xref: /openbmc/linux/fs/overlayfs/super.c (revision 7bcae826)
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 
10 #include <linux/fs.h>
11 #include <linux/namei.h>
12 #include <linux/xattr.h>
13 #include <linux/mount.h>
14 #include <linux/parser.h>
15 #include <linux/module.h>
16 #include <linux/statfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/posix_acl_xattr.h>
19 #include "overlayfs.h"
20 #include "ovl_entry.h"
21 
22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
23 MODULE_DESCRIPTION("Overlay filesystem");
24 MODULE_LICENSE("GPL");
25 
26 
27 struct ovl_dir_cache;
28 
29 #define OVL_MAX_STACK 500
30 
31 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
32 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
33 MODULE_PARM_DESC(ovl_redirect_dir_def,
34 		 "Default to on or off for the redirect_dir feature");
35 
36 static void ovl_dentry_release(struct dentry *dentry)
37 {
38 	struct ovl_entry *oe = dentry->d_fsdata;
39 
40 	if (oe) {
41 		unsigned int i;
42 
43 		dput(oe->__upperdentry);
44 		kfree(oe->redirect);
45 		for (i = 0; i < oe->numlower; i++)
46 			dput(oe->lowerstack[i].dentry);
47 		kfree_rcu(oe, rcu);
48 	}
49 }
50 
51 static struct dentry *ovl_d_real(struct dentry *dentry,
52 				 const struct inode *inode,
53 				 unsigned int open_flags)
54 {
55 	struct dentry *real;
56 
57 	if (!d_is_reg(dentry)) {
58 		if (!inode || inode == d_inode(dentry))
59 			return dentry;
60 		goto bug;
61 	}
62 
63 	if (d_is_negative(dentry))
64 		return dentry;
65 
66 	if (open_flags) {
67 		int err = ovl_open_maybe_copy_up(dentry, open_flags);
68 
69 		if (err)
70 			return ERR_PTR(err);
71 	}
72 
73 	real = ovl_dentry_upper(dentry);
74 	if (real && (!inode || inode == d_inode(real)))
75 		return real;
76 
77 	real = ovl_dentry_lower(dentry);
78 	if (!real)
79 		goto bug;
80 
81 	/* Handle recursion */
82 	real = d_real(real, inode, open_flags);
83 
84 	if (!inode || inode == d_inode(real))
85 		return real;
86 bug:
87 	WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
88 	     inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
89 	return dentry;
90 }
91 
92 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
93 {
94 	struct ovl_entry *oe = dentry->d_fsdata;
95 	unsigned int i;
96 	int ret = 1;
97 
98 	for (i = 0; i < oe->numlower; i++) {
99 		struct dentry *d = oe->lowerstack[i].dentry;
100 
101 		if (d->d_flags & DCACHE_OP_REVALIDATE) {
102 			ret = d->d_op->d_revalidate(d, flags);
103 			if (ret < 0)
104 				return ret;
105 			if (!ret) {
106 				if (!(flags & LOOKUP_RCU))
107 					d_invalidate(d);
108 				return -ESTALE;
109 			}
110 		}
111 	}
112 	return 1;
113 }
114 
115 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
116 {
117 	struct ovl_entry *oe = dentry->d_fsdata;
118 	unsigned int i;
119 	int ret = 1;
120 
121 	for (i = 0; i < oe->numlower; i++) {
122 		struct dentry *d = oe->lowerstack[i].dentry;
123 
124 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
125 			ret = d->d_op->d_weak_revalidate(d, flags);
126 			if (ret <= 0)
127 				break;
128 		}
129 	}
130 	return ret;
131 }
132 
133 static const struct dentry_operations ovl_dentry_operations = {
134 	.d_release = ovl_dentry_release,
135 	.d_real = ovl_d_real,
136 };
137 
138 static const struct dentry_operations ovl_reval_dentry_operations = {
139 	.d_release = ovl_dentry_release,
140 	.d_real = ovl_d_real,
141 	.d_revalidate = ovl_dentry_revalidate,
142 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
143 };
144 
145 static void ovl_put_super(struct super_block *sb)
146 {
147 	struct ovl_fs *ufs = sb->s_fs_info;
148 	unsigned i;
149 
150 	dput(ufs->workdir);
151 	mntput(ufs->upper_mnt);
152 	for (i = 0; i < ufs->numlower; i++)
153 		mntput(ufs->lower_mnt[i]);
154 	kfree(ufs->lower_mnt);
155 
156 	kfree(ufs->config.lowerdir);
157 	kfree(ufs->config.upperdir);
158 	kfree(ufs->config.workdir);
159 	put_cred(ufs->creator_cred);
160 	kfree(ufs);
161 }
162 
163 /**
164  * ovl_statfs
165  * @sb: The overlayfs super block
166  * @buf: The struct kstatfs to fill in with stats
167  *
168  * Get the filesystem statistics.  As writes always target the upper layer
169  * filesystem pass the statfs to the upper filesystem (if it exists)
170  */
171 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
172 {
173 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
174 	struct dentry *root_dentry = dentry->d_sb->s_root;
175 	struct path path;
176 	int err;
177 
178 	ovl_path_real(root_dentry, &path);
179 
180 	err = vfs_statfs(&path, buf);
181 	if (!err) {
182 		buf->f_namelen = ofs->namelen;
183 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
184 	}
185 
186 	return err;
187 }
188 
189 /**
190  * ovl_show_options
191  *
192  * Prints the mount options for a given superblock.
193  * Returns zero; does not fail.
194  */
195 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
196 {
197 	struct super_block *sb = dentry->d_sb;
198 	struct ovl_fs *ufs = sb->s_fs_info;
199 
200 	seq_show_option(m, "lowerdir", ufs->config.lowerdir);
201 	if (ufs->config.upperdir) {
202 		seq_show_option(m, "upperdir", ufs->config.upperdir);
203 		seq_show_option(m, "workdir", ufs->config.workdir);
204 	}
205 	if (ufs->config.default_permissions)
206 		seq_puts(m, ",default_permissions");
207 	if (ufs->config.redirect_dir != ovl_redirect_dir_def)
208 		seq_printf(m, ",redirect_dir=%s",
209 			   ufs->config.redirect_dir ? "on" : "off");
210 	return 0;
211 }
212 
213 static int ovl_remount(struct super_block *sb, int *flags, char *data)
214 {
215 	struct ovl_fs *ufs = sb->s_fs_info;
216 
217 	if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
218 		return -EROFS;
219 
220 	return 0;
221 }
222 
223 static const struct super_operations ovl_super_operations = {
224 	.put_super	= ovl_put_super,
225 	.statfs		= ovl_statfs,
226 	.show_options	= ovl_show_options,
227 	.remount_fs	= ovl_remount,
228 	.drop_inode	= generic_delete_inode,
229 };
230 
231 enum {
232 	OPT_LOWERDIR,
233 	OPT_UPPERDIR,
234 	OPT_WORKDIR,
235 	OPT_DEFAULT_PERMISSIONS,
236 	OPT_REDIRECT_DIR_ON,
237 	OPT_REDIRECT_DIR_OFF,
238 	OPT_ERR,
239 };
240 
241 static const match_table_t ovl_tokens = {
242 	{OPT_LOWERDIR,			"lowerdir=%s"},
243 	{OPT_UPPERDIR,			"upperdir=%s"},
244 	{OPT_WORKDIR,			"workdir=%s"},
245 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
246 	{OPT_REDIRECT_DIR_ON,		"redirect_dir=on"},
247 	{OPT_REDIRECT_DIR_OFF,		"redirect_dir=off"},
248 	{OPT_ERR,			NULL}
249 };
250 
251 static char *ovl_next_opt(char **s)
252 {
253 	char *sbegin = *s;
254 	char *p;
255 
256 	if (sbegin == NULL)
257 		return NULL;
258 
259 	for (p = sbegin; *p; p++) {
260 		if (*p == '\\') {
261 			p++;
262 			if (!*p)
263 				break;
264 		} else if (*p == ',') {
265 			*p = '\0';
266 			*s = p + 1;
267 			return sbegin;
268 		}
269 	}
270 	*s = NULL;
271 	return sbegin;
272 }
273 
274 static int ovl_parse_opt(char *opt, struct ovl_config *config)
275 {
276 	char *p;
277 
278 	while ((p = ovl_next_opt(&opt)) != NULL) {
279 		int token;
280 		substring_t args[MAX_OPT_ARGS];
281 
282 		if (!*p)
283 			continue;
284 
285 		token = match_token(p, ovl_tokens, args);
286 		switch (token) {
287 		case OPT_UPPERDIR:
288 			kfree(config->upperdir);
289 			config->upperdir = match_strdup(&args[0]);
290 			if (!config->upperdir)
291 				return -ENOMEM;
292 			break;
293 
294 		case OPT_LOWERDIR:
295 			kfree(config->lowerdir);
296 			config->lowerdir = match_strdup(&args[0]);
297 			if (!config->lowerdir)
298 				return -ENOMEM;
299 			break;
300 
301 		case OPT_WORKDIR:
302 			kfree(config->workdir);
303 			config->workdir = match_strdup(&args[0]);
304 			if (!config->workdir)
305 				return -ENOMEM;
306 			break;
307 
308 		case OPT_DEFAULT_PERMISSIONS:
309 			config->default_permissions = true;
310 			break;
311 
312 		case OPT_REDIRECT_DIR_ON:
313 			config->redirect_dir = true;
314 			break;
315 
316 		case OPT_REDIRECT_DIR_OFF:
317 			config->redirect_dir = false;
318 			break;
319 
320 		default:
321 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
322 			return -EINVAL;
323 		}
324 	}
325 
326 	/* Workdir is useless in non-upper mount */
327 	if (!config->upperdir && config->workdir) {
328 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
329 			config->workdir);
330 		kfree(config->workdir);
331 		config->workdir = NULL;
332 	}
333 
334 	return 0;
335 }
336 
337 #define OVL_WORKDIR_NAME "work"
338 
339 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
340 					 struct dentry *dentry)
341 {
342 	struct inode *dir = dentry->d_inode;
343 	struct dentry *work;
344 	int err;
345 	bool retried = false;
346 
347 	err = mnt_want_write(mnt);
348 	if (err)
349 		return ERR_PTR(err);
350 
351 	inode_lock_nested(dir, I_MUTEX_PARENT);
352 retry:
353 	work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
354 			      strlen(OVL_WORKDIR_NAME));
355 
356 	if (!IS_ERR(work)) {
357 		struct iattr attr = {
358 			.ia_valid = ATTR_MODE,
359 			.ia_mode = S_IFDIR | 0,
360 		};
361 
362 		if (work->d_inode) {
363 			err = -EEXIST;
364 			if (retried)
365 				goto out_dput;
366 
367 			retried = true;
368 			ovl_workdir_cleanup(dir, mnt, work, 0);
369 			dput(work);
370 			goto retry;
371 		}
372 
373 		err = ovl_create_real(dir, work,
374 				      &(struct cattr){.mode = S_IFDIR | 0},
375 				      NULL, true);
376 		if (err)
377 			goto out_dput;
378 
379 		/*
380 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
381 		 *
382 		 * a) success (there was a POSIX ACL xattr and was removed)
383 		 * b) -ENODATA (there was no POSIX ACL xattr)
384 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
385 		 *
386 		 * There are various other error values that could effectively
387 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
388 		 * if the xattr name is too long), but the set of filesystems
389 		 * allowed as upper are limited to "normal" ones, where checking
390 		 * for the above two errors is sufficient.
391 		 */
392 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
393 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
394 			goto out_dput;
395 
396 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
397 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
398 			goto out_dput;
399 
400 		/* Clear any inherited mode bits */
401 		inode_lock(work->d_inode);
402 		err = notify_change(work, &attr, NULL);
403 		inode_unlock(work->d_inode);
404 		if (err)
405 			goto out_dput;
406 	}
407 out_unlock:
408 	inode_unlock(dir);
409 	mnt_drop_write(mnt);
410 
411 	return work;
412 
413 out_dput:
414 	dput(work);
415 	work = ERR_PTR(err);
416 	goto out_unlock;
417 }
418 
419 static void ovl_unescape(char *s)
420 {
421 	char *d = s;
422 
423 	for (;; s++, d++) {
424 		if (*s == '\\')
425 			s++;
426 		*d = *s;
427 		if (!*s)
428 			break;
429 	}
430 }
431 
432 static int ovl_mount_dir_noesc(const char *name, struct path *path)
433 {
434 	int err = -EINVAL;
435 
436 	if (!*name) {
437 		pr_err("overlayfs: empty lowerdir\n");
438 		goto out;
439 	}
440 	err = kern_path(name, LOOKUP_FOLLOW, path);
441 	if (err) {
442 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
443 		goto out;
444 	}
445 	err = -EINVAL;
446 	if (ovl_dentry_weird(path->dentry)) {
447 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
448 		goto out_put;
449 	}
450 	if (!d_is_dir(path->dentry)) {
451 		pr_err("overlayfs: '%s' not a directory\n", name);
452 		goto out_put;
453 	}
454 	return 0;
455 
456 out_put:
457 	path_put(path);
458 out:
459 	return err;
460 }
461 
462 static int ovl_mount_dir(const char *name, struct path *path)
463 {
464 	int err = -ENOMEM;
465 	char *tmp = kstrdup(name, GFP_KERNEL);
466 
467 	if (tmp) {
468 		ovl_unescape(tmp);
469 		err = ovl_mount_dir_noesc(tmp, path);
470 
471 		if (!err)
472 			if (ovl_dentry_remote(path->dentry)) {
473 				pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
474 				       tmp);
475 				path_put(path);
476 				err = -EINVAL;
477 			}
478 		kfree(tmp);
479 	}
480 	return err;
481 }
482 
483 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
484 			     const char *name)
485 {
486 	struct kstatfs statfs;
487 	int err = vfs_statfs(path, &statfs);
488 
489 	if (err)
490 		pr_err("overlayfs: statfs failed on '%s'\n", name);
491 	else
492 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
493 
494 	return err;
495 }
496 
497 static int ovl_lower_dir(const char *name, struct path *path,
498 			 struct ovl_fs *ofs, int *stack_depth, bool *remote)
499 {
500 	int err;
501 
502 	err = ovl_mount_dir_noesc(name, path);
503 	if (err)
504 		goto out;
505 
506 	err = ovl_check_namelen(path, ofs, name);
507 	if (err)
508 		goto out_put;
509 
510 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
511 
512 	if (ovl_dentry_remote(path->dentry))
513 		*remote = true;
514 
515 	return 0;
516 
517 out_put:
518 	path_put(path);
519 out:
520 	return err;
521 }
522 
523 /* Workdir should not be subdir of upperdir and vice versa */
524 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
525 {
526 	bool ok = false;
527 
528 	if (workdir != upperdir) {
529 		ok = (lock_rename(workdir, upperdir) == NULL);
530 		unlock_rename(workdir, upperdir);
531 	}
532 	return ok;
533 }
534 
535 static unsigned int ovl_split_lowerdirs(char *str)
536 {
537 	unsigned int ctr = 1;
538 	char *s, *d;
539 
540 	for (s = d = str;; s++, d++) {
541 		if (*s == '\\') {
542 			s++;
543 		} else if (*s == ':') {
544 			*d = '\0';
545 			ctr++;
546 			continue;
547 		}
548 		*d = *s;
549 		if (!*s)
550 			break;
551 	}
552 	return ctr;
553 }
554 
555 static int __maybe_unused
556 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
557 			struct dentry *dentry, struct inode *inode,
558 			const char *name, void *buffer, size_t size)
559 {
560 	return ovl_xattr_get(dentry, handler->name, buffer, size);
561 }
562 
563 static int __maybe_unused
564 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
565 			struct dentry *dentry, struct inode *inode,
566 			const char *name, const void *value,
567 			size_t size, int flags)
568 {
569 	struct dentry *workdir = ovl_workdir(dentry);
570 	struct inode *realinode = ovl_inode_real(inode, NULL);
571 	struct posix_acl *acl = NULL;
572 	int err;
573 
574 	/* Check that everything is OK before copy-up */
575 	if (value) {
576 		acl = posix_acl_from_xattr(&init_user_ns, value, size);
577 		if (IS_ERR(acl))
578 			return PTR_ERR(acl);
579 	}
580 	err = -EOPNOTSUPP;
581 	if (!IS_POSIXACL(d_inode(workdir)))
582 		goto out_acl_release;
583 	if (!realinode->i_op->set_acl)
584 		goto out_acl_release;
585 	if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
586 		err = acl ? -EACCES : 0;
587 		goto out_acl_release;
588 	}
589 	err = -EPERM;
590 	if (!inode_owner_or_capable(inode))
591 		goto out_acl_release;
592 
593 	posix_acl_release(acl);
594 
595 	/*
596 	 * Check if sgid bit needs to be cleared (actual setacl operation will
597 	 * be done with mounter's capabilities and so that won't do it for us).
598 	 */
599 	if (unlikely(inode->i_mode & S_ISGID) &&
600 	    handler->flags == ACL_TYPE_ACCESS &&
601 	    !in_group_p(inode->i_gid) &&
602 	    !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
603 		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
604 
605 		err = ovl_setattr(dentry, &iattr);
606 		if (err)
607 			return err;
608 	}
609 
610 	err = ovl_xattr_set(dentry, handler->name, value, size, flags);
611 	if (!err)
612 		ovl_copyattr(ovl_inode_real(inode, NULL), inode);
613 
614 	return err;
615 
616 out_acl_release:
617 	posix_acl_release(acl);
618 	return err;
619 }
620 
621 static int ovl_own_xattr_get(const struct xattr_handler *handler,
622 			     struct dentry *dentry, struct inode *inode,
623 			     const char *name, void *buffer, size_t size)
624 {
625 	return -EOPNOTSUPP;
626 }
627 
628 static int ovl_own_xattr_set(const struct xattr_handler *handler,
629 			     struct dentry *dentry, struct inode *inode,
630 			     const char *name, const void *value,
631 			     size_t size, int flags)
632 {
633 	return -EOPNOTSUPP;
634 }
635 
636 static int ovl_other_xattr_get(const struct xattr_handler *handler,
637 			       struct dentry *dentry, struct inode *inode,
638 			       const char *name, void *buffer, size_t size)
639 {
640 	return ovl_xattr_get(dentry, name, buffer, size);
641 }
642 
643 static int ovl_other_xattr_set(const struct xattr_handler *handler,
644 			       struct dentry *dentry, struct inode *inode,
645 			       const char *name, const void *value,
646 			       size_t size, int flags)
647 {
648 	return ovl_xattr_set(dentry, name, value, size, flags);
649 }
650 
651 static const struct xattr_handler __maybe_unused
652 ovl_posix_acl_access_xattr_handler = {
653 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
654 	.flags = ACL_TYPE_ACCESS,
655 	.get = ovl_posix_acl_xattr_get,
656 	.set = ovl_posix_acl_xattr_set,
657 };
658 
659 static const struct xattr_handler __maybe_unused
660 ovl_posix_acl_default_xattr_handler = {
661 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
662 	.flags = ACL_TYPE_DEFAULT,
663 	.get = ovl_posix_acl_xattr_get,
664 	.set = ovl_posix_acl_xattr_set,
665 };
666 
667 static const struct xattr_handler ovl_own_xattr_handler = {
668 	.prefix	= OVL_XATTR_PREFIX,
669 	.get = ovl_own_xattr_get,
670 	.set = ovl_own_xattr_set,
671 };
672 
673 static const struct xattr_handler ovl_other_xattr_handler = {
674 	.prefix	= "", /* catch all */
675 	.get = ovl_other_xattr_get,
676 	.set = ovl_other_xattr_set,
677 };
678 
679 static const struct xattr_handler *ovl_xattr_handlers[] = {
680 #ifdef CONFIG_FS_POSIX_ACL
681 	&ovl_posix_acl_access_xattr_handler,
682 	&ovl_posix_acl_default_xattr_handler,
683 #endif
684 	&ovl_own_xattr_handler,
685 	&ovl_other_xattr_handler,
686 	NULL
687 };
688 
689 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
690 {
691 	struct path upperpath = { NULL, NULL };
692 	struct path workpath = { NULL, NULL };
693 	struct dentry *root_dentry;
694 	struct inode *realinode;
695 	struct ovl_entry *oe;
696 	struct ovl_fs *ufs;
697 	struct path *stack = NULL;
698 	char *lowertmp;
699 	char *lower;
700 	unsigned int numlower;
701 	unsigned int stacklen = 0;
702 	unsigned int i;
703 	bool remote = false;
704 	int err;
705 
706 	err = -ENOMEM;
707 	ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
708 	if (!ufs)
709 		goto out;
710 
711 	ufs->config.redirect_dir = ovl_redirect_dir_def;
712 	err = ovl_parse_opt((char *) data, &ufs->config);
713 	if (err)
714 		goto out_free_config;
715 
716 	err = -EINVAL;
717 	if (!ufs->config.lowerdir) {
718 		if (!silent)
719 			pr_err("overlayfs: missing 'lowerdir'\n");
720 		goto out_free_config;
721 	}
722 
723 	sb->s_stack_depth = 0;
724 	sb->s_maxbytes = MAX_LFS_FILESIZE;
725 	if (ufs->config.upperdir) {
726 		if (!ufs->config.workdir) {
727 			pr_err("overlayfs: missing 'workdir'\n");
728 			goto out_free_config;
729 		}
730 
731 		err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
732 		if (err)
733 			goto out_free_config;
734 
735 		/* Upper fs should not be r/o */
736 		if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
737 			pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
738 			err = -EINVAL;
739 			goto out_put_upperpath;
740 		}
741 
742 		err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
743 		if (err)
744 			goto out_put_upperpath;
745 
746 		err = ovl_mount_dir(ufs->config.workdir, &workpath);
747 		if (err)
748 			goto out_put_upperpath;
749 
750 		err = -EINVAL;
751 		if (upperpath.mnt != workpath.mnt) {
752 			pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
753 			goto out_put_workpath;
754 		}
755 		if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
756 			pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
757 			goto out_put_workpath;
758 		}
759 		sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
760 	}
761 	err = -ENOMEM;
762 	lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
763 	if (!lowertmp)
764 		goto out_put_workpath;
765 
766 	err = -EINVAL;
767 	stacklen = ovl_split_lowerdirs(lowertmp);
768 	if (stacklen > OVL_MAX_STACK) {
769 		pr_err("overlayfs: too many lower directories, limit is %d\n",
770 		       OVL_MAX_STACK);
771 		goto out_free_lowertmp;
772 	} else if (!ufs->config.upperdir && stacklen == 1) {
773 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
774 		goto out_free_lowertmp;
775 	}
776 
777 	err = -ENOMEM;
778 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
779 	if (!stack)
780 		goto out_free_lowertmp;
781 
782 	err = -EINVAL;
783 	lower = lowertmp;
784 	for (numlower = 0; numlower < stacklen; numlower++) {
785 		err = ovl_lower_dir(lower, &stack[numlower], ufs,
786 				    &sb->s_stack_depth, &remote);
787 		if (err)
788 			goto out_put_lowerpath;
789 
790 		lower = strchr(lower, '\0') + 1;
791 	}
792 
793 	err = -EINVAL;
794 	sb->s_stack_depth++;
795 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
796 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
797 		goto out_put_lowerpath;
798 	}
799 
800 	if (ufs->config.upperdir) {
801 		ufs->upper_mnt = clone_private_mount(&upperpath);
802 		err = PTR_ERR(ufs->upper_mnt);
803 		if (IS_ERR(ufs->upper_mnt)) {
804 			pr_err("overlayfs: failed to clone upperpath\n");
805 			goto out_put_lowerpath;
806 		}
807 		/* Don't inherit atime flags */
808 		ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
809 
810 		sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
811 
812 		ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
813 		err = PTR_ERR(ufs->workdir);
814 		if (IS_ERR(ufs->workdir)) {
815 			pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
816 				ufs->config.workdir, OVL_WORKDIR_NAME, -err);
817 			sb->s_flags |= MS_RDONLY;
818 			ufs->workdir = NULL;
819 		}
820 
821 		/*
822 		 * Upper should support d_type, else whiteouts are visible.
823 		 * Given workdir and upper are on same fs, we can do
824 		 * iterate_dir() on workdir. This check requires successful
825 		 * creation of workdir in previous step.
826 		 */
827 		if (ufs->workdir) {
828 			err = ovl_check_d_type_supported(&workpath);
829 			if (err < 0)
830 				goto out_put_workdir;
831 
832 			/*
833 			 * We allowed this configuration and don't want to
834 			 * break users over kernel upgrade. So warn instead
835 			 * of erroring out.
836 			 */
837 			if (!err)
838 				pr_warn("overlayfs: upper fs needs to support d_type.\n");
839 		}
840 	}
841 
842 	err = -ENOMEM;
843 	ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
844 	if (ufs->lower_mnt == NULL)
845 		goto out_put_workdir;
846 	for (i = 0; i < numlower; i++) {
847 		struct vfsmount *mnt = clone_private_mount(&stack[i]);
848 
849 		err = PTR_ERR(mnt);
850 		if (IS_ERR(mnt)) {
851 			pr_err("overlayfs: failed to clone lowerpath\n");
852 			goto out_put_lower_mnt;
853 		}
854 		/*
855 		 * Make lower_mnt R/O.  That way fchmod/fchown on lower file
856 		 * will fail instead of modifying lower fs.
857 		 */
858 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
859 
860 		ufs->lower_mnt[ufs->numlower] = mnt;
861 		ufs->numlower++;
862 	}
863 
864 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
865 	if (!ufs->upper_mnt)
866 		sb->s_flags |= MS_RDONLY;
867 
868 	if (remote)
869 		sb->s_d_op = &ovl_reval_dentry_operations;
870 	else
871 		sb->s_d_op = &ovl_dentry_operations;
872 
873 	ufs->creator_cred = prepare_creds();
874 	if (!ufs->creator_cred)
875 		goto out_put_lower_mnt;
876 
877 	err = -ENOMEM;
878 	oe = ovl_alloc_entry(numlower);
879 	if (!oe)
880 		goto out_put_cred;
881 
882 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
883 	sb->s_op = &ovl_super_operations;
884 	sb->s_xattr = ovl_xattr_handlers;
885 	sb->s_fs_info = ufs;
886 	sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
887 
888 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
889 	if (!root_dentry)
890 		goto out_free_oe;
891 
892 	mntput(upperpath.mnt);
893 	for (i = 0; i < numlower; i++)
894 		mntput(stack[i].mnt);
895 	path_put(&workpath);
896 	kfree(lowertmp);
897 
898 	oe->__upperdentry = upperpath.dentry;
899 	for (i = 0; i < numlower; i++) {
900 		oe->lowerstack[i].dentry = stack[i].dentry;
901 		oe->lowerstack[i].mnt = ufs->lower_mnt[i];
902 	}
903 	kfree(stack);
904 
905 	root_dentry->d_fsdata = oe;
906 
907 	realinode = d_inode(ovl_dentry_real(root_dentry));
908 	ovl_inode_init(d_inode(root_dentry), realinode, !!upperpath.dentry);
909 	ovl_copyattr(realinode, d_inode(root_dentry));
910 
911 	sb->s_root = root_dentry;
912 
913 	return 0;
914 
915 out_free_oe:
916 	kfree(oe);
917 out_put_cred:
918 	put_cred(ufs->creator_cred);
919 out_put_lower_mnt:
920 	for (i = 0; i < ufs->numlower; i++)
921 		mntput(ufs->lower_mnt[i]);
922 	kfree(ufs->lower_mnt);
923 out_put_workdir:
924 	dput(ufs->workdir);
925 	mntput(ufs->upper_mnt);
926 out_put_lowerpath:
927 	for (i = 0; i < numlower; i++)
928 		path_put(&stack[i]);
929 	kfree(stack);
930 out_free_lowertmp:
931 	kfree(lowertmp);
932 out_put_workpath:
933 	path_put(&workpath);
934 out_put_upperpath:
935 	path_put(&upperpath);
936 out_free_config:
937 	kfree(ufs->config.lowerdir);
938 	kfree(ufs->config.upperdir);
939 	kfree(ufs->config.workdir);
940 	kfree(ufs);
941 out:
942 	return err;
943 }
944 
945 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
946 				const char *dev_name, void *raw_data)
947 {
948 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
949 }
950 
951 static struct file_system_type ovl_fs_type = {
952 	.owner		= THIS_MODULE,
953 	.name		= "overlay",
954 	.mount		= ovl_mount,
955 	.kill_sb	= kill_anon_super,
956 };
957 MODULE_ALIAS_FS("overlay");
958 
959 static int __init ovl_init(void)
960 {
961 	return register_filesystem(&ovl_fs_type);
962 }
963 
964 static void __exit ovl_exit(void)
965 {
966 	unregister_filesystem(&ovl_fs_type);
967 }
968 
969 module_init(ovl_init);
970 module_exit(ovl_exit);
971