xref: /linux/fs/overlayfs/super.c (revision f2527d8f566a45fa00ee5abd04d1c9476d4d704f)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright (C) 2011 Novell Inc.
5  */
6 
7 #include <uapi/linux/magic.h>
8 #include <linux/fs.h>
9 #include <linux/namei.h>
10 #include <linux/xattr.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/module.h>
14 #include <linux/statfs.h>
15 #include <linux/seq_file.h>
16 #include <linux/posix_acl_xattr.h>
17 #include <linux/exportfs.h>
18 #include <linux/file.h>
19 #include "overlayfs.h"
20 
21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
22 MODULE_DESCRIPTION("Overlay filesystem");
23 MODULE_LICENSE("GPL");
24 
25 
26 struct ovl_dir_cache;
27 
28 #define OVL_MAX_STACK 500
29 
30 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
31 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
32 MODULE_PARM_DESC(redirect_dir,
33 		 "Default to on or off for the redirect_dir feature");
34 
35 static bool ovl_redirect_always_follow =
36 	IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
37 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
38 		   bool, 0644);
39 MODULE_PARM_DESC(redirect_always_follow,
40 		 "Follow redirects even if redirect_dir feature is turned off");
41 
42 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
43 module_param_named(index, ovl_index_def, bool, 0644);
44 MODULE_PARM_DESC(index,
45 		 "Default to on or off for the inodes index feature");
46 
47 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
48 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
49 MODULE_PARM_DESC(nfs_export,
50 		 "Default to on or off for the NFS export feature");
51 
52 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
53 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
54 MODULE_PARM_DESC(xino_auto,
55 		 "Auto enable xino feature");
56 
57 static void ovl_entry_stack_free(struct ovl_entry *oe)
58 {
59 	unsigned int i;
60 
61 	for (i = 0; i < oe->numlower; i++)
62 		dput(oe->lowerstack[i].dentry);
63 }
64 
65 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
66 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
67 MODULE_PARM_DESC(metacopy,
68 		 "Default to on or off for the metadata only copy up feature");
69 
70 static void ovl_dentry_release(struct dentry *dentry)
71 {
72 	struct ovl_entry *oe = dentry->d_fsdata;
73 
74 	if (oe) {
75 		ovl_entry_stack_free(oe);
76 		kfree_rcu(oe, rcu);
77 	}
78 }
79 
80 static struct dentry *ovl_d_real(struct dentry *dentry,
81 				 const struct inode *inode)
82 {
83 	struct dentry *real = NULL, *lower;
84 
85 	/* It's an overlay file */
86 	if (inode && d_inode(dentry) == inode)
87 		return dentry;
88 
89 	if (!d_is_reg(dentry)) {
90 		if (!inode || inode == d_inode(dentry))
91 			return dentry;
92 		goto bug;
93 	}
94 
95 	real = ovl_dentry_upper(dentry);
96 	if (real && (inode == d_inode(real)))
97 		return real;
98 
99 	if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
100 		return real;
101 
102 	lower = ovl_dentry_lowerdata(dentry);
103 	if (!lower)
104 		goto bug;
105 	real = lower;
106 
107 	/* Handle recursion */
108 	real = d_real(real, inode);
109 
110 	if (!inode || inode == d_inode(real))
111 		return real;
112 bug:
113 	WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
114 	     __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
115 	     inode ? inode->i_ino : 0, real,
116 	     real && d_inode(real) ? d_inode(real)->i_ino : 0);
117 	return dentry;
118 }
119 
120 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
121 {
122 	int ret = 1;
123 
124 	if (weak) {
125 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
126 			ret =  d->d_op->d_weak_revalidate(d, flags);
127 	} else if (d->d_flags & DCACHE_OP_REVALIDATE) {
128 		ret = d->d_op->d_revalidate(d, flags);
129 		if (!ret) {
130 			if (!(flags & LOOKUP_RCU))
131 				d_invalidate(d);
132 			ret = -ESTALE;
133 		}
134 	}
135 	return ret;
136 }
137 
138 static int ovl_dentry_revalidate_common(struct dentry *dentry,
139 					unsigned int flags, bool weak)
140 {
141 	struct ovl_entry *oe = dentry->d_fsdata;
142 	struct inode *inode = d_inode_rcu(dentry);
143 	struct dentry *upper;
144 	unsigned int i;
145 	int ret = 1;
146 
147 	/* Careful in RCU mode */
148 	if (!inode)
149 		return -ECHILD;
150 
151 	upper = ovl_i_dentry_upper(inode);
152 	if (upper)
153 		ret = ovl_revalidate_real(upper, flags, weak);
154 
155 	for (i = 0; ret > 0 && i < oe->numlower; i++) {
156 		ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
157 					  weak);
158 	}
159 	return ret;
160 }
161 
162 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
163 {
164 	return ovl_dentry_revalidate_common(dentry, flags, false);
165 }
166 
167 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
168 {
169 	return ovl_dentry_revalidate_common(dentry, flags, true);
170 }
171 
172 static const struct dentry_operations ovl_dentry_operations = {
173 	.d_release = ovl_dentry_release,
174 	.d_real = ovl_d_real,
175 	.d_revalidate = ovl_dentry_revalidate,
176 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
177 };
178 
179 static struct kmem_cache *ovl_inode_cachep;
180 
181 static struct inode *ovl_alloc_inode(struct super_block *sb)
182 {
183 	struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
184 
185 	if (!oi)
186 		return NULL;
187 
188 	oi->cache = NULL;
189 	oi->redirect = NULL;
190 	oi->version = 0;
191 	oi->flags = 0;
192 	oi->__upperdentry = NULL;
193 	oi->lowerpath.dentry = NULL;
194 	oi->lowerpath.layer = NULL;
195 	oi->lowerdata = NULL;
196 	mutex_init(&oi->lock);
197 
198 	return &oi->vfs_inode;
199 }
200 
201 static void ovl_free_inode(struct inode *inode)
202 {
203 	struct ovl_inode *oi = OVL_I(inode);
204 
205 	kfree(oi->redirect);
206 	mutex_destroy(&oi->lock);
207 	kmem_cache_free(ovl_inode_cachep, oi);
208 }
209 
210 static void ovl_destroy_inode(struct inode *inode)
211 {
212 	struct ovl_inode *oi = OVL_I(inode);
213 
214 	dput(oi->__upperdentry);
215 	dput(oi->lowerpath.dentry);
216 	if (S_ISDIR(inode->i_mode))
217 		ovl_dir_cache_free(inode);
218 	else
219 		iput(oi->lowerdata);
220 }
221 
222 static void ovl_free_fs(struct ovl_fs *ofs)
223 {
224 	struct vfsmount **mounts;
225 	unsigned i;
226 
227 	iput(ofs->workbasedir_trap);
228 	iput(ofs->indexdir_trap);
229 	iput(ofs->workdir_trap);
230 	dput(ofs->whiteout);
231 	dput(ofs->indexdir);
232 	dput(ofs->workdir);
233 	if (ofs->workdir_locked)
234 		ovl_inuse_unlock(ofs->workbasedir);
235 	dput(ofs->workbasedir);
236 	if (ofs->upperdir_locked)
237 		ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
238 
239 	/* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
240 	mounts = (struct vfsmount **) ofs->layers;
241 	for (i = 0; i < ofs->numlayer; i++) {
242 		iput(ofs->layers[i].trap);
243 		mounts[i] = ofs->layers[i].mnt;
244 	}
245 	kern_unmount_array(mounts, ofs->numlayer);
246 	kfree(ofs->layers);
247 	for (i = 0; i < ofs->numfs; i++)
248 		free_anon_bdev(ofs->fs[i].pseudo_dev);
249 	kfree(ofs->fs);
250 
251 	kfree(ofs->config.lowerdir);
252 	kfree(ofs->config.upperdir);
253 	kfree(ofs->config.workdir);
254 	kfree(ofs->config.redirect_mode);
255 	if (ofs->creator_cred)
256 		put_cred(ofs->creator_cred);
257 	kfree(ofs);
258 }
259 
260 static void ovl_put_super(struct super_block *sb)
261 {
262 	struct ovl_fs *ofs = sb->s_fs_info;
263 
264 	ovl_free_fs(ofs);
265 }
266 
267 /* Sync real dirty inodes in upper filesystem (if it exists) */
268 static int ovl_sync_fs(struct super_block *sb, int wait)
269 {
270 	struct ovl_fs *ofs = sb->s_fs_info;
271 	struct super_block *upper_sb;
272 	int ret;
273 
274 	ret = ovl_sync_status(ofs);
275 	/*
276 	 * We have to always set the err, because the return value isn't
277 	 * checked in syncfs, and instead indirectly return an error via
278 	 * the sb's writeback errseq, which VFS inspects after this call.
279 	 */
280 	if (ret < 0) {
281 		errseq_set(&sb->s_wb_err, -EIO);
282 		return -EIO;
283 	}
284 
285 	if (!ret)
286 		return ret;
287 
288 	/*
289 	 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
290 	 * All the super blocks will be iterated, including upper_sb.
291 	 *
292 	 * If this is a syncfs(2) call, then we do need to call
293 	 * sync_filesystem() on upper_sb, but enough if we do it when being
294 	 * called with wait == 1.
295 	 */
296 	if (!wait)
297 		return 0;
298 
299 	upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
300 
301 	down_read(&upper_sb->s_umount);
302 	ret = sync_filesystem(upper_sb);
303 	up_read(&upper_sb->s_umount);
304 
305 	return ret;
306 }
307 
308 /**
309  * ovl_statfs
310  * @dentry: The dentry to query
311  * @buf: The struct kstatfs to fill in with stats
312  *
313  * Get the filesystem statistics.  As writes always target the upper layer
314  * filesystem pass the statfs to the upper filesystem (if it exists)
315  */
316 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
317 {
318 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
319 	struct dentry *root_dentry = dentry->d_sb->s_root;
320 	struct path path;
321 	int err;
322 
323 	ovl_path_real(root_dentry, &path);
324 
325 	err = vfs_statfs(&path, buf);
326 	if (!err) {
327 		buf->f_namelen = ofs->namelen;
328 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
329 	}
330 
331 	return err;
332 }
333 
334 /* Will this overlay be forced to mount/remount ro? */
335 static bool ovl_force_readonly(struct ovl_fs *ofs)
336 {
337 	return (!ovl_upper_mnt(ofs) || !ofs->workdir);
338 }
339 
340 static const char *ovl_redirect_mode_def(void)
341 {
342 	return ovl_redirect_dir_def ? "on" : "off";
343 }
344 
345 static const char * const ovl_xino_str[] = {
346 	"off",
347 	"auto",
348 	"on",
349 };
350 
351 static inline int ovl_xino_def(void)
352 {
353 	return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
354 }
355 
356 /**
357  * ovl_show_options
358  * @m: the seq_file handle
359  * @dentry: The dentry to query
360  *
361  * Prints the mount options for a given superblock.
362  * Returns zero; does not fail.
363  */
364 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
365 {
366 	struct super_block *sb = dentry->d_sb;
367 	struct ovl_fs *ofs = sb->s_fs_info;
368 
369 	seq_show_option(m, "lowerdir", ofs->config.lowerdir);
370 	if (ofs->config.upperdir) {
371 		seq_show_option(m, "upperdir", ofs->config.upperdir);
372 		seq_show_option(m, "workdir", ofs->config.workdir);
373 	}
374 	if (ofs->config.default_permissions)
375 		seq_puts(m, ",default_permissions");
376 	if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
377 		seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
378 	if (ofs->config.index != ovl_index_def)
379 		seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
380 	if (!ofs->config.uuid)
381 		seq_puts(m, ",uuid=off");
382 	if (ofs->config.nfs_export != ovl_nfs_export_def)
383 		seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
384 						"on" : "off");
385 	if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
386 		seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
387 	if (ofs->config.metacopy != ovl_metacopy_def)
388 		seq_printf(m, ",metacopy=%s",
389 			   ofs->config.metacopy ? "on" : "off");
390 	if (ofs->config.ovl_volatile)
391 		seq_puts(m, ",volatile");
392 	if (ofs->config.userxattr)
393 		seq_puts(m, ",userxattr");
394 	return 0;
395 }
396 
397 static int ovl_remount(struct super_block *sb, int *flags, char *data)
398 {
399 	struct ovl_fs *ofs = sb->s_fs_info;
400 	struct super_block *upper_sb;
401 	int ret = 0;
402 
403 	if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
404 		return -EROFS;
405 
406 	if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
407 		upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
408 		if (ovl_should_sync(ofs)) {
409 			down_read(&upper_sb->s_umount);
410 			ret = sync_filesystem(upper_sb);
411 			up_read(&upper_sb->s_umount);
412 		}
413 	}
414 
415 	return ret;
416 }
417 
418 static const struct super_operations ovl_super_operations = {
419 	.alloc_inode	= ovl_alloc_inode,
420 	.free_inode	= ovl_free_inode,
421 	.destroy_inode	= ovl_destroy_inode,
422 	.drop_inode	= generic_delete_inode,
423 	.put_super	= ovl_put_super,
424 	.sync_fs	= ovl_sync_fs,
425 	.statfs		= ovl_statfs,
426 	.show_options	= ovl_show_options,
427 	.remount_fs	= ovl_remount,
428 };
429 
430 enum {
431 	OPT_LOWERDIR,
432 	OPT_UPPERDIR,
433 	OPT_WORKDIR,
434 	OPT_DEFAULT_PERMISSIONS,
435 	OPT_REDIRECT_DIR,
436 	OPT_INDEX_ON,
437 	OPT_INDEX_OFF,
438 	OPT_UUID_ON,
439 	OPT_UUID_OFF,
440 	OPT_NFS_EXPORT_ON,
441 	OPT_USERXATTR,
442 	OPT_NFS_EXPORT_OFF,
443 	OPT_XINO_ON,
444 	OPT_XINO_OFF,
445 	OPT_XINO_AUTO,
446 	OPT_METACOPY_ON,
447 	OPT_METACOPY_OFF,
448 	OPT_VOLATILE,
449 	OPT_ERR,
450 };
451 
452 static const match_table_t ovl_tokens = {
453 	{OPT_LOWERDIR,			"lowerdir=%s"},
454 	{OPT_UPPERDIR,			"upperdir=%s"},
455 	{OPT_WORKDIR,			"workdir=%s"},
456 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
457 	{OPT_REDIRECT_DIR,		"redirect_dir=%s"},
458 	{OPT_INDEX_ON,			"index=on"},
459 	{OPT_INDEX_OFF,			"index=off"},
460 	{OPT_USERXATTR,			"userxattr"},
461 	{OPT_UUID_ON,			"uuid=on"},
462 	{OPT_UUID_OFF,			"uuid=off"},
463 	{OPT_NFS_EXPORT_ON,		"nfs_export=on"},
464 	{OPT_NFS_EXPORT_OFF,		"nfs_export=off"},
465 	{OPT_XINO_ON,			"xino=on"},
466 	{OPT_XINO_OFF,			"xino=off"},
467 	{OPT_XINO_AUTO,			"xino=auto"},
468 	{OPT_METACOPY_ON,		"metacopy=on"},
469 	{OPT_METACOPY_OFF,		"metacopy=off"},
470 	{OPT_VOLATILE,			"volatile"},
471 	{OPT_ERR,			NULL}
472 };
473 
474 static char *ovl_next_opt(char **s)
475 {
476 	char *sbegin = *s;
477 	char *p;
478 
479 	if (sbegin == NULL)
480 		return NULL;
481 
482 	for (p = sbegin; *p; p++) {
483 		if (*p == '\\') {
484 			p++;
485 			if (!*p)
486 				break;
487 		} else if (*p == ',') {
488 			*p = '\0';
489 			*s = p + 1;
490 			return sbegin;
491 		}
492 	}
493 	*s = NULL;
494 	return sbegin;
495 }
496 
497 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
498 {
499 	if (strcmp(mode, "on") == 0) {
500 		config->redirect_dir = true;
501 		/*
502 		 * Does not make sense to have redirect creation without
503 		 * redirect following.
504 		 */
505 		config->redirect_follow = true;
506 	} else if (strcmp(mode, "follow") == 0) {
507 		config->redirect_follow = true;
508 	} else if (strcmp(mode, "off") == 0) {
509 		if (ovl_redirect_always_follow)
510 			config->redirect_follow = true;
511 	} else if (strcmp(mode, "nofollow") != 0) {
512 		pr_err("bad mount option \"redirect_dir=%s\"\n",
513 		       mode);
514 		return -EINVAL;
515 	}
516 
517 	return 0;
518 }
519 
520 static int ovl_parse_opt(char *opt, struct ovl_config *config)
521 {
522 	char *p;
523 	int err;
524 	bool metacopy_opt = false, redirect_opt = false;
525 	bool nfs_export_opt = false, index_opt = false;
526 
527 	config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
528 	if (!config->redirect_mode)
529 		return -ENOMEM;
530 
531 	while ((p = ovl_next_opt(&opt)) != NULL) {
532 		int token;
533 		substring_t args[MAX_OPT_ARGS];
534 
535 		if (!*p)
536 			continue;
537 
538 		token = match_token(p, ovl_tokens, args);
539 		switch (token) {
540 		case OPT_UPPERDIR:
541 			kfree(config->upperdir);
542 			config->upperdir = match_strdup(&args[0]);
543 			if (!config->upperdir)
544 				return -ENOMEM;
545 			break;
546 
547 		case OPT_LOWERDIR:
548 			kfree(config->lowerdir);
549 			config->lowerdir = match_strdup(&args[0]);
550 			if (!config->lowerdir)
551 				return -ENOMEM;
552 			break;
553 
554 		case OPT_WORKDIR:
555 			kfree(config->workdir);
556 			config->workdir = match_strdup(&args[0]);
557 			if (!config->workdir)
558 				return -ENOMEM;
559 			break;
560 
561 		case OPT_DEFAULT_PERMISSIONS:
562 			config->default_permissions = true;
563 			break;
564 
565 		case OPT_REDIRECT_DIR:
566 			kfree(config->redirect_mode);
567 			config->redirect_mode = match_strdup(&args[0]);
568 			if (!config->redirect_mode)
569 				return -ENOMEM;
570 			redirect_opt = true;
571 			break;
572 
573 		case OPT_INDEX_ON:
574 			config->index = true;
575 			index_opt = true;
576 			break;
577 
578 		case OPT_INDEX_OFF:
579 			config->index = false;
580 			index_opt = true;
581 			break;
582 
583 		case OPT_UUID_ON:
584 			config->uuid = true;
585 			break;
586 
587 		case OPT_UUID_OFF:
588 			config->uuid = false;
589 			break;
590 
591 		case OPT_NFS_EXPORT_ON:
592 			config->nfs_export = true;
593 			nfs_export_opt = true;
594 			break;
595 
596 		case OPT_NFS_EXPORT_OFF:
597 			config->nfs_export = false;
598 			nfs_export_opt = true;
599 			break;
600 
601 		case OPT_XINO_ON:
602 			config->xino = OVL_XINO_ON;
603 			break;
604 
605 		case OPT_XINO_OFF:
606 			config->xino = OVL_XINO_OFF;
607 			break;
608 
609 		case OPT_XINO_AUTO:
610 			config->xino = OVL_XINO_AUTO;
611 			break;
612 
613 		case OPT_METACOPY_ON:
614 			config->metacopy = true;
615 			metacopy_opt = true;
616 			break;
617 
618 		case OPT_METACOPY_OFF:
619 			config->metacopy = false;
620 			metacopy_opt = true;
621 			break;
622 
623 		case OPT_VOLATILE:
624 			config->ovl_volatile = true;
625 			break;
626 
627 		case OPT_USERXATTR:
628 			config->userxattr = true;
629 			break;
630 
631 		default:
632 			pr_err("unrecognized mount option \"%s\" or missing value\n",
633 					p);
634 			return -EINVAL;
635 		}
636 	}
637 
638 	/* Workdir/index are useless in non-upper mount */
639 	if (!config->upperdir) {
640 		if (config->workdir) {
641 			pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
642 				config->workdir);
643 			kfree(config->workdir);
644 			config->workdir = NULL;
645 		}
646 		if (config->index && index_opt) {
647 			pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
648 			index_opt = false;
649 		}
650 		config->index = false;
651 	}
652 
653 	if (!config->upperdir && config->ovl_volatile) {
654 		pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
655 		config->ovl_volatile = false;
656 	}
657 
658 	err = ovl_parse_redirect_mode(config, config->redirect_mode);
659 	if (err)
660 		return err;
661 
662 	/*
663 	 * This is to make the logic below simpler.  It doesn't make any other
664 	 * difference, since config->redirect_dir is only used for upper.
665 	 */
666 	if (!config->upperdir && config->redirect_follow)
667 		config->redirect_dir = true;
668 
669 	/* Resolve metacopy -> redirect_dir dependency */
670 	if (config->metacopy && !config->redirect_dir) {
671 		if (metacopy_opt && redirect_opt) {
672 			pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
673 			       config->redirect_mode);
674 			return -EINVAL;
675 		}
676 		if (redirect_opt) {
677 			/*
678 			 * There was an explicit redirect_dir=... that resulted
679 			 * in this conflict.
680 			 */
681 			pr_info("disabling metacopy due to redirect_dir=%s\n",
682 				config->redirect_mode);
683 			config->metacopy = false;
684 		} else {
685 			/* Automatically enable redirect otherwise. */
686 			config->redirect_follow = config->redirect_dir = true;
687 		}
688 	}
689 
690 	/* Resolve nfs_export -> index dependency */
691 	if (config->nfs_export && !config->index) {
692 		if (!config->upperdir && config->redirect_follow) {
693 			pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
694 			config->nfs_export = false;
695 		} else if (nfs_export_opt && index_opt) {
696 			pr_err("conflicting options: nfs_export=on,index=off\n");
697 			return -EINVAL;
698 		} else if (index_opt) {
699 			/*
700 			 * There was an explicit index=off that resulted
701 			 * in this conflict.
702 			 */
703 			pr_info("disabling nfs_export due to index=off\n");
704 			config->nfs_export = false;
705 		} else {
706 			/* Automatically enable index otherwise. */
707 			config->index = true;
708 		}
709 	}
710 
711 	/* Resolve nfs_export -> !metacopy dependency */
712 	if (config->nfs_export && config->metacopy) {
713 		if (nfs_export_opt && metacopy_opt) {
714 			pr_err("conflicting options: nfs_export=on,metacopy=on\n");
715 			return -EINVAL;
716 		}
717 		if (metacopy_opt) {
718 			/*
719 			 * There was an explicit metacopy=on that resulted
720 			 * in this conflict.
721 			 */
722 			pr_info("disabling nfs_export due to metacopy=on\n");
723 			config->nfs_export = false;
724 		} else {
725 			/*
726 			 * There was an explicit nfs_export=on that resulted
727 			 * in this conflict.
728 			 */
729 			pr_info("disabling metacopy due to nfs_export=on\n");
730 			config->metacopy = false;
731 		}
732 	}
733 
734 
735 	/* Resolve userxattr -> !redirect && !metacopy dependency */
736 	if (config->userxattr) {
737 		if (config->redirect_follow && redirect_opt) {
738 			pr_err("conflicting options: userxattr,redirect_dir=%s\n",
739 			       config->redirect_mode);
740 			return -EINVAL;
741 		}
742 		if (config->metacopy && metacopy_opt) {
743 			pr_err("conflicting options: userxattr,metacopy=on\n");
744 			return -EINVAL;
745 		}
746 		/*
747 		 * Silently disable default setting of redirect and metacopy.
748 		 * This shall be the default in the future as well: these
749 		 * options must be explicitly enabled if used together with
750 		 * userxattr.
751 		 */
752 		config->redirect_dir = config->redirect_follow = false;
753 		config->metacopy = false;
754 	}
755 
756 	return 0;
757 }
758 
759 #define OVL_WORKDIR_NAME "work"
760 #define OVL_INDEXDIR_NAME "index"
761 
762 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
763 					 const char *name, bool persist)
764 {
765 	struct inode *dir =  ofs->workbasedir->d_inode;
766 	struct vfsmount *mnt = ovl_upper_mnt(ofs);
767 	struct dentry *work;
768 	int err;
769 	bool retried = false;
770 
771 	inode_lock_nested(dir, I_MUTEX_PARENT);
772 retry:
773 	work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
774 
775 	if (!IS_ERR(work)) {
776 		struct iattr attr = {
777 			.ia_valid = ATTR_MODE,
778 			.ia_mode = S_IFDIR | 0,
779 		};
780 
781 		if (work->d_inode) {
782 			err = -EEXIST;
783 			if (retried)
784 				goto out_dput;
785 
786 			if (persist)
787 				goto out_unlock;
788 
789 			retried = true;
790 			err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
791 			dput(work);
792 			if (err == -EINVAL) {
793 				work = ERR_PTR(err);
794 				goto out_unlock;
795 			}
796 			goto retry;
797 		}
798 
799 		err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
800 		if (err)
801 			goto out_dput;
802 
803 		/* Weird filesystem returning with hashed negative (kernfs)? */
804 		err = -EINVAL;
805 		if (d_really_is_negative(work))
806 			goto out_dput;
807 
808 		/*
809 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
810 		 *
811 		 * a) success (there was a POSIX ACL xattr and was removed)
812 		 * b) -ENODATA (there was no POSIX ACL xattr)
813 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
814 		 *
815 		 * There are various other error values that could effectively
816 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
817 		 * if the xattr name is too long), but the set of filesystems
818 		 * allowed as upper are limited to "normal" ones, where checking
819 		 * for the above two errors is sufficient.
820 		 */
821 		err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
822 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
823 			goto out_dput;
824 
825 		err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
826 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
827 			goto out_dput;
828 
829 		/* Clear any inherited mode bits */
830 		inode_lock(work->d_inode);
831 		err = ovl_do_notify_change(ofs, work, &attr);
832 		inode_unlock(work->d_inode);
833 		if (err)
834 			goto out_dput;
835 	} else {
836 		err = PTR_ERR(work);
837 		goto out_err;
838 	}
839 out_unlock:
840 	inode_unlock(dir);
841 	return work;
842 
843 out_dput:
844 	dput(work);
845 out_err:
846 	pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
847 		ofs->config.workdir, name, -err);
848 	work = NULL;
849 	goto out_unlock;
850 }
851 
852 static void ovl_unescape(char *s)
853 {
854 	char *d = s;
855 
856 	for (;; s++, d++) {
857 		if (*s == '\\')
858 			s++;
859 		*d = *s;
860 		if (!*s)
861 			break;
862 	}
863 }
864 
865 static int ovl_mount_dir_noesc(const char *name, struct path *path)
866 {
867 	int err = -EINVAL;
868 
869 	if (!*name) {
870 		pr_err("empty lowerdir\n");
871 		goto out;
872 	}
873 	err = kern_path(name, LOOKUP_FOLLOW, path);
874 	if (err) {
875 		pr_err("failed to resolve '%s': %i\n", name, err);
876 		goto out;
877 	}
878 	err = -EINVAL;
879 	if (ovl_dentry_weird(path->dentry)) {
880 		pr_err("filesystem on '%s' not supported\n", name);
881 		goto out_put;
882 	}
883 	if (!d_is_dir(path->dentry)) {
884 		pr_err("'%s' not a directory\n", name);
885 		goto out_put;
886 	}
887 	return 0;
888 
889 out_put:
890 	path_put_init(path);
891 out:
892 	return err;
893 }
894 
895 static int ovl_mount_dir(const char *name, struct path *path)
896 {
897 	int err = -ENOMEM;
898 	char *tmp = kstrdup(name, GFP_KERNEL);
899 
900 	if (tmp) {
901 		ovl_unescape(tmp);
902 		err = ovl_mount_dir_noesc(tmp, path);
903 
904 		if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
905 			pr_err("filesystem on '%s' not supported as upperdir\n",
906 			       tmp);
907 			path_put_init(path);
908 			err = -EINVAL;
909 		}
910 		kfree(tmp);
911 	}
912 	return err;
913 }
914 
915 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
916 			     const char *name)
917 {
918 	struct kstatfs statfs;
919 	int err = vfs_statfs(path, &statfs);
920 
921 	if (err)
922 		pr_err("statfs failed on '%s'\n", name);
923 	else
924 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
925 
926 	return err;
927 }
928 
929 static int ovl_lower_dir(const char *name, struct path *path,
930 			 struct ovl_fs *ofs, int *stack_depth)
931 {
932 	int fh_type;
933 	int err;
934 
935 	err = ovl_mount_dir_noesc(name, path);
936 	if (err)
937 		return err;
938 
939 	err = ovl_check_namelen(path, ofs, name);
940 	if (err)
941 		return err;
942 
943 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
944 
945 	/*
946 	 * The inodes index feature and NFS export need to encode and decode
947 	 * file handles, so they require that all layers support them.
948 	 */
949 	fh_type = ovl_can_decode_fh(path->dentry->d_sb);
950 	if ((ofs->config.nfs_export ||
951 	     (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
952 		ofs->config.index = false;
953 		ofs->config.nfs_export = false;
954 		pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
955 			name);
956 	}
957 	/*
958 	 * Decoding origin file handle is required for persistent st_ino.
959 	 * Without persistent st_ino, xino=auto falls back to xino=off.
960 	 */
961 	if (ofs->config.xino == OVL_XINO_AUTO &&
962 	    ofs->config.upperdir && !fh_type) {
963 		ofs->config.xino = OVL_XINO_OFF;
964 		pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
965 			name);
966 	}
967 
968 	/* Check if lower fs has 32bit inode numbers */
969 	if (fh_type != FILEID_INO32_GEN)
970 		ofs->xino_mode = -1;
971 
972 	return 0;
973 }
974 
975 /* Workdir should not be subdir of upperdir and vice versa */
976 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
977 {
978 	bool ok = false;
979 
980 	if (workdir != upperdir) {
981 		ok = (lock_rename(workdir, upperdir) == NULL);
982 		unlock_rename(workdir, upperdir);
983 	}
984 	return ok;
985 }
986 
987 static unsigned int ovl_split_lowerdirs(char *str)
988 {
989 	unsigned int ctr = 1;
990 	char *s, *d;
991 
992 	for (s = d = str;; s++, d++) {
993 		if (*s == '\\') {
994 			s++;
995 		} else if (*s == ':') {
996 			*d = '\0';
997 			ctr++;
998 			continue;
999 		}
1000 		*d = *s;
1001 		if (!*s)
1002 			break;
1003 	}
1004 	return ctr;
1005 }
1006 
1007 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1008 			     struct dentry *dentry, struct inode *inode,
1009 			     const char *name, void *buffer, size_t size)
1010 {
1011 	return -EOPNOTSUPP;
1012 }
1013 
1014 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1015 			     struct user_namespace *mnt_userns,
1016 			     struct dentry *dentry, struct inode *inode,
1017 			     const char *name, const void *value,
1018 			     size_t size, int flags)
1019 {
1020 	return -EOPNOTSUPP;
1021 }
1022 
1023 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1024 			       struct dentry *dentry, struct inode *inode,
1025 			       const char *name, void *buffer, size_t size)
1026 {
1027 	return ovl_xattr_get(dentry, inode, name, buffer, size);
1028 }
1029 
1030 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1031 			       struct user_namespace *mnt_userns,
1032 			       struct dentry *dentry, struct inode *inode,
1033 			       const char *name, const void *value,
1034 			       size_t size, int flags)
1035 {
1036 	return ovl_xattr_set(dentry, inode, name, value, size, flags);
1037 }
1038 
1039 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1040 	.prefix	= OVL_XATTR_TRUSTED_PREFIX,
1041 	.get = ovl_own_xattr_get,
1042 	.set = ovl_own_xattr_set,
1043 };
1044 
1045 static const struct xattr_handler ovl_own_user_xattr_handler = {
1046 	.prefix	= OVL_XATTR_USER_PREFIX,
1047 	.get = ovl_own_xattr_get,
1048 	.set = ovl_own_xattr_set,
1049 };
1050 
1051 static const struct xattr_handler ovl_other_xattr_handler = {
1052 	.prefix	= "", /* catch all */
1053 	.get = ovl_other_xattr_get,
1054 	.set = ovl_other_xattr_set,
1055 };
1056 
1057 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1058 #ifdef CONFIG_FS_POSIX_ACL
1059 	&posix_acl_access_xattr_handler,
1060 	&posix_acl_default_xattr_handler,
1061 #endif
1062 	&ovl_own_trusted_xattr_handler,
1063 	&ovl_other_xattr_handler,
1064 	NULL
1065 };
1066 
1067 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1068 #ifdef CONFIG_FS_POSIX_ACL
1069 	&posix_acl_access_xattr_handler,
1070 	&posix_acl_default_xattr_handler,
1071 #endif
1072 	&ovl_own_user_xattr_handler,
1073 	&ovl_other_xattr_handler,
1074 	NULL
1075 };
1076 
1077 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1078 			  struct inode **ptrap, const char *name)
1079 {
1080 	struct inode *trap;
1081 	int err;
1082 
1083 	trap = ovl_get_trap_inode(sb, dir);
1084 	err = PTR_ERR_OR_ZERO(trap);
1085 	if (err) {
1086 		if (err == -ELOOP)
1087 			pr_err("conflicting %s path\n", name);
1088 		return err;
1089 	}
1090 
1091 	*ptrap = trap;
1092 	return 0;
1093 }
1094 
1095 /*
1096  * Determine how we treat concurrent use of upperdir/workdir based on the
1097  * index feature. This is papering over mount leaks of container runtimes,
1098  * for example, an old overlay mount is leaked and now its upperdir is
1099  * attempted to be used as a lower layer in a new overlay mount.
1100  */
1101 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1102 {
1103 	if (ofs->config.index) {
1104 		pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1105 		       name);
1106 		return -EBUSY;
1107 	} else {
1108 		pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1109 			name);
1110 		return 0;
1111 	}
1112 }
1113 
1114 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1115 			 struct ovl_layer *upper_layer, struct path *upperpath)
1116 {
1117 	struct vfsmount *upper_mnt;
1118 	int err;
1119 
1120 	err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1121 	if (err)
1122 		goto out;
1123 
1124 	/* Upperdir path should not be r/o */
1125 	if (__mnt_is_readonly(upperpath->mnt)) {
1126 		pr_err("upper fs is r/o, try multi-lower layers mount\n");
1127 		err = -EINVAL;
1128 		goto out;
1129 	}
1130 
1131 	err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1132 	if (err)
1133 		goto out;
1134 
1135 	err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1136 			     "upperdir");
1137 	if (err)
1138 		goto out;
1139 
1140 	upper_mnt = clone_private_mount(upperpath);
1141 	err = PTR_ERR(upper_mnt);
1142 	if (IS_ERR(upper_mnt)) {
1143 		pr_err("failed to clone upperpath\n");
1144 		goto out;
1145 	}
1146 
1147 	/* Don't inherit atime flags */
1148 	upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1149 	upper_layer->mnt = upper_mnt;
1150 	upper_layer->idx = 0;
1151 	upper_layer->fsid = 0;
1152 
1153 	/*
1154 	 * Inherit SB_NOSEC flag from upperdir.
1155 	 *
1156 	 * This optimization changes behavior when a security related attribute
1157 	 * (suid/sgid/security.*) is changed on an underlying layer.  This is
1158 	 * okay because we don't yet have guarantees in that case, but it will
1159 	 * need careful treatment once we want to honour changes to underlying
1160 	 * filesystems.
1161 	 */
1162 	if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1163 		sb->s_flags |= SB_NOSEC;
1164 
1165 	if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1166 		ofs->upperdir_locked = true;
1167 	} else {
1168 		err = ovl_report_in_use(ofs, "upperdir");
1169 		if (err)
1170 			goto out;
1171 	}
1172 
1173 	err = 0;
1174 out:
1175 	return err;
1176 }
1177 
1178 /*
1179  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1180  * negative values if error is encountered.
1181  */
1182 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1183 {
1184 	struct dentry *workdir = ofs->workdir;
1185 	struct inode *dir = d_inode(workdir);
1186 	struct dentry *temp;
1187 	struct dentry *dest;
1188 	struct dentry *whiteout;
1189 	struct name_snapshot name;
1190 	int err;
1191 
1192 	inode_lock_nested(dir, I_MUTEX_PARENT);
1193 
1194 	temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1195 	err = PTR_ERR(temp);
1196 	if (IS_ERR(temp))
1197 		goto out_unlock;
1198 
1199 	dest = ovl_lookup_temp(ofs, workdir);
1200 	err = PTR_ERR(dest);
1201 	if (IS_ERR(dest)) {
1202 		dput(temp);
1203 		goto out_unlock;
1204 	}
1205 
1206 	/* Name is inline and stable - using snapshot as a copy helper */
1207 	take_dentry_name_snapshot(&name, temp);
1208 	err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1209 	if (err) {
1210 		if (err == -EINVAL)
1211 			err = 0;
1212 		goto cleanup_temp;
1213 	}
1214 
1215 	whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1216 	err = PTR_ERR(whiteout);
1217 	if (IS_ERR(whiteout))
1218 		goto cleanup_temp;
1219 
1220 	err = ovl_is_whiteout(whiteout);
1221 
1222 	/* Best effort cleanup of whiteout and temp file */
1223 	if (err)
1224 		ovl_cleanup(ofs, dir, whiteout);
1225 	dput(whiteout);
1226 
1227 cleanup_temp:
1228 	ovl_cleanup(ofs, dir, temp);
1229 	release_dentry_name_snapshot(&name);
1230 	dput(temp);
1231 	dput(dest);
1232 
1233 out_unlock:
1234 	inode_unlock(dir);
1235 
1236 	return err;
1237 }
1238 
1239 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1240 					   struct dentry *parent,
1241 					   const char *name, umode_t mode)
1242 {
1243 	size_t len = strlen(name);
1244 	struct dentry *child;
1245 
1246 	inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1247 	child = ovl_lookup_upper(ofs, name, parent, len);
1248 	if (!IS_ERR(child) && !child->d_inode)
1249 		child = ovl_create_real(ofs, parent->d_inode, child,
1250 					OVL_CATTR(mode));
1251 	inode_unlock(parent->d_inode);
1252 	dput(parent);
1253 
1254 	return child;
1255 }
1256 
1257 /*
1258  * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1259  * present.
1260  */
1261 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1262 {
1263 	unsigned int ctr;
1264 	struct dentry *d = dget(ofs->workbasedir);
1265 	static const char *const volatile_path[] = {
1266 		OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1267 	};
1268 	const char *const *name = volatile_path;
1269 
1270 	for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1271 		d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1272 		if (IS_ERR(d))
1273 			return PTR_ERR(d);
1274 	}
1275 	dput(d);
1276 	return 0;
1277 }
1278 
1279 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1280 			    const struct path *workpath)
1281 {
1282 	struct vfsmount *mnt = ovl_upper_mnt(ofs);
1283 	struct dentry *workdir;
1284 	struct file *tmpfile;
1285 	bool rename_whiteout;
1286 	bool d_type;
1287 	int fh_type;
1288 	int err;
1289 
1290 	err = mnt_want_write(mnt);
1291 	if (err)
1292 		return err;
1293 
1294 	workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1295 	err = PTR_ERR(workdir);
1296 	if (IS_ERR_OR_NULL(workdir))
1297 		goto out;
1298 
1299 	ofs->workdir = workdir;
1300 
1301 	err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1302 	if (err)
1303 		goto out;
1304 
1305 	/*
1306 	 * Upper should support d_type, else whiteouts are visible.  Given
1307 	 * workdir and upper are on same fs, we can do iterate_dir() on
1308 	 * workdir. This check requires successful creation of workdir in
1309 	 * previous step.
1310 	 */
1311 	err = ovl_check_d_type_supported(workpath);
1312 	if (err < 0)
1313 		goto out;
1314 
1315 	d_type = err;
1316 	if (!d_type)
1317 		pr_warn("upper fs needs to support d_type.\n");
1318 
1319 	/* Check if upper/work fs supports O_TMPFILE */
1320 	tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1321 	ofs->tmpfile = !IS_ERR(tmpfile);
1322 	if (ofs->tmpfile)
1323 		fput(tmpfile);
1324 	else
1325 		pr_warn("upper fs does not support tmpfile.\n");
1326 
1327 
1328 	/* Check if upper/work fs supports RENAME_WHITEOUT */
1329 	err = ovl_check_rename_whiteout(ofs);
1330 	if (err < 0)
1331 		goto out;
1332 
1333 	rename_whiteout = err;
1334 	if (!rename_whiteout)
1335 		pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1336 
1337 	/*
1338 	 * Check if upper/work fs supports (trusted|user).overlay.* xattr
1339 	 */
1340 	err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1341 	if (err) {
1342 		pr_warn("failed to set xattr on upper\n");
1343 		ofs->noxattr = true;
1344 		if (ofs->config.index || ofs->config.metacopy) {
1345 			ofs->config.index = false;
1346 			ofs->config.metacopy = false;
1347 			pr_warn("...falling back to index=off,metacopy=off.\n");
1348 		}
1349 		/*
1350 		 * xattr support is required for persistent st_ino.
1351 		 * Without persistent st_ino, xino=auto falls back to xino=off.
1352 		 */
1353 		if (ofs->config.xino == OVL_XINO_AUTO) {
1354 			ofs->config.xino = OVL_XINO_OFF;
1355 			pr_warn("...falling back to xino=off.\n");
1356 		}
1357 		if (err == -EPERM && !ofs->config.userxattr)
1358 			pr_info("try mounting with 'userxattr' option\n");
1359 		err = 0;
1360 	} else {
1361 		ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1362 	}
1363 
1364 	/*
1365 	 * We allowed sub-optimal upper fs configuration and don't want to break
1366 	 * users over kernel upgrade, but we never allowed remote upper fs, so
1367 	 * we can enforce strict requirements for remote upper fs.
1368 	 */
1369 	if (ovl_dentry_remote(ofs->workdir) &&
1370 	    (!d_type || !rename_whiteout || ofs->noxattr)) {
1371 		pr_err("upper fs missing required features.\n");
1372 		err = -EINVAL;
1373 		goto out;
1374 	}
1375 
1376 	/*
1377 	 * For volatile mount, create a incompat/volatile/dirty file to keep
1378 	 * track of it.
1379 	 */
1380 	if (ofs->config.ovl_volatile) {
1381 		err = ovl_create_volatile_dirty(ofs);
1382 		if (err < 0) {
1383 			pr_err("Failed to create volatile/dirty file.\n");
1384 			goto out;
1385 		}
1386 	}
1387 
1388 	/* Check if upper/work fs supports file handles */
1389 	fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1390 	if (ofs->config.index && !fh_type) {
1391 		ofs->config.index = false;
1392 		pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1393 	}
1394 
1395 	/* Check if upper fs has 32bit inode numbers */
1396 	if (fh_type != FILEID_INO32_GEN)
1397 		ofs->xino_mode = -1;
1398 
1399 	/* NFS export of r/w mount depends on index */
1400 	if (ofs->config.nfs_export && !ofs->config.index) {
1401 		pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1402 		ofs->config.nfs_export = false;
1403 	}
1404 out:
1405 	mnt_drop_write(mnt);
1406 	return err;
1407 }
1408 
1409 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1410 			   const struct path *upperpath)
1411 {
1412 	int err;
1413 	struct path workpath = { };
1414 
1415 	err = ovl_mount_dir(ofs->config.workdir, &workpath);
1416 	if (err)
1417 		goto out;
1418 
1419 	err = -EINVAL;
1420 	if (upperpath->mnt != workpath.mnt) {
1421 		pr_err("workdir and upperdir must reside under the same mount\n");
1422 		goto out;
1423 	}
1424 	if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1425 		pr_err("workdir and upperdir must be separate subtrees\n");
1426 		goto out;
1427 	}
1428 
1429 	ofs->workbasedir = dget(workpath.dentry);
1430 
1431 	if (ovl_inuse_trylock(ofs->workbasedir)) {
1432 		ofs->workdir_locked = true;
1433 	} else {
1434 		err = ovl_report_in_use(ofs, "workdir");
1435 		if (err)
1436 			goto out;
1437 	}
1438 
1439 	err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1440 			     "workdir");
1441 	if (err)
1442 		goto out;
1443 
1444 	err = ovl_make_workdir(sb, ofs, &workpath);
1445 
1446 out:
1447 	path_put(&workpath);
1448 
1449 	return err;
1450 }
1451 
1452 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1453 			    struct ovl_entry *oe, const struct path *upperpath)
1454 {
1455 	struct vfsmount *mnt = ovl_upper_mnt(ofs);
1456 	struct dentry *indexdir;
1457 	int err;
1458 
1459 	err = mnt_want_write(mnt);
1460 	if (err)
1461 		return err;
1462 
1463 	/* Verify lower root is upper root origin */
1464 	err = ovl_verify_origin(ofs, upperpath->dentry,
1465 				oe->lowerstack[0].dentry, true);
1466 	if (err) {
1467 		pr_err("failed to verify upper root origin\n");
1468 		goto out;
1469 	}
1470 
1471 	/* index dir will act also as workdir */
1472 	iput(ofs->workdir_trap);
1473 	ofs->workdir_trap = NULL;
1474 	dput(ofs->workdir);
1475 	ofs->workdir = NULL;
1476 	indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1477 	if (IS_ERR(indexdir)) {
1478 		err = PTR_ERR(indexdir);
1479 	} else if (indexdir) {
1480 		ofs->indexdir = indexdir;
1481 		ofs->workdir = dget(indexdir);
1482 
1483 		err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1484 				     "indexdir");
1485 		if (err)
1486 			goto out;
1487 
1488 		/*
1489 		 * Verify upper root is exclusively associated with index dir.
1490 		 * Older kernels stored upper fh in ".overlay.origin"
1491 		 * xattr. If that xattr exists, verify that it is a match to
1492 		 * upper dir file handle. In any case, verify or set xattr
1493 		 * ".overlay.upper" to indicate that index may have
1494 		 * directory entries.
1495 		 */
1496 		if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1497 			err = ovl_verify_set_fh(ofs, ofs->indexdir,
1498 						OVL_XATTR_ORIGIN,
1499 						upperpath->dentry, true, false);
1500 			if (err)
1501 				pr_err("failed to verify index dir 'origin' xattr\n");
1502 		}
1503 		err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1504 				       true);
1505 		if (err)
1506 			pr_err("failed to verify index dir 'upper' xattr\n");
1507 
1508 		/* Cleanup bad/stale/orphan index entries */
1509 		if (!err)
1510 			err = ovl_indexdir_cleanup(ofs);
1511 	}
1512 	if (err || !ofs->indexdir)
1513 		pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1514 
1515 out:
1516 	mnt_drop_write(mnt);
1517 	return err;
1518 }
1519 
1520 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1521 {
1522 	unsigned int i;
1523 
1524 	if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1525 		return true;
1526 
1527 	/*
1528 	 * We allow using single lower with null uuid for index and nfs_export
1529 	 * for example to support those features with single lower squashfs.
1530 	 * To avoid regressions in setups of overlay with re-formatted lower
1531 	 * squashfs, do not allow decoding origin with lower null uuid unless
1532 	 * user opted-in to one of the new features that require following the
1533 	 * lower inode of non-dir upper.
1534 	 */
1535 	if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1536 		return false;
1537 
1538 	for (i = 0; i < ofs->numfs; i++) {
1539 		/*
1540 		 * We use uuid to associate an overlay lower file handle with a
1541 		 * lower layer, so we can accept lower fs with null uuid as long
1542 		 * as all lower layers with null uuid are on the same fs.
1543 		 * if we detect multiple lower fs with the same uuid, we
1544 		 * disable lower file handle decoding on all of them.
1545 		 */
1546 		if (ofs->fs[i].is_lower &&
1547 		    uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1548 			ofs->fs[i].bad_uuid = true;
1549 			return false;
1550 		}
1551 	}
1552 	return true;
1553 }
1554 
1555 /* Get a unique fsid for the layer */
1556 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1557 {
1558 	struct super_block *sb = path->mnt->mnt_sb;
1559 	unsigned int i;
1560 	dev_t dev;
1561 	int err;
1562 	bool bad_uuid = false;
1563 	bool warn = false;
1564 
1565 	for (i = 0; i < ofs->numfs; i++) {
1566 		if (ofs->fs[i].sb == sb)
1567 			return i;
1568 	}
1569 
1570 	if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1571 		bad_uuid = true;
1572 		if (ofs->config.xino == OVL_XINO_AUTO) {
1573 			ofs->config.xino = OVL_XINO_OFF;
1574 			warn = true;
1575 		}
1576 		if (ofs->config.index || ofs->config.nfs_export) {
1577 			ofs->config.index = false;
1578 			ofs->config.nfs_export = false;
1579 			warn = true;
1580 		}
1581 		if (warn) {
1582 			pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1583 				uuid_is_null(&sb->s_uuid) ? "null" :
1584 							    "conflicting",
1585 				path->dentry, ovl_xino_str[ofs->config.xino]);
1586 		}
1587 	}
1588 
1589 	err = get_anon_bdev(&dev);
1590 	if (err) {
1591 		pr_err("failed to get anonymous bdev for lowerpath\n");
1592 		return err;
1593 	}
1594 
1595 	ofs->fs[ofs->numfs].sb = sb;
1596 	ofs->fs[ofs->numfs].pseudo_dev = dev;
1597 	ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1598 
1599 	return ofs->numfs++;
1600 }
1601 
1602 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1603 			  struct path *stack, unsigned int numlower,
1604 			  struct ovl_layer *layers)
1605 {
1606 	int err;
1607 	unsigned int i;
1608 
1609 	err = -ENOMEM;
1610 	ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1611 	if (ofs->fs == NULL)
1612 		goto out;
1613 
1614 	/* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1615 	ofs->numfs++;
1616 
1617 	/*
1618 	 * All lower layers that share the same fs as upper layer, use the same
1619 	 * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1620 	 * only overlay to simplify ovl_fs_free().
1621 	 * is_lower will be set if upper fs is shared with a lower layer.
1622 	 */
1623 	err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1624 	if (err) {
1625 		pr_err("failed to get anonymous bdev for upper fs\n");
1626 		goto out;
1627 	}
1628 
1629 	if (ovl_upper_mnt(ofs)) {
1630 		ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1631 		ofs->fs[0].is_lower = false;
1632 	}
1633 
1634 	for (i = 0; i < numlower; i++) {
1635 		struct vfsmount *mnt;
1636 		struct inode *trap;
1637 		int fsid;
1638 
1639 		err = fsid = ovl_get_fsid(ofs, &stack[i]);
1640 		if (err < 0)
1641 			goto out;
1642 
1643 		/*
1644 		 * Check if lower root conflicts with this overlay layers before
1645 		 * checking if it is in-use as upperdir/workdir of "another"
1646 		 * mount, because we do not bother to check in ovl_is_inuse() if
1647 		 * the upperdir/workdir is in fact in-use by our
1648 		 * upperdir/workdir.
1649 		 */
1650 		err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1651 		if (err)
1652 			goto out;
1653 
1654 		if (ovl_is_inuse(stack[i].dentry)) {
1655 			err = ovl_report_in_use(ofs, "lowerdir");
1656 			if (err) {
1657 				iput(trap);
1658 				goto out;
1659 			}
1660 		}
1661 
1662 		mnt = clone_private_mount(&stack[i]);
1663 		err = PTR_ERR(mnt);
1664 		if (IS_ERR(mnt)) {
1665 			pr_err("failed to clone lowerpath\n");
1666 			iput(trap);
1667 			goto out;
1668 		}
1669 
1670 		/*
1671 		 * Make lower layers R/O.  That way fchmod/fchown on lower file
1672 		 * will fail instead of modifying lower fs.
1673 		 */
1674 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1675 
1676 		layers[ofs->numlayer].trap = trap;
1677 		layers[ofs->numlayer].mnt = mnt;
1678 		layers[ofs->numlayer].idx = ofs->numlayer;
1679 		layers[ofs->numlayer].fsid = fsid;
1680 		layers[ofs->numlayer].fs = &ofs->fs[fsid];
1681 		ofs->numlayer++;
1682 		ofs->fs[fsid].is_lower = true;
1683 	}
1684 
1685 	/*
1686 	 * When all layers on same fs, overlay can use real inode numbers.
1687 	 * With mount option "xino=<on|auto>", mounter declares that there are
1688 	 * enough free high bits in underlying fs to hold the unique fsid.
1689 	 * If overlayfs does encounter underlying inodes using the high xino
1690 	 * bits reserved for fsid, it emits a warning and uses the original
1691 	 * inode number or a non persistent inode number allocated from a
1692 	 * dedicated range.
1693 	 */
1694 	if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1695 		if (ofs->config.xino == OVL_XINO_ON)
1696 			pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1697 		ofs->xino_mode = 0;
1698 	} else if (ofs->config.xino == OVL_XINO_OFF) {
1699 		ofs->xino_mode = -1;
1700 	} else if (ofs->xino_mode < 0) {
1701 		/*
1702 		 * This is a roundup of number of bits needed for encoding
1703 		 * fsid, where fsid 0 is reserved for upper fs (even with
1704 		 * lower only overlay) +1 extra bit is reserved for the non
1705 		 * persistent inode number range that is used for resolving
1706 		 * xino lower bits overflow.
1707 		 */
1708 		BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1709 		ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1710 	}
1711 
1712 	if (ofs->xino_mode > 0) {
1713 		pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1714 			ofs->xino_mode);
1715 	}
1716 
1717 	err = 0;
1718 out:
1719 	return err;
1720 }
1721 
1722 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1723 				const char *lower, unsigned int numlower,
1724 				struct ovl_fs *ofs, struct ovl_layer *layers)
1725 {
1726 	int err;
1727 	struct path *stack = NULL;
1728 	unsigned int i;
1729 	struct ovl_entry *oe;
1730 
1731 	if (!ofs->config.upperdir && numlower == 1) {
1732 		pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1733 		return ERR_PTR(-EINVAL);
1734 	}
1735 
1736 	stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1737 	if (!stack)
1738 		return ERR_PTR(-ENOMEM);
1739 
1740 	err = -EINVAL;
1741 	for (i = 0; i < numlower; i++) {
1742 		err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1743 		if (err)
1744 			goto out_err;
1745 
1746 		lower = strchr(lower, '\0') + 1;
1747 	}
1748 
1749 	err = -EINVAL;
1750 	sb->s_stack_depth++;
1751 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1752 		pr_err("maximum fs stacking depth exceeded\n");
1753 		goto out_err;
1754 	}
1755 
1756 	err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1757 	if (err)
1758 		goto out_err;
1759 
1760 	err = -ENOMEM;
1761 	oe = ovl_alloc_entry(numlower);
1762 	if (!oe)
1763 		goto out_err;
1764 
1765 	for (i = 0; i < numlower; i++) {
1766 		oe->lowerstack[i].dentry = dget(stack[i].dentry);
1767 		oe->lowerstack[i].layer = &ofs->layers[i+1];
1768 	}
1769 
1770 out:
1771 	for (i = 0; i < numlower; i++)
1772 		path_put(&stack[i]);
1773 	kfree(stack);
1774 
1775 	return oe;
1776 
1777 out_err:
1778 	oe = ERR_PTR(err);
1779 	goto out;
1780 }
1781 
1782 /*
1783  * Check if this layer root is a descendant of:
1784  * - another layer of this overlayfs instance
1785  * - upper/work dir of any overlayfs instance
1786  */
1787 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1788 			   struct dentry *dentry, const char *name,
1789 			   bool is_lower)
1790 {
1791 	struct dentry *next = dentry, *parent;
1792 	int err = 0;
1793 
1794 	if (!dentry)
1795 		return 0;
1796 
1797 	parent = dget_parent(next);
1798 
1799 	/* Walk back ancestors to root (inclusive) looking for traps */
1800 	while (!err && parent != next) {
1801 		if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1802 			err = -ELOOP;
1803 			pr_err("overlapping %s path\n", name);
1804 		} else if (ovl_is_inuse(parent)) {
1805 			err = ovl_report_in_use(ofs, name);
1806 		}
1807 		next = parent;
1808 		parent = dget_parent(next);
1809 		dput(next);
1810 	}
1811 
1812 	dput(parent);
1813 
1814 	return err;
1815 }
1816 
1817 /*
1818  * Check if any of the layers or work dirs overlap.
1819  */
1820 static int ovl_check_overlapping_layers(struct super_block *sb,
1821 					struct ovl_fs *ofs)
1822 {
1823 	int i, err;
1824 
1825 	if (ovl_upper_mnt(ofs)) {
1826 		err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1827 				      "upperdir", false);
1828 		if (err)
1829 			return err;
1830 
1831 		/*
1832 		 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1833 		 * this instance and covers overlapping work and index dirs,
1834 		 * unless work or index dir have been moved since created inside
1835 		 * workbasedir.  In that case, we already have their traps in
1836 		 * inode cache and we will catch that case on lookup.
1837 		 */
1838 		err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1839 				      false);
1840 		if (err)
1841 			return err;
1842 	}
1843 
1844 	for (i = 1; i < ofs->numlayer; i++) {
1845 		err = ovl_check_layer(sb, ofs,
1846 				      ofs->layers[i].mnt->mnt_root,
1847 				      "lowerdir", true);
1848 		if (err)
1849 			return err;
1850 	}
1851 
1852 	return 0;
1853 }
1854 
1855 static struct dentry *ovl_get_root(struct super_block *sb,
1856 				   struct dentry *upperdentry,
1857 				   struct ovl_entry *oe)
1858 {
1859 	struct dentry *root;
1860 	struct ovl_path *lowerpath = &oe->lowerstack[0];
1861 	unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1862 	int fsid = lowerpath->layer->fsid;
1863 	struct ovl_inode_params oip = {
1864 		.upperdentry = upperdentry,
1865 		.lowerpath = lowerpath,
1866 	};
1867 
1868 	root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1869 	if (!root)
1870 		return NULL;
1871 
1872 	root->d_fsdata = oe;
1873 
1874 	if (upperdentry) {
1875 		/* Root inode uses upper st_ino/i_ino */
1876 		ino = d_inode(upperdentry)->i_ino;
1877 		fsid = 0;
1878 		ovl_dentry_set_upper_alias(root);
1879 		if (ovl_is_impuredir(sb, upperdentry))
1880 			ovl_set_flag(OVL_IMPURE, d_inode(root));
1881 	}
1882 
1883 	/* Root is always merge -> can have whiteouts */
1884 	ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1885 	ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1886 	ovl_set_upperdata(d_inode(root));
1887 	ovl_inode_init(d_inode(root), &oip, ino, fsid);
1888 	ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1889 
1890 	return root;
1891 }
1892 
1893 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1894 {
1895 	struct path upperpath = { };
1896 	struct dentry *root_dentry;
1897 	struct ovl_entry *oe;
1898 	struct ovl_fs *ofs;
1899 	struct ovl_layer *layers;
1900 	struct cred *cred;
1901 	char *splitlower = NULL;
1902 	unsigned int numlower;
1903 	int err;
1904 
1905 	err = -EIO;
1906 	if (WARN_ON(sb->s_user_ns != current_user_ns()))
1907 		goto out;
1908 
1909 	sb->s_d_op = &ovl_dentry_operations;
1910 
1911 	err = -ENOMEM;
1912 	ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1913 	if (!ofs)
1914 		goto out;
1915 
1916 	err = -ENOMEM;
1917 	ofs->creator_cred = cred = prepare_creds();
1918 	if (!cred)
1919 		goto out_err;
1920 
1921 	/* Is there a reason anyone would want not to share whiteouts? */
1922 	ofs->share_whiteout = true;
1923 
1924 	ofs->config.index = ovl_index_def;
1925 	ofs->config.uuid = true;
1926 	ofs->config.nfs_export = ovl_nfs_export_def;
1927 	ofs->config.xino = ovl_xino_def();
1928 	ofs->config.metacopy = ovl_metacopy_def;
1929 	err = ovl_parse_opt((char *) data, &ofs->config);
1930 	if (err)
1931 		goto out_err;
1932 
1933 	err = -EINVAL;
1934 	if (!ofs->config.lowerdir) {
1935 		if (!silent)
1936 			pr_err("missing 'lowerdir'\n");
1937 		goto out_err;
1938 	}
1939 
1940 	err = -ENOMEM;
1941 	splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1942 	if (!splitlower)
1943 		goto out_err;
1944 
1945 	err = -EINVAL;
1946 	numlower = ovl_split_lowerdirs(splitlower);
1947 	if (numlower > OVL_MAX_STACK) {
1948 		pr_err("too many lower directories, limit is %d\n",
1949 		       OVL_MAX_STACK);
1950 		goto out_err;
1951 	}
1952 
1953 	err = -ENOMEM;
1954 	layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1955 	if (!layers)
1956 		goto out_err;
1957 
1958 	ofs->layers = layers;
1959 	/* Layer 0 is reserved for upper even if there's no upper */
1960 	ofs->numlayer = 1;
1961 
1962 	sb->s_stack_depth = 0;
1963 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1964 	atomic_long_set(&ofs->last_ino, 1);
1965 	/* Assume underlying fs uses 32bit inodes unless proven otherwise */
1966 	if (ofs->config.xino != OVL_XINO_OFF) {
1967 		ofs->xino_mode = BITS_PER_LONG - 32;
1968 		if (!ofs->xino_mode) {
1969 			pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1970 			ofs->config.xino = OVL_XINO_OFF;
1971 		}
1972 	}
1973 
1974 	/* alloc/destroy_inode needed for setting up traps in inode cache */
1975 	sb->s_op = &ovl_super_operations;
1976 
1977 	if (ofs->config.upperdir) {
1978 		struct super_block *upper_sb;
1979 
1980 		err = -EINVAL;
1981 		if (!ofs->config.workdir) {
1982 			pr_err("missing 'workdir'\n");
1983 			goto out_err;
1984 		}
1985 
1986 		err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1987 		if (err)
1988 			goto out_err;
1989 
1990 		upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1991 		if (!ovl_should_sync(ofs)) {
1992 			ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1993 			if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1994 				err = -EIO;
1995 				pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1996 				goto out_err;
1997 			}
1998 		}
1999 
2000 		err = ovl_get_workdir(sb, ofs, &upperpath);
2001 		if (err)
2002 			goto out_err;
2003 
2004 		if (!ofs->workdir)
2005 			sb->s_flags |= SB_RDONLY;
2006 
2007 		sb->s_stack_depth = upper_sb->s_stack_depth;
2008 		sb->s_time_gran = upper_sb->s_time_gran;
2009 	}
2010 	oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2011 	err = PTR_ERR(oe);
2012 	if (IS_ERR(oe))
2013 		goto out_err;
2014 
2015 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
2016 	if (!ovl_upper_mnt(ofs))
2017 		sb->s_flags |= SB_RDONLY;
2018 
2019 	if (!ofs->config.uuid && ofs->numfs > 1) {
2020 		pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2021 		ofs->config.uuid = true;
2022 	}
2023 
2024 	if (!ovl_force_readonly(ofs) && ofs->config.index) {
2025 		err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2026 		if (err)
2027 			goto out_free_oe;
2028 
2029 		/* Force r/o mount with no index dir */
2030 		if (!ofs->indexdir)
2031 			sb->s_flags |= SB_RDONLY;
2032 	}
2033 
2034 	err = ovl_check_overlapping_layers(sb, ofs);
2035 	if (err)
2036 		goto out_free_oe;
2037 
2038 	/* Show index=off in /proc/mounts for forced r/o mount */
2039 	if (!ofs->indexdir) {
2040 		ofs->config.index = false;
2041 		if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2042 			pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2043 			ofs->config.nfs_export = false;
2044 		}
2045 	}
2046 
2047 	if (ofs->config.metacopy && ofs->config.nfs_export) {
2048 		pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2049 		ofs->config.nfs_export = false;
2050 	}
2051 
2052 	if (ofs->config.nfs_export)
2053 		sb->s_export_op = &ovl_export_operations;
2054 
2055 	/* Never override disk quota limits or use reserved space */
2056 	cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2057 
2058 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2059 	sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2060 		ovl_trusted_xattr_handlers;
2061 	sb->s_fs_info = ofs;
2062 	sb->s_flags |= SB_POSIXACL;
2063 	sb->s_iflags |= SB_I_SKIP_SYNC;
2064 
2065 	err = -ENOMEM;
2066 	root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2067 	if (!root_dentry)
2068 		goto out_free_oe;
2069 
2070 	mntput(upperpath.mnt);
2071 	kfree(splitlower);
2072 
2073 	sb->s_root = root_dentry;
2074 
2075 	return 0;
2076 
2077 out_free_oe:
2078 	ovl_entry_stack_free(oe);
2079 	kfree(oe);
2080 out_err:
2081 	kfree(splitlower);
2082 	path_put(&upperpath);
2083 	ovl_free_fs(ofs);
2084 out:
2085 	return err;
2086 }
2087 
2088 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2089 				const char *dev_name, void *raw_data)
2090 {
2091 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2092 }
2093 
2094 static struct file_system_type ovl_fs_type = {
2095 	.owner		= THIS_MODULE,
2096 	.name		= "overlay",
2097 	.fs_flags	= FS_USERNS_MOUNT,
2098 	.mount		= ovl_mount,
2099 	.kill_sb	= kill_anon_super,
2100 };
2101 MODULE_ALIAS_FS("overlay");
2102 
2103 static void ovl_inode_init_once(void *foo)
2104 {
2105 	struct ovl_inode *oi = foo;
2106 
2107 	inode_init_once(&oi->vfs_inode);
2108 }
2109 
2110 static int __init ovl_init(void)
2111 {
2112 	int err;
2113 
2114 	ovl_inode_cachep = kmem_cache_create("ovl_inode",
2115 					     sizeof(struct ovl_inode), 0,
2116 					     (SLAB_RECLAIM_ACCOUNT|
2117 					      SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2118 					     ovl_inode_init_once);
2119 	if (ovl_inode_cachep == NULL)
2120 		return -ENOMEM;
2121 
2122 	err = ovl_aio_request_cache_init();
2123 	if (!err) {
2124 		err = register_filesystem(&ovl_fs_type);
2125 		if (!err)
2126 			return 0;
2127 
2128 		ovl_aio_request_cache_destroy();
2129 	}
2130 	kmem_cache_destroy(ovl_inode_cachep);
2131 
2132 	return err;
2133 }
2134 
2135 static void __exit ovl_exit(void)
2136 {
2137 	unregister_filesystem(&ovl_fs_type);
2138 
2139 	/*
2140 	 * Make sure all delayed rcu free inodes are flushed before we
2141 	 * destroy cache.
2142 	 */
2143 	rcu_barrier();
2144 	kmem_cache_destroy(ovl_inode_cachep);
2145 	ovl_aio_request_cache_destroy();
2146 }
2147 
2148 module_init(ovl_init);
2149 module_exit(ovl_exit);
2150