xref: /freebsd/sys/fs/ext2fs/ext2_htree.c (revision 4f52dfbb8d6c4d446500c5b097e3806ec219fbd4)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010, 2012 Zheng Liu <lz@freebsd.org>
5  * Copyright (c) 2012, Vyacheslav Matyushin
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/systm.h>
35 #include <sys/namei.h>
36 #include <sys/bio.h>
37 #include <sys/buf.h>
38 #include <sys/endian.h>
39 #include <sys/mount.h>
40 #include <sys/vnode.h>
41 #include <sys/malloc.h>
42 #include <sys/dirent.h>
43 #include <sys/sysctl.h>
44 
45 #include <ufs/ufs/dir.h>
46 
47 #include <fs/ext2fs/inode.h>
48 #include <fs/ext2fs/ext2_mount.h>
49 #include <fs/ext2fs/ext2fs.h>
50 #include <fs/ext2fs/fs.h>
51 #include <fs/ext2fs/ext2_extern.h>
52 #include <fs/ext2fs/ext2_dinode.h>
53 #include <fs/ext2fs/ext2_dir.h>
54 #include <fs/ext2fs/htree.h>
55 
56 static void	ext2_append_entry(char *block, uint32_t blksize,
57 		    struct ext2fs_direct_2 *last_entry,
58 		    struct ext2fs_direct_2 *new_entry);
59 static int	ext2_htree_append_block(struct vnode *vp, char *data,
60 		    struct componentname *cnp, uint32_t blksize);
61 static int	ext2_htree_check_next(struct inode *ip, uint32_t hash,
62 		    const char *name, struct ext2fs_htree_lookup_info *info);
63 static int	ext2_htree_cmp_sort_entry(const void *e1, const void *e2);
64 static int	ext2_htree_find_leaf(struct inode *ip, const char *name,
65 		    int namelen, uint32_t *hash, uint8_t *hash_version,
66 		    struct ext2fs_htree_lookup_info *info);
67 static uint32_t ext2_htree_get_block(struct ext2fs_htree_entry *ep);
68 static uint16_t	ext2_htree_get_count(struct ext2fs_htree_entry *ep);
69 static uint32_t ext2_htree_get_hash(struct ext2fs_htree_entry *ep);
70 static uint16_t	ext2_htree_get_limit(struct ext2fs_htree_entry *ep);
71 static void	ext2_htree_insert_entry_to_level(struct ext2fs_htree_lookup_level *level,
72 		    uint32_t hash, uint32_t blk);
73 static void	ext2_htree_insert_entry(struct ext2fs_htree_lookup_info *info,
74 		    uint32_t hash, uint32_t blk);
75 static uint32_t	ext2_htree_node_limit(struct inode *ip);
76 static void	ext2_htree_set_block(struct ext2fs_htree_entry *ep,
77 		    uint32_t blk);
78 static void	ext2_htree_set_count(struct ext2fs_htree_entry *ep,
79 		    uint16_t cnt);
80 static void	ext2_htree_set_hash(struct ext2fs_htree_entry *ep,
81 		    uint32_t hash);
82 static void	ext2_htree_set_limit(struct ext2fs_htree_entry *ep,
83 		    uint16_t limit);
84 static int	ext2_htree_split_dirblock(char *block1, char *block2,
85 		    uint32_t blksize, uint32_t *hash_seed, uint8_t hash_version,
86 		    uint32_t *split_hash, struct  ext2fs_direct_2 *entry);
87 static void	ext2_htree_release(struct ext2fs_htree_lookup_info *info);
88 static uint32_t	ext2_htree_root_limit(struct inode *ip, int len);
89 static int	ext2_htree_writebuf(struct ext2fs_htree_lookup_info *info);
90 
91 int
92 ext2_htree_has_idx(struct inode *ip)
93 {
94 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
95 	    ip->i_flag & IN_E3INDEX)
96 		return (1);
97 	else
98 		return (0);
99 }
100 
101 static int
102 ext2_htree_check_next(struct inode *ip, uint32_t hash, const char *name,
103     struct ext2fs_htree_lookup_info *info)
104 {
105 	struct vnode *vp = ITOV(ip);
106 	struct ext2fs_htree_lookup_level *level;
107 	struct buf *bp;
108 	uint32_t next_hash;
109 	int idx = info->h_levels_num - 1;
110 	int levels = 0;
111 
112 	do {
113 		level = &info->h_levels[idx];
114 		level->h_entry++;
115 		if (level->h_entry < level->h_entries +
116 		    ext2_htree_get_count(level->h_entries))
117 			break;
118 		if (idx == 0)
119 			return (0);
120 		idx--;
121 		levels++;
122 	} while (1);
123 
124 	next_hash = ext2_htree_get_hash(level->h_entry);
125 	if ((hash & 1) == 0) {
126 		if (hash != (next_hash & ~1))
127 			return (0);
128 	}
129 
130 	while (levels > 0) {
131 		levels--;
132 		if (ext2_blkatoff(vp, ext2_htree_get_block(level->h_entry) *
133 		    ip->i_e2fs->e2fs_bsize, NULL, &bp) != 0)
134 			return (0);
135 		level = &info->h_levels[idx + 1];
136 		brelse(level->h_bp);
137 		level->h_bp = bp;
138 		level->h_entry = level->h_entries =
139 		    ((struct ext2fs_htree_node *)bp->b_data)->h_entries;
140 	}
141 
142 	return (1);
143 }
144 
145 static uint32_t
146 ext2_htree_get_block(struct ext2fs_htree_entry *ep)
147 {
148 	return (ep->h_blk & 0x00FFFFFF);
149 }
150 
151 static void
152 ext2_htree_set_block(struct ext2fs_htree_entry *ep, uint32_t blk)
153 {
154 	ep->h_blk = blk;
155 }
156 
157 static uint16_t
158 ext2_htree_get_count(struct ext2fs_htree_entry *ep)
159 {
160 	return (((struct ext2fs_htree_count *)(ep))->h_entries_num);
161 }
162 
163 static void
164 ext2_htree_set_count(struct ext2fs_htree_entry *ep, uint16_t cnt)
165 {
166 	((struct ext2fs_htree_count *)(ep))->h_entries_num = cnt;
167 }
168 
169 static uint32_t
170 ext2_htree_get_hash(struct ext2fs_htree_entry *ep)
171 {
172 	return (ep->h_hash);
173 }
174 
175 static uint16_t
176 ext2_htree_get_limit(struct ext2fs_htree_entry *ep)
177 {
178 	return (((struct ext2fs_htree_count *)(ep))->h_entries_max);
179 }
180 
181 static void
182 ext2_htree_set_hash(struct ext2fs_htree_entry *ep, uint32_t hash)
183 {
184 	ep->h_hash = hash;
185 }
186 
187 static void
188 ext2_htree_set_limit(struct ext2fs_htree_entry *ep, uint16_t limit)
189 {
190 	((struct ext2fs_htree_count *)(ep))->h_entries_max = limit;
191 }
192 
193 static void
194 ext2_htree_release(struct ext2fs_htree_lookup_info *info)
195 {
196 	u_int i;
197 
198 	for (i = 0; i < info->h_levels_num; i++) {
199 		struct buf *bp = info->h_levels[i].h_bp;
200 
201 		if (bp != NULL)
202 			brelse(bp);
203 	}
204 }
205 
206 static uint32_t
207 ext2_htree_root_limit(struct inode *ip, int len)
208 {
209 	uint32_t space;
210 
211 	space = ip->i_e2fs->e2fs_bsize - EXT2_DIR_REC_LEN(1) -
212 	    EXT2_DIR_REC_LEN(2) - len;
213 	return (space / sizeof(struct ext2fs_htree_entry));
214 }
215 
216 static uint32_t
217 ext2_htree_node_limit(struct inode *ip)
218 {
219 	struct m_ext2fs *fs;
220 	uint32_t space;
221 
222 	fs = ip->i_e2fs;
223 	space = fs->e2fs_bsize - EXT2_DIR_REC_LEN(0);
224 
225 	return (space / sizeof(struct ext2fs_htree_entry));
226 }
227 
228 static int
229 ext2_htree_find_leaf(struct inode *ip, const char *name, int namelen,
230     uint32_t *hash, uint8_t *hash_ver,
231     struct ext2fs_htree_lookup_info *info)
232 {
233 	struct vnode *vp;
234 	struct ext2fs *fs;
235 	struct m_ext2fs *m_fs;
236 	struct buf *bp = NULL;
237 	struct ext2fs_htree_root *rootp;
238 	struct ext2fs_htree_entry *entp, *start, *end, *middle, *found;
239 	struct ext2fs_htree_lookup_level *level_info;
240 	uint32_t hash_major = 0, hash_minor = 0;
241 	uint32_t levels, cnt;
242 	uint8_t hash_version;
243 
244 	if (name == NULL || info == NULL)
245 		return (-1);
246 
247 	vp = ITOV(ip);
248 	fs = ip->i_e2fs->e2fs;
249 	m_fs = ip->i_e2fs;
250 
251 	if (ext2_blkatoff(vp, 0, NULL, &bp) != 0)
252 		return (-1);
253 
254 	info->h_levels_num = 1;
255 	info->h_levels[0].h_bp = bp;
256 	rootp = (struct ext2fs_htree_root *)bp->b_data;
257 	if (rootp->h_info.h_hash_version != EXT2_HTREE_LEGACY &&
258 	    rootp->h_info.h_hash_version != EXT2_HTREE_HALF_MD4 &&
259 	    rootp->h_info.h_hash_version != EXT2_HTREE_TEA)
260 		goto error;
261 
262 	hash_version = rootp->h_info.h_hash_version;
263 	if (hash_version <= EXT2_HTREE_TEA)
264 		hash_version += m_fs->e2fs_uhash;
265 	*hash_ver = hash_version;
266 
267 	ext2_htree_hash(name, namelen, fs->e3fs_hash_seed,
268 	    hash_version, &hash_major, &hash_minor);
269 	*hash = hash_major;
270 
271 	if ((levels = rootp->h_info.h_ind_levels) > 1)
272 		goto error;
273 
274 	entp = (struct ext2fs_htree_entry *)(((char *)&rootp->h_info) +
275 	    rootp->h_info.h_info_len);
276 
277 	if (ext2_htree_get_limit(entp) !=
278 	    ext2_htree_root_limit(ip, rootp->h_info.h_info_len))
279 		goto error;
280 
281 	while (1) {
282 		cnt = ext2_htree_get_count(entp);
283 		if (cnt == 0 || cnt > ext2_htree_get_limit(entp))
284 			goto error;
285 
286 		start = entp + 1;
287 		end = entp + cnt - 1;
288 		while (start <= end) {
289 			middle = start + (end - start) / 2;
290 			if (ext2_htree_get_hash(middle) > hash_major)
291 				end = middle - 1;
292 			else
293 				start = middle + 1;
294 		}
295 		found = start - 1;
296 
297 		level_info = &(info->h_levels[info->h_levels_num - 1]);
298 		level_info->h_bp = bp;
299 		level_info->h_entries = entp;
300 		level_info->h_entry = found;
301 		if (levels == 0)
302 			return (0);
303 		levels--;
304 		if (ext2_blkatoff(vp,
305 		    ext2_htree_get_block(found) * m_fs->e2fs_bsize,
306 		    NULL, &bp) != 0)
307 			goto error;
308 		entp = ((struct ext2fs_htree_node *)bp->b_data)->h_entries;
309 		info->h_levels_num++;
310 		info->h_levels[info->h_levels_num - 1].h_bp = bp;
311 	}
312 
313 error:
314 	ext2_htree_release(info);
315 	return (-1);
316 }
317 
318 /*
319  * Try to lookup a directory entry in HTree index
320  */
321 int
322 ext2_htree_lookup(struct inode *ip, const char *name, int namelen,
323     struct buf **bpp, int *entryoffp, doff_t *offp,
324     doff_t *prevoffp, doff_t *endusefulp,
325     struct ext2fs_searchslot *ss)
326 {
327 	struct vnode *vp;
328 	struct ext2fs_htree_lookup_info info;
329 	struct ext2fs_htree_entry *leaf_node;
330 	struct m_ext2fs *m_fs;
331 	struct buf *bp;
332 	uint32_t blk;
333 	uint32_t dirhash;
334 	uint32_t bsize;
335 	uint8_t hash_version;
336 	int search_next;
337 	int found = 0;
338 
339 	m_fs = ip->i_e2fs;
340 	bsize = m_fs->e2fs_bsize;
341 	vp = ITOV(ip);
342 
343 	/* TODO: print error msg because we don't lookup '.' and '..' */
344 
345 	memset(&info, 0, sizeof(info));
346 	if (ext2_htree_find_leaf(ip, name, namelen, &dirhash,
347 	    &hash_version, &info))
348 		return (-1);
349 
350 	do {
351 		leaf_node = info.h_levels[info.h_levels_num - 1].h_entry;
352 		blk = ext2_htree_get_block(leaf_node);
353 		if (ext2_blkatoff(vp, blk * bsize, NULL, &bp) != 0) {
354 			ext2_htree_release(&info);
355 			return (-1);
356 		}
357 
358 		*offp = blk * bsize;
359 		*entryoffp = 0;
360 		*prevoffp = blk * bsize;
361 		*endusefulp = blk * bsize;
362 
363 		if (ss->slotstatus == NONE) {
364 			ss->slotoffset = -1;
365 			ss->slotfreespace = 0;
366 		}
367 
368 		if (ext2_search_dirblock(ip, bp->b_data, &found,
369 		    name, namelen, entryoffp, offp, prevoffp,
370 		    endusefulp, ss) != 0) {
371 			brelse(bp);
372 			ext2_htree_release(&info);
373 			return (-1);
374 		}
375 
376 		if (found) {
377 			*bpp = bp;
378 			ext2_htree_release(&info);
379 			return (0);
380 		}
381 
382 		brelse(bp);
383 		search_next = ext2_htree_check_next(ip, dirhash, name, &info);
384 	} while (search_next);
385 
386 	ext2_htree_release(&info);
387 	return (ENOENT);
388 }
389 
390 static int
391 ext2_htree_append_block(struct vnode *vp, char *data,
392     struct componentname *cnp, uint32_t blksize)
393 {
394 	struct iovec aiov;
395 	struct uio auio;
396 	struct inode *dp = VTOI(vp);
397 	uint64_t cursize, newsize;
398 	int error;
399 
400 	cursize = roundup(dp->i_size, blksize);
401 	newsize = cursize + blksize;
402 
403 	auio.uio_offset = cursize;
404 	auio.uio_resid = blksize;
405 	aiov.iov_len = blksize;
406 	aiov.iov_base = data;
407 	auio.uio_iov = &aiov;
408 	auio.uio_iovcnt = 1;
409 	auio.uio_rw = UIO_WRITE;
410 	auio.uio_segflg = UIO_SYSSPACE;
411 	error = VOP_WRITE(vp, &auio, IO_SYNC, cnp->cn_cred);
412 	if (!error)
413 		dp->i_size = newsize;
414 
415 	return (error);
416 }
417 
418 static int
419 ext2_htree_writebuf(struct ext2fs_htree_lookup_info *info)
420 {
421 	int i, error;
422 
423 	for (i = 0; i < info->h_levels_num; i++) {
424 		struct buf *bp = info->h_levels[i].h_bp;
425 
426 		error = bwrite(bp);
427 		if (error)
428 			return (error);
429 	}
430 
431 	return (0);
432 }
433 
434 static void
435 ext2_htree_insert_entry_to_level(struct ext2fs_htree_lookup_level *level,
436     uint32_t hash, uint32_t blk)
437 {
438 	struct ext2fs_htree_entry *target;
439 	int entries_num;
440 
441 	target = level->h_entry + 1;
442 	entries_num = ext2_htree_get_count(level->h_entries);
443 
444 	memmove(target + 1, target, (char *)(level->h_entries + entries_num) -
445 	    (char *)target);
446 	ext2_htree_set_block(target, blk);
447 	ext2_htree_set_hash(target, hash);
448 	ext2_htree_set_count(level->h_entries, entries_num + 1);
449 }
450 
451 /*
452  * Insert an index entry to the index node.
453  */
454 static void
455 ext2_htree_insert_entry(struct ext2fs_htree_lookup_info *info,
456     uint32_t hash, uint32_t blk)
457 {
458 	struct ext2fs_htree_lookup_level *level;
459 
460 	level = &info->h_levels[info->h_levels_num - 1];
461 	ext2_htree_insert_entry_to_level(level, hash, blk);
462 }
463 
464 /*
465  * Compare two entry sort descriptors by name hash value.
466  * This is used together with qsort.
467  */
468 static int
469 ext2_htree_cmp_sort_entry(const void *e1, const void *e2)
470 {
471 	const struct ext2fs_htree_sort_entry *entry1, *entry2;
472 
473 	entry1 = (const struct ext2fs_htree_sort_entry *)e1;
474 	entry2 = (const struct ext2fs_htree_sort_entry *)e2;
475 
476 	if (entry1->h_hash < entry2->h_hash)
477 		return (-1);
478 	if (entry1->h_hash > entry2->h_hash)
479 		return (1);
480 	return (0);
481 }
482 
483 /*
484  * Append an entry to the end of the directory block.
485  */
486 static void
487 ext2_append_entry(char *block, uint32_t blksize,
488     struct ext2fs_direct_2 *last_entry,
489     struct ext2fs_direct_2 *new_entry)
490 {
491 	uint16_t entry_len;
492 
493 	entry_len = EXT2_DIR_REC_LEN(last_entry->e2d_namlen);
494 	last_entry->e2d_reclen = entry_len;
495 	last_entry = (struct ext2fs_direct_2 *)((char *)last_entry + entry_len);
496 	new_entry->e2d_reclen = block + blksize - (char *)last_entry;
497 	memcpy(last_entry, new_entry, EXT2_DIR_REC_LEN(new_entry->e2d_namlen));
498 }
499 
500 /*
501  * Move half of entries from the old directory block to the new one.
502  */
503 static int
504 ext2_htree_split_dirblock(char *block1, char *block2, uint32_t blksize,
505     uint32_t *hash_seed, uint8_t hash_version,
506     uint32_t *split_hash, struct ext2fs_direct_2 *entry)
507 {
508 	int entry_cnt = 0;
509 	int size = 0;
510 	int i, k;
511 	uint32_t offset;
512 	uint16_t entry_len = 0;
513 	uint32_t entry_hash;
514 	struct ext2fs_direct_2 *ep, *last;
515 	char *dest;
516 	struct ext2fs_htree_sort_entry *sort_info;
517 
518 	ep = (struct ext2fs_direct_2 *)block1;
519 	dest = block2;
520 	sort_info = (struct ext2fs_htree_sort_entry *)
521 	    ((char *)block2 + blksize);
522 
523 	/*
524 	 * Calculate name hash value for the entry which is to be added.
525 	 */
526 	ext2_htree_hash(entry->e2d_name, entry->e2d_namlen, hash_seed,
527 	    hash_version, &entry_hash, NULL);
528 
529 	/*
530 	 * Fill in directory entry sort descriptors.
531 	 */
532 	while ((char *)ep < block1 + blksize) {
533 		if (ep->e2d_ino && ep->e2d_namlen) {
534 			entry_cnt++;
535 			sort_info--;
536 			sort_info->h_size = ep->e2d_reclen;
537 			sort_info->h_offset = (char *)ep - block1;
538 			ext2_htree_hash(ep->e2d_name, ep->e2d_namlen,
539 			    hash_seed, hash_version,
540 			    &sort_info->h_hash, NULL);
541 		}
542 		ep = (struct ext2fs_direct_2 *)
543 		    ((char *)ep + ep->e2d_reclen);
544 	}
545 
546 	/*
547 	 * Sort directory entry descriptors by name hash value.
548 	 */
549 	qsort(sort_info, entry_cnt, sizeof(struct ext2fs_htree_sort_entry),
550 	    ext2_htree_cmp_sort_entry);
551 
552 	/*
553 	 * Count the number of entries to move to directory block 2.
554 	 */
555 	for (i = entry_cnt - 1; i >= 0; i--) {
556 		if (sort_info[i].h_size + size > blksize / 2)
557 			break;
558 		size += sort_info[i].h_size;
559 	}
560 
561 	*split_hash = sort_info[i + 1].h_hash;
562 
563 	/*
564 	 * Set collision bit.
565 	 */
566 	if (*split_hash == sort_info[i].h_hash)
567 		*split_hash += 1;
568 
569 	/*
570 	 * Move half of directory entries from block 1 to block 2.
571 	 */
572 	for (k = i + 1; k < entry_cnt; k++) {
573 		ep = (struct ext2fs_direct_2 *)((char *)block1 +
574 		    sort_info[k].h_offset);
575 		entry_len = EXT2_DIR_REC_LEN(ep->e2d_namlen);
576 		memcpy(dest, ep, entry_len);
577 		((struct ext2fs_direct_2 *)dest)->e2d_reclen = entry_len;
578 		/* Mark directory entry as unused. */
579 		ep->e2d_ino = 0;
580 		dest += entry_len;
581 	}
582 	dest -= entry_len;
583 
584 	/* Shrink directory entries in block 1. */
585 	last = (struct ext2fs_direct_2 *)block1;
586 	entry_len = 0;
587 	for (offset = 0; offset < blksize; ) {
588 		ep = (struct ext2fs_direct_2 *)(block1 + offset);
589 		offset += ep->e2d_reclen;
590 		if (ep->e2d_ino) {
591 			last = (struct ext2fs_direct_2 *)
592 			    ((char *)last + entry_len);
593 			entry_len = EXT2_DIR_REC_LEN(ep->e2d_namlen);
594 			memcpy((void *)last, (void *)ep, entry_len);
595 			last->e2d_reclen = entry_len;
596 		}
597 	}
598 
599 	if (entry_hash >= *split_hash) {
600 		/* Add entry to block 2. */
601 		ext2_append_entry(block2, blksize,
602 		    (struct ext2fs_direct_2 *)dest, entry);
603 
604 		/* Adjust length field of last entry of block 1. */
605 		last->e2d_reclen = block1 + blksize - (char *)last;
606 	} else {
607 		/* Add entry to block 1. */
608 		ext2_append_entry(block1, blksize, last, entry);
609 
610 		/* Adjust length field of last entry of block 2. */
611 		((struct ext2fs_direct_2 *)dest)->e2d_reclen =
612 		    block2 + blksize - dest;
613 	}
614 
615 	return (0);
616 }
617 
618 /*
619  * Create an HTree index for a directory
620  */
621 int
622 ext2_htree_create_index(struct vnode *vp, struct componentname *cnp,
623     struct ext2fs_direct_2 *new_entry)
624 {
625 	struct buf *bp = NULL;
626 	struct inode *dp;
627 	struct ext2fs *fs;
628 	struct m_ext2fs *m_fs;
629 	struct ext2fs_direct_2 *ep, *dotdot;
630 	struct ext2fs_htree_root *root;
631 	struct ext2fs_htree_lookup_info info;
632 	uint32_t blksize, dirlen, split_hash;
633 	uint8_t hash_version;
634 	char *buf1 = NULL;
635 	char *buf2 = NULL;
636 	int error = 0;
637 
638 	dp = VTOI(vp);
639 	fs = dp->i_e2fs->e2fs;
640 	m_fs = dp->i_e2fs;
641 	blksize = m_fs->e2fs_bsize;
642 
643 	buf1 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
644 	buf2 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
645 
646 	if ((error = ext2_blkatoff(vp, 0, NULL, &bp)) != 0)
647 		goto out;
648 
649 	root = (struct ext2fs_htree_root *)bp->b_data;
650 	dotdot = (struct ext2fs_direct_2 *)((char *)&(root->h_dotdot));
651 	ep = (struct ext2fs_direct_2 *)((char *)dotdot + dotdot->e2d_reclen);
652 	dirlen = (char *)root + blksize - (char *)ep;
653 	memcpy(buf1, ep, dirlen);
654 	ep = (struct ext2fs_direct_2 *)buf1;
655 	while ((char *)ep < buf1 + dirlen)
656 		ep = (struct ext2fs_direct_2 *)
657 		    ((char *)ep + ep->e2d_reclen);
658 	ep->e2d_reclen = buf1 + blksize - (char *)ep;
659 
660 	dp->i_flag |= IN_E3INDEX;
661 
662 	/*
663 	 * Initialize index root.
664 	 */
665 	dotdot->e2d_reclen = blksize - EXT2_DIR_REC_LEN(1);
666 	memset(&root->h_info, 0, sizeof(root->h_info));
667 	root->h_info.h_hash_version = fs->e3fs_def_hash_version;
668 	root->h_info.h_info_len = sizeof(root->h_info);
669 	ext2_htree_set_block(root->h_entries, 1);
670 	ext2_htree_set_count(root->h_entries, 1);
671 	ext2_htree_set_limit(root->h_entries,
672 	    ext2_htree_root_limit(dp, sizeof(root->h_info)));
673 
674 	memset(&info, 0, sizeof(info));
675 	info.h_levels_num = 1;
676 	info.h_levels[0].h_entries = root->h_entries;
677 	info.h_levels[0].h_entry = root->h_entries;
678 
679 	hash_version = root->h_info.h_hash_version;
680 	if (hash_version <= EXT2_HTREE_TEA)
681 		hash_version += m_fs->e2fs_uhash;
682 	ext2_htree_split_dirblock(buf1, buf2, blksize, fs->e3fs_hash_seed,
683 	    hash_version, &split_hash, new_entry);
684 	ext2_htree_insert_entry(&info, split_hash, 2);
685 
686 	/*
687 	 * Write directory block 0.
688 	 */
689 	if (DOINGASYNC(vp)) {
690 		bdwrite(bp);
691 		error = 0;
692 	} else {
693 		error = bwrite(bp);
694 	}
695 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
696 	if (error)
697 		goto out;
698 
699 	/*
700 	 * Write directory block 1.
701 	 */
702 	error = ext2_htree_append_block(vp, buf1, cnp, blksize);
703 	if (error)
704 		goto out1;
705 
706 	/*
707 	 * Write directory block 2.
708 	 */
709 	error = ext2_htree_append_block(vp, buf2, cnp, blksize);
710 
711 	free(buf1, M_TEMP);
712 	free(buf2, M_TEMP);
713 	return (error);
714 out:
715 	if (bp != NULL)
716 		brelse(bp);
717 out1:
718 	free(buf1, M_TEMP);
719 	free(buf2, M_TEMP);
720 	return (error);
721 }
722 
723 /*
724  * Add an entry to the directory using htree index.
725  */
726 int
727 ext2_htree_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
728     struct componentname *cnp)
729 {
730 	struct ext2fs_htree_entry *entries, *leaf_node;
731 	struct ext2fs_htree_lookup_info info;
732 	struct buf *bp = NULL;
733 	struct ext2fs *fs;
734 	struct m_ext2fs *m_fs;
735 	struct inode *ip;
736 	uint16_t ent_num;
737 	uint32_t dirhash, split_hash;
738 	uint32_t blksize, blknum;
739 	uint64_t cursize, dirsize;
740 	uint8_t hash_version;
741 	char *newdirblock = NULL;
742 	char *newidxblock = NULL;
743 	struct ext2fs_htree_node *dst_node;
744 	struct ext2fs_htree_entry *dst_entries;
745 	struct ext2fs_htree_entry *root_entires;
746 	struct buf *dst_bp = NULL;
747 	int error, write_bp = 0, write_dst_bp = 0, write_info = 0;
748 
749 	ip = VTOI(dvp);
750 	m_fs = ip->i_e2fs;
751 	fs = m_fs->e2fs;
752 	blksize = m_fs->e2fs_bsize;
753 
754 	if (ip->i_count != 0)
755 		return ext2_add_entry(dvp, entry);
756 
757 	/* Target directory block is full, split it */
758 	memset(&info, 0, sizeof(info));
759 	error = ext2_htree_find_leaf(ip, entry->e2d_name, entry->e2d_namlen,
760 	    &dirhash, &hash_version, &info);
761 	if (error)
762 		return (error);
763 
764 	entries = info.h_levels[info.h_levels_num - 1].h_entries;
765 	ent_num = ext2_htree_get_count(entries);
766 	if (ent_num == ext2_htree_get_limit(entries)) {
767 		/* Split the index node. */
768 		root_entires = info.h_levels[0].h_entries;
769 		newidxblock = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
770 		dst_node = (struct ext2fs_htree_node *)newidxblock;
771 		memset(&dst_node->h_fake_dirent, 0,
772 		    sizeof(dst_node->h_fake_dirent));
773 		dst_node->h_fake_dirent.e2d_reclen = blksize;
774 
775 		cursize = roundup(ip->i_size, blksize);
776 		dirsize = cursize + blksize;
777 		blknum = dirsize / blksize - 1;
778 
779 		error = ext2_htree_append_block(dvp, newidxblock,
780 		    cnp, blksize);
781 		if (error)
782 			goto finish;
783 		error = ext2_blkatoff(dvp, cursize, NULL, &dst_bp);
784 		if (error)
785 			goto finish;
786 		dst_node = (struct ext2fs_htree_node *)dst_bp->b_data;
787 		dst_entries = dst_node->h_entries;
788 
789 		if (info.h_levels_num == 2) {
790 			uint16_t src_ent_num, dst_ent_num;
791 
792 			if (ext2_htree_get_count(root_entires) ==
793 			    ext2_htree_get_limit(root_entires)) {
794 				/* Directory index is full */
795 				error = EIO;
796 				goto finish;
797 			}
798 
799 			src_ent_num = ent_num / 2;
800 			dst_ent_num = ent_num - src_ent_num;
801 			split_hash = ext2_htree_get_hash(entries + src_ent_num);
802 
803 			/* Move half of index entries to the new index node */
804 			memcpy(dst_entries, entries + src_ent_num,
805 			    dst_ent_num * sizeof(struct ext2fs_htree_entry));
806 			ext2_htree_set_count(entries, src_ent_num);
807 			ext2_htree_set_count(dst_entries, dst_ent_num);
808 			ext2_htree_set_limit(dst_entries,
809 			    ext2_htree_node_limit(ip));
810 
811 			if (info.h_levels[1].h_entry >= entries + src_ent_num) {
812 				struct buf *tmp = info.h_levels[1].h_bp;
813 
814 				info.h_levels[1].h_bp = dst_bp;
815 				dst_bp = tmp;
816 
817 				info.h_levels[1].h_entry =
818 				    info.h_levels[1].h_entry -
819 				    (entries + src_ent_num) +
820 				    dst_entries;
821 				info.h_levels[1].h_entries = dst_entries;
822 			}
823 			ext2_htree_insert_entry_to_level(&info.h_levels[0],
824 			    split_hash, blknum);
825 
826 			/* Write new index node to disk */
827 			error = bwrite(dst_bp);
828 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
829 			if (error)
830 				goto finish;
831 			write_dst_bp = 1;
832 		} else {
833 			/* Create second level for htree index */
834 			struct ext2fs_htree_root *idx_root;
835 
836 			memcpy(dst_entries, entries,
837 			    ent_num * sizeof(struct ext2fs_htree_entry));
838 			ext2_htree_set_limit(dst_entries,
839 			    ext2_htree_node_limit(ip));
840 
841 			idx_root = (struct ext2fs_htree_root *)
842 			    info.h_levels[0].h_bp->b_data;
843 			idx_root->h_info.h_ind_levels = 1;
844 
845 			ext2_htree_set_count(entries, 1);
846 			ext2_htree_set_block(entries, blknum);
847 
848 			info.h_levels_num = 2;
849 			info.h_levels[1].h_entries = dst_entries;
850 			info.h_levels[1].h_entry = info.h_levels[0].h_entry -
851 			    info.h_levels[0].h_entries + dst_entries;
852 			info.h_levels[1].h_bp = dst_bp;
853 			dst_bp = NULL;
854 		}
855 	}
856 
857 	leaf_node = info.h_levels[info.h_levels_num - 1].h_entry;
858 	blknum = ext2_htree_get_block(leaf_node);
859 	error = ext2_blkatoff(dvp, blknum * blksize, NULL, &bp);
860 	if (error)
861 		goto finish;
862 
863 	/* Split target directory block */
864 	newdirblock = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
865 	ext2_htree_split_dirblock((char *)bp->b_data, newdirblock, blksize,
866 	    fs->e3fs_hash_seed, hash_version, &split_hash, entry);
867 	cursize = roundup(ip->i_size, blksize);
868 	dirsize = cursize + blksize;
869 	blknum = dirsize / blksize - 1;
870 
871 	/* Add index entry for the new directory block */
872 	ext2_htree_insert_entry(&info, split_hash, blknum);
873 
874 	/* Write the new directory block to the end of the directory */
875 	error = ext2_htree_append_block(dvp, newdirblock, cnp, blksize);
876 	if (error)
877 		goto finish;
878 
879 	/* Write the target directory block */
880 	error = bwrite(bp);
881 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
882 	if (error)
883 		goto finish;
884 	write_bp = 1;
885 
886 	/* Write the index block */
887 	error = ext2_htree_writebuf(&info);
888 	if (!error)
889 		write_info = 1;
890 
891 finish:
892 	if (dst_bp != NULL && !write_dst_bp)
893 		brelse(dst_bp);
894 	if (bp != NULL && !write_bp)
895 		brelse(bp);
896 	if (newdirblock != NULL)
897 		free(newdirblock, M_TEMP);
898 	if (newidxblock != NULL)
899 		free(newidxblock, M_TEMP);
900 	if (!write_info)
901 		ext2_htree_release(&info);
902 	return (error);
903 }
904