xref: /linux/fs/overlayfs/super.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
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 "overlayfs.h"
19 
20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
21 MODULE_DESCRIPTION("Overlay filesystem");
22 MODULE_LICENSE("GPL");
23 
24 
25 struct ovl_dir_cache;
26 
27 #define OVL_MAX_STACK 500
28 
29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
31 MODULE_PARM_DESC(redirect_dir,
32 		 "Default to on or off for the redirect_dir feature");
33 
34 static bool ovl_redirect_always_follow =
35 	IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
37 		   bool, 0644);
38 MODULE_PARM_DESC(redirect_always_follow,
39 		 "Follow redirects even if redirect_dir feature is turned off");
40 
41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
42 module_param_named(index, ovl_index_def, bool, 0644);
43 MODULE_PARM_DESC(index,
44 		 "Default to on or off for the inodes index feature");
45 
46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
48 MODULE_PARM_DESC(nfs_export,
49 		 "Default to on or off for the NFS export feature");
50 
51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
53 MODULE_PARM_DESC(xino_auto,
54 		 "Auto enable xino feature");
55 
56 static void ovl_entry_stack_free(struct ovl_entry *oe)
57 {
58 	unsigned int i;
59 
60 	for (i = 0; i < oe->numlower; i++)
61 		dput(oe->lowerstack[i].dentry);
62 }
63 
64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
66 MODULE_PARM_DESC(metacopy,
67 		 "Default to on or off for the metadata only copy up feature");
68 
69 static void ovl_dentry_release(struct dentry *dentry)
70 {
71 	struct ovl_entry *oe = dentry->d_fsdata;
72 
73 	if (oe) {
74 		ovl_entry_stack_free(oe);
75 		kfree_rcu(oe, rcu);
76 	}
77 }
78 
79 static struct dentry *ovl_d_real(struct dentry *dentry,
80 				 const struct inode *inode)
81 {
82 	struct dentry *real;
83 
84 	/* It's an overlay file */
85 	if (inode && d_inode(dentry) == inode)
86 		return dentry;
87 
88 	if (!d_is_reg(dentry)) {
89 		if (!inode || inode == d_inode(dentry))
90 			return dentry;
91 		goto bug;
92 	}
93 
94 	real = ovl_dentry_upper(dentry);
95 	if (real && (inode == d_inode(real)))
96 		return real;
97 
98 	if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
99 		return real;
100 
101 	real = ovl_dentry_lowerdata(dentry);
102 	if (!real)
103 		goto bug;
104 
105 	/* Handle recursion */
106 	real = d_real(real, inode);
107 
108 	if (!inode || inode == d_inode(real))
109 		return real;
110 bug:
111 	WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
112 	     inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
113 	return dentry;
114 }
115 
116 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
117 {
118 	struct ovl_entry *oe = dentry->d_fsdata;
119 	unsigned int i;
120 	int ret = 1;
121 
122 	for (i = 0; i < oe->numlower; i++) {
123 		struct dentry *d = oe->lowerstack[i].dentry;
124 
125 		if (d->d_flags & DCACHE_OP_REVALIDATE) {
126 			ret = d->d_op->d_revalidate(d, flags);
127 			if (ret < 0)
128 				return ret;
129 			if (!ret) {
130 				if (!(flags & LOOKUP_RCU))
131 					d_invalidate(d);
132 				return -ESTALE;
133 			}
134 		}
135 	}
136 	return 1;
137 }
138 
139 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
140 {
141 	struct ovl_entry *oe = dentry->d_fsdata;
142 	unsigned int i;
143 	int ret = 1;
144 
145 	for (i = 0; i < oe->numlower; i++) {
146 		struct dentry *d = oe->lowerstack[i].dentry;
147 
148 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
149 			ret = d->d_op->d_weak_revalidate(d, flags);
150 			if (ret <= 0)
151 				break;
152 		}
153 	}
154 	return ret;
155 }
156 
157 static const struct dentry_operations ovl_dentry_operations = {
158 	.d_release = ovl_dentry_release,
159 	.d_real = ovl_d_real,
160 };
161 
162 static const struct dentry_operations ovl_reval_dentry_operations = {
163 	.d_release = ovl_dentry_release,
164 	.d_real = ovl_d_real,
165 	.d_revalidate = ovl_dentry_revalidate,
166 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
167 };
168 
169 static struct kmem_cache *ovl_inode_cachep;
170 
171 static struct inode *ovl_alloc_inode(struct super_block *sb)
172 {
173 	struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
174 
175 	if (!oi)
176 		return NULL;
177 
178 	oi->cache = NULL;
179 	oi->redirect = NULL;
180 	oi->version = 0;
181 	oi->flags = 0;
182 	oi->__upperdentry = NULL;
183 	oi->lower = NULL;
184 	oi->lowerdata = NULL;
185 	mutex_init(&oi->lock);
186 
187 	return &oi->vfs_inode;
188 }
189 
190 static void ovl_free_inode(struct inode *inode)
191 {
192 	struct ovl_inode *oi = OVL_I(inode);
193 
194 	kfree(oi->redirect);
195 	mutex_destroy(&oi->lock);
196 	kmem_cache_free(ovl_inode_cachep, oi);
197 }
198 
199 static void ovl_destroy_inode(struct inode *inode)
200 {
201 	struct ovl_inode *oi = OVL_I(inode);
202 
203 	dput(oi->__upperdentry);
204 	iput(oi->lower);
205 	if (S_ISDIR(inode->i_mode))
206 		ovl_dir_cache_free(inode);
207 	else
208 		iput(oi->lowerdata);
209 }
210 
211 static void ovl_free_fs(struct ovl_fs *ofs)
212 {
213 	unsigned i;
214 
215 	iput(ofs->indexdir_trap);
216 	iput(ofs->workdir_trap);
217 	iput(ofs->upperdir_trap);
218 	dput(ofs->indexdir);
219 	dput(ofs->workdir);
220 	if (ofs->workdir_locked)
221 		ovl_inuse_unlock(ofs->workbasedir);
222 	dput(ofs->workbasedir);
223 	if (ofs->upperdir_locked)
224 		ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
225 	mntput(ofs->upper_mnt);
226 	for (i = 0; i < ofs->numlower; i++) {
227 		iput(ofs->lower_layers[i].trap);
228 		mntput(ofs->lower_layers[i].mnt);
229 	}
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 	.free_inode	= ovl_free_inode,
381 	.destroy_inode	= ovl_destroy_inode,
382 	.drop_inode	= generic_delete_inode,
383 	.put_super	= ovl_put_super,
384 	.sync_fs	= ovl_sync_fs,
385 	.statfs		= ovl_statfs,
386 	.show_options	= ovl_show_options,
387 	.remount_fs	= ovl_remount,
388 };
389 
390 enum {
391 	OPT_LOWERDIR,
392 	OPT_UPPERDIR,
393 	OPT_WORKDIR,
394 	OPT_DEFAULT_PERMISSIONS,
395 	OPT_REDIRECT_DIR,
396 	OPT_INDEX_ON,
397 	OPT_INDEX_OFF,
398 	OPT_NFS_EXPORT_ON,
399 	OPT_NFS_EXPORT_OFF,
400 	OPT_XINO_ON,
401 	OPT_XINO_OFF,
402 	OPT_XINO_AUTO,
403 	OPT_METACOPY_ON,
404 	OPT_METACOPY_OFF,
405 	OPT_ERR,
406 };
407 
408 static const match_table_t ovl_tokens = {
409 	{OPT_LOWERDIR,			"lowerdir=%s"},
410 	{OPT_UPPERDIR,			"upperdir=%s"},
411 	{OPT_WORKDIR,			"workdir=%s"},
412 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
413 	{OPT_REDIRECT_DIR,		"redirect_dir=%s"},
414 	{OPT_INDEX_ON,			"index=on"},
415 	{OPT_INDEX_OFF,			"index=off"},
416 	{OPT_NFS_EXPORT_ON,		"nfs_export=on"},
417 	{OPT_NFS_EXPORT_OFF,		"nfs_export=off"},
418 	{OPT_XINO_ON,			"xino=on"},
419 	{OPT_XINO_OFF,			"xino=off"},
420 	{OPT_XINO_AUTO,			"xino=auto"},
421 	{OPT_METACOPY_ON,		"metacopy=on"},
422 	{OPT_METACOPY_OFF,		"metacopy=off"},
423 	{OPT_ERR,			NULL}
424 };
425 
426 static char *ovl_next_opt(char **s)
427 {
428 	char *sbegin = *s;
429 	char *p;
430 
431 	if (sbegin == NULL)
432 		return NULL;
433 
434 	for (p = sbegin; *p; p++) {
435 		if (*p == '\\') {
436 			p++;
437 			if (!*p)
438 				break;
439 		} else if (*p == ',') {
440 			*p = '\0';
441 			*s = p + 1;
442 			return sbegin;
443 		}
444 	}
445 	*s = NULL;
446 	return sbegin;
447 }
448 
449 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
450 {
451 	if (strcmp(mode, "on") == 0) {
452 		config->redirect_dir = true;
453 		/*
454 		 * Does not make sense to have redirect creation without
455 		 * redirect following.
456 		 */
457 		config->redirect_follow = true;
458 	} else if (strcmp(mode, "follow") == 0) {
459 		config->redirect_follow = true;
460 	} else if (strcmp(mode, "off") == 0) {
461 		if (ovl_redirect_always_follow)
462 			config->redirect_follow = true;
463 	} else if (strcmp(mode, "nofollow") != 0) {
464 		pr_err("overlayfs: bad mount option \"redirect_dir=%s\"\n",
465 		       mode);
466 		return -EINVAL;
467 	}
468 
469 	return 0;
470 }
471 
472 static int ovl_parse_opt(char *opt, struct ovl_config *config)
473 {
474 	char *p;
475 	int err;
476 	bool metacopy_opt = false, redirect_opt = false;
477 
478 	config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
479 	if (!config->redirect_mode)
480 		return -ENOMEM;
481 
482 	while ((p = ovl_next_opt(&opt)) != NULL) {
483 		int token;
484 		substring_t args[MAX_OPT_ARGS];
485 
486 		if (!*p)
487 			continue;
488 
489 		token = match_token(p, ovl_tokens, args);
490 		switch (token) {
491 		case OPT_UPPERDIR:
492 			kfree(config->upperdir);
493 			config->upperdir = match_strdup(&args[0]);
494 			if (!config->upperdir)
495 				return -ENOMEM;
496 			break;
497 
498 		case OPT_LOWERDIR:
499 			kfree(config->lowerdir);
500 			config->lowerdir = match_strdup(&args[0]);
501 			if (!config->lowerdir)
502 				return -ENOMEM;
503 			break;
504 
505 		case OPT_WORKDIR:
506 			kfree(config->workdir);
507 			config->workdir = match_strdup(&args[0]);
508 			if (!config->workdir)
509 				return -ENOMEM;
510 			break;
511 
512 		case OPT_DEFAULT_PERMISSIONS:
513 			config->default_permissions = true;
514 			break;
515 
516 		case OPT_REDIRECT_DIR:
517 			kfree(config->redirect_mode);
518 			config->redirect_mode = match_strdup(&args[0]);
519 			if (!config->redirect_mode)
520 				return -ENOMEM;
521 			redirect_opt = true;
522 			break;
523 
524 		case OPT_INDEX_ON:
525 			config->index = true;
526 			break;
527 
528 		case OPT_INDEX_OFF:
529 			config->index = false;
530 			break;
531 
532 		case OPT_NFS_EXPORT_ON:
533 			config->nfs_export = true;
534 			break;
535 
536 		case OPT_NFS_EXPORT_OFF:
537 			config->nfs_export = false;
538 			break;
539 
540 		case OPT_XINO_ON:
541 			config->xino = OVL_XINO_ON;
542 			break;
543 
544 		case OPT_XINO_OFF:
545 			config->xino = OVL_XINO_OFF;
546 			break;
547 
548 		case OPT_XINO_AUTO:
549 			config->xino = OVL_XINO_AUTO;
550 			break;
551 
552 		case OPT_METACOPY_ON:
553 			config->metacopy = true;
554 			metacopy_opt = true;
555 			break;
556 
557 		case OPT_METACOPY_OFF:
558 			config->metacopy = false;
559 			break;
560 
561 		default:
562 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
563 			return -EINVAL;
564 		}
565 	}
566 
567 	/* Workdir is useless in non-upper mount */
568 	if (!config->upperdir && config->workdir) {
569 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
570 			config->workdir);
571 		kfree(config->workdir);
572 		config->workdir = NULL;
573 	}
574 
575 	err = ovl_parse_redirect_mode(config, config->redirect_mode);
576 	if (err)
577 		return err;
578 
579 	/*
580 	 * This is to make the logic below simpler.  It doesn't make any other
581 	 * difference, since config->redirect_dir is only used for upper.
582 	 */
583 	if (!config->upperdir && config->redirect_follow)
584 		config->redirect_dir = true;
585 
586 	/* Resolve metacopy -> redirect_dir dependency */
587 	if (config->metacopy && !config->redirect_dir) {
588 		if (metacopy_opt && redirect_opt) {
589 			pr_err("overlayfs: conflicting options: metacopy=on,redirect_dir=%s\n",
590 			       config->redirect_mode);
591 			return -EINVAL;
592 		}
593 		if (redirect_opt) {
594 			/*
595 			 * There was an explicit redirect_dir=... that resulted
596 			 * in this conflict.
597 			 */
598 			pr_info("overlayfs: disabling metacopy due to redirect_dir=%s\n",
599 				config->redirect_mode);
600 			config->metacopy = false;
601 		} else {
602 			/* Automatically enable redirect otherwise. */
603 			config->redirect_follow = config->redirect_dir = true;
604 		}
605 	}
606 
607 	return 0;
608 }
609 
610 #define OVL_WORKDIR_NAME "work"
611 #define OVL_INDEXDIR_NAME "index"
612 
613 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
614 					 const char *name, bool persist)
615 {
616 	struct inode *dir =  ofs->workbasedir->d_inode;
617 	struct vfsmount *mnt = ofs->upper_mnt;
618 	struct dentry *work;
619 	int err;
620 	bool retried = false;
621 	bool locked = false;
622 
623 	inode_lock_nested(dir, I_MUTEX_PARENT);
624 	locked = true;
625 
626 retry:
627 	work = lookup_one_len(name, ofs->workbasedir, strlen(name));
628 
629 	if (!IS_ERR(work)) {
630 		struct iattr attr = {
631 			.ia_valid = ATTR_MODE,
632 			.ia_mode = S_IFDIR | 0,
633 		};
634 
635 		if (work->d_inode) {
636 			err = -EEXIST;
637 			if (retried)
638 				goto out_dput;
639 
640 			if (persist)
641 				goto out_unlock;
642 
643 			retried = true;
644 			ovl_workdir_cleanup(dir, mnt, work, 0);
645 			dput(work);
646 			goto retry;
647 		}
648 
649 		work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
650 		err = PTR_ERR(work);
651 		if (IS_ERR(work))
652 			goto out_err;
653 
654 		/*
655 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
656 		 *
657 		 * a) success (there was a POSIX ACL xattr and was removed)
658 		 * b) -ENODATA (there was no POSIX ACL xattr)
659 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
660 		 *
661 		 * There are various other error values that could effectively
662 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
663 		 * if the xattr name is too long), but the set of filesystems
664 		 * allowed as upper are limited to "normal" ones, where checking
665 		 * for the above two errors is sufficient.
666 		 */
667 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
668 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
669 			goto out_dput;
670 
671 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
672 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
673 			goto out_dput;
674 
675 		/* Clear any inherited mode bits */
676 		inode_lock(work->d_inode);
677 		err = notify_change(work, &attr, NULL);
678 		inode_unlock(work->d_inode);
679 		if (err)
680 			goto out_dput;
681 	} else {
682 		err = PTR_ERR(work);
683 		goto out_err;
684 	}
685 out_unlock:
686 	if (locked)
687 		inode_unlock(dir);
688 
689 	return work;
690 
691 out_dput:
692 	dput(work);
693 out_err:
694 	pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
695 		ofs->config.workdir, name, -err);
696 	work = NULL;
697 	goto out_unlock;
698 }
699 
700 static void ovl_unescape(char *s)
701 {
702 	char *d = s;
703 
704 	for (;; s++, d++) {
705 		if (*s == '\\')
706 			s++;
707 		*d = *s;
708 		if (!*s)
709 			break;
710 	}
711 }
712 
713 static int ovl_mount_dir_noesc(const char *name, struct path *path)
714 {
715 	int err = -EINVAL;
716 
717 	if (!*name) {
718 		pr_err("overlayfs: empty lowerdir\n");
719 		goto out;
720 	}
721 	err = kern_path(name, LOOKUP_FOLLOW, path);
722 	if (err) {
723 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
724 		goto out;
725 	}
726 	err = -EINVAL;
727 	if (ovl_dentry_weird(path->dentry)) {
728 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
729 		goto out_put;
730 	}
731 	if (!d_is_dir(path->dentry)) {
732 		pr_err("overlayfs: '%s' not a directory\n", name);
733 		goto out_put;
734 	}
735 	return 0;
736 
737 out_put:
738 	path_put_init(path);
739 out:
740 	return err;
741 }
742 
743 static int ovl_mount_dir(const char *name, struct path *path)
744 {
745 	int err = -ENOMEM;
746 	char *tmp = kstrdup(name, GFP_KERNEL);
747 
748 	if (tmp) {
749 		ovl_unescape(tmp);
750 		err = ovl_mount_dir_noesc(tmp, path);
751 
752 		if (!err)
753 			if (ovl_dentry_remote(path->dentry)) {
754 				pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
755 				       tmp);
756 				path_put_init(path);
757 				err = -EINVAL;
758 			}
759 		kfree(tmp);
760 	}
761 	return err;
762 }
763 
764 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
765 			     const char *name)
766 {
767 	struct kstatfs statfs;
768 	int err = vfs_statfs(path, &statfs);
769 
770 	if (err)
771 		pr_err("overlayfs: statfs failed on '%s'\n", name);
772 	else
773 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
774 
775 	return err;
776 }
777 
778 static int ovl_lower_dir(const char *name, struct path *path,
779 			 struct ovl_fs *ofs, int *stack_depth, bool *remote)
780 {
781 	int fh_type;
782 	int err;
783 
784 	err = ovl_mount_dir_noesc(name, path);
785 	if (err)
786 		goto out;
787 
788 	err = ovl_check_namelen(path, ofs, name);
789 	if (err)
790 		goto out_put;
791 
792 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
793 
794 	if (ovl_dentry_remote(path->dentry))
795 		*remote = true;
796 
797 	/*
798 	 * The inodes index feature and NFS export need to encode and decode
799 	 * file handles, so they require that all layers support them.
800 	 */
801 	fh_type = ovl_can_decode_fh(path->dentry->d_sb);
802 	if ((ofs->config.nfs_export ||
803 	     (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
804 		ofs->config.index = false;
805 		ofs->config.nfs_export = false;
806 		pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
807 			name);
808 	}
809 
810 	/* Check if lower fs has 32bit inode numbers */
811 	if (fh_type != FILEID_INO32_GEN)
812 		ofs->xino_bits = 0;
813 
814 	return 0;
815 
816 out_put:
817 	path_put_init(path);
818 out:
819 	return err;
820 }
821 
822 /* Workdir should not be subdir of upperdir and vice versa */
823 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
824 {
825 	bool ok = false;
826 
827 	if (workdir != upperdir) {
828 		ok = (lock_rename(workdir, upperdir) == NULL);
829 		unlock_rename(workdir, upperdir);
830 	}
831 	return ok;
832 }
833 
834 static unsigned int ovl_split_lowerdirs(char *str)
835 {
836 	unsigned int ctr = 1;
837 	char *s, *d;
838 
839 	for (s = d = str;; s++, d++) {
840 		if (*s == '\\') {
841 			s++;
842 		} else if (*s == ':') {
843 			*d = '\0';
844 			ctr++;
845 			continue;
846 		}
847 		*d = *s;
848 		if (!*s)
849 			break;
850 	}
851 	return ctr;
852 }
853 
854 static int __maybe_unused
855 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
856 			struct dentry *dentry, struct inode *inode,
857 			const char *name, void *buffer, size_t size)
858 {
859 	return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
860 }
861 
862 static int __maybe_unused
863 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
864 			struct dentry *dentry, struct inode *inode,
865 			const char *name, const void *value,
866 			size_t size, int flags)
867 {
868 	struct dentry *workdir = ovl_workdir(dentry);
869 	struct inode *realinode = ovl_inode_real(inode);
870 	struct posix_acl *acl = NULL;
871 	int err;
872 
873 	/* Check that everything is OK before copy-up */
874 	if (value) {
875 		acl = posix_acl_from_xattr(&init_user_ns, value, size);
876 		if (IS_ERR(acl))
877 			return PTR_ERR(acl);
878 	}
879 	err = -EOPNOTSUPP;
880 	if (!IS_POSIXACL(d_inode(workdir)))
881 		goto out_acl_release;
882 	if (!realinode->i_op->set_acl)
883 		goto out_acl_release;
884 	if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
885 		err = acl ? -EACCES : 0;
886 		goto out_acl_release;
887 	}
888 	err = -EPERM;
889 	if (!inode_owner_or_capable(inode))
890 		goto out_acl_release;
891 
892 	posix_acl_release(acl);
893 
894 	/*
895 	 * Check if sgid bit needs to be cleared (actual setacl operation will
896 	 * be done with mounter's capabilities and so that won't do it for us).
897 	 */
898 	if (unlikely(inode->i_mode & S_ISGID) &&
899 	    handler->flags == ACL_TYPE_ACCESS &&
900 	    !in_group_p(inode->i_gid) &&
901 	    !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
902 		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
903 
904 		err = ovl_setattr(dentry, &iattr);
905 		if (err)
906 			return err;
907 	}
908 
909 	err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
910 	if (!err)
911 		ovl_copyattr(ovl_inode_real(inode), inode);
912 
913 	return err;
914 
915 out_acl_release:
916 	posix_acl_release(acl);
917 	return err;
918 }
919 
920 static int ovl_own_xattr_get(const struct xattr_handler *handler,
921 			     struct dentry *dentry, struct inode *inode,
922 			     const char *name, void *buffer, size_t size)
923 {
924 	return -EOPNOTSUPP;
925 }
926 
927 static int ovl_own_xattr_set(const struct xattr_handler *handler,
928 			     struct dentry *dentry, struct inode *inode,
929 			     const char *name, const void *value,
930 			     size_t size, int flags)
931 {
932 	return -EOPNOTSUPP;
933 }
934 
935 static int ovl_other_xattr_get(const struct xattr_handler *handler,
936 			       struct dentry *dentry, struct inode *inode,
937 			       const char *name, void *buffer, size_t size)
938 {
939 	return ovl_xattr_get(dentry, inode, name, buffer, size);
940 }
941 
942 static int ovl_other_xattr_set(const struct xattr_handler *handler,
943 			       struct dentry *dentry, struct inode *inode,
944 			       const char *name, const void *value,
945 			       size_t size, int flags)
946 {
947 	return ovl_xattr_set(dentry, inode, name, value, size, flags);
948 }
949 
950 static const struct xattr_handler __maybe_unused
951 ovl_posix_acl_access_xattr_handler = {
952 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
953 	.flags = ACL_TYPE_ACCESS,
954 	.get = ovl_posix_acl_xattr_get,
955 	.set = ovl_posix_acl_xattr_set,
956 };
957 
958 static const struct xattr_handler __maybe_unused
959 ovl_posix_acl_default_xattr_handler = {
960 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
961 	.flags = ACL_TYPE_DEFAULT,
962 	.get = ovl_posix_acl_xattr_get,
963 	.set = ovl_posix_acl_xattr_set,
964 };
965 
966 static const struct xattr_handler ovl_own_xattr_handler = {
967 	.prefix	= OVL_XATTR_PREFIX,
968 	.get = ovl_own_xattr_get,
969 	.set = ovl_own_xattr_set,
970 };
971 
972 static const struct xattr_handler ovl_other_xattr_handler = {
973 	.prefix	= "", /* catch all */
974 	.get = ovl_other_xattr_get,
975 	.set = ovl_other_xattr_set,
976 };
977 
978 static const struct xattr_handler *ovl_xattr_handlers[] = {
979 #ifdef CONFIG_FS_POSIX_ACL
980 	&ovl_posix_acl_access_xattr_handler,
981 	&ovl_posix_acl_default_xattr_handler,
982 #endif
983 	&ovl_own_xattr_handler,
984 	&ovl_other_xattr_handler,
985 	NULL
986 };
987 
988 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
989 			  struct inode **ptrap, const char *name)
990 {
991 	struct inode *trap;
992 	int err;
993 
994 	trap = ovl_get_trap_inode(sb, dir);
995 	err = PTR_ERR_OR_ZERO(trap);
996 	if (err) {
997 		if (err == -ELOOP)
998 			pr_err("overlayfs: conflicting %s path\n", name);
999 		return err;
1000 	}
1001 
1002 	*ptrap = trap;
1003 	return 0;
1004 }
1005 
1006 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1007 			 struct path *upperpath)
1008 {
1009 	struct vfsmount *upper_mnt;
1010 	int err;
1011 
1012 	err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1013 	if (err)
1014 		goto out;
1015 
1016 	/* Upper fs should not be r/o */
1017 	if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1018 		pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
1019 		err = -EINVAL;
1020 		goto out;
1021 	}
1022 
1023 	err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1024 	if (err)
1025 		goto out;
1026 
1027 	err = ovl_setup_trap(sb, upperpath->dentry, &ofs->upperdir_trap,
1028 			     "upperdir");
1029 	if (err)
1030 		goto out;
1031 
1032 	upper_mnt = clone_private_mount(upperpath);
1033 	err = PTR_ERR(upper_mnt);
1034 	if (IS_ERR(upper_mnt)) {
1035 		pr_err("overlayfs: failed to clone upperpath\n");
1036 		goto out;
1037 	}
1038 
1039 	/* Don't inherit atime flags */
1040 	upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1041 	ofs->upper_mnt = upper_mnt;
1042 
1043 	err = -EBUSY;
1044 	if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1045 		ofs->upperdir_locked = true;
1046 	} else if (ofs->config.index) {
1047 		pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
1048 		goto out;
1049 	} else {
1050 		pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1051 	}
1052 
1053 	err = 0;
1054 out:
1055 	return err;
1056 }
1057 
1058 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1059 			    struct path *workpath)
1060 {
1061 	struct vfsmount *mnt = ofs->upper_mnt;
1062 	struct dentry *temp;
1063 	int fh_type;
1064 	int err;
1065 
1066 	err = mnt_want_write(mnt);
1067 	if (err)
1068 		return err;
1069 
1070 	ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1071 	if (!ofs->workdir)
1072 		goto out;
1073 
1074 	err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1075 	if (err)
1076 		goto out;
1077 
1078 	/*
1079 	 * Upper should support d_type, else whiteouts are visible.  Given
1080 	 * workdir and upper are on same fs, we can do iterate_dir() on
1081 	 * workdir. This check requires successful creation of workdir in
1082 	 * previous step.
1083 	 */
1084 	err = ovl_check_d_type_supported(workpath);
1085 	if (err < 0)
1086 		goto out;
1087 
1088 	/*
1089 	 * We allowed this configuration and don't want to break users over
1090 	 * kernel upgrade. So warn instead of erroring out.
1091 	 */
1092 	if (!err)
1093 		pr_warn("overlayfs: upper fs needs to support d_type.\n");
1094 
1095 	/* Check if upper/work fs supports O_TMPFILE */
1096 	temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1097 	ofs->tmpfile = !IS_ERR(temp);
1098 	if (ofs->tmpfile)
1099 		dput(temp);
1100 	else
1101 		pr_warn("overlayfs: upper fs does not support tmpfile.\n");
1102 
1103 	/*
1104 	 * Check if upper/work fs supports trusted.overlay.* xattr
1105 	 */
1106 	err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1107 	if (err) {
1108 		ofs->noxattr = true;
1109 		ofs->config.index = false;
1110 		ofs->config.metacopy = false;
1111 		pr_warn("overlayfs: upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1112 		err = 0;
1113 	} else {
1114 		vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1115 	}
1116 
1117 	/* Check if upper/work fs supports file handles */
1118 	fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1119 	if (ofs->config.index && !fh_type) {
1120 		ofs->config.index = false;
1121 		pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1122 	}
1123 
1124 	/* Check if upper fs has 32bit inode numbers */
1125 	if (fh_type != FILEID_INO32_GEN)
1126 		ofs->xino_bits = 0;
1127 
1128 	/* NFS export of r/w mount depends on index */
1129 	if (ofs->config.nfs_export && !ofs->config.index) {
1130 		pr_warn("overlayfs: NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1131 		ofs->config.nfs_export = false;
1132 	}
1133 out:
1134 	mnt_drop_write(mnt);
1135 	return err;
1136 }
1137 
1138 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1139 			   struct path *upperpath)
1140 {
1141 	int err;
1142 	struct path workpath = { };
1143 
1144 	err = ovl_mount_dir(ofs->config.workdir, &workpath);
1145 	if (err)
1146 		goto out;
1147 
1148 	err = -EINVAL;
1149 	if (upperpath->mnt != workpath.mnt) {
1150 		pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1151 		goto out;
1152 	}
1153 	if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1154 		pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1155 		goto out;
1156 	}
1157 
1158 	ofs->workbasedir = dget(workpath.dentry);
1159 
1160 	err = -EBUSY;
1161 	if (ovl_inuse_trylock(ofs->workbasedir)) {
1162 		ofs->workdir_locked = true;
1163 	} else if (ofs->config.index) {
1164 		pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
1165 		goto out;
1166 	} else {
1167 		pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
1168 	}
1169 
1170 	err = ovl_make_workdir(sb, ofs, &workpath);
1171 
1172 out:
1173 	path_put(&workpath);
1174 
1175 	return err;
1176 }
1177 
1178 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1179 			    struct ovl_entry *oe, struct path *upperpath)
1180 {
1181 	struct vfsmount *mnt = ofs->upper_mnt;
1182 	int err;
1183 
1184 	err = mnt_want_write(mnt);
1185 	if (err)
1186 		return err;
1187 
1188 	/* Verify lower root is upper root origin */
1189 	err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1190 				true);
1191 	if (err) {
1192 		pr_err("overlayfs: failed to verify upper root origin\n");
1193 		goto out;
1194 	}
1195 
1196 	ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1197 	if (ofs->indexdir) {
1198 		err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1199 				     "indexdir");
1200 		if (err)
1201 			goto out;
1202 
1203 		/*
1204 		 * Verify upper root is exclusively associated with index dir.
1205 		 * Older kernels stored upper fh in "trusted.overlay.origin"
1206 		 * xattr. If that xattr exists, verify that it is a match to
1207 		 * upper dir file handle. In any case, verify or set xattr
1208 		 * "trusted.overlay.upper" to indicate that index may have
1209 		 * directory entries.
1210 		 */
1211 		if (ovl_check_origin_xattr(ofs->indexdir)) {
1212 			err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1213 						upperpath->dentry, true, false);
1214 			if (err)
1215 				pr_err("overlayfs: failed to verify index dir 'origin' xattr\n");
1216 		}
1217 		err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1218 		if (err)
1219 			pr_err("overlayfs: failed to verify index dir 'upper' xattr\n");
1220 
1221 		/* Cleanup bad/stale/orphan index entries */
1222 		if (!err)
1223 			err = ovl_indexdir_cleanup(ofs);
1224 	}
1225 	if (err || !ofs->indexdir)
1226 		pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1227 
1228 out:
1229 	mnt_drop_write(mnt);
1230 	return err;
1231 }
1232 
1233 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1234 {
1235 	unsigned int i;
1236 
1237 	if (!ofs->config.nfs_export && !(ofs->config.index && ofs->upper_mnt))
1238 		return true;
1239 
1240 	for (i = 0; i < ofs->numlowerfs; i++) {
1241 		/*
1242 		 * We use uuid to associate an overlay lower file handle with a
1243 		 * lower layer, so we can accept lower fs with null uuid as long
1244 		 * as all lower layers with null uuid are on the same fs.
1245 		 */
1246 		if (uuid_equal(&ofs->lower_fs[i].sb->s_uuid, uuid))
1247 			return false;
1248 	}
1249 	return true;
1250 }
1251 
1252 /* Get a unique fsid for the layer */
1253 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1254 {
1255 	struct super_block *sb = path->mnt->mnt_sb;
1256 	unsigned int i;
1257 	dev_t dev;
1258 	int err;
1259 
1260 	/* fsid 0 is reserved for upper fs even with non upper overlay */
1261 	if (ofs->upper_mnt && ofs->upper_mnt->mnt_sb == sb)
1262 		return 0;
1263 
1264 	for (i = 0; i < ofs->numlowerfs; i++) {
1265 		if (ofs->lower_fs[i].sb == sb)
1266 			return i + 1;
1267 	}
1268 
1269 	if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1270 		ofs->config.index = false;
1271 		ofs->config.nfs_export = false;
1272 		pr_warn("overlayfs: %s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1273 			uuid_is_null(&sb->s_uuid) ? "null" : "conflicting",
1274 			path->dentry);
1275 	}
1276 
1277 	err = get_anon_bdev(&dev);
1278 	if (err) {
1279 		pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1280 		return err;
1281 	}
1282 
1283 	ofs->lower_fs[ofs->numlowerfs].sb = sb;
1284 	ofs->lower_fs[ofs->numlowerfs].pseudo_dev = dev;
1285 	ofs->numlowerfs++;
1286 
1287 	return ofs->numlowerfs;
1288 }
1289 
1290 static int ovl_get_lower_layers(struct super_block *sb, struct ovl_fs *ofs,
1291 				struct path *stack, unsigned int numlower)
1292 {
1293 	int err;
1294 	unsigned int i;
1295 
1296 	err = -ENOMEM;
1297 	ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1298 				    GFP_KERNEL);
1299 	if (ofs->lower_layers == NULL)
1300 		goto out;
1301 
1302 	ofs->lower_fs = kcalloc(numlower, sizeof(struct ovl_sb),
1303 				GFP_KERNEL);
1304 	if (ofs->lower_fs == NULL)
1305 		goto out;
1306 
1307 	for (i = 0; i < numlower; i++) {
1308 		struct vfsmount *mnt;
1309 		struct inode *trap;
1310 		int fsid;
1311 
1312 		err = fsid = ovl_get_fsid(ofs, &stack[i]);
1313 		if (err < 0)
1314 			goto out;
1315 
1316 		err = -EBUSY;
1317 		if (ovl_is_inuse(stack[i].dentry)) {
1318 			pr_err("overlayfs: lowerdir is in-use as upperdir/workdir\n");
1319 			goto out;
1320 		}
1321 
1322 		err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1323 		if (err)
1324 			goto out;
1325 
1326 		mnt = clone_private_mount(&stack[i]);
1327 		err = PTR_ERR(mnt);
1328 		if (IS_ERR(mnt)) {
1329 			pr_err("overlayfs: failed to clone lowerpath\n");
1330 			iput(trap);
1331 			goto out;
1332 		}
1333 
1334 		/*
1335 		 * Make lower layers R/O.  That way fchmod/fchown on lower file
1336 		 * will fail instead of modifying lower fs.
1337 		 */
1338 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1339 
1340 		ofs->lower_layers[ofs->numlower].trap = trap;
1341 		ofs->lower_layers[ofs->numlower].mnt = mnt;
1342 		ofs->lower_layers[ofs->numlower].idx = i + 1;
1343 		ofs->lower_layers[ofs->numlower].fsid = fsid;
1344 		if (fsid) {
1345 			ofs->lower_layers[ofs->numlower].fs =
1346 				&ofs->lower_fs[fsid - 1];
1347 		}
1348 		ofs->numlower++;
1349 	}
1350 
1351 	/*
1352 	 * When all layers on same fs, overlay can use real inode numbers.
1353 	 * With mount option "xino=on", mounter declares that there are enough
1354 	 * free high bits in underlying fs to hold the unique fsid.
1355 	 * If overlayfs does encounter underlying inodes using the high xino
1356 	 * bits reserved for fsid, it emits a warning and uses the original
1357 	 * inode number.
1358 	 */
1359 	if (!ofs->numlowerfs || (ofs->numlowerfs == 1 && !ofs->upper_mnt)) {
1360 		ofs->xino_bits = 0;
1361 		ofs->config.xino = OVL_XINO_OFF;
1362 	} else if (ofs->config.xino == OVL_XINO_ON && !ofs->xino_bits) {
1363 		/*
1364 		 * This is a roundup of number of bits needed for numlowerfs+1
1365 		 * (i.e. ilog2(numlowerfs+1 - 1) + 1). fsid 0 is reserved for
1366 		 * upper fs even with non upper overlay.
1367 		 */
1368 		BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1369 		ofs->xino_bits = ilog2(ofs->numlowerfs) + 1;
1370 	}
1371 
1372 	if (ofs->xino_bits) {
1373 		pr_info("overlayfs: \"xino\" feature enabled using %d upper inode bits.\n",
1374 			ofs->xino_bits);
1375 	}
1376 
1377 	err = 0;
1378 out:
1379 	return err;
1380 }
1381 
1382 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1383 					    struct ovl_fs *ofs)
1384 {
1385 	int err;
1386 	char *lowertmp, *lower;
1387 	struct path *stack = NULL;
1388 	unsigned int stacklen, numlower = 0, i;
1389 	bool remote = false;
1390 	struct ovl_entry *oe;
1391 
1392 	err = -ENOMEM;
1393 	lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1394 	if (!lowertmp)
1395 		goto out_err;
1396 
1397 	err = -EINVAL;
1398 	stacklen = ovl_split_lowerdirs(lowertmp);
1399 	if (stacklen > OVL_MAX_STACK) {
1400 		pr_err("overlayfs: too many lower directories, limit is %d\n",
1401 		       OVL_MAX_STACK);
1402 		goto out_err;
1403 	} else if (!ofs->config.upperdir && stacklen == 1) {
1404 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1405 		goto out_err;
1406 	} else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1407 		   ofs->config.redirect_follow) {
1408 		pr_warn("overlayfs: NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1409 		ofs->config.nfs_export = false;
1410 	}
1411 
1412 	err = -ENOMEM;
1413 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1414 	if (!stack)
1415 		goto out_err;
1416 
1417 	err = -EINVAL;
1418 	lower = lowertmp;
1419 	for (numlower = 0; numlower < stacklen; numlower++) {
1420 		err = ovl_lower_dir(lower, &stack[numlower], ofs,
1421 				    &sb->s_stack_depth, &remote);
1422 		if (err)
1423 			goto out_err;
1424 
1425 		lower = strchr(lower, '\0') + 1;
1426 	}
1427 
1428 	err = -EINVAL;
1429 	sb->s_stack_depth++;
1430 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1431 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1432 		goto out_err;
1433 	}
1434 
1435 	err = ovl_get_lower_layers(sb, ofs, stack, numlower);
1436 	if (err)
1437 		goto out_err;
1438 
1439 	err = -ENOMEM;
1440 	oe = ovl_alloc_entry(numlower);
1441 	if (!oe)
1442 		goto out_err;
1443 
1444 	for (i = 0; i < numlower; i++) {
1445 		oe->lowerstack[i].dentry = dget(stack[i].dentry);
1446 		oe->lowerstack[i].layer = &ofs->lower_layers[i];
1447 	}
1448 
1449 	if (remote)
1450 		sb->s_d_op = &ovl_reval_dentry_operations;
1451 	else
1452 		sb->s_d_op = &ovl_dentry_operations;
1453 
1454 out:
1455 	for (i = 0; i < numlower; i++)
1456 		path_put(&stack[i]);
1457 	kfree(stack);
1458 	kfree(lowertmp);
1459 
1460 	return oe;
1461 
1462 out_err:
1463 	oe = ERR_PTR(err);
1464 	goto out;
1465 }
1466 
1467 /*
1468  * Check if this layer root is a descendant of:
1469  * - another layer of this overlayfs instance
1470  * - upper/work dir of any overlayfs instance
1471  */
1472 static int ovl_check_layer(struct super_block *sb, struct dentry *dentry,
1473 			   const char *name)
1474 {
1475 	struct dentry *next = dentry, *parent;
1476 	int err = 0;
1477 
1478 	if (!dentry)
1479 		return 0;
1480 
1481 	parent = dget_parent(next);
1482 
1483 	/* Walk back ancestors to root (inclusive) looking for traps */
1484 	while (!err && parent != next) {
1485 		if (ovl_is_inuse(parent)) {
1486 			err = -EBUSY;
1487 			pr_err("overlayfs: %s path overlapping in-use upperdir/workdir\n",
1488 			       name);
1489 		} else if (ovl_lookup_trap_inode(sb, parent)) {
1490 			err = -ELOOP;
1491 			pr_err("overlayfs: overlapping %s path\n", name);
1492 		}
1493 		next = parent;
1494 		parent = dget_parent(next);
1495 		dput(next);
1496 	}
1497 
1498 	dput(parent);
1499 
1500 	return err;
1501 }
1502 
1503 /*
1504  * Check if any of the layers or work dirs overlap.
1505  */
1506 static int ovl_check_overlapping_layers(struct super_block *sb,
1507 					struct ovl_fs *ofs)
1508 {
1509 	int i, err;
1510 
1511 	if (ofs->upper_mnt) {
1512 		err = ovl_check_layer(sb, ofs->upper_mnt->mnt_root, "upperdir");
1513 		if (err)
1514 			return err;
1515 
1516 		/*
1517 		 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1518 		 * this instance and covers overlapping work and index dirs,
1519 		 * unless work or index dir have been moved since created inside
1520 		 * workbasedir.  In that case, we already have their traps in
1521 		 * inode cache and we will catch that case on lookup.
1522 		 */
1523 		err = ovl_check_layer(sb, ofs->workbasedir, "workdir");
1524 		if (err)
1525 			return err;
1526 	}
1527 
1528 	for (i = 0; i < ofs->numlower; i++) {
1529 		err = ovl_check_layer(sb, ofs->lower_layers[i].mnt->mnt_root,
1530 				      "lowerdir");
1531 		if (err)
1532 			return err;
1533 	}
1534 
1535 	return 0;
1536 }
1537 
1538 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1539 {
1540 	struct path upperpath = { };
1541 	struct dentry *root_dentry;
1542 	struct ovl_entry *oe;
1543 	struct ovl_fs *ofs;
1544 	struct cred *cred;
1545 	int err;
1546 
1547 	err = -ENOMEM;
1548 	ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1549 	if (!ofs)
1550 		goto out;
1551 
1552 	ofs->creator_cred = cred = prepare_creds();
1553 	if (!cred)
1554 		goto out_err;
1555 
1556 	ofs->config.index = ovl_index_def;
1557 	ofs->config.nfs_export = ovl_nfs_export_def;
1558 	ofs->config.xino = ovl_xino_def();
1559 	ofs->config.metacopy = ovl_metacopy_def;
1560 	err = ovl_parse_opt((char *) data, &ofs->config);
1561 	if (err)
1562 		goto out_err;
1563 
1564 	err = -EINVAL;
1565 	if (!ofs->config.lowerdir) {
1566 		if (!silent)
1567 			pr_err("overlayfs: missing 'lowerdir'\n");
1568 		goto out_err;
1569 	}
1570 
1571 	sb->s_stack_depth = 0;
1572 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1573 	/* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1574 	if (ofs->config.xino != OVL_XINO_OFF)
1575 		ofs->xino_bits = BITS_PER_LONG - 32;
1576 
1577 	/* alloc/destroy_inode needed for setting up traps in inode cache */
1578 	sb->s_op = &ovl_super_operations;
1579 
1580 	if (ofs->config.upperdir) {
1581 		if (!ofs->config.workdir) {
1582 			pr_err("overlayfs: missing 'workdir'\n");
1583 			goto out_err;
1584 		}
1585 
1586 		err = ovl_get_upper(sb, ofs, &upperpath);
1587 		if (err)
1588 			goto out_err;
1589 
1590 		err = ovl_get_workdir(sb, ofs, &upperpath);
1591 		if (err)
1592 			goto out_err;
1593 
1594 		if (!ofs->workdir)
1595 			sb->s_flags |= SB_RDONLY;
1596 
1597 		sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1598 		sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1599 
1600 	}
1601 	oe = ovl_get_lowerstack(sb, ofs);
1602 	err = PTR_ERR(oe);
1603 	if (IS_ERR(oe))
1604 		goto out_err;
1605 
1606 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
1607 	if (!ofs->upper_mnt)
1608 		sb->s_flags |= SB_RDONLY;
1609 
1610 	if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1611 		err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1612 		if (err)
1613 			goto out_free_oe;
1614 
1615 		/* Force r/o mount with no index dir */
1616 		if (!ofs->indexdir) {
1617 			dput(ofs->workdir);
1618 			ofs->workdir = NULL;
1619 			sb->s_flags |= SB_RDONLY;
1620 		}
1621 
1622 	}
1623 
1624 	err = ovl_check_overlapping_layers(sb, ofs);
1625 	if (err)
1626 		goto out_free_oe;
1627 
1628 	/* Show index=off in /proc/mounts for forced r/o mount */
1629 	if (!ofs->indexdir) {
1630 		ofs->config.index = false;
1631 		if (ofs->upper_mnt && ofs->config.nfs_export) {
1632 			pr_warn("overlayfs: NFS export requires an index dir, falling back to nfs_export=off.\n");
1633 			ofs->config.nfs_export = false;
1634 		}
1635 	}
1636 
1637 	if (ofs->config.metacopy && ofs->config.nfs_export) {
1638 		pr_warn("overlayfs: NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1639 		ofs->config.nfs_export = false;
1640 	}
1641 
1642 	if (ofs->config.nfs_export)
1643 		sb->s_export_op = &ovl_export_operations;
1644 
1645 	/* Never override disk quota limits or use reserved space */
1646 	cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1647 
1648 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1649 	sb->s_xattr = ovl_xattr_handlers;
1650 	sb->s_fs_info = ofs;
1651 	sb->s_flags |= SB_POSIXACL;
1652 
1653 	err = -ENOMEM;
1654 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1655 	if (!root_dentry)
1656 		goto out_free_oe;
1657 
1658 	root_dentry->d_fsdata = oe;
1659 
1660 	mntput(upperpath.mnt);
1661 	if (upperpath.dentry) {
1662 		ovl_dentry_set_upper_alias(root_dentry);
1663 		if (ovl_is_impuredir(upperpath.dentry))
1664 			ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1665 	}
1666 
1667 	/* Root is always merge -> can have whiteouts */
1668 	ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1669 	ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1670 	ovl_set_upperdata(d_inode(root_dentry));
1671 	ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1672 		       ovl_dentry_lower(root_dentry), NULL);
1673 
1674 	sb->s_root = root_dentry;
1675 
1676 	return 0;
1677 
1678 out_free_oe:
1679 	ovl_entry_stack_free(oe);
1680 	kfree(oe);
1681 out_err:
1682 	path_put(&upperpath);
1683 	ovl_free_fs(ofs);
1684 out:
1685 	return err;
1686 }
1687 
1688 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1689 				const char *dev_name, void *raw_data)
1690 {
1691 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1692 }
1693 
1694 static struct file_system_type ovl_fs_type = {
1695 	.owner		= THIS_MODULE,
1696 	.name		= "overlay",
1697 	.mount		= ovl_mount,
1698 	.kill_sb	= kill_anon_super,
1699 };
1700 MODULE_ALIAS_FS("overlay");
1701 
1702 static void ovl_inode_init_once(void *foo)
1703 {
1704 	struct ovl_inode *oi = foo;
1705 
1706 	inode_init_once(&oi->vfs_inode);
1707 }
1708 
1709 static int __init ovl_init(void)
1710 {
1711 	int err;
1712 
1713 	ovl_inode_cachep = kmem_cache_create("ovl_inode",
1714 					     sizeof(struct ovl_inode), 0,
1715 					     (SLAB_RECLAIM_ACCOUNT|
1716 					      SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1717 					     ovl_inode_init_once);
1718 	if (ovl_inode_cachep == NULL)
1719 		return -ENOMEM;
1720 
1721 	err = register_filesystem(&ovl_fs_type);
1722 	if (err)
1723 		kmem_cache_destroy(ovl_inode_cachep);
1724 
1725 	return err;
1726 }
1727 
1728 static void __exit ovl_exit(void)
1729 {
1730 	unregister_filesystem(&ovl_fs_type);
1731 
1732 	/*
1733 	 * Make sure all delayed rcu free inodes are flushed before we
1734 	 * destroy cache.
1735 	 */
1736 	rcu_barrier();
1737 	kmem_cache_destroy(ovl_inode_cachep);
1738 
1739 }
1740 
1741 module_init(ovl_init);
1742 module_exit(ovl_exit);
1743