xref: /freebsd/sys/fs/ext2fs/ext2_htree.c (revision e1a528369708afb723290916ad8ea9c79399e933)
1 /*-
2  * Copyright (c) 2010, 2012 Zheng Liu <lz@freebsd.org>
3  * Copyright (c) 2012, Vyacheslav Matyushin
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  */
29 
30 #include <sys/param.h>
31 #include <sys/endian.h>
32 #include <sys/systm.h>
33 #include <sys/namei.h>
34 #include <sys/bio.h>
35 #include <sys/buf.h>
36 #include <sys/endian.h>
37 #include <sys/mount.h>
38 #include <sys/vnode.h>
39 #include <sys/malloc.h>
40 #include <sys/dirent.h>
41 #include <sys/sysctl.h>
42 
43 #include <ufs/ufs/dir.h>
44 
45 #include <fs/ext2fs/inode.h>
46 #include <fs/ext2fs/ext2_mount.h>
47 #include <fs/ext2fs/ext2fs.h>
48 #include <fs/ext2fs/fs.h>
49 #include <fs/ext2fs/ext2_extern.h>
50 #include <fs/ext2fs/ext2_dinode.h>
51 #include <fs/ext2fs/ext2_dir.h>
52 #include <fs/ext2fs/htree.h>
53 
54 static void	ext2_append_entry(char *block, uint32_t blksize,
55 		    struct ext2fs_direct_2 *last_entry,
56 		    struct ext2fs_direct_2 *new_entry);
57 static int	ext2_htree_append_block(struct vnode *vp, char *data,
58 		    struct componentname *cnp, uint32_t blksize);
59 static int	ext2_htree_check_next(struct inode *ip, uint32_t hash,
60 		    const char *name, struct ext2fs_htree_lookup_info *info);
61 static int	ext2_htree_cmp_sort_entry(const void *e1, const void *e2);
62 static int	ext2_htree_find_leaf(struct inode *ip, const char *name,
63 		    int namelen, uint32_t *hash, uint8_t *hash_verion,
64 		    struct ext2fs_htree_lookup_info *info);
65 static uint32_t ext2_htree_get_block(struct ext2fs_htree_entry *ep);
66 static uint16_t	ext2_htree_get_count(struct ext2fs_htree_entry *ep);
67 static uint32_t ext2_htree_get_hash(struct ext2fs_htree_entry *ep);
68 static uint16_t	ext2_htree_get_limit(struct ext2fs_htree_entry *ep);
69 static void	ext2_htree_insert_entry_to_level(struct ext2fs_htree_lookup_level *level,
70 		    uint32_t hash, uint32_t blk);
71 static void	ext2_htree_insert_entry(struct ext2fs_htree_lookup_info *info,
72 		    uint32_t hash, uint32_t blk);
73 static uint32_t	ext2_htree_node_limit(struct inode *ip);
74 static void	ext2_htree_set_block(struct ext2fs_htree_entry *ep,
75 		    uint32_t blk);
76 static void	ext2_htree_set_count(struct ext2fs_htree_entry *ep,
77 		    uint16_t cnt);
78 static void	ext2_htree_set_hash(struct ext2fs_htree_entry *ep,
79 		    uint32_t hash);
80 static void	ext2_htree_set_limit(struct ext2fs_htree_entry *ep,
81 		    uint16_t limit);
82 static int	ext2_htree_split_dirblock(char *block1, char *block2,
83 		    uint32_t blksize, uint32_t *hash_seed, uint8_t hash_version,
84 		    uint32_t *split_hash, struct  ext2fs_direct_2 *entry);
85 static void	ext2_htree_release(struct ext2fs_htree_lookup_info *info);
86 static uint32_t	ext2_htree_root_limit(struct inode *ip, int len);
87 static int	ext2_htree_writebuf(struct ext2fs_htree_lookup_info *info);
88 
89 int
90 ext2_htree_has_idx(struct inode *ip)
91 {
92 #ifdef EXT2FS_HTREE
93 	if (EXT2_HAS_COMPAT_FEATURE(ip->i_e2fs, EXT2F_COMPAT_DIRHASHINDEX) &&
94 	    ip->i_flags & EXT4_INDEX)
95 		return (1);
96 	else
97 #endif
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 	int i;
197 
198 	for (i = 0; i < info->h_levels_num; i++) {
199 		struct buf *bp = info->h_levels[i].h_bp;
200 		if (bp != NULL)
201 			brelse(bp);
202 	}
203 }
204 
205 static uint32_t
206 ext2_htree_root_limit(struct inode *ip, int len)
207 {
208 	uint32_t space;
209 
210 	space = ip->i_e2fs->e2fs_bsize - EXT2_DIR_REC_LEN(1) -
211 	    EXT2_DIR_REC_LEN(2) - len;
212 	return (space / sizeof(struct ext2fs_htree_entry));
213 }
214 
215 static uint32_t
216 ext2_htree_node_limit(struct inode *ip)
217 {
218 	struct m_ext2fs *fs;
219 	uint32_t space;
220 
221 	fs = ip->i_e2fs;
222 	space = fs->e2fs_bsize - EXT2_DIR_REC_LEN(0);
223 
224 	return (space / sizeof(struct ext2fs_htree_entry));
225 }
226 
227 static int
228 ext2_htree_find_leaf(struct inode *ip, const char *name, int namelen,
229 		     uint32_t *hash, uint8_t *hash_ver,
230 		     struct ext2fs_htree_lookup_info *info)
231 {
232 	struct vnode *vp;
233 	struct ext2fs *fs;
234 	struct m_ext2fs *m_fs;
235 	struct buf *bp = NULL;
236 	struct ext2fs_htree_root *rootp;
237 	struct ext2fs_htree_entry *entp, *start, *end, *middle, *found;
238 	struct ext2fs_htree_lookup_level *level_info;
239 	uint32_t hash_major = 0, hash_minor = 0;
240 	uint32_t levels, cnt;
241 	uint8_t hash_version;
242 
243 	if (name == NULL || info == NULL)
244 		return (-1);
245 
246 	vp = ITOV(ip);
247 	fs = ip->i_e2fs->e2fs;
248 	m_fs = ip->i_e2fs;
249 
250 	if (ext2_blkatoff(vp, 0, NULL, &bp) != 0)
251 		return (-1);
252 
253 	info->h_levels_num = 1;
254 	info->h_levels[0].h_bp = bp;
255 	rootp = (struct ext2fs_htree_root *)bp->b_data;
256 	if (rootp->h_info.h_hash_version != EXT2_HTREE_LEGACY &&
257 	    rootp->h_info.h_hash_version != EXT2_HTREE_HALF_MD4 &&
258 	    rootp->h_info.h_hash_version != EXT2_HTREE_TEA)
259 		goto error;
260 
261 	hash_version = rootp->h_info.h_hash_version;
262 	if (hash_version <= EXT2_HTREE_TEA)
263 		hash_version += m_fs->e2fs_uhash;
264 	*hash_ver = hash_version;
265 
266 	ext2_htree_hash(name, namelen, fs->e3fs_hash_seed,
267 	    hash_version, &hash_major, &hash_minor);
268 	*hash = hash_major;
269 
270 	if ((levels = rootp->h_info.h_ind_levels) > 1)
271 		goto error;
272 
273 	entp = (struct ext2fs_htree_entry *)(((char *)&rootp->h_info) +
274 	    rootp->h_info.h_info_len);
275 
276 	if (ext2_htree_get_limit(entp) !=
277 	    ext2_htree_root_limit(ip, rootp->h_info.h_info_len))
278 		goto error;
279 
280 	while (1) {
281 		cnt = ext2_htree_get_count(entp);
282 		if (cnt == 0 || cnt > ext2_htree_get_limit(entp))
283 			goto error;
284 
285 		start = entp + 1;
286 		end = entp + cnt - 1;
287 		while (start <= end) {
288 			middle = start + (end - start) / 2;
289 			if (ext2_htree_get_hash(middle) > hash_major)
290 				end = middle - 1;
291 			else
292 				start = middle + 1;
293 		}
294 		found = start - 1;
295 
296 		level_info = &(info->h_levels[info->h_levels_num - 1]);
297 		level_info->h_bp = bp;
298 		level_info->h_entries = entp;
299 		level_info->h_entry = found;
300 		if (levels == 0)
301 			return (0);
302 		levels--;
303 		if (ext2_blkatoff(vp,
304 		    ext2_htree_get_block(found) * m_fs->e2fs_bsize,
305 		    NULL, &bp) != 0)
306 			goto error;
307 		entp = ((struct ext2fs_htree_node *)bp->b_data)->h_entries;
308 		info->h_levels_num++;
309 		info->h_levels[info->h_levels_num - 1].h_bp = bp;
310 	}
311 
312 error:
313 	ext2_htree_release(info);
314 	return (-1);
315 }
316 
317 /*
318  * Try to lookup a directory entry in HTree index
319  */
320 int
321 ext2_htree_lookup(struct inode *ip, const char *name, int namelen,
322 		  struct buf **bpp, int *entryoffp, doff_t *offp,
323 		  doff_t *prevoffp, doff_t *endusefulp,
324 		  struct ext2fs_searchslot *ss)
325 {
326 	struct vnode *vp;
327 	struct ext2fs_htree_lookup_info info;
328 	struct ext2fs_htree_entry *leaf_node;
329 	struct m_ext2fs *m_fs;
330 	struct buf *bp;
331 	uint32_t blk;
332 	uint32_t dirhash;
333 	uint32_t bsize;
334 	uint8_t hash_version;
335 	int search_next;
336 	int found = 0;
337 
338 	m_fs = ip->i_e2fs;
339 	bsize = m_fs->e2fs_bsize;
340 	vp = ITOV(ip);
341 
342 	/* TODO: print error msg because we don't lookup '.' and '..' */
343 
344 	memset(&info, 0, sizeof(info));
345 	if (ext2_htree_find_leaf(ip, name, namelen, &dirhash,
346 	    &hash_version, &info))
347 		return (-1);
348 
349 	do {
350 		leaf_node = info.h_levels[info.h_levels_num - 1].h_entry;
351 		blk = ext2_htree_get_block(leaf_node);
352 		if (ext2_blkatoff(vp, blk * bsize, NULL, &bp) != 0) {
353 			ext2_htree_release(&info);
354 			return (-1);
355 		}
356 
357 		*offp = blk * bsize;
358 		*entryoffp = 0;
359 		*prevoffp = blk * bsize;
360 		*endusefulp = blk * bsize;
361 
362 		if (ss->slotstatus == NONE) {
363 			ss->slotoffset = -1;
364 			ss->slotfreespace = 0;
365 		}
366 
367 		if (ext2_search_dirblock(ip, bp->b_data, &found,
368 		    name, namelen, entryoffp, offp, prevoffp,
369 		    endusefulp, ss) != 0) {
370 			brelse(bp);
371 			ext2_htree_release(&info);
372 			return (-1);
373 		}
374 
375 		if (found) {
376 			*bpp = bp;
377 			ext2_htree_release(&info);
378 			return (0);
379 		}
380 
381 		brelse(bp);
382 		search_next = ext2_htree_check_next(ip, dirhash, name, &info);
383 	} while (search_next);
384 
385 	ext2_htree_release(&info);
386 	return (ENOENT);
387 }
388 
389 static int
390 ext2_htree_append_block(struct vnode *vp, char *data,
391 			struct componentname *cnp, uint32_t blksize)
392 {
393 	struct iovec aiov;
394 	struct uio auio;
395 	struct inode *dp = VTOI(vp);
396 	uint64_t cursize, newsize;
397 	int error;
398 
399 	cursize = roundup(dp->i_size, blksize);
400 	newsize = roundup(dp->i_size, blksize) + blksize;
401 
402 	auio.uio_offset = cursize;
403 	auio.uio_resid = blksize;
404 	aiov.iov_len = blksize;
405 	aiov.iov_base = data;
406 	auio.uio_iov = &aiov;
407 	auio.uio_iovcnt = 1;
408 	auio.uio_rw = UIO_WRITE;
409 	auio.uio_segflg = UIO_SYSSPACE;
410 	error = VOP_WRITE(vp, &auio, IO_SYNC, cnp->cn_cred);
411 	if (!error)
412 		dp->i_size = newsize;
413 
414 	return (error);
415 }
416 
417 static int
418 ext2_htree_writebuf(struct ext2fs_htree_lookup_info *info)
419 {
420 	int i, error;
421 
422 	for (i = 0; i < info->h_levels_num; i++) {
423 		struct buf *bp = info->h_levels[i].h_bp;
424 		error = bwrite(bp);
425 		if (error)
426 			return (error);
427 	}
428 
429 	return (0);
430 }
431 
432 static void
433 ext2_htree_insert_entry_to_level(struct ext2fs_htree_lookup_level *level,
434 				 uint32_t hash, uint32_t blk)
435 {
436 	struct ext2fs_htree_entry *target;
437 	int entries_num;
438 
439 	target = level->h_entry + 1;
440 	entries_num = ext2_htree_get_count(level->h_entries);
441 
442 	memmove(target + 1, target, (char *)(level->h_entries + entries_num) -
443 	    (char *)target);
444 	ext2_htree_set_block(target, blk);
445 	ext2_htree_set_hash(target, hash);
446 	ext2_htree_set_count(level->h_entries, entries_num + 1);
447 }
448 
449 /*
450  * Insert an index entry to the index node.
451  */
452 static void
453 ext2_htree_insert_entry(struct ext2fs_htree_lookup_info *info,
454 			uint32_t hash, uint32_t blk)
455 {
456 	struct ext2fs_htree_lookup_level *level;
457 
458 	level = &info->h_levels[info->h_levels_num - 1];
459 	ext2_htree_insert_entry_to_level(level, hash, blk);
460 }
461 
462 /*
463  * Compare two entry sort descriptors by name hash value.
464  * This is used together with qsort.
465  */
466 static int
467 ext2_htree_cmp_sort_entry(const void *e1, const void *e2)
468 {
469 	const struct ext2fs_htree_sort_entry *entry1, *entry2;
470 
471 	entry1 = (const struct ext2fs_htree_sort_entry *)e1;
472 	entry2 = (const struct ext2fs_htree_sort_entry *)e2;
473 
474 	if (entry1->h_hash < entry2->h_hash)
475 		return (-1);
476 	if (entry2->h_hash > entry2->h_hash)
477 		return (1);
478 	return (0);
479 }
480 
481 /*
482  * Append an entry to the end of the directory block.
483  */
484 static void
485 ext2_append_entry(char *block, uint32_t blksize,
486 		  struct ext2fs_direct_2 *last_entry,
487 		  struct ext2fs_direct_2 *new_entry)
488 {
489 	uint16_t entry_len;
490 
491 	entry_len = EXT2_DIR_REC_LEN(last_entry->e2d_namlen);
492 	last_entry->e2d_reclen = entry_len;
493 	last_entry = (struct ext2fs_direct_2 *)((char *)last_entry + entry_len);
494 	new_entry->e2d_reclen = block + blksize - (char *)last_entry;
495 	memcpy(last_entry, new_entry, EXT2_DIR_REC_LEN(new_entry->e2d_namlen));
496 }
497 
498 /*
499  * Move half of entries from the old directory block to the new one.
500  */
501 static int
502 ext2_htree_split_dirblock(char *block1, char *block2, uint32_t blksize,
503 			  uint32_t *hash_seed, uint8_t hash_version,
504 			  uint32_t *split_hash, struct ext2fs_direct_2 *entry)
505 {
506 	int entry_cnt = 0;
507 	int size = 0;
508 	int i, k;
509 	uint32_t offset;
510 	uint16_t entry_len = 0;
511 	uint32_t entry_hash;
512 	struct ext2fs_direct_2 *ep, *last;
513 	char *dest;
514 	struct ext2fs_htree_sort_entry *sort_info;
515 
516 	ep = (struct ext2fs_direct_2 *)block1;
517 	dest = block2;
518 	sort_info = (struct ext2fs_htree_sort_entry *)
519 	    ((char *)block2 + blksize);
520 
521 	/*
522 	 * Calculate name hash value for the entry which is to be added.
523 	 */
524 	ext2_htree_hash(entry->e2d_name, entry->e2d_namlen, hash_seed,
525 	    hash_version, &entry_hash, NULL);
526 
527 	/*
528 	 * Fill in directory entry sort descriptors.
529 	 */
530 	while ((char *)ep < block1 + blksize) {
531 		if (ep->e2d_ino && ep->e2d_namlen) {
532 			entry_cnt++;
533 			sort_info--;
534 			sort_info->h_size = ep->e2d_reclen;
535 			sort_info->h_offset = (char *)ep - block1;
536 			ext2_htree_hash(ep->e2d_name, ep->e2d_namlen,
537 			    hash_seed, hash_version,
538 			    &sort_info->h_hash, NULL);
539 		}
540 		ep = (struct ext2fs_direct_2 *)
541 		    ((char *)ep + ep->e2d_reclen);
542 	}
543 
544 	/*
545 	 * Sort directory entry descriptors by name hash value.
546 	 */
547 	qsort(sort_info, entry_cnt, sizeof(struct ext2fs_htree_sort_entry),
548 	    ext2_htree_cmp_sort_entry);
549 
550 	/*
551 	 * Count the number of entries to move to directory block 2.
552 	 */
553 	for (i = entry_cnt - 1; i >= 0; i--) {
554 		if (sort_info[i].h_size + size > blksize / 2)
555 			break;
556 		size += sort_info[i].h_size;
557 	}
558 
559 	*split_hash = sort_info[i + 1].h_hash;
560 
561 	/*
562 	 * Set collision bit.
563 	 */
564 	if (*split_hash == sort_info[i].h_hash)
565 		*split_hash += 1;
566 
567 	/*
568 	 * Move half of directory entries from block 1 to block 2.
569 	 */
570 	for (k = i + 1; k < entry_cnt; k++) {
571 		ep = (struct ext2fs_direct_2 *)((char *)block1 +
572 		    sort_info[k].h_offset);
573 		entry_len = EXT2_DIR_REC_LEN(ep->e2d_namlen);
574 		memcpy(dest, ep, entry_len);
575 		((struct ext2fs_direct_2 *)dest)->e2d_reclen = entry_len;
576 		/* Mark directory entry as unused. */
577 		ep->e2d_ino = 0;
578 		dest += entry_len;
579 	}
580 	dest -= entry_len;
581 
582 	/* Shrink directory entries in block 1. */
583 	last = (struct ext2fs_direct_2 *)block1;
584 	entry_len = EXT2_DIR_REC_LEN(last->e2d_namlen);
585 	for (offset = last->e2d_reclen; offset < blksize; ) {
586 		ep = (struct ext2fs_direct_2 *)(block1 + offset);
587 		offset += ep->e2d_reclen;
588 		if (last->e2d_ino) {
589 			/* Trim the existing slot */
590 			last->e2d_reclen = entry_len;
591 			last = (struct ext2fs_direct_2 *)
592 			   ((char *)last + entry_len);
593 		}
594 		entry_len = EXT2_DIR_REC_LEN(ep->e2d_namlen);
595 		memcpy((void *)last, (void *)ep, entry_len);
596 	}
597 
598 	if (entry_hash >= *split_hash) {
599 		/* Add entry to block 2. */
600 		ext2_append_entry(block2, blksize,
601 		    (struct ext2fs_direct_2 *)dest, entry);
602 
603 		/* Adjust length field of last entry of block 1. */
604 		last->e2d_reclen = block1 + blksize - (char *)last;
605 	} else {
606 		/* Add entry to block 1. */
607 		ext2_append_entry(block1, blksize, last, entry);
608 
609 		/* Adjust length field of last entry of block 2. */
610 		((struct ext2fs_direct_2 *)dest)->e2d_reclen =
611 		    block2 + blksize - dest;
612 	}
613 
614 	return (0);
615 }
616 
617 /*
618  * Create an HTree index for a directory
619  */
620 int
621 ext2_htree_create_index(struct vnode *vp, struct componentname *cnp,
622 			struct ext2fs_direct_2 *new_entry)
623 {
624 	struct buf *bp = NULL;
625 	struct inode *dp;
626 	struct ext2fs *fs;
627 	struct m_ext2fs *m_fs;
628 	struct ext2fs_direct_2 *ep, *dotdot;
629 	struct ext2fs_htree_root *root;
630 	struct ext2fs_htree_lookup_info info;
631 	uint32_t blksize, dirlen, split_hash;
632 	uint8_t hash_version;
633 	char *buf1 = NULL;
634 	char *buf2 = NULL;
635 	int error = 0;
636 
637 	dp = VTOI(vp);
638 	fs = dp->i_e2fs->e2fs;
639 	m_fs = dp->i_e2fs;
640 	blksize = m_fs->e2fs_bsize;
641 
642 	buf1 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
643 	buf2 = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
644 
645 	if ((error = ext2_blkatoff(vp, 0, NULL, &bp)) != 0)
646 		goto out;
647 
648 	root = (struct ext2fs_htree_root *)bp->b_data;
649 	dotdot = (struct ext2fs_direct_2 *)((char *)&(root->h_dotdot));
650 	ep = (struct ext2fs_direct_2 *)((char *)dotdot + dotdot->e2d_reclen);
651 	dirlen = (char *)root + blksize - (char *)ep;
652 	memcpy(buf1, ep, dirlen);
653 	ep = (struct ext2fs_direct_2 *)buf1;
654 	while ((char *)ep < buf1 + dirlen)
655 		ep = (struct ext2fs_direct_2 *)
656 		    ((char *)ep + ep->e2d_reclen);
657 	ep->e2d_reclen = buf1 + blksize - (char *)ep;
658 
659 	dp->i_flags |= EXT4_INDEX;
660 
661 	/*
662 	 * Initialize index root.
663 	 */
664 	dotdot->e2d_reclen = blksize - EXT2_DIR_REC_LEN(1);
665 	memset(&root->h_info, 0, sizeof(root->h_info));
666 	root->h_info.h_hash_version = fs->e3fs_def_hash_version;
667 	root->h_info.h_info_len = sizeof(root->h_info);
668 	ext2_htree_set_block(root->h_entries, 1);
669 	ext2_htree_set_count(root->h_entries, 1);
670 	ext2_htree_set_limit(root->h_entries,
671 	    ext2_htree_root_limit(dp, sizeof(root->h_info)));
672 
673 	memset(&info, 0, sizeof(info));
674 	info.h_levels_num = 1;
675 	info.h_levels[0].h_entries = root->h_entries;
676 	info.h_levels[0].h_entry = root->h_entries;
677 
678 	hash_version = root->h_info.h_hash_version;
679 	if (hash_version <= EXT2_HTREE_TEA)
680 		hash_version += m_fs->e2fs_uhash;
681 	ext2_htree_split_dirblock(buf1, buf2, blksize, fs->e3fs_hash_seed,
682 	    hash_version, &split_hash, new_entry);
683 	ext2_htree_insert_entry(&info, split_hash, 2);
684 
685 	/*
686 	 * Write directory block 0.
687 	 */
688 	if (DOINGASYNC(vp)) {
689 		bdwrite(bp);
690 		error = 0;
691 	} else {
692 		error = bwrite(bp);
693 	}
694 	dp->i_flag |= IN_CHANGE | IN_UPDATE;
695 	if (error)
696 		goto out;
697 
698 	/*
699 	 * Write directory block 1.
700 	 */
701 	error = ext2_htree_append_block(vp, buf1, cnp, blksize);
702 	if (error)
703 		goto out1;
704 
705 	/*
706 	 * Write directory block 2.
707 	 */
708 	error = ext2_htree_append_block(vp, buf2, cnp, blksize);
709 
710 	free(buf1, M_TEMP);
711 	free(buf2, M_TEMP);
712 	return (error);
713 out:
714 	if (bp != NULL)
715 		brelse(bp);
716 out1:
717 	free(buf1, M_TEMP);
718 	free(buf2, M_TEMP);
719 	return (error);
720 }
721 
722 /*
723  * Add an entry to the directory using htree index.
724  */
725 int
726 ext2_htree_add_entry(struct vnode *dvp, struct ext2fs_direct_2 *entry,
727 		     struct componentname *cnp)
728 {
729 	struct ext2fs_htree_entry *entries, *leaf_node;
730 	struct ext2fs_htree_lookup_info info;
731 	struct buf *bp = NULL;
732 	struct ext2fs *fs;
733 	struct m_ext2fs *m_fs;
734 	struct inode *ip;
735 	uint16_t ent_num;
736 	uint32_t dirhash, split_hash;
737 	uint32_t blksize, blknum;
738 	uint64_t cursize, dirsize;
739 	uint8_t hash_version;
740 	char *newdirblock = NULL;
741 	char *newidxblock = NULL;
742 	struct ext2fs_htree_node *dst_node;
743 	struct ext2fs_htree_entry *dst_entries;
744 	struct ext2fs_htree_entry *root_entires;
745 	struct buf *dst_bp = NULL;
746 	int error, write_bp = 0, write_dst_bp = 0, write_info = 0;
747 
748 	ip = VTOI(dvp);
749 	m_fs = ip->i_e2fs;
750 	fs = m_fs->e2fs;
751 	blksize = m_fs->e2fs_bsize;
752 
753 	if (ip->i_count != 0)
754 		return ext2_add_entry(dvp, entry);
755 
756 	/* Target directory block is full, split it */
757 	memset(&info, 0, sizeof(info));
758 	error = ext2_htree_find_leaf(ip, entry->e2d_name, entry->e2d_namlen,
759 	    &dirhash, &hash_version, &info);
760 	if (error)
761 		return (error);
762 
763 	entries = info.h_levels[info.h_levels_num - 1].h_entries;
764 	ent_num = ext2_htree_get_count(entries);
765 	if (ent_num == ext2_htree_get_limit(entries)) {
766 		/* Split the index node. */
767 		root_entires = info.h_levels[0].h_entries;
768 		newidxblock = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
769 		dst_node = (struct ext2fs_htree_node *)newidxblock;
770 		dst_entries = dst_node->h_entries;
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 = roundup(ip->i_size, blksize) + 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 				info.h_levels[1].h_bp = dst_bp;
814 				dst_bp = tmp;
815 
816 				info.h_levels[1].h_entry =
817 				    info.h_levels[1].h_entry -
818 				    (entries + src_ent_num) +
819 				    dst_entries;
820 				info.h_levels[1].h_entries = dst_entries;
821 			}
822 			ext2_htree_insert_entry_to_level(&info.h_levels[0],
823 			    split_hash, blknum);
824 
825 			/* Write new index node to disk */
826 			error = bwrite(dst_bp);
827 			ip->i_flag |= IN_CHANGE | IN_UPDATE;
828 			if (error)
829 				goto finish;
830 			write_dst_bp = 1;
831 		} else {
832 			/* Create second level for htree index */
833 			struct ext2fs_htree_root *idx_root;
834 
835 			memcpy(dst_entries, entries,
836 			    ent_num * sizeof(struct ext2fs_htree_entry));
837 			ext2_htree_set_limit(dst_entries,
838 			    ext2_htree_node_limit(ip));
839 
840 			idx_root = (struct ext2fs_htree_root *)
841 			    info.h_levels[0].h_bp->b_data;
842 			idx_root->h_info.h_ind_levels = 1;
843 
844 			ext2_htree_set_count(entries, 1);
845 			ext2_htree_set_block(entries, blknum);
846 
847 			info.h_levels_num = 2;
848 			info.h_levels[1].h_entries = dst_entries;
849 			info.h_levels[1].h_entry = info.h_levels[0].h_entry -
850 			    info.h_levels[0].h_entries + dst_entries;
851 			info.h_levels[1].h_bp = dst_bp;
852 		}
853 	}
854 
855 	leaf_node = info.h_levels[info.h_levels_num - 1].h_entry;
856 	blknum = ext2_htree_get_block(leaf_node);
857 	error = ext2_blkatoff(dvp, blknum * blksize, NULL, &bp);
858 	if (error)
859 		goto finish;
860 
861 	/* Split target directory block */
862 	newdirblock = malloc(blksize, M_TEMP, M_WAITOK | M_ZERO);
863 	ext2_htree_split_dirblock((char *)bp->b_data, newdirblock, blksize,
864 	    fs->e3fs_hash_seed, hash_version, &split_hash, entry);
865 	cursize = roundup(ip->i_size, blksize);
866 	dirsize = roundup(ip->i_size, blksize) + blksize;
867 	blknum = dirsize / blksize - 1;
868 
869 	/* Add index entry for the new directory block */
870 	ext2_htree_insert_entry(&info, split_hash, blknum);
871 
872 	/* Write the new directory block to the end of the directory */
873 	error = ext2_htree_append_block(dvp, newdirblock, cnp, blksize);
874 	if (error)
875 		goto finish;
876 
877 	/* Write the target directory block */
878 	error = bwrite(bp);
879 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
880 	if (error)
881 		goto finish;
882 	write_bp = 1;
883 
884 	/* Write the index block */
885 	error = ext2_htree_writebuf(&info);
886 	if (!error)
887 		write_info = 1;
888 
889 finish:
890 	if (dst_bp != NULL && !write_dst_bp)
891 		brelse(dst_bp);
892 	if (bp != NULL && !write_bp)
893 		brelse(bp);
894 	if (newdirblock != NULL)
895 		free(newdirblock, M_TEMP);
896 	if (newidxblock != NULL)
897 		free(newidxblock, M_TEMP);
898 	if (!write_info)
899 		ext2_htree_release(&info);
900 	return (error);
901 }
902