xref: /openbmc/linux/fs/overlayfs/super.c (revision 3b23dc52)
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 <uapi/linux/magic.h>
11 #include <linux/fs.h>
12 #include <linux/namei.h>
13 #include <linux/xattr.h>
14 #include <linux/mount.h>
15 #include <linux/parser.h>
16 #include <linux/module.h>
17 #include <linux/statfs.h>
18 #include <linux/seq_file.h>
19 #include <linux/posix_acl_xattr.h>
20 #include <linux/exportfs.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 
28 struct ovl_dir_cache;
29 
30 #define OVL_MAX_STACK 500
31 
32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
34 MODULE_PARM_DESC(ovl_redirect_dir_def,
35 		 "Default to on or off for the redirect_dir feature");
36 
37 static bool ovl_redirect_always_follow =
38 	IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
39 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
40 		   bool, 0644);
41 MODULE_PARM_DESC(ovl_redirect_always_follow,
42 		 "Follow redirects even if redirect_dir feature is turned off");
43 
44 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
45 module_param_named(index, ovl_index_def, bool, 0644);
46 MODULE_PARM_DESC(ovl_index_def,
47 		 "Default to on or off for the inodes index feature");
48 
49 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
50 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
51 MODULE_PARM_DESC(ovl_nfs_export_def,
52 		 "Default to on or off for the NFS export feature");
53 
54 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
55 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
56 MODULE_PARM_DESC(ovl_xino_auto_def,
57 		 "Auto enable xino feature");
58 
59 static void ovl_entry_stack_free(struct ovl_entry *oe)
60 {
61 	unsigned int i;
62 
63 	for (i = 0; i < oe->numlower; i++)
64 		dput(oe->lowerstack[i].dentry);
65 }
66 
67 static void ovl_dentry_release(struct dentry *dentry)
68 {
69 	struct ovl_entry *oe = dentry->d_fsdata;
70 
71 	if (oe) {
72 		ovl_entry_stack_free(oe);
73 		kfree_rcu(oe, rcu);
74 	}
75 }
76 
77 static int ovl_check_append_only(struct inode *inode, int flag)
78 {
79 	/*
80 	 * This test was moot in vfs may_open() because overlay inode does
81 	 * not have the S_APPEND flag, so re-check on real upper inode
82 	 */
83 	if (IS_APPEND(inode)) {
84 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
85 			return -EPERM;
86 		if (flag & O_TRUNC)
87 			return -EPERM;
88 	}
89 
90 	return 0;
91 }
92 
93 static struct dentry *ovl_d_real(struct dentry *dentry,
94 				 const struct inode *inode,
95 				 unsigned int open_flags, unsigned int flags)
96 {
97 	struct dentry *real;
98 	int err;
99 
100 	if (flags & D_REAL_UPPER)
101 		return ovl_dentry_upper(dentry);
102 
103 	if (!d_is_reg(dentry)) {
104 		if (!inode || inode == d_inode(dentry))
105 			return dentry;
106 		goto bug;
107 	}
108 
109 	if (open_flags) {
110 		err = ovl_open_maybe_copy_up(dentry, open_flags);
111 		if (err)
112 			return ERR_PTR(err);
113 	}
114 
115 	real = ovl_dentry_upper(dentry);
116 	if (real && (!inode || inode == d_inode(real))) {
117 		if (!inode) {
118 			err = ovl_check_append_only(d_inode(real), open_flags);
119 			if (err)
120 				return ERR_PTR(err);
121 		}
122 		return real;
123 	}
124 
125 	real = ovl_dentry_lower(dentry);
126 	if (!real)
127 		goto bug;
128 
129 	/* Handle recursion */
130 	real = d_real(real, inode, open_flags, 0);
131 
132 	if (!inode || inode == d_inode(real))
133 		return real;
134 bug:
135 	WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
136 	     inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
137 	return dentry;
138 }
139 
140 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
141 {
142 	struct ovl_entry *oe = dentry->d_fsdata;
143 	unsigned int i;
144 	int ret = 1;
145 
146 	for (i = 0; i < oe->numlower; i++) {
147 		struct dentry *d = oe->lowerstack[i].dentry;
148 
149 		if (d->d_flags & DCACHE_OP_REVALIDATE) {
150 			ret = d->d_op->d_revalidate(d, flags);
151 			if (ret < 0)
152 				return ret;
153 			if (!ret) {
154 				if (!(flags & LOOKUP_RCU))
155 					d_invalidate(d);
156 				return -ESTALE;
157 			}
158 		}
159 	}
160 	return 1;
161 }
162 
163 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
164 {
165 	struct ovl_entry *oe = dentry->d_fsdata;
166 	unsigned int i;
167 	int ret = 1;
168 
169 	for (i = 0; i < oe->numlower; i++) {
170 		struct dentry *d = oe->lowerstack[i].dentry;
171 
172 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
173 			ret = d->d_op->d_weak_revalidate(d, flags);
174 			if (ret <= 0)
175 				break;
176 		}
177 	}
178 	return ret;
179 }
180 
181 static const struct dentry_operations ovl_dentry_operations = {
182 	.d_release = ovl_dentry_release,
183 	.d_real = ovl_d_real,
184 };
185 
186 static const struct dentry_operations ovl_reval_dentry_operations = {
187 	.d_release = ovl_dentry_release,
188 	.d_real = ovl_d_real,
189 	.d_revalidate = ovl_dentry_revalidate,
190 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
191 };
192 
193 static struct kmem_cache *ovl_inode_cachep;
194 
195 static struct inode *ovl_alloc_inode(struct super_block *sb)
196 {
197 	struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
198 
199 	if (!oi)
200 		return NULL;
201 
202 	oi->cache = NULL;
203 	oi->redirect = NULL;
204 	oi->version = 0;
205 	oi->flags = 0;
206 	oi->__upperdentry = NULL;
207 	oi->lower = NULL;
208 	mutex_init(&oi->lock);
209 
210 	return &oi->vfs_inode;
211 }
212 
213 static void ovl_i_callback(struct rcu_head *head)
214 {
215 	struct inode *inode = container_of(head, struct inode, i_rcu);
216 
217 	kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
218 }
219 
220 static void ovl_destroy_inode(struct inode *inode)
221 {
222 	struct ovl_inode *oi = OVL_I(inode);
223 
224 	dput(oi->__upperdentry);
225 	iput(oi->lower);
226 	kfree(oi->redirect);
227 	ovl_dir_cache_free(inode);
228 	mutex_destroy(&oi->lock);
229 
230 	call_rcu(&inode->i_rcu, ovl_i_callback);
231 }
232 
233 static void ovl_free_fs(struct ovl_fs *ofs)
234 {
235 	unsigned i;
236 
237 	dput(ofs->indexdir);
238 	dput(ofs->workdir);
239 	if (ofs->workdir_locked)
240 		ovl_inuse_unlock(ofs->workbasedir);
241 	dput(ofs->workbasedir);
242 	if (ofs->upperdir_locked)
243 		ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
244 	mntput(ofs->upper_mnt);
245 	for (i = 0; i < ofs->numlower; i++)
246 		mntput(ofs->lower_layers[i].mnt);
247 	for (i = 0; i < ofs->numlowerfs; i++)
248 		free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
249 	kfree(ofs->lower_layers);
250 	kfree(ofs->lower_fs);
251 
252 	kfree(ofs->config.lowerdir);
253 	kfree(ofs->config.upperdir);
254 	kfree(ofs->config.workdir);
255 	kfree(ofs->config.redirect_mode);
256 	if (ofs->creator_cred)
257 		put_cred(ofs->creator_cred);
258 	kfree(ofs);
259 }
260 
261 static void ovl_put_super(struct super_block *sb)
262 {
263 	struct ovl_fs *ofs = sb->s_fs_info;
264 
265 	ovl_free_fs(ofs);
266 }
267 
268 /* Sync real dirty inodes in upper filesystem (if it exists) */
269 static int ovl_sync_fs(struct super_block *sb, int wait)
270 {
271 	struct ovl_fs *ofs = sb->s_fs_info;
272 	struct super_block *upper_sb;
273 	int ret;
274 
275 	if (!ofs->upper_mnt)
276 		return 0;
277 
278 	/*
279 	 * If this is a sync(2) call or an emergency sync, all the super blocks
280 	 * will be iterated, including upper_sb, so no need to do anything.
281 	 *
282 	 * If this is a syncfs(2) call, then we do need to call
283 	 * sync_filesystem() on upper_sb, but enough if we do it when being
284 	 * called with wait == 1.
285 	 */
286 	if (!wait)
287 		return 0;
288 
289 	upper_sb = ofs->upper_mnt->mnt_sb;
290 
291 	down_read(&upper_sb->s_umount);
292 	ret = sync_filesystem(upper_sb);
293 	up_read(&upper_sb->s_umount);
294 
295 	return ret;
296 }
297 
298 /**
299  * ovl_statfs
300  * @sb: The overlayfs super block
301  * @buf: The struct kstatfs to fill in with stats
302  *
303  * Get the filesystem statistics.  As writes always target the upper layer
304  * filesystem pass the statfs to the upper filesystem (if it exists)
305  */
306 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
307 {
308 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
309 	struct dentry *root_dentry = dentry->d_sb->s_root;
310 	struct path path;
311 	int err;
312 
313 	ovl_path_real(root_dentry, &path);
314 
315 	err = vfs_statfs(&path, buf);
316 	if (!err) {
317 		buf->f_namelen = ofs->namelen;
318 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
319 	}
320 
321 	return err;
322 }
323 
324 /* Will this overlay be forced to mount/remount ro? */
325 static bool ovl_force_readonly(struct ovl_fs *ofs)
326 {
327 	return (!ofs->upper_mnt || !ofs->workdir);
328 }
329 
330 static const char *ovl_redirect_mode_def(void)
331 {
332 	return ovl_redirect_dir_def ? "on" : "off";
333 }
334 
335 enum {
336 	OVL_XINO_OFF,
337 	OVL_XINO_AUTO,
338 	OVL_XINO_ON,
339 };
340 
341 static const char * const ovl_xino_str[] = {
342 	"off",
343 	"auto",
344 	"on",
345 };
346 
347 static inline int ovl_xino_def(void)
348 {
349 	return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
350 }
351 
352 /**
353  * ovl_show_options
354  *
355  * Prints the mount options for a given superblock.
356  * Returns zero; does not fail.
357  */
358 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
359 {
360 	struct super_block *sb = dentry->d_sb;
361 	struct ovl_fs *ofs = sb->s_fs_info;
362 
363 	seq_show_option(m, "lowerdir", ofs->config.lowerdir);
364 	if (ofs->config.upperdir) {
365 		seq_show_option(m, "upperdir", ofs->config.upperdir);
366 		seq_show_option(m, "workdir", ofs->config.workdir);
367 	}
368 	if (ofs->config.default_permissions)
369 		seq_puts(m, ",default_permissions");
370 	if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
371 		seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
372 	if (ofs->config.index != ovl_index_def)
373 		seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
374 	if (ofs->config.nfs_export != ovl_nfs_export_def)
375 		seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
376 						"on" : "off");
377 	if (ofs->config.xino != ovl_xino_def())
378 		seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
379 	return 0;
380 }
381 
382 static int ovl_remount(struct super_block *sb, int *flags, char *data)
383 {
384 	struct ovl_fs *ofs = sb->s_fs_info;
385 
386 	if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
387 		return -EROFS;
388 
389 	return 0;
390 }
391 
392 static const struct super_operations ovl_super_operations = {
393 	.alloc_inode	= ovl_alloc_inode,
394 	.destroy_inode	= ovl_destroy_inode,
395 	.drop_inode	= generic_delete_inode,
396 	.put_super	= ovl_put_super,
397 	.sync_fs	= ovl_sync_fs,
398 	.statfs		= ovl_statfs,
399 	.show_options	= ovl_show_options,
400 	.remount_fs	= ovl_remount,
401 };
402 
403 enum {
404 	OPT_LOWERDIR,
405 	OPT_UPPERDIR,
406 	OPT_WORKDIR,
407 	OPT_DEFAULT_PERMISSIONS,
408 	OPT_REDIRECT_DIR,
409 	OPT_INDEX_ON,
410 	OPT_INDEX_OFF,
411 	OPT_NFS_EXPORT_ON,
412 	OPT_NFS_EXPORT_OFF,
413 	OPT_XINO_ON,
414 	OPT_XINO_OFF,
415 	OPT_XINO_AUTO,
416 	OPT_ERR,
417 };
418 
419 static const match_table_t ovl_tokens = {
420 	{OPT_LOWERDIR,			"lowerdir=%s"},
421 	{OPT_UPPERDIR,			"upperdir=%s"},
422 	{OPT_WORKDIR,			"workdir=%s"},
423 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
424 	{OPT_REDIRECT_DIR,		"redirect_dir=%s"},
425 	{OPT_INDEX_ON,			"index=on"},
426 	{OPT_INDEX_OFF,			"index=off"},
427 	{OPT_NFS_EXPORT_ON,		"nfs_export=on"},
428 	{OPT_NFS_EXPORT_OFF,		"nfs_export=off"},
429 	{OPT_XINO_ON,			"xino=on"},
430 	{OPT_XINO_OFF,			"xino=off"},
431 	{OPT_XINO_AUTO,			"xino=auto"},
432 	{OPT_ERR,			NULL}
433 };
434 
435 static char *ovl_next_opt(char **s)
436 {
437 	char *sbegin = *s;
438 	char *p;
439 
440 	if (sbegin == NULL)
441 		return NULL;
442 
443 	for (p = sbegin; *p; p++) {
444 		if (*p == '\\') {
445 			p++;
446 			if (!*p)
447 				break;
448 		} else if (*p == ',') {
449 			*p = '\0';
450 			*s = p + 1;
451 			return sbegin;
452 		}
453 	}
454 	*s = NULL;
455 	return sbegin;
456 }
457 
458 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
459 {
460 	if (strcmp(mode, "on") == 0) {
461 		config->redirect_dir = true;
462 		/*
463 		 * Does not make sense to have redirect creation without
464 		 * redirect following.
465 		 */
466 		config->redirect_follow = true;
467 	} else if (strcmp(mode, "follow") == 0) {
468 		config->redirect_follow = true;
469 	} else if (strcmp(mode, "off") == 0) {
470 		if (ovl_redirect_always_follow)
471 			config->redirect_follow = true;
472 	} else if (strcmp(mode, "nofollow") != 0) {
473 		pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
474 		       mode);
475 		return -EINVAL;
476 	}
477 
478 	return 0;
479 }
480 
481 static int ovl_parse_opt(char *opt, struct ovl_config *config)
482 {
483 	char *p;
484 
485 	config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
486 	if (!config->redirect_mode)
487 		return -ENOMEM;
488 
489 	while ((p = ovl_next_opt(&opt)) != NULL) {
490 		int token;
491 		substring_t args[MAX_OPT_ARGS];
492 
493 		if (!*p)
494 			continue;
495 
496 		token = match_token(p, ovl_tokens, args);
497 		switch (token) {
498 		case OPT_UPPERDIR:
499 			kfree(config->upperdir);
500 			config->upperdir = match_strdup(&args[0]);
501 			if (!config->upperdir)
502 				return -ENOMEM;
503 			break;
504 
505 		case OPT_LOWERDIR:
506 			kfree(config->lowerdir);
507 			config->lowerdir = match_strdup(&args[0]);
508 			if (!config->lowerdir)
509 				return -ENOMEM;
510 			break;
511 
512 		case OPT_WORKDIR:
513 			kfree(config->workdir);
514 			config->workdir = match_strdup(&args[0]);
515 			if (!config->workdir)
516 				return -ENOMEM;
517 			break;
518 
519 		case OPT_DEFAULT_PERMISSIONS:
520 			config->default_permissions = true;
521 			break;
522 
523 		case OPT_REDIRECT_DIR:
524 			kfree(config->redirect_mode);
525 			config->redirect_mode = match_strdup(&args[0]);
526 			if (!config->redirect_mode)
527 				return -ENOMEM;
528 			break;
529 
530 		case OPT_INDEX_ON:
531 			config->index = true;
532 			break;
533 
534 		case OPT_INDEX_OFF:
535 			config->index = false;
536 			break;
537 
538 		case OPT_NFS_EXPORT_ON:
539 			config->nfs_export = true;
540 			break;
541 
542 		case OPT_NFS_EXPORT_OFF:
543 			config->nfs_export = false;
544 			break;
545 
546 		case OPT_XINO_ON:
547 			config->xino = OVL_XINO_ON;
548 			break;
549 
550 		case OPT_XINO_OFF:
551 			config->xino = OVL_XINO_OFF;
552 			break;
553 
554 		case OPT_XINO_AUTO:
555 			config->xino = OVL_XINO_AUTO;
556 			break;
557 
558 		default:
559 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
560 			return -EINVAL;
561 		}
562 	}
563 
564 	/* Workdir is useless in non-upper mount */
565 	if (!config->upperdir && config->workdir) {
566 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
567 			config->workdir);
568 		kfree(config->workdir);
569 		config->workdir = NULL;
570 	}
571 
572 	return ovl_parse_redirect_mode(config, config->redirect_mode);
573 }
574 
575 #define OVL_WORKDIR_NAME "work"
576 #define OVL_INDEXDIR_NAME "index"
577 
578 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
579 					 const char *name, bool persist)
580 {
581 	struct inode *dir =  ofs->workbasedir->d_inode;
582 	struct vfsmount *mnt = ofs->upper_mnt;
583 	struct dentry *work;
584 	int err;
585 	bool retried = false;
586 	bool locked = false;
587 
588 	inode_lock_nested(dir, I_MUTEX_PARENT);
589 	locked = true;
590 
591 retry:
592 	work = lookup_one_len(name, ofs->workbasedir, strlen(name));
593 
594 	if (!IS_ERR(work)) {
595 		struct iattr attr = {
596 			.ia_valid = ATTR_MODE,
597 			.ia_mode = S_IFDIR | 0,
598 		};
599 
600 		if (work->d_inode) {
601 			err = -EEXIST;
602 			if (retried)
603 				goto out_dput;
604 
605 			if (persist)
606 				goto out_unlock;
607 
608 			retried = true;
609 			ovl_workdir_cleanup(dir, mnt, work, 0);
610 			dput(work);
611 			goto retry;
612 		}
613 
614 		work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
615 		err = PTR_ERR(work);
616 		if (IS_ERR(work))
617 			goto out_err;
618 
619 		/*
620 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
621 		 *
622 		 * a) success (there was a POSIX ACL xattr and was removed)
623 		 * b) -ENODATA (there was no POSIX ACL xattr)
624 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
625 		 *
626 		 * There are various other error values that could effectively
627 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
628 		 * if the xattr name is too long), but the set of filesystems
629 		 * allowed as upper are limited to "normal" ones, where checking
630 		 * for the above two errors is sufficient.
631 		 */
632 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
633 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
634 			goto out_dput;
635 
636 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
637 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
638 			goto out_dput;
639 
640 		/* Clear any inherited mode bits */
641 		inode_lock(work->d_inode);
642 		err = notify_change(work, &attr, NULL);
643 		inode_unlock(work->d_inode);
644 		if (err)
645 			goto out_dput;
646 	} else {
647 		err = PTR_ERR(work);
648 		goto out_err;
649 	}
650 out_unlock:
651 	if (locked)
652 		inode_unlock(dir);
653 
654 	return work;
655 
656 out_dput:
657 	dput(work);
658 out_err:
659 	pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
660 		ofs->config.workdir, name, -err);
661 	work = NULL;
662 	goto out_unlock;
663 }
664 
665 static void ovl_unescape(char *s)
666 {
667 	char *d = s;
668 
669 	for (;; s++, d++) {
670 		if (*s == '\\')
671 			s++;
672 		*d = *s;
673 		if (!*s)
674 			break;
675 	}
676 }
677 
678 static int ovl_mount_dir_noesc(const char *name, struct path *path)
679 {
680 	int err = -EINVAL;
681 
682 	if (!*name) {
683 		pr_err("overlayfs: empty lowerdir\n");
684 		goto out;
685 	}
686 	err = kern_path(name, LOOKUP_FOLLOW, path);
687 	if (err) {
688 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
689 		goto out;
690 	}
691 	err = -EINVAL;
692 	if (ovl_dentry_weird(path->dentry)) {
693 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
694 		goto out_put;
695 	}
696 	if (!d_is_dir(path->dentry)) {
697 		pr_err("overlayfs: '%s' not a directory\n", name);
698 		goto out_put;
699 	}
700 	return 0;
701 
702 out_put:
703 	path_put_init(path);
704 out:
705 	return err;
706 }
707 
708 static int ovl_mount_dir(const char *name, struct path *path)
709 {
710 	int err = -ENOMEM;
711 	char *tmp = kstrdup(name, GFP_KERNEL);
712 
713 	if (tmp) {
714 		ovl_unescape(tmp);
715 		err = ovl_mount_dir_noesc(tmp, path);
716 
717 		if (!err)
718 			if (ovl_dentry_remote(path->dentry)) {
719 				pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
720 				       tmp);
721 				path_put_init(path);
722 				err = -EINVAL;
723 			}
724 		kfree(tmp);
725 	}
726 	return err;
727 }
728 
729 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
730 			     const char *name)
731 {
732 	struct kstatfs statfs;
733 	int err = vfs_statfs(path, &statfs);
734 
735 	if (err)
736 		pr_err("overlayfs: statfs failed on '%s'\n", name);
737 	else
738 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
739 
740 	return err;
741 }
742 
743 static int ovl_lower_dir(const char *name, struct path *path,
744 			 struct ovl_fs *ofs, int *stack_depth, bool *remote)
745 {
746 	int fh_type;
747 	int err;
748 
749 	err = ovl_mount_dir_noesc(name, path);
750 	if (err)
751 		goto out;
752 
753 	err = ovl_check_namelen(path, ofs, name);
754 	if (err)
755 		goto out_put;
756 
757 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
758 
759 	if (ovl_dentry_remote(path->dentry))
760 		*remote = true;
761 
762 	/*
763 	 * The inodes index feature and NFS export need to encode and decode
764 	 * file handles, so they require that all layers support them.
765 	 */
766 	fh_type = ovl_can_decode_fh(path->dentry->d_sb);
767 	if ((ofs->config.nfs_export ||
768 	     (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
769 		ofs->config.index = false;
770 		ofs->config.nfs_export = false;
771 		pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
772 			name);
773 	}
774 
775 	/* Check if lower fs has 32bit inode numbers */
776 	if (fh_type != FILEID_INO32_GEN)
777 		ofs->xino_bits = 0;
778 
779 	return 0;
780 
781 out_put:
782 	path_put_init(path);
783 out:
784 	return err;
785 }
786 
787 /* Workdir should not be subdir of upperdir and vice versa */
788 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
789 {
790 	bool ok = false;
791 
792 	if (workdir != upperdir) {
793 		ok = (lock_rename(workdir, upperdir) == NULL);
794 		unlock_rename(workdir, upperdir);
795 	}
796 	return ok;
797 }
798 
799 static unsigned int ovl_split_lowerdirs(char *str)
800 {
801 	unsigned int ctr = 1;
802 	char *s, *d;
803 
804 	for (s = d = str;; s++, d++) {
805 		if (*s == '\\') {
806 			s++;
807 		} else if (*s == ':') {
808 			*d = '\0';
809 			ctr++;
810 			continue;
811 		}
812 		*d = *s;
813 		if (!*s)
814 			break;
815 	}
816 	return ctr;
817 }
818 
819 static int __maybe_unused
820 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
821 			struct dentry *dentry, struct inode *inode,
822 			const char *name, void *buffer, size_t size)
823 {
824 	return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
825 }
826 
827 static int __maybe_unused
828 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
829 			struct dentry *dentry, struct inode *inode,
830 			const char *name, const void *value,
831 			size_t size, int flags)
832 {
833 	struct dentry *workdir = ovl_workdir(dentry);
834 	struct inode *realinode = ovl_inode_real(inode);
835 	struct posix_acl *acl = NULL;
836 	int err;
837 
838 	/* Check that everything is OK before copy-up */
839 	if (value) {
840 		acl = posix_acl_from_xattr(&init_user_ns, value, size);
841 		if (IS_ERR(acl))
842 			return PTR_ERR(acl);
843 	}
844 	err = -EOPNOTSUPP;
845 	if (!IS_POSIXACL(d_inode(workdir)))
846 		goto out_acl_release;
847 	if (!realinode->i_op->set_acl)
848 		goto out_acl_release;
849 	if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
850 		err = acl ? -EACCES : 0;
851 		goto out_acl_release;
852 	}
853 	err = -EPERM;
854 	if (!inode_owner_or_capable(inode))
855 		goto out_acl_release;
856 
857 	posix_acl_release(acl);
858 
859 	/*
860 	 * Check if sgid bit needs to be cleared (actual setacl operation will
861 	 * be done with mounter's capabilities and so that won't do it for us).
862 	 */
863 	if (unlikely(inode->i_mode & S_ISGID) &&
864 	    handler->flags == ACL_TYPE_ACCESS &&
865 	    !in_group_p(inode->i_gid) &&
866 	    !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
867 		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
868 
869 		err = ovl_setattr(dentry, &iattr);
870 		if (err)
871 			return err;
872 	}
873 
874 	err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
875 	if (!err)
876 		ovl_copyattr(ovl_inode_real(inode), inode);
877 
878 	return err;
879 
880 out_acl_release:
881 	posix_acl_release(acl);
882 	return err;
883 }
884 
885 static int ovl_own_xattr_get(const struct xattr_handler *handler,
886 			     struct dentry *dentry, struct inode *inode,
887 			     const char *name, void *buffer, size_t size)
888 {
889 	return -EOPNOTSUPP;
890 }
891 
892 static int ovl_own_xattr_set(const struct xattr_handler *handler,
893 			     struct dentry *dentry, struct inode *inode,
894 			     const char *name, const void *value,
895 			     size_t size, int flags)
896 {
897 	return -EOPNOTSUPP;
898 }
899 
900 static int ovl_other_xattr_get(const struct xattr_handler *handler,
901 			       struct dentry *dentry, struct inode *inode,
902 			       const char *name, void *buffer, size_t size)
903 {
904 	return ovl_xattr_get(dentry, inode, name, buffer, size);
905 }
906 
907 static int ovl_other_xattr_set(const struct xattr_handler *handler,
908 			       struct dentry *dentry, struct inode *inode,
909 			       const char *name, const void *value,
910 			       size_t size, int flags)
911 {
912 	return ovl_xattr_set(dentry, inode, name, value, size, flags);
913 }
914 
915 static const struct xattr_handler __maybe_unused
916 ovl_posix_acl_access_xattr_handler = {
917 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
918 	.flags = ACL_TYPE_ACCESS,
919 	.get = ovl_posix_acl_xattr_get,
920 	.set = ovl_posix_acl_xattr_set,
921 };
922 
923 static const struct xattr_handler __maybe_unused
924 ovl_posix_acl_default_xattr_handler = {
925 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
926 	.flags = ACL_TYPE_DEFAULT,
927 	.get = ovl_posix_acl_xattr_get,
928 	.set = ovl_posix_acl_xattr_set,
929 };
930 
931 static const struct xattr_handler ovl_own_xattr_handler = {
932 	.prefix	= OVL_XATTR_PREFIX,
933 	.get = ovl_own_xattr_get,
934 	.set = ovl_own_xattr_set,
935 };
936 
937 static const struct xattr_handler ovl_other_xattr_handler = {
938 	.prefix	= "", /* catch all */
939 	.get = ovl_other_xattr_get,
940 	.set = ovl_other_xattr_set,
941 };
942 
943 static const struct xattr_handler *ovl_xattr_handlers[] = {
944 #ifdef CONFIG_FS_POSIX_ACL
945 	&ovl_posix_acl_access_xattr_handler,
946 	&ovl_posix_acl_default_xattr_handler,
947 #endif
948 	&ovl_own_xattr_handler,
949 	&ovl_other_xattr_handler,
950 	NULL
951 };
952 
953 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
954 {
955 	struct vfsmount *upper_mnt;
956 	int err;
957 
958 	err = ovl_mount_dir(ofs->config.upperdir, upperpath);
959 	if (err)
960 		goto out;
961 
962 	/* Upper fs should not be r/o */
963 	if (sb_rdonly(upperpath->mnt->mnt_sb)) {
964 		pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
965 		err = -EINVAL;
966 		goto out;
967 	}
968 
969 	err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
970 	if (err)
971 		goto out;
972 
973 	err = -EBUSY;
974 	if (ovl_inuse_trylock(upperpath->dentry)) {
975 		ofs->upperdir_locked = true;
976 	} else if (ofs->config.index) {
977 		pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
978 		goto out;
979 	} else {
980 		pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
981 	}
982 
983 	upper_mnt = clone_private_mount(upperpath);
984 	err = PTR_ERR(upper_mnt);
985 	if (IS_ERR(upper_mnt)) {
986 		pr_err("overlayfs: failed to clone upperpath\n");
987 		goto out;
988 	}
989 
990 	/* Don't inherit atime flags */
991 	upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
992 	ofs->upper_mnt = upper_mnt;
993 	err = 0;
994 out:
995 	return err;
996 }
997 
998 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
999 {
1000 	struct vfsmount *mnt = ofs->upper_mnt;
1001 	struct dentry *temp;
1002 	int fh_type;
1003 	int err;
1004 
1005 	err = mnt_want_write(mnt);
1006 	if (err)
1007 		return err;
1008 
1009 	ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1010 	if (!ofs->workdir)
1011 		goto out;
1012 
1013 	/*
1014 	 * Upper should support d_type, else whiteouts are visible.  Given
1015 	 * workdir and upper are on same fs, we can do iterate_dir() on
1016 	 * workdir. This check requires successful creation of workdir in
1017 	 * previous step.
1018 	 */
1019 	err = ovl_check_d_type_supported(workpath);
1020 	if (err < 0)
1021 		goto out;
1022 
1023 	/*
1024 	 * We allowed this configuration and don't want to break users over
1025 	 * kernel upgrade. So warn instead of erroring out.
1026 	 */
1027 	if (!err)
1028 		pr_warn("overlayfs: upper fs needs to support d_type.\n");
1029 
1030 	/* Check if upper/work fs supports O_TMPFILE */
1031 	temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1032 	ofs->tmpfile = !IS_ERR(temp);
1033 	if (ofs->tmpfile)
1034 		dput(temp);
1035 	else
1036 		pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1037 
1038 	/*
1039 	 * Check if upper/work fs supports trusted.overlay.* xattr
1040 	 */
1041 	err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1042 	if (err) {
1043 		ofs->noxattr = true;
1044 		ofs->config.index = false;
1045 		pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off.\n");
1046 		err = 0;
1047 	} else {
1048 		vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1049 	}
1050 
1051 	/* Check if upper/work fs supports file handles */
1052 	fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1053 	if (ofs->config.index && !fh_type) {
1054 		ofs->config.index = false;
1055 		pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1056 	}
1057 
1058 	/* Check if upper fs has 32bit inode numbers */
1059 	if (fh_type != FILEID_INO32_GEN)
1060 		ofs->xino_bits = 0;
1061 
1062 	/* NFS export of r/w mount depends on index */
1063 	if (ofs->config.nfs_export && !ofs->config.index) {
1064 		pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1065 		ofs->config.nfs_export = false;
1066 	}
1067 
1068 out:
1069 	mnt_drop_write(mnt);
1070 	return err;
1071 }
1072 
1073 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1074 {
1075 	int err;
1076 	struct path workpath = { };
1077 
1078 	err = ovl_mount_dir(ofs->config.workdir, &workpath);
1079 	if (err)
1080 		goto out;
1081 
1082 	err = -EINVAL;
1083 	if (upperpath->mnt != workpath.mnt) {
1084 		pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1085 		goto out;
1086 	}
1087 	if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1088 		pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1089 		goto out;
1090 	}
1091 
1092 	err = -EBUSY;
1093 	if (ovl_inuse_trylock(workpath.dentry)) {
1094 		ofs->workdir_locked = true;
1095 	} else if (ofs->config.index) {
1096 		pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1097 		goto out;
1098 	} else {
1099 		pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1100 	}
1101 
1102 	ofs->workbasedir = dget(workpath.dentry);
1103 	err = ovl_make_workdir(ofs, &workpath);
1104 	if (err)
1105 		goto out;
1106 
1107 	err = 0;
1108 out:
1109 	path_put(&workpath);
1110 
1111 	return err;
1112 }
1113 
1114 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1115 			    struct path *upperpath)
1116 {
1117 	struct vfsmount *mnt = ofs->upper_mnt;
1118 	int err;
1119 
1120 	err = mnt_want_write(mnt);
1121 	if (err)
1122 		return err;
1123 
1124 	/* Verify lower root is upper root origin */
1125 	err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1126 				true);
1127 	if (err) {
1128 		pr_err("overlayfs: failed to verify upper root origin\n");
1129 		goto out;
1130 	}
1131 
1132 	ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1133 	if (ofs->indexdir) {
1134 		/*
1135 		 * Verify upper root is exclusively associated with index dir.
1136 		 * Older kernels stored upper fh in "trusted.overlay.origin"
1137 		 * xattr. If that xattr exists, verify that it is a match to
1138 		 * upper dir file handle. In any case, verify or set xattr
1139 		 * "trusted.overlay.upper" to indicate that index may have
1140 		 * directory entries.
1141 		 */
1142 		if (ovl_check_origin_xattr(ofs->indexdir)) {
1143 			err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1144 						upperpath->dentry, true, false);
1145 			if (err)
1146 				pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1147 		}
1148 		err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1149 		if (err)
1150 			pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1151 
1152 		/* Cleanup bad/stale/orphan index entries */
1153 		if (!err)
1154 			err = ovl_indexdir_cleanup(ofs);
1155 	}
1156 	if (err || !ofs->indexdir)
1157 		pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1158 
1159 out:
1160 	mnt_drop_write(mnt);
1161 	return err;
1162 }
1163 
1164 /* Get a unique fsid for the layer */
1165 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1166 {
1167 	unsigned int i;
1168 	dev_t dev;
1169 	int err;
1170 
1171 	/* fsid 0 is reserved for upper fs even with non upper overlay */
1172 	if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1173 		return 0;
1174 
1175 	for (i = 0; i < ofs->numlowerfs; i++) {
1176 		if (ofs->lower_fs[i].sb == sb)
1177 			return i + 1;
1178 	}
1179 
1180 	err = get_anon_bdev(&dev);
1181 	if (err) {
1182 		pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1183 		return err;
1184 	}
1185 
1186 	ofs->lower_fs[ofs->numlowerfs].sb = sb;
1187 	ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1188 	ofs->numlowerfs++;
1189 
1190 	return ofs->numlowerfs;
1191 }
1192 
1193 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1194 				unsigned int numlower)
1195 {
1196 	int err;
1197 	unsigned int i;
1198 
1199 	err = -ENOMEM;
1200 	ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1201 				    GFP_KERNEL);
1202 	if (ofs->lower_layers == NULL)
1203 		goto out;
1204 
1205 	ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1206 				GFP_KERNEL);
1207 	if (ofs->lower_fs == NULL)
1208 		goto out;
1209 
1210 	for (i = 0; i < numlower; i++) {
1211 		struct vfsmount *mnt;
1212 		int fsid;
1213 
1214 		err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1215 		if (err < 0)
1216 			goto out;
1217 
1218 		mnt = clone_private_mount(&stack[i]);
1219 		err = PTR_ERR(mnt);
1220 		if (IS_ERR(mnt)) {
1221 			pr_err("overlayfs: failed to clone lowerpath\n");
1222 			goto out;
1223 		}
1224 
1225 		/*
1226 		 * Make lower layers R/O.  That way fchmod/fchown on lower file
1227 		 * will fail instead of modifying lower fs.
1228 		 */
1229 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1230 
1231 		ofs->lower_layers[ofs->numlower].mnt = mnt;
1232 		ofs->lower_layers[ofs->numlower].idx = i + 1;
1233 		ofs->lower_layers[ofs->numlower].fsid = fsid;
1234 		if (fsid) {
1235 			ofs->lower_layers[ofs->numlower].fs =
1236 				&ofs->lower_fs[fsid - 1];
1237 		}
1238 		ofs->numlower++;
1239 	}
1240 
1241 	/*
1242 	 * When all layers on same fs, overlay can use real inode numbers.
1243 	 * With mount option "xino=on", mounter declares that there are enough
1244 	 * free high bits in underlying fs to hold the unique fsid.
1245 	 * If overlayfs does encounter underlying inodes using the high xino
1246 	 * bits reserved for fsid, it emits a warning and uses the original
1247 	 * inode number.
1248 	 */
1249 	if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1250 		ofs->xino_bits = 0;
1251 		ofs->config.xino = OVL_XINO_OFF;
1252 	} else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1253 		/*
1254 		 * This is a roundup of number of bits needed for numlowerfs+1
1255 		 * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1256 		 * upper fs even with non upper overlay.
1257 		 */
1258 		BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1259 		ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1260 	}
1261 
1262 	if (ofs->xino_bits) {
1263 		pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1264 			ofs->xino_bits);
1265 	}
1266 
1267 	err = 0;
1268 out:
1269 	return err;
1270 }
1271 
1272 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1273 					    struct ovl_fs *ofs)
1274 {
1275 	int err;
1276 	char *lowertmp, *lower;
1277 	struct path *stack = NULL;
1278 	unsigned int stacklen, numlower = 0, i;
1279 	bool remote = false;
1280 	struct ovl_entry *oe;
1281 
1282 	err = -ENOMEM;
1283 	lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1284 	if (!lowertmp)
1285 		goto out_err;
1286 
1287 	err = -EINVAL;
1288 	stacklen = ovl_split_lowerdirs(lowertmp);
1289 	if (stacklen > OVL_MAX_STACK) {
1290 		pr_err("overlayfs: too many lower directories, limit is %d\n",
1291 		       OVL_MAX_STACK);
1292 		goto out_err;
1293 	} else if (!ofs->config.upperdir && stacklen == 1) {
1294 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1295 		goto out_err;
1296 	} else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1297 		   ofs->config.redirect_follow) {
1298 		pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1299 		ofs->config.nfs_export = false;
1300 	}
1301 
1302 	err = -ENOMEM;
1303 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1304 	if (!stack)
1305 		goto out_err;
1306 
1307 	err = -EINVAL;
1308 	lower = lowertmp;
1309 	for (numlower = 0; numlower < stacklen; numlower++) {
1310 		err = ovl_lower_dir(lower, &stack[numlower], ofs,
1311 				    &sb->s_stack_depth, &remote);
1312 		if (err)
1313 			goto out_err;
1314 
1315 		lower = strchr(lower, '\0') + 1;
1316 	}
1317 
1318 	err = -EINVAL;
1319 	sb->s_stack_depth++;
1320 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1321 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1322 		goto out_err;
1323 	}
1324 
1325 	err = ovl_get_lower_layers(ofs, stack, numlower);
1326 	if (err)
1327 		goto out_err;
1328 
1329 	err = -ENOMEM;
1330 	oe = ovl_alloc_entry(numlower);
1331 	if (!oe)
1332 		goto out_err;
1333 
1334 	for (i = 0; i < numlower; i++) {
1335 		oe->lowerstack[i].dentry = dget(stack[i].dentry);
1336 		oe->lowerstack[i].layer = &ofs->lower_layers[i];
1337 	}
1338 
1339 	if (remote)
1340 		sb->s_d_op = &ovl_reval_dentry_operations;
1341 	else
1342 		sb->s_d_op = &ovl_dentry_operations;
1343 
1344 out:
1345 	for (i = 0; i < numlower; i++)
1346 		path_put(&stack[i]);
1347 	kfree(stack);
1348 	kfree(lowertmp);
1349 
1350 	return oe;
1351 
1352 out_err:
1353 	oe = ERR_PTR(err);
1354 	goto out;
1355 }
1356 
1357 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1358 {
1359 	struct path upperpath = { };
1360 	struct dentry *root_dentry;
1361 	struct ovl_entry *oe;
1362 	struct ovl_fs *ofs;
1363 	struct cred *cred;
1364 	int err;
1365 
1366 	err = -ENOMEM;
1367 	ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1368 	if (!ofs)
1369 		goto out;
1370 
1371 	ofs->creator_cred = cred = prepare_creds();
1372 	if (!cred)
1373 		goto out_err;
1374 
1375 	ofs->config.index = ovl_index_def;
1376 	ofs->config.nfs_export = ovl_nfs_export_def;
1377 	ofs->config.xino = ovl_xino_def();
1378 	err = ovl_parse_opt((char *) data, &ofs->config);
1379 	if (err)
1380 		goto out_err;
1381 
1382 	err = -EINVAL;
1383 	if (!ofs->config.lowerdir) {
1384 		if (!silent)
1385 			pr_err("overlayfs: missing 'lowerdir'\n");
1386 		goto out_err;
1387 	}
1388 
1389 	sb->s_stack_depth = 0;
1390 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1391 	/* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1392 	if (ofs->config.xino != OVL_XINO_OFF)
1393 		ofs->xino_bits = BITS_PER_LONG - 32;
1394 
1395 	if (ofs->config.upperdir) {
1396 		if (!ofs->config.workdir) {
1397 			pr_err("overlayfs: missing 'workdir'\n");
1398 			goto out_err;
1399 		}
1400 
1401 		err = ovl_get_upper(ofs, &upperpath);
1402 		if (err)
1403 			goto out_err;
1404 
1405 		err = ovl_get_workdir(ofs, &upperpath);
1406 		if (err)
1407 			goto out_err;
1408 
1409 		if (!ofs->workdir)
1410 			sb->s_flags |= SB_RDONLY;
1411 
1412 		sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1413 		sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1414 
1415 	}
1416 	oe = ovl_get_lowerstack(sb, ofs);
1417 	err = PTR_ERR(oe);
1418 	if (IS_ERR(oe))
1419 		goto out_err;
1420 
1421 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
1422 	if (!ofs->upper_mnt)
1423 		sb->s_flags |= SB_RDONLY;
1424 
1425 	if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1426 		err = ovl_get_indexdir(ofs, oe, &upperpath);
1427 		if (err)
1428 			goto out_free_oe;
1429 
1430 		/* Force r/o mount with no index dir */
1431 		if (!ofs->indexdir) {
1432 			dput(ofs->workdir);
1433 			ofs->workdir = NULL;
1434 			sb->s_flags |= SB_RDONLY;
1435 		}
1436 
1437 	}
1438 
1439 	/* Show index=off in /proc/mounts for forced r/o mount */
1440 	if (!ofs->indexdir) {
1441 		ofs->config.index = false;
1442 		if (ofs->upper_mnt && ofs->config.nfs_export) {
1443 			pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1444 			ofs->config.nfs_export = false;
1445 		}
1446 	}
1447 
1448 	if (ofs->config.nfs_export)
1449 		sb->s_export_op = &ovl_export_operations;
1450 
1451 	/* Never override disk quota limits or use reserved space */
1452 	cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1453 
1454 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1455 	sb->s_op = &ovl_super_operations;
1456 	sb->s_xattr = ovl_xattr_handlers;
1457 	sb->s_fs_info = ofs;
1458 	sb->s_flags |= SB_POSIXACL | SB_NOREMOTELOCK;
1459 
1460 	err = -ENOMEM;
1461 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1462 	if (!root_dentry)
1463 		goto out_free_oe;
1464 
1465 	root_dentry->d_fsdata = oe;
1466 
1467 	mntput(upperpath.mnt);
1468 	if (upperpath.dentry) {
1469 		ovl_dentry_set_upper_alias(root_dentry);
1470 		if (ovl_is_impuredir(upperpath.dentry))
1471 			ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1472 	}
1473 
1474 	/* Root is always merge -> can have whiteouts */
1475 	ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1476 	ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1477 	ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1478 		       ovl_dentry_lower(root_dentry));
1479 
1480 	sb->s_root = root_dentry;
1481 
1482 	return 0;
1483 
1484 out_free_oe:
1485 	ovl_entry_stack_free(oe);
1486 	kfree(oe);
1487 out_err:
1488 	path_put(&upperpath);
1489 	ovl_free_fs(ofs);
1490 out:
1491 	return err;
1492 }
1493 
1494 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1495 				const char *dev_name, void *raw_data)
1496 {
1497 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1498 }
1499 
1500 static struct file_system_type ovl_fs_type = {
1501 	.owner		= THIS_MODULE,
1502 	.name		= "overlay",
1503 	.mount		= ovl_mount,
1504 	.kill_sb	= kill_anon_super,
1505 };
1506 MODULE_ALIAS_FS("overlay");
1507 
1508 static void ovl_inode_init_once(void *foo)
1509 {
1510 	struct ovl_inode *oi = foo;
1511 
1512 	inode_init_once(&oi->vfs_inode);
1513 }
1514 
1515 static int __init ovl_init(void)
1516 {
1517 	int err;
1518 
1519 	ovl_inode_cachep = kmem_cache_create("ovl_inode",
1520 					     sizeof(struct ovl_inode), 0,
1521 					     (SLAB_RECLAIM_ACCOUNT|
1522 					      SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1523 					     ovl_inode_init_once);
1524 	if (ovl_inode_cachep == NULL)
1525 		return -ENOMEM;
1526 
1527 	err = register_filesystem(&ovl_fs_type);
1528 	if (err)
1529 		kmem_cache_destroy(ovl_inode_cachep);
1530 
1531 	return err;
1532 }
1533 
1534 static void __exit ovl_exit(void)
1535 {
1536 	unregister_filesystem(&ovl_fs_type);
1537 
1538 	/*
1539 	 * Make sure all delayed rcu free inodes are flushed before we
1540 	 * destroy cache.
1541 	 */
1542 	rcu_barrier();
1543 	kmem_cache_destroy(ovl_inode_cachep);
1544 
1545 }
1546 
1547 module_init(ovl_init);
1548 module_exit(ovl_exit);
1549