xref: /linux/fs/overlayfs/dir.c (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright (C) 2011 Novell Inc.
5  */
6 
7 #include <linux/fs.h>
8 #include <linux/namei.h>
9 #include <linux/xattr.h>
10 #include <linux/security.h>
11 #include <linux/cred.h>
12 #include <linux/module.h>
13 #include <linux/posix_acl.h>
14 #include <linux/posix_acl_xattr.h>
15 #include <linux/atomic.h>
16 #include <linux/ratelimit.h>
17 #include <linux/backing-file.h>
18 #include "overlayfs.h"
19 
20 static unsigned short ovl_redirect_max = 256;
21 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
22 MODULE_PARM_DESC(redirect_max,
23 		 "Maximum length of absolute redirect xattr value");
24 
25 static int ovl_set_redirect(struct dentry *dentry, bool samedir);
26 
27 static int ovl_cleanup_locked(struct ovl_fs *ofs, struct inode *wdir,
28 			      struct dentry *wdentry)
29 {
30 	int err;
31 
32 	dget(wdentry);
33 	if (d_is_dir(wdentry))
34 		err = ovl_do_rmdir(ofs, wdir, wdentry);
35 	else
36 		err = ovl_do_unlink(ofs, wdir, wdentry);
37 	dput(wdentry);
38 
39 	if (err) {
40 		pr_err("cleanup of '%pd2' failed (%i)\n",
41 		       wdentry, err);
42 	}
43 
44 	return err;
45 }
46 
47 int ovl_cleanup(struct ovl_fs *ofs, struct dentry *workdir,
48 		struct dentry *wdentry)
49 {
50 	wdentry = start_removing_dentry(workdir, wdentry);
51 	if (IS_ERR(wdentry))
52 		return PTR_ERR(wdentry);
53 
54 	ovl_cleanup_locked(ofs, workdir->d_inode, wdentry);
55 	end_removing(wdentry);
56 
57 	return 0;
58 }
59 
60 void ovl_tempname(char name[OVL_TEMPNAME_SIZE])
61 {
62 	static atomic_t temp_id = ATOMIC_INIT(0);
63 
64 	/* counter is allowed to wrap, since temp dentries are ephemeral */
65 	snprintf(name, OVL_TEMPNAME_SIZE, "#%x", atomic_inc_return(&temp_id));
66 }
67 
68 static struct dentry *ovl_start_creating_temp(struct ovl_fs *ofs,
69 					      struct dentry *workdir,
70 					      char name[OVL_TEMPNAME_SIZE])
71 {
72 	ovl_tempname(name);
73 	return start_creating(ovl_upper_mnt_idmap(ofs), workdir,
74 			      &QSTR(name));
75 }
76 
77 static struct dentry *ovl_whiteout(struct ovl_fs *ofs)
78 {
79 	int err;
80 	struct dentry *whiteout, *link;
81 	struct dentry *workdir = ofs->workdir;
82 	struct inode *wdir = workdir->d_inode;
83 	char name[OVL_TEMPNAME_SIZE];
84 
85 	guard(mutex)(&ofs->whiteout_lock);
86 
87 	if (!ofs->whiteout) {
88 		whiteout = ovl_start_creating_temp(ofs, workdir, name);
89 		if (IS_ERR(whiteout))
90 			return whiteout;
91 		err = ovl_do_whiteout(ofs, wdir, whiteout);
92 		if (!err)
93 			ofs->whiteout = dget(whiteout);
94 		end_creating(whiteout);
95 		if (err)
96 			return ERR_PTR(err);
97 	}
98 
99 	if (!ofs->no_shared_whiteout) {
100 		link = ovl_start_creating_temp(ofs, workdir, name);
101 		if (IS_ERR(link))
102 			return link;
103 		err = ovl_do_link(ofs, ofs->whiteout, wdir, link);
104 		if (!err)
105 			whiteout = dget(link);
106 		end_creating(link);
107 		if (!err)
108 			return whiteout;
109 
110 		if (err != -EMLINK) {
111 			pr_warn("Failed to link whiteout - disabling whiteout inode sharing(nlink=%u, err=%u)\n",
112 				ofs->whiteout->d_inode->i_nlink,
113 				err);
114 			ofs->no_shared_whiteout = true;
115 		}
116 	}
117 	whiteout = ofs->whiteout;
118 	ofs->whiteout = NULL;
119 	return whiteout;
120 }
121 
122 int ovl_cleanup_and_whiteout(struct ovl_fs *ofs, struct dentry *dir,
123 			     struct dentry *dentry)
124 {
125 	struct dentry *whiteout;
126 	struct renamedata rd = {};
127 	int err;
128 	int flags = 0;
129 
130 	whiteout = ovl_whiteout(ofs);
131 	err = PTR_ERR(whiteout);
132 	if (IS_ERR(whiteout))
133 		return err;
134 
135 	if (d_is_dir(dentry))
136 		flags = RENAME_EXCHANGE;
137 
138 	rd.mnt_idmap = ovl_upper_mnt_idmap(ofs);
139 	rd.old_parent = ofs->workdir;
140 	rd.new_parent = dir;
141 	rd.flags = flags;
142 	err = start_renaming_two_dentries(&rd, whiteout, dentry);
143 	if (!err) {
144 		err = ovl_do_rename_rd(&rd);
145 		end_renaming(&rd);
146 	}
147 	if (err)
148 		goto kill_whiteout;
149 	if (flags)
150 		ovl_cleanup(ofs, ofs->workdir, dentry);
151 
152 out:
153 	dput(whiteout);
154 	return err;
155 
156 kill_whiteout:
157 	ovl_cleanup(ofs, ofs->workdir, whiteout);
158 	goto out;
159 }
160 
161 struct dentry *ovl_create_real(struct ovl_fs *ofs, struct dentry *parent,
162 			       struct dentry *newdentry, struct qstr *qname,
163 			       struct ovl_cattr *attr)
164 {
165 	struct inode *dir = parent->d_inode;
166 	int err;
167 
168 	if (IS_ERR(newdentry))
169 		return newdentry;
170 
171 	err = -ESTALE;
172 	if (newdentry->d_inode)
173 		goto out;
174 
175 	if (attr->hardlink) {
176 		err = ovl_do_link(ofs, attr->hardlink, dir, newdentry);
177 	} else {
178 		switch (attr->mode & S_IFMT) {
179 		case S_IFREG:
180 			err = ovl_do_create(ofs, dir, newdentry, attr->mode);
181 			break;
182 
183 		case S_IFDIR:
184 			/* mkdir is special... */
185 			newdentry =  ovl_do_mkdir(ofs, dir, newdentry, attr->mode);
186 			err = PTR_ERR_OR_ZERO(newdentry);
187 			/* expect to inherit casefolding from workdir/upperdir */
188 			if (!err && ofs->casefold != ovl_dentry_casefolded(newdentry)) {
189 				pr_warn_ratelimited("wrong inherited casefold (%pd2)\n",
190 						    newdentry);
191 				err = -EINVAL;
192 			}
193 			break;
194 
195 		case S_IFCHR:
196 		case S_IFBLK:
197 		case S_IFIFO:
198 		case S_IFSOCK:
199 			err = ovl_do_mknod(ofs, dir, newdentry, attr->mode,
200 					   attr->rdev);
201 			break;
202 
203 		case S_IFLNK:
204 			err = ovl_do_symlink(ofs, dir, newdentry, attr->link);
205 			break;
206 
207 		default:
208 			err = -EPERM;
209 		}
210 	}
211 	if (err)
212 		goto out;
213 
214 	if (WARN_ON(!newdentry->d_inode)) {
215 		/*
216 		 * Not quite sure if non-instantiated dentry is legal or not.
217 		 * VFS doesn't seem to care so check and warn here.
218 		 */
219 		err = -EIO;
220 	} else if (d_unhashed(newdentry)) {
221 		struct dentry *d;
222 		/*
223 		 * Some filesystems (i.e. casefolded) may return an unhashed
224 		 * negative dentry from the ovl_start_creating_upper() call before
225 		 * ovl_create_real().
226 		 * In that case, lookup again after making the newdentry
227 		 * positive, so ovl_create_upper() always returns a hashed
228 		 * positive dentry.  We lookup using qname which should be
229 		 * the same name as newentry, but is certain not to change.
230 		 * As we have to drop the lock before the lookup a race
231 		 * could result in a lookup failure.  In that case we return
232 		 * an error.
233 		 */
234 		end_creating_keep(newdentry);
235 		d = ovl_start_creating_upper(ofs, parent, qname);
236 
237 		if (IS_ERR_OR_NULL(d)) {
238 			err = d ? PTR_ERR(d) : -ENOENT;
239 		} else if (d->d_inode != newdentry->d_inode) {
240 			err = -EIO;
241 		} else {
242 			dput(newdentry);
243 			return d;
244 		}
245 		end_creating(d);
246 		dput(newdentry);
247 		return ERR_PTR(err);
248 	}
249 out:
250 	if (err) {
251 		end_creating(newdentry);
252 		return ERR_PTR(err);
253 	}
254 	return newdentry;
255 }
256 
257 struct dentry *ovl_create_temp(struct ovl_fs *ofs, struct dentry *workdir,
258 			       struct ovl_cattr *attr)
259 {
260 	struct dentry *ret;
261 	char name[OVL_TEMPNAME_SIZE];
262 
263 	ret = ovl_start_creating_temp(ofs, workdir, name);
264 	if (IS_ERR(ret))
265 		return ret;
266 	ret = ovl_create_real(ofs, workdir, ret, &QSTR(name), attr);
267 	return end_creating_keep(ret);
268 }
269 
270 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
271 			       int xerr)
272 {
273 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
274 	int err;
275 
276 	err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
277 	if (!err)
278 		ovl_dentry_set_opaque(dentry);
279 
280 	return err;
281 }
282 
283 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
284 {
285 	/*
286 	 * Fail with -EIO when trying to create opaque dir and upper doesn't
287 	 * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
288 	 * return a specific error for noxattr case.
289 	 */
290 	return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
291 }
292 
293 /*
294  * Common operations required to be done after creation of file on upper.
295  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
296  * may not use to instantiate the new dentry.
297  */
298 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
299 			   struct dentry *newdentry, bool hardlink, struct file *tmpfile)
300 {
301 	struct ovl_inode_params oip = {
302 		.upperdentry = newdentry,
303 		.newinode = inode,
304 	};
305 
306 	ovl_dentry_set_upper_alias(dentry);
307 	ovl_dentry_init_reval(dentry, newdentry, NULL);
308 
309 	if (!hardlink) {
310 		/*
311 		 * ovl_obtain_alias() can be called after ovl_create_real()
312 		 * and before we get here, so we may get an inode from cache
313 		 * with the same real upperdentry that is not the inode we
314 		 * pre-allocated.  In this case we will use the cached inode
315 		 * to instantiate the new dentry.
316 		 *
317 		 * XXX: if we ever use ovl_obtain_alias() to decode directory
318 		 * file handles, need to use ovl_get_inode_locked() and
319 		 * d_instantiate_new() here to prevent from creating two
320 		 * hashed directory inode aliases.  We then need to return
321 		 * the obtained alias to ovl_mkdir().
322 		 */
323 		inode = ovl_get_inode(dentry->d_sb, &oip);
324 		if (IS_ERR(inode))
325 			return PTR_ERR(inode);
326 		if (inode == oip.newinode)
327 			ovl_set_flag(OVL_UPPERDATA, inode);
328 	} else {
329 		WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
330 		dput(newdentry);
331 		inc_nlink(inode);
332 	}
333 
334 	if (tmpfile)
335 		d_mark_tmpfile(tmpfile, inode);
336 
337 	d_instantiate(dentry, inode);
338 	if (inode != oip.newinode) {
339 		pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
340 				    dentry);
341 	}
342 
343 	/* Force lookup of new upper hardlink to find its lower */
344 	if (hardlink)
345 		d_drop(dentry);
346 
347 	return 0;
348 }
349 
350 static bool ovl_type_merge(struct dentry *dentry)
351 {
352 	return OVL_TYPE_MERGE(ovl_path_type(dentry));
353 }
354 
355 static bool ovl_type_origin(struct dentry *dentry)
356 {
357 	return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
358 }
359 
360 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
361 			    struct ovl_cattr *attr)
362 {
363 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
364 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
365 	struct dentry *newdentry;
366 	struct qstr qname = QSTR_LEN(dentry->d_name.name,
367 				     dentry->d_name.len);
368 	int err;
369 
370 	newdentry = ovl_start_creating_upper(ofs, upperdir,
371 					     &qname);
372 	if (IS_ERR(newdentry))
373 		return PTR_ERR(newdentry);
374 	newdentry = ovl_create_real(ofs, upperdir, newdentry, &qname, attr);
375 	if (IS_ERR(newdentry))
376 		return PTR_ERR(newdentry);
377 
378 	end_creating_keep(newdentry);
379 
380 	if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
381 	    !ovl_allow_offline_changes(ofs)) {
382 		/* Setting opaque here is just an optimization, allow to fail */
383 		ovl_set_opaque(dentry, newdentry);
384 	}
385 
386 	ovl_dir_modified(dentry->d_parent, false);
387 	err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink, NULL);
388 	if (err)
389 		goto out_cleanup;
390 	return 0;
391 
392 out_cleanup:
393 	ovl_cleanup(ofs, upperdir, newdentry);
394 	dput(newdentry);
395 	return err;
396 }
397 
398 static struct dentry *ovl_clear_empty(struct dentry *dentry,
399 				      struct list_head *list)
400 {
401 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
402 	struct dentry *workdir = ovl_workdir(dentry);
403 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
404 	struct renamedata rd = {};
405 	struct path upperpath;
406 	struct dentry *upper;
407 	struct dentry *opaquedir;
408 	struct kstat stat;
409 	int err;
410 
411 	if (WARN_ON(!workdir))
412 		return ERR_PTR(-EROFS);
413 
414 	ovl_path_upper(dentry, &upperpath);
415 	err = vfs_getattr(&upperpath, &stat,
416 			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
417 	if (err)
418 		goto out;
419 
420 	err = -ESTALE;
421 	if (!S_ISDIR(stat.mode))
422 		goto out;
423 	upper = upperpath.dentry;
424 
425 	opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
426 	err = PTR_ERR(opaquedir);
427 	if (IS_ERR(opaquedir))
428 		goto out;
429 
430 	rd.mnt_idmap = ovl_upper_mnt_idmap(ofs);
431 	rd.old_parent = workdir;
432 	rd.new_parent = upperdir;
433 	rd.flags = RENAME_EXCHANGE;
434 	err = start_renaming_two_dentries(&rd, opaquedir, upper);
435 	if (err)
436 		goto out_cleanup_unlocked;
437 
438 	err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
439 	if (err)
440 		goto out_cleanup;
441 
442 	err = ovl_set_opaque(dentry, opaquedir);
443 	if (err)
444 		goto out_cleanup;
445 
446 	inode_lock(opaquedir->d_inode);
447 	err = ovl_set_attr(ofs, opaquedir, &stat);
448 	inode_unlock(opaquedir->d_inode);
449 	if (err)
450 		goto out_cleanup;
451 
452 	err = ovl_do_rename_rd(&rd);
453 	end_renaming(&rd);
454 	if (err)
455 		goto out_cleanup_unlocked;
456 
457 	ovl_cleanup_whiteouts(ofs, upper, list);
458 	ovl_cleanup(ofs, workdir, upper);
459 
460 	/* dentry's upper doesn't match now, get rid of it */
461 	d_drop(dentry);
462 
463 	return opaquedir;
464 
465 out_cleanup:
466 	end_renaming(&rd);
467 out_cleanup_unlocked:
468 	ovl_cleanup(ofs, workdir, opaquedir);
469 	dput(opaquedir);
470 out:
471 	return ERR_PTR(err);
472 }
473 
474 static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
475 			     const char *acl_name, struct posix_acl *acl)
476 {
477 	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
478 		return 0;
479 
480 	return ovl_do_set_acl(ofs, upperdentry, acl_name, acl);
481 }
482 
483 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
484 				    struct ovl_cattr *cattr)
485 {
486 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
487 	struct dentry *workdir = ovl_workdir(dentry);
488 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
489 	struct renamedata rd = {};
490 	struct dentry *upper;
491 	struct dentry *newdentry;
492 	int err;
493 	struct posix_acl *acl, *default_acl;
494 	bool hardlink = !!cattr->hardlink;
495 
496 	if (WARN_ON(!workdir))
497 		return -EROFS;
498 
499 	if (!hardlink) {
500 		err = posix_acl_create(dentry->d_parent->d_inode,
501 				       &cattr->mode, &default_acl, &acl);
502 		if (err)
503 			return err;
504 	}
505 
506 	upper = ovl_lookup_upper_unlocked(ofs, dentry->d_name.name, upperdir,
507 					  dentry->d_name.len);
508 	err = PTR_ERR(upper);
509 	if (IS_ERR(upper))
510 		goto out;
511 
512 	err = -ESTALE;
513 	if (d_is_negative(upper) || !ovl_upper_is_whiteout(ofs, upper))
514 		goto out_dput;
515 
516 	newdentry = ovl_create_temp(ofs, workdir, cattr);
517 	err = PTR_ERR(newdentry);
518 	if (IS_ERR(newdentry))
519 		goto out_dput;
520 
521 	rd.mnt_idmap = ovl_upper_mnt_idmap(ofs);
522 	rd.old_parent = workdir;
523 	rd.new_parent = upperdir;
524 	rd.flags = 0;
525 	err = start_renaming_two_dentries(&rd, newdentry, upper);
526 	if (err)
527 		goto out_cleanup_unlocked;
528 
529 	/*
530 	 * mode could have been mutilated due to umask (e.g. sgid directory)
531 	 */
532 	if (!hardlink &&
533 	    !S_ISLNK(cattr->mode) &&
534 	    newdentry->d_inode->i_mode != cattr->mode) {
535 		struct iattr attr = {
536 			.ia_valid = ATTR_MODE,
537 			.ia_mode = cattr->mode,
538 		};
539 		inode_lock(newdentry->d_inode);
540 		err = ovl_do_notify_change(ofs, newdentry, &attr);
541 		inode_unlock(newdentry->d_inode);
542 		if (err)
543 			goto out_cleanup;
544 	}
545 	if (!hardlink) {
546 		err = ovl_set_upper_acl(ofs, newdentry,
547 					XATTR_NAME_POSIX_ACL_ACCESS, acl);
548 		if (err)
549 			goto out_cleanup;
550 
551 		err = ovl_set_upper_acl(ofs, newdentry,
552 					XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
553 		if (err)
554 			goto out_cleanup;
555 	}
556 
557 	if (!hardlink && S_ISDIR(cattr->mode)) {
558 		err = ovl_set_opaque(dentry, newdentry);
559 		if (err)
560 			goto out_cleanup;
561 
562 		rd.flags = RENAME_EXCHANGE;
563 		err = ovl_do_rename_rd(&rd);
564 		end_renaming(&rd);
565 		if (err)
566 			goto out_cleanup_unlocked;
567 
568 		ovl_cleanup(ofs, workdir, upper);
569 	} else {
570 		err = ovl_do_rename_rd(&rd);
571 		end_renaming(&rd);
572 		if (err)
573 			goto out_cleanup_unlocked;
574 	}
575 	ovl_dir_modified(dentry->d_parent, false);
576 	err = ovl_instantiate(dentry, inode, newdentry, hardlink, NULL);
577 	if (err) {
578 		ovl_cleanup(ofs, upperdir, newdentry);
579 		dput(newdentry);
580 	}
581 out_dput:
582 	dput(upper);
583 out:
584 	if (!hardlink) {
585 		posix_acl_release(acl);
586 		posix_acl_release(default_acl);
587 	}
588 	return err;
589 
590 out_cleanup:
591 	end_renaming(&rd);
592 out_cleanup_unlocked:
593 	ovl_cleanup(ofs, workdir, newdentry);
594 	dput(newdentry);
595 	goto out_dput;
596 }
597 
598 static const struct cred *ovl_override_creator_creds(const struct cred *original_creds,
599 						     struct dentry *dentry, struct inode *inode, umode_t mode)
600 {
601 	int err;
602 
603 	if (WARN_ON_ONCE(current->cred != ovl_creds(dentry->d_sb)))
604 		return ERR_PTR(-EINVAL);
605 
606 	CLASS(prepare_creds, override_cred)();
607 	if (!override_cred)
608 		return ERR_PTR(-ENOMEM);
609 
610 	override_cred->fsuid = inode->i_uid;
611 	override_cred->fsgid = inode->i_gid;
612 
613 	err = security_dentry_create_files_as(dentry, mode, &dentry->d_name,
614 					      original_creds, override_cred);
615 	if (err)
616 		return ERR_PTR(err);
617 
618 	return override_creds(no_free_ptr(override_cred));
619 }
620 
621 static void ovl_revert_creator_creds(const struct cred *old_cred)
622 {
623 	const struct cred *override_cred;
624 
625 	override_cred = revert_creds(old_cred);
626 	put_cred(override_cred);
627 }
628 
629 DEFINE_CLASS(ovl_override_creator_creds,
630 	     const struct cred *,
631 	     if (!IS_ERR_OR_NULL(_T)) ovl_revert_creator_creds(_T),
632 	     ovl_override_creator_creds(original_creds, dentry, inode, mode),
633 	     const struct cred *original_creds,
634 	     struct dentry *dentry,
635 	     struct inode *inode,
636 	     umode_t mode)
637 
638 static int ovl_create_handle_whiteouts(struct dentry *dentry,
639 				       struct inode *inode,
640 				       struct ovl_cattr *attr)
641 {
642 	if (!ovl_dentry_is_whiteout(dentry))
643 		return ovl_create_upper(dentry, inode, attr);
644 
645 	return ovl_create_over_whiteout(dentry, inode, attr);
646 }
647 
648 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
649 			      struct ovl_cattr *attr, bool origin)
650 {
651 	int err;
652 	struct dentry *parent = dentry->d_parent;
653 
654 	scoped_class(override_creds_ovl, original_creds, dentry->d_sb) {
655 		/*
656 		 * When linking a file with copy up origin into a new parent, mark the
657 		 * new parent dir "impure".
658 		 */
659 		if (origin) {
660 			err = ovl_set_impure(parent, ovl_dentry_upper(parent));
661 			if (err)
662 				return err;
663 		}
664 
665 		/*
666 		 * In the creation cases(create, mkdir, mknod, symlink),
667 		 * ovl should transfer current's fs{u,g}id to underlying
668 		 * fs. Because underlying fs want to initialize its new
669 		 * inode owner using current's fs{u,g}id. And in this
670 		 * case, the @inode is a new inode that is initialized
671 		 * in inode_init_owner() to current's fs{u,g}id. So use
672 		 * the inode's i_{u,g}id to override the cred's fs{u,g}id.
673 		 *
674 		 * But in the other hardlink case, ovl_link() does not
675 		 * create a new inode, so just use the ovl mounter's
676 		 * fs{u,g}id.
677 		 */
678 
679 		if (attr->hardlink)
680 			return ovl_create_handle_whiteouts(dentry, inode, attr);
681 
682 		scoped_class(ovl_override_creator_creds, cred, original_creds, dentry, inode, attr->mode) {
683 			if (IS_ERR(cred))
684 				return PTR_ERR(cred);
685 			return ovl_create_handle_whiteouts(dentry, inode, attr);
686 		}
687 	}
688 	return err;
689 }
690 
691 static int ovl_create_object(struct mnt_idmap *idmap, struct dentry *dentry,
692 			     int mode, dev_t rdev, const char *link)
693 {
694 	int err;
695 	struct inode *inode;
696 	struct ovl_cattr attr = {
697 		.rdev = rdev,
698 		.link = link,
699 	};
700 
701 	err = ovl_copy_up(dentry->d_parent);
702 	if (err)
703 		return err;
704 
705 	err = ovl_want_write(dentry);
706 	if (err)
707 		goto out;
708 
709 	/* Preallocate inode to be used by ovl_get_inode() */
710 	err = -ENOMEM;
711 	inode = ovl_new_inode(dentry->d_sb, mode, rdev);
712 	if (!inode)
713 		goto out_drop_write;
714 
715 	spin_lock(&inode->i_lock);
716 	inode_state_set(inode, I_CREATING);
717 	spin_unlock(&inode->i_lock);
718 
719 	inode_init_owner(idmap, inode, dentry->d_parent->d_inode, mode);
720 	attr.mode = inode->i_mode;
721 
722 	err = ovl_create_or_link(dentry, inode, &attr, false);
723 	/* Did we end up using the preallocated inode? */
724 	if (inode != d_inode(dentry))
725 		iput(inode);
726 
727 out_drop_write:
728 	ovl_drop_write(dentry);
729 out:
730 	return err;
731 }
732 
733 static int ovl_create(struct mnt_idmap *idmap, struct inode *dir,
734 		      struct dentry *dentry, umode_t mode)
735 {
736 	return ovl_create_object(idmap, dentry, (mode & 07777) | S_IFREG, 0, NULL);
737 }
738 
739 static struct dentry *ovl_mkdir(struct mnt_idmap *idmap, struct inode *dir,
740 				struct dentry *dentry, umode_t mode)
741 {
742 	return ERR_PTR(ovl_create_object(idmap, dentry, (mode & 07777) | S_IFDIR, 0, NULL));
743 }
744 
745 static int ovl_mknod(struct mnt_idmap *idmap, struct inode *dir,
746 		     struct dentry *dentry, umode_t mode, dev_t rdev)
747 {
748 	/* Don't allow creation of "whiteout" on overlay */
749 	if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
750 		return -EPERM;
751 
752 	return ovl_create_object(idmap, dentry, mode, rdev, NULL);
753 }
754 
755 static int ovl_symlink(struct mnt_idmap *idmap, struct inode *dir,
756 		       struct dentry *dentry, const char *link)
757 {
758 	return ovl_create_object(idmap, dentry, S_IFLNK, 0, link);
759 }
760 
761 static int ovl_set_link_redirect(struct dentry *dentry)
762 {
763 	with_ovl_creds(dentry->d_sb)
764 		return ovl_set_redirect(dentry, false);
765 }
766 
767 static int ovl_link(struct dentry *old, struct inode *newdir,
768 		    struct dentry *new)
769 {
770 	int err;
771 	struct inode *inode;
772 
773 	err = ovl_copy_up(old);
774 	if (err)
775 		goto out;
776 
777 	err = ovl_copy_up(new->d_parent);
778 	if (err)
779 		goto out;
780 
781 	err = ovl_nlink_start(old);
782 	if (err)
783 		goto out;
784 
785 	if (ovl_is_metacopy_dentry(old)) {
786 		err = ovl_set_link_redirect(old);
787 		if (err)
788 			goto out_nlink_end;
789 	}
790 
791 	inode = d_inode(old);
792 	ihold(inode);
793 
794 	err = ovl_create_or_link(new, inode,
795 			&(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
796 			ovl_type_origin(old));
797 	if (err)
798 		iput(inode);
799 
800 out_nlink_end:
801 	ovl_nlink_end(old);
802 out:
803 	return err;
804 }
805 
806 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
807 {
808 	return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
809 }
810 
811 static int ovl_remove_and_whiteout(struct dentry *dentry,
812 				   struct list_head *list)
813 {
814 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
815 	struct dentry *workdir = ovl_workdir(dentry);
816 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
817 	struct dentry *upper;
818 	struct dentry *opaquedir = NULL;
819 	int err;
820 
821 	if (WARN_ON(!workdir))
822 		return -EROFS;
823 
824 	if (!list_empty(list)) {
825 		opaquedir = ovl_clear_empty(dentry, list);
826 		err = PTR_ERR(opaquedir);
827 		if (IS_ERR(opaquedir))
828 			goto out;
829 	}
830 
831 	upper = ovl_lookup_upper_unlocked(ofs, dentry->d_name.name, upperdir,
832 					  dentry->d_name.len);
833 	err = PTR_ERR(upper);
834 	if (IS_ERR(upper))
835 		goto out_dput;
836 
837 	err = -ESTALE;
838 	if ((opaquedir && upper != opaquedir) ||
839 	    (!opaquedir && ovl_dentry_upper(dentry) &&
840 	     !ovl_matches_upper(dentry, upper))) {
841 		goto out_dput_upper;
842 	}
843 
844 	err = ovl_cleanup_and_whiteout(ofs, upperdir, upper);
845 	if (!err)
846 		ovl_dir_modified(dentry->d_parent, true);
847 
848 	d_drop(dentry);
849 out_dput_upper:
850 	dput(upper);
851 out_dput:
852 	dput(opaquedir);
853 out:
854 	return err;
855 }
856 
857 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
858 			    struct list_head *list)
859 {
860 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
861 	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
862 	struct inode *dir = upperdir->d_inode;
863 	struct dentry *upper;
864 	struct dentry *opaquedir = NULL;
865 	int err;
866 
867 	if (!list_empty(list)) {
868 		opaquedir = ovl_clear_empty(dentry, list);
869 		err = PTR_ERR(opaquedir);
870 		if (IS_ERR(opaquedir))
871 			goto out;
872 	}
873 
874 	upper = ovl_start_removing_upper(ofs, upperdir,
875 					 &QSTR_LEN(dentry->d_name.name,
876 						   dentry->d_name.len));
877 	err = PTR_ERR(upper);
878 	if (IS_ERR(upper))
879 		goto out_dput;
880 
881 	err = -ESTALE;
882 	if ((opaquedir && upper != opaquedir) ||
883 	    (!opaquedir && !ovl_matches_upper(dentry, upper)))
884 		goto out_unlock;
885 
886 	if (is_dir)
887 		err = ovl_do_rmdir(ofs, dir, upper);
888 	else
889 		err = ovl_do_unlink(ofs, dir, upper);
890 	ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
891 
892 	/*
893 	 * Keeping this dentry hashed would mean having to release
894 	 * upperpath/lowerpath, which could only be done if we are the
895 	 * sole user of this dentry.  Too tricky...  Just unhash for
896 	 * now.
897 	 */
898 	if (!err)
899 		d_drop(dentry);
900 out_unlock:
901 	end_removing(upper);
902 out_dput:
903 	dput(opaquedir);
904 out:
905 	return err;
906 }
907 
908 static bool ovl_pure_upper(struct dentry *dentry)
909 {
910 	return !ovl_dentry_lower(dentry) &&
911 	       !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
912 }
913 
914 static void ovl_drop_nlink(struct dentry *dentry)
915 {
916 	struct inode *inode = d_inode(dentry);
917 	struct dentry *alias;
918 
919 	/* Try to find another, hashed alias */
920 	spin_lock(&inode->i_lock);
921 	for_each_alias(alias, inode) {
922 		if (alias != dentry && !d_unhashed(alias))
923 			break;
924 	}
925 	spin_unlock(&inode->i_lock);
926 
927 	/*
928 	 * Changes to underlying layers may cause i_nlink to lose sync with
929 	 * reality.  In this case prevent the link count from going to zero
930 	 * prematurely.
931 	 */
932 	if (inode->i_nlink > !!alias)
933 		drop_nlink(inode);
934 }
935 
936 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
937 {
938 	int err;
939 	bool lower_positive = ovl_lower_positive(dentry);
940 	LIST_HEAD(list);
941 
942 	/* No need to clean pure upper removed by vfs_rmdir() */
943 	if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
944 		err = ovl_check_empty_dir(dentry, &list);
945 		if (err)
946 			goto out;
947 	}
948 
949 	err = ovl_copy_up(dentry->d_parent);
950 	if (err)
951 		goto out;
952 
953 	err = ovl_nlink_start(dentry);
954 	if (err)
955 		goto out;
956 
957 	with_ovl_creds(dentry->d_sb) {
958 		if (!lower_positive)
959 			err = ovl_remove_upper(dentry, is_dir, &list);
960 		else
961 			err = ovl_remove_and_whiteout(dentry, &list);
962 	}
963 	if (!err) {
964 		if (is_dir)
965 			clear_nlink(dentry->d_inode);
966 		else
967 			ovl_drop_nlink(dentry);
968 	}
969 	ovl_nlink_end(dentry);
970 
971 	/*
972 	 * Copy ctime
973 	 *
974 	 * Note: we fail to update ctime if there was no copy-up, only a
975 	 * whiteout
976 	 */
977 	if (ovl_dentry_upper(dentry))
978 		ovl_copyattr(d_inode(dentry));
979 
980 out:
981 	ovl_cache_free(&list);
982 	return err;
983 }
984 
985 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
986 {
987 	return ovl_do_remove(dentry, false);
988 }
989 
990 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
991 {
992 	return ovl_do_remove(dentry, true);
993 }
994 
995 static bool ovl_type_merge_or_lower(struct dentry *dentry)
996 {
997 	enum ovl_path_type type = ovl_path_type(dentry);
998 
999 	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
1000 }
1001 
1002 static bool ovl_can_move(struct dentry *dentry)
1003 {
1004 	return ovl_redirect_dir(OVL_FS(dentry->d_sb)) ||
1005 		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
1006 }
1007 
1008 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
1009 {
1010 	char *buf, *ret;
1011 	struct dentry *d, *tmp;
1012 	int buflen = ovl_redirect_max + 1;
1013 
1014 	if (!abs_redirect) {
1015 		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
1016 			       GFP_KERNEL);
1017 		goto out;
1018 	}
1019 
1020 	buf = ret = kmalloc(buflen, GFP_KERNEL);
1021 	if (!buf)
1022 		goto out;
1023 
1024 	buflen--;
1025 	buf[buflen] = '\0';
1026 	for (d = dget(dentry); !IS_ROOT(d);) {
1027 		const char *name;
1028 		int thislen;
1029 
1030 		spin_lock(&d->d_lock);
1031 		name = ovl_dentry_get_redirect(d);
1032 		if (name) {
1033 			thislen = strlen(name);
1034 		} else {
1035 			name = d->d_name.name;
1036 			thislen = d->d_name.len;
1037 		}
1038 
1039 		/* If path is too long, fall back to userspace move */
1040 		if (thislen + (name[0] != '/') > buflen) {
1041 			ret = ERR_PTR(-EXDEV);
1042 			spin_unlock(&d->d_lock);
1043 			goto out_put;
1044 		}
1045 
1046 		buflen -= thislen;
1047 		memcpy(&buf[buflen], name, thislen);
1048 		spin_unlock(&d->d_lock);
1049 		tmp = dget_parent(d);
1050 
1051 		dput(d);
1052 		d = tmp;
1053 
1054 		/* Absolute redirect: finished */
1055 		if (buf[buflen] == '/')
1056 			break;
1057 		buflen--;
1058 		buf[buflen] = '/';
1059 	}
1060 	ret = kstrdup(&buf[buflen], GFP_KERNEL);
1061 out_put:
1062 	dput(d);
1063 	kfree(buf);
1064 out:
1065 	return ret ? ret : ERR_PTR(-ENOMEM);
1066 }
1067 
1068 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1069 {
1070 	struct dentry *lowerdentry;
1071 
1072 	if (!samedir)
1073 		return true;
1074 
1075 	if (d_is_dir(dentry))
1076 		return false;
1077 
1078 	/*
1079 	 * For non-dir hardlinked files, we need absolute redirects
1080 	 * in general as two upper hardlinks could be in different
1081 	 * dirs. We could put a relative redirect now and convert
1082 	 * it to absolute redirect later. But when nlink > 1 and
1083 	 * indexing is on, that means relative redirect needs to be
1084 	 * converted to absolute during copy up of another lower
1085 	 * hardllink as well.
1086 	 *
1087 	 * So without optimizing too much, just check if lower is
1088 	 * a hard link or not. If lower is hard link, put absolute
1089 	 * redirect.
1090 	 */
1091 	lowerdentry = ovl_dentry_lower(dentry);
1092 	return (d_inode(lowerdentry)->i_nlink > 1);
1093 }
1094 
1095 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1096 {
1097 	int err;
1098 	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1099 	const char *redirect = ovl_dentry_get_redirect(dentry);
1100 	bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1101 
1102 	if (redirect && (!absolute_redirect || redirect[0] == '/'))
1103 		return 0;
1104 
1105 	redirect = ovl_get_redirect(dentry, absolute_redirect);
1106 	if (IS_ERR(redirect))
1107 		return PTR_ERR(redirect);
1108 
1109 	err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1110 				 OVL_XATTR_REDIRECT,
1111 				 redirect, strlen(redirect), -EXDEV);
1112 	if (!err) {
1113 		spin_lock(&dentry->d_lock);
1114 		ovl_dentry_set_redirect(dentry, redirect);
1115 		spin_unlock(&dentry->d_lock);
1116 	} else {
1117 		kfree(redirect);
1118 		pr_warn_ratelimited("failed to set redirect (%i)\n",
1119 				    err);
1120 		/* Fall back to userspace copy-up */
1121 		err = -EXDEV;
1122 	}
1123 	return err;
1124 }
1125 
1126 struct ovl_renamedata {
1127 	struct renamedata;
1128 	struct dentry *opaquedir;
1129 	bool cleanup_whiteout;
1130 	bool update_nlink;
1131 	bool overwrite;
1132 };
1133 
1134 static int ovl_rename_start(struct ovl_renamedata *ovlrd, struct list_head *list)
1135 {
1136 	struct dentry *old = ovlrd->old_dentry;
1137 	struct dentry *new = ovlrd->new_dentry;
1138 	bool is_dir = d_is_dir(old);
1139 	bool new_is_dir = d_is_dir(new);
1140 	int err;
1141 
1142 	if (ovlrd->flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1143 		return -EINVAL;
1144 
1145 	ovlrd->flags &= ~RENAME_NOREPLACE;
1146 
1147 	/* Don't copy up directory trees */
1148 	err = -EXDEV;
1149 	if (!ovl_can_move(old))
1150 		return err;
1151 	if (!ovlrd->overwrite && !ovl_can_move(new))
1152 		return err;
1153 
1154 	if (ovlrd->overwrite && new_is_dir && !ovl_pure_upper(new)) {
1155 		err = ovl_check_empty_dir(new, list);
1156 		if (err)
1157 			return err;
1158 	}
1159 
1160 	if (ovlrd->overwrite) {
1161 		if (ovl_lower_positive(old)) {
1162 			if (!ovl_dentry_is_whiteout(new)) {
1163 				/* Whiteout source */
1164 				ovlrd->flags |= RENAME_WHITEOUT;
1165 			} else {
1166 				/* Switch whiteouts */
1167 				ovlrd->flags |= RENAME_EXCHANGE;
1168 			}
1169 		} else if (is_dir && ovl_dentry_is_whiteout(new)) {
1170 			ovlrd->flags |= RENAME_EXCHANGE;
1171 			ovlrd->cleanup_whiteout = true;
1172 		}
1173 	}
1174 
1175 	err = ovl_copy_up(old);
1176 	if (err)
1177 		return err;
1178 
1179 	err = ovl_copy_up(new->d_parent);
1180 	if (err)
1181 		return err;
1182 
1183 	if (!ovlrd->overwrite) {
1184 		err = ovl_copy_up(new);
1185 		if (err)
1186 			return err;
1187 	} else if (d_inode(new)) {
1188 		err = ovl_nlink_start(new);
1189 		if (err)
1190 			return err;
1191 
1192 		ovlrd->update_nlink = true;
1193 	}
1194 
1195 	if (!ovlrd->update_nlink) {
1196 		/* ovl_nlink_start() took ovl_want_write() */
1197 		err = ovl_want_write(old);
1198 		if (err)
1199 			return err;
1200 	}
1201 
1202 	return 0;
1203 }
1204 
1205 static int ovl_rename_upper(struct ovl_renamedata *ovlrd, struct list_head *list)
1206 {
1207 	struct dentry *old = ovlrd->old_dentry;
1208 	struct dentry *new = ovlrd->new_dentry;
1209 	struct ovl_fs *ofs = OVL_FS(old->d_sb);
1210 	struct dentry *old_upperdir = ovl_dentry_upper(old->d_parent);
1211 	struct dentry *new_upperdir = ovl_dentry_upper(new->d_parent);
1212 	bool is_dir = d_is_dir(old);
1213 	bool new_is_dir = d_is_dir(new);
1214 	bool samedir = old->d_parent == new->d_parent;
1215 	struct renamedata rd = {};
1216 	struct dentry *de;
1217 	struct dentry *whiteout = NULL;
1218 	bool old_opaque, new_opaque;
1219 	int err;
1220 
1221 	if (!list_empty(list)) {
1222 		de = ovl_clear_empty(new, list);
1223 		if (IS_ERR(de))
1224 			return PTR_ERR(de);
1225 		ovlrd->opaquedir = de;
1226 	}
1227 
1228 	if (!samedir) {
1229 		/*
1230 		 * When moving a merge dir or non-dir with copy up origin into
1231 		 * a new parent, we are marking the new parent dir "impure".
1232 		 * When ovl_iterate() iterates an "impure" upper dir, it will
1233 		 * lookup the origin inodes of the entries to fill d_ino.
1234 		 */
1235 		if (ovl_type_origin(old)) {
1236 			err = ovl_set_impure(new->d_parent, new_upperdir);
1237 			if (err)
1238 				return err;
1239 		}
1240 		if (!ovlrd->overwrite && ovl_type_origin(new)) {
1241 			err = ovl_set_impure(old->d_parent, old_upperdir);
1242 			if (err)
1243 				return err;
1244 		}
1245 	}
1246 
1247 	rd.mnt_idmap = ovl_upper_mnt_idmap(ofs);
1248 	rd.old_parent = old_upperdir;
1249 	rd.new_parent = new_upperdir;
1250 	rd.flags = ovlrd->flags;
1251 
1252 	err = start_renaming(&rd, 0,
1253 			     &QSTR_LEN(old->d_name.name, old->d_name.len),
1254 			     &QSTR_LEN(new->d_name.name, new->d_name.len));
1255 	if (err)
1256 		return err;
1257 
1258 	err = -ESTALE;
1259 	if (!ovl_matches_upper(old, rd.old_dentry))
1260 		goto out_unlock;
1261 
1262 	old_opaque = ovl_dentry_is_opaque(old);
1263 	new_opaque = ovl_dentry_is_opaque(new);
1264 
1265 	err = -ESTALE;
1266 	if (d_inode(new) && ovl_dentry_upper(new)) {
1267 		if (ovlrd->opaquedir) {
1268 			if (rd.new_dentry != ovlrd->opaquedir)
1269 				goto out_unlock;
1270 		} else {
1271 			if (!ovl_matches_upper(new, rd.new_dentry))
1272 				goto out_unlock;
1273 		}
1274 	} else {
1275 		if (!d_is_negative(rd.new_dentry)) {
1276 			if (!new_opaque || !ovl_upper_is_whiteout(ofs, rd.new_dentry))
1277 				goto out_unlock;
1278 		} else {
1279 			if (ovlrd->flags & RENAME_EXCHANGE)
1280 				goto out_unlock;
1281 		}
1282 	}
1283 
1284 	if (rd.old_dentry->d_inode == rd.new_dentry->d_inode)
1285 		goto out_unlock;
1286 
1287 	err = 0;
1288 	if (ovl_type_merge_or_lower(old))
1289 		err = ovl_set_redirect(old, samedir);
1290 	else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1291 		err = ovl_set_opaque_xerr(old, rd.old_dentry, -EXDEV);
1292 	if (err)
1293 		goto out_unlock;
1294 
1295 	if (!ovlrd->overwrite && ovl_type_merge_or_lower(new))
1296 		err = ovl_set_redirect(new, samedir);
1297 	else if (!ovlrd->overwrite && new_is_dir && !new_opaque &&
1298 		 ovl_type_merge(old->d_parent))
1299 		err = ovl_set_opaque_xerr(new, rd.new_dentry, -EXDEV);
1300 	if (err)
1301 		goto out_unlock;
1302 
1303 	err = ovl_do_rename_rd(&rd);
1304 
1305 	if (!err && ovlrd->cleanup_whiteout)
1306 		whiteout = dget(rd.new_dentry);
1307 
1308 out_unlock:
1309 	end_renaming(&rd);
1310 
1311 	if (err)
1312 		return err;
1313 
1314 	if (whiteout) {
1315 		ovl_cleanup(ofs, old_upperdir, whiteout);
1316 		dput(whiteout);
1317 	}
1318 
1319 	if (ovlrd->overwrite && d_inode(new)) {
1320 		if (new_is_dir)
1321 			clear_nlink(d_inode(new));
1322 		else
1323 			ovl_drop_nlink(new);
1324 	}
1325 
1326 	ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1327 			 (!ovlrd->overwrite && ovl_type_origin(new)));
1328 	ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1329 			 (d_inode(new) && ovl_type_origin(new)));
1330 
1331 	/* copy ctime: */
1332 	ovl_copyattr(d_inode(old));
1333 	if (d_inode(new) && ovl_dentry_upper(new))
1334 		ovl_copyattr(d_inode(new));
1335 
1336 	return err;
1337 }
1338 
1339 static void ovl_rename_end(struct ovl_renamedata *ovlrd)
1340 {
1341 	if (ovlrd->update_nlink)
1342 		ovl_nlink_end(ovlrd->new_dentry);
1343 	else
1344 		ovl_drop_write(ovlrd->old_dentry);
1345 }
1346 
1347 static int ovl_rename(struct mnt_idmap *idmap, struct inode *olddir,
1348 		      struct dentry *old, struct inode *newdir,
1349 		      struct dentry *new, unsigned int flags)
1350 {
1351 	struct ovl_renamedata ovlrd = {
1352 		.old_parent		= old->d_parent,
1353 		.old_dentry		= old,
1354 		.new_parent		= new->d_parent,
1355 		.new_dentry		= new,
1356 		.flags			= flags,
1357 		.overwrite		= !(flags & RENAME_EXCHANGE),
1358 	};
1359 	LIST_HEAD(list);
1360 	int err;
1361 
1362 	err = ovl_rename_start(&ovlrd, &list);
1363 	if (!err) {
1364 		with_ovl_creds(old->d_sb)
1365 			err = ovl_rename_upper(&ovlrd, &list);
1366 		ovl_rename_end(&ovlrd);
1367 	}
1368 
1369 	dput(ovlrd.opaquedir);
1370 	ovl_cache_free(&list);
1371 	return err;
1372 }
1373 
1374 static int ovl_create_tmpfile(struct file *file, struct dentry *dentry,
1375 			      struct inode *inode, umode_t mode)
1376 {
1377 	struct path realparentpath;
1378 	struct file *realfile;
1379 	struct ovl_file *of;
1380 	struct dentry *newdentry;
1381 	/* It's okay to set O_NOATIME, since the owner will be current fsuid */
1382 	int flags = file->f_flags | OVL_OPEN_FLAGS;
1383 	int err;
1384 
1385 	scoped_class(override_creds_ovl, original_creds, dentry->d_sb) {
1386 		scoped_class(ovl_override_creator_creds, cred, original_creds, dentry, inode, mode) {
1387 			if (IS_ERR(cred))
1388 				return PTR_ERR(cred);
1389 
1390 			ovl_path_upper(dentry->d_parent, &realparentpath);
1391 			realfile = backing_tmpfile_open(file, flags, &realparentpath,
1392 							mode, current_cred());
1393 			err = PTR_ERR_OR_ZERO(realfile);
1394 			pr_debug("tmpfile/open(%pd2, 0%o) = %i\n", realparentpath.dentry, mode, err);
1395 			if (err)
1396 				return err;
1397 
1398 			of = ovl_file_alloc(realfile);
1399 			if (!of) {
1400 				fput(realfile);
1401 				return -ENOMEM;
1402 			}
1403 
1404 			/* ovl_instantiate() consumes the newdentry reference on success */
1405 			newdentry = dget(realfile->f_path.dentry);
1406 			err = ovl_instantiate(dentry, inode, newdentry, false, file);
1407 			if (!err) {
1408 				file->private_data = of;
1409 			} else {
1410 				dput(newdentry);
1411 				ovl_file_free(of);
1412 			}
1413 		}
1414 	}
1415 	return err;
1416 }
1417 
1418 static int ovl_dummy_open(struct inode *inode, struct file *file)
1419 {
1420 	return 0;
1421 }
1422 
1423 static int ovl_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
1424 		       struct file *file, umode_t mode)
1425 {
1426 	int err;
1427 	struct dentry *dentry = file->f_path.dentry;
1428 	struct inode *inode;
1429 
1430 	if (!OVL_FS(dentry->d_sb)->tmpfile)
1431 		return -EOPNOTSUPP;
1432 
1433 	err = ovl_copy_up(dentry->d_parent);
1434 	if (err)
1435 		return err;
1436 
1437 	err = ovl_want_write(dentry);
1438 	if (err)
1439 		return err;
1440 
1441 	err = -ENOMEM;
1442 	inode = ovl_new_inode(dentry->d_sb, mode, 0);
1443 	if (!inode)
1444 		goto drop_write;
1445 
1446 	inode_init_owner(idmap, inode, dir, mode);
1447 	err = ovl_create_tmpfile(file, dentry, inode, inode->i_mode);
1448 	if (err)
1449 		goto put_inode;
1450 
1451 	/*
1452 	 * Check if the preallocated inode was actually used.  Having something
1453 	 * else assigned to the dentry shouldn't happen as that would indicate
1454 	 * that the backing tmpfile "leaked" out of overlayfs.
1455 	 */
1456 	err = -EIO;
1457 	if (WARN_ON(inode != d_inode(dentry)))
1458 		goto put_realfile;
1459 
1460 	/* inode reference was transferred to dentry */
1461 	inode = NULL;
1462 	err = finish_open(file, dentry, ovl_dummy_open);
1463 put_realfile:
1464 	/* Without FMODE_OPENED ->release() won't be called on @file */
1465 	if (!(file->f_mode & FMODE_OPENED))
1466 		ovl_file_free(file->private_data);
1467 put_inode:
1468 	iput(inode);
1469 drop_write:
1470 	ovl_drop_write(dentry);
1471 	return err;
1472 }
1473 
1474 const struct inode_operations ovl_dir_inode_operations = {
1475 	.lookup		= ovl_lookup,
1476 	.mkdir		= ovl_mkdir,
1477 	.symlink	= ovl_symlink,
1478 	.unlink		= ovl_unlink,
1479 	.rmdir		= ovl_rmdir,
1480 	.rename		= ovl_rename,
1481 	.link		= ovl_link,
1482 	.setattr	= ovl_setattr,
1483 	.create		= ovl_create,
1484 	.mknod		= ovl_mknod,
1485 	.permission	= ovl_permission,
1486 	.getattr	= ovl_getattr,
1487 	.listxattr	= ovl_listxattr,
1488 	.get_inode_acl	= ovl_get_inode_acl,
1489 	.get_acl	= ovl_get_acl,
1490 	.set_acl	= ovl_set_acl,
1491 	.update_time	= ovl_update_time,
1492 	.fileattr_get	= ovl_fileattr_get,
1493 	.fileattr_set	= ovl_fileattr_set,
1494 	.tmpfile	= ovl_tmpfile,
1495 };
1496