xref: /linux/fs/overlayfs/super.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  *
3  * Copyright (C) 2011 Novell Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published by
7  * the Free Software Foundation.
8  */
9 
10 #include <uapi/linux/magic.h>
11 #include <linux/fs.h>
12 #include <linux/namei.h>
13 #include <linux/xattr.h>
14 #include <linux/mount.h>
15 #include <linux/parser.h>
16 #include <linux/module.h>
17 #include <linux/statfs.h>
18 #include <linux/seq_file.h>
19 #include <linux/posix_acl_xattr.h>
20 #include "overlayfs.h"
21 
22 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
23 MODULE_DESCRIPTION("Overlay filesystem");
24 MODULE_LICENSE("GPL");
25 
26 
27 struct ovl_dir_cache;
28 
29 #define OVL_MAX_STACK 500
30 
31 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
32 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
33 MODULE_PARM_DESC(ovl_redirect_dir_def,
34 		 "Default to on or off for the redirect_dir feature");
35 
36 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
37 module_param_named(index, ovl_index_def, bool, 0644);
38 MODULE_PARM_DESC(ovl_index_def,
39 		 "Default to on or off for the inodes index feature");
40 
41 static void ovl_entry_stack_free(struct ovl_entry *oe)
42 {
43 	unsigned int i;
44 
45 	for (i = 0; i < oe->numlower; i++)
46 		dput(oe->lowerstack[i].dentry);
47 }
48 
49 static void ovl_dentry_release(struct dentry *dentry)
50 {
51 	struct ovl_entry *oe = dentry->d_fsdata;
52 
53 	if (oe) {
54 		ovl_entry_stack_free(oe);
55 		kfree_rcu(oe, rcu);
56 	}
57 }
58 
59 static int ovl_check_append_only(struct inode *inode, int flag)
60 {
61 	/*
62 	 * This test was moot in vfs may_open() because overlay inode does
63 	 * not have the S_APPEND flag, so re-check on real upper inode
64 	 */
65 	if (IS_APPEND(inode)) {
66 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
67 			return -EPERM;
68 		if (flag & O_TRUNC)
69 			return -EPERM;
70 	}
71 
72 	return 0;
73 }
74 
75 static struct dentry *ovl_d_real(struct dentry *dentry,
76 				 const struct inode *inode,
77 				 unsigned int open_flags, unsigned int flags)
78 {
79 	struct dentry *real;
80 	int err;
81 
82 	if (flags & D_REAL_UPPER)
83 		return ovl_dentry_upper(dentry);
84 
85 	if (!d_is_reg(dentry)) {
86 		if (!inode || inode == d_inode(dentry))
87 			return dentry;
88 		goto bug;
89 	}
90 
91 	if (open_flags) {
92 		err = ovl_open_maybe_copy_up(dentry, open_flags);
93 		if (err)
94 			return ERR_PTR(err);
95 	}
96 
97 	real = ovl_dentry_upper(dentry);
98 	if (real && (!inode || inode == d_inode(real))) {
99 		if (!inode) {
100 			err = ovl_check_append_only(d_inode(real), open_flags);
101 			if (err)
102 				return ERR_PTR(err);
103 		}
104 		return real;
105 	}
106 
107 	real = ovl_dentry_lower(dentry);
108 	if (!real)
109 		goto bug;
110 
111 	/* Handle recursion */
112 	real = d_real(real, inode, open_flags, 0);
113 
114 	if (!inode || inode == d_inode(real))
115 		return real;
116 bug:
117 	WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
118 	     inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
119 	return dentry;
120 }
121 
122 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
123 {
124 	struct ovl_entry *oe = dentry->d_fsdata;
125 	unsigned int i;
126 	int ret = 1;
127 
128 	for (i = 0; i < oe->numlower; i++) {
129 		struct dentry *d = oe->lowerstack[i].dentry;
130 
131 		if (d->d_flags & DCACHE_OP_REVALIDATE) {
132 			ret = d->d_op->d_revalidate(d, flags);
133 			if (ret < 0)
134 				return ret;
135 			if (!ret) {
136 				if (!(flags & LOOKUP_RCU))
137 					d_invalidate(d);
138 				return -ESTALE;
139 			}
140 		}
141 	}
142 	return 1;
143 }
144 
145 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
146 {
147 	struct ovl_entry *oe = dentry->d_fsdata;
148 	unsigned int i;
149 	int ret = 1;
150 
151 	for (i = 0; i < oe->numlower; i++) {
152 		struct dentry *d = oe->lowerstack[i].dentry;
153 
154 		if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
155 			ret = d->d_op->d_weak_revalidate(d, flags);
156 			if (ret <= 0)
157 				break;
158 		}
159 	}
160 	return ret;
161 }
162 
163 static const struct dentry_operations ovl_dentry_operations = {
164 	.d_release = ovl_dentry_release,
165 	.d_real = ovl_d_real,
166 };
167 
168 static const struct dentry_operations ovl_reval_dentry_operations = {
169 	.d_release = ovl_dentry_release,
170 	.d_real = ovl_d_real,
171 	.d_revalidate = ovl_dentry_revalidate,
172 	.d_weak_revalidate = ovl_dentry_weak_revalidate,
173 };
174 
175 static struct kmem_cache *ovl_inode_cachep;
176 
177 static struct inode *ovl_alloc_inode(struct super_block *sb)
178 {
179 	struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
180 
181 	if (!oi)
182 		return NULL;
183 
184 	oi->cache = NULL;
185 	oi->redirect = NULL;
186 	oi->version = 0;
187 	oi->flags = 0;
188 	oi->__upperdentry = NULL;
189 	oi->lower = NULL;
190 	mutex_init(&oi->lock);
191 
192 	return &oi->vfs_inode;
193 }
194 
195 static void ovl_i_callback(struct rcu_head *head)
196 {
197 	struct inode *inode = container_of(head, struct inode, i_rcu);
198 
199 	kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
200 }
201 
202 static void ovl_destroy_inode(struct inode *inode)
203 {
204 	struct ovl_inode *oi = OVL_I(inode);
205 
206 	dput(oi->__upperdentry);
207 	kfree(oi->redirect);
208 	ovl_dir_cache_free(inode);
209 	mutex_destroy(&oi->lock);
210 
211 	call_rcu(&inode->i_rcu, ovl_i_callback);
212 }
213 
214 static void ovl_free_fs(struct ovl_fs *ofs)
215 {
216 	unsigned i;
217 
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 		mntput(ofs->lower_layers[i].mnt);
228 		free_anon_bdev(ofs->lower_layers[i].pseudo_dev);
229 	}
230 	kfree(ofs->lower_layers);
231 
232 	kfree(ofs->config.lowerdir);
233 	kfree(ofs->config.upperdir);
234 	kfree(ofs->config.workdir);
235 	if (ofs->creator_cred)
236 		put_cred(ofs->creator_cred);
237 	kfree(ofs);
238 }
239 
240 static void ovl_put_super(struct super_block *sb)
241 {
242 	struct ovl_fs *ofs = sb->s_fs_info;
243 
244 	ovl_free_fs(ofs);
245 }
246 
247 static int ovl_sync_fs(struct super_block *sb, int wait)
248 {
249 	struct ovl_fs *ofs = sb->s_fs_info;
250 	struct super_block *upper_sb;
251 	int ret;
252 
253 	if (!ofs->upper_mnt)
254 		return 0;
255 	upper_sb = ofs->upper_mnt->mnt_sb;
256 	if (!upper_sb->s_op->sync_fs)
257 		return 0;
258 
259 	/* real inodes have already been synced by sync_filesystem(ovl_sb) */
260 	down_read(&upper_sb->s_umount);
261 	ret = upper_sb->s_op->sync_fs(upper_sb, wait);
262 	up_read(&upper_sb->s_umount);
263 	return ret;
264 }
265 
266 /**
267  * ovl_statfs
268  * @sb: The overlayfs super block
269  * @buf: The struct kstatfs to fill in with stats
270  *
271  * Get the filesystem statistics.  As writes always target the upper layer
272  * filesystem pass the statfs to the upper filesystem (if it exists)
273  */
274 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
275 {
276 	struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
277 	struct dentry *root_dentry = dentry->d_sb->s_root;
278 	struct path path;
279 	int err;
280 
281 	ovl_path_real(root_dentry, &path);
282 
283 	err = vfs_statfs(&path, buf);
284 	if (!err) {
285 		buf->f_namelen = ofs->namelen;
286 		buf->f_type = OVERLAYFS_SUPER_MAGIC;
287 	}
288 
289 	return err;
290 }
291 
292 /* Will this overlay be forced to mount/remount ro? */
293 static bool ovl_force_readonly(struct ovl_fs *ofs)
294 {
295 	return (!ofs->upper_mnt || !ofs->workdir);
296 }
297 
298 /**
299  * ovl_show_options
300  *
301  * Prints the mount options for a given superblock.
302  * Returns zero; does not fail.
303  */
304 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
305 {
306 	struct super_block *sb = dentry->d_sb;
307 	struct ovl_fs *ofs = sb->s_fs_info;
308 
309 	seq_show_option(m, "lowerdir", ofs->config.lowerdir);
310 	if (ofs->config.upperdir) {
311 		seq_show_option(m, "upperdir", ofs->config.upperdir);
312 		seq_show_option(m, "workdir", ofs->config.workdir);
313 	}
314 	if (ofs->config.default_permissions)
315 		seq_puts(m, ",default_permissions");
316 	if (ofs->config.redirect_dir != ovl_redirect_dir_def)
317 		seq_printf(m, ",redirect_dir=%s",
318 			   ofs->config.redirect_dir ? "on" : "off");
319 	if (ofs->config.index != ovl_index_def)
320 		seq_printf(m, ",index=%s",
321 			   ofs->config.index ? "on" : "off");
322 	return 0;
323 }
324 
325 static int ovl_remount(struct super_block *sb, int *flags, char *data)
326 {
327 	struct ovl_fs *ofs = sb->s_fs_info;
328 
329 	if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
330 		return -EROFS;
331 
332 	return 0;
333 }
334 
335 static const struct super_operations ovl_super_operations = {
336 	.alloc_inode	= ovl_alloc_inode,
337 	.destroy_inode	= ovl_destroy_inode,
338 	.drop_inode	= generic_delete_inode,
339 	.put_super	= ovl_put_super,
340 	.sync_fs	= ovl_sync_fs,
341 	.statfs		= ovl_statfs,
342 	.show_options	= ovl_show_options,
343 	.remount_fs	= ovl_remount,
344 };
345 
346 enum {
347 	OPT_LOWERDIR,
348 	OPT_UPPERDIR,
349 	OPT_WORKDIR,
350 	OPT_DEFAULT_PERMISSIONS,
351 	OPT_REDIRECT_DIR_ON,
352 	OPT_REDIRECT_DIR_OFF,
353 	OPT_INDEX_ON,
354 	OPT_INDEX_OFF,
355 	OPT_ERR,
356 };
357 
358 static const match_table_t ovl_tokens = {
359 	{OPT_LOWERDIR,			"lowerdir=%s"},
360 	{OPT_UPPERDIR,			"upperdir=%s"},
361 	{OPT_WORKDIR,			"workdir=%s"},
362 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
363 	{OPT_REDIRECT_DIR_ON,		"redirect_dir=on"},
364 	{OPT_REDIRECT_DIR_OFF,		"redirect_dir=off"},
365 	{OPT_INDEX_ON,			"index=on"},
366 	{OPT_INDEX_OFF,			"index=off"},
367 	{OPT_ERR,			NULL}
368 };
369 
370 static char *ovl_next_opt(char **s)
371 {
372 	char *sbegin = *s;
373 	char *p;
374 
375 	if (sbegin == NULL)
376 		return NULL;
377 
378 	for (p = sbegin; *p; p++) {
379 		if (*p == '\\') {
380 			p++;
381 			if (!*p)
382 				break;
383 		} else if (*p == ',') {
384 			*p = '\0';
385 			*s = p + 1;
386 			return sbegin;
387 		}
388 	}
389 	*s = NULL;
390 	return sbegin;
391 }
392 
393 static int ovl_parse_opt(char *opt, struct ovl_config *config)
394 {
395 	char *p;
396 
397 	while ((p = ovl_next_opt(&opt)) != NULL) {
398 		int token;
399 		substring_t args[MAX_OPT_ARGS];
400 
401 		if (!*p)
402 			continue;
403 
404 		token = match_token(p, ovl_tokens, args);
405 		switch (token) {
406 		case OPT_UPPERDIR:
407 			kfree(config->upperdir);
408 			config->upperdir = match_strdup(&args[0]);
409 			if (!config->upperdir)
410 				return -ENOMEM;
411 			break;
412 
413 		case OPT_LOWERDIR:
414 			kfree(config->lowerdir);
415 			config->lowerdir = match_strdup(&args[0]);
416 			if (!config->lowerdir)
417 				return -ENOMEM;
418 			break;
419 
420 		case OPT_WORKDIR:
421 			kfree(config->workdir);
422 			config->workdir = match_strdup(&args[0]);
423 			if (!config->workdir)
424 				return -ENOMEM;
425 			break;
426 
427 		case OPT_DEFAULT_PERMISSIONS:
428 			config->default_permissions = true;
429 			break;
430 
431 		case OPT_REDIRECT_DIR_ON:
432 			config->redirect_dir = true;
433 			break;
434 
435 		case OPT_REDIRECT_DIR_OFF:
436 			config->redirect_dir = false;
437 			break;
438 
439 		case OPT_INDEX_ON:
440 			config->index = true;
441 			break;
442 
443 		case OPT_INDEX_OFF:
444 			config->index = false;
445 			break;
446 
447 		default:
448 			pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
449 			return -EINVAL;
450 		}
451 	}
452 
453 	/* Workdir is useless in non-upper mount */
454 	if (!config->upperdir && config->workdir) {
455 		pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
456 			config->workdir);
457 		kfree(config->workdir);
458 		config->workdir = NULL;
459 	}
460 
461 	return 0;
462 }
463 
464 #define OVL_WORKDIR_NAME "work"
465 #define OVL_INDEXDIR_NAME "index"
466 
467 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
468 					 const char *name, bool persist)
469 {
470 	struct inode *dir =  ofs->workbasedir->d_inode;
471 	struct vfsmount *mnt = ofs->upper_mnt;
472 	struct dentry *work;
473 	int err;
474 	bool retried = false;
475 	bool locked = false;
476 
477 	err = mnt_want_write(mnt);
478 	if (err)
479 		goto out_err;
480 
481 	inode_lock_nested(dir, I_MUTEX_PARENT);
482 	locked = true;
483 
484 retry:
485 	work = lookup_one_len(name, ofs->workbasedir, strlen(name));
486 
487 	if (!IS_ERR(work)) {
488 		struct iattr attr = {
489 			.ia_valid = ATTR_MODE,
490 			.ia_mode = S_IFDIR | 0,
491 		};
492 
493 		if (work->d_inode) {
494 			err = -EEXIST;
495 			if (retried)
496 				goto out_dput;
497 
498 			if (persist)
499 				goto out_unlock;
500 
501 			retried = true;
502 			ovl_workdir_cleanup(dir, mnt, work, 0);
503 			dput(work);
504 			goto retry;
505 		}
506 
507 		err = ovl_create_real(dir, work,
508 				      &(struct cattr){.mode = S_IFDIR | 0},
509 				      NULL, true);
510 		if (err)
511 			goto out_dput;
512 
513 		/*
514 		 * Try to remove POSIX ACL xattrs from workdir.  We are good if:
515 		 *
516 		 * a) success (there was a POSIX ACL xattr and was removed)
517 		 * b) -ENODATA (there was no POSIX ACL xattr)
518 		 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
519 		 *
520 		 * There are various other error values that could effectively
521 		 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
522 		 * if the xattr name is too long), but the set of filesystems
523 		 * allowed as upper are limited to "normal" ones, where checking
524 		 * for the above two errors is sufficient.
525 		 */
526 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
527 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
528 			goto out_dput;
529 
530 		err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
531 		if (err && err != -ENODATA && err != -EOPNOTSUPP)
532 			goto out_dput;
533 
534 		/* Clear any inherited mode bits */
535 		inode_lock(work->d_inode);
536 		err = notify_change(work, &attr, NULL);
537 		inode_unlock(work->d_inode);
538 		if (err)
539 			goto out_dput;
540 	} else {
541 		err = PTR_ERR(work);
542 		goto out_err;
543 	}
544 out_unlock:
545 	mnt_drop_write(mnt);
546 	if (locked)
547 		inode_unlock(dir);
548 
549 	return work;
550 
551 out_dput:
552 	dput(work);
553 out_err:
554 	pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
555 		ofs->config.workdir, name, -err);
556 	work = NULL;
557 	goto out_unlock;
558 }
559 
560 static void ovl_unescape(char *s)
561 {
562 	char *d = s;
563 
564 	for (;; s++, d++) {
565 		if (*s == '\\')
566 			s++;
567 		*d = *s;
568 		if (!*s)
569 			break;
570 	}
571 }
572 
573 static int ovl_mount_dir_noesc(const char *name, struct path *path)
574 {
575 	int err = -EINVAL;
576 
577 	if (!*name) {
578 		pr_err("overlayfs: empty lowerdir\n");
579 		goto out;
580 	}
581 	err = kern_path(name, LOOKUP_FOLLOW, path);
582 	if (err) {
583 		pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
584 		goto out;
585 	}
586 	err = -EINVAL;
587 	if (ovl_dentry_weird(path->dentry)) {
588 		pr_err("overlayfs: filesystem on '%s' not supported\n", name);
589 		goto out_put;
590 	}
591 	if (!d_is_dir(path->dentry)) {
592 		pr_err("overlayfs: '%s' not a directory\n", name);
593 		goto out_put;
594 	}
595 	return 0;
596 
597 out_put:
598 	path_put_init(path);
599 out:
600 	return err;
601 }
602 
603 static int ovl_mount_dir(const char *name, struct path *path)
604 {
605 	int err = -ENOMEM;
606 	char *tmp = kstrdup(name, GFP_KERNEL);
607 
608 	if (tmp) {
609 		ovl_unescape(tmp);
610 		err = ovl_mount_dir_noesc(tmp, path);
611 
612 		if (!err)
613 			if (ovl_dentry_remote(path->dentry)) {
614 				pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
615 				       tmp);
616 				path_put_init(path);
617 				err = -EINVAL;
618 			}
619 		kfree(tmp);
620 	}
621 	return err;
622 }
623 
624 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
625 			     const char *name)
626 {
627 	struct kstatfs statfs;
628 	int err = vfs_statfs(path, &statfs);
629 
630 	if (err)
631 		pr_err("overlayfs: statfs failed on '%s'\n", name);
632 	else
633 		ofs->namelen = max(ofs->namelen, statfs.f_namelen);
634 
635 	return err;
636 }
637 
638 static int ovl_lower_dir(const char *name, struct path *path,
639 			 struct ovl_fs *ofs, int *stack_depth, bool *remote)
640 {
641 	int err;
642 
643 	err = ovl_mount_dir_noesc(name, path);
644 	if (err)
645 		goto out;
646 
647 	err = ovl_check_namelen(path, ofs, name);
648 	if (err)
649 		goto out_put;
650 
651 	*stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
652 
653 	if (ovl_dentry_remote(path->dentry))
654 		*remote = true;
655 
656 	/*
657 	 * The inodes index feature needs to encode and decode file
658 	 * handles, so it requires that all layers support them.
659 	 */
660 	if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
661 		ofs->config.index = false;
662 		pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
663 	}
664 
665 	return 0;
666 
667 out_put:
668 	path_put_init(path);
669 out:
670 	return err;
671 }
672 
673 /* Workdir should not be subdir of upperdir and vice versa */
674 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
675 {
676 	bool ok = false;
677 
678 	if (workdir != upperdir) {
679 		ok = (lock_rename(workdir, upperdir) == NULL);
680 		unlock_rename(workdir, upperdir);
681 	}
682 	return ok;
683 }
684 
685 static unsigned int ovl_split_lowerdirs(char *str)
686 {
687 	unsigned int ctr = 1;
688 	char *s, *d;
689 
690 	for (s = d = str;; s++, d++) {
691 		if (*s == '\\') {
692 			s++;
693 		} else if (*s == ':') {
694 			*d = '\0';
695 			ctr++;
696 			continue;
697 		}
698 		*d = *s;
699 		if (!*s)
700 			break;
701 	}
702 	return ctr;
703 }
704 
705 static int __maybe_unused
706 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
707 			struct dentry *dentry, struct inode *inode,
708 			const char *name, void *buffer, size_t size)
709 {
710 	return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
711 }
712 
713 static int __maybe_unused
714 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
715 			struct dentry *dentry, struct inode *inode,
716 			const char *name, const void *value,
717 			size_t size, int flags)
718 {
719 	struct dentry *workdir = ovl_workdir(dentry);
720 	struct inode *realinode = ovl_inode_real(inode);
721 	struct posix_acl *acl = NULL;
722 	int err;
723 
724 	/* Check that everything is OK before copy-up */
725 	if (value) {
726 		acl = posix_acl_from_xattr(&init_user_ns, value, size);
727 		if (IS_ERR(acl))
728 			return PTR_ERR(acl);
729 	}
730 	err = -EOPNOTSUPP;
731 	if (!IS_POSIXACL(d_inode(workdir)))
732 		goto out_acl_release;
733 	if (!realinode->i_op->set_acl)
734 		goto out_acl_release;
735 	if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
736 		err = acl ? -EACCES : 0;
737 		goto out_acl_release;
738 	}
739 	err = -EPERM;
740 	if (!inode_owner_or_capable(inode))
741 		goto out_acl_release;
742 
743 	posix_acl_release(acl);
744 
745 	/*
746 	 * Check if sgid bit needs to be cleared (actual setacl operation will
747 	 * be done with mounter's capabilities and so that won't do it for us).
748 	 */
749 	if (unlikely(inode->i_mode & S_ISGID) &&
750 	    handler->flags == ACL_TYPE_ACCESS &&
751 	    !in_group_p(inode->i_gid) &&
752 	    !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
753 		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
754 
755 		err = ovl_setattr(dentry, &iattr);
756 		if (err)
757 			return err;
758 	}
759 
760 	err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
761 	if (!err)
762 		ovl_copyattr(ovl_inode_real(inode), inode);
763 
764 	return err;
765 
766 out_acl_release:
767 	posix_acl_release(acl);
768 	return err;
769 }
770 
771 static int ovl_own_xattr_get(const struct xattr_handler *handler,
772 			     struct dentry *dentry, struct inode *inode,
773 			     const char *name, void *buffer, size_t size)
774 {
775 	return -EOPNOTSUPP;
776 }
777 
778 static int ovl_own_xattr_set(const struct xattr_handler *handler,
779 			     struct dentry *dentry, struct inode *inode,
780 			     const char *name, const void *value,
781 			     size_t size, int flags)
782 {
783 	return -EOPNOTSUPP;
784 }
785 
786 static int ovl_other_xattr_get(const struct xattr_handler *handler,
787 			       struct dentry *dentry, struct inode *inode,
788 			       const char *name, void *buffer, size_t size)
789 {
790 	return ovl_xattr_get(dentry, inode, name, buffer, size);
791 }
792 
793 static int ovl_other_xattr_set(const struct xattr_handler *handler,
794 			       struct dentry *dentry, struct inode *inode,
795 			       const char *name, const void *value,
796 			       size_t size, int flags)
797 {
798 	return ovl_xattr_set(dentry, inode, name, value, size, flags);
799 }
800 
801 static const struct xattr_handler __maybe_unused
802 ovl_posix_acl_access_xattr_handler = {
803 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
804 	.flags = ACL_TYPE_ACCESS,
805 	.get = ovl_posix_acl_xattr_get,
806 	.set = ovl_posix_acl_xattr_set,
807 };
808 
809 static const struct xattr_handler __maybe_unused
810 ovl_posix_acl_default_xattr_handler = {
811 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
812 	.flags = ACL_TYPE_DEFAULT,
813 	.get = ovl_posix_acl_xattr_get,
814 	.set = ovl_posix_acl_xattr_set,
815 };
816 
817 static const struct xattr_handler ovl_own_xattr_handler = {
818 	.prefix	= OVL_XATTR_PREFIX,
819 	.get = ovl_own_xattr_get,
820 	.set = ovl_own_xattr_set,
821 };
822 
823 static const struct xattr_handler ovl_other_xattr_handler = {
824 	.prefix	= "", /* catch all */
825 	.get = ovl_other_xattr_get,
826 	.set = ovl_other_xattr_set,
827 };
828 
829 static const struct xattr_handler *ovl_xattr_handlers[] = {
830 #ifdef CONFIG_FS_POSIX_ACL
831 	&ovl_posix_acl_access_xattr_handler,
832 	&ovl_posix_acl_default_xattr_handler,
833 #endif
834 	&ovl_own_xattr_handler,
835 	&ovl_other_xattr_handler,
836 	NULL
837 };
838 
839 static int ovl_get_upper(struct ovl_fs *ofs, struct path *upperpath)
840 {
841 	struct vfsmount *upper_mnt;
842 	int err;
843 
844 	err = ovl_mount_dir(ofs->config.upperdir, upperpath);
845 	if (err)
846 		goto out;
847 
848 	/* Upper fs should not be r/o */
849 	if (sb_rdonly(upperpath->mnt->mnt_sb)) {
850 		pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
851 		err = -EINVAL;
852 		goto out;
853 	}
854 
855 	err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
856 	if (err)
857 		goto out;
858 
859 	err = -EBUSY;
860 	if (ovl_inuse_trylock(upperpath->dentry)) {
861 		ofs->upperdir_locked = true;
862 	} else if (ofs->config.index) {
863 		pr_err("overlayfs: upperdir is in-use by another mount, mount with '-o index=off' to override exclusive upperdir protection.\n");
864 		goto out;
865 	} else {
866 		pr_warn("overlayfs: upperdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
867 	}
868 
869 	upper_mnt = clone_private_mount(upperpath);
870 	err = PTR_ERR(upper_mnt);
871 	if (IS_ERR(upper_mnt)) {
872 		pr_err("overlayfs: failed to clone upperpath\n");
873 		goto out;
874 	}
875 
876 	/* Don't inherit atime flags */
877 	upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
878 	ofs->upper_mnt = upper_mnt;
879 	err = 0;
880 out:
881 	return err;
882 }
883 
884 static int ovl_make_workdir(struct ovl_fs *ofs, struct path *workpath)
885 {
886 	struct dentry *temp;
887 	int err;
888 
889 	ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
890 	if (!ofs->workdir)
891 		return 0;
892 
893 	/*
894 	 * Upper should support d_type, else whiteouts are visible.  Given
895 	 * workdir and upper are on same fs, we can do iterate_dir() on
896 	 * workdir. This check requires successful creation of workdir in
897 	 * previous step.
898 	 */
899 	err = ovl_check_d_type_supported(workpath);
900 	if (err < 0)
901 		return err;
902 
903 	/*
904 	 * We allowed this configuration and don't want to break users over
905 	 * kernel upgrade. So warn instead of erroring out.
906 	 */
907 	if (!err)
908 		pr_warn("overlayfs: upper fs needs to support d_type.\n");
909 
910 	/* Check if upper/work fs supports O_TMPFILE */
911 	temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
912 	ofs->tmpfile = !IS_ERR(temp);
913 	if (ofs->tmpfile)
914 		dput(temp);
915 	else
916 		pr_warn("overlayfs: upper fs does not support tmpfile.\n");
917 
918 	/*
919 	 * Check if upper/work fs supports trusted.overlay.* xattr
920 	 */
921 	err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
922 	if (err) {
923 		ofs->noxattr = true;
924 		pr_warn("overlayfs: upper fs does not support xattr.\n");
925 	} else {
926 		vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
927 	}
928 
929 	/* Check if upper/work fs supports file handles */
930 	if (ofs->config.index &&
931 	    !ovl_can_decode_fh(ofs->workdir->d_sb)) {
932 		ofs->config.index = false;
933 		pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
934 	}
935 
936 	return 0;
937 }
938 
939 static int ovl_get_workdir(struct ovl_fs *ofs, struct path *upperpath)
940 {
941 	int err;
942 	struct path workpath = { };
943 
944 	err = ovl_mount_dir(ofs->config.workdir, &workpath);
945 	if (err)
946 		goto out;
947 
948 	err = -EINVAL;
949 	if (upperpath->mnt != workpath.mnt) {
950 		pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
951 		goto out;
952 	}
953 	if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
954 		pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
955 		goto out;
956 	}
957 
958 	err = -EBUSY;
959 	if (ovl_inuse_trylock(workpath.dentry)) {
960 		ofs->workdir_locked = true;
961 	} else if (ofs->config.index) {
962 		pr_err("overlayfs: workdir is in-use by another mount, mount with '-o index=off' to override exclusive workdir protection.\n");
963 		goto out;
964 	} else {
965 		pr_warn("overlayfs: workdir is in-use by another mount, accessing files from both mounts will result in undefined behavior.\n");
966 	}
967 
968 	ofs->workbasedir = dget(workpath.dentry);
969 	err = ovl_make_workdir(ofs, &workpath);
970 	if (err)
971 		goto out;
972 
973 	err = 0;
974 out:
975 	path_put(&workpath);
976 
977 	return err;
978 }
979 
980 static int ovl_get_indexdir(struct ovl_fs *ofs, struct ovl_entry *oe,
981 			    struct path *upperpath)
982 {
983 	int err;
984 
985 	/* Verify lower root is upper root origin */
986 	err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
987 				false, true);
988 	if (err) {
989 		pr_err("overlayfs: failed to verify upper root origin\n");
990 		goto out;
991 	}
992 
993 	ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
994 	if (ofs->indexdir) {
995 		/* Verify upper root is index dir origin */
996 		err = ovl_verify_origin(ofs->indexdir, upperpath->dentry,
997 					true, true);
998 		if (err)
999 			pr_err("overlayfs: failed to verify index dir origin\n");
1000 
1001 		/* Cleanup bad/stale/orphan index entries */
1002 		if (!err)
1003 			err = ovl_indexdir_cleanup(ofs->indexdir,
1004 						   ofs->upper_mnt,
1005 						   oe->lowerstack,
1006 						   oe->numlower);
1007 	}
1008 	if (err || !ofs->indexdir)
1009 		pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1010 
1011 out:
1012 	return err;
1013 }
1014 
1015 static int ovl_get_lower_layers(struct ovl_fs *ofs, struct path *stack,
1016 				unsigned int numlower)
1017 {
1018 	int err;
1019 	unsigned int i;
1020 
1021 	err = -ENOMEM;
1022 	ofs->lower_layers = kcalloc(numlower, sizeof(struct ovl_layer),
1023 				    GFP_KERNEL);
1024 	if (ofs->lower_layers == NULL)
1025 		goto out;
1026 	for (i = 0; i < numlower; i++) {
1027 		struct vfsmount *mnt;
1028 		dev_t dev;
1029 
1030 		err = get_anon_bdev(&dev);
1031 		if (err) {
1032 			pr_err("overlayfs: failed to get anonymous bdev for lowerpath\n");
1033 			goto out;
1034 		}
1035 
1036 		mnt = clone_private_mount(&stack[i]);
1037 		err = PTR_ERR(mnt);
1038 		if (IS_ERR(mnt)) {
1039 			pr_err("overlayfs: failed to clone lowerpath\n");
1040 			free_anon_bdev(dev);
1041 			goto out;
1042 		}
1043 		/*
1044 		 * Make lower layers R/O.  That way fchmod/fchown on lower file
1045 		 * will fail instead of modifying lower fs.
1046 		 */
1047 		mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1048 
1049 		ofs->lower_layers[ofs->numlower].mnt = mnt;
1050 		ofs->lower_layers[ofs->numlower].pseudo_dev = dev;
1051 		ofs->numlower++;
1052 
1053 		/* Check if all lower layers are on same sb */
1054 		if (i == 0)
1055 			ofs->same_sb = mnt->mnt_sb;
1056 		else if (ofs->same_sb != mnt->mnt_sb)
1057 			ofs->same_sb = NULL;
1058 	}
1059 	err = 0;
1060 out:
1061 	return err;
1062 }
1063 
1064 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1065 					    struct ovl_fs *ofs)
1066 {
1067 	int err;
1068 	char *lowertmp, *lower;
1069 	struct path *stack = NULL;
1070 	unsigned int stacklen, numlower = 0, i;
1071 	bool remote = false;
1072 	struct ovl_entry *oe;
1073 
1074 	err = -ENOMEM;
1075 	lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1076 	if (!lowertmp)
1077 		goto out_err;
1078 
1079 	err = -EINVAL;
1080 	stacklen = ovl_split_lowerdirs(lowertmp);
1081 	if (stacklen > OVL_MAX_STACK) {
1082 		pr_err("overlayfs: too many lower directories, limit is %d\n",
1083 		       OVL_MAX_STACK);
1084 		goto out_err;
1085 	} else if (!ofs->config.upperdir && stacklen == 1) {
1086 		pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1087 		goto out_err;
1088 	}
1089 
1090 	err = -ENOMEM;
1091 	stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1092 	if (!stack)
1093 		goto out_err;
1094 
1095 	err = -EINVAL;
1096 	lower = lowertmp;
1097 	for (numlower = 0; numlower < stacklen; numlower++) {
1098 		err = ovl_lower_dir(lower, &stack[numlower], ofs,
1099 				    &sb->s_stack_depth, &remote);
1100 		if (err)
1101 			goto out_err;
1102 
1103 		lower = strchr(lower, '\0') + 1;
1104 	}
1105 
1106 	err = -EINVAL;
1107 	sb->s_stack_depth++;
1108 	if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1109 		pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1110 		goto out_err;
1111 	}
1112 
1113 	err = ovl_get_lower_layers(ofs, stack, numlower);
1114 	if (err)
1115 		goto out_err;
1116 
1117 	err = -ENOMEM;
1118 	oe = ovl_alloc_entry(numlower);
1119 	if (!oe)
1120 		goto out_err;
1121 
1122 	for (i = 0; i < numlower; i++) {
1123 		oe->lowerstack[i].dentry = dget(stack[i].dentry);
1124 		oe->lowerstack[i].layer = &ofs->lower_layers[i];
1125 	}
1126 
1127 	if (remote)
1128 		sb->s_d_op = &ovl_reval_dentry_operations;
1129 	else
1130 		sb->s_d_op = &ovl_dentry_operations;
1131 
1132 out:
1133 	for (i = 0; i < numlower; i++)
1134 		path_put(&stack[i]);
1135 	kfree(stack);
1136 	kfree(lowertmp);
1137 
1138 	return oe;
1139 
1140 out_err:
1141 	oe = ERR_PTR(err);
1142 	goto out;
1143 }
1144 
1145 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1146 {
1147 	struct path upperpath = { };
1148 	struct dentry *root_dentry;
1149 	struct ovl_entry *oe;
1150 	struct ovl_fs *ofs;
1151 	struct cred *cred;
1152 	int err;
1153 
1154 	err = -ENOMEM;
1155 	ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1156 	if (!ofs)
1157 		goto out;
1158 
1159 	ofs->creator_cred = cred = prepare_creds();
1160 	if (!cred)
1161 		goto out_err;
1162 
1163 	ofs->config.redirect_dir = ovl_redirect_dir_def;
1164 	ofs->config.index = ovl_index_def;
1165 	err = ovl_parse_opt((char *) data, &ofs->config);
1166 	if (err)
1167 		goto out_err;
1168 
1169 	err = -EINVAL;
1170 	if (!ofs->config.lowerdir) {
1171 		if (!silent)
1172 			pr_err("overlayfs: missing 'lowerdir'\n");
1173 		goto out_err;
1174 	}
1175 
1176 	sb->s_stack_depth = 0;
1177 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1178 	if (ofs->config.upperdir) {
1179 		if (!ofs->config.workdir) {
1180 			pr_err("overlayfs: missing 'workdir'\n");
1181 			goto out_err;
1182 		}
1183 
1184 		err = ovl_get_upper(ofs, &upperpath);
1185 		if (err)
1186 			goto out_err;
1187 
1188 		err = ovl_get_workdir(ofs, &upperpath);
1189 		if (err)
1190 			goto out_err;
1191 
1192 		if (!ofs->workdir)
1193 			sb->s_flags |= SB_RDONLY;
1194 
1195 		sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1196 		sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1197 
1198 	}
1199 	oe = ovl_get_lowerstack(sb, ofs);
1200 	err = PTR_ERR(oe);
1201 	if (IS_ERR(oe))
1202 		goto out_err;
1203 
1204 	/* If the upper fs is nonexistent, we mark overlayfs r/o too */
1205 	if (!ofs->upper_mnt)
1206 		sb->s_flags |= SB_RDONLY;
1207 	else if (ofs->upper_mnt->mnt_sb != ofs->same_sb)
1208 		ofs->same_sb = NULL;
1209 
1210 	if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1211 		err = ovl_get_indexdir(ofs, oe, &upperpath);
1212 		if (err)
1213 			goto out_free_oe;
1214 
1215 		if (!ofs->indexdir)
1216 			sb->s_flags |= SB_RDONLY;
1217 	}
1218 
1219 	/* Show index=off/on in /proc/mounts for any of the reasons above */
1220 	if (!ofs->indexdir)
1221 		ofs->config.index = false;
1222 
1223 	/* Never override disk quota limits or use reserved space */
1224 	cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1225 
1226 	sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1227 	sb->s_op = &ovl_super_operations;
1228 	sb->s_xattr = ovl_xattr_handlers;
1229 	sb->s_fs_info = ofs;
1230 	sb->s_flags |= SB_POSIXACL | SB_NOREMOTELOCK;
1231 
1232 	err = -ENOMEM;
1233 	root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1234 	if (!root_dentry)
1235 		goto out_free_oe;
1236 
1237 	mntput(upperpath.mnt);
1238 	if (upperpath.dentry) {
1239 		oe->has_upper = true;
1240 		if (ovl_is_impuredir(upperpath.dentry))
1241 			ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1242 	}
1243 
1244 	root_dentry->d_fsdata = oe;
1245 
1246 	/* Root is always merge -> can have whiteouts */
1247 	ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1248 	ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1249 		       ovl_dentry_lower(root_dentry));
1250 
1251 	sb->s_root = root_dentry;
1252 
1253 	return 0;
1254 
1255 out_free_oe:
1256 	ovl_entry_stack_free(oe);
1257 	kfree(oe);
1258 out_err:
1259 	path_put(&upperpath);
1260 	ovl_free_fs(ofs);
1261 out:
1262 	return err;
1263 }
1264 
1265 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1266 				const char *dev_name, void *raw_data)
1267 {
1268 	return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1269 }
1270 
1271 static struct file_system_type ovl_fs_type = {
1272 	.owner		= THIS_MODULE,
1273 	.name		= "overlay",
1274 	.mount		= ovl_mount,
1275 	.kill_sb	= kill_anon_super,
1276 };
1277 MODULE_ALIAS_FS("overlay");
1278 
1279 static void ovl_inode_init_once(void *foo)
1280 {
1281 	struct ovl_inode *oi = foo;
1282 
1283 	inode_init_once(&oi->vfs_inode);
1284 }
1285 
1286 static int __init ovl_init(void)
1287 {
1288 	int err;
1289 
1290 	ovl_inode_cachep = kmem_cache_create("ovl_inode",
1291 					     sizeof(struct ovl_inode), 0,
1292 					     (SLAB_RECLAIM_ACCOUNT|
1293 					      SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1294 					     ovl_inode_init_once);
1295 	if (ovl_inode_cachep == NULL)
1296 		return -ENOMEM;
1297 
1298 	err = register_filesystem(&ovl_fs_type);
1299 	if (err)
1300 		kmem_cache_destroy(ovl_inode_cachep);
1301 
1302 	return err;
1303 }
1304 
1305 static void __exit ovl_exit(void)
1306 {
1307 	unregister_filesystem(&ovl_fs_type);
1308 
1309 	/*
1310 	 * Make sure all delayed rcu free inodes are flushed before we
1311 	 * destroy cache.
1312 	 */
1313 	rcu_barrier();
1314 	kmem_cache_destroy(ovl_inode_cachep);
1315 
1316 }
1317 
1318 module_init(ovl_init);
1319 module_exit(ovl_exit);
1320