xref: /openbmc/linux/fs/overlayfs/dir.c (revision 727dede0)
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 <linux/fs.h>
11 #include <linux/namei.h>
12 #include <linux/xattr.h>
13 #include <linux/security.h>
14 #include <linux/cred.h>
15 #include <linux/module.h>
16 #include <linux/posix_acl.h>
17 #include <linux/posix_acl_xattr.h>
18 #include <linux/atomic.h>
19 #include <linux/ratelimit.h>
20 #include "overlayfs.h"
21 
22 static unsigned short ovl_redirect_max = 256;
23 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
24 MODULE_PARM_DESC(ovl_redirect_max,
25 		 "Maximum length of absolute redirect xattr value");
26 
27 int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
28 {
29 	int err;
30 
31 	dget(wdentry);
32 	if (d_is_dir(wdentry))
33 		err = ovl_do_rmdir(wdir, wdentry);
34 	else
35 		err = ovl_do_unlink(wdir, wdentry);
36 	dput(wdentry);
37 
38 	if (err) {
39 		pr_err("overlayfs: cleanup of '%pd2' failed (%i)\n",
40 		       wdentry, err);
41 	}
42 
43 	return err;
44 }
45 
46 struct dentry *ovl_lookup_temp(struct dentry *workdir)
47 {
48 	struct dentry *temp;
49 	char name[20];
50 	static atomic_t temp_id = ATOMIC_INIT(0);
51 
52 	/* counter is allowed to wrap, since temp dentries are ephemeral */
53 	snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
54 
55 	temp = lookup_one_len(name, workdir, strlen(name));
56 	if (!IS_ERR(temp) && temp->d_inode) {
57 		pr_err("overlayfs: workdir/%s already exists\n", name);
58 		dput(temp);
59 		temp = ERR_PTR(-EIO);
60 	}
61 
62 	return temp;
63 }
64 
65 /* caller holds i_mutex on workdir */
66 static struct dentry *ovl_whiteout(struct dentry *workdir,
67 				   struct dentry *dentry)
68 {
69 	int err;
70 	struct dentry *whiteout;
71 	struct inode *wdir = workdir->d_inode;
72 
73 	whiteout = ovl_lookup_temp(workdir);
74 	if (IS_ERR(whiteout))
75 		return whiteout;
76 
77 	err = ovl_do_whiteout(wdir, whiteout);
78 	if (err) {
79 		dput(whiteout);
80 		whiteout = ERR_PTR(err);
81 	}
82 
83 	return whiteout;
84 }
85 
86 int ovl_create_real(struct inode *dir, struct dentry *newdentry,
87 		    struct cattr *attr, struct dentry *hardlink, bool debug)
88 {
89 	int err;
90 
91 	if (newdentry->d_inode)
92 		return -ESTALE;
93 
94 	if (hardlink) {
95 		err = ovl_do_link(hardlink, dir, newdentry, debug);
96 	} else {
97 		switch (attr->mode & S_IFMT) {
98 		case S_IFREG:
99 			err = ovl_do_create(dir, newdentry, attr->mode, debug);
100 			break;
101 
102 		case S_IFDIR:
103 			err = ovl_do_mkdir(dir, newdentry, attr->mode, debug);
104 			break;
105 
106 		case S_IFCHR:
107 		case S_IFBLK:
108 		case S_IFIFO:
109 		case S_IFSOCK:
110 			err = ovl_do_mknod(dir, newdentry,
111 					   attr->mode, attr->rdev, debug);
112 			break;
113 
114 		case S_IFLNK:
115 			err = ovl_do_symlink(dir, newdentry, attr->link, debug);
116 			break;
117 
118 		default:
119 			err = -EPERM;
120 		}
121 	}
122 	if (!err && WARN_ON(!newdentry->d_inode)) {
123 		/*
124 		 * Not quite sure if non-instantiated dentry is legal or not.
125 		 * VFS doesn't seem to care so check and warn here.
126 		 */
127 		err = -ENOENT;
128 	}
129 	return err;
130 }
131 
132 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
133 			       int xerr)
134 {
135 	int err;
136 
137 	err = ovl_check_setxattr(dentry, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
138 	if (!err)
139 		ovl_dentry_set_opaque(dentry);
140 
141 	return err;
142 }
143 
144 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
145 {
146 	/*
147 	 * Fail with -EIO when trying to create opaque dir and upper doesn't
148 	 * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
149 	 * return a specific error for noxattr case.
150 	 */
151 	return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
152 }
153 
154 /* Common operations required to be done after creation of file on upper */
155 static void ovl_instantiate(struct dentry *dentry, struct inode *inode,
156 			    struct dentry *newdentry, bool hardlink)
157 {
158 	ovl_dentry_version_inc(dentry->d_parent, false);
159 	ovl_dentry_set_upper_alias(dentry);
160 	if (!hardlink) {
161 		ovl_inode_update(inode, newdentry);
162 		ovl_copyattr(newdentry->d_inode, inode);
163 	} else {
164 		WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
165 		dput(newdentry);
166 		inc_nlink(inode);
167 	}
168 	d_instantiate(dentry, inode);
169 	/* Force lookup of new upper hardlink to find its lower */
170 	if (hardlink)
171 		d_drop(dentry);
172 }
173 
174 static bool ovl_type_merge(struct dentry *dentry)
175 {
176 	return OVL_TYPE_MERGE(ovl_path_type(dentry));
177 }
178 
179 static bool ovl_type_origin(struct dentry *dentry)
180 {
181 	return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
182 }
183 
184 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
185 			    struct cattr *attr, struct dentry *hardlink)
186 {
187 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
188 	struct inode *udir = upperdir->d_inode;
189 	struct dentry *newdentry;
190 	int err;
191 
192 	if (!hardlink && !IS_POSIXACL(udir))
193 		attr->mode &= ~current_umask();
194 
195 	inode_lock_nested(udir, I_MUTEX_PARENT);
196 	newdentry = lookup_one_len(dentry->d_name.name, upperdir,
197 				   dentry->d_name.len);
198 	err = PTR_ERR(newdentry);
199 	if (IS_ERR(newdentry))
200 		goto out_unlock;
201 	err = ovl_create_real(udir, newdentry, attr, hardlink, false);
202 	if (err)
203 		goto out_dput;
204 
205 	if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry)) {
206 		/* Setting opaque here is just an optimization, allow to fail */
207 		ovl_set_opaque(dentry, newdentry);
208 	}
209 
210 	ovl_instantiate(dentry, inode, newdentry, !!hardlink);
211 	newdentry = NULL;
212 out_dput:
213 	dput(newdentry);
214 out_unlock:
215 	inode_unlock(udir);
216 	return err;
217 }
218 
219 static struct dentry *ovl_clear_empty(struct dentry *dentry,
220 				      struct list_head *list)
221 {
222 	struct dentry *workdir = ovl_workdir(dentry);
223 	struct inode *wdir = workdir->d_inode;
224 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
225 	struct inode *udir = upperdir->d_inode;
226 	struct path upperpath;
227 	struct dentry *upper;
228 	struct dentry *opaquedir;
229 	struct kstat stat;
230 	int err;
231 
232 	if (WARN_ON(!workdir))
233 		return ERR_PTR(-EROFS);
234 
235 	err = ovl_lock_rename_workdir(workdir, upperdir);
236 	if (err)
237 		goto out;
238 
239 	ovl_path_upper(dentry, &upperpath);
240 	err = vfs_getattr(&upperpath, &stat,
241 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
242 	if (err)
243 		goto out_unlock;
244 
245 	err = -ESTALE;
246 	if (!S_ISDIR(stat.mode))
247 		goto out_unlock;
248 	upper = upperpath.dentry;
249 	if (upper->d_parent->d_inode != udir)
250 		goto out_unlock;
251 
252 	opaquedir = ovl_lookup_temp(workdir);
253 	err = PTR_ERR(opaquedir);
254 	if (IS_ERR(opaquedir))
255 		goto out_unlock;
256 
257 	err = ovl_create_real(wdir, opaquedir,
258 			      &(struct cattr){.mode = stat.mode}, NULL, true);
259 	if (err)
260 		goto out_dput;
261 
262 	err = ovl_copy_xattr(upper, opaquedir);
263 	if (err)
264 		goto out_cleanup;
265 
266 	err = ovl_set_opaque(dentry, opaquedir);
267 	if (err)
268 		goto out_cleanup;
269 
270 	inode_lock(opaquedir->d_inode);
271 	err = ovl_set_attr(opaquedir, &stat);
272 	inode_unlock(opaquedir->d_inode);
273 	if (err)
274 		goto out_cleanup;
275 
276 	err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
277 	if (err)
278 		goto out_cleanup;
279 
280 	ovl_cleanup_whiteouts(upper, list);
281 	ovl_cleanup(wdir, upper);
282 	unlock_rename(workdir, upperdir);
283 
284 	/* dentry's upper doesn't match now, get rid of it */
285 	d_drop(dentry);
286 
287 	return opaquedir;
288 
289 out_cleanup:
290 	ovl_cleanup(wdir, opaquedir);
291 out_dput:
292 	dput(opaquedir);
293 out_unlock:
294 	unlock_rename(workdir, upperdir);
295 out:
296 	return ERR_PTR(err);
297 }
298 
299 static struct dentry *ovl_check_empty_and_clear(struct dentry *dentry)
300 {
301 	int err;
302 	struct dentry *ret = NULL;
303 	enum ovl_path_type type = ovl_path_type(dentry);
304 	LIST_HEAD(list);
305 
306 	err = ovl_check_empty_dir(dentry, &list);
307 	if (err) {
308 		ret = ERR_PTR(err);
309 		goto out_free;
310 	}
311 
312 	/*
313 	 * When removing an empty opaque directory, then it makes no sense to
314 	 * replace it with an exact replica of itself.
315 	 *
316 	 * If no upperdentry then skip clearing whiteouts.
317 	 *
318 	 * Can race with copy-up, since we don't hold the upperdir mutex.
319 	 * Doesn't matter, since copy-up can't create a non-empty directory
320 	 * from an empty one.
321 	 */
322 	if (OVL_TYPE_UPPER(type) && OVL_TYPE_MERGE(type))
323 		ret = ovl_clear_empty(dentry, &list);
324 
325 out_free:
326 	ovl_cache_free(&list);
327 
328 	return ret;
329 }
330 
331 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
332 			     const struct posix_acl *acl)
333 {
334 	void *buffer;
335 	size_t size;
336 	int err;
337 
338 	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
339 		return 0;
340 
341 	size = posix_acl_to_xattr(NULL, acl, NULL, 0);
342 	buffer = kmalloc(size, GFP_KERNEL);
343 	if (!buffer)
344 		return -ENOMEM;
345 
346 	size = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
347 	err = size;
348 	if (err < 0)
349 		goto out_free;
350 
351 	err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
352 out_free:
353 	kfree(buffer);
354 	return err;
355 }
356 
357 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
358 				    struct cattr *cattr,
359 				    struct dentry *hardlink)
360 {
361 	struct dentry *workdir = ovl_workdir(dentry);
362 	struct inode *wdir = workdir->d_inode;
363 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
364 	struct inode *udir = upperdir->d_inode;
365 	struct dentry *upper;
366 	struct dentry *newdentry;
367 	int err;
368 	struct posix_acl *acl, *default_acl;
369 
370 	if (WARN_ON(!workdir))
371 		return -EROFS;
372 
373 	if (!hardlink) {
374 		err = posix_acl_create(dentry->d_parent->d_inode,
375 				       &cattr->mode, &default_acl, &acl);
376 		if (err)
377 			return err;
378 	}
379 
380 	err = ovl_lock_rename_workdir(workdir, upperdir);
381 	if (err)
382 		goto out;
383 
384 	newdentry = ovl_lookup_temp(workdir);
385 	err = PTR_ERR(newdentry);
386 	if (IS_ERR(newdentry))
387 		goto out_unlock;
388 
389 	upper = lookup_one_len(dentry->d_name.name, upperdir,
390 			       dentry->d_name.len);
391 	err = PTR_ERR(upper);
392 	if (IS_ERR(upper))
393 		goto out_dput;
394 
395 	err = ovl_create_real(wdir, newdentry, cattr, hardlink, true);
396 	if (err)
397 		goto out_dput2;
398 
399 	/*
400 	 * mode could have been mutilated due to umask (e.g. sgid directory)
401 	 */
402 	if (!hardlink &&
403 	    !S_ISLNK(cattr->mode) &&
404 	    newdentry->d_inode->i_mode != cattr->mode) {
405 		struct iattr attr = {
406 			.ia_valid = ATTR_MODE,
407 			.ia_mode = cattr->mode,
408 		};
409 		inode_lock(newdentry->d_inode);
410 		err = notify_change(newdentry, &attr, NULL);
411 		inode_unlock(newdentry->d_inode);
412 		if (err)
413 			goto out_cleanup;
414 	}
415 	if (!hardlink) {
416 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
417 					acl);
418 		if (err)
419 			goto out_cleanup;
420 
421 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
422 					default_acl);
423 		if (err)
424 			goto out_cleanup;
425 	}
426 
427 	if (!hardlink && S_ISDIR(cattr->mode)) {
428 		err = ovl_set_opaque(dentry, newdentry);
429 		if (err)
430 			goto out_cleanup;
431 
432 		err = ovl_do_rename(wdir, newdentry, udir, upper,
433 				    RENAME_EXCHANGE);
434 		if (err)
435 			goto out_cleanup;
436 
437 		ovl_cleanup(wdir, upper);
438 	} else {
439 		err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
440 		if (err)
441 			goto out_cleanup;
442 	}
443 	ovl_instantiate(dentry, inode, newdentry, !!hardlink);
444 	newdentry = NULL;
445 out_dput2:
446 	dput(upper);
447 out_dput:
448 	dput(newdentry);
449 out_unlock:
450 	unlock_rename(workdir, upperdir);
451 out:
452 	if (!hardlink) {
453 		posix_acl_release(acl);
454 		posix_acl_release(default_acl);
455 	}
456 	return err;
457 
458 out_cleanup:
459 	ovl_cleanup(wdir, newdentry);
460 	goto out_dput2;
461 }
462 
463 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
464 			      struct cattr *attr, struct dentry *hardlink,
465 			      bool origin)
466 {
467 	int err;
468 	const struct cred *old_cred;
469 	struct cred *override_cred;
470 	struct dentry *parent = dentry->d_parent;
471 
472 	err = ovl_copy_up(parent);
473 	if (err)
474 		return err;
475 
476 	old_cred = ovl_override_creds(dentry->d_sb);
477 
478 	/*
479 	 * When linking a file with copy up origin into a new parent, mark the
480 	 * new parent dir "impure".
481 	 */
482 	if (origin) {
483 		err = ovl_set_impure(parent, ovl_dentry_upper(parent));
484 		if (err)
485 			goto out_revert_creds;
486 	}
487 
488 	err = -ENOMEM;
489 	override_cred = prepare_creds();
490 	if (override_cred) {
491 		override_cred->fsuid = inode->i_uid;
492 		override_cred->fsgid = inode->i_gid;
493 		if (!hardlink) {
494 			err = security_dentry_create_files_as(dentry,
495 					attr->mode, &dentry->d_name, old_cred,
496 					override_cred);
497 			if (err) {
498 				put_cred(override_cred);
499 				goto out_revert_creds;
500 			}
501 		}
502 		put_cred(override_creds(override_cred));
503 		put_cred(override_cred);
504 
505 		if (!ovl_dentry_is_whiteout(dentry))
506 			err = ovl_create_upper(dentry, inode, attr,
507 						hardlink);
508 		else
509 			err = ovl_create_over_whiteout(dentry, inode, attr,
510 							hardlink);
511 	}
512 out_revert_creds:
513 	revert_creds(old_cred);
514 	if (!err) {
515 		struct inode *realinode = d_inode(ovl_dentry_upper(dentry));
516 
517 		WARN_ON(inode->i_mode != realinode->i_mode);
518 		WARN_ON(!uid_eq(inode->i_uid, realinode->i_uid));
519 		WARN_ON(!gid_eq(inode->i_gid, realinode->i_gid));
520 	}
521 	return err;
522 }
523 
524 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
525 			     const char *link)
526 {
527 	int err;
528 	struct inode *inode;
529 	struct cattr attr = {
530 		.rdev = rdev,
531 		.link = link,
532 	};
533 
534 	err = ovl_want_write(dentry);
535 	if (err)
536 		goto out;
537 
538 	err = -ENOMEM;
539 	inode = ovl_new_inode(dentry->d_sb, mode, rdev);
540 	if (!inode)
541 		goto out_drop_write;
542 
543 	inode_init_owner(inode, dentry->d_parent->d_inode, mode);
544 	attr.mode = inode->i_mode;
545 
546 	err = ovl_create_or_link(dentry, inode, &attr, NULL, false);
547 	if (err)
548 		iput(inode);
549 
550 out_drop_write:
551 	ovl_drop_write(dentry);
552 out:
553 	return err;
554 }
555 
556 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
557 		      bool excl)
558 {
559 	return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
560 }
561 
562 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
563 {
564 	return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
565 }
566 
567 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
568 		     dev_t rdev)
569 {
570 	/* Don't allow creation of "whiteout" on overlay */
571 	if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
572 		return -EPERM;
573 
574 	return ovl_create_object(dentry, mode, rdev, NULL);
575 }
576 
577 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
578 		       const char *link)
579 {
580 	return ovl_create_object(dentry, S_IFLNK, 0, link);
581 }
582 
583 static int ovl_link(struct dentry *old, struct inode *newdir,
584 		    struct dentry *new)
585 {
586 	int err;
587 	bool locked = false;
588 	struct inode *inode;
589 
590 	err = ovl_want_write(old);
591 	if (err)
592 		goto out;
593 
594 	err = ovl_copy_up(old);
595 	if (err)
596 		goto out_drop_write;
597 
598 	err = ovl_nlink_start(old, &locked);
599 	if (err)
600 		goto out_drop_write;
601 
602 	inode = d_inode(old);
603 	ihold(inode);
604 
605 	err = ovl_create_or_link(new, inode, NULL, ovl_dentry_upper(old),
606 				 ovl_type_origin(old));
607 	if (err)
608 		iput(inode);
609 
610 	ovl_nlink_end(old, locked);
611 out_drop_write:
612 	ovl_drop_write(old);
613 out:
614 	return err;
615 }
616 
617 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
618 {
619 	return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
620 }
621 
622 static int ovl_remove_and_whiteout(struct dentry *dentry, bool is_dir)
623 {
624 	struct dentry *workdir = ovl_workdir(dentry);
625 	struct inode *wdir = workdir->d_inode;
626 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
627 	struct inode *udir = upperdir->d_inode;
628 	struct dentry *whiteout;
629 	struct dentry *upper;
630 	struct dentry *opaquedir = NULL;
631 	int err;
632 	int flags = 0;
633 
634 	if (WARN_ON(!workdir))
635 		return -EROFS;
636 
637 	if (is_dir) {
638 		opaquedir = ovl_check_empty_and_clear(dentry);
639 		err = PTR_ERR(opaquedir);
640 		if (IS_ERR(opaquedir))
641 			goto out;
642 	}
643 
644 	err = ovl_lock_rename_workdir(workdir, upperdir);
645 	if (err)
646 		goto out_dput;
647 
648 	upper = lookup_one_len(dentry->d_name.name, upperdir,
649 			       dentry->d_name.len);
650 	err = PTR_ERR(upper);
651 	if (IS_ERR(upper))
652 		goto out_unlock;
653 
654 	err = -ESTALE;
655 	if ((opaquedir && upper != opaquedir) ||
656 	    (!opaquedir && ovl_dentry_upper(dentry) &&
657 	     !ovl_matches_upper(dentry, upper))) {
658 		goto out_dput_upper;
659 	}
660 
661 	whiteout = ovl_whiteout(workdir, dentry);
662 	err = PTR_ERR(whiteout);
663 	if (IS_ERR(whiteout))
664 		goto out_dput_upper;
665 
666 	if (d_is_dir(upper))
667 		flags = RENAME_EXCHANGE;
668 
669 	err = ovl_do_rename(wdir, whiteout, udir, upper, flags);
670 	if (err)
671 		goto kill_whiteout;
672 	if (flags)
673 		ovl_cleanup(wdir, upper);
674 
675 	ovl_dentry_version_inc(dentry->d_parent, true);
676 out_d_drop:
677 	d_drop(dentry);
678 	dput(whiteout);
679 out_dput_upper:
680 	dput(upper);
681 out_unlock:
682 	unlock_rename(workdir, upperdir);
683 out_dput:
684 	dput(opaquedir);
685 out:
686 	return err;
687 
688 kill_whiteout:
689 	ovl_cleanup(wdir, whiteout);
690 	goto out_d_drop;
691 }
692 
693 static int ovl_remove_upper(struct dentry *dentry, bool is_dir)
694 {
695 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
696 	struct inode *dir = upperdir->d_inode;
697 	struct dentry *upper;
698 	struct dentry *opaquedir = NULL;
699 	int err;
700 
701 	/* Redirect dir can be !ovl_lower_positive && OVL_TYPE_MERGE */
702 	if (is_dir && ovl_dentry_get_redirect(dentry)) {
703 		opaquedir = ovl_check_empty_and_clear(dentry);
704 		err = PTR_ERR(opaquedir);
705 		if (IS_ERR(opaquedir))
706 			goto out;
707 	}
708 
709 	inode_lock_nested(dir, I_MUTEX_PARENT);
710 	upper = lookup_one_len(dentry->d_name.name, upperdir,
711 			       dentry->d_name.len);
712 	err = PTR_ERR(upper);
713 	if (IS_ERR(upper))
714 		goto out_unlock;
715 
716 	err = -ESTALE;
717 	if ((opaquedir && upper != opaquedir) ||
718 	    (!opaquedir && !ovl_matches_upper(dentry, upper)))
719 		goto out_dput_upper;
720 
721 	if (is_dir)
722 		err = vfs_rmdir(dir, upper);
723 	else
724 		err = vfs_unlink(dir, upper, NULL);
725 	ovl_dentry_version_inc(dentry->d_parent, ovl_type_origin(dentry));
726 
727 	/*
728 	 * Keeping this dentry hashed would mean having to release
729 	 * upperpath/lowerpath, which could only be done if we are the
730 	 * sole user of this dentry.  Too tricky...  Just unhash for
731 	 * now.
732 	 */
733 	if (!err)
734 		d_drop(dentry);
735 out_dput_upper:
736 	dput(upper);
737 out_unlock:
738 	inode_unlock(dir);
739 	dput(opaquedir);
740 out:
741 	return err;
742 }
743 
744 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
745 {
746 	int err;
747 	bool locked = false;
748 	const struct cred *old_cred;
749 
750 	err = ovl_want_write(dentry);
751 	if (err)
752 		goto out;
753 
754 	err = ovl_copy_up(dentry->d_parent);
755 	if (err)
756 		goto out_drop_write;
757 
758 	err = ovl_nlink_start(dentry, &locked);
759 	if (err)
760 		goto out_drop_write;
761 
762 	old_cred = ovl_override_creds(dentry->d_sb);
763 	if (!ovl_lower_positive(dentry))
764 		err = ovl_remove_upper(dentry, is_dir);
765 	else
766 		err = ovl_remove_and_whiteout(dentry, is_dir);
767 	revert_creds(old_cred);
768 	if (!err) {
769 		if (is_dir)
770 			clear_nlink(dentry->d_inode);
771 		else
772 			drop_nlink(dentry->d_inode);
773 	}
774 	ovl_nlink_end(dentry, locked);
775 out_drop_write:
776 	ovl_drop_write(dentry);
777 out:
778 	return err;
779 }
780 
781 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
782 {
783 	return ovl_do_remove(dentry, false);
784 }
785 
786 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
787 {
788 	return ovl_do_remove(dentry, true);
789 }
790 
791 static bool ovl_type_merge_or_lower(struct dentry *dentry)
792 {
793 	enum ovl_path_type type = ovl_path_type(dentry);
794 
795 	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
796 }
797 
798 static bool ovl_can_move(struct dentry *dentry)
799 {
800 	return ovl_redirect_dir(dentry->d_sb) ||
801 		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
802 }
803 
804 static char *ovl_get_redirect(struct dentry *dentry, bool samedir)
805 {
806 	char *buf, *ret;
807 	struct dentry *d, *tmp;
808 	int buflen = ovl_redirect_max + 1;
809 
810 	if (samedir) {
811 		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
812 			       GFP_KERNEL);
813 		goto out;
814 	}
815 
816 	buf = ret = kmalloc(buflen, GFP_KERNEL);
817 	if (!buf)
818 		goto out;
819 
820 	buflen--;
821 	buf[buflen] = '\0';
822 	for (d = dget(dentry); !IS_ROOT(d);) {
823 		const char *name;
824 		int thislen;
825 
826 		spin_lock(&d->d_lock);
827 		name = ovl_dentry_get_redirect(d);
828 		if (name) {
829 			thislen = strlen(name);
830 		} else {
831 			name = d->d_name.name;
832 			thislen = d->d_name.len;
833 		}
834 
835 		/* If path is too long, fall back to userspace move */
836 		if (thislen + (name[0] != '/') > buflen) {
837 			ret = ERR_PTR(-EXDEV);
838 			spin_unlock(&d->d_lock);
839 			goto out_put;
840 		}
841 
842 		buflen -= thislen;
843 		memcpy(&buf[buflen], name, thislen);
844 		tmp = dget_dlock(d->d_parent);
845 		spin_unlock(&d->d_lock);
846 
847 		dput(d);
848 		d = tmp;
849 
850 		/* Absolute redirect: finished */
851 		if (buf[buflen] == '/')
852 			break;
853 		buflen--;
854 		buf[buflen] = '/';
855 	}
856 	ret = kstrdup(&buf[buflen], GFP_KERNEL);
857 out_put:
858 	dput(d);
859 	kfree(buf);
860 out:
861 	return ret ? ret : ERR_PTR(-ENOMEM);
862 }
863 
864 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
865 {
866 	int err;
867 	const char *redirect = ovl_dentry_get_redirect(dentry);
868 
869 	if (redirect && (samedir || redirect[0] == '/'))
870 		return 0;
871 
872 	redirect = ovl_get_redirect(dentry, samedir);
873 	if (IS_ERR(redirect))
874 		return PTR_ERR(redirect);
875 
876 	err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
877 				 OVL_XATTR_REDIRECT,
878 				 redirect, strlen(redirect), -EXDEV);
879 	if (!err) {
880 		spin_lock(&dentry->d_lock);
881 		ovl_dentry_set_redirect(dentry, redirect);
882 		spin_unlock(&dentry->d_lock);
883 	} else {
884 		kfree(redirect);
885 		pr_warn_ratelimited("overlay: failed to set redirect (%i)\n", err);
886 		/* Fall back to userspace copy-up */
887 		err = -EXDEV;
888 	}
889 	return err;
890 }
891 
892 static int ovl_rename(struct inode *olddir, struct dentry *old,
893 		      struct inode *newdir, struct dentry *new,
894 		      unsigned int flags)
895 {
896 	int err;
897 	bool locked = false;
898 	struct dentry *old_upperdir;
899 	struct dentry *new_upperdir;
900 	struct dentry *olddentry;
901 	struct dentry *newdentry;
902 	struct dentry *trap;
903 	bool old_opaque;
904 	bool new_opaque;
905 	bool cleanup_whiteout = false;
906 	bool overwrite = !(flags & RENAME_EXCHANGE);
907 	bool is_dir = d_is_dir(old);
908 	bool new_is_dir = d_is_dir(new);
909 	bool samedir = olddir == newdir;
910 	struct dentry *opaquedir = NULL;
911 	const struct cred *old_cred = NULL;
912 
913 	err = -EINVAL;
914 	if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
915 		goto out;
916 
917 	flags &= ~RENAME_NOREPLACE;
918 
919 	/* Don't copy up directory trees */
920 	err = -EXDEV;
921 	if (!ovl_can_move(old))
922 		goto out;
923 	if (!overwrite && !ovl_can_move(new))
924 		goto out;
925 
926 	err = ovl_want_write(old);
927 	if (err)
928 		goto out;
929 
930 	err = ovl_copy_up(old);
931 	if (err)
932 		goto out_drop_write;
933 
934 	err = ovl_copy_up(new->d_parent);
935 	if (err)
936 		goto out_drop_write;
937 	if (!overwrite) {
938 		err = ovl_copy_up(new);
939 		if (err)
940 			goto out_drop_write;
941 	} else {
942 		err = ovl_nlink_start(new, &locked);
943 		if (err)
944 			goto out_drop_write;
945 	}
946 
947 	old_cred = ovl_override_creds(old->d_sb);
948 
949 	if (overwrite && new_is_dir && ovl_type_merge_or_lower(new)) {
950 		opaquedir = ovl_check_empty_and_clear(new);
951 		err = PTR_ERR(opaquedir);
952 		if (IS_ERR(opaquedir)) {
953 			opaquedir = NULL;
954 			goto out_revert_creds;
955 		}
956 	}
957 
958 	if (overwrite) {
959 		if (ovl_lower_positive(old)) {
960 			if (!ovl_dentry_is_whiteout(new)) {
961 				/* Whiteout source */
962 				flags |= RENAME_WHITEOUT;
963 			} else {
964 				/* Switch whiteouts */
965 				flags |= RENAME_EXCHANGE;
966 			}
967 		} else if (is_dir && ovl_dentry_is_whiteout(new)) {
968 			flags |= RENAME_EXCHANGE;
969 			cleanup_whiteout = true;
970 		}
971 	}
972 
973 	old_upperdir = ovl_dentry_upper(old->d_parent);
974 	new_upperdir = ovl_dentry_upper(new->d_parent);
975 
976 	if (!samedir) {
977 		/*
978 		 * When moving a merge dir or non-dir with copy up origin into
979 		 * a new parent, we are marking the new parent dir "impure".
980 		 * When ovl_iterate() iterates an "impure" upper dir, it will
981 		 * lookup the origin inodes of the entries to fill d_ino.
982 		 */
983 		if (ovl_type_origin(old)) {
984 			err = ovl_set_impure(new->d_parent, new_upperdir);
985 			if (err)
986 				goto out_revert_creds;
987 		}
988 		if (!overwrite && ovl_type_origin(new)) {
989 			err = ovl_set_impure(old->d_parent, old_upperdir);
990 			if (err)
991 				goto out_revert_creds;
992 		}
993 	}
994 
995 	trap = lock_rename(new_upperdir, old_upperdir);
996 
997 	olddentry = lookup_one_len(old->d_name.name, old_upperdir,
998 				   old->d_name.len);
999 	err = PTR_ERR(olddentry);
1000 	if (IS_ERR(olddentry))
1001 		goto out_unlock;
1002 
1003 	err = -ESTALE;
1004 	if (!ovl_matches_upper(old, olddentry))
1005 		goto out_dput_old;
1006 
1007 	newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1008 				   new->d_name.len);
1009 	err = PTR_ERR(newdentry);
1010 	if (IS_ERR(newdentry))
1011 		goto out_dput_old;
1012 
1013 	old_opaque = ovl_dentry_is_opaque(old);
1014 	new_opaque = ovl_dentry_is_opaque(new);
1015 
1016 	err = -ESTALE;
1017 	if (d_inode(new) && ovl_dentry_upper(new)) {
1018 		if (opaquedir) {
1019 			if (newdentry != opaquedir)
1020 				goto out_dput;
1021 		} else {
1022 			if (!ovl_matches_upper(new, newdentry))
1023 				goto out_dput;
1024 		}
1025 	} else {
1026 		if (!d_is_negative(newdentry) &&
1027 		    (!new_opaque || !ovl_is_whiteout(newdentry)))
1028 			goto out_dput;
1029 	}
1030 
1031 	if (olddentry == trap)
1032 		goto out_dput;
1033 	if (newdentry == trap)
1034 		goto out_dput;
1035 
1036 	if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
1037 		goto out_dput;
1038 
1039 	err = 0;
1040 	if (is_dir) {
1041 		if (ovl_type_merge_or_lower(old))
1042 			err = ovl_set_redirect(old, samedir);
1043 		else if (!old_opaque && ovl_type_merge(new->d_parent))
1044 			err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1045 		if (err)
1046 			goto out_dput;
1047 	}
1048 	if (!overwrite && new_is_dir) {
1049 		if (ovl_type_merge_or_lower(new))
1050 			err = ovl_set_redirect(new, samedir);
1051 		else if (!new_opaque && ovl_type_merge(old->d_parent))
1052 			err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1053 		if (err)
1054 			goto out_dput;
1055 	}
1056 
1057 	err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1058 			    new_upperdir->d_inode, newdentry, flags);
1059 	if (err)
1060 		goto out_dput;
1061 
1062 	if (cleanup_whiteout)
1063 		ovl_cleanup(old_upperdir->d_inode, newdentry);
1064 
1065 	if (overwrite && d_inode(new)) {
1066 		if (new_is_dir)
1067 			clear_nlink(d_inode(new));
1068 		else
1069 			drop_nlink(d_inode(new));
1070 	}
1071 
1072 	ovl_dentry_version_inc(old->d_parent,
1073 			       !overwrite && ovl_type_origin(new));
1074 	ovl_dentry_version_inc(new->d_parent, ovl_type_origin(old));
1075 
1076 out_dput:
1077 	dput(newdentry);
1078 out_dput_old:
1079 	dput(olddentry);
1080 out_unlock:
1081 	unlock_rename(new_upperdir, old_upperdir);
1082 out_revert_creds:
1083 	revert_creds(old_cred);
1084 	ovl_nlink_end(new, locked);
1085 out_drop_write:
1086 	ovl_drop_write(old);
1087 out:
1088 	dput(opaquedir);
1089 	return err;
1090 }
1091 
1092 const struct inode_operations ovl_dir_inode_operations = {
1093 	.lookup		= ovl_lookup,
1094 	.mkdir		= ovl_mkdir,
1095 	.symlink	= ovl_symlink,
1096 	.unlink		= ovl_unlink,
1097 	.rmdir		= ovl_rmdir,
1098 	.rename		= ovl_rename,
1099 	.link		= ovl_link,
1100 	.setattr	= ovl_setattr,
1101 	.create		= ovl_create,
1102 	.mknod		= ovl_mknod,
1103 	.permission	= ovl_permission,
1104 	.getattr	= ovl_getattr,
1105 	.listxattr	= ovl_listxattr,
1106 	.get_acl	= ovl_get_acl,
1107 	.update_time	= ovl_update_time,
1108 };
1109