dir.c revision 52d5a0c6
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 "overlayfs.h"
18
19static unsigned short ovl_redirect_max = 256;
20module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
21MODULE_PARM_DESC(redirect_max,
22		 "Maximum length of absolute redirect xattr value");
23
24static int ovl_set_redirect(struct dentry *dentry, bool samedir);
25
26int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
27{
28	int err;
29
30	dget(wdentry);
31	if (d_is_dir(wdentry))
32		err = ovl_do_rmdir(wdir, wdentry);
33	else
34		err = ovl_do_unlink(wdir, wdentry);
35	dput(wdentry);
36
37	if (err) {
38		pr_err("cleanup of '%pd2' failed (%i)\n",
39		       wdentry, err);
40	}
41
42	return err;
43}
44
45struct dentry *ovl_lookup_temp(struct dentry *workdir)
46{
47	struct dentry *temp;
48	char name[20];
49	static atomic_t temp_id = ATOMIC_INIT(0);
50
51	/* counter is allowed to wrap, since temp dentries are ephemeral */
52	snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
53
54	temp = lookup_one_len(name, workdir, strlen(name));
55	if (!IS_ERR(temp) && temp->d_inode) {
56		pr_err("workdir/%s already exists\n", name);
57		dput(temp);
58		temp = ERR_PTR(-EIO);
59	}
60
61	return temp;
62}
63
64/* caller holds i_mutex on workdir */
65static struct dentry *ovl_whiteout(struct ovl_fs *ofs)
66{
67	int err;
68	struct dentry *whiteout;
69	struct dentry *workdir = ofs->workdir;
70	struct inode *wdir = workdir->d_inode;
71
72	if (!ofs->whiteout) {
73		whiteout = ovl_lookup_temp(workdir);
74		if (IS_ERR(whiteout))
75			goto out;
76
77		err = ovl_do_whiteout(wdir, whiteout);
78		if (err) {
79			dput(whiteout);
80			whiteout = ERR_PTR(err);
81			goto out;
82		}
83		ofs->whiteout = whiteout;
84	}
85
86	if (ofs->share_whiteout) {
87		whiteout = ovl_lookup_temp(workdir);
88		if (IS_ERR(whiteout))
89			goto out;
90
91		err = ovl_do_link(ofs->whiteout, wdir, whiteout);
92		if (!err)
93			goto out;
94
95		if (err != -EMLINK) {
96			pr_warn("Failed to link whiteout - disabling whiteout inode sharing(nlink=%u, err=%i)\n",
97				ofs->whiteout->d_inode->i_nlink, err);
98			ofs->share_whiteout = false;
99		}
100		dput(whiteout);
101	}
102	whiteout = ofs->whiteout;
103	ofs->whiteout = NULL;
104out:
105	return whiteout;
106}
107
108/* Caller must hold i_mutex on both workdir and dir */
109int ovl_cleanup_and_whiteout(struct ovl_fs *ofs, struct inode *dir,
110			     struct dentry *dentry)
111{
112	struct inode *wdir = ofs->workdir->d_inode;
113	struct dentry *whiteout;
114	int err;
115	int flags = 0;
116
117	whiteout = ovl_whiteout(ofs);
118	err = PTR_ERR(whiteout);
119	if (IS_ERR(whiteout))
120		return err;
121
122	if (d_is_dir(dentry))
123		flags = RENAME_EXCHANGE;
124
125	err = ovl_do_rename(wdir, whiteout, dir, dentry, flags);
126	if (err)
127		goto kill_whiteout;
128	if (flags)
129		ovl_cleanup(wdir, dentry);
130
131out:
132	dput(whiteout);
133	return err;
134
135kill_whiteout:
136	ovl_cleanup(wdir, whiteout);
137	goto out;
138}
139
140static int ovl_mkdir_real(struct inode *dir, struct dentry **newdentry,
141			  umode_t mode)
142{
143	int err;
144	struct dentry *d, *dentry = *newdentry;
145
146	err = ovl_do_mkdir(dir, dentry, mode);
147	if (err)
148		return err;
149
150	if (likely(!d_unhashed(dentry)))
151		return 0;
152
153	/*
154	 * vfs_mkdir() may succeed and leave the dentry passed
155	 * to it unhashed and negative. If that happens, try to
156	 * lookup a new hashed and positive dentry.
157	 */
158	d = lookup_one_len(dentry->d_name.name, dentry->d_parent,
159			   dentry->d_name.len);
160	if (IS_ERR(d)) {
161		pr_warn("failed lookup after mkdir (%pd2, err=%i).\n",
162			dentry, err);
163		return PTR_ERR(d);
164	}
165	dput(dentry);
166	*newdentry = d;
167
168	return 0;
169}
170
171struct dentry *ovl_create_real(struct inode *dir, struct dentry *newdentry,
172			       struct ovl_cattr *attr)
173{
174	int err;
175
176	if (IS_ERR(newdentry))
177		return newdentry;
178
179	err = -ESTALE;
180	if (newdentry->d_inode)
181		goto out;
182
183	if (attr->hardlink) {
184		err = ovl_do_link(attr->hardlink, dir, newdentry);
185	} else {
186		switch (attr->mode & S_IFMT) {
187		case S_IFREG:
188			err = ovl_do_create(dir, newdentry, attr->mode);
189			break;
190
191		case S_IFDIR:
192			/* mkdir is special... */
193			err =  ovl_mkdir_real(dir, &newdentry, attr->mode);
194			break;
195
196		case S_IFCHR:
197		case S_IFBLK:
198		case S_IFIFO:
199		case S_IFSOCK:
200			err = ovl_do_mknod(dir, newdentry, attr->mode,
201					   attr->rdev);
202			break;
203
204		case S_IFLNK:
205			err = ovl_do_symlink(dir, newdentry, attr->link);
206			break;
207
208		default:
209			err = -EPERM;
210		}
211	}
212	if (!err && WARN_ON(!newdentry->d_inode)) {
213		/*
214		 * Not quite sure if non-instantiated dentry is legal or not.
215		 * VFS doesn't seem to care so check and warn here.
216		 */
217		err = -EIO;
218	}
219out:
220	if (err) {
221		dput(newdentry);
222		return ERR_PTR(err);
223	}
224	return newdentry;
225}
226
227struct dentry *ovl_create_temp(struct dentry *workdir, struct ovl_cattr *attr)
228{
229	return ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir),
230			       attr);
231}
232
233static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
234			       int xerr)
235{
236	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
237	int err;
238
239	err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
240	if (!err)
241		ovl_dentry_set_opaque(dentry);
242
243	return err;
244}
245
246static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
247{
248	/*
249	 * Fail with -EIO when trying to create opaque dir and upper doesn't
250	 * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
251	 * return a specific error for noxattr case.
252	 */
253	return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
254}
255
256/*
257 * Common operations required to be done after creation of file on upper.
258 * If @hardlink is false, then @inode is a pre-allocated inode, we may or
259 * may not use to instantiate the new dentry.
260 */
261static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
262			   struct dentry *newdentry, bool hardlink)
263{
264	struct ovl_inode_params oip = {
265		.upperdentry = newdentry,
266		.newinode = inode,
267	};
268
269	ovl_dir_modified(dentry->d_parent, false);
270	ovl_dentry_set_upper_alias(dentry);
271	ovl_dentry_update_reval(dentry, newdentry,
272			DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
273
274	if (!hardlink) {
275		/*
276		 * ovl_obtain_alias() can be called after ovl_create_real()
277		 * and before we get here, so we may get an inode from cache
278		 * with the same real upperdentry that is not the inode we
279		 * pre-allocated.  In this case we will use the cached inode
280		 * to instantiate the new dentry.
281		 *
282		 * XXX: if we ever use ovl_obtain_alias() to decode directory
283		 * file handles, need to use ovl_get_inode_locked() and
284		 * d_instantiate_new() here to prevent from creating two
285		 * hashed directory inode aliases.
286		 */
287		inode = ovl_get_inode(dentry->d_sb, &oip);
288		if (IS_ERR(inode))
289			return PTR_ERR(inode);
290		if (inode == oip.newinode)
291			ovl_set_flag(OVL_UPPERDATA, inode);
292	} else {
293		WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
294		dput(newdentry);
295		inc_nlink(inode);
296	}
297
298	d_instantiate(dentry, inode);
299	if (inode != oip.newinode) {
300		pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
301				    dentry);
302	}
303
304	/* Force lookup of new upper hardlink to find its lower */
305	if (hardlink)
306		d_drop(dentry);
307
308	return 0;
309}
310
311static bool ovl_type_merge(struct dentry *dentry)
312{
313	return OVL_TYPE_MERGE(ovl_path_type(dentry));
314}
315
316static bool ovl_type_origin(struct dentry *dentry)
317{
318	return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
319}
320
321static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
322			    struct ovl_cattr *attr)
323{
324	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
325	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
326	struct inode *udir = upperdir->d_inode;
327	struct dentry *newdentry;
328	int err;
329
330	if (!attr->hardlink && !IS_POSIXACL(udir))
331		attr->mode &= ~current_umask();
332
333	inode_lock_nested(udir, I_MUTEX_PARENT);
334	newdentry = ovl_create_real(udir,
335				    lookup_one_len(dentry->d_name.name,
336						   upperdir,
337						   dentry->d_name.len),
338				    attr);
339	err = PTR_ERR(newdentry);
340	if (IS_ERR(newdentry))
341		goto out_unlock;
342
343	if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
344	    !ovl_allow_offline_changes(ofs)) {
345		/* Setting opaque here is just an optimization, allow to fail */
346		ovl_set_opaque(dentry, newdentry);
347	}
348
349	err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
350	if (err)
351		goto out_cleanup;
352out_unlock:
353	inode_unlock(udir);
354	return err;
355
356out_cleanup:
357	ovl_cleanup(udir, newdentry);
358	dput(newdentry);
359	goto out_unlock;
360}
361
362static struct dentry *ovl_clear_empty(struct dentry *dentry,
363				      struct list_head *list)
364{
365	struct dentry *workdir = ovl_workdir(dentry);
366	struct inode *wdir = workdir->d_inode;
367	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
368	struct inode *udir = upperdir->d_inode;
369	struct path upperpath;
370	struct dentry *upper;
371	struct dentry *opaquedir;
372	struct kstat stat;
373	int err;
374
375	if (WARN_ON(!workdir))
376		return ERR_PTR(-EROFS);
377
378	err = ovl_lock_rename_workdir(workdir, upperdir);
379	if (err)
380		goto out;
381
382	ovl_path_upper(dentry, &upperpath);
383	err = vfs_getattr(&upperpath, &stat,
384			  STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
385	if (err)
386		goto out_unlock;
387
388	err = -ESTALE;
389	if (!S_ISDIR(stat.mode))
390		goto out_unlock;
391	upper = upperpath.dentry;
392	if (upper->d_parent->d_inode != udir)
393		goto out_unlock;
394
395	opaquedir = ovl_create_temp(workdir, OVL_CATTR(stat.mode));
396	err = PTR_ERR(opaquedir);
397	if (IS_ERR(opaquedir))
398		goto out_unlock;
399
400	err = ovl_copy_xattr(dentry->d_sb, upper, opaquedir);
401	if (err)
402		goto out_cleanup;
403
404	err = ovl_set_opaque(dentry, opaquedir);
405	if (err)
406		goto out_cleanup;
407
408	inode_lock(opaquedir->d_inode);
409	err = ovl_set_attr(opaquedir, &stat);
410	inode_unlock(opaquedir->d_inode);
411	if (err)
412		goto out_cleanup;
413
414	err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
415	if (err)
416		goto out_cleanup;
417
418	ovl_cleanup_whiteouts(upper, list);
419	ovl_cleanup(wdir, upper);
420	unlock_rename(workdir, upperdir);
421
422	/* dentry's upper doesn't match now, get rid of it */
423	d_drop(dentry);
424
425	return opaquedir;
426
427out_cleanup:
428	ovl_cleanup(wdir, opaquedir);
429	dput(opaquedir);
430out_unlock:
431	unlock_rename(workdir, upperdir);
432out:
433	return ERR_PTR(err);
434}
435
436static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
437			     const struct posix_acl *acl)
438{
439	void *buffer;
440	size_t size;
441	int err;
442
443	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
444		return 0;
445
446	size = posix_acl_xattr_size(acl->a_count);
447	buffer = kmalloc(size, GFP_KERNEL);
448	if (!buffer)
449		return -ENOMEM;
450
451	err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
452	if (err < 0)
453		goto out_free;
454
455	err = vfs_setxattr(&init_user_ns, upperdentry, name, buffer, size, XATTR_CREATE);
456out_free:
457	kfree(buffer);
458	return err;
459}
460
461static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
462				    struct ovl_cattr *cattr)
463{
464	struct dentry *workdir = ovl_workdir(dentry);
465	struct inode *wdir = workdir->d_inode;
466	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
467	struct inode *udir = upperdir->d_inode;
468	struct dentry *upper;
469	struct dentry *newdentry;
470	int err;
471	struct posix_acl *acl, *default_acl;
472	bool hardlink = !!cattr->hardlink;
473
474	if (WARN_ON(!workdir))
475		return -EROFS;
476
477	if (!hardlink) {
478		err = posix_acl_create(dentry->d_parent->d_inode,
479				       &cattr->mode, &default_acl, &acl);
480		if (err)
481			return err;
482	}
483
484	err = ovl_lock_rename_workdir(workdir, upperdir);
485	if (err)
486		goto out;
487
488	upper = lookup_one_len(dentry->d_name.name, upperdir,
489			       dentry->d_name.len);
490	err = PTR_ERR(upper);
491	if (IS_ERR(upper))
492		goto out_unlock;
493
494	err = -ESTALE;
495	if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
496		goto out_dput;
497
498	newdentry = ovl_create_temp(workdir, cattr);
499	err = PTR_ERR(newdentry);
500	if (IS_ERR(newdentry))
501		goto out_dput;
502
503	/*
504	 * mode could have been mutilated due to umask (e.g. sgid directory)
505	 */
506	if (!hardlink &&
507	    !S_ISLNK(cattr->mode) &&
508	    newdentry->d_inode->i_mode != cattr->mode) {
509		struct iattr attr = {
510			.ia_valid = ATTR_MODE,
511			.ia_mode = cattr->mode,
512		};
513		inode_lock(newdentry->d_inode);
514		err = notify_change(&init_user_ns, newdentry, &attr, NULL);
515		inode_unlock(newdentry->d_inode);
516		if (err)
517			goto out_cleanup;
518	}
519	if (!hardlink) {
520		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
521					acl);
522		if (err)
523			goto out_cleanup;
524
525		err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
526					default_acl);
527		if (err)
528			goto out_cleanup;
529	}
530
531	if (!hardlink && S_ISDIR(cattr->mode)) {
532		err = ovl_set_opaque(dentry, newdentry);
533		if (err)
534			goto out_cleanup;
535
536		err = ovl_do_rename(wdir, newdentry, udir, upper,
537				    RENAME_EXCHANGE);
538		if (err)
539			goto out_cleanup;
540
541		ovl_cleanup(wdir, upper);
542	} else {
543		err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
544		if (err)
545			goto out_cleanup;
546	}
547	err = ovl_instantiate(dentry, inode, newdentry, hardlink);
548	if (err) {
549		ovl_cleanup(udir, newdentry);
550		dput(newdentry);
551	}
552out_dput:
553	dput(upper);
554out_unlock:
555	unlock_rename(workdir, upperdir);
556out:
557	if (!hardlink) {
558		posix_acl_release(acl);
559		posix_acl_release(default_acl);
560	}
561	return err;
562
563out_cleanup:
564	ovl_cleanup(wdir, newdentry);
565	dput(newdentry);
566	goto out_dput;
567}
568
569static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
570			      struct ovl_cattr *attr, bool origin)
571{
572	int err;
573	const struct cred *old_cred;
574	struct cred *override_cred;
575	struct dentry *parent = dentry->d_parent;
576
577	err = ovl_copy_up(parent);
578	if (err)
579		return err;
580
581	old_cred = ovl_override_creds(dentry->d_sb);
582
583	/*
584	 * When linking a file with copy up origin into a new parent, mark the
585	 * new parent dir "impure".
586	 */
587	if (origin) {
588		err = ovl_set_impure(parent, ovl_dentry_upper(parent));
589		if (err)
590			goto out_revert_creds;
591	}
592
593	err = -ENOMEM;
594	override_cred = prepare_creds();
595	if (override_cred) {
596		override_cred->fsuid = inode->i_uid;
597		override_cred->fsgid = inode->i_gid;
598		if (!attr->hardlink) {
599			err = security_dentry_create_files_as(dentry,
600					attr->mode, &dentry->d_name, old_cred,
601					override_cred);
602			if (err) {
603				put_cred(override_cred);
604				goto out_revert_creds;
605			}
606		}
607		put_cred(override_creds(override_cred));
608		put_cred(override_cred);
609
610		if (!ovl_dentry_is_whiteout(dentry))
611			err = ovl_create_upper(dentry, inode, attr);
612		else
613			err = ovl_create_over_whiteout(dentry, inode, attr);
614	}
615out_revert_creds:
616	revert_creds(old_cred);
617	return err;
618}
619
620static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
621			     const char *link)
622{
623	int err;
624	struct inode *inode;
625	struct ovl_cattr attr = {
626		.rdev = rdev,
627		.link = link,
628	};
629
630	err = ovl_want_write(dentry);
631	if (err)
632		goto out;
633
634	/* Preallocate inode to be used by ovl_get_inode() */
635	err = -ENOMEM;
636	inode = ovl_new_inode(dentry->d_sb, mode, rdev);
637	if (!inode)
638		goto out_drop_write;
639
640	spin_lock(&inode->i_lock);
641	inode->i_state |= I_CREATING;
642	spin_unlock(&inode->i_lock);
643
644	inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
645	attr.mode = inode->i_mode;
646
647	err = ovl_create_or_link(dentry, inode, &attr, false);
648	/* Did we end up using the preallocated inode? */
649	if (inode != d_inode(dentry))
650		iput(inode);
651
652out_drop_write:
653	ovl_drop_write(dentry);
654out:
655	return err;
656}
657
658static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
659		      struct dentry *dentry, umode_t mode, bool excl)
660{
661	return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
662}
663
664static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
665		     struct dentry *dentry, umode_t mode)
666{
667	return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
668}
669
670static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
671		     struct dentry *dentry, umode_t mode, dev_t rdev)
672{
673	/* Don't allow creation of "whiteout" on overlay */
674	if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
675		return -EPERM;
676
677	return ovl_create_object(dentry, mode, rdev, NULL);
678}
679
680static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
681		       struct dentry *dentry, const char *link)
682{
683	return ovl_create_object(dentry, S_IFLNK, 0, link);
684}
685
686static int ovl_set_link_redirect(struct dentry *dentry)
687{
688	const struct cred *old_cred;
689	int err;
690
691	old_cred = ovl_override_creds(dentry->d_sb);
692	err = ovl_set_redirect(dentry, false);
693	revert_creds(old_cred);
694
695	return err;
696}
697
698static int ovl_link(struct dentry *old, struct inode *newdir,
699		    struct dentry *new)
700{
701	int err;
702	struct inode *inode;
703
704	err = ovl_want_write(old);
705	if (err)
706		goto out;
707
708	err = ovl_copy_up(old);
709	if (err)
710		goto out_drop_write;
711
712	err = ovl_copy_up(new->d_parent);
713	if (err)
714		goto out_drop_write;
715
716	if (ovl_is_metacopy_dentry(old)) {
717		err = ovl_set_link_redirect(old);
718		if (err)
719			goto out_drop_write;
720	}
721
722	err = ovl_nlink_start(old);
723	if (err)
724		goto out_drop_write;
725
726	inode = d_inode(old);
727	ihold(inode);
728
729	err = ovl_create_or_link(new, inode,
730			&(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
731			ovl_type_origin(old));
732	if (err)
733		iput(inode);
734
735	ovl_nlink_end(old);
736out_drop_write:
737	ovl_drop_write(old);
738out:
739	return err;
740}
741
742static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
743{
744	return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
745}
746
747static int ovl_remove_and_whiteout(struct dentry *dentry,
748				   struct list_head *list)
749{
750	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
751	struct dentry *workdir = ovl_workdir(dentry);
752	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
753	struct dentry *upper;
754	struct dentry *opaquedir = NULL;
755	int err;
756
757	if (WARN_ON(!workdir))
758		return -EROFS;
759
760	if (!list_empty(list)) {
761		opaquedir = ovl_clear_empty(dentry, list);
762		err = PTR_ERR(opaquedir);
763		if (IS_ERR(opaquedir))
764			goto out;
765	}
766
767	err = ovl_lock_rename_workdir(workdir, upperdir);
768	if (err)
769		goto out_dput;
770
771	upper = lookup_one_len(dentry->d_name.name, upperdir,
772			       dentry->d_name.len);
773	err = PTR_ERR(upper);
774	if (IS_ERR(upper))
775		goto out_unlock;
776
777	err = -ESTALE;
778	if ((opaquedir && upper != opaquedir) ||
779	    (!opaquedir && ovl_dentry_upper(dentry) &&
780	     !ovl_matches_upper(dentry, upper))) {
781		goto out_dput_upper;
782	}
783
784	err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
785	if (err)
786		goto out_d_drop;
787
788	ovl_dir_modified(dentry->d_parent, true);
789out_d_drop:
790	d_drop(dentry);
791out_dput_upper:
792	dput(upper);
793out_unlock:
794	unlock_rename(workdir, upperdir);
795out_dput:
796	dput(opaquedir);
797out:
798	return err;
799}
800
801static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
802			    struct list_head *list)
803{
804	struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
805	struct inode *dir = upperdir->d_inode;
806	struct dentry *upper;
807	struct dentry *opaquedir = NULL;
808	int err;
809
810	if (!list_empty(list)) {
811		opaquedir = ovl_clear_empty(dentry, list);
812		err = PTR_ERR(opaquedir);
813		if (IS_ERR(opaquedir))
814			goto out;
815	}
816
817	inode_lock_nested(dir, I_MUTEX_PARENT);
818	upper = lookup_one_len(dentry->d_name.name, upperdir,
819			       dentry->d_name.len);
820	err = PTR_ERR(upper);
821	if (IS_ERR(upper))
822		goto out_unlock;
823
824	err = -ESTALE;
825	if ((opaquedir && upper != opaquedir) ||
826	    (!opaquedir && !ovl_matches_upper(dentry, upper)))
827		goto out_dput_upper;
828
829	if (is_dir)
830		err = vfs_rmdir(&init_user_ns, dir, upper);
831	else
832		err = vfs_unlink(&init_user_ns, dir, upper, NULL);
833	ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
834
835	/*
836	 * Keeping this dentry hashed would mean having to release
837	 * upperpath/lowerpath, which could only be done if we are the
838	 * sole user of this dentry.  Too tricky...  Just unhash for
839	 * now.
840	 */
841	if (!err)
842		d_drop(dentry);
843out_dput_upper:
844	dput(upper);
845out_unlock:
846	inode_unlock(dir);
847	dput(opaquedir);
848out:
849	return err;
850}
851
852static bool ovl_pure_upper(struct dentry *dentry)
853{
854	return !ovl_dentry_lower(dentry) &&
855	       !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
856}
857
858static void ovl_drop_nlink(struct dentry *dentry)
859{
860	struct inode *inode = d_inode(dentry);
861	struct dentry *alias;
862
863	/* Try to find another, hashed alias */
864	spin_lock(&inode->i_lock);
865	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
866		if (alias != dentry && !d_unhashed(alias))
867			break;
868	}
869	spin_unlock(&inode->i_lock);
870
871	/*
872	 * Changes to underlying layers may cause i_nlink to lose sync with
873	 * reality.  In this case prevent the link count from going to zero
874	 * prematurely.
875	 */
876	if (inode->i_nlink > !!alias)
877		drop_nlink(inode);
878}
879
880static int ovl_do_remove(struct dentry *dentry, bool is_dir)
881{
882	int err;
883	const struct cred *old_cred;
884	struct dentry *upperdentry;
885	bool lower_positive = ovl_lower_positive(dentry);
886	LIST_HEAD(list);
887
888	/* No need to clean pure upper removed by vfs_rmdir() */
889	if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
890		err = ovl_check_empty_dir(dentry, &list);
891		if (err)
892			goto out;
893	}
894
895	err = ovl_want_write(dentry);
896	if (err)
897		goto out;
898
899	err = ovl_copy_up(dentry->d_parent);
900	if (err)
901		goto out_drop_write;
902
903	err = ovl_nlink_start(dentry);
904	if (err)
905		goto out_drop_write;
906
907	old_cred = ovl_override_creds(dentry->d_sb);
908	if (!lower_positive)
909		err = ovl_remove_upper(dentry, is_dir, &list);
910	else
911		err = ovl_remove_and_whiteout(dentry, &list);
912	revert_creds(old_cred);
913	if (!err) {
914		if (is_dir)
915			clear_nlink(dentry->d_inode);
916		else
917			ovl_drop_nlink(dentry);
918	}
919	ovl_nlink_end(dentry);
920
921	/*
922	 * Copy ctime
923	 *
924	 * Note: we fail to update ctime if there was no copy-up, only a
925	 * whiteout
926	 */
927	upperdentry = ovl_dentry_upper(dentry);
928	if (upperdentry)
929		ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
930
931out_drop_write:
932	ovl_drop_write(dentry);
933out:
934	ovl_cache_free(&list);
935	return err;
936}
937
938static int ovl_unlink(struct inode *dir, struct dentry *dentry)
939{
940	return ovl_do_remove(dentry, false);
941}
942
943static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
944{
945	return ovl_do_remove(dentry, true);
946}
947
948static bool ovl_type_merge_or_lower(struct dentry *dentry)
949{
950	enum ovl_path_type type = ovl_path_type(dentry);
951
952	return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
953}
954
955static bool ovl_can_move(struct dentry *dentry)
956{
957	return ovl_redirect_dir(dentry->d_sb) ||
958		!d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
959}
960
961static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
962{
963	char *buf, *ret;
964	struct dentry *d, *tmp;
965	int buflen = ovl_redirect_max + 1;
966
967	if (!abs_redirect) {
968		ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
969			       GFP_KERNEL);
970		goto out;
971	}
972
973	buf = ret = kmalloc(buflen, GFP_KERNEL);
974	if (!buf)
975		goto out;
976
977	buflen--;
978	buf[buflen] = '\0';
979	for (d = dget(dentry); !IS_ROOT(d);) {
980		const char *name;
981		int thislen;
982
983		spin_lock(&d->d_lock);
984		name = ovl_dentry_get_redirect(d);
985		if (name) {
986			thislen = strlen(name);
987		} else {
988			name = d->d_name.name;
989			thislen = d->d_name.len;
990		}
991
992		/* If path is too long, fall back to userspace move */
993		if (thislen + (name[0] != '/') > buflen) {
994			ret = ERR_PTR(-EXDEV);
995			spin_unlock(&d->d_lock);
996			goto out_put;
997		}
998
999		buflen -= thislen;
1000		memcpy(&buf[buflen], name, thislen);
1001		spin_unlock(&d->d_lock);
1002		tmp = dget_parent(d);
1003
1004		dput(d);
1005		d = tmp;
1006
1007		/* Absolute redirect: finished */
1008		if (buf[buflen] == '/')
1009			break;
1010		buflen--;
1011		buf[buflen] = '/';
1012	}
1013	ret = kstrdup(&buf[buflen], GFP_KERNEL);
1014out_put:
1015	dput(d);
1016	kfree(buf);
1017out:
1018	return ret ? ret : ERR_PTR(-ENOMEM);
1019}
1020
1021static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1022{
1023	struct dentry *lowerdentry;
1024
1025	if (!samedir)
1026		return true;
1027
1028	if (d_is_dir(dentry))
1029		return false;
1030
1031	/*
1032	 * For non-dir hardlinked files, we need absolute redirects
1033	 * in general as two upper hardlinks could be in different
1034	 * dirs. We could put a relative redirect now and convert
1035	 * it to absolute redirect later. But when nlink > 1 and
1036	 * indexing is on, that means relative redirect needs to be
1037	 * converted to absolute during copy up of another lower
1038	 * hardllink as well.
1039	 *
1040	 * So without optimizing too much, just check if lower is
1041	 * a hard link or not. If lower is hard link, put absolute
1042	 * redirect.
1043	 */
1044	lowerdentry = ovl_dentry_lower(dentry);
1045	return (d_inode(lowerdentry)->i_nlink > 1);
1046}
1047
1048static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1049{
1050	int err;
1051	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1052	const char *redirect = ovl_dentry_get_redirect(dentry);
1053	bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1054
1055	if (redirect && (!absolute_redirect || redirect[0] == '/'))
1056		return 0;
1057
1058	redirect = ovl_get_redirect(dentry, absolute_redirect);
1059	if (IS_ERR(redirect))
1060		return PTR_ERR(redirect);
1061
1062	err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1063				 OVL_XATTR_REDIRECT,
1064				 redirect, strlen(redirect), -EXDEV);
1065	if (!err) {
1066		spin_lock(&dentry->d_lock);
1067		ovl_dentry_set_redirect(dentry, redirect);
1068		spin_unlock(&dentry->d_lock);
1069	} else {
1070		kfree(redirect);
1071		pr_warn_ratelimited("failed to set redirect (%i)\n",
1072				    err);
1073		/* Fall back to userspace copy-up */
1074		err = -EXDEV;
1075	}
1076	return err;
1077}
1078
1079static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1080		      struct dentry *old, struct inode *newdir,
1081		      struct dentry *new, unsigned int flags)
1082{
1083	int err;
1084	struct dentry *old_upperdir;
1085	struct dentry *new_upperdir;
1086	struct dentry *olddentry;
1087	struct dentry *newdentry;
1088	struct dentry *trap;
1089	bool old_opaque;
1090	bool new_opaque;
1091	bool cleanup_whiteout = false;
1092	bool update_nlink = false;
1093	bool overwrite = !(flags & RENAME_EXCHANGE);
1094	bool is_dir = d_is_dir(old);
1095	bool new_is_dir = d_is_dir(new);
1096	bool samedir = olddir == newdir;
1097	struct dentry *opaquedir = NULL;
1098	const struct cred *old_cred = NULL;
1099	LIST_HEAD(list);
1100
1101	err = -EINVAL;
1102	if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1103		goto out;
1104
1105	flags &= ~RENAME_NOREPLACE;
1106
1107	/* Don't copy up directory trees */
1108	err = -EXDEV;
1109	if (!ovl_can_move(old))
1110		goto out;
1111	if (!overwrite && !ovl_can_move(new))
1112		goto out;
1113
1114	if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1115		err = ovl_check_empty_dir(new, &list);
1116		if (err)
1117			goto out;
1118	}
1119
1120	if (overwrite) {
1121		if (ovl_lower_positive(old)) {
1122			if (!ovl_dentry_is_whiteout(new)) {
1123				/* Whiteout source */
1124				flags |= RENAME_WHITEOUT;
1125			} else {
1126				/* Switch whiteouts */
1127				flags |= RENAME_EXCHANGE;
1128			}
1129		} else if (is_dir && ovl_dentry_is_whiteout(new)) {
1130			flags |= RENAME_EXCHANGE;
1131			cleanup_whiteout = true;
1132		}
1133	}
1134
1135	err = ovl_want_write(old);
1136	if (err)
1137		goto out;
1138
1139	err = ovl_copy_up(old);
1140	if (err)
1141		goto out_drop_write;
1142
1143	err = ovl_copy_up(new->d_parent);
1144	if (err)
1145		goto out_drop_write;
1146	if (!overwrite) {
1147		err = ovl_copy_up(new);
1148		if (err)
1149			goto out_drop_write;
1150	} else if (d_inode(new)) {
1151		err = ovl_nlink_start(new);
1152		if (err)
1153			goto out_drop_write;
1154
1155		update_nlink = true;
1156	}
1157
1158	old_cred = ovl_override_creds(old->d_sb);
1159
1160	if (!list_empty(&list)) {
1161		opaquedir = ovl_clear_empty(new, &list);
1162		err = PTR_ERR(opaquedir);
1163		if (IS_ERR(opaquedir)) {
1164			opaquedir = NULL;
1165			goto out_revert_creds;
1166		}
1167	}
1168
1169	old_upperdir = ovl_dentry_upper(old->d_parent);
1170	new_upperdir = ovl_dentry_upper(new->d_parent);
1171
1172	if (!samedir) {
1173		/*
1174		 * When moving a merge dir or non-dir with copy up origin into
1175		 * a new parent, we are marking the new parent dir "impure".
1176		 * When ovl_iterate() iterates an "impure" upper dir, it will
1177		 * lookup the origin inodes of the entries to fill d_ino.
1178		 */
1179		if (ovl_type_origin(old)) {
1180			err = ovl_set_impure(new->d_parent, new_upperdir);
1181			if (err)
1182				goto out_revert_creds;
1183		}
1184		if (!overwrite && ovl_type_origin(new)) {
1185			err = ovl_set_impure(old->d_parent, old_upperdir);
1186			if (err)
1187				goto out_revert_creds;
1188		}
1189	}
1190
1191	trap = lock_rename(new_upperdir, old_upperdir);
1192
1193	olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1194				   old->d_name.len);
1195	err = PTR_ERR(olddentry);
1196	if (IS_ERR(olddentry))
1197		goto out_unlock;
1198
1199	err = -ESTALE;
1200	if (!ovl_matches_upper(old, olddentry))
1201		goto out_dput_old;
1202
1203	newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1204				   new->d_name.len);
1205	err = PTR_ERR(newdentry);
1206	if (IS_ERR(newdentry))
1207		goto out_dput_old;
1208
1209	old_opaque = ovl_dentry_is_opaque(old);
1210	new_opaque = ovl_dentry_is_opaque(new);
1211
1212	err = -ESTALE;
1213	if (d_inode(new) && ovl_dentry_upper(new)) {
1214		if (opaquedir) {
1215			if (newdentry != opaquedir)
1216				goto out_dput;
1217		} else {
1218			if (!ovl_matches_upper(new, newdentry))
1219				goto out_dput;
1220		}
1221	} else {
1222		if (!d_is_negative(newdentry) &&
1223		    (!new_opaque || !ovl_is_whiteout(newdentry)))
1224			goto out_dput;
1225	}
1226
1227	if (olddentry == trap)
1228		goto out_dput;
1229	if (newdentry == trap)
1230		goto out_dput;
1231
1232	if (olddentry->d_inode == newdentry->d_inode)
1233		goto out_dput;
1234
1235	err = 0;
1236	if (ovl_type_merge_or_lower(old))
1237		err = ovl_set_redirect(old, samedir);
1238	else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1239		err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1240	if (err)
1241		goto out_dput;
1242
1243	if (!overwrite && ovl_type_merge_or_lower(new))
1244		err = ovl_set_redirect(new, samedir);
1245	else if (!overwrite && new_is_dir && !new_opaque &&
1246		 ovl_type_merge(old->d_parent))
1247		err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1248	if (err)
1249		goto out_dput;
1250
1251	err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1252			    new_upperdir->d_inode, newdentry, flags);
1253	if (err)
1254		goto out_dput;
1255
1256	if (cleanup_whiteout)
1257		ovl_cleanup(old_upperdir->d_inode, newdentry);
1258
1259	if (overwrite && d_inode(new)) {
1260		if (new_is_dir)
1261			clear_nlink(d_inode(new));
1262		else
1263			ovl_drop_nlink(new);
1264	}
1265
1266	ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1267			 (!overwrite && ovl_type_origin(new)));
1268	ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1269			 (d_inode(new) && ovl_type_origin(new)));
1270
1271	/* copy ctime: */
1272	ovl_copyattr(d_inode(olddentry), d_inode(old));
1273	if (d_inode(new) && ovl_dentry_upper(new))
1274		ovl_copyattr(d_inode(newdentry), d_inode(new));
1275
1276out_dput:
1277	dput(newdentry);
1278out_dput_old:
1279	dput(olddentry);
1280out_unlock:
1281	unlock_rename(new_upperdir, old_upperdir);
1282out_revert_creds:
1283	revert_creds(old_cred);
1284	if (update_nlink)
1285		ovl_nlink_end(new);
1286out_drop_write:
1287	ovl_drop_write(old);
1288out:
1289	dput(opaquedir);
1290	ovl_cache_free(&list);
1291	return err;
1292}
1293
1294const struct inode_operations ovl_dir_inode_operations = {
1295	.lookup		= ovl_lookup,
1296	.mkdir		= ovl_mkdir,
1297	.symlink	= ovl_symlink,
1298	.unlink		= ovl_unlink,
1299	.rmdir		= ovl_rmdir,
1300	.rename		= ovl_rename,
1301	.link		= ovl_link,
1302	.setattr	= ovl_setattr,
1303	.create		= ovl_create,
1304	.mknod		= ovl_mknod,
1305	.permission	= ovl_permission,
1306	.getattr	= ovl_getattr,
1307	.listxattr	= ovl_listxattr,
1308	.get_acl	= ovl_get_acl,
1309	.update_time	= ovl_update_time,
1310	.fileattr_get	= ovl_fileattr_get,
1311	.fileattr_set	= ovl_fileattr_set,
1312};
1313