xref: /openbmc/linux/fs/overlayfs/super.c (revision 8c0b9ee8)
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/security.h>
14 #include <linux/mount.h>
15 #include <linux/slab.h>
16 #include <linux/parser.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/statfs.h>
20 #include <linux/seq_file.h>
21 #include "overlayfs.h"
22 
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Overlay filesystem");
25 MODULE_LICENSE("GPL");
26 
27 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
28 
29 struct ovl_config {
30 	char *lowerdir;
31 	char *upperdir;
32 	char *workdir;
33 };
34 
35 /* private information held for overlayfs's superblock */
36 struct ovl_fs {
37 	struct vfsmount *upper_mnt;
38 	unsigned numlower;
39 	struct vfsmount **lower_mnt;
40 	struct dentry *workdir;
41 	long lower_namelen;
42 	/* pathnames of lower and upper dirs, for show_options */
43 	struct ovl_config config;
44 };
45 
46 struct ovl_dir_cache;
47 
48 /* private information held for every overlayfs dentry */
49 struct ovl_entry {
50 	struct dentry *__upperdentry;
51 	struct ovl_dir_cache *cache;
52 	union {
53 		struct {
54 			u64 version;
55 			bool opaque;
56 		};
57 		struct rcu_head rcu;
58 	};
59 	unsigned numlower;
60 	struct path lowerstack[];
61 };
62 
63 #define OVL_MAX_STACK 500
64 
65 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
66 {
67 	return oe->numlower ? oe->lowerstack[0].dentry : NULL;
68 }
69 
70 enum ovl_path_type ovl_path_type(struct dentry *dentry)
71 {
72 	struct ovl_entry *oe = dentry->d_fsdata;
73 	enum ovl_path_type type = 0;
74 
75 	if (oe->__upperdentry) {
76 		type = __OVL_PATH_UPPER;
77 
78 		if (oe->numlower) {
79 			if (S_ISDIR(dentry->d_inode->i_mode))
80 				type |= __OVL_PATH_MERGE;
81 		} else if (!oe->opaque) {
82 			type |= __OVL_PATH_PURE;
83 		}
84 	} else {
85 		if (oe->numlower > 1)
86 			type |= __OVL_PATH_MERGE;
87 	}
88 	return type;
89 }
90 
91 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
92 {
93 	return lockless_dereference(oe->__upperdentry);
94 }
95 
96 void ovl_path_upper(struct dentry *dentry, struct path *path)
97 {
98 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
99 	struct ovl_entry *oe = dentry->d_fsdata;
100 
101 	path->mnt = ofs->upper_mnt;
102 	path->dentry = ovl_upperdentry_dereference(oe);
103 }
104 
105 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
106 {
107 	enum ovl_path_type type = ovl_path_type(dentry);
108 
109 	if (!OVL_TYPE_UPPER(type))
110 		ovl_path_lower(dentry, path);
111 	else
112 		ovl_path_upper(dentry, path);
113 
114 	return type;
115 }
116 
117 struct dentry *ovl_dentry_upper(struct dentry *dentry)
118 {
119 	struct ovl_entry *oe = dentry->d_fsdata;
120 
121 	return ovl_upperdentry_dereference(oe);
122 }
123 
124 struct dentry *ovl_dentry_lower(struct dentry *dentry)
125 {
126 	struct ovl_entry *oe = dentry->d_fsdata;
127 
128 	return __ovl_dentry_lower(oe);
129 }
130 
131 struct dentry *ovl_dentry_real(struct dentry *dentry)
132 {
133 	struct ovl_entry *oe = dentry->d_fsdata;
134 	struct dentry *realdentry;
135 
136 	realdentry = ovl_upperdentry_dereference(oe);
137 	if (!realdentry)
138 		realdentry = __ovl_dentry_lower(oe);
139 
140 	return realdentry;
141 }
142 
143 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
144 {
145 	struct dentry *realdentry;
146 
147 	realdentry = ovl_upperdentry_dereference(oe);
148 	if (realdentry) {
149 		*is_upper = true;
150 	} else {
151 		realdentry = __ovl_dentry_lower(oe);
152 		*is_upper = false;
153 	}
154 	return realdentry;
155 }
156 
157 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
158 {
159 	struct ovl_entry *oe = dentry->d_fsdata;
160 
161 	return oe->cache;
162 }
163 
164 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
165 {
166 	struct ovl_entry *oe = dentry->d_fsdata;
167 
168 	oe->cache = cache;
169 }
170 
171 void ovl_path_lower(struct dentry *dentry, struct path *path)
172 {
173 	struct ovl_entry *oe = dentry->d_fsdata;
174 
175 	*path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
176 }
177 
178 int ovl_want_write(struct dentry *dentry)
179 {
180 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
181 	return mnt_want_write(ofs->upper_mnt);
182 }
183 
184 void ovl_drop_write(struct dentry *dentry)
185 {
186 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
187 	mnt_drop_write(ofs->upper_mnt);
188 }
189 
190 struct dentry *ovl_workdir(struct dentry *dentry)
191 {
192 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
193 	return ofs->workdir;
194 }
195 
196 bool ovl_dentry_is_opaque(struct dentry *dentry)
197 {
198 	struct ovl_entry *oe = dentry->d_fsdata;
199 	return oe->opaque;
200 }
201 
202 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
203 {
204 	struct ovl_entry *oe = dentry->d_fsdata;
205 	oe->opaque = opaque;
206 }
207 
208 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
209 {
210 	struct ovl_entry *oe = dentry->d_fsdata;
211 
212 	WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
213 	WARN_ON(oe->__upperdentry);
214 	BUG_ON(!upperdentry->d_inode);
215 	/*
216 	 * Make sure upperdentry is consistent before making it visible to
217 	 * ovl_upperdentry_dereference().
218 	 */
219 	smp_wmb();
220 	oe->__upperdentry = upperdentry;
221 }
222 
223 void ovl_dentry_version_inc(struct dentry *dentry)
224 {
225 	struct ovl_entry *oe = dentry->d_fsdata;
226 
227 	WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
228 	oe->version++;
229 }
230 
231 u64 ovl_dentry_version_get(struct dentry *dentry)
232 {
233 	struct ovl_entry *oe = dentry->d_fsdata;
234 
235 	WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
236 	return oe->version;
237 }
238 
239 bool ovl_is_whiteout(struct dentry *dentry)
240 {
241 	struct inode *inode = dentry->d_inode;
242 
243 	return inode && IS_WHITEOUT(inode);
244 }
245 
246 static bool ovl_is_opaquedir(struct dentry *dentry)
247 {
248 	int res;
249 	char val;
250 	struct inode *inode = dentry->d_inode;
251 
252 	if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
253 		return false;
254 
255 	res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
256 	if (res == 1 && val == 'y')
257 		return true;
258 
259 	return false;
260 }
261 
262 static void ovl_dentry_release(struct dentry *dentry)
263 {
264 	struct ovl_entry *oe = dentry->d_fsdata;
265 
266 	if (oe) {
267 		unsigned int i;
268 
269 		dput(oe->__upperdentry);
270 		for (i = 0; i < oe->numlower; i++)
271 			dput(oe->lowerstack[i].dentry);
272 		kfree_rcu(oe, rcu);
273 	}
274 }
275 
276 static const struct dentry_operations ovl_dentry_operations = {
277 	.d_release = ovl_dentry_release,
278 };
279 
280 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
281 {
282 	size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
283 	struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
284 
285 	if (oe)
286 		oe->numlower = numlower;
287 
288 	return oe;
289 }
290 
291 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
292 					     struct qstr *name)
293 {
294 	struct dentry *dentry;
295 
296 	mutex_lock(&dir->d_inode->i_mutex);
297 	dentry = lookup_one_len(name->name, dir, name->len);
298 	mutex_unlock(&dir->d_inode->i_mutex);
299 
300 	if (IS_ERR(dentry)) {
301 		if (PTR_ERR(dentry) == -ENOENT)
302 			dentry = NULL;
303 	} else if (!dentry->d_inode) {
304 		dput(dentry);
305 		dentry = NULL;
306 	}
307 	return dentry;
308 }
309 
310 /*
311  * Returns next layer in stack starting from top.
312  * Returns -1 if this is the last layer.
313  */
314 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
315 {
316 	struct ovl_entry *oe = dentry->d_fsdata;
317 
318 	BUG_ON(idx < 0);
319 	if (idx == 0) {
320 		ovl_path_upper(dentry, path);
321 		if (path->dentry)
322 			return oe->numlower ? 1 : -1;
323 		idx++;
324 	}
325 	BUG_ON(idx > oe->numlower);
326 	*path = oe->lowerstack[idx - 1];
327 
328 	return (idx < oe->numlower) ? idx + 1 : -1;
329 }
330 
331 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
332 			  unsigned int flags)
333 {
334 	struct ovl_entry *oe;
335 	struct ovl_entry *poe = dentry->d_parent->d_fsdata;
336 	struct path *stack = NULL;
337 	struct dentry *upperdir, *upperdentry = NULL;
338 	unsigned int ctr = 0;
339 	struct inode *inode = NULL;
340 	bool upperopaque = false;
341 	struct dentry *this, *prev = NULL;
342 	unsigned int i;
343 	int err;
344 
345 	upperdir = ovl_upperdentry_dereference(poe);
346 	if (upperdir) {
347 		this = ovl_lookup_real(upperdir, &dentry->d_name);
348 		err = PTR_ERR(this);
349 		if (IS_ERR(this))
350 			goto out;
351 
352 		if (this) {
353 			if (ovl_is_whiteout(this)) {
354 				dput(this);
355 				this = NULL;
356 				upperopaque = true;
357 			} else if (poe->numlower && ovl_is_opaquedir(this)) {
358 				upperopaque = true;
359 			}
360 		}
361 		upperdentry = prev = this;
362 	}
363 
364 	if (!upperopaque && poe->numlower) {
365 		err = -ENOMEM;
366 		stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
367 		if (!stack)
368 			goto out_put_upper;
369 	}
370 
371 	for (i = 0; !upperopaque && i < poe->numlower; i++) {
372 		bool opaque = false;
373 		struct path lowerpath = poe->lowerstack[i];
374 
375 		this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
376 		err = PTR_ERR(this);
377 		if (IS_ERR(this)) {
378 			/*
379 			 * If it's positive, then treat ENAMETOOLONG as ENOENT.
380 			 */
381 			if (err == -ENAMETOOLONG && (upperdentry || ctr))
382 				continue;
383 			goto out_put;
384 		}
385 		if (!this)
386 			continue;
387 		if (ovl_is_whiteout(this)) {
388 			dput(this);
389 			break;
390 		}
391 		/*
392 		 * Only makes sense to check opaque dir if this is not the
393 		 * lowermost layer.
394 		 */
395 		if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
396 			opaque = true;
397 
398 		if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
399 			     !S_ISDIR(this->d_inode->i_mode))) {
400 			/*
401 			 * FIXME: check for upper-opaqueness maybe better done
402 			 * in remove code.
403 			 */
404 			if (prev == upperdentry)
405 				upperopaque = true;
406 			dput(this);
407 			break;
408 		}
409 		/*
410 		 * If this is a non-directory then stop here.
411 		 */
412 		if (!S_ISDIR(this->d_inode->i_mode))
413 			opaque = true;
414 
415 		stack[ctr].dentry = this;
416 		stack[ctr].mnt = lowerpath.mnt;
417 		ctr++;
418 		prev = this;
419 		if (opaque)
420 			break;
421 	}
422 
423 	oe = ovl_alloc_entry(ctr);
424 	err = -ENOMEM;
425 	if (!oe)
426 		goto out_put;
427 
428 	if (upperdentry || ctr) {
429 		struct dentry *realdentry;
430 
431 		realdentry = upperdentry ? upperdentry : stack[0].dentry;
432 
433 		err = -ENOMEM;
434 		inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
435 				      oe);
436 		if (!inode)
437 			goto out_free_oe;
438 		ovl_copyattr(realdentry->d_inode, inode);
439 	}
440 
441 	oe->opaque = upperopaque;
442 	oe->__upperdentry = upperdentry;
443 	memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
444 	kfree(stack);
445 	dentry->d_fsdata = oe;
446 	d_add(dentry, inode);
447 
448 	return NULL;
449 
450 out_free_oe:
451 	kfree(oe);
452 out_put:
453 	for (i = 0; i < ctr; i++)
454 		dput(stack[i].dentry);
455 	kfree(stack);
456 out_put_upper:
457 	dput(upperdentry);
458 out:
459 	return ERR_PTR(err);
460 }
461 
462 struct file *ovl_path_open(struct path *path, int flags)
463 {
464 	return dentry_open(path, flags, current_cred());
465 }
466 
467 static void ovl_put_super(struct super_block *sb)
468 {
469 	struct ovl_fs *ufs = sb->s_fs_info;
470 	unsigned i;
471 
472 	dput(ufs->workdir);
473 	mntput(ufs->upper_mnt);
474 	for (i = 0; i < ufs->numlower; i++)
475 		mntput(ufs->lower_mnt[i]);
476 
477 	kfree(ufs->config.lowerdir);
478 	kfree(ufs->config.upperdir);
479 	kfree(ufs->config.workdir);
480 	kfree(ufs);
481 }
482 
483 /**
484  * ovl_statfs
485  * @sb: The overlayfs super block
486  * @buf: The struct kstatfs to fill in with stats
487  *
488  * Get the filesystem statistics.  As writes always target the upper layer
489  * filesystem pass the statfs to the upper filesystem (if it exists)
490  */
491 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
492 {
493 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
494 	struct dentry *root_dentry = dentry->d_sb->s_root;
495 	struct path path;
496 	int err;
497 
498 	ovl_path_real(root_dentry, &path);
499 
500 	err = vfs_statfs(&path, buf);
501 	if (!err) {
502 		buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
503 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
504 	}
505 
506 	return err;
507 }
508 
509 /**
510  * ovl_show_options
511  *
512  * Prints the mount options for a given superblock.
513  * Returns zero; does not fail.
514  */
515 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
516 {
517 	struct super_block *sb = dentry->d_sb;
518 	struct ovl_fs *ufs = sb->s_fs_info;
519 
520 	seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
521 	if (ufs->config.upperdir) {
522 		seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
523 		seq_printf(m, ",workdir=%s", ufs->config.workdir);
524 	}
525 	return 0;
526 }
527 
528 static int ovl_remount(struct super_block *sb, int *flags, char *data)
529 {
530 	struct ovl_fs *ufs = sb->s_fs_info;
531 
532 	if (!(*flags & MS_RDONLY) && !ufs->upper_mnt)
533 		return -EROFS;
534 
535 	return 0;
536 }
537 
538 static const struct super_operations ovl_super_operations = {
539 	.put_super	= ovl_put_super,
540 	.statfs		= ovl_statfs,
541 	.show_options	= ovl_show_options,
542 	.remount_fs	= ovl_remount,
543 };
544 
545 enum {
546 	OPT_LOWERDIR,
547 	OPT_UPPERDIR,
548 	OPT_WORKDIR,
549 	OPT_ERR,
550 };
551 
552 static const match_table_t ovl_tokens = {
553 	{OPT_LOWERDIR,			"lowerdir=%s"},
554 	{OPT_UPPERDIR,			"upperdir=%s"},
555 	{OPT_WORKDIR,			"workdir=%s"},
556 	{OPT_ERR,			NULL}
557 };
558 
559 static char *ovl_next_opt(char **s)
560 {
561 	char *sbegin = *s;
562 	char *p;
563 
564 	if (sbegin == NULL)
565 		return NULL;
566 
567 	for (p = sbegin; *p; p++) {
568 		if (*p == '\\') {
569 			p++;
570 			if (!*p)
571 				break;
572 		} else if (*p == ',') {
573 			*p = '\0';
574 			*s = p + 1;
575 			return sbegin;
576 		}
577 	}
578 	*s = NULL;
579 	return sbegin;
580 }
581 
582 static int ovl_parse_opt(char *opt, struct ovl_config *config)
583 {
584 	char *p;
585 
586 	while ((p = ovl_next_opt(&opt)) != NULL) {
587 		int token;
588 		substring_t args[MAX_OPT_ARGS];
589 
590 		if (!*p)
591 			continue;
592 
593 		token = match_token(p, ovl_tokens, args);
594 		switch (token) {
595 		case OPT_UPPERDIR:
596 			kfree(config->upperdir);
597 			config->upperdir = match_strdup(&args[0]);
598 			if (!config->upperdir)
599 				return -ENOMEM;
600 			break;
601 
602 		case OPT_LOWERDIR:
603 			kfree(config->lowerdir);
604 			config->lowerdir = match_strdup(&args[0]);
605 			if (!config->lowerdir)
606 				return -ENOMEM;
607 			break;
608 
609 		case OPT_WORKDIR:
610 			kfree(config->workdir);
611 			config->workdir = match_strdup(&args[0]);
612 			if (!config->workdir)
613 				return -ENOMEM;
614 			break;
615 
616 		default:
617 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
618 			return -EINVAL;
619 		}
620 	}
621 
622 	/* Workdir is useless in non-upper mount */
623 	if (!config->upperdir && config->workdir) {
624 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
625 			config->workdir);
626 		kfree(config->workdir);
627 		config->workdir = NULL;
628 	}
629 
630 	return 0;
631 }
632 
633 #define OVL_WORKDIR_NAME "work"
634 
635 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
636 					 struct dentry *dentry)
637 {
638 	struct inode *dir = dentry->d_inode;
639 	struct dentry *work;
640 	int err;
641 	bool retried = false;
642 
643 	err = mnt_want_write(mnt);
644 	if (err)
645 		return ERR_PTR(err);
646 
647 	mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
648 retry:
649 	work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
650 			      strlen(OVL_WORKDIR_NAME));
651 
652 	if (!IS_ERR(work)) {
653 		struct kstat stat = {
654 			.mode = S_IFDIR | 0,
655 		};
656 
657 		if (work->d_inode) {
658 			err = -EEXIST;
659 			if (retried)
660 				goto out_dput;
661 
662 			retried = true;
663 			ovl_cleanup(dir, work);
664 			dput(work);
665 			goto retry;
666 		}
667 
668 		err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
669 		if (err)
670 			goto out_dput;
671 	}
672 out_unlock:
673 	mutex_unlock(&dir->i_mutex);
674 	mnt_drop_write(mnt);
675 
676 	return work;
677 
678 out_dput:
679 	dput(work);
680 	work = ERR_PTR(err);
681 	goto out_unlock;
682 }
683 
684 static void ovl_unescape(char *s)
685 {
686 	char *d = s;
687 
688 	for (;; s++, d++) {
689 		if (*s == '\\')
690 			s++;
691 		*d = *s;
692 		if (!*s)
693 			break;
694 	}
695 }
696 
697 static bool ovl_is_allowed_fs_type(struct dentry *root)
698 {
699 	const struct dentry_operations *dop = root->d_op;
700 
701 	/*
702 	 * We don't support:
703 	 *  - automount filesystems
704 	 *  - filesystems with revalidate (FIXME for lower layer)
705 	 *  - filesystems with case insensitive names
706 	 */
707 	if (dop &&
708 	    (dop->d_manage || dop->d_automount ||
709 	     dop->d_revalidate || dop->d_weak_revalidate ||
710 	     dop->d_compare || dop->d_hash)) {
711 		return false;
712 	}
713 	return true;
714 }
715 
716 static int ovl_mount_dir_noesc(const char *name, struct path *path)
717 {
718 	int err = -EINVAL;
719 
720 	if (!*name) {
721 		pr_err("overlayfs: empty lowerdir\n");
722 		goto out;
723 	}
724 	err = kern_path(name, LOOKUP_FOLLOW, path);
725 	if (err) {
726 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
727 		goto out;
728 	}
729 	err = -EINVAL;
730 	if (!ovl_is_allowed_fs_type(path->dentry)) {
731 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
732 		goto out_put;
733 	}
734 	if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
735 		pr_err("overlayfs: '%s' not a directory\n", name);
736 		goto out_put;
737 	}
738 	return 0;
739 
740 out_put:
741 	path_put(path);
742 out:
743 	return err;
744 }
745 
746 static int ovl_mount_dir(const char *name, struct path *path)
747 {
748 	int err = -ENOMEM;
749 	char *tmp = kstrdup(name, GFP_KERNEL);
750 
751 	if (tmp) {
752 		ovl_unescape(tmp);
753 		err = ovl_mount_dir_noesc(tmp, path);
754 		kfree(tmp);
755 	}
756 	return err;
757 }
758 
759 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
760 			 int *stack_depth)
761 {
762 	int err;
763 	struct kstatfs statfs;
764 
765 	err = ovl_mount_dir_noesc(name, path);
766 	if (err)
767 		goto out;
768 
769 	err = vfs_statfs(path, &statfs);
770 	if (err) {
771 		pr_err("overlayfs: statfs failed on '%s'\n", name);
772 		goto out_put;
773 	}
774 	*namelen = max(*namelen, statfs.f_namelen);
775 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
776 
777 	return 0;
778 
779 out_put:
780 	path_put(path);
781 out:
782 	return err;
783 }
784 
785 /* Workdir should not be subdir of upperdir and vice versa */
786 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
787 {
788 	bool ok = false;
789 
790 	if (workdir != upperdir) {
791 		ok = (lock_rename(workdir, upperdir) == NULL);
792 		unlock_rename(workdir, upperdir);
793 	}
794 	return ok;
795 }
796 
797 static unsigned int ovl_split_lowerdirs(char *str)
798 {
799 	unsigned int ctr = 1;
800 	char *s, *d;
801 
802 	for (s = d = str;; s++, d++) {
803 		if (*s == '\\') {
804 			s++;
805 		} else if (*s == ':') {
806 			*d = '\0';
807 			ctr++;
808 			continue;
809 		}
810 		*d = *s;
811 		if (!*s)
812 			break;
813 	}
814 	return ctr;
815 }
816 
817 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
818 {
819 	struct path upperpath = { NULL, NULL };
820 	struct path workpath = { NULL, NULL };
821 	struct dentry *root_dentry;
822 	struct ovl_entry *oe;
823 	struct ovl_fs *ufs;
824 	struct path *stack = NULL;
825 	char *lowertmp;
826 	char *lower;
827 	unsigned int numlower;
828 	unsigned int stacklen = 0;
829 	unsigned int i;
830 	int err;
831 
832 	err = -ENOMEM;
833 	ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
834 	if (!ufs)
835 		goto out;
836 
837 	err = ovl_parse_opt((char *) data, &ufs->config);
838 	if (err)
839 		goto out_free_config;
840 
841 	err = -EINVAL;
842 	if (!ufs->config.lowerdir) {
843 		pr_err("overlayfs: missing 'lowerdir'\n");
844 		goto out_free_config;
845 	}
846 
847 	sb->s_stack_depth = 0;
848 	if (ufs->config.upperdir) {
849 		if (!ufs->config.workdir) {
850 			pr_err("overlayfs: missing 'workdir'\n");
851 			goto out_free_config;
852 		}
853 
854 		err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
855 		if (err)
856 			goto out_free_config;
857 
858 		/* Upper fs should not be r/o */
859 		if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
860 			pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
861 			err = -EINVAL;
862 			goto out_put_upperpath;
863 		}
864 
865 		err = ovl_mount_dir(ufs->config.workdir, &workpath);
866 		if (err)
867 			goto out_put_upperpath;
868 
869 		err = -EINVAL;
870 		if (upperpath.mnt != workpath.mnt) {
871 			pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
872 			goto out_put_workpath;
873 		}
874 		if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
875 			pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
876 			goto out_put_workpath;
877 		}
878 		sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
879 	}
880 	err = -ENOMEM;
881 	lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
882 	if (!lowertmp)
883 		goto out_put_workpath;
884 
885 	err = -EINVAL;
886 	stacklen = ovl_split_lowerdirs(lowertmp);
887 	if (stacklen > OVL_MAX_STACK) {
888 		pr_err("overlayfs: too many lower directries, limit is %d\n",
889 		       OVL_MAX_STACK);
890 		goto out_free_lowertmp;
891 	} else if (!ufs->config.upperdir && stacklen == 1) {
892 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
893 		goto out_free_lowertmp;
894 	}
895 
896 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
897 	if (!stack)
898 		goto out_free_lowertmp;
899 
900 	lower = lowertmp;
901 	for (numlower = 0; numlower < stacklen; numlower++) {
902 		err = ovl_lower_dir(lower, &stack[numlower],
903 				    &ufs->lower_namelen, &sb->s_stack_depth);
904 		if (err)
905 			goto out_put_lowerpath;
906 
907 		lower = strchr(lower, '\0') + 1;
908 	}
909 
910 	err = -EINVAL;
911 	sb->s_stack_depth++;
912 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
913 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
914 		goto out_put_lowerpath;
915 	}
916 
917 	if (ufs->config.upperdir) {
918 		ufs->upper_mnt = clone_private_mount(&upperpath);
919 		err = PTR_ERR(ufs->upper_mnt);
920 		if (IS_ERR(ufs->upper_mnt)) {
921 			pr_err("overlayfs: failed to clone upperpath\n");
922 			goto out_put_lowerpath;
923 		}
924 
925 		ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
926 		err = PTR_ERR(ufs->workdir);
927 		if (IS_ERR(ufs->workdir)) {
928 			pr_err("overlayfs: failed to create directory %s/%s\n",
929 			       ufs->config.workdir, OVL_WORKDIR_NAME);
930 			goto out_put_upper_mnt;
931 		}
932 	}
933 
934 	err = -ENOMEM;
935 	ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
936 	if (ufs->lower_mnt == NULL)
937 		goto out_put_workdir;
938 	for (i = 0; i < numlower; i++) {
939 		struct vfsmount *mnt = clone_private_mount(&stack[i]);
940 
941 		err = PTR_ERR(mnt);
942 		if (IS_ERR(mnt)) {
943 			pr_err("overlayfs: failed to clone lowerpath\n");
944 			goto out_put_lower_mnt;
945 		}
946 		/*
947 		 * Make lower_mnt R/O.  That way fchmod/fchown on lower file
948 		 * will fail instead of modifying lower fs.
949 		 */
950 		mnt->mnt_flags |= MNT_READONLY;
951 
952 		ufs->lower_mnt[ufs->numlower] = mnt;
953 		ufs->numlower++;
954 	}
955 
956 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
957 	if (!ufs->upper_mnt)
958 		sb->s_flags |= MS_RDONLY;
959 
960 	sb->s_d_op = &ovl_dentry_operations;
961 
962 	err = -ENOMEM;
963 	oe = ovl_alloc_entry(numlower);
964 	if (!oe)
965 		goto out_put_lower_mnt;
966 
967 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
968 	if (!root_dentry)
969 		goto out_free_oe;
970 
971 	mntput(upperpath.mnt);
972 	for (i = 0; i < numlower; i++)
973 		mntput(stack[i].mnt);
974 	path_put(&workpath);
975 	kfree(lowertmp);
976 
977 	oe->__upperdentry = upperpath.dentry;
978 	for (i = 0; i < numlower; i++) {
979 		oe->lowerstack[i].dentry = stack[i].dentry;
980 		oe->lowerstack[i].mnt = ufs->lower_mnt[i];
981 	}
982 
983 	root_dentry->d_fsdata = oe;
984 
985 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
986 	sb->s_op = &ovl_super_operations;
987 	sb->s_root = root_dentry;
988 	sb->s_fs_info = ufs;
989 
990 	return 0;
991 
992 out_free_oe:
993 	kfree(oe);
994 out_put_lower_mnt:
995 	for (i = 0; i < ufs->numlower; i++)
996 		mntput(ufs->lower_mnt[i]);
997 	kfree(ufs->lower_mnt);
998 out_put_workdir:
999 	dput(ufs->workdir);
1000 out_put_upper_mnt:
1001 	mntput(ufs->upper_mnt);
1002 out_put_lowerpath:
1003 	for (i = 0; i < numlower; i++)
1004 		path_put(&stack[i]);
1005 	kfree(stack);
1006 out_free_lowertmp:
1007 	kfree(lowertmp);
1008 out_put_workpath:
1009 	path_put(&workpath);
1010 out_put_upperpath:
1011 	path_put(&upperpath);
1012 out_free_config:
1013 	kfree(ufs->config.lowerdir);
1014 	kfree(ufs->config.upperdir);
1015 	kfree(ufs->config.workdir);
1016 	kfree(ufs);
1017 out:
1018 	return err;
1019 }
1020 
1021 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1022 				const char *dev_name, void *raw_data)
1023 {
1024 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1025 }
1026 
1027 static struct file_system_type ovl_fs_type = {
1028 	.owner		= THIS_MODULE,
1029 	.name		= "overlay",
1030 	.mount		= ovl_mount,
1031 	.kill_sb	= kill_anon_super,
1032 };
1033 MODULE_ALIAS_FS("overlay");
1034 
1035 static int __init ovl_init(void)
1036 {
1037 	return register_filesystem(&ovl_fs_type);
1038 }
1039 
1040 static void __exit ovl_exit(void)
1041 {
1042 	unregister_filesystem(&ovl_fs_type);
1043 }
1044 
1045 module_init(ovl_init);
1046 module_exit(ovl_exit);
1047