xref: /openbmc/linux/fs/overlayfs/super.c (revision da2ef666)
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 bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
68 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
69 MODULE_PARM_DESC(ovl_metacopy_def,
70 		 "Default to on or off for the metadata only copy up feature");
71 
72 static void ovl_dentry_release(struct dentry *dentry)
73 {
74 	struct ovl_entry *oe = dentry->d_fsdata;
75 
76 	if (oe) {
77 		ovl_entry_stack_free(oe);
78 		kfree_rcu(oe, rcu);
79 	}
80 }
81 
82 static struct dentry *ovl_d_real(struct dentry *dentry,
83 				 const struct inode *inode)
84 {
85 	struct dentry *real;
86 
87 	/* It's an overlay file */
88 	if (inode && d_inode(dentry) == inode)
89 		return dentry;
90 
91 	if (!d_is_reg(dentry)) {
92 		if (!inode || inode == d_inode(dentry))
93 			return dentry;
94 		goto bug;
95 	}
96 
97 	real = ovl_dentry_upper(dentry);
98 	if (real && (inode == d_inode(real)))
99 		return real;
100 
101 	if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
102 		return real;
103 
104 	real = ovl_dentry_lowerdata(dentry);
105 	if (!real)
106 		goto bug;
107 
108 	/* Handle recursion */
109 	real = d_real(real, inode);
110 
111 	if (!inode || inode == d_inode(real))
112 		return real;
113 bug:
114 	WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
115 	     inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
116 	return dentry;
117 }
118 
119 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
120 {
121 	struct ovl_entry *oe = dentry->d_fsdata;
122 	unsigned int i;
123 	int ret = 1;
124 
125 	for (i = 0; i < oe->numlower; i++) {
126 		struct dentry *d = oe->lowerstack[i].dentry;
127 
128 		if (d->d_flags & DCACHE_OP_REVALIDATE) {
129 			ret = d->d_op->d_revalidate(d, flags);
130 			if (ret < 0)
131 				return ret;
132 			if (!ret) {
133 				if (!(flags & LOOKUP_RCU))
134 					d_invalidate(d);
135 				return -ESTALE;
136 			}
137 		}
138 	}
139 	return 1;
140 }
141 
142 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
143 {
144 	struct ovl_entry *oe = dentry->d_fsdata;
145 	unsigned int i;
146 	int ret = 1;
147 
148 	for (i = 0; i < oe->numlower; i++) {
149 		struct dentry *d = oe->lowerstack[i].dentry;
150 
151 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
152 			ret = d->d_op->d_weak_revalidate(d, flags);
153 			if (ret <= 0)
154 				break;
155 		}
156 	}
157 	return ret;
158 }
159 
160 static const struct dentry_operations ovl_dentry_operations = {
161 	.d_release = ovl_dentry_release,
162 	.d_real = ovl_d_real,
163 };
164 
165 static const struct dentry_operations ovl_reval_dentry_operations = {
166 	.d_release = ovl_dentry_release,
167 	.d_real = ovl_d_real,
168 	.d_revalidate = ovl_dentry_revalidate,
169 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
170 };
171 
172 static struct kmem_cache *ovl_inode_cachep;
173 
174 static struct inode *ovl_alloc_inode(struct super_block *sb)
175 {
176 	struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
177 
178 	if (!oi)
179 		return NULL;
180 
181 	oi->cache = NULL;
182 	oi->redirect = NULL;
183 	oi->version = 0;
184 	oi->flags = 0;
185 	oi->__upperdentry = NULL;
186 	oi->lower = NULL;
187 	oi->lowerdata = NULL;
188 	mutex_init(&oi->lock);
189 
190 	return &oi->vfs_inode;
191 }
192 
193 static void ovl_i_callback(struct rcu_head *head)
194 {
195 	struct inode *inode = container_of(head, struct inode, i_rcu);
196 
197 	kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
198 }
199 
200 static void ovl_destroy_inode(struct inode *inode)
201 {
202 	struct ovl_inode *oi = OVL_I(inode);
203 
204 	dput(oi->__upperdentry);
205 	iput(oi->lower);
206 	if (S_ISDIR(inode->i_mode))
207 		ovl_dir_cache_free(inode);
208 	else
209 		iput(oi->lowerdata);
210 	kfree(oi->redirect);
211 	mutex_destroy(&oi->lock);
212 
213 	call_rcu(&inode->i_rcu, ovl_i_callback);
214 }
215 
216 static void ovl_free_fs(struct ovl_fs *ofs)
217 {
218 	unsigned i;
219 
220 	dput(ofs->indexdir);
221 	dput(ofs->workdir);
222 	if (ofs->workdir_locked)
223 		ovl_inuse_unlock(ofs->workbasedir);
224 	dput(ofs->workbasedir);
225 	if (ofs->upperdir_locked)
226 		ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
227 	mntput(ofs->upper_mnt);
228 	for (i = 0; i < ofs->numlower; i++)
229 		mntput(ofs->lower_layers[i].mnt);
230 	for (i = 0; i < ofs->numlowerfs; i++)
231 		free_anon_bdev(ofs->lower_fs[i].pseudo_dev);
232 	kfree(ofs->lower_layers);
233 	kfree(ofs->lower_fs);
234 
235 	kfree(ofs->config.lowerdir);
236 	kfree(ofs->config.upperdir);
237 	kfree(ofs->config.workdir);
238 	kfree(ofs->config.redirect_mode);
239 	if (ofs->creator_cred)
240 		put_cred(ofs->creator_cred);
241 	kfree(ofs);
242 }
243 
244 static void ovl_put_super(struct super_block *sb)
245 {
246 	struct ovl_fs *ofs = sb->s_fs_info;
247 
248 	ovl_free_fs(ofs);
249 }
250 
251 /* Sync real dirty inodes in upper filesystem (if it exists) */
252 static int ovl_sync_fs(struct super_block *sb, int wait)
253 {
254 	struct ovl_fs *ofs = sb->s_fs_info;
255 	struct super_block *upper_sb;
256 	int ret;
257 
258 	if (!ofs->upper_mnt)
259 		return 0;
260 
261 	/*
262 	 * If this is a sync(2) call or an emergency sync, all the super blocks
263 	 * will be iterated, including upper_sb, so no need to do anything.
264 	 *
265 	 * If this is a syncfs(2) call, then we do need to call
266 	 * sync_filesystem() on upper_sb, but enough if we do it when being
267 	 * called with wait == 1.
268 	 */
269 	if (!wait)
270 		return 0;
271 
272 	upper_sb = ofs->upper_mnt->mnt_sb;
273 
274 	down_read(&upper_sb->s_umount);
275 	ret = sync_filesystem(upper_sb);
276 	up_read(&upper_sb->s_umount);
277 
278 	return ret;
279 }
280 
281 /**
282  * ovl_statfs
283  * @sb: The overlayfs super block
284  * @buf: The struct kstatfs to fill in with stats
285  *
286  * Get the filesystem statistics.  As writes always target the upper layer
287  * filesystem pass the statfs to the upper filesystem (if it exists)
288  */
289 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
290 {
291 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
292 	struct dentry *root_dentry = dentry->d_sb->s_root;
293 	struct path path;
294 	int err;
295 
296 	ovl_path_real(root_dentry, &path);
297 
298 	err = vfs_statfs(&path, buf);
299 	if (!err) {
300 		buf->f_namelen = ofs->namelen;
301 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
302 	}
303 
304 	return err;
305 }
306 
307 /* Will this overlay be forced to mount/remount ro? */
308 static bool ovl_force_readonly(struct ovl_fs *ofs)
309 {
310 	return (!ofs->upper_mnt || !ofs->workdir);
311 }
312 
313 static const char *ovl_redirect_mode_def(void)
314 {
315 	return ovl_redirect_dir_def ? "on" : "off";
316 }
317 
318 enum {
319 	OVL_XINO_OFF,
320 	OVL_XINO_AUTO,
321 	OVL_XINO_ON,
322 };
323 
324 static const char * const ovl_xino_str[] = {
325 	"off",
326 	"auto",
327 	"on",
328 };
329 
330 static inline int ovl_xino_def(void)
331 {
332 	return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
333 }
334 
335 /**
336  * ovl_show_options
337  *
338  * Prints the mount options for a given superblock.
339  * Returns zero; does not fail.
340  */
341 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
342 {
343 	struct super_block *sb = dentry->d_sb;
344 	struct ovl_fs *ofs = sb->s_fs_info;
345 
346 	seq_show_option(m, "lowerdir", ofs->config.lowerdir);
347 	if (ofs->config.upperdir) {
348 		seq_show_option(m, "upperdir", ofs->config.upperdir);
349 		seq_show_option(m, "workdir", ofs->config.workdir);
350 	}
351 	if (ofs->config.default_permissions)
352 		seq_puts(m, ",default_permissions");
353 	if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
354 		seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
355 	if (ofs->config.index != ovl_index_def)
356 		seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
357 	if (ofs->config.nfs_export != ovl_nfs_export_def)
358 		seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
359 						"on" : "off");
360 	if (ofs->config.xino != ovl_xino_def())
361 		seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
362 	if (ofs->config.metacopy != ovl_metacopy_def)
363 		seq_printf(m, ",metacopy=%s",
364 			   ofs->config.metacopy ? "on" : "off");
365 	return 0;
366 }
367 
368 static int ovl_remount(struct super_block *sb, int *flags, char *data)
369 {
370 	struct ovl_fs *ofs = sb->s_fs_info;
371 
372 	if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
373 		return -EROFS;
374 
375 	return 0;
376 }
377 
378 static const struct super_operations ovl_super_operations = {
379 	.alloc_inode	= ovl_alloc_inode,
380 	.destroy_inode	= ovl_destroy_inode,
381 	.drop_inode	= generic_delete_inode,
382 	.put_super	= ovl_put_super,
383 	.sync_fs	= ovl_sync_fs,
384 	.statfs		= ovl_statfs,
385 	.show_options	= ovl_show_options,
386 	.remount_fs	= ovl_remount,
387 };
388 
389 enum {
390 	OPT_LOWERDIR,
391 	OPT_UPPERDIR,
392 	OPT_WORKDIR,
393 	OPT_DEFAULT_PERMISSIONS,
394 	OPT_REDIRECT_DIR,
395 	OPT_INDEX_ON,
396 	OPT_INDEX_OFF,
397 	OPT_NFS_EXPORT_ON,
398 	OPT_NFS_EXPORT_OFF,
399 	OPT_XINO_ON,
400 	OPT_XINO_OFF,
401 	OPT_XINO_AUTO,
402 	OPT_METACOPY_ON,
403 	OPT_METACOPY_OFF,
404 	OPT_ERR,
405 };
406 
407 static const match_table_t ovl_tokens = {
408 	{OPT_LOWERDIR,			"lowerdir=%s"},
409 	{OPT_UPPERDIR,			"upperdir=%s"},
410 	{OPT_WORKDIR,			"workdir=%s"},
411 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
412 	{OPT_REDIRECT_DIR,		"redirect_dir=%s"},
413 	{OPT_INDEX_ON,			"index=on"},
414 	{OPT_INDEX_OFF,			"index=off"},
415 	{OPT_NFS_EXPORT_ON,		"nfs_export=on"},
416 	{OPT_NFS_EXPORT_OFF,		"nfs_export=off"},
417 	{OPT_XINO_ON,			"xino=on"},
418 	{OPT_XINO_OFF,			"xino=off"},
419 	{OPT_XINO_AUTO,			"xino=auto"},
420 	{OPT_METACOPY_ON,		"metacopy=on"},
421 	{OPT_METACOPY_OFF,		"metacopy=off"},
422 	{OPT_ERR,			NULL}
423 };
424 
425 static char *ovl_next_opt(char **s)
426 {
427 	char *sbegin = *s;
428 	char *p;
429 
430 	if (sbegin == NULL)
431 		return NULL;
432 
433 	for (p = sbegin; *p; p++) {
434 		if (*p == '\\') {
435 			p++;
436 			if (!*p)
437 				break;
438 		} else if (*p == ',') {
439 			*p = '\0';
440 			*s = p + 1;
441 			return sbegin;
442 		}
443 	}
444 	*s = NULL;
445 	return sbegin;
446 }
447 
448 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
449 {
450 	if (strcmp(mode, "on") == 0) {
451 		config->redirect_dir = true;
452 		/*
453 		 * Does not make sense to have redirect creation without
454 		 * redirect following.
455 		 */
456 		config->redirect_follow = true;
457 	} else if (strcmp(mode, "follow") == 0) {
458 		config->redirect_follow = true;
459 	} else if (strcmp(mode, "off") == 0) {
460 		if (ovl_redirect_always_follow)
461 			config->redirect_follow = true;
462 	} else if (strcmp(mode, "nofollow") != 0) {
463 		pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
464 		       mode);
465 		return -EINVAL;
466 	}
467 
468 	return 0;
469 }
470 
471 static int ovl_parse_opt(char *opt, struct ovl_config *config)
472 {
473 	char *p;
474 	int err;
475 
476 	config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
477 	if (!config->redirect_mode)
478 		return -ENOMEM;
479 
480 	while ((p = ovl_next_opt(&opt)) != NULL) {
481 		int token;
482 		substring_t args[MAX_OPT_ARGS];
483 
484 		if (!*p)
485 			continue;
486 
487 		token = match_token(p, ovl_tokens, args);
488 		switch (token) {
489 		case OPT_UPPERDIR:
490 			kfree(config->upperdir);
491 			config->upperdir = match_strdup(&args[0]);
492 			if (!config->upperdir)
493 				return -ENOMEM;
494 			break;
495 
496 		case OPT_LOWERDIR:
497 			kfree(config->lowerdir);
498 			config->lowerdir = match_strdup(&args[0]);
499 			if (!config->lowerdir)
500 				return -ENOMEM;
501 			break;
502 
503 		case OPT_WORKDIR:
504 			kfree(config->workdir);
505 			config->workdir = match_strdup(&args[0]);
506 			if (!config->workdir)
507 				return -ENOMEM;
508 			break;
509 
510 		case OPT_DEFAULT_PERMISSIONS:
511 			config->default_permissions = true;
512 			break;
513 
514 		case OPT_REDIRECT_DIR:
515 			kfree(config->redirect_mode);
516 			config->redirect_mode = match_strdup(&args[0]);
517 			if (!config->redirect_mode)
518 				return -ENOMEM;
519 			break;
520 
521 		case OPT_INDEX_ON:
522 			config->index = true;
523 			break;
524 
525 		case OPT_INDEX_OFF:
526 			config->index = false;
527 			break;
528 
529 		case OPT_NFS_EXPORT_ON:
530 			config->nfs_export = true;
531 			break;
532 
533 		case OPT_NFS_EXPORT_OFF:
534 			config->nfs_export = false;
535 			break;
536 
537 		case OPT_XINO_ON:
538 			config->xino = OVL_XINO_ON;
539 			break;
540 
541 		case OPT_XINO_OFF:
542 			config->xino = OVL_XINO_OFF;
543 			break;
544 
545 		case OPT_XINO_AUTO:
546 			config->xino = OVL_XINO_AUTO;
547 			break;
548 
549 		case OPT_METACOPY_ON:
550 			config->metacopy = true;
551 			break;
552 
553 		case OPT_METACOPY_OFF:
554 			config->metacopy = false;
555 			break;
556 
557 		default:
558 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
559 			return -EINVAL;
560 		}
561 	}
562 
563 	/* Workdir is useless in non-upper mount */
564 	if (!config->upperdir && config->workdir) {
565 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
566 			config->workdir);
567 		kfree(config->workdir);
568 		config->workdir = NULL;
569 	}
570 
571 	err = ovl_parse_redirect_mode(config, config->redirect_mode);
572 	if (err)
573 		return err;
574 
575 	/* metacopy feature with upper requires redirect_dir=on */
576 	if (config->upperdir && config->metacopy && !config->redirect_dir) {
577 		pr_warn("overlayfs: metadata only copy up requires \"redirect_dir=on\", falling back to metacopy=off.\n");
578 		config->metacopy = false;
579 	} else if (config->metacopy && !config->redirect_follow) {
580 		pr_warn("overlayfs: metadata only copy up requires \"redirect_dir=follow\" on non-upper mount, falling back to metacopy=off.\n");
581 		config->metacopy = false;
582 	}
583 
584 	return 0;
585 }
586 
587 #define OVL_WORKDIR_NAME "work"
588 #define OVL_INDEXDIR_NAME "index"
589 
590 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
591 					 const char *name, bool persist)
592 {
593 	struct inode *dir =  ofs->workbasedir->d_inode;
594 	struct vfsmount *mnt = ofs->upper_mnt;
595 	struct dentry *work;
596 	int err;
597 	bool retried = false;
598 	bool locked = false;
599 
600 	inode_lock_nested(dir, I_MUTEX_PARENT);
601 	locked = true;
602 
603 retry:
604 	work = lookup_one_len(name, ofs->workbasedir, strlen(name));
605 
606 	if (!IS_ERR(work)) {
607 		struct iattr attr = {
608 			.ia_valid = ATTR_MODE,
609 			.ia_mode = S_IFDIR | 0,
610 		};
611 
612 		if (work->d_inode) {
613 			err = -EEXIST;
614 			if (retried)
615 				goto out_dput;
616 
617 			if (persist)
618 				goto out_unlock;
619 
620 			retried = true;
621 			ovl_workdir_cleanup(dir, mnt, work, 0);
622 			dput(work);
623 			goto retry;
624 		}
625 
626 		work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
627 		err = PTR_ERR(work);
628 		if (IS_ERR(work))
629 			goto out_err;
630 
631 		/*
632 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
633 		 *
634 		 * a) success (there was a POSIX ACL xattr and was removed)
635 		 * b) -ENODATA (there was no POSIX ACL xattr)
636 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
637 		 *
638 		 * There are various other error values that could effectively
639 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
640 		 * if the xattr name is too long), but the set of filesystems
641 		 * allowed as upper are limited to "normal" ones, where checking
642 		 * for the above two errors is sufficient.
643 		 */
644 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
645 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
646 			goto out_dput;
647 
648 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
649 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
650 			goto out_dput;
651 
652 		/* Clear any inherited mode bits */
653 		inode_lock(work->d_inode);
654 		err = notify_change(work, &attr, NULL);
655 		inode_unlock(work->d_inode);
656 		if (err)
657 			goto out_dput;
658 	} else {
659 		err = PTR_ERR(work);
660 		goto out_err;
661 	}
662 out_unlock:
663 	if (locked)
664 		inode_unlock(dir);
665 
666 	return work;
667 
668 out_dput:
669 	dput(work);
670 out_err:
671 	pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
672 		ofs->config.workdir, name, -err);
673 	work = NULL;
674 	goto out_unlock;
675 }
676 
677 static void ovl_unescape(char *s)
678 {
679 	char *d = s;
680 
681 	for (;; s++, d++) {
682 		if (*s == '\\')
683 			s++;
684 		*d = *s;
685 		if (!*s)
686 			break;
687 	}
688 }
689 
690 static int ovl_mount_dir_noesc(const char *name, struct path *path)
691 {
692 	int err = -EINVAL;
693 
694 	if (!*name) {
695 		pr_err("overlayfs: empty lowerdir\n");
696 		goto out;
697 	}
698 	err = kern_path(name, LOOKUP_FOLLOW, path);
699 	if (err) {
700 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
701 		goto out;
702 	}
703 	err = -EINVAL;
704 	if (ovl_dentry_weird(path->dentry)) {
705 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
706 		goto out_put;
707 	}
708 	if (!d_is_dir(path->dentry)) {
709 		pr_err("overlayfs: '%s' not a directory\n", name);
710 		goto out_put;
711 	}
712 	return 0;
713 
714 out_put:
715 	path_put_init(path);
716 out:
717 	return err;
718 }
719 
720 static int ovl_mount_dir(const char *name, struct path *path)
721 {
722 	int err = -ENOMEM;
723 	char *tmp = kstrdup(name, GFP_KERNEL);
724 
725 	if (tmp) {
726 		ovl_unescape(tmp);
727 		err = ovl_mount_dir_noesc(tmp, path);
728 
729 		if (!err)
730 			if (ovl_dentry_remote(path->dentry)) {
731 				pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
732 				       tmp);
733 				path_put_init(path);
734 				err = -EINVAL;
735 			}
736 		kfree(tmp);
737 	}
738 	return err;
739 }
740 
741 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
742 			     const char *name)
743 {
744 	struct kstatfs statfs;
745 	int err = vfs_statfs(path, &statfs);
746 
747 	if (err)
748 		pr_err("overlayfs: statfs failed on '%s'\n", name);
749 	else
750 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
751 
752 	return err;
753 }
754 
755 static int ovl_lower_dir(const char *name, struct path *path,
756 			 struct ovl_fs *ofs, int *stack_depth, bool *remote)
757 {
758 	int fh_type;
759 	int err;
760 
761 	err = ovl_mount_dir_noesc(name, path);
762 	if (err)
763 		goto out;
764 
765 	err = ovl_check_namelen(path, ofs, name);
766 	if (err)
767 		goto out_put;
768 
769 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
770 
771 	if (ovl_dentry_remote(path->dentry))
772 		*remote = true;
773 
774 	/*
775 	 * The inodes index feature and NFS export need to encode and decode
776 	 * file handles, so they require that all layers support them.
777 	 */
778 	fh_type = ovl_can_decode_fh(path->dentry->d_sb);
779 	if ((ofs->config.nfs_export ||
780 	     (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
781 		ofs->config.index = false;
782 		ofs->config.nfs_export = false;
783 		pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
784 			name);
785 	}
786 
787 	/* Check if lower fs has 32bit inode numbers */
788 	if (fh_type != FILEID_INO32_GEN)
789 		ofs->xino_bits = 0;
790 
791 	return 0;
792 
793 out_put:
794 	path_put_init(path);
795 out:
796 	return err;
797 }
798 
799 /* Workdir should not be subdir of upperdir and vice versa */
800 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
801 {
802 	bool ok = false;
803 
804 	if (workdir != upperdir) {
805 		ok = (lock_rename(workdir, upperdir) == NULL);
806 		unlock_rename(workdir, upperdir);
807 	}
808 	return ok;
809 }
810 
811 static unsigned int ovl_split_lowerdirs(char *str)
812 {
813 	unsigned int ctr = 1;
814 	char *s, *d;
815 
816 	for (s = d = str;; s++, d++) {
817 		if (*s == '\\') {
818 			s++;
819 		} else if (*s == ':') {
820 			*d = '\0';
821 			ctr++;
822 			continue;
823 		}
824 		*d = *s;
825 		if (!*s)
826 			break;
827 	}
828 	return ctr;
829 }
830 
831 static int __maybe_unused
832 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
833 			struct dentry *dentry, struct inode *inode,
834 			const char *name, void *buffer, size_t size)
835 {
836 	return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
837 }
838 
839 static int __maybe_unused
840 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
841 			struct dentry *dentry, struct inode *inode,
842 			const char *name, const void *value,
843 			size_t size, int flags)
844 {
845 	struct dentry *workdir = ovl_workdir(dentry);
846 	struct inode *realinode = ovl_inode_real(inode);
847 	struct posix_acl *acl = NULL;
848 	int err;
849 
850 	/* Check that everything is OK before copy-up */
851 	if (value) {
852 		acl = posix_acl_from_xattr(&init_user_ns, value, size);
853 		if (IS_ERR(acl))
854 			return PTR_ERR(acl);
855 	}
856 	err = -EOPNOTSUPP;
857 	if (!IS_POSIXACL(d_inode(workdir)))
858 		goto out_acl_release;
859 	if (!realinode->i_op->set_acl)
860 		goto out_acl_release;
861 	if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
862 		err = acl ? -EACCES : 0;
863 		goto out_acl_release;
864 	}
865 	err = -EPERM;
866 	if (!inode_owner_or_capable(inode))
867 		goto out_acl_release;
868 
869 	posix_acl_release(acl);
870 
871 	/*
872 	 * Check if sgid bit needs to be cleared (actual setacl operation will
873 	 * be done with mounter's capabilities and so that won't do it for us).
874 	 */
875 	if (unlikely(inode->i_mode & S_ISGID) &&
876 	    handler->flags == ACL_TYPE_ACCESS &&
877 	    !in_group_p(inode->i_gid) &&
878 	    !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
879 		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
880 
881 		err = ovl_setattr(dentry, &iattr);
882 		if (err)
883 			return err;
884 	}
885 
886 	err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
887 	if (!err)
888 		ovl_copyattr(ovl_inode_real(inode), inode);
889 
890 	return err;
891 
892 out_acl_release:
893 	posix_acl_release(acl);
894 	return err;
895 }
896 
897 static int ovl_own_xattr_get(const struct xattr_handler *handler,
898 			     struct dentry *dentry, struct inode *inode,
899 			     const char *name, void *buffer, size_t size)
900 {
901 	return -EOPNOTSUPP;
902 }
903 
904 static int ovl_own_xattr_set(const struct xattr_handler *handler,
905 			     struct dentry *dentry, struct inode *inode,
906 			     const char *name, const void *value,
907 			     size_t size, int flags)
908 {
909 	return -EOPNOTSUPP;
910 }
911 
912 static int ovl_other_xattr_get(const struct xattr_handler *handler,
913 			       struct dentry *dentry, struct inode *inode,
914 			       const char *name, void *buffer, size_t size)
915 {
916 	return ovl_xattr_get(dentry, inode, name, buffer, size);
917 }
918 
919 static int ovl_other_xattr_set(const struct xattr_handler *handler,
920 			       struct dentry *dentry, struct inode *inode,
921 			       const char *name, const void *value,
922 			       size_t size, int flags)
923 {
924 	return ovl_xattr_set(dentry, inode, name, value, size, flags);
925 }
926 
927 static const struct xattr_handler __maybe_unused
928 ovl_posix_acl_access_xattr_handler = {
929 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
930 	.flags = ACL_TYPE_ACCESS,
931 	.get = ovl_posix_acl_xattr_get,
932 	.set = ovl_posix_acl_xattr_set,
933 };
934 
935 static const struct xattr_handler __maybe_unused
936 ovl_posix_acl_default_xattr_handler = {
937 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
938 	.flags = ACL_TYPE_DEFAULT,
939 	.get = ovl_posix_acl_xattr_get,
940 	.set = ovl_posix_acl_xattr_set,
941 };
942 
943 static const struct xattr_handler ovl_own_xattr_handler = {
944 	.prefix	= OVL_XATTR_PREFIX,
945 	.get = ovl_own_xattr_get,
946 	.set = ovl_own_xattr_set,
947 };
948 
949 static const struct xattr_handler ovl_other_xattr_handler = {
950 	.prefix	= "", /* catch all */
951 	.get = ovl_other_xattr_get,
952 	.set = ovl_other_xattr_set,
953 };
954 
955 static const struct xattr_handler *ovl_xattr_handlers[] = {
956 #ifdef CONFIG_FS_POSIX_ACL
957 	&ovl_posix_acl_access_xattr_handler,
958 	&ovl_posix_acl_default_xattr_handler,
959 #endif
960 	&ovl_own_xattr_handler,
961 	&ovl_other_xattr_handler,
962 	NULL
963 };
964 
965 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
966 {
967 	struct vfsmount *upper_mnt;
968 	int err;
969 
970 	err = ovl_mount_dir(ofs->config.upperdir, upperpath);
971 	if (err)
972 		goto out;
973 
974 	/* Upper fs should not be r/o */
975 	if (sb_rdonly(upperpath->mnt->mnt_sb)) {
976 		pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
977 		err = -EINVAL;
978 		goto out;
979 	}
980 
981 	err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
982 	if (err)
983 		goto out;
984 
985 	err = -EBUSY;
986 	if (ovl_inuse_trylock(upperpath->dentry)) {
987 		ofs->upperdir_locked = true;
988 	} else if (ofs->config.index) {
989 		pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
990 		goto out;
991 	} else {
992 		pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
993 	}
994 
995 	upper_mnt = clone_private_mount(upperpath);
996 	err = PTR_ERR(upper_mnt);
997 	if (IS_ERR(upper_mnt)) {
998 		pr_err("overlayfs: failed to clone upperpath\n");
999 		goto out;
1000 	}
1001 
1002 	/* Don't inherit atime flags */
1003 	upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1004 	ofs->upper_mnt = upper_mnt;
1005 	err = 0;
1006 out:
1007 	return err;
1008 }
1009 
1010 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
1011 {
1012 	struct vfsmount *mnt = ofs->upper_mnt;
1013 	struct dentry *temp;
1014 	int fh_type;
1015 	int err;
1016 
1017 	err = mnt_want_write(mnt);
1018 	if (err)
1019 		return err;
1020 
1021 	ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1022 	if (!ofs->workdir)
1023 		goto out;
1024 
1025 	/*
1026 	 * Upper should support d_type, else whiteouts are visible.  Given
1027 	 * workdir and upper are on same fs, we can do iterate_dir() on
1028 	 * workdir. This check requires successful creation of workdir in
1029 	 * previous step.
1030 	 */
1031 	err = ovl_check_d_type_supported(workpath);
1032 	if (err < 0)
1033 		goto out;
1034 
1035 	/*
1036 	 * We allowed this configuration and don't want to break users over
1037 	 * kernel upgrade. So warn instead of erroring out.
1038 	 */
1039 	if (!err)
1040 		pr_warn("overlayfs: upper fs needs to support d_type.\n");
1041 
1042 	/* Check if upper/work fs supports O_TMPFILE */
1043 	temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1044 	ofs->tmpfile = !IS_ERR(temp);
1045 	if (ofs->tmpfile)
1046 		dput(temp);
1047 	else
1048 		pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1049 
1050 	/*
1051 	 * Check if upper/work fs supports trusted.overlay.* xattr
1052 	 */
1053 	err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1054 	if (err) {
1055 		ofs->noxattr = true;
1056 		ofs->config.index = false;
1057 		ofs->config.metacopy = false;
1058 		pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1059 		err = 0;
1060 	} else {
1061 		vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1062 	}
1063 
1064 	/* Check if upper/work fs supports file handles */
1065 	fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1066 	if (ofs->config.index && !fh_type) {
1067 		ofs->config.index = false;
1068 		pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1069 	}
1070 
1071 	/* Check if upper fs has 32bit inode numbers */
1072 	if (fh_type != FILEID_INO32_GEN)
1073 		ofs->xino_bits = 0;
1074 
1075 	/* NFS export of r/w mount depends on index */
1076 	if (ofs->config.nfs_export && !ofs->config.index) {
1077 		pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1078 		ofs->config.nfs_export = false;
1079 	}
1080 out:
1081 	mnt_drop_write(mnt);
1082 	return err;
1083 }
1084 
1085 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
1086 {
1087 	int err;
1088 	struct path workpath = { };
1089 
1090 	err = ovl_mount_dir(ofs->config.workdir, &workpath);
1091 	if (err)
1092 		goto out;
1093 
1094 	err = -EINVAL;
1095 	if (upperpath->mnt != workpath.mnt) {
1096 		pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1097 		goto out;
1098 	}
1099 	if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1100 		pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1101 		goto out;
1102 	}
1103 
1104 	err = -EBUSY;
1105 	if (ovl_inuse_trylock(workpath.dentry)) {
1106 		ofs->workdir_locked = true;
1107 	} else if (ofs->config.index) {
1108 		pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1109 		goto out;
1110 	} else {
1111 		pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1112 	}
1113 
1114 	ofs->workbasedir = dget(workpath.dentry);
1115 	err = ovl_make_workdir(ofs, &workpath);
1116 	if (err)
1117 		goto out;
1118 
1119 	err = 0;
1120 out:
1121 	path_put(&workpath);
1122 
1123 	return err;
1124 }
1125 
1126 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
1127 			    struct path *upperpath)
1128 {
1129 	struct vfsmount *mnt = ofs->upper_mnt;
1130 	int err;
1131 
1132 	err = mnt_want_write(mnt);
1133 	if (err)
1134 		return err;
1135 
1136 	/* Verify lower root is upper root origin */
1137 	err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1138 				true);
1139 	if (err) {
1140 		pr_err("overlayfs: failed to verify upper root origin\n");
1141 		goto out;
1142 	}
1143 
1144 	ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1145 	if (ofs->indexdir) {
1146 		/*
1147 		 * Verify upper root is exclusively associated with index dir.
1148 		 * Older kernels stored upper fh in "trusted.overlay.origin"
1149 		 * xattr. If that xattr exists, verify that it is a match to
1150 		 * upper dir file handle. In any case, verify or set xattr
1151 		 * "trusted.overlay.upper" to indicate that index may have
1152 		 * directory entries.
1153 		 */
1154 		if (ovl_check_origin_xattr(ofs->indexdir)) {
1155 			err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1156 						upperpath->dentry, true, false);
1157 			if (err)
1158 				pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1159 		}
1160 		err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1161 		if (err)
1162 			pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1163 
1164 		/* Cleanup bad/stale/orphan index entries */
1165 		if (!err)
1166 			err = ovl_indexdir_cleanup(ofs);
1167 	}
1168 	if (err || !ofs->indexdir)
1169 		pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1170 
1171 out:
1172 	mnt_drop_write(mnt);
1173 	return err;
1174 }
1175 
1176 /* Get a unique fsid for the layer */
1177 static int ovl_get_fsid(struct ovl_fs *ofs, struct super_block *sb)
1178 {
1179 	unsigned int i;
1180 	dev_t dev;
1181 	int err;
1182 
1183 	/* fsid 0 is reserved for upper fs even with non upper overlay */
1184 	if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1185 		return 0;
1186 
1187 	for (i = 0; i < ofs->numlowerfs; i++) {
1188 		if (ofs->lower_fs[i].sb == sb)
1189 			return i + 1;
1190 	}
1191 
1192 	err = get_anon_bdev(&dev);
1193 	if (err) {
1194 		pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1195 		return err;
1196 	}
1197 
1198 	ofs->lower_fs[ofs->numlowerfs].sb = sb;
1199 	ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1200 	ofs->numlowerfs++;
1201 
1202 	return ofs->numlowerfs;
1203 }
1204 
1205 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1206 				unsigned int numlower)
1207 {
1208 	int err;
1209 	unsigned int i;
1210 
1211 	err = -ENOMEM;
1212 	ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1213 				    GFP_KERNEL);
1214 	if (ofs->lower_layers == NULL)
1215 		goto out;
1216 
1217 	ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1218 				GFP_KERNEL);
1219 	if (ofs->lower_fs == NULL)
1220 		goto out;
1221 
1222 	for (i = 0; i < numlower; i++) {
1223 		struct vfsmount *mnt;
1224 		int fsid;
1225 
1226 		err = fsid = ovl_get_fsid(ofs, stack[i].mnt->mnt_sb);
1227 		if (err < 0)
1228 			goto out;
1229 
1230 		mnt = clone_private_mount(&stack[i]);
1231 		err = PTR_ERR(mnt);
1232 		if (IS_ERR(mnt)) {
1233 			pr_err("overlayfs: failed to clone lowerpath\n");
1234 			goto out;
1235 		}
1236 
1237 		/*
1238 		 * Make lower layers R/O.  That way fchmod/fchown on lower file
1239 		 * will fail instead of modifying lower fs.
1240 		 */
1241 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1242 
1243 		ofs->lower_layers[ofs->numlower].mnt = mnt;
1244 		ofs->lower_layers[ofs->numlower].idx = i + 1;
1245 		ofs->lower_layers[ofs->numlower].fsid = fsid;
1246 		if (fsid) {
1247 			ofs->lower_layers[ofs->numlower].fs =
1248 				&ofs->lower_fs[fsid - 1];
1249 		}
1250 		ofs->numlower++;
1251 	}
1252 
1253 	/*
1254 	 * When all layers on same fs, overlay can use real inode numbers.
1255 	 * With mount option "xino=on", mounter declares that there are enough
1256 	 * free high bits in underlying fs to hold the unique fsid.
1257 	 * If overlayfs does encounter underlying inodes using the high xino
1258 	 * bits reserved for fsid, it emits a warning and uses the original
1259 	 * inode number.
1260 	 */
1261 	if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1262 		ofs->xino_bits = 0;
1263 		ofs->config.xino = OVL_XINO_OFF;
1264 	} else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1265 		/*
1266 		 * This is a roundup of number of bits needed for numlowerfs+1
1267 		 * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1268 		 * upper fs even with non upper overlay.
1269 		 */
1270 		BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1271 		ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1272 	}
1273 
1274 	if (ofs->xino_bits) {
1275 		pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1276 			ofs->xino_bits);
1277 	}
1278 
1279 	err = 0;
1280 out:
1281 	return err;
1282 }
1283 
1284 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1285 					    struct ovl_fs *ofs)
1286 {
1287 	int err;
1288 	char *lowertmp, *lower;
1289 	struct path *stack = NULL;
1290 	unsigned int stacklen, numlower = 0, i;
1291 	bool remote = false;
1292 	struct ovl_entry *oe;
1293 
1294 	err = -ENOMEM;
1295 	lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1296 	if (!lowertmp)
1297 		goto out_err;
1298 
1299 	err = -EINVAL;
1300 	stacklen = ovl_split_lowerdirs(lowertmp);
1301 	if (stacklen > OVL_MAX_STACK) {
1302 		pr_err("overlayfs: too many lower directories, limit is %d\n",
1303 		       OVL_MAX_STACK);
1304 		goto out_err;
1305 	} else if (!ofs->config.upperdir && stacklen == 1) {
1306 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1307 		goto out_err;
1308 	} else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1309 		   ofs->config.redirect_follow) {
1310 		pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1311 		ofs->config.nfs_export = false;
1312 	}
1313 
1314 	err = -ENOMEM;
1315 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1316 	if (!stack)
1317 		goto out_err;
1318 
1319 	err = -EINVAL;
1320 	lower = lowertmp;
1321 	for (numlower = 0; numlower < stacklen; numlower++) {
1322 		err = ovl_lower_dir(lower, &stack[numlower], ofs,
1323 				    &sb->s_stack_depth, &remote);
1324 		if (err)
1325 			goto out_err;
1326 
1327 		lower = strchr(lower, '\0') + 1;
1328 	}
1329 
1330 	err = -EINVAL;
1331 	sb->s_stack_depth++;
1332 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1333 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1334 		goto out_err;
1335 	}
1336 
1337 	err = ovl_get_lower_layers(ofs, stack, numlower);
1338 	if (err)
1339 		goto out_err;
1340 
1341 	err = -ENOMEM;
1342 	oe = ovl_alloc_entry(numlower);
1343 	if (!oe)
1344 		goto out_err;
1345 
1346 	for (i = 0; i < numlower; i++) {
1347 		oe->lowerstack[i].dentry = dget(stack[i].dentry);
1348 		oe->lowerstack[i].layer = &ofs->lower_layers[i];
1349 	}
1350 
1351 	if (remote)
1352 		sb->s_d_op = &ovl_reval_dentry_operations;
1353 	else
1354 		sb->s_d_op = &ovl_dentry_operations;
1355 
1356 out:
1357 	for (i = 0; i < numlower; i++)
1358 		path_put(&stack[i]);
1359 	kfree(stack);
1360 	kfree(lowertmp);
1361 
1362 	return oe;
1363 
1364 out_err:
1365 	oe = ERR_PTR(err);
1366 	goto out;
1367 }
1368 
1369 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1370 {
1371 	struct path upperpath = { };
1372 	struct dentry *root_dentry;
1373 	struct ovl_entry *oe;
1374 	struct ovl_fs *ofs;
1375 	struct cred *cred;
1376 	int err;
1377 
1378 	err = -ENOMEM;
1379 	ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1380 	if (!ofs)
1381 		goto out;
1382 
1383 	ofs->creator_cred = cred = prepare_creds();
1384 	if (!cred)
1385 		goto out_err;
1386 
1387 	ofs->config.index = ovl_index_def;
1388 	ofs->config.nfs_export = ovl_nfs_export_def;
1389 	ofs->config.xino = ovl_xino_def();
1390 	ofs->config.metacopy = ovl_metacopy_def;
1391 	err = ovl_parse_opt((char *) data, &ofs->config);
1392 	if (err)
1393 		goto out_err;
1394 
1395 	err = -EINVAL;
1396 	if (!ofs->config.lowerdir) {
1397 		if (!silent)
1398 			pr_err("overlayfs: missing 'lowerdir'\n");
1399 		goto out_err;
1400 	}
1401 
1402 	sb->s_stack_depth = 0;
1403 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1404 	/* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1405 	if (ofs->config.xino != OVL_XINO_OFF)
1406 		ofs->xino_bits = BITS_PER_LONG - 32;
1407 
1408 	if (ofs->config.upperdir) {
1409 		if (!ofs->config.workdir) {
1410 			pr_err("overlayfs: missing 'workdir'\n");
1411 			goto out_err;
1412 		}
1413 
1414 		err = ovl_get_upper(ofs, &upperpath);
1415 		if (err)
1416 			goto out_err;
1417 
1418 		err = ovl_get_workdir(ofs, &upperpath);
1419 		if (err)
1420 			goto out_err;
1421 
1422 		if (!ofs->workdir)
1423 			sb->s_flags |= SB_RDONLY;
1424 
1425 		sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1426 		sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1427 
1428 	}
1429 	oe = ovl_get_lowerstack(sb, ofs);
1430 	err = PTR_ERR(oe);
1431 	if (IS_ERR(oe))
1432 		goto out_err;
1433 
1434 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
1435 	if (!ofs->upper_mnt)
1436 		sb->s_flags |= SB_RDONLY;
1437 
1438 	if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1439 		err = ovl_get_indexdir(ofs, oe, &upperpath);
1440 		if (err)
1441 			goto out_free_oe;
1442 
1443 		/* Force r/o mount with no index dir */
1444 		if (!ofs->indexdir) {
1445 			dput(ofs->workdir);
1446 			ofs->workdir = NULL;
1447 			sb->s_flags |= SB_RDONLY;
1448 		}
1449 
1450 	}
1451 
1452 	/* Show index=off in /proc/mounts for forced r/o mount */
1453 	if (!ofs->indexdir) {
1454 		ofs->config.index = false;
1455 		if (ofs->upper_mnt && ofs->config.nfs_export) {
1456 			pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1457 			ofs->config.nfs_export = false;
1458 		}
1459 	}
1460 
1461 	if (ofs->config.metacopy && ofs->config.nfs_export) {
1462 		pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1463 		ofs->config.nfs_export = false;
1464 	}
1465 
1466 	if (ofs->config.nfs_export)
1467 		sb->s_export_op = &ovl_export_operations;
1468 
1469 	/* Never override disk quota limits or use reserved space */
1470 	cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1471 
1472 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1473 	sb->s_op = &ovl_super_operations;
1474 	sb->s_xattr = ovl_xattr_handlers;
1475 	sb->s_fs_info = ofs;
1476 	sb->s_flags |= SB_POSIXACL;
1477 
1478 	err = -ENOMEM;
1479 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1480 	if (!root_dentry)
1481 		goto out_free_oe;
1482 
1483 	root_dentry->d_fsdata = oe;
1484 
1485 	mntput(upperpath.mnt);
1486 	if (upperpath.dentry) {
1487 		ovl_dentry_set_upper_alias(root_dentry);
1488 		if (ovl_is_impuredir(upperpath.dentry))
1489 			ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1490 	}
1491 
1492 	/* Root is always merge -> can have whiteouts */
1493 	ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1494 	ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1495 	ovl_set_upperdata(d_inode(root_dentry));
1496 	ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1497 		       ovl_dentry_lower(root_dentry), NULL);
1498 
1499 	sb->s_root = root_dentry;
1500 
1501 	return 0;
1502 
1503 out_free_oe:
1504 	ovl_entry_stack_free(oe);
1505 	kfree(oe);
1506 out_err:
1507 	path_put(&upperpath);
1508 	ovl_free_fs(ofs);
1509 out:
1510 	return err;
1511 }
1512 
1513 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1514 				const char *dev_name, void *raw_data)
1515 {
1516 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1517 }
1518 
1519 static struct file_system_type ovl_fs_type = {
1520 	.owner		= THIS_MODULE,
1521 	.name		= "overlay",
1522 	.mount		= ovl_mount,
1523 	.kill_sb	= kill_anon_super,
1524 };
1525 MODULE_ALIAS_FS("overlay");
1526 
1527 static void ovl_inode_init_once(void *foo)
1528 {
1529 	struct ovl_inode *oi = foo;
1530 
1531 	inode_init_once(&oi->vfs_inode);
1532 }
1533 
1534 static int __init ovl_init(void)
1535 {
1536 	int err;
1537 
1538 	ovl_inode_cachep = kmem_cache_create("ovl_inode",
1539 					     sizeof(struct ovl_inode), 0,
1540 					     (SLAB_RECLAIM_ACCOUNT|
1541 					      SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1542 					     ovl_inode_init_once);
1543 	if (ovl_inode_cachep == NULL)
1544 		return -ENOMEM;
1545 
1546 	err = register_filesystem(&ovl_fs_type);
1547 	if (err)
1548 		kmem_cache_destroy(ovl_inode_cachep);
1549 
1550 	return err;
1551 }
1552 
1553 static void __exit ovl_exit(void)
1554 {
1555 	unregister_filesystem(&ovl_fs_type);
1556 
1557 	/*
1558 	 * Make sure all delayed rcu free inodes are flushed before we
1559 	 * destroy cache.
1560 	 */
1561 	rcu_barrier();
1562 	kmem_cache_destroy(ovl_inode_cachep);
1563 
1564 }
1565 
1566 module_init(ovl_init);
1567 module_exit(ovl_exit);
1568