xref: /openbmc/linux/fs/ext4/extents.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3  * Written by Alex Tomas <alex@clusterfs.com>
4  *
5  * Architecture independence:
6  *   Copyright (c) 2005, Bull S.A.
7  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public Licens
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-
21  */
22 
23 /*
24  * Extents support for EXT4
25  *
26  * TODO:
27  *   - ext4*_error() should be used in some situations
28  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29  *   - smart tree reduction
30  */
31 
32 #include <linux/module.h>
33 #include <linux/fs.h>
34 #include <linux/time.h>
35 #include <linux/ext4_jbd2.h>
36 #include <linux/jbd.h>
37 #include <linux/highuid.h>
38 #include <linux/pagemap.h>
39 #include <linux/quotaops.h>
40 #include <linux/string.h>
41 #include <linux/slab.h>
42 #include <linux/ext4_fs_extents.h>
43 #include <asm/uaccess.h>
44 
45 
46 /*
47  * ext_pblock:
48  * combine low and high parts of physical block number into ext4_fsblk_t
49  */
50 static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
51 {
52 	ext4_fsblk_t block;
53 
54 	block = le32_to_cpu(ex->ee_start);
55 	block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
56 	return block;
57 }
58 
59 /*
60  * idx_pblock:
61  * combine low and high parts of a leaf physical block number into ext4_fsblk_t
62  */
63 static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
64 {
65 	ext4_fsblk_t block;
66 
67 	block = le32_to_cpu(ix->ei_leaf);
68 	block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
69 	return block;
70 }
71 
72 /*
73  * ext4_ext_store_pblock:
74  * stores a large physical block number into an extent struct,
75  * breaking it into parts
76  */
77 static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
78 {
79 	ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
80 	ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
81 }
82 
83 /*
84  * ext4_idx_store_pblock:
85  * stores a large physical block number into an index struct,
86  * breaking it into parts
87  */
88 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
89 {
90 	ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
91 	ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
92 }
93 
94 static int ext4_ext_check_header(const char *function, struct inode *inode,
95 				struct ext4_extent_header *eh)
96 {
97 	const char *error_msg = NULL;
98 
99 	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
100 		error_msg = "invalid magic";
101 		goto corrupted;
102 	}
103 	if (unlikely(eh->eh_max == 0)) {
104 		error_msg = "invalid eh_max";
105 		goto corrupted;
106 	}
107 	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
108 		error_msg = "invalid eh_entries";
109 		goto corrupted;
110 	}
111 	return 0;
112 
113 corrupted:
114 	ext4_error(inode->i_sb, function,
115 			"bad header in inode #%lu: %s - magic %x, "
116 			"entries %u, max %u, depth %u",
117 			inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
118 			le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
119 			le16_to_cpu(eh->eh_depth));
120 
121 	return -EIO;
122 }
123 
124 static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
125 {
126 	int err;
127 
128 	if (handle->h_buffer_credits > needed)
129 		return handle;
130 	if (!ext4_journal_extend(handle, needed))
131 		return handle;
132 	err = ext4_journal_restart(handle, needed);
133 
134 	return handle;
135 }
136 
137 /*
138  * could return:
139  *  - EROFS
140  *  - ENOMEM
141  */
142 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
143 				struct ext4_ext_path *path)
144 {
145 	if (path->p_bh) {
146 		/* path points to block */
147 		return ext4_journal_get_write_access(handle, path->p_bh);
148 	}
149 	/* path points to leaf/index in inode body */
150 	/* we use in-core data, no need to protect them */
151 	return 0;
152 }
153 
154 /*
155  * could return:
156  *  - EROFS
157  *  - ENOMEM
158  *  - EIO
159  */
160 static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
161 				struct ext4_ext_path *path)
162 {
163 	int err;
164 	if (path->p_bh) {
165 		/* path points to block */
166 		err = ext4_journal_dirty_metadata(handle, path->p_bh);
167 	} else {
168 		/* path points to leaf/index in inode body */
169 		err = ext4_mark_inode_dirty(handle, inode);
170 	}
171 	return err;
172 }
173 
174 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
175 			      struct ext4_ext_path *path,
176 			      ext4_fsblk_t block)
177 {
178 	struct ext4_inode_info *ei = EXT4_I(inode);
179 	ext4_fsblk_t bg_start;
180 	ext4_grpblk_t colour;
181 	int depth;
182 
183 	if (path) {
184 		struct ext4_extent *ex;
185 		depth = path->p_depth;
186 
187 		/* try to predict block placement */
188 		ex = path[depth].p_ext;
189 		if (ex)
190 			return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
191 
192 		/* it looks like index is empty;
193 		 * try to find starting block from index itself */
194 		if (path[depth].p_bh)
195 			return path[depth].p_bh->b_blocknr;
196 	}
197 
198 	/* OK. use inode's group */
199 	bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
200 		le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
201 	colour = (current->pid % 16) *
202 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
203 	return bg_start + colour + block;
204 }
205 
206 static ext4_fsblk_t
207 ext4_ext_new_block(handle_t *handle, struct inode *inode,
208 			struct ext4_ext_path *path,
209 			struct ext4_extent *ex, int *err)
210 {
211 	ext4_fsblk_t goal, newblock;
212 
213 	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
214 	newblock = ext4_new_block(handle, inode, goal, err);
215 	return newblock;
216 }
217 
218 static int ext4_ext_space_block(struct inode *inode)
219 {
220 	int size;
221 
222 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
223 			/ sizeof(struct ext4_extent);
224 #ifdef AGGRESSIVE_TEST
225 	if (size > 6)
226 		size = 6;
227 #endif
228 	return size;
229 }
230 
231 static int ext4_ext_space_block_idx(struct inode *inode)
232 {
233 	int size;
234 
235 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
236 			/ sizeof(struct ext4_extent_idx);
237 #ifdef AGGRESSIVE_TEST
238 	if (size > 5)
239 		size = 5;
240 #endif
241 	return size;
242 }
243 
244 static int ext4_ext_space_root(struct inode *inode)
245 {
246 	int size;
247 
248 	size = sizeof(EXT4_I(inode)->i_data);
249 	size -= sizeof(struct ext4_extent_header);
250 	size /= sizeof(struct ext4_extent);
251 #ifdef AGGRESSIVE_TEST
252 	if (size > 3)
253 		size = 3;
254 #endif
255 	return size;
256 }
257 
258 static int ext4_ext_space_root_idx(struct inode *inode)
259 {
260 	int size;
261 
262 	size = sizeof(EXT4_I(inode)->i_data);
263 	size -= sizeof(struct ext4_extent_header);
264 	size /= sizeof(struct ext4_extent_idx);
265 #ifdef AGGRESSIVE_TEST
266 	if (size > 4)
267 		size = 4;
268 #endif
269 	return size;
270 }
271 
272 #ifdef EXT_DEBUG
273 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
274 {
275 	int k, l = path->p_depth;
276 
277 	ext_debug("path:");
278 	for (k = 0; k <= l; k++, path++) {
279 		if (path->p_idx) {
280 		  ext_debug("  %d->%llu", le32_to_cpu(path->p_idx->ei_block),
281 			    idx_pblock(path->p_idx));
282 		} else if (path->p_ext) {
283 			ext_debug("  %d:%d:%llu ",
284 				  le32_to_cpu(path->p_ext->ee_block),
285 				  le16_to_cpu(path->p_ext->ee_len),
286 				  ext_pblock(path->p_ext));
287 		} else
288 			ext_debug("  []");
289 	}
290 	ext_debug("\n");
291 }
292 
293 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
294 {
295 	int depth = ext_depth(inode);
296 	struct ext4_extent_header *eh;
297 	struct ext4_extent *ex;
298 	int i;
299 
300 	if (!path)
301 		return;
302 
303 	eh = path[depth].p_hdr;
304 	ex = EXT_FIRST_EXTENT(eh);
305 
306 	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
307 		ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
308 			  le16_to_cpu(ex->ee_len), ext_pblock(ex));
309 	}
310 	ext_debug("\n");
311 }
312 #else
313 #define ext4_ext_show_path(inode,path)
314 #define ext4_ext_show_leaf(inode,path)
315 #endif
316 
317 static void ext4_ext_drop_refs(struct ext4_ext_path *path)
318 {
319 	int depth = path->p_depth;
320 	int i;
321 
322 	for (i = 0; i <= depth; i++, path++)
323 		if (path->p_bh) {
324 			brelse(path->p_bh);
325 			path->p_bh = NULL;
326 		}
327 }
328 
329 /*
330  * ext4_ext_binsearch_idx:
331  * binary search for the closest index of the given block
332  */
333 static void
334 ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
335 {
336 	struct ext4_extent_header *eh = path->p_hdr;
337 	struct ext4_extent_idx *r, *l, *m;
338 
339 	BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
340 	BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
341 	BUG_ON(le16_to_cpu(eh->eh_entries) <= 0);
342 
343 	ext_debug("binsearch for %d(idx):  ", block);
344 
345 	l = EXT_FIRST_INDEX(eh) + 1;
346 	r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
347 	while (l <= r) {
348 		m = l + (r - l) / 2;
349 		if (block < le32_to_cpu(m->ei_block))
350 			r = m - 1;
351 		else
352 			l = m + 1;
353 		ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
354 				m, m->ei_block, r, r->ei_block);
355 	}
356 
357 	path->p_idx = l - 1;
358 	ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
359 		  idx_block(path->p_idx));
360 
361 #ifdef CHECK_BINSEARCH
362 	{
363 		struct ext4_extent_idx *chix, *ix;
364 		int k;
365 
366 		chix = ix = EXT_FIRST_INDEX(eh);
367 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
368 		  if (k != 0 &&
369 		      le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
370 				printk("k=%d, ix=0x%p, first=0x%p\n", k,
371 					ix, EXT_FIRST_INDEX(eh));
372 				printk("%u <= %u\n",
373 				       le32_to_cpu(ix->ei_block),
374 				       le32_to_cpu(ix[-1].ei_block));
375 			}
376 			BUG_ON(k && le32_to_cpu(ix->ei_block)
377 				           <= le32_to_cpu(ix[-1].ei_block));
378 			if (block < le32_to_cpu(ix->ei_block))
379 				break;
380 			chix = ix;
381 		}
382 		BUG_ON(chix != path->p_idx);
383 	}
384 #endif
385 
386 }
387 
388 /*
389  * ext4_ext_binsearch:
390  * binary search for closest extent of the given block
391  */
392 static void
393 ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
394 {
395 	struct ext4_extent_header *eh = path->p_hdr;
396 	struct ext4_extent *r, *l, *m;
397 
398 	BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
399 	BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
400 
401 	if (eh->eh_entries == 0) {
402 		/*
403 		 * this leaf is empty:
404 		 * we get such a leaf in split/add case
405 		 */
406 		return;
407 	}
408 
409 	ext_debug("binsearch for %d:  ", block);
410 
411 	l = EXT_FIRST_EXTENT(eh) + 1;
412 	r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
413 
414 	while (l <= r) {
415 		m = l + (r - l) / 2;
416 		if (block < le32_to_cpu(m->ee_block))
417 			r = m - 1;
418 		else
419 			l = m + 1;
420 		ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
421 				m, m->ee_block, r, r->ee_block);
422 	}
423 
424 	path->p_ext = l - 1;
425 	ext_debug("  -> %d:%llu:%d ",
426 		        le32_to_cpu(path->p_ext->ee_block),
427 		        ext_pblock(path->p_ext),
428 			le16_to_cpu(path->p_ext->ee_len));
429 
430 #ifdef CHECK_BINSEARCH
431 	{
432 		struct ext4_extent *chex, *ex;
433 		int k;
434 
435 		chex = ex = EXT_FIRST_EXTENT(eh);
436 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
437 			BUG_ON(k && le32_to_cpu(ex->ee_block)
438 				          <= le32_to_cpu(ex[-1].ee_block));
439 			if (block < le32_to_cpu(ex->ee_block))
440 				break;
441 			chex = ex;
442 		}
443 		BUG_ON(chex != path->p_ext);
444 	}
445 #endif
446 
447 }
448 
449 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
450 {
451 	struct ext4_extent_header *eh;
452 
453 	eh = ext_inode_hdr(inode);
454 	eh->eh_depth = 0;
455 	eh->eh_entries = 0;
456 	eh->eh_magic = EXT4_EXT_MAGIC;
457 	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
458 	ext4_mark_inode_dirty(handle, inode);
459 	ext4_ext_invalidate_cache(inode);
460 	return 0;
461 }
462 
463 struct ext4_ext_path *
464 ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
465 {
466 	struct ext4_extent_header *eh;
467 	struct buffer_head *bh;
468 	short int depth, i, ppos = 0, alloc = 0;
469 
470 	eh = ext_inode_hdr(inode);
471 	BUG_ON(eh == NULL);
472 	if (ext4_ext_check_header(__FUNCTION__, inode, eh))
473 		return ERR_PTR(-EIO);
474 
475 	i = depth = ext_depth(inode);
476 
477 	/* account possible depth increase */
478 	if (!path) {
479 		path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
480 				GFP_NOFS);
481 		if (!path)
482 			return ERR_PTR(-ENOMEM);
483 		alloc = 1;
484 	}
485 	path[0].p_hdr = eh;
486 
487 	/* walk through the tree */
488 	while (i) {
489 		ext_debug("depth %d: num %d, max %d\n",
490 			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
491 		ext4_ext_binsearch_idx(inode, path + ppos, block);
492 		path[ppos].p_block = idx_pblock(path[ppos].p_idx);
493 		path[ppos].p_depth = i;
494 		path[ppos].p_ext = NULL;
495 
496 		bh = sb_bread(inode->i_sb, path[ppos].p_block);
497 		if (!bh)
498 			goto err;
499 
500 		eh = ext_block_hdr(bh);
501 		ppos++;
502 		BUG_ON(ppos > depth);
503 		path[ppos].p_bh = bh;
504 		path[ppos].p_hdr = eh;
505 		i--;
506 
507 		if (ext4_ext_check_header(__FUNCTION__, inode, eh))
508 			goto err;
509 	}
510 
511 	path[ppos].p_depth = i;
512 	path[ppos].p_hdr = eh;
513 	path[ppos].p_ext = NULL;
514 	path[ppos].p_idx = NULL;
515 
516 	if (ext4_ext_check_header(__FUNCTION__, inode, eh))
517 		goto err;
518 
519 	/* find extent */
520 	ext4_ext_binsearch(inode, path + ppos, block);
521 
522 	ext4_ext_show_path(inode, path);
523 
524 	return path;
525 
526 err:
527 	ext4_ext_drop_refs(path);
528 	if (alloc)
529 		kfree(path);
530 	return ERR_PTR(-EIO);
531 }
532 
533 /*
534  * ext4_ext_insert_index:
535  * insert new index [@logical;@ptr] into the block at @curp;
536  * check where to insert: before @curp or after @curp
537  */
538 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
539 				struct ext4_ext_path *curp,
540 				int logical, ext4_fsblk_t ptr)
541 {
542 	struct ext4_extent_idx *ix;
543 	int len, err;
544 
545 	err = ext4_ext_get_access(handle, inode, curp);
546 	if (err)
547 		return err;
548 
549 	BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
550 	len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
551 	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
552 		/* insert after */
553 		if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
554 			len = (len - 1) * sizeof(struct ext4_extent_idx);
555 			len = len < 0 ? 0 : len;
556 			ext_debug("insert new index %d after: %d. "
557 					"move %d from 0x%p to 0x%p\n",
558 					logical, ptr, len,
559 					(curp->p_idx + 1), (curp->p_idx + 2));
560 			memmove(curp->p_idx + 2, curp->p_idx + 1, len);
561 		}
562 		ix = curp->p_idx + 1;
563 	} else {
564 		/* insert before */
565 		len = len * sizeof(struct ext4_extent_idx);
566 		len = len < 0 ? 0 : len;
567 		ext_debug("insert new index %d before: %d. "
568 				"move %d from 0x%p to 0x%p\n",
569 				logical, ptr, len,
570 				curp->p_idx, (curp->p_idx + 1));
571 		memmove(curp->p_idx + 1, curp->p_idx, len);
572 		ix = curp->p_idx;
573 	}
574 
575 	ix->ei_block = cpu_to_le32(logical);
576 	ext4_idx_store_pblock(ix, ptr);
577 	curp->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(curp->p_hdr->eh_entries)+1);
578 
579 	BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
580 	                     > le16_to_cpu(curp->p_hdr->eh_max));
581 	BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
582 
583 	err = ext4_ext_dirty(handle, inode, curp);
584 	ext4_std_error(inode->i_sb, err);
585 
586 	return err;
587 }
588 
589 /*
590  * ext4_ext_split:
591  * inserts new subtree into the path, using free index entry
592  * at depth @at:
593  * - allocates all needed blocks (new leaf and all intermediate index blocks)
594  * - makes decision where to split
595  * - moves remaining extents and index entries (right to the split point)
596  *   into the newly allocated blocks
597  * - initializes subtree
598  */
599 static int ext4_ext_split(handle_t *handle, struct inode *inode,
600 				struct ext4_ext_path *path,
601 				struct ext4_extent *newext, int at)
602 {
603 	struct buffer_head *bh = NULL;
604 	int depth = ext_depth(inode);
605 	struct ext4_extent_header *neh;
606 	struct ext4_extent_idx *fidx;
607 	struct ext4_extent *ex;
608 	int i = at, k, m, a;
609 	ext4_fsblk_t newblock, oldblock;
610 	__le32 border;
611 	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
612 	int err = 0;
613 
614 	/* make decision: where to split? */
615 	/* FIXME: now decision is simplest: at current extent */
616 
617 	/* if current leaf will be split, then we should use
618 	 * border from split point */
619 	BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
620 	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
621 		border = path[depth].p_ext[1].ee_block;
622 		ext_debug("leaf will be split."
623 				" next leaf starts at %d\n",
624 			          le32_to_cpu(border));
625 	} else {
626 		border = newext->ee_block;
627 		ext_debug("leaf will be added."
628 				" next leaf starts at %d\n",
629 			        le32_to_cpu(border));
630 	}
631 
632 	/*
633 	 * If error occurs, then we break processing
634 	 * and mark filesystem read-only. index won't
635 	 * be inserted and tree will be in consistent
636 	 * state. Next mount will repair buffers too.
637 	 */
638 
639 	/*
640 	 * Get array to track all allocated blocks.
641 	 * We need this to handle errors and free blocks
642 	 * upon them.
643 	 */
644 	ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
645 	if (!ablocks)
646 		return -ENOMEM;
647 
648 	/* allocate all needed blocks */
649 	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
650 	for (a = 0; a < depth - at; a++) {
651 		newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
652 		if (newblock == 0)
653 			goto cleanup;
654 		ablocks[a] = newblock;
655 	}
656 
657 	/* initialize new leaf */
658 	newblock = ablocks[--a];
659 	BUG_ON(newblock == 0);
660 	bh = sb_getblk(inode->i_sb, newblock);
661 	if (!bh) {
662 		err = -EIO;
663 		goto cleanup;
664 	}
665 	lock_buffer(bh);
666 
667 	err = ext4_journal_get_create_access(handle, bh);
668 	if (err)
669 		goto cleanup;
670 
671 	neh = ext_block_hdr(bh);
672 	neh->eh_entries = 0;
673 	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
674 	neh->eh_magic = EXT4_EXT_MAGIC;
675 	neh->eh_depth = 0;
676 	ex = EXT_FIRST_EXTENT(neh);
677 
678 	/* move remainder of path[depth] to the new leaf */
679 	BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
680 	/* start copy from next extent */
681 	/* TODO: we could do it by single memmove */
682 	m = 0;
683 	path[depth].p_ext++;
684 	while (path[depth].p_ext <=
685 			EXT_MAX_EXTENT(path[depth].p_hdr)) {
686 		ext_debug("move %d:%llu:%d in new leaf %llu\n",
687 			        le32_to_cpu(path[depth].p_ext->ee_block),
688 			        ext_pblock(path[depth].p_ext),
689 			        le16_to_cpu(path[depth].p_ext->ee_len),
690 				newblock);
691 		/*memmove(ex++, path[depth].p_ext++,
692 				sizeof(struct ext4_extent));
693 		neh->eh_entries++;*/
694 		path[depth].p_ext++;
695 		m++;
696 	}
697 	if (m) {
698 		memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
699 		neh->eh_entries = cpu_to_le16(le16_to_cpu(neh->eh_entries)+m);
700 	}
701 
702 	set_buffer_uptodate(bh);
703 	unlock_buffer(bh);
704 
705 	err = ext4_journal_dirty_metadata(handle, bh);
706 	if (err)
707 		goto cleanup;
708 	brelse(bh);
709 	bh = NULL;
710 
711 	/* correct old leaf */
712 	if (m) {
713 		err = ext4_ext_get_access(handle, inode, path + depth);
714 		if (err)
715 			goto cleanup;
716 		path[depth].p_hdr->eh_entries =
717 		     cpu_to_le16(le16_to_cpu(path[depth].p_hdr->eh_entries)-m);
718 		err = ext4_ext_dirty(handle, inode, path + depth);
719 		if (err)
720 			goto cleanup;
721 
722 	}
723 
724 	/* create intermediate indexes */
725 	k = depth - at - 1;
726 	BUG_ON(k < 0);
727 	if (k)
728 		ext_debug("create %d intermediate indices\n", k);
729 	/* insert new index into current index block */
730 	/* current depth stored in i var */
731 	i = depth - 1;
732 	while (k--) {
733 		oldblock = newblock;
734 		newblock = ablocks[--a];
735 		bh = sb_getblk(inode->i_sb, (ext4_fsblk_t)newblock);
736 		if (!bh) {
737 			err = -EIO;
738 			goto cleanup;
739 		}
740 		lock_buffer(bh);
741 
742 		err = ext4_journal_get_create_access(handle, bh);
743 		if (err)
744 			goto cleanup;
745 
746 		neh = ext_block_hdr(bh);
747 		neh->eh_entries = cpu_to_le16(1);
748 		neh->eh_magic = EXT4_EXT_MAGIC;
749 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
750 		neh->eh_depth = cpu_to_le16(depth - i);
751 		fidx = EXT_FIRST_INDEX(neh);
752 		fidx->ei_block = border;
753 		ext4_idx_store_pblock(fidx, oldblock);
754 
755 		ext_debug("int.index at %d (block %llu): %lu -> %llu\n", i,
756 				newblock, (unsigned long) le32_to_cpu(border),
757 				oldblock);
758 		/* copy indexes */
759 		m = 0;
760 		path[i].p_idx++;
761 
762 		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
763 				EXT_MAX_INDEX(path[i].p_hdr));
764 		BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
765 				EXT_LAST_INDEX(path[i].p_hdr));
766 		while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
767 			ext_debug("%d: move %d:%d in new index %llu\n", i,
768 				        le32_to_cpu(path[i].p_idx->ei_block),
769 				        idx_pblock(path[i].p_idx),
770 				        newblock);
771 			/*memmove(++fidx, path[i].p_idx++,
772 					sizeof(struct ext4_extent_idx));
773 			neh->eh_entries++;
774 			BUG_ON(neh->eh_entries > neh->eh_max);*/
775 			path[i].p_idx++;
776 			m++;
777 		}
778 		if (m) {
779 			memmove(++fidx, path[i].p_idx - m,
780 				sizeof(struct ext4_extent_idx) * m);
781 			neh->eh_entries =
782 				cpu_to_le16(le16_to_cpu(neh->eh_entries) + m);
783 		}
784 		set_buffer_uptodate(bh);
785 		unlock_buffer(bh);
786 
787 		err = ext4_journal_dirty_metadata(handle, bh);
788 		if (err)
789 			goto cleanup;
790 		brelse(bh);
791 		bh = NULL;
792 
793 		/* correct old index */
794 		if (m) {
795 			err = ext4_ext_get_access(handle, inode, path + i);
796 			if (err)
797 				goto cleanup;
798 			path[i].p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path[i].p_hdr->eh_entries)-m);
799 			err = ext4_ext_dirty(handle, inode, path + i);
800 			if (err)
801 				goto cleanup;
802 		}
803 
804 		i--;
805 	}
806 
807 	/* insert new index */
808 	err = ext4_ext_insert_index(handle, inode, path + at,
809 				    le32_to_cpu(border), newblock);
810 
811 cleanup:
812 	if (bh) {
813 		if (buffer_locked(bh))
814 			unlock_buffer(bh);
815 		brelse(bh);
816 	}
817 
818 	if (err) {
819 		/* free all allocated blocks in error case */
820 		for (i = 0; i < depth; i++) {
821 			if (!ablocks[i])
822 				continue;
823 			ext4_free_blocks(handle, inode, ablocks[i], 1);
824 		}
825 	}
826 	kfree(ablocks);
827 
828 	return err;
829 }
830 
831 /*
832  * ext4_ext_grow_indepth:
833  * implements tree growing procedure:
834  * - allocates new block
835  * - moves top-level data (index block or leaf) into the new block
836  * - initializes new top-level, creating index that points to the
837  *   just created block
838  */
839 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
840 					struct ext4_ext_path *path,
841 					struct ext4_extent *newext)
842 {
843 	struct ext4_ext_path *curp = path;
844 	struct ext4_extent_header *neh;
845 	struct ext4_extent_idx *fidx;
846 	struct buffer_head *bh;
847 	ext4_fsblk_t newblock;
848 	int err = 0;
849 
850 	newblock = ext4_ext_new_block(handle, inode, path, newext, &err);
851 	if (newblock == 0)
852 		return err;
853 
854 	bh = sb_getblk(inode->i_sb, newblock);
855 	if (!bh) {
856 		err = -EIO;
857 		ext4_std_error(inode->i_sb, err);
858 		return err;
859 	}
860 	lock_buffer(bh);
861 
862 	err = ext4_journal_get_create_access(handle, bh);
863 	if (err) {
864 		unlock_buffer(bh);
865 		goto out;
866 	}
867 
868 	/* move top-level index/leaf into new block */
869 	memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
870 
871 	/* set size of new block */
872 	neh = ext_block_hdr(bh);
873 	/* old root could have indexes or leaves
874 	 * so calculate e_max right way */
875 	if (ext_depth(inode))
876 	  neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
877 	else
878 	  neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
879 	neh->eh_magic = EXT4_EXT_MAGIC;
880 	set_buffer_uptodate(bh);
881 	unlock_buffer(bh);
882 
883 	err = ext4_journal_dirty_metadata(handle, bh);
884 	if (err)
885 		goto out;
886 
887 	/* create index in new top-level index: num,max,pointer */
888 	err = ext4_ext_get_access(handle, inode, curp);
889 	if (err)
890 		goto out;
891 
892 	curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
893 	curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
894 	curp->p_hdr->eh_entries = cpu_to_le16(1);
895 	curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
896 	/* FIXME: it works, but actually path[0] can be index */
897 	curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
898 	ext4_idx_store_pblock(curp->p_idx, newblock);
899 
900 	neh = ext_inode_hdr(inode);
901 	fidx = EXT_FIRST_INDEX(neh);
902 	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
903 		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
904 		  le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
905 
906 	neh->eh_depth = cpu_to_le16(path->p_depth + 1);
907 	err = ext4_ext_dirty(handle, inode, curp);
908 out:
909 	brelse(bh);
910 
911 	return err;
912 }
913 
914 /*
915  * ext4_ext_create_new_leaf:
916  * finds empty index and adds new leaf.
917  * if no free index is found, then it requests in-depth growing.
918  */
919 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
920 					struct ext4_ext_path *path,
921 					struct ext4_extent *newext)
922 {
923 	struct ext4_ext_path *curp;
924 	int depth, i, err = 0;
925 
926 repeat:
927 	i = depth = ext_depth(inode);
928 
929 	/* walk up to the tree and look for free index entry */
930 	curp = path + depth;
931 	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
932 		i--;
933 		curp--;
934 	}
935 
936 	/* we use already allocated block for index block,
937 	 * so subsequent data blocks should be contiguous */
938 	if (EXT_HAS_FREE_INDEX(curp)) {
939 		/* if we found index with free entry, then use that
940 		 * entry: create all needed subtree and add new leaf */
941 		err = ext4_ext_split(handle, inode, path, newext, i);
942 
943 		/* refill path */
944 		ext4_ext_drop_refs(path);
945 		path = ext4_ext_find_extent(inode,
946 					    le32_to_cpu(newext->ee_block),
947 					    path);
948 		if (IS_ERR(path))
949 			err = PTR_ERR(path);
950 	} else {
951 		/* tree is full, time to grow in depth */
952 		err = ext4_ext_grow_indepth(handle, inode, path, newext);
953 		if (err)
954 			goto out;
955 
956 		/* refill path */
957 		ext4_ext_drop_refs(path);
958 		path = ext4_ext_find_extent(inode,
959 					    le32_to_cpu(newext->ee_block),
960 					    path);
961 		if (IS_ERR(path)) {
962 			err = PTR_ERR(path);
963 			goto out;
964 		}
965 
966 		/*
967 		 * only first (depth 0 -> 1) produces free space;
968 		 * in all other cases we have to split the grown tree
969 		 */
970 		depth = ext_depth(inode);
971 		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
972 			/* now we need to split */
973 			goto repeat;
974 		}
975 	}
976 
977 out:
978 	return err;
979 }
980 
981 /*
982  * ext4_ext_next_allocated_block:
983  * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
984  * NOTE: it considers block number from index entry as
985  * allocated block. Thus, index entries have to be consistent
986  * with leaves.
987  */
988 static unsigned long
989 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
990 {
991 	int depth;
992 
993 	BUG_ON(path == NULL);
994 	depth = path->p_depth;
995 
996 	if (depth == 0 && path->p_ext == NULL)
997 		return EXT_MAX_BLOCK;
998 
999 	while (depth >= 0) {
1000 		if (depth == path->p_depth) {
1001 			/* leaf */
1002 			if (path[depth].p_ext !=
1003 					EXT_LAST_EXTENT(path[depth].p_hdr))
1004 			  return le32_to_cpu(path[depth].p_ext[1].ee_block);
1005 		} else {
1006 			/* index */
1007 			if (path[depth].p_idx !=
1008 					EXT_LAST_INDEX(path[depth].p_hdr))
1009 			  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1010 		}
1011 		depth--;
1012 	}
1013 
1014 	return EXT_MAX_BLOCK;
1015 }
1016 
1017 /*
1018  * ext4_ext_next_leaf_block:
1019  * returns first allocated block from next leaf or EXT_MAX_BLOCK
1020  */
1021 static unsigned ext4_ext_next_leaf_block(struct inode *inode,
1022 					struct ext4_ext_path *path)
1023 {
1024 	int depth;
1025 
1026 	BUG_ON(path == NULL);
1027 	depth = path->p_depth;
1028 
1029 	/* zero-tree has no leaf blocks at all */
1030 	if (depth == 0)
1031 		return EXT_MAX_BLOCK;
1032 
1033 	/* go to index block */
1034 	depth--;
1035 
1036 	while (depth >= 0) {
1037 		if (path[depth].p_idx !=
1038 				EXT_LAST_INDEX(path[depth].p_hdr))
1039 		  return le32_to_cpu(path[depth].p_idx[1].ei_block);
1040 		depth--;
1041 	}
1042 
1043 	return EXT_MAX_BLOCK;
1044 }
1045 
1046 /*
1047  * ext4_ext_correct_indexes:
1048  * if leaf gets modified and modified extent is first in the leaf,
1049  * then we have to correct all indexes above.
1050  * TODO: do we need to correct tree in all cases?
1051  */
1052 int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1053 				struct ext4_ext_path *path)
1054 {
1055 	struct ext4_extent_header *eh;
1056 	int depth = ext_depth(inode);
1057 	struct ext4_extent *ex;
1058 	__le32 border;
1059 	int k, err = 0;
1060 
1061 	eh = path[depth].p_hdr;
1062 	ex = path[depth].p_ext;
1063 	BUG_ON(ex == NULL);
1064 	BUG_ON(eh == NULL);
1065 
1066 	if (depth == 0) {
1067 		/* there is no tree at all */
1068 		return 0;
1069 	}
1070 
1071 	if (ex != EXT_FIRST_EXTENT(eh)) {
1072 		/* we correct tree if first leaf got modified only */
1073 		return 0;
1074 	}
1075 
1076 	/*
1077 	 * TODO: we need correction if border is smaller than current one
1078 	 */
1079 	k = depth - 1;
1080 	border = path[depth].p_ext->ee_block;
1081 	err = ext4_ext_get_access(handle, inode, path + k);
1082 	if (err)
1083 		return err;
1084 	path[k].p_idx->ei_block = border;
1085 	err = ext4_ext_dirty(handle, inode, path + k);
1086 	if (err)
1087 		return err;
1088 
1089 	while (k--) {
1090 		/* change all left-side indexes */
1091 		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1092 			break;
1093 		err = ext4_ext_get_access(handle, inode, path + k);
1094 		if (err)
1095 			break;
1096 		path[k].p_idx->ei_block = border;
1097 		err = ext4_ext_dirty(handle, inode, path + k);
1098 		if (err)
1099 			break;
1100 	}
1101 
1102 	return err;
1103 }
1104 
1105 static int
1106 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1107 				struct ext4_extent *ex2)
1108 {
1109 	if (le32_to_cpu(ex1->ee_block) + le16_to_cpu(ex1->ee_len) !=
1110 			le32_to_cpu(ex2->ee_block))
1111 		return 0;
1112 
1113 	/*
1114 	 * To allow future support for preallocated extents to be added
1115 	 * as an RO_COMPAT feature, refuse to merge to extents if
1116 	 * this can result in the top bit of ee_len being set.
1117 	 */
1118 	if (le16_to_cpu(ex1->ee_len) + le16_to_cpu(ex2->ee_len) > EXT_MAX_LEN)
1119 		return 0;
1120 #ifdef AGGRESSIVE_TEST
1121 	if (le16_to_cpu(ex1->ee_len) >= 4)
1122 		return 0;
1123 #endif
1124 
1125 	if (ext_pblock(ex1) + le16_to_cpu(ex1->ee_len) == ext_pblock(ex2))
1126 		return 1;
1127 	return 0;
1128 }
1129 
1130 /*
1131  * ext4_ext_insert_extent:
1132  * tries to merge requsted extent into the existing extent or
1133  * inserts requested extent as new one into the tree,
1134  * creating new leaf in the no-space case.
1135  */
1136 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1137 				struct ext4_ext_path *path,
1138 				struct ext4_extent *newext)
1139 {
1140 	struct ext4_extent_header * eh;
1141 	struct ext4_extent *ex, *fex;
1142 	struct ext4_extent *nearex; /* nearest extent */
1143 	struct ext4_ext_path *npath = NULL;
1144 	int depth, len, err, next;
1145 
1146 	BUG_ON(newext->ee_len == 0);
1147 	depth = ext_depth(inode);
1148 	ex = path[depth].p_ext;
1149 	BUG_ON(path[depth].p_hdr == NULL);
1150 
1151 	/* try to insert block into found extent and return */
1152 	if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
1153 		ext_debug("append %d block to %d:%d (from %llu)\n",
1154 				le16_to_cpu(newext->ee_len),
1155 				le32_to_cpu(ex->ee_block),
1156 				le16_to_cpu(ex->ee_len), ext_pblock(ex));
1157 		err = ext4_ext_get_access(handle, inode, path + depth);
1158 		if (err)
1159 			return err;
1160 		ex->ee_len = cpu_to_le16(le16_to_cpu(ex->ee_len)
1161 					 + le16_to_cpu(newext->ee_len));
1162 		eh = path[depth].p_hdr;
1163 		nearex = ex;
1164 		goto merge;
1165 	}
1166 
1167 repeat:
1168 	depth = ext_depth(inode);
1169 	eh = path[depth].p_hdr;
1170 	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1171 		goto has_space;
1172 
1173 	/* probably next leaf has space for us? */
1174 	fex = EXT_LAST_EXTENT(eh);
1175 	next = ext4_ext_next_leaf_block(inode, path);
1176 	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1177 	    && next != EXT_MAX_BLOCK) {
1178 		ext_debug("next leaf block - %d\n", next);
1179 		BUG_ON(npath != NULL);
1180 		npath = ext4_ext_find_extent(inode, next, NULL);
1181 		if (IS_ERR(npath))
1182 			return PTR_ERR(npath);
1183 		BUG_ON(npath->p_depth != path->p_depth);
1184 		eh = npath[depth].p_hdr;
1185 		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1186 			ext_debug("next leaf isnt full(%d)\n",
1187 				  le16_to_cpu(eh->eh_entries));
1188 			path = npath;
1189 			goto repeat;
1190 		}
1191 		ext_debug("next leaf has no free space(%d,%d)\n",
1192 			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1193 	}
1194 
1195 	/*
1196 	 * There is no free space in the found leaf.
1197 	 * We're gonna add a new leaf in the tree.
1198 	 */
1199 	err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1200 	if (err)
1201 		goto cleanup;
1202 	depth = ext_depth(inode);
1203 	eh = path[depth].p_hdr;
1204 
1205 has_space:
1206 	nearex = path[depth].p_ext;
1207 
1208 	err = ext4_ext_get_access(handle, inode, path + depth);
1209 	if (err)
1210 		goto cleanup;
1211 
1212 	if (!nearex) {
1213 		/* there is no extent in this leaf, create first one */
1214 		ext_debug("first extent in the leaf: %d:%llu:%d\n",
1215 			        le32_to_cpu(newext->ee_block),
1216 			        ext_pblock(newext),
1217 			        le16_to_cpu(newext->ee_len));
1218 		path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1219 	} else if (le32_to_cpu(newext->ee_block)
1220 		           > le32_to_cpu(nearex->ee_block)) {
1221 /*		BUG_ON(newext->ee_block == nearex->ee_block); */
1222 		if (nearex != EXT_LAST_EXTENT(eh)) {
1223 			len = EXT_MAX_EXTENT(eh) - nearex;
1224 			len = (len - 1) * sizeof(struct ext4_extent);
1225 			len = len < 0 ? 0 : len;
1226 			ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
1227 					"move %d from 0x%p to 0x%p\n",
1228 				        le32_to_cpu(newext->ee_block),
1229 				        ext_pblock(newext),
1230 				        le16_to_cpu(newext->ee_len),
1231 					nearex, len, nearex + 1, nearex + 2);
1232 			memmove(nearex + 2, nearex + 1, len);
1233 		}
1234 		path[depth].p_ext = nearex + 1;
1235 	} else {
1236 		BUG_ON(newext->ee_block == nearex->ee_block);
1237 		len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1238 		len = len < 0 ? 0 : len;
1239 		ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
1240 				"move %d from 0x%p to 0x%p\n",
1241 				le32_to_cpu(newext->ee_block),
1242 				ext_pblock(newext),
1243 				le16_to_cpu(newext->ee_len),
1244 				nearex, len, nearex + 1, nearex + 2);
1245 		memmove(nearex + 1, nearex, len);
1246 		path[depth].p_ext = nearex;
1247 	}
1248 
1249 	eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
1250 	nearex = path[depth].p_ext;
1251 	nearex->ee_block = newext->ee_block;
1252 	nearex->ee_start = newext->ee_start;
1253 	nearex->ee_start_hi = newext->ee_start_hi;
1254 	nearex->ee_len = newext->ee_len;
1255 
1256 merge:
1257 	/* try to merge extents to the right */
1258 	while (nearex < EXT_LAST_EXTENT(eh)) {
1259 		if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1))
1260 			break;
1261 		/* merge with next extent! */
1262 		nearex->ee_len = cpu_to_le16(le16_to_cpu(nearex->ee_len)
1263 					     + le16_to_cpu(nearex[1].ee_len));
1264 		if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
1265 			len = (EXT_LAST_EXTENT(eh) - nearex - 1)
1266 					* sizeof(struct ext4_extent);
1267 			memmove(nearex + 1, nearex + 2, len);
1268 		}
1269 		eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1270 		BUG_ON(eh->eh_entries == 0);
1271 	}
1272 
1273 	/* try to merge extents to the left */
1274 
1275 	/* time to correct all indexes above */
1276 	err = ext4_ext_correct_indexes(handle, inode, path);
1277 	if (err)
1278 		goto cleanup;
1279 
1280 	err = ext4_ext_dirty(handle, inode, path + depth);
1281 
1282 cleanup:
1283 	if (npath) {
1284 		ext4_ext_drop_refs(npath);
1285 		kfree(npath);
1286 	}
1287 	ext4_ext_tree_changed(inode);
1288 	ext4_ext_invalidate_cache(inode);
1289 	return err;
1290 }
1291 
1292 int ext4_ext_walk_space(struct inode *inode, unsigned long block,
1293 			unsigned long num, ext_prepare_callback func,
1294 			void *cbdata)
1295 {
1296 	struct ext4_ext_path *path = NULL;
1297 	struct ext4_ext_cache cbex;
1298 	struct ext4_extent *ex;
1299 	unsigned long next, start = 0, end = 0;
1300 	unsigned long last = block + num;
1301 	int depth, exists, err = 0;
1302 
1303 	BUG_ON(func == NULL);
1304 	BUG_ON(inode == NULL);
1305 
1306 	while (block < last && block != EXT_MAX_BLOCK) {
1307 		num = last - block;
1308 		/* find extent for this block */
1309 		path = ext4_ext_find_extent(inode, block, path);
1310 		if (IS_ERR(path)) {
1311 			err = PTR_ERR(path);
1312 			path = NULL;
1313 			break;
1314 		}
1315 
1316 		depth = ext_depth(inode);
1317 		BUG_ON(path[depth].p_hdr == NULL);
1318 		ex = path[depth].p_ext;
1319 		next = ext4_ext_next_allocated_block(path);
1320 
1321 		exists = 0;
1322 		if (!ex) {
1323 			/* there is no extent yet, so try to allocate
1324 			 * all requested space */
1325 			start = block;
1326 			end = block + num;
1327 		} else if (le32_to_cpu(ex->ee_block) > block) {
1328 			/* need to allocate space before found extent */
1329 			start = block;
1330 			end = le32_to_cpu(ex->ee_block);
1331 			if (block + num < end)
1332 				end = block + num;
1333 		} else if (block >=
1334 			     le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len)) {
1335 			/* need to allocate space after found extent */
1336 			start = block;
1337 			end = block + num;
1338 			if (end >= next)
1339 				end = next;
1340 		} else if (block >= le32_to_cpu(ex->ee_block)) {
1341 			/*
1342 			 * some part of requested space is covered
1343 			 * by found extent
1344 			 */
1345 			start = block;
1346 			end = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len);
1347 			if (block + num < end)
1348 				end = block + num;
1349 			exists = 1;
1350 		} else {
1351 			BUG();
1352 		}
1353 		BUG_ON(end <= start);
1354 
1355 		if (!exists) {
1356 			cbex.ec_block = start;
1357 			cbex.ec_len = end - start;
1358 			cbex.ec_start = 0;
1359 			cbex.ec_type = EXT4_EXT_CACHE_GAP;
1360 		} else {
1361 		        cbex.ec_block = le32_to_cpu(ex->ee_block);
1362 		        cbex.ec_len = le16_to_cpu(ex->ee_len);
1363 		        cbex.ec_start = ext_pblock(ex);
1364 			cbex.ec_type = EXT4_EXT_CACHE_EXTENT;
1365 		}
1366 
1367 		BUG_ON(cbex.ec_len == 0);
1368 		err = func(inode, path, &cbex, cbdata);
1369 		ext4_ext_drop_refs(path);
1370 
1371 		if (err < 0)
1372 			break;
1373 		if (err == EXT_REPEAT)
1374 			continue;
1375 		else if (err == EXT_BREAK) {
1376 			err = 0;
1377 			break;
1378 		}
1379 
1380 		if (ext_depth(inode) != depth) {
1381 			/* depth was changed. we have to realloc path */
1382 			kfree(path);
1383 			path = NULL;
1384 		}
1385 
1386 		block = cbex.ec_block + cbex.ec_len;
1387 	}
1388 
1389 	if (path) {
1390 		ext4_ext_drop_refs(path);
1391 		kfree(path);
1392 	}
1393 
1394 	return err;
1395 }
1396 
1397 static void
1398 ext4_ext_put_in_cache(struct inode *inode, __u32 block,
1399 			__u32 len, __u32 start, int type)
1400 {
1401 	struct ext4_ext_cache *cex;
1402 	BUG_ON(len == 0);
1403 	cex = &EXT4_I(inode)->i_cached_extent;
1404 	cex->ec_type = type;
1405 	cex->ec_block = block;
1406 	cex->ec_len = len;
1407 	cex->ec_start = start;
1408 }
1409 
1410 /*
1411  * ext4_ext_put_gap_in_cache:
1412  * calculate boundaries of the gap that the requested block fits into
1413  * and cache this gap
1414  */
1415 static void
1416 ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
1417 				unsigned long block)
1418 {
1419 	int depth = ext_depth(inode);
1420 	unsigned long lblock, len;
1421 	struct ext4_extent *ex;
1422 
1423 	ex = path[depth].p_ext;
1424 	if (ex == NULL) {
1425 		/* there is no extent yet, so gap is [0;-] */
1426 		lblock = 0;
1427 		len = EXT_MAX_BLOCK;
1428 		ext_debug("cache gap(whole file):");
1429 	} else if (block < le32_to_cpu(ex->ee_block)) {
1430 		lblock = block;
1431 		len = le32_to_cpu(ex->ee_block) - block;
1432 		ext_debug("cache gap(before): %lu [%lu:%lu]",
1433 				(unsigned long) block,
1434 			        (unsigned long) le32_to_cpu(ex->ee_block),
1435 			        (unsigned long) le16_to_cpu(ex->ee_len));
1436 	} else if (block >= le32_to_cpu(ex->ee_block)
1437 		            + le16_to_cpu(ex->ee_len)) {
1438 	        lblock = le32_to_cpu(ex->ee_block)
1439 		         + le16_to_cpu(ex->ee_len);
1440 		len = ext4_ext_next_allocated_block(path);
1441 		ext_debug("cache gap(after): [%lu:%lu] %lu",
1442 			        (unsigned long) le32_to_cpu(ex->ee_block),
1443 			        (unsigned long) le16_to_cpu(ex->ee_len),
1444 				(unsigned long) block);
1445 		BUG_ON(len == lblock);
1446 		len = len - lblock;
1447 	} else {
1448 		lblock = len = 0;
1449 		BUG();
1450 	}
1451 
1452 	ext_debug(" -> %lu:%lu\n", (unsigned long) lblock, len);
1453 	ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1454 }
1455 
1456 static int
1457 ext4_ext_in_cache(struct inode *inode, unsigned long block,
1458 			struct ext4_extent *ex)
1459 {
1460 	struct ext4_ext_cache *cex;
1461 
1462 	cex = &EXT4_I(inode)->i_cached_extent;
1463 
1464 	/* has cache valid data? */
1465 	if (cex->ec_type == EXT4_EXT_CACHE_NO)
1466 		return EXT4_EXT_CACHE_NO;
1467 
1468 	BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1469 			cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1470 	if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
1471 	        ex->ee_block = cpu_to_le32(cex->ec_block);
1472 		ext4_ext_store_pblock(ex, cex->ec_start);
1473 	        ex->ee_len = cpu_to_le16(cex->ec_len);
1474 		ext_debug("%lu cached by %lu:%lu:%llu\n",
1475 				(unsigned long) block,
1476 				(unsigned long) cex->ec_block,
1477 				(unsigned long) cex->ec_len,
1478 				cex->ec_start);
1479 		return cex->ec_type;
1480 	}
1481 
1482 	/* not in cache */
1483 	return EXT4_EXT_CACHE_NO;
1484 }
1485 
1486 /*
1487  * ext4_ext_rm_idx:
1488  * removes index from the index block.
1489  * It's used in truncate case only, thus all requests are for
1490  * last index in the block only.
1491  */
1492 int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
1493 			struct ext4_ext_path *path)
1494 {
1495 	struct buffer_head *bh;
1496 	int err;
1497 	ext4_fsblk_t leaf;
1498 
1499 	/* free index block */
1500 	path--;
1501 	leaf = idx_pblock(path->p_idx);
1502 	BUG_ON(path->p_hdr->eh_entries == 0);
1503 	err = ext4_ext_get_access(handle, inode, path);
1504 	if (err)
1505 		return err;
1506 	path->p_hdr->eh_entries = cpu_to_le16(le16_to_cpu(path->p_hdr->eh_entries)-1);
1507 	err = ext4_ext_dirty(handle, inode, path);
1508 	if (err)
1509 		return err;
1510 	ext_debug("index is empty, remove it, free block %llu\n", leaf);
1511 	bh = sb_find_get_block(inode->i_sb, leaf);
1512 	ext4_forget(handle, 1, inode, bh, leaf);
1513 	ext4_free_blocks(handle, inode, leaf, 1);
1514 	return err;
1515 }
1516 
1517 /*
1518  * ext4_ext_calc_credits_for_insert:
1519  * This routine returns max. credits that the extent tree can consume.
1520  * It should be OK for low-performance paths like ->writepage()
1521  * To allow many writing processes to fit into a single transaction,
1522  * the caller should calculate credits under truncate_mutex and
1523  * pass the actual path.
1524  */
1525 int ext4_ext_calc_credits_for_insert(struct inode *inode,
1526 						struct ext4_ext_path *path)
1527 {
1528 	int depth, needed;
1529 
1530 	if (path) {
1531 		/* probably there is space in leaf? */
1532 		depth = ext_depth(inode);
1533 		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1534 				< le16_to_cpu(path[depth].p_hdr->eh_max))
1535 			return 1;
1536 	}
1537 
1538 	/*
1539 	 * given 32-bit logical block (4294967296 blocks), max. tree
1540 	 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
1541 	 * Let's also add one more level for imbalance.
1542 	 */
1543 	depth = 5;
1544 
1545 	/* allocation of new data block(s) */
1546 	needed = 2;
1547 
1548 	/*
1549 	 * tree can be full, so it would need to grow in depth:
1550 	 * we need one credit to modify old root, credits for
1551 	 * new root will be added in split accounting
1552 	 */
1553 	needed += 1;
1554 
1555 	/*
1556 	 * Index split can happen, we would need:
1557 	 *    allocate intermediate indexes (bitmap + group)
1558 	 *  + change two blocks at each level, but root (already included)
1559 	 */
1560 	needed += (depth * 2) + (depth * 2);
1561 
1562 	/* any allocation modifies superblock */
1563 	needed += 1;
1564 
1565 	return needed;
1566 }
1567 
1568 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1569 				struct ext4_extent *ex,
1570 				unsigned long from, unsigned long to)
1571 {
1572 	struct buffer_head *bh;
1573 	int i;
1574 
1575 #ifdef EXTENTS_STATS
1576 	{
1577 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1578 		unsigned short ee_len =  le16_to_cpu(ex->ee_len);
1579 		spin_lock(&sbi->s_ext_stats_lock);
1580 		sbi->s_ext_blocks += ee_len;
1581 		sbi->s_ext_extents++;
1582 		if (ee_len < sbi->s_ext_min)
1583 			sbi->s_ext_min = ee_len;
1584 		if (ee_len > sbi->s_ext_max)
1585 			sbi->s_ext_max = ee_len;
1586 		if (ext_depth(inode) > sbi->s_depth_max)
1587 			sbi->s_depth_max = ext_depth(inode);
1588 		spin_unlock(&sbi->s_ext_stats_lock);
1589 	}
1590 #endif
1591 	if (from >= le32_to_cpu(ex->ee_block)
1592 	    && to == le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1593 		/* tail removal */
1594 		unsigned long num;
1595 		ext4_fsblk_t start;
1596 		num = le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - from;
1597 		start = ext_pblock(ex) + le16_to_cpu(ex->ee_len) - num;
1598 		ext_debug("free last %lu blocks starting %llu\n", num, start);
1599 		for (i = 0; i < num; i++) {
1600 			bh = sb_find_get_block(inode->i_sb, start + i);
1601 			ext4_forget(handle, 0, inode, bh, start + i);
1602 		}
1603 		ext4_free_blocks(handle, inode, start, num);
1604 	} else if (from == le32_to_cpu(ex->ee_block)
1605 		   && to <= le32_to_cpu(ex->ee_block) + le16_to_cpu(ex->ee_len) - 1) {
1606 		printk("strange request: removal %lu-%lu from %u:%u\n",
1607 		       from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1608 	} else {
1609 		printk("strange request: removal(2) %lu-%lu from %u:%u\n",
1610 		       from, to, le32_to_cpu(ex->ee_block), le16_to_cpu(ex->ee_len));
1611 	}
1612 	return 0;
1613 }
1614 
1615 static int
1616 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1617 		struct ext4_ext_path *path, unsigned long start)
1618 {
1619 	int err = 0, correct_index = 0;
1620 	int depth = ext_depth(inode), credits;
1621 	struct ext4_extent_header *eh;
1622 	unsigned a, b, block, num;
1623 	unsigned long ex_ee_block;
1624 	unsigned short ex_ee_len;
1625 	struct ext4_extent *ex;
1626 
1627 	ext_debug("truncate since %lu in leaf\n", start);
1628 	if (!path[depth].p_hdr)
1629 		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1630 	eh = path[depth].p_hdr;
1631 	BUG_ON(eh == NULL);
1632 	BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
1633 	BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
1634 
1635 	/* find where to start removing */
1636 	ex = EXT_LAST_EXTENT(eh);
1637 
1638 	ex_ee_block = le32_to_cpu(ex->ee_block);
1639 	ex_ee_len = le16_to_cpu(ex->ee_len);
1640 
1641 	while (ex >= EXT_FIRST_EXTENT(eh) &&
1642 			ex_ee_block + ex_ee_len > start) {
1643 		ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1644 		path[depth].p_ext = ex;
1645 
1646 		a = ex_ee_block > start ? ex_ee_block : start;
1647 		b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1648 			ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1649 
1650 		ext_debug("  border %u:%u\n", a, b);
1651 
1652 		if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1653 			block = 0;
1654 			num = 0;
1655 			BUG();
1656 		} else if (a != ex_ee_block) {
1657 			/* remove tail of the extent */
1658 			block = ex_ee_block;
1659 			num = a - block;
1660 		} else if (b != ex_ee_block + ex_ee_len - 1) {
1661 			/* remove head of the extent */
1662 			block = a;
1663 			num = b - a;
1664 			/* there is no "make a hole" API yet */
1665 			BUG();
1666 		} else {
1667 			/* remove whole extent: excellent! */
1668 			block = ex_ee_block;
1669 			num = 0;
1670 			BUG_ON(a != ex_ee_block);
1671 			BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1672 		}
1673 
1674 		/* at present, extent can't cross block group: */
1675 		/* leaf + bitmap + group desc + sb + inode */
1676 		credits = 5;
1677 		if (ex == EXT_FIRST_EXTENT(eh)) {
1678 			correct_index = 1;
1679 			credits += (ext_depth(inode)) + 1;
1680 		}
1681 #ifdef CONFIG_QUOTA
1682 		credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1683 #endif
1684 
1685 		handle = ext4_ext_journal_restart(handle, credits);
1686 		if (IS_ERR(handle)) {
1687 			err = PTR_ERR(handle);
1688 			goto out;
1689 		}
1690 
1691 		err = ext4_ext_get_access(handle, inode, path + depth);
1692 		if (err)
1693 			goto out;
1694 
1695 		err = ext4_remove_blocks(handle, inode, ex, a, b);
1696 		if (err)
1697 			goto out;
1698 
1699 		if (num == 0) {
1700 			/* this extent is removed; mark slot entirely unused */
1701 			ext4_ext_store_pblock(ex, 0);
1702 			eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
1703 		}
1704 
1705 		ex->ee_block = cpu_to_le32(block);
1706 		ex->ee_len = cpu_to_le16(num);
1707 
1708 		err = ext4_ext_dirty(handle, inode, path + depth);
1709 		if (err)
1710 			goto out;
1711 
1712 		ext_debug("new extent: %u:%u:%llu\n", block, num,
1713 				ext_pblock(ex));
1714 		ex--;
1715 		ex_ee_block = le32_to_cpu(ex->ee_block);
1716 		ex_ee_len = le16_to_cpu(ex->ee_len);
1717 	}
1718 
1719 	if (correct_index && eh->eh_entries)
1720 		err = ext4_ext_correct_indexes(handle, inode, path);
1721 
1722 	/* if this leaf is free, then we should
1723 	 * remove it from index block above */
1724 	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1725 		err = ext4_ext_rm_idx(handle, inode, path + depth);
1726 
1727 out:
1728 	return err;
1729 }
1730 
1731 /*
1732  * ext4_ext_more_to_rm:
1733  * returns 1 if current index has to be freed (even partial)
1734  */
1735 static int
1736 ext4_ext_more_to_rm(struct ext4_ext_path *path)
1737 {
1738 	BUG_ON(path->p_idx == NULL);
1739 
1740 	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1741 		return 0;
1742 
1743 	/*
1744 	 * if truncate on deeper level happened, it wasn't partial,
1745 	 * so we have to consider current index for truncation
1746 	 */
1747 	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1748 		return 0;
1749 	return 1;
1750 }
1751 
1752 int ext4_ext_remove_space(struct inode *inode, unsigned long start)
1753 {
1754 	struct super_block *sb = inode->i_sb;
1755 	int depth = ext_depth(inode);
1756 	struct ext4_ext_path *path;
1757 	handle_t *handle;
1758 	int i = 0, err = 0;
1759 
1760 	ext_debug("truncate since %lu\n", start);
1761 
1762 	/* probably first extent we're gonna free will be last in block */
1763 	handle = ext4_journal_start(inode, depth + 1);
1764 	if (IS_ERR(handle))
1765 		return PTR_ERR(handle);
1766 
1767 	ext4_ext_invalidate_cache(inode);
1768 
1769 	/*
1770 	 * We start scanning from right side, freeing all the blocks
1771 	 * after i_size and walking into the tree depth-wise.
1772 	 */
1773 	path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_KERNEL);
1774 	if (path == NULL) {
1775 		ext4_journal_stop(handle);
1776 		return -ENOMEM;
1777 	}
1778 	path[0].p_hdr = ext_inode_hdr(inode);
1779 	if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) {
1780 		err = -EIO;
1781 		goto out;
1782 	}
1783 	path[0].p_depth = depth;
1784 
1785 	while (i >= 0 && err == 0) {
1786 		if (i == depth) {
1787 			/* this is leaf block */
1788 			err = ext4_ext_rm_leaf(handle, inode, path, start);
1789 			/* root level has p_bh == NULL, brelse() eats this */
1790 			brelse(path[i].p_bh);
1791 			path[i].p_bh = NULL;
1792 			i--;
1793 			continue;
1794 		}
1795 
1796 		/* this is index block */
1797 		if (!path[i].p_hdr) {
1798 			ext_debug("initialize header\n");
1799 			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
1800 			if (ext4_ext_check_header(__FUNCTION__, inode,
1801 							path[i].p_hdr)) {
1802 				err = -EIO;
1803 				goto out;
1804 			}
1805 		}
1806 
1807 		BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries)
1808 			   > le16_to_cpu(path[i].p_hdr->eh_max));
1809 		BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC);
1810 
1811 		if (!path[i].p_idx) {
1812 			/* this level hasn't been touched yet */
1813 			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
1814 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
1815 			ext_debug("init index ptr: hdr 0x%p, num %d\n",
1816 				  path[i].p_hdr,
1817 				  le16_to_cpu(path[i].p_hdr->eh_entries));
1818 		} else {
1819 			/* we were already here, see at next index */
1820 			path[i].p_idx--;
1821 		}
1822 
1823 		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
1824 				i, EXT_FIRST_INDEX(path[i].p_hdr),
1825 				path[i].p_idx);
1826 		if (ext4_ext_more_to_rm(path + i)) {
1827 			/* go to the next level */
1828 			ext_debug("move to level %d (block %llu)\n",
1829 				  i + 1, idx_pblock(path[i].p_idx));
1830 			memset(path + i + 1, 0, sizeof(*path));
1831 			path[i+1].p_bh =
1832 				sb_bread(sb, idx_pblock(path[i].p_idx));
1833 			if (!path[i+1].p_bh) {
1834 				/* should we reset i_size? */
1835 				err = -EIO;
1836 				break;
1837 			}
1838 
1839 			/* save actual number of indexes since this
1840 			 * number is changed at the next iteration */
1841 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
1842 			i++;
1843 		} else {
1844 			/* we finished processing this index, go up */
1845 			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
1846 				/* index is empty, remove it;
1847 				 * handle must be already prepared by the
1848 				 * truncatei_leaf() */
1849 				err = ext4_ext_rm_idx(handle, inode, path + i);
1850 			}
1851 			/* root level has p_bh == NULL, brelse() eats this */
1852 			brelse(path[i].p_bh);
1853 			path[i].p_bh = NULL;
1854 			i--;
1855 			ext_debug("return to level %d\n", i);
1856 		}
1857 	}
1858 
1859 	/* TODO: flexible tree reduction should be here */
1860 	if (path->p_hdr->eh_entries == 0) {
1861 		/*
1862 		 * truncate to zero freed all the tree,
1863 		 * so we need to correct eh_depth
1864 		 */
1865 		err = ext4_ext_get_access(handle, inode, path);
1866 		if (err == 0) {
1867 			ext_inode_hdr(inode)->eh_depth = 0;
1868 			ext_inode_hdr(inode)->eh_max =
1869 				cpu_to_le16(ext4_ext_space_root(inode));
1870 			err = ext4_ext_dirty(handle, inode, path);
1871 		}
1872 	}
1873 out:
1874 	ext4_ext_tree_changed(inode);
1875 	ext4_ext_drop_refs(path);
1876 	kfree(path);
1877 	ext4_journal_stop(handle);
1878 
1879 	return err;
1880 }
1881 
1882 /*
1883  * called at mount time
1884  */
1885 void ext4_ext_init(struct super_block *sb)
1886 {
1887 	/*
1888 	 * possible initialization would be here
1889 	 */
1890 
1891 	if (test_opt(sb, EXTENTS)) {
1892 		printk("EXT4-fs: file extents enabled");
1893 #ifdef AGGRESSIVE_TEST
1894 		printk(", aggressive tests");
1895 #endif
1896 #ifdef CHECK_BINSEARCH
1897 		printk(", check binsearch");
1898 #endif
1899 #ifdef EXTENTS_STATS
1900 		printk(", stats");
1901 #endif
1902 		printk("\n");
1903 #ifdef EXTENTS_STATS
1904 		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
1905 		EXT4_SB(sb)->s_ext_min = 1 << 30;
1906 		EXT4_SB(sb)->s_ext_max = 0;
1907 #endif
1908 	}
1909 }
1910 
1911 /*
1912  * called at umount time
1913  */
1914 void ext4_ext_release(struct super_block *sb)
1915 {
1916 	if (!test_opt(sb, EXTENTS))
1917 		return;
1918 
1919 #ifdef EXTENTS_STATS
1920 	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
1921 		struct ext4_sb_info *sbi = EXT4_SB(sb);
1922 		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
1923 			sbi->s_ext_blocks, sbi->s_ext_extents,
1924 			sbi->s_ext_blocks / sbi->s_ext_extents);
1925 		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
1926 			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
1927 	}
1928 #endif
1929 }
1930 
1931 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
1932 			ext4_fsblk_t iblock,
1933 			unsigned long max_blocks, struct buffer_head *bh_result,
1934 			int create, int extend_disksize)
1935 {
1936 	struct ext4_ext_path *path = NULL;
1937 	struct ext4_extent newex, *ex;
1938 	ext4_fsblk_t goal, newblock;
1939 	int err = 0, depth;
1940 	unsigned long allocated = 0;
1941 
1942 	__clear_bit(BH_New, &bh_result->b_state);
1943 	ext_debug("blocks %d/%lu requested for inode %u\n", (int) iblock,
1944 			max_blocks, (unsigned) inode->i_ino);
1945 	mutex_lock(&EXT4_I(inode)->truncate_mutex);
1946 
1947 	/* check in cache */
1948 	goal = ext4_ext_in_cache(inode, iblock, &newex);
1949 	if (goal) {
1950 		if (goal == EXT4_EXT_CACHE_GAP) {
1951 			if (!create) {
1952 				/* block isn't allocated yet and
1953 				 * user doesn't want to allocate it */
1954 				goto out2;
1955 			}
1956 			/* we should allocate requested block */
1957 		} else if (goal == EXT4_EXT_CACHE_EXTENT) {
1958 			/* block is already allocated */
1959 		        newblock = iblock
1960 		                   - le32_to_cpu(newex.ee_block)
1961 			           + ext_pblock(&newex);
1962 			/* number of remaining blocks in the extent */
1963 			allocated = le16_to_cpu(newex.ee_len) -
1964 					(iblock - le32_to_cpu(newex.ee_block));
1965 			goto out;
1966 		} else {
1967 			BUG();
1968 		}
1969 	}
1970 
1971 	/* find extent for this block */
1972 	path = ext4_ext_find_extent(inode, iblock, NULL);
1973 	if (IS_ERR(path)) {
1974 		err = PTR_ERR(path);
1975 		path = NULL;
1976 		goto out2;
1977 	}
1978 
1979 	depth = ext_depth(inode);
1980 
1981 	/*
1982 	 * consistent leaf must not be empty;
1983 	 * this situation is possible, though, _during_ tree modification;
1984 	 * this is why assert can't be put in ext4_ext_find_extent()
1985 	 */
1986 	BUG_ON(path[depth].p_ext == NULL && depth != 0);
1987 
1988 	ex = path[depth].p_ext;
1989 	if (ex) {
1990 	        unsigned long ee_block = le32_to_cpu(ex->ee_block);
1991 		ext4_fsblk_t ee_start = ext_pblock(ex);
1992 		unsigned short ee_len  = le16_to_cpu(ex->ee_len);
1993 
1994 		/*
1995 		 * Allow future support for preallocated extents to be added
1996 		 * as an RO_COMPAT feature:
1997 		 * Uninitialized extents are treated as holes, except that
1998 		 * we avoid (fail) allocating new blocks during a write.
1999 		 */
2000 		if (ee_len > EXT_MAX_LEN)
2001 			goto out2;
2002 		/* if found extent covers block, simply return it */
2003 	        if (iblock >= ee_block && iblock < ee_block + ee_len) {
2004 			newblock = iblock - ee_block + ee_start;
2005 			/* number of remaining blocks in the extent */
2006 			allocated = ee_len - (iblock - ee_block);
2007 			ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
2008 					ee_block, ee_len, newblock);
2009 			ext4_ext_put_in_cache(inode, ee_block, ee_len,
2010 						ee_start, EXT4_EXT_CACHE_EXTENT);
2011 			goto out;
2012 		}
2013 	}
2014 
2015 	/*
2016 	 * requested block isn't allocated yet;
2017 	 * we couldn't try to create block if create flag is zero
2018 	 */
2019 	if (!create) {
2020 		/* put just found gap into cache to speed up
2021 		 * subsequent requests */
2022 		ext4_ext_put_gap_in_cache(inode, path, iblock);
2023 		goto out2;
2024 	}
2025 	/*
2026 	 * Okay, we need to do block allocation.  Lazily initialize the block
2027 	 * allocation info here if necessary.
2028 	 */
2029 	if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2030 		ext4_init_block_alloc_info(inode);
2031 
2032 	/* allocate new block */
2033 	goal = ext4_ext_find_goal(inode, path, iblock);
2034 	allocated = max_blocks;
2035 	newblock = ext4_new_blocks(handle, inode, goal, &allocated, &err);
2036 	if (!newblock)
2037 		goto out2;
2038 	ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
2039 			goal, newblock, allocated);
2040 
2041 	/* try to insert new extent into found leaf and return */
2042 	newex.ee_block = cpu_to_le32(iblock);
2043 	ext4_ext_store_pblock(&newex, newblock);
2044 	newex.ee_len = cpu_to_le16(allocated);
2045 	err = ext4_ext_insert_extent(handle, inode, path, &newex);
2046 	if (err)
2047 		goto out2;
2048 
2049 	if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2050 		EXT4_I(inode)->i_disksize = inode->i_size;
2051 
2052 	/* previous routine could use block we allocated */
2053 	newblock = ext_pblock(&newex);
2054 	__set_bit(BH_New, &bh_result->b_state);
2055 
2056 	ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2057 				EXT4_EXT_CACHE_EXTENT);
2058 out:
2059 	if (allocated > max_blocks)
2060 		allocated = max_blocks;
2061 	ext4_ext_show_leaf(inode, path);
2062 	__set_bit(BH_Mapped, &bh_result->b_state);
2063 	bh_result->b_bdev = inode->i_sb->s_bdev;
2064 	bh_result->b_blocknr = newblock;
2065 out2:
2066 	if (path) {
2067 		ext4_ext_drop_refs(path);
2068 		kfree(path);
2069 	}
2070 	mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2071 
2072 	return err ? err : allocated;
2073 }
2074 
2075 void ext4_ext_truncate(struct inode * inode, struct page *page)
2076 {
2077 	struct address_space *mapping = inode->i_mapping;
2078 	struct super_block *sb = inode->i_sb;
2079 	unsigned long last_block;
2080 	handle_t *handle;
2081 	int err = 0;
2082 
2083 	/*
2084 	 * probably first extent we're gonna free will be last in block
2085 	 */
2086 	err = ext4_writepage_trans_blocks(inode) + 3;
2087 	handle = ext4_journal_start(inode, err);
2088 	if (IS_ERR(handle)) {
2089 		if (page) {
2090 			clear_highpage(page);
2091 			flush_dcache_page(page);
2092 			unlock_page(page);
2093 			page_cache_release(page);
2094 		}
2095 		return;
2096 	}
2097 
2098 	if (page)
2099 		ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2100 
2101 	mutex_lock(&EXT4_I(inode)->truncate_mutex);
2102 	ext4_ext_invalidate_cache(inode);
2103 
2104 	/*
2105 	 * TODO: optimization is possible here.
2106 	 * Probably we need not scan at all,
2107 	 * because page truncation is enough.
2108 	 */
2109 	if (ext4_orphan_add(handle, inode))
2110 		goto out_stop;
2111 
2112 	/* we have to know where to truncate from in crash case */
2113 	EXT4_I(inode)->i_disksize = inode->i_size;
2114 	ext4_mark_inode_dirty(handle, inode);
2115 
2116 	last_block = (inode->i_size + sb->s_blocksize - 1)
2117 			>> EXT4_BLOCK_SIZE_BITS(sb);
2118 	err = ext4_ext_remove_space(inode, last_block);
2119 
2120 	/* In a multi-transaction truncate, we only make the final
2121 	 * transaction synchronous. */
2122 	if (IS_SYNC(inode))
2123 		handle->h_sync = 1;
2124 
2125 out_stop:
2126 	/*
2127 	 * If this was a simple ftruncate() and the file will remain alive,
2128 	 * then we need to clear up the orphan record which we created above.
2129 	 * However, if this was a real unlink then we were called by
2130 	 * ext4_delete_inode(), and we allow that function to clean up the
2131 	 * orphan info for us.
2132 	 */
2133 	if (inode->i_nlink)
2134 		ext4_orphan_del(handle, inode);
2135 
2136 	mutex_unlock(&EXT4_I(inode)->truncate_mutex);
2137 	ext4_journal_stop(handle);
2138 }
2139 
2140 /*
2141  * ext4_ext_writepage_trans_blocks:
2142  * calculate max number of blocks we could modify
2143  * in order to allocate new block for an inode
2144  */
2145 int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2146 {
2147 	int needed;
2148 
2149 	needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2150 
2151 	/* caller wants to allocate num blocks, but note it includes sb */
2152 	needed = needed * num - (num - 1);
2153 
2154 #ifdef CONFIG_QUOTA
2155 	needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2156 #endif
2157 
2158 	return needed;
2159 }
2160 
2161 EXPORT_SYMBOL(ext4_mark_inode_dirty);
2162 EXPORT_SYMBOL(ext4_ext_invalidate_cache);
2163 EXPORT_SYMBOL(ext4_ext_insert_extent);
2164 EXPORT_SYMBOL(ext4_ext_walk_space);
2165 EXPORT_SYMBOL(ext4_ext_find_goal);
2166 EXPORT_SYMBOL(ext4_ext_calc_credits_for_insert);
2167 
2168