1/*	$NetBSD: ext2fs_rename.c,v 1.13 2023/08/26 05:22:50 riastradh Exp $	*/
2
3/*-
4 * Copyright (c) 2012 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Taylor R Campbell.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Ext2fs Rename
34 */
35
36#include <sys/cdefs.h>
37__KERNEL_RCSID(0, "$NetBSD: ext2fs_rename.c,v 1.13 2023/08/26 05:22:50 riastradh Exp $");
38
39#include <sys/param.h>
40#include <sys/buf.h>
41#include <sys/errno.h>
42#include <sys/kauth.h>
43#include <sys/mount.h>
44#include <sys/namei.h>
45#include <sys/vnode.h>
46#include <sys/vnode_if.h>
47#include <sys/dirent.h>
48
49#include <miscfs/genfs/genfs.h>
50
51#include <ufs/ext2fs/ext2fs.h>
52#include <ufs/ext2fs/ext2fs_dir.h>
53#include <ufs/ext2fs/ext2fs_extern.h>
54#include <ufs/ufs/inode.h>
55#include <ufs/ufs/ufs_extern.h>
56#include <ufs/ufs/ufsmount.h>
57
58/*
59 * Forward declarations
60 */
61static int ext2fs_sane_rename(struct vnode *, struct componentname *,
62    struct vnode *, struct componentname *,
63    kauth_cred_t, bool);
64static bool ext2fs_rename_ulr_overlap_p(const struct ufs_lookup_results *,
65    const struct ufs_lookup_results *);
66static int ext2fs_rename_recalculate_fulr(struct vnode *,
67    struct ufs_lookup_results *, const struct ufs_lookup_results *,
68    const struct componentname *);
69static bool ext2fs_rmdired_p(struct vnode *);
70static int ext2fs_read_dotdot(struct vnode *, kauth_cred_t, ino_t *);
71static int ext2fs_rename_replace_dotdot(struct vnode *,
72    struct vnode *, struct vnode *, kauth_cred_t);
73static int ext2fs_gro_lock_directory(struct mount *, struct vnode *);
74
75static const struct genfs_rename_ops ext2fs_genfs_rename_ops;
76
77/*
78 * ext2fs_sane_rename: The hairiest vop, with the saner API.
79 *
80 * Arguments:
81 *
82 * . fdvp (from directory vnode),
83 * . fcnp (from component name),
84 * . tdvp (to directory vnode),
85 * . tcnp (to component name),
86 * . cred (credentials structure), and
87 * . posixly_correct (flag for behaviour if target & source link same file).
88 *
89 * fdvp and tdvp may be the same, and must be referenced and unlocked.
90 */
91static int
92ext2fs_sane_rename(
93    struct vnode *fdvp, struct componentname *fcnp,
94    struct vnode *tdvp, struct componentname *tcnp,
95    kauth_cred_t cred, bool posixly_correct)
96{
97	struct ufs_lookup_results fulr, tulr;
98
99	return genfs_sane_rename(&ext2fs_genfs_rename_ops,
100	    fdvp, fcnp, &fulr, tdvp, tcnp, &tulr,
101	    cred, posixly_correct);
102}
103
104/*
105 * ext2fs_rename: The hairiest vop, with the insanest API.  Defer to
106 * genfs_insane_rename immediately.
107 */
108int
109ext2fs_rename(void *v)
110{
111
112	return genfs_insane_rename(v, &ext2fs_sane_rename);
113}
114
115/*
116 * ext2fs_gro_directory_empty_p: Return true if the directory vp is
117 * empty.  dvp is its parent.
118 *
119 * vp and dvp must be locked and referenced.
120 */
121static bool
122ext2fs_gro_directory_empty_p(struct mount *mp, kauth_cred_t cred,
123    struct vnode *vp, struct vnode *dvp)
124{
125
126	(void)mp;
127	KASSERT(mp != NULL);
128	KASSERT(vp != NULL);
129	KASSERT(dvp != NULL);
130	KASSERT(vp != dvp);
131	KASSERT(vp->v_mount == mp);
132	KASSERT(dvp->v_mount == mp);
133	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
134	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
135
136	return ext2fs_dirempty(VTOI(vp), VTOI(dvp)->i_number, cred);
137}
138
139/*
140 * ext2fs_gro_rename_check_possible: Check whether a rename is possible
141 * independent of credentials.
142 */
143static int
144ext2fs_gro_rename_check_possible(struct mount *mp,
145    struct vnode *fdvp, struct vnode *fvp,
146    struct vnode *tdvp, struct vnode *tvp)
147{
148
149	(void)mp;
150	KASSERT(mp != NULL);
151	KASSERT(fdvp != NULL);
152	KASSERT(fvp != NULL);
153	KASSERT(tdvp != NULL);
154	KASSERT(fdvp != fvp);
155	KASSERT(fdvp != tvp);
156	KASSERT(tdvp != fvp);
157	KASSERT(tdvp != tvp);
158	KASSERT(fvp != tvp);
159	KASSERT(fdvp->v_type == VDIR);
160	KASSERT(tdvp->v_type == VDIR);
161	KASSERT(fdvp->v_mount == mp);
162	KASSERT(fvp->v_mount == mp);
163	KASSERT(tdvp->v_mount == mp);
164	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
165	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
166	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
167	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
168	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
169
170	return genfs_ufslike_rename_check_possible(
171	    VTOI(fdvp)->i_e2fs_flags, VTOI(fvp)->i_e2fs_flags,
172	    VTOI(tdvp)->i_e2fs_flags, (tvp? VTOI(tvp)->i_e2fs_flags : 0),
173	    (tvp != NULL),
174	    EXT2_IMMUTABLE, EXT2_APPEND);
175}
176
177/*
178 * ext2fs_gro_rename_check_permitted: Check whether a rename is
179 * permitted given our credentials.
180 */
181static int
182ext2fs_gro_rename_check_permitted(struct mount *mp, kauth_cred_t cred,
183    struct vnode *fdvp, struct vnode *fvp,
184    struct vnode *tdvp, struct vnode *tvp)
185{
186
187	(void)mp;
188	KASSERT(mp != NULL);
189	KASSERT(fdvp != NULL);
190	KASSERT(fvp != NULL);
191	KASSERT(tdvp != NULL);
192	KASSERT(fdvp != fvp);
193	KASSERT(fdvp != tvp);
194	KASSERT(tdvp != fvp);
195	KASSERT(tdvp != tvp);
196	KASSERT(fvp != tvp);
197	KASSERT(fdvp->v_type == VDIR);
198	KASSERT(tdvp->v_type == VDIR);
199	KASSERT(fdvp->v_mount == mp);
200	KASSERT(fvp->v_mount == mp);
201	KASSERT(tdvp->v_mount == mp);
202	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
203	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
204	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
205	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
206	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
207
208	return genfs_ufslike_rename_check_permitted(cred,
209	    fdvp, VTOI(fdvp)->i_e2fs_mode, VTOI(fdvp)->i_uid,
210	    fvp, VTOI(fvp)->i_uid,
211	    tdvp, VTOI(tdvp)->i_e2fs_mode, VTOI(tdvp)->i_uid,
212	    tvp, (tvp? VTOI(tvp)->i_uid : 0));
213}
214
215/*
216 * ext2fs_gro_remove_check_possible: Check whether a remove is possible
217 * independent of credentials.
218 */
219static int
220ext2fs_gro_remove_check_possible(struct mount *mp,
221    struct vnode *dvp, struct vnode *vp)
222{
223
224	(void)mp;
225	KASSERT(mp != NULL);
226	KASSERT(dvp != NULL);
227	KASSERT(vp != NULL);
228	KASSERT(dvp != vp);
229	KASSERT(dvp->v_type == VDIR);
230	KASSERT(vp->v_type != VDIR);
231	KASSERT(dvp->v_mount == mp);
232	KASSERT(vp->v_mount == mp);
233	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
234	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
235
236	return genfs_ufslike_remove_check_possible(
237	    VTOI(dvp)->i_e2fs_flags, VTOI(vp)->i_e2fs_flags,
238	    EXT2_IMMUTABLE, EXT2_APPEND);
239}
240
241/*
242 * ext2fs_gro_remove_check_permitted: Check whether a remove is
243 * permitted given our credentials.
244 */
245static int
246ext2fs_gro_remove_check_permitted(struct mount *mp, kauth_cred_t cred,
247    struct vnode *dvp, struct vnode *vp)
248{
249
250	(void)mp;
251	KASSERT(mp != NULL);
252	KASSERT(dvp != NULL);
253	KASSERT(vp != NULL);
254	KASSERT(dvp != vp);
255	KASSERT(dvp->v_type == VDIR);
256	KASSERT(vp->v_type != VDIR);
257	KASSERT(dvp->v_mount == mp);
258	KASSERT(vp->v_mount == mp);
259	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
260	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
261
262	return genfs_ufslike_remove_check_permitted(cred,
263	    dvp, VTOI(dvp)->i_e2fs_mode, VTOI(dvp)->i_uid,
264	    vp, VTOI(vp)->i_uid);
265}
266
267/*
268 * ext2fs_gro_rename: Actually perform the rename operation.
269 */
270static int
271ext2fs_gro_rename(struct mount *mp, kauth_cred_t cred,
272    struct vnode *fdvp, struct componentname *fcnp,
273    void *fde, struct vnode *fvp,
274    struct vnode *tdvp, struct componentname *tcnp,
275    void *tde, struct vnode *tvp, nlink_t *tvp_nlinkp)
276{
277	struct ufs_lookup_results *fulr = fde;
278	struct ufs_lookup_results *tulr = tde;
279	bool directory_p, reparent_p;
280	int error;
281
282	(void)mp;
283	KASSERT(mp != NULL);
284	KASSERT(fdvp != NULL);
285	KASSERT(fcnp != NULL);
286	KASSERT(fulr != NULL);
287	KASSERT(fvp != NULL);
288	KASSERT(tdvp != NULL);
289	KASSERT(tcnp != NULL);
290	KASSERT(tulr != NULL);
291	KASSERT(fulr != tulr);
292	KASSERT(fdvp != fvp);
293	KASSERT(fdvp != tvp);
294	KASSERT(tdvp != fvp);
295	KASSERT(tdvp != tvp);
296	KASSERT(fvp != tvp);
297	KASSERT(fdvp->v_mount == mp);
298	KASSERT(fvp->v_mount == mp);
299	KASSERT(tdvp->v_mount == mp);
300	KASSERT((tvp == NULL) || (tvp->v_mount == mp));
301	KASSERT(VOP_ISLOCKED(fdvp) == LK_EXCLUSIVE);
302	KASSERT(VOP_ISLOCKED(fvp) == LK_EXCLUSIVE);
303	KASSERT(VOP_ISLOCKED(tdvp) == LK_EXCLUSIVE);
304	KASSERT((tvp == NULL) || (VOP_ISLOCKED(tvp) == LK_EXCLUSIVE));
305
306	/*
307	 * We shall need to temporarily bump the link count, so make
308	 * sure there is room to do so.
309	 */
310	if ((nlink_t)VTOI(fvp)->i_e2fs_nlink >= EXT2FS_LINK_MAX)
311		return EMLINK;
312
313	directory_p = (fvp->v_type == VDIR);
314	KASSERT(directory_p == ((VTOI(fvp)->i_e2fs_mode & IFMT) == IFDIR));
315	KASSERT((tvp == NULL) || (directory_p == (tvp->v_type == VDIR)));
316	KASSERT((tvp == NULL) || (directory_p ==
317		((VTOI(tvp)->i_e2fs_mode & IFMT) == IFDIR)));
318
319	reparent_p = (fdvp != tdvp);
320	KASSERT(reparent_p == (VTOI(fdvp)->i_number != VTOI(tdvp)->i_number));
321
322	/*
323	 * Commence hacking of the data on disk.
324	 */
325
326	/*
327	 * 1) Bump link count while we're moving stuff
328	 *    around.  If we crash somewhere before
329	 *    completing our work, the link count
330	 *    may be wrong, but correctable.
331	 */
332
333	KASSERT((nlink_t)VTOI(fvp)->i_e2fs_nlink < EXT2FS_LINK_MAX);
334	VTOI(fvp)->i_e2fs_nlink++;
335	VTOI(fvp)->i_flag |= IN_CHANGE;
336	error = ext2fs_update(fvp, NULL, NULL, UPDATE_WAIT);
337	if (error)
338		goto whymustithurtsomuch;
339
340	/*
341	 * 2) If target doesn't exist, link the target
342	 *    to the source and unlink the source.
343	 *    Otherwise, rewrite the target directory
344	 *    entry to reference the source inode and
345	 *    expunge the original entry's existence.
346	 */
347
348	if (tvp == NULL) {
349		/*
350		 * Account for ".." in new directory.
351		 * When source and destination have the same
352		 * parent we don't fool with the link count.
353		 */
354		if (directory_p && reparent_p) {
355			if ((nlink_t)VTOI(tdvp)->i_e2fs_nlink >= EXT2FS_LINK_MAX) {
356				error = EMLINK;
357				goto whymustithurtsomuch;
358			}
359			KASSERT((nlink_t)VTOI(tdvp)->i_e2fs_nlink < EXT2FS_LINK_MAX);
360			VTOI(tdvp)->i_e2fs_nlink++;
361			VTOI(tdvp)->i_flag |= IN_CHANGE;
362			error = ext2fs_update(tdvp, NULL, NULL, UPDATE_WAIT);
363			if (error) {
364				/*
365				 * Link count update didn't take --
366				 * back out the in-memory link count.
367				 */
368				KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
369				VTOI(tdvp)->i_e2fs_nlink--;
370				VTOI(tdvp)->i_flag |= IN_CHANGE;
371				goto whymustithurtsomuch;
372			}
373		}
374
375		error = ext2fs_direnter(VTOI(fvp), tdvp, tulr, tcnp);
376		if (error) {
377			if (directory_p && reparent_p) {
378				/*
379				 * Directory update didn't take, but
380				 * the link count update did -- back
381				 * out the in-memory link count and the
382				 * on-disk link count.
383				 */
384				KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
385				VTOI(tdvp)->i_e2fs_nlink--;
386				VTOI(tdvp)->i_flag |= IN_CHANGE;
387				(void)ext2fs_update(tdvp, NULL, NULL,
388				    UPDATE_WAIT);
389			}
390			goto whymustithurtsomuch;
391		}
392	} else {
393		if (directory_p)
394			/* XXX WTF?  Why purge here?  Why not purge others?  */
395			cache_purge(tdvp);
396
397		/*
398		 * Make the target directory's entry for tcnp point at
399		 * the source node.
400		 */
401		error = ext2fs_dirrewrite(VTOI(tdvp), tulr, VTOI(fvp), tcnp);
402		if (error)
403			goto whymustithurtsomuch;
404
405		/*
406		 * If the source and target are directories, and the
407		 * target is in the same directory as the source,
408		 * decrement the link count of the common parent
409		 * directory, since we are removing the target from
410		 * that directory.
411		 */
412		if (directory_p && !reparent_p) {
413			KASSERT(fdvp == tdvp);
414			/* XXX check, don't kassert */
415			KASSERT(0 < VTOI(tdvp)->i_e2fs_nlink);
416			VTOI(tdvp)->i_e2fs_nlink--;
417			VTOI(tdvp)->i_flag |= IN_CHANGE;
418		}
419
420		/*
421		 * Adjust the link count of the target to
422		 * reflect the dirrewrite above.  If this is
423		 * a directory it is empty and there are
424		 * no links to it, so we can squash the inode and
425		 * any space associated with it.  We disallowed
426		 * renaming over top of a directory with links to
427		 * it above, as the remaining link would point to
428		 * a directory without "." or ".." entries.
429		 */
430		/* XXX check, don't kassert */
431		KASSERT(0 < VTOI(tvp)->i_e2fs_nlink);
432		VTOI(tvp)->i_e2fs_nlink--;
433		if (directory_p) {
434			/*
435			 * XXX The ext2fs_dirempty call earlier does
436			 * not guarantee anything about nlink.
437			 */
438			if (VTOI(tvp)->i_e2fs_nlink != 1)
439				ufs_dirbad(VTOI(tvp), (doff_t)0,
440				    "hard-linked directory");
441			VTOI(tvp)->i_e2fs_nlink = 0;
442			error = ext2fs_truncate(tvp, (off_t)0, IO_SYNC, cred);
443#if 0			/* XXX This branch was not in ext2fs_rename!  */
444			if (error)
445				goto whymustithurtsomuch;
446#endif
447		}
448		*tvp_nlinkp = VTOI(tvp)->i_e2fs_nlink;
449		/*
450		 * XXX Why is this here, and not above the preceding
451		 * conditional?
452		 */
453		VTOI(tvp)->i_flag |= IN_CHANGE;
454	}
455
456	/*
457	 * If the source is a directory with a new parent, the link
458	 * count of the old parent directory must be decremented and
459	 * ".." set to point to the new parent.
460	 */
461	if (directory_p && reparent_p) {
462		error = ext2fs_rename_replace_dotdot(fvp, fdvp, tdvp, cred);
463		if (error)
464			goto whymustithurtsomuch;
465
466		/* XXX WTF?  Why purge here?  Why not purge others?  */
467		cache_purge(fdvp);
468	}
469
470	/*
471	 * 3) Unlink the source.
472	 */
473
474	/*
475	 * ext2fs_direnter may compact the directory in the process of
476	 * inserting a new entry.  That may invalidate fulr, which we
477	 * need in order to remove the old entry.  In that case, we
478	 * need to recalculate what fulr should be.
479	 */
480	if (!reparent_p && (tvp == NULL) &&
481	    ext2fs_rename_ulr_overlap_p(fulr, tulr)) {
482		error = ext2fs_rename_recalculate_fulr(fdvp, fulr, tulr, fcnp);
483#if 0				/* XXX */
484		if (error)	/* XXX Try to back out changes?  */
485			goto whymustithurtsomuch;
486#endif
487	}
488
489	error = ext2fs_dirremove(fdvp, fulr, fcnp);
490	if (error)
491		goto whymustithurtsomuch;
492
493#if 0				/* XXX */
494	genfs_rename_cache_purge(fdvp, fvp, tdvp, tvp);
495#endif
496
497whymustithurtsomuch:
498	KASSERT(0 < VTOI(fvp)->i_e2fs_nlink);
499	VTOI(fvp)->i_e2fs_nlink--;
500	VTOI(fvp)->i_flag |= IN_CHANGE;
501	return error;
502}
503
504/*
505 * ext2fs_rename_ulr_overlap_p: True iff tulr overlaps with fulr so
506 * that entering a directory entry at tulr may move fulr.
507 */
508static bool
509ext2fs_rename_ulr_overlap_p(const struct ufs_lookup_results *fulr,
510    const struct ufs_lookup_results *tulr)
511{
512	doff_t from_prev_start, from_prev_end, to_start, to_end;
513
514	KASSERT(fulr != NULL);
515	KASSERT(tulr != NULL);
516	KASSERT(fulr != tulr);
517
518	/*
519	 * fulr is from a DELETE lookup, so fulr->ulr_count is the size
520	 * of the preceding entry (d_reclen).
521	 */
522	from_prev_end = fulr->ulr_offset;
523	KASSERT(fulr->ulr_count <= from_prev_end);
524	from_prev_start = (from_prev_end - fulr->ulr_count);
525
526	/*
527	 * tulr is from a RENAME lookup, so tulr->ulr_count is the size
528	 * of the free space for an entry that we are about to fill.
529	 */
530	to_start = tulr->ulr_offset;
531	KASSERT(tulr->ulr_count < (EXT2FS_MAXDIRSIZE - to_start));
532	to_end = (to_start + tulr->ulr_count);
533
534	return
535	    (((to_start <= from_prev_start) && (from_prev_start < to_end)) ||
536		((to_start <= from_prev_end) && (from_prev_end < to_end)));
537}
538
539/*
540 * ext2fs_rename_recalculate_fulr: If we have just entered a directory
541 * into dvp at tulr, and we were about to remove one at fulr for an
542 * entry named fcnp, fulr may be invalid.  So, if necessary,
543 * recalculate it.
544 */
545static int
546ext2fs_rename_recalculate_fulr(struct vnode *dvp,
547    struct ufs_lookup_results *fulr, const struct ufs_lookup_results *tulr,
548    const struct componentname *fcnp)
549{
550	struct mount *mp;
551	struct ufsmount *ump;
552	/* XXX int is a silly type for this; blame ufsmount::um_dirblksiz.  */
553	int dirblksiz;
554	doff_t search_start, search_end;
555	doff_t offset;		/* Offset of entry we're examining.  */
556	struct buf *bp;		/* I/O block we're examining.  */
557	char *dirbuf;		/* Pointer into directory at search_start.  */
558	struct ext2fs_direct *ep; /* Pointer to the entry we're examining.  */
559	/* XXX direct::d_reclen is 16-bit;
560	 * ufs_lookup_results::ulr_reclen is 32-bit.  Blah.  */
561	uint32_t reclen;	/* Length of the entry we're examining.  */
562	uint32_t prev_reclen;	/* Length of the preceding entry.  */
563	int error;
564
565	KASSERT(dvp != NULL);
566	KASSERT(dvp->v_mount != NULL);
567	KASSERT(VTOI(dvp) != NULL);
568	KASSERT(fulr != NULL);
569	KASSERT(tulr != NULL);
570	KASSERT(fulr != tulr);
571	KASSERT(ext2fs_rename_ulr_overlap_p(fulr, tulr));
572
573	mp = dvp->v_mount;
574	ump = VFSTOUFS(mp);
575	KASSERT(ump != NULL);
576	KASSERT(ump == VTOI(dvp)->i_ump);
577
578	dirblksiz = ump->um_dirblksiz;
579	KASSERT(0 < dirblksiz);
580	KASSERT((dirblksiz & (dirblksiz - 1)) == 0);
581
582	/* A directory block may not span across multiple I/O blocks.  */
583	KASSERT(dirblksiz <= mp->mnt_stat.f_iosize);
584
585	/* Find the bounds of the search.  */
586	search_start = tulr->ulr_offset;
587	KASSERT(fulr->ulr_reclen < (EXT2FS_MAXDIRSIZE - fulr->ulr_offset));
588	search_end = (fulr->ulr_offset + fulr->ulr_reclen);
589
590	/* Compaction must happen only within a directory block. (*)  */
591	KASSERT(search_start <= search_end);
592	KASSERT((search_end - (search_start &~ (dirblksiz - 1))) <= dirblksiz);
593
594	dirbuf = NULL;
595	bp = NULL;
596	error = ext2fs_blkatoff(dvp, (off_t)search_start, &dirbuf, &bp);
597	if (error)
598		return error;
599	KASSERT(dirbuf != NULL);
600	KASSERT(bp != NULL);
601
602	/*
603	 * Guarantee we sha'n't go past the end of the buffer we got.
604	 * dirbuf is bp->b_data + (search_start & (iosize - 1)), and
605	 * the valid range is [bp->b_data, bp->b_data + bp->b_bcount).
606	 */
607	KASSERT((search_end - search_start) <=
608	    (bp->b_bcount - (search_start & (mp->mnt_stat.f_iosize - 1))));
609
610	prev_reclen = fulr->ulr_count;
611	offset = search_start;
612
613	/*
614	 * Search from search_start to search_end for the entry matching
615	 * fcnp, which must be there because we found it before and it
616	 * should only at most have moved earlier.
617	 */
618	for (;;) {
619		KASSERT(search_start <= offset);
620		KASSERT(offset < search_end);
621
622		/*
623		 * Examine the directory entry at offset.
624		 */
625		ep = (struct ext2fs_direct *)
626		    (dirbuf + (offset - search_start));
627		reclen = fs2h16(ep->e2d_reclen);
628
629		if (ep->e2d_ino == 0)
630			goto next;	/* Entry is unused.  */
631
632		if (fs2h32(ep->e2d_ino) == UFS_WINO)
633			goto next;	/* Entry is whiteout.  */
634
635		if (fcnp->cn_namelen != ep->e2d_namlen)
636			goto next;	/* Wrong name length.  */
637
638		if (memcmp(ep->e2d_name, fcnp->cn_nameptr, fcnp->cn_namelen))
639			goto next;	/* Wrong name.  */
640
641		/* Got it!  */
642		break;
643
644next:
645		if (! ((reclen < search_end) &&
646			(offset < (search_end - reclen)))) {
647			brelse(bp, 0);
648			return EIO;	/* XXX Panic?  What?  */
649		}
650
651		/* We may not move past the search end.  */
652		KASSERT(reclen < search_end);
653		KASSERT(offset < (search_end - reclen));
654
655		/*
656		 * We may not move across a directory block boundary;
657		 * see (*) above.
658		 */
659		KASSERT((offset &~ (dirblksiz - 1)) ==
660		    ((offset + reclen) &~ (dirblksiz - 1)));
661
662		prev_reclen = reclen;
663		offset += reclen;
664	}
665
666	/*
667	 * Found the entry.  Record where.
668	 */
669	fulr->ulr_offset = offset;
670	fulr->ulr_reclen = reclen;
671
672	/*
673	 * Record the preceding record length, but not if we're at the
674	 * start of a directory block.
675	 */
676	fulr->ulr_count = ((offset & (dirblksiz - 1))? prev_reclen : 0);
677
678	brelse(bp, 0);
679	return 0;
680}
681
682/*
683 * ext2fs_gro_remove: Rename an object over another link to itself,
684 * effectively removing just the original link.
685 */
686static int
687ext2fs_gro_remove(struct mount *mp, kauth_cred_t cred,
688    struct vnode *dvp, struct componentname *cnp, void *de, struct vnode *vp,
689    nlink_t *tvp_nlinkp)
690{
691	struct ufs_lookup_results *ulr = de;
692	int error;
693
694	(void)mp;
695	KASSERT(mp != NULL);
696	KASSERT(dvp != NULL);
697	KASSERT(cnp != NULL);
698	KASSERT(ulr != NULL);
699	KASSERT(vp != NULL);
700	KASSERT(dvp != vp);
701	KASSERT(dvp->v_mount == mp);
702	KASSERT(vp->v_mount == mp);
703	KASSERT(dvp->v_type == VDIR);
704	KASSERT(vp->v_type != VDIR);
705	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
706	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
707
708	error = ext2fs_dirremove(dvp, ulr, cnp);
709	if (error)
710		return error;
711
712	KASSERT(0 < VTOI(vp)->i_e2fs_nlink);
713	VTOI(vp)->i_e2fs_nlink--;
714	VTOI(vp)->i_flag |= IN_CHANGE;
715
716	*tvp_nlinkp = VTOI(vp)->i_e2fs_nlink;
717
718	return 0;
719}
720
721/*
722 * ext2fs_gro_lookup: Look up and save the lookup results.
723 */
724static int
725ext2fs_gro_lookup(struct mount *mp, struct vnode *dvp,
726    struct componentname *cnp, void *de_ret, struct vnode **vp_ret)
727{
728	struct ufs_lookup_results *ulr_ret = de_ret;
729	struct vnode *vp;
730	int error;
731
732	(void)mp;
733	KASSERT(mp != NULL);
734	KASSERT(dvp != NULL);
735	KASSERT(cnp != NULL);
736	KASSERT(ulr_ret != NULL);
737	KASSERT(vp_ret != NULL);
738	KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
739
740	/* Kludge cargo-culted from dholland's ufs_rename.  */
741	cnp->cn_flags &=~ MODMASK;
742	cnp->cn_flags |= (LOCKPARENT | LOCKLEAF);
743
744	error = relookup(dvp, &vp, cnp, 0 /* dummy */);
745	if ((error == 0) && (vp == NULL)) {
746		error = ENOENT;
747		goto out;
748	} else if (error) {
749		return error;
750	}
751
752	/*
753	 * Thanks to VFS insanity, relookup locks vp, which screws us
754	 * in various ways.
755	 */
756	KASSERT(vp != NULL);
757	VOP_UNLOCK(vp);
758
759out:	*ulr_ret = VTOI(dvp)->i_crap;
760	*vp_ret = vp;
761	return error;
762}
763
764/*
765 * ext2fs_rmdired_p: Check whether the directory vp has been rmdired.
766 *
767 * vp must be locked and referenced.
768 */
769static bool
770ext2fs_rmdired_p(struct vnode *vp)
771{
772
773	KASSERT(vp != NULL);
774	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
775	KASSERT(vp->v_type == VDIR);
776
777	/* XXX Is this correct?  */
778	return ext2fs_size(VTOI(vp)) == 0;
779}
780
781/*
782 * ext2fs_gro_genealogy: Analyze the genealogy of the source and target
783 * directories.
784 */
785static int
786ext2fs_gro_genealogy(struct mount *mp, kauth_cred_t cred,
787    struct vnode *fdvp, struct vnode *tdvp,
788    struct vnode **intermediate_node_ret)
789{
790	struct vnode *vp, *dvp;
791	ino_t dotdot_ino = -1;	/* XXX gcc 4.8.3: maybe-uninitialized */
792	int error;
793
794	KASSERT(mp != NULL);
795	KASSERT(fdvp != NULL);
796	KASSERT(tdvp != NULL);
797	KASSERT(fdvp != tdvp);
798	KASSERT(intermediate_node_ret != NULL);
799	KASSERT(fdvp->v_mount == mp);
800	KASSERT(tdvp->v_mount == mp);
801	KASSERT(fdvp->v_type == VDIR);
802	KASSERT(tdvp->v_type == VDIR);
803
804	/*
805	 * We need to provisionally lock tdvp to keep rmdir from
806	 * deleting it -- or any ancestor -- at an inopportune moment.
807	 */
808	error = ext2fs_gro_lock_directory(mp, tdvp);
809	if (error)
810		return error;
811
812	vp = tdvp;
813	vref(vp);
814
815	for (;;) {
816		KASSERT(vp != NULL);
817		KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
818		KASSERT(vp->v_mount == mp);
819		KASSERT(vp->v_type == VDIR);
820		KASSERT(!ext2fs_rmdired_p(vp));
821
822		/* Did we hit the root without finding fdvp?  */
823		if (VTOI(vp)->i_number == UFS_ROOTINO) {
824			vput(vp);
825			*intermediate_node_ret = NULL;
826			return 0;
827		}
828
829		error = ext2fs_read_dotdot(vp, cred, &dotdot_ino);
830		if (error) {
831			vput(vp);
832			return error;
833		}
834
835		/* Did we find that fdvp is an ancestor of tdvp?  */
836		if (VTOI(fdvp)->i_number == dotdot_ino) {
837			/* Unlock vp, but keep it referenced.  */
838			VOP_UNLOCK(vp);
839			*intermediate_node_ret = vp;
840			return 0;
841		}
842
843		/* Neither -- keep ascending the family tree.  */
844		error = vcache_get(mp, &dotdot_ino, sizeof(dotdot_ino), &dvp);
845		vput(vp);
846		if (error)
847			return error;
848		error = vn_lock(dvp, LK_EXCLUSIVE);
849		if (error) {
850			vrele(dvp);
851			return error;
852		}
853
854		KASSERT(dvp != NULL);
855		KASSERT(VOP_ISLOCKED(dvp) == LK_EXCLUSIVE);
856		vp = dvp;
857
858		if (vp->v_type != VDIR) {
859			/*
860			 * XXX Panic?  Print a warning?  Can this
861			 * happen if we lose the race I suspect to
862			 * exist above, and the `..' inode number has
863			 * been recycled?
864			 */
865			vput(vp);
866			return ENOTDIR;
867		}
868
869		if (ext2fs_rmdired_p(vp)) {
870			vput(vp);
871			return ENOENT;
872		}
873	}
874}
875
876/*
877 * ext2fs_read_dotdot: Store in *ino_ret the inode number of the parent
878 * of the directory vp.
879 */
880static int
881ext2fs_read_dotdot(struct vnode *vp, kauth_cred_t cred, ino_t *ino_ret)
882{
883	struct ext2fs_dirtemplate dirbuf;
884	int error;
885
886	KASSERT(vp != NULL);
887	KASSERT(ino_ret != NULL);
888	KASSERT(vp->v_type == VDIR);
889
890	error = ufs_bufio(UIO_READ, vp, &dirbuf, sizeof dirbuf, (off_t)0,
891	    IO_NODELOCKED, cred, NULL, NULL);
892	if (error)
893		return error;
894
895	if (dirbuf.dotdot_namlen != 2 ||
896	    dirbuf.dotdot_name[0] != '.' ||
897	    dirbuf.dotdot_name[1] != '.')
898		/* XXX Panic?  Print warning?  */
899		return ENOTDIR;
900
901	*ino_ret = fs2h32(dirbuf.dotdot_ino);
902	return 0;
903}
904
905/*
906 * ext2fs_rename_replace_dotdot: Change the target of the `..' entry of
907 * the directory vp from fdvp to tdvp.
908 */
909static int
910ext2fs_rename_replace_dotdot(struct vnode *vp,
911    struct vnode *fdvp, struct vnode *tdvp,
912    kauth_cred_t cred)
913{
914	struct ext2fs_dirtemplate dirbuf;
915	int error;
916
917	/* XXX Does it make sense to do this before the sanity checks below?  */
918	KASSERT(0 < VTOI(fdvp)->i_e2fs_nlink);
919	VTOI(fdvp)->i_e2fs_nlink--;
920	VTOI(fdvp)->i_flag |= IN_CHANGE;
921
922	error = ufs_bufio(UIO_READ, vp, &dirbuf, sizeof dirbuf, (off_t)0,
923	    IO_NODELOCKED, cred, NULL, NULL);
924	if (error)
925		return error;
926
927	if (dirbuf.dotdot_namlen != 2 ||
928	    dirbuf.dotdot_name[0] != '.' ||
929	    dirbuf.dotdot_name[1] != '.') {
930		ufs_dirbad(VTOI(vp), (doff_t)12, "bad `..' entry");
931		return 0;
932	}
933
934	if (fs2h32(dirbuf.dotdot_ino) != VTOI(fdvp)->i_number) {
935		ufs_dirbad(VTOI(vp), (doff_t)12,
936		    "`..' does not point at parent");
937		return 0;
938	}
939
940	dirbuf.dotdot_ino = h2fs32(VTOI(tdvp)->i_number);
941	/* XXX WTF?  Why not check error?  */
942	(void)ufs_bufio(UIO_WRITE, vp, &dirbuf, sizeof dirbuf, (off_t)0,
943	    (IO_NODELOCKED | IO_SYNC), cred, NULL, NULL);
944
945	return 0;
946}
947
948/*
949 * ext2fs_gro_lock_directory: Lock the directory vp, but fail if it has
950 * been rmdir'd.
951 */
952static int
953ext2fs_gro_lock_directory(struct mount *mp, struct vnode *vp)
954{
955
956	(void)mp;
957	KASSERT(mp != NULL);
958	KASSERT(vp != NULL);
959	KASSERT(vp->v_mount == mp);
960
961	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
962
963	if (ext2fs_rmdired_p(vp)) {
964		VOP_UNLOCK(vp);
965		return ENOENT;
966	}
967
968	return 0;
969}
970
971static const struct genfs_rename_ops ext2fs_genfs_rename_ops = {
972	.gro_directory_empty_p		= ext2fs_gro_directory_empty_p,
973	.gro_rename_check_possible	= ext2fs_gro_rename_check_possible,
974	.gro_rename_check_permitted	= ext2fs_gro_rename_check_permitted,
975	.gro_remove_check_possible	= ext2fs_gro_remove_check_possible,
976	.gro_remove_check_permitted	= ext2fs_gro_remove_check_permitted,
977	.gro_rename			= ext2fs_gro_rename,
978	.gro_remove			= ext2fs_gro_remove,
979	.gro_lookup			= ext2fs_gro_lookup,
980	.gro_genealogy			= ext2fs_gro_genealogy,
981	.gro_lock_directory		= ext2fs_gro_lock_directory,
982};
983