xref: /openbmc/linux/fs/overlayfs/dir.c (revision 4da722ca)
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);
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 int ovl_lock_rename_workdir(struct dentry *workdir,
220 				   struct dentry *upperdir)
221 {
222 	/* Workdir should not be the same as upperdir */
223 	if (workdir == upperdir)
224 		goto err;
225 
226 	/* Workdir should not be subdir of upperdir and vice versa */
227 	if (lock_rename(workdir, upperdir) != NULL)
228 		goto err_unlock;
229 
230 	return 0;
231 
232 err_unlock:
233 	unlock_rename(workdir, upperdir);
234 err:
235 	pr_err("overlayfs: failed to lock workdir+upperdir\n");
236 	return -EIO;
237 }
238 
239 static struct dentry *ovl_clear_empty(struct dentry *dentry,
240 				      struct list_head *list)
241 {
242 	struct dentry *workdir = ovl_workdir(dentry);
243 	struct inode *wdir = workdir->d_inode;
244 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
245 	struct inode *udir = upperdir->d_inode;
246 	struct path upperpath;
247 	struct dentry *upper;
248 	struct dentry *opaquedir;
249 	struct kstat stat;
250 	int err;
251 
252 	if (WARN_ON(!workdir))
253 		return ERR_PTR(-EROFS);
254 
255 	err = ovl_lock_rename_workdir(workdir, upperdir);
256 	if (err)
257 		goto out;
258 
259 	ovl_path_upper(dentry, &upperpath);
260 	err = vfs_getattr(&upperpath, &stat,
261 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
262 	if (err)
263 		goto out_unlock;
264 
265 	err = -ESTALE;
266 	if (!S_ISDIR(stat.mode))
267 		goto out_unlock;
268 	upper = upperpath.dentry;
269 	if (upper->d_parent->d_inode != udir)
270 		goto out_unlock;
271 
272 	opaquedir = ovl_lookup_temp(workdir);
273 	err = PTR_ERR(opaquedir);
274 	if (IS_ERR(opaquedir))
275 		goto out_unlock;
276 
277 	err = ovl_create_real(wdir, opaquedir,
278 			      &(struct cattr){.mode = stat.mode}, NULL, true);
279 	if (err)
280 		goto out_dput;
281 
282 	err = ovl_copy_xattr(upper, opaquedir);
283 	if (err)
284 		goto out_cleanup;
285 
286 	err = ovl_set_opaque(dentry, opaquedir);
287 	if (err)
288 		goto out_cleanup;
289 
290 	inode_lock(opaquedir->d_inode);
291 	err = ovl_set_attr(opaquedir, &stat);
292 	inode_unlock(opaquedir->d_inode);
293 	if (err)
294 		goto out_cleanup;
295 
296 	err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
297 	if (err)
298 		goto out_cleanup;
299 
300 	ovl_cleanup_whiteouts(upper, list);
301 	ovl_cleanup(wdir, upper);
302 	unlock_rename(workdir, upperdir);
303 
304 	/* dentry's upper doesn't match now, get rid of it */
305 	d_drop(dentry);
306 
307 	return opaquedir;
308 
309 out_cleanup:
310 	ovl_cleanup(wdir, opaquedir);
311 out_dput:
312 	dput(opaquedir);
313 out_unlock:
314 	unlock_rename(workdir, upperdir);
315 out:
316 	return ERR_PTR(err);
317 }
318 
319 static struct dentry *ovl_check_empty_and_clear(struct dentry *dentry)
320 {
321 	int err;
322 	struct dentry *ret = NULL;
323 	enum ovl_path_type type = ovl_path_type(dentry);
324 	LIST_HEAD(list);
325 
326 	err = ovl_check_empty_dir(dentry, &list);
327 	if (err) {
328 		ret = ERR_PTR(err);
329 		goto out_free;
330 	}
331 
332 	/*
333 	 * When removing an empty opaque directory, then it makes no sense to
334 	 * replace it with an exact replica of itself.
335 	 *
336 	 * If no upperdentry then skip clearing whiteouts.
337 	 *
338 	 * Can race with copy-up, since we don't hold the upperdir mutex.
339 	 * Doesn't matter, since copy-up can't create a non-empty directory
340 	 * from an empty one.
341 	 */
342 	if (OVL_TYPE_UPPER(type) && OVL_TYPE_MERGE(type))
343 		ret = ovl_clear_empty(dentry, &list);
344 
345 out_free:
346 	ovl_cache_free(&list);
347 
348 	return ret;
349 }
350 
351 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
352 			     const struct posix_acl *acl)
353 {
354 	void *buffer;
355 	size_t size;
356 	int err;
357 
358 	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
359 		return 0;
360 
361 	size = posix_acl_to_xattr(NULL, acl, NULL, 0);
362 	buffer = kmalloc(size, GFP_KERNEL);
363 	if (!buffer)
364 		return -ENOMEM;
365 
366 	size = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
367 	err = size;
368 	if (err < 0)
369 		goto out_free;
370 
371 	err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
372 out_free:
373 	kfree(buffer);
374 	return err;
375 }
376 
377 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
378 				    struct cattr *cattr,
379 				    struct dentry *hardlink)
380 {
381 	struct dentry *workdir = ovl_workdir(dentry);
382 	struct inode *wdir = workdir->d_inode;
383 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
384 	struct inode *udir = upperdir->d_inode;
385 	struct dentry *upper;
386 	struct dentry *newdentry;
387 	int err;
388 	struct posix_acl *acl, *default_acl;
389 
390 	if (WARN_ON(!workdir))
391 		return -EROFS;
392 
393 	if (!hardlink) {
394 		err = posix_acl_create(dentry->d_parent->d_inode,
395 				       &cattr->mode, &default_acl, &acl);
396 		if (err)
397 			return err;
398 	}
399 
400 	err = ovl_lock_rename_workdir(workdir, upperdir);
401 	if (err)
402 		goto out;
403 
404 	newdentry = ovl_lookup_temp(workdir);
405 	err = PTR_ERR(newdentry);
406 	if (IS_ERR(newdentry))
407 		goto out_unlock;
408 
409 	upper = lookup_one_len(dentry->d_name.name, upperdir,
410 			       dentry->d_name.len);
411 	err = PTR_ERR(upper);
412 	if (IS_ERR(upper))
413 		goto out_dput;
414 
415 	err = ovl_create_real(wdir, newdentry, cattr, hardlink, true);
416 	if (err)
417 		goto out_dput2;
418 
419 	/*
420 	 * mode could have been mutilated due to umask (e.g. sgid directory)
421 	 */
422 	if (!hardlink &&
423 	    !S_ISLNK(cattr->mode) &&
424 	    newdentry->d_inode->i_mode != cattr->mode) {
425 		struct iattr attr = {
426 			.ia_valid = ATTR_MODE,
427 			.ia_mode = cattr->mode,
428 		};
429 		inode_lock(newdentry->d_inode);
430 		err = notify_change(newdentry, &attr, NULL);
431 		inode_unlock(newdentry->d_inode);
432 		if (err)
433 			goto out_cleanup;
434 	}
435 	if (!hardlink) {
436 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
437 					acl);
438 		if (err)
439 			goto out_cleanup;
440 
441 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
442 					default_acl);
443 		if (err)
444 			goto out_cleanup;
445 	}
446 
447 	if (!hardlink && S_ISDIR(cattr->mode)) {
448 		err = ovl_set_opaque(dentry, newdentry);
449 		if (err)
450 			goto out_cleanup;
451 
452 		err = ovl_do_rename(wdir, newdentry, udir, upper,
453 				    RENAME_EXCHANGE);
454 		if (err)
455 			goto out_cleanup;
456 
457 		ovl_cleanup(wdir, upper);
458 	} else {
459 		err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
460 		if (err)
461 			goto out_cleanup;
462 	}
463 	ovl_instantiate(dentry, inode, newdentry, !!hardlink);
464 	newdentry = NULL;
465 out_dput2:
466 	dput(upper);
467 out_dput:
468 	dput(newdentry);
469 out_unlock:
470 	unlock_rename(workdir, upperdir);
471 out:
472 	if (!hardlink) {
473 		posix_acl_release(acl);
474 		posix_acl_release(default_acl);
475 	}
476 	return err;
477 
478 out_cleanup:
479 	ovl_cleanup(wdir, newdentry);
480 	goto out_dput2;
481 }
482 
483 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
484 			      struct cattr *attr, struct dentry *hardlink)
485 {
486 	int err;
487 	const struct cred *old_cred;
488 	struct cred *override_cred;
489 
490 	err = ovl_copy_up(dentry->d_parent);
491 	if (err)
492 		return err;
493 
494 	old_cred = ovl_override_creds(dentry->d_sb);
495 	err = -ENOMEM;
496 	override_cred = prepare_creds();
497 	if (override_cred) {
498 		override_cred->fsuid = inode->i_uid;
499 		override_cred->fsgid = inode->i_gid;
500 		if (!hardlink) {
501 			err = security_dentry_create_files_as(dentry,
502 					attr->mode, &dentry->d_name, old_cred,
503 					override_cred);
504 			if (err) {
505 				put_cred(override_cred);
506 				goto out_revert_creds;
507 			}
508 		}
509 		put_cred(override_creds(override_cred));
510 		put_cred(override_cred);
511 
512 		if (!ovl_dentry_is_whiteout(dentry))
513 			err = ovl_create_upper(dentry, inode, attr,
514 						hardlink);
515 		else
516 			err = ovl_create_over_whiteout(dentry, inode, attr,
517 							hardlink);
518 	}
519 out_revert_creds:
520 	revert_creds(old_cred);
521 	if (!err) {
522 		struct inode *realinode = d_inode(ovl_dentry_upper(dentry));
523 
524 		WARN_ON(inode->i_mode != realinode->i_mode);
525 		WARN_ON(!uid_eq(inode->i_uid, realinode->i_uid));
526 		WARN_ON(!gid_eq(inode->i_gid, realinode->i_gid));
527 	}
528 	return err;
529 }
530 
531 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
532 			     const char *link)
533 {
534 	int err;
535 	struct inode *inode;
536 	struct cattr attr = {
537 		.rdev = rdev,
538 		.link = link,
539 	};
540 
541 	err = ovl_want_write(dentry);
542 	if (err)
543 		goto out;
544 
545 	err = -ENOMEM;
546 	inode = ovl_new_inode(dentry->d_sb, mode, rdev);
547 	if (!inode)
548 		goto out_drop_write;
549 
550 	inode_init_owner(inode, dentry->d_parent->d_inode, mode);
551 	attr.mode = inode->i_mode;
552 
553 	err = ovl_create_or_link(dentry, inode, &attr, NULL);
554 	if (err)
555 		iput(inode);
556 
557 out_drop_write:
558 	ovl_drop_write(dentry);
559 out:
560 	return err;
561 }
562 
563 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
564 		      bool excl)
565 {
566 	return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
567 }
568 
569 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
570 {
571 	return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
572 }
573 
574 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
575 		     dev_t rdev)
576 {
577 	/* Don't allow creation of "whiteout" on overlay */
578 	if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
579 		return -EPERM;
580 
581 	return ovl_create_object(dentry, mode, rdev, NULL);
582 }
583 
584 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
585 		       const char *link)
586 {
587 	return ovl_create_object(dentry, S_IFLNK, 0, link);
588 }
589 
590 static int ovl_link(struct dentry *old, struct inode *newdir,
591 		    struct dentry *new)
592 {
593 	int err;
594 	bool locked = false;
595 	struct inode *inode;
596 
597 	err = ovl_want_write(old);
598 	if (err)
599 		goto out;
600 
601 	err = ovl_copy_up(old);
602 	if (err)
603 		goto out_drop_write;
604 
605 	err = ovl_nlink_start(old, &locked);
606 	if (err)
607 		goto out_drop_write;
608 
609 	inode = d_inode(old);
610 	ihold(inode);
611 
612 	err = ovl_create_or_link(new, inode, NULL, ovl_dentry_upper(old));
613 	if (err)
614 		iput(inode);
615 
616 	ovl_nlink_end(old, locked);
617 out_drop_write:
618 	ovl_drop_write(old);
619 out:
620 	return err;
621 }
622 
623 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
624 {
625 	return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
626 }
627 
628 static int ovl_remove_and_whiteout(struct dentry *dentry, bool is_dir)
629 {
630 	struct dentry *workdir = ovl_workdir(dentry);
631 	struct inode *wdir = workdir->d_inode;
632 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
633 	struct inode *udir = upperdir->d_inode;
634 	struct dentry *whiteout;
635 	struct dentry *upper;
636 	struct dentry *opaquedir = NULL;
637 	int err;
638 	int flags = 0;
639 
640 	if (WARN_ON(!workdir))
641 		return -EROFS;
642 
643 	if (is_dir) {
644 		opaquedir = ovl_check_empty_and_clear(dentry);
645 		err = PTR_ERR(opaquedir);
646 		if (IS_ERR(opaquedir))
647 			goto out;
648 	}
649 
650 	err = ovl_lock_rename_workdir(workdir, upperdir);
651 	if (err)
652 		goto out_dput;
653 
654 	upper = lookup_one_len(dentry->d_name.name, upperdir,
655 			       dentry->d_name.len);
656 	err = PTR_ERR(upper);
657 	if (IS_ERR(upper))
658 		goto out_unlock;
659 
660 	err = -ESTALE;
661 	if ((opaquedir && upper != opaquedir) ||
662 	    (!opaquedir && ovl_dentry_upper(dentry) &&
663 	     !ovl_matches_upper(dentry, upper))) {
664 		goto out_dput_upper;
665 	}
666 
667 	whiteout = ovl_whiteout(workdir, dentry);
668 	err = PTR_ERR(whiteout);
669 	if (IS_ERR(whiteout))
670 		goto out_dput_upper;
671 
672 	if (d_is_dir(upper))
673 		flags = RENAME_EXCHANGE;
674 
675 	err = ovl_do_rename(wdir, whiteout, udir, upper, flags);
676 	if (err)
677 		goto kill_whiteout;
678 	if (flags)
679 		ovl_cleanup(wdir, upper);
680 
681 	ovl_dentry_version_inc(dentry->d_parent);
682 out_d_drop:
683 	d_drop(dentry);
684 	dput(whiteout);
685 out_dput_upper:
686 	dput(upper);
687 out_unlock:
688 	unlock_rename(workdir, upperdir);
689 out_dput:
690 	dput(opaquedir);
691 out:
692 	return err;
693 
694 kill_whiteout:
695 	ovl_cleanup(wdir, whiteout);
696 	goto out_d_drop;
697 }
698 
699 static int ovl_remove_upper(struct dentry *dentry, bool is_dir)
700 {
701 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
702 	struct inode *dir = upperdir->d_inode;
703 	struct dentry *upper;
704 	struct dentry *opaquedir = NULL;
705 	int err;
706 
707 	/* Redirect dir can be !ovl_lower_positive && OVL_TYPE_MERGE */
708 	if (is_dir && ovl_dentry_get_redirect(dentry)) {
709 		opaquedir = ovl_check_empty_and_clear(dentry);
710 		err = PTR_ERR(opaquedir);
711 		if (IS_ERR(opaquedir))
712 			goto out;
713 	}
714 
715 	inode_lock_nested(dir, I_MUTEX_PARENT);
716 	upper = lookup_one_len(dentry->d_name.name, upperdir,
717 			       dentry->d_name.len);
718 	err = PTR_ERR(upper);
719 	if (IS_ERR(upper))
720 		goto out_unlock;
721 
722 	err = -ESTALE;
723 	if ((opaquedir && upper != opaquedir) ||
724 	    (!opaquedir && !ovl_matches_upper(dentry, upper)))
725 		goto out_dput_upper;
726 
727 	if (is_dir)
728 		err = vfs_rmdir(dir, upper);
729 	else
730 		err = vfs_unlink(dir, upper, NULL);
731 	ovl_dentry_version_inc(dentry->d_parent);
732 
733 	/*
734 	 * Keeping this dentry hashed would mean having to release
735 	 * upperpath/lowerpath, which could only be done if we are the
736 	 * sole user of this dentry.  Too tricky...  Just unhash for
737 	 * now.
738 	 */
739 	if (!err)
740 		d_drop(dentry);
741 out_dput_upper:
742 	dput(upper);
743 out_unlock:
744 	inode_unlock(dir);
745 	dput(opaquedir);
746 out:
747 	return err;
748 }
749 
750 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
751 {
752 	int err;
753 	bool locked = false;
754 	const struct cred *old_cred;
755 
756 	err = ovl_want_write(dentry);
757 	if (err)
758 		goto out;
759 
760 	err = ovl_copy_up(dentry->d_parent);
761 	if (err)
762 		goto out_drop_write;
763 
764 	err = ovl_nlink_start(dentry, &locked);
765 	if (err)
766 		goto out_drop_write;
767 
768 	old_cred = ovl_override_creds(dentry->d_sb);
769 	if (!ovl_lower_positive(dentry))
770 		err = ovl_remove_upper(dentry, is_dir);
771 	else
772 		err = ovl_remove_and_whiteout(dentry, is_dir);
773 	revert_creds(old_cred);
774 	if (!err) {
775 		if (is_dir)
776 			clear_nlink(dentry->d_inode);
777 		else
778 			drop_nlink(dentry->d_inode);
779 	}
780 	ovl_nlink_end(dentry, locked);
781 out_drop_write:
782 	ovl_drop_write(dentry);
783 out:
784 	return err;
785 }
786 
787 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
788 {
789 	return ovl_do_remove(dentry, false);
790 }
791 
792 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
793 {
794 	return ovl_do_remove(dentry, true);
795 }
796 
797 static bool ovl_type_merge_or_lower(struct dentry *dentry)
798 {
799 	enum ovl_path_type type = ovl_path_type(dentry);
800 
801 	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
802 }
803 
804 static bool ovl_can_move(struct dentry *dentry)
805 {
806 	return ovl_redirect_dir(dentry->d_sb) ||
807 		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
808 }
809 
810 static char *ovl_get_redirect(struct dentry *dentry, bool samedir)
811 {
812 	char *buf, *ret;
813 	struct dentry *d, *tmp;
814 	int buflen = ovl_redirect_max + 1;
815 
816 	if (samedir) {
817 		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
818 			       GFP_KERNEL);
819 		goto out;
820 	}
821 
822 	buf = ret = kmalloc(buflen, GFP_TEMPORARY);
823 	if (!buf)
824 		goto out;
825 
826 	buflen--;
827 	buf[buflen] = '\0';
828 	for (d = dget(dentry); !IS_ROOT(d);) {
829 		const char *name;
830 		int thislen;
831 
832 		spin_lock(&d->d_lock);
833 		name = ovl_dentry_get_redirect(d);
834 		if (name) {
835 			thislen = strlen(name);
836 		} else {
837 			name = d->d_name.name;
838 			thislen = d->d_name.len;
839 		}
840 
841 		/* If path is too long, fall back to userspace move */
842 		if (thislen + (name[0] != '/') > buflen) {
843 			ret = ERR_PTR(-EXDEV);
844 			spin_unlock(&d->d_lock);
845 			goto out_put;
846 		}
847 
848 		buflen -= thislen;
849 		memcpy(&buf[buflen], name, thislen);
850 		tmp = dget_dlock(d->d_parent);
851 		spin_unlock(&d->d_lock);
852 
853 		dput(d);
854 		d = tmp;
855 
856 		/* Absolute redirect: finished */
857 		if (buf[buflen] == '/')
858 			break;
859 		buflen--;
860 		buf[buflen] = '/';
861 	}
862 	ret = kstrdup(&buf[buflen], GFP_KERNEL);
863 out_put:
864 	dput(d);
865 	kfree(buf);
866 out:
867 	return ret ? ret : ERR_PTR(-ENOMEM);
868 }
869 
870 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
871 {
872 	int err;
873 	const char *redirect = ovl_dentry_get_redirect(dentry);
874 
875 	if (redirect && (samedir || redirect[0] == '/'))
876 		return 0;
877 
878 	redirect = ovl_get_redirect(dentry, samedir);
879 	if (IS_ERR(redirect))
880 		return PTR_ERR(redirect);
881 
882 	err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
883 				 OVL_XATTR_REDIRECT,
884 				 redirect, strlen(redirect), -EXDEV);
885 	if (!err) {
886 		spin_lock(&dentry->d_lock);
887 		ovl_dentry_set_redirect(dentry, redirect);
888 		spin_unlock(&dentry->d_lock);
889 	} else {
890 		kfree(redirect);
891 		pr_warn_ratelimited("overlay: failed to set redirect (%i)\n", err);
892 		/* Fall back to userspace copy-up */
893 		err = -EXDEV;
894 	}
895 	return err;
896 }
897 
898 static int ovl_rename(struct inode *olddir, struct dentry *old,
899 		      struct inode *newdir, struct dentry *new,
900 		      unsigned int flags)
901 {
902 	int err;
903 	bool locked = false;
904 	struct dentry *old_upperdir;
905 	struct dentry *new_upperdir;
906 	struct dentry *olddentry;
907 	struct dentry *newdentry;
908 	struct dentry *trap;
909 	bool old_opaque;
910 	bool new_opaque;
911 	bool cleanup_whiteout = false;
912 	bool overwrite = !(flags & RENAME_EXCHANGE);
913 	bool is_dir = d_is_dir(old);
914 	bool new_is_dir = d_is_dir(new);
915 	bool samedir = olddir == newdir;
916 	struct dentry *opaquedir = NULL;
917 	const struct cred *old_cred = NULL;
918 
919 	err = -EINVAL;
920 	if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
921 		goto out;
922 
923 	flags &= ~RENAME_NOREPLACE;
924 
925 	/* Don't copy up directory trees */
926 	err = -EXDEV;
927 	if (!ovl_can_move(old))
928 		goto out;
929 	if (!overwrite && !ovl_can_move(new))
930 		goto out;
931 
932 	err = ovl_want_write(old);
933 	if (err)
934 		goto out;
935 
936 	err = ovl_copy_up(old);
937 	if (err)
938 		goto out_drop_write;
939 
940 	err = ovl_copy_up(new->d_parent);
941 	if (err)
942 		goto out_drop_write;
943 	if (!overwrite) {
944 		err = ovl_copy_up(new);
945 		if (err)
946 			goto out_drop_write;
947 	} else {
948 		err = ovl_nlink_start(new, &locked);
949 		if (err)
950 			goto out_drop_write;
951 	}
952 
953 	old_cred = ovl_override_creds(old->d_sb);
954 
955 	if (overwrite && new_is_dir && ovl_type_merge_or_lower(new)) {
956 		opaquedir = ovl_check_empty_and_clear(new);
957 		err = PTR_ERR(opaquedir);
958 		if (IS_ERR(opaquedir)) {
959 			opaquedir = NULL;
960 			goto out_revert_creds;
961 		}
962 	}
963 
964 	if (overwrite) {
965 		if (ovl_lower_positive(old)) {
966 			if (!ovl_dentry_is_whiteout(new)) {
967 				/* Whiteout source */
968 				flags |= RENAME_WHITEOUT;
969 			} else {
970 				/* Switch whiteouts */
971 				flags |= RENAME_EXCHANGE;
972 			}
973 		} else if (is_dir && ovl_dentry_is_whiteout(new)) {
974 			flags |= RENAME_EXCHANGE;
975 			cleanup_whiteout = true;
976 		}
977 	}
978 
979 	old_upperdir = ovl_dentry_upper(old->d_parent);
980 	new_upperdir = ovl_dentry_upper(new->d_parent);
981 
982 	if (!samedir) {
983 		/*
984 		 * When moving a merge dir or non-dir with copy up origin into
985 		 * a new parent, we are marking the new parent dir "impure".
986 		 * When ovl_iterate() iterates an "impure" upper dir, it will
987 		 * lookup the origin inodes of the entries to fill d_ino.
988 		 */
989 		if (ovl_type_origin(old)) {
990 			err = ovl_set_impure(new->d_parent, new_upperdir);
991 			if (err)
992 				goto out_revert_creds;
993 		}
994 		if (!overwrite && ovl_type_origin(new)) {
995 			err = ovl_set_impure(old->d_parent, old_upperdir);
996 			if (err)
997 				goto out_revert_creds;
998 		}
999 	}
1000 
1001 	trap = lock_rename(new_upperdir, old_upperdir);
1002 
1003 	olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1004 				   old->d_name.len);
1005 	err = PTR_ERR(olddentry);
1006 	if (IS_ERR(olddentry))
1007 		goto out_unlock;
1008 
1009 	err = -ESTALE;
1010 	if (!ovl_matches_upper(old, olddentry))
1011 		goto out_dput_old;
1012 
1013 	newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1014 				   new->d_name.len);
1015 	err = PTR_ERR(newdentry);
1016 	if (IS_ERR(newdentry))
1017 		goto out_dput_old;
1018 
1019 	old_opaque = ovl_dentry_is_opaque(old);
1020 	new_opaque = ovl_dentry_is_opaque(new);
1021 
1022 	err = -ESTALE;
1023 	if (d_inode(new) && ovl_dentry_upper(new)) {
1024 		if (opaquedir) {
1025 			if (newdentry != opaquedir)
1026 				goto out_dput;
1027 		} else {
1028 			if (!ovl_matches_upper(new, newdentry))
1029 				goto out_dput;
1030 		}
1031 	} else {
1032 		if (!d_is_negative(newdentry) &&
1033 		    (!new_opaque || !ovl_is_whiteout(newdentry)))
1034 			goto out_dput;
1035 	}
1036 
1037 	if (olddentry == trap)
1038 		goto out_dput;
1039 	if (newdentry == trap)
1040 		goto out_dput;
1041 
1042 	if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
1043 		goto out_dput;
1044 
1045 	err = 0;
1046 	if (is_dir) {
1047 		if (ovl_type_merge_or_lower(old))
1048 			err = ovl_set_redirect(old, samedir);
1049 		else if (!old_opaque && ovl_type_merge(new->d_parent))
1050 			err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1051 		if (err)
1052 			goto out_dput;
1053 	}
1054 	if (!overwrite && new_is_dir) {
1055 		if (ovl_type_merge_or_lower(new))
1056 			err = ovl_set_redirect(new, samedir);
1057 		else if (!new_opaque && ovl_type_merge(old->d_parent))
1058 			err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1059 		if (err)
1060 			goto out_dput;
1061 	}
1062 
1063 	err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1064 			    new_upperdir->d_inode, newdentry, flags);
1065 	if (err)
1066 		goto out_dput;
1067 
1068 	if (cleanup_whiteout)
1069 		ovl_cleanup(old_upperdir->d_inode, newdentry);
1070 
1071 	if (overwrite && d_inode(new)) {
1072 		if (new_is_dir)
1073 			clear_nlink(d_inode(new));
1074 		else
1075 			drop_nlink(d_inode(new));
1076 	}
1077 
1078 	ovl_dentry_version_inc(old->d_parent);
1079 	ovl_dentry_version_inc(new->d_parent);
1080 
1081 out_dput:
1082 	dput(newdentry);
1083 out_dput_old:
1084 	dput(olddentry);
1085 out_unlock:
1086 	unlock_rename(new_upperdir, old_upperdir);
1087 out_revert_creds:
1088 	revert_creds(old_cred);
1089 	ovl_nlink_end(new, locked);
1090 out_drop_write:
1091 	ovl_drop_write(old);
1092 out:
1093 	dput(opaquedir);
1094 	return err;
1095 }
1096 
1097 const struct inode_operations ovl_dir_inode_operations = {
1098 	.lookup		= ovl_lookup,
1099 	.mkdir		= ovl_mkdir,
1100 	.symlink	= ovl_symlink,
1101 	.unlink		= ovl_unlink,
1102 	.rmdir		= ovl_rmdir,
1103 	.rename		= ovl_rename,
1104 	.link		= ovl_link,
1105 	.setattr	= ovl_setattr,
1106 	.create		= ovl_create,
1107 	.mknod		= ovl_mknod,
1108 	.permission	= ovl_permission,
1109 	.getattr	= ovl_getattr,
1110 	.listxattr	= ovl_listxattr,
1111 	.get_acl	= ovl_get_acl,
1112 	.update_time	= ovl_update_time,
1113 };
1114