR:
500 kfree(config->lowerdir);
501 config->lowerdir = match_strdup(&args[0]);
502 if (!config->lowerdir)
503 return -ENOMEM;
504 break;
505
506 case OPT_WORKDIR:
507 kfree(config->workdir);
508 config->workdir = match_strdup(&args[0]);
509 if (!config->workdir)
510 return -ENOMEM;
511 break;
512
513 case OPT_DEFAULT_PERMISSIONS:
514 config->default_permissions = true;
515 break;
516
517 case OPT_REDIRECT_DIR:
518 kfree(config->redirect_mode);
519 config->redirect_mode = match_strdup(&args[0]);
520 if (!config->redirect_mode)
521 return -ENOMEM;
522 redirect_opt = true;
523 break;
524
525 case OPT_INDEX_ON:
526 config->index = true;
527 break;
528
529 case OPT_INDEX_OFF:
530 config->index = false;
531 break;
532
533 case OPT_NFS_EXPORT_ON:
534 config->nfs_export = true;
535 break;
536
537 case OPT_NFS_EXPORT_OFF:
538 config->nfs_export = false;
539 break;
540
541 case OPT_XINO_ON:
542 config->xino = OVL_XINO_ON;
543 break;
544
545 case OPT_XINO_OFF:
546 config->xino = OVL_XINO_OFF;
547 break;
548
549 case OPT_XINO_AUTO:
550 config->xino = OVL_XINO_AUTO;
551 break;
552
553 case OPT_METACOPY_ON:
554 config->metacopy = true;
555 metacopy_opt = true;
556 break;
557
558 case OPT_METACOPY_OFF:
559 config->metacopy = false;
560 break;
561
562 default:
563 pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
564 return -EINVAL;
565 }
566 }
567
568 /* Workdir is useless in non-upper mount */
569 if (!config->upperdir && config->workdir) {
570 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
571 config->workdir);
572 kfree(config->workdir);
573 config->workdir = NULL;
574 }
575
576 err = ovl_parse_redirect_mode(config, config->redirect_mode);
577 if (err)
578 return err;
579
580 /*
581 * This is to make the logic below simpler. It doesn't make any other
582 * difference, since config->redirect_dir is only used for upper.
583 */
584 if (!config->upperdir && config->redirect_follow)
585 config->redirect_dir = true;
586
587 /* Resolve metacopy -> redirect_dir dependency */
588 if (config->metacopy && !config->redirect_dir) {
589 if (metacopy_opt && redirect_opt) {
590 pr_err("overlayfs: conflicting options: metacopy=on,redirect_dir=%s\n",
591 config->redirect_mode);
592 return -EINVAL;
593 }
594 if (redirect_opt) {
595 /*
596 * There was an explicit redirect_dir=... that resulted
597 * in this conflict.
598 */
599 pr_info("overlayfs: disabling metacopy due to redirect_dir=%s\n",
600 config->redirect_mode);
601 config->metacopy = false;
602 } else {
603 /* Automatically enable redirect otherwise. */
604 config->redirect_follow = config->redirect_dir = true;
605 }
606 }
607
608 return 0;
609 }
610
611 #define OVL_WORKDIR_NAME "work"
612 #define OVL_INDEXDIR_NAME "index"
613
614 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
615 const char *name, bool persist)
616 {
617 struct inode *dir = ofs->workbasedir->d_inode;
618 struct vfsmount *mnt = ofs->upper_mnt;
619 struct dentry *work;
620 int err;
621 bool retried = false;
622 bool locked = false;
623
624 inode_lock_nested(dir, I_MUTEX_PARENT);
625 locked = true;
626
627 retry:
628 work = lookup_one_len(name, ofs->workbasedir, strlen(name));
629
630 if (!IS_ERR(work)) {
631 struct iattr attr = {
632 .ia_valid = ATTR_MODE,
633 .ia_mode = S_IFDIR | 0,
634 };
635
636 if (work->d_inode) {
637 err = -EEXIST;
638 if (retried)
639 goto out_dput;
640
641 if (persist)
642 goto out_unlock;
643
644 retried = true;
645 ovl_workdir_cleanup(dir, mnt, work, 0);
646 dput(work);
647 goto retry;
648 }
649
650 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
651 err = PTR_ERR(work);
652 if (IS_ERR(work))
653 goto out_err;
654
655 /*
656 * Try to remove POSIX ACL xattrs from workdir. We are good if:
657 *
658 * a) success (there was a POSIX ACL xattr and was removed)
659 * b) -ENODATA (there was no POSIX ACL xattr)
660 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
661 *
662 * There are various other error values that could effectively
663 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
664 * if the xattr name is too long), but the set of filesystems
665 * allowed as upper are limited to "normal" ones, where checking
666 * for the above two errors is sufficient.
667 */
668 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
669 if (err && err != -ENODATA && err != -EOPNOTSUPP)
670 goto out_dput;
671
672 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
673 if (err && err != -ENODATA && err != -EOPNOTSUPP)
674 goto out_dput;
675
676 /* Clear any inherited mode bits */
677 inode_lock(work->d_inode);
678 err = notify_change(work, &attr, NULL);
679 inode_unlock(work->d_inode);
680 if (err)
681 goto out_dput;
682 } else {
683 err = PTR_ERR(work);
684 goto out_err;
685 }
686 out_unlock:
687 if (locked)
688 inode_unlock(dir);
689
690 return work;
691
692 out_dput:
693 dput(work);
694 out_err:
695 pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
696 ofs->config.workdir, name, -err);
697 work = NULL;
698 goto out_unlock;
699 }
700
701 static void ovl_unescape(char *s)
702 {
703 char *d = s;
704
705 for (;; s++, d++) {
706 if (*s == '\\')
707 s++;
708 *d = *s;
709 if (!*s)
710 break;
711 }
712 }
713
714 static int ovl_mount_dir_noesc(const char *name, struct path *path)
715 {
716 int err = -EINVAL;
717
718 if (!*name) {
719 pr_err("overlayfs: empty lowerdir\n");
720 goto out;
721 }
722 err = kern_path(name, LOOKUP_FOLLOW, path);
723 if (err) {
724 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
725 goto out;
726 }
727 err = -EINVAL;
728 if (ovl_dentry_weird(path->dentry)) {
729 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
730 goto out_put;
731 }
732 if (!d_is_dir(path->dentry)) {
733 pr_err("overlayfs: '%s' not a directory\n", name);
734 goto out_put;
735 }
736 return 0;
737
738 out_put:
739 path_put_init(path);
740 out:
741 return err;
742 }
743
744 static int ovl_mount_dir(const char *name, struct path *path)
745 {
746 int err = -ENOMEM;
747 char *tmp = kstrdup(name, GFP_KERNEL);
748
749 if (tmp) {
750 ovl_unescape(tmp);
751 err = ovl_mount_dir_noesc(tmp, path);
752
753 if (!err) {
754 if ((path->dentry->d_sb->s_magic != SHIFTFS_MAGIC) && ovl_dentry_remote(path->dentry)) {
755 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
756 tmp);
757 path_put_init(path);
758 err = -EINVAL;
759 }
760 }
761 kfree(tmp);
762 }
763 return err;
764 }
765
766 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
767 const char *name)
768 {
769 struct kstatfs statfs;
770 int err = vfs_statfs(path, &statfs);
771
772 if (err)
773 pr_err("overlayfs: statfs failed on '%s'\n", name);
774 else
775 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
776
777 return err;
778 }
779
780 static int ovl_lower_dir(const char *name, struct path *path,
781 struct ovl_fs *ofs, int *stack_depth, bool *remote)
782 {
783 int fh_type;
784 int err;
785
786 err = ovl_mount_dir_noesc(name, path);
787 if (err)
788 goto out;
789
790 err = ovl_check_namelen(path, ofs, name);
791 if (err)
792 goto out_put;
793
794 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
795
796 if (ovl_dentry_remote(path->dentry))
797 *remote = true;
798
799 /*
800 * The inodes index feature and NFS export need to encode and decode
801 * file handles, so they require that all layers support them.
802 */
803 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
804 if ((ofs->config.nfs_export ||
805 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
806 ofs->config.index = false;
807 ofs->config.nfs_export = false;
808 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
809 name);
810 }
811
812 /* Check if lower fs has 32bit inode numbers */
813 if (fh_type != FILEID_INO32_GEN)
814 ofs->xino_bits = 0;
815
816 return 0;
817
818 out_put:
819 path_put_init(path);
820 out:
821 return err;
822 }
823
824 /* Workdir should not be subdir of upperdir and vice versa */
825 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
826 {
827 bool ok = false;
828
829 if (workdir != upperdir) {
830 ok = (lock_rename(workdir, upperdir) == NULL);
831 unlock_rename(workdir, upperdir);
832 }
833 return ok;
834 }
835
836 static unsigned int ovl_split_lowerdirs(char *str)
837 {
838 unsigned int ctr = 1;
839 char *s, *d;
840
841 for (s = d = str;; s++, d++) {
842 if (*s == '\\') {
843 s++;
844 } else if (*s == ':') {
845 *d = '\0';
846 ctr++;
847 continue;
848 }
849 *d = *s;
850 if (!*s)
851 break;
852 }
853 return ctr;
854 }
855
856 static int __maybe_unused
857 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
858 struct dentry *dentry, struct inode *inode,
859 const char *name, void *buffer, size_t size)
860 {
861 return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
862 }
863
864 static int __maybe_unused
865 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
866 struct dentry *dentry, struct inode *inode,
867 const char *name, const void *value,
868 size_t size, int flags)
869 {
870 struct dentry *workdir = ovl_workdir(dentry);
871 struct inode *realinode = ovl_inode_real(inode);
872 struct posix_acl *acl = NULL;
873 int err;
874
875 /* Check that everything is OK before copy-up */
876 if (value) {
877 acl = posix_acl_from_xattr(&init_user_ns, value, size);
878 if (IS_ERR(acl))
879 return PTR_ERR(acl);
880 }
881 err = -EOPNOTSUPP;
882 if (!IS_POSIXACL(d_inode(workdir)))
883 goto out_acl_release;
884 if (!realinode->i_op->set_acl)
885 goto out_acl_release;
886 if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
887 err = acl ? -EACCES : 0;
888 goto out_acl_release;
889 }
890 err = -EPERM;
891 if (!inode_owner_or_capable(inode))
892 goto out_acl_release;
893
894 posix_acl_release(acl);
895
896 /*
897 * Check if sgid bit needs to be cleared (actual setacl operation will
898 * be done with mounter's capabilities and so that won't do it for us).
899 */
900 if (unlikely(inode->i_mode & S_ISGID) &&
901 handler->flags == ACL_TYPE_ACCESS &&
902 !in_group_p(inode->i_gid) &&
903 !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
904 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
905
906 err = ovl_setattr(dentry, &iattr);
907 if (err)
908 return err;
909 }
910
911 err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
912 if (!err)
913 ovl_copyattr(ovl_inode_real(inode), inode);
914
915 return err;
916
917 out_acl_release:
918 posix_acl_release(acl);
919 return err;
920 }
921
922 static int ovl_own_xattr_get(const struct xattr_handler *handler,
923 struct dentry *dentry, struct inode *inode,
924 const char *name, void *buffer, size_t size)
925 {
926 return -EOPNOTSUPP;
927 }
928
929 static int ovl_own_xattr_set(const struct xattr_handler *handler,
930 struct dentry *dentry, struct inode *inode,
931 const char *name, const void *value,
932 size_t size, int flags)
933 {
934 return -EOPNOTSUPP;
935 }
936
937 static int ovl_other_xattr_get(const struct xattr_handler *handler,
938 struct dentry *dentry, struct inode *inode,
939 const char *name, void *buffer, size_t size)
940 {
941 return ovl_xattr_get(dentry, inode, name, buffer, size);
942 }
943
944 static int ovl_other_xattr_set(const struct xattr_handler *handler,
945 struct dentry *dentry, struct inode *inode,
946 const char *name, const void *value,
947 size_t size, int flags)
948 {
949 return ovl_xattr_set(dentry, inode, name, value, size, flags);
950 }
951
952 static const struct xattr_handler __maybe_unused
953 ovl_posix_acl_access_xattr_handler = {
954 .name = XATTR_NAME_POSIX_ACL_ACCESS,
955 .flags = ACL_TYPE_ACCESS,
956 .get = ovl_posix_acl_xattr_get,
957 .set = ovl_posix_acl_xattr_set,
958 };
959
960 static const struct xattr_handler __maybe_unused
961 ovl_posix_acl_default_xattr_handler = {
962 .name = XATTR_NAME_POSIX_ACL_DEFAULT,
963 .flags = ACL_TYPE_DEFAULT,
964 .get = ovl_posix_acl_xattr_get,
965 .set = ovl_posix_acl_xattr_set,
966 };
967
968 static const struct xattr_handler ovl_own_xattr_handler = {
969 .prefix = OVL_XATTR_PREFIX,
970 .get = ovl_own_xattr_get,
971 .set = ovl_own_xattr_set,
972 };
973
974 static const struct xattr_handler ovl_other_xattr_handler = {
975 .prefix = "", /* catch all */
976 .get = ovl_other_xattr_get,
977 .set = ovl_other_xattr_set,
978 };
979
980 static const struct xattr_handler *ovl_xattr_handlers[] = {
981 #ifdef CONFIG_FS_POSIX_ACL
982 &ovl_posix_acl_access_xattr_handler,
983 &ovl_posix_acl_default_xattr_handler,
984 #endif
985 &ovl_own_xattr_handler,
986 &ovl_other_xattr_handler,
987 NULL
988 };
989
990 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
991 struct inode **ptrap, const char *name)
992 {
993 struct inode *trap;
994 int err;
995
996 trap = ovl_get_trap_inode(sb, dir);
997 err = PTR_ERR_OR_ZERO(trap);
998 if (err) {
999 if (err == -ELOOP)
1000 pr_err("overlayfs: conflicting %s path\n", name);
1001 return err;
1002 }
1003
1004 *ptrap = trap;
1005 return 0;
1006 }
1007
1008 /*
1009 * Determine how we treat concurrent use of upperdir/workdir based on the
1010 * index feature. This is papering over mount leaks of container runtimes,
1011 * for example, an old overlay mount is leaked and now its upperdir is
1012 * attempted to be used as a lower layer in a new overlay mount.
1013 */
1014 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1015 {
1016 if (ofs->config.index) {
1017 pr_err("overlayfs: %s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1018 name);
1019 return -EBUSY;
1020 } else {
1021 pr_warn("overlayfs: %s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1022 name);
1023 return 0;
1024 }
1025 }
1026
1027 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1028 struct path *upperpath)
1029 {
1030 struct vfsmount *upper_mnt;
1031 int err;
1032
1033 err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1034 if (err)
1035 goto out;
1036
1037 /* Upper fs should not be r/o */
1038 if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1039 pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1040 err = -EINVAL;
1041 goto out;
1042 }
1043
1044 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1045 if (err)
1046 goto out;
1047
1048 err = ovl_setup_trap(sb, upperpath->dentry, &ofs->upperdir_trap,
1049 "upperdir");
1050 if (err)
1051 goto out;
1052
1053 upper_mnt = clone_private_mount(upperpath);
1054 err = PTR_ERR(upper_mnt);
1055 if (IS_ERR(upper_mnt)) {
1056 pr_err("overlayfs: failed to clone upperpath\n");
1057 goto out;
1058 }
1059
1060 /* Don't inherit atime flags */
1061 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1062 ofs->upper_mnt = upper_mnt;
1063
1064 if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1065 ofs->upperdir_locked = true;
1066 } else {
1067 err = ovl_report_in_use(ofs, "upperdir");
1068 if (err)
1069 goto out;
1070 }
1071
1072 err = 0;
1073 out:
1074 return err;
1075 }
1076
1077 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1078 struct path *workpath)
1079 {
1080 struct vfsmount *mnt = ofs->upper_mnt;
1081 struct dentry *temp;
1082 int fh_type;
1083 int err;
1084
1085 err = mnt_want_write(mnt);
1086 if (err)
1087 return err;
1088
1089 ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1090 if (!ofs->workdir)
1091 goto out;
1092
1093 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1094 if (err)
1095 goto out;
1096
1097 /*
1098 * Upper should support d_type, else whiteouts are visible. Given
1099 * workdir and upper are on same fs, we can do iterate_dir() on
1100 * workdir. This check requires successful creation of workdir in
1101 * previous step.
1102 */
1103 err = ovl_check_d_type_supported(workpath);
1104 if (err < 0)
1105 goto out;
1106
1107 /*
1108 * We allowed this configuration and don't want to break users over
1109 * kernel upgrade. So warn instead of erroring out.
1110 */
1111 if (!err)
1112 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1113
1114 /* Check if upper/work fs supports O_TMPFILE */
1115 temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1116 ofs->tmpfile = !IS_ERR(temp);
1117 if (ofs->tmpfile)
1118 dput(temp);
1119 else
1120 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1121
1122 /*
1123 * Check if upper/work fs supports trusted.overlay.* xattr
1124 */
1125 err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1126 if (err) {
1127 ofs->noxattr = true;
1128 ofs->config.index = false;
1129 ofs->config.metacopy = false;
1130 pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1131 err = 0;
1132 } else {
1133 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1134 }
1135
1136 /* Check if upper/work fs supports file handles */
1137 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1138 if (ofs->config.index && !fh_type) {
1139 ofs->config.index = false;
1140 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1141 }
1142
1143 /* Check if upper fs has 32bit inode numbers */
1144 if (fh_type != FILEID_INO32_GEN)
1145 ofs->xino_bits = 0;
1146
1147 /* NFS export of r/w mount depends on index */
1148 if (ofs->config.nfs_export && !ofs->config.index) {
1149 pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1150 ofs->config.nfs_export = false;
1151 }
1152 out:
1153 mnt_drop_write(mnt);
1154 return err;
1155 }
1156
1157 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1158 struct path *upperpath)
1159 {
1160 int err;
1161 struct path workpath = { };
1162
1163 err = ovl_mount_dir(ofs->config.workdir, &workpath);
1164 if (err)
1165 goto out;
1166
1167 err = -EINVAL;
1168 if (upperpath->mnt != workpath.mnt) {
1169 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1170 goto out;
1171 }
1172 if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1173 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1174 goto out;
1175 }
1176
1177 ofs->workbasedir = dget(workpath.dentry);
1178
1179 if (ovl_inuse_trylock(ofs->workbasedir)) {
1180 ofs->workdir_locked = true;
1181 } else {
1182 err = ovl_report_in_use(ofs, "workdir");
1183 if (err)
1184 goto out;
1185 }
1186
1187 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1188 "workdir");
1189 if (err)
1190 goto out;
1191
1192 err = ovl_make_workdir(sb, ofs, &workpath);
1193
1194 out:
1195 path_put(&workpath);
1196
1197 return err;
1198 }
1199
1200 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1201 struct ovl_entry *oe, struct path *upperpath)
1202 {
1203 struct vfsmount *mnt = ofs->upper_mnt;
1204 int err;
1205
1206 err = mnt_want_write(mnt);
1207 if (err)
1208 return err;
1209
1210 /* Verify lower root is upper root origin */
1211 err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1212 true);
1213 if (err) {
1214 pr_err("overlayfs: failed to verify upper root origin\n");
1215 goto out;
1216 }
1217
1218 ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1219 if (ofs->indexdir) {
1220 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1221 "indexdir");
1222 if (err)
1223 goto out;
1224
1225 /*
1226 * Verify upper root is exclusively associated with index dir.
1227 * Older kernels stored upper fh in "trusted.overlay.origin"
1228 * xattr. If that xattr exists, verify that it is a match to
1229 * upper dir file handle. In any case, verify or set xattr
1230 * "trusted.overlay.upper" to indicate that index may have
1231 * directory entries.
1232 */
1233 if (ovl_check_origin_xattr(ofs->indexdir)) {
1234 err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1235 upperpath->dentry, true, false);
1236 if (err)
1237 pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1238 }
1239 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1240 if (err)
1241 pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1242
1243 /* Cleanup bad/stale/orphan index entries */
1244 if (!err)
1245 err = ovl_indexdir_cleanup(ofs);
1246 }
1247 if (err || !ofs->indexdir)
1248 pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1249
1250 out:
1251 mnt_drop_write(mnt);
1252 return err;
1253 }
1254
1255 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1256 {
1257 unsigned int i;
1258
1259 if (!ofs->config.nfs_export && !ofs->upper_mnt)
1260 return true;
1261
1262 for (i = 0; i < ofs->numlowerfs; i++) {
1263 /*
1264 * We use uuid to associate an overlay lower file handle with a
1265 * lower layer, so we can accept lower fs with null uuid as long
1266 * as all lower layers with null uuid are on the same fs.
1267 * if we detect multiple lower fs with the same uuid, we
1268 * disable lower file handle decoding on all of them.
1269 */
1270 if (uuid_equal(&ofs->lower_fs[i].sb->s_uuid, uuid)) {
1271 ofs->lower_fs[i].bad_uuid = true;
1272 return false;
1273 }
1274 }
1275 return true;
1276 }
1277
1278 /* Get a unique fsid for the layer */
1279 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1280 {
1281 struct super_block *sb = path->mnt->mnt_sb;
1282 unsigned int i;
1283 dev_t dev;
1284 int err;
1285 bool bad_uuid = false;
1286
1287 /* fsid 0 is reserved for upper fs even with non upper overlay */
1288 if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1289 return 0;
1290
1291 for (i = 0; i < ofs->numlowerfs; i++) {
1292 if (ofs->lower_fs[i].sb == sb)
1293 return i + 1;
1294 }
1295
1296 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1297 bad_uuid = true;
1298 if (ofs->config.index || ofs->config.nfs_export) {
1299 ofs->config.index = false;
1300 ofs->config.nfs_export = false;
1301 pr_warn("overlayfs: %s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1302 uuid_is_null(&sb->s_uuid) ? "null" :
1303 "conflicting",
1304 path->dentry);
1305 }
1306 }
1307
1308 err = get_anon_bdev(&dev);
1309 if (err) {
1310 pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1311 return err;
1312 }
1313
1314 ofs->lower_fs[ofs->numlowerfs].sb = sb;
1315 ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1316 ofs->lower_fs[ofs->numlowerfs].bad_uuid = bad_uuid;
1317 ofs->numlowerfs++;
1318
1319 return ofs->numlowerfs;
1320 }
1321
1322 static int ovl_get_lower_layers(struct super_block *sb, struct ovl_fs *ofs,
1323 struct path *stack, unsigned int numlower)
1324 {
1325 int err;
1326 unsigned int i;
1327
1328 err = -ENOMEM;
1329 ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1330 GFP_KERNEL);
1331 if (ofs->lower_layers == NULL)
1332 goto out;
1333
1334 ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1335 GFP_KERNEL);
1336 if (ofs->lower_fs == NULL)
1337 goto out;
1338
1339 for (i = 0; i < numlower; i++) {
1340 struct vfsmount *mnt;
1341 struct inode *trap;
1342 int fsid;
1343
1344 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1345 if (err < 0)
1346 goto out;
1347
1348 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1349 if (err)
1350 goto out;
1351
1352 if (ovl_is_inuse(stack[i].dentry)) {
1353 err = ovl_report_in_use(ofs, "lowerdir");
1354 if (err)
1355 goto out;
1356 }
1357
1358 mnt = clone_private_mount(&stack[i]);
1359 err = PTR_ERR(mnt);
1360 if (IS_ERR(mnt)) {
1361 pr_err("overlayfs: failed to clone lowerpath\n");
1362 iput(trap);
1363 goto out;
1364 }
1365
1366 /*
1367 * Make lower layers R/O. That way fchmod/fchown on lower file
1368 * will fail instead of modifying lower fs.
1369 */
1370 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1371
1372 ofs->lower_layers[ofs->numlower].trap = trap;
1373
1374 /*
1375 * If any lower mount is nosuid, force the ovl sb to also
1376 * be nosuid.
1377 */
1378 if (mnt->mnt_flags & MNT_NOSUID)
1379 sb->s_iflags |= SB_I_NOSUID;
1380
1381 ofs->lower_layers[ofs->numlower].mnt = mnt;
1382 ofs->lower_layers[ofs->numlower].idx = i + 1;
1383 ofs->lower_layers[ofs->numlower].fsid = fsid;
1384 if (fsid) {
1385 ofs->lower_layers[ofs->numlower].fs =
1386 &ofs->lower_fs[fsid - 1];
1387 }
1388 ofs->numlower++;
1389 }
1390
1391 /*
1392 * When all layers on same fs, overlay can use real inode numbers.
1393 * With mount option "xino=on", mounter declares that there are enough
1394 * free high bits in underlying fs to hold the unique fsid.
1395 * If overlayfs does encounter underlying inodes using the high xino
1396 * bits reserved for fsid, it emits a warning and uses the original
1397 * inode number.
1398 */
1399 if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1400 ofs->xino_bits = 0;
1401 ofs->config.xino = OVL_XINO_OFF;
1402 } else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1403 /*
1404 * This is a roundup of number of bits needed for numlowerfs+1
1405 * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1406 * upper fs even with non upper overlay.
1407 */
1408 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1409 ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1410 }
1411
1412 if (ofs->xino_bits) {
1413 pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1414 ofs->xino_bits);
1415 }
1416
1417 err = 0;
1418 out:
1419 return err;
1420 }
1421
1422 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1423 struct ovl_fs *ofs)
1424 {
1425 int err;
1426 char *lowertmp, *lower;
1427 struct path *stack = NULL;
1428 unsigned int stacklen, numlower = 0, i;
1429 bool remote = false;
1430 struct ovl_entry *oe;
1431
1432 err = -ENOMEM;
1433 lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1434 if (!lowertmp)
1435 goto out_err;
1436
1437 err = -EINVAL;
1438 stacklen = ovl_split_lowerdirs(lowertmp);
1439 if (stacklen > OVL_MAX_STACK) {
1440 pr_err("overlayfs: too many lower directories, limit is %d\n",
1441 OVL_MAX_STACK);
1442 goto out_err;
1443 } else if (!ofs->config.upperdir && stacklen == 1) {
1444 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1445 goto out_err;
1446 } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1447 ofs->config.redirect_follow) {
1448 pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1449 ofs->config.nfs_export = false;
1450 }
1451
1452 err = -ENOMEM;
1453 stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1454 if (!stack)
1455 goto out_err;
1456
1457 err = -EINVAL;
1458 lower = lowertmp;
1459 for (numlower = 0; numlower < stacklen; numlower++) {
1460 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1461 &sb->s_stack_depth, &remote);
1462 if (err)
1463 goto out_err;
1464
1465 lower = strchr(lower, '\0') + 1;
1466 }
1467
1468 err = -EINVAL;
1469 sb->s_stack_depth++;
1470 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1471 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1472 goto out_err;
1473 }
1474
1475 err = ovl_get_lower_layers(sb, ofs, stack, numlower);
1476 if (err)
1477 goto out_err;
1478
1479 err = -ENOMEM;
1480 oe = ovl_alloc_entry(numlower);
1481 if (!oe)
1482 goto out_err;
1483
1484 for (i = 0; i < numlower; i++) {
1485 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1486 oe->lowerstack[i].layer = &ofs->lower_layers[i];
1487 }
1488
1489 if (remote)
1490 sb->s_d_op = &ovl_reval_dentry_operations;
1491 else
1492 sb->s_d_op = &ovl_dentry_operations;
1493
1494 out:
1495 for (i = 0; i < numlower; i++)
1496 path_put(&stack[i]);
1497 kfree(stack);
1498 kfree(lowertmp);
1499
1500 return oe;
1501
1502 out_err:
1503 oe = ERR_PTR(err);
1504 goto out;
1505 }
1506
1507 /*
1508 * Check if this layer root is a descendant of:
1509 * - another layer of this overlayfs instance
1510 * - upper/work dir of any overlayfs instance
1511 */
1512 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1513 struct dentry *dentry, const char *name)
1514 {
1515 struct dentry *next = dentry, *parent;
1516 int err = 0;
1517
1518 if (!dentry)
1519 return 0;
1520
1521 parent = dget_parent(next);
1522
1523 /* Walk back ancestors to root (inclusive) looking for traps */
1524 while (!err && parent != next) {
1525 if (ovl_lookup_trap_inode(sb, parent)) {
1526 err = -ELOOP;
1527 pr_err("overlayfs: overlapping %s path\n", name);
1528 } else if (ovl_is_inuse(parent)) {
1529 err = ovl_report_in_use(ofs, name);
1530 }
1531 next = parent;
1532 parent = dget_parent(next);
1533 dput(next);
1534 }
1535
1536 dput(parent);
1537
1538 return err;
1539 }
1540
1541 /*
1542 * Check if any of the layers or work dirs overlap.
1543 */
1544 static int ovl_check_overlapping_layers(struct super_block *sb,
1545 struct ovl_fs *ofs)
1546 {
1547 int i, err;
1548
1549 if (ofs->upper_mnt) {
1550 err = ovl_check_layer(sb, ofs, ofs->upper_mnt->mnt_root,
1551 "upperdir");
1552 if (err)
1553 return err;
1554
1555 /*
1556 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1557 * this instance and covers overlapping work and index dirs,
1558 * unless work or index dir have been moved since created inside
1559 * workbasedir. In that case, we already have their traps in
1560 * inode cache and we will catch that case on lookup.
1561 */
1562 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1563 if (err)
1564 return err;
1565 }
1566
1567 for (i = 0; i < ofs->numlower; i++) {
1568 err = ovl_check_layer(sb, ofs,
1569 ofs->lower_layers[i].mnt->mnt_root,
1570 "lowerdir");
1571 if (err)
1572 return err;
1573 }
1574
1575 return 0;
1576 }
1577
1578 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1579 {
1580 struct path upperpath = { };
1581 struct dentry *root_dentry;
1582 struct ovl_entry *oe;
1583 struct ovl_fs *ofs;
1584 struct cred *cred;
1585 int err;
1586
1587 err = -ENOMEM;
1588 ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1589 if (!ofs)
1590 goto out;
1591
1592 ofs->creator_cred = cred = prepare_creds();
1593 if (!cred)
1594 goto out_err;
1595
1596 ofs->config.index = ovl_index_def;
1597 ofs->config.nfs_export = ovl_nfs_export_def;
1598 ofs->config.xino = ovl_xino_def();
1599 ofs->config.metacopy = ovl_metacopy_def;
1600 err = ovl_parse_opt((char *) data, &ofs->config);
1601 if (err)
1602 goto out_err;
1603
1604 err = -EINVAL;
1605 if (!ofs->config.lowerdir) {
1606 if (!silent)
1607 pr_err("overlayfs: missing 'lowerdir'\n");
1608 goto out_err;
1609 }
1610
1611 sb->s_stack_depth = 0;
1612 sb->s_maxbytes = MAX_LFS_FILESIZE;
1613 /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1614 if (ofs->config.xino != OVL_XINO_OFF)
1615 ofs->xino_bits = BITS_PER_LONG - 32;
1616
1617 /* alloc/destroy_inode needed for setting up traps in inode cache */
1618 sb->s_op = &ovl_super_operations;
1619
1620 if (ofs->config.upperdir) {
1621 if (!ofs->config.workdir) {
1622 pr_err("overlayfs: missing 'workdir'\n");
1623 goto out_err;
1624 }
1625
1626 err = ovl_get_upper(sb, ofs, &upperpath);
1627 if (err)
1628 goto out_err;
1629
1630 err = ovl_get_workdir(sb, ofs, &upperpath);
1631 if (err)
1632 goto out_err;
1633
1634 if (!ofs->workdir)
1635 sb->s_flags |= SB_RDONLY;
1636
1637 /*
1638 * If the upper mount is nosuid, force the ovl sb to also
1639 * be nosuid.
1640 */
1641 if (ofs->upper_mnt->mnt_flags & MNT_NOSUID)
1642 sb->s_iflags |= SB_I_NOSUID;
1643
1644 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1645 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1646
1647 }
1648 oe = ovl_get_lowerstack(sb, ofs);
1649 err = PTR_ERR(oe);
1650 if (IS_ERR(oe))
1651 goto out_err;
1652
1653 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1654 if (!ofs->upper_mnt)
1655 sb->s_flags |= SB_RDONLY;
1656
1657 if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1658 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1659 if (err)
1660 goto out_free_oe;
1661
1662 /* Force r/o mount with no index dir */
1663 if (!ofs->indexdir) {
1664 dput(ofs->workdir);
1665 ofs->workdir = NULL;
1666 sb->s_flags |= SB_RDONLY;
1667 }
1668
1669 }
1670
1671 err = ovl_check_overlapping_layers(sb, ofs);
1672 if (err)
1673 goto out_free_oe;
1674
1675 /* Show index=off in /proc/mounts for forced r/o mount */
1676 if (!ofs->indexdir) {
1677 ofs->config.index = false;
1678 if (ofs->upper_mnt && ofs->config.nfs_export) {
1679 pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1680 ofs->config.nfs_export = false;
1681 }
1682 }
1683
1684 if (ofs->config.metacopy && ofs->config.nfs_export) {
1685 pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1686 ofs->config.nfs_export = false;
1687 }
1688
1689 if (ofs->config.nfs_export)
1690 sb->s_export_op = &ovl_export_operations;
1691
1692 /* Never override disk quota limits or use reserved space */
1693 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1694
1695 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1696 sb->s_xattr = ovl_xattr_handlers;
1697 sb->s_fs_info = ofs;
1698 sb->s_flags |= SB_POSIXACL;
1699
1700 err = -ENOMEM;
1701 root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1702 if (!root_dentry)
1703 goto out_free_oe;
1704
1705 root_dentry->d_fsdata = oe;
1706
1707 mntput(upperpath.mnt);
1708 if (upperpath.dentry) {
1709 ovl_dentry_set_upper_alias(root_dentry);
1710 if (ovl_is_impuredir(upperpath.dentry))
1711 ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1712 }
1713
1714 /* Root is always merge -> can have whiteouts */
1715 ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1716 ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1717 ovl_set_upperdata(d_inode(root_dentry));
1718 ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1719 ovl_dentry_lower(root_dentry), NULL);
1720
1721 sb->s_root = root_dentry;
1722
1723 return 0;
1724
1725 out_free_oe:
1726 ovl_entry_stack_free(oe);
1727 kfree(oe);
1728 out_err:
1729 path_put(&upperpath);
1730 ovl_free_fs(ofs);
1731 out:
1732 return err;
1733 }
1734
1735 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1736 const char *dev_name, void *raw_data)
1737 {
1738 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1739 }
1740
1741 static struct file_system_type ovl_fs_type = {
1742 .owner = THIS_MODULE,
1743 .name = "overlay",
1744 .mount = ovl_mount,
1745 .kill_sb = kill_anon_super,
1746 .fs_flags = FS_USERNS_MOUNT,
1747 };
1748 MODULE_ALIAS_FS("overlay");
1749
1750 static void ovl_inode_init_once(void *foo)
1751 {
1752 struct ovl_inode *oi = foo;
1753
1754 inode_init_once(&oi->vfs_inode);
1755 }
1756
1757 static int __init ovl_init(void)
1758 {
1759 int err;
1760
1761 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1762 sizeof(struct ovl_inode), 0,
1763 (SLAB_RECLAIM_ACCOUNT|
1764 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1765 ovl_inode_init_once);
1766 if (ovl_inode_cachep == NULL)
1767 return -ENOMEM;
1768
1769 err = register_filesystem(&ovl_fs_type);
1770 if (err)
1771 kmem_cache_destroy(ovl_inode_cachep);
1772
1773 return err;
1774 }
1775
1776 static void __exit ovl_exit(void)
1777 {
1778 unregister_filesystem(&ovl_fs_type);
1779
1780 /*
1781 * Make sure all delayed rcu free inodes are flushed before we
1782 * destroy cache.
1783 */
1784 rcu_barrier();
1785 kmem_cache_destroy(ovl_inode_cachep);
1786
1787 }
1788
1789 module_init(ovl_init);
1790 module_exit(ovl_exit);
Ubuntu Eoan Linux kernel mirror