xref: /openbmc/linux/fs/overlayfs/dir.c (revision d236d361)
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 void 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 
44 struct dentry *ovl_lookup_temp(struct dentry *workdir, struct dentry *dentry)
45 {
46 	struct dentry *temp;
47 	char name[20];
48 	static atomic_t temp_id = ATOMIC_INIT(0);
49 
50 	/* counter is allowed to wrap, since temp dentries are ephemeral */
51 	snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
52 
53 	temp = lookup_one_len(name, workdir, strlen(name));
54 	if (!IS_ERR(temp) && temp->d_inode) {
55 		pr_err("overlayfs: workdir/%s already exists\n", name);
56 		dput(temp);
57 		temp = ERR_PTR(-EIO);
58 	}
59 
60 	return temp;
61 }
62 
63 /* caller holds i_mutex on workdir */
64 static struct dentry *ovl_whiteout(struct dentry *workdir,
65 				   struct dentry *dentry)
66 {
67 	int err;
68 	struct dentry *whiteout;
69 	struct inode *wdir = workdir->d_inode;
70 
71 	whiteout = ovl_lookup_temp(workdir, dentry);
72 	if (IS_ERR(whiteout))
73 		return whiteout;
74 
75 	err = ovl_do_whiteout(wdir, whiteout);
76 	if (err) {
77 		dput(whiteout);
78 		whiteout = ERR_PTR(err);
79 	}
80 
81 	return whiteout;
82 }
83 
84 int ovl_create_real(struct inode *dir, struct dentry *newdentry,
85 		    struct cattr *attr, struct dentry *hardlink, bool debug)
86 {
87 	int err;
88 
89 	if (newdentry->d_inode)
90 		return -ESTALE;
91 
92 	if (hardlink) {
93 		err = ovl_do_link(hardlink, dir, newdentry, debug);
94 	} else {
95 		switch (attr->mode & S_IFMT) {
96 		case S_IFREG:
97 			err = ovl_do_create(dir, newdentry, attr->mode, debug);
98 			break;
99 
100 		case S_IFDIR:
101 			err = ovl_do_mkdir(dir, newdentry, attr->mode, debug);
102 			break;
103 
104 		case S_IFCHR:
105 		case S_IFBLK:
106 		case S_IFIFO:
107 		case S_IFSOCK:
108 			err = ovl_do_mknod(dir, newdentry,
109 					   attr->mode, attr->rdev, debug);
110 			break;
111 
112 		case S_IFLNK:
113 			err = ovl_do_symlink(dir, newdentry, attr->link, debug);
114 			break;
115 
116 		default:
117 			err = -EPERM;
118 		}
119 	}
120 	if (!err && WARN_ON(!newdentry->d_inode)) {
121 		/*
122 		 * Not quite sure if non-instantiated dentry is legal or not.
123 		 * VFS doesn't seem to care so check and warn here.
124 		 */
125 		err = -ENOENT;
126 	}
127 	return err;
128 }
129 
130 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
131 {
132 	int err;
133 
134 	err = ovl_do_setxattr(upperdentry, OVL_XATTR_OPAQUE, "y", 1, 0);
135 	if (!err)
136 		ovl_dentry_set_opaque(dentry);
137 
138 	return err;
139 }
140 
141 /* Common operations required to be done after creation of file on upper */
142 static void ovl_instantiate(struct dentry *dentry, struct inode *inode,
143 			    struct dentry *newdentry, bool hardlink)
144 {
145 	ovl_dentry_version_inc(dentry->d_parent);
146 	ovl_dentry_update(dentry, newdentry);
147 	if (!hardlink) {
148 		ovl_inode_update(inode, d_inode(newdentry));
149 		ovl_copyattr(newdentry->d_inode, inode);
150 	} else {
151 		WARN_ON(ovl_inode_real(inode, NULL) != d_inode(newdentry));
152 		inc_nlink(inode);
153 	}
154 	d_instantiate(dentry, inode);
155 	/* Force lookup of new upper hardlink to find its lower */
156 	if (hardlink)
157 		d_drop(dentry);
158 }
159 
160 static bool ovl_type_merge(struct dentry *dentry)
161 {
162 	return OVL_TYPE_MERGE(ovl_path_type(dentry));
163 }
164 
165 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
166 			    struct cattr *attr, struct dentry *hardlink)
167 {
168 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
169 	struct inode *udir = upperdir->d_inode;
170 	struct dentry *newdentry;
171 	int err;
172 
173 	if (!hardlink && !IS_POSIXACL(udir))
174 		attr->mode &= ~current_umask();
175 
176 	inode_lock_nested(udir, I_MUTEX_PARENT);
177 	newdentry = lookup_one_len(dentry->d_name.name, upperdir,
178 				   dentry->d_name.len);
179 	err = PTR_ERR(newdentry);
180 	if (IS_ERR(newdentry))
181 		goto out_unlock;
182 	err = ovl_create_real(udir, newdentry, attr, hardlink, false);
183 	if (err)
184 		goto out_dput;
185 
186 	if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry)) {
187 		/* Setting opaque here is just an optimization, allow to fail */
188 		ovl_set_opaque(dentry, newdentry);
189 	}
190 
191 	ovl_instantiate(dentry, inode, newdentry, !!hardlink);
192 	newdentry = NULL;
193 out_dput:
194 	dput(newdentry);
195 out_unlock:
196 	inode_unlock(udir);
197 	return err;
198 }
199 
200 static int ovl_lock_rename_workdir(struct dentry *workdir,
201 				   struct dentry *upperdir)
202 {
203 	/* Workdir should not be the same as upperdir */
204 	if (workdir == upperdir)
205 		goto err;
206 
207 	/* Workdir should not be subdir of upperdir and vice versa */
208 	if (lock_rename(workdir, upperdir) != NULL)
209 		goto err_unlock;
210 
211 	return 0;
212 
213 err_unlock:
214 	unlock_rename(workdir, upperdir);
215 err:
216 	pr_err("overlayfs: failed to lock workdir+upperdir\n");
217 	return -EIO;
218 }
219 
220 static struct dentry *ovl_clear_empty(struct dentry *dentry,
221 				      struct list_head *list)
222 {
223 	struct dentry *workdir = ovl_workdir(dentry);
224 	struct inode *wdir = workdir->d_inode;
225 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
226 	struct inode *udir = upperdir->d_inode;
227 	struct path upperpath;
228 	struct dentry *upper;
229 	struct dentry *opaquedir;
230 	struct kstat stat;
231 	int err;
232 
233 	if (WARN_ON(!workdir))
234 		return ERR_PTR(-EROFS);
235 
236 	err = ovl_lock_rename_workdir(workdir, upperdir);
237 	if (err)
238 		goto out;
239 
240 	ovl_path_upper(dentry, &upperpath);
241 	err = vfs_getattr(&upperpath, &stat,
242 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
243 	if (err)
244 		goto out_unlock;
245 
246 	err = -ESTALE;
247 	if (!S_ISDIR(stat.mode))
248 		goto out_unlock;
249 	upper = upperpath.dentry;
250 	if (upper->d_parent->d_inode != udir)
251 		goto out_unlock;
252 
253 	opaquedir = ovl_lookup_temp(workdir, dentry);
254 	err = PTR_ERR(opaquedir);
255 	if (IS_ERR(opaquedir))
256 		goto out_unlock;
257 
258 	err = ovl_create_real(wdir, opaquedir,
259 			      &(struct cattr){.mode = stat.mode}, NULL, true);
260 	if (err)
261 		goto out_dput;
262 
263 	err = ovl_copy_xattr(upper, opaquedir);
264 	if (err)
265 		goto out_cleanup;
266 
267 	err = ovl_set_opaque(dentry, opaquedir);
268 	if (err)
269 		goto out_cleanup;
270 
271 	inode_lock(opaquedir->d_inode);
272 	err = ovl_set_attr(opaquedir, &stat);
273 	inode_unlock(opaquedir->d_inode);
274 	if (err)
275 		goto out_cleanup;
276 
277 	err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
278 	if (err)
279 		goto out_cleanup;
280 
281 	ovl_cleanup_whiteouts(upper, list);
282 	ovl_cleanup(wdir, upper);
283 	unlock_rename(workdir, upperdir);
284 
285 	/* dentry's upper doesn't match now, get rid of it */
286 	d_drop(dentry);
287 
288 	return opaquedir;
289 
290 out_cleanup:
291 	ovl_cleanup(wdir, opaquedir);
292 out_dput:
293 	dput(opaquedir);
294 out_unlock:
295 	unlock_rename(workdir, upperdir);
296 out:
297 	return ERR_PTR(err);
298 }
299 
300 static struct dentry *ovl_check_empty_and_clear(struct dentry *dentry)
301 {
302 	int err;
303 	struct dentry *ret = NULL;
304 	enum ovl_path_type type = ovl_path_type(dentry);
305 	LIST_HEAD(list);
306 
307 	err = ovl_check_empty_dir(dentry, &list);
308 	if (err) {
309 		ret = ERR_PTR(err);
310 		goto out_free;
311 	}
312 
313 	/*
314 	 * When removing an empty opaque directory, then it makes no sense to
315 	 * replace it with an exact replica of itself.
316 	 *
317 	 * If no upperdentry then skip clearing whiteouts.
318 	 *
319 	 * Can race with copy-up, since we don't hold the upperdir mutex.
320 	 * Doesn't matter, since copy-up can't create a non-empty directory
321 	 * from an empty one.
322 	 */
323 	if (OVL_TYPE_UPPER(type) && OVL_TYPE_MERGE(type))
324 		ret = ovl_clear_empty(dentry, &list);
325 
326 out_free:
327 	ovl_cache_free(&list);
328 
329 	return ret;
330 }
331 
332 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
333 			     const struct posix_acl *acl)
334 {
335 	void *buffer;
336 	size_t size;
337 	int err;
338 
339 	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
340 		return 0;
341 
342 	size = posix_acl_to_xattr(NULL, acl, NULL, 0);
343 	buffer = kmalloc(size, GFP_KERNEL);
344 	if (!buffer)
345 		return -ENOMEM;
346 
347 	size = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
348 	err = size;
349 	if (err < 0)
350 		goto out_free;
351 
352 	err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
353 out_free:
354 	kfree(buffer);
355 	return err;
356 }
357 
358 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
359 				    struct cattr *cattr,
360 				    struct dentry *hardlink)
361 {
362 	struct dentry *workdir = ovl_workdir(dentry);
363 	struct inode *wdir = workdir->d_inode;
364 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
365 	struct inode *udir = upperdir->d_inode;
366 	struct dentry *upper;
367 	struct dentry *newdentry;
368 	int err;
369 	struct posix_acl *acl, *default_acl;
370 
371 	if (WARN_ON(!workdir))
372 		return -EROFS;
373 
374 	if (!hardlink) {
375 		err = posix_acl_create(dentry->d_parent->d_inode,
376 				       &cattr->mode, &default_acl, &acl);
377 		if (err)
378 			return err;
379 	}
380 
381 	err = ovl_lock_rename_workdir(workdir, upperdir);
382 	if (err)
383 		goto out;
384 
385 	newdentry = ovl_lookup_temp(workdir, dentry);
386 	err = PTR_ERR(newdentry);
387 	if (IS_ERR(newdentry))
388 		goto out_unlock;
389 
390 	upper = lookup_one_len(dentry->d_name.name, upperdir,
391 			       dentry->d_name.len);
392 	err = PTR_ERR(upper);
393 	if (IS_ERR(upper))
394 		goto out_dput;
395 
396 	err = ovl_create_real(wdir, newdentry, cattr, hardlink, true);
397 	if (err)
398 		goto out_dput2;
399 
400 	/*
401 	 * mode could have been mutilated due to umask (e.g. sgid directory)
402 	 */
403 	if (!hardlink &&
404 	    !S_ISLNK(cattr->mode) &&
405 	    newdentry->d_inode->i_mode != cattr->mode) {
406 		struct iattr attr = {
407 			.ia_valid = ATTR_MODE,
408 			.ia_mode = cattr->mode,
409 		};
410 		inode_lock(newdentry->d_inode);
411 		err = notify_change(newdentry, &attr, NULL);
412 		inode_unlock(newdentry->d_inode);
413 		if (err)
414 			goto out_cleanup;
415 	}
416 	if (!hardlink) {
417 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
418 					acl);
419 		if (err)
420 			goto out_cleanup;
421 
422 		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
423 					default_acl);
424 		if (err)
425 			goto out_cleanup;
426 	}
427 
428 	if (!hardlink && S_ISDIR(cattr->mode)) {
429 		err = ovl_set_opaque(dentry, newdentry);
430 		if (err)
431 			goto out_cleanup;
432 
433 		err = ovl_do_rename(wdir, newdentry, udir, upper,
434 				    RENAME_EXCHANGE);
435 		if (err)
436 			goto out_cleanup;
437 
438 		ovl_cleanup(wdir, upper);
439 	} else {
440 		err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
441 		if (err)
442 			goto out_cleanup;
443 	}
444 	ovl_instantiate(dentry, inode, newdentry, !!hardlink);
445 	newdentry = NULL;
446 out_dput2:
447 	dput(upper);
448 out_dput:
449 	dput(newdentry);
450 out_unlock:
451 	unlock_rename(workdir, upperdir);
452 out:
453 	if (!hardlink) {
454 		posix_acl_release(acl);
455 		posix_acl_release(default_acl);
456 	}
457 	return err;
458 
459 out_cleanup:
460 	ovl_cleanup(wdir, newdentry);
461 	goto out_dput2;
462 }
463 
464 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
465 			      struct cattr *attr, struct dentry *hardlink)
466 {
467 	int err;
468 	const struct cred *old_cred;
469 	struct cred *override_cred;
470 
471 	err = ovl_copy_up(dentry->d_parent);
472 	if (err)
473 		return err;
474 
475 	old_cred = ovl_override_creds(dentry->d_sb);
476 	err = -ENOMEM;
477 	override_cred = prepare_creds();
478 	if (override_cred) {
479 		override_cred->fsuid = inode->i_uid;
480 		override_cred->fsgid = inode->i_gid;
481 		if (!hardlink) {
482 			err = security_dentry_create_files_as(dentry,
483 					attr->mode, &dentry->d_name, old_cred,
484 					override_cred);
485 			if (err) {
486 				put_cred(override_cred);
487 				goto out_revert_creds;
488 			}
489 		}
490 		put_cred(override_creds(override_cred));
491 		put_cred(override_cred);
492 
493 		if (!ovl_dentry_is_whiteout(dentry))
494 			err = ovl_create_upper(dentry, inode, attr,
495 						hardlink);
496 		else
497 			err = ovl_create_over_whiteout(dentry, inode, attr,
498 							hardlink);
499 	}
500 out_revert_creds:
501 	revert_creds(old_cred);
502 	if (!err) {
503 		struct inode *realinode = d_inode(ovl_dentry_upper(dentry));
504 
505 		WARN_ON(inode->i_mode != realinode->i_mode);
506 		WARN_ON(!uid_eq(inode->i_uid, realinode->i_uid));
507 		WARN_ON(!gid_eq(inode->i_gid, realinode->i_gid));
508 	}
509 	return err;
510 }
511 
512 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
513 			     const char *link)
514 {
515 	int err;
516 	struct inode *inode;
517 	struct cattr attr = {
518 		.rdev = rdev,
519 		.link = link,
520 	};
521 
522 	err = ovl_want_write(dentry);
523 	if (err)
524 		goto out;
525 
526 	err = -ENOMEM;
527 	inode = ovl_new_inode(dentry->d_sb, mode, rdev);
528 	if (!inode)
529 		goto out_drop_write;
530 
531 	inode_init_owner(inode, dentry->d_parent->d_inode, mode);
532 	attr.mode = inode->i_mode;
533 
534 	err = ovl_create_or_link(dentry, inode, &attr, NULL);
535 	if (err)
536 		iput(inode);
537 
538 out_drop_write:
539 	ovl_drop_write(dentry);
540 out:
541 	return err;
542 }
543 
544 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
545 		      bool excl)
546 {
547 	return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
548 }
549 
550 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
551 {
552 	return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
553 }
554 
555 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
556 		     dev_t rdev)
557 {
558 	/* Don't allow creation of "whiteout" on overlay */
559 	if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
560 		return -EPERM;
561 
562 	return ovl_create_object(dentry, mode, rdev, NULL);
563 }
564 
565 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
566 		       const char *link)
567 {
568 	return ovl_create_object(dentry, S_IFLNK, 0, link);
569 }
570 
571 static int ovl_link(struct dentry *old, struct inode *newdir,
572 		    struct dentry *new)
573 {
574 	int err;
575 	struct inode *inode;
576 
577 	err = ovl_want_write(old);
578 	if (err)
579 		goto out;
580 
581 	err = ovl_copy_up(old);
582 	if (err)
583 		goto out_drop_write;
584 
585 	inode = d_inode(old);
586 	ihold(inode);
587 
588 	err = ovl_create_or_link(new, inode, NULL, ovl_dentry_upper(old));
589 	if (err)
590 		iput(inode);
591 
592 out_drop_write:
593 	ovl_drop_write(old);
594 out:
595 	return err;
596 }
597 
598 static int ovl_remove_and_whiteout(struct dentry *dentry, bool is_dir)
599 {
600 	struct dentry *workdir = ovl_workdir(dentry);
601 	struct inode *wdir = workdir->d_inode;
602 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
603 	struct inode *udir = upperdir->d_inode;
604 	struct dentry *whiteout;
605 	struct dentry *upper;
606 	struct dentry *opaquedir = NULL;
607 	int err;
608 	int flags = 0;
609 
610 	if (WARN_ON(!workdir))
611 		return -EROFS;
612 
613 	if (is_dir) {
614 		opaquedir = ovl_check_empty_and_clear(dentry);
615 		err = PTR_ERR(opaquedir);
616 		if (IS_ERR(opaquedir))
617 			goto out;
618 	}
619 
620 	err = ovl_lock_rename_workdir(workdir, upperdir);
621 	if (err)
622 		goto out_dput;
623 
624 	upper = lookup_one_len(dentry->d_name.name, upperdir,
625 			       dentry->d_name.len);
626 	err = PTR_ERR(upper);
627 	if (IS_ERR(upper))
628 		goto out_unlock;
629 
630 	err = -ESTALE;
631 	if ((opaquedir && upper != opaquedir) ||
632 	    (!opaquedir && ovl_dentry_upper(dentry) &&
633 	     upper != ovl_dentry_upper(dentry))) {
634 		goto out_dput_upper;
635 	}
636 
637 	whiteout = ovl_whiteout(workdir, dentry);
638 	err = PTR_ERR(whiteout);
639 	if (IS_ERR(whiteout))
640 		goto out_dput_upper;
641 
642 	if (d_is_dir(upper))
643 		flags = RENAME_EXCHANGE;
644 
645 	err = ovl_do_rename(wdir, whiteout, udir, upper, flags);
646 	if (err)
647 		goto kill_whiteout;
648 	if (flags)
649 		ovl_cleanup(wdir, upper);
650 
651 	ovl_dentry_version_inc(dentry->d_parent);
652 out_d_drop:
653 	d_drop(dentry);
654 	dput(whiteout);
655 out_dput_upper:
656 	dput(upper);
657 out_unlock:
658 	unlock_rename(workdir, upperdir);
659 out_dput:
660 	dput(opaquedir);
661 out:
662 	return err;
663 
664 kill_whiteout:
665 	ovl_cleanup(wdir, whiteout);
666 	goto out_d_drop;
667 }
668 
669 static int ovl_remove_upper(struct dentry *dentry, bool is_dir)
670 {
671 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
672 	struct inode *dir = upperdir->d_inode;
673 	struct dentry *upper;
674 	struct dentry *opaquedir = NULL;
675 	int err;
676 
677 	/* Redirect dir can be !ovl_lower_positive && OVL_TYPE_MERGE */
678 	if (is_dir && ovl_dentry_get_redirect(dentry)) {
679 		opaquedir = ovl_check_empty_and_clear(dentry);
680 		err = PTR_ERR(opaquedir);
681 		if (IS_ERR(opaquedir))
682 			goto out;
683 	}
684 
685 	inode_lock_nested(dir, I_MUTEX_PARENT);
686 	upper = lookup_one_len(dentry->d_name.name, upperdir,
687 			       dentry->d_name.len);
688 	err = PTR_ERR(upper);
689 	if (IS_ERR(upper))
690 		goto out_unlock;
691 
692 	err = -ESTALE;
693 	if ((opaquedir && upper != opaquedir) ||
694 	    (!opaquedir && upper != ovl_dentry_upper(dentry)))
695 		goto out_dput_upper;
696 
697 	if (is_dir)
698 		err = vfs_rmdir(dir, upper);
699 	else
700 		err = vfs_unlink(dir, upper, NULL);
701 	ovl_dentry_version_inc(dentry->d_parent);
702 
703 	/*
704 	 * Keeping this dentry hashed would mean having to release
705 	 * upperpath/lowerpath, which could only be done if we are the
706 	 * sole user of this dentry.  Too tricky...  Just unhash for
707 	 * now.
708 	 */
709 	if (!err)
710 		d_drop(dentry);
711 out_dput_upper:
712 	dput(upper);
713 out_unlock:
714 	inode_unlock(dir);
715 	dput(opaquedir);
716 out:
717 	return err;
718 }
719 
720 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
721 {
722 	enum ovl_path_type type;
723 	int err;
724 	const struct cred *old_cred;
725 
726 	err = ovl_want_write(dentry);
727 	if (err)
728 		goto out;
729 
730 	err = ovl_copy_up(dentry->d_parent);
731 	if (err)
732 		goto out_drop_write;
733 
734 	type = ovl_path_type(dentry);
735 
736 	old_cred = ovl_override_creds(dentry->d_sb);
737 	if (!ovl_lower_positive(dentry))
738 		err = ovl_remove_upper(dentry, is_dir);
739 	else
740 		err = ovl_remove_and_whiteout(dentry, is_dir);
741 	revert_creds(old_cred);
742 	if (!err) {
743 		if (is_dir)
744 			clear_nlink(dentry->d_inode);
745 		else
746 			drop_nlink(dentry->d_inode);
747 	}
748 out_drop_write:
749 	ovl_drop_write(dentry);
750 out:
751 	return err;
752 }
753 
754 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
755 {
756 	return ovl_do_remove(dentry, false);
757 }
758 
759 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
760 {
761 	return ovl_do_remove(dentry, true);
762 }
763 
764 static bool ovl_type_merge_or_lower(struct dentry *dentry)
765 {
766 	enum ovl_path_type type = ovl_path_type(dentry);
767 
768 	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
769 }
770 
771 static bool ovl_can_move(struct dentry *dentry)
772 {
773 	return ovl_redirect_dir(dentry->d_sb) ||
774 		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
775 }
776 
777 static char *ovl_get_redirect(struct dentry *dentry, bool samedir)
778 {
779 	char *buf, *ret;
780 	struct dentry *d, *tmp;
781 	int buflen = ovl_redirect_max + 1;
782 
783 	if (samedir) {
784 		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
785 			       GFP_KERNEL);
786 		goto out;
787 	}
788 
789 	buf = ret = kmalloc(buflen, GFP_TEMPORARY);
790 	if (!buf)
791 		goto out;
792 
793 	buflen--;
794 	buf[buflen] = '\0';
795 	for (d = dget(dentry); !IS_ROOT(d);) {
796 		const char *name;
797 		int thislen;
798 
799 		spin_lock(&d->d_lock);
800 		name = ovl_dentry_get_redirect(d);
801 		if (name) {
802 			thislen = strlen(name);
803 		} else {
804 			name = d->d_name.name;
805 			thislen = d->d_name.len;
806 		}
807 
808 		/* If path is too long, fall back to userspace move */
809 		if (thislen + (name[0] != '/') > buflen) {
810 			ret = ERR_PTR(-EXDEV);
811 			spin_unlock(&d->d_lock);
812 			goto out_put;
813 		}
814 
815 		buflen -= thislen;
816 		memcpy(&buf[buflen], name, thislen);
817 		tmp = dget_dlock(d->d_parent);
818 		spin_unlock(&d->d_lock);
819 
820 		dput(d);
821 		d = tmp;
822 
823 		/* Absolute redirect: finished */
824 		if (buf[buflen] == '/')
825 			break;
826 		buflen--;
827 		buf[buflen] = '/';
828 	}
829 	ret = kstrdup(&buf[buflen], GFP_KERNEL);
830 out_put:
831 	dput(d);
832 	kfree(buf);
833 out:
834 	return ret ? ret : ERR_PTR(-ENOMEM);
835 }
836 
837 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
838 {
839 	int err;
840 	const char *redirect = ovl_dentry_get_redirect(dentry);
841 
842 	if (redirect && (samedir || redirect[0] == '/'))
843 		return 0;
844 
845 	redirect = ovl_get_redirect(dentry, samedir);
846 	if (IS_ERR(redirect))
847 		return PTR_ERR(redirect);
848 
849 	err = ovl_do_setxattr(ovl_dentry_upper(dentry), OVL_XATTR_REDIRECT,
850 			      redirect, strlen(redirect), 0);
851 	if (!err) {
852 		spin_lock(&dentry->d_lock);
853 		ovl_dentry_set_redirect(dentry, redirect);
854 		spin_unlock(&dentry->d_lock);
855 	} else {
856 		kfree(redirect);
857 		if (err == -EOPNOTSUPP)
858 			ovl_clear_redirect_dir(dentry->d_sb);
859 		else
860 			pr_warn_ratelimited("overlay: failed to set redirect (%i)\n", err);
861 		/* Fall back to userspace copy-up */
862 		err = -EXDEV;
863 	}
864 	return err;
865 }
866 
867 static int ovl_rename(struct inode *olddir, struct dentry *old,
868 		      struct inode *newdir, struct dentry *new,
869 		      unsigned int flags)
870 {
871 	int err;
872 	struct dentry *old_upperdir;
873 	struct dentry *new_upperdir;
874 	struct dentry *olddentry;
875 	struct dentry *newdentry;
876 	struct dentry *trap;
877 	bool old_opaque;
878 	bool new_opaque;
879 	bool cleanup_whiteout = false;
880 	bool overwrite = !(flags & RENAME_EXCHANGE);
881 	bool is_dir = d_is_dir(old);
882 	bool new_is_dir = d_is_dir(new);
883 	bool samedir = olddir == newdir;
884 	struct dentry *opaquedir = NULL;
885 	const struct cred *old_cred = NULL;
886 
887 	err = -EINVAL;
888 	if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
889 		goto out;
890 
891 	flags &= ~RENAME_NOREPLACE;
892 
893 	/* Don't copy up directory trees */
894 	err = -EXDEV;
895 	if (!ovl_can_move(old))
896 		goto out;
897 	if (!overwrite && !ovl_can_move(new))
898 		goto out;
899 
900 	err = ovl_want_write(old);
901 	if (err)
902 		goto out;
903 
904 	err = ovl_copy_up(old);
905 	if (err)
906 		goto out_drop_write;
907 
908 	err = ovl_copy_up(new->d_parent);
909 	if (err)
910 		goto out_drop_write;
911 	if (!overwrite) {
912 		err = ovl_copy_up(new);
913 		if (err)
914 			goto out_drop_write;
915 	}
916 
917 	old_cred = ovl_override_creds(old->d_sb);
918 
919 	if (overwrite && new_is_dir && ovl_type_merge_or_lower(new)) {
920 		opaquedir = ovl_check_empty_and_clear(new);
921 		err = PTR_ERR(opaquedir);
922 		if (IS_ERR(opaquedir)) {
923 			opaquedir = NULL;
924 			goto out_revert_creds;
925 		}
926 	}
927 
928 	if (overwrite) {
929 		if (ovl_lower_positive(old)) {
930 			if (!ovl_dentry_is_whiteout(new)) {
931 				/* Whiteout source */
932 				flags |= RENAME_WHITEOUT;
933 			} else {
934 				/* Switch whiteouts */
935 				flags |= RENAME_EXCHANGE;
936 			}
937 		} else if (is_dir && ovl_dentry_is_whiteout(new)) {
938 			flags |= RENAME_EXCHANGE;
939 			cleanup_whiteout = true;
940 		}
941 	}
942 
943 	old_upperdir = ovl_dentry_upper(old->d_parent);
944 	new_upperdir = ovl_dentry_upper(new->d_parent);
945 
946 	trap = lock_rename(new_upperdir, old_upperdir);
947 
948 	olddentry = lookup_one_len(old->d_name.name, old_upperdir,
949 				   old->d_name.len);
950 	err = PTR_ERR(olddentry);
951 	if (IS_ERR(olddentry))
952 		goto out_unlock;
953 
954 	err = -ESTALE;
955 	if (olddentry != ovl_dentry_upper(old))
956 		goto out_dput_old;
957 
958 	newdentry = lookup_one_len(new->d_name.name, new_upperdir,
959 				   new->d_name.len);
960 	err = PTR_ERR(newdentry);
961 	if (IS_ERR(newdentry))
962 		goto out_dput_old;
963 
964 	old_opaque = ovl_dentry_is_opaque(old);
965 	new_opaque = ovl_dentry_is_opaque(new);
966 
967 	err = -ESTALE;
968 	if (ovl_dentry_upper(new)) {
969 		if (opaquedir) {
970 			if (newdentry != opaquedir)
971 				goto out_dput;
972 		} else {
973 			if (newdentry != ovl_dentry_upper(new))
974 				goto out_dput;
975 		}
976 	} else {
977 		if (!d_is_negative(newdentry) &&
978 		    (!new_opaque || !ovl_is_whiteout(newdentry)))
979 			goto out_dput;
980 	}
981 
982 	if (olddentry == trap)
983 		goto out_dput;
984 	if (newdentry == trap)
985 		goto out_dput;
986 
987 	if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
988 		goto out_dput;
989 
990 	err = 0;
991 	if (is_dir) {
992 		if (ovl_type_merge_or_lower(old))
993 			err = ovl_set_redirect(old, samedir);
994 		else if (!old_opaque && ovl_type_merge(new->d_parent))
995 			err = ovl_set_opaque(old, olddentry);
996 		if (err)
997 			goto out_dput;
998 	}
999 	if (!overwrite && new_is_dir) {
1000 		if (ovl_type_merge_or_lower(new))
1001 			err = ovl_set_redirect(new, samedir);
1002 		else if (!new_opaque && ovl_type_merge(old->d_parent))
1003 			err = ovl_set_opaque(new, newdentry);
1004 		if (err)
1005 			goto out_dput;
1006 	}
1007 
1008 	err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1009 			    new_upperdir->d_inode, newdentry, flags);
1010 	if (err)
1011 		goto out_dput;
1012 
1013 	if (cleanup_whiteout)
1014 		ovl_cleanup(old_upperdir->d_inode, newdentry);
1015 
1016 	ovl_dentry_version_inc(old->d_parent);
1017 	ovl_dentry_version_inc(new->d_parent);
1018 
1019 out_dput:
1020 	dput(newdentry);
1021 out_dput_old:
1022 	dput(olddentry);
1023 out_unlock:
1024 	unlock_rename(new_upperdir, old_upperdir);
1025 out_revert_creds:
1026 	revert_creds(old_cred);
1027 out_drop_write:
1028 	ovl_drop_write(old);
1029 out:
1030 	dput(opaquedir);
1031 	return err;
1032 }
1033 
1034 const struct inode_operations ovl_dir_inode_operations = {
1035 	.lookup		= ovl_lookup,
1036 	.mkdir		= ovl_mkdir,
1037 	.symlink	= ovl_symlink,
1038 	.unlink		= ovl_unlink,
1039 	.rmdir		= ovl_rmdir,
1040 	.rename		= ovl_rename,
1041 	.link		= ovl_link,
1042 	.setattr	= ovl_setattr,
1043 	.create		= ovl_create,
1044 	.mknod		= ovl_mknod,
1045 	.permission	= ovl_permission,
1046 	.getattr	= ovl_getattr,
1047 	.listxattr	= ovl_listxattr,
1048 	.get_acl	= ovl_get_acl,
1049 	.update_time	= ovl_update_time,
1050 };
1051