xref: /freebsd/sys/fs/ext2fs/ext2_extents.c (revision a3557ef0)
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/endian.h>
40 #include <sys/conf.h>
41 #include <sys/sdt.h>
42 #include <sys/stat.h>
43 
44 #include <fs/ext2fs/ext2_mount.h>
45 #include <fs/ext2fs/fs.h>
46 #include <fs/ext2fs/inode.h>
47 #include <fs/ext2fs/ext2fs.h>
48 #include <fs/ext2fs/ext2_extents.h>
49 #include <fs/ext2fs/ext2_extern.h>
50 
51 SDT_PROVIDER_DECLARE(ext2fs);
52 /*
53  * ext2fs trace probe:
54  * arg0: verbosity. Higher numbers give more verbose messages
55  * arg1: Textual message
56  */
57 SDT_PROBE_DEFINE2(ext2fs, , trace, extents, "int", "char*");
58 
59 static MALLOC_DEFINE(M_EXT2EXTENTS, "ext2_extents", "EXT2 extents");
60 
61 #ifdef EXT2FS_PRINT_EXTENTS
62 static void
63 ext4_ext_print_extent(struct ext4_extent *ep)
64 {
65 
66 	printf("    ext %p => (blk %u len %u start %ju)\n",
67 	    ep, le32toh(ep->e_blk), le16toh(ep->e_len),
68 	    (uint64_t)le16toh(ep->e_start_hi) << 32 | le32toh(ep->e_start_lo));
69 }
70 
71 static void ext4_ext_print_header(struct inode *ip, struct ext4_extent_header *ehp);
72 
73 static void
74 ext4_ext_print_index(struct inode *ip, struct ext4_extent_index *ex, int do_walk)
75 {
76 	struct m_ext2fs *fs;
77 	struct buf *bp;
78 	int error;
79 
80 	fs = ip->i_e2fs;
81 
82 	printf("    index %p => (blk %u pblk %ju)\n",
83 	    ex, le32toh(ex->ei_blk), (uint64_t)le16toh(ex->ei_leaf_hi) << 32 |
84 	    le32toh(ex->ei_leaf_lo));
85 
86 	if(!do_walk)
87 		return;
88 
89 	if ((error = bread(ip->i_devvp,
90 	    fsbtodb(fs, ((uint64_t)le16toh(ex->ei_leaf_hi) << 32 |
91 	    le32toh(ex->ei_leaf_lo))), (int)fs->e2fs_bsize, NOCRED, &bp)) != 0) {
92 		brelse(bp);
93 		return;
94 	}
95 
96 	ext4_ext_print_header(ip, (struct ext4_extent_header *)bp->b_data);
97 
98 	brelse(bp);
99 
100 }
101 
102 static void
103 ext4_ext_print_header(struct inode *ip, struct ext4_extent_header *ehp)
104 {
105 	int i;
106 
107 	printf("header %p => (magic 0x%x entries %d max %d depth %d gen %d)\n",
108 	    ehp, le16toh(ehp->eh_magic), le16toh(ehp->eh_ecount),
109 	    le16toh(ehp->eh_max), le16toh(ehp->eh_depth), le32toh(ehp->eh_gen));
110 
111 	for (i = 0; i < le16toh(ehp->eh_ecount); i++)
112 		if (ehp->eh_depth != 0)
113 			ext4_ext_print_index(ip,
114 			    (struct ext4_extent_index *)(ehp + 1 + i), 1);
115 		else
116 			ext4_ext_print_extent((struct ext4_extent *)(ehp + 1 + i));
117 }
118 
119 static void
120 ext4_ext_print_path(struct inode *ip, struct ext4_extent_path *path)
121 {
122 	int k, l;
123 
124 	l = path->ep_depth;
125 
126 	printf("ip=%ju, Path:\n", ip->i_number);
127 	for (k = 0; k <= l; k++, path++) {
128 		if (path->ep_index) {
129 			ext4_ext_print_index(ip, path->ep_index, 0);
130 		} else if (path->ep_ext) {
131 			ext4_ext_print_extent(path->ep_ext);
132 		}
133 	}
134 }
135 
136 void
137 ext4_ext_print_extent_tree_status(struct inode *ip)
138 {
139 	struct ext4_extent_header *ehp;
140 
141 	ehp = (struct ext4_extent_header *)(char *)ip->i_db;
142 
143 	printf("Extent status:ip=%ju\n", ip->i_number);
144 	if (!(ip->i_flag & IN_E4EXTENTS))
145 		return;
146 
147 	ext4_ext_print_header(ip, ehp);
148 
149 	return;
150 }
151 #endif
152 
153 static inline struct ext4_extent_header *
154 ext4_ext_inode_header(struct inode *ip)
155 {
156 
157 	return ((struct ext4_extent_header *)ip->i_db);
158 }
159 
160 static inline struct ext4_extent_header *
161 ext4_ext_block_header(char *bdata)
162 {
163 
164 	return ((struct ext4_extent_header *)bdata);
165 }
166 
167 static inline unsigned short
168 ext4_ext_inode_depth(struct inode *ip)
169 {
170 	struct ext4_extent_header *ehp;
171 
172 	ehp = (struct ext4_extent_header *)ip->i_data;
173 	return (le16toh(ehp->eh_depth));
174 }
175 
176 static inline e4fs_daddr_t
177 ext4_ext_index_pblock(struct ext4_extent_index *index)
178 {
179 	e4fs_daddr_t blk;
180 
181 	blk = le32toh(index->ei_leaf_lo);
182 	blk |= (e4fs_daddr_t)le16toh(index->ei_leaf_hi) << 32;
183 
184 	return (blk);
185 }
186 
187 static inline void
188 ext4_index_store_pblock(struct ext4_extent_index *index, e4fs_daddr_t pb)
189 {
190 
191 	index->ei_leaf_lo = htole32(pb & 0xffffffff);
192 	index->ei_leaf_hi = htole16((pb >> 32) & 0xffff);
193 }
194 
195 
196 static inline e4fs_daddr_t
197 ext4_ext_extent_pblock(struct ext4_extent *extent)
198 {
199 	e4fs_daddr_t blk;
200 
201 	blk = le32toh(extent->e_start_lo);
202 	blk |= (e4fs_daddr_t)le16toh(extent->e_start_hi) << 32;
203 
204 	return (blk);
205 }
206 
207 static inline void
208 ext4_ext_store_pblock(struct ext4_extent *ex, e4fs_daddr_t pb)
209 {
210 
211 	ex->e_start_lo = htole32(pb & 0xffffffff);
212 	ex->e_start_hi = htole16((pb >> 32) & 0xffff);
213 }
214 
215 int
216 ext4_ext_in_cache(struct inode *ip, daddr_t lbn, struct ext4_extent *ep)
217 {
218 	struct ext4_extent_cache *ecp;
219 	int ret = EXT4_EXT_CACHE_NO;
220 
221 	ecp = &ip->i_ext_cache;
222 	if (ecp->ec_type == EXT4_EXT_CACHE_NO)
223 		return (ret);
224 
225 	if (lbn >= ecp->ec_blk && lbn < ecp->ec_blk + ecp->ec_len) {
226 		ep->e_blk = htole32(ecp->ec_blk);
227 		ep->e_start_lo = htole32(ecp->ec_start & 0xffffffff);
228 		ep->e_start_hi = htole16(ecp->ec_start >> 32 & 0xffff);
229 		ep->e_len = htole16(ecp->ec_len);
230 		ret = ecp->ec_type;
231 	}
232 	return (ret);
233 }
234 
235 static int
236 ext4_ext_check_header(struct inode *ip, struct ext4_extent_header *eh)
237 {
238 	struct m_ext2fs *fs;
239 	char *error_msg;
240 
241 	fs = ip->i_e2fs;
242 
243 	if (le16toh(eh->eh_magic) != EXT4_EXT_MAGIC) {
244 		error_msg = "header: invalid magic";
245 		goto corrupted;
246 	}
247 	if (eh->eh_max == 0) {
248 		error_msg = "header: invalid eh_max";
249 		goto corrupted;
250 	}
251 	if (le16toh(eh->eh_ecount) > le16toh(eh->eh_max)) {
252 		error_msg = "header: invalid eh_entries";
253 		goto corrupted;
254 	}
255 
256 	return (0);
257 
258 corrupted:
259 	SDT_PROBE2(ext2fs, , trace, extents, 1, error_msg);
260 	return (EIO);
261 }
262 
263 static void
264 ext4_ext_binsearch_index(struct ext4_extent_path *path, int blk)
265 {
266 	struct ext4_extent_header *eh;
267 	struct ext4_extent_index *r, *l, *m;
268 
269 	eh = path->ep_header;
270 
271 	KASSERT(le16toh(eh->eh_ecount) <= le16toh(eh->eh_max) &&
272 	    le16toh(eh->eh_ecount) > 0,
273 	    ("ext4_ext_binsearch_index: bad args"));
274 
275 	l = EXT_FIRST_INDEX(eh) + 1;
276 	r = EXT_FIRST_INDEX(eh) + le16toh(eh->eh_ecount) - 1;
277 	while (l <= r) {
278 		m = l + (r - l) / 2;
279 		if (blk < le32toh(m->ei_blk))
280 			r = m - 1;
281 		else
282 			l = m + 1;
283 	}
284 
285 	path->ep_index = l - 1;
286 }
287 
288 static void
289 ext4_ext_binsearch_ext(struct ext4_extent_path *path, int blk)
290 {
291 	struct ext4_extent_header *eh;
292 	struct ext4_extent *r, *l, *m;
293 
294 	eh = path->ep_header;
295 
296 	KASSERT(le16toh(eh->eh_ecount) <= le16toh(eh->eh_max),
297 	    ("ext4_ext_binsearch_ext: bad args"));
298 
299 	if (eh->eh_ecount == 0)
300 		return;
301 
302 	l = EXT_FIRST_EXTENT(eh) + 1;
303 	r = EXT_FIRST_EXTENT(eh) + le16toh(eh->eh_ecount) - 1;
304 
305 	while (l <= r) {
306 		m = l + (r - l) / 2;
307 		if (blk < le32toh(m->e_blk))
308 			r = m - 1;
309 		else
310 			l = m + 1;
311 	}
312 
313 	path->ep_ext = l - 1;
314 }
315 
316 static int
317 ext4_ext_fill_path_bdata(struct ext4_extent_path *path,
318     struct buf *bp, uint64_t blk)
319 {
320 
321 	KASSERT(path->ep_data == NULL,
322 	    ("ext4_ext_fill_path_bdata: bad ep_data"));
323 
324 	path->ep_data = malloc(bp->b_bufsize, M_EXT2EXTENTS, M_WAITOK);
325 	if (!path->ep_data)
326 		return (ENOMEM);
327 
328 	memcpy(path->ep_data, bp->b_data, bp->b_bufsize);
329 	path->ep_blk = blk;
330 
331 	return (0);
332 }
333 
334 static void
335 ext4_ext_fill_path_buf(struct ext4_extent_path *path, struct buf *bp)
336 {
337 
338 	KASSERT(path->ep_data != NULL,
339 	    ("ext4_ext_fill_path_buf: bad ep_data"));
340 
341 	memcpy(bp->b_data, path->ep_data, bp->b_bufsize);
342 }
343 
344 static void
345 ext4_ext_drop_refs(struct ext4_extent_path *path)
346 {
347 	int depth, i;
348 
349 	if (!path)
350 		return;
351 
352 	depth = path->ep_depth;
353 	for (i = 0; i <= depth; i++, path++)
354 		if (path->ep_data) {
355 			free(path->ep_data, M_EXT2EXTENTS);
356 			path->ep_data = NULL;
357 		}
358 }
359 
360 void
361 ext4_ext_path_free(struct ext4_extent_path *path)
362 {
363 
364 	if (!path)
365 		return;
366 
367 	ext4_ext_drop_refs(path);
368 	free(path, M_EXT2EXTENTS);
369 }
370 
371 int
372 ext4_ext_find_extent(struct inode *ip, daddr_t block,
373     struct ext4_extent_path **ppath)
374 {
375 	struct m_ext2fs *fs;
376 	struct ext4_extent_header *eh;
377 	struct ext4_extent_path *path;
378 	struct buf *bp;
379 	uint64_t blk;
380 	int error, depth, i, ppos, alloc;
381 
382 	fs = ip->i_e2fs;
383 	eh = ext4_ext_inode_header(ip);
384 	depth = ext4_ext_inode_depth(ip);
385 	ppos = 0;
386 	alloc = 0;
387 
388 	error = ext4_ext_check_header(ip, eh);
389 	if (error)
390 		return (error);
391 
392 	if (ppath == NULL)
393 		return (EINVAL);
394 
395 	path = *ppath;
396 	if (path == NULL) {
397 		path = malloc(EXT4_EXT_DEPTH_MAX *
398 		    sizeof(struct ext4_extent_path),
399 		    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
400 		if (!path)
401 			return (ENOMEM);
402 
403 		*ppath = path;
404 		alloc = 1;
405 	}
406 
407 	path[0].ep_header = eh;
408 	path[0].ep_data = NULL;
409 
410 	/* Walk through the tree. */
411 	i = depth;
412 	while (i) {
413 		ext4_ext_binsearch_index(&path[ppos], block);
414 		blk = ext4_ext_index_pblock(path[ppos].ep_index);
415 		path[ppos].ep_depth = i;
416 		path[ppos].ep_ext = NULL;
417 
418 		error = bread(ip->i_devvp, fsbtodb(ip->i_e2fs, blk),
419 		    ip->i_e2fs->e2fs_bsize, NOCRED, &bp);
420 		if (error) {
421 			goto error;
422 		}
423 
424 		ppos++;
425 		if (ppos > depth) {
426 			SDT_PROBE2(ext2fs, , trace, extents, 1,
427 			    "ppos > depth => extent corrupted");
428 			error = EIO;
429 			brelse(bp);
430 			goto error;
431 		}
432 
433 		ext4_ext_fill_path_bdata(&path[ppos], bp, blk);
434 		bqrelse(bp);
435 
436 		eh = ext4_ext_block_header(path[ppos].ep_data);
437 		if (ext4_ext_check_header(ip, eh) ||
438 		    ext2_extent_blk_csum_verify(ip, path[ppos].ep_data)) {
439 			error = EIO;
440 			goto error;
441 		}
442 
443 		path[ppos].ep_header = eh;
444 
445 		i--;
446 	}
447 
448 	error = ext4_ext_check_header(ip, eh);
449 	if (error)
450 		goto error;
451 
452 	/* Find extent. */
453 	path[ppos].ep_depth = i;
454 	path[ppos].ep_header = eh;
455 	path[ppos].ep_ext = NULL;
456 	path[ppos].ep_index = NULL;
457 	ext4_ext_binsearch_ext(&path[ppos], block);
458 	return (0);
459 
460 error:
461 	ext4_ext_drop_refs(path);
462 	if (alloc)
463 		free(path, M_EXT2EXTENTS);
464 
465 	*ppath = NULL;
466 
467 	return (error);
468 }
469 
470 static inline int
471 ext4_ext_space_root(struct inode *ip)
472 {
473 	int size;
474 
475 	size = sizeof(ip->i_data);
476 	size -= sizeof(struct ext4_extent_header);
477 	size /= sizeof(struct ext4_extent);
478 
479 	return (size);
480 }
481 
482 static inline int
483 ext4_ext_space_block(struct inode *ip)
484 {
485 	struct m_ext2fs *fs;
486 	int size;
487 
488 	fs = ip->i_e2fs;
489 
490 	size = (fs->e2fs_bsize - sizeof(struct ext4_extent_header)) /
491 	    sizeof(struct ext4_extent);
492 
493 	return (size);
494 }
495 
496 static inline int
497 ext4_ext_space_block_index(struct inode *ip)
498 {
499 	struct m_ext2fs *fs;
500 	int size;
501 
502 	fs = ip->i_e2fs;
503 
504 	size = (fs->e2fs_bsize - sizeof(struct ext4_extent_header)) /
505 	    sizeof(struct ext4_extent_index);
506 
507 	return (size);
508 }
509 
510 void
511 ext4_ext_tree_init(struct inode *ip)
512 {
513 	struct ext4_extent_header *ehp;
514 
515 	ip->i_flag |= IN_E4EXTENTS;
516 
517 	memset(ip->i_data, 0, EXT2_NDADDR + EXT2_NIADDR);
518 	ehp = (struct ext4_extent_header *)ip->i_data;
519 	ehp->eh_magic = htole16(EXT4_EXT_MAGIC);
520 	ehp->eh_max = htole16(ext4_ext_space_root(ip));
521 	ip->i_ext_cache.ec_type = EXT4_EXT_CACHE_NO;
522 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
523 	ext2_update(ip->i_vnode, 1);
524 }
525 
526 static inline void
527 ext4_ext_put_in_cache(struct inode *ip, uint32_t blk,
528 			uint32_t len, uint32_t start, int type)
529 {
530 
531 	KASSERT(len != 0, ("ext4_ext_put_in_cache: bad input"));
532 
533 	ip->i_ext_cache.ec_type = type;
534 	ip->i_ext_cache.ec_blk = blk;
535 	ip->i_ext_cache.ec_len = len;
536 	ip->i_ext_cache.ec_start = start;
537 }
538 
539 static e4fs_daddr_t
540 ext4_ext_blkpref(struct inode *ip, struct ext4_extent_path *path,
541     e4fs_daddr_t block)
542 {
543 	struct m_ext2fs *fs;
544 	struct ext4_extent *ex;
545 	e4fs_daddr_t bg_start;
546 	int depth;
547 
548 	fs = ip->i_e2fs;
549 
550 	if (path) {
551 		depth = path->ep_depth;
552 		ex = path[depth].ep_ext;
553 		if (ex) {
554 			e4fs_daddr_t pblk = ext4_ext_extent_pblock(ex);
555 			e2fs_daddr_t blk = le32toh(ex->e_blk);
556 
557 			if (block > blk)
558 				return (pblk + (block - blk));
559 			else
560 				return (pblk - (blk - block));
561 		}
562 
563 		/* Try to get block from index itself. */
564 		if (path[depth].ep_data)
565 			return (path[depth].ep_blk);
566 	}
567 
568 	/* Use inode's group. */
569 	bg_start = (ip->i_block_group * EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) +
570 	    le32toh(fs->e2fs->e2fs_first_dblock);
571 
572 	return (bg_start + block);
573 }
574 
575 static int inline
576 ext4_can_extents_be_merged(struct ext4_extent *ex1,
577     struct ext4_extent *ex2)
578 {
579 
580 	if (le32toh(ex1->e_blk) + le16toh(ex1->e_len) != le32toh(ex2->e_blk))
581 		return (0);
582 
583 	if (le16toh(ex1->e_len) + le16toh(ex2->e_len) > EXT4_MAX_LEN)
584 		return (0);
585 
586 	if (ext4_ext_extent_pblock(ex1) + le16toh(ex1->e_len) ==
587 	    ext4_ext_extent_pblock(ex2))
588 		return (1);
589 
590 	return (0);
591 }
592 
593 static unsigned
594 ext4_ext_next_leaf_block(struct inode *ip, struct ext4_extent_path *path)
595 {
596 	int depth = path->ep_depth;
597 
598 	/* Empty tree */
599 	if (depth == 0)
600 		return (EXT4_MAX_BLOCKS);
601 
602 	/* Go to indexes. */
603 	depth--;
604 
605 	while (depth >= 0) {
606 		if (path[depth].ep_index !=
607 		    EXT_LAST_INDEX(path[depth].ep_header))
608 			return (le32toh(path[depth].ep_index[1].ei_blk));
609 
610 		depth--;
611 	}
612 
613 	return (EXT4_MAX_BLOCKS);
614 }
615 
616 static int
617 ext4_ext_dirty(struct inode *ip, struct ext4_extent_path *path)
618 {
619 	struct m_ext2fs *fs;
620 	struct buf *bp;
621 	uint64_t blk;
622 	int error;
623 
624 	fs = ip->i_e2fs;
625 
626 	if (!path)
627 		return (EINVAL);
628 
629 	if (path->ep_data) {
630 		blk = path->ep_blk;
631 		bp = getblk(ip->i_devvp, fsbtodb(fs, blk),
632 		    fs->e2fs_bsize, 0, 0, 0);
633 		if (!bp)
634 			return (EIO);
635 		ext4_ext_fill_path_buf(path, bp);
636 		ext2_extent_blk_csum_set(ip, bp->b_data);
637 		error = bwrite(bp);
638 	} else {
639 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
640 		error = ext2_update(ip->i_vnode, 1);
641 	}
642 
643 	return (error);
644 }
645 
646 static int
647 ext4_ext_insert_index(struct inode *ip, struct ext4_extent_path *path,
648     uint32_t lblk, e4fs_daddr_t blk)
649 {
650 	struct m_ext2fs *fs;
651 	struct ext4_extent_index *idx;
652 	int len;
653 
654 	fs = ip->i_e2fs;
655 
656 	if (lblk == le32toh(path->ep_index->ei_blk)) {
657 		SDT_PROBE2(ext2fs, , trace, extents, 1,
658 		    "lblk == index blk => extent corrupted");
659 		return (EIO);
660 	}
661 
662 	if (le16toh(path->ep_header->eh_ecount) >=
663 	    le16toh(path->ep_header->eh_max)) {
664 		SDT_PROBE2(ext2fs, , trace, extents, 1,
665 		    "ecout > maxcount => extent corrupted");
666 		return (EIO);
667 	}
668 
669 	if (lblk > le32toh(path->ep_index->ei_blk)) {
670 		/* Insert after. */
671 		idx = path->ep_index + 1;
672 	} else {
673 		/* Insert before. */
674 		idx = path->ep_index;
675 	}
676 
677 	len = EXT_LAST_INDEX(path->ep_header) - idx + 1;
678 	if (len > 0)
679 		memmove(idx + 1, idx, len * sizeof(struct ext4_extent_index));
680 
681 	if (idx > EXT_MAX_INDEX(path->ep_header)) {
682 		SDT_PROBE2(ext2fs, , trace, extents, 1,
683 		    "index is out of range => extent corrupted");
684 		return (EIO);
685 	}
686 
687 	idx->ei_blk = htole32(lblk);
688 	ext4_index_store_pblock(idx, blk);
689 	path->ep_header->eh_ecount =
690 	    htole16(le16toh(path->ep_header->eh_ecount) + 1);
691 
692 	return (ext4_ext_dirty(ip, path));
693 }
694 
695 static e4fs_daddr_t
696 ext4_ext_alloc_meta(struct inode *ip)
697 {
698 	e4fs_daddr_t blk = ext2_alloc_meta(ip);
699 	if (blk) {
700 		ip->i_blocks += btodb(ip->i_e2fs->e2fs_bsize);
701 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
702 		ext2_update(ip->i_vnode, 1);
703 	}
704 
705 	return (blk);
706 }
707 
708 static void
709 ext4_ext_blkfree(struct inode *ip, uint64_t blk, int count, int flags)
710 {
711 	struct m_ext2fs *fs;
712 	int i, blocksreleased;
713 
714 	fs = ip->i_e2fs;
715 	blocksreleased = count;
716 
717 	for(i = 0; i < count; i++)
718 		ext2_blkfree(ip, blk + i, fs->e2fs_bsize);
719 
720 	if (ip->i_blocks >= blocksreleased)
721 		ip->i_blocks -= (btodb(fs->e2fs_bsize)*blocksreleased);
722 	else
723 		ip->i_blocks = 0;
724 
725 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
726 	ext2_update(ip->i_vnode, 1);
727 }
728 
729 static int
730 ext4_ext_split(struct inode *ip, struct ext4_extent_path *path,
731     struct ext4_extent *newext, int at)
732 {
733 	struct m_ext2fs *fs;
734 	struct  buf *bp;
735 	int depth = ext4_ext_inode_depth(ip);
736 	struct ext4_extent_header *neh;
737 	struct ext4_extent_index *fidx;
738 	struct ext4_extent *ex;
739 	int i = at, k, m, a;
740 	e4fs_daddr_t newblk, oldblk;
741 	uint32_t border;
742 	e4fs_daddr_t *ablks = NULL;
743 	int error = 0;
744 
745 	fs = ip->i_e2fs;
746 	bp = NULL;
747 
748 	/*
749 	 * We will split at current extent for now.
750 	 */
751 	if (path[depth].ep_ext > EXT_MAX_EXTENT(path[depth].ep_header)) {
752 		SDT_PROBE2(ext2fs, , trace, extents, 1,
753 		    "extent is out of range => extent corrupted");
754 		return (EIO);
755 	}
756 
757 	if (path[depth].ep_ext != EXT_MAX_EXTENT(path[depth].ep_header))
758 		border = le32toh(path[depth].ep_ext[1].e_blk);
759 	else
760 		border = le32toh(newext->e_blk);
761 
762 	/* Allocate new blocks. */
763 	ablks = malloc(sizeof(e4fs_daddr_t) * depth,
764 	    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
765 	if (!ablks)
766 		return (ENOMEM);
767 	for (a = 0; a < depth - at; a++) {
768 		newblk = ext4_ext_alloc_meta(ip);
769 		if (newblk == 0)
770 			goto cleanup;
771 		ablks[a] = newblk;
772 	}
773 
774 	newblk = ablks[--a];
775 	bp = getblk(ip->i_devvp, fsbtodb(fs, newblk), fs->e2fs_bsize, 0, 0, 0);
776 	if (!bp) {
777 		error = EIO;
778 		goto cleanup;
779 	}
780 
781 	neh = ext4_ext_block_header(bp->b_data);
782 	neh->eh_ecount = 0;
783 	neh->eh_max = le16toh(ext4_ext_space_block(ip));
784 	neh->eh_magic = le16toh(EXT4_EXT_MAGIC);
785 	neh->eh_depth = 0;
786 	ex = EXT_FIRST_EXTENT(neh);
787 
788 	if (le16toh(path[depth].ep_header->eh_ecount) !=
789 	    le16toh(path[depth].ep_header->eh_max)) {
790 		SDT_PROBE2(ext2fs, , trace, extents, 1,
791 		    "extents count out of range => extent corrupted");
792 		error = EIO;
793 		goto cleanup;
794 	}
795 
796 	/* Start copy from next extent. */
797 	m = 0;
798 	path[depth].ep_ext++;
799 	while (path[depth].ep_ext <= EXT_MAX_EXTENT(path[depth].ep_header)) {
800 		path[depth].ep_ext++;
801 		m++;
802 	}
803 	if (m) {
804 		memmove(ex, path[depth].ep_ext - m,
805 		    sizeof(struct ext4_extent) * m);
806 		neh->eh_ecount = htole16(le16toh(neh->eh_ecount) + m);
807 	}
808 
809 	ext2_extent_blk_csum_set(ip, bp->b_data);
810 	bwrite(bp);
811 	bp = NULL;
812 
813 	/* Fix old leaf. */
814 	if (m) {
815 		path[depth].ep_header->eh_ecount =
816 		    htole16(le16toh(path[depth].ep_header->eh_ecount) - m);
817 		ext4_ext_dirty(ip, path + depth);
818 	}
819 
820 	/* Create intermediate indexes. */
821 	k = depth - at - 1;
822 	KASSERT(k >= 0, ("ext4_ext_split: negative k"));
823 
824 	/* Insert new index into current index block. */
825 	i = depth - 1;
826 	while (k--) {
827 		oldblk = newblk;
828 		newblk = ablks[--a];
829 		error = bread(ip->i_devvp, fsbtodb(fs, newblk),
830 		    (int)fs->e2fs_bsize, NOCRED, &bp);
831 		if (error) {
832 			goto cleanup;
833 		}
834 
835 		neh = (struct ext4_extent_header *)bp->b_data;
836 		neh->eh_ecount = htole16(1);
837 		neh->eh_magic = htole16(EXT4_EXT_MAGIC);
838 		neh->eh_max = htole16(ext4_ext_space_block_index(ip));
839 		neh->eh_depth = htole16(depth - i);
840 		fidx = EXT_FIRST_INDEX(neh);
841 		fidx->ei_blk = htole32(border);
842 		ext4_index_store_pblock(fidx, oldblk);
843 
844 		m = 0;
845 		path[i].ep_index++;
846 		while (path[i].ep_index <= EXT_MAX_INDEX(path[i].ep_header)) {
847 			path[i].ep_index++;
848 			m++;
849 		}
850 		if (m) {
851 			memmove(++fidx, path[i].ep_index - m,
852 			    sizeof(struct ext4_extent_index) * m);
853 			neh->eh_ecount = htole16(le16toh(neh->eh_ecount) + m);
854 		}
855 
856 		ext2_extent_blk_csum_set(ip, bp->b_data);
857 		bwrite(bp);
858 		bp = NULL;
859 
860 		/* Fix old index. */
861 		if (m) {
862 			path[i].ep_header->eh_ecount =
863 			    htole16(le16toh(path[i].ep_header->eh_ecount) - m);
864 			ext4_ext_dirty(ip, path + i);
865 		}
866 
867 		i--;
868 	}
869 
870 	error = ext4_ext_insert_index(ip, path + at, border, newblk);
871 
872 cleanup:
873 	if (bp)
874 		brelse(bp);
875 
876 	if (error) {
877 		for (i = 0; i < depth; i++) {
878 			if (!ablks[i])
879 				continue;
880 			ext4_ext_blkfree(ip, ablks[i], 1, 0);
881 		}
882 	}
883 
884 	free(ablks, M_EXT2EXTENTS);
885 
886 	return (error);
887 }
888 
889 static int
890 ext4_ext_grow_indepth(struct inode *ip, struct ext4_extent_path *path,
891     struct ext4_extent *newext)
892 {
893 	struct m_ext2fs *fs;
894 	struct ext4_extent_path *curpath;
895 	struct ext4_extent_header *neh;
896 	struct buf *bp;
897 	e4fs_daddr_t newblk;
898 	int error = 0;
899 
900 	fs = ip->i_e2fs;
901 	curpath = path;
902 
903 	newblk = ext4_ext_alloc_meta(ip);
904 	if (newblk == 0)
905 		return (error);
906 
907 	bp = getblk(ip->i_devvp, fsbtodb(fs, newblk), fs->e2fs_bsize, 0, 0, 0);
908 	if (!bp)
909 		return (EIO);
910 
911 	/* Move top-level index/leaf into new block. */
912 	memmove(bp->b_data, curpath->ep_header, sizeof(ip->i_data));
913 
914 	/* Set size of new block */
915 	neh = ext4_ext_block_header(bp->b_data);
916 	neh->eh_magic = htole16(EXT4_EXT_MAGIC);
917 
918 	if (ext4_ext_inode_depth(ip))
919 		neh->eh_max = htole16(ext4_ext_space_block_index(ip));
920 	else
921 		neh->eh_max = htole16(ext4_ext_space_block(ip));
922 
923 	ext2_extent_blk_csum_set(ip, bp->b_data);
924 	error = bwrite(bp);
925 	if (error)
926 		goto out;
927 
928 	bp = NULL;
929 
930 	curpath->ep_header->eh_magic = htole16(EXT4_EXT_MAGIC);
931 	curpath->ep_header->eh_max = htole16(ext4_ext_space_root(ip));
932 	curpath->ep_header->eh_ecount = htole16(1);
933 	curpath->ep_index = EXT_FIRST_INDEX(curpath->ep_header);
934 	curpath->ep_index->ei_blk = EXT_FIRST_EXTENT(path[0].ep_header)->e_blk;
935 	ext4_index_store_pblock(curpath->ep_index, newblk);
936 
937 	neh = ext4_ext_inode_header(ip);
938 	neh->eh_depth = htole16(path->ep_depth + 1);
939 	ext4_ext_dirty(ip, curpath);
940 out:
941 	brelse(bp);
942 
943 	return (error);
944 }
945 
946 static int
947 ext4_ext_create_new_leaf(struct inode *ip, struct ext4_extent_path *path,
948     struct ext4_extent *newext)
949 {
950 	struct ext4_extent_path *curpath;
951 	int depth, i, error;
952 
953 repeat:
954 	i = depth = ext4_ext_inode_depth(ip);
955 
956 	/* Look for free index entry int the tree */
957 	curpath = path + depth;
958 	while (i > 0 && !EXT_HAS_FREE_INDEX(curpath)) {
959 		i--;
960 		curpath--;
961 	}
962 
963 	/*
964 	 * We use already allocated block for index block,
965 	 * so subsequent data blocks should be contiguous.
966 	 */
967 	if (EXT_HAS_FREE_INDEX(curpath)) {
968 		error = ext4_ext_split(ip, path, newext, i);
969 		if (error)
970 			goto out;
971 
972 		/* Refill path. */
973 		ext4_ext_drop_refs(path);
974 		error = ext4_ext_find_extent(ip, le32toh(newext->e_blk), &path);
975 		if (error)
976 			goto out;
977 	} else {
978 		/* Tree is full, do grow in depth. */
979 		error = ext4_ext_grow_indepth(ip, path, newext);
980 		if (error)
981 			goto out;
982 
983 		/* Refill path. */
984 		ext4_ext_drop_refs(path);
985 		error = ext4_ext_find_extent(ip, le32toh(newext->e_blk), &path);
986 		if (error)
987 			goto out;
988 
989 		/* Check and split tree if required. */
990 		depth = ext4_ext_inode_depth(ip);
991 		if (le16toh(path[depth].ep_header->eh_ecount) ==
992 		    le16toh(path[depth].ep_header->eh_max))
993 			goto repeat;
994 	}
995 
996 out:
997 	return (error);
998 }
999 
1000 static int
1001 ext4_ext_correct_indexes(struct inode *ip, struct ext4_extent_path *path)
1002 {
1003 	struct ext4_extent_header *eh;
1004 	struct ext4_extent *ex;
1005 	int32_t border;
1006 	int depth, k;
1007 
1008 	depth = ext4_ext_inode_depth(ip);
1009 	eh = path[depth].ep_header;
1010 	ex = path[depth].ep_ext;
1011 
1012 	if (ex == NULL || eh == NULL)
1013 		return (EIO);
1014 
1015 	if (!depth)
1016 		return (0);
1017 
1018 	/* We will correct tree if first leaf got modified only. */
1019 	if (ex != EXT_FIRST_EXTENT(eh))
1020 		return (0);
1021 
1022 	k = depth - 1;
1023 	border = le32toh(path[depth].ep_ext->e_blk);
1024 	path[k].ep_index->ei_blk = htole32(border);
1025 	ext4_ext_dirty(ip, path + k);
1026 	while (k--) {
1027 		/* Change all left-side indexes. */
1028 		if (path[k+1].ep_index != EXT_FIRST_INDEX(path[k+1].ep_header))
1029 			break;
1030 
1031 		path[k].ep_index->ei_blk = htole32(border);
1032 		ext4_ext_dirty(ip, path + k);
1033 	}
1034 
1035 	return (0);
1036 }
1037 
1038 static int
1039 ext4_ext_insert_extent(struct inode *ip, struct ext4_extent_path *path,
1040     struct ext4_extent *newext)
1041 {
1042 	struct ext4_extent_header * eh;
1043 	struct ext4_extent *ex, *nex, *nearex;
1044 	struct ext4_extent_path *npath;
1045 	int depth, len, error, next;
1046 
1047 	depth = ext4_ext_inode_depth(ip);
1048 	ex = path[depth].ep_ext;
1049 	npath = NULL;
1050 
1051 	if (htole16(newext->e_len) == 0 || path[depth].ep_header == NULL)
1052 		return (EINVAL);
1053 
1054 	/* Insert block into found extent. */
1055 	if (ex && ext4_can_extents_be_merged(ex, newext)) {
1056 		ex->e_len = htole16(le16toh(ex->e_len) + le16toh(newext->e_len));
1057 		eh = path[depth].ep_header;
1058 		nearex = ex;
1059 		goto merge;
1060 	}
1061 
1062 repeat:
1063 	depth = ext4_ext_inode_depth(ip);
1064 	eh = path[depth].ep_header;
1065 	if (le16toh(eh->eh_ecount) < le16toh(eh->eh_max))
1066 		goto has_space;
1067 
1068 	/* Try next leaf */
1069 	nex = EXT_LAST_EXTENT(eh);
1070 	next = ext4_ext_next_leaf_block(ip, path);
1071 	if (le32toh(newext->e_blk) > le32toh(nex->e_blk) && next !=
1072 	    EXT4_MAX_BLOCKS) {
1073 		KASSERT(npath == NULL,
1074 		    ("ext4_ext_insert_extent: bad path"));
1075 
1076 		error = ext4_ext_find_extent(ip, next, &npath);
1077 		if (error)
1078 			goto cleanup;
1079 
1080 		if (npath->ep_depth != path->ep_depth) {
1081 			error = EIO;
1082 			goto cleanup;
1083 		}
1084 
1085 		eh = npath[depth].ep_header;
1086 		if (le16toh(eh->eh_ecount) < le16toh(eh->eh_max)) {
1087 			path = npath;
1088 			goto repeat;
1089 		}
1090 	}
1091 
1092 	/*
1093 	 * There is no free space in the found leaf,
1094 	 * try to add a new leaf to the tree.
1095 	 */
1096 	error = ext4_ext_create_new_leaf(ip, path, newext);
1097 	if (error)
1098 		goto cleanup;
1099 
1100 	depth = ext4_ext_inode_depth(ip);
1101 	eh = path[depth].ep_header;
1102 
1103 has_space:
1104 	nearex = path[depth].ep_ext;
1105 	if (!nearex) {
1106 		/* Create new extent in the leaf. */
1107 		path[depth].ep_ext = EXT_FIRST_EXTENT(eh);
1108 	} else if (le32toh(newext->e_blk) > le32toh(nearex->e_blk)) {
1109 		if (nearex != EXT_LAST_EXTENT(eh)) {
1110 			len = EXT_MAX_EXTENT(eh) - nearex;
1111 			len = (len - 1) * sizeof(struct ext4_extent);
1112 			len = len < 0 ? 0 : len;
1113 			memmove(nearex + 2, nearex + 1, len);
1114 		}
1115 		path[depth].ep_ext = nearex + 1;
1116 	} else {
1117 		len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1118 		len = len < 0 ? 0 : len;
1119 		memmove(nearex + 1, nearex, len);
1120 		path[depth].ep_ext = nearex;
1121 	}
1122 
1123 	eh->eh_ecount = htole16(le16toh(eh->eh_ecount) + 1);
1124 	nearex = path[depth].ep_ext;
1125 	nearex->e_blk = newext->e_blk;
1126 	nearex->e_start_lo = newext->e_start_lo;
1127 	nearex->e_start_hi = newext->e_start_hi;
1128 	nearex->e_len = newext->e_len;
1129 
1130 merge:
1131 	/* Try to merge extents to the right. */
1132 	while (nearex < EXT_LAST_EXTENT(eh)) {
1133 		if (!ext4_can_extents_be_merged(nearex, nearex + 1))
1134 			break;
1135 
1136 		/* Merge with next extent. */
1137 		nearex->e_len = htole16(le16toh(nearex->e_len) +
1138 		    le16toh(nearex[1].e_len));
1139 		if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1140 			len = (EXT_LAST_EXTENT(eh) - nearex - 1) *
1141 			    sizeof(struct ext4_extent);
1142 			memmove(nearex + 1, nearex + 2, len);
1143 		}
1144 
1145 		eh->eh_ecount = htole16(le16toh(eh->eh_ecount) - 1);
1146 		KASSERT(le16toh(eh->eh_ecount) != 0,
1147 		    ("ext4_ext_insert_extent: bad ecount"));
1148 	}
1149 
1150 	/*
1151 	 * Try to merge extents to the left,
1152 	 * start from inexes correction.
1153 	 */
1154 	error = ext4_ext_correct_indexes(ip, path);
1155 	if (error)
1156 		goto cleanup;
1157 
1158 	ext4_ext_dirty(ip, path + depth);
1159 
1160 cleanup:
1161 	if (npath) {
1162 		ext4_ext_drop_refs(npath);
1163 		free(npath, M_EXT2EXTENTS);
1164 	}
1165 
1166 	ip->i_ext_cache.ec_type = EXT4_EXT_CACHE_NO;
1167 	return (error);
1168 }
1169 
1170 static e4fs_daddr_t
1171 ext4_new_blocks(struct inode *ip, daddr_t lbn, e4fs_daddr_t pref,
1172     struct ucred *cred, unsigned long *count, int *perror)
1173 {
1174 	struct m_ext2fs *fs;
1175 	e4fs_daddr_t newblk;
1176 
1177 	/*
1178 	 * We will allocate only single block for now.
1179 	 */
1180 	if (*count > 1)
1181 		return (0);
1182 
1183 	fs = ip->i_e2fs;
1184 	EXT2_LOCK(ip->i_ump);
1185 	*perror = ext2_alloc(ip, lbn, pref, (int)fs->e2fs_bsize, cred, &newblk);
1186 	if (*perror)
1187 		return (0);
1188 
1189 	if (newblk) {
1190 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
1191 		ext2_update(ip->i_vnode, 1);
1192 	}
1193 
1194 	return (newblk);
1195 }
1196 
1197 int
1198 ext4_ext_get_blocks(struct inode *ip, e4fs_daddr_t iblk,
1199     unsigned long max_blocks, struct ucred *cred, struct buf **bpp,
1200     int *pallocated, daddr_t *nb)
1201 {
1202 	struct m_ext2fs *fs;
1203 	struct buf *bp = NULL;
1204 	struct ext4_extent_path *path;
1205 	struct ext4_extent newex, *ex;
1206 	e4fs_daddr_t bpref, newblk = 0;
1207 	unsigned long allocated = 0;
1208 	int error = 0, depth;
1209 
1210 	if(bpp)
1211 		*bpp = NULL;
1212 	*pallocated = 0;
1213 
1214 	/* Check cache. */
1215 	path = NULL;
1216 	if ((bpref = ext4_ext_in_cache(ip, iblk, &newex))) {
1217 		if (bpref == EXT4_EXT_CACHE_IN) {
1218 			/* Block is already allocated. */
1219 			newblk = iblk - le32toh(newex.e_blk) +
1220 			    ext4_ext_extent_pblock(&newex);
1221 			allocated = le16toh(newex.e_len) - (iblk - le32toh(newex.e_blk));
1222 			goto out;
1223 		} else {
1224 			error = EIO;
1225 			goto out2;
1226 		}
1227 	}
1228 
1229 	error = ext4_ext_find_extent(ip, iblk, &path);
1230 	if (error) {
1231 		goto out2;
1232 	}
1233 
1234 	depth = ext4_ext_inode_depth(ip);
1235 	if (path[depth].ep_ext == NULL && depth != 0) {
1236 		error = EIO;
1237 		goto out2;
1238 	}
1239 
1240 	if ((ex = path[depth].ep_ext)) {
1241 		uint64_t lblk = le32toh(ex->e_blk);
1242 		uint16_t e_len  = le16toh(ex->e_len);
1243 		e4fs_daddr_t e_start = ext4_ext_extent_pblock(ex);
1244 
1245 		if (e_len > EXT4_MAX_LEN)
1246 			goto out2;
1247 
1248 		/* If we found extent covers block, simply return it. */
1249 		if (iblk >= lblk && iblk < lblk + e_len) {
1250 			newblk = iblk - lblk + e_start;
1251 			allocated = e_len - (iblk - lblk);
1252 			ext4_ext_put_in_cache(ip, lblk, e_len,
1253 			    e_start, EXT4_EXT_CACHE_IN);
1254 			goto out;
1255 		}
1256 	}
1257 
1258 	/* Allocate the new block. */
1259 	if (S_ISREG(ip->i_mode) && (!ip->i_next_alloc_block)) {
1260 		ip->i_next_alloc_goal = 0;
1261 	}
1262 
1263 	bpref = ext4_ext_blkpref(ip, path, iblk);
1264 	allocated = max_blocks;
1265 	newblk = ext4_new_blocks(ip, iblk, bpref, cred, &allocated, &error);
1266 	if (!newblk)
1267 		goto out2;
1268 
1269 	/* Try to insert new extent into found leaf and return. */
1270 	newex.e_blk = htole32(iblk);
1271 	ext4_ext_store_pblock(&newex, newblk);
1272 	newex.e_len = htole16(allocated);
1273 	error = ext4_ext_insert_extent(ip, path, &newex);
1274 	if (error)
1275 		goto out2;
1276 
1277 	newblk = ext4_ext_extent_pblock(&newex);
1278 	ext4_ext_put_in_cache(ip, iblk, allocated, newblk, EXT4_EXT_CACHE_IN);
1279 	*pallocated = 1;
1280 
1281 out:
1282 	if (allocated > max_blocks)
1283 		allocated = max_blocks;
1284 
1285 	if (bpp)
1286 	{
1287 		fs = ip->i_e2fs;
1288 		error = bread(ip->i_devvp, fsbtodb(fs, newblk),
1289 		    fs->e2fs_bsize, cred, &bp);
1290 		if (error) {
1291 			brelse(bp);
1292 		} else {
1293 			*bpp = bp;
1294 		}
1295 	}
1296 
1297 out2:
1298 	if (path) {
1299 		ext4_ext_drop_refs(path);
1300 		free(path, M_EXT2EXTENTS);
1301 	}
1302 
1303 	if (nb)
1304 		*nb = newblk;
1305 
1306 	return (error);
1307 }
1308 
1309 static inline uint16_t
1310 ext4_ext_get_actual_len(struct ext4_extent *ext)
1311 {
1312 
1313 	return (le16toh(ext->e_len) <= EXT_INIT_MAX_LEN ?
1314 	    le16toh(ext->e_len) : (le16toh(ext->e_len) - EXT_INIT_MAX_LEN));
1315 }
1316 
1317 static inline struct ext4_extent_header *
1318 ext4_ext_header(struct inode *ip)
1319 {
1320 
1321 	return ((struct ext4_extent_header *)ip->i_db);
1322 }
1323 
1324 static int
1325 ext4_remove_blocks(struct inode *ip, struct ext4_extent *ex,
1326     unsigned long from, unsigned long to)
1327 {
1328 	unsigned long num, start;
1329 
1330 	if (from >= le32toh(ex->e_blk) &&
1331 	    to == le32toh(ex->e_blk) + ext4_ext_get_actual_len(ex) - 1) {
1332 		/* Tail cleanup. */
1333 		num = le32toh(ex->e_blk) + ext4_ext_get_actual_len(ex) - from;
1334 		start = ext4_ext_extent_pblock(ex) +
1335 		    ext4_ext_get_actual_len(ex) - num;
1336 		ext4_ext_blkfree(ip, start, num, 0);
1337 	}
1338 
1339 	return (0);
1340 }
1341 
1342 static int
1343 ext4_ext_rm_index(struct inode *ip, struct ext4_extent_path *path)
1344 {
1345 	e4fs_daddr_t leaf;
1346 
1347 	/* Free index block. */
1348 	path--;
1349 	leaf = ext4_ext_index_pblock(path->ep_index);
1350 	KASSERT(path->ep_header->eh_ecount != 0,
1351 	    ("ext4_ext_rm_index: bad ecount"));
1352 	path->ep_header->eh_ecount =
1353 	    htole16(le16toh(path->ep_header->eh_ecount) - 1);
1354 	ext4_ext_dirty(ip, path);
1355 	ext4_ext_blkfree(ip, leaf, 1, 0);
1356 	return (0);
1357 }
1358 
1359 static int
1360 ext4_ext_rm_leaf(struct inode *ip, struct ext4_extent_path *path,
1361     uint64_t start)
1362 {
1363 	struct ext4_extent_header *eh;
1364 	struct ext4_extent *ex;
1365 	unsigned int a, b, block, num;
1366 	unsigned long ex_blk;
1367 	unsigned short ex_len;
1368 	int depth;
1369 	int error, correct_index;
1370 
1371 	depth = ext4_ext_inode_depth(ip);
1372 	if (!path[depth].ep_header) {
1373 		if (path[depth].ep_data == NULL)
1374 			return (EINVAL);
1375 		path[depth].ep_header =
1376 		    (struct ext4_extent_header* )path[depth].ep_data;
1377 	}
1378 
1379 	eh = path[depth].ep_header;
1380 	if (!eh) {
1381 		SDT_PROBE2(ext2fs, , trace, extents, 1,
1382 		    "bad header => extent corrupted");
1383 		return (EIO);
1384 	}
1385 
1386 	ex = EXT_LAST_EXTENT(eh);
1387 	ex_blk = le32toh(ex->e_blk);
1388 	ex_len = ext4_ext_get_actual_len(ex);
1389 
1390 	error = 0;
1391 	correct_index = 0;
1392 	while (ex >= EXT_FIRST_EXTENT(eh) && ex_blk + ex_len > start) {
1393 		path[depth].ep_ext = ex;
1394 		a = ex_blk > start ? ex_blk : start;
1395 		b = (uint64_t)ex_blk + ex_len - 1 <
1396 		    EXT4_MAX_BLOCKS ? ex_blk + ex_len - 1 : EXT4_MAX_BLOCKS;
1397 
1398 		if (a != ex_blk && b != ex_blk + ex_len - 1)
1399 			return (EINVAL);
1400 		else if (a != ex_blk) {
1401 			/* Remove tail of the extent. */
1402 			block = ex_blk;
1403 			num = a - block;
1404 		} else if (b != ex_blk + ex_len - 1) {
1405 			/* Remove head of the extent, not implemented. */
1406 			return (EINVAL);
1407 		} else {
1408 			/* Remove whole extent. */
1409 			block = ex_blk;
1410 			num = 0;
1411 		}
1412 
1413 		if (ex == EXT_FIRST_EXTENT(eh))
1414 			correct_index = 1;
1415 
1416 		error = ext4_remove_blocks(ip, ex, a, b);
1417 		if (error)
1418 			goto out;
1419 
1420 		if (num == 0) {
1421 			ext4_ext_store_pblock(ex, 0);
1422 			eh->eh_ecount = htole16(le16toh(eh->eh_ecount) - 1);
1423 		}
1424 
1425 		ex->e_blk = htole32(block);
1426 		ex->e_len = htole16(num);
1427 
1428 		ext4_ext_dirty(ip, path + depth);
1429 
1430 		ex--;
1431 		ex_blk = htole32(ex->e_blk);
1432 		ex_len = ext4_ext_get_actual_len(ex);
1433 	};
1434 
1435 	if (correct_index && le16toh(eh->eh_ecount))
1436 		error = ext4_ext_correct_indexes(ip, path);
1437 
1438 	/*
1439 	 * If this leaf is free, we should
1440 	 * remove it from index block above.
1441 	 */
1442 	if (error == 0 && eh->eh_ecount == 0 &&
1443 	    path[depth].ep_data != NULL)
1444 		error = ext4_ext_rm_index(ip, path + depth);
1445 
1446 out:
1447 	return (error);
1448 }
1449 
1450 static struct buf *
1451 ext4_read_extent_tree_block(struct inode *ip, e4fs_daddr_t pblk,
1452     int depth, int flags)
1453 {
1454 	struct m_ext2fs *fs;
1455 	struct ext4_extent_header *eh;
1456 	struct buf *bp;
1457 	int error;
1458 
1459 	fs = ip->i_e2fs;
1460 	error = bread(ip->i_devvp, fsbtodb(fs, pblk),
1461 	    fs->e2fs_bsize, NOCRED, &bp);
1462 	if (error) {
1463 		return (NULL);
1464 	}
1465 
1466 	eh = ext4_ext_block_header(bp->b_data);
1467 	if (le16toh(eh->eh_depth) != depth) {
1468 		SDT_PROBE2(ext2fs, , trace, extents, 1,
1469 		    "unexpected eh_depth");
1470 		goto err;
1471 	}
1472 
1473 	error = ext4_ext_check_header(ip, eh);
1474 	if (error)
1475 		goto err;
1476 
1477 	return (bp);
1478 
1479 err:
1480 	brelse(bp);
1481 	return (NULL);
1482 
1483 }
1484 
1485 static int inline
1486 ext4_ext_more_to_rm(struct ext4_extent_path *path)
1487 {
1488 
1489 	KASSERT(path->ep_index != NULL,
1490 	    ("ext4_ext_more_to_rm: bad index from path"));
1491 
1492 	if (path->ep_index < EXT_FIRST_INDEX(path->ep_header))
1493 		return (0);
1494 
1495 	if (le16toh(path->ep_header->eh_ecount) == path->index_count)
1496 		return (0);
1497 
1498 	return (1);
1499 }
1500 
1501 int
1502 ext4_ext_remove_space(struct inode *ip, off_t length, int flags,
1503     struct ucred *cred, struct thread *td)
1504 {
1505 	struct buf *bp;
1506 	struct ext4_extent_header *ehp;
1507 	struct ext4_extent_path *path;
1508 	int depth;
1509 	int i, error;
1510 
1511 	ehp = (struct ext4_extent_header *)ip->i_db;
1512 	depth = ext4_ext_inode_depth(ip);
1513 
1514 	error = ext4_ext_check_header(ip, ehp);
1515 	if(error)
1516 		return (error);
1517 
1518 	path = malloc(sizeof(struct ext4_extent_path) * (depth + 1),
1519 	    M_EXT2EXTENTS, M_WAITOK | M_ZERO);
1520 	if (!path)
1521 		return (ENOMEM);
1522 
1523 	path[0].ep_header = ehp;
1524 	path[0].ep_depth = depth;
1525 	i = 0;
1526 	while (error == 0 && i >= 0) {
1527 		if (i == depth) {
1528 			/* This is leaf. */
1529 			error = ext4_ext_rm_leaf(ip, path, length);
1530 			if (error)
1531 				break;
1532 			free(path[i].ep_data, M_EXT2EXTENTS);
1533 			path[i].ep_data = NULL;
1534 			i--;
1535 			continue;
1536 		}
1537 
1538 		/* This is index. */
1539 		if (!path[i].ep_header)
1540 			path[i].ep_header =
1541 			    (struct ext4_extent_header *)path[i].ep_data;
1542 
1543 		if (!path[i].ep_index) {
1544 			/* This level hasn't touched yet. */
1545 			path[i].ep_index = EXT_LAST_INDEX(path[i].ep_header);
1546 			path[i].index_count =
1547 			    le16toh(path[i].ep_header->eh_ecount) + 1;
1548 		} else {
1549 			/* We've already was here, see at next index. */
1550 			path[i].ep_index--;
1551 		}
1552 
1553 		if (ext4_ext_more_to_rm(path + i)) {
1554 			memset(path + i + 1, 0, sizeof(*path));
1555 			bp = ext4_read_extent_tree_block(ip,
1556 			    ext4_ext_index_pblock(path[i].ep_index),
1557 			    path[0].ep_depth - (i + 1), 0);
1558 			if (!bp) {
1559 				error = EIO;
1560 				break;
1561 			}
1562 
1563 			ext4_ext_fill_path_bdata(&path[i+1], bp,
1564 			    ext4_ext_index_pblock(path[i].ep_index));
1565 			brelse(bp);
1566 			path[i].index_count =
1567 			    le16toh(path[i].ep_header->eh_ecount);
1568 			i++;
1569 		} else {
1570 			if (path[i].ep_header->eh_ecount == 0 && i > 0) {
1571 				/* Index is empty, remove it. */
1572 				error = ext4_ext_rm_index(ip, path + i);
1573 			}
1574 			free(path[i].ep_data, M_EXT2EXTENTS);
1575 			path[i].ep_data = NULL;
1576 			i--;
1577 		}
1578 	}
1579 
1580 	if (path->ep_header->eh_ecount == 0) {
1581 		/*
1582 		 * Truncate the tree to zero.
1583 		 */
1584 		 ext4_ext_header(ip)->eh_depth = 0;
1585 		 ext4_ext_header(ip)->eh_max = htole16(ext4_ext_space_root(ip));
1586 		 ext4_ext_dirty(ip, path);
1587 	}
1588 
1589 	ext4_ext_drop_refs(path);
1590 	free(path, M_EXT2EXTENTS);
1591 
1592 	return (error);
1593 }
1594