xref: /freebsd/sys/fs/ext2fs/ext2_extents.c (revision d8a0fe102c0cfdfcd5b818f850eff09d8536c9bc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2010 Zheng Liu <lz@freebsd.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/types.h>
34 #include <sys/kernel.h>
35 #include <sys/malloc.h>
36 #include <sys/vnode.h>
37 #include <sys/bio.h>
38 #include <sys/buf.h>
39 #include <sys/conf.h>
40 #include <sys/stat.h>
41 
42 #include <fs/ext2fs/ext2_mount.h>
43 #include <fs/ext2fs/fs.h>
44 #include <fs/ext2fs/inode.h>
45 #include <fs/ext2fs/ext2fs.h>
46 #include <fs/ext2fs/ext2_extents.h>
47 #include <fs/ext2fs/ext2_extern.h>
48 
49 static MALLOC_DEFINE(M_EXT2EXTENTS, "ext2_extents", "EXT2 extents");
50 
51 #ifdef EXT2FS_DEBUG
52 static void
53 ext4_ext_print_extent(struct ext4_extent *ep)
54 {
55 
56 	printf("    ext %p => (blk %u len %u start %lu)\n",
57 	    ep, ep->e_blk, ep->e_len,
58 	    (uint64_t)ep->e_start_hi << 32 | ep->e_start_lo);
59 }
60 
61 static void ext4_ext_print_header(struct inode *ip, struct ext4_extent_header *ehp);
62 
63 static void
64 ext4_ext_print_index(struct inode *ip, struct ext4_extent_index *ex, int do_walk)
65 {
66 	struct m_ext2fs *fs;
67 	struct buf *bp;
68 	int error;
69 
70 	fs = ip->i_e2fs;
71 
72 	printf("    index %p => (blk %u pblk %lu)\n",
73 	    ex, ex->ei_blk, (uint64_t)ex->ei_leaf_hi << 32 | ex->ei_leaf_lo);
74 
75 	if(!do_walk)
76 		return;
77 
78 	if ((error = bread(ip->i_devvp,
79 	    fsbtodb(fs, ((uint64_t)ex->ei_leaf_hi << 32 | ex->ei_leaf_lo)),
80 	    (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
81 		brelse(bp);
82 		return;
83 	}
84 
85 	ext4_ext_print_header(ip, (struct ext4_extent_header *)bp->b_data);
86 
87 	brelse(bp);
88 
89 }
90 
91 static void
92 ext4_ext_print_header(struct inode *ip, struct ext4_extent_header *ehp)
93 {
94 	int i;
95 
96 	printf("header %p => (magic 0x%x entries %d max %d depth %d gen %d)\n",
97 	    ehp, ehp->eh_magic, ehp->eh_ecount, ehp->eh_max, ehp->eh_depth,
98 	    ehp->eh_gen);
99 
100 	for (i = 0; i < ehp->eh_ecount; i++)
101 		if (ehp->eh_depth != 0)
102 			ext4_ext_print_index(ip,
103 			    (struct ext4_extent_index *)(ehp + 1 + i), 1);
104 		else
105 			ext4_ext_print_extent((struct ext4_extent *)(ehp + 1 + i));
106 }
107 
108 static void
109 ext4_ext_print_path(struct inode *ip, struct ext4_extent_path *path)
110 {
111 	int k, l;
112 
113 	l = path->ep_depth
114 
115 	printf("ip=%d, Path:\n", ip->i_number);
116 	for (k = 0; k <= l; k++, path++) {
117 		if (path->ep_index) {
118 			ext4_ext_print_index(ip, path->ep_index, 0);
119 		} else if (path->ep_ext) {
120 			ext4_ext_print_extent(path->ep_ext);
121 		}
122 	}
123 }
124 
125 void
126 ext4_ext_print_extent_tree_status(struct inode * ip)
127 {
128 	struct m_ext2fs *fs;
129 	struct ext4_extent_header *ehp;
130 
131 	fs = ip->i_e2fs;
132 	ehp = (struct ext4_extent_header *)(char *)ip->i_db;
133 
134 	printf("Extent status:ip=%d\n", ip->i_number);
135 	if (!(ip->i_flag & IN_E4EXTENTS))
136 		return;
137 
138 	ext4_ext_print_header(ip, ehp);
139 
140 	return;
141 }
142 #endif
143 
144 static inline struct ext4_extent_header *
145 ext4_ext_inode_header(struct inode *ip)
146 {
147 
148 	return ((struct ext4_extent_header *)ip->i_db);
149 }
150 
151 static inline struct ext4_extent_header *
152 ext4_ext_block_header(char *bdata)
153 {
154 
155 	return ((struct ext4_extent_header *)bdata);
156 }
157 
158 static inline unsigned short
159 ext4_ext_inode_depth(struct inode *ip)
160 {
161 	struct ext4_extent_header *ehp;
162 
163 	ehp = (struct ext4_extent_header *)ip->i_data;
164 	return (ehp->eh_depth);
165 }
166 
167 static inline e4fs_daddr_t
168 ext4_ext_index_pblock(struct ext4_extent_index *index)
169 {
170 	e4fs_daddr_t blk;
171 
172 	blk = index->ei_leaf_lo;
173 	blk |= (e4fs_daddr_t)index->ei_leaf_hi << 32;
174 
175 	return (blk);
176 }
177 
178 static inline void
179 ext4_index_store_pblock(struct ext4_extent_index *index, e4fs_daddr_t pb)
180 {
181 
182 	index->ei_leaf_lo = pb & 0xffffffff;
183 	index->ei_leaf_hi = (pb >> 32) & 0xffff;
184 }
185 
186 
187 static inline e4fs_daddr_t
188 ext4_ext_extent_pblock(struct ext4_extent *extent)
189 {
190 	e4fs_daddr_t blk;
191 
192 	blk = extent->e_start_lo;
193 	blk |= (e4fs_daddr_t)extent->e_start_hi << 32;
194 
195 	return (blk);
196 }
197 
198 static inline void
199 ext4_ext_store_pblock(struct ext4_extent *ex, e4fs_daddr_t pb)
200 {
201 
202 	ex->e_start_lo = pb & 0xffffffff;
203 	ex->e_start_hi = (pb >> 32) & 0xffff;
204 }
205 
206 int
207 ext4_ext_in_cache(struct inode *ip, daddr_t lbn, struct ext4_extent *ep)
208 {
209 	struct ext4_extent_cache *ecp;
210 	int ret = EXT4_EXT_CACHE_NO;
211 
212 	ecp = &ip->i_ext_cache;
213 	if (ecp->ec_type == EXT4_EXT_CACHE_NO)
214 		return (ret);
215 
216 	if (lbn >= ecp->ec_blk && lbn < ecp->ec_blk + ecp->ec_len) {
217 		ep->e_blk = ecp->ec_blk;
218 		ep->e_start_lo = ecp->ec_start & 0xffffffff;
219 		ep->e_start_hi = ecp->ec_start >> 32 & 0xffff;
220 		ep->e_len = ecp->ec_len;
221 		ret = ecp->ec_type;
222 	}
223 	return (ret);
224 }
225 
226 static int
227 ext4_ext_check_header(struct inode *ip, struct ext4_extent_header *eh)
228 {
229 	struct m_ext2fs *fs;
230 	char *error_msg;
231 
232 	fs = ip->i_e2fs;
233 
234 	if (eh->eh_magic != EXT4_EXT_MAGIC) {
235 		error_msg = "invalid magic";
236 		goto corrupted;
237 	}
238 	if (eh->eh_max == 0) {
239 		error_msg = "invalid eh_max";
240 		goto corrupted;
241 	}
242 	if (eh->eh_ecount > eh->eh_max) {
243 		error_msg = "invalid eh_entries";
244 		goto corrupted;
245 	}
246 
247 	return (0);
248 
249 corrupted:
250 	ext2_fserr(fs, ip->i_uid, error_msg);
251 	return (EIO);
252 }
253 
254 static void
255 ext4_ext_binsearch_index(struct ext4_extent_path *path, int blk)
256 {
257 	struct ext4_extent_header *eh;
258 	struct ext4_extent_index *r, *l, *m;
259 
260 	eh = path->ep_header;
261 
262 	KASSERT(eh->eh_ecount <= eh->eh_max && eh->eh_ecount > 0,
263 	    ("ext4_ext_binsearch_index: bad args"));
264 
265 	l = EXT_FIRST_INDEX(eh) + 1;
266 	r = EXT_FIRST_INDEX(eh) + eh->eh_ecount - 1;
267 	while (l <= r) {
268 		m = l + (r - l) / 2;
269 		if (blk < m->ei_blk)
270 			r = m - 1;
271 		else
272 			l = m + 1;
273 	}
274 
275 	path->ep_index = l - 1;
276 }
277 
278 static void
279 ext4_ext_binsearch_ext(struct ext4_extent_path *path, int blk)
280 {
281 	struct ext4_extent_header *eh;
282 	struct ext4_extent *r, *l, *m;
283 
284 	eh = path->ep_header;
285 
286 	KASSERT(eh->eh_ecount <= eh->eh_max,
287 	    ("ext4_ext_binsearch_ext: bad args"));
288 
289 	if (eh->eh_ecount == 0)
290 		return;
291 
292 	l = EXT_FIRST_EXTENT(eh) + 1;
293 	r = EXT_FIRST_EXTENT(eh) + eh->eh_ecount - 1;
294 
295 	while (l <= r) {
296 		m = l + (r - l) / 2;
297 		if (blk < m->e_blk)
298 			r = m - 1;
299 		else
300 			l = m + 1;
301 	}
302 
303 	path->ep_ext = l - 1;
304 }
305 
306 static int
307 ext4_ext_fill_path_bdata(struct ext4_extent_path *path,
308     struct buf *bp, uint64_t blk)
309 {
310 
311 	KASSERT(path->ep_data == NULL,
312 	    ("ext4_ext_fill_path_bdata: bad ep_data"));
313 
314 	path->ep_data = malloc(bp->b_bufsize, M_EXT2EXTENTS, M_WAITOK);
315 	if (!path->ep_data)
316 		return (ENOMEM);
317 
318 	memcpy(path->ep_data, bp->b_data, bp->b_bufsize);
319 	path->ep_blk = blk;
320 
321 	return (0);
322 }
323 
324 static void
325 ext4_ext_fill_path_buf(struct ext4_extent_path *path, struct buf *bp)
326 {
327 
328 	KASSERT(path->ep_data != NULL,
329 	    ("ext4_ext_fill_path_buf: bad ep_data"));
330 
331 	memcpy(bp->b_data, path->ep_data, bp->b_bufsize);
332 }
333 
334 static void
335 ext4_ext_drop_refs(struct ext4_extent_path *path)
336 {
337 	int depth, i;
338 
339 	if (!path)
340 		return;
341 
342 	depth = path->ep_depth;
343 	for (i = 0; i <= depth; i++, path++)
344 		if (path->ep_data) {
345 			free(path->ep_data, M_EXT2EXTENTS);
346 			path->ep_data = NULL;
347 		}
348 }
349 
350 void
351 ext4_ext_path_free(struct ext4_extent_path *path)
352 {
353 
354 	if (!path)
355 		return;
356 
357 	ext4_ext_drop_refs(path);
358 	free(path, M_EXT2EXTENTS);
359 }
360 
361 int
362 ext4_ext_find_extent(struct inode *ip, daddr_t block,
363     struct ext4_extent_path **ppath)
364 {
365 	struct m_ext2fs *fs;
366 	struct ext4_extent_header *eh;
367 	struct ext4_extent_path *path;
368 	struct buf *bp;
369 	uint64_t blk;
370 	int error, depth, i, ppos, alloc;
371 
372 	fs = ip->i_e2fs;
373 	eh = ext4_ext_inode_header(ip);
374 	depth = ext4_ext_inode_depth(ip);
375 	ppos = 0;
376 	alloc = 0;
377 
378 	error = ext4_ext_check_header(ip, eh);
379 	if (error)
380 		return (error);
381 
382 	if (ppath == NULL)
383 		return (EINVAL);
384 
385 	path = *ppath;
386 	if (path == NULL) {
387 		path = malloc(EXT4_EXT_DEPTH_MAX *
388 		    sizeof(struct ext4_extent_path),
389 		    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
390 		if (!path)
391 			return (ENOMEM);
392 
393 		*ppath = path;
394 		alloc = 1;
395 	}
396 
397 	path[0].ep_header = eh;
398 	path[0].ep_data = NULL;
399 
400 	/* Walk through the tree. */
401 	i = depth;
402 	while (i) {
403 		ext4_ext_binsearch_index(&path[ppos], block);
404 		blk = ext4_ext_index_pblock(path[ppos].ep_index);
405 		path[ppos].ep_depth = i;
406 		path[ppos].ep_ext = NULL;
407 
408 		error = bread(ip->i_devvp, fsbtodb(ip->i_e2fs, blk),
409 		    ip->i_e2fs->e2fs_bsize, NOCRED, &bp);
410 		if (error) {
411 			brelse(bp);
412 			goto error;
413 		}
414 
415 		ppos++;
416 		if (ppos > depth) {
417 			ext2_fserr(fs, ip->i_uid,
418 			    "ppos > depth => extent corrupted");
419 			error = EIO;
420 			brelse(bp);
421 			goto error;
422 		}
423 
424 		ext4_ext_fill_path_bdata(&path[ppos], bp, blk);
425 		bqrelse(bp);
426 
427 		eh = ext4_ext_block_header(path[ppos].ep_data);
428 		error = ext4_ext_check_header(ip, eh);
429 		if (error)
430 			goto error;
431 
432 		path[ppos].ep_header = eh;
433 
434 		i--;
435 	}
436 
437 	error = ext4_ext_check_header(ip, eh);
438 	if (error)
439 		goto error;
440 
441 	/* Find extent. */
442 	path[ppos].ep_depth = i;
443 	path[ppos].ep_header = eh;
444 	path[ppos].ep_ext = NULL;
445 	path[ppos].ep_index = NULL;
446 	ext4_ext_binsearch_ext(&path[ppos], block);
447 	return (0);
448 
449 error:
450 	ext4_ext_drop_refs(path);
451 	if (alloc)
452 		free(path, M_EXT2EXTENTS);
453 
454 	*ppath = NULL;
455 
456 	return (error);
457 }
458 
459 static inline int
460 ext4_ext_space_root(struct inode *ip)
461 {
462 	int size;
463 
464 	size = sizeof(ip->i_data);
465 	size -= sizeof(struct ext4_extent_header);
466 	size /= sizeof(struct ext4_extent);
467 
468 	return (size);
469 }
470 
471 static inline int
472 ext4_ext_space_block(struct inode *ip)
473 {
474 	struct m_ext2fs *fs;
475 	int size;
476 
477 	fs = ip->i_e2fs;
478 
479 	size = (fs->e2fs_bsize - sizeof(struct ext4_extent_header)) /
480 	    sizeof(struct ext4_extent);
481 
482 	return (size);
483 }
484 
485 static inline int
486 ext4_ext_space_block_index(struct inode *ip)
487 {
488 	struct m_ext2fs *fs;
489 	int size;
490 
491 	fs = ip->i_e2fs;
492 
493 	size = (fs->e2fs_bsize - sizeof(struct ext4_extent_header)) /
494 	    sizeof(struct ext4_extent_index);
495 
496 	return (size);
497 }
498 
499 void
500 ext4_ext_tree_init(struct inode *ip)
501 {
502 	struct ext4_extent_header *ehp;
503 
504 	ip->i_flag |= IN_E4EXTENTS;
505 
506 	memset(ip->i_data, 0, EXT2_NDADDR + EXT2_NIADDR);
507 	ehp = (struct ext4_extent_header *)ip->i_data;
508 	ehp->eh_magic = EXT4_EXT_MAGIC;
509 	ehp->eh_max = ext4_ext_space_root(ip);
510 	ip->i_ext_cache.ec_type = EXT4_EXT_CACHE_NO;
511 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
512 	ext2_update(ip->i_vnode, 1);
513 }
514 
515 static inline void
516 ext4_ext_put_in_cache(struct inode *ip, uint32_t blk,
517 			uint32_t len, uint32_t start, int type)
518 {
519 
520 	KASSERT(len != 0, ("ext4_ext_put_in_cache: bad input"));
521 
522 	ip->i_ext_cache.ec_type = type;
523 	ip->i_ext_cache.ec_blk = blk;
524 	ip->i_ext_cache.ec_len = len;
525 	ip->i_ext_cache.ec_start = start;
526 }
527 
528 static e4fs_daddr_t
529 ext4_ext_blkpref(struct inode *ip, struct ext4_extent_path *path,
530     e4fs_daddr_t block)
531 {
532 	struct m_ext2fs *fs;
533 	struct ext4_extent *ex;
534 	e4fs_daddr_t bg_start;
535 	int depth;
536 
537 	fs = ip->i_e2fs;
538 
539 	if (path) {
540 		depth = path->ep_depth;
541 		ex = path[depth].ep_ext;
542 		if (ex) {
543 			e4fs_daddr_t pblk = ext4_ext_extent_pblock(ex);
544 			e2fs_daddr_t blk = ex->e_blk;
545 
546 			if (block > blk)
547 				return (pblk + (block - blk));
548 			else
549 				return (pblk - (blk - block));
550 		}
551 
552 		/* Try to get block from index itself. */
553 		if (path[depth].ep_data)
554 			return (path[depth].ep_blk);
555 	}
556 
557 	/* Use inode's group. */
558 	bg_start = (ip->i_block_group * EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) +
559 	    fs->e2fs->e2fs_first_dblock;
560 
561 	return (bg_start + block);
562 }
563 
564 static int inline
565 ext4_can_extents_be_merged(struct ext4_extent *ex1,
566     struct ext4_extent *ex2)
567 {
568 
569 	if (ex1->e_blk + ex1->e_len != ex2->e_blk)
570 		return (0);
571 
572 	if (ex1->e_len + ex2->e_len > EXT4_MAX_LEN)
573 		return (0);
574 
575 	if (ext4_ext_extent_pblock(ex1) + ex1->e_len ==
576 	    ext4_ext_extent_pblock(ex2))
577 		return (1);
578 
579 	return (0);
580 }
581 
582 static unsigned
583 ext4_ext_next_leaf_block(struct inode *ip, struct ext4_extent_path *path)
584 {
585 	int depth = path->ep_depth;
586 
587 	/* Empty tree */
588 	if (depth == 0)
589 		return (EXT4_MAX_BLOCKS);
590 
591 	/* Go to indexes. */
592 	depth--;
593 
594 	while (depth >= 0) {
595 		if (path[depth].ep_index !=
596 		    EXT_LAST_INDEX(path[depth].ep_header))
597 			return (path[depth].ep_index[1].ei_blk);
598 
599 		depth--;
600 	}
601 
602 	return (EXT4_MAX_BLOCKS);
603 }
604 
605 static int
606 ext4_ext_dirty(struct inode *ip, struct ext4_extent_path *path)
607 {
608 	struct m_ext2fs *fs;
609 	struct buf *bp;
610 	uint64_t blk;
611 	int error;
612 
613 	fs = ip->i_e2fs;
614 
615 	if (!path)
616 		return (EINVAL);
617 
618 	if (path->ep_data) {
619 		blk = path->ep_blk;
620 		bp = getblk(ip->i_devvp, fsbtodb(fs, blk),
621 		    fs->e2fs_bsize, 0, 0, 0);
622 		if (!bp)
623 			return (EIO);
624 		ext4_ext_fill_path_buf(path, bp);
625 		error = bwrite(bp);
626 	} else {
627 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
628 		error = ext2_update(ip->i_vnode, 1);
629 	}
630 
631 	return (error);
632 }
633 
634 static int
635 ext4_ext_insert_index(struct inode *ip, struct ext4_extent_path *path,
636     uint32_t lblk, e4fs_daddr_t blk)
637 {
638 	struct m_ext2fs *fs;
639 	struct ext4_extent_index *idx;
640 	int len;
641 
642 	fs = ip->i_e2fs;
643 
644 	if (lblk == path->ep_index->ei_blk) {
645 		ext2_fserr(fs, ip->i_uid,
646 		    "lblk == index blk => extent corrupted");
647 		return (EIO);
648 	}
649 
650 	if (path->ep_header->eh_ecount >= path->ep_header->eh_max) {
651 		ext2_fserr(fs, ip->i_uid,
652 		    "ecout > maxcount => extent corrupted");
653 		return (EIO);
654 	}
655 
656 	if (lblk > path->ep_index->ei_blk) {
657 		/* Insert after. */
658 		idx = path->ep_index + 1;
659 	} else {
660 		/* Insert before. */
661 		idx = path->ep_index;
662 	}
663 
664 	len = EXT_LAST_INDEX(path->ep_header) - idx + 1;
665 	if (len > 0)
666 		memmove(idx + 1, idx, len * sizeof(struct ext4_extent_index));
667 
668 	if (idx > EXT_MAX_INDEX(path->ep_header)) {
669 		ext2_fserr(fs, ip->i_uid,
670 		    "index is out of range => extent corrupted");
671 		return (EIO);
672 	}
673 
674 	idx->ei_blk = lblk;
675 	ext4_index_store_pblock(idx, blk);
676 	path->ep_header->eh_ecount++;
677 
678 	return (ext4_ext_dirty(ip, path));
679 }
680 
681 static e4fs_daddr_t
682 ext4_ext_alloc_meta(struct inode *ip)
683 {
684 	e4fs_daddr_t blk = ext2_alloc_meta(ip);
685 	if (blk) {
686 		ip->i_blocks += btodb(ip->i_e2fs->e2fs_bsize);
687 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
688 		ext2_update(ip->i_vnode, 1);
689 	}
690 
691 	return (blk);
692 }
693 
694 static void
695 ext4_ext_blkfree(struct inode *ip, uint64_t blk, int count, int flags)
696 {
697 	struct m_ext2fs *fs;
698 	int i, blocksreleased;
699 
700 	fs = ip->i_e2fs;
701 	blocksreleased = count;
702 
703 	for(i = 0; i < count; i++)
704 		ext2_blkfree(ip, blk + i, fs->e2fs_bsize);
705 
706 	if (ip->i_blocks >= blocksreleased)
707 		ip->i_blocks -= (btodb(fs->e2fs_bsize)*blocksreleased);
708 	else
709 		ip->i_blocks = 0;
710 
711 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
712 	ext2_update(ip->i_vnode, 1);
713 }
714 
715 static int
716 ext4_ext_split(struct inode *ip, struct ext4_extent_path *path,
717     struct ext4_extent *newext, int at)
718 {
719 	struct m_ext2fs *fs;
720 	struct  buf *bp;
721 	int depth = ext4_ext_inode_depth(ip);
722 	struct ext4_extent_header *neh;
723 	struct ext4_extent_index *fidx;
724 	struct ext4_extent *ex;
725 	int i = at, k, m, a;
726 	e4fs_daddr_t newblk, oldblk;
727 	uint32_t border;
728 	e4fs_daddr_t *ablks = NULL;
729 	int error = 0;
730 
731 	fs = ip->i_e2fs;
732 	bp = NULL;
733 
734 	/*
735 	 * We will split at current extent for now.
736 	 */
737 	if (path[depth].ep_ext > EXT_MAX_EXTENT(path[depth].ep_header)) {
738 		ext2_fserr(fs, ip->i_uid,
739 		    "extent is out of range => extent corrupted");
740 		return (EIO);
741 	}
742 
743 	if (path[depth].ep_ext != EXT_MAX_EXTENT(path[depth].ep_header))
744 		border = path[depth].ep_ext[1].e_blk;
745 	else
746 		border = newext->e_blk;
747 
748 	/* Allocate new blocks. */
749 	ablks = malloc(sizeof(e4fs_daddr_t) * depth,
750 	    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
751 	if (!ablks)
752 		return (ENOMEM);
753 	for (a = 0; a < depth - at; a++) {
754 		newblk = ext4_ext_alloc_meta(ip);
755 		if (newblk == 0)
756 			goto cleanup;
757 		ablks[a] = newblk;
758 	}
759 
760 	newblk = ablks[--a];
761 	bp = getblk(ip->i_devvp, fsbtodb(fs, newblk), fs->e2fs_bsize, 0, 0, 0);
762 	if (!bp) {
763 		error = EIO;
764 		goto cleanup;
765 	}
766 
767 	neh = ext4_ext_block_header(bp->b_data);
768 	neh->eh_ecount = 0;
769 	neh->eh_max = ext4_ext_space_block(ip);
770 	neh->eh_magic = EXT4_EXT_MAGIC;
771 	neh->eh_depth = 0;
772 	ex = EXT_FIRST_EXTENT(neh);
773 
774 	if (path[depth].ep_header->eh_ecount != path[depth].ep_header->eh_max) {
775 		ext2_fserr(fs, ip->i_uid,
776 		    "extents count out of range => extent corrupted");
777 		error = EIO;
778 		goto cleanup;
779 	}
780 
781 	/* Start copy from next extent. */
782 	m = 0;
783 	path[depth].ep_ext++;
784 	while (path[depth].ep_ext <= EXT_MAX_EXTENT(path[depth].ep_header)) {
785 		path[depth].ep_ext++;
786 		m++;
787 	}
788 	if (m) {
789 		memmove(ex, path[depth].ep_ext - m,
790 		    sizeof(struct ext4_extent) * m);
791 		neh->eh_ecount = neh->eh_ecount + m;
792 	}
793 
794 	bwrite(bp);
795 	bp = NULL;
796 
797 	/* Fix old leaf. */
798 	if (m) {
799 		path[depth].ep_header->eh_ecount =
800 		    path[depth].ep_header->eh_ecount - m;
801 		ext4_ext_dirty(ip, path + depth);
802 	}
803 
804 	/* Create intermediate indexes. */
805 	k = depth - at - 1;
806 	KASSERT(k >= 0, ("ext4_ext_split: negative k"));
807 
808 	/* Insert new index into current index block. */
809 	i = depth - 1;
810 	while (k--) {
811 		oldblk = newblk;
812 		newblk = ablks[--a];
813 		error = bread(ip->i_devvp, fsbtodb(fs, newblk),
814 		    (int)fs->e2fs_bsize, NOCRED, &bp);
815 		if (error) {
816 			brelse(bp);
817 			goto cleanup;
818 		}
819 
820 		neh = (struct ext4_extent_header *)bp->b_data;
821 		neh->eh_ecount = 1;
822 		neh->eh_magic = EXT4_EXT_MAGIC;
823 		neh->eh_max = ext4_ext_space_block_index(ip);
824 		neh->eh_depth = depth - i;
825 		fidx = EXT_FIRST_INDEX(neh);
826 		fidx->ei_blk = border;
827 		ext4_index_store_pblock(fidx, oldblk);
828 
829 		m = 0;
830 		path[i].ep_index++;
831 		while (path[i].ep_index <= EXT_MAX_INDEX(path[i].ep_header)) {
832 			path[i].ep_index++;
833 			m++;
834 		}
835 		if (m) {
836 			memmove(++fidx, path[i].ep_index - m,
837 			    sizeof(struct ext4_extent_index) * m);
838 			neh->eh_ecount = neh->eh_ecount + m;
839 		}
840 
841 		bwrite(bp);
842 		bp = NULL;
843 
844 		/* Fix old index. */
845 		if (m) {
846 			path[i].ep_header->eh_ecount =
847 			    path[i].ep_header->eh_ecount - m;
848 			ext4_ext_dirty(ip, path + i);
849 		}
850 
851 		i--;
852 	}
853 
854 	error = ext4_ext_insert_index(ip, path + at, border, newblk);
855 
856 cleanup:
857 	if (bp)
858 		brelse(bp);
859 
860 	if (error) {
861 		for (i = 0; i < depth; i++) {
862 			if (!ablks[i])
863 				continue;
864 			ext4_ext_blkfree(ip, ablks[i], 1, 0);
865 		}
866 	}
867 
868 	free(ablks, M_EXT2EXTENTS);
869 
870 	return (error);
871 }
872 
873 static int
874 ext4_ext_grow_indepth(struct inode *ip, struct ext4_extent_path *path,
875     struct ext4_extent *newext)
876 {
877 	struct m_ext2fs *fs;
878 	struct ext4_extent_path *curpath;
879 	struct ext4_extent_header *neh;
880 	struct ext4_extent_index *fidx;
881 	struct buf *bp;
882 	e4fs_daddr_t newblk;
883 	int error = 0;
884 
885 	fs = ip->i_e2fs;
886 	curpath = path;
887 
888 	newblk = ext4_ext_alloc_meta(ip);
889 	if (newblk == 0)
890 		return (error);
891 
892 	bp = getblk(ip->i_devvp, fsbtodb(fs, newblk), fs->e2fs_bsize, 0, 0, 0);
893 	if (!bp)
894 		return (EIO);
895 
896 	/* Move top-level index/leaf into new block. */
897 	memmove(bp->b_data, curpath->ep_header, sizeof(ip->i_data));
898 
899 	/* Set size of new block */
900 	neh = ext4_ext_block_header(bp->b_data);
901 	neh->eh_magic = EXT4_EXT_MAGIC;
902 
903 	if (ext4_ext_inode_depth(ip))
904 		neh->eh_max = ext4_ext_space_block_index(ip);
905 	else
906 		neh->eh_max = ext4_ext_space_block(ip);
907 
908 	error = bwrite(bp);
909 	if (error)
910 		goto out;
911 
912 	bp = NULL;
913 
914 	curpath->ep_header->eh_magic = EXT4_EXT_MAGIC;
915 	curpath->ep_header->eh_max = ext4_ext_space_root(ip);
916 	curpath->ep_header->eh_ecount = 1;
917 	curpath->ep_index = EXT_FIRST_INDEX(curpath->ep_header);
918 	curpath->ep_index->ei_blk = EXT_FIRST_EXTENT(path[0].ep_header)->e_blk;
919 	ext4_index_store_pblock(curpath->ep_index, newblk);
920 
921 	neh = ext4_ext_inode_header(ip);
922 	fidx = EXT_FIRST_INDEX(neh);
923 	neh->eh_depth = path->ep_depth + 1;
924 	ext4_ext_dirty(ip, curpath);
925 out:
926 	brelse(bp);
927 
928 	return (error);
929 }
930 
931 static int
932 ext4_ext_create_new_leaf(struct inode *ip, struct ext4_extent_path *path,
933     struct ext4_extent *newext)
934 {
935 	struct m_ext2fs *fs;
936 	struct ext4_extent_path *curpath;
937 	int depth, i, error;
938 
939 	fs = ip->i_e2fs;
940 
941 repeat:
942 	i = depth = ext4_ext_inode_depth(ip);
943 
944 	/* Look for free index entry int the tree */
945 	curpath = path + depth;
946 	while (i > 0 && !EXT_HAS_FREE_INDEX(curpath)) {
947 		i--;
948 		curpath--;
949 	}
950 
951 	/*
952 	 * We use already allocated block for index block,
953 	 * so subsequent data blocks should be contiguous.
954 	 */
955 	if (EXT_HAS_FREE_INDEX(curpath)) {
956 		error = ext4_ext_split(ip, path, newext, i);
957 		if (error)
958 			goto out;
959 
960 		/* Refill path. */
961 		ext4_ext_drop_refs(path);
962 		error = ext4_ext_find_extent(ip, newext->e_blk, &path);
963 		if (error)
964 			goto out;
965 	} else {
966 		/* Tree is full, do grow in depth. */
967 		error = ext4_ext_grow_indepth(ip, path, newext);
968 		if (error)
969 			goto out;
970 
971 		/* Refill path. */
972 		ext4_ext_drop_refs(path);
973 		error = ext4_ext_find_extent(ip, newext->e_blk, &path);
974 		if (error)
975 			goto out;
976 
977 		/* Check and split tree if required. */
978 		depth = ext4_ext_inode_depth(ip);
979 		if (path[depth].ep_header->eh_ecount ==
980 		    path[depth].ep_header->eh_max)
981 			goto repeat;
982 	}
983 
984 out:
985 	return (error);
986 }
987 
988 static int
989 ext4_ext_correct_indexes(struct inode *ip, struct ext4_extent_path *path)
990 {
991 	struct ext4_extent_header *eh;
992 	struct ext4_extent *ex;
993 	int32_t border;
994 	int depth, k;
995 
996 	depth = ext4_ext_inode_depth(ip);
997 	eh = path[depth].ep_header;
998 	ex = path[depth].ep_ext;
999 
1000 	if (ex == NULL || eh == NULL)
1001 		return (EIO);
1002 
1003 	if (!depth)
1004 		return (0);
1005 
1006 	/* We will correct tree if first leaf got modified only. */
1007 	if (ex != EXT_FIRST_EXTENT(eh))
1008 		return (0);
1009 
1010 	k = depth - 1;
1011 	border = path[depth].ep_ext->e_blk;
1012 	path[k].ep_index->ei_blk = border;
1013 	ext4_ext_dirty(ip, path + k);
1014 	while (k--) {
1015 		/* Change all left-side indexes. */
1016 		if (path[k+1].ep_index != EXT_FIRST_INDEX(path[k+1].ep_header))
1017 			break;
1018 
1019 		path[k].ep_index->ei_blk = border;
1020 		ext4_ext_dirty(ip, path + k);
1021 	}
1022 
1023 	return (0);
1024 }
1025 
1026 static int
1027 ext4_ext_insert_extent(struct inode *ip, struct ext4_extent_path *path,
1028     struct ext4_extent *newext)
1029 {
1030 	struct m_ext2fs *fs;
1031 	struct ext4_extent_header * eh;
1032 	struct ext4_extent *ex, *nex, *nearex;
1033 	struct ext4_extent_path *npath;
1034 	int depth, len, error, next;
1035 
1036 	fs = ip->i_e2fs;
1037 	depth = ext4_ext_inode_depth(ip);
1038 	ex = path[depth].ep_ext;
1039 	npath = NULL;
1040 
1041 	if (newext->e_len == 0 || path[depth].ep_header == NULL)
1042 		return (EINVAL);
1043 
1044 	/* Insert block into found extent. */
1045 	if (ex && ext4_can_extents_be_merged(ex, newext)) {
1046 		ex->e_len = ex->e_len + newext->e_len;
1047 		eh = path[depth].ep_header;
1048 		nearex = ex;
1049 		goto merge;
1050 	}
1051 
1052 repeat:
1053 	depth = ext4_ext_inode_depth(ip);
1054 	eh = path[depth].ep_header;
1055 	if (eh->eh_ecount < eh->eh_max)
1056 		goto has_space;
1057 
1058 	/* Try next leaf */
1059 	nex = EXT_LAST_EXTENT(eh);
1060 	next = ext4_ext_next_leaf_block(ip, path);
1061 	if (newext->e_blk > nex->e_blk && next != EXT4_MAX_BLOCKS) {
1062 		KASSERT(npath == NULL,
1063 		    ("ext4_ext_insert_extent: bad path"));
1064 
1065 		error = ext4_ext_find_extent(ip, next, &npath);
1066 		if (error)
1067 			goto cleanup;
1068 
1069 		if (npath->ep_depth != path->ep_depth) {
1070 			error = EIO;
1071 			goto cleanup;
1072 		}
1073 
1074 		eh = npath[depth].ep_header;
1075 		if (eh->eh_ecount < eh->eh_max) {
1076 			path = npath;
1077 			goto repeat;
1078 		}
1079 	}
1080 
1081 	/*
1082 	 * There is no free space in the found leaf,
1083 	 * try to add a new leaf to the tree.
1084 	 */
1085 	error = ext4_ext_create_new_leaf(ip, path, newext);
1086 	if (error)
1087 		goto cleanup;
1088 
1089 	depth = ext4_ext_inode_depth(ip);
1090 	eh = path[depth].ep_header;
1091 
1092 has_space:
1093 	nearex = path[depth].ep_ext;
1094 	if (!nearex) {
1095 		/* Create new extent in the leaf. */
1096 		path[depth].ep_ext = EXT_FIRST_EXTENT(eh);
1097 	} else if (newext->e_blk > nearex->e_blk) {
1098 		if (nearex != EXT_LAST_EXTENT(eh)) {
1099 			len = EXT_MAX_EXTENT(eh) - nearex;
1100 			len = (len - 1) * sizeof(struct ext4_extent);
1101 			len = len < 0 ? 0 : len;
1102 			memmove(nearex + 2, nearex + 1, len);
1103 		}
1104 		path[depth].ep_ext = nearex + 1;
1105 	} else {
1106 		len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1107 		len = len < 0 ? 0 : len;
1108 		memmove(nearex + 1, nearex, len);
1109 		path[depth].ep_ext = nearex;
1110 	}
1111 
1112 	eh->eh_ecount = eh->eh_ecount + 1;
1113 	nearex = path[depth].ep_ext;
1114 	nearex->e_blk = newext->e_blk;
1115 	nearex->e_start_lo = newext->e_start_lo;
1116 	nearex->e_start_hi = newext->e_start_hi;
1117 	nearex->e_len = newext->e_len;
1118 
1119 merge:
1120 	/* Try to merge extents to the right. */
1121 	while (nearex < EXT_LAST_EXTENT(eh)) {
1122 		if (!ext4_can_extents_be_merged(nearex, nearex + 1))
1123 			break;
1124 
1125 		/* Merge with next extent. */
1126 		nearex->e_len = nearex->e_len + nearex[1].e_len;
1127 		if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1128 			len = (EXT_LAST_EXTENT(eh) - nearex - 1) *
1129 			    sizeof(struct ext4_extent);
1130 			memmove(nearex + 1, nearex + 2, len);
1131 		}
1132 
1133 		eh->eh_ecount = eh->eh_ecount - 1;
1134 		KASSERT(eh->eh_ecount != 0,
1135 		    ("ext4_ext_insert_extent: bad ecount"));
1136 	}
1137 
1138 	/*
1139 	 * Try to merge extents to the left,
1140 	 * start from inexes correction.
1141 	 */
1142 	error = ext4_ext_correct_indexes(ip, path);
1143 	if (error)
1144 		goto cleanup;
1145 
1146 	ext4_ext_dirty(ip, path + depth);
1147 
1148 cleanup:
1149 	if (npath) {
1150 		ext4_ext_drop_refs(npath);
1151 		free(npath, M_EXT2EXTENTS);
1152 	}
1153 
1154 	ip->i_ext_cache.ec_type = EXT4_EXT_CACHE_NO;
1155 	return (error);
1156 }
1157 
1158 static e4fs_daddr_t
1159 ext4_new_blocks(struct inode *ip, daddr_t lbn, e4fs_daddr_t pref,
1160     struct ucred *cred, unsigned long *count, int *perror)
1161 {
1162 	struct m_ext2fs *fs;
1163 	struct ext2mount *ump;
1164 	e4fs_daddr_t newblk;
1165 
1166 	fs = ip->i_e2fs;
1167 	ump = ip->i_ump;
1168 
1169 	/*
1170 	 * We will allocate only single block for now.
1171 	 */
1172 	if (*count > 1)
1173 		return (0);
1174 
1175 	EXT2_LOCK(ip->i_ump);
1176 	*perror = ext2_alloc(ip, lbn, pref, (int)fs->e2fs_bsize, cred, &newblk);
1177 	if (*perror)
1178 		return (0);
1179 
1180 	if (newblk) {
1181 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1182 		ext2_update(ip->i_vnode, 1);
1183 	}
1184 
1185 	return (newblk);
1186 }
1187 
1188 int
1189 ext4_ext_get_blocks(struct inode *ip, e4fs_daddr_t iblk,
1190     unsigned long max_blocks, struct ucred *cred, struct buf **bpp,
1191     int *pallocated, uint32_t *nb)
1192 {
1193 	struct m_ext2fs *fs;
1194 	struct buf *bp = NULL;
1195 	struct ext4_extent_path *path;
1196 	struct ext4_extent newex, *ex;
1197 	e4fs_daddr_t bpref, newblk = 0;
1198 	unsigned long allocated = 0;
1199 	int error = 0, depth;
1200 
1201 	fs = ip->i_e2fs;
1202 	*pallocated = 0;
1203 	path = NULL;
1204 	if(bpp)
1205 		*bpp = NULL;
1206 
1207 	/* Check cache. */
1208 	if ((bpref = ext4_ext_in_cache(ip, iblk, &newex))) {
1209 		if (bpref == EXT4_EXT_CACHE_IN) {
1210 			/* Block is already allocated. */
1211 			newblk = iblk - newex.e_blk +
1212 			    ext4_ext_extent_pblock(&newex);
1213 			allocated = newex.e_len - (iblk - newex.e_blk);
1214 			goto out;
1215 		} else {
1216 			error = EIO;
1217 			goto out2;
1218 		}
1219 	}
1220 
1221 	error = ext4_ext_find_extent(ip, iblk, &path);
1222 	if (error) {
1223 		goto out2;
1224 	}
1225 
1226 	depth = ext4_ext_inode_depth(ip);
1227 	if (path[depth].ep_ext == NULL && depth != 0) {
1228 		error = EIO;
1229 		goto out2;
1230 	}
1231 
1232 	if ((ex = path[depth].ep_ext)) {
1233 		uint64_t lblk = ex->e_blk;
1234 		uint16_t e_len  = ex->e_len;
1235 		e4fs_daddr_t e_start = ext4_ext_extent_pblock(ex);
1236 
1237 		if (e_len > EXT4_MAX_LEN)
1238 			goto out2;
1239 
1240 		/* If we found extent covers block, simply return it. */
1241 		if (iblk >= lblk && iblk < lblk + e_len) {
1242 			newblk = iblk - lblk + e_start;
1243 			allocated = e_len - (iblk - lblk);
1244 			ext4_ext_put_in_cache(ip, lblk, e_len,
1245 			    e_start, EXT4_EXT_CACHE_IN);
1246 			goto out;
1247 		}
1248 	}
1249 
1250 	/* Allocate the new block. */
1251 	if (S_ISREG(ip->i_mode) && (!ip->i_next_alloc_block)) {
1252 		ip->i_next_alloc_goal = 0;
1253 	}
1254 
1255 	bpref = ext4_ext_blkpref(ip, path, iblk);
1256 	allocated = max_blocks;
1257 	newblk = ext4_new_blocks(ip, iblk, bpref, cred, &allocated, &error);
1258 	if (!newblk)
1259 		goto out2;
1260 
1261 	/* Try to insert new extent into found leaf and return. */
1262 	newex.e_blk = iblk;
1263 	ext4_ext_store_pblock(&newex, newblk);
1264 	newex.e_len = allocated;
1265 	error = ext4_ext_insert_extent(ip, path, &newex);
1266 	if (error)
1267 		goto out2;
1268 
1269 	newblk = ext4_ext_extent_pblock(&newex);
1270 	ext4_ext_put_in_cache(ip, iblk, allocated, newblk, EXT4_EXT_CACHE_IN);
1271 	*pallocated = 1;
1272 
1273 out:
1274 	if (allocated > max_blocks)
1275 		allocated = max_blocks;
1276 
1277 	if (bpp)
1278 	{
1279 		error = bread(ip->i_devvp, fsbtodb(fs, newblk),
1280 		    fs->e2fs_bsize, cred, &bp);
1281 		if (error) {
1282 			brelse(bp);
1283 		} else {
1284 			*bpp = bp;
1285 		}
1286 	}
1287 
1288 out2:
1289 	if (path) {
1290 		ext4_ext_drop_refs(path);
1291 		free(path, M_EXT2EXTENTS);
1292 	}
1293 
1294 	if (nb)
1295 		*nb = newblk;
1296 
1297 	return (error);
1298 }
1299 
1300 static inline uint16_t
1301 ext4_ext_get_actual_len(struct ext4_extent *ext)
1302 {
1303 
1304 	return (ext->e_len <= EXT_INIT_MAX_LEN ?
1305 	    ext->e_len : (ext->e_len - EXT_INIT_MAX_LEN));
1306 }
1307 
1308 static inline struct ext4_extent_header *
1309 ext4_ext_header(struct inode *ip)
1310 {
1311 
1312 	return (struct ext4_extent_header *)ip->i_db;
1313 }
1314 
1315 static int
1316 ext4_remove_blocks(struct inode *ip, struct ext4_extent *ex,
1317     unsigned long from, unsigned long to)
1318 {
1319 	unsigned long num, start;
1320 
1321 	if (from >= ex->e_blk &&
1322 	    to == ex->e_blk + ext4_ext_get_actual_len(ex) - 1) {
1323 		/* Tail cleanup. */
1324 		num = ex->e_blk + ext4_ext_get_actual_len(ex) - from;
1325 		start = ext4_ext_extent_pblock(ex) +
1326 		    ext4_ext_get_actual_len(ex) - num;
1327 		ext4_ext_blkfree(ip, start, num, 0);
1328 	}
1329 
1330 	return (0);
1331 }
1332 
1333 static int
1334 ext4_ext_rm_index(struct inode *ip, struct ext4_extent_path *path)
1335 {
1336 	e4fs_daddr_t leaf;
1337 
1338 	/* Free index block. */
1339 	path--;
1340 	leaf = ext4_ext_index_pblock(path->ep_index);
1341 	KASSERT(path->ep_header->eh_ecount != 0,
1342 	    ("ext4_ext_rm_index: bad ecount"));
1343 	path->ep_header->eh_ecount--;
1344 	ext4_ext_dirty(ip, path);
1345 	ext4_ext_blkfree(ip, leaf, 1, 0);
1346 	return (0);
1347 }
1348 
1349 static int
1350 ext4_ext_rm_leaf(struct inode *ip, struct ext4_extent_path *path,
1351     uint64_t start)
1352 {
1353 	struct m_ext2fs *fs;
1354 	int depth;
1355 	struct ext4_extent_header *eh;
1356 	unsigned int a, b, block, num;
1357 	unsigned long ex_blk;
1358 	unsigned short ex_len;
1359 	struct ext4_extent *ex;
1360 	int error, correct_index;
1361 
1362 	fs = ip->i_e2fs;
1363 	depth = ext4_ext_inode_depth(ip);
1364 	correct_index = 0;
1365 
1366 	if (!path[depth].ep_header) {
1367 		if (path[depth].ep_data == NULL)
1368 			return (EINVAL);
1369 		path[depth].ep_header =
1370 		    (struct ext4_extent_header* )path[depth].ep_data;
1371 	}
1372 
1373 	eh = path[depth].ep_header;
1374 	if (!eh) {
1375 		ext2_fserr(fs, ip->i_uid, "bad header => extent corrupted");
1376 		return (EIO);
1377 	}
1378 
1379 	ex = EXT_LAST_EXTENT(eh);
1380 	ex_blk = ex->e_blk;
1381 	ex_len = ext4_ext_get_actual_len(ex);
1382 
1383 	while (ex >= EXT_FIRST_EXTENT(eh) && ex_blk + ex_len > start) {
1384 		path[depth].ep_ext = ex;
1385 		a = ex_blk > start ? ex_blk : start;
1386 		b = (uint64_t)ex_blk + ex_len - 1 <
1387 		    EXT4_MAX_BLOCKS ? ex_blk + ex_len - 1 : EXT4_MAX_BLOCKS;
1388 
1389 		if (a != ex_blk && b != ex_blk + ex_len - 1)
1390 			return (EINVAL);
1391 		else if (a != ex_blk) {
1392 			/* Remove tail of the extent. */
1393 			block = ex_blk;
1394 			num = a - block;
1395 		} else if (b != ex_blk + ex_len - 1) {
1396 			/* Remove head of the extent, not implemented. */
1397 			return (EINVAL);
1398 		} else {
1399 			/* Remove whole extent. */
1400 			block = ex_blk;
1401 			num = 0;
1402 		}
1403 
1404 		if (ex == EXT_FIRST_EXTENT(eh))
1405 			correct_index = 1;
1406 
1407 		error = ext4_remove_blocks(ip, ex, a, b);
1408 		if (error)
1409 			goto out;
1410 
1411 		if (num == 0) {
1412 			ext4_ext_store_pblock(ex, 0);
1413 			eh->eh_ecount--;
1414 		}
1415 
1416 		ex->e_blk = block;
1417 		ex->e_len = num;
1418 
1419 		ext4_ext_dirty(ip, path + depth);
1420 
1421 		ex--;
1422 		ex_blk = ex->e_blk;
1423 		ex_len = ext4_ext_get_actual_len(ex);
1424 	};
1425 
1426 	if (correct_index && eh->eh_ecount)
1427 		error = ext4_ext_correct_indexes(ip, path);
1428 
1429 	/*
1430 	 * If this leaf is free, we should
1431 	 * remove it from index block above.
1432 	 */
1433 	if (error == 0 && eh->eh_ecount == 0 && path[depth].ep_data != NULL)
1434 		error = ext4_ext_rm_index(ip, path + depth);
1435 
1436 out:
1437 	return (error);
1438 }
1439 
1440 static struct buf *
1441 ext4_read_extent_tree_block(struct inode *ip, e4fs_daddr_t pblk,
1442     int depth, int flags)
1443 {
1444 	struct m_ext2fs *fs;
1445 	struct ext4_extent_header *eh;
1446 	struct buf *bp;
1447 	int error;
1448 
1449 	fs = ip->i_e2fs;
1450 
1451 	error = bread(ip->i_devvp, fsbtodb(fs, pblk),
1452 	    fs->e2fs_bsize, NOCRED, &bp);
1453 	if (error) {
1454 		brelse(bp);
1455 		return (NULL);
1456 	}
1457 
1458 	eh = ext4_ext_block_header(bp->b_data);
1459 	if (eh->eh_depth != depth) {
1460 		ext2_fserr(fs, ip->i_uid, "unexpected eh_depth");
1461 		goto err;
1462 	}
1463 
1464 	error = ext4_ext_check_header(ip, eh);
1465 	if (error)
1466 		goto err;
1467 
1468 	return (bp);
1469 
1470 err:
1471 	brelse(bp);
1472 	return (NULL);
1473 
1474 }
1475 
1476 static int inline
1477 ext4_ext_more_to_rm(struct ext4_extent_path *path)
1478 {
1479 
1480 	KASSERT(path->ep_index != NULL,
1481 	    ("ext4_ext_more_to_rm: bad index from path"));
1482 
1483 	if (path->ep_index < EXT_FIRST_INDEX(path->ep_header))
1484 		return (0);
1485 
1486 	if (path->ep_header->eh_ecount == path->index_count)
1487 		return (0);
1488 
1489 	return (1);
1490 }
1491 
1492 int
1493 ext4_ext_remove_space(struct inode *ip, off_t length, int flags,
1494     struct ucred *cred, struct thread *td)
1495 {
1496 	struct buf *bp;
1497 	struct ext4_extent_header *ehp;
1498 	struct ext4_extent_path *path;
1499 	int depth;
1500 	int i, error;
1501 
1502 	ehp = (struct ext4_extent_header *)ip->i_db;
1503 	depth = ext4_ext_inode_depth(ip);
1504 
1505 	error = ext4_ext_check_header(ip, ehp);
1506 	if(error)
1507 		return (error);
1508 
1509 	path = malloc(sizeof(struct ext4_extent_path) * (depth + 1),
1510 	    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
1511 	if (!path)
1512 		return (ENOMEM);
1513 
1514 	i = 0;
1515 	path[0].ep_header = ehp;
1516 	path[0].ep_depth = depth;
1517 	while (i >= 0 && error == 0) {
1518 		if (i == depth) {
1519 			/* This is leaf. */
1520 			error = ext4_ext_rm_leaf(ip, path, length);
1521 			if (error)
1522 				break;
1523 			free(path[i].ep_data, M_EXT2EXTENTS);
1524 			path[i].ep_data = NULL;
1525 			i--;
1526 			continue;
1527 		}
1528 
1529 		/* This is index. */
1530 		if (!path[i].ep_header)
1531 			path[i].ep_header =
1532 			    (struct ext4_extent_header *)path[i].ep_data;
1533 
1534 		if (!path[i].ep_index) {
1535 			/* This level hasn't touched yet. */
1536 			path[i].ep_index = EXT_LAST_INDEX(path[i].ep_header);
1537 			path[i].index_count = path[i].ep_header->eh_ecount + 1;
1538 		} else {
1539 			/* We've already was here, see at next index. */
1540 			path[i].ep_index--;
1541 		}
1542 
1543 		if (ext4_ext_more_to_rm(path + i)) {
1544 			memset(path + i + 1, 0, sizeof(*path));
1545 			bp = ext4_read_extent_tree_block(ip,
1546 			    ext4_ext_index_pblock(path[i].ep_index),
1547 			    path[0].ep_depth - (i + 1), 0);
1548 			if (!bp) {
1549 				error = EIO;
1550 				break;
1551 			}
1552 
1553 			ext4_ext_fill_path_bdata(&path[i+1], bp,
1554 			    ext4_ext_index_pblock(path[i].ep_index));
1555 			brelse(bp);
1556 			path[i].index_count = path[i].ep_header->eh_ecount;
1557 			i++;
1558 		} else {
1559 			if (path[i].ep_header->eh_ecount == 0 && i > 0) {
1560 				/* Index is empty, remove it. */
1561 				error = ext4_ext_rm_index(ip, path + i);
1562 			}
1563 			free(path[i].ep_data, M_EXT2EXTENTS);
1564 			path[i].ep_data = NULL;
1565 			i--;
1566 		}
1567 	}
1568 
1569 	if (path->ep_header->eh_ecount == 0) {
1570 		/*
1571 		 * Truncate the tree to zero.
1572 		 */
1573 		 ext4_ext_header(ip)->eh_depth = 0;
1574 		 ext4_ext_header(ip)->eh_max = ext4_ext_space_root(ip);
1575 		 ext4_ext_dirty(ip, path);
1576 
1577 	}
1578 
1579 	ext4_ext_drop_refs(path);
1580 	free(path, M_EXT2EXTENTS);
1581 
1582 	return (error);
1583 }
1584